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