Skip to contentSkip to conversation

vfs: have syncfs() return error when there are writeback errors

PATCH Superseded Jeff Layton 2 participants 7 messages Apr 15, 2020 – Apr 16, 2020

Subsystems (4)

v5 of 6 Superseded Latest is v6 →

lore
Share & tools
lore
#0 lore raw

v5 cover vfs: have syncfs() return error when there are writeback errors

v5:
- use RCU to ensure that bd_super doesn't go away while we're using it

This is the fifth iteration of this patchset. The main difference from v4 is that this one uses RCU to ensure validity of the bd_super pointer when we're marking it after a buffer_head writeback error.

Jeff Layton (2):
vfs: track per-sb writeback errors and report them to syncfs
buffer: record blockdev write errors in super_block that it backs

 drivers/dax/device.c    |  1 +
 fs/buffer.c             |  7 +++++++
 fs/file_table.c         |  1 +
 fs/open.c               |  3 +--
 fs/sync.c               |  6 ++++--
 include/linux/fs.h      | 16 ++++++++++++++++
 include/linux/pagemap.h |  5 ++++-
 7 files changed, 34 insertions(+), 5 deletions(-)
Signature · 1 line
2.25.2
#1 lore raw

PATCH v5 1/2 vfs: track per-sb writeback errors and report them to syncfs

From: Jeff Layton <[email protected]>

Usually we suggest that applications call fsync when they want to ensure that all data written to the file has made it to the backing store, but that can be inefficient when there are a lot of open files.

Calling syncfs on the filesystem can be more efficient in some situations, but the error reporting doesn't currently work the way most people expect. If a single inode on a filesystem reports a writeback error, syncfs won't necessarily return an error. syncfs only returns an error if __sync_blockdev fails, and on some filesystems that's a no-op.

It would be better if syncfs reported an error if there were any writeback failures. Then applications could call syncfs to see if there are any errors on any open files, and could then call fsync on all of the other descriptors to figure out which one failed.

This patch adds a new errseq_t to struct super_block, and has mapping_set_error also record writeback errors there.

To report those errors, we also need to keep an errseq_t in struct file to act as a cursor. This patch adds a dedicated field for that purpose, which slots nicely into 4 bytes of padding at the end of struct file on x86_64.

An earlier version of this patch used an O_PATH file descriptor to cue the kernel that the open file should track the superblock error and not the inode's writeback error.

I think that API is just too weird though. This is simpler and should make syncfs error reporting "just work" even if someone is multiplexing fsync and syncfs on the same fds.

Cc Andres Freund <[email protected]>
Cc Matthew Wilcox <[email protected]>
---
 drivers/dax/device.c    |  1 +
 fs/file_table.c         |  1 +
 fs/open.c               |  3 +--
 fs/sync.c               |  6 ++++--
 include/linux/fs.h      | 16 ++++++++++++++++
 include/linux/pagemap.h |  5 ++++-
 6 files changed, 27 insertions(+), 5 deletions(-)
drivers/dax/device.c +1 -0
index 1af823b2fe6b..4c0af2eb7e19 100644
@@ -377,6 +377,7 @@ static int dax_open(struct inode *inode, struct file *filp)
 	inode->i_mapping->a_ops = &dev_dax_aops;
 	filp->f_mapping = inode->i_mapping;
 	filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
+	filp->f_sb_err = file_sample_sb_err(filp);
 	filp->private_data = dev_dax;
 	inode->i_flags = S_DAX;
 
fs/file_table.c +1 -0
index 30d55c9a1744..676e620948d2 100644
@@ -198,6 +198,7 @@ static struct file *alloc_file(const struct path *path, int flags,
 	file->f_inode = path->dentry->d_inode;
 	file->f_mapping = path->dentry->d_inode->i_mapping;
 	file->f_wb_err = filemap_sample_wb_err(file->f_mapping);
+	file->f_sb_err = file_sample_sb_err(file);
 	if ((file->f_mode & FMODE_READ) &&
 	     likely(fop->read || fop->read_iter))
 		file->f_mode |= FMODE_CAN_READ;
fs/open.c +1 -2
index 719b320ede52..d9467a8a7f6a 100644
@@ -743,9 +743,8 @@ static int do_dentry_open(struct file *f,
 	path_get(&f->f_path);
 	f->f_inode = inode;
 	f->f_mapping = inode->i_mapping;
-
-	/* Ensure that we skip any errors that predate opening of the file */
 	f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
+	f->f_sb_err = file_sample_sb_err(f);
 
 	if (unlikely(f->f_flags & O_PATH)) {
 		f->f_mode = FMODE_PATH | FMODE_OPENED;
fs/sync.c +4 -2
index 4d1ff010bc5a..c6f6f5be5682 100644
@@ -161,7 +161,7 @@ SYSCALL_DEFINE1(syncfs, int, fd)
 {
 	struct fd f = fdget(fd);
 	struct super_block *sb;
-	int ret;
+	int ret, ret2;
 
 	if (!f.file)
 		return -EBADF;
@@ -171,8 +171,10 @@ SYSCALL_DEFINE1(syncfs, int, fd)
 	ret = sync_filesystem(sb);
 	up_read(&sb->s_umount);
 
+	ret2 = errseq_check_and_advance(&sb->s_wb_err, &f.file->f_sb_err);
+
 	fdput(f);
-	return ret;
+	return ret ? ret : ret2;
 }
 
 /**
include/linux/fs.h +16 -0
index 4f6f59b4f22a..5ad13cd6441c 100644
@@ -976,6 +976,7 @@ struct file {
 #endif /* #ifdef CONFIG_EPOLL */
 	struct address_space	*f_mapping;
 	errseq_t		f_wb_err;
+	errseq_t		f_sb_err; /* for syncfs */
 } __randomize_layout
   __attribute__((aligned(4)));	/* lest something weird decides that 2 is OK */
 
@@ -1520,6 +1521,9 @@ struct super_block {
 	/* Being remounted read-only */
 	int s_readonly_remount;
 
+	/* per-sb errseq_t for reporting writeback errors via syncfs */
+	errseq_t s_wb_err;
+
 	/* AIO completions deferred from interrupt context */
 	struct workqueue_struct *s_dio_done_wq;
 	struct hlist_head s_pins;
@@ -2827,6 +2831,18 @@ static inline errseq_t filemap_sample_wb_err(struct address_space *mapping)
 	return errseq_sample(&mapping->wb_err);
 }
 
+/**
+ * file_sample_sb_err - sample the current errseq_t to test for later errors
+ * @mapping: mapping to be sampled
+ *
+ * Grab the most current superblock-level errseq_t value for the given
+ * struct file.
+ */
+static inline errseq_t file_sample_sb_err(struct file *file)
+{
+	return errseq_sample(&file->f_path.dentry->d_sb->s_wb_err);
+}
+
 static inline int filemap_nr_thps(struct address_space *mapping)
 {
 #ifdef CONFIG_READ_ONLY_THP_FOR_FS
include/linux/pagemap.h +4 -1
index a8f7bd8ea1c6..d4409b13747e 100644
@@ -51,7 +51,10 @@ static inline void mapping_set_error(struct address_space *mapping, int error)
 		return;
 
 	/* Record in wb_err for checkers using errseq_t based tracking */
-	filemap_set_wb_err(mapping, error);
+	__filemap_set_wb_err(mapping, error);
+
+	/* Record it in superblock */
+	errseq_set(&mapping->host->i_sb->s_wb_err, error);
 
 	/* Record it in flags for now, for legacy callers */
 	if (error == -ENOSPC)
Signature · 1 line
2.25.2
#2 lore raw

PATCH v5 2/2 buffer: record blockdev write errors in super_block that it backs

In reply to Jeff Layton

From: Jeff Layton <[email protected]>

When syncing out a block device (a'la __sync_blockdev), any error encountered will only be recorded in the bd_inode's mapping. When the blockdev contains a filesystem however, we'd like to also record the error in the super_block that's stored there.

Make mark_buffer_write_io_error also record the error in the corresponding super_block when a writeback error occurs and the block device contains a mounted superblock.

Since superblocks are RCU freed, hold the rcu_read_lock to ensure that the superblock doesn't go away while we're marking it.

---
 fs/buffer.c | 7 +++++++
 1 file changed, 7 insertions(+)
fs/buffer.c +7 -0
index f73276d746bb..2a4a5cc20418 100644
@@ -1154,12 +1154,19 @@ EXPORT_SYMBOL(mark_buffer_dirty);
 
 void mark_buffer_write_io_error(struct buffer_head *bh)
 {
+	struct super_block *sb;
+
 	set_buffer_write_io_error(bh);
 	/* FIXME: do we need to set this in both places? */
 	if (bh->b_page && bh->b_page->mapping)
 		mapping_set_error(bh->b_page->mapping, -EIO);
 	if (bh->b_assoc_map)
 		mapping_set_error(bh->b_assoc_map, -EIO);
+	rcu_read_lock();
+	sb = bh->b_bdev->bd_super;
+	if (sb)
+		errseq_set(&sb->s_wb_err, -EIO);
+	rcu_read_unlock();
 }
 EXPORT_SYMBOL(mark_buffer_write_io_error);
 
Signature · 1 line
2.25.2
maintainer
#3 lore raw

Re: [PATCH v5 2/2] buffer: record blockdev write errors in super_block that it backs

Quoted 27 lines · Jeff Layton

From: Jeff Layton <[email protected]>

When syncing out a block device (a'la __sync_blockdev), any error encountered will only be recorded in the bd_inode's mapping. When the blockdev contains a filesystem however, we'd like to also record the error in the super_block that's stored there.

Make mark_buffer_write_io_error also record the error in the corresponding super_block when a writeback error occurs and the block device contains a mounted superblock.

Since superblocks are RCU freed, hold the rcu_read_lock to ensure that the superblock doesn't go away while we're marking it.

---
 fs/buffer.c | 7 +++++++
 1 file changed, 7 insertions(+)
fs/buffer.c +0 -0
index f73276d746bb..2a4a5cc20418 100644
@@ -1154,12 +1154,19 @@ EXPORT_SYMBOL(mark_buffer_dirty);
 
 void mark_buffer_write_io_error(struct buffer_head *bh)
 {

fs/buffer.c EXPORT_SYMBOL(mark_buffer_dirty); · line 1166

Quoted 10 lines
fs/buffer.c +4 -0
+	struct super_block *sb;
+
 	set_buffer_write_io_error(bh);
 	/* FIXME: do we need to set this in both places? */
 	if (bh->b_page && bh->b_page->mapping)
 		mapping_set_error(bh->b_page->mapping, -EIO);
 	if (bh->b_assoc_map)
 		mapping_set_error(bh->b_assoc_map, -EIO);
+	rcu_read_lock();
+	sb = bh->b_bdev->bd_super;

You still need READ_ONCE() here. Otherwise the dereference below can still result in refetch and NULL ptr deref.

Honza

Quoted 9 lines
fs/buffer.c +3 -0
+	if (sb)
+		errseq_set(&sb->s_wb_err, -EIO);
+	rcu_read_unlock();
 }
 EXPORT_SYMBOL(mark_buffer_write_io_error);
 
Signature · 1 line
2.25.2
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
#4 lore raw

Re: [PATCH v5 2/2] buffer: record blockdev write errors in super_block that it backs

Quoted 42 lines · Jan Kara
Quoted 27 lines · Jeff Layton

From: Jeff Layton <[email protected]>

When syncing out a block device (a'la __sync_blockdev), any error encountered will only be recorded in the bd_inode's mapping. When the blockdev contains a filesystem however, we'd like to also record the error in the super_block that's stored there.

Make mark_buffer_write_io_error also record the error in the corresponding super_block when a writeback error occurs and the block device contains a mounted superblock.

Since superblocks are RCU freed, hold the rcu_read_lock to ensure that the superblock doesn't go away while we're marking it.

---
 fs/buffer.c | 7 +++++++
 1 file changed, 7 insertions(+)
fs/buffer.c +0 -0
index f73276d746bb..2a4a5cc20418 100644
@@ -1154,12 +1154,19 @@ EXPORT_SYMBOL(mark_buffer_dirty);
 
 void mark_buffer_write_io_error(struct buffer_head *bh)
 {

fs/buffer.c EXPORT_SYMBOL(mark_buffer_dirty); · line 1166

Quoted 10 lines
fs/buffer.c +4 -0
+	struct super_block *sb;
+
 	set_buffer_write_io_error(bh);
 	/* FIXME: do we need to set this in both places? */
 	if (bh->b_page && bh->b_page->mapping)
 		mapping_set_error(bh->b_page->mapping, -EIO);
 	if (bh->b_assoc_map)
 		mapping_set_error(bh->b_assoc_map, -EIO);
+	rcu_read_lock();
+	sb = bh->b_bdev->bd_super;

You still need READ_ONCE() here. Otherwise the dereference below can still result in refetch and NULL ptr deref.

Quoted 2 lines

Honza

Huh? That seems like a really suspicious thing for the compiler/arch to do. We are checking that sb isn't NULL before we dereference it. Doesn't that imply a data dependency? How could the value of "sb" change after that?

I'm also not sure I understand how using READ_ONCE really helps there if we can't count on the value of a local variable not changing.

Quoted 9 lines
Quoted 9 lines
fs/buffer.c +3 -0
+	if (sb)
+		errseq_set(&sb->s_wb_err, -EIO);
+	rcu_read_unlock();
 }
 EXPORT_SYMBOL(mark_buffer_write_io_error);
 
Signature · 1 line
2.25.2
Signature · 1 line
Jeff Layton <[email protected]>
maintainer
#5 lore raw

Re: [PATCH v5 2/2] buffer: record blockdev write errors in super_block that it backs

Quoted 43 lines · Jeff Layton
Quoted 42 lines · Jan Kara
Quoted 27 lines · Jeff Layton

From: Jeff Layton <[email protected]>

When syncing out a block device (a'la __sync_blockdev), any error encountered will only be recorded in the bd_inode's mapping. When the blockdev contains a filesystem however, we'd like to also record the error in the super_block that's stored there.

Make mark_buffer_write_io_error also record the error in the corresponding super_block when a writeback error occurs and the block device contains a mounted superblock.

Since superblocks are RCU freed, hold the rcu_read_lock to ensure that the superblock doesn't go away while we're marking it.

---
 fs/buffer.c | 7 +++++++
 1 file changed, 7 insertions(+)
fs/buffer.c +0 -0
index f73276d746bb..2a4a5cc20418 100644
@@ -1154,12 +1154,19 @@ EXPORT_SYMBOL(mark_buffer_dirty);
 
 void mark_buffer_write_io_error(struct buffer_head *bh)
 {

fs/buffer.c EXPORT_SYMBOL(mark_buffer_dirty); · line 1166

Quoted 10 lines
fs/buffer.c +4 -0
+	struct super_block *sb;
+
 	set_buffer_write_io_error(bh);
 	/* FIXME: do we need to set this in both places? */
 	if (bh->b_page && bh->b_page->mapping)
 		mapping_set_error(bh->b_page->mapping, -EIO);
 	if (bh->b_assoc_map)
 		mapping_set_error(bh->b_assoc_map, -EIO);
+	rcu_read_lock();
+	sb = bh->b_bdev->bd_super;

You still need READ_ONCE() here. Otherwise the dereference below can still result in refetch and NULL ptr deref.

Quoted 7 lines
Quoted 2 lines

Honza

Huh? That seems like a really suspicious thing for the compiler/arch to do. We are checking that sb isn't NULL before we dereference it. Doesn't that imply a data dependency? How could the value of "sb" change after that?

Because the compiler is free to optimize the local variable away and actually compile the dereference below as bh->b_bdev->bd_super->s_wb_err (from C11 standard POV such code is equivalent since in C11 memory model it is assumed there are no concurrent accesses). And READ_ONCE() is a way to forbid compiler from doing such optimization - through 'volatile' keyword it tells the compiler there may be concurrent accesses happening and makes sure the value is really fetched into the local variable and used from there. There are good articles about this on LWN - I'd give you a link but LWN seems to be down today. But the latest article is about KCSAN and from there are links to older articles about compiler optimizations.

Quoted 2 lines

I'm also not sure I understand how using READ_ONCE really helps there if we can't count on the value of a local variable not changing.

I hope I've explained this above.

Honza

Quoted 13 lines
Quoted 9 lines
Quoted 9 lines
fs/buffer.c +3 -0
+	if (sb)
+		errseq_set(&sb->s_wb_err, -EIO);
+	rcu_read_unlock();
 }
 EXPORT_SYMBOL(mark_buffer_write_io_error);
 
Signature · 1 line
2.25.2
Signature · 1 line
Jeff Layton <[email protected]>
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
#6 lore raw

Re: [PATCH v5 2/2] buffer: record blockdev write errors in super_block that it backs

Quoted 63 lines · Jan Kara
Quoted 43 lines · Jeff Layton
Quoted 42 lines · Jan Kara
Quoted 27 lines · Jeff Layton

From: Jeff Layton <[email protected]>

When syncing out a block device (a'la __sync_blockdev), any error encountered will only be recorded in the bd_inode's mapping. When the blockdev contains a filesystem however, we'd like to also record the error in the super_block that's stored there.

Make mark_buffer_write_io_error also record the error in the corresponding super_block when a writeback error occurs and the block device contains a mounted superblock.

Since superblocks are RCU freed, hold the rcu_read_lock to ensure that the superblock doesn't go away while we're marking it.

---
 fs/buffer.c | 7 +++++++
 1 file changed, 7 insertions(+)
fs/buffer.c +0 -0
index f73276d746bb..2a4a5cc20418 100644
@@ -1154,12 +1154,19 @@ EXPORT_SYMBOL(mark_buffer_dirty);
 
 void mark_buffer_write_io_error(struct buffer_head *bh)
 {

fs/buffer.c EXPORT_SYMBOL(mark_buffer_dirty); · line 1166

Quoted 10 lines
fs/buffer.c +4 -0
+	struct super_block *sb;
+
 	set_buffer_write_io_error(bh);
 	/* FIXME: do we need to set this in both places? */
 	if (bh->b_page && bh->b_page->mapping)
 		mapping_set_error(bh->b_page->mapping, -EIO);
 	if (bh->b_assoc_map)
 		mapping_set_error(bh->b_assoc_map, -EIO);
+	rcu_read_lock();
+	sb = bh->b_bdev->bd_super;

You still need READ_ONCE() here. Otherwise the dereference below can still result in refetch and NULL ptr deref.

Quoted 7 lines
Quoted 2 lines

Honza

Huh? That seems like a really suspicious thing for the compiler/arch to do. We are checking that sb isn't NULL before we dereference it. Doesn't that imply a data dependency? How could the value of "sb" change after that?

Because the compiler is free to optimize the local variable away and actually compile the dereference below as bh->b_bdev->bd_super->s_wb_err (from C11 standard POV such code is equivalent since in C11 memory model it is assumed there are no concurrent accesses). And READ_ONCE() is a way to forbid compiler from doing such optimization - through 'volatile' keyword it tells the compiler there may be concurrent accesses happening and makes sure the value is really fetched into the local variable and used from there. There are good articles about this on LWN - I'd give you a link but LWN seems to be down today. But the latest article is about KCSAN and from there are links to older articles about compiler optimizations.

Quoted 5 lines
Quoted 2 lines

I'm also not sure I understand how using READ_ONCE really helps there if we can't count on the value of a local variable not changing.

I hope I've explained this above.

Got it. Thanks for the explanation. Now I'll have nightmares about all of the race conditions I've created in the past by making this assumption!

I'll send a v6 set in a few mins.

Quoted 13 lines
Quoted 13 lines
Quoted 9 lines
Quoted 9 lines
fs/buffer.c +3 -0
+	if (sb)
+		errseq_set(&sb->s_wb_err, -EIO);
+	rcu_read_unlock();
 }
 EXPORT_SYMBOL(mark_buffer_write_io_error);
 
Signature · 1 line
2.25.2
Signature · 1 line
Jeff Layton <[email protected]>
Signature · 1 line
Jeff Layton <[email protected]>