fsnotify: move fsnotify_open() hook into do_dentry_open()
Subsystems (4)
Amir Goldstein
reviewerPATCH 1/1 fsnotify: move fsnotify_open() hook into do_dentry_open()
fsnotify_open() hook is called only from high level system calls context and not called for the very many helpers to open files.
This may makes sense for many of the special file open cases, but it is inconsistent with fsnotify_close() hook that is called for every last fput() of on a file object with FMODE_OPENED.
As a result, it is possible to observe ACCESS, MODIFY and CLOSE events without ever observing an OPEN event.
Fix this inconsistency by replacing all the fsnotify_open() hooks with a single hook inside do_dentry_open().
If there are special cases that would like to opt-out of the possible overhead of fsnotify() call in fsnotify_open(), they would probably also want to avoid the overhead of fsnotify() call in the rest of the fsnotify hooks, so they should be opening that file with the __FMODE_NONOTIFY flag.
However, in the majority of those cases, the s_fsnotify_connectors optimization in fsnotify_parent() would be sufficient to avoid the overhead of fsnotify() call anyway.
---
Jan,
I found out about this problem when I tested the work to remove FMODE_NONOTIFY from overlayfs internal files - even after I enabled notifications on the underlying fs, the LTS tests [2] did not observe the OPEN events.
Because this change is independent of the ovl work and has implications on other subsystems as well (e.g. cachefiles), I think it is better if the change came through your tree.
This change has a potential to regress some micro-benchmarks, so if you could queue it up for soaking in linux-next that would be great.
Thanks,
Amir.
[1] https://lore.kernel.org/linux-fsdevel/[email protected]/
[2] https://github.com/amir73il/ltp/commits/ovl_encode_fid
fs/exec.c | 5 ----- fs/fhandle.c | 1 - fs/open.c | 6 +++++- io_uring/openclose.c | 1 - 4 files changed, 5 insertions(+), 8 deletions(-)
fs/exec.c +0 -5
@@ -152,8 +152,6 @@ SYSCALL_DEFINE1(uselib, const char __user *, library) path_noexec(&file->f_path))) goto exit; -fsnotify_open (file); - error = -ENOEXEC ;read_lock (&binfmt_lock);
@@ -934,9 +932,6 @@ static struct file *do_open_execat(int fd, struct filename *name, int flags)if (err) goto exit; -if (name->name[0 ] !='\0' ) -fsnotify_open (file); - out:return file;
fs/fhandle.c +0 -1
@@ -242,7 +242,6 @@ static long do_handle_open(int mountdirfd, struct file_handle __user *ufh, retval =PTR_ERR (file); }else { retval = fd; -fsnotify_open (file);fd_install (fd, file); }path_put (&path);
fs/open.c +5 -1
@@ -969,6 +969,11 @@ static int do_dentry_open(struct file *f, } } +/* +* Once we return a file with FMODE_OPENED, __fput() will call +* fsnotify_close(), so we need fsnotify_open() here for symetry. +*/ +fsnotify_open (f);return 0 ; cleanup_all:
@@ -1358,7 +1363,6 @@ static long do_sys_openat2(int dfd, const char __user *filename,put_unused_fd (fd); fd =PTR_ERR (f); }else { -fsnotify_open (f);fd_install (fd, f); } }
io_uring/openclose.c +0 -1
@@ -150,7 +150,6 @@ int io_openat2(struct io_kiocb *req, unsigned int issue_flags)if ((issue_flags &IO_URING_F_NONBLOCK ) && !nonblock_set) file->f_flags &= ~O_NONBLOCK ; -fsnotify_open (file);if (!fixed)fd_install (ret, file);
Signature · 1 line
2.34.1
Christian Brauner
maintainerQuoted 92 lines · Amir Goldstein
fsnotify_open()hook is called only from high level system calls context and not called for the very many helpers to open files.This may makes sense for many of the special file open cases, but it is inconsistent with
fsnotify_close()hook that is called for every lastfput()of on a file object withFMODE_OPENED.As a result, it is possible to observe ACCESS, MODIFY and CLOSE events without ever observing an OPEN event.
Fix this inconsistency by replacing all the
fsnotify_open()hooks with a single hook insidedo_dentry_open().If there are special cases that would like to opt-out of the possible overhead of
fsnotify()call infsnotify_open(), they would probably also want to avoid the overhead offsnotify()call in the rest of the fsnotify hooks, so they should be opening that file with the__FMODE_NONOTIFYflag.However, in the majority of those cases, the
s_fsnotify_connectorsoptimization infsnotify_parent()would be sufficient to avoid the overhead offsnotify()call anyway.Signed-off-by Amir Goldstein <[email protected]>---
Jan,
I found out about this problem when I tested the work to remove
FMODE_NONOTIFYfrom overlayfs internal files - even after I enabled notifications on the underlying fs, the LTS tests [2] did not observe the OPEN events.Because this change is independent of the ovl work and has implications on other subsystems as well (e.g. cachefiles), I think it is better if the change came through your tree.
This change has a potential to regress some micro-benchmarks, so if you could queue it up for soaking in linux-next that would be great.
Thanks,
Amir.[1] https://lore.kernel.org/linux-fsdevel/[email protected]/
[2] https://github.com/amir73il/ltp/commits/ovl_encode_fidfs/exec.c | 5 ----- fs/fhandle.c | 1 - fs/open.c | 6 +++++- io_uring/openclose.c | 1 - 4 files changed, 5 insertions(+), 8 deletions(-)fs/exec.c +0
-5@@ -152,8 +152,6 @@ SYSCALL_DEFINE1(uselib, const char __user *, library) path_noexec(&file->f_path))) goto exit; -fsnotify_open (file); - error = -ENOEXEC ;read_lock (&binfmt_lock);@@ -934,9 +932,6 @@ static struct file *do_open_execat(int fd, struct filename *name, int flags)if (err) goto exit; -if (name->name[0 ] !='\0' ) -fsnotify_open (file); - out:return file;fs/fhandle.c +0
-1@@ -242,7 +242,6 @@ static long do_handle_open(int mountdirfd, struct file_handle __user *ufh, retval =PTR_ERR (file); }else { retval = fd; -fsnotify_open (file);fd_install (fd, file); }path_put (&path);fs/open.c +0
-0
fs/open.c static int do_dentry_open(struct file *f, · line 974
Quoted 6 lines
fs/open.c +3
-0@@ -969,6 +969,11 @@ static int do_dentry_open(struct file *f, } } +/* +* Once we return a file with FMODE_OPENED, __fput() will call +* fsnotify_close(), so we need fsnotify_open() here for symetry.
s/symetry/symmetry/
Jan Kara
reviewerQuoted 18 lines · Amir Goldstein
fsnotify_open()hook is called only from high level system calls context and not called for the very many helpers to open files.This may makes sense for many of the special file open cases, but it is inconsistent with
fsnotify_close()hook that is called for every lastfput()of on a file object withFMODE_OPENED.As a result, it is possible to observe ACCESS, MODIFY and CLOSE events without ever observing an OPEN event.
Fix this inconsistency by replacing all the
fsnotify_open()hooks with a single hook insidedo_dentry_open().If there are special cases that would like to opt-out of the possible overhead of
fsnotify()call infsnotify_open(), they would probably also want to avoid the overhead offsnotify()call in the rest of the fsnotify hooks, so they should be opening that file with the__FMODE_NONOTIFYflag.
Quoted 5 lines
However, in the majority of those cases, the
s_fsnotify_connectorsoptimization infsnotify_parent()would be sufficient to avoid the overhead offsnotify()call anyway.Signed-off-by Amir Goldstein <[email protected]>
Thanks! The cleanup looks nice so I've applied it with the typo fixup from Christian. I have a slight worry this might break something subtle somewhere but after searching for a while I didn't find anything and the machine boots and ltp tests pass so it's worth a try :)
Honza
Quoted 103 lines
---
Jan,
I found out about this problem when I tested the work to remove
FMODE_NONOTIFYfrom overlayfs internal files - even after I enabled notifications on the underlying fs, the LTS tests [2] did not observe the OPEN events.Because this change is independent of the ovl work and has implications on other subsystems as well (e.g. cachefiles), I think it is better if the change came through your tree.
This change has a potential to regress some micro-benchmarks, so if you could queue it up for soaking in linux-next that would be great.
Thanks,
Amir.[1] https://lore.kernel.org/linux-fsdevel/[email protected]/
[2] https://github.com/amir73il/ltp/commits/ovl_encode_fidfs/exec.c | 5 ----- fs/fhandle.c | 1 - fs/open.c | 6 +++++- io_uring/openclose.c | 1 - 4 files changed, 5 insertions(+), 8 deletions(-)fs/exec.c +0
-5@@ -152,8 +152,6 @@ SYSCALL_DEFINE1(uselib, const char __user *, library) path_noexec(&file->f_path))) goto exit; -fsnotify_open (file); - error = -ENOEXEC ;read_lock (&binfmt_lock);@@ -934,9 +932,6 @@ static struct file *do_open_execat(int fd, struct filename *name, int flags)if (err) goto exit; -if (name->name[0 ] !='\0' ) -fsnotify_open (file); - out:return file;fs/fhandle.c +0
-1@@ -242,7 +242,6 @@ static long do_handle_open(int mountdirfd, struct file_handle __user *ufh, retval =PTR_ERR (file); }else { retval = fd; -fsnotify_open (file);fd_install (fd, file); }path_put (&path);fs/open.c +5
-1@@ -969,6 +969,11 @@ static int do_dentry_open(struct file *f, } } +/* +* Once we return a file with FMODE_OPENED, __fput() will call +* fsnotify_close(), so we need fsnotify_open() here for symetry. +*/ +fsnotify_open (f);return 0 ; cleanup_all:@@ -1358,7 +1363,6 @@ static long do_sys_openat2(int dfd, const char __user *filename,put_unused_fd (fd); fd =PTR_ERR (f); }else { -fsnotify_open (f);fd_install (fd, f); } }io_uring/openclose.c +0
-1@@ -150,7 +150,6 @@ int io_openat2(struct io_kiocb *req, unsigned int issue_flags)if ((issue_flags &IO_URING_F_NONBLOCK ) && !nonblock_set) file->f_flags &= ~O_NONBLOCK ; -fsnotify_open (file);if (!fixed)fd_install (ret, file);Signature · 1 line
2.34.1
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Christian Brauner
maintainerQuoted 23 lines · Jan Kara
Quoted 22 lines · Amir Goldstein
fsnotify_open()hook is called only from high level system calls context and not called for the very many helpers to open files.This may makes sense for many of the special file open cases, but it is inconsistent with
fsnotify_close()hook that is called for every lastfput()of on a file object withFMODE_OPENED.As a result, it is possible to observe ACCESS, MODIFY and CLOSE events without ever observing an OPEN event.
Fix this inconsistency by replacing all the
fsnotify_open()hooks with a single hook insidedo_dentry_open().If there are special cases that would like to opt-out of the possible overhead of
fsnotify()call infsnotify_open(), they would probably also want to avoid the overhead offsnotify()call in the rest of the fsnotify hooks, so they should be opening that file with the__FMODE_NONOTIFYflag.However, in the majority of those cases, the
s_fsnotify_connectorsoptimization infsnotify_parent()would be sufficient to avoid the overhead offsnotify()call anyway.
Quoted 6 lines
Quoted 1 line
Signed-off-by Amir Goldstein <[email protected]>Thanks! The cleanup looks nice so I've applied it with the typo fixup from Christian. I have a slight worry this might break something subtle somewhere but after searching for a while I didn't find anything and the machine boots and ltp tests pass so it's worth a try :)
Yep, I agree. If we can reduce cluttering multiple places with fsnotify_open() and instead move it to a central location it's a maintenance win in the long term.