2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "xfs_shared.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_inode.h"
30 #include "xfs_alloc.h"
31 #include "xfs_attr_remote.h"
32 #include "xfs_trans.h"
33 #include "xfs_inode_item.h"
36 #include "xfs_attr_leaf.h"
37 #include "xfs_error.h"
38 #include "xfs_quota.h"
39 #include "xfs_trace.h"
43 * Look at all the extents for this logical region,
44 * invalidate any buffers that are incore/in transactions.
47 xfs_attr3_leaf_freextent(
48 struct xfs_trans **trans,
53 struct xfs_bmbt_irec map;
63 * Roll through the "value", invalidating the attribute value's
70 * Try to remember where we decided to put the value.
73 error = xfs_bmapi_read(dp, (xfs_fileoff_t)tblkno, tblkcnt,
74 &map, &nmap, XFS_BMAPI_ATTRFORK);
79 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
82 * If it's a hole, these are already unmapped
83 * so there's nothing to invalidate.
85 if (map.br_startblock != HOLESTARTBLOCK) {
87 dblkno = XFS_FSB_TO_DADDR(dp->i_mount,
89 dblkcnt = XFS_FSB_TO_BB(dp->i_mount,
91 bp = xfs_trans_get_buf(*trans,
92 dp->i_mount->m_ddev_targp,
96 xfs_trans_binval(*trans, bp);
98 * Roll to next transaction.
100 error = xfs_trans_roll(trans, dp);
105 tblkno += map.br_blockcount;
106 tblkcnt -= map.br_blockcount;
113 * Invalidate all of the "remote" value regions pointed to by a particular
115 * Note that we must release the lock on the buffer so that we are not
116 * caught holding something that the logging code wants to flush to disk.
119 xfs_attr3_leaf_inactive(
120 struct xfs_trans **trans,
121 struct xfs_inode *dp,
124 struct xfs_attr_leafblock *leaf;
125 struct xfs_attr3_icleaf_hdr ichdr;
126 struct xfs_attr_leaf_entry *entry;
127 struct xfs_attr_leaf_name_remote *name_rmt;
128 struct xfs_attr_inactive_list *list;
129 struct xfs_attr_inactive_list *lp;
137 xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
140 * Count the number of "remote" value extents.
143 entry = xfs_attr3_leaf_entryp(leaf);
144 for (i = 0; i < ichdr.count; entry++, i++) {
145 if (be16_to_cpu(entry->nameidx) &&
146 ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
147 name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
148 if (name_rmt->valueblk)
154 * If there are no "remote" values, we're done.
157 xfs_trans_brelse(*trans, bp);
162 * Allocate storage for a list of all the "remote" value extents.
164 size = count * sizeof(xfs_attr_inactive_list_t);
165 list = kmem_alloc(size, KM_SLEEP);
168 * Identify each of the "remote" value extents.
171 entry = xfs_attr3_leaf_entryp(leaf);
172 for (i = 0; i < ichdr.count; entry++, i++) {
173 if (be16_to_cpu(entry->nameidx) &&
174 ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
175 name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
176 if (name_rmt->valueblk) {
177 lp->valueblk = be32_to_cpu(name_rmt->valueblk);
178 lp->valuelen = xfs_attr3_rmt_blocks(dp->i_mount,
179 be32_to_cpu(name_rmt->valuelen));
184 xfs_trans_brelse(*trans, bp); /* unlock for trans. in freextent() */
187 * Invalidate each of the "remote" value extents.
190 for (lp = list, i = 0; i < count; i++, lp++) {
191 tmp = xfs_attr3_leaf_freextent(trans, dp,
192 lp->valueblk, lp->valuelen);
195 error = tmp; /* save only the 1st errno */
203 * Recurse (gasp!) through the attribute nodes until we find leaves.
204 * We're doing a depth-first traversal in order to invalidate everything.
207 xfs_attr3_node_inactive(
208 struct xfs_trans **trans,
209 struct xfs_inode *dp,
213 xfs_da_blkinfo_t *info;
214 xfs_da_intnode_t *node;
215 xfs_dablk_t child_fsb;
216 xfs_daddr_t parent_blkno, child_blkno;
218 struct xfs_buf *child_bp;
219 struct xfs_da_node_entry *btree;
220 struct xfs_da3_icnode_hdr ichdr;
223 * Since this code is recursive (gasp!) we must protect ourselves.
225 if (level > XFS_DA_NODE_MAXDEPTH) {
226 xfs_trans_brelse(*trans, bp); /* no locks for later trans */
231 dp->d_ops->node_hdr_from_disk(&ichdr, node);
232 parent_blkno = bp->b_bn;
234 xfs_trans_brelse(*trans, bp);
237 btree = dp->d_ops->node_tree_p(node);
238 child_fsb = be32_to_cpu(btree[0].before);
239 xfs_trans_brelse(*trans, bp); /* no locks for later trans */
242 * If this is the node level just above the leaves, simply loop
243 * over the leaves removing all of them. If this is higher up
244 * in the tree, recurse downward.
246 for (i = 0; i < ichdr.count; i++) {
248 * Read the subsidiary block to see what we have to work with.
249 * Don't do this in a transaction. This is a depth-first
250 * traversal of the tree so we may deal with many blocks
251 * before we come back to this one.
253 error = xfs_da3_node_read(*trans, dp, child_fsb, -2, &child_bp,
258 /* save for re-read later */
259 child_blkno = XFS_BUF_ADDR(child_bp);
262 * Invalidate the subtree, however we have to.
264 info = child_bp->b_addr;
265 switch (info->magic) {
266 case cpu_to_be16(XFS_DA_NODE_MAGIC):
267 case cpu_to_be16(XFS_DA3_NODE_MAGIC):
268 error = xfs_attr3_node_inactive(trans, dp,
269 child_bp, level + 1);
271 case cpu_to_be16(XFS_ATTR_LEAF_MAGIC):
272 case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC):
273 error = xfs_attr3_leaf_inactive(trans, dp,
278 xfs_trans_brelse(*trans, child_bp);
285 * Remove the subsidiary block from the cache
288 error = xfs_da_get_buf(*trans, dp, 0, child_blkno,
289 &child_bp, XFS_ATTR_FORK);
292 xfs_trans_binval(*trans, child_bp);
296 * If we're not done, re-read the parent to get the next
297 * child block number.
299 if (i + 1 < ichdr.count) {
300 error = xfs_da3_node_read(*trans, dp, 0, parent_blkno,
304 child_fsb = be32_to_cpu(btree[i + 1].before);
305 xfs_trans_brelse(*trans, bp);
308 * Atomically commit the whole invalidate stuff.
310 error = xfs_trans_roll(trans, dp);
319 * Indiscriminately delete the entire attribute fork
321 * Recurse (gasp!) through the attribute nodes until we find leaves.
322 * We're doing a depth-first traversal in order to invalidate everything.
325 xfs_attr3_root_inactive(
326 struct xfs_trans **trans,
327 struct xfs_inode *dp)
329 struct xfs_da_blkinfo *info;
335 * Read block 0 to see what we have to work with.
336 * We only get here if we have extents, since we remove
337 * the extents in reverse order the extent containing
338 * block 0 must still be there.
340 error = xfs_da3_node_read(*trans, dp, 0, -1, &bp, XFS_ATTR_FORK);
346 * Invalidate the tree, even if the "tree" is only a single leaf block.
347 * This is a depth-first traversal!
350 switch (info->magic) {
351 case cpu_to_be16(XFS_DA_NODE_MAGIC):
352 case cpu_to_be16(XFS_DA3_NODE_MAGIC):
353 error = xfs_attr3_node_inactive(trans, dp, bp, 1);
355 case cpu_to_be16(XFS_ATTR_LEAF_MAGIC):
356 case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC):
357 error = xfs_attr3_leaf_inactive(trans, dp, bp);
361 xfs_trans_brelse(*trans, bp);
368 * Invalidate the incore copy of the root block.
370 error = xfs_da_get_buf(*trans, dp, 0, blkno, &bp, XFS_ATTR_FORK);
373 xfs_trans_binval(*trans, bp); /* remove from cache */
375 * Commit the invalidate and start the next transaction.
377 error = xfs_trans_roll(trans, dp);
383 xfs_attr_inactive(xfs_inode_t *dp)
390 ASSERT(! XFS_NOT_DQATTACHED(mp, dp));
392 xfs_ilock(dp, XFS_ILOCK_SHARED);
393 if (!xfs_inode_hasattr(dp) ||
394 dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
395 xfs_iunlock(dp, XFS_ILOCK_SHARED);
398 xfs_iunlock(dp, XFS_ILOCK_SHARED);
401 * Start our first transaction of the day.
403 * All future transactions during this code must be "chained" off
404 * this one via the trans_dup() call. All transactions will contain
405 * the inode, and the inode will always be marked with trans_ihold().
406 * Since the inode will be locked in all transactions, we must log
407 * the inode in every transaction to let it float upward through
410 trans = xfs_trans_alloc(mp, XFS_TRANS_ATTRINVAL);
411 error = xfs_trans_reserve(trans, &M_RES(mp)->tr_attrinval, 0, 0);
413 xfs_trans_cancel(trans, 0);
416 xfs_ilock(dp, XFS_ILOCK_EXCL);
419 * No need to make quota reservations here. We expect to release some
420 * blocks, not allocate, in the common case.
422 xfs_trans_ijoin(trans, dp, 0);
425 * Decide on what work routines to call based on the inode size.
427 if (!xfs_inode_hasattr(dp) ||
428 dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
432 error = xfs_attr3_root_inactive(&trans, dp);
436 error = xfs_itruncate_extents(&trans, dp, XFS_ATTR_FORK, 0);
440 error = xfs_trans_commit(trans, XFS_TRANS_RELEASE_LOG_RES);
441 xfs_iunlock(dp, XFS_ILOCK_EXCL);
446 xfs_trans_cancel(trans, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
447 xfs_iunlock(dp, XFS_ILOCK_EXCL);