New implementation, adding a compression to trace file. A basic infrastructure for compression support is added into the trace-cmd library. The zlib is used for compression, but more libraries and algorithms can be added. Trace data is commpressed and part of the trace file metadata: - ftrace events format - format of recorded events - information about the mapping of function addresses to the function names - trace_printk() format strings - information of the mapping a PID to a process name - options Note: not all trace-cmd commands are tested with these changes. These are verified to work with compressed trace files: trace-cmd record trace-cmd report trace-cmd dump trace-cmd agent trace-cmd profile trace-cmd split trace-cmd convert Todo list, in order to have complete trace file compression: - Add support for more compression libraries. - Test all trace-cmd commands with trace file v7. v6 changes: - Documentation upadate. - Bug fixes. - Coding style fixes. v5 changes: - Comress the latency trace data. - Add new "trace-cmd convert" subcommand. - Fixed bug with empty CPUs in the trace file. - A lot of minor bug fixes. v4 changes: - Tested and fixed profile and split subcommands with compression files. - Bug fixes. v3 changes: - Compress the trace data. - Added documentation to all compression APIs. - A few minor fixes. v2 changes: - Refactored compression APIs. - Moved the trace buffers description out of the trace options section. - Added compression of "options" section of the trace file. - Updated "trace-cmd list" to show available compression algorithms. - Tested with host-guest tracing. - Merged with the patchset that bumps the trace file version. Tzvetomir Stoyanov (VMware) (45): trace-cmd library: Remove unused private APIs for creating trace files trace-cmd library: Remove unused API tracecmd_update_option trace-cmd: Check if file version is supported trace-cmd library: Add new API to get file version of input handler trace-cmd library: Select the file version when writing trace file trace-cmd: Add APIs for library initialization and free trace-cmd library: Add support for compression algorithms trace-cmd list: Show supported compression algorithms trace-cmd library: Bump the trace file version to 7 trace-cmd library: Compress part of the trace file trace-cmd library: Read compressed trace file trace-cmd library: Add new API to get compression of input handler trace-cmd library: Inherit compression algorithm from input file trace-cmd library: Extend the create file APIs to support different compression trace-cmd record: Add new parameter --compression trace-cmd dump: Add support for trace files version 7 trace-cmd library: Add support for zlib compression library trace-cmd library: Hide the logic for updating buffer offset trace-cmd: Move buffers description outside of options trace-cmd library: Track the offset in the option section in the trace file trace-cmd library: Add compression of the option section of the trace file trace-cmd library: Refactor the logic for writing trace data in the file trace-cmd library: Add APIs for read and write compressed data in chunks trace-cmd: Compress trace data trace-cmd: Read compressed trace data trace-cmd library: Compress latency trace data trace-cmd: Read compressed latency trace data trace-cmd library: Reuse within the library the function that checks file state. trace-cmd library: Make tracecmd_copy_headers() to work with output handler trace-cmd: Do not use trace file compression with streams trace-cmd library: Add new API to get file version of output handler trace-cmd: Add file state parameter to tracecmd_copy trace-cmd: Copy CPU count in tracecmd_copy trace-cmd: Copy buffers description in tracecmd_copy trace-cmd: Copy options in tracecmd_copy trace-cmd library: Refactor the logic for writing CPU trace data trace-cmd library: Refactor the logic for writing CPU instance trace data trace-cmd: Copy trace data in tracecmd_copy trace-cmd: Add compression parameter to tracecmd_copy trace-cmd: Add new command "trace-cmd convert" trace-cmd record: Update man page trace-cmd: Add convert man page trace-cmd: Update bash completion trace-cmd list: Update the man page trace-cmd: Update trace.dat man page .../trace-cmd/trace-cmd-convert.1.txt | 60 ++ Documentation/trace-cmd/trace-cmd-list.1.txt | 3 + .../trace-cmd/trace-cmd-record.1.txt | 10 + Documentation/trace-cmd/trace-cmd.1.txt | 4 +- Documentation/trace-cmd/trace-cmd.dat.5.txt | 56 +- lib/trace-cmd/Makefile | 11 + .../include/private/trace-cmd-private.h | 92 +- lib/trace-cmd/include/trace-cmd-local.h | 44 +- lib/trace-cmd/trace-compress-zlib.c | 172 +++ lib/trace-cmd/trace-compress.c | 797 ++++++++++++++ lib/trace-cmd/trace-input.c | 999 ++++++++++++++---- lib/trace-cmd/trace-output.c | 877 ++++++++++----- lib/trace-cmd/trace-util.c | 60 ++ tracecmd/Makefile | 1 + tracecmd/include/trace-local.h | 2 + tracecmd/trace-cmd.bash | 37 + tracecmd/trace-cmd.c | 12 +- tracecmd/trace-convert.c | 94 ++ tracecmd/trace-dump.c | 162 ++- tracecmd/trace-list.c | 26 + tracecmd/trace-listen.c | 3 + tracecmd/trace-read.c | 14 +- tracecmd/trace-record.c | 48 +- tracecmd/trace-restore.c | 6 +- tracecmd/trace-split.c | 2 +- tracecmd/trace-stream.c | 2 +- tracecmd/trace-usage.c | 18 + 27 files changed, 3076 insertions(+), 536 deletions(-) create mode 100644 Documentation/trace-cmd/trace-cmd-convert.1.txt create mode 100644 lib/trace-cmd/trace-compress-zlib.c create mode 100644 lib/trace-cmd/trace-compress.c create mode 100644 tracecmd/trace-convert.c -- 2.31.1
Remove these unused private trace-cmd library APIs: tracecmd_create_file() tracecmd_create_file_glob() Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- .../include/private/trace-cmd-private.h | 6 ---- lib/trace-cmd/trace-output.c | 29 ------------------- 2 files changed, 35 deletions(-) diff --git a/lib/trace-cmd/include/private/trace-cmd-private.h b/lib/trace-cmd/include/private/trace-cmd-private.h index 5c2ab4ca..4bd36baa 100644 --- a/lib/trace-cmd/include/private/trace-cmd-private.h +++ b/lib/trace-cmd/include/private/trace-cmd-private.h @@ -263,12 +263,6 @@ struct tracecmd_option; struct tracecmd_msg_handle; struct tracecmd_output *tracecmd_create_file_latency(const char *output_file, int cpus); -struct tracecmd_output *tracecmd_create_file(const char *output_file, - int cpus, char * const *cpu_data_files); -struct tracecmd_output * -tracecmd_create_file_glob(const char *output_file, - int cpus, char * const *cpu_data_files, - struct tracecmd_event_list *event_globs); struct tracecmd_output * tracecmd_create_init_file_glob(const char *output_file, struct tracecmd_event_list *list); diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index a4a1eecc..9231c319 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -1614,35 +1614,6 @@ struct tracecmd_output *tracecmd_get_output_handle_fd(int fd) return NULL; } -struct tracecmd_output * -tracecmd_create_file_glob(const char *output_file, - int cpus, char * const *cpu_data_files, - struct tracecmd_event_list *list) -{ - struct tracecmd_output *handle; - - handle = create_file(output_file, NULL, NULL, NULL, list); - if (!handle) - return NULL; - - if (tracecmd_write_cmdlines(handle)) - return NULL; - - if (tracecmd_append_cpu_data(handle, cpus, cpu_data_files) < 0) { - tracecmd_output_close(handle); - return NULL; - } - - return handle; -} - -struct tracecmd_output *tracecmd_create_file(const char *output_file, - int cpus, char * const *cpu_data_files) -{ - return tracecmd_create_file_glob(output_file, cpus, - cpu_data_files, &all_event_list); -} - struct tracecmd_output *tracecmd_create_init_fd(int fd) { return create_file_fd(fd, NULL, NULL, NULL, &all_event_list, NULL); -- 2.31.1
Removed not used trace-cmd library API for updating the options section of already written trace file: tracecmd_update_option() Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- .../include/private/trace-cmd-private.h | 3 -- lib/trace-cmd/trace-output.c | 40 ------------------- 2 files changed, 43 deletions(-) diff --git a/lib/trace-cmd/include/private/trace-cmd-private.h b/lib/trace-cmd/include/private/trace-cmd-private.h index 4bd36baa..49539432 100644 --- a/lib/trace-cmd/include/private/trace-cmd-private.h +++ b/lib/trace-cmd/include/private/trace-cmd-private.h @@ -290,9 +290,6 @@ int tracecmd_write_cpus(struct tracecmd_output *handle, int cpus); int tracecmd_write_cmdlines(struct tracecmd_output *handle); int tracecmd_write_options(struct tracecmd_output *handle); int tracecmd_append_options(struct tracecmd_output *handle); -int tracecmd_update_option(struct tracecmd_output *handle, - struct tracecmd_option *option, int size, - const void *data); void tracecmd_output_close(struct tracecmd_output *handle); void tracecmd_output_free(struct tracecmd_output *handle); struct tracecmd_output *tracecmd_copy(struct tracecmd_input *ihandle, diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index 9231c319..c0aa58cf 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -1229,46 +1229,6 @@ int tracecmd_append_options(struct tracecmd_output *handle) return 0; } -int tracecmd_update_option(struct tracecmd_output *handle, - struct tracecmd_option *option, int size, - const void *data) -{ - tsize_t offset; - stsize_t ret; - - if (size > option->size) { - tracecmd_warning("Can't update option with more data than allocated"); - return -1; - } - - if (handle->file_state < TRACECMD_FILE_OPTIONS) { - /* Hasn't been written yet. Just update current pointer */ - option->size = size; - memcpy(option->data, data, size); - return 0; - } - - /* Save current offset */ - offset = lseek64(handle->fd, 0, SEEK_CUR); - - ret = lseek64(handle->fd, option->offset, SEEK_SET); - if (ret == (off64_t)-1) { - tracecmd_warning("could not seek to %lld\n", option->offset); - return -1; - } - - if (do_write_check(handle, data, size)) - return -1; - - ret = lseek64(handle->fd, offset, SEEK_SET); - if (ret == (off64_t)-1) { - tracecmd_warning("could not seek to %lld\n", offset); - return -1; - } - - return 0; -} - struct tracecmd_option * tracecmd_add_buffer_option(struct tracecmd_output *handle, const char *name, int cpus) -- 2.31.1
When reading a trace file, version of the file is ignored. This could case problems when bumping the version number because of changes in in the structure of the file. The old code should detect unsupported file version and should not try to read it. A new trace-cmd library API is added to check if version is supported: tracecmd_is_version_supported() Checks are added in the code to ensure not trying to read trace file from unsupported version. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/include/private/trace-cmd-private.h | 2 ++ lib/trace-cmd/trace-input.c | 10 ++++++++++ lib/trace-cmd/trace-util.c | 7 +++++++ tracecmd/trace-dump.c | 7 +++++++ 4 files changed, 26 insertions(+) diff --git a/lib/trace-cmd/include/private/trace-cmd-private.h b/lib/trace-cmd/include/private/trace-cmd-private.h index 49539432..6fc18938 100644 --- a/lib/trace-cmd/include/private/trace-cmd-private.h +++ b/lib/trace-cmd/include/private/trace-cmd-private.h @@ -42,6 +42,8 @@ void tracecmd_record_ref(struct tep_record *record); void tracecmd_set_debug(bool set_debug); bool tracecmd_get_debug(void); +bool tracecmd_is_version_supported(unsigned int version); + struct tracecmd_output; struct tracecmd_recorder; struct hook_list; diff --git a/lib/trace-cmd/trace-input.c b/lib/trace-cmd/trace-input.c index 5ee69b14..97ad0a5d 100644 --- a/lib/trace-cmd/trace-input.c +++ b/lib/trace-cmd/trace-input.c @@ -131,6 +131,7 @@ struct tracecmd_input { bool use_trace_clock; bool read_page; bool use_pipe; + int file_version; struct cpu_data *cpu_data; long long ts_offset; struct tsc2nsec tsc_calc; @@ -3269,6 +3270,7 @@ struct tracecmd_input *tracecmd_alloc_fd(int fd, int flags) unsigned int page_size; char *version; char buf[BUFSIZ]; + unsigned long ver; handle = malloc(sizeof(*handle)); if (!handle) @@ -3293,6 +3295,14 @@ struct tracecmd_input *tracecmd_alloc_fd(int fd, int flags) if (!version) goto failed_read; tracecmd_info("version = %s\n", version); + ver = strtol(version, NULL, 10); + if (!ver && errno) + goto failed_read; + if (!tracecmd_is_version_supported(ver)) { + tracecmd_warning("Unsupported file version %lu", ver); + goto failed_read; + } + handle->file_version = ver; free(version); if (do_read_check(handle, buf, 1)) diff --git a/lib/trace-cmd/trace-util.c b/lib/trace-cmd/trace-util.c index b65f9dec..b0c98c72 100644 --- a/lib/trace-cmd/trace-util.c +++ b/lib/trace-cmd/trace-util.c @@ -617,3 +617,10 @@ unsigned long long tracecmd_generate_traceid(void) free(str); return hash; } + +bool tracecmd_is_version_supported(unsigned int version) +{ + if (version <= FILE_VERSION) + return true; + return false; +} diff --git a/tracecmd/trace-dump.c b/tracecmd/trace-dump.c index 98425b98..03cc82b4 100644 --- a/tracecmd/trace-dump.c +++ b/tracecmd/trace-dump.c @@ -10,6 +10,7 @@ #include <getopt.h> #include <sys/stat.h> #include <fcntl.h> +#include <errno.h> #include "trace-local.h" @@ -145,6 +146,7 @@ static void dump_initial_format(int fd) char magic[] = TRACECMD_MAGIC; char buf[DUMP_SIZE]; int val4; + unsigned long ver; do_print(SUMMARY, "\t[Initial format]\n"); @@ -166,6 +168,11 @@ static void dump_initial_format(int fd) die("no version string"); do_print(SUMMARY, "\t\t%s\t[Version]\n", buf); + ver = strtol(buf, NULL, 10); + if (!ver && errno) + die("Invalid file version string %s", buf); + if (!tracecmd_is_version_supported(ver)) + die("Unsupported file version %lu", ver); /* get file endianness*/ if (read_file_bytes(fd, buf, 1)) -- 2.31.1
Added an API to get the version of the trace file, associated with given input file handler. tracecmd_get_in_file_version() Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/include/private/trace-cmd-private.h | 2 ++ lib/trace-cmd/trace-input.c | 9 +++++++++ 2 files changed, 11 insertions(+) diff --git a/lib/trace-cmd/include/private/trace-cmd-private.h b/lib/trace-cmd/include/private/trace-cmd-private.h index 6fc18938..01b12c47 100644 --- a/lib/trace-cmd/include/private/trace-cmd-private.h +++ b/lib/trace-cmd/include/private/trace-cmd-private.h @@ -232,6 +232,8 @@ int tracecmd_set_cursor(struct tracecmd_input *handle, unsigned long long tracecmd_get_cursor(struct tracecmd_input *handle, int cpu); +unsigned long tracecmd_get_in_file_version(struct tracecmd_input *handle); + int tracecmd_ftrace_overrides(struct tracecmd_input *handle, struct tracecmd_ftrace *finfo); bool tracecmd_get_use_trace_clock(struct tracecmd_input *handle); tracecmd_show_data_func diff --git a/lib/trace-cmd/trace-input.c b/lib/trace-cmd/trace-input.c index 97ad0a5d..46619ac7 100644 --- a/lib/trace-cmd/trace-input.c +++ b/lib/trace-cmd/trace-input.c @@ -4046,6 +4046,15 @@ struct tep_handle *tracecmd_get_tep(struct tracecmd_input *handle) return handle->pevent; } +/** + * tracecmd_get_in_file_version - return the trace.dat file version + * @handle: input handle for the trace.dat file + */ +unsigned long tracecmd_get_in_file_version(struct tracecmd_input *handle) +{ + return handle->file_version; +} + /** * tracecmd_get_use_trace_clock - return use_trace_clock * @handle: input handle for the trace.dat file -- 2.31.1
When a new output handler to trace file is allocated, select the proper file version. If this output handler is based on an existing input trace file handler, inherit the trace file version. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/trace-output.c | 19 ++++++++++++++++++- 1 file changed, 18 insertions(+), 1 deletion(-) diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index c0aa58cf..bacda23b 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -57,6 +57,7 @@ struct tracecmd_output { int nr_options; bool quiet; unsigned long file_state; + unsigned long file_version; struct list_head options; struct tracecmd_msg_handle *msg_handle; char *trace_clock; @@ -907,6 +908,17 @@ out_free: return ret; } +static int select_file_version(struct tracecmd_output *handle, + struct tracecmd_input *ihandle) +{ + if (ihandle) + handle->file_version = tracecmd_get_in_file_version(ihandle); + else + handle->file_version = FILE_VERSION; + + return 0; +} + static struct tracecmd_output * create_file_fd(int fd, struct tracecmd_input *ihandle, const char *tracing_dir, @@ -933,6 +945,9 @@ create_file_fd(int fd, struct tracecmd_input *ihandle, handle->msg_handle = msg_handle; + if (select_file_version(handle, ihandle)) + goto out_free; + list_head_init(&handle->options); buf[0] = 23; @@ -943,7 +958,8 @@ create_file_fd(int fd, struct tracecmd_input *ihandle, if (do_write_check(handle, buf, 10)) goto out_free; - if (do_write_check(handle, FILE_VERSION_STRING, strlen(FILE_VERSION_STRING) + 1)) + sprintf(buf, "%lu", handle->file_version); + if (do_write_check(handle, buf, strlen(buf) + 1)) goto out_free; /* get endian and page size */ @@ -1562,6 +1578,7 @@ struct tracecmd_output *tracecmd_get_output_handle_fd(int fd) handle->pevent = tracecmd_get_tep(ihandle); tep_ref(handle->pevent); handle->page_size = tracecmd_page_size(ihandle); + handle->file_version = tracecmd_get_in_file_version(ihandle); list_head_init(&handle->options); tracecmd_close(ihandle); -- 2.31.1
The trace-cmd library has no APIs for initialization and free of the whole library. Added these new APIs: tracecmd_lib_init() tracecmd_lib_free() and call them in trace-cmd main function. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/include/private/trace-cmd-private.h | 3 +++ lib/trace-cmd/trace-util.c | 9 +++++++++ tracecmd/trace-cmd.c | 11 ++++++++--- 3 files changed, 20 insertions(+), 3 deletions(-) diff --git a/lib/trace-cmd/include/private/trace-cmd-private.h b/lib/trace-cmd/include/private/trace-cmd-private.h index 01b12c47..be8b3c48 100644 --- a/lib/trace-cmd/include/private/trace-cmd-private.h +++ b/lib/trace-cmd/include/private/trace-cmd-private.h @@ -29,6 +29,9 @@ struct tep_plugin_list *trace_load_plugins(struct tep_handle *tep, int flags); +int tracecmd_lib_init(void); +void tracecmd_lib_free(void); + int *tracecmd_add_id(int *list, int id, int len); enum { diff --git a/lib/trace-cmd/trace-util.c b/lib/trace-cmd/trace-util.c index b0c98c72..61054ad2 100644 --- a/lib/trace-cmd/trace-util.c +++ b/lib/trace-cmd/trace-util.c @@ -624,3 +624,12 @@ bool tracecmd_is_version_supported(unsigned int version) return true; return false; } + +int tracecmd_lib_init(void) +{ + return 0; +} + +void tracecmd_lib_free(void) +{ +} diff --git a/tracecmd/trace-cmd.c b/tracecmd/trace-cmd.c index 00cdaa37..71c8f6d6 100644 --- a/tracecmd/trace-cmd.c +++ b/tracecmd/trace-cmd.c @@ -142,15 +142,20 @@ int main (int argc, char **argv) if (argc < 2) trace_usage(argc, argv); + tracecmd_lib_init(); + for (i = 0; i < ARRAY_SIZE(commands); ++i) { if (strcmp(argv[1], commands[i].name) == 0 ){ commands[i].run(argc, argv); - goto out; + break; } } + tracecmd_lib_free(); + /* No valid command found, show help */ - trace_usage(argc, argv); -out: + if (i == ARRAY_SIZE(commands)) + trace_usage(argc, argv); + exit(0); } -- 2.31.1
Added infrastructure to trace-cmd library for compression. Introduced various new APIs to work with this new functionality: struct tracecmd_compression tracecmd_compress_init() tracecmd_compress_free() tracecmd_compress_alloc() tracecmd_compress_destroy() tracecmd_compress_block() tracecmd_uncompress_block() tracecmd_compress_reset() tracecmd_compress_read() tracecmd_compress_pread() tracecmd_compress_write() tracecmd_compress_lseek() tracecmd_compress_proto_get_name() tracecmd_compress_is_supported() tracecmd_compress_protos_get() tracecmd_compress_proto_register() The compression algorithms are not part of this patch. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/Makefile | 1 + .../include/private/trace-cmd-private.h | 26 + lib/trace-cmd/include/trace-cmd-local.h | 2 + lib/trace-cmd/trace-compress.c | 589 ++++++++++++++++++ lib/trace-cmd/trace-util.c | 3 + 5 files changed, 621 insertions(+) create mode 100644 lib/trace-cmd/trace-compress.c diff --git a/lib/trace-cmd/Makefile b/lib/trace-cmd/Makefile index 17600318..bab4322d 100644 --- a/lib/trace-cmd/Makefile +++ b/lib/trace-cmd/Makefile @@ -25,6 +25,7 @@ ifeq ($(VSOCK_DEFINED), 1) OBJS += trace-timesync-ptp.o OBJS += trace-timesync-kvm.o endif +OBJS += trace-compress.o # Additional util objects OBJS += trace-blk-hack.o diff --git a/lib/trace-cmd/include/private/trace-cmd-private.h b/lib/trace-cmd/include/private/trace-cmd-private.h index be8b3c48..ec534d51 100644 --- a/lib/trace-cmd/include/private/trace-cmd-private.h +++ b/lib/trace-cmd/include/private/trace-cmd-private.h @@ -464,6 +464,32 @@ void tracecmd_tsync_free(struct tracecmd_time_sync *tsync); int tracecmd_write_guest_time_shift(struct tracecmd_output *handle, struct tracecmd_time_sync *tsync); +/* --- Compression --- */ +struct tracecmd_compression; +struct tracecmd_compression *tracecmd_compress_alloc(const char *name, const char *version, + int fd, struct tep_handle *tep, + struct tracecmd_msg_handle *msg_handle); +void tracecmd_compress_destroy(struct tracecmd_compression *handle); +int tracecmd_compress_block(struct tracecmd_compression *handle); +int tracecmd_uncompress_block(struct tracecmd_compression *handle); +void tracecmd_compress_reset(struct tracecmd_compression *handle); +int tracecmd_compress_read(struct tracecmd_compression *handle, char *dst, int len); +int tracecmd_compress_pread(struct tracecmd_compression *handle, char *dst, int len, off_t offset); +int tracecmd_compress_write(struct tracecmd_compression *handle, + const void *data, unsigned long long size); +off_t tracecmd_compress_lseek(struct tracecmd_compression *handle, off_t offset, int whence); +int tracecmd_compress_proto_get_name(struct tracecmd_compression *compress, + const char **name, const char **version); +bool tracecmd_compress_is_supported(const char *name, const char *version); +int tracecmd_compress_protos_get(char ***names, char ***versions); +int tracecmd_compress_proto_register(const char *name, const char *version, int weight, + int (*compress)(char *in, unsigned int in_bytes, + char *out, unsigned int *out_bytes), + int (*uncompress)(char *in, unsigned int in_bytes, + char *out, unsigned int *out_bytes), + unsigned int (*comress_size)(unsigned int bytes), + bool (*is_supported)(const char *name, const char *version)); + /* --- Plugin handling --- */ extern struct tep_plugin_option trace_ftrace_options[]; diff --git a/lib/trace-cmd/include/trace-cmd-local.h b/lib/trace-cmd/include/trace-cmd-local.h index 821b5cdb..c6591e56 100644 --- a/lib/trace-cmd/include/trace-cmd-local.h +++ b/lib/trace-cmd/include/trace-cmd-local.h @@ -31,5 +31,7 @@ void tracecmd_info(const char *fmt, ...); #endif #endif +void tracecmd_compress_init(void); +void tracecmd_compress_free(void); #endif /* _TRACE_CMD_LOCAL_H */ diff --git a/lib/trace-cmd/trace-compress.c b/lib/trace-cmd/trace-compress.c new file mode 100644 index 00000000..378a3e57 --- /dev/null +++ b/lib/trace-cmd/trace-compress.c @@ -0,0 +1,589 @@ +// SPDX-License-Identifier: LGPL-2.1 +/* + * Copyright (C) 2021, VMware, Tzvetomir Stoyanov tz.stoyanov@gmail.com> + * + */ +#include <stdlib.h> +#include <sys/time.h> +#include <fcntl.h> +#include <errno.h> +#include <unistd.h> + +#include "trace-cmd-private.h" +#include "trace-cmd-local.h" + +struct compress_proto { + struct compress_proto *next; + char *proto_name; + char *proto_version; + int weight; + + int (*compress_block)(char *in, unsigned int in_bytes, + char *out, unsigned int *out_bytes); + int (*uncompress_block)(char *in, unsigned int in_bytes, + char *out, unsigned int *out_bytes); + unsigned int (*compress_size)(unsigned int bytes); + bool (*is_supported)(const char *name, const char *version); +}; + +static struct compress_proto *proto_list; + +struct tracecmd_compression { + int fd; + unsigned int capacity; + unsigned long pointer; + char *buffer; + struct compress_proto *proto; + struct tep_handle *tep; + struct tracecmd_msg_handle *msg_handle; +}; + +static int read_fd(int fd, char *dst, int len) +{ + size_t size = 0; + int r; + + do { + r = read(fd, dst+size, len); + if (r > 0) { + size += r; + len -= r; + } else + break; + } while (r > 0); + + if (len) + return -1; + return size; +} + +static long long write_fd(int fd, const void *data, size_t size) +{ + long long tot = 0; + long long w; + + do { + w = write(fd, data + tot, size - tot); + tot += w; + + if (!w) + break; + if (w < 0) + return w; + } while (tot != size); + + return tot; +} + +static long long do_write(struct tracecmd_compression *handle, + const void *data, unsigned long long size) +{ + int ret; + + if (handle->msg_handle) { + ret = tracecmd_msg_data_send(handle->msg_handle, data, size); + if (ret) + return -1; + return size; + } + return write_fd(handle->fd, data, size); +} + +/** + * tracecmd_compress_lseek - Move the read/write pointer into the compression buffer + * @handle: compression handler + * @offset: number of bytes to move the pointer, can be negative or positive + * @whence: the starting position of the pointer movement, + * + * Returns the new file pointer on success, or -1 in case of an error. + */ +off_t tracecmd_compress_lseek(struct tracecmd_compression *handle, off_t offset, int whence) +{ + unsigned long p, extend; + char *buf; + + if (!handle || !handle->buffer) + return (off_t)-1; + + switch (whence) { + case SEEK_CUR: + p = handle->pointer + offset; + break; + case SEEK_END: + p = handle->capacity + offset; + break; + case SEEK_SET: + p = offset; + break; + default: + return (off_t)-1; + } + + if (p <= handle->capacity) { + handle->pointer = p; + } else { + extend = p - handle->capacity; + extend = extend < BUFSIZ ? BUFSIZ : extend; + buf = realloc(handle->buffer, handle->capacity + extend); + if (!buf) + return (off_t)-1; + handle->buffer = buf; + handle->capacity += extend; + handle->pointer = p; + } + + return p; +} + +static int compress_read(struct tracecmd_compression *handle, char *dst, int len) +{ + int s; + + if (handle->pointer + len > handle->capacity) + s = handle->capacity - handle->pointer; + else + s = len; + memcpy(dst, handle->buffer + handle->pointer, s); + + return s; +} + +/** + * tracecmd_compress_pread - pread() on compression buffer + * @handle: compression handler + * @dst: return, store the read data + * @len: length of data to be read + * @offset: offset in the buffer of data to be read + * + * Read a @len of data from the compression buffer at given @offset, + * without updating the buffer pointer. + * + * On success returns the number of bytes read, or -1 on failure. + */ +int tracecmd_compress_pread(struct tracecmd_compression *handle, char *dst, int len, off_t offset) +{ + int ret; + + if (!handle || !handle->buffer || offset > handle->capacity) + return -1; + + ret = tracecmd_compress_lseek(handle, offset, SEEK_SET); + if (ret < 0) + return ret; + return compress_read(handle, dst, len); +} + +/** + * tracecmd_compress_read - read() from compression buffer + * @handle: compression handler + * @dst: return, store the read data + * @len: length of data to be read + * + * Read a @len of data from the compression buffer + * + * On success returns the number of bytes read, or -1 on failure. + */ +int tracecmd_compress_read(struct tracecmd_compression *handle, char *dst, int len) +{ + int ret; + + if (!handle || !handle->buffer) + return -1; + + ret = compress_read(handle, dst, len); + if (ret > 0) + handle->pointer += ret; + + return ret; +} + +/** + * tracecmd_compress_reset - Reset the compression buffer + * @handle: compression handler + * + * Reset the compression buffer, any data currently in the buffer will be destroyed. + * + */ +void tracecmd_compress_reset(struct tracecmd_compression *handle) +{ + if (!handle) + return; + + free(handle->buffer); + handle->buffer = NULL; + handle->pointer = 0; + handle->capacity = 0; +} + +/** + * tracecmd_uncompress_block - uncompress a memory block + * @handle: compression handler + * + * Read compressed memory block from the file and uncompress it into internal buffer. + * The tracecmd_compress_read() can be used to read the uncompressed data from the buffer + * + * Returns 0 on success, or -1 in case of an error. + */ +int tracecmd_uncompress_block(struct tracecmd_compression *handle) +{ + unsigned int s_uncompressed; + unsigned int s_compressed; + char *bytes = NULL; + char buf[4]; + int ret; + + if (!handle || !handle->proto || !handle->proto->uncompress_block) + return -1; + tracecmd_compress_reset(handle); + + if (read(handle->fd, buf, 4) != 4) + return -1; + s_compressed = tep_read_number(handle->tep, buf, 4); + if (read(handle->fd, buf, 4) != 4) + return -1; + s_uncompressed = tep_read_number(handle->tep, buf, 4); + + handle->buffer = malloc(s_uncompressed); + if (!handle->buffer) + return -1; + bytes = malloc(s_compressed); + if (!bytes) + goto error; + + if (read_fd(handle->fd, bytes, s_compressed) < 0) + goto error; + ret = handle->proto->uncompress_block(bytes, s_compressed, + handle->buffer, &s_uncompressed); + if (ret) + goto error; + free(bytes); + handle->pointer = 0; + handle->capacity = s_uncompressed; + return 0; +error: + tracecmd_compress_reset(handle); + free(bytes); + return -1; +} + +/** + * tracecmd_compress_block - compress a memory block + * @handle: compression handler + * + * Compress the content of the internal memory buffer and write the compressed data in the file + * The tracecmd_compress_write() can be used to write data into the internal memory buffer, before + * calling this API. + * + * Returns 0 on success, or -1 in case of an error. + */ +int tracecmd_compress_block(struct tracecmd_compression *handle) +{ + unsigned int size; + char *buf; + int endian4; + int ret; + + if (!handle || !handle->proto || + !handle->proto->compress_size || !handle->proto->compress_block) + return -1; + + size = handle->proto->compress_size(handle->pointer); + buf = malloc(size); + if (!buf) + return -1; + ret = handle->proto->compress_block(handle->buffer, handle->pointer, buf, &size); + if (ret < 0) + goto out; + /* Write compressed data size */ + endian4 = tep_read_number(handle->tep, &size, 4); + ret = do_write(handle, &endian4, 4); + if (ret != 4) + goto out; + /* Write uncompressed data size */ + endian4 = tep_read_number(handle->tep, &handle->pointer, 4); + ret = do_write(handle, &endian4, 4); + if (ret != 4) + goto out; + /* Write compressed data */ + ret = do_write(handle, buf, size); + ret = ((ret == size) ? 0 : -1); +out: + tracecmd_compress_reset(handle); + free(buf); + return ret; +} + +/** + * tracecmd_compress_write - write() to compression buffer + * @handle: compression handler + * @data: data to be written + * @size: size of @data + * + * Write @data of @size in the compression buffer + * + * Returns 0 on success, or -1 on failure. + */ +int tracecmd_compress_write(struct tracecmd_compression *handle, + const void *data, unsigned long long size) +{ + char *buf; + int extend; + + if (!handle) + return -1; + + if (handle->capacity < handle->pointer + size) { + extend = (handle->pointer + size) - handle->capacity; + extend = extend < BUFSIZ ? BUFSIZ : extend; + buf = realloc(handle->buffer, handle->capacity + extend); + if (!buf) + return -1; + handle->buffer = buf; + handle->capacity += extend; + } + memcpy(&handle->buffer[handle->pointer], data, size); + handle->pointer += size; + return 0; +} + +/** + * tracecmd_compress_init - initialize the library with available compression algorithms + */ +void tracecmd_compress_init(void) +{ + struct timeval time; + + gettimeofday(&time, NULL); + srand((time.tv_sec * 1000) + (time.tv_usec / 1000)); +} + +static struct compress_proto *compress_proto_select(void) +{ + struct compress_proto *proto = proto_list; + struct compress_proto *selected = NULL; + + while (proto) { + if (!selected || selected->weight > proto->weight) + selected = proto; + proto = proto->next; + } + + return selected; +} + +/** + * tracecmd_compress_alloc - Allocate a new compression context + * @name: name of the compression algorithm, if NULL - auto select the best available algorithm + * @version: version of the compression algorithm, can be NULL + * @fd: file descriptor for reading / writing data + * @tep: tep handler, used to encode the data + * @msg_handle: message handler, use it for reading / writing data instead of @fd + * + * Returns NULL on failure or pointer to allocated compression context. + * The returned context must be freed by tracecmd_compress_destroy() + */ +struct tracecmd_compression *tracecmd_compress_alloc(const char *name, const char *version, + int fd, struct tep_handle *tep, + struct tracecmd_msg_handle *msg_handle) +{ + struct tracecmd_compression *new; + struct compress_proto *proto; + + if (name) { + proto = proto_list; + while (proto) { + if (proto->is_supported && proto->is_supported(name, version)) + break; + proto = proto->next; + } + } else { + proto = compress_proto_select(); + } + if (!proto) + return NULL; + + new = calloc(1, sizeof(*new)); + if (!new) + return NULL; + new->fd = fd; + new->tep = tep; + new->msg_handle = msg_handle; + new->proto = proto; + return new; +} + +/** + * tracecmd_compress_destroy - Free a compression context + * @handle: handle to the compression context that will be freed + */ +void tracecmd_compress_destroy(struct tracecmd_compression *handle) +{ + tracecmd_compress_reset(handle); + free(handle); +} + +/** + * tracecmd_compress_is_supported - check if compression algorithm with given name and + * version is supported + * @name: name of the compression algorithm. + * @version: version of the compression algorithm. + * + * Returns true if the algorithm with given name and version is supported or false if it is not. + */ +bool tracecmd_compress_is_supported(const char *name, const char *version) +{ + struct compress_proto *proto = proto_list; + + if (!name) + return NULL; + + while (proto) { + if (proto->is_supported && proto->is_supported(name, version)) + return true; + proto = proto->next; + } + return false; +} + +/** + * tracecmd_compress_proto_get_name - get name and version of compression algorithm + * @compress: compression handler. + * @name: return, name of the compression algorithm. + * @version: return, version of the compression algorithm. + * + * Returns 0 on success, or -1 in case of an error. If 0 is returned, the name and version of the + * algorithm are stored in @name and @version. The returned strings must *not* be freed. + */ +int tracecmd_compress_proto_get_name(struct tracecmd_compression *compress, + const char **name, const char **version) +{ + if (!compress || !compress->proto) + return -1; + if (name) + *name = compress->proto->proto_name; + if (version) + *version = compress->proto->proto_version; + return 0; +} + +/** + * tracecmd_compress_proto_register - register a new compression algorithm + * @name: name of the compression algorithm. + * @version: version of the compression algorithm. + * @weight: weight of the compression algorithm, lower is better. + * @compress: compression hook, called to compress a memory block. + * @uncompress: uncompression hook, called to uncompress a memory block. + * @compress_size: hook, called to get the required minimum size of the buffer for compression + * given number of bytes. + * @is_supported: check hook, called to check if compression with given name and version is + * supported by this plugin. + * + * Returns 0 on success, or -1 in case of an error. If algorithm with given name and version is + * already registered, -1 is returned. + */ +int tracecmd_compress_proto_register(const char *name, const char *version, int weight, + int (*compress)(char *in, unsigned int in_bytes, + char *out, unsigned int *out_bytes), + int (*uncompress)(char *in, unsigned int in_bytes, + char *out, unsigned int *out_bytes), + unsigned int (*compress_size)(unsigned int bytes), + bool (*is_supported)(const char *name, const char *version)) +{ + struct compress_proto *new; + + if (!name || !compress || !uncompress) + return -1; + if (tracecmd_compress_is_supported(name, version)) + return -1; + + new = calloc(1, sizeof(*new)); + if (!new) + return -1; + + new->proto_name = strdup(name); + if (!new->proto_name) + goto error; + new->proto_version = strdup(version); + if (!new->proto_version) + goto error; + new->compress_block = compress; + new->uncompress_block = uncompress; + new->compress_size = compress_size; + new->is_supported = is_supported; + new->weight = weight; + new->next = proto_list; + proto_list = new; + return 0; + +error: + free(new->proto_name); + free(new->proto_version); + free(new); + return -1; +} + +/** + * tracecmd_compress_free - free the library resources, related to available compression algorithms + * + */ +void tracecmd_compress_free(void) +{ + struct compress_proto *proto = proto_list; + struct compress_proto *del; + + while (proto) { + del = proto; + proto = proto->next; + free(del->proto_name); + free(del->proto_version); + free(del); + } + proto_list = NULL; +} + +/** + * tracecmd_compress_protos_get - get a list of all supported compression algorithms and versions + * @names: return, array with names of all supported compression algorithms + * @versions: return, array with versions of all supported compression algorithms + * + * On success, the size of @names and @versions arrays is returned. Those arrays are allocated by + * the API and must be freed with free() by the caller. Both arrays are with same size, each name + * from @names corresponds to a version from @versions. + * On error -1 is returned and @names and @versions arrays are not allocated. + */ +int tracecmd_compress_protos_get(char ***names, char ***versions) +{ + struct compress_proto *proto = proto_list; + char **n = NULL; + char **v = NULL; + int c, i; + + for (c = 0; proto; proto = proto->next) + c++; + + if (c < 1) + return c; + + n = calloc(c, sizeof(char *)); + if (!n) + goto error; + v = calloc(c, sizeof(char *)); + if (!v) + goto error; + + proto = proto_list; + for (i = 0; i < c && proto; i++) { + n[i] = proto->proto_name; + v[i] = proto->proto_version; + proto = proto->next; + } + + *names = n; + *versions = v; + return c; + +error: + free(n); + free(v); + return -1; +} diff --git a/lib/trace-cmd/trace-util.c b/lib/trace-cmd/trace-util.c index 61054ad2..136260c6 100644 --- a/lib/trace-cmd/trace-util.c +++ b/lib/trace-cmd/trace-util.c @@ -627,9 +627,12 @@ bool tracecmd_is_version_supported(unsigned int version) int tracecmd_lib_init(void) { + + tracecmd_compress_init(); return 0; } void tracecmd_lib_free(void) { + tracecmd_compress_free(); } -- 2.31.1
Add new parameter "trace-cmd list -c" to show supported compression algorithms. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- tracecmd/trace-list.c | 26 ++++++++++++++++++++++++++ tracecmd/trace-usage.c | 1 + 2 files changed, 27 insertions(+) diff --git a/tracecmd/trace-list.c b/tracecmd/trace-list.c index d060c810..900da73b 100644 --- a/tracecmd/trace-list.c +++ b/tracecmd/trace-list.c @@ -549,6 +549,24 @@ static void show_plugins(void) tep_free(pevent); } +static void show_compression(void) +{ + char **versions, **names; + int c, i; + + c = tracecmd_compress_protos_get(&names, &versions); + if (c <= 0) { + printf("No compression algorithms are supported\n"); + return; + } + printf("Supported compression algorithms:\n"); + for (i = 0; i < c; i++) + printf("\t%s, %s\n", names[i], versions[i]); + + free(names); + free(versions); +} + void trace_list(int argc, char **argv) { int events = 0; @@ -562,6 +580,7 @@ void trace_list(int argc, char **argv) int flags = 0; int systems = 0; int show_all = 1; + int compression = 0; int i; const char *arg; const char *funcre = NULL; @@ -626,6 +645,10 @@ void trace_list(int argc, char **argv) systems = 1; show_all = 0; break; + case 'c': + compression = 1; + show_all = 0; + break; case '-': if (strcmp(argv[i], "--debug") == 0) { tracecmd_set_debug(true); @@ -670,6 +693,8 @@ void trace_list(int argc, char **argv) show_clocks(); if (systems) show_systems(); + if (compression) + show_compression(); if (show_all) { printf("event systems:\n"); show_systems(); @@ -679,6 +704,7 @@ void trace_list(int argc, char **argv) show_tracers(); printf("\noptions:\n"); show_options(); + show_compression(); } return; diff --git a/tracecmd/trace-usage.c b/tracecmd/trace-usage.c index c0033ce8..cd8ca391 100644 --- a/tracecmd/trace-usage.c +++ b/tracecmd/trace-usage.c @@ -331,6 +331,7 @@ static struct usage_help usage_help[] = { " -O list plugin options\n" " -B list defined buffer instances\n" " -C list the defined clocks (and active one)\n" + " -c list the supported trace file compression algorithms\n" }, { "restore", -- 2.31.1
Adding a compression of the trace.dat file will change its structure. These changes are not backward compatible, the old trace-cmd binaries will not be able to read compressed trace files. Bumping the version to 7 will prevent old trace-cmd to read such files. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/include/trace-cmd-local.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lib/trace-cmd/include/trace-cmd-local.h b/lib/trace-cmd/include/trace-cmd-local.h index c6591e56..4d5d1455 100644 --- a/lib/trace-cmd/include/trace-cmd-local.h +++ b/lib/trace-cmd/include/trace-cmd-local.h @@ -15,7 +15,7 @@ void tracecmd_critical(const char *fmt, ...); void tracecmd_info(const char *fmt, ...); /* trace.dat file format version */ -#define FILE_VERSION 6 +#define FILE_VERSION 7 #define _STR(x) #x #define STR(x) _STR(x) -- 2.31.1
Compress part of the trace.dat file metadata. If there is compression support, chose file version 7 and compress these parts of the file: - ftrace events format - format of recorded events - information of the mapping of function addresses to the function names - trace_printk() format strings - information of the mapping a PID to a process name A new compression header is added in the file, right after the page size information. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/include/trace-cmd-local.h | 4 + lib/trace-cmd/trace-output.c | 145 ++++++++++++++++++++---- 2 files changed, 127 insertions(+), 22 deletions(-) diff --git a/lib/trace-cmd/include/trace-cmd-local.h b/lib/trace-cmd/include/trace-cmd-local.h index 4d5d1455..93613cd7 100644 --- a/lib/trace-cmd/include/trace-cmd-local.h +++ b/lib/trace-cmd/include/trace-cmd-local.h @@ -34,4 +34,8 @@ void tracecmd_info(const char *fmt, ...); void tracecmd_compress_init(void); void tracecmd_compress_free(void); +int out_compression_start(struct tracecmd_output *handle); +int out_compression_end(struct tracecmd_output *handle); +void out_compression_reset(struct tracecmd_output *handle); + #endif /* _TRACE_CMD_LOCAL_H */ diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index bacda23b..87816b37 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -58,6 +58,10 @@ struct tracecmd_output { bool quiet; unsigned long file_state; unsigned long file_version; + + bool do_compress; + struct tracecmd_compression *compress; + struct list_head options; struct tracecmd_msg_handle *msg_handle; char *trace_clock; @@ -78,12 +82,31 @@ struct list_event_system { static stsize_t do_write_check(struct tracecmd_output *handle, const void *data, tsize_t size) { + if (handle->do_compress) + return tracecmd_compress_write(handle->compress, data, size); + if (handle->msg_handle) return tracecmd_msg_data_send(handle->msg_handle, data, size); return __do_write_check(handle->fd, data, size); } +static inline int do_lseek(struct tracecmd_output *handle, off_t offset, int whence) +{ + if (handle->do_compress) + return tracecmd_compress_lseek(handle->compress, offset, whence); + else + return lseek64(handle->fd, offset, whence); +} + +static inline int do_preed(struct tracecmd_output *handle, void *dst, int len, off_t offset) +{ + if (handle->do_compress) + return tracecmd_compress_pread(handle->compress, dst, len, offset); + else + return pread(handle->fd, dst, len, offset); +} + static short convert_endian_2(struct tracecmd_output *handle, short val) { if (!handle->pevent) @@ -109,6 +132,31 @@ static unsigned long long convert_endian_8(struct tracecmd_output *handle, return tep_read_number(handle->pevent, &val, 8); } +__hidden void out_compression_reset(struct tracecmd_output *handle) +{ + if (handle->file_version < 7) + return; + tracecmd_compress_reset(handle->compress); + handle->do_compress = false; +} + +__hidden int out_compression_start(struct tracecmd_output *handle) +{ + if (handle->file_version < 7 || !handle->compress) + return 0; + tracecmd_compress_reset(handle->compress); + handle->do_compress = true; + return 0; +} + +__hidden int out_compression_end(struct tracecmd_output *handle) +{ + if (handle->file_version < 7) + return 0; + handle->do_compress = false; + return tracecmd_compress_block(handle->compress); +} + /** * tracecmd_set_quiet - Set if to print output to the screen * @quiet: If non zero, print no output to the screen @@ -159,6 +207,7 @@ void tracecmd_output_free(struct tracecmd_output *handle) free(option); } free(handle->trace_clock); + tracecmd_compress_destroy(handle->compress); free(handle); } @@ -659,13 +708,17 @@ static int read_ftrace_files(struct tracecmd_output *handle) } create_event_list_item(handle, &systems, &list); - + out_compression_start(handle); ret = copy_event_system(handle, systems); + if (!ret) + ret = out_compression_end(handle); + else + out_compression_reset(handle); free_list_events(systems); - handle->file_state = TRACECMD_FILE_FTRACE_EVENTS; - + if (!ret) + handle->file_state = TRACECMD_FILE_FTRACE_EVENTS; return ret; } @@ -715,6 +768,7 @@ static int read_event_files(struct tracecmd_output *handle, for (slist = systems; slist; slist = slist->next) count++; + out_compression_start(handle); ret = -1; endian4 = convert_endian_4(handle, count); if (do_write_check(handle, &endian4, 4)) @@ -730,8 +784,14 @@ static int read_event_files(struct tracecmd_output *handle, ret = copy_event_system(handle, slist); } - handle->file_state = TRACECMD_FILE_ALL_EVENTS; + if (!ret) + ret = out_compression_end(handle); out_free: + if (!ret) + handle->file_state = TRACECMD_FILE_ALL_EVENTS; + else + out_compression_reset(handle); + free_list_events(systems); return ret; @@ -793,20 +853,20 @@ static int read_proc_kallsyms(struct tracecmd_output *handle, if (kallsyms) path = kallsyms; - + out_compression_start(handle); ret = stat(path, &st); if (ret < 0) { /* not found */ size = 0; endian4 = convert_endian_4(handle, size); - if (do_write_check(handle, &endian4, 4)) - return -1; - return 0; + ret = do_write_check(handle, &endian4, 4); + goto out; } size = get_size(path); endian4 = convert_endian_4(handle, size); - if (do_write_check(handle, &endian4, 4)) - return -1; + ret = do_write_check(handle, &endian4, 4); + if (ret) + goto out; set_proc_kptr_restrict(0); check_size = copy_file(handle, path); @@ -814,13 +874,18 @@ static int read_proc_kallsyms(struct tracecmd_output *handle, errno = EINVAL; tracecmd_warning("error in size of file '%s'", path); set_proc_kptr_restrict(1); - return -1; + ret = -1; + goto out; } set_proc_kptr_restrict(1); - handle->file_state = TRACECMD_FILE_KALLSYMS; - - return 0; + ret = out_compression_end(handle); +out: + if (!ret) + handle->file_state = TRACECMD_FILE_KALLSYMS; + else + out_compression_reset(handle); + return ret; } static int read_ftrace_printk(struct tracecmd_output *handle) @@ -840,6 +905,7 @@ static int read_ftrace_printk(struct tracecmd_output *handle) if (!path) return -1; + out_compression_start(handle); ret = stat(path, &st); if (ret < 0) { /* not found */ @@ -861,11 +927,14 @@ static int read_ftrace_printk(struct tracecmd_output *handle) } out: - handle->file_state = TRACECMD_FILE_PRINTK; put_tracing_file(path); + if (out_compression_end(handle)) + return -1; + handle->file_state = TRACECMD_FILE_PRINTK; return 0; fail: put_tracing_file(path); + out_compression_reset(handle); return -1; } @@ -911,14 +980,39 @@ out_free: static int select_file_version(struct tracecmd_output *handle, struct tracecmd_input *ihandle) { - if (ihandle) + if (ihandle) { handle->file_version = tracecmd_get_in_file_version(ihandle); - else - handle->file_version = FILE_VERSION; + } else { + handle->compress = tracecmd_compress_alloc(NULL, NULL, handle->fd, + handle->pevent, handle->msg_handle); + if (handle->compress) + handle->file_version = 7; + else + handle->file_version = 6; + } return 0; } +static int write_compression_header(struct tracecmd_output *handle) +{ + const char *name = NULL; + const char *ver = NULL; + char *buf; + int ret; + + ret = tracecmd_compress_proto_get_name(handle->compress, &name, &ver); + if (ret < 0 || !name || !ver) + return -1; + ret = asprintf(&buf, "%s %s", name, ver); + if (ret < 0) + return -1; + ret = do_write_check(handle, buf, strlen(buf) + 1); + + free(buf); + return ret; +} + static struct tracecmd_output * create_file_fd(int fd, struct tracecmd_input *ihandle, const char *tracing_dir, @@ -992,8 +1086,9 @@ create_file_fd(int fd, struct tracecmd_input *ihandle, endian4 = convert_endian_4(handle, handle->page_size); if (do_write_check(handle, &endian4, 4)) goto out_free; + if (handle->file_version >= 7 && write_compression_header(handle)) + goto out_free; handle->file_state = TRACECMD_FILE_INIT; - if (ihandle) return handle; @@ -1285,11 +1380,17 @@ int tracecmd_write_cmdlines(struct tracecmd_output *handle) handle->file_state); return ret; } + out_compression_start(handle); + ret = save_tracing_file_data(handle, "saved_cmdlines"); - if (ret < 0) + if (ret < 0) { + out_compression_reset(handle); return ret; - handle->file_state = TRACECMD_FILE_CMD_LINES; - return 0; + } + ret = out_compression_end(handle); + if (!ret) + handle->file_state = TRACECMD_FILE_CMD_LINES; + return ret; } struct tracecmd_output *tracecmd_create_file_latency(const char *output_file, int cpus) -- 2.31.1
If a trace file version 7 is detected, read compressed sections of it. Read the compression header, check if the compression algorithm and version, used to compress the file, are supported and use it to uncompress these sections of the file: - ftrace events format - format of recorded events - information of the mapping of function addresses to the function names - trace_printk() format strings - information of the mapping a PID to a process name Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/include/trace-cmd-local.h | 3 + lib/trace-cmd/trace-input.c | 257 ++++++++++++++++-------- 2 files changed, 178 insertions(+), 82 deletions(-) diff --git a/lib/trace-cmd/include/trace-cmd-local.h b/lib/trace-cmd/include/trace-cmd-local.h index 93613cd7..419bfee0 100644 --- a/lib/trace-cmd/include/trace-cmd-local.h +++ b/lib/trace-cmd/include/trace-cmd-local.h @@ -38,4 +38,7 @@ int out_compression_start(struct tracecmd_output *handle); int out_compression_end(struct tracecmd_output *handle); void out_compression_reset(struct tracecmd_output *handle); +void in_uncompress_reset(struct tracecmd_input *handle); +int in_uncompress_block(struct tracecmd_input *handle); + #endif /* _TRACE_CMD_LOCAL_H */ diff --git a/lib/trace-cmd/trace-input.c b/lib/trace-cmd/trace-input.c index 46619ac7..43d731e0 100644 --- a/lib/trace-cmd/trace-input.c +++ b/lib/trace-cmd/trace-input.c @@ -136,6 +136,9 @@ struct tracecmd_input { long long ts_offset; struct tsc2nsec tsc_calc; + bool read_compress; + struct tracecmd_compression *compress; + struct host_trace_info host; double ts2secs; char * cpustats; @@ -255,12 +258,28 @@ static ssize_t do_read(struct tracecmd_input *handle, void *data, size_t size) return tot; } +static inline int do_lseek(struct tracecmd_input *handle, int offset, int whence) +{ + if (handle->read_compress) + return tracecmd_compress_lseek(handle->compress, offset, whence); + else + return lseek(handle->fd, offset, whence); +} + +static inline ssize_t do_read_compressed(struct tracecmd_input *handle, void *data, size_t size) +{ + if (handle->read_compress) + return tracecmd_compress_read(handle->compress, data, size); + else + return do_read(handle, data, size); +} + static ssize_t do_read_check(struct tracecmd_input *handle, void *data, size_t size) { ssize_t ret; - ret = do_read(handle, data, size); + ret = do_read_compressed(handle, data, size); if (ret < 0) return ret; if (ret != size) @@ -278,10 +297,8 @@ static char *read_string(struct tracecmd_input *handle) ssize_t r; for (;;) { - r = do_read(handle, buf, BUFSIZ); - if (r < 0) - goto fail; - if (!r) + r = do_read_compressed(handle, buf, BUFSIZ); + if (r <= 0) goto fail; for (i = 0; i < r; i++) { @@ -307,7 +324,7 @@ static char *read_string(struct tracecmd_input *handle) } /* move the file descriptor to the end of the string */ - r = lseek(handle->fd, -(r - (i+1)), SEEK_CUR); + r = do_lseek(handle, -(r - (i+1)), SEEK_CUR); if (r < 0) goto fail; @@ -359,6 +376,26 @@ static int read8(struct tracecmd_input *handle, unsigned long long *size) return 0; } +__hidden void in_uncompress_reset(struct tracecmd_input *handle) +{ + if (handle->compress && handle->file_version >= 7) { + handle->read_compress = false; + tracecmd_compress_reset(handle->compress); + } +} + +__hidden int in_uncompress_block(struct tracecmd_input *handle) +{ + int ret = 0; + + if (handle->compress && handle->file_version >= 7) { + ret = tracecmd_uncompress_block(handle->compress); + if (!ret) + handle->read_compress = true; + } + return ret; +} + static int read_header_files(struct tracecmd_input *handle) { struct tep_handle *pevent = handle->pevent; @@ -602,34 +639,40 @@ static int read_ftrace_files(struct tracecmd_input *handle, const char *regex) } } - if (read4(handle, &count) < 0) + if (in_uncompress_block(handle)) return -1; + ret = read4(handle, &count); + if (ret < 0) + goto out; + for (i = 0; i < count; i++) { - if (read8(handle, &size) < 0) - return -1; + ret = read8(handle, &size); + if (ret < 0) + goto out; ret = read_ftrace_file(handle, size, print_all, ereg); if (ret < 0) - return -1; + goto out; } handle->event_files_start = lseek64(handle->fd, 0, SEEK_CUR); + handle->file_state = TRACECMD_FILE_FTRACE_EVENTS; + ret = 0; +out: if (sreg) { regfree(sreg); regfree(ereg); } - - handle->file_state = TRACECMD_FILE_FTRACE_EVENTS; - - return 0; + in_uncompress_reset(handle); + return ret; } static int read_event_files(struct tracecmd_input *handle, const char *regex) { unsigned long long size; - char *system; + char *system = NULL; regex_t spreg; regex_t epreg; regex_t *sreg = NULL; @@ -654,13 +697,19 @@ static int read_event_files(struct tracecmd_input *handle, const char *regex) return -1; } - if (read4(handle, &systems) < 0) + if (in_uncompress_block(handle)) return -1; + ret = read4(handle, &systems); + if (ret < 0) + goto out; + for (i = 0; i < systems; i++) { system = read_string(handle); - if (!system) - return -1; + if (!system) { + ret = -1; + goto out; + } sys_printed = 0; print_all = 0; @@ -687,103 +736,117 @@ static int read_event_files(struct tracecmd_input *handle, const char *regex) } } - if (read4(handle, &count) < 0) - goto failed; + ret = read4(handle, &count); + if (ret < 0) + goto out; for (x=0; x < count; x++) { - if (read8(handle, &size) < 0) - goto failed; + ret = read8(handle, &size); + if (ret < 0) + goto out; ret = read_event_file(handle, system, size, print_all, &sys_printed, reg); if (ret < 0) - goto failed; + goto out; } free(system); - } - - if (sreg) { - regfree(sreg); - regfree(ereg); + system = NULL; } handle->file_state = TRACECMD_FILE_ALL_EVENTS; - - return 0; - - failed: + ret = 0; + out: + in_uncompress_reset(handle); if (sreg) { regfree(sreg); regfree(ereg); } free(system); - return -1; + return ret; } static int read_proc_kallsyms(struct tracecmd_input *handle) { - struct tep_handle *pevent = handle->pevent; + struct tep_handle *tep = handle->pevent; unsigned int size; - char *buf; + char *buf = NULL; + int ret; if (handle->file_state >= TRACECMD_FILE_KALLSYMS) return 0; - if (read4(handle, &size) < 0) + if (in_uncompress_block(handle)) return -1; - if (!size) - return 0; /* OK? */ - buf = malloc(size+1); - if (!buf) - return -1; - if (do_read_check(handle, buf, size)){ - free(buf); - return -1; + ret = read4(handle, &size); + if (ret < 0) + goto out; + if (!size) { + handle->file_state = TRACECMD_FILE_KALLSYMS; + goto out; /* OK? */ } - buf[size] = 0; - - tep_parse_kallsyms(pevent, buf); - free(buf); + buf = malloc(size+1); + if (!buf) { + ret = -1; + goto out; + } + ret = do_read_check(handle, buf, size); + if (ret < 0) + goto out; + buf[size] = 0; + tep_parse_kallsyms(tep, buf); handle->file_state = TRACECMD_FILE_KALLSYMS; - - return 0; + ret = 0; +out: + free(buf); + in_uncompress_reset(handle); + return ret; } static int read_ftrace_printk(struct tracecmd_input *handle) { unsigned int size; - char *buf; + char *buf = NULL; + int ret; if (handle->file_state >= TRACECMD_FILE_PRINTK) return 0; - if (read4(handle, &size) < 0) + if (in_uncompress_block(handle)) return -1; - if (!size) - return 0; /* OK? */ + + ret = read4(handle, &size); + if (ret < 0) + goto out; + if (!size) { + handle->file_state = TRACECMD_FILE_PRINTK; + goto out; /* OK? */ + } buf = malloc(size + 1); - if (!buf) - return -1; - if (do_read_check(handle, buf, size)) { - free(buf); - return -1; + if (!buf) { + ret = -1; + goto out; } + ret = do_read_check(handle, buf, size); + if (ret < 0) + goto out; buf[size] = 0; tep_parse_printk_formats(handle->pevent, buf); - - free(buf); - handle->file_state = TRACECMD_FILE_PRINTK; + ret = 0; - return 0; +out: + free(buf); + in_uncompress_reset(handle); + return ret; } static int read_and_parse_cmdlines(struct tracecmd_input *handle); @@ -2974,20 +3037,30 @@ static int read_and_parse_cmdlines(struct tracecmd_input *handle) { struct tep_handle *pevent = handle->pevent; unsigned long long size; - char *cmdlines; + char *cmdlines = NULL; + int ret; if (handle->file_state >= TRACECMD_FILE_CMD_LINES) return 0; - if (read_data_and_size(handle, &cmdlines, &size) < 0) + if (in_uncompress_block(handle)) return -1; + + ret = read_data_and_size(handle, &cmdlines, &size); + if (ret < 0) + goto out; + if (!size) { + handle->file_state = TRACECMD_FILE_CMD_LINES; + goto out; + } cmdlines[size] = 0; tep_parse_saved_cmdlines(pevent, cmdlines); - free(cmdlines); - handle->file_state = TRACECMD_FILE_CMD_LINES; - - return 0; + ret = 0; +out: + free(cmdlines); + in_uncompress_reset(handle); + return ret; } static void extract_trace_clock(struct tracecmd_input *handle, char *line) @@ -3268,7 +3341,9 @@ struct tracecmd_input *tracecmd_alloc_fd(int fd, int flags) struct tracecmd_input *handle; char test[] = TRACECMD_MAGIC; unsigned int page_size; - char *version; + size_t offset; + char *str = NULL; + char *zver; char buf[BUFSIZ]; unsigned long ver; @@ -3291,11 +3366,11 @@ struct tracecmd_input *tracecmd_alloc_fd(int fd, int flags) if (memcmp(buf, "tracing", 7) != 0) goto failed_read; - version = read_string(handle); - if (!version) + str = read_string(handle); + if (!str) goto failed_read; - tracecmd_info("version = %s\n", version); - ver = strtol(version, NULL, 10); + tracecmd_info("version = %s\n", str); + ver = strtol(str, NULL, 10); if (!ver && errno) goto failed_read; if (!tracecmd_is_version_supported(ver)) { @@ -3303,7 +3378,8 @@ struct tracecmd_input *tracecmd_alloc_fd(int fd, int flags) goto failed_read; } handle->file_version = ver; - free(version); + free(str); + str = NULL; if (do_read_check(handle, buf, 1)) goto failed_read; @@ -3328,20 +3404,36 @@ struct tracecmd_input *tracecmd_alloc_fd(int fd, int flags) read4(handle, &page_size); handle->page_size = page_size; - handle->header_files_start = - lseek64(handle->fd, 0, SEEK_CUR); + offset = lseek64(handle->fd, 0, SEEK_CUR); + handle->total_file_size = lseek64(handle->fd, 0, SEEK_END); + lseek64(handle->fd, offset, SEEK_SET); - handle->total_file_size = - lseek64(handle->fd, 0, SEEK_END); + if (handle->file_version >= 7) { + str = read_string(handle); + if (!str) + goto failed_read; + zver = strchr(str, ' '); + if (!zver) + goto failed_read; + *zver = '\0'; + handle->compress = tracecmd_compress_alloc(str, zver + 1, + handle->fd, handle->pevent, NULL); + if (!handle->compress) { + tracecmd_warning("Unsupported file compression %s %s", str, zver + 1); + goto failed_read; + } + free(str); + str = NULL; + } handle->header_files_start = - lseek64(handle->fd, handle->header_files_start, SEEK_SET); - + lseek64(handle->fd, 0, SEEK_CUR); handle->file_state = TRACECMD_FILE_INIT; return handle; failed_read: + free(str); free(handle); return NULL; @@ -3519,7 +3611,8 @@ void tracecmd_close(struct tracecmd_input *handle) if (handle->flags & TRACECMD_FL_BUFFER_INSTANCE) tracecmd_close(handle->parent); else { - /* Only main handle frees plugins and pevent */ + /* Only main handle frees plugins, pevent and compression context */ + tracecmd_compress_destroy(handle->compress); tep_unload_plugins(handle->plugin_list, handle->pevent); tep_free(handle->pevent); } -- 2.31.1
The new API returns name and version of the compression algorithm, used to compress the trace file associated with given input file handler: tracecmd_get_file_compress_proto() Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- .../include/private/trace-cmd-private.h | 2 ++ lib/trace-cmd/trace-input.c | 16 ++++++++++++++++ 2 files changed, 18 insertions(+) diff --git a/lib/trace-cmd/include/private/trace-cmd-private.h b/lib/trace-cmd/include/private/trace-cmd-private.h index ec534d51..624d0afb 100644 --- a/lib/trace-cmd/include/private/trace-cmd-private.h +++ b/lib/trace-cmd/include/private/trace-cmd-private.h @@ -236,6 +236,8 @@ unsigned long long tracecmd_get_cursor(struct tracecmd_input *handle, int cpu); unsigned long tracecmd_get_in_file_version(struct tracecmd_input *handle); +int tracecmd_get_file_compress_proto(struct tracecmd_input *handle, + const char **name, const char **version); int tracecmd_ftrace_overrides(struct tracecmd_input *handle, struct tracecmd_ftrace *finfo); bool tracecmd_get_use_trace_clock(struct tracecmd_input *handle); diff --git a/lib/trace-cmd/trace-input.c b/lib/trace-cmd/trace-input.c index 43d731e0..0b17b58d 100644 --- a/lib/trace-cmd/trace-input.c +++ b/lib/trace-cmd/trace-input.c @@ -4148,6 +4148,22 @@ unsigned long tracecmd_get_in_file_version(struct tracecmd_input *handle) return handle->file_version; } +/** + * tracecmd_get_file_compress_proto - get name and version of compression algorithm, + * used to compress the trace file + * @handle: input handle for the trace.dat file + * @name: return, name of the compression algorithm. + * @version: return, version of the compression algorithm. + * + * Returns 0 on success, or -1 in case of an error. If 0 is returned, the name and version of the + * algorithm are stored in @name and @version. The returned strings must *not* be freed. + */ +int tracecmd_get_file_compress_proto(struct tracecmd_input *handle, + const char **name, const char **version) +{ + return tracecmd_compress_proto_get_name(handle->compress, name, version); +} + /** * tracecmd_get_use_trace_clock - return use_trace_clock * @handle: input handle for the trace.dat file -- 2.31.1
When a new trace file output handler is allocated, based on given trace file input handler - use the same compression algorithm. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/trace-output.c | 32 +++++++++++++++++++++++++------- 1 file changed, 25 insertions(+), 7 deletions(-) diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index 87816b37..82b3e5df 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -980,17 +980,27 @@ out_free: static int select_file_version(struct tracecmd_output *handle, struct tracecmd_input *ihandle) { + const char *cname = NULL; + const char *cver = NULL; + if (ihandle) { handle->file_version = tracecmd_get_in_file_version(ihandle); - } else { - handle->compress = tracecmd_compress_alloc(NULL, NULL, handle->fd, - handle->pevent, handle->msg_handle); - if (handle->compress) - handle->file_version = 7; - else - handle->file_version = 6; + if (!tracecmd_get_file_compress_proto(ihandle, &cname, &cver)) { + handle->compress = tracecmd_compress_alloc(cname, cver, handle->fd, + handle->pevent, handle->msg_handle); + if (!handle->compress) + return -1; + } + return 0; } + handle->compress = tracecmd_compress_alloc(NULL, NULL, handle->fd, + handle->pevent, handle->msg_handle); + if (handle->compress) + handle->file_version = 7; + else + handle->file_version = 6; + return 0; } @@ -1645,6 +1655,8 @@ struct tracecmd_output *tracecmd_get_output_handle_fd(int fd) { struct tracecmd_output *handle = NULL; struct tracecmd_input *ihandle; + const char *cname = NULL; + const char *cver = NULL; int fd2; /* Move the file descriptor to the beginning */ @@ -1682,6 +1694,12 @@ struct tracecmd_output *tracecmd_get_output_handle_fd(int fd) handle->file_version = tracecmd_get_in_file_version(ihandle); list_head_init(&handle->options); + if (!tracecmd_get_file_compress_proto(ihandle, &cname, &cver)) { + handle->compress = tracecmd_compress_alloc(cname, cver, handle->fd, + handle->pevent, handle->msg_handle); + if (!handle->compress) + goto out_free; + } tracecmd_close(ihandle); return handle; -- 2.31.1
Added additional parameter for file compression to all trace-cmd library APIs for creating a new trace file. The caller could specify what compression algorithm should be used when creating a new trace file: tracecmd_create_file_latency tracecmd_create_init_file_glob tracecmd_create_init_fd_glob tracecmd_create_init_fd_msg tracecmd_create_init_file tracecmd_create_init_file_override tracecmd_create_init_fd The new parameter is of type string and can be one of: "any" - auto select the best available compression algorithm "none" - do not compress the trace file name - the name of the desired compression algorithms If a compression is used, trace file version 7 is selected, otherwise trace file version 6 is used. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- .../include/private/trace-cmd-private.h | 19 +++-- lib/trace-cmd/trace-output.c | 79 ++++++++++++------- tracecmd/trace-record.c | 12 ++- tracecmd/trace-restore.c | 4 +- tracecmd/trace-stream.c | 2 +- 5 files changed, 70 insertions(+), 46 deletions(-) diff --git a/lib/trace-cmd/include/private/trace-cmd-private.h b/lib/trace-cmd/include/private/trace-cmd-private.h index 624d0afb..ee73325c 100644 --- a/lib/trace-cmd/include/private/trace-cmd-private.h +++ b/lib/trace-cmd/include/private/trace-cmd-private.h @@ -271,20 +271,23 @@ struct tracecmd_event_list { struct tracecmd_option; struct tracecmd_msg_handle; -struct tracecmd_output *tracecmd_create_file_latency(const char *output_file, int cpus); +struct tracecmd_output *tracecmd_create_file_latency(const char *output_file, int cpus, + const char *compression); struct tracecmd_output * -tracecmd_create_init_file_glob(const char *output_file, - struct tracecmd_event_list *list); -struct tracecmd_output *tracecmd_create_init_fd(int fd); +tracecmd_create_init_file_glob(const char *output_file, struct tracecmd_event_list *list, + const char *compression); +struct tracecmd_output *tracecmd_create_init_fd(int fd, const char *compression); struct tracecmd_output * -tracecmd_create_init_fd_glob(int fd, struct tracecmd_event_list *list); +tracecmd_create_init_fd_glob(int fd, struct tracecmd_event_list *list, const char *compression); struct tracecmd_output * tracecmd_create_init_fd_msg(struct tracecmd_msg_handle *msg_handle, - struct tracecmd_event_list *list); -struct tracecmd_output *tracecmd_create_init_file(const char *output_file); + struct tracecmd_event_list *list, + const char *compression); +struct tracecmd_output *tracecmd_create_init_file(const char *output_file, const char *compression); struct tracecmd_output *tracecmd_create_init_file_override(const char *output_file, const char *tracing_dir, - const char *kallsyms); + const char *kallsyms, + const char *compression); struct tracecmd_option *tracecmd_add_option(struct tracecmd_output *handle, unsigned short id, int size, const void *data); diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index 82b3e5df..7c7d3d76 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -978,24 +978,41 @@ out_free: } static int select_file_version(struct tracecmd_output *handle, - struct tracecmd_input *ihandle) + struct tracecmd_input *ihandle, + const char *compression) { const char *cname = NULL; const char *cver = NULL; - if (ihandle) { - handle->file_version = tracecmd_get_in_file_version(ihandle); + if (compression) { /* If compression is specified - use it */ + if (!strcmp(compression, "any")) { + handle->compress = tracecmd_compress_alloc(NULL, NULL, handle->fd, + handle->pevent, handle->msg_handle); + if (!handle->compress) { + tracecmd_warning("No compression algorithms are not supported"); + return -1; + } + } else if (strcmp(compression, "none")) { + handle->compress = tracecmd_compress_alloc(compression, NULL, handle->fd, + handle->pevent, handle->msg_handle); + if (!handle->compress) { + tracecmd_warning("Compression algorithm %s is not supported", + compression); + return -1; + } + } + } else if (ihandle) { /* If there is input handler - inherit compression */ if (!tracecmd_get_file_compress_proto(ihandle, &cname, &cver)) { handle->compress = tracecmd_compress_alloc(cname, cver, handle->fd, handle->pevent, handle->msg_handle); if (!handle->compress) return -1; } - return 0; + } else { /* default - auto select the best available compression algorithm */ + handle->compress = tracecmd_compress_alloc(NULL, NULL, handle->fd, + handle->pevent, handle->msg_handle); } - handle->compress = tracecmd_compress_alloc(NULL, NULL, handle->fd, - handle->pevent, handle->msg_handle); if (handle->compress) handle->file_version = 7; else @@ -1028,7 +1045,8 @@ create_file_fd(int fd, struct tracecmd_input *ihandle, const char *tracing_dir, const char *kallsyms, struct tracecmd_event_list *list, - struct tracecmd_msg_handle *msg_handle) + struct tracecmd_msg_handle *msg_handle, + const char *compression) { struct tracecmd_output *handle; struct tep_handle *pevent; @@ -1049,7 +1067,7 @@ create_file_fd(int fd, struct tracecmd_input *ihandle, handle->msg_handle = msg_handle; - if (select_file_version(handle, ihandle)) + if (select_file_version(handle, ihandle, compression)) goto out_free; list_head_init(&handle->options); @@ -1128,7 +1146,8 @@ static struct tracecmd_output *create_file(const char *output_file, struct tracecmd_input *ihandle, const char *tracing_dir, const char *kallsyms, - struct tracecmd_event_list *list) + struct tracecmd_event_list *list, + const char *compression) { struct tracecmd_output *handle; int fd; @@ -1137,7 +1156,7 @@ static struct tracecmd_output *create_file(const char *output_file, if (fd < 0) return NULL; - handle = create_file_fd(fd, ihandle, tracing_dir, kallsyms, list, NULL); + handle = create_file_fd(fd, ihandle, tracing_dir, kallsyms, list, NULL, compression); if (!handle) { close(fd); unlink(output_file); @@ -1403,13 +1422,14 @@ int tracecmd_write_cmdlines(struct tracecmd_output *handle) return ret; } -struct tracecmd_output *tracecmd_create_file_latency(const char *output_file, int cpus) +struct tracecmd_output *tracecmd_create_file_latency(const char *output_file, int cpus, + const char *compression) { struct tracecmd_output *handle; char *path; int ret; - handle = create_file(output_file, NULL, NULL, NULL, &all_event_list); + handle = create_file(output_file, NULL, NULL, NULL, &all_event_list, compression); if (!handle) return NULL; @@ -1710,41 +1730,43 @@ struct tracecmd_output *tracecmd_get_output_handle_fd(int fd) return NULL; } -struct tracecmd_output *tracecmd_create_init_fd(int fd) +struct tracecmd_output *tracecmd_create_init_fd(int fd, const char *compression) { - return create_file_fd(fd, NULL, NULL, NULL, &all_event_list, NULL); + return create_file_fd(fd, NULL, NULL, NULL, &all_event_list, NULL, compression); } struct tracecmd_output * tracecmd_create_init_fd_msg(struct tracecmd_msg_handle *msg_handle, - struct tracecmd_event_list *list) + struct tracecmd_event_list *list, const char *compression) { - return create_file_fd(msg_handle->fd, NULL, NULL, NULL, list, msg_handle); + return create_file_fd(msg_handle->fd, NULL, NULL, NULL, + list, msg_handle, compression); } struct tracecmd_output * -tracecmd_create_init_fd_glob(int fd, struct tracecmd_event_list *list) +tracecmd_create_init_fd_glob(int fd, struct tracecmd_event_list *list, const char *compression) { - return create_file_fd(fd, NULL, NULL, NULL, list, NULL); + return create_file_fd(fd, NULL, NULL, NULL, list, NULL, compression); } struct tracecmd_output * -tracecmd_create_init_file_glob(const char *output_file, - struct tracecmd_event_list *list) +tracecmd_create_init_file_glob(const char *output_file, struct tracecmd_event_list *list, + const char *compression) { - return create_file(output_file, NULL, NULL, NULL, list); + return create_file(output_file, NULL, NULL, NULL, list, compression); } -struct tracecmd_output *tracecmd_create_init_file(const char *output_file) +struct tracecmd_output *tracecmd_create_init_file(const char *output_file, const char *compression) { - return create_file(output_file, NULL, NULL, NULL, &all_event_list); + return create_file(output_file, NULL, NULL, NULL, &all_event_list, compression); } struct tracecmd_output *tracecmd_create_init_file_override(const char *output_file, const char *tracing_dir, - const char *kallsyms) + const char *kallsyms, + const char *compression) { - return create_file(output_file, NULL, tracing_dir, kallsyms, &all_event_list); + return create_file(output_file, NULL, tracing_dir, kallsyms, &all_event_list, compression); } /** @@ -1756,12 +1778,13 @@ struct tracecmd_output *tracecmd_create_init_file_override(const char *output_fi * with the same characteristics (events and all) and returns * tracecmd_output handle to this new file. */ -struct tracecmd_output *tracecmd_copy(struct tracecmd_input *ihandle, - const char *file) +struct tracecmd_output *tracecmd_copy(struct tracecmd_input *ihandle, const char *file) { struct tracecmd_output *handle; + const char *compr_name = NULL; - handle = create_file(file, ihandle, NULL, NULL, &all_event_list); + tracecmd_get_file_compress_proto(ihandle, &compr_name, NULL); + handle = create_file(file, ihandle, NULL, NULL, &all_event_list, compr_name); if (!handle) return NULL; diff --git a/tracecmd/trace-record.c b/tracecmd/trace-record.c index 60ee5fb3..29f84b7c 100644 --- a/tracecmd/trace-record.c +++ b/tracecmd/trace-record.c @@ -3700,7 +3700,7 @@ setup_connection(struct buffer_instance *instance, struct common_record_context /* Now create the handle through this socket */ if (msg_handle->version == V3_PROTOCOL) { - network_handle = tracecmd_create_init_fd_msg(msg_handle, listed_events); + network_handle = tracecmd_create_init_fd_msg(msg_handle, listed_events, NULL); if (!network_handle) goto error; tracecmd_set_quiet(network_handle, quiet); @@ -3718,8 +3718,7 @@ setup_connection(struct buffer_instance *instance, struct common_record_context if (ret) goto error; } else { - network_handle = tracecmd_create_init_fd_glob(msg_handle->fd, - listed_events); + network_handle = tracecmd_create_init_fd_glob(msg_handle->fd, listed_events, NULL); if (!network_handle) goto error; tracecmd_set_quiet(network_handle, quiet); @@ -4067,8 +4066,7 @@ static void setup_agent(struct buffer_instance *instance, { struct tracecmd_output *network_handle; - network_handle = tracecmd_create_init_fd_msg(instance->msg_handle, - listed_events); + network_handle = tracecmd_create_init_fd_msg(instance->msg_handle, listed_events, NULL); add_options(network_handle, ctx); tracecmd_write_cmdlines(network_handle); tracecmd_write_cpus(network_handle, instance->cpu_count); @@ -4460,7 +4458,7 @@ static void record_data(struct common_record_context *ctx) return; if (latency) { - handle = tracecmd_create_file_latency(ctx->output, local_cpu_count); + handle = tracecmd_create_file_latency(ctx->output, local_cpu_count, NULL); tracecmd_set_quiet(handle, quiet); } else { if (!local_cpu_count) @@ -4491,7 +4489,7 @@ static void record_data(struct common_record_context *ctx) touch_file(temp_files[i]); } - handle = tracecmd_create_init_file_glob(ctx->output, listed_events); + handle = tracecmd_create_init_file_glob(ctx->output, listed_events, NULL); if (!handle) die("Error creating output file"); tracecmd_set_quiet(handle, quiet); diff --git a/tracecmd/trace-restore.c b/tracecmd/trace-restore.c index 280a37f0..96b4fb5d 100644 --- a/tracecmd/trace-restore.c +++ b/tracecmd/trace-restore.c @@ -91,7 +91,7 @@ void trace_restore (int argc, char **argv) } handle = tracecmd_create_init_file_override(output, tracing_dir, - kallsyms); + kallsyms, NULL); if (!handle) die("Unabled to create output file %s", output); if (tracecmd_write_cmdlines(handle) < 0) @@ -128,7 +128,7 @@ void trace_restore (int argc, char **argv) handle = tracecmd_copy(ihandle, output); tracecmd_close(ihandle); } else - handle = tracecmd_create_init_file(output); + handle = tracecmd_create_init_file(output, NULL); if (!handle) die("error writing to %s", output); diff --git a/tracecmd/trace-stream.c b/tracecmd/trace-stream.c index b47b208c..317de1b2 100644 --- a/tracecmd/trace-stream.c +++ b/tracecmd/trace-stream.c @@ -43,7 +43,7 @@ trace_stream_init(struct buffer_instance *instance, int cpu, int fd, int cpus, tfd = fileno(fp); ofd = dup(tfd); - trace_output = tracecmd_create_init_fd(ofd); + trace_output = tracecmd_create_init_fd(ofd, NULL); if (!trace_output) { fclose(fp); return NULL; -- 2.31.1
Added a new parameter "trace-cmd record --compression", can be used to select the desired compression algorithm for the trace output file. One of these strings can be passed: "any" - auto select the best available compression algorithm "none" - do not compress the trace file name - the name of the desired compression algorithms, available algorithms can be listed with "trace-cmd list -c" If a compression is used, trace file version 7 is selected, otherwise trace file version 6 is used. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- tracecmd/trace-record.c | 30 +++++++++++++++++++++++++----- tracecmd/trace-usage.c | 5 +++++ 2 files changed, 30 insertions(+), 5 deletions(-) diff --git a/tracecmd/trace-record.c b/tracecmd/trace-record.c index 29f84b7c..d3362e5b 100644 --- a/tracecmd/trace-record.c +++ b/tracecmd/trace-record.c @@ -199,6 +199,7 @@ struct common_record_context { char *date2ts; char *user; const char *clock; + const char *compression; struct tsc_nsec tsc2nsec; int data_flags; int tsync_loop_interval; @@ -3700,7 +3701,8 @@ setup_connection(struct buffer_instance *instance, struct common_record_context /* Now create the handle through this socket */ if (msg_handle->version == V3_PROTOCOL) { - network_handle = tracecmd_create_init_fd_msg(msg_handle, listed_events, NULL); + network_handle = tracecmd_create_init_fd_msg(msg_handle, + listed_events, ctx->compression); if (!network_handle) goto error; tracecmd_set_quiet(network_handle, quiet); @@ -3718,7 +3720,8 @@ setup_connection(struct buffer_instance *instance, struct common_record_context if (ret) goto error; } else { - network_handle = tracecmd_create_init_fd_glob(msg_handle->fd, listed_events, NULL); + network_handle = tracecmd_create_init_fd_glob(msg_handle->fd, + listed_events, ctx->compression); if (!network_handle) goto error; tracecmd_set_quiet(network_handle, quiet); @@ -4066,7 +4069,8 @@ static void setup_agent(struct buffer_instance *instance, { struct tracecmd_output *network_handle; - network_handle = tracecmd_create_init_fd_msg(instance->msg_handle, listed_events, NULL); + network_handle = tracecmd_create_init_fd_msg(instance->msg_handle, + listed_events, ctx->compression); add_options(network_handle, ctx); tracecmd_write_cmdlines(network_handle); tracecmd_write_cpus(network_handle, instance->cpu_count); @@ -4458,7 +4462,8 @@ static void record_data(struct common_record_context *ctx) return; if (latency) { - handle = tracecmd_create_file_latency(ctx->output, local_cpu_count, NULL); + handle = tracecmd_create_file_latency(ctx->output, + local_cpu_count, ctx->compression); tracecmd_set_quiet(handle, quiet); } else { if (!local_cpu_count) @@ -4489,7 +4494,8 @@ static void record_data(struct common_record_context *ctx) touch_file(temp_files[i]); } - handle = tracecmd_create_init_file_glob(ctx->output, listed_events, NULL); + handle = tracecmd_create_init_file_glob(ctx->output, + listed_events, ctx->compression); if (!handle) die("Error creating output file"); tracecmd_set_quiet(handle, quiet); @@ -5736,6 +5742,7 @@ void init_top_instance(void) } enum { + OPT_comporession = 239, OPT_tsc2nsec = 240, OPT_fork = 241, OPT_tsyncinterval = 242, @@ -6172,6 +6179,7 @@ static void parse_record_options(int argc, {"fork", no_argument, NULL, OPT_fork}, {"tsc2nsec", no_argument, NULL, OPT_tsc2nsec}, {"poll", no_argument, NULL, OPT_poll}, + {"compression", required_argument, NULL, OPT_comporession}, {NULL, 0, NULL, 0} }; @@ -6597,6 +6605,18 @@ static void parse_record_options(int argc, cmd_check_die(ctx, CMD_set, *(argv+1), "--poll"); recorder_flags |= TRACECMD_RECORD_POLL; break; + case OPT_comporession: + cmd_check_die(ctx, CMD_start, *(argv+1), "--compression"); + cmd_check_die(ctx, CMD_set, *(argv+1), "--compression"); + cmd_check_die(ctx, CMD_extract, *(argv+1), "--compression"); + cmd_check_die(ctx, CMD_stream, *(argv+1), "--compression"); + cmd_check_die(ctx, CMD_profile, *(argv+1), "--compression"); + cmd_check_die(ctx, CMD_record_agent, *(argv+1), "--compression"); + if (strcmp(optarg, "any") && strcmp(optarg, "none") && + !tracecmd_compress_is_supported(optarg, NULL)) + die("Compression algorithm %s is not supported", optarg); + ctx->compression = strdup(optarg); + break; case OPT_quiet: case 'q': quiet = true; diff --git a/tracecmd/trace-usage.c b/tracecmd/trace-usage.c index cd8ca391..c70f9919 100644 --- a/tracecmd/trace-usage.c +++ b/tracecmd/trace-usage.c @@ -69,6 +69,11 @@ static struct usage_help usage_help[] = { " If 0 is specified, no loop is performed - timestamps offset is calculated only twice," " at the beginnig and at the end of the trace\n" " --poll don't block while reading from the trace buffer\n" + " --compression compress the trace output file, one of these strings can be passed:\n" + " any - auto select the best available compression algorithm\n" + " none - do not compress the trace file\n" + " name - the name of the desired compression algorithms\n" + " available algorithms can be listed with trace-cmd list -c\n" }, { "set", -- 2.31.1
Added support to read and dump information of trace files version 7. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- tracecmd/trace-dump.c | 125 +++++++++++++++++++++++++++++++++++------- 1 file changed, 105 insertions(+), 20 deletions(-) diff --git a/tracecmd/trace-dump.c b/tracecmd/trace-dump.c index 03cc82b4..ed82104b 100644 --- a/tracecmd/trace-dump.c +++ b/tracecmd/trace-dump.c @@ -26,6 +26,9 @@ static struct tep_handle *tep; static unsigned int trace_cpus; static int has_clock; +static unsigned long file_version; +static bool read_compress; +static struct tracecmd_compression *compress; enum dump_items { SUMMARY = (1 << 0), @@ -52,46 +55,62 @@ enum dump_items verbosity; tracecmd_plog(fmt, ##__VA_ARGS__); \ } while (0) -static int read_file_string(int fd, char *dst, int len) +static int read_fd(int fd, char *dst, int len) { size_t size = 0; int r; do { - r = read(fd, dst+size, 1); + r = read(fd, dst+size, len); if (r > 0) { - size++; - len--; + size += r; + len -= r; } else break; - if (!dst[size - 1]) - break; - } while (r > 0 && len); + } while (r > 0); - if (!size || dst[size - 1]) + if (len) return -1; - return 0; + return size; } -static int read_file_bytes(int fd, char *dst, int len) +static int read_compressed(int fd, char *dst, int len) +{ + + if (read_compress) + return tracecmd_compress_read(compress, dst, len); + return read_fd(fd, dst, len); +} + +static int read_file_string(int fd, char *dst, int len) { size_t size = 0; int r; do { - r = read(fd, dst+size, len); + r = read_compressed(fd, dst+size, 1); if (r > 0) { - size += r; - len -= r; + size++; + len--; } else break; - } while (r > 0); + if (!dst[size - 1]) + break; + } while (r > 0 && len); - if (len) + if (!size || dst[size - 1]) return -1; return 0; } +static int read_file_bytes(int fd, char *dst, int len) +{ + int ret; + + ret = read_compressed(fd, dst, len); + return ret < 0 ? ret : 0; +} + static void read_dump_string(int fd, int size, enum dump_items id) { char buf[DUMP_SIZE]; @@ -146,7 +165,6 @@ static void dump_initial_format(int fd) char magic[] = TRACECMD_MAGIC; char buf[DUMP_SIZE]; int val4; - unsigned long ver; do_print(SUMMARY, "\t[Initial format]\n"); @@ -168,11 +186,11 @@ static void dump_initial_format(int fd) die("no version string"); do_print(SUMMARY, "\t\t%s\t[Version]\n", buf); - ver = strtol(buf, NULL, 10); - if (!ver && errno) + file_version = strtol(buf, NULL, 10); + if (!file_version && errno) die("Invalid file version string %s", buf); - if (!tracecmd_is_version_supported(ver)) - die("Unsupported file version %lu", ver); + if (!tracecmd_is_version_supported(file_version)) + die("Unsupported file version %lu", file_version); /* get file endianness*/ if (read_file_bytes(fd, buf, 1)) @@ -192,6 +210,29 @@ static void dump_initial_format(int fd) do_print(SUMMARY, "\t\t%d\t[Page size, bytes]\n", val4); } +static void dump_compress(int fd) +{ + char compr[DUMP_SIZE]; + char *ver = NULL; + + if (file_version < 7) + return; + + /* get compression header */ + if (read_file_string(fd, compr, DUMP_SIZE)) + die("no compression header"); + ver = strchr(compr, ' '); + if (!ver) + die("no compression version"); + *ver = '\0'; + do_print((SUMMARY), "\t\t%s\t[Compression algorithm]\n", compr); + do_print((SUMMARY), "\t\t%s\t[Compression version]\n", ver + 1); + + compress = tracecmd_compress_alloc(compr, ver + 1, fd, tep, NULL); + if (!compress) + die("cannot uncomress the file"); +} + static void dump_header_page(int fd) { unsigned long long size; @@ -234,11 +275,34 @@ static void dump_header_event(int fd) read_dump_string(fd, size, HEAD_EVENT); } +static void uncompress_reset(void) +{ + if (compress && file_version >= 7) { + read_compress = false; + tracecmd_compress_reset(compress); + } +} + +static int uncompress_block(void) +{ + int ret = 0; + + if (compress && file_version >= 7) { + ret = tracecmd_uncompress_block(compress); + if (!ret) + read_compress = true; + + } + return ret; +} + static void dump_ftrace_events_format(int fd) { unsigned long long size; unsigned int count; + if (uncompress_block()) + die("cannot uncompress file block"); do_print((SUMMARY | FTRACE_FORMAT), "\t[Ftrace format, "); if (read_file_number(fd, &count, 4)) die("cannot read the count of the ftrace events"); @@ -251,6 +315,7 @@ static void dump_ftrace_events_format(int fd) read_dump_string(fd, size, FTRACE_FORMAT); count--; } + uncompress_reset(); } static void dump_events_format(int fd) @@ -262,6 +327,9 @@ static void dump_events_format(int fd) do_print((SUMMARY | EVENT_FORMAT | EVENT_SYSTEMS), "\t[Events format, "); + if (uncompress_block()) + die("cannot uncompress file block"); + if (read_file_number(fd, &systems, 4)) die("cannot read the count of the event systems"); @@ -284,6 +352,7 @@ static void dump_events_format(int fd) } systems--; } + uncompress_reset(); } static void dump_kallsyms(int fd) @@ -292,12 +361,17 @@ static void dump_kallsyms(int fd) do_print((SUMMARY | KALLSYMS), "\t[Kallsyms, "); + if (uncompress_block()) + die("cannot uncompress file block"); + if (read_file_number(fd, &size, 4)) die("cannot read the size of the kallsyms"); do_print((SUMMARY | KALLSYMS), "%d bytes]\n", size); read_dump_string(fd, size, KALLSYMS); + + uncompress_reset(); } static void dump_printk(int fd) @@ -306,12 +380,17 @@ static void dump_printk(int fd) do_print((SUMMARY | TRACE_PRINTK), "\t[Trace printk, "); + if (uncompress_block()) + die("cannot uncompress file block"); + if (read_file_number(fd, &size, 4)) die("cannot read the size of the trace printk"); do_print((SUMMARY | TRACE_PRINTK), "%d bytes]\n", size); read_dump_string(fd, size, TRACE_PRINTK); + + uncompress_reset(); } static void dump_cmdlines(int fd) @@ -320,12 +399,17 @@ static void dump_cmdlines(int fd) do_print((SUMMARY | CMDLINES), "\t[Saved command lines, "); + if (uncompress_block()) + die("cannot uncompress file block"); + if (read_file_number(fd, &size, 8)) die("cannot read the size of the saved command lines"); do_print((SUMMARY | CMDLINES), "%d bytes]\n", size); read_dump_string(fd, size, CMDLINES); + + uncompress_reset(); } static void dump_cpus_count(int fd) @@ -669,6 +753,7 @@ static void dump_file(const char *file) do_print(SUMMARY, "\n Tracing meta data in file %s:\n", file); dump_initial_format(fd); + dump_compress(fd); dump_header_page(fd); dump_header_event(fd); dump_ftrace_events_format(fd); -- 2.31.1
If libz is available, use that library to provide trace file compression support. The library is detected runtime. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/Makefile | 10 ++ lib/trace-cmd/include/trace-cmd-local.h | 5 + lib/trace-cmd/trace-compress-zlib.c | 172 ++++++++++++++++++++++++ lib/trace-cmd/trace-compress.c | 8 ++ 4 files changed, 195 insertions(+) create mode 100644 lib/trace-cmd/trace-compress-zlib.c diff --git a/lib/trace-cmd/Makefile b/lib/trace-cmd/Makefile index bab4322d..83ba7016 100644 --- a/lib/trace-cmd/Makefile +++ b/lib/trace-cmd/Makefile @@ -7,6 +7,13 @@ ldir:=$(src)/lib/trace-cmd DEFAULT_TARGET = $(LIBTRACECMD_STATIC) +pound := \# +ZLIB_INSTALLED := $(shell if (printf "$(pound)include <zlib.h>\n void main(){deflateInit(NULL, Z_BEST_COMPRESSION);}" | $(CC) -o /dev/null -x c - -lz >/dev/null 2>&1) ; then echo 1; else echo 0 ; fi) +ifeq ($(ZLIB_INSTALLED), 1) +CFLAGS += -DHAVE_ZLIB +$(info Have zlib compression support) +endif + OBJS = OBJS += trace-hash.o OBJS += trace-hooks.o @@ -26,6 +33,9 @@ OBJS += trace-timesync-ptp.o OBJS += trace-timesync-kvm.o endif OBJS += trace-compress.o +ifeq ($(ZLIB_INSTALLED), 1) +OBJS += trace-compress-zlib.o +endif # Additional util objects OBJS += trace-blk-hack.o diff --git a/lib/trace-cmd/include/trace-cmd-local.h b/lib/trace-cmd/include/trace-cmd-local.h index 419bfee0..40a6a40d 100644 --- a/lib/trace-cmd/include/trace-cmd-local.h +++ b/lib/trace-cmd/include/trace-cmd-local.h @@ -31,6 +31,11 @@ void tracecmd_info(const char *fmt, ...); #endif #endif +#ifdef HAVE_ZLIB +int tracecmd_zlib_init(void); +void tracecmd_zlib_free(void); +#endif + void tracecmd_compress_init(void); void tracecmd_compress_free(void); diff --git a/lib/trace-cmd/trace-compress-zlib.c b/lib/trace-cmd/trace-compress-zlib.c new file mode 100644 index 00000000..3208d57b --- /dev/null +++ b/lib/trace-cmd/trace-compress-zlib.c @@ -0,0 +1,172 @@ +// SPDX-License-Identifier: LGPL-2.1 +/* + * Copyright (C) 2021, VMware, Tzvetomir Stoyanov tz.stoyanov@gmail.com> + * + */ +#include <stdlib.h> +#include <dlfcn.h> +#include <zlib.h> +#include <errno.h> + +#include "trace-cmd-private.h" + +#define __ZLIB_NAME "zlib" +#define __ZLIB_WEIGTH 10 +#define __ZLIB_FILE "libz.so" +#define ZLIB_FUNC_COMPRESS "compress2" +#define ZLIB_FUNC_DECOMOPRESS "uncompress" +#define ZLIB_FUNC_SIZE "compressBound" +#define ZLIB_FUNC_VERSION "zlibVersion" + +static void *zlib_handle; +static int (*_lib_compress)(unsigned char *out, unsigned long *out_bytes, + unsigned char *in, unsigned long in_bytes, int level); +static int (*_libz_decompress)(unsigned char *out, unsigned long *out_bytes, + unsigned char *in, unsigned long in_bytes); +static unsigned long (*_libz_compress_bound)(unsigned long in_bytes); +static const char *(*_libz_version)(void); + +static int zlib_compress(char *in, unsigned int in_bytes, + char *out, unsigned int *out_bytes) +{ + unsigned long out_size = *out_bytes; + int ret; + + if (!_lib_compress) + return -1; + + ret = _lib_compress((unsigned char *)out, &out_size, + (unsigned char *)in, (unsigned long)in_bytes, Z_BEST_COMPRESSION); + *out_bytes = out_size; + errno = 0; + switch (ret) { + case Z_OK: + return 0; + case Z_BUF_ERROR: + errno = -ENOBUFS; + break; + case Z_MEM_ERROR: + errno = -ENOMEM; + break; + case Z_STREAM_ERROR: + errno = -EINVAL; + break; + default: + errno = -EFAULT; + break; + } + + return -1; +} + +static int zlib_decompress(char *in, unsigned int in_bytes, + char *out, unsigned int *out_bytes) +{ + unsigned long out_size = *out_bytes; + int ret; + + if (!_libz_decompress) + return -1; + + ret = _libz_decompress((unsigned char *)out, &out_size, + (unsigned char *)in, (unsigned long)in_bytes); + *out_bytes = out_size; + errno = 0; + switch (ret) { + case Z_OK: + return 0; + case Z_BUF_ERROR: + errno = -ENOBUFS; + break; + case Z_MEM_ERROR: + errno = -ENOMEM; + break; + case Z_DATA_ERROR: + errno = -EINVAL; + break; + default: + errno = -EFAULT; + break; + } + + return -1; +} + +static unsigned int zlib_compress_bound(unsigned int in_bytes) +{ + if (!_libz_compress_bound) + return 0; + return _libz_compress_bound(in_bytes); +} + +static bool zlib_is_supported(const char *name, const char *version) +{ + const char *zver; + + if (!name) + return false; + if (strlen(name) != strlen(__ZLIB_NAME) || strcmp(name, __ZLIB_NAME)) + return false; + + if (!version) + return true; + + if (!_libz_version) + return false; + zver = _libz_version(); + if (!zver) + return false; + + /* Compare the major version number */ + if (atoi(version) <= atoi(zver)) + return true; + + return false; +} + +int tracecmd_zlib_init(void) +{ + if (zlib_handle) + return 0; + + zlib_handle = dlopen(__ZLIB_FILE, RTLD_NOW | RTLD_GLOBAL); + if (!zlib_handle) + return -1; + _lib_compress = dlsym(zlib_handle, ZLIB_FUNC_COMPRESS); + if (!_lib_compress) + goto error; + _libz_decompress = dlsym(zlib_handle, ZLIB_FUNC_DECOMOPRESS); + if (!_libz_decompress) + goto error; + _libz_compress_bound = dlsym(zlib_handle, ZLIB_FUNC_SIZE); + if (!_libz_compress_bound) + goto error; + _libz_version = dlsym(zlib_handle, ZLIB_FUNC_VERSION); + if (!_libz_version) + goto error; + + return tracecmd_compress_proto_register(__ZLIB_NAME, _libz_version(), __ZLIB_WEIGTH, + zlib_compress, zlib_decompress, + zlib_compress_bound, zlib_is_supported); + +error: + _lib_compress = NULL; + _libz_decompress = NULL; + _libz_version = NULL; + dlclose(zlib_handle); + zlib_handle = NULL; + return -1; +} + +void tracecmd_zlib_free(void) +{ + _lib_compress = NULL; + _libz_decompress = NULL; + _libz_version = NULL; + + if (zlib_handle) { + dlclose(zlib_handle); + zlib_handle = NULL; + } + +} diff --git a/lib/trace-cmd/trace-compress.c b/lib/trace-cmd/trace-compress.c index 378a3e57..039f60c5 100644 --- a/lib/trace-cmd/trace-compress.c +++ b/lib/trace-cmd/trace-compress.c @@ -355,6 +355,10 @@ void tracecmd_compress_init(void) gettimeofday(&time, NULL); srand((time.tv_sec * 1000) + (time.tv_usec / 1000)); + +#ifdef HAVE_ZLIB + tracecmd_zlib_init(); +#endif } static struct compress_proto *compress_proto_select(void) @@ -531,6 +535,10 @@ void tracecmd_compress_free(void) struct compress_proto *proto = proto_list; struct compress_proto *del; +#ifdef HAVE_ZLIB + tracecmd_zlib_free(); +#endif + while (proto) { del = proto; proto = proto->next; -- 2.31.1
When a trace buffer data are written in the trace file, the buffer option in the file metadata is updated with the file offset of the tracing data. Hide this logic into the trace-cmd library. Added new APIs: tracecmd_add_buffer_description() tracecmd_write_buffers_description() Changed APIs: tracecmd_append_buffer_cpu_data() Removed APIs: tracecmd_add_buffer_option() This internal refactoring is needed for changes, related to compression of the options sections of the trace file. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- .../include/private/trace-cmd-private.h | 8 +- lib/trace-cmd/trace-output.c | 84 +++++++++++++++++-- tracecmd/trace-record.c | 16 ++-- 3 files changed, 85 insertions(+), 23 deletions(-) diff --git a/lib/trace-cmd/include/private/trace-cmd-private.h b/lib/trace-cmd/include/private/trace-cmd-private.h index ee73325c..cbb578ec 100644 --- a/lib/trace-cmd/include/private/trace-cmd-private.h +++ b/lib/trace-cmd/include/private/trace-cmd-private.h @@ -295,8 +295,8 @@ struct tracecmd_option * tracecmd_add_option_v(struct tracecmd_output *handle, unsigned short id, const struct iovec *vector, int count); -struct tracecmd_option *tracecmd_add_buffer_option(struct tracecmd_output *handle, - const char *name, int cpus); +int tracecmd_add_buffer_description(struct tracecmd_output *handle, const char *name, int cpus); +int tracecmd_write_buffers_description(struct tracecmd_output *handle); int tracecmd_write_cpus(struct tracecmd_output *handle, int cpus); int tracecmd_write_cmdlines(struct tracecmd_output *handle); @@ -312,9 +312,7 @@ int tracecmd_write_cpu_data(struct tracecmd_output *handle, int tracecmd_append_cpu_data(struct tracecmd_output *handle, int cpus, char * const *cpu_data_files); int tracecmd_append_buffer_cpu_data(struct tracecmd_output *handle, - struct tracecmd_option *option, - int cpus, char * const *cpu_data_files); - + const char *name, int cpus, char * const *cpu_data_files); struct tracecmd_output *tracecmd_get_output_handle_fd(int fd); /* --- Reading the Fly Recorder Trace --- */ diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index 7c7d3d76..8f8ca164 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -44,6 +44,14 @@ struct tracecmd_option { struct list_head list; }; +struct tracecmd_buffer { + int cpus; + void *name; + tsize_t offset; + struct tracecmd_option *option; + struct list_head list; +}; + enum { OUTPUT_FL_SEND_META = (1 << 0), }; @@ -63,6 +71,7 @@ struct tracecmd_output { struct tracecmd_compression *compress; struct list_head options; + struct list_head buffers; struct tracecmd_msg_handle *msg_handle; char *trace_clock; }; @@ -189,6 +198,7 @@ bool tracecmd_get_quiet(struct tracecmd_output *handle) void tracecmd_output_free(struct tracecmd_output *handle) { struct tracecmd_option *option; + struct tracecmd_buffer *buffer; if (!handle) return; @@ -199,6 +209,13 @@ void tracecmd_output_free(struct tracecmd_output *handle) if (handle->pevent) tep_unref(handle->pevent); + while (!list_empty(&handle->buffers)) { + buffer = container_of(handle->buffers.next, + struct tracecmd_buffer, list); + list_del(&buffer->list); + free(buffer->name); + free(buffer); + } while (!list_empty(&handle->options)) { option = container_of(handle->options.next, struct tracecmd_option, list); @@ -1071,6 +1088,7 @@ create_file_fd(int fd, struct tracecmd_input *ihandle, goto out_free; list_head_init(&handle->options); + list_head_init(&handle->buffers); buf[0] = 23; buf[1] = 8; @@ -1369,9 +1387,8 @@ int tracecmd_append_options(struct tracecmd_output *handle) return 0; } -struct tracecmd_option * -tracecmd_add_buffer_option(struct tracecmd_output *handle, const char *name, - int cpus) +static struct tracecmd_option * +add_buffer_option(struct tracecmd_output *handle, const char *name, int cpus) { struct tracecmd_option *option; char *buf; @@ -1399,6 +1416,53 @@ tracecmd_add_buffer_option(struct tracecmd_output *handle, const char *name, return option; } +int tracecmd_add_buffer_description(struct tracecmd_output *handle, const char *name, int cpus) +{ + struct tracecmd_buffer *buf; + + buf = calloc(1, sizeof(struct tracecmd_buffer)); + if (!buf) + return -1; + buf->name = strdup(name); + buf->cpus = cpus; + if (!buf->name) { + free(buf); + return -1; + } + list_add_tail(&buf->list, &handle->buffers); + return 0; +} + +int tracecmd_write_buffers_description(struct tracecmd_output *handle) +{ + struct tracecmd_option *option; + struct tracecmd_buffer *buf; + + list_for_each_entry(buf, &handle->buffers, list) { + option = add_buffer_option(handle, buf->name, buf->cpus); + if (!option) + return -1; + buf->option = option; + } + return 0; +} + +static tsize_t get_buffer_file_offset(struct tracecmd_output *handle, const char *name) +{ + struct tracecmd_buffer *buf; + + list_for_each_entry(buf, &handle->buffers, list) { + if (strlen(name) == strlen(buf->name) && !strcmp(name, buf->name)) { + if (handle->file_version >= 7) + return buf->offset; + if (!buf->option) + break; + return buf->option->offset; + } + } + return 0; +} + int tracecmd_write_cmdlines(struct tracecmd_output *handle) { int ret; @@ -1643,18 +1707,23 @@ int tracecmd_append_cpu_data(struct tracecmd_output *handle, } int tracecmd_append_buffer_cpu_data(struct tracecmd_output *handle, - struct tracecmd_option *option, - int cpus, char * const *cpu_data_files) + const char *name, int cpus, char * const *cpu_data_files) { + tsize_t b_offset; tsize_t offset; stsize_t ret; + b_offset = get_buffer_file_offset(handle, name); + if (!b_offset) { + tracecmd_warning("Cannot find description for buffer %s\n", name); + return -1; + } offset = lseek64(handle->fd, 0, SEEK_CUR); /* Go to the option data, where will write the offest */ - ret = lseek64(handle->fd, option->offset, SEEK_SET); + ret = lseek64(handle->fd, b_offset, SEEK_SET); if (ret == (off64_t)-1) { - tracecmd_warning("could not seek to %lld\n", option->offset); + tracecmd_warning("could not seek to %lld\n", b_offset); return -1; } @@ -1713,6 +1782,7 @@ struct tracecmd_output *tracecmd_get_output_handle_fd(int fd) handle->page_size = tracecmd_page_size(ihandle); handle->file_version = tracecmd_get_in_file_version(ihandle); list_head_init(&handle->options); + list_head_init(&handle->buffers); if (!tracecmd_get_file_compress_proto(ihandle, &cname, &cver)) { handle->compress = tracecmd_compress_alloc(cname, cver, handle->fd, diff --git a/tracecmd/trace-record.c b/tracecmd/trace-record.c index d3362e5b..eff6f2f0 100644 --- a/tracecmd/trace-record.c +++ b/tracecmd/trace-record.c @@ -4152,7 +4152,6 @@ static void touch_file(const char *file) } static void append_buffer(struct tracecmd_output *handle, - struct tracecmd_option *buffer_option, struct buffer_instance *instance, char **temp_files) { @@ -4180,7 +4179,7 @@ static void append_buffer(struct tracecmd_output *handle, touch_file(temp_files[i]); } - tracecmd_append_buffer_cpu_data(handle, buffer_option, + tracecmd_append_buffer_cpu_data(handle, tracefs_instance_get_name(instance->tracefs), cpu_count, temp_files); for (i = 0; i < instance->cpu_count; i++) { @@ -4441,7 +4440,6 @@ static void write_guest_file(struct buffer_instance *instance) static void record_data(struct common_record_context *ctx) { - struct tracecmd_option **buffer_options; struct tracecmd_output *handle; struct buffer_instance *instance; bool local = false; @@ -4512,9 +4510,6 @@ static void record_data(struct common_record_context *ctx) } if (buffers) { - buffer_options = malloc(sizeof(*buffer_options) * buffers); - if (!buffer_options) - die("Failed to allocate buffer options"); i = 0; for_each_instance(instance) { int cpus = instance->cpu_count != local_cpu_count ? @@ -4522,10 +4517,9 @@ static void record_data(struct common_record_context *ctx) if (instance->msg_handle) continue; - - buffer_options[i++] = tracecmd_add_buffer_option(handle, - tracefs_instance_get_name(instance->tracefs), - cpus); + tracecmd_add_buffer_description(handle, + tracefs_instance_get_name(instance->tracefs), + cpus); add_buffer_stat(handle, instance); } } @@ -4560,7 +4554,7 @@ static void record_data(struct common_record_context *ctx) if (instance->msg_handle) continue; print_stat(instance); - append_buffer(handle, buffer_options[i++], instance, temp_files); + append_buffer(handle, instance, temp_files); } } -- 2.31.1
When a new buffer data are written in the trace file, the file offset is stored in the buffer description. As the buffer description is part of the trace options, which are already written in the file, the offset field in this option is updated with the new value. This causes a serious problem when option section of the file is compressed. Updating a value in already compressed section may change the size of this section, which can make the calculated offset of the buffer trace data invalid. To solve that problem and to enable the compression of the option section of the file, the structure of the trace file is changed: - In trace file v7, the description of the buffers is moved to a separate section in the trace file medatata. It is stored after the CPU count section and before the option section. The format of this new buffers description is: 4 bytes - count of the buffers, stored in the file. The top buffer is not counted. For each buffer: 8 bytes, offset of the trace data for this buffer in the file. null-terminated string, name of the buffer. The new section is mandatory, in case of no buffers - the count 0 is stored in the first 4 bytes. - In trace file v6, the description of the buffers is not changed. These changes are needed for compression of the options sections from the trace file. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- .../include/private/trace-cmd-private.h | 1 + lib/trace-cmd/include/trace-cmd-local.h | 3 + lib/trace-cmd/trace-input.c | 40 +++++++++ lib/trace-cmd/trace-output.c | 85 ++++++++++++++++++- tracecmd/trace-dump.c | 24 ++++++ tracecmd/trace-listen.c | 3 + tracecmd/trace-record.c | 4 + 7 files changed, 156 insertions(+), 4 deletions(-) diff --git a/lib/trace-cmd/include/private/trace-cmd-private.h b/lib/trace-cmd/include/private/trace-cmd-private.h index cbb578ec..faa8c533 100644 --- a/lib/trace-cmd/include/private/trace-cmd-private.h +++ b/lib/trace-cmd/include/private/trace-cmd-private.h @@ -112,6 +112,7 @@ enum tracecmd_file_states { TRACECMD_FILE_PRINTK, TRACECMD_FILE_CMD_LINES, TRACECMD_FILE_CPU_COUNT, + TRACECMD_FILE_BUFERS, TRACECMD_FILE_OPTIONS, TRACECMD_FILE_CPU_LATENCY, TRACECMD_FILE_CPU_FLYRECORD, diff --git a/lib/trace-cmd/include/trace-cmd-local.h b/lib/trace-cmd/include/trace-cmd-local.h index 40a6a40d..2f421103 100644 --- a/lib/trace-cmd/include/trace-cmd-local.h +++ b/lib/trace-cmd/include/trace-cmd-local.h @@ -46,4 +46,7 @@ void out_compression_reset(struct tracecmd_output *handle); void in_uncompress_reset(struct tracecmd_input *handle); int in_uncompress_block(struct tracecmd_input *handle); +int write_buffers_description_v7(struct tracecmd_output *handle); +int write_buffers_description_v6(struct tracecmd_output *handle); + #endif /* _TRACE_CMD_LOCAL_H */ diff --git a/lib/trace-cmd/trace-input.c b/lib/trace-cmd/trace-input.c index 0b17b58d..e953909f 100644 --- a/lib/trace-cmd/trace-input.c +++ b/lib/trace-cmd/trace-input.c @@ -881,6 +881,39 @@ static int read_cpus(struct tracecmd_input *handle) return 0; } +static int read_buffers(struct tracecmd_input *handle) +{ + struct input_buffer_instance *buffer; + unsigned long long offset; + unsigned int count; + int i; + + if (handle->file_version < 7 || handle->file_state >= TRACECMD_FILE_BUFERS) + return 0; + + if (read4(handle, &count) < 0) + return -1; + i = handle->nr_buffers; + handle->nr_buffers += count; + handle->buffers = realloc(handle->buffers, + sizeof(*handle->buffers) * handle->nr_buffers); + if (!handle->buffers) + return -1; + + for (; i < handle->nr_buffers; i++) { + buffer = &handle->buffers[i]; + if (read8(handle, &offset) < 0) + return -1; + buffer->offset = offset; + buffer->name = read_string(handle); + if (!buffer->name) + return -1; + } + + handle->file_state = TRACECMD_FILE_BUFERS; + return 0; +} + /** * tracecmd_read_headers - read the header information from trace.dat * @handle: input handle for the trace.dat file @@ -955,6 +988,13 @@ int tracecmd_read_headers(struct tracecmd_input *handle, if (state <= handle->file_state) return 0; + if (read_buffers(handle) < 0) + return -1; + + if (state <= handle->file_state) + return 0; + + if (read_options_type(handle) < 0) return -1; diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index 8f8ca164..7830f643 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -376,6 +376,17 @@ static int check_out_state(struct tracecmd_output *handle, int new_state) if (!handle) return -1; + if (handle->file_version < 7) { + switch (new_state) { + case TRACECMD_FILE_BUFERS: + tracecmd_warning("State %d is not supported in trace file version %d", + new_state, handle->file_version); + return -1; + default: + break; + } + } + switch (new_state) { case TRACECMD_FILE_HEADERS: case TRACECMD_FILE_FTRACE_EVENTS: @@ -384,10 +395,16 @@ static int check_out_state(struct tracecmd_output *handle, int new_state) case TRACECMD_FILE_PRINTK: case TRACECMD_FILE_CMD_LINES: case TRACECMD_FILE_CPU_COUNT: - case TRACECMD_FILE_OPTIONS: + case TRACECMD_FILE_BUFERS: if (handle->file_state == (new_state - 1)) return 0; break; + case TRACECMD_FILE_OPTIONS: + if (handle->file_version < 7 && handle->file_state == TRACECMD_FILE_CPU_COUNT) + return 0; + if (handle->file_version >= 7 && handle->file_state == TRACECMD_FILE_BUFERS) + return 0; + break; case TRACECMD_FILE_CPU_LATENCY: case TRACECMD_FILE_CPU_FLYRECORD: if (handle->file_state == TRACECMD_FILE_OPTIONS) @@ -1433,7 +1450,7 @@ int tracecmd_add_buffer_description(struct tracecmd_output *handle, const char * return 0; } -int tracecmd_write_buffers_description(struct tracecmd_output *handle) +__hidden int write_buffers_description_v6(struct tracecmd_output *handle) { struct tracecmd_option *option; struct tracecmd_buffer *buf; @@ -1447,6 +1464,63 @@ int tracecmd_write_buffers_description(struct tracecmd_output *handle) return 0; } +__hidden int write_buffers_description_v7(struct tracecmd_output *handle) +{ + struct tracecmd_buffer *buf; + char *data = NULL; + int count = 0; + int size; + int ret; + + if (!check_out_state(handle, TRACECMD_FILE_BUFERS)) { + tracecmd_warning("Cannot write buffers descriptions"); + return -1; + } + + list_for_each_entry(buf, &handle->buffers, list) { + count++; + } + + count = convert_endian_4(handle, count); + ret = do_write_check(handle, &count, 4); + if (ret < 0) + return ret; + + list_for_each_entry(buf, &handle->buffers, list) { + size = 8 + strlen(buf->name) + 1; + data = malloc(size); + if (!data) + goto error; + *(tsize_t *)data = 0; + strcpy(data + 8, buf->name); + /* Save the data location in case it needs to be updated */ + buf->offset = do_lseek(handle, 0, SEEK_CUR); + if (do_write_check(handle, data, size)) + goto error; + + /* + * In case a buffer instance has different number of CPUs as the + * local machine. + */ + if (buf->cpus) + tracecmd_add_option(handle, TRACECMD_OPTION_CPUCOUNT, + sizeof(int), &buf->cpus); + } + + handle->file_state = TRACECMD_FILE_BUFERS; + return 0; +error: + return -1; +} + +int tracecmd_write_buffers_description(struct tracecmd_output *handle) +{ + if (handle->file_version >= 7) + return write_buffers_description_v7(handle); + + return write_buffers_description_v6(handle); +} + static tsize_t get_buffer_file_offset(struct tracecmd_output *handle, const char *name) { struct tracecmd_buffer *buf; @@ -1505,7 +1579,8 @@ struct tracecmd_output *tracecmd_create_file_latency(const char *output_file, in if (tracecmd_write_cpus(handle, cpus) < 0) goto out_free; - + if (tracecmd_write_buffers_description(handle) < 0) + goto out_free; if (tracecmd_write_options(handle) < 0) goto out_free; @@ -1698,7 +1773,9 @@ int tracecmd_append_cpu_data(struct tracecmd_output *handle, ret = tracecmd_write_cpus(handle, cpus); if (ret) return ret; - + ret = tracecmd_write_buffers_description(handle); + if (ret) + return ret; ret = tracecmd_write_options(handle); if (ret) return ret; diff --git a/tracecmd/trace-dump.c b/tracecmd/trace-dump.c index ed82104b..56acb01d 100644 --- a/tracecmd/trace-dump.c +++ b/tracecmd/trace-dump.c @@ -420,6 +420,29 @@ static void dump_cpus_count(int fd) do_print(SUMMARY, "\t%d [CPUs with tracing data]\n", trace_cpus); } +static void dump_buffers(int fd) +{ + unsigned long long offset; + char name[DUMP_SIZE]; + int count; + + if (file_version < 7) + return; + + if (read_file_number(fd, &count, 4)) + die("cannot read the buffers count"); + do_print(SUMMARY, "\t%d [buffers with tracing data]\n", count); + + while (count) { + if (read_file_number(fd, &offset, 8)) + die("cannot read the buffer offset"); + if (read_file_string(fd, name, DUMP_SIZE)) + die("cannot read the buffer name"); + do_print(SUMMARY, "\t\t 0x%llx %s[data offset, buffer name]\n", offset, name); + count--; + } +} + static void dump_option_string(int fd, int size, char *desc) { do_print(OPTIONS, "\t\t[Option %s, %d bytes]\n", desc, size); @@ -762,6 +785,7 @@ static void dump_file(const char *file) dump_printk(fd); dump_cmdlines(fd); dump_cpus_count(fd); + dump_buffers(fd); dump_therest(fd); tep_free(tep); diff --git a/tracecmd/trace-listen.c b/tracecmd/trace-listen.c index 0ae1c948..62debdb6 100644 --- a/tracecmd/trace-listen.c +++ b/tracecmd/trace-listen.c @@ -604,6 +604,9 @@ static int put_together_file(int cpus, int ofd, const char *node, if (write_options) { ret = tracecmd_write_cpus(handle, cpus); + if (ret) + goto out; + ret = tracecmd_write_buffers_description(handle); if (ret) goto out; ret = tracecmd_write_options(handle); diff --git a/tracecmd/trace-record.c b/tracecmd/trace-record.c index eff6f2f0..2a379d09 100644 --- a/tracecmd/trace-record.c +++ b/tracecmd/trace-record.c @@ -3711,6 +3711,9 @@ setup_connection(struct buffer_instance *instance, struct common_record_context if (ret) goto error; ret = tracecmd_write_cpus(network_handle, instance->cpu_count); + if (ret) + goto error; + ret = tracecmd_write_buffers_description(network_handle); if (ret) goto error; ret = tracecmd_write_options(network_handle); @@ -4074,6 +4077,7 @@ static void setup_agent(struct buffer_instance *instance, add_options(network_handle, ctx); tracecmd_write_cmdlines(network_handle); tracecmd_write_cpus(network_handle, instance->cpu_count); + tracecmd_write_buffers_description(network_handle); tracecmd_write_options(network_handle); tracecmd_msg_finish_sending_data(instance->msg_handle); instance->network_handle = network_handle; -- 2.31.1
When reading / writing trace file, store the offset of the option section in the file. A new internal API is added, to get the offset of the option section from an input file handler: tracecmd_get_options_offset() These changes are needed for compression of the option section from the trace file. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/include/private/trace-cmd-private.h | 1 + lib/trace-cmd/trace-input.c | 11 +++++++++++ lib/trace-cmd/trace-output.c | 2 ++ 3 files changed, 14 insertions(+) diff --git a/lib/trace-cmd/include/private/trace-cmd-private.h b/lib/trace-cmd/include/private/trace-cmd-private.h index faa8c533..10089389 100644 --- a/lib/trace-cmd/include/private/trace-cmd-private.h +++ b/lib/trace-cmd/include/private/trace-cmd-private.h @@ -239,6 +239,7 @@ tracecmd_get_cursor(struct tracecmd_input *handle, int cpu); unsigned long tracecmd_get_in_file_version(struct tracecmd_input *handle); int tracecmd_get_file_compress_proto(struct tracecmd_input *handle, const char **name, const char **version); +size_t tracecmd_get_options_offset(struct tracecmd_input *handle); int tracecmd_ftrace_overrides(struct tracecmd_input *handle, struct tracecmd_ftrace *finfo); bool tracecmd_get_use_trace_clock(struct tracecmd_input *handle); diff --git a/lib/trace-cmd/trace-input.c b/lib/trace-cmd/trace-input.c index e953909f..b76f2b47 100644 --- a/lib/trace-cmd/trace-input.c +++ b/lib/trace-cmd/trace-input.c @@ -157,6 +157,7 @@ struct tracecmd_input { size_t header_files_start; size_t ftrace_files_start; size_t event_files_start; + size_t options_start; size_t total_file_size; /* For custom profilers. */ @@ -2759,6 +2760,7 @@ static int handle_options(struct tracecmd_input *handle) /* By default, use usecs, unless told otherwise */ handle->flags |= TRACECMD_FL_IN_USECS; + handle->options_start = lseek64(handle->fd, 0, SEEK_CUR); for (;;) { if (do_read_check(handle, &option, 2)) @@ -4213,6 +4215,15 @@ bool tracecmd_get_use_trace_clock(struct tracecmd_input *handle) return handle->use_trace_clock; } +/** + * tracecmd_get_options_offset - get offset of the options sections in the file + * @handle: input handle for the trace.dat file + */ +size_t tracecmd_get_options_offset(struct tracecmd_input *handle) +{ + return handle->options_start; +} + /** * tracecmd_get_show_data_func - return the show data func * @handle: input handle for the trace.dat file diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index 7830f643..bad234c4 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -66,6 +66,7 @@ struct tracecmd_output { bool quiet; unsigned long file_state; unsigned long file_version; + size_t options_start; bool do_compress; struct tracecmd_compression *compress; @@ -1858,6 +1859,7 @@ struct tracecmd_output *tracecmd_get_output_handle_fd(int fd) tep_ref(handle->pevent); handle->page_size = tracecmd_page_size(ihandle); handle->file_version = tracecmd_get_in_file_version(ihandle); + handle->options_start = tracecmd_get_options_offset(ihandle); list_head_init(&handle->options); list_head_init(&handle->buffers); -- 2.31.1
Comperss the option section of the trace file. This section is not big currently and compressing it does not reduce significantly the size of the file. This could be useful in the future as new options can be added, storing a potentially huge amount of data. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/include/trace-cmd-local.h | 1 + lib/trace-cmd/trace-input.c | 46 +++++++++++------ lib/trace-cmd/trace-output.c | 69 +++++++++++++++++-------- tracecmd/trace-dump.c | 16 ++++-- 4 files changed, 92 insertions(+), 40 deletions(-) diff --git a/lib/trace-cmd/include/trace-cmd-local.h b/lib/trace-cmd/include/trace-cmd-local.h index 2f421103..03d2a9a3 100644 --- a/lib/trace-cmd/include/trace-cmd-local.h +++ b/lib/trace-cmd/include/trace-cmd-local.h @@ -39,6 +39,7 @@ void tracecmd_zlib_free(void); void tracecmd_compress_init(void); void tracecmd_compress_free(void); +int out_uncompress_block(struct tracecmd_output *handle); int out_compression_start(struct tracecmd_output *handle); int out_compression_end(struct tracecmd_output *handle); void out_compression_reset(struct tracecmd_output *handle); diff --git a/lib/trace-cmd/trace-input.c b/lib/trace-cmd/trace-input.c index b76f2b47..8fff003e 100644 --- a/lib/trace-cmd/trace-input.c +++ b/lib/trace-cmd/trace-input.c @@ -2761,23 +2761,29 @@ static int handle_options(struct tracecmd_input *handle) /* By default, use usecs, unless told otherwise */ handle->flags |= TRACECMD_FL_IN_USECS; handle->options_start = lseek64(handle->fd, 0, SEEK_CUR); - + if (in_uncompress_block(handle)) + return -1; for (;;) { - if (do_read_check(handle, &option, 2)) - return -1; + ret = do_read_check(handle, &option, 2); + if (ret) + goto out; if (option == TRACECMD_OPTION_DONE) break; /* next 4 bytes is the size of the option */ - if (do_read_check(handle, &size, 4)) - return -1; + ret = do_read_check(handle, &size, 4); + if (ret) + goto out; size = tep_read_number(handle->pevent, &size, 4); buf = malloc(size); - if (!buf) - return -ENOMEM; - if (do_read_check(handle, buf, size)) - return -1; + if (!buf) { + ret = -ENOMEM; + goto out; + } + ret = do_read_check(handle, buf, size); + if (ret) + goto out; switch (option) { case TRACECMD_OPTION_DATE: @@ -2827,14 +2833,16 @@ static int handle_options(struct tracecmd_input *handle) buf + 8, 4); ret = tsync_cpu_offsets_load(handle, buf + 12, size - 12); if (ret < 0) - return ret; + goto out; tracecmd_enable_tsync(handle, true); break; case TRACECMD_OPTION_CPUSTAT: buf[size-1] = '\n'; cpustats = realloc(cpustats, cpustats_size + size + 1); - if (!cpustats) - return -ENOMEM; + if (!cpustats) { + ret = -ENOMEM; + goto out; + } memcpy(cpustats + cpustats_size, buf, size); cpustats_size += size; cpustats[cpustats_size] = 0; @@ -2844,14 +2852,17 @@ static int handle_options(struct tracecmd_input *handle) handle->nr_buffers++; handle->buffers = realloc(handle->buffers, sizeof(*handle->buffers) * handle->nr_buffers); - if (!handle->buffers) - return -ENOMEM; + if (!handle->buffers) { + ret = -ENOMEM; + goto out; + } buffer = &handle->buffers[handle->nr_buffers - 1]; buffer->name = strdup(buf + 8); if (!buffer->name) { free(handle->buffers); handle->buffers = NULL; - return -ENOMEM; + ret = -ENOMEM; + goto out; } offset = *(unsigned long long *)buf; buffer->offset = tep_read_number(handle->pevent, &offset, 8); @@ -2908,8 +2919,11 @@ static int handle_options(struct tracecmd_input *handle) } handle->cpustats = cpustats; + ret = 0; - return 0; +out: + in_uncompress_reset(handle); + return ret; } static int read_options_type(struct tracecmd_input *handle) diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index bad234c4..f7e6c3dd 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -150,6 +150,18 @@ __hidden void out_compression_reset(struct tracecmd_output *handle) handle->do_compress = false; } +__hidden int out_uncompress_block(struct tracecmd_output *handle) +{ + int ret = 0; + + if (handle->file_version < 7) + return 0; + ret = tracecmd_uncompress_block(handle->compress); + if (!ret) + handle->do_compress = true; + return ret; +} + __hidden int out_compression_start(struct tracecmd_output *handle) { if (handle->file_version < 7 || !handle->compress) @@ -1326,32 +1338,38 @@ int tracecmd_write_options(struct tracecmd_output *handle) if (do_write_check(handle, "options ", 10)) return -1; - + handle->options_start = lseek64(handle->fd, 0, SEEK_CUR); + out_compression_start(handle); list_for_each_entry(options, &handle->options, list) { endian2 = convert_endian_2(handle, options->id); if (do_write_check(handle, &endian2, 2)) - return -1; + goto error; endian4 = convert_endian_4(handle, options->size); if (do_write_check(handle, &endian4, 4)) - return -1; + goto error; /* Save the data location in case it needs to be updated */ - options->offset = lseek64(handle->fd, 0, SEEK_CUR); + options->offset = do_lseek(handle, 0, SEEK_CUR); if (do_write_check(handle, options->data, options->size)) - return -1; + goto error; } option = TRACECMD_OPTION_DONE; if (do_write_check(handle, &option, 2)) - return -1; + goto error; + if (out_compression_end(handle)) + goto error; handle->file_state = TRACECMD_FILE_OPTIONS; return 0; +error: + out_compression_reset(handle); + return -1; } int tracecmd_append_options(struct tracecmd_output *handle) @@ -1367,42 +1385,51 @@ int tracecmd_append_options(struct tracecmd_output *handle) * We can append only if options are already written and tracing data * is not yet written */ - if (handle->file_state != TRACECMD_FILE_OPTIONS) + if (handle->file_state != TRACECMD_FILE_OPTIONS || !handle->options_start) return -1; - - if (lseek64(handle->fd, 0, SEEK_END) == (off_t)-1) + if (lseek64(handle->fd, handle->options_start, SEEK_SET) == (off_t)-1) return -1; - offset = lseek64(handle->fd, -2, SEEK_CUR); - if (offset == (off_t)-1) + if (out_uncompress_block(handle)) return -1; - - r = pread(handle->fd, &option, 2, offset); + if (do_lseek(handle, 0, SEEK_END) == -1) + goto error; + offset = do_lseek(handle, -2, SEEK_CUR); + if (offset == (off_t)-1) + goto error; + r = do_preed(handle, &option, 2, offset); if (r != 2 || option != TRACECMD_OPTION_DONE) - return -1; - + goto error; list_for_each_entry(options, &handle->options, list) { endian2 = convert_endian_2(handle, options->id); if (do_write_check(handle, &endian2, 2)) - return -1; + goto error; endian4 = convert_endian_4(handle, options->size); if (do_write_check(handle, &endian4, 4)) - return -1; + goto error; /* Save the data location in case it needs to be updated */ - options->offset = lseek64(handle->fd, 0, SEEK_CUR); + options->offset = do_lseek(handle, 0, SEEK_CUR); if (do_write_check(handle, options->data, options->size)) - return -1; + goto error; } - option = TRACECMD_OPTION_DONE; if (do_write_check(handle, &option, 2)) - return -1; + goto error; + if (handle->file_version >= 7 && handle->compress) { + if (lseek64(handle->fd, handle->options_start, SEEK_SET) == (off_t)-1) + goto error; + if (out_compression_end(handle)) + goto error; + } return 0; +error: + out_compression_reset(handle); + return -1; } static struct tracecmd_option * diff --git a/tracecmd/trace-dump.c b/tracecmd/trace-dump.c index 56acb01d..8d0f2251 100644 --- a/tracecmd/trace-dump.c +++ b/tracecmd/trace-dump.c @@ -82,6 +82,13 @@ static int read_compressed(int fd, char *dst, int len) return read_fd(fd, dst, len); } +static int do_lseek(int fd, int offset, int whence) +{ + if (read_compress) + return tracecmd_compress_lseek(compress, offset, whence); + return lseek64(fd, offset, whence); +} + static int read_file_string(int fd, char *dst, int len) { size_t size = 0; @@ -625,6 +632,9 @@ static void dump_options(int fd) unsigned int size; int count = 0; + if (uncompress_block()) + die("cannot uncompress file block"); + for (;;) { if (read_file_number(fd, &option, 2)) die("cannot read the option id"); @@ -635,7 +645,7 @@ static void dump_options(int fd) count++; if (!DUMP_CHECK(OPTIONS) && !DUMP_CHECK(CLOCK) && !DUMP_CHECK(SUMMARY)) { - lseek64(fd, size, SEEK_CUR); + do_lseek(fd, size, SEEK_CUR); continue; } switch (option) { @@ -685,12 +695,12 @@ static void dump_options(int fd) default: do_print(OPTIONS, " %d %d\t[Unknown option, size - skipping]\n", option, size); - lseek64(fd, size, SEEK_CUR); + do_lseek(fd, size, SEEK_CUR); break; } } do_print(SUMMARY, "\t[%d options]\n", count); - + uncompress_reset(); } static void dump_latency(int fd) -- 2.31.1
Refactored the internal logic of tracecmd_write_cpu_data() API to be suitable for adding trace data compression. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/trace-output.c | 94 +++++++++++++++++------------------- 1 file changed, 44 insertions(+), 50 deletions(-) diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index f7e6c3dd..ed4ef638 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -1679,11 +1679,17 @@ static char *get_clock(struct tracecmd_output *handle) return handle->trace_clock; } +struct data_file_write { + int file_size; + off64_t soffset; + off64_t data_offset; + off64_t doffset; +}; + int tracecmd_write_cpu_data(struct tracecmd_output *handle, int cpus, char * const *cpu_data_files) { - off64_t *offsets = NULL; - unsigned long long *sizes = NULL; + struct data_file_write *data_files = NULL; off64_t offset; unsigned long long endian8; char *clock = NULL; @@ -1705,37 +1711,11 @@ int tracecmd_write_cpu_data(struct tracecmd_output *handle, if (do_write_check(handle, "flyrecord", 10)) goto out_free; - offsets = malloc(sizeof(*offsets) * cpus); - if (!offsets) - goto out_free; - sizes = malloc(sizeof(*sizes) * cpus); - if (!sizes) - goto out_free; - - offset = lseek64(handle->fd, 0, SEEK_CUR); - - /* hold any extra data for data */ - offset += cpus * (16); - - /* - * Unfortunately, the trace_clock data was placed after the - * cpu data, and wasn't accounted for with the offsets. - * We need to save room for the trace_clock file. This means - * we need to find the size of it before we define the final - * offsets. - */ - clock = get_clock(handle); - if (!clock) + data_files = calloc(cpus, sizeof(struct data_file_write)); + if (!data_files) goto out_free; - /* Save room for storing the size */ - offset += 8; - offset += strlen(clock); - /* 2 bytes for [] around the clock */ - offset += 2; - - /* Page align offset */ - offset = (offset + (handle->page_size - 1)) & ~(handle->page_size - 1); + /* Write 0 for trace data offset and size and store offsets of these fields */ for (i = 0; i < cpus; i++) { file = cpu_data_files[i]; ret = stat(file, &st); @@ -1743,53 +1723,67 @@ int tracecmd_write_cpu_data(struct tracecmd_output *handle, tracecmd_warning("can not stat '%s'", file); goto out_free; } - offsets[i] = offset; - sizes[i] = st.st_size; - offset += st.st_size; - offset = (offset + (handle->page_size - 1)) & ~(handle->page_size - 1); + data_files[i].file_size = st.st_size; - endian8 = convert_endian_8(handle, offsets[i]); + endian8 = 0; + data_files[i].doffset = lseek64(handle->fd, 0, SEEK_CUR); if (do_write_check(handle, &endian8, 8)) goto out_free; - endian8 = convert_endian_8(handle, sizes[i]); + data_files[i].soffset = lseek64(handle->fd, 0, SEEK_CUR); if (do_write_check(handle, &endian8, 8)) goto out_free; } - if (save_clock(handle, clock)) + clock = get_clock(handle); + if (clock && save_clock(handle, clock)) goto out_free; for (i = 0; i < cpus; i++) { + data_files[i].data_offset = lseek64(handle->fd, 0, SEEK_CUR); + /* Page align offset */ + data_files[i].data_offset = (data_files[i].data_offset + (handle->page_size - 1)) & ~(handle->page_size - 1); + data_files[i].data_offset = lseek64(handle->fd, data_files[i].data_offset, SEEK_SET); + if (data_files[i].data_offset == (off64_t)-1) + goto out_free; if (!tracecmd_get_quiet(handle)) fprintf(stderr, "CPU%d data recorded at offset=0x%llx\n", - i, (unsigned long long) offsets[i]); - offset = lseek64(handle->fd, offsets[i], SEEK_SET); - if (offset == (off64_t)-1) { - tracecmd_warning("could not seek to %lld\n", offsets[i]); - goto out_free; - } + i, (unsigned long long) data_files[i].data_offset); + check_size = copy_file(handle, cpu_data_files[i]); - if (check_size != sizes[i]) { + if (check_size != data_files[i].file_size) { errno = EINVAL; tracecmd_warning("did not match size of %lld to %lld", - check_size, sizes[i]); + check_size, data_files[i].file_size); goto out_free; } + /* Write the real CPU data offset inthe file */ + offset = lseek64(handle->fd, data_files[i].doffset, SEEK_SET); + endian8 = convert_endian_8(handle, data_files[i].data_offset); + if (do_write_check(handle, &endian8, 8)) + goto out_free; + /* Write the real CPU data size in the file */ + offset = lseek64(handle->fd, data_files[i].soffset, SEEK_SET); + endian8 = convert_endian_8(handle, check_size); + if (do_write_check(handle, &endian8, 8)) + goto out_free; + offset = data_files[i].data_offset + check_size; + offset = lseek64(handle->fd, offset, SEEK_SET); if (!tracecmd_get_quiet(handle)) fprintf(stderr, " %llu bytes in size\n", (unsigned long long)check_size); } - free(offsets); - free(sizes); + if (lseek64(handle->fd, 0, SEEK_END) == (off64_t)-1) + goto out_free; + free(data_files); handle->file_state = TRACECMD_FILE_CPU_FLYRECORD; return 0; out_free: - free(offsets); - free(sizes); + lseek64(handle->fd, 0, SEEK_END); + free(data_files); return -1; } -- 2.31.1
New compression APIs are added for writing and reading compressed data from / to files in chunks: tracecmd_compress_copy_from() tracecmd_uncompress_copy_to() Format of the compressed data, used by these APIs, is: - 4 bytes, chunks count - for each chunk: - 4 bytes, size of compressed data in this chunk - 4 bytes, uncompressed size of the data in this chunk - data, bytes of <size of compressed data in this chunk> Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- .../include/private/trace-cmd-private.h | 5 +- lib/trace-cmd/trace-compress.c | 200 ++++++++++++++++++ 2 files changed, 204 insertions(+), 1 deletion(-) diff --git a/lib/trace-cmd/include/private/trace-cmd-private.h b/lib/trace-cmd/include/private/trace-cmd-private.h index 10089389..067ba34d 100644 --- a/lib/trace-cmd/include/private/trace-cmd-private.h +++ b/lib/trace-cmd/include/private/trace-cmd-private.h @@ -494,7 +494,10 @@ int tracecmd_compress_proto_register(const char *name, const char *version, int char *out, unsigned int *out_bytes), unsigned int (*comress_size)(unsigned int bytes), bool (*is_supported)(const char *name, const char *version)); - +int tracecmd_compress_copy_from(struct tracecmd_compression *handle, int fd, int chunk_size, + unsigned long long *read_size, unsigned long long *write_size); +int tracecmd_uncompress_copy_to(struct tracecmd_compression *handle, int fd, + unsigned long long *read_size, unsigned long long *write_size); /* --- Plugin handling --- */ extern struct tep_plugin_option trace_ftrace_options[]; diff --git a/lib/trace-cmd/trace-compress.c b/lib/trace-cmd/trace-compress.c index 039f60c5..dfed7b0e 100644 --- a/lib/trace-cmd/trace-compress.c +++ b/lib/trace-cmd/trace-compress.c @@ -595,3 +595,203 @@ error: free(v); return -1; } + +/** + * tracecmd_compress_copy_from - Copy and compress data from a file + * @handle: compression handler + * @fd: file descriptor to uncompressed data to copy from + * @chunk_size: size of one compression chunk + * @read_size: in - max bytes to read from @fd, 0 to read till the EOF + * out - size of the uncompressed data read from @fd + * @write_size: return, size of the compressed data written into @handle + * + * This function reads uncompressed data from given @fd, compresses the data using the @handle + * compression context and writes the compressed data into the fd associated with the @handle. + * The data is compressed on chunks with given @chunk_size size. + * The compressed data is written in the format: + * - 4 bytes, chunks count + * - for each chunk: + * - 4 bytes, size of compressed data in this chunk + * - 4 bytes, uncompressed size of the data in this chunk + * - data, bytes of <size of compressed data in this chunk> + * + * On success 0 is returned, @read_size and @write_size are updated with the size of + * read and written data. + */ +int tracecmd_compress_copy_from(struct tracecmd_compression *handle, int fd, int chunk_size, + unsigned long long *read_size, unsigned long long *write_size) +{ + unsigned int rchunk = 0; + unsigned int chunks = 0; + unsigned int wsize = 0; + unsigned int rsize = 0; + unsigned int rmax = 0; + unsigned int csize; + unsigned int size; + unsigned int r; + off64_t offset; + char *buf_from; + char *buf_to; + int endian4; + int ret; + + if (!handle || !handle->proto || + !handle->proto->compress_block || !handle->proto->compress_size) + return 0; + if (read_size) + rmax = *read_size; + csize = handle->proto->compress_size(chunk_size); + buf_from = malloc(chunk_size); + if (!buf_from) + return -1; + buf_to = malloc(csize); + if (!buf_to) + return -1; + /* save the initial offset and write 0 chunks */ + offset = lseek64(handle->fd, 0, SEEK_CUR); + write_fd(handle->fd, &chunks, 4); + + do { + if (rmax > 0 && (rmax - rsize) < chunk_size) + rchunk = (rmax - rsize); + else + rchunk = chunk_size; + + r = read(fd, buf_from, rchunk); + if (r < 0 || (rmax > 0 && rsize >= rmax)) + break; + rsize += r; + size = csize; + if (r > 0) { + ret = handle->proto->compress_block(buf_from, r, buf_to, &size); + if (ret < 0) { + if (errno == EINTR) + continue; + break; + } + /* Write compressed data size */ + endian4 = tep_read_number(handle->tep, &size, 4); + ret = write_fd(handle->fd, &endian4, 4); + if (ret != 4) + break; + /* Write uncompressed data size */ + endian4 = tep_read_number(handle->tep, &r, 4); + ret = write_fd(handle->fd, &endian4, 4); + if (ret != 4) + break; + /* Write the compressed data */ + ret = write_fd(handle->fd, buf_to, size); + if (ret != size) + break; + /* data + compress header */ + wsize += (size + 8); + chunks++; + } + } while (r > 0); + free(buf_from); + free(buf_to); + if (r) + return -1; + if (lseek64(handle->fd, offset, SEEK_SET) == (off_t)-1) + return -1; + endian4 = tep_read_number(handle->tep, &chunks, 4); + /* write chunks count*/ + write_fd(handle->fd, &chunks, 4); + lseek64(handle->fd, offset, SEEK_SET); + if (lseek64(handle->fd, 0, SEEK_END) == (off_t)-1) + return -1; + if (read_size) + *read_size = rsize; + if (write_size) + *write_size = wsize; + return 0; +} + +/** + * tracecmd_uncompress_copy_to - Uncompress data and copy to a file + * @handle: compression handler + * @fd: file descriptor to uncompressed data to copy into + * @read_size: return, size of the compressed data read from @handle + * @write_size: return, size of the uncompressed data written into @fd + * + * This function reads compressed data from the fd, associated with @handle, uncompresses it + * using the @handle compression context and writes the uncompressed data into the fd. + * The compressed data must be in the format: + * - 4 bytes, chunks count + * - for each chunk: + * - 4 bytes, size of compressed data in this chunk + * - 4 bytes, uncompressed size of the data in this chunk + * - data, bytes of <size of compressed data in this chunk> + * + * On success 0 is returned, @read_size and @write_size are updated with the size of + * read and written data. + */ +int tracecmd_uncompress_copy_to(struct tracecmd_compression *handle, int fd, + unsigned long long *read_size, unsigned long long *write_size) +{ + unsigned int s_uncompressed; + unsigned int s_compressed; + unsigned int rsize = 0; + unsigned int wsize = 0; + char *bytes_out = NULL; + char *bytes_in = NULL; + int size_out; + int size_in; + int chunks; + char buf[4]; + char *tmp; + int ret; + + if (!handle || !handle->proto || !handle->proto->uncompress_block) + return -1; + + if (read(handle->fd, buf, 4) != 4) + return -1; + chunks = tep_read_number(handle->tep, buf, 4); + rsize += 4; + while (chunks) { + if (read(handle->fd, buf, 4) != 4) + break; + s_compressed = tep_read_number(handle->tep, buf, 4); + rsize += 4; + if (read(handle->fd, buf, 4) != 4) + break; + s_uncompressed = tep_read_number(handle->tep, buf, 4); + rsize += 4; + if (!bytes_in || size_in < s_compressed) { + tmp = realloc(bytes_in, s_compressed); + if (!tmp) + break; + bytes_in = tmp; + size_in = s_compressed; + } + + if (!bytes_out || size_out < s_uncompressed) { + tmp = realloc(bytes_out, s_uncompressed); + if (!tmp) + break; + bytes_out = tmp; + size_out = s_uncompressed; + } + + if (read_fd(handle->fd, bytes_in, s_compressed) < 0) + break; + rsize += s_compressed; + ret = handle->proto->uncompress_block(bytes_in, s_compressed, + bytes_out, &s_uncompressed); + if (ret) + break; + write_fd(fd, bytes_out, s_uncompressed); + wsize += s_uncompressed; + chunks--; + } + free(bytes_in); + free(bytes_out); + if (chunks) + return -1; + if (read_size) + *read_size = rsize; + if (write_size) + *write_size = wsize; + return 0; +} -- 2.31.1
When creating a trace.dat file of version 7, compress the trace data. The data is compressed in chunks. The chunk size is a compile time parameter, set by default to 1MB. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/trace-output.c | 50 +++++++++++++++++++++++++++++++----- 1 file changed, 43 insertions(+), 7 deletions(-) diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index ed4ef638..20a646af 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -321,6 +321,41 @@ static tsize_t copy_file(struct tracecmd_output *handle, return size; } +#define COMPRESS_CHUNK_SIZE (1*1024*1024) +static tsize_t copy_file_compress(struct tracecmd_output *handle, + const char *file, unsigned long long *write_size) +{ + unsigned long long rsize = 0; + unsigned long long wsize = 0; + tsize_t size; + int ret; + int fd; + + fd = open(file, O_RDONLY); + if (fd < 0) { + tracecmd_warning("Can't read '%s'", file); + return 0; + } + + if (handle->file_version >= 7) { + ret = tracecmd_compress_copy_from(handle->compress, fd, + COMPRESS_CHUNK_SIZE, &rsize, &wsize); + if (ret < 0) { + tracecmd_warning("Can't compress '%s'", file); + close(fd); + return 0; + } + size = rsize; + *write_size = wsize; + } else { + size = copy_file_fd(handle, fd); + *write_size = size; + } + + close(fd); + return size; +} + /* * Finds the path to the debugfs/tracing * Allocates the string and stores it. @@ -1692,8 +1727,9 @@ int tracecmd_write_cpu_data(struct tracecmd_output *handle, struct data_file_write *data_files = NULL; off64_t offset; unsigned long long endian8; + unsigned long long read_size; + unsigned long long write_size; char *clock = NULL; - off64_t check_size; char *file; struct stat st; int ret; @@ -1749,11 +1785,11 @@ int tracecmd_write_cpu_data(struct tracecmd_output *handle, fprintf(stderr, "CPU%d data recorded at offset=0x%llx\n", i, (unsigned long long) data_files[i].data_offset); - check_size = copy_file(handle, cpu_data_files[i]); - if (check_size != data_files[i].file_size) { + read_size = copy_file_compress(handle, cpu_data_files[i], &write_size); + if (read_size != data_files[i].file_size) { errno = EINVAL; tracecmd_warning("did not match size of %lld to %lld", - check_size, data_files[i].file_size); + read_size, data_files[i].file_size); goto out_free; } /* Write the real CPU data offset inthe file */ @@ -1763,14 +1799,14 @@ int tracecmd_write_cpu_data(struct tracecmd_output *handle, goto out_free; /* Write the real CPU data size in the file */ offset = lseek64(handle->fd, data_files[i].soffset, SEEK_SET); - endian8 = convert_endian_8(handle, check_size); + endian8 = convert_endian_8(handle, write_size); if (do_write_check(handle, &endian8, 8)) goto out_free; - offset = data_files[i].data_offset + check_size; + offset = data_files[i].data_offset + write_size; offset = lseek64(handle->fd, offset, SEEK_SET); if (!tracecmd_get_quiet(handle)) fprintf(stderr, " %llu bytes in size\n", - (unsigned long long)check_size); + (unsigned long long)write_size); } if (lseek64(handle->fd, 0, SEEK_END) == (off64_t)-1) -- 2.31.1
When reading a trace.dat file of version 7, uncompress the trace data. The trace data for each CPU is uncompressed in a temporary file, located in /tmp directory with prefix "trace_cpu_data". Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/trace-input.c | 74 +++++++++++++++++++++++++++++-------- tracecmd/trace-read.c | 8 ++++ 2 files changed, 67 insertions(+), 15 deletions(-) diff --git a/lib/trace-cmd/trace-input.c b/lib/trace-cmd/trace-input.c index 8fff003e..327082a2 100644 --- a/lib/trace-cmd/trace-input.c +++ b/lib/trace-cmd/trace-input.c @@ -54,6 +54,7 @@ struct page { #endif }; +#define COMPR_TEMP_FILE "/tmp/trace_cpu_dataXXXXXX" struct cpu_data { /* the first two never change */ unsigned long long file_offset; @@ -72,6 +73,10 @@ struct cpu_data { int page_cnt; int cpu; int pipe_fd; + + /* temporary file for uncompressed cpu data */ + int cfd; + char cfile[26]; /* strlen(COMPR_TEMP_FILE) */ }; struct input_buffer_instance { @@ -1080,6 +1085,7 @@ static void *allocate_page_map(struct tracecmd_input *handle, off64_t map_offset; void *map; int ret; + int fd; if (handle->read_page) { map = malloc(handle->page_size); @@ -1119,12 +1125,15 @@ static void *allocate_page_map(struct tracecmd_input *handle, map_size -= map_offset + map_size - (cpu_data->file_offset + cpu_data->file_size); + if (cpu_data->cfd >= 0) + fd = cpu_data->cfd; + else + fd = handle->fd; again: page_map->size = map_size; page_map->offset = map_offset; - page_map->map = mmap(NULL, map_size, PROT_READ, MAP_PRIVATE, - handle->fd, map_offset); + page_map->map = mmap(NULL, map_size, PROT_READ, MAP_PRIVATE, fd, map_offset); if (page_map->map == MAP_FAILED) { /* Try a smaller map */ @@ -2316,13 +2325,41 @@ tracecmd_read_prev(struct tracecmd_input *handle, struct tep_record *record) /* Not reached */ } +static int cpu_data_uncompress(struct tracecmd_input *handle, int cpu, unsigned long long *size) +{ + struct cpu_data *cpu_data; + + cpu_data = &handle->cpu_data[cpu]; + strcpy(cpu_data->cfile, COMPR_TEMP_FILE); + cpu_data->cfd = mkstemp(cpu_data->cfile); + if (cpu_data->cfd < 0) + return -1; + return tracecmd_uncompress_copy_to(handle->compress, cpu_data->cfd, NULL, size); +} + static int init_cpu(struct tracecmd_input *handle, int cpu) { struct cpu_data *cpu_data = &handle->cpu_data[cpu]; + unsigned long long size; + off64_t offset; int i; - cpu_data->offset = cpu_data->file_offset; - cpu_data->size = cpu_data->file_size; + if (handle->file_version >= 7 && cpu_data->file_size > 0) { + offset = lseek64(handle->fd, 0, SEEK_CUR); + if (lseek64(handle->fd, cpu_data->file_offset, SEEK_SET) == (off_t)-1) + return -1; + if (cpu_data_uncompress(handle, cpu, &size) < 0) + return -1; + cpu_data->offset = 0; + cpu_data->file_offset = 0; + cpu_data->file_size = size; + cpu_data->size = size; + if (lseek64(handle->fd, offset, SEEK_SET) == (off_t)-1) + return -1; + } else { + cpu_data->offset = cpu_data->file_offset; + cpu_data->size = cpu_data->file_size; + } cpu_data->timestamp = 0; list_head_init(&cpu_data->page_maps); @@ -3015,6 +3052,7 @@ static int read_cpu_data(struct tracecmd_input *handle) handle->cpu_data[cpu].file_offset = offset; handle->cpu_data[cpu].file_size = size; + handle->cpu_data[cpu].cfd = -1; if (size > max_size) max_size = size; @@ -3635,17 +3673,23 @@ void tracecmd_close(struct tracecmd_input *handle) /* The tracecmd_peek_data may have cached a record */ free_next(handle, cpu); free_page(handle, cpu); - if (handle->cpu_data && handle->cpu_data[cpu].kbuf) { - kbuffer_free(handle->cpu_data[cpu].kbuf); - if (handle->cpu_data[cpu].page_map) - free_page_map(handle->cpu_data[cpu].page_map); - - if (handle->cpu_data[cpu].page_cnt) - tracecmd_warning("%d pages still allocated on cpu %d%s", - handle->cpu_data[cpu].page_cnt, cpu, - show_records(handle->cpu_data[cpu].pages, - handle->cpu_data[cpu].nr_pages)); - free(handle->cpu_data[cpu].pages); + if (handle->cpu_data) { + if (handle->cpu_data[cpu].kbuf) { + kbuffer_free(handle->cpu_data[cpu].kbuf); + if (handle->cpu_data[cpu].page_map) + free_page_map(handle->cpu_data[cpu].page_map); + + if (handle->cpu_data[cpu].page_cnt) + tracecmd_warning("%d pages still allocated on cpu %d%s", + handle->cpu_data[cpu].page_cnt, cpu, + show_records(handle->cpu_data[cpu].pages, + handle->cpu_data[cpu].nr_pages)); + free(handle->cpu_data[cpu].pages); + } + if (handle->cpu_data[cpu].cfd >= 0) { + close(handle->cpu_data[cpu].cfd); + unlink(handle->cpu_data[cpu].cfile); + } } } diff --git a/tracecmd/trace-read.c b/tracecmd/trace-read.c index 0cf6e773..d605d05a 100644 --- a/tracecmd/trace-read.c +++ b/tracecmd/trace-read.c @@ -1363,7 +1363,14 @@ struct tracecmd_input *read_trace_header(const char *file, int flags) static void sig_end(int sig) { + struct handle_list *handles; + fprintf(stderr, "trace-cmd: Received SIGINT\n"); + + list_for_each_entry(handles, &handle_list, list) { + tracecmd_close(handles->handle); + } + exit(0); } @@ -1924,6 +1931,7 @@ void trace_report (int argc, char **argv) /* and version overrides uname! */ if (show_version) otype = OUTPUT_VERSION_ONLY; + read_data_info(&handle_list, otype, global, align_ts); list_for_each_entry(handles, &handle_list, list) { -- 2.31.1
When creating latency trace file of version 7, compress the latency trace data. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/trace-output.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index 20a646af..017b3d7d 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -1661,7 +1661,7 @@ struct tracecmd_output *tracecmd_create_file_latency(const char *output_file, in if (!path) goto out_free; - copy_file(handle, path); + copy_file_compress(handle, path, NULL); put_tracing_file(path); -- 2.31.1
When reading trace file version 7 with latency trace data, uncompress the data. A new trace-cmd API is introduced to handle that case: tracecmd_latency_data_read() Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- .../include/private/trace-cmd-private.h | 2 + lib/trace-cmd/trace-input.c | 42 ++++++++++++++++++- tracecmd/trace-read.c | 6 +-- 3 files changed, 46 insertions(+), 4 deletions(-) diff --git a/lib/trace-cmd/include/private/trace-cmd-private.h b/lib/trace-cmd/include/private/trace-cmd-private.h index 067ba34d..4a3eb1bc 100644 --- a/lib/trace-cmd/include/private/trace-cmd-private.h +++ b/lib/trace-cmd/include/private/trace-cmd-private.h @@ -208,6 +208,8 @@ tracecmd_peek_data_ref(struct tracecmd_input *handle, int cpu) return rec; } +int tracecmd_latency_data_read(struct tracecmd_input *handle, void *buf, size_t size); + struct tep_record * tracecmd_read_prev(struct tracecmd_input *handle, struct tep_record *record); diff --git a/lib/trace-cmd/trace-input.c b/lib/trace-cmd/trace-input.c index 327082a2..12876838 100644 --- a/lib/trace-cmd/trace-input.c +++ b/lib/trace-cmd/trace-input.c @@ -137,6 +137,9 @@ struct tracecmd_input { bool read_page; bool use_pipe; int file_version; + /* temporary file for uncompressed latency data */ + int lat_cfd; + char lat_cfile[26]; /* strlen(COMPR_TEMP_FILE) */ struct cpu_data *cpu_data; long long ts_offset; struct tsc2nsec tsc_calc; @@ -2995,6 +2998,35 @@ static int read_options_type(struct tracecmd_input *handle) return 0; } +int tracecmd_latency_data_read(struct tracecmd_input *handle, void *buf, size_t size) +{ + int fd, r; + + if (handle->file_state != TRACECMD_FILE_CPU_LATENCY) + return -1; + if (handle->lat_cfd >= 0) + fd = handle->lat_cfd; + else + fd = handle->fd; + r = read(fd, buf, size); + + return r; +} + +static int latency_data_uncompress(struct tracecmd_input *handle) +{ + int ret; + + strcpy(handle->lat_cfile, COMPR_TEMP_FILE); + handle->lat_cfd = mkstemp(handle->lat_cfile); + if (handle->lat_cfd < 0) + return -1; + ret = tracecmd_uncompress_copy_to(handle->compress, handle->lat_cfd, NULL, NULL); + if (!ret) + lseek64(handle->lat_cfd, 0, SEEK_SET); + return ret; +} + static int read_cpu_data(struct tracecmd_input *handle) { struct tep_handle *pevent = handle->pevent; @@ -3009,8 +3041,11 @@ static int read_cpu_data(struct tracecmd_input *handle) /* * Check if this is a latency report or not. */ - if (handle->file_state == TRACECMD_FILE_CPU_LATENCY) + if (handle->file_state == TRACECMD_FILE_CPU_LATENCY) { + if (handle->file_version >= 7 && latency_data_uncompress(handle)) + return -1; return 1; + } /* We expect this to be flyrecord */ if (handle->file_state != TRACECMD_FILE_CPU_FLYRECORD) @@ -3448,6 +3483,7 @@ struct tracecmd_input *tracecmd_alloc_fd(int fd, int flags) handle->fd = fd; handle->ref = 1; + handle->lat_cfd = -1; if (do_read_check(handle, buf, 3)) goto failed_read; @@ -3698,6 +3734,10 @@ void tracecmd_close(struct tracecmd_input *handle) free(handle->uname); free(handle->trace_clock); close(handle->fd); + if (handle->lat_cfd >= 0) { + close(handle->lat_cfd); + unlink(handle->lat_cfile); + } tracecmd_free_hooks(handle->hooks); handle->hooks = NULL; diff --git a/tracecmd/trace-read.c b/tracecmd/trace-read.c index d605d05a..dfb814b7 100644 --- a/tracecmd/trace-read.c +++ b/tracecmd/trace-read.c @@ -948,13 +948,13 @@ void trace_show_data(struct tracecmd_input *handle, struct tep_record *record) printf("\n"); } -static void read_rest(void) +static void read_latency(struct tracecmd_input *handle) { char buf[BUFSIZ + 1]; int r; do { - r = read(input_fd, buf, BUFSIZ); + r = tracecmd_latency_data_read(handle, buf, BUFSIZ); if (r > 0) { buf[r] = 0; printf("%s", buf); @@ -1241,7 +1241,7 @@ static void read_data_info(struct list_head *handle_list, enum output_type otype if (ret > 0) { if (multi_inputs) die("latency traces do not work with multiple inputs"); - read_rest(); + read_latency(handles->handle); return; } -- 2.31.1
Make the function, that checks if the next file state is valid, global for the tracmd-library, so it can be reused. It is important the same check logic to be used in the whole library. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/include/trace-cmd-local.h | 4 ++ lib/trace-cmd/trace-input.c | 5 ++ lib/trace-cmd/trace-output.c | 83 ++++++------------------- lib/trace-cmd/trace-util.c | 41 ++++++++++++ 4 files changed, 70 insertions(+), 63 deletions(-) diff --git a/lib/trace-cmd/include/trace-cmd-local.h b/lib/trace-cmd/include/trace-cmd-local.h index 03d2a9a3..d44b58a9 100644 --- a/lib/trace-cmd/include/trace-cmd-local.h +++ b/lib/trace-cmd/include/trace-cmd-local.h @@ -39,6 +39,10 @@ void tracecmd_zlib_free(void); void tracecmd_compress_init(void); void tracecmd_compress_free(void); +bool check_file_state(unsigned long file_version, int current_state, int new_state); +bool check_in_state(struct tracecmd_input *handle, int new_state); +bool check_out_state(struct tracecmd_output *handle, int new_state); + int out_uncompress_block(struct tracecmd_output *handle); int out_compression_start(struct tracecmd_output *handle); int out_compression_end(struct tracecmd_output *handle); diff --git a/lib/trace-cmd/trace-input.c b/lib/trace-cmd/trace-input.c index 12876838..d3c5ad18 100644 --- a/lib/trace-cmd/trace-input.c +++ b/lib/trace-cmd/trace-input.c @@ -4436,3 +4436,8 @@ int tracecmd_enable_tsync(struct tracecmd_input *handle, bool enable) return 0; } + +__hidden bool check_in_state(struct tracecmd_input *handle, int new_state) +{ + return check_file_state(handle->file_version, handle->file_state, new_state); +} diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index 017b3d7d..c41dfb8f 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -419,50 +419,6 @@ int tracecmd_ftrace_enable(int set) return ret; } -static int check_out_state(struct tracecmd_output *handle, int new_state) -{ - if (!handle) - return -1; - - if (handle->file_version < 7) { - switch (new_state) { - case TRACECMD_FILE_BUFERS: - tracecmd_warning("State %d is not supported in trace file version %d", - new_state, handle->file_version); - return -1; - default: - break; - } - } - - switch (new_state) { - case TRACECMD_FILE_HEADERS: - case TRACECMD_FILE_FTRACE_EVENTS: - case TRACECMD_FILE_ALL_EVENTS: - case TRACECMD_FILE_KALLSYMS: - case TRACECMD_FILE_PRINTK: - case TRACECMD_FILE_CMD_LINES: - case TRACECMD_FILE_CPU_COUNT: - case TRACECMD_FILE_BUFERS: - if (handle->file_state == (new_state - 1)) - return 0; - break; - case TRACECMD_FILE_OPTIONS: - if (handle->file_version < 7 && handle->file_state == TRACECMD_FILE_CPU_COUNT) - return 0; - if (handle->file_version >= 7 && handle->file_state == TRACECMD_FILE_BUFERS) - return 0; - break; - case TRACECMD_FILE_CPU_LATENCY: - case TRACECMD_FILE_CPU_FLYRECORD: - if (handle->file_state == TRACECMD_FILE_OPTIONS) - return 0; - break; - } - - return -1; -} - static int read_header_files(struct tracecmd_output *handle) { tsize_t size, check_size, endian8; @@ -471,7 +427,7 @@ static int read_header_files(struct tracecmd_output *handle) int fd; int ret; - if (check_out_state(handle, TRACECMD_FILE_HEADERS) < 0) { + if (!check_out_state(handle, TRACECMD_FILE_HEADERS)) { tracecmd_warning("Cannot read header files, unexpected state 0x%X", handle->file_state); return -1; @@ -783,7 +739,7 @@ static int read_ftrace_files(struct tracecmd_output *handle) struct tracecmd_event_list list = { .glob = "ftrace/*" }; int ret; - if (check_out_state(handle, TRACECMD_FILE_FTRACE_EVENTS) < 0) { + if (!check_out_state(handle, TRACECMD_FILE_FTRACE_EVENTS)) { tracecmd_warning("Cannot read ftrace files, unexpected state 0x%X", handle->file_state); return -1; @@ -828,7 +784,7 @@ static int read_event_files(struct tracecmd_output *handle, int endian4; int ret; - if (check_out_state(handle, TRACECMD_FILE_ALL_EVENTS) < 0) { + if (!check_out_state(handle, TRACECMD_FILE_ALL_EVENTS)) { tracecmd_warning("Cannot read event files, unexpected state 0x%X", handle->file_state); return -1; @@ -927,7 +883,7 @@ static int read_proc_kallsyms(struct tracecmd_output *handle, struct stat st; int ret; - if (check_out_state(handle, TRACECMD_FILE_KALLSYMS) < 0) { + if (!check_out_state(handle, TRACECMD_FILE_KALLSYMS)) { tracecmd_warning("Cannot read kallsyms, unexpected state 0x%X", handle->file_state); return -1; @@ -977,7 +933,7 @@ static int read_ftrace_printk(struct tracecmd_output *handle) char *path; int ret; - if (check_out_state(handle, TRACECMD_FILE_PRINTK) < 0) { + if (!check_out_state(handle, TRACECMD_FILE_PRINTK)) { tracecmd_warning("Cannot read printk, unexpected state 0x%X", handle->file_state); return -1; @@ -1339,11 +1295,10 @@ int tracecmd_write_cpus(struct tracecmd_output *handle, int cpus) { int ret; - ret = check_out_state(handle, TRACECMD_FILE_CPU_COUNT); - if (ret < 0) { + if (!check_out_state(handle, TRACECMD_FILE_CPU_COUNT)) { tracecmd_warning("Cannot write CPU count into the file, unexpected state 0x%X", handle->file_state); - return ret; + return -1; } cpus = convert_endian_4(handle, cpus); ret = do_write_check(handle, &cpus, 4); @@ -1359,16 +1314,14 @@ int tracecmd_write_options(struct tracecmd_output *handle) unsigned short option; unsigned short endian2; unsigned int endian4; - int ret; /* If already written, ignore */ if (handle->file_state == TRACECMD_FILE_OPTIONS) return 0; - ret = check_out_state(handle, TRACECMD_FILE_OPTIONS); - if (ret < 0) { + if (!check_out_state(handle, TRACECMD_FILE_OPTIONS)) { tracecmd_warning("Cannot write options into the file, unexpected state 0x%X", handle->file_state); - return ret; + return -1; } if (do_write_check(handle, "options ", 10)) @@ -1604,11 +1557,10 @@ int tracecmd_write_cmdlines(struct tracecmd_output *handle) { int ret; - ret = check_out_state(handle, TRACECMD_FILE_CMD_LINES); - if (ret < 0) { + if (!check_out_state(handle, TRACECMD_FILE_CMD_LINES)) { tracecmd_warning("Cannot write command lines into the file, unexpected state 0x%X", handle->file_state); - return ret; + return -1; } out_compression_start(handle); @@ -1628,7 +1580,6 @@ struct tracecmd_output *tracecmd_create_file_latency(const char *output_file, in { struct tracecmd_output *handle; char *path; - int ret; handle = create_file(output_file, NULL, NULL, NULL, &all_event_list, compression); if (!handle) @@ -1647,8 +1598,7 @@ struct tracecmd_output *tracecmd_create_file_latency(const char *output_file, in if (tracecmd_write_options(handle) < 0) goto out_free; - ret = check_out_state(handle, TRACECMD_FILE_CPU_LATENCY); - if (ret < 0) { + if (!check_out_state(handle, TRACECMD_FILE_CPU_LATENCY)) { tracecmd_warning("Cannot write latency data into the file, unexpected state 0x%X", handle->file_state); goto out_free; @@ -1737,7 +1687,9 @@ int tracecmd_write_cpu_data(struct tracecmd_output *handle, /* This can be called multiple times (when recording instances) */ ret = handle->file_state == TRACECMD_FILE_CPU_FLYRECORD ? 0 : - check_out_state(handle, TRACECMD_FILE_CPU_FLYRECORD); + check_file_state(handle->file_version, + handle->file_state, + TRACECMD_FILE_CPU_FLYRECORD); if (ret < 0) { tracecmd_warning("Cannot write trace data into the file, unexpected state 0x%X", handle->file_state); @@ -2006,3 +1958,8 @@ out_free: tracecmd_output_close(handle); return NULL; } + +__hidden bool check_out_state(struct tracecmd_output *handle, int new_state) +{ + return check_file_state(handle->file_version, handle->file_state, new_state); +} diff --git a/lib/trace-cmd/trace-util.c b/lib/trace-cmd/trace-util.c index 136260c6..087d37ab 100644 --- a/lib/trace-cmd/trace-util.c +++ b/lib/trace-cmd/trace-util.c @@ -636,3 +636,44 @@ void tracecmd_lib_free(void) { tracecmd_compress_free(); } + +__hidden bool check_file_state(unsigned long file_version, int current_state, int new_state) +{ + if (file_version < 7) { + switch (new_state) { + case TRACECMD_FILE_BUFERS: + tracecmd_warning("State %d is not supported in trace file version %d", + new_state, file_version); + return false; + default: + break; + } + } + + switch (new_state) { + case TRACECMD_FILE_HEADERS: + case TRACECMD_FILE_FTRACE_EVENTS: + case TRACECMD_FILE_ALL_EVENTS: + case TRACECMD_FILE_KALLSYMS: + case TRACECMD_FILE_PRINTK: + case TRACECMD_FILE_CMD_LINES: + case TRACECMD_FILE_CPU_COUNT: + case TRACECMD_FILE_BUFERS: + if (current_state == (new_state - 1)) + return true; + break; + case TRACECMD_FILE_OPTIONS: + if (file_version < 7 && current_state == TRACECMD_FILE_CPU_COUNT) + return true; + if (file_version >= 7 && current_state == TRACECMD_FILE_BUFERS) + return true; + break; + case TRACECMD_FILE_CPU_LATENCY: + case TRACECMD_FILE_CPU_FLYRECORD: + if (current_state == TRACECMD_FILE_OPTIONS) + return true; + break; + } + + return false; +} -- 2.31.1
When copying headers between two trace files, use output handler context instead of just a fd. Using output handler will allow to use file compression, if needed. This change fixes "trace-cmd split" command to work with trace files v7. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- .../include/private/trace-cmd-private.h | 3 +- lib/trace-cmd/include/trace-cmd-local.h | 4 + lib/trace-cmd/trace-input.c | 244 ++++++++++++------ lib/trace-cmd/trace-output.c | 13 +- 4 files changed, 174 insertions(+), 90 deletions(-) diff --git a/lib/trace-cmd/include/private/trace-cmd-private.h b/lib/trace-cmd/include/private/trace-cmd-private.h index 4a3eb1bc..65e80cec 100644 --- a/lib/trace-cmd/include/private/trace-cmd-private.h +++ b/lib/trace-cmd/include/private/trace-cmd-private.h @@ -170,7 +170,8 @@ int tracecmd_get_parsing_failures(struct tracecmd_input *handle); int tracecmd_long_size(struct tracecmd_input *handle); int tracecmd_page_size(struct tracecmd_input *handle); int tracecmd_cpus(struct tracecmd_input *handle); -int tracecmd_copy_headers(struct tracecmd_input *handle, int fd, +int tracecmd_copy_headers(struct tracecmd_input *in_handle, + struct tracecmd_output *out_handle, enum tracecmd_file_states start_state, enum tracecmd_file_states end_state); void tracecmd_set_flag(struct tracecmd_input *handle, int flag); diff --git a/lib/trace-cmd/include/trace-cmd-local.h b/lib/trace-cmd/include/trace-cmd-local.h index d44b58a9..8fb391fc 100644 --- a/lib/trace-cmd/include/trace-cmd-local.h +++ b/lib/trace-cmd/include/trace-cmd-local.h @@ -51,7 +51,11 @@ void out_compression_reset(struct tracecmd_output *handle); void in_uncompress_reset(struct tracecmd_input *handle); int in_uncompress_block(struct tracecmd_input *handle); +void out_set_file_state(struct tracecmd_output *handle, int new_state); + int write_buffers_description_v7(struct tracecmd_output *handle); int write_buffers_description_v6(struct tracecmd_output *handle); +long long do_write_check(struct tracecmd_output *handle, const void *data, long long size); + #endif /* _TRACE_CMD_LOCAL_H */ diff --git a/lib/trace-cmd/trace-input.c b/lib/trace-cmd/trace-input.c index d3c5ad18..74d1ebf7 100644 --- a/lib/trace-cmd/trace-input.c +++ b/lib/trace-cmd/trace-input.c @@ -3759,44 +3759,47 @@ void tracecmd_close(struct tracecmd_input *handle) free(handle); } -static int read_copy_size8(struct tracecmd_input *handle, int fd, unsigned long long *size) +static int read_copy_size8(struct tracecmd_input *in_handle, + struct tracecmd_output *out_handle, unsigned long long *size) { /* read size */ - if (do_read_check(handle, size, 8)) + if (do_read_check(in_handle, size, 8)) return -1; - if (__do_write_check(fd, size, 8)) + if (do_write_check(out_handle, size, 8)) return -1; - *size = tep_read_number(handle->pevent, size, 8); + *size = tep_read_number(in_handle->pevent, size, 8); return 0; } -static int read_copy_size4(struct tracecmd_input *handle, int fd, unsigned int *size) +static int read_copy_size4(struct tracecmd_input *in_handle, struct tracecmd_output *out_handle, + unsigned int *size) { /* read size */ - if (do_read_check(handle, size, 4)) + if (do_read_check(in_handle, size, 4)) return -1; - if (__do_write_check(fd, size, 4)) + if (do_write_check(out_handle, size, 4)) return -1; - *size = tep_read_number(handle->pevent, size, 4); + *size = tep_read_number(in_handle->pevent, size, 4); return 0; } -static int read_copy_data(struct tracecmd_input *handle, - unsigned long long size, int fd) +static int read_copy_data(struct tracecmd_input *in_handle, + unsigned long long size, + struct tracecmd_output *out_handle) { char *buf; buf = malloc(size); if (!buf) return -1; - if (do_read_check(handle, buf, size)) + if (do_read_check(in_handle, buf, size)) goto failed_read; - if (__do_write_check(fd, buf, size)) + if (do_write_check(out_handle, buf, size)) goto failed_read; free(buf); @@ -3808,65 +3811,82 @@ static int read_copy_data(struct tracecmd_input *handle, return -1; } -static int copy_header_files(struct tracecmd_input *handle, int fd) +static int copy_header_files(struct tracecmd_input *in_handle, + struct tracecmd_output *out_handle) { unsigned long long size; - if (handle->file_state != TRACECMD_FILE_HEADERS - 1) + if (!check_in_state(in_handle, TRACECMD_FILE_HEADERS) || + !check_out_state(out_handle, TRACECMD_FILE_HEADERS)) return -1; /* "header_page" */ - if (read_copy_data(handle, 12, fd) < 0) + if (read_copy_data(in_handle, 12, out_handle) < 0) return -1; - if (read_copy_size8(handle, fd, &size) < 0) + if (read_copy_size8(in_handle, out_handle, &size) < 0) return -1; - if (read_copy_data(handle, size, fd) < 0) + if (read_copy_data(in_handle, size, out_handle) < 0) return -1; /* "header_event" */ - if (read_copy_data(handle, 13, fd) < 0) + if (read_copy_data(in_handle, 13, out_handle) < 0) return -1; - if (read_copy_size8(handle, fd, &size) < 0) + if (read_copy_size8(in_handle, out_handle, &size) < 0) return -1; - if (read_copy_data(handle, size, fd) < 0) + if (read_copy_data(in_handle, size, out_handle) < 0) return -1; - handle->file_state = TRACECMD_FILE_HEADERS; + in_handle->file_state = TRACECMD_FILE_HEADERS; + out_set_file_state(out_handle, in_handle->file_state); return 0; } -static int copy_ftrace_files(struct tracecmd_input *handle, int fd) +static int copy_ftrace_files(struct tracecmd_input *in_handle, struct tracecmd_output *out_handle) { unsigned long long size; unsigned int count; unsigned int i; - if (handle->file_state != TRACECMD_FILE_FTRACE_EVENTS - 1) + if (!check_in_state(in_handle, TRACECMD_FILE_FTRACE_EVENTS) || + !check_out_state(out_handle, TRACECMD_FILE_FTRACE_EVENTS)) return -1; - if (read_copy_size4(handle, fd, &count) < 0) + if (in_uncompress_block(in_handle)) return -1; + out_compression_start(out_handle); + + if (read_copy_size4(in_handle, out_handle, &count) < 0) + goto error; for (i = 0; i < count; i++) { - if (read_copy_size8(handle, fd, &size) < 0) - return -1; + if (read_copy_size8(in_handle, out_handle, &size) < 0) + goto error; - if (read_copy_data(handle, size, fd) < 0) - return -1; + if (read_copy_data(in_handle, size, out_handle) < 0) + goto error; } - handle->file_state = TRACECMD_FILE_FTRACE_EVENTS; + in_uncompress_reset(in_handle); + in_handle->file_state = TRACECMD_FILE_FTRACE_EVENTS; + + if (out_compression_end(out_handle)) + goto error; + out_set_file_state(out_handle, in_handle->file_state); return 0; +error: + out_compression_reset(out_handle); + in_uncompress_reset(in_handle); + return -1; } -static int copy_event_files(struct tracecmd_input *handle, int fd) +static int copy_event_files(struct tracecmd_input *in_handle, struct tracecmd_output *out_handle) { unsigned long long size; char *system; @@ -3874,103 +3894,159 @@ static int copy_event_files(struct tracecmd_input *handle, int fd) unsigned int count; unsigned int i,x; - if (handle->file_state != TRACECMD_FILE_ALL_EVENTS - 1) + if (!check_in_state(in_handle, TRACECMD_FILE_ALL_EVENTS) || + !check_out_state(out_handle, TRACECMD_FILE_ALL_EVENTS)) return -1; - if (read_copy_size4(handle, fd, &systems) < 0) + if (in_uncompress_block(in_handle)) return -1; + out_compression_start(out_handle); + + if (read_copy_size4(in_handle, out_handle, &systems) < 0) + goto error; for (i = 0; i < systems; i++) { - system = read_string(handle); + system = read_string(in_handle); if (!system) - return -1; - if (__do_write_check(fd, system, strlen(system) + 1)) { + goto error; + if (do_write_check(out_handle, system, strlen(system) + 1)) { free(system); - return -1; + goto error; } free(system); - if (read_copy_size4(handle, fd, &count) < 0) - return -1; + if (read_copy_size4(in_handle, out_handle, &count) < 0) + goto error; for (x=0; x < count; x++) { - if (read_copy_size8(handle, fd, &size) < 0) - return -1; + if (read_copy_size8(in_handle, out_handle, &size) < 0) + goto error; - if (read_copy_data(handle, size, fd) < 0) - return -1; + if (read_copy_data(in_handle, size, out_handle) < 0) + goto error; } } - handle->file_state = TRACECMD_FILE_ALL_EVENTS; + in_uncompress_reset(in_handle); + in_handle->file_state = TRACECMD_FILE_ALL_EVENTS; + + if (out_compression_end(out_handle)) + goto error; + out_set_file_state(out_handle, in_handle->file_state); return 0; +error: + out_compression_reset(out_handle); + in_uncompress_reset(in_handle); + return -1; } -static int copy_proc_kallsyms(struct tracecmd_input *handle, int fd) +static int copy_proc_kallsyms(struct tracecmd_input *in_handle, struct tracecmd_output *out_handle) { unsigned int size; - if (handle->file_state != TRACECMD_FILE_KALLSYMS - 1) + if (!check_in_state(in_handle, TRACECMD_FILE_KALLSYMS) || + !check_out_state(out_handle, TRACECMD_FILE_KALLSYMS)) return -1; - if (read_copy_size4(handle, fd, &size) < 0) + if (in_uncompress_block(in_handle)) return -1; + out_compression_start(out_handle); + + if (read_copy_size4(in_handle, out_handle, &size) < 0) + goto error; if (!size) - return 0; /* OK? */ + goto out; /* OK? */ - if (read_copy_data(handle, size, fd) < 0) - return -1; + if (read_copy_data(in_handle, size, out_handle) < 0) + goto error; - handle->file_state = TRACECMD_FILE_KALLSYMS; +out: + in_uncompress_reset(in_handle); + in_handle->file_state = TRACECMD_FILE_KALLSYMS; + + if (out_compression_end(out_handle)) + goto error; + out_set_file_state(out_handle, in_handle->file_state); return 0; +error: + out_compression_reset(out_handle); + in_uncompress_reset(in_handle); + return -1; } -static int copy_ftrace_printk(struct tracecmd_input *handle, int fd) +static int copy_ftrace_printk(struct tracecmd_input *in_handle, struct tracecmd_output *out_handle) { unsigned int size; - if (handle->file_state != TRACECMD_FILE_PRINTK - 1) + if (!check_in_state(in_handle, TRACECMD_FILE_PRINTK) || + !check_out_state(out_handle, TRACECMD_FILE_PRINTK)) return -1; - if (read_copy_size4(handle, fd, &size) < 0) + if (in_uncompress_block(in_handle)) return -1; - if (!size) - return 0; /* OK? */ + out_compression_start(out_handle); - if (read_copy_data(handle, size, fd) < 0) - return -1; + if (read_copy_size4(in_handle, out_handle, &size) < 0) + goto error; + if (!size) + goto out; /* OK? */ - handle->file_state = TRACECMD_FILE_PRINTK; + if (read_copy_data(in_handle, size, out_handle) < 0) + goto error; +out: + in_uncompress_reset(in_handle); + in_handle->file_state = TRACECMD_FILE_PRINTK; + if (out_compression_end(out_handle)) + goto error; + out_set_file_state(out_handle, in_handle->file_state); return 0; +error: + out_compression_reset(out_handle); + in_uncompress_reset(in_handle); + return -1; } -static int copy_command_lines(struct tracecmd_input *handle, int fd) +static int copy_command_lines(struct tracecmd_input *in_handle, struct tracecmd_output *out_handle) { unsigned long long size; - if (handle->file_state != TRACECMD_FILE_CMD_LINES - 1) + if (!check_in_state(in_handle, TRACECMD_FILE_CMD_LINES) || + !check_out_state(out_handle, TRACECMD_FILE_CMD_LINES)) return -1; - if (read_copy_size8(handle, fd, &size) < 0) + if (in_uncompress_block(in_handle)) return -1; + out_compression_start(out_handle); + + if (read_copy_size8(in_handle, out_handle, &size) < 0) + goto error; if (!size) - return 0; /* OK? */ + goto out; /* OK? */ - if (read_copy_data(handle, size, fd) < 0) - return -1; + if (read_copy_data(in_handle, size, out_handle) < 0) + goto error; - handle->file_state = TRACECMD_FILE_CMD_LINES; +out: + in_uncompress_reset(in_handle); + in_handle->file_state = TRACECMD_FILE_CMD_LINES; + if (out_compression_end(out_handle)) + goto error; + out_set_file_state(out_handle, in_handle->file_state); return 0; +error: + out_compression_reset(out_handle); + in_uncompress_reset(in_handle); + return -1; } /** * tracecmd_copy_headers - Copy headers from a tracecmd_input handle to a file descriptor - * @handle: input handle for the trace.dat file to copy from. - * @fd: The file descriptor to copy to. + * @in_handle: input handle for the trace.dat file to copy from. + * @out_handle: output handle to the trace.dat file to copy to. * @start_state: The file state to start copying from (zero for the beginnig) * @end_state: The file state to stop at (zero for up to cmdlines) * @@ -3981,7 +4057,8 @@ static int copy_command_lines(struct tracecmd_input *handle, int fd) * NOTE: The input handle is also modified, and ends at the end * state as well. */ -int tracecmd_copy_headers(struct tracecmd_input *handle, int fd, +int tracecmd_copy_headers(struct tracecmd_input *in_handle, + struct tracecmd_output *out_handle, enum tracecmd_file_states start_state, enum tracecmd_file_states end_state) { @@ -3997,68 +4074,67 @@ int tracecmd_copy_headers(struct tracecmd_input *handle, int fd, if (end_state < TRACECMD_FILE_HEADERS) return 0; - - if (handle->file_state >= start_state) { + if (in_handle->file_state >= start_state) { /* Set the handle to just before the start state */ - lseek64(handle->fd, handle->header_files_start, SEEK_SET); + lseek64(in_handle->fd, in_handle->header_files_start, SEEK_SET); /* Now that the file handle has moved, change its state */ - handle->file_state = TRACECMD_FILE_INIT; + in_handle->file_state = TRACECMD_FILE_INIT; } /* Try to bring the input up to the start state - 1 */ - ret = tracecmd_read_headers(handle, start_state - 1); + ret = tracecmd_read_headers(in_handle, start_state - 1); if (ret < 0) goto out; switch (start_state) { case TRACECMD_FILE_HEADERS: - ret = copy_header_files(handle, fd); + ret = copy_header_files(in_handle, out_handle); if (ret < 0) goto out; /* fallthrough */ case TRACECMD_FILE_FTRACE_EVENTS: /* handle's state is now updating with the copies */ - if (end_state <= handle->file_state) + if (end_state <= in_handle->file_state) return 0; - ret = copy_ftrace_files(handle, fd); + ret = copy_ftrace_files(in_handle, out_handle); if (ret < 0) goto out; /* fallthrough */ case TRACECMD_FILE_ALL_EVENTS: - if (end_state <= handle->file_state) + if (end_state <= in_handle->file_state) return 0; - ret = copy_event_files(handle, fd); + ret = copy_event_files(in_handle, out_handle); if (ret < 0) goto out; /* fallthrough */ case TRACECMD_FILE_KALLSYMS: - if (end_state <= handle->file_state) + if (end_state <= in_handle->file_state) return 0; - ret = copy_proc_kallsyms(handle, fd); + ret = copy_proc_kallsyms(in_handle, out_handle); if (ret < 0) goto out; /* fallthrough */ case TRACECMD_FILE_PRINTK: - if (end_state <= handle->file_state) + if (end_state <= in_handle->file_state) return 0; - ret = copy_ftrace_printk(handle, fd); + ret = copy_ftrace_printk(in_handle, out_handle); if (ret < 0) goto out; /* fallthrough */ case TRACECMD_FILE_CMD_LINES: - if (end_state <= handle->file_state) + if (end_state <= in_handle->file_state) return 0; - ret = copy_command_lines(handle, fd); + ret = copy_command_lines(in_handle, out_handle); default: break; } diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index c41dfb8f..06d0738e 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -89,8 +89,8 @@ struct list_event_system { char *name; }; -static stsize_t -do_write_check(struct tracecmd_output *handle, const void *data, tsize_t size) +__hidden long long +do_write_check(struct tracecmd_output *handle, const void *data, long long size) { if (handle->do_compress) return tracecmd_compress_write(handle->compress, data, size); @@ -1946,11 +1946,9 @@ struct tracecmd_output *tracecmd_copy(struct tracecmd_input *ihandle, const char if (!handle) return NULL; - if (tracecmd_copy_headers(ihandle, handle->fd, 0, 0) < 0) + if (tracecmd_copy_headers(ihandle, handle, 0, 0) < 0) goto out_free; - handle->file_state = tracecmd_get_file_state(ihandle); - /* The file is all ready to have cpu data attached */ return handle; @@ -1959,6 +1957,11 @@ out_free: return NULL; } +__hidden void out_set_file_state(struct tracecmd_output *handle, int new_state) +{ + handle->file_state = new_state; +} + __hidden bool check_out_state(struct tracecmd_output *handle, int new_state) { return check_file_state(handle->file_version, handle->file_state, new_state); -- 2.31.1
When creating a temporary output handler for use with a stream, force it not not use trace file compression. As no data are saved in this file, there is no need to have a compression overhead. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- tracecmd/trace-stream.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tracecmd/trace-stream.c b/tracecmd/trace-stream.c index 317de1b2..1ca3a79d 100644 --- a/tracecmd/trace-stream.c +++ b/tracecmd/trace-stream.c @@ -43,7 +43,7 @@ trace_stream_init(struct buffer_instance *instance, int cpu, int fd, int cpus, tfd = fileno(fp); ofd = dup(tfd); - trace_output = tracecmd_create_init_fd(ofd, NULL); + trace_output = tracecmd_create_init_fd(ofd, "none"); if (!trace_output) { fclose(fp); return NULL; -- 2.31.1
Added an API to get the version of the trace file, associated with given output file handler. tracecmd_get_out_file_version() Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/include/private/trace-cmd-private.h | 1 + lib/trace-cmd/trace-output.c | 9 +++++++++ 2 files changed, 10 insertions(+) diff --git a/lib/trace-cmd/include/private/trace-cmd-private.h b/lib/trace-cmd/include/private/trace-cmd-private.h index 65e80cec..9a08a960 100644 --- a/lib/trace-cmd/include/private/trace-cmd-private.h +++ b/lib/trace-cmd/include/private/trace-cmd-private.h @@ -319,6 +319,7 @@ int tracecmd_append_cpu_data(struct tracecmd_output *handle, int tracecmd_append_buffer_cpu_data(struct tracecmd_output *handle, const char *name, int cpus, char * const *cpu_data_files); struct tracecmd_output *tracecmd_get_output_handle_fd(int fd); +unsigned long tracecmd_get_out_file_version(struct tracecmd_output *handle); /* --- Reading the Fly Recorder Trace --- */ diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index 06d0738e..9b83eb76 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -1966,3 +1966,12 @@ __hidden bool check_out_state(struct tracecmd_output *handle, int new_state) { return check_file_state(handle->file_version, handle->file_state, new_state); } + +/** + * tracecmd_get_out_file_version - return the trace.dat file version + * @handle: output handle for the trace.dat file + */ +unsigned long tracecmd_get_out_file_version(struct tracecmd_output *handle) +{ + return handle->file_version; +} -- 2.31.1
The tracecmd_copy() API is used to create output trace handler from given input trace handler and to copy data from it. Currently it is used by trace-cmd split and restore commands, that's why it is hardcoded what data to be copied from input to output handler. Addig desired output handler file state makes the API more generic and allows it to be used in more use cases. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/include/private/trace-cmd-private.h | 4 ++-- lib/trace-cmd/trace-output.c | 8 ++++++-- tracecmd/trace-restore.c | 2 +- tracecmd/trace-split.c | 2 +- 4 files changed, 10 insertions(+), 6 deletions(-) diff --git a/lib/trace-cmd/include/private/trace-cmd-private.h b/lib/trace-cmd/include/private/trace-cmd-private.h index 9a08a960..2c31be5f 100644 --- a/lib/trace-cmd/include/private/trace-cmd-private.h +++ b/lib/trace-cmd/include/private/trace-cmd-private.h @@ -309,8 +309,8 @@ int tracecmd_write_options(struct tracecmd_output *handle); int tracecmd_append_options(struct tracecmd_output *handle); void tracecmd_output_close(struct tracecmd_output *handle); void tracecmd_output_free(struct tracecmd_output *handle); -struct tracecmd_output *tracecmd_copy(struct tracecmd_input *ihandle, - const char *file); +struct tracecmd_output *tracecmd_copy(struct tracecmd_input *ihandle, const char *file, + enum tracecmd_file_states state); int tracecmd_write_cpu_data(struct tracecmd_output *handle, int cpus, char * const *cpu_data_files); diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index 9b83eb76..806c833c 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -1931,13 +1931,16 @@ struct tracecmd_output *tracecmd_create_init_file_override(const char *output_fi * tracecmd_copy - copy the headers of one trace.dat file for another * @ihandle: input handle of the trace.dat file to copy * @file: the trace.dat file to create + * @state: what data will be copied from the source handle * * Reads the header information and creates a new trace data file * with the same characteristics (events and all) and returns * tracecmd_output handle to this new file. */ -struct tracecmd_output *tracecmd_copy(struct tracecmd_input *ihandle, const char *file) +struct tracecmd_output *tracecmd_copy(struct tracecmd_input *ihandle, const char *file, + enum tracecmd_file_states state) { + enum tracecmd_file_states fstate; struct tracecmd_output *handle; const char *compr_name = NULL; @@ -1946,7 +1949,8 @@ struct tracecmd_output *tracecmd_copy(struct tracecmd_input *ihandle, const char if (!handle) return NULL; - if (tracecmd_copy_headers(ihandle, handle, 0, 0) < 0) + fstate = state > TRACECMD_FILE_CMD_LINES ? TRACECMD_FILE_CMD_LINES : state; + if (tracecmd_copy_headers(ihandle, handle, 0, fstate) < 0) goto out_free; /* The file is all ready to have cpu data attached */ diff --git a/tracecmd/trace-restore.c b/tracecmd/trace-restore.c index 96b4fb5d..7d439fd2 100644 --- a/tracecmd/trace-restore.c +++ b/tracecmd/trace-restore.c @@ -125,7 +125,7 @@ void trace_restore (int argc, char **argv) if (tracecmd_read_headers(ihandle, TRACECMD_FILE_CMD_LINES) < 0) die("error reading file %s headers", input); - handle = tracecmd_copy(ihandle, output); + handle = tracecmd_copy(ihandle, output, TRACECMD_FILE_CMD_LINES); tracecmd_close(ihandle); } else handle = tracecmd_create_init_file(output, NULL); diff --git a/tracecmd/trace-split.c b/tracecmd/trace-split.c index 8366d128..401e1d29 100644 --- a/tracecmd/trace-split.c +++ b/tracecmd/trace-split.c @@ -345,7 +345,7 @@ static double parse_file(struct tracecmd_input *handle, dir = dirname(output); base = basename(output); - ohandle = tracecmd_copy(handle, output_file); + ohandle = tracecmd_copy(handle, output_file, TRACECMD_FILE_CMD_LINES); cpus = tracecmd_cpus(handle); cpu_data = malloc(sizeof(*cpu_data) * cpus); -- 2.31.1
Extend the tracecmd_copy() API to support copying of CPU count from input to output trace hanlder. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/trace-input.c | 30 ++++++++++++++++++++++++++++++ lib/trace-cmd/trace-output.c | 2 +- 2 files changed, 31 insertions(+), 1 deletion(-) diff --git a/lib/trace-cmd/trace-input.c b/lib/trace-cmd/trace-input.c index 74d1ebf7..289102bf 100644 --- a/lib/trace-cmd/trace-input.c +++ b/lib/trace-cmd/trace-input.c @@ -4043,6 +4043,23 @@ error: return -1; } +static int copy_cpu_count(struct tracecmd_input *in_handle, struct tracecmd_output *out_handle) +{ + unsigned int size; + + if (!check_in_state(in_handle, TRACECMD_FILE_CPU_COUNT) || + !check_out_state(out_handle, TRACECMD_FILE_CPU_COUNT)) + return -1; + + if (read_copy_size4(in_handle, out_handle, &size) < 0) + return -1; + + in_handle->file_state = TRACECMD_FILE_CPU_COUNT; + out_set_file_state(out_handle, in_handle->file_state); + + return 0; +} + /** * tracecmd_copy_headers - Copy headers from a tracecmd_input handle to a file descriptor * @in_handle: input handle for the trace.dat file to copy from. @@ -4135,6 +4152,19 @@ int tracecmd_copy_headers(struct tracecmd_input *in_handle, return 0; ret = copy_command_lines(in_handle, out_handle); + if (ret < 0) + goto out; + + /* fallthrough */ + case TRACECMD_FILE_CPU_COUNT: + if (end_state <= in_handle->file_state) + return 0; + + ret = copy_cpu_count(in_handle, out_handle); + if (ret < 0) + goto out; + + /* fallthrough */ default: break; } diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index 806c833c..b4bdaeeb 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -1949,7 +1949,7 @@ struct tracecmd_output *tracecmd_copy(struct tracecmd_input *ihandle, const char if (!handle) return NULL; - fstate = state > TRACECMD_FILE_CMD_LINES ? TRACECMD_FILE_CMD_LINES : state; + fstate = state > TRACECMD_FILE_CPU_COUNT ? TRACECMD_FILE_CPU_COUNT : state; if (tracecmd_copy_headers(ihandle, handle, 0, fstate) < 0) goto out_free; -- 2.31.1
Extend the tracecmd_copy() API to support copying of trace buffers description from input to output trace handler. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- .../include/private/trace-cmd-private.h | 2 + lib/trace-cmd/trace-input.c | 38 +++++++++++++++++++ lib/trace-cmd/trace-output.c | 3 ++ 3 files changed, 43 insertions(+) diff --git a/lib/trace-cmd/include/private/trace-cmd-private.h b/lib/trace-cmd/include/private/trace-cmd-private.h index 2c31be5f..3735586d 100644 --- a/lib/trace-cmd/include/private/trace-cmd-private.h +++ b/lib/trace-cmd/include/private/trace-cmd-private.h @@ -174,6 +174,8 @@ int tracecmd_copy_headers(struct tracecmd_input *in_handle, struct tracecmd_output *out_handle, enum tracecmd_file_states start_state, enum tracecmd_file_states end_state); +int tracecmd_copy_buffer_descr(struct tracecmd_input *in_handle, + struct tracecmd_output *out_handle); void tracecmd_set_flag(struct tracecmd_input *handle, int flag); void tracecmd_clear_flag(struct tracecmd_input *handle, int flag); unsigned long tracecmd_get_flags(struct tracecmd_input *handle); diff --git a/lib/trace-cmd/trace-input.c b/lib/trace-cmd/trace-input.c index 289102bf..7d0a41c2 100644 --- a/lib/trace-cmd/trace-input.c +++ b/lib/trace-cmd/trace-input.c @@ -4173,6 +4173,44 @@ int tracecmd_copy_headers(struct tracecmd_input *in_handle, return ret < 0 ? -1 : 0; } +static int skip_buffers(struct tracecmd_input *handle) +{ + unsigned long long offset; + unsigned int count; + char *bname; + int i; + + if (read4(handle, &count) < 0) + return -1; + + for (i = 0; i < count; i++) { + if (read8(handle, &offset) < 0) + return -1; + bname = read_string(handle); + free(bname); + } + + handle->file_state = TRACECMD_FILE_BUFERS; + return 0; +} + +int tracecmd_copy_buffer_descr(struct tracecmd_input *in_handle, + struct tracecmd_output *out_handle) +{ + int i; + + if (in_handle->file_version >= 7) { + if (!check_in_state(in_handle, TRACECMD_FILE_BUFERS)) + return -1; + skip_buffers(in_handle); + } + for (i = 0; i < in_handle->nr_buffers; i++) + tracecmd_add_buffer_description(out_handle, + in_handle->buffers[i].name, 0); + + return tracecmd_write_buffers_description(out_handle); +} + /** * tracecmd_record_at_buffer_start - return true if record is first on subbuffer * @handle: input handle for the trace.dat file diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index b4bdaeeb..ca9024bf 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -1952,6 +1952,9 @@ struct tracecmd_output *tracecmd_copy(struct tracecmd_input *ihandle, const char fstate = state > TRACECMD_FILE_CPU_COUNT ? TRACECMD_FILE_CPU_COUNT : state; if (tracecmd_copy_headers(ihandle, handle, 0, fstate) < 0) goto out_free; + if (state >= TRACECMD_FILE_BUFERS && + tracecmd_copy_buffer_descr(ihandle, handle) < 0) + goto out_free; /* The file is all ready to have cpu data attached */ return handle; -- 2.31.1
Extend the tracecmd_copy() API to support copying of trace options section from input to output trace handler. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- .../include/private/trace-cmd-private.h | 2 + lib/trace-cmd/include/trace-cmd-local.h | 1 + lib/trace-cmd/trace-input.c | 80 +++++++++++++++++++ lib/trace-cmd/trace-output.c | 8 ++ 4 files changed, 91 insertions(+) diff --git a/lib/trace-cmd/include/private/trace-cmd-private.h b/lib/trace-cmd/include/private/trace-cmd-private.h index 3735586d..595ba688 100644 --- a/lib/trace-cmd/include/private/trace-cmd-private.h +++ b/lib/trace-cmd/include/private/trace-cmd-private.h @@ -176,6 +176,8 @@ int tracecmd_copy_headers(struct tracecmd_input *in_handle, enum tracecmd_file_states end_state); int tracecmd_copy_buffer_descr(struct tracecmd_input *in_handle, struct tracecmd_output *out_handle); +int tracecmd_copy_options(struct tracecmd_input *in_handle, + struct tracecmd_output *out_handle); void tracecmd_set_flag(struct tracecmd_input *handle, int flag); void tracecmd_clear_flag(struct tracecmd_input *handle, int flag); unsigned long tracecmd_get_flags(struct tracecmd_input *handle); diff --git a/lib/trace-cmd/include/trace-cmd-local.h b/lib/trace-cmd/include/trace-cmd-local.h index 8fb391fc..2e5682c0 100644 --- a/lib/trace-cmd/include/trace-cmd-local.h +++ b/lib/trace-cmd/include/trace-cmd-local.h @@ -52,6 +52,7 @@ void in_uncompress_reset(struct tracecmd_input *handle); int in_uncompress_block(struct tracecmd_input *handle); void out_set_file_state(struct tracecmd_output *handle, int new_state); +void out_save_options_offset(struct tracecmd_output *handle); int write_buffers_description_v7(struct tracecmd_output *handle); int write_buffers_description_v6(struct tracecmd_output *handle); diff --git a/lib/trace-cmd/trace-input.c b/lib/trace-cmd/trace-input.c index 7d0a41c2..794f8ea3 100644 --- a/lib/trace-cmd/trace-input.c +++ b/lib/trace-cmd/trace-input.c @@ -4211,6 +4211,86 @@ int tracecmd_copy_buffer_descr(struct tracecmd_input *in_handle, return tracecmd_write_buffers_description(out_handle); } +static int copy_options(struct tracecmd_input *in_handle, struct tracecmd_output *out_handle) +{ + unsigned short option; + unsigned short en2; + unsigned int size; + unsigned int en4; + + if (in_uncompress_block(in_handle)) + return -1; + out_save_options_offset(out_handle); + out_compression_start(out_handle); + for (;;) { + if (do_read_check(in_handle, &option, 2)) + goto error; + en2 = tep_read_number(in_handle->pevent, &option, 2); + if (en2 == TRACECMD_OPTION_DONE) + break; + /* next 4 bytes is the size of the option */ + if (do_read_check(in_handle, &size, 4)) + goto error; + en4 = tep_read_number(in_handle->pevent, &size, 4); + /* Do not copy buffers description, as there is a file specific offset */ + if (en2 == TRACECMD_OPTION_BUFFER) { + /* Skip the option */ + do_lseek(in_handle, en4, SEEK_CUR); + continue; + } + + if (do_write_check(out_handle, &option, 2)) + goto error; + if (do_write_check(out_handle, &size, 4)) + goto error; + if (read_copy_data(in_handle, en4, out_handle)) + goto error; + } + if (do_write_check(out_handle, &option, 2)) + goto error; + in_uncompress_reset(in_handle); + if (out_compression_end(out_handle)) + goto error; + in_handle->file_state = TRACECMD_FILE_OPTIONS; + out_set_file_state(out_handle, in_handle->file_state); + /* Append local options */ + return tracecmd_append_options(out_handle); +error: + out_compression_reset(out_handle); + in_uncompress_reset(in_handle); + return 0; +} + +int tracecmd_copy_options(struct tracecmd_input *in_handle, + struct tracecmd_output *out_handle) +{ + off64_t offset; + char buf[10]; + + if (!check_in_state(in_handle, TRACECMD_FILE_OPTIONS) || + !check_out_state(out_handle, TRACECMD_FILE_OPTIONS)) + return -1; + + /* Save where we currently are */ + offset = lseek64(in_handle->fd, 0, SEEK_CUR); + + if (do_read_check(in_handle, buf, 10)) + return -1; + /* check if this handles options */ + if (strncmp(buf, "options", 7) == 0) { + if (do_write_check(out_handle, "options ", 10)) + return -1; + if (copy_options(in_handle, out_handle) < 0) + return -1; + } else { + if (lseek64(in_handle->fd, offset, SEEK_SET) == (off_t)-1) + return -1; + in_handle->file_state = TRACECMD_FILE_OPTIONS; + out_set_file_state(out_handle, in_handle->file_state); + } + return 0; +} + /** * tracecmd_record_at_buffer_start - return true if record is first on subbuffer * @handle: input handle for the trace.dat file diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index ca9024bf..95ca483c 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -1955,6 +1955,9 @@ struct tracecmd_output *tracecmd_copy(struct tracecmd_input *ihandle, const char if (state >= TRACECMD_FILE_BUFERS && tracecmd_copy_buffer_descr(ihandle, handle) < 0) goto out_free; + if (state >= TRACECMD_FILE_OPTIONS && + tracecmd_copy_options(ihandle, handle) < 0) + goto out_free; /* The file is all ready to have cpu data attached */ return handle; @@ -1974,6 +1977,11 @@ __hidden bool check_out_state(struct tracecmd_output *handle, int new_state) return check_file_state(handle->file_version, handle->file_state, new_state); } +__hidden void out_save_options_offset(struct tracecmd_output *handle) +{ + handle->options_start = lseek64(handle->fd, 0, SEEK_CUR); +} + /** * tracecmd_get_out_file_version - return the trace.dat file version * @handle: output handle for the trace.dat file -- 2.31.1
Refactored the internal library logic for writing trace CPU data in the trace file. The existing logic copies trace data from a temporary file into the trace file. In order to reuse the code, modify it to support copying the trace data not only from temporary file, but from given fd at given offset. These changes will be used in extending the tracecmd_copy API. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/include/trace-cmd-local.h | 13 +++ lib/trace-cmd/trace-output.c | 142 +++++++++++++++++------- 2 files changed, 114 insertions(+), 41 deletions(-) diff --git a/lib/trace-cmd/include/trace-cmd-local.h b/lib/trace-cmd/include/trace-cmd-local.h index 2e5682c0..970087c7 100644 --- a/lib/trace-cmd/include/trace-cmd-local.h +++ b/lib/trace-cmd/include/trace-cmd-local.h @@ -53,10 +53,23 @@ int in_uncompress_block(struct tracecmd_input *handle); void out_set_file_state(struct tracecmd_output *handle, int new_state); void out_save_options_offset(struct tracecmd_output *handle); +unsigned long long out_copy_fd_compress(struct tracecmd_output *handle, + int fd, unsigned long long max, + unsigned long long *write_size); + int write_buffers_description_v7(struct tracecmd_output *handle); int write_buffers_description_v6(struct tracecmd_output *handle); long long do_write_check(struct tracecmd_output *handle, const void *data, long long size); +struct cpu_data_source { + int fd; + int size; + off64_t offset; +}; + +int out_write_cpu_data(struct tracecmd_output *handle, int cpus, + struct cpu_data_source *data); + #endif /* _TRACE_CMD_LOCAL_H */ diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index 95ca483c..e5da4cdb 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -286,18 +286,26 @@ static unsigned long get_size(const char *file) return size; } -static tsize_t copy_file_fd(struct tracecmd_output *handle, int fd) +static tsize_t copy_file_fd(struct tracecmd_output *handle, int fd, unsigned long long max) { + tsize_t rsize = 0; tsize_t size = 0; char buf[BUFSIZ]; stsize_t r; do { - r = read(fd, buf, BUFSIZ); + if (max > 0 && (max - size) < BUFSIZ) + rsize = (max - size); + else + rsize = BUFSIZ; + + r = read(fd, buf, rsize); if (r > 0) { size += r; if (do_write_check(handle, buf, r)) return 0; + if (max > 0 && size >= max) + break; } } while (r > 0); @@ -315,47 +323,61 @@ static tsize_t copy_file(struct tracecmd_output *handle, tracecmd_warning("Can't read '%s'", file); return 0; } - size = copy_file_fd(handle, fd); + size = copy_file_fd(handle, fd, 0); close(fd); return size; } #define COMPRESS_CHUNK_SIZE (1*1024*1024) -static tsize_t copy_file_compress(struct tracecmd_output *handle, - const char *file, unsigned long long *write_size) +__hidden unsigned long long out_copy_fd_compress(struct tracecmd_output *handle, + int fd, unsigned long long max, + unsigned long long *write_size) { unsigned long long rsize = 0; unsigned long long wsize = 0; - tsize_t size; + unsigned long long size; int ret; - int fd; - - fd = open(file, O_RDONLY); - if (fd < 0) { - tracecmd_warning("Can't read '%s'", file); - return 0; - } if (handle->file_version >= 7) { + rsize = max; ret = tracecmd_compress_copy_from(handle->compress, fd, COMPRESS_CHUNK_SIZE, &rsize, &wsize); - if (ret < 0) { - tracecmd_warning("Can't compress '%s'", file); - close(fd); + if (ret < 0) return 0; - } + size = rsize; - *write_size = wsize; + if (write_size) + *write_size = wsize; } else { - size = copy_file_fd(handle, fd); - *write_size = size; + size = copy_file_fd(handle, fd, max); + if (write_size) + *write_size = size; } - close(fd); return size; } +static tsize_t copy_file_compress(struct tracecmd_output *handle, + const char *file, unsigned long long *write_size) +{ + int ret; + int fd; + + fd = open(file, O_RDONLY); + if (fd < 0) { + tracecmd_warning("Can't read '%s'", file); + return 0; + } + + ret = out_copy_fd_compress(handle, fd, 0, write_size); + if (!ret) + tracecmd_warning("Can't compress '%s'", file); + + close(fd); + return ret; +} + /* * Finds the path to the debugfs/tracing * Allocates the string and stores it. @@ -467,7 +489,7 @@ static int read_header_files(struct tracecmd_output *handle) endian8 = convert_endian_8(handle, size); if (do_write_check(handle, &endian8, 8)) goto out_close; - check_size = copy_file_fd(handle, fd); + check_size = copy_file_fd(handle, fd, 0); close(fd); if (size != check_size) { tracecmd_warning("wrong size for '%s' size=%lld read=%lld", path, size, check_size); @@ -493,7 +515,7 @@ static int read_header_files(struct tracecmd_output *handle) endian8 = convert_endian_8(handle, size); if (do_write_check(handle, &endian8, 8)) goto out_close; - check_size = copy_file_fd(handle, fd); + check_size = copy_file_fd(handle, fd, 0); close(fd); if (size != check_size) { tracecmd_warning("wrong size for '%s'", path); @@ -1671,8 +1693,8 @@ struct data_file_write { off64_t doffset; }; -int tracecmd_write_cpu_data(struct tracecmd_output *handle, - int cpus, char * const *cpu_data_files) +__hidden int out_write_cpu_data(struct tracecmd_output *handle, + int cpus, struct cpu_data_source *data) { struct data_file_write *data_files = NULL; off64_t offset; @@ -1680,8 +1702,6 @@ int tracecmd_write_cpu_data(struct tracecmd_output *handle, unsigned long long read_size; unsigned long long write_size; char *clock = NULL; - char *file; - struct stat st; int ret; int i; @@ -1705,14 +1725,7 @@ int tracecmd_write_cpu_data(struct tracecmd_output *handle, /* Write 0 for trace data offset and size and store offsets of these fields */ for (i = 0; i < cpus; i++) { - file = cpu_data_files[i]; - ret = stat(file, &st); - if (ret < 0) { - tracecmd_warning("can not stat '%s'", file); - goto out_free; - } - data_files[i].file_size = st.st_size; - + data_files[i].file_size = data[i].size; endian8 = 0; data_files[i].doffset = lseek64(handle->fd, 0, SEEK_CUR); if (do_write_check(handle, &endian8, 8)) @@ -1736,13 +1749,19 @@ int tracecmd_write_cpu_data(struct tracecmd_output *handle, if (!tracecmd_get_quiet(handle)) fprintf(stderr, "CPU%d data recorded at offset=0x%llx\n", i, (unsigned long long) data_files[i].data_offset); - - read_size = copy_file_compress(handle, cpu_data_files[i], &write_size); - if (read_size != data_files[i].file_size) { - errno = EINVAL; - tracecmd_warning("did not match size of %lld to %lld", - read_size, data_files[i].file_size); + offset = lseek64(data[i].fd, data[i].offset, SEEK_SET); + if (offset == (off64_t)-1) goto out_free; + if (data[i].size) { + read_size = out_copy_fd_compress(handle, data[i].fd, data[i].size, &write_size); + if (read_size != data_files[i].file_size) { + errno = EINVAL; + tracecmd_warning("did not match size of %lld to %lld", + read_size, data_files[i].file_size); + goto out_free; + } + } else { + write_size = 0; } /* Write the real CPU data offset inthe file */ offset = lseek64(handle->fd, data_files[i].doffset, SEEK_SET); @@ -1775,6 +1794,47 @@ int tracecmd_write_cpu_data(struct tracecmd_output *handle, return -1; } +int tracecmd_write_cpu_data(struct tracecmd_output *handle, + int cpus, char * const *cpu_data_files) +{ + struct cpu_data_source *data; + struct stat st; + int ret; + int i; + + data = calloc(cpus, sizeof(struct cpu_data_source)); + if (!data) + return -1; + for (i = 0; i < cpus; i++) + data[i].fd = -1; + for (i = 0; i < cpus; i++) { + ret = stat(cpu_data_files[i], &st); + if (ret < 0) { + tracecmd_warning("can not stat '%s'", cpu_data_files[i]); + break; + } + data[i].fd = open(cpu_data_files[i], O_RDONLY); + if (data[i].fd < 0) { + tracecmd_warning("Can't read '%s'", data[i].fd); + break; + } + + data[i].size = st.st_size; + data[i].offset = 0; + } + + if (i < cpus) + ret = -1; + else + ret = out_write_cpu_data(handle, cpus, data); + for (i = 0; i < cpus; i++) { + if (data[i].fd >= 0) + close(data[i].fd); + } + free(data); + return ret; +} + int tracecmd_append_cpu_data(struct tracecmd_output *handle, int cpus, char * const *cpu_data_files) { -- 2.31.1
Added a helper function, internal to the library, for updating CPU instance offset in the trace file. This allows to reuse the logic inside the library, will be used in the tracecmd_copy API. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/include/trace-cmd-local.h | 2 ++ lib/trace-cmd/trace-output.c | 11 +++++++++-- 2 files changed, 11 insertions(+), 2 deletions(-) diff --git a/lib/trace-cmd/include/trace-cmd-local.h b/lib/trace-cmd/include/trace-cmd-local.h index 970087c7..3cb171ea 100644 --- a/lib/trace-cmd/include/trace-cmd-local.h +++ b/lib/trace-cmd/include/trace-cmd-local.h @@ -72,4 +72,6 @@ struct cpu_data_source { int out_write_cpu_data(struct tracecmd_output *handle, int cpus, struct cpu_data_source *data); +int out_update_buffer_cpu_offset(struct tracecmd_output *handle, const char *name); + #endif /* _TRACE_CMD_LOCAL_H */ diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index e5da4cdb..54134d03 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -1853,8 +1853,7 @@ int tracecmd_append_cpu_data(struct tracecmd_output *handle, return tracecmd_write_cpu_data(handle, cpus, cpu_data_files); } -int tracecmd_append_buffer_cpu_data(struct tracecmd_output *handle, - const char *name, int cpus, char * const *cpu_data_files) +__hidden int out_update_buffer_cpu_offset(struct tracecmd_output *handle, const char *name) { tsize_t b_offset; tsize_t offset; @@ -1883,6 +1882,14 @@ int tracecmd_append_buffer_cpu_data(struct tracecmd_output *handle, tracecmd_warning("could not seek to %lld\n", offset); return -1; } + return 0; +} + +int tracecmd_append_buffer_cpu_data(struct tracecmd_output *handle, + const char *name, int cpus, char * const *cpu_data_files) +{ + if (out_update_buffer_cpu_offset(handle, name)) + return -1; return tracecmd_write_cpu_data(handle, cpus, cpu_data_files); } -- 2.31.1
Extend the tracecmd_copy() API to support copying of trace data from input to output trace handler. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- .../include/private/trace-cmd-private.h | 2 + lib/trace-cmd/trace-input.c | 101 ++++++++++++++++++ lib/trace-cmd/trace-output.c | 3 + 3 files changed, 106 insertions(+) diff --git a/lib/trace-cmd/include/private/trace-cmd-private.h b/lib/trace-cmd/include/private/trace-cmd-private.h index 595ba688..79fef884 100644 --- a/lib/trace-cmd/include/private/trace-cmd-private.h +++ b/lib/trace-cmd/include/private/trace-cmd-private.h @@ -178,6 +178,8 @@ int tracecmd_copy_buffer_descr(struct tracecmd_input *in_handle, struct tracecmd_output *out_handle); int tracecmd_copy_options(struct tracecmd_input *in_handle, struct tracecmd_output *out_handle); +int tracecmd_copy_trace_data(struct tracecmd_input *in_handle, + struct tracecmd_output *out_handle); void tracecmd_set_flag(struct tracecmd_input *handle, int flag); void tracecmd_clear_flag(struct tracecmd_input *handle, int flag); unsigned long tracecmd_get_flags(struct tracecmd_input *handle); diff --git a/lib/trace-cmd/trace-input.c b/lib/trace-cmd/trace-input.c index 794f8ea3..2544d825 100644 --- a/lib/trace-cmd/trace-input.c +++ b/lib/trace-cmd/trace-input.c @@ -4291,6 +4291,107 @@ int tracecmd_copy_options(struct tracecmd_input *in_handle, return 0; } +static int copy_trace_latency(struct tracecmd_input *in_handle, + struct tracecmd_output *out_handle) +{ + int fd; + + if (do_write_check(out_handle, "latency ", 10)) + return -1; + + if (in_handle->lat_cfd >= 0) + fd = in_handle->lat_cfd; + else + fd = in_handle->fd; + + if (!out_copy_fd_compress(out_handle, fd, 0, NULL)) + return -1; + + out_set_file_state(out_handle, TRACECMD_FILE_CPU_LATENCY); + return 0; +} + +static int copy_trace_flyrecord_data(struct tracecmd_input *in_handle, + struct tracecmd_output *out_handle) +{ + struct cpu_data_source *data; + int ret; + int i; + + data = calloc(in_handle->cpus, sizeof(struct cpu_data_source)); + if (!data) + return -1; + for (i = 0; i < in_handle->cpus; i++) { + data[i].size = in_handle->cpu_data[i].file_size; + if (in_handle->cpu_data[i].cfd >= 0) { + data[i].fd = in_handle->cpu_data[i].cfd; + data[i].offset = 0; + } else { + data[i].fd = in_handle->fd; + data[i].offset = in_handle->cpu_data[i].file_offset; + } + } + ret = out_write_cpu_data(out_handle, in_handle->cpus, data); + + return ret; +} + +static int copy_trace_flyrecord(struct tracecmd_input *in_handle, + struct tracecmd_output *out_handle) +{ + struct tracecmd_input *instance; + const char *name; + int ret; + int i; + + /* top instance */ + ret = copy_trace_flyrecord_data(in_handle, out_handle); + if (ret) + return ret; + + for (i = 0; i < in_handle->nr_buffers; i++) { + name = tracecmd_buffer_instance_name(in_handle, i); + if (!name) + continue; + instance = tracecmd_buffer_instance_handle(in_handle, i); + if (!instance) + continue; + if (!tracecmd_get_quiet(out_handle)) + fprintf(stderr, "\nBuffer: %s\n\n", name); + + if (!out_update_buffer_cpu_offset(out_handle, name)) + copy_trace_flyrecord_data(instance, out_handle); + + tracecmd_close(instance); + } + + return 0; +} + +int tracecmd_copy_trace_data(struct tracecmd_input *in_handle, + struct tracecmd_output *out_handle) +{ + char buf[10]; + + if (!check_in_state(in_handle, TRACECMD_FILE_CPU_FLYRECORD) || + !check_out_state(out_handle, TRACECMD_FILE_CPU_FLYRECORD)) + return -1; + + tracecmd_set_out_clock(out_handle, in_handle->trace_clock); + + if (do_read_check(in_handle, buf, 10)) + return -1; + if (strncmp(buf, "latency", 7) == 0) { + in_handle->file_state = TRACECMD_FILE_CPU_LATENCY; + return copy_trace_latency(in_handle, out_handle); + } else if (strncmp(buf, "flyrecord", 9) == 0) { + in_handle->file_state = TRACECMD_FILE_CPU_FLYRECORD; + return copy_trace_flyrecord(in_handle, out_handle); + } + + return -1; +} + /** * tracecmd_record_at_buffer_start - return true if record is first on subbuffer * @handle: input handle for the trace.dat file diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index 54134d03..377c5863 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -2025,6 +2025,9 @@ struct tracecmd_output *tracecmd_copy(struct tracecmd_input *ihandle, const char if (state >= TRACECMD_FILE_OPTIONS && tracecmd_copy_options(ihandle, handle) < 0) goto out_free; + if (state >= TRACECMD_FILE_CPU_LATENCY && + tracecmd_copy_trace_data(ihandle, handle) < 0) + goto out_free; /* The file is all ready to have cpu data attached */ return handle; -- 2.31.1
Added compression parameter to the tracecmd_copy() API, which allows to select the compression of the created output file. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- lib/trace-cmd/include/private/trace-cmd-private.h | 2 +- lib/trace-cmd/trace-output.c | 11 +++++++---- tracecmd/trace-restore.c | 2 +- tracecmd/trace-split.c | 2 +- 4 files changed, 10 insertions(+), 7 deletions(-) diff --git a/lib/trace-cmd/include/private/trace-cmd-private.h b/lib/trace-cmd/include/private/trace-cmd-private.h index 79fef884..c8f0ee93 100644 --- a/lib/trace-cmd/include/private/trace-cmd-private.h +++ b/lib/trace-cmd/include/private/trace-cmd-private.h @@ -316,7 +316,7 @@ int tracecmd_append_options(struct tracecmd_output *handle); void tracecmd_output_close(struct tracecmd_output *handle); void tracecmd_output_free(struct tracecmd_output *handle); struct tracecmd_output *tracecmd_copy(struct tracecmd_input *ihandle, const char *file, - enum tracecmd_file_states state); + enum tracecmd_file_states state, const char *compression); int tracecmd_write_cpu_data(struct tracecmd_output *handle, int cpus, char * const *cpu_data_files); diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c index 377c5863..dae11ed3 100644 --- a/lib/trace-cmd/trace-output.c +++ b/lib/trace-cmd/trace-output.c @@ -1999,20 +1999,23 @@ struct tracecmd_output *tracecmd_create_init_file_override(const char *output_fi * @ihandle: input handle of the trace.dat file to copy * @file: the trace.dat file to create * @state: what data will be copied from the source handle + * @compression: compression of the output file, can be one of: + * NULL - inherit compression from the input file + * "any" - compress the output file with the best available algorithm + * "none" - do not compress the output file + * algorithm_name - compress the output file with specified algorithm * * Reads the header information and creates a new trace data file * with the same characteristics (events and all) and returns * tracecmd_output handle to this new file. */ struct tracecmd_output *tracecmd_copy(struct tracecmd_input *ihandle, const char *file, - enum tracecmd_file_states state) + enum tracecmd_file_states state, const char *compression) { enum tracecmd_file_states fstate; struct tracecmd_output *handle; - const char *compr_name = NULL; - tracecmd_get_file_compress_proto(ihandle, &compr_name, NULL); - handle = create_file(file, ihandle, NULL, NULL, &all_event_list, compr_name); + handle = create_file(file, ihandle, NULL, NULL, &all_event_list, compression); if (!handle) return NULL; diff --git a/tracecmd/trace-restore.c b/tracecmd/trace-restore.c index 7d439fd2..58913fa7 100644 --- a/tracecmd/trace-restore.c +++ b/tracecmd/trace-restore.c @@ -125,7 +125,7 @@ void trace_restore (int argc, char **argv) if (tracecmd_read_headers(ihandle, TRACECMD_FILE_CMD_LINES) < 0) die("error reading file %s headers", input); - handle = tracecmd_copy(ihandle, output, TRACECMD_FILE_CMD_LINES); + handle = tracecmd_copy(ihandle, output, TRACECMD_FILE_CMD_LINES, NULL); tracecmd_close(ihandle); } else handle = tracecmd_create_init_file(output, NULL); diff --git a/tracecmd/trace-split.c b/tracecmd/trace-split.c index 401e1d29..b98e6fc9 100644 --- a/tracecmd/trace-split.c +++ b/tracecmd/trace-split.c @@ -345,7 +345,7 @@ static double parse_file(struct tracecmd_input *handle, dir = dirname(output); base = basename(output); - ohandle = tracecmd_copy(handle, output_file, TRACECMD_FILE_CMD_LINES); + ohandle = tracecmd_copy(handle, output_file, TRACECMD_FILE_CMD_LINES, NULL); cpus = tracecmd_cpus(handle); cpu_data = malloc(sizeof(*cpu_data) * cpus); -- 2.31.1
A new trace-cmd subcommand is added, used to convert trace files: trace-cmd convert -i <input file> -o <output file> --compression <out file compression> The command reads the given input trace file and creates output trace file with given compression. The --compression parameter is optional: - no --compression parameter: the output file inherits the compression from the input file. --compression none - the output file is not compressed. --compression any - use the best available compression algorithm to compress the output file --compression <algorithm> - use the specified compression algorithm, if available, to compress the output file. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- tracecmd/Makefile | 1 + tracecmd/include/trace-local.h | 2 + tracecmd/trace-cmd.c | 1 + tracecmd/trace-convert.c | 94 ++++++++++++++++++++++++++++++++++ tracecmd/trace-usage.c | 12 +++++ 5 files changed, 110 insertions(+) create mode 100644 tracecmd/trace-convert.c diff --git a/tracecmd/Makefile b/tracecmd/Makefile index 80c69bbb..35086b71 100644 --- a/tracecmd/Makefile +++ b/tracecmd/Makefile @@ -36,6 +36,7 @@ TRACE_CMD_OBJS += trace-usage.o TRACE_CMD_OBJS += trace-dump.o TRACE_CMD_OBJS += trace-clear.o TRACE_CMD_OBJS += trace-vm.o +TRACE_CMD_OBJS += trace-convert.o ifeq ($(VSOCK_DEFINED), 1) TRACE_CMD_OBJS += trace-agent.o diff --git a/tracecmd/include/trace-local.h b/tracecmd/include/trace-local.h index e9a0aea8..13dab44c 100644 --- a/tracecmd/include/trace-local.h +++ b/tracecmd/include/trace-local.h @@ -110,6 +110,8 @@ void trace_usage(int argc, char **argv); void trace_dump(int argc, char **argv); +void trace_convert(int argc, char **argv); + int trace_record_agent(struct tracecmd_msg_handle *msg_handle, int cpus, int *fds, int argc, char **argv, bool use_fifos, diff --git a/tracecmd/trace-cmd.c b/tracecmd/trace-cmd.c index 71c8f6d6..e6274c35 100644 --- a/tracecmd/trace-cmd.c +++ b/tracecmd/trace-cmd.c @@ -130,6 +130,7 @@ struct command commands[] = { {"list", trace_list}, {"help", trace_usage}, {"dump", trace_dump}, + {"convert", trace_convert}, {"-h", trace_usage}, }; diff --git a/tracecmd/trace-convert.c b/tracecmd/trace-convert.c new file mode 100644 index 00000000..555e0c0a --- /dev/null +++ b/tracecmd/trace-convert.c @@ -0,0 +1,94 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * Copyright (C) 2021, VMware, Tzvetomir Stoyanov <tz.stoyanov@gmail.com> + */ +#include <stdlib.h> +#include <unistd.h> +#include <getopt.h> +#include <fcntl.h> +#include <errno.h> + +#include "trace-local.h" +#include "trace-cmd.h" +#include "trace-cmd-private.h" + +static void convert_file(const char *in, const char *out, char *compr) +{ + struct tracecmd_input *ihandle; + struct tracecmd_output *ohandle; + + ihandle = tracecmd_open(in, 0); + if (!ihandle) + die("error reading %s", in); + + ohandle = tracecmd_copy(ihandle, out, TRACECMD_FILE_CPU_FLYRECORD, compr); + if (!ohandle) + die("error writing %s", out); + tracecmd_output_close(ohandle); + tracecmd_close(ihandle); +} + +enum { + OPT_comporession = 255, +}; + +void trace_convert(int argc, char **argv) +{ + char *input_file = NULL; + char *output_file = NULL; + char *compression = NULL; + int c; + + if (argc < 2) + usage(argv); + + if (strcmp(argv[1], "convert") != 0) + usage(argv); + for (;;) { + int option_index = 0; + static struct option long_options[] = { + {"compression", required_argument, NULL, OPT_comporession}, + {"help", no_argument, NULL, '?'}, + {NULL, 0, NULL, 0} + }; + + c = getopt_long (argc-1, argv+1, "+hi:o:", long_options, &option_index); + if (c == -1) + break; + switch (c) { + case 'i': + if (input_file) + die("Only one input file is supported, %s already set", input_file); + input_file = optarg; + break; + case 'o': + if (output_file) + die("Only one output file is supported, %s already set", output_file); + output_file = optarg; + break; + case OPT_comporession: + if (strcmp(optarg, "any") && strcmp(optarg, "none") && + !tracecmd_compress_is_supported(optarg, NULL)) + die("Compression algorithm %s is not supported", optarg); + compression = optarg; + break; + case 'h': + case '?': + default: + usage(argv); + } + } + + if ((argc - optind) >= 2) { + if (output_file) + usage(argv); + output_file = argv[optind + 1]; + } + + if (!input_file) + input_file = DEFAULT_INPUT_FILE; + if (!output_file) + usage(argv); + + convert_file(input_file, output_file, compression); +} diff --git a/tracecmd/trace-usage.c b/tracecmd/trace-usage.c index c70f9919..d0d12e87 100644 --- a/tracecmd/trace-usage.c +++ b/tracecmd/trace-usage.c @@ -392,6 +392,18 @@ static struct usage_help usage_help[] = { " --clock trace clock, saved in the file\n" " -h, --help show usage information\n" }, + { + "convert", + "convert trace file to different version", + " %s convert [options]\n" + " -i input file, default is trace.dat\n" + " -o output file, mandatory parameter.\n" + " The output file can be specified also as last argument of the command\n" + " --compression compress the trace output file, one of these strings can be passed:\n" + " any - auto select the best available compression algorithm\n" + " none - do not compress the trace file\n" + " name - the name of the desired compression algorithms\n" + " available algorithms can be listed with trace-cmd list -c\n" }, { NULL, NULL, NULL } -- 2.31.1
Added documentation of the 'trace-cmd record --compression' parameter in the man page. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- Documentation/trace-cmd/trace-cmd-record.1.txt | 10 ++++++++++ 1 file changed, 10 insertions(+) diff --git a/Documentation/trace-cmd/trace-cmd-record.1.txt b/Documentation/trace-cmd/trace-cmd-record.1.txt index 5d063cc2..c5a7fb75 100644 --- a/Documentation/trace-cmd/trace-cmd-record.1.txt +++ b/Documentation/trace-cmd/trace-cmd-record.1.txt @@ -367,6 +367,16 @@ OPTIONS Traces are extracted by busy waiting, which will hog the CPUs, so only use when really needed. +*--compression*:: + Compression of the trace output file, one of these strings can be passed: + + 'any' - auto select the best available compression algorithm + + 'none' - do not compress the trace file + + 'name' - the name of the desired compression algorithms. Available algorithms can be listed with + trace-cmd list -c + EXAMPLES -------- -- 2.31.1
Added new man page, documenting the "trace-cmd convert" command. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- .../trace-cmd/trace-cmd-convert.1.txt | 60 +++++++++++++++++++ Documentation/trace-cmd/trace-cmd.1.txt | 4 +- 2 files changed, 63 insertions(+), 1 deletion(-) create mode 100644 Documentation/trace-cmd/trace-cmd-convert.1.txt diff --git a/Documentation/trace-cmd/trace-cmd-convert.1.txt b/Documentation/trace-cmd/trace-cmd-convert.1.txt new file mode 100644 index 00000000..e8d65843 --- /dev/null +++ b/Documentation/trace-cmd/trace-cmd-convert.1.txt @@ -0,0 +1,60 @@ +TRACE-CMD-CONVERT(1) +=================== + +NAME +---- +trace-cmd-convert - convert trace files + +SYNOPSIS +-------- +*trace-cmd convert* ['OPTIONS'] ['output-file'] + +DESCRIPTION +----------- +The trace-cmd(1) convert command converts trace file. It reads the input file and copies the data +into an output file. The output file may be in different format, depending on the command line +arguments. + +OPTIONS +------- +*-i* 'input-file':: + By default, trace-cmd convert will read the file 'trace.dat'. But the *-i* + option open up the given 'input-file' instead. +*-o* 'out-file':: + The name of the output file, this parameter is mandatory. Note, the output file may also be + specified as the last item on the command line. +*--compression*:: + Compression of the trace output file, one of these strings can be passed: + + 'any' - auto select the best available compression algorithm + + 'none' - do not compress the trace file + + 'name' - the name of the desired compression algorithms. Available algorithms can be listed with + trace-cmd list -c + +*--help*:: + Print usage information. + +EXAMPLES +-------- + +# trace-cmd convert --compression any trace_compress.dat + +SEE ALSO +-------- +trace-cmd(1), trace-cmd.dat(1) + +AUTHOR +------ +*Steven Rostedt* <rostedt@goodmis.org>, author of *trace-cmd*. +*Tzvetomir Stoyanov* <tz.stoyanov@gmail.com>, author of this man page. + +RESOURCES +--------- +https://git.kernel.org/pub/scm/utils/trace-cmd/trace-cmd.git/ + +COPYING +------- +Copyright \(C) 2021 VMware. Free use of this software is granted under +the terms of the GNU Public License (GPL). diff --git a/Documentation/trace-cmd/trace-cmd.1.txt b/Documentation/trace-cmd/trace-cmd.1.txt index b330b4fe..27c6e973 100644 --- a/Documentation/trace-cmd/trace-cmd.1.txt +++ b/Documentation/trace-cmd/trace-cmd.1.txt @@ -58,6 +58,8 @@ COMMANDS check-events - parse format strings for all trace events and return whether all formats are parseable + convert - convert trace files + OPTIONS ------- @@ -70,7 +72,7 @@ SEE ALSO -------- trace-cmd-record(1), trace-cmd-report(1), trace-cmd-hist(1), trace-cmd-start(1), trace-cmd-stop(1), trace-cmd-extract(1), trace-cmd-reset(1), -trace-cmd-restore(1), trace-cmd-stack(1), +trace-cmd-restore(1), trace-cmd-stack(1), trace-cmd-convert(1), trace-cmd-split(1), trace-cmd-list(1), trace-cmd-listen(1), trace-cmd.dat(5), trace-cmd-check-events(1) trace-cmd-stat(1) -- 2.31.1
Added support for trace-cmd convert and compression parameters in bash completion. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- tracecmd/trace-cmd.bash | 37 +++++++++++++++++++++++++++++++++++++ 1 file changed, 37 insertions(+) diff --git a/tracecmd/trace-cmd.bash b/tracecmd/trace-cmd.bash index b01c7a07..6639c143 100644 --- a/tracecmd/trace-cmd.bash +++ b/tracecmd/trace-cmd.bash @@ -64,6 +64,13 @@ plugin_options() COMPREPLY=( $(compgen -W "${opts}" -- "${cur}") ) } +compression_param() +{ + local opts=$(trace-cmd list -c | grep -v 'Supported' | cut -d "," -f1) + opts+=" any none " + COMPREPLY=( $(compgen -W "${opts}") ) +} + __trace_cmd_list_complete() { local prev=$1 @@ -181,6 +188,9 @@ __trace_cmd_record_complete() cmd_options record "$cur" fi ;; + --compression) + compression_param + ;; *) # stream start and profile do not show all options cmd_options record "$cur" @@ -222,6 +232,29 @@ __trace_cmd_dump_complete() esac } +__trace_cmd_convert_complete() +{ + local prev=$1 + local cur=$2 + shift 2 + local words=("$@") + + case "$prev" in + -i) + __show_files + ;; + -o) + __show_files + ;; + --compression) + compression_param + ;; + *) + cmd_options convert "$cur" + ;; + esac +} + __show_command_options() { local command="$1" @@ -298,6 +331,10 @@ _trace_cmd_complete() __trace_cmd_dump_complete "${prev}" "${cur}" ${words[@]} return 0 ;; + convert) + __trace_cmd_convert_complete "${prev}" "${cur}" ${words[@]} + return 0 + ;; *) __show_command_options "$w" "${prev}" "${cur}" ;; -- 2.31.1
Added documentation of the 'trace-cmd list -c' parameter in the man page. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- Documentation/trace-cmd/trace-cmd-list.1.txt | 3 +++ 1 file changed, 3 insertions(+) diff --git a/Documentation/trace-cmd/trace-cmd-list.1.txt b/Documentation/trace-cmd/trace-cmd-list.1.txt index a5c6b16c..b77e3460 100644 --- a/Documentation/trace-cmd/trace-cmd-list.1.txt +++ b/Documentation/trace-cmd/trace-cmd-list.1.txt @@ -71,6 +71,9 @@ OPTIONS List defined clocks that can be used with trace-cmd record -C. The one in brackets ([]) is the active clock. +*-c*:: + List the available trace file compression algorithms. + SEE ALSO -------- trace-cmd(1), trace-cmd-record(1), trace-cmd-report(1), trace-cmd-start(1), -- 2.31.1
Updated the trace.dat man page with the changes related to file version 7 and compression. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- Documentation/trace-cmd/trace-cmd.dat.5.txt | 56 ++++++++++++++++++--- 1 file changed, 50 insertions(+), 6 deletions(-) diff --git a/Documentation/trace-cmd/trace-cmd.dat.5.txt b/Documentation/trace-cmd/trace-cmd.dat.5.txt index 8d285353..e80d460e 100644 --- a/Documentation/trace-cmd/trace-cmd.dat.5.txt +++ b/Documentation/trace-cmd/trace-cmd.dat.5.txt @@ -52,6 +52,23 @@ INITIAL FORMAT The next 4 bytes are a 32-bit word that defines what the traced host machine page size was. + If the file version is 7 or greater, the compression header is + written next: + "name version\0" + where "name" and "version" are strings, name and version of the + compression algorithm used to compress the trace file. + +COMPRESSION FORMAT OF THE HEADER SECTIONS +----------------------------------------- + If the file version is 7 or greater, some header sections are compressed + with the compression algorithm, specified in the compression header. + The format of these compressed sections is: + <4 bytes> unsigned int, size of compressed data in the next block. + <4 bytes> unsigned int, size of uncompressed data. + <data> binary compressed data, with the specified size. + These sections must be uncompressed on reading. The described format of + the sections refers to the uncomperssed data. + HEADER INFO FORMAT ------------------ @@ -93,7 +110,8 @@ FTRACE EVENT FORMATS Directly after the header information comes the information about the Ftrace specific events. These are the events used by the Ftrace plugins - and are not enabled by the event tracing. + and are not enabled by the event tracing. If the file version is 7 or + greater, this section is compressed. The next 4 bytes contain a 32-bit word of the number of Ftrace event format files that are stored in the file. @@ -110,7 +128,8 @@ EVENT FORMATS ------------- Directly after the Ftrace formats comes the information about - the event layout. + the event layout. If the file version is 7 or greater, this section + is compressed. The next 4 bytes are a 32-bit word containing the number of event systems that are stored in the file. These are the @@ -137,7 +156,8 @@ KALLSYMS INFORMATION -------------------- Directly after the event formats comes the information of the mapping - of function addresses to the function names. + of function addresses to the function names. If the file version is 7 + or greater, this section is compressed. The next 4 bytes are a 32-bit word containing the size of the data holding the function mappings. @@ -154,6 +174,7 @@ TRACE_PRINTK INFORMATION store the format string outside the ring buffer. This information can be found in: debugfs/tracing/printk_formats + If the file version is 7 or greater, this section is compressed. The next 4 bytes are a 32-bit word containing the size of the data holding the printk formats. @@ -166,7 +187,8 @@ PROCESS INFORMATION ------------------- Directly after the trace_printk formats comes the information mapping - a PID to a process name. + a PID to a process name. If the file version is 7 or greater, this + section is compressed. The next 8 bytes contain a 64-bit word that holds the size of the data mapping the PID to a process name. @@ -193,10 +215,11 @@ REST OF TRACE-CMD HEADER "flyrecord\0" - If it is "options \0" then: + If it is "options \0" then follows a section with trace options. + If the file version is 7 or greater, this section is compressed. The next 2 bytes are a 16-bit word defining the current option. - If the the value is zero then there are no more options. + If the value is zero then there are no more options. Otherwise, the next 4 bytes contain a 32-bit word containing the option size. If the reader does not know how to handle the option @@ -206,6 +229,25 @@ REST OF TRACE-CMD HEADER The next option will be directly after the previous option, and the options ends with a zero in the option type field. +COMPRESSION FORMAT OF THE TRACE DATA +------------------------------------ + + If the file version is 7 or greater, the tarce data is compressed + with the compression algorithm, specified in the compression header. + The data is compressed in chunks. The size of one compression chunk + is defined when the file is written. The format of compressed trace + data is: + <4 bytes> unsigned int, count of chunks. + Follows the compressed chunks of givent count. For each chunk: + <4 bytes> unsigned int, size of compressed data in this chunk. + <4 bytes> unsigned int, size of uncompressed data. + <data> binary compressed data, with the specified size. + These chunks must be uncompressed on reading. The described format of + trace data refers to the uncomperssed data. + +TRACE DATA +---------- + The next 10 bytes after the options are one of the following: "latency \0" @@ -217,6 +259,7 @@ REST OF TRACE-CMD HEADER If the value is "latency \0", then the rest of the file is simply ASCII text that was taken from the target's: debugfs/tracing/trace + If the file version is 7 or greater, the latency data is compressed. If the value is "flyrecord\0", the following is present: @@ -232,6 +275,7 @@ REST OF TRACE-CMD HEADER CPU DATA -------- + If the file version is 7 or greater, the CPU data is compressed. The CPU data is located in the part of the file that is specified in the end of the header. Padding is placed between the header and the CPU data, placing the CPU data at a page aligned (target page) position -- 2.31.1
On Mon, 14 Jun 2021 10:49:47 +0300
"Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com> wrote:
> When reading a trace file, version of the file is ignored. This could
> case problems when bumping the version number because of changes in
> in the structure of the file. The old code should detect unsupported
> file version and should not try to read it.
> A new trace-cmd library API is added to check if version is supported:
> tracecmd_is_version_supported()
> Checks are added in the code to ensure not trying to read trace file
> from unsupported version.
After applying this patch, I get:
$ ./tracecmd/trace-cmd report
error reading header for trace.dat
-- Steve
On Mon, 21 Jun 2021 18:27:46 -0400
Steven Rostedt <rostedt@goodmis.org> wrote:
> After applying this patch, I get:
>
> $ ./tracecmd/trace-cmd report
> error reading header for trace.dat
Never mind, I tested on a leftover trace.dat that had version 7 already set!
So the patch did work :-)
-- Steve
On Mon, 14 Jun 2021 10:50:03 +0300 "Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com> wrote: > --- a/lib/trace-cmd/include/private/trace-cmd-private.h > +++ b/lib/trace-cmd/include/private/trace-cmd-private.h > @@ -112,6 +112,7 @@ enum tracecmd_file_states { > TRACECMD_FILE_PRINTK, > TRACECMD_FILE_CMD_LINES, > TRACECMD_FILE_CPU_COUNT, > + TRACECMD_FILE_BUFERS, Use the proper spelling of BUFFERS. Thanks, -- Steve > TRACECMD_FILE_OPTIONS, > TRACECMD_FILE_CPU_LATENCY, > TRACECMD_FILE_CPU_FLYRECORD,
On Mon, 14 Jun 2021 10:50:05 +0300
"Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com> wrote:
> Comperss the option section of the trace file. This section is not big
> currently and compressing it does not reduce significantly the size of
> the file. This could be useful in the future as new options can be
> added, storing a potentially huge amount of data.
I'm not sure we want to bother doing this. New options should never be
large, and if they are, then the option itself can be compressed when
created.
What good reason is there to compress the options?
I believe this is just adding more complexity than needed.
-- Steve
On Mon, 14 Jun 2021 10:50:06 +0300
"Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com> wrote:
> Refactored the internal logic of tracecmd_write_cpu_data() API to be
> suitable for adding trace data compression.
The change log should be more descriptive of what is being refactored.
-- Steve
On Mon, 14 Jun 2021 10:50:09 +0300
"Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com> wrote:
> When reading a trace.dat file of version 7, uncompress the trace data.
> The trace data for each CPU is uncompressed in a temporary file, located
> in /tmp directory with prefix "trace_cpu_data".
With large trace files, this will be an issue. Several systems setup the
/tmp directory as a ramfs file system (that is, it is locate in ram, and
not backed up on disk). If you have very large trace files, which you would
if you are going to bother compressing them, by uncompressing them into
/tmp, it could take up all the memory of the machine, or easily fill the
/tmp limit.
Simply uncompressing the entire trace data is not an option. The best we
can do is to uncompress on a as needed basis. That would require having
meta data that is stored to know what pages are compressed.
-- Steve
On Mon, 14 Jun 2021 10:50:17 +0300 "Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com> wrote: > Extend the tracecmd_copy() API to support copying of CPU count from > input to output trace hanlder. typo above. But there's no rationale for the change in the change log. It's simply saying what is being done, but not why it is being done. -- Steve > > Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> > --- >
On Mon, 14 Jun 2021 10:50:29 +0300 "Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com> wrote: > Updated the trace.dat man page with the changes related to file > version 7 and compression. > > Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> > --- > Documentation/trace-cmd/trace-cmd.dat.5.txt | 56 ++++++++++++++++++--- > 1 file changed, 50 insertions(+), 6 deletions(-) > > diff --git a/Documentation/trace-cmd/trace-cmd.dat.5.txt b/Documentation/trace-cmd/trace-cmd.dat.5.txt > index 8d285353..e80d460e 100644 > --- a/Documentation/trace-cmd/trace-cmd.dat.5.txt > +++ b/Documentation/trace-cmd/trace-cmd.dat.5.txt > @@ -52,6 +52,23 @@ INITIAL FORMAT > The next 4 bytes are a 32-bit word that defines what the traced > host machine page size was. > > + If the file version is 7 or greater, the compression header is > + written next: > + "name version\0" I wonder if we should make it: "name\0version\0" Also, I think "none" is acceptable, where none of the sections are compressed. If we add something special for a version 7 but don't want to compress, we need to support that. > + where "name" and "version" are strings, name and version of the > + compression algorithm used to compress the trace file. > + > +COMPRESSION FORMAT OF THE HEADER SECTIONS > +----------------------------------------- > + If the file version is 7 or greater, some header sections are compressed > + with the compression algorithm, specified in the compression header. > + The format of these compressed sections is: > + <4 bytes> unsigned int, size of compressed data in the next block. > + <4 bytes> unsigned int, size of uncompressed data. > + <data> binary compressed data, with the specified size. > + These sections must be uncompressed on reading. The described format of > + the sections refers to the uncomperssed data. I think each section should have a flag that states that it is compressed or not. That way we could have options that determine "what" gets compressed, and not have it be all or none. > + > HEADER INFO FORMAT > ------------------ > > @@ -93,7 +110,8 @@ FTRACE EVENT FORMATS > > Directly after the header information comes the information about > the Ftrace specific events. These are the events used by the Ftrace plugins > - and are not enabled by the event tracing. > + and are not enabled by the event tracing. If the file version is 7 or > + greater, this section is compressed. Perhaps add a single byte ahead of each section, where "0" is not compressed, and "1" is compressed? > > The next 4 bytes contain a 32-bit word of the number of Ftrace event > format files that are stored in the file. > @@ -110,7 +128,8 @@ EVENT FORMATS > ------------- > > Directly after the Ftrace formats comes the information about > - the event layout. > + the event layout. If the file version is 7 or greater, this section > + is compressed. > > The next 4 bytes are a 32-bit word containing the number of > event systems that are stored in the file. These are the > @@ -137,7 +156,8 @@ KALLSYMS INFORMATION > -------------------- > > Directly after the event formats comes the information of the mapping > - of function addresses to the function names. > + of function addresses to the function names. If the file version is 7 > + or greater, this section is compressed. > > The next 4 bytes are a 32-bit word containing the size of the > data holding the function mappings. > @@ -154,6 +174,7 @@ TRACE_PRINTK INFORMATION > store the format string outside the ring buffer. > This information can be found in: > debugfs/tracing/printk_formats > + If the file version is 7 or greater, this section is compressed. > > The next 4 bytes are a 32-bit word containing the size of the > data holding the printk formats. > @@ -166,7 +187,8 @@ PROCESS INFORMATION > ------------------- > > Directly after the trace_printk formats comes the information mapping > - a PID to a process name. > + a PID to a process name. If the file version is 7 or greater, this > + section is compressed. > > The next 8 bytes contain a 64-bit word that holds the size of the > data mapping the PID to a process name. > @@ -193,10 +215,11 @@ REST OF TRACE-CMD HEADER > > "flyrecord\0" > > - If it is "options \0" then: > + If it is "options \0" then follows a section with trace options. > + If the file version is 7 or greater, this section is compressed. > > The next 2 bytes are a 16-bit word defining the current option. > - If the the value is zero then there are no more options. > + If the value is zero then there are no more options. > > Otherwise, the next 4 bytes contain a 32-bit word containing the > option size. If the reader does not know how to handle the option > @@ -206,6 +229,25 @@ REST OF TRACE-CMD HEADER > The next option will be directly after the previous option, and > the options ends with a zero in the option type field. > > +COMPRESSION FORMAT OF THE TRACE DATA > +------------------------------------ > + > + If the file version is 7 or greater, the tarce data is compressed Typo "trace data" And this is where we definitely need to make it optional. We currently do not have a safe way to read this file. The "uncompress to /tmp" is not a reliable way to do this. And again, people can likely want to have the header compressed but not the data, due to speed in reading. -- Steve > + with the compression algorithm, specified in the compression header. > + The data is compressed in chunks. The size of one compression chunk > + is defined when the file is written. The format of compressed trace > + data is: > + <4 bytes> unsigned int, count of chunks. > + Follows the compressed chunks of givent count. For each chunk: > + <4 bytes> unsigned int, size of compressed data in this chunk. > + <4 bytes> unsigned int, size of uncompressed data. > + <data> binary compressed data, with the specified size. > + These chunks must be uncompressed on reading. The described format of > + trace data refers to the uncomperssed data. > + > +TRACE DATA > +---------- > + > The next 10 bytes after the options are one of the following: > > "latency \0" > @@ -217,6 +259,7 @@ REST OF TRACE-CMD HEADER > If the value is "latency \0", then the rest of the file is > simply ASCII text that was taken from the target's: > debugfs/tracing/trace > + If the file version is 7 or greater, the latency data is compressed. > > If the value is "flyrecord\0", the following is present: > > @@ -232,6 +275,7 @@ REST OF TRACE-CMD HEADER > CPU DATA > -------- > > + If the file version is 7 or greater, the CPU data is compressed. > The CPU data is located in the part of the file that is specified > in the end of the header. Padding is placed between the header and > the CPU data, placing the CPU data at a page aligned (target page) position
On Mon, 14 Jun 2021 10:50:01 +0300 "Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com> wrote: > If libz is available, use that library to provide trace file compression > support. The library is detected runtime. Why have the library detected at runtime? If it is detected, we can then have the library flags include -lz. Why use dlopen to load zlib? And not just include it? This seems rather fragile to try to get right. -- Steve > > Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> > ---
On Mon, 14 Jun 2021 10:50:02 +0300 "Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com> wrote: > When a trace buffer data are written in the trace file, the buffer > option in the file metadata is updated with the file offset of the > tracing data. Hide this logic into the trace-cmd library. > Added new APIs: > tracecmd_add_buffer_description() > tracecmd_write_buffers_description() > Changed APIs: > tracecmd_append_buffer_cpu_data() > Removed APIs: > tracecmd_add_buffer_option() So I was going to add this patch with some other ones from this series, but I'm holding off just due to the names. We are not writing a description, we are only adding name and cpus. We could change that to tracecmd_add_buffer_info() and tracecmd_write_buffer_info() ? As "info" is just information about the buffer, and name / cpu is that. But "description" usually means a bit more in depth information. Even though the write_buffer_info() could be doing more than one buffer, I think it still sounds best keeping it singular, as it is writing the buffer_info section. > > This internal refactoring is needed for changes, related to compression > of the options sections of the trace file. > > Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> > --- > .../include/private/trace-cmd-private.h | 8 +- > lib/trace-cmd/trace-output.c | 84 +++++++++++++++++-- > tracecmd/trace-record.c | 16 ++-- > 3 files changed, 85 insertions(+), 23 deletions(-) > > diff --git a/lib/trace-cmd/include/private/trace-cmd-private.h b/lib/trace-cmd/include/private/trace-cmd-private.h > index ee73325c..cbb578ec 100644 > --- a/lib/trace-cmd/include/private/trace-cmd-private.h > +++ b/lib/trace-cmd/include/private/trace-cmd-private.h > @@ -295,8 +295,8 @@ struct tracecmd_option * > tracecmd_add_option_v(struct tracecmd_output *handle, > unsigned short id, const struct iovec *vector, int count); > > -struct tracecmd_option *tracecmd_add_buffer_option(struct tracecmd_output *handle, > - const char *name, int cpus); > +int tracecmd_add_buffer_description(struct tracecmd_output *handle, const char *name, int cpus); > +int tracecmd_write_buffers_description(struct tracecmd_output *handle); > > int tracecmd_write_cpus(struct tracecmd_output *handle, int cpus); > int tracecmd_write_cmdlines(struct tracecmd_output *handle); > @@ -312,9 +312,7 @@ int tracecmd_write_cpu_data(struct tracecmd_output *handle, > int tracecmd_append_cpu_data(struct tracecmd_output *handle, > int cpus, char * const *cpu_data_files); > int tracecmd_append_buffer_cpu_data(struct tracecmd_output *handle, > - struct tracecmd_option *option, > - int cpus, char * const *cpu_data_files); > - > + const char *name, int cpus, char * const *cpu_data_files); > struct tracecmd_output *tracecmd_get_output_handle_fd(int fd); > > /* --- Reading the Fly Recorder Trace --- */ > diff --git a/lib/trace-cmd/trace-output.c b/lib/trace-cmd/trace-output.c > index 7c7d3d76..8f8ca164 100644 > --- a/lib/trace-cmd/trace-output.c > +++ b/lib/trace-cmd/trace-output.c > @@ -44,6 +44,14 @@ struct tracecmd_option { > struct list_head list; > }; > > +struct tracecmd_buffer { > + int cpus; > + void *name; > + tsize_t offset; > + struct tracecmd_option *option; > + struct list_head list; > +}; You can add another tab to the above structure to make it look better: struct tracecmd_buffer { int cpus; void *name; tsize_t offset; struct tracecmd_option *option; struct list_head list; }; -- Steve > + > enum { > OUTPUT_FL_SEND_META = (1 << 0), > }; > @@ -63,6 +71,7 @@ struct tracecmd_output { > struct tracecmd_compression *compress; > > struct list_head options; > + struct list_head buffers; > struct tracecmd_msg_handle *msg_handle; > char *trace_clock; > }; > @@ -189,6 +198,7 @@ bool tracecmd_get_quiet(struct tracecmd_output *handle) > void tracecmd_output_free(struct tracecmd_output *handle) > { > struct tracecmd_option *option; > + struct tracecmd_buffer *buffer; > > if (!handle) > return; > @@ -199,6 +209,13 @@ void tracecmd_output_free(struct tracecmd_output *handle) > if (handle->pevent) > tep_unref(handle->pevent); > > + while (!list_empty(&handle->buffers)) { > + buffer = container_of(handle->buffers.next, > + struct tracecmd_buffer, list); > + list_del(&buffer->list); > + free(buffer->name); > + free(buffer); > + } > while (!list_empty(&handle->options)) { > option = container_of(handle->options.next, > struct tracecmd_option, list); > @@ -1071,6 +1088,7 @@ create_file_fd(int fd, struct tracecmd_input *ihandle, > goto out_free; > > list_head_init(&handle->options); > + list_head_init(&handle->buffers); > > buf[0] = 23; > buf[1] = 8; > @@ -1369,9 +1387,8 @@ int tracecmd_append_options(struct tracecmd_output *handle) > return 0; > } > > -struct tracecmd_option * > -tracecmd_add_buffer_option(struct tracecmd_output *handle, const char *name, > - int cpus) > +static struct tracecmd_option * > +add_buffer_option(struct tracecmd_output *handle, const char *name, int cpus) > { > struct tracecmd_option *option; > char *buf; > @@ -1399,6 +1416,53 @@ tracecmd_add_buffer_option(struct tracecmd_output *handle, const char *name, > return option; > } > > +int tracecmd_add_buffer_description(struct tracecmd_output *handle, const char *name, int cpus) > +{ > + struct tracecmd_buffer *buf; > + > + buf = calloc(1, sizeof(struct tracecmd_buffer)); > + if (!buf) > + return -1; > + buf->name = strdup(name); > + buf->cpus = cpus; > + if (!buf->name) { > + free(buf); > + return -1; > + } > + list_add_tail(&buf->list, &handle->buffers); > + return 0; > +} > + > +int tracecmd_write_buffers_description(struct tracecmd_output *handle) > +{ > + struct tracecmd_option *option; > + struct tracecmd_buffer *buf; > + > + list_for_each_entry(buf, &handle->buffers, list) { > + option = add_buffer_option(handle, buf->name, buf->cpus); > + if (!option) > + return -1; > + buf->option = option; > + } > + return 0; > +} > + > +static tsize_t get_buffer_file_offset(struct tracecmd_output *handle, const char *name) > +{ > + struct tracecmd_buffer *buf; > + > + list_for_each_entry(buf, &handle->buffers, list) { > + if (strlen(name) == strlen(buf->name) && !strcmp(name, buf->name)) { > + if (handle->file_version >= 7) > + return buf->offset; > + if (!buf->option) > + break; > + return buf->option->offset; > + } > + } > + return 0; > +} > + > int tracecmd_write_cmdlines(struct tracecmd_output *handle) > { > int ret; > @@ -1643,18 +1707,23 @@ int tracecmd_append_cpu_data(struct tracecmd_output *handle, > } > > int tracecmd_append_buffer_cpu_data(struct tracecmd_output *handle, > - struct tracecmd_option *option, > - int cpus, char * const *cpu_data_files) > + const char *name, int cpus, char * const *cpu_data_files) > { > + tsize_t b_offset; > tsize_t offset; > stsize_t ret; > > + b_offset = get_buffer_file_offset(handle, name); > + if (!b_offset) { > + tracecmd_warning("Cannot find description for buffer %s\n", name); > + return -1; > + } > offset = lseek64(handle->fd, 0, SEEK_CUR); > > /* Go to the option data, where will write the offest */ > - ret = lseek64(handle->fd, option->offset, SEEK_SET); > + ret = lseek64(handle->fd, b_offset, SEEK_SET); > if (ret == (off64_t)-1) { > - tracecmd_warning("could not seek to %lld\n", option->offset); > + tracecmd_warning("could not seek to %lld\n", b_offset); > return -1; > } > > @@ -1713,6 +1782,7 @@ struct tracecmd_output *tracecmd_get_output_handle_fd(int fd) > handle->page_size = tracecmd_page_size(ihandle); > handle->file_version = tracecmd_get_in_file_version(ihandle); > list_head_init(&handle->options); > + list_head_init(&handle->buffers); > > if (!tracecmd_get_file_compress_proto(ihandle, &cname, &cver)) { > handle->compress = tracecmd_compress_alloc(cname, cver, handle->fd, > diff --git a/tracecmd/trace-record.c b/tracecmd/trace-record.c > index d3362e5b..eff6f2f0 100644 > --- a/tracecmd/trace-record.c > +++ b/tracecmd/trace-record.c > @@ -4152,7 +4152,6 @@ static void touch_file(const char *file) > } > > static void append_buffer(struct tracecmd_output *handle, > - struct tracecmd_option *buffer_option, > struct buffer_instance *instance, > char **temp_files) > { > @@ -4180,7 +4179,7 @@ static void append_buffer(struct tracecmd_output *handle, > touch_file(temp_files[i]); > } > > - tracecmd_append_buffer_cpu_data(handle, buffer_option, > + tracecmd_append_buffer_cpu_data(handle, tracefs_instance_get_name(instance->tracefs), > cpu_count, temp_files); > > for (i = 0; i < instance->cpu_count; i++) { > @@ -4441,7 +4440,6 @@ static void write_guest_file(struct buffer_instance *instance) > > static void record_data(struct common_record_context *ctx) > { > - struct tracecmd_option **buffer_options; > struct tracecmd_output *handle; > struct buffer_instance *instance; > bool local = false; > @@ -4512,9 +4510,6 @@ static void record_data(struct common_record_context *ctx) > } > > if (buffers) { > - buffer_options = malloc(sizeof(*buffer_options) * buffers); > - if (!buffer_options) > - die("Failed to allocate buffer options"); > i = 0; > for_each_instance(instance) { > int cpus = instance->cpu_count != local_cpu_count ? > @@ -4522,10 +4517,9 @@ static void record_data(struct common_record_context *ctx) > > if (instance->msg_handle) > continue; > - > - buffer_options[i++] = tracecmd_add_buffer_option(handle, > - tracefs_instance_get_name(instance->tracefs), > - cpus); > + tracecmd_add_buffer_description(handle, > + tracefs_instance_get_name(instance->tracefs), > + cpus); > add_buffer_stat(handle, instance); > } > } > @@ -4560,7 +4554,7 @@ static void record_data(struct common_record_context *ctx) > if (instance->msg_handle) > continue; > print_stat(instance); > - append_buffer(handle, buffer_options[i++], instance, temp_files); > + append_buffer(handle, instance, temp_files); > } > } >
Please rebase the series on top of upstream. I pushed up 6 of your patches from this series. And one bug fix that will affect the series as well. Also, we need to come up with a better option name than "--compression". One that also can handle an optional parameter. Perhaps just have it be "-z". We can have that be default zlib? Maybe, zlib is enough, and we don't need to have different compression types. -- Steve
On Tue, Jun 22, 2021 at 4:26 AM Steven Rostedt <rostedt@goodmis.org> wrote: > > On Mon, 14 Jun 2021 10:50:01 +0300 > "Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com> wrote: > > > If libz is available, use that library to provide trace file compression > > support. The library is detected runtime. > > Why have the library detected at runtime? > > If it is detected, we can then have the library flags include -lz. > > Why use dlopen to load zlib? And not just include it? > > This seems rather fragile to try to get right. The idea of this design is not to bring additional mandatory dependencies. I do not know if libz is available by default, but even if we assume it is - each additional compression library that is added will be a mandatory dependency. The compression is not a mandatory functionality, trace-cmd can work without it. Why do you think it is fragile, I think dlopen() uses the same core linker logic to find and load the library. The only difference I see is that using "-lz" leads to a mandatory dependency, trace-cmd will not run without it. > > -- Steve > > > > > > Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> > > --- -- Tzvetomir (Ceco) Stoyanov VMware Open Source Technology Center
On Tue, Jun 22, 2021 at 2:10 AM Steven Rostedt <rostedt@goodmis.org> wrote: > > On Mon, 14 Jun 2021 10:50:05 +0300 > "Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com> wrote: > > > Comperss the option section of the trace file. This section is not big > > currently and compressing it does not reduce significantly the size of > > the file. This could be useful in the future as new options can be > > added, storing a potentially huge amount of data. > > I'm not sure we want to bother doing this. New options should never be > large, and if they are, then the option itself can be compressed when > created. > > What good reason is there to compress the options? I think adding any new metadata should be done only by adding new options - this will not break the format. For example kernel debug information for parsing function's arguments - it will be huge. Having the entire option section compressed will simplify that. And it is consistent with the other metadata sections. > > I believe this is just adding more complexity than needed. > > -- Steve -- Tzvetomir (Ceco) Stoyanov VMware Open Source Technology Center
On Tue, Jun 22, 2021 at 2:23 AM Steven Rostedt <rostedt@goodmis.org> wrote: > > On Mon, 14 Jun 2021 10:50:09 +0300 > "Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com> wrote: > > > When reading a trace.dat file of version 7, uncompress the trace data. > > The trace data for each CPU is uncompressed in a temporary file, located > > in /tmp directory with prefix "trace_cpu_data". > > With large trace files, this will be an issue. Several systems setup the > /tmp directory as a ramfs file system (that is, it is locate in ram, and > not backed up on disk). If you have very large trace files, which you would > if you are going to bother compressing them, by uncompressing them into > /tmp, it could take up all the memory of the machine, or easily fill the > /tmp limit. There are a few possible approaches for solving that: - use the same directory where the input trace file is located - use an environment variable for user specified temp directory for these files - check if there is enough free space on the FS before uncompressing > > Simply uncompressing the entire trace data is not an option. The best we > can do is to uncompress on a as needed basis. That would require having > meta data that is stored to know what pages are compressed. > I can modify that logic to compress page by page, as the data is loaded by pages. Or use some of the above approaches ? > -- Steve -- Tzvetomir (Ceco) Stoyanov VMware Open Source Technology Center
On Tue, Jun 22, 2021 at 3:37 AM Steven Rostedt <rostedt@goodmis.org> wrote: > > On Mon, 14 Jun 2021 10:50:29 +0300 > "Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com> wrote: > > > Updated the trace.dat man page with the changes related to file > > version 7 and compression. > > > > Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> > > --- > > Documentation/trace-cmd/trace-cmd.dat.5.txt | 56 ++++++++++++++++++--- > > 1 file changed, 50 insertions(+), 6 deletions(-) > > > > diff --git a/Documentation/trace-cmd/trace-cmd.dat.5.txt b/Documentation/trace-cmd/trace-cmd.dat.5.txt > > index 8d285353..e80d460e 100644 > > --- a/Documentation/trace-cmd/trace-cmd.dat.5.txt > > +++ b/Documentation/trace-cmd/trace-cmd.dat.5.txt > > @@ -52,6 +52,23 @@ INITIAL FORMAT > > The next 4 bytes are a 32-bit word that defines what the traced > > host machine page size was. > > > > + If the file version is 7 or greater, the compression header is > > + written next: > > + "name version\0" > > I wonder if we should make it: "name\0version\0" > > Also, I think "none" is acceptable, where none of the sections are > compressed. If we add something special for a version 7 but don't want > to compress, we need to support that. > > > + where "name" and "version" are strings, name and version of the > > + compression algorithm used to compress the trace file. > > + > > +COMPRESSION FORMAT OF THE HEADER SECTIONS > > +----------------------------------------- > > + If the file version is 7 or greater, some header sections are compressed > > + with the compression algorithm, specified in the compression header. > > + The format of these compressed sections is: > > + <4 bytes> unsigned int, size of compressed data in the next block. > > + <4 bytes> unsigned int, size of uncompressed data. > > + <data> binary compressed data, with the specified size. > > + These sections must be uncompressed on reading. The described format of > > + the sections refers to the uncomperssed data. > > I think each section should have a flag that states that it is > compressed or not. That way we could have options that determine "what" > gets compressed, and not have it be all or none. I was thinking the same, but could not find a use case. That means to give control to the user to decide what parts should be compressed. This will complicate the implementation, new trace-cmd parameters should be added. As I couldn't thought of a use case, decided to go with the simpler approach. May be it makes sense only for the trace data, but the metadata should be always compressed if possible. > > > + > > HEADER INFO FORMAT > > ------------------ > > > > @@ -93,7 +110,8 @@ FTRACE EVENT FORMATS > > > > Directly after the header information comes the information about > > the Ftrace specific events. These are the events used by the Ftrace plugins > > - and are not enabled by the event tracing. > > + and are not enabled by the event tracing. If the file version is 7 or > > + greater, this section is compressed. > > Perhaps add a single byte ahead of each section, where "0" is not > compressed, and "1" is compressed? > > > > > The next 4 bytes contain a 32-bit word of the number of Ftrace event > > format files that are stored in the file. > > @@ -110,7 +128,8 @@ EVENT FORMATS > > ------------- > > > > Directly after the Ftrace formats comes the information about > > - the event layout. > > + the event layout. If the file version is 7 or greater, this section > > + is compressed. > > > > The next 4 bytes are a 32-bit word containing the number of > > event systems that are stored in the file. These are the > > @@ -137,7 +156,8 @@ KALLSYMS INFORMATION > > -------------------- > > > > Directly after the event formats comes the information of the mapping > > - of function addresses to the function names. > > + of function addresses to the function names. If the file version is 7 > > + or greater, this section is compressed. > > > > The next 4 bytes are a 32-bit word containing the size of the > > data holding the function mappings. > > @@ -154,6 +174,7 @@ TRACE_PRINTK INFORMATION > > store the format string outside the ring buffer. > > This information can be found in: > > debugfs/tracing/printk_formats > > + If the file version is 7 or greater, this section is compressed. > > > > The next 4 bytes are a 32-bit word containing the size of the > > data holding the printk formats. > > @@ -166,7 +187,8 @@ PROCESS INFORMATION > > ------------------- > > > > Directly after the trace_printk formats comes the information mapping > > - a PID to a process name. > > + a PID to a process name. If the file version is 7 or greater, this > > + section is compressed. > > > > The next 8 bytes contain a 64-bit word that holds the size of the > > data mapping the PID to a process name. > > @@ -193,10 +215,11 @@ REST OF TRACE-CMD HEADER > > > > "flyrecord\0" > > > > - If it is "options \0" then: > > + If it is "options \0" then follows a section with trace options. > > + If the file version is 7 or greater, this section is compressed. > > > > The next 2 bytes are a 16-bit word defining the current option. > > - If the the value is zero then there are no more options. > > + If the value is zero then there are no more options. > > > > Otherwise, the next 4 bytes contain a 32-bit word containing the > > option size. If the reader does not know how to handle the option > > @@ -206,6 +229,25 @@ REST OF TRACE-CMD HEADER > > The next option will be directly after the previous option, and > > the options ends with a zero in the option type field. > > > > +COMPRESSION FORMAT OF THE TRACE DATA > > +------------------------------------ > > + > > + If the file version is 7 or greater, the tarce data is compressed > > Typo "trace data" > > And this is where we definitely need to make it optional. We currently > do not have a safe way to read this file. The "uncompress to /tmp" is > not a reliable way to do this. And again, people can likely want to > have the header compressed but not the data, due to speed in reading. > > -- Steve > > > + with the compression algorithm, specified in the compression header. > > + The data is compressed in chunks. The size of one compression chunk > > + is defined when the file is written. The format of compressed trace > > + data is: > > + <4 bytes> unsigned int, count of chunks. > > + Follows the compressed chunks of givent count. For each chunk: > > + <4 bytes> unsigned int, size of compressed data in this chunk. > > + <4 bytes> unsigned int, size of uncompressed data. > > + <data> binary compressed data, with the specified size. > > + These chunks must be uncompressed on reading. The described format of > > + trace data refers to the uncomperssed data. > > + > > +TRACE DATA > > +---------- > > + > > The next 10 bytes after the options are one of the following: > > > > "latency \0" > > @@ -217,6 +259,7 @@ REST OF TRACE-CMD HEADER > > If the value is "latency \0", then the rest of the file is > > simply ASCII text that was taken from the target's: > > debugfs/tracing/trace > > + If the file version is 7 or greater, the latency data is compressed. > > > > If the value is "flyrecord\0", the following is present: > > > > @@ -232,6 +275,7 @@ REST OF TRACE-CMD HEADER > > CPU DATA > > -------- > > > > + If the file version is 7 or greater, the CPU data is compressed. > > The CPU data is located in the part of the file that is specified > > in the end of the header. Padding is placed between the header and > > the CPU data, placing the CPU data at a page aligned (target page) position > -- Tzvetomir (Ceco) Stoyanov VMware Open Source Technology Center
On Tue, 22 Jun 2021 13:29:36 +0300
Tzvetomir Stoyanov <tz.stoyanov@gmail.com> wrote:
> On Tue, Jun 22, 2021 at 4:26 AM Steven Rostedt <rostedt@goodmis.org> wrote:
> >
> > On Mon, 14 Jun 2021 10:50:01 +0300
> > "Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com> wrote:
> >
> > > If libz is available, use that library to provide trace file compression
> > > support. The library is detected runtime.
> >
> > Why have the library detected at runtime?
> >
> > If it is detected, we can then have the library flags include -lz.
> >
> > Why use dlopen to load zlib? And not just include it?
> >
> > This seems rather fragile to try to get right.
>
> The idea of this design is not to bring additional mandatory
> dependencies. I do not know if libz is available by default, but even
> if we assume it is - each additional compression library that is added
> will be a mandatory dependency. The compression is not a mandatory
> functionality, trace-cmd can work without it. Why do you think it is
> fragile, I think dlopen() uses the same core linker logic to find and
> load the library. The only difference I see is that using "-lz" leads
> to a mandatory dependency, trace-cmd will not run without it.
>
The dependency is determined at compile time, just like we do for the
python libraries, and dwarf, and anything else that might add a new
feature.
Let me explain a scenario of why I called it fragile. Let's say that
trace-cmd has no package dependency on zlib, but the system has some other
package that does have a dependency. Thus the package manager pulls in zlib
to satisfy this other package.
The user does a recording, and trace-cmd detects zlib, and compresses the
data.
Later, the user decides they do not need this other package and uninstalls
it. The package manager sees there's nothing that depends on zlib anymore,
and uninstalls zlib as well.
Then the user goes to read their trace.dat file, and suddenly trace-cmd
can't read it!
THAT is what I call fragile. And if something like that ever happened to
me, I would stop using whatever did that to me.
Manually pulling in system dynamic libraries with dlopen is something I
never heard of. The way to do this is make it a build time dependency. If
zlib exists, then define HAVE_ZLIB and allow compressions and everything
else. If it does not, then we don't support compression. Simple as that.
Making it a runtime dependency has a lot of issues, *especially* since one
run of trace-cmd (the reporting) depends on a previous run of trace-cmd
(the recording), and if the environment changes between the two, the user
will rightfully say WTF!
-- Steve
On Tue, 22 Jun 2021 13:50:44 +0300 Tzvetomir Stoyanov <tz.stoyanov@gmail.com> wrote: > > With large trace files, this will be an issue. Several systems setup the > > /tmp directory as a ramfs file system (that is, it is locate in ram, and > > not backed up on disk). If you have very large trace files, which you would > > if you are going to bother compressing them, by uncompressing them into > > /tmp, it could take up all the memory of the machine, or easily fill the > > /tmp limit. > > There are a few possible approaches for solving that: > - use the same directory where the input trace file is located I thought about that, but then decided against it, because there's a reason people compress it. If we have to uncompress it to read it, I can see people saying "why is it compressed in the first place?" When data is compressed to save disk space (which I consider this a case), then the reading has to uncompress it on a as-needed basis. > - use an environment variable for user specified temp directory for these files > - check if there is enough free space on the FS before uncompressing > > > > > Simply uncompressing the entire trace data is not an option. The best we > > can do is to uncompress on a as needed basis. That would require having > > meta data that is stored to know what pages are compressed. > > > I can modify that logic to compress page by page, as the data is > loaded by pages. Or use some of the above approaches ? Doing it page by page is probably the most logical solution. It will make it easier to manage without needing to create separate temporary files. I'm guessing we need an index of each page and where they start. We need a way to map the record offset to the page that contains it in such a way that tracecmd_read_at() still works. We could keep this in the file, or create it from the data. I'm thinking saving this as a section in the file would be good as it would be quicker for loading. Have a section for each CPU, that maps each page with their compressed offset in the file, and then just consider the page to be page size. Oh, which reminds me, we need to make sure that we don't use "getpagesize()" to determine the size of the page buffers, because I may be making the buffers more than a single page. It must use the header_page file in the events directory, because it that might change in the future! Anyway, we can have this: buffer_page_size: 4096 /* lets say the compressed data starts at 10,000 just to make this easier to explain. */ u64 cpu_array[0] 10000 <- page 1 (compress to 100 bytes) 10100 <- page 2 (compressed to 150 bytes) 10250 <- page 3 [...] But the record->offset should contain the offset of the uncompressed data. That is, if the record is on page 2 at offset 400 (uncompressed) then offset should be: record->offset = 14496 (10000 + 4096 + 400) Which would be calculated as: record->offset = cpu_data_start[cpu] + page * buffer_page_size + offset; This also means that cpu_array[1] has to save its uncompressed start. That is, even though it may start at 20,000 in the trace data file (10,000 more than the cpu_array[0] start). It's uncompressed location needs to account for all the cpu_array[0] pages, such that no two record's offsets will overlap if they are on different CPUs. cpu_data_start[0] = 10000 (but has 1000 pages, where 1000 * 4096 = 4,096,000) But even if cpu_array[1] starts at 20000, it has to account for the uncompressed cpu_array[0] data, thus we have: cpu_data_start[1] = 4106000 (4096000 + 10000) -- Steve
On Tue, 22 Jun 2021 14:05:25 +0300
Tzvetomir Stoyanov <tz.stoyanov@gmail.com> wrote:
> I was thinking the same, but could not find a use case. That means to
> give control to the user to decide what parts should be compressed.
> This will complicate the implementation, new trace-cmd parameters
> should be added. As I couldn't thought of a use case, decided to go
> with the simpler approach. May be it makes sense only for the trace
> data, but the metadata should be always compressed if possible.
I haven't though too hard for a use case, but the problem about going with
the simple approach is that if we come up with a use case, we can never
implement it.
I like to error on the side of flexibility at the cost of some complexity
at the beginning. You may not need that flexibility for years to come, but
when it happens, you'll be really glad you did it that way.
When going with the simple approach, and putting yourself in a corner, you
are setting yourself up for easy work now, but could potentially have a
real hard time, including complete rewrites, or just giving up on features,
in the future.
The use case I had was not to allow users to have a lot of options, but it
lets us know if we want to bother compressing everything or not. We can
perhaps find that compression takes up more time, and doesn't save us
anything. If we find that's the case, we can chose not to compress later.
By adding a flag that tells trace-cmd if a section is compressed or not, we
don't need to worry about that change. We could uncompress a section, and
everything will still "just work"!
-- Steve