xref: /openbmc/linux/fs/xfs/libxfs/xfs_bmap.c (revision b24413180f5600bcb3bb70fbed5cf186b60864bd)
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_sb.h"
26 #include "xfs_mount.h"
27 #include "xfs_defer.h"
28 #include "xfs_da_format.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_dir2.h"
31 #include "xfs_inode.h"
32 #include "xfs_btree.h"
33 #include "xfs_trans.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_extfree_item.h"
36 #include "xfs_alloc.h"
37 #include "xfs_bmap.h"
38 #include "xfs_bmap_util.h"
39 #include "xfs_bmap_btree.h"
40 #include "xfs_rtalloc.h"
41 #include "xfs_error.h"
42 #include "xfs_quota.h"
43 #include "xfs_trans_space.h"
44 #include "xfs_buf_item.h"
45 #include "xfs_trace.h"
46 #include "xfs_symlink.h"
47 #include "xfs_attr_leaf.h"
48 #include "xfs_filestream.h"
49 #include "xfs_rmap.h"
50 #include "xfs_ag_resv.h"
51 #include "xfs_refcount.h"
52 #include "xfs_icache.h"
53 
54 
55 kmem_zone_t		*xfs_bmap_free_item_zone;
56 
57 /*
58  * Miscellaneous helper functions
59  */
60 
61 /*
62  * Compute and fill in the value of the maximum depth of a bmap btree
63  * in this filesystem.  Done once, during mount.
64  */
65 void
66 xfs_bmap_compute_maxlevels(
67 	xfs_mount_t	*mp,		/* file system mount structure */
68 	int		whichfork)	/* data or attr fork */
69 {
70 	int		level;		/* btree level */
71 	uint		maxblocks;	/* max blocks at this level */
72 	uint		maxleafents;	/* max leaf entries possible */
73 	int		maxrootrecs;	/* max records in root block */
74 	int		minleafrecs;	/* min records in leaf block */
75 	int		minnoderecs;	/* min records in node block */
76 	int		sz;		/* root block size */
77 
78 	/*
79 	 * The maximum number of extents in a file, hence the maximum
80 	 * number of leaf entries, is controlled by the type of di_nextents
81 	 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
82 	 * (a signed 16-bit number, xfs_aextnum_t).
83 	 *
84 	 * Note that we can no longer assume that if we are in ATTR1 that
85 	 * the fork offset of all the inodes will be
86 	 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
87 	 * with ATTR2 and then mounted back with ATTR1, keeping the
88 	 * di_forkoff's fixed but probably at various positions. Therefore,
89 	 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
90 	 * of a minimum size available.
91 	 */
92 	if (whichfork == XFS_DATA_FORK) {
93 		maxleafents = MAXEXTNUM;
94 		sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
95 	} else {
96 		maxleafents = MAXAEXTNUM;
97 		sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
98 	}
99 	maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
100 	minleafrecs = mp->m_bmap_dmnr[0];
101 	minnoderecs = mp->m_bmap_dmnr[1];
102 	maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
103 	for (level = 1; maxblocks > 1; level++) {
104 		if (maxblocks <= maxrootrecs)
105 			maxblocks = 1;
106 		else
107 			maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
108 	}
109 	mp->m_bm_maxlevels[whichfork] = level;
110 }
111 
112 STATIC int				/* error */
113 xfs_bmbt_lookup_eq(
114 	struct xfs_btree_cur	*cur,
115 	xfs_fileoff_t		off,
116 	xfs_fsblock_t		bno,
117 	xfs_filblks_t		len,
118 	int			*stat)	/* success/failure */
119 {
120 	cur->bc_rec.b.br_startoff = off;
121 	cur->bc_rec.b.br_startblock = bno;
122 	cur->bc_rec.b.br_blockcount = len;
123 	return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
124 }
125 
126 STATIC int				/* error */
127 xfs_bmbt_lookup_ge(
128 	struct xfs_btree_cur	*cur,
129 	xfs_fileoff_t		off,
130 	xfs_fsblock_t		bno,
131 	xfs_filblks_t		len,
132 	int			*stat)	/* success/failure */
133 {
134 	cur->bc_rec.b.br_startoff = off;
135 	cur->bc_rec.b.br_startblock = bno;
136 	cur->bc_rec.b.br_blockcount = len;
137 	return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
138 }
139 
140 /*
141  * Check if the inode needs to be converted to btree format.
142  */
143 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
144 {
145 	return whichfork != XFS_COW_FORK &&
146 		XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
147 		XFS_IFORK_NEXTENTS(ip, whichfork) >
148 			XFS_IFORK_MAXEXT(ip, whichfork);
149 }
150 
151 /*
152  * Check if the inode should be converted to extent format.
153  */
154 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
155 {
156 	return whichfork != XFS_COW_FORK &&
157 		XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
158 		XFS_IFORK_NEXTENTS(ip, whichfork) <=
159 			XFS_IFORK_MAXEXT(ip, whichfork);
160 }
161 
162 /*
163  * Update the record referred to by cur to the value given
164  * by [off, bno, len, state].
165  * This either works (return 0) or gets an EFSCORRUPTED error.
166  */
167 STATIC int
168 xfs_bmbt_update(
169 	struct xfs_btree_cur	*cur,
170 	xfs_fileoff_t		off,
171 	xfs_fsblock_t		bno,
172 	xfs_filblks_t		len,
173 	xfs_exntst_t		state)
174 {
175 	union xfs_btree_rec	rec;
176 
177 	xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
178 	return xfs_btree_update(cur, &rec);
179 }
180 
181 /*
182  * Compute the worst-case number of indirect blocks that will be used
183  * for ip's delayed extent of length "len".
184  */
185 STATIC xfs_filblks_t
186 xfs_bmap_worst_indlen(
187 	xfs_inode_t	*ip,		/* incore inode pointer */
188 	xfs_filblks_t	len)		/* delayed extent length */
189 {
190 	int		level;		/* btree level number */
191 	int		maxrecs;	/* maximum record count at this level */
192 	xfs_mount_t	*mp;		/* mount structure */
193 	xfs_filblks_t	rval;		/* return value */
194 
195 	mp = ip->i_mount;
196 	maxrecs = mp->m_bmap_dmxr[0];
197 	for (level = 0, rval = 0;
198 	     level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
199 	     level++) {
200 		len += maxrecs - 1;
201 		do_div(len, maxrecs);
202 		rval += len;
203 		if (len == 1)
204 			return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
205 				level - 1;
206 		if (level == 0)
207 			maxrecs = mp->m_bmap_dmxr[1];
208 	}
209 	return rval;
210 }
211 
212 /*
213  * Calculate the default attribute fork offset for newly created inodes.
214  */
215 uint
216 xfs_default_attroffset(
217 	struct xfs_inode	*ip)
218 {
219 	struct xfs_mount	*mp = ip->i_mount;
220 	uint			offset;
221 
222 	if (mp->m_sb.sb_inodesize == 256) {
223 		offset = XFS_LITINO(mp, ip->i_d.di_version) -
224 				XFS_BMDR_SPACE_CALC(MINABTPTRS);
225 	} else {
226 		offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
227 	}
228 
229 	ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
230 	return offset;
231 }
232 
233 /*
234  * Helper routine to reset inode di_forkoff field when switching
235  * attribute fork from local to extent format - we reset it where
236  * possible to make space available for inline data fork extents.
237  */
238 STATIC void
239 xfs_bmap_forkoff_reset(
240 	xfs_inode_t	*ip,
241 	int		whichfork)
242 {
243 	if (whichfork == XFS_ATTR_FORK &&
244 	    ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
245 	    ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
246 	    ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
247 		uint	dfl_forkoff = xfs_default_attroffset(ip) >> 3;
248 
249 		if (dfl_forkoff > ip->i_d.di_forkoff)
250 			ip->i_d.di_forkoff = dfl_forkoff;
251 	}
252 }
253 
254 #ifdef DEBUG
255 STATIC struct xfs_buf *
256 xfs_bmap_get_bp(
257 	struct xfs_btree_cur	*cur,
258 	xfs_fsblock_t		bno)
259 {
260 	struct xfs_log_item_desc *lidp;
261 	int			i;
262 
263 	if (!cur)
264 		return NULL;
265 
266 	for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
267 		if (!cur->bc_bufs[i])
268 			break;
269 		if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
270 			return cur->bc_bufs[i];
271 	}
272 
273 	/* Chase down all the log items to see if the bp is there */
274 	list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
275 		struct xfs_buf_log_item	*bip;
276 		bip = (struct xfs_buf_log_item *)lidp->lid_item;
277 		if (bip->bli_item.li_type == XFS_LI_BUF &&
278 		    XFS_BUF_ADDR(bip->bli_buf) == bno)
279 			return bip->bli_buf;
280 	}
281 
282 	return NULL;
283 }
284 
285 STATIC void
286 xfs_check_block(
287 	struct xfs_btree_block	*block,
288 	xfs_mount_t		*mp,
289 	int			root,
290 	short			sz)
291 {
292 	int			i, j, dmxr;
293 	__be64			*pp, *thispa;	/* pointer to block address */
294 	xfs_bmbt_key_t		*prevp, *keyp;
295 
296 	ASSERT(be16_to_cpu(block->bb_level) > 0);
297 
298 	prevp = NULL;
299 	for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
300 		dmxr = mp->m_bmap_dmxr[0];
301 		keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
302 
303 		if (prevp) {
304 			ASSERT(be64_to_cpu(prevp->br_startoff) <
305 			       be64_to_cpu(keyp->br_startoff));
306 		}
307 		prevp = keyp;
308 
309 		/*
310 		 * Compare the block numbers to see if there are dups.
311 		 */
312 		if (root)
313 			pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
314 		else
315 			pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
316 
317 		for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
318 			if (root)
319 				thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
320 			else
321 				thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
322 			if (*thispa == *pp) {
323 				xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
324 					__func__, j, i,
325 					(unsigned long long)be64_to_cpu(*thispa));
326 				panic("%s: ptrs are equal in node\n",
327 					__func__);
328 			}
329 		}
330 	}
331 }
332 
333 /*
334  * Check that the extents for the inode ip are in the right order in all
335  * btree leaves. THis becomes prohibitively expensive for large extent count
336  * files, so don't bother with inodes that have more than 10,000 extents in
337  * them. The btree record ordering checks will still be done, so for such large
338  * bmapbt constructs that is going to catch most corruptions.
339  */
340 STATIC void
341 xfs_bmap_check_leaf_extents(
342 	xfs_btree_cur_t		*cur,	/* btree cursor or null */
343 	xfs_inode_t		*ip,		/* incore inode pointer */
344 	int			whichfork)	/* data or attr fork */
345 {
346 	struct xfs_btree_block	*block;	/* current btree block */
347 	xfs_fsblock_t		bno;	/* block # of "block" */
348 	xfs_buf_t		*bp;	/* buffer for "block" */
349 	int			error;	/* error return value */
350 	xfs_extnum_t		i=0, j;	/* index into the extents list */
351 	xfs_ifork_t		*ifp;	/* fork structure */
352 	int			level;	/* btree level, for checking */
353 	xfs_mount_t		*mp;	/* file system mount structure */
354 	__be64			*pp;	/* pointer to block address */
355 	xfs_bmbt_rec_t		*ep;	/* pointer to current extent */
356 	xfs_bmbt_rec_t		last = {0, 0}; /* last extent in prev block */
357 	xfs_bmbt_rec_t		*nextp;	/* pointer to next extent */
358 	int			bp_release = 0;
359 
360 	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
361 		return;
362 	}
363 
364 	/* skip large extent count inodes */
365 	if (ip->i_d.di_nextents > 10000)
366 		return;
367 
368 	bno = NULLFSBLOCK;
369 	mp = ip->i_mount;
370 	ifp = XFS_IFORK_PTR(ip, whichfork);
371 	block = ifp->if_broot;
372 	/*
373 	 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
374 	 */
375 	level = be16_to_cpu(block->bb_level);
376 	ASSERT(level > 0);
377 	xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
378 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
379 	bno = be64_to_cpu(*pp);
380 
381 	ASSERT(bno != NULLFSBLOCK);
382 	ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
383 	ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
384 
385 	/*
386 	 * Go down the tree until leaf level is reached, following the first
387 	 * pointer (leftmost) at each level.
388 	 */
389 	while (level-- > 0) {
390 		/* See if buf is in cur first */
391 		bp_release = 0;
392 		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
393 		if (!bp) {
394 			bp_release = 1;
395 			error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
396 						XFS_BMAP_BTREE_REF,
397 						&xfs_bmbt_buf_ops);
398 			if (error)
399 				goto error_norelse;
400 		}
401 		block = XFS_BUF_TO_BLOCK(bp);
402 		if (level == 0)
403 			break;
404 
405 		/*
406 		 * Check this block for basic sanity (increasing keys and
407 		 * no duplicate blocks).
408 		 */
409 
410 		xfs_check_block(block, mp, 0, 0);
411 		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
412 		bno = be64_to_cpu(*pp);
413 		XFS_WANT_CORRUPTED_GOTO(mp,
414 					XFS_FSB_SANITY_CHECK(mp, bno), error0);
415 		if (bp_release) {
416 			bp_release = 0;
417 			xfs_trans_brelse(NULL, bp);
418 		}
419 	}
420 
421 	/*
422 	 * Here with bp and block set to the leftmost leaf node in the tree.
423 	 */
424 	i = 0;
425 
426 	/*
427 	 * Loop over all leaf nodes checking that all extents are in the right order.
428 	 */
429 	for (;;) {
430 		xfs_fsblock_t	nextbno;
431 		xfs_extnum_t	num_recs;
432 
433 
434 		num_recs = xfs_btree_get_numrecs(block);
435 
436 		/*
437 		 * Read-ahead the next leaf block, if any.
438 		 */
439 
440 		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
441 
442 		/*
443 		 * Check all the extents to make sure they are OK.
444 		 * If we had a previous block, the last entry should
445 		 * conform with the first entry in this one.
446 		 */
447 
448 		ep = XFS_BMBT_REC_ADDR(mp, block, 1);
449 		if (i) {
450 			ASSERT(xfs_bmbt_disk_get_startoff(&last) +
451 			       xfs_bmbt_disk_get_blockcount(&last) <=
452 			       xfs_bmbt_disk_get_startoff(ep));
453 		}
454 		for (j = 1; j < num_recs; j++) {
455 			nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
456 			ASSERT(xfs_bmbt_disk_get_startoff(ep) +
457 			       xfs_bmbt_disk_get_blockcount(ep) <=
458 			       xfs_bmbt_disk_get_startoff(nextp));
459 			ep = nextp;
460 		}
461 
462 		last = *ep;
463 		i += num_recs;
464 		if (bp_release) {
465 			bp_release = 0;
466 			xfs_trans_brelse(NULL, bp);
467 		}
468 		bno = nextbno;
469 		/*
470 		 * If we've reached the end, stop.
471 		 */
472 		if (bno == NULLFSBLOCK)
473 			break;
474 
475 		bp_release = 0;
476 		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
477 		if (!bp) {
478 			bp_release = 1;
479 			error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
480 						XFS_BMAP_BTREE_REF,
481 						&xfs_bmbt_buf_ops);
482 			if (error)
483 				goto error_norelse;
484 		}
485 		block = XFS_BUF_TO_BLOCK(bp);
486 	}
487 
488 	return;
489 
490 error0:
491 	xfs_warn(mp, "%s: at error0", __func__);
492 	if (bp_release)
493 		xfs_trans_brelse(NULL, bp);
494 error_norelse:
495 	xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
496 		__func__, i);
497 	panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
498 	return;
499 }
500 
501 /*
502  * Add bmap trace insert entries for all the contents of the extent records.
503  */
504 void
505 xfs_bmap_trace_exlist(
506 	xfs_inode_t	*ip,		/* incore inode pointer */
507 	xfs_extnum_t	cnt,		/* count of entries in the list */
508 	int		whichfork,	/* data or attr or cow fork */
509 	unsigned long	caller_ip)
510 {
511 	xfs_extnum_t	idx;		/* extent record index */
512 	xfs_ifork_t	*ifp;		/* inode fork pointer */
513 	int		state = 0;
514 
515 	if (whichfork == XFS_ATTR_FORK)
516 		state |= BMAP_ATTRFORK;
517 	else if (whichfork == XFS_COW_FORK)
518 		state |= BMAP_COWFORK;
519 
520 	ifp = XFS_IFORK_PTR(ip, whichfork);
521 	ASSERT(cnt == xfs_iext_count(ifp));
522 	for (idx = 0; idx < cnt; idx++)
523 		trace_xfs_extlist(ip, idx, state, caller_ip);
524 }
525 
526 /*
527  * Validate that the bmbt_irecs being returned from bmapi are valid
528  * given the caller's original parameters.  Specifically check the
529  * ranges of the returned irecs to ensure that they only extend beyond
530  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
531  */
532 STATIC void
533 xfs_bmap_validate_ret(
534 	xfs_fileoff_t		bno,
535 	xfs_filblks_t		len,
536 	int			flags,
537 	xfs_bmbt_irec_t		*mval,
538 	int			nmap,
539 	int			ret_nmap)
540 {
541 	int			i;		/* index to map values */
542 
543 	ASSERT(ret_nmap <= nmap);
544 
545 	for (i = 0; i < ret_nmap; i++) {
546 		ASSERT(mval[i].br_blockcount > 0);
547 		if (!(flags & XFS_BMAPI_ENTIRE)) {
548 			ASSERT(mval[i].br_startoff >= bno);
549 			ASSERT(mval[i].br_blockcount <= len);
550 			ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
551 			       bno + len);
552 		} else {
553 			ASSERT(mval[i].br_startoff < bno + len);
554 			ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
555 			       bno);
556 		}
557 		ASSERT(i == 0 ||
558 		       mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
559 		       mval[i].br_startoff);
560 		ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
561 		       mval[i].br_startblock != HOLESTARTBLOCK);
562 		ASSERT(mval[i].br_state == XFS_EXT_NORM ||
563 		       mval[i].br_state == XFS_EXT_UNWRITTEN);
564 	}
565 }
566 
567 #else
568 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)		do { } while (0)
569 #define	xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)	do { } while (0)
570 #endif /* DEBUG */
571 
572 /*
573  * bmap free list manipulation functions
574  */
575 
576 /*
577  * Add the extent to the list of extents to be free at transaction end.
578  * The list is maintained sorted (by block number).
579  */
580 void
581 xfs_bmap_add_free(
582 	struct xfs_mount		*mp,
583 	struct xfs_defer_ops		*dfops,
584 	xfs_fsblock_t			bno,
585 	xfs_filblks_t			len,
586 	struct xfs_owner_info		*oinfo)
587 {
588 	struct xfs_extent_free_item	*new;		/* new element */
589 #ifdef DEBUG
590 	xfs_agnumber_t		agno;
591 	xfs_agblock_t		agbno;
592 
593 	ASSERT(bno != NULLFSBLOCK);
594 	ASSERT(len > 0);
595 	ASSERT(len <= MAXEXTLEN);
596 	ASSERT(!isnullstartblock(bno));
597 	agno = XFS_FSB_TO_AGNO(mp, bno);
598 	agbno = XFS_FSB_TO_AGBNO(mp, bno);
599 	ASSERT(agno < mp->m_sb.sb_agcount);
600 	ASSERT(agbno < mp->m_sb.sb_agblocks);
601 	ASSERT(len < mp->m_sb.sb_agblocks);
602 	ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
603 #endif
604 	ASSERT(xfs_bmap_free_item_zone != NULL);
605 
606 	new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
607 	new->xefi_startblock = bno;
608 	new->xefi_blockcount = (xfs_extlen_t)len;
609 	if (oinfo)
610 		new->xefi_oinfo = *oinfo;
611 	else
612 		xfs_rmap_skip_owner_update(&new->xefi_oinfo);
613 	trace_xfs_bmap_free_defer(mp, XFS_FSB_TO_AGNO(mp, bno), 0,
614 			XFS_FSB_TO_AGBNO(mp, bno), len);
615 	xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list);
616 }
617 
618 /*
619  * Inode fork format manipulation functions
620  */
621 
622 /*
623  * Transform a btree format file with only one leaf node, where the
624  * extents list will fit in the inode, into an extents format file.
625  * Since the file extents are already in-core, all we have to do is
626  * give up the space for the btree root and pitch the leaf block.
627  */
628 STATIC int				/* error */
629 xfs_bmap_btree_to_extents(
630 	xfs_trans_t		*tp,	/* transaction pointer */
631 	xfs_inode_t		*ip,	/* incore inode pointer */
632 	xfs_btree_cur_t		*cur,	/* btree cursor */
633 	int			*logflagsp, /* inode logging flags */
634 	int			whichfork)  /* data or attr fork */
635 {
636 	/* REFERENCED */
637 	struct xfs_btree_block	*cblock;/* child btree block */
638 	xfs_fsblock_t		cbno;	/* child block number */
639 	xfs_buf_t		*cbp;	/* child block's buffer */
640 	int			error;	/* error return value */
641 	xfs_ifork_t		*ifp;	/* inode fork data */
642 	xfs_mount_t		*mp;	/* mount point structure */
643 	__be64			*pp;	/* ptr to block address */
644 	struct xfs_btree_block	*rblock;/* root btree block */
645 	struct xfs_owner_info	oinfo;
646 
647 	mp = ip->i_mount;
648 	ifp = XFS_IFORK_PTR(ip, whichfork);
649 	ASSERT(whichfork != XFS_COW_FORK);
650 	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
651 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
652 	rblock = ifp->if_broot;
653 	ASSERT(be16_to_cpu(rblock->bb_level) == 1);
654 	ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
655 	ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
656 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
657 	cbno = be64_to_cpu(*pp);
658 	*logflagsp = 0;
659 #ifdef DEBUG
660 	if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
661 		return error;
662 #endif
663 	error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
664 				&xfs_bmbt_buf_ops);
665 	if (error)
666 		return error;
667 	cblock = XFS_BUF_TO_BLOCK(cbp);
668 	if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
669 		return error;
670 	xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
671 	xfs_bmap_add_free(mp, cur->bc_private.b.dfops, cbno, 1, &oinfo);
672 	ip->i_d.di_nblocks--;
673 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
674 	xfs_trans_binval(tp, cbp);
675 	if (cur->bc_bufs[0] == cbp)
676 		cur->bc_bufs[0] = NULL;
677 	xfs_iroot_realloc(ip, -1, whichfork);
678 	ASSERT(ifp->if_broot == NULL);
679 	ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
680 	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
681 	*logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
682 	return 0;
683 }
684 
685 /*
686  * Convert an extents-format file into a btree-format file.
687  * The new file will have a root block (in the inode) and a single child block.
688  */
689 STATIC int					/* error */
690 xfs_bmap_extents_to_btree(
691 	xfs_trans_t		*tp,		/* transaction pointer */
692 	xfs_inode_t		*ip,		/* incore inode pointer */
693 	xfs_fsblock_t		*firstblock,	/* first-block-allocated */
694 	struct xfs_defer_ops	*dfops,		/* blocks freed in xaction */
695 	xfs_btree_cur_t		**curp,		/* cursor returned to caller */
696 	int			wasdel,		/* converting a delayed alloc */
697 	int			*logflagsp,	/* inode logging flags */
698 	int			whichfork)	/* data or attr fork */
699 {
700 	struct xfs_btree_block	*ablock;	/* allocated (child) bt block */
701 	xfs_buf_t		*abp;		/* buffer for ablock */
702 	xfs_alloc_arg_t		args;		/* allocation arguments */
703 	xfs_bmbt_rec_t		*arp;		/* child record pointer */
704 	struct xfs_btree_block	*block;		/* btree root block */
705 	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
706 	xfs_bmbt_rec_host_t	*ep;		/* extent record pointer */
707 	int			error;		/* error return value */
708 	xfs_extnum_t		i, cnt;		/* extent record index */
709 	xfs_ifork_t		*ifp;		/* inode fork pointer */
710 	xfs_bmbt_key_t		*kp;		/* root block key pointer */
711 	xfs_mount_t		*mp;		/* mount structure */
712 	xfs_extnum_t		nextents;	/* number of file extents */
713 	xfs_bmbt_ptr_t		*pp;		/* root block address pointer */
714 
715 	mp = ip->i_mount;
716 	ASSERT(whichfork != XFS_COW_FORK);
717 	ifp = XFS_IFORK_PTR(ip, whichfork);
718 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
719 
720 	/*
721 	 * Make space in the inode incore.
722 	 */
723 	xfs_iroot_realloc(ip, 1, whichfork);
724 	ifp->if_flags |= XFS_IFBROOT;
725 
726 	/*
727 	 * Fill in the root.
728 	 */
729 	block = ifp->if_broot;
730 	xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
731 				 XFS_BTNUM_BMAP, 1, 1, ip->i_ino,
732 				 XFS_BTREE_LONG_PTRS);
733 	/*
734 	 * Need a cursor.  Can't allocate until bb_level is filled in.
735 	 */
736 	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
737 	cur->bc_private.b.firstblock = *firstblock;
738 	cur->bc_private.b.dfops = dfops;
739 	cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
740 	/*
741 	 * Convert to a btree with two levels, one record in root.
742 	 */
743 	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
744 	memset(&args, 0, sizeof(args));
745 	args.tp = tp;
746 	args.mp = mp;
747 	xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork);
748 	args.firstblock = *firstblock;
749 	if (*firstblock == NULLFSBLOCK) {
750 		args.type = XFS_ALLOCTYPE_START_BNO;
751 		args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
752 	} else if (dfops->dop_low) {
753 		args.type = XFS_ALLOCTYPE_START_BNO;
754 		args.fsbno = *firstblock;
755 	} else {
756 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
757 		args.fsbno = *firstblock;
758 	}
759 	args.minlen = args.maxlen = args.prod = 1;
760 	args.wasdel = wasdel;
761 	*logflagsp = 0;
762 	if ((error = xfs_alloc_vextent(&args))) {
763 		xfs_iroot_realloc(ip, -1, whichfork);
764 		xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
765 		return error;
766 	}
767 
768 	if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
769 		xfs_iroot_realloc(ip, -1, whichfork);
770 		xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
771 		return -ENOSPC;
772 	}
773 	/*
774 	 * Allocation can't fail, the space was reserved.
775 	 */
776 	ASSERT(*firstblock == NULLFSBLOCK ||
777 	       args.agno >= XFS_FSB_TO_AGNO(mp, *firstblock));
778 	*firstblock = cur->bc_private.b.firstblock = args.fsbno;
779 	cur->bc_private.b.allocated++;
780 	ip->i_d.di_nblocks++;
781 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
782 	abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
783 	/*
784 	 * Fill in the child block.
785 	 */
786 	abp->b_ops = &xfs_bmbt_buf_ops;
787 	ablock = XFS_BUF_TO_BLOCK(abp);
788 	xfs_btree_init_block_int(mp, ablock, abp->b_bn,
789 				XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
790 				XFS_BTREE_LONG_PTRS);
791 
792 	arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
793 	nextents =  xfs_iext_count(ifp);
794 	for (cnt = i = 0; i < nextents; i++) {
795 		ep = xfs_iext_get_ext(ifp, i);
796 		if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
797 			arp->l0 = cpu_to_be64(ep->l0);
798 			arp->l1 = cpu_to_be64(ep->l1);
799 			arp++; cnt++;
800 		}
801 	}
802 	ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
803 	xfs_btree_set_numrecs(ablock, cnt);
804 
805 	/*
806 	 * Fill in the root key and pointer.
807 	 */
808 	kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
809 	arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
810 	kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
811 	pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
812 						be16_to_cpu(block->bb_level)));
813 	*pp = cpu_to_be64(args.fsbno);
814 
815 	/*
816 	 * Do all this logging at the end so that
817 	 * the root is at the right level.
818 	 */
819 	xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
820 	xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
821 	ASSERT(*curp == NULL);
822 	*curp = cur;
823 	*logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
824 	return 0;
825 }
826 
827 /*
828  * Convert a local file to an extents file.
829  * This code is out of bounds for data forks of regular files,
830  * since the file data needs to get logged so things will stay consistent.
831  * (The bmap-level manipulations are ok, though).
832  */
833 void
834 xfs_bmap_local_to_extents_empty(
835 	struct xfs_inode	*ip,
836 	int			whichfork)
837 {
838 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
839 
840 	ASSERT(whichfork != XFS_COW_FORK);
841 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
842 	ASSERT(ifp->if_bytes == 0);
843 	ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
844 
845 	xfs_bmap_forkoff_reset(ip, whichfork);
846 	ifp->if_flags &= ~XFS_IFINLINE;
847 	ifp->if_flags |= XFS_IFEXTENTS;
848 	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
849 }
850 
851 
852 STATIC int				/* error */
853 xfs_bmap_local_to_extents(
854 	xfs_trans_t	*tp,		/* transaction pointer */
855 	xfs_inode_t	*ip,		/* incore inode pointer */
856 	xfs_fsblock_t	*firstblock,	/* first block allocated in xaction */
857 	xfs_extlen_t	total,		/* total blocks needed by transaction */
858 	int		*logflagsp,	/* inode logging flags */
859 	int		whichfork,
860 	void		(*init_fn)(struct xfs_trans *tp,
861 				   struct xfs_buf *bp,
862 				   struct xfs_inode *ip,
863 				   struct xfs_ifork *ifp))
864 {
865 	int		error = 0;
866 	int		flags;		/* logging flags returned */
867 	xfs_ifork_t	*ifp;		/* inode fork pointer */
868 	xfs_alloc_arg_t	args;		/* allocation arguments */
869 	xfs_buf_t	*bp;		/* buffer for extent block */
870 	struct xfs_bmbt_irec rec;
871 
872 	/*
873 	 * We don't want to deal with the case of keeping inode data inline yet.
874 	 * So sending the data fork of a regular inode is invalid.
875 	 */
876 	ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
877 	ifp = XFS_IFORK_PTR(ip, whichfork);
878 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
879 
880 	if (!ifp->if_bytes) {
881 		xfs_bmap_local_to_extents_empty(ip, whichfork);
882 		flags = XFS_ILOG_CORE;
883 		goto done;
884 	}
885 
886 	flags = 0;
887 	error = 0;
888 	ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) ==
889 								XFS_IFINLINE);
890 	memset(&args, 0, sizeof(args));
891 	args.tp = tp;
892 	args.mp = ip->i_mount;
893 	xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0);
894 	args.firstblock = *firstblock;
895 	/*
896 	 * Allocate a block.  We know we need only one, since the
897 	 * file currently fits in an inode.
898 	 */
899 	if (*firstblock == NULLFSBLOCK) {
900 		args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
901 		args.type = XFS_ALLOCTYPE_START_BNO;
902 	} else {
903 		args.fsbno = *firstblock;
904 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
905 	}
906 	args.total = total;
907 	args.minlen = args.maxlen = args.prod = 1;
908 	error = xfs_alloc_vextent(&args);
909 	if (error)
910 		goto done;
911 
912 	/* Can't fail, the space was reserved. */
913 	ASSERT(args.fsbno != NULLFSBLOCK);
914 	ASSERT(args.len == 1);
915 	*firstblock = args.fsbno;
916 	bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
917 
918 	/*
919 	 * Initialize the block, copy the data and log the remote buffer.
920 	 *
921 	 * The callout is responsible for logging because the remote format
922 	 * might differ from the local format and thus we don't know how much to
923 	 * log here. Note that init_fn must also set the buffer log item type
924 	 * correctly.
925 	 */
926 	init_fn(tp, bp, ip, ifp);
927 
928 	/* account for the change in fork size */
929 	xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
930 	xfs_bmap_local_to_extents_empty(ip, whichfork);
931 	flags |= XFS_ILOG_CORE;
932 
933 	rec.br_startoff = 0;
934 	rec.br_startblock = args.fsbno;
935 	rec.br_blockcount = 1;
936 	rec.br_state = XFS_EXT_NORM;
937 	xfs_iext_insert(ip, 0, 1, &rec, 0);
938 
939 	trace_xfs_bmap_post_update(ip, 0,
940 			whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
941 			_THIS_IP_);
942 	XFS_IFORK_NEXT_SET(ip, whichfork, 1);
943 	ip->i_d.di_nblocks = 1;
944 	xfs_trans_mod_dquot_byino(tp, ip,
945 		XFS_TRANS_DQ_BCOUNT, 1L);
946 	flags |= xfs_ilog_fext(whichfork);
947 
948 done:
949 	*logflagsp = flags;
950 	return error;
951 }
952 
953 /*
954  * Called from xfs_bmap_add_attrfork to handle btree format files.
955  */
956 STATIC int					/* error */
957 xfs_bmap_add_attrfork_btree(
958 	xfs_trans_t		*tp,		/* transaction pointer */
959 	xfs_inode_t		*ip,		/* incore inode pointer */
960 	xfs_fsblock_t		*firstblock,	/* first block allocated */
961 	struct xfs_defer_ops	*dfops,		/* blocks to free at commit */
962 	int			*flags)		/* inode logging flags */
963 {
964 	xfs_btree_cur_t		*cur;		/* btree cursor */
965 	int			error;		/* error return value */
966 	xfs_mount_t		*mp;		/* file system mount struct */
967 	int			stat;		/* newroot status */
968 
969 	mp = ip->i_mount;
970 	if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
971 		*flags |= XFS_ILOG_DBROOT;
972 	else {
973 		cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
974 		cur->bc_private.b.dfops = dfops;
975 		cur->bc_private.b.firstblock = *firstblock;
976 		if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
977 			goto error0;
978 		/* must be at least one entry */
979 		XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0);
980 		if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
981 			goto error0;
982 		if (stat == 0) {
983 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
984 			return -ENOSPC;
985 		}
986 		*firstblock = cur->bc_private.b.firstblock;
987 		cur->bc_private.b.allocated = 0;
988 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
989 	}
990 	return 0;
991 error0:
992 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
993 	return error;
994 }
995 
996 /*
997  * Called from xfs_bmap_add_attrfork to handle extents format files.
998  */
999 STATIC int					/* error */
1000 xfs_bmap_add_attrfork_extents(
1001 	xfs_trans_t		*tp,		/* transaction pointer */
1002 	xfs_inode_t		*ip,		/* incore inode pointer */
1003 	xfs_fsblock_t		*firstblock,	/* first block allocated */
1004 	struct xfs_defer_ops	*dfops,		/* blocks to free at commit */
1005 	int			*flags)		/* inode logging flags */
1006 {
1007 	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
1008 	int			error;		/* error return value */
1009 
1010 	if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1011 		return 0;
1012 	cur = NULL;
1013 	error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops, &cur, 0,
1014 		flags, XFS_DATA_FORK);
1015 	if (cur) {
1016 		cur->bc_private.b.allocated = 0;
1017 		xfs_btree_del_cursor(cur,
1018 			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1019 	}
1020 	return error;
1021 }
1022 
1023 /*
1024  * Called from xfs_bmap_add_attrfork to handle local format files. Each
1025  * different data fork content type needs a different callout to do the
1026  * conversion. Some are basic and only require special block initialisation
1027  * callouts for the data formating, others (directories) are so specialised they
1028  * handle everything themselves.
1029  *
1030  * XXX (dgc): investigate whether directory conversion can use the generic
1031  * formatting callout. It should be possible - it's just a very complex
1032  * formatter.
1033  */
1034 STATIC int					/* error */
1035 xfs_bmap_add_attrfork_local(
1036 	xfs_trans_t		*tp,		/* transaction pointer */
1037 	xfs_inode_t		*ip,		/* incore inode pointer */
1038 	xfs_fsblock_t		*firstblock,	/* first block allocated */
1039 	struct xfs_defer_ops	*dfops,		/* blocks to free at commit */
1040 	int			*flags)		/* inode logging flags */
1041 {
1042 	xfs_da_args_t		dargs;		/* args for dir/attr code */
1043 
1044 	if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1045 		return 0;
1046 
1047 	if (S_ISDIR(VFS_I(ip)->i_mode)) {
1048 		memset(&dargs, 0, sizeof(dargs));
1049 		dargs.geo = ip->i_mount->m_dir_geo;
1050 		dargs.dp = ip;
1051 		dargs.firstblock = firstblock;
1052 		dargs.dfops = dfops;
1053 		dargs.total = dargs.geo->fsbcount;
1054 		dargs.whichfork = XFS_DATA_FORK;
1055 		dargs.trans = tp;
1056 		return xfs_dir2_sf_to_block(&dargs);
1057 	}
1058 
1059 	if (S_ISLNK(VFS_I(ip)->i_mode))
1060 		return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1061 						 flags, XFS_DATA_FORK,
1062 						 xfs_symlink_local_to_remote);
1063 
1064 	/* should only be called for types that support local format data */
1065 	ASSERT(0);
1066 	return -EFSCORRUPTED;
1067 }
1068 
1069 /*
1070  * Convert inode from non-attributed to attributed.
1071  * Must not be in a transaction, ip must not be locked.
1072  */
1073 int						/* error code */
1074 xfs_bmap_add_attrfork(
1075 	xfs_inode_t		*ip,		/* incore inode pointer */
1076 	int			size,		/* space new attribute needs */
1077 	int			rsvd)		/* xact may use reserved blks */
1078 {
1079 	xfs_fsblock_t		firstblock;	/* 1st block/ag allocated */
1080 	struct xfs_defer_ops	dfops;		/* freed extent records */
1081 	xfs_mount_t		*mp;		/* mount structure */
1082 	xfs_trans_t		*tp;		/* transaction pointer */
1083 	int			blks;		/* space reservation */
1084 	int			version = 1;	/* superblock attr version */
1085 	int			logflags;	/* logging flags */
1086 	int			error;		/* error return value */
1087 
1088 	ASSERT(XFS_IFORK_Q(ip) == 0);
1089 
1090 	mp = ip->i_mount;
1091 	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1092 
1093 	blks = XFS_ADDAFORK_SPACE_RES(mp);
1094 
1095 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_addafork, blks, 0,
1096 			rsvd ? XFS_TRANS_RESERVE : 0, &tp);
1097 	if (error)
1098 		return error;
1099 
1100 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1101 	error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1102 			XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1103 			XFS_QMOPT_RES_REGBLKS);
1104 	if (error)
1105 		goto trans_cancel;
1106 	if (XFS_IFORK_Q(ip))
1107 		goto trans_cancel;
1108 	if (ip->i_d.di_anextents != 0) {
1109 		error = -EFSCORRUPTED;
1110 		goto trans_cancel;
1111 	}
1112 	if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1113 		/*
1114 		 * For inodes coming from pre-6.2 filesystems.
1115 		 */
1116 		ASSERT(ip->i_d.di_aformat == 0);
1117 		ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1118 	}
1119 
1120 	xfs_trans_ijoin(tp, ip, 0);
1121 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1122 
1123 	switch (ip->i_d.di_format) {
1124 	case XFS_DINODE_FMT_DEV:
1125 		ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1126 		break;
1127 	case XFS_DINODE_FMT_UUID:
1128 		ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1129 		break;
1130 	case XFS_DINODE_FMT_LOCAL:
1131 	case XFS_DINODE_FMT_EXTENTS:
1132 	case XFS_DINODE_FMT_BTREE:
1133 		ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1134 		if (!ip->i_d.di_forkoff)
1135 			ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1136 		else if (mp->m_flags & XFS_MOUNT_ATTR2)
1137 			version = 2;
1138 		break;
1139 	default:
1140 		ASSERT(0);
1141 		error = -EINVAL;
1142 		goto trans_cancel;
1143 	}
1144 
1145 	ASSERT(ip->i_afp == NULL);
1146 	ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1147 	ip->i_afp->if_flags = XFS_IFEXTENTS;
1148 	logflags = 0;
1149 	xfs_defer_init(&dfops, &firstblock);
1150 	switch (ip->i_d.di_format) {
1151 	case XFS_DINODE_FMT_LOCAL:
1152 		error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &dfops,
1153 			&logflags);
1154 		break;
1155 	case XFS_DINODE_FMT_EXTENTS:
1156 		error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1157 			&dfops, &logflags);
1158 		break;
1159 	case XFS_DINODE_FMT_BTREE:
1160 		error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &dfops,
1161 			&logflags);
1162 		break;
1163 	default:
1164 		error = 0;
1165 		break;
1166 	}
1167 	if (logflags)
1168 		xfs_trans_log_inode(tp, ip, logflags);
1169 	if (error)
1170 		goto bmap_cancel;
1171 	if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1172 	   (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1173 		bool log_sb = false;
1174 
1175 		spin_lock(&mp->m_sb_lock);
1176 		if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1177 			xfs_sb_version_addattr(&mp->m_sb);
1178 			log_sb = true;
1179 		}
1180 		if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1181 			xfs_sb_version_addattr2(&mp->m_sb);
1182 			log_sb = true;
1183 		}
1184 		spin_unlock(&mp->m_sb_lock);
1185 		if (log_sb)
1186 			xfs_log_sb(tp);
1187 	}
1188 
1189 	error = xfs_defer_finish(&tp, &dfops);
1190 	if (error)
1191 		goto bmap_cancel;
1192 	error = xfs_trans_commit(tp);
1193 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1194 	return error;
1195 
1196 bmap_cancel:
1197 	xfs_defer_cancel(&dfops);
1198 trans_cancel:
1199 	xfs_trans_cancel(tp);
1200 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1201 	return error;
1202 }
1203 
1204 /*
1205  * Internal and external extent tree search functions.
1206  */
1207 
1208 /*
1209  * Read in the extents to if_extents.
1210  * All inode fields are set up by caller, we just traverse the btree
1211  * and copy the records in. If the file system cannot contain unwritten
1212  * extents, the records are checked for no "state" flags.
1213  */
1214 int					/* error */
1215 xfs_bmap_read_extents(
1216 	xfs_trans_t		*tp,	/* transaction pointer */
1217 	xfs_inode_t		*ip,	/* incore inode */
1218 	int			whichfork) /* data or attr fork */
1219 {
1220 	struct xfs_btree_block	*block;	/* current btree block */
1221 	xfs_fsblock_t		bno;	/* block # of "block" */
1222 	xfs_buf_t		*bp;	/* buffer for "block" */
1223 	int			error;	/* error return value */
1224 	xfs_extnum_t		i, j;	/* index into the extents list */
1225 	xfs_ifork_t		*ifp;	/* fork structure */
1226 	int			level;	/* btree level, for checking */
1227 	xfs_mount_t		*mp;	/* file system mount structure */
1228 	__be64			*pp;	/* pointer to block address */
1229 	/* REFERENCED */
1230 	xfs_extnum_t		room;	/* number of entries there's room for */
1231 
1232 	mp = ip->i_mount;
1233 	ifp = XFS_IFORK_PTR(ip, whichfork);
1234 	block = ifp->if_broot;
1235 	/*
1236 	 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1237 	 */
1238 	level = be16_to_cpu(block->bb_level);
1239 	ASSERT(level > 0);
1240 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1241 	bno = be64_to_cpu(*pp);
1242 
1243 	/*
1244 	 * Go down the tree until leaf level is reached, following the first
1245 	 * pointer (leftmost) at each level.
1246 	 */
1247 	while (level-- > 0) {
1248 		error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1249 				XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1250 		if (error)
1251 			return error;
1252 		block = XFS_BUF_TO_BLOCK(bp);
1253 		if (level == 0)
1254 			break;
1255 		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1256 		bno = be64_to_cpu(*pp);
1257 		XFS_WANT_CORRUPTED_GOTO(mp,
1258 			XFS_FSB_SANITY_CHECK(mp, bno), error0);
1259 		xfs_trans_brelse(tp, bp);
1260 	}
1261 	/*
1262 	 * Here with bp and block set to the leftmost leaf node in the tree.
1263 	 */
1264 	room = xfs_iext_count(ifp);
1265 	i = 0;
1266 	/*
1267 	 * Loop over all leaf nodes.  Copy information to the extent records.
1268 	 */
1269 	for (;;) {
1270 		xfs_bmbt_rec_t	*frp;
1271 		xfs_fsblock_t	nextbno;
1272 		xfs_extnum_t	num_recs;
1273 
1274 		num_recs = xfs_btree_get_numrecs(block);
1275 		if (unlikely(i + num_recs > room)) {
1276 			ASSERT(i + num_recs <= room);
1277 			xfs_warn(ip->i_mount,
1278 				"corrupt dinode %Lu, (btree extents).",
1279 				(unsigned long long) ip->i_ino);
1280 			XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1281 				XFS_ERRLEVEL_LOW, ip->i_mount, block);
1282 			goto error0;
1283 		}
1284 		/*
1285 		 * Read-ahead the next leaf block, if any.
1286 		 */
1287 		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1288 		if (nextbno != NULLFSBLOCK)
1289 			xfs_btree_reada_bufl(mp, nextbno, 1,
1290 					     &xfs_bmbt_buf_ops);
1291 		/*
1292 		 * Copy records into the extent records.
1293 		 */
1294 		frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1295 		for (j = 0; j < num_recs; j++, i++, frp++) {
1296 			xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1297 			trp->l0 = be64_to_cpu(frp->l0);
1298 			trp->l1 = be64_to_cpu(frp->l1);
1299 			if (!xfs_bmbt_validate_extent(mp, whichfork, trp)) {
1300 				XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1301 						 XFS_ERRLEVEL_LOW, mp);
1302 				goto error0;
1303 			}
1304 		}
1305 		xfs_trans_brelse(tp, bp);
1306 		bno = nextbno;
1307 		/*
1308 		 * If we've reached the end, stop.
1309 		 */
1310 		if (bno == NULLFSBLOCK)
1311 			break;
1312 		error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1313 				XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1314 		if (error)
1315 			return error;
1316 		block = XFS_BUF_TO_BLOCK(bp);
1317 	}
1318 	if (i != XFS_IFORK_NEXTENTS(ip, whichfork))
1319 		return -EFSCORRUPTED;
1320 	ASSERT(i == xfs_iext_count(ifp));
1321 	XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1322 	return 0;
1323 error0:
1324 	xfs_trans_brelse(tp, bp);
1325 	return -EFSCORRUPTED;
1326 }
1327 
1328 /*
1329  * Returns the file-relative block number of the first unused block(s)
1330  * in the file with at least "len" logically contiguous blocks free.
1331  * This is the lowest-address hole if the file has holes, else the first block
1332  * past the end of file.
1333  * Return 0 if the file is currently local (in-inode).
1334  */
1335 int						/* error */
1336 xfs_bmap_first_unused(
1337 	xfs_trans_t	*tp,			/* transaction pointer */
1338 	xfs_inode_t	*ip,			/* incore inode */
1339 	xfs_extlen_t	len,			/* size of hole to find */
1340 	xfs_fileoff_t	*first_unused,		/* unused block */
1341 	int		whichfork)		/* data or attr fork */
1342 {
1343 	int		error;			/* error return value */
1344 	int		idx;			/* extent record index */
1345 	xfs_ifork_t	*ifp;			/* inode fork pointer */
1346 	xfs_fileoff_t	lastaddr;		/* last block number seen */
1347 	xfs_fileoff_t	lowest;			/* lowest useful block */
1348 	xfs_fileoff_t	max;			/* starting useful block */
1349 	xfs_extnum_t	nextents;		/* number of extent entries */
1350 
1351 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1352 	       XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1353 	       XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1354 	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1355 		*first_unused = 0;
1356 		return 0;
1357 	}
1358 	ifp = XFS_IFORK_PTR(ip, whichfork);
1359 	if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1360 	    (error = xfs_iread_extents(tp, ip, whichfork)))
1361 		return error;
1362 	lowest = *first_unused;
1363 	nextents = xfs_iext_count(ifp);
1364 	for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1365 		struct xfs_bmbt_irec got;
1366 
1367 		xfs_iext_get_extent(ifp, idx, &got);
1368 
1369 		/*
1370 		 * See if the hole before this extent will work.
1371 		 */
1372 		if (got.br_startoff >= lowest + len &&
1373 		    got.br_startoff - max >= len) {
1374 			*first_unused = max;
1375 			return 0;
1376 		}
1377 		lastaddr = got.br_startoff + got.br_blockcount;
1378 		max = XFS_FILEOFF_MAX(lastaddr, lowest);
1379 	}
1380 	*first_unused = max;
1381 	return 0;
1382 }
1383 
1384 /*
1385  * Returns the file-relative block number of the last block - 1 before
1386  * last_block (input value) in the file.
1387  * This is not based on i_size, it is based on the extent records.
1388  * Returns 0 for local files, as they do not have extent records.
1389  */
1390 int						/* error */
1391 xfs_bmap_last_before(
1392 	struct xfs_trans	*tp,		/* transaction pointer */
1393 	struct xfs_inode	*ip,		/* incore inode */
1394 	xfs_fileoff_t		*last_block,	/* last block */
1395 	int			whichfork)	/* data or attr fork */
1396 {
1397 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1398 	struct xfs_bmbt_irec	got;
1399 	xfs_extnum_t		idx;
1400 	int			error;
1401 
1402 	switch (XFS_IFORK_FORMAT(ip, whichfork)) {
1403 	case XFS_DINODE_FMT_LOCAL:
1404 		*last_block = 0;
1405 		return 0;
1406 	case XFS_DINODE_FMT_BTREE:
1407 	case XFS_DINODE_FMT_EXTENTS:
1408 		break;
1409 	default:
1410 		return -EIO;
1411 	}
1412 
1413 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1414 		error = xfs_iread_extents(tp, ip, whichfork);
1415 		if (error)
1416 			return error;
1417 	}
1418 
1419 	if (xfs_iext_lookup_extent(ip, ifp, *last_block - 1, &idx, &got)) {
1420 		if (got.br_startoff <= *last_block - 1)
1421 			return 0;
1422 	}
1423 
1424 	if (xfs_iext_get_extent(ifp, idx - 1, &got)) {
1425 		*last_block = got.br_startoff + got.br_blockcount;
1426 		return 0;
1427 	}
1428 
1429 	*last_block = 0;
1430 	return 0;
1431 }
1432 
1433 int
1434 xfs_bmap_last_extent(
1435 	struct xfs_trans	*tp,
1436 	struct xfs_inode	*ip,
1437 	int			whichfork,
1438 	struct xfs_bmbt_irec	*rec,
1439 	int			*is_empty)
1440 {
1441 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1442 	int			error;
1443 	int			nextents;
1444 
1445 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1446 		error = xfs_iread_extents(tp, ip, whichfork);
1447 		if (error)
1448 			return error;
1449 	}
1450 
1451 	nextents = xfs_iext_count(ifp);
1452 	if (nextents == 0) {
1453 		*is_empty = 1;
1454 		return 0;
1455 	}
1456 
1457 	xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1458 	*is_empty = 0;
1459 	return 0;
1460 }
1461 
1462 /*
1463  * Check the last inode extent to determine whether this allocation will result
1464  * in blocks being allocated at the end of the file. When we allocate new data
1465  * blocks at the end of the file which do not start at the previous data block,
1466  * we will try to align the new blocks at stripe unit boundaries.
1467  *
1468  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1469  * at, or past the EOF.
1470  */
1471 STATIC int
1472 xfs_bmap_isaeof(
1473 	struct xfs_bmalloca	*bma,
1474 	int			whichfork)
1475 {
1476 	struct xfs_bmbt_irec	rec;
1477 	int			is_empty;
1478 	int			error;
1479 
1480 	bma->aeof = false;
1481 	error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1482 				     &is_empty);
1483 	if (error)
1484 		return error;
1485 
1486 	if (is_empty) {
1487 		bma->aeof = true;
1488 		return 0;
1489 	}
1490 
1491 	/*
1492 	 * Check if we are allocation or past the last extent, or at least into
1493 	 * the last delayed allocated extent.
1494 	 */
1495 	bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1496 		(bma->offset >= rec.br_startoff &&
1497 		 isnullstartblock(rec.br_startblock));
1498 	return 0;
1499 }
1500 
1501 /*
1502  * Returns the file-relative block number of the first block past eof in
1503  * the file.  This is not based on i_size, it is based on the extent records.
1504  * Returns 0 for local files, as they do not have extent records.
1505  */
1506 int
1507 xfs_bmap_last_offset(
1508 	struct xfs_inode	*ip,
1509 	xfs_fileoff_t		*last_block,
1510 	int			whichfork)
1511 {
1512 	struct xfs_bmbt_irec	rec;
1513 	int			is_empty;
1514 	int			error;
1515 
1516 	*last_block = 0;
1517 
1518 	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1519 		return 0;
1520 
1521 	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1522 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1523 	       return -EIO;
1524 
1525 	error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1526 	if (error || is_empty)
1527 		return error;
1528 
1529 	*last_block = rec.br_startoff + rec.br_blockcount;
1530 	return 0;
1531 }
1532 
1533 /*
1534  * Returns whether the selected fork of the inode has exactly one
1535  * block or not.  For the data fork we check this matches di_size,
1536  * implying the file's range is 0..bsize-1.
1537  */
1538 int					/* 1=>1 block, 0=>otherwise */
1539 xfs_bmap_one_block(
1540 	xfs_inode_t	*ip,		/* incore inode */
1541 	int		whichfork)	/* data or attr fork */
1542 {
1543 	xfs_bmbt_rec_host_t *ep;	/* ptr to fork's extent */
1544 	xfs_ifork_t	*ifp;		/* inode fork pointer */
1545 	int		rval;		/* return value */
1546 	xfs_bmbt_irec_t	s;		/* internal version of extent */
1547 
1548 #ifndef DEBUG
1549 	if (whichfork == XFS_DATA_FORK)
1550 		return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1551 #endif	/* !DEBUG */
1552 	if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1553 		return 0;
1554 	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1555 		return 0;
1556 	ifp = XFS_IFORK_PTR(ip, whichfork);
1557 	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1558 	ep = xfs_iext_get_ext(ifp, 0);
1559 	xfs_bmbt_get_all(ep, &s);
1560 	rval = s.br_startoff == 0 && s.br_blockcount == 1;
1561 	if (rval && whichfork == XFS_DATA_FORK)
1562 		ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1563 	return rval;
1564 }
1565 
1566 /*
1567  * Extent tree manipulation functions used during allocation.
1568  */
1569 
1570 /*
1571  * Convert a delayed allocation to a real allocation.
1572  */
1573 STATIC int				/* error */
1574 xfs_bmap_add_extent_delay_real(
1575 	struct xfs_bmalloca	*bma,
1576 	int			whichfork)
1577 {
1578 	struct xfs_bmbt_irec	*new = &bma->got;
1579 	int			diff;	/* temp value */
1580 	xfs_bmbt_rec_host_t	*ep;	/* extent entry for idx */
1581 	int			error;	/* error return value */
1582 	int			i;	/* temp state */
1583 	xfs_ifork_t		*ifp;	/* inode fork pointer */
1584 	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
1585 	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
1586 					/* left is 0, right is 1, prev is 2 */
1587 	int			rval=0;	/* return value (logging flags) */
1588 	int			state = 0;/* state bits, accessed thru macros */
1589 	xfs_filblks_t		da_new; /* new count del alloc blocks used */
1590 	xfs_filblks_t		da_old; /* old count del alloc blocks used */
1591 	xfs_filblks_t		temp=0;	/* value for da_new calculations */
1592 	xfs_filblks_t		temp2=0;/* value for da_new calculations */
1593 	int			tmp_rval;	/* partial logging flags */
1594 	struct xfs_mount	*mp;
1595 	xfs_extnum_t		*nextents;
1596 
1597 	mp = bma->ip->i_mount;
1598 	ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1599 	ASSERT(whichfork != XFS_ATTR_FORK);
1600 	nextents = (whichfork == XFS_COW_FORK ? &bma->ip->i_cnextents :
1601 						&bma->ip->i_d.di_nextents);
1602 
1603 	ASSERT(bma->idx >= 0);
1604 	ASSERT(bma->idx <= xfs_iext_count(ifp));
1605 	ASSERT(!isnullstartblock(new->br_startblock));
1606 	ASSERT(!bma->cur ||
1607 	       (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1608 
1609 	XFS_STATS_INC(mp, xs_add_exlist);
1610 
1611 #define	LEFT		r[0]
1612 #define	RIGHT		r[1]
1613 #define	PREV		r[2]
1614 
1615 	if (whichfork == XFS_COW_FORK)
1616 		state |= BMAP_COWFORK;
1617 
1618 	/*
1619 	 * Set up a bunch of variables to make the tests simpler.
1620 	 */
1621 	ep = xfs_iext_get_ext(ifp, bma->idx);
1622 	xfs_bmbt_get_all(ep, &PREV);
1623 	new_endoff = new->br_startoff + new->br_blockcount;
1624 	ASSERT(PREV.br_startoff <= new->br_startoff);
1625 	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1626 
1627 	da_old = startblockval(PREV.br_startblock);
1628 	da_new = 0;
1629 
1630 	/*
1631 	 * Set flags determining what part of the previous delayed allocation
1632 	 * extent is being replaced by a real allocation.
1633 	 */
1634 	if (PREV.br_startoff == new->br_startoff)
1635 		state |= BMAP_LEFT_FILLING;
1636 	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1637 		state |= BMAP_RIGHT_FILLING;
1638 
1639 	/*
1640 	 * Check and set flags if this segment has a left neighbor.
1641 	 * Don't set contiguous if the combined extent would be too large.
1642 	 */
1643 	if (bma->idx > 0) {
1644 		state |= BMAP_LEFT_VALID;
1645 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
1646 
1647 		if (isnullstartblock(LEFT.br_startblock))
1648 			state |= BMAP_LEFT_DELAY;
1649 	}
1650 
1651 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1652 	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1653 	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1654 	    LEFT.br_state == new->br_state &&
1655 	    LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1656 		state |= BMAP_LEFT_CONTIG;
1657 
1658 	/*
1659 	 * Check and set flags if this segment has a right neighbor.
1660 	 * Don't set contiguous if the combined extent would be too large.
1661 	 * Also check for all-three-contiguous being too large.
1662 	 */
1663 	if (bma->idx < xfs_iext_count(ifp) - 1) {
1664 		state |= BMAP_RIGHT_VALID;
1665 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
1666 
1667 		if (isnullstartblock(RIGHT.br_startblock))
1668 			state |= BMAP_RIGHT_DELAY;
1669 	}
1670 
1671 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1672 	    new_endoff == RIGHT.br_startoff &&
1673 	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1674 	    new->br_state == RIGHT.br_state &&
1675 	    new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1676 	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1677 		       BMAP_RIGHT_FILLING)) !=
1678 		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1679 		       BMAP_RIGHT_FILLING) ||
1680 	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1681 			<= MAXEXTLEN))
1682 		state |= BMAP_RIGHT_CONTIG;
1683 
1684 	error = 0;
1685 	/*
1686 	 * Switch out based on the FILLING and CONTIG state bits.
1687 	 */
1688 	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1689 			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1690 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1691 	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1692 		/*
1693 		 * Filling in all of a previously delayed allocation extent.
1694 		 * The left and right neighbors are both contiguous with new.
1695 		 */
1696 		bma->idx--;
1697 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1698 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1699 			LEFT.br_blockcount + PREV.br_blockcount +
1700 			RIGHT.br_blockcount);
1701 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1702 
1703 		xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1704 		(*nextents)--;
1705 		if (bma->cur == NULL)
1706 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1707 		else {
1708 			rval = XFS_ILOG_CORE;
1709 			error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1710 					RIGHT.br_startblock,
1711 					RIGHT.br_blockcount, &i);
1712 			if (error)
1713 				goto done;
1714 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1715 			error = xfs_btree_delete(bma->cur, &i);
1716 			if (error)
1717 				goto done;
1718 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1719 			error = xfs_btree_decrement(bma->cur, 0, &i);
1720 			if (error)
1721 				goto done;
1722 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1723 			error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1724 					LEFT.br_startblock,
1725 					LEFT.br_blockcount +
1726 					PREV.br_blockcount +
1727 					RIGHT.br_blockcount, LEFT.br_state);
1728 			if (error)
1729 				goto done;
1730 		}
1731 		break;
1732 
1733 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1734 		/*
1735 		 * Filling in all of a previously delayed allocation extent.
1736 		 * The left neighbor is contiguous, the right is not.
1737 		 */
1738 		bma->idx--;
1739 
1740 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1741 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1742 			LEFT.br_blockcount + PREV.br_blockcount);
1743 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1744 
1745 		xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1746 		if (bma->cur == NULL)
1747 			rval = XFS_ILOG_DEXT;
1748 		else {
1749 			rval = 0;
1750 			error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1751 					LEFT.br_startblock, LEFT.br_blockcount,
1752 					&i);
1753 			if (error)
1754 				goto done;
1755 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1756 			error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1757 					LEFT.br_startblock,
1758 					LEFT.br_blockcount +
1759 					PREV.br_blockcount, LEFT.br_state);
1760 			if (error)
1761 				goto done;
1762 		}
1763 		break;
1764 
1765 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1766 		/*
1767 		 * Filling in all of a previously delayed allocation extent.
1768 		 * The right neighbor is contiguous, the left is not.
1769 		 */
1770 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1771 		xfs_bmbt_set_startblock(ep, new->br_startblock);
1772 		xfs_bmbt_set_blockcount(ep,
1773 			PREV.br_blockcount + RIGHT.br_blockcount);
1774 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1775 
1776 		xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1777 		if (bma->cur == NULL)
1778 			rval = XFS_ILOG_DEXT;
1779 		else {
1780 			rval = 0;
1781 			error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1782 					RIGHT.br_startblock,
1783 					RIGHT.br_blockcount, &i);
1784 			if (error)
1785 				goto done;
1786 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1787 			error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1788 					new->br_startblock,
1789 					PREV.br_blockcount +
1790 					RIGHT.br_blockcount, PREV.br_state);
1791 			if (error)
1792 				goto done;
1793 		}
1794 		break;
1795 
1796 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1797 		/*
1798 		 * Filling in all of a previously delayed allocation extent.
1799 		 * Neither the left nor right neighbors are contiguous with
1800 		 * the new one.
1801 		 */
1802 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1803 		xfs_bmbt_set_startblock(ep, new->br_startblock);
1804 		xfs_bmbt_set_state(ep, new->br_state);
1805 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1806 
1807 		(*nextents)++;
1808 		if (bma->cur == NULL)
1809 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1810 		else {
1811 			rval = XFS_ILOG_CORE;
1812 			error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1813 					new->br_startblock, new->br_blockcount,
1814 					&i);
1815 			if (error)
1816 				goto done;
1817 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1818 			bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1819 			error = xfs_btree_insert(bma->cur, &i);
1820 			if (error)
1821 				goto done;
1822 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1823 		}
1824 		break;
1825 
1826 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1827 		/*
1828 		 * Filling in the first part of a previous delayed allocation.
1829 		 * The left neighbor is contiguous.
1830 		 */
1831 		trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1832 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1833 			LEFT.br_blockcount + new->br_blockcount);
1834 		xfs_bmbt_set_startoff(ep,
1835 			PREV.br_startoff + new->br_blockcount);
1836 		trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1837 
1838 		temp = PREV.br_blockcount - new->br_blockcount;
1839 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1840 		xfs_bmbt_set_blockcount(ep, temp);
1841 		if (bma->cur == NULL)
1842 			rval = XFS_ILOG_DEXT;
1843 		else {
1844 			rval = 0;
1845 			error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1846 					LEFT.br_startblock, LEFT.br_blockcount,
1847 					&i);
1848 			if (error)
1849 				goto done;
1850 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1851 			error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1852 					LEFT.br_startblock,
1853 					LEFT.br_blockcount +
1854 					new->br_blockcount,
1855 					LEFT.br_state);
1856 			if (error)
1857 				goto done;
1858 		}
1859 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1860 			startblockval(PREV.br_startblock));
1861 		xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1862 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1863 
1864 		bma->idx--;
1865 		break;
1866 
1867 	case BMAP_LEFT_FILLING:
1868 		/*
1869 		 * Filling in the first part of a previous delayed allocation.
1870 		 * The left neighbor is not contiguous.
1871 		 */
1872 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1873 		xfs_bmbt_set_startoff(ep, new_endoff);
1874 		temp = PREV.br_blockcount - new->br_blockcount;
1875 		xfs_bmbt_set_blockcount(ep, temp);
1876 		xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
1877 		(*nextents)++;
1878 		if (bma->cur == NULL)
1879 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1880 		else {
1881 			rval = XFS_ILOG_CORE;
1882 			error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1883 					new->br_startblock, new->br_blockcount,
1884 					&i);
1885 			if (error)
1886 				goto done;
1887 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1888 			bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1889 			error = xfs_btree_insert(bma->cur, &i);
1890 			if (error)
1891 				goto done;
1892 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1893 		}
1894 
1895 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1896 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1897 					bma->firstblock, bma->dfops,
1898 					&bma->cur, 1, &tmp_rval, whichfork);
1899 			rval |= tmp_rval;
1900 			if (error)
1901 				goto done;
1902 		}
1903 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1904 			startblockval(PREV.br_startblock) -
1905 			(bma->cur ? bma->cur->bc_private.b.allocated : 0));
1906 		ep = xfs_iext_get_ext(ifp, bma->idx + 1);
1907 		xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1908 		trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1909 		break;
1910 
1911 	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1912 		/*
1913 		 * Filling in the last part of a previous delayed allocation.
1914 		 * The right neighbor is contiguous with the new allocation.
1915 		 */
1916 		temp = PREV.br_blockcount - new->br_blockcount;
1917 		trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1918 		xfs_bmbt_set_blockcount(ep, temp);
1919 		xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
1920 			new->br_startoff, new->br_startblock,
1921 			new->br_blockcount + RIGHT.br_blockcount,
1922 			RIGHT.br_state);
1923 		trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1924 		if (bma->cur == NULL)
1925 			rval = XFS_ILOG_DEXT;
1926 		else {
1927 			rval = 0;
1928 			error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1929 					RIGHT.br_startblock,
1930 					RIGHT.br_blockcount, &i);
1931 			if (error)
1932 				goto done;
1933 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1934 			error = xfs_bmbt_update(bma->cur, new->br_startoff,
1935 					new->br_startblock,
1936 					new->br_blockcount +
1937 					RIGHT.br_blockcount,
1938 					RIGHT.br_state);
1939 			if (error)
1940 				goto done;
1941 		}
1942 
1943 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1944 			startblockval(PREV.br_startblock));
1945 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1946 		xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1947 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1948 
1949 		bma->idx++;
1950 		break;
1951 
1952 	case BMAP_RIGHT_FILLING:
1953 		/*
1954 		 * Filling in the last part of a previous delayed allocation.
1955 		 * The right neighbor is not contiguous.
1956 		 */
1957 		temp = PREV.br_blockcount - new->br_blockcount;
1958 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1959 		xfs_bmbt_set_blockcount(ep, temp);
1960 		xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
1961 		(*nextents)++;
1962 		if (bma->cur == NULL)
1963 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1964 		else {
1965 			rval = XFS_ILOG_CORE;
1966 			error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1967 					new->br_startblock, new->br_blockcount,
1968 					&i);
1969 			if (error)
1970 				goto done;
1971 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1972 			bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1973 			error = xfs_btree_insert(bma->cur, &i);
1974 			if (error)
1975 				goto done;
1976 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1977 		}
1978 
1979 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1980 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1981 				bma->firstblock, bma->dfops, &bma->cur, 1,
1982 				&tmp_rval, whichfork);
1983 			rval |= tmp_rval;
1984 			if (error)
1985 				goto done;
1986 		}
1987 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1988 			startblockval(PREV.br_startblock) -
1989 			(bma->cur ? bma->cur->bc_private.b.allocated : 0));
1990 		ep = xfs_iext_get_ext(ifp, bma->idx);
1991 		xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1992 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1993 
1994 		bma->idx++;
1995 		break;
1996 
1997 	case 0:
1998 		/*
1999 		 * Filling in the middle part of a previous delayed allocation.
2000 		 * Contiguity is impossible here.
2001 		 * This case is avoided almost all the time.
2002 		 *
2003 		 * We start with a delayed allocation:
2004 		 *
2005 		 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2006 		 *  PREV @ idx
2007 		 *
2008 	         * and we are allocating:
2009 		 *                     +rrrrrrrrrrrrrrrrr+
2010 		 *			      new
2011 		 *
2012 		 * and we set it up for insertion as:
2013 		 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2014 		 *                            new
2015 		 *  PREV @ idx          LEFT              RIGHT
2016 		 *                      inserted at idx + 1
2017 		 */
2018 		temp = new->br_startoff - PREV.br_startoff;
2019 		temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2020 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2021 		xfs_bmbt_set_blockcount(ep, temp);	/* truncate PREV */
2022 		LEFT = *new;
2023 		RIGHT.br_state = PREV.br_state;
2024 		RIGHT.br_startblock = nullstartblock(
2025 				(int)xfs_bmap_worst_indlen(bma->ip, temp2));
2026 		RIGHT.br_startoff = new_endoff;
2027 		RIGHT.br_blockcount = temp2;
2028 		/* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2029 		xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2030 		(*nextents)++;
2031 		if (bma->cur == NULL)
2032 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2033 		else {
2034 			rval = XFS_ILOG_CORE;
2035 			error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2036 					new->br_startblock, new->br_blockcount,
2037 					&i);
2038 			if (error)
2039 				goto done;
2040 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2041 			bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2042 			error = xfs_btree_insert(bma->cur, &i);
2043 			if (error)
2044 				goto done;
2045 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2046 		}
2047 
2048 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2049 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2050 					bma->firstblock, bma->dfops, &bma->cur,
2051 					1, &tmp_rval, whichfork);
2052 			rval |= tmp_rval;
2053 			if (error)
2054 				goto done;
2055 		}
2056 		temp = xfs_bmap_worst_indlen(bma->ip, temp);
2057 		temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2058 		diff = (int)(temp + temp2 -
2059 			     (startblockval(PREV.br_startblock) -
2060 			      (bma->cur ?
2061 			       bma->cur->bc_private.b.allocated : 0)));
2062 		if (diff > 0) {
2063 			error = xfs_mod_fdblocks(bma->ip->i_mount,
2064 						 -((int64_t)diff), false);
2065 			ASSERT(!error);
2066 			if (error)
2067 				goto done;
2068 		}
2069 
2070 		ep = xfs_iext_get_ext(ifp, bma->idx);
2071 		xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2072 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2073 		trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2074 		xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2075 			nullstartblock((int)temp2));
2076 		trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2077 
2078 		bma->idx++;
2079 		da_new = temp + temp2;
2080 		break;
2081 
2082 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2083 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2084 	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2085 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2086 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2087 	case BMAP_LEFT_CONTIG:
2088 	case BMAP_RIGHT_CONTIG:
2089 		/*
2090 		 * These cases are all impossible.
2091 		 */
2092 		ASSERT(0);
2093 	}
2094 
2095 	/* add reverse mapping */
2096 	error = xfs_rmap_map_extent(mp, bma->dfops, bma->ip, whichfork, new);
2097 	if (error)
2098 		goto done;
2099 
2100 	/* convert to a btree if necessary */
2101 	if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2102 		int	tmp_logflags;	/* partial log flag return val */
2103 
2104 		ASSERT(bma->cur == NULL);
2105 		error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2106 				bma->firstblock, bma->dfops, &bma->cur,
2107 				da_old > 0, &tmp_logflags, whichfork);
2108 		bma->logflags |= tmp_logflags;
2109 		if (error)
2110 			goto done;
2111 	}
2112 
2113 	/* adjust for changes in reserved delayed indirect blocks */
2114 	if (da_old || da_new) {
2115 		temp = da_new;
2116 		if (bma->cur)
2117 			temp += bma->cur->bc_private.b.allocated;
2118 		if (temp < da_old)
2119 			xfs_mod_fdblocks(bma->ip->i_mount,
2120 					(int64_t)(da_old - temp), false);
2121 	}
2122 
2123 	/* clear out the allocated field, done with it now in any case. */
2124 	if (bma->cur)
2125 		bma->cur->bc_private.b.allocated = 0;
2126 
2127 	xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
2128 done:
2129 	if (whichfork != XFS_COW_FORK)
2130 		bma->logflags |= rval;
2131 	return error;
2132 #undef	LEFT
2133 #undef	RIGHT
2134 #undef	PREV
2135 }
2136 
2137 /*
2138  * Convert an unwritten allocation to a real allocation or vice versa.
2139  */
2140 STATIC int				/* error */
2141 xfs_bmap_add_extent_unwritten_real(
2142 	struct xfs_trans	*tp,
2143 	xfs_inode_t		*ip,	/* incore inode pointer */
2144 	int			whichfork,
2145 	xfs_extnum_t		*idx,	/* extent number to update/insert */
2146 	xfs_btree_cur_t		**curp,	/* if *curp is null, not a btree */
2147 	xfs_bmbt_irec_t		*new,	/* new data to add to file extents */
2148 	xfs_fsblock_t		*first,	/* pointer to firstblock variable */
2149 	struct xfs_defer_ops	*dfops,	/* list of extents to be freed */
2150 	int			*logflagsp) /* inode logging flags */
2151 {
2152 	xfs_btree_cur_t		*cur;	/* btree cursor */
2153 	xfs_bmbt_rec_host_t	*ep;	/* extent entry for idx */
2154 	int			error;	/* error return value */
2155 	int			i;	/* temp state */
2156 	xfs_ifork_t		*ifp;	/* inode fork pointer */
2157 	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
2158 	xfs_exntst_t		newext;	/* new extent state */
2159 	xfs_exntst_t		oldext;	/* old extent state */
2160 	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
2161 					/* left is 0, right is 1, prev is 2 */
2162 	int			rval=0;	/* return value (logging flags) */
2163 	int			state = 0;/* state bits, accessed thru macros */
2164 	struct xfs_mount	*mp = ip->i_mount;
2165 
2166 	*logflagsp = 0;
2167 
2168 	cur = *curp;
2169 	ifp = XFS_IFORK_PTR(ip, whichfork);
2170 	if (whichfork == XFS_COW_FORK)
2171 		state |= BMAP_COWFORK;
2172 
2173 	ASSERT(*idx >= 0);
2174 	ASSERT(*idx <= xfs_iext_count(ifp));
2175 	ASSERT(!isnullstartblock(new->br_startblock));
2176 
2177 	XFS_STATS_INC(mp, xs_add_exlist);
2178 
2179 #define	LEFT		r[0]
2180 #define	RIGHT		r[1]
2181 #define	PREV		r[2]
2182 
2183 	/*
2184 	 * Set up a bunch of variables to make the tests simpler.
2185 	 */
2186 	error = 0;
2187 	ep = xfs_iext_get_ext(ifp, *idx);
2188 	xfs_bmbt_get_all(ep, &PREV);
2189 	newext = new->br_state;
2190 	oldext = (newext == XFS_EXT_UNWRITTEN) ?
2191 		XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2192 	ASSERT(PREV.br_state == oldext);
2193 	new_endoff = new->br_startoff + new->br_blockcount;
2194 	ASSERT(PREV.br_startoff <= new->br_startoff);
2195 	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2196 
2197 	/*
2198 	 * Set flags determining what part of the previous oldext allocation
2199 	 * extent is being replaced by a newext allocation.
2200 	 */
2201 	if (PREV.br_startoff == new->br_startoff)
2202 		state |= BMAP_LEFT_FILLING;
2203 	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2204 		state |= BMAP_RIGHT_FILLING;
2205 
2206 	/*
2207 	 * Check and set flags if this segment has a left neighbor.
2208 	 * Don't set contiguous if the combined extent would be too large.
2209 	 */
2210 	if (*idx > 0) {
2211 		state |= BMAP_LEFT_VALID;
2212 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2213 
2214 		if (isnullstartblock(LEFT.br_startblock))
2215 			state |= BMAP_LEFT_DELAY;
2216 	}
2217 
2218 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2219 	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2220 	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2221 	    LEFT.br_state == newext &&
2222 	    LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2223 		state |= BMAP_LEFT_CONTIG;
2224 
2225 	/*
2226 	 * Check and set flags if this segment has a right neighbor.
2227 	 * Don't set contiguous if the combined extent would be too large.
2228 	 * Also check for all-three-contiguous being too large.
2229 	 */
2230 	if (*idx < xfs_iext_count(ifp) - 1) {
2231 		state |= BMAP_RIGHT_VALID;
2232 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2233 		if (isnullstartblock(RIGHT.br_startblock))
2234 			state |= BMAP_RIGHT_DELAY;
2235 	}
2236 
2237 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2238 	    new_endoff == RIGHT.br_startoff &&
2239 	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2240 	    newext == RIGHT.br_state &&
2241 	    new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2242 	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2243 		       BMAP_RIGHT_FILLING)) !=
2244 		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2245 		       BMAP_RIGHT_FILLING) ||
2246 	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2247 			<= MAXEXTLEN))
2248 		state |= BMAP_RIGHT_CONTIG;
2249 
2250 	/*
2251 	 * Switch out based on the FILLING and CONTIG state bits.
2252 	 */
2253 	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2254 			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2255 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2256 	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2257 		/*
2258 		 * Setting all of a previous oldext extent to newext.
2259 		 * The left and right neighbors are both contiguous with new.
2260 		 */
2261 		--*idx;
2262 
2263 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2264 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2265 			LEFT.br_blockcount + PREV.br_blockcount +
2266 			RIGHT.br_blockcount);
2267 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2268 
2269 		xfs_iext_remove(ip, *idx + 1, 2, state);
2270 		XFS_IFORK_NEXT_SET(ip, whichfork,
2271 				XFS_IFORK_NEXTENTS(ip, whichfork) - 2);
2272 		if (cur == NULL)
2273 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2274 		else {
2275 			rval = XFS_ILOG_CORE;
2276 			if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2277 					RIGHT.br_startblock,
2278 					RIGHT.br_blockcount, &i)))
2279 				goto done;
2280 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2281 			if ((error = xfs_btree_delete(cur, &i)))
2282 				goto done;
2283 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2284 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2285 				goto done;
2286 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2287 			if ((error = xfs_btree_delete(cur, &i)))
2288 				goto done;
2289 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2290 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2291 				goto done;
2292 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2293 			if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2294 				LEFT.br_startblock,
2295 				LEFT.br_blockcount + PREV.br_blockcount +
2296 				RIGHT.br_blockcount, LEFT.br_state)))
2297 				goto done;
2298 		}
2299 		break;
2300 
2301 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2302 		/*
2303 		 * Setting all of a previous oldext extent to newext.
2304 		 * The left neighbor is contiguous, the right is not.
2305 		 */
2306 		--*idx;
2307 
2308 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2309 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2310 			LEFT.br_blockcount + PREV.br_blockcount);
2311 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2312 
2313 		xfs_iext_remove(ip, *idx + 1, 1, state);
2314 		XFS_IFORK_NEXT_SET(ip, whichfork,
2315 				XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2316 		if (cur == NULL)
2317 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2318 		else {
2319 			rval = XFS_ILOG_CORE;
2320 			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2321 					PREV.br_startblock, PREV.br_blockcount,
2322 					&i)))
2323 				goto done;
2324 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2325 			if ((error = xfs_btree_delete(cur, &i)))
2326 				goto done;
2327 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2328 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2329 				goto done;
2330 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2331 			if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2332 				LEFT.br_startblock,
2333 				LEFT.br_blockcount + PREV.br_blockcount,
2334 				LEFT.br_state)))
2335 				goto done;
2336 		}
2337 		break;
2338 
2339 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2340 		/*
2341 		 * Setting all of a previous oldext extent to newext.
2342 		 * The right neighbor is contiguous, the left is not.
2343 		 */
2344 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2345 		xfs_bmbt_set_blockcount(ep,
2346 			PREV.br_blockcount + RIGHT.br_blockcount);
2347 		xfs_bmbt_set_state(ep, newext);
2348 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2349 		xfs_iext_remove(ip, *idx + 1, 1, state);
2350 		XFS_IFORK_NEXT_SET(ip, whichfork,
2351 				XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2352 		if (cur == NULL)
2353 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2354 		else {
2355 			rval = XFS_ILOG_CORE;
2356 			if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2357 					RIGHT.br_startblock,
2358 					RIGHT.br_blockcount, &i)))
2359 				goto done;
2360 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2361 			if ((error = xfs_btree_delete(cur, &i)))
2362 				goto done;
2363 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2364 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2365 				goto done;
2366 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2367 			if ((error = xfs_bmbt_update(cur, new->br_startoff,
2368 				new->br_startblock,
2369 				new->br_blockcount + RIGHT.br_blockcount,
2370 				newext)))
2371 				goto done;
2372 		}
2373 		break;
2374 
2375 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2376 		/*
2377 		 * Setting all of a previous oldext extent to newext.
2378 		 * Neither the left nor right neighbors are contiguous with
2379 		 * the new one.
2380 		 */
2381 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2382 		xfs_bmbt_set_state(ep, newext);
2383 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2384 
2385 		if (cur == NULL)
2386 			rval = XFS_ILOG_DEXT;
2387 		else {
2388 			rval = 0;
2389 			if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2390 					new->br_startblock, new->br_blockcount,
2391 					&i)))
2392 				goto done;
2393 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2394 			if ((error = xfs_bmbt_update(cur, new->br_startoff,
2395 				new->br_startblock, new->br_blockcount,
2396 				newext)))
2397 				goto done;
2398 		}
2399 		break;
2400 
2401 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2402 		/*
2403 		 * Setting the first part of a previous oldext extent to newext.
2404 		 * The left neighbor is contiguous.
2405 		 */
2406 		trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2407 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2408 			LEFT.br_blockcount + new->br_blockcount);
2409 		xfs_bmbt_set_startoff(ep,
2410 			PREV.br_startoff + new->br_blockcount);
2411 		trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2412 
2413 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2414 		xfs_bmbt_set_startblock(ep,
2415 			new->br_startblock + new->br_blockcount);
2416 		xfs_bmbt_set_blockcount(ep,
2417 			PREV.br_blockcount - new->br_blockcount);
2418 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2419 
2420 		--*idx;
2421 
2422 		if (cur == NULL)
2423 			rval = XFS_ILOG_DEXT;
2424 		else {
2425 			rval = 0;
2426 			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2427 					PREV.br_startblock, PREV.br_blockcount,
2428 					&i)))
2429 				goto done;
2430 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2431 			if ((error = xfs_bmbt_update(cur,
2432 				PREV.br_startoff + new->br_blockcount,
2433 				PREV.br_startblock + new->br_blockcount,
2434 				PREV.br_blockcount - new->br_blockcount,
2435 				oldext)))
2436 				goto done;
2437 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2438 				goto done;
2439 			error = xfs_bmbt_update(cur, LEFT.br_startoff,
2440 				LEFT.br_startblock,
2441 				LEFT.br_blockcount + new->br_blockcount,
2442 				LEFT.br_state);
2443 			if (error)
2444 				goto done;
2445 		}
2446 		break;
2447 
2448 	case BMAP_LEFT_FILLING:
2449 		/*
2450 		 * Setting the first part of a previous oldext extent to newext.
2451 		 * The left neighbor is not contiguous.
2452 		 */
2453 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2454 		ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2455 		xfs_bmbt_set_startoff(ep, new_endoff);
2456 		xfs_bmbt_set_blockcount(ep,
2457 			PREV.br_blockcount - new->br_blockcount);
2458 		xfs_bmbt_set_startblock(ep,
2459 			new->br_startblock + new->br_blockcount);
2460 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2461 
2462 		xfs_iext_insert(ip, *idx, 1, new, state);
2463 		XFS_IFORK_NEXT_SET(ip, whichfork,
2464 				XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
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(mp, i == 1, done);
2474 			if ((error = xfs_bmbt_update(cur,
2475 				PREV.br_startoff + new->br_blockcount,
2476 				PREV.br_startblock + new->br_blockcount,
2477 				PREV.br_blockcount - new->br_blockcount,
2478 				oldext)))
2479 				goto done;
2480 			cur->bc_rec.b = *new;
2481 			if ((error = xfs_btree_insert(cur, &i)))
2482 				goto done;
2483 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2484 		}
2485 		break;
2486 
2487 	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2488 		/*
2489 		 * Setting the last part of a previous oldext extent to newext.
2490 		 * The right neighbor is contiguous with the new allocation.
2491 		 */
2492 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2493 		xfs_bmbt_set_blockcount(ep,
2494 			PREV.br_blockcount - new->br_blockcount);
2495 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2496 
2497 		++*idx;
2498 
2499 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2500 		xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2501 			new->br_startoff, new->br_startblock,
2502 			new->br_blockcount + RIGHT.br_blockcount, newext);
2503 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2504 
2505 		if (cur == NULL)
2506 			rval = XFS_ILOG_DEXT;
2507 		else {
2508 			rval = 0;
2509 			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2510 					PREV.br_startblock,
2511 					PREV.br_blockcount, &i)))
2512 				goto done;
2513 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2514 			if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2515 				PREV.br_startblock,
2516 				PREV.br_blockcount - new->br_blockcount,
2517 				oldext)))
2518 				goto done;
2519 			if ((error = xfs_btree_increment(cur, 0, &i)))
2520 				goto done;
2521 			if ((error = xfs_bmbt_update(cur, new->br_startoff,
2522 				new->br_startblock,
2523 				new->br_blockcount + RIGHT.br_blockcount,
2524 				newext)))
2525 				goto done;
2526 		}
2527 		break;
2528 
2529 	case BMAP_RIGHT_FILLING:
2530 		/*
2531 		 * Setting the last part of a previous oldext extent to newext.
2532 		 * The right neighbor is not contiguous.
2533 		 */
2534 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2535 		xfs_bmbt_set_blockcount(ep,
2536 			PREV.br_blockcount - new->br_blockcount);
2537 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2538 
2539 		++*idx;
2540 		xfs_iext_insert(ip, *idx, 1, new, state);
2541 
2542 		XFS_IFORK_NEXT_SET(ip, whichfork,
2543 				XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2544 		if (cur == NULL)
2545 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2546 		else {
2547 			rval = XFS_ILOG_CORE;
2548 			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2549 					PREV.br_startblock, PREV.br_blockcount,
2550 					&i)))
2551 				goto done;
2552 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2553 			if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2554 				PREV.br_startblock,
2555 				PREV.br_blockcount - new->br_blockcount,
2556 				oldext)))
2557 				goto done;
2558 			if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2559 					new->br_startblock, new->br_blockcount,
2560 					&i)))
2561 				goto done;
2562 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2563 			cur->bc_rec.b.br_state = XFS_EXT_NORM;
2564 			if ((error = xfs_btree_insert(cur, &i)))
2565 				goto done;
2566 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2567 		}
2568 		break;
2569 
2570 	case 0:
2571 		/*
2572 		 * Setting the middle part of a previous oldext extent to
2573 		 * newext.  Contiguity is impossible here.
2574 		 * One extent becomes three extents.
2575 		 */
2576 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2577 		xfs_bmbt_set_blockcount(ep,
2578 			new->br_startoff - PREV.br_startoff);
2579 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2580 
2581 		r[0] = *new;
2582 		r[1].br_startoff = new_endoff;
2583 		r[1].br_blockcount =
2584 			PREV.br_startoff + PREV.br_blockcount - new_endoff;
2585 		r[1].br_startblock = new->br_startblock + new->br_blockcount;
2586 		r[1].br_state = oldext;
2587 
2588 		++*idx;
2589 		xfs_iext_insert(ip, *idx, 2, &r[0], state);
2590 
2591 		XFS_IFORK_NEXT_SET(ip, whichfork,
2592 				XFS_IFORK_NEXTENTS(ip, whichfork) + 2);
2593 		if (cur == NULL)
2594 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2595 		else {
2596 			rval = XFS_ILOG_CORE;
2597 			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2598 					PREV.br_startblock, PREV.br_blockcount,
2599 					&i)))
2600 				goto done;
2601 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2602 			/* new right extent - oldext */
2603 			if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2604 				r[1].br_startblock, r[1].br_blockcount,
2605 				r[1].br_state)))
2606 				goto done;
2607 			/* new left extent - oldext */
2608 			cur->bc_rec.b = PREV;
2609 			cur->bc_rec.b.br_blockcount =
2610 				new->br_startoff - PREV.br_startoff;
2611 			if ((error = xfs_btree_insert(cur, &i)))
2612 				goto done;
2613 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2614 			/*
2615 			 * Reset the cursor to the position of the new extent
2616 			 * we are about to insert as we can't trust it after
2617 			 * the previous insert.
2618 			 */
2619 			if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2620 					new->br_startblock, new->br_blockcount,
2621 					&i)))
2622 				goto done;
2623 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2624 			/* new middle extent - newext */
2625 			cur->bc_rec.b.br_state = new->br_state;
2626 			if ((error = xfs_btree_insert(cur, &i)))
2627 				goto done;
2628 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2629 		}
2630 		break;
2631 
2632 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2633 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2634 	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2635 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2636 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2637 	case BMAP_LEFT_CONTIG:
2638 	case BMAP_RIGHT_CONTIG:
2639 		/*
2640 		 * These cases are all impossible.
2641 		 */
2642 		ASSERT(0);
2643 	}
2644 
2645 	/* update reverse mappings */
2646 	error = xfs_rmap_convert_extent(mp, dfops, ip, whichfork, new);
2647 	if (error)
2648 		goto done;
2649 
2650 	/* convert to a btree if necessary */
2651 	if (xfs_bmap_needs_btree(ip, whichfork)) {
2652 		int	tmp_logflags;	/* partial log flag return val */
2653 
2654 		ASSERT(cur == NULL);
2655 		error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, &cur,
2656 				0, &tmp_logflags, whichfork);
2657 		*logflagsp |= tmp_logflags;
2658 		if (error)
2659 			goto done;
2660 	}
2661 
2662 	/* clear out the allocated field, done with it now in any case. */
2663 	if (cur) {
2664 		cur->bc_private.b.allocated = 0;
2665 		*curp = cur;
2666 	}
2667 
2668 	xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
2669 done:
2670 	*logflagsp |= rval;
2671 	return error;
2672 #undef	LEFT
2673 #undef	RIGHT
2674 #undef	PREV
2675 }
2676 
2677 /*
2678  * Convert a hole to a delayed allocation.
2679  */
2680 STATIC void
2681 xfs_bmap_add_extent_hole_delay(
2682 	xfs_inode_t		*ip,	/* incore inode pointer */
2683 	int			whichfork,
2684 	xfs_extnum_t		*idx,	/* extent number to update/insert */
2685 	xfs_bmbt_irec_t		*new)	/* new data to add to file extents */
2686 {
2687 	xfs_ifork_t		*ifp;	/* inode fork pointer */
2688 	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2689 	xfs_filblks_t		newlen=0;	/* new indirect size */
2690 	xfs_filblks_t		oldlen=0;	/* old indirect size */
2691 	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2692 	int			state;  /* state bits, accessed thru macros */
2693 	xfs_filblks_t		temp=0;	/* temp for indirect calculations */
2694 
2695 	ifp = XFS_IFORK_PTR(ip, whichfork);
2696 	state = 0;
2697 	if (whichfork == XFS_COW_FORK)
2698 		state |= BMAP_COWFORK;
2699 	ASSERT(isnullstartblock(new->br_startblock));
2700 
2701 	/*
2702 	 * Check and set flags if this segment has a left neighbor
2703 	 */
2704 	if (*idx > 0) {
2705 		state |= BMAP_LEFT_VALID;
2706 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2707 
2708 		if (isnullstartblock(left.br_startblock))
2709 			state |= BMAP_LEFT_DELAY;
2710 	}
2711 
2712 	/*
2713 	 * Check and set flags if the current (right) segment exists.
2714 	 * If it doesn't exist, we're converting the hole at end-of-file.
2715 	 */
2716 	if (*idx < xfs_iext_count(ifp)) {
2717 		state |= BMAP_RIGHT_VALID;
2718 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2719 
2720 		if (isnullstartblock(right.br_startblock))
2721 			state |= BMAP_RIGHT_DELAY;
2722 	}
2723 
2724 	/*
2725 	 * Set contiguity flags on the left and right neighbors.
2726 	 * Don't let extents get too large, even if the pieces are contiguous.
2727 	 */
2728 	if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2729 	    left.br_startoff + left.br_blockcount == new->br_startoff &&
2730 	    left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2731 		state |= BMAP_LEFT_CONTIG;
2732 
2733 	if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2734 	    new->br_startoff + new->br_blockcount == right.br_startoff &&
2735 	    new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2736 	    (!(state & BMAP_LEFT_CONTIG) ||
2737 	     (left.br_blockcount + new->br_blockcount +
2738 	      right.br_blockcount <= MAXEXTLEN)))
2739 		state |= BMAP_RIGHT_CONTIG;
2740 
2741 	/*
2742 	 * Switch out based on the contiguity flags.
2743 	 */
2744 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2745 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2746 		/*
2747 		 * New allocation is contiguous with delayed allocations
2748 		 * on the left and on the right.
2749 		 * Merge all three into a single extent record.
2750 		 */
2751 		--*idx;
2752 		temp = left.br_blockcount + new->br_blockcount +
2753 			right.br_blockcount;
2754 
2755 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2756 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2757 		oldlen = startblockval(left.br_startblock) +
2758 			startblockval(new->br_startblock) +
2759 			startblockval(right.br_startblock);
2760 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2761 					 oldlen);
2762 		xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2763 			nullstartblock((int)newlen));
2764 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2765 
2766 		xfs_iext_remove(ip, *idx + 1, 1, state);
2767 		break;
2768 
2769 	case BMAP_LEFT_CONTIG:
2770 		/*
2771 		 * New allocation is contiguous with a delayed allocation
2772 		 * on the left.
2773 		 * Merge the new allocation with the left neighbor.
2774 		 */
2775 		--*idx;
2776 		temp = left.br_blockcount + new->br_blockcount;
2777 
2778 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2779 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2780 		oldlen = startblockval(left.br_startblock) +
2781 			startblockval(new->br_startblock);
2782 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2783 					 oldlen);
2784 		xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2785 			nullstartblock((int)newlen));
2786 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2787 		break;
2788 
2789 	case BMAP_RIGHT_CONTIG:
2790 		/*
2791 		 * New allocation is contiguous with a delayed allocation
2792 		 * on the right.
2793 		 * Merge the new allocation with the right neighbor.
2794 		 */
2795 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2796 		temp = new->br_blockcount + right.br_blockcount;
2797 		oldlen = startblockval(new->br_startblock) +
2798 			startblockval(right.br_startblock);
2799 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2800 					 oldlen);
2801 		xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2802 			new->br_startoff,
2803 			nullstartblock((int)newlen), temp, right.br_state);
2804 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2805 		break;
2806 
2807 	case 0:
2808 		/*
2809 		 * New allocation is not contiguous with another
2810 		 * delayed allocation.
2811 		 * Insert a new entry.
2812 		 */
2813 		oldlen = newlen = 0;
2814 		xfs_iext_insert(ip, *idx, 1, new, state);
2815 		break;
2816 	}
2817 	if (oldlen != newlen) {
2818 		ASSERT(oldlen > newlen);
2819 		xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2820 				 false);
2821 		/*
2822 		 * Nothing to do for disk quota accounting here.
2823 		 */
2824 	}
2825 }
2826 
2827 /*
2828  * Convert a hole to a real allocation.
2829  */
2830 STATIC int				/* error */
2831 xfs_bmap_add_extent_hole_real(
2832 	struct xfs_trans	*tp,
2833 	struct xfs_inode	*ip,
2834 	int			whichfork,
2835 	xfs_extnum_t		*idx,
2836 	struct xfs_btree_cur	**curp,
2837 	struct xfs_bmbt_irec	*new,
2838 	xfs_fsblock_t		*first,
2839 	struct xfs_defer_ops	*dfops,
2840 	int			*logflagsp)
2841 {
2842 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
2843 	struct xfs_mount	*mp = ip->i_mount;
2844 	struct xfs_btree_cur	*cur = *curp;
2845 	int			error;	/* error return value */
2846 	int			i;	/* temp state */
2847 	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2848 	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2849 	int			rval=0;	/* return value (logging flags) */
2850 	int			state;	/* state bits, accessed thru macros */
2851 
2852 	ASSERT(*idx >= 0);
2853 	ASSERT(*idx <= xfs_iext_count(ifp));
2854 	ASSERT(!isnullstartblock(new->br_startblock));
2855 	ASSERT(!cur || !(cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2856 
2857 	XFS_STATS_INC(mp, xs_add_exlist);
2858 
2859 	state = 0;
2860 	if (whichfork == XFS_ATTR_FORK)
2861 		state |= BMAP_ATTRFORK;
2862 	if (whichfork == XFS_COW_FORK)
2863 		state |= BMAP_COWFORK;
2864 
2865 	/*
2866 	 * Check and set flags if this segment has a left neighbor.
2867 	 */
2868 	if (*idx > 0) {
2869 		state |= BMAP_LEFT_VALID;
2870 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2871 		if (isnullstartblock(left.br_startblock))
2872 			state |= BMAP_LEFT_DELAY;
2873 	}
2874 
2875 	/*
2876 	 * Check and set flags if this segment has a current value.
2877 	 * Not true if we're inserting into the "hole" at eof.
2878 	 */
2879 	if (*idx < xfs_iext_count(ifp)) {
2880 		state |= BMAP_RIGHT_VALID;
2881 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2882 		if (isnullstartblock(right.br_startblock))
2883 			state |= BMAP_RIGHT_DELAY;
2884 	}
2885 
2886 	/*
2887 	 * We're inserting a real allocation between "left" and "right".
2888 	 * Set the contiguity flags.  Don't let extents get too large.
2889 	 */
2890 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2891 	    left.br_startoff + left.br_blockcount == new->br_startoff &&
2892 	    left.br_startblock + left.br_blockcount == new->br_startblock &&
2893 	    left.br_state == new->br_state &&
2894 	    left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2895 		state |= BMAP_LEFT_CONTIG;
2896 
2897 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2898 	    new->br_startoff + new->br_blockcount == right.br_startoff &&
2899 	    new->br_startblock + new->br_blockcount == right.br_startblock &&
2900 	    new->br_state == right.br_state &&
2901 	    new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2902 	    (!(state & BMAP_LEFT_CONTIG) ||
2903 	     left.br_blockcount + new->br_blockcount +
2904 	     right.br_blockcount <= MAXEXTLEN))
2905 		state |= BMAP_RIGHT_CONTIG;
2906 
2907 	error = 0;
2908 	/*
2909 	 * Select which case we're in here, and implement it.
2910 	 */
2911 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2912 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2913 		/*
2914 		 * New allocation is contiguous with real allocations on the
2915 		 * left and on the right.
2916 		 * Merge all three into a single extent record.
2917 		 */
2918 		--*idx;
2919 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2920 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2921 			left.br_blockcount + new->br_blockcount +
2922 			right.br_blockcount);
2923 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2924 
2925 		xfs_iext_remove(ip, *idx + 1, 1, state);
2926 
2927 		XFS_IFORK_NEXT_SET(ip, whichfork,
2928 			XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2929 		if (cur == NULL) {
2930 			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2931 		} else {
2932 			rval = XFS_ILOG_CORE;
2933 			error = xfs_bmbt_lookup_eq(cur, right.br_startoff,
2934 					right.br_startblock, right.br_blockcount,
2935 					&i);
2936 			if (error)
2937 				goto done;
2938 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2939 			error = xfs_btree_delete(cur, &i);
2940 			if (error)
2941 				goto done;
2942 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2943 			error = xfs_btree_decrement(cur, 0, &i);
2944 			if (error)
2945 				goto done;
2946 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2947 			error = xfs_bmbt_update(cur, left.br_startoff,
2948 					left.br_startblock,
2949 					left.br_blockcount +
2950 						new->br_blockcount +
2951 						right.br_blockcount,
2952 					left.br_state);
2953 			if (error)
2954 				goto done;
2955 		}
2956 		break;
2957 
2958 	case BMAP_LEFT_CONTIG:
2959 		/*
2960 		 * New allocation is contiguous with a real allocation
2961 		 * on the left.
2962 		 * Merge the new allocation with the left neighbor.
2963 		 */
2964 		--*idx;
2965 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2966 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2967 			left.br_blockcount + new->br_blockcount);
2968 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2969 
2970 		if (cur == NULL) {
2971 			rval = xfs_ilog_fext(whichfork);
2972 		} else {
2973 			rval = 0;
2974 			error = xfs_bmbt_lookup_eq(cur, left.br_startoff,
2975 					left.br_startblock, left.br_blockcount,
2976 					&i);
2977 			if (error)
2978 				goto done;
2979 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2980 			error = xfs_bmbt_update(cur, left.br_startoff,
2981 					left.br_startblock,
2982 					left.br_blockcount +
2983 						new->br_blockcount,
2984 					left.br_state);
2985 			if (error)
2986 				goto done;
2987 		}
2988 		break;
2989 
2990 	case BMAP_RIGHT_CONTIG:
2991 		/*
2992 		 * New allocation is contiguous with a real allocation
2993 		 * on the right.
2994 		 * Merge the new allocation with the right neighbor.
2995 		 */
2996 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2997 		xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2998 			new->br_startoff, new->br_startblock,
2999 			new->br_blockcount + right.br_blockcount,
3000 			right.br_state);
3001 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
3002 
3003 		if (cur == NULL) {
3004 			rval = xfs_ilog_fext(whichfork);
3005 		} else {
3006 			rval = 0;
3007 			error = xfs_bmbt_lookup_eq(cur,
3008 					right.br_startoff,
3009 					right.br_startblock,
3010 					right.br_blockcount, &i);
3011 			if (error)
3012 				goto done;
3013 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3014 			error = xfs_bmbt_update(cur, new->br_startoff,
3015 					new->br_startblock,
3016 					new->br_blockcount +
3017 						right.br_blockcount,
3018 					right.br_state);
3019 			if (error)
3020 				goto done;
3021 		}
3022 		break;
3023 
3024 	case 0:
3025 		/*
3026 		 * New allocation is not contiguous with another
3027 		 * real allocation.
3028 		 * Insert a new entry.
3029 		 */
3030 		xfs_iext_insert(ip, *idx, 1, new, state);
3031 		XFS_IFORK_NEXT_SET(ip, whichfork,
3032 			XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3033 		if (cur == NULL) {
3034 			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3035 		} else {
3036 			rval = XFS_ILOG_CORE;
3037 			error = xfs_bmbt_lookup_eq(cur,
3038 					new->br_startoff,
3039 					new->br_startblock,
3040 					new->br_blockcount, &i);
3041 			if (error)
3042 				goto done;
3043 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
3044 			cur->bc_rec.b.br_state = new->br_state;
3045 			error = xfs_btree_insert(cur, &i);
3046 			if (error)
3047 				goto done;
3048 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3049 		}
3050 		break;
3051 	}
3052 
3053 	/* add reverse mapping */
3054 	error = xfs_rmap_map_extent(mp, dfops, ip, whichfork, new);
3055 	if (error)
3056 		goto done;
3057 
3058 	/* convert to a btree if necessary */
3059 	if (xfs_bmap_needs_btree(ip, whichfork)) {
3060 		int	tmp_logflags;	/* partial log flag return val */
3061 
3062 		ASSERT(cur == NULL);
3063 		error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, curp,
3064 				0, &tmp_logflags, whichfork);
3065 		*logflagsp |= tmp_logflags;
3066 		cur = *curp;
3067 		if (error)
3068 			goto done;
3069 	}
3070 
3071 	/* clear out the allocated field, done with it now in any case. */
3072 	if (cur)
3073 		cur->bc_private.b.allocated = 0;
3074 
3075 	xfs_bmap_check_leaf_extents(cur, ip, whichfork);
3076 done:
3077 	*logflagsp |= rval;
3078 	return error;
3079 }
3080 
3081 /*
3082  * Functions used in the extent read, allocate and remove paths
3083  */
3084 
3085 /*
3086  * Adjust the size of the new extent based on di_extsize and rt extsize.
3087  */
3088 int
3089 xfs_bmap_extsize_align(
3090 	xfs_mount_t	*mp,
3091 	xfs_bmbt_irec_t	*gotp,		/* next extent pointer */
3092 	xfs_bmbt_irec_t	*prevp,		/* previous extent pointer */
3093 	xfs_extlen_t	extsz,		/* align to this extent size */
3094 	int		rt,		/* is this a realtime inode? */
3095 	int		eof,		/* is extent at end-of-file? */
3096 	int		delay,		/* creating delalloc extent? */
3097 	int		convert,	/* overwriting unwritten extent? */
3098 	xfs_fileoff_t	*offp,		/* in/out: aligned offset */
3099 	xfs_extlen_t	*lenp)		/* in/out: aligned length */
3100 {
3101 	xfs_fileoff_t	orig_off;	/* original offset */
3102 	xfs_extlen_t	orig_alen;	/* original length */
3103 	xfs_fileoff_t	orig_end;	/* original off+len */
3104 	xfs_fileoff_t	nexto;		/* next file offset */
3105 	xfs_fileoff_t	prevo;		/* previous file offset */
3106 	xfs_fileoff_t	align_off;	/* temp for offset */
3107 	xfs_extlen_t	align_alen;	/* temp for length */
3108 	xfs_extlen_t	temp;		/* temp for calculations */
3109 
3110 	if (convert)
3111 		return 0;
3112 
3113 	orig_off = align_off = *offp;
3114 	orig_alen = align_alen = *lenp;
3115 	orig_end = orig_off + orig_alen;
3116 
3117 	/*
3118 	 * If this request overlaps an existing extent, then don't
3119 	 * attempt to perform any additional alignment.
3120 	 */
3121 	if (!delay && !eof &&
3122 	    (orig_off >= gotp->br_startoff) &&
3123 	    (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3124 		return 0;
3125 	}
3126 
3127 	/*
3128 	 * If the file offset is unaligned vs. the extent size
3129 	 * we need to align it.  This will be possible unless
3130 	 * the file was previously written with a kernel that didn't
3131 	 * perform this alignment, or if a truncate shot us in the
3132 	 * foot.
3133 	 */
3134 	temp = do_mod(orig_off, extsz);
3135 	if (temp) {
3136 		align_alen += temp;
3137 		align_off -= temp;
3138 	}
3139 
3140 	/* Same adjustment for the end of the requested area. */
3141 	temp = (align_alen % extsz);
3142 	if (temp)
3143 		align_alen += extsz - temp;
3144 
3145 	/*
3146 	 * For large extent hint sizes, the aligned extent might be larger than
3147 	 * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
3148 	 * the length back under MAXEXTLEN. The outer allocation loops handle
3149 	 * short allocation just fine, so it is safe to do this. We only want to
3150 	 * do it when we are forced to, though, because it means more allocation
3151 	 * operations are required.
3152 	 */
3153 	while (align_alen > MAXEXTLEN)
3154 		align_alen -= extsz;
3155 	ASSERT(align_alen <= MAXEXTLEN);
3156 
3157 	/*
3158 	 * If the previous block overlaps with this proposed allocation
3159 	 * then move the start forward without adjusting the length.
3160 	 */
3161 	if (prevp->br_startoff != NULLFILEOFF) {
3162 		if (prevp->br_startblock == HOLESTARTBLOCK)
3163 			prevo = prevp->br_startoff;
3164 		else
3165 			prevo = prevp->br_startoff + prevp->br_blockcount;
3166 	} else
3167 		prevo = 0;
3168 	if (align_off != orig_off && align_off < prevo)
3169 		align_off = prevo;
3170 	/*
3171 	 * If the next block overlaps with this proposed allocation
3172 	 * then move the start back without adjusting the length,
3173 	 * but not before offset 0.
3174 	 * This may of course make the start overlap previous block,
3175 	 * and if we hit the offset 0 limit then the next block
3176 	 * can still overlap too.
3177 	 */
3178 	if (!eof && gotp->br_startoff != NULLFILEOFF) {
3179 		if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3180 		    (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3181 			nexto = gotp->br_startoff + gotp->br_blockcount;
3182 		else
3183 			nexto = gotp->br_startoff;
3184 	} else
3185 		nexto = NULLFILEOFF;
3186 	if (!eof &&
3187 	    align_off + align_alen != orig_end &&
3188 	    align_off + align_alen > nexto)
3189 		align_off = nexto > align_alen ? nexto - align_alen : 0;
3190 	/*
3191 	 * If we're now overlapping the next or previous extent that
3192 	 * means we can't fit an extsz piece in this hole.  Just move
3193 	 * the start forward to the first valid spot and set
3194 	 * the length so we hit the end.
3195 	 */
3196 	if (align_off != orig_off && align_off < prevo)
3197 		align_off = prevo;
3198 	if (align_off + align_alen != orig_end &&
3199 	    align_off + align_alen > nexto &&
3200 	    nexto != NULLFILEOFF) {
3201 		ASSERT(nexto > prevo);
3202 		align_alen = nexto - align_off;
3203 	}
3204 
3205 	/*
3206 	 * If realtime, and the result isn't a multiple of the realtime
3207 	 * extent size we need to remove blocks until it is.
3208 	 */
3209 	if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3210 		/*
3211 		 * We're not covering the original request, or
3212 		 * we won't be able to once we fix the length.
3213 		 */
3214 		if (orig_off < align_off ||
3215 		    orig_end > align_off + align_alen ||
3216 		    align_alen - temp < orig_alen)
3217 			return -EINVAL;
3218 		/*
3219 		 * Try to fix it by moving the start up.
3220 		 */
3221 		if (align_off + temp <= orig_off) {
3222 			align_alen -= temp;
3223 			align_off += temp;
3224 		}
3225 		/*
3226 		 * Try to fix it by moving the end in.
3227 		 */
3228 		else if (align_off + align_alen - temp >= orig_end)
3229 			align_alen -= temp;
3230 		/*
3231 		 * Set the start to the minimum then trim the length.
3232 		 */
3233 		else {
3234 			align_alen -= orig_off - align_off;
3235 			align_off = orig_off;
3236 			align_alen -= align_alen % mp->m_sb.sb_rextsize;
3237 		}
3238 		/*
3239 		 * Result doesn't cover the request, fail it.
3240 		 */
3241 		if (orig_off < align_off || orig_end > align_off + align_alen)
3242 			return -EINVAL;
3243 	} else {
3244 		ASSERT(orig_off >= align_off);
3245 		/* see MAXEXTLEN handling above */
3246 		ASSERT(orig_end <= align_off + align_alen ||
3247 		       align_alen + extsz > MAXEXTLEN);
3248 	}
3249 
3250 #ifdef DEBUG
3251 	if (!eof && gotp->br_startoff != NULLFILEOFF)
3252 		ASSERT(align_off + align_alen <= gotp->br_startoff);
3253 	if (prevp->br_startoff != NULLFILEOFF)
3254 		ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3255 #endif
3256 
3257 	*lenp = align_alen;
3258 	*offp = align_off;
3259 	return 0;
3260 }
3261 
3262 #define XFS_ALLOC_GAP_UNITS	4
3263 
3264 void
3265 xfs_bmap_adjacent(
3266 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3267 {
3268 	xfs_fsblock_t	adjust;		/* adjustment to block numbers */
3269 	xfs_agnumber_t	fb_agno;	/* ag number of ap->firstblock */
3270 	xfs_mount_t	*mp;		/* mount point structure */
3271 	int		nullfb;		/* true if ap->firstblock isn't set */
3272 	int		rt;		/* true if inode is realtime */
3273 
3274 #define	ISVALID(x,y)	\
3275 	(rt ? \
3276 		(x) < mp->m_sb.sb_rblocks : \
3277 		XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3278 		XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3279 		XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3280 
3281 	mp = ap->ip->i_mount;
3282 	nullfb = *ap->firstblock == NULLFSBLOCK;
3283 	rt = XFS_IS_REALTIME_INODE(ap->ip) &&
3284 		xfs_alloc_is_userdata(ap->datatype);
3285 	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3286 	/*
3287 	 * If allocating at eof, and there's a previous real block,
3288 	 * try to use its last block as our starting point.
3289 	 */
3290 	if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3291 	    !isnullstartblock(ap->prev.br_startblock) &&
3292 	    ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3293 		    ap->prev.br_startblock)) {
3294 		ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3295 		/*
3296 		 * Adjust for the gap between prevp and us.
3297 		 */
3298 		adjust = ap->offset -
3299 			(ap->prev.br_startoff + ap->prev.br_blockcount);
3300 		if (adjust &&
3301 		    ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3302 			ap->blkno += adjust;
3303 	}
3304 	/*
3305 	 * If not at eof, then compare the two neighbor blocks.
3306 	 * Figure out whether either one gives us a good starting point,
3307 	 * and pick the better one.
3308 	 */
3309 	else if (!ap->eof) {
3310 		xfs_fsblock_t	gotbno;		/* right side block number */
3311 		xfs_fsblock_t	gotdiff=0;	/* right side difference */
3312 		xfs_fsblock_t	prevbno;	/* left side block number */
3313 		xfs_fsblock_t	prevdiff=0;	/* left side difference */
3314 
3315 		/*
3316 		 * If there's a previous (left) block, select a requested
3317 		 * start block based on it.
3318 		 */
3319 		if (ap->prev.br_startoff != NULLFILEOFF &&
3320 		    !isnullstartblock(ap->prev.br_startblock) &&
3321 		    (prevbno = ap->prev.br_startblock +
3322 			       ap->prev.br_blockcount) &&
3323 		    ISVALID(prevbno, ap->prev.br_startblock)) {
3324 			/*
3325 			 * Calculate gap to end of previous block.
3326 			 */
3327 			adjust = prevdiff = ap->offset -
3328 				(ap->prev.br_startoff +
3329 				 ap->prev.br_blockcount);
3330 			/*
3331 			 * Figure the startblock based on the previous block's
3332 			 * end and the gap size.
3333 			 * Heuristic!
3334 			 * If the gap is large relative to the piece we're
3335 			 * allocating, or using it gives us an invalid block
3336 			 * number, then just use the end of the previous block.
3337 			 */
3338 			if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3339 			    ISVALID(prevbno + prevdiff,
3340 				    ap->prev.br_startblock))
3341 				prevbno += adjust;
3342 			else
3343 				prevdiff += adjust;
3344 			/*
3345 			 * If the firstblock forbids it, can't use it,
3346 			 * must use default.
3347 			 */
3348 			if (!rt && !nullfb &&
3349 			    XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3350 				prevbno = NULLFSBLOCK;
3351 		}
3352 		/*
3353 		 * No previous block or can't follow it, just default.
3354 		 */
3355 		else
3356 			prevbno = NULLFSBLOCK;
3357 		/*
3358 		 * If there's a following (right) block, select a requested
3359 		 * start block based on it.
3360 		 */
3361 		if (!isnullstartblock(ap->got.br_startblock)) {
3362 			/*
3363 			 * Calculate gap to start of next block.
3364 			 */
3365 			adjust = gotdiff = ap->got.br_startoff - ap->offset;
3366 			/*
3367 			 * Figure the startblock based on the next block's
3368 			 * start and the gap size.
3369 			 */
3370 			gotbno = ap->got.br_startblock;
3371 			/*
3372 			 * Heuristic!
3373 			 * If the gap is large relative to the piece we're
3374 			 * allocating, or using it gives us an invalid block
3375 			 * number, then just use the start of the next block
3376 			 * offset by our length.
3377 			 */
3378 			if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3379 			    ISVALID(gotbno - gotdiff, gotbno))
3380 				gotbno -= adjust;
3381 			else if (ISVALID(gotbno - ap->length, gotbno)) {
3382 				gotbno -= ap->length;
3383 				gotdiff += adjust - ap->length;
3384 			} else
3385 				gotdiff += adjust;
3386 			/*
3387 			 * If the firstblock forbids it, can't use it,
3388 			 * must use default.
3389 			 */
3390 			if (!rt && !nullfb &&
3391 			    XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3392 				gotbno = NULLFSBLOCK;
3393 		}
3394 		/*
3395 		 * No next block, just default.
3396 		 */
3397 		else
3398 			gotbno = NULLFSBLOCK;
3399 		/*
3400 		 * If both valid, pick the better one, else the only good
3401 		 * one, else ap->blkno is already set (to 0 or the inode block).
3402 		 */
3403 		if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3404 			ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3405 		else if (prevbno != NULLFSBLOCK)
3406 			ap->blkno = prevbno;
3407 		else if (gotbno != NULLFSBLOCK)
3408 			ap->blkno = gotbno;
3409 	}
3410 #undef ISVALID
3411 }
3412 
3413 static int
3414 xfs_bmap_longest_free_extent(
3415 	struct xfs_trans	*tp,
3416 	xfs_agnumber_t		ag,
3417 	xfs_extlen_t		*blen,
3418 	int			*notinit)
3419 {
3420 	struct xfs_mount	*mp = tp->t_mountp;
3421 	struct xfs_perag	*pag;
3422 	xfs_extlen_t		longest;
3423 	int			error = 0;
3424 
3425 	pag = xfs_perag_get(mp, ag);
3426 	if (!pag->pagf_init) {
3427 		error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3428 		if (error)
3429 			goto out;
3430 
3431 		if (!pag->pagf_init) {
3432 			*notinit = 1;
3433 			goto out;
3434 		}
3435 	}
3436 
3437 	longest = xfs_alloc_longest_free_extent(mp, pag,
3438 				xfs_alloc_min_freelist(mp, pag),
3439 				xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
3440 	if (*blen < longest)
3441 		*blen = longest;
3442 
3443 out:
3444 	xfs_perag_put(pag);
3445 	return error;
3446 }
3447 
3448 static void
3449 xfs_bmap_select_minlen(
3450 	struct xfs_bmalloca	*ap,
3451 	struct xfs_alloc_arg	*args,
3452 	xfs_extlen_t		*blen,
3453 	int			notinit)
3454 {
3455 	if (notinit || *blen < ap->minlen) {
3456 		/*
3457 		 * Since we did a BUF_TRYLOCK above, it is possible that
3458 		 * there is space for this request.
3459 		 */
3460 		args->minlen = ap->minlen;
3461 	} else if (*blen < args->maxlen) {
3462 		/*
3463 		 * If the best seen length is less than the request length,
3464 		 * use the best as the minimum.
3465 		 */
3466 		args->minlen = *blen;
3467 	} else {
3468 		/*
3469 		 * Otherwise we've seen an extent as big as maxlen, use that
3470 		 * as the minimum.
3471 		 */
3472 		args->minlen = args->maxlen;
3473 	}
3474 }
3475 
3476 STATIC int
3477 xfs_bmap_btalloc_nullfb(
3478 	struct xfs_bmalloca	*ap,
3479 	struct xfs_alloc_arg	*args,
3480 	xfs_extlen_t		*blen)
3481 {
3482 	struct xfs_mount	*mp = ap->ip->i_mount;
3483 	xfs_agnumber_t		ag, startag;
3484 	int			notinit = 0;
3485 	int			error;
3486 
3487 	args->type = XFS_ALLOCTYPE_START_BNO;
3488 	args->total = ap->total;
3489 
3490 	startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3491 	if (startag == NULLAGNUMBER)
3492 		startag = ag = 0;
3493 
3494 	while (*blen < args->maxlen) {
3495 		error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3496 						     &notinit);
3497 		if (error)
3498 			return error;
3499 
3500 		if (++ag == mp->m_sb.sb_agcount)
3501 			ag = 0;
3502 		if (ag == startag)
3503 			break;
3504 	}
3505 
3506 	xfs_bmap_select_minlen(ap, args, blen, notinit);
3507 	return 0;
3508 }
3509 
3510 STATIC int
3511 xfs_bmap_btalloc_filestreams(
3512 	struct xfs_bmalloca	*ap,
3513 	struct xfs_alloc_arg	*args,
3514 	xfs_extlen_t		*blen)
3515 {
3516 	struct xfs_mount	*mp = ap->ip->i_mount;
3517 	xfs_agnumber_t		ag;
3518 	int			notinit = 0;
3519 	int			error;
3520 
3521 	args->type = XFS_ALLOCTYPE_NEAR_BNO;
3522 	args->total = ap->total;
3523 
3524 	ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3525 	if (ag == NULLAGNUMBER)
3526 		ag = 0;
3527 
3528 	error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3529 	if (error)
3530 		return error;
3531 
3532 	if (*blen < args->maxlen) {
3533 		error = xfs_filestream_new_ag(ap, &ag);
3534 		if (error)
3535 			return error;
3536 
3537 		error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3538 						     &notinit);
3539 		if (error)
3540 			return error;
3541 
3542 	}
3543 
3544 	xfs_bmap_select_minlen(ap, args, blen, notinit);
3545 
3546 	/*
3547 	 * Set the failure fallback case to look in the selected AG as stream
3548 	 * may have moved.
3549 	 */
3550 	ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3551 	return 0;
3552 }
3553 
3554 STATIC int
3555 xfs_bmap_btalloc(
3556 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3557 {
3558 	xfs_mount_t	*mp;		/* mount point structure */
3559 	xfs_alloctype_t	atype = 0;	/* type for allocation routines */
3560 	xfs_extlen_t	align = 0;	/* minimum allocation alignment */
3561 	xfs_agnumber_t	fb_agno;	/* ag number of ap->firstblock */
3562 	xfs_agnumber_t	ag;
3563 	xfs_alloc_arg_t	args;
3564 	xfs_extlen_t	blen;
3565 	xfs_extlen_t	nextminlen = 0;
3566 	int		nullfb;		/* true if ap->firstblock isn't set */
3567 	int		isaligned;
3568 	int		tryagain;
3569 	int		error;
3570 	int		stripe_align;
3571 
3572 	ASSERT(ap->length);
3573 
3574 	mp = ap->ip->i_mount;
3575 
3576 	/* stripe alignment for allocation is determined by mount parameters */
3577 	stripe_align = 0;
3578 	if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3579 		stripe_align = mp->m_swidth;
3580 	else if (mp->m_dalign)
3581 		stripe_align = mp->m_dalign;
3582 
3583 	if (ap->flags & XFS_BMAPI_COWFORK)
3584 		align = xfs_get_cowextsz_hint(ap->ip);
3585 	else if (xfs_alloc_is_userdata(ap->datatype))
3586 		align = xfs_get_extsz_hint(ap->ip);
3587 	if (align) {
3588 		error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3589 						align, 0, ap->eof, 0, ap->conv,
3590 						&ap->offset, &ap->length);
3591 		ASSERT(!error);
3592 		ASSERT(ap->length);
3593 	}
3594 
3595 
3596 	nullfb = *ap->firstblock == NULLFSBLOCK;
3597 	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3598 	if (nullfb) {
3599 		if (xfs_alloc_is_userdata(ap->datatype) &&
3600 		    xfs_inode_is_filestream(ap->ip)) {
3601 			ag = xfs_filestream_lookup_ag(ap->ip);
3602 			ag = (ag != NULLAGNUMBER) ? ag : 0;
3603 			ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3604 		} else {
3605 			ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3606 		}
3607 	} else
3608 		ap->blkno = *ap->firstblock;
3609 
3610 	xfs_bmap_adjacent(ap);
3611 
3612 	/*
3613 	 * If allowed, use ap->blkno; otherwise must use firstblock since
3614 	 * it's in the right allocation group.
3615 	 */
3616 	if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3617 		;
3618 	else
3619 		ap->blkno = *ap->firstblock;
3620 	/*
3621 	 * Normal allocation, done through xfs_alloc_vextent.
3622 	 */
3623 	tryagain = isaligned = 0;
3624 	memset(&args, 0, sizeof(args));
3625 	args.tp = ap->tp;
3626 	args.mp = mp;
3627 	args.fsbno = ap->blkno;
3628 	xfs_rmap_skip_owner_update(&args.oinfo);
3629 
3630 	/* Trim the allocation back to the maximum an AG can fit. */
3631 	args.maxlen = MIN(ap->length, mp->m_ag_max_usable);
3632 	args.firstblock = *ap->firstblock;
3633 	blen = 0;
3634 	if (nullfb) {
3635 		/*
3636 		 * Search for an allocation group with a single extent large
3637 		 * enough for the request.  If one isn't found, then adjust
3638 		 * the minimum allocation size to the largest space found.
3639 		 */
3640 		if (xfs_alloc_is_userdata(ap->datatype) &&
3641 		    xfs_inode_is_filestream(ap->ip))
3642 			error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3643 		else
3644 			error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3645 		if (error)
3646 			return error;
3647 	} else if (ap->dfops->dop_low) {
3648 		if (xfs_inode_is_filestream(ap->ip))
3649 			args.type = XFS_ALLOCTYPE_FIRST_AG;
3650 		else
3651 			args.type = XFS_ALLOCTYPE_START_BNO;
3652 		args.total = args.minlen = ap->minlen;
3653 	} else {
3654 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
3655 		args.total = ap->total;
3656 		args.minlen = ap->minlen;
3657 	}
3658 	/* apply extent size hints if obtained earlier */
3659 	if (align) {
3660 		args.prod = align;
3661 		if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3662 			args.mod = (xfs_extlen_t)(args.prod - args.mod);
3663 	} else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3664 		args.prod = 1;
3665 		args.mod = 0;
3666 	} else {
3667 		args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3668 		if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3669 			args.mod = (xfs_extlen_t)(args.prod - args.mod);
3670 	}
3671 	/*
3672 	 * If we are not low on available data blocks, and the
3673 	 * underlying logical volume manager is a stripe, and
3674 	 * the file offset is zero then try to allocate data
3675 	 * blocks on stripe unit boundary.
3676 	 * NOTE: ap->aeof is only set if the allocation length
3677 	 * is >= the stripe unit and the allocation offset is
3678 	 * at the end of file.
3679 	 */
3680 	if (!ap->dfops->dop_low && ap->aeof) {
3681 		if (!ap->offset) {
3682 			args.alignment = stripe_align;
3683 			atype = args.type;
3684 			isaligned = 1;
3685 			/*
3686 			 * Adjust for alignment
3687 			 */
3688 			if (blen > args.alignment && blen <= args.maxlen)
3689 				args.minlen = blen - args.alignment;
3690 			args.minalignslop = 0;
3691 		} else {
3692 			/*
3693 			 * First try an exact bno allocation.
3694 			 * If it fails then do a near or start bno
3695 			 * allocation with alignment turned on.
3696 			 */
3697 			atype = args.type;
3698 			tryagain = 1;
3699 			args.type = XFS_ALLOCTYPE_THIS_BNO;
3700 			args.alignment = 1;
3701 			/*
3702 			 * Compute the minlen+alignment for the
3703 			 * next case.  Set slop so that the value
3704 			 * of minlen+alignment+slop doesn't go up
3705 			 * between the calls.
3706 			 */
3707 			if (blen > stripe_align && blen <= args.maxlen)
3708 				nextminlen = blen - stripe_align;
3709 			else
3710 				nextminlen = args.minlen;
3711 			if (nextminlen + stripe_align > args.minlen + 1)
3712 				args.minalignslop =
3713 					nextminlen + stripe_align -
3714 					args.minlen - 1;
3715 			else
3716 				args.minalignslop = 0;
3717 		}
3718 	} else {
3719 		args.alignment = 1;
3720 		args.minalignslop = 0;
3721 	}
3722 	args.minleft = ap->minleft;
3723 	args.wasdel = ap->wasdel;
3724 	args.resv = XFS_AG_RESV_NONE;
3725 	args.datatype = ap->datatype;
3726 	if (ap->datatype & XFS_ALLOC_USERDATA_ZERO)
3727 		args.ip = ap->ip;
3728 
3729 	error = xfs_alloc_vextent(&args);
3730 	if (error)
3731 		return error;
3732 
3733 	if (tryagain && args.fsbno == NULLFSBLOCK) {
3734 		/*
3735 		 * Exact allocation failed. Now try with alignment
3736 		 * turned on.
3737 		 */
3738 		args.type = atype;
3739 		args.fsbno = ap->blkno;
3740 		args.alignment = stripe_align;
3741 		args.minlen = nextminlen;
3742 		args.minalignslop = 0;
3743 		isaligned = 1;
3744 		if ((error = xfs_alloc_vextent(&args)))
3745 			return error;
3746 	}
3747 	if (isaligned && args.fsbno == NULLFSBLOCK) {
3748 		/*
3749 		 * allocation failed, so turn off alignment and
3750 		 * try again.
3751 		 */
3752 		args.type = atype;
3753 		args.fsbno = ap->blkno;
3754 		args.alignment = 0;
3755 		if ((error = xfs_alloc_vextent(&args)))
3756 			return error;
3757 	}
3758 	if (args.fsbno == NULLFSBLOCK && nullfb &&
3759 	    args.minlen > ap->minlen) {
3760 		args.minlen = ap->minlen;
3761 		args.type = XFS_ALLOCTYPE_START_BNO;
3762 		args.fsbno = ap->blkno;
3763 		if ((error = xfs_alloc_vextent(&args)))
3764 			return error;
3765 	}
3766 	if (args.fsbno == NULLFSBLOCK && nullfb) {
3767 		args.fsbno = 0;
3768 		args.type = XFS_ALLOCTYPE_FIRST_AG;
3769 		args.total = ap->minlen;
3770 		if ((error = xfs_alloc_vextent(&args)))
3771 			return error;
3772 		ap->dfops->dop_low = true;
3773 	}
3774 	if (args.fsbno != NULLFSBLOCK) {
3775 		/*
3776 		 * check the allocation happened at the same or higher AG than
3777 		 * the first block that was allocated.
3778 		 */
3779 		ASSERT(*ap->firstblock == NULLFSBLOCK ||
3780 		       XFS_FSB_TO_AGNO(mp, *ap->firstblock) <=
3781 		       XFS_FSB_TO_AGNO(mp, args.fsbno));
3782 
3783 		ap->blkno = args.fsbno;
3784 		if (*ap->firstblock == NULLFSBLOCK)
3785 			*ap->firstblock = args.fsbno;
3786 		ASSERT(nullfb || fb_agno <= args.agno);
3787 		ap->length = args.len;
3788 		if (!(ap->flags & XFS_BMAPI_COWFORK))
3789 			ap->ip->i_d.di_nblocks += args.len;
3790 		xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3791 		if (ap->wasdel)
3792 			ap->ip->i_delayed_blks -= args.len;
3793 		/*
3794 		 * Adjust the disk quota also. This was reserved
3795 		 * earlier.
3796 		 */
3797 		xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3798 			ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3799 					XFS_TRANS_DQ_BCOUNT,
3800 			(long) args.len);
3801 	} else {
3802 		ap->blkno = NULLFSBLOCK;
3803 		ap->length = 0;
3804 	}
3805 	return 0;
3806 }
3807 
3808 /*
3809  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3810  * It figures out where to ask the underlying allocator to put the new extent.
3811  */
3812 STATIC int
3813 xfs_bmap_alloc(
3814 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3815 {
3816 	if (XFS_IS_REALTIME_INODE(ap->ip) &&
3817 	    xfs_alloc_is_userdata(ap->datatype))
3818 		return xfs_bmap_rtalloc(ap);
3819 	return xfs_bmap_btalloc(ap);
3820 }
3821 
3822 /* Trim extent to fit a logical block range. */
3823 void
3824 xfs_trim_extent(
3825 	struct xfs_bmbt_irec	*irec,
3826 	xfs_fileoff_t		bno,
3827 	xfs_filblks_t		len)
3828 {
3829 	xfs_fileoff_t		distance;
3830 	xfs_fileoff_t		end = bno + len;
3831 
3832 	if (irec->br_startoff + irec->br_blockcount <= bno ||
3833 	    irec->br_startoff >= end) {
3834 		irec->br_blockcount = 0;
3835 		return;
3836 	}
3837 
3838 	if (irec->br_startoff < bno) {
3839 		distance = bno - irec->br_startoff;
3840 		if (isnullstartblock(irec->br_startblock))
3841 			irec->br_startblock = DELAYSTARTBLOCK;
3842 		if (irec->br_startblock != DELAYSTARTBLOCK &&
3843 		    irec->br_startblock != HOLESTARTBLOCK)
3844 			irec->br_startblock += distance;
3845 		irec->br_startoff += distance;
3846 		irec->br_blockcount -= distance;
3847 	}
3848 
3849 	if (end < irec->br_startoff + irec->br_blockcount) {
3850 		distance = irec->br_startoff + irec->br_blockcount - end;
3851 		irec->br_blockcount -= distance;
3852 	}
3853 }
3854 
3855 /* trim extent to within eof */
3856 void
3857 xfs_trim_extent_eof(
3858 	struct xfs_bmbt_irec	*irec,
3859 	struct xfs_inode	*ip)
3860 
3861 {
3862 	xfs_trim_extent(irec, 0, XFS_B_TO_FSB(ip->i_mount,
3863 					      i_size_read(VFS_I(ip))));
3864 }
3865 
3866 /*
3867  * Trim the returned map to the required bounds
3868  */
3869 STATIC void
3870 xfs_bmapi_trim_map(
3871 	struct xfs_bmbt_irec	*mval,
3872 	struct xfs_bmbt_irec	*got,
3873 	xfs_fileoff_t		*bno,
3874 	xfs_filblks_t		len,
3875 	xfs_fileoff_t		obno,
3876 	xfs_fileoff_t		end,
3877 	int			n,
3878 	int			flags)
3879 {
3880 	if ((flags & XFS_BMAPI_ENTIRE) ||
3881 	    got->br_startoff + got->br_blockcount <= obno) {
3882 		*mval = *got;
3883 		if (isnullstartblock(got->br_startblock))
3884 			mval->br_startblock = DELAYSTARTBLOCK;
3885 		return;
3886 	}
3887 
3888 	if (obno > *bno)
3889 		*bno = obno;
3890 	ASSERT((*bno >= obno) || (n == 0));
3891 	ASSERT(*bno < end);
3892 	mval->br_startoff = *bno;
3893 	if (isnullstartblock(got->br_startblock))
3894 		mval->br_startblock = DELAYSTARTBLOCK;
3895 	else
3896 		mval->br_startblock = got->br_startblock +
3897 					(*bno - got->br_startoff);
3898 	/*
3899 	 * Return the minimum of what we got and what we asked for for
3900 	 * the length.  We can use the len variable here because it is
3901 	 * modified below and we could have been there before coming
3902 	 * here if the first part of the allocation didn't overlap what
3903 	 * was asked for.
3904 	 */
3905 	mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3906 			got->br_blockcount - (*bno - got->br_startoff));
3907 	mval->br_state = got->br_state;
3908 	ASSERT(mval->br_blockcount <= len);
3909 	return;
3910 }
3911 
3912 /*
3913  * Update and validate the extent map to return
3914  */
3915 STATIC void
3916 xfs_bmapi_update_map(
3917 	struct xfs_bmbt_irec	**map,
3918 	xfs_fileoff_t		*bno,
3919 	xfs_filblks_t		*len,
3920 	xfs_fileoff_t		obno,
3921 	xfs_fileoff_t		end,
3922 	int			*n,
3923 	int			flags)
3924 {
3925 	xfs_bmbt_irec_t	*mval = *map;
3926 
3927 	ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3928 	       ((mval->br_startoff + mval->br_blockcount) <= end));
3929 	ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3930 	       (mval->br_startoff < obno));
3931 
3932 	*bno = mval->br_startoff + mval->br_blockcount;
3933 	*len = end - *bno;
3934 	if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3935 		/* update previous map with new information */
3936 		ASSERT(mval->br_startblock == mval[-1].br_startblock);
3937 		ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3938 		ASSERT(mval->br_state == mval[-1].br_state);
3939 		mval[-1].br_blockcount = mval->br_blockcount;
3940 		mval[-1].br_state = mval->br_state;
3941 	} else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3942 		   mval[-1].br_startblock != DELAYSTARTBLOCK &&
3943 		   mval[-1].br_startblock != HOLESTARTBLOCK &&
3944 		   mval->br_startblock == mval[-1].br_startblock +
3945 					  mval[-1].br_blockcount &&
3946 		   ((flags & XFS_BMAPI_IGSTATE) ||
3947 			mval[-1].br_state == mval->br_state)) {
3948 		ASSERT(mval->br_startoff ==
3949 		       mval[-1].br_startoff + mval[-1].br_blockcount);
3950 		mval[-1].br_blockcount += mval->br_blockcount;
3951 	} else if (*n > 0 &&
3952 		   mval->br_startblock == DELAYSTARTBLOCK &&
3953 		   mval[-1].br_startblock == DELAYSTARTBLOCK &&
3954 		   mval->br_startoff ==
3955 		   mval[-1].br_startoff + mval[-1].br_blockcount) {
3956 		mval[-1].br_blockcount += mval->br_blockcount;
3957 		mval[-1].br_state = mval->br_state;
3958 	} else if (!((*n == 0) &&
3959 		     ((mval->br_startoff + mval->br_blockcount) <=
3960 		      obno))) {
3961 		mval++;
3962 		(*n)++;
3963 	}
3964 	*map = mval;
3965 }
3966 
3967 /*
3968  * Map file blocks to filesystem blocks without allocation.
3969  */
3970 int
3971 xfs_bmapi_read(
3972 	struct xfs_inode	*ip,
3973 	xfs_fileoff_t		bno,
3974 	xfs_filblks_t		len,
3975 	struct xfs_bmbt_irec	*mval,
3976 	int			*nmap,
3977 	int			flags)
3978 {
3979 	struct xfs_mount	*mp = ip->i_mount;
3980 	struct xfs_ifork	*ifp;
3981 	struct xfs_bmbt_irec	got;
3982 	xfs_fileoff_t		obno;
3983 	xfs_fileoff_t		end;
3984 	xfs_extnum_t		idx;
3985 	int			error;
3986 	bool			eof = false;
3987 	int			n = 0;
3988 	int			whichfork = xfs_bmapi_whichfork(flags);
3989 
3990 	ASSERT(*nmap >= 1);
3991 	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
3992 			   XFS_BMAPI_IGSTATE|XFS_BMAPI_COWFORK)));
3993 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
3994 
3995 	if (unlikely(XFS_TEST_ERROR(
3996 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3997 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
3998 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
3999 		XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4000 		return -EFSCORRUPTED;
4001 	}
4002 
4003 	if (XFS_FORCED_SHUTDOWN(mp))
4004 		return -EIO;
4005 
4006 	XFS_STATS_INC(mp, xs_blk_mapr);
4007 
4008 	ifp = XFS_IFORK_PTR(ip, whichfork);
4009 
4010 	/* No CoW fork?  Return a hole. */
4011 	if (whichfork == XFS_COW_FORK && !ifp) {
4012 		mval->br_startoff = bno;
4013 		mval->br_startblock = HOLESTARTBLOCK;
4014 		mval->br_blockcount = len;
4015 		mval->br_state = XFS_EXT_NORM;
4016 		*nmap = 1;
4017 		return 0;
4018 	}
4019 
4020 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4021 		error = xfs_iread_extents(NULL, ip, whichfork);
4022 		if (error)
4023 			return error;
4024 	}
4025 
4026 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &idx, &got))
4027 		eof = true;
4028 	end = bno + len;
4029 	obno = bno;
4030 
4031 	while (bno < end && n < *nmap) {
4032 		/* Reading past eof, act as though there's a hole up to end. */
4033 		if (eof)
4034 			got.br_startoff = end;
4035 		if (got.br_startoff > bno) {
4036 			/* Reading in a hole.  */
4037 			mval->br_startoff = bno;
4038 			mval->br_startblock = HOLESTARTBLOCK;
4039 			mval->br_blockcount =
4040 				XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4041 			mval->br_state = XFS_EXT_NORM;
4042 			bno += mval->br_blockcount;
4043 			len -= mval->br_blockcount;
4044 			mval++;
4045 			n++;
4046 			continue;
4047 		}
4048 
4049 		/* set up the extent map to return. */
4050 		xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4051 		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4052 
4053 		/* If we're done, stop now. */
4054 		if (bno >= end || n >= *nmap)
4055 			break;
4056 
4057 		/* Else go on to the next record. */
4058 		if (!xfs_iext_get_extent(ifp, ++idx, &got))
4059 			eof = true;
4060 	}
4061 	*nmap = n;
4062 	return 0;
4063 }
4064 
4065 /*
4066  * Add a delayed allocation extent to an inode. Blocks are reserved from the
4067  * global pool and the extent inserted into the inode in-core extent tree.
4068  *
4069  * On entry, got refers to the first extent beyond the offset of the extent to
4070  * allocate or eof is specified if no such extent exists. On return, got refers
4071  * to the extent record that was inserted to the inode fork.
4072  *
4073  * Note that the allocated extent may have been merged with contiguous extents
4074  * during insertion into the inode fork. Thus, got does not reflect the current
4075  * state of the inode fork on return. If necessary, the caller can use lastx to
4076  * look up the updated record in the inode fork.
4077  */
4078 int
4079 xfs_bmapi_reserve_delalloc(
4080 	struct xfs_inode	*ip,
4081 	int			whichfork,
4082 	xfs_fileoff_t		off,
4083 	xfs_filblks_t		len,
4084 	xfs_filblks_t		prealloc,
4085 	struct xfs_bmbt_irec	*got,
4086 	xfs_extnum_t		*lastx,
4087 	int			eof)
4088 {
4089 	struct xfs_mount	*mp = ip->i_mount;
4090 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
4091 	xfs_extlen_t		alen;
4092 	xfs_extlen_t		indlen;
4093 	char			rt = XFS_IS_REALTIME_INODE(ip);
4094 	xfs_extlen_t		extsz;
4095 	int			error;
4096 	xfs_fileoff_t		aoff = off;
4097 
4098 	/*
4099 	 * Cap the alloc length. Keep track of prealloc so we know whether to
4100 	 * tag the inode before we return.
4101 	 */
4102 	alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN);
4103 	if (!eof)
4104 		alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4105 	if (prealloc && alen >= len)
4106 		prealloc = alen - len;
4107 
4108 	/* Figure out the extent size, adjust alen */
4109 	if (whichfork == XFS_COW_FORK)
4110 		extsz = xfs_get_cowextsz_hint(ip);
4111 	else
4112 		extsz = xfs_get_extsz_hint(ip);
4113 	if (extsz) {
4114 		struct xfs_bmbt_irec	prev;
4115 
4116 		if (!xfs_iext_get_extent(ifp, *lastx - 1, &prev))
4117 			prev.br_startoff = NULLFILEOFF;
4118 
4119 		error = xfs_bmap_extsize_align(mp, got, &prev, extsz, rt, eof,
4120 					       1, 0, &aoff, &alen);
4121 		ASSERT(!error);
4122 	}
4123 
4124 	if (rt)
4125 		extsz = alen / mp->m_sb.sb_rextsize;
4126 
4127 	/*
4128 	 * Make a transaction-less quota reservation for delayed allocation
4129 	 * blocks.  This number gets adjusted later.  We return if we haven't
4130 	 * allocated blocks already inside this loop.
4131 	 */
4132 	error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4133 			rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4134 	if (error)
4135 		return error;
4136 
4137 	/*
4138 	 * Split changing sb for alen and indlen since they could be coming
4139 	 * from different places.
4140 	 */
4141 	indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4142 	ASSERT(indlen > 0);
4143 
4144 	if (rt) {
4145 		error = xfs_mod_frextents(mp, -((int64_t)extsz));
4146 	} else {
4147 		error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
4148 	}
4149 
4150 	if (error)
4151 		goto out_unreserve_quota;
4152 
4153 	error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
4154 	if (error)
4155 		goto out_unreserve_blocks;
4156 
4157 
4158 	ip->i_delayed_blks += alen;
4159 
4160 	got->br_startoff = aoff;
4161 	got->br_startblock = nullstartblock(indlen);
4162 	got->br_blockcount = alen;
4163 	got->br_state = XFS_EXT_NORM;
4164 
4165 	xfs_bmap_add_extent_hole_delay(ip, whichfork, lastx, got);
4166 
4167 	/*
4168 	 * Tag the inode if blocks were preallocated. Note that COW fork
4169 	 * preallocation can occur at the start or end of the extent, even when
4170 	 * prealloc == 0, so we must also check the aligned offset and length.
4171 	 */
4172 	if (whichfork == XFS_DATA_FORK && prealloc)
4173 		xfs_inode_set_eofblocks_tag(ip);
4174 	if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len))
4175 		xfs_inode_set_cowblocks_tag(ip);
4176 
4177 	return 0;
4178 
4179 out_unreserve_blocks:
4180 	if (rt)
4181 		xfs_mod_frextents(mp, extsz);
4182 	else
4183 		xfs_mod_fdblocks(mp, alen, false);
4184 out_unreserve_quota:
4185 	if (XFS_IS_QUOTA_ON(mp))
4186 		xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4187 				XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4188 	return error;
4189 }
4190 
4191 static int
4192 xfs_bmapi_allocate(
4193 	struct xfs_bmalloca	*bma)
4194 {
4195 	struct xfs_mount	*mp = bma->ip->i_mount;
4196 	int			whichfork = xfs_bmapi_whichfork(bma->flags);
4197 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4198 	int			tmp_logflags = 0;
4199 	int			error;
4200 
4201 	ASSERT(bma->length > 0);
4202 
4203 	/*
4204 	 * For the wasdelay case, we could also just allocate the stuff asked
4205 	 * for in this bmap call but that wouldn't be as good.
4206 	 */
4207 	if (bma->wasdel) {
4208 		bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4209 		bma->offset = bma->got.br_startoff;
4210 		if (bma->idx) {
4211 			xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4212 					 &bma->prev);
4213 		}
4214 	} else {
4215 		bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4216 		if (!bma->eof)
4217 			bma->length = XFS_FILBLKS_MIN(bma->length,
4218 					bma->got.br_startoff - bma->offset);
4219 	}
4220 
4221 	/*
4222 	 * Set the data type being allocated. For the data fork, the first data
4223 	 * in the file is treated differently to all other allocations. For the
4224 	 * attribute fork, we only need to ensure the allocated range is not on
4225 	 * the busy list.
4226 	 */
4227 	if (!(bma->flags & XFS_BMAPI_METADATA)) {
4228 		bma->datatype = XFS_ALLOC_NOBUSY;
4229 		if (whichfork == XFS_DATA_FORK) {
4230 			if (bma->offset == 0)
4231 				bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA;
4232 			else
4233 				bma->datatype |= XFS_ALLOC_USERDATA;
4234 		}
4235 		if (bma->flags & XFS_BMAPI_ZERO)
4236 			bma->datatype |= XFS_ALLOC_USERDATA_ZERO;
4237 	}
4238 
4239 	bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4240 
4241 	/*
4242 	 * Only want to do the alignment at the eof if it is userdata and
4243 	 * allocation length is larger than a stripe unit.
4244 	 */
4245 	if (mp->m_dalign && bma->length >= mp->m_dalign &&
4246 	    !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4247 		error = xfs_bmap_isaeof(bma, whichfork);
4248 		if (error)
4249 			return error;
4250 	}
4251 
4252 	error = xfs_bmap_alloc(bma);
4253 	if (error)
4254 		return error;
4255 
4256 	if (bma->cur)
4257 		bma->cur->bc_private.b.firstblock = *bma->firstblock;
4258 	if (bma->blkno == NULLFSBLOCK)
4259 		return 0;
4260 	if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4261 		bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4262 		bma->cur->bc_private.b.firstblock = *bma->firstblock;
4263 		bma->cur->bc_private.b.dfops = bma->dfops;
4264 	}
4265 	/*
4266 	 * Bump the number of extents we've allocated
4267 	 * in this call.
4268 	 */
4269 	bma->nallocs++;
4270 
4271 	if (bma->cur)
4272 		bma->cur->bc_private.b.flags =
4273 			bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4274 
4275 	bma->got.br_startoff = bma->offset;
4276 	bma->got.br_startblock = bma->blkno;
4277 	bma->got.br_blockcount = bma->length;
4278 	bma->got.br_state = XFS_EXT_NORM;
4279 
4280 	/*
4281 	 * In the data fork, a wasdelay extent has been initialized, so
4282 	 * shouldn't be flagged as unwritten.
4283 	 *
4284 	 * For the cow fork, however, we convert delalloc reservations
4285 	 * (extents allocated for speculative preallocation) to
4286 	 * allocated unwritten extents, and only convert the unwritten
4287 	 * extents to real extents when we're about to write the data.
4288 	 */
4289 	if ((!bma->wasdel || (bma->flags & XFS_BMAPI_COWFORK)) &&
4290 	    (bma->flags & XFS_BMAPI_PREALLOC) &&
4291 	    xfs_sb_version_hasextflgbit(&mp->m_sb))
4292 		bma->got.br_state = XFS_EXT_UNWRITTEN;
4293 
4294 	if (bma->wasdel)
4295 		error = xfs_bmap_add_extent_delay_real(bma, whichfork);
4296 	else
4297 		error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip,
4298 				whichfork, &bma->idx, &bma->cur, &bma->got,
4299 				bma->firstblock, bma->dfops, &bma->logflags);
4300 
4301 	bma->logflags |= tmp_logflags;
4302 	if (error)
4303 		return error;
4304 
4305 	/*
4306 	 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4307 	 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4308 	 * the neighbouring ones.
4309 	 */
4310 	xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4311 
4312 	ASSERT(bma->got.br_startoff <= bma->offset);
4313 	ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4314 	       bma->offset + bma->length);
4315 	ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4316 	       bma->got.br_state == XFS_EXT_UNWRITTEN);
4317 	return 0;
4318 }
4319 
4320 STATIC int
4321 xfs_bmapi_convert_unwritten(
4322 	struct xfs_bmalloca	*bma,
4323 	struct xfs_bmbt_irec	*mval,
4324 	xfs_filblks_t		len,
4325 	int			flags)
4326 {
4327 	int			whichfork = xfs_bmapi_whichfork(flags);
4328 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4329 	int			tmp_logflags = 0;
4330 	int			error;
4331 
4332 	/* check if we need to do unwritten->real conversion */
4333 	if (mval->br_state == XFS_EXT_UNWRITTEN &&
4334 	    (flags & XFS_BMAPI_PREALLOC))
4335 		return 0;
4336 
4337 	/* check if we need to do real->unwritten conversion */
4338 	if (mval->br_state == XFS_EXT_NORM &&
4339 	    (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4340 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4341 		return 0;
4342 
4343 	/*
4344 	 * Modify (by adding) the state flag, if writing.
4345 	 */
4346 	ASSERT(mval->br_blockcount <= len);
4347 	if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4348 		bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4349 					bma->ip, whichfork);
4350 		bma->cur->bc_private.b.firstblock = *bma->firstblock;
4351 		bma->cur->bc_private.b.dfops = bma->dfops;
4352 	}
4353 	mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4354 				? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4355 
4356 	/*
4357 	 * Before insertion into the bmbt, zero the range being converted
4358 	 * if required.
4359 	 */
4360 	if (flags & XFS_BMAPI_ZERO) {
4361 		error = xfs_zero_extent(bma->ip, mval->br_startblock,
4362 					mval->br_blockcount);
4363 		if (error)
4364 			return error;
4365 	}
4366 
4367 	error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork,
4368 			&bma->idx, &bma->cur, mval, bma->firstblock, bma->dfops,
4369 			&tmp_logflags);
4370 	/*
4371 	 * Log the inode core unconditionally in the unwritten extent conversion
4372 	 * path because the conversion might not have done so (e.g., if the
4373 	 * extent count hasn't changed). We need to make sure the inode is dirty
4374 	 * in the transaction for the sake of fsync(), even if nothing has
4375 	 * changed, because fsync() will not force the log for this transaction
4376 	 * unless it sees the inode pinned.
4377 	 *
4378 	 * Note: If we're only converting cow fork extents, there aren't
4379 	 * any on-disk updates to make, so we don't need to log anything.
4380 	 */
4381 	if (whichfork != XFS_COW_FORK)
4382 		bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4383 	if (error)
4384 		return error;
4385 
4386 	/*
4387 	 * Update our extent pointer, given that
4388 	 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4389 	 * of the neighbouring ones.
4390 	 */
4391 	xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4392 
4393 	/*
4394 	 * We may have combined previously unwritten space with written space,
4395 	 * so generate another request.
4396 	 */
4397 	if (mval->br_blockcount < len)
4398 		return -EAGAIN;
4399 	return 0;
4400 }
4401 
4402 /*
4403  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4404  * extent state if necessary.  Details behaviour is controlled by the flags
4405  * parameter.  Only allocates blocks from a single allocation group, to avoid
4406  * locking problems.
4407  *
4408  * The returned value in "firstblock" from the first call in a transaction
4409  * must be remembered and presented to subsequent calls in "firstblock".
4410  * An upper bound for the number of blocks to be allocated is supplied to
4411  * the first call in "total"; if no allocation group has that many free
4412  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4413  */
4414 int
4415 xfs_bmapi_write(
4416 	struct xfs_trans	*tp,		/* transaction pointer */
4417 	struct xfs_inode	*ip,		/* incore inode */
4418 	xfs_fileoff_t		bno,		/* starting file offs. mapped */
4419 	xfs_filblks_t		len,		/* length to map in file */
4420 	int			flags,		/* XFS_BMAPI_... */
4421 	xfs_fsblock_t		*firstblock,	/* first allocated block
4422 						   controls a.g. for allocs */
4423 	xfs_extlen_t		total,		/* total blocks needed */
4424 	struct xfs_bmbt_irec	*mval,		/* output: map values */
4425 	int			*nmap,		/* i/o: mval size/count */
4426 	struct xfs_defer_ops	*dfops)		/* i/o: list extents to free */
4427 {
4428 	struct xfs_mount	*mp = ip->i_mount;
4429 	struct xfs_ifork	*ifp;
4430 	struct xfs_bmalloca	bma = { NULL };	/* args for xfs_bmap_alloc */
4431 	xfs_fileoff_t		end;		/* end of mapped file region */
4432 	bool			eof = false;	/* after the end of extents */
4433 	int			error;		/* error return */
4434 	int			n;		/* current extent index */
4435 	xfs_fileoff_t		obno;		/* old block number (offset) */
4436 	int			whichfork;	/* data or attr fork */
4437 
4438 #ifdef DEBUG
4439 	xfs_fileoff_t		orig_bno;	/* original block number value */
4440 	int			orig_flags;	/* original flags arg value */
4441 	xfs_filblks_t		orig_len;	/* original value of len arg */
4442 	struct xfs_bmbt_irec	*orig_mval;	/* original value of mval */
4443 	int			orig_nmap;	/* original value of *nmap */
4444 
4445 	orig_bno = bno;
4446 	orig_len = len;
4447 	orig_flags = flags;
4448 	orig_mval = mval;
4449 	orig_nmap = *nmap;
4450 #endif
4451 	whichfork = xfs_bmapi_whichfork(flags);
4452 
4453 	ASSERT(*nmap >= 1);
4454 	ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4455 	ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4456 	ASSERT(tp != NULL ||
4457 	       (flags & (XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK)) ==
4458 			(XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK));
4459 	ASSERT(len > 0);
4460 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4461 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4462 	ASSERT(!(flags & XFS_BMAPI_REMAP));
4463 
4464 	/* zeroing is for currently only for data extents, not metadata */
4465 	ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4466 			(XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4467 	/*
4468 	 * we can allocate unwritten extents or pre-zero allocated blocks,
4469 	 * but it makes no sense to do both at once. This would result in
4470 	 * zeroing the unwritten extent twice, but it still being an
4471 	 * unwritten extent....
4472 	 */
4473 	ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4474 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4475 
4476 	if (unlikely(XFS_TEST_ERROR(
4477 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4478 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4479 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
4480 		XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4481 		return -EFSCORRUPTED;
4482 	}
4483 
4484 	if (XFS_FORCED_SHUTDOWN(mp))
4485 		return -EIO;
4486 
4487 	ifp = XFS_IFORK_PTR(ip, whichfork);
4488 
4489 	XFS_STATS_INC(mp, xs_blk_mapw);
4490 
4491 	if (*firstblock == NULLFSBLOCK) {
4492 		if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4493 			bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4494 		else
4495 			bma.minleft = 1;
4496 	} else {
4497 		bma.minleft = 0;
4498 	}
4499 
4500 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4501 		error = xfs_iread_extents(tp, ip, whichfork);
4502 		if (error)
4503 			goto error0;
4504 	}
4505 
4506 	n = 0;
4507 	end = bno + len;
4508 	obno = bno;
4509 
4510 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.idx, &bma.got))
4511 		eof = true;
4512 	if (!xfs_iext_get_extent(ifp, bma.idx - 1, &bma.prev))
4513 		bma.prev.br_startoff = NULLFILEOFF;
4514 	bma.tp = tp;
4515 	bma.ip = ip;
4516 	bma.total = total;
4517 	bma.datatype = 0;
4518 	bma.dfops = dfops;
4519 	bma.firstblock = firstblock;
4520 
4521 	while (bno < end && n < *nmap) {
4522 		bool			need_alloc = false, wasdelay = false;
4523 
4524 		/* in hole or beyoned EOF? */
4525 		if (eof || bma.got.br_startoff > bno) {
4526 			if (flags & XFS_BMAPI_DELALLOC) {
4527 				/*
4528 				 * For the COW fork we can reasonably get a
4529 				 * request for converting an extent that races
4530 				 * with other threads already having converted
4531 				 * part of it, as there converting COW to
4532 				 * regular blocks is not protected using the
4533 				 * IOLOCK.
4534 				 */
4535 				ASSERT(flags & XFS_BMAPI_COWFORK);
4536 				if (!(flags & XFS_BMAPI_COWFORK)) {
4537 					error = -EIO;
4538 					goto error0;
4539 				}
4540 
4541 				if (eof || bno >= end)
4542 					break;
4543 			} else {
4544 				need_alloc = true;
4545 			}
4546 		} else if (isnullstartblock(bma.got.br_startblock)) {
4547 			wasdelay = true;
4548 		}
4549 
4550 		/*
4551 		 * First, deal with the hole before the allocated space
4552 		 * that we found, if any.
4553 		 */
4554 		if (need_alloc || wasdelay) {
4555 			bma.eof = eof;
4556 			bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4557 			bma.wasdel = wasdelay;
4558 			bma.offset = bno;
4559 			bma.flags = flags;
4560 
4561 			/*
4562 			 * There's a 32/64 bit type mismatch between the
4563 			 * allocation length request (which can be 64 bits in
4564 			 * length) and the bma length request, which is
4565 			 * xfs_extlen_t and therefore 32 bits. Hence we have to
4566 			 * check for 32-bit overflows and handle them here.
4567 			 */
4568 			if (len > (xfs_filblks_t)MAXEXTLEN)
4569 				bma.length = MAXEXTLEN;
4570 			else
4571 				bma.length = len;
4572 
4573 			ASSERT(len > 0);
4574 			ASSERT(bma.length > 0);
4575 			error = xfs_bmapi_allocate(&bma);
4576 			if (error)
4577 				goto error0;
4578 			if (bma.blkno == NULLFSBLOCK)
4579 				break;
4580 
4581 			/*
4582 			 * If this is a CoW allocation, record the data in
4583 			 * the refcount btree for orphan recovery.
4584 			 */
4585 			if (whichfork == XFS_COW_FORK) {
4586 				error = xfs_refcount_alloc_cow_extent(mp, dfops,
4587 						bma.blkno, bma.length);
4588 				if (error)
4589 					goto error0;
4590 			}
4591 		}
4592 
4593 		/* Deal with the allocated space we found.  */
4594 		xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4595 							end, n, flags);
4596 
4597 		/* Execute unwritten extent conversion if necessary */
4598 		error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4599 		if (error == -EAGAIN)
4600 			continue;
4601 		if (error)
4602 			goto error0;
4603 
4604 		/* update the extent map to return */
4605 		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4606 
4607 		/*
4608 		 * If we're done, stop now.  Stop when we've allocated
4609 		 * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4610 		 * the transaction may get too big.
4611 		 */
4612 		if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4613 			break;
4614 
4615 		/* Else go on to the next record. */
4616 		bma.prev = bma.got;
4617 		if (!xfs_iext_get_extent(ifp, ++bma.idx, &bma.got))
4618 			eof = true;
4619 	}
4620 	*nmap = n;
4621 
4622 	/*
4623 	 * Transform from btree to extents, give it cur.
4624 	 */
4625 	if (xfs_bmap_wants_extents(ip, whichfork)) {
4626 		int		tmp_logflags = 0;
4627 
4628 		ASSERT(bma.cur);
4629 		error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4630 			&tmp_logflags, whichfork);
4631 		bma.logflags |= tmp_logflags;
4632 		if (error)
4633 			goto error0;
4634 	}
4635 
4636 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4637 	       XFS_IFORK_NEXTENTS(ip, whichfork) >
4638 		XFS_IFORK_MAXEXT(ip, whichfork));
4639 	error = 0;
4640 error0:
4641 	/*
4642 	 * Log everything.  Do this after conversion, there's no point in
4643 	 * logging the extent records if we've converted to btree format.
4644 	 */
4645 	if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4646 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4647 		bma.logflags &= ~xfs_ilog_fext(whichfork);
4648 	else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4649 		 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4650 		bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4651 	/*
4652 	 * Log whatever the flags say, even if error.  Otherwise we might miss
4653 	 * detecting a case where the data is changed, there's an error,
4654 	 * and it's not logged so we don't shutdown when we should.
4655 	 */
4656 	if (bma.logflags)
4657 		xfs_trans_log_inode(tp, ip, bma.logflags);
4658 
4659 	if (bma.cur) {
4660 		if (!error) {
4661 			ASSERT(*firstblock == NULLFSBLOCK ||
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 static int
4677 xfs_bmapi_remap(
4678 	struct xfs_trans	*tp,
4679 	struct xfs_inode	*ip,
4680 	xfs_fileoff_t		bno,
4681 	xfs_filblks_t		len,
4682 	xfs_fsblock_t		startblock,
4683 	struct xfs_defer_ops	*dfops)
4684 {
4685 	struct xfs_mount	*mp = ip->i_mount;
4686 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4687 	struct xfs_btree_cur	*cur = NULL;
4688 	xfs_fsblock_t		firstblock = NULLFSBLOCK;
4689 	struct xfs_bmbt_irec	got;
4690 	xfs_extnum_t		idx;
4691 	int			logflags = 0, error;
4692 
4693 	ASSERT(len > 0);
4694 	ASSERT(len <= (xfs_filblks_t)MAXEXTLEN);
4695 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4696 
4697 	if (unlikely(XFS_TEST_ERROR(
4698 	    (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4699 	     XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4700 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
4701 		XFS_ERROR_REPORT("xfs_bmapi_remap", XFS_ERRLEVEL_LOW, mp);
4702 		return -EFSCORRUPTED;
4703 	}
4704 
4705 	if (XFS_FORCED_SHUTDOWN(mp))
4706 		return -EIO;
4707 
4708 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4709 		error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4710 		if (error)
4711 			return error;
4712 	}
4713 
4714 	if (xfs_iext_lookup_extent(ip, ifp, bno, &idx, &got)) {
4715 		/* make sure we only reflink into a hole. */
4716 		ASSERT(got.br_startoff > bno);
4717 		ASSERT(got.br_startoff - bno >= len);
4718 	}
4719 
4720 	ip->i_d.di_nblocks += len;
4721 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4722 
4723 	if (ifp->if_flags & XFS_IFBROOT) {
4724 		cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
4725 		cur->bc_private.b.firstblock = firstblock;
4726 		cur->bc_private.b.dfops = dfops;
4727 		cur->bc_private.b.flags = 0;
4728 	}
4729 
4730 	got.br_startoff = bno;
4731 	got.br_startblock = startblock;
4732 	got.br_blockcount = len;
4733 	got.br_state = XFS_EXT_NORM;
4734 
4735 	error = xfs_bmap_add_extent_hole_real(tp, ip, XFS_DATA_FORK, &idx, &cur,
4736 			&got, &firstblock, dfops, &logflags);
4737 	if (error)
4738 		goto error0;
4739 
4740 	if (xfs_bmap_wants_extents(ip, XFS_DATA_FORK)) {
4741 		int		tmp_logflags = 0;
4742 
4743 		error = xfs_bmap_btree_to_extents(tp, ip, cur,
4744 			&tmp_logflags, XFS_DATA_FORK);
4745 		logflags |= tmp_logflags;
4746 	}
4747 
4748 error0:
4749 	if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS)
4750 		logflags &= ~XFS_ILOG_DEXT;
4751 	else if (ip->i_d.di_format != XFS_DINODE_FMT_BTREE)
4752 		logflags &= ~XFS_ILOG_DBROOT;
4753 
4754 	if (logflags)
4755 		xfs_trans_log_inode(tp, ip, logflags);
4756 	if (cur) {
4757 		xfs_btree_del_cursor(cur,
4758 				error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4759 	}
4760 	return error;
4761 }
4762 
4763 /*
4764  * When a delalloc extent is split (e.g., due to a hole punch), the original
4765  * indlen reservation must be shared across the two new extents that are left
4766  * behind.
4767  *
4768  * Given the original reservation and the worst case indlen for the two new
4769  * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4770  * reservation fairly across the two new extents. If necessary, steal available
4771  * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4772  * ores == 1). The number of stolen blocks is returned. The availability and
4773  * subsequent accounting of stolen blocks is the responsibility of the caller.
4774  */
4775 static xfs_filblks_t
4776 xfs_bmap_split_indlen(
4777 	xfs_filblks_t			ores,		/* original res. */
4778 	xfs_filblks_t			*indlen1,	/* ext1 worst indlen */
4779 	xfs_filblks_t			*indlen2,	/* ext2 worst indlen */
4780 	xfs_filblks_t			avail)		/* stealable blocks */
4781 {
4782 	xfs_filblks_t			len1 = *indlen1;
4783 	xfs_filblks_t			len2 = *indlen2;
4784 	xfs_filblks_t			nres = len1 + len2; /* new total res. */
4785 	xfs_filblks_t			stolen = 0;
4786 	xfs_filblks_t			resfactor;
4787 
4788 	/*
4789 	 * Steal as many blocks as we can to try and satisfy the worst case
4790 	 * indlen for both new extents.
4791 	 */
4792 	if (ores < nres && avail)
4793 		stolen = XFS_FILBLKS_MIN(nres - ores, avail);
4794 	ores += stolen;
4795 
4796 	 /* nothing else to do if we've satisfied the new reservation */
4797 	if (ores >= nres)
4798 		return stolen;
4799 
4800 	/*
4801 	 * We can't meet the total required reservation for the two extents.
4802 	 * Calculate the percent of the overall shortage between both extents
4803 	 * and apply this percentage to each of the requested indlen values.
4804 	 * This distributes the shortage fairly and reduces the chances that one
4805 	 * of the two extents is left with nothing when extents are repeatedly
4806 	 * split.
4807 	 */
4808 	resfactor = (ores * 100);
4809 	do_div(resfactor, nres);
4810 	len1 *= resfactor;
4811 	do_div(len1, 100);
4812 	len2 *= resfactor;
4813 	do_div(len2, 100);
4814 	ASSERT(len1 + len2 <= ores);
4815 	ASSERT(len1 < *indlen1 && len2 < *indlen2);
4816 
4817 	/*
4818 	 * Hand out the remainder to each extent. If one of the two reservations
4819 	 * is zero, we want to make sure that one gets a block first. The loop
4820 	 * below starts with len1, so hand len2 a block right off the bat if it
4821 	 * is zero.
4822 	 */
4823 	ores -= (len1 + len2);
4824 	ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores);
4825 	if (ores && !len2 && *indlen2) {
4826 		len2++;
4827 		ores--;
4828 	}
4829 	while (ores) {
4830 		if (len1 < *indlen1) {
4831 			len1++;
4832 			ores--;
4833 		}
4834 		if (!ores)
4835 			break;
4836 		if (len2 < *indlen2) {
4837 			len2++;
4838 			ores--;
4839 		}
4840 	}
4841 
4842 	*indlen1 = len1;
4843 	*indlen2 = len2;
4844 
4845 	return stolen;
4846 }
4847 
4848 int
4849 xfs_bmap_del_extent_delay(
4850 	struct xfs_inode	*ip,
4851 	int			whichfork,
4852 	xfs_extnum_t		*idx,
4853 	struct xfs_bmbt_irec	*got,
4854 	struct xfs_bmbt_irec	*del)
4855 {
4856 	struct xfs_mount	*mp = ip->i_mount;
4857 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
4858 	struct xfs_bmbt_irec	new;
4859 	int64_t			da_old, da_new, da_diff = 0;
4860 	xfs_fileoff_t		del_endoff, got_endoff;
4861 	xfs_filblks_t		got_indlen, new_indlen, stolen;
4862 	int			error = 0, state = 0;
4863 	bool			isrt;
4864 
4865 	XFS_STATS_INC(mp, xs_del_exlist);
4866 
4867 	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4868 	del_endoff = del->br_startoff + del->br_blockcount;
4869 	got_endoff = got->br_startoff + got->br_blockcount;
4870 	da_old = startblockval(got->br_startblock);
4871 	da_new = 0;
4872 
4873 	ASSERT(*idx >= 0);
4874 	ASSERT(*idx <= xfs_iext_count(ifp));
4875 	ASSERT(del->br_blockcount > 0);
4876 	ASSERT(got->br_startoff <= del->br_startoff);
4877 	ASSERT(got_endoff >= del_endoff);
4878 
4879 	if (isrt) {
4880 		uint64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount);
4881 
4882 		do_div(rtexts, mp->m_sb.sb_rextsize);
4883 		xfs_mod_frextents(mp, rtexts);
4884 	}
4885 
4886 	/*
4887 	 * Update the inode delalloc counter now and wait to update the
4888 	 * sb counters as we might have to borrow some blocks for the
4889 	 * indirect block accounting.
4890 	 */
4891 	error = xfs_trans_reserve_quota_nblks(NULL, ip,
4892 			-((long)del->br_blockcount), 0,
4893 			isrt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4894 	if (error)
4895 		return error;
4896 	ip->i_delayed_blks -= del->br_blockcount;
4897 
4898 	if (whichfork == XFS_COW_FORK)
4899 		state |= BMAP_COWFORK;
4900 
4901 	if (got->br_startoff == del->br_startoff)
4902 		state |= BMAP_LEFT_CONTIG;
4903 	if (got_endoff == del_endoff)
4904 		state |= BMAP_RIGHT_CONTIG;
4905 
4906 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
4907 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
4908 		/*
4909 		 * Matches the whole extent.  Delete the entry.
4910 		 */
4911 		xfs_iext_remove(ip, *idx, 1, state);
4912 		--*idx;
4913 		break;
4914 	case BMAP_LEFT_CONTIG:
4915 		/*
4916 		 * Deleting the first part of the extent.
4917 		 */
4918 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4919 		got->br_startoff = del_endoff;
4920 		got->br_blockcount -= del->br_blockcount;
4921 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4922 				got->br_blockcount), da_old);
4923 		got->br_startblock = nullstartblock((int)da_new);
4924 		xfs_iext_update_extent(ifp, *idx, got);
4925 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4926 		break;
4927 	case BMAP_RIGHT_CONTIG:
4928 		/*
4929 		 * Deleting the last part of the extent.
4930 		 */
4931 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4932 		got->br_blockcount = got->br_blockcount - del->br_blockcount;
4933 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4934 				got->br_blockcount), da_old);
4935 		got->br_startblock = nullstartblock((int)da_new);
4936 		xfs_iext_update_extent(ifp, *idx, got);
4937 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4938 		break;
4939 	case 0:
4940 		/*
4941 		 * Deleting the middle of the extent.
4942 		 *
4943 		 * Distribute the original indlen reservation across the two new
4944 		 * extents.  Steal blocks from the deleted extent if necessary.
4945 		 * Stealing blocks simply fudges the fdblocks accounting below.
4946 		 * Warn if either of the new indlen reservations is zero as this
4947 		 * can lead to delalloc problems.
4948 		 */
4949 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4950 
4951 		got->br_blockcount = del->br_startoff - got->br_startoff;
4952 		got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount);
4953 
4954 		new.br_blockcount = got_endoff - del_endoff;
4955 		new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount);
4956 
4957 		WARN_ON_ONCE(!got_indlen || !new_indlen);
4958 		stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen,
4959 						       del->br_blockcount);
4960 
4961 		got->br_startblock = nullstartblock((int)got_indlen);
4962 		xfs_iext_update_extent(ifp, *idx, got);
4963 		trace_xfs_bmap_post_update(ip, *idx, 0, _THIS_IP_);
4964 
4965 		new.br_startoff = del_endoff;
4966 		new.br_state = got->br_state;
4967 		new.br_startblock = nullstartblock((int)new_indlen);
4968 
4969 		++*idx;
4970 		xfs_iext_insert(ip, *idx, 1, &new, state);
4971 
4972 		da_new = got_indlen + new_indlen - stolen;
4973 		del->br_blockcount -= stolen;
4974 		break;
4975 	}
4976 
4977 	ASSERT(da_old >= da_new);
4978 	da_diff = da_old - da_new;
4979 	if (!isrt)
4980 		da_diff += del->br_blockcount;
4981 	if (da_diff)
4982 		xfs_mod_fdblocks(mp, da_diff, false);
4983 	return error;
4984 }
4985 
4986 void
4987 xfs_bmap_del_extent_cow(
4988 	struct xfs_inode	*ip,
4989 	xfs_extnum_t		*idx,
4990 	struct xfs_bmbt_irec	*got,
4991 	struct xfs_bmbt_irec	*del)
4992 {
4993 	struct xfs_mount	*mp = ip->i_mount;
4994 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
4995 	struct xfs_bmbt_irec	new;
4996 	xfs_fileoff_t		del_endoff, got_endoff;
4997 	int			state = BMAP_COWFORK;
4998 
4999 	XFS_STATS_INC(mp, xs_del_exlist);
5000 
5001 	del_endoff = del->br_startoff + del->br_blockcount;
5002 	got_endoff = got->br_startoff + got->br_blockcount;
5003 
5004 	ASSERT(*idx >= 0);
5005 	ASSERT(*idx <= xfs_iext_count(ifp));
5006 	ASSERT(del->br_blockcount > 0);
5007 	ASSERT(got->br_startoff <= del->br_startoff);
5008 	ASSERT(got_endoff >= del_endoff);
5009 	ASSERT(!isnullstartblock(got->br_startblock));
5010 
5011 	if (got->br_startoff == del->br_startoff)
5012 		state |= BMAP_LEFT_CONTIG;
5013 	if (got_endoff == del_endoff)
5014 		state |= BMAP_RIGHT_CONTIG;
5015 
5016 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
5017 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
5018 		/*
5019 		 * Matches the whole extent.  Delete the entry.
5020 		 */
5021 		xfs_iext_remove(ip, *idx, 1, state);
5022 		--*idx;
5023 		break;
5024 	case BMAP_LEFT_CONTIG:
5025 		/*
5026 		 * Deleting the first part of the extent.
5027 		 */
5028 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5029 		got->br_startoff = del_endoff;
5030 		got->br_blockcount -= del->br_blockcount;
5031 		got->br_startblock = del->br_startblock + del->br_blockcount;
5032 		xfs_iext_update_extent(ifp, *idx, got);
5033 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5034 		break;
5035 	case BMAP_RIGHT_CONTIG:
5036 		/*
5037 		 * Deleting the last part of the extent.
5038 		 */
5039 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5040 		got->br_blockcount -= del->br_blockcount;
5041 		xfs_iext_update_extent(ifp, *idx, got);
5042 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5043 		break;
5044 	case 0:
5045 		/*
5046 		 * Deleting the middle of the extent.
5047 		 */
5048 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5049 		got->br_blockcount = del->br_startoff - got->br_startoff;
5050 		xfs_iext_update_extent(ifp, *idx, got);
5051 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5052 
5053 		new.br_startoff = del_endoff;
5054 		new.br_blockcount = got_endoff - del_endoff;
5055 		new.br_state = got->br_state;
5056 		new.br_startblock = del->br_startblock + del->br_blockcount;
5057 
5058 		++*idx;
5059 		xfs_iext_insert(ip, *idx, 1, &new, state);
5060 		break;
5061 	}
5062 }
5063 
5064 /*
5065  * Called by xfs_bmapi to update file extent records and the btree
5066  * after removing space (or undoing a delayed allocation).
5067  */
5068 STATIC int				/* error */
5069 xfs_bmap_del_extent(
5070 	xfs_inode_t		*ip,	/* incore inode pointer */
5071 	xfs_trans_t		*tp,	/* current transaction pointer */
5072 	xfs_extnum_t		*idx,	/* extent number to update/delete */
5073 	struct xfs_defer_ops	*dfops,	/* list of extents to be freed */
5074 	xfs_btree_cur_t		*cur,	/* if null, not a btree */
5075 	xfs_bmbt_irec_t		*del,	/* data to remove from extents */
5076 	int			*logflagsp, /* inode logging flags */
5077 	int			whichfork, /* data or attr fork */
5078 	int			bflags)	/* bmapi flags */
5079 {
5080 	xfs_filblks_t		da_new;	/* new delay-alloc indirect blocks */
5081 	xfs_filblks_t		da_old;	/* old delay-alloc indirect blocks */
5082 	xfs_fsblock_t		del_endblock=0;	/* first block past del */
5083 	xfs_fileoff_t		del_endoff;	/* first offset past del */
5084 	int			delay;	/* current block is delayed allocated */
5085 	int			do_fx;	/* free extent at end of routine */
5086 	xfs_bmbt_rec_host_t	*ep;	/* current extent entry pointer */
5087 	int			error;	/* error return value */
5088 	int			flags;	/* inode logging flags */
5089 	xfs_bmbt_irec_t		got;	/* current extent entry */
5090 	xfs_fileoff_t		got_endoff;	/* first offset past got */
5091 	int			i;	/* temp state */
5092 	xfs_ifork_t		*ifp;	/* inode fork pointer */
5093 	xfs_mount_t		*mp;	/* mount structure */
5094 	xfs_filblks_t		nblks;	/* quota/sb block count */
5095 	xfs_bmbt_irec_t		new;	/* new record to be inserted */
5096 	/* REFERENCED */
5097 	uint			qfield;	/* quota field to update */
5098 	xfs_filblks_t		temp;	/* for indirect length calculations */
5099 	xfs_filblks_t		temp2;	/* for indirect length calculations */
5100 	int			state = 0;
5101 
5102 	mp = ip->i_mount;
5103 	XFS_STATS_INC(mp, xs_del_exlist);
5104 
5105 	if (whichfork == XFS_ATTR_FORK)
5106 		state |= BMAP_ATTRFORK;
5107 	else if (whichfork == XFS_COW_FORK)
5108 		state |= BMAP_COWFORK;
5109 
5110 	ifp = XFS_IFORK_PTR(ip, whichfork);
5111 	ASSERT((*idx >= 0) && (*idx < xfs_iext_count(ifp)));
5112 	ASSERT(del->br_blockcount > 0);
5113 	ep = xfs_iext_get_ext(ifp, *idx);
5114 	xfs_bmbt_get_all(ep, &got);
5115 	ASSERT(got.br_startoff <= del->br_startoff);
5116 	del_endoff = del->br_startoff + del->br_blockcount;
5117 	got_endoff = got.br_startoff + got.br_blockcount;
5118 	ASSERT(got_endoff >= del_endoff);
5119 	delay = isnullstartblock(got.br_startblock);
5120 	ASSERT(isnullstartblock(del->br_startblock) == delay);
5121 	flags = 0;
5122 	qfield = 0;
5123 	error = 0;
5124 	/*
5125 	 * If deleting a real allocation, must free up the disk space.
5126 	 */
5127 	if (!delay) {
5128 		flags = XFS_ILOG_CORE;
5129 		/*
5130 		 * Realtime allocation.  Free it and record di_nblocks update.
5131 		 */
5132 		if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
5133 			xfs_fsblock_t	bno;
5134 			xfs_filblks_t	len;
5135 
5136 			ASSERT(do_mod(del->br_blockcount,
5137 				      mp->m_sb.sb_rextsize) == 0);
5138 			ASSERT(do_mod(del->br_startblock,
5139 				      mp->m_sb.sb_rextsize) == 0);
5140 			bno = del->br_startblock;
5141 			len = del->br_blockcount;
5142 			do_div(bno, mp->m_sb.sb_rextsize);
5143 			do_div(len, mp->m_sb.sb_rextsize);
5144 			error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
5145 			if (error)
5146 				goto done;
5147 			do_fx = 0;
5148 			nblks = len * mp->m_sb.sb_rextsize;
5149 			qfield = XFS_TRANS_DQ_RTBCOUNT;
5150 		}
5151 		/*
5152 		 * Ordinary allocation.
5153 		 */
5154 		else {
5155 			do_fx = 1;
5156 			nblks = del->br_blockcount;
5157 			qfield = XFS_TRANS_DQ_BCOUNT;
5158 		}
5159 		/*
5160 		 * Set up del_endblock and cur for later.
5161 		 */
5162 		del_endblock = del->br_startblock + del->br_blockcount;
5163 		if (cur) {
5164 			if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
5165 					got.br_startblock, got.br_blockcount,
5166 					&i)))
5167 				goto done;
5168 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5169 		}
5170 		da_old = da_new = 0;
5171 	} else {
5172 		da_old = startblockval(got.br_startblock);
5173 		da_new = 0;
5174 		nblks = 0;
5175 		do_fx = 0;
5176 	}
5177 
5178 	/*
5179 	 * Set flag value to use in switch statement.
5180 	 * Left-contig is 2, right-contig is 1.
5181 	 */
5182 	switch (((got.br_startoff == del->br_startoff) << 1) |
5183 		(got_endoff == del_endoff)) {
5184 	case 3:
5185 		/*
5186 		 * Matches the whole extent.  Delete the entry.
5187 		 */
5188 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5189 		xfs_iext_remove(ip, *idx, 1,
5190 				whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
5191 		--*idx;
5192 		if (delay)
5193 			break;
5194 
5195 		XFS_IFORK_NEXT_SET(ip, whichfork,
5196 			XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5197 		flags |= XFS_ILOG_CORE;
5198 		if (!cur) {
5199 			flags |= xfs_ilog_fext(whichfork);
5200 			break;
5201 		}
5202 		if ((error = xfs_btree_delete(cur, &i)))
5203 			goto done;
5204 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5205 		break;
5206 
5207 	case 2:
5208 		/*
5209 		 * Deleting the first part of the extent.
5210 		 */
5211 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5212 		xfs_bmbt_set_startoff(ep, del_endoff);
5213 		temp = got.br_blockcount - del->br_blockcount;
5214 		xfs_bmbt_set_blockcount(ep, temp);
5215 		if (delay) {
5216 			temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
5217 				da_old);
5218 			xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5219 			trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5220 			da_new = temp;
5221 			break;
5222 		}
5223 		xfs_bmbt_set_startblock(ep, del_endblock);
5224 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5225 		if (!cur) {
5226 			flags |= xfs_ilog_fext(whichfork);
5227 			break;
5228 		}
5229 		if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
5230 				got.br_blockcount - del->br_blockcount,
5231 				got.br_state)))
5232 			goto done;
5233 		break;
5234 
5235 	case 1:
5236 		/*
5237 		 * Deleting the last part of the extent.
5238 		 */
5239 		temp = got.br_blockcount - del->br_blockcount;
5240 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5241 		xfs_bmbt_set_blockcount(ep, temp);
5242 		if (delay) {
5243 			temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
5244 				da_old);
5245 			xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5246 			trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5247 			da_new = temp;
5248 			break;
5249 		}
5250 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5251 		if (!cur) {
5252 			flags |= xfs_ilog_fext(whichfork);
5253 			break;
5254 		}
5255 		if ((error = xfs_bmbt_update(cur, got.br_startoff,
5256 				got.br_startblock,
5257 				got.br_blockcount - del->br_blockcount,
5258 				got.br_state)))
5259 			goto done;
5260 		break;
5261 
5262 	case 0:
5263 		/*
5264 		 * Deleting the middle of the extent.
5265 		 */
5266 		temp = del->br_startoff - got.br_startoff;
5267 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5268 		xfs_bmbt_set_blockcount(ep, temp);
5269 		new.br_startoff = del_endoff;
5270 		temp2 = got_endoff - del_endoff;
5271 		new.br_blockcount = temp2;
5272 		new.br_state = got.br_state;
5273 		if (!delay) {
5274 			new.br_startblock = del_endblock;
5275 			flags |= XFS_ILOG_CORE;
5276 			if (cur) {
5277 				if ((error = xfs_bmbt_update(cur,
5278 						got.br_startoff,
5279 						got.br_startblock, temp,
5280 						got.br_state)))
5281 					goto done;
5282 				if ((error = xfs_btree_increment(cur, 0, &i)))
5283 					goto done;
5284 				cur->bc_rec.b = new;
5285 				error = xfs_btree_insert(cur, &i);
5286 				if (error && error != -ENOSPC)
5287 					goto done;
5288 				/*
5289 				 * If get no-space back from btree insert,
5290 				 * it tried a split, and we have a zero
5291 				 * block reservation.
5292 				 * Fix up our state and return the error.
5293 				 */
5294 				if (error == -ENOSPC) {
5295 					/*
5296 					 * Reset the cursor, don't trust
5297 					 * it after any insert operation.
5298 					 */
5299 					if ((error = xfs_bmbt_lookup_eq(cur,
5300 							got.br_startoff,
5301 							got.br_startblock,
5302 							temp, &i)))
5303 						goto done;
5304 					XFS_WANT_CORRUPTED_GOTO(mp,
5305 								i == 1, done);
5306 					/*
5307 					 * Update the btree record back
5308 					 * to the original value.
5309 					 */
5310 					if ((error = xfs_bmbt_update(cur,
5311 							got.br_startoff,
5312 							got.br_startblock,
5313 							got.br_blockcount,
5314 							got.br_state)))
5315 						goto done;
5316 					/*
5317 					 * Reset the extent record back
5318 					 * to the original value.
5319 					 */
5320 					xfs_bmbt_set_blockcount(ep,
5321 						got.br_blockcount);
5322 					flags = 0;
5323 					error = -ENOSPC;
5324 					goto done;
5325 				}
5326 				XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5327 			} else
5328 				flags |= xfs_ilog_fext(whichfork);
5329 			XFS_IFORK_NEXT_SET(ip, whichfork,
5330 				XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5331 		} else {
5332 			xfs_filblks_t	stolen;
5333 			ASSERT(whichfork == XFS_DATA_FORK);
5334 
5335 			/*
5336 			 * Distribute the original indlen reservation across the
5337 			 * two new extents. Steal blocks from the deleted extent
5338 			 * if necessary. Stealing blocks simply fudges the
5339 			 * fdblocks accounting in xfs_bunmapi().
5340 			 */
5341 			temp = xfs_bmap_worst_indlen(ip, got.br_blockcount);
5342 			temp2 = xfs_bmap_worst_indlen(ip, new.br_blockcount);
5343 			stolen = xfs_bmap_split_indlen(da_old, &temp, &temp2,
5344 						       del->br_blockcount);
5345 			da_new = temp + temp2 - stolen;
5346 			del->br_blockcount -= stolen;
5347 
5348 			/*
5349 			 * Set the reservation for each extent. Warn if either
5350 			 * is zero as this can lead to delalloc problems.
5351 			 */
5352 			WARN_ON_ONCE(!temp || !temp2);
5353 			xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5354 			new.br_startblock = nullstartblock((int)temp2);
5355 		}
5356 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5357 		xfs_iext_insert(ip, *idx + 1, 1, &new, state);
5358 		++*idx;
5359 		break;
5360 	}
5361 
5362 	/* remove reverse mapping */
5363 	if (!delay) {
5364 		error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, del);
5365 		if (error)
5366 			goto done;
5367 	}
5368 
5369 	/*
5370 	 * If we need to, add to list of extents to delete.
5371 	 */
5372 	if (do_fx && !(bflags & XFS_BMAPI_REMAP)) {
5373 		if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
5374 			error = xfs_refcount_decrease_extent(mp, dfops, del);
5375 			if (error)
5376 				goto done;
5377 		} else
5378 			xfs_bmap_add_free(mp, dfops, del->br_startblock,
5379 					del->br_blockcount, NULL);
5380 	}
5381 
5382 	/*
5383 	 * Adjust inode # blocks in the file.
5384 	 */
5385 	if (nblks)
5386 		ip->i_d.di_nblocks -= nblks;
5387 	/*
5388 	 * Adjust quota data.
5389 	 */
5390 	if (qfield && !(bflags & XFS_BMAPI_REMAP))
5391 		xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5392 
5393 	/*
5394 	 * Account for change in delayed indirect blocks.
5395 	 * Nothing to do for disk quota accounting here.
5396 	 */
5397 	ASSERT(da_old >= da_new);
5398 	if (da_old > da_new)
5399 		xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new), false);
5400 done:
5401 	*logflagsp = flags;
5402 	return error;
5403 }
5404 
5405 /*
5406  * Unmap (remove) blocks from a file.
5407  * If nexts is nonzero then the number of extents to remove is limited to
5408  * that value.  If not all extents in the block range can be removed then
5409  * *done is set.
5410  */
5411 int						/* error */
5412 __xfs_bunmapi(
5413 	xfs_trans_t		*tp,		/* transaction pointer */
5414 	struct xfs_inode	*ip,		/* incore inode */
5415 	xfs_fileoff_t		bno,		/* starting offset to unmap */
5416 	xfs_filblks_t		*rlen,		/* i/o: amount remaining */
5417 	int			flags,		/* misc flags */
5418 	xfs_extnum_t		nexts,		/* number of extents max */
5419 	xfs_fsblock_t		*firstblock,	/* first allocated block
5420 						   controls a.g. for allocs */
5421 	struct xfs_defer_ops	*dfops)		/* i/o: deferred updates */
5422 {
5423 	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
5424 	xfs_bmbt_irec_t		del;		/* extent being deleted */
5425 	int			error;		/* error return value */
5426 	xfs_extnum_t		extno;		/* extent number in list */
5427 	xfs_bmbt_irec_t		got;		/* current extent record */
5428 	xfs_ifork_t		*ifp;		/* inode fork pointer */
5429 	int			isrt;		/* freeing in rt area */
5430 	xfs_extnum_t		lastx;		/* last extent index used */
5431 	int			logflags;	/* transaction logging flags */
5432 	xfs_extlen_t		mod;		/* rt extent offset */
5433 	xfs_mount_t		*mp;		/* mount structure */
5434 	xfs_fileoff_t		start;		/* first file offset deleted */
5435 	int			tmp_logflags;	/* partial logging flags */
5436 	int			wasdel;		/* was a delayed alloc extent */
5437 	int			whichfork;	/* data or attribute fork */
5438 	xfs_fsblock_t		sum;
5439 	xfs_filblks_t		len = *rlen;	/* length to unmap in file */
5440 	xfs_fileoff_t		max_len;
5441 	xfs_agnumber_t		prev_agno = NULLAGNUMBER, agno;
5442 
5443 	trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5444 
5445 	whichfork = xfs_bmapi_whichfork(flags);
5446 	ASSERT(whichfork != XFS_COW_FORK);
5447 	ifp = XFS_IFORK_PTR(ip, whichfork);
5448 	if (unlikely(
5449 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5450 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5451 		XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5452 				 ip->i_mount);
5453 		return -EFSCORRUPTED;
5454 	}
5455 	mp = ip->i_mount;
5456 	if (XFS_FORCED_SHUTDOWN(mp))
5457 		return -EIO;
5458 
5459 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5460 	ASSERT(len > 0);
5461 	ASSERT(nexts >= 0);
5462 
5463 	/*
5464 	 * Guesstimate how many blocks we can unmap without running the risk of
5465 	 * blowing out the transaction with a mix of EFIs and reflink
5466 	 * adjustments.
5467 	 */
5468 	if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK)
5469 		max_len = min(len, xfs_refcount_max_unmap(tp->t_log_res));
5470 	else
5471 		max_len = len;
5472 
5473 	if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5474 	    (error = xfs_iread_extents(tp, ip, whichfork)))
5475 		return error;
5476 	if (xfs_iext_count(ifp) == 0) {
5477 		*rlen = 0;
5478 		return 0;
5479 	}
5480 	XFS_STATS_INC(mp, xs_blk_unmap);
5481 	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5482 	start = bno;
5483 	bno = start + len - 1;
5484 
5485 	/*
5486 	 * Check to see if the given block number is past the end of the
5487 	 * file, back up to the last block if so...
5488 	 */
5489 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &lastx, &got)) {
5490 		ASSERT(lastx > 0);
5491 		xfs_iext_get_extent(ifp, --lastx, &got);
5492 		bno = got.br_startoff + got.br_blockcount - 1;
5493 	}
5494 
5495 	logflags = 0;
5496 	if (ifp->if_flags & XFS_IFBROOT) {
5497 		ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5498 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5499 		cur->bc_private.b.firstblock = *firstblock;
5500 		cur->bc_private.b.dfops = dfops;
5501 		cur->bc_private.b.flags = 0;
5502 	} else
5503 		cur = NULL;
5504 
5505 	if (isrt) {
5506 		/*
5507 		 * Synchronize by locking the bitmap inode.
5508 		 */
5509 		xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
5510 		xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5511 		xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
5512 		xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
5513 	}
5514 
5515 	extno = 0;
5516 	while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5517 	       (nexts == 0 || extno < nexts) && max_len > 0) {
5518 		/*
5519 		 * Is the found extent after a hole in which bno lives?
5520 		 * Just back up to the previous extent, if so.
5521 		 */
5522 		if (got.br_startoff > bno) {
5523 			if (--lastx < 0)
5524 				break;
5525 			xfs_iext_get_extent(ifp, lastx, &got);
5526 		}
5527 		/*
5528 		 * Is the last block of this extent before the range
5529 		 * we're supposed to delete?  If so, we're done.
5530 		 */
5531 		bno = XFS_FILEOFF_MIN(bno,
5532 			got.br_startoff + got.br_blockcount - 1);
5533 		if (bno < start)
5534 			break;
5535 		/*
5536 		 * Then deal with the (possibly delayed) allocated space
5537 		 * we found.
5538 		 */
5539 		del = got;
5540 		wasdel = isnullstartblock(del.br_startblock);
5541 
5542 		/*
5543 		 * Make sure we don't touch multiple AGF headers out of order
5544 		 * in a single transaction, as that could cause AB-BA deadlocks.
5545 		 */
5546 		if (!wasdel) {
5547 			agno = XFS_FSB_TO_AGNO(mp, del.br_startblock);
5548 			if (prev_agno != NULLAGNUMBER && prev_agno > agno)
5549 				break;
5550 			prev_agno = agno;
5551 		}
5552 		if (got.br_startoff < start) {
5553 			del.br_startoff = start;
5554 			del.br_blockcount -= start - got.br_startoff;
5555 			if (!wasdel)
5556 				del.br_startblock += start - got.br_startoff;
5557 		}
5558 		if (del.br_startoff + del.br_blockcount > bno + 1)
5559 			del.br_blockcount = bno + 1 - del.br_startoff;
5560 
5561 		/* How much can we safely unmap? */
5562 		if (max_len < del.br_blockcount) {
5563 			del.br_startoff += del.br_blockcount - max_len;
5564 			if (!wasdel)
5565 				del.br_startblock += del.br_blockcount - max_len;
5566 			del.br_blockcount = max_len;
5567 		}
5568 
5569 		sum = del.br_startblock + del.br_blockcount;
5570 		if (isrt &&
5571 		    (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5572 			/*
5573 			 * Realtime extent not lined up at the end.
5574 			 * The extent could have been split into written
5575 			 * and unwritten pieces, or we could just be
5576 			 * unmapping part of it.  But we can't really
5577 			 * get rid of part of a realtime extent.
5578 			 */
5579 			if (del.br_state == XFS_EXT_UNWRITTEN ||
5580 			    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5581 				/*
5582 				 * This piece is unwritten, or we're not
5583 				 * using unwritten extents.  Skip over it.
5584 				 */
5585 				ASSERT(bno >= mod);
5586 				bno -= mod > del.br_blockcount ?
5587 					del.br_blockcount : mod;
5588 				if (bno < got.br_startoff) {
5589 					if (--lastx >= 0)
5590 						xfs_bmbt_get_all(xfs_iext_get_ext(
5591 							ifp, lastx), &got);
5592 				}
5593 				continue;
5594 			}
5595 			/*
5596 			 * It's written, turn it unwritten.
5597 			 * This is better than zeroing it.
5598 			 */
5599 			ASSERT(del.br_state == XFS_EXT_NORM);
5600 			ASSERT(tp->t_blk_res > 0);
5601 			/*
5602 			 * If this spans a realtime extent boundary,
5603 			 * chop it back to the start of the one we end at.
5604 			 */
5605 			if (del.br_blockcount > mod) {
5606 				del.br_startoff += del.br_blockcount - mod;
5607 				del.br_startblock += del.br_blockcount - mod;
5608 				del.br_blockcount = mod;
5609 			}
5610 			del.br_state = XFS_EXT_UNWRITTEN;
5611 			error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5612 					whichfork, &lastx, &cur, &del,
5613 					firstblock, dfops, &logflags);
5614 			if (error)
5615 				goto error0;
5616 			goto nodelete;
5617 		}
5618 		if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5619 			/*
5620 			 * Realtime extent is lined up at the end but not
5621 			 * at the front.  We'll get rid of full extents if
5622 			 * we can.
5623 			 */
5624 			mod = mp->m_sb.sb_rextsize - mod;
5625 			if (del.br_blockcount > mod) {
5626 				del.br_blockcount -= mod;
5627 				del.br_startoff += mod;
5628 				del.br_startblock += mod;
5629 			} else if ((del.br_startoff == start &&
5630 				    (del.br_state == XFS_EXT_UNWRITTEN ||
5631 				     tp->t_blk_res == 0)) ||
5632 				   !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5633 				/*
5634 				 * Can't make it unwritten.  There isn't
5635 				 * a full extent here so just skip it.
5636 				 */
5637 				ASSERT(bno >= del.br_blockcount);
5638 				bno -= del.br_blockcount;
5639 				if (got.br_startoff > bno && --lastx >= 0)
5640 					xfs_iext_get_extent(ifp, lastx, &got);
5641 				continue;
5642 			} else if (del.br_state == XFS_EXT_UNWRITTEN) {
5643 				struct xfs_bmbt_irec	prev;
5644 
5645 				/*
5646 				 * This one is already unwritten.
5647 				 * It must have a written left neighbor.
5648 				 * Unwrite the killed part of that one and
5649 				 * try again.
5650 				 */
5651 				ASSERT(lastx > 0);
5652 				xfs_iext_get_extent(ifp, lastx - 1, &prev);
5653 				ASSERT(prev.br_state == XFS_EXT_NORM);
5654 				ASSERT(!isnullstartblock(prev.br_startblock));
5655 				ASSERT(del.br_startblock ==
5656 				       prev.br_startblock + prev.br_blockcount);
5657 				if (prev.br_startoff < start) {
5658 					mod = start - prev.br_startoff;
5659 					prev.br_blockcount -= mod;
5660 					prev.br_startblock += mod;
5661 					prev.br_startoff = start;
5662 				}
5663 				prev.br_state = XFS_EXT_UNWRITTEN;
5664 				lastx--;
5665 				error = xfs_bmap_add_extent_unwritten_real(tp,
5666 						ip, whichfork, &lastx, &cur,
5667 						&prev, firstblock, dfops,
5668 						&logflags);
5669 				if (error)
5670 					goto error0;
5671 				goto nodelete;
5672 			} else {
5673 				ASSERT(del.br_state == XFS_EXT_NORM);
5674 				del.br_state = XFS_EXT_UNWRITTEN;
5675 				error = xfs_bmap_add_extent_unwritten_real(tp,
5676 						ip, whichfork, &lastx, &cur,
5677 						&del, firstblock, dfops,
5678 						&logflags);
5679 				if (error)
5680 					goto error0;
5681 				goto nodelete;
5682 			}
5683 		}
5684 
5685 		/*
5686 		 * If it's the case where the directory code is running
5687 		 * with no block reservation, and the deleted block is in
5688 		 * the middle of its extent, and the resulting insert
5689 		 * of an extent would cause transformation to btree format,
5690 		 * then reject it.  The calling code will then swap
5691 		 * blocks around instead.
5692 		 * We have to do this now, rather than waiting for the
5693 		 * conversion to btree format, since the transaction
5694 		 * will be dirty.
5695 		 */
5696 		if (!wasdel && tp->t_blk_res == 0 &&
5697 		    XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5698 		    XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5699 			XFS_IFORK_MAXEXT(ip, whichfork) &&
5700 		    del.br_startoff > got.br_startoff &&
5701 		    del.br_startoff + del.br_blockcount <
5702 		    got.br_startoff + got.br_blockcount) {
5703 			error = -ENOSPC;
5704 			goto error0;
5705 		}
5706 
5707 		/*
5708 		 * Unreserve quota and update realtime free space, if
5709 		 * appropriate. If delayed allocation, update the inode delalloc
5710 		 * counter now and wait to update the sb counters as
5711 		 * xfs_bmap_del_extent() might need to borrow some blocks.
5712 		 */
5713 		if (wasdel) {
5714 			ASSERT(startblockval(del.br_startblock) > 0);
5715 			if (isrt) {
5716 				xfs_filblks_t rtexts;
5717 
5718 				rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5719 				do_div(rtexts, mp->m_sb.sb_rextsize);
5720 				xfs_mod_frextents(mp, (int64_t)rtexts);
5721 				(void)xfs_trans_reserve_quota_nblks(NULL,
5722 					ip, -((long)del.br_blockcount), 0,
5723 					XFS_QMOPT_RES_RTBLKS);
5724 			} else {
5725 				(void)xfs_trans_reserve_quota_nblks(NULL,
5726 					ip, -((long)del.br_blockcount), 0,
5727 					XFS_QMOPT_RES_REGBLKS);
5728 			}
5729 			ip->i_delayed_blks -= del.br_blockcount;
5730 			if (cur)
5731 				cur->bc_private.b.flags |=
5732 					XFS_BTCUR_BPRV_WASDEL;
5733 		} else if (cur)
5734 			cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5735 
5736 		error = xfs_bmap_del_extent(ip, tp, &lastx, dfops, cur, &del,
5737 				&tmp_logflags, whichfork, flags);
5738 		logflags |= tmp_logflags;
5739 		if (error)
5740 			goto error0;
5741 
5742 		if (!isrt && wasdel)
5743 			xfs_mod_fdblocks(mp, (int64_t)del.br_blockcount, false);
5744 
5745 		max_len -= del.br_blockcount;
5746 		bno = del.br_startoff - 1;
5747 nodelete:
5748 		/*
5749 		 * If not done go on to the next (previous) record.
5750 		 */
5751 		if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5752 			if (lastx >= 0) {
5753 				xfs_iext_get_extent(ifp, lastx, &got);
5754 				if (got.br_startoff > bno && --lastx >= 0)
5755 					xfs_iext_get_extent(ifp, lastx, &got);
5756 			}
5757 			extno++;
5758 		}
5759 	}
5760 	if (bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0)
5761 		*rlen = 0;
5762 	else
5763 		*rlen = bno - start + 1;
5764 
5765 	/*
5766 	 * Convert to a btree if necessary.
5767 	 */
5768 	if (xfs_bmap_needs_btree(ip, whichfork)) {
5769 		ASSERT(cur == NULL);
5770 		error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops,
5771 			&cur, 0, &tmp_logflags, whichfork);
5772 		logflags |= tmp_logflags;
5773 		if (error)
5774 			goto error0;
5775 	}
5776 	/*
5777 	 * transform from btree to extents, give it cur
5778 	 */
5779 	else if (xfs_bmap_wants_extents(ip, whichfork)) {
5780 		ASSERT(cur != NULL);
5781 		error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5782 			whichfork);
5783 		logflags |= tmp_logflags;
5784 		if (error)
5785 			goto error0;
5786 	}
5787 	/*
5788 	 * transform from extents to local?
5789 	 */
5790 	error = 0;
5791 error0:
5792 	/*
5793 	 * Log everything.  Do this after conversion, there's no point in
5794 	 * logging the extent records if we've converted to btree format.
5795 	 */
5796 	if ((logflags & xfs_ilog_fext(whichfork)) &&
5797 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5798 		logflags &= ~xfs_ilog_fext(whichfork);
5799 	else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5800 		 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5801 		logflags &= ~xfs_ilog_fbroot(whichfork);
5802 	/*
5803 	 * Log inode even in the error case, if the transaction
5804 	 * is dirty we'll need to shut down the filesystem.
5805 	 */
5806 	if (logflags)
5807 		xfs_trans_log_inode(tp, ip, logflags);
5808 	if (cur) {
5809 		if (!error) {
5810 			*firstblock = cur->bc_private.b.firstblock;
5811 			cur->bc_private.b.allocated = 0;
5812 		}
5813 		xfs_btree_del_cursor(cur,
5814 			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5815 	}
5816 	return error;
5817 }
5818 
5819 /* Unmap a range of a file. */
5820 int
5821 xfs_bunmapi(
5822 	xfs_trans_t		*tp,
5823 	struct xfs_inode	*ip,
5824 	xfs_fileoff_t		bno,
5825 	xfs_filblks_t		len,
5826 	int			flags,
5827 	xfs_extnum_t		nexts,
5828 	xfs_fsblock_t		*firstblock,
5829 	struct xfs_defer_ops	*dfops,
5830 	int			*done)
5831 {
5832 	int			error;
5833 
5834 	error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts, firstblock,
5835 			dfops);
5836 	*done = (len == 0);
5837 	return error;
5838 }
5839 
5840 /*
5841  * Determine whether an extent shift can be accomplished by a merge with the
5842  * extent that precedes the target hole of the shift.
5843  */
5844 STATIC bool
5845 xfs_bmse_can_merge(
5846 	struct xfs_bmbt_irec	*left,	/* preceding extent */
5847 	struct xfs_bmbt_irec	*got,	/* current extent to shift */
5848 	xfs_fileoff_t		shift)	/* shift fsb */
5849 {
5850 	xfs_fileoff_t		startoff;
5851 
5852 	startoff = got->br_startoff - shift;
5853 
5854 	/*
5855 	 * The extent, once shifted, must be adjacent in-file and on-disk with
5856 	 * the preceding extent.
5857 	 */
5858 	if ((left->br_startoff + left->br_blockcount != startoff) ||
5859 	    (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5860 	    (left->br_state != got->br_state) ||
5861 	    (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5862 		return false;
5863 
5864 	return true;
5865 }
5866 
5867 /*
5868  * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5869  * hole in the file. If an extent shift would result in the extent being fully
5870  * adjacent to the extent that currently precedes the hole, we can merge with
5871  * the preceding extent rather than do the shift.
5872  *
5873  * This function assumes the caller has verified a shift-by-merge is possible
5874  * with the provided extents via xfs_bmse_can_merge().
5875  */
5876 STATIC int
5877 xfs_bmse_merge(
5878 	struct xfs_inode		*ip,
5879 	int				whichfork,
5880 	xfs_fileoff_t			shift,		/* shift fsb */
5881 	int				current_ext,	/* idx of gotp */
5882 	struct xfs_bmbt_irec		*got,		/* extent to shift */
5883 	struct xfs_bmbt_irec		*left,		/* preceding extent */
5884 	struct xfs_btree_cur		*cur,
5885 	int				*logflags,	/* output */
5886 	struct xfs_defer_ops		*dfops)
5887 {
5888 	struct xfs_ifork		*ifp = XFS_IFORK_PTR(ip, whichfork);
5889 	struct xfs_bmbt_irec		new;
5890 	xfs_filblks_t			blockcount;
5891 	int				error, i;
5892 	struct xfs_mount		*mp = ip->i_mount;
5893 
5894 	blockcount = left->br_blockcount + got->br_blockcount;
5895 
5896 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5897 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5898 	ASSERT(xfs_bmse_can_merge(left, got, shift));
5899 
5900 	new = *left;
5901 	new.br_blockcount = blockcount;
5902 
5903 	/*
5904 	 * Update the on-disk extent count, the btree if necessary and log the
5905 	 * inode.
5906 	 */
5907 	XFS_IFORK_NEXT_SET(ip, whichfork,
5908 			   XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5909 	*logflags |= XFS_ILOG_CORE;
5910 	if (!cur) {
5911 		*logflags |= XFS_ILOG_DEXT;
5912 		goto done;
5913 	}
5914 
5915 	/* lookup and remove the extent to merge */
5916 	error = xfs_bmbt_lookup_eq(cur, got->br_startoff, got->br_startblock,
5917 				   got->br_blockcount, &i);
5918 	if (error)
5919 		return error;
5920 	XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5921 
5922 	error = xfs_btree_delete(cur, &i);
5923 	if (error)
5924 		return error;
5925 	XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5926 
5927 	/* lookup and update size of the previous extent */
5928 	error = xfs_bmbt_lookup_eq(cur, left->br_startoff, left->br_startblock,
5929 				   left->br_blockcount, &i);
5930 	if (error)
5931 		return error;
5932 	XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5933 
5934 	error = xfs_bmbt_update(cur, new.br_startoff, new.br_startblock,
5935 			        new.br_blockcount, new.br_state);
5936 	if (error)
5937 		return error;
5938 
5939 done:
5940 	xfs_iext_update_extent(ifp, current_ext - 1, &new);
5941 	xfs_iext_remove(ip, current_ext, 1, 0);
5942 
5943 	/* update reverse mapping. rmap functions merge the rmaps for us */
5944 	error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, got);
5945 	if (error)
5946 		return error;
5947 	memcpy(&new, got, sizeof(new));
5948 	new.br_startoff = left->br_startoff + left->br_blockcount;
5949 	return xfs_rmap_map_extent(mp, dfops, ip, whichfork, &new);
5950 }
5951 
5952 /*
5953  * Shift a single extent.
5954  */
5955 STATIC int
5956 xfs_bmse_shift_one(
5957 	struct xfs_inode		*ip,
5958 	int				whichfork,
5959 	xfs_fileoff_t			offset_shift_fsb,
5960 	int				*current_ext,
5961 	struct xfs_bmbt_irec		*got,
5962 	struct xfs_btree_cur		*cur,
5963 	int				*logflags,
5964 	enum shift_direction		direction,
5965 	struct xfs_defer_ops		*dfops)
5966 {
5967 	struct xfs_ifork		*ifp;
5968 	struct xfs_mount		*mp;
5969 	xfs_fileoff_t			startoff;
5970 	struct xfs_bmbt_irec		adj_irec, new;
5971 	int				error;
5972 	int				i;
5973 	int				total_extents;
5974 
5975 	mp = ip->i_mount;
5976 	ifp = XFS_IFORK_PTR(ip, whichfork);
5977 	total_extents = xfs_iext_count(ifp);
5978 
5979 	/* delalloc extents should be prevented by caller */
5980 	XFS_WANT_CORRUPTED_RETURN(mp, !isnullstartblock(got->br_startblock));
5981 
5982 	if (direction == SHIFT_LEFT) {
5983 		startoff = got->br_startoff - offset_shift_fsb;
5984 
5985 		/*
5986 		 * Check for merge if we've got an extent to the left,
5987 		 * otherwise make sure there's enough room at the start
5988 		 * of the file for the shift.
5989 		 */
5990 		if (!*current_ext) {
5991 			if (got->br_startoff < offset_shift_fsb)
5992 				return -EINVAL;
5993 			goto update_current_ext;
5994 		}
5995 
5996 		/*
5997 		 * grab the left extent and check for a large enough hole.
5998 		 */
5999 		xfs_iext_get_extent(ifp, *current_ext - 1, &adj_irec);
6000 		if (startoff < adj_irec.br_startoff + adj_irec.br_blockcount)
6001 			return -EINVAL;
6002 
6003 		/* check whether to merge the extent or shift it down */
6004 		if (xfs_bmse_can_merge(&adj_irec, got, offset_shift_fsb)) {
6005 			return xfs_bmse_merge(ip, whichfork, offset_shift_fsb,
6006 					      *current_ext, got, &adj_irec,
6007 					      cur, logflags, dfops);
6008 		}
6009 	} else {
6010 		startoff = got->br_startoff + offset_shift_fsb;
6011 		/* nothing to move if this is the last extent */
6012 		if (*current_ext >= (total_extents - 1))
6013 			goto update_current_ext;
6014 
6015 		/*
6016 		 * If this is not the last extent in the file, make sure there
6017 		 * is enough room between current extent and next extent for
6018 		 * accommodating the shift.
6019 		 */
6020 		xfs_iext_get_extent(ifp, *current_ext + 1, &adj_irec);
6021 		if (startoff + got->br_blockcount > adj_irec.br_startoff)
6022 			return -EINVAL;
6023 
6024 		/*
6025 		 * Unlike a left shift (which involves a hole punch),
6026 		 * a right shift does not modify extent neighbors
6027 		 * in any way. We should never find mergeable extents
6028 		 * in this scenario. Check anyways and warn if we
6029 		 * encounter two extents that could be one.
6030 		 */
6031 		if (xfs_bmse_can_merge(got, &adj_irec, offset_shift_fsb))
6032 			WARN_ON_ONCE(1);
6033 	}
6034 
6035 	/*
6036 	 * Increment the extent index for the next iteration, update the start
6037 	 * offset of the in-core extent and update the btree if applicable.
6038 	 */
6039 update_current_ext:
6040 	*logflags |= XFS_ILOG_CORE;
6041 
6042 	new = *got;
6043 	new.br_startoff = startoff;
6044 
6045 	if (cur) {
6046 		error = xfs_bmbt_lookup_eq(cur, got->br_startoff,
6047 				got->br_startblock, got->br_blockcount, &i);
6048 		if (error)
6049 			return error;
6050 		XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
6051 
6052 		error = xfs_bmbt_update(cur, new.br_startoff,
6053 				new.br_startblock, new.br_blockcount,
6054 				new.br_state);
6055 		if (error)
6056 			return error;
6057 	} else {
6058 		*logflags |= XFS_ILOG_DEXT;
6059 	}
6060 
6061 	xfs_iext_update_extent(ifp, *current_ext, &new);
6062 
6063 	if (direction == SHIFT_LEFT)
6064 		(*current_ext)++;
6065 	else
6066 		(*current_ext)--;
6067 
6068 	/* update reverse mapping */
6069 	error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, got);
6070 	if (error)
6071 		return error;
6072 	return xfs_rmap_map_extent(mp, dfops, ip, whichfork, &new);
6073 }
6074 
6075 /*
6076  * Shift extent records to the left/right to cover/create a hole.
6077  *
6078  * The maximum number of extents to be shifted in a single operation is
6079  * @num_exts. @stop_fsb specifies the file offset at which to stop shift and the
6080  * file offset where we've left off is returned in @next_fsb. @offset_shift_fsb
6081  * is the length by which each extent is shifted. If there is no hole to shift
6082  * the extents into, this will be considered invalid operation and we abort
6083  * immediately.
6084  */
6085 int
6086 xfs_bmap_shift_extents(
6087 	struct xfs_trans	*tp,
6088 	struct xfs_inode	*ip,
6089 	xfs_fileoff_t		*next_fsb,
6090 	xfs_fileoff_t		offset_shift_fsb,
6091 	int			*done,
6092 	xfs_fileoff_t		stop_fsb,
6093 	xfs_fsblock_t		*firstblock,
6094 	struct xfs_defer_ops	*dfops,
6095 	enum shift_direction	direction,
6096 	int			num_exts)
6097 {
6098 	struct xfs_btree_cur		*cur = NULL;
6099 	struct xfs_bmbt_irec            got;
6100 	struct xfs_mount		*mp = ip->i_mount;
6101 	struct xfs_ifork		*ifp;
6102 	xfs_extnum_t			nexts = 0;
6103 	xfs_extnum_t			current_ext;
6104 	xfs_extnum_t			total_extents;
6105 	xfs_extnum_t			stop_extent;
6106 	int				error = 0;
6107 	int				whichfork = XFS_DATA_FORK;
6108 	int				logflags = 0;
6109 
6110 	if (unlikely(XFS_TEST_ERROR(
6111 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
6112 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
6113 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
6114 		XFS_ERROR_REPORT("xfs_bmap_shift_extents",
6115 				 XFS_ERRLEVEL_LOW, mp);
6116 		return -EFSCORRUPTED;
6117 	}
6118 
6119 	if (XFS_FORCED_SHUTDOWN(mp))
6120 		return -EIO;
6121 
6122 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
6123 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6124 	ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT);
6125 
6126 	ifp = XFS_IFORK_PTR(ip, whichfork);
6127 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
6128 		/* Read in all the extents */
6129 		error = xfs_iread_extents(tp, ip, whichfork);
6130 		if (error)
6131 			return error;
6132 	}
6133 
6134 	if (ifp->if_flags & XFS_IFBROOT) {
6135 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6136 		cur->bc_private.b.firstblock = *firstblock;
6137 		cur->bc_private.b.dfops = dfops;
6138 		cur->bc_private.b.flags = 0;
6139 	}
6140 
6141 	/*
6142 	 * There may be delalloc extents in the data fork before the range we
6143 	 * are collapsing out, so we cannot use the count of real extents here.
6144 	 * Instead we have to calculate it from the incore fork.
6145 	 */
6146 	total_extents = xfs_iext_count(ifp);
6147 	if (total_extents == 0) {
6148 		*done = 1;
6149 		goto del_cursor;
6150 	}
6151 
6152 	/*
6153 	 * In case of first right shift, we need to initialize next_fsb
6154 	 */
6155 	if (*next_fsb == NULLFSBLOCK) {
6156 		ASSERT(direction == SHIFT_RIGHT);
6157 
6158 		current_ext = total_extents - 1;
6159 		xfs_iext_get_extent(ifp, current_ext, &got);
6160 		if (stop_fsb > got.br_startoff) {
6161 			*done = 1;
6162 			goto del_cursor;
6163 		}
6164 		*next_fsb = got.br_startoff;
6165 	} else {
6166 		/*
6167 		 * Look up the extent index for the fsb where we start shifting. We can
6168 		 * henceforth iterate with current_ext as extent list changes are locked
6169 		 * out via ilock.
6170 		 *
6171 		 * If next_fsb lies in a hole beyond which there are no extents we are
6172 		 * done.
6173 		 */
6174 		if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &current_ext,
6175 				&got)) {
6176 			*done = 1;
6177 			goto del_cursor;
6178 		}
6179 	}
6180 
6181 	/* Lookup the extent index at which we have to stop */
6182 	if (direction == SHIFT_RIGHT) {
6183 		struct xfs_bmbt_irec s;
6184 
6185 		xfs_iext_lookup_extent(ip, ifp, stop_fsb, &stop_extent, &s);
6186 		/* Make stop_extent exclusive of shift range */
6187 		stop_extent--;
6188 		if (current_ext <= stop_extent) {
6189 			error = -EIO;
6190 			goto del_cursor;
6191 		}
6192 	} else {
6193 		stop_extent = total_extents;
6194 		if (current_ext >= stop_extent) {
6195 			error = -EIO;
6196 			goto del_cursor;
6197 		}
6198 	}
6199 
6200 	while (nexts++ < num_exts) {
6201 		error = xfs_bmse_shift_one(ip, whichfork, offset_shift_fsb,
6202 					   &current_ext, &got, cur, &logflags,
6203 					   direction, dfops);
6204 		if (error)
6205 			goto del_cursor;
6206 		/*
6207 		 * If there was an extent merge during the shift, the extent
6208 		 * count can change. Update the total and grade the next record.
6209 		 */
6210 		if (direction == SHIFT_LEFT) {
6211 			total_extents = xfs_iext_count(ifp);
6212 			stop_extent = total_extents;
6213 		}
6214 
6215 		if (current_ext == stop_extent) {
6216 			*done = 1;
6217 			*next_fsb = NULLFSBLOCK;
6218 			break;
6219 		}
6220 		xfs_iext_get_extent(ifp, current_ext, &got);
6221 	}
6222 
6223 	if (!*done)
6224 		*next_fsb = got.br_startoff;
6225 
6226 del_cursor:
6227 	if (cur)
6228 		xfs_btree_del_cursor(cur,
6229 			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
6230 
6231 	if (logflags)
6232 		xfs_trans_log_inode(tp, ip, logflags);
6233 
6234 	return error;
6235 }
6236 
6237 /*
6238  * Splits an extent into two extents at split_fsb block such that it is
6239  * the first block of the current_ext. @current_ext is a target extent
6240  * to be split. @split_fsb is a block where the extents is split.
6241  * If split_fsb lies in a hole or the first block of extents, just return 0.
6242  */
6243 STATIC int
6244 xfs_bmap_split_extent_at(
6245 	struct xfs_trans	*tp,
6246 	struct xfs_inode	*ip,
6247 	xfs_fileoff_t		split_fsb,
6248 	xfs_fsblock_t		*firstfsb,
6249 	struct xfs_defer_ops	*dfops)
6250 {
6251 	int				whichfork = XFS_DATA_FORK;
6252 	struct xfs_btree_cur		*cur = NULL;
6253 	struct xfs_bmbt_irec		got;
6254 	struct xfs_bmbt_irec		new; /* split extent */
6255 	struct xfs_mount		*mp = ip->i_mount;
6256 	struct xfs_ifork		*ifp;
6257 	xfs_fsblock_t			gotblkcnt; /* new block count for got */
6258 	xfs_extnum_t			current_ext;
6259 	int				error = 0;
6260 	int				logflags = 0;
6261 	int				i = 0;
6262 
6263 	if (unlikely(XFS_TEST_ERROR(
6264 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
6265 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
6266 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
6267 		XFS_ERROR_REPORT("xfs_bmap_split_extent_at",
6268 				 XFS_ERRLEVEL_LOW, mp);
6269 		return -EFSCORRUPTED;
6270 	}
6271 
6272 	if (XFS_FORCED_SHUTDOWN(mp))
6273 		return -EIO;
6274 
6275 	ifp = XFS_IFORK_PTR(ip, whichfork);
6276 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
6277 		/* Read in all the extents */
6278 		error = xfs_iread_extents(tp, ip, whichfork);
6279 		if (error)
6280 			return error;
6281 	}
6282 
6283 	/*
6284 	 * If there are not extents, or split_fsb lies in a hole we are done.
6285 	 */
6286 	if (!xfs_iext_lookup_extent(ip, ifp, split_fsb, &current_ext, &got) ||
6287 	    got.br_startoff >= split_fsb)
6288 		return 0;
6289 
6290 	gotblkcnt = split_fsb - got.br_startoff;
6291 	new.br_startoff = split_fsb;
6292 	new.br_startblock = got.br_startblock + gotblkcnt;
6293 	new.br_blockcount = got.br_blockcount - gotblkcnt;
6294 	new.br_state = got.br_state;
6295 
6296 	if (ifp->if_flags & XFS_IFBROOT) {
6297 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6298 		cur->bc_private.b.firstblock = *firstfsb;
6299 		cur->bc_private.b.dfops = dfops;
6300 		cur->bc_private.b.flags = 0;
6301 		error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
6302 				got.br_startblock,
6303 				got.br_blockcount,
6304 				&i);
6305 		if (error)
6306 			goto del_cursor;
6307 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
6308 	}
6309 
6310 	got.br_blockcount = gotblkcnt;
6311 	xfs_iext_update_extent(ifp, current_ext, &got);
6312 
6313 	logflags = XFS_ILOG_CORE;
6314 	if (cur) {
6315 		error = xfs_bmbt_update(cur, got.br_startoff,
6316 				got.br_startblock,
6317 				got.br_blockcount,
6318 				got.br_state);
6319 		if (error)
6320 			goto del_cursor;
6321 	} else
6322 		logflags |= XFS_ILOG_DEXT;
6323 
6324 	/* Add new extent */
6325 	current_ext++;
6326 	xfs_iext_insert(ip, current_ext, 1, &new, 0);
6327 	XFS_IFORK_NEXT_SET(ip, whichfork,
6328 			   XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
6329 
6330 	if (cur) {
6331 		error = xfs_bmbt_lookup_eq(cur, new.br_startoff,
6332 				new.br_startblock, new.br_blockcount,
6333 				&i);
6334 		if (error)
6335 			goto del_cursor;
6336 		XFS_WANT_CORRUPTED_GOTO(mp, i == 0, del_cursor);
6337 		cur->bc_rec.b.br_state = new.br_state;
6338 
6339 		error = xfs_btree_insert(cur, &i);
6340 		if (error)
6341 			goto del_cursor;
6342 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
6343 	}
6344 
6345 	/*
6346 	 * Convert to a btree if necessary.
6347 	 */
6348 	if (xfs_bmap_needs_btree(ip, whichfork)) {
6349 		int tmp_logflags; /* partial log flag return val */
6350 
6351 		ASSERT(cur == NULL);
6352 		error = xfs_bmap_extents_to_btree(tp, ip, firstfsb, dfops,
6353 				&cur, 0, &tmp_logflags, whichfork);
6354 		logflags |= tmp_logflags;
6355 	}
6356 
6357 del_cursor:
6358 	if (cur) {
6359 		cur->bc_private.b.allocated = 0;
6360 		xfs_btree_del_cursor(cur,
6361 				error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
6362 	}
6363 
6364 	if (logflags)
6365 		xfs_trans_log_inode(tp, ip, logflags);
6366 	return error;
6367 }
6368 
6369 int
6370 xfs_bmap_split_extent(
6371 	struct xfs_inode        *ip,
6372 	xfs_fileoff_t           split_fsb)
6373 {
6374 	struct xfs_mount        *mp = ip->i_mount;
6375 	struct xfs_trans        *tp;
6376 	struct xfs_defer_ops    dfops;
6377 	xfs_fsblock_t           firstfsb;
6378 	int                     error;
6379 
6380 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
6381 			XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
6382 	if (error)
6383 		return error;
6384 
6385 	xfs_ilock(ip, XFS_ILOCK_EXCL);
6386 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
6387 
6388 	xfs_defer_init(&dfops, &firstfsb);
6389 
6390 	error = xfs_bmap_split_extent_at(tp, ip, split_fsb,
6391 			&firstfsb, &dfops);
6392 	if (error)
6393 		goto out;
6394 
6395 	error = xfs_defer_finish(&tp, &dfops);
6396 	if (error)
6397 		goto out;
6398 
6399 	return xfs_trans_commit(tp);
6400 
6401 out:
6402 	xfs_defer_cancel(&dfops);
6403 	xfs_trans_cancel(tp);
6404 	return error;
6405 }
6406 
6407 /* Deferred mapping is only for real extents in the data fork. */
6408 static bool
6409 xfs_bmap_is_update_needed(
6410 	struct xfs_bmbt_irec	*bmap)
6411 {
6412 	return  bmap->br_startblock != HOLESTARTBLOCK &&
6413 		bmap->br_startblock != DELAYSTARTBLOCK;
6414 }
6415 
6416 /* Record a bmap intent. */
6417 static int
6418 __xfs_bmap_add(
6419 	struct xfs_mount		*mp,
6420 	struct xfs_defer_ops		*dfops,
6421 	enum xfs_bmap_intent_type	type,
6422 	struct xfs_inode		*ip,
6423 	int				whichfork,
6424 	struct xfs_bmbt_irec		*bmap)
6425 {
6426 	int				error;
6427 	struct xfs_bmap_intent		*bi;
6428 
6429 	trace_xfs_bmap_defer(mp,
6430 			XFS_FSB_TO_AGNO(mp, bmap->br_startblock),
6431 			type,
6432 			XFS_FSB_TO_AGBNO(mp, bmap->br_startblock),
6433 			ip->i_ino, whichfork,
6434 			bmap->br_startoff,
6435 			bmap->br_blockcount,
6436 			bmap->br_state);
6437 
6438 	bi = kmem_alloc(sizeof(struct xfs_bmap_intent), KM_SLEEP | KM_NOFS);
6439 	INIT_LIST_HEAD(&bi->bi_list);
6440 	bi->bi_type = type;
6441 	bi->bi_owner = ip;
6442 	bi->bi_whichfork = whichfork;
6443 	bi->bi_bmap = *bmap;
6444 
6445 	error = xfs_defer_ijoin(dfops, bi->bi_owner);
6446 	if (error) {
6447 		kmem_free(bi);
6448 		return error;
6449 	}
6450 
6451 	xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list);
6452 	return 0;
6453 }
6454 
6455 /* Map an extent into a file. */
6456 int
6457 xfs_bmap_map_extent(
6458 	struct xfs_mount	*mp,
6459 	struct xfs_defer_ops	*dfops,
6460 	struct xfs_inode	*ip,
6461 	struct xfs_bmbt_irec	*PREV)
6462 {
6463 	if (!xfs_bmap_is_update_needed(PREV))
6464 		return 0;
6465 
6466 	return __xfs_bmap_add(mp, dfops, XFS_BMAP_MAP, ip,
6467 			XFS_DATA_FORK, PREV);
6468 }
6469 
6470 /* Unmap an extent out of a file. */
6471 int
6472 xfs_bmap_unmap_extent(
6473 	struct xfs_mount	*mp,
6474 	struct xfs_defer_ops	*dfops,
6475 	struct xfs_inode	*ip,
6476 	struct xfs_bmbt_irec	*PREV)
6477 {
6478 	if (!xfs_bmap_is_update_needed(PREV))
6479 		return 0;
6480 
6481 	return __xfs_bmap_add(mp, dfops, XFS_BMAP_UNMAP, ip,
6482 			XFS_DATA_FORK, PREV);
6483 }
6484 
6485 /*
6486  * Process one of the deferred bmap operations.  We pass back the
6487  * btree cursor to maintain our lock on the bmapbt between calls.
6488  */
6489 int
6490 xfs_bmap_finish_one(
6491 	struct xfs_trans		*tp,
6492 	struct xfs_defer_ops		*dfops,
6493 	struct xfs_inode		*ip,
6494 	enum xfs_bmap_intent_type	type,
6495 	int				whichfork,
6496 	xfs_fileoff_t			startoff,
6497 	xfs_fsblock_t			startblock,
6498 	xfs_filblks_t			*blockcount,
6499 	xfs_exntst_t			state)
6500 {
6501 	xfs_fsblock_t			firstfsb;
6502 	int				error = 0;
6503 
6504 	/*
6505 	 * firstfsb is tied to the transaction lifetime and is used to
6506 	 * ensure correct AG locking order and schedule work item
6507 	 * continuations.  XFS_BUI_MAX_FAST_EXTENTS (== 1) restricts us
6508 	 * to only making one bmap call per transaction, so it should
6509 	 * be safe to have it as a local variable here.
6510 	 */
6511 	firstfsb = NULLFSBLOCK;
6512 
6513 	trace_xfs_bmap_deferred(tp->t_mountp,
6514 			XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type,
6515 			XFS_FSB_TO_AGBNO(tp->t_mountp, startblock),
6516 			ip->i_ino, whichfork, startoff, *blockcount, state);
6517 
6518 	if (WARN_ON_ONCE(whichfork != XFS_DATA_FORK))
6519 		return -EFSCORRUPTED;
6520 
6521 	if (XFS_TEST_ERROR(false, tp->t_mountp,
6522 			XFS_ERRTAG_BMAP_FINISH_ONE))
6523 		return -EIO;
6524 
6525 	switch (type) {
6526 	case XFS_BMAP_MAP:
6527 		error = xfs_bmapi_remap(tp, ip, startoff, *blockcount,
6528 				startblock, dfops);
6529 		*blockcount = 0;
6530 		break;
6531 	case XFS_BMAP_UNMAP:
6532 		error = __xfs_bunmapi(tp, ip, startoff, blockcount,
6533 				XFS_BMAPI_REMAP, 1, &firstfsb, dfops);
6534 		break;
6535 	default:
6536 		ASSERT(0);
6537 		error = -EFSCORRUPTED;
6538 	}
6539 
6540 	return error;
6541 }
6542