vfs, overlayfs: Fix syncfs() to return error
Subsystems (2)
cover vfs, overlayfs: Fix syncfs() to return error
Hi,
This is V2 of patches which tries to fix syncfs() for overlayfs to return error code when sync_filesystem(upper_sb) returns error or there are writeback errors on upper_sb.
I posted V1 of patch here.
https://lore.kernel.org/linux-fsdevel/[email protected]/
This is just compile tested patch series. Trying to get early feedback to figure out what direction to move in to fix this issue.
Thanks
Vivek
Vivek Goyal (3):
vfs: add new f_op->syncfs vector
overlayfs: Implement f_op->syncfs() call
overlayfs: Check writeback errors w.r.t upper in ->syncfs()
fs/overlayfs/file.c | 1 + fs/overlayfs/overlayfs.h | 3 +++ fs/overlayfs/ovl_entry.h | 2 ++ fs/overlayfs/readdir.c | 1 + fs/overlayfs/super.c | 41 +++++++++++++++++++++++++++++++++++++++- fs/sync.c | 29 +++++++++++++++++++--------- include/linux/fs.h | 1 + 7 files changed, 68 insertions(+), 10 deletions(-)
Signature · 1 line
2.25.4
PATCH 1/3 vfs: add new f_op->syncfs vector
Current implementation of __sync_filesystem() ignores the return code from ->sync_fs(). I am not sure why that's the case. There must have been some historical reason for this.
Ignoring ->sync_fs() return code is problematic for overlayfs where it can return error if sync_filesystem() on upper super block failed. That error will simply be lost and sycnfs(overlay_fd), will get success (despite the fact it failed).
If we modify existing implementation, there is a concern that it will lead to user space visible behavior changes and break things. So instead implement a new file_operations->syncfs() call which will be called in syncfs() syscall path. Return code from this new call will be captured. And all the writeback error detection logic can go in there as well. Only filesystems which implement this call get affected by this change. Others continue to fallback to existing mechanism.
To be clear, I mean something like this (draft, untested) patch. You'd also need to add a new ->syncfs op for overlayfs, and that could just do a check_and_advance against the upper layer sb's errseq_t after calling sync_filesystem.
Vivek, fixed couple of minor compile errors in original patch.
--- fs/sync.c | 29 ++++++++++++++++++++--------- include/linux/fs.h | 1 + 2 files changed, 21 insertions(+), 9 deletions(-)
fs/sync.c +20 -9
@@ -155,27 +155,38 @@ void emergency_sync(void) } } +static int generic_syncfs (struct file *file) +{ +int ret, ret2; +struct super_block *sb = file->f_path.dentry->d_sb; + +down_read (&sb->s_umount); + ret =sync_filesystem (sb); +up_read (&sb->s_umount); + + ret2 =errseq_check_and_advance (&sb->s_wb_err, &file->f_sb_err); + +return ret ? ret : ret2; +} +/* * sync a single super */ SYSCALL_DEFINE1(syncfs, int, fd) {struct fd f =fdget (fd); -struct super_block *sb; -int ret, ret2; +int ret;if (!f.file)return -EBADF ; - sb = f.file->f_path.dentry->d_sb; - -down_read (&sb->s_umount); - ret =sync_filesystem (sb); -up_read (&sb->s_umount); - ret2 =errseq_check_and_advance (&sb->s_wb_err, &f.file->f_sb_err); +if (f.file->f_op->syncfs) + ret = f.file->f_op->syncfs (f.file); +else + ret =generic_syncfs (f.file);fdput (f); -return ret ? ret : ret2; +return ret; }/**
include/linux/fs.h +1 -0
@@ -1859,6 +1859,7 @@ struct file_operations { struct file *file_out, loff_t pos_out, loff_t len, unsigned int remap_flags);int (*fadvise)(struct file *,loff_t ,loff_t ,int ); +int (*syncfs)(struct file *); } __randomize_layout;struct inode_operations {
Signature · 1 line
2.25.4
Al Viro
maintainerRe: [PATCH 1/3] vfs: add new f_op->syncfs vector
[Christoph added to Cc...]
Quoted 9 lines · Vivek Goyal
Current implementation of
__sync_filesystem()ignores the return code from ->sync_fs(). I am not sure why that's the case. There must have been some historical reason for this.Ignoring ->sync_fs() return code is problematic for overlayfs where it can return error if
sync_filesystem()on upper super block failed. That error will simply be lost andsycnfs(overlay_fd), will get success (despite the fact it failed).
Quoted 8 lines
If we modify existing implementation, there is a concern that it will lead to user space visible behavior changes and break things. So instead implement a new
file_operations->syncfs()call which will be called insyncfs()syscall path. Return code from this new call will be captured. And all the writeback error detection logic can go in there as well. Only filesystems which implement this call get affected by this change. Others continue to fallback to existing mechanism.
That smells like a massive source of confusion down the road. I'd just looked through the existing instances; many always return 0, but quite a few sometimes try to return an error:
fs/btrfs/super.c:2412: .sync_fs = btrfs_sync_fs, fs/exfat/super.c:204: .sync_fs = exfat_sync_fs, fs/ext4/super.c:1674: .sync_fs = ext4_sync_fs, fs/f2fs/super.c:2480: .sync_fs = f2fs_sync_fs, fs/gfs2/super.c:1600: .sync_fs = gfs2_sync_fs, fs/hfsplus/super.c:368: .sync_fs = hfsplus_sync_fs, fs/nilfs2/super.c:689: .sync_fs = nilfs_sync_fs, fs/ocfs2/super.c:139: .sync_fs = ocfs2_sync_fs,
fs/overlayfs/super.c:399: .sync_fs = ovl_sync_fs,fs/ubifs/super.c:2052: .sync_fs = ubifs_sync_fs,
is the list of such. There are 4 method callers:dquot_quota_sync(), dquot_disable(), __sync_filesystem() and sync_fs_one_sb(). For sync_fs_one_sb() we want to ignore the return value; for __sync_filesystem() we almost certainly do *not* - it ends with return __sync_blockdev(sb->s_bdev, wait), after all. The question for that one is whether we want __sync_blockdev() called even in case of ->sync_fs() reporting a failure, and I suspect that it's safer to call it anyway and return the first error value we'd got. No idea about quota situation.
Jan Kara
reviewerRe: [PATCH 1/3] vfs: add new f_op->syncfs vector
Quoted 31 lines · Al Viro
[Christoph added to Cc...]
Quoted 9 lines · Vivek Goyal
Current implementation of
__sync_filesystem()ignores the return code from ->sync_fs(). I am not sure why that's the case. There must have been some historical reason for this.Ignoring ->sync_fs() return code is problematic for overlayfs where it can return error if
sync_filesystem()on upper super block failed. That error will simply be lost andsycnfs(overlay_fd), will get success (despite the fact it failed).Quoted 8 lines
If we modify existing implementation, there is a concern that it will lead to user space visible behavior changes and break things. So instead implement a new
file_operations->syncfs()call which will be called insyncfs()syscall path. Return code from this new call will be captured. And all the writeback error detection logic can go in there as well. Only filesystems which implement this call get affected by this change. Others continue to fallback to existing mechanism.That smells like a massive source of confusion down the road. I'd just looked through the existing instances; many always return 0, but quite a few sometimes try to return an error:
fs/btrfs/super.c:2412: .sync_fs = btrfs_sync_fs, fs/exfat/super.c:204: .sync_fs = exfat_sync_fs, fs/ext4/super.c:1674: .sync_fs = ext4_sync_fs, fs/f2fs/super.c:2480: .sync_fs = f2fs_sync_fs, fs/gfs2/super.c:1600: .sync_fs = gfs2_sync_fs, fs/hfsplus/super.c:368: .sync_fs = hfsplus_sync_fs, fs/nilfs2/super.c:689: .sync_fs = nilfs_sync_fs, fs/ocfs2/super.c:139: .sync_fs = ocfs2_sync_fs,
Quoted 12 lines
fs/overlayfs/super.c:399: .sync_fs =ovl_sync_fs,fs/ubifs/super.c:2052: .sync_fs =ubifs_sync_fs,
is the list of such. There are 4 method callers:dquot_quota_sync(),dquot_disable(),__sync_filesystem()andsync_fs_one_sb(). Forsync_fs_one_sb()we want to ignore the return value; for__sync_filesystem()we almost certainly do *not* - it ends with return__sync_blockdev(sb->s_bdev, wait), after all. The question for that one is whether we want__sync_blockdev()called even in case of ->sync_fs() reporting a failure, and I suspect that it's safer to call it anyway and return the first error value we'd got. No idea about quota situation.
WRT quota situation: All the ->sync_fs() calls there are due to cache coherency reasons (we need to get quota changes to disk, then prune quota files's page cache, and then userspace can read current quota structures from the disk). We don't want to fail dquot_disable() just because caches might be incoherent so ignoring ->sync_fs() return value there is fine. With dquot_quota_sync() it might make some sense to return the error - that's just a backend for Q_SYNC quotactl(2). OTOH I'm not sure anybody really cares - Q_SYNC is rarely used.
Honza
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Re: [PATCH 1/3] vfs: add new f_op->syncfs vector
Quoted 45 lines · Jan Kara
Quoted 31 lines · Al Viro
[Christoph added to Cc...]
Quoted 9 lines · Vivek Goyal
Current implementation of
__sync_filesystem()ignores the return code from ->sync_fs(). I am not sure why that's the case. There must have been some historical reason for this.Ignoring ->sync_fs() return code is problematic for overlayfs where it can return error if
sync_filesystem()on upper super block failed. That error will simply be lost andsycnfs(overlay_fd), will get success (despite the fact it failed).Quoted 8 lines
If we modify existing implementation, there is a concern that it will lead to user space visible behavior changes and break things. So instead implement a new
file_operations->syncfs()call which will be called insyncfs()syscall path. Return code from this new call will be captured. And all the writeback error detection logic can go in there as well. Only filesystems which implement this call get affected by this change. Others continue to fallback to existing mechanism.That smells like a massive source of confusion down the road. I'd just looked through the existing instances; many always return 0, but quite a few sometimes try to return an error:
fs/btrfs/super.c:2412: .sync_fs = btrfs_sync_fs, fs/exfat/super.c:204: .sync_fs = exfat_sync_fs, fs/ext4/super.c:1674: .sync_fs = ext4_sync_fs, fs/f2fs/super.c:2480: .sync_fs = f2fs_sync_fs, fs/gfs2/super.c:1600: .sync_fs = gfs2_sync_fs, fs/hfsplus/super.c:368: .sync_fs = hfsplus_sync_fs, fs/nilfs2/super.c:689: .sync_fs = nilfs_sync_fs, fs/ocfs2/super.c:139: .sync_fs = ocfs2_sync_fs,Quoted 12 lines
fs/overlayfs/super.c:399: .sync_fs =ovl_sync_fs,fs/ubifs/super.c:2052: .sync_fs =ubifs_sync_fs,
is the list of such. There are 4 method callers:dquot_quota_sync(),dquot_disable(),__sync_filesystem()andsync_fs_one_sb(). Forsync_fs_one_sb()we want to ignore the return value; for__sync_filesystem()we almost certainly do *not* - it ends with return__sync_blockdev(sb->s_bdev, wait), after all. The question for that one is whether we want__sync_blockdev()called even in case of ->sync_fs() reporting a failure, and I suspect that it's safer to call it anyway and return the first error value we'd got. No idea about quota situation.
Quoted 8 lines
WRT quota situation: All the ->sync_fs() calls there are due to cache coherency reasons (we need to get quota changes to disk, then prune quota files's page cache, and then userspace can read current quota structures from the disk). We don't want to fail
dquot_disable()just because caches might be incoherent so ignoring ->sync_fs() return value there is fine. Withdquot_quota_sync()it might make some sense to return the error - that's just a backend forQ_SYNCquotactl(2). OTOH I'm not sure anybody really cares -Q_SYNCis rarely used.
Thanks Jan. May be I will leave dquot_quota_sync() untouched for now. When somebody has a need to capture return code from ->sync_fs() there, it can be easily added.
Vivek
Re: [PATCH 1/3] vfs: add new f_op->syncfs vector
In reply to Al ViroQuoted 31 lines · Al Viro
[Christoph added to Cc...]
Quoted 9 lines · Vivek Goyal
Current implementation of
__sync_filesystem()ignores the return code from ->sync_fs(). I am not sure why that's the case. There must have been some historical reason for this.Ignoring ->sync_fs() return code is problematic for overlayfs where it can return error if
sync_filesystem()on upper super block failed. That error will simply be lost andsycnfs(overlay_fd), will get success (despite the fact it failed).Quoted 8 lines
If we modify existing implementation, there is a concern that it will lead to user space visible behavior changes and break things. So instead implement a new
file_operations->syncfs()call which will be called insyncfs()syscall path. Return code from this new call will be captured. And all the writeback error detection logic can go in there as well. Only filesystems which implement this call get affected by this change. Others continue to fallback to existing mechanism.That smells like a massive source of confusion down the road. I'd just looked through the existing instances; many always return 0, but quite a few sometimes try to return an error:
fs/btrfs/super.c:2412: .sync_fs = btrfs_sync_fs, fs/exfat/super.c:204: .sync_fs = exfat_sync_fs, fs/ext4/super.c:1674: .sync_fs = ext4_sync_fs, fs/f2fs/super.c:2480: .sync_fs = f2fs_sync_fs, fs/gfs2/super.c:1600: .sync_fs = gfs2_sync_fs, fs/hfsplus/super.c:368: .sync_fs = hfsplus_sync_fs, fs/nilfs2/super.c:689: .sync_fs = nilfs_sync_fs, fs/ocfs2/super.c:139: .sync_fs = ocfs2_sync_fs,
Quoted 11 lines
fs/overlayfs/super.c:399: .sync_fs =ovl_sync_fs,fs/ubifs/super.c:2052: .sync_fs =ubifs_sync_fs,
is the list of such. There are 4 method callers:dquot_quota_sync(),dquot_disable(),__sync_filesystem()andsync_fs_one_sb(). Forsync_fs_one_sb()we want to ignore the return value; for__sync_filesystem()we almost certainly do *not* - it ends with return__sync_blockdev(sb->s_bdev, wait), after all. The question for that one is whether we want__sync_blockdev()called even in case of ->sync_fs() reporting a failure, and I suspect that it's safer to call it anyway and return the first error value we'd got.
I posted V1 patch to do exactly above. In __sync_filesystem(), capture return code from ->sync_fs() but continue to call __sync_blockdev() and and return error code from ->sync_fs() if there is one otherwise return error code from __sync_blockdev().
https://lore.kernel.org/linux-fsdevel/[email protected]/
Thanks
Vivek
Quoted 2 lines
No idea about quota situation.
Re: [PATCH 1/3] vfs: add new f_op->syncfs vector
In reply to Al ViroQuoted 31 lines · Al Viro
[Christoph added to Cc...]
Quoted 9 lines · Vivek Goyal
Current implementation of
__sync_filesystem()ignores the return code from ->sync_fs(). I am not sure why that's the case. There must have been some historical reason for this.Ignoring ->sync_fs() return code is problematic for overlayfs where it can return error if
sync_filesystem()on upper super block failed. That error will simply be lost andsycnfs(overlay_fd), will get success (despite the fact it failed).Quoted 8 lines
If we modify existing implementation, there is a concern that it will lead to user space visible behavior changes and break things. So instead implement a new
file_operations->syncfs()call which will be called insyncfs()syscall path. Return code from this new call will be captured. And all the writeback error detection logic can go in there as well. Only filesystems which implement this call get affected by this change. Others continue to fallback to existing mechanism.That smells like a massive source of confusion down the road. I'd just looked through the existing instances; many always return 0, but quite a few sometimes try to return an error:
fs/btrfs/super.c:2412: .sync_fs = btrfs_sync_fs, fs/exfat/super.c:204: .sync_fs = exfat_sync_fs, fs/ext4/super.c:1674: .sync_fs = ext4_sync_fs, fs/f2fs/super.c:2480: .sync_fs = f2fs_sync_fs, fs/gfs2/super.c:1600: .sync_fs = gfs2_sync_fs, fs/hfsplus/super.c:368: .sync_fs = hfsplus_sync_fs, fs/nilfs2/super.c:689: .sync_fs = nilfs_sync_fs, fs/ocfs2/super.c:139: .sync_fs = ocfs2_sync_fs,
Quoted 13 lines
fs/overlayfs/super.c:399: .sync_fs =ovl_sync_fs,fs/ubifs/super.c:2052: .sync_fs =ubifs_sync_fs,
is the list of such. There are 4 method callers:dquot_quota_sync(),dquot_disable(),__sync_filesystem()andsync_fs_one_sb(). Forsync_fs_one_sb()we want to ignore the return value; for__sync_filesystem()we almost certainly do *not* - it ends with return__sync_blockdev(sb->s_bdev, wait), after all. The question for that one is whether we want__sync_blockdev()called even in case of ->sync_fs() reporting a failure, and I suspect that it's safer to call it anyway and return the first error value we'd got. No idea about quota situation.
The only problem there is that makes it a bit difficult to override the error return to syncfs, which is really what overlayfs wants to be able to do. Their syncfs syncs out the upper layer, so it makes sense to just have their file->f_sb_err track the upper layer's sb->s_wb_err.
You can plumb the errors from sync_fs all the way through to the syncfs syscall, but we can't currently tell whether we're doing the sync_fs op on behalf of sync(), syncfs() or something else entirely. We need to ensure that if it does return an error, that it doesn't get dropped on the floor.
I think it'd be simpler to just add f_op->syncfs and change s_op->sync_fs to a different name, to lessen the confusion. s_op->sync_fs sort of makes it look like you're implementing syncfs(2), but there's a bit more to it than that.
Maybe s_op->sync_filesystem? There are only about 113 instances "sync_fs" in the tree. Changing the name might also help highlight the fact that the return code won't be ignored like it used to be.
Signature · 1 line
Jeff Layton <[email protected]>
PATCH 2/3 overlayfs: Implement f_op->syncfs() call
In reply to Vivek GoyalProvide an implementation for ->syncfs(). Now if there is an error returned by sync_filesystem(upper_sb), it will be visible to user space. Currently in ovl_sync_fs() path, this error is ignored by VFS.
A later patch also adds logic to detect writeback error.
--- fs/overlayfs/file.c | 1 + fs/overlayfs/overlayfs.h | 3 +++ fs/overlayfs/readdir.c | 1 + fs/overlayfs/super.c | 30 ++++++++++++++++++++++++++++++ 4 files changed, 35 insertions(+)
fs/overlayfs/file.c +1 -0
@@ -806,6 +806,7 @@ const struct file_operations ovl_file_operations = { .copy_file_range = ovl_copy_file_range, .remap_file_range = ovl_remap_file_range, + .syncfs = ovl_syncfs, };int __init ovl_aio_request_cache_init (void )
fs/overlayfs/overlayfs.h +3 -0
@@ -520,3 +520,6 @@ int ovl_set_origin(struct dentry *dentry, struct dentry *lower,/* export.c */ extern const struct export_operations ovl_export_operations; + +/* super.c */ +int ovl_syncfs (struct file *file);
fs/overlayfs/readdir.c +1 -0
@@ -975,6 +975,7 @@ const struct file_operations ovl_dir_operations = {#ifdef CONFIG_COMPAT .compat_ioctl = ovl_compat_ioctl, #endif + .syncfs = ovl_syncfs, };int ovl_check_empty_dir (struct dentry *dentry,struct list_head *list)
fs/overlayfs/super.c +30 -0
@@ -286,6 +286,36 @@ static int ovl_sync_fs(struct super_block *sb, int wait)return ret; } +int ovl_syncfs (struct file *file) +{ +struct super_block *sb = file->f_path.dentry->d_sb; +struct ovl_fs *ofs = sb->s_fs_info; +struct super_block *upper_sb; +int ret; + + ret =0 ; +down_read (&sb->s_umount); +if (sb_rdonly (sb)) + goto out; + +if (!ovl_upper_mnt (ofs)) + goto out; + +if (!ovl_should_sync (ofs)) + goto out; + + upper_sb =ovl_upper_mnt (ofs)->mnt_sb; + +down_read (&upper_sb->s_umount); + ret =sync_filesystem (upper_sb); +up_read (&upper_sb->s_umount); + + +out: +up_read (&sb->s_umount); +return ret; +} +/** * ovl_statfs * @sb: The overlayfs super block
Signature · 1 line
2.25.4
PATCH 3/3 overlayfs: Check writeback errors w.r.t upper in ->syncfs()
In reply to Vivek GoyalCheck for writeback error on overlay super block w.r.t "struct file" passed in ->syncfs().
As of now real error happens on upper sb. So this patch first propagates error from upper sb to overlay sb and then checks error w.r.t struct file passed in.
Jeff, I know you prefer that I should rather file upper file and check error directly on on upper sb w.r.t this real upper file. While I was implementing that I thought what if file is on lower (and has not been copied up yet). In that case shall we not check writeback errors and return back to user space? That does not sound right though because, we are not checking for writeback errors on this file. Rather we are checking for any error on superblock. Upper might have an error and we should report it to user even if file in question is a lower file. And that's why I fell back to this approach. But I am open to change it if there are issues in this method.
--- fs/overlayfs/ovl_entry.h | 2 ++ fs/overlayfs/super.c | 15 ++++++++++++--- 2 files changed, 14 insertions(+), 3 deletions(-)
fs/overlayfs/ovl_entry.h +2 -0
@@ -79,6 +79,8 @@ struct ovl_fs {atomic_long_t last_ino;/* Whiteout dentry cache */ struct dentry *whiteout; +/* Protects multiple sb->s_wb_err update from upper_sb . */ +spinlock_t errseq_lock; };static inline struct vfsmount *ovl_upper_mnt (struct ovl_fs *ofs)
fs/overlayfs/super.c +12 -3
@@ -291,7 +291,7 @@ int ovl_syncfs(struct file *file)struct super_block *sb = file->f_path.dentry->d_sb;struct ovl_fs *ofs = sb->s_fs_info;struct super_block *upper_sb; -int ret; +int ret, ret2; ret =0 ;down_read (&sb->s_umount);
@@ -310,10 +310,18 @@ int ovl_syncfs(struct file *file) ret =sync_filesystem (upper_sb);up_read (&upper_sb->s_umount); +/* Update overlay sb->s_wb_err */ +if (errseq_check (&upper_sb->s_wb_err, sb->s_wb_err)) { +/* Upper sb has errors since last time */ +spin_lock (&ofs->errseq_lock); +errseq_check_and_advance (&upper_sb->s_wb_err, &sb->s_wb_err); +spin_unlock (&ofs->errseq_lock); + } + ret2 =errseq_check_and_advance (&sb->s_wb_err, &file->f_sb_err); out:up_read (&sb->s_umount); -return ret; +return ret ? ret : ret2; }/**
@@ -1903,6 +1911,7 @@ static int ovl_fill_super(struct super_block *sb, void *data, int silent)if (!cred) goto out_err; +spin_lock_init (&ofs->errseq_lock);/* Is there a reason anyone would want not to share whiteouts? */ ofs->share_whiteout = true;
@@ -1975,7 +1984,7 @@ static int ovl_fill_super(struct super_block *sb, void *data, int silent) sb->s_stack_depth =ovl_upper_mnt (ofs)->mnt_sb->s_stack_depth; sb->s_time_gran =ovl_upper_mnt (ofs)->mnt_sb->s_time_gran; - + sb->s_wb_err =errseq_sample (&ovl_upper_mnt (ofs)->mnt_sb->s_wb_err); } oe =ovl_get_lowerstack (sb, splitlower, numlower, ofs, layers); err =PTR_ERR (oe);
Signature · 1 line
2.25.4
Re: [PATCH 3/3] overlayfs: Check writeback errors w.r.t upper in ->syncfs()
Quoted 51 lines · Vivek Goyal
Check for writeback error on overlay super block w.r.t "struct file" passed in ->syncfs().
As of now real error happens on upper sb. So this patch first propagates error from upper sb to overlay sb and then checks error w.r.t struct file passed in.
Jeff, I know you prefer that I should rather file upper file and check error directly on on upper sb w.r.t this real upper file. While I was implementing that I thought what if file is on lower (and has not been copied up yet). In that case shall we not check writeback errors and return back to user space? That does not sound right though because, we are not checking for writeback errors on this file. Rather we are checking for any error on superblock. Upper might have an error and we should report it to user even if file in question is a lower file. And that's why I fell back to this approach. But I am open to change it if there are issues in this method.
Signed-off-by Vivek Goyal <[email protected]>--- fs/overlayfs/ovl_entry.h | 2 ++ fs/overlayfs/super.c | 15 ++++++++++++--- 2 files changed, 14 insertions(+), 3 deletions(-)fs/overlayfs/ovl_entry.h +2
-0@@ -79,6 +79,8 @@ struct ovl_fs {atomic_long_t last_ino;/* Whiteout dentry cache */ struct dentry *whiteout; +/* Protects multiple sb->s_wb_err update from upper_sb . */ +spinlock_t errseq_lock; };static inline struct vfsmount *ovl_upper_mnt (struct ovl_fs *ofs)fs/overlayfs/super.c +1
-1@@ -291,7 +291,7 @@ int ovl_syncfs(struct file *file)struct super_block *sb = file->f_path.dentry->d_sb;struct ovl_fs *ofs = sb->s_fs_info;struct super_block *upper_sb; -int ret; +int ret, ret2; ret =0 ;down_read (&sb->s_umount);
fs/overlayfs/super.c int ovl_syncfs(struct file *file) · line 319
Quoted 10 lines
fs/overlayfs/super.c +7
-0@@ -310,10 +310,18 @@ int ovl_syncfs(struct file *file) ret =sync_filesystem (upper_sb);up_read (&upper_sb->s_umount); +/* Update overlay sb->s_wb_err */ +if (errseq_check (&upper_sb->s_wb_err, sb->s_wb_err)) { +/* Upper sb has errors since last time */ +spin_lock (&ofs->errseq_lock); +errseq_check_and_advance (&upper_sb->s_wb_err, &sb->s_wb_err); +spin_unlock (&ofs->errseq_lock); + }
So, the problem here is that the resulting value in sb->s_wb_err is going to end up with the REPORTED flag set (using the naming in my latest set). So, a later opener of a file on sb->s_wb_err won't see it.
For instance, suppose you call sync() on the box and does the above check and advance. Then, you open the file and call syncfs() and get back no error because REPORTED flag was set when you opened. That error will then be lost.
Quoted 18 lines
fs/overlayfs/super.c +3
-1+ ret2 =errseq_check_and_advance (&sb->s_wb_err, &file->f_sb_err); out:up_read (&sb->s_umount); -return ret; +return ret ? ret : ret2; }/** @@ -1903,6 +1911,7 @@ static int ovl_fill_super(struct super_block *sb, void *data, int silent)if (!cred) goto out_err; +spin_lock_init (&ofs->errseq_lock);/* Is there a reason anyone would want not to share whiteouts? */ ofs->share_whiteout = true;
fs/overlayfs/super.c static int ovl_fill_super(struct super_block *sb, void *data, int silent) · line 1987
Quoted 5 lines
fs/overlayfs/super.c +1
-1@@ -1975,7 +1984,7 @@ static int ovl_fill_super(struct super_block *sb, void *data, int silent) sb->s_stack_depth =ovl_upper_mnt (ofs)->mnt_sb->s_stack_depth; sb->s_time_gran =ovl_upper_mnt (ofs)->mnt_sb->s_time_gran; - + sb->s_wb_err =errseq_sample (&ovl_upper_mnt (ofs)->mnt_sb->s_wb_err);
This will mark the error on the upper_sb as REPORTED, and that's not really that's the case if you're just using it set s_wb_err in the overlay. You might want to use errseq_peek in this situation.
Quoted 6 lines
fs/overlayfs/super.c +0
-0} oe =ovl_get_lowerstack (sb, splitlower, numlower, ofs, layers); err =PTR_ERR (oe);Signature · 1 line
2.25.4
Re: [PATCH 3/3] overlayfs: Check writeback errors w.r.t upper in ->syncfs()
Quoted 67 lines · Jeffrey Layton
Quoted 51 lines · Vivek Goyal
Check for writeback error on overlay super block w.r.t "struct file" passed in ->syncfs().
As of now real error happens on upper sb. So this patch first propagates error from upper sb to overlay sb and then checks error w.r.t struct file passed in.
Jeff, I know you prefer that I should rather file upper file and check error directly on on upper sb w.r.t this real upper file. While I was implementing that I thought what if file is on lower (and has not been copied up yet). In that case shall we not check writeback errors and return back to user space? That does not sound right though because, we are not checking for writeback errors on this file. Rather we are checking for any error on superblock. Upper might have an error and we should report it to user even if file in question is a lower file. And that's why I fell back to this approach. But I am open to change it if there are issues in this method.
Signed-off-by Vivek Goyal <[email protected]>--- fs/overlayfs/ovl_entry.h | 2 ++ fs/overlayfs/super.c | 15 ++++++++++++--- 2 files changed, 14 insertions(+), 3 deletions(-)fs/overlayfs/ovl_entry.h +2
-0@@ -79,6 +79,8 @@ struct ovl_fs {atomic_long_t last_ino;/* Whiteout dentry cache */ struct dentry *whiteout; +/* Protects multiple sb->s_wb_err update from upper_sb . */ +spinlock_t errseq_lock; };static inline struct vfsmount *ovl_upper_mnt (struct ovl_fs *ofs)fs/overlayfs/super.c +1
-1@@ -291,7 +291,7 @@ int ovl_syncfs(struct file *file)struct super_block *sb = file->f_path.dentry->d_sb;struct ovl_fs *ofs = sb->s_fs_info;struct super_block *upper_sb; -int ret; +int ret, ret2; ret =0 ;down_read (&sb->s_umount);fs/overlayfs/super.c int ovl_syncfs(struct file *file) · line 319
Quoted 10 lines
fs/overlayfs/super.c +7
-0@@ -310,10 +310,18 @@ int ovl_syncfs(struct file *file) ret =sync_filesystem (upper_sb);up_read (&upper_sb->s_umount); +/* Update overlay sb->s_wb_err */ +if (errseq_check (&upper_sb->s_wb_err, sb->s_wb_err)) { +/* Upper sb has errors since last time */ +spin_lock (&ofs->errseq_lock); +errseq_check_and_advance (&upper_sb->s_wb_err, &sb->s_wb_err); +spin_unlock (&ofs->errseq_lock); + }So, the problem here is that the resulting value in
sb->s_wb_erris going to end up with the REPORTED flag set (using the naming in my latest set). So, a later opener of a file onsb->s_wb_errwon't see it.
Quoted 4 lines
For instance, suppose you call
sync()on the box and does the above check and advance. Then, you open the file and callsyncfs()and get back no error because REPORTED flag was set when you opened. That error will then be lost.
Hi Jeff,
In this patch, I am doing this only in ->syncfs() path and not in ->sync_fs() path. IOW, errseq_check_and_advance() will take place only if there is a valid "struct file" passed in. That means there is a consumer of the error and that means it should be fine to set the sb->s_wb_err as SEEN/REPORTED, right?
If we end up plumbming "struct file" in existing ->sync_fs() routine, then I will call this only if a non NULL struct file has been passed in. Otherwise skip this step.
IOW, sync() call will not result in errseq_check_and_advance() instead a syncfs() call will.
Quoted 20 lines
Quoted 20 lines
fs/overlayfs/super.c +3
-1+ ret2 =errseq_check_and_advance (&sb->s_wb_err, &file->f_sb_err); out:up_read (&sb->s_umount); -return ret; +return ret ? ret : ret2; }/** @@ -1903,6 +1911,7 @@ static int ovl_fill_super(struct super_block *sb, void *data, int silent)if (!cred) goto out_err; +spin_lock_init (&ofs->errseq_lock);/* Is there a reason anyone would want not to share whiteouts? */ ofs->share_whiteout = true;@@ -1975,7 +1984,7 @@ static int ovl_fill_super(struct super_block *sb, void *data, int silent)
Quoted 8 lines
fs/overlayfs/super.c static int ovl_fill_super(struct super_block *sb, void *data, int silent) · line 1987
Quoted 4 lines
fs/overlayfs/super.c +1
-1sb->s_stack_depth =ovl_upper_mnt (ofs)->mnt_sb->s_stack_depth; sb->s_time_gran =ovl_upper_mnt (ofs)->mnt_sb->s_time_gran; - + sb->s_wb_err =errseq_sample (&ovl_upper_mnt (ofs)->mnt_sb->s_wb_err);This will mark the error on the
upper_sbas REPORTED, and that's not really that's the case if you're just using it sets_wb_errin the overlay. You might want to useerrseq_peekin this situation.
For now I am still looking at existing code and not new code. Because I belive that new code does not change existing behavior instead provides additional functionality to allow sampling the error without marking it seen as well as provide helper to not force seeing an unseen error.
So current errseq_sample() does not mark error SEEN. And if it is an unseen error, we will get 0 and be forced to see the error next time.
One small issue with this is that say upper has unseen error. Now we mount overlay and save that value in sb->s_wb_err (unseen). Say a file is opened on upper and error is now seen on upper. But we still have unseen error cached in overlay and if overlay fd is now opened, f->f_sb_err will be 0 and it will be forced to see err on next syncfs().
IOW, despite the fact that overlay fd was opened after upper sb had been marked seen, it still will see error. I think it probably is not a big issue.
Vivek
Quoted 8 lines
Quoted 7 lines
fs/overlayfs/super.c +0
-0} oe =ovl_get_lowerstack (sb, splitlower, numlower, ofs, layers); err =PTR_ERR (oe);Signature · 1 line
2.25.4
Re: [PATCH 3/3] overlayfs: Check writeback errors w.r.t upper in ->syncfs()
Quoted 81 lines · Vivek Goyal
Quoted 67 lines · Jeffrey Layton
Quoted 51 lines · Vivek Goyal
Check for writeback error on overlay super block w.r.t "struct file" passed in ->syncfs().
As of now real error happens on upper sb. So this patch first propagates error from upper sb to overlay sb and then checks error w.r.t struct file passed in.
Jeff, I know you prefer that I should rather file upper file and check error directly on on upper sb w.r.t this real upper file. While I was implementing that I thought what if file is on lower (and has not been copied up yet). In that case shall we not check writeback errors and return back to user space? That does not sound right though because, we are not checking for writeback errors on this file. Rather we are checking for any error on superblock. Upper might have an error and we should report it to user even if file in question is a lower file. And that's why I fell back to this approach. But I am open to change it if there are issues in this method.
Signed-off-by Vivek Goyal <[email protected]>--- fs/overlayfs/ovl_entry.h | 2 ++ fs/overlayfs/super.c | 15 ++++++++++++--- 2 files changed, 14 insertions(+), 3 deletions(-)fs/overlayfs/ovl_entry.h +2
-0@@ -79,6 +79,8 @@ struct ovl_fs {atomic_long_t last_ino;/* Whiteout dentry cache */ struct dentry *whiteout; +/* Protects multiple sb->s_wb_err update from upper_sb . */ +spinlock_t errseq_lock; };static inline struct vfsmount *ovl_upper_mnt (struct ovl_fs *ofs)fs/overlayfs/super.c +1
-1@@ -291,7 +291,7 @@ int ovl_syncfs(struct file *file)struct super_block *sb = file->f_path.dentry->d_sb;struct ovl_fs *ofs = sb->s_fs_info;struct super_block *upper_sb; -int ret; +int ret, ret2; ret =0 ;down_read (&sb->s_umount);fs/overlayfs/super.c int ovl_syncfs(struct file *file) · line 319
Quoted 10 lines
fs/overlayfs/super.c +7
-0@@ -310,10 +310,18 @@ int ovl_syncfs(struct file *file) ret =sync_filesystem (upper_sb);up_read (&upper_sb->s_umount); +/* Update overlay sb->s_wb_err */ +if (errseq_check (&upper_sb->s_wb_err, sb->s_wb_err)) { +/* Upper sb has errors since last time */ +spin_lock (&ofs->errseq_lock); +errseq_check_and_advance (&upper_sb->s_wb_err, &sb->s_wb_err); +spin_unlock (&ofs->errseq_lock); + }So, the problem here is that the resulting value in
sb->s_wb_erris going to end up with the REPORTED flag set (using the naming in my latest set). So, a later opener of a file onsb->s_wb_errwon't see it.Quoted 4 lines
For instance, suppose you call
sync()on the box and does the above check and advance. Then, you open the file and callsyncfs()and get back no error because REPORTED flag was set when you opened. That error will then be lost.Hi Jeff,
In this patch, I am doing this only in ->syncfs() path and not in ->sync_fs() path. IOW,
errseq_check_and_advance()will take place only if there is a valid "struct file" passed in. That means there is a consumer of the error and that means it should be fine to set thesb->s_wb_erras SEEN/REPORTED, right?
Quoted 7 lines
If we end up plumbming "struct file" in existing ->sync_fs() routine, then I will call this only if a non NULL struct file has been passed in. Otherwise skip this step.
IOW,
sync()call will not result inerrseq_check_and_advance()instead asyncfs()call will.
It still seems odd and I'm not sure you won't end up with weird corner cases due to the flag handling. If you're doing this in the new f_op->syncfs, then why bother with sb->s_wb_err at all? You can just do this, and avoid the overlayfs sb altogether:
if (errseq_check(&upper_sb->s_wb_err, file->f_sb_err)) {
/* Upper sb has errors since last time */
spin_lock(&file->f_lock);
errseq_check_and_advance(&upper_sb->s_wb_err, &file->f_sb_err);
spin_unlock(&file->f_lock);
}That's simpler than trying to propagate the error between two errseq_t's. You would need to sample the upper_sb->s_wb_err at open time in the overlayfs ->open handler though, to make sure you're tracking the right one.
Quoted 46 lines
Quoted 20 lines
Quoted 20 lines
fs/overlayfs/super.c +3
-1+ ret2 =errseq_check_and_advance (&sb->s_wb_err, &file->f_sb_err); out:up_read (&sb->s_umount); -return ret; +return ret ? ret : ret2; }/** @@ -1903,6 +1911,7 @@ static int ovl_fill_super(struct super_block *sb, void *data, int silent)if (!cred) goto out_err; +spin_lock_init (&ofs->errseq_lock);/* Is there a reason anyone would want not to share whiteouts? */ ofs->share_whiteout = true;@@ -1975,7 +1984,7 @@ static int ovl_fill_super(struct super_block *sb, void *data, int silent)Quoted 8 lines
fs/overlayfs/super.c static int ovl_fill_super(struct super_block *sb, void *data, int silent) · line 1987
Quoted 4 lines
fs/overlayfs/super.c +1
-1sb->s_stack_depth =ovl_upper_mnt (ofs)->mnt_sb->s_stack_depth; sb->s_time_gran =ovl_upper_mnt (ofs)->mnt_sb->s_time_gran; - + sb->s_wb_err =errseq_sample (&ovl_upper_mnt (ofs)->mnt_sb->s_wb_err);This will mark the error on the
upper_sbas REPORTED, and that's not really that's the case if you're just using it sets_wb_errin the overlay. You might want to useerrseq_peekin this situation.For now I am still looking at existing code and not new code. Because I belive that new code does not change existing behavior instead provides additional functionality to allow sampling the error without marking it seen as well as provide helper to not force seeing an unseen error.
So current
errseq_sample()does not mark error SEEN. And if it is an unseen error, we will get 0 and be forced to see the error next time.One small issue with this is that say upper has unseen error. Now we mount overlay and save that value in
sb->s_wb_err(unseen). Say a file is opened on upper and error is now seen on upper. But we still have unseen error cached in overlay and if overlay fd is now opened,f->f_sb_errwill be 0 and it will be forced to see err on nextsyncfs().
Quoted 4 lines
IOW, despite the fact that overlay fd was opened after upper sb had been marked seen, it still will see error. I think it probably is not a big issue.
Good point. I was thinking about the newer code that may mark it OBSERVED when you sample at open time.
Still, I think working with the overlayfs sb->s_wb_err is just adding complexity for little benefit. Assuming that writeback errors can only happen on the upper layer, you're better off avoiding it.
Quoted 10 lines
Quoted 9 lines
Quoted 8 lines
fs/overlayfs/super.c +0
-0} oe =ovl_get_lowerstack (sb, splitlower, numlower, ofs, layers); err =PTR_ERR (oe);Signature · 1 line
2.25.4
Re: [PATCH 3/3] overlayfs: Check writeback errors w.r.t upper in ->syncfs()
Quoted 102 lines · Jeff Layton
Quoted 81 lines · Vivek Goyal
Quoted 67 lines · Jeffrey Layton
Quoted 51 lines · Vivek Goyal
Check for writeback error on overlay super block w.r.t "struct file" passed in ->syncfs().
As of now real error happens on upper sb. So this patch first propagates error from upper sb to overlay sb and then checks error w.r.t struct file passed in.
Jeff, I know you prefer that I should rather file upper file and check error directly on on upper sb w.r.t this real upper file. While I was implementing that I thought what if file is on lower (and has not been copied up yet). In that case shall we not check writeback errors and return back to user space? That does not sound right though because, we are not checking for writeback errors on this file. Rather we are checking for any error on superblock. Upper might have an error and we should report it to user even if file in question is a lower file. And that's why I fell back to this approach. But I am open to change it if there are issues in this method.
Signed-off-by Vivek Goyal <[email protected]>--- fs/overlayfs/ovl_entry.h | 2 ++ fs/overlayfs/super.c | 15 ++++++++++++--- 2 files changed, 14 insertions(+), 3 deletions(-)fs/overlayfs/ovl_entry.h +2
-0@@ -79,6 +79,8 @@ struct ovl_fs {atomic_long_t last_ino;/* Whiteout dentry cache */ struct dentry *whiteout; +/* Protects multiple sb->s_wb_err update from upper_sb . */ +spinlock_t errseq_lock; };static inline struct vfsmount *ovl_upper_mnt (struct ovl_fs *ofs)fs/overlayfs/super.c +1
-1@@ -291,7 +291,7 @@ int ovl_syncfs(struct file *file)struct super_block *sb = file->f_path.dentry->d_sb;struct ovl_fs *ofs = sb->s_fs_info;struct super_block *upper_sb; -int ret; +int ret, ret2; ret =0 ;down_read (&sb->s_umount);fs/overlayfs/super.c int ovl_syncfs(struct file *file) · line 319
Quoted 10 lines
fs/overlayfs/super.c +7
-0@@ -310,10 +310,18 @@ int ovl_syncfs(struct file *file) ret =sync_filesystem (upper_sb);up_read (&upper_sb->s_umount); +/* Update overlay sb->s_wb_err */ +if (errseq_check (&upper_sb->s_wb_err, sb->s_wb_err)) { +/* Upper sb has errors since last time */ +spin_lock (&ofs->errseq_lock); +errseq_check_and_advance (&upper_sb->s_wb_err, &sb->s_wb_err); +spin_unlock (&ofs->errseq_lock); + }So, the problem here is that the resulting value in
sb->s_wb_erris going to end up with the REPORTED flag set (using the naming in my latest set). So, a later opener of a file onsb->s_wb_errwon't see it.Quoted 4 lines
For instance, suppose you call
sync()on the box and does the above check and advance. Then, you open the file and callsyncfs()and get back no error because REPORTED flag was set when you opened. That error will then be lost.Hi Jeff,
In this patch, I am doing this only in ->syncfs() path and not in ->sync_fs() path. IOW,
errseq_check_and_advance()will take place only if there is a valid "struct file" passed in. That means there is a consumer of the error and that means it should be fine to set thesb->s_wb_erras SEEN/REPORTED, right?Quoted 7 lines
If we end up plumbming "struct file" in existing ->sync_fs() routine, then I will call this only if a non NULL struct file has been passed in. Otherwise skip this step.
IOW,
sync()call will not result inerrseq_check_and_advance()instead asyncfs()call will.It still seems odd and I'm not sure you won't end up with weird corner cases due to the flag handling. If you're doing this in the new
f_op->syncfs, then why bother withsb->s_wb_errat all? You can just do this, and avoid the overlayfs sb altogether:if (errseq_check(&upper_sb->s_wb_err, file->f_sb_err)) { /* Upper sb has errors since last time */ spin_lock(&file->f_lock); errseq_check_and_advance(&upper_sb->s_wb_err, &file->f_sb_err); spin_unlock(&file->f_lock); }
Quoted 4 lines
That's simpler than trying to propagate the error between two errseq_t's. You would need to sample the
upper_sb->s_wb_errat open time in the overlayfs ->open handler though, to make sure you're tracking the right one.
IIUC, you are suggesting that when and overlay file is opened (lower or upper), always install current upper_sb->s_wb_err in f->f_sb_err. IOW, overide following VFS operations.
f->f_sb_err = file_sample_sb_err(f);
In ovl_open() and ovl_dir_open() with something like.
f->f_sb_err = errseq_sample(upper_sb->s_wb_err);
And then ->sync_fs() or ->syncfs(), can check for new errors w.r.t upper sb?
if (errseq_check(&upper_sb->s_wb_err, file->f_sb_err)) {
/* Upper sb has errors since last time */
spin_lock(&file->f_lock);
ret = errseq_check_and_advance(&upper_sb->s_wb_err, &file->f_sb_err);
spin_unlock(&file->f_lock);
}I guess I can try this. But if we don't update ovl_sb->s_wb_err, then question remains that how to avoid errseq_check_and_advance() call in SYSCALL(sycnfs). That will do more bad things in this case.
This will lead back to either creating new f_op->syncfs() where fs is responsible for writeback error checks (and not vfs). Or plumb "struct file" in exisitng ->sync_fs() and let filesystems do error checks (instead of VFS). This will be somewhat similar to your old proposal here.
https://lore.kernel.org/linux-fsdevel/[email protected]/
So advantage of updating ovl_sb->s_wb_err is that it reduces the churn needed in ->sync_fs() and moving errseq_check_and_advance() check out of vfs syncfs().
Quoted 52 lines
Quoted 47 lines
Quoted 21 lines
Quoted 21 lines
fs/overlayfs/super.c +3
-1+ ret2 =errseq_check_and_advance (&sb->s_wb_err, &file->f_sb_err); out:up_read (&sb->s_umount); -return ret; +return ret ? ret : ret2; }/** @@ -1903,6 +1911,7 @@ static int ovl_fill_super(struct super_block *sb, void *data, int silent)if (!cred) goto out_err; +spin_lock_init (&ofs->errseq_lock);/* Is there a reason anyone would want not to share whiteouts? */ ofs->share_whiteout = true;@@ -1975,7 +1984,7 @@ static int ovl_fill_super(struct super_block *sb, void *data, int silent)Quoted 8 lines
fs/overlayfs/super.c static int ovl_fill_super(struct super_block *sb, void *data, int silent) · line 1987
Quoted 4 lines
fs/overlayfs/super.c +1
-1sb->s_stack_depth =ovl_upper_mnt (ofs)->mnt_sb->s_stack_depth; sb->s_time_gran =ovl_upper_mnt (ofs)->mnt_sb->s_time_gran; - + sb->s_wb_err =errseq_sample (&ovl_upper_mnt (ofs)->mnt_sb->s_wb_err);This will mark the error on the
upper_sbas REPORTED, and that's not really that's the case if you're just using it sets_wb_errin the overlay. You might want to useerrseq_peekin this situation.For now I am still looking at existing code and not new code. Because I belive that new code does not change existing behavior instead provides additional functionality to allow sampling the error without marking it seen as well as provide helper to not force seeing an unseen error.
So current
errseq_sample()does not mark error SEEN. And if it is an unseen error, we will get 0 and be forced to see the error next time.One small issue with this is that say upper has unseen error. Now we mount overlay and save that value in
sb->s_wb_err(unseen). Say a file is opened on upper and error is now seen on upper. But we still have unseen error cached in overlay and if overlay fd is now opened,f->f_sb_errwill be 0 and it will be forced to see err on nextsyncfs().Quoted 4 lines
IOW, despite the fact that overlay fd was opened after upper sb had been marked seen, it still will see error. I think it probably is not a big issue.
Quoted 6 lines
Good point. I was thinking about the newer code that may mark it OBSERVED when you sample at open time.
Still, I think working with the overlayfs
sb->s_wb_erris just adding complexity for little benefit. Assuming that writeback errors can only happen on the upper layer, you're better off avoiding it.
If I want to avoid ovl_sb->s_wb_err updation, I will have to move ret2 = errseq_check_and_advance(&sb->s_wb_err, &f.file->f_sb_err); check in individual filesystems. And it will still not be same. Because currently after ->sync_fs() call, __sync_blockdev() is called and then we check for writeback errors. That means, I will have to move __sync_blockdev() also inside ->sync_fs().
Something like.
fs_sync_fs()
{
ret = do_fs_specific_sync_stuff();
ret2 = __sync_blockdev();
ret3 = errseq_check_and_advance(&sb->s_wb_err, &f.file->f_sb_err);
if (ret) {
return ret;
else
return ret2 ? ret2 : ret3;
}Does not look pretty.
Vivek
Re: [PATCH 3/3] overlayfs: Check writeback errors w.r.t upper in ->syncfs()
Quoted 138 lines · Vivek Goyal
Quoted 102 lines · Jeff Layton
Quoted 81 lines · Vivek Goyal
Quoted 67 lines · Jeffrey Layton
Quoted 51 lines · Vivek Goyal
Check for writeback error on overlay super block w.r.t "struct file" passed in ->syncfs().
As of now real error happens on upper sb. So this patch first propagates error from upper sb to overlay sb and then checks error w.r.t struct file passed in.
Jeff, I know you prefer that I should rather file upper file and check error directly on on upper sb w.r.t this real upper file. While I was implementing that I thought what if file is on lower (and has not been copied up yet). In that case shall we not check writeback errors and return back to user space? That does not sound right though because, we are not checking for writeback errors on this file. Rather we are checking for any error on superblock. Upper might have an error and we should report it to user even if file in question is a lower file. And that's why I fell back to this approach. But I am open to change it if there are issues in this method.
Signed-off-by Vivek Goyal <[email protected]>--- fs/overlayfs/ovl_entry.h | 2 ++ fs/overlayfs/super.c | 15 ++++++++++++--- 2 files changed, 14 insertions(+), 3 deletions(-)fs/overlayfs/ovl_entry.h +2
-0@@ -79,6 +79,8 @@ struct ovl_fs {atomic_long_t last_ino;/* Whiteout dentry cache */ struct dentry *whiteout; +/* Protects multiple sb->s_wb_err update from upper_sb . */ +spinlock_t errseq_lock; };static inline struct vfsmount *ovl_upper_mnt (struct ovl_fs *ofs)fs/overlayfs/super.c +1
-1@@ -291,7 +291,7 @@ int ovl_syncfs(struct file *file)struct super_block *sb = file->f_path.dentry->d_sb;struct ovl_fs *ofs = sb->s_fs_info;struct super_block *upper_sb; -int ret; +int ret, ret2; ret =0 ;down_read (&sb->s_umount);fs/overlayfs/super.c int ovl_syncfs(struct file *file) · line 319
Quoted 10 lines
fs/overlayfs/super.c +7
-0@@ -310,10 +310,18 @@ int ovl_syncfs(struct file *file) ret =sync_filesystem (upper_sb);up_read (&upper_sb->s_umount); +/* Update overlay sb->s_wb_err */ +if (errseq_check (&upper_sb->s_wb_err, sb->s_wb_err)) { +/* Upper sb has errors since last time */ +spin_lock (&ofs->errseq_lock); +errseq_check_and_advance (&upper_sb->s_wb_err, &sb->s_wb_err); +spin_unlock (&ofs->errseq_lock); + }So, the problem here is that the resulting value in
sb->s_wb_erris going to end up with the REPORTED flag set (using the naming in my latest set). So, a later opener of a file onsb->s_wb_errwon't see it.Quoted 4 lines
For instance, suppose you call
sync()on the box and does the above check and advance. Then, you open the file and callsyncfs()and get back no error because REPORTED flag was set when you opened. That error will then be lost.Hi Jeff,
In this patch, I am doing this only in ->syncfs() path and not in ->sync_fs() path. IOW,
errseq_check_and_advance()will take place only if there is a valid "struct file" passed in. That means there is a consumer of the error and that means it should be fine to set thesb->s_wb_erras SEEN/REPORTED, right?Quoted 7 lines
If we end up plumbming "struct file" in existing ->sync_fs() routine, then I will call this only if a non NULL struct file has been passed in. Otherwise skip this step.
IOW,
sync()call will not result inerrseq_check_and_advance()instead asyncfs()call will.It still seems odd and I'm not sure you won't end up with weird corner cases due to the flag handling. If you're doing this in the new
f_op->syncfs, then why bother withsb->s_wb_errat all? You can just do this, and avoid the overlayfs sb altogether:if (errseq_check(&upper_sb->s_wb_err, file->f_sb_err)) { /* Upper sb has errors since last time */ spin_lock(&file->f_lock); errseq_check_and_advance(&upper_sb->s_wb_err, &file->f_sb_err); spin_unlock(&file->f_lock); }Quoted 4 lines
That's simpler than trying to propagate the error between two errseq_t's. You would need to sample the
upper_sb->s_wb_errat open time in the overlayfs ->open handler though, to make sure you're tracking the right one.IIUC, you are suggesting that when and overlay file is opened (lower or upper), always install current
upper_sb->s_wb_errinf->f_sb_err. IOW, overide following VFS operations.
f->f_sb_err=file_sample_sb_err(f);In
ovl_open()andovl_dir_open()with something like.
f->f_sb_err=errseq_sample(upper_sb->s_wb_err);And then ->sync_fs() or ->syncfs(), can check for new errors w.r.t upper sb?
if (errseq_check(&upper_sb->s_wb_err, file->f_sb_err)) { /* Upper sb has errors since last time */ spin_lock(&file->f_lock); ret = errseq_check_and_advance(&upper_sb->s_wb_err, &file->f_sb_err); spin_unlock(&file->f_lock); }I guess I can try this. But if we don't update
ovl_sb->s_wb_err, then question remains that how to avoiderrseq_check_and_advance()call in SYSCALL(sycnfs). That will do more bad things in this case.This will lead back to either creating new
f_op->syncfs()where fs is responsible for writeback error checks (and not vfs). Or plumb "struct file" in exisitng ->sync_fs() and let filesystems do error checks (instead of VFS). This will be somewhat similar to your old proposal here.
Quoted 6 lines
https://lore.kernel.org/linux-fsdevel/[email protected]/
So advantage of updating
ovl_sb->s_wb_erris that it reduces the churn needed in ->sync_fs() and movingerrseq_check_and_advance()check out of vfssyncfs().
Correct.
The patch we're discussing here _does_ add a f_op->syncfs, which is why I was suggesting to do it that way.
Quoted 79 lines
Quoted 52 lines
Quoted 47 lines
Quoted 21 lines
Quoted 21 lines
fs/overlayfs/super.c +3
-1+ ret2 =errseq_check_and_advance (&sb->s_wb_err, &file->f_sb_err); out:up_read (&sb->s_umount); -return ret; +return ret ? ret : ret2; }/** @@ -1903,6 +1911,7 @@ static int ovl_fill_super(struct super_block *sb, void *data, int silent)if (!cred) goto out_err; +spin_lock_init (&ofs->errseq_lock);/* Is there a reason anyone would want not to share whiteouts? */ ofs->share_whiteout = true;@@ -1975,7 +1984,7 @@ static int ovl_fill_super(struct super_block *sb, void *data, int silent)Quoted 8 lines
fs/overlayfs/super.c static int ovl_fill_super(struct super_block *sb, void *data, int silent) · line 1987
Quoted 4 lines
fs/overlayfs/super.c +1
-1sb->s_stack_depth =ovl_upper_mnt (ofs)->mnt_sb->s_stack_depth; sb->s_time_gran =ovl_upper_mnt (ofs)->mnt_sb->s_time_gran; - + sb->s_wb_err =errseq_sample (&ovl_upper_mnt (ofs)->mnt_sb->s_wb_err);This will mark the error on the
upper_sbas REPORTED, and that's not really that's the case if you're just using it sets_wb_errin the overlay. You might want to useerrseq_peekin this situation.For now I am still looking at existing code and not new code. Because I belive that new code does not change existing behavior instead provides additional functionality to allow sampling the error without marking it seen as well as provide helper to not force seeing an unseen error.
So current
errseq_sample()does not mark error SEEN. And if it is an unseen error, we will get 0 and be forced to see the error next time.One small issue with this is that say upper has unseen error. Now we mount overlay and save that value in
sb->s_wb_err(unseen). Say a file is opened on upper and error is now seen on upper. But we still have unseen error cached in overlay and if overlay fd is now opened,f->f_sb_errwill be 0 and it will be forced to see err on nextsyncfs().Quoted 4 lines
IOW, despite the fact that overlay fd was opened after upper sb had been marked seen, it still will see error. I think it probably is not a big issue.
Quoted 6 lines
Good point. I was thinking about the newer code that may mark it OBSERVED when you sample at open time.
Still, I think working with the overlayfs
sb->s_wb_erris just adding complexity for little benefit. Assuming that writeback errors can only happen on the upper layer, you're better off avoiding it.If I want to avoid
ovl_sb->s_wb_errupdation, I will have to move ret2 =errseq_check_and_advance(&sb->s_wb_err, &f.file->f_sb_err); check in individual filesystems. And it will still not be same. Because currently after ->sync_fs() call,__sync_blockdev()is called and then we check for writeback errors. That means, I will have to move__sync_blockdev()also inside ->sync_fs().Something like.
fs_sync_fs(){ ret = do_fs_specific_sync_stuff(); ret2 = __sync_blockdev(); ret3 = errseq_check_and_advance(&sb->s_wb_err, &f.file->f_sb_err); if (ret) { return ret; else return ret2 ? ret2 : ret3; }
Quoted 2 lines
Does not look pretty.
If adding a new f_op approach isn't acceptable, then this is quite a bit more difficult, and you'll need to plumb the error through from ->sync_fs to the syncfs syscall wrapper somehow.
That's the main reason I'm advocating for a new f_op. It's a lot more straightforward, and I think it'll be less error-prone over the long haul.
-- Jeff
Re: [PATCH 3/3] overlayfs: Check writeback errors w.r.t upper in ->syncfs()
Quoted 3 lines · Jeffrey Layton
The patch we're discussing here _does_ add a
f_op->syncfs, which is why I was suggesting to do it that way.
I haven't thought through the issues to decide what I think of adding a new op, but I already know what I think of adding ->syncfs. Don't Do It. The name is much too easily confused with ->sync_fs.
If you call it ->sync_fs_return_error() it would be MUCH better.
And having said that, the solution becomes obvious. Add a new flag, either as another bit in 'int wait', or as a new bool. The new flag would be "return_error" - or whatever is appropriate.
NeilBrown
Re: [PATCH 3/3] overlayfs: Check writeback errors w.r.t upper in ->syncfs()
In reply to Vivek GoyalHi Vivek,
url: https://github.com/0day-ci/linux/commits/Vivek-Goyal/vfs-overlayfs-Fix-syncfs-to-return-error/20201217-073932 base: https://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/vfs.git overlayfs-next config: x86_64-randconfig-m001-20201217 (attached as .config) compiler: gcc-9 (Debian 9.3.0-15) 9.3.0
If you fix the issue, kindly add following tag as appropriate
smatch warnings:fs/overlayfs/super.c:327 ovl_syncfs() error: uninitialized symbol 'ret2'.
vim +/ret2 +327 fs/overlayfs/super.c
d0a04f1c11adc0e Vivek Goyal 2020-12-16 292 int ovl_syncfs(struct file *file)
d0a04f1c11adc0e Vivek Goyal 2020-12-16 293 {
d0a04f1c11adc0e Vivek Goyal 2020-12-16 294 struct super_block *sb = file->f_path.dentry->d_sb;
d0a04f1c11adc0e Vivek Goyal 2020-12-16 295 struct ovl_fs *ofs = sb->s_fs_info;
d0a04f1c11adc0e Vivek Goyal 2020-12-16 296 struct super_block *upper_sb;
0bed6122e561e0b Vivek Goyal 2020-12-16 297 int ret, ret2;
d0a04f1c11adc0e Vivek Goyal 2020-12-16 298
d0a04f1c11adc0e Vivek Goyal 2020-12-16 299 ret = 0;
d0a04f1c11adc0e Vivek Goyal 2020-12-16 300 down_read(&sb->s_umount);
d0a04f1c11adc0e Vivek Goyal 2020-12-16 301 if (sb_rdonly(sb))
d0a04f1c11adc0e Vivek Goyal 2020-12-16 302 goto out;
^^^^^^^^^"ret" is zero and "ret2" is uninitialized.
d0a04f1c11adc0e Vivek Goyal 2020-12-16 303
d0a04f1c11adc0e Vivek Goyal 2020-12-16 304 if (!ovl_upper_mnt(ofs))
d0a04f1c11adc0e Vivek Goyal 2020-12-16 305 goto out;
d0a04f1c11adc0e Vivek Goyal 2020-12-16 306
d0a04f1c11adc0e Vivek Goyal 2020-12-16 307 if (!ovl_should_sync(ofs))
d0a04f1c11adc0e Vivek Goyal 2020-12-16 308 goto out;
d0a04f1c11adc0e Vivek Goyal 2020-12-16 309
d0a04f1c11adc0e Vivek Goyal 2020-12-16 310 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
d0a04f1c11adc0e Vivek Goyal 2020-12-16 311
d0a04f1c11adc0e Vivek Goyal 2020-12-16 312 down_read(&upper_sb->s_umount);
d0a04f1c11adc0e Vivek Goyal 2020-12-16 313 ret = sync_filesystem(upper_sb);
d0a04f1c11adc0e Vivek Goyal 2020-12-16 314 up_read(&upper_sb->s_umount);
d0a04f1c11adc0e Vivek Goyal 2020-12-16 315
0bed6122e561e0b Vivek Goyal 2020-12-16 316 /* Update overlay sb->s_wb_err */
0bed6122e561e0b Vivek Goyal 2020-12-16 317 if (errseq_check(&upper_sb->s_wb_err, sb->s_wb_err)) {
0bed6122e561e0b Vivek Goyal 2020-12-16 318 /* Upper sb has errors since last time */
0bed6122e561e0b Vivek Goyal 2020-12-16 319 spin_lock(&ofs->errseq_lock);
0bed6122e561e0b Vivek Goyal 2020-12-16 320 errseq_check_and_advance(&upper_sb->s_wb_err, &sb->s_wb_err);
0bed6122e561e0b Vivek Goyal 2020-12-16 321 spin_unlock(&ofs->errseq_lock);
0bed6122e561e0b Vivek Goyal 2020-12-16 322 }
d0a04f1c11adc0e Vivek Goyal 2020-12-16 323
0bed6122e561e0b Vivek Goyal 2020-12-16 324 ret2 = errseq_check_and_advance(&sb->s_wb_err, &file->f_sb_err);
d0a04f1c11adc0e Vivek Goyal 2020-12-16 325 out:
d0a04f1c11adc0e Vivek Goyal 2020-12-16 326 up_read(&sb->s_umount);
0bed6122e561e0b Vivek Goyal 2020-12-16 @327 return ret ? ret : ret2;
^^^^So we are returning an uninitialized variable.
d0a04f1c11adc0e Vivek Goyal 2020-12-16 328 }
---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all(a)lists.01.org
_______________________________________________
kbuild mailing list -- kbuild(a)lists.01.org
To unsubscribe send an email to kbuild-leave(a)lists.01.org