fs: harden anon inodes
Subsystems (3)
Christian Brauner
maintainercover fs: harden anon inodes
* Anonymous inodes currently don't come with a proper mode causing issues in the kernel when we want to add useful VFS debug assert. Fix that by giving them a proper mode and masking it off when we report it to userspace which relies on them not having any mode.
* Anonymous inodes currently allow to change inode attributes because the VFS falls back to simple_setattr() if i_op->setattr isn't implemented. This means the ownership and mode for every single user of anon_inode_inode can be changed. Block that as it's either useless or actively harmful. If specific ownership is needed the respective subsystem should allocate anonymous inodes from their own private superblock.
* Port pidfs to the new anon_inode_{g,s}etattr() helpers.
* Add proper tests for anonymous inode behavior.
The anonymous inode specific fixes should ideally be backported to all LTS kernels.
---
Christian Brauner (9):
anon_inode: use a proper mode internally
pidfs: use anon_inode_getattr()
anon_inode: explicitly block ->setattr()
pidfs: use anon_inode_setattr()
anon_inode: raise SB_I_NODEV and SB_I_NOEXEC selftests/filesystems: add first test for anonymous inodes selftests/filesystems: add second test for anonymous inodes selftests/filesystems: add third test for anonymous inodes selftests/filesystems: add fourth test for anonymous inodes
fs/anon_inodes.c | 45 ++++++++++++++ fs/internal.h | 5 ++ fs/libfs.c | 2 +- fs/pidfs.c | 26 +------- tools/testing/selftests/filesystems/.gitignore | 1 + tools/testing/selftests/filesystems/Makefile | 2 +- .../selftests/filesystems/anon_inode_test.c | 69 ++++++++++++++++++++++ 7 files changed, 124 insertions(+), 26 deletions(-)
---
base-commit: 0af2f6be1b4281385b618cb86ad946eded089ac8
change-id: 20250407-work-anon_inode-e22bb1a74992
Christian Brauner
maintainerPATCH 1/9 anon_inode: use a proper mode internally
This allows the VFS to not trip over anonymous inodes and we can add asserts based on the mode into the vfs. When we report it to userspace we can simply hide the mode to avoid regressions. I've audited all direct callers of alloc_anon_inode() and only secretmen overrides i_mode and i_op inode operations but it already uses a regular file.
--- fs/anon_inodes.c | 36 ++++++++++++++++++++++++++++++++++++ fs/internal.h | 3 +++ fs/libfs.c | 2 +- 3 files changed, 40 insertions(+), 1 deletion(-)
fs/anon_inodes.c +36 -0
@@ -24,9 +24,43 @@#include <linux/uaccess.h> +#include "internal.h" +static struct vfsmount *anon_inode_mnt__ro_after_init ;static struct inode *anon_inode_inode__ro_after_init ; +/* +* User space expects anonymous inodes to have no file type in st_mode. +* +* In particular, 'lsof' has this legacy logic: +* +* type = s->st_mode & S_IFMT; +* switch (type) { +* ... +* case 0: +* if (!strcmp(p, "anon_inode")) +* Lf->ntype = Ntype = N_ANON_INODE; +* +* to detect our old anon_inode logic. +* +* Rather than mess with our internal sane inode data, just fix it +* up here in getattr() by masking off the format bits. +*/ +int anon_inode_getattr (struct mnt_idmap *idmap,const struct path *path, +struct kstat *stat,u32 request_mask, +unsigned int query_flags) +{ +struct inode *inode =d_inode (path->dentry); + +generic_fillattr (&nop_mnt_idmap, request_mask, inode, stat); + stat->mode &= ~S_IFMT ; +return 0 ; +} + +static const struct inode_operations anon_inode_operations = { + .getattr = anon_inode_getattr, +}; +/* * anon_inodefs_dname() is called from d_path(). */
@@ -66,6 +100,7 @@ static struct inode *anon_inode_make_secure_inode(if (IS_ERR (inode))return inode; inode->i_flags &= ~S_PRIVATE ; + inode->i_op = &anon_inode_operations; error =security_inode_init_security_anon (inode, &QSTR (name), context_inode);if (error) {
@@ -313,6 +348,7 @@ static int __init anon_inode_init(void) anon_inode_inode =alloc_anon_inode (anon_inode_mnt->mnt_sb);if (IS_ERR (anon_inode_inode))panic ("anon_inode_init() inode allocation failed (%ld)\n" ,PTR_ERR (anon_inode_inode)); + anon_inode_inode->i_op = &anon_inode_operations;return 0 ; }
fs/internal.h +3 -0
@@ -343,3 +343,6 @@ static inline bool path_mounted(const struct path *path)void file_f_owner_release (struct file *file);bool file_seek_cur_needs_f_lock (struct file *file);int statmount_mnt_idmap (struct mnt_idmap *idmap,struct seq_file *seq,bool uid_map); +int anon_inode_getattr (struct mnt_idmap *idmap,const struct path *path, +struct kstat *stat,u32 request_mask, +unsigned int query_flags);
fs/libfs.c +1 -1
@@ -1647,7 +1647,7 @@ struct inode *alloc_anon_inode(struct super_block *s)* that it already _is_ on the dirty list. */ inode->i_state =I_DIRTY ; - inode->i_mode =S_IRUSR |S_IWUSR ; + inode->i_mode =S_IFREG |S_IRUSR |S_IWUSR ; inode->i_uid =current_fsuid (); inode->i_gid =current_fsgid (); inode->i_flags |=S_PRIVATE ;
Signature · 1 line
2.47.2
Re: [PATCH 1/9] anon_inode: use a proper mode internally
Quoted 22 lines · Christian Brauner
This allows the VFS to not trip over anonymous inodes and we can add asserts based on the mode into the vfs. When we report it to userspace we can simply hide the mode to avoid regressions. I've audited all direct callers of
alloc_anon_inode()and only secretmen overridesi_modeandi_opinode operations but it already uses a regular file.Fixes af153bb63a336 ("vfs: catch invalid modes in may_open()")Cc <[email protected]> # all LTS kernelsReported-by [email protected]Signed-off-by Christian Brauner <[email protected]>--- fs/anon_inodes.c | 36 ++++++++++++++++++++++++++++++++++++ fs/internal.h | 3 +++ fs/libfs.c | 2 +- 3 files changed, 40 insertions(+), 1 deletion(-)fs/anon_inodes.c +0
-0
fs/anon_inodes.c line 33
Quoted 10 lines
fs/anon_inodes.c +4
-0@@ -24,9 +24,43 @@#include <linux/uaccess.h> +#include "internal.h" +static struct vfsmount *anon_inode_mnt__ro_after_init ;static struct inode *anon_inode_inode__ro_after_init ; +/* +* User space expects anonymous inodes to have no file type in st_mode.
Weird. Does anything actually depend on this?
ISTM that they should report a file type.
Quoted 74 lines
fs/anon_inodes.c +32
-0+* +* In particular, 'lsof' has this legacy logic: +* +* type = s->st_mode & S_IFMT; +* switch (type) { +* ... +* case 0: +* if (!strcmp(p, "anon_inode")) +* Lf->ntype = Ntype = N_ANON_INODE; +* +* to detect our old anon_inode logic. +* +* Rather than mess with our internal sane inode data, just fix it +* up here in getattr() by masking off the format bits. +*/ +int anon_inode_getattr (struct mnt_idmap *idmap,const struct path *path, +struct kstat *stat,u32 request_mask, +unsigned int query_flags) +{ +struct inode *inode =d_inode (path->dentry); + +generic_fillattr (&nop_mnt_idmap, request_mask, inode, stat); + stat->mode &= ~S_IFMT ; +return 0 ; +} + +static const struct inode_operations anon_inode_operations = { + .getattr = anon_inode_getattr, +}; +/* * anon_inodefs_dname() is called from d_path(). */ @@ -66,6 +100,7 @@ static struct inode *anon_inode_make_secure_inode(if (IS_ERR (inode))return inode; inode->i_flags &= ~S_PRIVATE ; + inode->i_op = &anon_inode_operations; error =security_inode_init_security_anon (inode, &QSTR (name), context_inode);if (error) {@@ -313,6 +348,7 @@ static int __init anon_inode_init(void) anon_inode_inode =alloc_anon_inode (anon_inode_mnt->mnt_sb);if (IS_ERR (anon_inode_inode))panic ("anon_inode_init() inode allocation failed (%ld)\n" ,PTR_ERR (anon_inode_inode)); + anon_inode_inode->i_op = &anon_inode_operations;return 0 ; }fs/internal.h +3
-0@@ -343,3 +343,6 @@ static inline bool path_mounted(const struct path *path)void file_f_owner_release (struct file *file);bool file_seek_cur_needs_f_lock (struct file *file);int statmount_mnt_idmap (struct mnt_idmap *idmap,struct seq_file *seq,bool uid_map); +int anon_inode_getattr (struct mnt_idmap *idmap,const struct path *path, +struct kstat *stat,u32 request_mask, +unsigned int query_flags);fs/libfs.c +1
-1@@ -1647,7 +1647,7 @@ struct inode *alloc_anon_inode(struct super_block *s)* that it already _is_ on the dirty list. */ inode->i_state =I_DIRTY ; - inode->i_mode =S_IRUSR |S_IWUSR ; + inode->i_mode =S_IFREG |S_IRUSR |S_IWUSR ; inode->i_uid =current_fsuid (); inode->i_gid =current_fsgid (); inode->i_flags |=S_PRIVATE ;
Signature · 1 line
Jeff Layton <[email protected]>
Christian Brauner
maintainerRe: [PATCH 1/9] anon_inode: use a proper mode internally
Quoted 31 lines · Jeff Layton
Quoted 30 lines · Christian Brauner
This allows the VFS to not trip over anonymous inodes and we can add asserts based on the mode into the vfs. When we report it to userspace we can simply hide the mode to avoid regressions. I've audited all direct callers of
alloc_anon_inode()and only secretmen overridesi_modeandi_opinode operations but it already uses a regular file.Fixes af153bb63a336 ("vfs: catch invalid modes in may_open()")Cc <[email protected]> # all LTS kernelsReported-by [email protected]Signed-off-by Christian Brauner <[email protected]>--- fs/anon_inodes.c | 36 ++++++++++++++++++++++++++++++++++++ fs/internal.h | 3 +++ fs/libfs.c | 2 +- 3 files changed, 40 insertions(+), 1 deletion(-)fs/anon_inodes.c +2
-0@@ -24,9 +24,43 @@#include <linux/uaccess.h> +#include "internal.h" +static struct vfsmount *anon_inode_mnt__ro_after_init ;static struct inode *anon_inode_inode__ro_after_init ;
Quoted 4 lines
fs/anon_inodes.c line 33
Quoted 2 lines
fs/anon_inodes.c +2
-0+/* +* User space expects anonymous inodes to have no file type in st_mode. Weird. Does anything actually depend on this?
Yeah, lsof failed and started complaining about this. They're checking the mode to recognize anonymous inodes. And tbf, it works to generically recognizer proper anonymous inodes because they came come from very different superblocks (dmabuf, drm, vfio, anon_inode_mnt etc.).
Quoted 2 lines
ISTM that they should report a file type.
I agree but that ship has probably sailed.
Re: [PATCH 1/9] anon_inode: use a proper mode internally
In reply to Christian BraunerQuoted 11 lines · Christian Brauner
This allows the VFS to not trip over anonymous inodes and we can add asserts based on the mode into the vfs. When we report it to userspace we can simply hide the mode to avoid regressions. I've audited all direct callers of
alloc_anon_inode()and only secretmen overridesi_modeandi_opinode operations but it already uses a regular file.Fixes af153bb63a336 ("vfs: catch invalid modes in may_open()")Cc <[email protected]> # all LTS kernelsReported-by [email protected]Signed-off-by Christian Brauner <[email protected]>
Looks good. Feel free to add:
Honza
Quoted 98 lines
--- fs/anon_inodes.c | 36 ++++++++++++++++++++++++++++++++++++ fs/internal.h | 3 +++ fs/libfs.c | 2 +- 3 files changed, 40 insertions(+), 1 deletion(-)fs/anon_inodes.c +36
-0@@ -24,9 +24,43 @@#include <linux/uaccess.h> +#include "internal.h" +static struct vfsmount *anon_inode_mnt__ro_after_init ;static struct inode *anon_inode_inode__ro_after_init ; +/* +* User space expects anonymous inodes to have no file type in st_mode. +* +* In particular, 'lsof' has this legacy logic: +* +* type = s->st_mode & S_IFMT; +* switch (type) { +* ... +* case 0: +* if (!strcmp(p, "anon_inode")) +* Lf->ntype = Ntype = N_ANON_INODE; +* +* to detect our old anon_inode logic. +* +* Rather than mess with our internal sane inode data, just fix it +* up here in getattr() by masking off the format bits. +*/ +int anon_inode_getattr (struct mnt_idmap *idmap,const struct path *path, +struct kstat *stat,u32 request_mask, +unsigned int query_flags) +{ +struct inode *inode =d_inode (path->dentry); + +generic_fillattr (&nop_mnt_idmap, request_mask, inode, stat); + stat->mode &= ~S_IFMT ; +return 0 ; +} + +static const struct inode_operations anon_inode_operations = { + .getattr = anon_inode_getattr, +}; +/* * anon_inodefs_dname() is called from d_path(). */ @@ -66,6 +100,7 @@ static struct inode *anon_inode_make_secure_inode(if (IS_ERR (inode))return inode; inode->i_flags &= ~S_PRIVATE ; + inode->i_op = &anon_inode_operations; error =security_inode_init_security_anon (inode, &QSTR (name), context_inode);if (error) {@@ -313,6 +348,7 @@ static int __init anon_inode_init(void) anon_inode_inode =alloc_anon_inode (anon_inode_mnt->mnt_sb);if (IS_ERR (anon_inode_inode))panic ("anon_inode_init() inode allocation failed (%ld)\n" ,PTR_ERR (anon_inode_inode)); + anon_inode_inode->i_op = &anon_inode_operations;return 0 ; }fs/internal.h +3
-0@@ -343,3 +343,6 @@ static inline bool path_mounted(const struct path *path)void file_f_owner_release (struct file *file);bool file_seek_cur_needs_f_lock (struct file *file);int statmount_mnt_idmap (struct mnt_idmap *idmap,struct seq_file *seq,bool uid_map); +int anon_inode_getattr (struct mnt_idmap *idmap,const struct path *path, +struct kstat *stat,u32 request_mask, +unsigned int query_flags);fs/libfs.c +1
-1@@ -1647,7 +1647,7 @@ struct inode *alloc_anon_inode(struct super_block *s)* that it already _is_ on the dirty list. */ inode->i_state =I_DIRTY ; - inode->i_mode =S_IRUSR |S_IWUSR ; + inode->i_mode =S_IFREG |S_IRUSR |S_IWUSR ; inode->i_uid =current_fsuid (); inode->i_gid =current_fsgid (); inode->i_flags |=S_PRIVATE ;Signature · 2 lines
-- 2.47.2
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Re: [PATCH 1/9] anon_inode: use a proper mode internally
In reply to Christian BraunerQuoted 5 lines · Christian Brauner
This allows the VFS to not trip over anonymous inodes and we can add asserts based on the mode into the vfs. When we report it to userspace we can simply hide the mode to avoid regressions. I've audited all direct callers of
alloc_anon_inode()and only secretmen overridesi_modeandi_opinode operations but it already uses a regular file.
We've been seeing failures in LTP's readadead01 in -next on arm64 platforms:
4601 07:43:36.192033 tst_test.c:1900: TINFO: LTP version: 20250130-1-g60fe84aaf 4602 07:43:36.201811 tst_test.c:1904: TINFO: Tested kernel: 6.15.0-rc1-next-20250410 #1 SMP PREEMPT Thu Apr 10 06:18:38 UTC 2025 aarch64 4603 07:43:36.208400 tst_kconfig.c:88: TINFO: Parsing kernel config '/proc/config.gz' 4604 07:43:36.218393 tst_test.c:1722: TINFO: Overall timeout per run is 0h 01m 30s 4605 07:43:36.223886 readahead01.c:36: TPASS: readahead() with fd = -1 : EBADF (9) 4606 07:43:36.229370 readahead01.c:43: TPASS: readahead() with invalid fd : EBADF (9) 4607 07:43:36.234998 readahead01.c:64: TPASS: readahead() on O_PATH file : EBADF (9) 4608 07:43:36.240527 readahead01.c:64: TPASS: readahead() on directory : EINVAL (22) 4609 07:43:36.246118 readahead01.c:64: TPASS: readahead() on /dev/zero : EINVAL (22) 4610 07:43:36.251530 readahead01.c:64: TPASS: readahead() on pipe read end : EINVAL (22) 4611 07:43:36.260007 readahead01.c:64: TPASS: readahead() on pipe write end : EBADF (9) 4612 07:43:36.265581 readahead01.c:64: TPASS: readahead() on unix socket : EINVAL (22) 4613 07:43:36.270928 readahead01.c:64: TPASS: readahead() on inet socket : EINVAL (22) 4614 07:43:36.276754 readahead01.c:64: TFAIL: readahead() on epoll succeeded 4615 07:43:36.279460 readahead01.c:64: TFAIL: readahead() on eventfd succeeded 4616 07:43:36.285053 readahead01.c:64: TFAIL: readahead() on signalfd succeeded 4617 07:43:36.290504 readahead01.c:64: TFAIL: readahead() on timerfd succeeded 4618 07:43:36.296220 readahead01.c:64: TFAIL: readahead() on fanotify succeeded 4619 07:43:36.301605 readahead01.c:64: TFAIL: readahead() on inotify succeeded 4620 07:43:36.307327 tst_fd.c:170: TCONF: Skipping userfaultfd: ENOSYS (38) 4621 07:43:36.312806 readahead01.c:64: TFAIL: readahead() on perf event succeeded 4622 07:43:36.318534 readahead01.c:64: TFAIL: readahead() on io uring succeeded 4623 07:43:36.321511 readahead01.c:64: TFAIL: readahead() on bpf map succeeded 4624 07:43:36.325711 readahead01.c:64: TFAIL: readahead() on fsopen succeeded 4625 07:43:36.331073 readahead01.c:64: TFAIL: readahead() on fspick succeeded 4626 07:43:36.336608 readahead01.c:64: TPASS: readahead() on open_tree : EBADF (9) 4627 07:43:36.336903 4628 07:43:36.339354 Summary: 4629 07:43:36.339641 passed 10 4630 07:43:36.342132 failed 11 4631 07:43:36.342420 broken 0 4632 07:43:36.342648 skipped 1 4633 07:43:36.344768 warnings 0
which bisected down to this patch, which is cfd86ef7e8e7b9e01 in -next:
git bisect start
# status: waiting for both good and bad commits
# bad: [29e7bf01ed8033c9a14ed0dc990dfe2736dbcd18] Add linux-next specific files for 20250410
git bisect bad 29e7bf01ed8033c9a14ed0dc990dfe2736dbcd18
# status: waiting for good commit(s), bad commit known
# good: [1785a3a7b96a52fae13880a5ba880a5f473eacb1] Merge branch 'for-linux-next-fixes' of https://gitlab.freedesktop.org/drm/misc/kernel.git
git bisect good 1785a3a7b96a52fae13880a5ba880a5f473eacb1
# bad: [793874436825ebf3dfeeac34b75682c234cf61ef] Merge branch 'for-linux-next' of https://gitlab.freedesktop.org/drm/misc/kernel.git
git bisect bad 793874436825ebf3dfeeac34b75682c234cf61ef
# bad: [f8b5c1664191e453611f77d36ba21b09bc468a2d] Merge branch 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/pci/pci.git
git bisect bad f8b5c1664191e453611f77d36ba21b09bc468a2d
# good: [100ac6e209fce471f3ff4d4e92f9d192fcfa7637] Merge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/qcom/linux.git
git bisect good 100ac6e209fce471f3ff4d4e92f9d192fcfa7637
# bad: [143ced925e31fe24e820866403276492f05efaa5] Merge branch 'vfs.all' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs.git
git bisect bad 143ced925e31fe24e820866403276492f05efaa5
# good: [b087fb728fdda75e1d3e83aa542d3aa025ac6c4a] Merge branch 'nfsd-next' of git://git.kernel.org/pub/scm/linux/kernel/git/cel/linux
git bisect good b087fb728fdda75e1d3e83aa542d3aa025ac6c4a
# good: [7ee85aeee98e85f72a663672267180218d1510db] Merge branch 'vfs-6.16.super' into vfs.all
git bisect good 7ee85aeee98e85f72a663672267180218d1510db
# bad: [d57e6ea6671b1ef0fcb09ccc52952c8a6bfb83c8] Merge branch 'vfs-6.16.misc' into vfs.all
git bisect bad d57e6ea6671b1ef0fcb09ccc52952c8a6bfb83c8
# bad: [25a6cc9a630b4b1b783903b23a3a97c5bf16bf41] selftests/filesystems: add open() test for anonymous inodes
git bisect bad 25a6cc9a630b4b1b783903b23a3a97c5bf16bf41
# bad: [c83b9024966090fe0df92aab16975b8d00089e1f] pidfs: use anon_inode_setattr()
git bisect bad c83b9024966090fe0df92aab16975b8d00089e1f
# bad: [cfd86ef7e8e7b9e015707e46479a6b1de141eed0] anon_inode: use a proper mode internally
git bisect bad cfd86ef7e8e7b9e015707e46479a6b1de141eed0
# good: [418556fa576ebbd644c7258a97b33203956ea232] docs: initramfs: update compression and mtime descriptions
git bisect good 418556fa576ebbd644c7258a97b33203956ea232
# first bad commit: [cfd86ef7e8e7b9e015707e46479a6b1de141eed0] anon_inode: use a proper mode internally
Christian Brauner
maintainerRe: [PATCH 1/9] anon_inode: use a proper mode internally
Quoted 9 lines · Mark Brown
Quoted 5 lines · Christian Brauner
This allows the VFS to not trip over anonymous inodes and we can add asserts based on the mode into the vfs. When we report it to userspace we can simply hide the mode to avoid regressions. I've audited all direct callers of
alloc_anon_inode()and only secretmen overridesi_modeandi_opinode operations but it already uses a regular file.We've been seeing failures in LTP's readadead01 in -next on arm64 platforms:
This fscking readhead garbage is driving me insane.
Ok, readahead skipped anonymous inodes because it's checking whether it is a regular file or not. We now make them regular files internally. Should be fixed in -next tomorrow.
Quoted 67 lines
4601 07:43:36.192033 tst_test.c:1900: TINFO: LTP version: 20250130-1-g60fe84aaf 4602 07:43:36.201811 tst_test.c:1904: TINFO: Tested kernel: 6.15.0-rc1-next-20250410 #1 SMP PREEMPT Thu Apr 10 06:18:38 UTC 2025 aarch64 4603 07:43:36.208400 tst_kconfig.c:88: TINFO: Parsing kernel config '/proc/config.gz' 4604 07:43:36.218393 tst_test.c:1722: TINFO: Overall timeout per run is 0h 01m 30s 4605 07:43:36.223886 readahead01.c:36: TPASS: readahead() with fd = -1 : EBADF (9) 4606 07:43:36.229370 readahead01.c:43: TPASS: readahead() with invalid fd : EBADF (9) 4607 07:43:36.234998 readahead01.c:64: TPASS: readahead() on O_PATH file : EBADF (9) 4608 07:43:36.240527 readahead01.c:64: TPASS: readahead() on directory : EINVAL (22) 4609 07:43:36.246118 readahead01.c:64: TPASS: readahead() on /dev/zero : EINVAL (22) 4610 07:43:36.251530 readahead01.c:64: TPASS: readahead() on pipe read end : EINVAL (22) 4611 07:43:36.260007 readahead01.c:64: TPASS: readahead() on pipe write end : EBADF (9) 4612 07:43:36.265581 readahead01.c:64: TPASS: readahead() on unix socket : EINVAL (22) 4613 07:43:36.270928 readahead01.c:64: TPASS: readahead() on inet socket : EINVAL (22) 4614 07:43:36.276754 readahead01.c:64: TFAIL: readahead() on epoll succeeded 4615 07:43:36.279460 readahead01.c:64: TFAIL: readahead() on eventfd succeeded 4616 07:43:36.285053 readahead01.c:64: TFAIL: readahead() on signalfd succeeded 4617 07:43:36.290504 readahead01.c:64: TFAIL: readahead() on timerfd succeeded 4618 07:43:36.296220 readahead01.c:64: TFAIL: readahead() on fanotify succeeded 4619 07:43:36.301605 readahead01.c:64: TFAIL: readahead() on inotify succeeded 4620 07:43:36.307327 tst_fd.c:170: TCONF: Skipping userfaultfd: ENOSYS (38) 4621 07:43:36.312806 readahead01.c:64: TFAIL: readahead() on perf event succeeded 4622 07:43:36.318534 readahead01.c:64: TFAIL: readahead() on io uring succeeded 4623 07:43:36.321511 readahead01.c:64: TFAIL: readahead() on bpf map succeeded 4624 07:43:36.325711 readahead01.c:64: TFAIL: readahead() on fsopen succeeded 4625 07:43:36.331073 readahead01.c:64: TFAIL: readahead() on fspick succeeded 4626 07:43:36.336608 readahead01.c:64: TPASS: readahead() on open_tree : EBADF (9) 4627 07:43:36.336903 4628 07:43:36.339354 Summary: 4629 07:43:36.339641 passed 10 4630 07:43:36.342132 failed 11 4631 07:43:36.342420 broken 0 4632 07:43:36.342648 skipped 1 4633 07:43:36.344768 warnings 0which bisected down to this patch, which is
cfd86ef7e8e7b9e01in -next:git bisect start
# status: waiting for both good and bad commits
# bad: [29e7bf01ed8033c9a14ed0dc990dfe2736dbcd18] Add linux-next specific files for 20250410
git bisect bad29e7bf01ed8033c9a14ed0dc990dfe2736dbcd18
# status: waiting for good commit(s), bad commit known
# good: [1785a3a7b96a52fae13880a5ba880a5f473eacb1] Merge branch 'for-linux-next-fixes' of https://gitlab.freedesktop.org/drm/misc/kernel.git
git bisect good1785a3a7b96a52fae13880a5ba880a5f473eacb1
# bad: [793874436825ebf3dfeeac34b75682c234cf61ef] Merge branch 'for-linux-next' of https://gitlab.freedesktop.org/drm/misc/kernel.git
git bisect bad793874436825ebf3dfeeac34b75682c234cf61ef
# bad: [f8b5c1664191e453611f77d36ba21b09bc468a2d] Merge branch 'next' ofgit://git.kernel.org/pub/scm/linux/kernel/git/pci/pci.git
git bisect badf8b5c1664191e453611f77d36ba21b09bc468a2d
# good: [100ac6e209fce471f3ff4d4e92f9d192fcfa7637] Merge branch 'for-next' ofgit://git.kernel.org/pub/scm/linux/kernel/git/qcom/linux.git
git bisect good100ac6e209fce471f3ff4d4e92f9d192fcfa7637
# bad: [143ced925e31fe24e820866403276492f05efaa5] Merge branch 'vfs.all' ofgit://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs.git
git bisect bad143ced925e31fe24e820866403276492f05efaa5
# good: [b087fb728fdda75e1d3e83aa542d3aa025ac6c4a] Merge branch 'nfsd-next' ofgit://git.kernel.org/pub/scm/linux/kernel/git/cel/linux
git bisect goodb087fb728fdda75e1d3e83aa542d3aa025ac6c4a
# good: [7ee85aeee98e85f72a663672267180218d1510db] Merge branch 'vfs-6.16.super' into vfs.all
git bisect good7ee85aeee98e85f72a663672267180218d1510db
# bad: [d57e6ea6671b1ef0fcb09ccc52952c8a6bfb83c8] Merge branch 'vfs-6.16.misc' into vfs.all
git bisect badd57e6ea6671b1ef0fcb09ccc52952c8a6bfb83c8
# bad: [25a6cc9a630b4b1b783903b23a3a97c5bf16bf41] selftests/filesystems: addopen()test for anonymous inodes
git bisect bad25a6cc9a630b4b1b783903b23a3a97c5bf16bf41
# bad: [c83b9024966090fe0df92aab16975b8d00089e1f] pidfs: useanon_inode_setattr()
git bisect badc83b9024966090fe0df92aab16975b8d00089e1f
# bad: [cfd86ef7e8e7b9e015707e46479a6b1de141eed0]anon_inode: use a proper mode internally
git bisect badcfd86ef7e8e7b9e015707e46479a6b1de141eed0
# good: [418556fa576ebbd644c7258a97b33203956ea232] docs: initramfs: update compression and mtime descriptions
git bisect good418556fa576ebbd644c7258a97b33203956ea232# first bad commit: [cfd86ef7e8e7b9e015707e46479a6b1de141eed0] anon_inode: use a proper mode internally
Re: [PATCH 1/9] anon_inode: use a proper mode internally
Quoted 8 lines · Christian Brauner
Quoted 7 lines · Mark Brown
Quoted 5 lines · Christian Brauner
This allows the VFS to not trip over anonymous inodes and we can add asserts based on the mode into the vfs. When we report it to userspace we can simply hide the mode to avoid regressions. I've audited all direct callers of
alloc_anon_inode()and only secretmen overridesi_modeandi_opinode operations but it already uses a regular file.
Quoted 7 lines
Quoted 2 lines
We've been seeing failures in LTP's readadead01 in -next on arm64 platforms:
This fscking readhead garbage is driving me insane.
Ok, readahead skipped anonymous inodes because it's checking whether it is a regular file or not. We now make them regular files internally. Should be fixed in -next tomorrow.
Is this the readahead syscall? Yeah that random check in the high level code looks odd if that's what is being triggered here.
Re: [PATCH 1/9] anon_inode: use a proper mode internally
In reply to Christian BraunerQuoted 11 lines · Christian Brauner
This allows the VFS to not trip over anonymous inodes and we can add asserts based on the mode into the vfs. When we report it to userspace we can simply hide the mode to avoid regressions. I've audited all direct callers of
alloc_anon_inode()and only secretmen overridesi_modeandi_opinode operations but it already uses a regular file.Fixes af153bb63a336 ("vfs: catch invalid modes in may_open()")Cc <[email protected]> # all LTS kernelsReported-by [email protected]Signed-off-by Christian Brauner <[email protected]>
Hi Christian and FSdevel list,
I'm reporting a regression introduced in the linux-next tree, which I've tracked down using git bisect to this commit.
Starting with linux-next tag next-20250408 (up to the latest next-20250417 I tested), attempting to start an SDDM Wayland session (using KDE Plasma's KWin) results in a black screen. The kwin_wayland and sddm-helper-start-wayland processes both enter a state consuming 100% CPU on a single core indefinitely, preventing the login screen from appearing. The last known good kernel is next-20250407.
Using strace on the kwin_wayland process revealed it's stuck in a tight loop involving ppoll and ioctl:
```
ioctl(29, FIONREAD, [0]) = 0
ppoll([{fd=10, events=POLLIN}, {fd=37, events=POLLIN}, {fd=9, events=POLLIN}, {fd=5, events=POLLIN}, {fd=29, events=POLLIN}, {fd=17, events=POLLIN}, {fd=3, events=POLLIN}], 7, NULL, NULL, 8) = 1 ([{fd=29, revents=POLLIN}])
ioctl(29, FIONREAD, [0]) = 0
ppoll([{fd=10, events=POLLIN}, {fd=37, events=POLLIN}, {fd=9, events=POLLIN}, {fd=5, events=POLLIN}, {fd=29, events=POLLIN}, {fd=17, events=POLLIN}, {fd=3, events=POLLIN}], 7, NULL, NULL, 8) = 1 ([{fd=29, revents=POLLIN}])
# ... this repeats indefinitely at high frequency
```Initially, I suspected a DRM issue, but checking the file descriptor confirmed it's an inotify instance:
``` $ sudo ls -l /proc/<kwin_pid>/fd/29 lr-x------ 1 sddm sddm 64 Apr 17 14:03 /proc/<kwin_pid>/fd/29 -> anon_inode:inotify ```
Reverting this commit resolves the issue, allowing SDDM and KWin Wayland to start normally.
Could you please take a look at this? Let me know if you need any further information or testing from my side.
Signature · 2 lines
Best regards, Xilin Wu <[email protected]>
Christian Brauner
maintainerRe: [PATCH 1/9] anon_inode: use a proper mode internally
Quoted 51 lines · Xilin Wu
Quoted 11 lines · Christian Brauner
This allows the VFS to not trip over anonymous inodes and we can add asserts based on the mode into the vfs. When we report it to userspace we can simply hide the mode to avoid regressions. I've audited all direct callers of
alloc_anon_inode()and only secretmen overridesi_modeandi_opinode operations but it already uses a regular file.Fixes af153bb63a336 ("vfs: catch invalid modes in may_open()")Cc <[email protected]> # all LTS kernelsReported-by [email protected]Signed-off-by Christian Brauner <[email protected]>Hi Christian and FSdevel list,
I'm reporting a regression introduced in the linux-next tree, which I've tracked down using
git bisectto this commit.Starting with
linux-nexttagnext-20250408(up to the latestnext-20250417I tested), attempting to start an SDDM Wayland session (using KDE Plasma's KWin) results in a black screen. Thekwin_waylandandsddm-helper-start-waylandprocesses both enter a state consuming 100% CPU on a single core indefinitely, preventing the login screen from appearing. The last known good kernel isnext-20250407.Using
straceon thekwin_waylandprocess revealed it's stuck in a tight loop involvingppollandioctl:``` ioctl(29, FIONREAD, [0]) = 0 ppoll([{fd=10, events=POLLIN}, {fd=37, events=POLLIN}, {fd=9, events=POLLIN}, {fd=5, events=POLLIN}, {fd=29, events=POLLIN}, {fd=17, events=POLLIN}, {fd=3, events=POLLIN}], 7, NULL, NULL, 8) = 1 ([{fd=29, revents=POLLIN}]) ioctl(29, FIONREAD, [0]) = 0 ppoll([{fd=10, events=POLLIN}, {fd=37, events=POLLIN}, {fd=9, events=POLLIN}, {fd=5, events=POLLIN}, {fd=29, events=POLLIN}, {fd=17, events=POLLIN}, {fd=3, events=POLLIN}], 7, NULL, NULL, 8) = 1 ([{fd=29, revents=POLLIN}]) # ... this repeats indefinitely at high frequency ```Initially, I suspected a DRM issue, but checking the file descriptor confirmed it's an
inotifyinstance:``` $ sudo ls -l /proc/<kwin_pid>/fd/29 lr-x------ 1 sddm sddm 64 Apr 17 14:03 /proc/<kwin_pid>/fd/29 -> anon_inode:inotify ```
Quoted 5 lines
Reverting this commit resolves the issue, allowing SDDM and KWin Wayland to start normally.
Could you please take a look at this? Let me know if you need any further information or testing from my side.
I'll take a look latest Tuesday.
Christian Brauner
maintainerRe: [PATCH 1/9] anon_inode: use a proper mode internally
Quoted 52 lines · Christian Brauner
Quoted 51 lines · Xilin Wu
Quoted 11 lines · Christian Brauner
This allows the VFS to not trip over anonymous inodes and we can add asserts based on the mode into the vfs. When we report it to userspace we can simply hide the mode to avoid regressions. I've audited all direct callers of
alloc_anon_inode()and only secretmen overridesi_modeandi_opinode operations but it already uses a regular file.Fixes af153bb63a336 ("vfs: catch invalid modes in may_open()")Cc <[email protected]> # all LTS kernelsReported-by [email protected]Signed-off-by Christian Brauner <[email protected]>Hi Christian and FSdevel list,
I'm reporting a regression introduced in the linux-next tree, which I've tracked down using
git bisectto this commit.Starting with
linux-nexttagnext-20250408(up to the latestnext-20250417I tested), attempting to start an SDDM Wayland session (using KDE Plasma's KWin) results in a black screen. Thekwin_waylandandsddm-helper-start-waylandprocesses both enter a state consuming 100% CPU on a single core indefinitely, preventing the login screen from appearing. The last known good kernel isnext-20250407.Using
straceon thekwin_waylandprocess revealed it's stuck in a tight loop involvingppollandioctl:``` ioctl(29, FIONREAD, [0]) = 0 ppoll([{fd=10, events=POLLIN}, {fd=37, events=POLLIN}, {fd=9, events=POLLIN}, {fd=5, events=POLLIN}, {fd=29, events=POLLIN}, {fd=17, events=POLLIN}, {fd=3, events=POLLIN}], 7, NULL, NULL, 8) = 1 ([{fd=29, revents=POLLIN}]) ioctl(29, FIONREAD, [0]) = 0 ppoll([{fd=10, events=POLLIN}, {fd=37, events=POLLIN}, {fd=9, events=POLLIN}, {fd=5, events=POLLIN}, {fd=29, events=POLLIN}, {fd=17, events=POLLIN}, {fd=3, events=POLLIN}], 7, NULL, NULL, 8) = 1 ([{fd=29, revents=POLLIN}]) # ... this repeats indefinitely at high frequency ```Initially, I suspected a DRM issue, but checking the file descriptor confirmed it's an
inotifyinstance:``` $ sudo ls -l /proc/<kwin_pid>/fd/29 lr-x------ 1 sddm sddm 64 Apr 17 14:03 /proc/<kwin_pid>/fd/29 -> anon_inode:inotify ```
Quoted 7 lines
Quoted 5 lines
Reverting this commit resolves the issue, allowing SDDM and KWin Wayland to start normally.
Could you please take a look at this? Let me know if you need any further information or testing from my side.
I'll take a look latest Tuesday.
Fixed. We need to ensure that we call the ioctl handler for anonymous inodes and not the generic FIONREAD call.
Christian Brauner
maintainerPATCH 2/9 pidfs: use anon_inode_getattr()
In reply to Christian BraunerSo far pidfs did use it's own version. Just use the generic version. We use our own wrappers because we're going to be implementing our own retrieval properties soon.
--- fs/pidfs.c | 24 +----------------------- 1 file changed, 1 insertion(+), 23 deletions(-)
fs/pidfs.c +1 -23
@@ -572,33 +572,11 @@ static int pidfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,return -EOPNOTSUPP ; } - -/* -* User space expects pidfs inodes to have no file type in st_mode. -* -* In particular, 'lsof' has this legacy logic: -* -* type = s->st_mode & S_IFMT; -* switch (type) { -* ... -* case 0: -* if (!strcmp(p, "anon_inode")) -* Lf->ntype = Ntype = N_ANON_INODE; -* -* to detect our old anon_inode logic. -* -* Rather than mess with our internal sane inode data, just fix it -* up here in getattr() by masking off the format bits. -*/ static int pidfs_getattr (struct mnt_idmap *idmap,const struct path *path,struct kstat *stat,u32 request_mask,unsigned int query_flags) { -struct inode *inode =d_inode (path->dentry); - -generic_fillattr (&nop_mnt_idmap, request_mask, inode, stat); - stat->mode &= ~S_IFMT ; -return 0 ; +return anon_inode_getattr (idmap, path, stat, request_mask, query_flags); }static const struct inode_operations pidfs_inode_operations = {
Signature · 1 line
2.47.2
Re: [PATCH 2/9] pidfs: use anon_inode_getattr()
Quoted 5 lines · Christian Brauner
So far pidfs did use it's own version. Just use the generic version. We use our own wrappers because we're going to be implementing our own retrieval properties soon.
Signed-off-by Christian Brauner <[email protected]>
Looks good. Feel free to add:
Honza
Quoted 47 lines
--- fs/pidfs.c | 24 +----------------------- 1 file changed, 1 insertion(+), 23 deletions(-)fs/pidfs.c +1
-23@@ -572,33 +572,11 @@ static int pidfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,return -EOPNOTSUPP ; } - -/* -* User space expects pidfs inodes to have no file type in st_mode. -* -* In particular, 'lsof' has this legacy logic: -* -* type = s->st_mode & S_IFMT; -* switch (type) { -* ... -* case 0: -* if (!strcmp(p, "anon_inode")) -* Lf->ntype = Ntype = N_ANON_INODE; -* -* to detect our old anon_inode logic. -* -* Rather than mess with our internal sane inode data, just fix it -* up here in getattr() by masking off the format bits. -*/ static int pidfs_getattr (struct mnt_idmap *idmap,const struct path *path,struct kstat *stat,u32 request_mask,unsigned int query_flags) { -struct inode *inode =d_inode (path->dentry); - -generic_fillattr (&nop_mnt_idmap, request_mask, inode, stat); - stat->mode &= ~S_IFMT ; -return 0 ; +return anon_inode_getattr (idmap, path, stat, request_mask, query_flags); }static const struct inode_operations pidfs_inode_operations = {Signature · 2 lines
-- 2.47.2
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Christian Brauner
maintainerPATCH 3/9 anon_inode: explicitly block ->setattr()
In reply to Christian BraunerIt is currently possible to change the mode and owner of the single anonymous inode in the kernel:
int main(int argc, char *argv[])
{
int ret, sfd;
sigset_t mask;
struct signalfd_siginfo fdsi;
sigemptyset(&mask);
sigaddset(&mask, SIGINT);
sigaddset(&mask, SIGQUIT);
ret = sigprocmask(SIG_BLOCK, &mask, NULL);
if (ret < 0)
_exit(1);
sfd = signalfd(-1, &mask, 0);
if (sfd < 0)
_exit(2);
ret = fchown(sfd, 5555, 5555);
if (ret < 0)
_exit(3);
ret = fchmod(sfd, 0777);
if (ret < 0)
_exit(3);
_exit(4);
}This is a bug. It's not really a meaningful one because anonymous inodes don't really figure into path lookup and they cannot be reopened via /proc/<pid>/fd/<nr> and can't be used for lookup itself. So they can only ever serve as direct references.
But it is still completely bogus to allow the mode and ownership or any of the properties of the anonymous inode to be changed. Block this!
--- fs/anon_inodes.c | 7 +++++++ fs/internal.h | 2 ++ 2 files changed, 9 insertions(+)
fs/anon_inodes.c +7 -0
@@ -57,8 +57,15 @@ int anon_inode_getattr(struct mnt_idmap *idmap, const struct path *path,return 0 ; } +int anon_inode_setattr (struct mnt_idmap *idmap,struct dentry *dentry, +struct iattr *attr) +{ +return -EOPNOTSUPP ; +} +static const struct inode_operations anon_inode_operations = { .getattr = anon_inode_getattr, + .setattr = anon_inode_setattr, };/*
fs/internal.h +2 -0
@@ -346,3 +346,5 @@ int statmount_mnt_idmap(struct mnt_idmap *idmap, struct seq_file *seq, bool uid_int anon_inode_getattr (struct mnt_idmap *idmap,const struct path *path,struct kstat *stat,u32 request_mask,unsigned int query_flags); +int anon_inode_setattr (struct mnt_idmap *idmap,struct dentry *dentry, +struct iattr *attr);
Signature · 1 line
2.47.2
Re: [PATCH 3/9] anon_inode: explicitly block ->setattr()
Quoted 37 lines · Christian Brauner
It is currently possible to change the mode and owner of the single anonymous inode in the kernel:
int
main(int argc, char *argv[]){ int ret, sfd; sigset_t mask; struct signalfd_siginfo fdsi; sigemptyset(&mask); sigaddset(&mask, SIGINT); sigaddset(&mask, SIGQUIT); ret = sigprocmask(SIG_BLOCK, &mask, NULL); if (ret < 0) _exit(1); sfd = signalfd(-1, &mask, 0); if (sfd < 0) _exit(2); ret = fchown(sfd, 5555, 5555); if (ret < 0) _exit(3); ret = fchmod(sfd, 0777); if (ret < 0) _exit(3); _exit(4); }This is a bug. It's not really a meaningful one because anonymous inodes don't really figure into path lookup and they cannot be reopened via /proc/<pid>/fd/<nr> and can't be used for lookup itself. So they can only ever serve as direct references.
Quoted 5 lines
But it is still completely bogus to allow the mode and ownership or any of the properties of the anonymous inode to be changed. Block this!
Cc <[email protected]> # all LTS kernelsSigned-off-by Christian Brauner <[email protected]>
Definitely. Feel free to add:
Honza
Quoted 39 lines
--- fs/anon_inodes.c | 7 +++++++ fs/internal.h | 2 ++ 2 files changed, 9 insertions(+)fs/anon_inodes.c +7
-0@@ -57,8 +57,15 @@ int anon_inode_getattr(struct mnt_idmap *idmap, const struct path *path,return 0 ; } +int anon_inode_setattr (struct mnt_idmap *idmap,struct dentry *dentry, +struct iattr *attr) +{ +return -EOPNOTSUPP ; +} +static const struct inode_operations anon_inode_operations = { .getattr = anon_inode_getattr, + .setattr = anon_inode_setattr, };/* fs/internal.h +2
-0@@ -346,3 +346,5 @@ int statmount_mnt_idmap(struct mnt_idmap *idmap, struct seq_file *seq, bool uid_int anon_inode_getattr (struct mnt_idmap *idmap,const struct path *path,struct kstat *stat,u32 request_mask,unsigned int query_flags); +int anon_inode_setattr (struct mnt_idmap *idmap,struct dentry *dentry, +struct iattr *attr);Signature · 2 lines
-- 2.47.2
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Christian Brauner
maintainerPATCH 4/9 pidfs: use anon_inode_setattr()
In reply to Christian BraunerSo far pidfs did use it's own version. Just use the generic version. We use our own wrappers because we're going to be implementing properties soon.
--- fs/pidfs.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
fs/pidfs.c +1 -1
@@ -569,7 +569,7 @@ static struct vfsmount *pidfs_mnt __ro_after_init;static int pidfs_setattr (struct mnt_idmap *idmap,struct dentry *dentry,struct iattr *attr) { -return -EOPNOTSUPP ; +return anon_inode_setattr (idmap, dentry, attr); } static int pidfs_getattr(struct mnt_idmap *idmap, const struct path *path,
Signature · 1 line
2.47.2
Re: [PATCH 4/9] pidfs: use anon_inode_setattr()
Quoted 5 lines · Christian Brauner
So far pidfs did use it's own version. Just use the generic version. We use our own wrappers because we're going to be implementing properties soon.
Signed-off-by Christian Brauner <[email protected]>
OK. Feel free to add:
Honza
Quoted 21 lines
--- fs/pidfs.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)fs/pidfs.c +1
-1@@ -569,7 +569,7 @@ static struct vfsmount *pidfs_mnt __ro_after_init;static int pidfs_setattr (struct mnt_idmap *idmap,struct dentry *dentry,struct iattr *attr) { -return -EOPNOTSUPP ; +return anon_inode_setattr (idmap, dentry, attr); } static int pidfs_getattr(struct mnt_idmap *idmap, const struct path *path,Signature · 2 lines
-- 2.47.2
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Christian Brauner
maintainerPATCH 5/9 anon_inode: raise SB_I_NODEV and SB_I_NOEXEC
In reply to Christian BraunerIt isn't possible to execute anoymous inodes because they cannot be opened in any way after they have been created. This includes execution:
execveat(fd_anon_inode, "", NULL, NULL, AT_EMPTY_PATH)
Anonymous inodes have inode->f_op set to no_open_fops which sets no_open() which returns ENXIO. That means any call to do_dentry_open() which is the endpoint of the do_open_execat() will fail. There's no chance to execute an anonymous inode. Unless a given subsystem overrides it ofc.
Howerver, we should still harden this and raise SB_I_NODEV and SB_I_NOEXEC on the superblock itself so that no one gets any creative ideas.
--- fs/anon_inodes.c | 2 ++ 1 file changed, 2 insertions(+)
fs/anon_inodes.c +2 -0
@@ -86,6 +86,8 @@ static int anon_inodefs_init_fs_context(struct fs_context *fc)struct pseudo_fs_context *ctx =init_pseudo (fc,ANON_INODE_FS_MAGIC );if (!ctx)return -ENOMEM ; + fc->s_iflags |=SB_I_NOEXEC ; + fc->s_iflags |=SB_I_NODEV ; ctx->dops = &anon_inodefs_dentry_operations;return 0 ; }
Signature · 1 line
2.47.2
Re: [PATCH 5/9] anon_inode: raise SB_I_NODEV and SB_I_NOEXEC
Quoted 1 line · Christian Brauner
It isn't possible to execute anoymous inodes because they cannot be
^^ anonymous
Quoted 11 lines
opened in any way after they have been created. This includes execution:
execveat(fd_anon_inode, "", NULL, NULL, AT_EMPTY_PATH)Anonymous inodes have
inode->f_opset tono_open_fopswhich setsno_open()which returns ENXIO. That means any call todo_dentry_open()which is the endpoint of thedo_open_execat()will fail. There's no chance to execute an anonymous inode. Unless a given subsystem overrides it ofc.Howerver, we should still harden this and raise
SB_I_NODEVand
^^^ However
Quoted 2 lines
SB_I_NOEXECon the superblock itself so that no one gets any creative ideas.
;)
Feel free to add:
Honza
Quoted 24 lines
Cc <[email protected]> # all LTS kernelsSigned-off-by Christian Brauner <[email protected]>--- fs/anon_inodes.c | 2 ++ 1 file changed, 2 insertions(+)fs/anon_inodes.c +2
-0@@ -86,6 +86,8 @@ static int anon_inodefs_init_fs_context(struct fs_context *fc)struct pseudo_fs_context *ctx =init_pseudo (fc,ANON_INODE_FS_MAGIC );if (!ctx)return -ENOMEM ; + fc->s_iflags |=SB_I_NOEXEC ; + fc->s_iflags |=SB_I_NODEV ; ctx->dops = &anon_inodefs_dentry_operations;return 0 ; }Signature · 2 lines
-- 2.47.2
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Christian Brauner
maintainerRe: [PATCH 5/9] anon_inode: raise SB_I_NODEV and SB_I_NOEXEC
Quoted 14 lines · Jan Kara
Quoted 1 line · Christian Brauner
It isn't possible to execute anoymous inodes because they cannot be
^^ anonymous
Quoted 10 lines
opened in any way after they have been created. This includes execution:
execveat(fd_anon_inode, "", NULL, NULL, AT_EMPTY_PATH)Anonymous inodes have
inode->f_opset tono_open_fopswhich setsno_open()which returns ENXIO. That means any call todo_dentry_open()which is the endpoint of thedo_open_execat()will fail. There's no chance to execute an anonymous inode. Unless a given subsystem overrides it ofc.
Quoted 7 lines
Quoted 1 line
Howerver, we should still harden this and raise
SB_I_NODEVand^^^ However
Quoted 2 lines
SB_I_NOEXECon the superblock itself so that no one gets any creative ideas.;)
I've told our new AI overloards to sprinkle-in some typos so no one realizes I've been mostly replaced by a bot.
Or I'm just generally tired so I fat-finger a lot more than usual. :D
Quoted 34 lines
Feel free to add:
Reviewed-by Jan Kara <[email protected]>Honza
Quoted 24 lines
Cc <[email protected]> # all LTS kernelsSigned-off-by Christian Brauner <[email protected]>--- fs/anon_inodes.c | 2 ++ 1 file changed, 2 insertions(+)fs/anon_inodes.c +2
-0@@ -86,6 +86,8 @@ static int anon_inodefs_init_fs_context(struct fs_context *fc)struct pseudo_fs_context *ctx =init_pseudo (fc,ANON_INODE_FS_MAGIC );if (!ctx)return -ENOMEM ; + fc->s_iflags |=SB_I_NOEXEC ; + fc->s_iflags |=SB_I_NODEV ; ctx->dops = &anon_inodefs_dentry_operations;return 0 ; }Signature · 2 lines
-- 2.47.2Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Christian Brauner
maintainerPATCH 6/9 selftests/filesystems: add first test for anonymous inodes
In reply to Christian BraunerTest that anonymous inodes cannot be chown()ed.
--- tools/testing/selftests/filesystems/.gitignore | 1 + tools/testing/selftests/filesystems/Makefile | 2 +- .../selftests/filesystems/anon_inode_test.c | 26 ++++++++++++++++++++++ 3 files changed, 28 insertions(+), 1 deletion(-)
tools/testing/selftests/filesystems/.gitignore +1 -0
@@ -2,3 +2,4 @@ dnotify_test devpts_pts file_stressor +anon_inode_test
tools/testing/selftests/filesystems/Makefile +1 -1
@@ -1,7 +1,7 @@ # SPDX-License-Identifier: GPL-2.0 CFLAGS += $(KHDR_INCLUDES) -TEST_GEN_PROGS := devpts_pts file_stressor +TEST_GEN_PROGS := devpts_pts file_stressor anon_inode_test TEST_GEN_PROGS_EXTENDED := dnotify_test include ../lib.mk
tools/testing/selftests/filesystems/anon_inode_test.c new file +26 -0
@@ -0,0 +1,26 @@ +// SPDX-License-Identifier: GPL-2.0 +#define _GNU_SOURCE +#define __SANE_USERSPACE_TYPES__ + +#include <fcntl.h> +#include <stdio.h> +#include <sys/stat.h> + +#include "../kselftest_harness.h" +#include "overlayfs/wrappers.h" + +TEST (anon_inode_no_chown) +{ +int fd_context; + + fd_context =sys_fsopen ("tmpfs" ,0 ); +ASSERT_GE (fd_context,0 ); + +ASSERT_LT (fchown (fd_context,1234 ,5678 ),0 ); +ASSERT_EQ (errno,EOPNOTSUPP ); + +EXPECT_EQ (close (fd_context),0 ); +} + +TEST_HARNESS_MAIN +
Signature · 1 line
2.47.2
Re: [PATCH 6/9] selftests/filesystems: add first test for anonymous inodes
Quoted 3 lines · Christian Brauner
Test that anonymous inodes cannot be chown()ed.
Signed-off-by Christian Brauner <[email protected]>
Looks good. Feel free to add:
Honza
Quoted 64 lines
--- tools/testing/selftests/filesystems/.gitignore | 1 + tools/testing/selftests/filesystems/Makefile | 2 +- .../selftests/filesystems/anon_inode_test.c | 26 ++++++++++++++++++++++ 3 files changed, 28 insertions(+), 1 deletion(-)tools/testing/selftests/filesystems/.gitignore +1
-0@@ -2,3 +2,4 @@ dnotify_test devpts_pts file_stressor +anon_inode_testtools/testing/selftests/filesystems/Makefile +1
-1@@ -1,7 +1,7 @@ # SPDX-License-Identifier: GPL-2.0 CFLAGS += $(KHDR_INCLUDES) -TEST_GEN_PROGS := devpts_pts file_stressor +TEST_GEN_PROGS := devpts_pts file_stressor anon_inode_test TEST_GEN_PROGS_EXTENDED := dnotify_test include ../lib.mktools/testing/selftests/filesystems/anon_inode_test.c new file +26
-0@@ -0,0 +1,26 @@ +// SPDX-License-Identifier: GPL-2.0 +#define _GNU_SOURCE +#define __SANE_USERSPACE_TYPES__ + +#include <fcntl.h> +#include <stdio.h> +#include <sys/stat.h> + +#include "../kselftest_harness.h" +#include "overlayfs/wrappers.h" + +TEST (anon_inode_no_chown) +{ +int fd_context; + + fd_context =sys_fsopen ("tmpfs" ,0 ); +ASSERT_GE (fd_context,0 ); + +ASSERT_LT (fchown (fd_context,1234 ,5678 ),0 ); +ASSERT_EQ (errno,EOPNOTSUPP ); + +EXPECT_EQ (close (fd_context),0 ); +} + +TEST_HARNESS_MAIN +Signature · 2 lines
-- 2.47.2
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Christian Brauner
maintainerPATCH 7/9 selftests/filesystems: add second test for anonymous inodes
In reply to Christian BraunerTest that anonymous inodes cannot be chmod()ed.
--- tools/testing/selftests/filesystems/anon_inode_test.c | 13 +++++++++++++ 1 file changed, 13 insertions(+)
tools/testing/selftests/filesystems/anon_inode_test.c +13 -0
@@ -22,5 +22,18 @@ TEST(anon_inode_no_chown)EXPECT_EQ (close (fd_context),0 ); } +TEST (anon_inode_no_chmod) +{ +int fd_context; + + fd_context =sys_fsopen ("tmpfs" ,0 ); +ASSERT_GE (fd_context,0 ); + +ASSERT_LT (fchmod (fd_context,0777 ),0 ); +ASSERT_EQ (errno,EOPNOTSUPP ); + +EXPECT_EQ (close (fd_context),0 ); +} +TEST_HARNESS_MAIN
Signature · 1 line
2.47.2
Re: [PATCH 7/9] selftests/filesystems: add second test for anonymous inodes
Quoted 3 lines · Christian Brauner
Test that anonymous inodes cannot be chmod()ed.
Signed-off-by Christian Brauner <[email protected]>
Looks good. Feel free to add:
Honza
Quoted 31 lines
--- tools/testing/selftests/filesystems/anon_inode_test.c | 13 +++++++++++++ 1 file changed, 13 insertions(+)tools/testing/selftests/filesystems/anon_inode_test.c +13
-0@@ -22,5 +22,18 @@ TEST(anon_inode_no_chown)EXPECT_EQ (close (fd_context),0 ); } +TEST (anon_inode_no_chmod) +{ +int fd_context; + + fd_context =sys_fsopen ("tmpfs" ,0 ); +ASSERT_GE (fd_context,0 ); + +ASSERT_LT (fchmod (fd_context,0777 ),0 ); +ASSERT_EQ (errno,EOPNOTSUPP ); + +EXPECT_EQ (close (fd_context),0 ); +} +TEST_HARNESS_MAIN Signature · 2 lines
-- 2.47.2
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Christian Brauner
maintainerPATCH 8/9 selftests/filesystems: add third test for anonymous inodes
In reply to Christian BraunerTest that anonymous inodes cannot be exec()ed.
--- tools/testing/selftests/filesystems/anon_inode_test.c | 13 +++++++++++++ 1 file changed, 13 insertions(+)
tools/testing/selftests/filesystems/anon_inode_test.c +13 -0
@@ -35,5 +35,18 @@ TEST(anon_inode_no_chmod)EXPECT_EQ (close (fd_context),0 ); } +TEST (anon_inode_no_exec) +{ +int fd_context; + + fd_context =sys_fsopen ("tmpfs" ,0 ); +ASSERT_GE (fd_context,0 ); + +ASSERT_LT (execveat (fd_context,"" ,NULL ,NULL ,AT_EMPTY_PATH ),0 ); +ASSERT_EQ (errno,EACCES ); + +EXPECT_EQ (close (fd_context),0 ); +} +TEST_HARNESS_MAIN
Signature · 1 line
2.47.2
Re: [PATCH 8/9] selftests/filesystems: add third test for anonymous inodes
Quoted 3 lines · Christian Brauner
Test that anonymous inodes cannot be exec()ed.
Signed-off-by Christian Brauner <[email protected]>
Looks good. Feel free to add:
Honza
Quoted 31 lines
--- tools/testing/selftests/filesystems/anon_inode_test.c | 13 +++++++++++++ 1 file changed, 13 insertions(+)tools/testing/selftests/filesystems/anon_inode_test.c +13
-0@@ -35,5 +35,18 @@ TEST(anon_inode_no_chmod)EXPECT_EQ (close (fd_context),0 ); } +TEST (anon_inode_no_exec) +{ +int fd_context; + + fd_context =sys_fsopen ("tmpfs" ,0 ); +ASSERT_GE (fd_context,0 ); + +ASSERT_LT (execveat (fd_context,"" ,NULL ,NULL ,AT_EMPTY_PATH ),0 ); +ASSERT_EQ (errno,EACCES ); + +EXPECT_EQ (close (fd_context),0 ); +} +TEST_HARNESS_MAIN Signature · 2 lines
-- 2.47.2
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Christian Brauner
maintainerPATCH 9/9 selftests/filesystems: add fourth test for anonymous inodes
In reply to Christian BraunerTest that anonymous inodes cannot be open()ed.
--- tools/testing/selftests/filesystems/anon_inode_test.c | 17 +++++++++++++++++ 1 file changed, 17 insertions(+)
tools/testing/selftests/filesystems/anon_inode_test.c +17 -0
@@ -48,5 +48,22 @@ TEST(anon_inode_no_exec)EXPECT_EQ (close (fd_context),0 ); } +TEST (anon_inode_no_open) +{ +int fd_context; + + fd_context =sys_fsopen ("tmpfs" ,0 ); +ASSERT_GE (fd_context,0 ); + +ASSERT_GE (dup2 (fd_context,500 ),0 ); +ASSERT_EQ (close (fd_context),0 ); + fd_context =500 ; + +ASSERT_LT (open ("/proc/self/fd/500" ,0 ),0 ); +ASSERT_EQ (errno,ENXIO ); + +EXPECT_EQ (close (fd_context),0 ); +} +TEST_HARNESS_MAIN
Signature · 1 line
2.47.2
Re: [PATCH 9/9] selftests/filesystems: add fourth test for anonymous inodes
Quoted 3 lines · Christian Brauner
Test that anonymous inodes cannot be open()ed.
Signed-off-by Christian Brauner <[email protected]>
Looks good. Feel free to add:
Honza
Quoted 35 lines
--- tools/testing/selftests/filesystems/anon_inode_test.c | 17 +++++++++++++++++ 1 file changed, 17 insertions(+)tools/testing/selftests/filesystems/anon_inode_test.c +17
-0@@ -48,5 +48,22 @@ TEST(anon_inode_no_exec)EXPECT_EQ (close (fd_context),0 ); } +TEST (anon_inode_no_open) +{ +int fd_context; + + fd_context =sys_fsopen ("tmpfs" ,0 ); +ASSERT_GE (fd_context,0 ); + +ASSERT_GE (dup2 (fd_context,500 ),0 ); +ASSERT_EQ (close (fd_context),0 ); + fd_context =500 ; + +ASSERT_LT (open ("/proc/self/fd/500" ,0 ),0 ); +ASSERT_EQ (errno,ENXIO ); + +EXPECT_EQ (close (fd_context),0 ); +} +TEST_HARNESS_MAIN Signature · 2 lines
-- 2.47.2
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Quoted 22 lines · Christian Brauner
* Anonymous inodes currently don't come with a proper mode causing issues in the kernel when we want to add useful VFS debug assert. Fix that by giving them a proper mode and masking it off when we report it to userspace which relies on them not having any mode.
* Anonymous inodes currently allow to change inode attributes because the VFS falls back to
simple_setattr()ifi_op->setattrisn't implemented. This means the ownership and mode for every single user ofanon_inode_inodecan be changed. Block that as it's either useless or actively harmful. If specific ownership is needed the respective subsystem should allocate anonymous inodes from their own private superblock.* Port pidfs to the new anon_inode_{g,s}etattr() helpers.
* Add proper tests for anonymous inode behavior.
The anonymous inode specific fixes should ideally be backported to all LTS kernels.
Signed-off-by Christian Brauner <[email protected]>
Quoted 12 lines
---
Christian Brauner (9):
anon_inode: use a proper mode internally
pidfs: useanon_inode_getattr()
anon_inode: explicitly block ->setattr()
pidfs: useanon_inode_setattr()
anon_inode: raiseSB_I_NODEVandSB_I_NOEXECselftests/filesystems: add first test for anonymous inodes selftests/filesystems: add second test for anonymous inodes selftests/filesystems: add third test for anonymous inodes selftests/filesystems: add fourth test for anonymous inodes
I have two nits, past that LGTM
1. I would add a comment explaining why S_IFREG in alloc_anon_inode()
2. commit messages for selftests could spell out what's being added instead of being counted, it's all one-liners
for example:
selftests/filesystems: validate that anonymous inodes cannot be chown()ed
Signature · 1 line
Mateusz Guzik <mjguzik gmail.com>
Christian Brauner
maintainerQuoted 36 lines · Mateusz Guzik
Quoted 22 lines · Christian Brauner
* Anonymous inodes currently don't come with a proper mode causing issues in the kernel when we want to add useful VFS debug assert. Fix that by giving them a proper mode and masking it off when we report it to userspace which relies on them not having any mode.
* Anonymous inodes currently allow to change inode attributes because the VFS falls back to
simple_setattr()ifi_op->setattrisn't implemented. This means the ownership and mode for every single user ofanon_inode_inodecan be changed. Block that as it's either useless or actively harmful. If specific ownership is needed the respective subsystem should allocate anonymous inodes from their own private superblock.* Port pidfs to the new anon_inode_{g,s}etattr() helpers.
* Add proper tests for anonymous inode behavior.
The anonymous inode specific fixes should ideally be backported to all LTS kernels.
Signed-off-by Christian Brauner <[email protected]>Quoted 12 lines
---
Christian Brauner (9):
anon_inode: use a proper mode internally
pidfs: useanon_inode_getattr()
anon_inode: explicitly block ->setattr()
pidfs: useanon_inode_setattr()
anon_inode: raiseSB_I_NODEVandSB_I_NOEXECselftests/filesystems: add first test for anonymous inodes selftests/filesystems: add second test for anonymous inodes selftests/filesystems: add third test for anonymous inodes selftests/filesystems: add fourth test for anonymous inodes
Quoted 3 lines
I have two nits, past that LGTM
1. I would add a comment explaining why
S_IFREGinalloc_anon_inode()
/*
* Historically anonymous inodes didn't have a type at all and
* userspace has come to rely on this. Internally they're just
* regular files but S_IFREG is masked off when reporting
* information to userspace.
*/
Quoted 2 lines
2. commit messages for selftests could spell out what's being added instead of being counted, it's all one-liners
I've replaced the count with chown(), chmod(), exec(), and open().
Thanks!
Quoted 41 lines · Christian Brauner
* Anonymous inodes currently don't come with a proper mode causing issues in the kernel when we want to add useful VFS debug assert. Fix that by giving them a proper mode and masking it off when we report it to userspace which relies on them not having any mode.
* Anonymous inodes currently allow to change inode attributes because the VFS falls back to
simple_setattr()ifi_op->setattrisn't implemented. This means the ownership and mode for every single user ofanon_inode_inodecan be changed. Block that as it's either useless or actively harmful. If specific ownership is needed the respective subsystem should allocate anonymous inodes from their own private superblock.* Port pidfs to the new anon_inode_{g,s}etattr() helpers.
* Add proper tests for anonymous inode behavior.
The anonymous inode specific fixes should ideally be backported to all LTS kernels.
Signed-off-by Christian Brauner <[email protected]>---
Christian Brauner (9):
anon_inode: use a proper mode internally
pidfs: useanon_inode_getattr()
anon_inode: explicitly block ->setattr()
pidfs: useanon_inode_setattr()
anon_inode: raiseSB_I_NODEVandSB_I_NOEXECselftests/filesystems: add first test for anonymous inodes selftests/filesystems: add second test for anonymous inodes selftests/filesystems: add third test for anonymous inodes selftests/filesystems: add fourth test for anonymous inodesfs/anon_inodes.c | 45 ++++++++++++++ fs/internal.h | 5 ++ fs/libfs.c | 2 +- fs/pidfs.c | 26 +------- tools/testing/selftests/filesystems/.gitignore | 1 + tools/testing/selftests/filesystems/Makefile | 2 +- .../selftests/filesystems/anon_inode_test.c | 69 ++++++++++++++++++++++ 7 files changed, 124 insertions(+), 26 deletions(-)
Quoted 4 lines
---
base-commit:0af2f6be1b4281385b618cb86ad946eded089ac8
change-id: 20250407-work-anon_inode-e22bb1a74992
This all looks like good changes to make. I'm still a little curious about what might be dependent on not seeing a file type in st_mode, but if we haven't traditionally reported one, it's probably safest to continue without doing so.