From 89c6c89af2ef41cb127c9694ef7783e585e96337 Mon Sep 17 00:00:00 2001 From: Dave Chinner Date: Mon, 30 Sep 2013 09:37:03 +1000 Subject: xfs: lockdep needs to know about 3 dquot-deep nesting Michael Semon reported that xfs/299 generated this lockdep warning: ============================================= [ INFO: possible recursive locking detected ] 3.12.0-rc2+ #2 Not tainted --------------------------------------------- touch/21072 is trying to acquire lock: (&xfs_dquot_other_class){+.+...}, at: [] xfs_trans_dqlockedjoin+0x57/0x64 but task is already holding lock: (&xfs_dquot_other_class){+.+...}, at: [] xfs_trans_dqlockedjoin+0x57/0x64 other info that might help us debug this: Possible unsafe locking scenario: CPU0 ---- lock(&xfs_dquot_other_class); lock(&xfs_dquot_other_class); *** DEADLOCK *** May be due to missing lock nesting notation 7 locks held by touch/21072: #0: (sb_writers#10){++++.+}, at: [] mnt_want_write+0x1e/0x3e #1: (&type->i_mutex_dir_key#4){+.+.+.}, at: [] do_last+0x245/0xe40 #2: (sb_internal#2){++++.+}, at: [] xfs_trans_alloc+0x1f/0x35 #3: (&(&ip->i_lock)->mr_lock/1){+.+...}, at: [] xfs_ilock+0x100/0x1f1 #4: (&(&ip->i_lock)->mr_lock){++++-.}, at: [] xfs_ilock_nowait+0x105/0x22f #5: (&dqp->q_qlock){+.+...}, at: [] xfs_trans_dqlockedjoin+0x57/0x64 #6: (&xfs_dquot_other_class){+.+...}, at: [] xfs_trans_dqlockedjoin+0x57/0x64 The lockdep annotation for dquot lock nesting only understands locking for user and "other" dquots, not user, group and quota dquots. Fix the annotations to match the locking heirarchy we now have. Reported-by: Michael L. Semon Signed-off-by: Dave Chinner Reviewed-by: Ben Myers Signed-off-by: Ben Myers (cherry picked from commit f112a049712a5c07de25d511c3c6587a2b1a015e) diff --git a/fs/xfs/xfs_dquot.c b/fs/xfs/xfs_dquot.c index 71520e6..1ee776d 100644 --- a/fs/xfs/xfs_dquot.c +++ b/fs/xfs/xfs_dquot.c @@ -64,7 +64,8 @@ int xfs_dqerror_mod = 33; struct kmem_zone *xfs_qm_dqtrxzone; static struct kmem_zone *xfs_qm_dqzone; -static struct lock_class_key xfs_dquot_other_class; +static struct lock_class_key xfs_dquot_group_class; +static struct lock_class_key xfs_dquot_project_class; /* * This is called to free all the memory associated with a dquot @@ -703,8 +704,20 @@ xfs_qm_dqread( * Make sure group quotas have a different lock class than user * quotas. */ - if (!(type & XFS_DQ_USER)) - lockdep_set_class(&dqp->q_qlock, &xfs_dquot_other_class); + switch (type) { + case XFS_DQ_USER: + /* uses the default lock class */ + break; + case XFS_DQ_GROUP: + lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class); + break; + case XFS_DQ_PROJ: + lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class); + break; + default: + ASSERT(0); + break; + } XFS_STATS_INC(xs_qm_dquot); -- cgit v0.10.2 From 6d313498f035abc9d8ad3a1b3295f133bfab9638 Mon Sep 17 00:00:00 2001 From: Dave Chinner Date: Mon, 30 Sep 2013 09:37:04 +1000 Subject: xfs: dirent dtype presence is dependent on directory magic numbers The determination of whether a directory entry contains a dtype field originally was dependent on the filesystem having CRCs enabled. This meant that the format for dtype beign enabled could be determined by checking the directory block magic number rather than doing a feature bit check. This was useful in that it meant that we didn't need to pass a struct xfs_mount around to functions that were already supplied with a directory block header. Unfortunately, the introduction of dtype fields into the v4 structure via a feature bit meant this "use the directory block magic number" method of discriminating the dirent entry sizes is broken. Hence we need to convert the places that use magic number checks to use feature bit checks so that they work correctly and not by chance. The current code works on v4 filesystems only because the dirent size roundup covers the extra byte needed by the dtype field in the places where this problem occurs. Signed-off-by: Dave Chinner Reviewed-by: Ben Myers Signed-off-by: Ben Myers (cherry picked from commit 367993e7c6428cb7617ab7653d61dca54e2fdede) diff --git a/fs/xfs/xfs_dir2_block.c b/fs/xfs/xfs_dir2_block.c index 0957aa9..12dad18 100644 --- a/fs/xfs/xfs_dir2_block.c +++ b/fs/xfs/xfs_dir2_block.c @@ -1158,7 +1158,7 @@ xfs_dir2_sf_to_block( /* * Create entry for . */ - dep = xfs_dir3_data_dot_entry_p(hdr); + dep = xfs_dir3_data_dot_entry_p(mp, hdr); dep->inumber = cpu_to_be64(dp->i_ino); dep->namelen = 1; dep->name[0] = '.'; @@ -1172,7 +1172,7 @@ xfs_dir2_sf_to_block( /* * Create entry for .. */ - dep = xfs_dir3_data_dotdot_entry_p(hdr); + dep = xfs_dir3_data_dotdot_entry_p(mp, hdr); dep->inumber = cpu_to_be64(xfs_dir2_sf_get_parent_ino(sfp)); dep->namelen = 2; dep->name[0] = dep->name[1] = '.'; @@ -1183,7 +1183,7 @@ xfs_dir2_sf_to_block( blp[1].hashval = cpu_to_be32(xfs_dir_hash_dotdot); blp[1].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp, (char *)dep - (char *)hdr)); - offset = xfs_dir3_data_first_offset(hdr); + offset = xfs_dir3_data_first_offset(mp); /* * Loop over existing entries, stuff them in. */ diff --git a/fs/xfs/xfs_dir2_format.h b/fs/xfs/xfs_dir2_format.h index a0961a6..9cf6738 100644 --- a/fs/xfs/xfs_dir2_format.h +++ b/fs/xfs/xfs_dir2_format.h @@ -497,69 +497,58 @@ xfs_dir3_data_unused_p(struct xfs_dir2_data_hdr *hdr) /* * Offsets of . and .. in data space (always block 0) * - * The macros are used for shortform directories as they have no headers to read - * the magic number out of. Shortform directories need to know the size of the - * data block header because the sfe embeds the block offset of the entry into - * it so that it doesn't change when format conversion occurs. Bad Things Happen - * if we don't follow this rule. - * * XXX: there is scope for significant optimisation of the logic here. Right * now we are checking for "dir3 format" over and over again. Ideally we should * only do it once for each operation. */ -#define XFS_DIR3_DATA_DOT_OFFSET(mp) \ - xfs_dir3_data_hdr_size(xfs_sb_version_hascrc(&(mp)->m_sb)) -#define XFS_DIR3_DATA_DOTDOT_OFFSET(mp) \ - (XFS_DIR3_DATA_DOT_OFFSET(mp) + xfs_dir3_data_entsize(mp, 1)) -#define XFS_DIR3_DATA_FIRST_OFFSET(mp) \ - (XFS_DIR3_DATA_DOTDOT_OFFSET(mp) + xfs_dir3_data_entsize(mp, 2)) - static inline xfs_dir2_data_aoff_t -xfs_dir3_data_dot_offset(struct xfs_dir2_data_hdr *hdr) +xfs_dir3_data_dot_offset(struct xfs_mount *mp) { - return xfs_dir3_data_entry_offset(hdr); + return xfs_dir3_data_hdr_size(xfs_sb_version_hascrc(&mp->m_sb)); } static inline xfs_dir2_data_aoff_t -xfs_dir3_data_dotdot_offset(struct xfs_dir2_data_hdr *hdr) +xfs_dir3_data_dotdot_offset(struct xfs_mount *mp) { - bool dir3 = hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) || - hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC); - return xfs_dir3_data_dot_offset(hdr) + - __xfs_dir3_data_entsize(dir3, 1); + return xfs_dir3_data_dot_offset(mp) + + xfs_dir3_data_entsize(mp, 1); } static inline xfs_dir2_data_aoff_t -xfs_dir3_data_first_offset(struct xfs_dir2_data_hdr *hdr) +xfs_dir3_data_first_offset(struct xfs_mount *mp) { - bool dir3 = hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) || - hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC); - return xfs_dir3_data_dotdot_offset(hdr) + - __xfs_dir3_data_entsize(dir3, 2); + return xfs_dir3_data_dotdot_offset(mp) + + xfs_dir3_data_entsize(mp, 2); } /* * location of . and .. in data space (always block 0) */ static inline struct xfs_dir2_data_entry * -xfs_dir3_data_dot_entry_p(struct xfs_dir2_data_hdr *hdr) +xfs_dir3_data_dot_entry_p( + struct xfs_mount *mp, + struct xfs_dir2_data_hdr *hdr) { return (struct xfs_dir2_data_entry *) - ((char *)hdr + xfs_dir3_data_dot_offset(hdr)); + ((char *)hdr + xfs_dir3_data_dot_offset(mp)); } static inline struct xfs_dir2_data_entry * -xfs_dir3_data_dotdot_entry_p(struct xfs_dir2_data_hdr *hdr) +xfs_dir3_data_dotdot_entry_p( + struct xfs_mount *mp, + struct xfs_dir2_data_hdr *hdr) { return (struct xfs_dir2_data_entry *) - ((char *)hdr + xfs_dir3_data_dotdot_offset(hdr)); + ((char *)hdr + xfs_dir3_data_dotdot_offset(mp)); } static inline struct xfs_dir2_data_entry * -xfs_dir3_data_first_entry_p(struct xfs_dir2_data_hdr *hdr) +xfs_dir3_data_first_entry_p( + struct xfs_mount *mp, + struct xfs_dir2_data_hdr *hdr) { return (struct xfs_dir2_data_entry *) - ((char *)hdr + xfs_dir3_data_first_offset(hdr)); + ((char *)hdr + xfs_dir3_data_first_offset(mp)); } /* diff --git a/fs/xfs/xfs_dir2_readdir.c b/fs/xfs/xfs_dir2_readdir.c index 8993ec1..8f84153 100644 --- a/fs/xfs/xfs_dir2_readdir.c +++ b/fs/xfs/xfs_dir2_readdir.c @@ -119,9 +119,9 @@ xfs_dir2_sf_getdents( * mp->m_dirdatablk. */ dot_offset = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk, - XFS_DIR3_DATA_DOT_OFFSET(mp)); + xfs_dir3_data_dot_offset(mp)); dotdot_offset = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk, - XFS_DIR3_DATA_DOTDOT_OFFSET(mp)); + xfs_dir3_data_dotdot_offset(mp)); /* * Put . entry unless we're starting past it. diff --git a/fs/xfs/xfs_dir2_sf.c b/fs/xfs/xfs_dir2_sf.c index bb6e284..3ef6d40 100644 --- a/fs/xfs/xfs_dir2_sf.c +++ b/fs/xfs/xfs_dir2_sf.c @@ -557,7 +557,7 @@ xfs_dir2_sf_addname_hard( * to insert the new entry. * If it's going to end up at the end then oldsfep will point there. */ - for (offset = XFS_DIR3_DATA_FIRST_OFFSET(mp), + for (offset = xfs_dir3_data_first_offset(mp), oldsfep = xfs_dir2_sf_firstentry(oldsfp), add_datasize = xfs_dir3_data_entsize(mp, args->namelen), eof = (char *)oldsfep == &buf[old_isize]; @@ -640,7 +640,7 @@ xfs_dir2_sf_addname_pick( sfp = (xfs_dir2_sf_hdr_t *)dp->i_df.if_u1.if_data; size = xfs_dir3_data_entsize(mp, args->namelen); - offset = XFS_DIR3_DATA_FIRST_OFFSET(mp); + offset = xfs_dir3_data_first_offset(mp); sfep = xfs_dir2_sf_firstentry(sfp); holefit = 0; /* @@ -713,7 +713,7 @@ xfs_dir2_sf_check( mp = dp->i_mount; sfp = (xfs_dir2_sf_hdr_t *)dp->i_df.if_u1.if_data; - offset = XFS_DIR3_DATA_FIRST_OFFSET(mp); + offset = xfs_dir3_data_first_offset(mp); ino = xfs_dir2_sf_get_parent_ino(sfp); i8count = ino > XFS_DIR2_MAX_SHORT_INUM; -- cgit v0.10.2 From 9b3b77fe661875f19ed748b67fb1eeb57d602b7e Mon Sep 17 00:00:00 2001 From: "tinguely@sgi.com" Date: Fri, 27 Sep 2013 09:00:55 -0500 Subject: xfs: fix memory leak in xlog_recover_add_to_trans Free the memory in error path of xlog_recover_add_to_trans(). Normally this memory is freed in recovery pass2, but is leaked in the error path. Signed-off-by: Mark Tinguely Reviewed-by: Eric Sandeen Signed-off-by: Ben Myers (cherry picked from commit 519ccb81ac1c8e3e4eed294acf93be00b43dcad6) diff --git a/fs/xfs/xfs_log_recover.c b/fs/xfs/xfs_log_recover.c index cc17987..4324058 100644 --- a/fs/xfs/xfs_log_recover.c +++ b/fs/xfs/xfs_log_recover.c @@ -1585,6 +1585,7 @@ xlog_recover_add_to_trans( "bad number of regions (%d) in inode log format", in_f->ilf_size); ASSERT(0); + free(ptr); return XFS_ERROR(EIO); } -- cgit v0.10.2 From b2a42f78ab475f4730300b0e9568bc3b2587d112 Mon Sep 17 00:00:00 2001 From: Thierry Reding Date: Tue, 1 Oct 2013 16:47:53 +0200 Subject: xfs: Use kmem_free() instead of free() This fixes a build failure caused by calling the free() function which does not exist in the Linux kernel. Signed-off-by: Thierry Reding Reviewed-by: Mark Tinguely Signed-off-by: Ben Myers (cherry picked from commit aaaae98022efa4f3c31042f1fdf9e7a0c5f04663) diff --git a/fs/xfs/xfs_log_recover.c b/fs/xfs/xfs_log_recover.c index 4324058..3979749 100644 --- a/fs/xfs/xfs_log_recover.c +++ b/fs/xfs/xfs_log_recover.c @@ -1585,7 +1585,7 @@ xlog_recover_add_to_trans( "bad number of regions (%d) in inode log format", in_f->ilf_size); ASSERT(0); - free(ptr); + kmem_free(ptr); return XFS_ERROR(EIO); } -- cgit v0.10.2