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