Loading...
Note: File does not exist in v3.1.
1/*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * Copyright (c) 2012 Red Hat, Inc.
4 * All Rights Reserved.
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 */
19#include "xfs.h"
20#include "xfs_fs.h"
21#include "xfs_shared.h"
22#include "xfs_format.h"
23#include "xfs_log_format.h"
24#include "xfs_trans_resv.h"
25#include "xfs_bit.h"
26#include "xfs_mount.h"
27#include "xfs_da_format.h"
28#include "xfs_defer.h"
29#include "xfs_inode.h"
30#include "xfs_btree.h"
31#include "xfs_trans.h"
32#include "xfs_extfree_item.h"
33#include "xfs_alloc.h"
34#include "xfs_bmap.h"
35#include "xfs_bmap_util.h"
36#include "xfs_bmap_btree.h"
37#include "xfs_rtalloc.h"
38#include "xfs_error.h"
39#include "xfs_quota.h"
40#include "xfs_trans_space.h"
41#include "xfs_trace.h"
42#include "xfs_icache.h"
43#include "xfs_log.h"
44#include "xfs_rmap_btree.h"
45#include "xfs_iomap.h"
46#include "xfs_reflink.h"
47#include "xfs_refcount.h"
48
49/* Kernel only BMAP related definitions and functions */
50
51/*
52 * Convert the given file system block to a disk block. We have to treat it
53 * differently based on whether the file is a real time file or not, because the
54 * bmap code does.
55 */
56xfs_daddr_t
57xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb)
58{
59 return (XFS_IS_REALTIME_INODE(ip) ? \
60 (xfs_daddr_t)XFS_FSB_TO_BB((ip)->i_mount, (fsb)) : \
61 XFS_FSB_TO_DADDR((ip)->i_mount, (fsb)));
62}
63
64/*
65 * Routine to zero an extent on disk allocated to the specific inode.
66 *
67 * The VFS functions take a linearised filesystem block offset, so we have to
68 * convert the sparse xfs fsb to the right format first.
69 * VFS types are real funky, too.
70 */
71int
72xfs_zero_extent(
73 struct xfs_inode *ip,
74 xfs_fsblock_t start_fsb,
75 xfs_off_t count_fsb)
76{
77 struct xfs_mount *mp = ip->i_mount;
78 xfs_daddr_t sector = xfs_fsb_to_db(ip, start_fsb);
79 sector_t block = XFS_BB_TO_FSBT(mp, sector);
80
81 return blkdev_issue_zeroout(xfs_find_bdev_for_inode(VFS_I(ip)),
82 block << (mp->m_super->s_blocksize_bits - 9),
83 count_fsb << (mp->m_super->s_blocksize_bits - 9),
84 GFP_NOFS, true);
85}
86
87int
88xfs_bmap_rtalloc(
89 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
90{
91 xfs_alloctype_t atype = 0; /* type for allocation routines */
92 int error; /* error return value */
93 xfs_mount_t *mp; /* mount point structure */
94 xfs_extlen_t prod = 0; /* product factor for allocators */
95 xfs_extlen_t ralen = 0; /* realtime allocation length */
96 xfs_extlen_t align; /* minimum allocation alignment */
97 xfs_rtblock_t rtb;
98
99 mp = ap->ip->i_mount;
100 align = xfs_get_extsz_hint(ap->ip);
101 prod = align / mp->m_sb.sb_rextsize;
102 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
103 align, 1, ap->eof, 0,
104 ap->conv, &ap->offset, &ap->length);
105 if (error)
106 return error;
107 ASSERT(ap->length);
108 ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
109
110 /*
111 * If the offset & length are not perfectly aligned
112 * then kill prod, it will just get us in trouble.
113 */
114 if (do_mod(ap->offset, align) || ap->length % align)
115 prod = 1;
116 /*
117 * Set ralen to be the actual requested length in rtextents.
118 */
119 ralen = ap->length / mp->m_sb.sb_rextsize;
120 /*
121 * If the old value was close enough to MAXEXTLEN that
122 * we rounded up to it, cut it back so it's valid again.
123 * Note that if it's a really large request (bigger than
124 * MAXEXTLEN), we don't hear about that number, and can't
125 * adjust the starting point to match it.
126 */
127 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
128 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
129
130 /*
131 * Lock out modifications to both the RT bitmap and summary inodes
132 */
133 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
134 xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
135 xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
136 xfs_trans_ijoin(ap->tp, mp->m_rsumip, XFS_ILOCK_EXCL);
137
138 /*
139 * If it's an allocation to an empty file at offset 0,
140 * pick an extent that will space things out in the rt area.
141 */
142 if (ap->eof && ap->offset == 0) {
143 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
144
145 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
146 if (error)
147 return error;
148 ap->blkno = rtx * mp->m_sb.sb_rextsize;
149 } else {
150 ap->blkno = 0;
151 }
152
153 xfs_bmap_adjacent(ap);
154
155 /*
156 * Realtime allocation, done through xfs_rtallocate_extent.
157 */
158 atype = ap->blkno == 0 ? XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
159 do_div(ap->blkno, mp->m_sb.sb_rextsize);
160 rtb = ap->blkno;
161 ap->length = ralen;
162 if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
163 &ralen, atype, ap->wasdel, prod, &rtb)))
164 return error;
165 if (rtb == NULLFSBLOCK && prod > 1 &&
166 (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1,
167 ap->length, &ralen, atype,
168 ap->wasdel, 1, &rtb)))
169 return error;
170 ap->blkno = rtb;
171 if (ap->blkno != NULLFSBLOCK) {
172 ap->blkno *= mp->m_sb.sb_rextsize;
173 ralen *= mp->m_sb.sb_rextsize;
174 ap->length = ralen;
175 ap->ip->i_d.di_nblocks += ralen;
176 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
177 if (ap->wasdel)
178 ap->ip->i_delayed_blks -= ralen;
179 /*
180 * Adjust the disk quota also. This was reserved
181 * earlier.
182 */
183 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
184 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
185 XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
186
187 /* Zero the extent if we were asked to do so */
188 if (ap->datatype & XFS_ALLOC_USERDATA_ZERO) {
189 error = xfs_zero_extent(ap->ip, ap->blkno, ap->length);
190 if (error)
191 return error;
192 }
193 } else {
194 ap->length = 0;
195 }
196 return 0;
197}
198
199/*
200 * Check if the endoff is outside the last extent. If so the caller will grow
201 * the allocation to a stripe unit boundary. All offsets are considered outside
202 * the end of file for an empty fork, so 1 is returned in *eof in that case.
203 */
204int
205xfs_bmap_eof(
206 struct xfs_inode *ip,
207 xfs_fileoff_t endoff,
208 int whichfork,
209 int *eof)
210{
211 struct xfs_bmbt_irec rec;
212 int error;
213
214 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
215 if (error || *eof)
216 return error;
217
218 *eof = endoff >= rec.br_startoff + rec.br_blockcount;
219 return 0;
220}
221
222/*
223 * Extent tree block counting routines.
224 */
225
226/*
227 * Count leaf blocks given a range of extent records.
228 */
229STATIC void
230xfs_bmap_count_leaves(
231 xfs_ifork_t *ifp,
232 xfs_extnum_t idx,
233 int numrecs,
234 int *count)
235{
236 int b;
237
238 for (b = 0; b < numrecs; b++) {
239 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
240 *count += xfs_bmbt_get_blockcount(frp);
241 }
242}
243
244/*
245 * Count leaf blocks given a range of extent records originally
246 * in btree format.
247 */
248STATIC void
249xfs_bmap_disk_count_leaves(
250 struct xfs_mount *mp,
251 struct xfs_btree_block *block,
252 int numrecs,
253 int *count)
254{
255 int b;
256 xfs_bmbt_rec_t *frp;
257
258 for (b = 1; b <= numrecs; b++) {
259 frp = XFS_BMBT_REC_ADDR(mp, block, b);
260 *count += xfs_bmbt_disk_get_blockcount(frp);
261 }
262}
263
264/*
265 * Recursively walks each level of a btree
266 * to count total fsblocks in use.
267 */
268STATIC int /* error */
269xfs_bmap_count_tree(
270 xfs_mount_t *mp, /* file system mount point */
271 xfs_trans_t *tp, /* transaction pointer */
272 xfs_ifork_t *ifp, /* inode fork pointer */
273 xfs_fsblock_t blockno, /* file system block number */
274 int levelin, /* level in btree */
275 int *count) /* Count of blocks */
276{
277 int error;
278 xfs_buf_t *bp, *nbp;
279 int level = levelin;
280 __be64 *pp;
281 xfs_fsblock_t bno = blockno;
282 xfs_fsblock_t nextbno;
283 struct xfs_btree_block *block, *nextblock;
284 int numrecs;
285
286 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF,
287 &xfs_bmbt_buf_ops);
288 if (error)
289 return error;
290 *count += 1;
291 block = XFS_BUF_TO_BLOCK(bp);
292
293 if (--level) {
294 /* Not at node above leaves, count this level of nodes */
295 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
296 while (nextbno != NULLFSBLOCK) {
297 error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp,
298 XFS_BMAP_BTREE_REF,
299 &xfs_bmbt_buf_ops);
300 if (error)
301 return error;
302 *count += 1;
303 nextblock = XFS_BUF_TO_BLOCK(nbp);
304 nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
305 xfs_trans_brelse(tp, nbp);
306 }
307
308 /* Dive to the next level */
309 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
310 bno = be64_to_cpu(*pp);
311 if (unlikely((error =
312 xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
313 xfs_trans_brelse(tp, bp);
314 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
315 XFS_ERRLEVEL_LOW, mp);
316 return -EFSCORRUPTED;
317 }
318 xfs_trans_brelse(tp, bp);
319 } else {
320 /* count all level 1 nodes and their leaves */
321 for (;;) {
322 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
323 numrecs = be16_to_cpu(block->bb_numrecs);
324 xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
325 xfs_trans_brelse(tp, bp);
326 if (nextbno == NULLFSBLOCK)
327 break;
328 bno = nextbno;
329 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
330 XFS_BMAP_BTREE_REF,
331 &xfs_bmbt_buf_ops);
332 if (error)
333 return error;
334 *count += 1;
335 block = XFS_BUF_TO_BLOCK(bp);
336 }
337 }
338 return 0;
339}
340
341/*
342 * Count fsblocks of the given fork.
343 */
344static int /* error */
345xfs_bmap_count_blocks(
346 xfs_trans_t *tp, /* transaction pointer */
347 xfs_inode_t *ip, /* incore inode */
348 int whichfork, /* data or attr fork */
349 int *count) /* out: count of blocks */
350{
351 struct xfs_btree_block *block; /* current btree block */
352 xfs_fsblock_t bno; /* block # of "block" */
353 xfs_ifork_t *ifp; /* fork structure */
354 int level; /* btree level, for checking */
355 xfs_mount_t *mp; /* file system mount structure */
356 __be64 *pp; /* pointer to block address */
357
358 bno = NULLFSBLOCK;
359 mp = ip->i_mount;
360 ifp = XFS_IFORK_PTR(ip, whichfork);
361 if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
362 xfs_bmap_count_leaves(ifp, 0, xfs_iext_count(ifp), count);
363 return 0;
364 }
365
366 /*
367 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
368 */
369 block = ifp->if_broot;
370 level = be16_to_cpu(block->bb_level);
371 ASSERT(level > 0);
372 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
373 bno = be64_to_cpu(*pp);
374 ASSERT(bno != NULLFSBLOCK);
375 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
376 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
377
378 if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
379 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
380 mp);
381 return -EFSCORRUPTED;
382 }
383
384 return 0;
385}
386
387/*
388 * returns 1 for success, 0 if we failed to map the extent.
389 */
390STATIC int
391xfs_getbmapx_fix_eof_hole(
392 xfs_inode_t *ip, /* xfs incore inode pointer */
393 int whichfork,
394 struct getbmapx *out, /* output structure */
395 int prealloced, /* this is a file with
396 * preallocated data space */
397 __int64_t end, /* last block requested */
398 xfs_fsblock_t startblock,
399 bool moretocome)
400{
401 __int64_t fixlen;
402 xfs_mount_t *mp; /* file system mount point */
403 xfs_ifork_t *ifp; /* inode fork pointer */
404 xfs_extnum_t lastx; /* last extent pointer */
405 xfs_fileoff_t fileblock;
406
407 if (startblock == HOLESTARTBLOCK) {
408 mp = ip->i_mount;
409 out->bmv_block = -1;
410 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
411 fixlen -= out->bmv_offset;
412 if (prealloced && out->bmv_offset + out->bmv_length == end) {
413 /* Came to hole at EOF. Trim it. */
414 if (fixlen <= 0)
415 return 0;
416 out->bmv_length = fixlen;
417 }
418 } else {
419 if (startblock == DELAYSTARTBLOCK)
420 out->bmv_block = -2;
421 else
422 out->bmv_block = xfs_fsb_to_db(ip, startblock);
423 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
424 ifp = XFS_IFORK_PTR(ip, whichfork);
425 if (!moretocome &&
426 xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
427 (lastx == xfs_iext_count(ifp) - 1))
428 out->bmv_oflags |= BMV_OF_LAST;
429 }
430
431 return 1;
432}
433
434/* Adjust the reported bmap around shared/unshared extent transitions. */
435STATIC int
436xfs_getbmap_adjust_shared(
437 struct xfs_inode *ip,
438 int whichfork,
439 struct xfs_bmbt_irec *map,
440 struct getbmapx *out,
441 struct xfs_bmbt_irec *next_map)
442{
443 struct xfs_mount *mp = ip->i_mount;
444 xfs_agnumber_t agno;
445 xfs_agblock_t agbno;
446 xfs_agblock_t ebno;
447 xfs_extlen_t elen;
448 xfs_extlen_t nlen;
449 int error;
450
451 next_map->br_startblock = NULLFSBLOCK;
452 next_map->br_startoff = NULLFILEOFF;
453 next_map->br_blockcount = 0;
454
455 /* Only written data blocks can be shared. */
456 if (!xfs_is_reflink_inode(ip) || whichfork != XFS_DATA_FORK ||
457 map->br_startblock == DELAYSTARTBLOCK ||
458 map->br_startblock == HOLESTARTBLOCK ||
459 ISUNWRITTEN(map))
460 return 0;
461
462 agno = XFS_FSB_TO_AGNO(mp, map->br_startblock);
463 agbno = XFS_FSB_TO_AGBNO(mp, map->br_startblock);
464 error = xfs_reflink_find_shared(mp, agno, agbno, map->br_blockcount,
465 &ebno, &elen, true);
466 if (error)
467 return error;
468
469 if (ebno == NULLAGBLOCK) {
470 /* No shared blocks at all. */
471 return 0;
472 } else if (agbno == ebno) {
473 /*
474 * Shared extent at (agbno, elen). Shrink the reported
475 * extent length and prepare to move the start of map[i]
476 * to agbno+elen, with the aim of (re)formatting the new
477 * map[i] the next time through the inner loop.
478 */
479 out->bmv_length = XFS_FSB_TO_BB(mp, elen);
480 out->bmv_oflags |= BMV_OF_SHARED;
481 if (elen != map->br_blockcount) {
482 *next_map = *map;
483 next_map->br_startblock += elen;
484 next_map->br_startoff += elen;
485 next_map->br_blockcount -= elen;
486 }
487 map->br_blockcount -= elen;
488 } else {
489 /*
490 * There's an unshared extent (agbno, ebno - agbno)
491 * followed by shared extent at (ebno, elen). Shrink
492 * the reported extent length to cover only the unshared
493 * extent and prepare to move up the start of map[i] to
494 * ebno, with the aim of (re)formatting the new map[i]
495 * the next time through the inner loop.
496 */
497 *next_map = *map;
498 nlen = ebno - agbno;
499 out->bmv_length = XFS_FSB_TO_BB(mp, nlen);
500 next_map->br_startblock += nlen;
501 next_map->br_startoff += nlen;
502 next_map->br_blockcount -= nlen;
503 map->br_blockcount -= nlen;
504 }
505
506 return 0;
507}
508
509/*
510 * Get inode's extents as described in bmv, and format for output.
511 * Calls formatter to fill the user's buffer until all extents
512 * are mapped, until the passed-in bmv->bmv_count slots have
513 * been filled, or until the formatter short-circuits the loop,
514 * if it is tracking filled-in extents on its own.
515 */
516int /* error code */
517xfs_getbmap(
518 xfs_inode_t *ip,
519 struct getbmapx *bmv, /* user bmap structure */
520 xfs_bmap_format_t formatter, /* format to user */
521 void *arg) /* formatter arg */
522{
523 __int64_t bmvend; /* last block requested */
524 int error = 0; /* return value */
525 __int64_t fixlen; /* length for -1 case */
526 int i; /* extent number */
527 int lock; /* lock state */
528 xfs_bmbt_irec_t *map; /* buffer for user's data */
529 xfs_mount_t *mp; /* file system mount point */
530 int nex; /* # of user extents can do */
531 int subnex; /* # of bmapi's can do */
532 int nmap; /* number of map entries */
533 struct getbmapx *out; /* output structure */
534 int whichfork; /* data or attr fork */
535 int prealloced; /* this is a file with
536 * preallocated data space */
537 int iflags; /* interface flags */
538 int bmapi_flags; /* flags for xfs_bmapi */
539 int cur_ext = 0;
540 struct xfs_bmbt_irec inject_map;
541
542 mp = ip->i_mount;
543 iflags = bmv->bmv_iflags;
544
545#ifndef DEBUG
546 /* Only allow CoW fork queries if we're debugging. */
547 if (iflags & BMV_IF_COWFORK)
548 return -EINVAL;
549#endif
550 if ((iflags & BMV_IF_ATTRFORK) && (iflags & BMV_IF_COWFORK))
551 return -EINVAL;
552
553 if (iflags & BMV_IF_ATTRFORK)
554 whichfork = XFS_ATTR_FORK;
555 else if (iflags & BMV_IF_COWFORK)
556 whichfork = XFS_COW_FORK;
557 else
558 whichfork = XFS_DATA_FORK;
559
560 switch (whichfork) {
561 case XFS_ATTR_FORK:
562 if (XFS_IFORK_Q(ip)) {
563 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
564 ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
565 ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
566 return -EINVAL;
567 } else if (unlikely(
568 ip->i_d.di_aformat != 0 &&
569 ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
570 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
571 ip->i_mount);
572 return -EFSCORRUPTED;
573 }
574
575 prealloced = 0;
576 fixlen = 1LL << 32;
577 break;
578 case XFS_COW_FORK:
579 if (ip->i_cformat != XFS_DINODE_FMT_EXTENTS)
580 return -EINVAL;
581
582 if (xfs_get_cowextsz_hint(ip)) {
583 prealloced = 1;
584 fixlen = mp->m_super->s_maxbytes;
585 } else {
586 prealloced = 0;
587 fixlen = XFS_ISIZE(ip);
588 }
589 break;
590 default:
591 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
592 ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
593 ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
594 return -EINVAL;
595
596 if (xfs_get_extsz_hint(ip) ||
597 ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
598 prealloced = 1;
599 fixlen = mp->m_super->s_maxbytes;
600 } else {
601 prealloced = 0;
602 fixlen = XFS_ISIZE(ip);
603 }
604 break;
605 }
606
607 if (bmv->bmv_length == -1) {
608 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
609 bmv->bmv_length =
610 max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
611 } else if (bmv->bmv_length == 0) {
612 bmv->bmv_entries = 0;
613 return 0;
614 } else if (bmv->bmv_length < 0) {
615 return -EINVAL;
616 }
617
618 nex = bmv->bmv_count - 1;
619 if (nex <= 0)
620 return -EINVAL;
621 bmvend = bmv->bmv_offset + bmv->bmv_length;
622
623
624 if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
625 return -ENOMEM;
626 out = kmem_zalloc_large(bmv->bmv_count * sizeof(struct getbmapx), 0);
627 if (!out)
628 return -ENOMEM;
629
630 xfs_ilock(ip, XFS_IOLOCK_SHARED);
631 switch (whichfork) {
632 case XFS_DATA_FORK:
633 if (!(iflags & BMV_IF_DELALLOC) &&
634 (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) {
635 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
636 if (error)
637 goto out_unlock_iolock;
638
639 /*
640 * Even after flushing the inode, there can still be
641 * delalloc blocks on the inode beyond EOF due to
642 * speculative preallocation. These are not removed
643 * until the release function is called or the inode
644 * is inactivated. Hence we cannot assert here that
645 * ip->i_delayed_blks == 0.
646 */
647 }
648
649 lock = xfs_ilock_data_map_shared(ip);
650 break;
651 case XFS_COW_FORK:
652 lock = XFS_ILOCK_SHARED;
653 xfs_ilock(ip, lock);
654 break;
655 case XFS_ATTR_FORK:
656 lock = xfs_ilock_attr_map_shared(ip);
657 break;
658 }
659
660 /*
661 * Don't let nex be bigger than the number of extents
662 * we can have assuming alternating holes and real extents.
663 */
664 if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
665 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
666
667 bmapi_flags = xfs_bmapi_aflag(whichfork);
668 if (!(iflags & BMV_IF_PREALLOC))
669 bmapi_flags |= XFS_BMAPI_IGSTATE;
670
671 /*
672 * Allocate enough space to handle "subnex" maps at a time.
673 */
674 error = -ENOMEM;
675 subnex = 16;
676 map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
677 if (!map)
678 goto out_unlock_ilock;
679
680 bmv->bmv_entries = 0;
681
682 if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
683 (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
684 error = 0;
685 goto out_free_map;
686 }
687
688 do {
689 nmap = (nex> subnex) ? subnex : nex;
690 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
691 XFS_BB_TO_FSB(mp, bmv->bmv_length),
692 map, &nmap, bmapi_flags);
693 if (error)
694 goto out_free_map;
695 ASSERT(nmap <= subnex);
696
697 for (i = 0; i < nmap && bmv->bmv_length &&
698 cur_ext < bmv->bmv_count - 1; i++) {
699 out[cur_ext].bmv_oflags = 0;
700 if (map[i].br_state == XFS_EXT_UNWRITTEN)
701 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
702 else if (map[i].br_startblock == DELAYSTARTBLOCK)
703 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
704 out[cur_ext].bmv_offset =
705 XFS_FSB_TO_BB(mp, map[i].br_startoff);
706 out[cur_ext].bmv_length =
707 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
708 out[cur_ext].bmv_unused1 = 0;
709 out[cur_ext].bmv_unused2 = 0;
710
711 /*
712 * delayed allocation extents that start beyond EOF can
713 * occur due to speculative EOF allocation when the
714 * delalloc extent is larger than the largest freespace
715 * extent at conversion time. These extents cannot be
716 * converted by data writeback, so can exist here even
717 * if we are not supposed to be finding delalloc
718 * extents.
719 */
720 if (map[i].br_startblock == DELAYSTARTBLOCK &&
721 map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip)))
722 ASSERT((iflags & BMV_IF_DELALLOC) != 0);
723
724 if (map[i].br_startblock == HOLESTARTBLOCK &&
725 whichfork == XFS_ATTR_FORK) {
726 /* came to the end of attribute fork */
727 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
728 goto out_free_map;
729 }
730
731 /* Is this a shared block? */
732 error = xfs_getbmap_adjust_shared(ip, whichfork,
733 &map[i], &out[cur_ext], &inject_map);
734 if (error)
735 goto out_free_map;
736
737 if (!xfs_getbmapx_fix_eof_hole(ip, whichfork,
738 &out[cur_ext], prealloced, bmvend,
739 map[i].br_startblock,
740 inject_map.br_startblock != NULLFSBLOCK))
741 goto out_free_map;
742
743 bmv->bmv_offset =
744 out[cur_ext].bmv_offset +
745 out[cur_ext].bmv_length;
746 bmv->bmv_length =
747 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
748
749 /*
750 * In case we don't want to return the hole,
751 * don't increase cur_ext so that we can reuse
752 * it in the next loop.
753 */
754 if ((iflags & BMV_IF_NO_HOLES) &&
755 map[i].br_startblock == HOLESTARTBLOCK) {
756 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
757 continue;
758 }
759
760 /*
761 * In order to report shared extents accurately,
762 * we report each distinct shared/unshared part
763 * of a single bmbt record using multiple bmap
764 * extents. To make that happen, we iterate the
765 * same map array item multiple times, each
766 * time trimming out the subextent that we just
767 * reported.
768 *
769 * Because of this, we must check the out array
770 * index (cur_ext) directly against bmv_count-1
771 * to avoid overflows.
772 */
773 if (inject_map.br_startblock != NULLFSBLOCK) {
774 map[i] = inject_map;
775 i--;
776 }
777 bmv->bmv_entries++;
778 cur_ext++;
779 }
780 } while (nmap && bmv->bmv_length && cur_ext < bmv->bmv_count - 1);
781
782 out_free_map:
783 kmem_free(map);
784 out_unlock_ilock:
785 xfs_iunlock(ip, lock);
786 out_unlock_iolock:
787 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
788
789 for (i = 0; i < cur_ext; i++) {
790 int full = 0; /* user array is full */
791
792 /* format results & advance arg */
793 error = formatter(&arg, &out[i], &full);
794 if (error || full)
795 break;
796 }
797
798 kmem_free(out);
799 return error;
800}
801
802/*
803 * dead simple method of punching delalyed allocation blocks from a range in
804 * the inode. Walks a block at a time so will be slow, but is only executed in
805 * rare error cases so the overhead is not critical. This will always punch out
806 * both the start and end blocks, even if the ranges only partially overlap
807 * them, so it is up to the caller to ensure that partial blocks are not
808 * passed in.
809 */
810int
811xfs_bmap_punch_delalloc_range(
812 struct xfs_inode *ip,
813 xfs_fileoff_t start_fsb,
814 xfs_fileoff_t length)
815{
816 xfs_fileoff_t remaining = length;
817 int error = 0;
818
819 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
820
821 do {
822 int done;
823 xfs_bmbt_irec_t imap;
824 int nimaps = 1;
825 xfs_fsblock_t firstblock;
826 struct xfs_defer_ops dfops;
827
828 /*
829 * Map the range first and check that it is a delalloc extent
830 * before trying to unmap the range. Otherwise we will be
831 * trying to remove a real extent (which requires a
832 * transaction) or a hole, which is probably a bad idea...
833 */
834 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
835 XFS_BMAPI_ENTIRE);
836
837 if (error) {
838 /* something screwed, just bail */
839 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
840 xfs_alert(ip->i_mount,
841 "Failed delalloc mapping lookup ino %lld fsb %lld.",
842 ip->i_ino, start_fsb);
843 }
844 break;
845 }
846 if (!nimaps) {
847 /* nothing there */
848 goto next_block;
849 }
850 if (imap.br_startblock != DELAYSTARTBLOCK) {
851 /* been converted, ignore */
852 goto next_block;
853 }
854 WARN_ON(imap.br_blockcount == 0);
855
856 /*
857 * Note: while we initialise the firstblock/dfops pair, they
858 * should never be used because blocks should never be
859 * allocated or freed for a delalloc extent and hence we need
860 * don't cancel or finish them after the xfs_bunmapi() call.
861 */
862 xfs_defer_init(&dfops, &firstblock);
863 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
864 &dfops, &done);
865 if (error)
866 break;
867
868 ASSERT(!xfs_defer_has_unfinished_work(&dfops));
869next_block:
870 start_fsb++;
871 remaining--;
872 } while(remaining > 0);
873
874 return error;
875}
876
877/*
878 * Test whether it is appropriate to check an inode for and free post EOF
879 * blocks. The 'force' parameter determines whether we should also consider
880 * regular files that are marked preallocated or append-only.
881 */
882bool
883xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
884{
885 /* prealloc/delalloc exists only on regular files */
886 if (!S_ISREG(VFS_I(ip)->i_mode))
887 return false;
888
889 /*
890 * Zero sized files with no cached pages and delalloc blocks will not
891 * have speculative prealloc/delalloc blocks to remove.
892 */
893 if (VFS_I(ip)->i_size == 0 &&
894 VFS_I(ip)->i_mapping->nrpages == 0 &&
895 ip->i_delayed_blks == 0)
896 return false;
897
898 /* If we haven't read in the extent list, then don't do it now. */
899 if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
900 return false;
901
902 /*
903 * Do not free real preallocated or append-only files unless the file
904 * has delalloc blocks and we are forced to remove them.
905 */
906 if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
907 if (!force || ip->i_delayed_blks == 0)
908 return false;
909
910 return true;
911}
912
913/*
914 * This is called by xfs_inactive to free any blocks beyond eof
915 * when the link count isn't zero and by xfs_dm_punch_hole() when
916 * punching a hole to EOF.
917 */
918int
919xfs_free_eofblocks(
920 struct xfs_inode *ip)
921{
922 struct xfs_trans *tp;
923 int error;
924 xfs_fileoff_t end_fsb;
925 xfs_fileoff_t last_fsb;
926 xfs_filblks_t map_len;
927 int nimaps;
928 struct xfs_bmbt_irec imap;
929 struct xfs_mount *mp = ip->i_mount;
930
931 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
932
933 /*
934 * Figure out if there are any blocks beyond the end
935 * of the file. If not, then there is nothing to do.
936 */
937 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
938 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
939 if (last_fsb <= end_fsb)
940 return 0;
941 map_len = last_fsb - end_fsb;
942
943 nimaps = 1;
944 xfs_ilock(ip, XFS_ILOCK_SHARED);
945 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
946 xfs_iunlock(ip, XFS_ILOCK_SHARED);
947
948 /*
949 * If there are blocks after the end of file, truncate the file to its
950 * current size to free them up.
951 */
952 if (!error && (nimaps != 0) &&
953 (imap.br_startblock != HOLESTARTBLOCK ||
954 ip->i_delayed_blks)) {
955 /*
956 * Attach the dquots to the inode up front.
957 */
958 error = xfs_qm_dqattach(ip, 0);
959 if (error)
960 return error;
961
962 /* wait on dio to ensure i_size has settled */
963 inode_dio_wait(VFS_I(ip));
964
965 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0,
966 &tp);
967 if (error) {
968 ASSERT(XFS_FORCED_SHUTDOWN(mp));
969 return error;
970 }
971
972 xfs_ilock(ip, XFS_ILOCK_EXCL);
973 xfs_trans_ijoin(tp, ip, 0);
974
975 /*
976 * Do not update the on-disk file size. If we update the
977 * on-disk file size and then the system crashes before the
978 * contents of the file are flushed to disk then the files
979 * may be full of holes (ie NULL files bug).
980 */
981 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK,
982 XFS_ISIZE(ip));
983 if (error) {
984 /*
985 * If we get an error at this point we simply don't
986 * bother truncating the file.
987 */
988 xfs_trans_cancel(tp);
989 } else {
990 error = xfs_trans_commit(tp);
991 if (!error)
992 xfs_inode_clear_eofblocks_tag(ip);
993 }
994
995 xfs_iunlock(ip, XFS_ILOCK_EXCL);
996 }
997 return error;
998}
999
1000int
1001xfs_alloc_file_space(
1002 struct xfs_inode *ip,
1003 xfs_off_t offset,
1004 xfs_off_t len,
1005 int alloc_type)
1006{
1007 xfs_mount_t *mp = ip->i_mount;
1008 xfs_off_t count;
1009 xfs_filblks_t allocated_fsb;
1010 xfs_filblks_t allocatesize_fsb;
1011 xfs_extlen_t extsz, temp;
1012 xfs_fileoff_t startoffset_fsb;
1013 xfs_fsblock_t firstfsb;
1014 int nimaps;
1015 int quota_flag;
1016 int rt;
1017 xfs_trans_t *tp;
1018 xfs_bmbt_irec_t imaps[1], *imapp;
1019 struct xfs_defer_ops dfops;
1020 uint qblocks, resblks, resrtextents;
1021 int error;
1022
1023 trace_xfs_alloc_file_space(ip);
1024
1025 if (XFS_FORCED_SHUTDOWN(mp))
1026 return -EIO;
1027
1028 error = xfs_qm_dqattach(ip, 0);
1029 if (error)
1030 return error;
1031
1032 if (len <= 0)
1033 return -EINVAL;
1034
1035 rt = XFS_IS_REALTIME_INODE(ip);
1036 extsz = xfs_get_extsz_hint(ip);
1037
1038 count = len;
1039 imapp = &imaps[0];
1040 nimaps = 1;
1041 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
1042 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
1043
1044 /*
1045 * Allocate file space until done or until there is an error
1046 */
1047 while (allocatesize_fsb && !error) {
1048 xfs_fileoff_t s, e;
1049
1050 /*
1051 * Determine space reservations for data/realtime.
1052 */
1053 if (unlikely(extsz)) {
1054 s = startoffset_fsb;
1055 do_div(s, extsz);
1056 s *= extsz;
1057 e = startoffset_fsb + allocatesize_fsb;
1058 if ((temp = do_mod(startoffset_fsb, extsz)))
1059 e += temp;
1060 if ((temp = do_mod(e, extsz)))
1061 e += extsz - temp;
1062 } else {
1063 s = 0;
1064 e = allocatesize_fsb;
1065 }
1066
1067 /*
1068 * The transaction reservation is limited to a 32-bit block
1069 * count, hence we need to limit the number of blocks we are
1070 * trying to reserve to avoid an overflow. We can't allocate
1071 * more than @nimaps extents, and an extent is limited on disk
1072 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
1073 */
1074 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
1075 if (unlikely(rt)) {
1076 resrtextents = qblocks = resblks;
1077 resrtextents /= mp->m_sb.sb_rextsize;
1078 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1079 quota_flag = XFS_QMOPT_RES_RTBLKS;
1080 } else {
1081 resrtextents = 0;
1082 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
1083 quota_flag = XFS_QMOPT_RES_REGBLKS;
1084 }
1085
1086 /*
1087 * Allocate and setup the transaction.
1088 */
1089 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks,
1090 resrtextents, 0, &tp);
1091
1092 /*
1093 * Check for running out of space
1094 */
1095 if (error) {
1096 /*
1097 * Free the transaction structure.
1098 */
1099 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1100 break;
1101 }
1102 xfs_ilock(ip, XFS_ILOCK_EXCL);
1103 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
1104 0, quota_flag);
1105 if (error)
1106 goto error1;
1107
1108 xfs_trans_ijoin(tp, ip, 0);
1109
1110 xfs_defer_init(&dfops, &firstfsb);
1111 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
1112 allocatesize_fsb, alloc_type, &firstfsb,
1113 resblks, imapp, &nimaps, &dfops);
1114 if (error)
1115 goto error0;
1116
1117 /*
1118 * Complete the transaction
1119 */
1120 error = xfs_defer_finish(&tp, &dfops, NULL);
1121 if (error)
1122 goto error0;
1123
1124 error = xfs_trans_commit(tp);
1125 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1126 if (error)
1127 break;
1128
1129 allocated_fsb = imapp->br_blockcount;
1130
1131 if (nimaps == 0) {
1132 error = -ENOSPC;
1133 break;
1134 }
1135
1136 startoffset_fsb += allocated_fsb;
1137 allocatesize_fsb -= allocated_fsb;
1138 }
1139
1140 return error;
1141
1142error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1143 xfs_defer_cancel(&dfops);
1144 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
1145
1146error1: /* Just cancel transaction */
1147 xfs_trans_cancel(tp);
1148 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1149 return error;
1150}
1151
1152static int
1153xfs_unmap_extent(
1154 struct xfs_inode *ip,
1155 xfs_fileoff_t startoffset_fsb,
1156 xfs_filblks_t len_fsb,
1157 int *done)
1158{
1159 struct xfs_mount *mp = ip->i_mount;
1160 struct xfs_trans *tp;
1161 struct xfs_defer_ops dfops;
1162 xfs_fsblock_t firstfsb;
1163 uint resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1164 int error;
1165
1166 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
1167 if (error) {
1168 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1169 return error;
1170 }
1171
1172 xfs_ilock(ip, XFS_ILOCK_EXCL);
1173 error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot, ip->i_gdquot,
1174 ip->i_pdquot, resblks, 0, XFS_QMOPT_RES_REGBLKS);
1175 if (error)
1176 goto out_trans_cancel;
1177
1178 xfs_trans_ijoin(tp, ip, 0);
1179
1180 xfs_defer_init(&dfops, &firstfsb);
1181 error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, &firstfsb,
1182 &dfops, done);
1183 if (error)
1184 goto out_bmap_cancel;
1185
1186 error = xfs_defer_finish(&tp, &dfops, ip);
1187 if (error)
1188 goto out_bmap_cancel;
1189
1190 error = xfs_trans_commit(tp);
1191out_unlock:
1192 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1193 return error;
1194
1195out_bmap_cancel:
1196 xfs_defer_cancel(&dfops);
1197out_trans_cancel:
1198 xfs_trans_cancel(tp);
1199 goto out_unlock;
1200}
1201
1202static int
1203xfs_adjust_extent_unmap_boundaries(
1204 struct xfs_inode *ip,
1205 xfs_fileoff_t *startoffset_fsb,
1206 xfs_fileoff_t *endoffset_fsb)
1207{
1208 struct xfs_mount *mp = ip->i_mount;
1209 struct xfs_bmbt_irec imap;
1210 int nimap, error;
1211 xfs_extlen_t mod = 0;
1212
1213 nimap = 1;
1214 error = xfs_bmapi_read(ip, *startoffset_fsb, 1, &imap, &nimap, 0);
1215 if (error)
1216 return error;
1217
1218 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1219 xfs_daddr_t block;
1220
1221 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1222 block = imap.br_startblock;
1223 mod = do_div(block, mp->m_sb.sb_rextsize);
1224 if (mod)
1225 *startoffset_fsb += mp->m_sb.sb_rextsize - mod;
1226 }
1227
1228 nimap = 1;
1229 error = xfs_bmapi_read(ip, *endoffset_fsb - 1, 1, &imap, &nimap, 0);
1230 if (error)
1231 return error;
1232
1233 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1234 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1235 mod++;
1236 if (mod && mod != mp->m_sb.sb_rextsize)
1237 *endoffset_fsb -= mod;
1238 }
1239
1240 return 0;
1241}
1242
1243static int
1244xfs_flush_unmap_range(
1245 struct xfs_inode *ip,
1246 xfs_off_t offset,
1247 xfs_off_t len)
1248{
1249 struct xfs_mount *mp = ip->i_mount;
1250 struct inode *inode = VFS_I(ip);
1251 xfs_off_t rounding, start, end;
1252 int error;
1253
1254 /* wait for the completion of any pending DIOs */
1255 inode_dio_wait(inode);
1256
1257 rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_SIZE);
1258 start = round_down(offset, rounding);
1259 end = round_up(offset + len, rounding) - 1;
1260
1261 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
1262 if (error)
1263 return error;
1264 truncate_pagecache_range(inode, start, end);
1265 return 0;
1266}
1267
1268int
1269xfs_free_file_space(
1270 struct xfs_inode *ip,
1271 xfs_off_t offset,
1272 xfs_off_t len)
1273{
1274 struct xfs_mount *mp = ip->i_mount;
1275 xfs_fileoff_t startoffset_fsb;
1276 xfs_fileoff_t endoffset_fsb;
1277 int done = 0, error;
1278
1279 trace_xfs_free_file_space(ip);
1280
1281 error = xfs_qm_dqattach(ip, 0);
1282 if (error)
1283 return error;
1284
1285 if (len <= 0) /* if nothing being freed */
1286 return 0;
1287
1288 error = xfs_flush_unmap_range(ip, offset, len);
1289 if (error)
1290 return error;
1291
1292 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
1293 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
1294
1295 /*
1296 * Need to zero the stuff we're not freeing, on disk. If it's a RT file
1297 * and we can't use unwritten extents then we actually need to ensure
1298 * to zero the whole extent, otherwise we just need to take of block
1299 * boundaries, and xfs_bunmapi will handle the rest.
1300 */
1301 if (XFS_IS_REALTIME_INODE(ip) &&
1302 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
1303 error = xfs_adjust_extent_unmap_boundaries(ip, &startoffset_fsb,
1304 &endoffset_fsb);
1305 if (error)
1306 return error;
1307 }
1308
1309 if (endoffset_fsb > startoffset_fsb) {
1310 while (!done) {
1311 error = xfs_unmap_extent(ip, startoffset_fsb,
1312 endoffset_fsb - startoffset_fsb, &done);
1313 if (error)
1314 return error;
1315 }
1316 }
1317
1318 /*
1319 * Now that we've unmap all full blocks we'll have to zero out any
1320 * partial block at the beginning and/or end. xfs_zero_range is
1321 * smart enough to skip any holes, including those we just created,
1322 * but we must take care not to zero beyond EOF and enlarge i_size.
1323 */
1324
1325 if (offset >= XFS_ISIZE(ip))
1326 return 0;
1327
1328 if (offset + len > XFS_ISIZE(ip))
1329 len = XFS_ISIZE(ip) - offset;
1330
1331 return xfs_zero_range(ip, offset, len, NULL);
1332}
1333
1334/*
1335 * Preallocate and zero a range of a file. This mechanism has the allocation
1336 * semantics of fallocate and in addition converts data in the range to zeroes.
1337 */
1338int
1339xfs_zero_file_space(
1340 struct xfs_inode *ip,
1341 xfs_off_t offset,
1342 xfs_off_t len)
1343{
1344 struct xfs_mount *mp = ip->i_mount;
1345 uint blksize;
1346 int error;
1347
1348 trace_xfs_zero_file_space(ip);
1349
1350 blksize = 1 << mp->m_sb.sb_blocklog;
1351
1352 /*
1353 * Punch a hole and prealloc the range. We use hole punch rather than
1354 * unwritten extent conversion for two reasons:
1355 *
1356 * 1.) Hole punch handles partial block zeroing for us.
1357 *
1358 * 2.) If prealloc returns ENOSPC, the file range is still zero-valued
1359 * by virtue of the hole punch.
1360 */
1361 error = xfs_free_file_space(ip, offset, len);
1362 if (error)
1363 goto out;
1364
1365 error = xfs_alloc_file_space(ip, round_down(offset, blksize),
1366 round_up(offset + len, blksize) -
1367 round_down(offset, blksize),
1368 XFS_BMAPI_PREALLOC);
1369out:
1370 return error;
1371
1372}
1373
1374/*
1375 * @next_fsb will keep track of the extent currently undergoing shift.
1376 * @stop_fsb will keep track of the extent at which we have to stop.
1377 * If we are shifting left, we will start with block (offset + len) and
1378 * shift each extent till last extent.
1379 * If we are shifting right, we will start with last extent inside file space
1380 * and continue until we reach the block corresponding to offset.
1381 */
1382static int
1383xfs_shift_file_space(
1384 struct xfs_inode *ip,
1385 xfs_off_t offset,
1386 xfs_off_t len,
1387 enum shift_direction direction)
1388{
1389 int done = 0;
1390 struct xfs_mount *mp = ip->i_mount;
1391 struct xfs_trans *tp;
1392 int error;
1393 struct xfs_defer_ops dfops;
1394 xfs_fsblock_t first_block;
1395 xfs_fileoff_t stop_fsb;
1396 xfs_fileoff_t next_fsb;
1397 xfs_fileoff_t shift_fsb;
1398 uint resblks;
1399
1400 ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT);
1401
1402 if (direction == SHIFT_LEFT) {
1403 /*
1404 * Reserve blocks to cover potential extent merges after left
1405 * shift operations.
1406 */
1407 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1408 next_fsb = XFS_B_TO_FSB(mp, offset + len);
1409 stop_fsb = XFS_B_TO_FSB(mp, VFS_I(ip)->i_size);
1410 } else {
1411 /*
1412 * If right shift, delegate the work of initialization of
1413 * next_fsb to xfs_bmap_shift_extent as it has ilock held.
1414 */
1415 resblks = 0;
1416 next_fsb = NULLFSBLOCK;
1417 stop_fsb = XFS_B_TO_FSB(mp, offset);
1418 }
1419
1420 shift_fsb = XFS_B_TO_FSB(mp, len);
1421
1422 /*
1423 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
1424 * into the accessible region of the file.
1425 */
1426 if (xfs_can_free_eofblocks(ip, true)) {
1427 error = xfs_free_eofblocks(ip);
1428 if (error)
1429 return error;
1430 }
1431
1432 /*
1433 * Writeback and invalidate cache for the remainder of the file as we're
1434 * about to shift down every extent from offset to EOF.
1435 */
1436 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1437 offset, -1);
1438 if (error)
1439 return error;
1440 error = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
1441 offset >> PAGE_SHIFT, -1);
1442 if (error)
1443 return error;
1444
1445 /*
1446 * The extent shiting code works on extent granularity. So, if
1447 * stop_fsb is not the starting block of extent, we need to split
1448 * the extent at stop_fsb.
1449 */
1450 if (direction == SHIFT_RIGHT) {
1451 error = xfs_bmap_split_extent(ip, stop_fsb);
1452 if (error)
1453 return error;
1454 }
1455
1456 while (!error && !done) {
1457 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0,
1458 &tp);
1459 if (error)
1460 break;
1461
1462 xfs_ilock(ip, XFS_ILOCK_EXCL);
1463 error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot,
1464 ip->i_gdquot, ip->i_pdquot, resblks, 0,
1465 XFS_QMOPT_RES_REGBLKS);
1466 if (error)
1467 goto out_trans_cancel;
1468
1469 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1470
1471 xfs_defer_init(&dfops, &first_block);
1472
1473 /*
1474 * We are using the write transaction in which max 2 bmbt
1475 * updates are allowed
1476 */
1477 error = xfs_bmap_shift_extents(tp, ip, &next_fsb, shift_fsb,
1478 &done, stop_fsb, &first_block, &dfops,
1479 direction, XFS_BMAP_MAX_SHIFT_EXTENTS);
1480 if (error)
1481 goto out_bmap_cancel;
1482
1483 error = xfs_defer_finish(&tp, &dfops, NULL);
1484 if (error)
1485 goto out_bmap_cancel;
1486
1487 error = xfs_trans_commit(tp);
1488 }
1489
1490 return error;
1491
1492out_bmap_cancel:
1493 xfs_defer_cancel(&dfops);
1494out_trans_cancel:
1495 xfs_trans_cancel(tp);
1496 return error;
1497}
1498
1499/*
1500 * xfs_collapse_file_space()
1501 * This routine frees disk space and shift extent for the given file.
1502 * The first thing we do is to free data blocks in the specified range
1503 * by calling xfs_free_file_space(). It would also sync dirty data
1504 * and invalidate page cache over the region on which collapse range
1505 * is working. And Shift extent records to the left to cover a hole.
1506 * RETURNS:
1507 * 0 on success
1508 * errno on error
1509 *
1510 */
1511int
1512xfs_collapse_file_space(
1513 struct xfs_inode *ip,
1514 xfs_off_t offset,
1515 xfs_off_t len)
1516{
1517 int error;
1518
1519 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1520 trace_xfs_collapse_file_space(ip);
1521
1522 error = xfs_free_file_space(ip, offset, len);
1523 if (error)
1524 return error;
1525
1526 return xfs_shift_file_space(ip, offset, len, SHIFT_LEFT);
1527}
1528
1529/*
1530 * xfs_insert_file_space()
1531 * This routine create hole space by shifting extents for the given file.
1532 * The first thing we do is to sync dirty data and invalidate page cache
1533 * over the region on which insert range is working. And split an extent
1534 * to two extents at given offset by calling xfs_bmap_split_extent.
1535 * And shift all extent records which are laying between [offset,
1536 * last allocated extent] to the right to reserve hole range.
1537 * RETURNS:
1538 * 0 on success
1539 * errno on error
1540 */
1541int
1542xfs_insert_file_space(
1543 struct xfs_inode *ip,
1544 loff_t offset,
1545 loff_t len)
1546{
1547 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1548 trace_xfs_insert_file_space(ip);
1549
1550 return xfs_shift_file_space(ip, offset, len, SHIFT_RIGHT);
1551}
1552
1553/*
1554 * We need to check that the format of the data fork in the temporary inode is
1555 * valid for the target inode before doing the swap. This is not a problem with
1556 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1557 * data fork depending on the space the attribute fork is taking so we can get
1558 * invalid formats on the target inode.
1559 *
1560 * E.g. target has space for 7 extents in extent format, temp inode only has
1561 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1562 * btree, but when swapped it needs to be in extent format. Hence we can't just
1563 * blindly swap data forks on attr2 filesystems.
1564 *
1565 * Note that we check the swap in both directions so that we don't end up with
1566 * a corrupt temporary inode, either.
1567 *
1568 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1569 * inode will prevent this situation from occurring, so all we do here is
1570 * reject and log the attempt. basically we are putting the responsibility on
1571 * userspace to get this right.
1572 */
1573static int
1574xfs_swap_extents_check_format(
1575 struct xfs_inode *ip, /* target inode */
1576 struct xfs_inode *tip) /* tmp inode */
1577{
1578
1579 /* Should never get a local format */
1580 if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
1581 tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
1582 return -EINVAL;
1583
1584 /*
1585 * if the target inode has less extents that then temporary inode then
1586 * why did userspace call us?
1587 */
1588 if (ip->i_d.di_nextents < tip->i_d.di_nextents)
1589 return -EINVAL;
1590
1591 /*
1592 * If we have to use the (expensive) rmap swap method, we can
1593 * handle any number of extents and any format.
1594 */
1595 if (xfs_sb_version_hasrmapbt(&ip->i_mount->m_sb))
1596 return 0;
1597
1598 /*
1599 * if the target inode is in extent form and the temp inode is in btree
1600 * form then we will end up with the target inode in the wrong format
1601 * as we already know there are less extents in the temp inode.
1602 */
1603 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1604 tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1605 return -EINVAL;
1606
1607 /* Check temp in extent form to max in target */
1608 if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1609 XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) >
1610 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1611 return -EINVAL;
1612
1613 /* Check target in extent form to max in temp */
1614 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1615 XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) >
1616 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1617 return -EINVAL;
1618
1619 /*
1620 * If we are in a btree format, check that the temp root block will fit
1621 * in the target and that it has enough extents to be in btree format
1622 * in the target.
1623 *
1624 * Note that we have to be careful to allow btree->extent conversions
1625 * (a common defrag case) which will occur when the temp inode is in
1626 * extent format...
1627 */
1628 if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1629 if (XFS_IFORK_BOFF(ip) &&
1630 XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip))
1631 return -EINVAL;
1632 if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <=
1633 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1634 return -EINVAL;
1635 }
1636
1637 /* Reciprocal target->temp btree format checks */
1638 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1639 if (XFS_IFORK_BOFF(tip) &&
1640 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
1641 return -EINVAL;
1642 if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <=
1643 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1644 return -EINVAL;
1645 }
1646
1647 return 0;
1648}
1649
1650static int
1651xfs_swap_extent_flush(
1652 struct xfs_inode *ip)
1653{
1654 int error;
1655
1656 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1657 if (error)
1658 return error;
1659 truncate_pagecache_range(VFS_I(ip), 0, -1);
1660
1661 /* Verify O_DIRECT for ftmp */
1662 if (VFS_I(ip)->i_mapping->nrpages)
1663 return -EINVAL;
1664 return 0;
1665}
1666
1667/*
1668 * Move extents from one file to another, when rmap is enabled.
1669 */
1670STATIC int
1671xfs_swap_extent_rmap(
1672 struct xfs_trans **tpp,
1673 struct xfs_inode *ip,
1674 struct xfs_inode *tip)
1675{
1676 struct xfs_bmbt_irec irec;
1677 struct xfs_bmbt_irec uirec;
1678 struct xfs_bmbt_irec tirec;
1679 xfs_fileoff_t offset_fsb;
1680 xfs_fileoff_t end_fsb;
1681 xfs_filblks_t count_fsb;
1682 xfs_fsblock_t firstfsb;
1683 struct xfs_defer_ops dfops;
1684 int error;
1685 xfs_filblks_t ilen;
1686 xfs_filblks_t rlen;
1687 int nimaps;
1688 __uint64_t tip_flags2;
1689
1690 /*
1691 * If the source file has shared blocks, we must flag the donor
1692 * file as having shared blocks so that we get the shared-block
1693 * rmap functions when we go to fix up the rmaps. The flags
1694 * will be switch for reals later.
1695 */
1696 tip_flags2 = tip->i_d.di_flags2;
1697 if (ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)
1698 tip->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK;
1699
1700 offset_fsb = 0;
1701 end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip)));
1702 count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
1703
1704 while (count_fsb) {
1705 /* Read extent from the donor file */
1706 nimaps = 1;
1707 error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec,
1708 &nimaps, 0);
1709 if (error)
1710 goto out;
1711 ASSERT(nimaps == 1);
1712 ASSERT(tirec.br_startblock != DELAYSTARTBLOCK);
1713
1714 trace_xfs_swap_extent_rmap_remap(tip, &tirec);
1715 ilen = tirec.br_blockcount;
1716
1717 /* Unmap the old blocks in the source file. */
1718 while (tirec.br_blockcount) {
1719 xfs_defer_init(&dfops, &firstfsb);
1720 trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec);
1721
1722 /* Read extent from the source file */
1723 nimaps = 1;
1724 error = xfs_bmapi_read(ip, tirec.br_startoff,
1725 tirec.br_blockcount, &irec,
1726 &nimaps, 0);
1727 if (error)
1728 goto out_defer;
1729 ASSERT(nimaps == 1);
1730 ASSERT(tirec.br_startoff == irec.br_startoff);
1731 trace_xfs_swap_extent_rmap_remap_piece(ip, &irec);
1732
1733 /* Trim the extent. */
1734 uirec = tirec;
1735 uirec.br_blockcount = rlen = min_t(xfs_filblks_t,
1736 tirec.br_blockcount,
1737 irec.br_blockcount);
1738 trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec);
1739
1740 /* Remove the mapping from the donor file. */
1741 error = xfs_bmap_unmap_extent((*tpp)->t_mountp, &dfops,
1742 tip, &uirec);
1743 if (error)
1744 goto out_defer;
1745
1746 /* Remove the mapping from the source file. */
1747 error = xfs_bmap_unmap_extent((*tpp)->t_mountp, &dfops,
1748 ip, &irec);
1749 if (error)
1750 goto out_defer;
1751
1752 /* Map the donor file's blocks into the source file. */
1753 error = xfs_bmap_map_extent((*tpp)->t_mountp, &dfops,
1754 ip, &uirec);
1755 if (error)
1756 goto out_defer;
1757
1758 /* Map the source file's blocks into the donor file. */
1759 error = xfs_bmap_map_extent((*tpp)->t_mountp, &dfops,
1760 tip, &irec);
1761 if (error)
1762 goto out_defer;
1763
1764 error = xfs_defer_finish(tpp, &dfops, ip);
1765 if (error)
1766 goto out_defer;
1767
1768 tirec.br_startoff += rlen;
1769 if (tirec.br_startblock != HOLESTARTBLOCK &&
1770 tirec.br_startblock != DELAYSTARTBLOCK)
1771 tirec.br_startblock += rlen;
1772 tirec.br_blockcount -= rlen;
1773 }
1774
1775 /* Roll on... */
1776 count_fsb -= ilen;
1777 offset_fsb += ilen;
1778 }
1779
1780 tip->i_d.di_flags2 = tip_flags2;
1781 return 0;
1782
1783out_defer:
1784 xfs_defer_cancel(&dfops);
1785out:
1786 trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_);
1787 tip->i_d.di_flags2 = tip_flags2;
1788 return error;
1789}
1790
1791/* Swap the extents of two files by swapping data forks. */
1792STATIC int
1793xfs_swap_extent_forks(
1794 struct xfs_trans *tp,
1795 struct xfs_inode *ip,
1796 struct xfs_inode *tip,
1797 int *src_log_flags,
1798 int *target_log_flags)
1799{
1800 struct xfs_ifork tempifp, *ifp, *tifp;
1801 int aforkblks = 0;
1802 int taforkblks = 0;
1803 xfs_extnum_t nextents;
1804 __uint64_t tmp;
1805 int error;
1806
1807 /*
1808 * Count the number of extended attribute blocks
1809 */
1810 if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
1811 (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1812 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK,
1813 &aforkblks);
1814 if (error)
1815 return error;
1816 }
1817 if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
1818 (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1819 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
1820 &taforkblks);
1821 if (error)
1822 return error;
1823 }
1824
1825 /*
1826 * Before we've swapped the forks, lets set the owners of the forks
1827 * appropriately. We have to do this as we are demand paging the btree
1828 * buffers, and so the validation done on read will expect the owner
1829 * field to be correctly set. Once we change the owners, we can swap the
1830 * inode forks.
1831 */
1832 if (ip->i_d.di_version == 3 &&
1833 ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1834 (*target_log_flags) |= XFS_ILOG_DOWNER;
1835 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK,
1836 tip->i_ino, NULL);
1837 if (error)
1838 return error;
1839 }
1840
1841 if (tip->i_d.di_version == 3 &&
1842 tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1843 (*src_log_flags) |= XFS_ILOG_DOWNER;
1844 error = xfs_bmbt_change_owner(tp, tip, XFS_DATA_FORK,
1845 ip->i_ino, NULL);
1846 if (error)
1847 return error;
1848 }
1849
1850 /*
1851 * Swap the data forks of the inodes
1852 */
1853 ifp = &ip->i_df;
1854 tifp = &tip->i_df;
1855 tempifp = *ifp; /* struct copy */
1856 *ifp = *tifp; /* struct copy */
1857 *tifp = tempifp; /* struct copy */
1858
1859 /*
1860 * Fix the on-disk inode values
1861 */
1862 tmp = (__uint64_t)ip->i_d.di_nblocks;
1863 ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
1864 tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
1865
1866 tmp = (__uint64_t) ip->i_d.di_nextents;
1867 ip->i_d.di_nextents = tip->i_d.di_nextents;
1868 tip->i_d.di_nextents = tmp;
1869
1870 tmp = (__uint64_t) ip->i_d.di_format;
1871 ip->i_d.di_format = tip->i_d.di_format;
1872 tip->i_d.di_format = tmp;
1873
1874 /*
1875 * The extents in the source inode could still contain speculative
1876 * preallocation beyond EOF (e.g. the file is open but not modified
1877 * while defrag is in progress). In that case, we need to copy over the
1878 * number of delalloc blocks the data fork in the source inode is
1879 * tracking beyond EOF so that when the fork is truncated away when the
1880 * temporary inode is unlinked we don't underrun the i_delayed_blks
1881 * counter on that inode.
1882 */
1883 ASSERT(tip->i_delayed_blks == 0);
1884 tip->i_delayed_blks = ip->i_delayed_blks;
1885 ip->i_delayed_blks = 0;
1886
1887 switch (ip->i_d.di_format) {
1888 case XFS_DINODE_FMT_EXTENTS:
1889 /*
1890 * If the extents fit in the inode, fix the pointer. Otherwise
1891 * it's already NULL or pointing to the extent.
1892 */
1893 nextents = xfs_iext_count(&ip->i_df);
1894 if (nextents <= XFS_INLINE_EXTS)
1895 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
1896 (*src_log_flags) |= XFS_ILOG_DEXT;
1897 break;
1898 case XFS_DINODE_FMT_BTREE:
1899 ASSERT(ip->i_d.di_version < 3 ||
1900 (*src_log_flags & XFS_ILOG_DOWNER));
1901 (*src_log_flags) |= XFS_ILOG_DBROOT;
1902 break;
1903 }
1904
1905 switch (tip->i_d.di_format) {
1906 case XFS_DINODE_FMT_EXTENTS:
1907 /*
1908 * If the extents fit in the inode, fix the pointer. Otherwise
1909 * it's already NULL or pointing to the extent.
1910 */
1911 nextents = xfs_iext_count(&tip->i_df);
1912 if (nextents <= XFS_INLINE_EXTS)
1913 tifp->if_u1.if_extents = tifp->if_u2.if_inline_ext;
1914 (*target_log_flags) |= XFS_ILOG_DEXT;
1915 break;
1916 case XFS_DINODE_FMT_BTREE:
1917 (*target_log_flags) |= XFS_ILOG_DBROOT;
1918 ASSERT(tip->i_d.di_version < 3 ||
1919 (*target_log_flags & XFS_ILOG_DOWNER));
1920 break;
1921 }
1922
1923 return 0;
1924}
1925
1926int
1927xfs_swap_extents(
1928 struct xfs_inode *ip, /* target inode */
1929 struct xfs_inode *tip, /* tmp inode */
1930 struct xfs_swapext *sxp)
1931{
1932 struct xfs_mount *mp = ip->i_mount;
1933 struct xfs_trans *tp;
1934 struct xfs_bstat *sbp = &sxp->sx_stat;
1935 int src_log_flags, target_log_flags;
1936 int error = 0;
1937 int lock_flags;
1938 struct xfs_ifork *cowfp;
1939 __uint64_t f;
1940 int resblks;
1941
1942 /*
1943 * Lock the inodes against other IO, page faults and truncate to
1944 * begin with. Then we can ensure the inodes are flushed and have no
1945 * page cache safely. Once we have done this we can take the ilocks and
1946 * do the rest of the checks.
1947 */
1948 lock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1949 lock_flags = XFS_MMAPLOCK_EXCL;
1950 xfs_lock_two_inodes(ip, tip, XFS_MMAPLOCK_EXCL);
1951
1952 /* Verify that both files have the same format */
1953 if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) {
1954 error = -EINVAL;
1955 goto out_unlock;
1956 }
1957
1958 /* Verify both files are either real-time or non-realtime */
1959 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1960 error = -EINVAL;
1961 goto out_unlock;
1962 }
1963
1964 error = xfs_swap_extent_flush(ip);
1965 if (error)
1966 goto out_unlock;
1967 error = xfs_swap_extent_flush(tip);
1968 if (error)
1969 goto out_unlock;
1970
1971 /*
1972 * Extent "swapping" with rmap requires a permanent reservation and
1973 * a block reservation because it's really just a remap operation
1974 * performed with log redo items!
1975 */
1976 if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
1977 /*
1978 * Conceptually this shouldn't affect the shape of either
1979 * bmbt, but since we atomically move extents one by one,
1980 * we reserve enough space to rebuild both trees.
1981 */
1982 resblks = XFS_SWAP_RMAP_SPACE_RES(mp,
1983 XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK),
1984 XFS_DATA_FORK) +
1985 XFS_SWAP_RMAP_SPACE_RES(mp,
1986 XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK),
1987 XFS_DATA_FORK);
1988 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks,
1989 0, 0, &tp);
1990 } else
1991 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0,
1992 0, 0, &tp);
1993 if (error)
1994 goto out_unlock;
1995
1996 /*
1997 * Lock and join the inodes to the tansaction so that transaction commit
1998 * or cancel will unlock the inodes from this point onwards.
1999 */
2000 xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
2001 lock_flags |= XFS_ILOCK_EXCL;
2002 xfs_trans_ijoin(tp, ip, 0);
2003 xfs_trans_ijoin(tp, tip, 0);
2004
2005
2006 /* Verify all data are being swapped */
2007 if (sxp->sx_offset != 0 ||
2008 sxp->sx_length != ip->i_d.di_size ||
2009 sxp->sx_length != tip->i_d.di_size) {
2010 error = -EFAULT;
2011 goto out_trans_cancel;
2012 }
2013
2014 trace_xfs_swap_extent_before(ip, 0);
2015 trace_xfs_swap_extent_before(tip, 1);
2016
2017 /* check inode formats now that data is flushed */
2018 error = xfs_swap_extents_check_format(ip, tip);
2019 if (error) {
2020 xfs_notice(mp,
2021 "%s: inode 0x%llx format is incompatible for exchanging.",
2022 __func__, ip->i_ino);
2023 goto out_trans_cancel;
2024 }
2025
2026 /*
2027 * Compare the current change & modify times with that
2028 * passed in. If they differ, we abort this swap.
2029 * This is the mechanism used to ensure the calling
2030 * process that the file was not changed out from
2031 * under it.
2032 */
2033 if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
2034 (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
2035 (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
2036 (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
2037 error = -EBUSY;
2038 goto out_trans_cancel;
2039 }
2040
2041 /*
2042 * Note the trickiness in setting the log flags - we set the owner log
2043 * flag on the opposite inode (i.e. the inode we are setting the new
2044 * owner to be) because once we swap the forks and log that, log
2045 * recovery is going to see the fork as owned by the swapped inode,
2046 * not the pre-swapped inodes.
2047 */
2048 src_log_flags = XFS_ILOG_CORE;
2049 target_log_flags = XFS_ILOG_CORE;
2050
2051 if (xfs_sb_version_hasrmapbt(&mp->m_sb))
2052 error = xfs_swap_extent_rmap(&tp, ip, tip);
2053 else
2054 error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags,
2055 &target_log_flags);
2056 if (error)
2057 goto out_trans_cancel;
2058
2059 /* Do we have to swap reflink flags? */
2060 if ((ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK) ^
2061 (tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)) {
2062 f = ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
2063 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
2064 ip->i_d.di_flags2 |= tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
2065 tip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
2066 tip->i_d.di_flags2 |= f & XFS_DIFLAG2_REFLINK;
2067 cowfp = ip->i_cowfp;
2068 ip->i_cowfp = tip->i_cowfp;
2069 tip->i_cowfp = cowfp;
2070 xfs_inode_set_cowblocks_tag(ip);
2071 xfs_inode_set_cowblocks_tag(tip);
2072 }
2073
2074 xfs_trans_log_inode(tp, ip, src_log_flags);
2075 xfs_trans_log_inode(tp, tip, target_log_flags);
2076
2077 /*
2078 * If this is a synchronous mount, make sure that the
2079 * transaction goes to disk before returning to the user.
2080 */
2081 if (mp->m_flags & XFS_MOUNT_WSYNC)
2082 xfs_trans_set_sync(tp);
2083
2084 error = xfs_trans_commit(tp);
2085
2086 trace_xfs_swap_extent_after(ip, 0);
2087 trace_xfs_swap_extent_after(tip, 1);
2088
2089out_unlock:
2090 xfs_iunlock(ip, lock_flags);
2091 xfs_iunlock(tip, lock_flags);
2092 unlock_two_nondirectories(VFS_I(ip), VFS_I(tip));
2093 return error;
2094
2095out_trans_cancel:
2096 xfs_trans_cancel(tp);
2097 goto out_unlock;
2098}