xref: /openbmc/linux/fs/xfs/libxfs/xfs_bmap.c (revision 07c7c6bf)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4  * All Rights Reserved.
5  */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_bit.h"
13 #include "xfs_sb.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_da_format.h"
17 #include "xfs_da_btree.h"
18 #include "xfs_dir2.h"
19 #include "xfs_inode.h"
20 #include "xfs_btree.h"
21 #include "xfs_trans.h"
22 #include "xfs_inode_item.h"
23 #include "xfs_extfree_item.h"
24 #include "xfs_alloc.h"
25 #include "xfs_bmap.h"
26 #include "xfs_bmap_util.h"
27 #include "xfs_bmap_btree.h"
28 #include "xfs_rtalloc.h"
29 #include "xfs_errortag.h"
30 #include "xfs_error.h"
31 #include "xfs_quota.h"
32 #include "xfs_trans_space.h"
33 #include "xfs_buf_item.h"
34 #include "xfs_trace.h"
35 #include "xfs_symlink.h"
36 #include "xfs_attr_leaf.h"
37 #include "xfs_filestream.h"
38 #include "xfs_rmap.h"
39 #include "xfs_ag_resv.h"
40 #include "xfs_refcount.h"
41 #include "xfs_icache.h"
42 
43 
44 kmem_zone_t		*xfs_bmap_free_item_zone;
45 
46 /*
47  * Miscellaneous helper functions
48  */
49 
50 /*
51  * Compute and fill in the value of the maximum depth of a bmap btree
52  * in this filesystem.  Done once, during mount.
53  */
54 void
55 xfs_bmap_compute_maxlevels(
56 	xfs_mount_t	*mp,		/* file system mount structure */
57 	int		whichfork)	/* data or attr fork */
58 {
59 	int		level;		/* btree level */
60 	uint		maxblocks;	/* max blocks at this level */
61 	uint		maxleafents;	/* max leaf entries possible */
62 	int		maxrootrecs;	/* max records in root block */
63 	int		minleafrecs;	/* min records in leaf block */
64 	int		minnoderecs;	/* min records in node block */
65 	int		sz;		/* root block size */
66 
67 	/*
68 	 * The maximum number of extents in a file, hence the maximum
69 	 * number of leaf entries, is controlled by the type of di_nextents
70 	 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
71 	 * (a signed 16-bit number, xfs_aextnum_t).
72 	 *
73 	 * Note that we can no longer assume that if we are in ATTR1 that
74 	 * the fork offset of all the inodes will be
75 	 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
76 	 * with ATTR2 and then mounted back with ATTR1, keeping the
77 	 * di_forkoff's fixed but probably at various positions. Therefore,
78 	 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
79 	 * of a minimum size available.
80 	 */
81 	if (whichfork == XFS_DATA_FORK) {
82 		maxleafents = MAXEXTNUM;
83 		sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
84 	} else {
85 		maxleafents = MAXAEXTNUM;
86 		sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
87 	}
88 	maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
89 	minleafrecs = mp->m_bmap_dmnr[0];
90 	minnoderecs = mp->m_bmap_dmnr[1];
91 	maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
92 	for (level = 1; maxblocks > 1; level++) {
93 		if (maxblocks <= maxrootrecs)
94 			maxblocks = 1;
95 		else
96 			maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
97 	}
98 	mp->m_bm_maxlevels[whichfork] = level;
99 }
100 
101 STATIC int				/* error */
102 xfs_bmbt_lookup_eq(
103 	struct xfs_btree_cur	*cur,
104 	struct xfs_bmbt_irec	*irec,
105 	int			*stat)	/* success/failure */
106 {
107 	cur->bc_rec.b = *irec;
108 	return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
109 }
110 
111 STATIC int				/* error */
112 xfs_bmbt_lookup_first(
113 	struct xfs_btree_cur	*cur,
114 	int			*stat)	/* success/failure */
115 {
116 	cur->bc_rec.b.br_startoff = 0;
117 	cur->bc_rec.b.br_startblock = 0;
118 	cur->bc_rec.b.br_blockcount = 0;
119 	return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
120 }
121 
122 /*
123  * Check if the inode needs to be converted to btree format.
124  */
125 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
126 {
127 	return whichfork != XFS_COW_FORK &&
128 		XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
129 		XFS_IFORK_NEXTENTS(ip, whichfork) >
130 			XFS_IFORK_MAXEXT(ip, whichfork);
131 }
132 
133 /*
134  * Check if the inode should be converted to extent format.
135  */
136 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
137 {
138 	return whichfork != XFS_COW_FORK &&
139 		XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
140 		XFS_IFORK_NEXTENTS(ip, whichfork) <=
141 			XFS_IFORK_MAXEXT(ip, whichfork);
142 }
143 
144 /*
145  * Update the record referred to by cur to the value given by irec
146  * This either works (return 0) or gets an EFSCORRUPTED error.
147  */
148 STATIC int
149 xfs_bmbt_update(
150 	struct xfs_btree_cur	*cur,
151 	struct xfs_bmbt_irec	*irec)
152 {
153 	union xfs_btree_rec	rec;
154 
155 	xfs_bmbt_disk_set_all(&rec.bmbt, irec);
156 	return xfs_btree_update(cur, &rec);
157 }
158 
159 /*
160  * Compute the worst-case number of indirect blocks that will be used
161  * for ip's delayed extent of length "len".
162  */
163 STATIC xfs_filblks_t
164 xfs_bmap_worst_indlen(
165 	xfs_inode_t	*ip,		/* incore inode pointer */
166 	xfs_filblks_t	len)		/* delayed extent length */
167 {
168 	int		level;		/* btree level number */
169 	int		maxrecs;	/* maximum record count at this level */
170 	xfs_mount_t	*mp;		/* mount structure */
171 	xfs_filblks_t	rval;		/* return value */
172 
173 	mp = ip->i_mount;
174 	maxrecs = mp->m_bmap_dmxr[0];
175 	for (level = 0, rval = 0;
176 	     level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
177 	     level++) {
178 		len += maxrecs - 1;
179 		do_div(len, maxrecs);
180 		rval += len;
181 		if (len == 1)
182 			return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
183 				level - 1;
184 		if (level == 0)
185 			maxrecs = mp->m_bmap_dmxr[1];
186 	}
187 	return rval;
188 }
189 
190 /*
191  * Calculate the default attribute fork offset for newly created inodes.
192  */
193 uint
194 xfs_default_attroffset(
195 	struct xfs_inode	*ip)
196 {
197 	struct xfs_mount	*mp = ip->i_mount;
198 	uint			offset;
199 
200 	if (mp->m_sb.sb_inodesize == 256) {
201 		offset = XFS_LITINO(mp, ip->i_d.di_version) -
202 				XFS_BMDR_SPACE_CALC(MINABTPTRS);
203 	} else {
204 		offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
205 	}
206 
207 	ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
208 	return offset;
209 }
210 
211 /*
212  * Helper routine to reset inode di_forkoff field when switching
213  * attribute fork from local to extent format - we reset it where
214  * possible to make space available for inline data fork extents.
215  */
216 STATIC void
217 xfs_bmap_forkoff_reset(
218 	xfs_inode_t	*ip,
219 	int		whichfork)
220 {
221 	if (whichfork == XFS_ATTR_FORK &&
222 	    ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
223 	    ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
224 		uint	dfl_forkoff = xfs_default_attroffset(ip) >> 3;
225 
226 		if (dfl_forkoff > ip->i_d.di_forkoff)
227 			ip->i_d.di_forkoff = dfl_forkoff;
228 	}
229 }
230 
231 #ifdef DEBUG
232 STATIC struct xfs_buf *
233 xfs_bmap_get_bp(
234 	struct xfs_btree_cur	*cur,
235 	xfs_fsblock_t		bno)
236 {
237 	struct xfs_log_item	*lip;
238 	int			i;
239 
240 	if (!cur)
241 		return NULL;
242 
243 	for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
244 		if (!cur->bc_bufs[i])
245 			break;
246 		if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
247 			return cur->bc_bufs[i];
248 	}
249 
250 	/* Chase down all the log items to see if the bp is there */
251 	list_for_each_entry(lip, &cur->bc_tp->t_items, li_trans) {
252 		struct xfs_buf_log_item	*bip = (struct xfs_buf_log_item *)lip;
253 
254 		if (bip->bli_item.li_type == XFS_LI_BUF &&
255 		    XFS_BUF_ADDR(bip->bli_buf) == bno)
256 			return bip->bli_buf;
257 	}
258 
259 	return NULL;
260 }
261 
262 STATIC void
263 xfs_check_block(
264 	struct xfs_btree_block	*block,
265 	xfs_mount_t		*mp,
266 	int			root,
267 	short			sz)
268 {
269 	int			i, j, dmxr;
270 	__be64			*pp, *thispa;	/* pointer to block address */
271 	xfs_bmbt_key_t		*prevp, *keyp;
272 
273 	ASSERT(be16_to_cpu(block->bb_level) > 0);
274 
275 	prevp = NULL;
276 	for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
277 		dmxr = mp->m_bmap_dmxr[0];
278 		keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
279 
280 		if (prevp) {
281 			ASSERT(be64_to_cpu(prevp->br_startoff) <
282 			       be64_to_cpu(keyp->br_startoff));
283 		}
284 		prevp = keyp;
285 
286 		/*
287 		 * Compare the block numbers to see if there are dups.
288 		 */
289 		if (root)
290 			pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
291 		else
292 			pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
293 
294 		for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
295 			if (root)
296 				thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
297 			else
298 				thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
299 			if (*thispa == *pp) {
300 				xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
301 					__func__, j, i,
302 					(unsigned long long)be64_to_cpu(*thispa));
303 				xfs_err(mp, "%s: ptrs are equal in node\n",
304 					__func__);
305 				xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
306 			}
307 		}
308 	}
309 }
310 
311 /*
312  * Check that the extents for the inode ip are in the right order in all
313  * btree leaves. THis becomes prohibitively expensive for large extent count
314  * files, so don't bother with inodes that have more than 10,000 extents in
315  * them. The btree record ordering checks will still be done, so for such large
316  * bmapbt constructs that is going to catch most corruptions.
317  */
318 STATIC void
319 xfs_bmap_check_leaf_extents(
320 	xfs_btree_cur_t		*cur,	/* btree cursor or null */
321 	xfs_inode_t		*ip,		/* incore inode pointer */
322 	int			whichfork)	/* data or attr fork */
323 {
324 	struct xfs_btree_block	*block;	/* current btree block */
325 	xfs_fsblock_t		bno;	/* block # of "block" */
326 	xfs_buf_t		*bp;	/* buffer for "block" */
327 	int			error;	/* error return value */
328 	xfs_extnum_t		i=0, j;	/* index into the extents list */
329 	struct xfs_ifork	*ifp;	/* fork structure */
330 	int			level;	/* btree level, for checking */
331 	xfs_mount_t		*mp;	/* file system mount structure */
332 	__be64			*pp;	/* pointer to block address */
333 	xfs_bmbt_rec_t		*ep;	/* pointer to current extent */
334 	xfs_bmbt_rec_t		last = {0, 0}; /* last extent in prev block */
335 	xfs_bmbt_rec_t		*nextp;	/* pointer to next extent */
336 	int			bp_release = 0;
337 
338 	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
339 		return;
340 	}
341 
342 	/* skip large extent count inodes */
343 	if (ip->i_d.di_nextents > 10000)
344 		return;
345 
346 	bno = NULLFSBLOCK;
347 	mp = ip->i_mount;
348 	ifp = XFS_IFORK_PTR(ip, whichfork);
349 	block = ifp->if_broot;
350 	/*
351 	 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
352 	 */
353 	level = be16_to_cpu(block->bb_level);
354 	ASSERT(level > 0);
355 	xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
356 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
357 	bno = be64_to_cpu(*pp);
358 
359 	ASSERT(bno != NULLFSBLOCK);
360 	ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
361 	ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
362 
363 	/*
364 	 * Go down the tree until leaf level is reached, following the first
365 	 * pointer (leftmost) at each level.
366 	 */
367 	while (level-- > 0) {
368 		/* See if buf is in cur first */
369 		bp_release = 0;
370 		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
371 		if (!bp) {
372 			bp_release = 1;
373 			error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
374 						XFS_BMAP_BTREE_REF,
375 						&xfs_bmbt_buf_ops);
376 			if (error)
377 				goto error_norelse;
378 		}
379 		block = XFS_BUF_TO_BLOCK(bp);
380 		if (level == 0)
381 			break;
382 
383 		/*
384 		 * Check this block for basic sanity (increasing keys and
385 		 * no duplicate blocks).
386 		 */
387 
388 		xfs_check_block(block, mp, 0, 0);
389 		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
390 		bno = be64_to_cpu(*pp);
391 		XFS_WANT_CORRUPTED_GOTO(mp,
392 					xfs_verify_fsbno(mp, bno), error0);
393 		if (bp_release) {
394 			bp_release = 0;
395 			xfs_trans_brelse(NULL, bp);
396 		}
397 	}
398 
399 	/*
400 	 * Here with bp and block set to the leftmost leaf node in the tree.
401 	 */
402 	i = 0;
403 
404 	/*
405 	 * Loop over all leaf nodes checking that all extents are in the right order.
406 	 */
407 	for (;;) {
408 		xfs_fsblock_t	nextbno;
409 		xfs_extnum_t	num_recs;
410 
411 
412 		num_recs = xfs_btree_get_numrecs(block);
413 
414 		/*
415 		 * Read-ahead the next leaf block, if any.
416 		 */
417 
418 		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
419 
420 		/*
421 		 * Check all the extents to make sure they are OK.
422 		 * If we had a previous block, the last entry should
423 		 * conform with the first entry in this one.
424 		 */
425 
426 		ep = XFS_BMBT_REC_ADDR(mp, block, 1);
427 		if (i) {
428 			ASSERT(xfs_bmbt_disk_get_startoff(&last) +
429 			       xfs_bmbt_disk_get_blockcount(&last) <=
430 			       xfs_bmbt_disk_get_startoff(ep));
431 		}
432 		for (j = 1; j < num_recs; j++) {
433 			nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
434 			ASSERT(xfs_bmbt_disk_get_startoff(ep) +
435 			       xfs_bmbt_disk_get_blockcount(ep) <=
436 			       xfs_bmbt_disk_get_startoff(nextp));
437 			ep = nextp;
438 		}
439 
440 		last = *ep;
441 		i += num_recs;
442 		if (bp_release) {
443 			bp_release = 0;
444 			xfs_trans_brelse(NULL, bp);
445 		}
446 		bno = nextbno;
447 		/*
448 		 * If we've reached the end, stop.
449 		 */
450 		if (bno == NULLFSBLOCK)
451 			break;
452 
453 		bp_release = 0;
454 		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
455 		if (!bp) {
456 			bp_release = 1;
457 			error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
458 						XFS_BMAP_BTREE_REF,
459 						&xfs_bmbt_buf_ops);
460 			if (error)
461 				goto error_norelse;
462 		}
463 		block = XFS_BUF_TO_BLOCK(bp);
464 	}
465 
466 	return;
467 
468 error0:
469 	xfs_warn(mp, "%s: at error0", __func__);
470 	if (bp_release)
471 		xfs_trans_brelse(NULL, bp);
472 error_norelse:
473 	xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
474 		__func__, i);
475 	xfs_err(mp, "%s: CORRUPTED BTREE OR SOMETHING", __func__);
476 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
477 	return;
478 }
479 
480 /*
481  * Validate that the bmbt_irecs being returned from bmapi are valid
482  * given the caller's original parameters.  Specifically check the
483  * ranges of the returned irecs to ensure that they only extend beyond
484  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
485  */
486 STATIC void
487 xfs_bmap_validate_ret(
488 	xfs_fileoff_t		bno,
489 	xfs_filblks_t		len,
490 	int			flags,
491 	xfs_bmbt_irec_t		*mval,
492 	int			nmap,
493 	int			ret_nmap)
494 {
495 	int			i;		/* index to map values */
496 
497 	ASSERT(ret_nmap <= nmap);
498 
499 	for (i = 0; i < ret_nmap; i++) {
500 		ASSERT(mval[i].br_blockcount > 0);
501 		if (!(flags & XFS_BMAPI_ENTIRE)) {
502 			ASSERT(mval[i].br_startoff >= bno);
503 			ASSERT(mval[i].br_blockcount <= len);
504 			ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
505 			       bno + len);
506 		} else {
507 			ASSERT(mval[i].br_startoff < bno + len);
508 			ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
509 			       bno);
510 		}
511 		ASSERT(i == 0 ||
512 		       mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
513 		       mval[i].br_startoff);
514 		ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
515 		       mval[i].br_startblock != HOLESTARTBLOCK);
516 		ASSERT(mval[i].br_state == XFS_EXT_NORM ||
517 		       mval[i].br_state == XFS_EXT_UNWRITTEN);
518 	}
519 }
520 
521 #else
522 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)		do { } while (0)
523 #define	xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)	do { } while (0)
524 #endif /* DEBUG */
525 
526 /*
527  * bmap free list manipulation functions
528  */
529 
530 /*
531  * Add the extent to the list of extents to be free at transaction end.
532  * The list is maintained sorted (by block number).
533  */
534 void
535 __xfs_bmap_add_free(
536 	struct xfs_trans		*tp,
537 	xfs_fsblock_t			bno,
538 	xfs_filblks_t			len,
539 	const struct xfs_owner_info	*oinfo,
540 	bool				skip_discard)
541 {
542 	struct xfs_extent_free_item	*new;		/* new element */
543 #ifdef DEBUG
544 	struct xfs_mount		*mp = tp->t_mountp;
545 	xfs_agnumber_t			agno;
546 	xfs_agblock_t			agbno;
547 
548 	ASSERT(bno != NULLFSBLOCK);
549 	ASSERT(len > 0);
550 	ASSERT(len <= MAXEXTLEN);
551 	ASSERT(!isnullstartblock(bno));
552 	agno = XFS_FSB_TO_AGNO(mp, bno);
553 	agbno = XFS_FSB_TO_AGBNO(mp, bno);
554 	ASSERT(agno < mp->m_sb.sb_agcount);
555 	ASSERT(agbno < mp->m_sb.sb_agblocks);
556 	ASSERT(len < mp->m_sb.sb_agblocks);
557 	ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
558 #endif
559 	ASSERT(xfs_bmap_free_item_zone != NULL);
560 
561 	new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
562 	new->xefi_startblock = bno;
563 	new->xefi_blockcount = (xfs_extlen_t)len;
564 	if (oinfo)
565 		new->xefi_oinfo = *oinfo;
566 	else
567 		new->xefi_oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
568 	new->xefi_skip_discard = skip_discard;
569 	trace_xfs_bmap_free_defer(tp->t_mountp,
570 			XFS_FSB_TO_AGNO(tp->t_mountp, bno), 0,
571 			XFS_FSB_TO_AGBNO(tp->t_mountp, bno), len);
572 	xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list);
573 }
574 
575 /*
576  * Inode fork format manipulation functions
577  */
578 
579 /*
580  * Convert the inode format to extent format if it currently is in btree format,
581  * but the extent list is small enough that it fits into the extent format.
582  *
583  * Since the extents are already in-core, all we have to do is give up the space
584  * for the btree root and pitch the leaf block.
585  */
586 STATIC int				/* error */
587 xfs_bmap_btree_to_extents(
588 	struct xfs_trans	*tp,	/* transaction pointer */
589 	struct xfs_inode	*ip,	/* incore inode pointer */
590 	struct xfs_btree_cur	*cur,	/* btree cursor */
591 	int			*logflagsp, /* inode logging flags */
592 	int			whichfork)  /* data or attr fork */
593 {
594 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
595 	struct xfs_mount	*mp = ip->i_mount;
596 	struct xfs_btree_block	*rblock = ifp->if_broot;
597 	struct xfs_btree_block	*cblock;/* child btree block */
598 	xfs_fsblock_t		cbno;	/* child block number */
599 	xfs_buf_t		*cbp;	/* child block's buffer */
600 	int			error;	/* error return value */
601 	__be64			*pp;	/* ptr to block address */
602 	struct xfs_owner_info	oinfo;
603 
604 	/* check if we actually need the extent format first: */
605 	if (!xfs_bmap_wants_extents(ip, whichfork))
606 		return 0;
607 
608 	ASSERT(cur);
609 	ASSERT(whichfork != XFS_COW_FORK);
610 	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
611 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
612 	ASSERT(be16_to_cpu(rblock->bb_level) == 1);
613 	ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
614 	ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
615 
616 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
617 	cbno = be64_to_cpu(*pp);
618 #ifdef DEBUG
619 	XFS_WANT_CORRUPTED_RETURN(cur->bc_mp,
620 			xfs_btree_check_lptr(cur, cbno, 1));
621 #endif
622 	error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
623 				&xfs_bmbt_buf_ops);
624 	if (error)
625 		return error;
626 	cblock = XFS_BUF_TO_BLOCK(cbp);
627 	if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
628 		return error;
629 	xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
630 	xfs_bmap_add_free(cur->bc_tp, cbno, 1, &oinfo);
631 	ip->i_d.di_nblocks--;
632 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
633 	xfs_trans_binval(tp, cbp);
634 	if (cur->bc_bufs[0] == cbp)
635 		cur->bc_bufs[0] = NULL;
636 	xfs_iroot_realloc(ip, -1, whichfork);
637 	ASSERT(ifp->if_broot == NULL);
638 	ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
639 	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
640 	*logflagsp |= XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
641 	return 0;
642 }
643 
644 /*
645  * Convert an extents-format file into a btree-format file.
646  * The new file will have a root block (in the inode) and a single child block.
647  */
648 STATIC int					/* error */
649 xfs_bmap_extents_to_btree(
650 	struct xfs_trans	*tp,		/* transaction pointer */
651 	struct xfs_inode	*ip,		/* incore inode pointer */
652 	struct xfs_btree_cur	**curp,		/* cursor returned to caller */
653 	int			wasdel,		/* converting a delayed alloc */
654 	int			*logflagsp,	/* inode logging flags */
655 	int			whichfork)	/* data or attr fork */
656 {
657 	struct xfs_btree_block	*ablock;	/* allocated (child) bt block */
658 	struct xfs_buf		*abp;		/* buffer for ablock */
659 	struct xfs_alloc_arg	args;		/* allocation arguments */
660 	struct xfs_bmbt_rec	*arp;		/* child record pointer */
661 	struct xfs_btree_block	*block;		/* btree root block */
662 	struct xfs_btree_cur	*cur;		/* bmap btree cursor */
663 	int			error;		/* error return value */
664 	struct xfs_ifork	*ifp;		/* inode fork pointer */
665 	struct xfs_bmbt_key	*kp;		/* root block key pointer */
666 	struct xfs_mount	*mp;		/* mount structure */
667 	xfs_bmbt_ptr_t		*pp;		/* root block address pointer */
668 	struct xfs_iext_cursor	icur;
669 	struct xfs_bmbt_irec	rec;
670 	xfs_extnum_t		cnt = 0;
671 
672 	mp = ip->i_mount;
673 	ASSERT(whichfork != XFS_COW_FORK);
674 	ifp = XFS_IFORK_PTR(ip, whichfork);
675 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
676 
677 	/*
678 	 * Make space in the inode incore. This needs to be undone if we fail
679 	 * to expand the root.
680 	 */
681 	xfs_iroot_realloc(ip, 1, whichfork);
682 	ifp->if_flags |= XFS_IFBROOT;
683 
684 	/*
685 	 * Fill in the root.
686 	 */
687 	block = ifp->if_broot;
688 	xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
689 				 XFS_BTNUM_BMAP, 1, 1, ip->i_ino,
690 				 XFS_BTREE_LONG_PTRS);
691 	/*
692 	 * Need a cursor.  Can't allocate until bb_level is filled in.
693 	 */
694 	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
695 	cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
696 	/*
697 	 * Convert to a btree with two levels, one record in root.
698 	 */
699 	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
700 	memset(&args, 0, sizeof(args));
701 	args.tp = tp;
702 	args.mp = mp;
703 	xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork);
704 	if (tp->t_firstblock == NULLFSBLOCK) {
705 		args.type = XFS_ALLOCTYPE_START_BNO;
706 		args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
707 	} else if (tp->t_flags & XFS_TRANS_LOWMODE) {
708 		args.type = XFS_ALLOCTYPE_START_BNO;
709 		args.fsbno = tp->t_firstblock;
710 	} else {
711 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
712 		args.fsbno = tp->t_firstblock;
713 	}
714 	args.minlen = args.maxlen = args.prod = 1;
715 	args.wasdel = wasdel;
716 	*logflagsp = 0;
717 	error = xfs_alloc_vextent(&args);
718 	if (error)
719 		goto out_root_realloc;
720 
721 	if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
722 		error = -ENOSPC;
723 		goto out_root_realloc;
724 	}
725 
726 	/*
727 	 * Allocation can't fail, the space was reserved.
728 	 */
729 	ASSERT(tp->t_firstblock == NULLFSBLOCK ||
730 	       args.agno >= XFS_FSB_TO_AGNO(mp, tp->t_firstblock));
731 	tp->t_firstblock = args.fsbno;
732 	cur->bc_private.b.allocated++;
733 	ip->i_d.di_nblocks++;
734 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
735 	abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
736 	if (!abp) {
737 		error = -EFSCORRUPTED;
738 		goto out_unreserve_dquot;
739 	}
740 
741 	/*
742 	 * Fill in the child block.
743 	 */
744 	abp->b_ops = &xfs_bmbt_buf_ops;
745 	ablock = XFS_BUF_TO_BLOCK(abp);
746 	xfs_btree_init_block_int(mp, ablock, abp->b_bn,
747 				XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
748 				XFS_BTREE_LONG_PTRS);
749 
750 	for_each_xfs_iext(ifp, &icur, &rec) {
751 		if (isnullstartblock(rec.br_startblock))
752 			continue;
753 		arp = XFS_BMBT_REC_ADDR(mp, ablock, 1 + cnt);
754 		xfs_bmbt_disk_set_all(arp, &rec);
755 		cnt++;
756 	}
757 	ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
758 	xfs_btree_set_numrecs(ablock, cnt);
759 
760 	/*
761 	 * Fill in the root key and pointer.
762 	 */
763 	kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
764 	arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
765 	kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
766 	pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
767 						be16_to_cpu(block->bb_level)));
768 	*pp = cpu_to_be64(args.fsbno);
769 
770 	/*
771 	 * Do all this logging at the end so that
772 	 * the root is at the right level.
773 	 */
774 	xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
775 	xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
776 	ASSERT(*curp == NULL);
777 	*curp = cur;
778 	*logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
779 	return 0;
780 
781 out_unreserve_dquot:
782 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
783 out_root_realloc:
784 	xfs_iroot_realloc(ip, -1, whichfork);
785 	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
786 	ASSERT(ifp->if_broot == NULL);
787 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
788 
789 	return error;
790 }
791 
792 /*
793  * Convert a local file to an extents file.
794  * This code is out of bounds for data forks of regular files,
795  * since the file data needs to get logged so things will stay consistent.
796  * (The bmap-level manipulations are ok, though).
797  */
798 void
799 xfs_bmap_local_to_extents_empty(
800 	struct xfs_inode	*ip,
801 	int			whichfork)
802 {
803 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
804 
805 	ASSERT(whichfork != XFS_COW_FORK);
806 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
807 	ASSERT(ifp->if_bytes == 0);
808 	ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
809 
810 	xfs_bmap_forkoff_reset(ip, whichfork);
811 	ifp->if_flags &= ~XFS_IFINLINE;
812 	ifp->if_flags |= XFS_IFEXTENTS;
813 	ifp->if_u1.if_root = NULL;
814 	ifp->if_height = 0;
815 	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
816 }
817 
818 
819 STATIC int				/* error */
820 xfs_bmap_local_to_extents(
821 	xfs_trans_t	*tp,		/* transaction pointer */
822 	xfs_inode_t	*ip,		/* incore inode pointer */
823 	xfs_extlen_t	total,		/* total blocks needed by transaction */
824 	int		*logflagsp,	/* inode logging flags */
825 	int		whichfork,
826 	void		(*init_fn)(struct xfs_trans *tp,
827 				   struct xfs_buf *bp,
828 				   struct xfs_inode *ip,
829 				   struct xfs_ifork *ifp))
830 {
831 	int		error = 0;
832 	int		flags;		/* logging flags returned */
833 	struct xfs_ifork *ifp;		/* inode fork pointer */
834 	xfs_alloc_arg_t	args;		/* allocation arguments */
835 	xfs_buf_t	*bp;		/* buffer for extent block */
836 	struct xfs_bmbt_irec rec;
837 	struct xfs_iext_cursor icur;
838 
839 	/*
840 	 * We don't want to deal with the case of keeping inode data inline yet.
841 	 * So sending the data fork of a regular inode is invalid.
842 	 */
843 	ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
844 	ifp = XFS_IFORK_PTR(ip, whichfork);
845 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
846 
847 	if (!ifp->if_bytes) {
848 		xfs_bmap_local_to_extents_empty(ip, whichfork);
849 		flags = XFS_ILOG_CORE;
850 		goto done;
851 	}
852 
853 	flags = 0;
854 	error = 0;
855 	ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS)) == XFS_IFINLINE);
856 	memset(&args, 0, sizeof(args));
857 	args.tp = tp;
858 	args.mp = ip->i_mount;
859 	xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0);
860 	/*
861 	 * Allocate a block.  We know we need only one, since the
862 	 * file currently fits in an inode.
863 	 */
864 	if (tp->t_firstblock == NULLFSBLOCK) {
865 		args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
866 		args.type = XFS_ALLOCTYPE_START_BNO;
867 	} else {
868 		args.fsbno = tp->t_firstblock;
869 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
870 	}
871 	args.total = total;
872 	args.minlen = args.maxlen = args.prod = 1;
873 	error = xfs_alloc_vextent(&args);
874 	if (error)
875 		goto done;
876 
877 	/* Can't fail, the space was reserved. */
878 	ASSERT(args.fsbno != NULLFSBLOCK);
879 	ASSERT(args.len == 1);
880 	tp->t_firstblock = args.fsbno;
881 	bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
882 
883 	/*
884 	 * Initialize the block, copy the data and log the remote buffer.
885 	 *
886 	 * The callout is responsible for logging because the remote format
887 	 * might differ from the local format and thus we don't know how much to
888 	 * log here. Note that init_fn must also set the buffer log item type
889 	 * correctly.
890 	 */
891 	init_fn(tp, bp, ip, ifp);
892 
893 	/* account for the change in fork size */
894 	xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
895 	xfs_bmap_local_to_extents_empty(ip, whichfork);
896 	flags |= XFS_ILOG_CORE;
897 
898 	ifp->if_u1.if_root = NULL;
899 	ifp->if_height = 0;
900 
901 	rec.br_startoff = 0;
902 	rec.br_startblock = args.fsbno;
903 	rec.br_blockcount = 1;
904 	rec.br_state = XFS_EXT_NORM;
905 	xfs_iext_first(ifp, &icur);
906 	xfs_iext_insert(ip, &icur, &rec, 0);
907 
908 	XFS_IFORK_NEXT_SET(ip, whichfork, 1);
909 	ip->i_d.di_nblocks = 1;
910 	xfs_trans_mod_dquot_byino(tp, ip,
911 		XFS_TRANS_DQ_BCOUNT, 1L);
912 	flags |= xfs_ilog_fext(whichfork);
913 
914 done:
915 	*logflagsp = flags;
916 	return error;
917 }
918 
919 /*
920  * Called from xfs_bmap_add_attrfork to handle btree format files.
921  */
922 STATIC int					/* error */
923 xfs_bmap_add_attrfork_btree(
924 	xfs_trans_t		*tp,		/* transaction pointer */
925 	xfs_inode_t		*ip,		/* incore inode pointer */
926 	int			*flags)		/* inode logging flags */
927 {
928 	xfs_btree_cur_t		*cur;		/* btree cursor */
929 	int			error;		/* error return value */
930 	xfs_mount_t		*mp;		/* file system mount struct */
931 	int			stat;		/* newroot status */
932 
933 	mp = ip->i_mount;
934 	if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
935 		*flags |= XFS_ILOG_DBROOT;
936 	else {
937 		cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
938 		error = xfs_bmbt_lookup_first(cur, &stat);
939 		if (error)
940 			goto error0;
941 		/* must be at least one entry */
942 		XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0);
943 		if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
944 			goto error0;
945 		if (stat == 0) {
946 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
947 			return -ENOSPC;
948 		}
949 		cur->bc_private.b.allocated = 0;
950 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
951 	}
952 	return 0;
953 error0:
954 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
955 	return error;
956 }
957 
958 /*
959  * Called from xfs_bmap_add_attrfork to handle extents format files.
960  */
961 STATIC int					/* error */
962 xfs_bmap_add_attrfork_extents(
963 	struct xfs_trans	*tp,		/* transaction pointer */
964 	struct xfs_inode	*ip,		/* incore inode pointer */
965 	int			*flags)		/* inode logging flags */
966 {
967 	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
968 	int			error;		/* error return value */
969 
970 	if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
971 		return 0;
972 	cur = NULL;
973 	error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, flags,
974 					  XFS_DATA_FORK);
975 	if (cur) {
976 		cur->bc_private.b.allocated = 0;
977 		xfs_btree_del_cursor(cur, error);
978 	}
979 	return error;
980 }
981 
982 /*
983  * Called from xfs_bmap_add_attrfork to handle local format files. Each
984  * different data fork content type needs a different callout to do the
985  * conversion. Some are basic and only require special block initialisation
986  * callouts for the data formating, others (directories) are so specialised they
987  * handle everything themselves.
988  *
989  * XXX (dgc): investigate whether directory conversion can use the generic
990  * formatting callout. It should be possible - it's just a very complex
991  * formatter.
992  */
993 STATIC int					/* error */
994 xfs_bmap_add_attrfork_local(
995 	struct xfs_trans	*tp,		/* transaction pointer */
996 	struct xfs_inode	*ip,		/* incore inode pointer */
997 	int			*flags)		/* inode logging flags */
998 {
999 	struct xfs_da_args	dargs;		/* args for dir/attr code */
1000 
1001 	if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1002 		return 0;
1003 
1004 	if (S_ISDIR(VFS_I(ip)->i_mode)) {
1005 		memset(&dargs, 0, sizeof(dargs));
1006 		dargs.geo = ip->i_mount->m_dir_geo;
1007 		dargs.dp = ip;
1008 		dargs.total = dargs.geo->fsbcount;
1009 		dargs.whichfork = XFS_DATA_FORK;
1010 		dargs.trans = tp;
1011 		return xfs_dir2_sf_to_block(&dargs);
1012 	}
1013 
1014 	if (S_ISLNK(VFS_I(ip)->i_mode))
1015 		return xfs_bmap_local_to_extents(tp, ip, 1, flags,
1016 						 XFS_DATA_FORK,
1017 						 xfs_symlink_local_to_remote);
1018 
1019 	/* should only be called for types that support local format data */
1020 	ASSERT(0);
1021 	return -EFSCORRUPTED;
1022 }
1023 
1024 /* Set an inode attr fork off based on the format */
1025 int
1026 xfs_bmap_set_attrforkoff(
1027 	struct xfs_inode	*ip,
1028 	int			size,
1029 	int			*version)
1030 {
1031 	switch (ip->i_d.di_format) {
1032 	case XFS_DINODE_FMT_DEV:
1033 		ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1034 		break;
1035 	case XFS_DINODE_FMT_LOCAL:
1036 	case XFS_DINODE_FMT_EXTENTS:
1037 	case XFS_DINODE_FMT_BTREE:
1038 		ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1039 		if (!ip->i_d.di_forkoff)
1040 			ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1041 		else if ((ip->i_mount->m_flags & XFS_MOUNT_ATTR2) && version)
1042 			*version = 2;
1043 		break;
1044 	default:
1045 		ASSERT(0);
1046 		return -EINVAL;
1047 	}
1048 
1049 	return 0;
1050 }
1051 
1052 /*
1053  * Convert inode from non-attributed to attributed.
1054  * Must not be in a transaction, ip must not be locked.
1055  */
1056 int						/* error code */
1057 xfs_bmap_add_attrfork(
1058 	xfs_inode_t		*ip,		/* incore inode pointer */
1059 	int			size,		/* space new attribute needs */
1060 	int			rsvd)		/* xact may use reserved blks */
1061 {
1062 	xfs_mount_t		*mp;		/* mount structure */
1063 	xfs_trans_t		*tp;		/* transaction pointer */
1064 	int			blks;		/* space reservation */
1065 	int			version = 1;	/* superblock attr version */
1066 	int			logflags;	/* logging flags */
1067 	int			error;		/* error return value */
1068 
1069 	ASSERT(XFS_IFORK_Q(ip) == 0);
1070 
1071 	mp = ip->i_mount;
1072 	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1073 
1074 	blks = XFS_ADDAFORK_SPACE_RES(mp);
1075 
1076 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_addafork, blks, 0,
1077 			rsvd ? XFS_TRANS_RESERVE : 0, &tp);
1078 	if (error)
1079 		return error;
1080 
1081 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1082 	error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1083 			XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1084 			XFS_QMOPT_RES_REGBLKS);
1085 	if (error)
1086 		goto trans_cancel;
1087 	if (XFS_IFORK_Q(ip))
1088 		goto trans_cancel;
1089 	if (ip->i_d.di_anextents != 0) {
1090 		error = -EFSCORRUPTED;
1091 		goto trans_cancel;
1092 	}
1093 	if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1094 		/*
1095 		 * For inodes coming from pre-6.2 filesystems.
1096 		 */
1097 		ASSERT(ip->i_d.di_aformat == 0);
1098 		ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1099 	}
1100 
1101 	xfs_trans_ijoin(tp, ip, 0);
1102 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1103 	error = xfs_bmap_set_attrforkoff(ip, size, &version);
1104 	if (error)
1105 		goto trans_cancel;
1106 	ASSERT(ip->i_afp == NULL);
1107 	ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1108 	ip->i_afp->if_flags = XFS_IFEXTENTS;
1109 	logflags = 0;
1110 	switch (ip->i_d.di_format) {
1111 	case XFS_DINODE_FMT_LOCAL:
1112 		error = xfs_bmap_add_attrfork_local(tp, ip, &logflags);
1113 		break;
1114 	case XFS_DINODE_FMT_EXTENTS:
1115 		error = xfs_bmap_add_attrfork_extents(tp, ip, &logflags);
1116 		break;
1117 	case XFS_DINODE_FMT_BTREE:
1118 		error = xfs_bmap_add_attrfork_btree(tp, ip, &logflags);
1119 		break;
1120 	default:
1121 		error = 0;
1122 		break;
1123 	}
1124 	if (logflags)
1125 		xfs_trans_log_inode(tp, ip, logflags);
1126 	if (error)
1127 		goto trans_cancel;
1128 	if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1129 	   (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1130 		bool log_sb = false;
1131 
1132 		spin_lock(&mp->m_sb_lock);
1133 		if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1134 			xfs_sb_version_addattr(&mp->m_sb);
1135 			log_sb = true;
1136 		}
1137 		if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1138 			xfs_sb_version_addattr2(&mp->m_sb);
1139 			log_sb = true;
1140 		}
1141 		spin_unlock(&mp->m_sb_lock);
1142 		if (log_sb)
1143 			xfs_log_sb(tp);
1144 	}
1145 
1146 	error = xfs_trans_commit(tp);
1147 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1148 	return error;
1149 
1150 trans_cancel:
1151 	xfs_trans_cancel(tp);
1152 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1153 	return error;
1154 }
1155 
1156 /*
1157  * Internal and external extent tree search functions.
1158  */
1159 
1160 /*
1161  * Read in extents from a btree-format inode.
1162  */
1163 int
1164 xfs_iread_extents(
1165 	struct xfs_trans	*tp,
1166 	struct xfs_inode	*ip,
1167 	int			whichfork)
1168 {
1169 	struct xfs_mount	*mp = ip->i_mount;
1170 	int			state = xfs_bmap_fork_to_state(whichfork);
1171 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1172 	xfs_extnum_t		nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
1173 	struct xfs_btree_block	*block = ifp->if_broot;
1174 	struct xfs_iext_cursor	icur;
1175 	struct xfs_bmbt_irec	new;
1176 	xfs_fsblock_t		bno;
1177 	struct xfs_buf		*bp;
1178 	xfs_extnum_t		i, j;
1179 	int			level;
1180 	__be64			*pp;
1181 	int			error;
1182 
1183 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1184 
1185 	if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1186 		XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
1187 		return -EFSCORRUPTED;
1188 	}
1189 
1190 	/*
1191 	 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1192 	 */
1193 	level = be16_to_cpu(block->bb_level);
1194 	if (unlikely(level == 0)) {
1195 		XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
1196 		return -EFSCORRUPTED;
1197 	}
1198 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1199 	bno = be64_to_cpu(*pp);
1200 
1201 	/*
1202 	 * Go down the tree until leaf level is reached, following the first
1203 	 * pointer (leftmost) at each level.
1204 	 */
1205 	while (level-- > 0) {
1206 		error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1207 				XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1208 		if (error)
1209 			goto out;
1210 		block = XFS_BUF_TO_BLOCK(bp);
1211 		if (level == 0)
1212 			break;
1213 		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1214 		bno = be64_to_cpu(*pp);
1215 		XFS_WANT_CORRUPTED_GOTO(mp,
1216 			xfs_verify_fsbno(mp, bno), out_brelse);
1217 		xfs_trans_brelse(tp, bp);
1218 	}
1219 
1220 	/*
1221 	 * Here with bp and block set to the leftmost leaf node in the tree.
1222 	 */
1223 	i = 0;
1224 	xfs_iext_first(ifp, &icur);
1225 
1226 	/*
1227 	 * Loop over all leaf nodes.  Copy information to the extent records.
1228 	 */
1229 	for (;;) {
1230 		xfs_bmbt_rec_t	*frp;
1231 		xfs_fsblock_t	nextbno;
1232 		xfs_extnum_t	num_recs;
1233 
1234 		num_recs = xfs_btree_get_numrecs(block);
1235 		if (unlikely(i + num_recs > nextents)) {
1236 			xfs_warn(ip->i_mount,
1237 				"corrupt dinode %Lu, (btree extents).",
1238 				(unsigned long long) ip->i_ino);
1239 			xfs_inode_verifier_error(ip, -EFSCORRUPTED,
1240 					__func__, block, sizeof(*block),
1241 					__this_address);
1242 			error = -EFSCORRUPTED;
1243 			goto out_brelse;
1244 		}
1245 		/*
1246 		 * Read-ahead the next leaf block, if any.
1247 		 */
1248 		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1249 		if (nextbno != NULLFSBLOCK)
1250 			xfs_btree_reada_bufl(mp, nextbno, 1,
1251 					     &xfs_bmbt_buf_ops);
1252 		/*
1253 		 * Copy records into the extent records.
1254 		 */
1255 		frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1256 		for (j = 0; j < num_recs; j++, frp++, i++) {
1257 			xfs_failaddr_t	fa;
1258 
1259 			xfs_bmbt_disk_get_all(frp, &new);
1260 			fa = xfs_bmap_validate_extent(ip, whichfork, &new);
1261 			if (fa) {
1262 				error = -EFSCORRUPTED;
1263 				xfs_inode_verifier_error(ip, error,
1264 						"xfs_iread_extents(2)",
1265 						frp, sizeof(*frp), fa);
1266 				goto out_brelse;
1267 			}
1268 			xfs_iext_insert(ip, &icur, &new, state);
1269 			trace_xfs_read_extent(ip, &icur, state, _THIS_IP_);
1270 			xfs_iext_next(ifp, &icur);
1271 		}
1272 		xfs_trans_brelse(tp, bp);
1273 		bno = nextbno;
1274 		/*
1275 		 * If we've reached the end, stop.
1276 		 */
1277 		if (bno == NULLFSBLOCK)
1278 			break;
1279 		error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1280 				XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1281 		if (error)
1282 			goto out;
1283 		block = XFS_BUF_TO_BLOCK(bp);
1284 	}
1285 
1286 	if (i != XFS_IFORK_NEXTENTS(ip, whichfork)) {
1287 		error = -EFSCORRUPTED;
1288 		goto out;
1289 	}
1290 	ASSERT(i == xfs_iext_count(ifp));
1291 
1292 	ifp->if_flags |= XFS_IFEXTENTS;
1293 	return 0;
1294 
1295 out_brelse:
1296 	xfs_trans_brelse(tp, bp);
1297 out:
1298 	xfs_iext_destroy(ifp);
1299 	return error;
1300 }
1301 
1302 /*
1303  * Returns the relative block number of the first unused block(s) in the given
1304  * fork with at least "len" logically contiguous blocks free.  This is the
1305  * lowest-address hole if the fork has holes, else the first block past the end
1306  * of fork.  Return 0 if the fork is currently local (in-inode).
1307  */
1308 int						/* error */
1309 xfs_bmap_first_unused(
1310 	struct xfs_trans	*tp,		/* transaction pointer */
1311 	struct xfs_inode	*ip,		/* incore inode */
1312 	xfs_extlen_t		len,		/* size of hole to find */
1313 	xfs_fileoff_t		*first_unused,	/* unused block */
1314 	int			whichfork)	/* data or attr fork */
1315 {
1316 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1317 	struct xfs_bmbt_irec	got;
1318 	struct xfs_iext_cursor	icur;
1319 	xfs_fileoff_t		lastaddr = 0;
1320 	xfs_fileoff_t		lowest, max;
1321 	int			error;
1322 
1323 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1324 	       XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1325 	       XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1326 
1327 	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1328 		*first_unused = 0;
1329 		return 0;
1330 	}
1331 
1332 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1333 		error = xfs_iread_extents(tp, ip, whichfork);
1334 		if (error)
1335 			return error;
1336 	}
1337 
1338 	lowest = max = *first_unused;
1339 	for_each_xfs_iext(ifp, &icur, &got) {
1340 		/*
1341 		 * See if the hole before this extent will work.
1342 		 */
1343 		if (got.br_startoff >= lowest + len &&
1344 		    got.br_startoff - max >= len)
1345 			break;
1346 		lastaddr = got.br_startoff + got.br_blockcount;
1347 		max = XFS_FILEOFF_MAX(lastaddr, lowest);
1348 	}
1349 
1350 	*first_unused = max;
1351 	return 0;
1352 }
1353 
1354 /*
1355  * Returns the file-relative block number of the last block - 1 before
1356  * last_block (input value) in the file.
1357  * This is not based on i_size, it is based on the extent records.
1358  * Returns 0 for local files, as they do not have extent records.
1359  */
1360 int						/* error */
1361 xfs_bmap_last_before(
1362 	struct xfs_trans	*tp,		/* transaction pointer */
1363 	struct xfs_inode	*ip,		/* incore inode */
1364 	xfs_fileoff_t		*last_block,	/* last block */
1365 	int			whichfork)	/* data or attr fork */
1366 {
1367 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1368 	struct xfs_bmbt_irec	got;
1369 	struct xfs_iext_cursor	icur;
1370 	int			error;
1371 
1372 	switch (XFS_IFORK_FORMAT(ip, whichfork)) {
1373 	case XFS_DINODE_FMT_LOCAL:
1374 		*last_block = 0;
1375 		return 0;
1376 	case XFS_DINODE_FMT_BTREE:
1377 	case XFS_DINODE_FMT_EXTENTS:
1378 		break;
1379 	default:
1380 		return -EIO;
1381 	}
1382 
1383 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1384 		error = xfs_iread_extents(tp, ip, whichfork);
1385 		if (error)
1386 			return error;
1387 	}
1388 
1389 	if (!xfs_iext_lookup_extent_before(ip, ifp, last_block, &icur, &got))
1390 		*last_block = 0;
1391 	return 0;
1392 }
1393 
1394 int
1395 xfs_bmap_last_extent(
1396 	struct xfs_trans	*tp,
1397 	struct xfs_inode	*ip,
1398 	int			whichfork,
1399 	struct xfs_bmbt_irec	*rec,
1400 	int			*is_empty)
1401 {
1402 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1403 	struct xfs_iext_cursor	icur;
1404 	int			error;
1405 
1406 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1407 		error = xfs_iread_extents(tp, ip, whichfork);
1408 		if (error)
1409 			return error;
1410 	}
1411 
1412 	xfs_iext_last(ifp, &icur);
1413 	if (!xfs_iext_get_extent(ifp, &icur, rec))
1414 		*is_empty = 1;
1415 	else
1416 		*is_empty = 0;
1417 	return 0;
1418 }
1419 
1420 /*
1421  * Check the last inode extent to determine whether this allocation will result
1422  * in blocks being allocated at the end of the file. When we allocate new data
1423  * blocks at the end of the file which do not start at the previous data block,
1424  * we will try to align the new blocks at stripe unit boundaries.
1425  *
1426  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1427  * at, or past the EOF.
1428  */
1429 STATIC int
1430 xfs_bmap_isaeof(
1431 	struct xfs_bmalloca	*bma,
1432 	int			whichfork)
1433 {
1434 	struct xfs_bmbt_irec	rec;
1435 	int			is_empty;
1436 	int			error;
1437 
1438 	bma->aeof = false;
1439 	error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1440 				     &is_empty);
1441 	if (error)
1442 		return error;
1443 
1444 	if (is_empty) {
1445 		bma->aeof = true;
1446 		return 0;
1447 	}
1448 
1449 	/*
1450 	 * Check if we are allocation or past the last extent, or at least into
1451 	 * the last delayed allocated extent.
1452 	 */
1453 	bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1454 		(bma->offset >= rec.br_startoff &&
1455 		 isnullstartblock(rec.br_startblock));
1456 	return 0;
1457 }
1458 
1459 /*
1460  * Returns the file-relative block number of the first block past eof in
1461  * the file.  This is not based on i_size, it is based on the extent records.
1462  * Returns 0 for local files, as they do not have extent records.
1463  */
1464 int
1465 xfs_bmap_last_offset(
1466 	struct xfs_inode	*ip,
1467 	xfs_fileoff_t		*last_block,
1468 	int			whichfork)
1469 {
1470 	struct xfs_bmbt_irec	rec;
1471 	int			is_empty;
1472 	int			error;
1473 
1474 	*last_block = 0;
1475 
1476 	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1477 		return 0;
1478 
1479 	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1480 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1481 	       return -EIO;
1482 
1483 	error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1484 	if (error || is_empty)
1485 		return error;
1486 
1487 	*last_block = rec.br_startoff + rec.br_blockcount;
1488 	return 0;
1489 }
1490 
1491 /*
1492  * Returns whether the selected fork of the inode has exactly one
1493  * block or not.  For the data fork we check this matches di_size,
1494  * implying the file's range is 0..bsize-1.
1495  */
1496 int					/* 1=>1 block, 0=>otherwise */
1497 xfs_bmap_one_block(
1498 	xfs_inode_t	*ip,		/* incore inode */
1499 	int		whichfork)	/* data or attr fork */
1500 {
1501 	struct xfs_ifork *ifp;		/* inode fork pointer */
1502 	int		rval;		/* return value */
1503 	xfs_bmbt_irec_t	s;		/* internal version of extent */
1504 	struct xfs_iext_cursor icur;
1505 
1506 #ifndef DEBUG
1507 	if (whichfork == XFS_DATA_FORK)
1508 		return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1509 #endif	/* !DEBUG */
1510 	if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1511 		return 0;
1512 	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1513 		return 0;
1514 	ifp = XFS_IFORK_PTR(ip, whichfork);
1515 	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1516 	xfs_iext_first(ifp, &icur);
1517 	xfs_iext_get_extent(ifp, &icur, &s);
1518 	rval = s.br_startoff == 0 && s.br_blockcount == 1;
1519 	if (rval && whichfork == XFS_DATA_FORK)
1520 		ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1521 	return rval;
1522 }
1523 
1524 /*
1525  * Extent tree manipulation functions used during allocation.
1526  */
1527 
1528 /*
1529  * Convert a delayed allocation to a real allocation.
1530  */
1531 STATIC int				/* error */
1532 xfs_bmap_add_extent_delay_real(
1533 	struct xfs_bmalloca	*bma,
1534 	int			whichfork)
1535 {
1536 	struct xfs_bmbt_irec	*new = &bma->got;
1537 	int			error;	/* error return value */
1538 	int			i;	/* temp state */
1539 	struct xfs_ifork	*ifp;	/* inode fork pointer */
1540 	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
1541 	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
1542 					/* left is 0, right is 1, prev is 2 */
1543 	int			rval=0;	/* return value (logging flags) */
1544 	int			state = xfs_bmap_fork_to_state(whichfork);
1545 	xfs_filblks_t		da_new; /* new count del alloc blocks used */
1546 	xfs_filblks_t		da_old; /* old count del alloc blocks used */
1547 	xfs_filblks_t		temp=0;	/* value for da_new calculations */
1548 	int			tmp_rval;	/* partial logging flags */
1549 	struct xfs_mount	*mp;
1550 	xfs_extnum_t		*nextents;
1551 	struct xfs_bmbt_irec	old;
1552 
1553 	mp = bma->ip->i_mount;
1554 	ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1555 	ASSERT(whichfork != XFS_ATTR_FORK);
1556 	nextents = (whichfork == XFS_COW_FORK ? &bma->ip->i_cnextents :
1557 						&bma->ip->i_d.di_nextents);
1558 
1559 	ASSERT(!isnullstartblock(new->br_startblock));
1560 	ASSERT(!bma->cur ||
1561 	       (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1562 
1563 	XFS_STATS_INC(mp, xs_add_exlist);
1564 
1565 #define	LEFT		r[0]
1566 #define	RIGHT		r[1]
1567 #define	PREV		r[2]
1568 
1569 	/*
1570 	 * Set up a bunch of variables to make the tests simpler.
1571 	 */
1572 	xfs_iext_get_extent(ifp, &bma->icur, &PREV);
1573 	new_endoff = new->br_startoff + new->br_blockcount;
1574 	ASSERT(isnullstartblock(PREV.br_startblock));
1575 	ASSERT(PREV.br_startoff <= new->br_startoff);
1576 	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1577 
1578 	da_old = startblockval(PREV.br_startblock);
1579 	da_new = 0;
1580 
1581 	/*
1582 	 * Set flags determining what part of the previous delayed allocation
1583 	 * extent is being replaced by a real allocation.
1584 	 */
1585 	if (PREV.br_startoff == new->br_startoff)
1586 		state |= BMAP_LEFT_FILLING;
1587 	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1588 		state |= BMAP_RIGHT_FILLING;
1589 
1590 	/*
1591 	 * Check and set flags if this segment has a left neighbor.
1592 	 * Don't set contiguous if the combined extent would be too large.
1593 	 */
1594 	if (xfs_iext_peek_prev_extent(ifp, &bma->icur, &LEFT)) {
1595 		state |= BMAP_LEFT_VALID;
1596 		if (isnullstartblock(LEFT.br_startblock))
1597 			state |= BMAP_LEFT_DELAY;
1598 	}
1599 
1600 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1601 	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1602 	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1603 	    LEFT.br_state == new->br_state &&
1604 	    LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1605 		state |= BMAP_LEFT_CONTIG;
1606 
1607 	/*
1608 	 * Check and set flags if this segment has a right neighbor.
1609 	 * Don't set contiguous if the combined extent would be too large.
1610 	 * Also check for all-three-contiguous being too large.
1611 	 */
1612 	if (xfs_iext_peek_next_extent(ifp, &bma->icur, &RIGHT)) {
1613 		state |= BMAP_RIGHT_VALID;
1614 		if (isnullstartblock(RIGHT.br_startblock))
1615 			state |= BMAP_RIGHT_DELAY;
1616 	}
1617 
1618 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1619 	    new_endoff == RIGHT.br_startoff &&
1620 	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1621 	    new->br_state == RIGHT.br_state &&
1622 	    new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1623 	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1624 		       BMAP_RIGHT_FILLING)) !=
1625 		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1626 		       BMAP_RIGHT_FILLING) ||
1627 	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1628 			<= MAXEXTLEN))
1629 		state |= BMAP_RIGHT_CONTIG;
1630 
1631 	error = 0;
1632 	/*
1633 	 * Switch out based on the FILLING and CONTIG state bits.
1634 	 */
1635 	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1636 			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1637 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1638 	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1639 		/*
1640 		 * Filling in all of a previously delayed allocation extent.
1641 		 * The left and right neighbors are both contiguous with new.
1642 		 */
1643 		LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
1644 
1645 		xfs_iext_remove(bma->ip, &bma->icur, state);
1646 		xfs_iext_remove(bma->ip, &bma->icur, state);
1647 		xfs_iext_prev(ifp, &bma->icur);
1648 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1649 		(*nextents)--;
1650 
1651 		if (bma->cur == NULL)
1652 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1653 		else {
1654 			rval = XFS_ILOG_CORE;
1655 			error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1656 			if (error)
1657 				goto done;
1658 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1659 			error = xfs_btree_delete(bma->cur, &i);
1660 			if (error)
1661 				goto done;
1662 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1663 			error = xfs_btree_decrement(bma->cur, 0, &i);
1664 			if (error)
1665 				goto done;
1666 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1667 			error = xfs_bmbt_update(bma->cur, &LEFT);
1668 			if (error)
1669 				goto done;
1670 		}
1671 		break;
1672 
1673 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1674 		/*
1675 		 * Filling in all of a previously delayed allocation extent.
1676 		 * The left neighbor is contiguous, the right is not.
1677 		 */
1678 		old = LEFT;
1679 		LEFT.br_blockcount += PREV.br_blockcount;
1680 
1681 		xfs_iext_remove(bma->ip, &bma->icur, state);
1682 		xfs_iext_prev(ifp, &bma->icur);
1683 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1684 
1685 		if (bma->cur == NULL)
1686 			rval = XFS_ILOG_DEXT;
1687 		else {
1688 			rval = 0;
1689 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1690 			if (error)
1691 				goto done;
1692 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1693 			error = xfs_bmbt_update(bma->cur, &LEFT);
1694 			if (error)
1695 				goto done;
1696 		}
1697 		break;
1698 
1699 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1700 		/*
1701 		 * Filling in all of a previously delayed allocation extent.
1702 		 * The right neighbor is contiguous, the left is not. Take care
1703 		 * with delay -> unwritten extent allocation here because the
1704 		 * delalloc record we are overwriting is always written.
1705 		 */
1706 		PREV.br_startblock = new->br_startblock;
1707 		PREV.br_blockcount += RIGHT.br_blockcount;
1708 		PREV.br_state = new->br_state;
1709 
1710 		xfs_iext_next(ifp, &bma->icur);
1711 		xfs_iext_remove(bma->ip, &bma->icur, state);
1712 		xfs_iext_prev(ifp, &bma->icur);
1713 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1714 
1715 		if (bma->cur == NULL)
1716 			rval = XFS_ILOG_DEXT;
1717 		else {
1718 			rval = 0;
1719 			error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1720 			if (error)
1721 				goto done;
1722 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1723 			error = xfs_bmbt_update(bma->cur, &PREV);
1724 			if (error)
1725 				goto done;
1726 		}
1727 		break;
1728 
1729 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1730 		/*
1731 		 * Filling in all of a previously delayed allocation extent.
1732 		 * Neither the left nor right neighbors are contiguous with
1733 		 * the new one.
1734 		 */
1735 		PREV.br_startblock = new->br_startblock;
1736 		PREV.br_state = new->br_state;
1737 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1738 
1739 		(*nextents)++;
1740 		if (bma->cur == NULL)
1741 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1742 		else {
1743 			rval = XFS_ILOG_CORE;
1744 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1745 			if (error)
1746 				goto done;
1747 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1748 			error = xfs_btree_insert(bma->cur, &i);
1749 			if (error)
1750 				goto done;
1751 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1752 		}
1753 		break;
1754 
1755 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1756 		/*
1757 		 * Filling in the first part of a previous delayed allocation.
1758 		 * The left neighbor is contiguous.
1759 		 */
1760 		old = LEFT;
1761 		temp = PREV.br_blockcount - new->br_blockcount;
1762 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1763 				startblockval(PREV.br_startblock));
1764 
1765 		LEFT.br_blockcount += new->br_blockcount;
1766 
1767 		PREV.br_blockcount = temp;
1768 		PREV.br_startoff += new->br_blockcount;
1769 		PREV.br_startblock = nullstartblock(da_new);
1770 
1771 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1772 		xfs_iext_prev(ifp, &bma->icur);
1773 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1774 
1775 		if (bma->cur == NULL)
1776 			rval = XFS_ILOG_DEXT;
1777 		else {
1778 			rval = 0;
1779 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1780 			if (error)
1781 				goto done;
1782 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1783 			error = xfs_bmbt_update(bma->cur, &LEFT);
1784 			if (error)
1785 				goto done;
1786 		}
1787 		break;
1788 
1789 	case BMAP_LEFT_FILLING:
1790 		/*
1791 		 * Filling in the first part of a previous delayed allocation.
1792 		 * The left neighbor is not contiguous.
1793 		 */
1794 		xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1795 		(*nextents)++;
1796 		if (bma->cur == NULL)
1797 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1798 		else {
1799 			rval = XFS_ILOG_CORE;
1800 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1801 			if (error)
1802 				goto done;
1803 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1804 			error = xfs_btree_insert(bma->cur, &i);
1805 			if (error)
1806 				goto done;
1807 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1808 		}
1809 
1810 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1811 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1812 					&bma->cur, 1, &tmp_rval, whichfork);
1813 			rval |= tmp_rval;
1814 			if (error)
1815 				goto done;
1816 		}
1817 
1818 		temp = PREV.br_blockcount - new->br_blockcount;
1819 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1820 			startblockval(PREV.br_startblock) -
1821 			(bma->cur ? bma->cur->bc_private.b.allocated : 0));
1822 
1823 		PREV.br_startoff = new_endoff;
1824 		PREV.br_blockcount = temp;
1825 		PREV.br_startblock = nullstartblock(da_new);
1826 		xfs_iext_next(ifp, &bma->icur);
1827 		xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1828 		xfs_iext_prev(ifp, &bma->icur);
1829 		break;
1830 
1831 	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1832 		/*
1833 		 * Filling in the last part of a previous delayed allocation.
1834 		 * The right neighbor is contiguous with the new allocation.
1835 		 */
1836 		old = RIGHT;
1837 		RIGHT.br_startoff = new->br_startoff;
1838 		RIGHT.br_startblock = new->br_startblock;
1839 		RIGHT.br_blockcount += new->br_blockcount;
1840 
1841 		if (bma->cur == NULL)
1842 			rval = XFS_ILOG_DEXT;
1843 		else {
1844 			rval = 0;
1845 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1846 			if (error)
1847 				goto done;
1848 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1849 			error = xfs_bmbt_update(bma->cur, &RIGHT);
1850 			if (error)
1851 				goto done;
1852 		}
1853 
1854 		temp = PREV.br_blockcount - new->br_blockcount;
1855 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1856 			startblockval(PREV.br_startblock));
1857 
1858 		PREV.br_blockcount = temp;
1859 		PREV.br_startblock = nullstartblock(da_new);
1860 
1861 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1862 		xfs_iext_next(ifp, &bma->icur);
1863 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &RIGHT);
1864 		break;
1865 
1866 	case BMAP_RIGHT_FILLING:
1867 		/*
1868 		 * Filling in the last part of a previous delayed allocation.
1869 		 * The right neighbor is not contiguous.
1870 		 */
1871 		xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1872 		(*nextents)++;
1873 		if (bma->cur == NULL)
1874 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1875 		else {
1876 			rval = XFS_ILOG_CORE;
1877 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1878 			if (error)
1879 				goto done;
1880 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1881 			error = xfs_btree_insert(bma->cur, &i);
1882 			if (error)
1883 				goto done;
1884 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1885 		}
1886 
1887 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1888 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1889 				&bma->cur, 1, &tmp_rval, whichfork);
1890 			rval |= tmp_rval;
1891 			if (error)
1892 				goto done;
1893 		}
1894 
1895 		temp = PREV.br_blockcount - new->br_blockcount;
1896 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1897 			startblockval(PREV.br_startblock) -
1898 			(bma->cur ? bma->cur->bc_private.b.allocated : 0));
1899 
1900 		PREV.br_startblock = nullstartblock(da_new);
1901 		PREV.br_blockcount = temp;
1902 		xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1903 		xfs_iext_next(ifp, &bma->icur);
1904 		break;
1905 
1906 	case 0:
1907 		/*
1908 		 * Filling in the middle part of a previous delayed allocation.
1909 		 * Contiguity is impossible here.
1910 		 * This case is avoided almost all the time.
1911 		 *
1912 		 * We start with a delayed allocation:
1913 		 *
1914 		 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1915 		 *  PREV @ idx
1916 		 *
1917 	         * and we are allocating:
1918 		 *                     +rrrrrrrrrrrrrrrrr+
1919 		 *			      new
1920 		 *
1921 		 * and we set it up for insertion as:
1922 		 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1923 		 *                            new
1924 		 *  PREV @ idx          LEFT              RIGHT
1925 		 *                      inserted at idx + 1
1926 		 */
1927 		old = PREV;
1928 
1929 		/* LEFT is the new middle */
1930 		LEFT = *new;
1931 
1932 		/* RIGHT is the new right */
1933 		RIGHT.br_state = PREV.br_state;
1934 		RIGHT.br_startoff = new_endoff;
1935 		RIGHT.br_blockcount =
1936 			PREV.br_startoff + PREV.br_blockcount - new_endoff;
1937 		RIGHT.br_startblock =
1938 			nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1939 					RIGHT.br_blockcount));
1940 
1941 		/* truncate PREV */
1942 		PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
1943 		PREV.br_startblock =
1944 			nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1945 					PREV.br_blockcount));
1946 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1947 
1948 		xfs_iext_next(ifp, &bma->icur);
1949 		xfs_iext_insert(bma->ip, &bma->icur, &RIGHT, state);
1950 		xfs_iext_insert(bma->ip, &bma->icur, &LEFT, state);
1951 		(*nextents)++;
1952 
1953 		if (bma->cur == NULL)
1954 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1955 		else {
1956 			rval = XFS_ILOG_CORE;
1957 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1958 			if (error)
1959 				goto done;
1960 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1961 			error = xfs_btree_insert(bma->cur, &i);
1962 			if (error)
1963 				goto done;
1964 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1965 		}
1966 
1967 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1968 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1969 					&bma->cur, 1, &tmp_rval, whichfork);
1970 			rval |= tmp_rval;
1971 			if (error)
1972 				goto done;
1973 		}
1974 
1975 		da_new = startblockval(PREV.br_startblock) +
1976 			 startblockval(RIGHT.br_startblock);
1977 		break;
1978 
1979 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1980 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1981 	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1982 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1983 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1984 	case BMAP_LEFT_CONTIG:
1985 	case BMAP_RIGHT_CONTIG:
1986 		/*
1987 		 * These cases are all impossible.
1988 		 */
1989 		ASSERT(0);
1990 	}
1991 
1992 	/* add reverse mapping unless caller opted out */
1993 	if (!(bma->flags & XFS_BMAPI_NORMAP)) {
1994 		error = xfs_rmap_map_extent(bma->tp, bma->ip, whichfork, new);
1995 		if (error)
1996 			goto done;
1997 	}
1998 
1999 	/* convert to a btree if necessary */
2000 	if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2001 		int	tmp_logflags;	/* partial log flag return val */
2002 
2003 		ASSERT(bma->cur == NULL);
2004 		error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2005 				&bma->cur, da_old > 0, &tmp_logflags,
2006 				whichfork);
2007 		bma->logflags |= tmp_logflags;
2008 		if (error)
2009 			goto done;
2010 	}
2011 
2012 	if (da_new != da_old)
2013 		xfs_mod_delalloc(mp, (int64_t)da_new - da_old);
2014 
2015 	if (bma->cur) {
2016 		da_new += bma->cur->bc_private.b.allocated;
2017 		bma->cur->bc_private.b.allocated = 0;
2018 	}
2019 
2020 	/* adjust for changes in reserved delayed indirect blocks */
2021 	if (da_new != da_old) {
2022 		ASSERT(state == 0 || da_new < da_old);
2023 		error = xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new),
2024 				false);
2025 	}
2026 
2027 	xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
2028 done:
2029 	if (whichfork != XFS_COW_FORK)
2030 		bma->logflags |= rval;
2031 	return error;
2032 #undef	LEFT
2033 #undef	RIGHT
2034 #undef	PREV
2035 }
2036 
2037 /*
2038  * Convert an unwritten allocation to a real allocation or vice versa.
2039  */
2040 int					/* error */
2041 xfs_bmap_add_extent_unwritten_real(
2042 	struct xfs_trans	*tp,
2043 	xfs_inode_t		*ip,	/* incore inode pointer */
2044 	int			whichfork,
2045 	struct xfs_iext_cursor	*icur,
2046 	xfs_btree_cur_t		**curp,	/* if *curp is null, not a btree */
2047 	xfs_bmbt_irec_t		*new,	/* new data to add to file extents */
2048 	int			*logflagsp) /* inode logging flags */
2049 {
2050 	xfs_btree_cur_t		*cur;	/* btree cursor */
2051 	int			error;	/* error return value */
2052 	int			i;	/* temp state */
2053 	struct xfs_ifork	*ifp;	/* inode fork pointer */
2054 	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
2055 	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
2056 					/* left is 0, right is 1, prev is 2 */
2057 	int			rval=0;	/* return value (logging flags) */
2058 	int			state = xfs_bmap_fork_to_state(whichfork);
2059 	struct xfs_mount	*mp = ip->i_mount;
2060 	struct xfs_bmbt_irec	old;
2061 
2062 	*logflagsp = 0;
2063 
2064 	cur = *curp;
2065 	ifp = XFS_IFORK_PTR(ip, whichfork);
2066 
2067 	ASSERT(!isnullstartblock(new->br_startblock));
2068 
2069 	XFS_STATS_INC(mp, xs_add_exlist);
2070 
2071 #define	LEFT		r[0]
2072 #define	RIGHT		r[1]
2073 #define	PREV		r[2]
2074 
2075 	/*
2076 	 * Set up a bunch of variables to make the tests simpler.
2077 	 */
2078 	error = 0;
2079 	xfs_iext_get_extent(ifp, icur, &PREV);
2080 	ASSERT(new->br_state != PREV.br_state);
2081 	new_endoff = new->br_startoff + new->br_blockcount;
2082 	ASSERT(PREV.br_startoff <= new->br_startoff);
2083 	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2084 
2085 	/*
2086 	 * Set flags determining what part of the previous oldext allocation
2087 	 * extent is being replaced by a newext allocation.
2088 	 */
2089 	if (PREV.br_startoff == new->br_startoff)
2090 		state |= BMAP_LEFT_FILLING;
2091 	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2092 		state |= BMAP_RIGHT_FILLING;
2093 
2094 	/*
2095 	 * Check and set flags if this segment has a left neighbor.
2096 	 * Don't set contiguous if the combined extent would be too large.
2097 	 */
2098 	if (xfs_iext_peek_prev_extent(ifp, icur, &LEFT)) {
2099 		state |= BMAP_LEFT_VALID;
2100 		if (isnullstartblock(LEFT.br_startblock))
2101 			state |= BMAP_LEFT_DELAY;
2102 	}
2103 
2104 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2105 	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2106 	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2107 	    LEFT.br_state == new->br_state &&
2108 	    LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2109 		state |= BMAP_LEFT_CONTIG;
2110 
2111 	/*
2112 	 * Check and set flags if this segment has a right neighbor.
2113 	 * Don't set contiguous if the combined extent would be too large.
2114 	 * Also check for all-three-contiguous being too large.
2115 	 */
2116 	if (xfs_iext_peek_next_extent(ifp, icur, &RIGHT)) {
2117 		state |= BMAP_RIGHT_VALID;
2118 		if (isnullstartblock(RIGHT.br_startblock))
2119 			state |= BMAP_RIGHT_DELAY;
2120 	}
2121 
2122 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2123 	    new_endoff == RIGHT.br_startoff &&
2124 	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2125 	    new->br_state == RIGHT.br_state &&
2126 	    new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2127 	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2128 		       BMAP_RIGHT_FILLING)) !=
2129 		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2130 		       BMAP_RIGHT_FILLING) ||
2131 	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2132 			<= MAXEXTLEN))
2133 		state |= BMAP_RIGHT_CONTIG;
2134 
2135 	/*
2136 	 * Switch out based on the FILLING and CONTIG state bits.
2137 	 */
2138 	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2139 			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2140 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2141 	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2142 		/*
2143 		 * Setting all of a previous oldext extent to newext.
2144 		 * The left and right neighbors are both contiguous with new.
2145 		 */
2146 		LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
2147 
2148 		xfs_iext_remove(ip, icur, state);
2149 		xfs_iext_remove(ip, icur, state);
2150 		xfs_iext_prev(ifp, icur);
2151 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2152 		XFS_IFORK_NEXT_SET(ip, whichfork,
2153 				XFS_IFORK_NEXTENTS(ip, whichfork) - 2);
2154 		if (cur == NULL)
2155 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2156 		else {
2157 			rval = XFS_ILOG_CORE;
2158 			error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2159 			if (error)
2160 				goto done;
2161 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2162 			if ((error = xfs_btree_delete(cur, &i)))
2163 				goto done;
2164 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2165 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2166 				goto done;
2167 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2168 			if ((error = xfs_btree_delete(cur, &i)))
2169 				goto done;
2170 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2171 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2172 				goto done;
2173 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2174 			error = xfs_bmbt_update(cur, &LEFT);
2175 			if (error)
2176 				goto done;
2177 		}
2178 		break;
2179 
2180 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2181 		/*
2182 		 * Setting all of a previous oldext extent to newext.
2183 		 * The left neighbor is contiguous, the right is not.
2184 		 */
2185 		LEFT.br_blockcount += PREV.br_blockcount;
2186 
2187 		xfs_iext_remove(ip, icur, state);
2188 		xfs_iext_prev(ifp, icur);
2189 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2190 		XFS_IFORK_NEXT_SET(ip, whichfork,
2191 				XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2192 		if (cur == NULL)
2193 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2194 		else {
2195 			rval = XFS_ILOG_CORE;
2196 			error = xfs_bmbt_lookup_eq(cur, &PREV, &i);
2197 			if (error)
2198 				goto done;
2199 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2200 			if ((error = xfs_btree_delete(cur, &i)))
2201 				goto done;
2202 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2203 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2204 				goto done;
2205 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2206 			error = xfs_bmbt_update(cur, &LEFT);
2207 			if (error)
2208 				goto done;
2209 		}
2210 		break;
2211 
2212 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2213 		/*
2214 		 * Setting all of a previous oldext extent to newext.
2215 		 * The right neighbor is contiguous, the left is not.
2216 		 */
2217 		PREV.br_blockcount += RIGHT.br_blockcount;
2218 		PREV.br_state = new->br_state;
2219 
2220 		xfs_iext_next(ifp, icur);
2221 		xfs_iext_remove(ip, icur, state);
2222 		xfs_iext_prev(ifp, icur);
2223 		xfs_iext_update_extent(ip, state, icur, &PREV);
2224 
2225 		XFS_IFORK_NEXT_SET(ip, whichfork,
2226 				XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2227 		if (cur == NULL)
2228 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2229 		else {
2230 			rval = XFS_ILOG_CORE;
2231 			error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2232 			if (error)
2233 				goto done;
2234 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2235 			if ((error = xfs_btree_delete(cur, &i)))
2236 				goto done;
2237 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2238 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2239 				goto done;
2240 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2241 			error = xfs_bmbt_update(cur, &PREV);
2242 			if (error)
2243 				goto done;
2244 		}
2245 		break;
2246 
2247 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2248 		/*
2249 		 * Setting all of a previous oldext extent to newext.
2250 		 * Neither the left nor right neighbors are contiguous with
2251 		 * the new one.
2252 		 */
2253 		PREV.br_state = new->br_state;
2254 		xfs_iext_update_extent(ip, state, icur, &PREV);
2255 
2256 		if (cur == NULL)
2257 			rval = XFS_ILOG_DEXT;
2258 		else {
2259 			rval = 0;
2260 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2261 			if (error)
2262 				goto done;
2263 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2264 			error = xfs_bmbt_update(cur, &PREV);
2265 			if (error)
2266 				goto done;
2267 		}
2268 		break;
2269 
2270 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2271 		/*
2272 		 * Setting the first part of a previous oldext extent to newext.
2273 		 * The left neighbor is contiguous.
2274 		 */
2275 		LEFT.br_blockcount += new->br_blockcount;
2276 
2277 		old = PREV;
2278 		PREV.br_startoff += new->br_blockcount;
2279 		PREV.br_startblock += new->br_blockcount;
2280 		PREV.br_blockcount -= new->br_blockcount;
2281 
2282 		xfs_iext_update_extent(ip, state, icur, &PREV);
2283 		xfs_iext_prev(ifp, icur);
2284 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2285 
2286 		if (cur == NULL)
2287 			rval = XFS_ILOG_DEXT;
2288 		else {
2289 			rval = 0;
2290 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2291 			if (error)
2292 				goto done;
2293 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2294 			error = xfs_bmbt_update(cur, &PREV);
2295 			if (error)
2296 				goto done;
2297 			error = xfs_btree_decrement(cur, 0, &i);
2298 			if (error)
2299 				goto done;
2300 			error = xfs_bmbt_update(cur, &LEFT);
2301 			if (error)
2302 				goto done;
2303 		}
2304 		break;
2305 
2306 	case BMAP_LEFT_FILLING:
2307 		/*
2308 		 * Setting the first part of a previous oldext extent to newext.
2309 		 * The left neighbor is not contiguous.
2310 		 */
2311 		old = PREV;
2312 		PREV.br_startoff += new->br_blockcount;
2313 		PREV.br_startblock += new->br_blockcount;
2314 		PREV.br_blockcount -= new->br_blockcount;
2315 
2316 		xfs_iext_update_extent(ip, state, icur, &PREV);
2317 		xfs_iext_insert(ip, icur, new, state);
2318 		XFS_IFORK_NEXT_SET(ip, whichfork,
2319 				XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2320 		if (cur == NULL)
2321 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2322 		else {
2323 			rval = XFS_ILOG_CORE;
2324 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2325 			if (error)
2326 				goto done;
2327 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2328 			error = xfs_bmbt_update(cur, &PREV);
2329 			if (error)
2330 				goto done;
2331 			cur->bc_rec.b = *new;
2332 			if ((error = xfs_btree_insert(cur, &i)))
2333 				goto done;
2334 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2335 		}
2336 		break;
2337 
2338 	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2339 		/*
2340 		 * Setting the last part of a previous oldext extent to newext.
2341 		 * The right neighbor is contiguous with the new allocation.
2342 		 */
2343 		old = PREV;
2344 		PREV.br_blockcount -= new->br_blockcount;
2345 
2346 		RIGHT.br_startoff = new->br_startoff;
2347 		RIGHT.br_startblock = new->br_startblock;
2348 		RIGHT.br_blockcount += new->br_blockcount;
2349 
2350 		xfs_iext_update_extent(ip, state, icur, &PREV);
2351 		xfs_iext_next(ifp, icur);
2352 		xfs_iext_update_extent(ip, state, icur, &RIGHT);
2353 
2354 		if (cur == NULL)
2355 			rval = XFS_ILOG_DEXT;
2356 		else {
2357 			rval = 0;
2358 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2359 			if (error)
2360 				goto done;
2361 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2362 			error = xfs_bmbt_update(cur, &PREV);
2363 			if (error)
2364 				goto done;
2365 			error = xfs_btree_increment(cur, 0, &i);
2366 			if (error)
2367 				goto done;
2368 			error = xfs_bmbt_update(cur, &RIGHT);
2369 			if (error)
2370 				goto done;
2371 		}
2372 		break;
2373 
2374 	case BMAP_RIGHT_FILLING:
2375 		/*
2376 		 * Setting the last part of a previous oldext extent to newext.
2377 		 * The right neighbor is not contiguous.
2378 		 */
2379 		old = PREV;
2380 		PREV.br_blockcount -= new->br_blockcount;
2381 
2382 		xfs_iext_update_extent(ip, state, icur, &PREV);
2383 		xfs_iext_next(ifp, icur);
2384 		xfs_iext_insert(ip, icur, new, state);
2385 
2386 		XFS_IFORK_NEXT_SET(ip, whichfork,
2387 				XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2388 		if (cur == NULL)
2389 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2390 		else {
2391 			rval = XFS_ILOG_CORE;
2392 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2393 			if (error)
2394 				goto done;
2395 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2396 			error = xfs_bmbt_update(cur, &PREV);
2397 			if (error)
2398 				goto done;
2399 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2400 			if (error)
2401 				goto done;
2402 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2403 			if ((error = xfs_btree_insert(cur, &i)))
2404 				goto done;
2405 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2406 		}
2407 		break;
2408 
2409 	case 0:
2410 		/*
2411 		 * Setting the middle part of a previous oldext extent to
2412 		 * newext.  Contiguity is impossible here.
2413 		 * One extent becomes three extents.
2414 		 */
2415 		old = PREV;
2416 		PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
2417 
2418 		r[0] = *new;
2419 		r[1].br_startoff = new_endoff;
2420 		r[1].br_blockcount =
2421 			old.br_startoff + old.br_blockcount - new_endoff;
2422 		r[1].br_startblock = new->br_startblock + new->br_blockcount;
2423 		r[1].br_state = PREV.br_state;
2424 
2425 		xfs_iext_update_extent(ip, state, icur, &PREV);
2426 		xfs_iext_next(ifp, icur);
2427 		xfs_iext_insert(ip, icur, &r[1], state);
2428 		xfs_iext_insert(ip, icur, &r[0], state);
2429 
2430 		XFS_IFORK_NEXT_SET(ip, whichfork,
2431 				XFS_IFORK_NEXTENTS(ip, whichfork) + 2);
2432 		if (cur == NULL)
2433 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2434 		else {
2435 			rval = XFS_ILOG_CORE;
2436 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2437 			if (error)
2438 				goto done;
2439 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2440 			/* new right extent - oldext */
2441 			error = xfs_bmbt_update(cur, &r[1]);
2442 			if (error)
2443 				goto done;
2444 			/* new left extent - oldext */
2445 			cur->bc_rec.b = PREV;
2446 			if ((error = xfs_btree_insert(cur, &i)))
2447 				goto done;
2448 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2449 			/*
2450 			 * Reset the cursor to the position of the new extent
2451 			 * we are about to insert as we can't trust it after
2452 			 * the previous insert.
2453 			 */
2454 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2455 			if (error)
2456 				goto done;
2457 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2458 			/* new middle extent - newext */
2459 			if ((error = xfs_btree_insert(cur, &i)))
2460 				goto done;
2461 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2462 		}
2463 		break;
2464 
2465 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2466 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2467 	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2468 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2469 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2470 	case BMAP_LEFT_CONTIG:
2471 	case BMAP_RIGHT_CONTIG:
2472 		/*
2473 		 * These cases are all impossible.
2474 		 */
2475 		ASSERT(0);
2476 	}
2477 
2478 	/* update reverse mappings */
2479 	error = xfs_rmap_convert_extent(mp, tp, ip, whichfork, new);
2480 	if (error)
2481 		goto done;
2482 
2483 	/* convert to a btree if necessary */
2484 	if (xfs_bmap_needs_btree(ip, whichfork)) {
2485 		int	tmp_logflags;	/* partial log flag return val */
2486 
2487 		ASSERT(cur == NULL);
2488 		error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
2489 				&tmp_logflags, whichfork);
2490 		*logflagsp |= tmp_logflags;
2491 		if (error)
2492 			goto done;
2493 	}
2494 
2495 	/* clear out the allocated field, done with it now in any case. */
2496 	if (cur) {
2497 		cur->bc_private.b.allocated = 0;
2498 		*curp = cur;
2499 	}
2500 
2501 	xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
2502 done:
2503 	*logflagsp |= rval;
2504 	return error;
2505 #undef	LEFT
2506 #undef	RIGHT
2507 #undef	PREV
2508 }
2509 
2510 /*
2511  * Convert a hole to a delayed allocation.
2512  */
2513 STATIC void
2514 xfs_bmap_add_extent_hole_delay(
2515 	xfs_inode_t		*ip,	/* incore inode pointer */
2516 	int			whichfork,
2517 	struct xfs_iext_cursor	*icur,
2518 	xfs_bmbt_irec_t		*new)	/* new data to add to file extents */
2519 {
2520 	struct xfs_ifork	*ifp;	/* inode fork pointer */
2521 	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2522 	xfs_filblks_t		newlen=0;	/* new indirect size */
2523 	xfs_filblks_t		oldlen=0;	/* old indirect size */
2524 	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2525 	int			state = xfs_bmap_fork_to_state(whichfork);
2526 	xfs_filblks_t		temp;	 /* temp for indirect calculations */
2527 
2528 	ifp = XFS_IFORK_PTR(ip, whichfork);
2529 	ASSERT(isnullstartblock(new->br_startblock));
2530 
2531 	/*
2532 	 * Check and set flags if this segment has a left neighbor
2533 	 */
2534 	if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2535 		state |= BMAP_LEFT_VALID;
2536 		if (isnullstartblock(left.br_startblock))
2537 			state |= BMAP_LEFT_DELAY;
2538 	}
2539 
2540 	/*
2541 	 * Check and set flags if the current (right) segment exists.
2542 	 * If it doesn't exist, we're converting the hole at end-of-file.
2543 	 */
2544 	if (xfs_iext_get_extent(ifp, icur, &right)) {
2545 		state |= BMAP_RIGHT_VALID;
2546 		if (isnullstartblock(right.br_startblock))
2547 			state |= BMAP_RIGHT_DELAY;
2548 	}
2549 
2550 	/*
2551 	 * Set contiguity flags on the left and right neighbors.
2552 	 * Don't let extents get too large, even if the pieces are contiguous.
2553 	 */
2554 	if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2555 	    left.br_startoff + left.br_blockcount == new->br_startoff &&
2556 	    left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2557 		state |= BMAP_LEFT_CONTIG;
2558 
2559 	if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2560 	    new->br_startoff + new->br_blockcount == right.br_startoff &&
2561 	    new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2562 	    (!(state & BMAP_LEFT_CONTIG) ||
2563 	     (left.br_blockcount + new->br_blockcount +
2564 	      right.br_blockcount <= MAXEXTLEN)))
2565 		state |= BMAP_RIGHT_CONTIG;
2566 
2567 	/*
2568 	 * Switch out based on the contiguity flags.
2569 	 */
2570 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2571 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2572 		/*
2573 		 * New allocation is contiguous with delayed allocations
2574 		 * on the left and on the right.
2575 		 * Merge all three into a single extent record.
2576 		 */
2577 		temp = left.br_blockcount + new->br_blockcount +
2578 			right.br_blockcount;
2579 
2580 		oldlen = startblockval(left.br_startblock) +
2581 			startblockval(new->br_startblock) +
2582 			startblockval(right.br_startblock);
2583 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2584 					 oldlen);
2585 		left.br_startblock = nullstartblock(newlen);
2586 		left.br_blockcount = temp;
2587 
2588 		xfs_iext_remove(ip, icur, state);
2589 		xfs_iext_prev(ifp, icur);
2590 		xfs_iext_update_extent(ip, state, icur, &left);
2591 		break;
2592 
2593 	case BMAP_LEFT_CONTIG:
2594 		/*
2595 		 * New allocation is contiguous with a delayed allocation
2596 		 * on the left.
2597 		 * Merge the new allocation with the left neighbor.
2598 		 */
2599 		temp = left.br_blockcount + new->br_blockcount;
2600 
2601 		oldlen = startblockval(left.br_startblock) +
2602 			startblockval(new->br_startblock);
2603 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2604 					 oldlen);
2605 		left.br_blockcount = temp;
2606 		left.br_startblock = nullstartblock(newlen);
2607 
2608 		xfs_iext_prev(ifp, icur);
2609 		xfs_iext_update_extent(ip, state, icur, &left);
2610 		break;
2611 
2612 	case BMAP_RIGHT_CONTIG:
2613 		/*
2614 		 * New allocation is contiguous with a delayed allocation
2615 		 * on the right.
2616 		 * Merge the new allocation with the right neighbor.
2617 		 */
2618 		temp = new->br_blockcount + right.br_blockcount;
2619 		oldlen = startblockval(new->br_startblock) +
2620 			startblockval(right.br_startblock);
2621 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2622 					 oldlen);
2623 		right.br_startoff = new->br_startoff;
2624 		right.br_startblock = nullstartblock(newlen);
2625 		right.br_blockcount = temp;
2626 		xfs_iext_update_extent(ip, state, icur, &right);
2627 		break;
2628 
2629 	case 0:
2630 		/*
2631 		 * New allocation is not contiguous with another
2632 		 * delayed allocation.
2633 		 * Insert a new entry.
2634 		 */
2635 		oldlen = newlen = 0;
2636 		xfs_iext_insert(ip, icur, new, state);
2637 		break;
2638 	}
2639 	if (oldlen != newlen) {
2640 		ASSERT(oldlen > newlen);
2641 		xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2642 				 false);
2643 		/*
2644 		 * Nothing to do for disk quota accounting here.
2645 		 */
2646 		xfs_mod_delalloc(ip->i_mount, (int64_t)newlen - oldlen);
2647 	}
2648 }
2649 
2650 /*
2651  * Convert a hole to a real allocation.
2652  */
2653 STATIC int				/* error */
2654 xfs_bmap_add_extent_hole_real(
2655 	struct xfs_trans	*tp,
2656 	struct xfs_inode	*ip,
2657 	int			whichfork,
2658 	struct xfs_iext_cursor	*icur,
2659 	struct xfs_btree_cur	**curp,
2660 	struct xfs_bmbt_irec	*new,
2661 	int			*logflagsp,
2662 	int			flags)
2663 {
2664 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
2665 	struct xfs_mount	*mp = ip->i_mount;
2666 	struct xfs_btree_cur	*cur = *curp;
2667 	int			error;	/* error return value */
2668 	int			i;	/* temp state */
2669 	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2670 	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2671 	int			rval=0;	/* return value (logging flags) */
2672 	int			state = xfs_bmap_fork_to_state(whichfork);
2673 	struct xfs_bmbt_irec	old;
2674 
2675 	ASSERT(!isnullstartblock(new->br_startblock));
2676 	ASSERT(!cur || !(cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2677 
2678 	XFS_STATS_INC(mp, xs_add_exlist);
2679 
2680 	/*
2681 	 * Check and set flags if this segment has a left neighbor.
2682 	 */
2683 	if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2684 		state |= BMAP_LEFT_VALID;
2685 		if (isnullstartblock(left.br_startblock))
2686 			state |= BMAP_LEFT_DELAY;
2687 	}
2688 
2689 	/*
2690 	 * Check and set flags if this segment has a current value.
2691 	 * Not true if we're inserting into the "hole" at eof.
2692 	 */
2693 	if (xfs_iext_get_extent(ifp, icur, &right)) {
2694 		state |= BMAP_RIGHT_VALID;
2695 		if (isnullstartblock(right.br_startblock))
2696 			state |= BMAP_RIGHT_DELAY;
2697 	}
2698 
2699 	/*
2700 	 * We're inserting a real allocation between "left" and "right".
2701 	 * Set the contiguity flags.  Don't let extents get too large.
2702 	 */
2703 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2704 	    left.br_startoff + left.br_blockcount == new->br_startoff &&
2705 	    left.br_startblock + left.br_blockcount == new->br_startblock &&
2706 	    left.br_state == new->br_state &&
2707 	    left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2708 		state |= BMAP_LEFT_CONTIG;
2709 
2710 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2711 	    new->br_startoff + new->br_blockcount == right.br_startoff &&
2712 	    new->br_startblock + new->br_blockcount == right.br_startblock &&
2713 	    new->br_state == right.br_state &&
2714 	    new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2715 	    (!(state & BMAP_LEFT_CONTIG) ||
2716 	     left.br_blockcount + new->br_blockcount +
2717 	     right.br_blockcount <= MAXEXTLEN))
2718 		state |= BMAP_RIGHT_CONTIG;
2719 
2720 	error = 0;
2721 	/*
2722 	 * Select which case we're in here, and implement it.
2723 	 */
2724 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2725 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2726 		/*
2727 		 * New allocation is contiguous with real allocations on the
2728 		 * left and on the right.
2729 		 * Merge all three into a single extent record.
2730 		 */
2731 		left.br_blockcount += new->br_blockcount + right.br_blockcount;
2732 
2733 		xfs_iext_remove(ip, icur, state);
2734 		xfs_iext_prev(ifp, icur);
2735 		xfs_iext_update_extent(ip, state, icur, &left);
2736 
2737 		XFS_IFORK_NEXT_SET(ip, whichfork,
2738 			XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2739 		if (cur == NULL) {
2740 			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2741 		} else {
2742 			rval = XFS_ILOG_CORE;
2743 			error = xfs_bmbt_lookup_eq(cur, &right, &i);
2744 			if (error)
2745 				goto done;
2746 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2747 			error = xfs_btree_delete(cur, &i);
2748 			if (error)
2749 				goto done;
2750 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2751 			error = xfs_btree_decrement(cur, 0, &i);
2752 			if (error)
2753 				goto done;
2754 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2755 			error = xfs_bmbt_update(cur, &left);
2756 			if (error)
2757 				goto done;
2758 		}
2759 		break;
2760 
2761 	case BMAP_LEFT_CONTIG:
2762 		/*
2763 		 * New allocation is contiguous with a real allocation
2764 		 * on the left.
2765 		 * Merge the new allocation with the left neighbor.
2766 		 */
2767 		old = left;
2768 		left.br_blockcount += new->br_blockcount;
2769 
2770 		xfs_iext_prev(ifp, icur);
2771 		xfs_iext_update_extent(ip, state, icur, &left);
2772 
2773 		if (cur == NULL) {
2774 			rval = xfs_ilog_fext(whichfork);
2775 		} else {
2776 			rval = 0;
2777 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2778 			if (error)
2779 				goto done;
2780 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2781 			error = xfs_bmbt_update(cur, &left);
2782 			if (error)
2783 				goto done;
2784 		}
2785 		break;
2786 
2787 	case BMAP_RIGHT_CONTIG:
2788 		/*
2789 		 * New allocation is contiguous with a real allocation
2790 		 * on the right.
2791 		 * Merge the new allocation with the right neighbor.
2792 		 */
2793 		old = right;
2794 
2795 		right.br_startoff = new->br_startoff;
2796 		right.br_startblock = new->br_startblock;
2797 		right.br_blockcount += new->br_blockcount;
2798 		xfs_iext_update_extent(ip, state, icur, &right);
2799 
2800 		if (cur == NULL) {
2801 			rval = xfs_ilog_fext(whichfork);
2802 		} else {
2803 			rval = 0;
2804 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2805 			if (error)
2806 				goto done;
2807 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2808 			error = xfs_bmbt_update(cur, &right);
2809 			if (error)
2810 				goto done;
2811 		}
2812 		break;
2813 
2814 	case 0:
2815 		/*
2816 		 * New allocation is not contiguous with another
2817 		 * real allocation.
2818 		 * Insert a new entry.
2819 		 */
2820 		xfs_iext_insert(ip, icur, new, state);
2821 		XFS_IFORK_NEXT_SET(ip, whichfork,
2822 			XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2823 		if (cur == NULL) {
2824 			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2825 		} else {
2826 			rval = XFS_ILOG_CORE;
2827 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2828 			if (error)
2829 				goto done;
2830 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2831 			error = xfs_btree_insert(cur, &i);
2832 			if (error)
2833 				goto done;
2834 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2835 		}
2836 		break;
2837 	}
2838 
2839 	/* add reverse mapping unless caller opted out */
2840 	if (!(flags & XFS_BMAPI_NORMAP)) {
2841 		error = xfs_rmap_map_extent(tp, ip, whichfork, new);
2842 		if (error)
2843 			goto done;
2844 	}
2845 
2846 	/* convert to a btree if necessary */
2847 	if (xfs_bmap_needs_btree(ip, whichfork)) {
2848 		int	tmp_logflags;	/* partial log flag return val */
2849 
2850 		ASSERT(cur == NULL);
2851 		error = xfs_bmap_extents_to_btree(tp, ip, curp, 0,
2852 				&tmp_logflags, whichfork);
2853 		*logflagsp |= tmp_logflags;
2854 		cur = *curp;
2855 		if (error)
2856 			goto done;
2857 	}
2858 
2859 	/* clear out the allocated field, done with it now in any case. */
2860 	if (cur)
2861 		cur->bc_private.b.allocated = 0;
2862 
2863 	xfs_bmap_check_leaf_extents(cur, ip, whichfork);
2864 done:
2865 	*logflagsp |= rval;
2866 	return error;
2867 }
2868 
2869 /*
2870  * Functions used in the extent read, allocate and remove paths
2871  */
2872 
2873 /*
2874  * Adjust the size of the new extent based on di_extsize and rt extsize.
2875  */
2876 int
2877 xfs_bmap_extsize_align(
2878 	xfs_mount_t	*mp,
2879 	xfs_bmbt_irec_t	*gotp,		/* next extent pointer */
2880 	xfs_bmbt_irec_t	*prevp,		/* previous extent pointer */
2881 	xfs_extlen_t	extsz,		/* align to this extent size */
2882 	int		rt,		/* is this a realtime inode? */
2883 	int		eof,		/* is extent at end-of-file? */
2884 	int		delay,		/* creating delalloc extent? */
2885 	int		convert,	/* overwriting unwritten extent? */
2886 	xfs_fileoff_t	*offp,		/* in/out: aligned offset */
2887 	xfs_extlen_t	*lenp)		/* in/out: aligned length */
2888 {
2889 	xfs_fileoff_t	orig_off;	/* original offset */
2890 	xfs_extlen_t	orig_alen;	/* original length */
2891 	xfs_fileoff_t	orig_end;	/* original off+len */
2892 	xfs_fileoff_t	nexto;		/* next file offset */
2893 	xfs_fileoff_t	prevo;		/* previous file offset */
2894 	xfs_fileoff_t	align_off;	/* temp for offset */
2895 	xfs_extlen_t	align_alen;	/* temp for length */
2896 	xfs_extlen_t	temp;		/* temp for calculations */
2897 
2898 	if (convert)
2899 		return 0;
2900 
2901 	orig_off = align_off = *offp;
2902 	orig_alen = align_alen = *lenp;
2903 	orig_end = orig_off + orig_alen;
2904 
2905 	/*
2906 	 * If this request overlaps an existing extent, then don't
2907 	 * attempt to perform any additional alignment.
2908 	 */
2909 	if (!delay && !eof &&
2910 	    (orig_off >= gotp->br_startoff) &&
2911 	    (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2912 		return 0;
2913 	}
2914 
2915 	/*
2916 	 * If the file offset is unaligned vs. the extent size
2917 	 * we need to align it.  This will be possible unless
2918 	 * the file was previously written with a kernel that didn't
2919 	 * perform this alignment, or if a truncate shot us in the
2920 	 * foot.
2921 	 */
2922 	div_u64_rem(orig_off, extsz, &temp);
2923 	if (temp) {
2924 		align_alen += temp;
2925 		align_off -= temp;
2926 	}
2927 
2928 	/* Same adjustment for the end of the requested area. */
2929 	temp = (align_alen % extsz);
2930 	if (temp)
2931 		align_alen += extsz - temp;
2932 
2933 	/*
2934 	 * For large extent hint sizes, the aligned extent might be larger than
2935 	 * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
2936 	 * the length back under MAXEXTLEN. The outer allocation loops handle
2937 	 * short allocation just fine, so it is safe to do this. We only want to
2938 	 * do it when we are forced to, though, because it means more allocation
2939 	 * operations are required.
2940 	 */
2941 	while (align_alen > MAXEXTLEN)
2942 		align_alen -= extsz;
2943 	ASSERT(align_alen <= MAXEXTLEN);
2944 
2945 	/*
2946 	 * If the previous block overlaps with this proposed allocation
2947 	 * then move the start forward without adjusting the length.
2948 	 */
2949 	if (prevp->br_startoff != NULLFILEOFF) {
2950 		if (prevp->br_startblock == HOLESTARTBLOCK)
2951 			prevo = prevp->br_startoff;
2952 		else
2953 			prevo = prevp->br_startoff + prevp->br_blockcount;
2954 	} else
2955 		prevo = 0;
2956 	if (align_off != orig_off && align_off < prevo)
2957 		align_off = prevo;
2958 	/*
2959 	 * If the next block overlaps with this proposed allocation
2960 	 * then move the start back without adjusting the length,
2961 	 * but not before offset 0.
2962 	 * This may of course make the start overlap previous block,
2963 	 * and if we hit the offset 0 limit then the next block
2964 	 * can still overlap too.
2965 	 */
2966 	if (!eof && gotp->br_startoff != NULLFILEOFF) {
2967 		if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2968 		    (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2969 			nexto = gotp->br_startoff + gotp->br_blockcount;
2970 		else
2971 			nexto = gotp->br_startoff;
2972 	} else
2973 		nexto = NULLFILEOFF;
2974 	if (!eof &&
2975 	    align_off + align_alen != orig_end &&
2976 	    align_off + align_alen > nexto)
2977 		align_off = nexto > align_alen ? nexto - align_alen : 0;
2978 	/*
2979 	 * If we're now overlapping the next or previous extent that
2980 	 * means we can't fit an extsz piece in this hole.  Just move
2981 	 * the start forward to the first valid spot and set
2982 	 * the length so we hit the end.
2983 	 */
2984 	if (align_off != orig_off && align_off < prevo)
2985 		align_off = prevo;
2986 	if (align_off + align_alen != orig_end &&
2987 	    align_off + align_alen > nexto &&
2988 	    nexto != NULLFILEOFF) {
2989 		ASSERT(nexto > prevo);
2990 		align_alen = nexto - align_off;
2991 	}
2992 
2993 	/*
2994 	 * If realtime, and the result isn't a multiple of the realtime
2995 	 * extent size we need to remove blocks until it is.
2996 	 */
2997 	if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2998 		/*
2999 		 * We're not covering the original request, or
3000 		 * we won't be able to once we fix the length.
3001 		 */
3002 		if (orig_off < align_off ||
3003 		    orig_end > align_off + align_alen ||
3004 		    align_alen - temp < orig_alen)
3005 			return -EINVAL;
3006 		/*
3007 		 * Try to fix it by moving the start up.
3008 		 */
3009 		if (align_off + temp <= orig_off) {
3010 			align_alen -= temp;
3011 			align_off += temp;
3012 		}
3013 		/*
3014 		 * Try to fix it by moving the end in.
3015 		 */
3016 		else if (align_off + align_alen - temp >= orig_end)
3017 			align_alen -= temp;
3018 		/*
3019 		 * Set the start to the minimum then trim the length.
3020 		 */
3021 		else {
3022 			align_alen -= orig_off - align_off;
3023 			align_off = orig_off;
3024 			align_alen -= align_alen % mp->m_sb.sb_rextsize;
3025 		}
3026 		/*
3027 		 * Result doesn't cover the request, fail it.
3028 		 */
3029 		if (orig_off < align_off || orig_end > align_off + align_alen)
3030 			return -EINVAL;
3031 	} else {
3032 		ASSERT(orig_off >= align_off);
3033 		/* see MAXEXTLEN handling above */
3034 		ASSERT(orig_end <= align_off + align_alen ||
3035 		       align_alen + extsz > MAXEXTLEN);
3036 	}
3037 
3038 #ifdef DEBUG
3039 	if (!eof && gotp->br_startoff != NULLFILEOFF)
3040 		ASSERT(align_off + align_alen <= gotp->br_startoff);
3041 	if (prevp->br_startoff != NULLFILEOFF)
3042 		ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3043 #endif
3044 
3045 	*lenp = align_alen;
3046 	*offp = align_off;
3047 	return 0;
3048 }
3049 
3050 #define XFS_ALLOC_GAP_UNITS	4
3051 
3052 void
3053 xfs_bmap_adjacent(
3054 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3055 {
3056 	xfs_fsblock_t	adjust;		/* adjustment to block numbers */
3057 	xfs_agnumber_t	fb_agno;	/* ag number of ap->firstblock */
3058 	xfs_mount_t	*mp;		/* mount point structure */
3059 	int		nullfb;		/* true if ap->firstblock isn't set */
3060 	int		rt;		/* true if inode is realtime */
3061 
3062 #define	ISVALID(x,y)	\
3063 	(rt ? \
3064 		(x) < mp->m_sb.sb_rblocks : \
3065 		XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3066 		XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3067 		XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3068 
3069 	mp = ap->ip->i_mount;
3070 	nullfb = ap->tp->t_firstblock == NULLFSBLOCK;
3071 	rt = XFS_IS_REALTIME_INODE(ap->ip) &&
3072 		xfs_alloc_is_userdata(ap->datatype);
3073 	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp,
3074 							ap->tp->t_firstblock);
3075 	/*
3076 	 * If allocating at eof, and there's a previous real block,
3077 	 * try to use its last block as our starting point.
3078 	 */
3079 	if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3080 	    !isnullstartblock(ap->prev.br_startblock) &&
3081 	    ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3082 		    ap->prev.br_startblock)) {
3083 		ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3084 		/*
3085 		 * Adjust for the gap between prevp and us.
3086 		 */
3087 		adjust = ap->offset -
3088 			(ap->prev.br_startoff + ap->prev.br_blockcount);
3089 		if (adjust &&
3090 		    ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3091 			ap->blkno += adjust;
3092 	}
3093 	/*
3094 	 * If not at eof, then compare the two neighbor blocks.
3095 	 * Figure out whether either one gives us a good starting point,
3096 	 * and pick the better one.
3097 	 */
3098 	else if (!ap->eof) {
3099 		xfs_fsblock_t	gotbno;		/* right side block number */
3100 		xfs_fsblock_t	gotdiff=0;	/* right side difference */
3101 		xfs_fsblock_t	prevbno;	/* left side block number */
3102 		xfs_fsblock_t	prevdiff=0;	/* left side difference */
3103 
3104 		/*
3105 		 * If there's a previous (left) block, select a requested
3106 		 * start block based on it.
3107 		 */
3108 		if (ap->prev.br_startoff != NULLFILEOFF &&
3109 		    !isnullstartblock(ap->prev.br_startblock) &&
3110 		    (prevbno = ap->prev.br_startblock +
3111 			       ap->prev.br_blockcount) &&
3112 		    ISVALID(prevbno, ap->prev.br_startblock)) {
3113 			/*
3114 			 * Calculate gap to end of previous block.
3115 			 */
3116 			adjust = prevdiff = ap->offset -
3117 				(ap->prev.br_startoff +
3118 				 ap->prev.br_blockcount);
3119 			/*
3120 			 * Figure the startblock based on the previous block's
3121 			 * end and the gap size.
3122 			 * Heuristic!
3123 			 * If the gap is large relative to the piece we're
3124 			 * allocating, or using it gives us an invalid block
3125 			 * number, then just use the end of the previous block.
3126 			 */
3127 			if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3128 			    ISVALID(prevbno + prevdiff,
3129 				    ap->prev.br_startblock))
3130 				prevbno += adjust;
3131 			else
3132 				prevdiff += adjust;
3133 			/*
3134 			 * If the firstblock forbids it, can't use it,
3135 			 * must use default.
3136 			 */
3137 			if (!rt && !nullfb &&
3138 			    XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3139 				prevbno = NULLFSBLOCK;
3140 		}
3141 		/*
3142 		 * No previous block or can't follow it, just default.
3143 		 */
3144 		else
3145 			prevbno = NULLFSBLOCK;
3146 		/*
3147 		 * If there's a following (right) block, select a requested
3148 		 * start block based on it.
3149 		 */
3150 		if (!isnullstartblock(ap->got.br_startblock)) {
3151 			/*
3152 			 * Calculate gap to start of next block.
3153 			 */
3154 			adjust = gotdiff = ap->got.br_startoff - ap->offset;
3155 			/*
3156 			 * Figure the startblock based on the next block's
3157 			 * start and the gap size.
3158 			 */
3159 			gotbno = ap->got.br_startblock;
3160 			/*
3161 			 * Heuristic!
3162 			 * If the gap is large relative to the piece we're
3163 			 * allocating, or using it gives us an invalid block
3164 			 * number, then just use the start of the next block
3165 			 * offset by our length.
3166 			 */
3167 			if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3168 			    ISVALID(gotbno - gotdiff, gotbno))
3169 				gotbno -= adjust;
3170 			else if (ISVALID(gotbno - ap->length, gotbno)) {
3171 				gotbno -= ap->length;
3172 				gotdiff += adjust - ap->length;
3173 			} else
3174 				gotdiff += adjust;
3175 			/*
3176 			 * If the firstblock forbids it, can't use it,
3177 			 * must use default.
3178 			 */
3179 			if (!rt && !nullfb &&
3180 			    XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3181 				gotbno = NULLFSBLOCK;
3182 		}
3183 		/*
3184 		 * No next block, just default.
3185 		 */
3186 		else
3187 			gotbno = NULLFSBLOCK;
3188 		/*
3189 		 * If both valid, pick the better one, else the only good
3190 		 * one, else ap->blkno is already set (to 0 or the inode block).
3191 		 */
3192 		if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3193 			ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3194 		else if (prevbno != NULLFSBLOCK)
3195 			ap->blkno = prevbno;
3196 		else if (gotbno != NULLFSBLOCK)
3197 			ap->blkno = gotbno;
3198 	}
3199 #undef ISVALID
3200 }
3201 
3202 static int
3203 xfs_bmap_longest_free_extent(
3204 	struct xfs_trans	*tp,
3205 	xfs_agnumber_t		ag,
3206 	xfs_extlen_t		*blen,
3207 	int			*notinit)
3208 {
3209 	struct xfs_mount	*mp = tp->t_mountp;
3210 	struct xfs_perag	*pag;
3211 	xfs_extlen_t		longest;
3212 	int			error = 0;
3213 
3214 	pag = xfs_perag_get(mp, ag);
3215 	if (!pag->pagf_init) {
3216 		error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3217 		if (error)
3218 			goto out;
3219 
3220 		if (!pag->pagf_init) {
3221 			*notinit = 1;
3222 			goto out;
3223 		}
3224 	}
3225 
3226 	longest = xfs_alloc_longest_free_extent(pag,
3227 				xfs_alloc_min_freelist(mp, pag),
3228 				xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
3229 	if (*blen < longest)
3230 		*blen = longest;
3231 
3232 out:
3233 	xfs_perag_put(pag);
3234 	return error;
3235 }
3236 
3237 static void
3238 xfs_bmap_select_minlen(
3239 	struct xfs_bmalloca	*ap,
3240 	struct xfs_alloc_arg	*args,
3241 	xfs_extlen_t		*blen,
3242 	int			notinit)
3243 {
3244 	if (notinit || *blen < ap->minlen) {
3245 		/*
3246 		 * Since we did a BUF_TRYLOCK above, it is possible that
3247 		 * there is space for this request.
3248 		 */
3249 		args->minlen = ap->minlen;
3250 	} else if (*blen < args->maxlen) {
3251 		/*
3252 		 * If the best seen length is less than the request length,
3253 		 * use the best as the minimum.
3254 		 */
3255 		args->minlen = *blen;
3256 	} else {
3257 		/*
3258 		 * Otherwise we've seen an extent as big as maxlen, use that
3259 		 * as the minimum.
3260 		 */
3261 		args->minlen = args->maxlen;
3262 	}
3263 }
3264 
3265 STATIC int
3266 xfs_bmap_btalloc_nullfb(
3267 	struct xfs_bmalloca	*ap,
3268 	struct xfs_alloc_arg	*args,
3269 	xfs_extlen_t		*blen)
3270 {
3271 	struct xfs_mount	*mp = ap->ip->i_mount;
3272 	xfs_agnumber_t		ag, startag;
3273 	int			notinit = 0;
3274 	int			error;
3275 
3276 	args->type = XFS_ALLOCTYPE_START_BNO;
3277 	args->total = ap->total;
3278 
3279 	startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3280 	if (startag == NULLAGNUMBER)
3281 		startag = ag = 0;
3282 
3283 	while (*blen < args->maxlen) {
3284 		error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3285 						     &notinit);
3286 		if (error)
3287 			return error;
3288 
3289 		if (++ag == mp->m_sb.sb_agcount)
3290 			ag = 0;
3291 		if (ag == startag)
3292 			break;
3293 	}
3294 
3295 	xfs_bmap_select_minlen(ap, args, blen, notinit);
3296 	return 0;
3297 }
3298 
3299 STATIC int
3300 xfs_bmap_btalloc_filestreams(
3301 	struct xfs_bmalloca	*ap,
3302 	struct xfs_alloc_arg	*args,
3303 	xfs_extlen_t		*blen)
3304 {
3305 	struct xfs_mount	*mp = ap->ip->i_mount;
3306 	xfs_agnumber_t		ag;
3307 	int			notinit = 0;
3308 	int			error;
3309 
3310 	args->type = XFS_ALLOCTYPE_NEAR_BNO;
3311 	args->total = ap->total;
3312 
3313 	ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3314 	if (ag == NULLAGNUMBER)
3315 		ag = 0;
3316 
3317 	error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3318 	if (error)
3319 		return error;
3320 
3321 	if (*blen < args->maxlen) {
3322 		error = xfs_filestream_new_ag(ap, &ag);
3323 		if (error)
3324 			return error;
3325 
3326 		error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3327 						     &notinit);
3328 		if (error)
3329 			return error;
3330 
3331 	}
3332 
3333 	xfs_bmap_select_minlen(ap, args, blen, notinit);
3334 
3335 	/*
3336 	 * Set the failure fallback case to look in the selected AG as stream
3337 	 * may have moved.
3338 	 */
3339 	ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3340 	return 0;
3341 }
3342 
3343 /* Update all inode and quota accounting for the allocation we just did. */
3344 static void
3345 xfs_bmap_btalloc_accounting(
3346 	struct xfs_bmalloca	*ap,
3347 	struct xfs_alloc_arg	*args)
3348 {
3349 	if (ap->flags & XFS_BMAPI_COWFORK) {
3350 		/*
3351 		 * COW fork blocks are in-core only and thus are treated as
3352 		 * in-core quota reservation (like delalloc blocks) even when
3353 		 * converted to real blocks. The quota reservation is not
3354 		 * accounted to disk until blocks are remapped to the data
3355 		 * fork. So if these blocks were previously delalloc, we
3356 		 * already have quota reservation and there's nothing to do
3357 		 * yet.
3358 		 */
3359 		if (ap->wasdel) {
3360 			xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len);
3361 			return;
3362 		}
3363 
3364 		/*
3365 		 * Otherwise, we've allocated blocks in a hole. The transaction
3366 		 * has acquired in-core quota reservation for this extent.
3367 		 * Rather than account these as real blocks, however, we reduce
3368 		 * the transaction quota reservation based on the allocation.
3369 		 * This essentially transfers the transaction quota reservation
3370 		 * to that of a delalloc extent.
3371 		 */
3372 		ap->ip->i_delayed_blks += args->len;
3373 		xfs_trans_mod_dquot_byino(ap->tp, ap->ip, XFS_TRANS_DQ_RES_BLKS,
3374 				-(long)args->len);
3375 		return;
3376 	}
3377 
3378 	/* data/attr fork only */
3379 	ap->ip->i_d.di_nblocks += args->len;
3380 	xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3381 	if (ap->wasdel) {
3382 		ap->ip->i_delayed_blks -= args->len;
3383 		xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len);
3384 	}
3385 	xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3386 		ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT : XFS_TRANS_DQ_BCOUNT,
3387 		args->len);
3388 }
3389 
3390 STATIC int
3391 xfs_bmap_btalloc(
3392 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3393 {
3394 	xfs_mount_t	*mp;		/* mount point structure */
3395 	xfs_alloctype_t	atype = 0;	/* type for allocation routines */
3396 	xfs_extlen_t	align = 0;	/* minimum allocation alignment */
3397 	xfs_agnumber_t	fb_agno;	/* ag number of ap->firstblock */
3398 	xfs_agnumber_t	ag;
3399 	xfs_alloc_arg_t	args;
3400 	xfs_fileoff_t	orig_offset;
3401 	xfs_extlen_t	orig_length;
3402 	xfs_extlen_t	blen;
3403 	xfs_extlen_t	nextminlen = 0;
3404 	int		nullfb;		/* true if ap->firstblock isn't set */
3405 	int		isaligned;
3406 	int		tryagain;
3407 	int		error;
3408 	int		stripe_align;
3409 
3410 	ASSERT(ap->length);
3411 	orig_offset = ap->offset;
3412 	orig_length = ap->length;
3413 
3414 	mp = ap->ip->i_mount;
3415 
3416 	/* stripe alignment for allocation is determined by mount parameters */
3417 	stripe_align = 0;
3418 	if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3419 		stripe_align = mp->m_swidth;
3420 	else if (mp->m_dalign)
3421 		stripe_align = mp->m_dalign;
3422 
3423 	if (ap->flags & XFS_BMAPI_COWFORK)
3424 		align = xfs_get_cowextsz_hint(ap->ip);
3425 	else if (xfs_alloc_is_userdata(ap->datatype))
3426 		align = xfs_get_extsz_hint(ap->ip);
3427 	if (align) {
3428 		error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3429 						align, 0, ap->eof, 0, ap->conv,
3430 						&ap->offset, &ap->length);
3431 		ASSERT(!error);
3432 		ASSERT(ap->length);
3433 	}
3434 
3435 
3436 	nullfb = ap->tp->t_firstblock == NULLFSBLOCK;
3437 	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp,
3438 							ap->tp->t_firstblock);
3439 	if (nullfb) {
3440 		if (xfs_alloc_is_userdata(ap->datatype) &&
3441 		    xfs_inode_is_filestream(ap->ip)) {
3442 			ag = xfs_filestream_lookup_ag(ap->ip);
3443 			ag = (ag != NULLAGNUMBER) ? ag : 0;
3444 			ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3445 		} else {
3446 			ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3447 		}
3448 	} else
3449 		ap->blkno = ap->tp->t_firstblock;
3450 
3451 	xfs_bmap_adjacent(ap);
3452 
3453 	/*
3454 	 * If allowed, use ap->blkno; otherwise must use firstblock since
3455 	 * it's in the right allocation group.
3456 	 */
3457 	if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3458 		;
3459 	else
3460 		ap->blkno = ap->tp->t_firstblock;
3461 	/*
3462 	 * Normal allocation, done through xfs_alloc_vextent.
3463 	 */
3464 	tryagain = isaligned = 0;
3465 	memset(&args, 0, sizeof(args));
3466 	args.tp = ap->tp;
3467 	args.mp = mp;
3468 	args.fsbno = ap->blkno;
3469 	args.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
3470 
3471 	/* Trim the allocation back to the maximum an AG can fit. */
3472 	args.maxlen = min(ap->length, mp->m_ag_max_usable);
3473 	blen = 0;
3474 	if (nullfb) {
3475 		/*
3476 		 * Search for an allocation group with a single extent large
3477 		 * enough for the request.  If one isn't found, then adjust
3478 		 * the minimum allocation size to the largest space found.
3479 		 */
3480 		if (xfs_alloc_is_userdata(ap->datatype) &&
3481 		    xfs_inode_is_filestream(ap->ip))
3482 			error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3483 		else
3484 			error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3485 		if (error)
3486 			return error;
3487 	} else if (ap->tp->t_flags & XFS_TRANS_LOWMODE) {
3488 		if (xfs_inode_is_filestream(ap->ip))
3489 			args.type = XFS_ALLOCTYPE_FIRST_AG;
3490 		else
3491 			args.type = XFS_ALLOCTYPE_START_BNO;
3492 		args.total = args.minlen = ap->minlen;
3493 	} else {
3494 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
3495 		args.total = ap->total;
3496 		args.minlen = ap->minlen;
3497 	}
3498 	/* apply extent size hints if obtained earlier */
3499 	if (align) {
3500 		args.prod = align;
3501 		div_u64_rem(ap->offset, args.prod, &args.mod);
3502 		if (args.mod)
3503 			args.mod = args.prod - args.mod;
3504 	} else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3505 		args.prod = 1;
3506 		args.mod = 0;
3507 	} else {
3508 		args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3509 		div_u64_rem(ap->offset, args.prod, &args.mod);
3510 		if (args.mod)
3511 			args.mod = args.prod - args.mod;
3512 	}
3513 	/*
3514 	 * If we are not low on available data blocks, and the
3515 	 * underlying logical volume manager is a stripe, and
3516 	 * the file offset is zero then try to allocate data
3517 	 * blocks on stripe unit boundary.
3518 	 * NOTE: ap->aeof is only set if the allocation length
3519 	 * is >= the stripe unit and the allocation offset is
3520 	 * at the end of file.
3521 	 */
3522 	if (!(ap->tp->t_flags & XFS_TRANS_LOWMODE) && ap->aeof) {
3523 		if (!ap->offset) {
3524 			args.alignment = stripe_align;
3525 			atype = args.type;
3526 			isaligned = 1;
3527 			/*
3528 			 * Adjust for alignment
3529 			 */
3530 			if (blen > args.alignment && blen <= args.maxlen)
3531 				args.minlen = blen - args.alignment;
3532 			args.minalignslop = 0;
3533 		} else {
3534 			/*
3535 			 * First try an exact bno allocation.
3536 			 * If it fails then do a near or start bno
3537 			 * allocation with alignment turned on.
3538 			 */
3539 			atype = args.type;
3540 			tryagain = 1;
3541 			args.type = XFS_ALLOCTYPE_THIS_BNO;
3542 			args.alignment = 1;
3543 			/*
3544 			 * Compute the minlen+alignment for the
3545 			 * next case.  Set slop so that the value
3546 			 * of minlen+alignment+slop doesn't go up
3547 			 * between the calls.
3548 			 */
3549 			if (blen > stripe_align && blen <= args.maxlen)
3550 				nextminlen = blen - stripe_align;
3551 			else
3552 				nextminlen = args.minlen;
3553 			if (nextminlen + stripe_align > args.minlen + 1)
3554 				args.minalignslop =
3555 					nextminlen + stripe_align -
3556 					args.minlen - 1;
3557 			else
3558 				args.minalignslop = 0;
3559 		}
3560 	} else {
3561 		args.alignment = 1;
3562 		args.minalignslop = 0;
3563 	}
3564 	args.minleft = ap->minleft;
3565 	args.wasdel = ap->wasdel;
3566 	args.resv = XFS_AG_RESV_NONE;
3567 	args.datatype = ap->datatype;
3568 	if (ap->datatype & XFS_ALLOC_USERDATA_ZERO)
3569 		args.ip = ap->ip;
3570 
3571 	error = xfs_alloc_vextent(&args);
3572 	if (error)
3573 		return error;
3574 
3575 	if (tryagain && args.fsbno == NULLFSBLOCK) {
3576 		/*
3577 		 * Exact allocation failed. Now try with alignment
3578 		 * turned on.
3579 		 */
3580 		args.type = atype;
3581 		args.fsbno = ap->blkno;
3582 		args.alignment = stripe_align;
3583 		args.minlen = nextminlen;
3584 		args.minalignslop = 0;
3585 		isaligned = 1;
3586 		if ((error = xfs_alloc_vextent(&args)))
3587 			return error;
3588 	}
3589 	if (isaligned && args.fsbno == NULLFSBLOCK) {
3590 		/*
3591 		 * allocation failed, so turn off alignment and
3592 		 * try again.
3593 		 */
3594 		args.type = atype;
3595 		args.fsbno = ap->blkno;
3596 		args.alignment = 0;
3597 		if ((error = xfs_alloc_vextent(&args)))
3598 			return error;
3599 	}
3600 	if (args.fsbno == NULLFSBLOCK && nullfb &&
3601 	    args.minlen > ap->minlen) {
3602 		args.minlen = ap->minlen;
3603 		args.type = XFS_ALLOCTYPE_START_BNO;
3604 		args.fsbno = ap->blkno;
3605 		if ((error = xfs_alloc_vextent(&args)))
3606 			return error;
3607 	}
3608 	if (args.fsbno == NULLFSBLOCK && nullfb) {
3609 		args.fsbno = 0;
3610 		args.type = XFS_ALLOCTYPE_FIRST_AG;
3611 		args.total = ap->minlen;
3612 		if ((error = xfs_alloc_vextent(&args)))
3613 			return error;
3614 		ap->tp->t_flags |= XFS_TRANS_LOWMODE;
3615 	}
3616 	if (args.fsbno != NULLFSBLOCK) {
3617 		/*
3618 		 * check the allocation happened at the same or higher AG than
3619 		 * the first block that was allocated.
3620 		 */
3621 		ASSERT(ap->tp->t_firstblock == NULLFSBLOCK ||
3622 		       XFS_FSB_TO_AGNO(mp, ap->tp->t_firstblock) <=
3623 		       XFS_FSB_TO_AGNO(mp, args.fsbno));
3624 
3625 		ap->blkno = args.fsbno;
3626 		if (ap->tp->t_firstblock == NULLFSBLOCK)
3627 			ap->tp->t_firstblock = args.fsbno;
3628 		ASSERT(nullfb || fb_agno <= args.agno);
3629 		ap->length = args.len;
3630 		/*
3631 		 * If the extent size hint is active, we tried to round the
3632 		 * caller's allocation request offset down to extsz and the
3633 		 * length up to another extsz boundary.  If we found a free
3634 		 * extent we mapped it in starting at this new offset.  If the
3635 		 * newly mapped space isn't long enough to cover any of the
3636 		 * range of offsets that was originally requested, move the
3637 		 * mapping up so that we can fill as much of the caller's
3638 		 * original request as possible.  Free space is apparently
3639 		 * very fragmented so we're unlikely to be able to satisfy the
3640 		 * hints anyway.
3641 		 */
3642 		if (ap->length <= orig_length)
3643 			ap->offset = orig_offset;
3644 		else if (ap->offset + ap->length < orig_offset + orig_length)
3645 			ap->offset = orig_offset + orig_length - ap->length;
3646 		xfs_bmap_btalloc_accounting(ap, &args);
3647 	} else {
3648 		ap->blkno = NULLFSBLOCK;
3649 		ap->length = 0;
3650 	}
3651 	return 0;
3652 }
3653 
3654 /*
3655  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3656  * It figures out where to ask the underlying allocator to put the new extent.
3657  */
3658 STATIC int
3659 xfs_bmap_alloc(
3660 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3661 {
3662 	if (XFS_IS_REALTIME_INODE(ap->ip) &&
3663 	    xfs_alloc_is_userdata(ap->datatype))
3664 		return xfs_bmap_rtalloc(ap);
3665 	return xfs_bmap_btalloc(ap);
3666 }
3667 
3668 /* Trim extent to fit a logical block range. */
3669 void
3670 xfs_trim_extent(
3671 	struct xfs_bmbt_irec	*irec,
3672 	xfs_fileoff_t		bno,
3673 	xfs_filblks_t		len)
3674 {
3675 	xfs_fileoff_t		distance;
3676 	xfs_fileoff_t		end = bno + len;
3677 
3678 	if (irec->br_startoff + irec->br_blockcount <= bno ||
3679 	    irec->br_startoff >= end) {
3680 		irec->br_blockcount = 0;
3681 		return;
3682 	}
3683 
3684 	if (irec->br_startoff < bno) {
3685 		distance = bno - irec->br_startoff;
3686 		if (isnullstartblock(irec->br_startblock))
3687 			irec->br_startblock = DELAYSTARTBLOCK;
3688 		if (irec->br_startblock != DELAYSTARTBLOCK &&
3689 		    irec->br_startblock != HOLESTARTBLOCK)
3690 			irec->br_startblock += distance;
3691 		irec->br_startoff += distance;
3692 		irec->br_blockcount -= distance;
3693 	}
3694 
3695 	if (end < irec->br_startoff + irec->br_blockcount) {
3696 		distance = irec->br_startoff + irec->br_blockcount - end;
3697 		irec->br_blockcount -= distance;
3698 	}
3699 }
3700 
3701 /*
3702  * Trim the returned map to the required bounds
3703  */
3704 STATIC void
3705 xfs_bmapi_trim_map(
3706 	struct xfs_bmbt_irec	*mval,
3707 	struct xfs_bmbt_irec	*got,
3708 	xfs_fileoff_t		*bno,
3709 	xfs_filblks_t		len,
3710 	xfs_fileoff_t		obno,
3711 	xfs_fileoff_t		end,
3712 	int			n,
3713 	int			flags)
3714 {
3715 	if ((flags & XFS_BMAPI_ENTIRE) ||
3716 	    got->br_startoff + got->br_blockcount <= obno) {
3717 		*mval = *got;
3718 		if (isnullstartblock(got->br_startblock))
3719 			mval->br_startblock = DELAYSTARTBLOCK;
3720 		return;
3721 	}
3722 
3723 	if (obno > *bno)
3724 		*bno = obno;
3725 	ASSERT((*bno >= obno) || (n == 0));
3726 	ASSERT(*bno < end);
3727 	mval->br_startoff = *bno;
3728 	if (isnullstartblock(got->br_startblock))
3729 		mval->br_startblock = DELAYSTARTBLOCK;
3730 	else
3731 		mval->br_startblock = got->br_startblock +
3732 					(*bno - got->br_startoff);
3733 	/*
3734 	 * Return the minimum of what we got and what we asked for for
3735 	 * the length.  We can use the len variable here because it is
3736 	 * modified below and we could have been there before coming
3737 	 * here if the first part of the allocation didn't overlap what
3738 	 * was asked for.
3739 	 */
3740 	mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3741 			got->br_blockcount - (*bno - got->br_startoff));
3742 	mval->br_state = got->br_state;
3743 	ASSERT(mval->br_blockcount <= len);
3744 	return;
3745 }
3746 
3747 /*
3748  * Update and validate the extent map to return
3749  */
3750 STATIC void
3751 xfs_bmapi_update_map(
3752 	struct xfs_bmbt_irec	**map,
3753 	xfs_fileoff_t		*bno,
3754 	xfs_filblks_t		*len,
3755 	xfs_fileoff_t		obno,
3756 	xfs_fileoff_t		end,
3757 	int			*n,
3758 	int			flags)
3759 {
3760 	xfs_bmbt_irec_t	*mval = *map;
3761 
3762 	ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3763 	       ((mval->br_startoff + mval->br_blockcount) <= end));
3764 	ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3765 	       (mval->br_startoff < obno));
3766 
3767 	*bno = mval->br_startoff + mval->br_blockcount;
3768 	*len = end - *bno;
3769 	if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3770 		/* update previous map with new information */
3771 		ASSERT(mval->br_startblock == mval[-1].br_startblock);
3772 		ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3773 		ASSERT(mval->br_state == mval[-1].br_state);
3774 		mval[-1].br_blockcount = mval->br_blockcount;
3775 		mval[-1].br_state = mval->br_state;
3776 	} else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3777 		   mval[-1].br_startblock != DELAYSTARTBLOCK &&
3778 		   mval[-1].br_startblock != HOLESTARTBLOCK &&
3779 		   mval->br_startblock == mval[-1].br_startblock +
3780 					  mval[-1].br_blockcount &&
3781 		   mval[-1].br_state == mval->br_state) {
3782 		ASSERT(mval->br_startoff ==
3783 		       mval[-1].br_startoff + mval[-1].br_blockcount);
3784 		mval[-1].br_blockcount += mval->br_blockcount;
3785 	} else if (*n > 0 &&
3786 		   mval->br_startblock == DELAYSTARTBLOCK &&
3787 		   mval[-1].br_startblock == DELAYSTARTBLOCK &&
3788 		   mval->br_startoff ==
3789 		   mval[-1].br_startoff + mval[-1].br_blockcount) {
3790 		mval[-1].br_blockcount += mval->br_blockcount;
3791 		mval[-1].br_state = mval->br_state;
3792 	} else if (!((*n == 0) &&
3793 		     ((mval->br_startoff + mval->br_blockcount) <=
3794 		      obno))) {
3795 		mval++;
3796 		(*n)++;
3797 	}
3798 	*map = mval;
3799 }
3800 
3801 /*
3802  * Map file blocks to filesystem blocks without allocation.
3803  */
3804 int
3805 xfs_bmapi_read(
3806 	struct xfs_inode	*ip,
3807 	xfs_fileoff_t		bno,
3808 	xfs_filblks_t		len,
3809 	struct xfs_bmbt_irec	*mval,
3810 	int			*nmap,
3811 	int			flags)
3812 {
3813 	struct xfs_mount	*mp = ip->i_mount;
3814 	struct xfs_ifork	*ifp;
3815 	struct xfs_bmbt_irec	got;
3816 	xfs_fileoff_t		obno;
3817 	xfs_fileoff_t		end;
3818 	struct xfs_iext_cursor	icur;
3819 	int			error;
3820 	bool			eof = false;
3821 	int			n = 0;
3822 	int			whichfork = xfs_bmapi_whichfork(flags);
3823 
3824 	ASSERT(*nmap >= 1);
3825 	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
3826 			   XFS_BMAPI_COWFORK)));
3827 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
3828 
3829 	if (unlikely(XFS_TEST_ERROR(
3830 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3831 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
3832 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
3833 		XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
3834 		return -EFSCORRUPTED;
3835 	}
3836 
3837 	if (XFS_FORCED_SHUTDOWN(mp))
3838 		return -EIO;
3839 
3840 	XFS_STATS_INC(mp, xs_blk_mapr);
3841 
3842 	ifp = XFS_IFORK_PTR(ip, whichfork);
3843 
3844 	/* No CoW fork?  Return a hole. */
3845 	if (whichfork == XFS_COW_FORK && !ifp) {
3846 		mval->br_startoff = bno;
3847 		mval->br_startblock = HOLESTARTBLOCK;
3848 		mval->br_blockcount = len;
3849 		mval->br_state = XFS_EXT_NORM;
3850 		*nmap = 1;
3851 		return 0;
3852 	}
3853 
3854 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
3855 		error = xfs_iread_extents(NULL, ip, whichfork);
3856 		if (error)
3857 			return error;
3858 	}
3859 
3860 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got))
3861 		eof = true;
3862 	end = bno + len;
3863 	obno = bno;
3864 
3865 	while (bno < end && n < *nmap) {
3866 		/* Reading past eof, act as though there's a hole up to end. */
3867 		if (eof)
3868 			got.br_startoff = end;
3869 		if (got.br_startoff > bno) {
3870 			/* Reading in a hole.  */
3871 			mval->br_startoff = bno;
3872 			mval->br_startblock = HOLESTARTBLOCK;
3873 			mval->br_blockcount =
3874 				XFS_FILBLKS_MIN(len, got.br_startoff - bno);
3875 			mval->br_state = XFS_EXT_NORM;
3876 			bno += mval->br_blockcount;
3877 			len -= mval->br_blockcount;
3878 			mval++;
3879 			n++;
3880 			continue;
3881 		}
3882 
3883 		/* set up the extent map to return. */
3884 		xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
3885 		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
3886 
3887 		/* If we're done, stop now. */
3888 		if (bno >= end || n >= *nmap)
3889 			break;
3890 
3891 		/* Else go on to the next record. */
3892 		if (!xfs_iext_next_extent(ifp, &icur, &got))
3893 			eof = true;
3894 	}
3895 	*nmap = n;
3896 	return 0;
3897 }
3898 
3899 /*
3900  * Add a delayed allocation extent to an inode. Blocks are reserved from the
3901  * global pool and the extent inserted into the inode in-core extent tree.
3902  *
3903  * On entry, got refers to the first extent beyond the offset of the extent to
3904  * allocate or eof is specified if no such extent exists. On return, got refers
3905  * to the extent record that was inserted to the inode fork.
3906  *
3907  * Note that the allocated extent may have been merged with contiguous extents
3908  * during insertion into the inode fork. Thus, got does not reflect the current
3909  * state of the inode fork on return. If necessary, the caller can use lastx to
3910  * look up the updated record in the inode fork.
3911  */
3912 int
3913 xfs_bmapi_reserve_delalloc(
3914 	struct xfs_inode	*ip,
3915 	int			whichfork,
3916 	xfs_fileoff_t		off,
3917 	xfs_filblks_t		len,
3918 	xfs_filblks_t		prealloc,
3919 	struct xfs_bmbt_irec	*got,
3920 	struct xfs_iext_cursor	*icur,
3921 	int			eof)
3922 {
3923 	struct xfs_mount	*mp = ip->i_mount;
3924 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
3925 	xfs_extlen_t		alen;
3926 	xfs_extlen_t		indlen;
3927 	int			error;
3928 	xfs_fileoff_t		aoff = off;
3929 
3930 	/*
3931 	 * Cap the alloc length. Keep track of prealloc so we know whether to
3932 	 * tag the inode before we return.
3933 	 */
3934 	alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN);
3935 	if (!eof)
3936 		alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
3937 	if (prealloc && alen >= len)
3938 		prealloc = alen - len;
3939 
3940 	/* Figure out the extent size, adjust alen */
3941 	if (whichfork == XFS_COW_FORK) {
3942 		struct xfs_bmbt_irec	prev;
3943 		xfs_extlen_t		extsz = xfs_get_cowextsz_hint(ip);
3944 
3945 		if (!xfs_iext_peek_prev_extent(ifp, icur, &prev))
3946 			prev.br_startoff = NULLFILEOFF;
3947 
3948 		error = xfs_bmap_extsize_align(mp, got, &prev, extsz, 0, eof,
3949 					       1, 0, &aoff, &alen);
3950 		ASSERT(!error);
3951 	}
3952 
3953 	/*
3954 	 * Make a transaction-less quota reservation for delayed allocation
3955 	 * blocks.  This number gets adjusted later.  We return if we haven't
3956 	 * allocated blocks already inside this loop.
3957 	 */
3958 	error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
3959 						XFS_QMOPT_RES_REGBLKS);
3960 	if (error)
3961 		return error;
3962 
3963 	/*
3964 	 * Split changing sb for alen and indlen since they could be coming
3965 	 * from different places.
3966 	 */
3967 	indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
3968 	ASSERT(indlen > 0);
3969 
3970 	error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
3971 	if (error)
3972 		goto out_unreserve_quota;
3973 
3974 	error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
3975 	if (error)
3976 		goto out_unreserve_blocks;
3977 
3978 
3979 	ip->i_delayed_blks += alen;
3980 	xfs_mod_delalloc(ip->i_mount, alen + indlen);
3981 
3982 	got->br_startoff = aoff;
3983 	got->br_startblock = nullstartblock(indlen);
3984 	got->br_blockcount = alen;
3985 	got->br_state = XFS_EXT_NORM;
3986 
3987 	xfs_bmap_add_extent_hole_delay(ip, whichfork, icur, got);
3988 
3989 	/*
3990 	 * Tag the inode if blocks were preallocated. Note that COW fork
3991 	 * preallocation can occur at the start or end of the extent, even when
3992 	 * prealloc == 0, so we must also check the aligned offset and length.
3993 	 */
3994 	if (whichfork == XFS_DATA_FORK && prealloc)
3995 		xfs_inode_set_eofblocks_tag(ip);
3996 	if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len))
3997 		xfs_inode_set_cowblocks_tag(ip);
3998 
3999 	return 0;
4000 
4001 out_unreserve_blocks:
4002 	xfs_mod_fdblocks(mp, alen, false);
4003 out_unreserve_quota:
4004 	if (XFS_IS_QUOTA_ON(mp))
4005 		xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0,
4006 						XFS_QMOPT_RES_REGBLKS);
4007 	return error;
4008 }
4009 
4010 static int
4011 xfs_bmapi_allocate(
4012 	struct xfs_bmalloca	*bma)
4013 {
4014 	struct xfs_mount	*mp = bma->ip->i_mount;
4015 	int			whichfork = xfs_bmapi_whichfork(bma->flags);
4016 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4017 	int			tmp_logflags = 0;
4018 	int			error;
4019 
4020 	ASSERT(bma->length > 0);
4021 
4022 	/*
4023 	 * For the wasdelay case, we could also just allocate the stuff asked
4024 	 * for in this bmap call but that wouldn't be as good.
4025 	 */
4026 	if (bma->wasdel) {
4027 		bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4028 		bma->offset = bma->got.br_startoff;
4029 		xfs_iext_peek_prev_extent(ifp, &bma->icur, &bma->prev);
4030 	} else {
4031 		bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4032 		if (!bma->eof)
4033 			bma->length = XFS_FILBLKS_MIN(bma->length,
4034 					bma->got.br_startoff - bma->offset);
4035 	}
4036 
4037 	/*
4038 	 * Set the data type being allocated. For the data fork, the first data
4039 	 * in the file is treated differently to all other allocations. For the
4040 	 * attribute fork, we only need to ensure the allocated range is not on
4041 	 * the busy list.
4042 	 */
4043 	if (!(bma->flags & XFS_BMAPI_METADATA)) {
4044 		bma->datatype = XFS_ALLOC_NOBUSY;
4045 		if (whichfork == XFS_DATA_FORK) {
4046 			if (bma->offset == 0)
4047 				bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA;
4048 			else
4049 				bma->datatype |= XFS_ALLOC_USERDATA;
4050 		}
4051 		if (bma->flags & XFS_BMAPI_ZERO)
4052 			bma->datatype |= XFS_ALLOC_USERDATA_ZERO;
4053 	}
4054 
4055 	bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4056 
4057 	/*
4058 	 * Only want to do the alignment at the eof if it is userdata and
4059 	 * allocation length is larger than a stripe unit.
4060 	 */
4061 	if (mp->m_dalign && bma->length >= mp->m_dalign &&
4062 	    !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4063 		error = xfs_bmap_isaeof(bma, whichfork);
4064 		if (error)
4065 			return error;
4066 	}
4067 
4068 	error = xfs_bmap_alloc(bma);
4069 	if (error)
4070 		return error;
4071 
4072 	if (bma->blkno == NULLFSBLOCK)
4073 		return 0;
4074 	if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur)
4075 		bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4076 	/*
4077 	 * Bump the number of extents we've allocated
4078 	 * in this call.
4079 	 */
4080 	bma->nallocs++;
4081 
4082 	if (bma->cur)
4083 		bma->cur->bc_private.b.flags =
4084 			bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4085 
4086 	bma->got.br_startoff = bma->offset;
4087 	bma->got.br_startblock = bma->blkno;
4088 	bma->got.br_blockcount = bma->length;
4089 	bma->got.br_state = XFS_EXT_NORM;
4090 
4091 	/*
4092 	 * In the data fork, a wasdelay extent has been initialized, so
4093 	 * shouldn't be flagged as unwritten.
4094 	 *
4095 	 * For the cow fork, however, we convert delalloc reservations
4096 	 * (extents allocated for speculative preallocation) to
4097 	 * allocated unwritten extents, and only convert the unwritten
4098 	 * extents to real extents when we're about to write the data.
4099 	 */
4100 	if ((!bma->wasdel || (bma->flags & XFS_BMAPI_COWFORK)) &&
4101 	    (bma->flags & XFS_BMAPI_PREALLOC))
4102 		bma->got.br_state = XFS_EXT_UNWRITTEN;
4103 
4104 	if (bma->wasdel)
4105 		error = xfs_bmap_add_extent_delay_real(bma, whichfork);
4106 	else
4107 		error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip,
4108 				whichfork, &bma->icur, &bma->cur, &bma->got,
4109 				&bma->logflags, bma->flags);
4110 
4111 	bma->logflags |= tmp_logflags;
4112 	if (error)
4113 		return error;
4114 
4115 	/*
4116 	 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4117 	 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4118 	 * the neighbouring ones.
4119 	 */
4120 	xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4121 
4122 	ASSERT(bma->got.br_startoff <= bma->offset);
4123 	ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4124 	       bma->offset + bma->length);
4125 	ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4126 	       bma->got.br_state == XFS_EXT_UNWRITTEN);
4127 	return 0;
4128 }
4129 
4130 STATIC int
4131 xfs_bmapi_convert_unwritten(
4132 	struct xfs_bmalloca	*bma,
4133 	struct xfs_bmbt_irec	*mval,
4134 	xfs_filblks_t		len,
4135 	int			flags)
4136 {
4137 	int			whichfork = xfs_bmapi_whichfork(flags);
4138 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4139 	int			tmp_logflags = 0;
4140 	int			error;
4141 
4142 	/* check if we need to do unwritten->real conversion */
4143 	if (mval->br_state == XFS_EXT_UNWRITTEN &&
4144 	    (flags & XFS_BMAPI_PREALLOC))
4145 		return 0;
4146 
4147 	/* check if we need to do real->unwritten conversion */
4148 	if (mval->br_state == XFS_EXT_NORM &&
4149 	    (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4150 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4151 		return 0;
4152 
4153 	/*
4154 	 * Modify (by adding) the state flag, if writing.
4155 	 */
4156 	ASSERT(mval->br_blockcount <= len);
4157 	if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4158 		bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4159 					bma->ip, whichfork);
4160 	}
4161 	mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4162 				? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4163 
4164 	/*
4165 	 * Before insertion into the bmbt, zero the range being converted
4166 	 * if required.
4167 	 */
4168 	if (flags & XFS_BMAPI_ZERO) {
4169 		error = xfs_zero_extent(bma->ip, mval->br_startblock,
4170 					mval->br_blockcount);
4171 		if (error)
4172 			return error;
4173 	}
4174 
4175 	error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork,
4176 			&bma->icur, &bma->cur, mval, &tmp_logflags);
4177 	/*
4178 	 * Log the inode core unconditionally in the unwritten extent conversion
4179 	 * path because the conversion might not have done so (e.g., if the
4180 	 * extent count hasn't changed). We need to make sure the inode is dirty
4181 	 * in the transaction for the sake of fsync(), even if nothing has
4182 	 * changed, because fsync() will not force the log for this transaction
4183 	 * unless it sees the inode pinned.
4184 	 *
4185 	 * Note: If we're only converting cow fork extents, there aren't
4186 	 * any on-disk updates to make, so we don't need to log anything.
4187 	 */
4188 	if (whichfork != XFS_COW_FORK)
4189 		bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4190 	if (error)
4191 		return error;
4192 
4193 	/*
4194 	 * Update our extent pointer, given that
4195 	 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4196 	 * of the neighbouring ones.
4197 	 */
4198 	xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4199 
4200 	/*
4201 	 * We may have combined previously unwritten space with written space,
4202 	 * so generate another request.
4203 	 */
4204 	if (mval->br_blockcount < len)
4205 		return -EAGAIN;
4206 	return 0;
4207 }
4208 
4209 static inline xfs_extlen_t
4210 xfs_bmapi_minleft(
4211 	struct xfs_trans	*tp,
4212 	struct xfs_inode	*ip,
4213 	int			fork)
4214 {
4215 	if (tp && tp->t_firstblock != NULLFSBLOCK)
4216 		return 0;
4217 	if (XFS_IFORK_FORMAT(ip, fork) != XFS_DINODE_FMT_BTREE)
4218 		return 1;
4219 	return be16_to_cpu(XFS_IFORK_PTR(ip, fork)->if_broot->bb_level) + 1;
4220 }
4221 
4222 /*
4223  * Log whatever the flags say, even if error.  Otherwise we might miss detecting
4224  * a case where the data is changed, there's an error, and it's not logged so we
4225  * don't shutdown when we should.  Don't bother logging extents/btree changes if
4226  * we converted to the other format.
4227  */
4228 static void
4229 xfs_bmapi_finish(
4230 	struct xfs_bmalloca	*bma,
4231 	int			whichfork,
4232 	int			error)
4233 {
4234 	if ((bma->logflags & xfs_ilog_fext(whichfork)) &&
4235 	    XFS_IFORK_FORMAT(bma->ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4236 		bma->logflags &= ~xfs_ilog_fext(whichfork);
4237 	else if ((bma->logflags & xfs_ilog_fbroot(whichfork)) &&
4238 		 XFS_IFORK_FORMAT(bma->ip, whichfork) != XFS_DINODE_FMT_BTREE)
4239 		bma->logflags &= ~xfs_ilog_fbroot(whichfork);
4240 
4241 	if (bma->logflags)
4242 		xfs_trans_log_inode(bma->tp, bma->ip, bma->logflags);
4243 	if (bma->cur)
4244 		xfs_btree_del_cursor(bma->cur, error);
4245 }
4246 
4247 /*
4248  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4249  * extent state if necessary.  Details behaviour is controlled by the flags
4250  * parameter.  Only allocates blocks from a single allocation group, to avoid
4251  * locking problems.
4252  */
4253 int
4254 xfs_bmapi_write(
4255 	struct xfs_trans	*tp,		/* transaction pointer */
4256 	struct xfs_inode	*ip,		/* incore inode */
4257 	xfs_fileoff_t		bno,		/* starting file offs. mapped */
4258 	xfs_filblks_t		len,		/* length to map in file */
4259 	int			flags,		/* XFS_BMAPI_... */
4260 	xfs_extlen_t		total,		/* total blocks needed */
4261 	struct xfs_bmbt_irec	*mval,		/* output: map values */
4262 	int			*nmap)		/* i/o: mval size/count */
4263 {
4264 	struct xfs_bmalloca	bma = {
4265 		.tp		= tp,
4266 		.ip		= ip,
4267 		.total		= total,
4268 	};
4269 	struct xfs_mount	*mp = ip->i_mount;
4270 	struct xfs_ifork	*ifp;
4271 	xfs_fileoff_t		end;		/* end of mapped file region */
4272 	bool			eof = false;	/* after the end of extents */
4273 	int			error;		/* error return */
4274 	int			n;		/* current extent index */
4275 	xfs_fileoff_t		obno;		/* old block number (offset) */
4276 	int			whichfork;	/* data or attr fork */
4277 
4278 #ifdef DEBUG
4279 	xfs_fileoff_t		orig_bno;	/* original block number value */
4280 	int			orig_flags;	/* original flags arg value */
4281 	xfs_filblks_t		orig_len;	/* original value of len arg */
4282 	struct xfs_bmbt_irec	*orig_mval;	/* original value of mval */
4283 	int			orig_nmap;	/* original value of *nmap */
4284 
4285 	orig_bno = bno;
4286 	orig_len = len;
4287 	orig_flags = flags;
4288 	orig_mval = mval;
4289 	orig_nmap = *nmap;
4290 #endif
4291 	whichfork = xfs_bmapi_whichfork(flags);
4292 
4293 	ASSERT(*nmap >= 1);
4294 	ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4295 	ASSERT(tp != NULL);
4296 	ASSERT(len > 0);
4297 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4298 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4299 	ASSERT(!(flags & XFS_BMAPI_REMAP));
4300 
4301 	/* zeroing is for currently only for data extents, not metadata */
4302 	ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4303 			(XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4304 	/*
4305 	 * we can allocate unwritten extents or pre-zero allocated blocks,
4306 	 * but it makes no sense to do both at once. This would result in
4307 	 * zeroing the unwritten extent twice, but it still being an
4308 	 * unwritten extent....
4309 	 */
4310 	ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4311 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4312 
4313 	if (unlikely(XFS_TEST_ERROR(
4314 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4315 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4316 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
4317 		XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4318 		return -EFSCORRUPTED;
4319 	}
4320 
4321 	if (XFS_FORCED_SHUTDOWN(mp))
4322 		return -EIO;
4323 
4324 	ifp = XFS_IFORK_PTR(ip, whichfork);
4325 
4326 	XFS_STATS_INC(mp, xs_blk_mapw);
4327 
4328 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4329 		error = xfs_iread_extents(tp, ip, whichfork);
4330 		if (error)
4331 			goto error0;
4332 	}
4333 
4334 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.icur, &bma.got))
4335 		eof = true;
4336 	if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4337 		bma.prev.br_startoff = NULLFILEOFF;
4338 	bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4339 
4340 	n = 0;
4341 	end = bno + len;
4342 	obno = bno;
4343 	while (bno < end && n < *nmap) {
4344 		bool			need_alloc = false, wasdelay = false;
4345 
4346 		/* in hole or beyond EOF? */
4347 		if (eof || bma.got.br_startoff > bno) {
4348 			/*
4349 			 * CoW fork conversions should /never/ hit EOF or
4350 			 * holes.  There should always be something for us
4351 			 * to work on.
4352 			 */
4353 			ASSERT(!((flags & XFS_BMAPI_CONVERT) &&
4354 			         (flags & XFS_BMAPI_COWFORK)));
4355 
4356 			need_alloc = true;
4357 		} else if (isnullstartblock(bma.got.br_startblock)) {
4358 			wasdelay = true;
4359 		}
4360 
4361 		/*
4362 		 * First, deal with the hole before the allocated space
4363 		 * that we found, if any.
4364 		 */
4365 		if (need_alloc || wasdelay) {
4366 			bma.eof = eof;
4367 			bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4368 			bma.wasdel = wasdelay;
4369 			bma.offset = bno;
4370 			bma.flags = flags;
4371 
4372 			/*
4373 			 * There's a 32/64 bit type mismatch between the
4374 			 * allocation length request (which can be 64 bits in
4375 			 * length) and the bma length request, which is
4376 			 * xfs_extlen_t and therefore 32 bits. Hence we have to
4377 			 * check for 32-bit overflows and handle them here.
4378 			 */
4379 			if (len > (xfs_filblks_t)MAXEXTLEN)
4380 				bma.length = MAXEXTLEN;
4381 			else
4382 				bma.length = len;
4383 
4384 			ASSERT(len > 0);
4385 			ASSERT(bma.length > 0);
4386 			error = xfs_bmapi_allocate(&bma);
4387 			if (error)
4388 				goto error0;
4389 			if (bma.blkno == NULLFSBLOCK)
4390 				break;
4391 
4392 			/*
4393 			 * If this is a CoW allocation, record the data in
4394 			 * the refcount btree for orphan recovery.
4395 			 */
4396 			if (whichfork == XFS_COW_FORK) {
4397 				error = xfs_refcount_alloc_cow_extent(tp,
4398 						bma.blkno, bma.length);
4399 				if (error)
4400 					goto error0;
4401 			}
4402 		}
4403 
4404 		/* Deal with the allocated space we found.  */
4405 		xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4406 							end, n, flags);
4407 
4408 		/* Execute unwritten extent conversion if necessary */
4409 		error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4410 		if (error == -EAGAIN)
4411 			continue;
4412 		if (error)
4413 			goto error0;
4414 
4415 		/* update the extent map to return */
4416 		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4417 
4418 		/*
4419 		 * If we're done, stop now.  Stop when we've allocated
4420 		 * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4421 		 * the transaction may get too big.
4422 		 */
4423 		if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4424 			break;
4425 
4426 		/* Else go on to the next record. */
4427 		bma.prev = bma.got;
4428 		if (!xfs_iext_next_extent(ifp, &bma.icur, &bma.got))
4429 			eof = true;
4430 	}
4431 	*nmap = n;
4432 
4433 	error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4434 			whichfork);
4435 	if (error)
4436 		goto error0;
4437 
4438 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4439 	       XFS_IFORK_NEXTENTS(ip, whichfork) >
4440 		XFS_IFORK_MAXEXT(ip, whichfork));
4441 	xfs_bmapi_finish(&bma, whichfork, 0);
4442 	xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4443 		orig_nmap, *nmap);
4444 	return 0;
4445 error0:
4446 	xfs_bmapi_finish(&bma, whichfork, error);
4447 	return error;
4448 }
4449 
4450 /*
4451  * Convert an existing delalloc extent to real blocks based on file offset. This
4452  * attempts to allocate the entire delalloc extent and may require multiple
4453  * invocations to allocate the target offset if a large enough physical extent
4454  * is not available.
4455  */
4456 int
4457 xfs_bmapi_convert_delalloc(
4458 	struct xfs_inode	*ip,
4459 	int			whichfork,
4460 	xfs_fileoff_t		offset_fsb,
4461 	struct xfs_bmbt_irec	*imap,
4462 	unsigned int		*seq)
4463 {
4464 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
4465 	struct xfs_mount	*mp = ip->i_mount;
4466 	struct xfs_bmalloca	bma = { NULL };
4467 	struct xfs_trans	*tp;
4468 	int			error;
4469 
4470 	/*
4471 	 * Space for the extent and indirect blocks was reserved when the
4472 	 * delalloc extent was created so there's no need to do so here.
4473 	 */
4474 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0,
4475 				XFS_TRANS_RESERVE, &tp);
4476 	if (error)
4477 		return error;
4478 
4479 	xfs_ilock(ip, XFS_ILOCK_EXCL);
4480 	xfs_trans_ijoin(tp, ip, 0);
4481 
4482 	if (!xfs_iext_lookup_extent(ip, ifp, offset_fsb, &bma.icur, &bma.got) ||
4483 	    bma.got.br_startoff > offset_fsb) {
4484 		/*
4485 		 * No extent found in the range we are trying to convert.  This
4486 		 * should only happen for the COW fork, where another thread
4487 		 * might have moved the extent to the data fork in the meantime.
4488 		 */
4489 		WARN_ON_ONCE(whichfork != XFS_COW_FORK);
4490 		error = -EAGAIN;
4491 		goto out_trans_cancel;
4492 	}
4493 
4494 	/*
4495 	 * If we find a real extent here we raced with another thread converting
4496 	 * the extent.  Just return the real extent at this offset.
4497 	 */
4498 	if (!isnullstartblock(bma.got.br_startblock)) {
4499 		*imap = bma.got;
4500 		*seq = READ_ONCE(ifp->if_seq);
4501 		goto out_trans_cancel;
4502 	}
4503 
4504 	bma.tp = tp;
4505 	bma.ip = ip;
4506 	bma.wasdel = true;
4507 	bma.offset = bma.got.br_startoff;
4508 	bma.length = max_t(xfs_filblks_t, bma.got.br_blockcount, MAXEXTLEN);
4509 	bma.total = XFS_EXTENTADD_SPACE_RES(ip->i_mount, XFS_DATA_FORK);
4510 	bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4511 	if (whichfork == XFS_COW_FORK)
4512 		bma.flags = XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC;
4513 
4514 	if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4515 		bma.prev.br_startoff = NULLFILEOFF;
4516 
4517 	error = xfs_bmapi_allocate(&bma);
4518 	if (error)
4519 		goto out_finish;
4520 
4521 	error = -ENOSPC;
4522 	if (WARN_ON_ONCE(bma.blkno == NULLFSBLOCK))
4523 		goto out_finish;
4524 	error = -EFSCORRUPTED;
4525 	if (WARN_ON_ONCE(!bma.got.br_startblock && !XFS_IS_REALTIME_INODE(ip)))
4526 		goto out_finish;
4527 
4528 	XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, bma.length));
4529 	XFS_STATS_INC(mp, xs_xstrat_quick);
4530 
4531 	ASSERT(!isnullstartblock(bma.got.br_startblock));
4532 	*imap = bma.got;
4533 	*seq = READ_ONCE(ifp->if_seq);
4534 
4535 	if (whichfork == XFS_COW_FORK) {
4536 		error = xfs_refcount_alloc_cow_extent(tp, bma.blkno,
4537 				bma.length);
4538 		if (error)
4539 			goto out_finish;
4540 	}
4541 
4542 	error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4543 			whichfork);
4544 	if (error)
4545 		goto out_finish;
4546 
4547 	xfs_bmapi_finish(&bma, whichfork, 0);
4548 	error = xfs_trans_commit(tp);
4549 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
4550 	return error;
4551 
4552 out_finish:
4553 	xfs_bmapi_finish(&bma, whichfork, error);
4554 out_trans_cancel:
4555 	xfs_trans_cancel(tp);
4556 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
4557 	return error;
4558 }
4559 
4560 int
4561 xfs_bmapi_remap(
4562 	struct xfs_trans	*tp,
4563 	struct xfs_inode	*ip,
4564 	xfs_fileoff_t		bno,
4565 	xfs_filblks_t		len,
4566 	xfs_fsblock_t		startblock,
4567 	int			flags)
4568 {
4569 	struct xfs_mount	*mp = ip->i_mount;
4570 	struct xfs_ifork	*ifp;
4571 	struct xfs_btree_cur	*cur = NULL;
4572 	struct xfs_bmbt_irec	got;
4573 	struct xfs_iext_cursor	icur;
4574 	int			whichfork = xfs_bmapi_whichfork(flags);
4575 	int			logflags = 0, error;
4576 
4577 	ifp = XFS_IFORK_PTR(ip, whichfork);
4578 	ASSERT(len > 0);
4579 	ASSERT(len <= (xfs_filblks_t)MAXEXTLEN);
4580 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4581 	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC |
4582 			   XFS_BMAPI_NORMAP)));
4583 	ASSERT((flags & (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC)) !=
4584 			(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC));
4585 
4586 	if (unlikely(XFS_TEST_ERROR(
4587 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4588 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4589 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
4590 		XFS_ERROR_REPORT("xfs_bmapi_remap", XFS_ERRLEVEL_LOW, mp);
4591 		return -EFSCORRUPTED;
4592 	}
4593 
4594 	if (XFS_FORCED_SHUTDOWN(mp))
4595 		return -EIO;
4596 
4597 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4598 		error = xfs_iread_extents(tp, ip, whichfork);
4599 		if (error)
4600 			return error;
4601 	}
4602 
4603 	if (xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
4604 		/* make sure we only reflink into a hole. */
4605 		ASSERT(got.br_startoff > bno);
4606 		ASSERT(got.br_startoff - bno >= len);
4607 	}
4608 
4609 	ip->i_d.di_nblocks += len;
4610 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4611 
4612 	if (ifp->if_flags & XFS_IFBROOT) {
4613 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
4614 		cur->bc_private.b.flags = 0;
4615 	}
4616 
4617 	got.br_startoff = bno;
4618 	got.br_startblock = startblock;
4619 	got.br_blockcount = len;
4620 	if (flags & XFS_BMAPI_PREALLOC)
4621 		got.br_state = XFS_EXT_UNWRITTEN;
4622 	else
4623 		got.br_state = XFS_EXT_NORM;
4624 
4625 	error = xfs_bmap_add_extent_hole_real(tp, ip, whichfork, &icur,
4626 			&cur, &got, &logflags, flags);
4627 	if (error)
4628 		goto error0;
4629 
4630 	error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags, whichfork);
4631 
4632 error0:
4633 	if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS)
4634 		logflags &= ~XFS_ILOG_DEXT;
4635 	else if (ip->i_d.di_format != XFS_DINODE_FMT_BTREE)
4636 		logflags &= ~XFS_ILOG_DBROOT;
4637 
4638 	if (logflags)
4639 		xfs_trans_log_inode(tp, ip, logflags);
4640 	if (cur)
4641 		xfs_btree_del_cursor(cur, error);
4642 	return error;
4643 }
4644 
4645 /*
4646  * When a delalloc extent is split (e.g., due to a hole punch), the original
4647  * indlen reservation must be shared across the two new extents that are left
4648  * behind.
4649  *
4650  * Given the original reservation and the worst case indlen for the two new
4651  * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4652  * reservation fairly across the two new extents. If necessary, steal available
4653  * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4654  * ores == 1). The number of stolen blocks is returned. The availability and
4655  * subsequent accounting of stolen blocks is the responsibility of the caller.
4656  */
4657 static xfs_filblks_t
4658 xfs_bmap_split_indlen(
4659 	xfs_filblks_t			ores,		/* original res. */
4660 	xfs_filblks_t			*indlen1,	/* ext1 worst indlen */
4661 	xfs_filblks_t			*indlen2,	/* ext2 worst indlen */
4662 	xfs_filblks_t			avail)		/* stealable blocks */
4663 {
4664 	xfs_filblks_t			len1 = *indlen1;
4665 	xfs_filblks_t			len2 = *indlen2;
4666 	xfs_filblks_t			nres = len1 + len2; /* new total res. */
4667 	xfs_filblks_t			stolen = 0;
4668 	xfs_filblks_t			resfactor;
4669 
4670 	/*
4671 	 * Steal as many blocks as we can to try and satisfy the worst case
4672 	 * indlen for both new extents.
4673 	 */
4674 	if (ores < nres && avail)
4675 		stolen = XFS_FILBLKS_MIN(nres - ores, avail);
4676 	ores += stolen;
4677 
4678 	 /* nothing else to do if we've satisfied the new reservation */
4679 	if (ores >= nres)
4680 		return stolen;
4681 
4682 	/*
4683 	 * We can't meet the total required reservation for the two extents.
4684 	 * Calculate the percent of the overall shortage between both extents
4685 	 * and apply this percentage to each of the requested indlen values.
4686 	 * This distributes the shortage fairly and reduces the chances that one
4687 	 * of the two extents is left with nothing when extents are repeatedly
4688 	 * split.
4689 	 */
4690 	resfactor = (ores * 100);
4691 	do_div(resfactor, nres);
4692 	len1 *= resfactor;
4693 	do_div(len1, 100);
4694 	len2 *= resfactor;
4695 	do_div(len2, 100);
4696 	ASSERT(len1 + len2 <= ores);
4697 	ASSERT(len1 < *indlen1 && len2 < *indlen2);
4698 
4699 	/*
4700 	 * Hand out the remainder to each extent. If one of the two reservations
4701 	 * is zero, we want to make sure that one gets a block first. The loop
4702 	 * below starts with len1, so hand len2 a block right off the bat if it
4703 	 * is zero.
4704 	 */
4705 	ores -= (len1 + len2);
4706 	ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores);
4707 	if (ores && !len2 && *indlen2) {
4708 		len2++;
4709 		ores--;
4710 	}
4711 	while (ores) {
4712 		if (len1 < *indlen1) {
4713 			len1++;
4714 			ores--;
4715 		}
4716 		if (!ores)
4717 			break;
4718 		if (len2 < *indlen2) {
4719 			len2++;
4720 			ores--;
4721 		}
4722 	}
4723 
4724 	*indlen1 = len1;
4725 	*indlen2 = len2;
4726 
4727 	return stolen;
4728 }
4729 
4730 int
4731 xfs_bmap_del_extent_delay(
4732 	struct xfs_inode	*ip,
4733 	int			whichfork,
4734 	struct xfs_iext_cursor	*icur,
4735 	struct xfs_bmbt_irec	*got,
4736 	struct xfs_bmbt_irec	*del)
4737 {
4738 	struct xfs_mount	*mp = ip->i_mount;
4739 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
4740 	struct xfs_bmbt_irec	new;
4741 	int64_t			da_old, da_new, da_diff = 0;
4742 	xfs_fileoff_t		del_endoff, got_endoff;
4743 	xfs_filblks_t		got_indlen, new_indlen, stolen;
4744 	int			state = xfs_bmap_fork_to_state(whichfork);
4745 	int			error = 0;
4746 	bool			isrt;
4747 
4748 	XFS_STATS_INC(mp, xs_del_exlist);
4749 
4750 	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4751 	del_endoff = del->br_startoff + del->br_blockcount;
4752 	got_endoff = got->br_startoff + got->br_blockcount;
4753 	da_old = startblockval(got->br_startblock);
4754 	da_new = 0;
4755 
4756 	ASSERT(del->br_blockcount > 0);
4757 	ASSERT(got->br_startoff <= del->br_startoff);
4758 	ASSERT(got_endoff >= del_endoff);
4759 
4760 	if (isrt) {
4761 		uint64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount);
4762 
4763 		do_div(rtexts, mp->m_sb.sb_rextsize);
4764 		xfs_mod_frextents(mp, rtexts);
4765 	}
4766 
4767 	/*
4768 	 * Update the inode delalloc counter now and wait to update the
4769 	 * sb counters as we might have to borrow some blocks for the
4770 	 * indirect block accounting.
4771 	 */
4772 	error = xfs_trans_reserve_quota_nblks(NULL, ip,
4773 			-((long)del->br_blockcount), 0,
4774 			isrt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4775 	if (error)
4776 		return error;
4777 	ip->i_delayed_blks -= del->br_blockcount;
4778 
4779 	if (got->br_startoff == del->br_startoff)
4780 		state |= BMAP_LEFT_FILLING;
4781 	if (got_endoff == del_endoff)
4782 		state |= BMAP_RIGHT_FILLING;
4783 
4784 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4785 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4786 		/*
4787 		 * Matches the whole extent.  Delete the entry.
4788 		 */
4789 		xfs_iext_remove(ip, icur, state);
4790 		xfs_iext_prev(ifp, icur);
4791 		break;
4792 	case BMAP_LEFT_FILLING:
4793 		/*
4794 		 * Deleting the first part of the extent.
4795 		 */
4796 		got->br_startoff = del_endoff;
4797 		got->br_blockcount -= del->br_blockcount;
4798 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4799 				got->br_blockcount), da_old);
4800 		got->br_startblock = nullstartblock((int)da_new);
4801 		xfs_iext_update_extent(ip, state, icur, got);
4802 		break;
4803 	case BMAP_RIGHT_FILLING:
4804 		/*
4805 		 * Deleting the last part of the extent.
4806 		 */
4807 		got->br_blockcount = got->br_blockcount - del->br_blockcount;
4808 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4809 				got->br_blockcount), da_old);
4810 		got->br_startblock = nullstartblock((int)da_new);
4811 		xfs_iext_update_extent(ip, state, icur, got);
4812 		break;
4813 	case 0:
4814 		/*
4815 		 * Deleting the middle of the extent.
4816 		 *
4817 		 * Distribute the original indlen reservation across the two new
4818 		 * extents.  Steal blocks from the deleted extent if necessary.
4819 		 * Stealing blocks simply fudges the fdblocks accounting below.
4820 		 * Warn if either of the new indlen reservations is zero as this
4821 		 * can lead to delalloc problems.
4822 		 */
4823 		got->br_blockcount = del->br_startoff - got->br_startoff;
4824 		got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount);
4825 
4826 		new.br_blockcount = got_endoff - del_endoff;
4827 		new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount);
4828 
4829 		WARN_ON_ONCE(!got_indlen || !new_indlen);
4830 		stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen,
4831 						       del->br_blockcount);
4832 
4833 		got->br_startblock = nullstartblock((int)got_indlen);
4834 
4835 		new.br_startoff = del_endoff;
4836 		new.br_state = got->br_state;
4837 		new.br_startblock = nullstartblock((int)new_indlen);
4838 
4839 		xfs_iext_update_extent(ip, state, icur, got);
4840 		xfs_iext_next(ifp, icur);
4841 		xfs_iext_insert(ip, icur, &new, state);
4842 
4843 		da_new = got_indlen + new_indlen - stolen;
4844 		del->br_blockcount -= stolen;
4845 		break;
4846 	}
4847 
4848 	ASSERT(da_old >= da_new);
4849 	da_diff = da_old - da_new;
4850 	if (!isrt)
4851 		da_diff += del->br_blockcount;
4852 	if (da_diff) {
4853 		xfs_mod_fdblocks(mp, da_diff, false);
4854 		xfs_mod_delalloc(mp, -da_diff);
4855 	}
4856 	return error;
4857 }
4858 
4859 void
4860 xfs_bmap_del_extent_cow(
4861 	struct xfs_inode	*ip,
4862 	struct xfs_iext_cursor	*icur,
4863 	struct xfs_bmbt_irec	*got,
4864 	struct xfs_bmbt_irec	*del)
4865 {
4866 	struct xfs_mount	*mp = ip->i_mount;
4867 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
4868 	struct xfs_bmbt_irec	new;
4869 	xfs_fileoff_t		del_endoff, got_endoff;
4870 	int			state = BMAP_COWFORK;
4871 
4872 	XFS_STATS_INC(mp, xs_del_exlist);
4873 
4874 	del_endoff = del->br_startoff + del->br_blockcount;
4875 	got_endoff = got->br_startoff + got->br_blockcount;
4876 
4877 	ASSERT(del->br_blockcount > 0);
4878 	ASSERT(got->br_startoff <= del->br_startoff);
4879 	ASSERT(got_endoff >= del_endoff);
4880 	ASSERT(!isnullstartblock(got->br_startblock));
4881 
4882 	if (got->br_startoff == del->br_startoff)
4883 		state |= BMAP_LEFT_FILLING;
4884 	if (got_endoff == del_endoff)
4885 		state |= BMAP_RIGHT_FILLING;
4886 
4887 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4888 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4889 		/*
4890 		 * Matches the whole extent.  Delete the entry.
4891 		 */
4892 		xfs_iext_remove(ip, icur, state);
4893 		xfs_iext_prev(ifp, icur);
4894 		break;
4895 	case BMAP_LEFT_FILLING:
4896 		/*
4897 		 * Deleting the first part of the extent.
4898 		 */
4899 		got->br_startoff = del_endoff;
4900 		got->br_blockcount -= del->br_blockcount;
4901 		got->br_startblock = del->br_startblock + del->br_blockcount;
4902 		xfs_iext_update_extent(ip, state, icur, got);
4903 		break;
4904 	case BMAP_RIGHT_FILLING:
4905 		/*
4906 		 * Deleting the last part of the extent.
4907 		 */
4908 		got->br_blockcount -= del->br_blockcount;
4909 		xfs_iext_update_extent(ip, state, icur, got);
4910 		break;
4911 	case 0:
4912 		/*
4913 		 * Deleting the middle of the extent.
4914 		 */
4915 		got->br_blockcount = del->br_startoff - got->br_startoff;
4916 
4917 		new.br_startoff = del_endoff;
4918 		new.br_blockcount = got_endoff - del_endoff;
4919 		new.br_state = got->br_state;
4920 		new.br_startblock = del->br_startblock + del->br_blockcount;
4921 
4922 		xfs_iext_update_extent(ip, state, icur, got);
4923 		xfs_iext_next(ifp, icur);
4924 		xfs_iext_insert(ip, icur, &new, state);
4925 		break;
4926 	}
4927 	ip->i_delayed_blks -= del->br_blockcount;
4928 }
4929 
4930 /*
4931  * Called by xfs_bmapi to update file extent records and the btree
4932  * after removing space.
4933  */
4934 STATIC int				/* error */
4935 xfs_bmap_del_extent_real(
4936 	xfs_inode_t		*ip,	/* incore inode pointer */
4937 	xfs_trans_t		*tp,	/* current transaction pointer */
4938 	struct xfs_iext_cursor	*icur,
4939 	xfs_btree_cur_t		*cur,	/* if null, not a btree */
4940 	xfs_bmbt_irec_t		*del,	/* data to remove from extents */
4941 	int			*logflagsp, /* inode logging flags */
4942 	int			whichfork, /* data or attr fork */
4943 	int			bflags)	/* bmapi flags */
4944 {
4945 	xfs_fsblock_t		del_endblock=0;	/* first block past del */
4946 	xfs_fileoff_t		del_endoff;	/* first offset past del */
4947 	int			do_fx;	/* free extent at end of routine */
4948 	int			error;	/* error return value */
4949 	int			flags = 0;/* inode logging flags */
4950 	struct xfs_bmbt_irec	got;	/* current extent entry */
4951 	xfs_fileoff_t		got_endoff;	/* first offset past got */
4952 	int			i;	/* temp state */
4953 	struct xfs_ifork	*ifp;	/* inode fork pointer */
4954 	xfs_mount_t		*mp;	/* mount structure */
4955 	xfs_filblks_t		nblks;	/* quota/sb block count */
4956 	xfs_bmbt_irec_t		new;	/* new record to be inserted */
4957 	/* REFERENCED */
4958 	uint			qfield;	/* quota field to update */
4959 	int			state = xfs_bmap_fork_to_state(whichfork);
4960 	struct xfs_bmbt_irec	old;
4961 
4962 	mp = ip->i_mount;
4963 	XFS_STATS_INC(mp, xs_del_exlist);
4964 
4965 	ifp = XFS_IFORK_PTR(ip, whichfork);
4966 	ASSERT(del->br_blockcount > 0);
4967 	xfs_iext_get_extent(ifp, icur, &got);
4968 	ASSERT(got.br_startoff <= del->br_startoff);
4969 	del_endoff = del->br_startoff + del->br_blockcount;
4970 	got_endoff = got.br_startoff + got.br_blockcount;
4971 	ASSERT(got_endoff >= del_endoff);
4972 	ASSERT(!isnullstartblock(got.br_startblock));
4973 	qfield = 0;
4974 	error = 0;
4975 
4976 	/*
4977 	 * If it's the case where the directory code is running with no block
4978 	 * reservation, and the deleted block is in the middle of its extent,
4979 	 * and the resulting insert of an extent would cause transformation to
4980 	 * btree format, then reject it.  The calling code will then swap blocks
4981 	 * around instead.  We have to do this now, rather than waiting for the
4982 	 * conversion to btree format, since the transaction will be dirty then.
4983 	 */
4984 	if (tp->t_blk_res == 0 &&
4985 	    XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
4986 	    XFS_IFORK_NEXTENTS(ip, whichfork) >=
4987 			XFS_IFORK_MAXEXT(ip, whichfork) &&
4988 	    del->br_startoff > got.br_startoff && del_endoff < got_endoff)
4989 		return -ENOSPC;
4990 
4991 	flags = XFS_ILOG_CORE;
4992 	if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
4993 		xfs_fsblock_t	bno;
4994 		xfs_filblks_t	len;
4995 		xfs_extlen_t	mod;
4996 
4997 		bno = div_u64_rem(del->br_startblock, mp->m_sb.sb_rextsize,
4998 				  &mod);
4999 		ASSERT(mod == 0);
5000 		len = div_u64_rem(del->br_blockcount, mp->m_sb.sb_rextsize,
5001 				  &mod);
5002 		ASSERT(mod == 0);
5003 
5004 		error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
5005 		if (error)
5006 			goto done;
5007 		do_fx = 0;
5008 		nblks = len * mp->m_sb.sb_rextsize;
5009 		qfield = XFS_TRANS_DQ_RTBCOUNT;
5010 	} else {
5011 		do_fx = 1;
5012 		nblks = del->br_blockcount;
5013 		qfield = XFS_TRANS_DQ_BCOUNT;
5014 	}
5015 
5016 	del_endblock = del->br_startblock + del->br_blockcount;
5017 	if (cur) {
5018 		error = xfs_bmbt_lookup_eq(cur, &got, &i);
5019 		if (error)
5020 			goto done;
5021 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5022 	}
5023 
5024 	if (got.br_startoff == del->br_startoff)
5025 		state |= BMAP_LEFT_FILLING;
5026 	if (got_endoff == del_endoff)
5027 		state |= BMAP_RIGHT_FILLING;
5028 
5029 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
5030 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
5031 		/*
5032 		 * Matches the whole extent.  Delete the entry.
5033 		 */
5034 		xfs_iext_remove(ip, icur, state);
5035 		xfs_iext_prev(ifp, icur);
5036 		XFS_IFORK_NEXT_SET(ip, whichfork,
5037 			XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5038 		flags |= XFS_ILOG_CORE;
5039 		if (!cur) {
5040 			flags |= xfs_ilog_fext(whichfork);
5041 			break;
5042 		}
5043 		if ((error = xfs_btree_delete(cur, &i)))
5044 			goto done;
5045 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5046 		break;
5047 	case BMAP_LEFT_FILLING:
5048 		/*
5049 		 * Deleting the first part of the extent.
5050 		 */
5051 		got.br_startoff = del_endoff;
5052 		got.br_startblock = del_endblock;
5053 		got.br_blockcount -= del->br_blockcount;
5054 		xfs_iext_update_extent(ip, state, icur, &got);
5055 		if (!cur) {
5056 			flags |= xfs_ilog_fext(whichfork);
5057 			break;
5058 		}
5059 		error = xfs_bmbt_update(cur, &got);
5060 		if (error)
5061 			goto done;
5062 		break;
5063 	case BMAP_RIGHT_FILLING:
5064 		/*
5065 		 * Deleting the last part of the extent.
5066 		 */
5067 		got.br_blockcount -= del->br_blockcount;
5068 		xfs_iext_update_extent(ip, state, icur, &got);
5069 		if (!cur) {
5070 			flags |= xfs_ilog_fext(whichfork);
5071 			break;
5072 		}
5073 		error = xfs_bmbt_update(cur, &got);
5074 		if (error)
5075 			goto done;
5076 		break;
5077 	case 0:
5078 		/*
5079 		 * Deleting the middle of the extent.
5080 		 */
5081 		old = got;
5082 
5083 		got.br_blockcount = del->br_startoff - got.br_startoff;
5084 		xfs_iext_update_extent(ip, state, icur, &got);
5085 
5086 		new.br_startoff = del_endoff;
5087 		new.br_blockcount = got_endoff - del_endoff;
5088 		new.br_state = got.br_state;
5089 		new.br_startblock = del_endblock;
5090 
5091 		flags |= XFS_ILOG_CORE;
5092 		if (cur) {
5093 			error = xfs_bmbt_update(cur, &got);
5094 			if (error)
5095 				goto done;
5096 			error = xfs_btree_increment(cur, 0, &i);
5097 			if (error)
5098 				goto done;
5099 			cur->bc_rec.b = new;
5100 			error = xfs_btree_insert(cur, &i);
5101 			if (error && error != -ENOSPC)
5102 				goto done;
5103 			/*
5104 			 * If get no-space back from btree insert, it tried a
5105 			 * split, and we have a zero block reservation.  Fix up
5106 			 * our state and return the error.
5107 			 */
5108 			if (error == -ENOSPC) {
5109 				/*
5110 				 * Reset the cursor, don't trust it after any
5111 				 * insert operation.
5112 				 */
5113 				error = xfs_bmbt_lookup_eq(cur, &got, &i);
5114 				if (error)
5115 					goto done;
5116 				XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5117 				/*
5118 				 * Update the btree record back
5119 				 * to the original value.
5120 				 */
5121 				error = xfs_bmbt_update(cur, &old);
5122 				if (error)
5123 					goto done;
5124 				/*
5125 				 * Reset the extent record back
5126 				 * to the original value.
5127 				 */
5128 				xfs_iext_update_extent(ip, state, icur, &old);
5129 				flags = 0;
5130 				error = -ENOSPC;
5131 				goto done;
5132 			}
5133 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5134 		} else
5135 			flags |= xfs_ilog_fext(whichfork);
5136 		XFS_IFORK_NEXT_SET(ip, whichfork,
5137 			XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5138 		xfs_iext_next(ifp, icur);
5139 		xfs_iext_insert(ip, icur, &new, state);
5140 		break;
5141 	}
5142 
5143 	/* remove reverse mapping */
5144 	error = xfs_rmap_unmap_extent(tp, ip, whichfork, del);
5145 	if (error)
5146 		goto done;
5147 
5148 	/*
5149 	 * If we need to, add to list of extents to delete.
5150 	 */
5151 	if (do_fx && !(bflags & XFS_BMAPI_REMAP)) {
5152 		if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
5153 			error = xfs_refcount_decrease_extent(tp, del);
5154 			if (error)
5155 				goto done;
5156 		} else {
5157 			__xfs_bmap_add_free(tp, del->br_startblock,
5158 					del->br_blockcount, NULL,
5159 					(bflags & XFS_BMAPI_NODISCARD) ||
5160 					del->br_state == XFS_EXT_UNWRITTEN);
5161 		}
5162 	}
5163 
5164 	/*
5165 	 * Adjust inode # blocks in the file.
5166 	 */
5167 	if (nblks)
5168 		ip->i_d.di_nblocks -= nblks;
5169 	/*
5170 	 * Adjust quota data.
5171 	 */
5172 	if (qfield && !(bflags & XFS_BMAPI_REMAP))
5173 		xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5174 
5175 done:
5176 	*logflagsp = flags;
5177 	return error;
5178 }
5179 
5180 /*
5181  * Unmap (remove) blocks from a file.
5182  * If nexts is nonzero then the number of extents to remove is limited to
5183  * that value.  If not all extents in the block range can be removed then
5184  * *done is set.
5185  */
5186 int						/* error */
5187 __xfs_bunmapi(
5188 	struct xfs_trans	*tp,		/* transaction pointer */
5189 	struct xfs_inode	*ip,		/* incore inode */
5190 	xfs_fileoff_t		start,		/* first file offset deleted */
5191 	xfs_filblks_t		*rlen,		/* i/o: amount remaining */
5192 	int			flags,		/* misc flags */
5193 	xfs_extnum_t		nexts)		/* number of extents max */
5194 {
5195 	struct xfs_btree_cur	*cur;		/* bmap btree cursor */
5196 	struct xfs_bmbt_irec	del;		/* extent being deleted */
5197 	int			error;		/* error return value */
5198 	xfs_extnum_t		extno;		/* extent number in list */
5199 	struct xfs_bmbt_irec	got;		/* current extent record */
5200 	struct xfs_ifork	*ifp;		/* inode fork pointer */
5201 	int			isrt;		/* freeing in rt area */
5202 	int			logflags;	/* transaction logging flags */
5203 	xfs_extlen_t		mod;		/* rt extent offset */
5204 	struct xfs_mount	*mp;		/* mount structure */
5205 	int			tmp_logflags;	/* partial logging flags */
5206 	int			wasdel;		/* was a delayed alloc extent */
5207 	int			whichfork;	/* data or attribute fork */
5208 	xfs_fsblock_t		sum;
5209 	xfs_filblks_t		len = *rlen;	/* length to unmap in file */
5210 	xfs_fileoff_t		max_len;
5211 	xfs_agnumber_t		prev_agno = NULLAGNUMBER, agno;
5212 	xfs_fileoff_t		end;
5213 	struct xfs_iext_cursor	icur;
5214 	bool			done = false;
5215 
5216 	trace_xfs_bunmap(ip, start, len, flags, _RET_IP_);
5217 
5218 	whichfork = xfs_bmapi_whichfork(flags);
5219 	ASSERT(whichfork != XFS_COW_FORK);
5220 	ifp = XFS_IFORK_PTR(ip, whichfork);
5221 	if (unlikely(
5222 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5223 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5224 		XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5225 				 ip->i_mount);
5226 		return -EFSCORRUPTED;
5227 	}
5228 	mp = ip->i_mount;
5229 	if (XFS_FORCED_SHUTDOWN(mp))
5230 		return -EIO;
5231 
5232 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5233 	ASSERT(len > 0);
5234 	ASSERT(nexts >= 0);
5235 
5236 	/*
5237 	 * Guesstimate how many blocks we can unmap without running the risk of
5238 	 * blowing out the transaction with a mix of EFIs and reflink
5239 	 * adjustments.
5240 	 */
5241 	if (tp && xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK)
5242 		max_len = min(len, xfs_refcount_max_unmap(tp->t_log_res));
5243 	else
5244 		max_len = len;
5245 
5246 	if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5247 	    (error = xfs_iread_extents(tp, ip, whichfork)))
5248 		return error;
5249 	if (xfs_iext_count(ifp) == 0) {
5250 		*rlen = 0;
5251 		return 0;
5252 	}
5253 	XFS_STATS_INC(mp, xs_blk_unmap);
5254 	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5255 	end = start + len;
5256 
5257 	if (!xfs_iext_lookup_extent_before(ip, ifp, &end, &icur, &got)) {
5258 		*rlen = 0;
5259 		return 0;
5260 	}
5261 	end--;
5262 
5263 	logflags = 0;
5264 	if (ifp->if_flags & XFS_IFBROOT) {
5265 		ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5266 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5267 		cur->bc_private.b.flags = 0;
5268 	} else
5269 		cur = NULL;
5270 
5271 	if (isrt) {
5272 		/*
5273 		 * Synchronize by locking the bitmap inode.
5274 		 */
5275 		xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
5276 		xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5277 		xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
5278 		xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
5279 	}
5280 
5281 	extno = 0;
5282 	while (end != (xfs_fileoff_t)-1 && end >= start &&
5283 	       (nexts == 0 || extno < nexts) && max_len > 0) {
5284 		/*
5285 		 * Is the found extent after a hole in which end lives?
5286 		 * Just back up to the previous extent, if so.
5287 		 */
5288 		if (got.br_startoff > end &&
5289 		    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5290 			done = true;
5291 			break;
5292 		}
5293 		/*
5294 		 * Is the last block of this extent before the range
5295 		 * we're supposed to delete?  If so, we're done.
5296 		 */
5297 		end = XFS_FILEOFF_MIN(end,
5298 			got.br_startoff + got.br_blockcount - 1);
5299 		if (end < start)
5300 			break;
5301 		/*
5302 		 * Then deal with the (possibly delayed) allocated space
5303 		 * we found.
5304 		 */
5305 		del = got;
5306 		wasdel = isnullstartblock(del.br_startblock);
5307 
5308 		/*
5309 		 * Make sure we don't touch multiple AGF headers out of order
5310 		 * in a single transaction, as that could cause AB-BA deadlocks.
5311 		 */
5312 		if (!wasdel) {
5313 			agno = XFS_FSB_TO_AGNO(mp, del.br_startblock);
5314 			if (prev_agno != NULLAGNUMBER && prev_agno > agno)
5315 				break;
5316 			prev_agno = agno;
5317 		}
5318 		if (got.br_startoff < start) {
5319 			del.br_startoff = start;
5320 			del.br_blockcount -= start - got.br_startoff;
5321 			if (!wasdel)
5322 				del.br_startblock += start - got.br_startoff;
5323 		}
5324 		if (del.br_startoff + del.br_blockcount > end + 1)
5325 			del.br_blockcount = end + 1 - del.br_startoff;
5326 
5327 		/* How much can we safely unmap? */
5328 		if (max_len < del.br_blockcount) {
5329 			del.br_startoff += del.br_blockcount - max_len;
5330 			if (!wasdel)
5331 				del.br_startblock += del.br_blockcount - max_len;
5332 			del.br_blockcount = max_len;
5333 		}
5334 
5335 		if (!isrt)
5336 			goto delete;
5337 
5338 		sum = del.br_startblock + del.br_blockcount;
5339 		div_u64_rem(sum, mp->m_sb.sb_rextsize, &mod);
5340 		if (mod) {
5341 			/*
5342 			 * Realtime extent not lined up at the end.
5343 			 * The extent could have been split into written
5344 			 * and unwritten pieces, or we could just be
5345 			 * unmapping part of it.  But we can't really
5346 			 * get rid of part of a realtime extent.
5347 			 */
5348 			if (del.br_state == XFS_EXT_UNWRITTEN) {
5349 				/*
5350 				 * This piece is unwritten, or we're not
5351 				 * using unwritten extents.  Skip over it.
5352 				 */
5353 				ASSERT(end >= mod);
5354 				end -= mod > del.br_blockcount ?
5355 					del.br_blockcount : mod;
5356 				if (end < got.br_startoff &&
5357 				    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5358 					done = true;
5359 					break;
5360 				}
5361 				continue;
5362 			}
5363 			/*
5364 			 * It's written, turn it unwritten.
5365 			 * This is better than zeroing it.
5366 			 */
5367 			ASSERT(del.br_state == XFS_EXT_NORM);
5368 			ASSERT(tp->t_blk_res > 0);
5369 			/*
5370 			 * If this spans a realtime extent boundary,
5371 			 * chop it back to the start of the one we end at.
5372 			 */
5373 			if (del.br_blockcount > mod) {
5374 				del.br_startoff += del.br_blockcount - mod;
5375 				del.br_startblock += del.br_blockcount - mod;
5376 				del.br_blockcount = mod;
5377 			}
5378 			del.br_state = XFS_EXT_UNWRITTEN;
5379 			error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5380 					whichfork, &icur, &cur, &del,
5381 					&logflags);
5382 			if (error)
5383 				goto error0;
5384 			goto nodelete;
5385 		}
5386 		div_u64_rem(del.br_startblock, mp->m_sb.sb_rextsize, &mod);
5387 		if (mod) {
5388 			/*
5389 			 * Realtime extent is lined up at the end but not
5390 			 * at the front.  We'll get rid of full extents if
5391 			 * we can.
5392 			 */
5393 			mod = mp->m_sb.sb_rextsize - mod;
5394 			if (del.br_blockcount > mod) {
5395 				del.br_blockcount -= mod;
5396 				del.br_startoff += mod;
5397 				del.br_startblock += mod;
5398 			} else if (del.br_startoff == start &&
5399 				   (del.br_state == XFS_EXT_UNWRITTEN ||
5400 				    tp->t_blk_res == 0)) {
5401 				/*
5402 				 * Can't make it unwritten.  There isn't
5403 				 * a full extent here so just skip it.
5404 				 */
5405 				ASSERT(end >= del.br_blockcount);
5406 				end -= del.br_blockcount;
5407 				if (got.br_startoff > end &&
5408 				    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5409 					done = true;
5410 					break;
5411 				}
5412 				continue;
5413 			} else if (del.br_state == XFS_EXT_UNWRITTEN) {
5414 				struct xfs_bmbt_irec	prev;
5415 
5416 				/*
5417 				 * This one is already unwritten.
5418 				 * It must have a written left neighbor.
5419 				 * Unwrite the killed part of that one and
5420 				 * try again.
5421 				 */
5422 				if (!xfs_iext_prev_extent(ifp, &icur, &prev))
5423 					ASSERT(0);
5424 				ASSERT(prev.br_state == XFS_EXT_NORM);
5425 				ASSERT(!isnullstartblock(prev.br_startblock));
5426 				ASSERT(del.br_startblock ==
5427 				       prev.br_startblock + prev.br_blockcount);
5428 				if (prev.br_startoff < start) {
5429 					mod = start - prev.br_startoff;
5430 					prev.br_blockcount -= mod;
5431 					prev.br_startblock += mod;
5432 					prev.br_startoff = start;
5433 				}
5434 				prev.br_state = XFS_EXT_UNWRITTEN;
5435 				error = xfs_bmap_add_extent_unwritten_real(tp,
5436 						ip, whichfork, &icur, &cur,
5437 						&prev, &logflags);
5438 				if (error)
5439 					goto error0;
5440 				goto nodelete;
5441 			} else {
5442 				ASSERT(del.br_state == XFS_EXT_NORM);
5443 				del.br_state = XFS_EXT_UNWRITTEN;
5444 				error = xfs_bmap_add_extent_unwritten_real(tp,
5445 						ip, whichfork, &icur, &cur,
5446 						&del, &logflags);
5447 				if (error)
5448 					goto error0;
5449 				goto nodelete;
5450 			}
5451 		}
5452 
5453 delete:
5454 		if (wasdel) {
5455 			error = xfs_bmap_del_extent_delay(ip, whichfork, &icur,
5456 					&got, &del);
5457 		} else {
5458 			error = xfs_bmap_del_extent_real(ip, tp, &icur, cur,
5459 					&del, &tmp_logflags, whichfork,
5460 					flags);
5461 			logflags |= tmp_logflags;
5462 		}
5463 
5464 		if (error)
5465 			goto error0;
5466 
5467 		max_len -= del.br_blockcount;
5468 		end = del.br_startoff - 1;
5469 nodelete:
5470 		/*
5471 		 * If not done go on to the next (previous) record.
5472 		 */
5473 		if (end != (xfs_fileoff_t)-1 && end >= start) {
5474 			if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5475 			    (got.br_startoff > end &&
5476 			     !xfs_iext_prev_extent(ifp, &icur, &got))) {
5477 				done = true;
5478 				break;
5479 			}
5480 			extno++;
5481 		}
5482 	}
5483 	if (done || end == (xfs_fileoff_t)-1 || end < start)
5484 		*rlen = 0;
5485 	else
5486 		*rlen = end - start + 1;
5487 
5488 	/*
5489 	 * Convert to a btree if necessary.
5490 	 */
5491 	if (xfs_bmap_needs_btree(ip, whichfork)) {
5492 		ASSERT(cur == NULL);
5493 		error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
5494 				&tmp_logflags, whichfork);
5495 		logflags |= tmp_logflags;
5496 	} else {
5497 		error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags,
5498 			whichfork);
5499 	}
5500 
5501 error0:
5502 	/*
5503 	 * Log everything.  Do this after conversion, there's no point in
5504 	 * logging the extent records if we've converted to btree format.
5505 	 */
5506 	if ((logflags & xfs_ilog_fext(whichfork)) &&
5507 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5508 		logflags &= ~xfs_ilog_fext(whichfork);
5509 	else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5510 		 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5511 		logflags &= ~xfs_ilog_fbroot(whichfork);
5512 	/*
5513 	 * Log inode even in the error case, if the transaction
5514 	 * is dirty we'll need to shut down the filesystem.
5515 	 */
5516 	if (logflags)
5517 		xfs_trans_log_inode(tp, ip, logflags);
5518 	if (cur) {
5519 		if (!error)
5520 			cur->bc_private.b.allocated = 0;
5521 		xfs_btree_del_cursor(cur, error);
5522 	}
5523 	return error;
5524 }
5525 
5526 /* Unmap a range of a file. */
5527 int
5528 xfs_bunmapi(
5529 	xfs_trans_t		*tp,
5530 	struct xfs_inode	*ip,
5531 	xfs_fileoff_t		bno,
5532 	xfs_filblks_t		len,
5533 	int			flags,
5534 	xfs_extnum_t		nexts,
5535 	int			*done)
5536 {
5537 	int			error;
5538 
5539 	error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts);
5540 	*done = (len == 0);
5541 	return error;
5542 }
5543 
5544 /*
5545  * Determine whether an extent shift can be accomplished by a merge with the
5546  * extent that precedes the target hole of the shift.
5547  */
5548 STATIC bool
5549 xfs_bmse_can_merge(
5550 	struct xfs_bmbt_irec	*left,	/* preceding extent */
5551 	struct xfs_bmbt_irec	*got,	/* current extent to shift */
5552 	xfs_fileoff_t		shift)	/* shift fsb */
5553 {
5554 	xfs_fileoff_t		startoff;
5555 
5556 	startoff = got->br_startoff - shift;
5557 
5558 	/*
5559 	 * The extent, once shifted, must be adjacent in-file and on-disk with
5560 	 * the preceding extent.
5561 	 */
5562 	if ((left->br_startoff + left->br_blockcount != startoff) ||
5563 	    (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5564 	    (left->br_state != got->br_state) ||
5565 	    (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5566 		return false;
5567 
5568 	return true;
5569 }
5570 
5571 /*
5572  * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5573  * hole in the file. If an extent shift would result in the extent being fully
5574  * adjacent to the extent that currently precedes the hole, we can merge with
5575  * the preceding extent rather than do the shift.
5576  *
5577  * This function assumes the caller has verified a shift-by-merge is possible
5578  * with the provided extents via xfs_bmse_can_merge().
5579  */
5580 STATIC int
5581 xfs_bmse_merge(
5582 	struct xfs_trans		*tp,
5583 	struct xfs_inode		*ip,
5584 	int				whichfork,
5585 	xfs_fileoff_t			shift,		/* shift fsb */
5586 	struct xfs_iext_cursor		*icur,
5587 	struct xfs_bmbt_irec		*got,		/* extent to shift */
5588 	struct xfs_bmbt_irec		*left,		/* preceding extent */
5589 	struct xfs_btree_cur		*cur,
5590 	int				*logflags)	/* output */
5591 {
5592 	struct xfs_bmbt_irec		new;
5593 	xfs_filblks_t			blockcount;
5594 	int				error, i;
5595 	struct xfs_mount		*mp = ip->i_mount;
5596 
5597 	blockcount = left->br_blockcount + got->br_blockcount;
5598 
5599 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5600 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5601 	ASSERT(xfs_bmse_can_merge(left, got, shift));
5602 
5603 	new = *left;
5604 	new.br_blockcount = blockcount;
5605 
5606 	/*
5607 	 * Update the on-disk extent count, the btree if necessary and log the
5608 	 * inode.
5609 	 */
5610 	XFS_IFORK_NEXT_SET(ip, whichfork,
5611 			   XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5612 	*logflags |= XFS_ILOG_CORE;
5613 	if (!cur) {
5614 		*logflags |= XFS_ILOG_DEXT;
5615 		goto done;
5616 	}
5617 
5618 	/* lookup and remove the extent to merge */
5619 	error = xfs_bmbt_lookup_eq(cur, got, &i);
5620 	if (error)
5621 		return error;
5622 	XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5623 
5624 	error = xfs_btree_delete(cur, &i);
5625 	if (error)
5626 		return error;
5627 	XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5628 
5629 	/* lookup and update size of the previous extent */
5630 	error = xfs_bmbt_lookup_eq(cur, left, &i);
5631 	if (error)
5632 		return error;
5633 	XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5634 
5635 	error = xfs_bmbt_update(cur, &new);
5636 	if (error)
5637 		return error;
5638 
5639 done:
5640 	xfs_iext_remove(ip, icur, 0);
5641 	xfs_iext_prev(XFS_IFORK_PTR(ip, whichfork), icur);
5642 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5643 			&new);
5644 
5645 	/* update reverse mapping. rmap functions merge the rmaps for us */
5646 	error = xfs_rmap_unmap_extent(tp, ip, whichfork, got);
5647 	if (error)
5648 		return error;
5649 	memcpy(&new, got, sizeof(new));
5650 	new.br_startoff = left->br_startoff + left->br_blockcount;
5651 	return xfs_rmap_map_extent(tp, ip, whichfork, &new);
5652 }
5653 
5654 static int
5655 xfs_bmap_shift_update_extent(
5656 	struct xfs_trans	*tp,
5657 	struct xfs_inode	*ip,
5658 	int			whichfork,
5659 	struct xfs_iext_cursor	*icur,
5660 	struct xfs_bmbt_irec	*got,
5661 	struct xfs_btree_cur	*cur,
5662 	int			*logflags,
5663 	xfs_fileoff_t		startoff)
5664 {
5665 	struct xfs_mount	*mp = ip->i_mount;
5666 	struct xfs_bmbt_irec	prev = *got;
5667 	int			error, i;
5668 
5669 	*logflags |= XFS_ILOG_CORE;
5670 
5671 	got->br_startoff = startoff;
5672 
5673 	if (cur) {
5674 		error = xfs_bmbt_lookup_eq(cur, &prev, &i);
5675 		if (error)
5676 			return error;
5677 		XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5678 
5679 		error = xfs_bmbt_update(cur, got);
5680 		if (error)
5681 			return error;
5682 	} else {
5683 		*logflags |= XFS_ILOG_DEXT;
5684 	}
5685 
5686 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5687 			got);
5688 
5689 	/* update reverse mapping */
5690 	error = xfs_rmap_unmap_extent(tp, ip, whichfork, &prev);
5691 	if (error)
5692 		return error;
5693 	return xfs_rmap_map_extent(tp, ip, whichfork, got);
5694 }
5695 
5696 int
5697 xfs_bmap_collapse_extents(
5698 	struct xfs_trans	*tp,
5699 	struct xfs_inode	*ip,
5700 	xfs_fileoff_t		*next_fsb,
5701 	xfs_fileoff_t		offset_shift_fsb,
5702 	bool			*done)
5703 {
5704 	int			whichfork = XFS_DATA_FORK;
5705 	struct xfs_mount	*mp = ip->i_mount;
5706 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
5707 	struct xfs_btree_cur	*cur = NULL;
5708 	struct xfs_bmbt_irec	got, prev;
5709 	struct xfs_iext_cursor	icur;
5710 	xfs_fileoff_t		new_startoff;
5711 	int			error = 0;
5712 	int			logflags = 0;
5713 
5714 	if (unlikely(XFS_TEST_ERROR(
5715 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5716 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5717 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
5718 		XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
5719 		return -EFSCORRUPTED;
5720 	}
5721 
5722 	if (XFS_FORCED_SHUTDOWN(mp))
5723 		return -EIO;
5724 
5725 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL));
5726 
5727 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5728 		error = xfs_iread_extents(tp, ip, whichfork);
5729 		if (error)
5730 			return error;
5731 	}
5732 
5733 	if (ifp->if_flags & XFS_IFBROOT) {
5734 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5735 		cur->bc_private.b.flags = 0;
5736 	}
5737 
5738 	if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5739 		*done = true;
5740 		goto del_cursor;
5741 	}
5742 	XFS_WANT_CORRUPTED_GOTO(mp, !isnullstartblock(got.br_startblock),
5743 				del_cursor);
5744 
5745 	new_startoff = got.br_startoff - offset_shift_fsb;
5746 	if (xfs_iext_peek_prev_extent(ifp, &icur, &prev)) {
5747 		if (new_startoff < prev.br_startoff + prev.br_blockcount) {
5748 			error = -EINVAL;
5749 			goto del_cursor;
5750 		}
5751 
5752 		if (xfs_bmse_can_merge(&prev, &got, offset_shift_fsb)) {
5753 			error = xfs_bmse_merge(tp, ip, whichfork,
5754 					offset_shift_fsb, &icur, &got, &prev,
5755 					cur, &logflags);
5756 			if (error)
5757 				goto del_cursor;
5758 			goto done;
5759 		}
5760 	} else {
5761 		if (got.br_startoff < offset_shift_fsb) {
5762 			error = -EINVAL;
5763 			goto del_cursor;
5764 		}
5765 	}
5766 
5767 	error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got,
5768 			cur, &logflags, new_startoff);
5769 	if (error)
5770 		goto del_cursor;
5771 
5772 done:
5773 	if (!xfs_iext_next_extent(ifp, &icur, &got)) {
5774 		*done = true;
5775 		goto del_cursor;
5776 	}
5777 
5778 	*next_fsb = got.br_startoff;
5779 del_cursor:
5780 	if (cur)
5781 		xfs_btree_del_cursor(cur, error);
5782 	if (logflags)
5783 		xfs_trans_log_inode(tp, ip, logflags);
5784 	return error;
5785 }
5786 
5787 /* Make sure we won't be right-shifting an extent past the maximum bound. */
5788 int
5789 xfs_bmap_can_insert_extents(
5790 	struct xfs_inode	*ip,
5791 	xfs_fileoff_t		off,
5792 	xfs_fileoff_t		shift)
5793 {
5794 	struct xfs_bmbt_irec	got;
5795 	int			is_empty;
5796 	int			error = 0;
5797 
5798 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5799 
5800 	if (XFS_FORCED_SHUTDOWN(ip->i_mount))
5801 		return -EIO;
5802 
5803 	xfs_ilock(ip, XFS_ILOCK_EXCL);
5804 	error = xfs_bmap_last_extent(NULL, ip, XFS_DATA_FORK, &got, &is_empty);
5805 	if (!error && !is_empty && got.br_startoff >= off &&
5806 	    ((got.br_startoff + shift) & BMBT_STARTOFF_MASK) < got.br_startoff)
5807 		error = -EINVAL;
5808 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
5809 
5810 	return error;
5811 }
5812 
5813 int
5814 xfs_bmap_insert_extents(
5815 	struct xfs_trans	*tp,
5816 	struct xfs_inode	*ip,
5817 	xfs_fileoff_t		*next_fsb,
5818 	xfs_fileoff_t		offset_shift_fsb,
5819 	bool			*done,
5820 	xfs_fileoff_t		stop_fsb)
5821 {
5822 	int			whichfork = XFS_DATA_FORK;
5823 	struct xfs_mount	*mp = ip->i_mount;
5824 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
5825 	struct xfs_btree_cur	*cur = NULL;
5826 	struct xfs_bmbt_irec	got, next;
5827 	struct xfs_iext_cursor	icur;
5828 	xfs_fileoff_t		new_startoff;
5829 	int			error = 0;
5830 	int			logflags = 0;
5831 
5832 	if (unlikely(XFS_TEST_ERROR(
5833 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5834 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5835 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
5836 		XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
5837 		return -EFSCORRUPTED;
5838 	}
5839 
5840 	if (XFS_FORCED_SHUTDOWN(mp))
5841 		return -EIO;
5842 
5843 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL));
5844 
5845 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5846 		error = xfs_iread_extents(tp, ip, whichfork);
5847 		if (error)
5848 			return error;
5849 	}
5850 
5851 	if (ifp->if_flags & XFS_IFBROOT) {
5852 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5853 		cur->bc_private.b.flags = 0;
5854 	}
5855 
5856 	if (*next_fsb == NULLFSBLOCK) {
5857 		xfs_iext_last(ifp, &icur);
5858 		if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5859 		    stop_fsb > got.br_startoff) {
5860 			*done = true;
5861 			goto del_cursor;
5862 		}
5863 	} else {
5864 		if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5865 			*done = true;
5866 			goto del_cursor;
5867 		}
5868 	}
5869 	XFS_WANT_CORRUPTED_GOTO(mp, !isnullstartblock(got.br_startblock),
5870 				del_cursor);
5871 
5872 	if (stop_fsb >= got.br_startoff + got.br_blockcount) {
5873 		error = -EIO;
5874 		goto del_cursor;
5875 	}
5876 
5877 	new_startoff = got.br_startoff + offset_shift_fsb;
5878 	if (xfs_iext_peek_next_extent(ifp, &icur, &next)) {
5879 		if (new_startoff + got.br_blockcount > next.br_startoff) {
5880 			error = -EINVAL;
5881 			goto del_cursor;
5882 		}
5883 
5884 		/*
5885 		 * Unlike a left shift (which involves a hole punch), a right
5886 		 * shift does not modify extent neighbors in any way.  We should
5887 		 * never find mergeable extents in this scenario.  Check anyways
5888 		 * and warn if we encounter two extents that could be one.
5889 		 */
5890 		if (xfs_bmse_can_merge(&got, &next, offset_shift_fsb))
5891 			WARN_ON_ONCE(1);
5892 	}
5893 
5894 	error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got,
5895 			cur, &logflags, new_startoff);
5896 	if (error)
5897 		goto del_cursor;
5898 
5899 	if (!xfs_iext_prev_extent(ifp, &icur, &got) ||
5900 	    stop_fsb >= got.br_startoff + got.br_blockcount) {
5901 		*done = true;
5902 		goto del_cursor;
5903 	}
5904 
5905 	*next_fsb = got.br_startoff;
5906 del_cursor:
5907 	if (cur)
5908 		xfs_btree_del_cursor(cur, error);
5909 	if (logflags)
5910 		xfs_trans_log_inode(tp, ip, logflags);
5911 	return error;
5912 }
5913 
5914 /*
5915  * Splits an extent into two extents at split_fsb block such that it is the
5916  * first block of the current_ext. @ext is a target extent to be split.
5917  * @split_fsb is a block where the extents is split.  If split_fsb lies in a
5918  * hole or the first block of extents, just return 0.
5919  */
5920 STATIC int
5921 xfs_bmap_split_extent_at(
5922 	struct xfs_trans	*tp,
5923 	struct xfs_inode	*ip,
5924 	xfs_fileoff_t		split_fsb)
5925 {
5926 	int				whichfork = XFS_DATA_FORK;
5927 	struct xfs_btree_cur		*cur = NULL;
5928 	struct xfs_bmbt_irec		got;
5929 	struct xfs_bmbt_irec		new; /* split extent */
5930 	struct xfs_mount		*mp = ip->i_mount;
5931 	struct xfs_ifork		*ifp;
5932 	xfs_fsblock_t			gotblkcnt; /* new block count for got */
5933 	struct xfs_iext_cursor		icur;
5934 	int				error = 0;
5935 	int				logflags = 0;
5936 	int				i = 0;
5937 
5938 	if (unlikely(XFS_TEST_ERROR(
5939 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5940 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5941 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
5942 		XFS_ERROR_REPORT("xfs_bmap_split_extent_at",
5943 				 XFS_ERRLEVEL_LOW, mp);
5944 		return -EFSCORRUPTED;
5945 	}
5946 
5947 	if (XFS_FORCED_SHUTDOWN(mp))
5948 		return -EIO;
5949 
5950 	ifp = XFS_IFORK_PTR(ip, whichfork);
5951 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5952 		/* Read in all the extents */
5953 		error = xfs_iread_extents(tp, ip, whichfork);
5954 		if (error)
5955 			return error;
5956 	}
5957 
5958 	/*
5959 	 * If there are not extents, or split_fsb lies in a hole we are done.
5960 	 */
5961 	if (!xfs_iext_lookup_extent(ip, ifp, split_fsb, &icur, &got) ||
5962 	    got.br_startoff >= split_fsb)
5963 		return 0;
5964 
5965 	gotblkcnt = split_fsb - got.br_startoff;
5966 	new.br_startoff = split_fsb;
5967 	new.br_startblock = got.br_startblock + gotblkcnt;
5968 	new.br_blockcount = got.br_blockcount - gotblkcnt;
5969 	new.br_state = got.br_state;
5970 
5971 	if (ifp->if_flags & XFS_IFBROOT) {
5972 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5973 		cur->bc_private.b.flags = 0;
5974 		error = xfs_bmbt_lookup_eq(cur, &got, &i);
5975 		if (error)
5976 			goto del_cursor;
5977 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
5978 	}
5979 
5980 	got.br_blockcount = gotblkcnt;
5981 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), &icur,
5982 			&got);
5983 
5984 	logflags = XFS_ILOG_CORE;
5985 	if (cur) {
5986 		error = xfs_bmbt_update(cur, &got);
5987 		if (error)
5988 			goto del_cursor;
5989 	} else
5990 		logflags |= XFS_ILOG_DEXT;
5991 
5992 	/* Add new extent */
5993 	xfs_iext_next(ifp, &icur);
5994 	xfs_iext_insert(ip, &icur, &new, 0);
5995 	XFS_IFORK_NEXT_SET(ip, whichfork,
5996 			   XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5997 
5998 	if (cur) {
5999 		error = xfs_bmbt_lookup_eq(cur, &new, &i);
6000 		if (error)
6001 			goto del_cursor;
6002 		XFS_WANT_CORRUPTED_GOTO(mp, i == 0, del_cursor);
6003 		error = xfs_btree_insert(cur, &i);
6004 		if (error)
6005 			goto del_cursor;
6006 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
6007 	}
6008 
6009 	/*
6010 	 * Convert to a btree if necessary.
6011 	 */
6012 	if (xfs_bmap_needs_btree(ip, whichfork)) {
6013 		int tmp_logflags; /* partial log flag return val */
6014 
6015 		ASSERT(cur == NULL);
6016 		error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
6017 				&tmp_logflags, whichfork);
6018 		logflags |= tmp_logflags;
6019 	}
6020 
6021 del_cursor:
6022 	if (cur) {
6023 		cur->bc_private.b.allocated = 0;
6024 		xfs_btree_del_cursor(cur, error);
6025 	}
6026 
6027 	if (logflags)
6028 		xfs_trans_log_inode(tp, ip, logflags);
6029 	return error;
6030 }
6031 
6032 int
6033 xfs_bmap_split_extent(
6034 	struct xfs_inode        *ip,
6035 	xfs_fileoff_t           split_fsb)
6036 {
6037 	struct xfs_mount        *mp = ip->i_mount;
6038 	struct xfs_trans        *tp;
6039 	int                     error;
6040 
6041 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
6042 			XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
6043 	if (error)
6044 		return error;
6045 
6046 	xfs_ilock(ip, XFS_ILOCK_EXCL);
6047 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
6048 
6049 	error = xfs_bmap_split_extent_at(tp, ip, split_fsb);
6050 	if (error)
6051 		goto out;
6052 
6053 	return xfs_trans_commit(tp);
6054 
6055 out:
6056 	xfs_trans_cancel(tp);
6057 	return error;
6058 }
6059 
6060 /* Deferred mapping is only for real extents in the data fork. */
6061 static bool
6062 xfs_bmap_is_update_needed(
6063 	struct xfs_bmbt_irec	*bmap)
6064 {
6065 	return  bmap->br_startblock != HOLESTARTBLOCK &&
6066 		bmap->br_startblock != DELAYSTARTBLOCK;
6067 }
6068 
6069 /* Record a bmap intent. */
6070 static int
6071 __xfs_bmap_add(
6072 	struct xfs_trans		*tp,
6073 	enum xfs_bmap_intent_type	type,
6074 	struct xfs_inode		*ip,
6075 	int				whichfork,
6076 	struct xfs_bmbt_irec		*bmap)
6077 {
6078 	struct xfs_bmap_intent		*bi;
6079 
6080 	trace_xfs_bmap_defer(tp->t_mountp,
6081 			XFS_FSB_TO_AGNO(tp->t_mountp, bmap->br_startblock),
6082 			type,
6083 			XFS_FSB_TO_AGBNO(tp->t_mountp, bmap->br_startblock),
6084 			ip->i_ino, whichfork,
6085 			bmap->br_startoff,
6086 			bmap->br_blockcount,
6087 			bmap->br_state);
6088 
6089 	bi = kmem_alloc(sizeof(struct xfs_bmap_intent), KM_SLEEP | KM_NOFS);
6090 	INIT_LIST_HEAD(&bi->bi_list);
6091 	bi->bi_type = type;
6092 	bi->bi_owner = ip;
6093 	bi->bi_whichfork = whichfork;
6094 	bi->bi_bmap = *bmap;
6095 
6096 	xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list);
6097 	return 0;
6098 }
6099 
6100 /* Map an extent into a file. */
6101 int
6102 xfs_bmap_map_extent(
6103 	struct xfs_trans	*tp,
6104 	struct xfs_inode	*ip,
6105 	struct xfs_bmbt_irec	*PREV)
6106 {
6107 	if (!xfs_bmap_is_update_needed(PREV))
6108 		return 0;
6109 
6110 	return __xfs_bmap_add(tp, XFS_BMAP_MAP, ip, XFS_DATA_FORK, PREV);
6111 }
6112 
6113 /* Unmap an extent out of a file. */
6114 int
6115 xfs_bmap_unmap_extent(
6116 	struct xfs_trans	*tp,
6117 	struct xfs_inode	*ip,
6118 	struct xfs_bmbt_irec	*PREV)
6119 {
6120 	if (!xfs_bmap_is_update_needed(PREV))
6121 		return 0;
6122 
6123 	return __xfs_bmap_add(tp, XFS_BMAP_UNMAP, ip, XFS_DATA_FORK, PREV);
6124 }
6125 
6126 /*
6127  * Process one of the deferred bmap operations.  We pass back the
6128  * btree cursor to maintain our lock on the bmapbt between calls.
6129  */
6130 int
6131 xfs_bmap_finish_one(
6132 	struct xfs_trans		*tp,
6133 	struct xfs_inode		*ip,
6134 	enum xfs_bmap_intent_type	type,
6135 	int				whichfork,
6136 	xfs_fileoff_t			startoff,
6137 	xfs_fsblock_t			startblock,
6138 	xfs_filblks_t			*blockcount,
6139 	xfs_exntst_t			state)
6140 {
6141 	int				error = 0;
6142 
6143 	ASSERT(tp->t_firstblock == NULLFSBLOCK);
6144 
6145 	trace_xfs_bmap_deferred(tp->t_mountp,
6146 			XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type,
6147 			XFS_FSB_TO_AGBNO(tp->t_mountp, startblock),
6148 			ip->i_ino, whichfork, startoff, *blockcount, state);
6149 
6150 	if (WARN_ON_ONCE(whichfork != XFS_DATA_FORK))
6151 		return -EFSCORRUPTED;
6152 
6153 	if (XFS_TEST_ERROR(false, tp->t_mountp,
6154 			XFS_ERRTAG_BMAP_FINISH_ONE))
6155 		return -EIO;
6156 
6157 	switch (type) {
6158 	case XFS_BMAP_MAP:
6159 		error = xfs_bmapi_remap(tp, ip, startoff, *blockcount,
6160 				startblock, 0);
6161 		*blockcount = 0;
6162 		break;
6163 	case XFS_BMAP_UNMAP:
6164 		error = __xfs_bunmapi(tp, ip, startoff, blockcount,
6165 				XFS_BMAPI_REMAP, 1);
6166 		break;
6167 	default:
6168 		ASSERT(0);
6169 		error = -EFSCORRUPTED;
6170 	}
6171 
6172 	return error;
6173 }
6174 
6175 /* Check that an inode's extent does not have invalid flags or bad ranges. */
6176 xfs_failaddr_t
6177 xfs_bmap_validate_extent(
6178 	struct xfs_inode	*ip,
6179 	int			whichfork,
6180 	struct xfs_bmbt_irec	*irec)
6181 {
6182 	struct xfs_mount	*mp = ip->i_mount;
6183 	xfs_fsblock_t		endfsb;
6184 	bool			isrt;
6185 
6186 	isrt = XFS_IS_REALTIME_INODE(ip);
6187 	endfsb = irec->br_startblock + irec->br_blockcount - 1;
6188 	if (isrt) {
6189 		if (!xfs_verify_rtbno(mp, irec->br_startblock))
6190 			return __this_address;
6191 		if (!xfs_verify_rtbno(mp, endfsb))
6192 			return __this_address;
6193 	} else {
6194 		if (!xfs_verify_fsbno(mp, irec->br_startblock))
6195 			return __this_address;
6196 		if (!xfs_verify_fsbno(mp, endfsb))
6197 			return __this_address;
6198 		if (XFS_FSB_TO_AGNO(mp, irec->br_startblock) !=
6199 		    XFS_FSB_TO_AGNO(mp, endfsb))
6200 			return __this_address;
6201 	}
6202 	if (irec->br_state != XFS_EXT_NORM && whichfork != XFS_DATA_FORK)
6203 		return __this_address;
6204 	return NULL;
6205 }
6206