]> err.no Git - linux-2.6/commitdiff
[PATCH] ext[34]: EA block reference count racing fix
authorMingming Cao <cmm@us.ibm.com>
Thu, 1 Mar 2007 04:13:35 +0000 (20:13 -0800)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Thu, 1 Mar 2007 22:53:38 +0000 (14:53 -0800)
There are race issues around ext[34] xattr block release code.

ext[34]_xattr_release_block() checks the reference count of xattr block
(h_refcount) and frees that xattr block if it is the last one reference it.
 Unlike ext2, the check of this counter is unprotected by any lock.
ext[34]_xattr_release_block() will free the mb_cache entry before freeing
that xattr block.  There is a small window between the check for the re
h_refcount ==1 and the call to mb_cache_entry_free().  During this small
window another inode might find this xattr block from the mbcache and reuse
it, racing a refcount updates.  The xattr block will later be freed by the
first inode without notice other inode is still use it.  Later if that
block is reallocated as a datablock for other file, then more serious
problem might happen.

We need put a lock around places checking the refount as well to avoid
racing issue.  Another place need this kind of protection is in
ext3_xattr_block_set(), where it will modify the xattr block content in-
the-fly if the refcount is 1 (means it's the only inode reference it).

This will also fix another issue: the xattr block may not get freed at all
if no lock is to protect the refcount check at the release time.  It is
possible that the last two inodes could release the shared xattr block at
the same time.  But both of them think they are not the last one so only
decreased the h_refcount without freeing xattr block at all.

We need to call lock_buffer() after ext3_journal_get_write_access() to
avoid deadlock (because the later will call lock_buffer()/unlock_buffer
() as well).

Signed-off-by: Mingming Cao <cmm@us.ibm.com>
Cc: Andreas Gruenbacher <agruen@suse.de>
Cc: <linux-ext4@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/ext3/xattr.c
fs/ext4/xattr.c

index 99857a400f4be217dcf92e5f16ac6c62783b62e2..12f7dda1232cb8b17749d6f3e87d6ea77a40bb0c 100644 (file)
@@ -475,8 +475,15 @@ ext3_xattr_release_block(handle_t *handle, struct inode *inode,
                         struct buffer_head *bh)
 {
        struct mb_cache_entry *ce = NULL;
+       int error = 0;
 
        ce = mb_cache_entry_get(ext3_xattr_cache, bh->b_bdev, bh->b_blocknr);
+       error = ext3_journal_get_write_access(handle, bh);
+       if (error)
+                goto out;
+
+       lock_buffer(bh);
+
        if (BHDR(bh)->h_refcount == cpu_to_le32(1)) {
                ea_bdebug(bh, "refcount now=0; freeing");
                if (ce)
@@ -485,21 +492,20 @@ ext3_xattr_release_block(handle_t *handle, struct inode *inode,
                get_bh(bh);
                ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
        } else {
-               if (ext3_journal_get_write_access(handle, bh) == 0) {
-                       lock_buffer(bh);
-                       BHDR(bh)->h_refcount = cpu_to_le32(
+               BHDR(bh)->h_refcount = cpu_to_le32(
                                le32_to_cpu(BHDR(bh)->h_refcount) - 1);
-                       ext3_journal_dirty_metadata(handle, bh);
-                       if (IS_SYNC(inode))
-                               handle->h_sync = 1;
-                       DQUOT_FREE_BLOCK(inode, 1);
-                       unlock_buffer(bh);
-                       ea_bdebug(bh, "refcount now=%d; releasing",
-                                 le32_to_cpu(BHDR(bh)->h_refcount));
-               }
+               error = ext3_journal_dirty_metadata(handle, bh);
+               handle->h_sync = 1;
+               DQUOT_FREE_BLOCK(inode, 1);
+               ea_bdebug(bh, "refcount now=%d; releasing",
+                         le32_to_cpu(BHDR(bh)->h_refcount));
                if (ce)
                        mb_cache_entry_release(ce);
        }
+       unlock_buffer(bh);
+out:
+       ext3_std_error(inode->i_sb, error);
+       return;
 }
 
 struct ext3_xattr_info {
@@ -675,7 +681,7 @@ ext3_xattr_block_set(handle_t *handle, struct inode *inode,
        struct buffer_head *new_bh = NULL;
        struct ext3_xattr_search *s = &bs->s;
        struct mb_cache_entry *ce = NULL;
-       int error;
+       int error = 0;
 
 #define header(x) ((struct ext3_xattr_header *)(x))
 
@@ -684,16 +690,17 @@ ext3_xattr_block_set(handle_t *handle, struct inode *inode,
        if (s->base) {
                ce = mb_cache_entry_get(ext3_xattr_cache, bs->bh->b_bdev,
                                        bs->bh->b_blocknr);
+               error = ext3_journal_get_write_access(handle, bs->bh);
+               if (error)
+                       goto cleanup;
+               lock_buffer(bs->bh);
+
                if (header(s->base)->h_refcount == cpu_to_le32(1)) {
                        if (ce) {
                                mb_cache_entry_free(ce);
                                ce = NULL;
                        }
                        ea_bdebug(bs->bh, "modifying in-place");
-                       error = ext3_journal_get_write_access(handle, bs->bh);
-                       if (error)
-                               goto cleanup;
-                       lock_buffer(bs->bh);
                        error = ext3_xattr_set_entry(i, s);
                        if (!error) {
                                if (!IS_LAST_ENTRY(s->first))
@@ -713,6 +720,9 @@ ext3_xattr_block_set(handle_t *handle, struct inode *inode,
                } else {
                        int offset = (char *)s->here - bs->bh->b_data;
 
+                       unlock_buffer(bs->bh);
+                       journal_release_buffer(handle, bs->bh);
+
                        if (ce) {
                                mb_cache_entry_release(ce);
                                ce = NULL;
index dc969c357aa1d4ef9e207054e506053b8efdc0c6..e832e96095b33c177e880db6370e1686153439f6 100644 (file)
@@ -475,8 +475,14 @@ ext4_xattr_release_block(handle_t *handle, struct inode *inode,
                         struct buffer_head *bh)
 {
        struct mb_cache_entry *ce = NULL;
+       int error = 0;
 
        ce = mb_cache_entry_get(ext4_xattr_cache, bh->b_bdev, bh->b_blocknr);
+       error = ext4_journal_get_write_access(handle, bh);
+       if (error)
+               goto out;
+
+       lock_buffer(bh);
        if (BHDR(bh)->h_refcount == cpu_to_le32(1)) {
                ea_bdebug(bh, "refcount now=0; freeing");
                if (ce)
@@ -485,21 +491,21 @@ ext4_xattr_release_block(handle_t *handle, struct inode *inode,
                get_bh(bh);
                ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
        } else {
-               if (ext4_journal_get_write_access(handle, bh) == 0) {
-                       lock_buffer(bh);
-                       BHDR(bh)->h_refcount = cpu_to_le32(
+               BHDR(bh)->h_refcount = cpu_to_le32(
                                le32_to_cpu(BHDR(bh)->h_refcount) - 1);
-                       ext4_journal_dirty_metadata(handle, bh);
-                       if (IS_SYNC(inode))
-                               handle->h_sync = 1;
-                       DQUOT_FREE_BLOCK(inode, 1);
-                       unlock_buffer(bh);
-                       ea_bdebug(bh, "refcount now=%d; releasing",
-                                 le32_to_cpu(BHDR(bh)->h_refcount));
-               }
+               error = ext4_journal_dirty_metadata(handle, bh);
+               if (IS_SYNC(inode))
+                       handle->h_sync = 1;
+               DQUOT_FREE_BLOCK(inode, 1);
+               ea_bdebug(bh, "refcount now=%d; releasing",
+                         le32_to_cpu(BHDR(bh)->h_refcount));
                if (ce)
                        mb_cache_entry_release(ce);
        }
+       unlock_buffer(bh);
+out:
+       ext4_std_error(inode->i_sb, error);
+       return;
 }
 
 struct ext4_xattr_info {
@@ -675,7 +681,7 @@ ext4_xattr_block_set(handle_t *handle, struct inode *inode,
        struct buffer_head *new_bh = NULL;
        struct ext4_xattr_search *s = &bs->s;
        struct mb_cache_entry *ce = NULL;
-       int error;
+       int error = 0;
 
 #define header(x) ((struct ext4_xattr_header *)(x))
 
@@ -684,16 +690,17 @@ ext4_xattr_block_set(handle_t *handle, struct inode *inode,
        if (s->base) {
                ce = mb_cache_entry_get(ext4_xattr_cache, bs->bh->b_bdev,
                                        bs->bh->b_blocknr);
+               error = ext4_journal_get_write_access(handle, bs->bh);
+               if (error)
+                       goto cleanup;
+               lock_buffer(bs->bh);
+
                if (header(s->base)->h_refcount == cpu_to_le32(1)) {
                        if (ce) {
                                mb_cache_entry_free(ce);
                                ce = NULL;
                        }
                        ea_bdebug(bs->bh, "modifying in-place");
-                       error = ext4_journal_get_write_access(handle, bs->bh);
-                       if (error)
-                               goto cleanup;
-                       lock_buffer(bs->bh);
                        error = ext4_xattr_set_entry(i, s);
                        if (!error) {
                                if (!IS_LAST_ENTRY(s->first))
@@ -713,6 +720,8 @@ ext4_xattr_block_set(handle_t *handle, struct inode *inode,
                } else {
                        int offset = (char *)s->here - bs->bh->b_data;
 
+                       unlock_buffer(bs->bh);
+                       jbd2_journal_release_buffer(handle, bs->bh);
                        if (ce) {
                                mb_cache_entry_release(ce);
                                ce = NULL;