[PATCH 1/4] linux user: make execfd global (like exec path) and keep it open

[PATCH 1/4] linux user: make execfd global (like exec path) and keep it open

[aladjev.andrew]

From: Andrew Aladjev <aladjev.andrew@gmail.com>

User opens /proc/self/exe symlink, than kernel should create /proc/self/fd/<execfd> symlink. We should be able to detect both exe and fd/<execfd> symlinks to provide common behaviour. The easiest solution is to make execfd global and keep it open. This solution looks acceptable because exec_path is already global. PS load_flt_binary is not closing bprm->fd, so load_elf_binary may not close it too (used symmetrically in loader_exec).
---
 linux-user/elfload.c |  3 ++-
 linux-user/exit.c    |  2 ++
 linux-user/main.c    |  2 +-
 linux-user/qemu.h    |  1 +
 linux-user/syscall.c | 16 ++++++++++++----
 5 files changed, 18 insertions(+), 6 deletions(-)

diff --git a/linux-user/elfload.c b/linux-user/elfload.c
index bab4237..4c347b0 100644
--- a/linux-user/elfload.c
+++ b/linux-user/elfload.c
@@ -2600,6 +2600,7 @@ static bool parse_elf_properties(int image_fd,
 
    IMAGE_NAME is the filename of the image, to use in error messages.
    IMAGE_FD is the open file descriptor for the image.
+   WARNING: IMAGE_FD won't be closed.
 
    BPRM_BUF is a copy of the beginning of the file; this of course
    contains the elf file header at offset 0.  It is assumed that this
@@ -2910,7 +2911,6 @@ static void load_elf_image(const char *image_name, int image_fd,
 
     mmap_unlock();
 
-    close(image_fd);
     return;
 
  exit_read:
@@ -2953,6 +2953,7 @@ static void load_elf_interp(const char *filename, struct image_info *info,
     }
 
     load_elf_image(filename, fd, info, NULL, bprm_buf);
+    close(fd);
 }
 
 static int symfind(const void *s0, const void *s1)
diff --git a/linux-user/exit.c b/linux-user/exit.c
index 70b3440..cc9cf38 100644
--- a/linux-user/exit.c
+++ b/linux-user/exit.c
@@ -28,6 +28,8 @@ extern void __gcov_dump(void);
 
 void preexit_cleanup(CPUArchState *env, int code)
 {
+    close(execfd);
+
 #ifdef CONFIG_GPROF
         _mcleanup();
 #endif
diff --git a/linux-user/main.c b/linux-user/main.c
index 81f48ff..d365335 100644
--- a/linux-user/main.c
+++ b/linux-user/main.c
@@ -50,6 +50,7 @@
 #include "crypto/init.h"
 
 char *exec_path;
+int execfd;
 
 int singlestep;
 static const char *argv0;
@@ -628,7 +629,6 @@ int main(int argc, char **argv, char **envp)
     int target_argc;
     int i;
     int ret;
-    int execfd;
     int log_mask;
     unsigned long max_reserved_va;
 
diff --git a/linux-user/qemu.h b/linux-user/qemu.h
index 52c9817..ec26730 100644
--- a/linux-user/qemu.h
+++ b/linux-user/qemu.h
@@ -160,6 +160,7 @@ typedef struct TaskState {
 } __attribute__((aligned(16))) TaskState;
 
 extern char *exec_path;
+extern int execfd;
 void init_task_state(TaskState *ts);
 void task_settid(TaskState *);
 void stop_all_tasks(void);
diff --git a/linux-user/syscall.c b/linux-user/syscall.c
index 389ec09..c171dea 100644
--- a/linux-user/syscall.c
+++ b/linux-user/syscall.c
@@ -8110,8 +8110,7 @@ static int do_openat(void *cpu_env, int dirfd, const char *pathname, int flags,
     };
 
     if (is_proc_myself(pathname, "exe")) {
-        int execfd = qemu_getauxval(AT_EXECFD);
-        return execfd ? execfd : safe_openat(dirfd, exec_path, flags, mode);
+        return execfd;
     }
 
     for (fake_open = fakes; fake_open->filename; fake_open++) {
@@ -8369,8 +8368,17 @@ static abi_long do_syscall1(void *cpu_env, int num, abi_long arg1,
         return ret;
 #endif
     case TARGET_NR_close:
-        fd_trans_unregister(arg1);
-        return get_errno(close(arg1));
+        {
+            int fd = arg1;
+
+            /* We don't need to close execfd, it will be closed on QEMU exit. */
+            if (fd == execfd) {
+                return 0;
+            }
+
+            fd_trans_unregister(fd);
+            return get_errno(close(fd));
+        }
 
     case TARGET_NR_brk:
         return do_brk(arg1);
-- 
2.26.2

[PATCH 2/4] linux user: moved is proc functions to separate file

[aladjev.andrew]

From: Andrew Aladjev <aladjev.andrew@gmail.com>

---
 linux-user/meson.build    |  1 +
 linux-user/syscall.c      | 33 +--------------------------------
 linux-user/syscall_proc.c | 32 ++++++++++++++++++++++++++++++++
 linux-user/syscall_proc.h |  7 +++++++
 4 files changed, 41 insertions(+), 32 deletions(-)
 create mode 100644 linux-user/syscall_proc.c
 create mode 100644 linux-user/syscall_proc.h

diff --git a/linux-user/meson.build b/linux-user/meson.build
index 7fe28d6..b6a204d 100644
--- a/linux-user/meson.build
+++ b/linux-user/meson.build
@@ -9,6 +9,7 @@ linux_user_ss.add(files(
   'signal.c',
   'strace.c',
   'syscall.c',
+  'syscall_proc.c',
   'uaccess.c',
   'uname.c',
 ))
diff --git a/linux-user/syscall.c b/linux-user/syscall.c
index c171dea..0ead34b 100644
--- a/linux-user/syscall.c
+++ b/linux-user/syscall.c
@@ -133,6 +133,7 @@
 #include "qapi/error.h"
 #include "fd-trans.h"
 #include "tcg/tcg.h"
+#include "syscall_proc.h"
 
 #ifndef CLONE_IO
 #define CLONE_IO                0x80000000      /* Clone io context */
@@ -7979,38 +7980,6 @@ static int open_self_auxv(void *cpu_env, int fd)
     return 0;
 }
 
-static int is_proc_myself(const char *filename, const char *entry)
-{
-    if (!strncmp(filename, "/proc/", strlen("/proc/"))) {
-        filename += strlen("/proc/");
-        if (!strncmp(filename, "self/", strlen("self/"))) {
-            filename += strlen("self/");
-        } else if (*filename >= '1' && *filename <= '9') {
-            char myself[80];
-            snprintf(myself, sizeof(myself), "%d/", getpid());
-            if (!strncmp(filename, myself, strlen(myself))) {
-                filename += strlen(myself);
-            } else {
-                return 0;
-            }
-        } else {
-            return 0;
-        }
-        if (!strcmp(filename, entry)) {
-            return 1;
-        }
-    }
-    return 0;
-}
-
-#if defined(HOST_WORDS_BIGENDIAN) != defined(TARGET_WORDS_BIGENDIAN) || \
-    defined(TARGET_SPARC) || defined(TARGET_M68K) || defined(TARGET_HPPA)
-static int is_proc(const char *filename, const char *entry)
-{
-    return strcmp(filename, entry) == 0;
-}
-#endif
-
 #if defined(HOST_WORDS_BIGENDIAN) != defined(TARGET_WORDS_BIGENDIAN)
 static int open_net_route(void *cpu_env, int fd)
 {
diff --git a/linux-user/syscall_proc.c b/linux-user/syscall_proc.c
new file mode 100644
index 0000000..34051a8
--- /dev/null
+++ b/linux-user/syscall_proc.c
@@ -0,0 +1,32 @@
+#include "qemu/osdep.h"
+
+#include "syscall_proc.h"
+
+int is_proc_myself(const char *filename, const char *entry)
+{
+    if (!strncmp(filename, "/proc/", strlen("/proc/"))) {
+        filename += strlen("/proc/");
+        if (!strncmp(filename, "self/", strlen("self/"))) {
+            filename += strlen("self/");
+        } else if (*filename >= '1' && *filename <= '9') {
+            char myself[80];
+            snprintf(myself, sizeof(myself), "%d/", getpid());
+            if (!strncmp(filename, myself, strlen(myself))) {
+                filename += strlen(myself);
+            } else {
+                return 0;
+            }
+        } else {
+            return 0;
+        }
+        if (!strcmp(filename, entry)) {
+            return 1;
+        }
+    }
+    return 0;
+}
+
+int is_proc(const char *filename, const char *entry)
+{
+    return strcmp(filename, entry) == 0;
+}
diff --git a/linux-user/syscall_proc.h b/linux-user/syscall_proc.h
new file mode 100644
index 0000000..3098af9
--- /dev/null
+++ b/linux-user/syscall_proc.h
@@ -0,0 +1,7 @@
+#ifndef SYSCALL_PROC_H
+#define SYSCALL_PROC_H
+
+int is_proc(const char *filename, const char *entry);
+int is_proc_myself(const char *filename, const char *entry);
+
+#endif
-- 
2.26.2

[PATCH 3/4] linux user: refactored is proc myself, added support for fd/<execfd>

[aladjev.andrew]

From: Andrew Aladjev <aladjev.andrew@gmail.com>

---
 linux-user/syscall.c      |  12 ++--
 linux-user/syscall_proc.c | 113 +++++++++++++++++++++++++++++++-------
 linux-user/syscall_proc.h |   5 +-
 3 files changed, 101 insertions(+), 29 deletions(-)

diff --git a/linux-user/syscall.c b/linux-user/syscall.c
index 0ead34b..00ae823 100644
--- a/linux-user/syscall.c
+++ b/linux-user/syscall.c
@@ -8067,18 +8067,18 @@ static int do_openat(void *cpu_env, int dirfd, const char *pathname, int flags,
         { "auxv", open_self_auxv, is_proc_myself },
         { "cmdline", open_self_cmdline, is_proc_myself },
 #if defined(HOST_WORDS_BIGENDIAN) != defined(TARGET_WORDS_BIGENDIAN)
-        { "/proc/net/route", open_net_route, is_proc },
+        { "net/route", open_net_route, is_proc },
 #endif
 #if defined(TARGET_SPARC) || defined(TARGET_HPPA)
-        { "/proc/cpuinfo", open_cpuinfo, is_proc },
+        { "cpuinfo", open_cpuinfo, is_proc },
 #endif
 #if defined(TARGET_M68K)
-        { "/proc/hardware", open_hardware, is_proc },
+        { "hardware", open_hardware, is_proc },
 #endif
         { NULL, NULL, NULL }
     };
 
-    if (is_proc_myself(pathname, "exe")) {
+    if (is_proc_myself_exe(pathname)) {
         return execfd;
     }
 
@@ -9603,7 +9603,7 @@ static abi_long do_syscall1(void *cpu_env, int num, abi_long arg1,
             } else if (!arg3) {
                 /* Short circuit this for the magic exe check. */
                 ret = -TARGET_EINVAL;
-            } else if (is_proc_myself((const char *)p, "exe")) {
+            } else if (is_proc_myself_exe((const char *)p)) {
                 char real[PATH_MAX], *temp;
                 temp = realpath(exec_path, real);
                 /* Return value is # of bytes that we wrote to the buffer. */
@@ -9632,7 +9632,7 @@ static abi_long do_syscall1(void *cpu_env, int num, abi_long arg1,
             p2 = lock_user(VERIFY_WRITE, arg3, arg4, 0);
             if (!p || !p2) {
                 ret = -TARGET_EFAULT;
-            } else if (is_proc_myself((const char *)p, "exe")) {
+            } else if (is_proc_myself_exe((const char *)p)) {
                 char real[PATH_MAX], *temp;
                 temp = realpath(exec_path, real);
                 ret = temp == NULL ? get_errno(-1) : strlen(real) ;
diff --git a/linux-user/syscall_proc.c b/linux-user/syscall_proc.c
index 34051a8..3cf986b 100644
--- a/linux-user/syscall_proc.c
+++ b/linux-user/syscall_proc.c
@@ -1,32 +1,103 @@
 #include "qemu/osdep.h"
+#include "elf.h"
 
 #include "syscall_proc.h"
+#include "qemu.h"
 
-int is_proc_myself(const char *filename, const char *entry)
+#define PROC "/proc/"
+#define SELF "self/"
+#define FD   "fd/"
+
+#define STARTS_WITH(path, CONSTANT) (              \
+    strlen(path) >= strlen(CONSTANT) &&            \
+    strncmp(path, CONSTANT, strlen(CONSTANT)) == 0 \
+)
+
+static inline char *scope_to_proc(const char *path)
 {
-    if (!strncmp(filename, "/proc/", strlen("/proc/"))) {
-        filename += strlen("/proc/");
-        if (!strncmp(filename, "self/", strlen("self/"))) {
-            filename += strlen("self/");
-        } else if (*filename >= '1' && *filename <= '9') {
-            char myself[80];
-            snprintf(myself, sizeof(myself), "%d/", getpid());
-            if (!strncmp(filename, myself, strlen(myself))) {
-                filename += strlen(myself);
-            } else {
-                return 0;
-            }
-        } else {
-            return 0;
-        }
-        if (!strcmp(filename, entry)) {
-            return 1;
+    if (STARTS_WITH(path, PROC)) {
+        return (char *)path + strlen(PROC);
+    }
+
+    return NULL;
+}
+
+static inline char *scope_to_proc_myself(const char *path)
+{
+    char *scope_path = scope_to_proc(path);
+    if (scope_path == NULL) {
+        return NULL;
+    }
+
+    if (STARTS_WITH(scope_path, SELF)) {
+        return scope_path + strlen(SELF);
+    }
+
+    if (strlen(scope_path) >= 1 &&
+        *scope_path >= '1' && *scope_path <= '9') {
+        char pid_path[80];
+        snprintf(pid_path, sizeof(pid_path), "%d/", getpid());
+        if (STARTS_WITH(scope_path, pid_path)) {
+            return scope_path + strlen(pid_path);
         }
     }
-    return 0;
+
+    return NULL;
+}
+
+int is_proc(const char *path, const char *entry)
+{
+    char *scope_path = scope_to_proc(path);
+    if (scope_path == NULL) {
+        return 0;
+    }
+
+    return strcmp(scope_path, entry) == 0;
 }
 
-int is_proc(const char *filename, const char *entry)
+int is_proc_myself(const char *path, const char *entry)
 {
-    return strcmp(filename, entry) == 0;
+    char *scope_path = scope_to_proc_myself(path);
+    if (scope_path == NULL) {
+        return 0;
+    }
+
+    return strcmp(scope_path, entry) == 0;
+}
+
+/*
+ * Kernel creates "fd/#{number}" link after opening "exe" link.
+ * Both "exe" and "fd/#{number}" can be used by application.
+ *
+ * Kernel can provide infinite amount of fd numbers.
+ * QEMU is going to always return single global execfd.
+ *
+ * So we need to check "exe" and "fd/#{execfd}" only.
+ */
+
+int is_proc_myself_exe(const char *path)
+{
+    char *scope_path = scope_to_proc_myself(path);
+    if (scope_path == NULL) {
+        return 0;
+    }
+
+    if (strcmp(scope_path, "exe") == 0) {
+        return 1;
+    }
+
+    if (STARTS_WITH(scope_path, FD)) {
+        scope_path += strlen(FD);
+
+        if (strlen(scope_path) >= 1 &&
+            *scope_path >= '1' && *scope_path <= '9') {
+            char execfd_path[80];
+            snprintf(execfd_path, sizeof(execfd_path), "%d", execfd);
+            if (strcmp(scope_path, execfd_path) == 0) {
+                return 1;
+            }
+        }
+    }
+
+    return 0;
 }
diff --git a/linux-user/syscall_proc.h b/linux-user/syscall_proc.h
index 3098af9..f0e59c0 100644
--- a/linux-user/syscall_proc.h
+++ b/linux-user/syscall_proc.h
@@ -1,7 +1,8 @@
 #ifndef SYSCALL_PROC_H
 #define SYSCALL_PROC_H
 
-int is_proc(const char *filename, const char *entry);
-int is_proc_myself(const char *filename, const char *entry);
+int is_proc(const char *path, const char *entry);
+int is_proc_myself(const char *path, const char *entry);
+int is_proc_myself_exe(const char *path);
 
 #endif
-- 
2.26.2

[PATCH 4/4] linux user: added tests for proc myself

[aladjev.andrew]

From: Andrew Aladjev <aladjev.andrew@gmail.com>

---
 tests/tcg/multiarch/linux-test.c | 85 ++++++++++++++++++++++++++++++++
 1 file changed, 85 insertions(+)

diff --git a/tests/tcg/multiarch/linux-test.c b/tests/tcg/multiarch/linux-test.c
index 96bbad5..4918d45 100644
--- a/tests/tcg/multiarch/linux-test.c
+++ b/tests/tcg/multiarch/linux-test.c
@@ -517,6 +517,89 @@ static void test_shm(void)
     chk_error(shmdt(ptr));
 }
 
+static void test_proc_myself_file(void)
+{
+    int fd1, fd2;
+    char link1[PATH_MAX], link2[PATH_MAX];
+    char buf1[PATH_MAX], buf2[PATH_MAX];
+    int buf1_length, buf2_length;
+
+    /* We can open any file that will always exist. */
+    const char *file_path = "/proc/self/comm";
+
+    char file_realpath[PATH_MAX];
+    if (realpath(file_path, file_realpath) == NULL) {
+        error("proc myself: invalid file");
+    }
+
+    fd1 = chk_error(open(file_path, O_RDONLY));
+    sprintf(link1, "/proc/self/fd/%d", fd1);
+
+    /* Lets try to open same file by first link. */
+    fd2 = chk_error(open(link1, O_RDONLY));
+    sprintf(link2, "/proc/self/fd/%d", fd2);
+
+    /* Two links should point to the same file path. */
+    buf1_length = chk_error(readlink(link1, buf1, PATH_MAX));
+    if (strlen(file_realpath) != buf1_length ||
+        strncmp(file_realpath, buf1, buf1_length) != 0) {
+        error("proc myself: invalid link");
+    }
+    buf2_length = chk_error(readlink(link2, buf2, PATH_MAX));
+    if (strlen(file_realpath) != buf2_length ||
+        strncmp(file_realpath, buf2, buf2_length) != 0) {
+        error("proc myself: invalid link");
+    }
+
+    /* We should be able to read same data from each fd. */
+    buf1_length = chk_error(read(fd1, buf1, PATH_MAX));
+    buf2_length = chk_error(read(fd2, buf2, PATH_MAX));
+    if (buf1_length == 0 || buf1_length != buf2_length ||
+        strncmp(buf1, buf2, buf2_length) != 0) {
+        error("proc myself: invalid file content");
+    }
+
+    chk_error(close(fd2));
+    chk_error(close(fd1));
+}
+
+static void test_proc_myself_exe(void)
+{
+    int fd1, fd2;
+    char link1[PATH_MAX], link2[PATH_MAX];
+    char buf1[PATH_MAX], buf2[PATH_MAX];
+    int buf1_length, buf2_length;
+
+    const char *exe_path = "/proc/self/exe";
+
+    char exe_realpath[PATH_MAX];
+    if (realpath(exe_path, exe_realpath) == NULL) {
+        error("proc myself: invalid exe");
+    }
+
+    fd1 = chk_error(open(exe_path, O_RDONLY));
+    sprintf(link1, "/proc/self/fd/%d", fd1);
+
+    /* Lets try to open link once again. */
+    fd2 = chk_error(open(link1, O_RDONLY));
+    sprintf(link2, "/proc/self/fd/%d", fd2);
+
+    /* Two links should point to the same exe path. */
+    buf1_length = chk_error(readlink(link1, buf1, PATH_MAX));
+    if (strlen(exe_realpath) != buf1_length ||
+        strncmp(exe_realpath, buf1, buf1_length) != 0) {
+        error("proc myself: invalid link");
+    }
+    buf2_length = chk_error(readlink(link2, buf2, PATH_MAX));
+    if (strlen(exe_realpath) != buf2_length ||
+        strncmp(exe_realpath, buf2, buf2_length) != 0) {
+        error("proc myself: invalid link");
+    }
+
+    chk_error(close(fd2));
+    chk_error(close(fd1));
+}
+
 int main(int argc, char **argv)
 {
     test_file();
@@ -532,5 +615,7 @@ int main(int argc, char **argv)
 
     test_signal();
     test_shm();
+    test_proc_myself_file();
+    test_proc_myself_exe();
     return 0;
 }
-- 
2.26.2