Revert "fsnotify: optionally pass access range in file permission hooks"
PATCH 1/1 Revert "fsnotify: optionally pass access range in file permission hooks"
This reverts commit d9e5d31084b024734e64307521414ef0ae1d5333.
This commit added an extra fsnotify call from rw_verify_area(), which can be a very hot path. In my testing, this adds roughly 5-6% extra overhead per IO, which is quite a lot. As a result, throughput of said test also drops by 5-6%, as it is CPU bound. Looking at perf, it's apparent why:
3.36% [kernel.vmlinux] [k] fsnotify
2.32% [kernel.vmlinux] [k] __fsnotify_parenwhich directly correlates with performance lost.
As the rationale for this patch isn't very strong, revert this commit for now and reclaim the performance.
--- fs/open.c | 4 ---- fs/read_write.c | 10 ++-------- fs/readdir.c | 4 ---- fs/remap_range.c | 8 +------- include/linux/fsnotify.h | 13 ++----------- security/security.c | 8 +++++++- 6 files changed, 12 insertions(+), 35 deletions(-)
fs/open.c +0 -4
@@ -304,10 +304,6 @@ int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)if (ret)return ret; - ret =fsnotify_file_area_perm (file,MAY_WRITE , &offset, len); -if (ret) -return ret; -if (S_ISFIFO (inode->i_mode))return -ESPIPE ;
fs/read_write.c +2 -8
@@ -354,9 +354,6 @@ SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,int rw_verify_area (int read_write,struct file *file,const loff_t *ppos,size_t count) { -int mask = read_write ==READ ?MAY_READ :MAY_WRITE ; -int ret; -if (unlikely ((ssize_t ) count <0 ))return -EINVAL ;
@@ -374,11 +371,8 @@ int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t } } - ret =security_file_permission (file, mask); -if (ret) -return ret; - -return fsnotify_file_area_perm (file, mask, ppos, count); +return security_file_permission (file, + read_write ==READ ?MAY_READ :MAY_WRITE ); }EXPORT_SYMBOL (rw_verify_area);
fs/readdir.c +0 -4
@@ -96,10 +96,6 @@ int iterate_dir(struct file *file, struct dir_context *ctx)if (res) goto out; - res =fsnotify_file_perm (file,MAY_READ ); -if (res) - goto out; - res =down_read_killable (&inode->i_rwsem);if (res) goto out;
fs/remap_range.c +1 -7
@@ -102,9 +102,7 @@ static int generic_remap_checks(struct file *file_in, loff_t pos_in,static int remap_verify_area (struct file *file,loff_t pos,loff_t len,bool write) { -int mask = write ?MAY_WRITE :MAY_READ ;loff_t tmp; -int ret;if (unlikely (pos <0 || len <0 ))return -EINVAL ;
@@ -112,11 +110,7 @@ static int remap_verify_area(struct file *file, loff_t pos, loff_t len,if (unlikely (check_add_overflow (pos, len, &tmp)))return -EINVAL ; - ret =security_file_permission (file, mask); -if (ret) -return ret; - -return fsnotify_file_area_perm (file, mask, &pos, len); +return security_file_permission (file, write ?MAY_WRITE :MAY_READ ); }/*
include/linux/fsnotify.h +2 -11
@@ -101,10 +101,9 @@ static inline int fsnotify_file(struct file *file, __u32 mask) }/* -* fsnotify_file_area_perm - permission hook before access to file range +* fsnotify_file_perm - permission hook before file access */ -static inline int fsnotify_file_area_perm (struct file *file,int perm_mask, -const loff_t *ppos,size_t count) +static inline int fsnotify_file_perm (struct file *file,int perm_mask) {__u32 fsnotify_mask =FS_ACCESS_PERM ;
@@ -121,14 +120,6 @@ static inline int fsnotify_file_area_perm(struct file *file, int perm_mask,return fsnotify_file (file, fsnotify_mask); } -/* -* fsnotify_file_perm - permission hook before file access -*/ -static inline int fsnotify_file_perm (struct file *file,int perm_mask) -{ -return fsnotify_file_area_perm (file, perm_mask,NULL ,0 ); -} -/* * fsnotify_open_perm - permission hook before file open */
security/security.c +7 -1
@@ -2580,7 +2580,13 @@ int security_kernfs_init_security(struct kernfs_node *kn_dir,*/ int security_file_permission (struct file *file,int mask) { -return call_int_hook (file_permission,0 , file, mask); +int ret; + + ret =call_int_hook (file_permission,0 , file, mask); +if (ret) +return ret; + +return fsnotify_file_perm (file, mask); }/**
Signature · 4 lines
2.43.0 -- Jens Axboe
Christian Brauner
maintainerQuoted 11 lines · Jens Axboe
This reverts commit
d9e5d31084b024734e64307521414ef0ae1d5333.This commit added an extra fsnotify call from
rw_verify_area(), which can be a very hot path. In my testing, this adds roughly 5-6% extra overhead per IO, which is quite a lot. As a result, throughput of said test also drops by 5-6%, as it is CPU bound. Looking at perf, it's apparent why:3.36% [kernel.vmlinux] [k] fsnotify 2.32% [kernel.vmlinux] [k] __fsnotify_paren
Quoted 7 lines
which directly correlates with performance lost.
As the rationale for this patch isn't very strong, revert this commit for now and reclaim the performance.
Signed-off-by Jens Axboe <[email protected]>---
Thanks for tracking this down! I think Amir, you, and I came to the conclusion that we can fix this in without having to revert. Amir is sending out a new patch later tonight. I'll get that fixed by the end of the week.
Christian
Quoted 17 lines · Christian Brauner
Quoted 16 lines · Jens Axboe
This reverts commit
d9e5d31084b024734e64307521414ef0ae1d5333.This commit added an extra fsnotify call from
rw_verify_area(), which can be a very hot path. In my testing, this adds roughly 5-6% extra overhead per IO, which is quite a lot. As a result, throughput of said test also drops by 5-6%, as it is CPU bound. Looking at perf, it's apparent why:3.36% [kernel.vmlinux] [k] fsnotify 2.32% [kernel.vmlinux] [k] __fsnotify_parenwhich directly correlates with performance lost.
As the rationale for this patch isn't very strong, revert this commit for now and reclaim the performance.
Quoted 7 lines
Quoted 2 lines
Signed-off-by Jens Axboe <[email protected]>---
Thanks for tracking this down! I think Amir, you, and I came to the conclusion that we can fix this in without having to revert. Amir is sending out a new patch later tonight. I'll get that fixed by the end of the week.
Yep, either revert or that patch fixes things for me.
Signature · 1 line
Jens Axboe