]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - fs/xfs/xfs_attr_inactive.c
ceph: fix up non-directory creation in SGID directories
[mirror_ubuntu-jammy-kernel.git] / fs / xfs / xfs_attr_inactive.c
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.h"
19 #include "xfs_attr_remote.h"
20 #include "xfs_trans.h"
21 #include "xfs_bmap.h"
22 #include "xfs_attr_leaf.h"
23 #include "xfs_quota.h"
24 #include "xfs_dir2.h"
25 #include "xfs_error.h"
26
27 /*
28 * Invalidate any incore buffers associated with this remote attribute value
29 * extent. We never log remote attribute value buffers, which means that they
30 * won't be attached to a transaction and are therefore safe to mark stale.
31 * The actual bunmapi will be taken care of later.
32 */
33 STATIC int
34 xfs_attr3_rmt_stale(
35 struct xfs_inode *dp,
36 xfs_dablk_t blkno,
37 int blkcnt)
38 {
39 struct xfs_bmbt_irec map;
40 int nmap;
41 int error;
42
43 /*
44 * Roll through the "value", invalidating the attribute value's
45 * blocks.
46 */
47 while (blkcnt > 0) {
48 /*
49 * Try to remember where we decided to put the value.
50 */
51 nmap = 1;
52 error = xfs_bmapi_read(dp, (xfs_fileoff_t)blkno, blkcnt,
53 &map, &nmap, XFS_BMAPI_ATTRFORK);
54 if (error)
55 return error;
56 if (XFS_IS_CORRUPT(dp->i_mount, nmap != 1))
57 return -EFSCORRUPTED;
58
59 /*
60 * Mark any incore buffers for the remote value as stale. We
61 * never log remote attr value buffers, so the buffer should be
62 * easy to kill.
63 */
64 error = xfs_attr_rmtval_stale(dp, &map, 0);
65 if (error)
66 return error;
67
68 blkno += map.br_blockcount;
69 blkcnt -= map.br_blockcount;
70 }
71
72 return 0;
73 }
74
75 /*
76 * Invalidate all of the "remote" value regions pointed to by a particular
77 * leaf block.
78 * Note that we must release the lock on the buffer so that we are not
79 * caught holding something that the logging code wants to flush to disk.
80 */
81 STATIC int
82 xfs_attr3_leaf_inactive(
83 struct xfs_trans **trans,
84 struct xfs_inode *dp,
85 struct xfs_buf *bp)
86 {
87 struct xfs_attr3_icleaf_hdr ichdr;
88 struct xfs_mount *mp = bp->b_mount;
89 struct xfs_attr_leafblock *leaf = bp->b_addr;
90 struct xfs_attr_leaf_entry *entry;
91 struct xfs_attr_leaf_name_remote *name_rmt;
92 int error = 0;
93 int i;
94
95 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
96
97 /*
98 * Find the remote value extents for this leaf and invalidate their
99 * incore buffers.
100 */
101 entry = xfs_attr3_leaf_entryp(leaf);
102 for (i = 0; i < ichdr.count; entry++, i++) {
103 int blkcnt;
104
105 if (!entry->nameidx || (entry->flags & XFS_ATTR_LOCAL))
106 continue;
107
108 name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
109 if (!name_rmt->valueblk)
110 continue;
111
112 blkcnt = xfs_attr3_rmt_blocks(dp->i_mount,
113 be32_to_cpu(name_rmt->valuelen));
114 error = xfs_attr3_rmt_stale(dp,
115 be32_to_cpu(name_rmt->valueblk), blkcnt);
116 if (error)
117 goto err;
118 }
119
120 xfs_trans_brelse(*trans, bp);
121 err:
122 return error;
123 }
124
125 /*
126 * Recurse (gasp!) through the attribute nodes until we find leaves.
127 * We're doing a depth-first traversal in order to invalidate everything.
128 */
129 STATIC int
130 xfs_attr3_node_inactive(
131 struct xfs_trans **trans,
132 struct xfs_inode *dp,
133 struct xfs_buf *bp,
134 int level)
135 {
136 struct xfs_mount *mp = dp->i_mount;
137 struct xfs_da_blkinfo *info;
138 xfs_dablk_t child_fsb;
139 xfs_daddr_t parent_blkno, child_blkno;
140 struct xfs_buf *child_bp;
141 struct xfs_da3_icnode_hdr ichdr;
142 int error, i;
143
144 /*
145 * Since this code is recursive (gasp!) we must protect ourselves.
146 */
147 if (level > XFS_DA_NODE_MAXDEPTH) {
148 xfs_buf_mark_corrupt(bp);
149 xfs_trans_brelse(*trans, bp); /* no locks for later trans */
150 return -EFSCORRUPTED;
151 }
152
153 xfs_da3_node_hdr_from_disk(dp->i_mount, &ichdr, bp->b_addr);
154 parent_blkno = xfs_buf_daddr(bp);
155 if (!ichdr.count) {
156 xfs_trans_brelse(*trans, bp);
157 return 0;
158 }
159 child_fsb = be32_to_cpu(ichdr.btree[0].before);
160 xfs_trans_brelse(*trans, bp); /* no locks for later trans */
161
162 /*
163 * If this is the node level just above the leaves, simply loop
164 * over the leaves removing all of them. If this is higher up
165 * in the tree, recurse downward.
166 */
167 for (i = 0; i < ichdr.count; i++) {
168 /*
169 * Read the subsidiary block to see what we have to work with.
170 * Don't do this in a transaction. This is a depth-first
171 * traversal of the tree so we may deal with many blocks
172 * before we come back to this one.
173 */
174 error = xfs_da3_node_read(*trans, dp, child_fsb, &child_bp,
175 XFS_ATTR_FORK);
176 if (error)
177 return error;
178
179 /* save for re-read later */
180 child_blkno = xfs_buf_daddr(child_bp);
181
182 /*
183 * Invalidate the subtree, however we have to.
184 */
185 info = child_bp->b_addr;
186 switch (info->magic) {
187 case cpu_to_be16(XFS_DA_NODE_MAGIC):
188 case cpu_to_be16(XFS_DA3_NODE_MAGIC):
189 error = xfs_attr3_node_inactive(trans, dp, child_bp,
190 level + 1);
191 break;
192 case cpu_to_be16(XFS_ATTR_LEAF_MAGIC):
193 case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC):
194 error = xfs_attr3_leaf_inactive(trans, dp, child_bp);
195 break;
196 default:
197 xfs_buf_mark_corrupt(child_bp);
198 xfs_trans_brelse(*trans, child_bp);
199 error = -EFSCORRUPTED;
200 break;
201 }
202 if (error)
203 return error;
204
205 /*
206 * Remove the subsidiary block from the cache and from the log.
207 */
208 error = xfs_trans_get_buf(*trans, mp->m_ddev_targp,
209 child_blkno,
210 XFS_FSB_TO_BB(mp, mp->m_attr_geo->fsbcount), 0,
211 &child_bp);
212 if (error)
213 return error;
214 error = bp->b_error;
215 if (error) {
216 xfs_trans_brelse(*trans, child_bp);
217 return error;
218 }
219 xfs_trans_binval(*trans, child_bp);
220
221 /*
222 * If we're not done, re-read the parent to get the next
223 * child block number.
224 */
225 if (i + 1 < ichdr.count) {
226 struct xfs_da3_icnode_hdr phdr;
227
228 error = xfs_da3_node_read_mapped(*trans, dp,
229 parent_blkno, &bp, XFS_ATTR_FORK);
230 if (error)
231 return error;
232 xfs_da3_node_hdr_from_disk(dp->i_mount, &phdr,
233 bp->b_addr);
234 child_fsb = be32_to_cpu(phdr.btree[i + 1].before);
235 xfs_trans_brelse(*trans, bp);
236 }
237 /*
238 * Atomically commit the whole invalidate stuff.
239 */
240 error = xfs_trans_roll_inode(trans, dp);
241 if (error)
242 return error;
243 }
244
245 return 0;
246 }
247
248 /*
249 * Indiscriminately delete the entire attribute fork
250 *
251 * Recurse (gasp!) through the attribute nodes until we find leaves.
252 * We're doing a depth-first traversal in order to invalidate everything.
253 */
254 static int
255 xfs_attr3_root_inactive(
256 struct xfs_trans **trans,
257 struct xfs_inode *dp)
258 {
259 struct xfs_mount *mp = dp->i_mount;
260 struct xfs_da_blkinfo *info;
261 struct xfs_buf *bp;
262 xfs_daddr_t blkno;
263 int error;
264
265 /*
266 * Read block 0 to see what we have to work with.
267 * We only get here if we have extents, since we remove
268 * the extents in reverse order the extent containing
269 * block 0 must still be there.
270 */
271 error = xfs_da3_node_read(*trans, dp, 0, &bp, XFS_ATTR_FORK);
272 if (error)
273 return error;
274 blkno = xfs_buf_daddr(bp);
275
276 /*
277 * Invalidate the tree, even if the "tree" is only a single leaf block.
278 * This is a depth-first traversal!
279 */
280 info = bp->b_addr;
281 switch (info->magic) {
282 case cpu_to_be16(XFS_DA_NODE_MAGIC):
283 case cpu_to_be16(XFS_DA3_NODE_MAGIC):
284 error = xfs_attr3_node_inactive(trans, dp, bp, 1);
285 break;
286 case cpu_to_be16(XFS_ATTR_LEAF_MAGIC):
287 case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC):
288 error = xfs_attr3_leaf_inactive(trans, dp, bp);
289 break;
290 default:
291 error = -EFSCORRUPTED;
292 xfs_buf_mark_corrupt(bp);
293 xfs_trans_brelse(*trans, bp);
294 break;
295 }
296 if (error)
297 return error;
298
299 /*
300 * Invalidate the incore copy of the root block.
301 */
302 error = xfs_trans_get_buf(*trans, mp->m_ddev_targp, blkno,
303 XFS_FSB_TO_BB(mp, mp->m_attr_geo->fsbcount), 0, &bp);
304 if (error)
305 return error;
306 error = bp->b_error;
307 if (error) {
308 xfs_trans_brelse(*trans, bp);
309 return error;
310 }
311 xfs_trans_binval(*trans, bp); /* remove from cache */
312 /*
313 * Commit the invalidate and start the next transaction.
314 */
315 error = xfs_trans_roll_inode(trans, dp);
316
317 return error;
318 }
319
320 /*
321 * xfs_attr_inactive kills all traces of an attribute fork on an inode. It
322 * removes both the on-disk and in-memory inode fork. Note that this also has to
323 * handle the condition of inodes without attributes but with an attribute fork
324 * configured, so we can't use xfs_inode_hasattr() here.
325 *
326 * The in-memory attribute fork is removed even on error.
327 */
328 int
329 xfs_attr_inactive(
330 struct xfs_inode *dp)
331 {
332 struct xfs_trans *trans;
333 struct xfs_mount *mp;
334 int lock_mode = XFS_ILOCK_SHARED;
335 int error = 0;
336
337 mp = dp->i_mount;
338 ASSERT(! XFS_NOT_DQATTACHED(mp, dp));
339
340 xfs_ilock(dp, lock_mode);
341 if (!XFS_IFORK_Q(dp))
342 goto out_destroy_fork;
343 xfs_iunlock(dp, lock_mode);
344
345 lock_mode = 0;
346
347 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_attrinval, 0, 0, 0, &trans);
348 if (error)
349 goto out_destroy_fork;
350
351 lock_mode = XFS_ILOCK_EXCL;
352 xfs_ilock(dp, lock_mode);
353
354 if (!XFS_IFORK_Q(dp))
355 goto out_cancel;
356
357 /*
358 * No need to make quota reservations here. We expect to release some
359 * blocks, not allocate, in the common case.
360 */
361 xfs_trans_ijoin(trans, dp, 0);
362
363 /*
364 * Invalidate and truncate the attribute fork extents. Make sure the
365 * fork actually has attributes as otherwise the invalidation has no
366 * blocks to read and returns an error. In this case, just do the fork
367 * removal below.
368 */
369 if (xfs_inode_hasattr(dp) &&
370 dp->i_afp->if_format != XFS_DINODE_FMT_LOCAL) {
371 error = xfs_attr3_root_inactive(&trans, dp);
372 if (error)
373 goto out_cancel;
374
375 error = xfs_itruncate_extents(&trans, dp, XFS_ATTR_FORK, 0);
376 if (error)
377 goto out_cancel;
378 }
379
380 /* Reset the attribute fork - this also destroys the in-core fork */
381 xfs_attr_fork_remove(dp, trans);
382
383 error = xfs_trans_commit(trans);
384 xfs_iunlock(dp, lock_mode);
385 return error;
386
387 out_cancel:
388 xfs_trans_cancel(trans);
389 out_destroy_fork:
390 /* kill the in-core attr fork before we drop the inode lock */
391 if (dp->i_afp) {
392 xfs_idestroy_fork(dp->i_afp);
393 kmem_cache_free(xfs_ifork_zone, dp->i_afp);
394 dp->i_afp = NULL;
395 }
396 if (lock_mode)
397 xfs_iunlock(dp, lock_mode);
398 return error;
399 }