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 bool
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 false;
149
150 if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid))
151 return false;
152 if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
153 return false;
154 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->info.lsn)))
155 return false;
156 } else {
157 if (ichdr.magic != XFS_DA_NODE_MAGIC)
158 return false;
159 }
160 if (ichdr.level == 0)
161 return false;
162 if (ichdr.level > XFS_DA_NODE_MAXDEPTH)
163 return false;
164 if (ichdr.count == 0)
165 return false;
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 false;
174
175 /* XXX: hash order check? */
176
177 return true;
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_fspriv;
186 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
187
188 if (!xfs_da3_node_verify(bp)) {
189 xfs_buf_ioerror(bp, -EFSCORRUPTED);
190 xfs_verifier_error(bp);
191 return;
192 }
193
194 if (!xfs_sb_version_hascrc(&mp->m_sb))
195 return;
196
197 if (bip)
198 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
199
200 xfs_buf_update_cksum(bp, XFS_DA3_NODE_CRC_OFF);
201}
202
203/*
204 * leaf/node format detection on trees is sketchy, so a node read can be done on
205 * leaf level blocks when detection identifies the tree as a node format tree
206 * incorrectly. In this case, we need to swap the verifier to match the correct
207 * format of the block being read.
208 */
209static void
210xfs_da3_node_read_verify(
211 struct xfs_buf *bp)
212{
213 struct xfs_da_blkinfo *info = bp->b_addr;
214
215 switch (be16_to_cpu(info->magic)) {
216 case XFS_DA3_NODE_MAGIC:
217 if (!xfs_buf_verify_cksum(bp, XFS_DA3_NODE_CRC_OFF)) {
218 xfs_buf_ioerror(bp, -EFSBADCRC);
219 break;
220 }
221 /* fall through */
222 case XFS_DA_NODE_MAGIC:
223 if (!xfs_da3_node_verify(bp)) {
224 xfs_buf_ioerror(bp, -EFSCORRUPTED);
225 break;
226 }
227 return;
228 case XFS_ATTR_LEAF_MAGIC:
229 case XFS_ATTR3_LEAF_MAGIC:
230 bp->b_ops = &xfs_attr3_leaf_buf_ops;
231 bp->b_ops->verify_read(bp);
232 return;
233 case XFS_DIR2_LEAFN_MAGIC:
234 case XFS_DIR3_LEAFN_MAGIC:
235 bp->b_ops = &xfs_dir3_leafn_buf_ops;
236 bp->b_ops->verify_read(bp);
237 return;
238 default:
239 xfs_buf_ioerror(bp, -EFSCORRUPTED);
240 break;
241 }
242
243 /* corrupt block */
244 xfs_verifier_error(bp);
245}
246
247const struct xfs_buf_ops xfs_da3_node_buf_ops = {
248 .name = "xfs_da3_node",
249 .verify_read = xfs_da3_node_read_verify,
250 .verify_write = xfs_da3_node_write_verify,
251};
252
253int
254xfs_da3_node_read(
255 struct xfs_trans *tp,
256 struct xfs_inode *dp,
257 xfs_dablk_t bno,
258 xfs_daddr_t mappedbno,
259 struct xfs_buf **bpp,
260 int which_fork)
261{
262 int err;
263
264 err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
265 which_fork, &xfs_da3_node_buf_ops);
266 if (!err && tp) {
267 struct xfs_da_blkinfo *info = (*bpp)->b_addr;
268 int type;
269
270 switch (be16_to_cpu(info->magic)) {
271 case XFS_DA_NODE_MAGIC:
272 case XFS_DA3_NODE_MAGIC:
273 type = XFS_BLFT_DA_NODE_BUF;
274 break;
275 case XFS_ATTR_LEAF_MAGIC:
276 case XFS_ATTR3_LEAF_MAGIC:
277 type = XFS_BLFT_ATTR_LEAF_BUF;
278 break;
279 case XFS_DIR2_LEAFN_MAGIC:
280 case XFS_DIR3_LEAFN_MAGIC:
281 type = XFS_BLFT_DIR_LEAFN_BUF;
282 break;
283 default:
284 type = 0;
285 ASSERT(0);
286 break;
287 }
288 xfs_trans_buf_set_type(tp, *bpp, type);
289 }
290 return err;
291}
292
293/*========================================================================
294 * Routines used for growing the Btree.
295 *========================================================================*/
296
297/*
298 * Create the initial contents of an intermediate node.
299 */
300int
301xfs_da3_node_create(
302 struct xfs_da_args *args,
303 xfs_dablk_t blkno,
304 int level,
305 struct xfs_buf **bpp,
306 int whichfork)
307{
308 struct xfs_da_intnode *node;
309 struct xfs_trans *tp = args->trans;
310 struct xfs_mount *mp = tp->t_mountp;
311 struct xfs_da3_icnode_hdr ichdr = {0};
312 struct xfs_buf *bp;
313 int error;
314 struct xfs_inode *dp = args->dp;
315
316 trace_xfs_da_node_create(args);
317 ASSERT(level <= XFS_DA_NODE_MAXDEPTH);
318
319 error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, whichfork);
320 if (error)
321 return error;
322 bp->b_ops = &xfs_da3_node_buf_ops;
323 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
324 node = bp->b_addr;
325
326 if (xfs_sb_version_hascrc(&mp->m_sb)) {
327 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
328
329 memset(hdr3, 0, sizeof(struct xfs_da3_node_hdr));
330 ichdr.magic = XFS_DA3_NODE_MAGIC;
331 hdr3->info.blkno = cpu_to_be64(bp->b_bn);
332 hdr3->info.owner = cpu_to_be64(args->dp->i_ino);
333 uuid_copy(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid);
334 } else {
335 ichdr.magic = XFS_DA_NODE_MAGIC;
336 }
337 ichdr.level = level;
338
339 dp->d_ops->node_hdr_to_disk(node, &ichdr);
340 xfs_trans_log_buf(tp, bp,
341 XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
342
343 *bpp = bp;
344 return 0;
345}
346
347/*
348 * Split a leaf node, rebalance, then possibly split
349 * intermediate nodes, rebalance, etc.
350 */
351int /* error */
352xfs_da3_split(
353 struct xfs_da_state *state)
354{
355 struct xfs_da_state_blk *oldblk;
356 struct xfs_da_state_blk *newblk;
357 struct xfs_da_state_blk *addblk;
358 struct xfs_da_intnode *node;
359 int max;
360 int action = 0;
361 int error;
362 int i;
363
364 trace_xfs_da_split(state->args);
365
366 /*
367 * Walk back up the tree splitting/inserting/adjusting as necessary.
368 * If we need to insert and there isn't room, split the node, then
369 * decide which fragment to insert the new block from below into.
370 * Note that we may split the root this way, but we need more fixup.
371 */
372 max = state->path.active - 1;
373 ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH));
374 ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC ||
375 state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
376
377 addblk = &state->path.blk[max]; /* initial dummy value */
378 for (i = max; (i >= 0) && addblk; state->path.active--, i--) {
379 oldblk = &state->path.blk[i];
380 newblk = &state->altpath.blk[i];
381
382 /*
383 * If a leaf node then
384 * Allocate a new leaf node, then rebalance across them.
385 * else if an intermediate node then
386 * We split on the last layer, must we split the node?
387 */
388 switch (oldblk->magic) {
389 case XFS_ATTR_LEAF_MAGIC:
390 error = xfs_attr3_leaf_split(state, oldblk, newblk);
391 if ((error != 0) && (error != -ENOSPC)) {
392 return error; /* GROT: attr is inconsistent */
393 }
394 if (!error) {
395 addblk = newblk;
396 break;
397 }
398 /*
399 * Entry wouldn't fit, split the leaf again. The new
400 * extrablk will be consumed by xfs_da3_node_split if
401 * the node is split.
402 */
403 state->extravalid = 1;
404 if (state->inleaf) {
405 state->extraafter = 0; /* before newblk */
406 trace_xfs_attr_leaf_split_before(state->args);
407 error = xfs_attr3_leaf_split(state, oldblk,
408 &state->extrablk);
409 } else {
410 state->extraafter = 1; /* after newblk */
411 trace_xfs_attr_leaf_split_after(state->args);
412 error = xfs_attr3_leaf_split(state, newblk,
413 &state->extrablk);
414 }
415 if (error)
416 return error; /* GROT: attr inconsistent */
417 addblk = newblk;
418 break;
419 case XFS_DIR2_LEAFN_MAGIC:
420 error = xfs_dir2_leafn_split(state, oldblk, newblk);
421 if (error)
422 return error;
423 addblk = newblk;
424 break;
425 case XFS_DA_NODE_MAGIC:
426 error = xfs_da3_node_split(state, oldblk, newblk, addblk,
427 max - i, &action);
428 addblk->bp = NULL;
429 if (error)
430 return error; /* GROT: dir is inconsistent */
431 /*
432 * Record the newly split block for the next time thru?
433 */
434 if (action)
435 addblk = newblk;
436 else
437 addblk = NULL;
438 break;
439 }
440
441 /*
442 * Update the btree to show the new hashval for this child.
443 */
444 xfs_da3_fixhashpath(state, &state->path);
445 }
446 if (!addblk)
447 return 0;
448
449 /*
450 * xfs_da3_node_split() should have consumed any extra blocks we added
451 * during a double leaf split in the attr fork. This is guaranteed as
452 * we can't be here if the attr fork only has a single leaf block.
453 */
454 ASSERT(state->extravalid == 0 ||
455 state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
456
457 /*
458 * Split the root node.
459 */
460 ASSERT(state->path.active == 0);
461 oldblk = &state->path.blk[0];
462 error = xfs_da3_root_split(state, oldblk, addblk);
463 if (error) {
464 addblk->bp = NULL;
465 return error; /* GROT: dir is inconsistent */
466 }
467
468 /*
469 * Update pointers to the node which used to be block 0 and just got
470 * bumped because of the addition of a new root node. Note that the
471 * original block 0 could be at any position in the list of blocks in
472 * the tree.
473 *
474 * Note: the magic numbers and sibling pointers are in the same physical
475 * place for both v2 and v3 headers (by design). Hence it doesn't matter
476 * which version of the xfs_da_intnode structure we use here as the
477 * result will be the same using either structure.
478 */
479 node = oldblk->bp->b_addr;
480 if (node->hdr.info.forw) {
481 ASSERT(be32_to_cpu(node->hdr.info.forw) == addblk->blkno);
482 node = addblk->bp->b_addr;
483 node->hdr.info.back = cpu_to_be32(oldblk->blkno);
484 xfs_trans_log_buf(state->args->trans, addblk->bp,
485 XFS_DA_LOGRANGE(node, &node->hdr.info,
486 sizeof(node->hdr.info)));
487 }
488 node = oldblk->bp->b_addr;
489 if (node->hdr.info.back) {
490 ASSERT(be32_to_cpu(node->hdr.info.back) == addblk->blkno);
491 node = addblk->bp->b_addr;
492 node->hdr.info.forw = cpu_to_be32(oldblk->blkno);
493 xfs_trans_log_buf(state->args->trans, addblk->bp,
494 XFS_DA_LOGRANGE(node, &node->hdr.info,
495 sizeof(node->hdr.info)));
496 }
497 addblk->bp = NULL;
498 return 0;
499}
500
501/*
502 * Split the root. We have to create a new root and point to the two
503 * parts (the split old root) that we just created. Copy block zero to
504 * the EOF, extending the inode in process.
505 */
506STATIC int /* error */
507xfs_da3_root_split(
508 struct xfs_da_state *state,
509 struct xfs_da_state_blk *blk1,
510 struct xfs_da_state_blk *blk2)
511{
512 struct xfs_da_intnode *node;
513 struct xfs_da_intnode *oldroot;
514 struct xfs_da_node_entry *btree;
515 struct xfs_da3_icnode_hdr nodehdr;
516 struct xfs_da_args *args;
517 struct xfs_buf *bp;
518 struct xfs_inode *dp;
519 struct xfs_trans *tp;
520 struct xfs_dir2_leaf *leaf;
521 xfs_dablk_t blkno;
522 int level;
523 int error;
524 int size;
525
526 trace_xfs_da_root_split(state->args);
527
528 /*
529 * Copy the existing (incorrect) block from the root node position
530 * to a free space somewhere.
531 */
532 args = state->args;
533 error = xfs_da_grow_inode(args, &blkno);
534 if (error)
535 return error;
536
537 dp = args->dp;
538 tp = args->trans;
539 error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork);
540 if (error)
541 return error;
542 node = bp->b_addr;
543 oldroot = blk1->bp->b_addr;
544 if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
545 oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC)) {
546 struct xfs_da3_icnode_hdr icnodehdr;
547
548 dp->d_ops->node_hdr_from_disk(&icnodehdr, oldroot);
549 btree = dp->d_ops->node_tree_p(oldroot);
550 size = (int)((char *)&btree[icnodehdr.count] - (char *)oldroot);
551 level = icnodehdr.level;
552
553 /*
554 * we are about to copy oldroot to bp, so set up the type
555 * of bp while we know exactly what it will be.
556 */
557 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
558 } else {
559 struct xfs_dir3_icleaf_hdr leafhdr;
560 struct xfs_dir2_leaf_entry *ents;
561
562 leaf = (xfs_dir2_leaf_t *)oldroot;
563 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
564 ents = dp->d_ops->leaf_ents_p(leaf);
565
566 ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
567 leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
568 size = (int)((char *)&ents[leafhdr.count] - (char *)leaf);
569 level = 0;
570
571 /*
572 * we are about to copy oldroot to bp, so set up the type
573 * of bp while we know exactly what it will be.
574 */
575 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
576 }
577
578 /*
579 * we can copy most of the information in the node from one block to
580 * another, but for CRC enabled headers we have to make sure that the
581 * block specific identifiers are kept intact. We update the buffer
582 * directly for this.
583 */
584 memcpy(node, oldroot, size);
585 if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
586 oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
587 struct xfs_da3_intnode *node3 = (struct xfs_da3_intnode *)node;
588
589 node3->hdr.info.blkno = cpu_to_be64(bp->b_bn);
590 }
591 xfs_trans_log_buf(tp, bp, 0, size - 1);
592
593 bp->b_ops = blk1->bp->b_ops;
594 xfs_trans_buf_copy_type(bp, blk1->bp);
595 blk1->bp = bp;
596 blk1->blkno = blkno;
597
598 /*
599 * Set up the new root node.
600 */
601 error = xfs_da3_node_create(args,
602 (args->whichfork == XFS_DATA_FORK) ? args->geo->leafblk : 0,
603 level + 1, &bp, args->whichfork);
604 if (error)
605 return error;
606
607 node = bp->b_addr;
608 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
609 btree = dp->d_ops->node_tree_p(node);
610 btree[0].hashval = cpu_to_be32(blk1->hashval);
611 btree[0].before = cpu_to_be32(blk1->blkno);
612 btree[1].hashval = cpu_to_be32(blk2->hashval);
613 btree[1].before = cpu_to_be32(blk2->blkno);
614 nodehdr.count = 2;
615 dp->d_ops->node_hdr_to_disk(node, &nodehdr);
616
617#ifdef DEBUG
618 if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
619 oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
620 ASSERT(blk1->blkno >= args->geo->leafblk &&
621 blk1->blkno < args->geo->freeblk);
622 ASSERT(blk2->blkno >= args->geo->leafblk &&
623 blk2->blkno < args->geo->freeblk);
624 }
625#endif
626
627 /* Header is already logged by xfs_da_node_create */
628 xfs_trans_log_buf(tp, bp,
629 XFS_DA_LOGRANGE(node, btree, sizeof(xfs_da_node_entry_t) * 2));
630
631 return 0;
632}
633
634/*
635 * Split the node, rebalance, then add the new entry.
636 */
637STATIC int /* error */
638xfs_da3_node_split(
639 struct xfs_da_state *state,
640 struct xfs_da_state_blk *oldblk,
641 struct xfs_da_state_blk *newblk,
642 struct xfs_da_state_blk *addblk,
643 int treelevel,
644 int *result)
645{
646 struct xfs_da_intnode *node;
647 struct xfs_da3_icnode_hdr nodehdr;
648 xfs_dablk_t blkno;
649 int newcount;
650 int error;
651 int useextra;
652 struct xfs_inode *dp = state->args->dp;
653
654 trace_xfs_da_node_split(state->args);
655
656 node = oldblk->bp->b_addr;
657 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
658
659 /*
660 * With V2 dirs the extra block is data or freespace.
661 */
662 useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK;
663 newcount = 1 + useextra;
664 /*
665 * Do we have to split the node?
666 */
667 if (nodehdr.count + newcount > state->args->geo->node_ents) {
668 /*
669 * Allocate a new node, add to the doubly linked chain of
670 * nodes, then move some of our excess entries into it.
671 */
672 error = xfs_da_grow_inode(state->args, &blkno);
673 if (error)
674 return error; /* GROT: dir is inconsistent */
675
676 error = xfs_da3_node_create(state->args, blkno, treelevel,
677 &newblk->bp, state->args->whichfork);
678 if (error)
679 return error; /* GROT: dir is inconsistent */
680 newblk->blkno = blkno;
681 newblk->magic = XFS_DA_NODE_MAGIC;
682 xfs_da3_node_rebalance(state, oldblk, newblk);
683 error = xfs_da3_blk_link(state, oldblk, newblk);
684 if (error)
685 return error;
686 *result = 1;
687 } else {
688 *result = 0;
689 }
690
691 /*
692 * Insert the new entry(s) into the correct block
693 * (updating last hashval in the process).
694 *
695 * xfs_da3_node_add() inserts BEFORE the given index,
696 * and as a result of using node_lookup_int() we always
697 * point to a valid entry (not after one), but a split
698 * operation always results in a new block whose hashvals
699 * FOLLOW the current block.
700 *
701 * If we had double-split op below us, then add the extra block too.
702 */
703 node = oldblk->bp->b_addr;
704 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
705 if (oldblk->index <= nodehdr.count) {
706 oldblk->index++;
707 xfs_da3_node_add(state, oldblk, addblk);
708 if (useextra) {
709 if (state->extraafter)
710 oldblk->index++;
711 xfs_da3_node_add(state, oldblk, &state->extrablk);
712 state->extravalid = 0;
713 }
714 } else {
715 newblk->index++;
716 xfs_da3_node_add(state, newblk, addblk);
717 if (useextra) {
718 if (state->extraafter)
719 newblk->index++;
720 xfs_da3_node_add(state, newblk, &state->extrablk);
721 state->extravalid = 0;
722 }
723 }
724
725 return 0;
726}
727
728/*
729 * Balance the btree elements between two intermediate nodes,
730 * usually one full and one empty.
731 *
732 * NOTE: if blk2 is empty, then it will get the upper half of blk1.
733 */
734STATIC void
735xfs_da3_node_rebalance(
736 struct xfs_da_state *state,
737 struct xfs_da_state_blk *blk1,
738 struct xfs_da_state_blk *blk2)
739{
740 struct xfs_da_intnode *node1;
741 struct xfs_da_intnode *node2;
742 struct xfs_da_intnode *tmpnode;
743 struct xfs_da_node_entry *btree1;
744 struct xfs_da_node_entry *btree2;
745 struct xfs_da_node_entry *btree_s;
746 struct xfs_da_node_entry *btree_d;
747 struct xfs_da3_icnode_hdr nodehdr1;
748 struct xfs_da3_icnode_hdr nodehdr2;
749 struct xfs_trans *tp;
750 int count;
751 int tmp;
752 int swap = 0;
753 struct xfs_inode *dp = state->args->dp;
754
755 trace_xfs_da_node_rebalance(state->args);
756
757 node1 = blk1->bp->b_addr;
758 node2 = blk2->bp->b_addr;
759 dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
760 dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
761 btree1 = dp->d_ops->node_tree_p(node1);
762 btree2 = dp->d_ops->node_tree_p(node2);
763
764 /*
765 * Figure out how many entries need to move, and in which direction.
766 * Swap the nodes around if that makes it simpler.
767 */
768 if (nodehdr1.count > 0 && nodehdr2.count > 0 &&
769 ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
770 (be32_to_cpu(btree2[nodehdr2.count - 1].hashval) <
771 be32_to_cpu(btree1[nodehdr1.count - 1].hashval)))) {
772 tmpnode = node1;
773 node1 = node2;
774 node2 = tmpnode;
775 dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
776 dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
777 btree1 = dp->d_ops->node_tree_p(node1);
778 btree2 = dp->d_ops->node_tree_p(node2);
779 swap = 1;
780 }
781
782 count = (nodehdr1.count - nodehdr2.count) / 2;
783 if (count == 0)
784 return;
785 tp = state->args->trans;
786 /*
787 * Two cases: high-to-low and low-to-high.
788 */
789 if (count > 0) {
790 /*
791 * Move elements in node2 up to make a hole.
792 */
793 tmp = nodehdr2.count;
794 if (tmp > 0) {
795 tmp *= (uint)sizeof(xfs_da_node_entry_t);
796 btree_s = &btree2[0];
797 btree_d = &btree2[count];
798 memmove(btree_d, btree_s, tmp);
799 }
800
801 /*
802 * Move the req'd B-tree elements from high in node1 to
803 * low in node2.
804 */
805 nodehdr2.count += count;
806 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
807 btree_s = &btree1[nodehdr1.count - count];
808 btree_d = &btree2[0];
809 memcpy(btree_d, btree_s, tmp);
810 nodehdr1.count -= count;
811 } else {
812 /*
813 * Move the req'd B-tree elements from low in node2 to
814 * high in node1.
815 */
816 count = -count;
817 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
818 btree_s = &btree2[0];
819 btree_d = &btree1[nodehdr1.count];
820 memcpy(btree_d, btree_s, tmp);
821 nodehdr1.count += count;
822
823 xfs_trans_log_buf(tp, blk1->bp,
824 XFS_DA_LOGRANGE(node1, btree_d, tmp));
825
826 /*
827 * Move elements in node2 down to fill the hole.
828 */
829 tmp = nodehdr2.count - count;
830 tmp *= (uint)sizeof(xfs_da_node_entry_t);
831 btree_s = &btree2[count];
832 btree_d = &btree2[0];
833 memmove(btree_d, btree_s, tmp);
834 nodehdr2.count -= count;
835 }
836
837 /*
838 * Log header of node 1 and all current bits of node 2.
839 */
840 dp->d_ops->node_hdr_to_disk(node1, &nodehdr1);
841 xfs_trans_log_buf(tp, blk1->bp,
842 XFS_DA_LOGRANGE(node1, &node1->hdr, dp->d_ops->node_hdr_size));
843
844 dp->d_ops->node_hdr_to_disk(node2, &nodehdr2);
845 xfs_trans_log_buf(tp, blk2->bp,
846 XFS_DA_LOGRANGE(node2, &node2->hdr,
847 dp->d_ops->node_hdr_size +
848 (sizeof(btree2[0]) * nodehdr2.count)));
849
850 /*
851 * Record the last hashval from each block for upward propagation.
852 * (note: don't use the swapped node pointers)
853 */
854 if (swap) {
855 node1 = blk1->bp->b_addr;
856 node2 = blk2->bp->b_addr;
857 dp->d_ops->node_hdr_from_disk(&nodehdr1, node1);
858 dp->d_ops->node_hdr_from_disk(&nodehdr2, node2);
859 btree1 = dp->d_ops->node_tree_p(node1);
860 btree2 = dp->d_ops->node_tree_p(node2);
861 }
862 blk1->hashval = be32_to_cpu(btree1[nodehdr1.count - 1].hashval);
863 blk2->hashval = be32_to_cpu(btree2[nodehdr2.count - 1].hashval);
864
865 /*
866 * Adjust the expected index for insertion.
867 */
868 if (blk1->index >= nodehdr1.count) {
869 blk2->index = blk1->index - nodehdr1.count;
870 blk1->index = nodehdr1.count + 1; /* make it invalid */
871 }
872}
873
874/*
875 * Add a new entry to an intermediate node.
876 */
877STATIC void
878xfs_da3_node_add(
879 struct xfs_da_state *state,
880 struct xfs_da_state_blk *oldblk,
881 struct xfs_da_state_blk *newblk)
882{
883 struct xfs_da_intnode *node;
884 struct xfs_da3_icnode_hdr nodehdr;
885 struct xfs_da_node_entry *btree;
886 int tmp;
887 struct xfs_inode *dp = state->args->dp;
888
889 trace_xfs_da_node_add(state->args);
890
891 node = oldblk->bp->b_addr;
892 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
893 btree = dp->d_ops->node_tree_p(node);
894
895 ASSERT(oldblk->index >= 0 && oldblk->index <= nodehdr.count);
896 ASSERT(newblk->blkno != 0);
897 if (state->args->whichfork == XFS_DATA_FORK)
898 ASSERT(newblk->blkno >= state->args->geo->leafblk &&
899 newblk->blkno < state->args->geo->freeblk);
900
901 /*
902 * We may need to make some room before we insert the new node.
903 */
904 tmp = 0;
905 if (oldblk->index < nodehdr.count) {
906 tmp = (nodehdr.count - oldblk->index) * (uint)sizeof(*btree);
907 memmove(&btree[oldblk->index + 1], &btree[oldblk->index], tmp);
908 }
909 btree[oldblk->index].hashval = cpu_to_be32(newblk->hashval);
910 btree[oldblk->index].before = cpu_to_be32(newblk->blkno);
911 xfs_trans_log_buf(state->args->trans, oldblk->bp,
912 XFS_DA_LOGRANGE(node, &btree[oldblk->index],
913 tmp + sizeof(*btree)));
914
915 nodehdr.count += 1;
916 dp->d_ops->node_hdr_to_disk(node, &nodehdr);
917 xfs_trans_log_buf(state->args->trans, oldblk->bp,
918 XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
919
920 /*
921 * Copy the last hash value from the oldblk to propagate upwards.
922 */
923 oldblk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
924}
925
926/*========================================================================
927 * Routines used for shrinking the Btree.
928 *========================================================================*/
929
930/*
931 * Deallocate an empty leaf node, remove it from its parent,
932 * possibly deallocating that block, etc...
933 */
934int
935xfs_da3_join(
936 struct xfs_da_state *state)
937{
938 struct xfs_da_state_blk *drop_blk;
939 struct xfs_da_state_blk *save_blk;
940 int action = 0;
941 int error;
942
943 trace_xfs_da_join(state->args);
944
945 drop_blk = &state->path.blk[ state->path.active-1 ];
946 save_blk = &state->altpath.blk[ state->path.active-1 ];
947 ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
948 ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
949 drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
950
951 /*
952 * Walk back up the tree joining/deallocating as necessary.
953 * When we stop dropping blocks, break out.
954 */
955 for ( ; state->path.active >= 2; drop_blk--, save_blk--,
956 state->path.active--) {
957 /*
958 * See if we can combine the block with a neighbor.
959 * (action == 0) => no options, just leave
960 * (action == 1) => coalesce, then unlink
961 * (action == 2) => block empty, unlink it
962 */
963 switch (drop_blk->magic) {
964 case XFS_ATTR_LEAF_MAGIC:
965 error = xfs_attr3_leaf_toosmall(state, &action);
966 if (error)
967 return error;
968 if (action == 0)
969 return 0;
970 xfs_attr3_leaf_unbalance(state, drop_blk, save_blk);
971 break;
972 case XFS_DIR2_LEAFN_MAGIC:
973 error = xfs_dir2_leafn_toosmall(state, &action);
974 if (error)
975 return error;
976 if (action == 0)
977 return 0;
978 xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
979 break;
980 case XFS_DA_NODE_MAGIC:
981 /*
982 * Remove the offending node, fixup hashvals,
983 * check for a toosmall neighbor.
984 */
985 xfs_da3_node_remove(state, drop_blk);
986 xfs_da3_fixhashpath(state, &state->path);
987 error = xfs_da3_node_toosmall(state, &action);
988 if (error)
989 return error;
990 if (action == 0)
991 return 0;
992 xfs_da3_node_unbalance(state, drop_blk, save_blk);
993 break;
994 }
995 xfs_da3_fixhashpath(state, &state->altpath);
996 error = xfs_da3_blk_unlink(state, drop_blk, save_blk);
997 xfs_da_state_kill_altpath(state);
998 if (error)
999 return error;
1000 error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
1001 drop_blk->bp);
1002 drop_blk->bp = NULL;
1003 if (error)
1004 return error;
1005 }
1006 /*
1007 * We joined all the way to the top. If it turns out that
1008 * we only have one entry in the root, make the child block
1009 * the new root.
1010 */
1011 xfs_da3_node_remove(state, drop_blk);
1012 xfs_da3_fixhashpath(state, &state->path);
1013 error = xfs_da3_root_join(state, &state->path.blk[0]);
1014 return error;
1015}
1016
1017#ifdef DEBUG
1018static void
1019xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level)
1020{
1021 __be16 magic = blkinfo->magic;
1022
1023 if (level == 1) {
1024 ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1025 magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
1026 magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
1027 magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
1028 } else {
1029 ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1030 magic == cpu_to_be16(XFS_DA3_NODE_MAGIC));
1031 }
1032 ASSERT(!blkinfo->forw);
1033 ASSERT(!blkinfo->back);
1034}
1035#else /* !DEBUG */
1036#define xfs_da_blkinfo_onlychild_validate(blkinfo, level)
1037#endif /* !DEBUG */
1038
1039/*
1040 * We have only one entry in the root. Copy the only remaining child of
1041 * the old root to block 0 as the new root node.
1042 */
1043STATIC int
1044xfs_da3_root_join(
1045 struct xfs_da_state *state,
1046 struct xfs_da_state_blk *root_blk)
1047{
1048 struct xfs_da_intnode *oldroot;
1049 struct xfs_da_args *args;
1050 xfs_dablk_t child;
1051 struct xfs_buf *bp;
1052 struct xfs_da3_icnode_hdr oldroothdr;
1053 struct xfs_da_node_entry *btree;
1054 int error;
1055 struct xfs_inode *dp = state->args->dp;
1056
1057 trace_xfs_da_root_join(state->args);
1058
1059 ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
1060
1061 args = state->args;
1062 oldroot = root_blk->bp->b_addr;
1063 dp->d_ops->node_hdr_from_disk(&oldroothdr, oldroot);
1064 ASSERT(oldroothdr.forw == 0);
1065 ASSERT(oldroothdr.back == 0);
1066
1067 /*
1068 * If the root has more than one child, then don't do anything.
1069 */
1070 if (oldroothdr.count > 1)
1071 return 0;
1072
1073 /*
1074 * Read in the (only) child block, then copy those bytes into
1075 * the root block's buffer and free the original child block.
1076 */
1077 btree = dp->d_ops->node_tree_p(oldroot);
1078 child = be32_to_cpu(btree[0].before);
1079 ASSERT(child != 0);
1080 error = xfs_da3_node_read(args->trans, dp, child, -1, &bp,
1081 args->whichfork);
1082 if (error)
1083 return error;
1084 xfs_da_blkinfo_onlychild_validate(bp->b_addr, oldroothdr.level);
1085
1086 /*
1087 * This could be copying a leaf back into the root block in the case of
1088 * there only being a single leaf block left in the tree. Hence we have
1089 * to update the b_ops pointer as well to match the buffer type change
1090 * that could occur. For dir3 blocks we also need to update the block
1091 * number in the buffer header.
1092 */
1093 memcpy(root_blk->bp->b_addr, bp->b_addr, args->geo->blksize);
1094 root_blk->bp->b_ops = bp->b_ops;
1095 xfs_trans_buf_copy_type(root_blk->bp, bp);
1096 if (oldroothdr.magic == XFS_DA3_NODE_MAGIC) {
1097 struct xfs_da3_blkinfo *da3 = root_blk->bp->b_addr;
1098 da3->blkno = cpu_to_be64(root_blk->bp->b_bn);
1099 }
1100 xfs_trans_log_buf(args->trans, root_blk->bp, 0,
1101 args->geo->blksize - 1);
1102 error = xfs_da_shrink_inode(args, child, bp);
1103 return error;
1104}
1105
1106/*
1107 * Check a node block and its neighbors to see if the block should be
1108 * collapsed into one or the other neighbor. Always keep the block
1109 * with the smaller block number.
1110 * If the current block is over 50% full, don't try to join it, return 0.
1111 * If the block is empty, fill in the state structure and return 2.
1112 * If it can be collapsed, fill in the state structure and return 1.
1113 * If nothing can be done, return 0.
1114 */
1115STATIC int
1116xfs_da3_node_toosmall(
1117 struct xfs_da_state *state,
1118 int *action)
1119{
1120 struct xfs_da_intnode *node;
1121 struct xfs_da_state_blk *blk;
1122 struct xfs_da_blkinfo *info;
1123 xfs_dablk_t blkno;
1124 struct xfs_buf *bp;
1125 struct xfs_da3_icnode_hdr nodehdr;
1126 int count;
1127 int forward;
1128 int error;
1129 int retval;
1130 int i;
1131 struct xfs_inode *dp = state->args->dp;
1132
1133 trace_xfs_da_node_toosmall(state->args);
1134
1135 /*
1136 * Check for the degenerate case of the block being over 50% full.
1137 * If so, it's not worth even looking to see if we might be able
1138 * to coalesce with a sibling.
1139 */
1140 blk = &state->path.blk[ state->path.active-1 ];
1141 info = blk->bp->b_addr;
1142 node = (xfs_da_intnode_t *)info;
1143 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1144 if (nodehdr.count > (state->args->geo->node_ents >> 1)) {
1145 *action = 0; /* blk over 50%, don't try to join */
1146 return 0; /* blk over 50%, don't try to join */
1147 }
1148
1149 /*
1150 * Check for the degenerate case of the block being empty.
1151 * If the block is empty, we'll simply delete it, no need to
1152 * coalesce it with a sibling block. We choose (arbitrarily)
1153 * to merge with the forward block unless it is NULL.
1154 */
1155 if (nodehdr.count == 0) {
1156 /*
1157 * Make altpath point to the block we want to keep and
1158 * path point to the block we want to drop (this one).
1159 */
1160 forward = (info->forw != 0);
1161 memcpy(&state->altpath, &state->path, sizeof(state->path));
1162 error = xfs_da3_path_shift(state, &state->altpath, forward,
1163 0, &retval);
1164 if (error)
1165 return error;
1166 if (retval) {
1167 *action = 0;
1168 } else {
1169 *action = 2;
1170 }
1171 return 0;
1172 }
1173
1174 /*
1175 * Examine each sibling block to see if we can coalesce with
1176 * at least 25% free space to spare. We need to figure out
1177 * whether to merge with the forward or the backward block.
1178 * We prefer coalescing with the lower numbered sibling so as
1179 * to shrink a directory over time.
1180 */
1181 count = state->args->geo->node_ents;
1182 count -= state->args->geo->node_ents >> 2;
1183 count -= nodehdr.count;
1184
1185 /* start with smaller blk num */
1186 forward = nodehdr.forw < nodehdr.back;
1187 for (i = 0; i < 2; forward = !forward, i++) {
1188 struct xfs_da3_icnode_hdr thdr;
1189 if (forward)
1190 blkno = nodehdr.forw;
1191 else
1192 blkno = nodehdr.back;
1193 if (blkno == 0)
1194 continue;
1195 error = xfs_da3_node_read(state->args->trans, dp,
1196 blkno, -1, &bp, state->args->whichfork);
1197 if (error)
1198 return error;
1199
1200 node = bp->b_addr;
1201 dp->d_ops->node_hdr_from_disk(&thdr, node);
1202 xfs_trans_brelse(state->args->trans, bp);
1203
1204 if (count - thdr.count >= 0)
1205 break; /* fits with at least 25% to spare */
1206 }
1207 if (i >= 2) {
1208 *action = 0;
1209 return 0;
1210 }
1211
1212 /*
1213 * Make altpath point to the block we want to keep (the lower
1214 * numbered block) and path point to the block we want to drop.
1215 */
1216 memcpy(&state->altpath, &state->path, sizeof(state->path));
1217 if (blkno < blk->blkno) {
1218 error = xfs_da3_path_shift(state, &state->altpath, forward,
1219 0, &retval);
1220 } else {
1221 error = xfs_da3_path_shift(state, &state->path, forward,
1222 0, &retval);
1223 }
1224 if (error)
1225 return error;
1226 if (retval) {
1227 *action = 0;
1228 return 0;
1229 }
1230 *action = 1;
1231 return 0;
1232}
1233
1234/*
1235 * Pick up the last hashvalue from an intermediate node.
1236 */
1237STATIC uint
1238xfs_da3_node_lasthash(
1239 struct xfs_inode *dp,
1240 struct xfs_buf *bp,
1241 int *count)
1242{
1243 struct xfs_da_intnode *node;
1244 struct xfs_da_node_entry *btree;
1245 struct xfs_da3_icnode_hdr nodehdr;
1246
1247 node = bp->b_addr;
1248 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1249 if (count)
1250 *count = nodehdr.count;
1251 if (!nodehdr.count)
1252 return 0;
1253 btree = dp->d_ops->node_tree_p(node);
1254 return be32_to_cpu(btree[nodehdr.count - 1].hashval);
1255}
1256
1257/*
1258 * Walk back up the tree adjusting hash values as necessary,
1259 * when we stop making changes, return.
1260 */
1261void
1262xfs_da3_fixhashpath(
1263 struct xfs_da_state *state,
1264 struct xfs_da_state_path *path)
1265{
1266 struct xfs_da_state_blk *blk;
1267 struct xfs_da_intnode *node;
1268 struct xfs_da_node_entry *btree;
1269 xfs_dahash_t lasthash=0;
1270 int level;
1271 int count;
1272 struct xfs_inode *dp = state->args->dp;
1273
1274 trace_xfs_da_fixhashpath(state->args);
1275
1276 level = path->active-1;
1277 blk = &path->blk[ level ];
1278 switch (blk->magic) {
1279 case XFS_ATTR_LEAF_MAGIC:
1280 lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
1281 if (count == 0)
1282 return;
1283 break;
1284 case XFS_DIR2_LEAFN_MAGIC:
1285 lasthash = xfs_dir2_leafn_lasthash(dp, blk->bp, &count);
1286 if (count == 0)
1287 return;
1288 break;
1289 case XFS_DA_NODE_MAGIC:
1290 lasthash = xfs_da3_node_lasthash(dp, blk->bp, &count);
1291 if (count == 0)
1292 return;
1293 break;
1294 }
1295 for (blk--, level--; level >= 0; blk--, level--) {
1296 struct xfs_da3_icnode_hdr nodehdr;
1297
1298 node = blk->bp->b_addr;
1299 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1300 btree = dp->d_ops->node_tree_p(node);
1301 if (be32_to_cpu(btree[blk->index].hashval) == lasthash)
1302 break;
1303 blk->hashval = lasthash;
1304 btree[blk->index].hashval = cpu_to_be32(lasthash);
1305 xfs_trans_log_buf(state->args->trans, blk->bp,
1306 XFS_DA_LOGRANGE(node, &btree[blk->index],
1307 sizeof(*btree)));
1308
1309 lasthash = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1310 }
1311}
1312
1313/*
1314 * Remove an entry from an intermediate node.
1315 */
1316STATIC void
1317xfs_da3_node_remove(
1318 struct xfs_da_state *state,
1319 struct xfs_da_state_blk *drop_blk)
1320{
1321 struct xfs_da_intnode *node;
1322 struct xfs_da3_icnode_hdr nodehdr;
1323 struct xfs_da_node_entry *btree;
1324 int index;
1325 int tmp;
1326 struct xfs_inode *dp = state->args->dp;
1327
1328 trace_xfs_da_node_remove(state->args);
1329
1330 node = drop_blk->bp->b_addr;
1331 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1332 ASSERT(drop_blk->index < nodehdr.count);
1333 ASSERT(drop_blk->index >= 0);
1334
1335 /*
1336 * Copy over the offending entry, or just zero it out.
1337 */
1338 index = drop_blk->index;
1339 btree = dp->d_ops->node_tree_p(node);
1340 if (index < nodehdr.count - 1) {
1341 tmp = nodehdr.count - index - 1;
1342 tmp *= (uint)sizeof(xfs_da_node_entry_t);
1343 memmove(&btree[index], &btree[index + 1], tmp);
1344 xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1345 XFS_DA_LOGRANGE(node, &btree[index], tmp));
1346 index = nodehdr.count - 1;
1347 }
1348 memset(&btree[index], 0, sizeof(xfs_da_node_entry_t));
1349 xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1350 XFS_DA_LOGRANGE(node, &btree[index], sizeof(btree[index])));
1351 nodehdr.count -= 1;
1352 dp->d_ops->node_hdr_to_disk(node, &nodehdr);
1353 xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1354 XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size));
1355
1356 /*
1357 * Copy the last hash value from the block to propagate upwards.
1358 */
1359 drop_blk->hashval = be32_to_cpu(btree[index - 1].hashval);
1360}
1361
1362/*
1363 * Unbalance the elements between two intermediate nodes,
1364 * move all Btree elements from one node into another.
1365 */
1366STATIC void
1367xfs_da3_node_unbalance(
1368 struct xfs_da_state *state,
1369 struct xfs_da_state_blk *drop_blk,
1370 struct xfs_da_state_blk *save_blk)
1371{
1372 struct xfs_da_intnode *drop_node;
1373 struct xfs_da_intnode *save_node;
1374 struct xfs_da_node_entry *drop_btree;
1375 struct xfs_da_node_entry *save_btree;
1376 struct xfs_da3_icnode_hdr drop_hdr;
1377 struct xfs_da3_icnode_hdr save_hdr;
1378 struct xfs_trans *tp;
1379 int sindex;
1380 int tmp;
1381 struct xfs_inode *dp = state->args->dp;
1382
1383 trace_xfs_da_node_unbalance(state->args);
1384
1385 drop_node = drop_blk->bp->b_addr;
1386 save_node = save_blk->bp->b_addr;
1387 dp->d_ops->node_hdr_from_disk(&drop_hdr, drop_node);
1388 dp->d_ops->node_hdr_from_disk(&save_hdr, save_node);
1389 drop_btree = dp->d_ops->node_tree_p(drop_node);
1390 save_btree = dp->d_ops->node_tree_p(save_node);
1391 tp = state->args->trans;
1392
1393 /*
1394 * If the dying block has lower hashvals, then move all the
1395 * elements in the remaining block up to make a hole.
1396 */
1397 if ((be32_to_cpu(drop_btree[0].hashval) <
1398 be32_to_cpu(save_btree[0].hashval)) ||
1399 (be32_to_cpu(drop_btree[drop_hdr.count - 1].hashval) <
1400 be32_to_cpu(save_btree[save_hdr.count - 1].hashval))) {
1401 /* XXX: check this - is memmove dst correct? */
1402 tmp = save_hdr.count * sizeof(xfs_da_node_entry_t);
1403 memmove(&save_btree[drop_hdr.count], &save_btree[0], tmp);
1404
1405 sindex = 0;
1406 xfs_trans_log_buf(tp, save_blk->bp,
1407 XFS_DA_LOGRANGE(save_node, &save_btree[0],
1408 (save_hdr.count + drop_hdr.count) *
1409 sizeof(xfs_da_node_entry_t)));
1410 } else {
1411 sindex = save_hdr.count;
1412 xfs_trans_log_buf(tp, save_blk->bp,
1413 XFS_DA_LOGRANGE(save_node, &save_btree[sindex],
1414 drop_hdr.count * sizeof(xfs_da_node_entry_t)));
1415 }
1416
1417 /*
1418 * Move all the B-tree elements from drop_blk to save_blk.
1419 */
1420 tmp = drop_hdr.count * (uint)sizeof(xfs_da_node_entry_t);
1421 memcpy(&save_btree[sindex], &drop_btree[0], tmp);
1422 save_hdr.count += drop_hdr.count;
1423
1424 dp->d_ops->node_hdr_to_disk(save_node, &save_hdr);
1425 xfs_trans_log_buf(tp, save_blk->bp,
1426 XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1427 dp->d_ops->node_hdr_size));
1428
1429 /*
1430 * Save the last hashval in the remaining block for upward propagation.
1431 */
1432 save_blk->hashval = be32_to_cpu(save_btree[save_hdr.count - 1].hashval);
1433}
1434
1435/*========================================================================
1436 * Routines used for finding things in the Btree.
1437 *========================================================================*/
1438
1439/*
1440 * Walk down the Btree looking for a particular filename, filling
1441 * in the state structure as we go.
1442 *
1443 * We will set the state structure to point to each of the elements
1444 * in each of the nodes where either the hashval is or should be.
1445 *
1446 * We support duplicate hashval's so for each entry in the current
1447 * node that could contain the desired hashval, descend. This is a
1448 * pruned depth-first tree search.
1449 */
1450int /* error */
1451xfs_da3_node_lookup_int(
1452 struct xfs_da_state *state,
1453 int *result)
1454{
1455 struct xfs_da_state_blk *blk;
1456 struct xfs_da_blkinfo *curr;
1457 struct xfs_da_intnode *node;
1458 struct xfs_da_node_entry *btree;
1459 struct xfs_da3_icnode_hdr nodehdr;
1460 struct xfs_da_args *args;
1461 xfs_dablk_t blkno;
1462 xfs_dahash_t hashval;
1463 xfs_dahash_t btreehashval;
1464 int probe;
1465 int span;
1466 int max;
1467 int error;
1468 int retval;
1469 struct xfs_inode *dp = state->args->dp;
1470
1471 args = state->args;
1472
1473 /*
1474 * Descend thru the B-tree searching each level for the right
1475 * node to use, until the right hashval is found.
1476 */
1477 blkno = (args->whichfork == XFS_DATA_FORK)? args->geo->leafblk : 0;
1478 for (blk = &state->path.blk[0], state->path.active = 1;
1479 state->path.active <= XFS_DA_NODE_MAXDEPTH;
1480 blk++, state->path.active++) {
1481 /*
1482 * Read the next node down in the tree.
1483 */
1484 blk->blkno = blkno;
1485 error = xfs_da3_node_read(args->trans, args->dp, blkno,
1486 -1, &blk->bp, args->whichfork);
1487 if (error) {
1488 blk->blkno = 0;
1489 state->path.active--;
1490 return error;
1491 }
1492 curr = blk->bp->b_addr;
1493 blk->magic = be16_to_cpu(curr->magic);
1494
1495 if (blk->magic == XFS_ATTR_LEAF_MAGIC ||
1496 blk->magic == XFS_ATTR3_LEAF_MAGIC) {
1497 blk->magic = XFS_ATTR_LEAF_MAGIC;
1498 blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1499 break;
1500 }
1501
1502 if (blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1503 blk->magic == XFS_DIR3_LEAFN_MAGIC) {
1504 blk->magic = XFS_DIR2_LEAFN_MAGIC;
1505 blk->hashval = xfs_dir2_leafn_lasthash(args->dp,
1506 blk->bp, NULL);
1507 break;
1508 }
1509
1510 blk->magic = XFS_DA_NODE_MAGIC;
1511
1512
1513 /*
1514 * Search an intermediate node for a match.
1515 */
1516 node = blk->bp->b_addr;
1517 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1518 btree = dp->d_ops->node_tree_p(node);
1519
1520 max = nodehdr.count;
1521 blk->hashval = be32_to_cpu(btree[max - 1].hashval);
1522
1523 /*
1524 * Binary search. (note: small blocks will skip loop)
1525 */
1526 probe = span = max / 2;
1527 hashval = args->hashval;
1528 while (span > 4) {
1529 span /= 2;
1530 btreehashval = be32_to_cpu(btree[probe].hashval);
1531 if (btreehashval < hashval)
1532 probe += span;
1533 else if (btreehashval > hashval)
1534 probe -= span;
1535 else
1536 break;
1537 }
1538 ASSERT((probe >= 0) && (probe < max));
1539 ASSERT((span <= 4) ||
1540 (be32_to_cpu(btree[probe].hashval) == hashval));
1541
1542 /*
1543 * Since we may have duplicate hashval's, find the first
1544 * matching hashval in the node.
1545 */
1546 while (probe > 0 &&
1547 be32_to_cpu(btree[probe].hashval) >= hashval) {
1548 probe--;
1549 }
1550 while (probe < max &&
1551 be32_to_cpu(btree[probe].hashval) < hashval) {
1552 probe++;
1553 }
1554
1555 /*
1556 * Pick the right block to descend on.
1557 */
1558 if (probe == max) {
1559 blk->index = max - 1;
1560 blkno = be32_to_cpu(btree[max - 1].before);
1561 } else {
1562 blk->index = probe;
1563 blkno = be32_to_cpu(btree[probe].before);
1564 }
1565 }
1566
1567 /*
1568 * A leaf block that ends in the hashval that we are interested in
1569 * (final hashval == search hashval) means that the next block may
1570 * contain more entries with the same hashval, shift upward to the
1571 * next leaf and keep searching.
1572 */
1573 for (;;) {
1574 if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1575 retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
1576 &blk->index, state);
1577 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1578 retval = xfs_attr3_leaf_lookup_int(blk->bp, args);
1579 blk->index = args->index;
1580 args->blkno = blk->blkno;
1581 } else {
1582 ASSERT(0);
1583 return -EFSCORRUPTED;
1584 }
1585 if (((retval == -ENOENT) || (retval == -ENOATTR)) &&
1586 (blk->hashval == args->hashval)) {
1587 error = xfs_da3_path_shift(state, &state->path, 1, 1,
1588 &retval);
1589 if (error)
1590 return error;
1591 if (retval == 0) {
1592 continue;
1593 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1594 /* path_shift() gives ENOENT */
1595 retval = -ENOATTR;
1596 }
1597 }
1598 break;
1599 }
1600 *result = retval;
1601 return 0;
1602}
1603
1604/*========================================================================
1605 * Utility routines.
1606 *========================================================================*/
1607
1608/*
1609 * Compare two intermediate nodes for "order".
1610 */
1611STATIC int
1612xfs_da3_node_order(
1613 struct xfs_inode *dp,
1614 struct xfs_buf *node1_bp,
1615 struct xfs_buf *node2_bp)
1616{
1617 struct xfs_da_intnode *node1;
1618 struct xfs_da_intnode *node2;
1619 struct xfs_da_node_entry *btree1;
1620 struct xfs_da_node_entry *btree2;
1621 struct xfs_da3_icnode_hdr node1hdr;
1622 struct xfs_da3_icnode_hdr node2hdr;
1623
1624 node1 = node1_bp->b_addr;
1625 node2 = node2_bp->b_addr;
1626 dp->d_ops->node_hdr_from_disk(&node1hdr, node1);
1627 dp->d_ops->node_hdr_from_disk(&node2hdr, node2);
1628 btree1 = dp->d_ops->node_tree_p(node1);
1629 btree2 = dp->d_ops->node_tree_p(node2);
1630
1631 if (node1hdr.count > 0 && node2hdr.count > 0 &&
1632 ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
1633 (be32_to_cpu(btree2[node2hdr.count - 1].hashval) <
1634 be32_to_cpu(btree1[node1hdr.count - 1].hashval)))) {
1635 return 1;
1636 }
1637 return 0;
1638}
1639
1640/*
1641 * Link a new block into a doubly linked list of blocks (of whatever type).
1642 */
1643int /* error */
1644xfs_da3_blk_link(
1645 struct xfs_da_state *state,
1646 struct xfs_da_state_blk *old_blk,
1647 struct xfs_da_state_blk *new_blk)
1648{
1649 struct xfs_da_blkinfo *old_info;
1650 struct xfs_da_blkinfo *new_info;
1651 struct xfs_da_blkinfo *tmp_info;
1652 struct xfs_da_args *args;
1653 struct xfs_buf *bp;
1654 int before = 0;
1655 int error;
1656 struct xfs_inode *dp = state->args->dp;
1657
1658 /*
1659 * Set up environment.
1660 */
1661 args = state->args;
1662 ASSERT(args != NULL);
1663 old_info = old_blk->bp->b_addr;
1664 new_info = new_blk->bp->b_addr;
1665 ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
1666 old_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1667 old_blk->magic == XFS_ATTR_LEAF_MAGIC);
1668
1669 switch (old_blk->magic) {
1670 case XFS_ATTR_LEAF_MAGIC:
1671 before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
1672 break;
1673 case XFS_DIR2_LEAFN_MAGIC:
1674 before = xfs_dir2_leafn_order(dp, old_blk->bp, new_blk->bp);
1675 break;
1676 case XFS_DA_NODE_MAGIC:
1677 before = xfs_da3_node_order(dp, old_blk->bp, new_blk->bp);
1678 break;
1679 }
1680
1681 /*
1682 * Link blocks in appropriate order.
1683 */
1684 if (before) {
1685 /*
1686 * Link new block in before existing block.
1687 */
1688 trace_xfs_da_link_before(args);
1689 new_info->forw = cpu_to_be32(old_blk->blkno);
1690 new_info->back = old_info->back;
1691 if (old_info->back) {
1692 error = xfs_da3_node_read(args->trans, dp,
1693 be32_to_cpu(old_info->back),
1694 -1, &bp, args->whichfork);
1695 if (error)
1696 return error;
1697 ASSERT(bp != NULL);
1698 tmp_info = bp->b_addr;
1699 ASSERT(tmp_info->magic == old_info->magic);
1700 ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno);
1701 tmp_info->forw = cpu_to_be32(new_blk->blkno);
1702 xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1703 }
1704 old_info->back = cpu_to_be32(new_blk->blkno);
1705 } else {
1706 /*
1707 * Link new block in after existing block.
1708 */
1709 trace_xfs_da_link_after(args);
1710 new_info->forw = old_info->forw;
1711 new_info->back = cpu_to_be32(old_blk->blkno);
1712 if (old_info->forw) {
1713 error = xfs_da3_node_read(args->trans, dp,
1714 be32_to_cpu(old_info->forw),
1715 -1, &bp, args->whichfork);
1716 if (error)
1717 return error;
1718 ASSERT(bp != NULL);
1719 tmp_info = bp->b_addr;
1720 ASSERT(tmp_info->magic == old_info->magic);
1721 ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno);
1722 tmp_info->back = cpu_to_be32(new_blk->blkno);
1723 xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1724 }
1725 old_info->forw = cpu_to_be32(new_blk->blkno);
1726 }
1727
1728 xfs_trans_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
1729 xfs_trans_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
1730 return 0;
1731}
1732
1733/*
1734 * Unlink a block from a doubly linked list of blocks.
1735 */
1736STATIC int /* error */
1737xfs_da3_blk_unlink(
1738 struct xfs_da_state *state,
1739 struct xfs_da_state_blk *drop_blk,
1740 struct xfs_da_state_blk *save_blk)
1741{
1742 struct xfs_da_blkinfo *drop_info;
1743 struct xfs_da_blkinfo *save_info;
1744 struct xfs_da_blkinfo *tmp_info;
1745 struct xfs_da_args *args;
1746 struct xfs_buf *bp;
1747 int error;
1748
1749 /*
1750 * Set up environment.
1751 */
1752 args = state->args;
1753 ASSERT(args != NULL);
1754 save_info = save_blk->bp->b_addr;
1755 drop_info = drop_blk->bp->b_addr;
1756 ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
1757 save_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1758 save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1759 ASSERT(save_blk->magic == drop_blk->magic);
1760 ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) ||
1761 (be32_to_cpu(save_info->back) == drop_blk->blkno));
1762 ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) ||
1763 (be32_to_cpu(drop_info->back) == save_blk->blkno));
1764
1765 /*
1766 * Unlink the leaf block from the doubly linked chain of leaves.
1767 */
1768 if (be32_to_cpu(save_info->back) == drop_blk->blkno) {
1769 trace_xfs_da_unlink_back(args);
1770 save_info->back = drop_info->back;
1771 if (drop_info->back) {
1772 error = xfs_da3_node_read(args->trans, args->dp,
1773 be32_to_cpu(drop_info->back),
1774 -1, &bp, args->whichfork);
1775 if (error)
1776 return error;
1777 ASSERT(bp != NULL);
1778 tmp_info = bp->b_addr;
1779 ASSERT(tmp_info->magic == save_info->magic);
1780 ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno);
1781 tmp_info->forw = cpu_to_be32(save_blk->blkno);
1782 xfs_trans_log_buf(args->trans, bp, 0,
1783 sizeof(*tmp_info) - 1);
1784 }
1785 } else {
1786 trace_xfs_da_unlink_forward(args);
1787 save_info->forw = drop_info->forw;
1788 if (drop_info->forw) {
1789 error = xfs_da3_node_read(args->trans, args->dp,
1790 be32_to_cpu(drop_info->forw),
1791 -1, &bp, args->whichfork);
1792 if (error)
1793 return error;
1794 ASSERT(bp != NULL);
1795 tmp_info = bp->b_addr;
1796 ASSERT(tmp_info->magic == save_info->magic);
1797 ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno);
1798 tmp_info->back = cpu_to_be32(save_blk->blkno);
1799 xfs_trans_log_buf(args->trans, bp, 0,
1800 sizeof(*tmp_info) - 1);
1801 }
1802 }
1803
1804 xfs_trans_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
1805 return 0;
1806}
1807
1808/*
1809 * Move a path "forward" or "!forward" one block at the current level.
1810 *
1811 * This routine will adjust a "path" to point to the next block
1812 * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1813 * Btree, including updating pointers to the intermediate nodes between
1814 * the new bottom and the root.
1815 */
1816int /* error */
1817xfs_da3_path_shift(
1818 struct xfs_da_state *state,
1819 struct xfs_da_state_path *path,
1820 int forward,
1821 int release,
1822 int *result)
1823{
1824 struct xfs_da_state_blk *blk;
1825 struct xfs_da_blkinfo *info;
1826 struct xfs_da_intnode *node;
1827 struct xfs_da_args *args;
1828 struct xfs_da_node_entry *btree;
1829 struct xfs_da3_icnode_hdr nodehdr;
1830 struct xfs_buf *bp;
1831 xfs_dablk_t blkno = 0;
1832 int level;
1833 int error;
1834 struct xfs_inode *dp = state->args->dp;
1835
1836 trace_xfs_da_path_shift(state->args);
1837
1838 /*
1839 * Roll up the Btree looking for the first block where our
1840 * current index is not at the edge of the block. Note that
1841 * we skip the bottom layer because we want the sibling block.
1842 */
1843 args = state->args;
1844 ASSERT(args != NULL);
1845 ASSERT(path != NULL);
1846 ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1847 level = (path->active-1) - 1; /* skip bottom layer in path */
1848 for (blk = &path->blk[level]; level >= 0; blk--, level--) {
1849 node = blk->bp->b_addr;
1850 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1851 btree = dp->d_ops->node_tree_p(node);
1852
1853 if (forward && (blk->index < nodehdr.count - 1)) {
1854 blk->index++;
1855 blkno = be32_to_cpu(btree[blk->index].before);
1856 break;
1857 } else if (!forward && (blk->index > 0)) {
1858 blk->index--;
1859 blkno = be32_to_cpu(btree[blk->index].before);
1860 break;
1861 }
1862 }
1863 if (level < 0) {
1864 *result = -ENOENT; /* we're out of our tree */
1865 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1866 return 0;
1867 }
1868
1869 /*
1870 * Roll down the edge of the subtree until we reach the
1871 * same depth we were at originally.
1872 */
1873 for (blk++, level++; level < path->active; blk++, level++) {
1874 /*
1875 * Read the next child block into a local buffer.
1876 */
1877 error = xfs_da3_node_read(args->trans, dp, blkno, -1, &bp,
1878 args->whichfork);
1879 if (error)
1880 return error;
1881
1882 /*
1883 * Release the old block (if it's dirty, the trans doesn't
1884 * actually let go) and swap the local buffer into the path
1885 * structure. This ensures failure of the above read doesn't set
1886 * a NULL buffer in an active slot in the path.
1887 */
1888 if (release)
1889 xfs_trans_brelse(args->trans, blk->bp);
1890 blk->blkno = blkno;
1891 blk->bp = bp;
1892
1893 info = blk->bp->b_addr;
1894 ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1895 info->magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
1896 info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1897 info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
1898 info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
1899 info->magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
1900
1901
1902 /*
1903 * Note: we flatten the magic number to a single type so we
1904 * don't have to compare against crc/non-crc types elsewhere.
1905 */
1906 switch (be16_to_cpu(info->magic)) {
1907 case XFS_DA_NODE_MAGIC:
1908 case XFS_DA3_NODE_MAGIC:
1909 blk->magic = XFS_DA_NODE_MAGIC;
1910 node = (xfs_da_intnode_t *)info;
1911 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
1912 btree = dp->d_ops->node_tree_p(node);
1913 blk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1914 if (forward)
1915 blk->index = 0;
1916 else
1917 blk->index = nodehdr.count - 1;
1918 blkno = be32_to_cpu(btree[blk->index].before);
1919 break;
1920 case XFS_ATTR_LEAF_MAGIC:
1921 case XFS_ATTR3_LEAF_MAGIC:
1922 blk->magic = XFS_ATTR_LEAF_MAGIC;
1923 ASSERT(level == path->active-1);
1924 blk->index = 0;
1925 blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1926 break;
1927 case XFS_DIR2_LEAFN_MAGIC:
1928 case XFS_DIR3_LEAFN_MAGIC:
1929 blk->magic = XFS_DIR2_LEAFN_MAGIC;
1930 ASSERT(level == path->active-1);
1931 blk->index = 0;
1932 blk->hashval = xfs_dir2_leafn_lasthash(args->dp,
1933 blk->bp, NULL);
1934 break;
1935 default:
1936 ASSERT(0);
1937 break;
1938 }
1939 }
1940 *result = 0;
1941 return 0;
1942}
1943
1944
1945/*========================================================================
1946 * Utility routines.
1947 *========================================================================*/
1948
1949/*
1950 * Implement a simple hash on a character string.
1951 * Rotate the hash value by 7 bits, then XOR each character in.
1952 * This is implemented with some source-level loop unrolling.
1953 */
1954xfs_dahash_t
1955xfs_da_hashname(const __uint8_t *name, int namelen)
1956{
1957 xfs_dahash_t hash;
1958
1959 /*
1960 * Do four characters at a time as long as we can.
1961 */
1962 for (hash = 0; namelen >= 4; namelen -= 4, name += 4)
1963 hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^
1964 (name[3] << 0) ^ rol32(hash, 7 * 4);
1965
1966 /*
1967 * Now do the rest of the characters.
1968 */
1969 switch (namelen) {
1970 case 3:
1971 return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^
1972 rol32(hash, 7 * 3);
1973 case 2:
1974 return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2);
1975 case 1:
1976 return (name[0] << 0) ^ rol32(hash, 7 * 1);
1977 default: /* case 0: */
1978 return hash;
1979 }
1980}
1981
1982enum xfs_dacmp
1983xfs_da_compname(
1984 struct xfs_da_args *args,
1985 const unsigned char *name,
1986 int len)
1987{
1988 return (args->namelen == len && memcmp(args->name, name, len) == 0) ?
1989 XFS_CMP_EXACT : XFS_CMP_DIFFERENT;
1990}
1991
1992static xfs_dahash_t
1993xfs_default_hashname(
1994 struct xfs_name *name)
1995{
1996 return xfs_da_hashname(name->name, name->len);
1997}
1998
1999const struct xfs_nameops xfs_default_nameops = {
2000 .hashname = xfs_default_hashname,
2001 .compname = xfs_da_compname
2002};
2003
2004int
2005xfs_da_grow_inode_int(
2006 struct xfs_da_args *args,
2007 xfs_fileoff_t *bno,
2008 int count)
2009{
2010 struct xfs_trans *tp = args->trans;
2011 struct xfs_inode *dp = args->dp;
2012 int w = args->whichfork;
2013 xfs_rfsblock_t nblks = dp->i_d.di_nblocks;
2014 struct xfs_bmbt_irec map, *mapp;
2015 int nmap, error, got, i, mapi;
2016
2017 /*
2018 * Find a spot in the file space to put the new block.
2019 */
2020 error = xfs_bmap_first_unused(tp, dp, count, bno, w);
2021 if (error)
2022 return error;
2023
2024 /*
2025 * Try mapping it in one filesystem block.
2026 */
2027 nmap = 1;
2028 ASSERT(args->firstblock != NULL);
2029 error = xfs_bmapi_write(tp, dp, *bno, count,
2030 xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG,
2031 args->firstblock, args->total, &map, &nmap,
2032 args->dfops);
2033 if (error)
2034 return error;
2035
2036 ASSERT(nmap <= 1);
2037 if (nmap == 1) {
2038 mapp = ↦
2039 mapi = 1;
2040 } else if (nmap == 0 && count > 1) {
2041 xfs_fileoff_t b;
2042 int c;
2043
2044 /*
2045 * If we didn't get it and the block might work if fragmented,
2046 * try without the CONTIG flag. Loop until we get it all.
2047 */
2048 mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP);
2049 for (b = *bno, mapi = 0; b < *bno + count; ) {
2050 nmap = MIN(XFS_BMAP_MAX_NMAP, count);
2051 c = (int)(*bno + count - b);
2052 error = xfs_bmapi_write(tp, dp, b, c,
2053 xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
2054 args->firstblock, args->total,
2055 &mapp[mapi], &nmap, args->dfops);
2056 if (error)
2057 goto out_free_map;
2058 if (nmap < 1)
2059 break;
2060 mapi += nmap;
2061 b = mapp[mapi - 1].br_startoff +
2062 mapp[mapi - 1].br_blockcount;
2063 }
2064 } else {
2065 mapi = 0;
2066 mapp = NULL;
2067 }
2068
2069 /*
2070 * Count the blocks we got, make sure it matches the total.
2071 */
2072 for (i = 0, got = 0; i < mapi; i++)
2073 got += mapp[i].br_blockcount;
2074 if (got != count || mapp[0].br_startoff != *bno ||
2075 mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
2076 *bno + count) {
2077 error = -ENOSPC;
2078 goto out_free_map;
2079 }
2080
2081 /* account for newly allocated blocks in reserved blocks total */
2082 args->total -= dp->i_d.di_nblocks - nblks;
2083
2084out_free_map:
2085 if (mapp != &map)
2086 kmem_free(mapp);
2087 return error;
2088}
2089
2090/*
2091 * Add a block to the btree ahead of the file.
2092 * Return the new block number to the caller.
2093 */
2094int
2095xfs_da_grow_inode(
2096 struct xfs_da_args *args,
2097 xfs_dablk_t *new_blkno)
2098{
2099 xfs_fileoff_t bno;
2100 int error;
2101
2102 trace_xfs_da_grow_inode(args);
2103
2104 bno = args->geo->leafblk;
2105 error = xfs_da_grow_inode_int(args, &bno, args->geo->fsbcount);
2106 if (!error)
2107 *new_blkno = (xfs_dablk_t)bno;
2108 return error;
2109}
2110
2111/*
2112 * Ick. We need to always be able to remove a btree block, even
2113 * if there's no space reservation because the filesystem is full.
2114 * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
2115 * It swaps the target block with the last block in the file. The
2116 * last block in the file can always be removed since it can't cause
2117 * a bmap btree split to do that.
2118 */
2119STATIC int
2120xfs_da3_swap_lastblock(
2121 struct xfs_da_args *args,
2122 xfs_dablk_t *dead_blknop,
2123 struct xfs_buf **dead_bufp)
2124{
2125 struct xfs_da_blkinfo *dead_info;
2126 struct xfs_da_blkinfo *sib_info;
2127 struct xfs_da_intnode *par_node;
2128 struct xfs_da_intnode *dead_node;
2129 struct xfs_dir2_leaf *dead_leaf2;
2130 struct xfs_da_node_entry *btree;
2131 struct xfs_da3_icnode_hdr par_hdr;
2132 struct xfs_inode *dp;
2133 struct xfs_trans *tp;
2134 struct xfs_mount *mp;
2135 struct xfs_buf *dead_buf;
2136 struct xfs_buf *last_buf;
2137 struct xfs_buf *sib_buf;
2138 struct xfs_buf *par_buf;
2139 xfs_dahash_t dead_hash;
2140 xfs_fileoff_t lastoff;
2141 xfs_dablk_t dead_blkno;
2142 xfs_dablk_t last_blkno;
2143 xfs_dablk_t sib_blkno;
2144 xfs_dablk_t par_blkno;
2145 int error;
2146 int w;
2147 int entno;
2148 int level;
2149 int dead_level;
2150
2151 trace_xfs_da_swap_lastblock(args);
2152
2153 dead_buf = *dead_bufp;
2154 dead_blkno = *dead_blknop;
2155 tp = args->trans;
2156 dp = args->dp;
2157 w = args->whichfork;
2158 ASSERT(w == XFS_DATA_FORK);
2159 mp = dp->i_mount;
2160 lastoff = args->geo->freeblk;
2161 error = xfs_bmap_last_before(tp, dp, &lastoff, w);
2162 if (error)
2163 return error;
2164 if (unlikely(lastoff == 0)) {
2165 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW,
2166 mp);
2167 return -EFSCORRUPTED;
2168 }
2169 /*
2170 * Read the last block in the btree space.
2171 */
2172 last_blkno = (xfs_dablk_t)lastoff - args->geo->fsbcount;
2173 error = xfs_da3_node_read(tp, dp, last_blkno, -1, &last_buf, w);
2174 if (error)
2175 return error;
2176 /*
2177 * Copy the last block into the dead buffer and log it.
2178 */
2179 memcpy(dead_buf->b_addr, last_buf->b_addr, args->geo->blksize);
2180 xfs_trans_log_buf(tp, dead_buf, 0, args->geo->blksize - 1);
2181 dead_info = dead_buf->b_addr;
2182 /*
2183 * Get values from the moved block.
2184 */
2185 if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
2186 dead_info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
2187 struct xfs_dir3_icleaf_hdr leafhdr;
2188 struct xfs_dir2_leaf_entry *ents;
2189
2190 dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
2191 dp->d_ops->leaf_hdr_from_disk(&leafhdr, dead_leaf2);
2192 ents = dp->d_ops->leaf_ents_p(dead_leaf2);
2193 dead_level = 0;
2194 dead_hash = be32_to_cpu(ents[leafhdr.count - 1].hashval);
2195 } else {
2196 struct xfs_da3_icnode_hdr deadhdr;
2197
2198 dead_node = (xfs_da_intnode_t *)dead_info;
2199 dp->d_ops->node_hdr_from_disk(&deadhdr, dead_node);
2200 btree = dp->d_ops->node_tree_p(dead_node);
2201 dead_level = deadhdr.level;
2202 dead_hash = be32_to_cpu(btree[deadhdr.count - 1].hashval);
2203 }
2204 sib_buf = par_buf = NULL;
2205 /*
2206 * If the moved block has a left sibling, fix up the pointers.
2207 */
2208 if ((sib_blkno = be32_to_cpu(dead_info->back))) {
2209 error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w);
2210 if (error)
2211 goto done;
2212 sib_info = sib_buf->b_addr;
2213 if (unlikely(
2214 be32_to_cpu(sib_info->forw) != last_blkno ||
2215 sib_info->magic != dead_info->magic)) {
2216 XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
2217 XFS_ERRLEVEL_LOW, mp);
2218 error = -EFSCORRUPTED;
2219 goto done;
2220 }
2221 sib_info->forw = cpu_to_be32(dead_blkno);
2222 xfs_trans_log_buf(tp, sib_buf,
2223 XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
2224 sizeof(sib_info->forw)));
2225 sib_buf = NULL;
2226 }
2227 /*
2228 * If the moved block has a right sibling, fix up the pointers.
2229 */
2230 if ((sib_blkno = be32_to_cpu(dead_info->forw))) {
2231 error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w);
2232 if (error)
2233 goto done;
2234 sib_info = sib_buf->b_addr;
2235 if (unlikely(
2236 be32_to_cpu(sib_info->back) != last_blkno ||
2237 sib_info->magic != dead_info->magic)) {
2238 XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
2239 XFS_ERRLEVEL_LOW, mp);
2240 error = -EFSCORRUPTED;
2241 goto done;
2242 }
2243 sib_info->back = cpu_to_be32(dead_blkno);
2244 xfs_trans_log_buf(tp, sib_buf,
2245 XFS_DA_LOGRANGE(sib_info, &sib_info->back,
2246 sizeof(sib_info->back)));
2247 sib_buf = NULL;
2248 }
2249 par_blkno = args->geo->leafblk;
2250 level = -1;
2251 /*
2252 * Walk down the tree looking for the parent of the moved block.
2253 */
2254 for (;;) {
2255 error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w);
2256 if (error)
2257 goto done;
2258 par_node = par_buf->b_addr;
2259 dp->d_ops->node_hdr_from_disk(&par_hdr, par_node);
2260 if (level >= 0 && level != par_hdr.level + 1) {
2261 XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
2262 XFS_ERRLEVEL_LOW, mp);
2263 error = -EFSCORRUPTED;
2264 goto done;
2265 }
2266 level = par_hdr.level;
2267 btree = dp->d_ops->node_tree_p(par_node);
2268 for (entno = 0;
2269 entno < par_hdr.count &&
2270 be32_to_cpu(btree[entno].hashval) < dead_hash;
2271 entno++)
2272 continue;
2273 if (entno == par_hdr.count) {
2274 XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
2275 XFS_ERRLEVEL_LOW, mp);
2276 error = -EFSCORRUPTED;
2277 goto done;
2278 }
2279 par_blkno = be32_to_cpu(btree[entno].before);
2280 if (level == dead_level + 1)
2281 break;
2282 xfs_trans_brelse(tp, par_buf);
2283 par_buf = NULL;
2284 }
2285 /*
2286 * We're in the right parent block.
2287 * Look for the right entry.
2288 */
2289 for (;;) {
2290 for (;
2291 entno < par_hdr.count &&
2292 be32_to_cpu(btree[entno].before) != last_blkno;
2293 entno++)
2294 continue;
2295 if (entno < par_hdr.count)
2296 break;
2297 par_blkno = par_hdr.forw;
2298 xfs_trans_brelse(tp, par_buf);
2299 par_buf = NULL;
2300 if (unlikely(par_blkno == 0)) {
2301 XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
2302 XFS_ERRLEVEL_LOW, mp);
2303 error = -EFSCORRUPTED;
2304 goto done;
2305 }
2306 error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w);
2307 if (error)
2308 goto done;
2309 par_node = par_buf->b_addr;
2310 dp->d_ops->node_hdr_from_disk(&par_hdr, par_node);
2311 if (par_hdr.level != level) {
2312 XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
2313 XFS_ERRLEVEL_LOW, mp);
2314 error = -EFSCORRUPTED;
2315 goto done;
2316 }
2317 btree = dp->d_ops->node_tree_p(par_node);
2318 entno = 0;
2319 }
2320 /*
2321 * Update the parent entry pointing to the moved block.
2322 */
2323 btree[entno].before = cpu_to_be32(dead_blkno);
2324 xfs_trans_log_buf(tp, par_buf,
2325 XFS_DA_LOGRANGE(par_node, &btree[entno].before,
2326 sizeof(btree[entno].before)));
2327 *dead_blknop = last_blkno;
2328 *dead_bufp = last_buf;
2329 return 0;
2330done:
2331 if (par_buf)
2332 xfs_trans_brelse(tp, par_buf);
2333 if (sib_buf)
2334 xfs_trans_brelse(tp, sib_buf);
2335 xfs_trans_brelse(tp, last_buf);
2336 return error;
2337}
2338
2339/*
2340 * Remove a btree block from a directory or attribute.
2341 */
2342int
2343xfs_da_shrink_inode(
2344 xfs_da_args_t *args,
2345 xfs_dablk_t dead_blkno,
2346 struct xfs_buf *dead_buf)
2347{
2348 xfs_inode_t *dp;
2349 int done, error, w, count;
2350 xfs_trans_t *tp;
2351
2352 trace_xfs_da_shrink_inode(args);
2353
2354 dp = args->dp;
2355 w = args->whichfork;
2356 tp = args->trans;
2357 count = args->geo->fsbcount;
2358 for (;;) {
2359 /*
2360 * Remove extents. If we get ENOSPC for a dir we have to move
2361 * the last block to the place we want to kill.
2362 */
2363 error = xfs_bunmapi(tp, dp, dead_blkno, count,
2364 xfs_bmapi_aflag(w), 0, args->firstblock,
2365 args->dfops, &done);
2366 if (error == -ENOSPC) {
2367 if (w != XFS_DATA_FORK)
2368 break;
2369 error = xfs_da3_swap_lastblock(args, &dead_blkno,
2370 &dead_buf);
2371 if (error)
2372 break;
2373 } else {
2374 break;
2375 }
2376 }
2377 xfs_trans_binval(tp, dead_buf);
2378 return error;
2379}
2380
2381/*
2382 * See if the mapping(s) for this btree block are valid, i.e.
2383 * don't contain holes, are logically contiguous, and cover the whole range.
2384 */
2385STATIC int
2386xfs_da_map_covers_blocks(
2387 int nmap,
2388 xfs_bmbt_irec_t *mapp,
2389 xfs_dablk_t bno,
2390 int count)
2391{
2392 int i;
2393 xfs_fileoff_t off;
2394
2395 for (i = 0, off = bno; i < nmap; i++) {
2396 if (mapp[i].br_startblock == HOLESTARTBLOCK ||
2397 mapp[i].br_startblock == DELAYSTARTBLOCK) {
2398 return 0;
2399 }
2400 if (off != mapp[i].br_startoff) {
2401 return 0;
2402 }
2403 off += mapp[i].br_blockcount;
2404 }
2405 return off == bno + count;
2406}
2407
2408/*
2409 * Convert a struct xfs_bmbt_irec to a struct xfs_buf_map.
2410 *
2411 * For the single map case, it is assumed that the caller has provided a pointer
2412 * to a valid xfs_buf_map. For the multiple map case, this function will
2413 * allocate the xfs_buf_map to hold all the maps and replace the caller's single
2414 * map pointer with the allocated map.
2415 */
2416static int
2417xfs_buf_map_from_irec(
2418 struct xfs_mount *mp,
2419 struct xfs_buf_map **mapp,
2420 int *nmaps,
2421 struct xfs_bmbt_irec *irecs,
2422 int nirecs)
2423{
2424 struct xfs_buf_map *map;
2425 int i;
2426
2427 ASSERT(*nmaps == 1);
2428 ASSERT(nirecs >= 1);
2429
2430 if (nirecs > 1) {
2431 map = kmem_zalloc(nirecs * sizeof(struct xfs_buf_map),
2432 KM_SLEEP | KM_NOFS);
2433 if (!map)
2434 return -ENOMEM;
2435 *mapp = map;
2436 }
2437
2438 *nmaps = nirecs;
2439 map = *mapp;
2440 for (i = 0; i < *nmaps; i++) {
2441 ASSERT(irecs[i].br_startblock != DELAYSTARTBLOCK &&
2442 irecs[i].br_startblock != HOLESTARTBLOCK);
2443 map[i].bm_bn = XFS_FSB_TO_DADDR(mp, irecs[i].br_startblock);
2444 map[i].bm_len = XFS_FSB_TO_BB(mp, irecs[i].br_blockcount);
2445 }
2446 return 0;
2447}
2448
2449/*
2450 * Map the block we are given ready for reading. There are three possible return
2451 * values:
2452 * -1 - will be returned if we land in a hole and mappedbno == -2 so the
2453 * caller knows not to execute a subsequent read.
2454 * 0 - if we mapped the block successfully
2455 * >0 - positive error number if there was an error.
2456 */
2457static int
2458xfs_dabuf_map(
2459 struct xfs_inode *dp,
2460 xfs_dablk_t bno,
2461 xfs_daddr_t mappedbno,
2462 int whichfork,
2463 struct xfs_buf_map **map,
2464 int *nmaps)
2465{
2466 struct xfs_mount *mp = dp->i_mount;
2467 int nfsb;
2468 int error = 0;
2469 struct xfs_bmbt_irec irec;
2470 struct xfs_bmbt_irec *irecs = &irec;
2471 int nirecs;
2472
2473 ASSERT(map && *map);
2474 ASSERT(*nmaps == 1);
2475
2476 if (whichfork == XFS_DATA_FORK)
2477 nfsb = mp->m_dir_geo->fsbcount;
2478 else
2479 nfsb = mp->m_attr_geo->fsbcount;
2480
2481 /*
2482 * Caller doesn't have a mapping. -2 means don't complain
2483 * if we land in a hole.
2484 */
2485 if (mappedbno == -1 || mappedbno == -2) {
2486 /*
2487 * Optimize the one-block case.
2488 */
2489 if (nfsb != 1)
2490 irecs = kmem_zalloc(sizeof(irec) * nfsb,
2491 KM_SLEEP | KM_NOFS);
2492
2493 nirecs = nfsb;
2494 error = xfs_bmapi_read(dp, (xfs_fileoff_t)bno, nfsb, irecs,
2495 &nirecs, xfs_bmapi_aflag(whichfork));
2496 if (error)
2497 goto out;
2498 } else {
2499 irecs->br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno);
2500 irecs->br_startoff = (xfs_fileoff_t)bno;
2501 irecs->br_blockcount = nfsb;
2502 irecs->br_state = 0;
2503 nirecs = 1;
2504 }
2505
2506 if (!xfs_da_map_covers_blocks(nirecs, irecs, bno, nfsb)) {
2507 error = mappedbno == -2 ? -1 : -EFSCORRUPTED;
2508 if (unlikely(error == -EFSCORRUPTED)) {
2509 if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
2510 int i;
2511 xfs_alert(mp, "%s: bno %lld dir: inode %lld",
2512 __func__, (long long)bno,
2513 (long long)dp->i_ino);
2514 for (i = 0; i < *nmaps; i++) {
2515 xfs_alert(mp,
2516"[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d",
2517 i,
2518 (long long)irecs[i].br_startoff,
2519 (long long)irecs[i].br_startblock,
2520 (long long)irecs[i].br_blockcount,
2521 irecs[i].br_state);
2522 }
2523 }
2524 XFS_ERROR_REPORT("xfs_da_do_buf(1)",
2525 XFS_ERRLEVEL_LOW, mp);
2526 }
2527 goto out;
2528 }
2529 error = xfs_buf_map_from_irec(mp, map, nmaps, irecs, nirecs);
2530out:
2531 if (irecs != &irec)
2532 kmem_free(irecs);
2533 return error;
2534}
2535
2536/*
2537 * Get a buffer for the dir/attr block.
2538 */
2539int
2540xfs_da_get_buf(
2541 struct xfs_trans *trans,
2542 struct xfs_inode *dp,
2543 xfs_dablk_t bno,
2544 xfs_daddr_t mappedbno,
2545 struct xfs_buf **bpp,
2546 int whichfork)
2547{
2548 struct xfs_buf *bp;
2549 struct xfs_buf_map map;
2550 struct xfs_buf_map *mapp;
2551 int nmap;
2552 int error;
2553
2554 *bpp = NULL;
2555 mapp = ↦
2556 nmap = 1;
2557 error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
2558 &mapp, &nmap);
2559 if (error) {
2560 /* mapping a hole is not an error, but we don't continue */
2561 if (error == -1)
2562 error = 0;
2563 goto out_free;
2564 }
2565
2566 bp = xfs_trans_get_buf_map(trans, dp->i_mount->m_ddev_targp,
2567 mapp, nmap, 0);
2568 error = bp ? bp->b_error : -EIO;
2569 if (error) {
2570 if (bp)
2571 xfs_trans_brelse(trans, bp);
2572 goto out_free;
2573 }
2574
2575 *bpp = bp;
2576
2577out_free:
2578 if (mapp != &map)
2579 kmem_free(mapp);
2580
2581 return error;
2582}
2583
2584/*
2585 * Get a buffer for the dir/attr block, fill in the contents.
2586 */
2587int
2588xfs_da_read_buf(
2589 struct xfs_trans *trans,
2590 struct xfs_inode *dp,
2591 xfs_dablk_t bno,
2592 xfs_daddr_t mappedbno,
2593 struct xfs_buf **bpp,
2594 int whichfork,
2595 const struct xfs_buf_ops *ops)
2596{
2597 struct xfs_buf *bp;
2598 struct xfs_buf_map map;
2599 struct xfs_buf_map *mapp;
2600 int nmap;
2601 int error;
2602
2603 *bpp = NULL;
2604 mapp = ↦
2605 nmap = 1;
2606 error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
2607 &mapp, &nmap);
2608 if (error) {
2609 /* mapping a hole is not an error, but we don't continue */
2610 if (error == -1)
2611 error = 0;
2612 goto out_free;
2613 }
2614
2615 error = xfs_trans_read_buf_map(dp->i_mount, trans,
2616 dp->i_mount->m_ddev_targp,
2617 mapp, nmap, 0, &bp, ops);
2618 if (error)
2619 goto out_free;
2620
2621 if (whichfork == XFS_ATTR_FORK)
2622 xfs_buf_set_ref(bp, XFS_ATTR_BTREE_REF);
2623 else
2624 xfs_buf_set_ref(bp, XFS_DIR_BTREE_REF);
2625 *bpp = bp;
2626out_free:
2627 if (mapp != &map)
2628 kmem_free(mapp);
2629
2630 return error;
2631}
2632
2633/*
2634 * Readahead the dir/attr block.
2635 */
2636int
2637xfs_da_reada_buf(
2638 struct xfs_inode *dp,
2639 xfs_dablk_t bno,
2640 xfs_daddr_t mappedbno,
2641 int whichfork,
2642 const struct xfs_buf_ops *ops)
2643{
2644 struct xfs_buf_map map;
2645 struct xfs_buf_map *mapp;
2646 int nmap;
2647 int error;
2648
2649 mapp = ↦
2650 nmap = 1;
2651 error = xfs_dabuf_map(dp, bno, mappedbno, whichfork,
2652 &mapp, &nmap);
2653 if (error) {
2654 /* mapping a hole is not an error, but we don't continue */
2655 if (error == -1)
2656 error = 0;
2657 goto out_free;
2658 }
2659
2660 mappedbno = mapp[0].bm_bn;
2661 xfs_buf_readahead_map(dp->i_mount->m_ddev_targp, mapp, nmap, ops);
2662
2663out_free:
2664 if (mapp != &map)
2665 kmem_free(mapp);
2666
2667 return error;
2668}
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
255static void
256xfs_da3_node_write_verify(
257 struct xfs_buf *bp)
258{
259 struct xfs_mount *mp = bp->b_mount;
260 struct xfs_buf_log_item *bip = bp->b_log_item;
261 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
262 xfs_failaddr_t fa;
263
264 fa = xfs_da3_node_verify(bp);
265 if (fa) {
266 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
267 return;
268 }
269
270 if (!xfs_has_crc(mp))
271 return;
272
273 if (bip)
274 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
275
276 xfs_buf_update_cksum(bp, XFS_DA3_NODE_CRC_OFF);
277}
278
279/*
280 * leaf/node format detection on trees is sketchy, so a node read can be done on
281 * leaf level blocks when detection identifies the tree as a node format tree
282 * incorrectly. In this case, we need to swap the verifier to match the correct
283 * format of the block being read.
284 */
285static void
286xfs_da3_node_read_verify(
287 struct xfs_buf *bp)
288{
289 struct xfs_da_blkinfo *info = bp->b_addr;
290 xfs_failaddr_t fa;
291
292 switch (be16_to_cpu(info->magic)) {
293 case XFS_DA3_NODE_MAGIC:
294 if (!xfs_buf_verify_cksum(bp, XFS_DA3_NODE_CRC_OFF)) {
295 xfs_verifier_error(bp, -EFSBADCRC,
296 __this_address);
297 break;
298 }
299 fallthrough;
300 case XFS_DA_NODE_MAGIC:
301 fa = xfs_da3_node_verify(bp);
302 if (fa)
303 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
304 return;
305 case XFS_ATTR_LEAF_MAGIC:
306 case XFS_ATTR3_LEAF_MAGIC:
307 bp->b_ops = &xfs_attr3_leaf_buf_ops;
308 bp->b_ops->verify_read(bp);
309 return;
310 case XFS_DIR2_LEAFN_MAGIC:
311 case XFS_DIR3_LEAFN_MAGIC:
312 bp->b_ops = &xfs_dir3_leafn_buf_ops;
313 bp->b_ops->verify_read(bp);
314 return;
315 default:
316 xfs_verifier_error(bp, -EFSCORRUPTED, __this_address);
317 break;
318 }
319}
320
321/* Verify the structure of a da3 block. */
322static xfs_failaddr_t
323xfs_da3_node_verify_struct(
324 struct xfs_buf *bp)
325{
326 struct xfs_da_blkinfo *info = bp->b_addr;
327
328 switch (be16_to_cpu(info->magic)) {
329 case XFS_DA3_NODE_MAGIC:
330 case XFS_DA_NODE_MAGIC:
331 return xfs_da3_node_verify(bp);
332 case XFS_ATTR_LEAF_MAGIC:
333 case XFS_ATTR3_LEAF_MAGIC:
334 bp->b_ops = &xfs_attr3_leaf_buf_ops;
335 return bp->b_ops->verify_struct(bp);
336 case XFS_DIR2_LEAFN_MAGIC:
337 case XFS_DIR3_LEAFN_MAGIC:
338 bp->b_ops = &xfs_dir3_leafn_buf_ops;
339 return bp->b_ops->verify_struct(bp);
340 default:
341 return __this_address;
342 }
343}
344
345const struct xfs_buf_ops xfs_da3_node_buf_ops = {
346 .name = "xfs_da3_node",
347 .magic16 = { cpu_to_be16(XFS_DA_NODE_MAGIC),
348 cpu_to_be16(XFS_DA3_NODE_MAGIC) },
349 .verify_read = xfs_da3_node_read_verify,
350 .verify_write = xfs_da3_node_write_verify,
351 .verify_struct = xfs_da3_node_verify_struct,
352};
353
354static int
355xfs_da3_node_set_type(
356 struct xfs_trans *tp,
357 struct xfs_inode *dp,
358 int whichfork,
359 struct xfs_buf *bp)
360{
361 struct xfs_da_blkinfo *info = bp->b_addr;
362
363 switch (be16_to_cpu(info->magic)) {
364 case XFS_DA_NODE_MAGIC:
365 case XFS_DA3_NODE_MAGIC:
366 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
367 return 0;
368 case XFS_ATTR_LEAF_MAGIC:
369 case XFS_ATTR3_LEAF_MAGIC:
370 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_ATTR_LEAF_BUF);
371 return 0;
372 case XFS_DIR2_LEAFN_MAGIC:
373 case XFS_DIR3_LEAFN_MAGIC:
374 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
375 return 0;
376 default:
377 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, tp->t_mountp,
378 info, sizeof(*info));
379 xfs_trans_brelse(tp, bp);
380 xfs_dirattr_mark_sick(dp, whichfork);
381 return -EFSCORRUPTED;
382 }
383}
384
385int
386xfs_da3_node_read(
387 struct xfs_trans *tp,
388 struct xfs_inode *dp,
389 xfs_dablk_t bno,
390 struct xfs_buf **bpp,
391 int whichfork)
392{
393 int error;
394
395 error = xfs_da_read_buf(tp, dp, bno, 0, bpp, whichfork,
396 &xfs_da3_node_buf_ops);
397 if (error || !*bpp || !tp)
398 return error;
399 return xfs_da3_node_set_type(tp, dp, whichfork, *bpp);
400}
401
402int
403xfs_da3_node_read_mapped(
404 struct xfs_trans *tp,
405 struct xfs_inode *dp,
406 xfs_daddr_t mappedbno,
407 struct xfs_buf **bpp,
408 int whichfork)
409{
410 struct xfs_mount *mp = dp->i_mount;
411 int error;
412
413 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, mappedbno,
414 XFS_FSB_TO_BB(mp, xfs_dabuf_nfsb(mp, whichfork)), 0,
415 bpp, &xfs_da3_node_buf_ops);
416 if (xfs_metadata_is_sick(error))
417 xfs_dirattr_mark_sick(dp, whichfork);
418 if (error || !*bpp)
419 return error;
420
421 if (whichfork == XFS_ATTR_FORK)
422 xfs_buf_set_ref(*bpp, XFS_ATTR_BTREE_REF);
423 else
424 xfs_buf_set_ref(*bpp, XFS_DIR_BTREE_REF);
425
426 if (!tp)
427 return 0;
428 return xfs_da3_node_set_type(tp, dp, whichfork, *bpp);
429}
430
431/*
432 * Copy src directory/attr leaf/node buffer to the dst.
433 * For v5 file systems make sure the right blkno is stamped in.
434 */
435void
436xfs_da_buf_copy(
437 struct xfs_buf *dst,
438 struct xfs_buf *src,
439 size_t size)
440{
441 struct xfs_da3_blkinfo *da3 = dst->b_addr;
442
443 memcpy(dst->b_addr, src->b_addr, size);
444 dst->b_ops = src->b_ops;
445 xfs_trans_buf_copy_type(dst, src);
446 if (xfs_has_crc(dst->b_mount))
447 da3->blkno = cpu_to_be64(xfs_buf_daddr(dst));
448}
449
450/*========================================================================
451 * Routines used for growing the Btree.
452 *========================================================================*/
453
454/*
455 * Create the initial contents of an intermediate node.
456 */
457int
458xfs_da3_node_create(
459 struct xfs_da_args *args,
460 xfs_dablk_t blkno,
461 int level,
462 struct xfs_buf **bpp,
463 int whichfork)
464{
465 struct xfs_da_intnode *node;
466 struct xfs_trans *tp = args->trans;
467 struct xfs_mount *mp = tp->t_mountp;
468 struct xfs_da3_icnode_hdr ichdr = {0};
469 struct xfs_buf *bp;
470 int error;
471 struct xfs_inode *dp = args->dp;
472
473 trace_xfs_da_node_create(args);
474 ASSERT(level <= XFS_DA_NODE_MAXDEPTH);
475
476 error = xfs_da_get_buf(tp, dp, blkno, &bp, whichfork);
477 if (error)
478 return error;
479 bp->b_ops = &xfs_da3_node_buf_ops;
480 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF);
481 node = bp->b_addr;
482
483 if (xfs_has_crc(mp)) {
484 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
485
486 memset(hdr3, 0, sizeof(struct xfs_da3_node_hdr));
487 ichdr.magic = XFS_DA3_NODE_MAGIC;
488 hdr3->info.blkno = cpu_to_be64(xfs_buf_daddr(bp));
489 hdr3->info.owner = cpu_to_be64(args->dp->i_ino);
490 uuid_copy(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid);
491 } else {
492 ichdr.magic = XFS_DA_NODE_MAGIC;
493 }
494 ichdr.level = level;
495
496 xfs_da3_node_hdr_to_disk(dp->i_mount, node, &ichdr);
497 xfs_trans_log_buf(tp, bp,
498 XFS_DA_LOGRANGE(node, &node->hdr, args->geo->node_hdr_size));
499
500 *bpp = bp;
501 return 0;
502}
503
504/*
505 * Split a leaf node, rebalance, then possibly split
506 * intermediate nodes, rebalance, etc.
507 */
508int /* error */
509xfs_da3_split(
510 struct xfs_da_state *state)
511{
512 struct xfs_da_state_blk *oldblk;
513 struct xfs_da_state_blk *newblk;
514 struct xfs_da_state_blk *addblk;
515 struct xfs_da_intnode *node;
516 int max;
517 int action = 0;
518 int error;
519 int i;
520
521 trace_xfs_da_split(state->args);
522
523 if (XFS_TEST_ERROR(false, state->mp, XFS_ERRTAG_DA_LEAF_SPLIT))
524 return -EIO;
525
526 /*
527 * Walk back up the tree splitting/inserting/adjusting as necessary.
528 * If we need to insert and there isn't room, split the node, then
529 * decide which fragment to insert the new block from below into.
530 * Note that we may split the root this way, but we need more fixup.
531 */
532 max = state->path.active - 1;
533 ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH));
534 ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC ||
535 state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
536
537 addblk = &state->path.blk[max]; /* initial dummy value */
538 for (i = max; (i >= 0) && addblk; state->path.active--, i--) {
539 oldblk = &state->path.blk[i];
540 newblk = &state->altpath.blk[i];
541
542 /*
543 * If a leaf node then
544 * Allocate a new leaf node, then rebalance across them.
545 * else if an intermediate node then
546 * We split on the last layer, must we split the node?
547 */
548 switch (oldblk->magic) {
549 case XFS_ATTR_LEAF_MAGIC:
550 error = xfs_attr3_leaf_split(state, oldblk, newblk);
551 if ((error != 0) && (error != -ENOSPC)) {
552 return error; /* GROT: attr is inconsistent */
553 }
554 if (!error) {
555 addblk = newblk;
556 break;
557 }
558 /*
559 * Entry wouldn't fit, split the leaf again. The new
560 * extrablk will be consumed by xfs_da3_node_split if
561 * the node is split.
562 */
563 state->extravalid = 1;
564 if (state->inleaf) {
565 state->extraafter = 0; /* before newblk */
566 trace_xfs_attr_leaf_split_before(state->args);
567 error = xfs_attr3_leaf_split(state, oldblk,
568 &state->extrablk);
569 } else {
570 state->extraafter = 1; /* after newblk */
571 trace_xfs_attr_leaf_split_after(state->args);
572 error = xfs_attr3_leaf_split(state, newblk,
573 &state->extrablk);
574 }
575 if (error)
576 return error; /* GROT: attr inconsistent */
577 addblk = newblk;
578 break;
579 case XFS_DIR2_LEAFN_MAGIC:
580 error = xfs_dir2_leafn_split(state, oldblk, newblk);
581 if (error)
582 return error;
583 addblk = newblk;
584 break;
585 case XFS_DA_NODE_MAGIC:
586 error = xfs_da3_node_split(state, oldblk, newblk, addblk,
587 max - i, &action);
588 addblk->bp = NULL;
589 if (error)
590 return error; /* GROT: dir is inconsistent */
591 /*
592 * Record the newly split block for the next time thru?
593 */
594 if (action)
595 addblk = newblk;
596 else
597 addblk = NULL;
598 break;
599 }
600
601 /*
602 * Update the btree to show the new hashval for this child.
603 */
604 xfs_da3_fixhashpath(state, &state->path);
605 }
606 if (!addblk)
607 return 0;
608
609 /*
610 * xfs_da3_node_split() should have consumed any extra blocks we added
611 * during a double leaf split in the attr fork. This is guaranteed as
612 * we can't be here if the attr fork only has a single leaf block.
613 */
614 ASSERT(state->extravalid == 0 ||
615 state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
616
617 /*
618 * Split the root node.
619 */
620 ASSERT(state->path.active == 0);
621 oldblk = &state->path.blk[0];
622 error = xfs_da3_root_split(state, oldblk, addblk);
623 if (error)
624 goto out;
625
626 /*
627 * Update pointers to the node which used to be block 0 and just got
628 * bumped because of the addition of a new root node. Note that the
629 * original block 0 could be at any position in the list of blocks in
630 * the tree.
631 *
632 * Note: the magic numbers and sibling pointers are in the same physical
633 * place for both v2 and v3 headers (by design). Hence it doesn't matter
634 * which version of the xfs_da_intnode structure we use here as the
635 * result will be the same using either structure.
636 */
637 node = oldblk->bp->b_addr;
638 if (node->hdr.info.forw) {
639 if (be32_to_cpu(node->hdr.info.forw) != addblk->blkno) {
640 xfs_buf_mark_corrupt(oldblk->bp);
641 xfs_da_mark_sick(state->args);
642 error = -EFSCORRUPTED;
643 goto out;
644 }
645 node = addblk->bp->b_addr;
646 node->hdr.info.back = cpu_to_be32(oldblk->blkno);
647 xfs_trans_log_buf(state->args->trans, addblk->bp,
648 XFS_DA_LOGRANGE(node, &node->hdr.info,
649 sizeof(node->hdr.info)));
650 }
651 node = oldblk->bp->b_addr;
652 if (node->hdr.info.back) {
653 if (be32_to_cpu(node->hdr.info.back) != addblk->blkno) {
654 xfs_buf_mark_corrupt(oldblk->bp);
655 xfs_da_mark_sick(state->args);
656 error = -EFSCORRUPTED;
657 goto out;
658 }
659 node = addblk->bp->b_addr;
660 node->hdr.info.forw = cpu_to_be32(oldblk->blkno);
661 xfs_trans_log_buf(state->args->trans, addblk->bp,
662 XFS_DA_LOGRANGE(node, &node->hdr.info,
663 sizeof(node->hdr.info)));
664 }
665out:
666 addblk->bp = NULL;
667 return error;
668}
669
670/*
671 * Split the root. We have to create a new root and point to the two
672 * parts (the split old root) that we just created. Copy block zero to
673 * the EOF, extending the inode in process.
674 */
675STATIC int /* error */
676xfs_da3_root_split(
677 struct xfs_da_state *state,
678 struct xfs_da_state_blk *blk1,
679 struct xfs_da_state_blk *blk2)
680{
681 struct xfs_da_intnode *node;
682 struct xfs_da_intnode *oldroot;
683 struct xfs_da_node_entry *btree;
684 struct xfs_da3_icnode_hdr nodehdr;
685 struct xfs_da_args *args;
686 struct xfs_buf *bp;
687 struct xfs_inode *dp;
688 struct xfs_trans *tp;
689 struct xfs_dir2_leaf *leaf;
690 xfs_dablk_t blkno;
691 int level;
692 int error;
693 int size;
694
695 trace_xfs_da_root_split(state->args);
696
697 /*
698 * Copy the existing (incorrect) block from the root node position
699 * to a free space somewhere.
700 */
701 args = state->args;
702 error = xfs_da_grow_inode(args, &blkno);
703 if (error)
704 return error;
705
706 dp = args->dp;
707 tp = args->trans;
708 error = xfs_da_get_buf(tp, dp, blkno, &bp, args->whichfork);
709 if (error)
710 return error;
711 node = bp->b_addr;
712 oldroot = blk1->bp->b_addr;
713 if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
714 oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC)) {
715 struct xfs_da3_icnode_hdr icnodehdr;
716
717 xfs_da3_node_hdr_from_disk(dp->i_mount, &icnodehdr, oldroot);
718 btree = icnodehdr.btree;
719 size = (int)((char *)&btree[icnodehdr.count] - (char *)oldroot);
720 level = icnodehdr.level;
721 } else {
722 struct xfs_dir3_icleaf_hdr leafhdr;
723
724 leaf = (xfs_dir2_leaf_t *)oldroot;
725 xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
726
727 ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
728 leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
729 size = (int)((char *)&leafhdr.ents[leafhdr.count] -
730 (char *)leaf);
731 level = 0;
732 }
733
734 /*
735 * Copy old root to new buffer and log it.
736 */
737 xfs_da_buf_copy(bp, blk1->bp, size);
738 xfs_trans_log_buf(tp, bp, 0, size - 1);
739
740 /*
741 * Update blk1 to point to new buffer.
742 */
743 blk1->bp = bp;
744 blk1->blkno = blkno;
745
746 /*
747 * Set up the new root node.
748 */
749 error = xfs_da3_node_create(args,
750 (args->whichfork == XFS_DATA_FORK) ? args->geo->leafblk : 0,
751 level + 1, &bp, args->whichfork);
752 if (error)
753 return error;
754
755 node = bp->b_addr;
756 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
757 btree = nodehdr.btree;
758 btree[0].hashval = cpu_to_be32(blk1->hashval);
759 btree[0].before = cpu_to_be32(blk1->blkno);
760 btree[1].hashval = cpu_to_be32(blk2->hashval);
761 btree[1].before = cpu_to_be32(blk2->blkno);
762 nodehdr.count = 2;
763 xfs_da3_node_hdr_to_disk(dp->i_mount, node, &nodehdr);
764
765#ifdef DEBUG
766 if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
767 oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
768 ASSERT(blk1->blkno >= args->geo->leafblk &&
769 blk1->blkno < args->geo->freeblk);
770 ASSERT(blk2->blkno >= args->geo->leafblk &&
771 blk2->blkno < args->geo->freeblk);
772 }
773#endif
774
775 /* Header is already logged by xfs_da_node_create */
776 xfs_trans_log_buf(tp, bp,
777 XFS_DA_LOGRANGE(node, btree, sizeof(xfs_da_node_entry_t) * 2));
778
779 return 0;
780}
781
782/*
783 * Split the node, rebalance, then add the new entry.
784 */
785STATIC int /* error */
786xfs_da3_node_split(
787 struct xfs_da_state *state,
788 struct xfs_da_state_blk *oldblk,
789 struct xfs_da_state_blk *newblk,
790 struct xfs_da_state_blk *addblk,
791 int treelevel,
792 int *result)
793{
794 struct xfs_da_intnode *node;
795 struct xfs_da3_icnode_hdr nodehdr;
796 xfs_dablk_t blkno;
797 int newcount;
798 int error;
799 int useextra;
800 struct xfs_inode *dp = state->args->dp;
801
802 trace_xfs_da_node_split(state->args);
803
804 node = oldblk->bp->b_addr;
805 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
806
807 /*
808 * With V2 dirs the extra block is data or freespace.
809 */
810 useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK;
811 newcount = 1 + useextra;
812 /*
813 * Do we have to split the node?
814 */
815 if (nodehdr.count + newcount > state->args->geo->node_ents) {
816 /*
817 * Allocate a new node, add to the doubly linked chain of
818 * nodes, then move some of our excess entries into it.
819 */
820 error = xfs_da_grow_inode(state->args, &blkno);
821 if (error)
822 return error; /* GROT: dir is inconsistent */
823
824 error = xfs_da3_node_create(state->args, blkno, treelevel,
825 &newblk->bp, state->args->whichfork);
826 if (error)
827 return error; /* GROT: dir is inconsistent */
828 newblk->blkno = blkno;
829 newblk->magic = XFS_DA_NODE_MAGIC;
830 xfs_da3_node_rebalance(state, oldblk, newblk);
831 error = xfs_da3_blk_link(state, oldblk, newblk);
832 if (error)
833 return error;
834 *result = 1;
835 } else {
836 *result = 0;
837 }
838
839 /*
840 * Insert the new entry(s) into the correct block
841 * (updating last hashval in the process).
842 *
843 * xfs_da3_node_add() inserts BEFORE the given index,
844 * and as a result of using node_lookup_int() we always
845 * point to a valid entry (not after one), but a split
846 * operation always results in a new block whose hashvals
847 * FOLLOW the current block.
848 *
849 * If we had double-split op below us, then add the extra block too.
850 */
851 node = oldblk->bp->b_addr;
852 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
853 if (oldblk->index <= nodehdr.count) {
854 oldblk->index++;
855 xfs_da3_node_add(state, oldblk, addblk);
856 if (useextra) {
857 if (state->extraafter)
858 oldblk->index++;
859 xfs_da3_node_add(state, oldblk, &state->extrablk);
860 state->extravalid = 0;
861 }
862 } else {
863 newblk->index++;
864 xfs_da3_node_add(state, newblk, addblk);
865 if (useextra) {
866 if (state->extraafter)
867 newblk->index++;
868 xfs_da3_node_add(state, newblk, &state->extrablk);
869 state->extravalid = 0;
870 }
871 }
872
873 return 0;
874}
875
876/*
877 * Balance the btree elements between two intermediate nodes,
878 * usually one full and one empty.
879 *
880 * NOTE: if blk2 is empty, then it will get the upper half of blk1.
881 */
882STATIC void
883xfs_da3_node_rebalance(
884 struct xfs_da_state *state,
885 struct xfs_da_state_blk *blk1,
886 struct xfs_da_state_blk *blk2)
887{
888 struct xfs_da_intnode *node1;
889 struct xfs_da_intnode *node2;
890 struct xfs_da_node_entry *btree1;
891 struct xfs_da_node_entry *btree2;
892 struct xfs_da_node_entry *btree_s;
893 struct xfs_da_node_entry *btree_d;
894 struct xfs_da3_icnode_hdr nodehdr1;
895 struct xfs_da3_icnode_hdr nodehdr2;
896 struct xfs_trans *tp;
897 int count;
898 int tmp;
899 int swap = 0;
900 struct xfs_inode *dp = state->args->dp;
901
902 trace_xfs_da_node_rebalance(state->args);
903
904 node1 = blk1->bp->b_addr;
905 node2 = blk2->bp->b_addr;
906 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr1, node1);
907 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr2, node2);
908 btree1 = nodehdr1.btree;
909 btree2 = nodehdr2.btree;
910
911 /*
912 * Figure out how many entries need to move, and in which direction.
913 * Swap the nodes around if that makes it simpler.
914 */
915 if (nodehdr1.count > 0 && nodehdr2.count > 0 &&
916 ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
917 (be32_to_cpu(btree2[nodehdr2.count - 1].hashval) <
918 be32_to_cpu(btree1[nodehdr1.count - 1].hashval)))) {
919 swap(node1, node2);
920 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr1, node1);
921 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr2, node2);
922 btree1 = nodehdr1.btree;
923 btree2 = nodehdr2.btree;
924 swap = 1;
925 }
926
927 count = (nodehdr1.count - nodehdr2.count) / 2;
928 if (count == 0)
929 return;
930 tp = state->args->trans;
931 /*
932 * Two cases: high-to-low and low-to-high.
933 */
934 if (count > 0) {
935 /*
936 * Move elements in node2 up to make a hole.
937 */
938 tmp = nodehdr2.count;
939 if (tmp > 0) {
940 tmp *= (uint)sizeof(xfs_da_node_entry_t);
941 btree_s = &btree2[0];
942 btree_d = &btree2[count];
943 memmove(btree_d, btree_s, tmp);
944 }
945
946 /*
947 * Move the req'd B-tree elements from high in node1 to
948 * low in node2.
949 */
950 nodehdr2.count += count;
951 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
952 btree_s = &btree1[nodehdr1.count - count];
953 btree_d = &btree2[0];
954 memcpy(btree_d, btree_s, tmp);
955 nodehdr1.count -= count;
956 } else {
957 /*
958 * Move the req'd B-tree elements from low in node2 to
959 * high in node1.
960 */
961 count = -count;
962 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
963 btree_s = &btree2[0];
964 btree_d = &btree1[nodehdr1.count];
965 memcpy(btree_d, btree_s, tmp);
966 nodehdr1.count += count;
967
968 xfs_trans_log_buf(tp, blk1->bp,
969 XFS_DA_LOGRANGE(node1, btree_d, tmp));
970
971 /*
972 * Move elements in node2 down to fill the hole.
973 */
974 tmp = nodehdr2.count - count;
975 tmp *= (uint)sizeof(xfs_da_node_entry_t);
976 btree_s = &btree2[count];
977 btree_d = &btree2[0];
978 memmove(btree_d, btree_s, tmp);
979 nodehdr2.count -= count;
980 }
981
982 /*
983 * Log header of node 1 and all current bits of node 2.
984 */
985 xfs_da3_node_hdr_to_disk(dp->i_mount, node1, &nodehdr1);
986 xfs_trans_log_buf(tp, blk1->bp,
987 XFS_DA_LOGRANGE(node1, &node1->hdr,
988 state->args->geo->node_hdr_size));
989
990 xfs_da3_node_hdr_to_disk(dp->i_mount, node2, &nodehdr2);
991 xfs_trans_log_buf(tp, blk2->bp,
992 XFS_DA_LOGRANGE(node2, &node2->hdr,
993 state->args->geo->node_hdr_size +
994 (sizeof(btree2[0]) * nodehdr2.count)));
995
996 /*
997 * Record the last hashval from each block for upward propagation.
998 * (note: don't use the swapped node pointers)
999 */
1000 if (swap) {
1001 node1 = blk1->bp->b_addr;
1002 node2 = blk2->bp->b_addr;
1003 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr1, node1);
1004 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr2, node2);
1005 btree1 = nodehdr1.btree;
1006 btree2 = nodehdr2.btree;
1007 }
1008 blk1->hashval = be32_to_cpu(btree1[nodehdr1.count - 1].hashval);
1009 blk2->hashval = be32_to_cpu(btree2[nodehdr2.count - 1].hashval);
1010
1011 /*
1012 * Adjust the expected index for insertion.
1013 */
1014 if (blk1->index >= nodehdr1.count) {
1015 blk2->index = blk1->index - nodehdr1.count;
1016 blk1->index = nodehdr1.count + 1; /* make it invalid */
1017 }
1018}
1019
1020/*
1021 * Add a new entry to an intermediate node.
1022 */
1023STATIC void
1024xfs_da3_node_add(
1025 struct xfs_da_state *state,
1026 struct xfs_da_state_blk *oldblk,
1027 struct xfs_da_state_blk *newblk)
1028{
1029 struct xfs_da_intnode *node;
1030 struct xfs_da3_icnode_hdr nodehdr;
1031 struct xfs_da_node_entry *btree;
1032 int tmp;
1033 struct xfs_inode *dp = state->args->dp;
1034
1035 trace_xfs_da_node_add(state->args);
1036
1037 node = oldblk->bp->b_addr;
1038 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
1039 btree = nodehdr.btree;
1040
1041 ASSERT(oldblk->index >= 0 && oldblk->index <= nodehdr.count);
1042 ASSERT(newblk->blkno != 0);
1043 if (state->args->whichfork == XFS_DATA_FORK)
1044 ASSERT(newblk->blkno >= state->args->geo->leafblk &&
1045 newblk->blkno < state->args->geo->freeblk);
1046
1047 /*
1048 * We may need to make some room before we insert the new node.
1049 */
1050 tmp = 0;
1051 if (oldblk->index < nodehdr.count) {
1052 tmp = (nodehdr.count - oldblk->index) * (uint)sizeof(*btree);
1053 memmove(&btree[oldblk->index + 1], &btree[oldblk->index], tmp);
1054 }
1055 btree[oldblk->index].hashval = cpu_to_be32(newblk->hashval);
1056 btree[oldblk->index].before = cpu_to_be32(newblk->blkno);
1057 xfs_trans_log_buf(state->args->trans, oldblk->bp,
1058 XFS_DA_LOGRANGE(node, &btree[oldblk->index],
1059 tmp + sizeof(*btree)));
1060
1061 nodehdr.count += 1;
1062 xfs_da3_node_hdr_to_disk(dp->i_mount, node, &nodehdr);
1063 xfs_trans_log_buf(state->args->trans, oldblk->bp,
1064 XFS_DA_LOGRANGE(node, &node->hdr,
1065 state->args->geo->node_hdr_size));
1066
1067 /*
1068 * Copy the last hash value from the oldblk to propagate upwards.
1069 */
1070 oldblk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1071}
1072
1073/*========================================================================
1074 * Routines used for shrinking the Btree.
1075 *========================================================================*/
1076
1077/*
1078 * Deallocate an empty leaf node, remove it from its parent,
1079 * possibly deallocating that block, etc...
1080 */
1081int
1082xfs_da3_join(
1083 struct xfs_da_state *state)
1084{
1085 struct xfs_da_state_blk *drop_blk;
1086 struct xfs_da_state_blk *save_blk;
1087 int action = 0;
1088 int error;
1089
1090 trace_xfs_da_join(state->args);
1091
1092 drop_blk = &state->path.blk[ state->path.active-1 ];
1093 save_blk = &state->altpath.blk[ state->path.active-1 ];
1094 ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
1095 ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
1096 drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1097
1098 /*
1099 * Walk back up the tree joining/deallocating as necessary.
1100 * When we stop dropping blocks, break out.
1101 */
1102 for ( ; state->path.active >= 2; drop_blk--, save_blk--,
1103 state->path.active--) {
1104 /*
1105 * See if we can combine the block with a neighbor.
1106 * (action == 0) => no options, just leave
1107 * (action == 1) => coalesce, then unlink
1108 * (action == 2) => block empty, unlink it
1109 */
1110 switch (drop_blk->magic) {
1111 case XFS_ATTR_LEAF_MAGIC:
1112 error = xfs_attr3_leaf_toosmall(state, &action);
1113 if (error)
1114 return error;
1115 if (action == 0)
1116 return 0;
1117 xfs_attr3_leaf_unbalance(state, drop_blk, save_blk);
1118 break;
1119 case XFS_DIR2_LEAFN_MAGIC:
1120 error = xfs_dir2_leafn_toosmall(state, &action);
1121 if (error)
1122 return error;
1123 if (action == 0)
1124 return 0;
1125 xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
1126 break;
1127 case XFS_DA_NODE_MAGIC:
1128 /*
1129 * Remove the offending node, fixup hashvals,
1130 * check for a toosmall neighbor.
1131 */
1132 xfs_da3_node_remove(state, drop_blk);
1133 xfs_da3_fixhashpath(state, &state->path);
1134 error = xfs_da3_node_toosmall(state, &action);
1135 if (error)
1136 return error;
1137 if (action == 0)
1138 return 0;
1139 xfs_da3_node_unbalance(state, drop_blk, save_blk);
1140 break;
1141 }
1142 xfs_da3_fixhashpath(state, &state->altpath);
1143 error = xfs_da3_blk_unlink(state, drop_blk, save_blk);
1144 xfs_da_state_kill_altpath(state);
1145 if (error)
1146 return error;
1147 error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
1148 drop_blk->bp);
1149 drop_blk->bp = NULL;
1150 if (error)
1151 return error;
1152 }
1153 /*
1154 * We joined all the way to the top. If it turns out that
1155 * we only have one entry in the root, make the child block
1156 * the new root.
1157 */
1158 xfs_da3_node_remove(state, drop_blk);
1159 xfs_da3_fixhashpath(state, &state->path);
1160 error = xfs_da3_root_join(state, &state->path.blk[0]);
1161 return error;
1162}
1163
1164#ifdef DEBUG
1165static void
1166xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level)
1167{
1168 __be16 magic = blkinfo->magic;
1169
1170 if (level == 1) {
1171 ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1172 magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
1173 magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
1174 magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
1175 } else {
1176 ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
1177 magic == cpu_to_be16(XFS_DA3_NODE_MAGIC));
1178 }
1179 ASSERT(!blkinfo->forw);
1180 ASSERT(!blkinfo->back);
1181}
1182#else /* !DEBUG */
1183#define xfs_da_blkinfo_onlychild_validate(blkinfo, level)
1184#endif /* !DEBUG */
1185
1186/*
1187 * We have only one entry in the root. Copy the only remaining child of
1188 * the old root to block 0 as the new root node.
1189 */
1190STATIC int
1191xfs_da3_root_join(
1192 struct xfs_da_state *state,
1193 struct xfs_da_state_blk *root_blk)
1194{
1195 struct xfs_da_intnode *oldroot;
1196 struct xfs_da_args *args;
1197 xfs_dablk_t child;
1198 struct xfs_buf *bp;
1199 struct xfs_da3_icnode_hdr oldroothdr;
1200 int error;
1201 struct xfs_inode *dp = state->args->dp;
1202
1203 trace_xfs_da_root_join(state->args);
1204
1205 ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
1206
1207 args = state->args;
1208 oldroot = root_blk->bp->b_addr;
1209 xfs_da3_node_hdr_from_disk(dp->i_mount, &oldroothdr, oldroot);
1210 ASSERT(oldroothdr.forw == 0);
1211 ASSERT(oldroothdr.back == 0);
1212
1213 /*
1214 * If the root has more than one child, then don't do anything.
1215 */
1216 if (oldroothdr.count > 1)
1217 return 0;
1218
1219 /*
1220 * Read in the (only) child block, then copy those bytes into
1221 * the root block's buffer and free the original child block.
1222 */
1223 child = be32_to_cpu(oldroothdr.btree[0].before);
1224 ASSERT(child != 0);
1225 error = xfs_da3_node_read(args->trans, dp, child, &bp, args->whichfork);
1226 if (error)
1227 return error;
1228 xfs_da_blkinfo_onlychild_validate(bp->b_addr, oldroothdr.level);
1229
1230 /*
1231 * Copy child to root buffer and log it.
1232 */
1233 xfs_da_buf_copy(root_blk->bp, bp, args->geo->blksize);
1234 xfs_trans_log_buf(args->trans, root_blk->bp, 0,
1235 args->geo->blksize - 1);
1236 /*
1237 * Now we can drop the child buffer.
1238 */
1239 error = xfs_da_shrink_inode(args, child, bp);
1240 return error;
1241}
1242
1243/*
1244 * Check a node block and its neighbors to see if the block should be
1245 * collapsed into one or the other neighbor. Always keep the block
1246 * with the smaller block number.
1247 * If the current block is over 50% full, don't try to join it, return 0.
1248 * If the block is empty, fill in the state structure and return 2.
1249 * If it can be collapsed, fill in the state structure and return 1.
1250 * If nothing can be done, return 0.
1251 */
1252STATIC int
1253xfs_da3_node_toosmall(
1254 struct xfs_da_state *state,
1255 int *action)
1256{
1257 struct xfs_da_intnode *node;
1258 struct xfs_da_state_blk *blk;
1259 struct xfs_da_blkinfo *info;
1260 xfs_dablk_t blkno;
1261 struct xfs_buf *bp;
1262 struct xfs_da3_icnode_hdr nodehdr;
1263 int count;
1264 int forward;
1265 int error;
1266 int retval;
1267 int i;
1268 struct xfs_inode *dp = state->args->dp;
1269
1270 trace_xfs_da_node_toosmall(state->args);
1271
1272 /*
1273 * Check for the degenerate case of the block being over 50% full.
1274 * If so, it's not worth even looking to see if we might be able
1275 * to coalesce with a sibling.
1276 */
1277 blk = &state->path.blk[ state->path.active-1 ];
1278 info = blk->bp->b_addr;
1279 node = (xfs_da_intnode_t *)info;
1280 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
1281 if (nodehdr.count > (state->args->geo->node_ents >> 1)) {
1282 *action = 0; /* blk over 50%, don't try to join */
1283 return 0; /* blk over 50%, don't try to join */
1284 }
1285
1286 /*
1287 * Check for the degenerate case of the block being empty.
1288 * If the block is empty, we'll simply delete it, no need to
1289 * coalesce it with a sibling block. We choose (arbitrarily)
1290 * to merge with the forward block unless it is NULL.
1291 */
1292 if (nodehdr.count == 0) {
1293 /*
1294 * Make altpath point to the block we want to keep and
1295 * path point to the block we want to drop (this one).
1296 */
1297 forward = (info->forw != 0);
1298 memcpy(&state->altpath, &state->path, sizeof(state->path));
1299 error = xfs_da3_path_shift(state, &state->altpath, forward,
1300 0, &retval);
1301 if (error)
1302 return error;
1303 if (retval) {
1304 *action = 0;
1305 } else {
1306 *action = 2;
1307 }
1308 return 0;
1309 }
1310
1311 /*
1312 * Examine each sibling block to see if we can coalesce with
1313 * at least 25% free space to spare. We need to figure out
1314 * whether to merge with the forward or the backward block.
1315 * We prefer coalescing with the lower numbered sibling so as
1316 * to shrink a directory over time.
1317 */
1318 count = state->args->geo->node_ents;
1319 count -= state->args->geo->node_ents >> 2;
1320 count -= nodehdr.count;
1321
1322 /* start with smaller blk num */
1323 forward = nodehdr.forw < nodehdr.back;
1324 for (i = 0; i < 2; forward = !forward, i++) {
1325 struct xfs_da3_icnode_hdr thdr;
1326 if (forward)
1327 blkno = nodehdr.forw;
1328 else
1329 blkno = nodehdr.back;
1330 if (blkno == 0)
1331 continue;
1332 error = xfs_da3_node_read(state->args->trans, dp, blkno, &bp,
1333 state->args->whichfork);
1334 if (error)
1335 return error;
1336
1337 node = bp->b_addr;
1338 xfs_da3_node_hdr_from_disk(dp->i_mount, &thdr, node);
1339 xfs_trans_brelse(state->args->trans, bp);
1340
1341 if (count - thdr.count >= 0)
1342 break; /* fits with at least 25% to spare */
1343 }
1344 if (i >= 2) {
1345 *action = 0;
1346 return 0;
1347 }
1348
1349 /*
1350 * Make altpath point to the block we want to keep (the lower
1351 * numbered block) and path point to the block we want to drop.
1352 */
1353 memcpy(&state->altpath, &state->path, sizeof(state->path));
1354 if (blkno < blk->blkno) {
1355 error = xfs_da3_path_shift(state, &state->altpath, forward,
1356 0, &retval);
1357 } else {
1358 error = xfs_da3_path_shift(state, &state->path, forward,
1359 0, &retval);
1360 }
1361 if (error)
1362 return error;
1363 if (retval) {
1364 *action = 0;
1365 return 0;
1366 }
1367 *action = 1;
1368 return 0;
1369}
1370
1371/*
1372 * Pick up the last hashvalue from an intermediate node.
1373 */
1374STATIC uint
1375xfs_da3_node_lasthash(
1376 struct xfs_inode *dp,
1377 struct xfs_buf *bp,
1378 int *count)
1379{
1380 struct xfs_da3_icnode_hdr nodehdr;
1381
1382 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, bp->b_addr);
1383 if (count)
1384 *count = nodehdr.count;
1385 if (!nodehdr.count)
1386 return 0;
1387 return be32_to_cpu(nodehdr.btree[nodehdr.count - 1].hashval);
1388}
1389
1390/*
1391 * Walk back up the tree adjusting hash values as necessary,
1392 * when we stop making changes, return.
1393 */
1394void
1395xfs_da3_fixhashpath(
1396 struct xfs_da_state *state,
1397 struct xfs_da_state_path *path)
1398{
1399 struct xfs_da_state_blk *blk;
1400 struct xfs_da_intnode *node;
1401 struct xfs_da_node_entry *btree;
1402 xfs_dahash_t lasthash=0;
1403 int level;
1404 int count;
1405 struct xfs_inode *dp = state->args->dp;
1406
1407 trace_xfs_da_fixhashpath(state->args);
1408
1409 level = path->active-1;
1410 blk = &path->blk[ level ];
1411 switch (blk->magic) {
1412 case XFS_ATTR_LEAF_MAGIC:
1413 lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
1414 if (count == 0)
1415 return;
1416 break;
1417 case XFS_DIR2_LEAFN_MAGIC:
1418 lasthash = xfs_dir2_leaf_lasthash(dp, blk->bp, &count);
1419 if (count == 0)
1420 return;
1421 break;
1422 case XFS_DA_NODE_MAGIC:
1423 lasthash = xfs_da3_node_lasthash(dp, blk->bp, &count);
1424 if (count == 0)
1425 return;
1426 break;
1427 }
1428 for (blk--, level--; level >= 0; blk--, level--) {
1429 struct xfs_da3_icnode_hdr nodehdr;
1430
1431 node = blk->bp->b_addr;
1432 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
1433 btree = nodehdr.btree;
1434 if (be32_to_cpu(btree[blk->index].hashval) == lasthash)
1435 break;
1436 blk->hashval = lasthash;
1437 btree[blk->index].hashval = cpu_to_be32(lasthash);
1438 xfs_trans_log_buf(state->args->trans, blk->bp,
1439 XFS_DA_LOGRANGE(node, &btree[blk->index],
1440 sizeof(*btree)));
1441
1442 lasthash = be32_to_cpu(btree[nodehdr.count - 1].hashval);
1443 }
1444}
1445
1446/*
1447 * Remove an entry from an intermediate node.
1448 */
1449STATIC void
1450xfs_da3_node_remove(
1451 struct xfs_da_state *state,
1452 struct xfs_da_state_blk *drop_blk)
1453{
1454 struct xfs_da_intnode *node;
1455 struct xfs_da3_icnode_hdr nodehdr;
1456 struct xfs_da_node_entry *btree;
1457 int index;
1458 int tmp;
1459 struct xfs_inode *dp = state->args->dp;
1460
1461 trace_xfs_da_node_remove(state->args);
1462
1463 node = drop_blk->bp->b_addr;
1464 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
1465 ASSERT(drop_blk->index < nodehdr.count);
1466 ASSERT(drop_blk->index >= 0);
1467
1468 /*
1469 * Copy over the offending entry, or just zero it out.
1470 */
1471 index = drop_blk->index;
1472 btree = nodehdr.btree;
1473 if (index < nodehdr.count - 1) {
1474 tmp = nodehdr.count - index - 1;
1475 tmp *= (uint)sizeof(xfs_da_node_entry_t);
1476 memmove(&btree[index], &btree[index + 1], tmp);
1477 xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1478 XFS_DA_LOGRANGE(node, &btree[index], tmp));
1479 index = nodehdr.count - 1;
1480 }
1481 memset(&btree[index], 0, sizeof(xfs_da_node_entry_t));
1482 xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1483 XFS_DA_LOGRANGE(node, &btree[index], sizeof(btree[index])));
1484 nodehdr.count -= 1;
1485 xfs_da3_node_hdr_to_disk(dp->i_mount, node, &nodehdr);
1486 xfs_trans_log_buf(state->args->trans, drop_blk->bp,
1487 XFS_DA_LOGRANGE(node, &node->hdr, state->args->geo->node_hdr_size));
1488
1489 /*
1490 * Copy the last hash value from the block to propagate upwards.
1491 */
1492 drop_blk->hashval = be32_to_cpu(btree[index - 1].hashval);
1493}
1494
1495/*
1496 * Unbalance the elements between two intermediate nodes,
1497 * move all Btree elements from one node into another.
1498 */
1499STATIC void
1500xfs_da3_node_unbalance(
1501 struct xfs_da_state *state,
1502 struct xfs_da_state_blk *drop_blk,
1503 struct xfs_da_state_blk *save_blk)
1504{
1505 struct xfs_da_intnode *drop_node;
1506 struct xfs_da_intnode *save_node;
1507 struct xfs_da_node_entry *drop_btree;
1508 struct xfs_da_node_entry *save_btree;
1509 struct xfs_da3_icnode_hdr drop_hdr;
1510 struct xfs_da3_icnode_hdr save_hdr;
1511 struct xfs_trans *tp;
1512 int sindex;
1513 int tmp;
1514 struct xfs_inode *dp = state->args->dp;
1515
1516 trace_xfs_da_node_unbalance(state->args);
1517
1518 drop_node = drop_blk->bp->b_addr;
1519 save_node = save_blk->bp->b_addr;
1520 xfs_da3_node_hdr_from_disk(dp->i_mount, &drop_hdr, drop_node);
1521 xfs_da3_node_hdr_from_disk(dp->i_mount, &save_hdr, save_node);
1522 drop_btree = drop_hdr.btree;
1523 save_btree = save_hdr.btree;
1524 tp = state->args->trans;
1525
1526 /*
1527 * If the dying block has lower hashvals, then move all the
1528 * elements in the remaining block up to make a hole.
1529 */
1530 if ((be32_to_cpu(drop_btree[0].hashval) <
1531 be32_to_cpu(save_btree[0].hashval)) ||
1532 (be32_to_cpu(drop_btree[drop_hdr.count - 1].hashval) <
1533 be32_to_cpu(save_btree[save_hdr.count - 1].hashval))) {
1534 /* XXX: check this - is memmove dst correct? */
1535 tmp = save_hdr.count * sizeof(xfs_da_node_entry_t);
1536 memmove(&save_btree[drop_hdr.count], &save_btree[0], tmp);
1537
1538 sindex = 0;
1539 xfs_trans_log_buf(tp, save_blk->bp,
1540 XFS_DA_LOGRANGE(save_node, &save_btree[0],
1541 (save_hdr.count + drop_hdr.count) *
1542 sizeof(xfs_da_node_entry_t)));
1543 } else {
1544 sindex = save_hdr.count;
1545 xfs_trans_log_buf(tp, save_blk->bp,
1546 XFS_DA_LOGRANGE(save_node, &save_btree[sindex],
1547 drop_hdr.count * sizeof(xfs_da_node_entry_t)));
1548 }
1549
1550 /*
1551 * Move all the B-tree elements from drop_blk to save_blk.
1552 */
1553 tmp = drop_hdr.count * (uint)sizeof(xfs_da_node_entry_t);
1554 memcpy(&save_btree[sindex], &drop_btree[0], tmp);
1555 save_hdr.count += drop_hdr.count;
1556
1557 xfs_da3_node_hdr_to_disk(dp->i_mount, save_node, &save_hdr);
1558 xfs_trans_log_buf(tp, save_blk->bp,
1559 XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1560 state->args->geo->node_hdr_size));
1561
1562 /*
1563 * Save the last hashval in the remaining block for upward propagation.
1564 */
1565 save_blk->hashval = be32_to_cpu(save_btree[save_hdr.count - 1].hashval);
1566}
1567
1568/*========================================================================
1569 * Routines used for finding things in the Btree.
1570 *========================================================================*/
1571
1572/*
1573 * Walk down the Btree looking for a particular filename, filling
1574 * in the state structure as we go.
1575 *
1576 * We will set the state structure to point to each of the elements
1577 * in each of the nodes where either the hashval is or should be.
1578 *
1579 * We support duplicate hashval's so for each entry in the current
1580 * node that could contain the desired hashval, descend. This is a
1581 * pruned depth-first tree search.
1582 */
1583int /* error */
1584xfs_da3_node_lookup_int(
1585 struct xfs_da_state *state,
1586 int *result)
1587{
1588 struct xfs_da_state_blk *blk;
1589 struct xfs_da_blkinfo *curr;
1590 struct xfs_da_intnode *node;
1591 struct xfs_da_node_entry *btree;
1592 struct xfs_da3_icnode_hdr nodehdr;
1593 struct xfs_da_args *args;
1594 xfs_dablk_t blkno;
1595 xfs_dahash_t hashval;
1596 xfs_dahash_t btreehashval;
1597 int probe;
1598 int span;
1599 int max;
1600 int error;
1601 int retval;
1602 unsigned int expected_level = 0;
1603 uint16_t magic;
1604 struct xfs_inode *dp = state->args->dp;
1605
1606 args = state->args;
1607
1608 /*
1609 * Descend thru the B-tree searching each level for the right
1610 * node to use, until the right hashval is found.
1611 */
1612 blkno = args->geo->leafblk;
1613 for (blk = &state->path.blk[0], state->path.active = 1;
1614 state->path.active <= XFS_DA_NODE_MAXDEPTH;
1615 blk++, state->path.active++) {
1616 /*
1617 * Read the next node down in the tree.
1618 */
1619 blk->blkno = blkno;
1620 error = xfs_da3_node_read(args->trans, args->dp, blkno,
1621 &blk->bp, args->whichfork);
1622 if (error) {
1623 blk->blkno = 0;
1624 state->path.active--;
1625 return error;
1626 }
1627 curr = blk->bp->b_addr;
1628 magic = be16_to_cpu(curr->magic);
1629
1630 if (magic == XFS_ATTR_LEAF_MAGIC ||
1631 magic == XFS_ATTR3_LEAF_MAGIC) {
1632 blk->magic = XFS_ATTR_LEAF_MAGIC;
1633 blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1634 break;
1635 }
1636
1637 if (magic == XFS_DIR2_LEAFN_MAGIC ||
1638 magic == XFS_DIR3_LEAFN_MAGIC) {
1639 blk->magic = XFS_DIR2_LEAFN_MAGIC;
1640 blk->hashval = xfs_dir2_leaf_lasthash(args->dp,
1641 blk->bp, NULL);
1642 break;
1643 }
1644
1645 if (magic != XFS_DA_NODE_MAGIC && magic != XFS_DA3_NODE_MAGIC) {
1646 xfs_buf_mark_corrupt(blk->bp);
1647 xfs_da_mark_sick(args);
1648 return -EFSCORRUPTED;
1649 }
1650
1651 blk->magic = XFS_DA_NODE_MAGIC;
1652
1653 /*
1654 * Search an intermediate node for a match.
1655 */
1656 node = blk->bp->b_addr;
1657 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node);
1658 btree = nodehdr.btree;
1659
1660 /* Tree taller than we can handle; bail out! */
1661 if (nodehdr.level >= XFS_DA_NODE_MAXDEPTH) {
1662 xfs_buf_mark_corrupt(blk->bp);
1663 xfs_da_mark_sick(args);
1664 return -EFSCORRUPTED;
1665 }
1666
1667 /* Check the level from the root. */
1668 if (blkno == args->geo->leafblk)
1669 expected_level = nodehdr.level - 1;
1670 else if (expected_level != nodehdr.level) {
1671 xfs_buf_mark_corrupt(blk->bp);
1672 xfs_da_mark_sick(args);
1673 return -EFSCORRUPTED;
1674 } else
1675 expected_level--;
1676
1677 max = nodehdr.count;
1678 blk->hashval = be32_to_cpu(btree[max - 1].hashval);
1679
1680 /*
1681 * Binary search. (note: small blocks will skip loop)
1682 */
1683 probe = span = max / 2;
1684 hashval = args->hashval;
1685 while (span > 4) {
1686 span /= 2;
1687 btreehashval = be32_to_cpu(btree[probe].hashval);
1688 if (btreehashval < hashval)
1689 probe += span;
1690 else if (btreehashval > hashval)
1691 probe -= span;
1692 else
1693 break;
1694 }
1695 ASSERT((probe >= 0) && (probe < max));
1696 ASSERT((span <= 4) ||
1697 (be32_to_cpu(btree[probe].hashval) == hashval));
1698
1699 /*
1700 * Since we may have duplicate hashval's, find the first
1701 * matching hashval in the node.
1702 */
1703 while (probe > 0 &&
1704 be32_to_cpu(btree[probe].hashval) >= hashval) {
1705 probe--;
1706 }
1707 while (probe < max &&
1708 be32_to_cpu(btree[probe].hashval) < hashval) {
1709 probe++;
1710 }
1711
1712 /*
1713 * Pick the right block to descend on.
1714 */
1715 if (probe == max) {
1716 blk->index = max - 1;
1717 blkno = be32_to_cpu(btree[max - 1].before);
1718 } else {
1719 blk->index = probe;
1720 blkno = be32_to_cpu(btree[probe].before);
1721 }
1722
1723 /* We can't point back to the root. */
1724 if (XFS_IS_CORRUPT(dp->i_mount, blkno == args->geo->leafblk)) {
1725 xfs_da_mark_sick(args);
1726 return -EFSCORRUPTED;
1727 }
1728 }
1729
1730 if (XFS_IS_CORRUPT(dp->i_mount, expected_level != 0)) {
1731 xfs_da_mark_sick(args);
1732 return -EFSCORRUPTED;
1733 }
1734
1735 /*
1736 * A leaf block that ends in the hashval that we are interested in
1737 * (final hashval == search hashval) means that the next block may
1738 * contain more entries with the same hashval, shift upward to the
1739 * next leaf and keep searching.
1740 */
1741 for (;;) {
1742 if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
1743 retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
1744 &blk->index, state);
1745 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1746 retval = xfs_attr3_leaf_lookup_int(blk->bp, args);
1747 blk->index = args->index;
1748 args->blkno = blk->blkno;
1749 } else {
1750 ASSERT(0);
1751 xfs_da_mark_sick(args);
1752 return -EFSCORRUPTED;
1753 }
1754 if (((retval == -ENOENT) || (retval == -ENOATTR)) &&
1755 (blk->hashval == args->hashval)) {
1756 error = xfs_da3_path_shift(state, &state->path, 1, 1,
1757 &retval);
1758 if (error)
1759 return error;
1760 if (retval == 0) {
1761 continue;
1762 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
1763 /* path_shift() gives ENOENT */
1764 retval = -ENOATTR;
1765 }
1766 }
1767 break;
1768 }
1769 *result = retval;
1770 return 0;
1771}
1772
1773/*========================================================================
1774 * Utility routines.
1775 *========================================================================*/
1776
1777/*
1778 * Compare two intermediate nodes for "order".
1779 */
1780STATIC int
1781xfs_da3_node_order(
1782 struct xfs_inode *dp,
1783 struct xfs_buf *node1_bp,
1784 struct xfs_buf *node2_bp)
1785{
1786 struct xfs_da_intnode *node1;
1787 struct xfs_da_intnode *node2;
1788 struct xfs_da_node_entry *btree1;
1789 struct xfs_da_node_entry *btree2;
1790 struct xfs_da3_icnode_hdr node1hdr;
1791 struct xfs_da3_icnode_hdr node2hdr;
1792
1793 node1 = node1_bp->b_addr;
1794 node2 = node2_bp->b_addr;
1795 xfs_da3_node_hdr_from_disk(dp->i_mount, &node1hdr, node1);
1796 xfs_da3_node_hdr_from_disk(dp->i_mount, &node2hdr, node2);
1797 btree1 = node1hdr.btree;
1798 btree2 = node2hdr.btree;
1799
1800 if (node1hdr.count > 0 && node2hdr.count > 0 &&
1801 ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) ||
1802 (be32_to_cpu(btree2[node2hdr.count - 1].hashval) <
1803 be32_to_cpu(btree1[node1hdr.count - 1].hashval)))) {
1804 return 1;
1805 }
1806 return 0;
1807}
1808
1809/*
1810 * Link a new block into a doubly linked list of blocks (of whatever type).
1811 */
1812int /* error */
1813xfs_da3_blk_link(
1814 struct xfs_da_state *state,
1815 struct xfs_da_state_blk *old_blk,
1816 struct xfs_da_state_blk *new_blk)
1817{
1818 struct xfs_da_blkinfo *old_info;
1819 struct xfs_da_blkinfo *new_info;
1820 struct xfs_da_blkinfo *tmp_info;
1821 struct xfs_da_args *args;
1822 struct xfs_buf *bp;
1823 int before = 0;
1824 int error;
1825 struct xfs_inode *dp = state->args->dp;
1826
1827 /*
1828 * Set up environment.
1829 */
1830 args = state->args;
1831 ASSERT(args != NULL);
1832 old_info = old_blk->bp->b_addr;
1833 new_info = new_blk->bp->b_addr;
1834 ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
1835 old_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1836 old_blk->magic == XFS_ATTR_LEAF_MAGIC);
1837
1838 switch (old_blk->magic) {
1839 case XFS_ATTR_LEAF_MAGIC:
1840 before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
1841 break;
1842 case XFS_DIR2_LEAFN_MAGIC:
1843 before = xfs_dir2_leafn_order(dp, old_blk->bp, new_blk->bp);
1844 break;
1845 case XFS_DA_NODE_MAGIC:
1846 before = xfs_da3_node_order(dp, old_blk->bp, new_blk->bp);
1847 break;
1848 }
1849
1850 /*
1851 * Link blocks in appropriate order.
1852 */
1853 if (before) {
1854 /*
1855 * Link new block in before existing block.
1856 */
1857 trace_xfs_da_link_before(args);
1858 new_info->forw = cpu_to_be32(old_blk->blkno);
1859 new_info->back = old_info->back;
1860 if (old_info->back) {
1861 error = xfs_da3_node_read(args->trans, dp,
1862 be32_to_cpu(old_info->back),
1863 &bp, args->whichfork);
1864 if (error)
1865 return error;
1866 ASSERT(bp != NULL);
1867 tmp_info = bp->b_addr;
1868 ASSERT(tmp_info->magic == old_info->magic);
1869 ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno);
1870 tmp_info->forw = cpu_to_be32(new_blk->blkno);
1871 xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1872 }
1873 old_info->back = cpu_to_be32(new_blk->blkno);
1874 } else {
1875 /*
1876 * Link new block in after existing block.
1877 */
1878 trace_xfs_da_link_after(args);
1879 new_info->forw = old_info->forw;
1880 new_info->back = cpu_to_be32(old_blk->blkno);
1881 if (old_info->forw) {
1882 error = xfs_da3_node_read(args->trans, dp,
1883 be32_to_cpu(old_info->forw),
1884 &bp, args->whichfork);
1885 if (error)
1886 return error;
1887 ASSERT(bp != NULL);
1888 tmp_info = bp->b_addr;
1889 ASSERT(tmp_info->magic == old_info->magic);
1890 ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno);
1891 tmp_info->back = cpu_to_be32(new_blk->blkno);
1892 xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1893 }
1894 old_info->forw = cpu_to_be32(new_blk->blkno);
1895 }
1896
1897 xfs_trans_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
1898 xfs_trans_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
1899 return 0;
1900}
1901
1902/*
1903 * Unlink a block from a doubly linked list of blocks.
1904 */
1905STATIC int /* error */
1906xfs_da3_blk_unlink(
1907 struct xfs_da_state *state,
1908 struct xfs_da_state_blk *drop_blk,
1909 struct xfs_da_state_blk *save_blk)
1910{
1911 struct xfs_da_blkinfo *drop_info;
1912 struct xfs_da_blkinfo *save_info;
1913 struct xfs_da_blkinfo *tmp_info;
1914 struct xfs_da_args *args;
1915 struct xfs_buf *bp;
1916 int error;
1917
1918 /*
1919 * Set up environment.
1920 */
1921 args = state->args;
1922 ASSERT(args != NULL);
1923 save_info = save_blk->bp->b_addr;
1924 drop_info = drop_blk->bp->b_addr;
1925 ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
1926 save_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1927 save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1928 ASSERT(save_blk->magic == drop_blk->magic);
1929 ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) ||
1930 (be32_to_cpu(save_info->back) == drop_blk->blkno));
1931 ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) ||
1932 (be32_to_cpu(drop_info->back) == save_blk->blkno));
1933
1934 /*
1935 * Unlink the leaf block from the doubly linked chain of leaves.
1936 */
1937 if (be32_to_cpu(save_info->back) == drop_blk->blkno) {
1938 trace_xfs_da_unlink_back(args);
1939 save_info->back = drop_info->back;
1940 if (drop_info->back) {
1941 error = xfs_da3_node_read(args->trans, args->dp,
1942 be32_to_cpu(drop_info->back),
1943 &bp, args->whichfork);
1944 if (error)
1945 return error;
1946 ASSERT(bp != NULL);
1947 tmp_info = bp->b_addr;
1948 ASSERT(tmp_info->magic == save_info->magic);
1949 ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno);
1950 tmp_info->forw = cpu_to_be32(save_blk->blkno);
1951 xfs_trans_log_buf(args->trans, bp, 0,
1952 sizeof(*tmp_info) - 1);
1953 }
1954 } else {
1955 trace_xfs_da_unlink_forward(args);
1956 save_info->forw = drop_info->forw;
1957 if (drop_info->forw) {
1958 error = xfs_da3_node_read(args->trans, args->dp,
1959 be32_to_cpu(drop_info->forw),
1960 &bp, args->whichfork);
1961 if (error)
1962 return error;
1963 ASSERT(bp != NULL);
1964 tmp_info = bp->b_addr;
1965 ASSERT(tmp_info->magic == save_info->magic);
1966 ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno);
1967 tmp_info->back = cpu_to_be32(save_blk->blkno);
1968 xfs_trans_log_buf(args->trans, bp, 0,
1969 sizeof(*tmp_info) - 1);
1970 }
1971 }
1972
1973 xfs_trans_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
1974 return 0;
1975}
1976
1977/*
1978 * Move a path "forward" or "!forward" one block at the current level.
1979 *
1980 * This routine will adjust a "path" to point to the next block
1981 * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1982 * Btree, including updating pointers to the intermediate nodes between
1983 * the new bottom and the root.
1984 */
1985int /* error */
1986xfs_da3_path_shift(
1987 struct xfs_da_state *state,
1988 struct xfs_da_state_path *path,
1989 int forward,
1990 int release,
1991 int *result)
1992{
1993 struct xfs_da_state_blk *blk;
1994 struct xfs_da_blkinfo *info;
1995 struct xfs_da_args *args;
1996 struct xfs_da_node_entry *btree;
1997 struct xfs_da3_icnode_hdr nodehdr;
1998 struct xfs_buf *bp;
1999 xfs_dablk_t blkno = 0;
2000 int level;
2001 int error;
2002 struct xfs_inode *dp = state->args->dp;
2003
2004 trace_xfs_da_path_shift(state->args);
2005
2006 /*
2007 * Roll up the Btree looking for the first block where our
2008 * current index is not at the edge of the block. Note that
2009 * we skip the bottom layer because we want the sibling block.
2010 */
2011 args = state->args;
2012 ASSERT(args != NULL);
2013 ASSERT(path != NULL);
2014 ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
2015 level = (path->active-1) - 1; /* skip bottom layer in path */
2016 for (; level >= 0; level--) {
2017 blk = &path->blk[level];
2018 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr,
2019 blk->bp->b_addr);
2020
2021 if (forward && (blk->index < nodehdr.count - 1)) {
2022 blk->index++;
2023 blkno = be32_to_cpu(nodehdr.btree[blk->index].before);
2024 break;
2025 } else if (!forward && (blk->index > 0)) {
2026 blk->index--;
2027 blkno = be32_to_cpu(nodehdr.btree[blk->index].before);
2028 break;
2029 }
2030 }
2031 if (level < 0) {
2032 *result = -ENOENT; /* we're out of our tree */
2033 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
2034 return 0;
2035 }
2036
2037 /*
2038 * Roll down the edge of the subtree until we reach the
2039 * same depth we were at originally.
2040 */
2041 for (blk++, level++; level < path->active; blk++, level++) {
2042 /*
2043 * Read the next child block into a local buffer.
2044 */
2045 error = xfs_da3_node_read(args->trans, dp, blkno, &bp,
2046 args->whichfork);
2047 if (error)
2048 return error;
2049
2050 /*
2051 * Release the old block (if it's dirty, the trans doesn't
2052 * actually let go) and swap the local buffer into the path
2053 * structure. This ensures failure of the above read doesn't set
2054 * a NULL buffer in an active slot in the path.
2055 */
2056 if (release)
2057 xfs_trans_brelse(args->trans, blk->bp);
2058 blk->blkno = blkno;
2059 blk->bp = bp;
2060
2061 info = blk->bp->b_addr;
2062 ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) ||
2063 info->magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) ||
2064 info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
2065 info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) ||
2066 info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
2067 info->magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
2068
2069
2070 /*
2071 * Note: we flatten the magic number to a single type so we
2072 * don't have to compare against crc/non-crc types elsewhere.
2073 */
2074 switch (be16_to_cpu(info->magic)) {
2075 case XFS_DA_NODE_MAGIC:
2076 case XFS_DA3_NODE_MAGIC:
2077 blk->magic = XFS_DA_NODE_MAGIC;
2078 xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr,
2079 bp->b_addr);
2080 btree = nodehdr.btree;
2081 blk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval);
2082 if (forward)
2083 blk->index = 0;
2084 else
2085 blk->index = nodehdr.count - 1;
2086 blkno = be32_to_cpu(btree[blk->index].before);
2087 break;
2088 case XFS_ATTR_LEAF_MAGIC:
2089 case XFS_ATTR3_LEAF_MAGIC:
2090 blk->magic = XFS_ATTR_LEAF_MAGIC;
2091 ASSERT(level == path->active-1);
2092 blk->index = 0;
2093 blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
2094 break;
2095 case XFS_DIR2_LEAFN_MAGIC:
2096 case XFS_DIR3_LEAFN_MAGIC:
2097 blk->magic = XFS_DIR2_LEAFN_MAGIC;
2098 ASSERT(level == path->active-1);
2099 blk->index = 0;
2100 blk->hashval = xfs_dir2_leaf_lasthash(args->dp,
2101 blk->bp, NULL);
2102 break;
2103 default:
2104 ASSERT(0);
2105 break;
2106 }
2107 }
2108 *result = 0;
2109 return 0;
2110}
2111
2112
2113/*========================================================================
2114 * Utility routines.
2115 *========================================================================*/
2116
2117/*
2118 * Implement a simple hash on a character string.
2119 * Rotate the hash value by 7 bits, then XOR each character in.
2120 * This is implemented with some source-level loop unrolling.
2121 */
2122xfs_dahash_t
2123xfs_da_hashname(const uint8_t *name, int namelen)
2124{
2125 xfs_dahash_t hash;
2126
2127 /*
2128 * Do four characters at a time as long as we can.
2129 */
2130 for (hash = 0; namelen >= 4; namelen -= 4, name += 4)
2131 hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^
2132 (name[3] << 0) ^ rol32(hash, 7 * 4);
2133
2134 /*
2135 * Now do the rest of the characters.
2136 */
2137 switch (namelen) {
2138 case 3:
2139 return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^
2140 rol32(hash, 7 * 3);
2141 case 2:
2142 return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2);
2143 case 1:
2144 return (name[0] << 0) ^ rol32(hash, 7 * 1);
2145 default: /* case 0: */
2146 return hash;
2147 }
2148}
2149
2150enum xfs_dacmp
2151xfs_da_compname(
2152 struct xfs_da_args *args,
2153 const unsigned char *name,
2154 int len)
2155{
2156 return (args->namelen == len && memcmp(args->name, name, len) == 0) ?
2157 XFS_CMP_EXACT : XFS_CMP_DIFFERENT;
2158}
2159
2160int
2161xfs_da_grow_inode_int(
2162 struct xfs_da_args *args,
2163 xfs_fileoff_t *bno,
2164 int count)
2165{
2166 struct xfs_trans *tp = args->trans;
2167 struct xfs_inode *dp = args->dp;
2168 int w = args->whichfork;
2169 xfs_rfsblock_t nblks = dp->i_nblocks;
2170 struct xfs_bmbt_irec map, *mapp;
2171 int nmap, error, got, i, mapi;
2172
2173 /*
2174 * Find a spot in the file space to put the new block.
2175 */
2176 error = xfs_bmap_first_unused(tp, dp, count, bno, w);
2177 if (error)
2178 return error;
2179
2180 /*
2181 * Try mapping it in one filesystem block.
2182 */
2183 nmap = 1;
2184 error = xfs_bmapi_write(tp, dp, *bno, count,
2185 xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG,
2186 args->total, &map, &nmap);
2187 if (error)
2188 return error;
2189
2190 ASSERT(nmap <= 1);
2191 if (nmap == 1) {
2192 mapp = ↦
2193 mapi = 1;
2194 } else if (nmap == 0 && count > 1) {
2195 xfs_fileoff_t b;
2196 int c;
2197
2198 /*
2199 * If we didn't get it and the block might work if fragmented,
2200 * try without the CONTIG flag. Loop until we get it all.
2201 */
2202 mapp = kmalloc(sizeof(*mapp) * count,
2203 GFP_KERNEL | __GFP_NOFAIL);
2204 for (b = *bno, mapi = 0; b < *bno + count; ) {
2205 c = (int)(*bno + count - b);
2206 nmap = min(XFS_BMAP_MAX_NMAP, c);
2207 error = xfs_bmapi_write(tp, dp, b, c,
2208 xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
2209 args->total, &mapp[mapi], &nmap);
2210 if (error)
2211 goto out_free_map;
2212 if (nmap < 1)
2213 break;
2214 mapi += nmap;
2215 b = mapp[mapi - 1].br_startoff +
2216 mapp[mapi - 1].br_blockcount;
2217 }
2218 } else {
2219 mapi = 0;
2220 mapp = NULL;
2221 }
2222
2223 /*
2224 * Count the blocks we got, make sure it matches the total.
2225 */
2226 for (i = 0, got = 0; i < mapi; i++)
2227 got += mapp[i].br_blockcount;
2228 if (got != count || mapp[0].br_startoff != *bno ||
2229 mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
2230 *bno + count) {
2231 error = -ENOSPC;
2232 goto out_free_map;
2233 }
2234
2235 /* account for newly allocated blocks in reserved blocks total */
2236 args->total -= dp->i_nblocks - nblks;
2237
2238out_free_map:
2239 if (mapp != &map)
2240 kfree(mapp);
2241 return error;
2242}
2243
2244/*
2245 * Add a block to the btree ahead of the file.
2246 * Return the new block number to the caller.
2247 */
2248int
2249xfs_da_grow_inode(
2250 struct xfs_da_args *args,
2251 xfs_dablk_t *new_blkno)
2252{
2253 xfs_fileoff_t bno;
2254 int error;
2255
2256 trace_xfs_da_grow_inode(args);
2257
2258 bno = args->geo->leafblk;
2259 error = xfs_da_grow_inode_int(args, &bno, args->geo->fsbcount);
2260 if (!error)
2261 *new_blkno = (xfs_dablk_t)bno;
2262 return error;
2263}
2264
2265/*
2266 * Ick. We need to always be able to remove a btree block, even
2267 * if there's no space reservation because the filesystem is full.
2268 * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
2269 * It swaps the target block with the last block in the file. The
2270 * last block in the file can always be removed since it can't cause
2271 * a bmap btree split to do that.
2272 */
2273STATIC int
2274xfs_da3_swap_lastblock(
2275 struct xfs_da_args *args,
2276 xfs_dablk_t *dead_blknop,
2277 struct xfs_buf **dead_bufp)
2278{
2279 struct xfs_da_blkinfo *dead_info;
2280 struct xfs_da_blkinfo *sib_info;
2281 struct xfs_da_intnode *par_node;
2282 struct xfs_da_intnode *dead_node;
2283 struct xfs_dir2_leaf *dead_leaf2;
2284 struct xfs_da_node_entry *btree;
2285 struct xfs_da3_icnode_hdr par_hdr;
2286 struct xfs_inode *dp;
2287 struct xfs_trans *tp;
2288 struct xfs_mount *mp;
2289 struct xfs_buf *dead_buf;
2290 struct xfs_buf *last_buf;
2291 struct xfs_buf *sib_buf;
2292 struct xfs_buf *par_buf;
2293 xfs_dahash_t dead_hash;
2294 xfs_fileoff_t lastoff;
2295 xfs_dablk_t dead_blkno;
2296 xfs_dablk_t last_blkno;
2297 xfs_dablk_t sib_blkno;
2298 xfs_dablk_t par_blkno;
2299 int error;
2300 int w;
2301 int entno;
2302 int level;
2303 int dead_level;
2304
2305 trace_xfs_da_swap_lastblock(args);
2306
2307 dead_buf = *dead_bufp;
2308 dead_blkno = *dead_blknop;
2309 tp = args->trans;
2310 dp = args->dp;
2311 w = args->whichfork;
2312 ASSERT(w == XFS_DATA_FORK);
2313 mp = dp->i_mount;
2314 lastoff = args->geo->freeblk;
2315 error = xfs_bmap_last_before(tp, dp, &lastoff, w);
2316 if (error)
2317 return error;
2318 if (XFS_IS_CORRUPT(mp, lastoff == 0)) {
2319 xfs_da_mark_sick(args);
2320 return -EFSCORRUPTED;
2321 }
2322 /*
2323 * Read the last block in the btree space.
2324 */
2325 last_blkno = (xfs_dablk_t)lastoff - args->geo->fsbcount;
2326 error = xfs_da3_node_read(tp, dp, last_blkno, &last_buf, w);
2327 if (error)
2328 return error;
2329 /*
2330 * Copy the last block into the dead buffer and log it.
2331 */
2332 xfs_da_buf_copy(dead_buf, last_buf, args->geo->blksize);
2333 xfs_trans_log_buf(tp, dead_buf, 0, args->geo->blksize - 1);
2334 dead_info = dead_buf->b_addr;
2335
2336 /*
2337 * Get values from the moved block.
2338 */
2339 if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
2340 dead_info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) {
2341 struct xfs_dir3_icleaf_hdr leafhdr;
2342 struct xfs_dir2_leaf_entry *ents;
2343
2344 dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
2345 xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr,
2346 dead_leaf2);
2347 ents = leafhdr.ents;
2348 dead_level = 0;
2349 dead_hash = be32_to_cpu(ents[leafhdr.count - 1].hashval);
2350 } else {
2351 struct xfs_da3_icnode_hdr deadhdr;
2352
2353 dead_node = (xfs_da_intnode_t *)dead_info;
2354 xfs_da3_node_hdr_from_disk(dp->i_mount, &deadhdr, dead_node);
2355 btree = deadhdr.btree;
2356 dead_level = deadhdr.level;
2357 dead_hash = be32_to_cpu(btree[deadhdr.count - 1].hashval);
2358 }
2359 sib_buf = par_buf = NULL;
2360 /*
2361 * If the moved block has a left sibling, fix up the pointers.
2362 */
2363 if ((sib_blkno = be32_to_cpu(dead_info->back))) {
2364 error = xfs_da3_node_read(tp, dp, sib_blkno, &sib_buf, w);
2365 if (error)
2366 goto done;
2367 sib_info = sib_buf->b_addr;
2368 if (XFS_IS_CORRUPT(mp,
2369 be32_to_cpu(sib_info->forw) != last_blkno ||
2370 sib_info->magic != dead_info->magic)) {
2371 xfs_da_mark_sick(args);
2372 error = -EFSCORRUPTED;
2373 goto done;
2374 }
2375 sib_info->forw = cpu_to_be32(dead_blkno);
2376 xfs_trans_log_buf(tp, sib_buf,
2377 XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
2378 sizeof(sib_info->forw)));
2379 sib_buf = NULL;
2380 }
2381 /*
2382 * If the moved block has a right sibling, fix up the pointers.
2383 */
2384 if ((sib_blkno = be32_to_cpu(dead_info->forw))) {
2385 error = xfs_da3_node_read(tp, dp, sib_blkno, &sib_buf, w);
2386 if (error)
2387 goto done;
2388 sib_info = sib_buf->b_addr;
2389 if (XFS_IS_CORRUPT(mp,
2390 be32_to_cpu(sib_info->back) != last_blkno ||
2391 sib_info->magic != dead_info->magic)) {
2392 xfs_da_mark_sick(args);
2393 error = -EFSCORRUPTED;
2394 goto done;
2395 }
2396 sib_info->back = cpu_to_be32(dead_blkno);
2397 xfs_trans_log_buf(tp, sib_buf,
2398 XFS_DA_LOGRANGE(sib_info, &sib_info->back,
2399 sizeof(sib_info->back)));
2400 sib_buf = NULL;
2401 }
2402 par_blkno = args->geo->leafblk;
2403 level = -1;
2404 /*
2405 * Walk down the tree looking for the parent of the moved block.
2406 */
2407 for (;;) {
2408 error = xfs_da3_node_read(tp, dp, par_blkno, &par_buf, w);
2409 if (error)
2410 goto done;
2411 par_node = par_buf->b_addr;
2412 xfs_da3_node_hdr_from_disk(dp->i_mount, &par_hdr, par_node);
2413 if (XFS_IS_CORRUPT(mp,
2414 level >= 0 && level != par_hdr.level + 1)) {
2415 xfs_da_mark_sick(args);
2416 error = -EFSCORRUPTED;
2417 goto done;
2418 }
2419 level = par_hdr.level;
2420 btree = par_hdr.btree;
2421 for (entno = 0;
2422 entno < par_hdr.count &&
2423 be32_to_cpu(btree[entno].hashval) < dead_hash;
2424 entno++)
2425 continue;
2426 if (XFS_IS_CORRUPT(mp, entno == par_hdr.count)) {
2427 xfs_da_mark_sick(args);
2428 error = -EFSCORRUPTED;
2429 goto done;
2430 }
2431 par_blkno = be32_to_cpu(btree[entno].before);
2432 if (level == dead_level + 1)
2433 break;
2434 xfs_trans_brelse(tp, par_buf);
2435 par_buf = NULL;
2436 }
2437 /*
2438 * We're in the right parent block.
2439 * Look for the right entry.
2440 */
2441 for (;;) {
2442 for (;
2443 entno < par_hdr.count &&
2444 be32_to_cpu(btree[entno].before) != last_blkno;
2445 entno++)
2446 continue;
2447 if (entno < par_hdr.count)
2448 break;
2449 par_blkno = par_hdr.forw;
2450 xfs_trans_brelse(tp, par_buf);
2451 par_buf = NULL;
2452 if (XFS_IS_CORRUPT(mp, par_blkno == 0)) {
2453 xfs_da_mark_sick(args);
2454 error = -EFSCORRUPTED;
2455 goto done;
2456 }
2457 error = xfs_da3_node_read(tp, dp, par_blkno, &par_buf, w);
2458 if (error)
2459 goto done;
2460 par_node = par_buf->b_addr;
2461 xfs_da3_node_hdr_from_disk(dp->i_mount, &par_hdr, par_node);
2462 if (XFS_IS_CORRUPT(mp, par_hdr.level != level)) {
2463 xfs_da_mark_sick(args);
2464 error = -EFSCORRUPTED;
2465 goto done;
2466 }
2467 btree = par_hdr.btree;
2468 entno = 0;
2469 }
2470 /*
2471 * Update the parent entry pointing to the moved block.
2472 */
2473 btree[entno].before = cpu_to_be32(dead_blkno);
2474 xfs_trans_log_buf(tp, par_buf,
2475 XFS_DA_LOGRANGE(par_node, &btree[entno].before,
2476 sizeof(btree[entno].before)));
2477 *dead_blknop = last_blkno;
2478 *dead_bufp = last_buf;
2479 return 0;
2480done:
2481 if (par_buf)
2482 xfs_trans_brelse(tp, par_buf);
2483 if (sib_buf)
2484 xfs_trans_brelse(tp, sib_buf);
2485 xfs_trans_brelse(tp, last_buf);
2486 return error;
2487}
2488
2489/*
2490 * Remove a btree block from a directory or attribute.
2491 */
2492int
2493xfs_da_shrink_inode(
2494 struct xfs_da_args *args,
2495 xfs_dablk_t dead_blkno,
2496 struct xfs_buf *dead_buf)
2497{
2498 struct xfs_inode *dp;
2499 int done, error, w, count;
2500 struct xfs_trans *tp;
2501
2502 trace_xfs_da_shrink_inode(args);
2503
2504 dp = args->dp;
2505 w = args->whichfork;
2506 tp = args->trans;
2507 count = args->geo->fsbcount;
2508 for (;;) {
2509 /*
2510 * Remove extents. If we get ENOSPC for a dir we have to move
2511 * the last block to the place we want to kill.
2512 */
2513 error = xfs_bunmapi(tp, dp, dead_blkno, count,
2514 xfs_bmapi_aflag(w), 0, &done);
2515 if (error == -ENOSPC) {
2516 if (w != XFS_DATA_FORK)
2517 break;
2518 error = xfs_da3_swap_lastblock(args, &dead_blkno,
2519 &dead_buf);
2520 if (error)
2521 break;
2522 } else {
2523 break;
2524 }
2525 }
2526 xfs_trans_binval(tp, dead_buf);
2527 return error;
2528}
2529
2530static int
2531xfs_dabuf_map(
2532 struct xfs_inode *dp,
2533 xfs_dablk_t bno,
2534 unsigned int flags,
2535 int whichfork,
2536 struct xfs_buf_map **mapp,
2537 int *nmaps)
2538{
2539 struct xfs_mount *mp = dp->i_mount;
2540 int nfsb = xfs_dabuf_nfsb(mp, whichfork);
2541 struct xfs_bmbt_irec irec, *irecs = &irec;
2542 struct xfs_buf_map *map = *mapp;
2543 xfs_fileoff_t off = bno;
2544 int error = 0, nirecs, i;
2545
2546 if (nfsb > 1)
2547 irecs = kzalloc(sizeof(irec) * nfsb,
2548 GFP_KERNEL | __GFP_NOLOCKDEP | __GFP_NOFAIL);
2549
2550 nirecs = nfsb;
2551 error = xfs_bmapi_read(dp, bno, nfsb, irecs, &nirecs,
2552 xfs_bmapi_aflag(whichfork));
2553 if (error)
2554 goto out_free_irecs;
2555
2556 /*
2557 * Use the caller provided map for the single map case, else allocate a
2558 * larger one that needs to be free by the caller.
2559 */
2560 if (nirecs > 1) {
2561 map = kzalloc(nirecs * sizeof(struct xfs_buf_map),
2562 GFP_KERNEL | __GFP_NOLOCKDEP | __GFP_NOFAIL);
2563 if (!map) {
2564 error = -ENOMEM;
2565 goto out_free_irecs;
2566 }
2567 *mapp = map;
2568 }
2569
2570 for (i = 0; i < nirecs; i++) {
2571 if (irecs[i].br_startblock == HOLESTARTBLOCK ||
2572 irecs[i].br_startblock == DELAYSTARTBLOCK)
2573 goto invalid_mapping;
2574 if (off != irecs[i].br_startoff)
2575 goto invalid_mapping;
2576
2577 map[i].bm_bn = XFS_FSB_TO_DADDR(mp, irecs[i].br_startblock);
2578 map[i].bm_len = XFS_FSB_TO_BB(mp, irecs[i].br_blockcount);
2579 off += irecs[i].br_blockcount;
2580 }
2581
2582 if (off != bno + nfsb)
2583 goto invalid_mapping;
2584
2585 *nmaps = nirecs;
2586out_free_irecs:
2587 if (irecs != &irec)
2588 kfree(irecs);
2589 return error;
2590
2591invalid_mapping:
2592 /* Caller ok with no mapping. */
2593 if (XFS_IS_CORRUPT(mp, !(flags & XFS_DABUF_MAP_HOLE_OK))) {
2594 xfs_dirattr_mark_sick(dp, whichfork);
2595 error = -EFSCORRUPTED;
2596 if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
2597 xfs_alert(mp, "%s: bno %u inode %llu",
2598 __func__, bno, dp->i_ino);
2599
2600 for (i = 0; i < nirecs; i++) {
2601 xfs_alert(mp,
2602"[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d",
2603 i, irecs[i].br_startoff,
2604 irecs[i].br_startblock,
2605 irecs[i].br_blockcount,
2606 irecs[i].br_state);
2607 }
2608 }
2609 } else {
2610 *nmaps = 0;
2611 }
2612 goto out_free_irecs;
2613}
2614
2615/*
2616 * Get a buffer for the dir/attr block.
2617 */
2618int
2619xfs_da_get_buf(
2620 struct xfs_trans *tp,
2621 struct xfs_inode *dp,
2622 xfs_dablk_t bno,
2623 struct xfs_buf **bpp,
2624 int whichfork)
2625{
2626 struct xfs_mount *mp = dp->i_mount;
2627 struct xfs_buf *bp;
2628 struct xfs_buf_map map, *mapp = ↦
2629 int nmap = 1;
2630 int error;
2631
2632 *bpp = NULL;
2633 error = xfs_dabuf_map(dp, bno, 0, whichfork, &mapp, &nmap);
2634 if (error || nmap == 0)
2635 goto out_free;
2636
2637 error = xfs_trans_get_buf_map(tp, mp->m_ddev_targp, mapp, nmap, 0, &bp);
2638 if (error)
2639 goto out_free;
2640
2641 *bpp = bp;
2642
2643out_free:
2644 if (mapp != &map)
2645 kfree(mapp);
2646
2647 return error;
2648}
2649
2650/*
2651 * Get a buffer for the dir/attr block, fill in the contents.
2652 */
2653int
2654xfs_da_read_buf(
2655 struct xfs_trans *tp,
2656 struct xfs_inode *dp,
2657 xfs_dablk_t bno,
2658 unsigned int flags,
2659 struct xfs_buf **bpp,
2660 int whichfork,
2661 const struct xfs_buf_ops *ops)
2662{
2663 struct xfs_mount *mp = dp->i_mount;
2664 struct xfs_buf *bp;
2665 struct xfs_buf_map map, *mapp = ↦
2666 int nmap = 1;
2667 int error;
2668
2669 *bpp = NULL;
2670 error = xfs_dabuf_map(dp, bno, flags, whichfork, &mapp, &nmap);
2671 if (error || !nmap)
2672 goto out_free;
2673
2674 error = xfs_trans_read_buf_map(mp, tp, mp->m_ddev_targp, mapp, nmap, 0,
2675 &bp, ops);
2676 if (xfs_metadata_is_sick(error))
2677 xfs_dirattr_mark_sick(dp, whichfork);
2678 if (error)
2679 goto out_free;
2680
2681 if (whichfork == XFS_ATTR_FORK)
2682 xfs_buf_set_ref(bp, XFS_ATTR_BTREE_REF);
2683 else
2684 xfs_buf_set_ref(bp, XFS_DIR_BTREE_REF);
2685 *bpp = bp;
2686out_free:
2687 if (mapp != &map)
2688 kfree(mapp);
2689
2690 return error;
2691}
2692
2693/*
2694 * Readahead the dir/attr block.
2695 */
2696int
2697xfs_da_reada_buf(
2698 struct xfs_inode *dp,
2699 xfs_dablk_t bno,
2700 unsigned int flags,
2701 int whichfork,
2702 const struct xfs_buf_ops *ops)
2703{
2704 struct xfs_buf_map map;
2705 struct xfs_buf_map *mapp;
2706 int nmap;
2707 int error;
2708
2709 mapp = ↦
2710 nmap = 1;
2711 error = xfs_dabuf_map(dp, bno, flags, whichfork, &mapp, &nmap);
2712 if (error || !nmap)
2713 goto out_free;
2714
2715 xfs_buf_readahead_map(dp->i_mount->m_ddev_targp, mapp, nmap, ops);
2716
2717out_free:
2718 if (mapp != &map)
2719 kfree(mapp);
2720
2721 return error;
2722}