xfs: tear down zoned sysfs before freeing zone info
PATCH v2 1/1 xfs: tear down zoned sysfs before freeing zone info
The zoned sysfs directory is currently registered as part of the generic per-mount sysfs setup, but the data it exposes has a narrower lifetime. mp->m_zone_info is allocated later by xfs_mount_zones() and freed by xfs_unmount_zones(), while the zoned sysfs kobject remains registered until xfs_mount_sysfs_del() runs near the end of xfs_unmountfs().
A read of nr_open_zones can therefore enter through the still-live sysfs kobject after xfs_unmount_zones() has freed mp->m_zone_info, leading to a use-after-free in nr_open_zones_show().
Make the zoned sysfs lifetime match the zone allocator lifetime. Create the zoned sysfs directory only after xfs_mount_zones() succeeds, and remove it before xfs_unmount_zones() frees m_zone_info during unmount and mount failure cleanup.
Validation reproduced this kernel report:
BUG: KASAN: slab-use-after-free in nr_open_zones_show+0x86/0x90 The buggy address belongs to the object at ffff88810b177800 which belongs to the cache kmalloc-1k of size 1024 The buggy address is located 160 bytes inside of freed 1024-byte region [ffff88810b177800, ffff88810b177c00) Read of size 4 Call trace: print_report+0xcd/0x620 nr_open_zones_show+0x86/0x90 (fs/xfs/xfs_sysfs.c:724) srso_alias_return_thunk+0x5/0xfbef5 __virt_addr_valid+0x20c/0x410 kasan_report+0xdd/0x110 sysfs_kf_seq_show+0x1bd/0x380 seq_read_iter+0x40f/0x11b0 lock_release+0xba/0x260 mark_held_locks+0x40/0x70 vfs_read+0x717/0xce0 __up_read+0x319/0x900 ksys_read+0xf8/0x1c0 do_user_addr_fault+0x3d0/0xbc0 trace_hardirqs_on_prepare+0x23/0xf0 do_syscall_64+0xc8/0x530 (arch/x86/entry/syscall_64.c:87) entry_SYSCALL_64_after_hwframe+0x74/0x7c Allocated by task stack: kasan_save_stack+0x33/0x60 kasan_save_track+0x14/0x30 __kasan_kmalloc+0xaa/0xb0 __kmalloc_cache_noprof+0x205/0x460 xfs_mount_zones+0x34c/0x2650 xfs_mountfs+0x1b97/0x1eb0 xfs_fs_fill_super+0xf2b/0x18a0 get_tree_bdev_flags+0x310/0x590 vfs_get_tree+0x8d/0x2e0 __x64_sys_fsconfig+0x61c/0xbc0 do_syscall_64+0xc8/0x530 (arch/x86/entry/syscall_64.c:87) entry_SYSCALL_64_after_hwframe+0x74/0x7c Freed by task stack: kasan_save_stack+0x33/0x60 kasan_save_track+0x14/0x30 kasan_save_free_info+0x3b/0x60 __kasan_slab_free+0x5f/0x80 kfree+0x20e/0x4c0 xfs_unmountfs+0x2fd/0x390 xfs_fs_put_super+0x60/0x110 generic_shutdown_super+0x143/0x4b0 kill_block_super+0x3b/0x90 xfs_kill_sb+0x12/0x50 deactivate_locked_super+0xa7/0x160 cleanup_mnt+0x218/0x420 task_work_run+0x11a/0x1f0 exit_to_user_mode_loop+0x13c/0x4f0 do_syscall_64+0x4a9/0x530 (arch/x86/entry/syscall_64.c:87) entry_SYSCALL_64_after_hwframe+0x74/0x7c
---
v2:
Tear down the zoned sysfs directory before freeing m_zone_info instead of serializing nr_open_zones_show() with s_umount.
fs/xfs/xfs_mount.c | 10 +++++++++- fs/xfs/xfs_sysfs.c | 28 +++++++++++++++++----------- fs/xfs/xfs_sysfs.h | 2 ++ 3 files changed, 28 insertions(+), 12 deletions(-)
fs/xfs/xfs_mount.c +9 -1
@@ -1178,6 +1178,9 @@ xfs_mountfs( error =xfs_mount_zones (mp);if (error) goto out_rtunmount; + error =xfs_zoned_sysfs_init (mp); +if (error) + goto out_unmount_zones; }/*
@@ -1233,6 +1236,9 @@ xfs_mountfs( out_agresv:xfs_fs_unreserve_ag_blocks (mp);xfs_qm_unmount_quotas (mp); +if (xfs_has_zoned (mp)) +xfs_zoned_sysfs_del (mp); + out_unmount_zones:if (xfs_has_zoned (mp))xfs_unmount_zones (mp); out_rtunmount:
@@ -1323,8 +1329,10 @@ xfs_unmountfs(xfs_zone_gc_stop (mp);xfs_fs_unreserve_ag_blocks (mp);xfs_qm_unmount_quotas (mp); -if (xfs_has_zoned (mp)) +if (xfs_has_zoned (mp)) { +xfs_zoned_sysfs_del (mp);xfs_unmount_zones (mp); + }xfs_rtunmount_inodes (mp);xfs_irele (mp->m_rootip);if (mp->m_metadirip)
fs/xfs/xfs_sysfs.c +17 -11
@@ -780,6 +780,23 @@ static const struct kobj_type xfs_zoned_ktype = { .default_groups = xfs_zoned_groups, }; +int +xfs_zoned_sysfs_init (struct xfs_mount *mp) +{ +if (!IS_ENABLED (CONFIG_XFS_RT ) || !xfs_has_zoned (mp)) +return 0 ; + +return xfs_sysfs_init (&mp->m_zoned_kobj, &xfs_zoned_ktype, + &mp->m_kobj,"zoned" ); +} + +void +xfs_zoned_sysfs_del (struct xfs_mount *mp) +{ +if (IS_ENABLED (CONFIG_XFS_RT ) &&xfs_has_zoned (mp)) +xfs_sysfs_del (&mp->m_zoned_kobj); +} +int xfs_mount_sysfs_init (struct xfs_mount *mp)
@@ -820,14 +837,6 @@ xfs_mount_sysfs_init(if (error) goto out_remove_error_dir; -if (IS_ENABLED (CONFIG_XFS_RT ) &&xfs_has_zoned (mp)) { -/* .../xfs/<dev>/zoned/ */ - error =xfs_sysfs_init (&mp->m_zoned_kobj, &xfs_zoned_ktype, - &mp->m_kobj,"zoned" ); -if (error) - goto out_remove_error_dir; - } -return 0 ; out_remove_error_dir:
@@ -846,9 +855,6 @@ xfs_mount_sysfs_del(struct xfs_error_cfg *cfg;int i, j; -if (IS_ENABLED (CONFIG_XFS_RT ) &&xfs_has_zoned (mp)) -xfs_sysfs_del (&mp->m_zoned_kobj); -for (i =0 ; i <XFS_ERR_CLASS_MAX ; i++) {for (j =0 ; j <XFS_ERR_ERRNO_MAX ; j++) { cfg = &mp->m_error_cfg[i][j];
fs/xfs/xfs_sysfs.h +2 -0
@@ -53,6 +53,8 @@ xfs_sysfs_del( }int xfs_mount_sysfs_init (struct xfs_mount *mp); +int xfs_zoned_sysfs_init (struct xfs_mount *mp); +void xfs_zoned_sysfs_del (struct xfs_mount *mp);void xfs_mount_sysfs_del (struct xfs_mount *mp);#endif /* __XFS_SYSFS_H__ */
Signature · 1 line
2.43.0
Quoted 14 lines · Cen Zhang
The zoned sysfs directory is currently registered as part of the generic per-mount sysfs setup, but the data it exposes has a narrower lifetime.
mp->m_zone_infois allocated later byxfs_mount_zones()and freed byxfs_unmount_zones(), while the zoned sysfs kobject remains registered untilxfs_mount_sysfs_del()runs near the end ofxfs_unmountfs().A read of
nr_open_zonescan therefore enter through the still-live sysfs kobject afterxfs_unmount_zones()has freedmp->m_zone_info, leading to a use-after-free innr_open_zones_show().Make the zoned sysfs lifetime match the zone allocator lifetime. Create the zoned sysfs directory only after
xfs_mount_zones()succeeds, and remove it beforexfs_unmount_zones()freesm_zone_infoduring unmount and mount failure cleanup.
[...]
Quoted 2 lines
Tear down the zoned sysfs directory before freeing
m_zone_infoinstead of serializingnr_open_zones_show()withs_umount.
Hmmm... What if the user already has entered the show function but gets preempted before referencing the zone info and umount proceeds ? That problem remains, no ?
Quoted 11 lines
fs/xfs/xfs_mount.c | 10 +++++++++- fs/xfs/xfs_sysfs.c | 28 +++++++++++++++++----------- fs/xfs/xfs_sysfs.h | 2 ++ 3 files changed, 28 insertions(+), 12 deletions(-)fs/xfs/xfs_mount.c +0
-0
fs/xfs/xfs_mount.c xfs_mountfs( · line 1183
Quoted 6 lines
fs/xfs/xfs_mount.c +3
-0@@ -1178,6 +1178,9 @@ xfs_mountfs( error =xfs_mount_zones (mp);if (error) goto out_rtunmount; + error =xfs_zoned_sysfs_init (mp); +if (error) + goto out_unmount_zones;
Why not call this in xfs_mount_zones() ?
fs/xfs/xfs_mount.c xfs_mountfs( · line 1240
Quoted 9 lines
fs/xfs/xfs_mount.c +2
-0}/* @@ -1233,6 +1236,9 @@ xfs_mountfs( out_agresv:xfs_fs_unreserve_ag_blocks (mp);xfs_qm_unmount_quotas (mp); +if (xfs_has_zoned (mp)) +xfs_zoned_sysfs_del (mp);
And this as the first thing in xfs_unmount_zones() ?
Another simpler approach could be to simply call xfs_mount_sysfs_del() earlier in xfs_unmountfs(). Though that may have other implications and I have not checked details.
Signature · 2 lines
Damien Le Moal Western Digital Research
Hi Damien,
Thanks for your review and suggestions.
Quoted 3 lines
Hmmm... What if the user already has entered the show function but gets preempted before referencing the zone info and umount proceeds ? That problem remains, no ?
I double checked the sysfs/kernfs paths, and I do not think that specific case remains with the current ordering. sysfs reads go through the kernfs seq_file path: kernfs_seq_start() takes an active reference before sysfs_kf_seq_show() calls the attribute's ->show method, and kernfs_seq_stop() drops it afterwards. On removal, kobject_del() reaches sysfs_remove_dir(), which calls kernfs_remove(); kernfs_remove() deactivates the subtree and kernfs_drain() waits for active references to drain.
So if nr_open_zones_show() has already been entered and then gets preempted, xfs_zoned_sysfs_del() should wait for that active callback to finish before returning. Only after that would xfs_unmount_zones() free m_zone_info.
Quoted 1 line
Why not call this in
xfs_mount_zones()?
Yes, that is cleaner. The zoned sysfs files expose zone allocator state, so putting xfs_zoned_sysfs_init() in xfs_mount_zones() makes the ownership and lifetime boundary more obvious. I will move it there for v3.
Quoted 1 line
And this as the first thing in
xfs_unmount_zones()?
Agreed. I will make xfs_unmount_zones() remove the zoned sysfs directory before zone GC teardown and before xfs_free_zone_info(). That keeps all zone-info users behind the same allocator teardown entry point.
One detail is that if xfs_zoned_sysfs_init() is called after xfs_zone_gc_mount() and then fails, the mount error path in xfs_mount_zones() must unwind the GC thread before freeing m_zone_info. I will handle that in the v3 error path.
Quoted 3 lines
Another simpler approach could be to simply call
xfs_mount_sysfs_del()earlier inxfs_unmountfs(). Though that may have other implications and I have not checked details.
I would rather avoid moving the whole xfs_mount_sysfs_del() earlier. The bug is specific to the zoned sysfs subtree because those files expose m_zone_info, whereas the rest of the per-mount sysfs tree has different state and lifetime. Moving only the zoned subtree into xfs_mount_zones()/xfs_unmount_zones() seems narrower and keeps unrelated sysfs teardown ordering unchanged.
Best Regards,
Cen Zhang