Loading...
1/*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
4 * All Rights Reserved.
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 */
19#include "xfs.h"
20#include "xfs_fs.h"
21#include "xfs_shared.h"
22#include "xfs_format.h"
23#include "xfs_log_format.h"
24#include "xfs_trans_resv.h"
25#include "xfs_bit.h"
26#include "xfs_mount.h"
27#include "xfs_da_format.h"
28#include "xfs_da_btree.h"
29#include "xfs_dir2.h"
30#include "xfs_dir2_priv.h"
31#include "xfs_inode.h"
32#include "xfs_trans.h"
33#include "xfs_inode_item.h"
34#include "xfs_alloc.h"
35#include "xfs_bmap.h"
36#include "xfs_attr.h"
37#include "xfs_attr_leaf.h"
38#include "xfs_error.h"
39#include "xfs_trace.h"
40#include "xfs_cksum.h"
41#include "xfs_buf_item.h"
42#include "xfs_log.h"
43
44/*
45 * xfs_da_btree.c
46 *
47 * Routines to implement directories as Btrees of hashed names.
48 */
49
50/*========================================================================
51 * Function prototypes for the kernel.
52 *========================================================================*/
53
54/*
55 * Routines used for growing the Btree.
56 */
57STATIC int xfs_da3_root_split(xfs_da_state_t *state,
58 xfs_da_state_blk_t *existing_root,
59 xfs_da_state_blk_t *new_child);
60STATIC int xfs_da3_node_split(xfs_da_state_t *state,
61 xfs_da_state_blk_t *existing_blk,
62 xfs_da_state_blk_t *split_blk,
63 xfs_da_state_blk_t *blk_to_add,
64 int treelevel,
65 int *result);
66STATIC void xfs_da3_node_rebalance(xfs_da_state_t *state,
67 xfs_da_state_blk_t *node_blk_1,
68 xfs_da_state_blk_t *node_blk_2);
69STATIC void xfs_da3_node_add(xfs_da_state_t *state,
70 xfs_da_state_blk_t *old_node_blk,
71 xfs_da_state_blk_t *new_node_blk);
72
73/*
74 * Routines used for shrinking the Btree.
75 */
76STATIC int xfs_da3_root_join(xfs_da_state_t *state,
77 xfs_da_state_blk_t *root_blk);
78STATIC int xfs_da3_node_toosmall(xfs_da_state_t *state, int *retval);
79STATIC void xfs_da3_node_remove(xfs_da_state_t *state,
80 xfs_da_state_blk_t *drop_blk);
81STATIC void xfs_da3_node_unbalance(xfs_da_state_t *state,
82 xfs_da_state_blk_t *src_node_blk,
83 xfs_da_state_blk_t *dst_node_blk);
84
85/*
86 * Utility routines.
87 */
88STATIC int xfs_da3_blk_unlink(xfs_da_state_t *state,
89 xfs_da_state_blk_t *drop_blk,
90 xfs_da_state_blk_t *save_blk);
91
92
93kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */
94
95/*
96 * Allocate a dir-state structure.
97 * We don't put them on the stack since they're large.
98 */
99xfs_da_state_t *
100xfs_da_state_alloc(void)
101{
102 return kmem_zone_zalloc(xfs_da_state_zone, KM_NOFS);
103}
104
105/*
106 * Kill the altpath contents of a da-state structure.
107 */
108STATIC void
109xfs_da_state_kill_altpath(xfs_da_state_t *state)
110{
111 int i;
112
113 for (i = 0; i < state->altpath.active; i++)
114 state->altpath.blk[i].bp = NULL;
115 state->altpath.active = 0;
116}
117
118/*
119 * Free a da-state structure.
120 */
121void
122xfs_da_state_free(xfs_da_state_t *state)
123{
124 xfs_da_state_kill_altpath(state);
125#ifdef DEBUG
126 memset((char *)state, 0, sizeof(*state));
127#endif /* DEBUG */
128 kmem_zone_free(xfs_da_state_zone, state);
129}
130
131static xfs_failaddr_t
132xfs_da3_node_verify(
133 struct xfs_buf *bp)
134{
135 struct xfs_mount *mp = bp->b_target->bt_mount;
136 struct xfs_da_intnode *hdr = bp->b_addr;
137 struct xfs_da3_icnode_hdr ichdr;
138 const struct xfs_dir_ops *ops;
139
140 ops = xfs_dir_get_ops(mp, NULL);
141
142 ops->node_hdr_from_disk(&ichdr, hdr);
143
144 if (xfs_sb_version_hascrc(&mp->m_sb)) {
145 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
146
147 if (ichdr.magic != XFS_DA3_NODE_MAGIC)
148 return __this_address;
149
150 if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid))
151 return __this_address;
152 if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
153 return __this_address;
154 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->info.lsn)))
155 return __this_address;
156 } else {
157 if (ichdr.magic != XFS_DA_NODE_MAGIC)
158 return __this_address;
159 }
160 if (ichdr.level == 0)
161 return __this_address;
162 if (ichdr.level > XFS_DA_NODE_MAXDEPTH)
163 return __this_address;
164 if (ichdr.count == 0)
165 return __this_address;
166
167 /*
168 * we don't know if the node is for and attribute or directory tree,
169 * so only fail if the count is outside both bounds
170 */
171 if (ichdr.count > mp->m_dir_geo->node_ents &&
172 ichdr.count > mp->m_attr_geo->node_ents)
173 return __this_address;
174
175 /* XXX: hash order check? */
176
177 return NULL;
178}
179
180static void
181xfs_da3_node_write_verify(
182 struct xfs_buf *bp)
183{
184 struct xfs_mount *mp = bp->b_target->bt_mount;
185 struct xfs_buf_log_item *bip = bp->b_log_item;
186 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
187 xfs_failaddr_t fa;
188
189 fa = xfs_da3_node_verify(bp);
190 if (fa) {
191 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
192 return;
193 }
194
195 if (!xfs_sb_version_hascrc(&mp->m_sb))
196 return;
197
198 if (bip)
199 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
200
201 xfs_buf_update_cksum(bp, XFS_DA3_NODE_CRC_OFF);
202}
203
204/*
205 * leaf/node format detection on trees is sketchy, so a node read can be done on
206 * leaf level blocks when detection identifies the tree as a node format tree
207 * incorrectly. In this case, we need to swap the verifier to match the correct
208 * format of the block being read.
209 */
210static void
211xfs_da3_node_read_verify(
212 struct xfs_buf *bp)
213{
214 struct xfs_da_blkinfo *info = bp->b_addr;
215 xfs_failaddr_t fa;
216
217 switch (be16_to_cpu(info->magic)) {
218 case XFS_DA3_NODE_MAGIC:
219 if (!xfs_buf_verify_cksum(bp, XFS_DA3_NODE_CRC_OFF)) {
220 xfs_verifier_error(bp, -EFSBADCRC,
221 __this_address);
222 break;
223 }
224 /* fall through */
225 case XFS_DA_NODE_MAGIC:
226 fa = xfs_da3_node_verify(bp);
227 if (fa)
228 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
229 return;
230 case XFS_ATTR_LEAF_MAGIC:
231 case XFS_ATTR3_LEAF_MAGIC:
232 bp->b_ops = &xfs_attr3_leaf_buf_ops;
233 bp->b_ops->verify_read(bp);
234 return;
235 case XFS_DIR2_LEAFN_MAGIC:
236 case XFS_DIR3_LEAFN_MAGIC:
237 bp->b_ops = &xfs_dir3_leafn_buf_ops;
238 bp->b_ops->verify_read(bp);
239 return;
240 default:
241 xfs_verifier_error(bp, -EFSCORRUPTED, __this_address);
242 break;
243 }
244}
245
246/* Verify the structure of a da3 block. */
247static xfs_failaddr_t
248xfs_da3_node_verify_struct(
249 struct xfs_buf *bp)
250{
251 struct xfs_da_blkinfo *info = bp->b_addr;
252
253 switch (be16_to_cpu(info->magic)) {
254 case XFS_DA3_NODE_MAGIC:
255 case XFS_DA_NODE_MAGIC:
256 return xfs_da3_node_verify(bp);
257 case XFS_ATTR_LEAF_MAGIC:
258 case XFS_ATTR3_LEAF_MAGIC:
259 bp->b_ops = &xfs_attr3_leaf_buf_ops;
260 return bp->b_ops->verify_struct(bp);
261 case XFS_DIR2_LEAFN_MAGIC:
262 case XFS_DIR3_LEAFN_MAGIC:
263 bp->b_ops = &xfs_dir3_leafn_buf_ops;
264 return bp->b_ops->verify_struct(bp);
265 default:
266 return __this_address;
267 }
268}
269
270const struct xfs_buf_ops xfs_da3_node_buf_ops = {
271 .name = "xfs_da3_node",
272 .verify_read = xfs_da3_node_read_verify,
273 .verify_write = xfs_da3_node_write_verify,
274 .verify_struct = xfs_da3_node_verify_struct,
275};
276
277int
278xfs_da3_node_read(
279 struct xfs_trans *tp,
280 struct xfs_inode *dp,
281 xfs_dablk_t bno,
282 xfs_daddr_t mappedbno,
283 struct xfs_buf **bpp,
284 int which_fork)
285{
286 int err;
287
288 err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
289 which_fork, &xfs_da3_node_buf_ops);
290 if (!err && tp && *bpp) {
291 struct xfs_da_blkinfo *info = (*bpp)->b_addr;
292 int type;
293
294 switch (be16_to_cpu(info->magic)) {
295 case XFS_DA_NODE_MAGIC:
296 case XFS_DA3_NODE_MAGIC:
297 type = XFS_BLFT_DA_NODE_BUF;
298 break;
299 case XFS_ATTR_LEAF_MAGIC:
300 case XFS_ATTR3_LEAF_MAGIC:
301 type = XFS_BLFT_ATTR_LEAF_BUF;
302 break;
303 case XFS_DIR2_LEAFN_MAGIC:
304 case XFS_DIR3_LEAFN_MAGIC:
305 type = XFS_BLFT_DIR_LEAFN_BUF;
306 break;
307 default:
308 type = 0;
309 ASSERT(0);
310 break;
311 }
312 xfs_trans_buf_set_type(tp, *bpp, type);
313 }
314 return err;
315}
316
317/*========================================================================
318 * Routines used for growing the Btree.
319 *========================================================================*/
320
321/*
322 * Create the initial contents of an intermediate node.
323 */
324int
325xfs_da3_node_create(
326 struct xfs_da_args *args,
327 xfs_dablk_t blkno,
328 int level,
329 struct xfs_buf **bpp,
330 int whichfork)
331{
332 struct xfs_da_intnode *node;
333 struct xfs_trans *tp = args->trans;
334 struct xfs_mount *mp = tp->t_mountp;
335 struct xfs_da3_icnode_hdr ichdr = {0};
336 struct xfs_buf *bp;
337 int error;
338 struct xfs_inode *dp = args->dp;
339
340 trace_xfs_da_node_create(args);
341 ASSERT(level <= XFS_DA_NODE_MAXDEPTH);
342
343 error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, whichfork);
344 if (error)
345 return error;
346 bp->b_ops = &xfs_da3_node_buf_ops;
347 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
348 node = bp->b_addr;
349
350 if (xfs_sb_version_hascrc(&mp->m_sb)) {
351 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
352
353 memset(hdr3, 0, sizeof(struct xfs_da3_node_hdr));
354 ichdr.magic = XFS_DA3_NODE_MAGIC;
355 hdr3->info.blkno = cpu_to_be64(bp->b_bn);
356 hdr3->info.owner = cpu_to_be64(args->dp->i_ino);
357 uuid_copy(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid);
358 } else {
359 ichdr.magic = XFS_DA_NODE_MAGIC;
360 }
361 ichdr.level = level;
362
363 dp->d_ops->node_hdr_to_disk(node, &ichdr);
364 xfs_trans_log_buf(tp, bp,
365 XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
366
367 *bpp = bp;
368 return 0;
369}
370
371/*
372 * Split a leaf node, rebalance, then possibly split
373 * intermediate nodes, rebalance, etc.
374 */
375int /* error */
376xfs_da3_split(
377 struct xfs_da_state *state)
378{
379 struct xfs_da_state_blk *oldblk;
380 struct xfs_da_state_blk *newblk;
381 struct xfs_da_state_blk *addblk;
382 struct xfs_da_intnode *node;
383 int max;
384 int action = 0;
385 int error;
386 int i;
387
388 trace_xfs_da_split(state->args);
389
390 /*
391 * Walk back up the tree splitting/inserting/adjusting as necessary.
392 * If we need to insert and there isn't room, split the node, then
393 * decide which fragment to insert the new block from below into.
394 * Note that we may split the root this way, but we need more fixup.
395 */
396 max = state->path.active - 1;
397 ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH));
398 ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC ||
399 state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
400
401 addblk = &state->path.blk[max]; /* initial dummy value */
402 for (i = max; (i >= 0) && addblk; state->path.active--, i--) {
403 oldblk = &state->path.blk[i];
404 newblk = &state->altpath.blk[i];
405
406 /*
407 * If a leaf node then
408 * Allocate a new leaf node, then rebalance across them.
409 * else if an intermediate node then
410 * We split on the last layer, must we split the node?
411 */
412 switch (oldblk->magic) {
413 case XFS_ATTR_LEAF_MAGIC:
414 error = xfs_attr3_leaf_split(state, oldblk, newblk);
415 if ((error != 0) && (error != -ENOSPC)) {
416 return error; /* GROT: attr is inconsistent */
417 }
418 if (!error) {
419 addblk = newblk;
420 break;
421 }
422 /*
423 * Entry wouldn't fit, split the leaf again. The new
424 * extrablk will be consumed by xfs_da3_node_split if
425 * the node is split.
426 */
427 state->extravalid = 1;
428 if (state->inleaf) {
429 state->extraafter = 0; /* before newblk */
430 trace_xfs_attr_leaf_split_before(state->args);
431 error = xfs_attr3_leaf_split(state, oldblk,
432 &state->extrablk);
433 } else {
434 state->extraafter = 1; /* after newblk */
435 trace_xfs_attr_leaf_split_after(state->args);
436 error = xfs_attr3_leaf_split(state, newblk,
437 &state->extrablk);
438 }
439 if (error)
440 return error; /* GROT: attr inconsistent */
441 addblk = newblk;
442 break;
443 case XFS_DIR2_LEAFN_MAGIC:
444 error = xfs_dir2_leafn_split(state, oldblk, newblk);
445 if (error)
446 return error;
447 addblk = newblk;
448 break;
449 case XFS_DA_NODE_MAGIC:
450 error = xfs_da3_node_split(state, oldblk, newblk, addblk,
451 max - i, &action);
452 addblk->bp = NULL;
453 if (error)
454 return error; /* GROT: dir is inconsistent */
455 /*
456 * Record the newly split block for the next time thru?
457 */
458 if (action)
459 addblk = newblk;
460 else
461 addblk = NULL;
462 break;
463 }
464
465 /*
466 * Update the btree to show the new hashval for this child.
467 */
468 xfs_da3_fixhashpath(state, &state->path);
469 }
470 if (!addblk)
471 return 0;
472
473 /*
474 * xfs_da3_node_split() should have consumed any extra blocks we added
475 * during a double leaf split in the attr fork. This is guaranteed as
476 * we can't be here if the attr fork only has a single leaf block.
477 */
478 ASSERT(state->extravalid == 0 ||
479 state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
480
481 /*
482 * Split the root node.
483 */
484 ASSERT(state->path.active == 0);
485 oldblk = &state->path.blk[0];
486 error = xfs_da3_root_split(state, oldblk, addblk);
487 if (error) {
488 addblk->bp = NULL;
489 return error; /* GROT: dir is inconsistent */
490 }
491
492 /*
493 * Update pointers to the node which used to be block 0 and just got
494 * bumped because of the addition of a new root node. Note that the
495 * original block 0 could be at any position in the list of blocks in
496 * the tree.
497 *
498 * Note: the magic numbers and sibling pointers are in the same physical
499 * place for both v2 and v3 headers (by design). Hence it doesn't matter
500 * which version of the xfs_da_intnode structure we use here as the
501 * result will be the same using either structure.
502 */
503 node = oldblk->bp->b_addr;
504 if (node->hdr.info.forw) {
505 ASSERT(be32_to_cpu(node->hdr.info.forw) == addblk->blkno);
506 node = addblk->bp->b_addr;
507 node->hdr.info.back = cpu_to_be32(oldblk->blkno);
508 xfs_trans_log_buf(state->args->trans, addblk->bp,
509 XFS_DA_LOGRANGE(node, &node->hdr.info,
510 sizeof(node->hdr.info)));
511 }
512 node = oldblk->bp->b_addr;
513 if (node->hdr.info.back) {
514 ASSERT(be32_to_cpu(node->hdr.info.back) == addblk->blkno);
515 node = addblk->bp->b_addr;
516 node->hdr.info.forw = cpu_to_be32(oldblk->blkno);
517 xfs_trans_log_buf(state->args->trans, addblk->bp,
518 XFS_DA_LOGRANGE(node, &node->hdr.info,
519 sizeof(node->hdr.info)));
520 }
521 addblk->bp = NULL;
522 return 0;
523}
524
525/*
526 * Split the root. We have to create a new root and point to the two
527 * parts (the split old root) that we just created. Copy block zero to
528 * the EOF, extending the inode in process.
529 */
530STATIC int /* error */
531xfs_da3_root_split(
532 struct xfs_da_state *state,
533 struct xfs_da_state_blk *blk1,
534 struct xfs_da_state_blk *blk2)
535{
536 struct xfs_da_intnode *node;
537 struct xfs_da_intnode *oldroot;
538 struct xfs_da_node_entry *btree;
539 struct xfs_da3_icnode_hdr nodehdr;
540 struct xfs_da_args *args;
541 struct xfs_buf *bp;
542 struct xfs_inode *dp;
543 struct xfs_trans *tp;
544 struct xfs_dir2_leaf *leaf;
545 xfs_dablk_t blkno;
546 int level;
547 int error;
548 int size;
549
550 trace_xfs_da_root_split(state->args);
551
552 /*
553 * Copy the existing (incorrect) block from the root node position
554 * to a free space somewhere.
555 */
556 args = state->args;
557 error = xfs_da_grow_inode(args, &blkno);
558 if (error)
559 return error;
560
561 dp = args->dp;
562 tp = args->trans;
563 error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork);
564 if (error)
565 return error;
566 node = bp->b_addr;
567 oldroot = blk1->bp->b_addr;
568 if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
569 oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC)) {
570 struct xfs_da3_icnode_hdr icnodehdr;
571
572 dp->d_ops->node_hdr_from_disk(&icnodehdr, oldroot);
573 btree = dp->d_ops->node_tree_p(oldroot);
574 size = (int)((char *)&btree[icnodehdr.count] - (char *)oldroot);
575 level = icnodehdr.level;
576
577 /*
578 * we are about to copy oldroot to bp, so set up the type
579 * of bp while we know exactly what it will be.
580 */
581 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
582 } else {
583 struct xfs_dir3_icleaf_hdr leafhdr;
584 struct xfs_dir2_leaf_entry *ents;
585
586 leaf = (xfs_dir2_leaf_t *)oldroot;
587 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
588 ents = dp->d_ops->leaf_ents_p(leaf);
589
590 ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
591 leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
592 size = (int)((char *)&ents[leafhdr.count] - (char *)leaf);
593 level = 0;
594
595 /*
596 * we are about to copy oldroot to bp, so set up the type
597 * of bp while we know exactly what it will be.
598 */
599 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
600 }
601
602 /*
603 * we can copy most of the information in the node from one block to
604 * another, but for CRC enabled headers we have to make sure that the
605 * block specific identifiers are kept intact. We update the buffer
606 * directly for this.
607 */
608 memcpy(node, oldroot, size);
609 if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
610 oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
611 struct xfs_da3_intnode *node3 = (struct xfs_da3_intnode *)node;
612
613 node3->hdr.info.blkno = cpu_to_be64(bp->b_bn);
614 }
615 xfs_trans_log_buf(tp, bp, 0, size - 1);
616
617 bp->b_ops = blk1->bp->b_ops;
618 xfs_trans_buf_copy_type(bp, blk1->bp);
619 blk1->bp = bp;
620 blk1->blkno = blkno;
621
622 /*
623 * Set up the new root node.
624 */
625 error = xfs_da3_node_create(args,
626 (args->whichfork == XFS_DATA_FORK) ? args->geo->leafblk : 0,
627 level + 1, &bp, args->whichfork);
628 if (error)
629 return error;
630
631 node = bp->b_addr;
632 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
633 btree = dp->d_ops->node_tree_p(node);
634 btree[0].hashval = cpu_to_be32(blk1->hashval);
635 btree[0].before = cpu_to_be32(blk1->blkno);
636 btree[1].hashval = cpu_to_be32(blk2->hashval);
637 btree[1].before = cpu_to_be32(blk2->blkno);
638 nodehdr.count = 2;
639 dp->d_ops->node_hdr_to_disk(node, &nodehdr);
640
641#ifdef DEBUG
642 if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
643 oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
644 ASSERT(blk1->blkno >= args->geo->leafblk &&
645 blk1->blkno < args->geo->freeblk);
646 ASSERT(blk2->blkno >= args->geo->leafblk &&
647 blk2->blkno < args->geo->freeblk);
648 }
649#endif
650
651 /* Header is already logged by xfs_da_node_create */
652 xfs_trans_log_buf(tp, bp,
653 XFS_DA_LOGRANGE(node, btree, sizeof(xfs_da_node_entry_t) * 2));
654
655 return 0;
656}
657
658/*
659 * Split the node, rebalance, then add the new entry.
660 */
661STATIC int /* error */
662xfs_da3_node_split(
663 struct xfs_da_state *state,
664 struct xfs_da_state_blk *oldblk,
665 struct xfs_da_state_blk *newblk,
666 struct xfs_da_state_blk *addblk,
667 int treelevel,
668 int *result)
669{
670 struct xfs_da_intnode *node;
671 struct xfs_da3_icnode_hdr nodehdr;
672 xfs_dablk_t blkno;
673 int newcount;
674 int error;
675 int useextra;
676 struct xfs_inode *dp = state->args->dp;
677
678 trace_xfs_da_node_split(state->args);
679
680 node = oldblk->bp->b_addr;
681 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
682
683 /*
684 * With V2 dirs the extra block is data or freespace.
685 */
686 useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK;
687 newcount = 1 + useextra;
688 /*
689 * Do we have to split the node?
690 */
691 if (nodehdr.count + newcount > state->args->geo->node_ents) {
692 /*
693 * Allocate a new node, add to the doubly linked chain of
694 * nodes, then move some of our excess entries into it.
695 */
696 error = xfs_da_grow_inode(state->args, &blkno);
697 if (error)
698 return error; /* GROT: dir is inconsistent */
699
700 error = xfs_da3_node_create(state->args, blkno, treelevel,
701 &newblk->bp, state->args->whichfork);
702 if (error)
703 return error; /* GROT: dir is inconsistent */
704 newblk->blkno = blkno;
705 newblk->magic = XFS_DA_NODE_MAGIC;
706 xfs_da3_node_rebalance(state, oldblk, newblk);
707 error = xfs_da3_blk_link(state, oldblk, newblk);
708 if (error)
709 return error;
710 *result = 1;
711 } else {
712 *result = 0;
713 }
714
715 /*
716 * Insert the new entry(s) into the correct block
717 * (updating last hashval in the process).
718 *
719 * xfs_da3_node_add() inserts BEFORE the given index,
720 * and as a result of using node_lookup_int() we always
721 * point to a valid entry (not after one), but a split
722 * operation always results in a new block whose hashvals
723 * FOLLOW the current block.
724 *
725 * If we had double-split op below us, then add the extra block too.
726 */
727 node = oldblk->bp->b_addr;
728 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
729 if (oldblk->index <= nodehdr.count) {
730 oldblk->index++;
731 xfs_da3_node_add(state, oldblk, addblk);
732 if (useextra) {
733 if (state->extraafter)
734 oldblk->index++;
735 xfs_da3_node_add(state, oldblk, &state->extrablk);
736 state->extravalid = 0;
737 }
738 } else {
739 newblk->index++;
740 xfs_da3_node_add(state, newblk, addblk);
741 if (useextra) {
742 if (state->extraafter)
743 newblk->index++;
744 xfs_da3_node_add(state, newblk, &state->extrablk);
745 state->extravalid = 0;
746 }
747 }
748
749 return 0;
750}
751
752/*
753 * Balance the btree elements between two intermediate nodes,
754 * usually one full and one empty.
755 *
756 * NOTE: if blk2 is empty, then it will get the upper half of blk1.
757 */
758STATIC void
759xfs_da3_node_rebalance(
760 struct xfs_da_state *state,
761 struct xfs_da_state_blk *blk1,
762 struct xfs_da_state_blk *blk2)
763{
764 struct xfs_da_intnode *node1;
765 struct xfs_da_intnode *node2;
766 struct xfs_da_intnode *tmpnode;
767 struct xfs_da_node_entry *btree1;
768 struct xfs_da_node_entry *btree2;
769 struct xfs_da_node_entry *btree_s;
770 struct xfs_da_node_entry *btree_d;
771 struct xfs_da3_icnode_hdr nodehdr1;
772 struct xfs_da3_icnode_hdr nodehdr2;
773 struct xfs_trans *tp;
774 int count;
775 int tmp;
776 int swap = 0;
777 struct xfs_inode *dp = state->args->dp;
778
779 trace_xfs_da_node_rebalance(state->args);
780
781 node1 = blk1->bp->b_addr;
782 node2 = blk2->bp->b_addr;
783 dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
784 dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
785 btree1 = dp->d_ops->node_tree_p(node1);
786 btree2 = dp->d_ops->node_tree_p(node2);
787
788 /*
789 * Figure out how many entries need to move, and in which direction.
790 * Swap the nodes around if that makes it simpler.
791 */
792 if (nodehdr1.count > 0 && nodehdr2.count > 0 &&
793 ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
794 (be32_to_cpu(btree2[nodehdr2.count - 1].hashval) <
795 be32_to_cpu(btree1[nodehdr1.count - 1].hashval)))) {
796 tmpnode = node1;
797 node1 = node2;
798 node2 = tmpnode;
799 dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
800 dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
801 btree1 = dp->d_ops->node_tree_p(node1);
802 btree2 = dp->d_ops->node_tree_p(node2);
803 swap = 1;
804 }
805
806 count = (nodehdr1.count - nodehdr2.count) / 2;
807 if (count == 0)
808 return;
809 tp = state->args->trans;
810 /*
811 * Two cases: high-to-low and low-to-high.
812 */
813 if (count > 0) {
814 /*
815 * Move elements in node2 up to make a hole.
816 */
817 tmp = nodehdr2.count;
818 if (tmp > 0) {
819 tmp *= (uint)sizeof(xfs_da_node_entry_t);
820 btree_s = &btree2[0];
821 btree_d = &btree2[count];
822 memmove(btree_d, btree_s, tmp);
823 }
824
825 /*
826 * Move the req'd B-tree elements from high in node1 to
827 * low in node2.
828 */
829 nodehdr2.count += count;
830 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
831 btree_s = &btree1[nodehdr1.count - count];
832 btree_d = &btree2[0];
833 memcpy(btree_d, btree_s, tmp);
834 nodehdr1.count -= count;
835 } else {
836 /*
837 * Move the req'd B-tree elements from low in node2 to
838 * high in node1.
839 */
840 count = -count;
841 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
842 btree_s = &btree2[0];
843 btree_d = &btree1[nodehdr1.count];
844 memcpy(btree_d, btree_s, tmp);
845 nodehdr1.count += count;
846
847 xfs_trans_log_buf(tp, blk1->bp,
848 XFS_DA_LOGRANGE(node1, btree_d, tmp));
849
850 /*
851 * Move elements in node2 down to fill the hole.
852 */
853 tmp = nodehdr2.count - count;
854 tmp *= (uint)sizeof(xfs_da_node_entry_t);
855 btree_s = &btree2[count];
856 btree_d = &btree2[0];
857 memmove(btree_d, btree_s, tmp);
858 nodehdr2.count -= count;
859 }
860
861 /*
862 * Log header of node 1 and all current bits of node 2.
863 */
864 dp->d_ops->node_hdr_to_disk(node1, &nodehdr1);
865 xfs_trans_log_buf(tp, blk1->bp,
866 XFS_DA_LOGRANGE(node1, &node1->hdr, dp->d_ops->node_hdr_size));
867
868 dp->d_ops->node_hdr_to_disk(node2, &nodehdr2);
869 xfs_trans_log_buf(tp, blk2->bp,
870 XFS_DA_LOGRANGE(node2, &node2->hdr,
871 dp->d_ops->node_hdr_size +
872 (sizeof(btree2[0]) * nodehdr2.count)));
873
874 /*
875 * Record the last hashval from each block for upward propagation.
876 * (note: don't use the swapped node pointers)
877 */
878 if (swap) {
879 node1 = blk1->bp->b_addr;
880 node2 = blk2->bp->b_addr;
881 dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
882 dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
883 btree1 = dp->d_ops->node_tree_p(node1);
884 btree2 = dp->d_ops->node_tree_p(node2);
885 }
886 blk1->hashval = be32_to_cpu(btree1[nodehdr1.count - 1].hashval);
887 blk2->hashval = be32_to_cpu(btree2[nodehdr2.count - 1].hashval);
888
889 /*
890 * Adjust the expected index for insertion.
891 */
892 if (blk1->index >= nodehdr1.count) {
893 blk2->index = blk1->index - nodehdr1.count;
894 blk1->index = nodehdr1.count + 1; /* make it invalid */
895 }
896}
897
898/*
899 * Add a new entry to an intermediate node.
900 */
901STATIC void
902xfs_da3_node_add(
903 struct xfs_da_state *state,
904 struct xfs_da_state_blk *oldblk,
905 struct xfs_da_state_blk *newblk)
906{
907 struct xfs_da_intnode *node;
908 struct xfs_da3_icnode_hdr nodehdr;
909 struct xfs_da_node_entry *btree;
910 int tmp;
911 struct xfs_inode *dp = state->args->dp;
912
913 trace_xfs_da_node_add(state->args);
914
915 node = oldblk->bp->b_addr;
916 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
917 btree = dp->d_ops->node_tree_p(node);
918
919 ASSERT(oldblk->index >= 0 && oldblk->index <= nodehdr.count);
920 ASSERT(newblk->blkno != 0);
921 if (state->args->whichfork == XFS_DATA_FORK)
922 ASSERT(newblk->blkno >= state->args->geo->leafblk &&
923 newblk->blkno < state->args->geo->freeblk);
924
925 /*
926 * We may need to make some room before we insert the new node.
927 */
928 tmp = 0;
929 if (oldblk->index < nodehdr.count) {
930 tmp = (nodehdr.count - oldblk->index) * (uint)sizeof(*btree);
931 memmove(&btree[oldblk->index + 1], &btree[oldblk->index], tmp);
932 }
933 btree[oldblk->index].hashval = cpu_to_be32(newblk->hashval);
934 btree[oldblk->index].before = cpu_to_be32(newblk->blkno);
935 xfs_trans_log_buf(state->args->trans, oldblk->bp,
936 XFS_DA_LOGRANGE(node, &btree[oldblk->index],
937 tmp + sizeof(*btree)));
938
939 nodehdr.count += 1;
940 dp->d_ops->node_hdr_to_disk(node, &nodehdr);
941 xfs_trans_log_buf(state->args->trans, oldblk->bp,
942 XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
943
944 /*
945 * Copy the last hash value from the oldblk to propagate upwards.
946 */
947 oldblk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
948}
949
950/*========================================================================
951 * Routines used for shrinking the Btree.
952 *========================================================================*/
953
954/*
955 * Deallocate an empty leaf node, remove it from its parent,
956 * possibly deallocating that block, etc...
957 */
958int
959xfs_da3_join(
960 struct xfs_da_state *state)
961{
962 struct xfs_da_state_blk *drop_blk;
963 struct xfs_da_state_blk *save_blk;
964 int action = 0;
965 int error;
966
967 trace_xfs_da_join(state->args);
968
969 drop_blk = &state->path.blk[ state->path.active-1 ];
970 save_blk = &state->altpath.blk[ state->path.active-1 ];
971 ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
972 ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
973 drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
974
975 /*
976 * Walk back up the tree joining/deallocating as necessary.
977 * When we stop dropping blocks, break out.
978 */
979 for ( ; state->path.active >= 2; drop_blk--, save_blk--,
980 state->path.active--) {
981 /*
982 * See if we can combine the block with a neighbor.
983 * (action == 0) => no options, just leave
984 * (action == 1) => coalesce, then unlink
985 * (action == 2) => block empty, unlink it
986 */
987 switch (drop_blk->magic) {
988 case XFS_ATTR_LEAF_MAGIC:
989 error = xfs_attr3_leaf_toosmall(state, &action);
990 if (error)
991 return error;
992 if (action == 0)
993 return 0;
994 xfs_attr3_leaf_unbalance(state, drop_blk, save_blk);
995 break;
996 case XFS_DIR2_LEAFN_MAGIC:
997 error = xfs_dir2_leafn_toosmall(state, &action);
998 if (error)
999 return error;
1000 if (action == 0)
1001 return 0;
1002 xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
1003 break;
1004 case XFS_DA_NODE_MAGIC:
1005 /*
1006 * Remove the offending node, fixup hashvals,
1007 * check for a toosmall neighbor.
1008 */
1009 xfs_da3_node_remove(state, drop_blk);
1010 xfs_da3_fixhashpath(state, &state->path);
1011 error = xfs_da3_node_toosmall(state, &action);
1012 if (error)
1013 return error;
1014 if (action == 0)
1015 return 0;
1016 xfs_da3_node_unbalance(state, drop_blk, save_blk);
1017 break;
1018 }
1019 xfs_da3_fixhashpath(state, &state->altpath);
1020 error = xfs_da3_blk_unlink(state, drop_blk, save_blk);
1021 xfs_da_state_kill_altpath(state);
1022 if (error)
1023 return error;
1024 error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
1025 drop_blk->bp);
1026 drop_blk->bp = NULL;
1027 if (error)
1028 return error;
1029 }
1030 /*
1031 * We joined all the way to the top. If it turns out that
1032 * we only have one entry in the root, make the child block
1033 * the new root.
1034 */
1035 xfs_da3_node_remove(state, drop_blk);
1036 xfs_da3_fixhashpath(state, &state->path);
1037 error = xfs_da3_root_join(state, &state->path.blk[0]);
1038 return error;
1039}
1040
1041#ifdef DEBUG
1042static void
1043xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level)
1044{
1045 __be16 magic = blkinfo->magic;
1046
1047 if (level == 1) {
1048 ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1049 magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
1050 magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
1051 magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
1052 } else {
1053 ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1054 magic == cpu_to_be16(XFS_DA3_NODE_MAGIC));
1055 }
1056 ASSERT(!blkinfo->forw);
1057 ASSERT(!blkinfo->back);
1058}
1059#else /* !DEBUG */
1060#define xfs_da_blkinfo_onlychild_validate(blkinfo, level)
1061#endif /* !DEBUG */
1062
1063/*
1064 * We have only one entry in the root. Copy the only remaining child of
1065 * the old root to block 0 as the new root node.
1066 */
1067STATIC int
1068xfs_da3_root_join(
1069 struct xfs_da_state *state,
1070 struct xfs_da_state_blk *root_blk)
1071{
1072 struct xfs_da_intnode *oldroot;
1073 struct xfs_da_args *args;
1074 xfs_dablk_t child;
1075 struct xfs_buf *bp;
1076 struct xfs_da3_icnode_hdr oldroothdr;
1077 struct xfs_da_node_entry *btree;
1078 int error;
1079 struct xfs_inode *dp = state->args->dp;
1080
1081 trace_xfs_da_root_join(state->args);
1082
1083 ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
1084
1085 args = state->args;
1086 oldroot = root_blk->bp->b_addr;
1087 dp->d_ops->node_hdr_from_disk(&oldroothdr, oldroot);
1088 ASSERT(oldroothdr.forw == 0);
1089 ASSERT(oldroothdr.back == 0);
1090
1091 /*
1092 * If the root has more than one child, then don't do anything.
1093 */
1094 if (oldroothdr.count > 1)
1095 return 0;
1096
1097 /*
1098 * Read in the (only) child block, then copy those bytes into
1099 * the root block's buffer and free the original child block.
1100 */
1101 btree = dp->d_ops->node_tree_p(oldroot);
1102 child = be32_to_cpu(btree[0].before);
1103 ASSERT(child != 0);
1104 error = xfs_da3_node_read(args->trans, dp, child, -1, &bp,
1105 args->whichfork);
1106 if (error)
1107 return error;
1108 xfs_da_blkinfo_onlychild_validate(bp->b_addr, oldroothdr.level);
1109
1110 /*
1111 * This could be copying a leaf back into the root block in the case of
1112 * there only being a single leaf block left in the tree. Hence we have
1113 * to update the b_ops pointer as well to match the buffer type change
1114 * that could occur. For dir3 blocks we also need to update the block
1115 * number in the buffer header.
1116 */
1117 memcpy(root_blk->bp->b_addr, bp->b_addr, args->geo->blksize);
1118 root_blk->bp->b_ops = bp->b_ops;
1119 xfs_trans_buf_copy_type(root_blk->bp, bp);
1120 if (oldroothdr.magic == XFS_DA3_NODE_MAGIC) {
1121 struct xfs_da3_blkinfo *da3 = root_blk->bp->b_addr;
1122 da3->blkno = cpu_to_be64(root_blk->bp->b_bn);
1123 }
1124 xfs_trans_log_buf(args->trans, root_blk->bp, 0,
1125 args->geo->blksize - 1);
1126 error = xfs_da_shrink_inode(args, child, bp);
1127 return error;
1128}
1129
1130/*
1131 * Check a node block and its neighbors to see if the block should be
1132 * collapsed into one or the other neighbor. Always keep the block
1133 * with the smaller block number.
1134 * If the current block is over 50% full, don't try to join it, return 0.
1135 * If the block is empty, fill in the state structure and return 2.
1136 * If it can be collapsed, fill in the state structure and return 1.
1137 * If nothing can be done, return 0.
1138 */
1139STATIC int
1140xfs_da3_node_toosmall(
1141 struct xfs_da_state *state,
1142 int *action)
1143{
1144 struct xfs_da_intnode *node;
1145 struct xfs_da_state_blk *blk;
1146 struct xfs_da_blkinfo *info;
1147 xfs_dablk_t blkno;
1148 struct xfs_buf *bp;
1149 struct xfs_da3_icnode_hdr nodehdr;
1150 int count;
1151 int forward;
1152 int error;
1153 int retval;
1154 int i;
1155 struct xfs_inode *dp = state->args->dp;
1156
1157 trace_xfs_da_node_toosmall(state->args);
1158
1159 /*
1160 * Check for the degenerate case of the block being over 50% full.
1161 * If so, it's not worth even looking to see if we might be able
1162 * to coalesce with a sibling.
1163 */
1164 blk = &state->path.blk[ state->path.active-1 ];
1165 info = blk->bp->b_addr;
1166 node = (xfs_da_intnode_t *)info;
1167 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1168 if (nodehdr.count > (state->args->geo->node_ents >> 1)) {
1169 *action = 0; /* blk over 50%, don't try to join */
1170 return 0; /* blk over 50%, don't try to join */
1171 }
1172
1173 /*
1174 * Check for the degenerate case of the block being empty.
1175 * If the block is empty, we'll simply delete it, no need to
1176 * coalesce it with a sibling block. We choose (arbitrarily)
1177 * to merge with the forward block unless it is NULL.
1178 */
1179 if (nodehdr.count == 0) {
1180 /*
1181 * Make altpath point to the block we want to keep and
1182 * path point to the block we want to drop (this one).
1183 */
1184 forward = (info->forw != 0);
1185 memcpy(&state->altpath, &state->path, sizeof(state->path));
1186 error = xfs_da3_path_shift(state, &state->altpath, forward,
1187 0, &retval);
1188 if (error)
1189 return error;
1190 if (retval) {
1191 *action = 0;
1192 } else {
1193 *action = 2;
1194 }
1195 return 0;
1196 }
1197
1198 /*
1199 * Examine each sibling block to see if we can coalesce with
1200 * at least 25% free space to spare. We need to figure out
1201 * whether to merge with the forward or the backward block.
1202 * We prefer coalescing with the lower numbered sibling so as
1203 * to shrink a directory over time.
1204 */
1205 count = state->args->geo->node_ents;
1206 count -= state->args->geo->node_ents >> 2;
1207 count -= nodehdr.count;
1208
1209 /* start with smaller blk num */
1210 forward = nodehdr.forw < nodehdr.back;
1211 for (i = 0; i < 2; forward = !forward, i++) {
1212 struct xfs_da3_icnode_hdr thdr;
1213 if (forward)
1214 blkno = nodehdr.forw;
1215 else
1216 blkno = nodehdr.back;
1217 if (blkno == 0)
1218 continue;
1219 error = xfs_da3_node_read(state->args->trans, dp,
1220 blkno, -1, &bp, state->args->whichfork);
1221 if (error)
1222 return error;
1223
1224 node = bp->b_addr;
1225 dp->d_ops->node_hdr_from_disk(&thdr, node);
1226 xfs_trans_brelse(state->args->trans, bp);
1227
1228 if (count - thdr.count >= 0)
1229 break; /* fits with at least 25% to spare */
1230 }
1231 if (i >= 2) {
1232 *action = 0;
1233 return 0;
1234 }
1235
1236 /*
1237 * Make altpath point to the block we want to keep (the lower
1238 * numbered block) and path point to the block we want to drop.
1239 */
1240 memcpy(&state->altpath, &state->path, sizeof(state->path));
1241 if (blkno < blk->blkno) {
1242 error = xfs_da3_path_shift(state, &state->altpath, forward,
1243 0, &retval);
1244 } else {
1245 error = xfs_da3_path_shift(state, &state->path, forward,
1246 0, &retval);
1247 }
1248 if (error)
1249 return error;
1250 if (retval) {
1251 *action = 0;
1252 return 0;
1253 }
1254 *action = 1;
1255 return 0;
1256}
1257
1258/*
1259 * Pick up the last hashvalue from an intermediate node.
1260 */
1261STATIC uint
1262xfs_da3_node_lasthash(
1263 struct xfs_inode *dp,
1264 struct xfs_buf *bp,
1265 int *count)
1266{
1267 struct xfs_da_intnode *node;
1268 struct xfs_da_node_entry *btree;
1269 struct xfs_da3_icnode_hdr nodehdr;
1270
1271 node = bp->b_addr;
1272 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1273 if (count)
1274 *count = nodehdr.count;
1275 if (!nodehdr.count)
1276 return 0;
1277 btree = dp->d_ops->node_tree_p(node);
1278 return be32_to_cpu(btree[nodehdr.count - 1].hashval);
1279}
1280
1281/*
1282 * Walk back up the tree adjusting hash values as necessary,
1283 * when we stop making changes, return.
1284 */
1285void
1286xfs_da3_fixhashpath(
1287 struct xfs_da_state *state,
1288 struct xfs_da_state_path *path)
1289{
1290 struct xfs_da_state_blk *blk;
1291 struct xfs_da_intnode *node;
1292 struct xfs_da_node_entry *btree;
1293 xfs_dahash_t lasthash=0;
1294 int level;
1295 int count;
1296 struct xfs_inode *dp = state->args->dp;
1297
1298 trace_xfs_da_fixhashpath(state->args);
1299
1300 level = path->active-1;
1301 blk = &path->blk[ level ];
1302 switch (blk->magic) {
1303 case XFS_ATTR_LEAF_MAGIC:
1304 lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
1305 if (count == 0)
1306 return;
1307 break;
1308 case XFS_DIR2_LEAFN_MAGIC:
1309 lasthash = xfs_dir2_leaf_lasthash(dp, blk->bp, &count);
1310 if (count == 0)
1311 return;
1312 break;
1313 case XFS_DA_NODE_MAGIC:
1314 lasthash = xfs_da3_node_lasthash(dp, blk->bp, &count);
1315 if (count == 0)
1316 return;
1317 break;
1318 }
1319 for (blk--, level--; level >= 0; blk--, level--) {
1320 struct xfs_da3_icnode_hdr nodehdr;
1321
1322 node = blk->bp->b_addr;
1323 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1324 btree = dp->d_ops->node_tree_p(node);
1325 if (be32_to_cpu(btree[blk->index].hashval) == lasthash)
1326 break;
1327 blk->hashval = lasthash;
1328 btree[blk->index].hashval = cpu_to_be32(lasthash);
1329 xfs_trans_log_buf(state->args->trans, blk->bp,
1330 XFS_DA_LOGRANGE(node, &btree[blk->index],
1331 sizeof(*btree)));
1332
1333 lasthash = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1334 }
1335}
1336
1337/*
1338 * Remove an entry from an intermediate node.
1339 */
1340STATIC void
1341xfs_da3_node_remove(
1342 struct xfs_da_state *state,
1343 struct xfs_da_state_blk *drop_blk)
1344{
1345 struct xfs_da_intnode *node;
1346 struct xfs_da3_icnode_hdr nodehdr;
1347 struct xfs_da_node_entry *btree;
1348 int index;
1349 int tmp;
1350 struct xfs_inode *dp = state->args->dp;
1351
1352 trace_xfs_da_node_remove(state->args);
1353
1354 node = drop_blk->bp->b_addr;
1355 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1356 ASSERT(drop_blk->index < nodehdr.count);
1357 ASSERT(drop_blk->index >= 0);
1358
1359 /*
1360 * Copy over the offending entry, or just zero it out.
1361 */
1362 index = drop_blk->index;
1363 btree = dp->d_ops->node_tree_p(node);
1364 if (index < nodehdr.count - 1) {
1365 tmp = nodehdr.count - index - 1;
1366 tmp *= (uint)sizeof(xfs_da_node_entry_t);
1367 memmove(&btree[index], &btree[index + 1], tmp);
1368 xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1369 XFS_DA_LOGRANGE(node, &btree[index], tmp));
1370 index = nodehdr.count - 1;
1371 }
1372 memset(&btree[index], 0, sizeof(xfs_da_node_entry_t));
1373 xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1374 XFS_DA_LOGRANGE(node, &btree[index], sizeof(btree[index])));
1375 nodehdr.count -= 1;
1376 dp->d_ops->node_hdr_to_disk(node, &nodehdr);
1377 xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1378 XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
1379
1380 /*
1381 * Copy the last hash value from the block to propagate upwards.
1382 */
1383 drop_blk->hashval = be32_to_cpu(btree[index - 1].hashval);
1384}
1385
1386/*
1387 * Unbalance the elements between two intermediate nodes,
1388 * move all Btree elements from one node into another.
1389 */
1390STATIC void
1391xfs_da3_node_unbalance(
1392 struct xfs_da_state *state,
1393 struct xfs_da_state_blk *drop_blk,
1394 struct xfs_da_state_blk *save_blk)
1395{
1396 struct xfs_da_intnode *drop_node;
1397 struct xfs_da_intnode *save_node;
1398 struct xfs_da_node_entry *drop_btree;
1399 struct xfs_da_node_entry *save_btree;
1400 struct xfs_da3_icnode_hdr drop_hdr;
1401 struct xfs_da3_icnode_hdr save_hdr;
1402 struct xfs_trans *tp;
1403 int sindex;
1404 int tmp;
1405 struct xfs_inode *dp = state->args->dp;
1406
1407 trace_xfs_da_node_unbalance(state->args);
1408
1409 drop_node = drop_blk->bp->b_addr;
1410 save_node = save_blk->bp->b_addr;
1411 dp->d_ops->node_hdr_from_disk(&drop_hdr, drop_node);
1412 dp->d_ops->node_hdr_from_disk(&save_hdr, save_node);
1413 drop_btree = dp->d_ops->node_tree_p(drop_node);
1414 save_btree = dp->d_ops->node_tree_p(save_node);
1415 tp = state->args->trans;
1416
1417 /*
1418 * If the dying block has lower hashvals, then move all the
1419 * elements in the remaining block up to make a hole.
1420 */
1421 if ((be32_to_cpu(drop_btree[0].hashval) <
1422 be32_to_cpu(save_btree[0].hashval)) ||
1423 (be32_to_cpu(drop_btree[drop_hdr.count - 1].hashval) <
1424 be32_to_cpu(save_btree[save_hdr.count - 1].hashval))) {
1425 /* XXX: check this - is memmove dst correct? */
1426 tmp = save_hdr.count * sizeof(xfs_da_node_entry_t);
1427 memmove(&save_btree[drop_hdr.count], &save_btree[0], tmp);
1428
1429 sindex = 0;
1430 xfs_trans_log_buf(tp, save_blk->bp,
1431 XFS_DA_LOGRANGE(save_node, &save_btree[0],
1432 (save_hdr.count + drop_hdr.count) *
1433 sizeof(xfs_da_node_entry_t)));
1434 } else {
1435 sindex = save_hdr.count;
1436 xfs_trans_log_buf(tp, save_blk->bp,
1437 XFS_DA_LOGRANGE(save_node, &save_btree[sindex],
1438 drop_hdr.count * sizeof(xfs_da_node_entry_t)));
1439 }
1440
1441 /*
1442 * Move all the B-tree elements from drop_blk to save_blk.
1443 */
1444 tmp = drop_hdr.count * (uint)sizeof(xfs_da_node_entry_t);
1445 memcpy(&save_btree[sindex], &drop_btree[0], tmp);
1446 save_hdr.count += drop_hdr.count;
1447
1448 dp->d_ops->node_hdr_to_disk(save_node, &save_hdr);
1449 xfs_trans_log_buf(tp, save_blk->bp,
1450 XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1451 dp->d_ops->node_hdr_size));
1452
1453 /*
1454 * Save the last hashval in the remaining block for upward propagation.
1455 */
1456 save_blk->hashval = be32_to_cpu(save_btree[save_hdr.count - 1].hashval);
1457}
1458
1459/*========================================================================
1460 * Routines used for finding things in the Btree.
1461 *========================================================================*/
1462
1463/*
1464 * Walk down the Btree looking for a particular filename, filling
1465 * in the state structure as we go.
1466 *
1467 * We will set the state structure to point to each of the elements
1468 * in each of the nodes where either the hashval is or should be.
1469 *
1470 * We support duplicate hashval's so for each entry in the current
1471 * node that could contain the desired hashval, descend. This is a
1472 * pruned depth-first tree search.
1473 */
1474int /* error */
1475xfs_da3_node_lookup_int(
1476 struct xfs_da_state *state,
1477 int *result)
1478{
1479 struct xfs_da_state_blk *blk;
1480 struct xfs_da_blkinfo *curr;
1481 struct xfs_da_intnode *node;
1482 struct xfs_da_node_entry *btree;
1483 struct xfs_da3_icnode_hdr nodehdr;
1484 struct xfs_da_args *args;
1485 xfs_dablk_t blkno;
1486 xfs_dahash_t hashval;
1487 xfs_dahash_t btreehashval;
1488 int probe;
1489 int span;
1490 int max;
1491 int error;
1492 int retval;
1493 unsigned int expected_level = 0;
1494 struct xfs_inode *dp = state->args->dp;
1495
1496 args = state->args;
1497
1498 /*
1499 * Descend thru the B-tree searching each level for the right
1500 * node to use, until the right hashval is found.
1501 */
1502 blkno = args->geo->leafblk;
1503 for (blk = &state->path.blk[0], state->path.active = 1;
1504 state->path.active <= XFS_DA_NODE_MAXDEPTH;
1505 blk++, state->path.active++) {
1506 /*
1507 * Read the next node down in the tree.
1508 */
1509 blk->blkno = blkno;
1510 error = xfs_da3_node_read(args->trans, args->dp, blkno,
1511 -1, &blk->bp, args->whichfork);
1512 if (error) {
1513 blk->blkno = 0;
1514 state->path.active--;
1515 return error;
1516 }
1517 curr = blk->bp->b_addr;
1518 blk->magic = be16_to_cpu(curr->magic);
1519
1520 if (blk->magic == XFS_ATTR_LEAF_MAGIC ||
1521 blk->magic == XFS_ATTR3_LEAF_MAGIC) {
1522 blk->magic = XFS_ATTR_LEAF_MAGIC;
1523 blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1524 break;
1525 }
1526
1527 if (blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1528 blk->magic == XFS_DIR3_LEAFN_MAGIC) {
1529 blk->magic = XFS_DIR2_LEAFN_MAGIC;
1530 blk->hashval = xfs_dir2_leaf_lasthash(args->dp,
1531 blk->bp, NULL);
1532 break;
1533 }
1534
1535 blk->magic = XFS_DA_NODE_MAGIC;
1536
1537
1538 /*
1539 * Search an intermediate node for a match.
1540 */
1541 node = blk->bp->b_addr;
1542 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1543 btree = dp->d_ops->node_tree_p(node);
1544
1545 /* Tree taller than we can handle; bail out! */
1546 if (nodehdr.level >= XFS_DA_NODE_MAXDEPTH)
1547 return -EFSCORRUPTED;
1548
1549 /* Check the level from the root. */
1550 if (blkno == args->geo->leafblk)
1551 expected_level = nodehdr.level - 1;
1552 else if (expected_level != nodehdr.level)
1553 return -EFSCORRUPTED;
1554 else
1555 expected_level--;
1556
1557 max = nodehdr.count;
1558 blk->hashval = be32_to_cpu(btree[max - 1].hashval);
1559
1560 /*
1561 * Binary search. (note: small blocks will skip loop)
1562 */
1563 probe = span = max / 2;
1564 hashval = args->hashval;
1565 while (span > 4) {
1566 span /= 2;
1567 btreehashval = be32_to_cpu(btree[probe].hashval);
1568 if (btreehashval < hashval)
1569 probe += span;
1570 else if (btreehashval > hashval)
1571 probe -= span;
1572 else
1573 break;
1574 }
1575 ASSERT((probe >= 0) && (probe < max));
1576 ASSERT((span <= 4) ||
1577 (be32_to_cpu(btree[probe].hashval) == hashval));
1578
1579 /*
1580 * Since we may have duplicate hashval's, find the first
1581 * matching hashval in the node.
1582 */
1583 while (probe > 0 &&
1584 be32_to_cpu(btree[probe].hashval) >= hashval) {
1585 probe--;
1586 }
1587 while (probe < max &&
1588 be32_to_cpu(btree[probe].hashval) < hashval) {
1589 probe++;
1590 }
1591
1592 /*
1593 * Pick the right block to descend on.
1594 */
1595 if (probe == max) {
1596 blk->index = max - 1;
1597 blkno = be32_to_cpu(btree[max - 1].before);
1598 } else {
1599 blk->index = probe;
1600 blkno = be32_to_cpu(btree[probe].before);
1601 }
1602
1603 /* We can't point back to the root. */
1604 if (blkno == args->geo->leafblk)
1605 return -EFSCORRUPTED;
1606 }
1607
1608 if (expected_level != 0)
1609 return -EFSCORRUPTED;
1610
1611 /*
1612 * A leaf block that ends in the hashval that we are interested in
1613 * (final hashval == search hashval) means that the next block may
1614 * contain more entries with the same hashval, shift upward to the
1615 * next leaf and keep searching.
1616 */
1617 for (;;) {
1618 if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1619 retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
1620 &blk->index, state);
1621 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1622 retval = xfs_attr3_leaf_lookup_int(blk->bp, args);
1623 blk->index = args->index;
1624 args->blkno = blk->blkno;
1625 } else {
1626 ASSERT(0);
1627 return -EFSCORRUPTED;
1628 }
1629 if (((retval == -ENOENT) || (retval == -ENOATTR)) &&
1630 (blk->hashval == args->hashval)) {
1631 error = xfs_da3_path_shift(state, &state->path, 1, 1,
1632 &retval);
1633 if (error)
1634 return error;
1635 if (retval == 0) {
1636 continue;
1637 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1638 /* path_shift() gives ENOENT */
1639 retval = -ENOATTR;
1640 }
1641 }
1642 break;
1643 }
1644 *result = retval;
1645 return 0;
1646}
1647
1648/*========================================================================
1649 * Utility routines.
1650 *========================================================================*/
1651
1652/*
1653 * Compare two intermediate nodes for "order".
1654 */
1655STATIC int
1656xfs_da3_node_order(
1657 struct xfs_inode *dp,
1658 struct xfs_buf *node1_bp,
1659 struct xfs_buf *node2_bp)
1660{
1661 struct xfs_da_intnode *node1;
1662 struct xfs_da_intnode *node2;
1663 struct xfs_da_node_entry *btree1;
1664 struct xfs_da_node_entry *btree2;
1665 struct xfs_da3_icnode_hdr node1hdr;
1666 struct xfs_da3_icnode_hdr node2hdr;
1667
1668 node1 = node1_bp->b_addr;
1669 node2 = node2_bp->b_addr;
1670 dp->d_ops->node_hdr_from_disk(&node1hdr, node1);
1671 dp->d_ops->node_hdr_from_disk(&node2hdr, node2);
1672 btree1 = dp->d_ops->node_tree_p(node1);
1673 btree2 = dp->d_ops->node_tree_p(node2);
1674
1675 if (node1hdr.count > 0 && node2hdr.count > 0 &&
1676 ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
1677 (be32_to_cpu(btree2[node2hdr.count - 1].hashval) <
1678 be32_to_cpu(btree1[node1hdr.count - 1].hashval)))) {
1679 return 1;
1680 }
1681 return 0;
1682}
1683
1684/*
1685 * Link a new block into a doubly linked list of blocks (of whatever type).
1686 */
1687int /* error */
1688xfs_da3_blk_link(
1689 struct xfs_da_state *state,
1690 struct xfs_da_state_blk *old_blk,
1691 struct xfs_da_state_blk *new_blk)
1692{
1693 struct xfs_da_blkinfo *old_info;
1694 struct xfs_da_blkinfo *new_info;
1695 struct xfs_da_blkinfo *tmp_info;
1696 struct xfs_da_args *args;
1697 struct xfs_buf *bp;
1698 int before = 0;
1699 int error;
1700 struct xfs_inode *dp = state->args->dp;
1701
1702 /*
1703 * Set up environment.
1704 */
1705 args = state->args;
1706 ASSERT(args != NULL);
1707 old_info = old_blk->bp->b_addr;
1708 new_info = new_blk->bp->b_addr;
1709 ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
1710 old_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1711 old_blk->magic == XFS_ATTR_LEAF_MAGIC);
1712
1713 switch (old_blk->magic) {
1714 case XFS_ATTR_LEAF_MAGIC:
1715 before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
1716 break;
1717 case XFS_DIR2_LEAFN_MAGIC:
1718 before = xfs_dir2_leafn_order(dp, old_blk->bp, new_blk->bp);
1719 break;
1720 case XFS_DA_NODE_MAGIC:
1721 before = xfs_da3_node_order(dp, old_blk->bp, new_blk->bp);
1722 break;
1723 }
1724
1725 /*
1726 * Link blocks in appropriate order.
1727 */
1728 if (before) {
1729 /*
1730 * Link new block in before existing block.
1731 */
1732 trace_xfs_da_link_before(args);
1733 new_info->forw = cpu_to_be32(old_blk->blkno);
1734 new_info->back = old_info->back;
1735 if (old_info->back) {
1736 error = xfs_da3_node_read(args->trans, dp,
1737 be32_to_cpu(old_info->back),
1738 -1, &bp, args->whichfork);
1739 if (error)
1740 return error;
1741 ASSERT(bp != NULL);
1742 tmp_info = bp->b_addr;
1743 ASSERT(tmp_info->magic == old_info->magic);
1744 ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno);
1745 tmp_info->forw = cpu_to_be32(new_blk->blkno);
1746 xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1747 }
1748 old_info->back = cpu_to_be32(new_blk->blkno);
1749 } else {
1750 /*
1751 * Link new block in after existing block.
1752 */
1753 trace_xfs_da_link_after(args);
1754 new_info->forw = old_info->forw;
1755 new_info->back = cpu_to_be32(old_blk->blkno);
1756 if (old_info->forw) {
1757 error = xfs_da3_node_read(args->trans, dp,
1758 be32_to_cpu(old_info->forw),
1759 -1, &bp, args->whichfork);
1760 if (error)
1761 return error;
1762 ASSERT(bp != NULL);
1763 tmp_info = bp->b_addr;
1764 ASSERT(tmp_info->magic == old_info->magic);
1765 ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno);
1766 tmp_info->back = cpu_to_be32(new_blk->blkno);
1767 xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1768 }
1769 old_info->forw = cpu_to_be32(new_blk->blkno);
1770 }
1771
1772 xfs_trans_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
1773 xfs_trans_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
1774 return 0;
1775}
1776
1777/*
1778 * Unlink a block from a doubly linked list of blocks.
1779 */
1780STATIC int /* error */
1781xfs_da3_blk_unlink(
1782 struct xfs_da_state *state,
1783 struct xfs_da_state_blk *drop_blk,
1784 struct xfs_da_state_blk *save_blk)
1785{
1786 struct xfs_da_blkinfo *drop_info;
1787 struct xfs_da_blkinfo *save_info;
1788 struct xfs_da_blkinfo *tmp_info;
1789 struct xfs_da_args *args;
1790 struct xfs_buf *bp;
1791 int error;
1792
1793 /*
1794 * Set up environment.
1795 */
1796 args = state->args;
1797 ASSERT(args != NULL);
1798 save_info = save_blk->bp->b_addr;
1799 drop_info = drop_blk->bp->b_addr;
1800 ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
1801 save_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1802 save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1803 ASSERT(save_blk->magic == drop_blk->magic);
1804 ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) ||
1805 (be32_to_cpu(save_info->back) == drop_blk->blkno));
1806 ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) ||
1807 (be32_to_cpu(drop_info->back) == save_blk->blkno));
1808
1809 /*
1810 * Unlink the leaf block from the doubly linked chain of leaves.
1811 */
1812 if (be32_to_cpu(save_info->back) == drop_blk->blkno) {
1813 trace_xfs_da_unlink_back(args);
1814 save_info->back = drop_info->back;
1815 if (drop_info->back) {
1816 error = xfs_da3_node_read(args->trans, args->dp,
1817 be32_to_cpu(drop_info->back),
1818 -1, &bp, args->whichfork);
1819 if (error)
1820 return error;
1821 ASSERT(bp != NULL);
1822 tmp_info = bp->b_addr;
1823 ASSERT(tmp_info->magic == save_info->magic);
1824 ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno);
1825 tmp_info->forw = cpu_to_be32(save_blk->blkno);
1826 xfs_trans_log_buf(args->trans, bp, 0,
1827 sizeof(*tmp_info) - 1);
1828 }
1829 } else {
1830 trace_xfs_da_unlink_forward(args);
1831 save_info->forw = drop_info->forw;
1832 if (drop_info->forw) {
1833 error = xfs_da3_node_read(args->trans, args->dp,
1834 be32_to_cpu(drop_info->forw),
1835 -1, &bp, args->whichfork);
1836 if (error)
1837 return error;
1838 ASSERT(bp != NULL);
1839 tmp_info = bp->b_addr;
1840 ASSERT(tmp_info->magic == save_info->magic);
1841 ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno);
1842 tmp_info->back = cpu_to_be32(save_blk->blkno);
1843 xfs_trans_log_buf(args->trans, bp, 0,
1844 sizeof(*tmp_info) - 1);
1845 }
1846 }
1847
1848 xfs_trans_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
1849 return 0;
1850}
1851
1852/*
1853 * Move a path "forward" or "!forward" one block at the current level.
1854 *
1855 * This routine will adjust a "path" to point to the next block
1856 * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1857 * Btree, including updating pointers to the intermediate nodes between
1858 * the new bottom and the root.
1859 */
1860int /* error */
1861xfs_da3_path_shift(
1862 struct xfs_da_state *state,
1863 struct xfs_da_state_path *path,
1864 int forward,
1865 int release,
1866 int *result)
1867{
1868 struct xfs_da_state_blk *blk;
1869 struct xfs_da_blkinfo *info;
1870 struct xfs_da_intnode *node;
1871 struct xfs_da_args *args;
1872 struct xfs_da_node_entry *btree;
1873 struct xfs_da3_icnode_hdr nodehdr;
1874 struct xfs_buf *bp;
1875 xfs_dablk_t blkno = 0;
1876 int level;
1877 int error;
1878 struct xfs_inode *dp = state->args->dp;
1879
1880 trace_xfs_da_path_shift(state->args);
1881
1882 /*
1883 * Roll up the Btree looking for the first block where our
1884 * current index is not at the edge of the block. Note that
1885 * we skip the bottom layer because we want the sibling block.
1886 */
1887 args = state->args;
1888 ASSERT(args != NULL);
1889 ASSERT(path != NULL);
1890 ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1891 level = (path->active-1) - 1; /* skip bottom layer in path */
1892 for (blk = &path->blk[level]; level >= 0; blk--, level--) {
1893 node = blk->bp->b_addr;
1894 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1895 btree = dp->d_ops->node_tree_p(node);
1896
1897 if (forward && (blk->index < nodehdr.count - 1)) {
1898 blk->index++;
1899 blkno = be32_to_cpu(btree[blk->index].before);
1900 break;
1901 } else if (!forward && (blk->index > 0)) {
1902 blk->index--;
1903 blkno = be32_to_cpu(btree[blk->index].before);
1904 break;
1905 }
1906 }
1907 if (level < 0) {
1908 *result = -ENOENT; /* we're out of our tree */
1909 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1910 return 0;
1911 }
1912
1913 /*
1914 * Roll down the edge of the subtree until we reach the
1915 * same depth we were at originally.
1916 */
1917 for (blk++, level++; level < path->active; blk++, level++) {
1918 /*
1919 * Read the next child block into a local buffer.
1920 */
1921 error = xfs_da3_node_read(args->trans, dp, blkno, -1, &bp,
1922 args->whichfork);
1923 if (error)
1924 return error;
1925
1926 /*
1927 * Release the old block (if it's dirty, the trans doesn't
1928 * actually let go) and swap the local buffer into the path
1929 * structure. This ensures failure of the above read doesn't set
1930 * a NULL buffer in an active slot in the path.
1931 */
1932 if (release)
1933 xfs_trans_brelse(args->trans, blk->bp);
1934 blk->blkno = blkno;
1935 blk->bp = bp;
1936
1937 info = blk->bp->b_addr;
1938 ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1939 info->magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
1940 info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1941 info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
1942 info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
1943 info->magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
1944
1945
1946 /*
1947 * Note: we flatten the magic number to a single type so we
1948 * don't have to compare against crc/non-crc types elsewhere.
1949 */
1950 switch (be16_to_cpu(info->magic)) {
1951 case XFS_DA_NODE_MAGIC:
1952 case XFS_DA3_NODE_MAGIC:
1953 blk->magic = XFS_DA_NODE_MAGIC;
1954 node = (xfs_da_intnode_t *)info;
1955 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1956 btree = dp->d_ops->node_tree_p(node);
1957 blk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1958 if (forward)
1959 blk->index = 0;
1960 else
1961 blk->index = nodehdr.count - 1;
1962 blkno = be32_to_cpu(btree[blk->index].before);
1963 break;
1964 case XFS_ATTR_LEAF_MAGIC:
1965 case XFS_ATTR3_LEAF_MAGIC:
1966 blk->magic = XFS_ATTR_LEAF_MAGIC;
1967 ASSERT(level == path->active-1);
1968 blk->index = 0;
1969 blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1970 break;
1971 case XFS_DIR2_LEAFN_MAGIC:
1972 case XFS_DIR3_LEAFN_MAGIC:
1973 blk->magic = XFS_DIR2_LEAFN_MAGIC;
1974 ASSERT(level == path->active-1);
1975 blk->index = 0;
1976 blk->hashval = xfs_dir2_leaf_lasthash(args->dp,
1977 blk->bp, NULL);
1978 break;
1979 default:
1980 ASSERT(0);
1981 break;
1982 }
1983 }
1984 *result = 0;
1985 return 0;
1986}
1987
1988
1989/*========================================================================
1990 * Utility routines.
1991 *========================================================================*/
1992
1993/*
1994 * Implement a simple hash on a character string.
1995 * Rotate the hash value by 7 bits, then XOR each character in.
1996 * This is implemented with some source-level loop unrolling.
1997 */
1998xfs_dahash_t
1999xfs_da_hashname(const uint8_t *name, int namelen)
2000{
2001 xfs_dahash_t hash;
2002
2003 /*
2004 * Do four characters at a time as long as we can.
2005 */
2006 for (hash = 0; namelen >= 4; namelen -= 4, name += 4)
2007 hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^
2008 (name[3] << 0) ^ rol32(hash, 7 * 4);
2009
2010 /*
2011 * Now do the rest of the characters.
2012 */
2013 switch (namelen) {
2014 case 3:
2015 return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^
2016 rol32(hash, 7 * 3);
2017 case 2:
2018 return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2);
2019 case 1:
2020 return (name[0] << 0) ^ rol32(hash, 7 * 1);
2021 default: /* case 0: */
2022 return hash;
2023 }
2024}
2025
2026enum xfs_dacmp
2027xfs_da_compname(
2028 struct xfs_da_args *args,
2029 const unsigned char *name,
2030 int len)
2031{
2032 return (args->namelen == len && memcmp(args->name, name, len) == 0) ?
2033 XFS_CMP_EXACT : XFS_CMP_DIFFERENT;
2034}
2035
2036static xfs_dahash_t
2037xfs_default_hashname(
2038 struct xfs_name *name)
2039{
2040 return xfs_da_hashname(name->name, name->len);
2041}
2042
2043const struct xfs_nameops xfs_default_nameops = {
2044 .hashname = xfs_default_hashname,
2045 .compname = xfs_da_compname
2046};
2047
2048int
2049xfs_da_grow_inode_int(
2050 struct xfs_da_args *args,
2051 xfs_fileoff_t *bno,
2052 int count)
2053{
2054 struct xfs_trans *tp = args->trans;
2055 struct xfs_inode *dp = args->dp;
2056 int w = args->whichfork;
2057 xfs_rfsblock_t nblks = dp->i_d.di_nblocks;
2058 struct xfs_bmbt_irec map, *mapp;
2059 int nmap, error, got, i, mapi;
2060
2061 /*
2062 * Find a spot in the file space to put the new block.
2063 */
2064 error = xfs_bmap_first_unused(tp, dp, count, bno, w);
2065 if (error)
2066 return error;
2067
2068 /*
2069 * Try mapping it in one filesystem block.
2070 */
2071 nmap = 1;
2072 ASSERT(args->firstblock != NULL);
2073 error = xfs_bmapi_write(tp, dp, *bno, count,
2074 xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG,
2075 args->firstblock, args->total, &map, &nmap,
2076 args->dfops);
2077 if (error)
2078 return error;
2079
2080 ASSERT(nmap <= 1);
2081 if (nmap == 1) {
2082 mapp = ↦
2083 mapi = 1;
2084 } else if (nmap == 0 && count > 1) {
2085 xfs_fileoff_t b;
2086 int c;
2087
2088 /*
2089 * If we didn't get it and the block might work if fragmented,
2090 * try without the CONTIG flag. Loop until we get it all.
2091 */
2092 mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP);
2093 for (b = *bno, mapi = 0; b < *bno + count; ) {
2094 nmap = MIN(XFS_BMAP_MAX_NMAP, count);
2095 c = (int)(*bno + count - b);
2096 error = xfs_bmapi_write(tp, dp, b, c,
2097 xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
2098 args->firstblock, args->total,
2099 &mapp[mapi], &nmap, args->dfops);
2100 if (error)
2101 goto out_free_map;
2102 if (nmap < 1)
2103 break;
2104 mapi += nmap;
2105 b = mapp[mapi - 1].br_startoff +
2106 mapp[mapi - 1].br_blockcount;
2107 }
2108 } else {
2109 mapi = 0;
2110 mapp = NULL;
2111 }
2112
2113 /*
2114 * Count the blocks we got, make sure it matches the total.
2115 */
2116 for (i = 0, got = 0; i < mapi; i++)
2117 got += mapp[i].br_blockcount;
2118 if (got != count || mapp[0].br_startoff != *bno ||
2119 mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
2120 *bno + count) {
2121 error = -ENOSPC;
2122 goto out_free_map;
2123 }
2124
2125 /* account for newly allocated blocks in reserved blocks total */
2126 args->total -= dp->i_d.di_nblocks - nblks;
2127
2128out_free_map:
2129 if (mapp != &map)
2130 kmem_free(mapp);
2131 return error;
2132}
2133
2134/*
2135 * Add a block to the btree ahead of the file.
2136 * Return the new block number to the caller.
2137 */
2138int
2139xfs_da_grow_inode(
2140 struct xfs_da_args *args,
2141 xfs_dablk_t *new_blkno)
2142{
2143 xfs_fileoff_t bno;
2144 int error;
2145
2146 trace_xfs_da_grow_inode(args);
2147
2148 bno = args->geo->leafblk;
2149 error = xfs_da_grow_inode_int(args, &bno, args->geo->fsbcount);
2150 if (!error)
2151 *new_blkno = (xfs_dablk_t)bno;
2152 return error;
2153}
2154
2155/*
2156 * Ick. We need to always be able to remove a btree block, even
2157 * if there's no space reservation because the filesystem is full.
2158 * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
2159 * It swaps the target block with the last block in the file. The
2160 * last block in the file can always be removed since it can't cause
2161 * a bmap btree split to do that.
2162 */
2163STATIC int
2164xfs_da3_swap_lastblock(
2165 struct xfs_da_args *args,
2166 xfs_dablk_t *dead_blknop,
2167 struct xfs_buf **dead_bufp)
2168{
2169 struct xfs_da_blkinfo *dead_info;
2170 struct xfs_da_blkinfo *sib_info;
2171 struct xfs_da_intnode *par_node;
2172 struct xfs_da_intnode *dead_node;
2173 struct xfs_dir2_leaf *dead_leaf2;
2174 struct xfs_da_node_entry *btree;
2175 struct xfs_da3_icnode_hdr par_hdr;
2176 struct xfs_inode *dp;
2177 struct xfs_trans *tp;
2178 struct xfs_mount *mp;
2179 struct xfs_buf *dead_buf;
2180 struct xfs_buf *last_buf;
2181 struct xfs_buf *sib_buf;
2182 struct xfs_buf *par_buf;
2183 xfs_dahash_t dead_hash;
2184 xfs_fileoff_t lastoff;
2185 xfs_dablk_t dead_blkno;
2186 xfs_dablk_t last_blkno;
2187 xfs_dablk_t sib_blkno;
2188 xfs_dablk_t par_blkno;
2189 int error;
2190 int w;
2191 int entno;
2192 int level;
2193 int dead_level;
2194
2195 trace_xfs_da_swap_lastblock(args);
2196
2197 dead_buf = *dead_bufp;
2198 dead_blkno = *dead_blknop;
2199 tp = args->trans;
2200 dp = args->dp;
2201 w = args->whichfork;
2202 ASSERT(w == XFS_DATA_FORK);
2203 mp = dp->i_mount;
2204 lastoff = args->geo->freeblk;
2205 error = xfs_bmap_last_before(tp, dp, &lastoff, w);
2206 if (error)
2207 return error;
2208 if (unlikely(lastoff == 0)) {
2209 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW,
2210 mp);
2211 return -EFSCORRUPTED;
2212 }
2213 /*
2214 * Read the last block in the btree space.
2215 */
2216 last_blkno = (xfs_dablk_t)lastoff - args->geo->fsbcount;
2217 error = xfs_da3_node_read(tp, dp, last_blkno, -1, &last_buf, w);
2218 if (error)
2219 return error;
2220 /*
2221 * Copy the last block into the dead buffer and log it.
2222 */
2223 memcpy(dead_buf->b_addr, last_buf->b_addr, args->geo->blksize);
2224 xfs_trans_log_buf(tp, dead_buf, 0, args->geo->blksize - 1);
2225 dead_info = dead_buf->b_addr;
2226 /*
2227 * Get values from the moved block.
2228 */
2229 if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
2230 dead_info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
2231 struct xfs_dir3_icleaf_hdr leafhdr;
2232 struct xfs_dir2_leaf_entry *ents;
2233
2234 dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
2235 dp->d_ops->leaf_hdr_from_disk(&leafhdr, dead_leaf2);
2236 ents = dp->d_ops->leaf_ents_p(dead_leaf2);
2237 dead_level = 0;
2238 dead_hash = be32_to_cpu(ents[leafhdr.count - 1].hashval);
2239 } else {
2240 struct xfs_da3_icnode_hdr deadhdr;
2241
2242 dead_node = (xfs_da_intnode_t *)dead_info;
2243 dp->d_ops->node_hdr_from_disk(&deadhdr, dead_node);
2244 btree = dp->d_ops->node_tree_p(dead_node);
2245 dead_level = deadhdr.level;
2246 dead_hash = be32_to_cpu(btree[deadhdr.count - 1].hashval);
2247 }
2248 sib_buf = par_buf = NULL;
2249 /*
2250 * If the moved block has a left sibling, fix up the pointers.
2251 */
2252 if ((sib_blkno = be32_to_cpu(dead_info->back))) {
2253 error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w);
2254 if (error)
2255 goto done;
2256 sib_info = sib_buf->b_addr;
2257 if (unlikely(
2258 be32_to_cpu(sib_info->forw) != last_blkno ||
2259 sib_info->magic != dead_info->magic)) {
2260 XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
2261 XFS_ERRLEVEL_LOW, mp);
2262 error = -EFSCORRUPTED;
2263 goto done;
2264 }
2265 sib_info->forw = cpu_to_be32(dead_blkno);
2266 xfs_trans_log_buf(tp, sib_buf,
2267 XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
2268 sizeof(sib_info->forw)));
2269 sib_buf = NULL;
2270 }
2271 /*
2272 * If the moved block has a right sibling, fix up the pointers.
2273 */
2274 if ((sib_blkno = be32_to_cpu(dead_info->forw))) {
2275 error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w);
2276 if (error)
2277 goto done;
2278 sib_info = sib_buf->b_addr;
2279 if (unlikely(
2280 be32_to_cpu(sib_info->back) != last_blkno ||
2281 sib_info->magic != dead_info->magic)) {
2282 XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
2283 XFS_ERRLEVEL_LOW, mp);
2284 error = -EFSCORRUPTED;
2285 goto done;
2286 }
2287 sib_info->back = cpu_to_be32(dead_blkno);
2288 xfs_trans_log_buf(tp, sib_buf,
2289 XFS_DA_LOGRANGE(sib_info, &sib_info->back,
2290 sizeof(sib_info->back)));
2291 sib_buf = NULL;
2292 }
2293 par_blkno = args->geo->leafblk;
2294 level = -1;
2295 /*
2296 * Walk down the tree looking for the parent of the moved block.
2297 */
2298 for (;;) {
2299 error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w);
2300 if (error)
2301 goto done;
2302 par_node = par_buf->b_addr;
2303 dp->d_ops->node_hdr_from_disk(&par_hdr, par_node);
2304 if (level >= 0 && level != par_hdr.level + 1) {
2305 XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
2306 XFS_ERRLEVEL_LOW, mp);
2307 error = -EFSCORRUPTED;
2308 goto done;
2309 }
2310 level = par_hdr.level;
2311 btree = dp->d_ops->node_tree_p(par_node);
2312 for (entno = 0;
2313 entno < par_hdr.count &&
2314 be32_to_cpu(btree[entno].hashval) < dead_hash;
2315 entno++)
2316 continue;
2317 if (entno == par_hdr.count) {
2318 XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
2319 XFS_ERRLEVEL_LOW, mp);
2320 error = -EFSCORRUPTED;
2321 goto done;
2322 }
2323 par_blkno = be32_to_cpu(btree[entno].before);
2324 if (level == dead_level + 1)
2325 break;
2326 xfs_trans_brelse(tp, par_buf);
2327 par_buf = NULL;
2328 }
2329 /*
2330 * We're in the right parent block.
2331 * Look for the right entry.
2332 */
2333 for (;;) {
2334 for (;
2335 entno < par_hdr.count &&
2336 be32_to_cpu(btree[entno].before) != last_blkno;
2337 entno++)
2338 continue;
2339 if (entno < par_hdr.count)
2340 break;
2341 par_blkno = par_hdr.forw;
2342 xfs_trans_brelse(tp, par_buf);
2343 par_buf = NULL;
2344 if (unlikely(par_blkno == 0)) {
2345 XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
2346 XFS_ERRLEVEL_LOW, mp);
2347 error = -EFSCORRUPTED;
2348 goto done;
2349 }
2350 error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w);
2351 if (error)
2352 goto done;
2353 par_node = par_buf->b_addr;
2354 dp->d_ops->node_hdr_from_disk(&par_hdr, par_node);
2355 if (par_hdr.level != level) {
2356 XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
2357 XFS_ERRLEVEL_LOW, mp);
2358 error = -EFSCORRUPTED;
2359 goto done;
2360 }
2361 btree = dp->d_ops->node_tree_p(par_node);
2362 entno = 0;
2363 }
2364 /*
2365 * Update the parent entry pointing to the moved block.
2366 */
2367 btree[entno].before = cpu_to_be32(dead_blkno);
2368 xfs_trans_log_buf(tp, par_buf,
2369 XFS_DA_LOGRANGE(par_node, &btree[entno].before,
2370 sizeof(btree[entno].before)));
2371 *dead_blknop = last_blkno;
2372 *dead_bufp = last_buf;
2373 return 0;
2374done:
2375 if (par_buf)
2376 xfs_trans_brelse(tp, par_buf);
2377 if (sib_buf)
2378 xfs_trans_brelse(tp, sib_buf);
2379 xfs_trans_brelse(tp, last_buf);
2380 return error;
2381}
2382
2383/*
2384 * Remove a btree block from a directory or attribute.
2385 */
2386int
2387xfs_da_shrink_inode(
2388 xfs_da_args_t *args,
2389 xfs_dablk_t dead_blkno,
2390 struct xfs_buf *dead_buf)
2391{
2392 xfs_inode_t *dp;
2393 int done, error, w, count;
2394 xfs_trans_t *tp;
2395
2396 trace_xfs_da_shrink_inode(args);
2397
2398 dp = args->dp;
2399 w = args->whichfork;
2400 tp = args->trans;
2401 count = args->geo->fsbcount;
2402 for (;;) {
2403 /*
2404 * Remove extents. If we get ENOSPC for a dir we have to move
2405 * the last block to the place we want to kill.
2406 */
2407 error = xfs_bunmapi(tp, dp, dead_blkno, count,
2408 xfs_bmapi_aflag(w), 0, args->firstblock,
2409 args->dfops, &done);
2410 if (error == -ENOSPC) {
2411 if (w != XFS_DATA_FORK)
2412 break;
2413 error = xfs_da3_swap_lastblock(args, &dead_blkno,
2414 &dead_buf);
2415 if (error)
2416 break;
2417 } else {
2418 break;
2419 }
2420 }
2421 xfs_trans_binval(tp, dead_buf);
2422 return error;
2423}
2424
2425/*
2426 * See if the mapping(s) for this btree block are valid, i.e.
2427 * don't contain holes, are logically contiguous, and cover the whole range.
2428 */
2429STATIC int
2430xfs_da_map_covers_blocks(
2431 int nmap,
2432 xfs_bmbt_irec_t *mapp,
2433 xfs_dablk_t bno,
2434 int count)
2435{
2436 int i;
2437 xfs_fileoff_t off;
2438
2439 for (i = 0, off = bno; i < nmap; i++) {
2440 if (mapp[i].br_startblock == HOLESTARTBLOCK ||
2441 mapp[i].br_startblock == DELAYSTARTBLOCK) {
2442 return 0;
2443 }
2444 if (off != mapp[i].br_startoff) {
2445 return 0;
2446 }
2447 off += mapp[i].br_blockcount;
2448 }
2449 return off == bno + count;
2450}
2451
2452/*
2453 * Convert a struct xfs_bmbt_irec to a struct xfs_buf_map.
2454 *
2455 * For the single map case, it is assumed that the caller has provided a pointer
2456 * to a valid xfs_buf_map. For the multiple map case, this function will
2457 * allocate the xfs_buf_map to hold all the maps and replace the caller's single
2458 * map pointer with the allocated map.
2459 */
2460static int
2461xfs_buf_map_from_irec(
2462 struct xfs_mount *mp,
2463 struct xfs_buf_map **mapp,
2464 int *nmaps,
2465 struct xfs_bmbt_irec *irecs,
2466 int nirecs)
2467{
2468 struct xfs_buf_map *map;
2469 int i;
2470
2471 ASSERT(*nmaps == 1);
2472 ASSERT(nirecs >= 1);
2473
2474 if (nirecs > 1) {
2475 map = kmem_zalloc(nirecs * sizeof(struct xfs_buf_map),
2476 KM_SLEEP | KM_NOFS);
2477 if (!map)
2478 return -ENOMEM;
2479 *mapp = map;
2480 }
2481
2482 *nmaps = nirecs;
2483 map = *mapp;
2484 for (i = 0; i < *nmaps; i++) {
2485 ASSERT(irecs[i].br_startblock != DELAYSTARTBLOCK &&
2486 irecs[i].br_startblock != HOLESTARTBLOCK);
2487 map[i].bm_bn = XFS_FSB_TO_DADDR(mp, irecs[i].br_startblock);
2488 map[i].bm_len = XFS_FSB_TO_BB(mp, irecs[i].br_blockcount);
2489 }
2490 return 0;
2491}
2492
2493/*
2494 * Map the block we are given ready for reading. There are three possible return
2495 * values:
2496 * -1 - will be returned if we land in a hole and mappedbno == -2 so the
2497 * caller knows not to execute a subsequent read.
2498 * 0 - if we mapped the block successfully
2499 * >0 - positive error number if there was an error.
2500 */
2501static int
2502xfs_dabuf_map(
2503 struct xfs_inode *dp,
2504 xfs_dablk_t bno,
2505 xfs_daddr_t mappedbno,
2506 int whichfork,
2507 struct xfs_buf_map **map,
2508 int *nmaps)
2509{
2510 struct xfs_mount *mp = dp->i_mount;
2511 int nfsb;
2512 int error = 0;
2513 struct xfs_bmbt_irec irec;
2514 struct xfs_bmbt_irec *irecs = &irec;
2515 int nirecs;
2516
2517 ASSERT(map && *map);
2518 ASSERT(*nmaps == 1);
2519
2520 if (whichfork == XFS_DATA_FORK)
2521 nfsb = mp->m_dir_geo->fsbcount;
2522 else
2523 nfsb = mp->m_attr_geo->fsbcount;
2524
2525 /*
2526 * Caller doesn't have a mapping. -2 means don't complain
2527 * if we land in a hole.
2528 */
2529 if (mappedbno == -1 || mappedbno == -2) {
2530 /*
2531 * Optimize the one-block case.
2532 */
2533 if (nfsb != 1)
2534 irecs = kmem_zalloc(sizeof(irec) * nfsb,
2535 KM_SLEEP | KM_NOFS);
2536
2537 nirecs = nfsb;
2538 error = xfs_bmapi_read(dp, (xfs_fileoff_t)bno, nfsb, irecs,
2539 &nirecs, xfs_bmapi_aflag(whichfork));
2540 if (error)
2541 goto out;
2542 } else {
2543 irecs->br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno);
2544 irecs->br_startoff = (xfs_fileoff_t)bno;
2545 irecs->br_blockcount = nfsb;
2546 irecs->br_state = 0;
2547 nirecs = 1;
2548 }
2549
2550 if (!xfs_da_map_covers_blocks(nirecs, irecs, bno, nfsb)) {
2551 error = mappedbno == -2 ? -1 : -EFSCORRUPTED;
2552 if (unlikely(error == -EFSCORRUPTED)) {
2553 if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
2554 int i;
2555 xfs_alert(mp, "%s: bno %lld dir: inode %lld",
2556 __func__, (long long)bno,
2557 (long long)dp->i_ino);
2558 for (i = 0; i < *nmaps; i++) {
2559 xfs_alert(mp,
2560"[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d",
2561 i,
2562 (long long)irecs[i].br_startoff,
2563 (long long)irecs[i].br_startblock,
2564 (long long)irecs[i].br_blockcount,
2565 irecs[i].br_state);
2566 }
2567 }
2568 XFS_ERROR_REPORT("xfs_da_do_buf(1)",
2569 XFS_ERRLEVEL_LOW, mp);
2570 }
2571 goto out;
2572 }
2573 error = xfs_buf_map_from_irec(mp, map, nmaps, irecs, nirecs);
2574out:
2575 if (irecs != &irec)
2576 kmem_free(irecs);
2577 return error;
2578}
2579
2580/*
2581 * Get a buffer for the dir/attr block.
2582 */
2583int
2584xfs_da_get_buf(
2585 struct xfs_trans *trans,
2586 struct xfs_inode *dp,
2587 xfs_dablk_t bno,
2588 xfs_daddr_t mappedbno,
2589 struct xfs_buf **bpp,
2590 int whichfork)
2591{
2592 struct xfs_buf *bp;
2593 struct xfs_buf_map map;
2594 struct xfs_buf_map *mapp;
2595 int nmap;
2596 int error;
2597
2598 *bpp = NULL;
2599 mapp = ↦
2600 nmap = 1;
2601 error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
2602 &mapp, &nmap);
2603 if (error) {
2604 /* mapping a hole is not an error, but we don't continue */
2605 if (error == -1)
2606 error = 0;
2607 goto out_free;
2608 }
2609
2610 bp = xfs_trans_get_buf_map(trans, dp->i_mount->m_ddev_targp,
2611 mapp, nmap, 0);
2612 error = bp ? bp->b_error : -EIO;
2613 if (error) {
2614 if (bp)
2615 xfs_trans_brelse(trans, bp);
2616 goto out_free;
2617 }
2618
2619 *bpp = bp;
2620
2621out_free:
2622 if (mapp != &map)
2623 kmem_free(mapp);
2624
2625 return error;
2626}
2627
2628/*
2629 * Get a buffer for the dir/attr block, fill in the contents.
2630 */
2631int
2632xfs_da_read_buf(
2633 struct xfs_trans *trans,
2634 struct xfs_inode *dp,
2635 xfs_dablk_t bno,
2636 xfs_daddr_t mappedbno,
2637 struct xfs_buf **bpp,
2638 int whichfork,
2639 const struct xfs_buf_ops *ops)
2640{
2641 struct xfs_buf *bp;
2642 struct xfs_buf_map map;
2643 struct xfs_buf_map *mapp;
2644 int nmap;
2645 int error;
2646
2647 *bpp = NULL;
2648 mapp = ↦
2649 nmap = 1;
2650 error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
2651 &mapp, &nmap);
2652 if (error) {
2653 /* mapping a hole is not an error, but we don't continue */
2654 if (error == -1)
2655 error = 0;
2656 goto out_free;
2657 }
2658
2659 error = xfs_trans_read_buf_map(dp->i_mount, trans,
2660 dp->i_mount->m_ddev_targp,
2661 mapp, nmap, 0, &bp, ops);
2662 if (error)
2663 goto out_free;
2664
2665 if (whichfork == XFS_ATTR_FORK)
2666 xfs_buf_set_ref(bp, XFS_ATTR_BTREE_REF);
2667 else
2668 xfs_buf_set_ref(bp, XFS_DIR_BTREE_REF);
2669 *bpp = bp;
2670out_free:
2671 if (mapp != &map)
2672 kmem_free(mapp);
2673
2674 return error;
2675}
2676
2677/*
2678 * Readahead the dir/attr block.
2679 */
2680int
2681xfs_da_reada_buf(
2682 struct xfs_inode *dp,
2683 xfs_dablk_t bno,
2684 xfs_daddr_t mappedbno,
2685 int whichfork,
2686 const struct xfs_buf_ops *ops)
2687{
2688 struct xfs_buf_map map;
2689 struct xfs_buf_map *mapp;
2690 int nmap;
2691 int error;
2692
2693 mapp = ↦
2694 nmap = 1;
2695 error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
2696 &mapp, &nmap);
2697 if (error) {
2698 /* mapping a hole is not an error, but we don't continue */
2699 if (error == -1)
2700 error = 0;
2701 goto out_free;
2702 }
2703
2704 mappedbno = mapp[0].bm_bn;
2705 xfs_buf_readahead_map(dp->i_mount->m_ddev_targp, mapp, nmap, ops);
2706
2707out_free:
2708 if (mapp != &map)
2709 kmem_free(mapp);
2710
2711 return error;
2712}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * Copyright (c) 2013 Red Hat, Inc.
5 * All Rights Reserved.
6 */
7#include "xfs.h"
8#include "xfs_fs.h"
9#include "xfs_shared.h"
10#include "xfs_format.h"
11#include "xfs_log_format.h"
12#include "xfs_trans_resv.h"
13#include "xfs_bit.h"
14#include "xfs_mount.h"
15#include "xfs_inode.h"
16#include "xfs_dir2.h"
17#include "xfs_dir2_priv.h"
18#include "xfs_trans.h"
19#include "xfs_bmap.h"
20#include "xfs_attr_leaf.h"
21#include "xfs_error.h"
22#include "xfs_trace.h"
23#include "xfs_buf_item.h"
24#include "xfs_log.h"
25#include "xfs_errortag.h"
26#include "xfs_health.h"
27
28/*
29 * xfs_da_btree.c
30 *
31 * Routines to implement directories as Btrees of hashed names.
32 */
33
34/*========================================================================
35 * Function prototypes for the kernel.
36 *========================================================================*/
37
38/*
39 * Routines used for growing the Btree.
40 */
41STATIC int xfs_da3_root_split(xfs_da_state_t *state,
42 xfs_da_state_blk_t *existing_root,
43 xfs_da_state_blk_t *new_child);
44STATIC int xfs_da3_node_split(xfs_da_state_t *state,
45 xfs_da_state_blk_t *existing_blk,
46 xfs_da_state_blk_t *split_blk,
47 xfs_da_state_blk_t *blk_to_add,
48 int treelevel,
49 int *result);
50STATIC void xfs_da3_node_rebalance(xfs_da_state_t *state,
51 xfs_da_state_blk_t *node_blk_1,
52 xfs_da_state_blk_t *node_blk_2);
53STATIC void xfs_da3_node_add(xfs_da_state_t *state,
54 xfs_da_state_blk_t *old_node_blk,
55 xfs_da_state_blk_t *new_node_blk);
56
57/*
58 * Routines used for shrinking the Btree.
59 */
60STATIC int xfs_da3_root_join(xfs_da_state_t *state,
61 xfs_da_state_blk_t *root_blk);
62STATIC int xfs_da3_node_toosmall(xfs_da_state_t *state, int *retval);
63STATIC void xfs_da3_node_remove(xfs_da_state_t *state,
64 xfs_da_state_blk_t *drop_blk);
65STATIC void xfs_da3_node_unbalance(xfs_da_state_t *state,
66 xfs_da_state_blk_t *src_node_blk,
67 xfs_da_state_blk_t *dst_node_blk);
68
69/*
70 * Utility routines.
71 */
72STATIC int xfs_da3_blk_unlink(xfs_da_state_t *state,
73 xfs_da_state_blk_t *drop_blk,
74 xfs_da_state_blk_t *save_blk);
75
76
77struct kmem_cache *xfs_da_state_cache; /* anchor for dir/attr state */
78
79/*
80 * Allocate a dir-state structure.
81 * We don't put them on the stack since they're large.
82 */
83struct xfs_da_state *
84xfs_da_state_alloc(
85 struct xfs_da_args *args)
86{
87 struct xfs_da_state *state;
88
89 state = kmem_cache_zalloc(xfs_da_state_cache,
90 GFP_KERNEL | __GFP_NOLOCKDEP | __GFP_NOFAIL);
91 state->args = args;
92 state->mp = args->dp->i_mount;
93 return state;
94}
95
96/*
97 * Kill the altpath contents of a da-state structure.
98 */
99STATIC void
100xfs_da_state_kill_altpath(xfs_da_state_t *state)
101{
102 int i;
103
104 for (i = 0; i < state->altpath.active; i++)
105 state->altpath.blk[i].bp = NULL;
106 state->altpath.active = 0;
107}
108
109/*
110 * Free a da-state structure.
111 */
112void
113xfs_da_state_free(xfs_da_state_t *state)
114{
115 xfs_da_state_kill_altpath(state);
116#ifdef DEBUG
117 memset((char *)state, 0, sizeof(*state));
118#endif /* DEBUG */
119 kmem_cache_free(xfs_da_state_cache, state);
120}
121
122void
123xfs_da_state_reset(
124 struct xfs_da_state *state,
125 struct xfs_da_args *args)
126{
127 xfs_da_state_kill_altpath(state);
128 memset(state, 0, sizeof(struct xfs_da_state));
129 state->args = args;
130 state->mp = state->args->dp->i_mount;
131}
132
133static inline int xfs_dabuf_nfsb(struct xfs_mount *mp, int whichfork)
134{
135 if (whichfork == XFS_DATA_FORK)
136 return mp->m_dir_geo->fsbcount;
137 return mp->m_attr_geo->fsbcount;
138}
139
140void
141xfs_da3_node_hdr_from_disk(
142 struct xfs_mount *mp,
143 struct xfs_da3_icnode_hdr *to,
144 struct xfs_da_intnode *from)
145{
146 if (xfs_has_crc(mp)) {
147 struct xfs_da3_intnode *from3 = (struct xfs_da3_intnode *)from;
148
149 to->forw = be32_to_cpu(from3->hdr.info.hdr.forw);
150 to->back = be32_to_cpu(from3->hdr.info.hdr.back);
151 to->magic = be16_to_cpu(from3->hdr.info.hdr.magic);
152 to->count = be16_to_cpu(from3->hdr.__count);
153 to->level = be16_to_cpu(from3->hdr.__level);
154 to->btree = from3->__btree;
155 ASSERT(to->magic == XFS_DA3_NODE_MAGIC);
156 } else {
157 to->forw = be32_to_cpu(from->hdr.info.forw);
158 to->back = be32_to_cpu(from->hdr.info.back);
159 to->magic = be16_to_cpu(from->hdr.info.magic);
160 to->count = be16_to_cpu(from->hdr.__count);
161 to->level = be16_to_cpu(from->hdr.__level);
162 to->btree = from->__btree;
163 ASSERT(to->magic == XFS_DA_NODE_MAGIC);
164 }
165}
166
167void
168xfs_da3_node_hdr_to_disk(
169 struct xfs_mount *mp,
170 struct xfs_da_intnode *to,
171 struct xfs_da3_icnode_hdr *from)
172{
173 if (xfs_has_crc(mp)) {
174 struct xfs_da3_intnode *to3 = (struct xfs_da3_intnode *)to;
175
176 ASSERT(from->magic == XFS_DA3_NODE_MAGIC);
177 to3->hdr.info.hdr.forw = cpu_to_be32(from->forw);
178 to3->hdr.info.hdr.back = cpu_to_be32(from->back);
179 to3->hdr.info.hdr.magic = cpu_to_be16(from->magic);
180 to3->hdr.__count = cpu_to_be16(from->count);
181 to3->hdr.__level = cpu_to_be16(from->level);
182 } else {
183 ASSERT(from->magic == XFS_DA_NODE_MAGIC);
184 to->hdr.info.forw = cpu_to_be32(from->forw);
185 to->hdr.info.back = cpu_to_be32(from->back);
186 to->hdr.info.magic = cpu_to_be16(from->magic);
187 to->hdr.__count = cpu_to_be16(from->count);
188 to->hdr.__level = cpu_to_be16(from->level);
189 }
190}
191
192/*
193 * Verify an xfs_da3_blkinfo structure. Note that the da3 fields are only
194 * accessible on v5 filesystems. This header format is common across da node,
195 * attr leaf and dir leaf blocks.
196 */
197xfs_failaddr_t
198xfs_da3_blkinfo_verify(
199 struct xfs_buf *bp,
200 struct xfs_da3_blkinfo *hdr3)
201{
202 struct xfs_mount *mp = bp->b_mount;
203 struct xfs_da_blkinfo *hdr = &hdr3->hdr;
204
205 if (!xfs_verify_magic16(bp, hdr->magic))
206 return __this_address;
207
208 if (xfs_has_crc(mp)) {
209 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_meta_uuid))
210 return __this_address;
211 if (be64_to_cpu(hdr3->blkno) != xfs_buf_daddr(bp))
212 return __this_address;
213 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->lsn)))
214 return __this_address;
215 }
216
217 return NULL;
218}
219
220static xfs_failaddr_t
221xfs_da3_node_verify(
222 struct xfs_buf *bp)
223{
224 struct xfs_mount *mp = bp->b_mount;
225 struct xfs_da_intnode *hdr = bp->b_addr;
226 struct xfs_da3_icnode_hdr ichdr;
227 xfs_failaddr_t fa;
228
229 xfs_da3_node_hdr_from_disk(mp, &ichdr, hdr);
230
231 fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
232 if (fa)
233 return fa;
234
235 if (ichdr.level == 0)
236 return __this_address;
237 if (ichdr.level > XFS_DA_NODE_MAXDEPTH)
238 return __this_address;
239 if (ichdr.count == 0)
240 return __this_address;
241
242 /*
243 * we don't know if the node is for and attribute or directory tree,
244 * so only fail if the count is outside both bounds
245 */
246 if (ichdr.count > mp->m_dir_geo->node_ents &&
247 ichdr.count > mp->m_attr_geo->node_ents)
248 return __this_address;
249
250 /* XXX: hash order check? */
251
252 return NULL;
253}
254
255xfs_failaddr_t
256xfs_da3_node_header_check(
257 struct xfs_buf *bp,
258 xfs_ino_t owner)
259{
260 struct xfs_mount *mp = bp->b_mount;
261
262 if (xfs_has_crc(mp)) {
263 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
264
265 if (hdr3->hdr.magic != cpu_to_be16(XFS_DA3_NODE_MAGIC))
266 return __this_address;
267
268 if (be64_to_cpu(hdr3->owner) != owner)
269 return __this_address;
270 }
271
272 return NULL;
273}
274
275xfs_failaddr_t
276xfs_da3_header_check(
277 struct xfs_buf *bp,
278 xfs_ino_t owner)
279{
280 struct xfs_mount *mp = bp->b_mount;
281 struct xfs_da_blkinfo *hdr = bp->b_addr;
282
283 if (!xfs_has_crc(mp))
284 return NULL;
285
286 switch (hdr->magic) {
287 case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC):
288 return xfs_attr3_leaf_header_check(bp, owner);
289 case cpu_to_be16(XFS_DA3_NODE_MAGIC):
290 return xfs_da3_node_header_check(bp, owner);
291 case cpu_to_be16(XFS_DIR3_LEAF1_MAGIC):
292 case cpu_to_be16(XFS_DIR3_LEAFN_MAGIC):
293 return xfs_dir3_leaf_header_check(bp, owner);
294 }
295
296 ASSERT(0);
297 return NULL;
298}
299
300static void
301xfs_da3_node_write_verify(
302 struct xfs_buf *bp)
303{
304 struct xfs_mount *mp = bp->b_mount;
305 struct xfs_buf_log_item *bip = bp->b_log_item;
306 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
307 xfs_failaddr_t fa;
308
309 fa = xfs_da3_node_verify(bp);
310 if (fa) {
311 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
312 return;
313 }
314
315 if (!xfs_has_crc(mp))
316 return;
317
318 if (bip)
319 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
320
321 xfs_buf_update_cksum(bp, XFS_DA3_NODE_CRC_OFF);
322}
323
324/*
325 * leaf/node format detection on trees is sketchy, so a node read can be done on
326 * leaf level blocks when detection identifies the tree as a node format tree
327 * incorrectly. In this case, we need to swap the verifier to match the correct
328 * format of the block being read.
329 */
330static void
331xfs_da3_node_read_verify(
332 struct xfs_buf *bp)
333{
334 struct xfs_da_blkinfo *info = bp->b_addr;
335 xfs_failaddr_t fa;
336
337 switch (be16_to_cpu(info->magic)) {
338 case XFS_DA3_NODE_MAGIC:
339 if (!xfs_buf_verify_cksum(bp, XFS_DA3_NODE_CRC_OFF)) {
340 xfs_verifier_error(bp, -EFSBADCRC,
341 __this_address);
342 break;
343 }
344 fallthrough;
345 case XFS_DA_NODE_MAGIC:
346 fa = xfs_da3_node_verify(bp);
347 if (fa)
348 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
349 return;
350 case XFS_ATTR_LEAF_MAGIC:
351 case XFS_ATTR3_LEAF_MAGIC:
352 bp->b_ops = &xfs_attr3_leaf_buf_ops;
353 bp->b_ops->verify_read(bp);
354 return;
355 case XFS_DIR2_LEAFN_MAGIC:
356 case XFS_DIR3_LEAFN_MAGIC:
357 bp->b_ops = &xfs_dir3_leafn_buf_ops;
358 bp->b_ops->verify_read(bp);
359 return;
360 default:
361 xfs_verifier_error(bp, -EFSCORRUPTED, __this_address);
362 break;
363 }
364}
365
366/* Verify the structure of a da3 block. */
367static xfs_failaddr_t
368xfs_da3_node_verify_struct(
369 struct xfs_buf *bp)
370{
371 struct xfs_da_blkinfo *info = bp->b_addr;
372
373 switch (be16_to_cpu(info->magic)) {
374 case XFS_DA3_NODE_MAGIC:
375 case XFS_DA_NODE_MAGIC:
376 return xfs_da3_node_verify(bp);
377 case XFS_ATTR_LEAF_MAGIC:
378 case XFS_ATTR3_LEAF_MAGIC:
379 bp->b_ops = &xfs_attr3_leaf_buf_ops;
380 return bp->b_ops->verify_struct(bp);
381 case XFS_DIR2_LEAFN_MAGIC:
382 case XFS_DIR3_LEAFN_MAGIC:
383 bp->b_ops = &xfs_dir3_leafn_buf_ops;
384 return bp->b_ops->verify_struct(bp);
385 default:
386 return __this_address;
387 }
388}
389
390const struct xfs_buf_ops xfs_da3_node_buf_ops = {
391 .name = "xfs_da3_node",
392 .magic16 = { cpu_to_be16(XFS_DA_NODE_MAGIC),
393 cpu_to_be16(XFS_DA3_NODE_MAGIC) },
394 .verify_read = xfs_da3_node_read_verify,
395 .verify_write = xfs_da3_node_write_verify,
396 .verify_struct = xfs_da3_node_verify_struct,
397};
398
399static int
400xfs_da3_node_set_type(
401 struct xfs_trans *tp,
402 struct xfs_inode *dp,
403 int whichfork,
404 struct xfs_buf *bp)
405{
406 struct xfs_da_blkinfo *info = bp->b_addr;
407
408 switch (be16_to_cpu(info->magic)) {
409 case XFS_DA_NODE_MAGIC:
410 case XFS_DA3_NODE_MAGIC:
411 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
412 return 0;
413 case XFS_ATTR_LEAF_MAGIC:
414 case XFS_ATTR3_LEAF_MAGIC:
415 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_ATTR_LEAF_BUF);
416 return 0;
417 case XFS_DIR2_LEAFN_MAGIC:
418 case XFS_DIR3_LEAFN_MAGIC:
419 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
420 return 0;
421 default:
422 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, tp->t_mountp,
423 info, sizeof(*info));
424 xfs_trans_brelse(tp, bp);
425 xfs_dirattr_mark_sick(dp, whichfork);
426 return -EFSCORRUPTED;
427 }
428}
429
430int
431xfs_da3_node_read(
432 struct xfs_trans *tp,
433 struct xfs_inode *dp,
434 xfs_dablk_t bno,
435 struct xfs_buf **bpp,
436 int whichfork)
437{
438 int error;
439
440 error = xfs_da_read_buf(tp, dp, bno, 0, bpp, whichfork,
441 &xfs_da3_node_buf_ops);
442 if (error || !*bpp || !tp)
443 return error;
444 return xfs_da3_node_set_type(tp, dp, whichfork, *bpp);
445}
446
447int
448xfs_da3_node_read_mapped(
449 struct xfs_trans *tp,
450 struct xfs_inode *dp,
451 xfs_daddr_t mappedbno,
452 struct xfs_buf **bpp,
453 int whichfork)
454{
455 struct xfs_mount *mp = dp->i_mount;
456 int error;
457
458 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, mappedbno,
459 XFS_FSB_TO_BB(mp, xfs_dabuf_nfsb(mp, whichfork)), 0,
460 bpp, &xfs_da3_node_buf_ops);
461 if (xfs_metadata_is_sick(error))
462 xfs_dirattr_mark_sick(dp, whichfork);
463 if (error || !*bpp)
464 return error;
465
466 if (whichfork == XFS_ATTR_FORK)
467 xfs_buf_set_ref(*bpp, XFS_ATTR_BTREE_REF);
468 else
469 xfs_buf_set_ref(*bpp, XFS_DIR_BTREE_REF);
470
471 if (!tp)
472 return 0;
473 return xfs_da3_node_set_type(tp, dp, whichfork, *bpp);
474}
475
476/*
477 * Copy src directory/attr leaf/node buffer to the dst.
478 * For v5 file systems make sure the right blkno is stamped in.
479 */
480void
481xfs_da_buf_copy(
482 struct xfs_buf *dst,
483 struct xfs_buf *src,
484 size_t size)
485{
486 struct xfs_da3_blkinfo *da3 = dst->b_addr;
487
488 memcpy(dst->b_addr, src->b_addr, size);
489 dst->b_ops = src->b_ops;
490 xfs_trans_buf_copy_type(dst, src);
491 if (xfs_has_crc(dst->b_mount))
492 da3->blkno = cpu_to_be64(xfs_buf_daddr(dst));
493}
494
495/*========================================================================
496 * Routines used for growing the Btree.
497 *========================================================================*/
498
499/*
500 * Create the initial contents of an intermediate node.
501 */
502int
503xfs_da3_node_create(
504 struct xfs_da_args *args,
505 xfs_dablk_t blkno,
506 int level,
507 struct xfs_buf **bpp,
508 int whichfork)
509{
510 struct xfs_da_intnode *node;
511 struct xfs_trans *tp = args->trans;
512 struct xfs_mount *mp = tp->t_mountp;
513 struct xfs_da3_icnode_hdr ichdr = {0};
514 struct xfs_buf *bp;
515 int error;
516 struct xfs_inode *dp = args->dp;
517
518 trace_xfs_da_node_create(args);
519 ASSERT(level <= XFS_DA_NODE_MAXDEPTH);
520
521 error = xfs_da_get_buf(tp, dp, blkno, &bp, whichfork);
522 if (error)
523 return error;
524 bp->b_ops = &xfs_da3_node_buf_ops;
525 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
526 node = bp->b_addr;
527
528 if (xfs_has_crc(mp)) {
529 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
530
531 memset(hdr3, 0, sizeof(struct xfs_da3_node_hdr));
532 ichdr.magic = XFS_DA3_NODE_MAGIC;
533 hdr3->info.blkno = cpu_to_be64(xfs_buf_daddr(bp));
534 hdr3->info.owner = cpu_to_be64(args->owner);
535 uuid_copy(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid);
536 } else {
537 ichdr.magic = XFS_DA_NODE_MAGIC;
538 }
539 ichdr.level = level;
540
541 xfs_da3_node_hdr_to_disk(dp->i_mount, node, &ichdr);
542 xfs_trans_log_buf(tp, bp,
543 XFS_DA_LOGRANGE(node, &node->hdr, args->geo->node_hdr_size));
544
545 *bpp = bp;
546 return 0;
547}
548
549/*
550 * Split a leaf node, rebalance, then possibly split
551 * intermediate nodes, rebalance, etc.
552 */
553int /* error */
554xfs_da3_split(
555 struct xfs_da_state *state)
556{
557 struct xfs_da_state_blk *oldblk;
558 struct xfs_da_state_blk *newblk;
559 struct xfs_da_state_blk *addblk;
560 struct xfs_da_intnode *node;
561 int max;
562 int action = 0;
563 int error;
564 int i;
565
566 trace_xfs_da_split(state->args);
567
568 if (XFS_TEST_ERROR(false, state->mp, XFS_ERRTAG_DA_LEAF_SPLIT))
569 return -EIO;
570
571 /*
572 * Walk back up the tree splitting/inserting/adjusting as necessary.
573 * If we need to insert and there isn't room, split the node, then
574 * decide which fragment to insert the new block from below into.
575 * Note that we may split the root this way, but we need more fixup.
576 */
577 max = state->path.active - 1;
578 ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH));
579 ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC ||
580 state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
581
582 addblk = &state->path.blk[max]; /* initial dummy value */
583 for (i = max; (i >= 0) && addblk; state->path.active--, i--) {
584 oldblk = &state->path.blk[i];
585 newblk = &state->altpath.blk[i];
586
587 /*
588 * If a leaf node then
589 * Allocate a new leaf node, then rebalance across them.
590 * else if an intermediate node then
591 * We split on the last layer, must we split the node?
592 */
593 switch (oldblk->magic) {
594 case XFS_ATTR_LEAF_MAGIC:
595 error = xfs_attr3_leaf_split(state, oldblk, newblk);
596 if (error < 0)
597 return error; /* GROT: attr is inconsistent */
598 if (!error) {
599 addblk = newblk;
600 break;
601 }
602 /*
603 * Entry wouldn't fit, split the leaf again. The new
604 * extrablk will be consumed by xfs_da3_node_split if
605 * the node is split.
606 */
607 state->extravalid = 1;
608 if (state->inleaf) {
609 state->extraafter = 0; /* before newblk */
610 trace_xfs_attr_leaf_split_before(state->args);
611 error = xfs_attr3_leaf_split(state, oldblk,
612 &state->extrablk);
613 } else {
614 state->extraafter = 1; /* after newblk */
615 trace_xfs_attr_leaf_split_after(state->args);
616 error = xfs_attr3_leaf_split(state, newblk,
617 &state->extrablk);
618 }
619 if (error == 1)
620 return -ENOSPC;
621 if (error)
622 return error; /* GROT: attr inconsistent */
623 addblk = newblk;
624 break;
625 case XFS_DIR2_LEAFN_MAGIC:
626 error = xfs_dir2_leafn_split(state, oldblk, newblk);
627 if (error)
628 return error;
629 addblk = newblk;
630 break;
631 case XFS_DA_NODE_MAGIC:
632 error = xfs_da3_node_split(state, oldblk, newblk, addblk,
633 max - i, &action);
634 addblk->bp = NULL;
635 if (error)
636 return error; /* GROT: dir is inconsistent */
637 /*
638 * Record the newly split block for the next time thru?
639 */
640 if (action)
641 addblk = newblk;
642 else
643 addblk = NULL;
644 break;
645 }
646
647 /*
648 * Update the btree to show the new hashval for this child.
649 */
650 xfs_da3_fixhashpath(state, &state->path);
651 }
652 if (!addblk)
653 return 0;
654
655 /*
656 * xfs_da3_node_split() should have consumed any extra blocks we added
657 * during a double leaf split in the attr fork. This is guaranteed as
658 * we can't be here if the attr fork only has a single leaf block.
659 */
660 ASSERT(state->extravalid == 0 ||
661 state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
662
663 /*
664 * Split the root node.
665 */
666 ASSERT(state->path.active == 0);
667 oldblk = &state->path.blk[0];
668 error = xfs_da3_root_split(state, oldblk, addblk);
669 if (error)
670 goto out;
671
672 /*
673 * Update pointers to the node which used to be block 0 and just got
674 * bumped because of the addition of a new root node. Note that the
675 * original block 0 could be at any position in the list of blocks in
676 * the tree.
677 *
678 * Note: the magic numbers and sibling pointers are in the same physical
679 * place for both v2 and v3 headers (by design). Hence it doesn't matter
680 * which version of the xfs_da_intnode structure we use here as the
681 * result will be the same using either structure.
682 */
683 node = oldblk->bp->b_addr;
684 if (node->hdr.info.forw) {
685 if (be32_to_cpu(node->hdr.info.forw) != addblk->blkno) {
686 xfs_buf_mark_corrupt(oldblk->bp);
687 xfs_da_mark_sick(state->args);
688 error = -EFSCORRUPTED;
689 goto out;
690 }
691 node = addblk->bp->b_addr;
692 node->hdr.info.back = cpu_to_be32(oldblk->blkno);
693 xfs_trans_log_buf(state->args->trans, addblk->bp,
694 XFS_DA_LOGRANGE(node, &node->hdr.info,
695 sizeof(node->hdr.info)));
696 }
697 node = oldblk->bp->b_addr;
698 if (node->hdr.info.back) {
699 if (be32_to_cpu(node->hdr.info.back) != addblk->blkno) {
700 xfs_buf_mark_corrupt(oldblk->bp);
701 xfs_da_mark_sick(state->args);
702 error = -EFSCORRUPTED;
703 goto out;
704 }
705 node = addblk->bp->b_addr;
706 node->hdr.info.forw = cpu_to_be32(oldblk->blkno);
707 xfs_trans_log_buf(state->args->trans, addblk->bp,
708 XFS_DA_LOGRANGE(node, &node->hdr.info,
709 sizeof(node->hdr.info)));
710 }
711out:
712 addblk->bp = NULL;
713 return error;
714}
715
716/*
717 * Split the root. We have to create a new root and point to the two
718 * parts (the split old root) that we just created. Copy block zero to
719 * the EOF, extending the inode in process.
720 */
721STATIC int /* error */
722xfs_da3_root_split(
723 struct xfs_da_state *state,
724 struct xfs_da_state_blk *blk1,
725 struct xfs_da_state_blk *blk2)
726{
727 struct xfs_da_intnode *node;
728 struct xfs_da_intnode *oldroot;
729 struct xfs_da_node_entry *btree;
730 struct xfs_da3_icnode_hdr nodehdr;
731 struct xfs_da_args *args;
732 struct xfs_buf *bp;
733 struct xfs_inode *dp;
734 struct xfs_trans *tp;
735 struct xfs_dir2_leaf *leaf;
736 xfs_dablk_t blkno;
737 int level;
738 int error;
739 int size;
740
741 trace_xfs_da_root_split(state->args);
742
743 /*
744 * Copy the existing (incorrect) block from the root node position
745 * to a free space somewhere.
746 */
747 args = state->args;
748 error = xfs_da_grow_inode(args, &blkno);
749 if (error)
750 return error;
751
752 dp = args->dp;
753 tp = args->trans;
754 error = xfs_da_get_buf(tp, dp, blkno, &bp, args->whichfork);
755 if (error)
756 return error;
757 node = bp->b_addr;
758 oldroot = blk1->bp->b_addr;
759 if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
760 oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC)) {
761 struct xfs_da3_icnode_hdr icnodehdr;
762
763 xfs_da3_node_hdr_from_disk(dp->i_mount, &icnodehdr, oldroot);
764 btree = icnodehdr.btree;
765 size = (int)((char *)&btree[icnodehdr.count] - (char *)oldroot);
766 level = icnodehdr.level;
767 } else {
768 struct xfs_dir3_icleaf_hdr leafhdr;
769
770 leaf = (xfs_dir2_leaf_t *)oldroot;
771 xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
772
773 ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
774 leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
775 size = (int)((char *)&leafhdr.ents[leafhdr.count] -
776 (char *)leaf);
777 level = 0;
778 }
779
780 /*
781 * Copy old root to new buffer and log it.
782 */
783 xfs_da_buf_copy(bp, blk1->bp, size);
784 xfs_trans_log_buf(tp, bp, 0, size - 1);
785
786 /*
787 * Update blk1 to point to new buffer.
788 */
789 blk1->bp = bp;
790 blk1->blkno = blkno;
791
792 /*
793 * Set up the new root node.
794 */
795 error = xfs_da3_node_create(args,
796 (args->whichfork == XFS_DATA_FORK) ? args->geo->leafblk : 0,
797 level + 1, &bp, args->whichfork);
798 if (error)
799 return error;
800
801 node = bp->b_addr;
802 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
803 btree = nodehdr.btree;
804 btree[0].hashval = cpu_to_be32(blk1->hashval);
805 btree[0].before = cpu_to_be32(blk1->blkno);
806 btree[1].hashval = cpu_to_be32(blk2->hashval);
807 btree[1].before = cpu_to_be32(blk2->blkno);
808 nodehdr.count = 2;
809 xfs_da3_node_hdr_to_disk(dp->i_mount, node, &nodehdr);
810
811#ifdef DEBUG
812 if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
813 oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
814 ASSERT(blk1->blkno >= args->geo->leafblk &&
815 blk1->blkno < args->geo->freeblk);
816 ASSERT(blk2->blkno >= args->geo->leafblk &&
817 blk2->blkno < args->geo->freeblk);
818 }
819#endif
820
821 /* Header is already logged by xfs_da_node_create */
822 xfs_trans_log_buf(tp, bp,
823 XFS_DA_LOGRANGE(node, btree, sizeof(xfs_da_node_entry_t) * 2));
824
825 return 0;
826}
827
828/*
829 * Split the node, rebalance, then add the new entry.
830 */
831STATIC int /* error */
832xfs_da3_node_split(
833 struct xfs_da_state *state,
834 struct xfs_da_state_blk *oldblk,
835 struct xfs_da_state_blk *newblk,
836 struct xfs_da_state_blk *addblk,
837 int treelevel,
838 int *result)
839{
840 struct xfs_da_intnode *node;
841 struct xfs_da3_icnode_hdr nodehdr;
842 xfs_dablk_t blkno;
843 int newcount;
844 int error;
845 int useextra;
846 struct xfs_inode *dp = state->args->dp;
847
848 trace_xfs_da_node_split(state->args);
849
850 node = oldblk->bp->b_addr;
851 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
852
853 /*
854 * With V2 dirs the extra block is data or freespace.
855 */
856 useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK;
857 newcount = 1 + useextra;
858 /*
859 * Do we have to split the node?
860 */
861 if (nodehdr.count + newcount > state->args->geo->node_ents) {
862 /*
863 * Allocate a new node, add to the doubly linked chain of
864 * nodes, then move some of our excess entries into it.
865 */
866 error = xfs_da_grow_inode(state->args, &blkno);
867 if (error)
868 return error; /* GROT: dir is inconsistent */
869
870 error = xfs_da3_node_create(state->args, blkno, treelevel,
871 &newblk->bp, state->args->whichfork);
872 if (error)
873 return error; /* GROT: dir is inconsistent */
874 newblk->blkno = blkno;
875 newblk->magic = XFS_DA_NODE_MAGIC;
876 xfs_da3_node_rebalance(state, oldblk, newblk);
877 error = xfs_da3_blk_link(state, oldblk, newblk);
878 if (error)
879 return error;
880 *result = 1;
881 } else {
882 *result = 0;
883 }
884
885 /*
886 * Insert the new entry(s) into the correct block
887 * (updating last hashval in the process).
888 *
889 * xfs_da3_node_add() inserts BEFORE the given index,
890 * and as a result of using node_lookup_int() we always
891 * point to a valid entry (not after one), but a split
892 * operation always results in a new block whose hashvals
893 * FOLLOW the current block.
894 *
895 * If we had double-split op below us, then add the extra block too.
896 */
897 node = oldblk->bp->b_addr;
898 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
899 if (oldblk->index <= nodehdr.count) {
900 oldblk->index++;
901 xfs_da3_node_add(state, oldblk, addblk);
902 if (useextra) {
903 if (state->extraafter)
904 oldblk->index++;
905 xfs_da3_node_add(state, oldblk, &state->extrablk);
906 state->extravalid = 0;
907 }
908 } else {
909 newblk->index++;
910 xfs_da3_node_add(state, newblk, addblk);
911 if (useextra) {
912 if (state->extraafter)
913 newblk->index++;
914 xfs_da3_node_add(state, newblk, &state->extrablk);
915 state->extravalid = 0;
916 }
917 }
918
919 return 0;
920}
921
922/*
923 * Balance the btree elements between two intermediate nodes,
924 * usually one full and one empty.
925 *
926 * NOTE: if blk2 is empty, then it will get the upper half of blk1.
927 */
928STATIC void
929xfs_da3_node_rebalance(
930 struct xfs_da_state *state,
931 struct xfs_da_state_blk *blk1,
932 struct xfs_da_state_blk *blk2)
933{
934 struct xfs_da_intnode *node1;
935 struct xfs_da_intnode *node2;
936 struct xfs_da_node_entry *btree1;
937 struct xfs_da_node_entry *btree2;
938 struct xfs_da_node_entry *btree_s;
939 struct xfs_da_node_entry *btree_d;
940 struct xfs_da3_icnode_hdr nodehdr1;
941 struct xfs_da3_icnode_hdr nodehdr2;
942 struct xfs_trans *tp;
943 int count;
944 int tmp;
945 int swap = 0;
946 struct xfs_inode *dp = state->args->dp;
947
948 trace_xfs_da_node_rebalance(state->args);
949
950 node1 = blk1->bp->b_addr;
951 node2 = blk2->bp->b_addr;
952 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr1, node1);
953 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr2, node2);
954 btree1 = nodehdr1.btree;
955 btree2 = nodehdr2.btree;
956
957 /*
958 * Figure out how many entries need to move, and in which direction.
959 * Swap the nodes around if that makes it simpler.
960 */
961 if (nodehdr1.count > 0 && nodehdr2.count > 0 &&
962 ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
963 (be32_to_cpu(btree2[nodehdr2.count - 1].hashval) <
964 be32_to_cpu(btree1[nodehdr1.count - 1].hashval)))) {
965 swap(node1, node2);
966 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr1, node1);
967 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr2, node2);
968 btree1 = nodehdr1.btree;
969 btree2 = nodehdr2.btree;
970 swap = 1;
971 }
972
973 count = (nodehdr1.count - nodehdr2.count) / 2;
974 if (count == 0)
975 return;
976 tp = state->args->trans;
977 /*
978 * Two cases: high-to-low and low-to-high.
979 */
980 if (count > 0) {
981 /*
982 * Move elements in node2 up to make a hole.
983 */
984 tmp = nodehdr2.count;
985 if (tmp > 0) {
986 tmp *= (uint)sizeof(xfs_da_node_entry_t);
987 btree_s = &btree2[0];
988 btree_d = &btree2[count];
989 memmove(btree_d, btree_s, tmp);
990 }
991
992 /*
993 * Move the req'd B-tree elements from high in node1 to
994 * low in node2.
995 */
996 nodehdr2.count += count;
997 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
998 btree_s = &btree1[nodehdr1.count - count];
999 btree_d = &btree2[0];
1000 memcpy(btree_d, btree_s, tmp);
1001 nodehdr1.count -= count;
1002 } else {
1003 /*
1004 * Move the req'd B-tree elements from low in node2 to
1005 * high in node1.
1006 */
1007 count = -count;
1008 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
1009 btree_s = &btree2[0];
1010 btree_d = &btree1[nodehdr1.count];
1011 memcpy(btree_d, btree_s, tmp);
1012 nodehdr1.count += count;
1013
1014 xfs_trans_log_buf(tp, blk1->bp,
1015 XFS_DA_LOGRANGE(node1, btree_d, tmp));
1016
1017 /*
1018 * Move elements in node2 down to fill the hole.
1019 */
1020 tmp = nodehdr2.count - count;
1021 tmp *= (uint)sizeof(xfs_da_node_entry_t);
1022 btree_s = &btree2[count];
1023 btree_d = &btree2[0];
1024 memmove(btree_d, btree_s, tmp);
1025 nodehdr2.count -= count;
1026 }
1027
1028 /*
1029 * Log header of node 1 and all current bits of node 2.
1030 */
1031 xfs_da3_node_hdr_to_disk(dp->i_mount, node1, &nodehdr1);
1032 xfs_trans_log_buf(tp, blk1->bp,
1033 XFS_DA_LOGRANGE(node1, &node1->hdr,
1034 state->args->geo->node_hdr_size));
1035
1036 xfs_da3_node_hdr_to_disk(dp->i_mount, node2, &nodehdr2);
1037 xfs_trans_log_buf(tp, blk2->bp,
1038 XFS_DA_LOGRANGE(node2, &node2->hdr,
1039 state->args->geo->node_hdr_size +
1040 (sizeof(btree2[0]) * nodehdr2.count)));
1041
1042 /*
1043 * Record the last hashval from each block for upward propagation.
1044 * (note: don't use the swapped node pointers)
1045 */
1046 if (swap) {
1047 node1 = blk1->bp->b_addr;
1048 node2 = blk2->bp->b_addr;
1049 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr1, node1);
1050 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr2, node2);
1051 btree1 = nodehdr1.btree;
1052 btree2 = nodehdr2.btree;
1053 }
1054 blk1->hashval = be32_to_cpu(btree1[nodehdr1.count - 1].hashval);
1055 blk2->hashval = be32_to_cpu(btree2[nodehdr2.count - 1].hashval);
1056
1057 /*
1058 * Adjust the expected index for insertion.
1059 */
1060 if (blk1->index >= nodehdr1.count) {
1061 blk2->index = blk1->index - nodehdr1.count;
1062 blk1->index = nodehdr1.count + 1; /* make it invalid */
1063 }
1064}
1065
1066/*
1067 * Add a new entry to an intermediate node.
1068 */
1069STATIC void
1070xfs_da3_node_add(
1071 struct xfs_da_state *state,
1072 struct xfs_da_state_blk *oldblk,
1073 struct xfs_da_state_blk *newblk)
1074{
1075 struct xfs_da_intnode *node;
1076 struct xfs_da3_icnode_hdr nodehdr;
1077 struct xfs_da_node_entry *btree;
1078 int tmp;
1079 struct xfs_inode *dp = state->args->dp;
1080
1081 trace_xfs_da_node_add(state->args);
1082
1083 node = oldblk->bp->b_addr;
1084 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
1085 btree = nodehdr.btree;
1086
1087 ASSERT(oldblk->index >= 0 && oldblk->index <= nodehdr.count);
1088 ASSERT(newblk->blkno != 0);
1089 if (state->args->whichfork == XFS_DATA_FORK)
1090 ASSERT(newblk->blkno >= state->args->geo->leafblk &&
1091 newblk->blkno < state->args->geo->freeblk);
1092
1093 /*
1094 * We may need to make some room before we insert the new node.
1095 */
1096 tmp = 0;
1097 if (oldblk->index < nodehdr.count) {
1098 tmp = (nodehdr.count - oldblk->index) * (uint)sizeof(*btree);
1099 memmove(&btree[oldblk->index + 1], &btree[oldblk->index], tmp);
1100 }
1101 btree[oldblk->index].hashval = cpu_to_be32(newblk->hashval);
1102 btree[oldblk->index].before = cpu_to_be32(newblk->blkno);
1103 xfs_trans_log_buf(state->args->trans, oldblk->bp,
1104 XFS_DA_LOGRANGE(node, &btree[oldblk->index],
1105 tmp + sizeof(*btree)));
1106
1107 nodehdr.count += 1;
1108 xfs_da3_node_hdr_to_disk(dp->i_mount, node, &nodehdr);
1109 xfs_trans_log_buf(state->args->trans, oldblk->bp,
1110 XFS_DA_LOGRANGE(node, &node->hdr,
1111 state->args->geo->node_hdr_size));
1112
1113 /*
1114 * Copy the last hash value from the oldblk to propagate upwards.
1115 */
1116 oldblk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1117}
1118
1119/*========================================================================
1120 * Routines used for shrinking the Btree.
1121 *========================================================================*/
1122
1123/*
1124 * Deallocate an empty leaf node, remove it from its parent,
1125 * possibly deallocating that block, etc...
1126 */
1127int
1128xfs_da3_join(
1129 struct xfs_da_state *state)
1130{
1131 struct xfs_da_state_blk *drop_blk;
1132 struct xfs_da_state_blk *save_blk;
1133 int action = 0;
1134 int error;
1135
1136 trace_xfs_da_join(state->args);
1137
1138 drop_blk = &state->path.blk[ state->path.active-1 ];
1139 save_blk = &state->altpath.blk[ state->path.active-1 ];
1140 ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
1141 ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
1142 drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1143
1144 /*
1145 * Walk back up the tree joining/deallocating as necessary.
1146 * When we stop dropping blocks, break out.
1147 */
1148 for ( ; state->path.active >= 2; drop_blk--, save_blk--,
1149 state->path.active--) {
1150 /*
1151 * See if we can combine the block with a neighbor.
1152 * (action == 0) => no options, just leave
1153 * (action == 1) => coalesce, then unlink
1154 * (action == 2) => block empty, unlink it
1155 */
1156 switch (drop_blk->magic) {
1157 case XFS_ATTR_LEAF_MAGIC:
1158 error = xfs_attr3_leaf_toosmall(state, &action);
1159 if (error)
1160 return error;
1161 if (action == 0)
1162 return 0;
1163 xfs_attr3_leaf_unbalance(state, drop_blk, save_blk);
1164 break;
1165 case XFS_DIR2_LEAFN_MAGIC:
1166 error = xfs_dir2_leafn_toosmall(state, &action);
1167 if (error)
1168 return error;
1169 if (action == 0)
1170 return 0;
1171 xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
1172 break;
1173 case XFS_DA_NODE_MAGIC:
1174 /*
1175 * Remove the offending node, fixup hashvals,
1176 * check for a toosmall neighbor.
1177 */
1178 xfs_da3_node_remove(state, drop_blk);
1179 xfs_da3_fixhashpath(state, &state->path);
1180 error = xfs_da3_node_toosmall(state, &action);
1181 if (error)
1182 return error;
1183 if (action == 0)
1184 return 0;
1185 xfs_da3_node_unbalance(state, drop_blk, save_blk);
1186 break;
1187 }
1188 xfs_da3_fixhashpath(state, &state->altpath);
1189 error = xfs_da3_blk_unlink(state, drop_blk, save_blk);
1190 xfs_da_state_kill_altpath(state);
1191 if (error)
1192 return error;
1193 error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
1194 drop_blk->bp);
1195 drop_blk->bp = NULL;
1196 if (error)
1197 return error;
1198 }
1199 /*
1200 * We joined all the way to the top. If it turns out that
1201 * we only have one entry in the root, make the child block
1202 * the new root.
1203 */
1204 xfs_da3_node_remove(state, drop_blk);
1205 xfs_da3_fixhashpath(state, &state->path);
1206 error = xfs_da3_root_join(state, &state->path.blk[0]);
1207 return error;
1208}
1209
1210#ifdef DEBUG
1211static void
1212xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level)
1213{
1214 __be16 magic = blkinfo->magic;
1215
1216 if (level == 1) {
1217 ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1218 magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
1219 magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
1220 magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
1221 } else {
1222 ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1223 magic == cpu_to_be16(XFS_DA3_NODE_MAGIC));
1224 }
1225 ASSERT(!blkinfo->forw);
1226 ASSERT(!blkinfo->back);
1227}
1228#else /* !DEBUG */
1229#define xfs_da_blkinfo_onlychild_validate(blkinfo, level)
1230#endif /* !DEBUG */
1231
1232/*
1233 * We have only one entry in the root. Copy the only remaining child of
1234 * the old root to block 0 as the new root node.
1235 */
1236STATIC int
1237xfs_da3_root_join(
1238 struct xfs_da_state *state,
1239 struct xfs_da_state_blk *root_blk)
1240{
1241 struct xfs_da_intnode *oldroot;
1242 struct xfs_da_args *args;
1243 xfs_dablk_t child;
1244 struct xfs_buf *bp;
1245 struct xfs_da3_icnode_hdr oldroothdr;
1246 int error;
1247 struct xfs_inode *dp = state->args->dp;
1248 xfs_failaddr_t fa;
1249
1250 trace_xfs_da_root_join(state->args);
1251
1252 ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
1253
1254 args = state->args;
1255 oldroot = root_blk->bp->b_addr;
1256 xfs_da3_node_hdr_from_disk(dp->i_mount, &oldroothdr, oldroot);
1257 ASSERT(oldroothdr.forw == 0);
1258 ASSERT(oldroothdr.back == 0);
1259
1260 /*
1261 * If the root has more than one child, then don't do anything.
1262 */
1263 if (oldroothdr.count > 1)
1264 return 0;
1265
1266 /*
1267 * Read in the (only) child block, then copy those bytes into
1268 * the root block's buffer and free the original child block.
1269 */
1270 child = be32_to_cpu(oldroothdr.btree[0].before);
1271 ASSERT(child != 0);
1272 error = xfs_da3_node_read(args->trans, dp, child, &bp, args->whichfork);
1273 if (error)
1274 return error;
1275 fa = xfs_da3_header_check(bp, args->owner);
1276 if (fa) {
1277 __xfs_buf_mark_corrupt(bp, fa);
1278 xfs_trans_brelse(args->trans, bp);
1279 xfs_da_mark_sick(args);
1280 return -EFSCORRUPTED;
1281 }
1282 xfs_da_blkinfo_onlychild_validate(bp->b_addr, oldroothdr.level);
1283
1284 /*
1285 * Copy child to root buffer and log it.
1286 */
1287 xfs_da_buf_copy(root_blk->bp, bp, args->geo->blksize);
1288 xfs_trans_log_buf(args->trans, root_blk->bp, 0,
1289 args->geo->blksize - 1);
1290 /*
1291 * Now we can drop the child buffer.
1292 */
1293 error = xfs_da_shrink_inode(args, child, bp);
1294 return error;
1295}
1296
1297/*
1298 * Check a node block and its neighbors to see if the block should be
1299 * collapsed into one or the other neighbor. Always keep the block
1300 * with the smaller block number.
1301 * If the current block is over 50% full, don't try to join it, return 0.
1302 * If the block is empty, fill in the state structure and return 2.
1303 * If it can be collapsed, fill in the state structure and return 1.
1304 * If nothing can be done, return 0.
1305 */
1306STATIC int
1307xfs_da3_node_toosmall(
1308 struct xfs_da_state *state,
1309 int *action)
1310{
1311 struct xfs_da_intnode *node;
1312 struct xfs_da_state_blk *blk;
1313 struct xfs_da_blkinfo *info;
1314 xfs_dablk_t blkno;
1315 struct xfs_buf *bp;
1316 xfs_failaddr_t fa;
1317 struct xfs_da3_icnode_hdr nodehdr;
1318 int count;
1319 int forward;
1320 int error;
1321 int retval;
1322 int i;
1323 struct xfs_inode *dp = state->args->dp;
1324
1325 trace_xfs_da_node_toosmall(state->args);
1326
1327 /*
1328 * Check for the degenerate case of the block being over 50% full.
1329 * If so, it's not worth even looking to see if we might be able
1330 * to coalesce with a sibling.
1331 */
1332 blk = &state->path.blk[ state->path.active-1 ];
1333 info = blk->bp->b_addr;
1334 node = (xfs_da_intnode_t *)info;
1335 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
1336 if (nodehdr.count > (state->args->geo->node_ents >> 1)) {
1337 *action = 0; /* blk over 50%, don't try to join */
1338 return 0; /* blk over 50%, don't try to join */
1339 }
1340
1341 /*
1342 * Check for the degenerate case of the block being empty.
1343 * If the block is empty, we'll simply delete it, no need to
1344 * coalesce it with a sibling block. We choose (arbitrarily)
1345 * to merge with the forward block unless it is NULL.
1346 */
1347 if (nodehdr.count == 0) {
1348 /*
1349 * Make altpath point to the block we want to keep and
1350 * path point to the block we want to drop (this one).
1351 */
1352 forward = (info->forw != 0);
1353 memcpy(&state->altpath, &state->path, sizeof(state->path));
1354 error = xfs_da3_path_shift(state, &state->altpath, forward,
1355 0, &retval);
1356 if (error)
1357 return error;
1358 if (retval) {
1359 *action = 0;
1360 } else {
1361 *action = 2;
1362 }
1363 return 0;
1364 }
1365
1366 /*
1367 * Examine each sibling block to see if we can coalesce with
1368 * at least 25% free space to spare. We need to figure out
1369 * whether to merge with the forward or the backward block.
1370 * We prefer coalescing with the lower numbered sibling so as
1371 * to shrink a directory over time.
1372 */
1373 count = state->args->geo->node_ents;
1374 count -= state->args->geo->node_ents >> 2;
1375 count -= nodehdr.count;
1376
1377 /* start with smaller blk num */
1378 forward = nodehdr.forw < nodehdr.back;
1379 for (i = 0; i < 2; forward = !forward, i++) {
1380 struct xfs_da3_icnode_hdr thdr;
1381 if (forward)
1382 blkno = nodehdr.forw;
1383 else
1384 blkno = nodehdr.back;
1385 if (blkno == 0)
1386 continue;
1387 error = xfs_da3_node_read(state->args->trans, dp, blkno, &bp,
1388 state->args->whichfork);
1389 if (error)
1390 return error;
1391 fa = xfs_da3_node_header_check(bp, state->args->owner);
1392 if (fa) {
1393 __xfs_buf_mark_corrupt(bp, fa);
1394 xfs_trans_brelse(state->args->trans, bp);
1395 xfs_da_mark_sick(state->args);
1396 return -EFSCORRUPTED;
1397 }
1398
1399 node = bp->b_addr;
1400 xfs_da3_node_hdr_from_disk(dp->i_mount, &thdr, node);
1401 xfs_trans_brelse(state->args->trans, bp);
1402
1403 if (count - thdr.count >= 0)
1404 break; /* fits with at least 25% to spare */
1405 }
1406 if (i >= 2) {
1407 *action = 0;
1408 return 0;
1409 }
1410
1411 /*
1412 * Make altpath point to the block we want to keep (the lower
1413 * numbered block) and path point to the block we want to drop.
1414 */
1415 memcpy(&state->altpath, &state->path, sizeof(state->path));
1416 if (blkno < blk->blkno) {
1417 error = xfs_da3_path_shift(state, &state->altpath, forward,
1418 0, &retval);
1419 } else {
1420 error = xfs_da3_path_shift(state, &state->path, forward,
1421 0, &retval);
1422 }
1423 if (error)
1424 return error;
1425 if (retval) {
1426 *action = 0;
1427 return 0;
1428 }
1429 *action = 1;
1430 return 0;
1431}
1432
1433/*
1434 * Pick up the last hashvalue from an intermediate node.
1435 */
1436STATIC uint
1437xfs_da3_node_lasthash(
1438 struct xfs_inode *dp,
1439 struct xfs_buf *bp,
1440 int *count)
1441{
1442 struct xfs_da3_icnode_hdr nodehdr;
1443
1444 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, bp->b_addr);
1445 if (count)
1446 *count = nodehdr.count;
1447 if (!nodehdr.count)
1448 return 0;
1449 return be32_to_cpu(nodehdr.btree[nodehdr.count - 1].hashval);
1450}
1451
1452/*
1453 * Walk back up the tree adjusting hash values as necessary,
1454 * when we stop making changes, return.
1455 */
1456void
1457xfs_da3_fixhashpath(
1458 struct xfs_da_state *state,
1459 struct xfs_da_state_path *path)
1460{
1461 struct xfs_da_state_blk *blk;
1462 struct xfs_da_intnode *node;
1463 struct xfs_da_node_entry *btree;
1464 xfs_dahash_t lasthash=0;
1465 int level;
1466 int count;
1467 struct xfs_inode *dp = state->args->dp;
1468
1469 trace_xfs_da_fixhashpath(state->args);
1470
1471 level = path->active-1;
1472 blk = &path->blk[ level ];
1473 switch (blk->magic) {
1474 case XFS_ATTR_LEAF_MAGIC:
1475 lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
1476 if (count == 0)
1477 return;
1478 break;
1479 case XFS_DIR2_LEAFN_MAGIC:
1480 lasthash = xfs_dir2_leaf_lasthash(dp, blk->bp, &count);
1481 if (count == 0)
1482 return;
1483 break;
1484 case XFS_DA_NODE_MAGIC:
1485 lasthash = xfs_da3_node_lasthash(dp, blk->bp, &count);
1486 if (count == 0)
1487 return;
1488 break;
1489 }
1490 for (blk--, level--; level >= 0; blk--, level--) {
1491 struct xfs_da3_icnode_hdr nodehdr;
1492
1493 node = blk->bp->b_addr;
1494 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
1495 btree = nodehdr.btree;
1496 if (be32_to_cpu(btree[blk->index].hashval) == lasthash)
1497 break;
1498 blk->hashval = lasthash;
1499 btree[blk->index].hashval = cpu_to_be32(lasthash);
1500 xfs_trans_log_buf(state->args->trans, blk->bp,
1501 XFS_DA_LOGRANGE(node, &btree[blk->index],
1502 sizeof(*btree)));
1503
1504 lasthash = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1505 }
1506}
1507
1508/*
1509 * Remove an entry from an intermediate node.
1510 */
1511STATIC void
1512xfs_da3_node_remove(
1513 struct xfs_da_state *state,
1514 struct xfs_da_state_blk *drop_blk)
1515{
1516 struct xfs_da_intnode *node;
1517 struct xfs_da3_icnode_hdr nodehdr;
1518 struct xfs_da_node_entry *btree;
1519 int index;
1520 int tmp;
1521 struct xfs_inode *dp = state->args->dp;
1522
1523 trace_xfs_da_node_remove(state->args);
1524
1525 node = drop_blk->bp->b_addr;
1526 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
1527 ASSERT(drop_blk->index < nodehdr.count);
1528 ASSERT(drop_blk->index >= 0);
1529
1530 /*
1531 * Copy over the offending entry, or just zero it out.
1532 */
1533 index = drop_blk->index;
1534 btree = nodehdr.btree;
1535 if (index < nodehdr.count - 1) {
1536 tmp = nodehdr.count - index - 1;
1537 tmp *= (uint)sizeof(xfs_da_node_entry_t);
1538 memmove(&btree[index], &btree[index + 1], tmp);
1539 xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1540 XFS_DA_LOGRANGE(node, &btree[index], tmp));
1541 index = nodehdr.count - 1;
1542 }
1543 memset(&btree[index], 0, sizeof(xfs_da_node_entry_t));
1544 xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1545 XFS_DA_LOGRANGE(node, &btree[index], sizeof(btree[index])));
1546 nodehdr.count -= 1;
1547 xfs_da3_node_hdr_to_disk(dp->i_mount, node, &nodehdr);
1548 xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1549 XFS_DA_LOGRANGE(node, &node->hdr, state->args->geo->node_hdr_size));
1550
1551 /*
1552 * Copy the last hash value from the block to propagate upwards.
1553 */
1554 drop_blk->hashval = be32_to_cpu(btree[index - 1].hashval);
1555}
1556
1557/*
1558 * Unbalance the elements between two intermediate nodes,
1559 * move all Btree elements from one node into another.
1560 */
1561STATIC void
1562xfs_da3_node_unbalance(
1563 struct xfs_da_state *state,
1564 struct xfs_da_state_blk *drop_blk,
1565 struct xfs_da_state_blk *save_blk)
1566{
1567 struct xfs_da_intnode *drop_node;
1568 struct xfs_da_intnode *save_node;
1569 struct xfs_da_node_entry *drop_btree;
1570 struct xfs_da_node_entry *save_btree;
1571 struct xfs_da3_icnode_hdr drop_hdr;
1572 struct xfs_da3_icnode_hdr save_hdr;
1573 struct xfs_trans *tp;
1574 int sindex;
1575 int tmp;
1576 struct xfs_inode *dp = state->args->dp;
1577
1578 trace_xfs_da_node_unbalance(state->args);
1579
1580 drop_node = drop_blk->bp->b_addr;
1581 save_node = save_blk->bp->b_addr;
1582 xfs_da3_node_hdr_from_disk(dp->i_mount, &drop_hdr, drop_node);
1583 xfs_da3_node_hdr_from_disk(dp->i_mount, &save_hdr, save_node);
1584 drop_btree = drop_hdr.btree;
1585 save_btree = save_hdr.btree;
1586 tp = state->args->trans;
1587
1588 /*
1589 * If the dying block has lower hashvals, then move all the
1590 * elements in the remaining block up to make a hole.
1591 */
1592 if ((be32_to_cpu(drop_btree[0].hashval) <
1593 be32_to_cpu(save_btree[0].hashval)) ||
1594 (be32_to_cpu(drop_btree[drop_hdr.count - 1].hashval) <
1595 be32_to_cpu(save_btree[save_hdr.count - 1].hashval))) {
1596 /* XXX: check this - is memmove dst correct? */
1597 tmp = save_hdr.count * sizeof(xfs_da_node_entry_t);
1598 memmove(&save_btree[drop_hdr.count], &save_btree[0], tmp);
1599
1600 sindex = 0;
1601 xfs_trans_log_buf(tp, save_blk->bp,
1602 XFS_DA_LOGRANGE(save_node, &save_btree[0],
1603 (save_hdr.count + drop_hdr.count) *
1604 sizeof(xfs_da_node_entry_t)));
1605 } else {
1606 sindex = save_hdr.count;
1607 xfs_trans_log_buf(tp, save_blk->bp,
1608 XFS_DA_LOGRANGE(save_node, &save_btree[sindex],
1609 drop_hdr.count * sizeof(xfs_da_node_entry_t)));
1610 }
1611
1612 /*
1613 * Move all the B-tree elements from drop_blk to save_blk.
1614 */
1615 tmp = drop_hdr.count * (uint)sizeof(xfs_da_node_entry_t);
1616 memcpy(&save_btree[sindex], &drop_btree[0], tmp);
1617 save_hdr.count += drop_hdr.count;
1618
1619 xfs_da3_node_hdr_to_disk(dp->i_mount, save_node, &save_hdr);
1620 xfs_trans_log_buf(tp, save_blk->bp,
1621 XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1622 state->args->geo->node_hdr_size));
1623
1624 /*
1625 * Save the last hashval in the remaining block for upward propagation.
1626 */
1627 save_blk->hashval = be32_to_cpu(save_btree[save_hdr.count - 1].hashval);
1628}
1629
1630/*========================================================================
1631 * Routines used for finding things in the Btree.
1632 *========================================================================*/
1633
1634/*
1635 * Walk down the Btree looking for a particular filename, filling
1636 * in the state structure as we go.
1637 *
1638 * We will set the state structure to point to each of the elements
1639 * in each of the nodes where either the hashval is or should be.
1640 *
1641 * We support duplicate hashval's so for each entry in the current
1642 * node that could contain the desired hashval, descend. This is a
1643 * pruned depth-first tree search.
1644 */
1645int /* error */
1646xfs_da3_node_lookup_int(
1647 struct xfs_da_state *state,
1648 int *result)
1649{
1650 struct xfs_da_state_blk *blk;
1651 struct xfs_da_blkinfo *curr;
1652 struct xfs_da_intnode *node;
1653 struct xfs_da_node_entry *btree;
1654 struct xfs_da3_icnode_hdr nodehdr;
1655 struct xfs_da_args *args;
1656 xfs_failaddr_t fa;
1657 xfs_dablk_t blkno;
1658 xfs_dahash_t hashval;
1659 xfs_dahash_t btreehashval;
1660 int probe;
1661 int span;
1662 int max;
1663 int error;
1664 int retval;
1665 unsigned int expected_level = 0;
1666 uint16_t magic;
1667 struct xfs_inode *dp = state->args->dp;
1668
1669 args = state->args;
1670
1671 /*
1672 * Descend thru the B-tree searching each level for the right
1673 * node to use, until the right hashval is found.
1674 */
1675 blkno = args->geo->leafblk;
1676 for (blk = &state->path.blk[0], state->path.active = 1;
1677 state->path.active <= XFS_DA_NODE_MAXDEPTH;
1678 blk++, state->path.active++) {
1679 /*
1680 * Read the next node down in the tree.
1681 */
1682 blk->blkno = blkno;
1683 error = xfs_da3_node_read(args->trans, args->dp, blkno,
1684 &blk->bp, args->whichfork);
1685 if (error) {
1686 blk->blkno = 0;
1687 state->path.active--;
1688 return error;
1689 }
1690 curr = blk->bp->b_addr;
1691 magic = be16_to_cpu(curr->magic);
1692
1693 if (magic == XFS_ATTR_LEAF_MAGIC ||
1694 magic == XFS_ATTR3_LEAF_MAGIC) {
1695 fa = xfs_attr3_leaf_header_check(blk->bp, args->owner);
1696 if (fa) {
1697 __xfs_buf_mark_corrupt(blk->bp, fa);
1698 xfs_da_mark_sick(args);
1699 return -EFSCORRUPTED;
1700 }
1701 blk->magic = XFS_ATTR_LEAF_MAGIC;
1702 blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1703 break;
1704 }
1705
1706 if (magic == XFS_DIR2_LEAFN_MAGIC ||
1707 magic == XFS_DIR3_LEAFN_MAGIC) {
1708 fa = xfs_dir3_leaf_header_check(blk->bp, args->owner);
1709 if (fa) {
1710 __xfs_buf_mark_corrupt(blk->bp, fa);
1711 xfs_da_mark_sick(args);
1712 return -EFSCORRUPTED;
1713 }
1714 blk->magic = XFS_DIR2_LEAFN_MAGIC;
1715 blk->hashval = xfs_dir2_leaf_lasthash(args->dp,
1716 blk->bp, NULL);
1717 break;
1718 }
1719
1720 if (magic != XFS_DA_NODE_MAGIC && magic != XFS_DA3_NODE_MAGIC) {
1721 xfs_buf_mark_corrupt(blk->bp);
1722 xfs_da_mark_sick(args);
1723 return -EFSCORRUPTED;
1724 }
1725
1726 fa = xfs_da3_node_header_check(blk->bp, args->owner);
1727 if (fa) {
1728 __xfs_buf_mark_corrupt(blk->bp, fa);
1729 xfs_da_mark_sick(args);
1730 return -EFSCORRUPTED;
1731 }
1732
1733 blk->magic = XFS_DA_NODE_MAGIC;
1734
1735 /*
1736 * Search an intermediate node for a match.
1737 */
1738 node = blk->bp->b_addr;
1739 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
1740 btree = nodehdr.btree;
1741
1742 /* Tree taller than we can handle; bail out! */
1743 if (nodehdr.level >= XFS_DA_NODE_MAXDEPTH) {
1744 xfs_buf_mark_corrupt(blk->bp);
1745 xfs_da_mark_sick(args);
1746 return -EFSCORRUPTED;
1747 }
1748
1749 /* Check the level from the root. */
1750 if (blkno == args->geo->leafblk)
1751 expected_level = nodehdr.level - 1;
1752 else if (expected_level != nodehdr.level) {
1753 xfs_buf_mark_corrupt(blk->bp);
1754 xfs_da_mark_sick(args);
1755 return -EFSCORRUPTED;
1756 } else
1757 expected_level--;
1758
1759 max = nodehdr.count;
1760 blk->hashval = be32_to_cpu(btree[max - 1].hashval);
1761
1762 /*
1763 * Binary search. (note: small blocks will skip loop)
1764 */
1765 probe = span = max / 2;
1766 hashval = args->hashval;
1767 while (span > 4) {
1768 span /= 2;
1769 btreehashval = be32_to_cpu(btree[probe].hashval);
1770 if (btreehashval < hashval)
1771 probe += span;
1772 else if (btreehashval > hashval)
1773 probe -= span;
1774 else
1775 break;
1776 }
1777 ASSERT((probe >= 0) && (probe < max));
1778 ASSERT((span <= 4) ||
1779 (be32_to_cpu(btree[probe].hashval) == hashval));
1780
1781 /*
1782 * Since we may have duplicate hashval's, find the first
1783 * matching hashval in the node.
1784 */
1785 while (probe > 0 &&
1786 be32_to_cpu(btree[probe].hashval) >= hashval) {
1787 probe--;
1788 }
1789 while (probe < max &&
1790 be32_to_cpu(btree[probe].hashval) < hashval) {
1791 probe++;
1792 }
1793
1794 /*
1795 * Pick the right block to descend on.
1796 */
1797 if (probe == max) {
1798 blk->index = max - 1;
1799 blkno = be32_to_cpu(btree[max - 1].before);
1800 } else {
1801 blk->index = probe;
1802 blkno = be32_to_cpu(btree[probe].before);
1803 }
1804
1805 /* We can't point back to the root. */
1806 if (XFS_IS_CORRUPT(dp->i_mount, blkno == args->geo->leafblk)) {
1807 xfs_da_mark_sick(args);
1808 return -EFSCORRUPTED;
1809 }
1810 }
1811
1812 if (XFS_IS_CORRUPT(dp->i_mount, expected_level != 0)) {
1813 xfs_da_mark_sick(args);
1814 return -EFSCORRUPTED;
1815 }
1816
1817 /*
1818 * A leaf block that ends in the hashval that we are interested in
1819 * (final hashval == search hashval) means that the next block may
1820 * contain more entries with the same hashval, shift upward to the
1821 * next leaf and keep searching.
1822 */
1823 for (;;) {
1824 if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1825 retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
1826 &blk->index, state);
1827 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1828 retval = xfs_attr3_leaf_lookup_int(blk->bp, args);
1829 blk->index = args->index;
1830 args->blkno = blk->blkno;
1831 } else {
1832 ASSERT(0);
1833 xfs_da_mark_sick(args);
1834 return -EFSCORRUPTED;
1835 }
1836 if (((retval == -ENOENT) || (retval == -ENOATTR)) &&
1837 (blk->hashval == args->hashval)) {
1838 error = xfs_da3_path_shift(state, &state->path, 1, 1,
1839 &retval);
1840 if (error)
1841 return error;
1842 if (retval == 0) {
1843 continue;
1844 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1845 /* path_shift() gives ENOENT */
1846 retval = -ENOATTR;
1847 }
1848 }
1849 break;
1850 }
1851 *result = retval;
1852 return 0;
1853}
1854
1855/*========================================================================
1856 * Utility routines.
1857 *========================================================================*/
1858
1859/*
1860 * Compare two intermediate nodes for "order".
1861 */
1862STATIC int
1863xfs_da3_node_order(
1864 struct xfs_inode *dp,
1865 struct xfs_buf *node1_bp,
1866 struct xfs_buf *node2_bp)
1867{
1868 struct xfs_da_intnode *node1;
1869 struct xfs_da_intnode *node2;
1870 struct xfs_da_node_entry *btree1;
1871 struct xfs_da_node_entry *btree2;
1872 struct xfs_da3_icnode_hdr node1hdr;
1873 struct xfs_da3_icnode_hdr node2hdr;
1874
1875 node1 = node1_bp->b_addr;
1876 node2 = node2_bp->b_addr;
1877 xfs_da3_node_hdr_from_disk(dp->i_mount, &node1hdr, node1);
1878 xfs_da3_node_hdr_from_disk(dp->i_mount, &node2hdr, node2);
1879 btree1 = node1hdr.btree;
1880 btree2 = node2hdr.btree;
1881
1882 if (node1hdr.count > 0 && node2hdr.count > 0 &&
1883 ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
1884 (be32_to_cpu(btree2[node2hdr.count - 1].hashval) <
1885 be32_to_cpu(btree1[node1hdr.count - 1].hashval)))) {
1886 return 1;
1887 }
1888 return 0;
1889}
1890
1891/*
1892 * Link a new block into a doubly linked list of blocks (of whatever type).
1893 */
1894int /* error */
1895xfs_da3_blk_link(
1896 struct xfs_da_state *state,
1897 struct xfs_da_state_blk *old_blk,
1898 struct xfs_da_state_blk *new_blk)
1899{
1900 struct xfs_da_blkinfo *old_info;
1901 struct xfs_da_blkinfo *new_info;
1902 struct xfs_da_blkinfo *tmp_info;
1903 struct xfs_da_args *args;
1904 struct xfs_buf *bp;
1905 xfs_failaddr_t fa;
1906 int before = 0;
1907 int error;
1908 struct xfs_inode *dp = state->args->dp;
1909
1910 /*
1911 * Set up environment.
1912 */
1913 args = state->args;
1914 ASSERT(args != NULL);
1915 old_info = old_blk->bp->b_addr;
1916 new_info = new_blk->bp->b_addr;
1917 ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
1918 old_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1919 old_blk->magic == XFS_ATTR_LEAF_MAGIC);
1920
1921 switch (old_blk->magic) {
1922 case XFS_ATTR_LEAF_MAGIC:
1923 before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
1924 break;
1925 case XFS_DIR2_LEAFN_MAGIC:
1926 before = xfs_dir2_leafn_order(dp, old_blk->bp, new_blk->bp);
1927 break;
1928 case XFS_DA_NODE_MAGIC:
1929 before = xfs_da3_node_order(dp, old_blk->bp, new_blk->bp);
1930 break;
1931 }
1932
1933 /*
1934 * Link blocks in appropriate order.
1935 */
1936 if (before) {
1937 /*
1938 * Link new block in before existing block.
1939 */
1940 trace_xfs_da_link_before(args);
1941 new_info->forw = cpu_to_be32(old_blk->blkno);
1942 new_info->back = old_info->back;
1943 if (old_info->back) {
1944 error = xfs_da3_node_read(args->trans, dp,
1945 be32_to_cpu(old_info->back),
1946 &bp, args->whichfork);
1947 if (error)
1948 return error;
1949 fa = xfs_da3_header_check(bp, args->owner);
1950 if (fa) {
1951 __xfs_buf_mark_corrupt(bp, fa);
1952 xfs_trans_brelse(args->trans, bp);
1953 xfs_da_mark_sick(args);
1954 return -EFSCORRUPTED;
1955 }
1956 ASSERT(bp != NULL);
1957 tmp_info = bp->b_addr;
1958 ASSERT(tmp_info->magic == old_info->magic);
1959 ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno);
1960 tmp_info->forw = cpu_to_be32(new_blk->blkno);
1961 xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1962 }
1963 old_info->back = cpu_to_be32(new_blk->blkno);
1964 } else {
1965 /*
1966 * Link new block in after existing block.
1967 */
1968 trace_xfs_da_link_after(args);
1969 new_info->forw = old_info->forw;
1970 new_info->back = cpu_to_be32(old_blk->blkno);
1971 if (old_info->forw) {
1972 error = xfs_da3_node_read(args->trans, dp,
1973 be32_to_cpu(old_info->forw),
1974 &bp, args->whichfork);
1975 if (error)
1976 return error;
1977 fa = xfs_da3_header_check(bp, args->owner);
1978 if (fa) {
1979 __xfs_buf_mark_corrupt(bp, fa);
1980 xfs_trans_brelse(args->trans, bp);
1981 xfs_da_mark_sick(args);
1982 return -EFSCORRUPTED;
1983 }
1984 ASSERT(bp != NULL);
1985 tmp_info = bp->b_addr;
1986 ASSERT(tmp_info->magic == old_info->magic);
1987 ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno);
1988 tmp_info->back = cpu_to_be32(new_blk->blkno);
1989 xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1990 }
1991 old_info->forw = cpu_to_be32(new_blk->blkno);
1992 }
1993
1994 xfs_trans_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
1995 xfs_trans_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
1996 return 0;
1997}
1998
1999/*
2000 * Unlink a block from a doubly linked list of blocks.
2001 */
2002STATIC int /* error */
2003xfs_da3_blk_unlink(
2004 struct xfs_da_state *state,
2005 struct xfs_da_state_blk *drop_blk,
2006 struct xfs_da_state_blk *save_blk)
2007{
2008 struct xfs_da_blkinfo *drop_info;
2009 struct xfs_da_blkinfo *save_info;
2010 struct xfs_da_blkinfo *tmp_info;
2011 struct xfs_da_args *args;
2012 struct xfs_buf *bp;
2013 xfs_failaddr_t fa;
2014 int error;
2015
2016 /*
2017 * Set up environment.
2018 */
2019 args = state->args;
2020 ASSERT(args != NULL);
2021 save_info = save_blk->bp->b_addr;
2022 drop_info = drop_blk->bp->b_addr;
2023 ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
2024 save_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
2025 save_blk->magic == XFS_ATTR_LEAF_MAGIC);
2026 ASSERT(save_blk->magic == drop_blk->magic);
2027 ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) ||
2028 (be32_to_cpu(save_info->back) == drop_blk->blkno));
2029 ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) ||
2030 (be32_to_cpu(drop_info->back) == save_blk->blkno));
2031
2032 /*
2033 * Unlink the leaf block from the doubly linked chain of leaves.
2034 */
2035 if (be32_to_cpu(save_info->back) == drop_blk->blkno) {
2036 trace_xfs_da_unlink_back(args);
2037 save_info->back = drop_info->back;
2038 if (drop_info->back) {
2039 error = xfs_da3_node_read(args->trans, args->dp,
2040 be32_to_cpu(drop_info->back),
2041 &bp, args->whichfork);
2042 if (error)
2043 return error;
2044 fa = xfs_da3_header_check(bp, args->owner);
2045 if (fa) {
2046 __xfs_buf_mark_corrupt(bp, fa);
2047 xfs_trans_brelse(args->trans, bp);
2048 xfs_da_mark_sick(args);
2049 return -EFSCORRUPTED;
2050 }
2051 ASSERT(bp != NULL);
2052 tmp_info = bp->b_addr;
2053 ASSERT(tmp_info->magic == save_info->magic);
2054 ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno);
2055 tmp_info->forw = cpu_to_be32(save_blk->blkno);
2056 xfs_trans_log_buf(args->trans, bp, 0,
2057 sizeof(*tmp_info) - 1);
2058 }
2059 } else {
2060 trace_xfs_da_unlink_forward(args);
2061 save_info->forw = drop_info->forw;
2062 if (drop_info->forw) {
2063 error = xfs_da3_node_read(args->trans, args->dp,
2064 be32_to_cpu(drop_info->forw),
2065 &bp, args->whichfork);
2066 if (error)
2067 return error;
2068 fa = xfs_da3_header_check(bp, args->owner);
2069 if (fa) {
2070 __xfs_buf_mark_corrupt(bp, fa);
2071 xfs_trans_brelse(args->trans, bp);
2072 xfs_da_mark_sick(args);
2073 return -EFSCORRUPTED;
2074 }
2075 ASSERT(bp != NULL);
2076 tmp_info = bp->b_addr;
2077 ASSERT(tmp_info->magic == save_info->magic);
2078 ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno);
2079 tmp_info->back = cpu_to_be32(save_blk->blkno);
2080 xfs_trans_log_buf(args->trans, bp, 0,
2081 sizeof(*tmp_info) - 1);
2082 }
2083 }
2084
2085 xfs_trans_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
2086 return 0;
2087}
2088
2089/*
2090 * Move a path "forward" or "!forward" one block at the current level.
2091 *
2092 * This routine will adjust a "path" to point to the next block
2093 * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
2094 * Btree, including updating pointers to the intermediate nodes between
2095 * the new bottom and the root.
2096 */
2097int /* error */
2098xfs_da3_path_shift(
2099 struct xfs_da_state *state,
2100 struct xfs_da_state_path *path,
2101 int forward,
2102 int release,
2103 int *result)
2104{
2105 struct xfs_da_state_blk *blk;
2106 struct xfs_da_blkinfo *info;
2107 struct xfs_da_args *args;
2108 struct xfs_da_node_entry *btree;
2109 struct xfs_da3_icnode_hdr nodehdr;
2110 struct xfs_buf *bp;
2111 xfs_failaddr_t fa;
2112 xfs_dablk_t blkno = 0;
2113 int level;
2114 int error;
2115 struct xfs_inode *dp = state->args->dp;
2116
2117 trace_xfs_da_path_shift(state->args);
2118
2119 /*
2120 * Roll up the Btree looking for the first block where our
2121 * current index is not at the edge of the block. Note that
2122 * we skip the bottom layer because we want the sibling block.
2123 */
2124 args = state->args;
2125 ASSERT(args != NULL);
2126 ASSERT(path != NULL);
2127 ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
2128 level = (path->active-1) - 1; /* skip bottom layer in path */
2129 for (; level >= 0; level--) {
2130 blk = &path->blk[level];
2131 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr,
2132 blk->bp->b_addr);
2133
2134 if (forward && (blk->index < nodehdr.count - 1)) {
2135 blk->index++;
2136 blkno = be32_to_cpu(nodehdr.btree[blk->index].before);
2137 break;
2138 } else if (!forward && (blk->index > 0)) {
2139 blk->index--;
2140 blkno = be32_to_cpu(nodehdr.btree[blk->index].before);
2141 break;
2142 }
2143 }
2144 if (level < 0) {
2145 *result = -ENOENT; /* we're out of our tree */
2146 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
2147 return 0;
2148 }
2149
2150 /*
2151 * Roll down the edge of the subtree until we reach the
2152 * same depth we were at originally.
2153 */
2154 for (blk++, level++; level < path->active; blk++, level++) {
2155 /*
2156 * Read the next child block into a local buffer.
2157 */
2158 error = xfs_da3_node_read(args->trans, dp, blkno, &bp,
2159 args->whichfork);
2160 if (error)
2161 return error;
2162
2163 /*
2164 * Release the old block (if it's dirty, the trans doesn't
2165 * actually let go) and swap the local buffer into the path
2166 * structure. This ensures failure of the above read doesn't set
2167 * a NULL buffer in an active slot in the path.
2168 */
2169 if (release)
2170 xfs_trans_brelse(args->trans, blk->bp);
2171 blk->blkno = blkno;
2172 blk->bp = bp;
2173
2174 info = blk->bp->b_addr;
2175 ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
2176 info->magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
2177 info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
2178 info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
2179 info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
2180 info->magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
2181
2182
2183 /*
2184 * Note: we flatten the magic number to a single type so we
2185 * don't have to compare against crc/non-crc types elsewhere.
2186 */
2187 switch (be16_to_cpu(info->magic)) {
2188 case XFS_DA_NODE_MAGIC:
2189 case XFS_DA3_NODE_MAGIC:
2190 fa = xfs_da3_node_header_check(blk->bp, args->owner);
2191 if (fa) {
2192 __xfs_buf_mark_corrupt(blk->bp, fa);
2193 xfs_da_mark_sick(args);
2194 return -EFSCORRUPTED;
2195 }
2196 blk->magic = XFS_DA_NODE_MAGIC;
2197 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr,
2198 bp->b_addr);
2199 btree = nodehdr.btree;
2200 blk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
2201 if (forward)
2202 blk->index = 0;
2203 else
2204 blk->index = nodehdr.count - 1;
2205 blkno = be32_to_cpu(btree[blk->index].before);
2206 break;
2207 case XFS_ATTR_LEAF_MAGIC:
2208 case XFS_ATTR3_LEAF_MAGIC:
2209 fa = xfs_attr3_leaf_header_check(blk->bp, args->owner);
2210 if (fa) {
2211 __xfs_buf_mark_corrupt(blk->bp, fa);
2212 xfs_da_mark_sick(args);
2213 return -EFSCORRUPTED;
2214 }
2215 blk->magic = XFS_ATTR_LEAF_MAGIC;
2216 ASSERT(level == path->active-1);
2217 blk->index = 0;
2218 blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
2219 break;
2220 case XFS_DIR2_LEAFN_MAGIC:
2221 case XFS_DIR3_LEAFN_MAGIC:
2222 fa = xfs_dir3_leaf_header_check(blk->bp, args->owner);
2223 if (fa) {
2224 __xfs_buf_mark_corrupt(blk->bp, fa);
2225 xfs_da_mark_sick(args);
2226 return -EFSCORRUPTED;
2227 }
2228 blk->magic = XFS_DIR2_LEAFN_MAGIC;
2229 ASSERT(level == path->active-1);
2230 blk->index = 0;
2231 blk->hashval = xfs_dir2_leaf_lasthash(args->dp,
2232 blk->bp, NULL);
2233 break;
2234 default:
2235 ASSERT(0);
2236 break;
2237 }
2238 }
2239 *result = 0;
2240 return 0;
2241}
2242
2243
2244/*========================================================================
2245 * Utility routines.
2246 *========================================================================*/
2247
2248/*
2249 * Implement a simple hash on a character string.
2250 * Rotate the hash value by 7 bits, then XOR each character in.
2251 * This is implemented with some source-level loop unrolling.
2252 */
2253xfs_dahash_t
2254xfs_da_hashname(const uint8_t *name, int namelen)
2255{
2256 xfs_dahash_t hash;
2257
2258 /*
2259 * Do four characters at a time as long as we can.
2260 */
2261 for (hash = 0; namelen >= 4; namelen -= 4, name += 4)
2262 hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^
2263 (name[3] << 0) ^ rol32(hash, 7 * 4);
2264
2265 /*
2266 * Now do the rest of the characters.
2267 */
2268 switch (namelen) {
2269 case 3:
2270 return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^
2271 rol32(hash, 7 * 3);
2272 case 2:
2273 return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2);
2274 case 1:
2275 return (name[0] << 0) ^ rol32(hash, 7 * 1);
2276 default: /* case 0: */
2277 return hash;
2278 }
2279}
2280
2281enum xfs_dacmp
2282xfs_da_compname(
2283 struct xfs_da_args *args,
2284 const unsigned char *name,
2285 int len)
2286{
2287 return (args->namelen == len && memcmp(args->name, name, len) == 0) ?
2288 XFS_CMP_EXACT : XFS_CMP_DIFFERENT;
2289}
2290
2291int
2292xfs_da_grow_inode_int(
2293 struct xfs_da_args *args,
2294 xfs_fileoff_t *bno,
2295 int count)
2296{
2297 struct xfs_trans *tp = args->trans;
2298 struct xfs_inode *dp = args->dp;
2299 int w = args->whichfork;
2300 xfs_rfsblock_t nblks = dp->i_nblocks;
2301 struct xfs_bmbt_irec map, *mapp = ↦
2302 int nmap, error, got, i, mapi = 1;
2303
2304 /*
2305 * Find a spot in the file space to put the new block.
2306 */
2307 error = xfs_bmap_first_unused(tp, dp, count, bno, w);
2308 if (error)
2309 return error;
2310
2311 /*
2312 * Try mapping it in one filesystem block.
2313 */
2314 nmap = 1;
2315 error = xfs_bmapi_write(tp, dp, *bno, count,
2316 xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG,
2317 args->total, &map, &nmap);
2318 if (error == -ENOSPC && count > 1) {
2319 xfs_fileoff_t b;
2320 int c;
2321
2322 /*
2323 * If we didn't get it and the block might work if fragmented,
2324 * try without the CONTIG flag. Loop until we get it all.
2325 */
2326 mapp = kmalloc(sizeof(*mapp) * count,
2327 GFP_KERNEL | __GFP_NOFAIL);
2328 for (b = *bno, mapi = 0; b < *bno + count; ) {
2329 c = (int)(*bno + count - b);
2330 nmap = min(XFS_BMAP_MAX_NMAP, c);
2331 error = xfs_bmapi_write(tp, dp, b, c,
2332 xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
2333 args->total, &mapp[mapi], &nmap);
2334 if (error)
2335 goto out_free_map;
2336 mapi += nmap;
2337 b = mapp[mapi - 1].br_startoff +
2338 mapp[mapi - 1].br_blockcount;
2339 }
2340 }
2341 if (error)
2342 goto out_free_map;
2343
2344 /*
2345 * Count the blocks we got, make sure it matches the total.
2346 */
2347 for (i = 0, got = 0; i < mapi; i++)
2348 got += mapp[i].br_blockcount;
2349 if (got != count || mapp[0].br_startoff != *bno ||
2350 mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
2351 *bno + count) {
2352 error = -ENOSPC;
2353 goto out_free_map;
2354 }
2355
2356 /* account for newly allocated blocks in reserved blocks total */
2357 args->total -= dp->i_nblocks - nblks;
2358
2359out_free_map:
2360 if (mapp != &map)
2361 kfree(mapp);
2362 return error;
2363}
2364
2365/*
2366 * Add a block to the btree ahead of the file.
2367 * Return the new block number to the caller.
2368 */
2369int
2370xfs_da_grow_inode(
2371 struct xfs_da_args *args,
2372 xfs_dablk_t *new_blkno)
2373{
2374 xfs_fileoff_t bno;
2375 int error;
2376
2377 trace_xfs_da_grow_inode(args);
2378
2379 bno = args->geo->leafblk;
2380 error = xfs_da_grow_inode_int(args, &bno, args->geo->fsbcount);
2381 if (!error)
2382 *new_blkno = (xfs_dablk_t)bno;
2383 return error;
2384}
2385
2386/*
2387 * Ick. We need to always be able to remove a btree block, even
2388 * if there's no space reservation because the filesystem is full.
2389 * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
2390 * It swaps the target block with the last block in the file. The
2391 * last block in the file can always be removed since it can't cause
2392 * a bmap btree split to do that.
2393 */
2394STATIC int
2395xfs_da3_swap_lastblock(
2396 struct xfs_da_args *args,
2397 xfs_dablk_t *dead_blknop,
2398 struct xfs_buf **dead_bufp)
2399{
2400 struct xfs_da_blkinfo *dead_info;
2401 struct xfs_da_blkinfo *sib_info;
2402 struct xfs_da_intnode *par_node;
2403 struct xfs_da_intnode *dead_node;
2404 struct xfs_dir2_leaf *dead_leaf2;
2405 struct xfs_da_node_entry *btree;
2406 struct xfs_da3_icnode_hdr par_hdr;
2407 struct xfs_inode *dp;
2408 struct xfs_trans *tp;
2409 struct xfs_mount *mp;
2410 struct xfs_buf *dead_buf;
2411 struct xfs_buf *last_buf;
2412 struct xfs_buf *sib_buf;
2413 struct xfs_buf *par_buf;
2414 xfs_failaddr_t fa;
2415 xfs_dahash_t dead_hash;
2416 xfs_fileoff_t lastoff;
2417 xfs_dablk_t dead_blkno;
2418 xfs_dablk_t last_blkno;
2419 xfs_dablk_t sib_blkno;
2420 xfs_dablk_t par_blkno;
2421 int error;
2422 int w;
2423 int entno;
2424 int level;
2425 int dead_level;
2426
2427 trace_xfs_da_swap_lastblock(args);
2428
2429 dead_buf = *dead_bufp;
2430 dead_blkno = *dead_blknop;
2431 tp = args->trans;
2432 dp = args->dp;
2433 w = args->whichfork;
2434 ASSERT(w == XFS_DATA_FORK);
2435 mp = dp->i_mount;
2436 lastoff = args->geo->freeblk;
2437 error = xfs_bmap_last_before(tp, dp, &lastoff, w);
2438 if (error)
2439 return error;
2440 if (XFS_IS_CORRUPT(mp, lastoff == 0)) {
2441 xfs_da_mark_sick(args);
2442 return -EFSCORRUPTED;
2443 }
2444 /*
2445 * Read the last block in the btree space.
2446 */
2447 last_blkno = (xfs_dablk_t)lastoff - args->geo->fsbcount;
2448 error = xfs_da3_node_read(tp, dp, last_blkno, &last_buf, w);
2449 if (error)
2450 return error;
2451 fa = xfs_da3_header_check(last_buf, args->owner);
2452 if (fa) {
2453 __xfs_buf_mark_corrupt(last_buf, fa);
2454 xfs_trans_brelse(tp, last_buf);
2455 xfs_da_mark_sick(args);
2456 return -EFSCORRUPTED;
2457 }
2458
2459 /*
2460 * Copy the last block into the dead buffer and log it.
2461 */
2462 xfs_da_buf_copy(dead_buf, last_buf, args->geo->blksize);
2463 xfs_trans_log_buf(tp, dead_buf, 0, args->geo->blksize - 1);
2464 dead_info = dead_buf->b_addr;
2465
2466 /*
2467 * Get values from the moved block.
2468 */
2469 if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
2470 dead_info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
2471 struct xfs_dir3_icleaf_hdr leafhdr;
2472 struct xfs_dir2_leaf_entry *ents;
2473
2474 dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
2475 xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr,
2476 dead_leaf2);
2477 ents = leafhdr.ents;
2478 dead_level = 0;
2479 dead_hash = be32_to_cpu(ents[leafhdr.count - 1].hashval);
2480 } else {
2481 struct xfs_da3_icnode_hdr deadhdr;
2482
2483 dead_node = (xfs_da_intnode_t *)dead_info;
2484 xfs_da3_node_hdr_from_disk(dp->i_mount, &deadhdr, dead_node);
2485 btree = deadhdr.btree;
2486 dead_level = deadhdr.level;
2487 dead_hash = be32_to_cpu(btree[deadhdr.count - 1].hashval);
2488 }
2489 sib_buf = par_buf = NULL;
2490 /*
2491 * If the moved block has a left sibling, fix up the pointers.
2492 */
2493 if ((sib_blkno = be32_to_cpu(dead_info->back))) {
2494 error = xfs_da3_node_read(tp, dp, sib_blkno, &sib_buf, w);
2495 if (error)
2496 goto done;
2497 fa = xfs_da3_header_check(sib_buf, args->owner);
2498 if (fa) {
2499 __xfs_buf_mark_corrupt(sib_buf, fa);
2500 xfs_da_mark_sick(args);
2501 error = -EFSCORRUPTED;
2502 goto done;
2503 }
2504 sib_info = sib_buf->b_addr;
2505 if (XFS_IS_CORRUPT(mp,
2506 be32_to_cpu(sib_info->forw) != last_blkno ||
2507 sib_info->magic != dead_info->magic)) {
2508 xfs_da_mark_sick(args);
2509 error = -EFSCORRUPTED;
2510 goto done;
2511 }
2512 sib_info->forw = cpu_to_be32(dead_blkno);
2513 xfs_trans_log_buf(tp, sib_buf,
2514 XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
2515 sizeof(sib_info->forw)));
2516 sib_buf = NULL;
2517 }
2518 /*
2519 * If the moved block has a right sibling, fix up the pointers.
2520 */
2521 if ((sib_blkno = be32_to_cpu(dead_info->forw))) {
2522 error = xfs_da3_node_read(tp, dp, sib_blkno, &sib_buf, w);
2523 if (error)
2524 goto done;
2525 fa = xfs_da3_header_check(sib_buf, args->owner);
2526 if (fa) {
2527 __xfs_buf_mark_corrupt(sib_buf, fa);
2528 xfs_da_mark_sick(args);
2529 error = -EFSCORRUPTED;
2530 goto done;
2531 }
2532 sib_info = sib_buf->b_addr;
2533 if (XFS_IS_CORRUPT(mp,
2534 be32_to_cpu(sib_info->back) != last_blkno ||
2535 sib_info->magic != dead_info->magic)) {
2536 xfs_da_mark_sick(args);
2537 error = -EFSCORRUPTED;
2538 goto done;
2539 }
2540 sib_info->back = cpu_to_be32(dead_blkno);
2541 xfs_trans_log_buf(tp, sib_buf,
2542 XFS_DA_LOGRANGE(sib_info, &sib_info->back,
2543 sizeof(sib_info->back)));
2544 sib_buf = NULL;
2545 }
2546 par_blkno = args->geo->leafblk;
2547 level = -1;
2548 /*
2549 * Walk down the tree looking for the parent of the moved block.
2550 */
2551 for (;;) {
2552 error = xfs_da3_node_read(tp, dp, par_blkno, &par_buf, w);
2553 if (error)
2554 goto done;
2555 fa = xfs_da3_node_header_check(par_buf, args->owner);
2556 if (fa) {
2557 __xfs_buf_mark_corrupt(par_buf, fa);
2558 xfs_da_mark_sick(args);
2559 error = -EFSCORRUPTED;
2560 goto done;
2561 }
2562 par_node = par_buf->b_addr;
2563 xfs_da3_node_hdr_from_disk(dp->i_mount, &par_hdr, par_node);
2564 if (XFS_IS_CORRUPT(mp,
2565 level >= 0 && level != par_hdr.level + 1)) {
2566 xfs_da_mark_sick(args);
2567 error = -EFSCORRUPTED;
2568 goto done;
2569 }
2570 level = par_hdr.level;
2571 btree = par_hdr.btree;
2572 for (entno = 0;
2573 entno < par_hdr.count &&
2574 be32_to_cpu(btree[entno].hashval) < dead_hash;
2575 entno++)
2576 continue;
2577 if (XFS_IS_CORRUPT(mp, entno == par_hdr.count)) {
2578 xfs_da_mark_sick(args);
2579 error = -EFSCORRUPTED;
2580 goto done;
2581 }
2582 par_blkno = be32_to_cpu(btree[entno].before);
2583 if (level == dead_level + 1)
2584 break;
2585 xfs_trans_brelse(tp, par_buf);
2586 par_buf = NULL;
2587 }
2588 /*
2589 * We're in the right parent block.
2590 * Look for the right entry.
2591 */
2592 for (;;) {
2593 for (;
2594 entno < par_hdr.count &&
2595 be32_to_cpu(btree[entno].before) != last_blkno;
2596 entno++)
2597 continue;
2598 if (entno < par_hdr.count)
2599 break;
2600 par_blkno = par_hdr.forw;
2601 xfs_trans_brelse(tp, par_buf);
2602 par_buf = NULL;
2603 if (XFS_IS_CORRUPT(mp, par_blkno == 0)) {
2604 xfs_da_mark_sick(args);
2605 error = -EFSCORRUPTED;
2606 goto done;
2607 }
2608 error = xfs_da3_node_read(tp, dp, par_blkno, &par_buf, w);
2609 if (error)
2610 goto done;
2611 fa = xfs_da3_node_header_check(par_buf, args->owner);
2612 if (fa) {
2613 __xfs_buf_mark_corrupt(par_buf, fa);
2614 xfs_da_mark_sick(args);
2615 error = -EFSCORRUPTED;
2616 goto done;
2617 }
2618 par_node = par_buf->b_addr;
2619 xfs_da3_node_hdr_from_disk(dp->i_mount, &par_hdr, par_node);
2620 if (XFS_IS_CORRUPT(mp, par_hdr.level != level)) {
2621 xfs_da_mark_sick(args);
2622 error = -EFSCORRUPTED;
2623 goto done;
2624 }
2625 btree = par_hdr.btree;
2626 entno = 0;
2627 }
2628 /*
2629 * Update the parent entry pointing to the moved block.
2630 */
2631 btree[entno].before = cpu_to_be32(dead_blkno);
2632 xfs_trans_log_buf(tp, par_buf,
2633 XFS_DA_LOGRANGE(par_node, &btree[entno].before,
2634 sizeof(btree[entno].before)));
2635 *dead_blknop = last_blkno;
2636 *dead_bufp = last_buf;
2637 return 0;
2638done:
2639 if (par_buf)
2640 xfs_trans_brelse(tp, par_buf);
2641 if (sib_buf)
2642 xfs_trans_brelse(tp, sib_buf);
2643 xfs_trans_brelse(tp, last_buf);
2644 return error;
2645}
2646
2647/*
2648 * Remove a btree block from a directory or attribute.
2649 */
2650int
2651xfs_da_shrink_inode(
2652 struct xfs_da_args *args,
2653 xfs_dablk_t dead_blkno,
2654 struct xfs_buf *dead_buf)
2655{
2656 struct xfs_inode *dp;
2657 int done, error, w, count;
2658 struct xfs_trans *tp;
2659
2660 trace_xfs_da_shrink_inode(args);
2661
2662 dp = args->dp;
2663 w = args->whichfork;
2664 tp = args->trans;
2665 count = args->geo->fsbcount;
2666 for (;;) {
2667 /*
2668 * Remove extents. If we get ENOSPC for a dir we have to move
2669 * the last block to the place we want to kill.
2670 */
2671 error = xfs_bunmapi(tp, dp, dead_blkno, count,
2672 xfs_bmapi_aflag(w), 0, &done);
2673 if (error == -ENOSPC) {
2674 if (w != XFS_DATA_FORK)
2675 break;
2676 error = xfs_da3_swap_lastblock(args, &dead_blkno,
2677 &dead_buf);
2678 if (error)
2679 break;
2680 } else {
2681 break;
2682 }
2683 }
2684 xfs_trans_binval(tp, dead_buf);
2685 return error;
2686}
2687
2688static int
2689xfs_dabuf_map(
2690 struct xfs_inode *dp,
2691 xfs_dablk_t bno,
2692 unsigned int flags,
2693 int whichfork,
2694 struct xfs_buf_map **mapp,
2695 int *nmaps)
2696{
2697 struct xfs_mount *mp = dp->i_mount;
2698 int nfsb = xfs_dabuf_nfsb(mp, whichfork);
2699 struct xfs_bmbt_irec irec, *irecs = &irec;
2700 struct xfs_buf_map *map = *mapp;
2701 xfs_fileoff_t off = bno;
2702 int error = 0, nirecs, i;
2703
2704 if (nfsb > 1)
2705 irecs = kzalloc(sizeof(irec) * nfsb,
2706 GFP_KERNEL | __GFP_NOLOCKDEP | __GFP_NOFAIL);
2707
2708 nirecs = nfsb;
2709 error = xfs_bmapi_read(dp, bno, nfsb, irecs, &nirecs,
2710 xfs_bmapi_aflag(whichfork));
2711 if (error)
2712 goto out_free_irecs;
2713
2714 /*
2715 * Use the caller provided map for the single map case, else allocate a
2716 * larger one that needs to be free by the caller.
2717 */
2718 if (nirecs > 1) {
2719 map = kzalloc(nirecs * sizeof(struct xfs_buf_map),
2720 GFP_KERNEL | __GFP_NOLOCKDEP | __GFP_NOFAIL);
2721 if (!map) {
2722 error = -ENOMEM;
2723 goto out_free_irecs;
2724 }
2725 *mapp = map;
2726 }
2727
2728 for (i = 0; i < nirecs; i++) {
2729 if (irecs[i].br_startblock == HOLESTARTBLOCK ||
2730 irecs[i].br_startblock == DELAYSTARTBLOCK)
2731 goto invalid_mapping;
2732 if (off != irecs[i].br_startoff)
2733 goto invalid_mapping;
2734
2735 map[i].bm_bn = XFS_FSB_TO_DADDR(mp, irecs[i].br_startblock);
2736 map[i].bm_len = XFS_FSB_TO_BB(mp, irecs[i].br_blockcount);
2737 off += irecs[i].br_blockcount;
2738 }
2739
2740 if (off != bno + nfsb)
2741 goto invalid_mapping;
2742
2743 *nmaps = nirecs;
2744out_free_irecs:
2745 if (irecs != &irec)
2746 kfree(irecs);
2747 return error;
2748
2749invalid_mapping:
2750 /* Caller ok with no mapping. */
2751 if (XFS_IS_CORRUPT(mp, !(flags & XFS_DABUF_MAP_HOLE_OK))) {
2752 xfs_dirattr_mark_sick(dp, whichfork);
2753 error = -EFSCORRUPTED;
2754 if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
2755 xfs_alert(mp, "%s: bno %u inode %llu",
2756 __func__, bno, dp->i_ino);
2757
2758 for (i = 0; i < nirecs; i++) {
2759 xfs_alert(mp,
2760"[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d",
2761 i, irecs[i].br_startoff,
2762 irecs[i].br_startblock,
2763 irecs[i].br_blockcount,
2764 irecs[i].br_state);
2765 }
2766 }
2767 } else {
2768 *nmaps = 0;
2769 }
2770 goto out_free_irecs;
2771}
2772
2773/*
2774 * Get a buffer for the dir/attr block.
2775 */
2776int
2777xfs_da_get_buf(
2778 struct xfs_trans *tp,
2779 struct xfs_inode *dp,
2780 xfs_dablk_t bno,
2781 struct xfs_buf **bpp,
2782 int whichfork)
2783{
2784 struct xfs_mount *mp = dp->i_mount;
2785 struct xfs_buf *bp;
2786 struct xfs_buf_map map, *mapp = ↦
2787 int nmap = 1;
2788 int error;
2789
2790 *bpp = NULL;
2791 error = xfs_dabuf_map(dp, bno, 0, whichfork, &mapp, &nmap);
2792 if (error || nmap == 0)
2793 goto out_free;
2794
2795 error = xfs_trans_get_buf_map(tp, mp->m_ddev_targp, mapp, nmap, 0, &bp);
2796 if (error)
2797 goto out_free;
2798
2799 *bpp = bp;
2800
2801out_free:
2802 if (mapp != &map)
2803 kfree(mapp);
2804
2805 return error;
2806}
2807
2808/*
2809 * Get a buffer for the dir/attr block, fill in the contents.
2810 */
2811int
2812xfs_da_read_buf(
2813 struct xfs_trans *tp,
2814 struct xfs_inode *dp,
2815 xfs_dablk_t bno,
2816 unsigned int flags,
2817 struct xfs_buf **bpp,
2818 int whichfork,
2819 const struct xfs_buf_ops *ops)
2820{
2821 struct xfs_mount *mp = dp->i_mount;
2822 struct xfs_buf *bp;
2823 struct xfs_buf_map map, *mapp = ↦
2824 int nmap = 1;
2825 int error;
2826
2827 *bpp = NULL;
2828 error = xfs_dabuf_map(dp, bno, flags, whichfork, &mapp, &nmap);
2829 if (error || !nmap)
2830 goto out_free;
2831
2832 error = xfs_trans_read_buf_map(mp, tp, mp->m_ddev_targp, mapp, nmap, 0,
2833 &bp, ops);
2834 if (xfs_metadata_is_sick(error))
2835 xfs_dirattr_mark_sick(dp, whichfork);
2836 if (error)
2837 goto out_free;
2838
2839 if (whichfork == XFS_ATTR_FORK)
2840 xfs_buf_set_ref(bp, XFS_ATTR_BTREE_REF);
2841 else
2842 xfs_buf_set_ref(bp, XFS_DIR_BTREE_REF);
2843 *bpp = bp;
2844out_free:
2845 if (mapp != &map)
2846 kfree(mapp);
2847
2848 return error;
2849}
2850
2851/*
2852 * Readahead the dir/attr block.
2853 */
2854int
2855xfs_da_reada_buf(
2856 struct xfs_inode *dp,
2857 xfs_dablk_t bno,
2858 unsigned int flags,
2859 int whichfork,
2860 const struct xfs_buf_ops *ops)
2861{
2862 struct xfs_buf_map map;
2863 struct xfs_buf_map *mapp;
2864 int nmap;
2865 int error;
2866
2867 mapp = ↦
2868 nmap = 1;
2869 error = xfs_dabuf_map(dp, bno, flags, whichfork, &mapp, &nmap);
2870 if (error || !nmap)
2871 goto out_free;
2872
2873 xfs_buf_readahead_map(dp->i_mount->m_ddev_targp, mapp, nmap, ops);
2874
2875out_free:
2876 if (mapp != &map)
2877 kfree(mapp);
2878
2879 return error;
2880}