diff mbox

xfs_io: Allow -P and -L to be given to open for O_PATH and O_NOFOLLOW

Message ID 859.1490824354@warthog.procyon.org.uk (mailing list archive)
State Superseded, archived
Headers show

Commit Message

David Howells March 29, 2017, 9:52 p.m. UTC 
Allow "open -P" to specify O_PATH so that paths which would otherwise be
unopenable might be opened for stat()'ing.  Such things include files that
would incur an access error or device files for which no corresponding
driver is available.

Allow "-L" to be given in conjunction with O_PATH to specify O_NOFOLLOW
also.

Signed-off-by: David Howells <dhowells@redhat.com>
---
 io/io.h           |    2 ++
 io/open.c         |   22 ++++++++++++++++++++--
 man/man8/xfs_io.8 |   12 +++++++++++-
 3 files changed, 33 insertions(+), 3 deletions(-)

--
To unsubscribe from this list: send the line "unsubscribe linux-xfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Comments

David Howells March 29, 2017, 11:57 p.m. UTC | #1 
Should XFS_IOC_FSGEOMETRY work on fds that are opened O_PATH on an XFS
filesystem?  Or should xfs_io not call that ioctl on O_PATH fds?

David
--
To unsubscribe from this list: send the line "unsubscribe linux-xfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Darrick J. Wong March 30, 2017, 12:23 a.m. UTC | #2 
On Thu, Mar 30, 2017 at 12:57:08AM +0100, David Howells wrote:
> Should XFS_IOC_FSGEOMETRY work on fds that are opened O_PATH on an XFS
> filesystem?  Or should xfs_io not call that ioctl on O_PATH fds?

Given that read/write don't work under O_PATH, I wouldn't expect ioctls
to work either.

I tried to write a two line program to call FSGEOMETRY to see what
actually happens but got so bogged down in trying to figure out what
magic incantation of C headers one needs to pull in the appropriate
structure typedefs for an ioctl I wasn't even trying to call that I gave
up.

Seriously, what the f*ck do you have to #include just to be able to use
O_PATH?

--D

> 
> David
> --
> To unsubscribe from this list: send the line "unsubscribe linux-xfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-xfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Eric Biggers March 30, 2017, 2:56 a.m. UTC | #3 
On Wed, Mar 29, 2017 at 05:23:28PM -0700, Darrick J. Wong wrote:
> On Thu, Mar 30, 2017 at 12:57:08AM +0100, David Howells wrote:
> > Should XFS_IOC_FSGEOMETRY work on fds that are opened O_PATH on an XFS
> > filesystem?  Or should xfs_io not call that ioctl on O_PATH fds?
> 
> Given that read/write don't work under O_PATH, I wouldn't expect ioctls
> to work either.
> 
> I tried to write a two line program to call FSGEOMETRY to see what
> actually happens but got so bogged down in trying to figure out what
> magic incantation of C headers one needs to pull in the appropriate
> structure typedefs for an ioctl I wasn't even trying to call that I gave
> up.
> 
> Seriously, what the f*ck do you have to #include just to be able to use
> O_PATH?

From open(2):

      "The O_DIRECT, O_NOATIME, O_PATH, and  O_TMPFILE  flags  are
      Linux-specific.  One must define _GNU_SOURCE to obtain their definitions."

Also, ioctl() with an O_PATH file descriptor always fails with EBADF because
sys_ioctl() uses fdget() rather than fdget_raw().  The filesystem isn't involved
at all.  This is intentional, I believe.

- Eric
--
To unsubscribe from this list: send the line "unsubscribe linux-xfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Darrick J. Wong March 30, 2017, 3:25 a.m. UTC | #4 
On Wed, Mar 29, 2017 at 07:56:20PM -0700, Eric Biggers wrote:
> On Wed, Mar 29, 2017 at 05:23:28PM -0700, Darrick J. Wong wrote:
> > On Thu, Mar 30, 2017 at 12:57:08AM +0100, David Howells wrote:
> > > Should XFS_IOC_FSGEOMETRY work on fds that are opened O_PATH on an XFS
> > > filesystem?  Or should xfs_io not call that ioctl on O_PATH fds?
> > 
> > Given that read/write don't work under O_PATH, I wouldn't expect ioctls
> > to work either.
> > 
> > I tried to write a two line program to call FSGEOMETRY to see what
> > actually happens but got so bogged down in trying to figure out what
> > magic incantation of C headers one needs to pull in the appropriate
> > structure typedefs for an ioctl I wasn't even trying to call that I gave
> > up.
> > 
> > Seriously, what the f*ck do you have to #include just to be able to use
> > O_PATH?
> 
> From open(2):
> 
>       "The O_DIRECT, O_NOATIME, O_PATH, and  O_TMPFILE  flags  are
>       Linux-specific.  One must define _GNU_SOURCE to obtain their definitions."

Ah, thank you.  I missed that, having skipped straight to the section on
O_PATH... ($deity I hate userspace)

> Also, ioctl() with an O_PATH file descriptor always fails with EBADF because
> sys_ioctl() uses fdget() rather than fdget_raw().  The filesystem isn't involved
> at all.  This is intentional, I believe.

Yes, it is.

--D

> 
> - Eric
--
To unsubscribe from this list: send the line "unsubscribe linux-xfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
David Howells March 30, 2017, 8:03 a.m. UTC | #5 
Eric Biggers <ebiggers3@gmail.com> wrote:

> Also, ioctl() with an O_PATH file descriptor always fails with EBADF because
> sys_ioctl() uses fdget() rather than fdget_raw().  The filesystem isn't
> involved at all.  This is intentional, I believe.

Yeah.  Apparently so, though some ioctls might make sense if they're asking
about the filesystem.

Anyway, I need to fix xfs_io for that now too if I'm going to use that.  I
wonder what xfs_io needs the geometry for.

I'm getting the impression xfs_io isn't really the tool I should be using for
this.

Dvid
--
To unsubscribe from this list: send the line "unsubscribe linux-xfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Eric Sandeen March 2, 2018, 3:27 a.m. UTC | #6 
On 3/29/17 4:52 PM, David Howells wrote:
> Allow "open -P" to specify O_PATH so that paths which would otherwise be
> unopenable might be opened for stat()'ing.  Such things include files that
> would incur an access error or device files for which no corresponding
> driver is available.
> 
> Allow "-L" to be given in conjunction with O_PATH to specify O_NOFOLLOW
> also.

Hey David - 

Trolling through old patches and finding lots of stuff that fell through
the cracks.  :(  This seems ok, but what is the usecase for it?  Was this
designed for some particular testcase?

Thanks,
-Eric

> Signed-off-by: David Howells <dhowells@redhat.com>
> ---
>  io/io.h           |    2 ++
>  io/open.c         |   22 ++++++++++++++++++++--
>  man/man8/xfs_io.8 |   12 +++++++++++-
>  3 files changed, 33 insertions(+), 3 deletions(-)
> 
> diff --git a/io/io.h b/io/io.h
> index 952bdb8..4399419 100644
> --- a/io/io.h
> +++ b/io/io.h
> @@ -38,6 +38,8 @@
>  #define IO_FOREIGN	(1<<7)
>  #define IO_NONBLOCK	(1<<8)
>  #define IO_TMPFILE	(1<<9)
> +#define IO_PATH		(1<<10)
> +#define IO_NOFOLLOW	(1<<11)
>  
>  /*
>   * Regular file I/O control
> diff --git a/io/open.c b/io/open.c
> index 2ed55cf..f7f508a 100644
> --- a/io/open.c
> +++ b/io/open.c
> @@ -72,6 +72,10 @@ openfile(
>  		oflags |= O_NONBLOCK;
>  	if (flags & IO_TMPFILE)
>  		oflags |= O_TMPFILE;
> +	if (flags & IO_PATH)
> +		oflags |= O_PATH;
> +	if (flags & IO_NOFOLLOW)
> +		oflags |= O_NOFOLLOW;
>  
>  	fd = open(path, oflags, mode);
>  	if (fd < 0) {
> @@ -179,6 +183,8 @@ open_help(void)
>  " -t -- open with O_TRUNC (truncate the file to zero length if it exists)\n"
>  " -R -- mark the file as a realtime XFS file immediately after opening it\n"
>  " -T -- open with O_TMPFILE (create a file not visible in the namespace)\n"
> +" -P -- open with O_PATH (create an fd that is merely a location reference)\n"
> +" -L -- open with O_NOFOLLOW (don't follow symlink)\n"
>  " Note1: usually read/write direct IO requests must be blocksize aligned;\n"
>  "        some kernels, however, allow sectorsize alignment for direct IO.\n"
>  " Note2: the bmap for non-regular files can be obtained provided the file\n"
> @@ -203,7 +209,7 @@ open_f(
>  		return 0;
>  	}
>  
> -	while ((c = getopt(argc, argv, "FRTacdfm:nrstx")) != EOF) {
> +	while ((c = getopt(argc, argv, "FLPRTacdfm:nrstx")) != EOF) {
>  		switch (c) {
>  		case 'F':
>  			/* Ignored / deprecated now, handled automatically */
> @@ -244,6 +250,12 @@ open_f(
>  		case 'T':
>  			flags |= IO_TMPFILE;
>  			break;
> +		case 'P':
> +			flags |= IO_PATH;
> +			break;
> +		case 'L':
> +			flags |= IO_NOFOLLOW;
> +			break;
>  		default:
>  			return command_usage(&open_cmd);
>  		}
> @@ -257,6 +269,12 @@ open_f(
>  		return -1;
>  	}
>  
> +	if ((flags & (IO_PATH|IO_NOFOLLOW)) &&
> +	    (flags & ~(IO_PATH|IO_NOFOLLOW))) {
> +		fprintf(stderr, _("-P and -L are incompatible with the other options\n"));
> +		return -1;
> +	}
> +
>  	fd = openfile(argv[optind], &geometry, flags, mode);
>  	if (fd < 0)
>  		return 0;
> @@ -772,7 +790,7 @@ open_init(void)
>  	open_cmd.argmax = -1;
>  	open_cmd.flags = CMD_NOMAP_OK | CMD_NOFILE_OK |
>  			 CMD_FOREIGN_OK | CMD_FLAG_ONESHOT;
> -	open_cmd.args = _("[-acdrstxT] [-m mode] [path]");
> +	open_cmd.args = _("[-acdrstxRTPL] [-m mode] [path]");
>  	open_cmd.oneline = _("open the file specified by path");
>  	open_cmd.help = open_help;
>  
> diff --git a/man/man8/xfs_io.8 b/man/man8/xfs_io.8
> index 022f0ea..e77be40 100644
> --- a/man/man8/xfs_io.8
> +++ b/man/man8/xfs_io.8
> @@ -122,7 +122,7 @@ command for more details on any command.
>  Display a list of all open files and (optionally) switch to an alternate
>  current open file.
>  .TP
> -.BI "open [[ \-acdfrstRT ] " path " ]"
> +.BI "open [[ \-acdfrstRTPL ] " path " ]"
>  Closes the current file, and opens the file specified by
>  .I path
>  instead. Without any arguments, displays statistics about the current
> @@ -164,6 +164,16 @@ option.
>  .B \-R
>  marks the file as a realtime XFS file after
>  opening it, if it is not already marked as such.
> +.TP
> +.B \-P
> +opens the path as a referent only (O_PATH).  This is incompatible with other
> +flags specifying other O_xxx flags apart from
> +.BR \-L .
> +.TP
> +.B \-L
> +doesn't follow symlinks (O_NOFOLLOW).  This is incompatible with other
> +flags specifying other O_xxx flags apart from
> +.BR \-P .
>  .PD
>  .RE
>  .TP
> --
> To unsubscribe from this list: send the line "unsubscribe linux-xfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 
--
To unsubscribe from this list: send the line "unsubscribe linux-xfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

diff --git a/io/io.h b/io/io.h
index 952bdb8..4399419 100644
--- a/io/io.h
+++ b/io/io.h
@@ -38,6 +38,8 @@ 
 #define IO_FOREIGN	(1<<7)
 #define IO_NONBLOCK	(1<<8)
 #define IO_TMPFILE	(1<<9)
+#define IO_PATH		(1<<10)
+#define IO_NOFOLLOW	(1<<11)
 
 /*
  * Regular file I/O control
diff --git a/io/open.c b/io/open.c
index 2ed55cf..f7f508a 100644
--- a/io/open.c
+++ b/io/open.c
@@ -72,6 +72,10 @@  openfile(
 		oflags |= O_NONBLOCK;
 	if (flags & IO_TMPFILE)
 		oflags |= O_TMPFILE;
+	if (flags & IO_PATH)
+		oflags |= O_PATH;
+	if (flags & IO_NOFOLLOW)
+		oflags |= O_NOFOLLOW;
 
 	fd = open(path, oflags, mode);
 	if (fd < 0) {
@@ -179,6 +183,8 @@  open_help(void)
 " -t -- open with O_TRUNC (truncate the file to zero length if it exists)\n"
 " -R -- mark the file as a realtime XFS file immediately after opening it\n"
 " -T -- open with O_TMPFILE (create a file not visible in the namespace)\n"
+" -P -- open with O_PATH (create an fd that is merely a location reference)\n"
+" -L -- open with O_NOFOLLOW (don't follow symlink)\n"
 " Note1: usually read/write direct IO requests must be blocksize aligned;\n"
 "        some kernels, however, allow sectorsize alignment for direct IO.\n"
 " Note2: the bmap for non-regular files can be obtained provided the file\n"
@@ -203,7 +209,7 @@  open_f(
 		return 0;
 	}
 
-	while ((c = getopt(argc, argv, "FRTacdfm:nrstx")) != EOF) {
+	while ((c = getopt(argc, argv, "FLPRTacdfm:nrstx")) != EOF) {
 		switch (c) {
 		case 'F':
 			/* Ignored / deprecated now, handled automatically */
@@ -244,6 +250,12 @@  open_f(
 		case 'T':
 			flags |= IO_TMPFILE;
 			break;
+		case 'P':
+			flags |= IO_PATH;
+			break;
+		case 'L':
+			flags |= IO_NOFOLLOW;
+			break;
 		default:
 			return command_usage(&open_cmd);
 		}
@@ -257,6 +269,12 @@  open_f(
 		return -1;
 	}
 
+	if ((flags & (IO_PATH|IO_NOFOLLOW)) &&
+	    (flags & ~(IO_PATH|IO_NOFOLLOW))) {
+		fprintf(stderr, _("-P and -L are incompatible with the other options\n"));
+		return -1;
+	}
+
 	fd = openfile(argv[optind], &geometry, flags, mode);
 	if (fd < 0)
 		return 0;
@@ -772,7 +790,7 @@  open_init(void)
 	open_cmd.argmax = -1;
 	open_cmd.flags = CMD_NOMAP_OK | CMD_NOFILE_OK |
 			 CMD_FOREIGN_OK | CMD_FLAG_ONESHOT;
-	open_cmd.args = _("[-acdrstxT] [-m mode] [path]");
+	open_cmd.args = _("[-acdrstxRTPL] [-m mode] [path]");
 	open_cmd.oneline = _("open the file specified by path");
 	open_cmd.help = open_help;
 
diff --git a/man/man8/xfs_io.8 b/man/man8/xfs_io.8
index 022f0ea..e77be40 100644
--- a/man/man8/xfs_io.8
+++ b/man/man8/xfs_io.8
@@ -122,7 +122,7 @@  command for more details on any command.
 Display a list of all open files and (optionally) switch to an alternate
 current open file.
 .TP
-.BI "open [[ \-acdfrstRT ] " path " ]"
+.BI "open [[ \-acdfrstRTPL ] " path " ]"
 Closes the current file, and opens the file specified by
 .I path
 instead. Without any arguments, displays statistics about the current
@@ -164,6 +164,16 @@  option.
 .B \-R
 marks the file as a realtime XFS file after
 opening it, if it is not already marked as such.
+.TP
+.B \-P
+opens the path as a referent only (O_PATH).  This is incompatible with other
+flags specifying other O_xxx flags apart from
+.BR \-L .
+.TP
+.B \-L
+doesn't follow symlinks (O_NOFOLLOW).  This is incompatible with other
+flags specifying other O_xxx flags apart from
+.BR \-P .
 .PD
 .RE
 .TP