]>
Commit | Line | Data |
---|---|---|
2bd0ea18 | 1 | /* |
da23017d NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
5000d01d | 4 | * |
da23017d NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
2bd0ea18 | 7 | * published by the Free Software Foundation. |
5000d01d | 8 | * |
da23017d NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
5000d01d | 13 | * |
da23017d NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
2bd0ea18 NS |
17 | */ |
18 | ||
19 | #include <xfs.h> | |
20 | ||
21 | /* | |
22 | * xfs_da_btree.c | |
23 | * | |
24 | * Routines to implement directories as Btrees of hashed names. | |
25 | */ | |
26 | ||
5e656dbb BN |
27 | /*======================================================================== |
28 | * Function prototypes for the kernel. | |
29 | *========================================================================*/ | |
30 | ||
31 | /* | |
32 | * Routines used for growing the Btree. | |
33 | */ | |
34 | STATIC int xfs_da_root_split(xfs_da_state_t *state, | |
35 | xfs_da_state_blk_t *existing_root, | |
36 | xfs_da_state_blk_t *new_child); | |
37 | STATIC int xfs_da_node_split(xfs_da_state_t *state, | |
38 | xfs_da_state_blk_t *existing_blk, | |
39 | xfs_da_state_blk_t *split_blk, | |
40 | xfs_da_state_blk_t *blk_to_add, | |
41 | int treelevel, | |
42 | int *result); | |
43 | STATIC void xfs_da_node_rebalance(xfs_da_state_t *state, | |
44 | xfs_da_state_blk_t *node_blk_1, | |
45 | xfs_da_state_blk_t *node_blk_2); | |
46 | STATIC void xfs_da_node_add(xfs_da_state_t *state, | |
47 | xfs_da_state_blk_t *old_node_blk, | |
48 | xfs_da_state_blk_t *new_node_blk); | |
49 | ||
50 | /* | |
51 | * Routines used for shrinking the Btree. | |
52 | */ | |
53 | STATIC int xfs_da_root_join(xfs_da_state_t *state, | |
54 | xfs_da_state_blk_t *root_blk); | |
55 | STATIC int xfs_da_node_toosmall(xfs_da_state_t *state, int *retval); | |
56 | STATIC void xfs_da_node_remove(xfs_da_state_t *state, | |
57 | xfs_da_state_blk_t *drop_blk); | |
58 | STATIC void xfs_da_node_unbalance(xfs_da_state_t *state, | |
59 | xfs_da_state_blk_t *src_node_blk, | |
60 | xfs_da_state_blk_t *dst_node_blk); | |
61 | ||
62 | /* | |
63 | * Utility routines. | |
64 | */ | |
a2ceac1f DC |
65 | STATIC uint xfs_da_node_lasthash(struct xfs_buf *bp, int *count); |
66 | STATIC int xfs_da_node_order(struct xfs_buf *node1_bp, | |
67 | struct xfs_buf *node2_bp); | |
5e656dbb BN |
68 | STATIC int xfs_da_blk_unlink(xfs_da_state_t *state, |
69 | xfs_da_state_blk_t *drop_blk, | |
70 | xfs_da_state_blk_t *save_blk); | |
71 | STATIC void xfs_da_state_kill_altpath(xfs_da_state_t *state); | |
72 | ||
a2ceac1f DC |
73 | static void |
74 | xfs_da_node_verify( | |
75 | struct xfs_buf *bp) | |
76 | { | |
77 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
78 | struct xfs_da_node_hdr *hdr = bp->b_addr; | |
79 | int block_ok = 0; | |
80 | ||
81 | block_ok = hdr->info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC); | |
82 | block_ok = block_ok && | |
83 | be16_to_cpu(hdr->level) > 0 && | |
84 | be16_to_cpu(hdr->count) > 0 ; | |
85 | if (!block_ok) { | |
86 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, hdr); | |
87 | xfs_buf_ioerror(bp, EFSCORRUPTED); | |
88 | } | |
89 | ||
90 | } | |
91 | ||
92 | static void | |
93 | xfs_da_node_write_verify( | |
94 | struct xfs_buf *bp) | |
95 | { | |
96 | xfs_da_node_verify(bp); | |
97 | } | |
98 | ||
99 | /* | |
100 | * leaf/node format detection on trees is sketchy, so a node read can be done on | |
101 | * leaf level blocks when detection identifies the tree as a node format tree | |
102 | * incorrectly. In this case, we need to swap the verifier to match the correct | |
103 | * format of the block being read. | |
104 | */ | |
105 | static void | |
106 | xfs_da_node_read_verify( | |
107 | struct xfs_buf *bp) | |
108 | { | |
109 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
110 | struct xfs_da_blkinfo *info = bp->b_addr; | |
111 | ||
112 | switch (be16_to_cpu(info->magic)) { | |
113 | case XFS_DA_NODE_MAGIC: | |
114 | xfs_da_node_verify(bp); | |
115 | break; | |
116 | case XFS_ATTR_LEAF_MAGIC: | |
117 | bp->b_ops = &xfs_attr_leaf_buf_ops; | |
118 | bp->b_ops->verify_read(bp); | |
119 | return; | |
120 | case XFS_DIR2_LEAFN_MAGIC: | |
121 | bp->b_ops = &xfs_dir2_leafn_buf_ops; | |
122 | bp->b_ops->verify_read(bp); | |
123 | return; | |
124 | default: | |
125 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, | |
126 | mp, info); | |
127 | xfs_buf_ioerror(bp, EFSCORRUPTED); | |
128 | break; | |
129 | } | |
130 | } | |
131 | ||
132 | const struct xfs_buf_ops xfs_da_node_buf_ops = { | |
133 | .verify_read = xfs_da_node_read_verify, | |
134 | .verify_write = xfs_da_node_write_verify, | |
135 | }; | |
136 | ||
137 | ||
138 | int | |
139 | xfs_da_node_read( | |
140 | struct xfs_trans *tp, | |
141 | struct xfs_inode *dp, | |
142 | xfs_dablk_t bno, | |
143 | xfs_daddr_t mappedbno, | |
144 | struct xfs_buf **bpp, | |
145 | int which_fork) | |
146 | { | |
147 | return xfs_da_read_buf(tp, dp, bno, mappedbno, bpp, | |
148 | which_fork, &xfs_da_node_buf_ops); | |
149 | } | |
150 | ||
2bd0ea18 NS |
151 | /*======================================================================== |
152 | * Routines used for growing the Btree. | |
153 | *========================================================================*/ | |
154 | ||
155 | /* | |
156 | * Create the initial contents of an intermediate node. | |
157 | */ | |
158 | int | |
159 | xfs_da_node_create(xfs_da_args_t *args, xfs_dablk_t blkno, int level, | |
a2ceac1f | 160 | struct xfs_buf **bpp, int whichfork) |
2bd0ea18 NS |
161 | { |
162 | xfs_da_intnode_t *node; | |
a2ceac1f | 163 | struct xfs_buf *bp; |
2bd0ea18 NS |
164 | int error; |
165 | xfs_trans_t *tp; | |
166 | ||
a2ceac1f DC |
167 | trace_xfs_da_node_create(args); |
168 | ||
2bd0ea18 NS |
169 | tp = args->trans; |
170 | error = xfs_da_get_buf(tp, args->dp, blkno, -1, &bp, whichfork); | |
171 | if (error) | |
172 | return(error); | |
173 | ASSERT(bp != NULL); | |
a2ceac1f | 174 | node = bp->b_addr; |
46eca962 NS |
175 | node->hdr.info.forw = 0; |
176 | node->hdr.info.back = 0; | |
5e656dbb | 177 | node->hdr.info.magic = cpu_to_be16(XFS_DA_NODE_MAGIC); |
46eca962 NS |
178 | node->hdr.info.pad = 0; |
179 | node->hdr.count = 0; | |
5e656dbb | 180 | node->hdr.level = cpu_to_be16(level); |
2bd0ea18 | 181 | |
a2ceac1f | 182 | xfs_trans_log_buf(tp, bp, |
2bd0ea18 NS |
183 | XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr))); |
184 | ||
a2ceac1f | 185 | bp->b_ops = &xfs_da_node_buf_ops; |
2bd0ea18 NS |
186 | *bpp = bp; |
187 | return(0); | |
188 | } | |
189 | ||
190 | /* | |
191 | * Split a leaf node, rebalance, then possibly split | |
192 | * intermediate nodes, rebalance, etc. | |
193 | */ | |
194 | int /* error */ | |
195 | xfs_da_split(xfs_da_state_t *state) | |
196 | { | |
197 | xfs_da_state_blk_t *oldblk, *newblk, *addblk; | |
198 | xfs_da_intnode_t *node; | |
a2ceac1f | 199 | struct xfs_buf *bp; |
2bd0ea18 NS |
200 | int max, action, error, i; |
201 | ||
a2ceac1f DC |
202 | trace_xfs_da_split(state->args); |
203 | ||
2bd0ea18 NS |
204 | /* |
205 | * Walk back up the tree splitting/inserting/adjusting as necessary. | |
206 | * If we need to insert and there isn't room, split the node, then | |
207 | * decide which fragment to insert the new block from below into. | |
208 | * Note that we may split the root this way, but we need more fixup. | |
209 | */ | |
210 | max = state->path.active - 1; | |
211 | ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH)); | |
212 | ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC || | |
5e656dbb | 213 | state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC); |
2bd0ea18 NS |
214 | |
215 | addblk = &state->path.blk[max]; /* initial dummy value */ | |
216 | for (i = max; (i >= 0) && addblk; state->path.active--, i--) { | |
217 | oldblk = &state->path.blk[i]; | |
218 | newblk = &state->altpath.blk[i]; | |
219 | ||
220 | /* | |
221 | * If a leaf node then | |
222 | * Allocate a new leaf node, then rebalance across them. | |
223 | * else if an intermediate node then | |
224 | * We split on the last layer, must we split the node? | |
225 | */ | |
226 | switch (oldblk->magic) { | |
227 | case XFS_ATTR_LEAF_MAGIC: | |
2bd0ea18 NS |
228 | error = xfs_attr_leaf_split(state, oldblk, newblk); |
229 | if ((error != 0) && (error != ENOSPC)) { | |
230 | return(error); /* GROT: attr is inconsistent */ | |
231 | } | |
232 | if (!error) { | |
233 | addblk = newblk; | |
234 | break; | |
235 | } | |
236 | /* | |
237 | * Entry wouldn't fit, split the leaf again. | |
238 | */ | |
239 | state->extravalid = 1; | |
240 | if (state->inleaf) { | |
241 | state->extraafter = 0; /* before newblk */ | |
a2ceac1f | 242 | trace_xfs_attr_leaf_split_before(state->args); |
2bd0ea18 NS |
243 | error = xfs_attr_leaf_split(state, oldblk, |
244 | &state->extrablk); | |
245 | } else { | |
246 | state->extraafter = 1; /* after newblk */ | |
a2ceac1f | 247 | trace_xfs_attr_leaf_split_after(state->args); |
2bd0ea18 NS |
248 | error = xfs_attr_leaf_split(state, newblk, |
249 | &state->extrablk); | |
250 | } | |
251 | if (error) | |
252 | return(error); /* GROT: attr inconsistent */ | |
253 | addblk = newblk; | |
254 | break; | |
2bd0ea18 | 255 | case XFS_DIR2_LEAFN_MAGIC: |
2bd0ea18 NS |
256 | error = xfs_dir2_leafn_split(state, oldblk, newblk); |
257 | if (error) | |
258 | return error; | |
259 | addblk = newblk; | |
260 | break; | |
261 | case XFS_DA_NODE_MAGIC: | |
262 | error = xfs_da_node_split(state, oldblk, newblk, addblk, | |
263 | max - i, &action); | |
2bd0ea18 NS |
264 | addblk->bp = NULL; |
265 | if (error) | |
266 | return(error); /* GROT: dir is inconsistent */ | |
267 | /* | |
268 | * Record the newly split block for the next time thru? | |
269 | */ | |
270 | if (action) | |
271 | addblk = newblk; | |
272 | else | |
273 | addblk = NULL; | |
274 | break; | |
275 | } | |
276 | ||
277 | /* | |
278 | * Update the btree to show the new hashval for this child. | |
279 | */ | |
280 | xfs_da_fixhashpath(state, &state->path); | |
2bd0ea18 NS |
281 | } |
282 | if (!addblk) | |
283 | return(0); | |
284 | ||
285 | /* | |
286 | * Split the root node. | |
287 | */ | |
288 | ASSERT(state->path.active == 0); | |
289 | oldblk = &state->path.blk[0]; | |
290 | error = xfs_da_root_split(state, oldblk, addblk); | |
291 | if (error) { | |
2bd0ea18 NS |
292 | addblk->bp = NULL; |
293 | return(error); /* GROT: dir is inconsistent */ | |
294 | } | |
295 | ||
296 | /* | |
297 | * Update pointers to the node which used to be block 0 and | |
298 | * just got bumped because of the addition of a new root node. | |
299 | * There might be three blocks involved if a double split occurred, | |
300 | * and the original block 0 could be at any position in the list. | |
301 | */ | |
302 | ||
a2ceac1f | 303 | node = oldblk->bp->b_addr; |
46eca962 | 304 | if (node->hdr.info.forw) { |
5e656dbb | 305 | if (be32_to_cpu(node->hdr.info.forw) == addblk->blkno) { |
2bd0ea18 NS |
306 | bp = addblk->bp; |
307 | } else { | |
308 | ASSERT(state->extravalid); | |
309 | bp = state->extrablk.bp; | |
310 | } | |
a2ceac1f | 311 | node = bp->b_addr; |
5e656dbb | 312 | node->hdr.info.back = cpu_to_be32(oldblk->blkno); |
a2ceac1f | 313 | xfs_trans_log_buf(state->args->trans, bp, |
2bd0ea18 NS |
314 | XFS_DA_LOGRANGE(node, &node->hdr.info, |
315 | sizeof(node->hdr.info))); | |
316 | } | |
a2ceac1f | 317 | node = oldblk->bp->b_addr; |
5e656dbb BN |
318 | if (node->hdr.info.back) { |
319 | if (be32_to_cpu(node->hdr.info.back) == addblk->blkno) { | |
2bd0ea18 NS |
320 | bp = addblk->bp; |
321 | } else { | |
322 | ASSERT(state->extravalid); | |
323 | bp = state->extrablk.bp; | |
324 | } | |
a2ceac1f | 325 | node = bp->b_addr; |
5e656dbb | 326 | node->hdr.info.forw = cpu_to_be32(oldblk->blkno); |
a2ceac1f | 327 | xfs_trans_log_buf(state->args->trans, bp, |
2bd0ea18 NS |
328 | XFS_DA_LOGRANGE(node, &node->hdr.info, |
329 | sizeof(node->hdr.info))); | |
330 | } | |
2bd0ea18 NS |
331 | addblk->bp = NULL; |
332 | return(0); | |
333 | } | |
334 | ||
335 | /* | |
336 | * Split the root. We have to create a new root and point to the two | |
337 | * parts (the split old root) that we just created. Copy block zero to | |
338 | * the EOF, extending the inode in process. | |
339 | */ | |
340 | STATIC int /* error */ | |
341 | xfs_da_root_split(xfs_da_state_t *state, xfs_da_state_blk_t *blk1, | |
342 | xfs_da_state_blk_t *blk2) | |
343 | { | |
344 | xfs_da_intnode_t *node, *oldroot; | |
345 | xfs_da_args_t *args; | |
346 | xfs_dablk_t blkno; | |
a2ceac1f | 347 | struct xfs_buf *bp; |
2bd0ea18 NS |
348 | int error, size; |
349 | xfs_inode_t *dp; | |
350 | xfs_trans_t *tp; | |
351 | xfs_mount_t *mp; | |
352 | xfs_dir2_leaf_t *leaf; | |
353 | ||
a2ceac1f DC |
354 | trace_xfs_da_root_split(state->args); |
355 | ||
2bd0ea18 NS |
356 | /* |
357 | * Copy the existing (incorrect) block from the root node position | |
358 | * to a free space somewhere. | |
359 | */ | |
360 | args = state->args; | |
361 | ASSERT(args != NULL); | |
362 | error = xfs_da_grow_inode(args, &blkno); | |
363 | if (error) | |
364 | return(error); | |
365 | dp = args->dp; | |
366 | tp = args->trans; | |
367 | mp = state->mp; | |
368 | error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork); | |
369 | if (error) | |
370 | return(error); | |
371 | ASSERT(bp != NULL); | |
a2ceac1f DC |
372 | node = bp->b_addr; |
373 | oldroot = blk1->bp->b_addr; | |
374 | if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)) { | |
5e656dbb | 375 | size = (int)((char *)&oldroot->btree[be16_to_cpu(oldroot->hdr.count)] - |
2bd0ea18 NS |
376 | (char *)oldroot); |
377 | } else { | |
a2ceac1f | 378 | ASSERT(oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC)); |
2bd0ea18 | 379 | leaf = (xfs_dir2_leaf_t *)oldroot; |
5e656dbb | 380 | size = (int)((char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] - |
2bd0ea18 NS |
381 | (char *)leaf); |
382 | } | |
32181a02 | 383 | memcpy(node, oldroot, size); |
a2ceac1f DC |
384 | xfs_trans_log_buf(tp, bp, 0, size - 1); |
385 | ||
386 | bp->b_ops = blk1->bp->b_ops; | |
2bd0ea18 NS |
387 | blk1->bp = bp; |
388 | blk1->blkno = blkno; | |
389 | ||
390 | /* | |
391 | * Set up the new root node. | |
392 | */ | |
393 | error = xfs_da_node_create(args, | |
5e656dbb BN |
394 | (args->whichfork == XFS_DATA_FORK) ? mp->m_dirleafblk : 0, |
395 | be16_to_cpu(node->hdr.level) + 1, &bp, args->whichfork); | |
2bd0ea18 NS |
396 | if (error) |
397 | return(error); | |
a2ceac1f | 398 | node = bp->b_addr; |
5e656dbb BN |
399 | node->btree[0].hashval = cpu_to_be32(blk1->hashval); |
400 | node->btree[0].before = cpu_to_be32(blk1->blkno); | |
401 | node->btree[1].hashval = cpu_to_be32(blk2->hashval); | |
402 | node->btree[1].before = cpu_to_be32(blk2->blkno); | |
403 | node->hdr.count = cpu_to_be16(2); | |
6bef826c NS |
404 | |
405 | #ifdef DEBUG | |
a2ceac1f | 406 | if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC)) { |
2bd0ea18 NS |
407 | ASSERT(blk1->blkno >= mp->m_dirleafblk && |
408 | blk1->blkno < mp->m_dirfreeblk); | |
409 | ASSERT(blk2->blkno >= mp->m_dirleafblk && | |
410 | blk2->blkno < mp->m_dirfreeblk); | |
411 | } | |
6bef826c NS |
412 | #endif |
413 | ||
2bd0ea18 | 414 | /* Header is already logged by xfs_da_node_create */ |
a2ceac1f | 415 | xfs_trans_log_buf(tp, bp, |
2bd0ea18 NS |
416 | XFS_DA_LOGRANGE(node, node->btree, |
417 | sizeof(xfs_da_node_entry_t) * 2)); | |
2bd0ea18 NS |
418 | |
419 | return(0); | |
420 | } | |
421 | ||
422 | /* | |
423 | * Split the node, rebalance, then add the new entry. | |
424 | */ | |
425 | STATIC int /* error */ | |
426 | xfs_da_node_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk, | |
427 | xfs_da_state_blk_t *newblk, | |
428 | xfs_da_state_blk_t *addblk, | |
429 | int treelevel, int *result) | |
430 | { | |
431 | xfs_da_intnode_t *node; | |
432 | xfs_dablk_t blkno; | |
433 | int newcount, error; | |
434 | int useextra; | |
435 | ||
a2ceac1f DC |
436 | trace_xfs_da_node_split(state->args); |
437 | ||
438 | node = oldblk->bp->b_addr; | |
439 | ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); | |
2bd0ea18 NS |
440 | |
441 | /* | |
5e656dbb | 442 | * With V2 dirs the extra block is data or freespace. |
2bd0ea18 | 443 | */ |
5e656dbb | 444 | useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK; |
2bd0ea18 NS |
445 | newcount = 1 + useextra; |
446 | /* | |
447 | * Do we have to split the node? | |
448 | */ | |
5e656dbb | 449 | if ((be16_to_cpu(node->hdr.count) + newcount) > state->node_ents) { |
2bd0ea18 NS |
450 | /* |
451 | * Allocate a new node, add to the doubly linked chain of | |
452 | * nodes, then move some of our excess entries into it. | |
453 | */ | |
454 | error = xfs_da_grow_inode(state->args, &blkno); | |
455 | if (error) | |
456 | return(error); /* GROT: dir is inconsistent */ | |
5000d01d | 457 | |
2bd0ea18 NS |
458 | error = xfs_da_node_create(state->args, blkno, treelevel, |
459 | &newblk->bp, state->args->whichfork); | |
460 | if (error) | |
461 | return(error); /* GROT: dir is inconsistent */ | |
462 | newblk->blkno = blkno; | |
463 | newblk->magic = XFS_DA_NODE_MAGIC; | |
464 | xfs_da_node_rebalance(state, oldblk, newblk); | |
465 | error = xfs_da_blk_link(state, oldblk, newblk); | |
466 | if (error) | |
467 | return(error); | |
468 | *result = 1; | |
469 | } else { | |
470 | *result = 0; | |
471 | } | |
472 | ||
473 | /* | |
474 | * Insert the new entry(s) into the correct block | |
475 | * (updating last hashval in the process). | |
476 | * | |
477 | * xfs_da_node_add() inserts BEFORE the given index, | |
478 | * and as a result of using node_lookup_int() we always | |
479 | * point to a valid entry (not after one), but a split | |
480 | * operation always results in a new block whose hashvals | |
481 | * FOLLOW the current block. | |
482 | * | |
483 | * If we had double-split op below us, then add the extra block too. | |
484 | */ | |
a2ceac1f | 485 | node = oldblk->bp->b_addr; |
5e656dbb | 486 | if (oldblk->index <= be16_to_cpu(node->hdr.count)) { |
2bd0ea18 NS |
487 | oldblk->index++; |
488 | xfs_da_node_add(state, oldblk, addblk); | |
489 | if (useextra) { | |
490 | if (state->extraafter) | |
491 | oldblk->index++; | |
492 | xfs_da_node_add(state, oldblk, &state->extrablk); | |
493 | state->extravalid = 0; | |
494 | } | |
495 | } else { | |
496 | newblk->index++; | |
497 | xfs_da_node_add(state, newblk, addblk); | |
498 | if (useextra) { | |
499 | if (state->extraafter) | |
500 | newblk->index++; | |
501 | xfs_da_node_add(state, newblk, &state->extrablk); | |
502 | state->extravalid = 0; | |
503 | } | |
504 | } | |
505 | ||
506 | return(0); | |
507 | } | |
508 | ||
509 | /* | |
510 | * Balance the btree elements between two intermediate nodes, | |
511 | * usually one full and one empty. | |
512 | * | |
513 | * NOTE: if blk2 is empty, then it will get the upper half of blk1. | |
514 | */ | |
515 | STATIC void | |
516 | xfs_da_node_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1, | |
517 | xfs_da_state_blk_t *blk2) | |
518 | { | |
519 | xfs_da_intnode_t *node1, *node2, *tmpnode; | |
520 | xfs_da_node_entry_t *btree_s, *btree_d; | |
521 | int count, tmp; | |
522 | xfs_trans_t *tp; | |
523 | ||
a2ceac1f DC |
524 | trace_xfs_da_node_rebalance(state->args); |
525 | ||
526 | node1 = blk1->bp->b_addr; | |
527 | node2 = blk2->bp->b_addr; | |
2bd0ea18 NS |
528 | /* |
529 | * Figure out how many entries need to move, and in which direction. | |
530 | * Swap the nodes around if that makes it simpler. | |
531 | */ | |
5e656dbb BN |
532 | if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) && |
533 | ((be32_to_cpu(node2->btree[0].hashval) < be32_to_cpu(node1->btree[0].hashval)) || | |
534 | (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) < | |
535 | be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) { | |
2bd0ea18 NS |
536 | tmpnode = node1; |
537 | node1 = node2; | |
538 | node2 = tmpnode; | |
539 | } | |
a2ceac1f DC |
540 | ASSERT(node1->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
541 | ASSERT(node2->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); | |
5e656dbb | 542 | count = (be16_to_cpu(node1->hdr.count) - be16_to_cpu(node2->hdr.count)) / 2; |
2bd0ea18 NS |
543 | if (count == 0) |
544 | return; | |
545 | tp = state->args->trans; | |
546 | /* | |
547 | * Two cases: high-to-low and low-to-high. | |
548 | */ | |
549 | if (count > 0) { | |
550 | /* | |
551 | * Move elements in node2 up to make a hole. | |
552 | */ | |
5e656dbb | 553 | if ((tmp = be16_to_cpu(node2->hdr.count)) > 0) { |
2bd0ea18 NS |
554 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
555 | btree_s = &node2->btree[0]; | |
556 | btree_d = &node2->btree[count]; | |
32181a02 | 557 | memmove(btree_d, btree_s, tmp); |
2bd0ea18 NS |
558 | } |
559 | ||
560 | /* | |
561 | * Move the req'd B-tree elements from high in node1 to | |
562 | * low in node2. | |
563 | */ | |
5e656dbb | 564 | be16_add_cpu(&node2->hdr.count, count); |
2bd0ea18 | 565 | tmp = count * (uint)sizeof(xfs_da_node_entry_t); |
5e656dbb | 566 | btree_s = &node1->btree[be16_to_cpu(node1->hdr.count) - count]; |
2bd0ea18 | 567 | btree_d = &node2->btree[0]; |
32181a02 | 568 | memcpy(btree_d, btree_s, tmp); |
5e656dbb | 569 | be16_add_cpu(&node1->hdr.count, -count); |
2bd0ea18 NS |
570 | } else { |
571 | /* | |
572 | * Move the req'd B-tree elements from low in node2 to | |
573 | * high in node1. | |
574 | */ | |
575 | count = -count; | |
576 | tmp = count * (uint)sizeof(xfs_da_node_entry_t); | |
577 | btree_s = &node2->btree[0]; | |
5e656dbb | 578 | btree_d = &node1->btree[be16_to_cpu(node1->hdr.count)]; |
32181a02 | 579 | memcpy(btree_d, btree_s, tmp); |
5e656dbb | 580 | be16_add_cpu(&node1->hdr.count, count); |
a2ceac1f | 581 | xfs_trans_log_buf(tp, blk1->bp, |
2bd0ea18 NS |
582 | XFS_DA_LOGRANGE(node1, btree_d, tmp)); |
583 | ||
584 | /* | |
585 | * Move elements in node2 down to fill the hole. | |
586 | */ | |
5e656dbb | 587 | tmp = be16_to_cpu(node2->hdr.count) - count; |
2bd0ea18 NS |
588 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
589 | btree_s = &node2->btree[count]; | |
590 | btree_d = &node2->btree[0]; | |
32181a02 | 591 | memmove(btree_d, btree_s, tmp); |
5e656dbb | 592 | be16_add_cpu(&node2->hdr.count, -count); |
2bd0ea18 NS |
593 | } |
594 | ||
595 | /* | |
596 | * Log header of node 1 and all current bits of node 2. | |
597 | */ | |
a2ceac1f | 598 | xfs_trans_log_buf(tp, blk1->bp, |
2bd0ea18 | 599 | XFS_DA_LOGRANGE(node1, &node1->hdr, sizeof(node1->hdr))); |
a2ceac1f | 600 | xfs_trans_log_buf(tp, blk2->bp, |
2bd0ea18 NS |
601 | XFS_DA_LOGRANGE(node2, &node2->hdr, |
602 | sizeof(node2->hdr) + | |
5e656dbb | 603 | sizeof(node2->btree[0]) * be16_to_cpu(node2->hdr.count))); |
2bd0ea18 NS |
604 | |
605 | /* | |
606 | * Record the last hashval from each block for upward propagation. | |
607 | * (note: don't use the swapped node pointers) | |
608 | */ | |
a2ceac1f DC |
609 | node1 = blk1->bp->b_addr; |
610 | node2 = blk2->bp->b_addr; | |
5e656dbb BN |
611 | blk1->hashval = be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval); |
612 | blk2->hashval = be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval); | |
2bd0ea18 NS |
613 | |
614 | /* | |
615 | * Adjust the expected index for insertion. | |
616 | */ | |
5e656dbb BN |
617 | if (blk1->index >= be16_to_cpu(node1->hdr.count)) { |
618 | blk2->index = blk1->index - be16_to_cpu(node1->hdr.count); | |
619 | blk1->index = be16_to_cpu(node1->hdr.count) + 1; /* make it invalid */ | |
2bd0ea18 NS |
620 | } |
621 | } | |
622 | ||
623 | /* | |
624 | * Add a new entry to an intermediate node. | |
625 | */ | |
626 | STATIC void | |
627 | xfs_da_node_add(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk, | |
628 | xfs_da_state_blk_t *newblk) | |
629 | { | |
630 | xfs_da_intnode_t *node; | |
631 | xfs_da_node_entry_t *btree; | |
632 | int tmp; | |
2bd0ea18 | 633 | |
a2ceac1f DC |
634 | trace_xfs_da_node_add(state->args); |
635 | ||
636 | node = oldblk->bp->b_addr; | |
637 | ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); | |
5e656dbb | 638 | ASSERT((oldblk->index >= 0) && (oldblk->index <= be16_to_cpu(node->hdr.count))); |
2bd0ea18 | 639 | ASSERT(newblk->blkno != 0); |
5e656dbb | 640 | if (state->args->whichfork == XFS_DATA_FORK) |
56b2de80 DC |
641 | ASSERT(newblk->blkno >= state->mp->m_dirleafblk && |
642 | newblk->blkno < state->mp->m_dirfreeblk); | |
2bd0ea18 NS |
643 | |
644 | /* | |
645 | * We may need to make some room before we insert the new node. | |
646 | */ | |
647 | tmp = 0; | |
648 | btree = &node->btree[ oldblk->index ]; | |
5e656dbb BN |
649 | if (oldblk->index < be16_to_cpu(node->hdr.count)) { |
650 | tmp = (be16_to_cpu(node->hdr.count) - oldblk->index) * (uint)sizeof(*btree); | |
32181a02 | 651 | memmove(btree + 1, btree, tmp); |
2bd0ea18 | 652 | } |
5e656dbb BN |
653 | btree->hashval = cpu_to_be32(newblk->hashval); |
654 | btree->before = cpu_to_be32(newblk->blkno); | |
a2ceac1f | 655 | xfs_trans_log_buf(state->args->trans, oldblk->bp, |
2bd0ea18 | 656 | XFS_DA_LOGRANGE(node, btree, tmp + sizeof(*btree))); |
5e656dbb | 657 | be16_add_cpu(&node->hdr.count, 1); |
a2ceac1f | 658 | xfs_trans_log_buf(state->args->trans, oldblk->bp, |
2bd0ea18 NS |
659 | XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr))); |
660 | ||
661 | /* | |
662 | * Copy the last hash value from the oldblk to propagate upwards. | |
663 | */ | |
5e656dbb | 664 | oldblk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1 ].hashval); |
2bd0ea18 NS |
665 | } |
666 | ||
667 | /*======================================================================== | |
668 | * Routines used for shrinking the Btree. | |
669 | *========================================================================*/ | |
670 | ||
671 | /* | |
672 | * Deallocate an empty leaf node, remove it from its parent, | |
673 | * possibly deallocating that block, etc... | |
674 | */ | |
675 | int | |
676 | xfs_da_join(xfs_da_state_t *state) | |
677 | { | |
678 | xfs_da_state_blk_t *drop_blk, *save_blk; | |
679 | int action, error; | |
680 | ||
a2ceac1f DC |
681 | trace_xfs_da_join(state->args); |
682 | ||
2bd0ea18 NS |
683 | action = 0; |
684 | drop_blk = &state->path.blk[ state->path.active-1 ]; | |
685 | save_blk = &state->altpath.blk[ state->path.active-1 ]; | |
686 | ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC); | |
687 | ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC || | |
5e656dbb | 688 | drop_blk->magic == XFS_DIR2_LEAFN_MAGIC); |
2bd0ea18 NS |
689 | |
690 | /* | |
691 | * Walk back up the tree joining/deallocating as necessary. | |
692 | * When we stop dropping blocks, break out. | |
693 | */ | |
694 | for ( ; state->path.active >= 2; drop_blk--, save_blk--, | |
695 | state->path.active--) { | |
696 | /* | |
697 | * See if we can combine the block with a neighbor. | |
698 | * (action == 0) => no options, just leave | |
699 | * (action == 1) => coalesce, then unlink | |
700 | * (action == 2) => block empty, unlink it | |
701 | */ | |
702 | switch (drop_blk->magic) { | |
703 | case XFS_ATTR_LEAF_MAGIC: | |
2bd0ea18 | 704 | error = xfs_attr_leaf_toosmall(state, &action); |
2bd0ea18 NS |
705 | if (error) |
706 | return(error); | |
707 | if (action == 0) | |
708 | return(0); | |
2bd0ea18 | 709 | xfs_attr_leaf_unbalance(state, drop_blk, save_blk); |
2bd0ea18 | 710 | break; |
2bd0ea18 | 711 | case XFS_DIR2_LEAFN_MAGIC: |
2bd0ea18 NS |
712 | error = xfs_dir2_leafn_toosmall(state, &action); |
713 | if (error) | |
714 | return error; | |
715 | if (action == 0) | |
716 | return 0; | |
717 | xfs_dir2_leafn_unbalance(state, drop_blk, save_blk); | |
718 | break; | |
719 | case XFS_DA_NODE_MAGIC: | |
720 | /* | |
721 | * Remove the offending node, fixup hashvals, | |
722 | * check for a toosmall neighbor. | |
723 | */ | |
724 | xfs_da_node_remove(state, drop_blk); | |
725 | xfs_da_fixhashpath(state, &state->path); | |
726 | error = xfs_da_node_toosmall(state, &action); | |
727 | if (error) | |
728 | return(error); | |
729 | if (action == 0) | |
730 | return 0; | |
731 | xfs_da_node_unbalance(state, drop_blk, save_blk); | |
732 | break; | |
733 | } | |
734 | xfs_da_fixhashpath(state, &state->altpath); | |
735 | error = xfs_da_blk_unlink(state, drop_blk, save_blk); | |
736 | xfs_da_state_kill_altpath(state); | |
737 | if (error) | |
738 | return(error); | |
739 | error = xfs_da_shrink_inode(state->args, drop_blk->blkno, | |
740 | drop_blk->bp); | |
741 | drop_blk->bp = NULL; | |
742 | if (error) | |
743 | return(error); | |
744 | } | |
745 | /* | |
746 | * We joined all the way to the top. If it turns out that | |
747 | * we only have one entry in the root, make the child block | |
748 | * the new root. | |
749 | */ | |
750 | xfs_da_node_remove(state, drop_blk); | |
751 | xfs_da_fixhashpath(state, &state->path); | |
752 | error = xfs_da_root_join(state, &state->path.blk[0]); | |
753 | return(error); | |
754 | } | |
755 | ||
a2ceac1f DC |
756 | #ifdef DEBUG |
757 | static void | |
758 | xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level) | |
759 | { | |
760 | __be16 magic = blkinfo->magic; | |
761 | ||
762 | if (level == 1) { | |
763 | ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || | |
764 | magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC)); | |
765 | } else | |
766 | ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); | |
767 | ASSERT(!blkinfo->forw); | |
768 | ASSERT(!blkinfo->back); | |
769 | } | |
770 | #else /* !DEBUG */ | |
771 | #define xfs_da_blkinfo_onlychild_validate(blkinfo, level) | |
772 | #endif /* !DEBUG */ | |
773 | ||
2bd0ea18 | 774 | /* |
dfc130f3 | 775 | * We have only one entry in the root. Copy the only remaining child of |
2bd0ea18 NS |
776 | * the old root to block 0 as the new root node. |
777 | */ | |
778 | STATIC int | |
779 | xfs_da_root_join(xfs_da_state_t *state, xfs_da_state_blk_t *root_blk) | |
780 | { | |
781 | xfs_da_intnode_t *oldroot; | |
2bd0ea18 NS |
782 | xfs_da_args_t *args; |
783 | xfs_dablk_t child; | |
a2ceac1f | 784 | struct xfs_buf *bp; |
2bd0ea18 NS |
785 | int error; |
786 | ||
a2ceac1f DC |
787 | trace_xfs_da_root_join(state->args); |
788 | ||
2bd0ea18 NS |
789 | args = state->args; |
790 | ASSERT(args != NULL); | |
791 | ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC); | |
a2ceac1f DC |
792 | oldroot = root_blk->bp->b_addr; |
793 | ASSERT(oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); | |
46eca962 NS |
794 | ASSERT(!oldroot->hdr.info.forw); |
795 | ASSERT(!oldroot->hdr.info.back); | |
2bd0ea18 NS |
796 | |
797 | /* | |
798 | * If the root has more than one child, then don't do anything. | |
799 | */ | |
5e656dbb | 800 | if (be16_to_cpu(oldroot->hdr.count) > 1) |
2bd0ea18 NS |
801 | return(0); |
802 | ||
803 | /* | |
804 | * Read in the (only) child block, then copy those bytes into | |
805 | * the root block's buffer and free the original child block. | |
806 | */ | |
5e656dbb | 807 | child = be32_to_cpu(oldroot->btree[0].before); |
2bd0ea18 | 808 | ASSERT(child != 0); |
a2ceac1f | 809 | error = xfs_da_node_read(args->trans, args->dp, child, -1, &bp, |
2bd0ea18 NS |
810 | args->whichfork); |
811 | if (error) | |
812 | return(error); | |
813 | ASSERT(bp != NULL); | |
a2ceac1f DC |
814 | xfs_da_blkinfo_onlychild_validate(bp->b_addr, |
815 | be16_to_cpu(oldroot->hdr.level)); | |
816 | ||
817 | /* | |
818 | * This could be copying a leaf back into the root block in the case of | |
819 | * there only being a single leaf block left in the tree. Hence we have | |
820 | * to update the b_ops pointer as well to match the buffer type change | |
821 | * that could occur. | |
822 | */ | |
823 | memcpy(root_blk->bp->b_addr, bp->b_addr, state->blocksize); | |
824 | root_blk->bp->b_ops = bp->b_ops; | |
825 | xfs_trans_log_buf(args->trans, root_blk->bp, 0, state->blocksize - 1); | |
2bd0ea18 NS |
826 | error = xfs_da_shrink_inode(args, child, bp); |
827 | return(error); | |
828 | } | |
829 | ||
830 | /* | |
831 | * Check a node block and its neighbors to see if the block should be | |
832 | * collapsed into one or the other neighbor. Always keep the block | |
833 | * with the smaller block number. | |
834 | * If the current block is over 50% full, don't try to join it, return 0. | |
835 | * If the block is empty, fill in the state structure and return 2. | |
836 | * If it can be collapsed, fill in the state structure and return 1. | |
837 | * If nothing can be done, return 0. | |
838 | */ | |
839 | STATIC int | |
840 | xfs_da_node_toosmall(xfs_da_state_t *state, int *action) | |
841 | { | |
842 | xfs_da_intnode_t *node; | |
843 | xfs_da_state_blk_t *blk; | |
844 | xfs_da_blkinfo_t *info; | |
845 | int count, forward, error, retval, i; | |
846 | xfs_dablk_t blkno; | |
a2ceac1f DC |
847 | struct xfs_buf *bp; |
848 | ||
849 | trace_xfs_da_node_toosmall(state->args); | |
2bd0ea18 NS |
850 | |
851 | /* | |
852 | * Check for the degenerate case of the block being over 50% full. | |
853 | * If so, it's not worth even looking to see if we might be able | |
854 | * to coalesce with a sibling. | |
855 | */ | |
856 | blk = &state->path.blk[ state->path.active-1 ]; | |
a2ceac1f DC |
857 | info = blk->bp->b_addr; |
858 | ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); | |
2bd0ea18 | 859 | node = (xfs_da_intnode_t *)info; |
5e656dbb | 860 | count = be16_to_cpu(node->hdr.count); |
6bef826c | 861 | if (count > (state->node_ents >> 1)) { |
4ca431fc NS |
862 | *action = 0; /* blk over 50%, don't try to join */ |
863 | return(0); /* blk over 50%, don't try to join */ | |
2bd0ea18 NS |
864 | } |
865 | ||
866 | /* | |
867 | * Check for the degenerate case of the block being empty. | |
868 | * If the block is empty, we'll simply delete it, no need to | |
5e656dbb | 869 | * coalesce it with a sibling block. We choose (arbitrarily) |
2bd0ea18 NS |
870 | * to merge with the forward block unless it is NULL. |
871 | */ | |
872 | if (count == 0) { | |
873 | /* | |
874 | * Make altpath point to the block we want to keep and | |
875 | * path point to the block we want to drop (this one). | |
876 | */ | |
5e656dbb | 877 | forward = (info->forw != 0); |
32181a02 | 878 | memcpy(&state->altpath, &state->path, sizeof(state->path)); |
2bd0ea18 NS |
879 | error = xfs_da_path_shift(state, &state->altpath, forward, |
880 | 0, &retval); | |
881 | if (error) | |
882 | return(error); | |
883 | if (retval) { | |
884 | *action = 0; | |
885 | } else { | |
886 | *action = 2; | |
887 | } | |
888 | return(0); | |
889 | } | |
890 | ||
891 | /* | |
892 | * Examine each sibling block to see if we can coalesce with | |
893 | * at least 25% free space to spare. We need to figure out | |
894 | * whether to merge with the forward or the backward block. | |
895 | * We prefer coalescing with the lower numbered sibling so as | |
896 | * to shrink a directory over time. | |
897 | */ | |
898 | /* start with smaller blk num */ | |
5e656dbb | 899 | forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back)); |
2bd0ea18 NS |
900 | for (i = 0; i < 2; forward = !forward, i++) { |
901 | if (forward) | |
5e656dbb | 902 | blkno = be32_to_cpu(info->forw); |
2bd0ea18 | 903 | else |
5e656dbb | 904 | blkno = be32_to_cpu(info->back); |
2bd0ea18 NS |
905 | if (blkno == 0) |
906 | continue; | |
a2ceac1f | 907 | error = xfs_da_node_read(state->args->trans, state->args->dp, |
2bd0ea18 NS |
908 | blkno, -1, &bp, state->args->whichfork); |
909 | if (error) | |
910 | return(error); | |
911 | ASSERT(bp != NULL); | |
912 | ||
913 | node = (xfs_da_intnode_t *)info; | |
6bef826c NS |
914 | count = state->node_ents; |
915 | count -= state->node_ents >> 2; | |
5e656dbb | 916 | count -= be16_to_cpu(node->hdr.count); |
a2ceac1f DC |
917 | node = bp->b_addr; |
918 | ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); | |
5e656dbb | 919 | count -= be16_to_cpu(node->hdr.count); |
a2ceac1f | 920 | xfs_trans_brelse(state->args->trans, bp); |
2bd0ea18 NS |
921 | if (count >= 0) |
922 | break; /* fits with at least 25% to spare */ | |
923 | } | |
924 | if (i >= 2) { | |
925 | *action = 0; | |
926 | return(0); | |
927 | } | |
928 | ||
929 | /* | |
930 | * Make altpath point to the block we want to keep (the lower | |
931 | * numbered block) and path point to the block we want to drop. | |
932 | */ | |
32181a02 | 933 | memcpy(&state->altpath, &state->path, sizeof(state->path)); |
2bd0ea18 NS |
934 | if (blkno < blk->blkno) { |
935 | error = xfs_da_path_shift(state, &state->altpath, forward, | |
936 | 0, &retval); | |
937 | if (error) { | |
938 | return(error); | |
939 | } | |
940 | if (retval) { | |
941 | *action = 0; | |
942 | return(0); | |
943 | } | |
944 | } else { | |
945 | error = xfs_da_path_shift(state, &state->path, forward, | |
946 | 0, &retval); | |
947 | if (error) { | |
948 | return(error); | |
949 | } | |
950 | if (retval) { | |
951 | *action = 0; | |
952 | return(0); | |
953 | } | |
954 | } | |
955 | *action = 1; | |
956 | return(0); | |
957 | } | |
958 | ||
2bd0ea18 NS |
959 | /* |
960 | * Walk back up the tree adjusting hash values as necessary, | |
961 | * when we stop making changes, return. | |
962 | */ | |
963 | void | |
964 | xfs_da_fixhashpath(xfs_da_state_t *state, xfs_da_state_path_t *path) | |
965 | { | |
966 | xfs_da_state_blk_t *blk; | |
967 | xfs_da_intnode_t *node; | |
968 | xfs_da_node_entry_t *btree; | |
0e266570 | 969 | xfs_dahash_t lasthash=0; |
2bd0ea18 NS |
970 | int level, count; |
971 | ||
a2ceac1f DC |
972 | trace_xfs_da_fixhashpath(state->args); |
973 | ||
2bd0ea18 NS |
974 | level = path->active-1; |
975 | blk = &path->blk[ level ]; | |
976 | switch (blk->magic) { | |
2bd0ea18 NS |
977 | case XFS_ATTR_LEAF_MAGIC: |
978 | lasthash = xfs_attr_leaf_lasthash(blk->bp, &count); | |
979 | if (count == 0) | |
980 | return; | |
981 | break; | |
2bd0ea18 | 982 | case XFS_DIR2_LEAFN_MAGIC: |
2bd0ea18 NS |
983 | lasthash = xfs_dir2_leafn_lasthash(blk->bp, &count); |
984 | if (count == 0) | |
985 | return; | |
986 | break; | |
987 | case XFS_DA_NODE_MAGIC: | |
988 | lasthash = xfs_da_node_lasthash(blk->bp, &count); | |
989 | if (count == 0) | |
990 | return; | |
991 | break; | |
992 | } | |
993 | for (blk--, level--; level >= 0; blk--, level--) { | |
a2ceac1f DC |
994 | node = blk->bp->b_addr; |
995 | ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); | |
2bd0ea18 | 996 | btree = &node->btree[ blk->index ]; |
5e656dbb | 997 | if (be32_to_cpu(btree->hashval) == lasthash) |
2bd0ea18 NS |
998 | break; |
999 | blk->hashval = lasthash; | |
5e656dbb | 1000 | btree->hashval = cpu_to_be32(lasthash); |
a2ceac1f | 1001 | xfs_trans_log_buf(state->args->trans, blk->bp, |
2bd0ea18 NS |
1002 | XFS_DA_LOGRANGE(node, btree, sizeof(*btree))); |
1003 | ||
5e656dbb | 1004 | lasthash = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval); |
2bd0ea18 NS |
1005 | } |
1006 | } | |
1007 | ||
2bd0ea18 NS |
1008 | /* |
1009 | * Remove an entry from an intermediate node. | |
1010 | */ | |
1011 | STATIC void | |
1012 | xfs_da_node_remove(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk) | |
1013 | { | |
1014 | xfs_da_intnode_t *node; | |
1015 | xfs_da_node_entry_t *btree; | |
1016 | int tmp; | |
1017 | ||
a2ceac1f DC |
1018 | trace_xfs_da_node_remove(state->args); |
1019 | ||
1020 | node = drop_blk->bp->b_addr; | |
5e656dbb | 1021 | ASSERT(drop_blk->index < be16_to_cpu(node->hdr.count)); |
2bd0ea18 NS |
1022 | ASSERT(drop_blk->index >= 0); |
1023 | ||
1024 | /* | |
1025 | * Copy over the offending entry, or just zero it out. | |
1026 | */ | |
1027 | btree = &node->btree[drop_blk->index]; | |
5e656dbb BN |
1028 | if (drop_blk->index < (be16_to_cpu(node->hdr.count)-1)) { |
1029 | tmp = be16_to_cpu(node->hdr.count) - drop_blk->index - 1; | |
2bd0ea18 | 1030 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
32181a02 | 1031 | memmove(btree, btree + 1, tmp); |
a2ceac1f | 1032 | xfs_trans_log_buf(state->args->trans, drop_blk->bp, |
2bd0ea18 | 1033 | XFS_DA_LOGRANGE(node, btree, tmp)); |
5e656dbb | 1034 | btree = &node->btree[be16_to_cpu(node->hdr.count)-1]; |
2bd0ea18 | 1035 | } |
32181a02 | 1036 | memset((char *)btree, 0, sizeof(xfs_da_node_entry_t)); |
a2ceac1f | 1037 | xfs_trans_log_buf(state->args->trans, drop_blk->bp, |
2bd0ea18 | 1038 | XFS_DA_LOGRANGE(node, btree, sizeof(*btree))); |
5e656dbb | 1039 | be16_add_cpu(&node->hdr.count, -1); |
a2ceac1f | 1040 | xfs_trans_log_buf(state->args->trans, drop_blk->bp, |
2bd0ea18 NS |
1041 | XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr))); |
1042 | ||
1043 | /* | |
1044 | * Copy the last hash value from the block to propagate upwards. | |
1045 | */ | |
1046 | btree--; | |
5e656dbb | 1047 | drop_blk->hashval = be32_to_cpu(btree->hashval); |
2bd0ea18 NS |
1048 | } |
1049 | ||
1050 | /* | |
1051 | * Unbalance the btree elements between two intermediate nodes, | |
1052 | * move all Btree elements from one node into another. | |
1053 | */ | |
1054 | STATIC void | |
1055 | xfs_da_node_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk, | |
1056 | xfs_da_state_blk_t *save_blk) | |
1057 | { | |
1058 | xfs_da_intnode_t *drop_node, *save_node; | |
1059 | xfs_da_node_entry_t *btree; | |
1060 | int tmp; | |
1061 | xfs_trans_t *tp; | |
1062 | ||
a2ceac1f DC |
1063 | trace_xfs_da_node_unbalance(state->args); |
1064 | ||
1065 | drop_node = drop_blk->bp->b_addr; | |
1066 | save_node = save_blk->bp->b_addr; | |
1067 | ASSERT(drop_node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); | |
1068 | ASSERT(save_node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); | |
2bd0ea18 NS |
1069 | tp = state->args->trans; |
1070 | ||
1071 | /* | |
1072 | * If the dying block has lower hashvals, then move all the | |
1073 | * elements in the remaining block up to make a hole. | |
1074 | */ | |
5e656dbb BN |
1075 | if ((be32_to_cpu(drop_node->btree[0].hashval) < be32_to_cpu(save_node->btree[ 0 ].hashval)) || |
1076 | (be32_to_cpu(drop_node->btree[be16_to_cpu(drop_node->hdr.count)-1].hashval) < | |
1077 | be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval))) | |
2bd0ea18 | 1078 | { |
5e656dbb BN |
1079 | btree = &save_node->btree[be16_to_cpu(drop_node->hdr.count)]; |
1080 | tmp = be16_to_cpu(save_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t); | |
32181a02 | 1081 | memmove(btree, &save_node->btree[0], tmp); |
2bd0ea18 | 1082 | btree = &save_node->btree[0]; |
a2ceac1f | 1083 | xfs_trans_log_buf(tp, save_blk->bp, |
2bd0ea18 | 1084 | XFS_DA_LOGRANGE(save_node, btree, |
5e656dbb | 1085 | (be16_to_cpu(save_node->hdr.count) + be16_to_cpu(drop_node->hdr.count)) * |
2bd0ea18 NS |
1086 | sizeof(xfs_da_node_entry_t))); |
1087 | } else { | |
5e656dbb | 1088 | btree = &save_node->btree[be16_to_cpu(save_node->hdr.count)]; |
a2ceac1f | 1089 | xfs_trans_log_buf(tp, save_blk->bp, |
2bd0ea18 | 1090 | XFS_DA_LOGRANGE(save_node, btree, |
5e656dbb | 1091 | be16_to_cpu(drop_node->hdr.count) * |
2bd0ea18 NS |
1092 | sizeof(xfs_da_node_entry_t))); |
1093 | } | |
1094 | ||
1095 | /* | |
1096 | * Move all the B-tree elements from drop_blk to save_blk. | |
1097 | */ | |
5e656dbb | 1098 | tmp = be16_to_cpu(drop_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t); |
32181a02 | 1099 | memcpy(btree, &drop_node->btree[0], tmp); |
5e656dbb | 1100 | be16_add_cpu(&save_node->hdr.count, be16_to_cpu(drop_node->hdr.count)); |
2bd0ea18 | 1101 | |
a2ceac1f | 1102 | xfs_trans_log_buf(tp, save_blk->bp, |
2bd0ea18 NS |
1103 | XFS_DA_LOGRANGE(save_node, &save_node->hdr, |
1104 | sizeof(save_node->hdr))); | |
1105 | ||
1106 | /* | |
1107 | * Save the last hashval in the remaining block for upward propagation. | |
1108 | */ | |
5e656dbb | 1109 | save_blk->hashval = be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval); |
2bd0ea18 NS |
1110 | } |
1111 | ||
2bd0ea18 NS |
1112 | /*======================================================================== |
1113 | * Routines used for finding things in the Btree. | |
1114 | *========================================================================*/ | |
1115 | ||
1116 | /* | |
1117 | * Walk down the Btree looking for a particular filename, filling | |
1118 | * in the state structure as we go. | |
1119 | * | |
1120 | * We will set the state structure to point to each of the elements | |
1121 | * in each of the nodes where either the hashval is or should be. | |
1122 | * | |
1123 | * We support duplicate hashval's so for each entry in the current | |
1124 | * node that could contain the desired hashval, descend. This is a | |
1125 | * pruned depth-first tree search. | |
1126 | */ | |
1127 | int /* error */ | |
1128 | xfs_da_node_lookup_int(xfs_da_state_t *state, int *result) | |
1129 | { | |
1130 | xfs_da_state_blk_t *blk; | |
1131 | xfs_da_blkinfo_t *curr; | |
1132 | xfs_da_intnode_t *node; | |
1133 | xfs_da_node_entry_t *btree; | |
1134 | xfs_dablk_t blkno; | |
1135 | int probe, span, max, error, retval; | |
5e656dbb | 1136 | xfs_dahash_t hashval, btreehashval; |
2bd0ea18 NS |
1137 | xfs_da_args_t *args; |
1138 | ||
1139 | args = state->args; | |
32a82561 | 1140 | |
2bd0ea18 NS |
1141 | /* |
1142 | * Descend thru the B-tree searching each level for the right | |
1143 | * node to use, until the right hashval is found. | |
1144 | */ | |
5e656dbb | 1145 | blkno = (args->whichfork == XFS_DATA_FORK)? state->mp->m_dirleafblk : 0; |
2bd0ea18 NS |
1146 | for (blk = &state->path.blk[0], state->path.active = 1; |
1147 | state->path.active <= XFS_DA_NODE_MAXDEPTH; | |
1148 | blk++, state->path.active++) { | |
1149 | /* | |
1150 | * Read the next node down in the tree. | |
1151 | */ | |
1152 | blk->blkno = blkno; | |
a2ceac1f | 1153 | error = xfs_da_node_read(args->trans, args->dp, blkno, |
32a82561 | 1154 | -1, &blk->bp, args->whichfork); |
2bd0ea18 NS |
1155 | if (error) { |
1156 | blk->blkno = 0; | |
1157 | state->path.active--; | |
1158 | return(error); | |
1159 | } | |
a2ceac1f | 1160 | curr = blk->bp->b_addr; |
5e656dbb BN |
1161 | blk->magic = be16_to_cpu(curr->magic); |
1162 | ASSERT(blk->magic == XFS_DA_NODE_MAGIC || | |
1163 | blk->magic == XFS_DIR2_LEAFN_MAGIC || | |
1164 | blk->magic == XFS_ATTR_LEAF_MAGIC); | |
2bd0ea18 NS |
1165 | |
1166 | /* | |
1167 | * Search an intermediate node for a match. | |
1168 | */ | |
5e656dbb | 1169 | if (blk->magic == XFS_DA_NODE_MAGIC) { |
a2ceac1f | 1170 | node = blk->bp->b_addr; |
5e656dbb BN |
1171 | max = be16_to_cpu(node->hdr.count); |
1172 | blk->hashval = be32_to_cpu(node->btree[max-1].hashval); | |
2bd0ea18 NS |
1173 | |
1174 | /* | |
1175 | * Binary search. (note: small blocks will skip loop) | |
1176 | */ | |
2bd0ea18 | 1177 | probe = span = max / 2; |
32a82561 | 1178 | hashval = args->hashval; |
2bd0ea18 NS |
1179 | for (btree = &node->btree[probe]; span > 4; |
1180 | btree = &node->btree[probe]) { | |
1181 | span /= 2; | |
5e656dbb BN |
1182 | btreehashval = be32_to_cpu(btree->hashval); |
1183 | if (btreehashval < hashval) | |
2bd0ea18 | 1184 | probe += span; |
5e656dbb | 1185 | else if (btreehashval > hashval) |
2bd0ea18 NS |
1186 | probe -= span; |
1187 | else | |
1188 | break; | |
1189 | } | |
1190 | ASSERT((probe >= 0) && (probe < max)); | |
5e656dbb | 1191 | ASSERT((span <= 4) || (be32_to_cpu(btree->hashval) == hashval)); |
2bd0ea18 NS |
1192 | |
1193 | /* | |
1194 | * Since we may have duplicate hashval's, find the first | |
1195 | * matching hashval in the node. | |
1196 | */ | |
5e656dbb | 1197 | while ((probe > 0) && (be32_to_cpu(btree->hashval) >= hashval)) { |
2bd0ea18 NS |
1198 | btree--; |
1199 | probe--; | |
1200 | } | |
5e656dbb | 1201 | while ((probe < max) && (be32_to_cpu(btree->hashval) < hashval)) { |
2bd0ea18 NS |
1202 | btree++; |
1203 | probe++; | |
1204 | } | |
1205 | ||
1206 | /* | |
1207 | * Pick the right block to descend on. | |
1208 | */ | |
1209 | if (probe == max) { | |
1210 | blk->index = max-1; | |
5e656dbb | 1211 | blkno = be32_to_cpu(node->btree[max-1].before); |
2bd0ea18 NS |
1212 | } else { |
1213 | blk->index = probe; | |
5e656dbb | 1214 | blkno = be32_to_cpu(btree->before); |
2bd0ea18 | 1215 | } |
5e656dbb | 1216 | } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) { |
2bd0ea18 NS |
1217 | blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL); |
1218 | break; | |
5e656dbb | 1219 | } else if (blk->magic == XFS_DIR2_LEAFN_MAGIC) { |
2bd0ea18 NS |
1220 | blk->hashval = xfs_dir2_leafn_lasthash(blk->bp, NULL); |
1221 | break; | |
1222 | } | |
1223 | } | |
1224 | ||
1225 | /* | |
1226 | * A leaf block that ends in the hashval that we are interested in | |
1227 | * (final hashval == search hashval) means that the next block may | |
1228 | * contain more entries with the same hashval, shift upward to the | |
1229 | * next leaf and keep searching. | |
1230 | */ | |
1231 | for (;;) { | |
5e656dbb | 1232 | if (blk->magic == XFS_DIR2_LEAFN_MAGIC) { |
32a82561 | 1233 | retval = xfs_dir2_leafn_lookup_int(blk->bp, args, |
2bd0ea18 | 1234 | &blk->index, state); |
5e656dbb | 1235 | } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) { |
32a82561 NS |
1236 | retval = xfs_attr_leaf_lookup_int(blk->bp, args); |
1237 | blk->index = args->index; | |
1238 | args->blkno = blk->blkno; | |
5e656dbb BN |
1239 | } else { |
1240 | ASSERT(0); | |
1241 | return XFS_ERROR(EFSCORRUPTED); | |
2bd0ea18 | 1242 | } |
2bd0ea18 | 1243 | if (((retval == ENOENT) || (retval == ENOATTR)) && |
32a82561 | 1244 | (blk->hashval == args->hashval)) { |
2bd0ea18 NS |
1245 | error = xfs_da_path_shift(state, &state->path, 1, 1, |
1246 | &retval); | |
1247 | if (error) | |
1248 | return(error); | |
1249 | if (retval == 0) { | |
1250 | continue; | |
5e656dbb | 1251 | } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) { |
2bd0ea18 NS |
1252 | /* path_shift() gives ENOENT */ |
1253 | retval = XFS_ERROR(ENOATTR); | |
1254 | } | |
2bd0ea18 NS |
1255 | } |
1256 | break; | |
1257 | } | |
1258 | *result = retval; | |
5000d01d | 1259 | return(0); |
2bd0ea18 NS |
1260 | } |
1261 | ||
2bd0ea18 NS |
1262 | /*======================================================================== |
1263 | * Utility routines. | |
1264 | *========================================================================*/ | |
1265 | ||
1266 | /* | |
1267 | * Link a new block into a doubly linked list of blocks (of whatever type). | |
1268 | */ | |
1269 | int /* error */ | |
1270 | xfs_da_blk_link(xfs_da_state_t *state, xfs_da_state_blk_t *old_blk, | |
1271 | xfs_da_state_blk_t *new_blk) | |
1272 | { | |
1273 | xfs_da_blkinfo_t *old_info, *new_info, *tmp_info; | |
1274 | xfs_da_args_t *args; | |
0e266570 | 1275 | int before=0, error; |
a2ceac1f | 1276 | struct xfs_buf *bp; |
2bd0ea18 NS |
1277 | |
1278 | /* | |
1279 | * Set up environment. | |
1280 | */ | |
1281 | args = state->args; | |
1282 | ASSERT(args != NULL); | |
a2ceac1f DC |
1283 | old_info = old_blk->bp->b_addr; |
1284 | new_info = new_blk->bp->b_addr; | |
2bd0ea18 | 1285 | ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC || |
5e656dbb | 1286 | old_blk->magic == XFS_DIR2_LEAFN_MAGIC || |
2bd0ea18 | 1287 | old_blk->magic == XFS_ATTR_LEAF_MAGIC); |
5e656dbb BN |
1288 | ASSERT(old_blk->magic == be16_to_cpu(old_info->magic)); |
1289 | ASSERT(new_blk->magic == be16_to_cpu(new_info->magic)); | |
2bd0ea18 NS |
1290 | ASSERT(old_blk->magic == new_blk->magic); |
1291 | ||
1292 | switch (old_blk->magic) { | |
2bd0ea18 NS |
1293 | case XFS_ATTR_LEAF_MAGIC: |
1294 | before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp); | |
1295 | break; | |
2bd0ea18 | 1296 | case XFS_DIR2_LEAFN_MAGIC: |
2bd0ea18 NS |
1297 | before = xfs_dir2_leafn_order(old_blk->bp, new_blk->bp); |
1298 | break; | |
1299 | case XFS_DA_NODE_MAGIC: | |
1300 | before = xfs_da_node_order(old_blk->bp, new_blk->bp); | |
1301 | break; | |
1302 | } | |
1303 | ||
1304 | /* | |
1305 | * Link blocks in appropriate order. | |
1306 | */ | |
1307 | if (before) { | |
1308 | /* | |
1309 | * Link new block in before existing block. | |
1310 | */ | |
a2ceac1f | 1311 | trace_xfs_da_link_before(args); |
5e656dbb BN |
1312 | new_info->forw = cpu_to_be32(old_blk->blkno); |
1313 | new_info->back = old_info->back; | |
1314 | if (old_info->back) { | |
a2ceac1f | 1315 | error = xfs_da_node_read(args->trans, args->dp, |
5e656dbb BN |
1316 | be32_to_cpu(old_info->back), |
1317 | -1, &bp, args->whichfork); | |
2bd0ea18 NS |
1318 | if (error) |
1319 | return(error); | |
1320 | ASSERT(bp != NULL); | |
a2ceac1f | 1321 | tmp_info = bp->b_addr; |
5e656dbb BN |
1322 | ASSERT(be16_to_cpu(tmp_info->magic) == be16_to_cpu(old_info->magic)); |
1323 | ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno); | |
1324 | tmp_info->forw = cpu_to_be32(new_blk->blkno); | |
a2ceac1f | 1325 | xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1); |
2bd0ea18 | 1326 | } |
5e656dbb | 1327 | old_info->back = cpu_to_be32(new_blk->blkno); |
2bd0ea18 NS |
1328 | } else { |
1329 | /* | |
1330 | * Link new block in after existing block. | |
1331 | */ | |
a2ceac1f | 1332 | trace_xfs_da_link_after(args); |
5e656dbb BN |
1333 | new_info->forw = old_info->forw; |
1334 | new_info->back = cpu_to_be32(old_blk->blkno); | |
1335 | if (old_info->forw) { | |
a2ceac1f | 1336 | error = xfs_da_node_read(args->trans, args->dp, |
5e656dbb BN |
1337 | be32_to_cpu(old_info->forw), |
1338 | -1, &bp, args->whichfork); | |
2bd0ea18 NS |
1339 | if (error) |
1340 | return(error); | |
1341 | ASSERT(bp != NULL); | |
a2ceac1f | 1342 | tmp_info = bp->b_addr; |
5e656dbb BN |
1343 | ASSERT(tmp_info->magic == old_info->magic); |
1344 | ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno); | |
1345 | tmp_info->back = cpu_to_be32(new_blk->blkno); | |
a2ceac1f | 1346 | xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1); |
2bd0ea18 | 1347 | } |
5e656dbb | 1348 | old_info->forw = cpu_to_be32(new_blk->blkno); |
2bd0ea18 NS |
1349 | } |
1350 | ||
a2ceac1f DC |
1351 | xfs_trans_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1); |
1352 | xfs_trans_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1); | |
2bd0ea18 NS |
1353 | return(0); |
1354 | } | |
1355 | ||
2bd0ea18 NS |
1356 | /* |
1357 | * Compare two intermediate nodes for "order". | |
1358 | */ | |
1359 | STATIC int | |
a2ceac1f DC |
1360 | xfs_da_node_order( |
1361 | struct xfs_buf *node1_bp, | |
1362 | struct xfs_buf *node2_bp) | |
2bd0ea18 NS |
1363 | { |
1364 | xfs_da_intnode_t *node1, *node2; | |
1365 | ||
a2ceac1f DC |
1366 | node1 = node1_bp->b_addr; |
1367 | node2 = node2_bp->b_addr; | |
1368 | ASSERT(node1->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) && | |
1369 | node2->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); | |
5e656dbb BN |
1370 | if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) && |
1371 | ((be32_to_cpu(node2->btree[0].hashval) < | |
1372 | be32_to_cpu(node1->btree[0].hashval)) || | |
1373 | (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) < | |
1374 | be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) { | |
2bd0ea18 NS |
1375 | return(1); |
1376 | } | |
1377 | return(0); | |
1378 | } | |
1379 | ||
2bd0ea18 NS |
1380 | /* |
1381 | * Pick up the last hashvalue from an intermediate node. | |
1382 | */ | |
1383 | STATIC uint | |
a2ceac1f DC |
1384 | xfs_da_node_lasthash( |
1385 | struct xfs_buf *bp, | |
1386 | int *count) | |
2bd0ea18 NS |
1387 | { |
1388 | xfs_da_intnode_t *node; | |
1389 | ||
a2ceac1f DC |
1390 | node = bp->b_addr; |
1391 | ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); | |
2bd0ea18 | 1392 | if (count) |
5e656dbb | 1393 | *count = be16_to_cpu(node->hdr.count); |
46eca962 | 1394 | if (!node->hdr.count) |
2bd0ea18 | 1395 | return(0); |
5e656dbb | 1396 | return be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval); |
2bd0ea18 NS |
1397 | } |
1398 | ||
1399 | /* | |
1400 | * Unlink a block from a doubly linked list of blocks. | |
1401 | */ | |
5e656dbb | 1402 | STATIC int /* error */ |
2bd0ea18 NS |
1403 | xfs_da_blk_unlink(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk, |
1404 | xfs_da_state_blk_t *save_blk) | |
1405 | { | |
1406 | xfs_da_blkinfo_t *drop_info, *save_info, *tmp_info; | |
1407 | xfs_da_args_t *args; | |
a2ceac1f | 1408 | struct xfs_buf *bp; |
2bd0ea18 NS |
1409 | int error; |
1410 | ||
1411 | /* | |
1412 | * Set up environment. | |
1413 | */ | |
1414 | args = state->args; | |
1415 | ASSERT(args != NULL); | |
a2ceac1f DC |
1416 | save_info = save_blk->bp->b_addr; |
1417 | drop_info = drop_blk->bp->b_addr; | |
2bd0ea18 | 1418 | ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC || |
5e656dbb | 1419 | save_blk->magic == XFS_DIR2_LEAFN_MAGIC || |
2bd0ea18 | 1420 | save_blk->magic == XFS_ATTR_LEAF_MAGIC); |
5e656dbb BN |
1421 | ASSERT(save_blk->magic == be16_to_cpu(save_info->magic)); |
1422 | ASSERT(drop_blk->magic == be16_to_cpu(drop_info->magic)); | |
2bd0ea18 | 1423 | ASSERT(save_blk->magic == drop_blk->magic); |
5e656dbb BN |
1424 | ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) || |
1425 | (be32_to_cpu(save_info->back) == drop_blk->blkno)); | |
1426 | ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) || | |
1427 | (be32_to_cpu(drop_info->back) == save_blk->blkno)); | |
2bd0ea18 NS |
1428 | |
1429 | /* | |
1430 | * Unlink the leaf block from the doubly linked chain of leaves. | |
1431 | */ | |
5e656dbb | 1432 | if (be32_to_cpu(save_info->back) == drop_blk->blkno) { |
a2ceac1f | 1433 | trace_xfs_da_unlink_back(args); |
5e656dbb BN |
1434 | save_info->back = drop_info->back; |
1435 | if (drop_info->back) { | |
a2ceac1f | 1436 | error = xfs_da_node_read(args->trans, args->dp, |
5e656dbb BN |
1437 | be32_to_cpu(drop_info->back), |
1438 | -1, &bp, args->whichfork); | |
2bd0ea18 NS |
1439 | if (error) |
1440 | return(error); | |
1441 | ASSERT(bp != NULL); | |
a2ceac1f | 1442 | tmp_info = bp->b_addr; |
5e656dbb BN |
1443 | ASSERT(tmp_info->magic == save_info->magic); |
1444 | ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno); | |
1445 | tmp_info->forw = cpu_to_be32(save_blk->blkno); | |
a2ceac1f | 1446 | xfs_trans_log_buf(args->trans, bp, 0, |
2bd0ea18 | 1447 | sizeof(*tmp_info) - 1); |
2bd0ea18 NS |
1448 | } |
1449 | } else { | |
a2ceac1f | 1450 | trace_xfs_da_unlink_forward(args); |
5e656dbb BN |
1451 | save_info->forw = drop_info->forw; |
1452 | if (drop_info->forw) { | |
a2ceac1f | 1453 | error = xfs_da_node_read(args->trans, args->dp, |
5e656dbb BN |
1454 | be32_to_cpu(drop_info->forw), |
1455 | -1, &bp, args->whichfork); | |
2bd0ea18 NS |
1456 | if (error) |
1457 | return(error); | |
1458 | ASSERT(bp != NULL); | |
a2ceac1f | 1459 | tmp_info = bp->b_addr; |
5e656dbb BN |
1460 | ASSERT(tmp_info->magic == save_info->magic); |
1461 | ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno); | |
1462 | tmp_info->back = cpu_to_be32(save_blk->blkno); | |
a2ceac1f | 1463 | xfs_trans_log_buf(args->trans, bp, 0, |
2bd0ea18 | 1464 | sizeof(*tmp_info) - 1); |
2bd0ea18 NS |
1465 | } |
1466 | } | |
1467 | ||
a2ceac1f | 1468 | xfs_trans_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1); |
2bd0ea18 NS |
1469 | return(0); |
1470 | } | |
1471 | ||
1472 | /* | |
1473 | * Move a path "forward" or "!forward" one block at the current level. | |
1474 | * | |
1475 | * This routine will adjust a "path" to point to the next block | |
1476 | * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the | |
1477 | * Btree, including updating pointers to the intermediate nodes between | |
1478 | * the new bottom and the root. | |
1479 | */ | |
1480 | int /* error */ | |
1481 | xfs_da_path_shift(xfs_da_state_t *state, xfs_da_state_path_t *path, | |
1482 | int forward, int release, int *result) | |
1483 | { | |
1484 | xfs_da_state_blk_t *blk; | |
1485 | xfs_da_blkinfo_t *info; | |
1486 | xfs_da_intnode_t *node; | |
1487 | xfs_da_args_t *args; | |
0e266570 | 1488 | xfs_dablk_t blkno=0; |
2bd0ea18 NS |
1489 | int level, error; |
1490 | ||
a2ceac1f DC |
1491 | trace_xfs_da_path_shift(state->args); |
1492 | ||
2bd0ea18 NS |
1493 | /* |
1494 | * Roll up the Btree looking for the first block where our | |
1495 | * current index is not at the edge of the block. Note that | |
1496 | * we skip the bottom layer because we want the sibling block. | |
1497 | */ | |
1498 | args = state->args; | |
1499 | ASSERT(args != NULL); | |
1500 | ASSERT(path != NULL); | |
1501 | ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH)); | |
1502 | level = (path->active-1) - 1; /* skip bottom layer in path */ | |
1503 | for (blk = &path->blk[level]; level >= 0; blk--, level--) { | |
1504 | ASSERT(blk->bp != NULL); | |
a2ceac1f DC |
1505 | node = blk->bp->b_addr; |
1506 | ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); | |
5e656dbb | 1507 | if (forward && (blk->index < be16_to_cpu(node->hdr.count)-1)) { |
2bd0ea18 | 1508 | blk->index++; |
5e656dbb | 1509 | blkno = be32_to_cpu(node->btree[blk->index].before); |
2bd0ea18 NS |
1510 | break; |
1511 | } else if (!forward && (blk->index > 0)) { | |
1512 | blk->index--; | |
5e656dbb | 1513 | blkno = be32_to_cpu(node->btree[blk->index].before); |
2bd0ea18 NS |
1514 | break; |
1515 | } | |
1516 | } | |
1517 | if (level < 0) { | |
1518 | *result = XFS_ERROR(ENOENT); /* we're out of our tree */ | |
5e656dbb | 1519 | ASSERT(args->op_flags & XFS_DA_OP_OKNOENT); |
2bd0ea18 NS |
1520 | return(0); |
1521 | } | |
1522 | ||
1523 | /* | |
1524 | * Roll down the edge of the subtree until we reach the | |
1525 | * same depth we were at originally. | |
1526 | */ | |
1527 | for (blk++, level++; level < path->active; blk++, level++) { | |
1528 | /* | |
1529 | * Release the old block. | |
1530 | * (if it's dirty, trans won't actually let go) | |
1531 | */ | |
1532 | if (release) | |
a2ceac1f | 1533 | xfs_trans_brelse(args->trans, blk->bp); |
2bd0ea18 NS |
1534 | |
1535 | /* | |
1536 | * Read the next child block. | |
1537 | */ | |
1538 | blk->blkno = blkno; | |
a2ceac1f DC |
1539 | error = xfs_da_node_read(args->trans, args->dp, blkno, -1, |
1540 | &blk->bp, args->whichfork); | |
2bd0ea18 NS |
1541 | if (error) |
1542 | return(error); | |
1543 | ASSERT(blk->bp != NULL); | |
a2ceac1f DC |
1544 | info = blk->bp->b_addr; |
1545 | ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) || | |
1546 | info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || | |
1547 | info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC)); | |
5e656dbb BN |
1548 | blk->magic = be16_to_cpu(info->magic); |
1549 | if (blk->magic == XFS_DA_NODE_MAGIC) { | |
2bd0ea18 | 1550 | node = (xfs_da_intnode_t *)info; |
5e656dbb | 1551 | blk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval); |
2bd0ea18 NS |
1552 | if (forward) |
1553 | blk->index = 0; | |
1554 | else | |
5e656dbb BN |
1555 | blk->index = be16_to_cpu(node->hdr.count)-1; |
1556 | blkno = be32_to_cpu(node->btree[blk->index].before); | |
2bd0ea18 NS |
1557 | } else { |
1558 | ASSERT(level == path->active-1); | |
1559 | blk->index = 0; | |
1560 | switch(blk->magic) { | |
2bd0ea18 NS |
1561 | case XFS_ATTR_LEAF_MAGIC: |
1562 | blk->hashval = xfs_attr_leaf_lasthash(blk->bp, | |
1563 | NULL); | |
1564 | break; | |
2bd0ea18 | 1565 | case XFS_DIR2_LEAFN_MAGIC: |
2bd0ea18 NS |
1566 | blk->hashval = xfs_dir2_leafn_lasthash(blk->bp, |
1567 | NULL); | |
1568 | break; | |
1569 | default: | |
1570 | ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC || | |
5e656dbb | 1571 | blk->magic == XFS_DIR2_LEAFN_MAGIC); |
2bd0ea18 NS |
1572 | break; |
1573 | } | |
1574 | } | |
1575 | } | |
1576 | *result = 0; | |
1577 | return(0); | |
1578 | } | |
1579 | ||
1580 | ||
1581 | /*======================================================================== | |
1582 | * Utility routines. | |
1583 | *========================================================================*/ | |
1584 | ||
1585 | /* | |
1586 | * Implement a simple hash on a character string. | |
1587 | * Rotate the hash value by 7 bits, then XOR each character in. | |
1588 | * This is implemented with some source-level loop unrolling. | |
1589 | */ | |
1590 | xfs_dahash_t | |
56b2de80 | 1591 | xfs_da_hashname(const __uint8_t *name, int namelen) |
2bd0ea18 NS |
1592 | { |
1593 | xfs_dahash_t hash; | |
1594 | ||
2bd0ea18 NS |
1595 | /* |
1596 | * Do four characters at a time as long as we can. | |
1597 | */ | |
f9c48a87 | 1598 | for (hash = 0; namelen >= 4; namelen -= 4, name += 4) |
2bd0ea18 | 1599 | hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^ |
f9c48a87 NS |
1600 | (name[3] << 0) ^ rol32(hash, 7 * 4); |
1601 | ||
2bd0ea18 NS |
1602 | /* |
1603 | * Now do the rest of the characters. | |
1604 | */ | |
1605 | switch (namelen) { | |
1606 | case 3: | |
1607 | return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^ | |
f9c48a87 | 1608 | rol32(hash, 7 * 3); |
2bd0ea18 | 1609 | case 2: |
f9c48a87 | 1610 | return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2); |
2bd0ea18 | 1611 | case 1: |
f9c48a87 NS |
1612 | return (name[0] << 0) ^ rol32(hash, 7 * 1); |
1613 | default: /* case 0: */ | |
2bd0ea18 NS |
1614 | return hash; |
1615 | } | |
2bd0ea18 NS |
1616 | } |
1617 | ||
51ca7008 | 1618 | enum xfs_dacmp |
5e656dbb BN |
1619 | xfs_da_compname( |
1620 | struct xfs_da_args *args, | |
56b2de80 DC |
1621 | const unsigned char *name, |
1622 | int len) | |
51ca7008 | 1623 | { |
5e656dbb | 1624 | return (args->namelen == len && memcmp(args->name, name, len) == 0) ? |
51ca7008 BN |
1625 | XFS_CMP_EXACT : XFS_CMP_DIFFERENT; |
1626 | } | |
1627 | ||
5e656dbb BN |
1628 | static xfs_dahash_t |
1629 | xfs_default_hashname( | |
1630 | struct xfs_name *name) | |
1631 | { | |
1632 | return xfs_da_hashname(name->name, name->len); | |
1633 | } | |
1634 | ||
51ca7008 | 1635 | const struct xfs_nameops xfs_default_nameops = { |
5e656dbb | 1636 | .hashname = xfs_default_hashname, |
51ca7008 BN |
1637 | .compname = xfs_da_compname |
1638 | }; | |
1639 | ||
2bd0ea18 | 1640 | int |
a2ceac1f DC |
1641 | xfs_da_grow_inode_int( |
1642 | struct xfs_da_args *args, | |
1643 | xfs_fileoff_t *bno, | |
1644 | int count) | |
2bd0ea18 | 1645 | { |
a2ceac1f DC |
1646 | struct xfs_trans *tp = args->trans; |
1647 | struct xfs_inode *dp = args->dp; | |
1648 | int w = args->whichfork; | |
1649 | xfs_drfsbno_t nblks = dp->i_d.di_nblocks; | |
1650 | struct xfs_bmbt_irec map, *mapp; | |
1651 | int nmap, error, got, i, mapi; | |
56b2de80 | 1652 | |
2bd0ea18 NS |
1653 | /* |
1654 | * Find a spot in the file space to put the new block. | |
1655 | */ | |
a2ceac1f DC |
1656 | error = xfs_bmap_first_unused(tp, dp, count, bno, w); |
1657 | if (error) | |
2bd0ea18 | 1658 | return error; |
a2ceac1f | 1659 | |
2bd0ea18 NS |
1660 | /* |
1661 | * Try mapping it in one filesystem block. | |
1662 | */ | |
1663 | nmap = 1; | |
1664 | ASSERT(args->firstblock != NULL); | |
a2ceac1f DC |
1665 | error = xfs_bmapi_write(tp, dp, *bno, count, |
1666 | xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG, | |
2bd0ea18 | 1667 | args->firstblock, args->total, &map, &nmap, |
a2ceac1f DC |
1668 | args->flist); |
1669 | if (error) | |
2bd0ea18 | 1670 | return error; |
a2ceac1f | 1671 | |
2bd0ea18 NS |
1672 | ASSERT(nmap <= 1); |
1673 | if (nmap == 1) { | |
1674 | mapp = ↦ | |
1675 | mapi = 1; | |
a2ceac1f DC |
1676 | } else if (nmap == 0 && count > 1) { |
1677 | xfs_fileoff_t b; | |
1678 | int c; | |
1679 | ||
1680 | /* | |
1681 | * If we didn't get it and the block might work if fragmented, | |
1682 | * try without the CONTIG flag. Loop until we get it all. | |
1683 | */ | |
2bd0ea18 | 1684 | mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP); |
a2ceac1f | 1685 | for (b = *bno, mapi = 0; b < *bno + count; ) { |
2bd0ea18 | 1686 | nmap = MIN(XFS_BMAP_MAX_NMAP, count); |
a2ceac1f DC |
1687 | c = (int)(*bno + count - b); |
1688 | error = xfs_bmapi_write(tp, dp, b, c, | |
1689 | xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA, | |
2bd0ea18 | 1690 | args->firstblock, args->total, |
a2ceac1f DC |
1691 | &mapp[mapi], &nmap, args->flist); |
1692 | if (error) | |
1693 | goto out_free_map; | |
2bd0ea18 NS |
1694 | if (nmap < 1) |
1695 | break; | |
1696 | mapi += nmap; | |
1697 | b = mapp[mapi - 1].br_startoff + | |
1698 | mapp[mapi - 1].br_blockcount; | |
1699 | } | |
1700 | } else { | |
2bd0ea18 NS |
1701 | mapi = 0; |
1702 | mapp = NULL; | |
1703 | } | |
a2ceac1f | 1704 | |
2bd0ea18 NS |
1705 | /* |
1706 | * Count the blocks we got, make sure it matches the total. | |
1707 | */ | |
1708 | for (i = 0, got = 0; i < mapi; i++) | |
1709 | got += mapp[i].br_blockcount; | |
a2ceac1f | 1710 | if (got != count || mapp[0].br_startoff != *bno || |
2bd0ea18 | 1711 | mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount != |
a2ceac1f DC |
1712 | *bno + count) { |
1713 | error = XFS_ERROR(ENOSPC); | |
1714 | goto out_free_map; | |
2bd0ea18 | 1715 | } |
a2ceac1f | 1716 | |
56b2de80 DC |
1717 | /* account for newly allocated blocks in reserved blocks total */ |
1718 | args->total -= dp->i_d.di_nblocks - nblks; | |
a2ceac1f DC |
1719 | |
1720 | out_free_map: | |
1721 | if (mapp != &map) | |
1722 | kmem_free(mapp); | |
1723 | return error; | |
1724 | } | |
1725 | ||
1726 | /* | |
1727 | * Add a block to the btree ahead of the file. | |
1728 | * Return the new block number to the caller. | |
1729 | */ | |
1730 | int | |
1731 | xfs_da_grow_inode( | |
1732 | struct xfs_da_args *args, | |
1733 | xfs_dablk_t *new_blkno) | |
1734 | { | |
1735 | xfs_fileoff_t bno; | |
1736 | int count; | |
1737 | int error; | |
1738 | ||
1739 | trace_xfs_da_grow_inode(args); | |
1740 | ||
1741 | if (args->whichfork == XFS_DATA_FORK) { | |
1742 | bno = args->dp->i_mount->m_dirleafblk; | |
1743 | count = args->dp->i_mount->m_dirblkfsbs; | |
1744 | } else { | |
1745 | bno = 0; | |
1746 | count = 1; | |
1747 | } | |
1748 | ||
1749 | error = xfs_da_grow_inode_int(args, &bno, count); | |
1750 | if (!error) | |
1751 | *new_blkno = (xfs_dablk_t)bno; | |
1752 | return error; | |
2bd0ea18 NS |
1753 | } |
1754 | ||
2bd0ea18 | 1755 | /* |
dfc130f3 | 1756 | * Ick. We need to always be able to remove a btree block, even |
2bd0ea18 NS |
1757 | * if there's no space reservation because the filesystem is full. |
1758 | * This is called if xfs_bunmapi on a btree block fails due to ENOSPC. | |
1759 | * It swaps the target block with the last block in the file. The | |
1760 | * last block in the file can always be removed since it can't cause | |
1761 | * a bmap btree split to do that. | |
1762 | */ | |
1763 | STATIC int | |
a2ceac1f DC |
1764 | xfs_da_swap_lastblock( |
1765 | xfs_da_args_t *args, | |
1766 | xfs_dablk_t *dead_blknop, | |
1767 | struct xfs_buf **dead_bufp) | |
2bd0ea18 NS |
1768 | { |
1769 | xfs_dablk_t dead_blkno, last_blkno, sib_blkno, par_blkno; | |
a2ceac1f | 1770 | struct xfs_buf *dead_buf, *last_buf, *sib_buf, *par_buf; |
2bd0ea18 NS |
1771 | xfs_fileoff_t lastoff; |
1772 | xfs_inode_t *ip; | |
1773 | xfs_trans_t *tp; | |
1774 | xfs_mount_t *mp; | |
1775 | int error, w, entno, level, dead_level; | |
1776 | xfs_da_blkinfo_t *dead_info, *sib_info; | |
1777 | xfs_da_intnode_t *par_node, *dead_node; | |
2bd0ea18 NS |
1778 | xfs_dir2_leaf_t *dead_leaf2; |
1779 | xfs_dahash_t dead_hash; | |
1780 | ||
a2ceac1f DC |
1781 | trace_xfs_da_swap_lastblock(args); |
1782 | ||
2bd0ea18 NS |
1783 | dead_buf = *dead_bufp; |
1784 | dead_blkno = *dead_blknop; | |
1785 | tp = args->trans; | |
1786 | ip = args->dp; | |
1787 | w = args->whichfork; | |
1788 | ASSERT(w == XFS_DATA_FORK); | |
1789 | mp = ip->i_mount; | |
5e656dbb BN |
1790 | lastoff = mp->m_dirfreeblk; |
1791 | error = xfs_bmap_last_before(tp, ip, &lastoff, w); | |
2bd0ea18 NS |
1792 | if (error) |
1793 | return error; | |
4ca431fc NS |
1794 | if (unlikely(lastoff == 0)) { |
1795 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW, | |
1796 | mp); | |
2bd0ea18 | 1797 | return XFS_ERROR(EFSCORRUPTED); |
4ca431fc | 1798 | } |
2bd0ea18 NS |
1799 | /* |
1800 | * Read the last block in the btree space. | |
1801 | */ | |
1802 | last_blkno = (xfs_dablk_t)lastoff - mp->m_dirblkfsbs; | |
a2ceac1f DC |
1803 | error = xfs_da_node_read(tp, ip, last_blkno, -1, &last_buf, w); |
1804 | if (error) | |
2bd0ea18 NS |
1805 | return error; |
1806 | /* | |
1807 | * Copy the last block into the dead buffer and log it. | |
1808 | */ | |
a2ceac1f DC |
1809 | memcpy(dead_buf->b_addr, last_buf->b_addr, mp->m_dirblksize); |
1810 | xfs_trans_log_buf(tp, dead_buf, 0, mp->m_dirblksize - 1); | |
1811 | dead_info = dead_buf->b_addr; | |
2bd0ea18 NS |
1812 | /* |
1813 | * Get values from the moved block. | |
1814 | */ | |
a2ceac1f | 1815 | if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC)) { |
2bd0ea18 NS |
1816 | dead_leaf2 = (xfs_dir2_leaf_t *)dead_info; |
1817 | dead_level = 0; | |
5e656dbb | 1818 | dead_hash = be32_to_cpu(dead_leaf2->ents[be16_to_cpu(dead_leaf2->hdr.count) - 1].hashval); |
2bd0ea18 | 1819 | } else { |
a2ceac1f | 1820 | ASSERT(dead_info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
2bd0ea18 | 1821 | dead_node = (xfs_da_intnode_t *)dead_info; |
5e656dbb BN |
1822 | dead_level = be16_to_cpu(dead_node->hdr.level); |
1823 | dead_hash = be32_to_cpu(dead_node->btree[be16_to_cpu(dead_node->hdr.count) - 1].hashval); | |
2bd0ea18 NS |
1824 | } |
1825 | sib_buf = par_buf = NULL; | |
1826 | /* | |
1827 | * If the moved block has a left sibling, fix up the pointers. | |
1828 | */ | |
5e656dbb | 1829 | if ((sib_blkno = be32_to_cpu(dead_info->back))) { |
a2ceac1f DC |
1830 | error = xfs_da_node_read(tp, ip, sib_blkno, -1, &sib_buf, w); |
1831 | if (error) | |
2bd0ea18 | 1832 | goto done; |
a2ceac1f | 1833 | sib_info = sib_buf->b_addr; |
4ca431fc | 1834 | if (unlikely( |
5e656dbb BN |
1835 | be32_to_cpu(sib_info->forw) != last_blkno || |
1836 | sib_info->magic != dead_info->magic)) { | |
4ca431fc NS |
1837 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)", |
1838 | XFS_ERRLEVEL_LOW, mp); | |
2bd0ea18 NS |
1839 | error = XFS_ERROR(EFSCORRUPTED); |
1840 | goto done; | |
1841 | } | |
5e656dbb | 1842 | sib_info->forw = cpu_to_be32(dead_blkno); |
a2ceac1f | 1843 | xfs_trans_log_buf(tp, sib_buf, |
2bd0ea18 NS |
1844 | XFS_DA_LOGRANGE(sib_info, &sib_info->forw, |
1845 | sizeof(sib_info->forw))); | |
2bd0ea18 NS |
1846 | sib_buf = NULL; |
1847 | } | |
1848 | /* | |
1849 | * If the moved block has a right sibling, fix up the pointers. | |
1850 | */ | |
5e656dbb | 1851 | if ((sib_blkno = be32_to_cpu(dead_info->forw))) { |
a2ceac1f DC |
1852 | error = xfs_da_node_read(tp, ip, sib_blkno, -1, &sib_buf, w); |
1853 | if (error) | |
2bd0ea18 | 1854 | goto done; |
a2ceac1f | 1855 | sib_info = sib_buf->b_addr; |
4ca431fc | 1856 | if (unlikely( |
5e656dbb BN |
1857 | be32_to_cpu(sib_info->back) != last_blkno || |
1858 | sib_info->magic != dead_info->magic)) { | |
4ca431fc NS |
1859 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)", |
1860 | XFS_ERRLEVEL_LOW, mp); | |
2bd0ea18 NS |
1861 | error = XFS_ERROR(EFSCORRUPTED); |
1862 | goto done; | |
1863 | } | |
5e656dbb | 1864 | sib_info->back = cpu_to_be32(dead_blkno); |
a2ceac1f | 1865 | xfs_trans_log_buf(tp, sib_buf, |
2bd0ea18 NS |
1866 | XFS_DA_LOGRANGE(sib_info, &sib_info->back, |
1867 | sizeof(sib_info->back))); | |
2bd0ea18 NS |
1868 | sib_buf = NULL; |
1869 | } | |
5e656dbb | 1870 | par_blkno = mp->m_dirleafblk; |
2bd0ea18 NS |
1871 | level = -1; |
1872 | /* | |
1873 | * Walk down the tree looking for the parent of the moved block. | |
1874 | */ | |
1875 | for (;;) { | |
a2ceac1f DC |
1876 | error = xfs_da_node_read(tp, ip, par_blkno, -1, &par_buf, w); |
1877 | if (error) | |
2bd0ea18 | 1878 | goto done; |
a2ceac1f DC |
1879 | par_node = par_buf->b_addr; |
1880 | if (unlikely(par_node->hdr.info.magic != | |
1881 | cpu_to_be16(XFS_DA_NODE_MAGIC) || | |
5e656dbb | 1882 | (level >= 0 && level != be16_to_cpu(par_node->hdr.level) + 1))) { |
4ca431fc NS |
1883 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)", |
1884 | XFS_ERRLEVEL_LOW, mp); | |
2bd0ea18 NS |
1885 | error = XFS_ERROR(EFSCORRUPTED); |
1886 | goto done; | |
1887 | } | |
5e656dbb | 1888 | level = be16_to_cpu(par_node->hdr.level); |
2bd0ea18 | 1889 | for (entno = 0; |
5e656dbb BN |
1890 | entno < be16_to_cpu(par_node->hdr.count) && |
1891 | be32_to_cpu(par_node->btree[entno].hashval) < dead_hash; | |
2bd0ea18 NS |
1892 | entno++) |
1893 | continue; | |
5e656dbb | 1894 | if (unlikely(entno == be16_to_cpu(par_node->hdr.count))) { |
4ca431fc NS |
1895 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)", |
1896 | XFS_ERRLEVEL_LOW, mp); | |
2bd0ea18 NS |
1897 | error = XFS_ERROR(EFSCORRUPTED); |
1898 | goto done; | |
1899 | } | |
5e656dbb | 1900 | par_blkno = be32_to_cpu(par_node->btree[entno].before); |
2bd0ea18 NS |
1901 | if (level == dead_level + 1) |
1902 | break; | |
a2ceac1f | 1903 | xfs_trans_brelse(tp, par_buf); |
2bd0ea18 NS |
1904 | par_buf = NULL; |
1905 | } | |
1906 | /* | |
1907 | * We're in the right parent block. | |
1908 | * Look for the right entry. | |
1909 | */ | |
1910 | for (;;) { | |
1911 | for (; | |
5e656dbb BN |
1912 | entno < be16_to_cpu(par_node->hdr.count) && |
1913 | be32_to_cpu(par_node->btree[entno].before) != last_blkno; | |
2bd0ea18 NS |
1914 | entno++) |
1915 | continue; | |
5e656dbb | 1916 | if (entno < be16_to_cpu(par_node->hdr.count)) |
2bd0ea18 | 1917 | break; |
5e656dbb | 1918 | par_blkno = be32_to_cpu(par_node->hdr.info.forw); |
a2ceac1f | 1919 | xfs_trans_brelse(tp, par_buf); |
2bd0ea18 | 1920 | par_buf = NULL; |
4ca431fc NS |
1921 | if (unlikely(par_blkno == 0)) { |
1922 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)", | |
1923 | XFS_ERRLEVEL_LOW, mp); | |
2bd0ea18 NS |
1924 | error = XFS_ERROR(EFSCORRUPTED); |
1925 | goto done; | |
1926 | } | |
a2ceac1f DC |
1927 | error = xfs_da_node_read(tp, ip, par_blkno, -1, &par_buf, w); |
1928 | if (error) | |
2bd0ea18 | 1929 | goto done; |
a2ceac1f | 1930 | par_node = par_buf->b_addr; |
4ca431fc | 1931 | if (unlikely( |
5e656dbb | 1932 | be16_to_cpu(par_node->hdr.level) != level || |
a2ceac1f | 1933 | par_node->hdr.info.magic != cpu_to_be16(XFS_DA_NODE_MAGIC))) { |
4ca431fc NS |
1934 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)", |
1935 | XFS_ERRLEVEL_LOW, mp); | |
2bd0ea18 NS |
1936 | error = XFS_ERROR(EFSCORRUPTED); |
1937 | goto done; | |
1938 | } | |
1939 | entno = 0; | |
1940 | } | |
1941 | /* | |
1942 | * Update the parent entry pointing to the moved block. | |
1943 | */ | |
5e656dbb | 1944 | par_node->btree[entno].before = cpu_to_be32(dead_blkno); |
a2ceac1f | 1945 | xfs_trans_log_buf(tp, par_buf, |
2bd0ea18 NS |
1946 | XFS_DA_LOGRANGE(par_node, &par_node->btree[entno].before, |
1947 | sizeof(par_node->btree[entno].before))); | |
2bd0ea18 NS |
1948 | *dead_blknop = last_blkno; |
1949 | *dead_bufp = last_buf; | |
1950 | return 0; | |
1951 | done: | |
1952 | if (par_buf) | |
a2ceac1f | 1953 | xfs_trans_brelse(tp, par_buf); |
2bd0ea18 | 1954 | if (sib_buf) |
a2ceac1f DC |
1955 | xfs_trans_brelse(tp, sib_buf); |
1956 | xfs_trans_brelse(tp, last_buf); | |
2bd0ea18 NS |
1957 | return error; |
1958 | } | |
1959 | ||
1960 | /* | |
1961 | * Remove a btree block from a directory or attribute. | |
1962 | */ | |
1963 | int | |
a2ceac1f DC |
1964 | xfs_da_shrink_inode( |
1965 | xfs_da_args_t *args, | |
1966 | xfs_dablk_t dead_blkno, | |
1967 | struct xfs_buf *dead_buf) | |
2bd0ea18 NS |
1968 | { |
1969 | xfs_inode_t *dp; | |
1970 | int done, error, w, count; | |
2bd0ea18 NS |
1971 | xfs_trans_t *tp; |
1972 | xfs_mount_t *mp; | |
1973 | ||
a2ceac1f DC |
1974 | trace_xfs_da_shrink_inode(args); |
1975 | ||
2bd0ea18 NS |
1976 | dp = args->dp; |
1977 | w = args->whichfork; | |
1978 | tp = args->trans; | |
1979 | mp = dp->i_mount; | |
5e656dbb | 1980 | if (w == XFS_DATA_FORK) |
2bd0ea18 NS |
1981 | count = mp->m_dirblkfsbs; |
1982 | else | |
1983 | count = 1; | |
1984 | for (;;) { | |
1985 | /* | |
1986 | * Remove extents. If we get ENOSPC for a dir we have to move | |
1987 | * the last block to the place we want to kill. | |
1988 | */ | |
1989 | if ((error = xfs_bunmapi(tp, dp, dead_blkno, count, | |
56b2de80 DC |
1990 | xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA, |
1991 | 0, args->firstblock, args->flist, | |
2bd0ea18 NS |
1992 | &done)) == ENOSPC) { |
1993 | if (w != XFS_DATA_FORK) | |
5e656dbb | 1994 | break; |
0e266570 NS |
1995 | if ((error = xfs_da_swap_lastblock(args, &dead_blkno, |
1996 | &dead_buf))) | |
5e656dbb BN |
1997 | break; |
1998 | } else { | |
2bd0ea18 | 1999 | break; |
2bd0ea18 NS |
2000 | } |
2001 | } | |
a2ceac1f | 2002 | xfs_trans_binval(tp, dead_buf); |
2bd0ea18 NS |
2003 | return error; |
2004 | } | |
2005 | ||
2006 | /* | |
2007 | * See if the mapping(s) for this btree block are valid, i.e. | |
2008 | * don't contain holes, are logically contiguous, and cover the whole range. | |
2009 | */ | |
2010 | STATIC int | |
2011 | xfs_da_map_covers_blocks( | |
2012 | int nmap, | |
dfc130f3 | 2013 | xfs_bmbt_irec_t *mapp, |
2bd0ea18 NS |
2014 | xfs_dablk_t bno, |
2015 | int count) | |
2016 | { | |
2017 | int i; | |
2018 | xfs_fileoff_t off; | |
2019 | ||
2020 | for (i = 0, off = bno; i < nmap; i++) { | |
2021 | if (mapp[i].br_startblock == HOLESTARTBLOCK || | |
2022 | mapp[i].br_startblock == DELAYSTARTBLOCK) { | |
2bd0ea18 NS |
2023 | return 0; |
2024 | } | |
2025 | if (off != mapp[i].br_startoff) { | |
2bd0ea18 NS |
2026 | return 0; |
2027 | } | |
2028 | off += mapp[i].br_blockcount; | |
2029 | } | |
2030 | return off == bno + count; | |
2031 | } | |
2032 | ||
2033 | /* | |
a2ceac1f DC |
2034 | * Convert a struct xfs_bmbt_irec to a struct xfs_buf_map. |
2035 | * | |
2036 | * For the single map case, it is assumed that the caller has provided a pointer | |
2037 | * to a valid xfs_buf_map. For the multiple map case, this function will | |
2038 | * allocate the xfs_buf_map to hold all the maps and replace the caller's single | |
2039 | * map pointer with the allocated map. | |
2bd0ea18 | 2040 | */ |
a2ceac1f DC |
2041 | static int |
2042 | xfs_buf_map_from_irec( | |
2043 | struct xfs_mount *mp, | |
2044 | struct xfs_buf_map **mapp, | |
2045 | unsigned int *nmaps, | |
2046 | struct xfs_bmbt_irec *irecs, | |
2047 | unsigned int nirecs) | |
2bd0ea18 | 2048 | { |
a2ceac1f DC |
2049 | struct xfs_buf_map *map; |
2050 | int i; | |
2051 | ||
2052 | ASSERT(*nmaps == 1); | |
2053 | ASSERT(nirecs >= 1); | |
2054 | ||
2055 | if (nirecs > 1) { | |
2056 | map = kmem_zalloc(nirecs * sizeof(struct xfs_buf_map), KM_SLEEP); | |
2057 | if (!map) | |
2058 | return ENOMEM; | |
2059 | *mapp = map; | |
2060 | } | |
2061 | ||
2062 | *nmaps = nirecs; | |
2063 | map = *mapp; | |
2064 | for (i = 0; i < *nmaps; i++) { | |
2065 | ASSERT(irecs[i].br_startblock != DELAYSTARTBLOCK && | |
2066 | irecs[i].br_startblock != HOLESTARTBLOCK); | |
2067 | map[i].bm_bn = XFS_FSB_TO_DADDR(mp, irecs[i].br_startblock); | |
2068 | map[i].bm_len = XFS_FSB_TO_BB(mp, irecs[i].br_blockcount); | |
2069 | } | |
2070 | return 0; | |
2071 | } | |
2072 | ||
2073 | /* | |
2074 | * Map the block we are given ready for reading. There are three possible return | |
2075 | * values: | |
2076 | * -1 - will be returned if we land in a hole and mappedbno == -2 so the | |
2077 | * caller knows not to execute a subsequent read. | |
2078 | * 0 - if we mapped the block successfully | |
2079 | * >0 - positive error number if there was an error. | |
2080 | */ | |
2081 | static int | |
2082 | xfs_dabuf_map( | |
2083 | struct xfs_trans *trans, | |
2084 | struct xfs_inode *dp, | |
2085 | xfs_dablk_t bno, | |
2086 | xfs_daddr_t mappedbno, | |
2087 | int whichfork, | |
2088 | struct xfs_buf_map **map, | |
2089 | int *nmaps) | |
2090 | { | |
2091 | struct xfs_mount *mp = dp->i_mount; | |
2092 | int nfsb; | |
2093 | int error = 0; | |
2094 | struct xfs_bmbt_irec irec; | |
2095 | struct xfs_bmbt_irec *irecs = &irec; | |
2096 | int nirecs; | |
2097 | ||
2098 | ASSERT(map && *map); | |
2099 | ASSERT(*nmaps == 1); | |
2bd0ea18 | 2100 | |
5e656dbb | 2101 | nfsb = (whichfork == XFS_DATA_FORK) ? mp->m_dirblkfsbs : 1; |
a2ceac1f | 2102 | |
2bd0ea18 NS |
2103 | /* |
2104 | * Caller doesn't have a mapping. -2 means don't complain | |
2105 | * if we land in a hole. | |
2106 | */ | |
2107 | if (mappedbno == -1 || mappedbno == -2) { | |
2108 | /* | |
2109 | * Optimize the one-block case. | |
2110 | */ | |
a2ceac1f DC |
2111 | if (nfsb != 1) |
2112 | irecs = kmem_zalloc(sizeof(irec) * nfsb, KM_SLEEP); | |
2bd0ea18 | 2113 | |
a2ceac1f DC |
2114 | nirecs = nfsb; |
2115 | error = xfs_bmapi_read(dp, (xfs_fileoff_t)bno, nfsb, irecs, | |
2116 | &nirecs, xfs_bmapi_aflag(whichfork)); | |
2117 | if (error) | |
2118 | goto out; | |
2bd0ea18 | 2119 | } else { |
a2ceac1f DC |
2120 | irecs->br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno); |
2121 | irecs->br_startoff = (xfs_fileoff_t)bno; | |
2122 | irecs->br_blockcount = nfsb; | |
2123 | irecs->br_state = 0; | |
2124 | nirecs = 1; | |
2bd0ea18 | 2125 | } |
a2ceac1f DC |
2126 | |
2127 | if (!xfs_da_map_covers_blocks(nirecs, irecs, bno, nfsb)) { | |
2128 | error = mappedbno == -2 ? -1 : XFS_ERROR(EFSCORRUPTED); | |
4ca431fc NS |
2129 | if (unlikely(error == EFSCORRUPTED)) { |
2130 | if (xfs_error_level >= XFS_ERRLEVEL_LOW) { | |
a2ceac1f DC |
2131 | int i; |
2132 | xfs_alert(mp, "%s: bno %lld dir: inode %lld", | |
2133 | __func__, (long long)bno, | |
4ca431fc | 2134 | (long long)dp->i_ino); |
a2ceac1f DC |
2135 | for (i = 0; i < *nmaps; i++) { |
2136 | xfs_alert(mp, | |
2137 | "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d", | |
4ca431fc | 2138 | i, |
a2ceac1f DC |
2139 | (long long)irecs[i].br_startoff, |
2140 | (long long)irecs[i].br_startblock, | |
2141 | (long long)irecs[i].br_blockcount, | |
2142 | irecs[i].br_state); | |
4ca431fc NS |
2143 | } |
2144 | } | |
2145 | XFS_ERROR_REPORT("xfs_da_do_buf(1)", | |
2146 | XFS_ERRLEVEL_LOW, mp); | |
2147 | } | |
a2ceac1f | 2148 | goto out; |
2bd0ea18 | 2149 | } |
a2ceac1f DC |
2150 | error = xfs_buf_map_from_irec(mp, map, nmaps, irecs, nirecs); |
2151 | out: | |
2152 | if (irecs != &irec) | |
2153 | kmem_free(irecs); | |
2154 | return error; | |
2155 | } | |
2156 | ||
2157 | /* | |
2158 | * Get a buffer for the dir/attr block. | |
2159 | */ | |
2160 | int | |
2161 | xfs_da_get_buf( | |
2162 | struct xfs_trans *trans, | |
2163 | struct xfs_inode *dp, | |
2164 | xfs_dablk_t bno, | |
2165 | xfs_daddr_t mappedbno, | |
2166 | struct xfs_buf **bpp, | |
2167 | int whichfork) | |
2168 | { | |
2169 | struct xfs_buf *bp; | |
2170 | struct xfs_buf_map map; | |
2171 | struct xfs_buf_map *mapp; | |
2172 | int nmap; | |
2173 | int error; | |
2174 | ||
2175 | *bpp = NULL; | |
2176 | mapp = ↦ | |
2177 | nmap = 1; | |
2178 | error = xfs_dabuf_map(trans, dp, bno, mappedbno, whichfork, | |
2179 | &mapp, &nmap); | |
2180 | if (error) { | |
2181 | /* mapping a hole is not an error, but we don't continue */ | |
2182 | if (error == -1) | |
2bd0ea18 | 2183 | error = 0; |
a2ceac1f | 2184 | goto out_free; |
2bd0ea18 | 2185 | } |
a2ceac1f DC |
2186 | |
2187 | bp = xfs_trans_get_buf_map(trans, dp->i_mount->m_ddev_targp, | |
2188 | mapp, nmap, 0); | |
2189 | error = bp ? bp->b_error : XFS_ERROR(EIO); | |
2190 | if (error) { | |
2191 | xfs_trans_brelse(trans, bp); | |
2192 | goto out_free; | |
2193 | } | |
2194 | ||
2195 | *bpp = bp; | |
2196 | ||
2197 | out_free: | |
2198 | if (mapp != &map) | |
2199 | kmem_free(mapp); | |
2200 | ||
2201 | return error; | |
2202 | } | |
2203 | ||
2204 | /* | |
2205 | * Get a buffer for the dir/attr block, fill in the contents. | |
2206 | */ | |
2207 | int | |
2208 | xfs_da_read_buf( | |
2209 | struct xfs_trans *trans, | |
2210 | struct xfs_inode *dp, | |
2211 | xfs_dablk_t bno, | |
2212 | xfs_daddr_t mappedbno, | |
2213 | struct xfs_buf **bpp, | |
2214 | int whichfork, | |
2215 | const struct xfs_buf_ops *ops) | |
2216 | { | |
2217 | struct xfs_buf *bp; | |
2218 | struct xfs_buf_map map; | |
2219 | struct xfs_buf_map *mapp; | |
2220 | int nmap; | |
2221 | int error; | |
2222 | ||
2223 | *bpp = NULL; | |
2224 | mapp = ↦ | |
2225 | nmap = 1; | |
2226 | error = xfs_dabuf_map(trans, dp, bno, mappedbno, whichfork, | |
2227 | &mapp, &nmap); | |
2228 | if (error) { | |
2229 | /* mapping a hole is not an error, but we don't continue */ | |
2230 | if (error == -1) | |
2231 | error = 0; | |
2232 | goto out_free; | |
2233 | } | |
2234 | ||
2235 | error = xfs_trans_read_buf_map(dp->i_mount, trans, | |
2236 | dp->i_mount->m_ddev_targp, | |
2237 | mapp, nmap, 0, &bp, ops); | |
2238 | if (error) | |
2239 | goto out_free; | |
2240 | ||
2241 | if (whichfork == XFS_ATTR_FORK) | |
2242 | xfs_buf_set_ref(bp, XFS_ATTR_BTREE_REF); | |
2bd0ea18 | 2243 | else |
a2ceac1f DC |
2244 | xfs_buf_set_ref(bp, XFS_DIR_BTREE_REF); |
2245 | ||
2bd0ea18 | 2246 | /* |
a2ceac1f DC |
2247 | * This verification code will be moved to a CRC verification callback |
2248 | * function so just leave it here unchanged until then. | |
2bd0ea18 | 2249 | */ |
a2ceac1f DC |
2250 | { |
2251 | xfs_dir2_data_hdr_t *hdr = bp->b_addr; | |
2252 | xfs_dir2_free_t *free = bp->b_addr; | |
2253 | xfs_da_blkinfo_t *info = bp->b_addr; | |
87718ce2 | 2254 | uint magic, magic1; |
a2ceac1f | 2255 | struct xfs_mount *mp = dp->i_mount; |
2bd0ea18 | 2256 | |
5e656dbb | 2257 | magic = be16_to_cpu(info->magic); |
a2ceac1f | 2258 | magic1 = be32_to_cpu(hdr->magic); |
4ca431fc NS |
2259 | if (unlikely( |
2260 | XFS_TEST_ERROR((magic != XFS_DA_NODE_MAGIC) && | |
87718ce2 NS |
2261 | (magic != XFS_ATTR_LEAF_MAGIC) && |
2262 | (magic != XFS_DIR2_LEAF1_MAGIC) && | |
2263 | (magic != XFS_DIR2_LEAFN_MAGIC) && | |
2264 | (magic1 != XFS_DIR2_BLOCK_MAGIC) && | |
2265 | (magic1 != XFS_DIR2_DATA_MAGIC) && | |
a2ceac1f | 2266 | (free->hdr.magic != cpu_to_be32(XFS_DIR2_FREE_MAGIC)), |
2bd0ea18 | 2267 | mp, XFS_ERRTAG_DA_READ_BUF, |
4ca431fc | 2268 | XFS_RANDOM_DA_READ_BUF))) { |
a2ceac1f | 2269 | trace_xfs_da_btree_corrupt(bp, _RET_IP_); |
4ca431fc NS |
2270 | XFS_CORRUPTION_ERROR("xfs_da_do_buf(2)", |
2271 | XFS_ERRLEVEL_LOW, mp, info); | |
2bd0ea18 | 2272 | error = XFS_ERROR(EFSCORRUPTED); |
a2ceac1f DC |
2273 | xfs_trans_brelse(trans, bp); |
2274 | goto out_free; | |
2bd0ea18 NS |
2275 | } |
2276 | } | |
a2ceac1f DC |
2277 | *bpp = bp; |
2278 | out_free: | |
2bd0ea18 | 2279 | if (mapp != &map) |
5e656dbb | 2280 | kmem_free(mapp); |
2bd0ea18 | 2281 | |
a2ceac1f | 2282 | return error; |
2bd0ea18 NS |
2283 | } |
2284 | ||
2285 | /* | |
5e656dbb | 2286 | * Readahead the dir/attr block. |
2bd0ea18 | 2287 | */ |
5e656dbb BN |
2288 | xfs_daddr_t |
2289 | xfs_da_reada_buf( | |
a2ceac1f DC |
2290 | struct xfs_trans *trans, |
2291 | struct xfs_inode *dp, | |
2292 | xfs_dablk_t bno, | |
2293 | xfs_daddr_t mappedbno, | |
2294 | int whichfork, | |
2295 | const struct xfs_buf_ops *ops) | |
2bd0ea18 | 2296 | { |
a2ceac1f DC |
2297 | struct xfs_buf_map map; |
2298 | struct xfs_buf_map *mapp; | |
2299 | int nmap; | |
2300 | int error; | |
2bd0ea18 | 2301 | |
a2ceac1f DC |
2302 | mapp = ↦ |
2303 | nmap = 1; | |
2304 | error = xfs_dabuf_map(trans, dp, bno, mappedbno, whichfork, | |
2305 | &mapp, &nmap); | |
2306 | if (error) { | |
2307 | /* mapping a hole is not an error, but we don't continue */ | |
2308 | if (error == -1) | |
2309 | error = 0; | |
2310 | goto out_free; | |
2311 | } | |
2312 | ||
2313 | mappedbno = mapp[0].bm_bn; | |
2314 | xfs_buf_readahead_map(dp->i_mount->m_ddev_targp, mapp, nmap, ops); | |
2315 | ||
2316 | out_free: | |
2317 | if (mapp != &map) | |
2318 | kmem_free(mapp); | |
2319 | ||
2320 | if (error) | |
5e656dbb | 2321 | return -1; |
a2ceac1f | 2322 | return mappedbno; |
2bd0ea18 NS |
2323 | } |
2324 | ||
5e656dbb | 2325 | kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */ |
2bd0ea18 NS |
2326 | |
2327 | /* | |
2328 | * Allocate a dir-state structure. | |
2329 | * We don't put them on the stack since they're large. | |
2330 | */ | |
2331 | xfs_da_state_t * | |
2332 | xfs_da_state_alloc(void) | |
2333 | { | |
56b2de80 | 2334 | return kmem_zone_zalloc(xfs_da_state_zone, KM_NOFS); |
2bd0ea18 NS |
2335 | } |
2336 | ||
2337 | /* | |
2338 | * Kill the altpath contents of a da-state structure. | |
2339 | */ | |
5e656dbb | 2340 | STATIC void |
2bd0ea18 NS |
2341 | xfs_da_state_kill_altpath(xfs_da_state_t *state) |
2342 | { | |
2343 | int i; | |
2344 | ||
a2ceac1f DC |
2345 | for (i = 0; i < state->altpath.active; i++) |
2346 | state->altpath.blk[i].bp = NULL; | |
2bd0ea18 NS |
2347 | state->altpath.active = 0; |
2348 | } | |
2349 | ||
2350 | /* | |
2351 | * Free a da-state structure. | |
2352 | */ | |
2353 | void | |
2354 | xfs_da_state_free(xfs_da_state_t *state) | |
2355 | { | |
2bd0ea18 | 2356 | xfs_da_state_kill_altpath(state); |
2bd0ea18 | 2357 | #ifdef DEBUG |
32181a02 | 2358 | memset((char *)state, 0, sizeof(*state)); |
2bd0ea18 NS |
2359 | #endif /* DEBUG */ |
2360 | kmem_zone_free(xfs_da_state_zone, state); | |
2361 | } |