autofs4: deadlock during create
Ian Kent
maintainerPATCH 1/1 autofs4: deadlock during create
Hi,
Due to inconsistent locking in the VFS between calls to lookup and revalidate deadlock can occur in the automounter.
The inconsistency is that the directory inode mutex is held for both lookup and revalidate calls when called via lookup_hash whereas it is held only for lookup during a path walk. Consequently, if the mutex is held during a call to revalidate autofs4 can't release the mutex to callback the daemon as it can't know whether it owns the mutex.
This situation happens when a process tries to create a directory within an automount and a second process also tries to create the same directory between the lookup and the mkdir. Since the first process has dropped the mutex for the daemon callback, the second process takes it during revalidate leading to deadlock between the autofs daemon and the second process when the daemon tries to create the mount point directory.
After spending quite a bit of time trying to resolve this on more than one occassion, using rather complex and ulgy approaches, it turns out that just delaying the hashing of the dentry until the create operation work fine.
---
fs/autofs4/root.c renamed from fs/autofs4/root.c.lookup-delay-hash +13 -13
@@ -587,19 +587,20 @@ static struct dentry *autofs4_lookup(str unhashed =autofs4_lookup_unhashed (sbi, dentry->d_parent, &dentry->d_name);if (!unhashed) {/* -* Mark the dentry incomplete, but add it. This is needed so -* that the VFS layer knows about the dentry, and we can count -* on catching any lookups through the revalidate. -* -* Let all the hard work be done by the revalidate function that -* needs to be able to do this anyway.. -* -* We need to do this before we release the directory semaphore. +* Mark the dentry incomplete but don't hash it. We do this +* to serialize our inode creation operations (symlink and +* mkdir) which prevents deadlock during the callback to +* the daemon. Subsequent user space lookups for the same +* dentry are placed on the wait queue while the daemon +* itself is allowed passage unresticted so the create +* operation itself can then hash the dentry. Finally, +* we check for the hashed dentry and return the newly +* hashed dentry. */ dentry->d_op = &autofs4_root_dentry_operations; dentry->d_fsdata =NULL ; -d_add (dentry,NULL ); +d_instantiate (dentry,NULL ); }else {struct autofs_info *ino =autofs4_dentry_ino (unhashed);DPRINTK ("rehash %p with %p" , dentry, unhashed);
@@ -615,7 +616,6 @@ static struct dentry *autofs4_lookup(strautofs4_wait (sbi, unhashed,NFY_NONE );DPRINTK ("request completed" ); } -d_rehash (unhashed); dentry = unhashed; }
@@ -658,7 +658,7 @@ static struct dentry *autofs4_lookup(str* for all system calls, but it should be OK for the operations * we permit from an autofs. */ -if (dentry->d_inode &&d_unhashed (dentry)) { +if (!oz_mode &&d_unhashed (dentry)) {/* * A user space application can (and has done in the past) * remove and re-create this directory during the callback.
@@ -716,7 +716,7 @@ static int autofs4_dir_symlink(struct instrcpy (cp, symname); inode =autofs4_get_inode (dir->i_sb, ino); -d_instantiate (dentry, inode); +d_add (dentry, inode);if (dir == dir->i_sb->s_root->d_inode) dentry->d_op = &autofs4_root_dentry_operations;
@@ -844,7 +844,7 @@ static int autofs4_dir_mkdir(struct inodreturn -ENOSPC ; inode =autofs4_get_inode (dir->i_sb, ino); -d_instantiate (dentry, inode); +d_add (dentry, inode);if (dir == dir->i_sb->s_root->d_inode) dentry->d_op = &autofs4_root_dentry_operations;
Quoted 19 lines · Ian Kent
Hi,
Due to inconsistent locking in the VFS between calls to lookup and revalidate deadlock can occur in the automounter.
The inconsistency is that the directory inode mutex is held for both lookup and revalidate calls when called via
lookup_hashwhereas it is held only for lookup during a path walk. Consequently, if the mutex is held during a call to revalidate autofs4 can't release the mutex to callback the daemon as it can't know whether it owns the mutex.This situation happens when a process tries to create a directory within an automount and a second process also tries to create the same directory between the lookup and the mkdir. Since the first process has dropped the mutex for the daemon callback, the second process takes it during revalidate leading to deadlock between the autofs daemon and the second process when the daemon tries to create the mount point directory.
Quoted 6 lines
After spending quite a bit of time trying to resolve this on more than one occassion, using rather complex and ulgy approaches, it turns out that just delaying the hashing of the dentry until the create operation work fine.
Signed-off-by Ian Kent <[email protected]>
-Jeff
Ian Kent
maintainerQuoted 6 lines · Jeff Moyer
Quoted 5 lines · Ian Kent
After spending quite a bit of time trying to resolve this on more than one occassion, using rather complex and ulgy approaches, it turns out that just delaying the hashing of the dentry until the create operation work fine.
There seems to be an unexpected problem with the patch after removal of the d_rehash line left in the original by mistake.
I'm working on it.
Sorry all.
Ian
Ian Kent
maintainerQuoted 11 lines · Ian Kent
Quoted 6 lines · Jeff Moyer
Quoted 5 lines · Ian Kent
After spending quite a bit of time trying to resolve this on more than one occassion, using rather complex and ulgy approaches, it turns out that just delaying the hashing of the dentry until the create operation work fine.
There seems to be an unexpected problem with the patch after removal of the
d_rehashline left in the original by mistake.
I see you haven't got to this yet, probably because I used your old email address, so to save hassle, here is the final patch and the description.
Due to inconsistent locking in the VFS between calls to lookup and revalidate deadlock can occur in the automounter.
The inconsistency is that the directory inode mutex is held for both lookup and revalidate calls when called via lookup_hash whereas it is held only for lookup during a path walk. Consequently, if the mutex is held during a call to revalidate autofs4 can't release the mutex to callback the daemon as it can't know whether it owns the mutex.
This situation happens when a process tries to create a directory within an automount and a second process also tries to create the same directory between the lookup and the mkdir. Since the first process has dropped the mutex for the daemon callback, the second process takes it during revalidate leading to deadlock between the autofs daemon and the second process when the daemon tries to create the mount point directory.
After spending quite a bit of time trying to resolve this on more than one occassion, using rather complex and ulgy approaches, it turns out that just delaying the hashing of the dentry until the create operation work fine.
---
fs/autofs4/root.c renamed from fs/autofs4/root.c.lookup-delay-hash +17 -14
@@ -587,19 +587,20 @@ static struct dentry *autofs4_lookup(str unhashed =autofs4_lookup_unhashed (sbi, dentry->d_parent, &dentry->d_name);if (!unhashed) {/* -* Mark the dentry incomplete, but add it. This is needed so -* that the VFS layer knows about the dentry, and we can count -* on catching any lookups through the revalidate. -* -* Let all the hard work be done by the revalidate function that -* needs to be able to do this anyway.. -* -* We need to do this before we release the directory semaphore. +* Mark the dentry incomplete but don't hash it. We do this +* to serialize our inode creation operations (symlink and +* mkdir) which prevents deadlock during the callback to +* the daemon. Subsequent user space lookups for the same +* dentry are placed on the wait queue while the daemon +* itself is allowed passage unresticted so the create +* operation itself can then hash the dentry. Finally, +* we check for the hashed dentry and return the newly +* hashed dentry. */ dentry->d_op = &autofs4_root_dentry_operations; dentry->d_fsdata =NULL ; -d_add (dentry,NULL ); +d_instantiate (dentry,NULL ); }else {struct autofs_info *ino =autofs4_dentry_ino (unhashed);DPRINTK ("rehash %p with %p" , dentry, unhashed);
@@ -607,15 +608,17 @@ static struct dentry *autofs4_lookup(str* If we are racing with expire the request might not * be quite complete but the directory has been removed * so it must have been successful, so just wait for it. +* We need to ensure the AUTOFS_INF_EXPIRING flag is clear +* before continuing as revalidate may fail when calling +* try_to_fill_dentry (returning EAGAIN) if we don't. */ -if (ino && (ino->flags &AUTOFS_INF_EXPIRING )) { +while (ino && (ino->flags &AUTOFS_INF_EXPIRING )) {DPRINTK ("wait for incomplete expire %p name=%.*s" , unhashed, unhashed->d_name.len, unhashed->d_name.name);autofs4_wait (sbi, unhashed,NFY_NONE );DPRINTK ("request completed" ); } -d_rehash (unhashed); dentry = unhashed; }
@@ -658,7 +661,7 @@ static struct dentry *autofs4_lookup(str* for all system calls, but it should be OK for the operations * we permit from an autofs. */ -if (dentry->d_inode &&d_unhashed (dentry)) { +if (!oz_mode &&d_unhashed (dentry)) {/* * A user space application can (and has done in the past) * remove and re-create this directory during the callback.
@@ -716,7 +719,7 @@ static int autofs4_dir_symlink(struct instrcpy (cp, symname); inode =autofs4_get_inode (dir->i_sb, ino); -d_instantiate (dentry, inode); +d_add (dentry, inode);if (dir == dir->i_sb->s_root->d_inode) dentry->d_op = &autofs4_root_dentry_operations;
@@ -844,7 +847,7 @@ static int autofs4_dir_mkdir(struct inodreturn -ENOSPC ; inode =autofs4_get_inode (dir->i_sb, ino); -d_instantiate (dentry, inode); +d_add (dentry, inode);if (dir == dir->i_sb->s_root->d_inode) dentry->d_op = &autofs4_root_dentry_operations;
Quoted 9 lines · On Fri, 17 Aug 2007 11:56:11 +0800 Ian Kent <[email protected]> wrote:
Quoted 8 lines · Ian Kent
Quoted 6 lines · Jeff Moyer
Quoted 5 lines · Ian Kent
After spending quite a bit of time trying to resolve this on more than one occassion, using rather complex and ulgy approaches, it turns out that just delaying the hashing of the dentry until the create operation work fine.
Quoted 6 lines
Quoted 3 lines
There seems to be an unexpected problem with the patch after removal of the
d_rehashline left in the original by mistake.I see you haven't got to this yet, probably because I used your old email address,
Actually my cursor has advanced that far, but you said it didn't work right yet.
Quoted 2 lines
so to save hassle, here is the final patch and the description.
OK, thanks.