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

TOMOYO Linux Cross Reference
Linux/fs/xfs/libxfs/xfs_bmap.c

Version: ~ [ linux-5.12-rc1 ] ~ [ linux-5.11.2 ] ~ [ linux-5.10.19 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.101 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.177 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.222 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.258 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.258 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ 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  * All Rights Reserved.
  5  */
  6 #include "xfs.h"
  7 #include "xfs_fs.h"
  8 #include "xfs_shared.h"
  9 #include "xfs_format.h"
 10 #include "xfs_log_format.h"
 11 #include "xfs_trans_resv.h"
 12 #include "xfs_bit.h"
 13 #include "xfs_sb.h"
 14 #include "xfs_mount.h"
 15 #include "xfs_defer.h"
 16 #include "xfs_dir2.h"
 17 #include "xfs_inode.h"
 18 #include "xfs_btree.h"
 19 #include "xfs_trans.h"
 20 #include "xfs_alloc.h"
 21 #include "xfs_bmap.h"
 22 #include "xfs_bmap_util.h"
 23 #include "xfs_bmap_btree.h"
 24 #include "xfs_rtalloc.h"
 25 #include "xfs_errortag.h"
 26 #include "xfs_error.h"
 27 #include "xfs_quota.h"
 28 #include "xfs_trans_space.h"
 29 #include "xfs_buf_item.h"
 30 #include "xfs_trace.h"
 31 #include "xfs_attr_leaf.h"
 32 #include "xfs_filestream.h"
 33 #include "xfs_rmap.h"
 34 #include "xfs_ag_resv.h"
 35 #include "xfs_refcount.h"
 36 #include "xfs_icache.h"
 37 
 38 
 39 kmem_zone_t             *xfs_bmap_free_item_zone;
 40 
 41 /*
 42  * Miscellaneous helper functions
 43  */
 44 
 45 /*
 46  * Compute and fill in the value of the maximum depth of a bmap btree
 47  * in this filesystem.  Done once, during mount.
 48  */
 49 void
 50 xfs_bmap_compute_maxlevels(
 51         xfs_mount_t     *mp,            /* file system mount structure */
 52         int             whichfork)      /* data or attr fork */
 53 {
 54         int             level;          /* btree level */
 55         uint            maxblocks;      /* max blocks at this level */
 56         uint            maxleafents;    /* max leaf entries possible */
 57         int             maxrootrecs;    /* max records in root block */
 58         int             minleafrecs;    /* min records in leaf block */
 59         int             minnoderecs;    /* min records in node block */
 60         int             sz;             /* root block size */
 61 
 62         /*
 63          * The maximum number of extents in a file, hence the maximum
 64          * number of leaf entries, is controlled by the type of di_nextents
 65          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
 66          * (a signed 16-bit number, xfs_aextnum_t).
 67          *
 68          * Note that we can no longer assume that if we are in ATTR1 that
 69          * the fork offset of all the inodes will be
 70          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
 71          * with ATTR2 and then mounted back with ATTR1, keeping the
 72          * di_forkoff's fixed but probably at various positions. Therefore,
 73          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
 74          * of a minimum size available.
 75          */
 76         if (whichfork == XFS_DATA_FORK) {
 77                 maxleafents = MAXEXTNUM;
 78                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
 79         } else {
 80                 maxleafents = MAXAEXTNUM;
 81                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
 82         }
 83         maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
 84         minleafrecs = mp->m_bmap_dmnr[0];
 85         minnoderecs = mp->m_bmap_dmnr[1];
 86         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
 87         for (level = 1; maxblocks > 1; level++) {
 88                 if (maxblocks <= maxrootrecs)
 89                         maxblocks = 1;
 90                 else
 91                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
 92         }
 93         mp->m_bm_maxlevels[whichfork] = level;
 94 }
 95 
 96 STATIC int                              /* error */
 97 xfs_bmbt_lookup_eq(
 98         struct xfs_btree_cur    *cur,
 99         struct xfs_bmbt_irec    *irec,
100         int                     *stat)  /* success/failure */
101 {
102         cur->bc_rec.b = *irec;
103         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
104 }
105 
106 STATIC int                              /* error */
107 xfs_bmbt_lookup_first(
108         struct xfs_btree_cur    *cur,
109         int                     *stat)  /* success/failure */
110 {
111         cur->bc_rec.b.br_startoff = 0;
112         cur->bc_rec.b.br_startblock = 0;
113         cur->bc_rec.b.br_blockcount = 0;
114         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
115 }
116 
117 /*
118  * Check if the inode needs to be converted to btree format.
119  */
120 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
121 {
122         return whichfork != XFS_COW_FORK &&
123                 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
124                 XFS_IFORK_NEXTENTS(ip, whichfork) >
125                         XFS_IFORK_MAXEXT(ip, whichfork);
126 }
127 
128 /*
129  * Check if the inode should be converted to extent format.
130  */
131 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
132 {
133         return whichfork != XFS_COW_FORK &&
134                 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
135                 XFS_IFORK_NEXTENTS(ip, whichfork) <=
136                         XFS_IFORK_MAXEXT(ip, whichfork);
137 }
138 
139 /*
140  * Update the record referred to by cur to the value given by irec
141  * This either works (return 0) or gets an EFSCORRUPTED error.
142  */
143 STATIC int
144 xfs_bmbt_update(
145         struct xfs_btree_cur    *cur,
146         struct xfs_bmbt_irec    *irec)
147 {
148         union xfs_btree_rec     rec;
149 
150         xfs_bmbt_disk_set_all(&rec.bmbt, irec);
151         return xfs_btree_update(cur, &rec);
152 }
153 
154 /*
155  * Compute the worst-case number of indirect blocks that will be used
156  * for ip's delayed extent of length "len".
157  */
158 STATIC xfs_filblks_t
159 xfs_bmap_worst_indlen(
160         xfs_inode_t     *ip,            /* incore inode pointer */
161         xfs_filblks_t   len)            /* delayed extent length */
162 {
163         int             level;          /* btree level number */
164         int             maxrecs;        /* maximum record count at this level */
165         xfs_mount_t     *mp;            /* mount structure */
166         xfs_filblks_t   rval;           /* return value */
167 
168         mp = ip->i_mount;
169         maxrecs = mp->m_bmap_dmxr[0];
170         for (level = 0, rval = 0;
171              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
172              level++) {
173                 len += maxrecs - 1;
174                 do_div(len, maxrecs);
175                 rval += len;
176                 if (len == 1)
177                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
178                                 level - 1;
179                 if (level == 0)
180                         maxrecs = mp->m_bmap_dmxr[1];
181         }
182         return rval;
183 }
184 
185 /*
186  * Calculate the default attribute fork offset for newly created inodes.
187  */
188 uint
189 xfs_default_attroffset(
190         struct xfs_inode        *ip)
191 {
192         struct xfs_mount        *mp = ip->i_mount;
193         uint                    offset;
194 
195         if (mp->m_sb.sb_inodesize == 256) {
196                 offset = XFS_LITINO(mp, ip->i_d.di_version) -
197                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
198         } else {
199                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
200         }
201 
202         ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
203         return offset;
204 }
205 
206 /*
207  * Helper routine to reset inode di_forkoff field when switching
208  * attribute fork from local to extent format - we reset it where
209  * possible to make space available for inline data fork extents.
210  */
211 STATIC void
212 xfs_bmap_forkoff_reset(
213         xfs_inode_t     *ip,
214         int             whichfork)
215 {
216         if (whichfork == XFS_ATTR_FORK &&
217             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
218             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
219                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
220 
221                 if (dfl_forkoff > ip->i_d.di_forkoff)
222                         ip->i_d.di_forkoff = dfl_forkoff;
223         }
224 }
225 
226 #ifdef DEBUG
227 STATIC struct xfs_buf *
228 xfs_bmap_get_bp(
229         struct xfs_btree_cur    *cur,
230         xfs_fsblock_t           bno)
231 {
232         struct xfs_log_item     *lip;
233         int                     i;
234 
235         if (!cur)
236                 return NULL;
237 
238         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
239                 if (!cur->bc_bufs[i])
240                         break;
241                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
242                         return cur->bc_bufs[i];
243         }
244 
245         /* Chase down all the log items to see if the bp is there */
246         list_for_each_entry(lip, &cur->bc_tp->t_items, li_trans) {
247                 struct xfs_buf_log_item *bip = (struct xfs_buf_log_item *)lip;
248 
249                 if (bip->bli_item.li_type == XFS_LI_BUF &&
250                     XFS_BUF_ADDR(bip->bli_buf) == bno)
251                         return bip->bli_buf;
252         }
253 
254         return NULL;
255 }
256 
257 STATIC void
258 xfs_check_block(
259         struct xfs_btree_block  *block,
260         xfs_mount_t             *mp,
261         int                     root,
262         short                   sz)
263 {
264         int                     i, j, dmxr;
265         __be64                  *pp, *thispa;   /* pointer to block address */
266         xfs_bmbt_key_t          *prevp, *keyp;
267 
268         ASSERT(be16_to_cpu(block->bb_level) > 0);
269 
270         prevp = NULL;
271         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
272                 dmxr = mp->m_bmap_dmxr[0];
273                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
274 
275                 if (prevp) {
276                         ASSERT(be64_to_cpu(prevp->br_startoff) <
277                                be64_to_cpu(keyp->br_startoff));
278                 }
279                 prevp = keyp;
280 
281                 /*
282                  * Compare the block numbers to see if there are dups.
283                  */
284                 if (root)
285                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
286                 else
287                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
288 
289                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
290                         if (root)
291                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
292                         else
293                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
294                         if (*thispa == *pp) {
295                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
296                                         __func__, j, i,
297                                         (unsigned long long)be64_to_cpu(*thispa));
298                                 xfs_err(mp, "%s: ptrs are equal in node\n",
299                                         __func__);
300                                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
301                         }
302                 }
303         }
304 }
305 
306 /*
307  * Check that the extents for the inode ip are in the right order in all
308  * btree leaves. THis becomes prohibitively expensive for large extent count
309  * files, so don't bother with inodes that have more than 10,000 extents in
310  * them. The btree record ordering checks will still be done, so for such large
311  * bmapbt constructs that is going to catch most corruptions.
312  */
313 STATIC void
314 xfs_bmap_check_leaf_extents(
315         xfs_btree_cur_t         *cur,   /* btree cursor or null */
316         xfs_inode_t             *ip,            /* incore inode pointer */
317         int                     whichfork)      /* data or attr fork */
318 {
319         struct xfs_btree_block  *block; /* current btree block */
320         xfs_fsblock_t           bno;    /* block # of "block" */
321         xfs_buf_t               *bp;    /* buffer for "block" */
322         int                     error;  /* error return value */
323         xfs_extnum_t            i=0, j; /* index into the extents list */
324         struct xfs_ifork        *ifp;   /* fork structure */
325         int                     level;  /* btree level, for checking */
326         xfs_mount_t             *mp;    /* file system mount structure */
327         __be64                  *pp;    /* pointer to block address */
328         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
329         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
330         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
331         int                     bp_release = 0;
332 
333         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
334                 return;
335         }
336 
337         /* skip large extent count inodes */
338         if (ip->i_d.di_nextents > 10000)
339                 return;
340 
341         bno = NULLFSBLOCK;
342         mp = ip->i_mount;
343         ifp = XFS_IFORK_PTR(ip, whichfork);
344         block = ifp->if_broot;
345         /*
346          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
347          */
348         level = be16_to_cpu(block->bb_level);
349         ASSERT(level > 0);
350         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
351         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
352         bno = be64_to_cpu(*pp);
353 
354         ASSERT(bno != NULLFSBLOCK);
355         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
356         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
357 
358         /*
359          * Go down the tree until leaf level is reached, following the first
360          * pointer (leftmost) at each level.
361          */
362         while (level-- > 0) {
363                 /* See if buf is in cur first */
364                 bp_release = 0;
365                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
366                 if (!bp) {
367                         bp_release = 1;
368                         error = xfs_btree_read_bufl(mp, NULL, bno, &bp,
369                                                 XFS_BMAP_BTREE_REF,
370                                                 &xfs_bmbt_buf_ops);
371                         if (error)
372                                 goto error_norelse;
373                 }
374                 block = XFS_BUF_TO_BLOCK(bp);
375                 if (level == 0)
376                         break;
377 
378                 /*
379                  * Check this block for basic sanity (increasing keys and
380                  * no duplicate blocks).
381                  */
382 
383                 xfs_check_block(block, mp, 0, 0);
384                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
385                 bno = be64_to_cpu(*pp);
386                 XFS_WANT_CORRUPTED_GOTO(mp,
387                                         xfs_verify_fsbno(mp, bno), error0);
388                 if (bp_release) {
389                         bp_release = 0;
390                         xfs_trans_brelse(NULL, bp);
391                 }
392         }
393 
394         /*
395          * Here with bp and block set to the leftmost leaf node in the tree.
396          */
397         i = 0;
398 
399         /*
400          * Loop over all leaf nodes checking that all extents are in the right order.
401          */
402         for (;;) {
403                 xfs_fsblock_t   nextbno;
404                 xfs_extnum_t    num_recs;
405 
406 
407                 num_recs = xfs_btree_get_numrecs(block);
408 
409                 /*
410                  * Read-ahead the next leaf block, if any.
411                  */
412 
413                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
414 
415                 /*
416                  * Check all the extents to make sure they are OK.
417                  * If we had a previous block, the last entry should
418                  * conform with the first entry in this one.
419                  */
420 
421                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
422                 if (i) {
423                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
424                                xfs_bmbt_disk_get_blockcount(&last) <=
425                                xfs_bmbt_disk_get_startoff(ep));
426                 }
427                 for (j = 1; j < num_recs; j++) {
428                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
429                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
430                                xfs_bmbt_disk_get_blockcount(ep) <=
431                                xfs_bmbt_disk_get_startoff(nextp));
432                         ep = nextp;
433                 }
434 
435                 last = *ep;
436                 i += num_recs;
437                 if (bp_release) {
438                         bp_release = 0;
439                         xfs_trans_brelse(NULL, bp);
440                 }
441                 bno = nextbno;
442                 /*
443                  * If we've reached the end, stop.
444                  */
445                 if (bno == NULLFSBLOCK)
446                         break;
447 
448                 bp_release = 0;
449                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
450                 if (!bp) {
451                         bp_release = 1;
452                         error = xfs_btree_read_bufl(mp, NULL, bno, &bp,
453                                                 XFS_BMAP_BTREE_REF,
454                                                 &xfs_bmbt_buf_ops);
455                         if (error)
456                                 goto error_norelse;
457                 }
458                 block = XFS_BUF_TO_BLOCK(bp);
459         }
460 
461         return;
462 
463 error0:
464         xfs_warn(mp, "%s: at error0", __func__);
465         if (bp_release)
466                 xfs_trans_brelse(NULL, bp);
467 error_norelse:
468         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
469                 __func__, i);
470         xfs_err(mp, "%s: CORRUPTED BTREE OR SOMETHING", __func__);
471         xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
472         return;
473 }
474 
475 /*
476  * Validate that the bmbt_irecs being returned from bmapi are valid
477  * given the caller's original parameters.  Specifically check the
478  * ranges of the returned irecs to ensure that they only extend beyond
479  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
480  */
481 STATIC void
482 xfs_bmap_validate_ret(
483         xfs_fileoff_t           bno,
484         xfs_filblks_t           len,
485         int                     flags,
486         xfs_bmbt_irec_t         *mval,
487         int                     nmap,
488         int                     ret_nmap)
489 {
490         int                     i;              /* index to map values */
491 
492         ASSERT(ret_nmap <= nmap);
493 
494         for (i = 0; i < ret_nmap; i++) {
495                 ASSERT(mval[i].br_blockcount > 0);
496                 if (!(flags & XFS_BMAPI_ENTIRE)) {
497                         ASSERT(mval[i].br_startoff >= bno);
498                         ASSERT(mval[i].br_blockcount <= len);
499                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
500                                bno + len);
501                 } else {
502                         ASSERT(mval[i].br_startoff < bno + len);
503                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
504                                bno);
505                 }
506                 ASSERT(i == 0 ||
507                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
508                        mval[i].br_startoff);
509                 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
510                        mval[i].br_startblock != HOLESTARTBLOCK);
511                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
512                        mval[i].br_state == XFS_EXT_UNWRITTEN);
513         }
514 }
515 
516 #else
517 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)         do { } while (0)
518 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)    do { } while (0)
519 #endif /* DEBUG */
520 
521 /*
522  * bmap free list manipulation functions
523  */
524 
525 /*
526  * Add the extent to the list of extents to be free at transaction end.
527  * The list is maintained sorted (by block number).
528  */
529 void
530 __xfs_bmap_add_free(
531         struct xfs_trans                *tp,
532         xfs_fsblock_t                   bno,
533         xfs_filblks_t                   len,
534         const struct xfs_owner_info     *oinfo,
535         bool                            skip_discard)
536 {
537         struct xfs_extent_free_item     *new;           /* new element */
538 #ifdef DEBUG
539         struct xfs_mount                *mp = tp->t_mountp;
540         xfs_agnumber_t                  agno;
541         xfs_agblock_t                   agbno;
542 
543         ASSERT(bno != NULLFSBLOCK);
544         ASSERT(len > 0);
545         ASSERT(len <= MAXEXTLEN);
546         ASSERT(!isnullstartblock(bno));
547         agno = XFS_FSB_TO_AGNO(mp, bno);
548         agbno = XFS_FSB_TO_AGBNO(mp, bno);
549         ASSERT(agno < mp->m_sb.sb_agcount);
550         ASSERT(agbno < mp->m_sb.sb_agblocks);
551         ASSERT(len < mp->m_sb.sb_agblocks);
552         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
553 #endif
554         ASSERT(xfs_bmap_free_item_zone != NULL);
555 
556         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
557         new->xefi_startblock = bno;
558         new->xefi_blockcount = (xfs_extlen_t)len;
559         if (oinfo)
560                 new->xefi_oinfo = *oinfo;
561         else
562                 new->xefi_oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
563         new->xefi_skip_discard = skip_discard;
564         trace_xfs_bmap_free_defer(tp->t_mountp,
565                         XFS_FSB_TO_AGNO(tp->t_mountp, bno), 0,
566                         XFS_FSB_TO_AGBNO(tp->t_mountp, bno), len);
567         xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list);
568 }
569 
570 /*
571  * Inode fork format manipulation functions
572  */
573 
574 /*
575  * Convert the inode format to extent format if it currently is in btree format,
576  * but the extent list is small enough that it fits into the extent format.
577  *
578  * Since the extents are already in-core, all we have to do is give up the space
579  * for the btree root and pitch the leaf block.
580  */
581 STATIC int                              /* error */
582 xfs_bmap_btree_to_extents(
583         struct xfs_trans        *tp,    /* transaction pointer */
584         struct xfs_inode        *ip,    /* incore inode pointer */
585         struct xfs_btree_cur    *cur,   /* btree cursor */
586         int                     *logflagsp, /* inode logging flags */
587         int                     whichfork)  /* data or attr fork */
588 {
589         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
590         struct xfs_mount        *mp = ip->i_mount;
591         struct xfs_btree_block  *rblock = ifp->if_broot;
592         struct xfs_btree_block  *cblock;/* child btree block */
593         xfs_fsblock_t           cbno;   /* child block number */
594         xfs_buf_t               *cbp;   /* child block's buffer */
595         int                     error;  /* error return value */
596         __be64                  *pp;    /* ptr to block address */
597         struct xfs_owner_info   oinfo;
598 
599         /* check if we actually need the extent format first: */
600         if (!xfs_bmap_wants_extents(ip, whichfork))
601                 return 0;
602 
603         ASSERT(cur);
604         ASSERT(whichfork != XFS_COW_FORK);
605         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
606         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
607         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
608         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
609         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
610 
611         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
612         cbno = be64_to_cpu(*pp);
613 #ifdef DEBUG
614         XFS_WANT_CORRUPTED_RETURN(cur->bc_mp,
615                         xfs_btree_check_lptr(cur, cbno, 1));
616 #endif
617         error = xfs_btree_read_bufl(mp, tp, cbno, &cbp, XFS_BMAP_BTREE_REF,
618                                 &xfs_bmbt_buf_ops);
619         if (error)
620                 return error;
621         cblock = XFS_BUF_TO_BLOCK(cbp);
622         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
623                 return error;
624         xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
625         xfs_bmap_add_free(cur->bc_tp, cbno, 1, &oinfo);
626         ip->i_d.di_nblocks--;
627         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
628         xfs_trans_binval(tp, cbp);
629         if (cur->bc_bufs[0] == cbp)
630                 cur->bc_bufs[0] = NULL;
631         xfs_iroot_realloc(ip, -1, whichfork);
632         ASSERT(ifp->if_broot == NULL);
633         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
634         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
635         *logflagsp |= XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
636         return 0;
637 }
638 
639 /*
640  * Convert an extents-format file into a btree-format file.
641  * The new file will have a root block (in the inode) and a single child block.
642  */
643 STATIC int                                      /* error */
644 xfs_bmap_extents_to_btree(
645         struct xfs_trans        *tp,            /* transaction pointer */
646         struct xfs_inode        *ip,            /* incore inode pointer */
647         struct xfs_btree_cur    **curp,         /* cursor returned to caller */
648         int                     wasdel,         /* converting a delayed alloc */
649         int                     *logflagsp,     /* inode logging flags */
650         int                     whichfork)      /* data or attr fork */
651 {
652         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
653         struct xfs_buf          *abp;           /* buffer for ablock */
654         struct xfs_alloc_arg    args;           /* allocation arguments */
655         struct xfs_bmbt_rec     *arp;           /* child record pointer */
656         struct xfs_btree_block  *block;         /* btree root block */
657         struct xfs_btree_cur    *cur;           /* bmap btree cursor */
658         int                     error;          /* error return value */
659         struct xfs_ifork        *ifp;           /* inode fork pointer */
660         struct xfs_bmbt_key     *kp;            /* root block key pointer */
661         struct xfs_mount        *mp;            /* mount structure */
662         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
663         struct xfs_iext_cursor  icur;
664         struct xfs_bmbt_irec    rec;
665         xfs_extnum_t            cnt = 0;
666 
667         mp = ip->i_mount;
668         ASSERT(whichfork != XFS_COW_FORK);
669         ifp = XFS_IFORK_PTR(ip, whichfork);
670         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
671 
672         /*
673          * Make space in the inode incore. This needs to be undone if we fail
674          * to expand the root.
675          */
676         xfs_iroot_realloc(ip, 1, whichfork);
677         ifp->if_flags |= XFS_IFBROOT;
678 
679         /*
680          * Fill in the root.
681          */
682         block = ifp->if_broot;
683         xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
684                                  XFS_BTNUM_BMAP, 1, 1, ip->i_ino,
685                                  XFS_BTREE_LONG_PTRS);
686         /*
687          * Need a cursor.  Can't allocate until bb_level is filled in.
688          */
689         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
690         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
691         /*
692          * Convert to a btree with two levels, one record in root.
693          */
694         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
695         memset(&args, 0, sizeof(args));
696         args.tp = tp;
697         args.mp = mp;
698         xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork);
699         if (tp->t_firstblock == NULLFSBLOCK) {
700                 args.type = XFS_ALLOCTYPE_START_BNO;
701                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
702         } else if (tp->t_flags & XFS_TRANS_LOWMODE) {
703                 args.type = XFS_ALLOCTYPE_START_BNO;
704                 args.fsbno = tp->t_firstblock;
705         } else {
706                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
707                 args.fsbno = tp->t_firstblock;
708         }
709         args.minlen = args.maxlen = args.prod = 1;
710         args.wasdel = wasdel;
711         *logflagsp = 0;
712         error = xfs_alloc_vextent(&args);
713         if (error)
714                 goto out_root_realloc;
715 
716         if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
717                 error = -ENOSPC;
718                 goto out_root_realloc;
719         }
720 
721         /*
722          * Allocation can't fail, the space was reserved.
723          */
724         ASSERT(tp->t_firstblock == NULLFSBLOCK ||
725                args.agno >= XFS_FSB_TO_AGNO(mp, tp->t_firstblock));
726         tp->t_firstblock = args.fsbno;
727         cur->bc_private.b.allocated++;
728         ip->i_d.di_nblocks++;
729         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
730         abp = xfs_btree_get_bufl(mp, tp, args.fsbno);
731         if (!abp) {
732                 error = -EFSCORRUPTED;
733                 goto out_unreserve_dquot;
734         }
735 
736         /*
737          * Fill in the child block.
738          */
739         abp->b_ops = &xfs_bmbt_buf_ops;
740         ablock = XFS_BUF_TO_BLOCK(abp);
741         xfs_btree_init_block_int(mp, ablock, abp->b_bn,
742                                 XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
743                                 XFS_BTREE_LONG_PTRS);
744 
745         for_each_xfs_iext(ifp, &icur, &rec) {
746                 if (isnullstartblock(rec.br_startblock))
747                         continue;
748                 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1 + cnt);
749                 xfs_bmbt_disk_set_all(arp, &rec);
750                 cnt++;
751         }
752         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
753         xfs_btree_set_numrecs(ablock, cnt);
754 
755         /*
756          * Fill in the root key and pointer.
757          */
758         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
759         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
760         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
761         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
762                                                 be16_to_cpu(block->bb_level)));
763         *pp = cpu_to_be64(args.fsbno);
764 
765         /*
766          * Do all this logging at the end so that
767          * the root is at the right level.
768          */
769         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
770         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
771         ASSERT(*curp == NULL);
772         *curp = cur;
773         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
774         return 0;
775 
776 out_unreserve_dquot:
777         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
778 out_root_realloc:
779         xfs_iroot_realloc(ip, -1, whichfork);
780         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
781         ASSERT(ifp->if_broot == NULL);
782         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
783 
784         return error;
785 }
786 
787 /*
788  * Convert a local file to an extents file.
789  * This code is out of bounds for data forks of regular files,
790  * since the file data needs to get logged so things will stay consistent.
791  * (The bmap-level manipulations are ok, though).
792  */
793 void
794 xfs_bmap_local_to_extents_empty(
795         struct xfs_inode        *ip,
796         int                     whichfork)
797 {
798         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
799 
800         ASSERT(whichfork != XFS_COW_FORK);
801         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
802         ASSERT(ifp->if_bytes == 0);
803         ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
804 
805         xfs_bmap_forkoff_reset(ip, whichfork);
806         ifp->if_flags &= ~XFS_IFINLINE;
807         ifp->if_flags |= XFS_IFEXTENTS;
808         ifp->if_u1.if_root = NULL;
809         ifp->if_height = 0;
810         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
811 }
812 
813 
814 STATIC int                              /* error */
815 xfs_bmap_local_to_extents(
816         xfs_trans_t     *tp,            /* transaction pointer */
817         xfs_inode_t     *ip,            /* incore inode pointer */
818         xfs_extlen_t    total,          /* total blocks needed by transaction */
819         int             *logflagsp,     /* inode logging flags */
820         int             whichfork,
821         void            (*init_fn)(struct xfs_trans *tp,
822                                    struct xfs_buf *bp,
823                                    struct xfs_inode *ip,
824                                    struct xfs_ifork *ifp))
825 {
826         int             error = 0;
827         int             flags;          /* logging flags returned */
828         struct xfs_ifork *ifp;          /* inode fork pointer */
829         xfs_alloc_arg_t args;           /* allocation arguments */
830         xfs_buf_t       *bp;            /* buffer for extent block */
831         struct xfs_bmbt_irec rec;
832         struct xfs_iext_cursor icur;
833 
834         /*
835          * We don't want to deal with the case of keeping inode data inline yet.
836          * So sending the data fork of a regular inode is invalid.
837          */
838         ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
839         ifp = XFS_IFORK_PTR(ip, whichfork);
840         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
841 
842         if (!ifp->if_bytes) {
843                 xfs_bmap_local_to_extents_empty(ip, whichfork);
844                 flags = XFS_ILOG_CORE;
845                 goto done;
846         }
847 
848         flags = 0;
849         error = 0;
850         ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS)) == XFS_IFINLINE);
851         memset(&args, 0, sizeof(args));
852         args.tp = tp;
853         args.mp = ip->i_mount;
854         xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0);
855         /*
856          * Allocate a block.  We know we need only one, since the
857          * file currently fits in an inode.
858          */
859         if (tp->t_firstblock == NULLFSBLOCK) {
860                 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
861                 args.type = XFS_ALLOCTYPE_START_BNO;
862         } else {
863                 args.fsbno = tp->t_firstblock;
864                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
865         }
866         args.total = total;
867         args.minlen = args.maxlen = args.prod = 1;
868         error = xfs_alloc_vextent(&args);
869         if (error)
870                 goto done;
871 
872         /* Can't fail, the space was reserved. */
873         ASSERT(args.fsbno != NULLFSBLOCK);
874         ASSERT(args.len == 1);
875         tp->t_firstblock = args.fsbno;
876         bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno);
877 
878         /*
879          * Initialize the block, copy the data and log the remote buffer.
880          *
881          * The callout is responsible for logging because the remote format
882          * might differ from the local format and thus we don't know how much to
883          * log here. Note that init_fn must also set the buffer log item type
884          * correctly.
885          */
886         init_fn(tp, bp, ip, ifp);
887 
888         /* account for the change in fork size */
889         xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
890         xfs_bmap_local_to_extents_empty(ip, whichfork);
891         flags |= XFS_ILOG_CORE;
892 
893         ifp->if_u1.if_root = NULL;
894         ifp->if_height = 0;
895 
896         rec.br_startoff = 0;
897         rec.br_startblock = args.fsbno;
898         rec.br_blockcount = 1;
899         rec.br_state = XFS_EXT_NORM;
900         xfs_iext_first(ifp, &icur);
901         xfs_iext_insert(ip, &icur, &rec, 0);
902 
903         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
904         ip->i_d.di_nblocks = 1;
905         xfs_trans_mod_dquot_byino(tp, ip,
906                 XFS_TRANS_DQ_BCOUNT, 1L);
907         flags |= xfs_ilog_fext(whichfork);
908 
909 done:
910         *logflagsp = flags;
911         return error;
912 }
913 
914 /*
915  * Called from xfs_bmap_add_attrfork to handle btree format files.
916  */
917 STATIC int                                      /* error */
918 xfs_bmap_add_attrfork_btree(
919         xfs_trans_t             *tp,            /* transaction pointer */
920         xfs_inode_t             *ip,            /* incore inode pointer */
921         int                     *flags)         /* inode logging flags */
922 {
923         xfs_btree_cur_t         *cur;           /* btree cursor */
924         int                     error;          /* error return value */
925         xfs_mount_t             *mp;            /* file system mount struct */
926         int                     stat;           /* newroot status */
927 
928         mp = ip->i_mount;
929         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
930                 *flags |= XFS_ILOG_DBROOT;
931         else {
932                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
933                 error = xfs_bmbt_lookup_first(cur, &stat);
934                 if (error)
935                         goto error0;
936                 /* must be at least one entry */
937                 XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0);
938                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
939                         goto error0;
940                 if (stat == 0) {
941                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
942                         return -ENOSPC;
943                 }
944                 cur->bc_private.b.allocated = 0;
945                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
946         }
947         return 0;
948 error0:
949         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
950         return error;
951 }
952 
953 /*
954  * Called from xfs_bmap_add_attrfork to handle extents format files.
955  */
956 STATIC int                                      /* error */
957 xfs_bmap_add_attrfork_extents(
958         struct xfs_trans        *tp,            /* transaction pointer */
959         struct xfs_inode        *ip,            /* incore inode pointer */
960         int                     *flags)         /* inode logging flags */
961 {
962         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
963         int                     error;          /* error return value */
964 
965         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
966                 return 0;
967         cur = NULL;
968         error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, flags,
969                                           XFS_DATA_FORK);
970         if (cur) {
971                 cur->bc_private.b.allocated = 0;
972                 xfs_btree_del_cursor(cur, error);
973         }
974         return error;
975 }
976 
977 /*
978  * Called from xfs_bmap_add_attrfork to handle local format files. Each
979  * different data fork content type needs a different callout to do the
980  * conversion. Some are basic and only require special block initialisation
981  * callouts for the data formating, others (directories) are so specialised they
982  * handle everything themselves.
983  *
984  * XXX (dgc): investigate whether directory conversion can use the generic
985  * formatting callout. It should be possible - it's just a very complex
986  * formatter.
987  */
988 STATIC int                                      /* error */
989 xfs_bmap_add_attrfork_local(
990         struct xfs_trans        *tp,            /* transaction pointer */
991         struct xfs_inode        *ip,            /* incore inode pointer */
992         int                     *flags)         /* inode logging flags */
993 {
994         struct xfs_da_args      dargs;          /* args for dir/attr code */
995 
996         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
997                 return 0;
998 
999         if (S_ISDIR(VFS_I(ip)->i_mode)) {
1000                 memset(&dargs, 0, sizeof(dargs));
1001                 dargs.geo = ip->i_mount->m_dir_geo;
1002                 dargs.dp = ip;
1003                 dargs.total = dargs.geo->fsbcount;
1004                 dargs.whichfork = XFS_DATA_FORK;
1005                 dargs.trans = tp;
1006                 return xfs_dir2_sf_to_block(&dargs);
1007         }
1008 
1009         if (S_ISLNK(VFS_I(ip)->i_mode))
1010                 return xfs_bmap_local_to_extents(tp, ip, 1, flags,
1011                                                  XFS_DATA_FORK,
1012                                                  xfs_symlink_local_to_remote);
1013 
1014         /* should only be called for types that support local format data */
1015         ASSERT(0);
1016         return -EFSCORRUPTED;
1017 }
1018 
1019 /* Set an inode attr fork off based on the format */
1020 int
1021 xfs_bmap_set_attrforkoff(
1022         struct xfs_inode        *ip,
1023         int                     size,
1024         int                     *version)
1025 {
1026         switch (ip->i_d.di_format) {
1027         case XFS_DINODE_FMT_DEV:
1028                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1029                 break;
1030         case XFS_DINODE_FMT_LOCAL:
1031         case XFS_DINODE_FMT_EXTENTS:
1032         case XFS_DINODE_FMT_BTREE:
1033                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1034                 if (!ip->i_d.di_forkoff)
1035                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1036                 else if ((ip->i_mount->m_flags & XFS_MOUNT_ATTR2) && version)
1037                         *version = 2;
1038                 break;
1039         default:
1040                 ASSERT(0);
1041                 return -EINVAL;
1042         }
1043 
1044         return 0;
1045 }
1046 
1047 /*
1048  * Convert inode from non-attributed to attributed.
1049  * Must not be in a transaction, ip must not be locked.
1050  */
1051 int                                             /* error code */
1052 xfs_bmap_add_attrfork(
1053         xfs_inode_t             *ip,            /* incore inode pointer */
1054         int                     size,           /* space new attribute needs */
1055         int                     rsvd)           /* xact may use reserved blks */
1056 {
1057         xfs_mount_t             *mp;            /* mount structure */
1058         xfs_trans_t             *tp;            /* transaction pointer */
1059         int                     blks;           /* space reservation */
1060         int                     version = 1;    /* superblock attr version */
1061         int                     logflags;       /* logging flags */
1062         int                     error;          /* error return value */
1063 
1064         ASSERT(XFS_IFORK_Q(ip) == 0);
1065 
1066         mp = ip->i_mount;
1067         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1068 
1069         blks = XFS_ADDAFORK_SPACE_RES(mp);
1070 
1071         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_addafork, blks, 0,
1072                         rsvd ? XFS_TRANS_RESERVE : 0, &tp);
1073         if (error)
1074                 return error;
1075 
1076         xfs_ilock(ip, XFS_ILOCK_EXCL);
1077         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1078                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1079                         XFS_QMOPT_RES_REGBLKS);
1080         if (error)
1081                 goto trans_cancel;
1082         if (XFS_IFORK_Q(ip))
1083                 goto trans_cancel;
1084         if (ip->i_d.di_anextents != 0) {
1085                 error = -EFSCORRUPTED;
1086                 goto trans_cancel;
1087         }
1088         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1089                 /*
1090                  * For inodes coming from pre-6.2 filesystems.
1091                  */
1092                 ASSERT(ip->i_d.di_aformat == 0);
1093                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1094         }
1095 
1096         xfs_trans_ijoin(tp, ip, 0);
1097         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1098         error = xfs_bmap_set_attrforkoff(ip, size, &version);
1099         if (error)
1100                 goto trans_cancel;
1101         ASSERT(ip->i_afp == NULL);
1102         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1103         ip->i_afp->if_flags = XFS_IFEXTENTS;
1104         logflags = 0;
1105         switch (ip->i_d.di_format) {
1106         case XFS_DINODE_FMT_LOCAL:
1107                 error = xfs_bmap_add_attrfork_local(tp, ip, &logflags);
1108                 break;
1109         case XFS_DINODE_FMT_EXTENTS:
1110                 error = xfs_bmap_add_attrfork_extents(tp, ip, &logflags);
1111                 break;
1112         case XFS_DINODE_FMT_BTREE:
1113                 error = xfs_bmap_add_attrfork_btree(tp, ip, &logflags);
1114                 break;
1115         default:
1116                 error = 0;
1117                 break;
1118         }
1119         if (logflags)
1120                 xfs_trans_log_inode(tp, ip, logflags);
1121         if (error)
1122                 goto trans_cancel;
1123         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1124            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1125                 bool log_sb = false;
1126 
1127                 spin_lock(&mp->m_sb_lock);
1128                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1129                         xfs_sb_version_addattr(&mp->m_sb);
1130                         log_sb = true;
1131                 }
1132                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1133                         xfs_sb_version_addattr2(&mp->m_sb);
1134                         log_sb = true;
1135                 }
1136                 spin_unlock(&mp->m_sb_lock);
1137                 if (log_sb)
1138                         xfs_log_sb(tp);
1139         }
1140 
1141         error = xfs_trans_commit(tp);
1142         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1143         return error;
1144 
1145 trans_cancel:
1146         xfs_trans_cancel(tp);
1147         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1148         return error;
1149 }
1150 
1151 /*
1152  * Internal and external extent tree search functions.
1153  */
1154 
1155 /*
1156  * Read in extents from a btree-format inode.
1157  */
1158 int
1159 xfs_iread_extents(
1160         struct xfs_trans        *tp,
1161         struct xfs_inode        *ip,
1162         int                     whichfork)
1163 {
1164         struct xfs_mount        *mp = ip->i_mount;
1165         int                     state = xfs_bmap_fork_to_state(whichfork);
1166         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1167         xfs_extnum_t            nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
1168         struct xfs_btree_block  *block = ifp->if_broot;
1169         struct xfs_iext_cursor  icur;
1170         struct xfs_bmbt_irec    new;
1171         xfs_fsblock_t           bno;
1172         struct xfs_buf          *bp;
1173         xfs_extnum_t            i, j;
1174         int                     level;
1175         __be64                  *pp;
1176         int                     error;
1177 
1178         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1179 
1180         if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1181                 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
1182                 return -EFSCORRUPTED;
1183         }
1184 
1185         /*
1186          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1187          */
1188         level = be16_to_cpu(block->bb_level);
1189         if (unlikely(level == 0)) {
1190                 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
1191                 return -EFSCORRUPTED;
1192         }
1193         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1194         bno = be64_to_cpu(*pp);
1195 
1196         /*
1197          * Go down the tree until leaf level is reached, following the first
1198          * pointer (leftmost) at each level.
1199          */
1200         while (level-- > 0) {
1201                 error = xfs_btree_read_bufl(mp, tp, bno, &bp,
1202                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1203                 if (error)
1204                         goto out;
1205                 block = XFS_BUF_TO_BLOCK(bp);
1206                 if (level == 0)
1207                         break;
1208                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1209                 bno = be64_to_cpu(*pp);
1210                 XFS_WANT_CORRUPTED_GOTO(mp,
1211                         xfs_verify_fsbno(mp, bno), out_brelse);
1212                 xfs_trans_brelse(tp, bp);
1213         }
1214 
1215         /*
1216          * Here with bp and block set to the leftmost leaf node in the tree.
1217          */
1218         i = 0;
1219         xfs_iext_first(ifp, &icur);
1220 
1221         /*
1222          * Loop over all leaf nodes.  Copy information to the extent records.
1223          */
1224         for (;;) {
1225                 xfs_bmbt_rec_t  *frp;
1226                 xfs_fsblock_t   nextbno;
1227                 xfs_extnum_t    num_recs;
1228 
1229                 num_recs = xfs_btree_get_numrecs(block);
1230                 if (unlikely(i + num_recs > nextents)) {
1231                         xfs_warn(ip->i_mount,
1232                                 "corrupt dinode %Lu, (btree extents).",
1233                                 (unsigned long long) ip->i_ino);
1234                         xfs_inode_verifier_error(ip, -EFSCORRUPTED,
1235                                         __func__, block, sizeof(*block),
1236                                         __this_address);
1237                         error = -EFSCORRUPTED;
1238                         goto out_brelse;
1239                 }
1240                 /*
1241                  * Read-ahead the next leaf block, if any.
1242                  */
1243                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1244                 if (nextbno != NULLFSBLOCK)
1245                         xfs_btree_reada_bufl(mp, nextbno, 1,
1246                                              &xfs_bmbt_buf_ops);
1247                 /*
1248                  * Copy records into the extent records.
1249                  */
1250                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1251                 for (j = 0; j < num_recs; j++, frp++, i++) {
1252                         xfs_failaddr_t  fa;
1253 
1254                         xfs_bmbt_disk_get_all(frp, &new);
1255                         fa = xfs_bmap_validate_extent(ip, whichfork, &new);
1256                         if (fa) {
1257                                 error = -EFSCORRUPTED;
1258                                 xfs_inode_verifier_error(ip, error,
1259                                                 "xfs_iread_extents(2)",
1260                                                 frp, sizeof(*frp), fa);
1261                                 goto out_brelse;
1262                         }
1263                         xfs_iext_insert(ip, &icur, &new, state);
1264                         trace_xfs_read_extent(ip, &icur, state, _THIS_IP_);
1265                         xfs_iext_next(ifp, &icur);
1266                 }
1267                 xfs_trans_brelse(tp, bp);
1268                 bno = nextbno;
1269                 /*
1270                  * If we've reached the end, stop.
1271                  */
1272                 if (bno == NULLFSBLOCK)
1273                         break;
1274                 error = xfs_btree_read_bufl(mp, tp, bno, &bp,
1275                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1276                 if (error)
1277                         goto out;
1278                 block = XFS_BUF_TO_BLOCK(bp);
1279         }
1280 
1281         if (i != XFS_IFORK_NEXTENTS(ip, whichfork)) {
1282                 error = -EFSCORRUPTED;
1283                 goto out;
1284         }
1285         ASSERT(i == xfs_iext_count(ifp));
1286 
1287         ifp->if_flags |= XFS_IFEXTENTS;
1288         return 0;
1289 
1290 out_brelse:
1291         xfs_trans_brelse(tp, bp);
1292 out:
1293         xfs_iext_destroy(ifp);
1294         return error;
1295 }
1296 
1297 /*
1298  * Returns the relative block number of the first unused block(s) in the given
1299  * fork with at least "len" logically contiguous blocks free.  This is the
1300  * lowest-address hole if the fork has holes, else the first block past the end
1301  * of fork.  Return 0 if the fork is currently local (in-inode).
1302  */
1303 int                                             /* error */
1304 xfs_bmap_first_unused(
1305         struct xfs_trans        *tp,            /* transaction pointer */
1306         struct xfs_inode        *ip,            /* incore inode */
1307         xfs_extlen_t            len,            /* size of hole to find */
1308         xfs_fileoff_t           *first_unused,  /* unused block */
1309         int                     whichfork)      /* data or attr fork */
1310 {
1311         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1312         struct xfs_bmbt_irec    got;
1313         struct xfs_iext_cursor  icur;
1314         xfs_fileoff_t           lastaddr = 0;
1315         xfs_fileoff_t           lowest, max;
1316         int                     error;
1317 
1318         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1319                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1320                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1321 
1322         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1323                 *first_unused = 0;
1324                 return 0;
1325         }
1326 
1327         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1328                 error = xfs_iread_extents(tp, ip, whichfork);
1329                 if (error)
1330                         return error;
1331         }
1332 
1333         lowest = max = *first_unused;
1334         for_each_xfs_iext(ifp, &icur, &got) {
1335                 /*
1336                  * See if the hole before this extent will work.
1337                  */
1338                 if (got.br_startoff >= lowest + len &&
1339                     got.br_startoff - max >= len)
1340                         break;
1341                 lastaddr = got.br_startoff + got.br_blockcount;
1342                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
1343         }
1344 
1345         *first_unused = max;
1346         return 0;
1347 }
1348 
1349 /*
1350  * Returns the file-relative block number of the last block - 1 before
1351  * last_block (input value) in the file.
1352  * This is not based on i_size, it is based on the extent records.
1353  * Returns 0 for local files, as they do not have extent records.
1354  */
1355 int                                             /* error */
1356 xfs_bmap_last_before(
1357         struct xfs_trans        *tp,            /* transaction pointer */
1358         struct xfs_inode        *ip,            /* incore inode */
1359         xfs_fileoff_t           *last_block,    /* last block */
1360         int                     whichfork)      /* data or attr fork */
1361 {
1362         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1363         struct xfs_bmbt_irec    got;
1364         struct xfs_iext_cursor  icur;
1365         int                     error;
1366 
1367         switch (XFS_IFORK_FORMAT(ip, whichfork)) {
1368         case XFS_DINODE_FMT_LOCAL:
1369                 *last_block = 0;
1370                 return 0;
1371         case XFS_DINODE_FMT_BTREE:
1372         case XFS_DINODE_FMT_EXTENTS:
1373                 break;
1374         default:
1375                 return -EIO;
1376         }
1377 
1378         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1379                 error = xfs_iread_extents(tp, ip, whichfork);
1380                 if (error)
1381                         return error;
1382         }
1383 
1384         if (!xfs_iext_lookup_extent_before(ip, ifp, last_block, &icur, &got))
1385                 *last_block = 0;
1386         return 0;
1387 }
1388 
1389 int
1390 xfs_bmap_last_extent(
1391         struct xfs_trans        *tp,
1392         struct xfs_inode        *ip,
1393         int                     whichfork,
1394         struct xfs_bmbt_irec    *rec,
1395         int                     *is_empty)
1396 {
1397         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1398         struct xfs_iext_cursor  icur;
1399         int                     error;
1400 
1401         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1402                 error = xfs_iread_extents(tp, ip, whichfork);
1403                 if (error)
1404                         return error;
1405         }
1406 
1407         xfs_iext_last(ifp, &icur);
1408         if (!xfs_iext_get_extent(ifp, &icur, rec))
1409                 *is_empty = 1;
1410         else
1411                 *is_empty = 0;
1412         return 0;
1413 }
1414 
1415 /*
1416  * Check the last inode extent to determine whether this allocation will result
1417  * in blocks being allocated at the end of the file. When we allocate new data
1418  * blocks at the end of the file which do not start at the previous data block,
1419  * we will try to align the new blocks at stripe unit boundaries.
1420  *
1421  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1422  * at, or past the EOF.
1423  */
1424 STATIC int
1425 xfs_bmap_isaeof(
1426         struct xfs_bmalloca     *bma,
1427         int                     whichfork)
1428 {
1429         struct xfs_bmbt_irec    rec;
1430         int                     is_empty;
1431         int                     error;
1432 
1433         bma->aeof = false;
1434         error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1435                                      &is_empty);
1436         if (error)
1437                 return error;
1438 
1439         if (is_empty) {
1440                 bma->aeof = true;
1441                 return 0;
1442         }
1443 
1444         /*
1445          * Check if we are allocation or past the last extent, or at least into
1446          * the last delayed allocated extent.
1447          */
1448         bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1449                 (bma->offset >= rec.br_startoff &&
1450                  isnullstartblock(rec.br_startblock));
1451         return 0;
1452 }
1453 
1454 /*
1455  * Returns the file-relative block number of the first block past eof in
1456  * the file.  This is not based on i_size, it is based on the extent records.
1457  * Returns 0 for local files, as they do not have extent records.
1458  */
1459 int
1460 xfs_bmap_last_offset(
1461         struct xfs_inode        *ip,
1462         xfs_fileoff_t           *last_block,
1463         int                     whichfork)
1464 {
1465         struct xfs_bmbt_irec    rec;
1466         int                     is_empty;
1467         int                     error;
1468 
1469         *last_block = 0;
1470 
1471         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1472                 return 0;
1473 
1474         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1475             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1476                return -EIO;
1477 
1478         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1479         if (error || is_empty)
1480                 return error;
1481 
1482         *last_block = rec.br_startoff + rec.br_blockcount;
1483         return 0;
1484 }
1485 
1486 /*
1487  * Returns whether the selected fork of the inode has exactly one
1488  * block or not.  For the data fork we check this matches di_size,
1489  * implying the file's range is 0..bsize-1.
1490  */
1491 int                                     /* 1=>1 block, 0=>otherwise */
1492 xfs_bmap_one_block(
1493         xfs_inode_t     *ip,            /* incore inode */
1494         int             whichfork)      /* data or attr fork */
1495 {
1496         struct xfs_ifork *ifp;          /* inode fork pointer */
1497         int             rval;           /* return value */
1498         xfs_bmbt_irec_t s;              /* internal version of extent */
1499         struct xfs_iext_cursor icur;
1500 
1501 #ifndef DEBUG
1502         if (whichfork == XFS_DATA_FORK)
1503                 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1504 #endif  /* !DEBUG */
1505         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1506                 return 0;
1507         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1508                 return 0;
1509         ifp = XFS_IFORK_PTR(ip, whichfork);
1510         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1511         xfs_iext_first(ifp, &icur);
1512         xfs_iext_get_extent(ifp, &icur, &s);
1513         rval = s.br_startoff == 0 && s.br_blockcount == 1;
1514         if (rval && whichfork == XFS_DATA_FORK)
1515                 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1516         return rval;
1517 }
1518 
1519 /*
1520  * Extent tree manipulation functions used during allocation.
1521  */
1522 
1523 /*
1524  * Convert a delayed allocation to a real allocation.
1525  */
1526 STATIC int                              /* error */
1527 xfs_bmap_add_extent_delay_real(
1528         struct xfs_bmalloca     *bma,
1529         int                     whichfork)
1530 {
1531         struct xfs_bmbt_irec    *new = &bma->got;
1532         int                     error;  /* error return value */
1533         int                     i;      /* temp state */
1534         struct xfs_ifork        *ifp;   /* inode fork pointer */
1535         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1536         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1537                                         /* left is 0, right is 1, prev is 2 */
1538         int                     rval=0; /* return value (logging flags) */
1539         int                     state = xfs_bmap_fork_to_state(whichfork);
1540         xfs_filblks_t           da_new; /* new count del alloc blocks used */
1541         xfs_filblks_t           da_old; /* old count del alloc blocks used */
1542         xfs_filblks_t           temp=0; /* value for da_new calculations */
1543         int                     tmp_rval;       /* partial logging flags */
1544         struct xfs_mount        *mp;
1545         xfs_extnum_t            *nextents;
1546         struct xfs_bmbt_irec    old;
1547 
1548         mp = bma->ip->i_mount;
1549         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1550         ASSERT(whichfork != XFS_ATTR_FORK);
1551         nextents = (whichfork == XFS_COW_FORK ? &bma->ip->i_cnextents :
1552                                                 &bma->ip->i_d.di_nextents);
1553 
1554         ASSERT(!isnullstartblock(new->br_startblock));
1555         ASSERT(!bma->cur ||
1556                (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1557 
1558         XFS_STATS_INC(mp, xs_add_exlist);
1559 
1560 #define LEFT            r[0]
1561 #define RIGHT           r[1]
1562 #define PREV            r[2]
1563 
1564         /*
1565          * Set up a bunch of variables to make the tests simpler.
1566          */
1567         xfs_iext_get_extent(ifp, &bma->icur, &PREV);
1568         new_endoff = new->br_startoff + new->br_blockcount;
1569         ASSERT(isnullstartblock(PREV.br_startblock));
1570         ASSERT(PREV.br_startoff <= new->br_startoff);
1571         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1572 
1573         da_old = startblockval(PREV.br_startblock);
1574         da_new = 0;
1575 
1576         /*
1577          * Set flags determining what part of the previous delayed allocation
1578          * extent is being replaced by a real allocation.
1579          */
1580         if (PREV.br_startoff == new->br_startoff)
1581                 state |= BMAP_LEFT_FILLING;
1582         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1583                 state |= BMAP_RIGHT_FILLING;
1584 
1585         /*
1586          * Check and set flags if this segment has a left neighbor.
1587          * Don't set contiguous if the combined extent would be too large.
1588          */
1589         if (xfs_iext_peek_prev_extent(ifp, &bma->icur, &LEFT)) {
1590                 state |= BMAP_LEFT_VALID;
1591                 if (isnullstartblock(LEFT.br_startblock))
1592                         state |= BMAP_LEFT_DELAY;
1593         }
1594 
1595         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1596             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1597             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1598             LEFT.br_state == new->br_state &&
1599             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1600                 state |= BMAP_LEFT_CONTIG;
1601 
1602         /*
1603          * Check and set flags if this segment has a right neighbor.
1604          * Don't set contiguous if the combined extent would be too large.
1605          * Also check for all-three-contiguous being too large.
1606          */
1607         if (xfs_iext_peek_next_extent(ifp, &bma->icur, &RIGHT)) {
1608                 state |= BMAP_RIGHT_VALID;
1609                 if (isnullstartblock(RIGHT.br_startblock))
1610                         state |= BMAP_RIGHT_DELAY;
1611         }
1612 
1613         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1614             new_endoff == RIGHT.br_startoff &&
1615             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1616             new->br_state == RIGHT.br_state &&
1617             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1618             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1619                        BMAP_RIGHT_FILLING)) !=
1620                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1621                        BMAP_RIGHT_FILLING) ||
1622              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1623                         <= MAXEXTLEN))
1624                 state |= BMAP_RIGHT_CONTIG;
1625 
1626         error = 0;
1627         /*
1628          * Switch out based on the FILLING and CONTIG state bits.
1629          */
1630         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1631                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1632         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1633              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1634                 /*
1635                  * Filling in all of a previously delayed allocation extent.
1636                  * The left and right neighbors are both contiguous with new.
1637                  */
1638                 LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
1639 
1640                 xfs_iext_remove(bma->ip, &bma->icur, state);
1641                 xfs_iext_remove(bma->ip, &bma->icur, state);
1642                 xfs_iext_prev(ifp, &bma->icur);
1643                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1644                 (*nextents)--;
1645 
1646                 if (bma->cur == NULL)
1647                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1648                 else {
1649                         rval = XFS_ILOG_CORE;
1650                         error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1651                         if (error)
1652                                 goto done;
1653                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1654                         error = xfs_btree_delete(bma->cur, &i);
1655                         if (error)
1656                                 goto done;
1657                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1658                         error = xfs_btree_decrement(bma->cur, 0, &i);
1659                         if (error)
1660                                 goto done;
1661                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1662                         error = xfs_bmbt_update(bma->cur, &LEFT);
1663                         if (error)
1664                                 goto done;
1665                 }
1666                 break;
1667 
1668         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1669                 /*
1670                  * Filling in all of a previously delayed allocation extent.
1671                  * The left neighbor is contiguous, the right is not.
1672                  */
1673                 old = LEFT;
1674                 LEFT.br_blockcount += PREV.br_blockcount;
1675 
1676                 xfs_iext_remove(bma->ip, &bma->icur, state);
1677                 xfs_iext_prev(ifp, &bma->icur);
1678                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1679 
1680                 if (bma->cur == NULL)
1681                         rval = XFS_ILOG_DEXT;
1682                 else {
1683                         rval = 0;
1684                         error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1685                         if (error)
1686                                 goto done;
1687                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1688                         error = xfs_bmbt_update(bma->cur, &LEFT);
1689                         if (error)
1690                                 goto done;
1691                 }
1692                 break;
1693 
1694         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1695                 /*
1696                  * Filling in all of a previously delayed allocation extent.
1697                  * The right neighbor is contiguous, the left is not. Take care
1698                  * with delay -> unwritten extent allocation here because the
1699                  * delalloc record we are overwriting is always written.
1700                  */
1701                 PREV.br_startblock = new->br_startblock;
1702                 PREV.br_blockcount += RIGHT.br_blockcount;
1703                 PREV.br_state = new->br_state;
1704 
1705                 xfs_iext_next(ifp, &bma->icur);
1706                 xfs_iext_remove(bma->ip, &bma->icur, state);
1707                 xfs_iext_prev(ifp, &bma->icur);
1708                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1709 
1710                 if (bma->cur == NULL)
1711                         rval = XFS_ILOG_DEXT;
1712                 else {
1713                         rval = 0;
1714                         error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1715                         if (error)
1716                                 goto done;
1717                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1718                         error = xfs_bmbt_update(bma->cur, &PREV);
1719                         if (error)
1720                                 goto done;
1721                 }
1722                 break;
1723 
1724         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1725                 /*
1726                  * Filling in all of a previously delayed allocation extent.
1727                  * Neither the left nor right neighbors are contiguous with
1728                  * the new one.
1729                  */
1730                 PREV.br_startblock = new->br_startblock;
1731                 PREV.br_state = new->br_state;
1732                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1733 
1734                 (*nextents)++;
1735                 if (bma->cur == NULL)
1736                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1737                 else {
1738                         rval = XFS_ILOG_CORE;
1739                         error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1740                         if (error)
1741                                 goto done;
1742                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1743                         error = xfs_btree_insert(bma->cur, &i);
1744                         if (error)
1745                                 goto done;
1746                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1747                 }
1748                 break;
1749 
1750         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1751                 /*
1752                  * Filling in the first part of a previous delayed allocation.
1753                  * The left neighbor is contiguous.
1754                  */
1755                 old = LEFT;
1756                 temp = PREV.br_blockcount - new->br_blockcount;
1757                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1758                                 startblockval(PREV.br_startblock));
1759 
1760                 LEFT.br_blockcount += new->br_blockcount;
1761 
1762                 PREV.br_blockcount = temp;
1763                 PREV.br_startoff += new->br_blockcount;
1764                 PREV.br_startblock = nullstartblock(da_new);
1765 
1766                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1767                 xfs_iext_prev(ifp, &bma->icur);
1768                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1769 
1770                 if (bma->cur == NULL)
1771                         rval = XFS_ILOG_DEXT;
1772                 else {
1773                         rval = 0;
1774                         error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1775                         if (error)
1776                                 goto done;
1777                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1778                         error = xfs_bmbt_update(bma->cur, &LEFT);
1779                         if (error)
1780                                 goto done;
1781                 }
1782                 break;
1783 
1784         case BMAP_LEFT_FILLING:
1785                 /*
1786                  * Filling in the first part of a previous delayed allocation.
1787                  * The left neighbor is not contiguous.
1788                  */
1789                 xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1790                 (*nextents)++;
1791                 if (bma->cur == NULL)
1792                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1793                 else {
1794                         rval = XFS_ILOG_CORE;
1795                         error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1796                         if (error)
1797                                 goto done;
1798                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1799                         error = xfs_btree_insert(bma->cur, &i);
1800                         if (error)
1801                                 goto done;
1802                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1803                 }
1804 
1805                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1806                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1807                                         &bma->cur, 1, &tmp_rval, whichfork);
1808                         rval |= tmp_rval;
1809                         if (error)
1810                                 goto done;
1811                 }
1812 
1813                 temp = PREV.br_blockcount - new->br_blockcount;
1814                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1815                         startblockval(PREV.br_startblock) -
1816                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
1817 
1818                 PREV.br_startoff = new_endoff;
1819                 PREV.br_blockcount = temp;
1820                 PREV.br_startblock = nullstartblock(da_new);
1821                 xfs_iext_next(ifp, &bma->icur);
1822                 xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1823                 xfs_iext_prev(ifp, &bma->icur);
1824                 break;
1825 
1826         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1827                 /*
1828                  * Filling in the last part of a previous delayed allocation.
1829                  * The right neighbor is contiguous with the new allocation.
1830                  */
1831                 old = RIGHT;
1832                 RIGHT.br_startoff = new->br_startoff;
1833                 RIGHT.br_startblock = new->br_startblock;
1834                 RIGHT.br_blockcount += new->br_blockcount;
1835 
1836                 if (bma->cur == NULL)
1837                         rval = XFS_ILOG_DEXT;
1838                 else {
1839                         rval = 0;
1840                         error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1841                         if (error)
1842                                 goto done;
1843                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1844                         error = xfs_bmbt_update(bma->cur, &RIGHT);
1845                         if (error)
1846                                 goto done;
1847                 }
1848 
1849                 temp = PREV.br_blockcount - new->br_blockcount;
1850                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1851                         startblockval(PREV.br_startblock));
1852 
1853                 PREV.br_blockcount = temp;
1854                 PREV.br_startblock = nullstartblock(da_new);
1855 
1856                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1857                 xfs_iext_next(ifp, &bma->icur);
1858                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &RIGHT);
1859                 break;
1860 
1861         case BMAP_RIGHT_FILLING:
1862                 /*
1863                  * Filling in the last part of a previous delayed allocation.
1864                  * The right neighbor is not contiguous.
1865                  */
1866                 xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1867                 (*nextents)++;
1868                 if (bma->cur == NULL)
1869                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1870                 else {
1871                         rval = XFS_ILOG_CORE;
1872                         error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1873                         if (error)
1874                                 goto done;
1875                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1876                         error = xfs_btree_insert(bma->cur, &i);
1877                         if (error)
1878                                 goto done;
1879                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1880                 }
1881 
1882                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1883                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1884                                 &bma->cur, 1, &tmp_rval, whichfork);
1885                         rval |= tmp_rval;
1886                         if (error)
1887                                 goto done;
1888                 }
1889 
1890                 temp = PREV.br_blockcount - new->br_blockcount;
1891                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1892                         startblockval(PREV.br_startblock) -
1893                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
1894 
1895                 PREV.br_startblock = nullstartblock(da_new);
1896                 PREV.br_blockcount = temp;
1897                 xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1898                 xfs_iext_next(ifp, &bma->icur);
1899                 break;
1900 
1901         case 0:
1902                 /*
1903                  * Filling in the middle part of a previous delayed allocation.
1904                  * Contiguity is impossible here.
1905                  * This case is avoided almost all the time.
1906                  *
1907                  * We start with a delayed allocation:
1908                  *
1909                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1910                  *  PREV @ idx
1911                  *
1912                  * and we are allocating:
1913                  *                     +rrrrrrrrrrrrrrrrr+
1914                  *                            new
1915                  *
1916                  * and we set it up for insertion as:
1917                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1918                  *                            new
1919                  *  PREV @ idx          LEFT              RIGHT
1920                  *                      inserted at idx + 1
1921                  */
1922                 old = PREV;
1923 
1924                 /* LEFT is the new middle */
1925                 LEFT = *new;
1926 
1927                 /* RIGHT is the new right */
1928                 RIGHT.br_state = PREV.br_state;
1929                 RIGHT.br_startoff = new_endoff;
1930                 RIGHT.br_blockcount =
1931                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1932                 RIGHT.br_startblock =
1933                         nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1934                                         RIGHT.br_blockcount));
1935 
1936                 /* truncate PREV */
1937                 PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
1938                 PREV.br_startblock =
1939                         nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1940                                         PREV.br_blockcount));
1941                 xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1942 
1943                 xfs_iext_next(ifp, &bma->icur);
1944                 xfs_iext_insert(bma->ip, &bma->icur, &RIGHT, state);
1945                 xfs_iext_insert(bma->ip, &bma->icur, &LEFT, state);
1946                 (*nextents)++;
1947 
1948                 if (bma->cur == NULL)
1949                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1950                 else {
1951                         rval = XFS_ILOG_CORE;
1952                         error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1953                         if (error)
1954                                 goto done;
1955                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1956                         error = xfs_btree_insert(bma->cur, &i);
1957                         if (error)
1958                                 goto done;
1959                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1960                 }
1961 
1962                 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1963                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1964                                         &bma->cur, 1, &tmp_rval, whichfork);
1965                         rval |= tmp_rval;
1966                         if (error)
1967                                 goto done;
1968                 }
1969 
1970                 da_new = startblockval(PREV.br_startblock) +
1971                          startblockval(RIGHT.br_startblock);
1972                 break;
1973 
1974         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1975         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1976         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1977         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1978         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1979         case BMAP_LEFT_CONTIG:
1980         case BMAP_RIGHT_CONTIG:
1981                 /*
1982                  * These cases are all impossible.
1983                  */
1984                 ASSERT(0);
1985         }
1986 
1987         /* add reverse mapping unless caller opted out */
1988         if (!(bma->flags & XFS_BMAPI_NORMAP)) {
1989                 error = xfs_rmap_map_extent(bma->tp, bma->ip, whichfork, new);
1990                 if (error)
1991                         goto done;
1992         }
1993 
1994         /* convert to a btree if necessary */
1995         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1996                 int     tmp_logflags;   /* partial log flag return val */
1997 
1998                 ASSERT(bma->cur == NULL);
1999                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2000                                 &bma->cur, da_old > 0, &tmp_logflags,
2001                                 whichfork);
2002                 bma->logflags |= tmp_logflags;
2003                 if (error)
2004                         goto done;
2005         }
2006 
2007         if (da_new != da_old)
2008                 xfs_mod_delalloc(mp, (int64_t)da_new - da_old);
2009 
2010         if (bma->cur) {
2011                 da_new += bma->cur->bc_private.b.allocated;
2012                 bma->cur->bc_private.b.allocated = 0;
2013         }
2014 
2015         /* adjust for changes in reserved delayed indirect blocks */
2016         if (da_new != da_old) {
2017                 ASSERT(state == 0 || da_new < da_old);
2018                 error = xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new),
2019                                 false);
2020         }
2021 
2022         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
2023 done:
2024         if (whichfork != XFS_COW_FORK)
2025                 bma->logflags |= rval;
2026         return error;
2027 #undef  LEFT
2028 #undef  RIGHT
2029 #undef  PREV
2030 }
2031 
2032 /*
2033  * Convert an unwritten allocation to a real allocation or vice versa.
2034  */
2035 int                                     /* error */
2036 xfs_bmap_add_extent_unwritten_real(
2037         struct xfs_trans        *tp,
2038         xfs_inode_t             *ip,    /* incore inode pointer */
2039         int                     whichfork,
2040         struct xfs_iext_cursor  *icur,
2041         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
2042         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
2043         int                     *logflagsp) /* inode logging flags */
2044 {
2045         xfs_btree_cur_t         *cur;   /* btree cursor */
2046         int                     error;  /* error return value */
2047         int                     i;      /* temp state */
2048         struct xfs_ifork        *ifp;   /* inode fork pointer */
2049         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
2050         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
2051                                         /* left is 0, right is 1, prev is 2 */
2052         int                     rval=0; /* return value (logging flags) */
2053         int                     state = xfs_bmap_fork_to_state(whichfork);
2054         struct xfs_mount        *mp = ip->i_mount;
2055         struct xfs_bmbt_irec    old;
2056 
2057         *logflagsp = 0;
2058 
2059         cur = *curp;
2060         ifp = XFS_IFORK_PTR(ip, whichfork);
2061 
2062         ASSERT(!isnullstartblock(new->br_startblock));
2063 
2064         XFS_STATS_INC(mp, xs_add_exlist);
2065 
2066 #define LEFT            r[0]
2067 #define RIGHT           r[1]
2068 #define PREV            r[2]
2069 
2070         /*
2071          * Set up a bunch of variables to make the tests simpler.
2072          */
2073         error = 0;
2074         xfs_iext_get_extent(ifp, icur, &PREV);
2075         ASSERT(new->br_state != PREV.br_state);
2076         new_endoff = new->br_startoff + new->br_blockcount;
2077         ASSERT(PREV.br_startoff <= new->br_startoff);
2078         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2079 
2080         /*
2081          * Set flags determining what part of the previous oldext allocation
2082          * extent is being replaced by a newext allocation.
2083          */
2084         if (PREV.br_startoff == new->br_startoff)
2085                 state |= BMAP_LEFT_FILLING;
2086         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2087                 state |= BMAP_RIGHT_FILLING;
2088 
2089         /*
2090          * Check and set flags if this segment has a left neighbor.
2091          * Don't set contiguous if the combined extent would be too large.
2092          */
2093         if (xfs_iext_peek_prev_extent(ifp, icur, &LEFT)) {
2094                 state |= BMAP_LEFT_VALID;
2095                 if (isnullstartblock(LEFT.br_startblock))
2096                         state |= BMAP_LEFT_DELAY;
2097         }
2098 
2099         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2100             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2101             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2102             LEFT.br_state == new->br_state &&
2103             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2104                 state |= BMAP_LEFT_CONTIG;
2105 
2106         /*
2107          * Check and set flags if this segment has a right neighbor.
2108          * Don't set contiguous if the combined extent would be too large.
2109          * Also check for all-three-contiguous being too large.
2110          */
2111         if (xfs_iext_peek_next_extent(ifp, icur, &RIGHT)) {
2112                 state |= BMAP_RIGHT_VALID;
2113                 if (isnullstartblock(RIGHT.br_startblock))
2114                         state |= BMAP_RIGHT_DELAY;
2115         }
2116 
2117         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2118             new_endoff == RIGHT.br_startoff &&
2119             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2120             new->br_state == RIGHT.br_state &&
2121             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2122             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2123                        BMAP_RIGHT_FILLING)) !=
2124                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2125                        BMAP_RIGHT_FILLING) ||
2126              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2127                         <= MAXEXTLEN))
2128                 state |= BMAP_RIGHT_CONTIG;
2129 
2130         /*
2131          * Switch out based on the FILLING and CONTIG state bits.
2132          */
2133         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2134                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2135         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2136              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2137                 /*
2138                  * Setting all of a previous oldext extent to newext.
2139                  * The left and right neighbors are both contiguous with new.
2140                  */
2141                 LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
2142 
2143                 xfs_iext_remove(ip, icur, state);
2144                 xfs_iext_remove(ip, icur, state);
2145                 xfs_iext_prev(ifp, icur);
2146                 xfs_iext_update_extent(ip, state, icur, &LEFT);
2147                 XFS_IFORK_NEXT_SET(ip, whichfork,
2148                                 XFS_IFORK_NEXTENTS(ip, whichfork) - 2);
2149                 if (cur == NULL)
2150                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2151                 else {
2152                         rval = XFS_ILOG_CORE;
2153                         error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2154                         if (error)
2155                                 goto done;
2156                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2157                         if ((error = xfs_btree_delete(cur, &i)))
2158                                 goto done;
2159                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2160                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2161                                 goto done;
2162                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2163                         if ((error = xfs_btree_delete(cur, &i)))
2164                                 goto done;
2165                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2166                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2167                                 goto done;
2168                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2169                         error = xfs_bmbt_update(cur, &LEFT);
2170                         if (error)
2171                                 goto done;
2172                 }
2173                 break;
2174 
2175         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2176                 /*
2177                  * Setting all of a previous oldext extent to newext.
2178                  * The left neighbor is contiguous, the right is not.
2179                  */
2180                 LEFT.br_blockcount += PREV.br_blockcount;
2181 
2182                 xfs_iext_remove(ip, icur, state);
2183                 xfs_iext_prev(ifp, icur);
2184                 xfs_iext_update_extent(ip, state, icur, &LEFT);
2185                 XFS_IFORK_NEXT_SET(ip, whichfork,
2186                                 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2187                 if (cur == NULL)
2188                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2189                 else {
2190                         rval = XFS_ILOG_CORE;
2191                         error = xfs_bmbt_lookup_eq(cur, &PREV, &i);
2192                         if (error)
2193                                 goto done;
2194                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2195                         if ((error = xfs_btree_delete(cur, &i)))
2196                                 goto done;
2197                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2198                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2199                                 goto done;
2200                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2201                         error = xfs_bmbt_update(cur, &LEFT);
2202                         if (error)
2203                                 goto done;
2204                 }
2205                 break;
2206 
2207         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2208                 /*
2209                  * Setting all of a previous oldext extent to newext.
2210                  * The right neighbor is contiguous, the left is not.
2211                  */
2212                 PREV.br_blockcount += RIGHT.br_blockcount;
2213                 PREV.br_state = new->br_state;
2214 
2215                 xfs_iext_next(ifp, icur);
2216                 xfs_iext_remove(ip, icur, state);
2217                 xfs_iext_prev(ifp, icur);
2218                 xfs_iext_update_extent(ip, state, icur, &PREV);
2219 
2220                 XFS_IFORK_NEXT_SET(ip, whichfork,
2221                                 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2222                 if (cur == NULL)
2223                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2224                 else {
2225                         rval = XFS_ILOG_CORE;
2226                         error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2227                         if (error)
2228                                 goto done;
2229                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2230                         if ((error = xfs_btree_delete(cur, &i)))
2231                                 goto done;
2232                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2233                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2234                                 goto done;
2235                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2236                         error = xfs_bmbt_update(cur, &PREV);
2237                         if (error)
2238                                 goto done;
2239                 }
2240                 break;
2241 
2242         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2243                 /*
2244                  * Setting all of a previous oldext extent to newext.
2245                  * Neither the left nor right neighbors are contiguous with
2246                  * the new one.
2247                  */
2248                 PREV.br_state = new->br_state;
2249                 xfs_iext_update_extent(ip, state, icur, &PREV);
2250 
2251                 if (cur == NULL)
2252                         rval = XFS_ILOG_DEXT;
2253                 else {
2254                         rval = 0;
2255                         error = xfs_bmbt_lookup_eq(cur, new, &i);
2256                         if (error)
2257                                 goto done;
2258                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2259                         error = xfs_bmbt_update(cur, &PREV);
2260                         if (error)
2261                                 goto done;
2262                 }
2263                 break;
2264 
2265         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2266                 /*
2267                  * Setting the first part of a previous oldext extent to newext.
2268                  * The left neighbor is contiguous.
2269                  */
2270                 LEFT.br_blockcount += new->br_blockcount;
2271 
2272                 old = PREV;
2273                 PREV.br_startoff += new->br_blockcount;
2274                 PREV.br_startblock += new->br_blockcount;
2275                 PREV.br_blockcount -= new->br_blockcount;
2276 
2277                 xfs_iext_update_extent(ip, state, icur, &PREV);
2278                 xfs_iext_prev(ifp, icur);
2279                 xfs_iext_update_extent(ip, state, icur, &LEFT);
2280 
2281                 if (cur == NULL)
2282                         rval = XFS_ILOG_DEXT;
2283                 else {
2284                         rval = 0;
2285                         error = xfs_bmbt_lookup_eq(cur, &old, &i);
2286                         if (error)
2287                                 goto done;
2288                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2289                         error = xfs_bmbt_update(cur, &PREV);
2290                         if (error)
2291                                 goto done;
2292                         error = xfs_btree_decrement(cur, 0, &i);
2293                         if (error)
2294                                 goto done;
2295                         error = xfs_bmbt_update(cur, &LEFT);
2296                         if (error)
2297                                 goto done;
2298                 }
2299                 break;
2300 
2301         case BMAP_LEFT_FILLING:
2302                 /*
2303                  * Setting the first part of a previous oldext extent to newext.
2304                  * The left neighbor is not contiguous.
2305                  */
2306                 old = PREV;
2307                 PREV.br_startoff += new->br_blockcount;
2308                 PREV.br_startblock += new->br_blockcount;
2309                 PREV.br_blockcount -= new->br_blockcount;
2310 
2311                 xfs_iext_update_extent(ip, state, icur, &PREV);
2312                 xfs_iext_insert(ip, icur, new, state);
2313                 XFS_IFORK_NEXT_SET(ip, whichfork,
2314                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2315                 if (cur == NULL)
2316                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2317                 else {
2318                         rval = XFS_ILOG_CORE;
2319                         error = xfs_bmbt_lookup_eq(cur, &old, &i);
2320                         if (error)
2321                                 goto done;
2322                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2323                         error = xfs_bmbt_update(cur, &PREV);
2324                         if (error)
2325                                 goto done;
2326                         cur->bc_rec.b = *new;
2327                         if ((error = xfs_btree_insert(cur, &i)))
2328                                 goto done;
2329                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2330                 }
2331                 break;
2332 
2333         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2334                 /*
2335                  * Setting the last part of a previous oldext extent to newext.
2336                  * The right neighbor is contiguous with the new allocation.
2337                  */
2338                 old = PREV;
2339                 PREV.br_blockcount -= new->br_blockcount;
2340 
2341                 RIGHT.br_startoff = new->br_startoff;
2342                 RIGHT.br_startblock = new->br_startblock;
2343                 RIGHT.br_blockcount += new->br_blockcount;
2344 
2345                 xfs_iext_update_extent(ip, state, icur, &PREV);
2346                 xfs_iext_next(ifp, icur);
2347                 xfs_iext_update_extent(ip, state, icur, &RIGHT);
2348 
2349                 if (cur == NULL)
2350                         rval = XFS_ILOG_DEXT;
2351                 else {
2352                         rval = 0;
2353                         error = xfs_bmbt_lookup_eq(cur, &old, &i);
2354                         if (error)
2355                                 goto done;
2356                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2357                         error = xfs_bmbt_update(cur, &PREV);
2358                         if (error)
2359                                 goto done;
2360                         error = xfs_btree_increment(cur, 0, &i);
2361                         if (error)
2362                                 goto done;
2363                         error = xfs_bmbt_update(cur, &RIGHT);
2364                         if (error)
2365                                 goto done;
2366                 }
2367                 break;
2368 
2369         case BMAP_RIGHT_FILLING:
2370                 /*
2371                  * Setting the last part of a previous oldext extent to newext.
2372                  * The right neighbor is not contiguous.
2373                  */
2374                 old = PREV;
2375                 PREV.br_blockcount -= new->br_blockcount;
2376 
2377                 xfs_iext_update_extent(ip, state, icur, &PREV);
2378                 xfs_iext_next(ifp, icur);
2379                 xfs_iext_insert(ip, icur, new, state);
2380 
2381                 XFS_IFORK_NEXT_SET(ip, whichfork,
2382                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2383                 if (cur == NULL)
2384                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2385                 else {
2386                         rval = XFS_ILOG_CORE;
2387                         error = xfs_bmbt_lookup_eq(cur, &old, &i);
2388                         if (error)
2389                                 goto done;
2390                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2391                         error = xfs_bmbt_update(cur, &PREV);
2392                         if (error)
2393                                 goto done;
2394                         error = xfs_bmbt_lookup_eq(cur, new, &i);
2395                         if (error)
2396                                 goto done;
2397                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2398                         if ((error = xfs_btree_insert(cur, &i)))
2399                                 goto done;
2400                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2401                 }
2402                 break;
2403 
2404         case 0:
2405                 /*
2406                  * Setting the middle part of a previous oldext extent to
2407                  * newext.  Contiguity is impossible here.
2408                  * One extent becomes three extents.
2409                  */
2410                 old = PREV;
2411                 PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
2412 
2413                 r[0] = *new;
2414                 r[1].br_startoff = new_endoff;
2415                 r[1].br_blockcount =
2416                         old.br_startoff + old.br_blockcount - new_endoff;
2417                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
2418                 r[1].br_state = PREV.br_state;
2419 
2420                 xfs_iext_update_extent(ip, state, icur, &PREV);
2421                 xfs_iext_next(ifp, icur);
2422                 xfs_iext_insert(ip, icur, &r[1], state);
2423                 xfs_iext_insert(ip, icur, &r[0], state);
2424 
2425                 XFS_IFORK_NEXT_SET(ip, whichfork,
2426                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 2);
2427                 if (cur == NULL)
2428                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2429                 else {
2430                         rval = XFS_ILOG_CORE;
2431                         error = xfs_bmbt_lookup_eq(cur, &old, &i);
2432                         if (error)
2433                                 goto done;
2434                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2435                         /* new right extent - oldext */
2436                         error = xfs_bmbt_update(cur, &r[1]);
2437                         if (error)
2438                                 goto done;
2439                         /* new left extent - oldext */
2440                         cur->bc_rec.b = PREV;
2441                         if ((error = xfs_btree_insert(cur, &i)))
2442                                 goto done;
2443                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2444                         /*
2445                          * Reset the cursor to the position of the new extent
2446                          * we are about to insert as we can't trust it after
2447                          * the previous insert.
2448                          */
2449                         error = xfs_bmbt_lookup_eq(cur, new, &i);
2450                         if (error)
2451                                 goto done;
2452                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2453                         /* new middle extent - newext */
2454                         if ((error = xfs_btree_insert(cur, &i)))
2455                                 goto done;
2456                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2457                 }
2458                 break;
2459 
2460         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2461         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2462         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2463         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2464         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2465         case BMAP_LEFT_CONTIG:
2466         case BMAP_RIGHT_CONTIG:
2467                 /*
2468                  * These cases are all impossible.
2469                  */
2470                 ASSERT(0);
2471         }
2472 
2473         /* update reverse mappings */
2474         error = xfs_rmap_convert_extent(mp, tp, ip, whichfork, new);
2475         if (error)
2476                 goto done;
2477 
2478         /* convert to a btree if necessary */
2479         if (xfs_bmap_needs_btree(ip, whichfork)) {
2480                 int     tmp_logflags;   /* partial log flag return val */
2481 
2482                 ASSERT(cur == NULL);
2483                 error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
2484                                 &tmp_logflags, whichfork);
2485                 *logflagsp |= tmp_logflags;
2486                 if (error)
2487                         goto done;
2488         }
2489 
2490         /* clear out the allocated field, done with it now in any case. */
2491         if (cur) {
2492                 cur->bc_private.b.allocated = 0;
2493                 *curp = cur;
2494         }
2495 
2496         xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
2497 done:
2498         *logflagsp |= rval;
2499         return error;
2500 #undef  LEFT
2501 #undef  RIGHT
2502 #undef  PREV
2503 }
2504 
2505 /*
2506  * Convert a hole to a delayed allocation.
2507  */
2508 STATIC void
2509 xfs_bmap_add_extent_hole_delay(
2510         xfs_inode_t             *ip,    /* incore inode pointer */
2511         int                     whichfork,
2512         struct xfs_iext_cursor  *icur,
2513         xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
2514 {
2515         struct xfs_ifork        *ifp;   /* inode fork pointer */
2516         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2517         xfs_filblks_t           newlen=0;       /* new indirect size */
2518         xfs_filblks_t           oldlen=0;       /* old indirect size */
2519         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2520         int                     state = xfs_bmap_fork_to_state(whichfork);
2521         xfs_filblks_t           temp;    /* temp for indirect calculations */
2522 
2523         ifp = XFS_IFORK_PTR(ip, whichfork);
2524         ASSERT(isnullstartblock(new->br_startblock));
2525 
2526         /*
2527          * Check and set flags if this segment has a left neighbor
2528          */
2529         if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2530                 state |= BMAP_LEFT_VALID;
2531                 if (isnullstartblock(left.br_startblock))
2532                         state |= BMAP_LEFT_DELAY;
2533         }
2534 
2535         /*
2536          * Check and set flags if the current (right) segment exists.
2537          * If it doesn't exist, we're converting the hole at end-of-file.
2538          */
2539         if (xfs_iext_get_extent(ifp, icur, &right)) {
2540                 state |= BMAP_RIGHT_VALID;
2541                 if (isnullstartblock(right.br_startblock))
2542                         state |= BMAP_RIGHT_DELAY;
2543         }
2544 
2545         /*
2546          * Set contiguity flags on the left and right neighbors.
2547          * Don't let extents get too large, even if the pieces are contiguous.
2548          */
2549         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2550             left.br_startoff + left.br_blockcount == new->br_startoff &&
2551             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2552                 state |= BMAP_LEFT_CONTIG;
2553 
2554         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2555             new->br_startoff + new->br_blockcount == right.br_startoff &&
2556             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2557             (!(state & BMAP_LEFT_CONTIG) ||
2558              (left.br_blockcount + new->br_blockcount +
2559               right.br_blockcount <= MAXEXTLEN)))
2560                 state |= BMAP_RIGHT_CONTIG;
2561 
2562         /*
2563          * Switch out based on the contiguity flags.
2564          */
2565         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2566         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2567                 /*
2568                  * New allocation is contiguous with delayed allocations
2569                  * on the left and on the right.
2570                  * Merge all three into a single extent record.
2571                  */
2572                 temp = left.br_blockcount + new->br_blockcount +
2573                         right.br_blockcount;
2574 
2575                 oldlen = startblockval(left.br_startblock) +
2576                         startblockval(new->br_startblock) +
2577                         startblockval(right.br_startblock);
2578                 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2579                                          oldlen);
2580                 left.br_startblock = nullstartblock(newlen);
2581                 left.br_blockcount = temp;
2582 
2583                 xfs_iext_remove(ip, icur, state);
2584                 xfs_iext_prev(ifp, icur);
2585                 xfs_iext_update_extent(ip, state, icur, &left);
2586                 break;
2587 
2588         case BMAP_LEFT_CONTIG:
2589                 /*
2590                  * New allocation is contiguous with a delayed allocation
2591                  * on the left.
2592                  * Merge the new allocation with the left neighbor.
2593                  */
2594                 temp = left.br_blockcount + new->br_blockcount;
2595 
2596                 oldlen = startblockval(left.br_startblock) +
2597                         startblockval(new->br_startblock);
2598                 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2599                                          oldlen);
2600                 left.br_blockcount = temp;
2601                 left.br_startblock = nullstartblock(newlen);
2602 
2603                 xfs_iext_prev(ifp, icur);
2604                 xfs_iext_update_extent(ip, state, icur, &left);
2605                 break;
2606 
2607         case BMAP_RIGHT_CONTIG:
2608                 /*
2609                  * New allocation is contiguous with a delayed allocation
2610                  * on the right.
2611                  * Merge the new allocation with the right neighbor.
2612                  */
2613                 temp = new->br_blockcount + right.br_blockcount;
2614                 oldlen = startblockval(new->br_startblock) +
2615                         startblockval(right.br_startblock);
2616                 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2617                                          oldlen);
2618                 right.br_startoff = new->br_startoff;
2619                 right.br_startblock = nullstartblock(newlen);
2620                 right.br_blockcount = temp;
2621                 xfs_iext_update_extent(ip, state, icur, &right);
2622                 break;
2623 
2624         case 0:
2625                 /*
2626                  * New allocation is not contiguous with another
2627                  * delayed allocation.
2628                  * Insert a new entry.
2629                  */
2630                 oldlen = newlen = 0;
2631                 xfs_iext_insert(ip, icur, new, state);
2632                 break;
2633         }
2634         if (oldlen != newlen) {
2635                 ASSERT(oldlen > newlen);
2636                 xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2637                                  false);
2638                 /*
2639                  * Nothing to do for disk quota accounting here.
2640                  */
2641                 xfs_mod_delalloc(ip->i_mount, (int64_t)newlen - oldlen);
2642         }
2643 }
2644 
2645 /*
2646  * Convert a hole to a real allocation.
2647  */
2648 STATIC int                              /* error */
2649 xfs_bmap_add_extent_hole_real(
2650         struct xfs_trans        *tp,
2651         struct xfs_inode        *ip,
2652         int                     whichfork,
2653         struct xfs_iext_cursor  *icur,
2654         struct xfs_btree_cur    **curp,
2655         struct xfs_bmbt_irec    *new,
2656         int                     *logflagsp,
2657         int                     flags)
2658 {
2659         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
2660         struct xfs_mount        *mp = ip->i_mount;
2661         struct xfs_btree_cur    *cur = *curp;
2662         int                     error;  /* error return value */
2663         int                     i;      /* temp state */
2664         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2665         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2666         int                     rval=0; /* return value (logging flags) */
2667         int                     state = xfs_bmap_fork_to_state(whichfork);
2668         struct xfs_bmbt_irec    old;
2669 
2670         ASSERT(!isnullstartblock(new->br_startblock));
2671         ASSERT(!cur || !(cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2672 
2673         XFS_STATS_INC(mp, xs_add_exlist);
2674 
2675         /*
2676          * Check and set flags if this segment has a left neighbor.
2677          */
2678         if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2679                 state |= BMAP_LEFT_VALID;
2680                 if (isnullstartblock(left.br_startblock))
2681                         state |= BMAP_LEFT_DELAY;
2682         }
2683 
2684         /*
2685          * Check and set flags if this segment has a current value.
2686          * Not true if we're inserting into the "hole" at eof.
2687          */
2688         if (xfs_iext_get_extent(ifp, icur, &right)) {
2689                 state |= BMAP_RIGHT_VALID;
2690                 if (isnullstartblock(right.br_startblock))
2691                         state |= BMAP_RIGHT_DELAY;
2692         }
2693 
2694         /*
2695          * We're inserting a real allocation between "left" and "right".
2696          * Set the contiguity flags.  Don't let extents get too large.
2697          */
2698         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2699             left.br_startoff + left.br_blockcount == new->br_startoff &&
2700             left.br_startblock + left.br_blockcount == new->br_startblock &&
2701             left.br_state == new->br_state &&
2702             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2703                 state |= BMAP_LEFT_CONTIG;
2704 
2705         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2706             new->br_startoff + new->br_blockcount == right.br_startoff &&
2707             new->br_startblock + new->br_blockcount == right.br_startblock &&
2708             new->br_state == right.br_state &&
2709             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2710             (!(state & BMAP_LEFT_CONTIG) ||
2711              left.br_blockcount + new->br_blockcount +
2712              right.br_blockcount <= MAXEXTLEN))
2713                 state |= BMAP_RIGHT_CONTIG;
2714 
2715         error = 0;
2716         /*
2717          * Select which case we're in here, and implement it.
2718          */
2719         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2720         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2721                 /*
2722                  * New allocation is contiguous with real allocations on the
2723                  * left and on the right.
2724                  * Merge all three into a single extent record.
2725                  */
2726                 left.br_blockcount += new->br_blockcount + right.br_blockcount;
2727 
2728                 xfs_iext_remove(ip, icur, state);
2729                 xfs_iext_prev(ifp, icur);
2730                 xfs_iext_update_extent(ip, state, icur, &left);
2731 
2732                 XFS_IFORK_NEXT_SET(ip, whichfork,
2733                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2734                 if (cur == NULL) {
2735                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2736                 } else {
2737                         rval = XFS_ILOG_CORE;
2738                         error = xfs_bmbt_lookup_eq(cur, &right, &i);
2739                         if (error)
2740                                 goto done;
2741                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2742                         error = xfs_btree_delete(cur, &i);
2743                         if (error)
2744                                 goto done;
2745                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2746                         error = xfs_btree_decrement(cur, 0, &i);
2747                         if (error)
2748                                 goto done;
2749                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2750                         error = xfs_bmbt_update(cur, &left);
2751                         if (error)
2752                                 goto done;
2753                 }
2754                 break;
2755 
2756         case BMAP_LEFT_CONTIG:
2757                 /*
2758                  * New allocation is contiguous with a real allocation
2759                  * on the left.
2760                  * Merge the new allocation with the left neighbor.
2761                  */
2762                 old = left;
2763                 left.br_blockcount += new->br_blockcount;
2764 
2765                 xfs_iext_prev(ifp, icur);
2766                 xfs_iext_update_extent(ip, state, icur, &left);
2767 
2768                 if (cur == NULL) {
2769                         rval = xfs_ilog_fext(whichfork);
2770                 } else {
2771                         rval = 0;
2772                         error = xfs_bmbt_lookup_eq(cur, &old, &i);
2773                         if (error)
2774                                 goto done;
2775                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2776                         error = xfs_bmbt_update(cur, &left);
2777                         if (error)
2778                                 goto done;
2779                 }
2780                 break;
2781 
2782         case BMAP_RIGHT_CONTIG:
2783                 /*
2784                  * New allocation is contiguous with a real allocation
2785                  * on the right.
2786                  * Merge the new allocation with the right neighbor.
2787                  */
2788                 old = right;
2789 
2790                 right.br_startoff = new->br_startoff;
2791                 right.br_startblock = new->br_startblock;
2792                 right.br_blockcount += new->br_blockcount;
2793                 xfs_iext_update_extent(ip, state, icur, &right);
2794 
2795                 if (cur == NULL) {
2796                         rval = xfs_ilog_fext(whichfork);
2797                 } else {
2798                         rval = 0;
2799                         error = xfs_bmbt_lookup_eq(cur, &old, &i);
2800                         if (error)
2801                                 goto done;
2802                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2803                         error = xfs_bmbt_update(cur, &right);
2804                         if (error)
2805                                 goto done;
2806                 }
2807                 break;
2808 
2809         case 0:
2810                 /*
2811                  * New allocation is not contiguous with another
2812                  * real allocation.
2813                  * Insert a new entry.
2814                  */
2815                 xfs_iext_insert(ip, icur, new, state);
2816                 XFS_IFORK_NEXT_SET(ip, whichfork,
2817                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2818                 if (cur == NULL) {
2819                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2820                 } else {
2821                         rval = XFS_ILOG_CORE;
2822                         error = xfs_bmbt_lookup_eq(cur, new, &i);
2823                         if (error)
2824                                 goto done;
2825                         XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2826                         error = xfs_btree_insert(cur, &i);
2827                         if (error)
2828                                 goto done;
2829                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2830                 }
2831                 break;
2832         }
2833 
2834         /* add reverse mapping unless caller opted out */
2835         if (!(flags & XFS_BMAPI_NORMAP)) {
2836                 error = xfs_rmap_map_extent(tp, ip, whichfork, new);
2837                 if (error)
2838                         goto done;
2839         }
2840 
2841         /* convert to a btree if necessary */
2842         if (xfs_bmap_needs_btree(ip, whichfork)) {
2843                 int     tmp_logflags;   /* partial log flag return val */
2844 
2845                 ASSERT(cur == NULL);
2846                 error = xfs_bmap_extents_to_btree(tp, ip, curp, 0,
2847                                 &tmp_logflags, whichfork);
2848                 *logflagsp |= tmp_logflags;
2849                 cur = *curp;
2850                 if (error)
2851                         goto done;
2852         }
2853 
2854         /* clear out the allocated field, done with it now in any case. */
2855         if (cur)
2856                 cur->bc_private.b.allocated = 0;
2857 
2858         xfs_bmap_check_leaf_extents(cur, ip, whichfork);
2859 done:
2860         *logflagsp |= rval;
2861         return error;
2862 }
2863 
2864 /*
2865  * Functions used in the extent read, allocate and remove paths
2866  */
2867 
2868 /*
2869  * Adjust the size of the new extent based on di_extsize and rt extsize.
2870  */
2871 int
2872 xfs_bmap_extsize_align(
2873         xfs_mount_t     *mp,
2874         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
2875         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
2876         xfs_extlen_t    extsz,          /* align to this extent size */
2877         int             rt,             /* is this a realtime inode? */
2878         int             eof,            /* is extent at end-of-file? */
2879         int             delay,          /* creating delalloc extent? */
2880         int             convert,        /* overwriting unwritten extent? */
2881         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
2882         xfs_extlen_t    *lenp)          /* in/out: aligned length */
2883 {
2884         xfs_fileoff_t   orig_off;       /* original offset */
2885         xfs_extlen_t    orig_alen;      /* original length */
2886         xfs_fileoff_t   orig_end;       /* original off+len */
2887         xfs_fileoff_t   nexto;          /* next file offset */
2888         xfs_fileoff_t   prevo;          /* previous file offset */
2889         xfs_fileoff_t   align_off;      /* temp for offset */
2890         xfs_extlen_t    align_alen;     /* temp for length */
2891         xfs_extlen_t    temp;           /* temp for calculations */
2892 
2893         if (convert)
2894                 return 0;
2895 
2896         orig_off = align_off = *offp;
2897         orig_alen = align_alen = *lenp;
2898         orig_end = orig_off + orig_alen;
2899 
2900         /*
2901          * If this request overlaps an existing extent, then don't
2902          * attempt to perform any additional alignment.
2903          */
2904         if (!delay && !eof &&
2905             (orig_off >= gotp->br_startoff) &&
2906             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2907                 return 0;
2908         }
2909 
2910         /*
2911          * If the file offset is unaligned vs. the extent size
2912          * we need to align it.  This will be possible unless
2913          * the file was previously written with a kernel that didn't
2914          * perform this alignment, or if a truncate shot us in the
2915          * foot.
2916          */
2917         div_u64_rem(orig_off, extsz, &temp);
2918         if (temp) {
2919                 align_alen += temp;
2920                 align_off -= temp;
2921         }
2922 
2923         /* Same adjustment for the end of the requested area. */
2924         temp = (align_alen % extsz);
2925         if (temp)
2926                 align_alen += extsz - temp;
2927 
2928         /*
2929          * For large extent hint sizes, the aligned extent might be larger than
2930          * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
2931          * the length back under MAXEXTLEN. The outer allocation loops handle
2932          * short allocation just fine, so it is safe to do this. We only want to
2933          * do it when we are forced to, though, because it means more allocation
2934          * operations are required.
2935          */
2936         while (align_alen > MAXEXTLEN)
2937                 align_alen -= extsz;
2938         ASSERT(align_alen <= MAXEXTLEN);
2939 
2940         /*
2941          * If the previous block overlaps with this proposed allocation
2942          * then move the start forward without adjusting the length.
2943          */
2944         if (prevp->br_startoff != NULLFILEOFF) {
2945                 if (prevp->br_startblock == HOLESTARTBLOCK)
2946                         prevo = prevp->br_startoff;
2947                 else
2948                         prevo = prevp->br_startoff + prevp->br_blockcount;
2949         } else
2950                 prevo = 0;
2951         if (align_off != orig_off && align_off < prevo)
2952                 align_off = prevo;
2953         /*
2954          * If the next block overlaps with this proposed allocation
2955          * then move the start back without adjusting the length,
2956          * but not before offset 0.
2957          * This may of course make the start overlap previous block,
2958          * and if we hit the offset 0 limit then the next block
2959          * can still overlap too.
2960          */
2961         if (!eof && gotp->br_startoff != NULLFILEOFF) {
2962                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2963                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2964                         nexto = gotp->br_startoff + gotp->br_blockcount;
2965                 else
2966                         nexto = gotp->br_startoff;
2967         } else
2968                 nexto = NULLFILEOFF;
2969         if (!eof &&
2970             align_off + align_alen != orig_end &&
2971             align_off + align_alen > nexto)
2972                 align_off = nexto > align_alen ? nexto - align_alen : 0;
2973         /*
2974          * If we're now overlapping the next or previous extent that
2975          * means we can't fit an extsz piece in this hole.  Just move
2976          * the start forward to the first valid spot and set
2977          * the length so we hit the end.
2978          */
2979         if (align_off != orig_off && align_off < prevo)
2980                 align_off = prevo;
2981         if (align_off + align_alen != orig_end &&
2982             align_off + align_alen > nexto &&
2983             nexto != NULLFILEOFF) {
2984                 ASSERT(nexto > prevo);
2985                 align_alen = nexto - align_off;
2986         }
2987 
2988         /*
2989          * If realtime, and the result isn't a multiple of the realtime
2990          * extent size we need to remove blocks until it is.
2991          */
2992         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2993                 /*
2994                  * We're not covering the original request, or
2995                  * we won't be able to once we fix the length.
2996                  */
2997                 if (orig_off < align_off ||
2998                     orig_end > align_off + align_alen ||
2999                     align_alen - temp < orig_alen)
3000                         return -EINVAL;
3001                 /*
3002                  * Try to fix it by moving the start up.
3003                  */
3004                 if (align_off + temp <= orig_off) {
3005                         align_alen -= temp;
3006                         align_off += temp;
3007                 }
3008                 /*
3009                  * Try to fix it by moving the end in.
3010                  */
3011                 else if (align_off + align_alen - temp >= orig_end)
3012                         align_alen -= temp;
3013                 /*
3014                  * Set the start to the minimum then trim the length.
3015                  */
3016                 else {
3017                         align_alen -= orig_off - align_off;
3018                         align_off = orig_off;
3019                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
3020                 }
3021                 /*
3022                  * Result doesn't cover the request, fail it.
3023                  */
3024                 if (orig_off < align_off || orig_end > align_off + align_alen)
3025                         return -EINVAL;
3026         } else {
3027                 ASSERT(orig_off >= align_off);
3028                 /* see MAXEXTLEN handling above */
3029                 ASSERT(orig_end <= align_off + align_alen ||
3030                        align_alen + extsz > MAXEXTLEN);
3031         }
3032 
3033 #ifdef DEBUG
3034         if (!eof && gotp->br_startoff != NULLFILEOFF)
3035                 ASSERT(align_off + align_alen <= gotp->br_startoff);
3036         if (prevp->br_startoff != NULLFILEOFF)
3037                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3038 #endif
3039 
3040         *lenp = align_alen;
3041         *offp = align_off;
3042         return 0;
3043 }
3044 
3045 #define XFS_ALLOC_GAP_UNITS     4
3046 
3047 void
3048 xfs_bmap_adjacent(
3049         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3050 {
3051         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
3052         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3053         xfs_mount_t     *mp;            /* mount point structure */
3054         int             nullfb;         /* true if ap->firstblock isn't set */
3055         int             rt;             /* true if inode is realtime */
3056 
3057 #define ISVALID(x,y)    \
3058         (rt ? \
3059                 (x) < mp->m_sb.sb_rblocks : \
3060                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3061                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3062                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3063 
3064         mp = ap->ip->i_mount;
3065         nullfb = ap->tp->t_firstblock == NULLFSBLOCK;
3066         rt = XFS_IS_REALTIME_INODE(ap->ip) &&
3067                 xfs_alloc_is_userdata(ap->datatype);
3068         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp,
3069                                                         ap->tp->t_firstblock);
3070         /*
3071          * If allocating at eof, and there's a previous real block,
3072          * try to use its last block as our starting point.
3073          */
3074         if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3075             !isnullstartblock(ap->prev.br_startblock) &&
3076             ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3077                     ap->prev.br_startblock)) {
3078                 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3079                 /*
3080                  * Adjust for the gap between prevp and us.
3081                  */
3082                 adjust = ap->offset -
3083                         (ap->prev.br_startoff + ap->prev.br_blockcount);
3084                 if (adjust &&
3085                     ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3086                         ap->blkno += adjust;
3087         }
3088         /*
3089          * If not at eof, then compare the two neighbor blocks.
3090          * Figure out whether either one gives us a good starting point,
3091          * and pick the better one.
3092          */
3093         else if (!ap->eof) {
3094                 xfs_fsblock_t   gotbno;         /* right side block number */
3095                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
3096                 xfs_fsblock_t   prevbno;        /* left side block number */
3097                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
3098 
3099                 /*
3100                  * If there's a previous (left) block, select a requested
3101                  * start block based on it.
3102                  */
3103                 if (ap->prev.br_startoff != NULLFILEOFF &&
3104                     !isnullstartblock(ap->prev.br_startblock) &&
3105                     (prevbno = ap->prev.br_startblock +
3106                                ap->prev.br_blockcount) &&
3107                     ISVALID(prevbno, ap->prev.br_startblock)) {
3108                         /*
3109                          * Calculate gap to end of previous block.
3110                          */
3111                         adjust = prevdiff = ap->offset -
3112                                 (ap->prev.br_startoff +
3113                                  ap->prev.br_blockcount);
3114                         /*
3115                          * Figure the startblock based on the previous block's
3116                          * end and the gap size.
3117                          * Heuristic!
3118                          * If the gap is large relative to the piece we're
3119                          * allocating, or using it gives us an invalid block
3120                          * number, then just use the end of the previous block.
3121                          */
3122                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3123                             ISVALID(prevbno + prevdiff,
3124                                     ap->prev.br_startblock))
3125                                 prevbno += adjust;
3126                         else
3127                                 prevdiff += adjust;
3128                         /*
3129                          * If the firstblock forbids it, can't use it,
3130                          * must use default.
3131                          */
3132                         if (!rt && !nullfb &&
3133                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3134                                 prevbno = NULLFSBLOCK;
3135                 }
3136                 /*
3137                  * No previous block or can't follow it, just default.
3138                  */
3139                 else
3140                         prevbno = NULLFSBLOCK;
3141                 /*
3142                  * If there's a following (right) block, select a requested
3143                  * start block based on it.
3144                  */
3145                 if (!isnullstartblock(ap->got.br_startblock)) {
3146                         /*
3147                          * Calculate gap to start of next block.
3148                          */
3149                         adjust = gotdiff = ap->got.br_startoff - ap->offset;
3150                         /*
3151                          * Figure the startblock based on the next block's
3152                          * start and the gap size.
3153                          */
3154                         gotbno = ap->got.br_startblock;
3155                         /*
3156                          * Heuristic!
3157                          * If the gap is large relative to the piece we're
3158                          * allocating, or using it gives us an invalid block
3159                          * number, then just use the start of the next block
3160                          * offset by our length.
3161                          */
3162                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3163                             ISVALID(gotbno - gotdiff, gotbno))
3164                                 gotbno -= adjust;
3165                         else if (ISVALID(gotbno - ap->length, gotbno)) {
3166                                 gotbno -= ap->length;
3167                                 gotdiff += adjust - ap->length;
3168                         } else
3169                                 gotdiff += adjust;
3170                         /*
3171                          * If the firstblock forbids it, can't use it,
3172                          * must use default.
3173                          */
3174                         if (!rt && !nullfb &&
3175                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3176                                 gotbno = NULLFSBLOCK;
3177                 }
3178                 /*
3179                  * No next block, just default.
3180                  */
3181                 else
3182                         gotbno = NULLFSBLOCK;
3183                 /*
3184                  * If both valid, pick the better one, else the only good
3185                  * one, else ap->blkno is already set (to 0 or the inode block).
3186                  */
3187                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3188                         ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3189                 else if (prevbno != NULLFSBLOCK)
3190                         ap->blkno = prevbno;
3191                 else if (gotbno != NULLFSBLOCK)
3192                         ap->blkno = gotbno;
3193         }
3194 #undef ISVALID
3195 }
3196 
3197 static int
3198 xfs_bmap_longest_free_extent(
3199         struct xfs_trans        *tp,
3200         xfs_agnumber_t          ag,
3201         xfs_extlen_t            *blen,
3202         int                     *notinit)
3203 {
3204         struct xfs_mount        *mp = tp->t_mountp;
3205         struct xfs_perag        *pag;
3206         xfs_extlen_t            longest;
3207         int                     error = 0;
3208 
3209         pag = xfs_perag_get(mp, ag);
3210         if (!pag->pagf_init) {
3211                 error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3212                 if (error)
3213                         goto out;
3214 
3215                 if (!pag->pagf_init) {
3216                         *notinit = 1;
3217                         goto out;
3218                 }
3219         }
3220 
3221         longest = xfs_alloc_longest_free_extent(pag,
3222                                 xfs_alloc_min_freelist(mp, pag),
3223                                 xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
3224         if (*blen < longest)
3225                 *blen = longest;
3226 
3227 out:
3228         xfs_perag_put(pag);
3229         return error;
3230 }
3231 
3232 static void
3233 xfs_bmap_select_minlen(
3234         struct xfs_bmalloca     *ap,
3235         struct xfs_alloc_arg    *args,
3236         xfs_extlen_t            *blen,
3237         int                     notinit)
3238 {
3239         if (notinit || *blen < ap->minlen) {
3240                 /*
3241                  * Since we did a BUF_TRYLOCK above, it is possible that
3242                  * there is space for this request.
3243                  */
3244                 args->minlen = ap->minlen;
3245         } else if (*blen < args->maxlen) {
3246                 /*
3247                  * If the best seen length is less than the request length,
3248                  * use the best as the minimum.
3249                  */
3250                 args->minlen = *blen;
3251         } else {
3252                 /*
3253                  * Otherwise we've seen an extent as big as maxlen, use that
3254                  * as the minimum.
3255                  */
3256                 args->minlen = args->maxlen;
3257         }
3258 }
3259 
3260 STATIC int
3261 xfs_bmap_btalloc_nullfb(
3262         struct xfs_bmalloca     *ap,
3263         struct xfs_alloc_arg    *args,
3264         xfs_extlen_t            *blen)
3265 {
3266         struct xfs_mount        *mp = ap->ip->i_mount;
3267         xfs_agnumber_t          ag, startag;
3268         int                     notinit = 0;
3269         int                     error;
3270 
3271         args->type = XFS_ALLOCTYPE_START_BNO;
3272         args->total = ap->total;
3273 
3274         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3275         if (startag == NULLAGNUMBER)
3276                 startag = ag = 0;
3277 
3278         while (*blen < args->maxlen) {
3279                 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3280                                                      &notinit);
3281                 if (error)
3282                         return error;
3283 
3284                 if (++ag == mp->m_sb.sb_agcount)
3285                         ag = 0;
3286                 if (ag == startag)
3287                         break;
3288         }
3289 
3290         xfs_bmap_select_minlen(ap, args, blen, notinit);
3291         return 0;
3292 }
3293 
3294 STATIC int
3295 xfs_bmap_btalloc_filestreams(
3296         struct xfs_bmalloca     *ap,
3297         struct xfs_alloc_arg    *args,
3298         xfs_extlen_t            *blen)
3299 {
3300         struct xfs_mount        *mp = ap->ip->i_mount;
3301         xfs_agnumber_t          ag;
3302         int                     notinit = 0;
3303         int                     error;
3304 
3305         args->type = XFS_ALLOCTYPE_NEAR_BNO;
3306         args->total = ap->total;
3307 
3308         ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3309         if (ag == NULLAGNUMBER)
3310                 ag = 0;
3311 
3312         error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3313         if (error)
3314                 return error;
3315 
3316         if (*blen < args->maxlen) {
3317                 error = xfs_filestream_new_ag(ap, &ag);
3318                 if (error)
3319                         return error;
3320 
3321                 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3322                                                      &notinit);
3323                 if (error)
3324                         return error;
3325 
3326         }
3327 
3328         xfs_bmap_select_minlen(ap, args, blen, notinit);
3329 
3330         /*
3331          * Set the failure fallback case to look in the selected AG as stream
3332          * may have moved.
3333          */
3334         ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3335         return 0;
3336 }
3337 
3338 /* Update all inode and quota accounting for the allocation we just did. */
3339 static void
3340 xfs_bmap_btalloc_accounting(
3341         struct xfs_bmalloca     *ap,
3342         struct xfs_alloc_arg    *args)
3343 {
3344         if (ap->flags & XFS_BMAPI_COWFORK) {
3345                 /*
3346                  * COW fork blocks are in-core only and thus are treated as
3347                  * in-core quota reservation (like delalloc blocks) even when
3348                  * converted to real blocks. The quota reservation is not
3349                  * accounted to disk until blocks are remapped to the data
3350                  * fork. So if these blocks were previously delalloc, we
3351                  * already have quota reservation and there's nothing to do
3352                  * yet.
3353                  */
3354                 if (ap->wasdel) {
3355                         xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len);
3356                         return;
3357                 }
3358 
3359                 /*
3360                  * Otherwise, we've allocated blocks in a hole. The transaction
3361                  * has acquired in-core quota reservation for this extent.
3362                  * Rather than account these as real blocks, however, we reduce
3363                  * the transaction quota reservation based on the allocation.
3364                  * This essentially transfers the transaction quota reservation
3365                  * to that of a delalloc extent.
3366                  */
3367                 ap->ip->i_delayed_blks += args->len;
3368                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip, XFS_TRANS_DQ_RES_BLKS,
3369                                 -(long)args->len);
3370                 return;
3371         }
3372 
3373         /* data/attr fork only */
3374         ap->ip->i_d.di_nblocks += args->len;
3375         xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3376         if (ap->wasdel) {
3377                 ap->ip->i_delayed_blks -= args->len;
3378                 xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len);
3379         }
3380         xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3381                 ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT : XFS_TRANS_DQ_BCOUNT,
3382                 args->len);
3383 }
3384 
3385 STATIC int
3386 xfs_bmap_btalloc(
3387         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3388 {
3389         xfs_mount_t     *mp;            /* mount point structure */
3390         xfs_alloctype_t atype = 0;      /* type for allocation routines */
3391         xfs_extlen_t    align = 0;      /* minimum allocation alignment */
3392         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3393         xfs_agnumber_t  ag;
3394         xfs_alloc_arg_t args;
3395         xfs_fileoff_t   orig_offset;
3396         xfs_extlen_t    orig_length;
3397         xfs_extlen_t    blen;
3398         xfs_extlen_t    nextminlen = 0;
3399         int             nullfb;         /* true if ap->firstblock isn't set */
3400         int             isaligned;
3401         int             tryagain;
3402         int             error;
3403         int             stripe_align;
3404 
3405         ASSERT(ap->length);
3406         orig_offset = ap->offset;
3407         orig_length = ap->length;
3408 
3409         mp = ap->ip->i_mount;
3410 
3411         /* stripe alignment for allocation is determined by mount parameters */
3412         stripe_align = 0;
3413         if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3414                 stripe_align = mp->m_swidth;
3415         else if (mp->m_dalign)
3416                 stripe_align = mp->m_dalign;
3417 
3418         if (ap->flags & XFS_BMAPI_COWFORK)
3419                 align = xfs_get_cowextsz_hint(ap->ip);
3420         else if (xfs_alloc_is_userdata(ap->datatype))
3421                 align = xfs_get_extsz_hint(ap->ip);
3422         if (align) {
3423                 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3424                                                 align, 0, ap->eof, 0, ap->conv,
3425                                                 &ap->offset, &ap->length);
3426                 ASSERT(!error);
3427                 ASSERT(ap->length);
3428         }
3429 
3430 
3431         nullfb = ap->tp->t_firstblock == NULLFSBLOCK;
3432         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp,
3433                                                         ap->tp->t_firstblock);
3434         if (nullfb) {
3435                 if (xfs_alloc_is_userdata(ap->datatype) &&
3436                     xfs_inode_is_filestream(ap->ip)) {
3437                         ag = xfs_filestream_lookup_ag(ap->ip);
3438                         ag = (ag != NULLAGNUMBER) ? ag : 0;
3439                         ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3440                 } else {
3441                         ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3442                 }
3443         } else
3444                 ap->blkno = ap->tp->t_firstblock;
3445 
3446         xfs_bmap_adjacent(ap);
3447 
3448         /*
3449          * If allowed, use ap->blkno; otherwise must use firstblock since
3450          * it's in the right allocation group.
3451          */
3452         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3453                 ;
3454         else
3455                 ap->blkno = ap->tp->t_firstblock;
3456         /*
3457          * Normal allocation, done through xfs_alloc_vextent.
3458          */
3459         tryagain = isaligned = 0;
3460         memset(&args, 0, sizeof(args));
3461         args.tp = ap->tp;
3462         args.mp = mp;
3463         args.fsbno = ap->blkno;
3464         args.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
3465 
3466         /* Trim the allocation back to the maximum an AG can fit. */
3467         args.maxlen = min(ap->length, mp->m_ag_max_usable);
3468         blen = 0;
3469         if (nullfb) {
3470                 /*
3471                  * Search for an allocation group with a single extent large
3472                  * enough for the request.  If one isn't found, then adjust
3473                  * the minimum allocation size to the largest space found.
3474                  */
3475                 if (xfs_alloc_is_userdata(ap->datatype) &&
3476                     xfs_inode_is_filestream(ap->ip))
3477                         error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3478                 else
3479                         error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3480                 if (error)
3481                         return error;
3482         } else if (ap->tp->t_flags & XFS_TRANS_LOWMODE) {
3483                 if (xfs_inode_is_filestream(ap->ip))
3484                         args.type = XFS_ALLOCTYPE_FIRST_AG;
3485                 else
3486                         args.type = XFS_ALLOCTYPE_START_BNO;
3487                 args.total = args.minlen = ap->minlen;
3488         } else {
3489                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3490                 args.total = ap->total;
3491                 args.minlen = ap->minlen;
3492         }
3493         /* apply extent size hints if obtained earlier */
3494         if (align) {
3495                 args.prod = align;
3496                 div_u64_rem(ap->offset, args.prod, &args.mod);
3497                 if (args.mod)
3498                         args.mod = args.prod - args.mod;
3499         } else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3500                 args.prod = 1;
3501                 args.mod = 0;
3502         } else {
3503                 args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3504                 div_u64_rem(ap->offset, args.prod, &args.mod);
3505                 if (args.mod)
3506                         args.mod = args.prod - args.mod;
3507         }
3508         /*
3509          * If we are not low on available data blocks, and the
3510          * underlying logical volume manager is a stripe, and
3511          * the file offset is zero then try to allocate data
3512          * blocks on stripe unit boundary.
3513          * NOTE: ap->aeof is only set if the allocation length
3514          * is >= the stripe unit and the allocation offset is
3515          * at the end of file.
3516          */
3517         if (!(ap->tp->t_flags & XFS_TRANS_LOWMODE) && ap->aeof) {
3518                 if (!ap->offset) {
3519                         args.alignment = stripe_align;
3520                         atype = args.type;
3521                         isaligned = 1;
3522                         /*
3523                          * Adjust for alignment
3524                          */
3525                         if (blen > args.alignment && blen <= args.maxlen)
3526                                 args.minlen = blen - args.alignment;
3527                         args.minalignslop = 0;
3528                 } else {
3529                         /*
3530                          * First try an exact bno allocation.
3531                          * If it fails then do a near or start bno
3532                          * allocation with alignment turned on.
3533                          */
3534                         atype = args.type;
3535                         tryagain = 1;
3536                         args.type = XFS_ALLOCTYPE_THIS_BNO;
3537                         args.alignment = 1;
3538                         /*
3539                          * Compute the minlen+alignment for the
3540                          * next case.  Set slop so that the value
3541                          * of minlen+alignment+slop doesn't go up
3542                          * between the calls.
3543                          */
3544                         if (blen > stripe_align && blen <= args.maxlen)
3545                                 nextminlen = blen - stripe_align;
3546                         else
3547                                 nextminlen = args.minlen;
3548                         if (nextminlen + stripe_align > args.minlen + 1)
3549                                 args.minalignslop =
3550                                         nextminlen + stripe_align -
3551                                         args.minlen - 1;
3552                         else
3553                                 args.minalignslop = 0;
3554                 }
3555         } else {
3556                 args.alignment = 1;
3557                 args.minalignslop = 0;
3558         }
3559         args.minleft = ap->minleft;
3560         args.wasdel = ap->wasdel;
3561         args.resv = XFS_AG_RESV_NONE;
3562         args.datatype = ap->datatype;
3563         if (ap->datatype & XFS_ALLOC_USERDATA_ZERO)
3564                 args.ip = ap->ip;
3565 
3566         error = xfs_alloc_vextent(&args);
3567         if (error)
3568                 return error;
3569 
3570         if (tryagain && args.fsbno == NULLFSBLOCK) {
3571                 /*
3572                  * Exact allocation failed. Now try with alignment
3573                  * turned on.
3574                  */
3575                 args.type = atype;
3576                 args.fsbno = ap->blkno;
3577                 args.alignment = stripe_align;
3578                 args.minlen = nextminlen;
3579                 args.minalignslop = 0;
3580                 isaligned = 1;
3581                 if ((error = xfs_alloc_vextent(&args)))
3582                         return error;
3583         }
3584         if (isaligned && args.fsbno == NULLFSBLOCK) {
3585                 /*
3586                  * allocation failed, so turn off alignment and
3587                  * try again.
3588                  */
3589                 args.type = atype;
3590                 args.fsbno = ap->blkno;
3591                 args.alignment = 0;
3592                 if ((error = xfs_alloc_vextent(&args)))
3593                         return error;
3594         }
3595         if (args.fsbno == NULLFSBLOCK && nullfb &&
3596             args.minlen > ap->minlen) {
3597                 args.minlen = ap->minlen;
3598                 args.type = XFS_ALLOCTYPE_START_BNO;
3599                 args.fsbno = ap->blkno;
3600                 if ((error = xfs_alloc_vextent(&args)))
3601                         return error;
3602         }
3603         if (args.fsbno == NULLFSBLOCK && nullfb) {
3604                 args.fsbno = 0;
3605                 args.type = XFS_ALLOCTYPE_FIRST_AG;
3606                 args.total = ap->minlen;
3607                 if ((error = xfs_alloc_vextent(&args)))
3608                         return error;
3609                 ap->tp->t_flags |= XFS_TRANS_LOWMODE;
3610         }
3611         if (args.fsbno != NULLFSBLOCK) {
3612                 /*
3613                  * check the allocation happened at the same or higher AG than
3614                  * the first block that was allocated.
3615                  */
3616                 ASSERT(ap->tp->t_firstblock == NULLFSBLOCK ||
3617                        XFS_FSB_TO_AGNO(mp, ap->tp->t_firstblock) <=
3618                        XFS_FSB_TO_AGNO(mp, args.fsbno));
3619 
3620                 ap->blkno = args.fsbno;
3621                 if (ap->tp->t_firstblock == NULLFSBLOCK)
3622                         ap->tp->t_firstblock = args.fsbno;
3623                 ASSERT(nullfb || fb_agno <= args.agno);
3624                 ap->length = args.len;
3625                 /*
3626                  * If the extent size hint is active, we tried to round the
3627                  * caller's allocation request offset down to extsz and the
3628                  * length up to another extsz boundary.  If we found a free
3629                  * extent we mapped it in starting at this new offset.  If the
3630                  * newly mapped space isn't long enough to cover any of the
3631                  * range of offsets that was originally requested, move the
3632                  * mapping up so that we can fill as much of the caller's
3633                  * original request as possible.  Free space is apparently
3634                  * very fragmented so we're unlikely to be able to satisfy the
3635                  * hints anyway.
3636                  */
3637                 if (ap->length <= orig_length)
3638                         ap->offset = orig_offset;
3639                 else if (ap->offset + ap->length < orig_offset + orig_length)
3640                         ap->offset = orig_offset + orig_length - ap->length;
3641                 xfs_bmap_btalloc_accounting(ap, &args);
3642         } else {
3643                 ap->blkno = NULLFSBLOCK;
3644                 ap->length = 0;
3645         }
3646         return 0;
3647 }
3648 
3649 /*
3650  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3651  * It figures out where to ask the underlying allocator to put the new extent.
3652  */
3653 STATIC int
3654 xfs_bmap_alloc(
3655         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3656 {
3657         if (XFS_IS_REALTIME_INODE(ap->ip) &&
3658             xfs_alloc_is_userdata(ap->datatype))
3659                 return xfs_bmap_rtalloc(ap);
3660         return xfs_bmap_btalloc(ap);
3661 }
3662 
3663 /* Trim extent to fit a logical block range. */
3664 void
3665 xfs_trim_extent(
3666         struct xfs_bmbt_irec    *irec,
3667         xfs_fileoff_t           bno,
3668         xfs_filblks_t           len)
3669 {
3670         xfs_fileoff_t           distance;
3671         xfs_fileoff_t           end = bno + len;
3672 
3673         if (irec->br_startoff + irec->br_blockcount <= bno ||
3674             irec->br_startoff >= end) {
3675                 irec->br_blockcount = 0;
3676                 return;
3677         }
3678 
3679         if (irec->br_startoff < bno) {
3680                 distance = bno - irec->br_startoff;
3681                 if (isnullstartblock(irec->br_startblock))
3682                         irec->br_startblock = DELAYSTARTBLOCK;
3683                 if (irec->br_startblock != DELAYSTARTBLOCK &&
3684                     irec->br_startblock != HOLESTARTBLOCK)
3685                         irec->br_startblock += distance;
3686                 irec->br_startoff += distance;
3687                 irec->br_blockcount -= distance;
3688         }
3689 
3690         if (end < irec->br_startoff + irec->br_blockcount) {
3691                 distance = irec->br_startoff + irec->br_blockcount - end;
3692                 irec->br_blockcount -= distance;
3693         }
3694 }
3695 
3696 /*
3697  * Trim the returned map to the required bounds
3698  */
3699 STATIC void
3700 xfs_bmapi_trim_map(
3701         struct xfs_bmbt_irec    *mval,
3702         struct xfs_bmbt_irec    *got,
3703         xfs_fileoff_t           *bno,
3704         xfs_filblks_t           len,
3705         xfs_fileoff_t           obno,
3706         xfs_fileoff_t           end,
3707         int                     n,
3708         int                     flags)
3709 {
3710         if ((flags & XFS_BMAPI_ENTIRE) ||
3711             got->br_startoff + got->br_blockcount <= obno) {
3712                 *mval = *got;
3713                 if (isnullstartblock(got->br_startblock))
3714                         mval->br_startblock = DELAYSTARTBLOCK;
3715                 return;
3716         }
3717 
3718         if (obno > *bno)
3719                 *bno = obno;
3720         ASSERT((*bno >= obno) || (n == 0));
3721         ASSERT(*bno < end);
3722         mval->br_startoff = *bno;
3723         if (isnullstartblock(got->br_startblock))
3724                 mval->br_startblock = DELAYSTARTBLOCK;
3725         else
3726                 mval->br_startblock = got->br_startblock +
3727                                         (*bno - got->br_startoff);
3728         /*
3729          * Return the minimum of what we got and what we asked for for
3730          * the length.  We can use the len variable here because it is
3731          * modified below and we could have been there before coming
3732          * here if the first part of the allocation didn't overlap what
3733          * was asked for.
3734          */
3735         mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3736                         got->br_blockcount - (*bno - got->br_startoff));
3737         mval->br_state = got->br_state;
3738         ASSERT(mval->br_blockcount <= len);
3739         return;
3740 }
3741 
3742 /*
3743  * Update and validate the extent map to return
3744  */
3745 STATIC void
3746 xfs_bmapi_update_map(
3747         struct xfs_bmbt_irec    **map,
3748         xfs_fileoff_t           *bno,
3749         xfs_filblks_t           *len,
3750         xfs_fileoff_t           obno,
3751         xfs_fileoff_t           end,
3752         int                     *n,
3753         int                     flags)
3754 {
3755         xfs_bmbt_irec_t *mval = *map;
3756 
3757         ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3758                ((mval->br_startoff + mval->br_blockcount) <= end));
3759         ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3760                (mval->br_startoff < obno));
3761 
3762         *bno = mval->br_startoff + mval->br_blockcount;
3763         *len = end - *bno;
3764         if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3765                 /* update previous map with new information */
3766                 ASSERT(mval->br_startblock == mval[-1].br_startblock);
3767                 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3768                 ASSERT(mval->br_state == mval[-1].br_state);
3769                 mval[-1].br_blockcount = mval->br_blockcount;
3770                 mval[-1].br_state = mval->br_state;
3771         } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3772                    mval[-1].br_startblock != DELAYSTARTBLOCK &&
3773                    mval[-1].br_startblock != HOLESTARTBLOCK &&
3774                    mval->br_startblock == mval[-1].br_startblock +
3775                                           mval[-1].br_blockcount &&
3776                    mval[-1].br_state == mval->br_state) {
3777                 ASSERT(mval->br_startoff ==
3778                        mval[-1].br_startoff + mval[-1].br_blockcount);
3779                 mval[-1].br_blockcount += mval->br_blockcount;
3780         } else if (*n > 0 &&
3781                    mval->br_startblock == DELAYSTARTBLOCK &&
3782                    mval[-1].br_startblock == DELAYSTARTBLOCK &&
3783                    mval->br_startoff ==
3784                    mval[-1].br_startoff + mval[-1].br_blockcount) {
3785                 mval[-1].br_blockcount += mval->br_blockcount;
3786                 mval[-1].br_state = mval->br_state;
3787         } else if (!((*n == 0) &&
3788                      ((mval->br_startoff + mval->br_blockcount) <=
3789                       obno))) {
3790                 mval++;
3791                 (*n)++;
3792         }
3793         *map = mval;
3794 }
3795 
3796 /*
3797  * Map file blocks to filesystem blocks without allocation.
3798  */
3799 int
3800 xfs_bmapi_read(
3801         struct xfs_inode        *ip,
3802         xfs_fileoff_t           bno,
3803         xfs_filblks_t           len,
3804         struct xfs_bmbt_irec    *mval,
3805         int                     *nmap,
3806         int                     flags)
3807 {
3808         struct xfs_mount        *mp = ip->i_mount;
3809         struct xfs_ifork        *ifp;
3810         struct xfs_bmbt_irec    got;
3811         xfs_fileoff_t           obno;
3812         xfs_fileoff_t           end;
3813         struct xfs_iext_cursor  icur;
3814         int                     error;
3815         bool                    eof = false;
3816         int                     n = 0;
3817         int                     whichfork = xfs_bmapi_whichfork(flags);
3818 
3819         ASSERT(*nmap >= 1);
3820         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
3821                            XFS_BMAPI_COWFORK)));
3822         ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
3823 
3824         if (unlikely(XFS_TEST_ERROR(
3825             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3826              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
3827              mp, XFS_ERRTAG_BMAPIFORMAT))) {
3828                 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
3829                 return -EFSCORRUPTED;
3830         }
3831 
3832         if (XFS_FORCED_SHUTDOWN(mp))
3833                 return -EIO;
3834 
3835         XFS_STATS_INC(mp, xs_blk_mapr);
3836 
3837         ifp = XFS_IFORK_PTR(ip, whichfork);
3838         if (!ifp) {
3839                 /* No CoW fork?  Return a hole. */
3840                 if (whichfork == XFS_COW_FORK) {
3841                         mval->br_startoff = bno;
3842                         mval->br_startblock = HOLESTARTBLOCK;
3843                         mval->br_blockcount = len;
3844                         mval->br_state = XFS_EXT_NORM;
3845                         *nmap = 1;
3846                         return 0;
3847                 }
3848 
3849                 /*
3850                  * A missing attr ifork implies that the inode says we're in
3851                  * extents or btree format but failed to pass the inode fork
3852                  * verifier while trying to load it.  Treat that as a file
3853                  * corruption too.
3854                  */
3855 #ifdef DEBUG
3856                 xfs_alert(mp, "%s: inode %llu missing fork %d",
3857                                 __func__, ip->i_ino, whichfork);
3858 #endif /* DEBUG */
3859                 return -EFSCORRUPTED;
3860         }
3861 
3862         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
3863                 error = xfs_iread_extents(NULL, ip, whichfork);
3864                 if (error)
3865                         return error;
3866         }
3867 
3868         if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got))
3869                 eof = true;
3870         end = bno + len;
3871         obno = bno;
3872 
3873         while (bno < end && n < *nmap) {
3874                 /* Reading past eof, act as though there's a hole up to end. */
3875                 if (eof)
3876                         got.br_startoff = end;
3877                 if (got.br_startoff > bno) {
3878                         /* Reading in a hole.  */
3879                         mval->br_startoff = bno;
3880                         mval->br_startblock = HOLESTARTBLOCK;
3881                         mval->br_blockcount =
3882                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
3883                         mval->br_state = XFS_EXT_NORM;
3884                         bno += mval->br_blockcount;
3885                         len -= mval->br_blockcount;
3886                         mval++;
3887                         n++;
3888                         continue;
3889                 }
3890 
3891                 /* set up the extent map to return. */
3892                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
3893                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
3894 
3895                 /* If we're done, stop now. */
3896                 if (bno >= end || n >= *nmap)
3897                         break;
3898 
3899                 /* Else go on to the next record. */
3900                 if (!xfs_iext_next_extent(ifp, &icur, &got))
3901                         eof = true;
3902         }
3903         *nmap = n;
3904         return 0;
3905 }
3906 
3907 /*
3908  * Add a delayed allocation extent to an inode. Blocks are reserved from the
3909  * global pool and the extent inserted into the inode in-core extent tree.
3910  *
3911  * On entry, got refers to the first extent beyond the offset of the extent to
3912  * allocate or eof is specified if no such extent exists. On return, got refers
3913  * to the extent record that was inserted to the inode fork.
3914  *
3915  * Note that the allocated extent may have been merged with contiguous extents
3916  * during insertion into the inode fork. Thus, got does not reflect the current
3917  * state of the inode fork on return. If necessary, the caller can use lastx to
3918  * look up the updated record in the inode fork.
3919  */
3920 int
3921 xfs_bmapi_reserve_delalloc(
3922         struct xfs_inode        *ip,
3923         int                     whichfork,
3924         xfs_fileoff_t           off,
3925         xfs_filblks_t           len,
3926         xfs_filblks_t           prealloc,
3927         struct xfs_bmbt_irec    *got,
3928         struct xfs_iext_cursor  *icur,
3929         int                     eof)
3930 {
3931         struct xfs_mount        *mp = ip->i_mount;
3932         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
3933         xfs_extlen_t            alen;
3934         xfs_extlen_t            indlen;
3935         int                     error;
3936         xfs_fileoff_t           aoff = off;
3937 
3938         /*
3939          * Cap the alloc length. Keep track of prealloc so we know whether to
3940          * tag the inode before we return.
3941          */
3942         alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN);
3943         if (!eof)
3944                 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
3945         if (prealloc && alen >= len)
3946                 prealloc = alen - len;
3947 
3948         /* Figure out the extent size, adjust alen */
3949         if (whichfork == XFS_COW_FORK) {
3950                 struct xfs_bmbt_irec    prev;
3951                 xfs_extlen_t            extsz = xfs_get_cowextsz_hint(ip);
3952 
3953                 if (!xfs_iext_peek_prev_extent(ifp, icur, &prev))
3954                         prev.br_startoff = NULLFILEOFF;
3955 
3956                 error = xfs_bmap_extsize_align(mp, got, &prev, extsz, 0, eof,
3957                                                1, 0, &aoff, &alen);
3958                 ASSERT(!error);
3959         }
3960 
3961         /*
3962          * Make a transaction-less quota reservation for delayed allocation
3963          * blocks.  This number gets adjusted later.  We return if we haven't
3964          * allocated blocks already inside this loop.
3965          */
3966         error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
3967                                                 XFS_QMOPT_RES_REGBLKS);
3968         if (error)
3969                 return error;
3970 
3971         /*
3972          * Split changing sb for alen and indlen since they could be coming
3973          * from different places.
3974          */
3975         indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
3976         ASSERT(indlen > 0);
3977 
3978         error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
3979         if (error)
3980                 goto out_unreserve_quota;
3981 
3982         error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
3983         if (error)
3984                 goto out_unreserve_blocks;
3985 
3986 
3987         ip->i_delayed_blks += alen;
3988         xfs_mod_delalloc(ip->i_mount, alen + indlen);
3989 
3990         got->br_startoff = aoff;
3991         got->br_startblock = nullstartblock(indlen);
3992         got->br_blockcount = alen;
3993         got->br_state = XFS_EXT_NORM;
3994 
3995         xfs_bmap_add_extent_hole_delay(ip, whichfork, icur, got);
3996 
3997         /*
3998          * Tag the inode if blocks were preallocated. Note that COW fork
3999          * preallocation can occur at the start or end of the extent, even when
4000          * prealloc == 0, so we must also check the aligned offset and length.
4001          */
4002         if (whichfork == XFS_DATA_FORK && prealloc)
4003                 xfs_inode_set_eofblocks_tag(ip);
4004         if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len))
4005                 xfs_inode_set_cowblocks_tag(ip);
4006 
4007         return 0;
4008 
4009 out_unreserve_blocks:
4010         xfs_mod_fdblocks(mp, alen, false);
4011 out_unreserve_quota:
4012         if (XFS_IS_QUOTA_ON(mp))
4013                 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0,
4014                                                 XFS_QMOPT_RES_REGBLKS);
4015         return error;
4016 }
4017 
4018 static int
4019 xfs_bmapi_allocate(
4020         struct xfs_bmalloca     *bma)
4021 {
4022         struct xfs_mount        *mp = bma->ip->i_mount;
4023         int                     whichfork = xfs_bmapi_whichfork(bma->flags);
4024         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4025         int                     tmp_logflags = 0;
4026         int                     error;
4027 
4028         ASSERT(bma->length > 0);
4029 
4030         /*
4031          * For the wasdelay case, we could also just allocate the stuff asked
4032          * for in this bmap call but that wouldn't be as good.
4033          */
4034         if (bma->wasdel) {
4035                 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4036                 bma->offset = bma->got.br_startoff;
4037                 xfs_iext_peek_prev_extent(ifp, &bma->icur, &bma->prev);
4038         } else {
4039                 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4040                 if (!bma->eof)
4041                         bma->length = XFS_FILBLKS_MIN(bma->length,
4042                                         bma->got.br_startoff - bma->offset);
4043         }
4044 
4045         /*
4046          * Set the data type being allocated. For the data fork, the first data
4047          * in the file is treated differently to all other allocations. For the
4048          * attribute fork, we only need to ensure the allocated range is not on
4049          * the busy list.
4050          */
4051         if (!(bma->flags & XFS_BMAPI_METADATA)) {
4052                 bma->datatype = XFS_ALLOC_NOBUSY;
4053                 if (whichfork == XFS_DATA_FORK) {
4054                         if (bma->offset == 0)
4055                                 bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA;
4056                         else
4057                                 bma->datatype |= XFS_ALLOC_USERDATA;
4058                 }
4059                 if (bma->flags & XFS_BMAPI_ZERO)
4060                         bma->datatype |= XFS_ALLOC_USERDATA_ZERO;
4061         }
4062 
4063         bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4064 
4065         /*
4066          * Only want to do the alignment at the eof if it is userdata and
4067          * allocation length is larger than a stripe unit.
4068          */
4069         if (mp->m_dalign && bma->length >= mp->m_dalign &&
4070             !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4071                 error = xfs_bmap_isaeof(bma, whichfork);
4072                 if (error)
4073                         return error;
4074         }
4075 
4076         error = xfs_bmap_alloc(bma);
4077         if (error)
4078                 return error;
4079 
4080         if (bma->blkno == NULLFSBLOCK)
4081                 return 0;
4082         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur)
4083                 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4084         /*
4085          * Bump the number of extents we've allocated
4086          * in this call.
4087          */
4088         bma->nallocs++;
4089 
4090         if (bma->cur)
4091                 bma->cur->bc_private.b.flags =
4092                         bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4093 
4094         bma->got.br_startoff = bma->offset;
4095         bma->got.br_startblock = bma->blkno;
4096         bma->got.br_blockcount = bma->length;
4097         bma->got.br_state = XFS_EXT_NORM;
4098 
4099         /*
4100          * In the data fork, a wasdelay extent has been initialized, so
4101          * shouldn't be flagged as unwritten.
4102          *
4103          * For the cow fork, however, we convert delalloc reservations
4104          * (extents allocated for speculative preallocation) to
4105          * allocated unwritten extents, and only convert the unwritten
4106          * extents to real extents when we're about to write the data.
4107          */
4108         if ((!bma->wasdel || (bma->flags & XFS_BMAPI_COWFORK)) &&
4109             (bma->flags & XFS_BMAPI_PREALLOC))
4110                 bma->got.br_state = XFS_EXT_UNWRITTEN;
4111 
4112         if (bma->wasdel)
4113                 error = xfs_bmap_add_extent_delay_real(bma, whichfork);
4114         else
4115                 error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip,
4116                                 whichfork, &bma->icur, &bma->cur, &bma->got,
4117                                 &bma->logflags, bma->flags);
4118 
4119         bma->logflags |= tmp_logflags;
4120         if (error)
4121                 return error;
4122 
4123         /*
4124          * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4125          * or xfs_bmap_add_extent_hole_real might have merged it into one of
4126          * the neighbouring ones.
4127          */
4128         xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4129 
4130         ASSERT(bma->got.br_startoff <= bma->offset);
4131         ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4132                bma->offset + bma->length);
4133         ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4134                bma->got.br_state == XFS_EXT_UNWRITTEN);
4135         return 0;
4136 }
4137 
4138 STATIC int
4139 xfs_bmapi_convert_unwritten(
4140         struct xfs_bmalloca     *bma,
4141         struct xfs_bmbt_irec    *mval,
4142         xfs_filblks_t           len,
4143         int                     flags)
4144 {
4145         int                     whichfork = xfs_bmapi_whichfork(flags);
4146         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4147         int                     tmp_logflags = 0;
4148         int                     error;
4149 
4150         /* check if we need to do unwritten->real conversion */
4151         if (mval->br_state == XFS_EXT_UNWRITTEN &&
4152             (flags & XFS_BMAPI_PREALLOC))
4153                 return 0;
4154 
4155         /* check if we need to do real->unwritten conversion */
4156         if (mval->br_state == XFS_EXT_NORM &&
4157             (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4158                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4159                 return 0;
4160 
4161         /*
4162          * Modify (by adding) the state flag, if writing.
4163          */
4164         ASSERT(mval->br_blockcount <= len);
4165         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4166                 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4167                                         bma->ip, whichfork);
4168         }
4169         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4170                                 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4171 
4172         /*
4173          * Before insertion into the bmbt, zero the range being converted
4174          * if required.
4175          */
4176         if (flags & XFS_BMAPI_ZERO) {
4177                 error = xfs_zero_extent(bma->ip, mval->br_startblock,
4178                                         mval->br_blockcount);
4179                 if (error)
4180                         return error;
4181         }
4182 
4183         error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork,
4184                         &bma->icur, &bma->cur, mval, &tmp_logflags);
4185         /*
4186          * Log the inode core unconditionally in the unwritten extent conversion
4187          * path because the conversion might not have done so (e.g., if the
4188          * extent count hasn't changed). We need to make sure the inode is dirty
4189          * in the transaction for the sake of fsync(), even if nothing has
4190          * changed, because fsync() will not force the log for this transaction
4191          * unless it sees the inode pinned.
4192          *
4193          * Note: If we're only converting cow fork extents, there aren't
4194          * any on-disk updates to make, so we don't need to log anything.
4195          */
4196         if (whichfork != XFS_COW_FORK)
4197                 bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4198         if (error)
4199                 return error;
4200 
4201         /*
4202          * Update our extent pointer, given that
4203          * xfs_bmap_add_extent_unwritten_real might have merged it into one
4204          * of the neighbouring ones.
4205          */
4206         xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4207 
4208         /*
4209          * We may have combined previously unwritten space with written space,
4210          * so generate another request.
4211          */
4212         if (mval->br_blockcount < len)
4213                 return -EAGAIN;
4214         return 0;
4215 }
4216 
4217 static inline xfs_extlen_t
4218 xfs_bmapi_minleft(
4219         struct xfs_trans        *tp,
4220         struct xfs_inode        *ip,
4221         int                     fork)
4222 {
4223         if (tp && tp->t_firstblock != NULLFSBLOCK)
4224                 return 0;
4225         if (XFS_IFORK_FORMAT(ip, fork) != XFS_DINODE_FMT_BTREE)
4226                 return 1;
4227         return be16_to_cpu(XFS_IFORK_PTR(ip, fork)->if_broot->bb_level) + 1;
4228 }
4229 
4230 /*
4231  * Log whatever the flags say, even if error.  Otherwise we might miss detecting
4232  * a case where the data is changed, there's an error, and it's not logged so we
4233  * don't shutdown when we should.  Don't bother logging extents/btree changes if
4234  * we converted to the other format.
4235  */
4236 static void
4237 xfs_bmapi_finish(
4238         struct xfs_bmalloca     *bma,
4239         int                     whichfork,
4240         int                     error)
4241 {
4242         if ((bma->logflags & xfs_ilog_fext(whichfork)) &&
4243             XFS_IFORK_FORMAT(bma->ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4244                 bma->logflags &= ~xfs_ilog_fext(whichfork);
4245         else if ((bma->logflags & xfs_ilog_fbroot(whichfork)) &&
4246                  XFS_IFORK_FORMAT(bma->ip, whichfork) != XFS_DINODE_FMT_BTREE)
4247                 bma->logflags &= ~xfs_ilog_fbroot(whichfork);
4248 
4249         if (bma->logflags)
4250                 xfs_trans_log_inode(bma->tp, bma->ip, bma->logflags);
4251         if (bma->cur)
4252                 xfs_btree_del_cursor(bma->cur, error);
4253 }
4254 
4255 /*
4256  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4257  * extent state if necessary.  Details behaviour is controlled by the flags
4258  * parameter.  Only allocates blocks from a single allocation group, to avoid
4259  * locking problems.
4260  */
4261 int
4262 xfs_bmapi_write(
4263         struct xfs_trans        *tp,            /* transaction pointer */
4264         struct xfs_inode        *ip,            /* incore inode */
4265         xfs_fileoff_t           bno,            /* starting file offs. mapped */
4266         xfs_filblks_t           len,            /* length to map in file */
4267         int                     flags,          /* XFS_BMAPI_... */
4268         xfs_extlen_t            total,          /* total blocks needed */
4269         struct xfs_bmbt_irec    *mval,          /* output: map values */
4270         int                     *nmap)          /* i/o: mval size/count */
4271 {
4272         struct xfs_bmalloca     bma = {
4273                 .tp             = tp,
4274                 .ip             = ip,
4275                 .total          = total,
4276         };
4277         struct xfs_mount        *mp = ip->i_mount;
4278         struct xfs_ifork        *ifp;
4279         xfs_fileoff_t           end;            /* end of mapped file region */
4280         bool                    eof = false;    /* after the end of extents */
4281         int                     error;          /* error return */
4282         int                     n;              /* current extent index */
4283         xfs_fileoff_t           obno;           /* old block number (offset) */
4284         int                     whichfork;      /* data or attr fork */
4285 
4286 #ifdef DEBUG
4287         xfs_fileoff_t           orig_bno;       /* original block number value */
4288         int                     orig_flags;     /* original flags arg value */
4289         xfs_filblks_t           orig_len;       /* original value of len arg */
4290         struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4291         int                     orig_nmap;      /* original value of *nmap */
4292 
4293         orig_bno = bno;
4294         orig_len = len;
4295         orig_flags = flags;
4296         orig_mval = mval;
4297         orig_nmap = *nmap;
4298 #endif
4299         whichfork = xfs_bmapi_whichfork(flags);
4300 
4301         ASSERT(*nmap >= 1);
4302         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4303         ASSERT(tp != NULL);
4304         ASSERT(len > 0);
4305         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4306         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4307         ASSERT(!(flags & XFS_BMAPI_REMAP));
4308 
4309         /* zeroing is for currently only for data extents, not metadata */
4310         ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4311                         (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4312         /*
4313          * we can allocate unwritten extents or pre-zero allocated blocks,
4314          * but it makes no sense to do both at once. This would result in
4315          * zeroing the unwritten extent twice, but it still being an
4316          * unwritten extent....
4317          */
4318         ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4319                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4320 
4321         if (unlikely(XFS_TEST_ERROR(
4322             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4323              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4324              mp, XFS_ERRTAG_BMAPIFORMAT))) {
4325                 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4326                 return -EFSCORRUPTED;
4327         }
4328 
4329         if (XFS_FORCED_SHUTDOWN(mp))
4330                 return -EIO;
4331 
4332         ifp = XFS_IFORK_PTR(ip, whichfork);
4333 
4334         XFS_STATS_INC(mp, xs_blk_mapw);
4335 
4336         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4337                 error = xfs_iread_extents(tp, ip, whichfork);
4338                 if (error)
4339                         goto error0;
4340         }
4341 
4342         if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.icur, &bma.got))
4343                 eof = true;
4344         if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4345                 bma.prev.br_startoff = NULLFILEOFF;
4346         bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4347 
4348         n = 0;
4349         end = bno + len;
4350         obno = bno;
4351         while (bno < end && n < *nmap) {
4352                 bool                    need_alloc = false, wasdelay = false;
4353 
4354                 /* in hole or beyond EOF? */
4355                 if (eof || bma.got.br_startoff > bno) {
4356                         /*
4357                          * CoW fork conversions should /never/ hit EOF or
4358                          * holes.  There should always be something for us
4359                          * to work on.
4360                          */
4361                         ASSERT(!((flags & XFS_BMAPI_CONVERT) &&
4362                                  (flags & XFS_BMAPI_COWFORK)));
4363 
4364                         need_alloc = true;
4365                 } else if (isnullstartblock(bma.got.br_startblock)) {
4366                         wasdelay = true;
4367                 }
4368 
4369                 /*
4370                  * First, deal with the hole before the allocated space
4371                  * that we found, if any.
4372                  */
4373                 if (need_alloc || wasdelay) {
4374                         bma.eof = eof;
4375                         bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4376                         bma.wasdel = wasdelay;
4377                         bma.offset = bno;
4378                         bma.flags = flags;
4379 
4380                         /*
4381                          * There's a 32/64 bit type mismatch between the
4382                          * allocation length request (which can be 64 bits in
4383                          * length) and the bma length request, which is
4384                          * xfs_extlen_t and therefore 32 bits. Hence we have to
4385                          * check for 32-bit overflows and handle them here.
4386                          */
4387                         if (len > (xfs_filblks_t)MAXEXTLEN)
4388                                 bma.length = MAXEXTLEN;
4389                         else
4390                                 bma.length = len;
4391 
4392                         ASSERT(len > 0);
4393                         ASSERT(bma.length > 0);
4394                         error = xfs_bmapi_allocate(&bma);
4395                         if (error)
4396                                 goto error0;
4397                         if (bma.blkno == NULLFSBLOCK)
4398                                 break;
4399 
4400                         /*
4401                          * If this is a CoW allocation, record the data in
4402                          * the refcount btree for orphan recovery.
4403                          */
4404                         if (whichfork == XFS_COW_FORK) {
4405                                 error = xfs_refcount_alloc_cow_extent(tp,
4406                                                 bma.blkno, bma.length);
4407                                 if (error)
4408                                         goto error0;
4409                         }
4410                 }
4411 
4412                 /* Deal with the allocated space we found.  */
4413                 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4414                                                         end, n, flags);
4415 
4416                 /* Execute unwritten extent conversion if necessary */
4417                 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4418                 if (error == -EAGAIN)
4419                         continue;
4420                 if (error)
4421                         goto error0;
4422 
4423                 /* update the extent map to return */
4424                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4425 
4426                 /*
4427                  * If we're done, stop now.  Stop when we've allocated
4428                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4429                  * the transaction may get too big.
4430                  */
4431                 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4432                         break;
4433 
4434                 /* Else go on to the next record. */
4435                 bma.prev = bma.got;
4436                 if (!xfs_iext_next_extent(ifp, &bma.icur, &bma.got))
4437                         eof = true;
4438         }
4439         *nmap = n;
4440 
4441         error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4442                         whichfork);
4443         if (error)
4444                 goto error0;
4445 
4446         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4447                XFS_IFORK_NEXTENTS(ip, whichfork) >
4448                 XFS_IFORK_MAXEXT(ip, whichfork));
4449         xfs_bmapi_finish(&bma, whichfork, 0);
4450         xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4451                 orig_nmap, *nmap);
4452         return 0;
4453 error0:
4454         xfs_bmapi_finish(&bma, whichfork, error);
4455         return error;
4456 }
4457 
4458 /*
4459  * Convert an existing delalloc extent to real blocks based on file offset. This
4460  * attempts to allocate the entire delalloc extent and may require multiple
4461  * invocations to allocate the target offset if a large enough physical extent
4462  * is not available.
4463  */
4464 int
4465 xfs_bmapi_convert_delalloc(
4466         struct xfs_inode        *ip,
4467         int                     whichfork,
4468         xfs_fileoff_t           offset_fsb,
4469         struct xfs_bmbt_irec    *imap,
4470         unsigned int            *seq)
4471 {
4472         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
4473         struct xfs_mount        *mp = ip->i_mount;
4474         struct xfs_bmalloca     bma = { NULL };
4475         struct xfs_trans        *tp;
4476         int                     error;
4477 
4478         /*
4479          * Space for the extent and indirect blocks was reserved when the
4480          * delalloc extent was created so there's no need to do so here.
4481          */
4482         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0,
4483                                 XFS_TRANS_RESERVE, &tp);
4484         if (error)
4485                 return error;
4486 
4487         xfs_ilock(ip, XFS_ILOCK_EXCL);
4488         xfs_trans_ijoin(tp, ip, 0);
4489 
4490         if (!xfs_iext_lookup_extent(ip, ifp, offset_fsb, &bma.icur, &bma.got) ||
4491             bma.got.br_startoff > offset_fsb) {
4492                 /*
4493                  * No extent found in the range we are trying to convert.  This
4494                  * should only happen for the COW fork, where another thread
4495                  * might have moved the extent to the data fork in the meantime.
4496                  */
4497                 WARN_ON_ONCE(whichfork != XFS_COW_FORK);
4498                 error = -EAGAIN;
4499                 goto out_trans_cancel;
4500         }
4501 
4502         /*
4503          * If we find a real extent here we raced with another thread converting
4504          * the extent.  Just return the real extent at this offset.
4505          */
4506         if (!isnullstartblock(bma.got.br_startblock)) {
4507                 *imap = bma.got;
4508                 *seq = READ_ONCE(ifp->if_seq);
4509                 goto out_trans_cancel;
4510         }
4511 
4512         bma.tp = tp;
4513         bma.ip = ip;
4514         bma.wasdel = true;
4515         bma.offset = bma.got.br_startoff;
4516         bma.length = max_t(xfs_filblks_t, bma.got.br_blockcount, MAXEXTLEN);
4517         bma.total = XFS_EXTENTADD_SPACE_RES(ip->i_mount, XFS_DATA_FORK);
4518         bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4519         if (whichfork == XFS_COW_FORK)
4520                 bma.flags = XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC;
4521 
4522         if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4523                 bma.prev.br_startoff = NULLFILEOFF;
4524 
4525         error = xfs_bmapi_allocate(&bma);
4526         if (error)
4527                 goto out_finish;
4528 
4529         error = -ENOSPC;
4530         if (WARN_ON_ONCE(bma.blkno == NULLFSBLOCK))
4531                 goto out_finish;
4532         error = -EFSCORRUPTED;
4533         if (WARN_ON_ONCE(!bma.got.br_startblock && !XFS_IS_REALTIME_INODE(ip)))
4534                 goto out_finish;
4535 
4536         XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, bma.length));
4537         XFS_STATS_INC(mp, xs_xstrat_quick);
4538 
4539         ASSERT(!isnullstartblock(bma.got.br_startblock));
4540         *imap = bma.got;
4541         *seq = READ_ONCE(ifp->if_seq);
4542 
4543         if (whichfork == XFS_COW_FORK) {
4544                 error = xfs_refcount_alloc_cow_extent(tp, bma.blkno,
4545                                 bma.length);
4546                 if (error)
4547                         goto out_finish;
4548         }
4549 
4550         error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4551                         whichfork);
4552         if (error)
4553                 goto out_finish;
4554 
4555         xfs_bmapi_finish(&bma, whichfork, 0);
4556         error = xfs_trans_commit(tp);
4557         xfs_iunlock(ip, XFS_ILOCK_EXCL);
4558         return error;
4559 
4560 out_finish:
4561         xfs_bmapi_finish(&bma, whichfork, error);
4562 out_trans_cancel:
4563         xfs_trans_cancel(tp);
4564         xfs_iunlock(ip, XFS_ILOCK_EXCL);
4565         return error;
4566 }
4567 
4568 int
4569 xfs_bmapi_remap(
4570         struct xfs_trans        *tp,
4571         struct xfs_inode        *ip,
4572         xfs_fileoff_t           bno,
4573         xfs_filblks_t           len,
4574         xfs_fsblock_t           startblock,
4575         int                     flags)
4576 {
4577         struct xfs_mount        *mp = ip->i_mount;
4578         struct xfs_ifork        *ifp;
4579         struct xfs_btree_cur    *cur = NULL;
4580         struct xfs_bmbt_irec    got;
4581         struct xfs_iext_cursor  icur;
4582         int                     whichfork = xfs_bmapi_whichfork(flags);
4583         int                     logflags = 0, error;
4584 
4585         ifp = XFS_IFORK_PTR(ip, whichfork);
4586         ASSERT(len > 0);
4587         ASSERT(len <= (xfs_filblks_t)MAXEXTLEN);
4588         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4589         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC |
4590                            XFS_BMAPI_NORMAP)));
4591         ASSERT((flags & (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC)) !=
4592                         (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC));
4593 
4594         if (unlikely(XFS_TEST_ERROR(
4595             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4596              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4597              mp, XFS_ERRTAG_BMAPIFORMAT))) {
4598                 XFS_ERROR_REPORT("xfs_bmapi_remap", XFS_ERRLEVEL_LOW, mp);
4599                 return -EFSCORRUPTED;
4600         }
4601 
4602         if (XFS_FORCED_SHUTDOWN(mp))
4603                 return -EIO;
4604 
4605         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4606                 error = xfs_iread_extents(tp, ip, whichfork);
4607                 if (error)
4608                         return error;
4609         }
4610 
4611         if (xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
4612                 /* make sure we only reflink into a hole. */
4613                 ASSERT(got.br_startoff > bno);
4614                 ASSERT(got.br_startoff - bno >= len);
4615         }
4616 
4617         ip->i_d.di_nblocks += len;
4618         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4619 
4620         if (ifp->if_flags & XFS_IFBROOT) {
4621                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
4622                 cur->bc_private.b.flags = 0;
4623         }
4624 
4625         got.br_startoff = bno;
4626         got.br_startblock = startblock;
4627         got.br_blockcount = len;
4628         if (flags & XFS_BMAPI_PREALLOC)
4629                 got.br_state = XFS_EXT_UNWRITTEN;
4630         else
4631                 got.br_state = XFS_EXT_NORM;
4632 
4633         error = xfs_bmap_add_extent_hole_real(tp, ip, whichfork, &icur,
4634                         &cur, &got, &logflags, flags);
4635         if (error)
4636                 goto error0;
4637 
4638         error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags, whichfork);
4639 
4640 error0:
4641         if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS)
4642                 logflags &= ~XFS_ILOG_DEXT;
4643         else if (ip->i_d.di_format != XFS_DINODE_FMT_BTREE)
4644                 logflags &= ~XFS_ILOG_DBROOT;
4645 
4646         if (logflags)
4647                 xfs_trans_log_inode(tp, ip, logflags);
4648         if (cur)
4649                 xfs_btree_del_cursor(cur, error);
4650         return error;
4651 }
4652 
4653 /*
4654  * When a delalloc extent is split (e.g., due to a hole punch), the original
4655  * indlen reservation must be shared across the two new extents that are left
4656  * behind.
4657  *
4658  * Given the original reservation and the worst case indlen for the two new
4659  * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4660  * reservation fairly across the two new extents. If necessary, steal available
4661  * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4662  * ores == 1). The number of stolen blocks is returned. The availability and
4663  * subsequent accounting of stolen blocks is the responsibility of the caller.
4664  */
4665 static xfs_filblks_t
4666 xfs_bmap_split_indlen(
4667         xfs_filblks_t                   ores,           /* original res. */
4668         xfs_filblks_t                   *indlen1,       /* ext1 worst indlen */
4669         xfs_filblks_t                   *indlen2,       /* ext2 worst indlen */
4670         xfs_filblks_t                   avail)          /* stealable blocks */
4671 {
4672         xfs_filblks_t                   len1 = *indlen1;
4673         xfs_filblks_t                   len2 = *indlen2;
4674         xfs_filblks_t                   nres = len1 + len2; /* new total res. */
4675         xfs_filblks_t                   stolen = 0;
4676         xfs_filblks_t                   resfactor;
4677 
4678         /*
4679          * Steal as many blocks as we can to try and satisfy the worst case
4680          * indlen for both new extents.
4681          */
4682         if (ores < nres && avail)
4683                 stolen = XFS_FILBLKS_MIN(nres - ores, avail);
4684         ores += stolen;
4685 
4686          /* nothing else to do if we've satisfied the new reservation */
4687         if (ores >= nres)
4688                 return stolen;
4689 
4690         /*
4691          * We can't meet the total required reservation for the two extents.
4692          * Calculate the percent of the overall shortage between both extents
4693          * and apply this percentage to each of the requested indlen values.
4694          * This distributes the shortage fairly and reduces the chances that one
4695          * of the two extents is left with nothing when extents are repeatedly
4696          * split.
4697          */
4698         resfactor = (ores * 100);
4699         do_div(resfactor, nres);
4700         len1 *= resfactor;
4701         do_div(len1, 100);
4702         len2 *= resfactor;
4703         do_div(len2, 100);
4704         ASSERT(len1 + len2 <= ores);
4705         ASSERT(len1 < *indlen1 && len2 < *indlen2);
4706 
4707         /*
4708          * Hand out the remainder to each extent. If one of the two reservations
4709          * is zero, we want to make sure that one gets a block first. The loop
4710          * below starts with len1, so hand len2 a block right off the bat if it
4711          * is zero.
4712          */
4713         ores -= (len1 + len2);
4714         ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores);
4715         if (ores && !len2 && *indlen2) {
4716                 len2++;
4717                 ores--;
4718         }
4719         while (ores) {
4720                 if (len1 < *indlen1) {
4721                         len1++;
4722                         ores--;
4723                 }
4724                 if (!ores)
4725                         break;
4726                 if (len2 < *indlen2) {
4727                         len2++;
4728                         ores--;
4729                 }
4730         }
4731 
4732         *indlen1 = len1;
4733         *indlen2 = len2;
4734 
4735         return stolen;
4736 }
4737 
4738 int
4739 xfs_bmap_del_extent_delay(
4740         struct xfs_inode        *ip,
4741         int                     whichfork,
4742         struct xfs_iext_cursor  *icur,
4743         struct xfs_bmbt_irec    *got,
4744         struct xfs_bmbt_irec    *del)
4745 {
4746         struct xfs_mount        *mp = ip->i_mount;
4747         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
4748         struct xfs_bmbt_irec    new;
4749         int64_t                 da_old, da_new, da_diff = 0;
4750         xfs_fileoff_t           del_endoff, got_endoff;
4751         xfs_filblks_t           got_indlen, new_indlen, stolen;
4752         int                     state = xfs_bmap_fork_to_state(whichfork);
4753         int                     error = 0;
4754         bool                    isrt;
4755 
4756         XFS_STATS_INC(mp, xs_del_exlist);
4757 
4758         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4759         del_endoff = del->br_startoff + del->br_blockcount;
4760         got_endoff = got->br_startoff + got->br_blockcount;
4761         da_old = startblockval(got->br_startblock);
4762         da_new = 0;
4763 
4764         ASSERT(del->br_blockcount > 0);
4765         ASSERT(got->br_startoff <= del->br_startoff);
4766         ASSERT(got_endoff >= del_endoff);
4767 
4768         if (isrt) {
4769                 uint64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount);
4770 
4771                 do_div(rtexts, mp->m_sb.sb_rextsize);
4772                 xfs_mod_frextents(mp, rtexts);
4773         }
4774 
4775         /*
4776          * Update the inode delalloc counter now and wait to update the
4777          * sb counters as we might have to borrow some blocks for the
4778          * indirect block accounting.
4779          */
4780         error = xfs_trans_reserve_quota_nblks(NULL, ip,
4781                         -((long)del->br_blockcount), 0,
4782                         isrt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4783         if (error)
4784                 return error;
4785         ip->i_delayed_blks -= del->br_blockcount;
4786 
4787         if (got->br_startoff == del->br_startoff)
4788                 state |= BMAP_LEFT_FILLING;
4789         if (got_endoff == del_endoff)
4790                 state |= BMAP_RIGHT_FILLING;
4791 
4792         switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4793         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4794                 /*
4795                  * Matches the whole extent.  Delete the entry.
4796                  */
4797                 xfs_iext_remove(ip, icur, state);
4798                 xfs_iext_prev(ifp, icur);
4799                 break;
4800         case BMAP_LEFT_FILLING:
4801                 /*
4802                  * Deleting the first part of the extent.
4803                  */
4804                 got->br_startoff = del_endoff;
4805                 got->br_blockcount -= del->br_blockcount;
4806                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4807                                 got->br_blockcount), da_old);
4808                 got->br_startblock = nullstartblock((int)da_new);
4809                 xfs_iext_update_extent(ip, state, icur, got);
4810                 break;
4811         case BMAP_RIGHT_FILLING:
4812                 /*
4813                  * Deleting the last part of the extent.
4814                  */
4815                 got->br_blockcount = got->br_blockcount - del->br_blockcount;
4816                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4817                                 got->br_blockcount), da_old);
4818                 got->br_startblock = nullstartblock((int)da_new);
4819                 xfs_iext_update_extent(ip, state, icur, got);
4820                 break;
4821         case 0:
4822                 /*
4823                  * Deleting the middle of the extent.
4824                  *
4825                  * Distribute the original indlen reservation across the two new
4826                  * extents.  Steal blocks from the deleted extent if necessary.
4827                  * Stealing blocks simply fudges the fdblocks accounting below.
4828                  * Warn if either of the new indlen reservations is zero as this
4829                  * can lead to delalloc problems.
4830                  */
4831                 got->br_blockcount = del->br_startoff - got->br_startoff;
4832                 got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount);
4833 
4834                 new.br_blockcount = got_endoff - del_endoff;
4835                 new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount);
4836 
4837                 WARN_ON_ONCE(!got_indlen || !new_indlen);
4838                 stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen,
4839                                                        del->br_blockcount);
4840 
4841                 got->br_startblock = nullstartblock((int)got_indlen);
4842 
4843                 new.br_startoff = del_endoff;
4844                 new.br_state = got->br_state;
4845                 new.br_startblock = nullstartblock((int)new_indlen);
4846 
4847                 xfs_iext_update_extent(ip, state, icur, got);
4848                 xfs_iext_next(ifp, icur);
4849                 xfs_iext_insert(ip, icur, &new, state);
4850 
4851                 da_new = got_indlen + new_indlen - stolen;
4852                 del->br_blockcount -= stolen;
4853                 break;
4854         }
4855 
4856         ASSERT(da_old >= da_new);
4857         da_diff = da_old - da_new;
4858         if (!isrt)
4859                 da_diff += del->br_blockcount;
4860         if (da_diff) {
4861                 xfs_mod_fdblocks(mp, da_diff, false);
4862                 xfs_mod_delalloc(mp, -da_diff);
4863         }
4864         return error;
4865 }
4866 
4867 void
4868 xfs_bmap_del_extent_cow(
4869         struct xfs_inode        *ip,
4870         struct xfs_iext_cursor  *icur,
4871         struct xfs_bmbt_irec    *got,
4872         struct xfs_bmbt_irec    *del)
4873 {
4874         struct xfs_mount        *mp = ip->i_mount;
4875         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
4876         struct xfs_bmbt_irec    new;
4877         xfs_fileoff_t           del_endoff, got_endoff;
4878         int                     state = BMAP_COWFORK;
4879 
4880         XFS_STATS_INC(mp, xs_del_exlist);
4881 
4882         del_endoff = del->br_startoff + del->br_blockcount;
4883         got_endoff = got->br_startoff + got->br_blockcount;
4884 
4885         ASSERT(del->br_blockcount > 0);
4886         ASSERT(got->br_startoff <= del->br_startoff);
4887         ASSERT(got_endoff >= del_endoff);
4888         ASSERT(!isnullstartblock(got->br_startblock));
4889 
4890         if (got->br_startoff == del->br_startoff)
4891                 state |= BMAP_LEFT_FILLING;
4892         if (got_endoff == del_endoff)
4893                 state |= BMAP_RIGHT_FILLING;
4894 
4895         switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4896         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4897                 /*
4898                  * Matches the whole extent.  Delete the entry.
4899                  */
4900                 xfs_iext_remove(ip, icur, state);
4901                 xfs_iext_prev(ifp, icur);
4902                 break;
4903         case BMAP_LEFT_FILLING:
4904                 /*
4905                  * Deleting the first part of the extent.
4906                  */
4907                 got->br_startoff = del_endoff;
4908                 got->br_blockcount -= del->br_blockcount;
4909                 got->br_startblock = del->br_startblock + del->br_blockcount;
4910                 xfs_iext_update_extent(ip, state, icur, got);
4911                 break;
4912         case BMAP_RIGHT_FILLING:
4913                 /*
4914                  * Deleting the last part of the extent.
4915                  */
4916                 got->br_blockcount -= del->br_blockcount;
4917                 xfs_iext_update_extent(ip, state, icur, got);
4918                 break;
4919         case 0:
4920                 /*
4921                  * Deleting the middle of the extent.
4922                  */
4923                 got->br_blockcount = del->br_startoff - got->br_startoff;
4924 
4925                 new.br_startoff = del_endoff;
4926                 new.br_blockcount = got_endoff - del_endoff;
4927                 new.br_state = got->br_state;
4928                 new.br_startblock = del->br_startblock + del->br_blockcount;
4929 
4930                 xfs_iext_update_extent(ip, state, icur, got);
4931                 xfs_iext_next(ifp, icur);
4932                 xfs_iext_insert(ip, icur, &new, state);
4933                 break;
4934         }
4935         ip->i_delayed_blks -= del->br_blockcount;
4936 }
4937 
4938 /*
4939  * Called by xfs_bmapi to update file extent records and the btree
4940  * after removing space.
4941  */
4942 STATIC int                              /* error */
4943 xfs_bmap_del_extent_real(
4944         xfs_inode_t             *ip,    /* incore inode pointer */
4945         xfs_trans_t             *tp,    /* current transaction pointer */
4946         struct xfs_iext_cursor  *icur,
4947         xfs_btree_cur_t         *cur,   /* if null, not a btree */
4948         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
4949         int                     *logflagsp, /* inode logging flags */
4950         int                     whichfork, /* data or attr fork */
4951         int                     bflags) /* bmapi flags */
4952 {
4953         xfs_fsblock_t           del_endblock=0; /* first block past del */
4954         xfs_fileoff_t           del_endoff;     /* first offset past del */
4955         int                     do_fx;  /* free extent at end of routine */
4956         int                     error;  /* error return value */
4957         int                     flags = 0;/* inode logging flags */
4958         struct xfs_bmbt_irec    got;    /* current extent entry */
4959         xfs_fileoff_t           got_endoff;     /* first offset past got */
4960         int                     i;      /* temp state */
4961         struct xfs_ifork        *ifp;   /* inode fork pointer */
4962         xfs_mount_t             *mp;    /* mount structure */
4963         xfs_filblks_t           nblks;  /* quota/sb block count */
4964         xfs_bmbt_irec_t         new;    /* new record to be inserted */
4965         /* REFERENCED */
4966         uint                    qfield; /* quota field to update */
4967         int                     state = xfs_bmap_fork_to_state(whichfork);
4968         struct xfs_bmbt_irec    old;
4969 
4970         mp = ip->i_mount;
4971         XFS_STATS_INC(mp, xs_del_exlist);
4972 
4973         ifp = XFS_IFORK_PTR(ip, whichfork);
4974         ASSERT(del->br_blockcount > 0);
4975         xfs_iext_get_extent(ifp, icur, &got);
4976         ASSERT(got.br_startoff <= del->br_startoff);
4977         del_endoff = del->br_startoff + del->br_blockcount;
4978         got_endoff = got.br_startoff + got.br_blockcount;
4979         ASSERT(got_endoff >= del_endoff);
4980         ASSERT(!isnullstartblock(got.br_startblock));
4981         qfield = 0;
4982         error = 0;
4983 
4984         /*
4985          * If it's the case where the directory code is running with no block
4986          * reservation, and the deleted block is in the middle of its extent,
4987          * and the resulting insert of an extent would cause transformation to
4988          * btree format, then reject it.  The calling code will then swap blocks
4989          * around instead.  We have to do this now, rather than waiting for the
4990          * conversion to btree format, since the transaction will be dirty then.
4991          */
4992         if (tp->t_blk_res == 0 &&
4993             XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
4994             XFS_IFORK_NEXTENTS(ip, whichfork) >=
4995                         XFS_IFORK_MAXEXT(ip, whichfork) &&
4996             del->br_startoff > got.br_startoff && del_endoff < got_endoff)
4997                 return -ENOSPC;
4998 
4999         flags = XFS_ILOG_CORE;
5000         if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
5001                 xfs_fsblock_t   bno;
5002                 xfs_filblks_t   len;
5003                 xfs_extlen_t    mod;
5004 
5005                 bno = div_u64_rem(del->br_startblock, mp->m_sb.sb_rextsize,
5006                                   &mod);
5007                 ASSERT(mod == 0);
5008                 len = div_u64_rem(del->br_blockcount, mp->m_sb.sb_rextsize,
5009                                   &mod);
5010                 ASSERT(mod == 0);
5011 
5012                 error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
5013                 if (error)
5014                         goto done;
5015                 do_fx = 0;
5016                 nblks = len * mp->m_sb.sb_rextsize;
5017                 qfield = XFS_TRANS_DQ_RTBCOUNT;
5018         } else {
5019                 do_fx = 1;
5020                 nblks = del->br_blockcount;
5021                 qfield = XFS_TRANS_DQ_BCOUNT;
5022         }
5023 
5024         del_endblock = del->br_startblock + del->br_blockcount;
5025         if (cur) {
5026                 error = xfs_bmbt_lookup_eq(cur, &got, &i);
5027                 if (error)
5028                         goto done;
5029                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5030         }
5031 
5032         if (got.br_startoff == del->br_startoff)
5033                 state |= BMAP_LEFT_FILLING;
5034         if (got_endoff == del_endoff)
5035                 state |= BMAP_RIGHT_FILLING;
5036 
5037         switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
5038         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
5039                 /*
5040                  * Matches the whole extent.  Delete the entry.
5041                  */
5042                 xfs_iext_remove(ip, icur, state);
5043                 xfs_iext_prev(ifp, icur);
5044                 XFS_IFORK_NEXT_SET(ip, whichfork,
5045                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5046                 flags |= XFS_ILOG_CORE;
5047                 if (!cur) {
5048                         flags |= xfs_ilog_fext(whichfork);
5049                         break;
5050                 }
5051                 if ((error = xfs_btree_delete(cur, &i)))
5052                         goto done;
5053                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5054                 break;
5055         case BMAP_LEFT_FILLING:
5056                 /*
5057                  * Deleting the first part of the extent.
5058                  */
5059                 got.br_startoff = del_endoff;
5060                 got.br_startblock = del_endblock;
5061                 got.br_blockcount -= del->br_blockcount;
5062                 xfs_iext_update_extent(ip, state, icur, &got);
5063                 if (!cur) {
5064                         flags |= xfs_ilog_fext(whichfork);
5065                         break;
5066                 }
5067                 error = xfs_bmbt_update(cur, &got);
5068                 if (error)
5069                         goto done;
5070                 break;
5071         case BMAP_RIGHT_FILLING:
5072                 /*
5073                  * Deleting the last part of the extent.
5074                  */
5075                 got.br_blockcount -= del->br_blockcount;
5076                 xfs_iext_update_extent(ip, state, icur, &got);
5077                 if (!cur) {
5078                         flags |= xfs_ilog_fext(whichfork);
5079                         break;
5080                 }
5081                 error = xfs_bmbt_update(cur, &got);
5082                 if (error)
5083                         goto done;
5084                 break;
5085         case 0:
5086                 /*
5087                  * Deleting the middle of the extent.
5088                  */
5089                 old = got;
5090 
5091                 got.br_blockcount = del->br_startoff - got.br_startoff;
5092                 xfs_iext_update_extent(ip, state, icur, &got);
5093 
5094                 new.br_startoff = del_endoff;
5095                 new.br_blockcount = got_endoff - del_endoff;
5096                 new.br_state = got.br_state;
5097                 new.br_startblock = del_endblock;
5098 
5099                 flags |= XFS_ILOG_CORE;
5100                 if (cur) {
5101                         error = xfs_bmbt_update(cur, &got);
5102                         if (error)
5103                                 goto done;
5104                         error = xfs_btree_increment(cur, 0, &i);
5105                         if (error)
5106                                 goto done;
5107                         cur->bc_rec.b = new;
5108                         error = xfs_btree_insert(cur, &i);
5109                         if (error && error != -ENOSPC)
5110                                 goto done;
5111                         /*
5112                          * If get no-space back from btree insert, it tried a
5113                          * split, and we have a zero block reservation.  Fix up
5114                          * our state and return the error.
5115                          */
5116                         if (error == -ENOSPC) {
5117                                 /*
5118                                  * Reset the cursor, don't trust it after any
5119                                  * insert operation.
5120                                  */
5121                                 error = xfs_bmbt_lookup_eq(cur, &got, &i);
5122                                 if (error)
5123                                         goto done;
5124                                 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5125                                 /*
5126                                  * Update the btree record back
5127                                  * to the original value.
5128                                  */
5129                                 error = xfs_bmbt_update(cur, &old);
5130                                 if (error)
5131                                         goto done;
5132                                 /*
5133                                  * Reset the extent record back
5134                                  * to the original value.
5135                                  */
5136                                 xfs_iext_update_extent(ip, state, icur, &old);
5137                                 flags = 0;
5138                                 error = -ENOSPC;
5139                                 goto done;
5140                         }
5141                         XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5142                 } else
5143                         flags |= xfs_ilog_fext(whichfork);
5144                 XFS_IFORK_NEXT_SET(ip, whichfork,
5145                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5146                 xfs_iext_next(ifp, icur);
5147                 xfs_iext_insert(ip, icur, &new, state);
5148                 break;
5149         }
5150 
5151         /* remove reverse mapping */
5152         error = xfs_rmap_unmap_extent(tp, ip, whichfork, del);
5153         if (error)
5154                 goto done;
5155 
5156         /*
5157          * If we need to, add to list of extents to delete.
5158          */
5159         if (do_fx && !(bflags & XFS_BMAPI_REMAP)) {
5160                 if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
5161                         error = xfs_refcount_decrease_extent(tp, del);
5162                         if (error)
5163                                 goto done;
5164                 } else {
5165                         __xfs_bmap_add_free(tp, del->br_startblock,
5166                                         del->br_blockcount, NULL,
5167                                         (bflags & XFS_BMAPI_NODISCARD) ||
5168                                         del->br_state == XFS_EXT_UNWRITTEN);
5169                 }
5170         }
5171 
5172         /*
5173          * Adjust inode # blocks in the file.
5174          */
5175         if (nblks)
5176                 ip->i_d.di_nblocks -= nblks;
5177         /*
5178          * Adjust quota data.
5179          */
5180         if (qfield && !(bflags & XFS_BMAPI_REMAP))
5181                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5182 
5183 done:
5184         *logflagsp = flags;
5185         return error;
5186 }
5187 
5188 /*
5189  * Unmap (remove) blocks from a file.
5190  * If nexts is nonzero then the number of extents to remove is limited to
5191  * that value.  If not all extents in the block range can be removed then
5192  * *done is set.
5193  */
5194 int                                             /* error */
5195 __xfs_bunmapi(
5196         struct xfs_trans        *tp,            /* transaction pointer */
5197         struct xfs_inode        *ip,            /* incore inode */
5198         xfs_fileoff_t           start,          /* first file offset deleted */
5199         xfs_filblks_t           *rlen,          /* i/o: amount remaining */
5200         int                     flags,          /* misc flags */
5201         xfs_extnum_t            nexts)          /* number of extents max */
5202 {
5203         struct xfs_btree_cur    *cur;           /* bmap btree cursor */
5204         struct xfs_bmbt_irec    del;            /* extent being deleted */
5205         int                     error;          /* error return value */
5206         xfs_extnum_t            extno;          /* extent number in list */
5207         struct xfs_bmbt_irec    got;            /* current extent record */
5208         struct xfs_ifork        *ifp;           /* inode fork pointer */
5209         int                     isrt;           /* freeing in rt area */
5210         int                     logflags;       /* transaction logging flags */
5211         xfs_extlen_t            mod;            /* rt extent offset */
5212         struct xfs_mount        *mp;            /* mount structure */
5213         int                     tmp_logflags;   /* partial logging flags */
5214         int                     wasdel;         /* was a delayed alloc extent */
5215         int                     whichfork;      /* data or attribute fork */
5216         xfs_fsblock_t           sum;
5217         xfs_filblks_t           len = *rlen;    /* length to unmap in file */
5218         xfs_fileoff_t           max_len;
5219         xfs_agnumber_t          prev_agno = NULLAGNUMBER, agno;
5220         xfs_fileoff_t           end;
5221         struct xfs_iext_cursor  icur;
5222         bool                    done = false;
5223 
5224         trace_xfs_bunmap(ip, start, len, flags, _RET_IP_);
5225 
5226         whichfork = xfs_bmapi_whichfork(flags);
5227         ASSERT(whichfork != XFS_COW_FORK);
5228         ifp = XFS_IFORK_PTR(ip, whichfork);
5229         if (unlikely(
5230             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5231             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5232                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5233                                  ip->i_mount);
5234                 return -EFSCORRUPTED;
5235         }
5236         mp = ip->i_mount;
5237         if (XFS_FORCED_SHUTDOWN(mp))
5238                 return -EIO;
5239 
5240         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5241         ASSERT(len > 0);
5242         ASSERT(nexts >= 0);
5243 
5244         /*
5245          * Guesstimate how many blocks we can unmap without running the risk of
5246          * blowing out the transaction with a mix of EFIs and reflink
5247          * adjustments.
5248          */
5249         if (tp && xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK)
5250                 max_len = min(len, xfs_refcount_max_unmap(tp->t_log_res));
5251         else
5252                 max_len = len;
5253 
5254         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5255             (error = xfs_iread_extents(tp, ip, whichfork)))
5256                 return error;
5257         if (xfs_iext_count(ifp) == 0) {
5258                 *rlen = 0;
5259                 return 0;
5260         }
5261         XFS_STATS_INC(mp, xs_blk_unmap);
5262         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5263         end = start + len;
5264 
5265         if (!xfs_iext_lookup_extent_before(ip, ifp, &end, &icur, &got)) {
5266                 *rlen = 0;
5267                 return 0;
5268         }
5269         end--;
5270 
5271         logflags = 0;
5272         if (ifp->if_flags & XFS_IFBROOT) {
5273                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5274                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5275                 cur->bc_private.b.flags = 0;
5276         } else
5277                 cur = NULL;
5278 
5279         if (isrt) {
5280                 /*
5281                  * Synchronize by locking the bitmap inode.
5282                  */
5283                 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
5284                 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5285                 xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
5286                 xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
5287         }
5288 
5289         extno = 0;
5290         while (end != (xfs_fileoff_t)-1 && end >= start &&
5291                (nexts == 0 || extno < nexts) && max_len > 0) {
5292                 /*
5293                  * Is the found extent after a hole in which end lives?
5294                  * Just back up to the previous extent, if so.
5295                  */
5296                 if (got.br_startoff > end &&
5297                     !xfs_iext_prev_extent(ifp, &icur, &got)) {
5298                         done = true;
5299                         break;
5300                 }
5301                 /*
5302                  * Is the last block of this extent before the range
5303                  * we're supposed to delete?  If so, we're done.
5304                  */
5305                 end = XFS_FILEOFF_MIN(end,
5306                         got.br_startoff + got.br_blockcount - 1);
5307                 if (end < start)
5308                         break;
5309                 /*
5310                  * Then deal with the (possibly delayed) allocated space
5311                  * we found.
5312                  */
5313                 del = got;
5314                 wasdel = isnullstartblock(del.br_startblock);
5315 
5316                 /*
5317                  * Make sure we don't touch multiple AGF headers out of order
5318                  * in a single transaction, as that could cause AB-BA deadlocks.
5319                  */
5320                 if (!wasdel) {
5321                         agno = XFS_FSB_TO_AGNO(mp, del.br_startblock);
5322                         if (prev_agno != NULLAGNUMBER && prev_agno > agno)
5323                                 break;
5324                         prev_agno = agno;
5325                 }
5326                 if (got.br_startoff < start) {
5327                         del.br_startoff = start;
5328                         del.br_blockcount -= start - got.br_startoff;
5329                         if (!wasdel)
5330                                 del.br_startblock += start - got.br_startoff;
5331                 }
5332                 if (del.br_startoff + del.br_blockcount > end + 1)
5333                         del.br_blockcount = end + 1 - del.br_startoff;
5334 
5335                 /* How much can we safely unmap? */
5336                 if (max_len < del.br_blockcount) {
5337                         del.br_startoff += del.br_blockcount - max_len;
5338                         if (!wasdel)
5339                                 del.br_startblock += del.br_blockcount - max_len;
5340                         del.br_blockcount = max_len;
5341                 }
5342 
5343                 if (!isrt)
5344                         goto delete;
5345 
5346                 sum = del.br_startblock + del.br_blockcount;
5347                 div_u64_rem(sum, mp->m_sb.sb_rextsize, &mod);
5348                 if (mod) {
5349                         /*
5350                          * Realtime extent not lined up at the end.
5351                          * The extent could have been split into written
5352                          * and unwritten pieces, or we could just be
5353                          * unmapping part of it.  But we can't really
5354                          * get rid of part of a realtime extent.
5355                          */
5356                         if (del.br_state == XFS_EXT_UNWRITTEN) {
5357                                 /*
5358                                  * This piece is unwritten, or we're not
5359                                  * using unwritten extents.  Skip over it.
5360                                  */
5361                                 ASSERT(end >= mod);
5362                                 end -= mod > del.br_blockcount ?
5363                                         del.br_blockcount : mod;
5364                                 if (end < got.br_startoff &&
5365                                     !xfs_iext_prev_extent(ifp, &icur, &got)) {
5366                                         done = true;
5367                                         break;
5368                                 }
5369                                 continue;
5370                         }
5371                         /*
5372                          * It's written, turn it unwritten.
5373                          * This is better than zeroing it.
5374                          */
5375                         ASSERT(del.br_state == XFS_EXT_NORM);
5376                         ASSERT(tp->t_blk_res > 0);
5377                         /*
5378                          * If this spans a realtime extent boundary,
5379                          * chop it back to the start of the one we end at.
5380                          */
5381                         if (del.br_blockcount > mod) {
5382                                 del.br_startoff += del.br_blockcount - mod;
5383                                 del.br_startblock += del.br_blockcount - mod;
5384                                 del.br_blockcount = mod;
5385                         }
5386                         del.br_state = XFS_EXT_UNWRITTEN;
5387                         error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5388                                         whichfork, &icur, &cur, &del,
5389                                         &logflags);
5390                         if (error)
5391                                 goto error0;
5392                         goto nodelete;
5393                 }
5394                 div_u64_rem(del.br_startblock, mp->m_sb.sb_rextsize, &mod);
5395                 if (mod) {
5396                         /*
5397                          * Realtime extent is lined up at the end but not
5398                          * at the front.  We'll get rid of full extents if
5399                          * we can.
5400                          */
5401                         mod = mp->m_sb.sb_rextsize - mod;
5402                         if (del.br_blockcount > mod) {
5403                                 del.br_blockcount -= mod;
5404                                 del.br_startoff += mod;
5405                                 del.br_startblock += mod;
5406                         } else if (del.br_startoff == start &&
5407                                    (del.br_state == XFS_EXT_UNWRITTEN ||
5408                                     tp->t_blk_res == 0)) {
5409                                 /*
5410                                  * Can't make it unwritten.  There isn't
5411                                  * a full extent here so just skip it.
5412                                  */
5413                                 ASSERT(end >= del.br_blockcount);
5414                                 end -= del.br_blockcount;
5415                                 if (got.br_startoff > end &&
5416                                     !xfs_iext_prev_extent(ifp, &icur, &got)) {
5417                                         done = true;
5418                                         break;
5419                                 }
5420                                 continue;
5421                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5422                                 struct xfs_bmbt_irec    prev;
5423 
5424                                 /*
5425                                  * This one is already unwritten.
5426                                  * It must have a written left neighbor.
5427                                  * Unwrite the killed part of that one and
5428                                  * try again.
5429                                  */
5430                                 if (!xfs_iext_prev_extent(ifp, &icur, &prev))
5431                                         ASSERT(0);
5432                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5433                                 ASSERT(!isnullstartblock(prev.br_startblock));
5434                                 ASSERT(del.br_startblock ==
5435                                        prev.br_startblock + prev.br_blockcount);
5436                                 if (prev.br_startoff < start) {
5437                                         mod = start - prev.br_startoff;
5438                                         prev.br_blockcount -= mod;
5439                                         prev.br_startblock += mod;
5440                                         prev.br_startoff = start;
5441                                 }
5442                                 prev.br_state = XFS_EXT_UNWRITTEN;
5443                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5444                                                 ip, whichfork, &icur, &cur,
5445                                                 &prev, &logflags);
5446                                 if (error)
5447                                         goto error0;
5448                                 goto nodelete;
5449                         } else {
5450                                 ASSERT(del.br_state == XFS_EXT_NORM);
5451                                 del.br_state = XFS_EXT_UNWRITTEN;
5452                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5453                                                 ip, whichfork, &icur, &cur,
5454                                                 &del, &logflags);
5455                                 if (error)
5456                                         goto error0;
5457                                 goto nodelete;
5458                         }
5459                 }
5460 
5461 delete:
5462                 if (wasdel) {
5463                         error = xfs_bmap_del_extent_delay(ip, whichfork, &icur,
5464                                         &got, &del);
5465                 } else {
5466                         error = xfs_bmap_del_extent_real(ip, tp, &icur, cur,
5467                                         &del, &tmp_logflags, whichfork,
5468                                         flags);
5469                         logflags |= tmp_logflags;
5470                 }
5471 
5472                 if (error)
5473                         goto error0;
5474 
5475                 max_len -= del.br_blockcount;
5476                 end = del.br_startoff - 1;
5477 nodelete:
5478                 /*
5479                  * If not done go on to the next (previous) record.
5480                  */
5481                 if (end != (xfs_fileoff_t)-1 && end >= start) {
5482                         if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5483                             (got.br_startoff > end &&
5484                              !xfs_iext_prev_extent(ifp, &icur, &got))) {
5485                                 done = true;
5486                                 break;
5487                         }
5488                         extno++;
5489                 }
5490         }
5491         if (done || end == (xfs_fileoff_t)-1 || end < start)
5492                 *rlen = 0;
5493         else
5494                 *rlen = end - start + 1;
5495 
5496         /*
5497          * Convert to a btree if necessary.
5498          */
5499         if (xfs_bmap_needs_btree(ip, whichfork)) {
5500                 ASSERT(cur == NULL);
5501                 error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
5502                                 &tmp_logflags, whichfork);
5503                 logflags |= tmp_logflags;
5504         } else {
5505                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags,
5506                         whichfork);
5507         }
5508 
5509 error0:
5510         /*
5511          * Log everything.  Do this after conversion, there's no point in
5512          * logging the extent records if we've converted to btree format.
5513          */
5514         if ((logflags & xfs_ilog_fext(whichfork)) &&
5515             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5516                 logflags &= ~xfs_ilog_fext(whichfork);
5517         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5518                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5519                 logflags &= ~xfs_ilog_fbroot(whichfork);
5520         /*
5521          * Log inode even in the error case, if the transaction
5522          * is dirty we'll need to shut down the filesystem.
5523          */
5524         if (logflags)
5525                 xfs_trans_log_inode(tp, ip, logflags);
5526         if (cur) {
5527                 if (!error)
5528                         cur->bc_private.b.allocated = 0;
5529                 xfs_btree_del_cursor(cur, error);
5530         }
5531         return error;
5532 }
5533 
5534 /* Unmap a range of a file. */
5535 int
5536 xfs_bunmapi(
5537         xfs_trans_t             *tp,
5538         struct xfs_inode        *ip,
5539         xfs_fileoff_t           bno,
5540         xfs_filblks_t           len,
5541         int                     flags,
5542         xfs_extnum_t            nexts,
5543         int                     *done)
5544 {
5545         int                     error;
5546 
5547         error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts);
5548         *done = (len == 0);
5549         return error;
5550 }
5551 
5552 /*
5553  * Determine whether an extent shift can be accomplished by a merge with the
5554  * extent that precedes the target hole of the shift.
5555  */
5556 STATIC bool
5557 xfs_bmse_can_merge(
5558         struct xfs_bmbt_irec    *left,  /* preceding extent */
5559         struct xfs_bmbt_irec    *got,   /* current extent to shift */
5560         xfs_fileoff_t           shift)  /* shift fsb */
5561 {
5562         xfs_fileoff_t           startoff;
5563 
5564         startoff = got->br_startoff - shift;
5565 
5566         /*
5567          * The extent, once shifted, must be adjacent in-file and on-disk with
5568          * the preceding extent.
5569          */
5570         if ((left->br_startoff + left->br_blockcount != startoff) ||
5571             (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5572             (left->br_state != got->br_state) ||
5573             (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5574                 return false;
5575 
5576         return true;
5577 }
5578 
5579 /*
5580  * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5581  * hole in the file. If an extent shift would result in the extent being fully
5582  * adjacent to the extent that currently precedes the hole, we can merge with
5583  * the preceding extent rather than do the shift.
5584  *
5585  * This function assumes the caller has verified a shift-by-merge is possible
5586  * with the provided extents via xfs_bmse_can_merge().
5587  */
5588 STATIC int
5589 xfs_bmse_merge(
5590         struct xfs_trans                *tp,
5591         struct xfs_inode                *ip,
5592         int                             whichfork,
5593         xfs_fileoff_t                   shift,          /* shift fsb */
5594         struct xfs_iext_cursor          *icur,
5595         struct xfs_bmbt_irec            *got,           /* extent to shift */
5596         struct xfs_bmbt_irec            *left,          /* preceding extent */
5597         struct xfs_btree_cur            *cur,
5598         int                             *logflags)      /* output */
5599 {
5600         struct xfs_bmbt_irec            new;
5601         xfs_filblks_t                   blockcount;
5602         int                             error, i;
5603         struct xfs_mount                *mp = ip->i_mount;
5604 
5605         blockcount = left->br_blockcount + got->br_blockcount;
5606 
5607         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5608         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5609         ASSERT(xfs_bmse_can_merge(left, got, shift));
5610 
5611         new = *left;
5612         new.br_blockcount = blockcount;