aboutsummaryrefslogtreecommitdiffstats
path: root/fs/xfs
diff options
context:
space:
mode:
authorIngo Molnar <mingo@elte.hu>2009-09-04 08:44:16 -0400
committerIngo Molnar <mingo@elte.hu>2009-09-04 08:44:16 -0400
commit695a461296e5df148c99ac087b9e1cb380f4db15 (patch)
tree951893036fdc0b7bae0e17bc739ac8ffe909781d /fs/xfs
parentc7084b35eb1a4d3353a501508baf9d3d82822c93 (diff)
parent2b681fafcc50fea6304ed418667c9d04282acb73 (diff)
Merge branch 'amd-iommu/2.6.32' of git://git.kernel.org/pub/scm/linux/kernel/git/joro/linux-2.6-iommu into core/iommu
Diffstat (limited to 'fs/xfs')
-rw-r--r--fs/xfs/linux-2.6/xfs_aops.c8
-rw-r--r--fs/xfs/linux-2.6/xfs_buf.c2
-rw-r--r--fs/xfs/linux-2.6/xfs_iops.c4
-rw-r--r--fs/xfs/linux-2.6/xfs_sync.c13
-rw-r--r--fs/xfs/linux-2.6/xfs_sync.h1
-rw-r--r--fs/xfs/xfs_attr.c8
-rw-r--r--fs/xfs/xfs_bmap.c2
-rw-r--r--fs/xfs/xfs_btree.c4
-rw-r--r--fs/xfs/xfs_da_btree.c6
-rw-r--r--fs/xfs/xfs_dir2.c2
-rw-r--r--fs/xfs/xfs_fsops.c20
-rw-r--r--fs/xfs/xfs_iget.c253
-rw-r--r--fs/xfs/xfs_inode.c10
-rw-r--r--fs/xfs/xfs_inode.h17
-rw-r--r--fs/xfs/xfs_log.c2
-rw-r--r--fs/xfs/xfs_vnodeops.c4
16 files changed, 193 insertions, 163 deletions
diff --git a/fs/xfs/linux-2.6/xfs_aops.c b/fs/xfs/linux-2.6/xfs_aops.c
index 7ec89fc05b2b..aecf2519db76 100644
--- a/fs/xfs/linux-2.6/xfs_aops.c
+++ b/fs/xfs/linux-2.6/xfs_aops.c
@@ -1268,6 +1268,14 @@ xfs_vm_writepage(
1268 if (!page_has_buffers(page)) 1268 if (!page_has_buffers(page))
1269 create_empty_buffers(page, 1 << inode->i_blkbits, 0); 1269 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
1270 1270
1271
1272 /*
1273 * VM calculation for nr_to_write seems off. Bump it way
1274 * up, this gets simple streaming writes zippy again.
1275 * To be reviewed again after Jens' writeback changes.
1276 */
1277 wbc->nr_to_write *= 4;
1278
1271 /* 1279 /*
1272 * Convert delayed allocate, unwritten or unmapped space 1280 * Convert delayed allocate, unwritten or unmapped space
1273 * to real space and flush out to disk. 1281 * to real space and flush out to disk.
diff --git a/fs/xfs/linux-2.6/xfs_buf.c b/fs/xfs/linux-2.6/xfs_buf.c
index 0c93c7ef3d18..965df1227d64 100644
--- a/fs/xfs/linux-2.6/xfs_buf.c
+++ b/fs/xfs/linux-2.6/xfs_buf.c
@@ -770,7 +770,7 @@ xfs_buf_associate_memory(
770 bp->b_pages = NULL; 770 bp->b_pages = NULL;
771 bp->b_addr = mem; 771 bp->b_addr = mem;
772 772
773 rval = _xfs_buf_get_pages(bp, page_count, 0); 773 rval = _xfs_buf_get_pages(bp, page_count, XBF_DONT_BLOCK);
774 if (rval) 774 if (rval)
775 return rval; 775 return rval;
776 776
diff --git a/fs/xfs/linux-2.6/xfs_iops.c b/fs/xfs/linux-2.6/xfs_iops.c
index 58973bb46038..8070b34cc287 100644
--- a/fs/xfs/linux-2.6/xfs_iops.c
+++ b/fs/xfs/linux-2.6/xfs_iops.c
@@ -680,8 +680,8 @@ xfs_vn_fiemap(
680 else 680 else
681 bm.bmv_length = BTOBB(length); 681 bm.bmv_length = BTOBB(length);
682 682
683 /* our formatter will tell xfs_getbmap when to stop. */ 683 /* We add one because in getbmap world count includes the header */
684 bm.bmv_count = MAXEXTNUM; 684 bm.bmv_count = fieinfo->fi_extents_max + 1;
685 bm.bmv_iflags = BMV_IF_PREALLOC; 685 bm.bmv_iflags = BMV_IF_PREALLOC;
686 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) 686 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
687 bm.bmv_iflags |= BMV_IF_ATTRFORK; 687 bm.bmv_iflags |= BMV_IF_ATTRFORK;
diff --git a/fs/xfs/linux-2.6/xfs_sync.c b/fs/xfs/linux-2.6/xfs_sync.c
index b619d6b8ca43..98ef624d9baf 100644
--- a/fs/xfs/linux-2.6/xfs_sync.c
+++ b/fs/xfs/linux-2.6/xfs_sync.c
@@ -708,6 +708,16 @@ xfs_reclaim_inode(
708 return 0; 708 return 0;
709} 709}
710 710
711void
712__xfs_inode_set_reclaim_tag(
713 struct xfs_perag *pag,
714 struct xfs_inode *ip)
715{
716 radix_tree_tag_set(&pag->pag_ici_root,
717 XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino),
718 XFS_ICI_RECLAIM_TAG);
719}
720
711/* 721/*
712 * We set the inode flag atomically with the radix tree tag. 722 * We set the inode flag atomically with the radix tree tag.
713 * Once we get tag lookups on the radix tree, this inode flag 723 * Once we get tag lookups on the radix tree, this inode flag
@@ -722,8 +732,7 @@ xfs_inode_set_reclaim_tag(
722 732
723 read_lock(&pag->pag_ici_lock); 733 read_lock(&pag->pag_ici_lock);
724 spin_lock(&ip->i_flags_lock); 734 spin_lock(&ip->i_flags_lock);
725 radix_tree_tag_set(&pag->pag_ici_root, 735 __xfs_inode_set_reclaim_tag(pag, ip);
726 XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG);
727 __xfs_iflags_set(ip, XFS_IRECLAIMABLE); 736 __xfs_iflags_set(ip, XFS_IRECLAIMABLE);
728 spin_unlock(&ip->i_flags_lock); 737 spin_unlock(&ip->i_flags_lock);
729 read_unlock(&pag->pag_ici_lock); 738 read_unlock(&pag->pag_ici_lock);
diff --git a/fs/xfs/linux-2.6/xfs_sync.h b/fs/xfs/linux-2.6/xfs_sync.h
index 2a10301c99c7..59120602588a 100644
--- a/fs/xfs/linux-2.6/xfs_sync.h
+++ b/fs/xfs/linux-2.6/xfs_sync.h
@@ -48,6 +48,7 @@ int xfs_reclaim_inode(struct xfs_inode *ip, int locked, int sync_mode);
48int xfs_reclaim_inodes(struct xfs_mount *mp, int mode); 48int xfs_reclaim_inodes(struct xfs_mount *mp, int mode);
49 49
50void xfs_inode_set_reclaim_tag(struct xfs_inode *ip); 50void xfs_inode_set_reclaim_tag(struct xfs_inode *ip);
51void __xfs_inode_set_reclaim_tag(struct xfs_perag *pag, struct xfs_inode *ip);
51void xfs_inode_clear_reclaim_tag(struct xfs_inode *ip); 52void xfs_inode_clear_reclaim_tag(struct xfs_inode *ip);
52void __xfs_inode_clear_reclaim_tag(struct xfs_mount *mp, struct xfs_perag *pag, 53void __xfs_inode_clear_reclaim_tag(struct xfs_mount *mp, struct xfs_perag *pag,
53 struct xfs_inode *ip); 54 struct xfs_inode *ip);
diff --git a/fs/xfs/xfs_attr.c b/fs/xfs/xfs_attr.c
index db15feb906ff..4ece1906bd41 100644
--- a/fs/xfs/xfs_attr.c
+++ b/fs/xfs/xfs_attr.c
@@ -2010,7 +2010,9 @@ xfs_attr_rmtval_get(xfs_da_args_t *args)
2010 dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock); 2010 dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
2011 blkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount); 2011 blkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
2012 error = xfs_read_buf(mp, mp->m_ddev_targp, dblkno, 2012 error = xfs_read_buf(mp, mp->m_ddev_targp, dblkno,
2013 blkcnt, XFS_BUF_LOCK, &bp); 2013 blkcnt,
2014 XFS_BUF_LOCK | XBF_DONT_BLOCK,
2015 &bp);
2014 if (error) 2016 if (error)
2015 return(error); 2017 return(error);
2016 2018
@@ -2141,8 +2143,8 @@ xfs_attr_rmtval_set(xfs_da_args_t *args)
2141 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock), 2143 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2142 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount); 2144 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2143 2145
2144 bp = xfs_buf_get_flags(mp->m_ddev_targp, dblkno, 2146 bp = xfs_buf_get_flags(mp->m_ddev_targp, dblkno, blkcnt,
2145 blkcnt, XFS_BUF_LOCK); 2147 XFS_BUF_LOCK | XBF_DONT_BLOCK);
2146 ASSERT(bp); 2148 ASSERT(bp);
2147 ASSERT(!XFS_BUF_GETERROR(bp)); 2149 ASSERT(!XFS_BUF_GETERROR(bp));
2148 2150
diff --git a/fs/xfs/xfs_bmap.c b/fs/xfs/xfs_bmap.c
index 7928b9983c1d..8ee5b5a76a2a 100644
--- a/fs/xfs/xfs_bmap.c
+++ b/fs/xfs/xfs_bmap.c
@@ -6009,7 +6009,7 @@ xfs_getbmap(
6009 */ 6009 */
6010 error = ENOMEM; 6010 error = ENOMEM;
6011 subnex = 16; 6011 subnex = 16;
6012 map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL); 6012 map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
6013 if (!map) 6013 if (!map)
6014 goto out_unlock_ilock; 6014 goto out_unlock_ilock;
6015 6015
diff --git a/fs/xfs/xfs_btree.c b/fs/xfs/xfs_btree.c
index e9df99574829..26717388acf5 100644
--- a/fs/xfs/xfs_btree.c
+++ b/fs/xfs/xfs_btree.c
@@ -120,8 +120,8 @@ xfs_btree_check_sblock(
120 XFS_RANDOM_BTREE_CHECK_SBLOCK))) { 120 XFS_RANDOM_BTREE_CHECK_SBLOCK))) {
121 if (bp) 121 if (bp)
122 xfs_buftrace("SBTREE ERROR", bp); 122 xfs_buftrace("SBTREE ERROR", bp);
123 XFS_ERROR_REPORT("xfs_btree_check_sblock", XFS_ERRLEVEL_LOW, 123 XFS_CORRUPTION_ERROR("xfs_btree_check_sblock",
124 cur->bc_mp); 124 XFS_ERRLEVEL_LOW, cur->bc_mp, block);
125 return XFS_ERROR(EFSCORRUPTED); 125 return XFS_ERROR(EFSCORRUPTED);
126 } 126 }
127 return 0; 127 return 0;
diff --git a/fs/xfs/xfs_da_btree.c b/fs/xfs/xfs_da_btree.c
index 9ff6e57a5075..2847bbc1c534 100644
--- a/fs/xfs/xfs_da_btree.c
+++ b/fs/xfs/xfs_da_btree.c
@@ -2201,7 +2201,7 @@ kmem_zone_t *xfs_dabuf_zone; /* dabuf zone */
2201xfs_da_state_t * 2201xfs_da_state_t *
2202xfs_da_state_alloc(void) 2202xfs_da_state_alloc(void)
2203{ 2203{
2204 return kmem_zone_zalloc(xfs_da_state_zone, KM_SLEEP); 2204 return kmem_zone_zalloc(xfs_da_state_zone, KM_NOFS);
2205} 2205}
2206 2206
2207/* 2207/*
@@ -2261,9 +2261,9 @@ xfs_da_buf_make(int nbuf, xfs_buf_t **bps, inst_t *ra)
2261 int off; 2261 int off;
2262 2262
2263 if (nbuf == 1) 2263 if (nbuf == 1)
2264 dabuf = kmem_zone_alloc(xfs_dabuf_zone, KM_SLEEP); 2264 dabuf = kmem_zone_alloc(xfs_dabuf_zone, KM_NOFS);
2265 else 2265 else
2266 dabuf = kmem_alloc(XFS_DA_BUF_SIZE(nbuf), KM_SLEEP); 2266 dabuf = kmem_alloc(XFS_DA_BUF_SIZE(nbuf), KM_NOFS);
2267 dabuf->dirty = 0; 2267 dabuf->dirty = 0;
2268#ifdef XFS_DABUF_DEBUG 2268#ifdef XFS_DABUF_DEBUG
2269 dabuf->ra = ra; 2269 dabuf->ra = ra;
diff --git a/fs/xfs/xfs_dir2.c b/fs/xfs/xfs_dir2.c
index c657bec6d951..bb1d58eb3982 100644
--- a/fs/xfs/xfs_dir2.c
+++ b/fs/xfs/xfs_dir2.c
@@ -256,7 +256,7 @@ xfs_dir_cilookup_result(
256 !(args->op_flags & XFS_DA_OP_CILOOKUP)) 256 !(args->op_flags & XFS_DA_OP_CILOOKUP))
257 return EEXIST; 257 return EEXIST;
258 258
259 args->value = kmem_alloc(len, KM_MAYFAIL); 259 args->value = kmem_alloc(len, KM_NOFS | KM_MAYFAIL);
260 if (!args->value) 260 if (!args->value)
261 return ENOMEM; 261 return ENOMEM;
262 262
diff --git a/fs/xfs/xfs_fsops.c b/fs/xfs/xfs_fsops.c
index cbd451bb4848..2d0b3e1da9e6 100644
--- a/fs/xfs/xfs_fsops.c
+++ b/fs/xfs/xfs_fsops.c
@@ -167,17 +167,25 @@ xfs_growfs_data_private(
167 new = nb - mp->m_sb.sb_dblocks; 167 new = nb - mp->m_sb.sb_dblocks;
168 oagcount = mp->m_sb.sb_agcount; 168 oagcount = mp->m_sb.sb_agcount;
169 if (nagcount > oagcount) { 169 if (nagcount > oagcount) {
170 void *new_perag, *old_perag;
171
170 xfs_filestream_flush(mp); 172 xfs_filestream_flush(mp);
173
174 new_perag = kmem_zalloc(sizeof(xfs_perag_t) * nagcount,
175 KM_MAYFAIL);
176 if (!new_perag)
177 return XFS_ERROR(ENOMEM);
178
171 down_write(&mp->m_peraglock); 179 down_write(&mp->m_peraglock);
172 mp->m_perag = kmem_realloc(mp->m_perag, 180 memcpy(new_perag, mp->m_perag, sizeof(xfs_perag_t) * oagcount);
173 sizeof(xfs_perag_t) * nagcount, 181 old_perag = mp->m_perag;
174 sizeof(xfs_perag_t) * oagcount, 182 mp->m_perag = new_perag;
175 KM_SLEEP); 183
176 memset(&mp->m_perag[oagcount], 0,
177 (nagcount - oagcount) * sizeof(xfs_perag_t));
178 mp->m_flags |= XFS_MOUNT_32BITINODES; 184 mp->m_flags |= XFS_MOUNT_32BITINODES;
179 nagimax = xfs_initialize_perag(mp, nagcount); 185 nagimax = xfs_initialize_perag(mp, nagcount);
180 up_write(&mp->m_peraglock); 186 up_write(&mp->m_peraglock);
187
188 kmem_free(old_perag);
181 } 189 }
182 tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFS); 190 tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFS);
183 tp->t_flags |= XFS_TRANS_RESERVE; 191 tp->t_flags |= XFS_TRANS_RESERVE;
diff --git a/fs/xfs/xfs_iget.c b/fs/xfs/xfs_iget.c
index 5fcec6f020a7..ecbf8b4d2e2e 100644
--- a/fs/xfs/xfs_iget.c
+++ b/fs/xfs/xfs_iget.c
@@ -64,6 +64,10 @@ xfs_inode_alloc(
64 ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP); 64 ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP);
65 if (!ip) 65 if (!ip)
66 return NULL; 66 return NULL;
67 if (inode_init_always(mp->m_super, VFS_I(ip))) {
68 kmem_zone_free(xfs_inode_zone, ip);
69 return NULL;
70 }
67 71
68 ASSERT(atomic_read(&ip->i_iocount) == 0); 72 ASSERT(atomic_read(&ip->i_iocount) == 0);
69 ASSERT(atomic_read(&ip->i_pincount) == 0); 73 ASSERT(atomic_read(&ip->i_pincount) == 0);
@@ -105,17 +109,6 @@ xfs_inode_alloc(
105#ifdef XFS_DIR2_TRACE 109#ifdef XFS_DIR2_TRACE
106 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_NOFS); 110 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_NOFS);
107#endif 111#endif
108 /*
109 * Now initialise the VFS inode. We do this after the xfs_inode
110 * initialisation as internal failures will result in ->destroy_inode
111 * being called and that will pass down through the reclaim path and
112 * free the XFS inode. This path requires the XFS inode to already be
113 * initialised. Hence if this call fails, the xfs_inode has already
114 * been freed and we should not reference it at all in the error
115 * handling.
116 */
117 if (!inode_init_always(mp->m_super, VFS_I(ip)))
118 return NULL;
119 112
120 /* prevent anyone from using this yet */ 113 /* prevent anyone from using this yet */
121 VFS_I(ip)->i_state = I_NEW|I_LOCK; 114 VFS_I(ip)->i_state = I_NEW|I_LOCK;
@@ -123,6 +116,71 @@ xfs_inode_alloc(
123 return ip; 116 return ip;
124} 117}
125 118
119STATIC void
120xfs_inode_free(
121 struct xfs_inode *ip)
122{
123 switch (ip->i_d.di_mode & S_IFMT) {
124 case S_IFREG:
125 case S_IFDIR:
126 case S_IFLNK:
127 xfs_idestroy_fork(ip, XFS_DATA_FORK);
128 break;
129 }
130
131 if (ip->i_afp)
132 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
133
134#ifdef XFS_INODE_TRACE
135 ktrace_free(ip->i_trace);
136#endif
137#ifdef XFS_BMAP_TRACE
138 ktrace_free(ip->i_xtrace);
139#endif
140#ifdef XFS_BTREE_TRACE
141 ktrace_free(ip->i_btrace);
142#endif
143#ifdef XFS_RW_TRACE
144 ktrace_free(ip->i_rwtrace);
145#endif
146#ifdef XFS_ILOCK_TRACE
147 ktrace_free(ip->i_lock_trace);
148#endif
149#ifdef XFS_DIR2_TRACE
150 ktrace_free(ip->i_dir_trace);
151#endif
152
153 if (ip->i_itemp) {
154 /*
155 * Only if we are shutting down the fs will we see an
156 * inode still in the AIL. If it is there, we should remove
157 * it to prevent a use-after-free from occurring.
158 */
159 xfs_log_item_t *lip = &ip->i_itemp->ili_item;
160 struct xfs_ail *ailp = lip->li_ailp;
161
162 ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
163 XFS_FORCED_SHUTDOWN(ip->i_mount));
164 if (lip->li_flags & XFS_LI_IN_AIL) {
165 spin_lock(&ailp->xa_lock);
166 if (lip->li_flags & XFS_LI_IN_AIL)
167 xfs_trans_ail_delete(ailp, lip);
168 else
169 spin_unlock(&ailp->xa_lock);
170 }
171 xfs_inode_item_destroy(ip);
172 ip->i_itemp = NULL;
173 }
174
175 /* asserts to verify all state is correct here */
176 ASSERT(atomic_read(&ip->i_iocount) == 0);
177 ASSERT(atomic_read(&ip->i_pincount) == 0);
178 ASSERT(!spin_is_locked(&ip->i_flags_lock));
179 ASSERT(completion_done(&ip->i_flush));
180
181 kmem_zone_free(xfs_inode_zone, ip);
182}
183
126/* 184/*
127 * Check the validity of the inode we just found it the cache 185 * Check the validity of the inode we just found it the cache
128 */ 186 */
@@ -133,80 +191,82 @@ xfs_iget_cache_hit(
133 int flags, 191 int flags,
134 int lock_flags) __releases(pag->pag_ici_lock) 192 int lock_flags) __releases(pag->pag_ici_lock)
135{ 193{
194 struct inode *inode = VFS_I(ip);
136 struct xfs_mount *mp = ip->i_mount; 195 struct xfs_mount *mp = ip->i_mount;
137 int error = EAGAIN; 196 int error;
197
198 spin_lock(&ip->i_flags_lock);
138 199
139 /* 200 /*
140 * If INEW is set this inode is being set up 201 * If we are racing with another cache hit that is currently
141 * If IRECLAIM is set this inode is being torn down 202 * instantiating this inode or currently recycling it out of
142 * Pause and try again. 203 * reclaimabe state, wait for the initialisation to complete
204 * before continuing.
205 *
206 * XXX(hch): eventually we should do something equivalent to
207 * wait_on_inode to wait for these flags to be cleared
208 * instead of polling for it.
143 */ 209 */
144 if (xfs_iflags_test(ip, (XFS_INEW|XFS_IRECLAIM))) { 210 if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) {
145 XFS_STATS_INC(xs_ig_frecycle); 211 XFS_STATS_INC(xs_ig_frecycle);
212 error = EAGAIN;
146 goto out_error; 213 goto out_error;
147 } 214 }
148 215
149 /* If IRECLAIMABLE is set, we've torn down the vfs inode part */ 216 /*
150 if (xfs_iflags_test(ip, XFS_IRECLAIMABLE)) { 217 * If lookup is racing with unlink return an error immediately.
151 218 */
152 /* 219 if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) {
153 * If lookup is racing with unlink, then we should return an 220 error = ENOENT;
154 * error immediately so we don't remove it from the reclaim 221 goto out_error;
155 * list and potentially leak the inode. 222 }
156 */
157 if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
158 error = ENOENT;
159 goto out_error;
160 }
161 223
224 /*
225 * If IRECLAIMABLE is set, we've torn down the VFS inode already.
226 * Need to carefully get it back into useable state.
227 */
228 if (ip->i_flags & XFS_IRECLAIMABLE) {
162 xfs_itrace_exit_tag(ip, "xfs_iget.alloc"); 229 xfs_itrace_exit_tag(ip, "xfs_iget.alloc");
163 230
164 /* 231 /*
165 * We need to re-initialise the VFS inode as it has been 232 * We need to set XFS_INEW atomically with clearing the
166 * 'freed' by the VFS. Do this here so we can deal with 233 * reclaimable tag so that we do have an indicator of the
167 * errors cleanly, then tag it so it can be set up correctly 234 * inode still being initialized.
168 * later.
169 */ 235 */
170 if (!inode_init_always(mp->m_super, VFS_I(ip))) { 236 ip->i_flags |= XFS_INEW;
171 error = ENOMEM; 237 ip->i_flags &= ~XFS_IRECLAIMABLE;
172 goto out_error; 238 __xfs_inode_clear_reclaim_tag(mp, pag, ip);
173 }
174 239
175 /* 240 spin_unlock(&ip->i_flags_lock);
176 * We must set the XFS_INEW flag before clearing the 241 read_unlock(&pag->pag_ici_lock);
177 * XFS_IRECLAIMABLE flag so that if a racing lookup does
178 * not find the XFS_IRECLAIMABLE above but has the igrab()
179 * below succeed we can safely check XFS_INEW to detect
180 * that this inode is still being initialised.
181 */
182 xfs_iflags_set(ip, XFS_INEW);
183 xfs_iflags_clear(ip, XFS_IRECLAIMABLE);
184 242
185 /* clear the radix tree reclaim flag as well. */ 243 error = -inode_init_always(mp->m_super, inode);
186 __xfs_inode_clear_reclaim_tag(mp, pag, ip); 244 if (error) {
187 } else if (!igrab(VFS_I(ip))) { 245 /*
246 * Re-initializing the inode failed, and we are in deep
247 * trouble. Try to re-add it to the reclaim list.
248 */
249 read_lock(&pag->pag_ici_lock);
250 spin_lock(&ip->i_flags_lock);
251
252 ip->i_flags &= ~XFS_INEW;
253 ip->i_flags |= XFS_IRECLAIMABLE;
254 __xfs_inode_set_reclaim_tag(pag, ip);
255 goto out_error;
256 }
257 inode->i_state = I_LOCK|I_NEW;
258 } else {
188 /* If the VFS inode is being torn down, pause and try again. */ 259 /* If the VFS inode is being torn down, pause and try again. */
189 XFS_STATS_INC(xs_ig_frecycle); 260 if (!igrab(inode)) {
190 goto out_error; 261 error = EAGAIN;
191 } else if (xfs_iflags_test(ip, XFS_INEW)) { 262 goto out_error;
192 /* 263 }
193 * We are racing with another cache hit that is
194 * currently recycling this inode out of the XFS_IRECLAIMABLE
195 * state. Wait for the initialisation to complete before
196 * continuing.
197 */
198 wait_on_inode(VFS_I(ip));
199 }
200 264
201 if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) { 265 /* We've got a live one. */
202 error = ENOENT; 266 spin_unlock(&ip->i_flags_lock);
203 iput(VFS_I(ip)); 267 read_unlock(&pag->pag_ici_lock);
204 goto out_error;
205 } 268 }
206 269
207 /* We've got a live one. */
208 read_unlock(&pag->pag_ici_lock);
209
210 if (lock_flags != 0) 270 if (lock_flags != 0)
211 xfs_ilock(ip, lock_flags); 271 xfs_ilock(ip, lock_flags);
212 272
@@ -216,6 +276,7 @@ xfs_iget_cache_hit(
216 return 0; 276 return 0;
217 277
218out_error: 278out_error:
279 spin_unlock(&ip->i_flags_lock);
219 read_unlock(&pag->pag_ici_lock); 280 read_unlock(&pag->pag_ici_lock);
220 return error; 281 return error;
221} 282}
@@ -299,7 +360,8 @@ out_preload_end:
299 if (lock_flags) 360 if (lock_flags)
300 xfs_iunlock(ip, lock_flags); 361 xfs_iunlock(ip, lock_flags);
301out_destroy: 362out_destroy:
302 xfs_destroy_inode(ip); 363 __destroy_inode(VFS_I(ip));
364 xfs_inode_free(ip);
303 return error; 365 return error;
304} 366}
305 367
@@ -504,62 +566,7 @@ xfs_ireclaim(
504 xfs_qm_dqdetach(ip); 566 xfs_qm_dqdetach(ip);
505 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); 567 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
506 568
507 switch (ip->i_d.di_mode & S_IFMT) { 569 xfs_inode_free(ip);
508 case S_IFREG:
509 case S_IFDIR:
510 case S_IFLNK:
511 xfs_idestroy_fork(ip, XFS_DATA_FORK);
512 break;
513 }
514
515 if (ip->i_afp)
516 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
517
518#ifdef XFS_INODE_TRACE
519 ktrace_free(ip->i_trace);
520#endif
521#ifdef XFS_BMAP_TRACE
522 ktrace_free(ip->i_xtrace);
523#endif
524#ifdef XFS_BTREE_TRACE
525 ktrace_free(ip->i_btrace);
526#endif
527#ifdef XFS_RW_TRACE
528 ktrace_free(ip->i_rwtrace);
529#endif
530#ifdef XFS_ILOCK_TRACE
531 ktrace_free(ip->i_lock_trace);
532#endif
533#ifdef XFS_DIR2_TRACE
534 ktrace_free(ip->i_dir_trace);
535#endif
536 if (ip->i_itemp) {
537 /*
538 * Only if we are shutting down the fs will we see an
539 * inode still in the AIL. If it is there, we should remove
540 * it to prevent a use-after-free from occurring.
541 */
542 xfs_log_item_t *lip = &ip->i_itemp->ili_item;
543 struct xfs_ail *ailp = lip->li_ailp;
544
545 ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
546 XFS_FORCED_SHUTDOWN(ip->i_mount));
547 if (lip->li_flags & XFS_LI_IN_AIL) {
548 spin_lock(&ailp->xa_lock);
549 if (lip->li_flags & XFS_LI_IN_AIL)
550 xfs_trans_ail_delete(ailp, lip);
551 else
552 spin_unlock(&ailp->xa_lock);
553 }
554 xfs_inode_item_destroy(ip);
555 ip->i_itemp = NULL;
556 }
557 /* asserts to verify all state is correct here */
558 ASSERT(atomic_read(&ip->i_iocount) == 0);
559 ASSERT(atomic_read(&ip->i_pincount) == 0);
560 ASSERT(!spin_is_locked(&ip->i_flags_lock));
561 ASSERT(completion_done(&ip->i_flush));
562 kmem_zone_free(xfs_inode_zone, ip);
563} 570}
564 571
565/* 572/*
diff --git a/fs/xfs/xfs_inode.c b/fs/xfs/xfs_inode.c
index 1f22d65fed0a..da428b3fe0f5 100644
--- a/fs/xfs/xfs_inode.c
+++ b/fs/xfs/xfs_inode.c
@@ -343,6 +343,16 @@ xfs_iformat(
343 return XFS_ERROR(EFSCORRUPTED); 343 return XFS_ERROR(EFSCORRUPTED);
344 } 344 }
345 345
346 if (unlikely((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) &&
347 !ip->i_mount->m_rtdev_targp)) {
348 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
349 "corrupt dinode %Lu, has realtime flag set.",
350 ip->i_ino);
351 XFS_CORRUPTION_ERROR("xfs_iformat(realtime)",
352 XFS_ERRLEVEL_LOW, ip->i_mount, dip);
353 return XFS_ERROR(EFSCORRUPTED);
354 }
355
346 switch (ip->i_d.di_mode & S_IFMT) { 356 switch (ip->i_d.di_mode & S_IFMT) {
347 case S_IFIFO: 357 case S_IFIFO:
348 case S_IFCHR: 358 case S_IFCHR:
diff --git a/fs/xfs/xfs_inode.h b/fs/xfs/xfs_inode.h
index 1804f866a71d..65f24a3cc992 100644
--- a/fs/xfs/xfs_inode.h
+++ b/fs/xfs/xfs_inode.h
@@ -310,23 +310,6 @@ static inline struct inode *VFS_I(struct xfs_inode *ip)
310} 310}
311 311
312/* 312/*
313 * Get rid of a partially initialized inode.
314 *
315 * We have to go through destroy_inode to make sure allocations
316 * from init_inode_always like the security data are undone.
317 *
318 * We mark the inode bad so that it takes the short cut in
319 * the reclaim path instead of going through the flush path
320 * which doesn't make sense for an inode that has never seen the
321 * light of day.
322 */
323static inline void xfs_destroy_inode(struct xfs_inode *ip)
324{
325 make_bad_inode(VFS_I(ip));
326 return destroy_inode(VFS_I(ip));
327}
328
329/*
330 * i_flags helper functions 313 * i_flags helper functions
331 */ 314 */
332static inline void 315static inline void
diff --git a/fs/xfs/xfs_log.c b/fs/xfs/xfs_log.c
index 3750f04ede0b..9dbdff3ea484 100644
--- a/fs/xfs/xfs_log.c
+++ b/fs/xfs/xfs_log.c
@@ -3180,7 +3180,7 @@ try_again:
3180STATIC void 3180STATIC void
3181xlog_state_want_sync(xlog_t *log, xlog_in_core_t *iclog) 3181xlog_state_want_sync(xlog_t *log, xlog_in_core_t *iclog)
3182{ 3182{
3183 ASSERT(spin_is_locked(&log->l_icloglock)); 3183 assert_spin_locked(&log->l_icloglock);
3184 3184
3185 if (iclog->ic_state == XLOG_STATE_ACTIVE) { 3185 if (iclog->ic_state == XLOG_STATE_ACTIVE) {
3186 xlog_state_switch_iclogs(log, iclog, 0); 3186 xlog_state_switch_iclogs(log, iclog, 0);
diff --git a/fs/xfs/xfs_vnodeops.c b/fs/xfs/xfs_vnodeops.c
index c4eca5ed5dab..492d75bae2bf 100644
--- a/fs/xfs/xfs_vnodeops.c
+++ b/fs/xfs/xfs_vnodeops.c
@@ -538,7 +538,9 @@ xfs_readlink_bmap(
538 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); 538 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
539 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); 539 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
540 540
541 bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0); 541 bp = xfs_buf_read_flags(mp->m_ddev_targp, d, BTOBB(byte_cnt),
542 XBF_LOCK | XBF_MAPPED |
543 XBF_DONT_BLOCK);
542 error = XFS_BUF_GETERROR(bp); 544 error = XFS_BUF_GETERROR(bp);
543 if (error) { 545 if (error) {
544 xfs_ioerror_alert("xfs_readlink", 546 xfs_ioerror_alert("xfs_readlink",