?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Implemented inode prefetch
[pohmelfs.git] / fs / xfs / xfs_da_btree.c
blob77c74257c2a3b7e631f9ebfa6e25f6b1f3bcf46d
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
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
7 * published by the Free Software Foundation.
8 *
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.
13 *
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
17 */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_mount.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_dir2.h"
31 #include "xfs_dir2_format.h"
32 #include "xfs_dir2_priv.h"
33 #include "xfs_dinode.h"
34 #include "xfs_inode.h"
35 #include "xfs_inode_item.h"
36 #include "xfs_alloc.h"
37 #include "xfs_bmap.h"
38 #include "xfs_attr.h"
39 #include "xfs_attr_leaf.h"
40 #include "xfs_error.h"
41 #include "xfs_trace.h"
42
43 /*
44 * xfs_da_btree.c
45 *
46 * Routines to implement directories as Btrees of hashed names.
47 */
48
49 /*========================================================================
50 * Function prototypes for the kernel.
51 *========================================================================*/
52
53 /*
54 * Routines used for growing the Btree.
55 */
56 STATIC int xfs_da_root_split(xfs_da_state_t *state,
57 xfs_da_state_blk_t *existing_root,
58 xfs_da_state_blk_t *new_child);
59 STATIC int xfs_da_node_split(xfs_da_state_t *state,
60 xfs_da_state_blk_t *existing_blk,
61 xfs_da_state_blk_t *split_blk,
62 xfs_da_state_blk_t *blk_to_add,
63 int treelevel,
64 int *result);
65 STATIC void xfs_da_node_rebalance(xfs_da_state_t *state,
66 xfs_da_state_blk_t *node_blk_1,
67 xfs_da_state_blk_t *node_blk_2);
68 STATIC void xfs_da_node_add(xfs_da_state_t *state,
69 xfs_da_state_blk_t *old_node_blk,
70 xfs_da_state_blk_t *new_node_blk);
71
72 /*
73 * Routines used for shrinking the Btree.
74 */
75 STATIC int xfs_da_root_join(xfs_da_state_t *state,
76 xfs_da_state_blk_t *root_blk);
77 STATIC int xfs_da_node_toosmall(xfs_da_state_t *state, int *retval);
78 STATIC void xfs_da_node_remove(xfs_da_state_t *state,
79 xfs_da_state_blk_t *drop_blk);
80 STATIC void xfs_da_node_unbalance(xfs_da_state_t *state,
81 xfs_da_state_blk_t *src_node_blk,
82 xfs_da_state_blk_t *dst_node_blk);
83
84 /*
85 * Utility routines.
86 */
87 STATIC uint xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count);
88 STATIC int xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp);
89 STATIC xfs_dabuf_t *xfs_da_buf_make(int nbuf, xfs_buf_t **bps);
90 STATIC int xfs_da_blk_unlink(xfs_da_state_t *state,
91 xfs_da_state_blk_t *drop_blk,
92 xfs_da_state_blk_t *save_blk);
93 STATIC void xfs_da_state_kill_altpath(xfs_da_state_t *state);
94
95 /*========================================================================
96 * Routines used for growing the Btree.
97 *========================================================================*/
98
99 /*
100 * Create the initial contents of an intermediate node.
101 */
102 int
103 xfs_da_node_create(xfs_da_args_t *args, xfs_dablk_t blkno, int level,
104 xfs_dabuf_t **bpp, int whichfork)
105 {
106 xfs_da_intnode_t *node;
107 xfs_dabuf_t *bp;
108 int error;
109 xfs_trans_t *tp;
110
111 tp = args->trans;
112 error = xfs_da_get_buf(tp, args->dp, blkno, -1, &bp, whichfork);
113 if (error)
114 return(error);
115 ASSERT(bp != NULL);
116 node = bp->data;
117 node->hdr.info.forw = 0;
118 node->hdr.info.back = 0;
119 node->hdr.info.magic = cpu_to_be16(XFS_DA_NODE_MAGIC);
120 node->hdr.info.pad = 0;
121 node->hdr.count = 0;
122 node->hdr.level = cpu_to_be16(level);
123
124 xfs_da_log_buf(tp, bp,
125 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
126
127 *bpp = bp;
128 return(0);
129 }
130
131 /*
132 * Split a leaf node, rebalance, then possibly split
133 * intermediate nodes, rebalance, etc.
134 */
135 int /* error */
136 xfs_da_split(xfs_da_state_t *state)
137 {
138 xfs_da_state_blk_t *oldblk, *newblk, *addblk;
139 xfs_da_intnode_t *node;
140 xfs_dabuf_t *bp;
141 int max, action, error, i;
142
143 /*
144 * Walk back up the tree splitting/inserting/adjusting as necessary.
145 * If we need to insert and there isn't room, split the node, then
146 * decide which fragment to insert the new block from below into.
147 * Note that we may split the root this way, but we need more fixup.
148 */
149 max = state->path.active - 1;
150 ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH));
151 ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC ||
152 state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
153
154 addblk = &state->path.blk[max]; /* initial dummy value */
155 for (i = max; (i >= 0) && addblk; state->path.active--, i--) {
156 oldblk = &state->path.blk[i];
157 newblk = &state->altpath.blk[i];
158
159 /*
160 * If a leaf node then
161 * Allocate a new leaf node, then rebalance across them.
162 * else if an intermediate node then
163 * We split on the last layer, must we split the node?
164 */
165 switch (oldblk->magic) {
166 case XFS_ATTR_LEAF_MAGIC:
167 error = xfs_attr_leaf_split(state, oldblk, newblk);
168 if ((error != 0) && (error != ENOSPC)) {
169 return(error); /* GROT: attr is inconsistent */
170 }
171 if (!error) {
172 addblk = newblk;
173 break;
174 }
175 /*
176 * Entry wouldn't fit, split the leaf again.
177 */
178 state->extravalid = 1;
179 if (state->inleaf) {
180 state->extraafter = 0; /* before newblk */
181 error = xfs_attr_leaf_split(state, oldblk,
182 &state->extrablk);
183 } else {
184 state->extraafter = 1; /* after newblk */
185 error = xfs_attr_leaf_split(state, newblk,
186 &state->extrablk);
187 }
188 if (error)
189 return(error); /* GROT: attr inconsistent */
190 addblk = newblk;
191 break;
192 case XFS_DIR2_LEAFN_MAGIC:
193 error = xfs_dir2_leafn_split(state, oldblk, newblk);
194 if (error)
195 return error;
196 addblk = newblk;
197 break;
198 case XFS_DA_NODE_MAGIC:
199 error = xfs_da_node_split(state, oldblk, newblk, addblk,
200 max - i, &action);
201 xfs_da_buf_done(addblk->bp);
202 addblk->bp = NULL;
203 if (error)
204 return(error); /* GROT: dir is inconsistent */
205 /*
206 * Record the newly split block for the next time thru?
207 */
208 if (action)
209 addblk = newblk;
210 else
211 addblk = NULL;
212 break;
213 }
214
215 /*
216 * Update the btree to show the new hashval for this child.
217 */
218 xfs_da_fixhashpath(state, &state->path);
219 /*
220 * If we won't need this block again, it's getting dropped
221 * from the active path by the loop control, so we need
222 * to mark it done now.
223 */
224 if (i > 0 || !addblk)
225 xfs_da_buf_done(oldblk->bp);
226 }
227 if (!addblk)
228 return(0);
229
230 /*
231 * Split the root node.
232 */
233 ASSERT(state->path.active == 0);
234 oldblk = &state->path.blk[0];
235 error = xfs_da_root_split(state, oldblk, addblk);
236 if (error) {
237 xfs_da_buf_done(oldblk->bp);
238 xfs_da_buf_done(addblk->bp);
239 addblk->bp = NULL;
240 return(error); /* GROT: dir is inconsistent */
241 }
242
243 /*
244 * Update pointers to the node which used to be block 0 and
245 * just got bumped because of the addition of a new root node.
246 * There might be three blocks involved if a double split occurred,
247 * and the original block 0 could be at any position in the list.
248 */
249
250 node = oldblk->bp->data;
251 if (node->hdr.info.forw) {
252 if (be32_to_cpu(node->hdr.info.forw) == addblk->blkno) {
253 bp = addblk->bp;
254 } else {
255 ASSERT(state->extravalid);
256 bp = state->extrablk.bp;
257 }
258 node = bp->data;
259 node->hdr.info.back = cpu_to_be32(oldblk->blkno);
260 xfs_da_log_buf(state->args->trans, bp,
261 XFS_DA_LOGRANGE(node, &node->hdr.info,
262 sizeof(node->hdr.info)));
263 }
264 node = oldblk->bp->data;
265 if (node->hdr.info.back) {
266 if (be32_to_cpu(node->hdr.info.back) == addblk->blkno) {
267 bp = addblk->bp;
268 } else {
269 ASSERT(state->extravalid);
270 bp = state->extrablk.bp;
271 }
272 node = bp->data;
273 node->hdr.info.forw = cpu_to_be32(oldblk->blkno);
274 xfs_da_log_buf(state->args->trans, bp,
275 XFS_DA_LOGRANGE(node, &node->hdr.info,
276 sizeof(node->hdr.info)));
277 }
278 xfs_da_buf_done(oldblk->bp);
279 xfs_da_buf_done(addblk->bp);
280 addblk->bp = NULL;
281 return(0);
282 }
283
284 /*
285 * Split the root. We have to create a new root and point to the two
286 * parts (the split old root) that we just created. Copy block zero to
287 * the EOF, extending the inode in process.
288 */
289 STATIC int /* error */
290 xfs_da_root_split(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
291 xfs_da_state_blk_t *blk2)
292 {
293 xfs_da_intnode_t *node, *oldroot;
294 xfs_da_args_t *args;
295 xfs_dablk_t blkno;
296 xfs_dabuf_t *bp;
297 int error, size;
298 xfs_inode_t *dp;
299 xfs_trans_t *tp;
300 xfs_mount_t *mp;
301 xfs_dir2_leaf_t *leaf;
302
303 /*
304 * Copy the existing (incorrect) block from the root node position
305 * to a free space somewhere.
306 */
307 args = state->args;
308 ASSERT(args != NULL);
309 error = xfs_da_grow_inode(args, &blkno);
310 if (error)
311 return(error);
312 dp = args->dp;
313 tp = args->trans;
314 mp = state->mp;
315 error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork);
316 if (error)
317 return(error);
318 ASSERT(bp != NULL);
319 node = bp->data;
320 oldroot = blk1->bp->data;
321 if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)) {
322 size = (int)((char *)&oldroot->btree[be16_to_cpu(oldroot->hdr.count)] -
323 (char *)oldroot);
324 } else {
325 ASSERT(oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
326 leaf = (xfs_dir2_leaf_t *)oldroot;
327 size = (int)((char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] -
328 (char *)leaf);
329 }
330 memcpy(node, oldroot, size);
331 xfs_da_log_buf(tp, bp, 0, size - 1);
332 xfs_da_buf_done(blk1->bp);
333 blk1->bp = bp;
334 blk1->blkno = blkno;
335
336 /*
337 * Set up the new root node.
338 */
339 error = xfs_da_node_create(args,
340 (args->whichfork == XFS_DATA_FORK) ? mp->m_dirleafblk : 0,
341 be16_to_cpu(node->hdr.level) + 1, &bp, args->whichfork);
342 if (error)
343 return(error);
344 node = bp->data;
345 node->btree[0].hashval = cpu_to_be32(blk1->hashval);
346 node->btree[0].before = cpu_to_be32(blk1->blkno);
347 node->btree[1].hashval = cpu_to_be32(blk2->hashval);
348 node->btree[1].before = cpu_to_be32(blk2->blkno);
349 node->hdr.count = cpu_to_be16(2);
350
351 #ifdef DEBUG
352 if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC)) {
353 ASSERT(blk1->blkno >= mp->m_dirleafblk &&
354 blk1->blkno < mp->m_dirfreeblk);
355 ASSERT(blk2->blkno >= mp->m_dirleafblk &&
356 blk2->blkno < mp->m_dirfreeblk);
357 }
358 #endif
359
360 /* Header is already logged by xfs_da_node_create */
361 xfs_da_log_buf(tp, bp,
362 XFS_DA_LOGRANGE(node, node->btree,
363 sizeof(xfs_da_node_entry_t) * 2));
364 xfs_da_buf_done(bp);
365
366 return(0);
367 }
368
369 /*
370 * Split the node, rebalance, then add the new entry.
371 */
372 STATIC int /* error */
373 xfs_da_node_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
374 xfs_da_state_blk_t *newblk,
375 xfs_da_state_blk_t *addblk,
376 int treelevel, int *result)
377 {
378 xfs_da_intnode_t *node;
379 xfs_dablk_t blkno;
380 int newcount, error;
381 int useextra;
382
383 node = oldblk->bp->data;
384 ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
385
386 /*
387 * With V2 dirs the extra block is data or freespace.
388 */
389 useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK;
390 newcount = 1 + useextra;
391 /*
392 * Do we have to split the node?
393 */
394 if ((be16_to_cpu(node->hdr.count) + newcount) > state->node_ents) {
395 /*
396 * Allocate a new node, add to the doubly linked chain of
397 * nodes, then move some of our excess entries into it.
398 */
399 error = xfs_da_grow_inode(state->args, &blkno);
400 if (error)
401 return(error); /* GROT: dir is inconsistent */
402
403 error = xfs_da_node_create(state->args, blkno, treelevel,
404 &newblk->bp, state->args->whichfork);
405 if (error)
406 return(error); /* GROT: dir is inconsistent */
407 newblk->blkno = blkno;
408 newblk->magic = XFS_DA_NODE_MAGIC;
409 xfs_da_node_rebalance(state, oldblk, newblk);
410 error = xfs_da_blk_link(state, oldblk, newblk);
411 if (error)
412 return(error);
413 *result = 1;
414 } else {
415 *result = 0;
416 }
417
418 /*
419 * Insert the new entry(s) into the correct block
420 * (updating last hashval in the process).
421 *
422 * xfs_da_node_add() inserts BEFORE the given index,
423 * and as a result of using node_lookup_int() we always
424 * point to a valid entry (not after one), but a split
425 * operation always results in a new block whose hashvals
426 * FOLLOW the current block.
427 *
428 * If we had double-split op below us, then add the extra block too.
429 */
430 node = oldblk->bp->data;
431 if (oldblk->index <= be16_to_cpu(node->hdr.count)) {
432 oldblk->index++;
433 xfs_da_node_add(state, oldblk, addblk);
434 if (useextra) {
435 if (state->extraafter)
436 oldblk->index++;
437 xfs_da_node_add(state, oldblk, &state->extrablk);
438 state->extravalid = 0;
439 }
440 } else {
441 newblk->index++;
442 xfs_da_node_add(state, newblk, addblk);
443 if (useextra) {
444 if (state->extraafter)
445 newblk->index++;
446 xfs_da_node_add(state, newblk, &state->extrablk);
447 state->extravalid = 0;
448 }
449 }
450
451 return(0);
452 }
453
454 /*
455 * Balance the btree elements between two intermediate nodes,
456 * usually one full and one empty.
457 *
458 * NOTE: if blk2 is empty, then it will get the upper half of blk1.
459 */
460 STATIC void
461 xfs_da_node_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
462 xfs_da_state_blk_t *blk2)
463 {
464 xfs_da_intnode_t *node1, *node2, *tmpnode;
465 xfs_da_node_entry_t *btree_s, *btree_d;
466 int count, tmp;
467 xfs_trans_t *tp;
468
469 node1 = blk1->bp->data;
470 node2 = blk2->bp->data;
471 /*
472 * Figure out how many entries need to move, and in which direction.
473 * Swap the nodes around if that makes it simpler.
474 */
475 if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) &&
476 ((be32_to_cpu(node2->btree[0].hashval) < be32_to_cpu(node1->btree[0].hashval)) ||
477 (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) <
478 be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) {
479 tmpnode = node1;
480 node1 = node2;
481 node2 = tmpnode;
482 }
483 ASSERT(node1->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
484 ASSERT(node2->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
485 count = (be16_to_cpu(node1->hdr.count) - be16_to_cpu(node2->hdr.count)) / 2;
486 if (count == 0)
487 return;
488 tp = state->args->trans;
489 /*
490 * Two cases: high-to-low and low-to-high.
491 */
492 if (count > 0) {
493 /*
494 * Move elements in node2 up to make a hole.
495 */
496 if ((tmp = be16_to_cpu(node2->hdr.count)) > 0) {
497 tmp *= (uint)sizeof(xfs_da_node_entry_t);
498 btree_s = &node2->btree[0];
499 btree_d = &node2->btree[count];
500 memmove(btree_d, btree_s, tmp);
501 }
502
503 /*
504 * Move the req'd B-tree elements from high in node1 to
505 * low in node2.
506 */
507 be16_add_cpu(&node2->hdr.count, count);
508 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
509 btree_s = &node1->btree[be16_to_cpu(node1->hdr.count) - count];
510 btree_d = &node2->btree[0];
511 memcpy(btree_d, btree_s, tmp);
512 be16_add_cpu(&node1->hdr.count, -count);
513 } else {
514 /*
515 * Move the req'd B-tree elements from low in node2 to
516 * high in node1.
517 */
518 count = -count;
519 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
520 btree_s = &node2->btree[0];
521 btree_d = &node1->btree[be16_to_cpu(node1->hdr.count)];
522 memcpy(btree_d, btree_s, tmp);
523 be16_add_cpu(&node1->hdr.count, count);
524 xfs_da_log_buf(tp, blk1->bp,
525 XFS_DA_LOGRANGE(node1, btree_d, tmp));
526
527 /*
528 * Move elements in node2 down to fill the hole.
529 */
530 tmp = be16_to_cpu(node2->hdr.count) - count;
531 tmp *= (uint)sizeof(xfs_da_node_entry_t);
532 btree_s = &node2->btree[count];
533 btree_d = &node2->btree[0];
534 memmove(btree_d, btree_s, tmp);
535 be16_add_cpu(&node2->hdr.count, -count);
536 }
537
538 /*
539 * Log header of node 1 and all current bits of node 2.
540 */
541 xfs_da_log_buf(tp, blk1->bp,
542 XFS_DA_LOGRANGE(node1, &node1->hdr, sizeof(node1->hdr)));
543 xfs_da_log_buf(tp, blk2->bp,
544 XFS_DA_LOGRANGE(node2, &node2->hdr,
545 sizeof(node2->hdr) +
546 sizeof(node2->btree[0]) * be16_to_cpu(node2->hdr.count)));
547
548 /*
549 * Record the last hashval from each block for upward propagation.
550 * (note: don't use the swapped node pointers)
551 */
552 node1 = blk1->bp->data;
553 node2 = blk2->bp->data;
554 blk1->hashval = be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval);
555 blk2->hashval = be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval);
556
557 /*
558 * Adjust the expected index for insertion.
559 */
560 if (blk1->index >= be16_to_cpu(node1->hdr.count)) {
561 blk2->index = blk1->index - be16_to_cpu(node1->hdr.count);
562 blk1->index = be16_to_cpu(node1->hdr.count) + 1; /* make it invalid */
563 }
564 }
565
566 /*
567 * Add a new entry to an intermediate node.
568 */
569 STATIC void
570 xfs_da_node_add(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
571 xfs_da_state_blk_t *newblk)
572 {
573 xfs_da_intnode_t *node;
574 xfs_da_node_entry_t *btree;
575 int tmp;
576
577 node = oldblk->bp->data;
578 ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
579 ASSERT((oldblk->index >= 0) && (oldblk->index <= be16_to_cpu(node->hdr.count)));
580 ASSERT(newblk->blkno != 0);
581 if (state->args->whichfork == XFS_DATA_FORK)
582 ASSERT(newblk->blkno >= state->mp->m_dirleafblk &&
583 newblk->blkno < state->mp->m_dirfreeblk);
584
585 /*
586 * We may need to make some room before we insert the new node.
587 */
588 tmp = 0;
589 btree = &node->btree[ oldblk->index ];
590 if (oldblk->index < be16_to_cpu(node->hdr.count)) {
591 tmp = (be16_to_cpu(node->hdr.count) - oldblk->index) * (uint)sizeof(*btree);
592 memmove(btree + 1, btree, tmp);
593 }
594 btree->hashval = cpu_to_be32(newblk->hashval);
595 btree->before = cpu_to_be32(newblk->blkno);
596 xfs_da_log_buf(state->args->trans, oldblk->bp,
597 XFS_DA_LOGRANGE(node, btree, tmp + sizeof(*btree)));
598 be16_add_cpu(&node->hdr.count, 1);
599 xfs_da_log_buf(state->args->trans, oldblk->bp,
600 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
601
602 /*
603 * Copy the last hash value from the oldblk to propagate upwards.
604 */
605 oldblk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1 ].hashval);
606 }
607
608 /*========================================================================
609 * Routines used for shrinking the Btree.
610 *========================================================================*/
611
612 /*
613 * Deallocate an empty leaf node, remove it from its parent,
614 * possibly deallocating that block, etc...
615 */
616 int
617 xfs_da_join(xfs_da_state_t *state)
618 {
619 xfs_da_state_blk_t *drop_blk, *save_blk;
620 int action, error;
621
622 action = 0;
623 drop_blk = &state->path.blk[ state->path.active-1 ];
624 save_blk = &state->altpath.blk[ state->path.active-1 ];
625 ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
626 ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
627 drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
628
629 /*
630 * Walk back up the tree joining/deallocating as necessary.
631 * When we stop dropping blocks, break out.
632 */
633 for ( ; state->path.active >= 2; drop_blk--, save_blk--,
634 state->path.active--) {
635 /*
636 * See if we can combine the block with a neighbor.
637 * (action == 0) => no options, just leave
638 * (action == 1) => coalesce, then unlink
639 * (action == 2) => block empty, unlink it
640 */
641 switch (drop_blk->magic) {
642 case XFS_ATTR_LEAF_MAGIC:
643 error = xfs_attr_leaf_toosmall(state, &action);
644 if (error)
645 return(error);
646 if (action == 0)
647 return(0);
648 xfs_attr_leaf_unbalance(state, drop_blk, save_blk);
649 break;
650 case XFS_DIR2_LEAFN_MAGIC:
651 error = xfs_dir2_leafn_toosmall(state, &action);
652 if (error)
653 return error;
654 if (action == 0)
655 return 0;
656 xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
657 break;
658 case XFS_DA_NODE_MAGIC:
659 /*
660 * Remove the offending node, fixup hashvals,
661 * check for a toosmall neighbor.
662 */
663 xfs_da_node_remove(state, drop_blk);
664 xfs_da_fixhashpath(state, &state->path);
665 error = xfs_da_node_toosmall(state, &action);
666 if (error)
667 return(error);
668 if (action == 0)
669 return 0;
670 xfs_da_node_unbalance(state, drop_blk, save_blk);
671 break;
672 }
673 xfs_da_fixhashpath(state, &state->altpath);
674 error = xfs_da_blk_unlink(state, drop_blk, save_blk);
675 xfs_da_state_kill_altpath(state);
676 if (error)
677 return(error);
678 error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
679 drop_blk->bp);
680 drop_blk->bp = NULL;
681 if (error)
682 return(error);
683 }
684 /*
685 * We joined all the way to the top. If it turns out that
686 * we only have one entry in the root, make the child block
687 * the new root.
688 */
689 xfs_da_node_remove(state, drop_blk);
690 xfs_da_fixhashpath(state, &state->path);
691 error = xfs_da_root_join(state, &state->path.blk[0]);
692 return(error);
693 }
694
695 #ifdef DEBUG
696 static void
697 xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level)
698 {
699 __be16 magic = blkinfo->magic;
700
701 if (level == 1) {
702 ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
703 magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
704 } else
705 ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
706 ASSERT(!blkinfo->forw);
707 ASSERT(!blkinfo->back);
708 }
709 #else /* !DEBUG */
710 #define xfs_da_blkinfo_onlychild_validate(blkinfo, level)
711 #endif /* !DEBUG */
712
713 /*
714 * We have only one entry in the root. Copy the only remaining child of
715 * the old root to block 0 as the new root node.
716 */
717 STATIC int
718 xfs_da_root_join(xfs_da_state_t *state, xfs_da_state_blk_t *root_blk)
719 {
720 xfs_da_intnode_t *oldroot;
721 xfs_da_args_t *args;
722 xfs_dablk_t child;
723 xfs_dabuf_t *bp;
724 int error;
725
726 args = state->args;
727 ASSERT(args != NULL);
728 ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
729 oldroot = root_blk->bp->data;
730 ASSERT(oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
731 ASSERT(!oldroot->hdr.info.forw);
732 ASSERT(!oldroot->hdr.info.back);
733
734 /*
735 * If the root has more than one child, then don't do anything.
736 */
737 if (be16_to_cpu(oldroot->hdr.count) > 1)
738 return(0);
739
740 /*
741 * Read in the (only) child block, then copy those bytes into
742 * the root block's buffer and free the original child block.
743 */
744 child = be32_to_cpu(oldroot->btree[0].before);
745 ASSERT(child != 0);
746 error = xfs_da_read_buf(args->trans, args->dp, child, -1, &bp,
747 args->whichfork);
748 if (error)
749 return(error);
750 ASSERT(bp != NULL);
751 xfs_da_blkinfo_onlychild_validate(bp->data,
752 be16_to_cpu(oldroot->hdr.level));
753
754 memcpy(root_blk->bp->data, bp->data, state->blocksize);
755 xfs_da_log_buf(args->trans, root_blk->bp, 0, state->blocksize - 1);
756 error = xfs_da_shrink_inode(args, child, bp);
757 return(error);
758 }
759
760 /*
761 * Check a node block and its neighbors to see if the block should be
762 * collapsed into one or the other neighbor. Always keep the block
763 * with the smaller block number.
764 * If the current block is over 50% full, don't try to join it, return 0.
765 * If the block is empty, fill in the state structure and return 2.
766 * If it can be collapsed, fill in the state structure and return 1.
767 * If nothing can be done, return 0.
768 */
769 STATIC int
770 xfs_da_node_toosmall(xfs_da_state_t *state, int *action)
771 {
772 xfs_da_intnode_t *node;
773 xfs_da_state_blk_t *blk;
774 xfs_da_blkinfo_t *info;
775 int count, forward, error, retval, i;
776 xfs_dablk_t blkno;
777 xfs_dabuf_t *bp;
778
779 /*
780 * Check for the degenerate case of the block being over 50% full.
781 * If so, it's not worth even looking to see if we might be able
782 * to coalesce with a sibling.
783 */
784 blk = &state->path.blk[ state->path.active-1 ];
785 info = blk->bp->data;
786 ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
787 node = (xfs_da_intnode_t *)info;
788 count = be16_to_cpu(node->hdr.count);
789 if (count > (state->node_ents >> 1)) {
790 *action = 0; /* blk over 50%, don't try to join */
791 return(0); /* blk over 50%, don't try to join */
792 }
793
794 /*
795 * Check for the degenerate case of the block being empty.
796 * If the block is empty, we'll simply delete it, no need to
797 * coalesce it with a sibling block. We choose (arbitrarily)
798 * to merge with the forward block unless it is NULL.
799 */
800 if (count == 0) {
801 /*
802 * Make altpath point to the block we want to keep and
803 * path point to the block we want to drop (this one).
804 */
805 forward = (info->forw != 0);
806 memcpy(&state->altpath, &state->path, sizeof(state->path));
807 error = xfs_da_path_shift(state, &state->altpath, forward,
808 0, &retval);
809 if (error)
810 return(error);
811 if (retval) {
812 *action = 0;
813 } else {
814 *action = 2;
815 }
816 return(0);
817 }
818
819 /*
820 * Examine each sibling block to see if we can coalesce with
821 * at least 25% free space to spare. We need to figure out
822 * whether to merge with the forward or the backward block.
823 * We prefer coalescing with the lower numbered sibling so as
824 * to shrink a directory over time.
825 */
826 /* start with smaller blk num */
827 forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back));
828 for (i = 0; i < 2; forward = !forward, i++) {
829 if (forward)
830 blkno = be32_to_cpu(info->forw);
831 else
832 blkno = be32_to_cpu(info->back);
833 if (blkno == 0)
834 continue;
835 error = xfs_da_read_buf(state->args->trans, state->args->dp,
836 blkno, -1, &bp, state->args->whichfork);
837 if (error)
838 return(error);
839 ASSERT(bp != NULL);
840
841 node = (xfs_da_intnode_t *)info;
842 count = state->node_ents;
843 count -= state->node_ents >> 2;
844 count -= be16_to_cpu(node->hdr.count);
845 node = bp->data;
846 ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
847 count -= be16_to_cpu(node->hdr.count);
848 xfs_da_brelse(state->args->trans, bp);
849 if (count >= 0)
850 break; /* fits with at least 25% to spare */
851 }
852 if (i >= 2) {
853 *action = 0;
854 return(0);
855 }
856
857 /*
858 * Make altpath point to the block we want to keep (the lower
859 * numbered block) and path point to the block we want to drop.
860 */
861 memcpy(&state->altpath, &state->path, sizeof(state->path));
862 if (blkno < blk->blkno) {
863 error = xfs_da_path_shift(state, &state->altpath, forward,
864 0, &retval);
865 if (error) {
866 return(error);
867 }
868 if (retval) {
869 *action = 0;
870 return(0);
871 }
872 } else {
873 error = xfs_da_path_shift(state, &state->path, forward,
874 0, &retval);
875 if (error) {
876 return(error);
877 }
878 if (retval) {
879 *action = 0;
880 return(0);
881 }
882 }
883 *action = 1;
884 return(0);
885 }
886
887 /*
888 * Walk back up the tree adjusting hash values as necessary,
889 * when we stop making changes, return.
890 */
891 void
892 xfs_da_fixhashpath(xfs_da_state_t *state, xfs_da_state_path_t *path)
893 {
894 xfs_da_state_blk_t *blk;
895 xfs_da_intnode_t *node;
896 xfs_da_node_entry_t *btree;
897 xfs_dahash_t lasthash=0;
898 int level, count;
899
900 level = path->active-1;
901 blk = &path->blk[ level ];
902 switch (blk->magic) {
903 case XFS_ATTR_LEAF_MAGIC:
904 lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
905 if (count == 0)
906 return;
907 break;
908 case XFS_DIR2_LEAFN_MAGIC:
909 lasthash = xfs_dir2_leafn_lasthash(blk->bp, &count);
910 if (count == 0)
911 return;
912 break;
913 case XFS_DA_NODE_MAGIC:
914 lasthash = xfs_da_node_lasthash(blk->bp, &count);
915 if (count == 0)
916 return;
917 break;
918 }
919 for (blk--, level--; level >= 0; blk--, level--) {
920 node = blk->bp->data;
921 ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
922 btree = &node->btree[ blk->index ];
923 if (be32_to_cpu(btree->hashval) == lasthash)
924 break;
925 blk->hashval = lasthash;
926 btree->hashval = cpu_to_be32(lasthash);
927 xfs_da_log_buf(state->args->trans, blk->bp,
928 XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
929
930 lasthash = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
931 }
932 }
933
934 /*
935 * Remove an entry from an intermediate node.
936 */
937 STATIC void
938 xfs_da_node_remove(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk)
939 {
940 xfs_da_intnode_t *node;
941 xfs_da_node_entry_t *btree;
942 int tmp;
943
944 node = drop_blk->bp->data;
945 ASSERT(drop_blk->index < be16_to_cpu(node->hdr.count));
946 ASSERT(drop_blk->index >= 0);
947
948 /*
949 * Copy over the offending entry, or just zero it out.
950 */
951 btree = &node->btree[drop_blk->index];
952 if (drop_blk->index < (be16_to_cpu(node->hdr.count)-1)) {
953 tmp = be16_to_cpu(node->hdr.count) - drop_blk->index - 1;
954 tmp *= (uint)sizeof(xfs_da_node_entry_t);
955 memmove(btree, btree + 1, tmp);
956 xfs_da_log_buf(state->args->trans, drop_blk->bp,
957 XFS_DA_LOGRANGE(node, btree, tmp));
958 btree = &node->btree[be16_to_cpu(node->hdr.count)-1];
959 }
960 memset((char *)btree, 0, sizeof(xfs_da_node_entry_t));
961 xfs_da_log_buf(state->args->trans, drop_blk->bp,
962 XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
963 be16_add_cpu(&node->hdr.count, -1);
964 xfs_da_log_buf(state->args->trans, drop_blk->bp,
965 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
966
967 /*
968 * Copy the last hash value from the block to propagate upwards.
969 */
970 btree--;
971 drop_blk->hashval = be32_to_cpu(btree->hashval);
972 }
973
974 /*
975 * Unbalance the btree elements between two intermediate nodes,
976 * move all Btree elements from one node into another.
977 */
978 STATIC void
979 xfs_da_node_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
980 xfs_da_state_blk_t *save_blk)
981 {
982 xfs_da_intnode_t *drop_node, *save_node;
983 xfs_da_node_entry_t *btree;
984 int tmp;
985 xfs_trans_t *tp;
986
987 drop_node = drop_blk->bp->data;
988 save_node = save_blk->bp->data;
989 ASSERT(drop_node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
990 ASSERT(save_node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
991 tp = state->args->trans;
992
993 /*
994 * If the dying block has lower hashvals, then move all the
995 * elements in the remaining block up to make a hole.
996 */
997 if ((be32_to_cpu(drop_node->btree[0].hashval) < be32_to_cpu(save_node->btree[ 0 ].hashval)) ||
998 (be32_to_cpu(drop_node->btree[be16_to_cpu(drop_node->hdr.count)-1].hashval) <
999 be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval)))
1000 {
1001 btree = &save_node->btree[be16_to_cpu(drop_node->hdr.count)];
1002 tmp = be16_to_cpu(save_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t);
1003 memmove(btree, &save_node->btree[0], tmp);
1004 btree = &save_node->btree[0];
1005 xfs_da_log_buf(tp, save_blk->bp,
1006 XFS_DA_LOGRANGE(save_node, btree,
1007 (be16_to_cpu(save_node->hdr.count) + be16_to_cpu(drop_node->hdr.count)) *
1008 sizeof(xfs_da_node_entry_t)));
1009 } else {
1010 btree = &save_node->btree[be16_to_cpu(save_node->hdr.count)];
1011 xfs_da_log_buf(tp, save_blk->bp,
1012 XFS_DA_LOGRANGE(save_node, btree,
1013 be16_to_cpu(drop_node->hdr.count) *
1014 sizeof(xfs_da_node_entry_t)));
1015 }
1016
1017 /*
1018 * Move all the B-tree elements from drop_blk to save_blk.
1019 */
1020 tmp = be16_to_cpu(drop_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t);
1021 memcpy(btree, &drop_node->btree[0], tmp);
1022 be16_add_cpu(&save_node->hdr.count, be16_to_cpu(drop_node->hdr.count));
1023
1024 xfs_da_log_buf(tp, save_blk->bp,
1025 XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1026 sizeof(save_node->hdr)));
1027
1028 /*
1029 * Save the last hashval in the remaining block for upward propagation.
1030 */
1031 save_blk->hashval = be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval);
1032 }
1033
1034 /*========================================================================
1035 * Routines used for finding things in the Btree.
1036 *========================================================================*/
1037
1038 /*
1039 * Walk down the Btree looking for a particular filename, filling
1040 * in the state structure as we go.
1041 *
1042 * We will set the state structure to point to each of the elements
1043 * in each of the nodes where either the hashval is or should be.
1044 *
1045 * We support duplicate hashval's so for each entry in the current
1046 * node that could contain the desired hashval, descend. This is a
1047 * pruned depth-first tree search.
1048 */
1049 int /* error */
1050 xfs_da_node_lookup_int(xfs_da_state_t *state, int *result)
1051 {
1052 xfs_da_state_blk_t *blk;
1053 xfs_da_blkinfo_t *curr;
1054 xfs_da_intnode_t *node;
1055 xfs_da_node_entry_t *btree;
1056 xfs_dablk_t blkno;
1057 int probe, span, max, error, retval;
1058 xfs_dahash_t hashval, btreehashval;
1059 xfs_da_args_t *args;
1060
1061 args = state->args;
1062
1063 /*
1064 * Descend thru the B-tree searching each level for the right
1065 * node to use, until the right hashval is found.
1066 */
1067 blkno = (args->whichfork == XFS_DATA_FORK)? state->mp->m_dirleafblk : 0;
1068 for (blk = &state->path.blk[0], state->path.active = 1;
1069 state->path.active <= XFS_DA_NODE_MAXDEPTH;
1070 blk++, state->path.active++) {
1071 /*
1072 * Read the next node down in the tree.
1073 */
1074 blk->blkno = blkno;
1075 error = xfs_da_read_buf(args->trans, args->dp, blkno,
1076 -1, &blk->bp, args->whichfork);
1077 if (error) {
1078 blk->blkno = 0;
1079 state->path.active--;
1080 return(error);
1081 }
1082 curr = blk->bp->data;
1083 blk->magic = be16_to_cpu(curr->magic);
1084 ASSERT(blk->magic == XFS_DA_NODE_MAGIC ||
1085 blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1086 blk->magic == XFS_ATTR_LEAF_MAGIC);
1087
1088 /*
1089 * Search an intermediate node for a match.
1090 */
1091 if (blk->magic == XFS_DA_NODE_MAGIC) {
1092 node = blk->bp->data;
1093 max = be16_to_cpu(node->hdr.count);
1094 blk->hashval = be32_to_cpu(node->btree[max-1].hashval);
1095
1096 /*
1097 * Binary search. (note: small blocks will skip loop)
1098 */
1099 probe = span = max / 2;
1100 hashval = args->hashval;
1101 for (btree = &node->btree[probe]; span > 4;
1102 btree = &node->btree[probe]) {
1103 span /= 2;
1104 btreehashval = be32_to_cpu(btree->hashval);
1105 if (btreehashval < hashval)
1106 probe += span;
1107 else if (btreehashval > hashval)
1108 probe -= span;
1109 else
1110 break;
1111 }
1112 ASSERT((probe >= 0) && (probe < max));
1113 ASSERT((span <= 4) || (be32_to_cpu(btree->hashval) == hashval));
1114
1115 /*
1116 * Since we may have duplicate hashval's, find the first
1117 * matching hashval in the node.
1118 */
1119 while ((probe > 0) && (be32_to_cpu(btree->hashval) >= hashval)) {
1120 btree--;
1121 probe--;
1122 }
1123 while ((probe < max) && (be32_to_cpu(btree->hashval) < hashval)) {
1124 btree++;
1125 probe++;
1126 }
1127
1128 /*
1129 * Pick the right block to descend on.
1130 */
1131 if (probe == max) {
1132 blk->index = max-1;
1133 blkno = be32_to_cpu(node->btree[max-1].before);
1134 } else {
1135 blk->index = probe;
1136 blkno = be32_to_cpu(btree->before);
1137 }
1138 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1139 blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1140 break;
1141 } else if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1142 blk->hashval = xfs_dir2_leafn_lasthash(blk->bp, NULL);
1143 break;
1144 }
1145 }
1146
1147 /*
1148 * A leaf block that ends in the hashval that we are interested in
1149 * (final hashval == search hashval) means that the next block may
1150 * contain more entries with the same hashval, shift upward to the
1151 * next leaf and keep searching.
1152 */
1153 for (;;) {
1154 if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1155 retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
1156 &blk->index, state);
1157 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1158 retval = xfs_attr_leaf_lookup_int(blk->bp, args);
1159 blk->index = args->index;
1160 args->blkno = blk->blkno;
1161 } else {
1162 ASSERT(0);
1163 return XFS_ERROR(EFSCORRUPTED);
1164 }
1165 if (((retval == ENOENT) || (retval == ENOATTR)) &&
1166 (blk->hashval == args->hashval)) {
1167 error = xfs_da_path_shift(state, &state->path, 1, 1,
1168 &retval);
1169 if (error)
1170 return(error);
1171 if (retval == 0) {
1172 continue;
1173 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1174 /* path_shift() gives ENOENT */
1175 retval = XFS_ERROR(ENOATTR);
1176 }
1177 }
1178 break;
1179 }
1180 *result = retval;
1181 return(0);
1182 }
1183
1184 /*========================================================================
1185 * Utility routines.
1186 *========================================================================*/
1187
1188 /*
1189 * Link a new block into a doubly linked list of blocks (of whatever type).
1190 */
1191 int /* error */
1192 xfs_da_blk_link(xfs_da_state_t *state, xfs_da_state_blk_t *old_blk,
1193 xfs_da_state_blk_t *new_blk)
1194 {
1195 xfs_da_blkinfo_t *old_info, *new_info, *tmp_info;
1196 xfs_da_args_t *args;
1197 int before=0, error;
1198 xfs_dabuf_t *bp;
1199
1200 /*
1201 * Set up environment.
1202 */
1203 args = state->args;
1204 ASSERT(args != NULL);
1205 old_info = old_blk->bp->data;
1206 new_info = new_blk->bp->data;
1207 ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
1208 old_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1209 old_blk->magic == XFS_ATTR_LEAF_MAGIC);
1210 ASSERT(old_blk->magic == be16_to_cpu(old_info->magic));
1211 ASSERT(new_blk->magic == be16_to_cpu(new_info->magic));
1212 ASSERT(old_blk->magic == new_blk->magic);
1213
1214 switch (old_blk->magic) {
1215 case XFS_ATTR_LEAF_MAGIC:
1216 before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
1217 break;
1218 case XFS_DIR2_LEAFN_MAGIC:
1219 before = xfs_dir2_leafn_order(old_blk->bp, new_blk->bp);
1220 break;
1221 case XFS_DA_NODE_MAGIC:
1222 before = xfs_da_node_order(old_blk->bp, new_blk->bp);
1223 break;
1224 }
1225
1226 /*
1227 * Link blocks in appropriate order.
1228 */
1229 if (before) {
1230 /*
1231 * Link new block in before existing block.
1232 */
1233 new_info->forw = cpu_to_be32(old_blk->blkno);
1234 new_info->back = old_info->back;
1235 if (old_info->back) {
1236 error = xfs_da_read_buf(args->trans, args->dp,
1237 be32_to_cpu(old_info->back),
1238 -1, &bp, args->whichfork);
1239 if (error)
1240 return(error);
1241 ASSERT(bp != NULL);
1242 tmp_info = bp->data;
1243 ASSERT(be16_to_cpu(tmp_info->magic) == be16_to_cpu(old_info->magic));
1244 ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno);
1245 tmp_info->forw = cpu_to_be32(new_blk->blkno);
1246 xfs_da_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1247 xfs_da_buf_done(bp);
1248 }
1249 old_info->back = cpu_to_be32(new_blk->blkno);
1250 } else {
1251 /*
1252 * Link new block in after existing block.
1253 */
1254 new_info->forw = old_info->forw;
1255 new_info->back = cpu_to_be32(old_blk->blkno);
1256 if (old_info->forw) {
1257 error = xfs_da_read_buf(args->trans, args->dp,
1258 be32_to_cpu(old_info->forw),
1259 -1, &bp, args->whichfork);
1260 if (error)
1261 return(error);
1262 ASSERT(bp != NULL);
1263 tmp_info = bp->data;
1264 ASSERT(tmp_info->magic == old_info->magic);
1265 ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno);
1266 tmp_info->back = cpu_to_be32(new_blk->blkno);
1267 xfs_da_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1268 xfs_da_buf_done(bp);
1269 }
1270 old_info->forw = cpu_to_be32(new_blk->blkno);
1271 }
1272
1273 xfs_da_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
1274 xfs_da_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
1275 return(0);
1276 }
1277
1278 /*
1279 * Compare two intermediate nodes for "order".
1280 */
1281 STATIC int
1282 xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp)
1283 {
1284 xfs_da_intnode_t *node1, *node2;
1285
1286 node1 = node1_bp->data;
1287 node2 = node2_bp->data;
1288 ASSERT(node1->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) &&
1289 node2->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
1290 if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) &&
1291 ((be32_to_cpu(node2->btree[0].hashval) <
1292 be32_to_cpu(node1->btree[0].hashval)) ||
1293 (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) <
1294 be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) {
1295 return(1);
1296 }
1297 return(0);
1298 }
1299
1300 /*
1301 * Pick up the last hashvalue from an intermediate node.
1302 */
1303 STATIC uint
1304 xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count)
1305 {
1306 xfs_da_intnode_t *node;
1307
1308 node = bp->data;
1309 ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
1310 if (count)
1311 *count = be16_to_cpu(node->hdr.count);
1312 if (!node->hdr.count)
1313 return(0);
1314 return be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
1315 }
1316
1317 /*
1318 * Unlink a block from a doubly linked list of blocks.
1319 */
1320 STATIC int /* error */
1321 xfs_da_blk_unlink(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1322 xfs_da_state_blk_t *save_blk)
1323 {
1324 xfs_da_blkinfo_t *drop_info, *save_info, *tmp_info;
1325 xfs_da_args_t *args;
1326 xfs_dabuf_t *bp;
1327 int error;
1328
1329 /*
1330 * Set up environment.
1331 */
1332 args = state->args;
1333 ASSERT(args != NULL);
1334 save_info = save_blk->bp->data;
1335 drop_info = drop_blk->bp->data;
1336 ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
1337 save_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1338 save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1339 ASSERT(save_blk->magic == be16_to_cpu(save_info->magic));
1340 ASSERT(drop_blk->magic == be16_to_cpu(drop_info->magic));
1341 ASSERT(save_blk->magic == drop_blk->magic);
1342 ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) ||
1343 (be32_to_cpu(save_info->back) == drop_blk->blkno));
1344 ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) ||
1345 (be32_to_cpu(drop_info->back) == save_blk->blkno));
1346
1347 /*
1348 * Unlink the leaf block from the doubly linked chain of leaves.
1349 */
1350 if (be32_to_cpu(save_info->back) == drop_blk->blkno) {
1351 save_info->back = drop_info->back;
1352 if (drop_info->back) {
1353 error = xfs_da_read_buf(args->trans, args->dp,
1354 be32_to_cpu(drop_info->back),
1355 -1, &bp, args->whichfork);
1356 if (error)
1357 return(error);
1358 ASSERT(bp != NULL);
1359 tmp_info = bp->data;
1360 ASSERT(tmp_info->magic == save_info->magic);
1361 ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno);
1362 tmp_info->forw = cpu_to_be32(save_blk->blkno);
1363 xfs_da_log_buf(args->trans, bp, 0,
1364 sizeof(*tmp_info) - 1);
1365 xfs_da_buf_done(bp);
1366 }
1367 } else {
1368 save_info->forw = drop_info->forw;
1369 if (drop_info->forw) {
1370 error = xfs_da_read_buf(args->trans, args->dp,
1371 be32_to_cpu(drop_info->forw),
1372 -1, &bp, args->whichfork);
1373 if (error)
1374 return(error);
1375 ASSERT(bp != NULL);
1376 tmp_info = bp->data;
1377 ASSERT(tmp_info->magic == save_info->magic);
1378 ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno);
1379 tmp_info->back = cpu_to_be32(save_blk->blkno);
1380 xfs_da_log_buf(args->trans, bp, 0,
1381 sizeof(*tmp_info) - 1);
1382 xfs_da_buf_done(bp);
1383 }
1384 }
1385
1386 xfs_da_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
1387 return(0);
1388 }
1389
1390 /*
1391 * Move a path "forward" or "!forward" one block at the current level.
1392 *
1393 * This routine will adjust a "path" to point to the next block
1394 * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1395 * Btree, including updating pointers to the intermediate nodes between
1396 * the new bottom and the root.
1397 */
1398 int /* error */
1399 xfs_da_path_shift(xfs_da_state_t *state, xfs_da_state_path_t *path,
1400 int forward, int release, int *result)
1401 {
1402 xfs_da_state_blk_t *blk;
1403 xfs_da_blkinfo_t *info;
1404 xfs_da_intnode_t *node;
1405 xfs_da_args_t *args;
1406 xfs_dablk_t blkno=0;
1407 int level, error;
1408
1409 /*
1410 * Roll up the Btree looking for the first block where our
1411 * current index is not at the edge of the block. Note that
1412 * we skip the bottom layer because we want the sibling block.
1413 */
1414 args = state->args;
1415 ASSERT(args != NULL);
1416 ASSERT(path != NULL);
1417 ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1418 level = (path->active-1) - 1; /* skip bottom layer in path */
1419 for (blk = &path->blk[level]; level >= 0; blk--, level--) {
1420 ASSERT(blk->bp != NULL);
1421 node = blk->bp->data;
1422 ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
1423 if (forward && (blk->index < be16_to_cpu(node->hdr.count)-1)) {
1424 blk->index++;
1425 blkno = be32_to_cpu(node->btree[blk->index].before);
1426 break;
1427 } else if (!forward && (blk->index > 0)) {
1428 blk->index--;
1429 blkno = be32_to_cpu(node->btree[blk->index].before);
1430 break;
1431 }
1432 }
1433 if (level < 0) {
1434 *result = XFS_ERROR(ENOENT); /* we're out of our tree */
1435 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1436 return(0);
1437 }
1438
1439 /*
1440 * Roll down the edge of the subtree until we reach the
1441 * same depth we were at originally.
1442 */
1443 for (blk++, level++; level < path->active; blk++, level++) {
1444 /*
1445 * Release the old block.
1446 * (if it's dirty, trans won't actually let go)
1447 */
1448 if (release)
1449 xfs_da_brelse(args->trans, blk->bp);
1450
1451 /*
1452 * Read the next child block.
1453 */
1454 blk->blkno = blkno;
1455 error = xfs_da_read_buf(args->trans, args->dp, blkno, -1,
1456 &blk->bp, args->whichfork);
1457 if (error)
1458 return(error);
1459 ASSERT(blk->bp != NULL);
1460 info = blk->bp->data;
1461 ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1462 info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1463 info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1464 blk->magic = be16_to_cpu(info->magic);
1465 if (blk->magic == XFS_DA_NODE_MAGIC) {
1466 node = (xfs_da_intnode_t *)info;
1467 blk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
1468 if (forward)
1469 blk->index = 0;
1470 else
1471 blk->index = be16_to_cpu(node->hdr.count)-1;
1472 blkno = be32_to_cpu(node->btree[blk->index].before);
1473 } else {
1474 ASSERT(level == path->active-1);
1475 blk->index = 0;
1476 switch(blk->magic) {
1477 case XFS_ATTR_LEAF_MAGIC:
1478 blk->hashval = xfs_attr_leaf_lasthash(blk->bp,
1479 NULL);
1480 break;
1481 case XFS_DIR2_LEAFN_MAGIC:
1482 blk->hashval = xfs_dir2_leafn_lasthash(blk->bp,
1483 NULL);
1484 break;
1485 default:
1486 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC ||
1487 blk->magic == XFS_DIR2_LEAFN_MAGIC);
1488 break;
1489 }
1490 }
1491 }
1492 *result = 0;
1493 return(0);
1494 }
1495
1496
1497 /*========================================================================
1498 * Utility routines.
1499 *========================================================================*/
1500
1501 /*
1502 * Implement a simple hash on a character string.
1503 * Rotate the hash value by 7 bits, then XOR each character in.
1504 * This is implemented with some source-level loop unrolling.
1505 */
1506 xfs_dahash_t
1507 xfs_da_hashname(const __uint8_t *name, int namelen)
1508 {
1509 xfs_dahash_t hash;
1510
1511 /*
1512 * Do four characters at a time as long as we can.
1513 */
1514 for (hash = 0; namelen >= 4; namelen -= 4, name += 4)
1515 hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^
1516 (name[3] << 0) ^ rol32(hash, 7 * 4);
1517
1518 /*
1519 * Now do the rest of the characters.
1520 */
1521 switch (namelen) {
1522 case 3:
1523 return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^
1524 rol32(hash, 7 * 3);
1525 case 2:
1526 return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2);
1527 case 1:
1528 return (name[0] << 0) ^ rol32(hash, 7 * 1);
1529 default: /* case 0: */
1530 return hash;
1531 }
1532 }
1533
1534 enum xfs_dacmp
1535 xfs_da_compname(
1536 struct xfs_da_args *args,
1537 const unsigned char *name,
1538 int len)
1539 {
1540 return (args->namelen == len && memcmp(args->name, name, len) == 0) ?
1541 XFS_CMP_EXACT : XFS_CMP_DIFFERENT;
1542 }
1543
1544 static xfs_dahash_t
1545 xfs_default_hashname(
1546 struct xfs_name *name)
1547 {
1548 return xfs_da_hashname(name->name, name->len);
1549 }
1550
1551 const struct xfs_nameops xfs_default_nameops = {
1552 .hashname = xfs_default_hashname,
1553 .compname = xfs_da_compname
1554 };
1555
1556 int
1557 xfs_da_grow_inode_int(
1558 struct xfs_da_args *args,
1559 xfs_fileoff_t *bno,
1560 int count)
1561 {
1562 struct xfs_trans *tp = args->trans;
1563 struct xfs_inode *dp = args->dp;
1564 int w = args->whichfork;
1565 xfs_drfsbno_t nblks = dp->i_d.di_nblocks;
1566 struct xfs_bmbt_irec map, *mapp;
1567 int nmap, error, got, i, mapi;
1568
1569 /*
1570 * Find a spot in the file space to put the new block.
1571 */
1572 error = xfs_bmap_first_unused(tp, dp, count, bno, w);
1573 if (error)
1574 return error;
1575
1576 /*
1577 * Try mapping it in one filesystem block.
1578 */
1579 nmap = 1;
1580 ASSERT(args->firstblock != NULL);
1581 error = xfs_bmapi_write(tp, dp, *bno, count,
1582 xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG,
1583 args->firstblock, args->total, &map, &nmap,
1584 args->flist);
1585 if (error)
1586 return error;
1587
1588 ASSERT(nmap <= 1);
1589 if (nmap == 1) {
1590 mapp = &map;
1591 mapi = 1;
1592 } else if (nmap == 0 && count > 1) {
1593 xfs_fileoff_t b;
1594 int c;
1595
1596 /*
1597 * If we didn't get it and the block might work if fragmented,
1598 * try without the CONTIG flag. Loop until we get it all.
1599 */
1600 mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP);
1601 for (b = *bno, mapi = 0; b < *bno + count; ) {
1602 nmap = MIN(XFS_BMAP_MAX_NMAP, count);
1603 c = (int)(*bno + count - b);
1604 error = xfs_bmapi_write(tp, dp, b, c,
1605 xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
1606 args->firstblock, args->total,
1607 &mapp[mapi], &nmap, args->flist);
1608 if (error)
1609 goto out_free_map;
1610 if (nmap < 1)
1611 break;
1612 mapi += nmap;
1613 b = mapp[mapi - 1].br_startoff +
1614 mapp[mapi - 1].br_blockcount;
1615 }
1616 } else {
1617 mapi = 0;
1618 mapp = NULL;
1619 }
1620
1621 /*
1622 * Count the blocks we got, make sure it matches the total.
1623 */
1624 for (i = 0, got = 0; i < mapi; i++)
1625 got += mapp[i].br_blockcount;
1626 if (got != count || mapp[0].br_startoff != *bno ||
1627 mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
1628 *bno + count) {
1629 error = XFS_ERROR(ENOSPC);
1630 goto out_free_map;
1631 }
1632
1633 /* account for newly allocated blocks in reserved blocks total */
1634 args->total -= dp->i_d.di_nblocks - nblks;
1635
1636 out_free_map:
1637 if (mapp != &map)
1638 kmem_free(mapp);
1639 return error;
1640 }
1641
1642 /*
1643 * Add a block to the btree ahead of the file.
1644 * Return the new block number to the caller.
1645 */
1646 int
1647 xfs_da_grow_inode(
1648 struct xfs_da_args *args,
1649 xfs_dablk_t *new_blkno)
1650 {
1651 xfs_fileoff_t bno;
1652 int count;
1653 int error;
1654
1655 if (args->whichfork == XFS_DATA_FORK) {
1656 bno = args->dp->i_mount->m_dirleafblk;
1657 count = args->dp->i_mount->m_dirblkfsbs;
1658 } else {
1659 bno = 0;
1660 count = 1;
1661 }
1662
1663 error = xfs_da_grow_inode_int(args, &bno, count);
1664 if (!error)
1665 *new_blkno = (xfs_dablk_t)bno;
1666 return error;
1667 }
1668
1669 /*
1670 * Ick. We need to always be able to remove a btree block, even
1671 * if there's no space reservation because the filesystem is full.
1672 * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
1673 * It swaps the target block with the last block in the file. The
1674 * last block in the file can always be removed since it can't cause
1675 * a bmap btree split to do that.
1676 */
1677 STATIC int
1678 xfs_da_swap_lastblock(xfs_da_args_t *args, xfs_dablk_t *dead_blknop,
1679 xfs_dabuf_t **dead_bufp)
1680 {
1681 xfs_dablk_t dead_blkno, last_blkno, sib_blkno, par_blkno;
1682 xfs_dabuf_t *dead_buf, *last_buf, *sib_buf, *par_buf;
1683 xfs_fileoff_t lastoff;
1684 xfs_inode_t *ip;
1685 xfs_trans_t *tp;
1686 xfs_mount_t *mp;
1687 int error, w, entno, level, dead_level;
1688 xfs_da_blkinfo_t *dead_info, *sib_info;
1689 xfs_da_intnode_t *par_node, *dead_node;
1690 xfs_dir2_leaf_t *dead_leaf2;
1691 xfs_dahash_t dead_hash;
1692
1693 dead_buf = *dead_bufp;
1694 dead_blkno = *dead_blknop;
1695 tp = args->trans;
1696 ip = args->dp;
1697 w = args->whichfork;
1698 ASSERT(w == XFS_DATA_FORK);
1699 mp = ip->i_mount;
1700 lastoff = mp->m_dirfreeblk;
1701 error = xfs_bmap_last_before(tp, ip, &lastoff, w);
1702 if (error)
1703 return error;
1704 if (unlikely(lastoff == 0)) {
1705 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW,
1706 mp);
1707 return XFS_ERROR(EFSCORRUPTED);
1708 }
1709 /*
1710 * Read the last block in the btree space.
1711 */
1712 last_blkno = (xfs_dablk_t)lastoff - mp->m_dirblkfsbs;
1713 if ((error = xfs_da_read_buf(tp, ip, last_blkno, -1, &last_buf, w)))
1714 return error;
1715 /*
1716 * Copy the last block into the dead buffer and log it.
1717 */
1718 memcpy(dead_buf->data, last_buf->data, mp->m_dirblksize);
1719 xfs_da_log_buf(tp, dead_buf, 0, mp->m_dirblksize - 1);
1720 dead_info = dead_buf->data;
1721 /*
1722 * Get values from the moved block.
1723 */
1724 if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC)) {
1725 dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
1726 dead_level = 0;
1727 dead_hash = be32_to_cpu(dead_leaf2->ents[be16_to_cpu(dead_leaf2->hdr.count) - 1].hashval);
1728 } else {
1729 ASSERT(dead_info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
1730 dead_node = (xfs_da_intnode_t *)dead_info;
1731 dead_level = be16_to_cpu(dead_node->hdr.level);
1732 dead_hash = be32_to_cpu(dead_node->btree[be16_to_cpu(dead_node->hdr.count) - 1].hashval);
1733 }
1734 sib_buf = par_buf = NULL;
1735 /*
1736 * If the moved block has a left sibling, fix up the pointers.
1737 */
1738 if ((sib_blkno = be32_to_cpu(dead_info->back))) {
1739 if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w)))
1740 goto done;
1741 sib_info = sib_buf->data;
1742 if (unlikely(
1743 be32_to_cpu(sib_info->forw) != last_blkno ||
1744 sib_info->magic != dead_info->magic)) {
1745 XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
1746 XFS_ERRLEVEL_LOW, mp);
1747 error = XFS_ERROR(EFSCORRUPTED);
1748 goto done;
1749 }
1750 sib_info->forw = cpu_to_be32(dead_blkno);
1751 xfs_da_log_buf(tp, sib_buf,
1752 XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
1753 sizeof(sib_info->forw)));
1754 xfs_da_buf_done(sib_buf);
1755 sib_buf = NULL;
1756 }
1757 /*
1758 * If the moved block has a right sibling, fix up the pointers.
1759 */
1760 if ((sib_blkno = be32_to_cpu(dead_info->forw))) {
1761 if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w)))
1762 goto done;
1763 sib_info = sib_buf->data;
1764 if (unlikely(
1765 be32_to_cpu(sib_info->back) != last_blkno ||
1766 sib_info->magic != dead_info->magic)) {
1767 XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
1768 XFS_ERRLEVEL_LOW, mp);
1769 error = XFS_ERROR(EFSCORRUPTED);
1770 goto done;
1771 }
1772 sib_info->back = cpu_to_be32(dead_blkno);
1773 xfs_da_log_buf(tp, sib_buf,
1774 XFS_DA_LOGRANGE(sib_info, &sib_info->back,
1775 sizeof(sib_info->back)));
1776 xfs_da_buf_done(sib_buf);
1777 sib_buf = NULL;
1778 }
1779 par_blkno = mp->m_dirleafblk;
1780 level = -1;
1781 /*
1782 * Walk down the tree looking for the parent of the moved block.
1783 */
1784 for (;;) {
1785 if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w)))
1786 goto done;
1787 par_node = par_buf->data;
1788 if (unlikely(par_node->hdr.info.magic !=
1789 cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1790 (level >= 0 && level != be16_to_cpu(par_node->hdr.level) + 1))) {
1791 XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
1792 XFS_ERRLEVEL_LOW, mp);
1793 error = XFS_ERROR(EFSCORRUPTED);
1794 goto done;
1795 }
1796 level = be16_to_cpu(par_node->hdr.level);
1797 for (entno = 0;
1798 entno < be16_to_cpu(par_node->hdr.count) &&
1799 be32_to_cpu(par_node->btree[entno].hashval) < dead_hash;
1800 entno++)
1801 continue;
1802 if (unlikely(entno == be16_to_cpu(par_node->hdr.count))) {
1803 XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
1804 XFS_ERRLEVEL_LOW, mp);
1805 error = XFS_ERROR(EFSCORRUPTED);
1806 goto done;
1807 }
1808 par_blkno = be32_to_cpu(par_node->btree[entno].before);
1809 if (level == dead_level + 1)
1810 break;
1811 xfs_da_brelse(tp, par_buf);
1812 par_buf = NULL;
1813 }
1814 /*
1815 * We're in the right parent block.
1816 * Look for the right entry.
1817 */
1818 for (;;) {
1819 for (;
1820 entno < be16_to_cpu(par_node->hdr.count) &&
1821 be32_to_cpu(par_node->btree[entno].before) != last_blkno;
1822 entno++)
1823 continue;
1824 if (entno < be16_to_cpu(par_node->hdr.count))
1825 break;
1826 par_blkno = be32_to_cpu(par_node->hdr.info.forw);
1827 xfs_da_brelse(tp, par_buf);
1828 par_buf = NULL;
1829 if (unlikely(par_blkno == 0)) {
1830 XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
1831 XFS_ERRLEVEL_LOW, mp);
1832 error = XFS_ERROR(EFSCORRUPTED);
1833 goto done;
1834 }
1835 if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w)))
1836 goto done;
1837 par_node = par_buf->data;
1838 if (unlikely(
1839 be16_to_cpu(par_node->hdr.level) != level ||
1840 par_node->hdr.info.magic != cpu_to_be16(XFS_DA_NODE_MAGIC))) {
1841 XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
1842 XFS_ERRLEVEL_LOW, mp);
1843 error = XFS_ERROR(EFSCORRUPTED);
1844 goto done;
1845 }
1846 entno = 0;
1847 }
1848 /*
1849 * Update the parent entry pointing to the moved block.
1850 */
1851 par_node->btree[entno].before = cpu_to_be32(dead_blkno);
1852 xfs_da_log_buf(tp, par_buf,
1853 XFS_DA_LOGRANGE(par_node, &par_node->btree[entno].before,
1854 sizeof(par_node->btree[entno].before)));
1855 xfs_da_buf_done(par_buf);
1856 xfs_da_buf_done(dead_buf);
1857 *dead_blknop = last_blkno;
1858 *dead_bufp = last_buf;
1859 return 0;
1860 done:
1861 if (par_buf)
1862 xfs_da_brelse(tp, par_buf);
1863 if (sib_buf)
1864 xfs_da_brelse(tp, sib_buf);
1865 xfs_da_brelse(tp, last_buf);
1866 return error;
1867 }
1868
1869 /*
1870 * Remove a btree block from a directory or attribute.
1871 */
1872 int
1873 xfs_da_shrink_inode(xfs_da_args_t *args, xfs_dablk_t dead_blkno,
1874 xfs_dabuf_t *dead_buf)
1875 {
1876 xfs_inode_t *dp;
1877 int done, error, w, count;
1878 xfs_trans_t *tp;
1879 xfs_mount_t *mp;
1880
1881 dp = args->dp;
1882 w = args->whichfork;
1883 tp = args->trans;
1884 mp = dp->i_mount;
1885 if (w == XFS_DATA_FORK)
1886 count = mp->m_dirblkfsbs;
1887 else
1888 count = 1;
1889 for (;;) {
1890 /*
1891 * Remove extents. If we get ENOSPC for a dir we have to move
1892 * the last block to the place we want to kill.
1893 */
1894 if ((error = xfs_bunmapi(tp, dp, dead_blkno, count,
1895 xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
1896 0, args->firstblock, args->flist,
1897 &done)) == ENOSPC) {
1898 if (w != XFS_DATA_FORK)
1899 break;
1900 if ((error = xfs_da_swap_lastblock(args, &dead_blkno,
1901 &dead_buf)))
1902 break;
1903 } else {
1904 break;
1905 }
1906 }
1907 xfs_da_binval(tp, dead_buf);
1908 return error;
1909 }
1910
1911 /*
1912 * See if the mapping(s) for this btree block are valid, i.e.
1913 * don't contain holes, are logically contiguous, and cover the whole range.
1914 */
1915 STATIC int
1916 xfs_da_map_covers_blocks(
1917 int nmap,
1918 xfs_bmbt_irec_t *mapp,
1919 xfs_dablk_t bno,
1920 int count)
1921 {
1922 int i;
1923 xfs_fileoff_t off;
1924
1925 for (i = 0, off = bno; i < nmap; i++) {
1926 if (mapp[i].br_startblock == HOLESTARTBLOCK ||
1927 mapp[i].br_startblock == DELAYSTARTBLOCK) {
1928 return 0;
1929 }
1930 if (off != mapp[i].br_startoff) {
1931 return 0;
1932 }
1933 off += mapp[i].br_blockcount;
1934 }
1935 return off == bno + count;
1936 }
1937
1938 /*
1939 * Make a dabuf.
1940 * Used for get_buf, read_buf, read_bufr, and reada_buf.
1941 */
1942 STATIC int
1943 xfs_da_do_buf(
1944 xfs_trans_t *trans,
1945 xfs_inode_t *dp,
1946 xfs_dablk_t bno,
1947 xfs_daddr_t *mappedbnop,
1948 xfs_dabuf_t **bpp,
1949 int whichfork,
1950 int caller)
1951 {
1952 xfs_buf_t *bp = NULL;
1953 xfs_buf_t **bplist;
1954 int error=0;
1955 int i;
1956 xfs_bmbt_irec_t map;
1957 xfs_bmbt_irec_t *mapp;
1958 xfs_daddr_t mappedbno;
1959 xfs_mount_t *mp;
1960 int nbplist=0;
1961 int nfsb;
1962 int nmap;
1963 xfs_dabuf_t *rbp;
1964
1965 mp = dp->i_mount;
1966 nfsb = (whichfork == XFS_DATA_FORK) ? mp->m_dirblkfsbs : 1;
1967 mappedbno = *mappedbnop;
1968 /*
1969 * Caller doesn't have a mapping. -2 means don't complain
1970 * if we land in a hole.
1971 */
1972 if (mappedbno == -1 || mappedbno == -2) {
1973 /*
1974 * Optimize the one-block case.
1975 */
1976 if (nfsb == 1)
1977 mapp = &map;
1978 else
1979 mapp = kmem_alloc(sizeof(*mapp) * nfsb, KM_SLEEP);
1980
1981 nmap = nfsb;
1982 error = xfs_bmapi_read(dp, (xfs_fileoff_t)bno, nfsb, mapp,
1983 &nmap, xfs_bmapi_aflag(whichfork));
1984 if (error)
1985 goto exit0;
1986 } else {
1987 map.br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno);
1988 map.br_startoff = (xfs_fileoff_t)bno;
1989 map.br_blockcount = nfsb;
1990 mapp = &map;
1991 nmap = 1;
1992 }
1993 if (!xfs_da_map_covers_blocks(nmap, mapp, bno, nfsb)) {
1994 error = mappedbno == -2 ? 0 : XFS_ERROR(EFSCORRUPTED);
1995 if (unlikely(error == EFSCORRUPTED)) {
1996 if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
1997 xfs_alert(mp, "%s: bno %lld dir: inode %lld",
1998 __func__, (long long)bno,
1999 (long long)dp->i_ino);
2000 for (i = 0; i < nmap; i++) {
2001 xfs_alert(mp,
2002 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d",
2003 i,
2004 (long long)mapp[i].br_startoff,
2005 (long long)mapp[i].br_startblock,
2006 (long long)mapp[i].br_blockcount,
2007 mapp[i].br_state);
2008 }
2009 }
2010 XFS_ERROR_REPORT("xfs_da_do_buf(1)",
2011 XFS_ERRLEVEL_LOW, mp);
2012 }
2013 goto exit0;
2014 }
2015 if (caller != 3 && nmap > 1) {
2016 bplist = kmem_alloc(sizeof(*bplist) * nmap, KM_SLEEP);
2017 nbplist = 0;
2018 } else
2019 bplist = NULL;
2020 /*
2021 * Turn the mapping(s) into buffer(s).
2022 */
2023 for (i = 0; i < nmap; i++) {
2024 int nmapped;
2025
2026 mappedbno = XFS_FSB_TO_DADDR(mp, mapp[i].br_startblock);
2027 if (i == 0)
2028 *mappedbnop = mappedbno;
2029 nmapped = (int)XFS_FSB_TO_BB(mp, mapp[i].br_blockcount);
2030 switch (caller) {
2031 case 0:
2032 bp = xfs_trans_get_buf(trans, mp->m_ddev_targp,
2033 mappedbno, nmapped, 0);
2034 error = bp ? bp->b_error : XFS_ERROR(EIO);
2035 break;
2036 case 1:
2037 case 2:
2038 bp = NULL;
2039 error = xfs_trans_read_buf(mp, trans, mp->m_ddev_targp,
2040 mappedbno, nmapped, 0, &bp);
2041 break;
2042 case 3:
2043 xfs_buf_readahead(mp->m_ddev_targp, mappedbno, nmapped);
2044 error = 0;
2045 bp = NULL;
2046 break;
2047 }
2048 if (error) {
2049 if (bp)
2050 xfs_trans_brelse(trans, bp);
2051 goto exit1;
2052 }
2053 if (!bp)
2054 continue;
2055 if (caller == 1) {
2056 if (whichfork == XFS_ATTR_FORK)
2057 xfs_buf_set_ref(bp, XFS_ATTR_BTREE_REF);
2058 else
2059 xfs_buf_set_ref(bp, XFS_DIR_BTREE_REF);
2060 }
2061 if (bplist) {
2062 bplist[nbplist++] = bp;
2063 }
2064 }
2065 /*
2066 * Build a dabuf structure.
2067 */
2068 if (bplist) {
2069 rbp = xfs_da_buf_make(nbplist, bplist);
2070 } else if (bp)
2071 rbp = xfs_da_buf_make(1, &bp);
2072 else
2073 rbp = NULL;
2074 /*
2075 * For read_buf, check the magic number.
2076 */
2077 if (caller == 1) {
2078 xfs_dir2_data_hdr_t *hdr = rbp->data;
2079 xfs_dir2_free_t *free = rbp->data;
2080 xfs_da_blkinfo_t *info = rbp->data;
2081 uint magic, magic1;
2082
2083 magic = be16_to_cpu(info->magic);
2084 magic1 = be32_to_cpu(hdr->magic);
2085 if (unlikely(
2086 XFS_TEST_ERROR((magic != XFS_DA_NODE_MAGIC) &&
2087 (magic != XFS_ATTR_LEAF_MAGIC) &&
2088 (magic != XFS_DIR2_LEAF1_MAGIC) &&
2089 (magic != XFS_DIR2_LEAFN_MAGIC) &&
2090 (magic1 != XFS_DIR2_BLOCK_MAGIC) &&
2091 (magic1 != XFS_DIR2_DATA_MAGIC) &&
2092 (free->hdr.magic != cpu_to_be32(XFS_DIR2_FREE_MAGIC)),
2093 mp, XFS_ERRTAG_DA_READ_BUF,
2094 XFS_RANDOM_DA_READ_BUF))) {
2095 trace_xfs_da_btree_corrupt(rbp->bps[0], _RET_IP_);
2096 XFS_CORRUPTION_ERROR("xfs_da_do_buf(2)",
2097 XFS_ERRLEVEL_LOW, mp, info);
2098 error = XFS_ERROR(EFSCORRUPTED);
2099 xfs_da_brelse(trans, rbp);
2100 nbplist = 0;
2101 goto exit1;
2102 }
2103 }
2104 if (bplist) {
2105 kmem_free(bplist);
2106 }
2107 if (mapp != &map) {
2108 kmem_free(mapp);
2109 }
2110 if (bpp)
2111 *bpp = rbp;
2112 return 0;
2113 exit1:
2114 if (bplist) {
2115 for (i = 0; i < nbplist; i++)
2116 xfs_trans_brelse(trans, bplist[i]);
2117 kmem_free(bplist);
2118 }
2119 exit0:
2120 if (mapp != &map)
2121 kmem_free(mapp);
2122 if (bpp)
2123 *bpp = NULL;
2124 return error;
2125 }
2126
2127 /*
2128 * Get a buffer for the dir/attr block.
2129 */
2130 int
2131 xfs_da_get_buf(
2132 xfs_trans_t *trans,
2133 xfs_inode_t *dp,
2134 xfs_dablk_t bno,
2135 xfs_daddr_t mappedbno,
2136 xfs_dabuf_t **bpp,
2137 int whichfork)
2138 {
2139 return xfs_da_do_buf(trans, dp, bno, &mappedbno, bpp, whichfork, 0);
2140 }
2141
2142 /*
2143 * Get a buffer for the dir/attr block, fill in the contents.
2144 */
2145 int
2146 xfs_da_read_buf(
2147 xfs_trans_t *trans,
2148 xfs_inode_t *dp,
2149 xfs_dablk_t bno,
2150 xfs_daddr_t mappedbno,
2151 xfs_dabuf_t **bpp,
2152 int whichfork)
2153 {
2154 return xfs_da_do_buf(trans, dp, bno, &mappedbno, bpp, whichfork, 1);
2155 }
2156
2157 /*
2158 * Readahead the dir/attr block.
2159 */
2160 xfs_daddr_t
2161 xfs_da_reada_buf(
2162 xfs_trans_t *trans,
2163 xfs_inode_t *dp,
2164 xfs_dablk_t bno,
2165 int whichfork)
2166 {
2167 xfs_daddr_t rval;
2168
2169 rval = -1;
2170 if (xfs_da_do_buf(trans, dp, bno, &rval, NULL, whichfork, 3))
2171 return -1;
2172 else
2173 return rval;
2174 }
2175
2176 kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */
2177 kmem_zone_t *xfs_dabuf_zone; /* dabuf zone */
2178
2179 /*
2180 * Allocate a dir-state structure.
2181 * We don't put them on the stack since they're large.
2182 */
2183 xfs_da_state_t *
2184 xfs_da_state_alloc(void)
2185 {
2186 return kmem_zone_zalloc(xfs_da_state_zone, KM_NOFS);
2187 }
2188
2189 /*
2190 * Kill the altpath contents of a da-state structure.
2191 */
2192 STATIC void
2193 xfs_da_state_kill_altpath(xfs_da_state_t *state)
2194 {
2195 int i;
2196
2197 for (i = 0; i < state->altpath.active; i++) {
2198 if (state->altpath.blk[i].bp) {
2199 if (state->altpath.blk[i].bp != state->path.blk[i].bp)
2200 xfs_da_buf_done(state->altpath.blk[i].bp);
2201 state->altpath.blk[i].bp = NULL;
2202 }
2203 }
2204 state->altpath.active = 0;
2205 }
2206
2207 /*
2208 * Free a da-state structure.
2209 */
2210 void
2211 xfs_da_state_free(xfs_da_state_t *state)
2212 {
2213 int i;
2214
2215 xfs_da_state_kill_altpath(state);
2216 for (i = 0; i < state->path.active; i++) {
2217 if (state->path.blk[i].bp)
2218 xfs_da_buf_done(state->path.blk[i].bp);
2219 }
2220 if (state->extravalid && state->extrablk.bp)
2221 xfs_da_buf_done(state->extrablk.bp);
2222 #ifdef DEBUG
2223 memset((char *)state, 0, sizeof(*state));
2224 #endif /* DEBUG */
2225 kmem_zone_free(xfs_da_state_zone, state);
2226 }
2227
2228 /*
2229 * Create a dabuf.
2230 */
2231 /* ARGSUSED */
2232 STATIC xfs_dabuf_t *
2233 xfs_da_buf_make(int nbuf, xfs_buf_t **bps)
2234 {
2235 xfs_buf_t *bp;
2236 xfs_dabuf_t *dabuf;
2237 int i;
2238 int off;
2239
2240 if (nbuf == 1)
2241 dabuf = kmem_zone_alloc(xfs_dabuf_zone, KM_NOFS);
2242 else
2243 dabuf = kmem_alloc(XFS_DA_BUF_SIZE(nbuf), KM_NOFS);
2244 dabuf->dirty = 0;
2245 if (nbuf == 1) {
2246 dabuf->nbuf = 1;
2247 bp = bps[0];
2248 dabuf->bbcount = (short)BTOBB(XFS_BUF_COUNT(bp));
2249 dabuf->data = bp->b_addr;
2250 dabuf->bps[0] = bp;
2251 } else {
2252 dabuf->nbuf = nbuf;
2253 for (i = 0, dabuf->bbcount = 0; i < nbuf; i++) {
2254 dabuf->bps[i] = bp = bps[i];
2255 dabuf->bbcount += BTOBB(XFS_BUF_COUNT(bp));
2256 }
2257 dabuf->data = kmem_alloc(BBTOB(dabuf->bbcount), KM_SLEEP);
2258 for (i = off = 0; i < nbuf; i++, off += XFS_BUF_COUNT(bp)) {
2259 bp = bps[i];
2260 memcpy((char *)dabuf->data + off, bp->b_addr,
2261 XFS_BUF_COUNT(bp));
2262 }
2263 }
2264 return dabuf;
2265 }
2266
2267 /*
2268 * Un-dirty a dabuf.
2269 */
2270 STATIC void
2271 xfs_da_buf_clean(xfs_dabuf_t *dabuf)
2272 {
2273 xfs_buf_t *bp;
2274 int i;
2275 int off;
2276
2277 if (dabuf->dirty) {
2278 ASSERT(dabuf->nbuf > 1);
2279 dabuf->dirty = 0;
2280 for (i = off = 0; i < dabuf->nbuf;
2281 i++, off += XFS_BUF_COUNT(bp)) {
2282 bp = dabuf->bps[i];
2283 memcpy(bp->b_addr, dabuf->data + off,
2284 XFS_BUF_COUNT(bp));
2285 }
2286 }
2287 }
2288
2289 /*
2290 * Release a dabuf.
2291 */
2292 void
2293 xfs_da_buf_done(xfs_dabuf_t *dabuf)
2294 {
2295 ASSERT(dabuf);
2296 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2297 if (dabuf->dirty)
2298 xfs_da_buf_clean(dabuf);
2299 if (dabuf->nbuf > 1) {
2300 kmem_free(dabuf->data);
2301 kmem_free(dabuf);
2302 } else {
2303 kmem_zone_free(xfs_dabuf_zone, dabuf);
2304 }
2305 }
2306
2307 /*
2308 * Log transaction from a dabuf.
2309 */
2310 void
2311 xfs_da_log_buf(xfs_trans_t *tp, xfs_dabuf_t *dabuf, uint first, uint last)
2312 {
2313 xfs_buf_t *bp;
2314 uint f;
2315 int i;
2316 uint l;
2317 int off;
2318
2319 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2320 if (dabuf->nbuf == 1) {
2321 ASSERT(dabuf->data == dabuf->bps[0]->b_addr);
2322 xfs_trans_log_buf(tp, dabuf->bps[0], first, last);
2323 return;
2324 }
2325 dabuf->dirty = 1;
2326 ASSERT(first <= last);
2327 for (i = off = 0; i < dabuf->nbuf; i++, off += XFS_BUF_COUNT(bp)) {
2328 bp = dabuf->bps[i];
2329 f = off;
2330 l = f + XFS_BUF_COUNT(bp) - 1;
2331 if (f < first)
2332 f = first;
2333 if (l > last)
2334 l = last;
2335 if (f <= l)
2336 xfs_trans_log_buf(tp, bp, f - off, l - off);
2337 /*
2338 * B_DONE is set by xfs_trans_log buf.
2339 * If we don't set it on a new buffer (get not read)
2340 * then if we don't put anything in the buffer it won't
2341 * be set, and at commit it it released into the cache,
2342 * and then a read will fail.
2343 */
2344 else if (!(XFS_BUF_ISDONE(bp)))
2345 XFS_BUF_DONE(bp);
2346 }
2347 ASSERT(last < off);
2348 }
2349
2350 /*
2351 * Release dabuf from a transaction.
2352 * Have to free up the dabuf before the buffers are released,
2353 * since the synchronization on the dabuf is really the lock on the buffer.
2354 */
2355 void
2356 xfs_da_brelse(xfs_trans_t *tp, xfs_dabuf_t *dabuf)
2357 {
2358 xfs_buf_t *bp;
2359 xfs_buf_t **bplist;
2360 int i;
2361 int nbuf;
2362
2363 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2364 if ((nbuf = dabuf->nbuf) == 1) {
2365 bplist = &bp;
2366 bp = dabuf->bps[0];
2367 } else {
2368 bplist = kmem_alloc(nbuf * sizeof(*bplist), KM_SLEEP);
2369 memcpy(bplist, dabuf->bps, nbuf * sizeof(*bplist));
2370 }
2371 xfs_da_buf_done(dabuf);
2372 for (i = 0; i < nbuf; i++)
2373 xfs_trans_brelse(tp, bplist[i]);
2374 if (bplist != &bp)
2375 kmem_free(bplist);
2376 }
2377
2378 /*
2379 * Invalidate dabuf from a transaction.
2380 */
2381 void
2382 xfs_da_binval(xfs_trans_t *tp, xfs_dabuf_t *dabuf)
2383 {
2384 xfs_buf_t *bp;
2385 xfs_buf_t **bplist;
2386 int i;
2387 int nbuf;
2388
2389 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2390 if ((nbuf = dabuf->nbuf) == 1) {
2391 bplist = &bp;
2392 bp = dabuf->bps[0];
2393 } else {
2394 bplist = kmem_alloc(nbuf * sizeof(*bplist), KM_SLEEP);
2395 memcpy(bplist, dabuf->bps, nbuf * sizeof(*bplist));
2396 }
2397 xfs_da_buf_done(dabuf);
2398 for (i = 0; i < nbuf; i++)
2399 xfs_trans_binval(tp, bplist[i]);
2400 if (bplist != &bp)
2401 kmem_free(bplist);
2402 }
2403
2404 /*
2405 * Get the first daddr from a dabuf.
2406 */
2407 xfs_daddr_t
2408 xfs_da_blkno(xfs_dabuf_t *dabuf)
2409 {
2410 ASSERT(dabuf->nbuf);
2411 ASSERT(dabuf->data);
2412 return XFS_BUF_ADDR(dabuf->bps[0]);
2413 }