projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
xfs: remove unused delta tracking code in xfs_bmapi
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / xfs / xfs_da_btree.c
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_dir2.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_dir2_sf.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_alloc.h"
36 #include "xfs_bmap.h"
37 #include "xfs_attr.h"
38 #include "xfs_attr_leaf.h"
39 #include "xfs_dir2_data.h"
40 #include "xfs_dir2_leaf.h"
41 #include "xfs_dir2_block.h"
42 #include "xfs_dir2_node.h"
43 #include "xfs_error.h"
44 #include "xfs_trace.h"
45
46 /*
47 * xfs_da_btree.c
48 *
49 * Routines to implement directories as Btrees of hashed names.
50 */
51
52 /*========================================================================
53 * Function prototypes for the kernel.
54 *========================================================================*/
55
56 /*
57 * Routines used for growing the Btree.
58 */
59 STATIC int xfs_da_root_split(xfs_da_state_t *state,
60 xfs_da_state_blk_t *existing_root,
61 xfs_da_state_blk_t *new_child);
62 STATIC int xfs_da_node_split(xfs_da_state_t *state,
63 xfs_da_state_blk_t *existing_blk,
64 xfs_da_state_blk_t *split_blk,
65 xfs_da_state_blk_t *blk_to_add,
66 int treelevel,
67 int *result);
68 STATIC void xfs_da_node_rebalance(xfs_da_state_t *state,
69 xfs_da_state_blk_t *node_blk_1,
70 xfs_da_state_blk_t *node_blk_2);
71 STATIC void xfs_da_node_add(xfs_da_state_t *state,
72 xfs_da_state_blk_t *old_node_blk,
73 xfs_da_state_blk_t *new_node_blk);
74
75 /*
76 * Routines used for shrinking the Btree.
77 */
78 STATIC int xfs_da_root_join(xfs_da_state_t *state,
79 xfs_da_state_blk_t *root_blk);
80 STATIC int xfs_da_node_toosmall(xfs_da_state_t *state, int *retval);
81 STATIC void xfs_da_node_remove(xfs_da_state_t *state,
82 xfs_da_state_blk_t *drop_blk);
83 STATIC void xfs_da_node_unbalance(xfs_da_state_t *state,
84 xfs_da_state_blk_t *src_node_blk,
85 xfs_da_state_blk_t *dst_node_blk);
86
87 /*
88 * Utility routines.
89 */
90 STATIC uint xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count);
91 STATIC int xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp);
92 STATIC xfs_dabuf_t *xfs_da_buf_make(int nbuf, xfs_buf_t **bps, inst_t *ra);
93 STATIC int xfs_da_blk_unlink(xfs_da_state_t *state,
94 xfs_da_state_blk_t *drop_blk,
95 xfs_da_state_blk_t *save_blk);
96 STATIC void xfs_da_state_kill_altpath(xfs_da_state_t *state);
97
98 /*========================================================================
99 * Routines used for growing the Btree.
100 *========================================================================*/
101
102 /*
103 * Create the initial contents of an intermediate node.
104 */
105 int
106 xfs_da_node_create(xfs_da_args_t *args, xfs_dablk_t blkno, int level,
107 xfs_dabuf_t **bpp, int whichfork)
108 {
109 xfs_da_intnode_t *node;
110 xfs_dabuf_t *bp;
111 int error;
112 xfs_trans_t *tp;
113
114 tp = args->trans;
115 error = xfs_da_get_buf(tp, args->dp, blkno, -1, &bp, whichfork);
116 if (error)
117 return(error);
118 ASSERT(bp != NULL);
119 node = bp->data;
120 node->hdr.info.forw = 0;
121 node->hdr.info.back = 0;
122 node->hdr.info.magic = cpu_to_be16(XFS_DA_NODE_MAGIC);
123 node->hdr.info.pad = 0;
124 node->hdr.count = 0;
125 node->hdr.level = cpu_to_be16(level);
126
127 xfs_da_log_buf(tp, bp,
128 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
129
130 *bpp = bp;
131 return(0);
132 }
133
134 /*
135 * Split a leaf node, rebalance, then possibly split
136 * intermediate nodes, rebalance, etc.
137 */
138 int /* error */
139 xfs_da_split(xfs_da_state_t *state)
140 {
141 xfs_da_state_blk_t *oldblk, *newblk, *addblk;
142 xfs_da_intnode_t *node;
143 xfs_dabuf_t *bp;
144 int max, action, error, i;
145
146 /*
147 * Walk back up the tree splitting/inserting/adjusting as necessary.
148 * If we need to insert and there isn't room, split the node, then
149 * decide which fragment to insert the new block from below into.
150 * Note that we may split the root this way, but we need more fixup.
151 */
152 max = state->path.active - 1;
153 ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH));
154 ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC ||
155 state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
156
157 addblk = &state->path.blk[max]; /* initial dummy value */
158 for (i = max; (i >= 0) && addblk; state->path.active--, i--) {
159 oldblk = &state->path.blk[i];
160 newblk = &state->altpath.blk[i];
161
162 /*
163 * If a leaf node then
164 * Allocate a new leaf node, then rebalance across them.
165 * else if an intermediate node then
166 * We split on the last layer, must we split the node?
167 */
168 switch (oldblk->magic) {
169 case XFS_ATTR_LEAF_MAGIC:
170 error = xfs_attr_leaf_split(state, oldblk, newblk);
171 if ((error != 0) && (error != ENOSPC)) {
172 return(error); /* GROT: attr is inconsistent */
173 }
174 if (!error) {
175 addblk = newblk;
176 break;
177 }
178 /*
179 * Entry wouldn't fit, split the leaf again.
180 */
181 state->extravalid = 1;
182 if (state->inleaf) {
183 state->extraafter = 0; /* before newblk */
184 error = xfs_attr_leaf_split(state, oldblk,
185 &state->extrablk);
186 } else {
187 state->extraafter = 1; /* after newblk */
188 error = xfs_attr_leaf_split(state, newblk,
189 &state->extrablk);
190 }
191 if (error)
192 return(error); /* GROT: attr inconsistent */
193 addblk = newblk;
194 break;
195 case XFS_DIR2_LEAFN_MAGIC:
196 error = xfs_dir2_leafn_split(state, oldblk, newblk);
197 if (error)
198 return error;
199 addblk = newblk;
200 break;
201 case XFS_DA_NODE_MAGIC:
202 error = xfs_da_node_split(state, oldblk, newblk, addblk,
203 max - i, &action);
204 xfs_da_buf_done(addblk->bp);
205 addblk->bp = NULL;
206 if (error)
207 return(error); /* GROT: dir is inconsistent */
208 /*
209 * Record the newly split block for the next time thru?
210 */
211 if (action)
212 addblk = newblk;
213 else
214 addblk = NULL;
215 break;
216 }
217
218 /*
219 * Update the btree to show the new hashval for this child.
220 */
221 xfs_da_fixhashpath(state, &state->path);
222 /*
223 * If we won't need this block again, it's getting dropped
224 * from the active path by the loop control, so we need
225 * to mark it done now.
226 */
227 if (i > 0 || !addblk)
228 xfs_da_buf_done(oldblk->bp);
229 }
230 if (!addblk)
231 return(0);
232
233 /*
234 * Split the root node.
235 */
236 ASSERT(state->path.active == 0);
237 oldblk = &state->path.blk[0];
238 error = xfs_da_root_split(state, oldblk, addblk);
239 if (error) {
240 xfs_da_buf_done(oldblk->bp);
241 xfs_da_buf_done(addblk->bp);
242 addblk->bp = NULL;
243 return(error); /* GROT: dir is inconsistent */
244 }
245
246 /*
247 * Update pointers to the node which used to be block 0 and
248 * just got bumped because of the addition of a new root node.
249 * There might be three blocks involved if a double split occurred,
250 * and the original block 0 could be at any position in the list.
251 */
252
253 node = oldblk->bp->data;
254 if (node->hdr.info.forw) {
255 if (be32_to_cpu(node->hdr.info.forw) == addblk->blkno) {
256 bp = addblk->bp;
257 } else {
258 ASSERT(state->extravalid);
259 bp = state->extrablk.bp;
260 }
261 node = bp->data;
262 node->hdr.info.back = cpu_to_be32(oldblk->blkno);
263 xfs_da_log_buf(state->args->trans, bp,
264 XFS_DA_LOGRANGE(node, &node->hdr.info,
265 sizeof(node->hdr.info)));
266 }
267 node = oldblk->bp->data;
268 if (node->hdr.info.back) {
269 if (be32_to_cpu(node->hdr.info.back) == addblk->blkno) {
270 bp = addblk->bp;
271 } else {
272 ASSERT(state->extravalid);
273 bp = state->extrablk.bp;
274 }
275 node = bp->data;
276 node->hdr.info.forw = cpu_to_be32(oldblk->blkno);
277 xfs_da_log_buf(state->args->trans, bp,
278 XFS_DA_LOGRANGE(node, &node->hdr.info,
279 sizeof(node->hdr.info)));
280 }
281 xfs_da_buf_done(oldblk->bp);
282 xfs_da_buf_done(addblk->bp);
283 addblk->bp = NULL;
284 return(0);
285 }
286
287 /*
288 * Split the root. We have to create a new root and point to the two
289 * parts (the split old root) that we just created. Copy block zero to
290 * the EOF, extending the inode in process.
291 */
292 STATIC int /* error */
293 xfs_da_root_split(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
294 xfs_da_state_blk_t *blk2)
295 {
296 xfs_da_intnode_t *node, *oldroot;
297 xfs_da_args_t *args;
298 xfs_dablk_t blkno;
299 xfs_dabuf_t *bp;
300 int error, size;
301 xfs_inode_t *dp;
302 xfs_trans_t *tp;
303 xfs_mount_t *mp;
304 xfs_dir2_leaf_t *leaf;
305
306 /*
307 * Copy the existing (incorrect) block from the root node position
308 * to a free space somewhere.
309 */
310 args = state->args;
311 ASSERT(args != NULL);
312 error = xfs_da_grow_inode(args, &blkno);
313 if (error)
314 return(error);
315 dp = args->dp;
316 tp = args->trans;
317 mp = state->mp;
318 error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork);
319 if (error)
320 return(error);
321 ASSERT(bp != NULL);
322 node = bp->data;
323 oldroot = blk1->bp->data;
324 if (be16_to_cpu(oldroot->hdr.info.magic) == XFS_DA_NODE_MAGIC) {
325 size = (int)((char *)&oldroot->btree[be16_to_cpu(oldroot->hdr.count)] -
326 (char *)oldroot);
327 } else {
328 ASSERT(be16_to_cpu(oldroot->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
329 leaf = (xfs_dir2_leaf_t *)oldroot;
330 size = (int)((char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] -
331 (char *)leaf);
332 }
333 memcpy(node, oldroot, size);
334 xfs_da_log_buf(tp, bp, 0, size - 1);
335 xfs_da_buf_done(blk1->bp);
336 blk1->bp = bp;
337 blk1->blkno = blkno;
338
339 /*
340 * Set up the new root node.
341 */
342 error = xfs_da_node_create(args,
343 (args->whichfork == XFS_DATA_FORK) ? mp->m_dirleafblk : 0,
344 be16_to_cpu(node->hdr.level) + 1, &bp, args->whichfork);
345 if (error)
346 return(error);
347 node = bp->data;
348 node->btree[0].hashval = cpu_to_be32(blk1->hashval);
349 node->btree[0].before = cpu_to_be32(blk1->blkno);
350 node->btree[1].hashval = cpu_to_be32(blk2->hashval);
351 node->btree[1].before = cpu_to_be32(blk2->blkno);
352 node->hdr.count = cpu_to_be16(2);
353
354 #ifdef DEBUG
355 if (be16_to_cpu(oldroot->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC) {
356 ASSERT(blk1->blkno >= mp->m_dirleafblk &&
357 blk1->blkno < mp->m_dirfreeblk);
358 ASSERT(blk2->blkno >= mp->m_dirleafblk &&
359 blk2->blkno < mp->m_dirfreeblk);
360 }
361 #endif
362
363 /* Header is already logged by xfs_da_node_create */
364 xfs_da_log_buf(tp, bp,
365 XFS_DA_LOGRANGE(node, node->btree,
366 sizeof(xfs_da_node_entry_t) * 2));
367 xfs_da_buf_done(bp);
368
369 return(0);
370 }
371
372 /*
373 * Split the node, rebalance, then add the new entry.
374 */
375 STATIC int /* error */
376 xfs_da_node_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
377 xfs_da_state_blk_t *newblk,
378 xfs_da_state_blk_t *addblk,
379 int treelevel, int *result)
380 {
381 xfs_da_intnode_t *node;
382 xfs_dablk_t blkno;
383 int newcount, error;
384 int useextra;
385
386 node = oldblk->bp->data;
387 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
388
389 /*
390 * With V2 dirs the extra block is data or freespace.
391 */
392 useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK;
393 newcount = 1 + useextra;
394 /*
395 * Do we have to split the node?
396 */
397 if ((be16_to_cpu(node->hdr.count) + newcount) > state->node_ents) {
398 /*
399 * Allocate a new node, add to the doubly linked chain of
400 * nodes, then move some of our excess entries into it.
401 */
402 error = xfs_da_grow_inode(state->args, &blkno);
403 if (error)
404 return(error); /* GROT: dir is inconsistent */
405
406 error = xfs_da_node_create(state->args, blkno, treelevel,
407 &newblk->bp, state->args->whichfork);
408 if (error)
409 return(error); /* GROT: dir is inconsistent */
410 newblk->blkno = blkno;
411 newblk->magic = XFS_DA_NODE_MAGIC;
412 xfs_da_node_rebalance(state, oldblk, newblk);
413 error = xfs_da_blk_link(state, oldblk, newblk);
414 if (error)
415 return(error);
416 *result = 1;
417 } else {
418 *result = 0;
419 }
420
421 /*
422 * Insert the new entry(s) into the correct block
423 * (updating last hashval in the process).
424 *
425 * xfs_da_node_add() inserts BEFORE the given index,
426 * and as a result of using node_lookup_int() we always
427 * point to a valid entry (not after one), but a split
428 * operation always results in a new block whose hashvals
429 * FOLLOW the current block.
430 *
431 * If we had double-split op below us, then add the extra block too.
432 */
433 node = oldblk->bp->data;
434 if (oldblk->index <= be16_to_cpu(node->hdr.count)) {
435 oldblk->index++;
436 xfs_da_node_add(state, oldblk, addblk);
437 if (useextra) {
438 if (state->extraafter)
439 oldblk->index++;
440 xfs_da_node_add(state, oldblk, &state->extrablk);
441 state->extravalid = 0;
442 }
443 } else {
444 newblk->index++;
445 xfs_da_node_add(state, newblk, addblk);
446 if (useextra) {
447 if (state->extraafter)
448 newblk->index++;
449 xfs_da_node_add(state, newblk, &state->extrablk);
450 state->extravalid = 0;
451 }
452 }
453
454 return(0);
455 }
456
457 /*
458 * Balance the btree elements between two intermediate nodes,
459 * usually one full and one empty.
460 *
461 * NOTE: if blk2 is empty, then it will get the upper half of blk1.
462 */
463 STATIC void
464 xfs_da_node_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
465 xfs_da_state_blk_t *blk2)
466 {
467 xfs_da_intnode_t *node1, *node2, *tmpnode;
468 xfs_da_node_entry_t *btree_s, *btree_d;
469 int count, tmp;
470 xfs_trans_t *tp;
471
472 node1 = blk1->bp->data;
473 node2 = blk2->bp->data;
474 /*
475 * Figure out how many entries need to move, and in which direction.
476 * Swap the nodes around if that makes it simpler.
477 */
478 if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) &&
479 ((be32_to_cpu(node2->btree[0].hashval) < be32_to_cpu(node1->btree[0].hashval)) ||
480 (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) <
481 be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) {
482 tmpnode = node1;
483 node1 = node2;
484 node2 = tmpnode;
485 }
486 ASSERT(be16_to_cpu(node1->hdr.info.magic) == XFS_DA_NODE_MAGIC);
487 ASSERT(be16_to_cpu(node2->hdr.info.magic) == XFS_DA_NODE_MAGIC);
488 count = (be16_to_cpu(node1->hdr.count) - be16_to_cpu(node2->hdr.count)) / 2;
489 if (count == 0)
490 return;
491 tp = state->args->trans;
492 /*
493 * Two cases: high-to-low and low-to-high.
494 */
495 if (count > 0) {
496 /*
497 * Move elements in node2 up to make a hole.
498 */
499 if ((tmp = be16_to_cpu(node2->hdr.count)) > 0) {
500 tmp *= (uint)sizeof(xfs_da_node_entry_t);
501 btree_s = &node2->btree[0];
502 btree_d = &node2->btree[count];
503 memmove(btree_d, btree_s, tmp);
504 }
505
506 /*
507 * Move the req'd B-tree elements from high in node1 to
508 * low in node2.
509 */
510 be16_add_cpu(&node2->hdr.count, count);
511 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
512 btree_s = &node1->btree[be16_to_cpu(node1->hdr.count) - count];
513 btree_d = &node2->btree[0];
514 memcpy(btree_d, btree_s, tmp);
515 be16_add_cpu(&node1->hdr.count, -count);
516 } else {
517 /*
518 * Move the req'd B-tree elements from low in node2 to
519 * high in node1.
520 */
521 count = -count;
522 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
523 btree_s = &node2->btree[0];
524 btree_d = &node1->btree[be16_to_cpu(node1->hdr.count)];
525 memcpy(btree_d, btree_s, tmp);
526 be16_add_cpu(&node1->hdr.count, count);
527 xfs_da_log_buf(tp, blk1->bp,
528 XFS_DA_LOGRANGE(node1, btree_d, tmp));
529
530 /*
531 * Move elements in node2 down to fill the hole.
532 */
533 tmp = be16_to_cpu(node2->hdr.count) - count;
534 tmp *= (uint)sizeof(xfs_da_node_entry_t);
535 btree_s = &node2->btree[count];
536 btree_d = &node2->btree[0];
537 memmove(btree_d, btree_s, tmp);
538 be16_add_cpu(&node2->hdr.count, -count);
539 }
540
541 /*
542 * Log header of node 1 and all current bits of node 2.
543 */
544 xfs_da_log_buf(tp, blk1->bp,
545 XFS_DA_LOGRANGE(node1, &node1->hdr, sizeof(node1->hdr)));
546 xfs_da_log_buf(tp, blk2->bp,
547 XFS_DA_LOGRANGE(node2, &node2->hdr,
548 sizeof(node2->hdr) +
549 sizeof(node2->btree[0]) * be16_to_cpu(node2->hdr.count)));
550
551 /*
552 * Record the last hashval from each block for upward propagation.
553 * (note: don't use the swapped node pointers)
554 */
555 node1 = blk1->bp->data;
556 node2 = blk2->bp->data;
557 blk1->hashval = be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval);
558 blk2->hashval = be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval);
559
560 /*
561 * Adjust the expected index for insertion.
562 */
563 if (blk1->index >= be16_to_cpu(node1->hdr.count)) {
564 blk2->index = blk1->index - be16_to_cpu(node1->hdr.count);
565 blk1->index = be16_to_cpu(node1->hdr.count) + 1; /* make it invalid */
566 }
567 }
568
569 /*
570 * Add a new entry to an intermediate node.
571 */
572 STATIC void
573 xfs_da_node_add(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
574 xfs_da_state_blk_t *newblk)
575 {
576 xfs_da_intnode_t *node;
577 xfs_da_node_entry_t *btree;
578 int tmp;
579 xfs_mount_t *mp;
580
581 node = oldblk->bp->data;
582 mp = state->mp;
583 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
584 ASSERT((oldblk->index >= 0) && (oldblk->index <= be16_to_cpu(node->hdr.count)));
585 ASSERT(newblk->blkno != 0);
586 if (state->args->whichfork == XFS_DATA_FORK)
587 ASSERT(newblk->blkno >= mp->m_dirleafblk &&
588 newblk->blkno < mp->m_dirfreeblk);
589
590 /*
591 * We may need to make some room before we insert the new node.
592 */
593 tmp = 0;
594 btree = &node->btree[ oldblk->index ];
595 if (oldblk->index < be16_to_cpu(node->hdr.count)) {
596 tmp = (be16_to_cpu(node->hdr.count) - oldblk->index) * (uint)sizeof(*btree);
597 memmove(btree + 1, btree, tmp);
598 }
599 btree->hashval = cpu_to_be32(newblk->hashval);
600 btree->before = cpu_to_be32(newblk->blkno);
601 xfs_da_log_buf(state->args->trans, oldblk->bp,
602 XFS_DA_LOGRANGE(node, btree, tmp + sizeof(*btree)));
603 be16_add_cpu(&node->hdr.count, 1);
604 xfs_da_log_buf(state->args->trans, oldblk->bp,
605 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
606
607 /*
608 * Copy the last hash value from the oldblk to propagate upwards.
609 */
610 oldblk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1 ].hashval);
611 }
612
613 /*========================================================================
614 * Routines used for shrinking the Btree.
615 *========================================================================*/
616
617 /*
618 * Deallocate an empty leaf node, remove it from its parent,
619 * possibly deallocating that block, etc...
620 */
621 int
622 xfs_da_join(xfs_da_state_t *state)
623 {
624 xfs_da_state_blk_t *drop_blk, *save_blk;
625 int action, error;
626
627 action = 0;
628 drop_blk = &state->path.blk[ state->path.active-1 ];
629 save_blk = &state->altpath.blk[ state->path.active-1 ];
630 ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
631 ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
632 drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
633
634 /*
635 * Walk back up the tree joining/deallocating as necessary.
636 * When we stop dropping blocks, break out.
637 */
638 for ( ; state->path.active >= 2; drop_blk--, save_blk--,
639 state->path.active--) {
640 /*
641 * See if we can combine the block with a neighbor.
642 * (action == 0) => no options, just leave
643 * (action == 1) => coalesce, then unlink
644 * (action == 2) => block empty, unlink it
645 */
646 switch (drop_blk->magic) {
647 case XFS_ATTR_LEAF_MAGIC:
648 error = xfs_attr_leaf_toosmall(state, &action);
649 if (error)
650 return(error);
651 if (action == 0)
652 return(0);
653 xfs_attr_leaf_unbalance(state, drop_blk, save_blk);
654 break;
655 case XFS_DIR2_LEAFN_MAGIC:
656 error = xfs_dir2_leafn_toosmall(state, &action);
657 if (error)
658 return error;
659 if (action == 0)
660 return 0;
661 xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
662 break;
663 case XFS_DA_NODE_MAGIC:
664 /*
665 * Remove the offending node, fixup hashvals,
666 * check for a toosmall neighbor.
667 */
668 xfs_da_node_remove(state, drop_blk);
669 xfs_da_fixhashpath(state, &state->path);
670 error = xfs_da_node_toosmall(state, &action);
671 if (error)
672 return(error);
673 if (action == 0)
674 return 0;
675 xfs_da_node_unbalance(state, drop_blk, save_blk);
676 break;
677 }
678 xfs_da_fixhashpath(state, &state->altpath);
679 error = xfs_da_blk_unlink(state, drop_blk, save_blk);
680 xfs_da_state_kill_altpath(state);
681 if (error)
682 return(error);
683 error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
684 drop_blk->bp);
685 drop_blk->bp = NULL;
686 if (error)
687 return(error);
688 }
689 /*
690 * We joined all the way to the top. If it turns out that
691 * we only have one entry in the root, make the child block
692 * the new root.
693 */
694 xfs_da_node_remove(state, drop_blk);
695 xfs_da_fixhashpath(state, &state->path);
696 error = xfs_da_root_join(state, &state->path.blk[0]);
697 return(error);
698 }
699
700 /*
701 * We have only one entry in the root. Copy the only remaining child of
702 * the old root to block 0 as the new root node.
703 */
704 STATIC int
705 xfs_da_root_join(xfs_da_state_t *state, xfs_da_state_blk_t *root_blk)
706 {
707 xfs_da_intnode_t *oldroot;
708 /* REFERENCED */
709 xfs_da_blkinfo_t *blkinfo;
710 xfs_da_args_t *args;
711 xfs_dablk_t child;
712 xfs_dabuf_t *bp;
713 int error;
714
715 args = state->args;
716 ASSERT(args != NULL);
717 ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
718 oldroot = root_blk->bp->data;
719 ASSERT(be16_to_cpu(oldroot->hdr.info.magic) == XFS_DA_NODE_MAGIC);
720 ASSERT(!oldroot->hdr.info.forw);
721 ASSERT(!oldroot->hdr.info.back);
722
723 /*
724 * If the root has more than one child, then don't do anything.
725 */
726 if (be16_to_cpu(oldroot->hdr.count) > 1)
727 return(0);
728
729 /*
730 * Read in the (only) child block, then copy those bytes into
731 * the root block's buffer and free the original child block.
732 */
733 child = be32_to_cpu(oldroot->btree[0].before);
734 ASSERT(child != 0);
735 error = xfs_da_read_buf(args->trans, args->dp, child, -1, &bp,
736 args->whichfork);
737 if (error)
738 return(error);
739 ASSERT(bp != NULL);
740 blkinfo = bp->data;
741 if (be16_to_cpu(oldroot->hdr.level) == 1) {
742 ASSERT(be16_to_cpu(blkinfo->magic) == XFS_DIR2_LEAFN_MAGIC ||
743 be16_to_cpu(blkinfo->magic) == XFS_ATTR_LEAF_MAGIC);
744 } else {
745 ASSERT(be16_to_cpu(blkinfo->magic) == XFS_DA_NODE_MAGIC);
746 }
747 ASSERT(!blkinfo->forw);
748 ASSERT(!blkinfo->back);
749 memcpy(root_blk->bp->data, bp->data, state->blocksize);
750 xfs_da_log_buf(args->trans, root_blk->bp, 0, state->blocksize - 1);
751 error = xfs_da_shrink_inode(args, child, bp);
752 return(error);
753 }
754
755 /*
756 * Check a node block and its neighbors to see if the block should be
757 * collapsed into one or the other neighbor. Always keep the block
758 * with the smaller block number.
759 * If the current block is over 50% full, don't try to join it, return 0.
760 * If the block is empty, fill in the state structure and return 2.
761 * If it can be collapsed, fill in the state structure and return 1.
762 * If nothing can be done, return 0.
763 */
764 STATIC int
765 xfs_da_node_toosmall(xfs_da_state_t *state, int *action)
766 {
767 xfs_da_intnode_t *node;
768 xfs_da_state_blk_t *blk;
769 xfs_da_blkinfo_t *info;
770 int count, forward, error, retval, i;
771 xfs_dablk_t blkno;
772 xfs_dabuf_t *bp;
773
774 /*
775 * Check for the degenerate case of the block being over 50% full.
776 * If so, it's not worth even looking to see if we might be able
777 * to coalesce with a sibling.
778 */
779 blk = &state->path.blk[ state->path.active-1 ];
780 info = blk->bp->data;
781 ASSERT(be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC);
782 node = (xfs_da_intnode_t *)info;
783 count = be16_to_cpu(node->hdr.count);
784 if (count > (state->node_ents >> 1)) {
785 *action = 0; /* blk over 50%, don't try to join */
786 return(0); /* blk over 50%, don't try to join */
787 }
788
789 /*
790 * Check for the degenerate case of the block being empty.
791 * If the block is empty, we'll simply delete it, no need to
792 * coalesce it with a sibling block. We choose (arbitrarily)
793 * to merge with the forward block unless it is NULL.
794 */
795 if (count == 0) {
796 /*
797 * Make altpath point to the block we want to keep and
798 * path point to the block we want to drop (this one).
799 */
800 forward = (info->forw != 0);
801 memcpy(&state->altpath, &state->path, sizeof(state->path));
802 error = xfs_da_path_shift(state, &state->altpath, forward,
803 0, &retval);
804 if (error)
805 return(error);
806 if (retval) {
807 *action = 0;
808 } else {
809 *action = 2;
810 }
811 return(0);
812 }
813
814 /*
815 * Examine each sibling block to see if we can coalesce with
816 * at least 25% free space to spare. We need to figure out
817 * whether to merge with the forward or the backward block.
818 * We prefer coalescing with the lower numbered sibling so as
819 * to shrink a directory over time.
820 */
821 /* start with smaller blk num */
822 forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back));
823 for (i = 0; i < 2; forward = !forward, i++) {
824 if (forward)
825 blkno = be32_to_cpu(info->forw);
826 else
827 blkno = be32_to_cpu(info->back);
828 if (blkno == 0)
829 continue;
830 error = xfs_da_read_buf(state->args->trans, state->args->dp,
831 blkno, -1, &bp, state->args->whichfork);
832 if (error)
833 return(error);
834 ASSERT(bp != NULL);
835
836 node = (xfs_da_intnode_t *)info;
837 count = state->node_ents;
838 count -= state->node_ents >> 2;
839 count -= be16_to_cpu(node->hdr.count);
840 node = bp->data;
841 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
842 count -= be16_to_cpu(node->hdr.count);
843 xfs_da_brelse(state->args->trans, bp);
844 if (count >= 0)
845 break; /* fits with at least 25% to spare */
846 }
847 if (i >= 2) {
848 *action = 0;
849 return(0);
850 }
851
852 /*
853 * Make altpath point to the block we want to keep (the lower
854 * numbered block) and path point to the block we want to drop.
855 */
856 memcpy(&state->altpath, &state->path, sizeof(state->path));
857 if (blkno < blk->blkno) {
858 error = xfs_da_path_shift(state, &state->altpath, forward,
859 0, &retval);
860 if (error) {
861 return(error);
862 }
863 if (retval) {
864 *action = 0;
865 return(0);
866 }
867 } else {
868 error = xfs_da_path_shift(state, &state->path, forward,
869 0, &retval);
870 if (error) {
871 return(error);
872 }
873 if (retval) {
874 *action = 0;
875 return(0);
876 }
877 }
878 *action = 1;
879 return(0);
880 }
881
882 /*
883 * Walk back up the tree adjusting hash values as necessary,
884 * when we stop making changes, return.
885 */
886 void
887 xfs_da_fixhashpath(xfs_da_state_t *state, xfs_da_state_path_t *path)
888 {
889 xfs_da_state_blk_t *blk;
890 xfs_da_intnode_t *node;
891 xfs_da_node_entry_t *btree;
892 xfs_dahash_t lasthash=0;
893 int level, count;
894
895 level = path->active-1;
896 blk = &path->blk[ level ];
897 switch (blk->magic) {
898 case XFS_ATTR_LEAF_MAGIC:
899 lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
900 if (count == 0)
901 return;
902 break;
903 case XFS_DIR2_LEAFN_MAGIC:
904 lasthash = xfs_dir2_leafn_lasthash(blk->bp, &count);
905 if (count == 0)
906 return;
907 break;
908 case XFS_DA_NODE_MAGIC:
909 lasthash = xfs_da_node_lasthash(blk->bp, &count);
910 if (count == 0)
911 return;
912 break;
913 }
914 for (blk--, level--; level >= 0; blk--, level--) {
915 node = blk->bp->data;
916 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
917 btree = &node->btree[ blk->index ];
918 if (be32_to_cpu(btree->hashval) == lasthash)
919 break;
920 blk->hashval = lasthash;
921 btree->hashval = cpu_to_be32(lasthash);
922 xfs_da_log_buf(state->args->trans, blk->bp,
923 XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
924
925 lasthash = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
926 }
927 }
928
929 /*
930 * Remove an entry from an intermediate node.
931 */
932 STATIC void
933 xfs_da_node_remove(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk)
934 {
935 xfs_da_intnode_t *node;
936 xfs_da_node_entry_t *btree;
937 int tmp;
938
939 node = drop_blk->bp->data;
940 ASSERT(drop_blk->index < be16_to_cpu(node->hdr.count));
941 ASSERT(drop_blk->index >= 0);
942
943 /*
944 * Copy over the offending entry, or just zero it out.
945 */
946 btree = &node->btree[drop_blk->index];
947 if (drop_blk->index < (be16_to_cpu(node->hdr.count)-1)) {
948 tmp = be16_to_cpu(node->hdr.count) - drop_blk->index - 1;
949 tmp *= (uint)sizeof(xfs_da_node_entry_t);
950 memmove(btree, btree + 1, tmp);
951 xfs_da_log_buf(state->args->trans, drop_blk->bp,
952 XFS_DA_LOGRANGE(node, btree, tmp));
953 btree = &node->btree[be16_to_cpu(node->hdr.count)-1];
954 }
955 memset((char *)btree, 0, sizeof(xfs_da_node_entry_t));
956 xfs_da_log_buf(state->args->trans, drop_blk->bp,
957 XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
958 be16_add_cpu(&node->hdr.count, -1);
959 xfs_da_log_buf(state->args->trans, drop_blk->bp,
960 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
961
962 /*
963 * Copy the last hash value from the block to propagate upwards.
964 */
965 btree--;
966 drop_blk->hashval = be32_to_cpu(btree->hashval);
967 }
968
969 /*
970 * Unbalance the btree elements between two intermediate nodes,
971 * move all Btree elements from one node into another.
972 */
973 STATIC void
974 xfs_da_node_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
975 xfs_da_state_blk_t *save_blk)
976 {
977 xfs_da_intnode_t *drop_node, *save_node;
978 xfs_da_node_entry_t *btree;
979 int tmp;
980 xfs_trans_t *tp;
981
982 drop_node = drop_blk->bp->data;
983 save_node = save_blk->bp->data;
984 ASSERT(be16_to_cpu(drop_node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
985 ASSERT(be16_to_cpu(save_node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
986 tp = state->args->trans;
987
988 /*
989 * If the dying block has lower hashvals, then move all the
990 * elements in the remaining block up to make a hole.
991 */
992 if ((be32_to_cpu(drop_node->btree[0].hashval) < be32_to_cpu(save_node->btree[ 0 ].hashval)) ||
993 (be32_to_cpu(drop_node->btree[be16_to_cpu(drop_node->hdr.count)-1].hashval) <
994 be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval)))
995 {
996 btree = &save_node->btree[be16_to_cpu(drop_node->hdr.count)];
997 tmp = be16_to_cpu(save_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t);
998 memmove(btree, &save_node->btree[0], tmp);
999 btree = &save_node->btree[0];
1000 xfs_da_log_buf(tp, save_blk->bp,
1001 XFS_DA_LOGRANGE(save_node, btree,
1002 (be16_to_cpu(save_node->hdr.count) + be16_to_cpu(drop_node->hdr.count)) *
1003 sizeof(xfs_da_node_entry_t)));
1004 } else {
1005 btree = &save_node->btree[be16_to_cpu(save_node->hdr.count)];
1006 xfs_da_log_buf(tp, save_blk->bp,
1007 XFS_DA_LOGRANGE(save_node, btree,
1008 be16_to_cpu(drop_node->hdr.count) *
1009 sizeof(xfs_da_node_entry_t)));
1010 }
1011
1012 /*
1013 * Move all the B-tree elements from drop_blk to save_blk.
1014 */
1015 tmp = be16_to_cpu(drop_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t);
1016 memcpy(btree, &drop_node->btree[0], tmp);
1017 be16_add_cpu(&save_node->hdr.count, be16_to_cpu(drop_node->hdr.count));
1018
1019 xfs_da_log_buf(tp, save_blk->bp,
1020 XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1021 sizeof(save_node->hdr)));
1022
1023 /*
1024 * Save the last hashval in the remaining block for upward propagation.
1025 */
1026 save_blk->hashval = be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval);
1027 }
1028
1029 /*========================================================================
1030 * Routines used for finding things in the Btree.
1031 *========================================================================*/
1032
1033 /*
1034 * Walk down the Btree looking for a particular filename, filling
1035 * in the state structure as we go.
1036 *
1037 * We will set the state structure to point to each of the elements
1038 * in each of the nodes where either the hashval is or should be.
1039 *
1040 * We support duplicate hashval's so for each entry in the current
1041 * node that could contain the desired hashval, descend. This is a
1042 * pruned depth-first tree search.
1043 */
1044 int /* error */
1045 xfs_da_node_lookup_int(xfs_da_state_t *state, int *result)
1046 {
1047 xfs_da_state_blk_t *blk;
1048 xfs_da_blkinfo_t *curr;
1049 xfs_da_intnode_t *node;
1050 xfs_da_node_entry_t *btree;
1051 xfs_dablk_t blkno;
1052 int probe, span, max, error, retval;
1053 xfs_dahash_t hashval, btreehashval;
1054 xfs_da_args_t *args;
1055
1056 args = state->args;
1057
1058 /*
1059 * Descend thru the B-tree searching each level for the right
1060 * node to use, until the right hashval is found.
1061 */
1062 blkno = (args->whichfork == XFS_DATA_FORK)? state->mp->m_dirleafblk : 0;
1063 for (blk = &state->path.blk[0], state->path.active = 1;
1064 state->path.active <= XFS_DA_NODE_MAXDEPTH;
1065 blk++, state->path.active++) {
1066 /*
1067 * Read the next node down in the tree.
1068 */
1069 blk->blkno = blkno;
1070 error = xfs_da_read_buf(args->trans, args->dp, blkno,
1071 -1, &blk->bp, args->whichfork);
1072 if (error) {
1073 blk->blkno = 0;
1074 state->path.active--;
1075 return(error);
1076 }
1077 curr = blk->bp->data;
1078 blk->magic = be16_to_cpu(curr->magic);
1079 ASSERT(blk->magic == XFS_DA_NODE_MAGIC ||
1080 blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1081 blk->magic == XFS_ATTR_LEAF_MAGIC);
1082
1083 /*
1084 * Search an intermediate node for a match.
1085 */
1086 if (blk->magic == XFS_DA_NODE_MAGIC) {
1087 node = blk->bp->data;
1088 max = be16_to_cpu(node->hdr.count);
1089 blk->hashval = be32_to_cpu(node->btree[max-1].hashval);
1090
1091 /*
1092 * Binary search. (note: small blocks will skip loop)
1093 */
1094 probe = span = max / 2;
1095 hashval = args->hashval;
1096 for (btree = &node->btree[probe]; span > 4;
1097 btree = &node->btree[probe]) {
1098 span /= 2;
1099 btreehashval = be32_to_cpu(btree->hashval);
1100 if (btreehashval < hashval)
1101 probe += span;
1102 else if (btreehashval > hashval)
1103 probe -= span;
1104 else
1105 break;
1106 }
1107 ASSERT((probe >= 0) && (probe < max));
1108 ASSERT((span <= 4) || (be32_to_cpu(btree->hashval) == hashval));
1109
1110 /*
1111 * Since we may have duplicate hashval's, find the first
1112 * matching hashval in the node.
1113 */
1114 while ((probe > 0) && (be32_to_cpu(btree->hashval) >= hashval)) {
1115 btree--;
1116 probe--;
1117 }
1118 while ((probe < max) && (be32_to_cpu(btree->hashval) < hashval)) {
1119 btree++;
1120 probe++;
1121 }
1122
1123 /*
1124 * Pick the right block to descend on.
1125 */
1126 if (probe == max) {
1127 blk->index = max-1;
1128 blkno = be32_to_cpu(node->btree[max-1].before);
1129 } else {
1130 blk->index = probe;
1131 blkno = be32_to_cpu(btree->before);
1132 }
1133 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1134 blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1135 break;
1136 } else if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1137 blk->hashval = xfs_dir2_leafn_lasthash(blk->bp, NULL);
1138 break;
1139 }
1140 }
1141
1142 /*
1143 * A leaf block that ends in the hashval that we are interested in
1144 * (final hashval == search hashval) means that the next block may
1145 * contain more entries with the same hashval, shift upward to the
1146 * next leaf and keep searching.
1147 */
1148 for (;;) {
1149 if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1150 retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
1151 &blk->index, state);
1152 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1153 retval = xfs_attr_leaf_lookup_int(blk->bp, args);
1154 blk->index = args->index;
1155 args->blkno = blk->blkno;
1156 } else {
1157 ASSERT(0);
1158 return XFS_ERROR(EFSCORRUPTED);
1159 }
1160 if (((retval == ENOENT) || (retval == ENOATTR)) &&
1161 (blk->hashval == args->hashval)) {
1162 error = xfs_da_path_shift(state, &state->path, 1, 1,
1163 &retval);
1164 if (error)
1165 return(error);
1166 if (retval == 0) {
1167 continue;
1168 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1169 /* path_shift() gives ENOENT */
1170 retval = XFS_ERROR(ENOATTR);
1171 }
1172 }
1173 break;
1174 }
1175 *result = retval;
1176 return(0);
1177 }
1178
1179 /*========================================================================
1180 * Utility routines.
1181 *========================================================================*/
1182
1183 /*
1184 * Link a new block into a doubly linked list of blocks (of whatever type).
1185 */
1186 int /* error */
1187 xfs_da_blk_link(xfs_da_state_t *state, xfs_da_state_blk_t *old_blk,
1188 xfs_da_state_blk_t *new_blk)
1189 {
1190 xfs_da_blkinfo_t *old_info, *new_info, *tmp_info;
1191 xfs_da_args_t *args;
1192 int before=0, error;
1193 xfs_dabuf_t *bp;
1194
1195 /*
1196 * Set up environment.
1197 */
1198 args = state->args;
1199 ASSERT(args != NULL);
1200 old_info = old_blk->bp->data;
1201 new_info = new_blk->bp->data;
1202 ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
1203 old_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1204 old_blk->magic == XFS_ATTR_LEAF_MAGIC);
1205 ASSERT(old_blk->magic == be16_to_cpu(old_info->magic));
1206 ASSERT(new_blk->magic == be16_to_cpu(new_info->magic));
1207 ASSERT(old_blk->magic == new_blk->magic);
1208
1209 switch (old_blk->magic) {
1210 case XFS_ATTR_LEAF_MAGIC:
1211 before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
1212 break;
1213 case XFS_DIR2_LEAFN_MAGIC:
1214 before = xfs_dir2_leafn_order(old_blk->bp, new_blk->bp);
1215 break;
1216 case XFS_DA_NODE_MAGIC:
1217 before = xfs_da_node_order(old_blk->bp, new_blk->bp);
1218 break;
1219 }
1220
1221 /*
1222 * Link blocks in appropriate order.
1223 */
1224 if (before) {
1225 /*
1226 * Link new block in before existing block.
1227 */
1228 new_info->forw = cpu_to_be32(old_blk->blkno);
1229 new_info->back = old_info->back;
1230 if (old_info->back) {
1231 error = xfs_da_read_buf(args->trans, args->dp,
1232 be32_to_cpu(old_info->back),
1233 -1, &bp, args->whichfork);
1234 if (error)
1235 return(error);
1236 ASSERT(bp != NULL);
1237 tmp_info = bp->data;
1238 ASSERT(be16_to_cpu(tmp_info->magic) == be16_to_cpu(old_info->magic));
1239 ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno);
1240 tmp_info->forw = cpu_to_be32(new_blk->blkno);
1241 xfs_da_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1242 xfs_da_buf_done(bp);
1243 }
1244 old_info->back = cpu_to_be32(new_blk->blkno);
1245 } else {
1246 /*
1247 * Link new block in after existing block.
1248 */
1249 new_info->forw = old_info->forw;
1250 new_info->back = cpu_to_be32(old_blk->blkno);
1251 if (old_info->forw) {
1252 error = xfs_da_read_buf(args->trans, args->dp,
1253 be32_to_cpu(old_info->forw),
1254 -1, &bp, args->whichfork);
1255 if (error)
1256 return(error);
1257 ASSERT(bp != NULL);
1258 tmp_info = bp->data;
1259 ASSERT(tmp_info->magic == old_info->magic);
1260 ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno);
1261 tmp_info->back = cpu_to_be32(new_blk->blkno);
1262 xfs_da_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1263 xfs_da_buf_done(bp);
1264 }
1265 old_info->forw = cpu_to_be32(new_blk->blkno);
1266 }
1267
1268 xfs_da_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
1269 xfs_da_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
1270 return(0);
1271 }
1272
1273 /*
1274 * Compare two intermediate nodes for "order".
1275 */
1276 STATIC int
1277 xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp)
1278 {
1279 xfs_da_intnode_t *node1, *node2;
1280
1281 node1 = node1_bp->data;
1282 node2 = node2_bp->data;
1283 ASSERT((be16_to_cpu(node1->hdr.info.magic) == XFS_DA_NODE_MAGIC) &&
1284 (be16_to_cpu(node2->hdr.info.magic) == XFS_DA_NODE_MAGIC));
1285 if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) &&
1286 ((be32_to_cpu(node2->btree[0].hashval) <
1287 be32_to_cpu(node1->btree[0].hashval)) ||
1288 (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) <
1289 be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) {
1290 return(1);
1291 }
1292 return(0);
1293 }
1294
1295 /*
1296 * Pick up the last hashvalue from an intermediate node.
1297 */
1298 STATIC uint
1299 xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count)
1300 {
1301 xfs_da_intnode_t *node;
1302
1303 node = bp->data;
1304 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
1305 if (count)
1306 *count = be16_to_cpu(node->hdr.count);
1307 if (!node->hdr.count)
1308 return(0);
1309 return be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
1310 }
1311
1312 /*
1313 * Unlink a block from a doubly linked list of blocks.
1314 */
1315 STATIC int /* error */
1316 xfs_da_blk_unlink(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1317 xfs_da_state_blk_t *save_blk)
1318 {
1319 xfs_da_blkinfo_t *drop_info, *save_info, *tmp_info;
1320 xfs_da_args_t *args;
1321 xfs_dabuf_t *bp;
1322 int error;
1323
1324 /*
1325 * Set up environment.
1326 */
1327 args = state->args;
1328 ASSERT(args != NULL);
1329 save_info = save_blk->bp->data;
1330 drop_info = drop_blk->bp->data;
1331 ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
1332 save_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1333 save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1334 ASSERT(save_blk->magic == be16_to_cpu(save_info->magic));
1335 ASSERT(drop_blk->magic == be16_to_cpu(drop_info->magic));
1336 ASSERT(save_blk->magic == drop_blk->magic);
1337 ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) ||
1338 (be32_to_cpu(save_info->back) == drop_blk->blkno));
1339 ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) ||
1340 (be32_to_cpu(drop_info->back) == save_blk->blkno));
1341
1342 /*
1343 * Unlink the leaf block from the doubly linked chain of leaves.
1344 */
1345 if (be32_to_cpu(save_info->back) == drop_blk->blkno) {
1346 save_info->back = drop_info->back;
1347 if (drop_info->back) {
1348 error = xfs_da_read_buf(args->trans, args->dp,
1349 be32_to_cpu(drop_info->back),
1350 -1, &bp, args->whichfork);
1351 if (error)
1352 return(error);
1353 ASSERT(bp != NULL);
1354 tmp_info = bp->data;
1355 ASSERT(tmp_info->magic == save_info->magic);
1356 ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno);
1357 tmp_info->forw = cpu_to_be32(save_blk->blkno);
1358 xfs_da_log_buf(args->trans, bp, 0,
1359 sizeof(*tmp_info) - 1);
1360 xfs_da_buf_done(bp);
1361 }
1362 } else {
1363 save_info->forw = drop_info->forw;
1364 if (drop_info->forw) {
1365 error = xfs_da_read_buf(args->trans, args->dp,
1366 be32_to_cpu(drop_info->forw),
1367 -1, &bp, args->whichfork);
1368 if (error)
1369 return(error);
1370 ASSERT(bp != NULL);
1371 tmp_info = bp->data;
1372 ASSERT(tmp_info->magic == save_info->magic);
1373 ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno);
1374 tmp_info->back = cpu_to_be32(save_blk->blkno);
1375 xfs_da_log_buf(args->trans, bp, 0,
1376 sizeof(*tmp_info) - 1);
1377 xfs_da_buf_done(bp);
1378 }
1379 }
1380
1381 xfs_da_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
1382 return(0);
1383 }
1384
1385 /*
1386 * Move a path "forward" or "!forward" one block at the current level.
1387 *
1388 * This routine will adjust a "path" to point to the next block
1389 * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1390 * Btree, including updating pointers to the intermediate nodes between
1391 * the new bottom and the root.
1392 */
1393 int /* error */
1394 xfs_da_path_shift(xfs_da_state_t *state, xfs_da_state_path_t *path,
1395 int forward, int release, int *result)
1396 {
1397 xfs_da_state_blk_t *blk;
1398 xfs_da_blkinfo_t *info;
1399 xfs_da_intnode_t *node;
1400 xfs_da_args_t *args;
1401 xfs_dablk_t blkno=0;
1402 int level, error;
1403
1404 /*
1405 * Roll up the Btree looking for the first block where our
1406 * current index is not at the edge of the block. Note that
1407 * we skip the bottom layer because we want the sibling block.
1408 */
1409 args = state->args;
1410 ASSERT(args != NULL);
1411 ASSERT(path != NULL);
1412 ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1413 level = (path->active-1) - 1; /* skip bottom layer in path */
1414 for (blk = &path->blk[level]; level >= 0; blk--, level--) {
1415 ASSERT(blk->bp != NULL);
1416 node = blk->bp->data;
1417 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
1418 if (forward && (blk->index < be16_to_cpu(node->hdr.count)-1)) {
1419 blk->index++;
1420 blkno = be32_to_cpu(node->btree[blk->index].before);
1421 break;
1422 } else if (!forward && (blk->index > 0)) {
1423 blk->index--;
1424 blkno = be32_to_cpu(node->btree[blk->index].before);
1425 break;
1426 }
1427 }
1428 if (level < 0) {
1429 *result = XFS_ERROR(ENOENT); /* we're out of our tree */
1430 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1431 return(0);
1432 }
1433
1434 /*
1435 * Roll down the edge of the subtree until we reach the
1436 * same depth we were at originally.
1437 */
1438 for (blk++, level++; level < path->active; blk++, level++) {
1439 /*
1440 * Release the old block.
1441 * (if it's dirty, trans won't actually let go)
1442 */
1443 if (release)
1444 xfs_da_brelse(args->trans, blk->bp);
1445
1446 /*
1447 * Read the next child block.
1448 */
1449 blk->blkno = blkno;
1450 error = xfs_da_read_buf(args->trans, args->dp, blkno, -1,
1451 &blk->bp, args->whichfork);
1452 if (error)
1453 return(error);
1454 ASSERT(blk->bp != NULL);
1455 info = blk->bp->data;
1456 ASSERT(be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC ||
1457 be16_to_cpu(info->magic) == XFS_DIR2_LEAFN_MAGIC ||
1458 be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC);
1459 blk->magic = be16_to_cpu(info->magic);
1460 if (blk->magic == XFS_DA_NODE_MAGIC) {
1461 node = (xfs_da_intnode_t *)info;
1462 blk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
1463 if (forward)
1464 blk->index = 0;
1465 else
1466 blk->index = be16_to_cpu(node->hdr.count)-1;
1467 blkno = be32_to_cpu(node->btree[blk->index].before);
1468 } else {
1469 ASSERT(level == path->active-1);
1470 blk->index = 0;
1471 switch(blk->magic) {
1472 case XFS_ATTR_LEAF_MAGIC:
1473 blk->hashval = xfs_attr_leaf_lasthash(blk->bp,
1474 NULL);
1475 break;
1476 case XFS_DIR2_LEAFN_MAGIC:
1477 blk->hashval = xfs_dir2_leafn_lasthash(blk->bp,
1478 NULL);
1479 break;
1480 default:
1481 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC ||
1482 blk->magic == XFS_DIR2_LEAFN_MAGIC);
1483 break;
1484 }
1485 }
1486 }
1487 *result = 0;
1488 return(0);
1489 }
1490
1491
1492 /*========================================================================
1493 * Utility routines.
1494 *========================================================================*/
1495
1496 /*
1497 * Implement a simple hash on a character string.
1498 * Rotate the hash value by 7 bits, then XOR each character in.
1499 * This is implemented with some source-level loop unrolling.
1500 */
1501 xfs_dahash_t
1502 xfs_da_hashname(const __uint8_t *name, int namelen)
1503 {
1504 xfs_dahash_t hash;
1505
1506 /*
1507 * Do four characters at a time as long as we can.
1508 */
1509 for (hash = 0; namelen >= 4; namelen -= 4, name += 4)
1510 hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^
1511 (name[3] << 0) ^ rol32(hash, 7 * 4);
1512
1513 /*
1514 * Now do the rest of the characters.
1515 */
1516 switch (namelen) {
1517 case 3:
1518 return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^
1519 rol32(hash, 7 * 3);
1520 case 2:
1521 return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2);
1522 case 1:
1523 return (name[0] << 0) ^ rol32(hash, 7 * 1);
1524 default: /* case 0: */
1525 return hash;
1526 }
1527 }
1528
1529 enum xfs_dacmp
1530 xfs_da_compname(
1531 struct xfs_da_args *args,
1532 const unsigned char *name,
1533 int len)
1534 {
1535 return (args->namelen == len && memcmp(args->name, name, len) == 0) ?
1536 XFS_CMP_EXACT : XFS_CMP_DIFFERENT;
1537 }
1538
1539 static xfs_dahash_t
1540 xfs_default_hashname(
1541 struct xfs_name *name)
1542 {
1543 return xfs_da_hashname(name->name, name->len);
1544 }
1545
1546 const struct xfs_nameops xfs_default_nameops = {
1547 .hashname = xfs_default_hashname,
1548 .compname = xfs_da_compname
1549 };
1550
1551 /*
1552 * Add a block to the btree ahead of the file.
1553 * Return the new block number to the caller.
1554 */
1555 int
1556 xfs_da_grow_inode(xfs_da_args_t *args, xfs_dablk_t *new_blkno)
1557 {
1558 xfs_fileoff_t bno, b;
1559 xfs_bmbt_irec_t map;
1560 xfs_bmbt_irec_t *mapp;
1561 xfs_inode_t *dp;
1562 int nmap, error, w, count, c, got, i, mapi;
1563 xfs_trans_t *tp;
1564 xfs_mount_t *mp;
1565 xfs_drfsbno_t nblks;
1566
1567 dp = args->dp;
1568 mp = dp->i_mount;
1569 w = args->whichfork;
1570 tp = args->trans;
1571 nblks = dp->i_d.di_nblocks;
1572
1573 /*
1574 * For new directories adjust the file offset and block count.
1575 */
1576 if (w == XFS_DATA_FORK) {
1577 bno = mp->m_dirleafblk;
1578 count = mp->m_dirblkfsbs;
1579 } else {
1580 bno = 0;
1581 count = 1;
1582 }
1583 /*
1584 * Find a spot in the file space to put the new block.
1585 */
1586 if ((error = xfs_bmap_first_unused(tp, dp, count, &bno, w)))
1587 return error;
1588 if (w == XFS_DATA_FORK)
1589 ASSERT(bno >= mp->m_dirleafblk && bno < mp->m_dirfreeblk);
1590 /*
1591 * Try mapping it in one filesystem block.
1592 */
1593 nmap = 1;
1594 ASSERT(args->firstblock != NULL);
1595 if ((error = xfs_bmapi(tp, dp, bno, count,
1596 xfs_bmapi_aflag(w)|XFS_BMAPI_WRITE|XFS_BMAPI_METADATA|
1597 XFS_BMAPI_CONTIG,
1598 args->firstblock, args->total, &map, &nmap,
1599 args->flist))) {
1600 return error;
1601 }
1602 ASSERT(nmap <= 1);
1603 if (nmap == 1) {
1604 mapp = &map;
1605 mapi = 1;
1606 }
1607 /*
1608 * If we didn't get it and the block might work if fragmented,
1609 * try without the CONTIG flag. Loop until we get it all.
1610 */
1611 else if (nmap == 0 && count > 1) {
1612 mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP);
1613 for (b = bno, mapi = 0; b < bno + count; ) {
1614 nmap = MIN(XFS_BMAP_MAX_NMAP, count);
1615 c = (int)(bno + count - b);
1616 if ((error = xfs_bmapi(tp, dp, b, c,
1617 xfs_bmapi_aflag(w)|XFS_BMAPI_WRITE|
1618 XFS_BMAPI_METADATA,
1619 args->firstblock, args->total,
1620 &mapp[mapi], &nmap, args->flist))) {
1621 kmem_free(mapp);
1622 return error;
1623 }
1624 if (nmap < 1)
1625 break;
1626 mapi += nmap;
1627 b = mapp[mapi - 1].br_startoff +
1628 mapp[mapi - 1].br_blockcount;
1629 }
1630 } else {
1631 mapi = 0;
1632 mapp = NULL;
1633 }
1634 /*
1635 * Count the blocks we got, make sure it matches the total.
1636 */
1637 for (i = 0, got = 0; i < mapi; i++)
1638 got += mapp[i].br_blockcount;
1639 if (got != count || mapp[0].br_startoff != bno ||
1640 mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
1641 bno + count) {
1642 if (mapp != &map)
1643 kmem_free(mapp);
1644 return XFS_ERROR(ENOSPC);
1645 }
1646 if (mapp != &map)
1647 kmem_free(mapp);
1648 /* account for newly allocated blocks in reserved blocks total */
1649 args->total -= dp->i_d.di_nblocks - nblks;
1650 *new_blkno = (xfs_dablk_t)bno;
1651 return 0;
1652 }
1653
1654 /*
1655 * Ick. We need to always be able to remove a btree block, even
1656 * if there's no space reservation because the filesystem is full.
1657 * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
1658 * It swaps the target block with the last block in the file. The
1659 * last block in the file can always be removed since it can't cause
1660 * a bmap btree split to do that.
1661 */
1662 STATIC int
1663 xfs_da_swap_lastblock(xfs_da_args_t *args, xfs_dablk_t *dead_blknop,
1664 xfs_dabuf_t **dead_bufp)
1665 {
1666 xfs_dablk_t dead_blkno, last_blkno, sib_blkno, par_blkno;
1667 xfs_dabuf_t *dead_buf, *last_buf, *sib_buf, *par_buf;
1668 xfs_fileoff_t lastoff;
1669 xfs_inode_t *ip;
1670 xfs_trans_t *tp;
1671 xfs_mount_t *mp;
1672 int error, w, entno, level, dead_level;
1673 xfs_da_blkinfo_t *dead_info, *sib_info;
1674 xfs_da_intnode_t *par_node, *dead_node;
1675 xfs_dir2_leaf_t *dead_leaf2;
1676 xfs_dahash_t dead_hash;
1677
1678 dead_buf = *dead_bufp;
1679 dead_blkno = *dead_blknop;
1680 tp = args->trans;
1681 ip = args->dp;
1682 w = args->whichfork;
1683 ASSERT(w == XFS_DATA_FORK);
1684 mp = ip->i_mount;
1685 lastoff = mp->m_dirfreeblk;
1686 error = xfs_bmap_last_before(tp, ip, &lastoff, w);
1687 if (error)
1688 return error;
1689 if (unlikely(lastoff == 0)) {
1690 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW,
1691 mp);
1692 return XFS_ERROR(EFSCORRUPTED);
1693 }
1694 /*
1695 * Read the last block in the btree space.
1696 */
1697 last_blkno = (xfs_dablk_t)lastoff - mp->m_dirblkfsbs;
1698 if ((error = xfs_da_read_buf(tp, ip, last_blkno, -1, &last_buf, w)))
1699 return error;
1700 /*
1701 * Copy the last block into the dead buffer and log it.
1702 */
1703 memcpy(dead_buf->data, last_buf->data, mp->m_dirblksize);
1704 xfs_da_log_buf(tp, dead_buf, 0, mp->m_dirblksize - 1);
1705 dead_info = dead_buf->data;
1706 /*
1707 * Get values from the moved block.
1708 */
1709 if (be16_to_cpu(dead_info->magic) == XFS_DIR2_LEAFN_MAGIC) {
1710 dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
1711 dead_level = 0;
1712 dead_hash = be32_to_cpu(dead_leaf2->ents[be16_to_cpu(dead_leaf2->hdr.count) - 1].hashval);
1713 } else {
1714 ASSERT(be16_to_cpu(dead_info->magic) == XFS_DA_NODE_MAGIC);
1715 dead_node = (xfs_da_intnode_t *)dead_info;
1716 dead_level = be16_to_cpu(dead_node->hdr.level);
1717 dead_hash = be32_to_cpu(dead_node->btree[be16_to_cpu(dead_node->hdr.count) - 1].hashval);
1718 }
1719 sib_buf = par_buf = NULL;
1720 /*
1721 * If the moved block has a left sibling, fix up the pointers.
1722 */
1723 if ((sib_blkno = be32_to_cpu(dead_info->back))) {
1724 if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w)))
1725 goto done;
1726 sib_info = sib_buf->data;
1727 if (unlikely(
1728 be32_to_cpu(sib_info->forw) != last_blkno ||
1729 sib_info->magic != dead_info->magic)) {
1730 XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
1731 XFS_ERRLEVEL_LOW, mp);
1732 error = XFS_ERROR(EFSCORRUPTED);
1733 goto done;
1734 }
1735 sib_info->forw = cpu_to_be32(dead_blkno);
1736 xfs_da_log_buf(tp, sib_buf,
1737 XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
1738 sizeof(sib_info->forw)));
1739 xfs_da_buf_done(sib_buf);
1740 sib_buf = NULL;
1741 }
1742 /*
1743 * If the moved block has a right sibling, fix up the pointers.
1744 */
1745 if ((sib_blkno = be32_to_cpu(dead_info->forw))) {
1746 if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w)))
1747 goto done;
1748 sib_info = sib_buf->data;
1749 if (unlikely(
1750 be32_to_cpu(sib_info->back) != last_blkno ||
1751 sib_info->magic != dead_info->magic)) {
1752 XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
1753 XFS_ERRLEVEL_LOW, mp);
1754 error = XFS_ERROR(EFSCORRUPTED);
1755 goto done;
1756 }
1757 sib_info->back = cpu_to_be32(dead_blkno);
1758 xfs_da_log_buf(tp, sib_buf,
1759 XFS_DA_LOGRANGE(sib_info, &sib_info->back,
1760 sizeof(sib_info->back)));
1761 xfs_da_buf_done(sib_buf);
1762 sib_buf = NULL;
1763 }
1764 par_blkno = mp->m_dirleafblk;
1765 level = -1;
1766 /*
1767 * Walk down the tree looking for the parent of the moved block.
1768 */
1769 for (;;) {
1770 if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w)))
1771 goto done;
1772 par_node = par_buf->data;
1773 if (unlikely(
1774 be16_to_cpu(par_node->hdr.info.magic) != XFS_DA_NODE_MAGIC ||
1775 (level >= 0 && level != be16_to_cpu(par_node->hdr.level) + 1))) {
1776 XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
1777 XFS_ERRLEVEL_LOW, mp);
1778 error = XFS_ERROR(EFSCORRUPTED);
1779 goto done;
1780 }
1781 level = be16_to_cpu(par_node->hdr.level);
1782 for (entno = 0;
1783 entno < be16_to_cpu(par_node->hdr.count) &&
1784 be32_to_cpu(par_node->btree[entno].hashval) < dead_hash;
1785 entno++)
1786 continue;
1787 if (unlikely(entno == be16_to_cpu(par_node->hdr.count))) {
1788 XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
1789 XFS_ERRLEVEL_LOW, mp);
1790 error = XFS_ERROR(EFSCORRUPTED);
1791 goto done;
1792 }
1793 par_blkno = be32_to_cpu(par_node->btree[entno].before);
1794 if (level == dead_level + 1)
1795 break;
1796 xfs_da_brelse(tp, par_buf);
1797 par_buf = NULL;
1798 }
1799 /*
1800 * We're in the right parent block.
1801 * Look for the right entry.
1802 */
1803 for (;;) {
1804 for (;
1805 entno < be16_to_cpu(par_node->hdr.count) &&
1806 be32_to_cpu(par_node->btree[entno].before) != last_blkno;
1807 entno++)
1808 continue;
1809 if (entno < be16_to_cpu(par_node->hdr.count))
1810 break;
1811 par_blkno = be32_to_cpu(par_node->hdr.info.forw);
1812 xfs_da_brelse(tp, par_buf);
1813 par_buf = NULL;
1814 if (unlikely(par_blkno == 0)) {
1815 XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
1816 XFS_ERRLEVEL_LOW, mp);
1817 error = XFS_ERROR(EFSCORRUPTED);
1818 goto done;
1819 }
1820 if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w)))
1821 goto done;
1822 par_node = par_buf->data;
1823 if (unlikely(
1824 be16_to_cpu(par_node->hdr.level) != level ||
1825 be16_to_cpu(par_node->hdr.info.magic) != XFS_DA_NODE_MAGIC)) {
1826 XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
1827 XFS_ERRLEVEL_LOW, mp);
1828 error = XFS_ERROR(EFSCORRUPTED);
1829 goto done;
1830 }
1831 entno = 0;
1832 }
1833 /*
1834 * Update the parent entry pointing to the moved block.
1835 */
1836 par_node->btree[entno].before = cpu_to_be32(dead_blkno);
1837 xfs_da_log_buf(tp, par_buf,
1838 XFS_DA_LOGRANGE(par_node, &par_node->btree[entno].before,
1839 sizeof(par_node->btree[entno].before)));
1840 xfs_da_buf_done(par_buf);
1841 xfs_da_buf_done(dead_buf);
1842 *dead_blknop = last_blkno;
1843 *dead_bufp = last_buf;
1844 return 0;
1845 done:
1846 if (par_buf)
1847 xfs_da_brelse(tp, par_buf);
1848 if (sib_buf)
1849 xfs_da_brelse(tp, sib_buf);
1850 xfs_da_brelse(tp, last_buf);
1851 return error;
1852 }
1853
1854 /*
1855 * Remove a btree block from a directory or attribute.
1856 */
1857 int
1858 xfs_da_shrink_inode(xfs_da_args_t *args, xfs_dablk_t dead_blkno,
1859 xfs_dabuf_t *dead_buf)
1860 {
1861 xfs_inode_t *dp;
1862 int done, error, w, count;
1863 xfs_trans_t *tp;
1864 xfs_mount_t *mp;
1865
1866 dp = args->dp;
1867 w = args->whichfork;
1868 tp = args->trans;
1869 mp = dp->i_mount;
1870 if (w == XFS_DATA_FORK)
1871 count = mp->m_dirblkfsbs;
1872 else
1873 count = 1;
1874 for (;;) {
1875 /*
1876 * Remove extents. If we get ENOSPC for a dir we have to move
1877 * the last block to the place we want to kill.
1878 */
1879 if ((error = xfs_bunmapi(tp, dp, dead_blkno, count,
1880 xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
1881 0, args->firstblock, args->flist,
1882 &done)) == ENOSPC) {
1883 if (w != XFS_DATA_FORK)
1884 break;
1885 if ((error = xfs_da_swap_lastblock(args, &dead_blkno,
1886 &dead_buf)))
1887 break;
1888 } else {
1889 break;
1890 }
1891 }
1892 xfs_da_binval(tp, dead_buf);
1893 return error;
1894 }
1895
1896 /*
1897 * See if the mapping(s) for this btree block are valid, i.e.
1898 * don't contain holes, are logically contiguous, and cover the whole range.
1899 */
1900 STATIC int
1901 xfs_da_map_covers_blocks(
1902 int nmap,
1903 xfs_bmbt_irec_t *mapp,
1904 xfs_dablk_t bno,
1905 int count)
1906 {
1907 int i;
1908 xfs_fileoff_t off;
1909
1910 for (i = 0, off = bno; i < nmap; i++) {
1911 if (mapp[i].br_startblock == HOLESTARTBLOCK ||
1912 mapp[i].br_startblock == DELAYSTARTBLOCK) {
1913 return 0;
1914 }
1915 if (off != mapp[i].br_startoff) {
1916 return 0;
1917 }
1918 off += mapp[i].br_blockcount;
1919 }
1920 return off == bno + count;
1921 }
1922
1923 /*
1924 * Make a dabuf.
1925 * Used for get_buf, read_buf, read_bufr, and reada_buf.
1926 */
1927 STATIC int
1928 xfs_da_do_buf(
1929 xfs_trans_t *trans,
1930 xfs_inode_t *dp,
1931 xfs_dablk_t bno,
1932 xfs_daddr_t *mappedbnop,
1933 xfs_dabuf_t **bpp,
1934 int whichfork,
1935 int caller,
1936 inst_t *ra)
1937 {
1938 xfs_buf_t *bp = NULL;
1939 xfs_buf_t **bplist;
1940 int error=0;
1941 int i;
1942 xfs_bmbt_irec_t map;
1943 xfs_bmbt_irec_t *mapp;
1944 xfs_daddr_t mappedbno;
1945 xfs_mount_t *mp;
1946 int nbplist=0;
1947 int nfsb;
1948 int nmap;
1949 xfs_dabuf_t *rbp;
1950
1951 mp = dp->i_mount;
1952 nfsb = (whichfork == XFS_DATA_FORK) ? mp->m_dirblkfsbs : 1;
1953 mappedbno = *mappedbnop;
1954 /*
1955 * Caller doesn't have a mapping. -2 means don't complain
1956 * if we land in a hole.
1957 */
1958 if (mappedbno == -1 || mappedbno == -2) {
1959 /*
1960 * Optimize the one-block case.
1961 */
1962 if (nfsb == 1) {
1963 xfs_fsblock_t fsb;
1964
1965 if ((error =
1966 xfs_bmapi_single(trans, dp, whichfork, &fsb,
1967 (xfs_fileoff_t)bno))) {
1968 return error;
1969 }
1970 mapp = &map;
1971 if (fsb == NULLFSBLOCK) {
1972 nmap = 0;
1973 } else {
1974 map.br_startblock = fsb;
1975 map.br_startoff = (xfs_fileoff_t)bno;
1976 map.br_blockcount = 1;
1977 nmap = 1;
1978 }
1979 } else {
1980 mapp = kmem_alloc(sizeof(*mapp) * nfsb, KM_SLEEP);
1981 nmap = nfsb;
1982 if ((error = xfs_bmapi(trans, dp, (xfs_fileoff_t)bno,
1983 nfsb,
1984 XFS_BMAPI_METADATA |
1985 xfs_bmapi_aflag(whichfork),
1986 NULL, 0, mapp, &nmap, NULL)))
1987 goto exit0;
1988 }
1989 } else {
1990 map.br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno);
1991 map.br_startoff = (xfs_fileoff_t)bno;
1992 map.br_blockcount = nfsb;
1993 mapp = &map;
1994 nmap = 1;
1995 }
1996 if (!xfs_da_map_covers_blocks(nmap, mapp, bno, nfsb)) {
1997 error = mappedbno == -2 ? 0 : XFS_ERROR(EFSCORRUPTED);
1998 if (unlikely(error == EFSCORRUPTED)) {
1999 if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
2000 cmn_err(CE_ALERT, "xfs_da_do_buf: bno %lld\n",
2001 (long long)bno);
2002 cmn_err(CE_ALERT, "dir: inode %lld\n",
2003 (long long)dp->i_ino);
2004 for (i = 0; i < nmap; i++) {
2005 cmn_err(CE_ALERT,
2006 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d\n",
2007 i,
2008 (long long)mapp[i].br_startoff,
2009 (long long)mapp[i].br_startblock,
2010 (long long)mapp[i].br_blockcount,
2011 mapp[i].br_state);
2012 }
2013 }
2014 XFS_ERROR_REPORT("xfs_da_do_buf(1)",
2015 XFS_ERRLEVEL_LOW, mp);
2016 }
2017 goto exit0;
2018 }
2019 if (caller != 3 && nmap > 1) {
2020 bplist = kmem_alloc(sizeof(*bplist) * nmap, KM_SLEEP);
2021 nbplist = 0;
2022 } else
2023 bplist = NULL;
2024 /*
2025 * Turn the mapping(s) into buffer(s).
2026 */
2027 for (i = 0; i < nmap; i++) {
2028 int nmapped;
2029
2030 mappedbno = XFS_FSB_TO_DADDR(mp, mapp[i].br_startblock);
2031 if (i == 0)
2032 *mappedbnop = mappedbno;
2033 nmapped = (int)XFS_FSB_TO_BB(mp, mapp[i].br_blockcount);
2034 switch (caller) {
2035 case 0:
2036 bp = xfs_trans_get_buf(trans, mp->m_ddev_targp,
2037 mappedbno, nmapped, 0);
2038 error = bp ? XFS_BUF_GETERROR(bp) : XFS_ERROR(EIO);
2039 break;
2040 case 1:
2041 case 2:
2042 bp = NULL;
2043 error = xfs_trans_read_buf(mp, trans, mp->m_ddev_targp,
2044 mappedbno, nmapped, 0, &bp);
2045 break;
2046 case 3:
2047 xfs_baread(mp->m_ddev_targp, mappedbno, nmapped);
2048 error = 0;
2049 bp = NULL;
2050 break;
2051 }
2052 if (error) {
2053 if (bp)
2054 xfs_trans_brelse(trans, bp);
2055 goto exit1;
2056 }
2057 if (!bp)
2058 continue;
2059 if (caller == 1) {
2060 if (whichfork == XFS_ATTR_FORK) {
2061 XFS_BUF_SET_VTYPE_REF(bp, B_FS_ATTR_BTREE,
2062 XFS_ATTR_BTREE_REF);
2063 } else {
2064 XFS_BUF_SET_VTYPE_REF(bp, B_FS_DIR_BTREE,
2065 XFS_DIR_BTREE_REF);
2066 }
2067 }
2068 if (bplist) {
2069 bplist[nbplist++] = bp;
2070 }
2071 }
2072 /*
2073 * Build a dabuf structure.
2074 */
2075 if (bplist) {
2076 rbp = xfs_da_buf_make(nbplist, bplist, ra);
2077 } else if (bp)
2078 rbp = xfs_da_buf_make(1, &bp, ra);
2079 else
2080 rbp = NULL;
2081 /*
2082 * For read_buf, check the magic number.
2083 */
2084 if (caller == 1) {
2085 xfs_dir2_data_t *data;
2086 xfs_dir2_free_t *free;
2087 xfs_da_blkinfo_t *info;
2088 uint magic, magic1;
2089
2090 info = rbp->data;
2091 data = rbp->data;
2092 free = rbp->data;
2093 magic = be16_to_cpu(info->magic);
2094 magic1 = be32_to_cpu(data->hdr.magic);
2095 if (unlikely(
2096 XFS_TEST_ERROR((magic != XFS_DA_NODE_MAGIC) &&
2097 (magic != XFS_ATTR_LEAF_MAGIC) &&
2098 (magic != XFS_DIR2_LEAF1_MAGIC) &&
2099 (magic != XFS_DIR2_LEAFN_MAGIC) &&
2100 (magic1 != XFS_DIR2_BLOCK_MAGIC) &&
2101 (magic1 != XFS_DIR2_DATA_MAGIC) &&
2102 (be32_to_cpu(free->hdr.magic) != XFS_DIR2_FREE_MAGIC),
2103 mp, XFS_ERRTAG_DA_READ_BUF,
2104 XFS_RANDOM_DA_READ_BUF))) {
2105 trace_xfs_da_btree_corrupt(rbp->bps[0], _RET_IP_);
2106 XFS_CORRUPTION_ERROR("xfs_da_do_buf(2)",
2107 XFS_ERRLEVEL_LOW, mp, info);
2108 error = XFS_ERROR(EFSCORRUPTED);
2109 xfs_da_brelse(trans, rbp);
2110 nbplist = 0;
2111 goto exit1;
2112 }
2113 }
2114 if (bplist) {
2115 kmem_free(bplist);
2116 }
2117 if (mapp != &map) {
2118 kmem_free(mapp);
2119 }
2120 if (bpp)
2121 *bpp = rbp;
2122 return 0;
2123 exit1:
2124 if (bplist) {
2125 for (i = 0; i < nbplist; i++)
2126 xfs_trans_brelse(trans, bplist[i]);
2127 kmem_free(bplist);
2128 }
2129 exit0:
2130 if (mapp != &map)
2131 kmem_free(mapp);
2132 if (bpp)
2133 *bpp = NULL;
2134 return error;
2135 }
2136
2137 /*
2138 * Get a buffer for the dir/attr block.
2139 */
2140 int
2141 xfs_da_get_buf(
2142 xfs_trans_t *trans,
2143 xfs_inode_t *dp,
2144 xfs_dablk_t bno,
2145 xfs_daddr_t mappedbno,
2146 xfs_dabuf_t **bpp,
2147 int whichfork)
2148 {
2149 return xfs_da_do_buf(trans, dp, bno, &mappedbno, bpp, whichfork, 0,
2150 (inst_t *)__return_address);
2151 }
2152
2153 /*
2154 * Get a buffer for the dir/attr block, fill in the contents.
2155 */
2156 int
2157 xfs_da_read_buf(
2158 xfs_trans_t *trans,
2159 xfs_inode_t *dp,
2160 xfs_dablk_t bno,
2161 xfs_daddr_t mappedbno,
2162 xfs_dabuf_t **bpp,
2163 int whichfork)
2164 {
2165 return xfs_da_do_buf(trans, dp, bno, &mappedbno, bpp, whichfork, 1,
2166 (inst_t *)__return_address);
2167 }
2168
2169 /*
2170 * Readahead the dir/attr block.
2171 */
2172 xfs_daddr_t
2173 xfs_da_reada_buf(
2174 xfs_trans_t *trans,
2175 xfs_inode_t *dp,
2176 xfs_dablk_t bno,
2177 int whichfork)
2178 {
2179 xfs_daddr_t rval;
2180
2181 rval = -1;
2182 if (xfs_da_do_buf(trans, dp, bno, &rval, NULL, whichfork, 3,
2183 (inst_t *)__return_address))
2184 return -1;
2185 else
2186 return rval;
2187 }
2188
2189 kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */
2190 kmem_zone_t *xfs_dabuf_zone; /* dabuf zone */
2191
2192 /*
2193 * Allocate a dir-state structure.
2194 * We don't put them on the stack since they're large.
2195 */
2196 xfs_da_state_t *
2197 xfs_da_state_alloc(void)
2198 {
2199 return kmem_zone_zalloc(xfs_da_state_zone, KM_NOFS);
2200 }
2201
2202 /*
2203 * Kill the altpath contents of a da-state structure.
2204 */
2205 STATIC void
2206 xfs_da_state_kill_altpath(xfs_da_state_t *state)
2207 {
2208 int i;
2209
2210 for (i = 0; i < state->altpath.active; i++) {
2211 if (state->altpath.blk[i].bp) {
2212 if (state->altpath.blk[i].bp != state->path.blk[i].bp)
2213 xfs_da_buf_done(state->altpath.blk[i].bp);
2214 state->altpath.blk[i].bp = NULL;
2215 }
2216 }
2217 state->altpath.active = 0;
2218 }
2219
2220 /*
2221 * Free a da-state structure.
2222 */
2223 void
2224 xfs_da_state_free(xfs_da_state_t *state)
2225 {
2226 int i;
2227
2228 xfs_da_state_kill_altpath(state);
2229 for (i = 0; i < state->path.active; i++) {
2230 if (state->path.blk[i].bp)
2231 xfs_da_buf_done(state->path.blk[i].bp);
2232 }
2233 if (state->extravalid && state->extrablk.bp)
2234 xfs_da_buf_done(state->extrablk.bp);
2235 #ifdef DEBUG
2236 memset((char *)state, 0, sizeof(*state));
2237 #endif /* DEBUG */
2238 kmem_zone_free(xfs_da_state_zone, state);
2239 }
2240
2241 #ifdef XFS_DABUF_DEBUG
2242 xfs_dabuf_t *xfs_dabuf_global_list;
2243 static DEFINE_SPINLOCK(xfs_dabuf_global_lock);
2244 #endif
2245
2246 /*
2247 * Create a dabuf.
2248 */
2249 /* ARGSUSED */
2250 STATIC xfs_dabuf_t *
2251 xfs_da_buf_make(int nbuf, xfs_buf_t **bps, inst_t *ra)
2252 {
2253 xfs_buf_t *bp;
2254 xfs_dabuf_t *dabuf;
2255 int i;
2256 int off;
2257
2258 if (nbuf == 1)
2259 dabuf = kmem_zone_alloc(xfs_dabuf_zone, KM_NOFS);
2260 else
2261 dabuf = kmem_alloc(XFS_DA_BUF_SIZE(nbuf), KM_NOFS);
2262 dabuf->dirty = 0;
2263 #ifdef XFS_DABUF_DEBUG
2264 dabuf->ra = ra;
2265 dabuf->target = XFS_BUF_TARGET(bps[0]);
2266 dabuf->blkno = XFS_BUF_ADDR(bps[0]);
2267 #endif
2268 if (nbuf == 1) {
2269 dabuf->nbuf = 1;
2270 bp = bps[0];
2271 dabuf->bbcount = (short)BTOBB(XFS_BUF_COUNT(bp));
2272 dabuf->data = XFS_BUF_PTR(bp);
2273 dabuf->bps[0] = bp;
2274 } else {
2275 dabuf->nbuf = nbuf;
2276 for (i = 0, dabuf->bbcount = 0; i < nbuf; i++) {
2277 dabuf->bps[i] = bp = bps[i];
2278 dabuf->bbcount += BTOBB(XFS_BUF_COUNT(bp));
2279 }
2280 dabuf->data = kmem_alloc(BBTOB(dabuf->bbcount), KM_SLEEP);
2281 for (i = off = 0; i < nbuf; i++, off += XFS_BUF_COUNT(bp)) {
2282 bp = bps[i];
2283 memcpy((char *)dabuf->data + off, XFS_BUF_PTR(bp),
2284 XFS_BUF_COUNT(bp));
2285 }
2286 }
2287 #ifdef XFS_DABUF_DEBUG
2288 {
2289 xfs_dabuf_t *p;
2290
2291 spin_lock(&xfs_dabuf_global_lock);
2292 for (p = xfs_dabuf_global_list; p; p = p->next) {
2293 ASSERT(p->blkno != dabuf->blkno ||
2294 p->target != dabuf->target);
2295 }
2296 dabuf->prev = NULL;
2297 if (xfs_dabuf_global_list)
2298 xfs_dabuf_global_list->prev = dabuf;
2299 dabuf->next = xfs_dabuf_global_list;
2300 xfs_dabuf_global_list = dabuf;
2301 spin_unlock(&xfs_dabuf_global_lock);
2302 }
2303 #endif
2304 return dabuf;
2305 }
2306
2307 /*
2308 * Un-dirty a dabuf.
2309 */
2310 STATIC void
2311 xfs_da_buf_clean(xfs_dabuf_t *dabuf)
2312 {
2313 xfs_buf_t *bp;
2314 int i;
2315 int off;
2316
2317 if (dabuf->dirty) {
2318 ASSERT(dabuf->nbuf > 1);
2319 dabuf->dirty = 0;
2320 for (i = off = 0; i < dabuf->nbuf;
2321 i++, off += XFS_BUF_COUNT(bp)) {
2322 bp = dabuf->bps[i];
2323 memcpy(XFS_BUF_PTR(bp), (char *)dabuf->data + off,
2324 XFS_BUF_COUNT(bp));
2325 }
2326 }
2327 }
2328
2329 /*
2330 * Release a dabuf.
2331 */
2332 void
2333 xfs_da_buf_done(xfs_dabuf_t *dabuf)
2334 {
2335 ASSERT(dabuf);
2336 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2337 if (dabuf->dirty)
2338 xfs_da_buf_clean(dabuf);
2339 if (dabuf->nbuf > 1)
2340 kmem_free(dabuf->data);
2341 #ifdef XFS_DABUF_DEBUG
2342 {
2343 spin_lock(&xfs_dabuf_global_lock);
2344 if (dabuf->prev)
2345 dabuf->prev->next = dabuf->next;
2346 else
2347 xfs_dabuf_global_list = dabuf->next;
2348 if (dabuf->next)
2349 dabuf->next->prev = dabuf->prev;
2350 spin_unlock(&xfs_dabuf_global_lock);
2351 }
2352 memset(dabuf, 0, XFS_DA_BUF_SIZE(dabuf->nbuf));
2353 #endif
2354 if (dabuf->nbuf == 1)
2355 kmem_zone_free(xfs_dabuf_zone, dabuf);
2356 else
2357 kmem_free(dabuf);
2358 }
2359
2360 /*
2361 * Log transaction from a dabuf.
2362 */
2363 void
2364 xfs_da_log_buf(xfs_trans_t *tp, xfs_dabuf_t *dabuf, uint first, uint last)
2365 {
2366 xfs_buf_t *bp;
2367 uint f;
2368 int i;
2369 uint l;
2370 int off;
2371
2372 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2373 if (dabuf->nbuf == 1) {
2374 ASSERT(dabuf->data == (void *)XFS_BUF_PTR(dabuf->bps[0]));
2375 xfs_trans_log_buf(tp, dabuf->bps[0], first, last);
2376 return;
2377 }
2378 dabuf->dirty = 1;
2379 ASSERT(first <= last);
2380 for (i = off = 0; i < dabuf->nbuf; i++, off += XFS_BUF_COUNT(bp)) {
2381 bp = dabuf->bps[i];
2382 f = off;
2383 l = f + XFS_BUF_COUNT(bp) - 1;
2384 if (f < first)
2385 f = first;
2386 if (l > last)
2387 l = last;
2388 if (f <= l)
2389 xfs_trans_log_buf(tp, bp, f - off, l - off);
2390 /*
2391 * B_DONE is set by xfs_trans_log buf.
2392 * If we don't set it on a new buffer (get not read)
2393 * then if we don't put anything in the buffer it won't
2394 * be set, and at commit it it released into the cache,
2395 * and then a read will fail.
2396 */
2397 else if (!(XFS_BUF_ISDONE(bp)))
2398 XFS_BUF_DONE(bp);
2399 }
2400 ASSERT(last < off);
2401 }
2402
2403 /*
2404 * Release dabuf from a transaction.
2405 * Have to free up the dabuf before the buffers are released,
2406 * since the synchronization on the dabuf is really the lock on the buffer.
2407 */
2408 void
2409 xfs_da_brelse(xfs_trans_t *tp, xfs_dabuf_t *dabuf)
2410 {
2411 xfs_buf_t *bp;
2412 xfs_buf_t **bplist;
2413 int i;
2414 int nbuf;
2415
2416 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2417 if ((nbuf = dabuf->nbuf) == 1) {
2418 bplist = &bp;
2419 bp = dabuf->bps[0];
2420 } else {
2421 bplist = kmem_alloc(nbuf * sizeof(*bplist), KM_SLEEP);
2422 memcpy(bplist, dabuf->bps, nbuf * sizeof(*bplist));
2423 }
2424 xfs_da_buf_done(dabuf);
2425 for (i = 0; i < nbuf; i++)
2426 xfs_trans_brelse(tp, bplist[i]);
2427 if (bplist != &bp)
2428 kmem_free(bplist);
2429 }
2430
2431 /*
2432 * Invalidate dabuf from a transaction.
2433 */
2434 void
2435 xfs_da_binval(xfs_trans_t *tp, xfs_dabuf_t *dabuf)
2436 {
2437 xfs_buf_t *bp;
2438 xfs_buf_t **bplist;
2439 int i;
2440 int nbuf;
2441
2442 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2443 if ((nbuf = dabuf->nbuf) == 1) {
2444 bplist = &bp;
2445 bp = dabuf->bps[0];
2446 } else {
2447 bplist = kmem_alloc(nbuf * sizeof(*bplist), KM_SLEEP);
2448 memcpy(bplist, dabuf->bps, nbuf * sizeof(*bplist));
2449 }
2450 xfs_da_buf_done(dabuf);
2451 for (i = 0; i < nbuf; i++)
2452 xfs_trans_binval(tp, bplist[i]);
2453 if (bplist != &bp)
2454 kmem_free(bplist);
2455 }
2456
2457 /*
2458 * Get the first daddr from a dabuf.
2459 */
2460 xfs_daddr_t
2461 xfs_da_blkno(xfs_dabuf_t *dabuf)
2462 {
2463 ASSERT(dabuf->nbuf);
2464 ASSERT(dabuf->data);
2465 return XFS_BUF_ADDR(dabuf->bps[0]);
2466 }
kernel source for telekom puls (alcatel/mediatek)