Loading...
Note: File does not exist in v3.1.
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * Copyright (c) 2012 Red Hat, Inc.
5 * All Rights Reserved.
6 */
7#include "xfs.h"
8#include "xfs_fs.h"
9#include "xfs_shared.h"
10#include "xfs_format.h"
11#include "xfs_log_format.h"
12#include "xfs_trans_resv.h"
13#include "xfs_bit.h"
14#include "xfs_mount.h"
15#include "xfs_defer.h"
16#include "xfs_inode.h"
17#include "xfs_btree.h"
18#include "xfs_trans.h"
19#include "xfs_alloc.h"
20#include "xfs_bmap.h"
21#include "xfs_bmap_util.h"
22#include "xfs_bmap_btree.h"
23#include "xfs_rtalloc.h"
24#include "xfs_error.h"
25#include "xfs_quota.h"
26#include "xfs_trans_space.h"
27#include "xfs_trace.h"
28#include "xfs_icache.h"
29#include "xfs_iomap.h"
30#include "xfs_reflink.h"
31
32/* Kernel only BMAP related definitions and functions */
33
34/*
35 * Convert the given file system block to a disk block. We have to treat it
36 * differently based on whether the file is a real time file or not, because the
37 * bmap code does.
38 */
39xfs_daddr_t
40xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb)
41{
42 if (XFS_IS_REALTIME_INODE(ip))
43 return XFS_FSB_TO_BB(ip->i_mount, fsb);
44 return XFS_FSB_TO_DADDR(ip->i_mount, fsb);
45}
46
47/*
48 * Routine to zero an extent on disk allocated to the specific inode.
49 *
50 * The VFS functions take a linearised filesystem block offset, so we have to
51 * convert the sparse xfs fsb to the right format first.
52 * VFS types are real funky, too.
53 */
54int
55xfs_zero_extent(
56 struct xfs_inode *ip,
57 xfs_fsblock_t start_fsb,
58 xfs_off_t count_fsb)
59{
60 struct xfs_mount *mp = ip->i_mount;
61 xfs_daddr_t sector = xfs_fsb_to_db(ip, start_fsb);
62 sector_t block = XFS_BB_TO_FSBT(mp, sector);
63
64 return blkdev_issue_zeroout(xfs_find_bdev_for_inode(VFS_I(ip)),
65 block << (mp->m_super->s_blocksize_bits - 9),
66 count_fsb << (mp->m_super->s_blocksize_bits - 9),
67 GFP_NOFS, 0);
68}
69
70#ifdef CONFIG_XFS_RT
71int
72xfs_bmap_rtalloc(
73 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
74{
75 int error; /* error return value */
76 xfs_mount_t *mp; /* mount point structure */
77 xfs_extlen_t prod = 0; /* product factor for allocators */
78 xfs_extlen_t mod = 0; /* product factor for allocators */
79 xfs_extlen_t ralen = 0; /* realtime allocation length */
80 xfs_extlen_t align; /* minimum allocation alignment */
81 xfs_rtblock_t rtb;
82
83 mp = ap->ip->i_mount;
84 align = xfs_get_extsz_hint(ap->ip);
85 prod = align / mp->m_sb.sb_rextsize;
86 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
87 align, 1, ap->eof, 0,
88 ap->conv, &ap->offset, &ap->length);
89 if (error)
90 return error;
91 ASSERT(ap->length);
92 ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
93
94 /*
95 * If the offset & length are not perfectly aligned
96 * then kill prod, it will just get us in trouble.
97 */
98 div_u64_rem(ap->offset, align, &mod);
99 if (mod || ap->length % align)
100 prod = 1;
101 /*
102 * Set ralen to be the actual requested length in rtextents.
103 */
104 ralen = ap->length / mp->m_sb.sb_rextsize;
105 /*
106 * If the old value was close enough to MAXEXTLEN that
107 * we rounded up to it, cut it back so it's valid again.
108 * Note that if it's a really large request (bigger than
109 * MAXEXTLEN), we don't hear about that number, and can't
110 * adjust the starting point to match it.
111 */
112 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
113 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
114
115 /*
116 * Lock out modifications to both the RT bitmap and summary inodes
117 */
118 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
119 xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
120 xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
121 xfs_trans_ijoin(ap->tp, mp->m_rsumip, XFS_ILOCK_EXCL);
122
123 /*
124 * If it's an allocation to an empty file at offset 0,
125 * pick an extent that will space things out in the rt area.
126 */
127 if (ap->eof && ap->offset == 0) {
128 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
129
130 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
131 if (error)
132 return error;
133 ap->blkno = rtx * mp->m_sb.sb_rextsize;
134 } else {
135 ap->blkno = 0;
136 }
137
138 xfs_bmap_adjacent(ap);
139
140 /*
141 * Realtime allocation, done through xfs_rtallocate_extent.
142 */
143 do_div(ap->blkno, mp->m_sb.sb_rextsize);
144 rtb = ap->blkno;
145 ap->length = ralen;
146 error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
147 &ralen, ap->wasdel, prod, &rtb);
148 if (error)
149 return error;
150
151 ap->blkno = rtb;
152 if (ap->blkno != NULLFSBLOCK) {
153 ap->blkno *= mp->m_sb.sb_rextsize;
154 ralen *= mp->m_sb.sb_rextsize;
155 ap->length = ralen;
156 ap->ip->i_d.di_nblocks += ralen;
157 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
158 if (ap->wasdel)
159 ap->ip->i_delayed_blks -= ralen;
160 /*
161 * Adjust the disk quota also. This was reserved
162 * earlier.
163 */
164 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
165 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
166 XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
167
168 /* Zero the extent if we were asked to do so */
169 if (ap->datatype & XFS_ALLOC_USERDATA_ZERO) {
170 error = xfs_zero_extent(ap->ip, ap->blkno, ap->length);
171 if (error)
172 return error;
173 }
174 } else {
175 ap->length = 0;
176 }
177 return 0;
178}
179#endif /* CONFIG_XFS_RT */
180
181/*
182 * Check if the endoff is outside the last extent. If so the caller will grow
183 * the allocation to a stripe unit boundary. All offsets are considered outside
184 * the end of file for an empty fork, so 1 is returned in *eof in that case.
185 */
186int
187xfs_bmap_eof(
188 struct xfs_inode *ip,
189 xfs_fileoff_t endoff,
190 int whichfork,
191 int *eof)
192{
193 struct xfs_bmbt_irec rec;
194 int error;
195
196 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
197 if (error || *eof)
198 return error;
199
200 *eof = endoff >= rec.br_startoff + rec.br_blockcount;
201 return 0;
202}
203
204/*
205 * Extent tree block counting routines.
206 */
207
208/*
209 * Count leaf blocks given a range of extent records. Delayed allocation
210 * extents are not counted towards the totals.
211 */
212xfs_extnum_t
213xfs_bmap_count_leaves(
214 struct xfs_ifork *ifp,
215 xfs_filblks_t *count)
216{
217 struct xfs_iext_cursor icur;
218 struct xfs_bmbt_irec got;
219 xfs_extnum_t numrecs = 0;
220
221 for_each_xfs_iext(ifp, &icur, &got) {
222 if (!isnullstartblock(got.br_startblock)) {
223 *count += got.br_blockcount;
224 numrecs++;
225 }
226 }
227
228 return numrecs;
229}
230
231/*
232 * Count leaf blocks given a range of extent records originally
233 * in btree format.
234 */
235STATIC void
236xfs_bmap_disk_count_leaves(
237 struct xfs_mount *mp,
238 struct xfs_btree_block *block,
239 int numrecs,
240 xfs_filblks_t *count)
241{
242 int b;
243 xfs_bmbt_rec_t *frp;
244
245 for (b = 1; b <= numrecs; b++) {
246 frp = XFS_BMBT_REC_ADDR(mp, block, b);
247 *count += xfs_bmbt_disk_get_blockcount(frp);
248 }
249}
250
251/*
252 * Recursively walks each level of a btree
253 * to count total fsblocks in use.
254 */
255STATIC int
256xfs_bmap_count_tree(
257 struct xfs_mount *mp,
258 struct xfs_trans *tp,
259 struct xfs_ifork *ifp,
260 xfs_fsblock_t blockno,
261 int levelin,
262 xfs_extnum_t *nextents,
263 xfs_filblks_t *count)
264{
265 int error;
266 struct xfs_buf *bp, *nbp;
267 int level = levelin;
268 __be64 *pp;
269 xfs_fsblock_t bno = blockno;
270 xfs_fsblock_t nextbno;
271 struct xfs_btree_block *block, *nextblock;
272 int numrecs;
273
274 error = xfs_btree_read_bufl(mp, tp, bno, &bp, XFS_BMAP_BTREE_REF,
275 &xfs_bmbt_buf_ops);
276 if (error)
277 return error;
278 *count += 1;
279 block = XFS_BUF_TO_BLOCK(bp);
280
281 if (--level) {
282 /* Not at node above leaves, count this level of nodes */
283 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
284 while (nextbno != NULLFSBLOCK) {
285 error = xfs_btree_read_bufl(mp, tp, nextbno, &nbp,
286 XFS_BMAP_BTREE_REF,
287 &xfs_bmbt_buf_ops);
288 if (error)
289 return error;
290 *count += 1;
291 nextblock = XFS_BUF_TO_BLOCK(nbp);
292 nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
293 xfs_trans_brelse(tp, nbp);
294 }
295
296 /* Dive to the next level */
297 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
298 bno = be64_to_cpu(*pp);
299 error = xfs_bmap_count_tree(mp, tp, ifp, bno, level, nextents,
300 count);
301 if (error) {
302 xfs_trans_brelse(tp, bp);
303 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
304 XFS_ERRLEVEL_LOW, mp);
305 return -EFSCORRUPTED;
306 }
307 xfs_trans_brelse(tp, bp);
308 } else {
309 /* count all level 1 nodes and their leaves */
310 for (;;) {
311 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
312 numrecs = be16_to_cpu(block->bb_numrecs);
313 (*nextents) += numrecs;
314 xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
315 xfs_trans_brelse(tp, bp);
316 if (nextbno == NULLFSBLOCK)
317 break;
318 bno = nextbno;
319 error = xfs_btree_read_bufl(mp, tp, bno, &bp,
320 XFS_BMAP_BTREE_REF,
321 &xfs_bmbt_buf_ops);
322 if (error)
323 return error;
324 *count += 1;
325 block = XFS_BUF_TO_BLOCK(bp);
326 }
327 }
328 return 0;
329}
330
331/*
332 * Count fsblocks of the given fork. Delayed allocation extents are
333 * not counted towards the totals.
334 */
335int
336xfs_bmap_count_blocks(
337 struct xfs_trans *tp,
338 struct xfs_inode *ip,
339 int whichfork,
340 xfs_extnum_t *nextents,
341 xfs_filblks_t *count)
342{
343 struct xfs_mount *mp; /* file system mount structure */
344 __be64 *pp; /* pointer to block address */
345 struct xfs_btree_block *block; /* current btree block */
346 struct xfs_ifork *ifp; /* fork structure */
347 xfs_fsblock_t bno; /* block # of "block" */
348 int level; /* btree level, for checking */
349 int error;
350
351 bno = NULLFSBLOCK;
352 mp = ip->i_mount;
353 *nextents = 0;
354 *count = 0;
355 ifp = XFS_IFORK_PTR(ip, whichfork);
356 if (!ifp)
357 return 0;
358
359 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
360 case XFS_DINODE_FMT_EXTENTS:
361 *nextents = xfs_bmap_count_leaves(ifp, count);
362 return 0;
363 case XFS_DINODE_FMT_BTREE:
364 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
365 error = xfs_iread_extents(tp, ip, whichfork);
366 if (error)
367 return error;
368 }
369
370 /*
371 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
372 */
373 block = ifp->if_broot;
374 level = be16_to_cpu(block->bb_level);
375 ASSERT(level > 0);
376 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
377 bno = be64_to_cpu(*pp);
378 ASSERT(bno != NULLFSBLOCK);
379 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
380 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
381
382 error = xfs_bmap_count_tree(mp, tp, ifp, bno, level,
383 nextents, count);
384 if (error) {
385 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)",
386 XFS_ERRLEVEL_LOW, mp);
387 return -EFSCORRUPTED;
388 }
389 return 0;
390 }
391
392 return 0;
393}
394
395static int
396xfs_getbmap_report_one(
397 struct xfs_inode *ip,
398 struct getbmapx *bmv,
399 struct kgetbmap *out,
400 int64_t bmv_end,
401 struct xfs_bmbt_irec *got)
402{
403 struct kgetbmap *p = out + bmv->bmv_entries;
404 bool shared = false;
405 int error;
406
407 error = xfs_reflink_trim_around_shared(ip, got, &shared);
408 if (error)
409 return error;
410
411 if (isnullstartblock(got->br_startblock) ||
412 got->br_startblock == DELAYSTARTBLOCK) {
413 /*
414 * Delalloc extents that start beyond EOF can occur due to
415 * speculative EOF allocation when the delalloc extent is larger
416 * than the largest freespace extent at conversion time. These
417 * extents cannot be converted by data writeback, so can exist
418 * here even if we are not supposed to be finding delalloc
419 * extents.
420 */
421 if (got->br_startoff < XFS_B_TO_FSB(ip->i_mount, XFS_ISIZE(ip)))
422 ASSERT((bmv->bmv_iflags & BMV_IF_DELALLOC) != 0);
423
424 p->bmv_oflags |= BMV_OF_DELALLOC;
425 p->bmv_block = -2;
426 } else {
427 p->bmv_block = xfs_fsb_to_db(ip, got->br_startblock);
428 }
429
430 if (got->br_state == XFS_EXT_UNWRITTEN &&
431 (bmv->bmv_iflags & BMV_IF_PREALLOC))
432 p->bmv_oflags |= BMV_OF_PREALLOC;
433
434 if (shared)
435 p->bmv_oflags |= BMV_OF_SHARED;
436
437 p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, got->br_startoff);
438 p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, got->br_blockcount);
439
440 bmv->bmv_offset = p->bmv_offset + p->bmv_length;
441 bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset);
442 bmv->bmv_entries++;
443 return 0;
444}
445
446static void
447xfs_getbmap_report_hole(
448 struct xfs_inode *ip,
449 struct getbmapx *bmv,
450 struct kgetbmap *out,
451 int64_t bmv_end,
452 xfs_fileoff_t bno,
453 xfs_fileoff_t end)
454{
455 struct kgetbmap *p = out + bmv->bmv_entries;
456
457 if (bmv->bmv_iflags & BMV_IF_NO_HOLES)
458 return;
459
460 p->bmv_block = -1;
461 p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, bno);
462 p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, end - bno);
463
464 bmv->bmv_offset = p->bmv_offset + p->bmv_length;
465 bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset);
466 bmv->bmv_entries++;
467}
468
469static inline bool
470xfs_getbmap_full(
471 struct getbmapx *bmv)
472{
473 return bmv->bmv_length == 0 || bmv->bmv_entries >= bmv->bmv_count - 1;
474}
475
476static bool
477xfs_getbmap_next_rec(
478 struct xfs_bmbt_irec *rec,
479 xfs_fileoff_t total_end)
480{
481 xfs_fileoff_t end = rec->br_startoff + rec->br_blockcount;
482
483 if (end == total_end)
484 return false;
485
486 rec->br_startoff += rec->br_blockcount;
487 if (!isnullstartblock(rec->br_startblock) &&
488 rec->br_startblock != DELAYSTARTBLOCK)
489 rec->br_startblock += rec->br_blockcount;
490 rec->br_blockcount = total_end - end;
491 return true;
492}
493
494/*
495 * Get inode's extents as described in bmv, and format for output.
496 * Calls formatter to fill the user's buffer until all extents
497 * are mapped, until the passed-in bmv->bmv_count slots have
498 * been filled, or until the formatter short-circuits the loop,
499 * if it is tracking filled-in extents on its own.
500 */
501int /* error code */
502xfs_getbmap(
503 struct xfs_inode *ip,
504 struct getbmapx *bmv, /* user bmap structure */
505 struct kgetbmap *out)
506{
507 struct xfs_mount *mp = ip->i_mount;
508 int iflags = bmv->bmv_iflags;
509 int whichfork, lock, error = 0;
510 int64_t bmv_end, max_len;
511 xfs_fileoff_t bno, first_bno;
512 struct xfs_ifork *ifp;
513 struct xfs_bmbt_irec got, rec;
514 xfs_filblks_t len;
515 struct xfs_iext_cursor icur;
516
517 if (bmv->bmv_iflags & ~BMV_IF_VALID)
518 return -EINVAL;
519#ifndef DEBUG
520 /* Only allow CoW fork queries if we're debugging. */
521 if (iflags & BMV_IF_COWFORK)
522 return -EINVAL;
523#endif
524 if ((iflags & BMV_IF_ATTRFORK) && (iflags & BMV_IF_COWFORK))
525 return -EINVAL;
526
527 if (bmv->bmv_length < -1)
528 return -EINVAL;
529 bmv->bmv_entries = 0;
530 if (bmv->bmv_length == 0)
531 return 0;
532
533 if (iflags & BMV_IF_ATTRFORK)
534 whichfork = XFS_ATTR_FORK;
535 else if (iflags & BMV_IF_COWFORK)
536 whichfork = XFS_COW_FORK;
537 else
538 whichfork = XFS_DATA_FORK;
539 ifp = XFS_IFORK_PTR(ip, whichfork);
540
541 xfs_ilock(ip, XFS_IOLOCK_SHARED);
542 switch (whichfork) {
543 case XFS_ATTR_FORK:
544 if (!XFS_IFORK_Q(ip))
545 goto out_unlock_iolock;
546
547 max_len = 1LL << 32;
548 lock = xfs_ilock_attr_map_shared(ip);
549 break;
550 case XFS_COW_FORK:
551 /* No CoW fork? Just return */
552 if (!ifp)
553 goto out_unlock_iolock;
554
555 if (xfs_get_cowextsz_hint(ip))
556 max_len = mp->m_super->s_maxbytes;
557 else
558 max_len = XFS_ISIZE(ip);
559
560 lock = XFS_ILOCK_SHARED;
561 xfs_ilock(ip, lock);
562 break;
563 case XFS_DATA_FORK:
564 if (!(iflags & BMV_IF_DELALLOC) &&
565 (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) {
566 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
567 if (error)
568 goto out_unlock_iolock;
569
570 /*
571 * Even after flushing the inode, there can still be
572 * delalloc blocks on the inode beyond EOF due to
573 * speculative preallocation. These are not removed
574 * until the release function is called or the inode
575 * is inactivated. Hence we cannot assert here that
576 * ip->i_delayed_blks == 0.
577 */
578 }
579
580 if (xfs_get_extsz_hint(ip) ||
581 (ip->i_d.di_flags &
582 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))
583 max_len = mp->m_super->s_maxbytes;
584 else
585 max_len = XFS_ISIZE(ip);
586
587 lock = xfs_ilock_data_map_shared(ip);
588 break;
589 }
590
591 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
592 case XFS_DINODE_FMT_EXTENTS:
593 case XFS_DINODE_FMT_BTREE:
594 break;
595 case XFS_DINODE_FMT_LOCAL:
596 /* Local format inode forks report no extents. */
597 goto out_unlock_ilock;
598 default:
599 error = -EINVAL;
600 goto out_unlock_ilock;
601 }
602
603 if (bmv->bmv_length == -1) {
604 max_len = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, max_len));
605 bmv->bmv_length = max(0LL, max_len - bmv->bmv_offset);
606 }
607
608 bmv_end = bmv->bmv_offset + bmv->bmv_length;
609
610 first_bno = bno = XFS_BB_TO_FSBT(mp, bmv->bmv_offset);
611 len = XFS_BB_TO_FSB(mp, bmv->bmv_length);
612
613 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
614 error = xfs_iread_extents(NULL, ip, whichfork);
615 if (error)
616 goto out_unlock_ilock;
617 }
618
619 if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
620 /*
621 * Report a whole-file hole if the delalloc flag is set to
622 * stay compatible with the old implementation.
623 */
624 if (iflags & BMV_IF_DELALLOC)
625 xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
626 XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
627 goto out_unlock_ilock;
628 }
629
630 while (!xfs_getbmap_full(bmv)) {
631 xfs_trim_extent(&got, first_bno, len);
632
633 /*
634 * Report an entry for a hole if this extent doesn't directly
635 * follow the previous one.
636 */
637 if (got.br_startoff > bno) {
638 xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
639 got.br_startoff);
640 if (xfs_getbmap_full(bmv))
641 break;
642 }
643
644 /*
645 * In order to report shared extents accurately, we report each
646 * distinct shared / unshared part of a single bmbt record with
647 * an individual getbmapx record.
648 */
649 bno = got.br_startoff + got.br_blockcount;
650 rec = got;
651 do {
652 error = xfs_getbmap_report_one(ip, bmv, out, bmv_end,
653 &rec);
654 if (error || xfs_getbmap_full(bmv))
655 goto out_unlock_ilock;
656 } while (xfs_getbmap_next_rec(&rec, bno));
657
658 if (!xfs_iext_next_extent(ifp, &icur, &got)) {
659 xfs_fileoff_t end = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
660
661 out[bmv->bmv_entries - 1].bmv_oflags |= BMV_OF_LAST;
662
663 if (whichfork != XFS_ATTR_FORK && bno < end &&
664 !xfs_getbmap_full(bmv)) {
665 xfs_getbmap_report_hole(ip, bmv, out, bmv_end,
666 bno, end);
667 }
668 break;
669 }
670
671 if (bno >= first_bno + len)
672 break;
673 }
674
675out_unlock_ilock:
676 xfs_iunlock(ip, lock);
677out_unlock_iolock:
678 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
679 return error;
680}
681
682/*
683 * Dead simple method of punching delalyed allocation blocks from a range in
684 * the inode. This will always punch out both the start and end blocks, even
685 * if the ranges only partially overlap them, so it is up to the caller to
686 * ensure that partial blocks are not passed in.
687 */
688int
689xfs_bmap_punch_delalloc_range(
690 struct xfs_inode *ip,
691 xfs_fileoff_t start_fsb,
692 xfs_fileoff_t length)
693{
694 struct xfs_ifork *ifp = &ip->i_df;
695 xfs_fileoff_t end_fsb = start_fsb + length;
696 struct xfs_bmbt_irec got, del;
697 struct xfs_iext_cursor icur;
698 int error = 0;
699
700 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
701
702 xfs_ilock(ip, XFS_ILOCK_EXCL);
703 if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got))
704 goto out_unlock;
705
706 while (got.br_startoff + got.br_blockcount > start_fsb) {
707 del = got;
708 xfs_trim_extent(&del, start_fsb, length);
709
710 /*
711 * A delete can push the cursor forward. Step back to the
712 * previous extent on non-delalloc or extents outside the
713 * target range.
714 */
715 if (!del.br_blockcount ||
716 !isnullstartblock(del.br_startblock)) {
717 if (!xfs_iext_prev_extent(ifp, &icur, &got))
718 break;
719 continue;
720 }
721
722 error = xfs_bmap_del_extent_delay(ip, XFS_DATA_FORK, &icur,
723 &got, &del);
724 if (error || !xfs_iext_get_extent(ifp, &icur, &got))
725 break;
726 }
727
728out_unlock:
729 xfs_iunlock(ip, XFS_ILOCK_EXCL);
730 return error;
731}
732
733/*
734 * Test whether it is appropriate to check an inode for and free post EOF
735 * blocks. The 'force' parameter determines whether we should also consider
736 * regular files that are marked preallocated or append-only.
737 */
738bool
739xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
740{
741 /* prealloc/delalloc exists only on regular files */
742 if (!S_ISREG(VFS_I(ip)->i_mode))
743 return false;
744
745 /*
746 * Zero sized files with no cached pages and delalloc blocks will not
747 * have speculative prealloc/delalloc blocks to remove.
748 */
749 if (VFS_I(ip)->i_size == 0 &&
750 VFS_I(ip)->i_mapping->nrpages == 0 &&
751 ip->i_delayed_blks == 0)
752 return false;
753
754 /* If we haven't read in the extent list, then don't do it now. */
755 if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
756 return false;
757
758 /*
759 * Do not free real preallocated or append-only files unless the file
760 * has delalloc blocks and we are forced to remove them.
761 */
762 if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
763 if (!force || ip->i_delayed_blks == 0)
764 return false;
765
766 return true;
767}
768
769/*
770 * This is called to free any blocks beyond eof. The caller must hold
771 * IOLOCK_EXCL unless we are in the inode reclaim path and have the only
772 * reference to the inode.
773 */
774int
775xfs_free_eofblocks(
776 struct xfs_inode *ip)
777{
778 struct xfs_trans *tp;
779 int error;
780 xfs_fileoff_t end_fsb;
781 xfs_fileoff_t last_fsb;
782 xfs_filblks_t map_len;
783 int nimaps;
784 struct xfs_bmbt_irec imap;
785 struct xfs_mount *mp = ip->i_mount;
786
787 /*
788 * Figure out if there are any blocks beyond the end
789 * of the file. If not, then there is nothing to do.
790 */
791 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
792 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
793 if (last_fsb <= end_fsb)
794 return 0;
795 map_len = last_fsb - end_fsb;
796
797 nimaps = 1;
798 xfs_ilock(ip, XFS_ILOCK_SHARED);
799 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
800 xfs_iunlock(ip, XFS_ILOCK_SHARED);
801
802 /*
803 * If there are blocks after the end of file, truncate the file to its
804 * current size to free them up.
805 */
806 if (!error && (nimaps != 0) &&
807 (imap.br_startblock != HOLESTARTBLOCK ||
808 ip->i_delayed_blks)) {
809 /*
810 * Attach the dquots to the inode up front.
811 */
812 error = xfs_qm_dqattach(ip);
813 if (error)
814 return error;
815
816 /* wait on dio to ensure i_size has settled */
817 inode_dio_wait(VFS_I(ip));
818
819 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0,
820 &tp);
821 if (error) {
822 ASSERT(XFS_FORCED_SHUTDOWN(mp));
823 return error;
824 }
825
826 xfs_ilock(ip, XFS_ILOCK_EXCL);
827 xfs_trans_ijoin(tp, ip, 0);
828
829 /*
830 * Do not update the on-disk file size. If we update the
831 * on-disk file size and then the system crashes before the
832 * contents of the file are flushed to disk then the files
833 * may be full of holes (ie NULL files bug).
834 */
835 error = xfs_itruncate_extents_flags(&tp, ip, XFS_DATA_FORK,
836 XFS_ISIZE(ip), XFS_BMAPI_NODISCARD);
837 if (error) {
838 /*
839 * If we get an error at this point we simply don't
840 * bother truncating the file.
841 */
842 xfs_trans_cancel(tp);
843 } else {
844 error = xfs_trans_commit(tp);
845 if (!error)
846 xfs_inode_clear_eofblocks_tag(ip);
847 }
848
849 xfs_iunlock(ip, XFS_ILOCK_EXCL);
850 }
851 return error;
852}
853
854int
855xfs_alloc_file_space(
856 struct xfs_inode *ip,
857 xfs_off_t offset,
858 xfs_off_t len,
859 int alloc_type)
860{
861 xfs_mount_t *mp = ip->i_mount;
862 xfs_off_t count;
863 xfs_filblks_t allocated_fsb;
864 xfs_filblks_t allocatesize_fsb;
865 xfs_extlen_t extsz, temp;
866 xfs_fileoff_t startoffset_fsb;
867 xfs_fileoff_t endoffset_fsb;
868 int nimaps;
869 int quota_flag;
870 int rt;
871 xfs_trans_t *tp;
872 xfs_bmbt_irec_t imaps[1], *imapp;
873 uint qblocks, resblks, resrtextents;
874 int error;
875
876 trace_xfs_alloc_file_space(ip);
877
878 if (XFS_FORCED_SHUTDOWN(mp))
879 return -EIO;
880
881 error = xfs_qm_dqattach(ip);
882 if (error)
883 return error;
884
885 if (len <= 0)
886 return -EINVAL;
887
888 rt = XFS_IS_REALTIME_INODE(ip);
889 extsz = xfs_get_extsz_hint(ip);
890
891 count = len;
892 imapp = &imaps[0];
893 nimaps = 1;
894 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
895 endoffset_fsb = XFS_B_TO_FSB(mp, offset + count);
896 allocatesize_fsb = endoffset_fsb - startoffset_fsb;
897
898 /*
899 * Allocate file space until done or until there is an error
900 */
901 while (allocatesize_fsb && !error) {
902 xfs_fileoff_t s, e;
903
904 /*
905 * Determine space reservations for data/realtime.
906 */
907 if (unlikely(extsz)) {
908 s = startoffset_fsb;
909 do_div(s, extsz);
910 s *= extsz;
911 e = startoffset_fsb + allocatesize_fsb;
912 div_u64_rem(startoffset_fsb, extsz, &temp);
913 if (temp)
914 e += temp;
915 div_u64_rem(e, extsz, &temp);
916 if (temp)
917 e += extsz - temp;
918 } else {
919 s = 0;
920 e = allocatesize_fsb;
921 }
922
923 /*
924 * The transaction reservation is limited to a 32-bit block
925 * count, hence we need to limit the number of blocks we are
926 * trying to reserve to avoid an overflow. We can't allocate
927 * more than @nimaps extents, and an extent is limited on disk
928 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
929 */
930 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
931 if (unlikely(rt)) {
932 resrtextents = qblocks = resblks;
933 resrtextents /= mp->m_sb.sb_rextsize;
934 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
935 quota_flag = XFS_QMOPT_RES_RTBLKS;
936 } else {
937 resrtextents = 0;
938 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
939 quota_flag = XFS_QMOPT_RES_REGBLKS;
940 }
941
942 /*
943 * Allocate and setup the transaction.
944 */
945 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks,
946 resrtextents, 0, &tp);
947
948 /*
949 * Check for running out of space
950 */
951 if (error) {
952 /*
953 * Free the transaction structure.
954 */
955 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
956 break;
957 }
958 xfs_ilock(ip, XFS_ILOCK_EXCL);
959 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
960 0, quota_flag);
961 if (error)
962 goto error1;
963
964 xfs_trans_ijoin(tp, ip, 0);
965
966 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
967 allocatesize_fsb, alloc_type, resblks,
968 imapp, &nimaps);
969 if (error)
970 goto error0;
971
972 /*
973 * Complete the transaction
974 */
975 error = xfs_trans_commit(tp);
976 xfs_iunlock(ip, XFS_ILOCK_EXCL);
977 if (error)
978 break;
979
980 allocated_fsb = imapp->br_blockcount;
981
982 if (nimaps == 0) {
983 error = -ENOSPC;
984 break;
985 }
986
987 startoffset_fsb += allocated_fsb;
988 allocatesize_fsb -= allocated_fsb;
989 }
990
991 return error;
992
993error0: /* unlock inode, unreserve quota blocks, cancel trans */
994 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
995
996error1: /* Just cancel transaction */
997 xfs_trans_cancel(tp);
998 xfs_iunlock(ip, XFS_ILOCK_EXCL);
999 return error;
1000}
1001
1002static int
1003xfs_unmap_extent(
1004 struct xfs_inode *ip,
1005 xfs_fileoff_t startoffset_fsb,
1006 xfs_filblks_t len_fsb,
1007 int *done)
1008{
1009 struct xfs_mount *mp = ip->i_mount;
1010 struct xfs_trans *tp;
1011 uint resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1012 int error;
1013
1014 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
1015 if (error) {
1016 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1017 return error;
1018 }
1019
1020 xfs_ilock(ip, XFS_ILOCK_EXCL);
1021 error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot, ip->i_gdquot,
1022 ip->i_pdquot, resblks, 0, XFS_QMOPT_RES_REGBLKS);
1023 if (error)
1024 goto out_trans_cancel;
1025
1026 xfs_trans_ijoin(tp, ip, 0);
1027
1028 error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, done);
1029 if (error)
1030 goto out_trans_cancel;
1031
1032 error = xfs_trans_commit(tp);
1033out_unlock:
1034 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1035 return error;
1036
1037out_trans_cancel:
1038 xfs_trans_cancel(tp);
1039 goto out_unlock;
1040}
1041
1042int
1043xfs_flush_unmap_range(
1044 struct xfs_inode *ip,
1045 xfs_off_t offset,
1046 xfs_off_t len)
1047{
1048 struct xfs_mount *mp = ip->i_mount;
1049 struct inode *inode = VFS_I(ip);
1050 xfs_off_t rounding, start, end;
1051 int error;
1052
1053 /* wait for the completion of any pending DIOs */
1054 inode_dio_wait(inode);
1055
1056 rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_SIZE);
1057 start = round_down(offset, rounding);
1058 end = round_up(offset + len, rounding) - 1;
1059
1060 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
1061 if (error)
1062 return error;
1063 truncate_pagecache_range(inode, start, end);
1064 return 0;
1065}
1066
1067int
1068xfs_free_file_space(
1069 struct xfs_inode *ip,
1070 xfs_off_t offset,
1071 xfs_off_t len)
1072{
1073 struct xfs_mount *mp = ip->i_mount;
1074 xfs_fileoff_t startoffset_fsb;
1075 xfs_fileoff_t endoffset_fsb;
1076 int done = 0, error;
1077
1078 trace_xfs_free_file_space(ip);
1079
1080 error = xfs_qm_dqattach(ip);
1081 if (error)
1082 return error;
1083
1084 if (len <= 0) /* if nothing being freed */
1085 return 0;
1086
1087 error = xfs_flush_unmap_range(ip, offset, len);
1088 if (error)
1089 return error;
1090
1091 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
1092 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
1093
1094 /*
1095 * Need to zero the stuff we're not freeing, on disk.
1096 */
1097 if (endoffset_fsb > startoffset_fsb) {
1098 while (!done) {
1099 error = xfs_unmap_extent(ip, startoffset_fsb,
1100 endoffset_fsb - startoffset_fsb, &done);
1101 if (error)
1102 return error;
1103 }
1104 }
1105
1106 /*
1107 * Now that we've unmap all full blocks we'll have to zero out any
1108 * partial block at the beginning and/or end. iomap_zero_range is smart
1109 * enough to skip any holes, including those we just created, but we
1110 * must take care not to zero beyond EOF and enlarge i_size.
1111 */
1112 if (offset >= XFS_ISIZE(ip))
1113 return 0;
1114 if (offset + len > XFS_ISIZE(ip))
1115 len = XFS_ISIZE(ip) - offset;
1116 error = iomap_zero_range(VFS_I(ip), offset, len, NULL, &xfs_iomap_ops);
1117 if (error)
1118 return error;
1119
1120 /*
1121 * If we zeroed right up to EOF and EOF straddles a page boundary we
1122 * must make sure that the post-EOF area is also zeroed because the
1123 * page could be mmap'd and iomap_zero_range doesn't do that for us.
1124 * Writeback of the eof page will do this, albeit clumsily.
1125 */
1126 if (offset + len >= XFS_ISIZE(ip) && offset_in_page(offset + len) > 0) {
1127 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1128 round_down(offset + len, PAGE_SIZE), LLONG_MAX);
1129 }
1130
1131 return error;
1132}
1133
1134/*
1135 * Preallocate and zero a range of a file. This mechanism has the allocation
1136 * semantics of fallocate and in addition converts data in the range to zeroes.
1137 */
1138int
1139xfs_zero_file_space(
1140 struct xfs_inode *ip,
1141 xfs_off_t offset,
1142 xfs_off_t len)
1143{
1144 struct xfs_mount *mp = ip->i_mount;
1145 uint blksize;
1146 int error;
1147
1148 trace_xfs_zero_file_space(ip);
1149
1150 blksize = 1 << mp->m_sb.sb_blocklog;
1151
1152 /*
1153 * Punch a hole and prealloc the range. We use hole punch rather than
1154 * unwritten extent conversion for two reasons:
1155 *
1156 * 1.) Hole punch handles partial block zeroing for us.
1157 *
1158 * 2.) If prealloc returns ENOSPC, the file range is still zero-valued
1159 * by virtue of the hole punch.
1160 */
1161 error = xfs_free_file_space(ip, offset, len);
1162 if (error || xfs_is_always_cow_inode(ip))
1163 return error;
1164
1165 return xfs_alloc_file_space(ip, round_down(offset, blksize),
1166 round_up(offset + len, blksize) -
1167 round_down(offset, blksize),
1168 XFS_BMAPI_PREALLOC);
1169}
1170
1171static int
1172xfs_prepare_shift(
1173 struct xfs_inode *ip,
1174 loff_t offset)
1175{
1176 int error;
1177
1178 /*
1179 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
1180 * into the accessible region of the file.
1181 */
1182 if (xfs_can_free_eofblocks(ip, true)) {
1183 error = xfs_free_eofblocks(ip);
1184 if (error)
1185 return error;
1186 }
1187
1188 /*
1189 * Writeback and invalidate cache for the remainder of the file as we're
1190 * about to shift down every extent from offset to EOF.
1191 */
1192 error = xfs_flush_unmap_range(ip, offset, XFS_ISIZE(ip));
1193 if (error)
1194 return error;
1195
1196 /*
1197 * Clean out anything hanging around in the cow fork now that
1198 * we've flushed all the dirty data out to disk to avoid having
1199 * CoW extents at the wrong offsets.
1200 */
1201 if (xfs_inode_has_cow_data(ip)) {
1202 error = xfs_reflink_cancel_cow_range(ip, offset, NULLFILEOFF,
1203 true);
1204 if (error)
1205 return error;
1206 }
1207
1208 return 0;
1209}
1210
1211/*
1212 * xfs_collapse_file_space()
1213 * This routine frees disk space and shift extent for the given file.
1214 * The first thing we do is to free data blocks in the specified range
1215 * by calling xfs_free_file_space(). It would also sync dirty data
1216 * and invalidate page cache over the region on which collapse range
1217 * is working. And Shift extent records to the left to cover a hole.
1218 * RETURNS:
1219 * 0 on success
1220 * errno on error
1221 *
1222 */
1223int
1224xfs_collapse_file_space(
1225 struct xfs_inode *ip,
1226 xfs_off_t offset,
1227 xfs_off_t len)
1228{
1229 struct xfs_mount *mp = ip->i_mount;
1230 struct xfs_trans *tp;
1231 int error;
1232 xfs_fileoff_t next_fsb = XFS_B_TO_FSB(mp, offset + len);
1233 xfs_fileoff_t shift_fsb = XFS_B_TO_FSB(mp, len);
1234 uint resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1235 bool done = false;
1236
1237 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1238 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
1239
1240 trace_xfs_collapse_file_space(ip);
1241
1242 error = xfs_free_file_space(ip, offset, len);
1243 if (error)
1244 return error;
1245
1246 error = xfs_prepare_shift(ip, offset);
1247 if (error)
1248 return error;
1249
1250 while (!error && !done) {
1251 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0,
1252 &tp);
1253 if (error)
1254 break;
1255
1256 xfs_ilock(ip, XFS_ILOCK_EXCL);
1257 error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot,
1258 ip->i_gdquot, ip->i_pdquot, resblks, 0,
1259 XFS_QMOPT_RES_REGBLKS);
1260 if (error)
1261 goto out_trans_cancel;
1262 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1263
1264 error = xfs_bmap_collapse_extents(tp, ip, &next_fsb, shift_fsb,
1265 &done);
1266 if (error)
1267 goto out_trans_cancel;
1268
1269 error = xfs_trans_commit(tp);
1270 }
1271
1272 return error;
1273
1274out_trans_cancel:
1275 xfs_trans_cancel(tp);
1276 return error;
1277}
1278
1279/*
1280 * xfs_insert_file_space()
1281 * This routine create hole space by shifting extents for the given file.
1282 * The first thing we do is to sync dirty data and invalidate page cache
1283 * over the region on which insert range is working. And split an extent
1284 * to two extents at given offset by calling xfs_bmap_split_extent.
1285 * And shift all extent records which are laying between [offset,
1286 * last allocated extent] to the right to reserve hole range.
1287 * RETURNS:
1288 * 0 on success
1289 * errno on error
1290 */
1291int
1292xfs_insert_file_space(
1293 struct xfs_inode *ip,
1294 loff_t offset,
1295 loff_t len)
1296{
1297 struct xfs_mount *mp = ip->i_mount;
1298 struct xfs_trans *tp;
1299 int error;
1300 xfs_fileoff_t stop_fsb = XFS_B_TO_FSB(mp, offset);
1301 xfs_fileoff_t next_fsb = NULLFSBLOCK;
1302 xfs_fileoff_t shift_fsb = XFS_B_TO_FSB(mp, len);
1303 bool done = false;
1304
1305 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1306 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
1307
1308 trace_xfs_insert_file_space(ip);
1309
1310 error = xfs_bmap_can_insert_extents(ip, stop_fsb, shift_fsb);
1311 if (error)
1312 return error;
1313
1314 error = xfs_prepare_shift(ip, offset);
1315 if (error)
1316 return error;
1317
1318 /*
1319 * The extent shifting code works on extent granularity. So, if stop_fsb
1320 * is not the starting block of extent, we need to split the extent at
1321 * stop_fsb.
1322 */
1323 error = xfs_bmap_split_extent(ip, stop_fsb);
1324 if (error)
1325 return error;
1326
1327 while (!error && !done) {
1328 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0,
1329 &tp);
1330 if (error)
1331 break;
1332
1333 xfs_ilock(ip, XFS_ILOCK_EXCL);
1334 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1335 error = xfs_bmap_insert_extents(tp, ip, &next_fsb, shift_fsb,
1336 &done, stop_fsb);
1337 if (error)
1338 goto out_trans_cancel;
1339
1340 error = xfs_trans_commit(tp);
1341 }
1342
1343 return error;
1344
1345out_trans_cancel:
1346 xfs_trans_cancel(tp);
1347 return error;
1348}
1349
1350/*
1351 * We need to check that the format of the data fork in the temporary inode is
1352 * valid for the target inode before doing the swap. This is not a problem with
1353 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1354 * data fork depending on the space the attribute fork is taking so we can get
1355 * invalid formats on the target inode.
1356 *
1357 * E.g. target has space for 7 extents in extent format, temp inode only has
1358 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1359 * btree, but when swapped it needs to be in extent format. Hence we can't just
1360 * blindly swap data forks on attr2 filesystems.
1361 *
1362 * Note that we check the swap in both directions so that we don't end up with
1363 * a corrupt temporary inode, either.
1364 *
1365 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1366 * inode will prevent this situation from occurring, so all we do here is
1367 * reject and log the attempt. basically we are putting the responsibility on
1368 * userspace to get this right.
1369 */
1370static int
1371xfs_swap_extents_check_format(
1372 struct xfs_inode *ip, /* target inode */
1373 struct xfs_inode *tip) /* tmp inode */
1374{
1375
1376 /* Should never get a local format */
1377 if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
1378 tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
1379 return -EINVAL;
1380
1381 /*
1382 * if the target inode has less extents that then temporary inode then
1383 * why did userspace call us?
1384 */
1385 if (ip->i_d.di_nextents < tip->i_d.di_nextents)
1386 return -EINVAL;
1387
1388 /*
1389 * If we have to use the (expensive) rmap swap method, we can
1390 * handle any number of extents and any format.
1391 */
1392 if (xfs_sb_version_hasrmapbt(&ip->i_mount->m_sb))
1393 return 0;
1394
1395 /*
1396 * if the target inode is in extent form and the temp inode is in btree
1397 * form then we will end up with the target inode in the wrong format
1398 * as we already know there are less extents in the temp inode.
1399 */
1400 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1401 tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1402 return -EINVAL;
1403
1404 /* Check temp in extent form to max in target */
1405 if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1406 XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) >
1407 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1408 return -EINVAL;
1409
1410 /* Check target in extent form to max in temp */
1411 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1412 XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) >
1413 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1414 return -EINVAL;
1415
1416 /*
1417 * If we are in a btree format, check that the temp root block will fit
1418 * in the target and that it has enough extents to be in btree format
1419 * in the target.
1420 *
1421 * Note that we have to be careful to allow btree->extent conversions
1422 * (a common defrag case) which will occur when the temp inode is in
1423 * extent format...
1424 */
1425 if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1426 if (XFS_IFORK_Q(ip) &&
1427 XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip))
1428 return -EINVAL;
1429 if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <=
1430 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1431 return -EINVAL;
1432 }
1433
1434 /* Reciprocal target->temp btree format checks */
1435 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1436 if (XFS_IFORK_Q(tip) &&
1437 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
1438 return -EINVAL;
1439 if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <=
1440 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1441 return -EINVAL;
1442 }
1443
1444 return 0;
1445}
1446
1447static int
1448xfs_swap_extent_flush(
1449 struct xfs_inode *ip)
1450{
1451 int error;
1452
1453 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1454 if (error)
1455 return error;
1456 truncate_pagecache_range(VFS_I(ip), 0, -1);
1457
1458 /* Verify O_DIRECT for ftmp */
1459 if (VFS_I(ip)->i_mapping->nrpages)
1460 return -EINVAL;
1461 return 0;
1462}
1463
1464/*
1465 * Move extents from one file to another, when rmap is enabled.
1466 */
1467STATIC int
1468xfs_swap_extent_rmap(
1469 struct xfs_trans **tpp,
1470 struct xfs_inode *ip,
1471 struct xfs_inode *tip)
1472{
1473 struct xfs_trans *tp = *tpp;
1474 struct xfs_bmbt_irec irec;
1475 struct xfs_bmbt_irec uirec;
1476 struct xfs_bmbt_irec tirec;
1477 xfs_fileoff_t offset_fsb;
1478 xfs_fileoff_t end_fsb;
1479 xfs_filblks_t count_fsb;
1480 int error;
1481 xfs_filblks_t ilen;
1482 xfs_filblks_t rlen;
1483 int nimaps;
1484 uint64_t tip_flags2;
1485
1486 /*
1487 * If the source file has shared blocks, we must flag the donor
1488 * file as having shared blocks so that we get the shared-block
1489 * rmap functions when we go to fix up the rmaps. The flags
1490 * will be switch for reals later.
1491 */
1492 tip_flags2 = tip->i_d.di_flags2;
1493 if (ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)
1494 tip->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK;
1495
1496 offset_fsb = 0;
1497 end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip)));
1498 count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
1499
1500 while (count_fsb) {
1501 /* Read extent from the donor file */
1502 nimaps = 1;
1503 error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec,
1504 &nimaps, 0);
1505 if (error)
1506 goto out;
1507 ASSERT(nimaps == 1);
1508 ASSERT(tirec.br_startblock != DELAYSTARTBLOCK);
1509
1510 trace_xfs_swap_extent_rmap_remap(tip, &tirec);
1511 ilen = tirec.br_blockcount;
1512
1513 /* Unmap the old blocks in the source file. */
1514 while (tirec.br_blockcount) {
1515 ASSERT(tp->t_firstblock == NULLFSBLOCK);
1516 trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec);
1517
1518 /* Read extent from the source file */
1519 nimaps = 1;
1520 error = xfs_bmapi_read(ip, tirec.br_startoff,
1521 tirec.br_blockcount, &irec,
1522 &nimaps, 0);
1523 if (error)
1524 goto out;
1525 ASSERT(nimaps == 1);
1526 ASSERT(tirec.br_startoff == irec.br_startoff);
1527 trace_xfs_swap_extent_rmap_remap_piece(ip, &irec);
1528
1529 /* Trim the extent. */
1530 uirec = tirec;
1531 uirec.br_blockcount = rlen = min_t(xfs_filblks_t,
1532 tirec.br_blockcount,
1533 irec.br_blockcount);
1534 trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec);
1535
1536 /* Remove the mapping from the donor file. */
1537 xfs_bmap_unmap_extent(tp, tip, &uirec);
1538
1539 /* Remove the mapping from the source file. */
1540 xfs_bmap_unmap_extent(tp, ip, &irec);
1541
1542 /* Map the donor file's blocks into the source file. */
1543 xfs_bmap_map_extent(tp, ip, &uirec);
1544
1545 /* Map the source file's blocks into the donor file. */
1546 xfs_bmap_map_extent(tp, tip, &irec);
1547
1548 error = xfs_defer_finish(tpp);
1549 tp = *tpp;
1550 if (error)
1551 goto out;
1552
1553 tirec.br_startoff += rlen;
1554 if (tirec.br_startblock != HOLESTARTBLOCK &&
1555 tirec.br_startblock != DELAYSTARTBLOCK)
1556 tirec.br_startblock += rlen;
1557 tirec.br_blockcount -= rlen;
1558 }
1559
1560 /* Roll on... */
1561 count_fsb -= ilen;
1562 offset_fsb += ilen;
1563 }
1564
1565 tip->i_d.di_flags2 = tip_flags2;
1566 return 0;
1567
1568out:
1569 trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_);
1570 tip->i_d.di_flags2 = tip_flags2;
1571 return error;
1572}
1573
1574/* Swap the extents of two files by swapping data forks. */
1575STATIC int
1576xfs_swap_extent_forks(
1577 struct xfs_trans *tp,
1578 struct xfs_inode *ip,
1579 struct xfs_inode *tip,
1580 int *src_log_flags,
1581 int *target_log_flags)
1582{
1583 xfs_filblks_t aforkblks = 0;
1584 xfs_filblks_t taforkblks = 0;
1585 xfs_extnum_t junk;
1586 uint64_t tmp;
1587 int error;
1588
1589 /*
1590 * Count the number of extended attribute blocks
1591 */
1592 if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
1593 (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1594 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &junk,
1595 &aforkblks);
1596 if (error)
1597 return error;
1598 }
1599 if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
1600 (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1601 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK, &junk,
1602 &taforkblks);
1603 if (error)
1604 return error;
1605 }
1606
1607 /*
1608 * Btree format (v3) inodes have the inode number stamped in the bmbt
1609 * block headers. We can't start changing the bmbt blocks until the
1610 * inode owner change is logged so recovery does the right thing in the
1611 * event of a crash. Set the owner change log flags now and leave the
1612 * bmbt scan as the last step.
1613 */
1614 if (ip->i_d.di_version == 3 &&
1615 ip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1616 (*target_log_flags) |= XFS_ILOG_DOWNER;
1617 if (tip->i_d.di_version == 3 &&
1618 tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1619 (*src_log_flags) |= XFS_ILOG_DOWNER;
1620
1621 /*
1622 * Swap the data forks of the inodes
1623 */
1624 swap(ip->i_df, tip->i_df);
1625
1626 /*
1627 * Fix the on-disk inode values
1628 */
1629 tmp = (uint64_t)ip->i_d.di_nblocks;
1630 ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
1631 tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
1632
1633 swap(ip->i_d.di_nextents, tip->i_d.di_nextents);
1634 swap(ip->i_d.di_format, tip->i_d.di_format);
1635
1636 /*
1637 * The extents in the source inode could still contain speculative
1638 * preallocation beyond EOF (e.g. the file is open but not modified
1639 * while defrag is in progress). In that case, we need to copy over the
1640 * number of delalloc blocks the data fork in the source inode is
1641 * tracking beyond EOF so that when the fork is truncated away when the
1642 * temporary inode is unlinked we don't underrun the i_delayed_blks
1643 * counter on that inode.
1644 */
1645 ASSERT(tip->i_delayed_blks == 0);
1646 tip->i_delayed_blks = ip->i_delayed_blks;
1647 ip->i_delayed_blks = 0;
1648
1649 switch (ip->i_d.di_format) {
1650 case XFS_DINODE_FMT_EXTENTS:
1651 (*src_log_flags) |= XFS_ILOG_DEXT;
1652 break;
1653 case XFS_DINODE_FMT_BTREE:
1654 ASSERT(ip->i_d.di_version < 3 ||
1655 (*src_log_flags & XFS_ILOG_DOWNER));
1656 (*src_log_flags) |= XFS_ILOG_DBROOT;
1657 break;
1658 }
1659
1660 switch (tip->i_d.di_format) {
1661 case XFS_DINODE_FMT_EXTENTS:
1662 (*target_log_flags) |= XFS_ILOG_DEXT;
1663 break;
1664 case XFS_DINODE_FMT_BTREE:
1665 (*target_log_flags) |= XFS_ILOG_DBROOT;
1666 ASSERT(tip->i_d.di_version < 3 ||
1667 (*target_log_flags & XFS_ILOG_DOWNER));
1668 break;
1669 }
1670
1671 return 0;
1672}
1673
1674/*
1675 * Fix up the owners of the bmbt blocks to refer to the current inode. The
1676 * change owner scan attempts to order all modified buffers in the current
1677 * transaction. In the event of ordered buffer failure, the offending buffer is
1678 * physically logged as a fallback and the scan returns -EAGAIN. We must roll
1679 * the transaction in this case to replenish the fallback log reservation and
1680 * restart the scan. This process repeats until the scan completes.
1681 */
1682static int
1683xfs_swap_change_owner(
1684 struct xfs_trans **tpp,
1685 struct xfs_inode *ip,
1686 struct xfs_inode *tmpip)
1687{
1688 int error;
1689 struct xfs_trans *tp = *tpp;
1690
1691 do {
1692 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK, ip->i_ino,
1693 NULL);
1694 /* success or fatal error */
1695 if (error != -EAGAIN)
1696 break;
1697
1698 error = xfs_trans_roll(tpp);
1699 if (error)
1700 break;
1701 tp = *tpp;
1702
1703 /*
1704 * Redirty both inodes so they can relog and keep the log tail
1705 * moving forward.
1706 */
1707 xfs_trans_ijoin(tp, ip, 0);
1708 xfs_trans_ijoin(tp, tmpip, 0);
1709 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1710 xfs_trans_log_inode(tp, tmpip, XFS_ILOG_CORE);
1711 } while (true);
1712
1713 return error;
1714}
1715
1716int
1717xfs_swap_extents(
1718 struct xfs_inode *ip, /* target inode */
1719 struct xfs_inode *tip, /* tmp inode */
1720 struct xfs_swapext *sxp)
1721{
1722 struct xfs_mount *mp = ip->i_mount;
1723 struct xfs_trans *tp;
1724 struct xfs_bstat *sbp = &sxp->sx_stat;
1725 int src_log_flags, target_log_flags;
1726 int error = 0;
1727 int lock_flags;
1728 uint64_t f;
1729 int resblks = 0;
1730
1731 /*
1732 * Lock the inodes against other IO, page faults and truncate to
1733 * begin with. Then we can ensure the inodes are flushed and have no
1734 * page cache safely. Once we have done this we can take the ilocks and
1735 * do the rest of the checks.
1736 */
1737 lock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1738 lock_flags = XFS_MMAPLOCK_EXCL;
1739 xfs_lock_two_inodes(ip, XFS_MMAPLOCK_EXCL, tip, XFS_MMAPLOCK_EXCL);
1740
1741 /* Verify that both files have the same format */
1742 if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) {
1743 error = -EINVAL;
1744 goto out_unlock;
1745 }
1746
1747 /* Verify both files are either real-time or non-realtime */
1748 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1749 error = -EINVAL;
1750 goto out_unlock;
1751 }
1752
1753 error = xfs_swap_extent_flush(ip);
1754 if (error)
1755 goto out_unlock;
1756 error = xfs_swap_extent_flush(tip);
1757 if (error)
1758 goto out_unlock;
1759
1760 if (xfs_inode_has_cow_data(tip)) {
1761 error = xfs_reflink_cancel_cow_range(tip, 0, NULLFILEOFF, true);
1762 if (error)
1763 return error;
1764 }
1765
1766 /*
1767 * Extent "swapping" with rmap requires a permanent reservation and
1768 * a block reservation because it's really just a remap operation
1769 * performed with log redo items!
1770 */
1771 if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
1772 int w = XFS_DATA_FORK;
1773 uint32_t ipnext = XFS_IFORK_NEXTENTS(ip, w);
1774 uint32_t tipnext = XFS_IFORK_NEXTENTS(tip, w);
1775
1776 /*
1777 * Conceptually this shouldn't affect the shape of either bmbt,
1778 * but since we atomically move extents one by one, we reserve
1779 * enough space to rebuild both trees.
1780 */
1781 resblks = XFS_SWAP_RMAP_SPACE_RES(mp, ipnext, w);
1782 resblks += XFS_SWAP_RMAP_SPACE_RES(mp, tipnext, w);
1783
1784 /*
1785 * Handle the corner case where either inode might straddle the
1786 * btree format boundary. If so, the inode could bounce between
1787 * btree <-> extent format on unmap -> remap cycles, freeing and
1788 * allocating a bmapbt block each time.
1789 */
1790 if (ipnext == (XFS_IFORK_MAXEXT(ip, w) + 1))
1791 resblks += XFS_IFORK_MAXEXT(ip, w);
1792 if (tipnext == (XFS_IFORK_MAXEXT(tip, w) + 1))
1793 resblks += XFS_IFORK_MAXEXT(tip, w);
1794 }
1795 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
1796 if (error)
1797 goto out_unlock;
1798
1799 /*
1800 * Lock and join the inodes to the tansaction so that transaction commit
1801 * or cancel will unlock the inodes from this point onwards.
1802 */
1803 xfs_lock_two_inodes(ip, XFS_ILOCK_EXCL, tip, XFS_ILOCK_EXCL);
1804 lock_flags |= XFS_ILOCK_EXCL;
1805 xfs_trans_ijoin(tp, ip, 0);
1806 xfs_trans_ijoin(tp, tip, 0);
1807
1808
1809 /* Verify all data are being swapped */
1810 if (sxp->sx_offset != 0 ||
1811 sxp->sx_length != ip->i_d.di_size ||
1812 sxp->sx_length != tip->i_d.di_size) {
1813 error = -EFAULT;
1814 goto out_trans_cancel;
1815 }
1816
1817 trace_xfs_swap_extent_before(ip, 0);
1818 trace_xfs_swap_extent_before(tip, 1);
1819
1820 /* check inode formats now that data is flushed */
1821 error = xfs_swap_extents_check_format(ip, tip);
1822 if (error) {
1823 xfs_notice(mp,
1824 "%s: inode 0x%llx format is incompatible for exchanging.",
1825 __func__, ip->i_ino);
1826 goto out_trans_cancel;
1827 }
1828
1829 /*
1830 * Compare the current change & modify times with that
1831 * passed in. If they differ, we abort this swap.
1832 * This is the mechanism used to ensure the calling
1833 * process that the file was not changed out from
1834 * under it.
1835 */
1836 if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
1837 (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
1838 (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
1839 (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
1840 error = -EBUSY;
1841 goto out_trans_cancel;
1842 }
1843
1844 /*
1845 * Note the trickiness in setting the log flags - we set the owner log
1846 * flag on the opposite inode (i.e. the inode we are setting the new
1847 * owner to be) because once we swap the forks and log that, log
1848 * recovery is going to see the fork as owned by the swapped inode,
1849 * not the pre-swapped inodes.
1850 */
1851 src_log_flags = XFS_ILOG_CORE;
1852 target_log_flags = XFS_ILOG_CORE;
1853
1854 if (xfs_sb_version_hasrmapbt(&mp->m_sb))
1855 error = xfs_swap_extent_rmap(&tp, ip, tip);
1856 else
1857 error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags,
1858 &target_log_flags);
1859 if (error)
1860 goto out_trans_cancel;
1861
1862 /* Do we have to swap reflink flags? */
1863 if ((ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK) ^
1864 (tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)) {
1865 f = ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
1866 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1867 ip->i_d.di_flags2 |= tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
1868 tip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1869 tip->i_d.di_flags2 |= f & XFS_DIFLAG2_REFLINK;
1870 }
1871
1872 /* Swap the cow forks. */
1873 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1874 ASSERT(ip->i_cformat == XFS_DINODE_FMT_EXTENTS);
1875 ASSERT(tip->i_cformat == XFS_DINODE_FMT_EXTENTS);
1876
1877 swap(ip->i_cnextents, tip->i_cnextents);
1878 swap(ip->i_cowfp, tip->i_cowfp);
1879
1880 if (ip->i_cowfp && ip->i_cowfp->if_bytes)
1881 xfs_inode_set_cowblocks_tag(ip);
1882 else
1883 xfs_inode_clear_cowblocks_tag(ip);
1884 if (tip->i_cowfp && tip->i_cowfp->if_bytes)
1885 xfs_inode_set_cowblocks_tag(tip);
1886 else
1887 xfs_inode_clear_cowblocks_tag(tip);
1888 }
1889
1890 xfs_trans_log_inode(tp, ip, src_log_flags);
1891 xfs_trans_log_inode(tp, tip, target_log_flags);
1892
1893 /*
1894 * The extent forks have been swapped, but crc=1,rmapbt=0 filesystems
1895 * have inode number owner values in the bmbt blocks that still refer to
1896 * the old inode. Scan each bmbt to fix up the owner values with the
1897 * inode number of the current inode.
1898 */
1899 if (src_log_flags & XFS_ILOG_DOWNER) {
1900 error = xfs_swap_change_owner(&tp, ip, tip);
1901 if (error)
1902 goto out_trans_cancel;
1903 }
1904 if (target_log_flags & XFS_ILOG_DOWNER) {
1905 error = xfs_swap_change_owner(&tp, tip, ip);
1906 if (error)
1907 goto out_trans_cancel;
1908 }
1909
1910 /*
1911 * If this is a synchronous mount, make sure that the
1912 * transaction goes to disk before returning to the user.
1913 */
1914 if (mp->m_flags & XFS_MOUNT_WSYNC)
1915 xfs_trans_set_sync(tp);
1916
1917 error = xfs_trans_commit(tp);
1918
1919 trace_xfs_swap_extent_after(ip, 0);
1920 trace_xfs_swap_extent_after(tip, 1);
1921
1922out_unlock:
1923 xfs_iunlock(ip, lock_flags);
1924 xfs_iunlock(tip, lock_flags);
1925 unlock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1926 return error;
1927
1928out_trans_cancel:
1929 xfs_trans_cancel(tp);
1930 goto out_unlock;
1931}