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_inode.h"
27#include "xfs_trans.h"
28#include "xfs_inode_item.h"
29#include "xfs_alloc.h"
30#include "xfs_btree.h"
31#include "xfs_bmap_btree.h"
32#include "xfs_bmap.h"
33#include "xfs_error.h"
34#include "xfs_quota.h"
35#include "xfs_trace.h"
36#include "xfs_cksum.h"
37
38/*
39 * Determine the extent state.
40 */
41/* ARGSUSED */
42STATIC xfs_exntst_t
43xfs_extent_state(
44 xfs_filblks_t blks,
45 int extent_flag)
46{
47 if (extent_flag) {
48 ASSERT(blks != 0); /* saved for DMIG */
49 return XFS_EXT_UNWRITTEN;
50 }
51 return XFS_EXT_NORM;
52}
53
54/*
55 * Convert on-disk form of btree root to in-memory form.
56 */
57void
58xfs_bmdr_to_bmbt(
59 struct xfs_inode *ip,
60 xfs_bmdr_block_t *dblock,
61 int dblocklen,
62 struct xfs_btree_block *rblock,
63 int rblocklen)
64{
65 struct xfs_mount *mp = ip->i_mount;
66 int dmxr;
67 xfs_bmbt_key_t *fkp;
68 __be64 *fpp;
69 xfs_bmbt_key_t *tkp;
70 __be64 *tpp;
71
72 if (xfs_sb_version_hascrc(&mp->m_sb))
73 xfs_btree_init_block_int(mp, rblock, XFS_BUF_DADDR_NULL,
74 XFS_BMAP_CRC_MAGIC, 0, 0, ip->i_ino,
75 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
76 else
77 xfs_btree_init_block_int(mp, rblock, XFS_BUF_DADDR_NULL,
78 XFS_BMAP_MAGIC, 0, 0, ip->i_ino,
79 XFS_BTREE_LONG_PTRS);
80
81 rblock->bb_level = dblock->bb_level;
82 ASSERT(be16_to_cpu(rblock->bb_level) > 0);
83 rblock->bb_numrecs = dblock->bb_numrecs;
84 dmxr = xfs_bmdr_maxrecs(dblocklen, 0);
85 fkp = XFS_BMDR_KEY_ADDR(dblock, 1);
86 tkp = XFS_BMBT_KEY_ADDR(mp, rblock, 1);
87 fpp = XFS_BMDR_PTR_ADDR(dblock, 1, dmxr);
88 tpp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, rblocklen);
89 dmxr = be16_to_cpu(dblock->bb_numrecs);
90 memcpy(tkp, fkp, sizeof(*fkp) * dmxr);
91 memcpy(tpp, fpp, sizeof(*fpp) * dmxr);
92}
93
94/*
95 * Convert a compressed bmap extent record to an uncompressed form.
96 * This code must be in sync with the routines xfs_bmbt_get_startoff,
97 * xfs_bmbt_get_startblock, xfs_bmbt_get_blockcount and xfs_bmbt_get_state.
98 */
99STATIC void
100__xfs_bmbt_get_all(
101 __uint64_t l0,
102 __uint64_t l1,
103 xfs_bmbt_irec_t *s)
104{
105 int ext_flag;
106 xfs_exntst_t st;
107
108 ext_flag = (int)(l0 >> (64 - BMBT_EXNTFLAG_BITLEN));
109 s->br_startoff = ((xfs_fileoff_t)l0 &
110 xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN)) >> 9;
111 s->br_startblock = (((xfs_fsblock_t)l0 & xfs_mask64lo(9)) << 43) |
112 (((xfs_fsblock_t)l1) >> 21);
113 s->br_blockcount = (xfs_filblks_t)(l1 & xfs_mask64lo(21));
114 /* This is xfs_extent_state() in-line */
115 if (ext_flag) {
116 ASSERT(s->br_blockcount != 0); /* saved for DMIG */
117 st = XFS_EXT_UNWRITTEN;
118 } else
119 st = XFS_EXT_NORM;
120 s->br_state = st;
121}
122
123void
124xfs_bmbt_get_all(
125 xfs_bmbt_rec_host_t *r,
126 xfs_bmbt_irec_t *s)
127{
128 __xfs_bmbt_get_all(r->l0, r->l1, s);
129}
130
131/*
132 * Extract the blockcount field from an in memory bmap extent record.
133 */
134xfs_filblks_t
135xfs_bmbt_get_blockcount(
136 xfs_bmbt_rec_host_t *r)
137{
138 return (xfs_filblks_t)(r->l1 & xfs_mask64lo(21));
139}
140
141/*
142 * Extract the startblock field from an in memory bmap extent record.
143 */
144xfs_fsblock_t
145xfs_bmbt_get_startblock(
146 xfs_bmbt_rec_host_t *r)
147{
148 return (((xfs_fsblock_t)r->l0 & xfs_mask64lo(9)) << 43) |
149 (((xfs_fsblock_t)r->l1) >> 21);
150}
151
152/*
153 * Extract the startoff field from an in memory bmap extent record.
154 */
155xfs_fileoff_t
156xfs_bmbt_get_startoff(
157 xfs_bmbt_rec_host_t *r)
158{
159 return ((xfs_fileoff_t)r->l0 &
160 xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN)) >> 9;
161}
162
163xfs_exntst_t
164xfs_bmbt_get_state(
165 xfs_bmbt_rec_host_t *r)
166{
167 int ext_flag;
168
169 ext_flag = (int)((r->l0) >> (64 - BMBT_EXNTFLAG_BITLEN));
170 return xfs_extent_state(xfs_bmbt_get_blockcount(r),
171 ext_flag);
172}
173
174/*
175 * Extract the blockcount field from an on disk bmap extent record.
176 */
177xfs_filblks_t
178xfs_bmbt_disk_get_blockcount(
179 xfs_bmbt_rec_t *r)
180{
181 return (xfs_filblks_t)(be64_to_cpu(r->l1) & xfs_mask64lo(21));
182}
183
184/*
185 * Extract the startoff field from a disk format bmap extent record.
186 */
187xfs_fileoff_t
188xfs_bmbt_disk_get_startoff(
189 xfs_bmbt_rec_t *r)
190{
191 return ((xfs_fileoff_t)be64_to_cpu(r->l0) &
192 xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN)) >> 9;
193}
194
195
196/*
197 * Set all the fields in a bmap extent record from the arguments.
198 */
199void
200xfs_bmbt_set_allf(
201 xfs_bmbt_rec_host_t *r,
202 xfs_fileoff_t startoff,
203 xfs_fsblock_t startblock,
204 xfs_filblks_t blockcount,
205 xfs_exntst_t state)
206{
207 int extent_flag = (state == XFS_EXT_NORM) ? 0 : 1;
208
209 ASSERT(state == XFS_EXT_NORM || state == XFS_EXT_UNWRITTEN);
210 ASSERT((startoff & xfs_mask64hi(64-BMBT_STARTOFF_BITLEN)) == 0);
211 ASSERT((blockcount & xfs_mask64hi(64-BMBT_BLOCKCOUNT_BITLEN)) == 0);
212
213 ASSERT((startblock & xfs_mask64hi(64-BMBT_STARTBLOCK_BITLEN)) == 0);
214
215 r->l0 = ((xfs_bmbt_rec_base_t)extent_flag << 63) |
216 ((xfs_bmbt_rec_base_t)startoff << 9) |
217 ((xfs_bmbt_rec_base_t)startblock >> 43);
218 r->l1 = ((xfs_bmbt_rec_base_t)startblock << 21) |
219 ((xfs_bmbt_rec_base_t)blockcount &
220 (xfs_bmbt_rec_base_t)xfs_mask64lo(21));
221}
222
223/*
224 * Set all the fields in a bmap extent record from the uncompressed form.
225 */
226void
227xfs_bmbt_set_all(
228 xfs_bmbt_rec_host_t *r,
229 xfs_bmbt_irec_t *s)
230{
231 xfs_bmbt_set_allf(r, s->br_startoff, s->br_startblock,
232 s->br_blockcount, s->br_state);
233}
234
235
236/*
237 * Set all the fields in a disk format bmap extent record from the arguments.
238 */
239void
240xfs_bmbt_disk_set_allf(
241 xfs_bmbt_rec_t *r,
242 xfs_fileoff_t startoff,
243 xfs_fsblock_t startblock,
244 xfs_filblks_t blockcount,
245 xfs_exntst_t state)
246{
247 int extent_flag = (state == XFS_EXT_NORM) ? 0 : 1;
248
249 ASSERT(state == XFS_EXT_NORM || state == XFS_EXT_UNWRITTEN);
250 ASSERT((startoff & xfs_mask64hi(64-BMBT_STARTOFF_BITLEN)) == 0);
251 ASSERT((blockcount & xfs_mask64hi(64-BMBT_BLOCKCOUNT_BITLEN)) == 0);
252 ASSERT((startblock & xfs_mask64hi(64-BMBT_STARTBLOCK_BITLEN)) == 0);
253
254 r->l0 = cpu_to_be64(
255 ((xfs_bmbt_rec_base_t)extent_flag << 63) |
256 ((xfs_bmbt_rec_base_t)startoff << 9) |
257 ((xfs_bmbt_rec_base_t)startblock >> 43));
258 r->l1 = cpu_to_be64(
259 ((xfs_bmbt_rec_base_t)startblock << 21) |
260 ((xfs_bmbt_rec_base_t)blockcount &
261 (xfs_bmbt_rec_base_t)xfs_mask64lo(21)));
262}
263
264/*
265 * Set all the fields in a bmap extent record from the uncompressed form.
266 */
267STATIC void
268xfs_bmbt_disk_set_all(
269 xfs_bmbt_rec_t *r,
270 xfs_bmbt_irec_t *s)
271{
272 xfs_bmbt_disk_set_allf(r, s->br_startoff, s->br_startblock,
273 s->br_blockcount, s->br_state);
274}
275
276/*
277 * Set the blockcount field in a bmap extent record.
278 */
279void
280xfs_bmbt_set_blockcount(
281 xfs_bmbt_rec_host_t *r,
282 xfs_filblks_t v)
283{
284 ASSERT((v & xfs_mask64hi(43)) == 0);
285 r->l1 = (r->l1 & (xfs_bmbt_rec_base_t)xfs_mask64hi(43)) |
286 (xfs_bmbt_rec_base_t)(v & xfs_mask64lo(21));
287}
288
289/*
290 * Set the startblock field in a bmap extent record.
291 */
292void
293xfs_bmbt_set_startblock(
294 xfs_bmbt_rec_host_t *r,
295 xfs_fsblock_t v)
296{
297 ASSERT((v & xfs_mask64hi(12)) == 0);
298 r->l0 = (r->l0 & (xfs_bmbt_rec_base_t)xfs_mask64hi(55)) |
299 (xfs_bmbt_rec_base_t)(v >> 43);
300 r->l1 = (r->l1 & (xfs_bmbt_rec_base_t)xfs_mask64lo(21)) |
301 (xfs_bmbt_rec_base_t)(v << 21);
302}
303
304/*
305 * Set the startoff field in a bmap extent record.
306 */
307void
308xfs_bmbt_set_startoff(
309 xfs_bmbt_rec_host_t *r,
310 xfs_fileoff_t v)
311{
312 ASSERT((v & xfs_mask64hi(9)) == 0);
313 r->l0 = (r->l0 & (xfs_bmbt_rec_base_t) xfs_mask64hi(1)) |
314 ((xfs_bmbt_rec_base_t)v << 9) |
315 (r->l0 & (xfs_bmbt_rec_base_t)xfs_mask64lo(9));
316}
317
318/*
319 * Set the extent state field in a bmap extent record.
320 */
321void
322xfs_bmbt_set_state(
323 xfs_bmbt_rec_host_t *r,
324 xfs_exntst_t v)
325{
326 ASSERT(v == XFS_EXT_NORM || v == XFS_EXT_UNWRITTEN);
327 if (v == XFS_EXT_NORM)
328 r->l0 &= xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN);
329 else
330 r->l0 |= xfs_mask64hi(BMBT_EXNTFLAG_BITLEN);
331}
332
333/*
334 * Convert in-memory form of btree root to on-disk form.
335 */
336void
337xfs_bmbt_to_bmdr(
338 struct xfs_mount *mp,
339 struct xfs_btree_block *rblock,
340 int rblocklen,
341 xfs_bmdr_block_t *dblock,
342 int dblocklen)
343{
344 int dmxr;
345 xfs_bmbt_key_t *fkp;
346 __be64 *fpp;
347 xfs_bmbt_key_t *tkp;
348 __be64 *tpp;
349
350 if (xfs_sb_version_hascrc(&mp->m_sb)) {
351 ASSERT(rblock->bb_magic == cpu_to_be32(XFS_BMAP_CRC_MAGIC));
352 ASSERT(uuid_equal(&rblock->bb_u.l.bb_uuid,
353 &mp->m_sb.sb_meta_uuid));
354 ASSERT(rblock->bb_u.l.bb_blkno ==
355 cpu_to_be64(XFS_BUF_DADDR_NULL));
356 } else
357 ASSERT(rblock->bb_magic == cpu_to_be32(XFS_BMAP_MAGIC));
358 ASSERT(rblock->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK));
359 ASSERT(rblock->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK));
360 ASSERT(rblock->bb_level != 0);
361 dblock->bb_level = rblock->bb_level;
362 dblock->bb_numrecs = rblock->bb_numrecs;
363 dmxr = xfs_bmdr_maxrecs(dblocklen, 0);
364 fkp = XFS_BMBT_KEY_ADDR(mp, rblock, 1);
365 tkp = XFS_BMDR_KEY_ADDR(dblock, 1);
366 fpp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, rblocklen);
367 tpp = XFS_BMDR_PTR_ADDR(dblock, 1, dmxr);
368 dmxr = be16_to_cpu(dblock->bb_numrecs);
369 memcpy(tkp, fkp, sizeof(*fkp) * dmxr);
370 memcpy(tpp, fpp, sizeof(*fpp) * dmxr);
371}
372
373/*
374 * Check extent records, which have just been read, for
375 * any bit in the extent flag field. ASSERT on debug
376 * kernels, as this condition should not occur.
377 * Return an error condition (1) if any flags found,
378 * otherwise return 0.
379 */
380
381int
382xfs_check_nostate_extents(
383 xfs_ifork_t *ifp,
384 xfs_extnum_t idx,
385 xfs_extnum_t num)
386{
387 for (; num > 0; num--, idx++) {
388 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
389 if ((ep->l0 >>
390 (64 - BMBT_EXNTFLAG_BITLEN)) != 0) {
391 ASSERT(0);
392 return 1;
393 }
394 }
395 return 0;
396}
397
398
399STATIC struct xfs_btree_cur *
400xfs_bmbt_dup_cursor(
401 struct xfs_btree_cur *cur)
402{
403 struct xfs_btree_cur *new;
404
405 new = xfs_bmbt_init_cursor(cur->bc_mp, cur->bc_tp,
406 cur->bc_private.b.ip, cur->bc_private.b.whichfork);
407
408 /*
409 * Copy the firstblock, flist, and flags values,
410 * since init cursor doesn't get them.
411 */
412 new->bc_private.b.firstblock = cur->bc_private.b.firstblock;
413 new->bc_private.b.flist = cur->bc_private.b.flist;
414 new->bc_private.b.flags = cur->bc_private.b.flags;
415
416 return new;
417}
418
419STATIC void
420xfs_bmbt_update_cursor(
421 struct xfs_btree_cur *src,
422 struct xfs_btree_cur *dst)
423{
424 ASSERT((dst->bc_private.b.firstblock != NULLFSBLOCK) ||
425 (dst->bc_private.b.ip->i_d.di_flags & XFS_DIFLAG_REALTIME));
426 ASSERT(dst->bc_private.b.flist == src->bc_private.b.flist);
427
428 dst->bc_private.b.allocated += src->bc_private.b.allocated;
429 dst->bc_private.b.firstblock = src->bc_private.b.firstblock;
430
431 src->bc_private.b.allocated = 0;
432}
433
434STATIC int
435xfs_bmbt_alloc_block(
436 struct xfs_btree_cur *cur,
437 union xfs_btree_ptr *start,
438 union xfs_btree_ptr *new,
439 int *stat)
440{
441 xfs_alloc_arg_t args; /* block allocation args */
442 int error; /* error return value */
443
444 memset(&args, 0, sizeof(args));
445 args.tp = cur->bc_tp;
446 args.mp = cur->bc_mp;
447 args.fsbno = cur->bc_private.b.firstblock;
448 args.firstblock = args.fsbno;
449
450 if (args.fsbno == NULLFSBLOCK) {
451 args.fsbno = be64_to_cpu(start->l);
452 args.type = XFS_ALLOCTYPE_START_BNO;
453 /*
454 * Make sure there is sufficient room left in the AG to
455 * complete a full tree split for an extent insert. If
456 * we are converting the middle part of an extent then
457 * we may need space for two tree splits.
458 *
459 * We are relying on the caller to make the correct block
460 * reservation for this operation to succeed. If the
461 * reservation amount is insufficient then we may fail a
462 * block allocation here and corrupt the filesystem.
463 */
464 args.minleft = args.tp->t_blk_res;
465 } else if (cur->bc_private.b.flist->xbf_low) {
466 args.type = XFS_ALLOCTYPE_START_BNO;
467 } else {
468 args.type = XFS_ALLOCTYPE_NEAR_BNO;
469 }
470
471 args.minlen = args.maxlen = args.prod = 1;
472 args.wasdel = cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL;
473 if (!args.wasdel && args.tp->t_blk_res == 0) {
474 error = -ENOSPC;
475 goto error0;
476 }
477 error = xfs_alloc_vextent(&args);
478 if (error)
479 goto error0;
480
481 if (args.fsbno == NULLFSBLOCK && args.minleft) {
482 /*
483 * Could not find an AG with enough free space to satisfy
484 * a full btree split. Try again without minleft and if
485 * successful activate the lowspace algorithm.
486 */
487 args.fsbno = 0;
488 args.type = XFS_ALLOCTYPE_FIRST_AG;
489 args.minleft = 0;
490 error = xfs_alloc_vextent(&args);
491 if (error)
492 goto error0;
493 cur->bc_private.b.flist->xbf_low = 1;
494 }
495 if (args.fsbno == NULLFSBLOCK) {
496 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
497 *stat = 0;
498 return 0;
499 }
500 ASSERT(args.len == 1);
501 cur->bc_private.b.firstblock = args.fsbno;
502 cur->bc_private.b.allocated++;
503 cur->bc_private.b.ip->i_d.di_nblocks++;
504 xfs_trans_log_inode(args.tp, cur->bc_private.b.ip, XFS_ILOG_CORE);
505 xfs_trans_mod_dquot_byino(args.tp, cur->bc_private.b.ip,
506 XFS_TRANS_DQ_BCOUNT, 1L);
507
508 new->l = cpu_to_be64(args.fsbno);
509
510 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
511 *stat = 1;
512 return 0;
513
514 error0:
515 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
516 return error;
517}
518
519STATIC int
520xfs_bmbt_free_block(
521 struct xfs_btree_cur *cur,
522 struct xfs_buf *bp)
523{
524 struct xfs_mount *mp = cur->bc_mp;
525 struct xfs_inode *ip = cur->bc_private.b.ip;
526 struct xfs_trans *tp = cur->bc_tp;
527 xfs_fsblock_t fsbno = XFS_DADDR_TO_FSB(mp, XFS_BUF_ADDR(bp));
528
529 xfs_bmap_add_free(fsbno, 1, cur->bc_private.b.flist, mp);
530 ip->i_d.di_nblocks--;
531
532 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
533 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
534 return 0;
535}
536
537STATIC int
538xfs_bmbt_get_minrecs(
539 struct xfs_btree_cur *cur,
540 int level)
541{
542 if (level == cur->bc_nlevels - 1) {
543 struct xfs_ifork *ifp;
544
545 ifp = XFS_IFORK_PTR(cur->bc_private.b.ip,
546 cur->bc_private.b.whichfork);
547
548 return xfs_bmbt_maxrecs(cur->bc_mp,
549 ifp->if_broot_bytes, level == 0) / 2;
550 }
551
552 return cur->bc_mp->m_bmap_dmnr[level != 0];
553}
554
555int
556xfs_bmbt_get_maxrecs(
557 struct xfs_btree_cur *cur,
558 int level)
559{
560 if (level == cur->bc_nlevels - 1) {
561 struct xfs_ifork *ifp;
562
563 ifp = XFS_IFORK_PTR(cur->bc_private.b.ip,
564 cur->bc_private.b.whichfork);
565
566 return xfs_bmbt_maxrecs(cur->bc_mp,
567 ifp->if_broot_bytes, level == 0);
568 }
569
570 return cur->bc_mp->m_bmap_dmxr[level != 0];
571
572}
573
574/*
575 * Get the maximum records we could store in the on-disk format.
576 *
577 * For non-root nodes this is equivalent to xfs_bmbt_get_maxrecs, but
578 * for the root node this checks the available space in the dinode fork
579 * so that we can resize the in-memory buffer to match it. After a
580 * resize to the maximum size this function returns the same value
581 * as xfs_bmbt_get_maxrecs for the root node, too.
582 */
583STATIC int
584xfs_bmbt_get_dmaxrecs(
585 struct xfs_btree_cur *cur,
586 int level)
587{
588 if (level != cur->bc_nlevels - 1)
589 return cur->bc_mp->m_bmap_dmxr[level != 0];
590 return xfs_bmdr_maxrecs(cur->bc_private.b.forksize, level == 0);
591}
592
593STATIC void
594xfs_bmbt_init_key_from_rec(
595 union xfs_btree_key *key,
596 union xfs_btree_rec *rec)
597{
598 key->bmbt.br_startoff =
599 cpu_to_be64(xfs_bmbt_disk_get_startoff(&rec->bmbt));
600}
601
602STATIC void
603xfs_bmbt_init_rec_from_key(
604 union xfs_btree_key *key,
605 union xfs_btree_rec *rec)
606{
607 ASSERT(key->bmbt.br_startoff != 0);
608
609 xfs_bmbt_disk_set_allf(&rec->bmbt, be64_to_cpu(key->bmbt.br_startoff),
610 0, 0, XFS_EXT_NORM);
611}
612
613STATIC void
614xfs_bmbt_init_rec_from_cur(
615 struct xfs_btree_cur *cur,
616 union xfs_btree_rec *rec)
617{
618 xfs_bmbt_disk_set_all(&rec->bmbt, &cur->bc_rec.b);
619}
620
621STATIC void
622xfs_bmbt_init_ptr_from_cur(
623 struct xfs_btree_cur *cur,
624 union xfs_btree_ptr *ptr)
625{
626 ptr->l = 0;
627}
628
629STATIC __int64_t
630xfs_bmbt_key_diff(
631 struct xfs_btree_cur *cur,
632 union xfs_btree_key *key)
633{
634 return (__int64_t)be64_to_cpu(key->bmbt.br_startoff) -
635 cur->bc_rec.b.br_startoff;
636}
637
638static bool
639xfs_bmbt_verify(
640 struct xfs_buf *bp)
641{
642 struct xfs_mount *mp = bp->b_target->bt_mount;
643 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
644 unsigned int level;
645
646 switch (block->bb_magic) {
647 case cpu_to_be32(XFS_BMAP_CRC_MAGIC):
648 if (!xfs_sb_version_hascrc(&mp->m_sb))
649 return false;
650 if (!uuid_equal(&block->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid))
651 return false;
652 if (be64_to_cpu(block->bb_u.l.bb_blkno) != bp->b_bn)
653 return false;
654 /*
655 * XXX: need a better way of verifying the owner here. Right now
656 * just make sure there has been one set.
657 */
658 if (be64_to_cpu(block->bb_u.l.bb_owner) == 0)
659 return false;
660 /* fall through */
661 case cpu_to_be32(XFS_BMAP_MAGIC):
662 break;
663 default:
664 return false;
665 }
666
667 /*
668 * numrecs and level verification.
669 *
670 * We don't know what fork we belong to, so just verify that the level
671 * is less than the maximum of the two. Later checks will be more
672 * precise.
673 */
674 level = be16_to_cpu(block->bb_level);
675 if (level > max(mp->m_bm_maxlevels[0], mp->m_bm_maxlevels[1]))
676 return false;
677 if (be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
678 return false;
679
680 /* sibling pointer verification */
681 if (!block->bb_u.l.bb_leftsib ||
682 (block->bb_u.l.bb_leftsib != cpu_to_be64(NULLFSBLOCK) &&
683 !XFS_FSB_SANITY_CHECK(mp, be64_to_cpu(block->bb_u.l.bb_leftsib))))
684 return false;
685 if (!block->bb_u.l.bb_rightsib ||
686 (block->bb_u.l.bb_rightsib != cpu_to_be64(NULLFSBLOCK) &&
687 !XFS_FSB_SANITY_CHECK(mp, be64_to_cpu(block->bb_u.l.bb_rightsib))))
688 return false;
689
690 return true;
691}
692
693static void
694xfs_bmbt_read_verify(
695 struct xfs_buf *bp)
696{
697 if (!xfs_btree_lblock_verify_crc(bp))
698 xfs_buf_ioerror(bp, -EFSBADCRC);
699 else if (!xfs_bmbt_verify(bp))
700 xfs_buf_ioerror(bp, -EFSCORRUPTED);
701
702 if (bp->b_error) {
703 trace_xfs_btree_corrupt(bp, _RET_IP_);
704 xfs_verifier_error(bp);
705 }
706}
707
708static void
709xfs_bmbt_write_verify(
710 struct xfs_buf *bp)
711{
712 if (!xfs_bmbt_verify(bp)) {
713 trace_xfs_btree_corrupt(bp, _RET_IP_);
714 xfs_buf_ioerror(bp, -EFSCORRUPTED);
715 xfs_verifier_error(bp);
716 return;
717 }
718 xfs_btree_lblock_calc_crc(bp);
719}
720
721const struct xfs_buf_ops xfs_bmbt_buf_ops = {
722 .name = "xfs_bmbt",
723 .verify_read = xfs_bmbt_read_verify,
724 .verify_write = xfs_bmbt_write_verify,
725};
726
727
728#if defined(DEBUG) || defined(XFS_WARN)
729STATIC int
730xfs_bmbt_keys_inorder(
731 struct xfs_btree_cur *cur,
732 union xfs_btree_key *k1,
733 union xfs_btree_key *k2)
734{
735 return be64_to_cpu(k1->bmbt.br_startoff) <
736 be64_to_cpu(k2->bmbt.br_startoff);
737}
738
739STATIC int
740xfs_bmbt_recs_inorder(
741 struct xfs_btree_cur *cur,
742 union xfs_btree_rec *r1,
743 union xfs_btree_rec *r2)
744{
745 return xfs_bmbt_disk_get_startoff(&r1->bmbt) +
746 xfs_bmbt_disk_get_blockcount(&r1->bmbt) <=
747 xfs_bmbt_disk_get_startoff(&r2->bmbt);
748}
749#endif /* DEBUG */
750
751static const struct xfs_btree_ops xfs_bmbt_ops = {
752 .rec_len = sizeof(xfs_bmbt_rec_t),
753 .key_len = sizeof(xfs_bmbt_key_t),
754
755 .dup_cursor = xfs_bmbt_dup_cursor,
756 .update_cursor = xfs_bmbt_update_cursor,
757 .alloc_block = xfs_bmbt_alloc_block,
758 .free_block = xfs_bmbt_free_block,
759 .get_maxrecs = xfs_bmbt_get_maxrecs,
760 .get_minrecs = xfs_bmbt_get_minrecs,
761 .get_dmaxrecs = xfs_bmbt_get_dmaxrecs,
762 .init_key_from_rec = xfs_bmbt_init_key_from_rec,
763 .init_rec_from_key = xfs_bmbt_init_rec_from_key,
764 .init_rec_from_cur = xfs_bmbt_init_rec_from_cur,
765 .init_ptr_from_cur = xfs_bmbt_init_ptr_from_cur,
766 .key_diff = xfs_bmbt_key_diff,
767 .buf_ops = &xfs_bmbt_buf_ops,
768#if defined(DEBUG) || defined(XFS_WARN)
769 .keys_inorder = xfs_bmbt_keys_inorder,
770 .recs_inorder = xfs_bmbt_recs_inorder,
771#endif
772};
773
774/*
775 * Allocate a new bmap btree cursor.
776 */
777struct xfs_btree_cur * /* new bmap btree cursor */
778xfs_bmbt_init_cursor(
779 struct xfs_mount *mp, /* file system mount point */
780 struct xfs_trans *tp, /* transaction pointer */
781 struct xfs_inode *ip, /* inode owning the btree */
782 int whichfork) /* data or attr fork */
783{
784 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
785 struct xfs_btree_cur *cur;
786
787 cur = kmem_zone_zalloc(xfs_btree_cur_zone, KM_SLEEP);
788
789 cur->bc_tp = tp;
790 cur->bc_mp = mp;
791 cur->bc_nlevels = be16_to_cpu(ifp->if_broot->bb_level) + 1;
792 cur->bc_btnum = XFS_BTNUM_BMAP;
793 cur->bc_blocklog = mp->m_sb.sb_blocklog;
794
795 cur->bc_ops = &xfs_bmbt_ops;
796 cur->bc_flags = XFS_BTREE_LONG_PTRS | XFS_BTREE_ROOT_IN_INODE;
797 if (xfs_sb_version_hascrc(&mp->m_sb))
798 cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
799
800 cur->bc_private.b.forksize = XFS_IFORK_SIZE(ip, whichfork);
801 cur->bc_private.b.ip = ip;
802 cur->bc_private.b.firstblock = NULLFSBLOCK;
803 cur->bc_private.b.flist = NULL;
804 cur->bc_private.b.allocated = 0;
805 cur->bc_private.b.flags = 0;
806 cur->bc_private.b.whichfork = whichfork;
807
808 return cur;
809}
810
811/*
812 * Calculate number of records in a bmap btree block.
813 */
814int
815xfs_bmbt_maxrecs(
816 struct xfs_mount *mp,
817 int blocklen,
818 int leaf)
819{
820 blocklen -= XFS_BMBT_BLOCK_LEN(mp);
821
822 if (leaf)
823 return blocklen / sizeof(xfs_bmbt_rec_t);
824 return blocklen / (sizeof(xfs_bmbt_key_t) + sizeof(xfs_bmbt_ptr_t));
825}
826
827/*
828 * Calculate number of records in a bmap btree inode root.
829 */
830int
831xfs_bmdr_maxrecs(
832 int blocklen,
833 int leaf)
834{
835 blocklen -= sizeof(xfs_bmdr_block_t);
836
837 if (leaf)
838 return blocklen / sizeof(xfs_bmdr_rec_t);
839 return blocklen / (sizeof(xfs_bmdr_key_t) + sizeof(xfs_bmdr_ptr_t));
840}
841
842/*
843 * Change the owner of a btree format fork fo the inode passed in. Change it to
844 * the owner of that is passed in so that we can change owners before or after
845 * we switch forks between inodes. The operation that the caller is doing will
846 * determine whether is needs to change owner before or after the switch.
847 *
848 * For demand paged transactional modification, the fork switch should be done
849 * after reading in all the blocks, modifying them and pinning them in the
850 * transaction. For modification when the buffers are already pinned in memory,
851 * the fork switch can be done before changing the owner as we won't need to
852 * validate the owner until the btree buffers are unpinned and writes can occur
853 * again.
854 *
855 * For recovery based ownership change, there is no transactional context and
856 * so a buffer list must be supplied so that we can record the buffers that we
857 * modified for the caller to issue IO on.
858 */
859int
860xfs_bmbt_change_owner(
861 struct xfs_trans *tp,
862 struct xfs_inode *ip,
863 int whichfork,
864 xfs_ino_t new_owner,
865 struct list_head *buffer_list)
866{
867 struct xfs_btree_cur *cur;
868 int error;
869
870 ASSERT(tp || buffer_list);
871 ASSERT(!(tp && buffer_list));
872 if (whichfork == XFS_DATA_FORK)
873 ASSERT(ip->i_d.di_format == XFS_DINODE_FMT_BTREE);
874 else
875 ASSERT(ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE);
876
877 cur = xfs_bmbt_init_cursor(ip->i_mount, tp, ip, whichfork);
878 if (!cur)
879 return -ENOMEM;
880
881 error = xfs_btree_change_owner(cur, new_owner, buffer_list);
882 xfs_btree_del_cursor(cur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
883 return error;
884}