Hello David, et al.,
I merged your statx(2) page, and edited somewhat heavily.
(The merged page source has been pushed to Git.)
Could you please carefully review the text below, in case
I added any errors.
There is one question in a FIXME below. Could you please
take a look at that also.
Your proposed page duplicated a lot of content from stat(2).
I like to avoid such redundancy, so I move the common pieces
into a new page, inode(7), and reworked stat(2) and statx(2).
Cheers,
Michael
NAME
statx - get file status (extended)
SYNOPSIS
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <fcntl.h> /* Definition of AT_* constants */
int statx(int dirfd, const char *pathname, int flags,
unsigned int mask, struct statx *buf);
Note: There is no glibc wrapper for renameat2(); see NOTES.
DESCRIPTION
This function returns information about a file, storing it in
the buffer pointed to by buf. The returned buffer is a struc‐
ture of the following type:
struct statx {
__u32 stx_mask; /* Mask of bits indicating
filled fields */
__u32 stx_blksize; /* Block size for filesystem I/O */
__u64 stx_attributes; /* Extra file attribute indicators */
__u32 stx_nlink; /* Number of hard links */
__u32 stx_uid; /* User ID of owner */
__u32 stx_gid; /* Group ID of owner */
__u16 stx_mode; /* File type and mode */
__u64 stx_ino; /* Inode number */
__u64 stx_size; /* Total size in bytes */
__u64 stx_blocks; /* Number of 512B blocks allocated */
/* The following fields are file timestamps */
struct statx_timestamp stx_atime; /* Last access */
struct statx_timestamp stx_btime; /* Creation */
struct statx_timestamp stx_ctime; /* Last status change */
struct statx_timestamp stx_mtime; /* Last modification */
/* If this file represents a device, then the next two
fields contain the ID of the device */
__u32 stx_rdev_major; /* Major ID */
__u32 stx_rdev_minor; /* Minor ID */
/* The next two fields contain the ID of the device
containing the filesystem where the file resides */
__u32 stx_dev_major; /* Major ID */
__u32 stx_dev_minor; /* Minor ID */
};
The file timestamps are structures of the following type:
struct statx_timestamp {
__s64 tv_sec; /* Seconds since the Epoch (UNIX time) */
__s32 tv_nsec; /* Nanoseconds before or since tv_sec */
};
(Note that reserved space and padding is omitted.)
Invoking statx():
To access a file's status, no permissions are required on the
file itself, but in the case of statx() with a pathname, exe‐
cute (search) permission is required on all of the directories
in pathname that lead to the file.
statx() uses pathname, dirfd, and flags identify the target
file in one of the following ways:
An absolute pathname
If pathname begins with a slash, then it is an absolute
pathname that identifies the target file. In this case,
dirfd is ignored.
A relative pathname
If pathname is a string that begins with a character
other than a slash and dirfd is AT_FDCWD, then pathname
is a relative pathname that is interpreted relative to
the process's current working directory.
A pathname interpreted relative to a directory file descriptor
If pathname is a string that begins with a character
other than a slash and dirfd is a file descriptor that
refers to a directory, then pathname is a relative path‐
name that is interpreted relative to the directory
referred to by dirfd.
By file descriptor
If pathname is NULL, then the target file is the one
referred to by the file descriptor dirfd. dirfd may
refer to any type of file, not just a directory. (The
AT_EMPTY_PATH flag described below provides similar
functionality.)
┌─────────────────────────────────────────────────────┐
│FIXME │
├─────────────────────────────────────────────────────┤
│It appears that there are two different ways of │
│doing the same thing: specifying the file to be │
│stat'ed via a file descriptor. Either, we specify │
│'pathname' as NULL, or we specify 'pathname' as an │
│empty string and include the AT_EMPTY_PATH flag. │
│What's the rationale for having two ways of doing │
│this? │
└─────────────────────────────────────────────────────┘
flags can be used to influence a pathname-based lookup. A
value for flags is constructed by ORing together zero or more
of the following constants:
AT_EMPTY_PATH
If pathname is an empty string, operate on the file
referred to by dirfd (which may have been obtained using
the open(2) O_PATH flag). If dirfd is AT_FDCWD, the
call operates on the current working directory. In this
case, dirfd can refer to any type of file, not just a
directory. This flag is Linux-specific; define
_GNU_SOURCE to obtain its definition.
AT_NO_AUTOMOUNT
Don't automount the terminal ("basename") component of
pathname if it is a directory that is an automount
point. This allows the caller to gather attributes of
an automount point (rather than the location it would
mount). This flag can be used in tools that scan direc‐
tories to prevent mass-automounting of a directory of
automount points. The AT_NO_AUTOMOUNT flag has no
effect if the mount point has already been mounted over.
This flag is Linux-specific; define _GNU_SOURCE to
obtain its definition.
AT_SYMLINK_NOFOLLOW
If pathname is a symbolic link, do not dereference it:
instead return information about the link itself, like
lstat(2).
flags can also be used to control what sort of synchronization
the kernel will do when querying a file on a remote filesystem.
This is done by ORing in one of the following values:
AT_STATX_SYNC_AS_STAT
Do whatever stat(2) does. This is the default and is
very much filesystem specific.
AT_STATX_FORCE_SYNC
Force the attributes to be synchronized with the server.
This may require that a network filesystem perform a
data writeback to get the timestamps correct.
AT_STATX_DONT_SYNC
Don't synchronize anything, but rather just take what‐
ever the system has cached if possible. This may mean
that the information returned is approximate, but, on a
network filesystem, it may not involve a round trip to
the server - even if no lease is held.
The mask argument to statx() is used to tell the kernel which
fields the caller is interested in. mask is an ORed combina‐
tion of the following constants:
STATX_TYPE Want stx_mode & S_IFMT
STATX_MODE Want stx_mode & ~S_IFMT
STATX_NLINK Want stx_nlink
STATX_UID Want stx_uid
STATX_GID Want stx_gid
STATX_ATIME Want stx_atime
STATX_MTIME Want stx_mtime
STATX_CTIME Want stx_ctime
STATX_INO Want stx_ino
STATX_SIZE Want stx_size
STATX_BLOCKS Want stx_blocks
STATX_BASIC_STATS [All of the above]
STATX_BTIME Want stx_btime
STATX_ALL [All currently available fields]
Note the kernel does not reject values in mask other than the
above. Instead, it simply informs the caller which values are
supported by this kernel and filesystem via the statx.stx_mask
field. Therefore, do not simply set mask to UINT_MAX (all bits
set), as one or more bits may, in the future, be used to spec‐
ify an extension to the buffer.
The returned information
The status information for the target file is returned in the
statx structure pointed to by buf. Included in this is
stx_mask which indicates what other information has been
returned. stx_mask has the same format as the mask argument
and bits are set in it to indicate which fields have been
filled in.
It should be noted that the kernel may return fields that
weren't requested and may fail to return fields that were
requested, depending on what the backing filesystem supports.
In either case, stx_mask will not be equal mask.
If a filesystem does not support a field or if it has an unrep‐
resentable value (for instance, a file with an exotic type),
then the mask bit corresponding to that field will be cleared
in stx_mask even if the user asked for it and a dummy value
will be filled in for compatibility purposes if one is avail‐
able (e.g., a dummy UID and GID may be specified to mount under
some circumstances).
A filesystem may also fill in fields that the caller didn't ask
for if it has values for them available and the information is
available at no extra cost. If this happens, the corresponding
bits will be set in stx_mask.
Note: for performance and simplicity reasons, different fields
in the statx structure may contain state information from dif‐
ferent moments during the execution of the system call. For
example, if stx_mode or stx_uid is changed by another process
by calling chmod(2) or chown(2), stat() might return the old
stx_mode together with the new stx_uid, or the old stx_uid
together with the new stx_mode.
Apart from stx_mask (which is described above), the fields in
the statx structure are:
stx_mode
The file type and mode. See inode(7) for details.
stx_size
The size of the file (if it is a regular file or a sym‐
bolic link) in bytes. The size of a symbolic link is
the length of the pathname it contains, without a termi‐
nating null byte.
stx_blocks
The number of blocks allocated to the file on the
medium, in 512-byte units. (This may be smaller than
stx_size/512 when the file has holes.)
stx_blksize
The "preferred" block size for efficient filesystem I/O.
(Writing to a file in smaller chunks may cause an inef‐
ficient read-modify-rewrite.)
stx_nlink
The number of hard links on a file.
stx_uid
The user ID of the file's owner.
stx_gid
The ID of the group that may access the file.
stx_dev_major and stx_dev_minor
The device on which this file (inode) resides.
stx_rdev_major and stx_rdev_minor
The device that this file (inode) represents if the file
is of block or character device type.
stx_attributes
Further status information about the file (see below for
more information).
stx_atime
The file's last access timestamp.
stx_btime
The file's creation timestamp.
stx_ctime
The file's last status change timestamp.
stx_mtime
The file's last modification timestamp.
For further information on the above fields, see inode(7).
File attributes
The stx_attributes field contains a set of ORed flags that
indicate additional attributes of the file:
STATX_ATTR_COMPRESSED
The file is compressed by the fs and may take extra
resources to access.
STATX_ATTR_IMMUTABLE
The file cannot be modified: it cannot be deleted or
renamed, no hard links can be created to this file and
no data can be written to it. See chattr(1).
STATX_ATTR_APPEND
The file can only be opened in append mode for writing.
Random access writing is not permitted. See chattr(1).
STATX_ATTR_NODUMP
File is not a candidate for backup when a backup program
such as dump(8) is run. See chattr(1).
STATX_ATTR_ENCRYPTED
A key is required for the file to be encrypted by the
filesystem.
RETURN VALUE
On success, zero is returned. On error, -1 is returned, and
errno is set appropriately.
ERRORS
EACCES Search permission is denied for one of the directories
in the path prefix of pathname. (See also path_resolu‐
tion(7).)
EBADF dirfd is not a valid open file descriptor.
EFAULT Bad address.
EINVAL Invalid flag specified in flags.
ELOOP Too many symbolic links encountered while traversing the
pathname.
ENAMETOOLONG
pathname is too long.
ENOENT A component of pathname does not exist, or pathname is
an empty string and AT_EMPTY_PATH was not specified in
flags.
ENOMEM Out of memory (i.e., kernel memory).
ENOTDIR
A component of the path prefix of pathname is not a
directory or pathname is relative and dirfd is a file
descriptor referring to a file other than a directory.
VERSIONS
statx() was added to Linux in kernel 4.11.
CONFORMING TO
statx() is Linux specific.
NOTES
Glibc does not (yet) provide a wrapper for the statx() system
call; call it using syscall(2).
SEE ALSO
ls(1), stat(1), access(2), chmod(2), chown(2), stat(2), read‐
link(2), utime(2), capabilities(7), inode(7), symlink(7)
--
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Linux/UNIX System Programming Training: http://man7.org/training/
Hi Michael
On Tue, Apr 25, 2017 at 01:14:26PM +0200, Michael Kerrisk (man-pages) wrote:
>
> [...]
>
> Could you please carefully review the text below, in case
> I added any errors.
Just a few comments below.
> [...]
>
> Invoking statx():
> To access a file's status, no permissions are required on the
> file itself, but in the case of statx() with a pathname, exe‐
> cute (search) permission is required on all of the directories
> in pathname that lead to the file.
>
> statx() uses pathname, dirfd, and flags identify the target
This should be:
"statx() uses pathname, dirfd, and flags *to* identify the target"
> [...]
>
> AT_STATX_SYNC_AS_STAT
> Do whatever stat(2) does. This is the default and is
> very much filesystem specific.
Since we write "Linux-specific" further above we should probably use
"very much filesystem-specific."
here for consistency.
> AT_STATX_FORCE_SYNC
> Force the attributes to be synchronized with the server.
> This may require that a network filesystem perform a
> data writeback to get the timestamps correct.
>
> AT_STATX_DONT_SYNC
> Don't synchronize anything, but rather just take what‐
> ever the system has cached if possible. This may mean
> that the information returned is approximate, but, on a
> network filesystem, it may not involve a round trip to
> the server - even if no lease is held.
>
> The mask argument to statx() is used to tell the kernel which
> fields the caller is interested in. mask is an ORed combina‐
> tion of the following constants:
>
> STATX_TYPE Want stx_mode & S_IFMT
> STATX_MODE Want stx_mode & ~S_IFMT
> STATX_NLINK Want stx_nlink
> STATX_UID Want stx_uid
> STATX_GID Want stx_gid
> STATX_ATIME Want stx_atime
> STATX_MTIME Want stx_mtime
> STATX_CTIME Want stx_ctime
> STATX_INO Want stx_ino
> STATX_SIZE Want stx_size
> STATX_BLOCKS Want stx_blocks
> STATX_BASIC_STATS [All of the above]
> STATX_BTIME Want stx_btime
> STATX_ALL [All currently available fields]
>
> Note the kernel does not reject values in mask other than the
We should probably either insert a colon here
"Note: the kernel does not reject values in mask other than the..."
or reformulate the sentence like this.
"Note that the kernel does not reject values in mask other than the..."
> [...]
>
> stx_gid
> The ID of the group that may access the file.
This group ID is the gid of the file owner's primary group, no? At least
that's what the field comment in the DESCRIPTION implies.
I think it would be more accurate to write:
"The ID of the file owner's primary group"
> [...]
>
> STATX_ATTR_COMPRESSED
> The file is compressed by the fs and may take extra
We write out 'filesystem' everywhere else so I think we should replace 'fs'
with it here as well.
> CONFORMING TO
> statx() is Linux specific.
For consistency we should write:
"statx() is Linux-specific".
I wrote the changes in-line but if you prefer I can 'git send-email'
a patch as well.
Cheers and thanks for all the hard work!
Silvan
Hello Silvan,
On 04/25/2017 08:50 PM, Silvan Jegen wrote:
> Hi Michael
>
> On Tue, Apr 25, 2017 at 01:14:26PM +0200, Michael Kerrisk (man-pages) wrote:
>>
>> [...]
>>
>> Could you please carefully review the text below, in case
>> I added any errors.
>
> Just a few comments below.
>
>
>> [...]
>>
>> Invoking statx():
>> To access a file's status, no permissions are required on the
>> file itself, but in the case of statx() with a pathname, exe‐
>> cute (search) permission is required on all of the directories
>> in pathname that lead to the file.
>>
>> statx() uses pathname, dirfd, and flags identify the target
>
> This should be:
>
> "statx() uses pathname, dirfd, and flags *to* identify the target"
Fixed.
>> [...]
>>
>> AT_STATX_SYNC_AS_STAT
>> Do whatever stat(2) does. This is the default and is
>> very much filesystem specific.
>
> Since we write "Linux-specific" further above we should probably use
>
> "very much filesystem-specific."
Fixed.
> here for consistency.
>
>
>> AT_STATX_FORCE_SYNC
>> Force the attributes to be synchronized with the server.
>> This may require that a network filesystem perform a
>> data writeback to get the timestamps correct.
>>
>> AT_STATX_DONT_SYNC
>> Don't synchronize anything, but rather just take what‐
>> ever the system has cached if possible. This may mean
>> that the information returned is approximate, but, on a
>> network filesystem, it may not involve a round trip to
>> the server - even if no lease is held.
>>
>> The mask argument to statx() is used to tell the kernel which
>> fields the caller is interested in. mask is an ORed combina‐
>> tion of the following constants:
>>
>> STATX_TYPE Want stx_mode & S_IFMT
>> STATX_MODE Want stx_mode & ~S_IFMT
>> STATX_NLINK Want stx_nlink
>> STATX_UID Want stx_uid
>> STATX_GID Want stx_gid
>> STATX_ATIME Want stx_atime
>> STATX_MTIME Want stx_mtime
>> STATX_CTIME Want stx_ctime
>> STATX_INO Want stx_ino
>> STATX_SIZE Want stx_size
>> STATX_BLOCKS Want stx_blocks
>> STATX_BASIC_STATS [All of the above]
>> STATX_BTIME Want stx_btime
>> STATX_ALL [All currently available fields]
>>
>> Note the kernel does not reject values in mask other than the
>
> We should probably either insert a colon here
>
> "Note: the kernel does not reject values in mask other than the..."
>
> or reformulate the sentence like this.
>
> "Note that the kernel does not reject values in mask other than the..."
I changed to the second suggestion.
>
>
>> [...]
>>
>> stx_gid
>> The ID of the group that may access the file.
>
> This group ID is the gid of the file owner's primary group, no? At least
> that's what the field comment in the DESCRIPTION implies.
>
> I think it would be more accurate to write:
>
> "The ID of the file owner's primary group"
I made it:
This field contains the ID of the group owner of the file.
>
>> [...]
>>
>> STATX_ATTR_COMPRESSED
>> The file is compressed by the fs and may take extra
>
> We write out 'filesystem' everywhere else so I think we should replace 'fs'
> with it here as well.
Indeed! (Fixed.)
>> CONFORMING TO
>> statx() is Linux specific.
>
> For consistency we should write:
>
> "statx() is Linux-specific".
Fixed.
> I wrote the changes in-line but if you prefer I can 'git send-email'
> a patch as well.
This form of feedback is fine.
> Cheers and thanks for all the hard work!
You're welcome. Thanks for checking it.
Cheers,
Michael
--
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Linux/UNIX System Programming Training: http://man7.org/training/
On Tue, Apr 25, 2017 at 01:14:26PM +0200, Michael Kerrisk (man-pages) wrote:
[...]
> The file timestamps are structures of the following type:
>
> struct statx_timestamp {
> __s64 tv_sec; /* Seconds since the Epoch (UNIX time) */
> __s32 tv_nsec; /* Nanoseconds before or since tv_sec */
> };
Nanoseconds since tv_sec. The "before tv_sec" is not implemented yet
and I hope nobody is going to implement it for the reasons I mentioned
in https://marc.info/?i=20170422192743.GA17005%40altlinux.org
--
ldv
On 04/25/2017 10:06 PM, Dmitry V. Levin wrote:
> On Tue, Apr 25, 2017 at 01:14:26PM +0200, Michael Kerrisk (man-pages) wrote:
> [...]
>> The file timestamps are structures of the following type:
>>
>> struct statx_timestamp {
>> __s64 tv_sec; /* Seconds since the Epoch (UNIX time) */
>> __s32 tv_nsec; /* Nanoseconds before or since tv_sec */
>> };
>
> Nanoseconds since tv_sec. The "before tv_sec" is not implemented yet
> and I hope nobody is going to implement it for the reasons I mentioned
> in https://marc.info/?i=20170422192743.GA17005%40altlinux.org
Thanks Dmitry. Fixed.
Cheers,
Michael
--
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Linux/UNIX System Programming Training: http://man7.org/training/
At 2017-04-26T07:42:12+0200, Michael Kerrisk (man-pages) wrote:
> On 04/25/2017 10:06 PM, Dmitry V. Levin wrote:
> > On Tue, Apr 25, 2017 at 01:14:26PM +0200, Michael Kerrisk (man-pages) wrote:
> > [...]
> >> The file timestamps are structures of the following type:
> >>
> >> struct statx_timestamp {
> >> __s64 tv_sec; /* Seconds since the Epoch (UNIX time) */
> >> __s32 tv_nsec; /* Nanoseconds before or since tv_sec */
> >> };
> >
> > Nanoseconds since tv_sec. The "before tv_sec" is not implemented yet
> > and I hope nobody is going to implement it for the reasons I mentioned
> > in https://marc.info/?i=20170422192743.GA17005%40altlinux.org
>
> Thanks Dmitry. Fixed.
If you give a programmer a signed type, he or she will cheerfully stuff
negative values into it, perhaps to carry out-of-band information as
with IEEE NaNs and tagged pointers.
The user should be admonished against using the full range of the type
in any case ([0, 1e9 - 1]).
Regards,
Branden
Michael Kerrisk (man-pages) <[email protected]> wrote:
> Note: There is no glibc wrapper for renameat2(); see NOTES.
statx, not renameat2.
> __u64 stx_blocks; /* Number of 512B blocks allocated */
The following needs to be added in here:
__u64 stx_attributes_mask; /* Mask to show what's supported in stx_attributes */
This indicates what stx_attributes the VFS and filesystem actually support.
> __s32 tv_nsec; /* Nanoseconds before or since tv_sec */
If you're going to do Dmitry's suggestion, then this needs to be __u32 and you
should remove "before or".
> statx() uses pathname, dirfd, and flags identify the target
> file in one of the following ways:
"to identify"
> A pathname interpreted relative to a directory file descriptor
I think this is too wordy for a heading and it almost seems to merge into the
description paragraph since it almost ends at the same right margin. How
about:
A directory-relative pathname
> By file descriptor
> If pathname is NULL, then the target file is the one
> referred to by the file descriptor dirfd. dirfd may
> refer to any type of file, not just a directory. (The
> AT_EMPTY_PATH flag described below provides similar
> functionality.)
>
> ┌───────────────────────────────┐
> │FIXME │
> ├───────────────────────────────┤
> │It appears that there are two different ways of │
> │doing the same thing: specifying the file to be │
> │stat'ed via a file descriptor. Either, we specify │
> │'pathname' as NULL, or we specify 'pathname' as an │
> │empty string and include the AT_EMPTY_PATH flag. │
> │What's the rationale for having two ways of doing │
> │this? │
> └───────────────────────────────┘
AT_EMPTY_PATH wasn't there back in 2010. I could eliminate the:
statx(fd, NULL, 0, ...);
option in favour of:
statx(fd, "", AT_EMPTY_PATH, ...);
Any thoughts either way, Al?
It would seem that AT_EMPTY_PATH should be redundant, though, since you can
just set the pathname pointer to NULL.
> The mask argument to statx() is used to tell the kernel which
> fields the caller is interested in. mask is an ORed combina‐
> tion of the following constants:
>
> STATX_TYPE Want stx_mode & S_IFMT
> ...
> STATX_ALL [All currently available fields]
>
> Note the kernel does not reject values in mask other than the
> above. Instead, it simply informs the caller which values are
> supported by this kernel and filesystem via the statx.stx_mask
> field. Therefore, do not simply set mask to UINT_MAX (all bits
> set), as one or more bits may, in the future, be used to spec‐
> ify an extension to the buffer.
Is it worth mentioning STATX__RESERVED here, I wonder? It's not something
that you can actually use (you'll get EINVAL if you try), and will be renamed
should it become used.
> It should be noted that the kernel may return fields that
> weren't requested and may fail to return fields that were
> requested, depending on what the backing filesystem supports.
Maybe add "and can be safely ignored" in there somewhere since this seems to
be upsetting people.
> If a filesystem does not support a field or if it has an unrep‐
> resentable value (for instance, a file with an exotic type),
> then the mask bit corresponding to that field will be cleared
> in stx_mask even if the user asked for it and a dummy value
> will be filled in for compatibility purposes if one is avail‐
> able (e.g., a dummy UID and GID may be specified to mount under
> some circumstances).
I don't promise a dummy value for any "extended" field other than zero.
> Apart from stx_mask (which is described above), the fields in
> the statx structure are:
>
> stx_mode
> The file type and mode. See inode(7) for details.
> ...
Should this list either be in alphabetical order or offset-in-struct order?
This needs to be added:
stx_attributes_mask
A mask indicating which bits in stx_attributes are supported by
the VFS and the filesystem.
> File attributes
> The stx_attributes field contains a set of ORed flags that
> indicate additional attributes of the file:
Probably should mention stx_attributes_mask here also. Perhaps:
The stx_attributes field contains a set of ORed flags that
indicate additional attributes of the file. Note that any
attribute that is not indicated as supported by
stx_attributes_mask has no usable value here. The bits in
stx_attributes_mask correspond bit-by-bit to stx_attributes.
David
Hello David,
Thanks for your comments. Some questions below.
On 04/26/2017 01:35 PM, David Howells wrote:
>
> Michael Kerrisk (man-pages) <[email protected]> wrote:
>
>> Note: There is no glibc wrapper for renameat2(); see NOTES.
>
> statx, not renameat2.
Already reported, and fixed.
>
>> __u64 stx_blocks; /* Number of 512B blocks allocated */
>
> The following needs to be added in here:
>
> __u64 stx_attributes_mask; /* Mask to show what's supported in stx_attributes */
Done.
> This indicates what stx_attributes the VFS and filesystem actually support.
>
>> __s32 tv_nsec; /* Nanoseconds before or since tv_sec */
>
> If you're going to do Dmitry's suggestion, then this needs to be __u32 and you
> should remove "before or".
I think the question is rather: what is going to be done to the API?
Will it be changed as Dmitry suggests?
>> statx() uses pathname, dirfd, and flags identify the target
>> file in one of the following ways:
>
> "to identify"
Already reported, and fixed.
>> A pathname interpreted relative to a directory file descriptor
>
> I think this is too wordy for a heading and it almost seems to merge into the
> description paragraph since it almost ends at the same right margin. How
> about:
>
> A directory-relative pathname
Yes. Changed as you suggest.
>> By file descriptor
>> If pathname is NULL, then the target file is the one
>> referred to by the file descriptor dirfd. dirfd may
>> refer to any type of file, not just a directory. (The
>> AT_EMPTY_PATH flag described below provides similar
>> functionality.)
>>
>> ┌───────────────────────────────┐
>> │FIXME │
>> ├───────────────────────────────┤
>> │It appears that there are two different ways of │
>> │doing the same thing: specifying the file to be │
>> │stat'ed via a file descriptor. Either, we specify │
>> │'pathname' as NULL, or we specify 'pathname' as an │
>> │empty string and include the AT_EMPTY_PATH flag. │
>> │What's the rationale for having two ways of doing │
>> │this? │
>> └───────────────────────────────┘
>
> AT_EMPTY_PATH wasn't there back in 2010. I could eliminate the:
>
> statx(fd, NULL, 0, ...);
>
> option in favour of:
>
> statx(fd, "", AT_EMPTY_PATH, ...);
>
> Any thoughts either way, Al?
>
> It would seem that AT_EMPTY_PATH should be redundant, though, since you can
> just set the pathname pointer to NULL.
Having two ways to do something is odd, and redundant. Note
of the other APIs that provide this functionality do so
in both ways, AFAIK. It's not a big problem, but certainly
strange. If you settle on having just one, then I'd say
choose AT_EMPTY_PATH.
>> The mask argument to statx() is used to tell the kernel which
>> fields the caller is interested in. mask is an ORed combina‐
>> tion of the following constants:
>>
>> STATX_TYPE Want stx_mode & S_IFMT
>> ...
>> STATX_ALL [All currently available fields]
>>
>> Note the kernel does not reject values in mask other than the
>> above. Instead, it simply informs the caller which values are
>> supported by this kernel and filesystem via the statx.stx_mask
>> field. Therefore, do not simply set mask to UINT_MAX (all bits
>> set), as one or more bits may, in the future, be used to spec‐
>> ify an extension to the buffer.
>
> Is it worth mentioning STATX__RESERVED here, I wonder? It's not something
> that you can actually use (you'll get EINVAL if you try), and will be renamed
> should it become used.
Under ERRORS I added:
.TP
.B EINVAL
Reserved flag specified in
.IR mask .
Okay?
>> It should be noted that the kernel may return fields that
>> weren't requested and may fail to return fields that were
>> requested, depending on what the backing filesystem supports.
>
> Maybe add "and can be safely ignored" in there somewhere since this seems to
> be upsetting people.
You say "in there somewhere", but it's not quite clear to me which piece
this applies to. Could you propose a wording please.
>
>> If a filesystem does not support a field or if it has an unrep‐
>> resentable value (for instance, a file with an exotic type),
>> then the mask bit corresponding to that field will be cleared
>> in stx_mask even if the user asked for it and a dummy value
>> will be filled in for compatibility purposes if one is avail‐
>> able (e.g., a dummy UID and GID may be specified to mount under
>> some circumstances).
>
> I don't promise a dummy value for any "extended" field other than zero.
I don't know what you mean to say here. Do you mean some
text in the page should change?
>> Apart from stx_mask (which is described above), the fields in
>> the statx structure are:
>>
>> stx_mode
>> The file type and mode. See inode(7) for details.
>> ...
>
> Should this list either be in alphabetical order or offset-in-struct order?
Probably the same order as the struct.
> This needs to be added:
>
> stx_attributes_mask
> A mask indicating which bits in stx_attributes are supported by
> the VFS and the filesystem.
Added.
>> File attributes
>> The stx_attributes field contains a set of ORed flags that
>> indicate additional attributes of the file:
>
> Probably should mention stx_attributes_mask here also. Perhaps:
>
> The stx_attributes field contains a set of ORed flags that
> indicate additional attributes of the file. Note that any
> attribute that is not indicated as supported by
> stx_attributes_mask has no usable value here. The bits in
> stx_attributes_mask correspond bit-by-bit to stx_attributes.
Added. But, what does "no usable value here" mean? (The relationship
between stx_attributes_mask and stx_attributes still isn't
so clear to me.
Cheers,
Michael
--
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Linux/UNIX System Programming Training: http://man7.org/training/
Michael Kerrisk (man-pages) <[email protected]> wrote:
> > This indicates what stx_attributes the VFS and filesystem actually support.
> >
> >> __s32 tv_nsec; /* Nanoseconds before or since tv_sec */
> >
> > If you're going to do Dmitry's suggestion, then this needs to be __u32 and you
> > should remove "before or".
>
> I think the question is rather: what is going to be done to the API?
> Will it be changed as Dmitry suggests?
I've forwarded Dmitry's patch to this effect.
> Having two ways to do something is odd, and redundant. Note
> of the other APIs that provide this functionality do so
> in both ways, AFAIK. It's not a big problem, but certainly
> strange. If you settle on having just one, then I'd say
> choose AT_EMPTY_PATH.
If I choose that, I presume I would have to give EINVAL if the path is NULL or
anything other than ""?
> Under ERRORS I added:
>
> .TP
> .B EINVAL
> Reserved flag specified in
> .IR mask .
>
> Okay?
That's fine.
> >> It should be noted that the kernel may return fields that
> >> weren't requested and may fail to return fields that were
> >> requested, depending on what the backing filesystem supports.
> >
> > Maybe add "and can be safely ignored" in there somewhere since this seems to
> > be upsetting people.
>
> You say "in there somewhere", but it's not quite clear to me which piece
> this applies to. Could you propose a wording please.
Can you do footnotes in roff?
It should be noted that the kernel may return fields that
weren't requested[*] and may fail to return fields that were
requested, depending on what the backing filesystem supports.
[*] These can be safely ignored.
Or maybe:
It should be noted that the kernel may return fields that
weren't requested and may fail to return fields that were
requested, depending on what the backing filesystem supports.
Fields that are given values despite being unrequested can just
be ignored.
> >> If a filesystem does not support a field or if it has an unrep‐
> >> resentable value (for instance, a file with an exotic type),
> >> then the mask bit corresponding to that field will be cleared
> >> in stx_mask even if the user asked for it and a dummy value
> >> will be filled in for compatibility purposes if one is avail‐
> >> able (e.g., a dummy UID and GID may be specified to mount under
> >> some circumstances).
> >
> > I don't promise a dummy value for any "extended" field other than zero.
>
> I don't know what you mean to say here. Do you mean some
> text in the page should change?
The paragraph promises a "dummy value will be filled in for compatibility
purposes if one is available", but doesn't place any restriction on the fields
towhich this applies. This is only true of the basic stat fields; all other
fields will be cleared if not set.
Maybe we can just leave this as is. We're not promising a dummy field will
*always* be emplaced. We can always say that they're just not available for
extended fields if someone asks.
Maybe the best thing to do is to simply add "and cleared otherwise." to the
end of the paragraph.
> > Should this list either be in alphabetical order or offset-in-struct order?
>
> Probably the same order as the struct.
Sounds good.
> Added. But, what does "no usable value here" mean? (The relationship
> between stx_attributes_mask and stx_attributes still isn't
> so clear to me.
It's not so obvious with the bits that are currently defined. But I have a
patch that adds Windows attribute bits also (for cifs, ntfs, fat, ...). What
does it mean, say, if the archive bit is clear? Does it mean that archive
isn't set in the fs or that the fs doesn't support it?
Further, I have plans to add a 'setattrx' syscall that takes a statx struct
and calls notify_change() with its contents in the kernel. If I do that, I
need to indicate to notify_change() what changes should be effected. stx_mask
covers most of the fields, but not stx_attributes. Some of these attributes
would be alterable.
Would you prefer it to be reverted for the moment?
David
On Wed, Apr 26, 2017 at 12:35:08PM +0100, David Howells wrote:
> AT_EMPTY_PATH wasn't there back in 2010. I could eliminate the:
>
> statx(fd, NULL, 0, ...);
>
> option in favour of:
>
> statx(fd, "", AT_EMPTY_PATH, ...);
>
> Any thoughts either way, Al?
>
> It would seem that AT_EMPTY_PATH should be redundant, though, since you can
> just set the pathname pointer to NULL.
NULL pathname pointer means an error for a lot of existing syscalls, so
if you want to turn them into wrappers for ...at() ones at libc level,
you'd need to do special-casing of NULL both kernel-side and in libc wrappers.
Requiring "" + AT_EMPTY_PATH means a single dereference of userland pointer.
OTOH, that's not a terrible burden...
Hi David,
On 26 April 2017 at 17:10, David Howells <[email protected]> wrote:
> Michael Kerrisk (man-pages) <[email protected]> wrote:
>
>> > This indicates what stx_attributes the VFS and filesystem actually support.
>> >
>> >> __s32 tv_nsec; /* Nanoseconds before or since tv_sec */
>> >
>> > If you're going to do Dmitry's suggestion, then this needs to be __u32 and you
>> > should remove "before or".
>>
>> I think the question is rather: what is going to be done to the API?
>> Will it be changed as Dmitry suggests?
>
> I've forwarded Dmitry's patch to this effect.
The man page now corresponds.
>> Having two ways to do something is odd, and redundant. Note
>> of the other APIs that provide this functionality do so
>> in both ways, AFAIK. It's not a big problem, but certainly
>> strange. If you settle on having just one, then I'd say
>> choose AT_EMPTY_PATH.
>
> If I choose that, I presume I would have to give EINVAL if the path is NULL or
> anything other than ""?
AFAICS, just set lookup_flags to include LOOKUP_EMPTY and
getname_flags() does the rest. (Essentially, AT_EMPTY_PATH is a safety
catch for an empty path: if the path is nonempty, it is interpreted as
usual, othewise if it is empty, you get ENOENT unless AT_EMPTY_PATH is
also set.
>> Under ERRORS I added:
>>
>> .TP
>> .B EINVAL
>> Reserved flag specified in
>> .IR mask .
>>
>> Okay?
>
> That's fine.
Thanks.
>> >> It should be noted that the kernel may return fields that
>> >> weren't requested and may fail to return fields that were
>> >> requested, depending on what the backing filesystem supports.
>> >
>> > Maybe add "and can be safely ignored" in there somewhere since this seems to
>> > be upsetting people.
>>
>> You say "in there somewhere", but it's not quite clear to me which piece
>> this applies to. Could you propose a wording please.
>
> Can you do footnotes in roff?
>
> It should be noted that the kernel may return fields that
> weren't requested[*] and may fail to return fields that were
> requested, depending on what the backing filesystem supports.
>
> [*] These can be safely ignored.
>
> Or maybe:
>
> It should be noted that the kernel may return fields that
> weren't requested and may fail to return fields that were
> requested, depending on what the backing filesystem supports.
> Fields that are given values despite being unrequested can just
> be ignored.
I took the second approach.
>> >> If a filesystem does not support a field or if it has an unrep‐
>> >> resentable value (for instance, a file with an exotic type),
>> >> then the mask bit corresponding to that field will be cleared
>> >> in stx_mask even if the user asked for it and a dummy value
>> >> will be filled in for compatibility purposes if one is avail‐
>> >> able (e.g., a dummy UID and GID may be specified to mount under
>> >> some circumstances).
>> >
>> > I don't promise a dummy value for any "extended" field other than zero.
>>
>> I don't know what you mean to say here. Do you mean some
>> text in the page should change?
>
> The paragraph promises a "dummy value will be filled in for compatibility
> purposes if one is available", but doesn't place any restriction on the fields
> towhich this applies. This is only true of the basic stat fields; all other
> fields will be cleared if not set.
>
> Maybe we can just leave this as is. We're not promising a dummy field will
> *always* be emplaced. We can always say that they're just not available for
> extended fields if someone asks.
>
> Maybe the best thing to do is to simply add "and cleared otherwise." to the
> end of the paragraph.
Two points:
* You do realize the text about "dummy values" was your original text?
* Adding "and cleared otherwise" to end of the paragraph doesn't make
sense. I'll leave the text as is, but if you want to propose a more
complete phrasing, let me know.
>> > Should this list either be in alphabetical order or offset-in-struct order?
>>
>> Probably the same order as the struct.
>
> Sounds good.
Already done.
>> Added. But, what does "no usable value here" mean? (The relationship
>> between stx_attributes_mask and stx_attributes still isn't
>> so clear to me.
>
> It's not so obvious with the bits that are currently defined. But I have a
> patch that adds Windows attribute bits also (for cifs, ntfs, fat, ...). What
> does it mean, say, if the archive bit is clear? Does it mean that archive
> isn't set in the fs or that the fs doesn't support it?
>
> Further, I have plans to add a 'setattrx' syscall that takes a statx struct
> and calls notify_change() with its contents in the kernel. If I do that, I
> need to indicate to notify_change() what changes should be effected. stx_mask
> covers most of the fields, but not stx_attributes. Some of these attributes
> would be alterable.
>
> Would you prefer it to be reverted for the moment?
To what does "it" refer?
Anyway, I think we do need some better text describing these two
fields and the difference between them. Can you come up with
something?
Cheers,
Michael
--
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Linux/UNIX System Programming Training: http://man7.org/training/