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

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

Version: ~ [ linux-5.6-rc3 ] ~ [ linux-5.5.6 ] ~ [ linux-5.4.22 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.106 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.171 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.214 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.214 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.82 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ 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 /*
  2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
  3  * All Rights Reserved.
  4  *
  5  * This program is free software; you can redistribute it and/or
  6  * modify it under the terms of the GNU General Public License as
  7  * published by the Free Software Foundation.
  8  *
  9  * This program is distributed in the hope that it would be useful,
 10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
 11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 12  * GNU General Public License for more details.
 13  *
 14  * You should have received a copy of the GNU General Public License
 15  * along with this program; if not, write the Free Software Foundation,
 16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 17  */
 18 #include "xfs.h"
 19 #include "xfs_fs.h"
 20 #include "xfs_shared.h"
 21 #include "xfs_format.h"
 22 #include "xfs_log_format.h"
 23 #include "xfs_trans_resv.h"
 24 #include "xfs_bit.h"
 25 #include "xfs_inum.h"
 26 #include "xfs_sb.h"
 27 #include "xfs_ag.h"
 28 #include "xfs_mount.h"
 29 #include "xfs_da_format.h"
 30 #include "xfs_da_btree.h"
 31 #include "xfs_dir2.h"
 32 #include "xfs_inode.h"
 33 #include "xfs_btree.h"
 34 #include "xfs_trans.h"
 35 #include "xfs_inode_item.h"
 36 #include "xfs_extfree_item.h"
 37 #include "xfs_alloc.h"
 38 #include "xfs_bmap.h"
 39 #include "xfs_bmap_util.h"
 40 #include "xfs_bmap_btree.h"
 41 #include "xfs_rtalloc.h"
 42 #include "xfs_error.h"
 43 #include "xfs_quota.h"
 44 #include "xfs_trans_space.h"
 45 #include "xfs_buf_item.h"
 46 #include "xfs_trace.h"
 47 #include "xfs_symlink.h"
 48 #include "xfs_attr_leaf.h"
 49 #include "xfs_dinode.h"
 50 #include "xfs_filestream.h"
 51 
 52 
 53 kmem_zone_t             *xfs_bmap_free_item_zone;
 54 
 55 /*
 56  * Miscellaneous helper functions
 57  */
 58 
 59 /*
 60  * Compute and fill in the value of the maximum depth of a bmap btree
 61  * in this filesystem.  Done once, during mount.
 62  */
 63 void
 64 xfs_bmap_compute_maxlevels(
 65         xfs_mount_t     *mp,            /* file system mount structure */
 66         int             whichfork)      /* data or attr fork */
 67 {
 68         int             level;          /* btree level */
 69         uint            maxblocks;      /* max blocks at this level */
 70         uint            maxleafents;    /* max leaf entries possible */
 71         int             maxrootrecs;    /* max records in root block */
 72         int             minleafrecs;    /* min records in leaf block */
 73         int             minnoderecs;    /* min records in node block */
 74         int             sz;             /* root block size */
 75 
 76         /*
 77          * The maximum number of extents in a file, hence the maximum
 78          * number of leaf entries, is controlled by the type of di_nextents
 79          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
 80          * (a signed 16-bit number, xfs_aextnum_t).
 81          *
 82          * Note that we can no longer assume that if we are in ATTR1 that
 83          * the fork offset of all the inodes will be
 84          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
 85          * with ATTR2 and then mounted back with ATTR1, keeping the
 86          * di_forkoff's fixed but probably at various positions. Therefore,
 87          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
 88          * of a minimum size available.
 89          */
 90         if (whichfork == XFS_DATA_FORK) {
 91                 maxleafents = MAXEXTNUM;
 92                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
 93         } else {
 94                 maxleafents = MAXAEXTNUM;
 95                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
 96         }
 97         maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
 98         minleafrecs = mp->m_bmap_dmnr[0];
 99         minnoderecs = mp->m_bmap_dmnr[1];
100         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
101         for (level = 1; maxblocks > 1; level++) {
102                 if (maxblocks <= maxrootrecs)
103                         maxblocks = 1;
104                 else
105                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
106         }
107         mp->m_bm_maxlevels[whichfork] = level;
108 }
109 
110 STATIC int                              /* error */
111 xfs_bmbt_lookup_eq(
112         struct xfs_btree_cur    *cur,
113         xfs_fileoff_t           off,
114         xfs_fsblock_t           bno,
115         xfs_filblks_t           len,
116         int                     *stat)  /* success/failure */
117 {
118         cur->bc_rec.b.br_startoff = off;
119         cur->bc_rec.b.br_startblock = bno;
120         cur->bc_rec.b.br_blockcount = len;
121         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
122 }
123 
124 STATIC int                              /* error */
125 xfs_bmbt_lookup_ge(
126         struct xfs_btree_cur    *cur,
127         xfs_fileoff_t           off,
128         xfs_fsblock_t           bno,
129         xfs_filblks_t           len,
130         int                     *stat)  /* success/failure */
131 {
132         cur->bc_rec.b.br_startoff = off;
133         cur->bc_rec.b.br_startblock = bno;
134         cur->bc_rec.b.br_blockcount = len;
135         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
136 }
137 
138 /*
139  * Check if the inode needs to be converted to btree format.
140  */
141 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
142 {
143         return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
144                 XFS_IFORK_NEXTENTS(ip, whichfork) >
145                         XFS_IFORK_MAXEXT(ip, whichfork);
146 }
147 
148 /*
149  * Check if the inode should be converted to extent format.
150  */
151 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
152 {
153         return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
154                 XFS_IFORK_NEXTENTS(ip, whichfork) <=
155                         XFS_IFORK_MAXEXT(ip, whichfork);
156 }
157 
158 /*
159  * Update the record referred to by cur to the value given
160  * by [off, bno, len, state].
161  * This either works (return 0) or gets an EFSCORRUPTED error.
162  */
163 STATIC int
164 xfs_bmbt_update(
165         struct xfs_btree_cur    *cur,
166         xfs_fileoff_t           off,
167         xfs_fsblock_t           bno,
168         xfs_filblks_t           len,
169         xfs_exntst_t            state)
170 {
171         union xfs_btree_rec     rec;
172 
173         xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
174         return xfs_btree_update(cur, &rec);
175 }
176 
177 /*
178  * Compute the worst-case number of indirect blocks that will be used
179  * for ip's delayed extent of length "len".
180  */
181 STATIC xfs_filblks_t
182 xfs_bmap_worst_indlen(
183         xfs_inode_t     *ip,            /* incore inode pointer */
184         xfs_filblks_t   len)            /* delayed extent length */
185 {
186         int             level;          /* btree level number */
187         int             maxrecs;        /* maximum record count at this level */
188         xfs_mount_t     *mp;            /* mount structure */
189         xfs_filblks_t   rval;           /* return value */
190 
191         mp = ip->i_mount;
192         maxrecs = mp->m_bmap_dmxr[0];
193         for (level = 0, rval = 0;
194              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
195              level++) {
196                 len += maxrecs - 1;
197                 do_div(len, maxrecs);
198                 rval += len;
199                 if (len == 1)
200                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
201                                 level - 1;
202                 if (level == 0)
203                         maxrecs = mp->m_bmap_dmxr[1];
204         }
205         return rval;
206 }
207 
208 /*
209  * Calculate the default attribute fork offset for newly created inodes.
210  */
211 uint
212 xfs_default_attroffset(
213         struct xfs_inode        *ip)
214 {
215         struct xfs_mount        *mp = ip->i_mount;
216         uint                    offset;
217 
218         if (mp->m_sb.sb_inodesize == 256) {
219                 offset = XFS_LITINO(mp, ip->i_d.di_version) -
220                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
221         } else {
222                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
223         }
224 
225         ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
226         return offset;
227 }
228 
229 /*
230  * Helper routine to reset inode di_forkoff field when switching
231  * attribute fork from local to extent format - we reset it where
232  * possible to make space available for inline data fork extents.
233  */
234 STATIC void
235 xfs_bmap_forkoff_reset(
236         xfs_mount_t     *mp,
237         xfs_inode_t     *ip,
238         int             whichfork)
239 {
240         if (whichfork == XFS_ATTR_FORK &&
241             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
242             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
243             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
244                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
245 
246                 if (dfl_forkoff > ip->i_d.di_forkoff)
247                         ip->i_d.di_forkoff = dfl_forkoff;
248         }
249 }
250 
251 /*
252  * Debug/sanity checking code
253  */
254 
255 STATIC int
256 xfs_bmap_sanity_check(
257         struct xfs_mount        *mp,
258         struct xfs_buf          *bp,
259         int                     level)
260 {
261         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
262 
263         if (block->bb_magic != cpu_to_be32(XFS_BMAP_CRC_MAGIC) &&
264             block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC))
265                 return 0;
266 
267         if (be16_to_cpu(block->bb_level) != level ||
268             be16_to_cpu(block->bb_numrecs) == 0 ||
269             be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
270                 return 0;
271 
272         return 1;
273 }
274 
275 #ifdef DEBUG
276 STATIC struct xfs_buf *
277 xfs_bmap_get_bp(
278         struct xfs_btree_cur    *cur,
279         xfs_fsblock_t           bno)
280 {
281         struct xfs_log_item_desc *lidp;
282         int                     i;
283 
284         if (!cur)
285                 return NULL;
286 
287         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
288                 if (!cur->bc_bufs[i])
289                         break;
290                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
291                         return cur->bc_bufs[i];
292         }
293 
294         /* Chase down all the log items to see if the bp is there */
295         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
296                 struct xfs_buf_log_item *bip;
297                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
298                 if (bip->bli_item.li_type == XFS_LI_BUF &&
299                     XFS_BUF_ADDR(bip->bli_buf) == bno)
300                         return bip->bli_buf;
301         }
302 
303         return NULL;
304 }
305 
306 STATIC void
307 xfs_check_block(
308         struct xfs_btree_block  *block,
309         xfs_mount_t             *mp,
310         int                     root,
311         short                   sz)
312 {
313         int                     i, j, dmxr;
314         __be64                  *pp, *thispa;   /* pointer to block address */
315         xfs_bmbt_key_t          *prevp, *keyp;
316 
317         ASSERT(be16_to_cpu(block->bb_level) > 0);
318 
319         prevp = NULL;
320         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
321                 dmxr = mp->m_bmap_dmxr[0];
322                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
323 
324                 if (prevp) {
325                         ASSERT(be64_to_cpu(prevp->br_startoff) <
326                                be64_to_cpu(keyp->br_startoff));
327                 }
328                 prevp = keyp;
329 
330                 /*
331                  * Compare the block numbers to see if there are dups.
332                  */
333                 if (root)
334                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
335                 else
336                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
337 
338                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
339                         if (root)
340                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
341                         else
342                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
343                         if (*thispa == *pp) {
344                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
345                                         __func__, j, i,
346                                         (unsigned long long)be64_to_cpu(*thispa));
347                                 panic("%s: ptrs are equal in node\n",
348                                         __func__);
349                         }
350                 }
351         }
352 }
353 
354 /*
355  * Check that the extents for the inode ip are in the right order in all
356  * btree leaves.
357  */
358 
359 STATIC void
360 xfs_bmap_check_leaf_extents(
361         xfs_btree_cur_t         *cur,   /* btree cursor or null */
362         xfs_inode_t             *ip,            /* incore inode pointer */
363         int                     whichfork)      /* data or attr fork */
364 {
365         struct xfs_btree_block  *block; /* current btree block */
366         xfs_fsblock_t           bno;    /* block # of "block" */
367         xfs_buf_t               *bp;    /* buffer for "block" */
368         int                     error;  /* error return value */
369         xfs_extnum_t            i=0, j; /* index into the extents list */
370         xfs_ifork_t             *ifp;   /* fork structure */
371         int                     level;  /* btree level, for checking */
372         xfs_mount_t             *mp;    /* file system mount structure */
373         __be64                  *pp;    /* pointer to block address */
374         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
375         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
376         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
377         int                     bp_release = 0;
378 
379         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
380                 return;
381         }
382 
383         bno = NULLFSBLOCK;
384         mp = ip->i_mount;
385         ifp = XFS_IFORK_PTR(ip, whichfork);
386         block = ifp->if_broot;
387         /*
388          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
389          */
390         level = be16_to_cpu(block->bb_level);
391         ASSERT(level > 0);
392         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
393         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
394         bno = be64_to_cpu(*pp);
395 
396         ASSERT(bno != NULLDFSBNO);
397         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
398         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
399 
400         /*
401          * Go down the tree until leaf level is reached, following the first
402          * pointer (leftmost) at each level.
403          */
404         while (level-- > 0) {
405                 /* See if buf is in cur first */
406                 bp_release = 0;
407                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
408                 if (!bp) {
409                         bp_release = 1;
410                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
411                                                 XFS_BMAP_BTREE_REF,
412                                                 &xfs_bmbt_buf_ops);
413                         if (error)
414                                 goto error_norelse;
415                 }
416                 block = XFS_BUF_TO_BLOCK(bp);
417                 XFS_WANT_CORRUPTED_GOTO(
418                         xfs_bmap_sanity_check(mp, bp, level),
419                         error0);
420                 if (level == 0)
421                         break;
422 
423                 /*
424                  * Check this block for basic sanity (increasing keys and
425                  * no duplicate blocks).
426                  */
427 
428                 xfs_check_block(block, mp, 0, 0);
429                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
430                 bno = be64_to_cpu(*pp);
431                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
432                 if (bp_release) {
433                         bp_release = 0;
434                         xfs_trans_brelse(NULL, bp);
435                 }
436         }
437 
438         /*
439          * Here with bp and block set to the leftmost leaf node in the tree.
440          */
441         i = 0;
442 
443         /*
444          * Loop over all leaf nodes checking that all extents are in the right order.
445          */
446         for (;;) {
447                 xfs_fsblock_t   nextbno;
448                 xfs_extnum_t    num_recs;
449 
450 
451                 num_recs = xfs_btree_get_numrecs(block);
452 
453                 /*
454                  * Read-ahead the next leaf block, if any.
455                  */
456 
457                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
458 
459                 /*
460                  * Check all the extents to make sure they are OK.
461                  * If we had a previous block, the last entry should
462                  * conform with the first entry in this one.
463                  */
464 
465                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
466                 if (i) {
467                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
468                                xfs_bmbt_disk_get_blockcount(&last) <=
469                                xfs_bmbt_disk_get_startoff(ep));
470                 }
471                 for (j = 1; j < num_recs; j++) {
472                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
473                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
474                                xfs_bmbt_disk_get_blockcount(ep) <=
475                                xfs_bmbt_disk_get_startoff(nextp));
476                         ep = nextp;
477                 }
478 
479                 last = *ep;
480                 i += num_recs;
481                 if (bp_release) {
482                         bp_release = 0;
483                         xfs_trans_brelse(NULL, bp);
484                 }
485                 bno = nextbno;
486                 /*
487                  * If we've reached the end, stop.
488                  */
489                 if (bno == NULLFSBLOCK)
490                         break;
491 
492                 bp_release = 0;
493                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
494                 if (!bp) {
495                         bp_release = 1;
496                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
497                                                 XFS_BMAP_BTREE_REF,
498                                                 &xfs_bmbt_buf_ops);
499                         if (error)
500                                 goto error_norelse;
501                 }
502                 block = XFS_BUF_TO_BLOCK(bp);
503         }
504         if (bp_release) {
505                 bp_release = 0;
506                 xfs_trans_brelse(NULL, bp);
507         }
508         return;
509 
510 error0:
511         xfs_warn(mp, "%s: at error0", __func__);
512         if (bp_release)
513                 xfs_trans_brelse(NULL, bp);
514 error_norelse:
515         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
516                 __func__, i);
517         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
518         return;
519 }
520 
521 /*
522  * Add bmap trace insert entries for all the contents of the extent records.
523  */
524 void
525 xfs_bmap_trace_exlist(
526         xfs_inode_t     *ip,            /* incore inode pointer */
527         xfs_extnum_t    cnt,            /* count of entries in the list */
528         int             whichfork,      /* data or attr fork */
529         unsigned long   caller_ip)
530 {
531         xfs_extnum_t    idx;            /* extent record index */
532         xfs_ifork_t     *ifp;           /* inode fork pointer */
533         int             state = 0;
534 
535         if (whichfork == XFS_ATTR_FORK)
536                 state |= BMAP_ATTRFORK;
537 
538         ifp = XFS_IFORK_PTR(ip, whichfork);
539         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
540         for (idx = 0; idx < cnt; idx++)
541                 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
542 }
543 
544 /*
545  * Validate that the bmbt_irecs being returned from bmapi are valid
546  * given the caller's original parameters.  Specifically check the
547  * ranges of the returned irecs to ensure that they only extend beyond
548  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
549  */
550 STATIC void
551 xfs_bmap_validate_ret(
552         xfs_fileoff_t           bno,
553         xfs_filblks_t           len,
554         int                     flags,
555         xfs_bmbt_irec_t         *mval,
556         int                     nmap,
557         int                     ret_nmap)
558 {
559         int                     i;              /* index to map values */
560 
561         ASSERT(ret_nmap <= nmap);
562 
563         for (i = 0; i < ret_nmap; i++) {
564                 ASSERT(mval[i].br_blockcount > 0);
565                 if (!(flags & XFS_BMAPI_ENTIRE)) {
566                         ASSERT(mval[i].br_startoff >= bno);
567                         ASSERT(mval[i].br_blockcount <= len);
568                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
569                                bno + len);
570                 } else {
571                         ASSERT(mval[i].br_startoff < bno + len);
572                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
573                                bno);
574                 }
575                 ASSERT(i == 0 ||
576                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
577                        mval[i].br_startoff);
578                 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
579                        mval[i].br_startblock != HOLESTARTBLOCK);
580                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
581                        mval[i].br_state == XFS_EXT_UNWRITTEN);
582         }
583 }
584 
585 #else
586 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)         do { } while (0)
587 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
588 #endif /* DEBUG */
589 
590 /*
591  * bmap free list manipulation functions
592  */
593 
594 /*
595  * Add the extent to the list of extents to be free at transaction end.
596  * The list is maintained sorted (by block number).
597  */
598 void
599 xfs_bmap_add_free(
600         xfs_fsblock_t           bno,            /* fs block number of extent */
601         xfs_filblks_t           len,            /* length of extent */
602         xfs_bmap_free_t         *flist,         /* list of extents */
603         xfs_mount_t             *mp)            /* mount point structure */
604 {
605         xfs_bmap_free_item_t    *cur;           /* current (next) element */
606         xfs_bmap_free_item_t    *new;           /* new element */
607         xfs_bmap_free_item_t    *prev;          /* previous element */
608 #ifdef DEBUG
609         xfs_agnumber_t          agno;
610         xfs_agblock_t           agbno;
611 
612         ASSERT(bno != NULLFSBLOCK);
613         ASSERT(len > 0);
614         ASSERT(len <= MAXEXTLEN);
615         ASSERT(!isnullstartblock(bno));
616         agno = XFS_FSB_TO_AGNO(mp, bno);
617         agbno = XFS_FSB_TO_AGBNO(mp, bno);
618         ASSERT(agno < mp->m_sb.sb_agcount);
619         ASSERT(agbno < mp->m_sb.sb_agblocks);
620         ASSERT(len < mp->m_sb.sb_agblocks);
621         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
622 #endif
623         ASSERT(xfs_bmap_free_item_zone != NULL);
624         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
625         new->xbfi_startblock = bno;
626         new->xbfi_blockcount = (xfs_extlen_t)len;
627         for (prev = NULL, cur = flist->xbf_first;
628              cur != NULL;
629              prev = cur, cur = cur->xbfi_next) {
630                 if (cur->xbfi_startblock >= bno)
631                         break;
632         }
633         if (prev)
634                 prev->xbfi_next = new;
635         else
636                 flist->xbf_first = new;
637         new->xbfi_next = cur;
638         flist->xbf_count++;
639 }
640 
641 /*
642  * Remove the entry "free" from the free item list.  Prev points to the
643  * previous entry, unless "free" is the head of the list.
644  */
645 void
646 xfs_bmap_del_free(
647         xfs_bmap_free_t         *flist, /* free item list header */
648         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
649         xfs_bmap_free_item_t    *free)  /* list item to be freed */
650 {
651         if (prev)
652                 prev->xbfi_next = free->xbfi_next;
653         else
654                 flist->xbf_first = free->xbfi_next;
655         flist->xbf_count--;
656         kmem_zone_free(xfs_bmap_free_item_zone, free);
657 }
658 
659 /*
660  * Free up any items left in the list.
661  */
662 void
663 xfs_bmap_cancel(
664         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
665 {
666         xfs_bmap_free_item_t    *free;  /* free list item */
667         xfs_bmap_free_item_t    *next;
668 
669         if (flist->xbf_count == 0)
670                 return;
671         ASSERT(flist->xbf_first != NULL);
672         for (free = flist->xbf_first; free; free = next) {
673                 next = free->xbfi_next;
674                 xfs_bmap_del_free(flist, NULL, free);
675         }
676         ASSERT(flist->xbf_count == 0);
677 }
678 
679 /*
680  * Inode fork format manipulation functions
681  */
682 
683 /*
684  * Transform a btree format file with only one leaf node, where the
685  * extents list will fit in the inode, into an extents format file.
686  * Since the file extents are already in-core, all we have to do is
687  * give up the space for the btree root and pitch the leaf block.
688  */
689 STATIC int                              /* error */
690 xfs_bmap_btree_to_extents(
691         xfs_trans_t             *tp,    /* transaction pointer */
692         xfs_inode_t             *ip,    /* incore inode pointer */
693         xfs_btree_cur_t         *cur,   /* btree cursor */
694         int                     *logflagsp, /* inode logging flags */
695         int                     whichfork)  /* data or attr fork */
696 {
697         /* REFERENCED */
698         struct xfs_btree_block  *cblock;/* child btree block */
699         xfs_fsblock_t           cbno;   /* child block number */
700         xfs_buf_t               *cbp;   /* child block's buffer */
701         int                     error;  /* error return value */
702         xfs_ifork_t             *ifp;   /* inode fork data */
703         xfs_mount_t             *mp;    /* mount point structure */
704         __be64                  *pp;    /* ptr to block address */
705         struct xfs_btree_block  *rblock;/* root btree block */
706 
707         mp = ip->i_mount;
708         ifp = XFS_IFORK_PTR(ip, whichfork);
709         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
710         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
711         rblock = ifp->if_broot;
712         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
713         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
714         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
715         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
716         cbno = be64_to_cpu(*pp);
717         *logflagsp = 0;
718 #ifdef DEBUG
719         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
720                 return error;
721 #endif
722         error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
723                                 &xfs_bmbt_buf_ops);
724         if (error)
725                 return error;
726         cblock = XFS_BUF_TO_BLOCK(cbp);
727         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
728                 return error;
729         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
730         ip->i_d.di_nblocks--;
731         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
732         xfs_trans_binval(tp, cbp);
733         if (cur->bc_bufs[0] == cbp)
734                 cur->bc_bufs[0] = NULL;
735         xfs_iroot_realloc(ip, -1, whichfork);
736         ASSERT(ifp->if_broot == NULL);
737         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
738         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
739         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
740         return 0;
741 }
742 
743 /*
744  * Convert an extents-format file into a btree-format file.
745  * The new file will have a root block (in the inode) and a single child block.
746  */
747 STATIC int                                      /* error */
748 xfs_bmap_extents_to_btree(
749         xfs_trans_t             *tp,            /* transaction pointer */
750         xfs_inode_t             *ip,            /* incore inode pointer */
751         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
752         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
753         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
754         int                     wasdel,         /* converting a delayed alloc */
755         int                     *logflagsp,     /* inode logging flags */
756         int                     whichfork)      /* data or attr fork */
757 {
758         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
759         xfs_buf_t               *abp;           /* buffer for ablock */
760         xfs_alloc_arg_t         args;           /* allocation arguments */
761         xfs_bmbt_rec_t          *arp;           /* child record pointer */
762         struct xfs_btree_block  *block;         /* btree root block */
763         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
764         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
765         int                     error;          /* error return value */
766         xfs_extnum_t            i, cnt;         /* extent record index */
767         xfs_ifork_t             *ifp;           /* inode fork pointer */
768         xfs_bmbt_key_t          *kp;            /* root block key pointer */
769         xfs_mount_t             *mp;            /* mount structure */
770         xfs_extnum_t            nextents;       /* number of file extents */
771         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
772 
773         mp = ip->i_mount;
774         ifp = XFS_IFORK_PTR(ip, whichfork);
775         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
776 
777         /*
778          * Make space in the inode incore.
779          */
780         xfs_iroot_realloc(ip, 1, whichfork);
781         ifp->if_flags |= XFS_IFBROOT;
782 
783         /*
784          * Fill in the root.
785          */
786         block = ifp->if_broot;
787         if (xfs_sb_version_hascrc(&mp->m_sb))
788                 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
789                                  XFS_BMAP_CRC_MAGIC, 1, 1, ip->i_ino,
790                                  XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
791         else
792                 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
793                                  XFS_BMAP_MAGIC, 1, 1, ip->i_ino,
794                                  XFS_BTREE_LONG_PTRS);
795 
796         /*
797          * Need a cursor.  Can't allocate until bb_level is filled in.
798          */
799         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
800         cur->bc_private.b.firstblock = *firstblock;
801         cur->bc_private.b.flist = flist;
802         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
803         /*
804          * Convert to a btree with two levels, one record in root.
805          */
806         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
807         memset(&args, 0, sizeof(args));
808         args.tp = tp;
809         args.mp = mp;
810         args.firstblock = *firstblock;
811         if (*firstblock == NULLFSBLOCK) {
812                 args.type = XFS_ALLOCTYPE_START_BNO;
813                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
814         } else if (flist->xbf_low) {
815                 args.type = XFS_ALLOCTYPE_START_BNO;
816                 args.fsbno = *firstblock;
817         } else {
818                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
819                 args.fsbno = *firstblock;
820         }
821         args.minlen = args.maxlen = args.prod = 1;
822         args.wasdel = wasdel;
823         *logflagsp = 0;
824         if ((error = xfs_alloc_vextent(&args))) {
825                 xfs_iroot_realloc(ip, -1, whichfork);
826                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
827                 return error;
828         }
829         /*
830          * Allocation can't fail, the space was reserved.
831          */
832         ASSERT(args.fsbno != NULLFSBLOCK);
833         ASSERT(*firstblock == NULLFSBLOCK ||
834                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
835                (flist->xbf_low &&
836                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
837         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
838         cur->bc_private.b.allocated++;
839         ip->i_d.di_nblocks++;
840         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
841         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
842         /*
843          * Fill in the child block.
844          */
845         abp->b_ops = &xfs_bmbt_buf_ops;
846         ablock = XFS_BUF_TO_BLOCK(abp);
847         if (xfs_sb_version_hascrc(&mp->m_sb))
848                 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
849                                 XFS_BMAP_CRC_MAGIC, 0, 0, ip->i_ino,
850                                 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
851         else
852                 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
853                                 XFS_BMAP_MAGIC, 0, 0, ip->i_ino,
854                                 XFS_BTREE_LONG_PTRS);
855 
856         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
857         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
858         for (cnt = i = 0; i < nextents; i++) {
859                 ep = xfs_iext_get_ext(ifp, i);
860                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
861                         arp->l0 = cpu_to_be64(ep->l0);
862                         arp->l1 = cpu_to_be64(ep->l1);
863                         arp++; cnt++;
864                 }
865         }
866         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
867         xfs_btree_set_numrecs(ablock, cnt);
868 
869         /*
870          * Fill in the root key and pointer.
871          */
872         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
873         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
874         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
875         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
876                                                 be16_to_cpu(block->bb_level)));
877         *pp = cpu_to_be64(args.fsbno);
878 
879         /*
880          * Do all this logging at the end so that
881          * the root is at the right level.
882          */
883         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
884         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
885         ASSERT(*curp == NULL);
886         *curp = cur;
887         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
888         return 0;
889 }
890 
891 /*
892  * Convert a local file to an extents file.
893  * This code is out of bounds for data forks of regular files,
894  * since the file data needs to get logged so things will stay consistent.
895  * (The bmap-level manipulations are ok, though).
896  */
897 void
898 xfs_bmap_local_to_extents_empty(
899         struct xfs_inode        *ip,
900         int                     whichfork)
901 {
902         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
903 
904         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
905         ASSERT(ifp->if_bytes == 0);
906         ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
907 
908         xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
909         ifp->if_flags &= ~XFS_IFINLINE;
910         ifp->if_flags |= XFS_IFEXTENTS;
911         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
912 }
913 
914 
915 STATIC int                              /* error */
916 xfs_bmap_local_to_extents(
917         xfs_trans_t     *tp,            /* transaction pointer */
918         xfs_inode_t     *ip,            /* incore inode pointer */
919         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
920         xfs_extlen_t    total,          /* total blocks needed by transaction */
921         int             *logflagsp,     /* inode logging flags */
922         int             whichfork,
923         void            (*init_fn)(struct xfs_trans *tp,
924                                    struct xfs_buf *bp,
925                                    struct xfs_inode *ip,
926                                    struct xfs_ifork *ifp))
927 {
928         int             error = 0;
929         int             flags;          /* logging flags returned */
930         xfs_ifork_t     *ifp;           /* inode fork pointer */
931         xfs_alloc_arg_t args;           /* allocation arguments */
932         xfs_buf_t       *bp;            /* buffer for extent block */
933         xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
934 
935         /*
936          * We don't want to deal with the case of keeping inode data inline yet.
937          * So sending the data fork of a regular inode is invalid.
938          */
939         ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
940         ifp = XFS_IFORK_PTR(ip, whichfork);
941         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
942 
943         if (!ifp->if_bytes) {
944                 xfs_bmap_local_to_extents_empty(ip, whichfork);
945                 flags = XFS_ILOG_CORE;
946                 goto done;
947         }
948 
949         flags = 0;
950         error = 0;
951         ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) ==
952                                                                 XFS_IFINLINE);
953         memset(&args, 0, sizeof(args));
954         args.tp = tp;
955         args.mp = ip->i_mount;
956         args.firstblock = *firstblock;
957         /*
958          * Allocate a block.  We know we need only one, since the
959          * file currently fits in an inode.
960          */
961         if (*firstblock == NULLFSBLOCK) {
962                 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
963                 args.type = XFS_ALLOCTYPE_START_BNO;
964         } else {
965                 args.fsbno = *firstblock;
966                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
967         }
968         args.total = total;
969         args.minlen = args.maxlen = args.prod = 1;
970         error = xfs_alloc_vextent(&args);
971         if (error)
972                 goto done;
973 
974         /* Can't fail, the space was reserved. */
975         ASSERT(args.fsbno != NULLFSBLOCK);
976         ASSERT(args.len == 1);
977         *firstblock = args.fsbno;
978         bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
979 
980         /* initialise the block and copy the data */
981         init_fn(tp, bp, ip, ifp);
982 
983         /* account for the change in fork size and log everything */
984         xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
985         xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
986         xfs_bmap_local_to_extents_empty(ip, whichfork);
987         flags |= XFS_ILOG_CORE;
988 
989         xfs_iext_add(ifp, 0, 1);
990         ep = xfs_iext_get_ext(ifp, 0);
991         xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
992         trace_xfs_bmap_post_update(ip, 0,
993                         whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
994                         _THIS_IP_);
995         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
996         ip->i_d.di_nblocks = 1;
997         xfs_trans_mod_dquot_byino(tp, ip,
998                 XFS_TRANS_DQ_BCOUNT, 1L);
999         flags |= xfs_ilog_fext(whichfork);
1000 
1001 done:
1002         *logflagsp = flags;
1003         return error;
1004 }
1005 
1006 /*
1007  * Called from xfs_bmap_add_attrfork to handle btree format files.
1008  */
1009 STATIC int                                      /* error */
1010 xfs_bmap_add_attrfork_btree(
1011         xfs_trans_t             *tp,            /* transaction pointer */
1012         xfs_inode_t             *ip,            /* incore inode pointer */
1013         xfs_fsblock_t           *firstblock,    /* first block allocated */
1014         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
1015         int                     *flags)         /* inode logging flags */
1016 {
1017         xfs_btree_cur_t         *cur;           /* btree cursor */
1018         int                     error;          /* error return value */
1019         xfs_mount_t             *mp;            /* file system mount struct */
1020         int                     stat;           /* newroot status */
1021 
1022         mp = ip->i_mount;
1023         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
1024                 *flags |= XFS_ILOG_DBROOT;
1025         else {
1026                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
1027                 cur->bc_private.b.flist = flist;
1028                 cur->bc_private.b.firstblock = *firstblock;
1029                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
1030                         goto error0;
1031                 /* must be at least one entry */
1032                 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
1033                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
1034                         goto error0;
1035                 if (stat == 0) {
1036                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1037                         return XFS_ERROR(ENOSPC);
1038                 }
1039                 *firstblock = cur->bc_private.b.firstblock;
1040                 cur->bc_private.b.allocated = 0;
1041                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1042         }
1043         return 0;
1044 error0:
1045         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1046         return error;
1047 }
1048 
1049 /*
1050  * Called from xfs_bmap_add_attrfork to handle extents format files.
1051  */
1052 STATIC int                                      /* error */
1053 xfs_bmap_add_attrfork_extents(
1054         xfs_trans_t             *tp,            /* transaction pointer */
1055         xfs_inode_t             *ip,            /* incore inode pointer */
1056         xfs_fsblock_t           *firstblock,    /* first block allocated */
1057         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
1058         int                     *flags)         /* inode logging flags */
1059 {
1060         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
1061         int                     error;          /* error return value */
1062 
1063         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1064                 return 0;
1065         cur = NULL;
1066         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
1067                 flags, XFS_DATA_FORK);
1068         if (cur) {
1069                 cur->bc_private.b.allocated = 0;
1070                 xfs_btree_del_cursor(cur,
1071                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1072         }
1073         return error;
1074 }
1075 
1076 /*
1077  * Called from xfs_bmap_add_attrfork to handle local format files. Each
1078  * different data fork content type needs a different callout to do the
1079  * conversion. Some are basic and only require special block initialisation
1080  * callouts for the data formating, others (directories) are so specialised they
1081  * handle everything themselves.
1082  *
1083  * XXX (dgc): investigate whether directory conversion can use the generic
1084  * formatting callout. It should be possible - it's just a very complex
1085  * formatter.
1086  */
1087 STATIC int                                      /* error */
1088 xfs_bmap_add_attrfork_local(
1089         xfs_trans_t             *tp,            /* transaction pointer */
1090         xfs_inode_t             *ip,            /* incore inode pointer */
1091         xfs_fsblock_t           *firstblock,    /* first block allocated */
1092         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
1093         int                     *flags)         /* inode logging flags */
1094 {
1095         xfs_da_args_t           dargs;          /* args for dir/attr code */
1096 
1097         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1098                 return 0;
1099 
1100         if (S_ISDIR(ip->i_d.di_mode)) {
1101                 memset(&dargs, 0, sizeof(dargs));
1102                 dargs.dp = ip;
1103                 dargs.firstblock = firstblock;
1104                 dargs.flist = flist;
1105                 dargs.total = ip->i_mount->m_dirblkfsbs;
1106                 dargs.whichfork = XFS_DATA_FORK;
1107                 dargs.trans = tp;
1108                 return xfs_dir2_sf_to_block(&dargs);
1109         }
1110 
1111         if (S_ISLNK(ip->i_d.di_mode))
1112                 return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1113                                                  flags, XFS_DATA_FORK,
1114                                                  xfs_symlink_local_to_remote);
1115 
1116         /* should only be called for types that support local format data */
1117         ASSERT(0);
1118         return EFSCORRUPTED;
1119 }
1120 
1121 /*
1122  * Convert inode from non-attributed to attributed.
1123  * Must not be in a transaction, ip must not be locked.
1124  */
1125 int                                             /* error code */
1126 xfs_bmap_add_attrfork(
1127         xfs_inode_t             *ip,            /* incore inode pointer */
1128         int                     size,           /* space new attribute needs */
1129         int                     rsvd)           /* xact may use reserved blks */
1130 {
1131         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
1132         xfs_bmap_free_t         flist;          /* freed extent records */
1133         xfs_mount_t             *mp;            /* mount structure */
1134         xfs_trans_t             *tp;            /* transaction pointer */
1135         int                     blks;           /* space reservation */
1136         int                     version = 1;    /* superblock attr version */
1137         int                     committed;      /* xaction was committed */
1138         int                     logflags;       /* logging flags */
1139         int                     error;          /* error return value */
1140         int                     cancel_flags = 0;
1141 
1142         ASSERT(XFS_IFORK_Q(ip) == 0);
1143 
1144         mp = ip->i_mount;
1145         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1146         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
1147         blks = XFS_ADDAFORK_SPACE_RES(mp);
1148         if (rsvd)
1149                 tp->t_flags |= XFS_TRANS_RESERVE;
1150         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_addafork, blks, 0);
1151         if (error) {
1152                 xfs_trans_cancel(tp, 0);
1153                 return error;
1154         }
1155         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1156         xfs_ilock(ip, XFS_ILOCK_EXCL);
1157         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1158                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1159                         XFS_QMOPT_RES_REGBLKS);
1160         if (error)
1161                 goto trans_cancel;
1162         cancel_flags |= XFS_TRANS_ABORT;
1163         if (XFS_IFORK_Q(ip))
1164                 goto trans_cancel;
1165         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1166                 /*
1167                  * For inodes coming from pre-6.2 filesystems.
1168                  */
1169                 ASSERT(ip->i_d.di_aformat == 0);
1170                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1171         }
1172         ASSERT(ip->i_d.di_anextents == 0);
1173 
1174         xfs_trans_ijoin(tp, ip, 0);
1175         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1176 
1177         switch (ip->i_d.di_format) {
1178         case XFS_DINODE_FMT_DEV:
1179                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1180                 break;
1181         case XFS_DINODE_FMT_UUID:
1182                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1183                 break;
1184         case XFS_DINODE_FMT_LOCAL:
1185         case XFS_DINODE_FMT_EXTENTS:
1186         case XFS_DINODE_FMT_BTREE:
1187                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1188                 if (!ip->i_d.di_forkoff)
1189                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1190                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
1191                         version = 2;
1192                 break;
1193         default:
1194                 ASSERT(0);
1195                 error = XFS_ERROR(EINVAL);
1196                 goto trans_cancel;
1197         }
1198 
1199         ASSERT(ip->i_afp == NULL);
1200         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1201         ip->i_afp->if_flags = XFS_IFEXTENTS;
1202         logflags = 0;
1203         xfs_bmap_init(&flist, &firstblock);
1204         switch (ip->i_d.di_format) {
1205         case XFS_DINODE_FMT_LOCAL:
1206                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
1207                         &logflags);
1208                 break;
1209         case XFS_DINODE_FMT_EXTENTS:
1210                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1211                         &flist, &logflags);
1212                 break;
1213         case XFS_DINODE_FMT_BTREE:
1214                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
1215                         &logflags);
1216                 break;
1217         default:
1218                 error = 0;
1219                 break;
1220         }
1221         if (logflags)
1222                 xfs_trans_log_inode(tp, ip, logflags);
1223         if (error)
1224                 goto bmap_cancel;
1225         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1226            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1227                 __int64_t sbfields = 0;
1228 
1229                 spin_lock(&mp->m_sb_lock);
1230                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1231                         xfs_sb_version_addattr(&mp->m_sb);
1232                         sbfields |= XFS_SB_VERSIONNUM;
1233                 }
1234                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1235                         xfs_sb_version_addattr2(&mp->m_sb);
1236                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
1237                 }
1238                 if (sbfields) {
1239                         spin_unlock(&mp->m_sb_lock);
1240                         xfs_mod_sb(tp, sbfields);
1241                 } else
1242                         spin_unlock(&mp->m_sb_lock);
1243         }
1244 
1245         error = xfs_bmap_finish(&tp, &flist, &committed);
1246         if (error)
1247                 goto bmap_cancel;
1248         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1249         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1250         return error;
1251 
1252 bmap_cancel:
1253         xfs_bmap_cancel(&flist);
1254 trans_cancel:
1255         xfs_trans_cancel(tp, cancel_flags);
1256         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1257         return error;
1258 }
1259 
1260 /*
1261  * Internal and external extent tree search functions.
1262  */
1263 
1264 /*
1265  * Read in the extents to if_extents.
1266  * All inode fields are set up by caller, we just traverse the btree
1267  * and copy the records in. If the file system cannot contain unwritten
1268  * extents, the records are checked for no "state" flags.
1269  */
1270 int                                     /* error */
1271 xfs_bmap_read_extents(
1272         xfs_trans_t             *tp,    /* transaction pointer */
1273         xfs_inode_t             *ip,    /* incore inode */
1274         int                     whichfork) /* data or attr fork */
1275 {
1276         struct xfs_btree_block  *block; /* current btree block */
1277         xfs_fsblock_t           bno;    /* block # of "block" */
1278         xfs_buf_t               *bp;    /* buffer for "block" */
1279         int                     error;  /* error return value */
1280         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
1281         xfs_extnum_t            i, j;   /* index into the extents list */
1282         xfs_ifork_t             *ifp;   /* fork structure */
1283         int                     level;  /* btree level, for checking */
1284         xfs_mount_t             *mp;    /* file system mount structure */
1285         __be64                  *pp;    /* pointer to block address */
1286         /* REFERENCED */
1287         xfs_extnum_t            room;   /* number of entries there's room for */
1288 
1289         bno = NULLFSBLOCK;
1290         mp = ip->i_mount;
1291         ifp = XFS_IFORK_PTR(ip, whichfork);
1292         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
1293                                         XFS_EXTFMT_INODE(ip);
1294         block = ifp->if_broot;
1295         /*
1296          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1297          */
1298         level = be16_to_cpu(block->bb_level);
1299         ASSERT(level > 0);
1300         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1301         bno = be64_to_cpu(*pp);
1302         ASSERT(bno != NULLDFSBNO);
1303         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
1304         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
1305         /*
1306          * Go down the tree until leaf level is reached, following the first
1307          * pointer (leftmost) at each level.
1308          */
1309         while (level-- > 0) {
1310                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1311                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1312                 if (error)
1313                         return error;
1314                 block = XFS_BUF_TO_BLOCK(bp);
1315                 XFS_WANT_CORRUPTED_GOTO(
1316                         xfs_bmap_sanity_check(mp, bp, level),
1317                         error0);
1318                 if (level == 0)
1319                         break;
1320                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1321                 bno = be64_to_cpu(*pp);
1322                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
1323                 xfs_trans_brelse(tp, bp);
1324         }
1325         /*
1326          * Here with bp and block set to the leftmost leaf node in the tree.
1327          */
1328         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1329         i = 0;
1330         /*
1331          * Loop over all leaf nodes.  Copy information to the extent records.
1332          */
1333         for (;;) {
1334                 xfs_bmbt_rec_t  *frp;
1335                 xfs_fsblock_t   nextbno;
1336                 xfs_extnum_t    num_recs;
1337                 xfs_extnum_t    start;
1338 
1339                 num_recs = xfs_btree_get_numrecs(block);
1340                 if (unlikely(i + num_recs > room)) {
1341                         ASSERT(i + num_recs <= room);
1342                         xfs_warn(ip->i_mount,
1343                                 "corrupt dinode %Lu, (btree extents).",
1344                                 (unsigned long long) ip->i_ino);
1345                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1346                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
1347                         goto error0;
1348                 }
1349                 XFS_WANT_CORRUPTED_GOTO(
1350                         xfs_bmap_sanity_check(mp, bp, 0),
1351                         error0);
1352                 /*
1353                  * Read-ahead the next leaf block, if any.
1354                  */
1355                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1356                 if (nextbno != NULLFSBLOCK)
1357                         xfs_btree_reada_bufl(mp, nextbno, 1,
1358                                              &xfs_bmbt_buf_ops);
1359                 /*
1360                  * Copy records into the extent records.
1361                  */
1362                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1363                 start = i;
1364                 for (j = 0; j < num_recs; j++, i++, frp++) {
1365                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1366                         trp->l0 = be64_to_cpu(frp->l0);
1367                         trp->l1 = be64_to_cpu(frp->l1);
1368                 }
1369                 if (exntf == XFS_EXTFMT_NOSTATE) {
1370                         /*
1371                          * Check all attribute bmap btree records and
1372                          * any "older" data bmap btree records for a
1373                          * set bit in the "extent flag" position.
1374                          */
1375                         if (unlikely(xfs_check_nostate_extents(ifp,
1376                                         start, num_recs))) {
1377                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1378                                                  XFS_ERRLEVEL_LOW,
1379                                                  ip->i_mount);
1380                                 goto error0;
1381                         }
1382                 }
1383                 xfs_trans_brelse(tp, bp);
1384                 bno = nextbno;
1385                 /*
1386                  * If we've reached the end, stop.
1387                  */
1388                 if (bno == NULLFSBLOCK)
1389                         break;
1390                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1391                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1392                 if (error)
1393                         return error;
1394                 block = XFS_BUF_TO_BLOCK(bp);
1395         }
1396         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
1397         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
1398         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1399         return 0;
1400 error0:
1401         xfs_trans_brelse(tp, bp);
1402         return XFS_ERROR(EFSCORRUPTED);
1403 }
1404 
1405 
1406 /*
1407  * Search the extent records for the entry containing block bno.
1408  * If bno lies in a hole, point to the next entry.  If bno lies
1409  * past eof, *eofp will be set, and *prevp will contain the last
1410  * entry (null if none).  Else, *lastxp will be set to the index
1411  * of the found entry; *gotp will contain the entry.
1412  */
1413 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
1414 xfs_bmap_search_multi_extents(
1415         xfs_ifork_t     *ifp,           /* inode fork pointer */
1416         xfs_fileoff_t   bno,            /* block number searched for */
1417         int             *eofp,          /* out: end of file found */
1418         xfs_extnum_t    *lastxp,        /* out: last extent index */
1419         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1420         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1421 {
1422         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
1423         xfs_extnum_t    lastx;          /* last extent index */
1424 
1425         /*
1426          * Initialize the extent entry structure to catch access to
1427          * uninitialized br_startblock field.
1428          */
1429         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
1430         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
1431         gotp->br_state = XFS_EXT_INVALID;
1432 #if XFS_BIG_BLKNOS
1433         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
1434 #else
1435         gotp->br_startblock = 0xffffa5a5;
1436 #endif
1437         prevp->br_startoff = NULLFILEOFF;
1438 
1439         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
1440         if (lastx > 0) {
1441                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
1442         }
1443         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1444                 xfs_bmbt_get_all(ep, gotp);
1445                 *eofp = 0;
1446         } else {
1447                 if (lastx > 0) {
1448                         *gotp = *prevp;
1449                 }
1450                 *eofp = 1;
1451                 ep = NULL;
1452         }
1453         *lastxp = lastx;
1454         return ep;
1455 }
1456 
1457 /*
1458  * Search the extents list for the inode, for the extent containing bno.
1459  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
1460  * *eofp will be set, and *prevp will contain the last entry (null if none).
1461  * Else, *lastxp will be set to the index of the found
1462  * entry; *gotp will contain the entry.
1463  */
1464 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
1465 xfs_bmap_search_extents(
1466         xfs_inode_t     *ip,            /* incore inode pointer */
1467         xfs_fileoff_t   bno,            /* block number searched for */
1468         int             fork,           /* data or attr fork */
1469         int             *eofp,          /* out: end of file found */
1470         xfs_extnum_t    *lastxp,        /* out: last extent index */
1471         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1472         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1473 {
1474         xfs_ifork_t     *ifp;           /* inode fork pointer */
1475         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
1476 
1477         XFS_STATS_INC(xs_look_exlist);
1478         ifp = XFS_IFORK_PTR(ip, fork);
1479 
1480         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
1481 
1482         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
1483                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
1484                 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
1485                                 "Access to block zero in inode %llu "
1486                                 "start_block: %llx start_off: %llx "
1487                                 "blkcnt: %llx extent-state: %x lastx: %x",
1488                         (unsigned long long)ip->i_ino,
1489                         (unsigned long long)gotp->br_startblock,
1490                         (unsigned long long)gotp->br_startoff,
1491                         (unsigned long long)gotp->br_blockcount,
1492                         gotp->br_state, *lastxp);
1493                 *lastxp = NULLEXTNUM;
1494                 *eofp = 1;
1495                 return NULL;
1496         }
1497         return ep;
1498 }
1499 
1500 /*
1501  * Returns the file-relative block number of the first unused block(s)
1502  * in the file with at least "len" logically contiguous blocks free.
1503  * This is the lowest-address hole if the file has holes, else the first block
1504  * past the end of file.
1505  * Return 0 if the file is currently local (in-inode).
1506  */
1507 int                                             /* error */
1508 xfs_bmap_first_unused(
1509         xfs_trans_t     *tp,                    /* transaction pointer */
1510         xfs_inode_t     *ip,                    /* incore inode */
1511         xfs_extlen_t    len,                    /* size of hole to find */
1512         xfs_fileoff_t   *first_unused,          /* unused block */
1513         int             whichfork)              /* data or attr fork */
1514 {
1515         int             error;                  /* error return value */
1516         int             idx;                    /* extent record index */
1517         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1518         xfs_fileoff_t   lastaddr;               /* last block number seen */
1519         xfs_fileoff_t   lowest;                 /* lowest useful block */
1520         xfs_fileoff_t   max;                    /* starting useful block */
1521         xfs_fileoff_t   off;                    /* offset for this block */
1522         xfs_extnum_t    nextents;               /* number of extent entries */
1523 
1524         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1525                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1526                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1527         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1528                 *first_unused = 0;
1529                 return 0;
1530         }
1531         ifp = XFS_IFORK_PTR(ip, whichfork);
1532         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1533             (error = xfs_iread_extents(tp, ip, whichfork)))
1534                 return error;
1535         lowest = *first_unused;
1536         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1537         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1538                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
1539                 off = xfs_bmbt_get_startoff(ep);
1540                 /*
1541                  * See if the hole before this extent will work.
1542                  */
1543                 if (off >= lowest + len && off - max >= len) {
1544                         *first_unused = max;
1545                         return 0;
1546                 }
1547                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
1548                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
1549         }
1550         *first_unused = max;
1551         return 0;
1552 }
1553 
1554 /*
1555  * Returns the file-relative block number of the last block - 1 before
1556  * last_block (input value) in the file.
1557  * This is not based on i_size, it is based on the extent records.
1558  * Returns 0 for local files, as they do not have extent records.
1559  */
1560 int                                             /* error */
1561 xfs_bmap_last_before(
1562         xfs_trans_t     *tp,                    /* transaction pointer */
1563         xfs_inode_t     *ip,                    /* incore inode */
1564         xfs_fileoff_t   *last_block,            /* last block */
1565         int             whichfork)              /* data or attr fork */
1566 {
1567         xfs_fileoff_t   bno;                    /* input file offset */
1568         int             eof;                    /* hit end of file */
1569         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
1570         int             error;                  /* error return value */
1571         xfs_bmbt_irec_t got;                    /* current extent value */
1572         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1573         xfs_extnum_t    lastx;                  /* last extent used */
1574         xfs_bmbt_irec_t prev;                   /* previous extent value */
1575 
1576         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1577             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
1578             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
1579                return XFS_ERROR(EIO);
1580         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1581                 *last_block = 0;
1582                 return 0;
1583         }
1584         ifp = XFS_IFORK_PTR(ip, whichfork);
1585         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1586             (error = xfs_iread_extents(tp, ip, whichfork)))
1587                 return error;
1588         bno = *last_block - 1;
1589         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
1590                 &prev);
1591         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
1592                 if (prev.br_startoff == NULLFILEOFF)
1593                         *last_block = 0;
1594                 else
1595                         *last_block = prev.br_startoff + prev.br_blockcount;
1596         }
1597         /*
1598          * Otherwise *last_block is already the right answer.
1599          */
1600         return 0;
1601 }
1602 
1603 int
1604 xfs_bmap_last_extent(
1605         struct xfs_trans        *tp,
1606         struct xfs_inode        *ip,
1607         int                     whichfork,
1608         struct xfs_bmbt_irec    *rec,
1609         int                     *is_empty)
1610 {
1611         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
1612         int                     error;
1613         int                     nextents;
1614 
1615         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1616                 error = xfs_iread_extents(tp, ip, whichfork);
1617                 if (error)
1618                         return error;
1619         }
1620 
1621         nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
1622         if (nextents == 0) {
1623                 *is_empty = 1;
1624                 return 0;
1625         }
1626 
1627         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1628         *is_empty = 0;
1629         return 0;
1630 }
1631 
1632 /*
1633  * Check the last inode extent to determine whether this allocation will result
1634  * in blocks being allocated at the end of the file. When we allocate new data
1635  * blocks at the end of the file which do not start at the previous data block,
1636  * we will try to align the new blocks at stripe unit boundaries.
1637  *
1638  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1639  * at, or past the EOF.
1640  */
1641 STATIC int
1642 xfs_bmap_isaeof(
1643         struct xfs_bmalloca     *bma,
1644         int                     whichfork)
1645 {
1646         struct xfs_bmbt_irec    rec;
1647         int                     is_empty;
1648         int                     error;
1649 
1650         bma->aeof = 0;
1651         error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1652                                      &is_empty);
1653         if (error)
1654                 return error;
1655 
1656         if (is_empty) {
1657                 bma->aeof = 1;
1658                 return 0;
1659         }
1660 
1661         /*
1662          * Check if we are allocation or past the last extent, or at least into
1663          * the last delayed allocated extent.
1664          */
1665         bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1666                 (bma->offset >= rec.br_startoff &&
1667                  isnullstartblock(rec.br_startblock));
1668         return 0;
1669 }
1670 
1671 /*
1672  * Returns the file-relative block number of the first block past eof in
1673  * the file.  This is not based on i_size, it is based on the extent records.
1674  * Returns 0 for local files, as they do not have extent records.
1675  */
1676 int
1677 xfs_bmap_last_offset(
1678         struct xfs_trans        *tp,
1679         struct xfs_inode        *ip,
1680         xfs_fileoff_t           *last_block,
1681         int                     whichfork)
1682 {
1683         struct xfs_bmbt_irec    rec;
1684         int                     is_empty;
1685         int                     error;
1686 
1687         *last_block = 0;
1688 
1689         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1690                 return 0;
1691 
1692         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1693             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1694                return XFS_ERROR(EIO);
1695 
1696         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1697         if (error || is_empty)
1698                 return error;
1699 
1700         *last_block = rec.br_startoff + rec.br_blockcount;
1701         return 0;
1702 }
1703 
1704 /*
1705  * Returns whether the selected fork of the inode has exactly one
1706  * block or not.  For the data fork we check this matches di_size,
1707  * implying the file's range is 0..bsize-1.
1708  */
1709 int                                     /* 1=>1 block, 0=>otherwise */
1710 xfs_bmap_one_block(
1711         xfs_inode_t     *ip,            /* incore inode */
1712         int             whichfork)      /* data or attr fork */
1713 {
1714         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
1715         xfs_ifork_t     *ifp;           /* inode fork pointer */
1716         int             rval;           /* return value */
1717         xfs_bmbt_irec_t s;              /* internal version of extent */
1718 
1719 #ifndef DEBUG
1720         if (whichfork == XFS_DATA_FORK)
1721                 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1722 #endif  /* !DEBUG */
1723         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1724                 return 0;
1725         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1726                 return 0;
1727         ifp = XFS_IFORK_PTR(ip, whichfork);
1728         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1729         ep = xfs_iext_get_ext(ifp, 0);
1730         xfs_bmbt_get_all(ep, &s);
1731         rval = s.br_startoff == 0 && s.br_blockcount == 1;
1732         if (rval && whichfork == XFS_DATA_FORK)
1733                 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1734         return rval;
1735 }
1736 
1737 /*
1738  * Extent tree manipulation functions used during allocation.
1739  */
1740 
1741 /*
1742  * Convert a delayed allocation to a real allocation.
1743  */
1744 STATIC int                              /* error */
1745 xfs_bmap_add_extent_delay_real(
1746         struct xfs_bmalloca     *bma)
1747 {
1748         struct xfs_bmbt_irec    *new = &bma->got;
1749         int                     diff;   /* temp value */
1750         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1751         int                     error;  /* error return value */
1752         int                     i;      /* temp state */
1753         xfs_ifork_t             *ifp;   /* inode fork pointer */
1754         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1755         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1756                                         /* left is 0, right is 1, prev is 2 */
1757         int                     rval=0; /* return value (logging flags) */
1758         int                     state = 0;/* state bits, accessed thru macros */
1759         xfs_filblks_t           da_new; /* new count del alloc blocks used */
1760         xfs_filblks_t           da_old; /* old count del alloc blocks used */
1761         xfs_filblks_t           temp=0; /* value for da_new calculations */
1762         xfs_filblks_t           temp2=0;/* value for da_new calculations */
1763         int                     tmp_rval;       /* partial logging flags */
1764 
1765         ifp = XFS_IFORK_PTR(bma->ip, XFS_DATA_FORK);
1766 
1767         ASSERT(bma->idx >= 0);
1768         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1769         ASSERT(!isnullstartblock(new->br_startblock));
1770         ASSERT(!bma->cur ||
1771                (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1772 
1773         XFS_STATS_INC(xs_add_exlist);
1774 
1775 #define LEFT            r[0]
1776 #define RIGHT           r[1]
1777 #define PREV            r[2]
1778 
1779         /*
1780          * Set up a bunch of variables to make the tests simpler.
1781          */
1782         ep = xfs_iext_get_ext(ifp, bma->idx);
1783         xfs_bmbt_get_all(ep, &PREV);
1784         new_endoff = new->br_startoff + new->br_blockcount;
1785         ASSERT(PREV.br_startoff <= new->br_startoff);
1786         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1787 
1788         da_old = startblockval(PREV.br_startblock);
1789         da_new = 0;
1790 
1791         /*
1792          * Set flags determining what part of the previous delayed allocation
1793          * extent is being replaced by a real allocation.
1794          */
1795         if (PREV.br_startoff == new->br_startoff)
1796                 state |= BMAP_LEFT_FILLING;
1797         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1798                 state |= BMAP_RIGHT_FILLING;
1799 
1800         /*
1801          * Check and set flags if this segment has a left neighbor.
1802          * Don't set contiguous if the combined extent would be too large.
1803          */
1804         if (bma->idx > 0) {
1805                 state |= BMAP_LEFT_VALID;
1806                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
1807 
1808                 if (isnullstartblock(LEFT.br_startblock))
1809                         state |= BMAP_LEFT_DELAY;
1810         }
1811 
1812         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1813             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1814             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1815             LEFT.br_state == new->br_state &&
1816             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1817                 state |= BMAP_LEFT_CONTIG;
1818 
1819         /*
1820          * Check and set flags if this segment has a right neighbor.
1821          * Don't set contiguous if the combined extent would be too large.
1822          * Also check for all-three-contiguous being too large.
1823          */
1824         if (bma->idx < bma->ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1825                 state |= BMAP_RIGHT_VALID;
1826                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
1827 
1828                 if (isnullstartblock(RIGHT.br_startblock))
1829                         state |= BMAP_RIGHT_DELAY;
1830         }
1831 
1832         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1833             new_endoff == RIGHT.br_startoff &&
1834             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1835             new->br_state == RIGHT.br_state &&
1836             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1837             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1838                        BMAP_RIGHT_FILLING)) !=
1839                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1840                        BMAP_RIGHT_FILLING) ||
1841              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1842                         <= MAXEXTLEN))
1843                 state |= BMAP_RIGHT_CONTIG;
1844 
1845         error = 0;
1846         /*
1847          * Switch out based on the FILLING and CONTIG state bits.
1848          */
1849         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1850                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1851         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1852              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1853                 /*
1854                  * Filling in all of a previously delayed allocation extent.
1855                  * The left and right neighbors are both contiguous with new.
1856                  */
1857                 bma->idx--;
1858                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1859                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1860                         LEFT.br_blockcount + PREV.br_blockcount +
1861                         RIGHT.br_blockcount);
1862                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1863 
1864                 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1865                 bma->ip->i_d.di_nextents--;
1866                 if (bma->cur == NULL)
1867                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1868                 else {
1869                         rval = XFS_ILOG_CORE;
1870                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1871                                         RIGHT.br_startblock,
1872                                         RIGHT.br_blockcount, &i);
1873                         if (error)
1874                                 goto done;
1875                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1876                         error = xfs_btree_delete(bma->cur, &i);
1877                         if (error)
1878                                 goto done;
1879                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1880                         error = xfs_btree_decrement(bma->cur, 0, &i);
1881                         if (error)
1882                                 goto done;
1883                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1884                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1885                                         LEFT.br_startblock,
1886                                         LEFT.br_blockcount +
1887                                         PREV.br_blockcount +
1888                                         RIGHT.br_blockcount, LEFT.br_state);
1889                         if (error)
1890                                 goto done;
1891                 }
1892                 break;
1893 
1894         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1895                 /*
1896                  * Filling in all of a previously delayed allocation extent.
1897                  * The left neighbor is contiguous, the right is not.
1898                  */
1899                 bma->idx--;
1900 
1901                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1902                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1903                         LEFT.br_blockcount + PREV.br_blockcount);
1904                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1905 
1906                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1907                 if (bma->cur == NULL)
1908                         rval = XFS_ILOG_DEXT;
1909                 else {
1910                         rval = 0;
1911                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1912                                         LEFT.br_startblock, LEFT.br_blockcount,
1913                                         &i);
1914                         if (error)
1915                                 goto done;
1916                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1917                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1918                                         LEFT.br_startblock,
1919                                         LEFT.br_blockcount +
1920                                         PREV.br_blockcount, LEFT.br_state);
1921                         if (error)
1922                                 goto done;
1923                 }
1924                 break;
1925 
1926         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1927                 /*
1928                  * Filling in all of a previously delayed allocation extent.
1929                  * The right neighbor is contiguous, the left is not.
1930                  */
1931                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1932                 xfs_bmbt_set_startblock(ep, new->br_startblock);
1933                 xfs_bmbt_set_blockcount(ep,
1934                         PREV.br_blockcount + RIGHT.br_blockcount);
1935                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1936 
1937                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1938                 if (bma->cur == NULL)
1939                         rval = XFS_ILOG_DEXT;
1940                 else {
1941                         rval = 0;
1942                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1943                                         RIGHT.br_startblock,
1944                                         RIGHT.br_blockcount, &i);
1945                         if (error)
1946                                 goto done;
1947                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1948                         error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1949                                         new->br_startblock,
1950                                         PREV.br_blockcount +
1951                                         RIGHT.br_blockcount, PREV.br_state);
1952                         if (error)
1953                                 goto done;
1954                 }
1955                 break;
1956 
1957         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1958                 /*
1959                  * Filling in all of a previously delayed allocation extent.
1960                  * Neither the left nor right neighbors are contiguous with
1961                  * the new one.
1962                  */
1963                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1964                 xfs_bmbt_set_startblock(ep, new->br_startblock);
1965                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1966 
1967                 bma->ip->i_d.di_nextents++;
1968                 if (bma->cur == NULL)
1969                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1970                 else {
1971                         rval = XFS_ILOG_CORE;
1972                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1973                                         new->br_startblock, new->br_blockcount,
1974                                         &i);
1975                         if (error)
1976                                 goto done;
1977                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1978                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1979                         error = xfs_btree_insert(bma->cur, &i);
1980                         if (error)
1981                                 goto done;
1982                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1983                 }
1984                 break;
1985 
1986         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1987                 /*
1988                  * Filling in the first part of a previous delayed allocation.
1989                  * The left neighbor is contiguous.
1990                  */
1991                 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1992                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1993                         LEFT.br_blockcount + new->br_blockcount);
1994                 xfs_bmbt_set_startoff(ep,
1995                         PREV.br_startoff + new->br_blockcount);
1996                 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1997 
1998                 temp = PREV.br_blockcount - new->br_blockcount;
1999                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2000                 xfs_bmbt_set_blockcount(ep, temp);
2001                 if (bma->cur == NULL)
2002                         rval = XFS_ILOG_DEXT;
2003                 else {
2004                         rval = 0;
2005                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
2006                                         LEFT.br_startblock, LEFT.br_blockcount,
2007                                         &i);
2008                         if (error)
2009                                 goto done;
2010                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2011                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
2012                                         LEFT.br_startblock,
2013                                         LEFT.br_blockcount +
2014                                         new->br_blockcount,
2015                                         LEFT.br_state);
2016                         if (error)
2017                                 goto done;
2018                 }
2019                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2020                         startblockval(PREV.br_startblock));
2021                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2022                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2023 
2024                 bma->idx--;
2025                 break;
2026 
2027         case BMAP_LEFT_FILLING:
2028                 /*
2029                  * Filling in the first part of a previous delayed allocation.
2030                  * The left neighbor is not contiguous.
2031                  */
2032                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2033                 xfs_bmbt_set_startoff(ep, new_endoff);
2034                 temp = PREV.br_blockcount - new->br_blockcount;
2035                 xfs_bmbt_set_blockcount(ep, temp);
2036                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
2037                 bma->ip->i_d.di_nextents++;
2038                 if (bma->cur == NULL)
2039                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2040                 else {
2041                         rval = XFS_ILOG_CORE;
2042                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2043                                         new->br_startblock, new->br_blockcount,
2044                                         &i);
2045                         if (error)
2046                                 goto done;
2047                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2048                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2049                         error = xfs_btree_insert(bma->cur, &i);
2050                         if (error)
2051                                 goto done;
2052                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2053                 }
2054 
2055                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2056                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2057                                         bma->firstblock, bma->flist,
2058                                         &bma->cur, 1, &tmp_rval, XFS_DATA_FORK);
2059                         rval |= tmp_rval;
2060                         if (error)
2061                                 goto done;
2062                 }
2063                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2064                         startblockval(PREV.br_startblock) -
2065                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2066                 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
2067                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2068                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2069                 break;
2070 
2071         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2072                 /*
2073                  * Filling in the last part of a previous delayed allocation.
2074                  * The right neighbor is contiguous with the new allocation.
2075                  */
2076                 temp = PREV.br_blockcount - new->br_blockcount;
2077                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2078                 xfs_bmbt_set_blockcount(ep, temp);
2079                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
2080                         new->br_startoff, new->br_startblock,
2081                         new->br_blockcount + RIGHT.br_blockcount,
2082                         RIGHT.br_state);
2083                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2084                 if (bma->cur == NULL)
2085                         rval = XFS_ILOG_DEXT;
2086                 else {
2087                         rval = 0;
2088                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2089                                         RIGHT.br_startblock,
2090                                         RIGHT.br_blockcount, &i);
2091                         if (error)
2092                                 goto done;
2093                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2094                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
2095                                         new->br_startblock,
2096                                         new->br_blockcount +
2097                                         RIGHT.br_blockcount,
2098                                         RIGHT.br_state);
2099                         if (error)
2100                                 goto done;
2101                 }
2102 
2103                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2104                         startblockval(PREV.br_startblock));
2105                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2106                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2107                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2108 
2109                 bma->idx++;
2110                 break;
2111 
2112         case BMAP_RIGHT_FILLING:
2113                 /*
2114                  * Filling in the last part of a previous delayed allocation.
2115                  * The right neighbor is not contiguous.
2116                  */
2117                 temp = PREV.br_blockcount - new->br_blockcount;
2118                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2119                 xfs_bmbt_set_blockcount(ep, temp);
2120                 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
2121                 bma->ip->i_d.di_nextents++;
2122                 if (bma->cur == NULL)
2123                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2124                 else {
2125                         rval = XFS_ILOG_CORE;
2126                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2127                                         new->br_startblock, new->br_blockcount,
2128                                         &i);
2129                         if (error)
2130                                 goto done;
2131                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2132                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2133                         error = xfs_btree_insert(bma->cur, &i);
2134                         if (error)
2135                                 goto done;
2136                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2137                 }
2138 
2139                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2140                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2141                                 bma->firstblock, bma->flist, &bma->cur, 1,
2142                                 &tmp_rval, XFS_DATA_FORK);
2143                         rval |= tmp_rval;
2144                         if (error)
2145                                 goto done;
2146                 }
2147                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2148                         startblockval(PREV.br_startblock) -
2149                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2150                 ep = xfs_iext_get_ext(ifp, bma->idx);
2151                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2152                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2153 
2154                 bma->idx++;
2155                 break;
2156 
2157         case 0:
2158                 /*
2159                  * Filling in the middle part of a previous delayed allocation.
2160                  * Contiguity is impossible here.
2161                  * This case is avoided almost all the time.
2162                  *
2163                  * We start with a delayed allocation:
2164                  *
2165                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2166                  *  PREV @ idx
2167                  *
2168                  * and we are allocating:
2169                  *                     +rrrrrrrrrrrrrrrrr+
2170                  *                            new
2171                  *
2172                  * and we set it up for insertion as:
2173                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2174                  *                            new
2175                  *  PREV @ idx          LEFT              RIGHT
2176                  *                      inserted at idx + 1
2177                  */
2178                 temp = new->br_startoff - PREV.br_startoff;
2179                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2180                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2181                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
2182                 LEFT = *new;
2183                 RIGHT.br_state = PREV.br_state;
2184                 RIGHT.br_startblock = nullstartblock(
2185                                 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
2186                 RIGHT.br_startoff = new_endoff;
2187                 RIGHT.br_blockcount = temp2;
2188                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2189                 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2190                 bma->ip->i_d.di_nextents++;
2191                 if (bma->cur == NULL)
2192                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2193                 else {
2194                         rval = XFS_ILOG_CORE;
2195                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2196                                         new->br_startblock, new->br_blockcount,
2197                                         &i);
2198                         if (error)
2199                                 goto done;
2200                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2201                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2202                         error = xfs_btree_insert(bma->cur, &i);
2203                         if (error)
2204                                 goto done;
2205                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2206                 }
2207 
2208                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2209                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2210                                         bma->firstblock, bma->flist, &bma->cur,
2211                                         1, &tmp_rval, XFS_DATA_FORK);
2212                         rval |= tmp_rval;
2213                         if (error)
2214                                 goto done;
2215                 }
2216                 temp = xfs_bmap_worst_indlen(bma->ip, temp);
2217                 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2218                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
2219                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2220                 if (diff > 0) {
2221                         error = xfs_icsb_modify_counters(bma->ip->i_mount,
2222                                         XFS_SBS_FDBLOCKS,
2223                                         -((int64_t)diff), 0);
2224                         ASSERT(!error);
2225                         if (error)
2226                                 goto done;
2227                 }
2228 
2229                 ep = xfs_iext_get_ext(ifp, bma->idx);
2230                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2231                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2232                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2233                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2234                         nullstartblock((int)temp2));
2235                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2236 
2237                 bma->idx++;
2238                 da_new = temp + temp2;
2239                 break;
2240 
2241         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2242         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2243         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2244         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2245         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2246         case BMAP_LEFT_CONTIG:
2247         case BMAP_RIGHT_CONTIG:
2248                 /*
2249                  * These cases are all impossible.
2250                  */
2251                 ASSERT(0);
2252         }
2253 
2254         /* convert to a btree if necessary */
2255         if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2256                 int     tmp_logflags;   /* partial log flag return val */
2257 
2258                 ASSERT(bma->cur == NULL);
2259                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2260                                 bma->firstblock, bma->flist, &bma->cur,
2261                                 da_old > 0, &tmp_logflags, XFS_DATA_FORK);
2262                 bma->logflags |= tmp_logflags;
2263                 if (error)
2264                         goto done;
2265         }
2266 
2267         /* adjust for changes in reserved delayed indirect blocks */
2268         if (da_old || da_new) {
2269                 temp = da_new;
2270                 if (bma->cur)
2271                         temp += bma->cur->bc_private.b.allocated;
2272                 ASSERT(temp <= da_old);
2273                 if (temp < da_old)
2274                         xfs_icsb_modify_counters(bma->ip->i_mount,
2275                                         XFS_SBS_FDBLOCKS,
2276                                         (int64_t)(da_old - temp), 0);
2277         }
2278 
2279         /* clear out the allocated field, done with it now in any case. */
2280         if (bma->cur)
2281                 bma->cur->bc_private.b.allocated = 0;
2282 
2283         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, XFS_DATA_FORK);
2284 done:
2285         bma->logflags |= rval;
2286         return error;
2287 #undef  LEFT
2288 #undef  RIGHT
2289 #undef  PREV
2290 }
2291 
2292 /*
2293  * Convert an unwritten allocation to a real allocation or vice versa.
2294  */
2295 STATIC int                              /* error */
2296 xfs_bmap_add_extent_unwritten_real(
2297         struct xfs_trans        *tp,
2298         xfs_inode_t             *ip,    /* incore inode pointer */
2299         xfs_extnum_t            *idx,   /* extent number to update/insert */
2300         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
2301         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
2302         xfs_fsblock_t           *first, /* pointer to firstblock variable */
2303         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2304         int                     *logflagsp) /* inode logging flags */
2305 {
2306         xfs_btree_cur_t         *cur;   /* btree cursor */
2307         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
2308         int                     error;  /* error return value */
2309         int                     i;      /* temp state */
2310         xfs_ifork_t             *ifp;   /* inode fork pointer */
2311         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
2312         xfs_exntst_t            newext; /* new extent state */
2313         xfs_exntst_t            oldext; /* old extent state */
2314         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
2315                                         /* left is 0, right is 1, prev is 2 */
2316         int                     rval=0; /* return value (logging flags) */
2317         int                     state = 0;/* state bits, accessed thru macros */
2318 
2319         *logflagsp = 0;
2320 
2321         cur = *curp;
2322         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2323 
2324         ASSERT(*idx >= 0);
2325         ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2326         ASSERT(!isnullstartblock(new->br_startblock));
2327 
2328         XFS_STATS_INC(xs_add_exlist);
2329 
2330 #define LEFT            r[0]
2331 #define RIGHT           r[1]
2332 #define PREV            r[2]
2333 
2334         /*
2335          * Set up a bunch of variables to make the tests simpler.
2336          */
2337         error = 0;
2338         ep = xfs_iext_get_ext(ifp, *idx);
2339         xfs_bmbt_get_all(ep, &PREV);
2340         newext = new->br_state;
2341         oldext = (newext == XFS_EXT_UNWRITTEN) ?
2342                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2343         ASSERT(PREV.br_state == oldext);
2344         new_endoff = new->br_startoff + new->br_blockcount;
2345         ASSERT(PREV.br_startoff <= new->br_startoff);
2346         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2347 
2348         /*
2349          * Set flags determining what part of the previous oldext allocation
2350          * extent is being replaced by a newext allocation.
2351          */
2352         if (PREV.br_startoff == new->br_startoff)
2353                 state |= BMAP_LEFT_FILLING;
2354         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2355                 state |= BMAP_RIGHT_FILLING;
2356 
2357         /*
2358          * Check and set flags if this segment has a left neighbor.
2359          * Don't set contiguous if the combined extent would be too large.
2360          */
2361         if (*idx > 0) {
2362                 state |= BMAP_LEFT_VALID;
2363                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2364 
2365                 if (isnullstartblock(LEFT.br_startblock))
2366                         state |= BMAP_LEFT_DELAY;
2367         }
2368 
2369         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2370             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2371             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2372             LEFT.br_state == newext &&
2373             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2374                 state |= BMAP_LEFT_CONTIG;
2375 
2376         /*
2377          * Check and set flags if this segment has a right neighbor.
2378          * Don't set contiguous if the combined extent would be too large.
2379          * Also check for all-three-contiguous being too large.
2380          */
2381         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
2382                 state |= BMAP_RIGHT_VALID;
2383                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2384                 if (isnullstartblock(RIGHT.br_startblock))
2385                         state |= BMAP_RIGHT_DELAY;
2386         }
2387 
2388         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2389             new_endoff == RIGHT.br_startoff &&
2390             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2391             newext == RIGHT.br_state &&
2392             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2393             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2394                        BMAP_RIGHT_FILLING)) !=
2395                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2396                        BMAP_RIGHT_FILLING) ||
2397              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2398                         <= MAXEXTLEN))
2399                 state |= BMAP_RIGHT_CONTIG;
2400 
2401         /*
2402          * Switch out based on the FILLING and CONTIG state bits.
2403          */
2404         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2405                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2406         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2407              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2408                 /*
2409                  * Setting all of a previous oldext extent to newext.
2410                  * The left and right neighbors are both contiguous with new.
2411                  */
2412                 --*idx;
2413 
2414                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2415                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2416                         LEFT.br_blockcount + PREV.br_blockcount +
2417                         RIGHT.br_blockcount);
2418                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2419 
2420                 xfs_iext_remove(ip, *idx + 1, 2, state);
2421                 ip->i_d.di_nextents -= 2;
2422                 if (cur == NULL)
2423                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2424                 else {
2425                         rval = XFS_ILOG_CORE;
2426                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2427                                         RIGHT.br_startblock,
2428                                         RIGHT.br_blockcount, &i)))
2429                                 goto done;
2430                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2431                         if ((error = xfs_btree_delete(cur, &i)))
2432                                 goto done;
2433                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2434                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2435                                 goto done;
2436                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2437                         if ((error = xfs_btree_delete(cur, &i)))
2438                                 goto done;
2439                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2440                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2441                                 goto done;
2442                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2443                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2444                                 LEFT.br_startblock,
2445                                 LEFT.br_blockcount + PREV.br_blockcount +
2446                                 RIGHT.br_blockcount, LEFT.br_state)))
2447                                 goto done;
2448                 }
2449                 break;
2450 
2451         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2452                 /*
2453                  * Setting all of a previous oldext extent to newext.
2454                  * The left neighbor is contiguous, the right is not.
2455                  */
2456                 --*idx;
2457 
2458                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2459                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2460                         LEFT.br_blockcount + PREV.br_blockcount);
2461                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2462 
2463                 xfs_iext_remove(ip, *idx + 1, 1, state);
2464                 ip->i_d.di_nextents--;
2465                 if (cur == NULL)
2466                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2467                 else {
2468                         rval = XFS_ILOG_CORE;
2469                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2470                                         PREV.br_startblock, PREV.br_blockcount,
2471                                         &i)))
2472                                 goto done;
2473                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2474                         if ((error = xfs_btree_delete(cur, &i)))
2475                                 goto done;
2476                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2477                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2478                                 goto done;
2479                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2480                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2481                                 LEFT.br_startblock,
2482                                 LEFT.br_blockcount + PREV.br_blockcount,
2483                                 LEFT.br_state)))
2484                                 goto done;
2485                 }
2486                 break;
2487 
2488         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2489                 /*
2490                  * Setting all of a previous oldext extent to newext.
2491                  * The right neighbor is contiguous, the left is not.
2492                  */
2493                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2494                 xfs_bmbt_set_blockcount(ep,
2495                         PREV.br_blockcount + RIGHT.br_blockcount);
2496                 xfs_bmbt_set_state(ep, newext);
2497                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2498                 xfs_iext_remove(ip, *idx + 1, 1, state);
2499                 ip->i_d.di_nextents--;
2500                 if (cur == NULL)
2501                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2502                 else {
2503                         rval = XFS_ILOG_CORE;
2504                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2505                                         RIGHT.br_startblock,
2506                                         RIGHT.br_blockcount, &i)))
2507                                 goto done;
2508                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2509                         if ((error = xfs_btree_delete(cur, &i)))
2510                                 goto done;
2511                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2512                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2513                                 goto done;
2514                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2515                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2516                                 new->br_startblock,
2517                                 new->br_blockcount + RIGHT.br_blockcount,
2518                                 newext)))
2519                                 goto done;
2520                 }
2521                 break;
2522 
2523         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2524                 /*
2525                  * Setting all of a previous oldext extent to newext.
2526                  * Neither the left nor right neighbors are contiguous with
2527                  * the new one.
2528                  */
2529                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2530                 xfs_bmbt_set_state(ep, newext);
2531                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2532 
2533                 if (cur == NULL)
2534                         rval = XFS_ILOG_DEXT;
2535                 else {
2536                         rval = 0;
2537                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2538                                         new->br_startblock, new->br_blockcount,
2539                                         &i)))
2540                                 goto done;
2541                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2542                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2543                                 new->br_startblock, new->br_blockcount,
2544                                 newext)))
2545                                 goto done;
2546                 }
2547                 break;
2548 
2549         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2550                 /*
2551                  * Setting the first part of a previous oldext extent to newext.
2552                  * The left neighbor is contiguous.
2553                  */
2554                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2555                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2556                         LEFT.br_blockcount + new->br_blockcount);
2557                 xfs_bmbt_set_startoff(ep,
2558                         PREV.br_startoff + new->br_blockcount);
2559                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2560 
2561                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2562                 xfs_bmbt_set_startblock(ep,
2563                         new->br_startblock + new->br_blockcount);
2564                 xfs_bmbt_set_blockcount(ep,
2565                         PREV.br_blockcount - new->br_blockcount);
2566                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2567 
2568                 --*idx;
2569 
2570                 if (cur == NULL)
2571                         rval = XFS_ILOG_DEXT;
2572                 else {
2573                         rval = 0;
2574                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2575                                         PREV.br_startblock, PREV.br_blockcount,
2576                                         &i)))
2577                                 goto done;
2578                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2579                         if ((error = xfs_bmbt_update(cur,
2580                                 PREV.br_startoff + new->br_blockcount,
2581                                 PREV.br_startblock + new->br_blockcount,
2582                                 PREV.br_blockcount - new->br_blockcount,
2583                                 oldext)))
2584                                 goto done;
2585                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2586                                 goto done;
2587                         error = xfs_bmbt_update(cur, LEFT.br_startoff,
2588                                 LEFT.br_startblock,
2589                                 LEFT.br_blockcount + new->br_blockcount,
2590                                 LEFT.br_state);
2591                         if (error)
2592                                 goto done;
2593                 }
2594                 break;
2595 
2596         case BMAP_LEFT_FILLING:
2597                 /*
2598                  * Setting the first part of a previous oldext extent to newext.
2599                  * The left neighbor is not contiguous.
2600                  */
2601                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2602                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2603                 xfs_bmbt_set_startoff(ep, new_endoff);
2604                 xfs_bmbt_set_blockcount(ep,
2605                         PREV.br_blockcount - new->br_blockcount);
2606                 xfs_bmbt_set_startblock(ep,
2607                         new->br_startblock + new->br_blockcount);
2608                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2609 
2610                 xfs_iext_insert(ip, *idx, 1, new, state);
2611                 ip->i_d.di_nextents++;
2612                 if (cur == NULL)
2613                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2614                 else {
2615                         rval = XFS_ILOG_CORE;
2616                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2617                                         PREV.br_startblock, PREV.br_blockcount,
2618                                         &i)))
2619                                 goto done;
2620                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2621                         if ((error = xfs_bmbt_update(cur,
2622                                 PREV.br_startoff + new->br_blockcount,
2623                                 PREV.br_startblock + new->br_blockcount,
2624                                 PREV.br_blockcount - new->br_blockcount,
2625                                 oldext)))
2626                                 goto done;
2627                         cur->bc_rec.b = *new;
2628                         if ((error = xfs_btree_insert(cur, &i)))
2629                                 goto done;
2630                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2631                 }
2632                 break;
2633 
2634         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2635                 /*
2636                  * Setting the last part of a previous oldext extent to newext.
2637                  * The right neighbor is contiguous with the new allocation.
2638                  */
2639                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2640                 xfs_bmbt_set_blockcount(ep,
2641                         PREV.br_blockcount - new->br_blockcount);
2642                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2643 
2644                 ++*idx;
2645 
2646                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2647                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2648                         new->br_startoff, new->br_startblock,
2649                         new->br_blockcount + RIGHT.br_blockcount, newext);
2650                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2651 
2652                 if (cur == NULL)
2653                         rval = XFS_ILOG_DEXT;
2654                 else {
2655                         rval = 0;
2656                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2657                                         PREV.br_startblock,
2658                                         PREV.br_blockcount, &i)))
2659                                 goto done;
2660                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2661                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2662                                 PREV.br_startblock,
2663                                 PREV.br_blockcount - new->br_blockcount,
2664                                 oldext)))
2665                                 goto done;
2666                         if ((error = xfs_btree_increment(cur, 0, &i)))
2667                                 goto done;
2668                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2669                                 new->br_startblock,
2670                                 new->br_blockcount + RIGHT.br_blockcount,
2671                                 newext)))
2672                                 goto done;
2673                 }
2674                 break;
2675 
2676         case BMAP_RIGHT_FILLING:
2677                 /*
2678                  * Setting the last part of a previous oldext extent to newext.
2679                  * The right neighbor is not contiguous.
2680                  */
2681                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2682                 xfs_bmbt_set_blockcount(ep,
2683                         PREV.br_blockcount - new->br_blockcount);
2684                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2685 
2686                 ++*idx;
2687                 xfs_iext_insert(ip, *idx, 1, new, state);
2688 
2689                 ip->i_d.di_nextents++;
2690                 if (cur == NULL)
2691                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2692                 else {
2693                         rval = XFS_ILOG_CORE;
2694                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2695                                         PREV.br_startblock, PREV.br_blockcount,
2696                                         &i)))
2697                                 goto done;
2698                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2699                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2700                                 PREV.br_startblock,
2701                                 PREV.br_blockcount - new->br_blockcount,
2702                                 oldext)))
2703                                 goto done;
2704                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2705                                         new->br_startblock, new->br_blockcount,
2706                                         &i)))
2707                                 goto done;
2708                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2709                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
2710                         if ((error = xfs_btree_insert(cur, &i)))
2711                                 goto done;
2712                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2713                 }
2714                 break;
2715 
2716         case 0:
2717                 /*
2718                  * Setting the middle part of a previous oldext extent to
2719                  * newext.  Contiguity is impossible here.
2720                  * One extent becomes three extents.
2721                  */
2722                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2723                 xfs_bmbt_set_blockcount(ep,
2724                         new->br_startoff - PREV.br_startoff);
2725                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2726 
2727                 r[0] = *new;
2728                 r[1].br_startoff = new_endoff;
2729                 r[1].br_blockcount =
2730                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
2731                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
2732                 r[1].br_state = oldext;
2733 
2734                 ++*idx;
2735                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
2736 
2737                 ip->i_d.di_nextents += 2;
2738                 if (cur == NULL)
2739                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2740                 else {
2741                         rval = XFS_ILOG_CORE;
2742                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2743                                         PREV.br_startblock, PREV.br_blockcount,
2744                                         &i)))
2745                                 goto done;
2746                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2747                         /* new right extent - oldext */
2748                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2749                                 r[1].br_startblock, r[1].br_blockcount,
2750                                 r[1].br_state)))
2751                                 goto done;
2752                         /* new left extent - oldext */
2753                         cur->bc_rec.b = PREV;
2754                         cur->bc_rec.b.br_blockcount =
2755                                 new->br_startoff - PREV.br_startoff;
2756                         if ((error = xfs_btree_insert(cur, &i)))
2757                                 goto done;
2758                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2759                         /*
2760                          * Reset the cursor to the position of the new extent
2761                          * we are about to insert as we can't trust it after
2762                          * the previous insert.
2763                          */
2764                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2765                                         new->br_startblock, new->br_blockcount,
2766                                         &i)))
2767                                 goto done;
2768                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2769                         /* new middle extent - newext */
2770                         cur->bc_rec.b.br_state = new->br_state;
2771                         if ((error = xfs_btree_insert(cur, &i)))
2772                                 goto done;
2773                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2774                 }
2775                 break;
2776 
2777         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2778         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2779         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2780         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2781         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2782         case BMAP_LEFT_CONTIG:
2783         case BMAP_RIGHT_CONTIG:
2784                 /*
2785                  * These cases are all impossible.
2786                  */
2787                 ASSERT(0);
2788         }
2789 
2790         /* convert to a btree if necessary */
2791         if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
2792                 int     tmp_logflags;   /* partial log flag return val */
2793 
2794                 ASSERT(cur == NULL);
2795                 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
2796                                 0, &tmp_logflags, XFS_DATA_FORK);
2797                 *logflagsp |= tmp_logflags;
2798                 if (error)
2799                         goto done;
2800         }
2801 
2802         /* clear out the allocated field, done with it now in any case. */
2803         if (cur) {
2804                 cur->bc_private.b.allocated = 0;
2805                 *curp = cur;
2806         }
2807 
2808         xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
2809 done:
2810         *logflagsp |= rval;
2811         return error;
2812 #undef  LEFT
2813 #undef  RIGHT
2814 #undef  PREV
2815 }
2816 
2817 /*
2818  * Convert a hole to a delayed allocation.
2819  */
2820 STATIC void
2821 xfs_bmap_add_extent_hole_delay(
2822         xfs_inode_t             *ip,    /* incore inode pointer */
2823         xfs_extnum_t            *idx,   /* extent number to update/insert */
2824         xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
2825 {
2826         xfs_ifork_t             *ifp;   /* inode fork pointer */
2827         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2828         xfs_filblks_t           newlen=0;       /* new indirect size */
2829         xfs_filblks_t           oldlen=0;       /* old indirect size */
2830         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2831         int                     state;  /* state bits, accessed thru macros */
2832         xfs_filblks_t           temp=0; /* temp for indirect calculations */
2833 
2834         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2835         state = 0;
2836         ASSERT(isnullstartblock(new->br_startblock));
2837 
2838         /*
2839          * Check and set flags if this segment has a left neighbor
2840          */
2841         if (*idx > 0) {
2842                 state |= BMAP_LEFT_VALID;
2843                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2844 
2845                 if (isnullstartblock(left.br_startblock))
2846                         state |= BMAP_LEFT_DELAY;
2847         }
2848 
2849         /*
2850          * Check and set flags if the current (right) segment exists.
2851          * If it doesn't exist, we're converting the hole at end-of-file.
2852          */
2853         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2854                 state |= BMAP_RIGHT_VALID;
2855                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2856 
2857                 if (isnullstartblock(right.br_startblock))
2858                         state |= BMAP_RIGHT_DELAY;
2859         }
2860 
2861         /*
2862          * Set contiguity flags on the left and right neighbors.
2863          * Don't let extents get too large, even if the pieces are contiguous.
2864          */
2865         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2866             left.br_startoff + left.br_blockcount == new->br_startoff &&
2867             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2868                 state |= BMAP_LEFT_CONTIG;
2869 
2870         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2871             new->br_startoff + new->br_blockcount == right.br_startoff &&
2872             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2873             (!(state & BMAP_LEFT_CONTIG) ||
2874              (left.br_blockcount + new->br_blockcount +
2875               right.br_blockcount <= MAXEXTLEN)))
2876                 state |= BMAP_RIGHT_CONTIG;
2877 
2878         /*
2879          * Switch out based on the contiguity flags.
2880          */
2881         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2882         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2883                 /*
2884                  * New allocation is contiguous with delayed allocations
2885                  * on the left and on the right.
2886                  * Merge all three into a single extent record.
2887                  */
2888                 --*idx;
2889                 temp = left.br_blockcount + new->br_blockcount +
2890                         right.br_blockcount;
2891 
2892                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2893                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2894                 oldlen = startblockval(left.br_startblock) +
2895                         startblockval(new->br_startblock) +
2896                         startblockval(right.br_startblock);
2897                 newlen = xfs_bmap_worst_indlen(ip, temp);
2898                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2899                         nullstartblock((int)newlen));
2900                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2901 
2902                 xfs_iext_remove(ip, *idx + 1, 1, state);
2903                 break;
2904 
2905         case BMAP_LEFT_CONTIG:
2906                 /*
2907                  * New allocation is contiguous with a delayed allocation
2908                  * on the left.
2909                  * Merge the new allocation with the left neighbor.
2910                  */
2911                 --*idx;
2912                 temp = left.br_blockcount + new->br_blockcount;
2913 
2914                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2915                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2916                 oldlen = startblockval(left.br_startblock) +
2917                         startblockval(new->br_startblock);
2918                 newlen = xfs_bmap_worst_indlen(ip, temp);
2919                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2920                         nullstartblock((int)newlen));
2921                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2922                 break;
2923 
2924         case BMAP_RIGHT_CONTIG:
2925                 /*
2926                  * New allocation is contiguous with a delayed allocation
2927                  * on the right.
2928                  * Merge the new allocation with the right neighbor.
2929                  */
2930                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2931                 temp = new->br_blockcount + right.br_blockcount;
2932                 oldlen = startblockval(new->br_startblock) +
2933                         startblockval(right.br_startblock);
2934                 newlen = xfs_bmap_worst_indlen(ip, temp);
2935                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2936                         new->br_startoff,
2937                         nullstartblock((int)newlen), temp, right.br_state);
2938                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2939                 break;
2940 
2941         case 0:
2942                 /*
2943                  * New allocation is not contiguous with another
2944                  * delayed allocation.
2945                  * Insert a new entry.
2946                  */
2947                 oldlen = newlen = 0;
2948                 xfs_iext_insert(ip, *idx, 1, new, state);
2949                 break;
2950         }
2951         if (oldlen != newlen) {
2952                 ASSERT(oldlen > newlen);
2953                 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
2954                         (int64_t)(oldlen - newlen), 0);
2955                 /*
2956                  * Nothing to do for disk quota accounting here.
2957                  */
2958         }
2959 }
2960 
2961 /*
2962  * Convert a hole to a real allocation.
2963  */
2964 STATIC int                              /* error */
2965 xfs_bmap_add_extent_hole_real(
2966         struct xfs_bmalloca     *bma,
2967         int                     whichfork)
2968 {
2969         struct xfs_bmbt_irec    *new = &bma->got;
2970         int                     error;  /* error return value */
2971         int                     i;      /* temp state */
2972         xfs_ifork_t             *ifp;   /* inode fork pointer */
2973         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2974         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2975         int                     rval=0; /* return value (logging flags) */
2976         int                     state;  /* state bits, accessed thru macros */
2977 
2978         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
2979 
2980         ASSERT(bma->idx >= 0);
2981         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2982         ASSERT(!isnullstartblock(new->br_startblock));
2983         ASSERT(!bma->cur ||
2984                !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2985 
2986         XFS_STATS_INC(xs_add_exlist);
2987 
2988         state = 0;
2989         if (whichfork == XFS_ATTR_FORK)
2990                 state |= BMAP_ATTRFORK;
2991 
2992         /*
2993          * Check and set flags if this segment has a left neighbor.
2994          */
2995         if (bma->idx > 0) {
2996                 state |= BMAP_LEFT_VALID;
2997                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
2998                 if (isnullstartblock(left.br_startblock))
2999                         state |= BMAP_LEFT_DELAY;
3000         }
3001 
3002         /*
3003          * Check and set flags if this segment has a current value.
3004          * Not true if we're inserting into the "hole" at eof.
3005          */
3006         if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
3007                 state |= BMAP_RIGHT_VALID;
3008                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
3009                 if (isnullstartblock(right.br_startblock))
3010                         state |= BMAP_RIGHT_DELAY;
3011         }
3012 
3013         /*
3014          * We're inserting a real allocation between "left" and "right".
3015          * Set the contiguity flags.  Don't let extents get too large.
3016          */
3017         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
3018             left.br_startoff + left.br_blockcount == new->br_startoff &&
3019             left.br_startblock + left.br_blockcount == new->br_startblock &&
3020             left.br_state == new->br_state &&
3021             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
3022                 state |= BMAP_LEFT_CONTIG;
3023 
3024         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
3025             new->br_startoff + new->br_blockcount == right.br_startoff &&
3026             new->br_startblock + new->br_blockcount == right.br_startblock &&
3027             new->br_state == right.br_state &&
3028             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
3029             (!(state & BMAP_LEFT_CONTIG) ||
3030              left.br_blockcount + new->br_blockcount +
3031              right.br_blockcount <= MAXEXTLEN))
3032                 state |= BMAP_RIGHT_CONTIG;
3033 
3034         error = 0;
3035         /*
3036          * Select which case we're in here, and implement it.
3037          */
3038         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
3039         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3040                 /*
3041                  * New allocation is contiguous with real allocations on the
3042                  * left and on the right.
3043                  * Merge all three into a single extent record.
3044                  */
3045                 --bma->idx;
3046                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3047                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3048                         left.br_blockcount + new->br_blockcount +
3049                         right.br_blockcount);
3050                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3051 
3052                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
3053 
3054                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3055                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
3056                 if (bma->cur == NULL) {
3057                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3058                 } else {
3059                         rval = XFS_ILOG_CORE;
3060                         error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
3061                                         right.br_startblock, right.br_blockcount,
3062                                         &i);
3063                         if (error)
3064                                 goto done;
3065                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3066                         error = xfs_btree_delete(bma->cur, &i);
3067                         if (error)
3068                                 goto done;
3069                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3070                         error = xfs_btree_decrement(bma->cur, 0, &i);
3071                         if (error)
3072                                 goto done;
3073                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3074                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
3075                                         left.br_startblock,
3076                                         left.br_blockcount +
3077                                                 new->br_blockcount +
3078                                                 right.br_blockcount,
3079                                         left.br_state);
3080                         if (error)
3081                                 goto done;
3082                 }
3083                 break;
3084 
3085         case BMAP_LEFT_CONTIG:
3086                 /*
3087                  * New allocation is contiguous with a real allocation
3088                  * on the left.
3089                  * Merge the new allocation with the left neighbor.
3090                  */
3091                 --bma->idx;
3092                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3093                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3094                         left.br_blockcount + new->br_blockcount);
3095                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3096 
3097                 if (bma->cur == NULL) {
3098                         rval = xfs_ilog_fext(whichfork);
3099                 } else {
3100                         rval = 0;
3101                         error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
3102                                         left.br_startblock, left.br_blockcount,
3103                                         &i);
3104                         if (error)
3105                                 goto done;
3106                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3107                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
3108                                         left.br_startblock,
3109                                         left.br_blockcount +
3110                                                 new->br_blockcount,
3111                                         left.br_state);
3112                         if (error)
3113                                 goto done;
3114                 }
3115                 break;
3116 
3117         case BMAP_RIGHT_CONTIG:
3118                 /*
3119                  * New allocation is contiguous with a real allocation
3120                  * on the right.
3121                  * Merge the new allocation with the right neighbor.
3122                  */
3123                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3124                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
3125                         new->br_startoff, new->br_startblock,
3126                         new->br_blockcount + right.br_blockcount,
3127                         right.br_state);
3128                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3129 
3130                 if (bma->cur == NULL) {
3131                         rval = xfs_ilog_fext(whichfork);
3132                 } else {
3133                         rval = 0;
3134                         error = xfs_bmbt_lookup_eq(bma->cur,
3135                                         right.br_startoff,
3136                                         right.br_startblock,
3137                                         right.br_blockcount, &i);
3138                         if (error)
3139                                 goto done;
3140                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3141                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
3142                                         new->br_startblock,
3143                                         new->br_blockcount +
3144                                                 right.br_blockcount,
3145                                         right.br_state);
3146                         if (error)
3147                                 goto done;
3148                 }
3149                 break;
3150 
3151         case 0:
3152                 /*
3153                  * New allocation is not contiguous with another
3154                  * real allocation.
3155                  * Insert a new entry.
3156                  */
3157                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
3158                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3159                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
3160                 if (bma->cur == NULL) {
3161                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3162                 } else {
3163                         rval = XFS_ILOG_CORE;
3164                         error = xfs_bmbt_lookup_eq(bma->cur,
3165                                         new->br_startoff,
3166                                         new->br_startblock,
3167                                         new->br_blockcount, &i);
3168                         if (error)
3169                                 goto done;
3170                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
3171                         bma->cur->bc_rec.b.br_state = new->br_state;
3172                         error = xfs_btree_insert(bma->cur, &i);
3173                         if (error)
3174                                 goto done;
3175                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3176                 }
3177                 break;
3178         }
3179 
3180         /* convert to a btree if necessary */
3181         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
3182                 int     tmp_logflags;   /* partial log flag return val */
3183 
3184                 ASSERT(bma->cur == NULL);
3185                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
3186                                 bma->firstblock, bma->flist, &bma->cur,
3187                                 0, &tmp_logflags, whichfork);
3188                 bma->logflags |= tmp_logflags;
3189                 if (error)
3190                         goto done;
3191         }
3192 
3193         /* clear out the allocated field, done with it now in any case. */
3194         if (bma->cur)
3195                 bma->cur->bc_private.b.allocated = 0;
3196 
3197         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
3198 done:
3199         bma->logflags |= rval;
3200         return error;
3201 }
3202 
3203 /*
3204  * Functions used in the extent read, allocate and remove paths
3205  */
3206 
3207 /*
3208  * Adjust the size of the new extent based on di_extsize and rt extsize.
3209  */
3210 int
3211 xfs_bmap_extsize_align(
3212         xfs_mount_t     *mp,
3213         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
3214         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
3215         xfs_extlen_t    extsz,          /* align to this extent size */
3216         int             rt,             /* is this a realtime inode? */
3217         int             eof,            /* is extent at end-of-file? */
3218         int             delay,          /* creating delalloc extent? */
3219         int             convert,        /* overwriting unwritten extent? */
3220         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
3221         xfs_extlen_t    *lenp)          /* in/out: aligned length */
3222 {
3223         xfs_fileoff_t   orig_off;       /* original offset */
3224         xfs_extlen_t    orig_alen;      /* original length */
3225         xfs_fileoff_t   orig_end;       /* original off+len */
3226         xfs_fileoff_t   nexto;          /* next file offset */
3227         xfs_fileoff_t   prevo;          /* previous file offset */
3228         xfs_fileoff_t   align_off;      /* temp for offset */
3229         xfs_extlen_t    align_alen;     /* temp for length */
3230         xfs_extlen_t    temp;           /* temp for calculations */
3231 
3232         if (convert)
3233                 return 0;
3234 
3235         orig_off = align_off = *offp;
3236         orig_alen = align_alen = *lenp;
3237         orig_end = orig_off + orig_alen;
3238 
3239         /*
3240          * If this request overlaps an existing extent, then don't
3241          * attempt to perform any additional alignment.
3242          */
3243         if (!delay && !eof &&
3244             (orig_off >= gotp->br_startoff) &&
3245             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3246                 return 0;
3247         }
3248 
3249         /*
3250          * If the file offset is unaligned vs. the extent size
3251          * we need to align it.  This will be possible unless
3252          * the file was previously written with a kernel that didn't
3253          * perform this alignment, or if a truncate shot us in the
3254          * foot.
3255          */
3256         temp = do_mod(orig_off, extsz);
3257         if (temp) {
3258                 align_alen += temp;
3259                 align_off -= temp;
3260         }
3261         /*
3262          * Same adjustment for the end of the requested area.
3263          */
3264         if ((temp = (align_alen % extsz))) {
3265                 align_alen += extsz - temp;
3266         }
3267         /*
3268          * If the previous block overlaps with this proposed allocation
3269          * then move the start forward without adjusting the length.
3270          */
3271         if (prevp->br_startoff != NULLFILEOFF) {
3272                 if (prevp->br_startblock == HOLESTARTBLOCK)
3273                         prevo = prevp->br_startoff;
3274                 else
3275                         prevo = prevp->br_startoff + prevp->br_blockcount;
3276         } else
3277                 prevo = 0;
3278         if (align_off != orig_off && align_off < prevo)
3279                 align_off = prevo;
3280         /*
3281          * If the next block overlaps with this proposed allocation
3282          * then move the start back without adjusting the length,
3283          * but not before offset 0.
3284          * This may of course make the start overlap previous block,
3285          * and if we hit the offset 0 limit then the next block
3286          * can still overlap too.
3287          */
3288         if (!eof && gotp->br_startoff != NULLFILEOFF) {
3289                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3290                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3291                         nexto = gotp->br_startoff + gotp->br_blockcount;
3292                 else
3293                         nexto = gotp->br_startoff;
3294         } else
3295                 nexto = NULLFILEOFF;
3296         if (!eof &&
3297             align_off + align_alen != orig_end &&
3298             align_off + align_alen > nexto)
3299                 align_off = nexto > align_alen ? nexto - align_alen : 0;
3300         /*
3301          * If we're now overlapping the next or previous extent that
3302          * means we can't fit an extsz piece in this hole.  Just move
3303          * the start forward to the first valid spot and set
3304          * the length so we hit the end.
3305          */
3306         if (align_off != orig_off && align_off < prevo)
3307                 align_off = prevo;
3308         if (align_off + align_alen != orig_end &&
3309             align_off + align_alen > nexto &&
3310             nexto != NULLFILEOFF) {
3311                 ASSERT(nexto > prevo);
3312                 align_alen = nexto - align_off;
3313         }
3314 
3315         /*
3316          * If realtime, and the result isn't a multiple of the realtime
3317          * extent size we need to remove blocks until it is.
3318          */
3319         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3320                 /*
3321                  * We're not covering the original request, or
3322                  * we won't be able to once we fix the length.
3323                  */
3324                 if (orig_off < align_off ||
3325                     orig_end > align_off + align_alen ||
3326                     align_alen - temp < orig_alen)
3327                         return XFS_ERROR(EINVAL);
3328                 /*
3329                  * Try to fix it by moving the start up.
3330                  */
3331                 if (align_off + temp <= orig_off) {
3332                         align_alen -= temp;
3333                         align_off += temp;
3334                 }
3335                 /*
3336                  * Try to fix it by moving the end in.
3337                  */
3338                 else if (align_off + align_alen - temp >= orig_end)
3339                         align_alen -= temp;
3340                 /*
3341                  * Set the start to the minimum then trim the length.
3342                  */
3343                 else {
3344                         align_alen -= orig_off - align_off;
3345                         align_off = orig_off;
3346                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
3347                 }
3348                 /*
3349                  * Result doesn't cover the request, fail it.
3350                  */
3351                 if (orig_off < align_off || orig_end > align_off + align_alen)
3352                         return XFS_ERROR(EINVAL);
3353         } else {
3354                 ASSERT(orig_off >= align_off);
3355                 ASSERT(orig_end <= align_off + align_alen);
3356         }
3357 
3358 #ifdef DEBUG
3359         if (!eof && gotp->br_startoff != NULLFILEOFF)
3360                 ASSERT(align_off + align_alen <= gotp->br_startoff);
3361         if (prevp->br_startoff != NULLFILEOFF)
3362                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3363 #endif
3364 
3365         *lenp = align_alen;
3366         *offp = align_off;
3367         return 0;
3368 }
3369 
3370 #define XFS_ALLOC_GAP_UNITS     4
3371 
3372 void
3373 xfs_bmap_adjacent(
3374         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3375 {
3376         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
3377         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3378         xfs_mount_t     *mp;            /* mount point structure */
3379         int             nullfb;         /* true if ap->firstblock isn't set */
3380         int             rt;             /* true if inode is realtime */
3381 
3382 #define ISVALID(x,y)    \
3383         (rt ? \
3384                 (x) < mp->m_sb.sb_rblocks : \
3385                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3386                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3387                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3388 
3389         mp = ap->ip->i_mount;
3390         nullfb = *ap->firstblock == NULLFSBLOCK;
3391         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
3392         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3393         /*
3394          * If allocating at eof, and there's a previous real block,
3395          * try to use its last block as our starting point.
3396          */
3397         if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3398             !isnullstartblock(ap->prev.br_startblock) &&
3399             ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3400                     ap->prev.br_startblock)) {
3401                 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3402                 /*
3403                  * Adjust for the gap between prevp and us.
3404                  */
3405                 adjust = ap->offset -
3406                         (ap->prev.br_startoff + ap->prev.br_blockcount);
3407                 if (adjust &&
3408                     ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3409                         ap->blkno += adjust;
3410         }
3411         /*
3412          * If not at eof, then compare the two neighbor blocks.
3413          * Figure out whether either one gives us a good starting point,
3414          * and pick the better one.
3415          */
3416         else if (!ap->eof) {
3417                 xfs_fsblock_t   gotbno;         /* right side block number */
3418                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
3419                 xfs_fsblock_t   prevbno;        /* left side block number */
3420                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
3421 
3422                 /*
3423                  * If there's a previous (left) block, select a requested
3424                  * start block based on it.
3425                  */
3426                 if (ap->prev.br_startoff != NULLFILEOFF &&
3427                     !isnullstartblock(ap->prev.br_startblock) &&
3428                     (prevbno = ap->prev.br_startblock +
3429                                ap->prev.br_blockcount) &&
3430                     ISVALID(prevbno, ap->prev.br_startblock)) {
3431                         /*
3432                          * Calculate gap to end of previous block.
3433                          */
3434                         adjust = prevdiff = ap->offset -
3435                                 (ap->prev.br_startoff +
3436                                  ap->prev.br_blockcount);
3437                         /*
3438                          * Figure the startblock based on the previous block's
3439                          * end and the gap size.
3440                          * Heuristic!
3441                          * If the gap is large relative to the piece we're
3442                          * allocating, or using it gives us an invalid block
3443                          * number, then just use the end of the previous block.
3444                          */
3445                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3446                             ISVALID(prevbno + prevdiff,
3447                                     ap->prev.br_startblock))
3448                                 prevbno += adjust;
3449                         else
3450                                 prevdiff += adjust;
3451                         /*
3452                          * If the firstblock forbids it, can't use it,
3453                          * must use default.
3454                          */
3455                         if (!rt && !nullfb &&
3456                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3457                                 prevbno = NULLFSBLOCK;
3458                 }
3459                 /*
3460                  * No previous block or can't follow it, just default.
3461                  */
3462                 else
3463                         prevbno = NULLFSBLOCK;
3464                 /*
3465                  * If there's a following (right) block, select a requested
3466                  * start block based on it.
3467                  */
3468                 if (!isnullstartblock(ap->got.br_startblock)) {
3469                         /*
3470                          * Calculate gap to start of next block.
3471                          */
3472                         adjust = gotdiff = ap->got.br_startoff - ap->offset;
3473                         /*
3474                          * Figure the startblock based on the next block's
3475                          * start and the gap size.
3476                          */
3477                         gotbno = ap->got.br_startblock;
3478                         /*
3479                          * Heuristic!
3480                          * If the gap is large relative to the piece we're
3481                          * allocating, or using it gives us an invalid block
3482                          * number, then just use the start of the next block
3483                          * offset by our length.
3484                          */
3485                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3486                             ISVALID(gotbno - gotdiff, gotbno))
3487                                 gotbno -= adjust;
3488                         else if (ISVALID(gotbno - ap->length, gotbno)) {
3489                                 gotbno -= ap->length;
3490                                 gotdiff += adjust - ap->length;
3491                         } else
3492                                 gotdiff += adjust;
3493                         /*
3494                          * If the firstblock forbids it, can't use it,
3495                          * must use default.
3496                          */
3497                         if (!rt && !nullfb &&
3498                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3499                                 gotbno = NULLFSBLOCK;
3500                 }
3501                 /*
3502                  * No next block, just default.
3503                  */
3504                 else
3505                         gotbno = NULLFSBLOCK;
3506                 /*
3507                  * If both valid, pick the better one, else the only good
3508                  * one, else ap->blkno is already set (to 0 or the inode block).
3509                  */
3510                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3511                         ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3512                 else if (prevbno != NULLFSBLOCK)
3513                         ap->blkno = prevbno;
3514                 else if (gotbno != NULLFSBLOCK)
3515                         ap->blkno = gotbno;
3516         }
3517 #undef ISVALID
3518 }
3519 
3520 STATIC int
3521 xfs_bmap_btalloc_nullfb(
3522         struct xfs_bmalloca     *ap,
3523         struct xfs_alloc_arg    *args,
3524         xfs_extlen_t            *blen)
3525 {
3526         struct xfs_mount        *mp = ap->ip->i_mount;
3527         struct xfs_perag        *pag;
3528         xfs_agnumber_t          ag, startag;
3529         int                     notinit = 0;
3530         int                     error;
3531 
3532         if (ap->userdata && xfs_inode_is_filestream(ap->ip))
3533                 args->type = XFS_ALLOCTYPE_NEAR_BNO;
3534         else
3535                 args->type = XFS_ALLOCTYPE_START_BNO;
3536         args->total = ap->total;
3537 
3538         /*
3539          * Search for an allocation group with a single extent large enough
3540          * for the request.  If one isn't found, then adjust the minimum
3541          * allocation size to the largest space found.
3542          */
3543         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3544         if (startag == NULLAGNUMBER)
3545                 startag = ag = 0;
3546 
3547         pag = xfs_perag_get(mp, ag);
3548         while (*blen < args->maxlen) {
3549                 if (!pag->pagf_init) {
3550                         error = xfs_alloc_pagf_init(mp, args->tp, ag,
3551                                                     XFS_ALLOC_FLAG_TRYLOCK);
3552                         if (error) {
3553                                 xfs_perag_put(pag);
3554                                 return error;
3555                         }
3556                 }
3557 
3558                 /*
3559                  * See xfs_alloc_fix_freelist...
3560                  */
3561                 if (pag->pagf_init) {
3562                         xfs_extlen_t    longest;
3563                         longest = xfs_alloc_longest_free_extent(mp, pag);
3564                         if (*blen < longest)
3565                                 *blen = longest;
3566                 } else
3567                         notinit = 1;
3568 
3569                 if (xfs_inode_is_filestream(ap->ip)) {
3570                         if (*blen >= args->maxlen)
3571                                 break;
3572 
3573                         if (ap->userdata) {
3574                                 /*
3575                                  * If startag is an invalid AG, we've
3576                                  * come here once before and
3577                                  * xfs_filestream_new_ag picked the
3578                                  * best currently available.
3579                                  *
3580                                  * Don't continue looping, since we
3581                                  * could loop forever.
3582                                  */
3583                                 if (startag == NULLAGNUMBER)
3584                                         break;
3585 
3586                                 error = xfs_filestream_new_ag(ap, &ag);
3587                                 xfs_perag_put(pag);
3588                                 if (error)
3589                                         return error;
3590 
3591                                 /* loop again to set 'blen'*/
3592                                 startag = NULLAGNUMBER;
3593                                 pag = xfs_perag_get(mp, ag);
3594                                 continue;
3595                         }
3596                 }
3597                 if (++ag == mp->m_sb.sb_agcount)
3598                         ag = 0;
3599                 if (ag == startag)
3600                         break;
3601                 xfs_perag_put(pag);
3602                 pag = xfs_perag_get(mp, ag);
3603         }
3604         xfs_perag_put(pag);
3605 
3606         /*
3607          * Since the above loop did a BUF_TRYLOCK, it is
3608          * possible that there is space for this request.
3609          */
3610         if (notinit || *blen < ap->minlen)
3611                 args->minlen = ap->minlen;
3612         /*
3613          * If the best seen length is less than the request
3614          * length, use the best as the minimum.
3615          */
3616         else if (*blen < args->maxlen)
3617                 args->minlen = *blen;
3618         /*
3619          * Otherwise we've seen an extent as big as maxlen,
3620          * use that as the minimum.
3621          */
3622         else
3623                 args->minlen = args->maxlen;
3624 
3625         /*
3626          * set the failure fallback case to look in the selected
3627          * AG as the stream may have moved.
3628          */
3629         if (xfs_inode_is_filestream(ap->ip))
3630                 ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3631 
3632         return 0;
3633 }
3634 
3635 STATIC int
3636 xfs_bmap_btalloc(
3637         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3638 {
3639         xfs_mount_t     *mp;            /* mount point structure */
3640         xfs_alloctype_t atype = 0;      /* type for allocation routines */
3641         xfs_extlen_t    align;          /* minimum allocation alignment */
3642         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
3643         xfs_agnumber_t  ag;
3644         xfs_alloc_arg_t args;
3645         xfs_extlen_t    blen;
3646         xfs_extlen_t    nextminlen = 0;
3647         int             nullfb;         /* true if ap->firstblock isn't set */
3648         int             isaligned;
3649         int             tryagain;
3650         int             error;
3651         int             stripe_align;
3652 
3653         ASSERT(ap->length);
3654 
3655         mp = ap->ip->i_mount;
3656 
3657         /* stripe alignment for allocation is determined by mount parameters */
3658         stripe_align = 0;
3659         if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3660                 stripe_align = mp->m_swidth;
3661         else if (mp->m_dalign)
3662                 stripe_align = mp->m_dalign;
3663 
3664         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
3665         if (unlikely(align)) {
3666                 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3667                                                 align, 0, ap->eof, 0, ap->conv,
3668                                                 &ap->offset, &ap->length);
3669                 ASSERT(!error);
3670                 ASSERT(ap->length);
3671         }
3672 
3673 
3674         nullfb = *ap->firstblock == NULLFSBLOCK;
3675         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3676         if (nullfb) {
3677                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
3678                         ag = xfs_filestream_lookup_ag(ap->ip);
3679                         ag = (ag != NULLAGNUMBER) ? ag : 0;
3680                         ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3681                 } else {
3682                         ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3683                 }
3684         } else
3685                 ap->blkno = *ap->firstblock;
3686 
3687         xfs_bmap_adjacent(ap);
3688 
3689         /*
3690          * If allowed, use ap->blkno; otherwise must use firstblock since
3691          * it's in the right allocation group.
3692          */
3693         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3694                 ;
3695         else
3696                 ap->blkno = *ap->firstblock;
3697         /*
3698          * Normal allocation, done through xfs_alloc_vextent.
3699          */
3700         tryagain = isaligned = 0;
3701         memset(&args, 0, sizeof(args));
3702         args.tp = ap->tp;
3703         args.mp = mp;
3704         args.fsbno = ap->blkno;
3705 
3706         /* Trim the allocation back to the maximum an AG can fit. */
3707         args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
3708         args.firstblock = *ap->firstblock;
3709         blen = 0;
3710         if (nullfb) {
3711                 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3712                 if (error)
3713                         return error;
3714         } else if (ap->flist->xbf_low) {
3715                 if (xfs_inode_is_filestream(ap->ip))
3716                         args.type = XFS_ALLOCTYPE_FIRST_AG;
3717                 else
3718                         args.type = XFS_ALLOCTYPE_START_BNO;
3719                 args.total = args.minlen = ap->minlen;
3720         } else {
3721                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3722                 args.total = ap->total;
3723                 args.minlen = ap->minlen;
3724         }
3725         /* apply extent size hints if obtained earlier */
3726         if (unlikely(align)) {
3727                 args.prod = align;
3728                 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3729                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
3730         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
3731                 args.prod = 1;
3732                 args.mod = 0;
3733         } else {
3734                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
3735                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3736                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
3737         }
3738         /*
3739          * If we are not low on available data blocks, and the
3740          * underlying logical volume manager is a stripe, and
3741          * the file offset is zero then try to allocate data
3742          * blocks on stripe unit boundary.
3743          * NOTE: ap->aeof is only set if the allocation length
3744          * is >= the stripe unit and the allocation offset is
3745          * at the end of file.
3746          */
3747         if (!ap->flist->xbf_low && ap->aeof) {
3748                 if (!ap->offset) {
3749                         args.alignment = stripe_align;
3750                         atype = args.type;
3751                         isaligned = 1;
3752                         /*
3753                          * Adjust for alignment
3754                          */
3755                         if (blen > args.alignment && blen <= args.maxlen)
3756                                 args.minlen = blen - args.alignment;
3757                         args.minalignslop = 0;
3758                 } else {
3759                         /*
3760                          * First try an exact bno allocation.
3761                          * If it fails then do a near or start bno
3762                          * allocation with alignment turned on.
3763                          */
3764                         atype = args.type;
3765                         tryagain = 1;
3766                         args.type = XFS_ALLOCTYPE_THIS_BNO;
3767                         args.alignment = 1;
3768                         /*
3769                          * Compute the minlen+alignment for the
3770                          * next case.  Set slop so that the value
3771                          * of minlen+alignment+slop doesn't go up
3772                          * between the calls.
3773                          */
3774                         if (blen > stripe_align && blen <= args.maxlen)
3775                                 nextminlen = blen - stripe_align;
3776                         else
3777                                 nextminlen = args.minlen;
3778                         if (nextminlen + stripe_align > args.minlen + 1)
3779                                 args.minalignslop =
3780                                         nextminlen + stripe_align -
3781                                         args.minlen - 1;
3782                         else
3783                                 args.minalignslop = 0;
3784                 }
3785         } else {
3786                 args.alignment = 1;
3787                 args.minalignslop = 0;
3788         }
3789         args.minleft = ap->minleft;
3790         args.wasdel = ap->wasdel;
3791         args.isfl = 0;
3792         args.userdata = ap->userdata;
3793         if ((error = xfs_alloc_vextent(&args)))
3794                 return error;
3795         if (tryagain && args.fsbno == NULLFSBLOCK) {
3796                 /*
3797                  * Exact allocation failed. Now try with alignment
3798                  * turned on.
3799                  */
3800                 args.type = atype;
3801                 args.fsbno = ap->blkno;
3802                 args.alignment = stripe_align;
3803                 args.minlen = nextminlen;
3804                 args.minalignslop = 0;
3805                 isaligned = 1;
3806                 if ((error = xfs_alloc_vextent(&args)))
3807                         return error;
3808         }
3809         if (isaligned && args.fsbno == NULLFSBLOCK) {
3810                 /*
3811                  * allocation failed, so turn off alignment and
3812                  * try again.
3813                  */
3814                 args.type = atype;
3815                 args.fsbno = ap->blkno;
3816                 args.alignment = 0;
3817                 if ((error = xfs_alloc_vextent(&args)))
3818                         return error;
3819         }
3820         if (args.fsbno == NULLFSBLOCK && nullfb &&
3821             args.minlen > ap->minlen) {
3822                 args.minlen = ap->minlen;
3823                 args.type = XFS_ALLOCTYPE_START_BNO;
3824                 args.fsbno = ap->blkno;
3825                 if ((error = xfs_alloc_vextent(&args)))
3826                         return error;
3827         }
3828         if (args.fsbno == NULLFSBLOCK && nullfb) {
3829                 args.fsbno = 0;
3830                 args.type = XFS_ALLOCTYPE_FIRST_AG;
3831                 args.total = ap->minlen;
3832                 args.minleft = 0;
3833                 if ((error = xfs_alloc_vextent(&args)))
3834                         return error;
3835                 ap->flist->xbf_low = 1;
3836         }
3837         if (args.fsbno != NULLFSBLOCK) {
3838                 /*
3839                  * check the allocation happened at the same or higher AG than
3840                  * the first block that was allocated.
3841                  */
3842                 ASSERT(*ap->firstblock == NULLFSBLOCK ||
3843                        XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
3844                        XFS_FSB_TO_AGNO(mp, args.fsbno) ||
3845                        (ap->flist->xbf_low &&
3846                         XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
3847                         XFS_FSB_TO_AGNO(mp, args.fsbno)));
3848 
3849                 ap->blkno = args.fsbno;
3850                 if (*ap->firstblock == NULLFSBLOCK)
3851                         *ap->firstblock = args.fsbno;
3852                 ASSERT(nullfb || fb_agno == args.agno ||
3853                        (ap->flist->xbf_low && fb_agno < args.agno));
3854                 ap->length = args.len;
3855                 ap->ip->i_d.di_nblocks += args.len;
3856                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3857                 if (ap->wasdel)
3858                         ap->ip->i_delayed_blks -= args.len;
3859                 /*
3860                  * Adjust the disk quota also. This was reserved
3861                  * earlier.
3862                  */
3863                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3864                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3865                                         XFS_TRANS_DQ_BCOUNT,
3866                         (long) args.len);
3867         } else {
3868                 ap->blkno = NULLFSBLOCK;
3869                 ap->length = 0;
3870         }
3871         return 0;
3872 }
3873 
3874 /*
3875  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3876  * It figures out where to ask the underlying allocator to put the new extent.
3877  */
3878 STATIC int
3879 xfs_bmap_alloc(
3880         struct xfs_bmalloca     *ap)    /* bmap alloc argument struct */
3881 {
3882         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
3883                 return xfs_bmap_rtalloc(ap);
3884         return xfs_bmap_btalloc(ap);
3885 }
3886 
3887 /*
3888  * Trim the returned map to the required bounds
3889  */
3890 STATIC void
3891 xfs_bmapi_trim_map(
3892         struct xfs_bmbt_irec    *mval,
3893         struct xfs_bmbt_irec    *got,
3894         xfs_fileoff_t           *bno,
3895         xfs_filblks_t           len,
3896         xfs_fileoff_t           obno,
3897         xfs_fileoff_t           end,
3898         int                     n,
3899         int                     flags)
3900 {
3901         if ((flags & XFS_BMAPI_ENTIRE) ||
3902             got->br_startoff + got->br_blockcount <= obno) {
3903                 *mval = *got;
3904                 if (isnullstartblock(got->br_startblock))
3905                         mval->br_startblock = DELAYSTARTBLOCK;
3906                 return;
3907         }
3908 
3909         if (obno > *bno)
3910                 *bno = obno;
3911         ASSERT((*bno >= obno) || (n == 0));
3912         ASSERT(*bno < end);
3913         mval->br_startoff = *bno;
3914         if (isnullstartblock(got->br_startblock))
3915                 mval->br_startblock = DELAYSTARTBLOCK;
3916         else
3917                 mval->br_startblock = got->br_startblock +
3918                                         (*bno - got->br_startoff);
3919         /*
3920          * Return the minimum of what we got and what we asked for for
3921          * the length.  We can use the len variable here because it is
3922          * modified below and we could have been there before coming
3923          * here if the first part of the allocation didn't overlap what
3924          * was asked for.
3925          */
3926         mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3927                         got->br_blockcount - (*bno - got->br_startoff));
3928         mval->br_state = got->br_state;
3929         ASSERT(mval->br_blockcount <= len);
3930         return;
3931 }
3932 
3933 /*
3934  * Update and validate the extent map to return
3935  */
3936 STATIC void
3937 xfs_bmapi_update_map(
3938         struct xfs_bmbt_irec    **map,
3939         xfs_fileoff_t           *bno,
3940         xfs_filblks_t           *len,
3941         xfs_fileoff_t           obno,
3942         xfs_fileoff_t           end,
3943         int                     *n,
3944         int                     flags)
3945 {
3946         xfs_bmbt_irec_t *mval = *map;
3947 
3948         ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3949                ((mval->br_startoff + mval->br_blockcount) <= end));
3950         ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3951                (mval->br_startoff < obno));
3952 
3953         *bno = mval->br_startoff + mval->br_blockcount;
3954         *len = end - *bno;
3955         if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3956                 /* update previous map with new information */
3957                 ASSERT(mval->br_startblock == mval[-1].br_startblock);
3958                 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3959                 ASSERT(mval->br_state == mval[-1].br_state);
3960                 mval[-1].br_blockcount = mval->br_blockcount;
3961                 mval[-1].br_state = mval->br_state;
3962         } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3963                    mval[-1].br_startblock != DELAYSTARTBLOCK &&
3964                    mval[-1].br_startblock != HOLESTARTBLOCK &&
3965                    mval->br_startblock == mval[-1].br_startblock +
3966                                           mval[-1].br_blockcount &&
3967                    ((flags & XFS_BMAPI_IGSTATE) ||
3968                         mval[-1].br_state == mval->br_state)) {
3969                 ASSERT(mval->br_startoff ==
3970                        mval[-1].br_startoff + mval[-1].br_blockcount);
3971                 mval[-1].br_blockcount += mval->br_blockcount;
3972         } else if (*n > 0 &&
3973                    mval->br_startblock == DELAYSTARTBLOCK &&
3974                    mval[-1].br_startblock == DELAYSTARTBLOCK &&
3975                    mval->br_startoff ==
3976                    mval[-1].br_startoff + mval[-1].br_blockcount) {
3977                 mval[-1].br_blockcount += mval->br_blockcount;
3978                 mval[-1].br_state = mval->br_state;
3979         } else if (!((*n == 0) &&
3980                      ((mval->br_startoff + mval->br_blockcount) <=
3981                       obno))) {
3982                 mval++;
3983                 (*n)++;
3984         }
3985         *map = mval;
3986 }
3987 
3988 /*
3989  * Map file blocks to filesystem blocks without allocation.
3990  */
3991 int
3992 xfs_bmapi_read(
3993         struct xfs_inode        *ip,
3994         xfs_fileoff_t           bno,
3995         xfs_filblks_t           len,
3996         struct xfs_bmbt_irec    *mval,
3997         int                     *nmap,
3998         int                     flags)
3999 {
4000         struct xfs_mount        *mp = ip->i_mount;
4001         struct xfs_ifork        *ifp;
4002         struct xfs_bmbt_irec    got;
4003         struct xfs_bmbt_irec    prev;
4004         xfs_fileoff_t           obno;
4005         xfs_fileoff_t           end;
4006         xfs_extnum_t            lastx;
4007         int                     error;
4008         int                     eof;
4009         int                     n = 0;
4010         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4011                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4012 
4013         ASSERT(*nmap >= 1);
4014         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4015                            XFS_BMAPI_IGSTATE)));
4016 
4017         if (unlikely(XFS_TEST_ERROR(
4018             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4019              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4020              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4021                 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4022                 return XFS_ERROR(EFSCORRUPTED);
4023         }
4024 
4025         if (XFS_FORCED_SHUTDOWN(mp))
4026                 return XFS_ERROR(EIO);
4027 
4028         XFS_STATS_INC(xs_blk_mapr);
4029 
4030         ifp = XFS_IFORK_PTR(ip, whichfork);
4031 
4032         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4033                 error = xfs_iread_extents(NULL, ip, whichfork);
4034                 if (error)
4035                         return error;
4036         }
4037 
4038         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4039         end = bno + len;
4040         obno = bno;
4041 
4042         while (bno < end && n < *nmap) {
4043                 /* Reading past eof, act as though there's a hole up to end. */
4044                 if (eof)
4045                         got.br_startoff = end;
4046                 if (got.br_startoff > bno) {
4047                         /* Reading in a hole.  */
4048                         mval->br_startoff = bno;
4049                         mval->br_startblock = HOLESTARTBLOCK;
4050                         mval->br_blockcount =
4051                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4052                         mval->br_state = XFS_EXT_NORM;
4053                         bno += mval->br_blockcount;
4054                         len -= mval->br_blockcount;
4055                         mval++;
4056                         n++;
4057                         continue;
4058                 }
4059 
4060                 /* set up the extent map to return. */
4061                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4062                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4063 
4064                 /* If we're done, stop now. */
4065                 if (bno >= end || n >= *nmap)
4066                         break;
4067 
4068                 /* Else go on to the next record. */
4069                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4070                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4071                 else
4072                         eof = 1;
4073         }
4074         *nmap = n;
4075         return 0;
4076 }
4077 
4078 STATIC int
4079 xfs_bmapi_reserve_delalloc(
4080         struct xfs_inode        *ip,
4081         xfs_fileoff_t           aoff,
4082         xfs_filblks_t           len,
4083         struct xfs_bmbt_irec    *got,
4084         struct xfs_bmbt_irec    *prev,
4085         xfs_extnum_t            *lastx,
4086         int                     eof)
4087 {
4088         struct xfs_mount        *mp = ip->i_mount;
4089         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4090         xfs_extlen_t            alen;
4091         xfs_extlen_t            indlen;
4092         char                    rt = XFS_IS_REALTIME_INODE(ip);
4093         xfs_extlen_t            extsz;
4094         int                     error;
4095 
4096         alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4097         if (!eof)
4098                 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4099 
4100         /* Figure out the extent size, adjust alen */
4101         extsz = xfs_get_extsz_hint(ip);
4102         if (extsz) {
4103                 /*
4104                  * Make sure we don't exceed a single extent length when we
4105                  * align the extent by reducing length we are going to
4106                  * allocate by the maximum amount extent size aligment may
4107                  * require.
4108                  */
4109                 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4110                 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4111                                                1, 0, &aoff, &alen);
4112                 ASSERT(!error);
4113         }
4114 
4115         if (rt)
4116                 extsz = alen / mp->m_sb.sb_rextsize;
4117 
4118         /*
4119          * Make a transaction-less quota reservation for delayed allocation
4120          * blocks.  This number gets adjusted later.  We return if we haven't
4121          * allocated blocks already inside this loop.
4122          */
4123         error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4124                         rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4125         if (error)
4126                 return error;
4127 
4128         /*
4129          * Split changing sb for alen and indlen since they could be coming
4130          * from different places.
4131          */
4132         indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4133         ASSERT(indlen > 0);
4134 
4135         if (rt) {
4136                 error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4137                                           -((int64_t)extsz), 0);
4138         } else {
4139                 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4140                                                  -((int64_t)alen), 0);
4141         }
4142 
4143         if (error)
4144                 goto out_unreserve_quota;
4145 
4146         error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4147                                          -((int64_t)indlen), 0);
4148         if (error)
4149                 goto out_unreserve_blocks;
4150 
4151 
4152         ip->i_delayed_blks += alen;
4153 
4154         got->br_startoff = aoff;
4155         got->br_startblock = nullstartblock(indlen);
4156         got->br_blockcount = alen;
4157         got->br_state = XFS_EXT_NORM;
4158         xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4159 
4160         /*
4161          * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4162          * might have merged it into one of the neighbouring ones.
4163          */
4164         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4165 
4166         ASSERT(got->br_startoff <= aoff);
4167         ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4168         ASSERT(isnullstartblock(got->br_startblock));
4169         ASSERT(got->br_state == XFS_EXT_NORM);
4170         return 0;
4171 
4172 out_unreserve_blocks:
4173         if (rt)
4174                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4175         else
4176                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4177 out_unreserve_quota:
4178         if (XFS_IS_QUOTA_ON(mp))
4179                 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4180                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4181         return error;
4182 }
4183 
4184 /*
4185  * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4186  */
4187 int
4188 xfs_bmapi_delay(
4189         struct xfs_inode        *ip,    /* incore inode */
4190         xfs_fileoff_t           bno,    /* starting file offs. mapped */
4191         xfs_filblks_t           len,    /* length to map in file */
4192         struct xfs_bmbt_irec    *mval,  /* output: map values */
4193         int                     *nmap,  /* i/o: mval size/count */
4194         int                     flags)  /* XFS_BMAPI_... */
4195 {
4196         struct xfs_mount        *mp = ip->i_mount;
4197         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4198         struct xfs_bmbt_irec    got;    /* current file extent record */
4199         struct xfs_bmbt_irec    prev;   /* previous file extent record */
4200         xfs_fileoff_t           obno;   /* old block number (offset) */
4201         xfs_fileoff_t           end;    /* end of mapped file region */
4202         xfs_extnum_t            lastx;  /* last useful extent number */
4203         int                     eof;    /* we've hit the end of extents */
4204         int                     n = 0;  /* current extent index */
4205         int                     error = 0;
4206 
4207         ASSERT(*nmap >= 1);
4208         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4209         ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4210 
4211         if (unlikely(XFS_TEST_ERROR(
4212             (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4213              XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4214              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4215                 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4216                 return XFS_ERROR(EFSCORRUPTED);
4217         }
4218 
4219         if (XFS_FORCED_SHUTDOWN(mp))
4220                 return XFS_ERROR(EIO);
4221 
4222         XFS_STATS_INC(xs_blk_mapw);
4223 
4224         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4225                 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4226                 if (error)
4227                         return error;
4228         }
4229 
4230         xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4231         end = bno + len;
4232         obno = bno;
4233 
4234         while (bno < end && n < *nmap) {
4235                 if (eof || got.br_startoff > bno) {
4236                         error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4237                                                            &prev, &lastx, eof);
4238                         if (error) {
4239                                 if (n == 0) {
4240                                         *nmap = 0;
4241                                         return error;
4242                                 }
4243                                 break;
4244                         }
4245                 }
4246 
4247                 /* set up the extent map to return. */
4248                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4249                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4250 
4251                 /* If we're done, stop now. */
4252                 if (bno >= end || n >= *nmap)
4253                         break;
4254 
4255                 /* Else go on to the next record. */
4256                 prev = got;
4257                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4258                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4259                 else
4260                         eof = 1;
4261         }
4262 
4263         *nmap = n;
4264         return 0;
4265 }
4266 
4267 
4268 int
4269 __xfs_bmapi_allocate(
4270         struct xfs_bmalloca     *bma)
4271 {
4272         struct xfs_mount        *mp = bma->ip->i_mount;
4273         int                     whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
4274                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4275         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4276         int                     tmp_logflags = 0;
4277         int                     error;
4278 
4279         ASSERT(bma->length > 0);
4280 
4281         /*
4282          * For the wasdelay case, we could also just allocate the stuff asked
4283          * for in this bmap call but that wouldn't be as good.
4284          */
4285         if (bma->wasdel) {
4286                 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4287                 bma->offset = bma->got.br_startoff;
4288                 if (bma->idx != NULLEXTNUM && bma->idx) {
4289                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4290                                          &bma->prev);
4291                 }
4292         } else {
4293                 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4294                 if (!bma->eof)
4295                         bma->length = XFS_FILBLKS_MIN(bma->length,
4296                                         bma->got.br_startoff - bma->offset);
4297         }
4298 
4299         /*
4300          * Indicate if this is the first user data in the file, or just any
4301          * user data.
4302          */
4303         if (!(bma->flags & XFS_BMAPI_METADATA)) {
4304                 bma->userdata = (bma->offset == 0) ?
4305                         XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4306         }
4307 
4308         bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4309 
4310         /*
4311          * Only want to do the alignment at the eof if it is userdata and
4312          * allocation length is larger than a stripe unit.
4313          */
4314         if (mp->m_dalign && bma->length >= mp->m_dalign &&
4315             !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4316                 error = xfs_bmap_isaeof(bma, whichfork);
4317                 if (error)
4318                         return error;
4319         }
4320 
4321         error = xfs_bmap_alloc(bma);
4322         if (error)
4323                 return error;
4324 
4325         if (bma->flist->xbf_low)
4326                 bma->minleft = 0;
4327         if (bma->cur)
4328                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4329         if (bma->blkno == NULLFSBLOCK)
4330                 return 0;
4331         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4332                 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4333                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4334                 bma->cur->bc_private.b.flist = bma->flist;
4335         }
4336         /*
4337          * Bump the number of extents we've allocated
4338          * in this call.
4339          */
4340         bma->nallocs++;
4341 
4342         if (bma->cur)
4343                 bma->cur->bc_private.b.flags =
4344                         bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4345 
4346         bma->got.br_startoff = bma->offset;
4347         bma->got.br_startblock = bma->blkno;
4348         bma->got.br_blockcount = bma->length;
4349         bma->got.br_state = XFS_EXT_NORM;
4350 
4351         /*
4352          * A wasdelay extent has been initialized, so shouldn't be flagged
4353          * as unwritten.
4354          */
4355         if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4356             xfs_sb_version_hasextflgbit(&mp->m_sb))
4357                 bma->got.br_state = XFS_EXT_UNWRITTEN;
4358 
4359         if (bma->wasdel)
4360                 error = xfs_bmap_add_extent_delay_real(bma);
4361         else
4362                 error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4363 
4364         bma->logflags |= tmp_logflags;
4365         if (error)
4366                 return error;
4367 
4368         /*
4369          * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4370          * or xfs_bmap_add_extent_hole_real might have merged it into one of
4371          * the neighbouring ones.
4372          */
4373         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4374 
4375         ASSERT(bma->got.br_startoff <= bma->offset);
4376         ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4377                bma->offset + bma->length);
4378         ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4379                bma->got.br_state == XFS_EXT_UNWRITTEN);
4380         return 0;
4381 }
4382 
4383 STATIC int
4384 xfs_bmapi_convert_unwritten(
4385         struct xfs_bmalloca     *bma,
4386         struct xfs_bmbt_irec    *mval,
4387         xfs_filblks_t           len,
4388         int                     flags)
4389 {
4390         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4391                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4392         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4393         int                     tmp_logflags = 0;
4394         int                     error;
4395 
4396         /* check if we need to do unwritten->real conversion */
4397         if (mval->br_state == XFS_EXT_UNWRITTEN &&
4398             (flags & XFS_BMAPI_PREALLOC))
4399                 return 0;
4400 
4401         /* check if we need to do real->unwritten conversion */
4402         if (mval->br_state == XFS_EXT_NORM &&
4403             (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4404                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4405                 return 0;
4406 
4407         /*
4408          * Modify (by adding) the state flag, if writing.
4409          */
4410         ASSERT(mval->br_blockcount <= len);
4411         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4412                 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4413                                         bma->ip, whichfork);
4414                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4415                 bma->cur->bc_private.b.flist = bma->flist;
4416         }
4417         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4418                                 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4419 
4420         error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4421                         &bma->cur, mval, bma->firstblock, bma->flist,
4422                         &tmp_logflags);
4423         bma->logflags |= tmp_logflags;
4424         if (error)
4425                 return error;
4426 
4427         /*
4428          * Update our extent pointer, given that
4429          * xfs_bmap_add_extent_unwritten_real might have merged it into one
4430          * of the neighbouring ones.
4431          */
4432         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4433 
4434         /*
4435          * We may have combined previously unwritten space with written space,
4436          * so generate another request.
4437          */
4438         if (mval->br_blockcount < len)
4439                 return EAGAIN;
4440         return 0;
4441 }
4442 
4443 /*
4444  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4445  * extent state if necessary.  Details behaviour is controlled by the flags
4446  * parameter.  Only allocates blocks from a single allocation group, to avoid
4447  * locking problems.
4448  *
4449  * The returned value in "firstblock" from the first call in a transaction
4450  * must be remembered and presented to subsequent calls in "firstblock".
4451  * An upper bound for the number of blocks to be allocated is supplied to
4452  * the first call in "total"; if no allocation group has that many free
4453  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4454  */
4455 int
4456 xfs_bmapi_write(
4457         struct xfs_trans        *tp,            /* transaction pointer */
4458         struct xfs_inode        *ip,            /* incore inode */
4459         xfs_fileoff_t           bno,            /* starting file offs. mapped */
4460         xfs_filblks_t           len,            /* length to map in file */
4461         int                     flags,          /* XFS_BMAPI_... */
4462         xfs_fsblock_t           *firstblock,    /* first allocated block
4463                                                    controls a.g. for allocs */
4464         xfs_extlen_t            total,          /* total blocks needed */
4465         struct xfs_bmbt_irec    *mval,          /* output: map values */
4466         int                     *nmap,          /* i/o: mval size/count */
4467         struct xfs_bmap_free    *flist)         /* i/o: list extents to free */
4468 {
4469         struct xfs_mount        *mp = ip->i_mount;
4470         struct xfs_ifork        *ifp;
4471         struct xfs_bmalloca     bma = { NULL }; /* args for xfs_bmap_alloc */
4472         xfs_fileoff_t           end;            /* end of mapped file region */
4473         int                     eof;            /* after the end of extents */
4474         int                     error;          /* error return */
4475         int                     n;              /* current extent index */
4476         xfs_fileoff_t           obno;           /* old block number (offset) */
4477         int                     whichfork;      /* data or attr fork */
4478         char                    inhole;         /* current location is hole in file */
4479         char                    wasdelay;       /* old extent was delayed */
4480 
4481 #ifdef DEBUG
4482         xfs_fileoff_t           orig_bno;       /* original block number value */
4483         int                     orig_flags;     /* original flags arg value */
4484         xfs_filblks_t           orig_len;       /* original value of len arg */
4485         struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4486         int                     orig_nmap;      /* original value of *nmap */
4487 
4488         orig_bno = bno;
4489         orig_len = len;
4490         orig_flags = flags;
4491         orig_mval = mval;
4492         orig_nmap = *nmap;
4493 #endif
4494         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4495                 XFS_ATTR_FORK : XFS_DATA_FORK;
4496 
4497         ASSERT(*nmap >= 1);
4498         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4499         ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4500         ASSERT(tp != NULL);
4501         ASSERT(len > 0);
4502         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4503 
4504         if (unlikely(XFS_TEST_ERROR(
4505             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4506              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4507              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4508                 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4509                 return XFS_ERROR(EFSCORRUPTED);
4510         }
4511 
4512         if (XFS_FORCED_SHUTDOWN(mp))
4513                 return XFS_ERROR(EIO);
4514 
4515         ifp = XFS_IFORK_PTR(ip, whichfork);
4516 
4517         XFS_STATS_INC(xs_blk_mapw);
4518 
4519         if (*firstblock == NULLFSBLOCK) {
4520                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4521                         bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4522                 else
4523                         bma.minleft = 1;
4524         } else {
4525                 bma.minleft = 0;
4526         }
4527 
4528         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4529                 error = xfs_iread_extents(tp, ip, whichfork);
4530                 if (error)
4531                         goto error0;
4532         }
4533 
4534         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4535                                 &bma.prev);
4536         n = 0;
4537         end = bno + len;
4538         obno = bno;
4539 
4540         bma.tp = tp;
4541         bma.ip = ip;
4542         bma.total = total;
4543         bma.userdata = 0;
4544         bma.flist = flist;
4545         bma.firstblock = firstblock;
4546 
4547         if (flags & XFS_BMAPI_STACK_SWITCH)
4548                 bma.stack_switch = 1;
4549 
4550         while (bno < end && n < *nmap) {
4551                 inhole = eof || bma.got.br_startoff > bno;
4552                 wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4553 
4554                 /*
4555                  * First, deal with the hole before the allocated space
4556                  * that we found, if any.
4557                  */
4558                 if (inhole || wasdelay) {
4559                         bma.eof = eof;
4560                         bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4561                         bma.wasdel = wasdelay;
4562                         bma.offset = bno;
4563                         bma.flags = flags;
4564 
4565                         /*
4566                          * There's a 32/64 bit type mismatch between the
4567                          * allocation length request (which can be 64 bits in
4568                          * length) and the bma length request, which is
4569                          * xfs_extlen_t and therefore 32 bits. Hence we have to
4570                          * check for 32-bit overflows and handle them here.
4571                          */
4572                         if (len > (xfs_filblks_t)MAXEXTLEN)
4573                                 bma.length = MAXEXTLEN;
4574                         else
4575                                 bma.length = len;
4576 
4577                         ASSERT(len > 0);
4578                         ASSERT(bma.length > 0);
4579                         error = xfs_bmapi_allocate(&bma);
4580                         if (error)
4581                                 goto error0;
4582                         if (bma.blkno == NULLFSBLOCK)
4583                                 break;
4584                 }
4585 
4586                 /* Deal with the allocated space we found.  */
4587                 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4588                                                         end, n, flags);
4589 
4590                 /* Execute unwritten extent conversion if necessary */
4591                 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4592                 if (error == EAGAIN)
4593                         continue;
4594                 if (error)
4595                         goto error0;
4596 
4597                 /* update the extent map to return */
4598                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4599 
4600                 /*
4601                  * If we're done, stop now.  Stop when we've allocated
4602                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4603                  * the transaction may get too big.
4604                  */
4605                 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4606                         break;
4607 
4608                 /* Else go on to the next record. */
4609                 bma.prev = bma.got;
4610                 if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
4611                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
4612                                          &bma.got);
4613                 } else
4614                         eof = 1;
4615         }
4616         *nmap = n;
4617 
4618         /*
4619          * Transform from btree to extents, give it cur.
4620          */
4621         if (xfs_bmap_wants_extents(ip, whichfork)) {
4622                 int             tmp_logflags = 0;
4623 
4624                 ASSERT(bma.cur);
4625                 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4626                         &tmp_logflags, whichfork);
4627                 bma.logflags |= tmp_logflags;
4628                 if (error)
4629                         goto error0;
4630         }
4631 
4632         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4633                XFS_IFORK_NEXTENTS(ip, whichfork) >
4634                 XFS_IFORK_MAXEXT(ip, whichfork));
4635         error = 0;
4636 error0:
4637         /*
4638          * Log everything.  Do this after conversion, there's no point in
4639          * logging the extent records if we've converted to btree format.
4640          */
4641         if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4642             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4643                 bma.logflags &= ~xfs_ilog_fext(whichfork);
4644         else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4645                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4646                 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4647         /*
4648          * Log whatever the flags say, even if error.  Otherwise we might miss
4649          * detecting a case where the data is changed, there's an error,
4650          * and it's not logged so we don't shutdown when we should.
4651          */
4652         if (bma.logflags)
4653                 xfs_trans_log_inode(tp, ip, bma.logflags);
4654 
4655         if (bma.cur) {
4656                 if (!error) {
4657                         ASSERT(*firstblock == NULLFSBLOCK ||
4658                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
4659                                XFS_FSB_TO_AGNO(mp,
4660                                        bma.cur->bc_private.b.firstblock) ||
4661                                (flist->xbf_low &&
4662                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
4663                                 XFS_FSB_TO_AGNO(mp,
4664                                         bma.cur->bc_private.b.firstblock)));
4665                         *firstblock = bma.cur->bc_private.b.firstblock;
4666                 }
4667                 xfs_btree_del_cursor(bma.cur,
4668                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4669         }
4670         if (!error)
4671                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4672                         orig_nmap, *nmap);
4673         return error;
4674 }
4675 
4676 /*
4677  * Called by xfs_bmapi to update file extent records and the btree
4678  * after removing space (or undoing a delayed allocation).
4679  */
4680 STATIC int                              /* error */
4681 xfs_bmap_del_extent(
4682         xfs_inode_t             *ip,    /* incore inode pointer */
4683         xfs_trans_t             *tp,    /* current transaction pointer */
4684         xfs_extnum_t            *idx,   /* extent number to update/delete */
4685         xfs_bmap_free_t         *flist, /* list of extents to be freed */
4686         xfs_btree_cur_t         *cur,   /* if null, not a btree */
4687         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
4688         int                     *logflagsp, /* inode logging flags */
4689         int                     whichfork) /* data or attr fork */
4690 {
4691         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
4692         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
4693         xfs_fsblock_t           del_endblock=0; /* first block past del */
4694         xfs_fileoff_t           del_endoff;     /* first offset past del */
4695         int                     delay;  /* current block is delayed allocated */
4696         int                     do_fx;  /* free extent at end of routine */
4697         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
4698         int                     error;  /* error return value */
4699         int                     flags;  /* inode logging flags */
4700         xfs_bmbt_irec_t         got;    /* current extent entry */
4701         xfs_fileoff_t           got_endoff;     /* first offset past got */
4702         int                     i;      /* temp state */
4703         xfs_ifork_t             *ifp;   /* inode fork pointer */
4704         xfs_mount_t             *mp;    /* mount structure */
4705         xfs_filblks_t           nblks;  /* quota/sb block count */
4706         xfs_bmbt_irec_t         new;    /* new record to be inserted */
4707         /* REFERENCED */
4708         uint                    qfield; /* quota field to update */
4709         xfs_filblks_t           temp;   /* for indirect length calculations */
4710         xfs_filblks_t           temp2;  /* for indirect length calculations */
4711         int                     state = 0;
4712 
4713         XFS_STATS_INC(xs_del_exlist);
4714 
4715         if (whichfork == XFS_ATTR_FORK)
4716                 state |= BMAP_ATTRFORK;
4717 
4718         mp = ip->i_mount;
4719         ifp = XFS_IFORK_PTR(ip, whichfork);
4720         ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
4721                 (uint)sizeof(xfs_bmbt_rec_t)));
4722         ASSERT(del->br_blockcount > 0);
4723         ep = xfs_iext_get_ext(ifp, *idx);
4724         xfs_bmbt_get_all(ep, &got);
4725         ASSERT(got.br_startoff <= del->br_startoff);
4726         del_endoff = del->br_startoff + del->br_blockcount;
4727         got_endoff = got.br_startoff + got.br_blockcount;
4728         ASSERT(got_endoff >= del_endoff);
4729         delay = isnullstartblock(got.br_startblock);
4730         ASSERT(isnullstartblock(del->br_startblock) == delay);
4731         flags = 0;
4732         qfield = 0;
4733         error = 0;
4734         /*
4735          * If deleting a real allocation, must free up the disk space.
4736          */
4737         if (!delay) {
4738                 flags = XFS_ILOG_CORE;
4739                 /*
4740                  * Realtime allocation.  Free it and record di_nblocks update.
4741                  */
4742                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
4743                         xfs_fsblock_t   bno;
4744                         xfs_filblks_t   len;
4745 
4746                         ASSERT(do_mod(del->br_blockcount,
4747                                       mp->m_sb.sb_rextsize) == 0);
4748                         ASSERT(do_mod(del->br_startblock,
4749                                       mp->m_sb.sb_rextsize) == 0);
4750                         bno = del->br_startblock;
4751                         len = del->br_blockcount;
4752                         do_div(bno, mp->m_sb.sb_rextsize);
4753                         do_div(len, mp->m_sb.sb_rextsize);
4754                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
4755                         if (error)
4756                                 goto done;
4757                         do_fx = 0;
4758                         nblks = len * mp->m_sb.sb_rextsize;
4759                         qfield = XFS_TRANS_DQ_RTBCOUNT;
4760                 }
4761                 /*
4762                  * Ordinary allocation.
4763                  */
4764                 else {
4765                         do_fx = 1;
4766                         nblks = del->br_blockcount;
4767                         qfield = XFS_TRANS_DQ_BCOUNT;
4768                 }
4769                 /*
4770                  * Set up del_endblock and cur for later.
4771                  */
4772                 del_endblock = del->br_startblock + del->br_blockcount;
4773                 if (cur) {
4774                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
4775                                         got.br_startblock, got.br_blockcount,
4776                                         &i)))
4777                                 goto done;
4778                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4779                 }
4780                 da_old = da_new = 0;
4781         } else {
4782                 da_old = startblockval(got.br_startblock);
4783                 da_new = 0;
4784                 nblks = 0;
4785                 do_fx = 0;
4786         }
4787         /*
4788          * Set flag value to use in switch statement.
4789          * Left-contig is 2, right-contig is 1.
4790          */
4791         switch (((got.br_startoff == del->br_startoff) << 1) |
4792                 (got_endoff == del_endoff)) {
4793         case 3:
4794                 /*
4795                  * Matches the whole extent.  Delete the entry.
4796                  */
4797                 xfs_iext_remove(ip, *idx, 1,
4798                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
4799                 --*idx;
4800                 if (delay)
4801                         break;
4802 
4803                 XFS_IFORK_NEXT_SET(ip, whichfork,
4804                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
4805                 flags |= XFS_ILOG_CORE;
4806                 if (!cur) {
4807                         flags |= xfs_ilog_fext(whichfork);
4808                         break;
4809                 }
4810                 if ((error = xfs_btree_delete(cur, &i)))
4811                         goto done;
4812                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4813                 break;
4814 
4815         case 2:
4816                 /*
4817                  * Deleting the first part of the extent.
4818                  */
4819                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4820                 xfs_bmbt_set_startoff(ep, del_endoff);
4821                 temp = got.br_blockcount - del->br_blockcount;
4822                 xfs_bmbt_set_blockcount(ep, temp);
4823                 if (delay) {
4824                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4825                                 da_old);
4826                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4827                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4828                         da_new = temp;
4829                         break;
4830                 }
4831                 xfs_bmbt_set_startblock(ep, del_endblock);
4832                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4833                 if (!cur) {
4834                         flags |= xfs_ilog_fext(whichfork);
4835                         break;
4836                 }
4837                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
4838                                 got.br_blockcount - del->br_blockcount,
4839                                 got.br_state)))
4840                         goto done;
4841                 break;
4842 
4843         case 1:
4844                 /*
4845                  * Deleting the last part of the extent.
4846                  */
4847                 temp = got.br_blockcount - del->br_blockcount;
4848                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4849                 xfs_bmbt_set_blockcount(ep, temp);
4850                 if (delay) {
4851                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4852                                 da_old);
4853                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4854                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4855                         da_new = temp;
4856                         break;
4857                 }
4858                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4859                 if (!cur) {
4860                         flags |= xfs_ilog_fext(whichfork);
4861                         break;
4862                 }
4863                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
4864                                 got.br_startblock,
4865                                 got.br_blockcount - del->br_blockcount,
4866                                 got.br_state)))
4867                         goto done;
4868                 break;
4869 
4870         case 0:
4871                 /*
4872                  * Deleting the middle of the extent.
4873                  */
4874                 temp = del->br_startoff - got.br_startoff;
4875                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4876                 xfs_bmbt_set_blockcount(ep, temp);
4877                 new.br_startoff = del_endoff;
4878                 temp2 = got_endoff - del_endoff;
4879                 new.br_blockcount = temp2;
4880                 new.br_state = got.br_state;
4881                 if (!delay) {
4882                         new.br_startblock = del_endblock;
4883                         flags |= XFS_ILOG_CORE;
4884                         if (cur) {
4885                                 if ((error = xfs_bmbt_update(cur,
4886                                                 got.br_startoff,
4887                                                 got.br_startblock, temp,
4888                                                 got.br_state)))
4889                                         goto done;
4890                                 if ((error = xfs_btree_increment(cur, 0, &i)))
4891                                         goto done;
4892                                 cur->bc_rec.b = new;
4893                                 error = xfs_btree_insert(cur, &i);
4894                                 if (error && error != ENOSPC)
4895                                         goto done;
4896                                 /*
4897                                  * If get no-space back from btree insert,
4898                                  * it tried a split, and we have a zero
4899                                  * block reservation.
4900                                  * Fix up our state and return the error.
4901                                  */
4902                                 if (error == ENOSPC) {
4903                                         /*
4904                                          * Reset the cursor, don't trust
4905                                          * it after any insert operation.
4906                                          */
4907                                         if ((error = xfs_bmbt_lookup_eq(cur,
4908                                                         got.br_startoff,
4909                                                         got.br_startblock,
4910                                                         temp, &i)))
4911                                                 goto done;
4912                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4913                                         /*
4914                                          * Update the btree record back
4915                                          * to the original value.
4916                                          */
4917                                         if ((error = xfs_bmbt_update(cur,
4918                                                         got.br_startoff,
4919                                                         got.br_startblock,
4920                                                         got.br_blockcount,
4921                                                         got.br_state)))
4922                                                 goto done;
4923                                         /*
4924                                          * Reset the extent record back
4925                                          * to the original value.
4926                                          */
4927                                         xfs_bmbt_set_blockcount(ep,
4928                                                 got.br_blockcount);
4929                                         flags = 0;
4930                                         error = XFS_ERROR(ENOSPC);
4931                                         goto done;
4932                                 }
4933                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4934                         } else
4935                                 flags |= xfs_ilog_fext(whichfork);
4936                         XFS_IFORK_NEXT_SET(ip, whichfork,
4937                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
4938                 } else {
4939                         ASSERT(whichfork == XFS_DATA_FORK);
4940                         temp = xfs_bmap_worst_indlen(ip, temp);
4941                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4942                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
4943                         new.br_startblock = nullstartblock((int)temp2);
4944                         da_new = temp + temp2;
4945                         while (da_new > da_old) {
4946                                 if (temp) {
4947                                         temp--;
4948                                         da_new--;
4949                                         xfs_bmbt_set_startblock(ep,
4950                                                 nullstartblock((int)temp));
4951                                 }
4952                                 if (da_new == da_old)
4953                                         break;
4954                                 if (temp2) {
4955                                         temp2--;
4956                                         da_new--;
4957                                         new.br_startblock =
4958                                                 nullstartblock((int)temp2);
4959                                 }
4960                         }
4961                 }
4962                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4963                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
4964                 ++*idx;
4965                 break;
4966         }
4967         /*
4968          * If we need to, add to list of extents to delete.
4969          */
4970         if (do_fx)
4971                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
4972                         mp);
4973         /*
4974          * Adjust inode # blocks in the file.
4975          */
4976         if (nblks)
4977                 ip->i_d.di_nblocks -= nblks;
4978         /*
4979          * Adjust quota data.
4980          */
4981         if (qfield)
4982                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
4983 
4984         /*
4985          * Account for change in delayed indirect blocks.
4986          * Nothing to do for disk quota accounting here.
4987          */
4988         ASSERT(da_old >= da_new);
4989         if (da_old > da_new) {
4990                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4991                         (int64_t)(da_old - da_new), 0);
4992         }
4993 done:
4994         *logflagsp = flags;
4995         return error;
4996 }
4997 
4998 /*
4999  * Unmap (remove) blocks from a file.
5000  * If nexts is nonzero then the number of extents to remove is limited to
5001  * that value.  If not all extents in the block range can be removed then
5002  * *done is set.
5003  */
5004 int                                             /* error */
5005 xfs_bunmapi(
5006         xfs_trans_t             *tp,            /* transaction pointer */
5007         struct xfs_inode        *ip,            /* incore inode */
5008         xfs_fileoff_t           bno,            /* starting offset to unmap */
5009         xfs_filblks_t           len,            /* length to unmap in file */
5010         int                     flags,          /* misc flags */
5011         xfs_extnum_t            nexts,          /* number of extents max */
5012         xfs_fsblock_t           *firstblock,    /* first allocated block
5013                                                    controls a.g. for allocs */
5014         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5015         int                     *done)          /* set if not done yet */
5016 {
5017         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5018         xfs_bmbt_irec_t         del;            /* extent being deleted */
5019         int                     eof;            /* is deleting at eof */
5020         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
5021         int                     error;          /* error return value */
5022         xfs_extnum_t            extno;          /* extent number in list */
5023         xfs_bmbt_irec_t         got;            /* current extent record */
5024         xfs_ifork_t             *ifp;           /* inode fork pointer */
5025         int                     isrt;           /* freeing in rt area */
5026         xfs_extnum_t            lastx;          /* last extent index used */
5027         int                     logflags;       /* transaction logging flags */
5028         xfs_extlen_t            mod;            /* rt extent offset */
5029         xfs_mount_t             *mp;            /* mount structure */
5030         xfs_extnum_t            nextents;       /* number of file extents */
5031         xfs_bmbt_irec_t         prev;           /* previous extent record */
5032         xfs_fileoff_t           start;          /* first file offset deleted */
5033         int                     tmp_logflags;   /* partial logging flags */
5034         int                     wasdel;         /* was a delayed alloc extent */
5035         int                     whichfork;      /* data or attribute fork */
5036         xfs_fsblock_t           sum;
5037 
5038         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5039 
5040         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5041                 XFS_ATTR_FORK : XFS_DATA_FORK;
5042         ifp = XFS_IFORK_PTR(ip, whichfork);
5043         if (unlikely(
5044             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5045             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5046                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5047                                  ip->i_mount);
5048                 return XFS_ERROR(EFSCORRUPTED);
5049         }
5050         mp = ip->i_mount;
5051         if (XFS_FORCED_SHUTDOWN(mp))
5052                 return XFS_ERROR(EIO);
5053 
5054         ASSERT(len > 0);
5055         ASSERT(nexts >= 0);
5056 
5057         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5058             (error = xfs_iread_extents(tp, ip, whichfork)))
5059                 return error;
5060         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5061         if (nextents == 0) {
5062                 *done = 1;
5063                 return 0;
5064         }
5065         XFS_STATS_INC(xs_blk_unmap);
5066         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5067         start = bno;
5068         bno = start + len - 1;
5069         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5070                 &prev);
5071 
5072         /*
5073          * Check to see if the given block number is past the end of the
5074          * file, back up to the last block if so...
5075          */
5076         if (eof) {
5077                 ep = xfs_iext_get_ext(ifp, --lastx);
5078                 xfs_bmbt_get_all(ep, &got);
5079                 bno = got.br_startoff + got.br_blockcount - 1;
5080         }
5081         logflags = 0;
5082         if (ifp->if_flags & XFS_IFBROOT) {
5083                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5084                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5085                 cur->bc_private.b.firstblock = *firstblock;
5086                 cur->bc_private.b.flist = flist;
5087                 cur->bc_private.b.flags = 0;
5088         } else
5089                 cur = NULL;
5090 
5091         if (isrt) {
5092                 /*
5093                  * Synchronize by locking the bitmap inode.
5094                  */
5095                 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
5096                 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5097         }
5098 
5099         extno = 0;
5100         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5101                (nexts == 0 || extno < nexts)) {
5102                 /*
5103                  * Is the found extent after a hole in which bno lives?
5104                  * Just back up to the previous extent, if so.
5105                  */
5106                 if (got.br_startoff > bno) {
5107                         if (--lastx < 0)
5108                                 break;
5109                         ep = xfs_iext_get_ext(ifp, lastx);
5110                         xfs_bmbt_get_all(ep, &got);
5111                 }
5112                 /*
5113                  * Is the last block of this extent before the range
5114                  * we're supposed to delete?  If so, we're done.
5115                  */
5116                 bno = XFS_FILEOFF_MIN(bno,
5117                         got.br_startoff + got.br_blockcount - 1);
5118                 if (bno < start)
5119                         break;
5120                 /*
5121                  * Then deal with the (possibly delayed) allocated space
5122                  * we found.
5123                  */
5124                 ASSERT(ep != NULL);
5125                 del = got;
5126                 wasdel = isnullstartblock(del.br_startblock);
5127                 if (got.br_startoff < start) {
5128                         del.br_startoff = start;
5129                         del.br_blockcount -= start - got.br_startoff;
5130                         if (!wasdel)
5131                                 del.br_startblock += start - got.br_startoff;
5132                 }
5133                 if (del.br_startoff + del.br_blockcount > bno + 1)
5134                         del.br_blockcount = bno + 1 - del.br_startoff;
5135                 sum = del.br_startblock + del.br_blockcount;
5136                 if (isrt &&
5137                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5138                         /*
5139                          * Realtime extent not lined up at the end.
5140                          * The extent could have been split into written
5141                          * and unwritten pieces, or we could just be
5142                          * unmapping part of it.  But we can't really
5143                          * get rid of part of a realtime extent.
5144                          */
5145                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5146                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5147                                 /*
5148                                  * This piece is unwritten, or we're not
5149                                  * using unwritten extents.  Skip over it.
5150                                  */
5151                                 ASSERT(bno >= mod);
5152                                 bno -= mod > del.br_blockcount ?
5153                                         del.br_blockcount : mod;
5154                                 if (bno < got.br_startoff) {
5155                                         if (--lastx >= 0)
5156                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5157                                                         ifp, lastx), &got);
5158                                 }
5159                                 continue;
5160                         }
5161                         /*
5162                          * It's written, turn it unwritten.
5163                          * This is better than zeroing it.
5164                          */
5165                         ASSERT(del.br_state == XFS_EXT_NORM);
5166                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5167                         /*
5168                          * If this spans a realtime extent boundary,
5169                          * chop it back to the start of the one we end at.
5170                          */
5171                         if (del.br_blockcount > mod) {
5172                                 del.br_startoff += del.br_blockcount - mod;
5173                                 del.br_startblock += del.br_blockcount - mod;
5174                                 del.br_blockcount = mod;
5175                         }
5176                         del.br_state = XFS_EXT_UNWRITTEN;
5177                         error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5178                                         &lastx, &cur, &del, firstblock, flist,
5179                                         &logflags);
5180                         if (error)
5181                                 goto error0;
5182                         goto nodelete;
5183                 }
5184                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5185                         /*
5186                          * Realtime extent is lined up at the end but not
5187                          * at the front.  We'll get rid of full extents if
5188                          * we can.
5189                          */
5190                         mod = mp->m_sb.sb_rextsize - mod;
5191                         if (del.br_blockcount > mod) {
5192                                 del.br_blockcount -= mod;
5193                                 del.br_startoff += mod;
5194                                 del.br_startblock += mod;
5195                         } else if ((del.br_startoff == start &&
5196                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5197                                      xfs_trans_get_block_res(tp) == 0)) ||
5198                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5199                                 /*
5200                                  * Can't make it unwritten.  There isn't
5201                                  * a full extent here so just skip it.
5202                                  */
5203                                 ASSERT(bno >= del.br_blockcount);
5204                                 bno -= del.br_blockcount;
5205                                 if (got.br_startoff > bno) {
5206                                         if (--lastx >= 0) {
5207                                                 ep = xfs_iext_get_ext(ifp,
5208                                                                       lastx);
5209                                                 xfs_bmbt_get_all(ep, &got);
5210                                         }
5211                                 }
5212                                 continue;
5213                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5214                                 /*
5215                                  * This one is already unwritten.
5216                                  * It must have a written left neighbor.
5217                                  * Unwrite the killed part of that one and
5218                                  * try again.
5219                                  */
5220                                 ASSERT(lastx > 0);
5221                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5222                                                 lastx - 1), &prev);
5223                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5224                                 ASSERT(!isnullstartblock(prev.br_startblock));
5225                                 ASSERT(del.br_startblock ==
5226                                        prev.br_startblock + prev.br_blockcount);
5227                                 if (prev.br_startoff < start) {
5228                                         mod = start - prev.br_startoff;
5229                                         prev.br_blockcount -= mod;
5230                                         prev.br_startblock += mod;
5231                                         prev.br_startoff = start;
5232                                 }
5233                                 prev.br_state = XFS_EXT_UNWRITTEN;
5234                                 lastx--;
5235                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5236                                                 ip, &lastx, &cur, &prev,
5237                                                 firstblock, flist, &logflags);
5238                                 if (error)
5239                                         goto error0;
5240                                 goto nodelete;
5241                         } else {
5242                                 ASSERT(del.br_state == XFS_EXT_NORM);
5243                                 del.br_state = XFS_EXT_UNWRITTEN;
5244                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5245                                                 ip, &lastx, &cur, &del,
5246                                                 firstblock, flist, &logflags);
5247                                 if (error)
5248                                         goto error0;
5249                                 goto nodelete;
5250                         }
5251                 }
5252                 if (wasdel) {
5253                         ASSERT(startblockval(del.br_startblock) > 0);
5254                         /* Update realtime/data freespace, unreserve quota */
5255                         if (isrt) {
5256                                 xfs_filblks_t rtexts;
5257 
5258                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5259                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5260                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5261                                                 (int64_t)rtexts, 0);
5262                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5263                                         ip, -((long)del.br_blockcount), 0,
5264                                         XFS_QMOPT_RES_RTBLKS);
5265                         } else {
5266                                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5267                                                 (int64_t)del.br_blockcount, 0);
5268                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5269                                         ip, -((long)del.br_blockcount), 0,
5270                                         XFS_QMOPT_RES_REGBLKS);
5271                         }
5272                         ip->i_delayed_blks -= del.br_blockcount;
5273                         if (cur)
5274                                 cur->bc_private.b.flags |=
5275                                         XFS_BTCUR_BPRV_WASDEL;
5276                 } else if (cur)
5277                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5278                 /*
5279                  * If it's the case where the directory code is running
5280                  * with no block reservation, and the deleted block is in
5281                  * the middle of its extent, and the resulting insert
5282                  * of an extent would cause transformation to btree format,
5283                  * then reject it.  The calling code will then swap
5284                  * blocks around instead.
5285                  * We have to do this now, rather than waiting for the
5286                  * conversion to btree format, since the transaction
5287                  * will be dirty.
5288                  */
5289                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5290                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5291                     XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5292                         XFS_IFORK_MAXEXT(ip, whichfork) &&
5293                     del.br_startoff > got.br_startoff &&
5294                     del.br_startoff + del.br_blockcount <
5295                     got.br_startoff + got.br_blockcount) {
5296                         error = XFS_ERROR(ENOSPC);
5297                         goto error0;
5298                 }
5299                 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5300                                 &tmp_logflags, whichfork);
5301                 logflags |= tmp_logflags;
5302                 if (error)
5303                         goto error0;
5304                 bno = del.br_startoff - 1;
5305 nodelete:
5306                 /*
5307                  * If not done go on to the next (previous) record.
5308                  */
5309                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5310                         if (lastx >= 0) {
5311                                 ep = xfs_iext_get_ext(ifp, lastx);
5312                                 if (xfs_bmbt_get_startoff(ep) > bno) {
5313                                         if (--lastx >= 0)
5314                                                 ep = xfs_iext_get_ext(ifp,
5315                                                                       lastx);
5316                                 }
5317                                 xfs_bmbt_get_all(ep, &got);
5318                         }
5319                         extno++;
5320                 }
5321         }
5322         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5323 
5324         /*
5325          * Convert to a btree if necessary.
5326          */
5327         if (xfs_bmap_needs_btree(ip, whichfork)) {
5328                 ASSERT(cur == NULL);
5329                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5330                         &cur, 0, &tmp_logflags, whichfork);
5331                 logflags |= tmp_logflags;
5332                 if (error)
5333                         goto error0;
5334         }
5335         /*
5336          * transform from btree to extents, give it cur
5337          */
5338         else if (xfs_bmap_wants_extents(ip, whichfork)) {
5339                 ASSERT(cur != NULL);
5340                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5341                         whichfork);
5342                 logflags |= tmp_logflags;
5343                 if (error)
5344                         goto error0;
5345         }
5346         /*
5347          * transform from extents to local?
5348          */
5349         error = 0;
5350 error0:
5351         /*
5352          * Log everything.  Do this after conversion, there's no point in
5353          * logging the extent records if we've converted to btree format.
5354          */
5355         if ((logflags & xfs_ilog_fext(whichfork)) &&
5356             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5357                 logflags &= ~xfs_ilog_fext(whichfork);
5358         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5359                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5360                 logflags &= ~xfs_ilog_fbroot(whichfork);
5361         /*
5362          * Log inode even in the error case, if the transaction
5363          * is dirty we'll need to shut down the filesystem.
5364          */
5365         if (logflags)
5366                 xfs_trans_log_inode(tp, ip, logflags);
5367         if (cur) {
5368                 if (!error) {
5369                         *firstblock = cur->bc_private.b.firstblock;
5370                         cur->bc_private.b.allocated = 0;
5371                 }
5372                 xfs_btree_del_cursor(cur,
5373                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5374         }
5375         return error;
5376 }
5377 

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

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

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

osdn.jp