* [PATCH] xfs_io: implement 'inode' command V5
@ 2015-11-26 15:46 Carlos Maiolino
2015-11-30 13:22 ` Brian Foster
0 siblings, 1 reply; 4+ messages in thread
From: Carlos Maiolino @ 2015-11-26 15:46 UTC (permalink / raw)
To: xfs
Implements a new xfs_io command, named 'inode', which is supposed to be
used to query information about inode's existence and its physical size
in the filesystem.
Supported options:
Default: -- Return true(1) or false(0) if any inode greater than
32bits has been found in the filesystem
-v -- verbose mode
Display the number and the physical size (in bits)
of the largest inode in the filesystem
[num] -- Return true(1) or false(0) if the inode [num] is in use
-n [num] -- Return the next valid inode after [num]
No manpage sent because there were changes in the supported options and its
descriptions.
I'll send the manpage after the options and descriptions are reviewed.
- Changelog
V3:
- Merge all 3 patches from the V2 together in a single patch
- Rework of '-n [num]' and 'num' only arguments algorithm
- Argument -n now relies on bulkreq.count to check for next inodes, not
on bstat.bs_ino anymore.
- for loop in ret_lsize or ret_largest case, now relies on count being 0
to break the loop
V4:
- Refactor inode_f function to reduce its size and easier logic
- Implement error handlers for invalid command combination (hopefully
all invalid combinations).
- use a single xfs_inogrp array for keep track of inodes
- Fix missing newline in inode_help()
- Rewrite help message in inode_help()
- Fix indentation
V5:
- Reduce the amount of options
- remove igrp_rec variable, and use igroup[lastgrp] directly to get
information from the last inode groups returned by ioctl
Signed-off-by: Carlos Maiolino <cmaiolino@redhat.com>
---
io/open.c | 143 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
1 file changed, 143 insertions(+)
diff --git a/io/open.c b/io/open.c
index ac5a5e0..1e38ea8 100644
--- a/io/open.c
+++ b/io/open.c
@@ -20,6 +20,7 @@
#include "input.h"
#include "init.h"
#include "io.h"
+#include "libxfs.h"
#ifndef __O_TMPFILE
#if defined __alpha__
@@ -44,6 +45,7 @@ static cmdinfo_t statfs_cmd;
static cmdinfo_t chproj_cmd;
static cmdinfo_t lsproj_cmd;
static cmdinfo_t extsize_cmd;
+static cmdinfo_t inode_cmd;
static prid_t prid;
static long extsize;
@@ -750,6 +752,136 @@ statfs_f(
return 0;
}
+static void
+inode_help(void)
+{
+ printf(_(
+"\n"
+"Query physical information about the inode"
+"\n"
+" Default: -- Return true(1) or false(0) if any inode greater than\n"
+" 32bits has been found in the filesystem\n"
+" -v -- verbose mode\n"
+" Display the number and the physical size (in bits)\n"
+" of the largest inode in the filesystem\n"
+"[num] -- Return true(1) or false(0) if the inode [num] is in use\n"
+" -n [num] -- Return the next valid inode after [num]\n"
+"\n"));
+}
+
+static int
+inode_f(
+ int argc,
+ char **argv)
+{
+ __s32 count = 0;
+ __s32 lastgrp = 0;
+ __u64 last = 0;
+ __u64 lastino = 0;
+ __u64 userino = 0;
+ char *p;
+ int c;
+ int verbose = 0;
+ int ret_next = 0;
+ int cmd = 0;
+ struct xfs_inogrp igroup[1024];
+ struct xfs_fsop_bulkreq bulkreq;
+ struct xfs_bstat bstat;
+
+ while ((c = getopt(argc, argv, "nv")) != EOF) {
+ switch (c) {
+ case 'v':
+ verbose = 1;
+ break;
+ case 'n':
+ ret_next = 1;
+ break;
+ default:
+ return command_usage(&inode_cmd);
+ }
+ }
+
+ if (ret_next && verbose)
+ return command_usage(&inode_cmd);
+
+ if (optind < argc) {
+ if (verbose)
+ return command_usage(&inode_cmd);
+
+ if (ret_next) {
+ cmd = XFS_IOC_FSBULKSTAT;
+ } else {
+ if (argc > 2)
+ return command_usage(&inode_cmd);
+ else
+ cmd = XFS_IOC_FSBULKSTAT_SINGLE;
+ }
+
+ userino = strtoull(argv[optind], &p, 10);
+ if ((*p != '\0')) {
+ printf(_("[num] must be a numeric value\n"));
+ exitcode = 1;
+ return 0;
+ }
+
+ bulkreq.lastip = &userino;
+ bulkreq.icount = 1;
+ bulkreq.ubuffer = &bstat;
+ bulkreq.ocount = &count;
+
+ if (xfsctl(file->name, file->fd, cmd, &bulkreq)) {
+ if (errno == EINVAL) {
+ if (!ret_next)
+ printf("0\n");
+ } else {
+ perror("xfsctl");
+ }
+ exitcode = 1;
+ return 0;
+ }
+
+ if (ret_next) {
+ printf("%llu\n", bstat.bs_ino);
+ return 0;
+ } else {
+ /* Inode number used*/
+ printf("1\n");
+ return 0;
+ }
+ }
+
+ bulkreq.lastip = &last;
+ bulkreq.icount = 1024; /* User-defined maybe!? */
+ bulkreq.ubuffer = &igroup;
+ bulkreq.ocount = &count;
+
+ for (;;) {
+ if (xfsctl(file->name, file->fd, XFS_IOC_FSINUMBERS,
+ &bulkreq)) {
+ perror("XFS_IOC_FSINUMBERS");
+ exitcode = 1;
+ return 0;
+ }
+
+ if (count == 0)
+ break;
+
+ lastgrp = count;
+ }
+
+ lastgrp--;
+ lastino = igroup[lastgrp].xi_startino +
+ xfs_highbit64(igroup[lastgrp].xi_allocmask);
+
+ if (verbose)
+ printf("%llu:%d\n", lastino,
+ lastino > XFS_MAXINUMBER_32 ? 64 : 32);
+ else
+ printf("%d\n", lastino > XFS_MAXINUMBER_32 ? 1 : 0);
+
+ return 0;
+}
+
void
open_init(void)
{
@@ -815,6 +947,16 @@ open_init(void)
_("get/set preferred extent size (in bytes) for the open file");
extsize_cmd.help = extsize_help;
+ inode_cmd.name = "inode";
+ inode_cmd.cfunc = inode_f;
+ inode_cmd.args = _("[-n | -v] [num]");
+ inode_cmd.argmin = 0;
+ inode_cmd.argmax = 2;
+ inode_cmd.flags = CMD_NOMAP_OK;
+ inode_cmd.oneline =
+ _("Query inode number usage in the filesystem");
+ inode_cmd.help = inode_help;
+
add_command(&open_cmd);
add_command(&stat_cmd);
add_command(&close_cmd);
@@ -822,4 +964,5 @@ open_init(void)
add_command(&chproj_cmd);
add_command(&lsproj_cmd);
add_command(&extsize_cmd);
+ add_command(&inode_cmd);
}
--
2.4.3
_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs
^ permalink raw reply related [flat|nested] 4+ messages in thread
* Re: [PATCH] xfs_io: implement 'inode' command V5
2015-11-26 15:46 [PATCH] xfs_io: implement 'inode' command V5 Carlos Maiolino
@ 2015-11-30 13:22 ` Brian Foster
2015-11-30 14:26 ` Carlos Maiolino
0 siblings, 1 reply; 4+ messages in thread
From: Brian Foster @ 2015-11-30 13:22 UTC (permalink / raw)
To: Carlos Maiolino; +Cc: xfs
On Thu, Nov 26, 2015 at 04:46:35PM +0100, Carlos Maiolino wrote:
> Implements a new xfs_io command, named 'inode', which is supposed to be
> used to query information about inode's existence and its physical size
> in the filesystem.
>
> Supported options:
>
> Default: -- Return true(1) or false(0) if any inode greater than
> 32bits has been found in the filesystem
> -v -- verbose mode
> Display the number and the physical size (in bits)
> of the largest inode in the filesystem
> [num] -- Return true(1) or false(0) if the inode [num] is in use
> -n [num] -- Return the next valid inode after [num]
>
> No manpage sent because there were changes in the supported options and its
> descriptions.
> I'll send the manpage after the options and descriptions are reviewed.
>
> - Changelog
>
> V3:
> - Merge all 3 patches from the V2 together in a single patch
> - Rework of '-n [num]' and 'num' only arguments algorithm
> - Argument -n now relies on bulkreq.count to check for next inodes, not
> on bstat.bs_ino anymore.
> - for loop in ret_lsize or ret_largest case, now relies on count being 0
> to break the loop
>
> V4:
> - Refactor inode_f function to reduce its size and easier logic
> - Implement error handlers for invalid command combination (hopefully
> all invalid combinations).
> - use a single xfs_inogrp array for keep track of inodes
> - Fix missing newline in inode_help()
> - Rewrite help message in inode_help()
> - Fix indentation
>
> V5:
> - Reduce the amount of options
> - remove igrp_rec variable, and use igroup[lastgrp] directly to get
> information from the last inode groups returned by ioctl
>
> Signed-off-by: Carlos Maiolino <cmaiolino@redhat.com>
> ---
> io/open.c | 143 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> 1 file changed, 143 insertions(+)
>
> diff --git a/io/open.c b/io/open.c
> index ac5a5e0..1e38ea8 100644
> --- a/io/open.c
> +++ b/io/open.c
> @@ -20,6 +20,7 @@
> #include "input.h"
> #include "init.h"
> #include "io.h"
> +#include "libxfs.h"
>
> #ifndef __O_TMPFILE
> #if defined __alpha__
> @@ -44,6 +45,7 @@ static cmdinfo_t statfs_cmd;
> static cmdinfo_t chproj_cmd;
> static cmdinfo_t lsproj_cmd;
> static cmdinfo_t extsize_cmd;
> +static cmdinfo_t inode_cmd;
> static prid_t prid;
> static long extsize;
>
> @@ -750,6 +752,136 @@ statfs_f(
> return 0;
> }
>
> +static void
> +inode_help(void)
> +{
> + printf(_(
> +"\n"
> +"Query physical information about the inode"
> +"\n"
> +" Default: -- Return true(1) or false(0) if any inode greater than\n"
> +" 32bits has been found in the filesystem\n"
> +" -v -- verbose mode\n"
> +" Display the number and the physical size (in bits)\n"
> +" of the largest inode in the filesystem\n"
> +"[num] -- Return true(1) or false(0) if the inode [num] is in use\n"
> +" -n [num] -- Return the next valid inode after [num]\n"
> +"\n"));
> +}
> +
> +static int
> +inode_f(
> + int argc,
> + char **argv)
> +{
> + __s32 count = 0;
> + __s32 lastgrp = 0;
> + __u64 last = 0;
> + __u64 lastino = 0;
> + __u64 userino = 0;
> + char *p;
> + int c;
> + int verbose = 0;
> + int ret_next = 0;
> + int cmd = 0;
> + struct xfs_inogrp igroup[1024];
> + struct xfs_fsop_bulkreq bulkreq;
> + struct xfs_bstat bstat;
> +
> + while ((c = getopt(argc, argv, "nv")) != EOF) {
I think we want "n:v" here since -n expects an argument, even if we
don't process the arg here.
> + switch (c) {
> + case 'v':
> + verbose = 1;
> + break;
> + case 'n':
> + ret_next = 1;
> + break;
> + default:
> + return command_usage(&inode_cmd);
> + }
> + }
> +
> + if (ret_next && verbose)
> + return command_usage(&inode_cmd);
> +
Why is this not supported? Hmm, I see that -n returns an inode number
and otherwise we print 0/1 or <inode>:<size> with -v. Perhaps this would
be easier if the command semantics/output were more consistent. E.g.,
"inode": print 0/1 based on largest inode size
"inode -v": print <ino>:<size> of largest inode
"inode <ino>": print <ino> if inode exists
"inode -v <ino>": print <ino>:<size> if inode exists
"inode -n <ino>": print <next ino> if next inode exists
"inode -nv <ino>": print <next ino>:<size> if next inode exists
In other words, the default behavior is to identify the 32-bit/64-bit
state of the fs. If an inode is provided, we print the inode number if
the inode exists. The -n flags alters this behavior to find the next
inode. The -v flag alters the previous two situations to also print the
inode size.
> + if (optind < argc) {
A comment above this check to explain what it means (i.e., user passed
an inode number) would be nice.
> + if (verbose)
> + return command_usage(&inode_cmd);
Also, why is this not supported (see above)?
> +
> + if (ret_next) {
> + cmd = XFS_IOC_FSBULKSTAT;
> + } else {
> + if (argc > 2)
> + return command_usage(&inode_cmd);
> + else
> + cmd = XFS_IOC_FSBULKSTAT_SINGLE;
> + }
> +
> + userino = strtoull(argv[optind], &p, 10);
> + if ((*p != '\0')) {
> + printf(_("[num] must be a numeric value\n"));
> + exitcode = 1;
> + return 0;
> + }
> +
> + bulkreq.lastip = &userino;
> + bulkreq.icount = 1;
> + bulkreq.ubuffer = &bstat;
> + bulkreq.ocount = &count;
> +
> + if (xfsctl(file->name, file->fd, cmd, &bulkreq)) {
> + if (errno == EINVAL) {
> + if (!ret_next)
> + printf("0\n");
> + } else {
> + perror("xfsctl");
> + }
> + exitcode = 1;
> + return 0;
> + }
> +
> + if (ret_next) {
> + printf("%llu\n", bstat.bs_ino);
> + return 0;
> + } else {
> + /* Inode number used*/
> + printf("1\n");
> + return 0;
> + }
The return 0 can go after the if/else.
> + }
> +
/*
* The user has not provided an inode number. Therefore, find
* the largest inode in the fs.
*/
Brian
> + bulkreq.lastip = &last;
> + bulkreq.icount = 1024; /* User-defined maybe!? */
> + bulkreq.ubuffer = &igroup;
> + bulkreq.ocount = &count;
> +
> + for (;;) {
> + if (xfsctl(file->name, file->fd, XFS_IOC_FSINUMBERS,
> + &bulkreq)) {
> + perror("XFS_IOC_FSINUMBERS");
> + exitcode = 1;
> + return 0;
> + }
> +
> + if (count == 0)
> + break;
> +
> + lastgrp = count;
> + }
> +
> + lastgrp--;
> + lastino = igroup[lastgrp].xi_startino +
> + xfs_highbit64(igroup[lastgrp].xi_allocmask);
> +
> + if (verbose)
> + printf("%llu:%d\n", lastino,
> + lastino > XFS_MAXINUMBER_32 ? 64 : 32);
> + else
> + printf("%d\n", lastino > XFS_MAXINUMBER_32 ? 1 : 0);
> +
> + return 0;
> +}
> +
> void
> open_init(void)
> {
> @@ -815,6 +947,16 @@ open_init(void)
> _("get/set preferred extent size (in bytes) for the open file");
> extsize_cmd.help = extsize_help;
>
> + inode_cmd.name = "inode";
> + inode_cmd.cfunc = inode_f;
> + inode_cmd.args = _("[-n | -v] [num]");
> + inode_cmd.argmin = 0;
> + inode_cmd.argmax = 2;
> + inode_cmd.flags = CMD_NOMAP_OK;
> + inode_cmd.oneline =
> + _("Query inode number usage in the filesystem");
> + inode_cmd.help = inode_help;
> +
> add_command(&open_cmd);
> add_command(&stat_cmd);
> add_command(&close_cmd);
> @@ -822,4 +964,5 @@ open_init(void)
> add_command(&chproj_cmd);
> add_command(&lsproj_cmd);
> add_command(&extsize_cmd);
> + add_command(&inode_cmd);
> }
> --
> 2.4.3
>
> _______________________________________________
> xfs mailing list
> xfs@oss.sgi.com
> http://oss.sgi.com/mailman/listinfo/xfs
_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH] xfs_io: implement 'inode' command V5
2015-11-30 13:22 ` Brian Foster
@ 2015-11-30 14:26 ` Carlos Maiolino
2015-11-30 16:16 ` Brian Foster
0 siblings, 1 reply; 4+ messages in thread
From: Carlos Maiolino @ 2015-11-30 14:26 UTC (permalink / raw)
To: Brian Foster; +Cc: xfs
>
> I think we want "n:v" here since -n expects an argument, even if we
> don't process the arg here.
Using getopt() to handle the -n argument, will make the inode command having 2
different entry points for the same argument, i.e. the inode number. One as an
argument for -n, and another as an argument for the command itself, like:
inode -n <num>
inode <num>
We need to handle [num] as a stand-alone argument anyway, so, I just don't think
we need to handle the same argument in different ways, which I achieved by not
using [num] as a getopt() argument, but instead, handling [num] 'manually'
according to the options used in getopt().
Not sure if I could be clear or get things more confused :)
> > + if (ret_next && verbose)
> > + return command_usage(&inode_cmd);
> > +
>
> Why is this not supported? Hmm, I see that -n returns an inode number
> and otherwise we print 0/1 or <inode>:<size> with -v. Perhaps this would
> be easier if the command semantics/output were more consistent. E.g.,
>
> "inode": print 0/1 based on largest inode size
> "inode -v": print <ino>:<size> of largest inode
> "inode <ino>": print <ino> if inode exists
> "inode -v <ino>": print <ino>:<size> if inode exists
I thought about this, but I decided to not do it because the command looks a bit
redundant for me when 'inode <ino' was returning 0 or 1. Returning the inode
number itself, if it exists, makes more sense to have a -v option here too.
> "inode -n <ino>": print <next ino> if next inode exists
> "inode -nv <ino>": print <next ino>:<size> if next inode exists
Just FYI, if the 'next inode' doesn't exist (i.e. using the last fs inode as
argument), the ioctl will return 0 in bstat.bs_ino, which, I choose to leave it
as-is, and adding this observation to the man page, instead of returning a
messag like "no more inodes in the fs".
I decided to leave it as-is, because for usage would be easier to parse a '0'
return value from -n argument, than parsing an error message which has the same
meaning of a zeroed return.
Anyway, I'm going add -v to the another options, just please take a look at my
replies regarding the 'inode -n' return value and the reason I didn't use
getopt() to handle -n argument and if you agree or not, so I'll rewrite the
patch to v6 based on this.
Cheers o>
--
Carlos
_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH] xfs_io: implement 'inode' command V5
2015-11-30 14:26 ` Carlos Maiolino
@ 2015-11-30 16:16 ` Brian Foster
0 siblings, 0 replies; 4+ messages in thread
From: Brian Foster @ 2015-11-30 16:16 UTC (permalink / raw)
To: xfs
On Mon, Nov 30, 2015 at 03:26:22PM +0100, Carlos Maiolino wrote:
> >
> > I think we want "n:v" here since -n expects an argument, even if we
> > don't process the arg here.
>
> Using getopt() to handle the -n argument, will make the inode command having 2
> different entry points for the same argument, i.e. the inode number. One as an
> argument for -n, and another as an argument for the command itself, like:
>
> inode -n <num>
> inode <num>
>
> We need to handle [num] as a stand-alone argument anyway, so, I just don't think
> we need to handle the same argument in different ways, which I achieved by not
> using [num] as a getopt() argument, but instead, handling [num] 'manually'
> according to the options used in getopt().
>
> Not sure if I could be clear or get things more confused :)
>
Sure, but I'm just referring to the error case when the user passes -n
without an argument. This should return an error but it doesn't at the
moment. I'm assuming that using "n:" would ensure the error message is
printed without disrupting the other code (e.g., continue to process
[num] manually even though "n:" is passed to getopt()). Is that the
case? If not, the error could be detected/handled manually as well.
Either way, a comment would also be useful here to document the special
handling as you note above.
>
> > > + if (ret_next && verbose)
> > > + return command_usage(&inode_cmd);
> > > +
> >
> > Why is this not supported? Hmm, I see that -n returns an inode number
> > and otherwise we print 0/1 or <inode>:<size> with -v. Perhaps this would
> > be easier if the command semantics/output were more consistent. E.g.,
> >
> > "inode": print 0/1 based on largest inode size
> > "inode -v": print <ino>:<size> of largest inode
> > "inode <ino>": print <ino> if inode exists
> > "inode -v <ino>": print <ino>:<size> if inode exists
>
> I thought about this, but I decided to not do it because the command looks a bit
> redundant for me when 'inode <ino' was returning 0 or 1. Returning the inode
> number itself, if it exists, makes more sense to have a -v option here too.
>
Not sure I follow... AFAICT the command semantics change depending on
whether an inode number is passed or not (irrespective of -n and -v). If
not, we're looking to see if the largest inode is 32-bit or 64-bit. If
an inode number is passed, we're checking to see if an inode exists.
Brian
> > "inode -n <ino>": print <next ino> if next inode exists
> > "inode -nv <ino>": print <next ino>:<size> if next inode exists
>
> Just FYI, if the 'next inode' doesn't exist (i.e. using the last fs inode as
> argument), the ioctl will return 0 in bstat.bs_ino, which, I choose to leave it
> as-is, and adding this observation to the man page, instead of returning a
> messag like "no more inodes in the fs".
>
> I decided to leave it as-is, because for usage would be easier to parse a '0'
> return value from -n argument, than parsing an error message which has the same
> meaning of a zeroed return.
>
>
> Anyway, I'm going add -v to the another options, just please take a look at my
> replies regarding the 'inode -n' return value and the reason I didn't use
> getopt() to handle -n argument and if you agree or not, so I'll rewrite the
> patch to v6 based on this.
>
> Cheers o>
>
> --
> Carlos
_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2015-11-30 16:16 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2015-11-26 15:46 [PATCH] xfs_io: implement 'inode' command V5 Carlos Maiolino
2015-11-30 13:22 ` Brian Foster
2015-11-30 14:26 ` Carlos Maiolino
2015-11-30 16:16 ` Brian Foster
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.