Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6#include "xfs.h"
7#include "xfs_fs.h"
8#include "xfs_shared.h"
9#include "xfs_format.h"
10#include "xfs_log_format.h"
11#include "xfs_trans_resv.h"
12#include "xfs_bit.h"
13#include "xfs_mount.h"
14#include "xfs_inode.h"
15#include "xfs_trans.h"
16#include "xfs_alloc.h"
17#include "xfs_btree.h"
18#include "xfs_bmap_btree.h"
19#include "xfs_bmap.h"
20#include "xfs_error.h"
21#include "xfs_quota.h"
22#include "xfs_trace.h"
23#include "xfs_rmap.h"
24
25static struct kmem_cache *xfs_bmbt_cur_cache;
26
27/*
28 * Convert on-disk form of btree root to in-memory form.
29 */
30void
31xfs_bmdr_to_bmbt(
32 struct xfs_inode *ip,
33 xfs_bmdr_block_t *dblock,
34 int dblocklen,
35 struct xfs_btree_block *rblock,
36 int rblocklen)
37{
38 struct xfs_mount *mp = ip->i_mount;
39 int dmxr;
40 xfs_bmbt_key_t *fkp;
41 __be64 *fpp;
42 xfs_bmbt_key_t *tkp;
43 __be64 *tpp;
44
45 xfs_btree_init_block_int(mp, rblock, XFS_BUF_DADDR_NULL,
46 XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
47 XFS_BTREE_LONG_PTRS);
48 rblock->bb_level = dblock->bb_level;
49 ASSERT(be16_to_cpu(rblock->bb_level) > 0);
50 rblock->bb_numrecs = dblock->bb_numrecs;
51 dmxr = xfs_bmdr_maxrecs(dblocklen, 0);
52 fkp = XFS_BMDR_KEY_ADDR(dblock, 1);
53 tkp = XFS_BMBT_KEY_ADDR(mp, rblock, 1);
54 fpp = XFS_BMDR_PTR_ADDR(dblock, 1, dmxr);
55 tpp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, rblocklen);
56 dmxr = be16_to_cpu(dblock->bb_numrecs);
57 memcpy(tkp, fkp, sizeof(*fkp) * dmxr);
58 memcpy(tpp, fpp, sizeof(*fpp) * dmxr);
59}
60
61void
62xfs_bmbt_disk_get_all(
63 const struct xfs_bmbt_rec *rec,
64 struct xfs_bmbt_irec *irec)
65{
66 uint64_t l0 = get_unaligned_be64(&rec->l0);
67 uint64_t l1 = get_unaligned_be64(&rec->l1);
68
69 irec->br_startoff = (l0 & xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN)) >> 9;
70 irec->br_startblock = ((l0 & xfs_mask64lo(9)) << 43) | (l1 >> 21);
71 irec->br_blockcount = l1 & xfs_mask64lo(21);
72 if (l0 >> (64 - BMBT_EXNTFLAG_BITLEN))
73 irec->br_state = XFS_EXT_UNWRITTEN;
74 else
75 irec->br_state = XFS_EXT_NORM;
76}
77
78/*
79 * Extract the blockcount field from an on disk bmap extent record.
80 */
81xfs_filblks_t
82xfs_bmbt_disk_get_blockcount(
83 const struct xfs_bmbt_rec *r)
84{
85 return (xfs_filblks_t)(be64_to_cpu(r->l1) & xfs_mask64lo(21));
86}
87
88/*
89 * Extract the startoff field from a disk format bmap extent record.
90 */
91xfs_fileoff_t
92xfs_bmbt_disk_get_startoff(
93 const struct xfs_bmbt_rec *r)
94{
95 return ((xfs_fileoff_t)be64_to_cpu(r->l0) &
96 xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN)) >> 9;
97}
98
99/*
100 * Set all the fields in a bmap extent record from the uncompressed form.
101 */
102void
103xfs_bmbt_disk_set_all(
104 struct xfs_bmbt_rec *r,
105 struct xfs_bmbt_irec *s)
106{
107 int extent_flag = (s->br_state != XFS_EXT_NORM);
108
109 ASSERT(s->br_state == XFS_EXT_NORM || s->br_state == XFS_EXT_UNWRITTEN);
110 ASSERT(!(s->br_startoff & xfs_mask64hi(64-BMBT_STARTOFF_BITLEN)));
111 ASSERT(!(s->br_blockcount & xfs_mask64hi(64-BMBT_BLOCKCOUNT_BITLEN)));
112 ASSERT(!(s->br_startblock & xfs_mask64hi(64-BMBT_STARTBLOCK_BITLEN)));
113
114 put_unaligned_be64(
115 ((xfs_bmbt_rec_base_t)extent_flag << 63) |
116 ((xfs_bmbt_rec_base_t)s->br_startoff << 9) |
117 ((xfs_bmbt_rec_base_t)s->br_startblock >> 43), &r->l0);
118 put_unaligned_be64(
119 ((xfs_bmbt_rec_base_t)s->br_startblock << 21) |
120 ((xfs_bmbt_rec_base_t)s->br_blockcount &
121 (xfs_bmbt_rec_base_t)xfs_mask64lo(21)), &r->l1);
122}
123
124/*
125 * Convert in-memory form of btree root to on-disk form.
126 */
127void
128xfs_bmbt_to_bmdr(
129 struct xfs_mount *mp,
130 struct xfs_btree_block *rblock,
131 int rblocklen,
132 xfs_bmdr_block_t *dblock,
133 int dblocklen)
134{
135 int dmxr;
136 xfs_bmbt_key_t *fkp;
137 __be64 *fpp;
138 xfs_bmbt_key_t *tkp;
139 __be64 *tpp;
140
141 if (xfs_has_crc(mp)) {
142 ASSERT(rblock->bb_magic == cpu_to_be32(XFS_BMAP_CRC_MAGIC));
143 ASSERT(uuid_equal(&rblock->bb_u.l.bb_uuid,
144 &mp->m_sb.sb_meta_uuid));
145 ASSERT(rblock->bb_u.l.bb_blkno ==
146 cpu_to_be64(XFS_BUF_DADDR_NULL));
147 } else
148 ASSERT(rblock->bb_magic == cpu_to_be32(XFS_BMAP_MAGIC));
149 ASSERT(rblock->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK));
150 ASSERT(rblock->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK));
151 ASSERT(rblock->bb_level != 0);
152 dblock->bb_level = rblock->bb_level;
153 dblock->bb_numrecs = rblock->bb_numrecs;
154 dmxr = xfs_bmdr_maxrecs(dblocklen, 0);
155 fkp = XFS_BMBT_KEY_ADDR(mp, rblock, 1);
156 tkp = XFS_BMDR_KEY_ADDR(dblock, 1);
157 fpp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, rblocklen);
158 tpp = XFS_BMDR_PTR_ADDR(dblock, 1, dmxr);
159 dmxr = be16_to_cpu(dblock->bb_numrecs);
160 memcpy(tkp, fkp, sizeof(*fkp) * dmxr);
161 memcpy(tpp, fpp, sizeof(*fpp) * dmxr);
162}
163
164STATIC struct xfs_btree_cur *
165xfs_bmbt_dup_cursor(
166 struct xfs_btree_cur *cur)
167{
168 struct xfs_btree_cur *new;
169
170 new = xfs_bmbt_init_cursor(cur->bc_mp, cur->bc_tp,
171 cur->bc_ino.ip, cur->bc_ino.whichfork);
172
173 /*
174 * Copy the firstblock, dfops, and flags values,
175 * since init cursor doesn't get them.
176 */
177 new->bc_ino.flags = cur->bc_ino.flags;
178
179 return new;
180}
181
182STATIC void
183xfs_bmbt_update_cursor(
184 struct xfs_btree_cur *src,
185 struct xfs_btree_cur *dst)
186{
187 ASSERT((dst->bc_tp->t_firstblock != NULLFSBLOCK) ||
188 (dst->bc_ino.ip->i_diflags & XFS_DIFLAG_REALTIME));
189
190 dst->bc_ino.allocated += src->bc_ino.allocated;
191 dst->bc_tp->t_firstblock = src->bc_tp->t_firstblock;
192
193 src->bc_ino.allocated = 0;
194}
195
196STATIC int
197xfs_bmbt_alloc_block(
198 struct xfs_btree_cur *cur,
199 const union xfs_btree_ptr *start,
200 union xfs_btree_ptr *new,
201 int *stat)
202{
203 xfs_alloc_arg_t args; /* block allocation args */
204 int error; /* error return value */
205
206 memset(&args, 0, sizeof(args));
207 args.tp = cur->bc_tp;
208 args.mp = cur->bc_mp;
209 args.fsbno = cur->bc_tp->t_firstblock;
210 xfs_rmap_ino_bmbt_owner(&args.oinfo, cur->bc_ino.ip->i_ino,
211 cur->bc_ino.whichfork);
212
213 if (args.fsbno == NULLFSBLOCK) {
214 args.fsbno = be64_to_cpu(start->l);
215 args.type = XFS_ALLOCTYPE_START_BNO;
216 /*
217 * Make sure there is sufficient room left in the AG to
218 * complete a full tree split for an extent insert. If
219 * we are converting the middle part of an extent then
220 * we may need space for two tree splits.
221 *
222 * We are relying on the caller to make the correct block
223 * reservation for this operation to succeed. If the
224 * reservation amount is insufficient then we may fail a
225 * block allocation here and corrupt the filesystem.
226 */
227 args.minleft = args.tp->t_blk_res;
228 } else if (cur->bc_tp->t_flags & XFS_TRANS_LOWMODE) {
229 args.type = XFS_ALLOCTYPE_START_BNO;
230 } else {
231 args.type = XFS_ALLOCTYPE_NEAR_BNO;
232 }
233
234 args.minlen = args.maxlen = args.prod = 1;
235 args.wasdel = cur->bc_ino.flags & XFS_BTCUR_BMBT_WASDEL;
236 if (!args.wasdel && args.tp->t_blk_res == 0) {
237 error = -ENOSPC;
238 goto error0;
239 }
240 error = xfs_alloc_vextent(&args);
241 if (error)
242 goto error0;
243
244 if (args.fsbno == NULLFSBLOCK && args.minleft) {
245 /*
246 * Could not find an AG with enough free space to satisfy
247 * a full btree split. Try again and if
248 * successful activate the lowspace algorithm.
249 */
250 args.fsbno = 0;
251 args.type = XFS_ALLOCTYPE_FIRST_AG;
252 error = xfs_alloc_vextent(&args);
253 if (error)
254 goto error0;
255 cur->bc_tp->t_flags |= XFS_TRANS_LOWMODE;
256 }
257 if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
258 *stat = 0;
259 return 0;
260 }
261
262 ASSERT(args.len == 1);
263 cur->bc_tp->t_firstblock = args.fsbno;
264 cur->bc_ino.allocated++;
265 cur->bc_ino.ip->i_nblocks++;
266 xfs_trans_log_inode(args.tp, cur->bc_ino.ip, XFS_ILOG_CORE);
267 xfs_trans_mod_dquot_byino(args.tp, cur->bc_ino.ip,
268 XFS_TRANS_DQ_BCOUNT, 1L);
269
270 new->l = cpu_to_be64(args.fsbno);
271
272 *stat = 1;
273 return 0;
274
275 error0:
276 return error;
277}
278
279STATIC int
280xfs_bmbt_free_block(
281 struct xfs_btree_cur *cur,
282 struct xfs_buf *bp)
283{
284 struct xfs_mount *mp = cur->bc_mp;
285 struct xfs_inode *ip = cur->bc_ino.ip;
286 struct xfs_trans *tp = cur->bc_tp;
287 xfs_fsblock_t fsbno = XFS_DADDR_TO_FSB(mp, xfs_buf_daddr(bp));
288 struct xfs_owner_info oinfo;
289
290 xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, cur->bc_ino.whichfork);
291 xfs_free_extent_later(cur->bc_tp, fsbno, 1, &oinfo);
292 ip->i_nblocks--;
293
294 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
295 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
296 return 0;
297}
298
299STATIC int
300xfs_bmbt_get_minrecs(
301 struct xfs_btree_cur *cur,
302 int level)
303{
304 if (level == cur->bc_nlevels - 1) {
305 struct xfs_ifork *ifp;
306
307 ifp = xfs_ifork_ptr(cur->bc_ino.ip,
308 cur->bc_ino.whichfork);
309
310 return xfs_bmbt_maxrecs(cur->bc_mp,
311 ifp->if_broot_bytes, level == 0) / 2;
312 }
313
314 return cur->bc_mp->m_bmap_dmnr[level != 0];
315}
316
317int
318xfs_bmbt_get_maxrecs(
319 struct xfs_btree_cur *cur,
320 int level)
321{
322 if (level == cur->bc_nlevels - 1) {
323 struct xfs_ifork *ifp;
324
325 ifp = xfs_ifork_ptr(cur->bc_ino.ip,
326 cur->bc_ino.whichfork);
327
328 return xfs_bmbt_maxrecs(cur->bc_mp,
329 ifp->if_broot_bytes, level == 0);
330 }
331
332 return cur->bc_mp->m_bmap_dmxr[level != 0];
333
334}
335
336/*
337 * Get the maximum records we could store in the on-disk format.
338 *
339 * For non-root nodes this is equivalent to xfs_bmbt_get_maxrecs, but
340 * for the root node this checks the available space in the dinode fork
341 * so that we can resize the in-memory buffer to match it. After a
342 * resize to the maximum size this function returns the same value
343 * as xfs_bmbt_get_maxrecs for the root node, too.
344 */
345STATIC int
346xfs_bmbt_get_dmaxrecs(
347 struct xfs_btree_cur *cur,
348 int level)
349{
350 if (level != cur->bc_nlevels - 1)
351 return cur->bc_mp->m_bmap_dmxr[level != 0];
352 return xfs_bmdr_maxrecs(cur->bc_ino.forksize, level == 0);
353}
354
355STATIC void
356xfs_bmbt_init_key_from_rec(
357 union xfs_btree_key *key,
358 const union xfs_btree_rec *rec)
359{
360 key->bmbt.br_startoff =
361 cpu_to_be64(xfs_bmbt_disk_get_startoff(&rec->bmbt));
362}
363
364STATIC void
365xfs_bmbt_init_high_key_from_rec(
366 union xfs_btree_key *key,
367 const union xfs_btree_rec *rec)
368{
369 key->bmbt.br_startoff = cpu_to_be64(
370 xfs_bmbt_disk_get_startoff(&rec->bmbt) +
371 xfs_bmbt_disk_get_blockcount(&rec->bmbt) - 1);
372}
373
374STATIC void
375xfs_bmbt_init_rec_from_cur(
376 struct xfs_btree_cur *cur,
377 union xfs_btree_rec *rec)
378{
379 xfs_bmbt_disk_set_all(&rec->bmbt, &cur->bc_rec.b);
380}
381
382STATIC void
383xfs_bmbt_init_ptr_from_cur(
384 struct xfs_btree_cur *cur,
385 union xfs_btree_ptr *ptr)
386{
387 ptr->l = 0;
388}
389
390STATIC int64_t
391xfs_bmbt_key_diff(
392 struct xfs_btree_cur *cur,
393 const union xfs_btree_key *key)
394{
395 return (int64_t)be64_to_cpu(key->bmbt.br_startoff) -
396 cur->bc_rec.b.br_startoff;
397}
398
399STATIC int64_t
400xfs_bmbt_diff_two_keys(
401 struct xfs_btree_cur *cur,
402 const union xfs_btree_key *k1,
403 const union xfs_btree_key *k2)
404{
405 uint64_t a = be64_to_cpu(k1->bmbt.br_startoff);
406 uint64_t b = be64_to_cpu(k2->bmbt.br_startoff);
407
408 /*
409 * Note: This routine previously casted a and b to int64 and subtracted
410 * them to generate a result. This lead to problems if b was the
411 * "maximum" key value (all ones) being signed incorrectly, hence this
412 * somewhat less efficient version.
413 */
414 if (a > b)
415 return 1;
416 if (b > a)
417 return -1;
418 return 0;
419}
420
421static xfs_failaddr_t
422xfs_bmbt_verify(
423 struct xfs_buf *bp)
424{
425 struct xfs_mount *mp = bp->b_mount;
426 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
427 xfs_failaddr_t fa;
428 unsigned int level;
429
430 if (!xfs_verify_magic(bp, block->bb_magic))
431 return __this_address;
432
433 if (xfs_has_crc(mp)) {
434 /*
435 * XXX: need a better way of verifying the owner here. Right now
436 * just make sure there has been one set.
437 */
438 fa = xfs_btree_lblock_v5hdr_verify(bp, XFS_RMAP_OWN_UNKNOWN);
439 if (fa)
440 return fa;
441 }
442
443 /*
444 * numrecs and level verification.
445 *
446 * We don't know what fork we belong to, so just verify that the level
447 * is less than the maximum of the two. Later checks will be more
448 * precise.
449 */
450 level = be16_to_cpu(block->bb_level);
451 if (level > max(mp->m_bm_maxlevels[0], mp->m_bm_maxlevels[1]))
452 return __this_address;
453
454 return xfs_btree_lblock_verify(bp, mp->m_bmap_dmxr[level != 0]);
455}
456
457static void
458xfs_bmbt_read_verify(
459 struct xfs_buf *bp)
460{
461 xfs_failaddr_t fa;
462
463 if (!xfs_btree_lblock_verify_crc(bp))
464 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
465 else {
466 fa = xfs_bmbt_verify(bp);
467 if (fa)
468 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
469 }
470
471 if (bp->b_error)
472 trace_xfs_btree_corrupt(bp, _RET_IP_);
473}
474
475static void
476xfs_bmbt_write_verify(
477 struct xfs_buf *bp)
478{
479 xfs_failaddr_t fa;
480
481 fa = xfs_bmbt_verify(bp);
482 if (fa) {
483 trace_xfs_btree_corrupt(bp, _RET_IP_);
484 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
485 return;
486 }
487 xfs_btree_lblock_calc_crc(bp);
488}
489
490const struct xfs_buf_ops xfs_bmbt_buf_ops = {
491 .name = "xfs_bmbt",
492 .magic = { cpu_to_be32(XFS_BMAP_MAGIC),
493 cpu_to_be32(XFS_BMAP_CRC_MAGIC) },
494 .verify_read = xfs_bmbt_read_verify,
495 .verify_write = xfs_bmbt_write_verify,
496 .verify_struct = xfs_bmbt_verify,
497};
498
499
500STATIC int
501xfs_bmbt_keys_inorder(
502 struct xfs_btree_cur *cur,
503 const union xfs_btree_key *k1,
504 const union xfs_btree_key *k2)
505{
506 return be64_to_cpu(k1->bmbt.br_startoff) <
507 be64_to_cpu(k2->bmbt.br_startoff);
508}
509
510STATIC int
511xfs_bmbt_recs_inorder(
512 struct xfs_btree_cur *cur,
513 const union xfs_btree_rec *r1,
514 const union xfs_btree_rec *r2)
515{
516 return xfs_bmbt_disk_get_startoff(&r1->bmbt) +
517 xfs_bmbt_disk_get_blockcount(&r1->bmbt) <=
518 xfs_bmbt_disk_get_startoff(&r2->bmbt);
519}
520
521static const struct xfs_btree_ops xfs_bmbt_ops = {
522 .rec_len = sizeof(xfs_bmbt_rec_t),
523 .key_len = sizeof(xfs_bmbt_key_t),
524
525 .dup_cursor = xfs_bmbt_dup_cursor,
526 .update_cursor = xfs_bmbt_update_cursor,
527 .alloc_block = xfs_bmbt_alloc_block,
528 .free_block = xfs_bmbt_free_block,
529 .get_maxrecs = xfs_bmbt_get_maxrecs,
530 .get_minrecs = xfs_bmbt_get_minrecs,
531 .get_dmaxrecs = xfs_bmbt_get_dmaxrecs,
532 .init_key_from_rec = xfs_bmbt_init_key_from_rec,
533 .init_high_key_from_rec = xfs_bmbt_init_high_key_from_rec,
534 .init_rec_from_cur = xfs_bmbt_init_rec_from_cur,
535 .init_ptr_from_cur = xfs_bmbt_init_ptr_from_cur,
536 .key_diff = xfs_bmbt_key_diff,
537 .diff_two_keys = xfs_bmbt_diff_two_keys,
538 .buf_ops = &xfs_bmbt_buf_ops,
539 .keys_inorder = xfs_bmbt_keys_inorder,
540 .recs_inorder = xfs_bmbt_recs_inorder,
541};
542
543/*
544 * Allocate a new bmap btree cursor.
545 */
546struct xfs_btree_cur * /* new bmap btree cursor */
547xfs_bmbt_init_cursor(
548 struct xfs_mount *mp, /* file system mount point */
549 struct xfs_trans *tp, /* transaction pointer */
550 struct xfs_inode *ip, /* inode owning the btree */
551 int whichfork) /* data or attr fork */
552{
553 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
554 struct xfs_btree_cur *cur;
555 ASSERT(whichfork != XFS_COW_FORK);
556
557 cur = xfs_btree_alloc_cursor(mp, tp, XFS_BTNUM_BMAP,
558 mp->m_bm_maxlevels[whichfork], xfs_bmbt_cur_cache);
559 cur->bc_nlevels = be16_to_cpu(ifp->if_broot->bb_level) + 1;
560 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_bmbt_2);
561
562 cur->bc_ops = &xfs_bmbt_ops;
563 cur->bc_flags = XFS_BTREE_LONG_PTRS | XFS_BTREE_ROOT_IN_INODE;
564 if (xfs_has_crc(mp))
565 cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
566
567 cur->bc_ino.forksize = xfs_inode_fork_size(ip, whichfork);
568 cur->bc_ino.ip = ip;
569 cur->bc_ino.allocated = 0;
570 cur->bc_ino.flags = 0;
571 cur->bc_ino.whichfork = whichfork;
572
573 return cur;
574}
575
576/* Calculate number of records in a block mapping btree block. */
577static inline unsigned int
578xfs_bmbt_block_maxrecs(
579 unsigned int blocklen,
580 bool leaf)
581{
582 if (leaf)
583 return blocklen / sizeof(xfs_bmbt_rec_t);
584 return blocklen / (sizeof(xfs_bmbt_key_t) + sizeof(xfs_bmbt_ptr_t));
585}
586
587/*
588 * Calculate number of records in a bmap btree block.
589 */
590int
591xfs_bmbt_maxrecs(
592 struct xfs_mount *mp,
593 int blocklen,
594 int leaf)
595{
596 blocklen -= XFS_BMBT_BLOCK_LEN(mp);
597 return xfs_bmbt_block_maxrecs(blocklen, leaf);
598}
599
600/*
601 * Calculate the maximum possible height of the btree that the on-disk format
602 * supports. This is used for sizing structures large enough to support every
603 * possible configuration of a filesystem that might get mounted.
604 */
605unsigned int
606xfs_bmbt_maxlevels_ondisk(void)
607{
608 unsigned int minrecs[2];
609 unsigned int blocklen;
610
611 blocklen = min(XFS_MIN_BLOCKSIZE - XFS_BTREE_SBLOCK_LEN,
612 XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN);
613
614 minrecs[0] = xfs_bmbt_block_maxrecs(blocklen, true) / 2;
615 minrecs[1] = xfs_bmbt_block_maxrecs(blocklen, false) / 2;
616
617 /* One extra level for the inode root. */
618 return xfs_btree_compute_maxlevels(minrecs,
619 XFS_MAX_EXTCNT_DATA_FORK_LARGE) + 1;
620}
621
622/*
623 * Calculate number of records in a bmap btree inode root.
624 */
625int
626xfs_bmdr_maxrecs(
627 int blocklen,
628 int leaf)
629{
630 blocklen -= sizeof(xfs_bmdr_block_t);
631
632 if (leaf)
633 return blocklen / sizeof(xfs_bmdr_rec_t);
634 return blocklen / (sizeof(xfs_bmdr_key_t) + sizeof(xfs_bmdr_ptr_t));
635}
636
637/*
638 * Change the owner of a btree format fork fo the inode passed in. Change it to
639 * the owner of that is passed in so that we can change owners before or after
640 * we switch forks between inodes. The operation that the caller is doing will
641 * determine whether is needs to change owner before or after the switch.
642 *
643 * For demand paged transactional modification, the fork switch should be done
644 * after reading in all the blocks, modifying them and pinning them in the
645 * transaction. For modification when the buffers are already pinned in memory,
646 * the fork switch can be done before changing the owner as we won't need to
647 * validate the owner until the btree buffers are unpinned and writes can occur
648 * again.
649 *
650 * For recovery based ownership change, there is no transactional context and
651 * so a buffer list must be supplied so that we can record the buffers that we
652 * modified for the caller to issue IO on.
653 */
654int
655xfs_bmbt_change_owner(
656 struct xfs_trans *tp,
657 struct xfs_inode *ip,
658 int whichfork,
659 xfs_ino_t new_owner,
660 struct list_head *buffer_list)
661{
662 struct xfs_btree_cur *cur;
663 int error;
664
665 ASSERT(tp || buffer_list);
666 ASSERT(!(tp && buffer_list));
667 ASSERT(xfs_ifork_ptr(ip, whichfork)->if_format == XFS_DINODE_FMT_BTREE);
668
669 cur = xfs_bmbt_init_cursor(ip->i_mount, tp, ip, whichfork);
670 cur->bc_ino.flags |= XFS_BTCUR_BMBT_INVALID_OWNER;
671
672 error = xfs_btree_change_owner(cur, new_owner, buffer_list);
673 xfs_btree_del_cursor(cur, error);
674 return error;
675}
676
677/* Calculate the bmap btree size for some records. */
678unsigned long long
679xfs_bmbt_calc_size(
680 struct xfs_mount *mp,
681 unsigned long long len)
682{
683 return xfs_btree_calc_size(mp->m_bmap_dmnr, len);
684}
685
686int __init
687xfs_bmbt_init_cur_cache(void)
688{
689 xfs_bmbt_cur_cache = kmem_cache_create("xfs_bmbt_cur",
690 xfs_btree_cur_sizeof(xfs_bmbt_maxlevels_ondisk()),
691 0, 0, NULL);
692
693 if (!xfs_bmbt_cur_cache)
694 return -ENOMEM;
695 return 0;
696}
697
698void
699xfs_bmbt_destroy_cur_cache(void)
700{
701 kmem_cache_destroy(xfs_bmbt_cur_cache);
702 xfs_bmbt_cur_cache = NULL;
703}
1/*
2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18#include "xfs.h"
19#include "xfs_fs.h"
20#include "xfs_shared.h"
21#include "xfs_format.h"
22#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
24#include "xfs_bit.h"
25#include "xfs_mount.h"
26#include "xfs_defer.h"
27#include "xfs_inode.h"
28#include "xfs_trans.h"
29#include "xfs_inode_item.h"
30#include "xfs_alloc.h"
31#include "xfs_btree.h"
32#include "xfs_bmap_btree.h"
33#include "xfs_bmap.h"
34#include "xfs_error.h"
35#include "xfs_quota.h"
36#include "xfs_trace.h"
37#include "xfs_cksum.h"
38#include "xfs_rmap.h"
39
40/*
41 * Convert on-disk form of btree root to in-memory form.
42 */
43void
44xfs_bmdr_to_bmbt(
45 struct xfs_inode *ip,
46 xfs_bmdr_block_t *dblock,
47 int dblocklen,
48 struct xfs_btree_block *rblock,
49 int rblocklen)
50{
51 struct xfs_mount *mp = ip->i_mount;
52 int dmxr;
53 xfs_bmbt_key_t *fkp;
54 __be64 *fpp;
55 xfs_bmbt_key_t *tkp;
56 __be64 *tpp;
57
58 xfs_btree_init_block_int(mp, rblock, XFS_BUF_DADDR_NULL,
59 XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
60 XFS_BTREE_LONG_PTRS);
61 rblock->bb_level = dblock->bb_level;
62 ASSERT(be16_to_cpu(rblock->bb_level) > 0);
63 rblock->bb_numrecs = dblock->bb_numrecs;
64 dmxr = xfs_bmdr_maxrecs(dblocklen, 0);
65 fkp = XFS_BMDR_KEY_ADDR(dblock, 1);
66 tkp = XFS_BMBT_KEY_ADDR(mp, rblock, 1);
67 fpp = XFS_BMDR_PTR_ADDR(dblock, 1, dmxr);
68 tpp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, rblocklen);
69 dmxr = be16_to_cpu(dblock->bb_numrecs);
70 memcpy(tkp, fkp, sizeof(*fkp) * dmxr);
71 memcpy(tpp, fpp, sizeof(*fpp) * dmxr);
72}
73
74void
75xfs_bmbt_disk_get_all(
76 struct xfs_bmbt_rec *rec,
77 struct xfs_bmbt_irec *irec)
78{
79 uint64_t l0 = get_unaligned_be64(&rec->l0);
80 uint64_t l1 = get_unaligned_be64(&rec->l1);
81
82 irec->br_startoff = (l0 & xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN)) >> 9;
83 irec->br_startblock = ((l0 & xfs_mask64lo(9)) << 43) | (l1 >> 21);
84 irec->br_blockcount = l1 & xfs_mask64lo(21);
85 if (l0 >> (64 - BMBT_EXNTFLAG_BITLEN))
86 irec->br_state = XFS_EXT_UNWRITTEN;
87 else
88 irec->br_state = XFS_EXT_NORM;
89}
90
91/*
92 * Extract the blockcount field from an on disk bmap extent record.
93 */
94xfs_filblks_t
95xfs_bmbt_disk_get_blockcount(
96 xfs_bmbt_rec_t *r)
97{
98 return (xfs_filblks_t)(be64_to_cpu(r->l1) & xfs_mask64lo(21));
99}
100
101/*
102 * Extract the startoff field from a disk format bmap extent record.
103 */
104xfs_fileoff_t
105xfs_bmbt_disk_get_startoff(
106 xfs_bmbt_rec_t *r)
107{
108 return ((xfs_fileoff_t)be64_to_cpu(r->l0) &
109 xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN)) >> 9;
110}
111
112/*
113 * Set all the fields in a bmap extent record from the uncompressed form.
114 */
115void
116xfs_bmbt_disk_set_all(
117 struct xfs_bmbt_rec *r,
118 struct xfs_bmbt_irec *s)
119{
120 int extent_flag = (s->br_state != XFS_EXT_NORM);
121
122 ASSERT(s->br_state == XFS_EXT_NORM || s->br_state == XFS_EXT_UNWRITTEN);
123 ASSERT(!(s->br_startoff & xfs_mask64hi(64-BMBT_STARTOFF_BITLEN)));
124 ASSERT(!(s->br_blockcount & xfs_mask64hi(64-BMBT_BLOCKCOUNT_BITLEN)));
125 ASSERT(!(s->br_startblock & xfs_mask64hi(64-BMBT_STARTBLOCK_BITLEN)));
126
127 put_unaligned_be64(
128 ((xfs_bmbt_rec_base_t)extent_flag << 63) |
129 ((xfs_bmbt_rec_base_t)s->br_startoff << 9) |
130 ((xfs_bmbt_rec_base_t)s->br_startblock >> 43), &r->l0);
131 put_unaligned_be64(
132 ((xfs_bmbt_rec_base_t)s->br_startblock << 21) |
133 ((xfs_bmbt_rec_base_t)s->br_blockcount &
134 (xfs_bmbt_rec_base_t)xfs_mask64lo(21)), &r->l1);
135}
136
137/*
138 * Convert in-memory form of btree root to on-disk form.
139 */
140void
141xfs_bmbt_to_bmdr(
142 struct xfs_mount *mp,
143 struct xfs_btree_block *rblock,
144 int rblocklen,
145 xfs_bmdr_block_t *dblock,
146 int dblocklen)
147{
148 int dmxr;
149 xfs_bmbt_key_t *fkp;
150 __be64 *fpp;
151 xfs_bmbt_key_t *tkp;
152 __be64 *tpp;
153
154 if (xfs_sb_version_hascrc(&mp->m_sb)) {
155 ASSERT(rblock->bb_magic == cpu_to_be32(XFS_BMAP_CRC_MAGIC));
156 ASSERT(uuid_equal(&rblock->bb_u.l.bb_uuid,
157 &mp->m_sb.sb_meta_uuid));
158 ASSERT(rblock->bb_u.l.bb_blkno ==
159 cpu_to_be64(XFS_BUF_DADDR_NULL));
160 } else
161 ASSERT(rblock->bb_magic == cpu_to_be32(XFS_BMAP_MAGIC));
162 ASSERT(rblock->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK));
163 ASSERT(rblock->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK));
164 ASSERT(rblock->bb_level != 0);
165 dblock->bb_level = rblock->bb_level;
166 dblock->bb_numrecs = rblock->bb_numrecs;
167 dmxr = xfs_bmdr_maxrecs(dblocklen, 0);
168 fkp = XFS_BMBT_KEY_ADDR(mp, rblock, 1);
169 tkp = XFS_BMDR_KEY_ADDR(dblock, 1);
170 fpp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, rblocklen);
171 tpp = XFS_BMDR_PTR_ADDR(dblock, 1, dmxr);
172 dmxr = be16_to_cpu(dblock->bb_numrecs);
173 memcpy(tkp, fkp, sizeof(*fkp) * dmxr);
174 memcpy(tpp, fpp, sizeof(*fpp) * dmxr);
175}
176
177STATIC struct xfs_btree_cur *
178xfs_bmbt_dup_cursor(
179 struct xfs_btree_cur *cur)
180{
181 struct xfs_btree_cur *new;
182
183 new = xfs_bmbt_init_cursor(cur->bc_mp, cur->bc_tp,
184 cur->bc_private.b.ip, cur->bc_private.b.whichfork);
185
186 /*
187 * Copy the firstblock, dfops, and flags values,
188 * since init cursor doesn't get them.
189 */
190 new->bc_private.b.firstblock = cur->bc_private.b.firstblock;
191 new->bc_private.b.dfops = cur->bc_private.b.dfops;
192 new->bc_private.b.flags = cur->bc_private.b.flags;
193
194 return new;
195}
196
197STATIC void
198xfs_bmbt_update_cursor(
199 struct xfs_btree_cur *src,
200 struct xfs_btree_cur *dst)
201{
202 ASSERT((dst->bc_private.b.firstblock != NULLFSBLOCK) ||
203 (dst->bc_private.b.ip->i_d.di_flags & XFS_DIFLAG_REALTIME));
204 ASSERT(dst->bc_private.b.dfops == src->bc_private.b.dfops);
205
206 dst->bc_private.b.allocated += src->bc_private.b.allocated;
207 dst->bc_private.b.firstblock = src->bc_private.b.firstblock;
208
209 src->bc_private.b.allocated = 0;
210}
211
212STATIC int
213xfs_bmbt_alloc_block(
214 struct xfs_btree_cur *cur,
215 union xfs_btree_ptr *start,
216 union xfs_btree_ptr *new,
217 int *stat)
218{
219 xfs_alloc_arg_t args; /* block allocation args */
220 int error; /* error return value */
221
222 memset(&args, 0, sizeof(args));
223 args.tp = cur->bc_tp;
224 args.mp = cur->bc_mp;
225 args.fsbno = cur->bc_private.b.firstblock;
226 args.firstblock = args.fsbno;
227 xfs_rmap_ino_bmbt_owner(&args.oinfo, cur->bc_private.b.ip->i_ino,
228 cur->bc_private.b.whichfork);
229
230 if (args.fsbno == NULLFSBLOCK) {
231 args.fsbno = be64_to_cpu(start->l);
232 args.type = XFS_ALLOCTYPE_START_BNO;
233 /*
234 * Make sure there is sufficient room left in the AG to
235 * complete a full tree split for an extent insert. If
236 * we are converting the middle part of an extent then
237 * we may need space for two tree splits.
238 *
239 * We are relying on the caller to make the correct block
240 * reservation for this operation to succeed. If the
241 * reservation amount is insufficient then we may fail a
242 * block allocation here and corrupt the filesystem.
243 */
244 args.minleft = args.tp->t_blk_res;
245 } else if (cur->bc_private.b.dfops->dop_low) {
246 args.type = XFS_ALLOCTYPE_START_BNO;
247 } else {
248 args.type = XFS_ALLOCTYPE_NEAR_BNO;
249 }
250
251 args.minlen = args.maxlen = args.prod = 1;
252 args.wasdel = cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL;
253 if (!args.wasdel && args.tp->t_blk_res == 0) {
254 error = -ENOSPC;
255 goto error0;
256 }
257 error = xfs_alloc_vextent(&args);
258 if (error)
259 goto error0;
260
261 if (args.fsbno == NULLFSBLOCK && args.minleft) {
262 /*
263 * Could not find an AG with enough free space to satisfy
264 * a full btree split. Try again and if
265 * successful activate the lowspace algorithm.
266 */
267 args.fsbno = 0;
268 args.type = XFS_ALLOCTYPE_FIRST_AG;
269 error = xfs_alloc_vextent(&args);
270 if (error)
271 goto error0;
272 cur->bc_private.b.dfops->dop_low = true;
273 }
274 if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
275 *stat = 0;
276 return 0;
277 }
278
279 ASSERT(args.len == 1);
280 cur->bc_private.b.firstblock = args.fsbno;
281 cur->bc_private.b.allocated++;
282 cur->bc_private.b.ip->i_d.di_nblocks++;
283 xfs_trans_log_inode(args.tp, cur->bc_private.b.ip, XFS_ILOG_CORE);
284 xfs_trans_mod_dquot_byino(args.tp, cur->bc_private.b.ip,
285 XFS_TRANS_DQ_BCOUNT, 1L);
286
287 new->l = cpu_to_be64(args.fsbno);
288
289 *stat = 1;
290 return 0;
291
292 error0:
293 return error;
294}
295
296STATIC int
297xfs_bmbt_free_block(
298 struct xfs_btree_cur *cur,
299 struct xfs_buf *bp)
300{
301 struct xfs_mount *mp = cur->bc_mp;
302 struct xfs_inode *ip = cur->bc_private.b.ip;
303 struct xfs_trans *tp = cur->bc_tp;
304 xfs_fsblock_t fsbno = XFS_DADDR_TO_FSB(mp, XFS_BUF_ADDR(bp));
305 struct xfs_owner_info oinfo;
306
307 xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, cur->bc_private.b.whichfork);
308 xfs_bmap_add_free(mp, cur->bc_private.b.dfops, fsbno, 1, &oinfo);
309 ip->i_d.di_nblocks--;
310
311 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
312 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
313 return 0;
314}
315
316STATIC int
317xfs_bmbt_get_minrecs(
318 struct xfs_btree_cur *cur,
319 int level)
320{
321 if (level == cur->bc_nlevels - 1) {
322 struct xfs_ifork *ifp;
323
324 ifp = XFS_IFORK_PTR(cur->bc_private.b.ip,
325 cur->bc_private.b.whichfork);
326
327 return xfs_bmbt_maxrecs(cur->bc_mp,
328 ifp->if_broot_bytes, level == 0) / 2;
329 }
330
331 return cur->bc_mp->m_bmap_dmnr[level != 0];
332}
333
334int
335xfs_bmbt_get_maxrecs(
336 struct xfs_btree_cur *cur,
337 int level)
338{
339 if (level == cur->bc_nlevels - 1) {
340 struct xfs_ifork *ifp;
341
342 ifp = XFS_IFORK_PTR(cur->bc_private.b.ip,
343 cur->bc_private.b.whichfork);
344
345 return xfs_bmbt_maxrecs(cur->bc_mp,
346 ifp->if_broot_bytes, level == 0);
347 }
348
349 return cur->bc_mp->m_bmap_dmxr[level != 0];
350
351}
352
353/*
354 * Get the maximum records we could store in the on-disk format.
355 *
356 * For non-root nodes this is equivalent to xfs_bmbt_get_maxrecs, but
357 * for the root node this checks the available space in the dinode fork
358 * so that we can resize the in-memory buffer to match it. After a
359 * resize to the maximum size this function returns the same value
360 * as xfs_bmbt_get_maxrecs for the root node, too.
361 */
362STATIC int
363xfs_bmbt_get_dmaxrecs(
364 struct xfs_btree_cur *cur,
365 int level)
366{
367 if (level != cur->bc_nlevels - 1)
368 return cur->bc_mp->m_bmap_dmxr[level != 0];
369 return xfs_bmdr_maxrecs(cur->bc_private.b.forksize, level == 0);
370}
371
372STATIC void
373xfs_bmbt_init_key_from_rec(
374 union xfs_btree_key *key,
375 union xfs_btree_rec *rec)
376{
377 key->bmbt.br_startoff =
378 cpu_to_be64(xfs_bmbt_disk_get_startoff(&rec->bmbt));
379}
380
381STATIC void
382xfs_bmbt_init_high_key_from_rec(
383 union xfs_btree_key *key,
384 union xfs_btree_rec *rec)
385{
386 key->bmbt.br_startoff = cpu_to_be64(
387 xfs_bmbt_disk_get_startoff(&rec->bmbt) +
388 xfs_bmbt_disk_get_blockcount(&rec->bmbt) - 1);
389}
390
391STATIC void
392xfs_bmbt_init_rec_from_cur(
393 struct xfs_btree_cur *cur,
394 union xfs_btree_rec *rec)
395{
396 xfs_bmbt_disk_set_all(&rec->bmbt, &cur->bc_rec.b);
397}
398
399STATIC void
400xfs_bmbt_init_ptr_from_cur(
401 struct xfs_btree_cur *cur,
402 union xfs_btree_ptr *ptr)
403{
404 ptr->l = 0;
405}
406
407STATIC int64_t
408xfs_bmbt_key_diff(
409 struct xfs_btree_cur *cur,
410 union xfs_btree_key *key)
411{
412 return (int64_t)be64_to_cpu(key->bmbt.br_startoff) -
413 cur->bc_rec.b.br_startoff;
414}
415
416STATIC int64_t
417xfs_bmbt_diff_two_keys(
418 struct xfs_btree_cur *cur,
419 union xfs_btree_key *k1,
420 union xfs_btree_key *k2)
421{
422 return (int64_t)be64_to_cpu(k1->bmbt.br_startoff) -
423 be64_to_cpu(k2->bmbt.br_startoff);
424}
425
426static xfs_failaddr_t
427xfs_bmbt_verify(
428 struct xfs_buf *bp)
429{
430 struct xfs_mount *mp = bp->b_target->bt_mount;
431 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
432 xfs_failaddr_t fa;
433 unsigned int level;
434
435 switch (block->bb_magic) {
436 case cpu_to_be32(XFS_BMAP_CRC_MAGIC):
437 /*
438 * XXX: need a better way of verifying the owner here. Right now
439 * just make sure there has been one set.
440 */
441 fa = xfs_btree_lblock_v5hdr_verify(bp, XFS_RMAP_OWN_UNKNOWN);
442 if (fa)
443 return fa;
444 /* fall through */
445 case cpu_to_be32(XFS_BMAP_MAGIC):
446 break;
447 default:
448 return __this_address;
449 }
450
451 /*
452 * numrecs and level verification.
453 *
454 * We don't know what fork we belong to, so just verify that the level
455 * is less than the maximum of the two. Later checks will be more
456 * precise.
457 */
458 level = be16_to_cpu(block->bb_level);
459 if (level > max(mp->m_bm_maxlevels[0], mp->m_bm_maxlevels[1]))
460 return __this_address;
461
462 return xfs_btree_lblock_verify(bp, mp->m_bmap_dmxr[level != 0]);
463}
464
465static void
466xfs_bmbt_read_verify(
467 struct xfs_buf *bp)
468{
469 xfs_failaddr_t fa;
470
471 if (!xfs_btree_lblock_verify_crc(bp))
472 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
473 else {
474 fa = xfs_bmbt_verify(bp);
475 if (fa)
476 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
477 }
478
479 if (bp->b_error)
480 trace_xfs_btree_corrupt(bp, _RET_IP_);
481}
482
483static void
484xfs_bmbt_write_verify(
485 struct xfs_buf *bp)
486{
487 xfs_failaddr_t fa;
488
489 fa = xfs_bmbt_verify(bp);
490 if (fa) {
491 trace_xfs_btree_corrupt(bp, _RET_IP_);
492 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
493 return;
494 }
495 xfs_btree_lblock_calc_crc(bp);
496}
497
498const struct xfs_buf_ops xfs_bmbt_buf_ops = {
499 .name = "xfs_bmbt",
500 .verify_read = xfs_bmbt_read_verify,
501 .verify_write = xfs_bmbt_write_verify,
502 .verify_struct = xfs_bmbt_verify,
503};
504
505
506STATIC int
507xfs_bmbt_keys_inorder(
508 struct xfs_btree_cur *cur,
509 union xfs_btree_key *k1,
510 union xfs_btree_key *k2)
511{
512 return be64_to_cpu(k1->bmbt.br_startoff) <
513 be64_to_cpu(k2->bmbt.br_startoff);
514}
515
516STATIC int
517xfs_bmbt_recs_inorder(
518 struct xfs_btree_cur *cur,
519 union xfs_btree_rec *r1,
520 union xfs_btree_rec *r2)
521{
522 return xfs_bmbt_disk_get_startoff(&r1->bmbt) +
523 xfs_bmbt_disk_get_blockcount(&r1->bmbt) <=
524 xfs_bmbt_disk_get_startoff(&r2->bmbt);
525}
526
527static const struct xfs_btree_ops xfs_bmbt_ops = {
528 .rec_len = sizeof(xfs_bmbt_rec_t),
529 .key_len = sizeof(xfs_bmbt_key_t),
530
531 .dup_cursor = xfs_bmbt_dup_cursor,
532 .update_cursor = xfs_bmbt_update_cursor,
533 .alloc_block = xfs_bmbt_alloc_block,
534 .free_block = xfs_bmbt_free_block,
535 .get_maxrecs = xfs_bmbt_get_maxrecs,
536 .get_minrecs = xfs_bmbt_get_minrecs,
537 .get_dmaxrecs = xfs_bmbt_get_dmaxrecs,
538 .init_key_from_rec = xfs_bmbt_init_key_from_rec,
539 .init_high_key_from_rec = xfs_bmbt_init_high_key_from_rec,
540 .init_rec_from_cur = xfs_bmbt_init_rec_from_cur,
541 .init_ptr_from_cur = xfs_bmbt_init_ptr_from_cur,
542 .key_diff = xfs_bmbt_key_diff,
543 .diff_two_keys = xfs_bmbt_diff_two_keys,
544 .buf_ops = &xfs_bmbt_buf_ops,
545 .keys_inorder = xfs_bmbt_keys_inorder,
546 .recs_inorder = xfs_bmbt_recs_inorder,
547};
548
549/*
550 * Allocate a new bmap btree cursor.
551 */
552struct xfs_btree_cur * /* new bmap btree cursor */
553xfs_bmbt_init_cursor(
554 struct xfs_mount *mp, /* file system mount point */
555 struct xfs_trans *tp, /* transaction pointer */
556 struct xfs_inode *ip, /* inode owning the btree */
557 int whichfork) /* data or attr fork */
558{
559 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
560 struct xfs_btree_cur *cur;
561 ASSERT(whichfork != XFS_COW_FORK);
562
563 cur = kmem_zone_zalloc(xfs_btree_cur_zone, KM_NOFS);
564
565 cur->bc_tp = tp;
566 cur->bc_mp = mp;
567 cur->bc_nlevels = be16_to_cpu(ifp->if_broot->bb_level) + 1;
568 cur->bc_btnum = XFS_BTNUM_BMAP;
569 cur->bc_blocklog = mp->m_sb.sb_blocklog;
570 cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_bmbt_2);
571
572 cur->bc_ops = &xfs_bmbt_ops;
573 cur->bc_flags = XFS_BTREE_LONG_PTRS | XFS_BTREE_ROOT_IN_INODE;
574 if (xfs_sb_version_hascrc(&mp->m_sb))
575 cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
576
577 cur->bc_private.b.forksize = XFS_IFORK_SIZE(ip, whichfork);
578 cur->bc_private.b.ip = ip;
579 cur->bc_private.b.firstblock = NULLFSBLOCK;
580 cur->bc_private.b.dfops = NULL;
581 cur->bc_private.b.allocated = 0;
582 cur->bc_private.b.flags = 0;
583 cur->bc_private.b.whichfork = whichfork;
584
585 return cur;
586}
587
588/*
589 * Calculate number of records in a bmap btree block.
590 */
591int
592xfs_bmbt_maxrecs(
593 struct xfs_mount *mp,
594 int blocklen,
595 int leaf)
596{
597 blocklen -= XFS_BMBT_BLOCK_LEN(mp);
598
599 if (leaf)
600 return blocklen / sizeof(xfs_bmbt_rec_t);
601 return blocklen / (sizeof(xfs_bmbt_key_t) + sizeof(xfs_bmbt_ptr_t));
602}
603
604/*
605 * Calculate number of records in a bmap btree inode root.
606 */
607int
608xfs_bmdr_maxrecs(
609 int blocklen,
610 int leaf)
611{
612 blocklen -= sizeof(xfs_bmdr_block_t);
613
614 if (leaf)
615 return blocklen / sizeof(xfs_bmdr_rec_t);
616 return blocklen / (sizeof(xfs_bmdr_key_t) + sizeof(xfs_bmdr_ptr_t));
617}
618
619/*
620 * Change the owner of a btree format fork fo the inode passed in. Change it to
621 * the owner of that is passed in so that we can change owners before or after
622 * we switch forks between inodes. The operation that the caller is doing will
623 * determine whether is needs to change owner before or after the switch.
624 *
625 * For demand paged transactional modification, the fork switch should be done
626 * after reading in all the blocks, modifying them and pinning them in the
627 * transaction. For modification when the buffers are already pinned in memory,
628 * the fork switch can be done before changing the owner as we won't need to
629 * validate the owner until the btree buffers are unpinned and writes can occur
630 * again.
631 *
632 * For recovery based ownership change, there is no transactional context and
633 * so a buffer list must be supplied so that we can record the buffers that we
634 * modified for the caller to issue IO on.
635 */
636int
637xfs_bmbt_change_owner(
638 struct xfs_trans *tp,
639 struct xfs_inode *ip,
640 int whichfork,
641 xfs_ino_t new_owner,
642 struct list_head *buffer_list)
643{
644 struct xfs_btree_cur *cur;
645 int error;
646
647 ASSERT(tp || buffer_list);
648 ASSERT(!(tp && buffer_list));
649 if (whichfork == XFS_DATA_FORK)
650 ASSERT(ip->i_d.di_format == XFS_DINODE_FMT_BTREE);
651 else
652 ASSERT(ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE);
653
654 cur = xfs_bmbt_init_cursor(ip->i_mount, tp, ip, whichfork);
655 if (!cur)
656 return -ENOMEM;
657 cur->bc_private.b.flags |= XFS_BTCUR_BPRV_INVALID_OWNER;
658
659 error = xfs_btree_change_owner(cur, new_owner, buffer_list);
660 xfs_btree_del_cursor(cur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
661 return error;
662}