1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * Copyright (c) 2013 Red Hat, Inc.
5 * All Rights Reserved.
6 */
7 #include "xfs.h"
8 #include "xfs_fs.h"
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_bit.h"
14 #include "xfs_mount.h"
15 #include "xfs_da_format.h"
16 #include "xfs_da_btree.h"
17 #include "xfs_inode.h"
18 #include "xfs_attr_remote.h"
19 #include "xfs_trans.h"
20 #include "xfs_bmap.h"
21 #include "xfs_attr.h"
22 #include "xfs_attr_leaf.h"
23 #include "xfs_quota.h"
24 #include "xfs_dir2.h"
25
26 /*
27 * Look at all the extents for this logical region,
28 * invalidate any buffers that are incore/in transactions.
29 */
30 STATIC int
31 xfs_attr3_leaf_freextent(
32 struct xfs_trans **trans,
33 struct xfs_inode *dp,
34 xfs_dablk_t blkno,
35 int blkcnt)
36 {
37 struct xfs_bmbt_irec map;
38 struct xfs_buf *bp;
39 xfs_dablk_t tblkno;
40 xfs_daddr_t dblkno;
41 int tblkcnt;
42 int dblkcnt;
43 int nmap;
44 int error;
45
46 /*
47 * Roll through the "value", invalidating the attribute value's
48 * blocks.
49 */
50 tblkno = blkno;
51 tblkcnt = blkcnt;
52 while (tblkcnt > 0) {
53 /*
54 * Try to remember where we decided to put the value.
55 */
56 nmap = 1;
57 error = xfs_bmapi_read(dp, (xfs_fileoff_t)tblkno, tblkcnt,
58 &map, &nmap, XFS_BMAPI_ATTRFORK);
59 if (error) {
60 return error;
61 }
62 ASSERT(nmap == 1);
63 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
64
65 /*
66 * If it's a hole, these are already unmapped
67 * so there's nothing to invalidate.
68 */
69 if (map.br_startblock != HOLESTARTBLOCK) {
70
71 dblkno = XFS_FSB_TO_DADDR(dp->i_mount,
72 map.br_startblock);
73 dblkcnt = XFS_FSB_TO_BB(dp->i_mount,
74 map.br_blockcount);
75 bp = xfs_trans_get_buf(*trans,
76 dp->i_mount->m_ddev_targp,
77 dblkno, dblkcnt, 0);
78 if (!bp)
79 return -ENOMEM;
80 xfs_trans_binval(*trans, bp);
81 /*
82 * Roll to next transaction.
83 */
84 error = xfs_trans_roll_inode(trans, dp);
85 if (error)
86 return error;
87 }
88
89 tblkno += map.br_blockcount;
90 tblkcnt -= map.br_blockcount;
91 }
92
93 return 0;
94 }
95
96 /*
97 * Invalidate all of the "remote" value regions pointed to by a particular
98 * leaf block.
99 * Note that we must release the lock on the buffer so that we are not
100 * caught holding something that the logging code wants to flush to disk.
101 */
102 STATIC int
103 xfs_attr3_leaf_inactive(
104 struct xfs_trans **trans,
105 struct xfs_inode *dp,
106 struct xfs_buf *bp)
107 {
108 struct xfs_attr_leafblock *leaf;
109 struct xfs_attr3_icleaf_hdr ichdr;
110 struct xfs_attr_leaf_entry *entry;
111 struct xfs_attr_leaf_name_remote *name_rmt;
112 struct xfs_attr_inactive_list *list;
113 struct xfs_attr_inactive_list *lp;
114 int error;
115 int count;
116 int size;
117 int tmp;
118 int i;
119 struct xfs_mount *mp = bp->b_mount;
120
121 leaf = bp->b_addr;
122 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
123
124 /*
125 * Count the number of "remote" value extents.
126 */
127 count = 0;
128 entry = xfs_attr3_leaf_entryp(leaf);
129 for (i = 0; i < ichdr.count; entry++, i++) {
130 if (be16_to_cpu(entry->nameidx) &&
131 ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
132 name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
133 if (name_rmt->valueblk)
134 count++;
135 }
136 }
137
138 /*
139 * If there are no "remote" values, we're done.
140 */
141 if (count == 0) {
142 xfs_trans_brelse(*trans, bp);
143 return 0;
144 }
145
146 /*
147 * Allocate storage for a list of all the "remote" value extents.
148 */
149 size = count * sizeof(xfs_attr_inactive_list_t);
150 list = kmem_alloc(size, 0);
151
152 /*
153 * Identify each of the "remote" value extents.
154 */
155 lp = list;
156 entry = xfs_attr3_leaf_entryp(leaf);
157 for (i = 0; i < ichdr.count; entry++, i++) {
158 if (be16_to_cpu(entry->nameidx) &&
159 ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
160 name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
161 if (name_rmt->valueblk) {
162 lp->valueblk = be32_to_cpu(name_rmt->valueblk);
163 lp->valuelen = xfs_attr3_rmt_blocks(dp->i_mount,
164 be32_to_cpu(name_rmt->valuelen));
165 lp++;
166 }
167 }
168 }
169 xfs_trans_brelse(*trans, bp); /* unlock for trans. in freextent() */
170
171 /*
172 * Invalidate each of the "remote" value extents.
173 */
174 error = 0;
175 for (lp = list, i = 0; i < count; i++, lp++) {
176 tmp = xfs_attr3_leaf_freextent(trans, dp,
177 lp->valueblk, lp->valuelen);
178
179 if (error == 0)
180 error = tmp; /* save only the 1st errno */
181 }
182
183 kmem_free(list);
184 return error;
185 }
186
187 /*
188 * Recurse (gasp!) through the attribute nodes until we find leaves.
189 * We're doing a depth-first traversal in order to invalidate everything.
190 */
191 STATIC int
192 xfs_attr3_node_inactive(
193 struct xfs_trans **trans,
194 struct xfs_inode *dp,
195 struct xfs_buf *bp,
196 int level)
197 {
198 xfs_da_blkinfo_t *info;
199 xfs_da_intnode_t *node;
200 xfs_dablk_t child_fsb;
201 xfs_daddr_t parent_blkno, child_blkno;
202 int error, i;
203 struct xfs_buf *child_bp;
204 struct xfs_da_node_entry *btree;
205 struct xfs_da3_icnode_hdr ichdr;
206
207 /*
208 * Since this code is recursive (gasp!) we must protect ourselves.
209 */
210 if (level > XFS_DA_NODE_MAXDEPTH) {
211 xfs_trans_brelse(*trans, bp); /* no locks for later trans */
212 return -EIO;
213 }
214
215 node = bp->b_addr;
216 dp->d_ops->node_hdr_from_disk(&ichdr, node);
217 parent_blkno = bp->b_bn;
218 if (!ichdr.count) {
219 xfs_trans_brelse(*trans, bp);
220 return 0;
221 }
222 btree = dp->d_ops->node_tree_p(node);
223 child_fsb = be32_to_cpu(btree[0].before);
224 xfs_trans_brelse(*trans, bp); /* no locks for later trans */
225
226 /*
227 * If this is the node level just above the leaves, simply loop
228 * over the leaves removing all of them. If this is higher up
229 * in the tree, recurse downward.
230 */
231 for (i = 0; i < ichdr.count; i++) {
232 /*
233 * Read the subsidiary block to see what we have to work with.
234 * Don't do this in a transaction. This is a depth-first
235 * traversal of the tree so we may deal with many blocks
236 * before we come back to this one.
237 */
238 error = xfs_da3_node_read(*trans, dp, child_fsb, -1, &child_bp,
239 XFS_ATTR_FORK);
240 if (error)
241 return error;
242
243 /* save for re-read later */
244 child_blkno = XFS_BUF_ADDR(child_bp);
245
246 /*
247 * Invalidate the subtree, however we have to.
248 */
249 info = child_bp->b_addr;
250 switch (info->magic) {
251 case cpu_to_be16(XFS_DA_NODE_MAGIC):
252 case cpu_to_be16(XFS_DA3_NODE_MAGIC):
253 error = xfs_attr3_node_inactive(trans, dp, child_bp,
254 level + 1);
255 break;
256 case cpu_to_be16(XFS_ATTR_LEAF_MAGIC):
257 case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC):
258 error = xfs_attr3_leaf_inactive(trans, dp, child_bp);
259 break;
260 default:
261 error = -EIO;
262 xfs_trans_brelse(*trans, child_bp);
263 break;
264 }
265 if (error)
266 return error;
267
268 /*
269 * Remove the subsidiary block from the cache and from the log.
270 */
271 error = xfs_da_get_buf(*trans, dp, 0, child_blkno, &child_bp,
272 XFS_ATTR_FORK);
273 if (error)
274 return error;
275 xfs_trans_binval(*trans, child_bp);
276
277 /*
278 * If we're not done, re-read the parent to get the next
279 * child block number.
280 */
281 if (i + 1 < ichdr.count) {
282 error = xfs_da3_node_read(*trans, dp, 0, parent_blkno,
283 &bp, XFS_ATTR_FORK);
284 if (error)
285 return error;
286 node = bp->b_addr;
287 btree = dp->d_ops->node_tree_p(node);
288 child_fsb = be32_to_cpu(btree[i + 1].before);
289 xfs_trans_brelse(*trans, bp);
290 }
291 /*
292 * Atomically commit the whole invalidate stuff.
293 */
294 error = xfs_trans_roll_inode(trans, dp);
295 if (error)
296 return error;
297 }
298
299 return 0;
300 }
301
302 /*
303 * Indiscriminately delete the entire attribute fork
304 *
305 * Recurse (gasp!) through the attribute nodes until we find leaves.
306 * We're doing a depth-first traversal in order to invalidate everything.
307 */
308 static int
309 xfs_attr3_root_inactive(
310 struct xfs_trans **trans,
311 struct xfs_inode *dp)
312 {
313 struct xfs_da_blkinfo *info;
314 struct xfs_buf *bp;
315 xfs_daddr_t blkno;
316 int error;
317
318 /*
319 * Read block 0 to see what we have to work with.
320 * We only get here if we have extents, since we remove
321 * the extents in reverse order the extent containing
322 * block 0 must still be there.
323 */
324 error = xfs_da3_node_read(*trans, dp, 0, -1, &bp, XFS_ATTR_FORK);
325 if (error)
326 return error;
327 blkno = bp->b_bn;
328
329 /*
330 * Invalidate the tree, even if the "tree" is only a single leaf block.
331 * This is a depth-first traversal!
332 */
333 info = bp->b_addr;
334 switch (info->magic) {
335 case cpu_to_be16(XFS_DA_NODE_MAGIC):
336 case cpu_to_be16(XFS_DA3_NODE_MAGIC):
337 error = xfs_attr3_node_inactive(trans, dp, bp, 1);
338 break;
339 case cpu_to_be16(XFS_ATTR_LEAF_MAGIC):
340 case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC):
341 error = xfs_attr3_leaf_inactive(trans, dp, bp);
342 break;
343 default:
344 error = -EIO;
345 xfs_trans_brelse(*trans, bp);
346 break;
347 }
348 if (error)
349 return error;
350
351 /*
352 * Invalidate the incore copy of the root block.
353 */
354 error = xfs_da_get_buf(*trans, dp, 0, blkno, &bp, XFS_ATTR_FORK);
355 if (error)
356 return error;
357 xfs_trans_binval(*trans, bp); /* remove from cache */
358 /*
359 * Commit the invalidate and start the next transaction.
360 */
361 error = xfs_trans_roll_inode(trans, dp);
362
363 return error;
364 }
365
366 /*
367 * xfs_attr_inactive kills all traces of an attribute fork on an inode. It
368 * removes both the on-disk and in-memory inode fork. Note that this also has to
369 * handle the condition of inodes without attributes but with an attribute fork
370 * configured, so we can't use xfs_inode_hasattr() here.
371 *
372 * The in-memory attribute fork is removed even on error.
373 */
374 int
375 xfs_attr_inactive(
376 struct xfs_inode *dp)
377 {
378 struct xfs_trans *trans;
379 struct xfs_mount *mp;
380 int lock_mode = XFS_ILOCK_SHARED;
381 int error = 0;
382
383 mp = dp->i_mount;
384 ASSERT(! XFS_NOT_DQATTACHED(mp, dp));
385
386 xfs_ilock(dp, lock_mode);
387 if (!XFS_IFORK_Q(dp))
388 goto out_destroy_fork;
389 xfs_iunlock(dp, lock_mode);
390
391 lock_mode = 0;
392
393 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_attrinval, 0, 0, 0, &trans);
394 if (error)
395 goto out_destroy_fork;
396
397 lock_mode = XFS_ILOCK_EXCL;
398 xfs_ilock(dp, lock_mode);
399
400 if (!XFS_IFORK_Q(dp))
401 goto out_cancel;
402
403 /*
404 * No need to make quota reservations here. We expect to release some
405 * blocks, not allocate, in the common case.
406 */
407 xfs_trans_ijoin(trans, dp, 0);
408
409 /*
410 * Invalidate and truncate the attribute fork extents. Make sure the
411 * fork actually has attributes as otherwise the invalidation has no
412 * blocks to read and returns an error. In this case, just do the fork
413 * removal below.
414 */
415 if (xfs_inode_hasattr(dp) &&
416 dp->i_d.di_aformat != XFS_DINODE_FMT_LOCAL) {
417 error = xfs_attr3_root_inactive(&trans, dp);
418 if (error)
419 goto out_cancel;
420
421 error = xfs_itruncate_extents(&trans, dp, XFS_ATTR_FORK, 0);
422 if (error)
423 goto out_cancel;
424 }
425
426 /* Reset the attribute fork - this also destroys the in-core fork */
427 xfs_attr_fork_remove(dp, trans);
428
429 error = xfs_trans_commit(trans);
430 xfs_iunlock(dp, lock_mode);
431 return error;
432
433 out_cancel:
434 xfs_trans_cancel(trans);
435 out_destroy_fork:
436 /* kill the in-core attr fork before we drop the inode lock */
437 if (dp->i_afp)
438 xfs_idestroy_fork(dp, XFS_ATTR_FORK);
439 if (lock_mode)
440 xfs_iunlock(dp, lock_mode);
441 return error;
442 }