fixes in methods exposed to rcu pathwalk

RCU pathwalk requirements for methods hadn't been
documented well enough; in particular, the implications of races with fs shutdown need to be spelled out.

I tried to put together something that would resemble
such documentation (see below). Unfortunately, it turns out that we have some places that need fixing; all but two cases (ntfs3 and ceph) are hopefully dealt with in the patch series (in followups). I would obviously like to hear from fs maintainers involved - most of the fixes are trivial, but...

Please, review and comment (both the docs below and
the patches)

================================================================================

Exposure of filesystem code to lockless environment, or
the ways in which RCU pathwalk might make filesystem maintainer life painful.

Filesystem methods can usually count upon VFS-provided warranties
regarding the stability of objects they are called to act upon; at the very least, they can expect the dentries/inodes/superblocks involved to remain live throughout the operation.

Life would be much more painful without that; however, such
warranties do not come for free. The problem is that access patterns are heavily biased; every system call getting an absolute pathname will have to start at root directory, etc. Having each of them in effect write "I'd been here" on the same memory objects would cost quite a bit. As the result, we try to keep the fast path stores-free, bumping no
refcounts and taking no locks. Details are described elsewhere, but the
bottom line for filesystems is that some methods may be called with much looser warranties than usual. Of course, from the filesystem POV each of those is a potential source of headache - you are asked to operate on an object that might start to be torn down right under you, possibly along with the filesystem instance it lives on.
[[ footnote:

it *is* possible for a lazy pathwalk to run into a dentry on a filesystem that is getting shut down. Here's how:
* have two threads sharing fs_struct chdir'ed on that filesystem.
* lazy-umount it, so that the only thing holding it alive is
cwd of these threads.
* one of the threads does relative pathname resolution
and gets to e.g. ->d_hash(). It's holding rcu_read_lock().
* at the same time another thread does fchdir(), moving to
different directory. set_fs_pwd() flips to the new place and does mntput() on the old one. No RCU delays here, we calculate the refcount of that mount and see that we are dropping the last reference. We make sure that the first thread will fail when it comes to legitimize_mnt() and do this:

			init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
			if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME))
				return;

As we leave the syscall, we have __cleanup_mnt() run; it calls cleanup_mnt() on our mount, which hits deactivate_super(). That was the last reference to superblock.

Voila - we have a filesystem shutdown right under the nose of
a thread running in ->d_hash() of something on that filesystem.
Mutatis mutandis, one can arrange the same for other methods
called by rcu pathwalk.

It's not easy to hit (especially if you want to get through the
entire ->kill_sb() before the first thread gets through ->d_hash()), and it's probably impossible on the real hardware; on KVM it might be borderline doable. However, it is possible and I would not swear that other ways of arranging the same thing are equally hard to hit.

The bottom line: methods that can be called in RCU mode need to
be careful about the per-superblock objects destruction.
]]

The list of the methods that could run into that fun:

method | indication that the call is unsafe | unstable objects
->d_hash(d, ...) | none - any call might be | d
->d_compare(d, ...) | none - any call might be | d
->d_revalidate(d, f) | f & LOOKUP_RCU | d
->d_manage(d, f) | f | d
->permission(i, m) | m & MAY_NOT_BLOCK | i
->get_link(d, i, ...) | d == NULL | i
->get_inode_acl(i, t, f)| f == LOOKUP_RCU | i

Additionally, callback set by set_delayed_call() from unsafe call of
->get_link() will be run in the same environment; that one is usually not a problem, though.

For the sake of completeness, three of LSM methods
(->inode_permission(), ->inode_follow_link() and ->task_to_inode()) might be called in similar environment, but that's a problem for LSM crowd, not for filesystem folks.

Any method call is, of course, required not to crash - no stepping on
freed memory, etc. All of the unsafe calls listed above are done under rcu_read_lock(), so they are not allowed to block. Further requirements vary between the methods.

Before going through the list of affected methods, several notes on
the things that _are_ guaranteed:
* if a reference to struct dentry is passed to such call, it will
not be freed until the method returns. The same goes for a reference to
struct inode and to struct super_block pointed to by ->d_sb or ->i_sb members of dentry and inode resp. Any of those might be in process of being torn down or enter such state right under us; the entire point of those unsafe calls is that we make them without telling anyone they'd need to wait for us.
* following ->d_parent and ->d_inode of such dentries is fine,
provided that it's done by READ_ONCE() (for ->d_inode the preferred form is d_inode_rcu(dentry)). The value of ->d_parent is never going to be NULL and it will again point to a struct dentry that will not be freed until the method call finishes. The value of ->d_inode might be NULL; if non-NULL, it'll be pointing to a struct inode that will not be freed until the method call finishes.
* none of the inodes passed to an unsafe call could have reached
fs/inode.c:evict() before the caller grabbed rcu_read_lock().
* for inodes 'not freed' means 'not entered ->free_inode()', so
anything that won't be destroyed until ->free_inode() is safe to access. Anything synchronously destroyed in ->evict_inode() or ->destroy_inode() is not safe; however, one can count upon the call_rcu() callbacks issued in those yet to be entered. Note that unlike dentries and superblocks, inodes are embedded into filesystem-private objects; anything stored directly in the containing object is safe to access.
* for dentries anything destroyed by ->d_prune() (synchronously or
not) is not safe; the same goes for the things synchronously destroyed by ->d_release(). However, call_rcu() callbacks issued in ->d_release() are yet to be entered.
* for superblocks we can count upon call_rcu() callbacks issued
from inside the ->kill_sb() (including ->put_super()) yet to be entered.
* NOTE: we can not count upon the things like ->d_parent
being positive (or a directory); a race with rename()+rmdir()+mknod() and you might find a FIFO as parent's inode. NULL is even easier - just have the dentry *and* its ex-parent already past dentry_kill() (which is a normal situation for eviction on memory pressure) and there you go. Normally such pathologies are prevented by the locking (and dentry refcounting), but... the entire point of that stuff is to avoid informing anyone that we are there, so those mechanisms are bypassed. What's more, if dentry is not pinned by refcount, grabbing its ->d_lock will *not* suffice to prevent that kind of mess - the scenario with eviction by memory pressure won't be prevented by that; you might have grabbed ->d_lock only after the dentry_kill() had released it, and at that point ->d_parent still points to what used to be the parent, but there's nothing to prevent its eviction.

	1. ->d_compare().

	For ->d_compare() we just need to make sure it won't crash

when called for dying dentry - an incorrect return value won't harm the caller in such case. False positives and false negatives alike - the callers take care of that. To be pedantic, make that "false positives do not cause problems unless they have ->d_manage()", but ->d_manage() is present only on autofs and there's no autofs ->d_compare() instances.

[[ footnote:
Some callers prevent being called for dying dentry (holding
->d_lock and having verified !d_unhashed() or finding it in the list of inode's aliases under ->i_lock). For those the scenario in question simply cannot arise.
Some follow the match with lockref_get_not_dead() and treat
the failure as mismatch. That takes care of false positives, and false negatives on dying dentry are still correct - we simply pretend to have lost the race.
The only caller that does not fit into the classes above is
__d_lookup_rcu_op_compare(). There we sample ->d_seq and verify !d_unhashed() before calling ->d_compare(). That is not enough to prevent dentry from starting to die right under us; however, the sampled value of ->d_seq will be rechecked when the caller gets to step_into(), so for a false positive we will end up with a mismatch. The corner case around ->d_manage() is due to the handle_mounts() done before step_into() gets to ->d_seq validation...
]]

	There is no indication that ->d_compare() is called in RCU mode;

the majority of callers are such, anyway, so we need to cope with that. VFS guarantees that dentry won't be freed under us; the same goes for the superblock pointed to by its ->d_sb. Name points to memory object that won't get freed under us and length does not exceed the size of that object. The contents of that object is *NOT* guaranteed to be stable; d_move() might race with us, modifying the name. However, in that case we are free to return an arbitrary result - the callers will take care of both false positives and false negatives in such case. The name we are comparing dentry with (passed in qstr) is stable, thankfully...

If we need to access any other data, it's up to the filesystem
to protect it. In practice it means that destruction of fs-private part of superblock (and possibly unicode tables hanging off it, etc.) might need to be RCU-delayed.

*IF* you want the behaviour that varies depending upon the parent
directory, you get to be very careful with READ_ONCE() and watch out for the object lifetimes.

Basically, if the things get that tricky, ask for help.
Currently there are two such instances in the tree - proc_sys_compare() and generic_ci_d_compare(). Both are... special.

	2. ->d_hash().

	For ->d_hash() on a dying dentry we are free to report any hash

value; the only extra requirement is that we should not return stray hard errors. In other words, if we return anything other than 0 or -ECHILD, we'd better make sure that this error would've been correct before the parent started dying. Since ->d_hash() error-reporting is usually done to reject unacceptable names (too long, contain unsuitable characters for this filesystem, etc.), that's really not a problem - hard errors depend only upon the name, not the parent.

Again, VFS guarantees that freeing of dentry and of the superblock
pointed to by dentry->d_sb won't happen under us. The name passed to us (in qstr) is stable. If you need anything beyond that, you are in the same situation as with ->d_compare(). Might want to RCU-delay freeing private part of superblock (if that's what we need to access), might want the same for some objects hanging off that (unicode tables, etc.). If you need something beyond that - ask for help.

3. ->d_revalidate().

For this one we do have an indication of call being unsafe -
flags & LOOKUP_RCU. With ->d_revalidate we are always allowed to bail out and return -ECHILD; that will have the caller drop out of RCU mode. We definitely need to do that if revalidate would require any kind of IO, mutex-taking, etc.; we can't block in RCU mode.

	Quite a few instances of ->d_revalidate() simply treat LOOKUP_RCU

in flags as "return -ECHILD and be done with that"; it's guaranteed to do the right thing, but you lose the benefits of RCU pathwalks whenever you run into such dentry.

Same as with the previous methods, we are guaranteed that
dentry and dentry->d_sb won't be freed under us. We are also guaranteed that ->d_parent (which is *not* stable, so use READ_ONCE) points to a struct dentry that won't get freed under us. As always with ->d_parent, it's not NULL - for a detached dentry it will point to dentry itself. d_inode_rcu() of dentry and its parent will be either NULL or will
point to a struct inode that won't get freed under us. Anything beyond
than that is not guaranteed. We may find parent to be negative - it can happen if we race with d_move() and removal of old parent. In that case just return -ECHILD and be done with that.

On non-RCU side you could use dget_parent() instead - that
would give a positive dentry and its ->d_inode would remain stable. dget_parent() has to be paired with dput(), though, so it's not usable in RCU mode.

If you need fs-private objects associated with dentry, its parent
inode(s) or superblock - see the general notes above on how to access those.

4. ->d_manage()

Can be called in RCU mode; gets an argument telling it if it has
been called so. Pretty much autofs-only; for everyone's sanity sake, don't inflict more of those on the kernel. Definitely don't do that without asking first...

5. ->permission()

Can be called in RCU mode; that is indicated by MAY_NOT_BLOCK
in mask, and it can only happen for MAY_EXEC checks on directories. In RCU mode it is not allowed to block, and it is allowed to bail out by returning -ECHILD. It might be called for an inode that is getting
torn down, possibly along with its filesystem. Errors other than -ECHILD
should only be returned if they would've been returned in non-RCU mode; several instances in procfs currently (6.5) run afoul of that one. That's an instructive example, BTW - what happens is that proc_pid_permission() uses proc_get_task() to find the relevant process. proc_get_task() uses PID reference stored in struct proc_inode our inode is embedded into; inode can't have been freed yet, so fetching ->pid member in that is safe. However, using the value you've fetched is a different story
- proc_evict_inode() would have passed it to put_pid() and replaced it with NULL. Unsafe caller has no way to tell if that is happening right under it. Solution: stop zeroing ->pid in proc_evict_inode() and move put_pid() from proc_pid_evict_inode() to proc_free_inode(). That's not all that is needed (there's access to procfs-private part of superblock as well), but it does make a good example of how such stuff can be dealt with.

Note that idmap argument is safe on all calls - its destruction
is rcu-delayed.

The amount of headache is seriously reduced (for now) by the fact
that a lot of instances boil down to generic_permission() (which will do the right thing in RCU mode) when mask is MAY_EXEC | MAY_NOT_BLOCK. If we ever extend RCU mode to other ->permission() callers, the thing will get interesting; that's not likely to happen, though, unless access(2) goes there [this is NOT a suggestion, folks].

6. ->get_link()

Again, this can be called in RCU mode. Even if your
->d_revalidate() always returns -ECHILD in RCU mode and kicks the pathwalk out of it, you can't assume that ->get_link() won't be reached; binding a symlink on a regular file on another filesystem is possible
and that's all it takes for RCU pathwalk to get there. NULL dentry
argument is an indicator of unsafe call; if you can't handle it, just return ERR_PTR(-ECHILD). Any allocations you need to do (and with this method you really might need that) should be done with GFP_ATOMIC in the unsafe case.

Whatever you pass to set_delayed_call() is going to be called
in the same mode as ->get_link() itself; not a problem for most of the instances. The string you return needs to stay there until the callback gets called or, if no callback is set, until at least the freeing of inode. As usual, for an unsafe call the inode might be in process of teardown, possibly along with the hosting filesystem. The usual considerations apply. The same, BTW, applies to whatever you set in ->i_link - it must stay around at least until ->free_inode().

7. ->get_inode_acl()

Very limited exposure for that one - unsafe call is possible
only if you explicitly set ACL_DONT_CACHE as cached ACL value. Only two filesystems (fuse and overlayfs) even bother. Unsafe call is indicated by explicit flag (the third argument of the method), bailout is done by returning ERR_PTR(-CHILD) and the usual considerations apply for any access to data structures you might need to do.

Patchset summary follows; patches themselves - in followups.

1/15) rcu pathwalk: prevent bogus hard errors from may_lookup()
That's one in the core pathwalk logics; basically, treat
errors from RCU permission() calls as "trust hard errors only if the object is still there". More for the sake of ->permission() instances (similar to ->d_compare() et.al. the only thing they need to care about when called on inode that is getting torn down is not to crash; specific error value doesn't matter), but there's also a narrow race where RCU pathwalk yields EACCES while refwalk would fail with ENOTDIR.

2/15) exfat: move freeing sbi, upcase table and dropping nls into rcu-delayed helper

	Several UAF in their ->d_hash/->d_compare

3/15) affs: free affs_sb_info with kfree_rcu()

	UAF in ->d_hash/->d_compare

4/15) hfsplus: switch to rcu-delayed unloading of nls and freeing ->s_fs_info

	UAF and oops in ->d_hash/->d_compare

5/15) cifs_get_link(): bail out in unsafe case

	GFP_KERNEL allocation under rcu_read_lock in ->get_link()

6/15) procfs: move dropping pde and pid from ->evict_inode() to ->free_inode() 7/15) procfs: make freeing proc_fs_info rcu-delayed

	UAF in ->d_revalidate/->permission

8/15) gfs2: fix an oops in gfs2_permission()
What it says...

9/15) nfs: make nfs_set_verifier() safe for use in RCU pathwalk

	race and oops in nfs_set_verifier() when used in RCU mode.
	Short version of the story - holding ->d_lock does *not*
	guarantee that ->d_parent->d_inode is non-NULL, unless you
	have your dentry pinned.

10/15) nfs: fix UAF on pathwalk running into umount

	Several UAF in ->d_revalidate/->permission/->get_link

11/15) fuse: fix UAF in rcu pathwalks

	UAF in ->permission/->get_link; accesses ->s_fs_info->fc,
	and ->s_fs_info points to struct fuse_mount which gets
	freed synchronously by fuse_mount_destroy(), from the end of
	->kill_sb().  It proceeds to accessing ->fc, but that part
	is safe - ->fc freeing is done via kfree_rcu() (called via
	->fc->release()) after the refcount of ->fc drops to zero.
	That can't happen until the call of fuse_conn_put() (from
	fuse_mount_destroy() from ->kill_sb()), so anything rcu-delayed
	from there won't get freed until the end of rcu pathwalk.
		
	Unfortunately, we also dereference fc->user_ns (pass it
	to current_in_userns).  That gets dropped via put_user_ns()
	(non-rcu-delayed) from the final fuse_conn_put() and it
	needs to be delayed.

	Possible solution: make freeing in ->release() synchronous
	and do call_rcu(delayed_release, &fc->rcu), with
	delayed_release() doing put_user_ns() and calling
	->release().
	This one is relatively intrusive and I'd really like comments
	from Miklos...

12/15) afs: fix __afs_break_callback() / afs_drop_open_mmap() race
->d_revalidate/->permission manage to step on this one. We might end up doing __afs_break_callback() there,
and it could race with afs_drop_open_mmap(), leading to stray queued work on object that might be about to be freed, with nothing to flush or cancel the sucker.
To fix that race, make sure that afs_drop_open_mmap() will only do the final decrement while under ->cb_lock; since the entire __afs_break_callback() is done under the same, it will either see zero mmap count and do nothing, or it will finish queue_work() before afs_drop_open_mmap() gets to its flush_work().

13/15) overlayfs: move freeing ovl_entry past rcu delay

	Used to be done by kfree_rcu() from ->d_release(),
	got moved to ->destroy_inode() without any delays;
	move freeing to ->free_inode().

14/15) ovl_dentry_revalidate_common(): fetch inode once

	d_inode_rcu() is right - we might be in rcu pathwalk;
	however, OVL_E() hides plain d_inode() on the same dentry...

15/15) overlayfs: make use of ->layers safe in rcu pathwalk
ovl_permission() accesses ->layers[...].mnt; we can't have ->layers freed without an RCU delay on fs shutdown. Fortunately, kern_unmount_array() used to drop those mounts does include an RCU delay, so freeing is delayed; unfortunately, the array passed to kern_unmount_array() is formed by mangling ->layers contents and that happens without any delays.

Use a separate array instead; local if we have a few layers, kmalloc'ed if there's a lot of them. If allocation fails, fall back to kern_unmount() for individual mounts; it's not a fast path by any stretch of imagination.

If lazy call of ->permission() returns a hard error, check that try_to_unlazy() succeeds before returning it. That both makes life easier for ->permission() instances and closes the race in ENOTDIR handling - it is possible that positive d_can_lookup() seen in link_path_walk() applies to the state *after* unlink() + mkdir(), while nd->inode matches the state prior to that.

Normally seeing e.g. EACCES from permission check in rcu pathwalk means that with some timings non-rcu pathwalk would've run into the same; however, running into a non-executable regular file in the middle of a pathname would not get to permission check - it would fail with ENOTDIR instead.

---
 fs/namei.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

index 567ee547492b..2494561a21fe 100644
@@ -1717,7 +1717,11 @@ static inline int may_lookup(struct mnt_idmap *idmap,
 {
 	if (nd->flags & LOOKUP_RCU) {
 		int err = inode_permission(idmap, nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
-		if (err != -ECHILD || !try_to_unlazy(nd))
+		if (!err)		// success, keep going
+			return 0;
+		if (!try_to_unlazy(nd))
+			return -ECHILD;	// redo it all non-lazy
+		if (err != -ECHILD)	// hard error
 			return err;
 	}
 	return inode_permission(idmap, nd->inode, MAY_EXEC);

2.39.2

That stuff can be accessed by ->d_hash()/->d_compare(); as it is, we have a hard-to-hit UAF if rcu pathwalk manages to get into ->d_hash() on a filesystem that is in process of getting shut down.

Besides, having nls and upcase table cleanup moved from ->put_super() towards the place where sbi is freed makes for simpler failure exits.

---
 fs/exfat/exfat_fs.h |  1 +
 fs/exfat/nls.c      | 14 ++++----------
 fs/exfat/super.c    | 20 +++++++++++---------
 3 files changed, 16 insertions(+), 19 deletions(-)

index f55498e5c23d..22e17b0a66e8 100644
@@ -273,6 +273,7 @@ struct exfat_sb_info {
 
 	spinlock_t inode_hash_lock;
 	struct hlist_head inode_hashtable[EXFAT_HASH_SIZE];
+	struct rcu_head rcu;
 };
 
 #define EXFAT_CACHE_VALID	0

index 705710f93e2d..afdf13c34ff5 100644
@@ -655,7 +655,6 @@ static int exfat_load_upcase_table(struct super_block *sb,
 	unsigned int sect_size = sb->s_blocksize;
 	unsigned int i, index = 0;
 	u32 chksum = 0;
-	int ret;
 	unsigned char skip = false;
 	unsigned short *upcase_table;
 

@@ -673,8 +672,7 @@ static int exfat_load_upcase_table(struct super_block *sb,
 		if (!bh) {
 			exfat_err(sb, "failed to read sector(0x%llx)",
 				  (unsigned long long)sector);
-			ret = -EIO;
-			goto free_table;
+			return -EIO;
 		}
 		sector++;
 		for (i = 0; i < sect_size && index <= 0xFFFF; i += 2) {

@@ -701,15 +699,12 @@ static int exfat_load_upcase_table(struct super_block *sb,
 
 	exfat_err(sb, "failed to load upcase table (idx : 0x%08x, chksum : 0x%08x, utbl_chksum : 0x%08x)",
 		  index, chksum, utbl_checksum);
-	ret = -EINVAL;
-free_table:
-	exfat_free_upcase_table(sbi);
-	return ret;
+	return -EINVAL;
 }
 
 static int exfat_load_default_upcase_table(struct super_block *sb)
 {
-	int i, ret = -EIO;
+	int i;
 	struct exfat_sb_info *sbi = EXFAT_SB(sb);
 	unsigned char skip = false;
 	unsigned short uni = 0, *upcase_table;

@@ -740,8 +735,7 @@ static int exfat_load_default_upcase_table(struct super_block *sb)
 		return 0;
 
 	/* FATAL error: default upcase table has error */
-	exfat_free_upcase_table(sbi);
-	return ret;
+	return -EIO;
 }
 
 int exfat_create_upcase_table(struct super_block *sb)

index 2778bd9b631e..593cfff8c6f4 100644
@@ -39,9 +39,6 @@ static void exfat_put_super(struct super_block *sb)
 	exfat_free_bitmap(sbi);
 	brelse(sbi->boot_bh);
 	mutex_unlock(&sbi->s_lock);
-
-	unload_nls(sbi->nls_io);
-	exfat_free_upcase_table(sbi);
 }
 
 static int exfat_sync_fs(struct super_block *sb, int wait)

@@ -593,7 +590,7 @@ static int __exfat_fill_super(struct super_block *sb)
 	ret = exfat_load_bitmap(sb);
 	if (ret) {
 		exfat_err(sb, "failed to load alloc-bitmap");
-		goto free_upcase_table;
+		goto free_bh;
 	}
 
 	ret = exfat_count_used_clusters(sb, &sbi->used_clusters);

@@ -606,8 +603,6 @@ static int __exfat_fill_super(struct super_block *sb)
 
 free_alloc_bitmap:
 	exfat_free_bitmap(sbi);
-free_upcase_table:
-	exfat_free_upcase_table(sbi);
 free_bh:
 	brelse(sbi->boot_bh);
 	return ret;

@@ -694,12 +689,10 @@ static int exfat_fill_super(struct super_block *sb, struct fs_context *fc)
 	sb->s_root = NULL;
 
 free_table:
-	exfat_free_upcase_table(sbi);
 	exfat_free_bitmap(sbi);
 	brelse(sbi->boot_bh);
 
 check_nls_io:
-	unload_nls(sbi->nls_io);
 	return err;
 }
 

@@ -764,13 +757,22 @@ static int exfat_init_fs_context(struct fs_context *fc)
 	return 0;
 }
 
+static void delayed_free(struct rcu_head *p)
+{
+	struct exfat_sb_info *sbi = container_of(p, struct exfat_sb_info, rcu);
+
+	unload_nls(sbi->nls_io);
+	exfat_free_upcase_table(sbi);
+	exfat_free_sbi(sbi);
+}
+
 static void exfat_kill_sb(struct super_block *sb)
 {
 	struct exfat_sb_info *sbi = sb->s_fs_info;
 
 	kill_block_super(sb);
 	if (sbi)
-		exfat_free_sbi(sbi);
+		call_rcu(&sbi->rcu, delayed_free);
 }
 
 static struct file_system_type exfat_fs_type = {

2.39.2

I don't disagree with moving the freeing, but the RCU-delay makes me go "hmm".

Is there some reason why we can't try to do this in generic code? The umount code already does RCU delays for other things, I get the feeling that we should have a RCu delay between "put_super" and "kkill_sb".

Could we move the ->kill_sb() call into destroy_super_work(), which is already RCU-delayed, for example?

It feels wrong to have the filesystems have to deal with the vfs layer doing RCU-lookups.

Linus

	For one thing, ->kill_sb() might do tons of IO.  And we really want

to have that done before umount(2) returns to userland, so that part can't be offloaded via schedule_work()...

one of the flags in it is used by ->d_hash()/->d_compare()

---
 fs/affs/affs.h  | 1 +
 fs/affs/super.c | 2 +-
 2 files changed, 2 insertions(+), 1 deletion(-)

index 60685ec76d98..2e612834329a 100644
@@ -105,6 +105,7 @@ struct affs_sb_info {
 	int work_queued;		/* non-zero delayed work is queued */
 	struct delayed_work sb_work;	/* superblock flush delayed work */
 	spinlock_t work_lock;		/* protects sb_work and work_queued */
+	struct rcu_head rcu;
 };
 
 #define AFFS_MOUNT_SF_INTL		0x0001 /* International filesystem. */

index 58b391446ae1..b56a95cf414a 100644
@@ -640,7 +640,7 @@ static void affs_kill_sb(struct super_block *sb)
 		affs_brelse(sbi->s_root_bh);
 		kfree(sbi->s_prefix);
 		mutex_destroy(&sbi->s_bmlock);
-		kfree(sbi);
+		kfree_rcu(sbi, rcu);
 	}
 }
 

2.39.2

->d_hash() and ->d_compare() use those, so we need to delay freeing them.

---
 fs/hfsplus/hfsplus_fs.h |  1 +
 fs/hfsplus/super.c      | 12 +++++++++---
 2 files changed, 10 insertions(+), 3 deletions(-)

index 7ededcb720c1..012a3d003fbe 100644
@@ -190,6 +190,7 @@ struct hfsplus_sb_info {
 	int work_queued;               /* non-zero delayed work is queued */
 	struct delayed_work sync_work; /* FS sync delayed work */
 	spinlock_t work_lock;          /* protects sync_work and work_queued */
+	struct rcu_head rcu;
 };
 
 #define HFSPLUS_SB_WRITEBACKUP	0

index 1986b4f18a90..97920202790f 100644
@@ -277,6 +277,14 @@ void hfsplus_mark_mdb_dirty(struct super_block *sb)
 	spin_unlock(&sbi->work_lock);
 }
 
+static void delayed_free(struct rcu_head *p)
+{
+	struct hfsplus_sb_info *sbi = container_of(p, struct hfsplus_sb_info, rcu);
+
+	unload_nls(sbi->nls);
+	kfree(sbi);
+}
+
 static void hfsplus_put_super(struct super_block *sb)
 {
 	struct hfsplus_sb_info *sbi = HFSPLUS_SB(sb);

@@ -302,9 +310,7 @@ static void hfsplus_put_super(struct super_block *sb)
 	hfs_btree_close(sbi->ext_tree);
 	kfree(sbi->s_vhdr_buf);
 	kfree(sbi->s_backup_vhdr_buf);
-	unload_nls(sbi->nls);
-	kfree(sb->s_fs_info);
-	sb->s_fs_info = NULL;
+	call_rcu(&sbi->rcu, delayed_free);
 }
 
 static int hfsplus_statfs(struct dentry *dentry, struct kstatfs *buf)

2.39.2

Instead of all this duplicatio in the file system, can we please just add a

	struct nls_table *s_nls;

to struct super_block and RCU free it in common code and drop all the code in the file systems?

It makes no sense for most of the filesystems, for one thing (note that any use in ->lookup() does not warrant rcu delays). What's more, how do you formulate the rules for what goes in that field when filesystem uses more than one nls_table?

Consider e.g. HFS; two separate nls_table (nls_disk, nls_io), neither needs RCU delays of any sort. On VFAT, for that matter - again, two tables, one needs RCU delay, another doesn't (both get dropped from the same helper, so both get it).

BTW, is there any reason not to have synchronize_rcu() in delete_module(2), just before calling ->exit()?

It's not a hot path, unless something really weird is going on, and it would get rid of the need to delay unload_nls() calls...

We already have one - it's hidden in free_module().

It's done after the kallsyms list removal. Is that too late?

Note that module *loading* actually has a few other synchronize_rcu() calls in the failure path. In fact, load_module() has *two* "synchronize_rcu()" calls:

- after ftrace_release_mod(mod)

- before releasing module_mutex

for the failure path, but there's only one for module unloading.

Adding one to after ftrace_release_mod() - where we have that async_synchronize_full() - would make module unloading match the loading error path.

But the synchronize_rcu calls do seem to be a bit randomly sprinkled about. Maybe the one in free_module() is already sufficient for nls?

Linus

Never mind. You want it before the ->exit(). That makes sense, and there is no matching code path for the module load failure case (since that implies no - or failed - init()).

Linus

->d_revalidate() bails out there, anyway. It's not enough to prevent getting into ->get_link() in RCU mode, but that could happen only in a very contrieved setup. Not worth trying to do anything fancy here unless ->d_revalidate() stops kicking out of RCU mode at least in some cases.

---
 fs/smb/client/cifsfs.c | 3 +++
 1 file changed, 3 insertions(+)

index 22869cda1356..2b044e47a3a6 100644
@@ -1170,6 +1170,9 @@ const char *cifs_get_link(struct dentry *dentry, struct inode *inode,
 {
 	char *target_path;
 
+	if (!dentry)
+		return ERR_PTR(-ECHILD);
+
 	target_path = kmalloc(PATH_MAX, GFP_KERNEL);
 	if (!target_path)
 		return ERR_PTR(-ENOMEM);

2.39.2

that keeps both around until struct inode is freed, making access to them safe from rcu-pathwalk

---
 fs/proc/base.c  |  2 --
 fs/proc/inode.c | 19 ++++++++-----------
 2 files changed, 8 insertions(+), 13 deletions(-)

index ffd54617c354..8e93b11a0fed 100644
@@ -1881,8 +1881,6 @@ void proc_pid_evict_inode(struct proc_inode *ei)
 		hlist_del_init_rcu(&ei->sibling_inodes);
 		spin_unlock(&pid->lock);
 	}
-
-	put_pid(pid);
 }
 
 struct inode *proc_pid_make_inode(struct super_block *sb,

index 532dc9d240f7..c83e1a55da42 100644
@@ -30,7 +30,6 @@
 
 static void proc_evict_inode(struct inode *inode)
 {
-	struct proc_dir_entry *de;
 	struct ctl_table_header *head;
 	struct proc_inode *ei = PROC_I(inode);
 

@@ -38,17 +37,8 @@ static void proc_evict_inode(struct inode *inode)
 	clear_inode(inode);
 
 	/* Stop tracking associated processes */
-	if (ei->pid) {
+	if (ei->pid)
 		proc_pid_evict_inode(ei);
-		ei->pid = NULL;
-	}
-
-	/* Let go of any associated proc directory entry */
-	de = ei->pde;
-	if (de) {
-		pde_put(de);
-		ei->pde = NULL;
-	}
 
 	head = ei->sysctl;
 	if (head) {

@@ -80,6 +70,13 @@ static struct inode *proc_alloc_inode(struct super_block *sb)
 
 static void proc_free_inode(struct inode *inode)
 {
+	struct proc_inode *ei = PROC_I(inode);
+
+	if (ei->pid)
+		put_pid(ei->pid);
+	/* Let go of any associated proc directory entry */
+	if (ei->pde)
+		pde_put(ei->pde);
 	kmem_cache_free(proc_inode_cachep, PROC_I(inode));
 }
 

2.39.2

makes proc_pid_ns() safe from rcu pathwalk (put_pid_ns() is still synchronous, but that's not a problem - it does rcu-delay everything that needs to be)

---
 fs/proc/root.c          | 2 +-
 include/linux/proc_fs.h | 1 +
 2 files changed, 2 insertions(+), 1 deletion(-)

index 9191248f2dac..d3148e1d883b 100644
@@ -271,7 +271,7 @@ static void proc_kill_sb(struct super_block *sb)
 
 	kill_anon_super(sb);
 	put_pid_ns(fs_info->pid_ns);
-	kfree(fs_info);
+	kfree_rcu(fs_info, rcu);
 }
 
 static struct file_system_type proc_fs_type = {

index de407e7c3b55..0b2a89854440 100644
@@ -65,6 +65,7 @@ struct proc_fs_info {
 	kgid_t pid_gid;
 	enum proc_hidepid hide_pid;
 	enum proc_pidonly pidonly;
+	struct rcu_head rcu;
 };
 
 static inline struct proc_fs_info *proc_sb_info(struct super_block *sb)

2.39.2

in RCU mode we might race with gfs2_evict_inode(), which zeroes ->i_gl. Freeing of the object it points to is RCU-delayed, so if we manage to fetch the pointer before it's been replaced with NULL, we are fine. Check if we'd fetched NULL and treat that as "bail out and tell the caller to get out of RCU mode".

---
 fs/gfs2/inode.c | 6 ++++--
 fs/gfs2/super.c | 2 +-
 2 files changed, 5 insertions(+), 3 deletions(-)

index 0eac04507904..e2432c327599 100644
@@ -1868,14 +1868,16 @@ int gfs2_permission(struct mnt_idmap *idmap, struct inode *inode,
 {
 	struct gfs2_inode *ip;
 	struct gfs2_holder i_gh;
+	struct gfs2_glock *gl;
 	int error;
 
 	gfs2_holder_mark_uninitialized(&i_gh);
 	ip = GFS2_I(inode);
-	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
+	gl = rcu_dereference(ip->i_gl);
+	if (!gl || gfs2_glock_is_locked_by_me(gl) == NULL) {
 		if (mask & MAY_NOT_BLOCK)
 			return -ECHILD;
-		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
+		error = gfs2_glock_nq_init(gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
 		if (error)
 			return error;
 	}

index 02d93da21b2b..0dd5641990b9 100644
@@ -1550,7 +1550,7 @@ static void gfs2_evict_inode(struct inode *inode)
 		wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
 		gfs2_glock_add_to_lru(ip->i_gl);
 		gfs2_glock_put_eventually(ip->i_gl);
-		ip->i_gl = NULL;
+		rcu_assign_pointer(ip->i_gl, NULL);
 	}
 }
 

2.39.2

We shouldn't observe NULL ->i_gl unless we are in RCU mode, which means we'll bail out without reaching gfs2_glock_nq_init()...

Something like

	if (unlikely(!gl)) {
		/* inode is getting torn down, must be RCU mode */
		WARN_ON_ONCE(!(mask & MAY_NOT_BLOCK));
		return -ECHILD;
	}

might be less confusing way to express that...

Am Mo., 2. Okt. 2023 um 19:09 Uhr schrieb Al Viro <[email protected]>:

Looking good, thanks. I'll queue it up.

Could you please send such fixes to the filesystem-specific list in the future (scripts/get_maintainer.pl)?

Thanks,
Andreas

nfs_set_verifier() relies upon dentry being pinned; if that's the case, grabbing ->d_lock stabilizes ->d_parent and guarantees that ->d_parent points to a positive dentry. For something we'd run into in RCU mode that is *not* true - dentry might've been through dentry_kill() just as we grabbed ->d_lock, with its parent going through the same just as we get to into nfs_set_verifier_locked(). It might get to detaching inode (and zeroing ->d_inode) before nfs_set_verifier_locked() gets to fetching that; we get an oops as the result.

That can happen in nfs{,4} ->d_revalidate(); we check that parent is positive, but that's done before we get to nfs_set_verifier() and it's possible for memory pressure to pick our dentry as eviction candidate by that time. If that happens, back-to-back attempts to kill dentry and its parent are quite normal.

---
 fs/nfs/dir.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

index e6a51fd94fea..8ffc1f78ba51 100644
@@ -1431,9 +1431,9 @@ static bool nfs_verifier_is_delegated(struct dentry *dentry)
 static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
 {
 	struct inode *inode = d_inode(dentry);
-	struct inode *dir = d_inode(dentry->d_parent);
+	struct inode *dir = d_inode_rcu(dentry->d_parent);
 
-	if (!nfs_verify_change_attribute(dir, verf))
+	if (!dir || !nfs_verify_change_attribute(dir, verf))
 		return;
 	if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
 		nfs_set_verifier_delegated(&verf);

2.39.2

NFS ->d_revalidate(), ->permission() and ->get_link() need to access some parts of nfs_server when called in RCU mode:

	server->flags
	server->caps
	*(server->io_stats)
and, worst of all, call
	server->nfs_client->rpc_ops->have_delegation

(the last one - as NFS_PROTO(inode)->have_delegation()). We really don't want to RCU-delay the entire nfs_free_server() (it would have to be done with schedule_work() from RCU callback, since it can't be made to run from interrupt context), but actual freeing of nfs_server and ->io_stats can be done via call_rcu() just fine. nfs_client part is handled simply by making nfs_free_client() use kfree_rcu().

---
 fs/nfs/client.c           | 13 ++++++++++---
 include/linux/nfs_fs_sb.h |  2 ++
 2 files changed, 12 insertions(+), 3 deletions(-)

index 44eca51b2808..fbdc9ca80f71 100644
@@ -246,7 +246,7 @@ void nfs_free_client(struct nfs_client *clp)
 	put_nfs_version(clp->cl_nfs_mod);
 	kfree(clp->cl_hostname);
 	kfree(clp->cl_acceptor);
-	kfree(clp);
+	kfree_rcu(clp, rcu);
 }
 EXPORT_SYMBOL_GPL(nfs_free_client);
 

@@ -1006,6 +1006,14 @@ struct nfs_server *nfs_alloc_server(void)
 }
 EXPORT_SYMBOL_GPL(nfs_alloc_server);
 
+static void delayed_free(struct rcu_head *p)
+{
+	struct nfs_server *server = container_of(p, struct nfs_server, rcu);
+
+	nfs_free_iostats(server->io_stats);
+	kfree(server);
+}
+
 /*
  * Free up a server record
  */

@@ -1031,10 +1039,9 @@ void nfs_free_server(struct nfs_server *server)
 
 	ida_destroy(&server->lockowner_id);
 	ida_destroy(&server->openowner_id);
-	nfs_free_iostats(server->io_stats);
 	put_cred(server->cred);
-	kfree(server);
 	nfs_release_automount_timer();
+	call_rcu(&server->rcu, delayed_free);
 }
 EXPORT_SYMBOL_GPL(nfs_free_server);
 

index cd628c4b011e..4bd16da65c05 100644
@@ -124,6 +124,7 @@ struct nfs_client {
 	char			cl_ipaddr[48];
 	struct net		*cl_net;
 	struct list_head	pending_cb_stateids;
+	struct rcu_head		rcu;
 };
 
 /*

@@ -264,6 +265,7 @@ struct nfs_server {
 	const struct cred	*cred;
 	bool			has_sec_mnt_opts;
 	struct kobject		kobj;
+	struct rcu_head		rcu;
 };
 
 /* Server capabilities */

2.39.2

->permission(), ->get_link() and ->inode_get_acl() might dereference ->s_fs_info (and, in case of ->permission(), ->s_fs_info->fc->user_ns as well) when called from rcu pathwalk.

Freeing ->s_fs_info->fc is rcu-delayed; we need to make freeing ->s_fs_info and dropping ->user_ns rcu-delayed too.

---
 fs/fuse/cuse.c   |  3 +--
 fs/fuse/fuse_i.h |  1 +
 fs/fuse/inode.c  | 15 +++++++++++----
 3 files changed, 13 insertions(+), 6 deletions(-)

index 91e89e68177e..b6cad106c37e 100644
@@ -474,8 +474,7 @@ static int cuse_send_init(struct cuse_conn *cc)
 
 static void cuse_fc_release(struct fuse_conn *fc)
 {
-	struct cuse_conn *cc = fc_to_cc(fc);
-	kfree_rcu(cc, fc.rcu);
+	kfree(fc_to_cc(fc));
 }
 
 /**

index bf0b85d0b95c..0c45014fbb03 100644
@@ -873,6 +873,7 @@ struct fuse_mount {
 
 	/* Entry on fc->mounts */
 	struct list_head fc_entry;
+	struct rcu_head rcu;
 };
 
 static inline struct fuse_mount *get_fuse_mount_super(struct super_block *sb)

index 2e4eb7cf26fb..e13c9312cb55 100644
@@ -881,6 +881,14 @@ void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
 }
 EXPORT_SYMBOL_GPL(fuse_conn_init);
 
+static void delayed_release(struct rcu_head *p)
+{
+	struct fuse_conn *fc = container_of(p, struct fuse_conn, rcu);
+
+	put_user_ns(fc->user_ns);
+	fc->release(fc);
+}
+
 void fuse_conn_put(struct fuse_conn *fc)
 {
 	if (refcount_dec_and_test(&fc->count)) {

@@ -892,13 +900,12 @@ void fuse_conn_put(struct fuse_conn *fc)
 		if (fiq->ops->release)
 			fiq->ops->release(fiq);
 		put_pid_ns(fc->pid_ns);
-		put_user_ns(fc->user_ns);
 		bucket = rcu_dereference_protected(fc->curr_bucket, 1);
 		if (bucket) {
 			WARN_ON(atomic_read(&bucket->count) != 1);
 			kfree(bucket);
 		}
-		fc->release(fc);
+		call_rcu(&fc->rcu, delayed_release);
 	}
 }
 EXPORT_SYMBOL_GPL(fuse_conn_put);

@@ -1316,7 +1323,7 @@ EXPORT_SYMBOL_GPL(fuse_send_init);
 void fuse_free_conn(struct fuse_conn *fc)
 {
 	WARN_ON(!list_empty(&fc->devices));
-	kfree_rcu(fc, rcu);
+	kfree(fc);
 }
 EXPORT_SYMBOL_GPL(fuse_free_conn);
 

@@ -1833,7 +1840,7 @@ static void fuse_sb_destroy(struct super_block *sb)
 void fuse_mount_destroy(struct fuse_mount *fm)
 {
 	fuse_conn_put(fm->fc);
-	kfree(fm);
+	kfree_rcu(fm, rcu);
 }
 EXPORT_SYMBOL(fuse_mount_destroy);
 

2.39.2

In __afs_break_callback() we might check ->cb_nr_mmap and if it's non-zero do queue_work(&vnode->cb_work). In afs_drop_open_mmap() we decrement ->cb_nr_mmap and do flush_work(&vnode->cb_work) if it reaches zero.

The trouble is, there's nothing to prevent __afs_break_callback() from seeing ->cb_nr_mmap before the decrement and do queue_work() after both the decrement and flush_work(). If that happens, we might be in trouble - vnode might get freed before the queued work runs.

__afs_break_callback() is always done under ->cb_lock, so let's make sure that ->cb_nr_mmap can change from non-zero to zero while holding ->cb_lock (the spinlock component of it - it's a seqlock and we don't need to mess with the counter).

---
 fs/afs/file.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

index d37dd201752b..0012ea300eb5 100644
@@ -529,13 +529,17 @@ static void afs_add_open_mmap(struct afs_vnode *vnode)
 
 static void afs_drop_open_mmap(struct afs_vnode *vnode)
 {
-	if (!atomic_dec_and_test(&vnode->cb_nr_mmap))
+	if (atomic_add_unless(&vnode->cb_nr_mmap, -1, 1))
 		return;
 
 	down_write(&vnode->volume->cell->fs_open_mmaps_lock);
 
-	if (atomic_read(&vnode->cb_nr_mmap) == 0)
+	read_seqlock_excl(&vnode->cb_lock);
+	// the only place where ->cb_nr_mmap may hit 0
+	// see __afs_break_callback() for the other side...
+	if (atomic_dec_and_test(&vnode->cb_nr_mmap))
 		list_del_init(&vnode->cb_mmap_link);
+	read_sequnlock_excl(&vnode->cb_lock);
 
 	up_write(&vnode->volume->cell->fs_open_mmaps_lock);
 	flush_work(&vnode->cb_work);

2.39.2

... into ->free_inode(), that is.

---
 fs/overlayfs/super.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

index def266b5e2a3..f09184b865ec 100644
@@ -167,6 +167,7 @@ static void ovl_free_inode(struct inode *inode)
 	struct ovl_inode *oi = OVL_I(inode);
 
 	kfree(oi->redirect);
+	kfree(oi->oe);
 	mutex_destroy(&oi->lock);
 	kmem_cache_free(ovl_inode_cachep, oi);
 }

@@ -176,7 +177,7 @@ static void ovl_destroy_inode(struct inode *inode)
 	struct ovl_inode *oi = OVL_I(inode);
 
 	dput(oi->__upperdentry);
-	ovl_free_entry(oi->oe);
+	ovl_stack_put(ovl_lowerstack(oi->oe), ovl_numlower(oi->oe));
 	if (S_ISDIR(inode->i_mode))
 		ovl_dir_cache_free(inode);
 	else

2.39.2

d_inode_rcu() is right - we might be in rcu pathwalk; however, OVL_E() hides plain d_inode() on the same dentry...

---
 fs/overlayfs/super.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

index f09184b865ec..905d3aaf4e55 100644
@@ -104,8 +104,8 @@ static int ovl_revalidate_real(struct dentry *d, unsigned int flags, bool weak)
 static int ovl_dentry_revalidate_common(struct dentry *dentry,
 					unsigned int flags, bool weak)
 {
-	struct ovl_entry *oe = OVL_E(dentry);
-	struct ovl_path *lowerstack = ovl_lowerstack(oe);
+	struct ovl_entry *oe;
+	struct ovl_path *lowerstack;
 	struct inode *inode = d_inode_rcu(dentry);
 	struct dentry *upper;
 	unsigned int i;

@@ -115,6 +115,8 @@ static int ovl_dentry_revalidate_common(struct dentry *dentry,
 	if (!inode)
 		return -ECHILD;
 
+	oe = OVL_I_E(inode);
+	lowerstack = ovl_lowerstack(oe);
 	upper = ovl_i_dentry_upper(inode);
 	if (upper)
 		ret = ovl_revalidate_real(upper, flags, weak);

2.39.2

ovl_permission() accesses ->layers[...].mnt; we can't have ->layers freed without an RCU delay on fs shutdown. Fortunately, kern_unmount_array() used to drop those mounts does include an RCU delay, so freeing is delayed; unfortunately, the array passed to kern_unmount_array() is formed by mangling ->layers contents and that happens without any delays.

Use a separate array instead; local if we have a few layers, kmalloc'ed if there's a lot of them. If allocation fails, fall back to kern_unmount() for individual mounts; it's not a fast path by any stretch of imagination.

---
 fs/overlayfs/ovl_entry.h |  1 -
 fs/overlayfs/params.c    | 26 ++++++++++++++++++++------
 2 files changed, 20 insertions(+), 7 deletions(-)

index e9539f98e86a..618b63bb7987 100644
@@ -30,7 +30,6 @@ struct ovl_sb {
 };
 
 struct ovl_layer {
-	/* ovl_free_fs() relies on @mnt being the first member! */
 	struct vfsmount *mnt;
 	/* Trap in ovl inode cache */
 	struct inode *trap;

index b9355bb6d75a..ab594fd407b4 100644
@@ -738,8 +738,15 @@ int ovl_init_fs_context(struct fs_context *fc)
 void ovl_free_fs(struct ovl_fs *ofs)
 {
 	struct vfsmount **mounts;
+	struct vfsmount *m[16];
+	unsigned n = ofs->numlayer;
 	unsigned i;
 
+	if (n > 16)
+		mounts = kmalloc_array(n, sizeof(struct mount *), GFP_KERNEL);
+	else
+		mounts = m;
+
 	iput(ofs->workbasedir_trap);
 	iput(ofs->indexdir_trap);
 	iput(ofs->workdir_trap);

@@ -752,14 +759,21 @@ void ovl_free_fs(struct ovl_fs *ofs)
 	if (ofs->upperdir_locked)
 		ovl_inuse_unlock(ovl_upper_mnt(ofs)->mnt_root);
 
-	/* Hack!  Reuse ofs->layers as a vfsmount array before freeing it */
-	mounts = (struct vfsmount **) ofs->layers;
-	for (i = 0; i < ofs->numlayer; i++) {
+	for (i = 0; i < n; i++) {
 		iput(ofs->layers[i].trap);
-		mounts[i] = ofs->layers[i].mnt;
-		kfree(ofs->layers[i].name);
+		if (unlikely(!mounts))
+			kern_unmount(ofs->layers[i].mnt);
+		else
+			mounts[i] = ofs->layers[i].mnt;
 	}
-	kern_unmount_array(mounts, ofs->numlayer);
+	if (mounts) {
+		kern_unmount_array(mounts, n);
+		if (mounts != m)
+			kfree(mounts);
+	}
+	// by this point we had an RCU delay from kern_unmount{_array,}()
+	for (i = 0; i < n; i++)
+		kfree(ofs->layers[i].name);
 	kfree(ofs->layers);
 	for (i = 0; i < ofs->numfs; i++)
 		free_anon_bdev(ofs->fs[i].pseudo_dev);

2.39.2

If that is the case, then having 3 different code paths seems quite an excessive over optimization...

I think there is a better way -
layout the mounts array linearly in ofs->mounts[] to begin with, remove .mnt out of ovl_layer and use ofs->mounts[layer->idx] to get to a layer's mount.

I can try to write this patch to see how it ends up looking.

Thanks,
Amir.

Fine by me; about the only problem I see is the cache footprint... How many layers do you consider the normal use, actually?

I've also considered just allocating the extra space for ofs->mounts[] at super creation time rather than on super destruction. I just cannot get myself to be bothered with this cleanup code because of saving memory of ovl_fs.

However, looking closer, we have a wasfull layer->name pointer, which is only ever used for ovl_show_options() (to print the original requested layer path from mount options).

So I am inclined to move these rarely accessed pointers to ofs->layer_names[], which can be used for the temp array for kern_unmount_array() because freeing name does not need RCU delay AFAICT (?).

I will take a swing at it.

Define "normal".
Currently, we have #define OVL_MAX_STACK 500
and being able to create an overlay with many layers was cited as one of the requirements to drive the move to new mount api:

https://lore.kernel.org/linux-unionfs/[email protected]/

I am not really familiar with the user workloads that need that many layers and why. It's obviously not the common case.

Thanks,
Amir.

AFAICS, it doesn't. The only user after the setup is done is ->show_options(), i.e. show_vfsmnt() and show_mountinfo(). Anyone who tries to use those without making sure that struct mount is not going to come apart under them will have far worse troubles...

FWIW, I'm perfectly fine with having these fixes go through your tree; they are independent from the rest of the series.

However, ovl_lowerstack(oe) does not appear to be stable in RCU walk...

Ah, you fixed that in another patch.
If you are going to be sending this to Linus, please add

I was going to send some fixes this week anyway, so I can pick those through the overlayfs tree if you like.

Thanks,
Amir.

Al,

From all your series, the two ovl fixes are for rather new regressions (v6.5) so I queued your two regression fixes (13,14) and my own version of patch 15 to go into linux-next via overlayfs tree [1] and I will send them to Linus later this week, so they can make their way to 6.5.y.

Thanks,
Amir.

[1] https://git.kernel.org/pub/scm/linux/kernel/git/overlayfs/vfs.git/log/?h=ovl-fixes

PS: ceph ->d_revalidate() uses ->s_fs_info and stuff hanging off it (mdsc) and I'm not familiar enough with the codebase to do anything useful with it; AFAICS, there are UAF on fs shutdown, but I'd rather leave that one to ceph folks. NTFS3 ->d_hash() and ->d_compare() do blocking allocations, and that stuff can be called under rcu_read_lock(); ->d_compare() pretty much always is called so...