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