[PATCH v5 0/5] Improve simple directory offset wrap behavior

[PATCH v5 0/5] Improve simple directory offset wrap behavior

[cel]

From: Chuck Lever <chuck.lever@oracle.com>

The purpose of this series is to construct a set of upstream fixes
that can be backported to v6.6 to address CVE-2024-46701.

Changes since v4:
- Refactor and rename find_next_sibling_locked()

This series (against v6.13-rc2) has been pushed to:

https://git.kernel.org/pub/scm/linux/kernel/git/cel/linux.git/log/?h=tmpfs-fixes

Chuck Lever (5):
  libfs: Return ENOSPC when the directory offset range is exhausted
  Revert "libfs: Add simple_offset_empty()"
  Revert "libfs: fix infinite directory reads for offset dir"
  libfs: Replace simple_offset end-of-directory detection
  libfs: Use d_children list to iterate simple_offset directories

 fs/libfs.c         | 163 ++++++++++++++++++++++-----------------------
 include/linux/fs.h |   1 -
 mm/shmem.c         |   4 +-
 3 files changed, 81 insertions(+), 87 deletions(-)

-- 
2.47.0

[PATCH v5 1/5] libfs: Return ENOSPC when the directory offset range is exhausted

[cel]

From: Chuck Lever <chuck.lever@oracle.com>

Testing shows that the EBUSY error return from mtree_alloc_cyclic()
leaks into user space. The ERRORS section of "man creat(2)" says:

>	EBUSY	O_EXCL was specified in flags and pathname refers
>		to a block device that is in use by the system
>		(e.g., it is mounted).

ENOSPC is closer to what applications expect in this situation.

Note that the normal range of simple directory offset values is
2..2^63, so hitting this error is going to be rare to impossible.

Fixes: 6faddda69f62 ("libfs: Add directory operations for stable offsets")
Cc: <stable@vger.kernel.org> # v6.9+
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: Yang Erkun <yangerkun@huawei.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---
 fs/libfs.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/fs/libfs.c b/fs/libfs.c
index 748ac5923154..f6d04c69f195 100644
--- a/fs/libfs.c
+++ b/fs/libfs.c
@@ -292,7 +292,9 @@ int simple_offset_add(struct offset_ctx *octx, struct dentry *dentry)
 
 	ret = mtree_alloc_cyclic(&octx->mt, &offset, dentry, DIR_OFFSET_MIN,
 				 LONG_MAX, &octx->next_offset, GFP_KERNEL);
-	if (ret < 0)
+	if (unlikely(ret == -EBUSY))
+		return -ENOSPC;
+	if (unlikely(ret < 0))
 		return ret;
 
 	offset_set(dentry, offset);
-- 
2.47.0

RE: [PATCH v5 1/5] libfs: Return ENOSPC when the directory offset range is exhausted

[David Laight]

From: cel@kernel.org
> Sent: 15 December 2024 18:58
> 
> From: Chuck Lever <chuck.lever@oracle.com>
> 
> Testing shows that the EBUSY error return from mtree_alloc_cyclic()
> leaks into user space. The ERRORS section of "man creat(2)" says:
> 
> >	EBUSY	O_EXCL was specified in flags and pathname refers
> >		to a block device that is in use by the system
> >		(e.g., it is mounted).
> 
> ENOSPC is closer to what applications expect in this situation.
> 
> Note that the normal range of simple directory offset values is
> 2..2^63, so hitting this error is going to be rare to impossible.
> 
> Fixes: 6faddda69f62 ("libfs: Add directory operations for stable offsets")
> Cc: <stable@vger.kernel.org> # v6.9+
> Reviewed-by: Jeff Layton <jlayton@kernel.org>
> Reviewed-by: Yang Erkun <yangerkun@huawei.com>
> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
> ---
>  fs/libfs.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/fs/libfs.c b/fs/libfs.c
> index 748ac5923154..f6d04c69f195 100644
> --- a/fs/libfs.c
> +++ b/fs/libfs.c
> @@ -292,7 +292,9 @@ int simple_offset_add(struct offset_ctx *octx, struct dentry *dentry)
> 
>  	ret = mtree_alloc_cyclic(&octx->mt, &offset, dentry, DIR_OFFSET_MIN,
>  				 LONG_MAX, &octx->next_offset, GFP_KERNEL);
> -	if (ret < 0)
> +	if (unlikely(ret == -EBUSY))
> +		return -ENOSPC;
> +	if (unlikely(ret < 0))
>  		return ret;

You've just added an extra comparison to a hot path.
Doing:
	if (ret < 0)
		return ret == -EBUSY ? -ENOSPC : ret;
would be better.

	David

> 
>  	offset_set(dentry, offset);
> --
> 2.47.0
> 

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)

Re: [PATCH v5 1/5] libfs: Return ENOSPC when the directory offset range is exhausted

[Pratyush Yadav]

On Sun, Dec 15 2024, David Laight wrote:

> From: cel@kernel.org
>> Sent: 15 December 2024 18:58
>> 
>> From: Chuck Lever <chuck.lever@oracle.com>
>> 
>> Testing shows that the EBUSY error return from mtree_alloc_cyclic()
>> leaks into user space. The ERRORS section of "man creat(2)" says:
>> 
>> >	EBUSY	O_EXCL was specified in flags and pathname refers
>> >		to a block device that is in use by the system
>> >		(e.g., it is mounted).
>> 
>> ENOSPC is closer to what applications expect in this situation.
>> 
>> Note that the normal range of simple directory offset values is
>> 2..2^63, so hitting this error is going to be rare to impossible.
>> 
>> Fixes: 6faddda69f62 ("libfs: Add directory operations for stable offsets")
>> Cc: <stable@vger.kernel.org> # v6.9+
>> Reviewed-by: Jeff Layton <jlayton@kernel.org>
>> Reviewed-by: Yang Erkun <yangerkun@huawei.com>
>> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
>> ---
>>  fs/libfs.c | 4 +++-
>>  1 file changed, 3 insertions(+), 1 deletion(-)
>> 
>> diff --git a/fs/libfs.c b/fs/libfs.c
>> index 748ac5923154..f6d04c69f195 100644
>> --- a/fs/libfs.c
>> +++ b/fs/libfs.c
>> @@ -292,7 +292,9 @@ int simple_offset_add(struct offset_ctx *octx, struct dentry *dentry)
>> 
>>  	ret = mtree_alloc_cyclic(&octx->mt, &offset, dentry, DIR_OFFSET_MIN,
>>  				 LONG_MAX, &octx->next_offset, GFP_KERNEL);
>> -	if (ret < 0)
>> +	if (unlikely(ret == -EBUSY))
>> +		return -ENOSPC;
>> +	if (unlikely(ret < 0))
>>  		return ret;
>
> You've just added an extra comparison to a hot path.
> Doing:
> 	if (ret < 0)
> 		return ret == -EBUSY ? -ENOSPC : ret;
> would be better.

This also has two comparisons: one for ret < 0 and another for ret ==
-EBUSY. So I don't see a difference. I was curious to see if compilers
can somehow optimize one or the other, so I ran the two on godbolt and I
see no real difference between the two: https://godbolt.org/z/9Gav6b6Mf

-- 
Regards,
Pratyush Yadav

Re: [PATCH v5 1/5] libfs: Return ENOSPC when the directory offset range is exhausted

[Chuck Lever]

On 12/16/24 8:39 AM, Pratyush Yadav wrote:
> On Sun, Dec 15 2024, David Laight wrote:
> 
>> From: cel@kernel.org
>>> Sent: 15 December 2024 18:58
>>>
>>> From: Chuck Lever <chuck.lever@oracle.com>
>>>
>>> Testing shows that the EBUSY error return from mtree_alloc_cyclic()
>>> leaks into user space. The ERRORS section of "man creat(2)" says:
>>>
>>>> 	EBUSY	O_EXCL was specified in flags and pathname refers
>>>> 		to a block device that is in use by the system
>>>> 		(e.g., it is mounted).
>>>
>>> ENOSPC is closer to what applications expect in this situation.
>>>
>>> Note that the normal range of simple directory offset values is
>>> 2..2^63, so hitting this error is going to be rare to impossible.
>>>
>>> Fixes: 6faddda69f62 ("libfs: Add directory operations for stable offsets")
>>> Cc: <stable@vger.kernel.org> # v6.9+
>>> Reviewed-by: Jeff Layton <jlayton@kernel.org>
>>> Reviewed-by: Yang Erkun <yangerkun@huawei.com>
>>> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
>>> ---
>>>   fs/libfs.c | 4 +++-
>>>   1 file changed, 3 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/fs/libfs.c b/fs/libfs.c
>>> index 748ac5923154..f6d04c69f195 100644
>>> --- a/fs/libfs.c
>>> +++ b/fs/libfs.c
>>> @@ -292,7 +292,9 @@ int simple_offset_add(struct offset_ctx *octx, struct dentry *dentry)
>>>
>>>   	ret = mtree_alloc_cyclic(&octx->mt, &offset, dentry, DIR_OFFSET_MIN,
>>>   				 LONG_MAX, &octx->next_offset, GFP_KERNEL);
>>> -	if (ret < 0)
>>> +	if (unlikely(ret == -EBUSY))
>>> +		return -ENOSPC;
>>> +	if (unlikely(ret < 0))
>>>   		return ret;
>>
>> You've just added an extra comparison to a hot path.
>> Doing:
>> 	if (ret < 0)
>> 		return ret == -EBUSY ? -ENOSPC : ret;
>> would be better.
> 
> This also has two comparisons: one for ret < 0 and another for ret ==
> -EBUSY. So I don't see a difference. I was curious to see if compilers
> can somehow optimize one or the other, so I ran the two on godbolt and I
> see no real difference between the two: https://godbolt.org/z/9Gav6b6Mf

In my version, both comparisons are done every time through this flow.
David's version changes it so that only one comparison is done unless
@ret is less than zero (which is rare).

I've updated simple_offset_add() in my tree to use David's version.


-- 
Chuck Lever

[PATCH v5 2/5] Revert "libfs: Add simple_offset_empty()"

[cel]

From: Chuck Lever <chuck.lever@oracle.com>

simple_empty() and simple_offset_empty() perform the same task.
The latter's use as a canary to find bugs has not found any new
issues. A subsequent patch will remove the use of the mtree for
iterating directory contents, so revert back to using a similar
mechanism for determining whether a directory is indeed empty.

Only one such mechanism is ever needed.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---
 fs/libfs.c         | 32 --------------------------------
 include/linux/fs.h |  1 -
 mm/shmem.c         |  4 ++--
 3 files changed, 2 insertions(+), 35 deletions(-)

diff --git a/fs/libfs.c b/fs/libfs.c
index f6d04c69f195..9bcc97ffae48 100644
--- a/fs/libfs.c
+++ b/fs/libfs.c
@@ -331,38 +331,6 @@ void simple_offset_remove(struct offset_ctx *octx, struct dentry *dentry)
 	offset_set(dentry, 0);
 }
 
-/**
- * simple_offset_empty - Check if a dentry can be unlinked
- * @dentry: dentry to be tested
- *
- * Returns 0 if @dentry is a non-empty directory; otherwise returns 1.
- */
-int simple_offset_empty(struct dentry *dentry)
-{
-	struct inode *inode = d_inode(dentry);
-	struct offset_ctx *octx;
-	struct dentry *child;
-	unsigned long index;
-	int ret = 1;
-
-	if (!inode || !S_ISDIR(inode->i_mode))
-		return ret;
-
-	index = DIR_OFFSET_MIN;
-	octx = inode->i_op->get_offset_ctx(inode);
-	mt_for_each(&octx->mt, child, index, LONG_MAX) {
-		spin_lock(&child->d_lock);
-		if (simple_positive(child)) {
-			spin_unlock(&child->d_lock);
-			ret = 0;
-			break;
-		}
-		spin_unlock(&child->d_lock);
-	}
-
-	return ret;
-}
-
 /**
  * simple_offset_rename - handle directory offsets for rename
  * @old_dir: parent directory of source entry
diff --git a/include/linux/fs.h b/include/linux/fs.h
index 7e29433c5ecc..f7efc6866ebc 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -3468,7 +3468,6 @@ struct offset_ctx {
 void simple_offset_init(struct offset_ctx *octx);
 int simple_offset_add(struct offset_ctx *octx, struct dentry *dentry);
 void simple_offset_remove(struct offset_ctx *octx, struct dentry *dentry);
-int simple_offset_empty(struct dentry *dentry);
 int simple_offset_rename(struct inode *old_dir, struct dentry *old_dentry,
 			 struct inode *new_dir, struct dentry *new_dentry);
 int simple_offset_rename_exchange(struct inode *old_dir,
diff --git a/mm/shmem.c b/mm/shmem.c
index ccb9629a0f70..274c2666f457 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -3818,7 +3818,7 @@ static int shmem_unlink(struct inode *dir, struct dentry *dentry)
 
 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
 {
-	if (!simple_offset_empty(dentry))
+	if (!simple_empty(dentry))
 		return -ENOTEMPTY;
 
 	drop_nlink(d_inode(dentry));
@@ -3875,7 +3875,7 @@ static int shmem_rename2(struct mnt_idmap *idmap,
 		return simple_offset_rename_exchange(old_dir, old_dentry,
 						     new_dir, new_dentry);
 
-	if (!simple_offset_empty(new_dentry))
+	if (!simple_empty(new_dentry))
 		return -ENOTEMPTY;
 
 	if (flags & RENAME_WHITEOUT) {
-- 
2.47.0

[PATCH v5 3/5] Revert "libfs: fix infinite directory reads for offset dir"

[cel]

From: Chuck Lever <chuck.lever@oracle.com>

The current directory offset allocator (based on mtree_alloc_cyclic)
stores the next offset value to return in octx->next_offset. This
mechanism typically returns values that increase monotonically over
time. Eventually, though, the newly allocated offset value wraps
back to a low number (say, 2) which is smaller than other already-
allocated offset values.

Yu Kuai <yukuai3@huawei.com> reports that, after commit 64a7ce76fb90
("libfs: fix infinite directory reads for offset dir"), if a
directory's offset allocator wraps, existing entries are no longer
visible via readdir/getdents because offset_readdir() stops listing
entries once an entry's offset is larger than octx->next_offset.
These entries vanish persistently -- they can be looked up, but will
never again appear in readdir(3) output.

The reason for this is that the commit treats directory offsets as
monotonically increasing integer values rather than opaque cookies,
and introduces this comparison:

	if (dentry2offset(dentry) >= last_index) {

On 64-bit platforms, the directory offset value upper bound is
2^63 - 1. Directory offsets will monotonically increase for millions
of years without wrapping.

On 32-bit platforms, however, LONG_MAX is 2^31 - 1. The allocator
can wrap after only a few weeks (at worst).

Revert commit 64a7ce76fb90 ("libfs: fix infinite directory reads for
offset dir") to prepare for a fix that can work properly on 32-bit
systems and might apply to recent LTS kernels where shmem employs
the simple_offset mechanism.

Reported-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---
 fs/libfs.c | 35 +++++++++++------------------------
 1 file changed, 11 insertions(+), 24 deletions(-)

diff --git a/fs/libfs.c b/fs/libfs.c
index 9bcc97ffae48..fdf6ba305112 100644
--- a/fs/libfs.c
+++ b/fs/libfs.c
@@ -424,14 +424,6 @@ void simple_offset_destroy(struct offset_ctx *octx)
 	mtree_destroy(&octx->mt);
 }
 
-static int offset_dir_open(struct inode *inode, struct file *file)
-{
-	struct offset_ctx *ctx = inode->i_op->get_offset_ctx(inode);
-
-	file->private_data = (void *)ctx->next_offset;
-	return 0;
-}
-
 /**
  * offset_dir_llseek - Advance the read position of a directory descriptor
  * @file: an open directory whose position is to be updated
@@ -445,9 +437,6 @@ static int offset_dir_open(struct inode *inode, struct file *file)
  */
 static loff_t offset_dir_llseek(struct file *file, loff_t offset, int whence)
 {
-	struct inode *inode = file->f_inode;
-	struct offset_ctx *ctx = inode->i_op->get_offset_ctx(inode);
-
 	switch (whence) {
 	case SEEK_CUR:
 		offset += file->f_pos;
@@ -461,8 +450,7 @@ static loff_t offset_dir_llseek(struct file *file, loff_t offset, int whence)
 	}
 
 	/* In this case, ->private_data is protected by f_pos_lock */
-	if (!offset)
-		file->private_data = (void *)ctx->next_offset;
+	file->private_data = NULL;
 	return vfs_setpos(file, offset, LONG_MAX);
 }
 
@@ -493,7 +481,7 @@ static bool offset_dir_emit(struct dir_context *ctx, struct dentry *dentry)
 			  inode->i_ino, fs_umode_to_dtype(inode->i_mode));
 }
 
-static void offset_iterate_dir(struct inode *inode, struct dir_context *ctx, long last_index)
+static void *offset_iterate_dir(struct inode *inode, struct dir_context *ctx)
 {
 	struct offset_ctx *octx = inode->i_op->get_offset_ctx(inode);
 	struct dentry *dentry;
@@ -501,21 +489,17 @@ static void offset_iterate_dir(struct inode *inode, struct dir_context *ctx, lon
 	while (true) {
 		dentry = offset_find_next(octx, ctx->pos);
 		if (!dentry)
-			return;
-
-		if (dentry2offset(dentry) >= last_index) {
-			dput(dentry);
-			return;
-		}
+			return ERR_PTR(-ENOENT);
 
 		if (!offset_dir_emit(ctx, dentry)) {
 			dput(dentry);
-			return;
+			break;
 		}
 
 		ctx->pos = dentry2offset(dentry) + 1;
 		dput(dentry);
 	}
+	return NULL;
 }
 
 /**
@@ -542,19 +526,22 @@ static void offset_iterate_dir(struct inode *inode, struct dir_context *ctx, lon
 static int offset_readdir(struct file *file, struct dir_context *ctx)
 {
 	struct dentry *dir = file->f_path.dentry;
-	long last_index = (long)file->private_data;
 
 	lockdep_assert_held(&d_inode(dir)->i_rwsem);
 
 	if (!dir_emit_dots(file, ctx))
 		return 0;
 
-	offset_iterate_dir(d_inode(dir), ctx, last_index);
+	/* In this case, ->private_data is protected by f_pos_lock */
+	if (ctx->pos == DIR_OFFSET_MIN)
+		file->private_data = NULL;
+	else if (file->private_data == ERR_PTR(-ENOENT))
+		return 0;
+	file->private_data = offset_iterate_dir(d_inode(dir), ctx);
 	return 0;
 }
 
 const struct file_operations simple_offset_dir_operations = {
-	.open		= offset_dir_open,
 	.llseek		= offset_dir_llseek,
 	.iterate_shared	= offset_readdir,
 	.read		= generic_read_dir,
-- 
2.47.0

[PATCH v5 4/5] libfs: Replace simple_offset end-of-directory detection

[cel]

From: Chuck Lever <chuck.lever@oracle.com>

According to getdents(3), the d_off field in each returned directory
entry points to the next entry in the directory. The d_off field in
the last returned entry in the readdir buffer must contain a valid
offset value, but if it points to an actual directory entry, then
readdir/getdents can loop.

This patch introduces a specific fixed offset value that is placed
in the d_off field of the last entry in a directory. Some user space
applications assume that the EOD offset value is larger than the
offsets of real directory entries, so the largest possible offset
value is reserved for this purpose. This new value is never
allocated by simple_offset_add().

When ->iterate_dir() returns, getdents{64} inserts the ctx->pos
value into the d_off field of the last valid entry in the readdir
buffer. When it hits EOD, offset_readdir() sets ctx->pos to the EOD
offset value so the last entry is updated to point to the EOD marker.

When trying to read the entry at the EOD offset, offset_readdir()
terminates immediately.

It is worth noting that using a Maple tree for directory offset
value allocation does not guarantee a 63-bit range of values --
on platforms where "long" is a 32-bit type, the directory offset
value range is still 0..(2^31 - 1).

Fixes: 796432efab1e ("libfs: getdents() should return 0 after reaching EOD")
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---
 fs/libfs.c | 38 ++++++++++++++++++++++----------------
 1 file changed, 22 insertions(+), 16 deletions(-)

diff --git a/fs/libfs.c b/fs/libfs.c
index fdf6ba305112..00dfcfa97edf 100644
--- a/fs/libfs.c
+++ b/fs/libfs.c
@@ -245,9 +245,16 @@ const struct inode_operations simple_dir_inode_operations = {
 };
 EXPORT_SYMBOL(simple_dir_inode_operations);
 
-/* 0 is '.', 1 is '..', so always start with offset 2 or more */
+/* simple_offset_add() allocation range */
 enum {
-	DIR_OFFSET_MIN	= 2,
+	DIR_OFFSET_MIN		= 2,
+	DIR_OFFSET_MAX		= LONG_MAX - 1,
+};
+
+/* simple_offset_add() never assigns these to a dentry */
+enum {
+	DIR_OFFSET_EOD		= LONG_MAX,	/* Marks EOD */
+
 };
 
 static void offset_set(struct dentry *dentry, long offset)
@@ -291,7 +298,8 @@ int simple_offset_add(struct offset_ctx *octx, struct dentry *dentry)
 		return -EBUSY;
 
 	ret = mtree_alloc_cyclic(&octx->mt, &offset, dentry, DIR_OFFSET_MIN,
-				 LONG_MAX, &octx->next_offset, GFP_KERNEL);
+				 DIR_OFFSET_MAX, &octx->next_offset,
+				 GFP_KERNEL);
 	if (unlikely(ret == -EBUSY))
 		return -ENOSPC;
 	if (unlikely(ret < 0))
@@ -449,8 +457,6 @@ static loff_t offset_dir_llseek(struct file *file, loff_t offset, int whence)
 		return -EINVAL;
 	}
 
-	/* In this case, ->private_data is protected by f_pos_lock */
-	file->private_data = NULL;
 	return vfs_setpos(file, offset, LONG_MAX);
 }
 
@@ -460,7 +466,7 @@ static struct dentry *offset_find_next(struct offset_ctx *octx, loff_t offset)
 	struct dentry *child, *found = NULL;
 
 	rcu_read_lock();
-	child = mas_find(&mas, LONG_MAX);
+	child = mas_find(&mas, DIR_OFFSET_MAX);
 	if (!child)
 		goto out;
 	spin_lock(&child->d_lock);
@@ -481,7 +487,7 @@ static bool offset_dir_emit(struct dir_context *ctx, struct dentry *dentry)
 			  inode->i_ino, fs_umode_to_dtype(inode->i_mode));
 }
 
-static void *offset_iterate_dir(struct inode *inode, struct dir_context *ctx)
+static void offset_iterate_dir(struct inode *inode, struct dir_context *ctx)
 {
 	struct offset_ctx *octx = inode->i_op->get_offset_ctx(inode);
 	struct dentry *dentry;
@@ -489,7 +495,7 @@ static void *offset_iterate_dir(struct inode *inode, struct dir_context *ctx)
 	while (true) {
 		dentry = offset_find_next(octx, ctx->pos);
 		if (!dentry)
-			return ERR_PTR(-ENOENT);
+			goto out_eod;
 
 		if (!offset_dir_emit(ctx, dentry)) {
 			dput(dentry);
@@ -499,7 +505,10 @@ static void *offset_iterate_dir(struct inode *inode, struct dir_context *ctx)
 		ctx->pos = dentry2offset(dentry) + 1;
 		dput(dentry);
 	}
-	return NULL;
+	return;
+
+out_eod:
+	ctx->pos = DIR_OFFSET_EOD;
 }
 
 /**
@@ -519,6 +528,8 @@ static void *offset_iterate_dir(struct inode *inode, struct dir_context *ctx)
  *
  * On return, @ctx->pos contains an offset that will read the next entry
  * in this directory when offset_readdir() is called again with @ctx.
+ * Caller places this value in the d_off field of the last entry in the
+ * user's buffer.
  *
  * Return values:
  *   %0 - Complete
@@ -531,13 +542,8 @@ static int offset_readdir(struct file *file, struct dir_context *ctx)
 
 	if (!dir_emit_dots(file, ctx))
 		return 0;
-
-	/* In this case, ->private_data is protected by f_pos_lock */
-	if (ctx->pos == DIR_OFFSET_MIN)
-		file->private_data = NULL;
-	else if (file->private_data == ERR_PTR(-ENOENT))
-		return 0;
-	file->private_data = offset_iterate_dir(d_inode(dir), ctx);
+	if (ctx->pos != DIR_OFFSET_EOD)
+		offset_iterate_dir(d_inode(dir), ctx);
 	return 0;
 }
 
-- 
2.47.0

[PATCH v5 5/5] libfs: Use d_children list to iterate simple_offset directories

[cel]

From: Chuck Lever <chuck.lever@oracle.com>

The mtree mechanism has been effective at creating directory offsets
that are stable over multiple opendir instances. However, it has not
been able to handle the subtleties of renames that are concurrent
with readdir.

Instead of using the mtree to emit entries in the order of their
offset values, use it only to map incoming ctx->pos to a starting
entry. Then use the directory's d_children list, which is already
maintained properly by the dcache, to find the next child to emit.

One of the sneaky things about this is that when the mtree-allocated
offset value wraps (which is very rare), looking up ctx->pos++ is
not going to find the next entry; it will return NULL. Instead, by
following the d_children list, the offset values can appear in any
order but all of the entries in the directory will be visited
eventually.

Note also that the readdir() is guaranteed to reach the tail of this
list. Entries are added only at the head of d_children, and readdir
walks from its current position in that list towards its tail.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---
 fs/libfs.c | 84 +++++++++++++++++++++++++++++++++++++-----------------
 1 file changed, 58 insertions(+), 26 deletions(-)

diff --git a/fs/libfs.c b/fs/libfs.c
index 00dfcfa97edf..cc1226bf2b56 100644
--- a/fs/libfs.c
+++ b/fs/libfs.c
@@ -247,12 +247,13 @@ EXPORT_SYMBOL(simple_dir_inode_operations);
 
 /* simple_offset_add() allocation range */
 enum {
-	DIR_OFFSET_MIN		= 2,
+	DIR_OFFSET_MIN		= 3,
 	DIR_OFFSET_MAX		= LONG_MAX - 1,
 };
 
 /* simple_offset_add() never assigns these to a dentry */
 enum {
+	DIR_OFFSET_FIRST	= 2,		/* Find first real entry */
 	DIR_OFFSET_EOD		= LONG_MAX,	/* Marks EOD */
 
 };
@@ -460,51 +461,82 @@ static loff_t offset_dir_llseek(struct file *file, loff_t offset, int whence)
 	return vfs_setpos(file, offset, LONG_MAX);
 }
 
-static struct dentry *offset_find_next(struct offset_ctx *octx, loff_t offset)
+static struct dentry *find_positive_dentry(struct dentry *parent,
+					   struct dentry *dentry,
+					   bool next)
 {
-	MA_STATE(mas, &octx->mt, offset, offset);
+	struct dentry *found = NULL;
+
+	spin_lock(&parent->d_lock);
+	if (next)
+		dentry = d_next_sibling(dentry);
+	else if (!dentry)
+		dentry = d_first_child(parent);
+	hlist_for_each_entry_from(dentry, d_sib) {
+		if (!simple_positive(dentry))
+			continue;
+		spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
+		if (simple_positive(dentry))
+			found = dget_dlock(dentry);
+		spin_unlock(&dentry->d_lock);
+		if (likely(found))
+			break;
+	}
+	spin_unlock(&parent->d_lock);
+	return found;
+}
+
+static noinline_for_stack struct dentry *
+offset_dir_lookup(struct dentry *parent, loff_t offset)
+{
+	struct inode *inode = d_inode(parent);
+	struct offset_ctx *octx = inode->i_op->get_offset_ctx(inode);
 	struct dentry *child, *found = NULL;
 
-	rcu_read_lock();
-	child = mas_find(&mas, DIR_OFFSET_MAX);
-	if (!child)
-		goto out;
-	spin_lock(&child->d_lock);
-	if (simple_positive(child))
-		found = dget_dlock(child);
-	spin_unlock(&child->d_lock);
-out:
-	rcu_read_unlock();
+	MA_STATE(mas, &octx->mt, offset, offset);
+
+	if (offset == DIR_OFFSET_FIRST)
+		found = find_positive_dentry(parent, NULL, false);
+	else {
+		rcu_read_lock();
+		child = mas_find(&mas, DIR_OFFSET_MAX);
+		found = find_positive_dentry(parent, child, false);
+		rcu_read_unlock();
+	}
 	return found;
 }
 
 static bool offset_dir_emit(struct dir_context *ctx, struct dentry *dentry)
 {
 	struct inode *inode = d_inode(dentry);
-	long offset = dentry2offset(dentry);
 
-	return ctx->actor(ctx, dentry->d_name.name, dentry->d_name.len, offset,
-			  inode->i_ino, fs_umode_to_dtype(inode->i_mode));
+	return dir_emit(ctx, dentry->d_name.name, dentry->d_name.len,
+			inode->i_ino, fs_umode_to_dtype(inode->i_mode));
 }
 
-static void offset_iterate_dir(struct inode *inode, struct dir_context *ctx)
+static void offset_iterate_dir(struct file *file, struct dir_context *ctx)
 {
-	struct offset_ctx *octx = inode->i_op->get_offset_ctx(inode);
+	struct dentry *dir = file->f_path.dentry;
 	struct dentry *dentry;
 
+	dentry = offset_dir_lookup(dir, ctx->pos);
+	if (!dentry)
+		goto out_eod;
 	while (true) {
-		dentry = offset_find_next(octx, ctx->pos);
-		if (!dentry)
-			goto out_eod;
+		struct dentry *next;
 
-		if (!offset_dir_emit(ctx, dentry)) {
-			dput(dentry);
+		ctx->pos = dentry2offset(dentry);
+		if (!offset_dir_emit(ctx, dentry))
 			break;
-		}
 
-		ctx->pos = dentry2offset(dentry) + 1;
+		next = find_positive_dentry(dir, dentry, true);
 		dput(dentry);
+
+		if (!next)
+			goto out_eod;
+		dentry = next;
 	}
+	dput(dentry);
 	return;
 
 out_eod:
@@ -543,7 +575,7 @@ static int offset_readdir(struct file *file, struct dir_context *ctx)
 	if (!dir_emit_dots(file, ctx))
 		return 0;
 	if (ctx->pos != DIR_OFFSET_EOD)
-		offset_iterate_dir(d_inode(dir), ctx);
+		offset_iterate_dir(file, ctx);
 	return 0;
 }
 
-- 
2.47.0