blob: 541d59f5e65822270fa225c93aa94bbf9a687b06 [file] [log] [blame]
Dave Chinner68988112013-08-12 20:49:42 +10001/*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
Dave Chinnerc24b5df2013-08-12 20:49:45 +10003 * Copyright (c) 2012 Red Hat, Inc.
Dave Chinner68988112013-08-12 20:49:42 +10004 * 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_format.h"
22#include "xfs_bit.h"
23#include "xfs_log.h"
24#include "xfs_inum.h"
25#include "xfs_trans.h"
26#include "xfs_sb.h"
27#include "xfs_ag.h"
28#include "xfs_mount.h"
29#include "xfs_da_btree.h"
30#include "xfs_bmap_btree.h"
31#include "xfs_alloc_btree.h"
32#include "xfs_ialloc_btree.h"
33#include "xfs_dinode.h"
34#include "xfs_inode.h"
35#include "xfs_btree.h"
36#include "xfs_extfree_item.h"
37#include "xfs_alloc.h"
38#include "xfs_bmap.h"
39#include "xfs_bmap_util.h"
40#include "xfs_rtalloc.h"
41#include "xfs_error.h"
42#include "xfs_quota.h"
43#include "xfs_trans_space.h"
44#include "xfs_trace.h"
Dave Chinnerc24b5df2013-08-12 20:49:45 +100045#include "xfs_icache.h"
Dave Chinner68988112013-08-12 20:49:42 +100046
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 */
Jie Liu3d3c8b52013-08-12 20:49:59 +100081 struct xfs_trans_res tres; /* new log reservation */
Dave Chinner68988112013-08-12 20:49:42 +100082 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);
Jie Liu3d3c8b52013-08-12 20:49:59 +100096
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;
Dave Chinner68988112013-08-12 20:49:42 +1000100 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
Jie Liu3d3c8b52013-08-12 20:49:59 +1000117 error = xfs_trans_reserve(ntp, &tres, 0, 0);
118 if (error)
Dave Chinner68988112013-08-12 20:49:42 +1000119 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 return args->result;
291}
292
293/*
294 * Check if the endoff is outside the last extent. If so the caller will grow
295 * the allocation to a stripe unit boundary. All offsets are considered outside
296 * the end of file for an empty fork, so 1 is returned in *eof in that case.
297 */
298int
299xfs_bmap_eof(
300 struct xfs_inode *ip,
301 xfs_fileoff_t endoff,
302 int whichfork,
303 int *eof)
304{
305 struct xfs_bmbt_irec rec;
306 int error;
307
308 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
309 if (error || *eof)
310 return error;
311
312 *eof = endoff >= rec.br_startoff + rec.br_blockcount;
313 return 0;
314}
315
316/*
317 * Extent tree block counting routines.
318 */
319
320/*
321 * Count leaf blocks given a range of extent records.
322 */
323STATIC void
324xfs_bmap_count_leaves(
325 xfs_ifork_t *ifp,
326 xfs_extnum_t idx,
327 int numrecs,
328 int *count)
329{
330 int b;
331
332 for (b = 0; b < numrecs; b++) {
333 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
334 *count += xfs_bmbt_get_blockcount(frp);
335 }
336}
337
338/*
339 * Count leaf blocks given a range of extent records originally
340 * in btree format.
341 */
342STATIC void
343xfs_bmap_disk_count_leaves(
344 struct xfs_mount *mp,
345 struct xfs_btree_block *block,
346 int numrecs,
347 int *count)
348{
349 int b;
350 xfs_bmbt_rec_t *frp;
351
352 for (b = 1; b <= numrecs; b++) {
353 frp = XFS_BMBT_REC_ADDR(mp, block, b);
354 *count += xfs_bmbt_disk_get_blockcount(frp);
355 }
356}
357
358/*
359 * Recursively walks each level of a btree
Zhi Yong Wu8be11e92013-08-12 03:14:52 +0000360 * to count total fsblocks in use.
Dave Chinner68988112013-08-12 20:49:42 +1000361 */
362STATIC int /* error */
363xfs_bmap_count_tree(
364 xfs_mount_t *mp, /* file system mount point */
365 xfs_trans_t *tp, /* transaction pointer */
366 xfs_ifork_t *ifp, /* inode fork pointer */
367 xfs_fsblock_t blockno, /* file system block number */
368 int levelin, /* level in btree */
369 int *count) /* Count of blocks */
370{
371 int error;
372 xfs_buf_t *bp, *nbp;
373 int level = levelin;
374 __be64 *pp;
375 xfs_fsblock_t bno = blockno;
376 xfs_fsblock_t nextbno;
377 struct xfs_btree_block *block, *nextblock;
378 int numrecs;
379
380 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF,
381 &xfs_bmbt_buf_ops);
382 if (error)
383 return error;
384 *count += 1;
385 block = XFS_BUF_TO_BLOCK(bp);
386
387 if (--level) {
388 /* Not at node above leaves, count this level of nodes */
389 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
390 while (nextbno != NULLFSBLOCK) {
391 error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp,
392 XFS_BMAP_BTREE_REF,
393 &xfs_bmbt_buf_ops);
394 if (error)
395 return error;
396 *count += 1;
397 nextblock = XFS_BUF_TO_BLOCK(nbp);
398 nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
399 xfs_trans_brelse(tp, nbp);
400 }
401
402 /* Dive to the next level */
403 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
404 bno = be64_to_cpu(*pp);
405 if (unlikely((error =
406 xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
407 xfs_trans_brelse(tp, bp);
408 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
409 XFS_ERRLEVEL_LOW, mp);
410 return XFS_ERROR(EFSCORRUPTED);
411 }
412 xfs_trans_brelse(tp, bp);
413 } else {
414 /* count all level 1 nodes and their leaves */
415 for (;;) {
416 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
417 numrecs = be16_to_cpu(block->bb_numrecs);
418 xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
419 xfs_trans_brelse(tp, bp);
420 if (nextbno == NULLFSBLOCK)
421 break;
422 bno = nextbno;
423 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
424 XFS_BMAP_BTREE_REF,
425 &xfs_bmbt_buf_ops);
426 if (error)
427 return error;
428 *count += 1;
429 block = XFS_BUF_TO_BLOCK(bp);
430 }
431 }
432 return 0;
433}
434
435/*
436 * Count fsblocks of the given fork.
437 */
438int /* error */
439xfs_bmap_count_blocks(
440 xfs_trans_t *tp, /* transaction pointer */
441 xfs_inode_t *ip, /* incore inode */
442 int whichfork, /* data or attr fork */
443 int *count) /* out: count of blocks */
444{
445 struct xfs_btree_block *block; /* current btree block */
446 xfs_fsblock_t bno; /* block # of "block" */
447 xfs_ifork_t *ifp; /* fork structure */
448 int level; /* btree level, for checking */
449 xfs_mount_t *mp; /* file system mount structure */
450 __be64 *pp; /* pointer to block address */
451
452 bno = NULLFSBLOCK;
453 mp = ip->i_mount;
454 ifp = XFS_IFORK_PTR(ip, whichfork);
455 if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
456 xfs_bmap_count_leaves(ifp, 0,
457 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
458 count);
459 return 0;
460 }
461
462 /*
463 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
464 */
465 block = ifp->if_broot;
466 level = be16_to_cpu(block->bb_level);
467 ASSERT(level > 0);
468 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
469 bno = be64_to_cpu(*pp);
470 ASSERT(bno != NULLDFSBNO);
471 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
472 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
473
474 if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
475 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
476 mp);
477 return XFS_ERROR(EFSCORRUPTED);
478 }
479
480 return 0;
481}
482
483/*
484 * returns 1 for success, 0 if we failed to map the extent.
485 */
486STATIC int
487xfs_getbmapx_fix_eof_hole(
488 xfs_inode_t *ip, /* xfs incore inode pointer */
489 struct getbmapx *out, /* output structure */
490 int prealloced, /* this is a file with
491 * preallocated data space */
492 __int64_t end, /* last block requested */
493 xfs_fsblock_t startblock)
494{
495 __int64_t fixlen;
496 xfs_mount_t *mp; /* file system mount point */
497 xfs_ifork_t *ifp; /* inode fork pointer */
498 xfs_extnum_t lastx; /* last extent pointer */
499 xfs_fileoff_t fileblock;
500
501 if (startblock == HOLESTARTBLOCK) {
502 mp = ip->i_mount;
503 out->bmv_block = -1;
504 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
505 fixlen -= out->bmv_offset;
506 if (prealloced && out->bmv_offset + out->bmv_length == end) {
507 /* Came to hole at EOF. Trim it. */
508 if (fixlen <= 0)
509 return 0;
510 out->bmv_length = fixlen;
511 }
512 } else {
513 if (startblock == DELAYSTARTBLOCK)
514 out->bmv_block = -2;
515 else
516 out->bmv_block = xfs_fsb_to_db(ip, startblock);
517 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
518 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
519 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
520 (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
521 out->bmv_oflags |= BMV_OF_LAST;
522 }
523
524 return 1;
525}
526
527/*
528 * Get inode's extents as described in bmv, and format for output.
529 * Calls formatter to fill the user's buffer until all extents
530 * are mapped, until the passed-in bmv->bmv_count slots have
531 * been filled, or until the formatter short-circuits the loop,
532 * if it is tracking filled-in extents on its own.
533 */
534int /* error code */
535xfs_getbmap(
536 xfs_inode_t *ip,
537 struct getbmapx *bmv, /* user bmap structure */
538 xfs_bmap_format_t formatter, /* format to user */
539 void *arg) /* formatter arg */
540{
541 __int64_t bmvend; /* last block requested */
542 int error = 0; /* return value */
543 __int64_t fixlen; /* length for -1 case */
544 int i; /* extent number */
545 int lock; /* lock state */
546 xfs_bmbt_irec_t *map; /* buffer for user's data */
547 xfs_mount_t *mp; /* file system mount point */
548 int nex; /* # of user extents can do */
549 int nexleft; /* # of user extents left */
550 int subnex; /* # of bmapi's can do */
551 int nmap; /* number of map entries */
552 struct getbmapx *out; /* output structure */
553 int whichfork; /* data or attr fork */
554 int prealloced; /* this is a file with
555 * preallocated data space */
556 int iflags; /* interface flags */
557 int bmapi_flags; /* flags for xfs_bmapi */
558 int cur_ext = 0;
559
560 mp = ip->i_mount;
561 iflags = bmv->bmv_iflags;
562 whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
563
564 if (whichfork == XFS_ATTR_FORK) {
565 if (XFS_IFORK_Q(ip)) {
566 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
567 ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
568 ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
569 return XFS_ERROR(EINVAL);
570 } else if (unlikely(
571 ip->i_d.di_aformat != 0 &&
572 ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
573 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
574 ip->i_mount);
575 return XFS_ERROR(EFSCORRUPTED);
576 }
577
578 prealloced = 0;
579 fixlen = 1LL << 32;
580 } else {
581 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
582 ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
583 ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
584 return XFS_ERROR(EINVAL);
585
586 if (xfs_get_extsz_hint(ip) ||
587 ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
588 prealloced = 1;
589 fixlen = mp->m_super->s_maxbytes;
590 } else {
591 prealloced = 0;
592 fixlen = XFS_ISIZE(ip);
593 }
594 }
595
596 if (bmv->bmv_length == -1) {
597 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
598 bmv->bmv_length =
599 max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
600 } else if (bmv->bmv_length == 0) {
601 bmv->bmv_entries = 0;
602 return 0;
603 } else if (bmv->bmv_length < 0) {
604 return XFS_ERROR(EINVAL);
605 }
606
607 nex = bmv->bmv_count - 1;
608 if (nex <= 0)
609 return XFS_ERROR(EINVAL);
610 bmvend = bmv->bmv_offset + bmv->bmv_length;
611
612
613 if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
614 return XFS_ERROR(ENOMEM);
615 out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
616 if (!out) {
617 out = kmem_zalloc_large(bmv->bmv_count *
618 sizeof(struct getbmapx));
619 if (!out)
620 return XFS_ERROR(ENOMEM);
621 }
622
623 xfs_ilock(ip, XFS_IOLOCK_SHARED);
624 if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
625 if (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size) {
626 error = -filemap_write_and_wait(VFS_I(ip)->i_mapping);
627 if (error)
628 goto out_unlock_iolock;
629 }
630 /*
631 * even after flushing the inode, there can still be delalloc
632 * blocks on the inode beyond EOF due to speculative
633 * preallocation. These are not removed until the release
634 * function is called or the inode is inactivated. Hence we
635 * cannot assert here that ip->i_delayed_blks == 0.
636 */
637 }
638
639 lock = xfs_ilock_map_shared(ip);
640
641 /*
642 * Don't let nex be bigger than the number of extents
643 * we can have assuming alternating holes and real extents.
644 */
645 if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
646 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
647
648 bmapi_flags = xfs_bmapi_aflag(whichfork);
649 if (!(iflags & BMV_IF_PREALLOC))
650 bmapi_flags |= XFS_BMAPI_IGSTATE;
651
652 /*
653 * Allocate enough space to handle "subnex" maps at a time.
654 */
655 error = ENOMEM;
656 subnex = 16;
657 map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
658 if (!map)
659 goto out_unlock_ilock;
660
661 bmv->bmv_entries = 0;
662
663 if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
664 (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
665 error = 0;
666 goto out_free_map;
667 }
668
669 nexleft = nex;
670
671 do {
672 nmap = (nexleft > subnex) ? subnex : nexleft;
673 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
674 XFS_BB_TO_FSB(mp, bmv->bmv_length),
675 map, &nmap, bmapi_flags);
676 if (error)
677 goto out_free_map;
678 ASSERT(nmap <= subnex);
679
680 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
681 out[cur_ext].bmv_oflags = 0;
682 if (map[i].br_state == XFS_EXT_UNWRITTEN)
683 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
684 else if (map[i].br_startblock == DELAYSTARTBLOCK)
685 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
686 out[cur_ext].bmv_offset =
687 XFS_FSB_TO_BB(mp, map[i].br_startoff);
688 out[cur_ext].bmv_length =
689 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
690 out[cur_ext].bmv_unused1 = 0;
691 out[cur_ext].bmv_unused2 = 0;
692
693 /*
694 * delayed allocation extents that start beyond EOF can
695 * occur due to speculative EOF allocation when the
696 * delalloc extent is larger than the largest freespace
697 * extent at conversion time. These extents cannot be
698 * converted by data writeback, so can exist here even
699 * if we are not supposed to be finding delalloc
700 * extents.
701 */
702 if (map[i].br_startblock == DELAYSTARTBLOCK &&
703 map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip)))
704 ASSERT((iflags & BMV_IF_DELALLOC) != 0);
705
706 if (map[i].br_startblock == HOLESTARTBLOCK &&
707 whichfork == XFS_ATTR_FORK) {
708 /* came to the end of attribute fork */
709 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
710 goto out_free_map;
711 }
712
713 if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
714 prealloced, bmvend,
715 map[i].br_startblock))
716 goto out_free_map;
717
718 bmv->bmv_offset =
719 out[cur_ext].bmv_offset +
720 out[cur_ext].bmv_length;
721 bmv->bmv_length =
722 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
723
724 /*
725 * In case we don't want to return the hole,
726 * don't increase cur_ext so that we can reuse
727 * it in the next loop.
728 */
729 if ((iflags & BMV_IF_NO_HOLES) &&
730 map[i].br_startblock == HOLESTARTBLOCK) {
731 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
732 continue;
733 }
734
735 nexleft--;
736 bmv->bmv_entries++;
737 cur_ext++;
738 }
739 } while (nmap && nexleft && bmv->bmv_length);
740
741 out_free_map:
742 kmem_free(map);
743 out_unlock_ilock:
744 xfs_iunlock_map_shared(ip, lock);
745 out_unlock_iolock:
746 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
747
748 for (i = 0; i < cur_ext; i++) {
749 int full = 0; /* user array is full */
750
751 /* format results & advance arg */
752 error = formatter(&arg, &out[i], &full);
753 if (error || full)
754 break;
755 }
756
757 if (is_vmalloc_addr(out))
758 kmem_free_large(out);
759 else
760 kmem_free(out);
761 return error;
762}
763
764/*
765 * dead simple method of punching delalyed allocation blocks from a range in
766 * the inode. Walks a block at a time so will be slow, but is only executed in
Zhi Yong Wuad4809b2013-08-12 03:14:55 +0000767 * rare error cases so the overhead is not critical. This will always punch out
Dave Chinner68988112013-08-12 20:49:42 +1000768 * both the start and end blocks, even if the ranges only partially overlap
769 * them, so it is up to the caller to ensure that partial blocks are not
770 * passed in.
771 */
772int
773xfs_bmap_punch_delalloc_range(
774 struct xfs_inode *ip,
775 xfs_fileoff_t start_fsb,
776 xfs_fileoff_t length)
777{
778 xfs_fileoff_t remaining = length;
779 int error = 0;
780
781 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
782
783 do {
784 int done;
785 xfs_bmbt_irec_t imap;
786 int nimaps = 1;
787 xfs_fsblock_t firstblock;
788 xfs_bmap_free_t flist;
789
790 /*
791 * Map the range first and check that it is a delalloc extent
792 * before trying to unmap the range. Otherwise we will be
793 * trying to remove a real extent (which requires a
794 * transaction) or a hole, which is probably a bad idea...
795 */
796 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
797 XFS_BMAPI_ENTIRE);
798
799 if (error) {
800 /* something screwed, just bail */
801 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
802 xfs_alert(ip->i_mount,
803 "Failed delalloc mapping lookup ino %lld fsb %lld.",
804 ip->i_ino, start_fsb);
805 }
806 break;
807 }
808 if (!nimaps) {
809 /* nothing there */
810 goto next_block;
811 }
812 if (imap.br_startblock != DELAYSTARTBLOCK) {
813 /* been converted, ignore */
814 goto next_block;
815 }
816 WARN_ON(imap.br_blockcount == 0);
817
818 /*
819 * Note: while we initialise the firstblock/flist pair, they
820 * should never be used because blocks should never be
821 * allocated or freed for a delalloc extent and hence we need
822 * don't cancel or finish them after the xfs_bunmapi() call.
823 */
824 xfs_bmap_init(&flist, &firstblock);
825 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
826 &flist, &done);
827 if (error)
828 break;
829
830 ASSERT(!flist.xbf_count && !flist.xbf_first);
831next_block:
832 start_fsb++;
833 remaining--;
834 } while(remaining > 0);
835
836 return error;
837}
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000838
839/*
840 * Test whether it is appropriate to check an inode for and free post EOF
841 * blocks. The 'force' parameter determines whether we should also consider
842 * regular files that are marked preallocated or append-only.
843 */
844bool
845xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
846{
847 /* prealloc/delalloc exists only on regular files */
848 if (!S_ISREG(ip->i_d.di_mode))
849 return false;
850
851 /*
852 * Zero sized files with no cached pages and delalloc blocks will not
853 * have speculative prealloc/delalloc blocks to remove.
854 */
855 if (VFS_I(ip)->i_size == 0 &&
856 VN_CACHED(VFS_I(ip)) == 0 &&
857 ip->i_delayed_blks == 0)
858 return false;
859
860 /* If we haven't read in the extent list, then don't do it now. */
861 if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
862 return false;
863
864 /*
865 * Do not free real preallocated or append-only files unless the file
866 * has delalloc blocks and we are forced to remove them.
867 */
868 if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
869 if (!force || ip->i_delayed_blks == 0)
870 return false;
871
872 return true;
873}
874
875/*
876 * This is called by xfs_inactive to free any blocks beyond eof
877 * when the link count isn't zero and by xfs_dm_punch_hole() when
878 * punching a hole to EOF.
879 */
880int
881xfs_free_eofblocks(
882 xfs_mount_t *mp,
883 xfs_inode_t *ip,
884 bool need_iolock)
885{
886 xfs_trans_t *tp;
887 int error;
888 xfs_fileoff_t end_fsb;
889 xfs_fileoff_t last_fsb;
890 xfs_filblks_t map_len;
891 int nimaps;
892 xfs_bmbt_irec_t imap;
893
894 /*
895 * Figure out if there are any blocks beyond the end
896 * of the file. If not, then there is nothing to do.
897 */
898 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
899 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
900 if (last_fsb <= end_fsb)
901 return 0;
902 map_len = last_fsb - end_fsb;
903
904 nimaps = 1;
905 xfs_ilock(ip, XFS_ILOCK_SHARED);
906 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
907 xfs_iunlock(ip, XFS_ILOCK_SHARED);
908
909 if (!error && (nimaps != 0) &&
910 (imap.br_startblock != HOLESTARTBLOCK ||
911 ip->i_delayed_blks)) {
912 /*
913 * Attach the dquots to the inode up front.
914 */
915 error = xfs_qm_dqattach(ip, 0);
916 if (error)
917 return error;
918
919 /*
920 * There are blocks after the end of file.
921 * Free them up now by truncating the file to
922 * its current size.
923 */
924 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
925
926 if (need_iolock) {
927 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
928 xfs_trans_cancel(tp, 0);
929 return EAGAIN;
930 }
931 }
932
Jie Liu3d3c8b52013-08-12 20:49:59 +1000933 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000934 if (error) {
935 ASSERT(XFS_FORCED_SHUTDOWN(mp));
936 xfs_trans_cancel(tp, 0);
937 if (need_iolock)
938 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
939 return error;
940 }
941
942 xfs_ilock(ip, XFS_ILOCK_EXCL);
943 xfs_trans_ijoin(tp, ip, 0);
944
945 /*
946 * Do not update the on-disk file size. If we update the
947 * on-disk file size and then the system crashes before the
948 * contents of the file are flushed to disk then the files
949 * may be full of holes (ie NULL files bug).
950 */
951 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK,
952 XFS_ISIZE(ip));
953 if (error) {
954 /*
955 * If we get an error at this point we simply don't
956 * bother truncating the file.
957 */
958 xfs_trans_cancel(tp,
959 (XFS_TRANS_RELEASE_LOG_RES |
960 XFS_TRANS_ABORT));
961 } else {
962 error = xfs_trans_commit(tp,
963 XFS_TRANS_RELEASE_LOG_RES);
964 if (!error)
965 xfs_inode_clear_eofblocks_tag(ip);
966 }
967
968 xfs_iunlock(ip, XFS_ILOCK_EXCL);
969 if (need_iolock)
970 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
971 }
972 return error;
973}
974
975/*
976 * xfs_alloc_file_space()
977 * This routine allocates disk space for the given file.
978 *
979 * If alloc_type == 0, this request is for an ALLOCSP type
980 * request which will change the file size. In this case, no
981 * DMAPI event will be generated by the call. A TRUNCATE event
982 * will be generated later by xfs_setattr.
983 *
984 * If alloc_type != 0, this request is for a RESVSP type
985 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
986 * lower block boundary byte address is less than the file's
987 * length.
988 *
989 * RETURNS:
990 * 0 on success
991 * errno on error
992 *
993 */
994STATIC int
995xfs_alloc_file_space(
996 xfs_inode_t *ip,
997 xfs_off_t offset,
998 xfs_off_t len,
999 int alloc_type,
1000 int attr_flags)
1001{
1002 xfs_mount_t *mp = ip->i_mount;
1003 xfs_off_t count;
1004 xfs_filblks_t allocated_fsb;
1005 xfs_filblks_t allocatesize_fsb;
1006 xfs_extlen_t extsz, temp;
1007 xfs_fileoff_t startoffset_fsb;
1008 xfs_fsblock_t firstfsb;
1009 int nimaps;
1010 int quota_flag;
1011 int rt;
1012 xfs_trans_t *tp;
1013 xfs_bmbt_irec_t imaps[1], *imapp;
1014 xfs_bmap_free_t free_list;
1015 uint qblocks, resblks, resrtextents;
1016 int committed;
1017 int error;
1018
1019 trace_xfs_alloc_file_space(ip);
1020
1021 if (XFS_FORCED_SHUTDOWN(mp))
1022 return XFS_ERROR(EIO);
1023
1024 error = xfs_qm_dqattach(ip, 0);
1025 if (error)
1026 return error;
1027
1028 if (len <= 0)
1029 return XFS_ERROR(EINVAL);
1030
1031 rt = XFS_IS_REALTIME_INODE(ip);
1032 extsz = xfs_get_extsz_hint(ip);
1033
1034 count = len;
1035 imapp = &imaps[0];
1036 nimaps = 1;
1037 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
1038 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
1039
1040 /*
1041 * Allocate file space until done or until there is an error
1042 */
1043 while (allocatesize_fsb && !error) {
1044 xfs_fileoff_t s, e;
1045
1046 /*
1047 * Determine space reservations for data/realtime.
1048 */
1049 if (unlikely(extsz)) {
1050 s = startoffset_fsb;
1051 do_div(s, extsz);
1052 s *= extsz;
1053 e = startoffset_fsb + allocatesize_fsb;
1054 if ((temp = do_mod(startoffset_fsb, extsz)))
1055 e += temp;
1056 if ((temp = do_mod(e, extsz)))
1057 e += extsz - temp;
1058 } else {
1059 s = 0;
1060 e = allocatesize_fsb;
1061 }
1062
1063 /*
1064 * The transaction reservation is limited to a 32-bit block
1065 * count, hence we need to limit the number of blocks we are
1066 * trying to reserve to avoid an overflow. We can't allocate
1067 * more than @nimaps extents, and an extent is limited on disk
1068 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
1069 */
1070 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
1071 if (unlikely(rt)) {
1072 resrtextents = qblocks = resblks;
1073 resrtextents /= mp->m_sb.sb_rextsize;
1074 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1075 quota_flag = XFS_QMOPT_RES_RTBLKS;
1076 } else {
1077 resrtextents = 0;
1078 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
1079 quota_flag = XFS_QMOPT_RES_REGBLKS;
1080 }
1081
1082 /*
1083 * Allocate and setup the transaction.
1084 */
1085 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
Jie Liu3d3c8b52013-08-12 20:49:59 +10001086 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
1087 resblks, resrtextents);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001088 /*
1089 * Check for running out of space
1090 */
1091 if (error) {
1092 /*
1093 * Free the transaction structure.
1094 */
1095 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1096 xfs_trans_cancel(tp, 0);
1097 break;
1098 }
1099 xfs_ilock(ip, XFS_ILOCK_EXCL);
1100 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
1101 0, quota_flag);
1102 if (error)
1103 goto error1;
1104
1105 xfs_trans_ijoin(tp, ip, 0);
1106
1107 xfs_bmap_init(&free_list, &firstfsb);
1108 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
1109 allocatesize_fsb, alloc_type, &firstfsb,
1110 0, imapp, &nimaps, &free_list);
1111 if (error) {
1112 goto error0;
1113 }
1114
1115 /*
1116 * Complete the transaction
1117 */
1118 error = xfs_bmap_finish(&tp, &free_list, &committed);
1119 if (error) {
1120 goto error0;
1121 }
1122
1123 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1124 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1125 if (error) {
1126 break;
1127 }
1128
1129 allocated_fsb = imapp->br_blockcount;
1130
1131 if (nimaps == 0) {
1132 error = XFS_ERROR(ENOSPC);
1133 break;
1134 }
1135
1136 startoffset_fsb += allocated_fsb;
1137 allocatesize_fsb -= allocated_fsb;
1138 }
1139
1140 return error;
1141
1142error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1143 xfs_bmap_cancel(&free_list);
1144 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
1145
1146error1: /* Just cancel transaction */
1147 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1148 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1149 return error;
1150}
1151
1152/*
1153 * Zero file bytes between startoff and endoff inclusive.
1154 * The iolock is held exclusive and no blocks are buffered.
1155 *
1156 * This function is used by xfs_free_file_space() to zero
1157 * partial blocks when the range to free is not block aligned.
1158 * When unreserving space with boundaries that are not block
1159 * aligned we round up the start and round down the end
1160 * boundaries and then use this function to zero the parts of
1161 * the blocks that got dropped during the rounding.
1162 */
1163STATIC int
1164xfs_zero_remaining_bytes(
1165 xfs_inode_t *ip,
1166 xfs_off_t startoff,
1167 xfs_off_t endoff)
1168{
1169 xfs_bmbt_irec_t imap;
1170 xfs_fileoff_t offset_fsb;
1171 xfs_off_t lastoffset;
1172 xfs_off_t offset;
1173 xfs_buf_t *bp;
1174 xfs_mount_t *mp = ip->i_mount;
1175 int nimap;
1176 int error = 0;
1177
1178 /*
1179 * Avoid doing I/O beyond eof - it's not necessary
1180 * since nothing can read beyond eof. The space will
1181 * be zeroed when the file is extended anyway.
1182 */
1183 if (startoff >= XFS_ISIZE(ip))
1184 return 0;
1185
1186 if (endoff > XFS_ISIZE(ip))
1187 endoff = XFS_ISIZE(ip);
1188
1189 bp = xfs_buf_get_uncached(XFS_IS_REALTIME_INODE(ip) ?
1190 mp->m_rtdev_targp : mp->m_ddev_targp,
1191 BTOBB(mp->m_sb.sb_blocksize), 0);
1192 if (!bp)
1193 return XFS_ERROR(ENOMEM);
1194
1195 xfs_buf_unlock(bp);
1196
1197 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
1198 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1199 nimap = 1;
1200 error = xfs_bmapi_read(ip, offset_fsb, 1, &imap, &nimap, 0);
1201 if (error || nimap < 1)
1202 break;
1203 ASSERT(imap.br_blockcount >= 1);
1204 ASSERT(imap.br_startoff == offset_fsb);
1205 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
1206 if (lastoffset > endoff)
1207 lastoffset = endoff;
1208 if (imap.br_startblock == HOLESTARTBLOCK)
1209 continue;
1210 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1211 if (imap.br_state == XFS_EXT_UNWRITTEN)
1212 continue;
1213 XFS_BUF_UNDONE(bp);
1214 XFS_BUF_UNWRITE(bp);
1215 XFS_BUF_READ(bp);
1216 XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock));
1217 xfsbdstrat(mp, bp);
1218 error = xfs_buf_iowait(bp);
1219 if (error) {
1220 xfs_buf_ioerror_alert(bp,
1221 "xfs_zero_remaining_bytes(read)");
1222 break;
1223 }
1224 memset(bp->b_addr +
1225 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
1226 0, lastoffset - offset + 1);
1227 XFS_BUF_UNDONE(bp);
1228 XFS_BUF_UNREAD(bp);
1229 XFS_BUF_WRITE(bp);
1230 xfsbdstrat(mp, bp);
1231 error = xfs_buf_iowait(bp);
1232 if (error) {
1233 xfs_buf_ioerror_alert(bp,
1234 "xfs_zero_remaining_bytes(write)");
1235 break;
1236 }
1237 }
1238 xfs_buf_free(bp);
1239 return error;
1240}
1241
1242/*
1243 * xfs_free_file_space()
1244 * This routine frees disk space for the given file.
1245 *
1246 * This routine is only called by xfs_change_file_space
1247 * for an UNRESVSP type call.
1248 *
1249 * RETURNS:
1250 * 0 on success
1251 * errno on error
1252 *
1253 */
1254STATIC int
1255xfs_free_file_space(
1256 xfs_inode_t *ip,
1257 xfs_off_t offset,
1258 xfs_off_t len,
1259 int attr_flags)
1260{
1261 int committed;
1262 int done;
1263 xfs_fileoff_t endoffset_fsb;
1264 int error;
1265 xfs_fsblock_t firstfsb;
1266 xfs_bmap_free_t free_list;
1267 xfs_bmbt_irec_t imap;
1268 xfs_off_t ioffset;
1269 xfs_extlen_t mod=0;
1270 xfs_mount_t *mp;
1271 int nimap;
1272 uint resblks;
1273 xfs_off_t rounding;
1274 int rt;
1275 xfs_fileoff_t startoffset_fsb;
1276 xfs_trans_t *tp;
1277 int need_iolock = 1;
1278
1279 mp = ip->i_mount;
1280
1281 trace_xfs_free_file_space(ip);
1282
1283 error = xfs_qm_dqattach(ip, 0);
1284 if (error)
1285 return error;
1286
1287 error = 0;
1288 if (len <= 0) /* if nothing being freed */
1289 return error;
1290 rt = XFS_IS_REALTIME_INODE(ip);
1291 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
1292 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
1293
1294 if (attr_flags & XFS_ATTR_NOLOCK)
1295 need_iolock = 0;
1296 if (need_iolock) {
1297 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1298 /* wait for the completion of any pending DIOs */
1299 inode_dio_wait(VFS_I(ip));
1300 }
1301
1302 rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
1303 ioffset = offset & ~(rounding - 1);
1304 error = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1305 ioffset, -1);
1306 if (error)
1307 goto out_unlock_iolock;
1308 truncate_pagecache_range(VFS_I(ip), ioffset, -1);
1309
1310 /*
1311 * Need to zero the stuff we're not freeing, on disk.
1312 * If it's a realtime file & can't use unwritten extents then we
1313 * actually need to zero the extent edges. Otherwise xfs_bunmapi
1314 * will take care of it for us.
1315 */
1316 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
1317 nimap = 1;
1318 error = xfs_bmapi_read(ip, startoffset_fsb, 1,
1319 &imap, &nimap, 0);
1320 if (error)
1321 goto out_unlock_iolock;
1322 ASSERT(nimap == 0 || nimap == 1);
1323 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1324 xfs_daddr_t block;
1325
1326 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1327 block = imap.br_startblock;
1328 mod = do_div(block, mp->m_sb.sb_rextsize);
1329 if (mod)
1330 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
1331 }
1332 nimap = 1;
1333 error = xfs_bmapi_read(ip, endoffset_fsb - 1, 1,
1334 &imap, &nimap, 0);
1335 if (error)
1336 goto out_unlock_iolock;
1337 ASSERT(nimap == 0 || nimap == 1);
1338 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1339 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1340 mod++;
1341 if (mod && (mod != mp->m_sb.sb_rextsize))
1342 endoffset_fsb -= mod;
1343 }
1344 }
1345 if ((done = (endoffset_fsb <= startoffset_fsb)))
1346 /*
1347 * One contiguous piece to clear
1348 */
1349 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
1350 else {
1351 /*
1352 * Some full blocks, possibly two pieces to clear
1353 */
1354 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
1355 error = xfs_zero_remaining_bytes(ip, offset,
1356 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
1357 if (!error &&
1358 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
1359 error = xfs_zero_remaining_bytes(ip,
1360 XFS_FSB_TO_B(mp, endoffset_fsb),
1361 offset + len - 1);
1362 }
1363
1364 /*
1365 * free file space until done or until there is an error
1366 */
1367 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1368 while (!error && !done) {
1369
1370 /*
1371 * allocate and setup the transaction. Allow this
1372 * transaction to dip into the reserve blocks to ensure
1373 * the freeing of the space succeeds at ENOSPC.
1374 */
1375 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1376 tp->t_flags |= XFS_TRANS_RESERVE;
Jie Liu3d3c8b52013-08-12 20:49:59 +10001377 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, resblks, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001378
1379 /*
1380 * check for running out of space
1381 */
1382 if (error) {
1383 /*
1384 * Free the transaction structure.
1385 */
1386 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1387 xfs_trans_cancel(tp, 0);
1388 break;
1389 }
1390 xfs_ilock(ip, XFS_ILOCK_EXCL);
1391 error = xfs_trans_reserve_quota(tp, mp,
1392 ip->i_udquot, ip->i_gdquot, ip->i_pdquot,
1393 resblks, 0, XFS_QMOPT_RES_REGBLKS);
1394 if (error)
1395 goto error1;
1396
1397 xfs_trans_ijoin(tp, ip, 0);
1398
1399 /*
1400 * issue the bunmapi() call to free the blocks
1401 */
1402 xfs_bmap_init(&free_list, &firstfsb);
1403 error = xfs_bunmapi(tp, ip, startoffset_fsb,
1404 endoffset_fsb - startoffset_fsb,
1405 0, 2, &firstfsb, &free_list, &done);
1406 if (error) {
1407 goto error0;
1408 }
1409
1410 /*
1411 * complete the transaction
1412 */
1413 error = xfs_bmap_finish(&tp, &free_list, &committed);
1414 if (error) {
1415 goto error0;
1416 }
1417
1418 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1419 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1420 }
1421
1422 out_unlock_iolock:
1423 if (need_iolock)
1424 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1425 return error;
1426
1427 error0:
1428 xfs_bmap_cancel(&free_list);
1429 error1:
1430 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1431 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
1432 XFS_ILOCK_EXCL);
1433 return error;
1434}
1435
1436
1437STATIC int
1438xfs_zero_file_space(
1439 struct xfs_inode *ip,
1440 xfs_off_t offset,
1441 xfs_off_t len,
1442 int attr_flags)
1443{
1444 struct xfs_mount *mp = ip->i_mount;
1445 uint granularity;
1446 xfs_off_t start_boundary;
1447 xfs_off_t end_boundary;
1448 int error;
1449
1450 granularity = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
1451
1452 /*
1453 * Round the range of extents we are going to convert inwards. If the
1454 * offset is aligned, then it doesn't get changed so we zero from the
1455 * start of the block offset points to.
1456 */
1457 start_boundary = round_up(offset, granularity);
1458 end_boundary = round_down(offset + len, granularity);
1459
1460 ASSERT(start_boundary >= offset);
1461 ASSERT(end_boundary <= offset + len);
1462
1463 if (!(attr_flags & XFS_ATTR_NOLOCK))
1464 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1465
1466 if (start_boundary < end_boundary - 1) {
1467 /* punch out the page cache over the conversion range */
1468 truncate_pagecache_range(VFS_I(ip), start_boundary,
1469 end_boundary - 1);
1470 /* convert the blocks */
1471 error = xfs_alloc_file_space(ip, start_boundary,
1472 end_boundary - start_boundary - 1,
1473 XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT,
1474 attr_flags);
1475 if (error)
1476 goto out_unlock;
1477
1478 /* We've handled the interior of the range, now for the edges */
1479 if (start_boundary != offset)
1480 error = xfs_iozero(ip, offset, start_boundary - offset);
1481 if (error)
1482 goto out_unlock;
1483
1484 if (end_boundary != offset + len)
1485 error = xfs_iozero(ip, end_boundary,
1486 offset + len - end_boundary);
1487
1488 } else {
1489 /*
1490 * It's either a sub-granularity range or the range spanned lies
1491 * partially across two adjacent blocks.
1492 */
1493 error = xfs_iozero(ip, offset, len);
1494 }
1495
1496out_unlock:
1497 if (!(attr_flags & XFS_ATTR_NOLOCK))
1498 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1499 return error;
1500
1501}
1502
1503/*
1504 * xfs_change_file_space()
1505 * This routine allocates or frees disk space for the given file.
1506 * The user specified parameters are checked for alignment and size
1507 * limitations.
1508 *
1509 * RETURNS:
1510 * 0 on success
1511 * errno on error
1512 *
1513 */
1514int
1515xfs_change_file_space(
1516 xfs_inode_t *ip,
1517 int cmd,
1518 xfs_flock64_t *bf,
1519 xfs_off_t offset,
1520 int attr_flags)
1521{
1522 xfs_mount_t *mp = ip->i_mount;
1523 int clrprealloc;
1524 int error;
1525 xfs_fsize_t fsize;
1526 int setprealloc;
1527 xfs_off_t startoffset;
1528 xfs_trans_t *tp;
1529 struct iattr iattr;
1530
1531 if (!S_ISREG(ip->i_d.di_mode))
1532 return XFS_ERROR(EINVAL);
1533
1534 switch (bf->l_whence) {
1535 case 0: /*SEEK_SET*/
1536 break;
1537 case 1: /*SEEK_CUR*/
1538 bf->l_start += offset;
1539 break;
1540 case 2: /*SEEK_END*/
1541 bf->l_start += XFS_ISIZE(ip);
1542 break;
1543 default:
1544 return XFS_ERROR(EINVAL);
1545 }
1546
1547 /*
1548 * length of <= 0 for resv/unresv/zero is invalid. length for
1549 * alloc/free is ignored completely and we have no idea what userspace
1550 * might have set it to, so set it to zero to allow range
1551 * checks to pass.
1552 */
1553 switch (cmd) {
1554 case XFS_IOC_ZERO_RANGE:
1555 case XFS_IOC_RESVSP:
1556 case XFS_IOC_RESVSP64:
1557 case XFS_IOC_UNRESVSP:
1558 case XFS_IOC_UNRESVSP64:
1559 if (bf->l_len <= 0)
1560 return XFS_ERROR(EINVAL);
1561 break;
1562 default:
1563 bf->l_len = 0;
1564 break;
1565 }
1566
1567 if (bf->l_start < 0 ||
1568 bf->l_start > mp->m_super->s_maxbytes ||
1569 bf->l_start + bf->l_len < 0 ||
1570 bf->l_start + bf->l_len >= mp->m_super->s_maxbytes)
1571 return XFS_ERROR(EINVAL);
1572
1573 bf->l_whence = 0;
1574
1575 startoffset = bf->l_start;
1576 fsize = XFS_ISIZE(ip);
1577
1578 setprealloc = clrprealloc = 0;
1579 switch (cmd) {
1580 case XFS_IOC_ZERO_RANGE:
1581 error = xfs_zero_file_space(ip, startoffset, bf->l_len,
1582 attr_flags);
1583 if (error)
1584 return error;
1585 setprealloc = 1;
1586 break;
1587
1588 case XFS_IOC_RESVSP:
1589 case XFS_IOC_RESVSP64:
1590 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
1591 XFS_BMAPI_PREALLOC, attr_flags);
1592 if (error)
1593 return error;
1594 setprealloc = 1;
1595 break;
1596
1597 case XFS_IOC_UNRESVSP:
1598 case XFS_IOC_UNRESVSP64:
1599 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
1600 attr_flags)))
1601 return error;
1602 break;
1603
1604 case XFS_IOC_ALLOCSP:
1605 case XFS_IOC_ALLOCSP64:
1606 case XFS_IOC_FREESP:
1607 case XFS_IOC_FREESP64:
1608 /*
1609 * These operations actually do IO when extending the file, but
1610 * the allocation is done seperately to the zeroing that is
1611 * done. This set of operations need to be serialised against
1612 * other IO operations, such as truncate and buffered IO. We
1613 * need to take the IOLOCK here to serialise the allocation and
1614 * zeroing IO to prevent other IOLOCK holders (e.g. getbmap,
1615 * truncate, direct IO) from racing against the transient
1616 * allocated but not written state we can have here.
1617 */
1618 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1619 if (startoffset > fsize) {
1620 error = xfs_alloc_file_space(ip, fsize,
1621 startoffset - fsize, 0,
1622 attr_flags | XFS_ATTR_NOLOCK);
1623 if (error) {
1624 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1625 break;
1626 }
1627 }
1628
1629 iattr.ia_valid = ATTR_SIZE;
1630 iattr.ia_size = startoffset;
1631
1632 error = xfs_setattr_size(ip, &iattr,
1633 attr_flags | XFS_ATTR_NOLOCK);
1634 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1635
1636 if (error)
1637 return error;
1638
1639 clrprealloc = 1;
1640 break;
1641
1642 default:
1643 ASSERT(0);
1644 return XFS_ERROR(EINVAL);
1645 }
1646
1647 /*
1648 * update the inode timestamp, mode, and prealloc flag bits
1649 */
1650 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
Jie Liu3d3c8b52013-08-12 20:49:59 +10001651 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_writeid, 0, 0);
1652 if (error) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001653 xfs_trans_cancel(tp, 0);
1654 return error;
1655 }
1656
1657 xfs_ilock(ip, XFS_ILOCK_EXCL);
1658 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1659
1660 if ((attr_flags & XFS_ATTR_DMI) == 0) {
1661 ip->i_d.di_mode &= ~S_ISUID;
1662
1663 /*
1664 * Note that we don't have to worry about mandatory
1665 * file locking being disabled here because we only
1666 * clear the S_ISGID bit if the Group execute bit is
1667 * on, but if it was on then mandatory locking wouldn't
1668 * have been enabled.
1669 */
1670 if (ip->i_d.di_mode & S_IXGRP)
1671 ip->i_d.di_mode &= ~S_ISGID;
1672
1673 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1674 }
1675 if (setprealloc)
1676 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
1677 else if (clrprealloc)
1678 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
1679
1680 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1681 if (attr_flags & XFS_ATTR_SYNC)
1682 xfs_trans_set_sync(tp);
1683 return xfs_trans_commit(tp, 0);
1684}
Dave Chinnera133d952013-08-12 20:49:48 +10001685
1686/*
1687 * We need to check that the format of the data fork in the temporary inode is
1688 * valid for the target inode before doing the swap. This is not a problem with
1689 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1690 * data fork depending on the space the attribute fork is taking so we can get
1691 * invalid formats on the target inode.
1692 *
1693 * E.g. target has space for 7 extents in extent format, temp inode only has
1694 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1695 * btree, but when swapped it needs to be in extent format. Hence we can't just
1696 * blindly swap data forks on attr2 filesystems.
1697 *
1698 * Note that we check the swap in both directions so that we don't end up with
1699 * a corrupt temporary inode, either.
1700 *
1701 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1702 * inode will prevent this situation from occurring, so all we do here is
1703 * reject and log the attempt. basically we are putting the responsibility on
1704 * userspace to get this right.
1705 */
1706static int
1707xfs_swap_extents_check_format(
1708 xfs_inode_t *ip, /* target inode */
1709 xfs_inode_t *tip) /* tmp inode */
1710{
1711
1712 /* Should never get a local format */
1713 if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
1714 tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
1715 return EINVAL;
1716
1717 /*
1718 * if the target inode has less extents that then temporary inode then
1719 * why did userspace call us?
1720 */
1721 if (ip->i_d.di_nextents < tip->i_d.di_nextents)
1722 return EINVAL;
1723
1724 /*
1725 * if the target inode is in extent form and the temp inode is in btree
1726 * form then we will end up with the target inode in the wrong format
1727 * as we already know there are less extents in the temp inode.
1728 */
1729 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1730 tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1731 return EINVAL;
1732
1733 /* Check temp in extent form to max in target */
1734 if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1735 XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) >
1736 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1737 return EINVAL;
1738
1739 /* Check target in extent form to max in temp */
1740 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1741 XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) >
1742 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1743 return EINVAL;
1744
1745 /*
1746 * If we are in a btree format, check that the temp root block will fit
1747 * in the target and that it has enough extents to be in btree format
1748 * in the target.
1749 *
1750 * Note that we have to be careful to allow btree->extent conversions
1751 * (a common defrag case) which will occur when the temp inode is in
1752 * extent format...
1753 */
1754 if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1755 if (XFS_IFORK_BOFF(ip) &&
1756 XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip))
1757 return EINVAL;
1758 if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <=
1759 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1760 return EINVAL;
1761 }
1762
1763 /* Reciprocal target->temp btree format checks */
1764 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1765 if (XFS_IFORK_BOFF(tip) &&
1766 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
1767 return EINVAL;
1768 if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <=
1769 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1770 return EINVAL;
1771 }
1772
1773 return 0;
1774}
1775
1776int
1777xfs_swap_extents(
1778 xfs_inode_t *ip, /* target inode */
1779 xfs_inode_t *tip, /* tmp inode */
1780 xfs_swapext_t *sxp)
1781{
1782 xfs_mount_t *mp = ip->i_mount;
1783 xfs_trans_t *tp;
1784 xfs_bstat_t *sbp = &sxp->sx_stat;
1785 xfs_ifork_t *tempifp, *ifp, *tifp;
1786 int src_log_flags, target_log_flags;
1787 int error = 0;
1788 int aforkblks = 0;
1789 int taforkblks = 0;
1790 __uint64_t tmp;
1791
1792 /*
1793 * We have no way of updating owner information in the BMBT blocks for
1794 * each inode on CRC enabled filesystems, so to avoid corrupting the
1795 * this metadata we simply don't allow extent swaps to occur.
1796 */
1797 if (xfs_sb_version_hascrc(&mp->m_sb))
1798 return XFS_ERROR(EINVAL);
1799
1800 tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL);
1801 if (!tempifp) {
1802 error = XFS_ERROR(ENOMEM);
1803 goto out;
1804 }
1805
1806 /*
1807 * we have to do two separate lock calls here to keep lockdep
1808 * happy. If we try to get all the locks in one call, lock will
1809 * report false positives when we drop the ILOCK and regain them
1810 * below.
1811 */
1812 xfs_lock_two_inodes(ip, tip, XFS_IOLOCK_EXCL);
1813 xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
1814
1815 /* Verify that both files have the same format */
1816 if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) {
1817 error = XFS_ERROR(EINVAL);
1818 goto out_unlock;
1819 }
1820
1821 /* Verify both files are either real-time or non-realtime */
1822 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1823 error = XFS_ERROR(EINVAL);
1824 goto out_unlock;
1825 }
1826
1827 error = -filemap_write_and_wait(VFS_I(tip)->i_mapping);
1828 if (error)
1829 goto out_unlock;
1830 truncate_pagecache_range(VFS_I(tip), 0, -1);
1831
1832 /* Verify O_DIRECT for ftmp */
1833 if (VN_CACHED(VFS_I(tip)) != 0) {
1834 error = XFS_ERROR(EINVAL);
1835 goto out_unlock;
1836 }
1837
1838 /* Verify all data are being swapped */
1839 if (sxp->sx_offset != 0 ||
1840 sxp->sx_length != ip->i_d.di_size ||
1841 sxp->sx_length != tip->i_d.di_size) {
1842 error = XFS_ERROR(EFAULT);
1843 goto out_unlock;
1844 }
1845
1846 trace_xfs_swap_extent_before(ip, 0);
1847 trace_xfs_swap_extent_before(tip, 1);
1848
1849 /* check inode formats now that data is flushed */
1850 error = xfs_swap_extents_check_format(ip, tip);
1851 if (error) {
1852 xfs_notice(mp,
1853 "%s: inode 0x%llx format is incompatible for exchanging.",
1854 __func__, ip->i_ino);
1855 goto out_unlock;
1856 }
1857
1858 /*
1859 * Compare the current change & modify times with that
1860 * passed in. If they differ, we abort this swap.
1861 * This is the mechanism used to ensure the calling
1862 * process that the file was not changed out from
1863 * under it.
1864 */
1865 if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
1866 (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
1867 (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
1868 (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
1869 error = XFS_ERROR(EBUSY);
1870 goto out_unlock;
1871 }
1872
1873 /* We need to fail if the file is memory mapped. Once we have tossed
1874 * all existing pages, the page fault will have no option
1875 * but to go to the filesystem for pages. By making the page fault call
1876 * vop_read (or write in the case of autogrow) they block on the iolock
1877 * until we have switched the extents.
1878 */
1879 if (VN_MAPPED(VFS_I(ip))) {
1880 error = XFS_ERROR(EBUSY);
1881 goto out_unlock;
1882 }
1883
1884 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1885 xfs_iunlock(tip, XFS_ILOCK_EXCL);
1886
1887 /*
1888 * There is a race condition here since we gave up the
1889 * ilock. However, the data fork will not change since
1890 * we have the iolock (locked for truncation too) so we
1891 * are safe. We don't really care if non-io related
1892 * fields change.
1893 */
1894 truncate_pagecache_range(VFS_I(ip), 0, -1);
1895
1896 tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT);
Jie Liu3d3c8b52013-08-12 20:49:59 +10001897 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
1898 if (error) {
Dave Chinnera133d952013-08-12 20:49:48 +10001899 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1900 xfs_iunlock(tip, XFS_IOLOCK_EXCL);
1901 xfs_trans_cancel(tp, 0);
1902 goto out;
1903 }
1904 xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
1905
1906 /*
1907 * Count the number of extended attribute blocks
1908 */
1909 if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
1910 (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1911 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
1912 if (error)
1913 goto out_trans_cancel;
1914 }
1915 if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
1916 (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1917 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
1918 &taforkblks);
1919 if (error)
1920 goto out_trans_cancel;
1921 }
1922
1923 /*
1924 * Swap the data forks of the inodes
1925 */
1926 ifp = &ip->i_df;
1927 tifp = &tip->i_df;
1928 *tempifp = *ifp; /* struct copy */
1929 *ifp = *tifp; /* struct copy */
1930 *tifp = *tempifp; /* struct copy */
1931
1932 /*
1933 * Fix the on-disk inode values
1934 */
1935 tmp = (__uint64_t)ip->i_d.di_nblocks;
1936 ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
1937 tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
1938
1939 tmp = (__uint64_t) ip->i_d.di_nextents;
1940 ip->i_d.di_nextents = tip->i_d.di_nextents;
1941 tip->i_d.di_nextents = tmp;
1942
1943 tmp = (__uint64_t) ip->i_d.di_format;
1944 ip->i_d.di_format = tip->i_d.di_format;
1945 tip->i_d.di_format = tmp;
1946
1947 /*
1948 * The extents in the source inode could still contain speculative
1949 * preallocation beyond EOF (e.g. the file is open but not modified
1950 * while defrag is in progress). In that case, we need to copy over the
1951 * number of delalloc blocks the data fork in the source inode is
1952 * tracking beyond EOF so that when the fork is truncated away when the
1953 * temporary inode is unlinked we don't underrun the i_delayed_blks
1954 * counter on that inode.
1955 */
1956 ASSERT(tip->i_delayed_blks == 0);
1957 tip->i_delayed_blks = ip->i_delayed_blks;
1958 ip->i_delayed_blks = 0;
1959
1960 src_log_flags = XFS_ILOG_CORE;
1961 switch (ip->i_d.di_format) {
1962 case XFS_DINODE_FMT_EXTENTS:
1963 /* If the extents fit in the inode, fix the
1964 * pointer. Otherwise it's already NULL or
1965 * pointing to the extent.
1966 */
1967 if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) {
1968 ifp->if_u1.if_extents =
1969 ifp->if_u2.if_inline_ext;
1970 }
1971 src_log_flags |= XFS_ILOG_DEXT;
1972 break;
1973 case XFS_DINODE_FMT_BTREE:
1974 src_log_flags |= XFS_ILOG_DBROOT;
1975 break;
1976 }
1977
1978 target_log_flags = XFS_ILOG_CORE;
1979 switch (tip->i_d.di_format) {
1980 case XFS_DINODE_FMT_EXTENTS:
1981 /* If the extents fit in the inode, fix the
1982 * pointer. Otherwise it's already NULL or
1983 * pointing to the extent.
1984 */
1985 if (tip->i_d.di_nextents <= XFS_INLINE_EXTS) {
1986 tifp->if_u1.if_extents =
1987 tifp->if_u2.if_inline_ext;
1988 }
1989 target_log_flags |= XFS_ILOG_DEXT;
1990 break;
1991 case XFS_DINODE_FMT_BTREE:
1992 target_log_flags |= XFS_ILOG_DBROOT;
1993 break;
1994 }
1995
1996
1997 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1998 xfs_trans_ijoin(tp, tip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1999
2000 xfs_trans_log_inode(tp, ip, src_log_flags);
2001 xfs_trans_log_inode(tp, tip, target_log_flags);
2002
2003 /*
2004 * If this is a synchronous mount, make sure that the
2005 * transaction goes to disk before returning to the user.
2006 */
2007 if (mp->m_flags & XFS_MOUNT_WSYNC)
2008 xfs_trans_set_sync(tp);
2009
2010 error = xfs_trans_commit(tp, 0);
2011
2012 trace_xfs_swap_extent_after(ip, 0);
2013 trace_xfs_swap_extent_after(tip, 1);
2014out:
2015 kmem_free(tempifp);
2016 return error;
2017
2018out_unlock:
2019 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
2020 xfs_iunlock(tip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
2021 goto out;
2022
2023out_trans_cancel:
2024 xfs_trans_cancel(tp, 0);
2025 goto out_unlock;
2026}