Loading...
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_sb.h"
27#include "xfs_ag.h"
28#include "xfs_mount.h"
29#include "xfs_da_format.h"
30#include "xfs_inode.h"
31#include "xfs_btree.h"
32#include "xfs_trans.h"
33#include "xfs_extfree_item.h"
34#include "xfs_alloc.h"
35#include "xfs_bmap.h"
36#include "xfs_bmap_util.h"
37#include "xfs_bmap_btree.h"
38#include "xfs_rtalloc.h"
39#include "xfs_error.h"
40#include "xfs_quota.h"
41#include "xfs_trans_space.h"
42#include "xfs_trace.h"
43#include "xfs_icache.h"
44#include "xfs_log.h"
45#include "xfs_dinode.h"
46
47/* Kernel only BMAP related definitions and functions */
48
49/*
50 * Convert the given file system block to a disk block. We have to treat it
51 * differently based on whether the file is a real time file or not, because the
52 * bmap code does.
53 */
54xfs_daddr_t
55xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb)
56{
57 return (XFS_IS_REALTIME_INODE(ip) ? \
58 (xfs_daddr_t)XFS_FSB_TO_BB((ip)->i_mount, (fsb)) : \
59 XFS_FSB_TO_DADDR((ip)->i_mount, (fsb)));
60}
61
62/*
63 * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
64 * caller. Frees all the extents that need freeing, which must be done
65 * last due to locking considerations. We never free any extents in
66 * the first transaction.
67 *
68 * Return 1 if the given transaction was committed and a new one
69 * started, and 0 otherwise in the committed parameter.
70 */
71int /* error */
72xfs_bmap_finish(
73 xfs_trans_t **tp, /* transaction pointer addr */
74 xfs_bmap_free_t *flist, /* i/o: list extents to free */
75 int *committed) /* xact committed or not */
76{
77 xfs_efd_log_item_t *efd; /* extent free data */
78 xfs_efi_log_item_t *efi; /* extent free intention */
79 int error; /* error return value */
80 xfs_bmap_free_item_t *free; /* free extent item */
81 struct xfs_trans_res tres; /* new log reservation */
82 xfs_mount_t *mp; /* filesystem mount structure */
83 xfs_bmap_free_item_t *next; /* next item on free list */
84 xfs_trans_t *ntp; /* new transaction pointer */
85
86 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
87 if (flist->xbf_count == 0) {
88 *committed = 0;
89 return 0;
90 }
91 ntp = *tp;
92 efi = xfs_trans_get_efi(ntp, flist->xbf_count);
93 for (free = flist->xbf_first; free; free = free->xbfi_next)
94 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
95 free->xbfi_blockcount);
96
97 tres.tr_logres = ntp->t_log_res;
98 tres.tr_logcount = ntp->t_log_count;
99 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
100 ntp = xfs_trans_dup(*tp);
101 error = xfs_trans_commit(*tp, 0);
102 *tp = ntp;
103 *committed = 1;
104 /*
105 * We have a new transaction, so we should return committed=1,
106 * even though we're returning an error.
107 */
108 if (error)
109 return error;
110
111 /*
112 * transaction commit worked ok so we can drop the extra ticket
113 * reference that we gained in xfs_trans_dup()
114 */
115 xfs_log_ticket_put(ntp->t_ticket);
116
117 error = xfs_trans_reserve(ntp, &tres, 0, 0);
118 if (error)
119 return error;
120 efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
121 for (free = flist->xbf_first; free != NULL; free = next) {
122 next = free->xbfi_next;
123 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
124 free->xbfi_blockcount))) {
125 /*
126 * The bmap free list will be cleaned up at a
127 * higher level. The EFI will be canceled when
128 * this transaction is aborted.
129 * Need to force shutdown here to make sure it
130 * happens, since this transaction may not be
131 * dirty yet.
132 */
133 mp = ntp->t_mountp;
134 if (!XFS_FORCED_SHUTDOWN(mp))
135 xfs_force_shutdown(mp,
136 (error == EFSCORRUPTED) ?
137 SHUTDOWN_CORRUPT_INCORE :
138 SHUTDOWN_META_IO_ERROR);
139 return error;
140 }
141 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
142 free->xbfi_blockcount);
143 xfs_bmap_del_free(flist, NULL, free);
144 }
145 return 0;
146}
147
148int
149xfs_bmap_rtalloc(
150 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
151{
152 xfs_alloctype_t atype = 0; /* type for allocation routines */
153 int error; /* error return value */
154 xfs_mount_t *mp; /* mount point structure */
155 xfs_extlen_t prod = 0; /* product factor for allocators */
156 xfs_extlen_t ralen = 0; /* realtime allocation length */
157 xfs_extlen_t align; /* minimum allocation alignment */
158 xfs_rtblock_t rtb;
159
160 mp = ap->ip->i_mount;
161 align = xfs_get_extsz_hint(ap->ip);
162 prod = align / mp->m_sb.sb_rextsize;
163 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
164 align, 1, ap->eof, 0,
165 ap->conv, &ap->offset, &ap->length);
166 if (error)
167 return error;
168 ASSERT(ap->length);
169 ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
170
171 /*
172 * If the offset & length are not perfectly aligned
173 * then kill prod, it will just get us in trouble.
174 */
175 if (do_mod(ap->offset, align) || ap->length % align)
176 prod = 1;
177 /*
178 * Set ralen to be the actual requested length in rtextents.
179 */
180 ralen = ap->length / mp->m_sb.sb_rextsize;
181 /*
182 * If the old value was close enough to MAXEXTLEN that
183 * we rounded up to it, cut it back so it's valid again.
184 * Note that if it's a really large request (bigger than
185 * MAXEXTLEN), we don't hear about that number, and can't
186 * adjust the starting point to match it.
187 */
188 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
189 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
190
191 /*
192 * Lock out other modifications to the RT bitmap inode.
193 */
194 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
195 xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
196
197 /*
198 * If it's an allocation to an empty file at offset 0,
199 * pick an extent that will space things out in the rt area.
200 */
201 if (ap->eof && ap->offset == 0) {
202 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
203
204 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
205 if (error)
206 return error;
207 ap->blkno = rtx * mp->m_sb.sb_rextsize;
208 } else {
209 ap->blkno = 0;
210 }
211
212 xfs_bmap_adjacent(ap);
213
214 /*
215 * Realtime allocation, done through xfs_rtallocate_extent.
216 */
217 atype = ap->blkno == 0 ? XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
218 do_div(ap->blkno, mp->m_sb.sb_rextsize);
219 rtb = ap->blkno;
220 ap->length = ralen;
221 if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
222 &ralen, atype, ap->wasdel, prod, &rtb)))
223 return error;
224 if (rtb == NULLFSBLOCK && prod > 1 &&
225 (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1,
226 ap->length, &ralen, atype,
227 ap->wasdel, 1, &rtb)))
228 return error;
229 ap->blkno = rtb;
230 if (ap->blkno != NULLFSBLOCK) {
231 ap->blkno *= mp->m_sb.sb_rextsize;
232 ralen *= mp->m_sb.sb_rextsize;
233 ap->length = ralen;
234 ap->ip->i_d.di_nblocks += ralen;
235 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
236 if (ap->wasdel)
237 ap->ip->i_delayed_blks -= ralen;
238 /*
239 * Adjust the disk quota also. This was reserved
240 * earlier.
241 */
242 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
243 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
244 XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
245 } else {
246 ap->length = 0;
247 }
248 return 0;
249}
250
251/*
252 * Stack switching interfaces for allocation
253 */
254static void
255xfs_bmapi_allocate_worker(
256 struct work_struct *work)
257{
258 struct xfs_bmalloca *args = container_of(work,
259 struct xfs_bmalloca, work);
260 unsigned long pflags;
261
262 /* we are in a transaction context here */
263 current_set_flags_nested(&pflags, PF_FSTRANS);
264
265 args->result = __xfs_bmapi_allocate(args);
266 complete(args->done);
267
268 current_restore_flags_nested(&pflags, PF_FSTRANS);
269}
270
271/*
272 * Some allocation requests often come in with little stack to work on. Push
273 * them off to a worker thread so there is lots of stack to use. Otherwise just
274 * call directly to avoid the context switch overhead here.
275 */
276int
277xfs_bmapi_allocate(
278 struct xfs_bmalloca *args)
279{
280 DECLARE_COMPLETION_ONSTACK(done);
281
282 if (!args->stack_switch)
283 return __xfs_bmapi_allocate(args);
284
285
286 args->done = &done;
287 INIT_WORK_ONSTACK(&args->work, xfs_bmapi_allocate_worker);
288 queue_work(xfs_alloc_wq, &args->work);
289 wait_for_completion(&done);
290 destroy_work_on_stack(&args->work);
291 return args->result;
292}
293
294/*
295 * Check if the endoff is outside the last extent. If so the caller will grow
296 * the allocation to a stripe unit boundary. All offsets are considered outside
297 * the end of file for an empty fork, so 1 is returned in *eof in that case.
298 */
299int
300xfs_bmap_eof(
301 struct xfs_inode *ip,
302 xfs_fileoff_t endoff,
303 int whichfork,
304 int *eof)
305{
306 struct xfs_bmbt_irec rec;
307 int error;
308
309 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
310 if (error || *eof)
311 return error;
312
313 *eof = endoff >= rec.br_startoff + rec.br_blockcount;
314 return 0;
315}
316
317/*
318 * Extent tree block counting routines.
319 */
320
321/*
322 * Count leaf blocks given a range of extent records.
323 */
324STATIC void
325xfs_bmap_count_leaves(
326 xfs_ifork_t *ifp,
327 xfs_extnum_t idx,
328 int numrecs,
329 int *count)
330{
331 int b;
332
333 for (b = 0; b < numrecs; b++) {
334 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
335 *count += xfs_bmbt_get_blockcount(frp);
336 }
337}
338
339/*
340 * Count leaf blocks given a range of extent records originally
341 * in btree format.
342 */
343STATIC void
344xfs_bmap_disk_count_leaves(
345 struct xfs_mount *mp,
346 struct xfs_btree_block *block,
347 int numrecs,
348 int *count)
349{
350 int b;
351 xfs_bmbt_rec_t *frp;
352
353 for (b = 1; b <= numrecs; b++) {
354 frp = XFS_BMBT_REC_ADDR(mp, block, b);
355 *count += xfs_bmbt_disk_get_blockcount(frp);
356 }
357}
358
359/*
360 * Recursively walks each level of a btree
361 * to count total fsblocks in use.
362 */
363STATIC int /* error */
364xfs_bmap_count_tree(
365 xfs_mount_t *mp, /* file system mount point */
366 xfs_trans_t *tp, /* transaction pointer */
367 xfs_ifork_t *ifp, /* inode fork pointer */
368 xfs_fsblock_t blockno, /* file system block number */
369 int levelin, /* level in btree */
370 int *count) /* Count of blocks */
371{
372 int error;
373 xfs_buf_t *bp, *nbp;
374 int level = levelin;
375 __be64 *pp;
376 xfs_fsblock_t bno = blockno;
377 xfs_fsblock_t nextbno;
378 struct xfs_btree_block *block, *nextblock;
379 int numrecs;
380
381 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF,
382 &xfs_bmbt_buf_ops);
383 if (error)
384 return error;
385 *count += 1;
386 block = XFS_BUF_TO_BLOCK(bp);
387
388 if (--level) {
389 /* Not at node above leaves, count this level of nodes */
390 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
391 while (nextbno != NULLFSBLOCK) {
392 error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp,
393 XFS_BMAP_BTREE_REF,
394 &xfs_bmbt_buf_ops);
395 if (error)
396 return error;
397 *count += 1;
398 nextblock = XFS_BUF_TO_BLOCK(nbp);
399 nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
400 xfs_trans_brelse(tp, nbp);
401 }
402
403 /* Dive to the next level */
404 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
405 bno = be64_to_cpu(*pp);
406 if (unlikely((error =
407 xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
408 xfs_trans_brelse(tp, bp);
409 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
410 XFS_ERRLEVEL_LOW, mp);
411 return XFS_ERROR(EFSCORRUPTED);
412 }
413 xfs_trans_brelse(tp, bp);
414 } else {
415 /* count all level 1 nodes and their leaves */
416 for (;;) {
417 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
418 numrecs = be16_to_cpu(block->bb_numrecs);
419 xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
420 xfs_trans_brelse(tp, bp);
421 if (nextbno == NULLFSBLOCK)
422 break;
423 bno = nextbno;
424 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
425 XFS_BMAP_BTREE_REF,
426 &xfs_bmbt_buf_ops);
427 if (error)
428 return error;
429 *count += 1;
430 block = XFS_BUF_TO_BLOCK(bp);
431 }
432 }
433 return 0;
434}
435
436/*
437 * Count fsblocks of the given fork.
438 */
439int /* error */
440xfs_bmap_count_blocks(
441 xfs_trans_t *tp, /* transaction pointer */
442 xfs_inode_t *ip, /* incore inode */
443 int whichfork, /* data or attr fork */
444 int *count) /* out: count of blocks */
445{
446 struct xfs_btree_block *block; /* current btree block */
447 xfs_fsblock_t bno; /* block # of "block" */
448 xfs_ifork_t *ifp; /* fork structure */
449 int level; /* btree level, for checking */
450 xfs_mount_t *mp; /* file system mount structure */
451 __be64 *pp; /* pointer to block address */
452
453 bno = NULLFSBLOCK;
454 mp = ip->i_mount;
455 ifp = XFS_IFORK_PTR(ip, whichfork);
456 if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
457 xfs_bmap_count_leaves(ifp, 0,
458 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
459 count);
460 return 0;
461 }
462
463 /*
464 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
465 */
466 block = ifp->if_broot;
467 level = be16_to_cpu(block->bb_level);
468 ASSERT(level > 0);
469 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
470 bno = be64_to_cpu(*pp);
471 ASSERT(bno != NULLDFSBNO);
472 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
473 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
474
475 if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
476 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
477 mp);
478 return XFS_ERROR(EFSCORRUPTED);
479 }
480
481 return 0;
482}
483
484/*
485 * returns 1 for success, 0 if we failed to map the extent.
486 */
487STATIC int
488xfs_getbmapx_fix_eof_hole(
489 xfs_inode_t *ip, /* xfs incore inode pointer */
490 struct getbmapx *out, /* output structure */
491 int prealloced, /* this is a file with
492 * preallocated data space */
493 __int64_t end, /* last block requested */
494 xfs_fsblock_t startblock)
495{
496 __int64_t fixlen;
497 xfs_mount_t *mp; /* file system mount point */
498 xfs_ifork_t *ifp; /* inode fork pointer */
499 xfs_extnum_t lastx; /* last extent pointer */
500 xfs_fileoff_t fileblock;
501
502 if (startblock == HOLESTARTBLOCK) {
503 mp = ip->i_mount;
504 out->bmv_block = -1;
505 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
506 fixlen -= out->bmv_offset;
507 if (prealloced && out->bmv_offset + out->bmv_length == end) {
508 /* Came to hole at EOF. Trim it. */
509 if (fixlen <= 0)
510 return 0;
511 out->bmv_length = fixlen;
512 }
513 } else {
514 if (startblock == DELAYSTARTBLOCK)
515 out->bmv_block = -2;
516 else
517 out->bmv_block = xfs_fsb_to_db(ip, startblock);
518 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
519 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
520 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
521 (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
522 out->bmv_oflags |= BMV_OF_LAST;
523 }
524
525 return 1;
526}
527
528/*
529 * Get inode's extents as described in bmv, and format for output.
530 * Calls formatter to fill the user's buffer until all extents
531 * are mapped, until the passed-in bmv->bmv_count slots have
532 * been filled, or until the formatter short-circuits the loop,
533 * if it is tracking filled-in extents on its own.
534 */
535int /* error code */
536xfs_getbmap(
537 xfs_inode_t *ip,
538 struct getbmapx *bmv, /* user bmap structure */
539 xfs_bmap_format_t formatter, /* format to user */
540 void *arg) /* formatter arg */
541{
542 __int64_t bmvend; /* last block requested */
543 int error = 0; /* return value */
544 __int64_t fixlen; /* length for -1 case */
545 int i; /* extent number */
546 int lock; /* lock state */
547 xfs_bmbt_irec_t *map; /* buffer for user's data */
548 xfs_mount_t *mp; /* file system mount point */
549 int nex; /* # of user extents can do */
550 int nexleft; /* # of user extents left */
551 int subnex; /* # of bmapi's can do */
552 int nmap; /* number of map entries */
553 struct getbmapx *out; /* output structure */
554 int whichfork; /* data or attr fork */
555 int prealloced; /* this is a file with
556 * preallocated data space */
557 int iflags; /* interface flags */
558 int bmapi_flags; /* flags for xfs_bmapi */
559 int cur_ext = 0;
560
561 mp = ip->i_mount;
562 iflags = bmv->bmv_iflags;
563 whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
564
565 if (whichfork == XFS_ATTR_FORK) {
566 if (XFS_IFORK_Q(ip)) {
567 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
568 ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
569 ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
570 return XFS_ERROR(EINVAL);
571 } else if (unlikely(
572 ip->i_d.di_aformat != 0 &&
573 ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
574 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
575 ip->i_mount);
576 return XFS_ERROR(EFSCORRUPTED);
577 }
578
579 prealloced = 0;
580 fixlen = 1LL << 32;
581 } else {
582 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
583 ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
584 ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
585 return XFS_ERROR(EINVAL);
586
587 if (xfs_get_extsz_hint(ip) ||
588 ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
589 prealloced = 1;
590 fixlen = mp->m_super->s_maxbytes;
591 } else {
592 prealloced = 0;
593 fixlen = XFS_ISIZE(ip);
594 }
595 }
596
597 if (bmv->bmv_length == -1) {
598 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
599 bmv->bmv_length =
600 max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
601 } else if (bmv->bmv_length == 0) {
602 bmv->bmv_entries = 0;
603 return 0;
604 } else if (bmv->bmv_length < 0) {
605 return XFS_ERROR(EINVAL);
606 }
607
608 nex = bmv->bmv_count - 1;
609 if (nex <= 0)
610 return XFS_ERROR(EINVAL);
611 bmvend = bmv->bmv_offset + bmv->bmv_length;
612
613
614 if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
615 return XFS_ERROR(ENOMEM);
616 out = kmem_zalloc_large(bmv->bmv_count * sizeof(struct getbmapx), 0);
617 if (!out)
618 return XFS_ERROR(ENOMEM);
619
620 xfs_ilock(ip, XFS_IOLOCK_SHARED);
621 if (whichfork == XFS_DATA_FORK) {
622 if (!(iflags & BMV_IF_DELALLOC) &&
623 (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) {
624 error = -filemap_write_and_wait(VFS_I(ip)->i_mapping);
625 if (error)
626 goto out_unlock_iolock;
627
628 /*
629 * Even after flushing the inode, there can still be
630 * delalloc blocks on the inode beyond EOF due to
631 * speculative preallocation. These are not removed
632 * until the release function is called or the inode
633 * is inactivated. Hence we cannot assert here that
634 * ip->i_delayed_blks == 0.
635 */
636 }
637
638 lock = xfs_ilock_data_map_shared(ip);
639 } else {
640 lock = xfs_ilock_attr_map_shared(ip);
641 }
642
643 /*
644 * Don't let nex be bigger than the number of extents
645 * we can have assuming alternating holes and real extents.
646 */
647 if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
648 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
649
650 bmapi_flags = xfs_bmapi_aflag(whichfork);
651 if (!(iflags & BMV_IF_PREALLOC))
652 bmapi_flags |= XFS_BMAPI_IGSTATE;
653
654 /*
655 * Allocate enough space to handle "subnex" maps at a time.
656 */
657 error = ENOMEM;
658 subnex = 16;
659 map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
660 if (!map)
661 goto out_unlock_ilock;
662
663 bmv->bmv_entries = 0;
664
665 if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
666 (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
667 error = 0;
668 goto out_free_map;
669 }
670
671 nexleft = nex;
672
673 do {
674 nmap = (nexleft > subnex) ? subnex : nexleft;
675 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
676 XFS_BB_TO_FSB(mp, bmv->bmv_length),
677 map, &nmap, bmapi_flags);
678 if (error)
679 goto out_free_map;
680 ASSERT(nmap <= subnex);
681
682 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
683 out[cur_ext].bmv_oflags = 0;
684 if (map[i].br_state == XFS_EXT_UNWRITTEN)
685 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
686 else if (map[i].br_startblock == DELAYSTARTBLOCK)
687 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
688 out[cur_ext].bmv_offset =
689 XFS_FSB_TO_BB(mp, map[i].br_startoff);
690 out[cur_ext].bmv_length =
691 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
692 out[cur_ext].bmv_unused1 = 0;
693 out[cur_ext].bmv_unused2 = 0;
694
695 /*
696 * delayed allocation extents that start beyond EOF can
697 * occur due to speculative EOF allocation when the
698 * delalloc extent is larger than the largest freespace
699 * extent at conversion time. These extents cannot be
700 * converted by data writeback, so can exist here even
701 * if we are not supposed to be finding delalloc
702 * extents.
703 */
704 if (map[i].br_startblock == DELAYSTARTBLOCK &&
705 map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip)))
706 ASSERT((iflags & BMV_IF_DELALLOC) != 0);
707
708 if (map[i].br_startblock == HOLESTARTBLOCK &&
709 whichfork == XFS_ATTR_FORK) {
710 /* came to the end of attribute fork */
711 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
712 goto out_free_map;
713 }
714
715 if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
716 prealloced, bmvend,
717 map[i].br_startblock))
718 goto out_free_map;
719
720 bmv->bmv_offset =
721 out[cur_ext].bmv_offset +
722 out[cur_ext].bmv_length;
723 bmv->bmv_length =
724 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
725
726 /*
727 * In case we don't want to return the hole,
728 * don't increase cur_ext so that we can reuse
729 * it in the next loop.
730 */
731 if ((iflags & BMV_IF_NO_HOLES) &&
732 map[i].br_startblock == HOLESTARTBLOCK) {
733 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
734 continue;
735 }
736
737 nexleft--;
738 bmv->bmv_entries++;
739 cur_ext++;
740 }
741 } while (nmap && nexleft && bmv->bmv_length);
742
743 out_free_map:
744 kmem_free(map);
745 out_unlock_ilock:
746 xfs_iunlock(ip, lock);
747 out_unlock_iolock:
748 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
749
750 for (i = 0; i < cur_ext; i++) {
751 int full = 0; /* user array is full */
752
753 /* format results & advance arg */
754 error = formatter(&arg, &out[i], &full);
755 if (error || full)
756 break;
757 }
758
759 kmem_free(out);
760 return error;
761}
762
763/*
764 * dead simple method of punching delalyed allocation blocks from a range in
765 * the inode. Walks a block at a time so will be slow, but is only executed in
766 * rare error cases so the overhead is not critical. This will always punch out
767 * both the start and end blocks, even if the ranges only partially overlap
768 * them, so it is up to the caller to ensure that partial blocks are not
769 * passed in.
770 */
771int
772xfs_bmap_punch_delalloc_range(
773 struct xfs_inode *ip,
774 xfs_fileoff_t start_fsb,
775 xfs_fileoff_t length)
776{
777 xfs_fileoff_t remaining = length;
778 int error = 0;
779
780 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
781
782 do {
783 int done;
784 xfs_bmbt_irec_t imap;
785 int nimaps = 1;
786 xfs_fsblock_t firstblock;
787 xfs_bmap_free_t flist;
788
789 /*
790 * Map the range first and check that it is a delalloc extent
791 * before trying to unmap the range. Otherwise we will be
792 * trying to remove a real extent (which requires a
793 * transaction) or a hole, which is probably a bad idea...
794 */
795 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
796 XFS_BMAPI_ENTIRE);
797
798 if (error) {
799 /* something screwed, just bail */
800 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
801 xfs_alert(ip->i_mount,
802 "Failed delalloc mapping lookup ino %lld fsb %lld.",
803 ip->i_ino, start_fsb);
804 }
805 break;
806 }
807 if (!nimaps) {
808 /* nothing there */
809 goto next_block;
810 }
811 if (imap.br_startblock != DELAYSTARTBLOCK) {
812 /* been converted, ignore */
813 goto next_block;
814 }
815 WARN_ON(imap.br_blockcount == 0);
816
817 /*
818 * Note: while we initialise the firstblock/flist pair, they
819 * should never be used because blocks should never be
820 * allocated or freed for a delalloc extent and hence we need
821 * don't cancel or finish them after the xfs_bunmapi() call.
822 */
823 xfs_bmap_init(&flist, &firstblock);
824 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
825 &flist, &done);
826 if (error)
827 break;
828
829 ASSERT(!flist.xbf_count && !flist.xbf_first);
830next_block:
831 start_fsb++;
832 remaining--;
833 } while(remaining > 0);
834
835 return error;
836}
837
838/*
839 * Test whether it is appropriate to check an inode for and free post EOF
840 * blocks. The 'force' parameter determines whether we should also consider
841 * regular files that are marked preallocated or append-only.
842 */
843bool
844xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
845{
846 /* prealloc/delalloc exists only on regular files */
847 if (!S_ISREG(ip->i_d.di_mode))
848 return false;
849
850 /*
851 * Zero sized files with no cached pages and delalloc blocks will not
852 * have speculative prealloc/delalloc blocks to remove.
853 */
854 if (VFS_I(ip)->i_size == 0 &&
855 VN_CACHED(VFS_I(ip)) == 0 &&
856 ip->i_delayed_blks == 0)
857 return false;
858
859 /* If we haven't read in the extent list, then don't do it now. */
860 if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
861 return false;
862
863 /*
864 * Do not free real preallocated or append-only files unless the file
865 * has delalloc blocks and we are forced to remove them.
866 */
867 if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
868 if (!force || ip->i_delayed_blks == 0)
869 return false;
870
871 return true;
872}
873
874/*
875 * This is called by xfs_inactive to free any blocks beyond eof
876 * when the link count isn't zero and by xfs_dm_punch_hole() when
877 * punching a hole to EOF.
878 */
879int
880xfs_free_eofblocks(
881 xfs_mount_t *mp,
882 xfs_inode_t *ip,
883 bool need_iolock)
884{
885 xfs_trans_t *tp;
886 int error;
887 xfs_fileoff_t end_fsb;
888 xfs_fileoff_t last_fsb;
889 xfs_filblks_t map_len;
890 int nimaps;
891 xfs_bmbt_irec_t imap;
892
893 /*
894 * Figure out if there are any blocks beyond the end
895 * of the file. If not, then there is nothing to do.
896 */
897 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
898 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
899 if (last_fsb <= end_fsb)
900 return 0;
901 map_len = last_fsb - end_fsb;
902
903 nimaps = 1;
904 xfs_ilock(ip, XFS_ILOCK_SHARED);
905 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
906 xfs_iunlock(ip, XFS_ILOCK_SHARED);
907
908 if (!error && (nimaps != 0) &&
909 (imap.br_startblock != HOLESTARTBLOCK ||
910 ip->i_delayed_blks)) {
911 /*
912 * Attach the dquots to the inode up front.
913 */
914 error = xfs_qm_dqattach(ip, 0);
915 if (error)
916 return error;
917
918 /*
919 * There are blocks after the end of file.
920 * Free them up now by truncating the file to
921 * its current size.
922 */
923 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
924
925 if (need_iolock) {
926 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
927 xfs_trans_cancel(tp, 0);
928 return EAGAIN;
929 }
930 }
931
932 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
933 if (error) {
934 ASSERT(XFS_FORCED_SHUTDOWN(mp));
935 xfs_trans_cancel(tp, 0);
936 if (need_iolock)
937 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
938 return error;
939 }
940
941 xfs_ilock(ip, XFS_ILOCK_EXCL);
942 xfs_trans_ijoin(tp, ip, 0);
943
944 /*
945 * Do not update the on-disk file size. If we update the
946 * on-disk file size and then the system crashes before the
947 * contents of the file are flushed to disk then the files
948 * may be full of holes (ie NULL files bug).
949 */
950 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK,
951 XFS_ISIZE(ip));
952 if (error) {
953 /*
954 * If we get an error at this point we simply don't
955 * bother truncating the file.
956 */
957 xfs_trans_cancel(tp,
958 (XFS_TRANS_RELEASE_LOG_RES |
959 XFS_TRANS_ABORT));
960 } else {
961 error = xfs_trans_commit(tp,
962 XFS_TRANS_RELEASE_LOG_RES);
963 if (!error)
964 xfs_inode_clear_eofblocks_tag(ip);
965 }
966
967 xfs_iunlock(ip, XFS_ILOCK_EXCL);
968 if (need_iolock)
969 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
970 }
971 return error;
972}
973
974int
975xfs_alloc_file_space(
976 struct xfs_inode *ip,
977 xfs_off_t offset,
978 xfs_off_t len,
979 int alloc_type)
980{
981 xfs_mount_t *mp = ip->i_mount;
982 xfs_off_t count;
983 xfs_filblks_t allocated_fsb;
984 xfs_filblks_t allocatesize_fsb;
985 xfs_extlen_t extsz, temp;
986 xfs_fileoff_t startoffset_fsb;
987 xfs_fsblock_t firstfsb;
988 int nimaps;
989 int quota_flag;
990 int rt;
991 xfs_trans_t *tp;
992 xfs_bmbt_irec_t imaps[1], *imapp;
993 xfs_bmap_free_t free_list;
994 uint qblocks, resblks, resrtextents;
995 int committed;
996 int error;
997
998 trace_xfs_alloc_file_space(ip);
999
1000 if (XFS_FORCED_SHUTDOWN(mp))
1001 return XFS_ERROR(EIO);
1002
1003 error = xfs_qm_dqattach(ip, 0);
1004 if (error)
1005 return error;
1006
1007 if (len <= 0)
1008 return XFS_ERROR(EINVAL);
1009
1010 rt = XFS_IS_REALTIME_INODE(ip);
1011 extsz = xfs_get_extsz_hint(ip);
1012
1013 count = len;
1014 imapp = &imaps[0];
1015 nimaps = 1;
1016 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
1017 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
1018
1019 /*
1020 * Allocate file space until done or until there is an error
1021 */
1022 while (allocatesize_fsb && !error) {
1023 xfs_fileoff_t s, e;
1024
1025 /*
1026 * Determine space reservations for data/realtime.
1027 */
1028 if (unlikely(extsz)) {
1029 s = startoffset_fsb;
1030 do_div(s, extsz);
1031 s *= extsz;
1032 e = startoffset_fsb + allocatesize_fsb;
1033 if ((temp = do_mod(startoffset_fsb, extsz)))
1034 e += temp;
1035 if ((temp = do_mod(e, extsz)))
1036 e += extsz - temp;
1037 } else {
1038 s = 0;
1039 e = allocatesize_fsb;
1040 }
1041
1042 /*
1043 * The transaction reservation is limited to a 32-bit block
1044 * count, hence we need to limit the number of blocks we are
1045 * trying to reserve to avoid an overflow. We can't allocate
1046 * more than @nimaps extents, and an extent is limited on disk
1047 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
1048 */
1049 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
1050 if (unlikely(rt)) {
1051 resrtextents = qblocks = resblks;
1052 resrtextents /= mp->m_sb.sb_rextsize;
1053 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1054 quota_flag = XFS_QMOPT_RES_RTBLKS;
1055 } else {
1056 resrtextents = 0;
1057 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
1058 quota_flag = XFS_QMOPT_RES_REGBLKS;
1059 }
1060
1061 /*
1062 * Allocate and setup the transaction.
1063 */
1064 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1065 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
1066 resblks, resrtextents);
1067 /*
1068 * Check for running out of space
1069 */
1070 if (error) {
1071 /*
1072 * Free the transaction structure.
1073 */
1074 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1075 xfs_trans_cancel(tp, 0);
1076 break;
1077 }
1078 xfs_ilock(ip, XFS_ILOCK_EXCL);
1079 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
1080 0, quota_flag);
1081 if (error)
1082 goto error1;
1083
1084 xfs_trans_ijoin(tp, ip, 0);
1085
1086 xfs_bmap_init(&free_list, &firstfsb);
1087 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
1088 allocatesize_fsb, alloc_type, &firstfsb,
1089 0, imapp, &nimaps, &free_list);
1090 if (error) {
1091 goto error0;
1092 }
1093
1094 /*
1095 * Complete the transaction
1096 */
1097 error = xfs_bmap_finish(&tp, &free_list, &committed);
1098 if (error) {
1099 goto error0;
1100 }
1101
1102 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1103 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1104 if (error) {
1105 break;
1106 }
1107
1108 allocated_fsb = imapp->br_blockcount;
1109
1110 if (nimaps == 0) {
1111 error = XFS_ERROR(ENOSPC);
1112 break;
1113 }
1114
1115 startoffset_fsb += allocated_fsb;
1116 allocatesize_fsb -= allocated_fsb;
1117 }
1118
1119 return error;
1120
1121error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1122 xfs_bmap_cancel(&free_list);
1123 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
1124
1125error1: /* Just cancel transaction */
1126 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1127 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1128 return error;
1129}
1130
1131/*
1132 * Zero file bytes between startoff and endoff inclusive.
1133 * The iolock is held exclusive and no blocks are buffered.
1134 *
1135 * This function is used by xfs_free_file_space() to zero
1136 * partial blocks when the range to free is not block aligned.
1137 * When unreserving space with boundaries that are not block
1138 * aligned we round up the start and round down the end
1139 * boundaries and then use this function to zero the parts of
1140 * the blocks that got dropped during the rounding.
1141 */
1142STATIC int
1143xfs_zero_remaining_bytes(
1144 xfs_inode_t *ip,
1145 xfs_off_t startoff,
1146 xfs_off_t endoff)
1147{
1148 xfs_bmbt_irec_t imap;
1149 xfs_fileoff_t offset_fsb;
1150 xfs_off_t lastoffset;
1151 xfs_off_t offset;
1152 xfs_buf_t *bp;
1153 xfs_mount_t *mp = ip->i_mount;
1154 int nimap;
1155 int error = 0;
1156
1157 /*
1158 * Avoid doing I/O beyond eof - it's not necessary
1159 * since nothing can read beyond eof. The space will
1160 * be zeroed when the file is extended anyway.
1161 */
1162 if (startoff >= XFS_ISIZE(ip))
1163 return 0;
1164
1165 if (endoff > XFS_ISIZE(ip))
1166 endoff = XFS_ISIZE(ip);
1167
1168 bp = xfs_buf_get_uncached(XFS_IS_REALTIME_INODE(ip) ?
1169 mp->m_rtdev_targp : mp->m_ddev_targp,
1170 BTOBB(mp->m_sb.sb_blocksize), 0);
1171 if (!bp)
1172 return XFS_ERROR(ENOMEM);
1173
1174 xfs_buf_unlock(bp);
1175
1176 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
1177 uint lock_mode;
1178
1179 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1180 nimap = 1;
1181
1182 lock_mode = xfs_ilock_data_map_shared(ip);
1183 error = xfs_bmapi_read(ip, offset_fsb, 1, &imap, &nimap, 0);
1184 xfs_iunlock(ip, lock_mode);
1185
1186 if (error || nimap < 1)
1187 break;
1188 ASSERT(imap.br_blockcount >= 1);
1189 ASSERT(imap.br_startoff == offset_fsb);
1190 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
1191 if (lastoffset > endoff)
1192 lastoffset = endoff;
1193 if (imap.br_startblock == HOLESTARTBLOCK)
1194 continue;
1195 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1196 if (imap.br_state == XFS_EXT_UNWRITTEN)
1197 continue;
1198 XFS_BUF_UNDONE(bp);
1199 XFS_BUF_UNWRITE(bp);
1200 XFS_BUF_READ(bp);
1201 XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock));
1202
1203 if (XFS_FORCED_SHUTDOWN(mp)) {
1204 error = XFS_ERROR(EIO);
1205 break;
1206 }
1207 xfs_buf_iorequest(bp);
1208 error = xfs_buf_iowait(bp);
1209 if (error) {
1210 xfs_buf_ioerror_alert(bp,
1211 "xfs_zero_remaining_bytes(read)");
1212 break;
1213 }
1214 memset(bp->b_addr +
1215 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
1216 0, lastoffset - offset + 1);
1217 XFS_BUF_UNDONE(bp);
1218 XFS_BUF_UNREAD(bp);
1219 XFS_BUF_WRITE(bp);
1220
1221 if (XFS_FORCED_SHUTDOWN(mp)) {
1222 error = XFS_ERROR(EIO);
1223 break;
1224 }
1225 xfs_buf_iorequest(bp);
1226 error = xfs_buf_iowait(bp);
1227 if (error) {
1228 xfs_buf_ioerror_alert(bp,
1229 "xfs_zero_remaining_bytes(write)");
1230 break;
1231 }
1232 }
1233 xfs_buf_free(bp);
1234 return error;
1235}
1236
1237int
1238xfs_free_file_space(
1239 struct xfs_inode *ip,
1240 xfs_off_t offset,
1241 xfs_off_t len)
1242{
1243 int committed;
1244 int done;
1245 xfs_fileoff_t endoffset_fsb;
1246 int error;
1247 xfs_fsblock_t firstfsb;
1248 xfs_bmap_free_t free_list;
1249 xfs_bmbt_irec_t imap;
1250 xfs_off_t ioffset;
1251 xfs_extlen_t mod=0;
1252 xfs_mount_t *mp;
1253 int nimap;
1254 uint resblks;
1255 xfs_off_t rounding;
1256 int rt;
1257 xfs_fileoff_t startoffset_fsb;
1258 xfs_trans_t *tp;
1259
1260 mp = ip->i_mount;
1261
1262 trace_xfs_free_file_space(ip);
1263
1264 error = xfs_qm_dqattach(ip, 0);
1265 if (error)
1266 return error;
1267
1268 error = 0;
1269 if (len <= 0) /* if nothing being freed */
1270 return error;
1271 rt = XFS_IS_REALTIME_INODE(ip);
1272 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
1273 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
1274
1275 /* wait for the completion of any pending DIOs */
1276 inode_dio_wait(VFS_I(ip));
1277
1278 rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
1279 ioffset = offset & ~(rounding - 1);
1280 error = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1281 ioffset, -1);
1282 if (error)
1283 goto out;
1284 truncate_pagecache_range(VFS_I(ip), ioffset, -1);
1285
1286 /*
1287 * Need to zero the stuff we're not freeing, on disk.
1288 * If it's a realtime file & can't use unwritten extents then we
1289 * actually need to zero the extent edges. Otherwise xfs_bunmapi
1290 * will take care of it for us.
1291 */
1292 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
1293 nimap = 1;
1294 error = xfs_bmapi_read(ip, startoffset_fsb, 1,
1295 &imap, &nimap, 0);
1296 if (error)
1297 goto out;
1298 ASSERT(nimap == 0 || nimap == 1);
1299 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1300 xfs_daddr_t block;
1301
1302 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1303 block = imap.br_startblock;
1304 mod = do_div(block, mp->m_sb.sb_rextsize);
1305 if (mod)
1306 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
1307 }
1308 nimap = 1;
1309 error = xfs_bmapi_read(ip, endoffset_fsb - 1, 1,
1310 &imap, &nimap, 0);
1311 if (error)
1312 goto out;
1313 ASSERT(nimap == 0 || nimap == 1);
1314 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1315 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1316 mod++;
1317 if (mod && (mod != mp->m_sb.sb_rextsize))
1318 endoffset_fsb -= mod;
1319 }
1320 }
1321 if ((done = (endoffset_fsb <= startoffset_fsb)))
1322 /*
1323 * One contiguous piece to clear
1324 */
1325 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
1326 else {
1327 /*
1328 * Some full blocks, possibly two pieces to clear
1329 */
1330 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
1331 error = xfs_zero_remaining_bytes(ip, offset,
1332 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
1333 if (!error &&
1334 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
1335 error = xfs_zero_remaining_bytes(ip,
1336 XFS_FSB_TO_B(mp, endoffset_fsb),
1337 offset + len - 1);
1338 }
1339
1340 /*
1341 * free file space until done or until there is an error
1342 */
1343 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1344 while (!error && !done) {
1345
1346 /*
1347 * allocate and setup the transaction. Allow this
1348 * transaction to dip into the reserve blocks to ensure
1349 * the freeing of the space succeeds at ENOSPC.
1350 */
1351 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1352 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, resblks, 0);
1353
1354 /*
1355 * check for running out of space
1356 */
1357 if (error) {
1358 /*
1359 * Free the transaction structure.
1360 */
1361 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1362 xfs_trans_cancel(tp, 0);
1363 break;
1364 }
1365 xfs_ilock(ip, XFS_ILOCK_EXCL);
1366 error = xfs_trans_reserve_quota(tp, mp,
1367 ip->i_udquot, ip->i_gdquot, ip->i_pdquot,
1368 resblks, 0, XFS_QMOPT_RES_REGBLKS);
1369 if (error)
1370 goto error1;
1371
1372 xfs_trans_ijoin(tp, ip, 0);
1373
1374 /*
1375 * issue the bunmapi() call to free the blocks
1376 */
1377 xfs_bmap_init(&free_list, &firstfsb);
1378 error = xfs_bunmapi(tp, ip, startoffset_fsb,
1379 endoffset_fsb - startoffset_fsb,
1380 0, 2, &firstfsb, &free_list, &done);
1381 if (error) {
1382 goto error0;
1383 }
1384
1385 /*
1386 * complete the transaction
1387 */
1388 error = xfs_bmap_finish(&tp, &free_list, &committed);
1389 if (error) {
1390 goto error0;
1391 }
1392
1393 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1394 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1395 }
1396
1397 out:
1398 return error;
1399
1400 error0:
1401 xfs_bmap_cancel(&free_list);
1402 error1:
1403 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1404 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1405 goto out;
1406}
1407
1408
1409int
1410xfs_zero_file_space(
1411 struct xfs_inode *ip,
1412 xfs_off_t offset,
1413 xfs_off_t len)
1414{
1415 struct xfs_mount *mp = ip->i_mount;
1416 uint granularity;
1417 xfs_off_t start_boundary;
1418 xfs_off_t end_boundary;
1419 int error;
1420
1421 trace_xfs_zero_file_space(ip);
1422
1423 granularity = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
1424
1425 /*
1426 * Round the range of extents we are going to convert inwards. If the
1427 * offset is aligned, then it doesn't get changed so we zero from the
1428 * start of the block offset points to.
1429 */
1430 start_boundary = round_up(offset, granularity);
1431 end_boundary = round_down(offset + len, granularity);
1432
1433 ASSERT(start_boundary >= offset);
1434 ASSERT(end_boundary <= offset + len);
1435
1436 if (start_boundary < end_boundary - 1) {
1437 /*
1438 * punch out delayed allocation blocks and the page cache over
1439 * the conversion range
1440 */
1441 xfs_ilock(ip, XFS_ILOCK_EXCL);
1442 error = xfs_bmap_punch_delalloc_range(ip,
1443 XFS_B_TO_FSBT(mp, start_boundary),
1444 XFS_B_TO_FSB(mp, end_boundary - start_boundary));
1445 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1446 truncate_pagecache_range(VFS_I(ip), start_boundary,
1447 end_boundary - 1);
1448
1449 /* convert the blocks */
1450 error = xfs_alloc_file_space(ip, start_boundary,
1451 end_boundary - start_boundary - 1,
1452 XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT);
1453 if (error)
1454 goto out;
1455
1456 /* We've handled the interior of the range, now for the edges */
1457 if (start_boundary != offset) {
1458 error = xfs_iozero(ip, offset, start_boundary - offset);
1459 if (error)
1460 goto out;
1461 }
1462
1463 if (end_boundary != offset + len)
1464 error = xfs_iozero(ip, end_boundary,
1465 offset + len - end_boundary);
1466
1467 } else {
1468 /*
1469 * It's either a sub-granularity range or the range spanned lies
1470 * partially across two adjacent blocks.
1471 */
1472 error = xfs_iozero(ip, offset, len);
1473 }
1474
1475out:
1476 return error;
1477
1478}
1479
1480/*
1481 * xfs_collapse_file_space()
1482 * This routine frees disk space and shift extent for the given file.
1483 * The first thing we do is to free data blocks in the specified range
1484 * by calling xfs_free_file_space(). It would also sync dirty data
1485 * and invalidate page cache over the region on which collapse range
1486 * is working. And Shift extent records to the left to cover a hole.
1487 * RETURNS:
1488 * 0 on success
1489 * errno on error
1490 *
1491 */
1492int
1493xfs_collapse_file_space(
1494 struct xfs_inode *ip,
1495 xfs_off_t offset,
1496 xfs_off_t len)
1497{
1498 int done = 0;
1499 struct xfs_mount *mp = ip->i_mount;
1500 struct xfs_trans *tp;
1501 int error;
1502 xfs_extnum_t current_ext = 0;
1503 struct xfs_bmap_free free_list;
1504 xfs_fsblock_t first_block;
1505 int committed;
1506 xfs_fileoff_t start_fsb;
1507 xfs_fileoff_t shift_fsb;
1508
1509 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1510
1511 trace_xfs_collapse_file_space(ip);
1512
1513 start_fsb = XFS_B_TO_FSB(mp, offset + len);
1514 shift_fsb = XFS_B_TO_FSB(mp, len);
1515
1516 error = xfs_free_file_space(ip, offset, len);
1517 if (error)
1518 return error;
1519
1520 while (!error && !done) {
1521 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1522 tp->t_flags |= XFS_TRANS_RESERVE;
1523 /*
1524 * We would need to reserve permanent block for transaction.
1525 * This will come into picture when after shifting extent into
1526 * hole we found that adjacent extents can be merged which
1527 * may lead to freeing of a block during record update.
1528 */
1529 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
1530 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0);
1531 if (error) {
1532 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1533 xfs_trans_cancel(tp, 0);
1534 break;
1535 }
1536
1537 xfs_ilock(ip, XFS_ILOCK_EXCL);
1538 error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot,
1539 ip->i_gdquot, ip->i_pdquot,
1540 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0,
1541 XFS_QMOPT_RES_REGBLKS);
1542 if (error)
1543 goto out;
1544
1545 xfs_trans_ijoin(tp, ip, 0);
1546
1547 xfs_bmap_init(&free_list, &first_block);
1548
1549 /*
1550 * We are using the write transaction in which max 2 bmbt
1551 * updates are allowed
1552 */
1553 error = xfs_bmap_shift_extents(tp, ip, &done, start_fsb,
1554 shift_fsb, ¤t_ext,
1555 &first_block, &free_list,
1556 XFS_BMAP_MAX_SHIFT_EXTENTS);
1557 if (error)
1558 goto out;
1559
1560 error = xfs_bmap_finish(&tp, &free_list, &committed);
1561 if (error)
1562 goto out;
1563
1564 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1565 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1566 }
1567
1568 return error;
1569
1570out:
1571 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1572 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1573 return error;
1574}
1575
1576/*
1577 * We need to check that the format of the data fork in the temporary inode is
1578 * valid for the target inode before doing the swap. This is not a problem with
1579 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1580 * data fork depending on the space the attribute fork is taking so we can get
1581 * invalid formats on the target inode.
1582 *
1583 * E.g. target has space for 7 extents in extent format, temp inode only has
1584 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1585 * btree, but when swapped it needs to be in extent format. Hence we can't just
1586 * blindly swap data forks on attr2 filesystems.
1587 *
1588 * Note that we check the swap in both directions so that we don't end up with
1589 * a corrupt temporary inode, either.
1590 *
1591 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1592 * inode will prevent this situation from occurring, so all we do here is
1593 * reject and log the attempt. basically we are putting the responsibility on
1594 * userspace to get this right.
1595 */
1596static int
1597xfs_swap_extents_check_format(
1598 xfs_inode_t *ip, /* target inode */
1599 xfs_inode_t *tip) /* tmp inode */
1600{
1601
1602 /* Should never get a local format */
1603 if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
1604 tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
1605 return EINVAL;
1606
1607 /*
1608 * if the target inode has less extents that then temporary inode then
1609 * why did userspace call us?
1610 */
1611 if (ip->i_d.di_nextents < tip->i_d.di_nextents)
1612 return EINVAL;
1613
1614 /*
1615 * if the target inode is in extent form and the temp inode is in btree
1616 * form then we will end up with the target inode in the wrong format
1617 * as we already know there are less extents in the temp inode.
1618 */
1619 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1620 tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1621 return EINVAL;
1622
1623 /* Check temp in extent form to max in target */
1624 if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1625 XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) >
1626 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1627 return EINVAL;
1628
1629 /* Check target in extent form to max in temp */
1630 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1631 XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) >
1632 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1633 return EINVAL;
1634
1635 /*
1636 * If we are in a btree format, check that the temp root block will fit
1637 * in the target and that it has enough extents to be in btree format
1638 * in the target.
1639 *
1640 * Note that we have to be careful to allow btree->extent conversions
1641 * (a common defrag case) which will occur when the temp inode is in
1642 * extent format...
1643 */
1644 if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1645 if (XFS_IFORK_BOFF(ip) &&
1646 XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip))
1647 return EINVAL;
1648 if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <=
1649 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1650 return EINVAL;
1651 }
1652
1653 /* Reciprocal target->temp btree format checks */
1654 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1655 if (XFS_IFORK_BOFF(tip) &&
1656 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
1657 return EINVAL;
1658 if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <=
1659 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1660 return EINVAL;
1661 }
1662
1663 return 0;
1664}
1665
1666int
1667xfs_swap_extents(
1668 xfs_inode_t *ip, /* target inode */
1669 xfs_inode_t *tip, /* tmp inode */
1670 xfs_swapext_t *sxp)
1671{
1672 xfs_mount_t *mp = ip->i_mount;
1673 xfs_trans_t *tp;
1674 xfs_bstat_t *sbp = &sxp->sx_stat;
1675 xfs_ifork_t *tempifp, *ifp, *tifp;
1676 int src_log_flags, target_log_flags;
1677 int error = 0;
1678 int aforkblks = 0;
1679 int taforkblks = 0;
1680 __uint64_t tmp;
1681
1682 tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL);
1683 if (!tempifp) {
1684 error = XFS_ERROR(ENOMEM);
1685 goto out;
1686 }
1687
1688 /*
1689 * we have to do two separate lock calls here to keep lockdep
1690 * happy. If we try to get all the locks in one call, lock will
1691 * report false positives when we drop the ILOCK and regain them
1692 * below.
1693 */
1694 xfs_lock_two_inodes(ip, tip, XFS_IOLOCK_EXCL);
1695 xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
1696
1697 /* Verify that both files have the same format */
1698 if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) {
1699 error = XFS_ERROR(EINVAL);
1700 goto out_unlock;
1701 }
1702
1703 /* Verify both files are either real-time or non-realtime */
1704 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1705 error = XFS_ERROR(EINVAL);
1706 goto out_unlock;
1707 }
1708
1709 error = -filemap_write_and_wait(VFS_I(tip)->i_mapping);
1710 if (error)
1711 goto out_unlock;
1712 truncate_pagecache_range(VFS_I(tip), 0, -1);
1713
1714 /* Verify O_DIRECT for ftmp */
1715 if (VN_CACHED(VFS_I(tip)) != 0) {
1716 error = XFS_ERROR(EINVAL);
1717 goto out_unlock;
1718 }
1719
1720 /* Verify all data are being swapped */
1721 if (sxp->sx_offset != 0 ||
1722 sxp->sx_length != ip->i_d.di_size ||
1723 sxp->sx_length != tip->i_d.di_size) {
1724 error = XFS_ERROR(EFAULT);
1725 goto out_unlock;
1726 }
1727
1728 trace_xfs_swap_extent_before(ip, 0);
1729 trace_xfs_swap_extent_before(tip, 1);
1730
1731 /* check inode formats now that data is flushed */
1732 error = xfs_swap_extents_check_format(ip, tip);
1733 if (error) {
1734 xfs_notice(mp,
1735 "%s: inode 0x%llx format is incompatible for exchanging.",
1736 __func__, ip->i_ino);
1737 goto out_unlock;
1738 }
1739
1740 /*
1741 * Compare the current change & modify times with that
1742 * passed in. If they differ, we abort this swap.
1743 * This is the mechanism used to ensure the calling
1744 * process that the file was not changed out from
1745 * under it.
1746 */
1747 if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
1748 (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
1749 (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
1750 (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
1751 error = XFS_ERROR(EBUSY);
1752 goto out_unlock;
1753 }
1754
1755 /* We need to fail if the file is memory mapped. Once we have tossed
1756 * all existing pages, the page fault will have no option
1757 * but to go to the filesystem for pages. By making the page fault call
1758 * vop_read (or write in the case of autogrow) they block on the iolock
1759 * until we have switched the extents.
1760 */
1761 if (VN_MAPPED(VFS_I(ip))) {
1762 error = XFS_ERROR(EBUSY);
1763 goto out_unlock;
1764 }
1765
1766 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1767 xfs_iunlock(tip, XFS_ILOCK_EXCL);
1768
1769 /*
1770 * There is a race condition here since we gave up the
1771 * ilock. However, the data fork will not change since
1772 * we have the iolock (locked for truncation too) so we
1773 * are safe. We don't really care if non-io related
1774 * fields change.
1775 */
1776 truncate_pagecache_range(VFS_I(ip), 0, -1);
1777
1778 tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT);
1779 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
1780 if (error) {
1781 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1782 xfs_iunlock(tip, XFS_IOLOCK_EXCL);
1783 xfs_trans_cancel(tp, 0);
1784 goto out;
1785 }
1786 xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
1787
1788 /*
1789 * Count the number of extended attribute blocks
1790 */
1791 if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
1792 (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1793 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
1794 if (error)
1795 goto out_trans_cancel;
1796 }
1797 if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
1798 (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1799 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
1800 &taforkblks);
1801 if (error)
1802 goto out_trans_cancel;
1803 }
1804
1805 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1806 xfs_trans_ijoin(tp, tip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1807
1808 /*
1809 * Before we've swapped the forks, lets set the owners of the forks
1810 * appropriately. We have to do this as we are demand paging the btree
1811 * buffers, and so the validation done on read will expect the owner
1812 * field to be correctly set. Once we change the owners, we can swap the
1813 * inode forks.
1814 *
1815 * Note the trickiness in setting the log flags - we set the owner log
1816 * flag on the opposite inode (i.e. the inode we are setting the new
1817 * owner to be) because once we swap the forks and log that, log
1818 * recovery is going to see the fork as owned by the swapped inode,
1819 * not the pre-swapped inodes.
1820 */
1821 src_log_flags = XFS_ILOG_CORE;
1822 target_log_flags = XFS_ILOG_CORE;
1823 if (ip->i_d.di_version == 3 &&
1824 ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1825 target_log_flags |= XFS_ILOG_DOWNER;
1826 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK,
1827 tip->i_ino, NULL);
1828 if (error)
1829 goto out_trans_cancel;
1830 }
1831
1832 if (tip->i_d.di_version == 3 &&
1833 tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1834 src_log_flags |= XFS_ILOG_DOWNER;
1835 error = xfs_bmbt_change_owner(tp, tip, XFS_DATA_FORK,
1836 ip->i_ino, NULL);
1837 if (error)
1838 goto out_trans_cancel;
1839 }
1840
1841 /*
1842 * Swap the data forks of the inodes
1843 */
1844 ifp = &ip->i_df;
1845 tifp = &tip->i_df;
1846 *tempifp = *ifp; /* struct copy */
1847 *ifp = *tifp; /* struct copy */
1848 *tifp = *tempifp; /* struct copy */
1849
1850 /*
1851 * Fix the on-disk inode values
1852 */
1853 tmp = (__uint64_t)ip->i_d.di_nblocks;
1854 ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
1855 tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
1856
1857 tmp = (__uint64_t) ip->i_d.di_nextents;
1858 ip->i_d.di_nextents = tip->i_d.di_nextents;
1859 tip->i_d.di_nextents = tmp;
1860
1861 tmp = (__uint64_t) ip->i_d.di_format;
1862 ip->i_d.di_format = tip->i_d.di_format;
1863 tip->i_d.di_format = tmp;
1864
1865 /*
1866 * The extents in the source inode could still contain speculative
1867 * preallocation beyond EOF (e.g. the file is open but not modified
1868 * while defrag is in progress). In that case, we need to copy over the
1869 * number of delalloc blocks the data fork in the source inode is
1870 * tracking beyond EOF so that when the fork is truncated away when the
1871 * temporary inode is unlinked we don't underrun the i_delayed_blks
1872 * counter on that inode.
1873 */
1874 ASSERT(tip->i_delayed_blks == 0);
1875 tip->i_delayed_blks = ip->i_delayed_blks;
1876 ip->i_delayed_blks = 0;
1877
1878 switch (ip->i_d.di_format) {
1879 case XFS_DINODE_FMT_EXTENTS:
1880 /* If the extents fit in the inode, fix the
1881 * pointer. Otherwise it's already NULL or
1882 * pointing to the extent.
1883 */
1884 if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) {
1885 ifp->if_u1.if_extents =
1886 ifp->if_u2.if_inline_ext;
1887 }
1888 src_log_flags |= XFS_ILOG_DEXT;
1889 break;
1890 case XFS_DINODE_FMT_BTREE:
1891 ASSERT(ip->i_d.di_version < 3 ||
1892 (src_log_flags & XFS_ILOG_DOWNER));
1893 src_log_flags |= XFS_ILOG_DBROOT;
1894 break;
1895 }
1896
1897 switch (tip->i_d.di_format) {
1898 case XFS_DINODE_FMT_EXTENTS:
1899 /* If the extents fit in the inode, fix the
1900 * pointer. Otherwise it's already NULL or
1901 * pointing to the extent.
1902 */
1903 if (tip->i_d.di_nextents <= XFS_INLINE_EXTS) {
1904 tifp->if_u1.if_extents =
1905 tifp->if_u2.if_inline_ext;
1906 }
1907 target_log_flags |= XFS_ILOG_DEXT;
1908 break;
1909 case XFS_DINODE_FMT_BTREE:
1910 target_log_flags |= XFS_ILOG_DBROOT;
1911 ASSERT(tip->i_d.di_version < 3 ||
1912 (target_log_flags & XFS_ILOG_DOWNER));
1913 break;
1914 }
1915
1916 xfs_trans_log_inode(tp, ip, src_log_flags);
1917 xfs_trans_log_inode(tp, tip, target_log_flags);
1918
1919 /*
1920 * If this is a synchronous mount, make sure that the
1921 * transaction goes to disk before returning to the user.
1922 */
1923 if (mp->m_flags & XFS_MOUNT_WSYNC)
1924 xfs_trans_set_sync(tp);
1925
1926 error = xfs_trans_commit(tp, 0);
1927
1928 trace_xfs_swap_extent_after(ip, 0);
1929 trace_xfs_swap_extent_after(tip, 1);
1930out:
1931 kmem_free(tempifp);
1932 return error;
1933
1934out_unlock:
1935 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1936 xfs_iunlock(tip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1937 goto out;
1938
1939out_trans_cancel:
1940 xfs_trans_cancel(tp, 0);
1941 goto out_unlock;
1942}
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#include "xfs_rtbitmap.h"
32
33/* Kernel only BMAP related definitions and functions */
34
35/*
36 * Convert the given file system block to a disk block. We have to treat it
37 * differently based on whether the file is a real time file or not, because the
38 * bmap code does.
39 */
40xfs_daddr_t
41xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb)
42{
43 if (XFS_IS_REALTIME_INODE(ip))
44 return XFS_FSB_TO_BB(ip->i_mount, fsb);
45 return XFS_FSB_TO_DADDR(ip->i_mount, fsb);
46}
47
48/*
49 * Routine to zero an extent on disk allocated to the specific inode.
50 *
51 * The VFS functions take a linearised filesystem block offset, so we have to
52 * convert the sparse xfs fsb to the right format first.
53 * VFS types are real funky, too.
54 */
55int
56xfs_zero_extent(
57 struct xfs_inode *ip,
58 xfs_fsblock_t start_fsb,
59 xfs_off_t count_fsb)
60{
61 struct xfs_mount *mp = ip->i_mount;
62 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
63 xfs_daddr_t sector = xfs_fsb_to_db(ip, start_fsb);
64 sector_t block = XFS_BB_TO_FSBT(mp, sector);
65
66 return blkdev_issue_zeroout(target->bt_bdev,
67 block << (mp->m_super->s_blocksize_bits - 9),
68 count_fsb << (mp->m_super->s_blocksize_bits - 9),
69 GFP_KERNEL, 0);
70}
71
72/*
73 * Extent tree block counting routines.
74 */
75
76/*
77 * Count leaf blocks given a range of extent records. Delayed allocation
78 * extents are not counted towards the totals.
79 */
80xfs_extnum_t
81xfs_bmap_count_leaves(
82 struct xfs_ifork *ifp,
83 xfs_filblks_t *count)
84{
85 struct xfs_iext_cursor icur;
86 struct xfs_bmbt_irec got;
87 xfs_extnum_t numrecs = 0;
88
89 for_each_xfs_iext(ifp, &icur, &got) {
90 if (!isnullstartblock(got.br_startblock)) {
91 *count += got.br_blockcount;
92 numrecs++;
93 }
94 }
95
96 return numrecs;
97}
98
99/*
100 * Count fsblocks of the given fork. Delayed allocation extents are
101 * not counted towards the totals.
102 */
103int
104xfs_bmap_count_blocks(
105 struct xfs_trans *tp,
106 struct xfs_inode *ip,
107 int whichfork,
108 xfs_extnum_t *nextents,
109 xfs_filblks_t *count)
110{
111 struct xfs_mount *mp = ip->i_mount;
112 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
113 struct xfs_btree_cur *cur;
114 xfs_extlen_t btblocks = 0;
115 int error;
116
117 *nextents = 0;
118 *count = 0;
119
120 if (!ifp)
121 return 0;
122
123 switch (ifp->if_format) {
124 case XFS_DINODE_FMT_BTREE:
125 error = xfs_iread_extents(tp, ip, whichfork);
126 if (error)
127 return error;
128
129 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
130 error = xfs_btree_count_blocks(cur, &btblocks);
131 xfs_btree_del_cursor(cur, error);
132 if (error)
133 return error;
134
135 /*
136 * xfs_btree_count_blocks includes the root block contained in
137 * the inode fork in @btblocks, so subtract one because we're
138 * only interested in allocated disk blocks.
139 */
140 *count += btblocks - 1;
141
142 fallthrough;
143 case XFS_DINODE_FMT_EXTENTS:
144 *nextents = xfs_bmap_count_leaves(ifp, count);
145 break;
146 }
147
148 return 0;
149}
150
151static int
152xfs_getbmap_report_one(
153 struct xfs_inode *ip,
154 struct getbmapx *bmv,
155 struct kgetbmap *out,
156 int64_t bmv_end,
157 struct xfs_bmbt_irec *got)
158{
159 struct kgetbmap *p = out + bmv->bmv_entries;
160 bool shared = false;
161 int error;
162
163 error = xfs_reflink_trim_around_shared(ip, got, &shared);
164 if (error)
165 return error;
166
167 if (isnullstartblock(got->br_startblock) ||
168 got->br_startblock == DELAYSTARTBLOCK) {
169 /*
170 * Take the flush completion as being a point-in-time snapshot
171 * where there are no delalloc extents, and if any new ones
172 * have been created racily, just skip them as being 'after'
173 * the flush and so don't get reported.
174 */
175 if (!(bmv->bmv_iflags & BMV_IF_DELALLOC))
176 return 0;
177
178 p->bmv_oflags |= BMV_OF_DELALLOC;
179 p->bmv_block = -2;
180 } else {
181 p->bmv_block = xfs_fsb_to_db(ip, got->br_startblock);
182 }
183
184 if (got->br_state == XFS_EXT_UNWRITTEN &&
185 (bmv->bmv_iflags & BMV_IF_PREALLOC))
186 p->bmv_oflags |= BMV_OF_PREALLOC;
187
188 if (shared)
189 p->bmv_oflags |= BMV_OF_SHARED;
190
191 p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, got->br_startoff);
192 p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, got->br_blockcount);
193
194 bmv->bmv_offset = p->bmv_offset + p->bmv_length;
195 bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset);
196 bmv->bmv_entries++;
197 return 0;
198}
199
200static void
201xfs_getbmap_report_hole(
202 struct xfs_inode *ip,
203 struct getbmapx *bmv,
204 struct kgetbmap *out,
205 int64_t bmv_end,
206 xfs_fileoff_t bno,
207 xfs_fileoff_t end)
208{
209 struct kgetbmap *p = out + bmv->bmv_entries;
210
211 if (bmv->bmv_iflags & BMV_IF_NO_HOLES)
212 return;
213
214 p->bmv_block = -1;
215 p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, bno);
216 p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, end - bno);
217
218 bmv->bmv_offset = p->bmv_offset + p->bmv_length;
219 bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset);
220 bmv->bmv_entries++;
221}
222
223static inline bool
224xfs_getbmap_full(
225 struct getbmapx *bmv)
226{
227 return bmv->bmv_length == 0 || bmv->bmv_entries >= bmv->bmv_count - 1;
228}
229
230static bool
231xfs_getbmap_next_rec(
232 struct xfs_bmbt_irec *rec,
233 xfs_fileoff_t total_end)
234{
235 xfs_fileoff_t end = rec->br_startoff + rec->br_blockcount;
236
237 if (end == total_end)
238 return false;
239
240 rec->br_startoff += rec->br_blockcount;
241 if (!isnullstartblock(rec->br_startblock) &&
242 rec->br_startblock != DELAYSTARTBLOCK)
243 rec->br_startblock += rec->br_blockcount;
244 rec->br_blockcount = total_end - end;
245 return true;
246}
247
248/*
249 * Get inode's extents as described in bmv, and format for output.
250 * Calls formatter to fill the user's buffer until all extents
251 * are mapped, until the passed-in bmv->bmv_count slots have
252 * been filled, or until the formatter short-circuits the loop,
253 * if it is tracking filled-in extents on its own.
254 */
255int /* error code */
256xfs_getbmap(
257 struct xfs_inode *ip,
258 struct getbmapx *bmv, /* user bmap structure */
259 struct kgetbmap *out)
260{
261 struct xfs_mount *mp = ip->i_mount;
262 int iflags = bmv->bmv_iflags;
263 int whichfork, lock, error = 0;
264 int64_t bmv_end, max_len;
265 xfs_fileoff_t bno, first_bno;
266 struct xfs_ifork *ifp;
267 struct xfs_bmbt_irec got, rec;
268 xfs_filblks_t len;
269 struct xfs_iext_cursor icur;
270
271 if (bmv->bmv_iflags & ~BMV_IF_VALID)
272 return -EINVAL;
273#ifndef DEBUG
274 /* Only allow CoW fork queries if we're debugging. */
275 if (iflags & BMV_IF_COWFORK)
276 return -EINVAL;
277#endif
278 if ((iflags & BMV_IF_ATTRFORK) && (iflags & BMV_IF_COWFORK))
279 return -EINVAL;
280
281 if (bmv->bmv_length < -1)
282 return -EINVAL;
283 bmv->bmv_entries = 0;
284 if (bmv->bmv_length == 0)
285 return 0;
286
287 if (iflags & BMV_IF_ATTRFORK)
288 whichfork = XFS_ATTR_FORK;
289 else if (iflags & BMV_IF_COWFORK)
290 whichfork = XFS_COW_FORK;
291 else
292 whichfork = XFS_DATA_FORK;
293
294 xfs_ilock(ip, XFS_IOLOCK_SHARED);
295 switch (whichfork) {
296 case XFS_ATTR_FORK:
297 lock = xfs_ilock_attr_map_shared(ip);
298 if (!xfs_inode_has_attr_fork(ip))
299 goto out_unlock_ilock;
300
301 max_len = 1LL << 32;
302 break;
303 case XFS_COW_FORK:
304 lock = XFS_ILOCK_SHARED;
305 xfs_ilock(ip, lock);
306
307 /* No CoW fork? Just return */
308 if (!xfs_ifork_ptr(ip, whichfork))
309 goto out_unlock_ilock;
310
311 if (xfs_get_cowextsz_hint(ip))
312 max_len = mp->m_super->s_maxbytes;
313 else
314 max_len = XFS_ISIZE(ip);
315 break;
316 case XFS_DATA_FORK:
317 if (!(iflags & BMV_IF_DELALLOC) &&
318 (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_disk_size)) {
319 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
320 if (error)
321 goto out_unlock_iolock;
322
323 /*
324 * Even after flushing the inode, there can still be
325 * delalloc blocks on the inode beyond EOF due to
326 * speculative preallocation. These are not removed
327 * until the release function is called or the inode
328 * is inactivated. Hence we cannot assert here that
329 * ip->i_delayed_blks == 0.
330 */
331 }
332
333 if (xfs_get_extsz_hint(ip) ||
334 (ip->i_diflags &
335 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))
336 max_len = mp->m_super->s_maxbytes;
337 else
338 max_len = XFS_ISIZE(ip);
339
340 lock = xfs_ilock_data_map_shared(ip);
341 break;
342 }
343
344 ifp = xfs_ifork_ptr(ip, whichfork);
345
346 switch (ifp->if_format) {
347 case XFS_DINODE_FMT_EXTENTS:
348 case XFS_DINODE_FMT_BTREE:
349 break;
350 case XFS_DINODE_FMT_LOCAL:
351 /* Local format inode forks report no extents. */
352 goto out_unlock_ilock;
353 default:
354 error = -EINVAL;
355 goto out_unlock_ilock;
356 }
357
358 if (bmv->bmv_length == -1) {
359 max_len = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, max_len));
360 bmv->bmv_length = max(0LL, max_len - bmv->bmv_offset);
361 }
362
363 bmv_end = bmv->bmv_offset + bmv->bmv_length;
364
365 first_bno = bno = XFS_BB_TO_FSBT(mp, bmv->bmv_offset);
366 len = XFS_BB_TO_FSB(mp, bmv->bmv_length);
367
368 error = xfs_iread_extents(NULL, ip, whichfork);
369 if (error)
370 goto out_unlock_ilock;
371
372 if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
373 /*
374 * Report a whole-file hole if the delalloc flag is set to
375 * stay compatible with the old implementation.
376 */
377 if (iflags & BMV_IF_DELALLOC)
378 xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
379 XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
380 goto out_unlock_ilock;
381 }
382
383 while (!xfs_getbmap_full(bmv)) {
384 xfs_trim_extent(&got, first_bno, len);
385
386 /*
387 * Report an entry for a hole if this extent doesn't directly
388 * follow the previous one.
389 */
390 if (got.br_startoff > bno) {
391 xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
392 got.br_startoff);
393 if (xfs_getbmap_full(bmv))
394 break;
395 }
396
397 /*
398 * In order to report shared extents accurately, we report each
399 * distinct shared / unshared part of a single bmbt record with
400 * an individual getbmapx record.
401 */
402 bno = got.br_startoff + got.br_blockcount;
403 rec = got;
404 do {
405 error = xfs_getbmap_report_one(ip, bmv, out, bmv_end,
406 &rec);
407 if (error || xfs_getbmap_full(bmv))
408 goto out_unlock_ilock;
409 } while (xfs_getbmap_next_rec(&rec, bno));
410
411 if (!xfs_iext_next_extent(ifp, &icur, &got)) {
412 xfs_fileoff_t end = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
413
414 if (bmv->bmv_entries > 0)
415 out[bmv->bmv_entries - 1].bmv_oflags |=
416 BMV_OF_LAST;
417
418 if (whichfork != XFS_ATTR_FORK && bno < end &&
419 !xfs_getbmap_full(bmv)) {
420 xfs_getbmap_report_hole(ip, bmv, out, bmv_end,
421 bno, end);
422 }
423 break;
424 }
425
426 if (bno >= first_bno + len)
427 break;
428 }
429
430out_unlock_ilock:
431 xfs_iunlock(ip, lock);
432out_unlock_iolock:
433 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
434 return error;
435}
436
437/*
438 * Dead simple method of punching delalyed allocation blocks from a range in
439 * the inode. This will always punch out both the start and end blocks, even
440 * if the ranges only partially overlap them, so it is up to the caller to
441 * ensure that partial blocks are not passed in.
442 */
443int
444xfs_bmap_punch_delalloc_range(
445 struct xfs_inode *ip,
446 xfs_off_t start_byte,
447 xfs_off_t end_byte)
448{
449 struct xfs_mount *mp = ip->i_mount;
450 struct xfs_ifork *ifp = &ip->i_df;
451 xfs_fileoff_t start_fsb = XFS_B_TO_FSBT(mp, start_byte);
452 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, end_byte);
453 struct xfs_bmbt_irec got, del;
454 struct xfs_iext_cursor icur;
455 int error = 0;
456
457 ASSERT(!xfs_need_iread_extents(ifp));
458
459 xfs_ilock(ip, XFS_ILOCK_EXCL);
460 if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got))
461 goto out_unlock;
462
463 while (got.br_startoff + got.br_blockcount > start_fsb) {
464 del = got;
465 xfs_trim_extent(&del, start_fsb, end_fsb - start_fsb);
466
467 /*
468 * A delete can push the cursor forward. Step back to the
469 * previous extent on non-delalloc or extents outside the
470 * target range.
471 */
472 if (!del.br_blockcount ||
473 !isnullstartblock(del.br_startblock)) {
474 if (!xfs_iext_prev_extent(ifp, &icur, &got))
475 break;
476 continue;
477 }
478
479 error = xfs_bmap_del_extent_delay(ip, XFS_DATA_FORK, &icur,
480 &got, &del);
481 if (error || !xfs_iext_get_extent(ifp, &icur, &got))
482 break;
483 }
484
485out_unlock:
486 xfs_iunlock(ip, XFS_ILOCK_EXCL);
487 return error;
488}
489
490/*
491 * Test whether it is appropriate to check an inode for and free post EOF
492 * blocks. The 'force' parameter determines whether we should also consider
493 * regular files that are marked preallocated or append-only.
494 */
495bool
496xfs_can_free_eofblocks(
497 struct xfs_inode *ip,
498 bool force)
499{
500 struct xfs_bmbt_irec imap;
501 struct xfs_mount *mp = ip->i_mount;
502 xfs_fileoff_t end_fsb;
503 xfs_fileoff_t last_fsb;
504 int nimaps = 1;
505 int error;
506
507 /*
508 * Caller must either hold the exclusive io lock; or be inactivating
509 * the inode, which guarantees there are no other users of the inode.
510 */
511 if (!(VFS_I(ip)->i_state & I_FREEING))
512 xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL);
513
514 /* prealloc/delalloc exists only on regular files */
515 if (!S_ISREG(VFS_I(ip)->i_mode))
516 return false;
517
518 /*
519 * Zero sized files with no cached pages and delalloc blocks will not
520 * have speculative prealloc/delalloc blocks to remove.
521 */
522 if (VFS_I(ip)->i_size == 0 &&
523 VFS_I(ip)->i_mapping->nrpages == 0 &&
524 ip->i_delayed_blks == 0)
525 return false;
526
527 /* If we haven't read in the extent list, then don't do it now. */
528 if (xfs_need_iread_extents(&ip->i_df))
529 return false;
530
531 /*
532 * Do not free real preallocated or append-only files unless the file
533 * has delalloc blocks and we are forced to remove them.
534 */
535 if (ip->i_diflags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
536 if (!force || ip->i_delayed_blks == 0)
537 return false;
538
539 /*
540 * Do not try to free post-EOF blocks if EOF is beyond the end of the
541 * range supported by the page cache, because the truncation will loop
542 * forever.
543 */
544 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
545 if (XFS_IS_REALTIME_INODE(ip) && mp->m_sb.sb_rextsize > 1)
546 end_fsb = xfs_rtb_roundup_rtx(mp, end_fsb);
547 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
548 if (last_fsb <= end_fsb)
549 return false;
550
551 /*
552 * Look up the mapping for the first block past EOF. If we can't find
553 * it, there's nothing to free.
554 */
555 xfs_ilock(ip, XFS_ILOCK_SHARED);
556 error = xfs_bmapi_read(ip, end_fsb, last_fsb - end_fsb, &imap, &nimaps,
557 0);
558 xfs_iunlock(ip, XFS_ILOCK_SHARED);
559 if (error || nimaps == 0)
560 return false;
561
562 /*
563 * If there's a real mapping there or there are delayed allocation
564 * reservations, then we have post-EOF blocks to try to free.
565 */
566 return imap.br_startblock != HOLESTARTBLOCK || ip->i_delayed_blks;
567}
568
569/*
570 * This is called to free any blocks beyond eof. The caller must hold
571 * IOLOCK_EXCL unless we are in the inode reclaim path and have the only
572 * reference to the inode.
573 */
574int
575xfs_free_eofblocks(
576 struct xfs_inode *ip)
577{
578 struct xfs_trans *tp;
579 struct xfs_mount *mp = ip->i_mount;
580 int error;
581
582 /* Attach the dquots to the inode up front. */
583 error = xfs_qm_dqattach(ip);
584 if (error)
585 return error;
586
587 /* Wait on dio to ensure i_size has settled. */
588 inode_dio_wait(VFS_I(ip));
589
590 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
591 if (error) {
592 ASSERT(xfs_is_shutdown(mp));
593 return error;
594 }
595
596 xfs_ilock(ip, XFS_ILOCK_EXCL);
597 xfs_trans_ijoin(tp, ip, 0);
598
599 /*
600 * Do not update the on-disk file size. If we update the on-disk file
601 * size and then the system crashes before the contents of the file are
602 * flushed to disk then the files may be full of holes (ie NULL files
603 * bug).
604 */
605 error = xfs_itruncate_extents_flags(&tp, ip, XFS_DATA_FORK,
606 XFS_ISIZE(ip), XFS_BMAPI_NODISCARD);
607 if (error)
608 goto err_cancel;
609
610 error = xfs_trans_commit(tp);
611 if (error)
612 goto out_unlock;
613
614 xfs_inode_clear_eofblocks_tag(ip);
615 goto out_unlock;
616
617err_cancel:
618 /*
619 * If we get an error at this point we simply don't
620 * bother truncating the file.
621 */
622 xfs_trans_cancel(tp);
623out_unlock:
624 xfs_iunlock(ip, XFS_ILOCK_EXCL);
625 return error;
626}
627
628int
629xfs_alloc_file_space(
630 struct xfs_inode *ip,
631 xfs_off_t offset,
632 xfs_off_t len)
633{
634 xfs_mount_t *mp = ip->i_mount;
635 xfs_off_t count;
636 xfs_filblks_t allocatesize_fsb;
637 xfs_extlen_t extsz, temp;
638 xfs_fileoff_t startoffset_fsb;
639 xfs_fileoff_t endoffset_fsb;
640 int rt;
641 xfs_trans_t *tp;
642 xfs_bmbt_irec_t imaps[1], *imapp;
643 int error;
644
645 trace_xfs_alloc_file_space(ip);
646
647 if (xfs_is_shutdown(mp))
648 return -EIO;
649
650 error = xfs_qm_dqattach(ip);
651 if (error)
652 return error;
653
654 if (len <= 0)
655 return -EINVAL;
656
657 rt = XFS_IS_REALTIME_INODE(ip);
658 extsz = xfs_get_extsz_hint(ip);
659
660 count = len;
661 imapp = &imaps[0];
662 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
663 endoffset_fsb = XFS_B_TO_FSB(mp, offset + count);
664 allocatesize_fsb = endoffset_fsb - startoffset_fsb;
665
666 /*
667 * Allocate file space until done or until there is an error
668 */
669 while (allocatesize_fsb && !error) {
670 xfs_fileoff_t s, e;
671 unsigned int dblocks, rblocks, resblks;
672 int nimaps = 1;
673
674 /*
675 * Determine space reservations for data/realtime.
676 */
677 if (unlikely(extsz)) {
678 s = startoffset_fsb;
679 do_div(s, extsz);
680 s *= extsz;
681 e = startoffset_fsb + allocatesize_fsb;
682 div_u64_rem(startoffset_fsb, extsz, &temp);
683 if (temp)
684 e += temp;
685 div_u64_rem(e, extsz, &temp);
686 if (temp)
687 e += extsz - temp;
688 } else {
689 s = 0;
690 e = allocatesize_fsb;
691 }
692
693 /*
694 * The transaction reservation is limited to a 32-bit block
695 * count, hence we need to limit the number of blocks we are
696 * trying to reserve to avoid an overflow. We can't allocate
697 * more than @nimaps extents, and an extent is limited on disk
698 * to XFS_BMBT_MAX_EXTLEN (21 bits), so use that to enforce the
699 * limit.
700 */
701 resblks = min_t(xfs_fileoff_t, (e - s),
702 (XFS_MAX_BMBT_EXTLEN * nimaps));
703 if (unlikely(rt)) {
704 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
705 rblocks = resblks;
706 } else {
707 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
708 rblocks = 0;
709 }
710
711 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write,
712 dblocks, rblocks, false, &tp);
713 if (error)
714 break;
715
716 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
717 XFS_IEXT_ADD_NOSPLIT_CNT);
718 if (error == -EFBIG)
719 error = xfs_iext_count_upgrade(tp, ip,
720 XFS_IEXT_ADD_NOSPLIT_CNT);
721 if (error)
722 goto error;
723
724 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
725 allocatesize_fsb, XFS_BMAPI_PREALLOC, 0, imapp,
726 &nimaps);
727 if (error)
728 goto error;
729
730 ip->i_diflags |= XFS_DIFLAG_PREALLOC;
731 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
732
733 error = xfs_trans_commit(tp);
734 xfs_iunlock(ip, XFS_ILOCK_EXCL);
735 if (error)
736 break;
737
738 /*
739 * If the allocator cannot find a single free extent large
740 * enough to cover the start block of the requested range,
741 * xfs_bmapi_write will return 0 but leave *nimaps set to 0.
742 *
743 * In that case we simply need to keep looping with the same
744 * startoffset_fsb so that one of the following allocations
745 * will eventually reach the requested range.
746 */
747 if (nimaps) {
748 startoffset_fsb += imapp->br_blockcount;
749 allocatesize_fsb -= imapp->br_blockcount;
750 }
751 }
752
753 return error;
754
755error:
756 xfs_trans_cancel(tp);
757 xfs_iunlock(ip, XFS_ILOCK_EXCL);
758 return error;
759}
760
761static int
762xfs_unmap_extent(
763 struct xfs_inode *ip,
764 xfs_fileoff_t startoffset_fsb,
765 xfs_filblks_t len_fsb,
766 int *done)
767{
768 struct xfs_mount *mp = ip->i_mount;
769 struct xfs_trans *tp;
770 uint resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
771 int error;
772
773 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, resblks, 0,
774 false, &tp);
775 if (error)
776 return error;
777
778 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
779 XFS_IEXT_PUNCH_HOLE_CNT);
780 if (error == -EFBIG)
781 error = xfs_iext_count_upgrade(tp, ip, XFS_IEXT_PUNCH_HOLE_CNT);
782 if (error)
783 goto out_trans_cancel;
784
785 error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, done);
786 if (error)
787 goto out_trans_cancel;
788
789 error = xfs_trans_commit(tp);
790out_unlock:
791 xfs_iunlock(ip, XFS_ILOCK_EXCL);
792 return error;
793
794out_trans_cancel:
795 xfs_trans_cancel(tp);
796 goto out_unlock;
797}
798
799/* Caller must first wait for the completion of any pending DIOs if required. */
800int
801xfs_flush_unmap_range(
802 struct xfs_inode *ip,
803 xfs_off_t offset,
804 xfs_off_t len)
805{
806 struct xfs_mount *mp = ip->i_mount;
807 struct inode *inode = VFS_I(ip);
808 xfs_off_t rounding, start, end;
809 int error;
810
811 rounding = max_t(xfs_off_t, mp->m_sb.sb_blocksize, PAGE_SIZE);
812 start = round_down(offset, rounding);
813 end = round_up(offset + len, rounding) - 1;
814
815 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
816 if (error)
817 return error;
818 truncate_pagecache_range(inode, start, end);
819 return 0;
820}
821
822int
823xfs_free_file_space(
824 struct xfs_inode *ip,
825 xfs_off_t offset,
826 xfs_off_t len)
827{
828 struct xfs_mount *mp = ip->i_mount;
829 xfs_fileoff_t startoffset_fsb;
830 xfs_fileoff_t endoffset_fsb;
831 int done = 0, error;
832
833 trace_xfs_free_file_space(ip);
834
835 error = xfs_qm_dqattach(ip);
836 if (error)
837 return error;
838
839 if (len <= 0) /* if nothing being freed */
840 return 0;
841
842 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
843 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
844
845 /* We can only free complete realtime extents. */
846 if (XFS_IS_REALTIME_INODE(ip) && mp->m_sb.sb_rextsize > 1) {
847 startoffset_fsb = xfs_rtb_roundup_rtx(mp, startoffset_fsb);
848 endoffset_fsb = xfs_rtb_rounddown_rtx(mp, endoffset_fsb);
849 }
850
851 /*
852 * Need to zero the stuff we're not freeing, on disk.
853 */
854 if (endoffset_fsb > startoffset_fsb) {
855 while (!done) {
856 error = xfs_unmap_extent(ip, startoffset_fsb,
857 endoffset_fsb - startoffset_fsb, &done);
858 if (error)
859 return error;
860 }
861 }
862
863 /*
864 * Now that we've unmap all full blocks we'll have to zero out any
865 * partial block at the beginning and/or end. xfs_zero_range is smart
866 * enough to skip any holes, including those we just created, but we
867 * must take care not to zero beyond EOF and enlarge i_size.
868 */
869 if (offset >= XFS_ISIZE(ip))
870 return 0;
871 if (offset + len > XFS_ISIZE(ip))
872 len = XFS_ISIZE(ip) - offset;
873 error = xfs_zero_range(ip, offset, len, NULL);
874 if (error)
875 return error;
876
877 /*
878 * If we zeroed right up to EOF and EOF straddles a page boundary we
879 * must make sure that the post-EOF area is also zeroed because the
880 * page could be mmap'd and xfs_zero_range doesn't do that for us.
881 * Writeback of the eof page will do this, albeit clumsily.
882 */
883 if (offset + len >= XFS_ISIZE(ip) && offset_in_page(offset + len) > 0) {
884 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
885 round_down(offset + len, PAGE_SIZE), LLONG_MAX);
886 }
887
888 return error;
889}
890
891static int
892xfs_prepare_shift(
893 struct xfs_inode *ip,
894 loff_t offset)
895{
896 struct xfs_mount *mp = ip->i_mount;
897 int error;
898
899 /*
900 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
901 * into the accessible region of the file.
902 */
903 if (xfs_can_free_eofblocks(ip, true)) {
904 error = xfs_free_eofblocks(ip);
905 if (error)
906 return error;
907 }
908
909 /*
910 * Shift operations must stabilize the start block offset boundary along
911 * with the full range of the operation. If we don't, a COW writeback
912 * completion could race with an insert, front merge with the start
913 * extent (after split) during the shift and corrupt the file. Start
914 * with the block just prior to the start to stabilize the boundary.
915 */
916 offset = round_down(offset, mp->m_sb.sb_blocksize);
917 if (offset)
918 offset -= mp->m_sb.sb_blocksize;
919
920 /*
921 * Writeback and invalidate cache for the remainder of the file as we're
922 * about to shift down every extent from offset to EOF.
923 */
924 error = xfs_flush_unmap_range(ip, offset, XFS_ISIZE(ip));
925 if (error)
926 return error;
927
928 /*
929 * Clean out anything hanging around in the cow fork now that
930 * we've flushed all the dirty data out to disk to avoid having
931 * CoW extents at the wrong offsets.
932 */
933 if (xfs_inode_has_cow_data(ip)) {
934 error = xfs_reflink_cancel_cow_range(ip, offset, NULLFILEOFF,
935 true);
936 if (error)
937 return error;
938 }
939
940 return 0;
941}
942
943/*
944 * xfs_collapse_file_space()
945 * This routine frees disk space and shift extent for the given file.
946 * The first thing we do is to free data blocks in the specified range
947 * by calling xfs_free_file_space(). It would also sync dirty data
948 * and invalidate page cache over the region on which collapse range
949 * is working. And Shift extent records to the left to cover a hole.
950 * RETURNS:
951 * 0 on success
952 * errno on error
953 *
954 */
955int
956xfs_collapse_file_space(
957 struct xfs_inode *ip,
958 xfs_off_t offset,
959 xfs_off_t len)
960{
961 struct xfs_mount *mp = ip->i_mount;
962 struct xfs_trans *tp;
963 int error;
964 xfs_fileoff_t next_fsb = XFS_B_TO_FSB(mp, offset + len);
965 xfs_fileoff_t shift_fsb = XFS_B_TO_FSB(mp, len);
966 bool done = false;
967
968 xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL);
969
970 trace_xfs_collapse_file_space(ip);
971
972 error = xfs_free_file_space(ip, offset, len);
973 if (error)
974 return error;
975
976 error = xfs_prepare_shift(ip, offset);
977 if (error)
978 return error;
979
980 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
981 if (error)
982 return error;
983
984 xfs_ilock(ip, XFS_ILOCK_EXCL);
985 xfs_trans_ijoin(tp, ip, 0);
986
987 while (!done) {
988 error = xfs_bmap_collapse_extents(tp, ip, &next_fsb, shift_fsb,
989 &done);
990 if (error)
991 goto out_trans_cancel;
992 if (done)
993 break;
994
995 /* finish any deferred frees and roll the transaction */
996 error = xfs_defer_finish(&tp);
997 if (error)
998 goto out_trans_cancel;
999 }
1000
1001 error = xfs_trans_commit(tp);
1002 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1003 return error;
1004
1005out_trans_cancel:
1006 xfs_trans_cancel(tp);
1007 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1008 return error;
1009}
1010
1011/*
1012 * xfs_insert_file_space()
1013 * This routine create hole space by shifting extents for the given file.
1014 * The first thing we do is to sync dirty data and invalidate page cache
1015 * over the region on which insert range is working. And split an extent
1016 * to two extents at given offset by calling xfs_bmap_split_extent.
1017 * And shift all extent records which are laying between [offset,
1018 * last allocated extent] to the right to reserve hole range.
1019 * RETURNS:
1020 * 0 on success
1021 * errno on error
1022 */
1023int
1024xfs_insert_file_space(
1025 struct xfs_inode *ip,
1026 loff_t offset,
1027 loff_t len)
1028{
1029 struct xfs_mount *mp = ip->i_mount;
1030 struct xfs_trans *tp;
1031 int error;
1032 xfs_fileoff_t stop_fsb = XFS_B_TO_FSB(mp, offset);
1033 xfs_fileoff_t next_fsb = NULLFSBLOCK;
1034 xfs_fileoff_t shift_fsb = XFS_B_TO_FSB(mp, len);
1035 bool done = false;
1036
1037 xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL);
1038
1039 trace_xfs_insert_file_space(ip);
1040
1041 error = xfs_bmap_can_insert_extents(ip, stop_fsb, shift_fsb);
1042 if (error)
1043 return error;
1044
1045 error = xfs_prepare_shift(ip, offset);
1046 if (error)
1047 return error;
1048
1049 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
1050 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
1051 if (error)
1052 return error;
1053
1054 xfs_ilock(ip, XFS_ILOCK_EXCL);
1055 xfs_trans_ijoin(tp, ip, 0);
1056
1057 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
1058 XFS_IEXT_PUNCH_HOLE_CNT);
1059 if (error == -EFBIG)
1060 error = xfs_iext_count_upgrade(tp, ip, XFS_IEXT_PUNCH_HOLE_CNT);
1061 if (error)
1062 goto out_trans_cancel;
1063
1064 /*
1065 * The extent shifting code works on extent granularity. So, if stop_fsb
1066 * is not the starting block of extent, we need to split the extent at
1067 * stop_fsb.
1068 */
1069 error = xfs_bmap_split_extent(tp, ip, stop_fsb);
1070 if (error)
1071 goto out_trans_cancel;
1072
1073 do {
1074 error = xfs_defer_finish(&tp);
1075 if (error)
1076 goto out_trans_cancel;
1077
1078 error = xfs_bmap_insert_extents(tp, ip, &next_fsb, shift_fsb,
1079 &done, stop_fsb);
1080 if (error)
1081 goto out_trans_cancel;
1082 } while (!done);
1083
1084 error = xfs_trans_commit(tp);
1085 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1086 return error;
1087
1088out_trans_cancel:
1089 xfs_trans_cancel(tp);
1090 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1091 return error;
1092}
1093
1094/*
1095 * We need to check that the format of the data fork in the temporary inode is
1096 * valid for the target inode before doing the swap. This is not a problem with
1097 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1098 * data fork depending on the space the attribute fork is taking so we can get
1099 * invalid formats on the target inode.
1100 *
1101 * E.g. target has space for 7 extents in extent format, temp inode only has
1102 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1103 * btree, but when swapped it needs to be in extent format. Hence we can't just
1104 * blindly swap data forks on attr2 filesystems.
1105 *
1106 * Note that we check the swap in both directions so that we don't end up with
1107 * a corrupt temporary inode, either.
1108 *
1109 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1110 * inode will prevent this situation from occurring, so all we do here is
1111 * reject and log the attempt. basically we are putting the responsibility on
1112 * userspace to get this right.
1113 */
1114static int
1115xfs_swap_extents_check_format(
1116 struct xfs_inode *ip, /* target inode */
1117 struct xfs_inode *tip) /* tmp inode */
1118{
1119 struct xfs_ifork *ifp = &ip->i_df;
1120 struct xfs_ifork *tifp = &tip->i_df;
1121
1122 /* User/group/project quota ids must match if quotas are enforced. */
1123 if (XFS_IS_QUOTA_ON(ip->i_mount) &&
1124 (!uid_eq(VFS_I(ip)->i_uid, VFS_I(tip)->i_uid) ||
1125 !gid_eq(VFS_I(ip)->i_gid, VFS_I(tip)->i_gid) ||
1126 ip->i_projid != tip->i_projid))
1127 return -EINVAL;
1128
1129 /* Should never get a local format */
1130 if (ifp->if_format == XFS_DINODE_FMT_LOCAL ||
1131 tifp->if_format == XFS_DINODE_FMT_LOCAL)
1132 return -EINVAL;
1133
1134 /*
1135 * if the target inode has less extents that then temporary inode then
1136 * why did userspace call us?
1137 */
1138 if (ifp->if_nextents < tifp->if_nextents)
1139 return -EINVAL;
1140
1141 /*
1142 * If we have to use the (expensive) rmap swap method, we can
1143 * handle any number of extents and any format.
1144 */
1145 if (xfs_has_rmapbt(ip->i_mount))
1146 return 0;
1147
1148 /*
1149 * if the target inode is in extent form and the temp inode is in btree
1150 * form then we will end up with the target inode in the wrong format
1151 * as we already know there are less extents in the temp inode.
1152 */
1153 if (ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1154 tifp->if_format == XFS_DINODE_FMT_BTREE)
1155 return -EINVAL;
1156
1157 /* Check temp in extent form to max in target */
1158 if (tifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1159 tifp->if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1160 return -EINVAL;
1161
1162 /* Check target in extent form to max in temp */
1163 if (ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1164 ifp->if_nextents > XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1165 return -EINVAL;
1166
1167 /*
1168 * If we are in a btree format, check that the temp root block will fit
1169 * in the target and that it has enough extents to be in btree format
1170 * in the target.
1171 *
1172 * Note that we have to be careful to allow btree->extent conversions
1173 * (a common defrag case) which will occur when the temp inode is in
1174 * extent format...
1175 */
1176 if (tifp->if_format == XFS_DINODE_FMT_BTREE) {
1177 if (xfs_inode_has_attr_fork(ip) &&
1178 XFS_BMAP_BMDR_SPACE(tifp->if_broot) > xfs_inode_fork_boff(ip))
1179 return -EINVAL;
1180 if (tifp->if_nextents <= XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1181 return -EINVAL;
1182 }
1183
1184 /* Reciprocal target->temp btree format checks */
1185 if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
1186 if (xfs_inode_has_attr_fork(tip) &&
1187 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > xfs_inode_fork_boff(tip))
1188 return -EINVAL;
1189 if (ifp->if_nextents <= XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1190 return -EINVAL;
1191 }
1192
1193 return 0;
1194}
1195
1196static int
1197xfs_swap_extent_flush(
1198 struct xfs_inode *ip)
1199{
1200 int error;
1201
1202 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1203 if (error)
1204 return error;
1205 truncate_pagecache_range(VFS_I(ip), 0, -1);
1206
1207 /* Verify O_DIRECT for ftmp */
1208 if (VFS_I(ip)->i_mapping->nrpages)
1209 return -EINVAL;
1210 return 0;
1211}
1212
1213/*
1214 * Move extents from one file to another, when rmap is enabled.
1215 */
1216STATIC int
1217xfs_swap_extent_rmap(
1218 struct xfs_trans **tpp,
1219 struct xfs_inode *ip,
1220 struct xfs_inode *tip)
1221{
1222 struct xfs_trans *tp = *tpp;
1223 struct xfs_bmbt_irec irec;
1224 struct xfs_bmbt_irec uirec;
1225 struct xfs_bmbt_irec tirec;
1226 xfs_fileoff_t offset_fsb;
1227 xfs_fileoff_t end_fsb;
1228 xfs_filblks_t count_fsb;
1229 int error;
1230 xfs_filblks_t ilen;
1231 xfs_filblks_t rlen;
1232 int nimaps;
1233 uint64_t tip_flags2;
1234
1235 /*
1236 * If the source file has shared blocks, we must flag the donor
1237 * file as having shared blocks so that we get the shared-block
1238 * rmap functions when we go to fix up the rmaps. The flags
1239 * will be switch for reals later.
1240 */
1241 tip_flags2 = tip->i_diflags2;
1242 if (ip->i_diflags2 & XFS_DIFLAG2_REFLINK)
1243 tip->i_diflags2 |= XFS_DIFLAG2_REFLINK;
1244
1245 offset_fsb = 0;
1246 end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip)));
1247 count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
1248
1249 while (count_fsb) {
1250 /* Read extent from the donor file */
1251 nimaps = 1;
1252 error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec,
1253 &nimaps, 0);
1254 if (error)
1255 goto out;
1256 ASSERT(nimaps == 1);
1257 ASSERT(tirec.br_startblock != DELAYSTARTBLOCK);
1258
1259 trace_xfs_swap_extent_rmap_remap(tip, &tirec);
1260 ilen = tirec.br_blockcount;
1261
1262 /* Unmap the old blocks in the source file. */
1263 while (tirec.br_blockcount) {
1264 ASSERT(tp->t_highest_agno == NULLAGNUMBER);
1265 trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec);
1266
1267 /* Read extent from the source file */
1268 nimaps = 1;
1269 error = xfs_bmapi_read(ip, tirec.br_startoff,
1270 tirec.br_blockcount, &irec,
1271 &nimaps, 0);
1272 if (error)
1273 goto out;
1274 ASSERT(nimaps == 1);
1275 ASSERT(tirec.br_startoff == irec.br_startoff);
1276 trace_xfs_swap_extent_rmap_remap_piece(ip, &irec);
1277
1278 /* Trim the extent. */
1279 uirec = tirec;
1280 uirec.br_blockcount = rlen = min_t(xfs_filblks_t,
1281 tirec.br_blockcount,
1282 irec.br_blockcount);
1283 trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec);
1284
1285 if (xfs_bmap_is_real_extent(&uirec)) {
1286 error = xfs_iext_count_may_overflow(ip,
1287 XFS_DATA_FORK,
1288 XFS_IEXT_SWAP_RMAP_CNT);
1289 if (error == -EFBIG)
1290 error = xfs_iext_count_upgrade(tp, ip,
1291 XFS_IEXT_SWAP_RMAP_CNT);
1292 if (error)
1293 goto out;
1294 }
1295
1296 if (xfs_bmap_is_real_extent(&irec)) {
1297 error = xfs_iext_count_may_overflow(tip,
1298 XFS_DATA_FORK,
1299 XFS_IEXT_SWAP_RMAP_CNT);
1300 if (error == -EFBIG)
1301 error = xfs_iext_count_upgrade(tp, ip,
1302 XFS_IEXT_SWAP_RMAP_CNT);
1303 if (error)
1304 goto out;
1305 }
1306
1307 /* Remove the mapping from the donor file. */
1308 xfs_bmap_unmap_extent(tp, tip, XFS_DATA_FORK, &uirec);
1309
1310 /* Remove the mapping from the source file. */
1311 xfs_bmap_unmap_extent(tp, ip, XFS_DATA_FORK, &irec);
1312
1313 /* Map the donor file's blocks into the source file. */
1314 xfs_bmap_map_extent(tp, ip, XFS_DATA_FORK, &uirec);
1315
1316 /* Map the source file's blocks into the donor file. */
1317 xfs_bmap_map_extent(tp, tip, XFS_DATA_FORK, &irec);
1318
1319 error = xfs_defer_finish(tpp);
1320 tp = *tpp;
1321 if (error)
1322 goto out;
1323
1324 tirec.br_startoff += rlen;
1325 if (tirec.br_startblock != HOLESTARTBLOCK &&
1326 tirec.br_startblock != DELAYSTARTBLOCK)
1327 tirec.br_startblock += rlen;
1328 tirec.br_blockcount -= rlen;
1329 }
1330
1331 /* Roll on... */
1332 count_fsb -= ilen;
1333 offset_fsb += ilen;
1334 }
1335
1336 tip->i_diflags2 = tip_flags2;
1337 return 0;
1338
1339out:
1340 trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_);
1341 tip->i_diflags2 = tip_flags2;
1342 return error;
1343}
1344
1345/* Swap the extents of two files by swapping data forks. */
1346STATIC int
1347xfs_swap_extent_forks(
1348 struct xfs_trans *tp,
1349 struct xfs_inode *ip,
1350 struct xfs_inode *tip,
1351 int *src_log_flags,
1352 int *target_log_flags)
1353{
1354 xfs_filblks_t aforkblks = 0;
1355 xfs_filblks_t taforkblks = 0;
1356 xfs_extnum_t junk;
1357 uint64_t tmp;
1358 int error;
1359
1360 /*
1361 * Count the number of extended attribute blocks
1362 */
1363 if (xfs_inode_has_attr_fork(ip) && ip->i_af.if_nextents > 0 &&
1364 ip->i_af.if_format != XFS_DINODE_FMT_LOCAL) {
1365 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &junk,
1366 &aforkblks);
1367 if (error)
1368 return error;
1369 }
1370 if (xfs_inode_has_attr_fork(tip) && tip->i_af.if_nextents > 0 &&
1371 tip->i_af.if_format != XFS_DINODE_FMT_LOCAL) {
1372 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK, &junk,
1373 &taforkblks);
1374 if (error)
1375 return error;
1376 }
1377
1378 /*
1379 * Btree format (v3) inodes have the inode number stamped in the bmbt
1380 * block headers. We can't start changing the bmbt blocks until the
1381 * inode owner change is logged so recovery does the right thing in the
1382 * event of a crash. Set the owner change log flags now and leave the
1383 * bmbt scan as the last step.
1384 */
1385 if (xfs_has_v3inodes(ip->i_mount)) {
1386 if (ip->i_df.if_format == XFS_DINODE_FMT_BTREE)
1387 (*target_log_flags) |= XFS_ILOG_DOWNER;
1388 if (tip->i_df.if_format == XFS_DINODE_FMT_BTREE)
1389 (*src_log_flags) |= XFS_ILOG_DOWNER;
1390 }
1391
1392 /*
1393 * Swap the data forks of the inodes
1394 */
1395 swap(ip->i_df, tip->i_df);
1396
1397 /*
1398 * Fix the on-disk inode values
1399 */
1400 tmp = (uint64_t)ip->i_nblocks;
1401 ip->i_nblocks = tip->i_nblocks - taforkblks + aforkblks;
1402 tip->i_nblocks = tmp + taforkblks - aforkblks;
1403
1404 /*
1405 * The extents in the source inode could still contain speculative
1406 * preallocation beyond EOF (e.g. the file is open but not modified
1407 * while defrag is in progress). In that case, we need to copy over the
1408 * number of delalloc blocks the data fork in the source inode is
1409 * tracking beyond EOF so that when the fork is truncated away when the
1410 * temporary inode is unlinked we don't underrun the i_delayed_blks
1411 * counter on that inode.
1412 */
1413 ASSERT(tip->i_delayed_blks == 0);
1414 tip->i_delayed_blks = ip->i_delayed_blks;
1415 ip->i_delayed_blks = 0;
1416
1417 switch (ip->i_df.if_format) {
1418 case XFS_DINODE_FMT_EXTENTS:
1419 (*src_log_flags) |= XFS_ILOG_DEXT;
1420 break;
1421 case XFS_DINODE_FMT_BTREE:
1422 ASSERT(!xfs_has_v3inodes(ip->i_mount) ||
1423 (*src_log_flags & XFS_ILOG_DOWNER));
1424 (*src_log_flags) |= XFS_ILOG_DBROOT;
1425 break;
1426 }
1427
1428 switch (tip->i_df.if_format) {
1429 case XFS_DINODE_FMT_EXTENTS:
1430 (*target_log_flags) |= XFS_ILOG_DEXT;
1431 break;
1432 case XFS_DINODE_FMT_BTREE:
1433 (*target_log_flags) |= XFS_ILOG_DBROOT;
1434 ASSERT(!xfs_has_v3inodes(ip->i_mount) ||
1435 (*target_log_flags & XFS_ILOG_DOWNER));
1436 break;
1437 }
1438
1439 return 0;
1440}
1441
1442/*
1443 * Fix up the owners of the bmbt blocks to refer to the current inode. The
1444 * change owner scan attempts to order all modified buffers in the current
1445 * transaction. In the event of ordered buffer failure, the offending buffer is
1446 * physically logged as a fallback and the scan returns -EAGAIN. We must roll
1447 * the transaction in this case to replenish the fallback log reservation and
1448 * restart the scan. This process repeats until the scan completes.
1449 */
1450static int
1451xfs_swap_change_owner(
1452 struct xfs_trans **tpp,
1453 struct xfs_inode *ip,
1454 struct xfs_inode *tmpip)
1455{
1456 int error;
1457 struct xfs_trans *tp = *tpp;
1458
1459 do {
1460 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK, ip->i_ino,
1461 NULL);
1462 /* success or fatal error */
1463 if (error != -EAGAIN)
1464 break;
1465
1466 error = xfs_trans_roll(tpp);
1467 if (error)
1468 break;
1469 tp = *tpp;
1470
1471 /*
1472 * Redirty both inodes so they can relog and keep the log tail
1473 * moving forward.
1474 */
1475 xfs_trans_ijoin(tp, ip, 0);
1476 xfs_trans_ijoin(tp, tmpip, 0);
1477 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1478 xfs_trans_log_inode(tp, tmpip, XFS_ILOG_CORE);
1479 } while (true);
1480
1481 return error;
1482}
1483
1484int
1485xfs_swap_extents(
1486 struct xfs_inode *ip, /* target inode */
1487 struct xfs_inode *tip, /* tmp inode */
1488 struct xfs_swapext *sxp)
1489{
1490 struct xfs_mount *mp = ip->i_mount;
1491 struct xfs_trans *tp;
1492 struct xfs_bstat *sbp = &sxp->sx_stat;
1493 int src_log_flags, target_log_flags;
1494 int error = 0;
1495 uint64_t f;
1496 int resblks = 0;
1497 unsigned int flags = 0;
1498 struct timespec64 ctime, mtime;
1499
1500 /*
1501 * Lock the inodes against other IO, page faults and truncate to
1502 * begin with. Then we can ensure the inodes are flushed and have no
1503 * page cache safely. Once we have done this we can take the ilocks and
1504 * do the rest of the checks.
1505 */
1506 lock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1507 filemap_invalidate_lock_two(VFS_I(ip)->i_mapping,
1508 VFS_I(tip)->i_mapping);
1509
1510 /* Verify that both files have the same format */
1511 if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) {
1512 error = -EINVAL;
1513 goto out_unlock;
1514 }
1515
1516 /* Verify both files are either real-time or non-realtime */
1517 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1518 error = -EINVAL;
1519 goto out_unlock;
1520 }
1521
1522 error = xfs_qm_dqattach(ip);
1523 if (error)
1524 goto out_unlock;
1525
1526 error = xfs_qm_dqattach(tip);
1527 if (error)
1528 goto out_unlock;
1529
1530 error = xfs_swap_extent_flush(ip);
1531 if (error)
1532 goto out_unlock;
1533 error = xfs_swap_extent_flush(tip);
1534 if (error)
1535 goto out_unlock;
1536
1537 if (xfs_inode_has_cow_data(tip)) {
1538 error = xfs_reflink_cancel_cow_range(tip, 0, NULLFILEOFF, true);
1539 if (error)
1540 goto out_unlock;
1541 }
1542
1543 /*
1544 * Extent "swapping" with rmap requires a permanent reservation and
1545 * a block reservation because it's really just a remap operation
1546 * performed with log redo items!
1547 */
1548 if (xfs_has_rmapbt(mp)) {
1549 int w = XFS_DATA_FORK;
1550 uint32_t ipnext = ip->i_df.if_nextents;
1551 uint32_t tipnext = tip->i_df.if_nextents;
1552
1553 /*
1554 * Conceptually this shouldn't affect the shape of either bmbt,
1555 * but since we atomically move extents one by one, we reserve
1556 * enough space to rebuild both trees.
1557 */
1558 resblks = XFS_SWAP_RMAP_SPACE_RES(mp, ipnext, w);
1559 resblks += XFS_SWAP_RMAP_SPACE_RES(mp, tipnext, w);
1560
1561 /*
1562 * If either inode straddles a bmapbt block allocation boundary,
1563 * the rmapbt algorithm triggers repeated allocs and frees as
1564 * extents are remapped. This can exhaust the block reservation
1565 * prematurely and cause shutdown. Return freed blocks to the
1566 * transaction reservation to counter this behavior.
1567 */
1568 flags |= XFS_TRANS_RES_FDBLKS;
1569 }
1570 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, flags,
1571 &tp);
1572 if (error)
1573 goto out_unlock;
1574
1575 /*
1576 * Lock and join the inodes to the tansaction so that transaction commit
1577 * or cancel will unlock the inodes from this point onwards.
1578 */
1579 xfs_lock_two_inodes(ip, XFS_ILOCK_EXCL, tip, XFS_ILOCK_EXCL);
1580 xfs_trans_ijoin(tp, ip, 0);
1581 xfs_trans_ijoin(tp, tip, 0);
1582
1583
1584 /* Verify all data are being swapped */
1585 if (sxp->sx_offset != 0 ||
1586 sxp->sx_length != ip->i_disk_size ||
1587 sxp->sx_length != tip->i_disk_size) {
1588 error = -EFAULT;
1589 goto out_trans_cancel;
1590 }
1591
1592 trace_xfs_swap_extent_before(ip, 0);
1593 trace_xfs_swap_extent_before(tip, 1);
1594
1595 /* check inode formats now that data is flushed */
1596 error = xfs_swap_extents_check_format(ip, tip);
1597 if (error) {
1598 xfs_notice(mp,
1599 "%s: inode 0x%llx format is incompatible for exchanging.",
1600 __func__, ip->i_ino);
1601 goto out_trans_cancel;
1602 }
1603
1604 /*
1605 * Compare the current change & modify times with that
1606 * passed in. If they differ, we abort this swap.
1607 * This is the mechanism used to ensure the calling
1608 * process that the file was not changed out from
1609 * under it.
1610 */
1611 ctime = inode_get_ctime(VFS_I(ip));
1612 mtime = inode_get_mtime(VFS_I(ip));
1613 if ((sbp->bs_ctime.tv_sec != ctime.tv_sec) ||
1614 (sbp->bs_ctime.tv_nsec != ctime.tv_nsec) ||
1615 (sbp->bs_mtime.tv_sec != mtime.tv_sec) ||
1616 (sbp->bs_mtime.tv_nsec != mtime.tv_nsec)) {
1617 error = -EBUSY;
1618 goto out_trans_cancel;
1619 }
1620
1621 /*
1622 * Note the trickiness in setting the log flags - we set the owner log
1623 * flag on the opposite inode (i.e. the inode we are setting the new
1624 * owner to be) because once we swap the forks and log that, log
1625 * recovery is going to see the fork as owned by the swapped inode,
1626 * not the pre-swapped inodes.
1627 */
1628 src_log_flags = XFS_ILOG_CORE;
1629 target_log_flags = XFS_ILOG_CORE;
1630
1631 if (xfs_has_rmapbt(mp))
1632 error = xfs_swap_extent_rmap(&tp, ip, tip);
1633 else
1634 error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags,
1635 &target_log_flags);
1636 if (error)
1637 goto out_trans_cancel;
1638
1639 /* Do we have to swap reflink flags? */
1640 if ((ip->i_diflags2 & XFS_DIFLAG2_REFLINK) ^
1641 (tip->i_diflags2 & XFS_DIFLAG2_REFLINK)) {
1642 f = ip->i_diflags2 & XFS_DIFLAG2_REFLINK;
1643 ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1644 ip->i_diflags2 |= tip->i_diflags2 & XFS_DIFLAG2_REFLINK;
1645 tip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1646 tip->i_diflags2 |= f & XFS_DIFLAG2_REFLINK;
1647 }
1648
1649 /* Swap the cow forks. */
1650 if (xfs_has_reflink(mp)) {
1651 ASSERT(!ip->i_cowfp ||
1652 ip->i_cowfp->if_format == XFS_DINODE_FMT_EXTENTS);
1653 ASSERT(!tip->i_cowfp ||
1654 tip->i_cowfp->if_format == XFS_DINODE_FMT_EXTENTS);
1655
1656 swap(ip->i_cowfp, tip->i_cowfp);
1657
1658 if (ip->i_cowfp && ip->i_cowfp->if_bytes)
1659 xfs_inode_set_cowblocks_tag(ip);
1660 else
1661 xfs_inode_clear_cowblocks_tag(ip);
1662 if (tip->i_cowfp && tip->i_cowfp->if_bytes)
1663 xfs_inode_set_cowblocks_tag(tip);
1664 else
1665 xfs_inode_clear_cowblocks_tag(tip);
1666 }
1667
1668 xfs_trans_log_inode(tp, ip, src_log_flags);
1669 xfs_trans_log_inode(tp, tip, target_log_flags);
1670
1671 /*
1672 * The extent forks have been swapped, but crc=1,rmapbt=0 filesystems
1673 * have inode number owner values in the bmbt blocks that still refer to
1674 * the old inode. Scan each bmbt to fix up the owner values with the
1675 * inode number of the current inode.
1676 */
1677 if (src_log_flags & XFS_ILOG_DOWNER) {
1678 error = xfs_swap_change_owner(&tp, ip, tip);
1679 if (error)
1680 goto out_trans_cancel;
1681 }
1682 if (target_log_flags & XFS_ILOG_DOWNER) {
1683 error = xfs_swap_change_owner(&tp, tip, ip);
1684 if (error)
1685 goto out_trans_cancel;
1686 }
1687
1688 /*
1689 * If this is a synchronous mount, make sure that the
1690 * transaction goes to disk before returning to the user.
1691 */
1692 if (xfs_has_wsync(mp))
1693 xfs_trans_set_sync(tp);
1694
1695 error = xfs_trans_commit(tp);
1696
1697 trace_xfs_swap_extent_after(ip, 0);
1698 trace_xfs_swap_extent_after(tip, 1);
1699
1700out_unlock_ilock:
1701 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1702 xfs_iunlock(tip, XFS_ILOCK_EXCL);
1703out_unlock:
1704 filemap_invalidate_unlock_two(VFS_I(ip)->i_mapping,
1705 VFS_I(tip)->i_mapping);
1706 unlock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1707 return error;
1708
1709out_trans_cancel:
1710 xfs_trans_cancel(tp);
1711 goto out_unlock_ilock;
1712}