~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/xfs/xfs_bmap_util.c

Version: ~ [ linux-5.1-rc1 ] ~ [ linux-5.0.3 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.30 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.107 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.164 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.176 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.136 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.63 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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_da_format.h"
 16 #include "xfs_defer.h"
 17 #include "xfs_inode.h"
 18 #include "xfs_btree.h"
 19 #include "xfs_trans.h"
 20 #include "xfs_extfree_item.h"
 21 #include "xfs_alloc.h"
 22 #include "xfs_bmap.h"
 23 #include "xfs_bmap_util.h"
 24 #include "xfs_bmap_btree.h"
 25 #include "xfs_rtalloc.h"
 26 #include "xfs_error.h"
 27 #include "xfs_quota.h"
 28 #include "xfs_trans_space.h"
 29 #include "xfs_trace.h"
 30 #include "xfs_icache.h"
 31 #include "xfs_log.h"
 32 #include "xfs_rmap_btree.h"
 33 #include "xfs_iomap.h"
 34 #include "xfs_reflink.h"
 35 #include "xfs_refcount.h"
 36 
 37 /* Kernel only BMAP related definitions and functions */
 38 
 39 /*
 40  * Convert the given file system block to a disk block.  We have to treat it
 41  * differently based on whether the file is a real time file or not, because the
 42  * bmap code does.
 43  */
 44 xfs_daddr_t
 45 xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb)
 46 {
 47         return (XFS_IS_REALTIME_INODE(ip) ? \
 48                  (xfs_daddr_t)XFS_FSB_TO_BB((ip)->i_mount, (fsb)) : \
 49                  XFS_FSB_TO_DADDR((ip)->i_mount, (fsb)));
 50 }
 51 
 52 /*
 53  * Routine to zero an extent on disk allocated to the specific inode.
 54  *
 55  * The VFS functions take a linearised filesystem block offset, so we have to
 56  * convert the sparse xfs fsb to the right format first.
 57  * VFS types are real funky, too.
 58  */
 59 int
 60 xfs_zero_extent(
 61         struct xfs_inode *ip,
 62         xfs_fsblock_t   start_fsb,
 63         xfs_off_t       count_fsb)
 64 {
 65         struct xfs_mount *mp = ip->i_mount;
 66         xfs_daddr_t     sector = xfs_fsb_to_db(ip, start_fsb);
 67         sector_t        block = XFS_BB_TO_FSBT(mp, sector);
 68 
 69         return blkdev_issue_zeroout(xfs_find_bdev_for_inode(VFS_I(ip)),
 70                 block << (mp->m_super->s_blocksize_bits - 9),
 71                 count_fsb << (mp->m_super->s_blocksize_bits - 9),
 72                 GFP_NOFS, 0);
 73 }
 74 
 75 #ifdef CONFIG_XFS_RT
 76 int
 77 xfs_bmap_rtalloc(
 78         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
 79 {
 80         int             error;          /* error return value */
 81         xfs_mount_t     *mp;            /* mount point structure */
 82         xfs_extlen_t    prod = 0;       /* product factor for allocators */
 83         xfs_extlen_t    mod = 0;        /* product factor for allocators */
 84         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
 85         xfs_extlen_t    align;          /* minimum allocation alignment */
 86         xfs_rtblock_t   rtb;
 87 
 88         mp = ap->ip->i_mount;
 89         align = xfs_get_extsz_hint(ap->ip);
 90         prod = align / mp->m_sb.sb_rextsize;
 91         error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
 92                                         align, 1, ap->eof, 0,
 93                                         ap->conv, &ap->offset, &ap->length);
 94         if (error)
 95                 return error;
 96         ASSERT(ap->length);
 97         ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
 98 
 99         /*
100          * If the offset & length are not perfectly aligned
101          * then kill prod, it will just get us in trouble.
102          */
103         div_u64_rem(ap->offset, align, &mod);
104         if (mod || ap->length % align)
105                 prod = 1;
106         /*
107          * Set ralen to be the actual requested length in rtextents.
108          */
109         ralen = ap->length / mp->m_sb.sb_rextsize;
110         /*
111          * If the old value was close enough to MAXEXTLEN that
112          * we rounded up to it, cut it back so it's valid again.
113          * Note that if it's a really large request (bigger than
114          * MAXEXTLEN), we don't hear about that number, and can't
115          * adjust the starting point to match it.
116          */
117         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
118                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
119 
120         /*
121          * Lock out modifications to both the RT bitmap and summary inodes
122          */
123         xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
124         xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
125         xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
126         xfs_trans_ijoin(ap->tp, mp->m_rsumip, XFS_ILOCK_EXCL);
127 
128         /*
129          * If it's an allocation to an empty file at offset 0,
130          * pick an extent that will space things out in the rt area.
131          */
132         if (ap->eof && ap->offset == 0) {
133                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
134 
135                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
136                 if (error)
137                         return error;
138                 ap->blkno = rtx * mp->m_sb.sb_rextsize;
139         } else {
140                 ap->blkno = 0;
141         }
142 
143         xfs_bmap_adjacent(ap);
144 
145         /*
146          * Realtime allocation, done through xfs_rtallocate_extent.
147          */
148         do_div(ap->blkno, mp->m_sb.sb_rextsize);
149         rtb = ap->blkno;
150         ap->length = ralen;
151         error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
152                                 &ralen, ap->wasdel, prod, &rtb);
153         if (error)
154                 return error;
155 
156         ap->blkno = rtb;
157         if (ap->blkno != NULLFSBLOCK) {
158                 ap->blkno *= mp->m_sb.sb_rextsize;
159                 ralen *= mp->m_sb.sb_rextsize;
160                 ap->length = ralen;
161                 ap->ip->i_d.di_nblocks += ralen;
162                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
163                 if (ap->wasdel)
164                         ap->ip->i_delayed_blks -= ralen;
165                 /*
166                  * Adjust the disk quota also. This was reserved
167                  * earlier.
168                  */
169                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
170                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
171                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
172 
173                 /* Zero the extent if we were asked to do so */
174                 if (ap->datatype & XFS_ALLOC_USERDATA_ZERO) {
175                         error = xfs_zero_extent(ap->ip, ap->blkno, ap->length);
176                         if (error)
177                                 return error;
178                 }
179         } else {
180                 ap->length = 0;
181         }
182         return 0;
183 }
184 #endif /* CONFIG_XFS_RT */
185 
186 /*
187  * Check if the endoff is outside the last extent. If so the caller will grow
188  * the allocation to a stripe unit boundary.  All offsets are considered outside
189  * the end of file for an empty fork, so 1 is returned in *eof in that case.
190  */
191 int
192 xfs_bmap_eof(
193         struct xfs_inode        *ip,
194         xfs_fileoff_t           endoff,
195         int                     whichfork,
196         int                     *eof)
197 {
198         struct xfs_bmbt_irec    rec;
199         int                     error;
200 
201         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
202         if (error || *eof)
203                 return error;
204 
205         *eof = endoff >= rec.br_startoff + rec.br_blockcount;
206         return 0;
207 }
208 
209 /*
210  * Extent tree block counting routines.
211  */
212 
213 /*
214  * Count leaf blocks given a range of extent records.  Delayed allocation
215  * extents are not counted towards the totals.
216  */
217 xfs_extnum_t
218 xfs_bmap_count_leaves(
219         struct xfs_ifork        *ifp,
220         xfs_filblks_t           *count)
221 {
222         struct xfs_iext_cursor  icur;
223         struct xfs_bmbt_irec    got;
224         xfs_extnum_t            numrecs = 0;
225 
226         for_each_xfs_iext(ifp, &icur, &got) {
227                 if (!isnullstartblock(got.br_startblock)) {
228                         *count += got.br_blockcount;
229                         numrecs++;
230                 }
231         }
232 
233         return numrecs;
234 }
235 
236 /*
237  * Count leaf blocks given a range of extent records originally
238  * in btree format.
239  */
240 STATIC void
241 xfs_bmap_disk_count_leaves(
242         struct xfs_mount        *mp,
243         struct xfs_btree_block  *block,
244         int                     numrecs,
245         xfs_filblks_t           *count)
246 {
247         int             b;
248         xfs_bmbt_rec_t  *frp;
249 
250         for (b = 1; b <= numrecs; b++) {
251                 frp = XFS_BMBT_REC_ADDR(mp, block, b);
252                 *count += xfs_bmbt_disk_get_blockcount(frp);
253         }
254 }
255 
256 /*
257  * Recursively walks each level of a btree
258  * to count total fsblocks in use.
259  */
260 STATIC int
261 xfs_bmap_count_tree(
262         struct xfs_mount        *mp,
263         struct xfs_trans        *tp,
264         struct xfs_ifork        *ifp,
265         xfs_fsblock_t           blockno,
266         int                     levelin,
267         xfs_extnum_t            *nextents,
268         xfs_filblks_t           *count)
269 {
270         int                     error;
271         struct xfs_buf          *bp, *nbp;
272         int                     level = levelin;
273         __be64                  *pp;
274         xfs_fsblock_t           bno = blockno;
275         xfs_fsblock_t           nextbno;
276         struct xfs_btree_block  *block, *nextblock;
277         int                     numrecs;
278 
279         error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF,
280                                                 &xfs_bmbt_buf_ops);
281         if (error)
282                 return error;
283         *count += 1;
284         block = XFS_BUF_TO_BLOCK(bp);
285 
286         if (--level) {
287                 /* Not at node above leaves, count this level of nodes */
288                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
289                 while (nextbno != NULLFSBLOCK) {
290                         error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp,
291                                                 XFS_BMAP_BTREE_REF,
292                                                 &xfs_bmbt_buf_ops);
293                         if (error)
294                                 return error;
295                         *count += 1;
296                         nextblock = XFS_BUF_TO_BLOCK(nbp);
297                         nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
298                         xfs_trans_brelse(tp, nbp);
299                 }
300 
301                 /* Dive to the next level */
302                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
303                 bno = be64_to_cpu(*pp);
304                 error = xfs_bmap_count_tree(mp, tp, ifp, bno, level, nextents,
305                                 count);
306                 if (error) {
307                         xfs_trans_brelse(tp, bp);
308                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
309                                          XFS_ERRLEVEL_LOW, mp);
310                         return -EFSCORRUPTED;
311                 }
312                 xfs_trans_brelse(tp, bp);
313         } else {
314                 /* count all level 1 nodes and their leaves */
315                 for (;;) {
316                         nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
317                         numrecs = be16_to_cpu(block->bb_numrecs);
318                         (*nextents) += numrecs;
319                         xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
320                         xfs_trans_brelse(tp, bp);
321                         if (nextbno == NULLFSBLOCK)
322                                 break;
323                         bno = nextbno;
324                         error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
325                                                 XFS_BMAP_BTREE_REF,
326                                                 &xfs_bmbt_buf_ops);
327                         if (error)
328                                 return error;
329                         *count += 1;
330                         block = XFS_BUF_TO_BLOCK(bp);
331                 }
332         }
333         return 0;
334 }
335 
336 /*
337  * Count fsblocks of the given fork.  Delayed allocation extents are
338  * not counted towards the totals.
339  */
340 int
341 xfs_bmap_count_blocks(
342         struct xfs_trans        *tp,
343         struct xfs_inode        *ip,
344         int                     whichfork,
345         xfs_extnum_t            *nextents,
346         xfs_filblks_t           *count)
347 {
348         struct xfs_mount        *mp;    /* file system mount structure */
349         __be64                  *pp;    /* pointer to block address */
350         struct xfs_btree_block  *block; /* current btree block */
351         struct xfs_ifork        *ifp;   /* fork structure */
352         xfs_fsblock_t           bno;    /* block # of "block" */
353         int                     level;  /* btree level, for checking */
354         int                     error;
355 
356         bno = NULLFSBLOCK;
357         mp = ip->i_mount;
358         *nextents = 0;
359         *count = 0;
360         ifp = XFS_IFORK_PTR(ip, whichfork);
361         if (!ifp)
362                 return 0;
363 
364         switch (XFS_IFORK_FORMAT(ip, whichfork)) {
365         case XFS_DINODE_FMT_EXTENTS:
366                 *nextents = xfs_bmap_count_leaves(ifp, count);
367                 return 0;
368         case XFS_DINODE_FMT_BTREE:
369                 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
370                         error = xfs_iread_extents(tp, ip, whichfork);
371                         if (error)
372                                 return error;
373                 }
374 
375                 /*
376                  * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
377                  */
378                 block = ifp->if_broot;
379                 level = be16_to_cpu(block->bb_level);
380                 ASSERT(level > 0);
381                 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
382                 bno = be64_to_cpu(*pp);
383                 ASSERT(bno != NULLFSBLOCK);
384                 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
385                 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
386 
387                 error = xfs_bmap_count_tree(mp, tp, ifp, bno, level,
388                                 nextents, count);
389                 if (error) {
390                         XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)",
391                                         XFS_ERRLEVEL_LOW, mp);
392                         return -EFSCORRUPTED;
393                 }
394                 return 0;
395         }
396 
397         return 0;
398 }
399 
400 static int
401 xfs_getbmap_report_one(
402         struct xfs_inode        *ip,
403         struct getbmapx         *bmv,
404         struct kgetbmap         *out,
405         int64_t                 bmv_end,
406         struct xfs_bmbt_irec    *got)
407 {
408         struct kgetbmap         *p = out + bmv->bmv_entries;
409         bool                    shared = false;
410         int                     error;
411 
412         error = xfs_reflink_trim_around_shared(ip, got, &shared);
413         if (error)
414                 return error;
415 
416         if (isnullstartblock(got->br_startblock) ||
417             got->br_startblock == DELAYSTARTBLOCK) {
418                 /*
419                  * Delalloc extents that start beyond EOF can occur due to
420                  * speculative EOF allocation when the delalloc extent is larger
421                  * than the largest freespace extent at conversion time.  These
422                  * extents cannot be converted by data writeback, so can exist
423                  * here even if we are not supposed to be finding delalloc
424                  * extents.
425                  */
426                 if (got->br_startoff < XFS_B_TO_FSB(ip->i_mount, XFS_ISIZE(ip)))
427                         ASSERT((bmv->bmv_iflags & BMV_IF_DELALLOC) != 0);
428 
429                 p->bmv_oflags |= BMV_OF_DELALLOC;
430                 p->bmv_block = -2;
431         } else {
432                 p->bmv_block = xfs_fsb_to_db(ip, got->br_startblock);
433         }
434 
435         if (got->br_state == XFS_EXT_UNWRITTEN &&
436             (bmv->bmv_iflags & BMV_IF_PREALLOC))
437                 p->bmv_oflags |= BMV_OF_PREALLOC;
438 
439         if (shared)
440                 p->bmv_oflags |= BMV_OF_SHARED;
441 
442         p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, got->br_startoff);
443         p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, got->br_blockcount);
444 
445         bmv->bmv_offset = p->bmv_offset + p->bmv_length;
446         bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset);
447         bmv->bmv_entries++;
448         return 0;
449 }
450 
451 static void
452 xfs_getbmap_report_hole(
453         struct xfs_inode        *ip,
454         struct getbmapx         *bmv,
455         struct kgetbmap         *out,
456         int64_t                 bmv_end,
457         xfs_fileoff_t           bno,
458         xfs_fileoff_t           end)
459 {
460         struct kgetbmap         *p = out + bmv->bmv_entries;
461 
462         if (bmv->bmv_iflags & BMV_IF_NO_HOLES)
463                 return;
464 
465         p->bmv_block = -1;
466         p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, bno);
467         p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, end - bno);
468 
469         bmv->bmv_offset = p->bmv_offset + p->bmv_length;
470         bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset);
471         bmv->bmv_entries++;
472 }
473 
474 static inline bool
475 xfs_getbmap_full(
476         struct getbmapx         *bmv)
477 {
478         return bmv->bmv_length == 0 || bmv->bmv_entries >= bmv->bmv_count - 1;
479 }
480 
481 static bool
482 xfs_getbmap_next_rec(
483         struct xfs_bmbt_irec    *rec,
484         xfs_fileoff_t           total_end)
485 {
486         xfs_fileoff_t           end = rec->br_startoff + rec->br_blockcount;
487 
488         if (end == total_end)
489                 return false;
490 
491         rec->br_startoff += rec->br_blockcount;
492         if (!isnullstartblock(rec->br_startblock) &&
493             rec->br_startblock != DELAYSTARTBLOCK)
494                 rec->br_startblock += rec->br_blockcount;
495         rec->br_blockcount = total_end - end;
496         return true;
497 }
498 
499 /*
500  * Get inode's extents as described in bmv, and format for output.
501  * Calls formatter to fill the user's buffer until all extents
502  * are mapped, until the passed-in bmv->bmv_count slots have
503  * been filled, or until the formatter short-circuits the loop,
504  * if it is tracking filled-in extents on its own.
505  */
506 int                                             /* error code */
507 xfs_getbmap(
508         struct xfs_inode        *ip,
509         struct getbmapx         *bmv,           /* user bmap structure */
510         struct kgetbmap         *out)
511 {
512         struct xfs_mount        *mp = ip->i_mount;
513         int                     iflags = bmv->bmv_iflags;
514         int                     whichfork, lock, error = 0;
515         int64_t                 bmv_end, max_len;
516         xfs_fileoff_t           bno, first_bno;
517         struct xfs_ifork        *ifp;
518         struct xfs_bmbt_irec    got, rec;
519         xfs_filblks_t           len;
520         struct xfs_iext_cursor  icur;
521 
522         if (bmv->bmv_iflags & ~BMV_IF_VALID)
523                 return -EINVAL;
524 #ifndef DEBUG
525         /* Only allow CoW fork queries if we're debugging. */
526         if (iflags & BMV_IF_COWFORK)
527                 return -EINVAL;
528 #endif
529         if ((iflags & BMV_IF_ATTRFORK) && (iflags & BMV_IF_COWFORK))
530                 return -EINVAL;
531 
532         if (bmv->bmv_length < -1)
533                 return -EINVAL;
534         bmv->bmv_entries = 0;
535         if (bmv->bmv_length == 0)
536                 return 0;
537 
538         if (iflags & BMV_IF_ATTRFORK)
539                 whichfork = XFS_ATTR_FORK;
540         else if (iflags & BMV_IF_COWFORK)
541                 whichfork = XFS_COW_FORK;
542         else
543                 whichfork = XFS_DATA_FORK;
544         ifp = XFS_IFORK_PTR(ip, whichfork);
545 
546         xfs_ilock(ip, XFS_IOLOCK_SHARED);
547         switch (whichfork) {
548         case XFS_ATTR_FORK:
549                 if (!XFS_IFORK_Q(ip))
550                         goto out_unlock_iolock;
551 
552                 max_len = 1LL << 32;
553                 lock = xfs_ilock_attr_map_shared(ip);
554                 break;
555         case XFS_COW_FORK:
556                 /* No CoW fork? Just return */
557                 if (!ifp)
558                         goto out_unlock_iolock;
559 
560                 if (xfs_get_cowextsz_hint(ip))
561                         max_len = mp->m_super->s_maxbytes;
562                 else
563                         max_len = XFS_ISIZE(ip);
564 
565                 lock = XFS_ILOCK_SHARED;
566                 xfs_ilock(ip, lock);
567                 break;
568         case XFS_DATA_FORK:
569                 if (!(iflags & BMV_IF_DELALLOC) &&
570                     (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) {
571                         error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
572                         if (error)
573                                 goto out_unlock_iolock;
574 
575                         /*
576                          * Even after flushing the inode, there can still be
577                          * delalloc blocks on the inode beyond EOF due to
578                          * speculative preallocation.  These are not removed
579                          * until the release function is called or the inode
580                          * is inactivated.  Hence we cannot assert here that
581                          * ip->i_delayed_blks == 0.
582                          */
583                 }
584 
585                 if (xfs_get_extsz_hint(ip) ||
586                     (ip->i_d.di_flags &
587                      (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))
588                         max_len = mp->m_super->s_maxbytes;
589                 else
590                         max_len = XFS_ISIZE(ip);
591 
592                 lock = xfs_ilock_data_map_shared(ip);
593                 break;
594         }
595 
596         switch (XFS_IFORK_FORMAT(ip, whichfork)) {
597         case XFS_DINODE_FMT_EXTENTS:
598         case XFS_DINODE_FMT_BTREE:
599                 break;
600         case XFS_DINODE_FMT_LOCAL:
601                 /* Local format inode forks report no extents. */
602                 goto out_unlock_ilock;
603         default:
604                 error = -EINVAL;
605                 goto out_unlock_ilock;
606         }
607 
608         if (bmv->bmv_length == -1) {
609                 max_len = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, max_len));
610                 bmv->bmv_length = max(0LL, max_len - bmv->bmv_offset);
611         }
612 
613         bmv_end = bmv->bmv_offset + bmv->bmv_length;
614 
615         first_bno = bno = XFS_BB_TO_FSBT(mp, bmv->bmv_offset);
616         len = XFS_BB_TO_FSB(mp, bmv->bmv_length);
617 
618         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
619                 error = xfs_iread_extents(NULL, ip, whichfork);
620                 if (error)
621                         goto out_unlock_ilock;
622         }
623 
624         if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
625                 /*
626                  * Report a whole-file hole if the delalloc flag is set to
627                  * stay compatible with the old implementation.
628                  */
629                 if (iflags & BMV_IF_DELALLOC)
630                         xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
631                                         XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
632                 goto out_unlock_ilock;
633         }
634 
635         while (!xfs_getbmap_full(bmv)) {
636                 xfs_trim_extent(&got, first_bno, len);
637 
638                 /*
639                  * Report an entry for a hole if this extent doesn't directly
640                  * follow the previous one.
641                  */
642                 if (got.br_startoff > bno) {
643                         xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
644                                         got.br_startoff);
645                         if (xfs_getbmap_full(bmv))
646                                 break;
647                 }
648 
649                 /*
650                  * In order to report shared extents accurately, we report each
651                  * distinct shared / unshared part of a single bmbt record with
652                  * an individual getbmapx record.
653                  */
654                 bno = got.br_startoff + got.br_blockcount;
655                 rec = got;
656                 do {
657                         error = xfs_getbmap_report_one(ip, bmv, out, bmv_end,
658                                         &rec);
659                         if (error || xfs_getbmap_full(bmv))
660                                 goto out_unlock_ilock;
661                 } while (xfs_getbmap_next_rec(&rec, bno));
662 
663                 if (!xfs_iext_next_extent(ifp, &icur, &got)) {
664                         xfs_fileoff_t   end = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
665 
666                         out[bmv->bmv_entries - 1].bmv_oflags |= BMV_OF_LAST;
667 
668                         if (whichfork != XFS_ATTR_FORK && bno < end &&
669                             !xfs_getbmap_full(bmv)) {
670                                 xfs_getbmap_report_hole(ip, bmv, out, bmv_end,
671                                                 bno, end);
672                         }
673                         break;
674                 }
675 
676                 if (bno >= first_bno + len)
677                         break;
678         }
679 
680 out_unlock_ilock:
681         xfs_iunlock(ip, lock);
682 out_unlock_iolock:
683         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
684         return error;
685 }
686 
687 /*
688  * Dead simple method of punching delalyed allocation blocks from a range in
689  * the inode.  This will always punch out both the start and end blocks, even
690  * if the ranges only partially overlap them, so it is up to the caller to
691  * ensure that partial blocks are not passed in.
692  */
693 int
694 xfs_bmap_punch_delalloc_range(
695         struct xfs_inode        *ip,
696         xfs_fileoff_t           start_fsb,
697         xfs_fileoff_t           length)
698 {
699         struct xfs_ifork        *ifp = &ip->i_df;
700         xfs_fileoff_t           end_fsb = start_fsb + length;
701         struct xfs_bmbt_irec    got, del;
702         struct xfs_iext_cursor  icur;
703         int                     error = 0;
704 
705         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
706 
707         xfs_ilock(ip, XFS_ILOCK_EXCL);
708         if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got))
709                 goto out_unlock;
710 
711         while (got.br_startoff + got.br_blockcount > start_fsb) {
712                 del = got;
713                 xfs_trim_extent(&del, start_fsb, length);
714 
715                 /*
716                  * A delete can push the cursor forward. Step back to the
717                  * previous extent on non-delalloc or extents outside the
718                  * target range.
719                  */
720                 if (!del.br_blockcount ||
721                     !isnullstartblock(del.br_startblock)) {
722                         if (!xfs_iext_prev_extent(ifp, &icur, &got))
723                                 break;
724                         continue;
725                 }
726 
727                 error = xfs_bmap_del_extent_delay(ip, XFS_DATA_FORK, &icur,
728                                                   &got, &del);
729                 if (error || !xfs_iext_get_extent(ifp, &icur, &got))
730                         break;
731         }
732 
733 out_unlock:
734         xfs_iunlock(ip, XFS_ILOCK_EXCL);
735         return error;
736 }
737 
738 /*
739  * Test whether it is appropriate to check an inode for and free post EOF
740  * blocks. The 'force' parameter determines whether we should also consider
741  * regular files that are marked preallocated or append-only.
742  */
743 bool
744 xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
745 {
746         /* prealloc/delalloc exists only on regular files */
747         if (!S_ISREG(VFS_I(ip)->i_mode))
748                 return false;
749 
750         /*
751          * Zero sized files with no cached pages and delalloc blocks will not
752          * have speculative prealloc/delalloc blocks to remove.
753          */
754         if (VFS_I(ip)->i_size == 0 &&
755             VFS_I(ip)->i_mapping->nrpages == 0 &&
756             ip->i_delayed_blks == 0)
757                 return false;
758 
759         /* If we haven't read in the extent list, then don't do it now. */
760         if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
761                 return false;
762 
763         /*
764          * Do not free real preallocated or append-only files unless the file
765          * has delalloc blocks and we are forced to remove them.
766          */
767         if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
768                 if (!force || ip->i_delayed_blks == 0)
769                         return false;
770 
771         return true;
772 }
773 
774 /*
775  * This is called to free any blocks beyond eof. The caller must hold
776  * IOLOCK_EXCL unless we are in the inode reclaim path and have the only
777  * reference to the inode.
778  */
779 int
780 xfs_free_eofblocks(
781         struct xfs_inode        *ip)
782 {
783         struct xfs_trans        *tp;
784         int                     error;
785         xfs_fileoff_t           end_fsb;
786         xfs_fileoff_t           last_fsb;
787         xfs_filblks_t           map_len;
788         int                     nimaps;
789         struct xfs_bmbt_irec    imap;
790         struct xfs_mount        *mp = ip->i_mount;
791 
792         /*
793          * Figure out if there are any blocks beyond the end
794          * of the file.  If not, then there is nothing to do.
795          */
796         end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
797         last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
798         if (last_fsb <= end_fsb)
799                 return 0;
800         map_len = last_fsb - end_fsb;
801 
802         nimaps = 1;
803         xfs_ilock(ip, XFS_ILOCK_SHARED);
804         error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
805         xfs_iunlock(ip, XFS_ILOCK_SHARED);
806 
807         /*
808          * If there are blocks after the end of file, truncate the file to its
809          * current size to free them up.
810          */
811         if (!error && (nimaps != 0) &&
812             (imap.br_startblock != HOLESTARTBLOCK ||
813              ip->i_delayed_blks)) {
814                 /*
815                  * Attach the dquots to the inode up front.
816                  */
817                 error = xfs_qm_dqattach(ip);
818                 if (error)
819                         return error;
820 
821                 /* wait on dio to ensure i_size has settled */
822                 inode_dio_wait(VFS_I(ip));
823 
824                 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0,
825                                 &tp);
826                 if (error) {
827                         ASSERT(XFS_FORCED_SHUTDOWN(mp));
828                         return error;
829                 }
830 
831                 xfs_ilock(ip, XFS_ILOCK_EXCL);
832                 xfs_trans_ijoin(tp, ip, 0);
833 
834                 /*
835                  * Do not update the on-disk file size.  If we update the
836                  * on-disk file size and then the system crashes before the
837                  * contents of the file are flushed to disk then the files
838                  * may be full of holes (ie NULL files bug).
839                  */
840                 error = xfs_itruncate_extents_flags(&tp, ip, XFS_DATA_FORK,
841                                         XFS_ISIZE(ip), XFS_BMAPI_NODISCARD);
842                 if (error) {
843                         /*
844                          * If we get an error at this point we simply don't
845                          * bother truncating the file.
846                          */
847                         xfs_trans_cancel(tp);
848                 } else {
849                         error = xfs_trans_commit(tp);
850                         if (!error)
851                                 xfs_inode_clear_eofblocks_tag(ip);
852                 }
853 
854                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
855         }
856         return error;
857 }
858 
859 int
860 xfs_alloc_file_space(
861         struct xfs_inode        *ip,
862         xfs_off_t               offset,
863         xfs_off_t               len,
864         int                     alloc_type)
865 {
866         xfs_mount_t             *mp = ip->i_mount;
867         xfs_off_t               count;
868         xfs_filblks_t           allocated_fsb;
869         xfs_filblks_t           allocatesize_fsb;
870         xfs_extlen_t            extsz, temp;
871         xfs_fileoff_t           startoffset_fsb;
872         int                     nimaps;
873         int                     quota_flag;
874         int                     rt;
875         xfs_trans_t             *tp;
876         xfs_bmbt_irec_t         imaps[1], *imapp;
877         uint                    qblocks, resblks, resrtextents;
878         int                     error;
879 
880         trace_xfs_alloc_file_space(ip);
881 
882         if (XFS_FORCED_SHUTDOWN(mp))
883                 return -EIO;
884 
885         error = xfs_qm_dqattach(ip);
886         if (error)
887                 return error;
888 
889         if (len <= 0)
890                 return -EINVAL;
891 
892         rt = XFS_IS_REALTIME_INODE(ip);
893         extsz = xfs_get_extsz_hint(ip);
894 
895         count = len;
896         imapp = &imaps[0];
897         nimaps = 1;
898         startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
899         allocatesize_fsb = XFS_B_TO_FSB(mp, count);
900 
901         /*
902          * Allocate file space until done or until there is an error
903          */
904         while (allocatesize_fsb && !error) {
905                 xfs_fileoff_t   s, e;
906 
907                 /*
908                  * Determine space reservations for data/realtime.
909                  */
910                 if (unlikely(extsz)) {
911                         s = startoffset_fsb;
912                         do_div(s, extsz);
913                         s *= extsz;
914                         e = startoffset_fsb + allocatesize_fsb;
915                         div_u64_rem(startoffset_fsb, extsz, &temp);
916                         if (temp)
917                                 e += temp;
918                         div_u64_rem(e, extsz, &temp);
919                         if (temp)
920                                 e += extsz - temp;
921                 } else {
922                         s = 0;
923                         e = allocatesize_fsb;
924                 }
925 
926                 /*
927                  * The transaction reservation is limited to a 32-bit block
928                  * count, hence we need to limit the number of blocks we are
929                  * trying to reserve to avoid an overflow. We can't allocate
930                  * more than @nimaps extents, and an extent is limited on disk
931                  * to MAXEXTLEN (21 bits), so use that to enforce the limit.
932                  */
933                 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
934                 if (unlikely(rt)) {
935                         resrtextents = qblocks = resblks;
936                         resrtextents /= mp->m_sb.sb_rextsize;
937                         resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
938                         quota_flag = XFS_QMOPT_RES_RTBLKS;
939                 } else {
940                         resrtextents = 0;
941                         resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
942                         quota_flag = XFS_QMOPT_RES_REGBLKS;
943                 }
944 
945                 /*
946                  * Allocate and setup the transaction.
947                  */
948                 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks,
949                                 resrtextents, 0, &tp);
950 
951                 /*
952                  * Check for running out of space
953                  */
954                 if (error) {
955                         /*
956                          * Free the transaction structure.
957                          */
958                         ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
959                         break;
960                 }
961                 xfs_ilock(ip, XFS_ILOCK_EXCL);
962                 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
963                                                       0, quota_flag);
964                 if (error)
965                         goto error1;
966 
967                 xfs_trans_ijoin(tp, ip, 0);
968 
969                 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
970                                         allocatesize_fsb, alloc_type, resblks,
971                                         imapp, &nimaps);
972                 if (error)
973                         goto error0;
974 
975                 /*
976                  * Complete the transaction
977                  */
978                 error = xfs_trans_commit(tp);
979                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
980                 if (error)
981                         break;
982 
983                 allocated_fsb = imapp->br_blockcount;
984 
985                 if (nimaps == 0) {
986                         error = -ENOSPC;
987                         break;
988                 }
989 
990                 startoffset_fsb += allocated_fsb;
991                 allocatesize_fsb -= allocated_fsb;
992         }
993 
994         return error;
995 
996 error0: /* unlock inode, unreserve quota blocks, cancel trans */
997         xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
998 
999 error1: /* Just cancel transaction */
1000         xfs_trans_cancel(tp);
1001         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1002         return error;
1003 }
1004 
1005 static int
1006 xfs_unmap_extent(
1007         struct xfs_inode        *ip,
1008         xfs_fileoff_t           startoffset_fsb,
1009         xfs_filblks_t           len_fsb,
1010         int                     *done)
1011 {
1012         struct xfs_mount        *mp = ip->i_mount;
1013         struct xfs_trans        *tp;
1014         uint                    resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1015         int                     error;
1016 
1017         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
1018         if (error) {
1019                 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1020                 return error;
1021         }
1022 
1023         xfs_ilock(ip, XFS_ILOCK_EXCL);
1024         error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot, ip->i_gdquot,
1025                         ip->i_pdquot, resblks, 0, XFS_QMOPT_RES_REGBLKS);
1026         if (error)
1027                 goto out_trans_cancel;
1028 
1029         xfs_trans_ijoin(tp, ip, 0);
1030 
1031         error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, done);
1032         if (error)
1033                 goto out_trans_cancel;
1034 
1035         error = xfs_trans_commit(tp);
1036 out_unlock:
1037         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1038         return error;
1039 
1040 out_trans_cancel:
1041         xfs_trans_cancel(tp);
1042         goto out_unlock;
1043 }
1044 
1045 int
1046 xfs_flush_unmap_range(
1047         struct xfs_inode        *ip,
1048         xfs_off_t               offset,
1049         xfs_off_t               len)
1050 {
1051         struct xfs_mount        *mp = ip->i_mount;
1052         struct inode            *inode = VFS_I(ip);
1053         xfs_off_t               rounding, start, end;
1054         int                     error;
1055 
1056         /* wait for the completion of any pending DIOs */
1057         inode_dio_wait(inode);
1058 
1059         rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_SIZE);
1060         start = round_down(offset, rounding);
1061         end = round_up(offset + len, rounding) - 1;
1062 
1063         error = filemap_write_and_wait_range(inode->i_mapping, start, end);
1064         if (error)
1065                 return error;
1066         truncate_pagecache_range(inode, start, end);
1067         return 0;
1068 }
1069 
1070 int
1071 xfs_free_file_space(
1072         struct xfs_inode        *ip,
1073         xfs_off_t               offset,
1074         xfs_off_t               len)
1075 {
1076         struct xfs_mount        *mp = ip->i_mount;
1077         xfs_fileoff_t           startoffset_fsb;
1078         xfs_fileoff_t           endoffset_fsb;
1079         int                     done = 0, error;
1080 
1081         trace_xfs_free_file_space(ip);
1082 
1083         error = xfs_qm_dqattach(ip);
1084         if (error)
1085                 return error;
1086 
1087         if (len <= 0)   /* if nothing being freed */
1088                 return 0;
1089 
1090         error = xfs_flush_unmap_range(ip, offset, len);
1091         if (error)
1092                 return error;
1093 
1094         startoffset_fsb = XFS_B_TO_FSB(mp, offset);
1095         endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
1096 
1097         /*
1098          * Need to zero the stuff we're not freeing, on disk.
1099          */
1100         if (endoffset_fsb > startoffset_fsb) {
1101                 while (!done) {
1102                         error = xfs_unmap_extent(ip, startoffset_fsb,
1103                                         endoffset_fsb - startoffset_fsb, &done);
1104                         if (error)
1105                                 return error;
1106                 }
1107         }
1108 
1109         /*
1110          * Now that we've unmap all full blocks we'll have to zero out any
1111          * partial block at the beginning and/or end.  iomap_zero_range is smart
1112          * enough to skip any holes, including those we just created, but we
1113          * must take care not to zero beyond EOF and enlarge i_size.
1114          */
1115         if (offset >= XFS_ISIZE(ip))
1116                 return 0;
1117         if (offset + len > XFS_ISIZE(ip))
1118                 len = XFS_ISIZE(ip) - offset;
1119         error = iomap_zero_range(VFS_I(ip), offset, len, NULL, &xfs_iomap_ops);
1120         if (error)
1121                 return error;
1122 
1123         /*
1124          * If we zeroed right up to EOF and EOF straddles a page boundary we
1125          * must make sure that the post-EOF area is also zeroed because the
1126          * page could be mmap'd and iomap_zero_range doesn't do that for us.
1127          * Writeback of the eof page will do this, albeit clumsily.
1128          */
1129         if (offset + len >= XFS_ISIZE(ip) && offset_in_page(offset + len) > 0) {
1130                 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1131                                 round_down(offset + len, PAGE_SIZE), LLONG_MAX);
1132         }
1133 
1134         return error;
1135 }
1136 
1137 /*
1138  * Preallocate and zero a range of a file. This mechanism has the allocation
1139  * semantics of fallocate and in addition converts data in the range to zeroes.
1140  */
1141 int
1142 xfs_zero_file_space(
1143         struct xfs_inode        *ip,
1144         xfs_off_t               offset,
1145         xfs_off_t               len)
1146 {
1147         struct xfs_mount        *mp = ip->i_mount;
1148         uint                    blksize;
1149         int                     error;
1150 
1151         trace_xfs_zero_file_space(ip);
1152 
1153         blksize = 1 << mp->m_sb.sb_blocklog;
1154 
1155         /*
1156          * Punch a hole and prealloc the range. We use hole punch rather than
1157          * unwritten extent conversion for two reasons:
1158          *
1159          * 1.) Hole punch handles partial block zeroing for us.
1160          *
1161          * 2.) If prealloc returns ENOSPC, the file range is still zero-valued
1162          * by virtue of the hole punch.
1163          */
1164         error = xfs_free_file_space(ip, offset, len);
1165         if (error || xfs_is_always_cow_inode(ip))
1166                 return error;
1167 
1168         return xfs_alloc_file_space(ip, round_down(offset, blksize),
1169                                      round_up(offset + len, blksize) -
1170                                      round_down(offset, blksize),
1171                                      XFS_BMAPI_PREALLOC);
1172 }
1173 
1174 static int
1175 xfs_prepare_shift(
1176         struct xfs_inode        *ip,
1177         loff_t                  offset)
1178 {
1179         int                     error;
1180 
1181         /*
1182          * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
1183          * into the accessible region of the file.
1184          */
1185         if (xfs_can_free_eofblocks(ip, true)) {
1186                 error = xfs_free_eofblocks(ip);
1187                 if (error)
1188                         return error;
1189         }
1190 
1191         /*
1192          * Writeback and invalidate cache for the remainder of the file as we're
1193          * about to shift down every extent from offset to EOF.
1194          */
1195         error = xfs_flush_unmap_range(ip, offset, XFS_ISIZE(ip));
1196 
1197         /*
1198          * Clean out anything hanging around in the cow fork now that
1199          * we've flushed all the dirty data out to disk to avoid having
1200          * CoW extents at the wrong offsets.
1201          */
1202         if (xfs_inode_has_cow_data(ip)) {
1203                 error = xfs_reflink_cancel_cow_range(ip, offset, NULLFILEOFF,
1204                                 true);
1205                 if (error)
1206                         return error;
1207         }
1208 
1209         return 0;
1210 }
1211 
1212 /*
1213  * xfs_collapse_file_space()
1214  *      This routine frees disk space and shift extent for the given file.
1215  *      The first thing we do is to free data blocks in the specified range
1216  *      by calling xfs_free_file_space(). It would also sync dirty data
1217  *      and invalidate page cache over the region on which collapse range
1218  *      is working. And Shift extent records to the left to cover a hole.
1219  * RETURNS:
1220  *      0 on success
1221  *      errno on error
1222  *
1223  */
1224 int
1225 xfs_collapse_file_space(
1226         struct xfs_inode        *ip,
1227         xfs_off_t               offset,
1228         xfs_off_t               len)
1229 {
1230         struct xfs_mount        *mp = ip->i_mount;
1231         struct xfs_trans        *tp;
1232         int                     error;
1233         xfs_fileoff_t           next_fsb = XFS_B_TO_FSB(mp, offset + len);
1234         xfs_fileoff_t           shift_fsb = XFS_B_TO_FSB(mp, len);
1235         uint                    resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1236         bool                    done = false;
1237 
1238         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1239         ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
1240 
1241         trace_xfs_collapse_file_space(ip);
1242 
1243         error = xfs_free_file_space(ip, offset, len);
1244         if (error)
1245                 return error;
1246 
1247         error = xfs_prepare_shift(ip, offset);
1248         if (error)
1249                 return error;
1250 
1251         while (!error && !done) {
1252                 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0,
1253                                         &tp);
1254                 if (error)
1255                         break;
1256 
1257                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1258                 error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot,
1259                                 ip->i_gdquot, ip->i_pdquot, resblks, 0,
1260                                 XFS_QMOPT_RES_REGBLKS);
1261                 if (error)
1262                         goto out_trans_cancel;
1263                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1264 
1265                 error = xfs_bmap_collapse_extents(tp, ip, &next_fsb, shift_fsb,
1266                                 &done);
1267                 if (error)
1268                         goto out_trans_cancel;
1269 
1270                 error = xfs_trans_commit(tp);
1271         }
1272 
1273         return error;
1274 
1275 out_trans_cancel:
1276         xfs_trans_cancel(tp);
1277         return error;
1278 }
1279 
1280 /*
1281  * xfs_insert_file_space()
1282  *      This routine create hole space by shifting extents for the given file.
1283  *      The first thing we do is to sync dirty data and invalidate page cache
1284  *      over the region on which insert range is working. And split an extent
1285  *      to two extents at given offset by calling xfs_bmap_split_extent.
1286  *      And shift all extent records which are laying between [offset,
1287  *      last allocated extent] to the right to reserve hole range.
1288  * RETURNS:
1289  *      0 on success
1290  *      errno on error
1291  */
1292 int
1293 xfs_insert_file_space(
1294         struct xfs_inode        *ip,
1295         loff_t                  offset,
1296         loff_t                  len)
1297 {
1298         struct xfs_mount        *mp = ip->i_mount;
1299         struct xfs_trans        *tp;
1300         int                     error;
1301         xfs_fileoff_t           stop_fsb = XFS_B_TO_FSB(mp, offset);
1302         xfs_fileoff_t           next_fsb = NULLFSBLOCK;
1303         xfs_fileoff_t           shift_fsb = XFS_B_TO_FSB(mp, len);
1304         bool                    done = false;
1305 
1306         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1307         ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
1308 
1309         trace_xfs_insert_file_space(ip);
1310 
1311         error = xfs_bmap_can_insert_extents(ip, stop_fsb, shift_fsb);
1312         if (error)
1313                 return error;
1314 
1315         error = xfs_prepare_shift(ip, offset);
1316         if (error)
1317                 return error;
1318 
1319         /*
1320          * The extent shifting code works on extent granularity. So, if stop_fsb
1321          * is not the starting block of extent, we need to split the extent at
1322          * stop_fsb.
1323          */
1324         error = xfs_bmap_split_extent(ip, stop_fsb);
1325         if (error)
1326                 return error;
1327 
1328         while (!error && !done) {
1329                 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0,
1330                                         &tp);
1331                 if (error)
1332                         break;
1333 
1334                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1335                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1336                 error = xfs_bmap_insert_extents(tp, ip, &next_fsb, shift_fsb,
1337                                 &done, stop_fsb);
1338                 if (error)
1339                         goto out_trans_cancel;
1340 
1341                 error = xfs_trans_commit(tp);
1342         }
1343 
1344         return error;
1345 
1346 out_trans_cancel:
1347         xfs_trans_cancel(tp);
1348         return error;
1349 }
1350 
1351 /*
1352  * We need to check that the format of the data fork in the temporary inode is
1353  * valid for the target inode before doing the swap. This is not a problem with
1354  * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1355  * data fork depending on the space the attribute fork is taking so we can get
1356  * invalid formats on the target inode.
1357  *
1358  * E.g. target has space for 7 extents in extent format, temp inode only has
1359  * space for 6.  If we defragment down to 7 extents, then the tmp format is a
1360  * btree, but when swapped it needs to be in extent format. Hence we can't just
1361  * blindly swap data forks on attr2 filesystems.
1362  *
1363  * Note that we check the swap in both directions so that we don't end up with
1364  * a corrupt temporary inode, either.
1365  *
1366  * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1367  * inode will prevent this situation from occurring, so all we do here is
1368  * reject and log the attempt. basically we are putting the responsibility on
1369  * userspace to get this right.
1370  */
1371 static int
1372 xfs_swap_extents_check_format(
1373         struct xfs_inode        *ip,    /* target inode */
1374         struct xfs_inode        *tip)   /* tmp inode */
1375 {
1376 
1377         /* Should never get a local format */
1378         if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
1379             tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
1380                 return -EINVAL;
1381 
1382         /*
1383          * if the target inode has less extents that then temporary inode then
1384          * why did userspace call us?
1385          */
1386         if (ip->i_d.di_nextents < tip->i_d.di_nextents)
1387                 return -EINVAL;
1388 
1389         /*
1390          * If we have to use the (expensive) rmap swap method, we can
1391          * handle any number of extents and any format.
1392          */
1393         if (xfs_sb_version_hasrmapbt(&ip->i_mount->m_sb))
1394                 return 0;
1395 
1396         /*
1397          * if the target inode is in extent form and the temp inode is in btree
1398          * form then we will end up with the target inode in the wrong format
1399          * as we already know there are less extents in the temp inode.
1400          */
1401         if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1402             tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1403                 return -EINVAL;
1404 
1405         /* Check temp in extent form to max in target */
1406         if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1407             XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) >
1408                         XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1409                 return -EINVAL;
1410 
1411         /* Check target in extent form to max in temp */
1412         if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1413             XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) >
1414                         XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1415                 return -EINVAL;
1416 
1417         /*
1418          * If we are in a btree format, check that the temp root block will fit
1419          * in the target and that it has enough extents to be in btree format
1420          * in the target.
1421          *
1422          * Note that we have to be careful to allow btree->extent conversions
1423          * (a common defrag case) which will occur when the temp inode is in
1424          * extent format...
1425          */
1426         if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1427                 if (XFS_IFORK_Q(ip) &&
1428                     XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip))
1429                         return -EINVAL;
1430                 if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <=
1431                     XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1432                         return -EINVAL;
1433         }
1434 
1435         /* Reciprocal target->temp btree format checks */
1436         if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1437                 if (XFS_IFORK_Q(tip) &&
1438                     XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
1439                         return -EINVAL;
1440                 if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <=
1441                     XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1442                         return -EINVAL;
1443         }
1444 
1445         return 0;
1446 }
1447 
1448 static int
1449 xfs_swap_extent_flush(
1450         struct xfs_inode        *ip)
1451 {
1452         int     error;
1453 
1454         error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1455         if (error)
1456                 return error;
1457         truncate_pagecache_range(VFS_I(ip), 0, -1);
1458 
1459         /* Verify O_DIRECT for ftmp */
1460         if (VFS_I(ip)->i_mapping->nrpages)
1461                 return -EINVAL;
1462         return 0;
1463 }
1464 
1465 /*
1466  * Move extents from one file to another, when rmap is enabled.
1467  */
1468 STATIC int
1469 xfs_swap_extent_rmap(
1470         struct xfs_trans                **tpp,
1471         struct xfs_inode                *ip,
1472         struct xfs_inode                *tip)
1473 {
1474         struct xfs_trans                *tp = *tpp;
1475         struct xfs_bmbt_irec            irec;
1476         struct xfs_bmbt_irec            uirec;
1477         struct xfs_bmbt_irec            tirec;
1478         xfs_fileoff_t                   offset_fsb;
1479         xfs_fileoff_t                   end_fsb;
1480         xfs_filblks_t                   count_fsb;
1481         int                             error;
1482         xfs_filblks_t                   ilen;
1483         xfs_filblks_t                   rlen;
1484         int                             nimaps;
1485         uint64_t                        tip_flags2;
1486 
1487         /*
1488          * If the source file has shared blocks, we must flag the donor
1489          * file as having shared blocks so that we get the shared-block
1490          * rmap functions when we go to fix up the rmaps.  The flags
1491          * will be switch for reals later.
1492          */
1493         tip_flags2 = tip->i_d.di_flags2;
1494         if (ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)
1495                 tip->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK;
1496 
1497         offset_fsb = 0;
1498         end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip)));
1499         count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
1500 
1501         while (count_fsb) {
1502                 /* Read extent from the donor file */
1503                 nimaps = 1;
1504                 error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec,
1505                                 &nimaps, 0);
1506                 if (error)
1507                         goto out;
1508                 ASSERT(nimaps == 1);
1509                 ASSERT(tirec.br_startblock != DELAYSTARTBLOCK);
1510 
1511                 trace_xfs_swap_extent_rmap_remap(tip, &tirec);
1512                 ilen = tirec.br_blockcount;
1513 
1514                 /* Unmap the old blocks in the source file. */
1515                 while (tirec.br_blockcount) {
1516                         ASSERT(tp->t_firstblock == NULLFSBLOCK);
1517                         trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec);
1518 
1519                         /* Read extent from the source file */
1520                         nimaps = 1;
1521                         error = xfs_bmapi_read(ip, tirec.br_startoff,
1522                                         tirec.br_blockcount, &irec,
1523                                         &nimaps, 0);
1524                         if (error)
1525                                 goto out;
1526                         ASSERT(nimaps == 1);
1527                         ASSERT(tirec.br_startoff == irec.br_startoff);
1528                         trace_xfs_swap_extent_rmap_remap_piece(ip, &irec);
1529 
1530                         /* Trim the extent. */
1531                         uirec = tirec;
1532                         uirec.br_blockcount = rlen = min_t(xfs_filblks_t,
1533                                         tirec.br_blockcount,
1534                                         irec.br_blockcount);
1535                         trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec);
1536 
1537                         /* Remove the mapping from the donor file. */
1538                         error = xfs_bmap_unmap_extent(tp, tip, &uirec);
1539                         if (error)
1540                                 goto out;
1541 
1542                         /* Remove the mapping from the source file. */
1543                         error = xfs_bmap_unmap_extent(tp, ip, &irec);
1544                         if (error)
1545                                 goto out;
1546 
1547                         /* Map the donor file's blocks into the source file. */
1548                         error = xfs_bmap_map_extent(tp, ip, &uirec);
1549                         if (error)
1550                                 goto out;
1551 
1552                         /* Map the source file's blocks into the donor file. */
1553                         error = xfs_bmap_map_extent(tp, tip, &irec);
1554                         if (error)
1555                                 goto out;
1556 
1557                         error = xfs_defer_finish(tpp);
1558                         tp = *tpp;
1559                         if (error)
1560                                 goto out;
1561 
1562                         tirec.br_startoff += rlen;
1563                         if (tirec.br_startblock != HOLESTARTBLOCK &&
1564                             tirec.br_startblock != DELAYSTARTBLOCK)
1565                                 tirec.br_startblock += rlen;
1566                         tirec.br_blockcount -= rlen;
1567                 }
1568 
1569                 /* Roll on... */
1570                 count_fsb -= ilen;
1571                 offset_fsb += ilen;
1572         }
1573 
1574         tip->i_d.di_flags2 = tip_flags2;
1575         return 0;
1576 
1577 out:
1578         trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_);
1579         tip->i_d.di_flags2 = tip_flags2;
1580         return error;
1581 }
1582 
1583 /* Swap the extents of two files by swapping data forks. */
1584 STATIC int
1585 xfs_swap_extent_forks(
1586         struct xfs_trans        *tp,
1587         struct xfs_inode        *ip,
1588         struct xfs_inode        *tip,
1589         int                     *src_log_flags,
1590         int                     *target_log_flags)
1591 {
1592         xfs_filblks_t           aforkblks = 0;
1593         xfs_filblks_t           taforkblks = 0;
1594         xfs_extnum_t            junk;
1595         uint64_t                tmp;
1596         int                     error;
1597 
1598         /*
1599          * Count the number of extended attribute blocks
1600          */
1601         if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
1602              (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1603                 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &junk,
1604                                 &aforkblks);
1605                 if (error)
1606                         return error;
1607         }
1608         if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
1609              (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1610                 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK, &junk,
1611                                 &taforkblks);
1612                 if (error)
1613                         return error;
1614         }
1615 
1616         /*
1617          * Btree format (v3) inodes have the inode number stamped in the bmbt
1618          * block headers. We can't start changing the bmbt blocks until the
1619          * inode owner change is logged so recovery does the right thing in the
1620          * event of a crash. Set the owner change log flags now and leave the
1621          * bmbt scan as the last step.
1622          */
1623         if (ip->i_d.di_version == 3 &&
1624             ip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1625                 (*target_log_flags) |= XFS_ILOG_DOWNER;
1626         if (tip->i_d.di_version == 3 &&
1627             tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1628                 (*src_log_flags) |= XFS_ILOG_DOWNER;
1629 
1630         /*
1631          * Swap the data forks of the inodes
1632          */
1633         swap(ip->i_df, tip->i_df);
1634 
1635         /*
1636          * Fix the on-disk inode values
1637          */
1638         tmp = (uint64_t)ip->i_d.di_nblocks;
1639         ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
1640         tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
1641 
1642         swap(ip->i_d.di_nextents, tip->i_d.di_nextents);
1643         swap(ip->i_d.di_format, tip->i_d.di_format);
1644 
1645         /*
1646          * The extents in the source inode could still contain speculative
1647          * preallocation beyond EOF (e.g. the file is open but not modified
1648          * while defrag is in progress). In that case, we need to copy over the
1649          * number of delalloc blocks the data fork in the source inode is
1650          * tracking beyond EOF so that when the fork is truncated away when the
1651          * temporary inode is unlinked we don't underrun the i_delayed_blks
1652          * counter on that inode.
1653          */
1654         ASSERT(tip->i_delayed_blks == 0);
1655         tip->i_delayed_blks = ip->i_delayed_blks;
1656         ip->i_delayed_blks = 0;
1657 
1658         switch (ip->i_d.di_format) {
1659         case XFS_DINODE_FMT_EXTENTS:
1660                 (*src_log_flags) |= XFS_ILOG_DEXT;
1661                 break;
1662         case XFS_DINODE_FMT_BTREE:
1663                 ASSERT(ip->i_d.di_version < 3 ||
1664                        (*src_log_flags & XFS_ILOG_DOWNER));
1665                 (*src_log_flags) |= XFS_ILOG_DBROOT;
1666                 break;
1667         }
1668 
1669         switch (tip->i_d.di_format) {
1670         case XFS_DINODE_FMT_EXTENTS:
1671                 (*target_log_flags) |= XFS_ILOG_DEXT;
1672                 break;
1673         case XFS_DINODE_FMT_BTREE:
1674                 (*target_log_flags) |= XFS_ILOG_DBROOT;
1675                 ASSERT(tip->i_d.di_version < 3 ||
1676                        (*target_log_flags & XFS_ILOG_DOWNER));
1677                 break;
1678         }
1679 
1680         return 0;
1681 }
1682 
1683 /*
1684  * Fix up the owners of the bmbt blocks to refer to the current inode. The
1685  * change owner scan attempts to order all modified buffers in the current
1686  * transaction. In the event of ordered buffer failure, the offending buffer is
1687  * physically logged as a fallback and the scan returns -EAGAIN. We must roll
1688  * the transaction in this case to replenish the fallback log reservation and
1689  * restart the scan. This process repeats until the scan completes.
1690  */
1691 static int
1692 xfs_swap_change_owner(
1693         struct xfs_trans        **tpp,
1694         struct xfs_inode        *ip,
1695         struct xfs_inode        *tmpip)
1696 {
1697         int                     error;
1698         struct xfs_trans        *tp = *tpp;
1699 
1700         do {
1701                 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK, ip->i_ino,
1702                                               NULL);
1703                 /* success or fatal error */
1704                 if (error != -EAGAIN)
1705                         break;
1706 
1707                 error = xfs_trans_roll(tpp);
1708                 if (error)
1709                         break;
1710                 tp = *tpp;
1711 
1712                 /*
1713                  * Redirty both inodes so they can relog and keep the log tail
1714                  * moving forward.
1715                  */
1716                 xfs_trans_ijoin(tp, ip, 0);
1717                 xfs_trans_ijoin(tp, tmpip, 0);
1718                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1719                 xfs_trans_log_inode(tp, tmpip, XFS_ILOG_CORE);
1720         } while (true);
1721 
1722         return error;
1723 }
1724 
1725 int
1726 xfs_swap_extents(
1727         struct xfs_inode        *ip,    /* target inode */
1728         struct xfs_inode        *tip,   /* tmp inode */
1729         struct xfs_swapext      *sxp)
1730 {
1731         struct xfs_mount        *mp = ip->i_mount;
1732         struct xfs_trans        *tp;
1733         struct xfs_bstat        *sbp = &sxp->sx_stat;
1734         int                     src_log_flags, target_log_flags;
1735         int                     error = 0;
1736         int                     lock_flags;
1737         uint64_t                f;
1738         int                     resblks = 0;
1739 
1740         /*
1741          * Lock the inodes against other IO, page faults and truncate to
1742          * begin with.  Then we can ensure the inodes are flushed and have no
1743          * page cache safely. Once we have done this we can take the ilocks and
1744          * do the rest of the checks.
1745          */
1746         lock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1747         lock_flags = XFS_MMAPLOCK_EXCL;
1748         xfs_lock_two_inodes(ip, XFS_MMAPLOCK_EXCL, tip, XFS_MMAPLOCK_EXCL);
1749 
1750         /* Verify that both files have the same format */
1751         if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) {
1752                 error = -EINVAL;
1753                 goto out_unlock;
1754         }
1755 
1756         /* Verify both files are either real-time or non-realtime */
1757         if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1758                 error = -EINVAL;
1759                 goto out_unlock;
1760         }
1761 
1762         error = xfs_swap_extent_flush(ip);
1763         if (error)
1764                 goto out_unlock;
1765         error = xfs_swap_extent_flush(tip);
1766         if (error)
1767                 goto out_unlock;
1768 
1769         if (xfs_inode_has_cow_data(tip)) {
1770                 error = xfs_reflink_cancel_cow_range(tip, 0, NULLFILEOFF, true);
1771                 if (error)
1772                         return error;
1773         }
1774 
1775         /*
1776          * Extent "swapping" with rmap requires a permanent reservation and
1777          * a block reservation because it's really just a remap operation
1778          * performed with log redo items!
1779          */
1780         if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
1781                 int             w       = XFS_DATA_FORK;
1782                 uint32_t        ipnext  = XFS_IFORK_NEXTENTS(ip, w);
1783                 uint32_t        tipnext = XFS_IFORK_NEXTENTS(tip, w);
1784 
1785                 /*
1786                  * Conceptually this shouldn't affect the shape of either bmbt,
1787                  * but since we atomically move extents one by one, we reserve
1788                  * enough space to rebuild both trees.
1789                  */
1790                 resblks = XFS_SWAP_RMAP_SPACE_RES(mp, ipnext, w);
1791                 resblks +=  XFS_SWAP_RMAP_SPACE_RES(mp, tipnext, w);
1792 
1793                 /*
1794                  * Handle the corner case where either inode might straddle the
1795                  * btree format boundary. If so, the inode could bounce between
1796                  * btree <-> extent format on unmap -> remap cycles, freeing and
1797                  * allocating a bmapbt block each time.
1798                  */
1799                 if (ipnext == (XFS_IFORK_MAXEXT(ip, w) + 1))
1800                         resblks += XFS_IFORK_MAXEXT(ip, w);
1801                 if (tipnext == (XFS_IFORK_MAXEXT(tip, w) + 1))
1802                         resblks += XFS_IFORK_MAXEXT(tip, w);
1803         }
1804         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
1805         if (error)
1806                 goto out_unlock;
1807 
1808         /*
1809          * Lock and join the inodes to the tansaction so that transaction commit
1810          * or cancel will unlock the inodes from this point onwards.
1811          */
1812         xfs_lock_two_inodes(ip, XFS_ILOCK_EXCL, tip, XFS_ILOCK_EXCL);
1813         lock_flags |= XFS_ILOCK_EXCL;
1814         xfs_trans_ijoin(tp, ip, 0);
1815         xfs_trans_ijoin(tp, tip, 0);
1816 
1817 
1818         /* Verify all data are being swapped */
1819         if (sxp->sx_offset != 0 ||
1820             sxp->sx_length != ip->i_d.di_size ||
1821             sxp->sx_length != tip->i_d.di_size) {
1822                 error = -EFAULT;
1823                 goto out_trans_cancel;
1824         }
1825 
1826         trace_xfs_swap_extent_before(ip, 0);
1827         trace_xfs_swap_extent_before(tip, 1);
1828 
1829         /* check inode formats now that data is flushed */
1830         error = xfs_swap_extents_check_format(ip, tip);
1831         if (error) {
1832                 xfs_notice(mp,
1833                     "%s: inode 0x%llx format is incompatible for exchanging.",
1834                                 __func__, ip->i_ino);
1835                 goto out_trans_cancel;
1836         }
1837 
1838         /*
1839          * Compare the current change & modify times with that
1840          * passed in.  If they differ, we abort this swap.
1841          * This is the mechanism used to ensure the calling
1842          * process that the file was not changed out from
1843          * under it.
1844          */
1845         if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
1846             (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
1847             (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
1848             (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
1849                 error = -EBUSY;
1850                 goto out_trans_cancel;
1851         }
1852 
1853         /*
1854          * Note the trickiness in setting the log flags - we set the owner log
1855          * flag on the opposite inode (i.e. the inode we are setting the new
1856          * owner to be) because once we swap the forks and log that, log
1857          * recovery is going to see the fork as owned by the swapped inode,
1858          * not the pre-swapped inodes.
1859          */
1860         src_log_flags = XFS_ILOG_CORE;
1861         target_log_flags = XFS_ILOG_CORE;
1862 
1863         if (xfs_sb_version_hasrmapbt(&mp->m_sb))
1864                 error = xfs_swap_extent_rmap(&tp, ip, tip);
1865         else
1866                 error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags,
1867                                 &target_log_flags);
1868         if (error)
1869                 goto out_trans_cancel;
1870 
1871         /* Do we have to swap reflink flags? */
1872         if ((ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK) ^
1873             (tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)) {
1874                 f = ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
1875                 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1876                 ip->i_d.di_flags2 |= tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
1877                 tip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1878                 tip->i_d.di_flags2 |= f & XFS_DIFLAG2_REFLINK;
1879         }
1880 
1881         /* Swap the cow forks. */
1882         if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1883                 ASSERT(ip->i_cformat == XFS_DINODE_FMT_EXTENTS);
1884                 ASSERT(tip->i_cformat == XFS_DINODE_FMT_EXTENTS);
1885 
1886                 swap(ip->i_cnextents, tip->i_cnextents);
1887                 swap(ip->i_cowfp, tip->i_cowfp);
1888 
1889                 if (ip->i_cowfp && ip->i_cowfp->if_bytes)
1890                         xfs_inode_set_cowblocks_tag(ip);
1891                 else
1892                         xfs_inode_clear_cowblocks_tag(ip);
1893                 if (tip->i_cowfp && tip->i_cowfp->if_bytes)
1894                         xfs_inode_set_cowblocks_tag(tip);
1895                 else
1896                         xfs_inode_clear_cowblocks_tag(tip);
1897         }
1898 
1899         xfs_trans_log_inode(tp, ip,  src_log_flags);
1900         xfs_trans_log_inode(tp, tip, target_log_flags);
1901 
1902         /*
1903          * The extent forks have been swapped, but crc=1,rmapbt=0 filesystems
1904          * have inode number owner values in the bmbt blocks that still refer to
1905          * the old inode. Scan each bmbt to fix up the owner values with the
1906          * inode number of the current inode.
1907          */
1908         if (src_log_flags & XFS_ILOG_DOWNER) {
1909                 error = xfs_swap_change_owner(&tp, ip, tip);
1910                 if (error)
1911                         goto out_trans_cancel;
1912         }
1913         if (target_log_flags & XFS_ILOG_DOWNER) {
1914                 error = xfs_swap_change_owner(&tp, tip, ip);
1915                 if (error)
1916                         goto out_trans_cancel;
1917         }
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);
1927 
1928         trace_xfs_swap_extent_after(ip, 0);
1929         trace_xfs_swap_extent_after(tip, 1);
1930 
1931 out_unlock:
1932         xfs_iunlock(ip, lock_flags);
1933         xfs_iunlock(tip, lock_flags);
1934         unlock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1935         return error;
1936 
1937 out_trans_cancel:
1938         xfs_trans_cancel(tp);
1939         goto out_unlock;
1940 }
1941 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp