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 | bp = NULL; |
162 | |
163 | /* |
164 | * If this is the node level just above the leaves, simply loop |
165 | * over the leaves removing all of them. If this is higher up |
166 | * in the tree, recurse downward. |
167 | */ |
168 | for (i = 0; i < ichdr.count; i++) { |
169 | /* |
170 | * Read the subsidiary block to see what we have to work with. |
171 | * Don't do this in a transaction. This is a depth-first |
172 | * traversal of the tree so we may deal with many blocks |
173 | * before we come back to this one. |
174 | */ |
175 | error = xfs_da3_node_read(*trans, dp, child_fsb, &child_bp, |
176 | XFS_ATTR_FORK); |
177 | if (error) |
178 | return error; |
179 | |
180 | /* save for re-read later */ |
181 | child_blkno = xfs_buf_daddr(bp: child_bp); |
182 | |
183 | /* |
184 | * Invalidate the subtree, however we have to. |
185 | */ |
186 | info = child_bp->b_addr; |
187 | switch (info->magic) { |
188 | case cpu_to_be16(XFS_DA_NODE_MAGIC): |
189 | case cpu_to_be16(XFS_DA3_NODE_MAGIC): |
190 | error = xfs_attr3_node_inactive(trans, dp, bp: child_bp, |
191 | level: level + 1); |
192 | break; |
193 | case cpu_to_be16(XFS_ATTR_LEAF_MAGIC): |
194 | case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC): |
195 | error = xfs_attr3_leaf_inactive(trans, dp, bp: child_bp); |
196 | break; |
197 | default: |
198 | xfs_buf_mark_corrupt(child_bp); |
199 | xfs_trans_brelse(*trans, child_bp); |
200 | error = -EFSCORRUPTED; |
201 | break; |
202 | } |
203 | if (error) |
204 | return error; |
205 | |
206 | /* |
207 | * Remove the subsidiary block from the cache and from the log. |
208 | */ |
209 | error = xfs_trans_get_buf(tp: *trans, target: mp->m_ddev_targp, |
210 | blkno: child_blkno, |
211 | numblks: XFS_FSB_TO_BB(mp, mp->m_attr_geo->fsbcount), flags: 0, |
212 | bpp: &child_bp); |
213 | if (error) |
214 | return error; |
215 | xfs_trans_binval(*trans, child_bp); |
216 | child_bp = NULL; |
217 | |
218 | /* |
219 | * If we're not done, re-read the parent to get the next |
220 | * child block number. |
221 | */ |
222 | if (i + 1 < ichdr.count) { |
223 | struct xfs_da3_icnode_hdr phdr; |
224 | |
225 | error = xfs_da3_node_read_mapped(*trans, dp, |
226 | parent_blkno, &bp, XFS_ATTR_FORK); |
227 | if (error) |
228 | return error; |
229 | xfs_da3_node_hdr_from_disk(dp->i_mount, &phdr, |
230 | bp->b_addr); |
231 | child_fsb = be32_to_cpu(phdr.btree[i + 1].before); |
232 | xfs_trans_brelse(*trans, bp); |
233 | bp = NULL; |
234 | } |
235 | /* |
236 | * Atomically commit the whole invalidate stuff. |
237 | */ |
238 | error = xfs_trans_roll_inode(trans, dp); |
239 | if (error) |
240 | return error; |
241 | } |
242 | |
243 | return 0; |
244 | } |
245 | |
246 | /* |
247 | * Indiscriminately delete the entire attribute fork |
248 | * |
249 | * Recurse (gasp!) through the attribute nodes until we find leaves. |
250 | * We're doing a depth-first traversal in order to invalidate everything. |
251 | */ |
252 | static int |
253 | xfs_attr3_root_inactive( |
254 | struct xfs_trans **trans, |
255 | struct xfs_inode *dp) |
256 | { |
257 | struct xfs_mount *mp = dp->i_mount; |
258 | struct xfs_da_blkinfo *info; |
259 | struct xfs_buf *bp; |
260 | xfs_daddr_t blkno; |
261 | int error; |
262 | |
263 | /* |
264 | * Read block 0 to see what we have to work with. |
265 | * We only get here if we have extents, since we remove |
266 | * the extents in reverse order the extent containing |
267 | * block 0 must still be there. |
268 | */ |
269 | error = xfs_da3_node_read(*trans, dp, 0, &bp, XFS_ATTR_FORK); |
270 | if (error) |
271 | return error; |
272 | blkno = xfs_buf_daddr(bp); |
273 | |
274 | /* |
275 | * Invalidate the tree, even if the "tree" is only a single leaf block. |
276 | * This is a depth-first traversal! |
277 | */ |
278 | info = bp->b_addr; |
279 | switch (info->magic) { |
280 | case cpu_to_be16(XFS_DA_NODE_MAGIC): |
281 | case cpu_to_be16(XFS_DA3_NODE_MAGIC): |
282 | error = xfs_attr3_node_inactive(trans, dp, bp, level: 1); |
283 | break; |
284 | case cpu_to_be16(XFS_ATTR_LEAF_MAGIC): |
285 | case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC): |
286 | error = xfs_attr3_leaf_inactive(trans, dp, bp); |
287 | break; |
288 | default: |
289 | error = -EFSCORRUPTED; |
290 | xfs_buf_mark_corrupt(bp); |
291 | xfs_trans_brelse(*trans, bp); |
292 | break; |
293 | } |
294 | if (error) |
295 | return error; |
296 | |
297 | /* |
298 | * Invalidate the incore copy of the root block. |
299 | */ |
300 | error = xfs_trans_get_buf(tp: *trans, target: mp->m_ddev_targp, blkno, |
301 | numblks: XFS_FSB_TO_BB(mp, mp->m_attr_geo->fsbcount), flags: 0, bpp: &bp); |
302 | if (error) |
303 | return error; |
304 | error = bp->b_error; |
305 | if (error) { |
306 | xfs_trans_brelse(*trans, bp); |
307 | return error; |
308 | } |
309 | xfs_trans_binval(*trans, bp); /* remove from cache */ |
310 | /* |
311 | * Commit the invalidate and start the next transaction. |
312 | */ |
313 | error = xfs_trans_roll_inode(trans, dp); |
314 | |
315 | return error; |
316 | } |
317 | |
318 | /* |
319 | * xfs_attr_inactive kills all traces of an attribute fork on an inode. It |
320 | * removes both the on-disk and in-memory inode fork. Note that this also has to |
321 | * handle the condition of inodes without attributes but with an attribute fork |
322 | * configured, so we can't use xfs_inode_hasattr() here. |
323 | * |
324 | * The in-memory attribute fork is removed even on error. |
325 | */ |
326 | int |
327 | xfs_attr_inactive( |
328 | struct xfs_inode *dp) |
329 | { |
330 | struct xfs_trans *trans; |
331 | struct xfs_mount *mp; |
332 | int lock_mode = XFS_ILOCK_SHARED; |
333 | int error = 0; |
334 | |
335 | mp = dp->i_mount; |
336 | |
337 | xfs_ilock(dp, lock_mode); |
338 | if (!xfs_inode_has_attr_fork(ip: dp)) |
339 | goto out_destroy_fork; |
340 | xfs_iunlock(dp, lock_mode); |
341 | |
342 | lock_mode = 0; |
343 | |
344 | error = xfs_trans_alloc(mp, resp: &M_RES(mp)->tr_attrinval, blocks: 0, rtextents: 0, flags: 0, tpp: &trans); |
345 | if (error) |
346 | goto out_destroy_fork; |
347 | |
348 | lock_mode = XFS_ILOCK_EXCL; |
349 | xfs_ilock(dp, lock_mode); |
350 | |
351 | if (!xfs_inode_has_attr_fork(ip: dp)) |
352 | goto out_cancel; |
353 | |
354 | /* |
355 | * No need to make quota reservations here. We expect to release some |
356 | * blocks, not allocate, in the common case. |
357 | */ |
358 | xfs_trans_ijoin(trans, dp, 0); |
359 | |
360 | /* |
361 | * Invalidate and truncate the attribute fork extents. Make sure the |
362 | * fork actually has xattr blocks as otherwise the invalidation has no |
363 | * blocks to read and returns an error. In this case, just do the fork |
364 | * removal below. |
365 | */ |
366 | if (dp->i_af.if_nextents > 0) { |
367 | error = xfs_attr3_root_inactive(trans: &trans, dp); |
368 | if (error) |
369 | goto out_cancel; |
370 | |
371 | error = xfs_itruncate_extents(&trans, dp, XFS_ATTR_FORK, 0); |
372 | if (error) |
373 | goto out_cancel; |
374 | } |
375 | |
376 | /* Reset the attribute fork - this also destroys the in-core fork */ |
377 | xfs_attr_fork_remove(dp, trans); |
378 | |
379 | error = xfs_trans_commit(trans); |
380 | xfs_iunlock(dp, lock_mode); |
381 | return error; |
382 | |
383 | out_cancel: |
384 | xfs_trans_cancel(trans); |
385 | out_destroy_fork: |
386 | /* kill the in-core attr fork before we drop the inode lock */ |
387 | xfs_ifork_zap_attr(dp); |
388 | if (lock_mode) |
389 | xfs_iunlock(dp, lock_mode); |
390 | return error; |
391 | } |
392 | |