xref: /linux/fs/xfs/libxfs/xfs_bmap.c (revision e21f9e2e862e9eb3dd64eaddb6256b3e5098660f)
1 /*
2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_bit.h"
25 #include "xfs_sb.h"
26 #include "xfs_mount.h"
27 #include "xfs_defer.h"
28 #include "xfs_da_format.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_dir2.h"
31 #include "xfs_inode.h"
32 #include "xfs_btree.h"
33 #include "xfs_trans.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_extfree_item.h"
36 #include "xfs_alloc.h"
37 #include "xfs_bmap.h"
38 #include "xfs_bmap_util.h"
39 #include "xfs_bmap_btree.h"
40 #include "xfs_rtalloc.h"
41 #include "xfs_errortag.h"
42 #include "xfs_error.h"
43 #include "xfs_quota.h"
44 #include "xfs_trans_space.h"
45 #include "xfs_buf_item.h"
46 #include "xfs_trace.h"
47 #include "xfs_symlink.h"
48 #include "xfs_attr_leaf.h"
49 #include "xfs_filestream.h"
50 #include "xfs_rmap.h"
51 #include "xfs_ag_resv.h"
52 #include "xfs_refcount.h"
53 #include "xfs_icache.h"
54 
55 
56 kmem_zone_t		*xfs_bmap_free_item_zone;
57 
58 /*
59  * Miscellaneous helper functions
60  */
61 
62 /*
63  * Compute and fill in the value of the maximum depth of a bmap btree
64  * in this filesystem.  Done once, during mount.
65  */
66 void
67 xfs_bmap_compute_maxlevels(
68 	xfs_mount_t	*mp,		/* file system mount structure */
69 	int		whichfork)	/* data or attr fork */
70 {
71 	int		level;		/* btree level */
72 	uint		maxblocks;	/* max blocks at this level */
73 	uint		maxleafents;	/* max leaf entries possible */
74 	int		maxrootrecs;	/* max records in root block */
75 	int		minleafrecs;	/* min records in leaf block */
76 	int		minnoderecs;	/* min records in node block */
77 	int		sz;		/* root block size */
78 
79 	/*
80 	 * The maximum number of extents in a file, hence the maximum
81 	 * number of leaf entries, is controlled by the type of di_nextents
82 	 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
83 	 * (a signed 16-bit number, xfs_aextnum_t).
84 	 *
85 	 * Note that we can no longer assume that if we are in ATTR1 that
86 	 * the fork offset of all the inodes will be
87 	 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
88 	 * with ATTR2 and then mounted back with ATTR1, keeping the
89 	 * di_forkoff's fixed but probably at various positions. Therefore,
90 	 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
91 	 * of a minimum size available.
92 	 */
93 	if (whichfork == XFS_DATA_FORK) {
94 		maxleafents = MAXEXTNUM;
95 		sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
96 	} else {
97 		maxleafents = MAXAEXTNUM;
98 		sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
99 	}
100 	maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
101 	minleafrecs = mp->m_bmap_dmnr[0];
102 	minnoderecs = mp->m_bmap_dmnr[1];
103 	maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
104 	for (level = 1; maxblocks > 1; level++) {
105 		if (maxblocks <= maxrootrecs)
106 			maxblocks = 1;
107 		else
108 			maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
109 	}
110 	mp->m_bm_maxlevels[whichfork] = level;
111 }
112 
113 STATIC int				/* error */
114 xfs_bmbt_lookup_eq(
115 	struct xfs_btree_cur	*cur,
116 	struct xfs_bmbt_irec	*irec,
117 	int			*stat)	/* success/failure */
118 {
119 	cur->bc_rec.b = *irec;
120 	return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
121 }
122 
123 STATIC int				/* error */
124 xfs_bmbt_lookup_first(
125 	struct xfs_btree_cur	*cur,
126 	int			*stat)	/* success/failure */
127 {
128 	cur->bc_rec.b.br_startoff = 0;
129 	cur->bc_rec.b.br_startblock = 0;
130 	cur->bc_rec.b.br_blockcount = 0;
131 	return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
132 }
133 
134 /*
135  * Check if the inode needs to be converted to btree format.
136  */
137 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
138 {
139 	return whichfork != XFS_COW_FORK &&
140 		XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
141 		XFS_IFORK_NEXTENTS(ip, whichfork) >
142 			XFS_IFORK_MAXEXT(ip, whichfork);
143 }
144 
145 /*
146  * Check if the inode should be converted to extent format.
147  */
148 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
149 {
150 	return whichfork != XFS_COW_FORK &&
151 		XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
152 		XFS_IFORK_NEXTENTS(ip, whichfork) <=
153 			XFS_IFORK_MAXEXT(ip, whichfork);
154 }
155 
156 /*
157  * Update the record referred to by cur to the value given by irec
158  * This either works (return 0) or gets an EFSCORRUPTED error.
159  */
160 STATIC int
161 xfs_bmbt_update(
162 	struct xfs_btree_cur	*cur,
163 	struct xfs_bmbt_irec	*irec)
164 {
165 	union xfs_btree_rec	rec;
166 
167 	xfs_bmbt_disk_set_all(&rec.bmbt, irec);
168 	return xfs_btree_update(cur, &rec);
169 }
170 
171 /*
172  * Compute the worst-case number of indirect blocks that will be used
173  * for ip's delayed extent of length "len".
174  */
175 STATIC xfs_filblks_t
176 xfs_bmap_worst_indlen(
177 	xfs_inode_t	*ip,		/* incore inode pointer */
178 	xfs_filblks_t	len)		/* delayed extent length */
179 {
180 	int		level;		/* btree level number */
181 	int		maxrecs;	/* maximum record count at this level */
182 	xfs_mount_t	*mp;		/* mount structure */
183 	xfs_filblks_t	rval;		/* return value */
184 
185 	mp = ip->i_mount;
186 	maxrecs = mp->m_bmap_dmxr[0];
187 	for (level = 0, rval = 0;
188 	     level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
189 	     level++) {
190 		len += maxrecs - 1;
191 		do_div(len, maxrecs);
192 		rval += len;
193 		if (len == 1)
194 			return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
195 				level - 1;
196 		if (level == 0)
197 			maxrecs = mp->m_bmap_dmxr[1];
198 	}
199 	return rval;
200 }
201 
202 /*
203  * Calculate the default attribute fork offset for newly created inodes.
204  */
205 uint
206 xfs_default_attroffset(
207 	struct xfs_inode	*ip)
208 {
209 	struct xfs_mount	*mp = ip->i_mount;
210 	uint			offset;
211 
212 	if (mp->m_sb.sb_inodesize == 256) {
213 		offset = XFS_LITINO(mp, ip->i_d.di_version) -
214 				XFS_BMDR_SPACE_CALC(MINABTPTRS);
215 	} else {
216 		offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
217 	}
218 
219 	ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
220 	return offset;
221 }
222 
223 /*
224  * Helper routine to reset inode di_forkoff field when switching
225  * attribute fork from local to extent format - we reset it where
226  * possible to make space available for inline data fork extents.
227  */
228 STATIC void
229 xfs_bmap_forkoff_reset(
230 	xfs_inode_t	*ip,
231 	int		whichfork)
232 {
233 	if (whichfork == XFS_ATTR_FORK &&
234 	    ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
235 	    ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
236 		uint	dfl_forkoff = xfs_default_attroffset(ip) >> 3;
237 
238 		if (dfl_forkoff > ip->i_d.di_forkoff)
239 			ip->i_d.di_forkoff = dfl_forkoff;
240 	}
241 }
242 
243 #ifdef DEBUG
244 STATIC struct xfs_buf *
245 xfs_bmap_get_bp(
246 	struct xfs_btree_cur	*cur,
247 	xfs_fsblock_t		bno)
248 {
249 	struct xfs_log_item_desc *lidp;
250 	int			i;
251 
252 	if (!cur)
253 		return NULL;
254 
255 	for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
256 		if (!cur->bc_bufs[i])
257 			break;
258 		if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
259 			return cur->bc_bufs[i];
260 	}
261 
262 	/* Chase down all the log items to see if the bp is there */
263 	list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
264 		struct xfs_buf_log_item	*bip;
265 		bip = (struct xfs_buf_log_item *)lidp->lid_item;
266 		if (bip->bli_item.li_type == XFS_LI_BUF &&
267 		    XFS_BUF_ADDR(bip->bli_buf) == bno)
268 			return bip->bli_buf;
269 	}
270 
271 	return NULL;
272 }
273 
274 STATIC void
275 xfs_check_block(
276 	struct xfs_btree_block	*block,
277 	xfs_mount_t		*mp,
278 	int			root,
279 	short			sz)
280 {
281 	int			i, j, dmxr;
282 	__be64			*pp, *thispa;	/* pointer to block address */
283 	xfs_bmbt_key_t		*prevp, *keyp;
284 
285 	ASSERT(be16_to_cpu(block->bb_level) > 0);
286 
287 	prevp = NULL;
288 	for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
289 		dmxr = mp->m_bmap_dmxr[0];
290 		keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
291 
292 		if (prevp) {
293 			ASSERT(be64_to_cpu(prevp->br_startoff) <
294 			       be64_to_cpu(keyp->br_startoff));
295 		}
296 		prevp = keyp;
297 
298 		/*
299 		 * Compare the block numbers to see if there are dups.
300 		 */
301 		if (root)
302 			pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
303 		else
304 			pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
305 
306 		for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
307 			if (root)
308 				thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
309 			else
310 				thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
311 			if (*thispa == *pp) {
312 				xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
313 					__func__, j, i,
314 					(unsigned long long)be64_to_cpu(*thispa));
315 				panic("%s: ptrs are equal in node\n",
316 					__func__);
317 			}
318 		}
319 	}
320 }
321 
322 /*
323  * Check that the extents for the inode ip are in the right order in all
324  * btree leaves. THis becomes prohibitively expensive for large extent count
325  * files, so don't bother with inodes that have more than 10,000 extents in
326  * them. The btree record ordering checks will still be done, so for such large
327  * bmapbt constructs that is going to catch most corruptions.
328  */
329 STATIC void
330 xfs_bmap_check_leaf_extents(
331 	xfs_btree_cur_t		*cur,	/* btree cursor or null */
332 	xfs_inode_t		*ip,		/* incore inode pointer */
333 	int			whichfork)	/* data or attr fork */
334 {
335 	struct xfs_btree_block	*block;	/* current btree block */
336 	xfs_fsblock_t		bno;	/* block # of "block" */
337 	xfs_buf_t		*bp;	/* buffer for "block" */
338 	int			error;	/* error return value */
339 	xfs_extnum_t		i=0, j;	/* index into the extents list */
340 	xfs_ifork_t		*ifp;	/* fork structure */
341 	int			level;	/* btree level, for checking */
342 	xfs_mount_t		*mp;	/* file system mount structure */
343 	__be64			*pp;	/* pointer to block address */
344 	xfs_bmbt_rec_t		*ep;	/* pointer to current extent */
345 	xfs_bmbt_rec_t		last = {0, 0}; /* last extent in prev block */
346 	xfs_bmbt_rec_t		*nextp;	/* pointer to next extent */
347 	int			bp_release = 0;
348 
349 	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
350 		return;
351 	}
352 
353 	/* skip large extent count inodes */
354 	if (ip->i_d.di_nextents > 10000)
355 		return;
356 
357 	bno = NULLFSBLOCK;
358 	mp = ip->i_mount;
359 	ifp = XFS_IFORK_PTR(ip, whichfork);
360 	block = ifp->if_broot;
361 	/*
362 	 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
363 	 */
364 	level = be16_to_cpu(block->bb_level);
365 	ASSERT(level > 0);
366 	xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
367 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
368 	bno = be64_to_cpu(*pp);
369 
370 	ASSERT(bno != NULLFSBLOCK);
371 	ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
372 	ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
373 
374 	/*
375 	 * Go down the tree until leaf level is reached, following the first
376 	 * pointer (leftmost) at each level.
377 	 */
378 	while (level-- > 0) {
379 		/* See if buf is in cur first */
380 		bp_release = 0;
381 		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
382 		if (!bp) {
383 			bp_release = 1;
384 			error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
385 						XFS_BMAP_BTREE_REF,
386 						&xfs_bmbt_buf_ops);
387 			if (error)
388 				goto error_norelse;
389 		}
390 		block = XFS_BUF_TO_BLOCK(bp);
391 		if (level == 0)
392 			break;
393 
394 		/*
395 		 * Check this block for basic sanity (increasing keys and
396 		 * no duplicate blocks).
397 		 */
398 
399 		xfs_check_block(block, mp, 0, 0);
400 		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
401 		bno = be64_to_cpu(*pp);
402 		XFS_WANT_CORRUPTED_GOTO(mp,
403 					xfs_verify_fsbno(mp, bno), error0);
404 		if (bp_release) {
405 			bp_release = 0;
406 			xfs_trans_brelse(NULL, bp);
407 		}
408 	}
409 
410 	/*
411 	 * Here with bp and block set to the leftmost leaf node in the tree.
412 	 */
413 	i = 0;
414 
415 	/*
416 	 * Loop over all leaf nodes checking that all extents are in the right order.
417 	 */
418 	for (;;) {
419 		xfs_fsblock_t	nextbno;
420 		xfs_extnum_t	num_recs;
421 
422 
423 		num_recs = xfs_btree_get_numrecs(block);
424 
425 		/*
426 		 * Read-ahead the next leaf block, if any.
427 		 */
428 
429 		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
430 
431 		/*
432 		 * Check all the extents to make sure they are OK.
433 		 * If we had a previous block, the last entry should
434 		 * conform with the first entry in this one.
435 		 */
436 
437 		ep = XFS_BMBT_REC_ADDR(mp, block, 1);
438 		if (i) {
439 			ASSERT(xfs_bmbt_disk_get_startoff(&last) +
440 			       xfs_bmbt_disk_get_blockcount(&last) <=
441 			       xfs_bmbt_disk_get_startoff(ep));
442 		}
443 		for (j = 1; j < num_recs; j++) {
444 			nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
445 			ASSERT(xfs_bmbt_disk_get_startoff(ep) +
446 			       xfs_bmbt_disk_get_blockcount(ep) <=
447 			       xfs_bmbt_disk_get_startoff(nextp));
448 			ep = nextp;
449 		}
450 
451 		last = *ep;
452 		i += num_recs;
453 		if (bp_release) {
454 			bp_release = 0;
455 			xfs_trans_brelse(NULL, bp);
456 		}
457 		bno = nextbno;
458 		/*
459 		 * If we've reached the end, stop.
460 		 */
461 		if (bno == NULLFSBLOCK)
462 			break;
463 
464 		bp_release = 0;
465 		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
466 		if (!bp) {
467 			bp_release = 1;
468 			error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
469 						XFS_BMAP_BTREE_REF,
470 						&xfs_bmbt_buf_ops);
471 			if (error)
472 				goto error_norelse;
473 		}
474 		block = XFS_BUF_TO_BLOCK(bp);
475 	}
476 
477 	return;
478 
479 error0:
480 	xfs_warn(mp, "%s: at error0", __func__);
481 	if (bp_release)
482 		xfs_trans_brelse(NULL, bp);
483 error_norelse:
484 	xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
485 		__func__, i);
486 	panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
487 	return;
488 }
489 
490 /*
491  * Validate that the bmbt_irecs being returned from bmapi are valid
492  * given the caller's original parameters.  Specifically check the
493  * ranges of the returned irecs to ensure that they only extend beyond
494  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
495  */
496 STATIC void
497 xfs_bmap_validate_ret(
498 	xfs_fileoff_t		bno,
499 	xfs_filblks_t		len,
500 	int			flags,
501 	xfs_bmbt_irec_t		*mval,
502 	int			nmap,
503 	int			ret_nmap)
504 {
505 	int			i;		/* index to map values */
506 
507 	ASSERT(ret_nmap <= nmap);
508 
509 	for (i = 0; i < ret_nmap; i++) {
510 		ASSERT(mval[i].br_blockcount > 0);
511 		if (!(flags & XFS_BMAPI_ENTIRE)) {
512 			ASSERT(mval[i].br_startoff >= bno);
513 			ASSERT(mval[i].br_blockcount <= len);
514 			ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
515 			       bno + len);
516 		} else {
517 			ASSERT(mval[i].br_startoff < bno + len);
518 			ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
519 			       bno);
520 		}
521 		ASSERT(i == 0 ||
522 		       mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
523 		       mval[i].br_startoff);
524 		ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
525 		       mval[i].br_startblock != HOLESTARTBLOCK);
526 		ASSERT(mval[i].br_state == XFS_EXT_NORM ||
527 		       mval[i].br_state == XFS_EXT_UNWRITTEN);
528 	}
529 }
530 
531 #else
532 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)		do { } while (0)
533 #define	xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)	do { } while (0)
534 #endif /* DEBUG */
535 
536 /*
537  * bmap free list manipulation functions
538  */
539 
540 /*
541  * Add the extent to the list of extents to be free at transaction end.
542  * The list is maintained sorted (by block number).
543  */
544 void
545 xfs_bmap_add_free(
546 	struct xfs_mount		*mp,
547 	struct xfs_defer_ops		*dfops,
548 	xfs_fsblock_t			bno,
549 	xfs_filblks_t			len,
550 	struct xfs_owner_info		*oinfo)
551 {
552 	struct xfs_extent_free_item	*new;		/* new element */
553 #ifdef DEBUG
554 	xfs_agnumber_t		agno;
555 	xfs_agblock_t		agbno;
556 
557 	ASSERT(bno != NULLFSBLOCK);
558 	ASSERT(len > 0);
559 	ASSERT(len <= MAXEXTLEN);
560 	ASSERT(!isnullstartblock(bno));
561 	agno = XFS_FSB_TO_AGNO(mp, bno);
562 	agbno = XFS_FSB_TO_AGBNO(mp, bno);
563 	ASSERT(agno < mp->m_sb.sb_agcount);
564 	ASSERT(agbno < mp->m_sb.sb_agblocks);
565 	ASSERT(len < mp->m_sb.sb_agblocks);
566 	ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
567 #endif
568 	ASSERT(xfs_bmap_free_item_zone != NULL);
569 
570 	new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
571 	new->xefi_startblock = bno;
572 	new->xefi_blockcount = (xfs_extlen_t)len;
573 	if (oinfo)
574 		new->xefi_oinfo = *oinfo;
575 	else
576 		xfs_rmap_skip_owner_update(&new->xefi_oinfo);
577 	trace_xfs_bmap_free_defer(mp, XFS_FSB_TO_AGNO(mp, bno), 0,
578 			XFS_FSB_TO_AGBNO(mp, bno), len);
579 	xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list);
580 }
581 
582 /*
583  * Inode fork format manipulation functions
584  */
585 
586 /*
587  * Transform a btree format file with only one leaf node, where the
588  * extents list will fit in the inode, into an extents format file.
589  * Since the file extents are already in-core, all we have to do is
590  * give up the space for the btree root and pitch the leaf block.
591  */
592 STATIC int				/* error */
593 xfs_bmap_btree_to_extents(
594 	xfs_trans_t		*tp,	/* transaction pointer */
595 	xfs_inode_t		*ip,	/* incore inode pointer */
596 	xfs_btree_cur_t		*cur,	/* btree cursor */
597 	int			*logflagsp, /* inode logging flags */
598 	int			whichfork)  /* data or attr fork */
599 {
600 	/* REFERENCED */
601 	struct xfs_btree_block	*cblock;/* child btree block */
602 	xfs_fsblock_t		cbno;	/* child block number */
603 	xfs_buf_t		*cbp;	/* child block's buffer */
604 	int			error;	/* error return value */
605 	xfs_ifork_t		*ifp;	/* inode fork data */
606 	xfs_mount_t		*mp;	/* mount point structure */
607 	__be64			*pp;	/* ptr to block address */
608 	struct xfs_btree_block	*rblock;/* root btree block */
609 	struct xfs_owner_info	oinfo;
610 
611 	mp = ip->i_mount;
612 	ifp = XFS_IFORK_PTR(ip, whichfork);
613 	ASSERT(whichfork != XFS_COW_FORK);
614 	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
615 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
616 	rblock = ifp->if_broot;
617 	ASSERT(be16_to_cpu(rblock->bb_level) == 1);
618 	ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
619 	ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
620 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
621 	cbno = be64_to_cpu(*pp);
622 	*logflagsp = 0;
623 #ifdef DEBUG
624 	XFS_WANT_CORRUPTED_RETURN(cur->bc_mp,
625 			xfs_btree_check_lptr(cur, cbno, 1));
626 #endif
627 	error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
628 				&xfs_bmbt_buf_ops);
629 	if (error)
630 		return error;
631 	cblock = XFS_BUF_TO_BLOCK(cbp);
632 	if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
633 		return error;
634 	xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
635 	xfs_bmap_add_free(mp, cur->bc_private.b.dfops, cbno, 1, &oinfo);
636 	ip->i_d.di_nblocks--;
637 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
638 	xfs_trans_binval(tp, cbp);
639 	if (cur->bc_bufs[0] == cbp)
640 		cur->bc_bufs[0] = NULL;
641 	xfs_iroot_realloc(ip, -1, whichfork);
642 	ASSERT(ifp->if_broot == NULL);
643 	ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
644 	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
645 	*logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
646 	return 0;
647 }
648 
649 /*
650  * Convert an extents-format file into a btree-format file.
651  * The new file will have a root block (in the inode) and a single child block.
652  */
653 STATIC int					/* error */
654 xfs_bmap_extents_to_btree(
655 	xfs_trans_t		*tp,		/* transaction pointer */
656 	xfs_inode_t		*ip,		/* incore inode pointer */
657 	xfs_fsblock_t		*firstblock,	/* first-block-allocated */
658 	struct xfs_defer_ops	*dfops,		/* blocks freed in xaction */
659 	xfs_btree_cur_t		**curp,		/* cursor returned to caller */
660 	int			wasdel,		/* converting a delayed alloc */
661 	int			*logflagsp,	/* inode logging flags */
662 	int			whichfork)	/* data or attr fork */
663 {
664 	struct xfs_btree_block	*ablock;	/* allocated (child) bt block */
665 	xfs_buf_t		*abp;		/* buffer for ablock */
666 	xfs_alloc_arg_t		args;		/* allocation arguments */
667 	xfs_bmbt_rec_t		*arp;		/* child record pointer */
668 	struct xfs_btree_block	*block;		/* btree root block */
669 	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
670 	int			error;		/* error return value */
671 	xfs_ifork_t		*ifp;		/* inode fork pointer */
672 	xfs_bmbt_key_t		*kp;		/* root block key pointer */
673 	xfs_mount_t		*mp;		/* mount structure */
674 	xfs_bmbt_ptr_t		*pp;		/* root block address pointer */
675 	struct xfs_iext_cursor	icur;
676 	struct xfs_bmbt_irec	rec;
677 	xfs_extnum_t		cnt = 0;
678 
679 	mp = ip->i_mount;
680 	ASSERT(whichfork != XFS_COW_FORK);
681 	ifp = XFS_IFORK_PTR(ip, whichfork);
682 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
683 
684 	/*
685 	 * Make space in the inode incore.
686 	 */
687 	xfs_iroot_realloc(ip, 1, whichfork);
688 	ifp->if_flags |= XFS_IFBROOT;
689 
690 	/*
691 	 * Fill in the root.
692 	 */
693 	block = ifp->if_broot;
694 	xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
695 				 XFS_BTNUM_BMAP, 1, 1, ip->i_ino,
696 				 XFS_BTREE_LONG_PTRS);
697 	/*
698 	 * Need a cursor.  Can't allocate until bb_level is filled in.
699 	 */
700 	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
701 	cur->bc_private.b.firstblock = *firstblock;
702 	cur->bc_private.b.dfops = dfops;
703 	cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
704 	/*
705 	 * Convert to a btree with two levels, one record in root.
706 	 */
707 	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
708 	memset(&args, 0, sizeof(args));
709 	args.tp = tp;
710 	args.mp = mp;
711 	xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork);
712 	args.firstblock = *firstblock;
713 	if (*firstblock == NULLFSBLOCK) {
714 		args.type = XFS_ALLOCTYPE_START_BNO;
715 		args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
716 	} else if (dfops->dop_low) {
717 		args.type = XFS_ALLOCTYPE_START_BNO;
718 		args.fsbno = *firstblock;
719 	} else {
720 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
721 		args.fsbno = *firstblock;
722 	}
723 	args.minlen = args.maxlen = args.prod = 1;
724 	args.wasdel = wasdel;
725 	*logflagsp = 0;
726 	if ((error = xfs_alloc_vextent(&args))) {
727 		xfs_iroot_realloc(ip, -1, whichfork);
728 		xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
729 		return error;
730 	}
731 
732 	if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
733 		xfs_iroot_realloc(ip, -1, whichfork);
734 		xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
735 		return -ENOSPC;
736 	}
737 	/*
738 	 * Allocation can't fail, the space was reserved.
739 	 */
740 	ASSERT(*firstblock == NULLFSBLOCK ||
741 	       args.agno >= XFS_FSB_TO_AGNO(mp, *firstblock));
742 	*firstblock = cur->bc_private.b.firstblock = args.fsbno;
743 	cur->bc_private.b.allocated++;
744 	ip->i_d.di_nblocks++;
745 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
746 	abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
747 	/*
748 	 * Fill in the child block.
749 	 */
750 	abp->b_ops = &xfs_bmbt_buf_ops;
751 	ablock = XFS_BUF_TO_BLOCK(abp);
752 	xfs_btree_init_block_int(mp, ablock, abp->b_bn,
753 				XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
754 				XFS_BTREE_LONG_PTRS);
755 
756 	for_each_xfs_iext(ifp, &icur, &rec) {
757 		if (isnullstartblock(rec.br_startblock))
758 			continue;
759 		arp = XFS_BMBT_REC_ADDR(mp, ablock, 1 + cnt);
760 		xfs_bmbt_disk_set_all(arp, &rec);
761 		cnt++;
762 	}
763 	ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
764 	xfs_btree_set_numrecs(ablock, cnt);
765 
766 	/*
767 	 * Fill in the root key and pointer.
768 	 */
769 	kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
770 	arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
771 	kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
772 	pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
773 						be16_to_cpu(block->bb_level)));
774 	*pp = cpu_to_be64(args.fsbno);
775 
776 	/*
777 	 * Do all this logging at the end so that
778 	 * the root is at the right level.
779 	 */
780 	xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
781 	xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
782 	ASSERT(*curp == NULL);
783 	*curp = cur;
784 	*logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
785 	return 0;
786 }
787 
788 /*
789  * Convert a local file to an extents file.
790  * This code is out of bounds for data forks of regular files,
791  * since the file data needs to get logged so things will stay consistent.
792  * (The bmap-level manipulations are ok, though).
793  */
794 void
795 xfs_bmap_local_to_extents_empty(
796 	struct xfs_inode	*ip,
797 	int			whichfork)
798 {
799 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
800 
801 	ASSERT(whichfork != XFS_COW_FORK);
802 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
803 	ASSERT(ifp->if_bytes == 0);
804 	ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
805 
806 	xfs_bmap_forkoff_reset(ip, whichfork);
807 	ifp->if_flags &= ~XFS_IFINLINE;
808 	ifp->if_flags |= XFS_IFEXTENTS;
809 	ifp->if_u1.if_root = NULL;
810 	ifp->if_height = 0;
811 	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
812 }
813 
814 
815 STATIC int				/* error */
816 xfs_bmap_local_to_extents(
817 	xfs_trans_t	*tp,		/* transaction pointer */
818 	xfs_inode_t	*ip,		/* incore inode pointer */
819 	xfs_fsblock_t	*firstblock,	/* first block allocated in xaction */
820 	xfs_extlen_t	total,		/* total blocks needed by transaction */
821 	int		*logflagsp,	/* inode logging flags */
822 	int		whichfork,
823 	void		(*init_fn)(struct xfs_trans *tp,
824 				   struct xfs_buf *bp,
825 				   struct xfs_inode *ip,
826 				   struct xfs_ifork *ifp))
827 {
828 	int		error = 0;
829 	int		flags;		/* logging flags returned */
830 	xfs_ifork_t	*ifp;		/* inode fork pointer */
831 	xfs_alloc_arg_t	args;		/* allocation arguments */
832 	xfs_buf_t	*bp;		/* buffer for extent block */
833 	struct xfs_bmbt_irec rec;
834 	struct xfs_iext_cursor icur;
835 
836 	/*
837 	 * We don't want to deal with the case of keeping inode data inline yet.
838 	 * So sending the data fork of a regular inode is invalid.
839 	 */
840 	ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
841 	ifp = XFS_IFORK_PTR(ip, whichfork);
842 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
843 
844 	if (!ifp->if_bytes) {
845 		xfs_bmap_local_to_extents_empty(ip, whichfork);
846 		flags = XFS_ILOG_CORE;
847 		goto done;
848 	}
849 
850 	flags = 0;
851 	error = 0;
852 	ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS)) == XFS_IFINLINE);
853 	memset(&args, 0, sizeof(args));
854 	args.tp = tp;
855 	args.mp = ip->i_mount;
856 	xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0);
857 	args.firstblock = *firstblock;
858 	/*
859 	 * Allocate a block.  We know we need only one, since the
860 	 * file currently fits in an inode.
861 	 */
862 	if (*firstblock == NULLFSBLOCK) {
863 		args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
864 		args.type = XFS_ALLOCTYPE_START_BNO;
865 	} else {
866 		args.fsbno = *firstblock;
867 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
868 	}
869 	args.total = total;
870 	args.minlen = args.maxlen = args.prod = 1;
871 	error = xfs_alloc_vextent(&args);
872 	if (error)
873 		goto done;
874 
875 	/* Can't fail, the space was reserved. */
876 	ASSERT(args.fsbno != NULLFSBLOCK);
877 	ASSERT(args.len == 1);
878 	*firstblock = args.fsbno;
879 	bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
880 
881 	/*
882 	 * Initialize the block, copy the data and log the remote buffer.
883 	 *
884 	 * The callout is responsible for logging because the remote format
885 	 * might differ from the local format and thus we don't know how much to
886 	 * log here. Note that init_fn must also set the buffer log item type
887 	 * correctly.
888 	 */
889 	init_fn(tp, bp, ip, ifp);
890 
891 	/* account for the change in fork size */
892 	xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
893 	xfs_bmap_local_to_extents_empty(ip, whichfork);
894 	flags |= XFS_ILOG_CORE;
895 
896 	ifp->if_u1.if_root = NULL;
897 	ifp->if_height = 0;
898 
899 	rec.br_startoff = 0;
900 	rec.br_startblock = args.fsbno;
901 	rec.br_blockcount = 1;
902 	rec.br_state = XFS_EXT_NORM;
903 	xfs_iext_first(ifp, &icur);
904 	xfs_iext_insert(ip, &icur, &rec, 0);
905 
906 	XFS_IFORK_NEXT_SET(ip, whichfork, 1);
907 	ip->i_d.di_nblocks = 1;
908 	xfs_trans_mod_dquot_byino(tp, ip,
909 		XFS_TRANS_DQ_BCOUNT, 1L);
910 	flags |= xfs_ilog_fext(whichfork);
911 
912 done:
913 	*logflagsp = flags;
914 	return error;
915 }
916 
917 /*
918  * Called from xfs_bmap_add_attrfork to handle btree format files.
919  */
920 STATIC int					/* error */
921 xfs_bmap_add_attrfork_btree(
922 	xfs_trans_t		*tp,		/* transaction pointer */
923 	xfs_inode_t		*ip,		/* incore inode pointer */
924 	xfs_fsblock_t		*firstblock,	/* first block allocated */
925 	struct xfs_defer_ops	*dfops,		/* blocks to free at commit */
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 		cur->bc_private.b.dfops = dfops;
939 		cur->bc_private.b.firstblock = *firstblock;
940 		error = xfs_bmbt_lookup_first(cur, &stat);
941 		if (error)
942 			goto error0;
943 		/* must be at least one entry */
944 		XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0);
945 		if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
946 			goto error0;
947 		if (stat == 0) {
948 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
949 			return -ENOSPC;
950 		}
951 		*firstblock = cur->bc_private.b.firstblock;
952 		cur->bc_private.b.allocated = 0;
953 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
954 	}
955 	return 0;
956 error0:
957 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
958 	return error;
959 }
960 
961 /*
962  * Called from xfs_bmap_add_attrfork to handle extents format files.
963  */
964 STATIC int					/* error */
965 xfs_bmap_add_attrfork_extents(
966 	xfs_trans_t		*tp,		/* transaction pointer */
967 	xfs_inode_t		*ip,		/* incore inode pointer */
968 	xfs_fsblock_t		*firstblock,	/* first block allocated */
969 	struct xfs_defer_ops	*dfops,		/* blocks to free at commit */
970 	int			*flags)		/* inode logging flags */
971 {
972 	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
973 	int			error;		/* error return value */
974 
975 	if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
976 		return 0;
977 	cur = NULL;
978 	error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops, &cur, 0,
979 		flags, XFS_DATA_FORK);
980 	if (cur) {
981 		cur->bc_private.b.allocated = 0;
982 		xfs_btree_del_cursor(cur,
983 			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
984 	}
985 	return error;
986 }
987 
988 /*
989  * Called from xfs_bmap_add_attrfork to handle local format files. Each
990  * different data fork content type needs a different callout to do the
991  * conversion. Some are basic and only require special block initialisation
992  * callouts for the data formating, others (directories) are so specialised they
993  * handle everything themselves.
994  *
995  * XXX (dgc): investigate whether directory conversion can use the generic
996  * formatting callout. It should be possible - it's just a very complex
997  * formatter.
998  */
999 STATIC int					/* error */
1000 xfs_bmap_add_attrfork_local(
1001 	xfs_trans_t		*tp,		/* transaction pointer */
1002 	xfs_inode_t		*ip,		/* incore inode pointer */
1003 	xfs_fsblock_t		*firstblock,	/* first block allocated */
1004 	struct xfs_defer_ops	*dfops,		/* blocks to free at commit */
1005 	int			*flags)		/* inode logging flags */
1006 {
1007 	xfs_da_args_t		dargs;		/* args for dir/attr code */
1008 
1009 	if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1010 		return 0;
1011 
1012 	if (S_ISDIR(VFS_I(ip)->i_mode)) {
1013 		memset(&dargs, 0, sizeof(dargs));
1014 		dargs.geo = ip->i_mount->m_dir_geo;
1015 		dargs.dp = ip;
1016 		dargs.firstblock = firstblock;
1017 		dargs.dfops = dfops;
1018 		dargs.total = dargs.geo->fsbcount;
1019 		dargs.whichfork = XFS_DATA_FORK;
1020 		dargs.trans = tp;
1021 		return xfs_dir2_sf_to_block(&dargs);
1022 	}
1023 
1024 	if (S_ISLNK(VFS_I(ip)->i_mode))
1025 		return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1026 						 flags, XFS_DATA_FORK,
1027 						 xfs_symlink_local_to_remote);
1028 
1029 	/* should only be called for types that support local format data */
1030 	ASSERT(0);
1031 	return -EFSCORRUPTED;
1032 }
1033 
1034 /*
1035  * Convert inode from non-attributed to attributed.
1036  * Must not be in a transaction, ip must not be locked.
1037  */
1038 int						/* error code */
1039 xfs_bmap_add_attrfork(
1040 	xfs_inode_t		*ip,		/* incore inode pointer */
1041 	int			size,		/* space new attribute needs */
1042 	int			rsvd)		/* xact may use reserved blks */
1043 {
1044 	xfs_fsblock_t		firstblock;	/* 1st block/ag allocated */
1045 	struct xfs_defer_ops	dfops;		/* freed extent records */
1046 	xfs_mount_t		*mp;		/* mount structure */
1047 	xfs_trans_t		*tp;		/* transaction pointer */
1048 	int			blks;		/* space reservation */
1049 	int			version = 1;	/* superblock attr version */
1050 	int			logflags;	/* logging flags */
1051 	int			error;		/* error return value */
1052 
1053 	ASSERT(XFS_IFORK_Q(ip) == 0);
1054 
1055 	mp = ip->i_mount;
1056 	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1057 
1058 	blks = XFS_ADDAFORK_SPACE_RES(mp);
1059 
1060 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_addafork, blks, 0,
1061 			rsvd ? XFS_TRANS_RESERVE : 0, &tp);
1062 	if (error)
1063 		return error;
1064 
1065 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1066 	error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1067 			XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1068 			XFS_QMOPT_RES_REGBLKS);
1069 	if (error)
1070 		goto trans_cancel;
1071 	if (XFS_IFORK_Q(ip))
1072 		goto trans_cancel;
1073 	if (ip->i_d.di_anextents != 0) {
1074 		error = -EFSCORRUPTED;
1075 		goto trans_cancel;
1076 	}
1077 	if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1078 		/*
1079 		 * For inodes coming from pre-6.2 filesystems.
1080 		 */
1081 		ASSERT(ip->i_d.di_aformat == 0);
1082 		ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1083 	}
1084 
1085 	xfs_trans_ijoin(tp, ip, 0);
1086 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1087 
1088 	switch (ip->i_d.di_format) {
1089 	case XFS_DINODE_FMT_DEV:
1090 		ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1091 		break;
1092 	case XFS_DINODE_FMT_LOCAL:
1093 	case XFS_DINODE_FMT_EXTENTS:
1094 	case XFS_DINODE_FMT_BTREE:
1095 		ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1096 		if (!ip->i_d.di_forkoff)
1097 			ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1098 		else if (mp->m_flags & XFS_MOUNT_ATTR2)
1099 			version = 2;
1100 		break;
1101 	default:
1102 		ASSERT(0);
1103 		error = -EINVAL;
1104 		goto trans_cancel;
1105 	}
1106 
1107 	ASSERT(ip->i_afp == NULL);
1108 	ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1109 	ip->i_afp->if_flags = XFS_IFEXTENTS;
1110 	logflags = 0;
1111 	xfs_defer_init(&dfops, &firstblock);
1112 	switch (ip->i_d.di_format) {
1113 	case XFS_DINODE_FMT_LOCAL:
1114 		error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &dfops,
1115 			&logflags);
1116 		break;
1117 	case XFS_DINODE_FMT_EXTENTS:
1118 		error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1119 			&dfops, &logflags);
1120 		break;
1121 	case XFS_DINODE_FMT_BTREE:
1122 		error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &dfops,
1123 			&logflags);
1124 		break;
1125 	default:
1126 		error = 0;
1127 		break;
1128 	}
1129 	if (logflags)
1130 		xfs_trans_log_inode(tp, ip, logflags);
1131 	if (error)
1132 		goto bmap_cancel;
1133 	if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1134 	   (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1135 		bool log_sb = false;
1136 
1137 		spin_lock(&mp->m_sb_lock);
1138 		if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1139 			xfs_sb_version_addattr(&mp->m_sb);
1140 			log_sb = true;
1141 		}
1142 		if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1143 			xfs_sb_version_addattr2(&mp->m_sb);
1144 			log_sb = true;
1145 		}
1146 		spin_unlock(&mp->m_sb_lock);
1147 		if (log_sb)
1148 			xfs_log_sb(tp);
1149 	}
1150 
1151 	error = xfs_defer_finish(&tp, &dfops);
1152 	if (error)
1153 		goto bmap_cancel;
1154 	error = xfs_trans_commit(tp);
1155 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1156 	return error;
1157 
1158 bmap_cancel:
1159 	xfs_defer_cancel(&dfops);
1160 trans_cancel:
1161 	xfs_trans_cancel(tp);
1162 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1163 	return error;
1164 }
1165 
1166 /*
1167  * Internal and external extent tree search functions.
1168  */
1169 
1170 /*
1171  * Read in extents from a btree-format inode.
1172  */
1173 int
1174 xfs_iread_extents(
1175 	struct xfs_trans	*tp,
1176 	struct xfs_inode	*ip,
1177 	int			whichfork)
1178 {
1179 	struct xfs_mount	*mp = ip->i_mount;
1180 	int			state = xfs_bmap_fork_to_state(whichfork);
1181 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1182 	xfs_extnum_t		nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
1183 	struct xfs_btree_block	*block = ifp->if_broot;
1184 	struct xfs_iext_cursor	icur;
1185 	struct xfs_bmbt_irec	new;
1186 	xfs_fsblock_t		bno;
1187 	struct xfs_buf		*bp;
1188 	xfs_extnum_t		i, j;
1189 	int			level;
1190 	__be64			*pp;
1191 	int			error;
1192 
1193 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1194 
1195 	if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1196 		XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
1197 		return -EFSCORRUPTED;
1198 	}
1199 
1200 	/*
1201 	 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1202 	 */
1203 	level = be16_to_cpu(block->bb_level);
1204 	ASSERT(level > 0);
1205 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1206 	bno = be64_to_cpu(*pp);
1207 
1208 	/*
1209 	 * Go down the tree until leaf level is reached, following the first
1210 	 * pointer (leftmost) at each level.
1211 	 */
1212 	while (level-- > 0) {
1213 		error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1214 				XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1215 		if (error)
1216 			goto out;
1217 		block = XFS_BUF_TO_BLOCK(bp);
1218 		if (level == 0)
1219 			break;
1220 		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1221 		bno = be64_to_cpu(*pp);
1222 		XFS_WANT_CORRUPTED_GOTO(mp,
1223 			xfs_verify_fsbno(mp, bno), out_brelse);
1224 		xfs_trans_brelse(tp, bp);
1225 	}
1226 
1227 	/*
1228 	 * Here with bp and block set to the leftmost leaf node in the tree.
1229 	 */
1230 	i = 0;
1231 	xfs_iext_first(ifp, &icur);
1232 
1233 	/*
1234 	 * Loop over all leaf nodes.  Copy information to the extent records.
1235 	 */
1236 	for (;;) {
1237 		xfs_bmbt_rec_t	*frp;
1238 		xfs_fsblock_t	nextbno;
1239 		xfs_extnum_t	num_recs;
1240 
1241 		num_recs = xfs_btree_get_numrecs(block);
1242 		if (unlikely(i + num_recs > nextents)) {
1243 			ASSERT(i + num_recs <= nextents);
1244 			xfs_warn(ip->i_mount,
1245 				"corrupt dinode %Lu, (btree extents).",
1246 				(unsigned long long) ip->i_ino);
1247 			xfs_inode_verifier_error(ip, -EFSCORRUPTED,
1248 					__func__, block, sizeof(*block),
1249 					__this_address);
1250 			error = -EFSCORRUPTED;
1251 			goto out_brelse;
1252 		}
1253 		/*
1254 		 * Read-ahead the next leaf block, if any.
1255 		 */
1256 		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1257 		if (nextbno != NULLFSBLOCK)
1258 			xfs_btree_reada_bufl(mp, nextbno, 1,
1259 					     &xfs_bmbt_buf_ops);
1260 		/*
1261 		 * Copy records into the extent records.
1262 		 */
1263 		frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1264 		for (j = 0; j < num_recs; j++, frp++, i++) {
1265 			xfs_failaddr_t	fa;
1266 
1267 			xfs_bmbt_disk_get_all(frp, &new);
1268 			fa = xfs_bmap_validate_extent(ip, whichfork, &new);
1269 			if (fa) {
1270 				error = -EFSCORRUPTED;
1271 				xfs_inode_verifier_error(ip, error,
1272 						"xfs_iread_extents(2)",
1273 						frp, sizeof(*frp), fa);
1274 				goto out_brelse;
1275 			}
1276 			xfs_iext_insert(ip, &icur, &new, state);
1277 			trace_xfs_read_extent(ip, &icur, state, _THIS_IP_);
1278 			xfs_iext_next(ifp, &icur);
1279 		}
1280 		xfs_trans_brelse(tp, bp);
1281 		bno = nextbno;
1282 		/*
1283 		 * If we've reached the end, stop.
1284 		 */
1285 		if (bno == NULLFSBLOCK)
1286 			break;
1287 		error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1288 				XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1289 		if (error)
1290 			goto out;
1291 		block = XFS_BUF_TO_BLOCK(bp);
1292 	}
1293 
1294 	if (i != XFS_IFORK_NEXTENTS(ip, whichfork)) {
1295 		error = -EFSCORRUPTED;
1296 		goto out;
1297 	}
1298 	ASSERT(i == xfs_iext_count(ifp));
1299 
1300 	ifp->if_flags |= XFS_IFEXTENTS;
1301 	return 0;
1302 
1303 out_brelse:
1304 	xfs_trans_brelse(tp, bp);
1305 out:
1306 	xfs_iext_destroy(ifp);
1307 	return error;
1308 }
1309 
1310 /*
1311  * Returns the relative block number of the first unused block(s) in the given
1312  * fork with at least "len" logically contiguous blocks free.  This is the
1313  * lowest-address hole if the fork has holes, else the first block past the end
1314  * of fork.  Return 0 if the fork is currently local (in-inode).
1315  */
1316 int						/* error */
1317 xfs_bmap_first_unused(
1318 	struct xfs_trans	*tp,		/* transaction pointer */
1319 	struct xfs_inode	*ip,		/* incore inode */
1320 	xfs_extlen_t		len,		/* size of hole to find */
1321 	xfs_fileoff_t		*first_unused,	/* unused block */
1322 	int			whichfork)	/* data or attr fork */
1323 {
1324 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1325 	struct xfs_bmbt_irec	got;
1326 	struct xfs_iext_cursor	icur;
1327 	xfs_fileoff_t		lastaddr = 0;
1328 	xfs_fileoff_t		lowest, max;
1329 	int			error;
1330 
1331 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1332 	       XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1333 	       XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1334 
1335 	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1336 		*first_unused = 0;
1337 		return 0;
1338 	}
1339 
1340 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1341 		error = xfs_iread_extents(tp, ip, whichfork);
1342 		if (error)
1343 			return error;
1344 	}
1345 
1346 	lowest = max = *first_unused;
1347 	for_each_xfs_iext(ifp, &icur, &got) {
1348 		/*
1349 		 * See if the hole before this extent will work.
1350 		 */
1351 		if (got.br_startoff >= lowest + len &&
1352 		    got.br_startoff - max >= len)
1353 			break;
1354 		lastaddr = got.br_startoff + got.br_blockcount;
1355 		max = XFS_FILEOFF_MAX(lastaddr, lowest);
1356 	}
1357 
1358 	*first_unused = max;
1359 	return 0;
1360 }
1361 
1362 /*
1363  * Returns the file-relative block number of the last block - 1 before
1364  * last_block (input value) in the file.
1365  * This is not based on i_size, it is based on the extent records.
1366  * Returns 0 for local files, as they do not have extent records.
1367  */
1368 int						/* error */
1369 xfs_bmap_last_before(
1370 	struct xfs_trans	*tp,		/* transaction pointer */
1371 	struct xfs_inode	*ip,		/* incore inode */
1372 	xfs_fileoff_t		*last_block,	/* last block */
1373 	int			whichfork)	/* data or attr fork */
1374 {
1375 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1376 	struct xfs_bmbt_irec	got;
1377 	struct xfs_iext_cursor	icur;
1378 	int			error;
1379 
1380 	switch (XFS_IFORK_FORMAT(ip, whichfork)) {
1381 	case XFS_DINODE_FMT_LOCAL:
1382 		*last_block = 0;
1383 		return 0;
1384 	case XFS_DINODE_FMT_BTREE:
1385 	case XFS_DINODE_FMT_EXTENTS:
1386 		break;
1387 	default:
1388 		return -EIO;
1389 	}
1390 
1391 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1392 		error = xfs_iread_extents(tp, ip, whichfork);
1393 		if (error)
1394 			return error;
1395 	}
1396 
1397 	if (!xfs_iext_lookup_extent_before(ip, ifp, last_block, &icur, &got))
1398 		*last_block = 0;
1399 	return 0;
1400 }
1401 
1402 int
1403 xfs_bmap_last_extent(
1404 	struct xfs_trans	*tp,
1405 	struct xfs_inode	*ip,
1406 	int			whichfork,
1407 	struct xfs_bmbt_irec	*rec,
1408 	int			*is_empty)
1409 {
1410 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1411 	struct xfs_iext_cursor	icur;
1412 	int			error;
1413 
1414 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1415 		error = xfs_iread_extents(tp, ip, whichfork);
1416 		if (error)
1417 			return error;
1418 	}
1419 
1420 	xfs_iext_last(ifp, &icur);
1421 	if (!xfs_iext_get_extent(ifp, &icur, rec))
1422 		*is_empty = 1;
1423 	else
1424 		*is_empty = 0;
1425 	return 0;
1426 }
1427 
1428 /*
1429  * Check the last inode extent to determine whether this allocation will result
1430  * in blocks being allocated at the end of the file. When we allocate new data
1431  * blocks at the end of the file which do not start at the previous data block,
1432  * we will try to align the new blocks at stripe unit boundaries.
1433  *
1434  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1435  * at, or past the EOF.
1436  */
1437 STATIC int
1438 xfs_bmap_isaeof(
1439 	struct xfs_bmalloca	*bma,
1440 	int			whichfork)
1441 {
1442 	struct xfs_bmbt_irec	rec;
1443 	int			is_empty;
1444 	int			error;
1445 
1446 	bma->aeof = false;
1447 	error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1448 				     &is_empty);
1449 	if (error)
1450 		return error;
1451 
1452 	if (is_empty) {
1453 		bma->aeof = true;
1454 		return 0;
1455 	}
1456 
1457 	/*
1458 	 * Check if we are allocation or past the last extent, or at least into
1459 	 * the last delayed allocated extent.
1460 	 */
1461 	bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1462 		(bma->offset >= rec.br_startoff &&
1463 		 isnullstartblock(rec.br_startblock));
1464 	return 0;
1465 }
1466 
1467 /*
1468  * Returns the file-relative block number of the first block past eof in
1469  * the file.  This is not based on i_size, it is based on the extent records.
1470  * Returns 0 for local files, as they do not have extent records.
1471  */
1472 int
1473 xfs_bmap_last_offset(
1474 	struct xfs_inode	*ip,
1475 	xfs_fileoff_t		*last_block,
1476 	int			whichfork)
1477 {
1478 	struct xfs_bmbt_irec	rec;
1479 	int			is_empty;
1480 	int			error;
1481 
1482 	*last_block = 0;
1483 
1484 	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1485 		return 0;
1486 
1487 	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1488 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1489 	       return -EIO;
1490 
1491 	error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1492 	if (error || is_empty)
1493 		return error;
1494 
1495 	*last_block = rec.br_startoff + rec.br_blockcount;
1496 	return 0;
1497 }
1498 
1499 /*
1500  * Returns whether the selected fork of the inode has exactly one
1501  * block or not.  For the data fork we check this matches di_size,
1502  * implying the file's range is 0..bsize-1.
1503  */
1504 int					/* 1=>1 block, 0=>otherwise */
1505 xfs_bmap_one_block(
1506 	xfs_inode_t	*ip,		/* incore inode */
1507 	int		whichfork)	/* data or attr fork */
1508 {
1509 	xfs_ifork_t	*ifp;		/* inode fork pointer */
1510 	int		rval;		/* return value */
1511 	xfs_bmbt_irec_t	s;		/* internal version of extent */
1512 	struct xfs_iext_cursor icur;
1513 
1514 #ifndef DEBUG
1515 	if (whichfork == XFS_DATA_FORK)
1516 		return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1517 #endif	/* !DEBUG */
1518 	if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1519 		return 0;
1520 	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1521 		return 0;
1522 	ifp = XFS_IFORK_PTR(ip, whichfork);
1523 	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1524 	xfs_iext_first(ifp, &icur);
1525 	xfs_iext_get_extent(ifp, &icur, &s);
1526 	rval = s.br_startoff == 0 && s.br_blockcount == 1;
1527 	if (rval && whichfork == XFS_DATA_FORK)
1528 		ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1529 	return rval;
1530 }
1531 
1532 /*
1533  * Extent tree manipulation functions used during allocation.
1534  */
1535 
1536 /*
1537  * Convert a delayed allocation to a real allocation.
1538  */
1539 STATIC int				/* error */
1540 xfs_bmap_add_extent_delay_real(
1541 	struct xfs_bmalloca	*bma,
1542 	int			whichfork)
1543 {
1544 	struct xfs_bmbt_irec	*new = &bma->got;
1545 	int			error;	/* error return value */
1546 	int			i;	/* temp state */
1547 	xfs_ifork_t		*ifp;	/* inode fork pointer */
1548 	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
1549 	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
1550 					/* left is 0, right is 1, prev is 2 */
1551 	int			rval=0;	/* return value (logging flags) */
1552 	int			state = xfs_bmap_fork_to_state(whichfork);
1553 	xfs_filblks_t		da_new; /* new count del alloc blocks used */
1554 	xfs_filblks_t		da_old; /* old count del alloc blocks used */
1555 	xfs_filblks_t		temp=0;	/* value for da_new calculations */
1556 	int			tmp_rval;	/* partial logging flags */
1557 	struct xfs_mount	*mp;
1558 	xfs_extnum_t		*nextents;
1559 	struct xfs_bmbt_irec	old;
1560 
1561 	mp = bma->ip->i_mount;
1562 	ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1563 	ASSERT(whichfork != XFS_ATTR_FORK);
1564 	nextents = (whichfork == XFS_COW_FORK ? &bma->ip->i_cnextents :
1565 						&bma->ip->i_d.di_nextents);
1566 
1567 	ASSERT(!isnullstartblock(new->br_startblock));
1568 	ASSERT(!bma->cur ||
1569 	       (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1570 
1571 	XFS_STATS_INC(mp, xs_add_exlist);
1572 
1573 #define	LEFT		r[0]
1574 #define	RIGHT		r[1]
1575 #define	PREV		r[2]
1576 
1577 	/*
1578 	 * Set up a bunch of variables to make the tests simpler.
1579 	 */
1580 	xfs_iext_get_extent(ifp, &bma->icur, &PREV);
1581 	new_endoff = new->br_startoff + new->br_blockcount;
1582 	ASSERT(isnullstartblock(PREV.br_startblock));
1583 	ASSERT(PREV.br_startoff <= new->br_startoff);
1584 	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1585 
1586 	da_old = startblockval(PREV.br_startblock);
1587 	da_new = 0;
1588 
1589 	/*
1590 	 * Set flags determining what part of the previous delayed allocation
1591 	 * extent is being replaced by a real allocation.
1592 	 */
1593 	if (PREV.br_startoff == new->br_startoff)
1594 		state |= BMAP_LEFT_FILLING;
1595 	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1596 		state |= BMAP_RIGHT_FILLING;
1597 
1598 	/*
1599 	 * Check and set flags if this segment has a left neighbor.
1600 	 * Don't set contiguous if the combined extent would be too large.
1601 	 */
1602 	if (xfs_iext_peek_prev_extent(ifp, &bma->icur, &LEFT)) {
1603 		state |= BMAP_LEFT_VALID;
1604 		if (isnullstartblock(LEFT.br_startblock))
1605 			state |= BMAP_LEFT_DELAY;
1606 	}
1607 
1608 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1609 	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1610 	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1611 	    LEFT.br_state == new->br_state &&
1612 	    LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1613 		state |= BMAP_LEFT_CONTIG;
1614 
1615 	/*
1616 	 * Check and set flags if this segment has a right neighbor.
1617 	 * Don't set contiguous if the combined extent would be too large.
1618 	 * Also check for all-three-contiguous being too large.
1619 	 */
1620 	if (xfs_iext_peek_next_extent(ifp, &bma->icur, &RIGHT)) {
1621 		state |= BMAP_RIGHT_VALID;
1622 		if (isnullstartblock(RIGHT.br_startblock))
1623 			state |= BMAP_RIGHT_DELAY;
1624 	}
1625 
1626 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1627 	    new_endoff == RIGHT.br_startoff &&
1628 	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1629 	    new->br_state == RIGHT.br_state &&
1630 	    new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1631 	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1632 		       BMAP_RIGHT_FILLING)) !=
1633 		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1634 		       BMAP_RIGHT_FILLING) ||
1635 	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1636 			<= MAXEXTLEN))
1637 		state |= BMAP_RIGHT_CONTIG;
1638 
1639 	error = 0;
1640 	/*
1641 	 * Switch out based on the FILLING and CONTIG state bits.
1642 	 */
1643 	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1644 			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1645 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1646 	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1647 		/*
1648 		 * Filling in all of a previously delayed allocation extent.
1649 		 * The left and right neighbors are both contiguous with new.
1650 		 */
1651 		LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
1652 
1653 		xfs_iext_remove(bma->ip, &bma->icur, state);
1654 		xfs_iext_remove(bma->ip, &bma->icur, state);
1655 		xfs_iext_prev(ifp, &bma->icur);
1656 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1657 		(*nextents)--;
1658 
1659 		if (bma->cur == NULL)
1660 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1661 		else {
1662 			rval = XFS_ILOG_CORE;
1663 			error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1664 			if (error)
1665 				goto done;
1666 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1667 			error = xfs_btree_delete(bma->cur, &i);
1668 			if (error)
1669 				goto done;
1670 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1671 			error = xfs_btree_decrement(bma->cur, 0, &i);
1672 			if (error)
1673 				goto done;
1674 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1675 			error = xfs_bmbt_update(bma->cur, &LEFT);
1676 			if (error)
1677 				goto done;
1678 		}
1679 		break;
1680 
1681 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1682 		/*
1683 		 * Filling in all of a previously delayed allocation extent.
1684 		 * The left neighbor is contiguous, the right is not.
1685 		 */
1686 		old = LEFT;
1687 		LEFT.br_blockcount += PREV.br_blockcount;
1688 
1689 		xfs_iext_remove(bma->ip, &bma->icur, state);
1690 		xfs_iext_prev(ifp, &bma->icur);
1691 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1692 
1693 		if (bma->cur == NULL)
1694 			rval = XFS_ILOG_DEXT;
1695 		else {
1696 			rval = 0;
1697 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1698 			if (error)
1699 				goto done;
1700 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1701 			error = xfs_bmbt_update(bma->cur, &LEFT);
1702 			if (error)
1703 				goto done;
1704 		}
1705 		break;
1706 
1707 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1708 		/*
1709 		 * Filling in all of a previously delayed allocation extent.
1710 		 * The right neighbor is contiguous, the left is not.
1711 		 */
1712 		PREV.br_startblock = new->br_startblock;
1713 		PREV.br_blockcount += RIGHT.br_blockcount;
1714 
1715 		xfs_iext_next(ifp, &bma->icur);
1716 		xfs_iext_remove(bma->ip, &bma->icur, state);
1717 		xfs_iext_prev(ifp, &bma->icur);
1718 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1719 
1720 		if (bma->cur == NULL)
1721 			rval = XFS_ILOG_DEXT;
1722 		else {
1723 			rval = 0;
1724 			error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1725 			if (error)
1726 				goto done;
1727 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1728 			error = xfs_bmbt_update(bma->cur, &PREV);
1729 			if (error)
1730 				goto done;
1731 		}
1732 		break;
1733 
1734 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1735 		/*
1736 		 * Filling in all of a previously delayed allocation extent.
1737 		 * Neither the left nor right neighbors are contiguous with
1738 		 * the new one.
1739 		 */
1740 		PREV.br_startblock = new->br_startblock;
1741 		PREV.br_state = new->br_state;
1742 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1743 
1744 		(*nextents)++;
1745 		if (bma->cur == NULL)
1746 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1747 		else {
1748 			rval = XFS_ILOG_CORE;
1749 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1750 			if (error)
1751 				goto done;
1752 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1753 			error = xfs_btree_insert(bma->cur, &i);
1754 			if (error)
1755 				goto done;
1756 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1757 		}
1758 		break;
1759 
1760 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1761 		/*
1762 		 * Filling in the first part of a previous delayed allocation.
1763 		 * The left neighbor is contiguous.
1764 		 */
1765 		old = LEFT;
1766 		temp = PREV.br_blockcount - new->br_blockcount;
1767 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1768 				startblockval(PREV.br_startblock));
1769 
1770 		LEFT.br_blockcount += new->br_blockcount;
1771 
1772 		PREV.br_blockcount = temp;
1773 		PREV.br_startoff += new->br_blockcount;
1774 		PREV.br_startblock = nullstartblock(da_new);
1775 
1776 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1777 		xfs_iext_prev(ifp, &bma->icur);
1778 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1779 
1780 		if (bma->cur == NULL)
1781 			rval = XFS_ILOG_DEXT;
1782 		else {
1783 			rval = 0;
1784 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1785 			if (error)
1786 				goto done;
1787 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1788 			error = xfs_bmbt_update(bma->cur, &LEFT);
1789 			if (error)
1790 				goto done;
1791 		}
1792 		break;
1793 
1794 	case BMAP_LEFT_FILLING:
1795 		/*
1796 		 * Filling in the first part of a previous delayed allocation.
1797 		 * The left neighbor is not contiguous.
1798 		 */
1799 		xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1800 		(*nextents)++;
1801 		if (bma->cur == NULL)
1802 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1803 		else {
1804 			rval = XFS_ILOG_CORE;
1805 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1806 			if (error)
1807 				goto done;
1808 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1809 			error = xfs_btree_insert(bma->cur, &i);
1810 			if (error)
1811 				goto done;
1812 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1813 		}
1814 
1815 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1816 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1817 					bma->firstblock, bma->dfops,
1818 					&bma->cur, 1, &tmp_rval, whichfork);
1819 			rval |= tmp_rval;
1820 			if (error)
1821 				goto done;
1822 		}
1823 
1824 		temp = PREV.br_blockcount - new->br_blockcount;
1825 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1826 			startblockval(PREV.br_startblock) -
1827 			(bma->cur ? bma->cur->bc_private.b.allocated : 0));
1828 
1829 		PREV.br_startoff = new_endoff;
1830 		PREV.br_blockcount = temp;
1831 		PREV.br_startblock = nullstartblock(da_new);
1832 		xfs_iext_next(ifp, &bma->icur);
1833 		xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1834 		xfs_iext_prev(ifp, &bma->icur);
1835 		break;
1836 
1837 	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1838 		/*
1839 		 * Filling in the last part of a previous delayed allocation.
1840 		 * The right neighbor is contiguous with the new allocation.
1841 		 */
1842 		old = RIGHT;
1843 		RIGHT.br_startoff = new->br_startoff;
1844 		RIGHT.br_startblock = new->br_startblock;
1845 		RIGHT.br_blockcount += new->br_blockcount;
1846 
1847 		if (bma->cur == NULL)
1848 			rval = XFS_ILOG_DEXT;
1849 		else {
1850 			rval = 0;
1851 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1852 			if (error)
1853 				goto done;
1854 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1855 			error = xfs_bmbt_update(bma->cur, &RIGHT);
1856 			if (error)
1857 				goto done;
1858 		}
1859 
1860 		temp = PREV.br_blockcount - new->br_blockcount;
1861 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1862 			startblockval(PREV.br_startblock));
1863 
1864 		PREV.br_blockcount = temp;
1865 		PREV.br_startblock = nullstartblock(da_new);
1866 
1867 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1868 		xfs_iext_next(ifp, &bma->icur);
1869 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &RIGHT);
1870 		break;
1871 
1872 	case BMAP_RIGHT_FILLING:
1873 		/*
1874 		 * Filling in the last part of a previous delayed allocation.
1875 		 * The right neighbor is not contiguous.
1876 		 */
1877 		xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1878 		(*nextents)++;
1879 		if (bma->cur == NULL)
1880 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1881 		else {
1882 			rval = XFS_ILOG_CORE;
1883 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1884 			if (error)
1885 				goto done;
1886 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1887 			error = xfs_btree_insert(bma->cur, &i);
1888 			if (error)
1889 				goto done;
1890 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1891 		}
1892 
1893 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1894 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1895 				bma->firstblock, bma->dfops, &bma->cur, 1,
1896 				&tmp_rval, whichfork);
1897 			rval |= tmp_rval;
1898 			if (error)
1899 				goto done;
1900 		}
1901 
1902 		temp = PREV.br_blockcount - new->br_blockcount;
1903 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1904 			startblockval(PREV.br_startblock) -
1905 			(bma->cur ? bma->cur->bc_private.b.allocated : 0));
1906 
1907 		PREV.br_startblock = nullstartblock(da_new);
1908 		PREV.br_blockcount = temp;
1909 		xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1910 		xfs_iext_next(ifp, &bma->icur);
1911 		break;
1912 
1913 	case 0:
1914 		/*
1915 		 * Filling in the middle part of a previous delayed allocation.
1916 		 * Contiguity is impossible here.
1917 		 * This case is avoided almost all the time.
1918 		 *
1919 		 * We start with a delayed allocation:
1920 		 *
1921 		 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1922 		 *  PREV @ idx
1923 		 *
1924 	         * and we are allocating:
1925 		 *                     +rrrrrrrrrrrrrrrrr+
1926 		 *			      new
1927 		 *
1928 		 * and we set it up for insertion as:
1929 		 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1930 		 *                            new
1931 		 *  PREV @ idx          LEFT              RIGHT
1932 		 *                      inserted at idx + 1
1933 		 */
1934 		old = PREV;
1935 
1936 		/* LEFT is the new middle */
1937 		LEFT = *new;
1938 
1939 		/* RIGHT is the new right */
1940 		RIGHT.br_state = PREV.br_state;
1941 		RIGHT.br_startoff = new_endoff;
1942 		RIGHT.br_blockcount =
1943 			PREV.br_startoff + PREV.br_blockcount - new_endoff;
1944 		RIGHT.br_startblock =
1945 			nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1946 					RIGHT.br_blockcount));
1947 
1948 		/* truncate PREV */
1949 		PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
1950 		PREV.br_startblock =
1951 			nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1952 					PREV.br_blockcount));
1953 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1954 
1955 		xfs_iext_next(ifp, &bma->icur);
1956 		xfs_iext_insert(bma->ip, &bma->icur, &RIGHT, state);
1957 		xfs_iext_insert(bma->ip, &bma->icur, &LEFT, state);
1958 		(*nextents)++;
1959 
1960 		if (bma->cur == NULL)
1961 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1962 		else {
1963 			rval = XFS_ILOG_CORE;
1964 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1965 			if (error)
1966 				goto done;
1967 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1968 			error = xfs_btree_insert(bma->cur, &i);
1969 			if (error)
1970 				goto done;
1971 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1972 		}
1973 
1974 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1975 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1976 					bma->firstblock, bma->dfops, &bma->cur,
1977 					1, &tmp_rval, whichfork);
1978 			rval |= tmp_rval;
1979 			if (error)
1980 				goto done;
1981 		}
1982 
1983 		da_new = startblockval(PREV.br_startblock) +
1984 			 startblockval(RIGHT.br_startblock);
1985 		break;
1986 
1987 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1988 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1989 	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1990 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1991 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1992 	case BMAP_LEFT_CONTIG:
1993 	case BMAP_RIGHT_CONTIG:
1994 		/*
1995 		 * These cases are all impossible.
1996 		 */
1997 		ASSERT(0);
1998 	}
1999 
2000 	/* add reverse mapping */
2001 	error = xfs_rmap_map_extent(mp, bma->dfops, bma->ip, whichfork, new);
2002 	if (error)
2003 		goto done;
2004 
2005 	/* convert to a btree if necessary */
2006 	if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2007 		int	tmp_logflags;	/* partial log flag return val */
2008 
2009 		ASSERT(bma->cur == NULL);
2010 		error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2011 				bma->firstblock, bma->dfops, &bma->cur,
2012 				da_old > 0, &tmp_logflags, whichfork);
2013 		bma->logflags |= tmp_logflags;
2014 		if (error)
2015 			goto done;
2016 	}
2017 
2018 	if (bma->cur) {
2019 		da_new += bma->cur->bc_private.b.allocated;
2020 		bma->cur->bc_private.b.allocated = 0;
2021 	}
2022 
2023 	/* adjust for changes in reserved delayed indirect blocks */
2024 	if (da_new != da_old) {
2025 		ASSERT(state == 0 || da_new < da_old);
2026 		error = xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new),
2027 				false);
2028 	}
2029 
2030 	xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
2031 done:
2032 	if (whichfork != XFS_COW_FORK)
2033 		bma->logflags |= rval;
2034 	return error;
2035 #undef	LEFT
2036 #undef	RIGHT
2037 #undef	PREV
2038 }
2039 
2040 /*
2041  * Convert an unwritten allocation to a real allocation or vice versa.
2042  */
2043 STATIC int				/* error */
2044 xfs_bmap_add_extent_unwritten_real(
2045 	struct xfs_trans	*tp,
2046 	xfs_inode_t		*ip,	/* incore inode pointer */
2047 	int			whichfork,
2048 	struct xfs_iext_cursor	*icur,
2049 	xfs_btree_cur_t		**curp,	/* if *curp is null, not a btree */
2050 	xfs_bmbt_irec_t		*new,	/* new data to add to file extents */
2051 	xfs_fsblock_t		*first,	/* pointer to firstblock variable */
2052 	struct xfs_defer_ops	*dfops,	/* list of extents to be freed */
2053 	int			*logflagsp) /* inode logging flags */
2054 {
2055 	xfs_btree_cur_t		*cur;	/* btree cursor */
2056 	int			error;	/* error return value */
2057 	int			i;	/* temp state */
2058 	xfs_ifork_t		*ifp;	/* inode fork pointer */
2059 	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
2060 	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
2061 					/* left is 0, right is 1, prev is 2 */
2062 	int			rval=0;	/* return value (logging flags) */
2063 	int			state = xfs_bmap_fork_to_state(whichfork);
2064 	struct xfs_mount	*mp = ip->i_mount;
2065 	struct xfs_bmbt_irec	old;
2066 
2067 	*logflagsp = 0;
2068 
2069 	cur = *curp;
2070 	ifp = XFS_IFORK_PTR(ip, whichfork);
2071 
2072 	ASSERT(!isnullstartblock(new->br_startblock));
2073 
2074 	XFS_STATS_INC(mp, xs_add_exlist);
2075 
2076 #define	LEFT		r[0]
2077 #define	RIGHT		r[1]
2078 #define	PREV		r[2]
2079 
2080 	/*
2081 	 * Set up a bunch of variables to make the tests simpler.
2082 	 */
2083 	error = 0;
2084 	xfs_iext_get_extent(ifp, icur, &PREV);
2085 	ASSERT(new->br_state != PREV.br_state);
2086 	new_endoff = new->br_startoff + new->br_blockcount;
2087 	ASSERT(PREV.br_startoff <= new->br_startoff);
2088 	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2089 
2090 	/*
2091 	 * Set flags determining what part of the previous oldext allocation
2092 	 * extent is being replaced by a newext allocation.
2093 	 */
2094 	if (PREV.br_startoff == new->br_startoff)
2095 		state |= BMAP_LEFT_FILLING;
2096 	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2097 		state |= BMAP_RIGHT_FILLING;
2098 
2099 	/*
2100 	 * Check and set flags if this segment has a left neighbor.
2101 	 * Don't set contiguous if the combined extent would be too large.
2102 	 */
2103 	if (xfs_iext_peek_prev_extent(ifp, icur, &LEFT)) {
2104 		state |= BMAP_LEFT_VALID;
2105 		if (isnullstartblock(LEFT.br_startblock))
2106 			state |= BMAP_LEFT_DELAY;
2107 	}
2108 
2109 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2110 	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2111 	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2112 	    LEFT.br_state == new->br_state &&
2113 	    LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2114 		state |= BMAP_LEFT_CONTIG;
2115 
2116 	/*
2117 	 * Check and set flags if this segment has a right neighbor.
2118 	 * Don't set contiguous if the combined extent would be too large.
2119 	 * Also check for all-three-contiguous being too large.
2120 	 */
2121 	if (xfs_iext_peek_next_extent(ifp, icur, &RIGHT)) {
2122 		state |= BMAP_RIGHT_VALID;
2123 		if (isnullstartblock(RIGHT.br_startblock))
2124 			state |= BMAP_RIGHT_DELAY;
2125 	}
2126 
2127 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2128 	    new_endoff == RIGHT.br_startoff &&
2129 	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2130 	    new->br_state == RIGHT.br_state &&
2131 	    new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2132 	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2133 		       BMAP_RIGHT_FILLING)) !=
2134 		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2135 		       BMAP_RIGHT_FILLING) ||
2136 	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2137 			<= MAXEXTLEN))
2138 		state |= BMAP_RIGHT_CONTIG;
2139 
2140 	/*
2141 	 * Switch out based on the FILLING and CONTIG state bits.
2142 	 */
2143 	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2144 			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2145 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2146 	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2147 		/*
2148 		 * Setting all of a previous oldext extent to newext.
2149 		 * The left and right neighbors are both contiguous with new.
2150 		 */
2151 		LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
2152 
2153 		xfs_iext_remove(ip, icur, state);
2154 		xfs_iext_remove(ip, icur, state);
2155 		xfs_iext_prev(ifp, icur);
2156 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2157 		XFS_IFORK_NEXT_SET(ip, whichfork,
2158 				XFS_IFORK_NEXTENTS(ip, whichfork) - 2);
2159 		if (cur == NULL)
2160 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2161 		else {
2162 			rval = XFS_ILOG_CORE;
2163 			error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2164 			if (error)
2165 				goto done;
2166 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2167 			if ((error = xfs_btree_delete(cur, &i)))
2168 				goto done;
2169 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2170 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2171 				goto done;
2172 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2173 			if ((error = xfs_btree_delete(cur, &i)))
2174 				goto done;
2175 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2176 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2177 				goto done;
2178 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2179 			error = xfs_bmbt_update(cur, &LEFT);
2180 			if (error)
2181 				goto done;
2182 		}
2183 		break;
2184 
2185 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2186 		/*
2187 		 * Setting all of a previous oldext extent to newext.
2188 		 * The left neighbor is contiguous, the right is not.
2189 		 */
2190 		LEFT.br_blockcount += PREV.br_blockcount;
2191 
2192 		xfs_iext_remove(ip, icur, state);
2193 		xfs_iext_prev(ifp, icur);
2194 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2195 		XFS_IFORK_NEXT_SET(ip, whichfork,
2196 				XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2197 		if (cur == NULL)
2198 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2199 		else {
2200 			rval = XFS_ILOG_CORE;
2201 			error = xfs_bmbt_lookup_eq(cur, &PREV, &i);
2202 			if (error)
2203 				goto done;
2204 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2205 			if ((error = xfs_btree_delete(cur, &i)))
2206 				goto done;
2207 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2208 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2209 				goto done;
2210 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2211 			error = xfs_bmbt_update(cur, &LEFT);
2212 			if (error)
2213 				goto done;
2214 		}
2215 		break;
2216 
2217 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2218 		/*
2219 		 * Setting all of a previous oldext extent to newext.
2220 		 * The right neighbor is contiguous, the left is not.
2221 		 */
2222 		PREV.br_blockcount += RIGHT.br_blockcount;
2223 		PREV.br_state = new->br_state;
2224 
2225 		xfs_iext_next(ifp, icur);
2226 		xfs_iext_remove(ip, icur, state);
2227 		xfs_iext_prev(ifp, icur);
2228 		xfs_iext_update_extent(ip, state, icur, &PREV);
2229 
2230 		XFS_IFORK_NEXT_SET(ip, whichfork,
2231 				XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2232 		if (cur == NULL)
2233 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2234 		else {
2235 			rval = XFS_ILOG_CORE;
2236 			error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2237 			if (error)
2238 				goto done;
2239 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2240 			if ((error = xfs_btree_delete(cur, &i)))
2241 				goto done;
2242 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2243 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2244 				goto done;
2245 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2246 			error = xfs_bmbt_update(cur, &PREV);
2247 			if (error)
2248 				goto done;
2249 		}
2250 		break;
2251 
2252 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2253 		/*
2254 		 * Setting all of a previous oldext extent to newext.
2255 		 * Neither the left nor right neighbors are contiguous with
2256 		 * the new one.
2257 		 */
2258 		PREV.br_state = new->br_state;
2259 		xfs_iext_update_extent(ip, state, icur, &PREV);
2260 
2261 		if (cur == NULL)
2262 			rval = XFS_ILOG_DEXT;
2263 		else {
2264 			rval = 0;
2265 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2266 			if (error)
2267 				goto done;
2268 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2269 			error = xfs_bmbt_update(cur, &PREV);
2270 			if (error)
2271 				goto done;
2272 		}
2273 		break;
2274 
2275 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2276 		/*
2277 		 * Setting the first part of a previous oldext extent to newext.
2278 		 * The left neighbor is contiguous.
2279 		 */
2280 		LEFT.br_blockcount += new->br_blockcount;
2281 
2282 		old = PREV;
2283 		PREV.br_startoff += new->br_blockcount;
2284 		PREV.br_startblock += new->br_blockcount;
2285 		PREV.br_blockcount -= new->br_blockcount;
2286 
2287 		xfs_iext_update_extent(ip, state, icur, &PREV);
2288 		xfs_iext_prev(ifp, icur);
2289 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2290 
2291 		if (cur == NULL)
2292 			rval = XFS_ILOG_DEXT;
2293 		else {
2294 			rval = 0;
2295 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2296 			if (error)
2297 				goto done;
2298 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2299 			error = xfs_bmbt_update(cur, &PREV);
2300 			if (error)
2301 				goto done;
2302 			error = xfs_btree_decrement(cur, 0, &i);
2303 			if (error)
2304 				goto done;
2305 			error = xfs_bmbt_update(cur, &LEFT);
2306 			if (error)
2307 				goto done;
2308 		}
2309 		break;
2310 
2311 	case BMAP_LEFT_FILLING:
2312 		/*
2313 		 * Setting the first part of a previous oldext extent to newext.
2314 		 * The left neighbor is not contiguous.
2315 		 */
2316 		old = PREV;
2317 		PREV.br_startoff += new->br_blockcount;
2318 		PREV.br_startblock += new->br_blockcount;
2319 		PREV.br_blockcount -= new->br_blockcount;
2320 
2321 		xfs_iext_update_extent(ip, state, icur, &PREV);
2322 		xfs_iext_insert(ip, icur, new, state);
2323 		XFS_IFORK_NEXT_SET(ip, whichfork,
2324 				XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2325 		if (cur == NULL)
2326 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2327 		else {
2328 			rval = XFS_ILOG_CORE;
2329 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2330 			if (error)
2331 				goto done;
2332 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2333 			error = xfs_bmbt_update(cur, &PREV);
2334 			if (error)
2335 				goto done;
2336 			cur->bc_rec.b = *new;
2337 			if ((error = xfs_btree_insert(cur, &i)))
2338 				goto done;
2339 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2340 		}
2341 		break;
2342 
2343 	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2344 		/*
2345 		 * Setting the last part of a previous oldext extent to newext.
2346 		 * The right neighbor is contiguous with the new allocation.
2347 		 */
2348 		old = PREV;
2349 		PREV.br_blockcount -= new->br_blockcount;
2350 
2351 		RIGHT.br_startoff = new->br_startoff;
2352 		RIGHT.br_startblock = new->br_startblock;
2353 		RIGHT.br_blockcount += new->br_blockcount;
2354 
2355 		xfs_iext_update_extent(ip, state, icur, &PREV);
2356 		xfs_iext_next(ifp, icur);
2357 		xfs_iext_update_extent(ip, state, icur, &RIGHT);
2358 
2359 		if (cur == NULL)
2360 			rval = XFS_ILOG_DEXT;
2361 		else {
2362 			rval = 0;
2363 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2364 			if (error)
2365 				goto done;
2366 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2367 			error = xfs_bmbt_update(cur, &PREV);
2368 			if (error)
2369 				goto done;
2370 			error = xfs_btree_increment(cur, 0, &i);
2371 			if (error)
2372 				goto done;
2373 			error = xfs_bmbt_update(cur, &RIGHT);
2374 			if (error)
2375 				goto done;
2376 		}
2377 		break;
2378 
2379 	case BMAP_RIGHT_FILLING:
2380 		/*
2381 		 * Setting the last part of a previous oldext extent to newext.
2382 		 * The right neighbor is not contiguous.
2383 		 */
2384 		old = PREV;
2385 		PREV.br_blockcount -= new->br_blockcount;
2386 
2387 		xfs_iext_update_extent(ip, state, icur, &PREV);
2388 		xfs_iext_next(ifp, icur);
2389 		xfs_iext_insert(ip, icur, new, state);
2390 
2391 		XFS_IFORK_NEXT_SET(ip, whichfork,
2392 				XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2393 		if (cur == NULL)
2394 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2395 		else {
2396 			rval = XFS_ILOG_CORE;
2397 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2398 			if (error)
2399 				goto done;
2400 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2401 			error = xfs_bmbt_update(cur, &PREV);
2402 			if (error)
2403 				goto done;
2404 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2405 			if (error)
2406 				goto done;
2407 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2408 			if ((error = xfs_btree_insert(cur, &i)))
2409 				goto done;
2410 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2411 		}
2412 		break;
2413 
2414 	case 0:
2415 		/*
2416 		 * Setting the middle part of a previous oldext extent to
2417 		 * newext.  Contiguity is impossible here.
2418 		 * One extent becomes three extents.
2419 		 */
2420 		old = PREV;
2421 		PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
2422 
2423 		r[0] = *new;
2424 		r[1].br_startoff = new_endoff;
2425 		r[1].br_blockcount =
2426 			old.br_startoff + old.br_blockcount - new_endoff;
2427 		r[1].br_startblock = new->br_startblock + new->br_blockcount;
2428 		r[1].br_state = PREV.br_state;
2429 
2430 		xfs_iext_update_extent(ip, state, icur, &PREV);
2431 		xfs_iext_next(ifp, icur);
2432 		xfs_iext_insert(ip, icur, &r[1], state);
2433 		xfs_iext_insert(ip, icur, &r[0], state);
2434 
2435 		XFS_IFORK_NEXT_SET(ip, whichfork,
2436 				XFS_IFORK_NEXTENTS(ip, whichfork) + 2);
2437 		if (cur == NULL)
2438 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2439 		else {
2440 			rval = XFS_ILOG_CORE;
2441 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2442 			if (error)
2443 				goto done;
2444 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2445 			/* new right extent - oldext */
2446 			error = xfs_bmbt_update(cur, &r[1]);
2447 			if (error)
2448 				goto done;
2449 			/* new left extent - oldext */
2450 			cur->bc_rec.b = PREV;
2451 			if ((error = xfs_btree_insert(cur, &i)))
2452 				goto done;
2453 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2454 			/*
2455 			 * Reset the cursor to the position of the new extent
2456 			 * we are about to insert as we can't trust it after
2457 			 * the previous insert.
2458 			 */
2459 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2460 			if (error)
2461 				goto done;
2462 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2463 			/* new middle extent - newext */
2464 			if ((error = xfs_btree_insert(cur, &i)))
2465 				goto done;
2466 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2467 		}
2468 		break;
2469 
2470 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2471 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2472 	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2473 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2474 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2475 	case BMAP_LEFT_CONTIG:
2476 	case BMAP_RIGHT_CONTIG:
2477 		/*
2478 		 * These cases are all impossible.
2479 		 */
2480 		ASSERT(0);
2481 	}
2482 
2483 	/* update reverse mappings */
2484 	error = xfs_rmap_convert_extent(mp, dfops, ip, whichfork, new);
2485 	if (error)
2486 		goto done;
2487 
2488 	/* convert to a btree if necessary */
2489 	if (xfs_bmap_needs_btree(ip, whichfork)) {
2490 		int	tmp_logflags;	/* partial log flag return val */
2491 
2492 		ASSERT(cur == NULL);
2493 		error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, &cur,
2494 				0, &tmp_logflags, whichfork);
2495 		*logflagsp |= tmp_logflags;
2496 		if (error)
2497 			goto done;
2498 	}
2499 
2500 	/* clear out the allocated field, done with it now in any case. */
2501 	if (cur) {
2502 		cur->bc_private.b.allocated = 0;
2503 		*curp = cur;
2504 	}
2505 
2506 	xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
2507 done:
2508 	*logflagsp |= rval;
2509 	return error;
2510 #undef	LEFT
2511 #undef	RIGHT
2512 #undef	PREV
2513 }
2514 
2515 /*
2516  * Convert a hole to a delayed allocation.
2517  */
2518 STATIC void
2519 xfs_bmap_add_extent_hole_delay(
2520 	xfs_inode_t		*ip,	/* incore inode pointer */
2521 	int			whichfork,
2522 	struct xfs_iext_cursor	*icur,
2523 	xfs_bmbt_irec_t		*new)	/* new data to add to file extents */
2524 {
2525 	xfs_ifork_t		*ifp;	/* inode fork pointer */
2526 	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2527 	xfs_filblks_t		newlen=0;	/* new indirect size */
2528 	xfs_filblks_t		oldlen=0;	/* old indirect size */
2529 	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2530 	int			state = xfs_bmap_fork_to_state(whichfork);
2531 	xfs_filblks_t		temp;	 /* temp for indirect calculations */
2532 
2533 	ifp = XFS_IFORK_PTR(ip, whichfork);
2534 	ASSERT(isnullstartblock(new->br_startblock));
2535 
2536 	/*
2537 	 * Check and set flags if this segment has a left neighbor
2538 	 */
2539 	if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2540 		state |= BMAP_LEFT_VALID;
2541 		if (isnullstartblock(left.br_startblock))
2542 			state |= BMAP_LEFT_DELAY;
2543 	}
2544 
2545 	/*
2546 	 * Check and set flags if the current (right) segment exists.
2547 	 * If it doesn't exist, we're converting the hole at end-of-file.
2548 	 */
2549 	if (xfs_iext_get_extent(ifp, icur, &right)) {
2550 		state |= BMAP_RIGHT_VALID;
2551 		if (isnullstartblock(right.br_startblock))
2552 			state |= BMAP_RIGHT_DELAY;
2553 	}
2554 
2555 	/*
2556 	 * Set contiguity flags on the left and right neighbors.
2557 	 * Don't let extents get too large, even if the pieces are contiguous.
2558 	 */
2559 	if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2560 	    left.br_startoff + left.br_blockcount == new->br_startoff &&
2561 	    left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2562 		state |= BMAP_LEFT_CONTIG;
2563 
2564 	if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2565 	    new->br_startoff + new->br_blockcount == right.br_startoff &&
2566 	    new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2567 	    (!(state & BMAP_LEFT_CONTIG) ||
2568 	     (left.br_blockcount + new->br_blockcount +
2569 	      right.br_blockcount <= MAXEXTLEN)))
2570 		state |= BMAP_RIGHT_CONTIG;
2571 
2572 	/*
2573 	 * Switch out based on the contiguity flags.
2574 	 */
2575 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2576 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2577 		/*
2578 		 * New allocation is contiguous with delayed allocations
2579 		 * on the left and on the right.
2580 		 * Merge all three into a single extent record.
2581 		 */
2582 		temp = left.br_blockcount + new->br_blockcount +
2583 			right.br_blockcount;
2584 
2585 		oldlen = startblockval(left.br_startblock) +
2586 			startblockval(new->br_startblock) +
2587 			startblockval(right.br_startblock);
2588 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2589 					 oldlen);
2590 		left.br_startblock = nullstartblock(newlen);
2591 		left.br_blockcount = temp;
2592 
2593 		xfs_iext_remove(ip, icur, state);
2594 		xfs_iext_prev(ifp, icur);
2595 		xfs_iext_update_extent(ip, state, icur, &left);
2596 		break;
2597 
2598 	case BMAP_LEFT_CONTIG:
2599 		/*
2600 		 * New allocation is contiguous with a delayed allocation
2601 		 * on the left.
2602 		 * Merge the new allocation with the left neighbor.
2603 		 */
2604 		temp = left.br_blockcount + new->br_blockcount;
2605 
2606 		oldlen = startblockval(left.br_startblock) +
2607 			startblockval(new->br_startblock);
2608 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2609 					 oldlen);
2610 		left.br_blockcount = temp;
2611 		left.br_startblock = nullstartblock(newlen);
2612 
2613 		xfs_iext_prev(ifp, icur);
2614 		xfs_iext_update_extent(ip, state, icur, &left);
2615 		break;
2616 
2617 	case BMAP_RIGHT_CONTIG:
2618 		/*
2619 		 * New allocation is contiguous with a delayed allocation
2620 		 * on the right.
2621 		 * Merge the new allocation with the right neighbor.
2622 		 */
2623 		temp = new->br_blockcount + right.br_blockcount;
2624 		oldlen = startblockval(new->br_startblock) +
2625 			startblockval(right.br_startblock);
2626 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2627 					 oldlen);
2628 		right.br_startoff = new->br_startoff;
2629 		right.br_startblock = nullstartblock(newlen);
2630 		right.br_blockcount = temp;
2631 		xfs_iext_update_extent(ip, state, icur, &right);
2632 		break;
2633 
2634 	case 0:
2635 		/*
2636 		 * New allocation is not contiguous with another
2637 		 * delayed allocation.
2638 		 * Insert a new entry.
2639 		 */
2640 		oldlen = newlen = 0;
2641 		xfs_iext_insert(ip, icur, new, state);
2642 		break;
2643 	}
2644 	if (oldlen != newlen) {
2645 		ASSERT(oldlen > newlen);
2646 		xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2647 				 false);
2648 		/*
2649 		 * Nothing to do for disk quota accounting here.
2650 		 */
2651 	}
2652 }
2653 
2654 /*
2655  * Convert a hole to a real allocation.
2656  */
2657 STATIC int				/* error */
2658 xfs_bmap_add_extent_hole_real(
2659 	struct xfs_trans	*tp,
2660 	struct xfs_inode	*ip,
2661 	int			whichfork,
2662 	struct xfs_iext_cursor	*icur,
2663 	struct xfs_btree_cur	**curp,
2664 	struct xfs_bmbt_irec	*new,
2665 	xfs_fsblock_t		*first,
2666 	struct xfs_defer_ops	*dfops,
2667 	int			*logflagsp)
2668 {
2669 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
2670 	struct xfs_mount	*mp = ip->i_mount;
2671 	struct xfs_btree_cur	*cur = *curp;
2672 	int			error;	/* error return value */
2673 	int			i;	/* temp state */
2674 	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2675 	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2676 	int			rval=0;	/* return value (logging flags) */
2677 	int			state = xfs_bmap_fork_to_state(whichfork);
2678 	struct xfs_bmbt_irec	old;
2679 
2680 	ASSERT(!isnullstartblock(new->br_startblock));
2681 	ASSERT(!cur || !(cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2682 
2683 	XFS_STATS_INC(mp, xs_add_exlist);
2684 
2685 	/*
2686 	 * Check and set flags if this segment has a left neighbor.
2687 	 */
2688 	if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2689 		state |= BMAP_LEFT_VALID;
2690 		if (isnullstartblock(left.br_startblock))
2691 			state |= BMAP_LEFT_DELAY;
2692 	}
2693 
2694 	/*
2695 	 * Check and set flags if this segment has a current value.
2696 	 * Not true if we're inserting into the "hole" at eof.
2697 	 */
2698 	if (xfs_iext_get_extent(ifp, icur, &right)) {
2699 		state |= BMAP_RIGHT_VALID;
2700 		if (isnullstartblock(right.br_startblock))
2701 			state |= BMAP_RIGHT_DELAY;
2702 	}
2703 
2704 	/*
2705 	 * We're inserting a real allocation between "left" and "right".
2706 	 * Set the contiguity flags.  Don't let extents get too large.
2707 	 */
2708 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2709 	    left.br_startoff + left.br_blockcount == new->br_startoff &&
2710 	    left.br_startblock + left.br_blockcount == new->br_startblock &&
2711 	    left.br_state == new->br_state &&
2712 	    left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2713 		state |= BMAP_LEFT_CONTIG;
2714 
2715 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2716 	    new->br_startoff + new->br_blockcount == right.br_startoff &&
2717 	    new->br_startblock + new->br_blockcount == right.br_startblock &&
2718 	    new->br_state == right.br_state &&
2719 	    new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2720 	    (!(state & BMAP_LEFT_CONTIG) ||
2721 	     left.br_blockcount + new->br_blockcount +
2722 	     right.br_blockcount <= MAXEXTLEN))
2723 		state |= BMAP_RIGHT_CONTIG;
2724 
2725 	error = 0;
2726 	/*
2727 	 * Select which case we're in here, and implement it.
2728 	 */
2729 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2730 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2731 		/*
2732 		 * New allocation is contiguous with real allocations on the
2733 		 * left and on the right.
2734 		 * Merge all three into a single extent record.
2735 		 */
2736 		left.br_blockcount += new->br_blockcount + right.br_blockcount;
2737 
2738 		xfs_iext_remove(ip, icur, state);
2739 		xfs_iext_prev(ifp, icur);
2740 		xfs_iext_update_extent(ip, state, icur, &left);
2741 
2742 		XFS_IFORK_NEXT_SET(ip, whichfork,
2743 			XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2744 		if (cur == NULL) {
2745 			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2746 		} else {
2747 			rval = XFS_ILOG_CORE;
2748 			error = xfs_bmbt_lookup_eq(cur, &right, &i);
2749 			if (error)
2750 				goto done;
2751 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2752 			error = xfs_btree_delete(cur, &i);
2753 			if (error)
2754 				goto done;
2755 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2756 			error = xfs_btree_decrement(cur, 0, &i);
2757 			if (error)
2758 				goto done;
2759 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2760 			error = xfs_bmbt_update(cur, &left);
2761 			if (error)
2762 				goto done;
2763 		}
2764 		break;
2765 
2766 	case BMAP_LEFT_CONTIG:
2767 		/*
2768 		 * New allocation is contiguous with a real allocation
2769 		 * on the left.
2770 		 * Merge the new allocation with the left neighbor.
2771 		 */
2772 		old = left;
2773 		left.br_blockcount += new->br_blockcount;
2774 
2775 		xfs_iext_prev(ifp, icur);
2776 		xfs_iext_update_extent(ip, state, icur, &left);
2777 
2778 		if (cur == NULL) {
2779 			rval = xfs_ilog_fext(whichfork);
2780 		} else {
2781 			rval = 0;
2782 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2783 			if (error)
2784 				goto done;
2785 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2786 			error = xfs_bmbt_update(cur, &left);
2787 			if (error)
2788 				goto done;
2789 		}
2790 		break;
2791 
2792 	case BMAP_RIGHT_CONTIG:
2793 		/*
2794 		 * New allocation is contiguous with a real allocation
2795 		 * on the right.
2796 		 * Merge the new allocation with the right neighbor.
2797 		 */
2798 		old = right;
2799 
2800 		right.br_startoff = new->br_startoff;
2801 		right.br_startblock = new->br_startblock;
2802 		right.br_blockcount += new->br_blockcount;
2803 		xfs_iext_update_extent(ip, state, icur, &right);
2804 
2805 		if (cur == NULL) {
2806 			rval = xfs_ilog_fext(whichfork);
2807 		} else {
2808 			rval = 0;
2809 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2810 			if (error)
2811 				goto done;
2812 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2813 			error = xfs_bmbt_update(cur, &right);
2814 			if (error)
2815 				goto done;
2816 		}
2817 		break;
2818 
2819 	case 0:
2820 		/*
2821 		 * New allocation is not contiguous with another
2822 		 * real allocation.
2823 		 * Insert a new entry.
2824 		 */
2825 		xfs_iext_insert(ip, icur, new, state);
2826 		XFS_IFORK_NEXT_SET(ip, whichfork,
2827 			XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2828 		if (cur == NULL) {
2829 			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2830 		} else {
2831 			rval = XFS_ILOG_CORE;
2832 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2833 			if (error)
2834 				goto done;
2835 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2836 			error = xfs_btree_insert(cur, &i);
2837 			if (error)
2838 				goto done;
2839 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2840 		}
2841 		break;
2842 	}
2843 
2844 	/* add reverse mapping */
2845 	error = xfs_rmap_map_extent(mp, dfops, ip, whichfork, new);
2846 	if (error)
2847 		goto done;
2848 
2849 	/* convert to a btree if necessary */
2850 	if (xfs_bmap_needs_btree(ip, whichfork)) {
2851 		int	tmp_logflags;	/* partial log flag return val */
2852 
2853 		ASSERT(cur == NULL);
2854 		error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, curp,
2855 				0, &tmp_logflags, whichfork);
2856 		*logflagsp |= tmp_logflags;
2857 		cur = *curp;
2858 		if (error)
2859 			goto done;
2860 	}
2861 
2862 	/* clear out the allocated field, done with it now in any case. */
2863 	if (cur)
2864 		cur->bc_private.b.allocated = 0;
2865 
2866 	xfs_bmap_check_leaf_extents(cur, ip, whichfork);
2867 done:
2868 	*logflagsp |= rval;
2869 	return error;
2870 }
2871 
2872 /*
2873  * Functions used in the extent read, allocate and remove paths
2874  */
2875 
2876 /*
2877  * Adjust the size of the new extent based on di_extsize and rt extsize.
2878  */
2879 int
2880 xfs_bmap_extsize_align(
2881 	xfs_mount_t	*mp,
2882 	xfs_bmbt_irec_t	*gotp,		/* next extent pointer */
2883 	xfs_bmbt_irec_t	*prevp,		/* previous extent pointer */
2884 	xfs_extlen_t	extsz,		/* align to this extent size */
2885 	int		rt,		/* is this a realtime inode? */
2886 	int		eof,		/* is extent at end-of-file? */
2887 	int		delay,		/* creating delalloc extent? */
2888 	int		convert,	/* overwriting unwritten extent? */
2889 	xfs_fileoff_t	*offp,		/* in/out: aligned offset */
2890 	xfs_extlen_t	*lenp)		/* in/out: aligned length */
2891 {
2892 	xfs_fileoff_t	orig_off;	/* original offset */
2893 	xfs_extlen_t	orig_alen;	/* original length */
2894 	xfs_fileoff_t	orig_end;	/* original off+len */
2895 	xfs_fileoff_t	nexto;		/* next file offset */
2896 	xfs_fileoff_t	prevo;		/* previous file offset */
2897 	xfs_fileoff_t	align_off;	/* temp for offset */
2898 	xfs_extlen_t	align_alen;	/* temp for length */
2899 	xfs_extlen_t	temp;		/* temp for calculations */
2900 
2901 	if (convert)
2902 		return 0;
2903 
2904 	orig_off = align_off = *offp;
2905 	orig_alen = align_alen = *lenp;
2906 	orig_end = orig_off + orig_alen;
2907 
2908 	/*
2909 	 * If this request overlaps an existing extent, then don't
2910 	 * attempt to perform any additional alignment.
2911 	 */
2912 	if (!delay && !eof &&
2913 	    (orig_off >= gotp->br_startoff) &&
2914 	    (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2915 		return 0;
2916 	}
2917 
2918 	/*
2919 	 * If the file offset is unaligned vs. the extent size
2920 	 * we need to align it.  This will be possible unless
2921 	 * the file was previously written with a kernel that didn't
2922 	 * perform this alignment, or if a truncate shot us in the
2923 	 * foot.
2924 	 */
2925 	temp = do_mod(orig_off, extsz);
2926 	if (temp) {
2927 		align_alen += temp;
2928 		align_off -= temp;
2929 	}
2930 
2931 	/* Same adjustment for the end of the requested area. */
2932 	temp = (align_alen % extsz);
2933 	if (temp)
2934 		align_alen += extsz - temp;
2935 
2936 	/*
2937 	 * For large extent hint sizes, the aligned extent might be larger than
2938 	 * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
2939 	 * the length back under MAXEXTLEN. The outer allocation loops handle
2940 	 * short allocation just fine, so it is safe to do this. We only want to
2941 	 * do it when we are forced to, though, because it means more allocation
2942 	 * operations are required.
2943 	 */
2944 	while (align_alen > MAXEXTLEN)
2945 		align_alen -= extsz;
2946 	ASSERT(align_alen <= MAXEXTLEN);
2947 
2948 	/*
2949 	 * If the previous block overlaps with this proposed allocation
2950 	 * then move the start forward without adjusting the length.
2951 	 */
2952 	if (prevp->br_startoff != NULLFILEOFF) {
2953 		if (prevp->br_startblock == HOLESTARTBLOCK)
2954 			prevo = prevp->br_startoff;
2955 		else
2956 			prevo = prevp->br_startoff + prevp->br_blockcount;
2957 	} else
2958 		prevo = 0;
2959 	if (align_off != orig_off && align_off < prevo)
2960 		align_off = prevo;
2961 	/*
2962 	 * If the next block overlaps with this proposed allocation
2963 	 * then move the start back without adjusting the length,
2964 	 * but not before offset 0.
2965 	 * This may of course make the start overlap previous block,
2966 	 * and if we hit the offset 0 limit then the next block
2967 	 * can still overlap too.
2968 	 */
2969 	if (!eof && gotp->br_startoff != NULLFILEOFF) {
2970 		if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2971 		    (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2972 			nexto = gotp->br_startoff + gotp->br_blockcount;
2973 		else
2974 			nexto = gotp->br_startoff;
2975 	} else
2976 		nexto = NULLFILEOFF;
2977 	if (!eof &&
2978 	    align_off + align_alen != orig_end &&
2979 	    align_off + align_alen > nexto)
2980 		align_off = nexto > align_alen ? nexto - align_alen : 0;
2981 	/*
2982 	 * If we're now overlapping the next or previous extent that
2983 	 * means we can't fit an extsz piece in this hole.  Just move
2984 	 * the start forward to the first valid spot and set
2985 	 * the length so we hit the end.
2986 	 */
2987 	if (align_off != orig_off && align_off < prevo)
2988 		align_off = prevo;
2989 	if (align_off + align_alen != orig_end &&
2990 	    align_off + align_alen > nexto &&
2991 	    nexto != NULLFILEOFF) {
2992 		ASSERT(nexto > prevo);
2993 		align_alen = nexto - align_off;
2994 	}
2995 
2996 	/*
2997 	 * If realtime, and the result isn't a multiple of the realtime
2998 	 * extent size we need to remove blocks until it is.
2999 	 */
3000 	if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3001 		/*
3002 		 * We're not covering the original request, or
3003 		 * we won't be able to once we fix the length.
3004 		 */
3005 		if (orig_off < align_off ||
3006 		    orig_end > align_off + align_alen ||
3007 		    align_alen - temp < orig_alen)
3008 			return -EINVAL;
3009 		/*
3010 		 * Try to fix it by moving the start up.
3011 		 */
3012 		if (align_off + temp <= orig_off) {
3013 			align_alen -= temp;
3014 			align_off += temp;
3015 		}
3016 		/*
3017 		 * Try to fix it by moving the end in.
3018 		 */
3019 		else if (align_off + align_alen - temp >= orig_end)
3020 			align_alen -= temp;
3021 		/*
3022 		 * Set the start to the minimum then trim the length.
3023 		 */
3024 		else {
3025 			align_alen -= orig_off - align_off;
3026 			align_off = orig_off;
3027 			align_alen -= align_alen % mp->m_sb.sb_rextsize;
3028 		}
3029 		/*
3030 		 * Result doesn't cover the request, fail it.
3031 		 */
3032 		if (orig_off < align_off || orig_end > align_off + align_alen)
3033 			return -EINVAL;
3034 	} else {
3035 		ASSERT(orig_off >= align_off);
3036 		/* see MAXEXTLEN handling above */
3037 		ASSERT(orig_end <= align_off + align_alen ||
3038 		       align_alen + extsz > MAXEXTLEN);
3039 	}
3040 
3041 #ifdef DEBUG
3042 	if (!eof && gotp->br_startoff != NULLFILEOFF)
3043 		ASSERT(align_off + align_alen <= gotp->br_startoff);
3044 	if (prevp->br_startoff != NULLFILEOFF)
3045 		ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3046 #endif
3047 
3048 	*lenp = align_alen;
3049 	*offp = align_off;
3050 	return 0;
3051 }
3052 
3053 #define XFS_ALLOC_GAP_UNITS	4
3054 
3055 void
3056 xfs_bmap_adjacent(
3057 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3058 {
3059 	xfs_fsblock_t	adjust;		/* adjustment to block numbers */
3060 	xfs_agnumber_t	fb_agno;	/* ag number of ap->firstblock */
3061 	xfs_mount_t	*mp;		/* mount point structure */
3062 	int		nullfb;		/* true if ap->firstblock isn't set */
3063 	int		rt;		/* true if inode is realtime */
3064 
3065 #define	ISVALID(x,y)	\
3066 	(rt ? \
3067 		(x) < mp->m_sb.sb_rblocks : \
3068 		XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3069 		XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3070 		XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3071 
3072 	mp = ap->ip->i_mount;
3073 	nullfb = *ap->firstblock == NULLFSBLOCK;
3074 	rt = XFS_IS_REALTIME_INODE(ap->ip) &&
3075 		xfs_alloc_is_userdata(ap->datatype);
3076 	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3077 	/*
3078 	 * If allocating at eof, and there's a previous real block,
3079 	 * try to use its last block as our starting point.
3080 	 */
3081 	if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3082 	    !isnullstartblock(ap->prev.br_startblock) &&
3083 	    ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3084 		    ap->prev.br_startblock)) {
3085 		ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3086 		/*
3087 		 * Adjust for the gap between prevp and us.
3088 		 */
3089 		adjust = ap->offset -
3090 			(ap->prev.br_startoff + ap->prev.br_blockcount);
3091 		if (adjust &&
3092 		    ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3093 			ap->blkno += adjust;
3094 	}
3095 	/*
3096 	 * If not at eof, then compare the two neighbor blocks.
3097 	 * Figure out whether either one gives us a good starting point,
3098 	 * and pick the better one.
3099 	 */
3100 	else if (!ap->eof) {
3101 		xfs_fsblock_t	gotbno;		/* right side block number */
3102 		xfs_fsblock_t	gotdiff=0;	/* right side difference */
3103 		xfs_fsblock_t	prevbno;	/* left side block number */
3104 		xfs_fsblock_t	prevdiff=0;	/* left side difference */
3105 
3106 		/*
3107 		 * If there's a previous (left) block, select a requested
3108 		 * start block based on it.
3109 		 */
3110 		if (ap->prev.br_startoff != NULLFILEOFF &&
3111 		    !isnullstartblock(ap->prev.br_startblock) &&
3112 		    (prevbno = ap->prev.br_startblock +
3113 			       ap->prev.br_blockcount) &&
3114 		    ISVALID(prevbno, ap->prev.br_startblock)) {
3115 			/*
3116 			 * Calculate gap to end of previous block.
3117 			 */
3118 			adjust = prevdiff = ap->offset -
3119 				(ap->prev.br_startoff +
3120 				 ap->prev.br_blockcount);
3121 			/*
3122 			 * Figure the startblock based on the previous block's
3123 			 * end and the gap size.
3124 			 * Heuristic!
3125 			 * If the gap is large relative to the piece we're
3126 			 * allocating, or using it gives us an invalid block
3127 			 * number, then just use the end of the previous block.
3128 			 */
3129 			if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3130 			    ISVALID(prevbno + prevdiff,
3131 				    ap->prev.br_startblock))
3132 				prevbno += adjust;
3133 			else
3134 				prevdiff += adjust;
3135 			/*
3136 			 * If the firstblock forbids it, can't use it,
3137 			 * must use default.
3138 			 */
3139 			if (!rt && !nullfb &&
3140 			    XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3141 				prevbno = NULLFSBLOCK;
3142 		}
3143 		/*
3144 		 * No previous block or can't follow it, just default.
3145 		 */
3146 		else
3147 			prevbno = NULLFSBLOCK;
3148 		/*
3149 		 * If there's a following (right) block, select a requested
3150 		 * start block based on it.
3151 		 */
3152 		if (!isnullstartblock(ap->got.br_startblock)) {
3153 			/*
3154 			 * Calculate gap to start of next block.
3155 			 */
3156 			adjust = gotdiff = ap->got.br_startoff - ap->offset;
3157 			/*
3158 			 * Figure the startblock based on the next block's
3159 			 * start and the gap size.
3160 			 */
3161 			gotbno = ap->got.br_startblock;
3162 			/*
3163 			 * Heuristic!
3164 			 * If the gap is large relative to the piece we're
3165 			 * allocating, or using it gives us an invalid block
3166 			 * number, then just use the start of the next block
3167 			 * offset by our length.
3168 			 */
3169 			if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3170 			    ISVALID(gotbno - gotdiff, gotbno))
3171 				gotbno -= adjust;
3172 			else if (ISVALID(gotbno - ap->length, gotbno)) {
3173 				gotbno -= ap->length;
3174 				gotdiff += adjust - ap->length;
3175 			} else
3176 				gotdiff += adjust;
3177 			/*
3178 			 * If the firstblock forbids it, can't use it,
3179 			 * must use default.
3180 			 */
3181 			if (!rt && !nullfb &&
3182 			    XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3183 				gotbno = NULLFSBLOCK;
3184 		}
3185 		/*
3186 		 * No next block, just default.
3187 		 */
3188 		else
3189 			gotbno = NULLFSBLOCK;
3190 		/*
3191 		 * If both valid, pick the better one, else the only good
3192 		 * one, else ap->blkno is already set (to 0 or the inode block).
3193 		 */
3194 		if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3195 			ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3196 		else if (prevbno != NULLFSBLOCK)
3197 			ap->blkno = prevbno;
3198 		else if (gotbno != NULLFSBLOCK)
3199 			ap->blkno = gotbno;
3200 	}
3201 #undef ISVALID
3202 }
3203 
3204 static int
3205 xfs_bmap_longest_free_extent(
3206 	struct xfs_trans	*tp,
3207 	xfs_agnumber_t		ag,
3208 	xfs_extlen_t		*blen,
3209 	int			*notinit)
3210 {
3211 	struct xfs_mount	*mp = tp->t_mountp;
3212 	struct xfs_perag	*pag;
3213 	xfs_extlen_t		longest;
3214 	int			error = 0;
3215 
3216 	pag = xfs_perag_get(mp, ag);
3217 	if (!pag->pagf_init) {
3218 		error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3219 		if (error)
3220 			goto out;
3221 
3222 		if (!pag->pagf_init) {
3223 			*notinit = 1;
3224 			goto out;
3225 		}
3226 	}
3227 
3228 	longest = xfs_alloc_longest_free_extent(pag,
3229 				xfs_alloc_min_freelist(mp, pag),
3230 				xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
3231 	if (*blen < longest)
3232 		*blen = longest;
3233 
3234 out:
3235 	xfs_perag_put(pag);
3236 	return error;
3237 }
3238 
3239 static void
3240 xfs_bmap_select_minlen(
3241 	struct xfs_bmalloca	*ap,
3242 	struct xfs_alloc_arg	*args,
3243 	xfs_extlen_t		*blen,
3244 	int			notinit)
3245 {
3246 	if (notinit || *blen < ap->minlen) {
3247 		/*
3248 		 * Since we did a BUF_TRYLOCK above, it is possible that
3249 		 * there is space for this request.
3250 		 */
3251 		args->minlen = ap->minlen;
3252 	} else if (*blen < args->maxlen) {
3253 		/*
3254 		 * If the best seen length is less than the request length,
3255 		 * use the best as the minimum.
3256 		 */
3257 		args->minlen = *blen;
3258 	} else {
3259 		/*
3260 		 * Otherwise we've seen an extent as big as maxlen, use that
3261 		 * as the minimum.
3262 		 */
3263 		args->minlen = args->maxlen;
3264 	}
3265 }
3266 
3267 STATIC int
3268 xfs_bmap_btalloc_nullfb(
3269 	struct xfs_bmalloca	*ap,
3270 	struct xfs_alloc_arg	*args,
3271 	xfs_extlen_t		*blen)
3272 {
3273 	struct xfs_mount	*mp = ap->ip->i_mount;
3274 	xfs_agnumber_t		ag, startag;
3275 	int			notinit = 0;
3276 	int			error;
3277 
3278 	args->type = XFS_ALLOCTYPE_START_BNO;
3279 	args->total = ap->total;
3280 
3281 	startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3282 	if (startag == NULLAGNUMBER)
3283 		startag = ag = 0;
3284 
3285 	while (*blen < args->maxlen) {
3286 		error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3287 						     &notinit);
3288 		if (error)
3289 			return error;
3290 
3291 		if (++ag == mp->m_sb.sb_agcount)
3292 			ag = 0;
3293 		if (ag == startag)
3294 			break;
3295 	}
3296 
3297 	xfs_bmap_select_minlen(ap, args, blen, notinit);
3298 	return 0;
3299 }
3300 
3301 STATIC int
3302 xfs_bmap_btalloc_filestreams(
3303 	struct xfs_bmalloca	*ap,
3304 	struct xfs_alloc_arg	*args,
3305 	xfs_extlen_t		*blen)
3306 {
3307 	struct xfs_mount	*mp = ap->ip->i_mount;
3308 	xfs_agnumber_t		ag;
3309 	int			notinit = 0;
3310 	int			error;
3311 
3312 	args->type = XFS_ALLOCTYPE_NEAR_BNO;
3313 	args->total = ap->total;
3314 
3315 	ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3316 	if (ag == NULLAGNUMBER)
3317 		ag = 0;
3318 
3319 	error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3320 	if (error)
3321 		return error;
3322 
3323 	if (*blen < args->maxlen) {
3324 		error = xfs_filestream_new_ag(ap, &ag);
3325 		if (error)
3326 			return error;
3327 
3328 		error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3329 						     &notinit);
3330 		if (error)
3331 			return error;
3332 
3333 	}
3334 
3335 	xfs_bmap_select_minlen(ap, args, blen, notinit);
3336 
3337 	/*
3338 	 * Set the failure fallback case to look in the selected AG as stream
3339 	 * may have moved.
3340 	 */
3341 	ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3342 	return 0;
3343 }
3344 
3345 /* Update all inode and quota accounting for the allocation we just did. */
3346 static void
3347 xfs_bmap_btalloc_accounting(
3348 	struct xfs_bmalloca	*ap,
3349 	struct xfs_alloc_arg	*args)
3350 {
3351 	if (ap->flags & XFS_BMAPI_COWFORK) {
3352 		/*
3353 		 * COW fork blocks are in-core only and thus are treated as
3354 		 * in-core quota reservation (like delalloc blocks) even when
3355 		 * converted to real blocks. The quota reservation is not
3356 		 * accounted to disk until blocks are remapped to the data
3357 		 * fork. So if these blocks were previously delalloc, we
3358 		 * already have quota reservation and there's nothing to do
3359 		 * yet.
3360 		 */
3361 		if (ap->wasdel)
3362 			return;
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_trans_mod_dquot_byino(ap->tp, ap->ip,
3384 		ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT : XFS_TRANS_DQ_BCOUNT,
3385 		args->len);
3386 }
3387 
3388 STATIC int
3389 xfs_bmap_btalloc(
3390 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3391 {
3392 	xfs_mount_t	*mp;		/* mount point structure */
3393 	xfs_alloctype_t	atype = 0;	/* type for allocation routines */
3394 	xfs_extlen_t	align = 0;	/* minimum allocation alignment */
3395 	xfs_agnumber_t	fb_agno;	/* ag number of ap->firstblock */
3396 	xfs_agnumber_t	ag;
3397 	xfs_alloc_arg_t	args;
3398 	xfs_fileoff_t	orig_offset;
3399 	xfs_extlen_t	orig_length;
3400 	xfs_extlen_t	blen;
3401 	xfs_extlen_t	nextminlen = 0;
3402 	int		nullfb;		/* true if ap->firstblock isn't set */
3403 	int		isaligned;
3404 	int		tryagain;
3405 	int		error;
3406 	int		stripe_align;
3407 
3408 	ASSERT(ap->length);
3409 	orig_offset = ap->offset;
3410 	orig_length = ap->length;
3411 
3412 	mp = ap->ip->i_mount;
3413 
3414 	/* stripe alignment for allocation is determined by mount parameters */
3415 	stripe_align = 0;
3416 	if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3417 		stripe_align = mp->m_swidth;
3418 	else if (mp->m_dalign)
3419 		stripe_align = mp->m_dalign;
3420 
3421 	if (ap->flags & XFS_BMAPI_COWFORK)
3422 		align = xfs_get_cowextsz_hint(ap->ip);
3423 	else if (xfs_alloc_is_userdata(ap->datatype))
3424 		align = xfs_get_extsz_hint(ap->ip);
3425 	if (align) {
3426 		error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3427 						align, 0, ap->eof, 0, ap->conv,
3428 						&ap->offset, &ap->length);
3429 		ASSERT(!error);
3430 		ASSERT(ap->length);
3431 	}
3432 
3433 
3434 	nullfb = *ap->firstblock == NULLFSBLOCK;
3435 	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3436 	if (nullfb) {
3437 		if (xfs_alloc_is_userdata(ap->datatype) &&
3438 		    xfs_inode_is_filestream(ap->ip)) {
3439 			ag = xfs_filestream_lookup_ag(ap->ip);
3440 			ag = (ag != NULLAGNUMBER) ? ag : 0;
3441 			ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3442 		} else {
3443 			ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3444 		}
3445 	} else
3446 		ap->blkno = *ap->firstblock;
3447 
3448 	xfs_bmap_adjacent(ap);
3449 
3450 	/*
3451 	 * If allowed, use ap->blkno; otherwise must use firstblock since
3452 	 * it's in the right allocation group.
3453 	 */
3454 	if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3455 		;
3456 	else
3457 		ap->blkno = *ap->firstblock;
3458 	/*
3459 	 * Normal allocation, done through xfs_alloc_vextent.
3460 	 */
3461 	tryagain = isaligned = 0;
3462 	memset(&args, 0, sizeof(args));
3463 	args.tp = ap->tp;
3464 	args.mp = mp;
3465 	args.fsbno = ap->blkno;
3466 	xfs_rmap_skip_owner_update(&args.oinfo);
3467 
3468 	/* Trim the allocation back to the maximum an AG can fit. */
3469 	args.maxlen = MIN(ap->length, mp->m_ag_max_usable);
3470 	args.firstblock = *ap->firstblock;
3471 	blen = 0;
3472 	if (nullfb) {
3473 		/*
3474 		 * Search for an allocation group with a single extent large
3475 		 * enough for the request.  If one isn't found, then adjust
3476 		 * the minimum allocation size to the largest space found.
3477 		 */
3478 		if (xfs_alloc_is_userdata(ap->datatype) &&
3479 		    xfs_inode_is_filestream(ap->ip))
3480 			error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3481 		else
3482 			error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3483 		if (error)
3484 			return error;
3485 	} else if (ap->dfops->dop_low) {
3486 		if (xfs_inode_is_filestream(ap->ip))
3487 			args.type = XFS_ALLOCTYPE_FIRST_AG;
3488 		else
3489 			args.type = XFS_ALLOCTYPE_START_BNO;
3490 		args.total = args.minlen = ap->minlen;
3491 	} else {
3492 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
3493 		args.total = ap->total;
3494 		args.minlen = ap->minlen;
3495 	}
3496 	/* apply extent size hints if obtained earlier */
3497 	if (align) {
3498 		args.prod = align;
3499 		if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3500 			args.mod = (xfs_extlen_t)(args.prod - args.mod);
3501 	} else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3502 		args.prod = 1;
3503 		args.mod = 0;
3504 	} else {
3505 		args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3506 		if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3507 			args.mod = (xfs_extlen_t)(args.prod - args.mod);
3508 	}
3509 	/*
3510 	 * If we are not low on available data blocks, and the
3511 	 * underlying logical volume manager is a stripe, and
3512 	 * the file offset is zero then try to allocate data
3513 	 * blocks on stripe unit boundary.
3514 	 * NOTE: ap->aeof is only set if the allocation length
3515 	 * is >= the stripe unit and the allocation offset is
3516 	 * at the end of file.
3517 	 */
3518 	if (!ap->dfops->dop_low && ap->aeof) {
3519 		if (!ap->offset) {
3520 			args.alignment = stripe_align;
3521 			atype = args.type;
3522 			isaligned = 1;
3523 			/*
3524 			 * Adjust for alignment
3525 			 */
3526 			if (blen > args.alignment && blen <= args.maxlen)
3527 				args.minlen = blen - args.alignment;
3528 			args.minalignslop = 0;
3529 		} else {
3530 			/*
3531 			 * First try an exact bno allocation.
3532 			 * If it fails then do a near or start bno
3533 			 * allocation with alignment turned on.
3534 			 */
3535 			atype = args.type;
3536 			tryagain = 1;
3537 			args.type = XFS_ALLOCTYPE_THIS_BNO;
3538 			args.alignment = 1;
3539 			/*
3540 			 * Compute the minlen+alignment for the
3541 			 * next case.  Set slop so that the value
3542 			 * of minlen+alignment+slop doesn't go up
3543 			 * between the calls.
3544 			 */
3545 			if (blen > stripe_align && blen <= args.maxlen)
3546 				nextminlen = blen - stripe_align;
3547 			else
3548 				nextminlen = args.minlen;
3549 			if (nextminlen + stripe_align > args.minlen + 1)
3550 				args.minalignslop =
3551 					nextminlen + stripe_align -
3552 					args.minlen - 1;
3553 			else
3554 				args.minalignslop = 0;
3555 		}
3556 	} else {
3557 		args.alignment = 1;
3558 		args.minalignslop = 0;
3559 	}
3560 	args.minleft = ap->minleft;
3561 	args.wasdel = ap->wasdel;
3562 	args.resv = XFS_AG_RESV_NONE;
3563 	args.datatype = ap->datatype;
3564 	if (ap->datatype & XFS_ALLOC_USERDATA_ZERO)
3565 		args.ip = ap->ip;
3566 
3567 	error = xfs_alloc_vextent(&args);
3568 	if (error)
3569 		return error;
3570 
3571 	if (tryagain && args.fsbno == NULLFSBLOCK) {
3572 		/*
3573 		 * Exact allocation failed. Now try with alignment
3574 		 * turned on.
3575 		 */
3576 		args.type = atype;
3577 		args.fsbno = ap->blkno;
3578 		args.alignment = stripe_align;
3579 		args.minlen = nextminlen;
3580 		args.minalignslop = 0;
3581 		isaligned = 1;
3582 		if ((error = xfs_alloc_vextent(&args)))
3583 			return error;
3584 	}
3585 	if (isaligned && args.fsbno == NULLFSBLOCK) {
3586 		/*
3587 		 * allocation failed, so turn off alignment and
3588 		 * try again.
3589 		 */
3590 		args.type = atype;
3591 		args.fsbno = ap->blkno;
3592 		args.alignment = 0;
3593 		if ((error = xfs_alloc_vextent(&args)))
3594 			return error;
3595 	}
3596 	if (args.fsbno == NULLFSBLOCK && nullfb &&
3597 	    args.minlen > ap->minlen) {
3598 		args.minlen = ap->minlen;
3599 		args.type = XFS_ALLOCTYPE_START_BNO;
3600 		args.fsbno = ap->blkno;
3601 		if ((error = xfs_alloc_vextent(&args)))
3602 			return error;
3603 	}
3604 	if (args.fsbno == NULLFSBLOCK && nullfb) {
3605 		args.fsbno = 0;
3606 		args.type = XFS_ALLOCTYPE_FIRST_AG;
3607 		args.total = ap->minlen;
3608 		if ((error = xfs_alloc_vextent(&args)))
3609 			return error;
3610 		ap->dfops->dop_low = true;
3611 	}
3612 	if (args.fsbno != NULLFSBLOCK) {
3613 		/*
3614 		 * check the allocation happened at the same or higher AG than
3615 		 * the first block that was allocated.
3616 		 */
3617 		ASSERT(*ap->firstblock == NULLFSBLOCK ||
3618 		       XFS_FSB_TO_AGNO(mp, *ap->firstblock) <=
3619 		       XFS_FSB_TO_AGNO(mp, args.fsbno));
3620 
3621 		ap->blkno = args.fsbno;
3622 		if (*ap->firstblock == NULLFSBLOCK)
3623 			*ap->firstblock = args.fsbno;
3624 		ASSERT(nullfb || fb_agno <= args.agno);
3625 		ap->length = args.len;
3626 		/*
3627 		 * If the extent size hint is active, we tried to round the
3628 		 * caller's allocation request offset down to extsz and the
3629 		 * length up to another extsz boundary.  If we found a free
3630 		 * extent we mapped it in starting at this new offset.  If the
3631 		 * newly mapped space isn't long enough to cover any of the
3632 		 * range of offsets that was originally requested, move the
3633 		 * mapping up so that we can fill as much of the caller's
3634 		 * original request as possible.  Free space is apparently
3635 		 * very fragmented so we're unlikely to be able to satisfy the
3636 		 * hints anyway.
3637 		 */
3638 		if (ap->length <= orig_length)
3639 			ap->offset = orig_offset;
3640 		else if (ap->offset + ap->length < orig_offset + orig_length)
3641 			ap->offset = orig_offset + orig_length - ap->length;
3642 		xfs_bmap_btalloc_accounting(ap, &args);
3643 	} else {
3644 		ap->blkno = NULLFSBLOCK;
3645 		ap->length = 0;
3646 	}
3647 	return 0;
3648 }
3649 
3650 /*
3651  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3652  * It figures out where to ask the underlying allocator to put the new extent.
3653  */
3654 STATIC int
3655 xfs_bmap_alloc(
3656 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3657 {
3658 	if (XFS_IS_REALTIME_INODE(ap->ip) &&
3659 	    xfs_alloc_is_userdata(ap->datatype))
3660 		return xfs_bmap_rtalloc(ap);
3661 	return xfs_bmap_btalloc(ap);
3662 }
3663 
3664 /* Trim extent to fit a logical block range. */
3665 void
3666 xfs_trim_extent(
3667 	struct xfs_bmbt_irec	*irec,
3668 	xfs_fileoff_t		bno,
3669 	xfs_filblks_t		len)
3670 {
3671 	xfs_fileoff_t		distance;
3672 	xfs_fileoff_t		end = bno + len;
3673 
3674 	if (irec->br_startoff + irec->br_blockcount <= bno ||
3675 	    irec->br_startoff >= end) {
3676 		irec->br_blockcount = 0;
3677 		return;
3678 	}
3679 
3680 	if (irec->br_startoff < bno) {
3681 		distance = bno - irec->br_startoff;
3682 		if (isnullstartblock(irec->br_startblock))
3683 			irec->br_startblock = DELAYSTARTBLOCK;
3684 		if (irec->br_startblock != DELAYSTARTBLOCK &&
3685 		    irec->br_startblock != HOLESTARTBLOCK)
3686 			irec->br_startblock += distance;
3687 		irec->br_startoff += distance;
3688 		irec->br_blockcount -= distance;
3689 	}
3690 
3691 	if (end < irec->br_startoff + irec->br_blockcount) {
3692 		distance = irec->br_startoff + irec->br_blockcount - end;
3693 		irec->br_blockcount -= distance;
3694 	}
3695 }
3696 
3697 /* trim extent to within eof */
3698 void
3699 xfs_trim_extent_eof(
3700 	struct xfs_bmbt_irec	*irec,
3701 	struct xfs_inode	*ip)
3702 
3703 {
3704 	xfs_trim_extent(irec, 0, XFS_B_TO_FSB(ip->i_mount,
3705 					      i_size_read(VFS_I(ip))));
3706 }
3707 
3708 /*
3709  * Trim the returned map to the required bounds
3710  */
3711 STATIC void
3712 xfs_bmapi_trim_map(
3713 	struct xfs_bmbt_irec	*mval,
3714 	struct xfs_bmbt_irec	*got,
3715 	xfs_fileoff_t		*bno,
3716 	xfs_filblks_t		len,
3717 	xfs_fileoff_t		obno,
3718 	xfs_fileoff_t		end,
3719 	int			n,
3720 	int			flags)
3721 {
3722 	if ((flags & XFS_BMAPI_ENTIRE) ||
3723 	    got->br_startoff + got->br_blockcount <= obno) {
3724 		*mval = *got;
3725 		if (isnullstartblock(got->br_startblock))
3726 			mval->br_startblock = DELAYSTARTBLOCK;
3727 		return;
3728 	}
3729 
3730 	if (obno > *bno)
3731 		*bno = obno;
3732 	ASSERT((*bno >= obno) || (n == 0));
3733 	ASSERT(*bno < end);
3734 	mval->br_startoff = *bno;
3735 	if (isnullstartblock(got->br_startblock))
3736 		mval->br_startblock = DELAYSTARTBLOCK;
3737 	else
3738 		mval->br_startblock = got->br_startblock +
3739 					(*bno - got->br_startoff);
3740 	/*
3741 	 * Return the minimum of what we got and what we asked for for
3742 	 * the length.  We can use the len variable here because it is
3743 	 * modified below and we could have been there before coming
3744 	 * here if the first part of the allocation didn't overlap what
3745 	 * was asked for.
3746 	 */
3747 	mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3748 			got->br_blockcount - (*bno - got->br_startoff));
3749 	mval->br_state = got->br_state;
3750 	ASSERT(mval->br_blockcount <= len);
3751 	return;
3752 }
3753 
3754 /*
3755  * Update and validate the extent map to return
3756  */
3757 STATIC void
3758 xfs_bmapi_update_map(
3759 	struct xfs_bmbt_irec	**map,
3760 	xfs_fileoff_t		*bno,
3761 	xfs_filblks_t		*len,
3762 	xfs_fileoff_t		obno,
3763 	xfs_fileoff_t		end,
3764 	int			*n,
3765 	int			flags)
3766 {
3767 	xfs_bmbt_irec_t	*mval = *map;
3768 
3769 	ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3770 	       ((mval->br_startoff + mval->br_blockcount) <= end));
3771 	ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3772 	       (mval->br_startoff < obno));
3773 
3774 	*bno = mval->br_startoff + mval->br_blockcount;
3775 	*len = end - *bno;
3776 	if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3777 		/* update previous map with new information */
3778 		ASSERT(mval->br_startblock == mval[-1].br_startblock);
3779 		ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3780 		ASSERT(mval->br_state == mval[-1].br_state);
3781 		mval[-1].br_blockcount = mval->br_blockcount;
3782 		mval[-1].br_state = mval->br_state;
3783 	} else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3784 		   mval[-1].br_startblock != DELAYSTARTBLOCK &&
3785 		   mval[-1].br_startblock != HOLESTARTBLOCK &&
3786 		   mval->br_startblock == mval[-1].br_startblock +
3787 					  mval[-1].br_blockcount &&
3788 		   ((flags & XFS_BMAPI_IGSTATE) ||
3789 			mval[-1].br_state == mval->br_state)) {
3790 		ASSERT(mval->br_startoff ==
3791 		       mval[-1].br_startoff + mval[-1].br_blockcount);
3792 		mval[-1].br_blockcount += mval->br_blockcount;
3793 	} else if (*n > 0 &&
3794 		   mval->br_startblock == DELAYSTARTBLOCK &&
3795 		   mval[-1].br_startblock == DELAYSTARTBLOCK &&
3796 		   mval->br_startoff ==
3797 		   mval[-1].br_startoff + mval[-1].br_blockcount) {
3798 		mval[-1].br_blockcount += mval->br_blockcount;
3799 		mval[-1].br_state = mval->br_state;
3800 	} else if (!((*n == 0) &&
3801 		     ((mval->br_startoff + mval->br_blockcount) <=
3802 		      obno))) {
3803 		mval++;
3804 		(*n)++;
3805 	}
3806 	*map = mval;
3807 }
3808 
3809 /*
3810  * Map file blocks to filesystem blocks without allocation.
3811  */
3812 int
3813 xfs_bmapi_read(
3814 	struct xfs_inode	*ip,
3815 	xfs_fileoff_t		bno,
3816 	xfs_filblks_t		len,
3817 	struct xfs_bmbt_irec	*mval,
3818 	int			*nmap,
3819 	int			flags)
3820 {
3821 	struct xfs_mount	*mp = ip->i_mount;
3822 	struct xfs_ifork	*ifp;
3823 	struct xfs_bmbt_irec	got;
3824 	xfs_fileoff_t		obno;
3825 	xfs_fileoff_t		end;
3826 	struct xfs_iext_cursor	icur;
3827 	int			error;
3828 	bool			eof = false;
3829 	int			n = 0;
3830 	int			whichfork = xfs_bmapi_whichfork(flags);
3831 
3832 	ASSERT(*nmap >= 1);
3833 	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
3834 			   XFS_BMAPI_IGSTATE|XFS_BMAPI_COWFORK)));
3835 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
3836 
3837 	if (unlikely(XFS_TEST_ERROR(
3838 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3839 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
3840 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
3841 		XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
3842 		return -EFSCORRUPTED;
3843 	}
3844 
3845 	if (XFS_FORCED_SHUTDOWN(mp))
3846 		return -EIO;
3847 
3848 	XFS_STATS_INC(mp, xs_blk_mapr);
3849 
3850 	ifp = XFS_IFORK_PTR(ip, whichfork);
3851 
3852 	/* No CoW fork?  Return a hole. */
3853 	if (whichfork == XFS_COW_FORK && !ifp) {
3854 		mval->br_startoff = bno;
3855 		mval->br_startblock = HOLESTARTBLOCK;
3856 		mval->br_blockcount = len;
3857 		mval->br_state = XFS_EXT_NORM;
3858 		*nmap = 1;
3859 		return 0;
3860 	}
3861 
3862 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
3863 		error = xfs_iread_extents(NULL, ip, whichfork);
3864 		if (error)
3865 			return error;
3866 	}
3867 
3868 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got))
3869 		eof = true;
3870 	end = bno + len;
3871 	obno = bno;
3872 
3873 	while (bno < end && n < *nmap) {
3874 		/* Reading past eof, act as though there's a hole up to end. */
3875 		if (eof)
3876 			got.br_startoff = end;
3877 		if (got.br_startoff > bno) {
3878 			/* Reading in a hole.  */
3879 			mval->br_startoff = bno;
3880 			mval->br_startblock = HOLESTARTBLOCK;
3881 			mval->br_blockcount =
3882 				XFS_FILBLKS_MIN(len, got.br_startoff - bno);
3883 			mval->br_state = XFS_EXT_NORM;
3884 			bno += mval->br_blockcount;
3885 			len -= mval->br_blockcount;
3886 			mval++;
3887 			n++;
3888 			continue;
3889 		}
3890 
3891 		/* set up the extent map to return. */
3892 		xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
3893 		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
3894 
3895 		/* If we're done, stop now. */
3896 		if (bno >= end || n >= *nmap)
3897 			break;
3898 
3899 		/* Else go on to the next record. */
3900 		if (!xfs_iext_next_extent(ifp, &icur, &got))
3901 			eof = true;
3902 	}
3903 	*nmap = n;
3904 	return 0;
3905 }
3906 
3907 /*
3908  * Add a delayed allocation extent to an inode. Blocks are reserved from the
3909  * global pool and the extent inserted into the inode in-core extent tree.
3910  *
3911  * On entry, got refers to the first extent beyond the offset of the extent to
3912  * allocate or eof is specified if no such extent exists. On return, got refers
3913  * to the extent record that was inserted to the inode fork.
3914  *
3915  * Note that the allocated extent may have been merged with contiguous extents
3916  * during insertion into the inode fork. Thus, got does not reflect the current
3917  * state of the inode fork on return. If necessary, the caller can use lastx to
3918  * look up the updated record in the inode fork.
3919  */
3920 int
3921 xfs_bmapi_reserve_delalloc(
3922 	struct xfs_inode	*ip,
3923 	int			whichfork,
3924 	xfs_fileoff_t		off,
3925 	xfs_filblks_t		len,
3926 	xfs_filblks_t		prealloc,
3927 	struct xfs_bmbt_irec	*got,
3928 	struct xfs_iext_cursor	*icur,
3929 	int			eof)
3930 {
3931 	struct xfs_mount	*mp = ip->i_mount;
3932 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
3933 	xfs_extlen_t		alen;
3934 	xfs_extlen_t		indlen;
3935 	int			error;
3936 	xfs_fileoff_t		aoff = off;
3937 
3938 	/*
3939 	 * Cap the alloc length. Keep track of prealloc so we know whether to
3940 	 * tag the inode before we return.
3941 	 */
3942 	alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN);
3943 	if (!eof)
3944 		alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
3945 	if (prealloc && alen >= len)
3946 		prealloc = alen - len;
3947 
3948 	/* Figure out the extent size, adjust alen */
3949 	if (whichfork == XFS_COW_FORK) {
3950 		struct xfs_bmbt_irec	prev;
3951 		xfs_extlen_t		extsz = xfs_get_cowextsz_hint(ip);
3952 
3953 		if (!xfs_iext_peek_prev_extent(ifp, icur, &prev))
3954 			prev.br_startoff = NULLFILEOFF;
3955 
3956 		error = xfs_bmap_extsize_align(mp, got, &prev, extsz, 0, eof,
3957 					       1, 0, &aoff, &alen);
3958 		ASSERT(!error);
3959 	}
3960 
3961 	/*
3962 	 * Make a transaction-less quota reservation for delayed allocation
3963 	 * blocks.  This number gets adjusted later.  We return if we haven't
3964 	 * allocated blocks already inside this loop.
3965 	 */
3966 	error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
3967 						XFS_QMOPT_RES_REGBLKS);
3968 	if (error)
3969 		return error;
3970 
3971 	/*
3972 	 * Split changing sb for alen and indlen since they could be coming
3973 	 * from different places.
3974 	 */
3975 	indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
3976 	ASSERT(indlen > 0);
3977 
3978 	error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
3979 	if (error)
3980 		goto out_unreserve_quota;
3981 
3982 	error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
3983 	if (error)
3984 		goto out_unreserve_blocks;
3985 
3986 
3987 	ip->i_delayed_blks += alen;
3988 
3989 	got->br_startoff = aoff;
3990 	got->br_startblock = nullstartblock(indlen);
3991 	got->br_blockcount = alen;
3992 	got->br_state = XFS_EXT_NORM;
3993 
3994 	xfs_bmap_add_extent_hole_delay(ip, whichfork, icur, got);
3995 
3996 	/*
3997 	 * Tag the inode if blocks were preallocated. Note that COW fork
3998 	 * preallocation can occur at the start or end of the extent, even when
3999 	 * prealloc == 0, so we must also check the aligned offset and length.
4000 	 */
4001 	if (whichfork == XFS_DATA_FORK && prealloc)
4002 		xfs_inode_set_eofblocks_tag(ip);
4003 	if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len))
4004 		xfs_inode_set_cowblocks_tag(ip);
4005 
4006 	return 0;
4007 
4008 out_unreserve_blocks:
4009 	xfs_mod_fdblocks(mp, alen, false);
4010 out_unreserve_quota:
4011 	if (XFS_IS_QUOTA_ON(mp))
4012 		xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0,
4013 						XFS_QMOPT_RES_REGBLKS);
4014 	return error;
4015 }
4016 
4017 static int
4018 xfs_bmapi_allocate(
4019 	struct xfs_bmalloca	*bma)
4020 {
4021 	struct xfs_mount	*mp = bma->ip->i_mount;
4022 	int			whichfork = xfs_bmapi_whichfork(bma->flags);
4023 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4024 	int			tmp_logflags = 0;
4025 	int			error;
4026 
4027 	ASSERT(bma->length > 0);
4028 
4029 	/*
4030 	 * For the wasdelay case, we could also just allocate the stuff asked
4031 	 * for in this bmap call but that wouldn't be as good.
4032 	 */
4033 	if (bma->wasdel) {
4034 		bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4035 		bma->offset = bma->got.br_startoff;
4036 		xfs_iext_peek_prev_extent(ifp, &bma->icur, &bma->prev);
4037 	} else {
4038 		bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4039 		if (!bma->eof)
4040 			bma->length = XFS_FILBLKS_MIN(bma->length,
4041 					bma->got.br_startoff - bma->offset);
4042 	}
4043 
4044 	/*
4045 	 * Set the data type being allocated. For the data fork, the first data
4046 	 * in the file is treated differently to all other allocations. For the
4047 	 * attribute fork, we only need to ensure the allocated range is not on
4048 	 * the busy list.
4049 	 */
4050 	if (!(bma->flags & XFS_BMAPI_METADATA)) {
4051 		bma->datatype = XFS_ALLOC_NOBUSY;
4052 		if (whichfork == XFS_DATA_FORK) {
4053 			if (bma->offset == 0)
4054 				bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA;
4055 			else
4056 				bma->datatype |= XFS_ALLOC_USERDATA;
4057 		}
4058 		if (bma->flags & XFS_BMAPI_ZERO)
4059 			bma->datatype |= XFS_ALLOC_USERDATA_ZERO;
4060 	}
4061 
4062 	bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4063 
4064 	/*
4065 	 * Only want to do the alignment at the eof if it is userdata and
4066 	 * allocation length is larger than a stripe unit.
4067 	 */
4068 	if (mp->m_dalign && bma->length >= mp->m_dalign &&
4069 	    !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4070 		error = xfs_bmap_isaeof(bma, whichfork);
4071 		if (error)
4072 			return error;
4073 	}
4074 
4075 	error = xfs_bmap_alloc(bma);
4076 	if (error)
4077 		return error;
4078 
4079 	if (bma->cur)
4080 		bma->cur->bc_private.b.firstblock = *bma->firstblock;
4081 	if (bma->blkno == NULLFSBLOCK)
4082 		return 0;
4083 	if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4084 		bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4085 		bma->cur->bc_private.b.firstblock = *bma->firstblock;
4086 		bma->cur->bc_private.b.dfops = bma->dfops;
4087 	}
4088 	/*
4089 	 * Bump the number of extents we've allocated
4090 	 * in this call.
4091 	 */
4092 	bma->nallocs++;
4093 
4094 	if (bma->cur)
4095 		bma->cur->bc_private.b.flags =
4096 			bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4097 
4098 	bma->got.br_startoff = bma->offset;
4099 	bma->got.br_startblock = bma->blkno;
4100 	bma->got.br_blockcount = bma->length;
4101 	bma->got.br_state = XFS_EXT_NORM;
4102 
4103 	/*
4104 	 * In the data fork, a wasdelay extent has been initialized, so
4105 	 * shouldn't be flagged as unwritten.
4106 	 *
4107 	 * For the cow fork, however, we convert delalloc reservations
4108 	 * (extents allocated for speculative preallocation) to
4109 	 * allocated unwritten extents, and only convert the unwritten
4110 	 * extents to real extents when we're about to write the data.
4111 	 */
4112 	if ((!bma->wasdel || (bma->flags & XFS_BMAPI_COWFORK)) &&
4113 	    (bma->flags & XFS_BMAPI_PREALLOC) &&
4114 	    xfs_sb_version_hasextflgbit(&mp->m_sb))
4115 		bma->got.br_state = XFS_EXT_UNWRITTEN;
4116 
4117 	if (bma->wasdel)
4118 		error = xfs_bmap_add_extent_delay_real(bma, whichfork);
4119 	else
4120 		error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip,
4121 				whichfork, &bma->icur, &bma->cur, &bma->got,
4122 				bma->firstblock, bma->dfops, &bma->logflags);
4123 
4124 	bma->logflags |= tmp_logflags;
4125 	if (error)
4126 		return error;
4127 
4128 	/*
4129 	 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4130 	 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4131 	 * the neighbouring ones.
4132 	 */
4133 	xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4134 
4135 	ASSERT(bma->got.br_startoff <= bma->offset);
4136 	ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4137 	       bma->offset + bma->length);
4138 	ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4139 	       bma->got.br_state == XFS_EXT_UNWRITTEN);
4140 	return 0;
4141 }
4142 
4143 STATIC int
4144 xfs_bmapi_convert_unwritten(
4145 	struct xfs_bmalloca	*bma,
4146 	struct xfs_bmbt_irec	*mval,
4147 	xfs_filblks_t		len,
4148 	int			flags)
4149 {
4150 	int			whichfork = xfs_bmapi_whichfork(flags);
4151 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4152 	int			tmp_logflags = 0;
4153 	int			error;
4154 
4155 	/* check if we need to do unwritten->real conversion */
4156 	if (mval->br_state == XFS_EXT_UNWRITTEN &&
4157 	    (flags & XFS_BMAPI_PREALLOC))
4158 		return 0;
4159 
4160 	/* check if we need to do real->unwritten conversion */
4161 	if (mval->br_state == XFS_EXT_NORM &&
4162 	    (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4163 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4164 		return 0;
4165 
4166 	/*
4167 	 * Modify (by adding) the state flag, if writing.
4168 	 */
4169 	ASSERT(mval->br_blockcount <= len);
4170 	if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4171 		bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4172 					bma->ip, whichfork);
4173 		bma->cur->bc_private.b.firstblock = *bma->firstblock;
4174 		bma->cur->bc_private.b.dfops = bma->dfops;
4175 	}
4176 	mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4177 				? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4178 
4179 	/*
4180 	 * Before insertion into the bmbt, zero the range being converted
4181 	 * if required.
4182 	 */
4183 	if (flags & XFS_BMAPI_ZERO) {
4184 		error = xfs_zero_extent(bma->ip, mval->br_startblock,
4185 					mval->br_blockcount);
4186 		if (error)
4187 			return error;
4188 	}
4189 
4190 	error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork,
4191 			&bma->icur, &bma->cur, mval, bma->firstblock,
4192 			bma->dfops, &tmp_logflags);
4193 	/*
4194 	 * Log the inode core unconditionally in the unwritten extent conversion
4195 	 * path because the conversion might not have done so (e.g., if the
4196 	 * extent count hasn't changed). We need to make sure the inode is dirty
4197 	 * in the transaction for the sake of fsync(), even if nothing has
4198 	 * changed, because fsync() will not force the log for this transaction
4199 	 * unless it sees the inode pinned.
4200 	 *
4201 	 * Note: If we're only converting cow fork extents, there aren't
4202 	 * any on-disk updates to make, so we don't need to log anything.
4203 	 */
4204 	if (whichfork != XFS_COW_FORK)
4205 		bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4206 	if (error)
4207 		return error;
4208 
4209 	/*
4210 	 * Update our extent pointer, given that
4211 	 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4212 	 * of the neighbouring ones.
4213 	 */
4214 	xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4215 
4216 	/*
4217 	 * We may have combined previously unwritten space with written space,
4218 	 * so generate another request.
4219 	 */
4220 	if (mval->br_blockcount < len)
4221 		return -EAGAIN;
4222 	return 0;
4223 }
4224 
4225 /*
4226  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4227  * extent state if necessary.  Details behaviour is controlled by the flags
4228  * parameter.  Only allocates blocks from a single allocation group, to avoid
4229  * locking problems.
4230  *
4231  * The returned value in "firstblock" from the first call in a transaction
4232  * must be remembered and presented to subsequent calls in "firstblock".
4233  * An upper bound for the number of blocks to be allocated is supplied to
4234  * the first call in "total"; if no allocation group has that many free
4235  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4236  */
4237 int
4238 xfs_bmapi_write(
4239 	struct xfs_trans	*tp,		/* transaction pointer */
4240 	struct xfs_inode	*ip,		/* incore inode */
4241 	xfs_fileoff_t		bno,		/* starting file offs. mapped */
4242 	xfs_filblks_t		len,		/* length to map in file */
4243 	int			flags,		/* XFS_BMAPI_... */
4244 	xfs_fsblock_t		*firstblock,	/* first allocated block
4245 						   controls a.g. for allocs */
4246 	xfs_extlen_t		total,		/* total blocks needed */
4247 	struct xfs_bmbt_irec	*mval,		/* output: map values */
4248 	int			*nmap,		/* i/o: mval size/count */
4249 	struct xfs_defer_ops	*dfops)		/* i/o: list extents to free */
4250 {
4251 	struct xfs_mount	*mp = ip->i_mount;
4252 	struct xfs_ifork	*ifp;
4253 	struct xfs_bmalloca	bma = { NULL };	/* args for xfs_bmap_alloc */
4254 	xfs_fileoff_t		end;		/* end of mapped file region */
4255 	bool			eof = false;	/* after the end of extents */
4256 	int			error;		/* error return */
4257 	int			n;		/* current extent index */
4258 	xfs_fileoff_t		obno;		/* old block number (offset) */
4259 	int			whichfork;	/* data or attr fork */
4260 
4261 #ifdef DEBUG
4262 	xfs_fileoff_t		orig_bno;	/* original block number value */
4263 	int			orig_flags;	/* original flags arg value */
4264 	xfs_filblks_t		orig_len;	/* original value of len arg */
4265 	struct xfs_bmbt_irec	*orig_mval;	/* original value of mval */
4266 	int			orig_nmap;	/* original value of *nmap */
4267 
4268 	orig_bno = bno;
4269 	orig_len = len;
4270 	orig_flags = flags;
4271 	orig_mval = mval;
4272 	orig_nmap = *nmap;
4273 #endif
4274 	whichfork = xfs_bmapi_whichfork(flags);
4275 
4276 	ASSERT(*nmap >= 1);
4277 	ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4278 	ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4279 	ASSERT(tp != NULL ||
4280 	       (flags & (XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK)) ==
4281 			(XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK));
4282 	ASSERT(len > 0);
4283 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4284 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4285 	ASSERT(!(flags & XFS_BMAPI_REMAP));
4286 
4287 	/* zeroing is for currently only for data extents, not metadata */
4288 	ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4289 			(XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4290 	/*
4291 	 * we can allocate unwritten extents or pre-zero allocated blocks,
4292 	 * but it makes no sense to do both at once. This would result in
4293 	 * zeroing the unwritten extent twice, but it still being an
4294 	 * unwritten extent....
4295 	 */
4296 	ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4297 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4298 
4299 	if (unlikely(XFS_TEST_ERROR(
4300 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4301 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4302 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
4303 		XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4304 		return -EFSCORRUPTED;
4305 	}
4306 
4307 	if (XFS_FORCED_SHUTDOWN(mp))
4308 		return -EIO;
4309 
4310 	ifp = XFS_IFORK_PTR(ip, whichfork);
4311 
4312 	XFS_STATS_INC(mp, xs_blk_mapw);
4313 
4314 	if (*firstblock == NULLFSBLOCK) {
4315 		if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4316 			bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4317 		else
4318 			bma.minleft = 1;
4319 	} else {
4320 		bma.minleft = 0;
4321 	}
4322 
4323 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4324 		error = xfs_iread_extents(tp, ip, whichfork);
4325 		if (error)
4326 			goto error0;
4327 	}
4328 
4329 	n = 0;
4330 	end = bno + len;
4331 	obno = bno;
4332 
4333 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.icur, &bma.got))
4334 		eof = true;
4335 	if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4336 		bma.prev.br_startoff = NULLFILEOFF;
4337 	bma.tp = tp;
4338 	bma.ip = ip;
4339 	bma.total = total;
4340 	bma.datatype = 0;
4341 	bma.dfops = dfops;
4342 	bma.firstblock = firstblock;
4343 
4344 	while (bno < end && n < *nmap) {
4345 		bool			need_alloc = false, wasdelay = false;
4346 
4347 		/* in hole or beyond EOF? */
4348 		if (eof || bma.got.br_startoff > bno) {
4349 			/*
4350 			 * CoW fork conversions should /never/ hit EOF or
4351 			 * holes.  There should always be something for us
4352 			 * to work on.
4353 			 */
4354 			ASSERT(!((flags & XFS_BMAPI_CONVERT) &&
4355 			         (flags & XFS_BMAPI_COWFORK)));
4356 
4357 			if (flags & XFS_BMAPI_DELALLOC) {
4358 				/*
4359 				 * For the COW fork we can reasonably get a
4360 				 * request for converting an extent that races
4361 				 * with other threads already having converted
4362 				 * part of it, as there converting COW to
4363 				 * regular blocks is not protected using the
4364 				 * IOLOCK.
4365 				 */
4366 				ASSERT(flags & XFS_BMAPI_COWFORK);
4367 				if (!(flags & XFS_BMAPI_COWFORK)) {
4368 					error = -EIO;
4369 					goto error0;
4370 				}
4371 
4372 				if (eof || bno >= end)
4373 					break;
4374 			} else {
4375 				need_alloc = true;
4376 			}
4377 		} else if (isnullstartblock(bma.got.br_startblock)) {
4378 			wasdelay = true;
4379 		}
4380 
4381 		/*
4382 		 * First, deal with the hole before the allocated space
4383 		 * that we found, if any.
4384 		 */
4385 		if ((need_alloc || wasdelay) &&
4386 		    !(flags & XFS_BMAPI_CONVERT_ONLY)) {
4387 			bma.eof = eof;
4388 			bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4389 			bma.wasdel = wasdelay;
4390 			bma.offset = bno;
4391 			bma.flags = flags;
4392 
4393 			/*
4394 			 * There's a 32/64 bit type mismatch between the
4395 			 * allocation length request (which can be 64 bits in
4396 			 * length) and the bma length request, which is
4397 			 * xfs_extlen_t and therefore 32 bits. Hence we have to
4398 			 * check for 32-bit overflows and handle them here.
4399 			 */
4400 			if (len > (xfs_filblks_t)MAXEXTLEN)
4401 				bma.length = MAXEXTLEN;
4402 			else
4403 				bma.length = len;
4404 
4405 			ASSERT(len > 0);
4406 			ASSERT(bma.length > 0);
4407 			error = xfs_bmapi_allocate(&bma);
4408 			if (error)
4409 				goto error0;
4410 			if (bma.blkno == NULLFSBLOCK)
4411 				break;
4412 
4413 			/*
4414 			 * If this is a CoW allocation, record the data in
4415 			 * the refcount btree for orphan recovery.
4416 			 */
4417 			if (whichfork == XFS_COW_FORK) {
4418 				error = xfs_refcount_alloc_cow_extent(mp, dfops,
4419 						bma.blkno, bma.length);
4420 				if (error)
4421 					goto error0;
4422 			}
4423 		}
4424 
4425 		/* Deal with the allocated space we found.  */
4426 		xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4427 							end, n, flags);
4428 
4429 		/* Execute unwritten extent conversion if necessary */
4430 		error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4431 		if (error == -EAGAIN)
4432 			continue;
4433 		if (error)
4434 			goto error0;
4435 
4436 		/* update the extent map to return */
4437 		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4438 
4439 		/*
4440 		 * If we're done, stop now.  Stop when we've allocated
4441 		 * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4442 		 * the transaction may get too big.
4443 		 */
4444 		if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4445 			break;
4446 
4447 		/* Else go on to the next record. */
4448 		bma.prev = bma.got;
4449 		if (!xfs_iext_next_extent(ifp, &bma.icur, &bma.got))
4450 			eof = true;
4451 	}
4452 	*nmap = n;
4453 
4454 	/*
4455 	 * Transform from btree to extents, give it cur.
4456 	 */
4457 	if (xfs_bmap_wants_extents(ip, whichfork)) {
4458 		int		tmp_logflags = 0;
4459 
4460 		ASSERT(bma.cur);
4461 		error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4462 			&tmp_logflags, whichfork);
4463 		bma.logflags |= tmp_logflags;
4464 		if (error)
4465 			goto error0;
4466 	}
4467 
4468 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4469 	       XFS_IFORK_NEXTENTS(ip, whichfork) >
4470 		XFS_IFORK_MAXEXT(ip, whichfork));
4471 	error = 0;
4472 error0:
4473 	/*
4474 	 * Log everything.  Do this after conversion, there's no point in
4475 	 * logging the extent records if we've converted to btree format.
4476 	 */
4477 	if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4478 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4479 		bma.logflags &= ~xfs_ilog_fext(whichfork);
4480 	else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4481 		 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4482 		bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4483 	/*
4484 	 * Log whatever the flags say, even if error.  Otherwise we might miss
4485 	 * detecting a case where the data is changed, there's an error,
4486 	 * and it's not logged so we don't shutdown when we should.
4487 	 */
4488 	if (bma.logflags)
4489 		xfs_trans_log_inode(tp, ip, bma.logflags);
4490 
4491 	if (bma.cur) {
4492 		if (!error) {
4493 			ASSERT(*firstblock == NULLFSBLOCK ||
4494 			       XFS_FSB_TO_AGNO(mp, *firstblock) <=
4495 			       XFS_FSB_TO_AGNO(mp,
4496 				       bma.cur->bc_private.b.firstblock));
4497 			*firstblock = bma.cur->bc_private.b.firstblock;
4498 		}
4499 		xfs_btree_del_cursor(bma.cur,
4500 			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4501 	}
4502 	if (!error)
4503 		xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4504 			orig_nmap, *nmap);
4505 	return error;
4506 }
4507 
4508 static int
4509 xfs_bmapi_remap(
4510 	struct xfs_trans	*tp,
4511 	struct xfs_inode	*ip,
4512 	xfs_fileoff_t		bno,
4513 	xfs_filblks_t		len,
4514 	xfs_fsblock_t		startblock,
4515 	struct xfs_defer_ops	*dfops)
4516 {
4517 	struct xfs_mount	*mp = ip->i_mount;
4518 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4519 	struct xfs_btree_cur	*cur = NULL;
4520 	xfs_fsblock_t		firstblock = NULLFSBLOCK;
4521 	struct xfs_bmbt_irec	got;
4522 	struct xfs_iext_cursor	icur;
4523 	int			logflags = 0, error;
4524 
4525 	ASSERT(len > 0);
4526 	ASSERT(len <= (xfs_filblks_t)MAXEXTLEN);
4527 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4528 
4529 	if (unlikely(XFS_TEST_ERROR(
4530 	    (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4531 	     XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4532 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
4533 		XFS_ERROR_REPORT("xfs_bmapi_remap", XFS_ERRLEVEL_LOW, mp);
4534 		return -EFSCORRUPTED;
4535 	}
4536 
4537 	if (XFS_FORCED_SHUTDOWN(mp))
4538 		return -EIO;
4539 
4540 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4541 		error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4542 		if (error)
4543 			return error;
4544 	}
4545 
4546 	if (xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
4547 		/* make sure we only reflink into a hole. */
4548 		ASSERT(got.br_startoff > bno);
4549 		ASSERT(got.br_startoff - bno >= len);
4550 	}
4551 
4552 	ip->i_d.di_nblocks += len;
4553 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4554 
4555 	if (ifp->if_flags & XFS_IFBROOT) {
4556 		cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
4557 		cur->bc_private.b.firstblock = firstblock;
4558 		cur->bc_private.b.dfops = dfops;
4559 		cur->bc_private.b.flags = 0;
4560 	}
4561 
4562 	got.br_startoff = bno;
4563 	got.br_startblock = startblock;
4564 	got.br_blockcount = len;
4565 	got.br_state = XFS_EXT_NORM;
4566 
4567 	error = xfs_bmap_add_extent_hole_real(tp, ip, XFS_DATA_FORK, &icur,
4568 			&cur, &got, &firstblock, dfops, &logflags);
4569 	if (error)
4570 		goto error0;
4571 
4572 	if (xfs_bmap_wants_extents(ip, XFS_DATA_FORK)) {
4573 		int		tmp_logflags = 0;
4574 
4575 		error = xfs_bmap_btree_to_extents(tp, ip, cur,
4576 			&tmp_logflags, XFS_DATA_FORK);
4577 		logflags |= tmp_logflags;
4578 	}
4579 
4580 error0:
4581 	if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS)
4582 		logflags &= ~XFS_ILOG_DEXT;
4583 	else if (ip->i_d.di_format != XFS_DINODE_FMT_BTREE)
4584 		logflags &= ~XFS_ILOG_DBROOT;
4585 
4586 	if (logflags)
4587 		xfs_trans_log_inode(tp, ip, logflags);
4588 	if (cur) {
4589 		xfs_btree_del_cursor(cur,
4590 				error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4591 	}
4592 	return error;
4593 }
4594 
4595 /*
4596  * When a delalloc extent is split (e.g., due to a hole punch), the original
4597  * indlen reservation must be shared across the two new extents that are left
4598  * behind.
4599  *
4600  * Given the original reservation and the worst case indlen for the two new
4601  * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4602  * reservation fairly across the two new extents. If necessary, steal available
4603  * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4604  * ores == 1). The number of stolen blocks is returned. The availability and
4605  * subsequent accounting of stolen blocks is the responsibility of the caller.
4606  */
4607 static xfs_filblks_t
4608 xfs_bmap_split_indlen(
4609 	xfs_filblks_t			ores,		/* original res. */
4610 	xfs_filblks_t			*indlen1,	/* ext1 worst indlen */
4611 	xfs_filblks_t			*indlen2,	/* ext2 worst indlen */
4612 	xfs_filblks_t			avail)		/* stealable blocks */
4613 {
4614 	xfs_filblks_t			len1 = *indlen1;
4615 	xfs_filblks_t			len2 = *indlen2;
4616 	xfs_filblks_t			nres = len1 + len2; /* new total res. */
4617 	xfs_filblks_t			stolen = 0;
4618 	xfs_filblks_t			resfactor;
4619 
4620 	/*
4621 	 * Steal as many blocks as we can to try and satisfy the worst case
4622 	 * indlen for both new extents.
4623 	 */
4624 	if (ores < nres && avail)
4625 		stolen = XFS_FILBLKS_MIN(nres - ores, avail);
4626 	ores += stolen;
4627 
4628 	 /* nothing else to do if we've satisfied the new reservation */
4629 	if (ores >= nres)
4630 		return stolen;
4631 
4632 	/*
4633 	 * We can't meet the total required reservation for the two extents.
4634 	 * Calculate the percent of the overall shortage between both extents
4635 	 * and apply this percentage to each of the requested indlen values.
4636 	 * This distributes the shortage fairly and reduces the chances that one
4637 	 * of the two extents is left with nothing when extents are repeatedly
4638 	 * split.
4639 	 */
4640 	resfactor = (ores * 100);
4641 	do_div(resfactor, nres);
4642 	len1 *= resfactor;
4643 	do_div(len1, 100);
4644 	len2 *= resfactor;
4645 	do_div(len2, 100);
4646 	ASSERT(len1 + len2 <= ores);
4647 	ASSERT(len1 < *indlen1 && len2 < *indlen2);
4648 
4649 	/*
4650 	 * Hand out the remainder to each extent. If one of the two reservations
4651 	 * is zero, we want to make sure that one gets a block first. The loop
4652 	 * below starts with len1, so hand len2 a block right off the bat if it
4653 	 * is zero.
4654 	 */
4655 	ores -= (len1 + len2);
4656 	ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores);
4657 	if (ores && !len2 && *indlen2) {
4658 		len2++;
4659 		ores--;
4660 	}
4661 	while (ores) {
4662 		if (len1 < *indlen1) {
4663 			len1++;
4664 			ores--;
4665 		}
4666 		if (!ores)
4667 			break;
4668 		if (len2 < *indlen2) {
4669 			len2++;
4670 			ores--;
4671 		}
4672 	}
4673 
4674 	*indlen1 = len1;
4675 	*indlen2 = len2;
4676 
4677 	return stolen;
4678 }
4679 
4680 int
4681 xfs_bmap_del_extent_delay(
4682 	struct xfs_inode	*ip,
4683 	int			whichfork,
4684 	struct xfs_iext_cursor	*icur,
4685 	struct xfs_bmbt_irec	*got,
4686 	struct xfs_bmbt_irec	*del)
4687 {
4688 	struct xfs_mount	*mp = ip->i_mount;
4689 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
4690 	struct xfs_bmbt_irec	new;
4691 	int64_t			da_old, da_new, da_diff = 0;
4692 	xfs_fileoff_t		del_endoff, got_endoff;
4693 	xfs_filblks_t		got_indlen, new_indlen, stolen;
4694 	int			state = xfs_bmap_fork_to_state(whichfork);
4695 	int			error = 0;
4696 	bool			isrt;
4697 
4698 	XFS_STATS_INC(mp, xs_del_exlist);
4699 
4700 	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4701 	del_endoff = del->br_startoff + del->br_blockcount;
4702 	got_endoff = got->br_startoff + got->br_blockcount;
4703 	da_old = startblockval(got->br_startblock);
4704 	da_new = 0;
4705 
4706 	ASSERT(del->br_blockcount > 0);
4707 	ASSERT(got->br_startoff <= del->br_startoff);
4708 	ASSERT(got_endoff >= del_endoff);
4709 
4710 	if (isrt) {
4711 		uint64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount);
4712 
4713 		do_div(rtexts, mp->m_sb.sb_rextsize);
4714 		xfs_mod_frextents(mp, rtexts);
4715 	}
4716 
4717 	/*
4718 	 * Update the inode delalloc counter now and wait to update the
4719 	 * sb counters as we might have to borrow some blocks for the
4720 	 * indirect block accounting.
4721 	 */
4722 	error = xfs_trans_reserve_quota_nblks(NULL, ip,
4723 			-((long)del->br_blockcount), 0,
4724 			isrt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4725 	if (error)
4726 		return error;
4727 	ip->i_delayed_blks -= del->br_blockcount;
4728 
4729 	if (got->br_startoff == del->br_startoff)
4730 		state |= BMAP_LEFT_FILLING;
4731 	if (got_endoff == del_endoff)
4732 		state |= BMAP_RIGHT_FILLING;
4733 
4734 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4735 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4736 		/*
4737 		 * Matches the whole extent.  Delete the entry.
4738 		 */
4739 		xfs_iext_remove(ip, icur, state);
4740 		xfs_iext_prev(ifp, icur);
4741 		break;
4742 	case BMAP_LEFT_FILLING:
4743 		/*
4744 		 * Deleting the first part of the extent.
4745 		 */
4746 		got->br_startoff = del_endoff;
4747 		got->br_blockcount -= del->br_blockcount;
4748 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4749 				got->br_blockcount), da_old);
4750 		got->br_startblock = nullstartblock((int)da_new);
4751 		xfs_iext_update_extent(ip, state, icur, got);
4752 		break;
4753 	case BMAP_RIGHT_FILLING:
4754 		/*
4755 		 * Deleting the last part of the extent.
4756 		 */
4757 		got->br_blockcount = got->br_blockcount - del->br_blockcount;
4758 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4759 				got->br_blockcount), da_old);
4760 		got->br_startblock = nullstartblock((int)da_new);
4761 		xfs_iext_update_extent(ip, state, icur, got);
4762 		break;
4763 	case 0:
4764 		/*
4765 		 * Deleting the middle of the extent.
4766 		 *
4767 		 * Distribute the original indlen reservation across the two new
4768 		 * extents.  Steal blocks from the deleted extent if necessary.
4769 		 * Stealing blocks simply fudges the fdblocks accounting below.
4770 		 * Warn if either of the new indlen reservations is zero as this
4771 		 * can lead to delalloc problems.
4772 		 */
4773 		got->br_blockcount = del->br_startoff - got->br_startoff;
4774 		got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount);
4775 
4776 		new.br_blockcount = got_endoff - del_endoff;
4777 		new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount);
4778 
4779 		WARN_ON_ONCE(!got_indlen || !new_indlen);
4780 		stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen,
4781 						       del->br_blockcount);
4782 
4783 		got->br_startblock = nullstartblock((int)got_indlen);
4784 
4785 		new.br_startoff = del_endoff;
4786 		new.br_state = got->br_state;
4787 		new.br_startblock = nullstartblock((int)new_indlen);
4788 
4789 		xfs_iext_update_extent(ip, state, icur, got);
4790 		xfs_iext_next(ifp, icur);
4791 		xfs_iext_insert(ip, icur, &new, state);
4792 
4793 		da_new = got_indlen + new_indlen - stolen;
4794 		del->br_blockcount -= stolen;
4795 		break;
4796 	}
4797 
4798 	ASSERT(da_old >= da_new);
4799 	da_diff = da_old - da_new;
4800 	if (!isrt)
4801 		da_diff += del->br_blockcount;
4802 	if (da_diff)
4803 		xfs_mod_fdblocks(mp, da_diff, false);
4804 	return error;
4805 }
4806 
4807 void
4808 xfs_bmap_del_extent_cow(
4809 	struct xfs_inode	*ip,
4810 	struct xfs_iext_cursor	*icur,
4811 	struct xfs_bmbt_irec	*got,
4812 	struct xfs_bmbt_irec	*del)
4813 {
4814 	struct xfs_mount	*mp = ip->i_mount;
4815 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
4816 	struct xfs_bmbt_irec	new;
4817 	xfs_fileoff_t		del_endoff, got_endoff;
4818 	int			state = BMAP_COWFORK;
4819 
4820 	XFS_STATS_INC(mp, xs_del_exlist);
4821 
4822 	del_endoff = del->br_startoff + del->br_blockcount;
4823 	got_endoff = got->br_startoff + got->br_blockcount;
4824 
4825 	ASSERT(del->br_blockcount > 0);
4826 	ASSERT(got->br_startoff <= del->br_startoff);
4827 	ASSERT(got_endoff >= del_endoff);
4828 	ASSERT(!isnullstartblock(got->br_startblock));
4829 
4830 	if (got->br_startoff == del->br_startoff)
4831 		state |= BMAP_LEFT_FILLING;
4832 	if (got_endoff == del_endoff)
4833 		state |= BMAP_RIGHT_FILLING;
4834 
4835 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4836 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4837 		/*
4838 		 * Matches the whole extent.  Delete the entry.
4839 		 */
4840 		xfs_iext_remove(ip, icur, state);
4841 		xfs_iext_prev(ifp, icur);
4842 		break;
4843 	case BMAP_LEFT_FILLING:
4844 		/*
4845 		 * Deleting the first part of the extent.
4846 		 */
4847 		got->br_startoff = del_endoff;
4848 		got->br_blockcount -= del->br_blockcount;
4849 		got->br_startblock = del->br_startblock + del->br_blockcount;
4850 		xfs_iext_update_extent(ip, state, icur, got);
4851 		break;
4852 	case BMAP_RIGHT_FILLING:
4853 		/*
4854 		 * Deleting the last part of the extent.
4855 		 */
4856 		got->br_blockcount -= del->br_blockcount;
4857 		xfs_iext_update_extent(ip, state, icur, got);
4858 		break;
4859 	case 0:
4860 		/*
4861 		 * Deleting the middle of the extent.
4862 		 */
4863 		got->br_blockcount = del->br_startoff - got->br_startoff;
4864 
4865 		new.br_startoff = del_endoff;
4866 		new.br_blockcount = got_endoff - del_endoff;
4867 		new.br_state = got->br_state;
4868 		new.br_startblock = del->br_startblock + del->br_blockcount;
4869 
4870 		xfs_iext_update_extent(ip, state, icur, got);
4871 		xfs_iext_next(ifp, icur);
4872 		xfs_iext_insert(ip, icur, &new, state);
4873 		break;
4874 	}
4875 	ip->i_delayed_blks -= del->br_blockcount;
4876 }
4877 
4878 /*
4879  * Called by xfs_bmapi to update file extent records and the btree
4880  * after removing space.
4881  */
4882 STATIC int				/* error */
4883 xfs_bmap_del_extent_real(
4884 	xfs_inode_t		*ip,	/* incore inode pointer */
4885 	xfs_trans_t		*tp,	/* current transaction pointer */
4886 	struct xfs_iext_cursor	*icur,
4887 	struct xfs_defer_ops	*dfops,	/* list of extents to be freed */
4888 	xfs_btree_cur_t		*cur,	/* if null, not a btree */
4889 	xfs_bmbt_irec_t		*del,	/* data to remove from extents */
4890 	int			*logflagsp, /* inode logging flags */
4891 	int			whichfork, /* data or attr fork */
4892 	int			bflags)	/* bmapi flags */
4893 {
4894 	xfs_fsblock_t		del_endblock=0;	/* first block past del */
4895 	xfs_fileoff_t		del_endoff;	/* first offset past del */
4896 	int			do_fx;	/* free extent at end of routine */
4897 	int			error;	/* error return value */
4898 	int			flags = 0;/* inode logging flags */
4899 	struct xfs_bmbt_irec	got;	/* current extent entry */
4900 	xfs_fileoff_t		got_endoff;	/* first offset past got */
4901 	int			i;	/* temp state */
4902 	xfs_ifork_t		*ifp;	/* inode fork pointer */
4903 	xfs_mount_t		*mp;	/* mount structure */
4904 	xfs_filblks_t		nblks;	/* quota/sb block count */
4905 	xfs_bmbt_irec_t		new;	/* new record to be inserted */
4906 	/* REFERENCED */
4907 	uint			qfield;	/* quota field to update */
4908 	int			state = xfs_bmap_fork_to_state(whichfork);
4909 	struct xfs_bmbt_irec	old;
4910 
4911 	mp = ip->i_mount;
4912 	XFS_STATS_INC(mp, xs_del_exlist);
4913 
4914 	ifp = XFS_IFORK_PTR(ip, whichfork);
4915 	ASSERT(del->br_blockcount > 0);
4916 	xfs_iext_get_extent(ifp, icur, &got);
4917 	ASSERT(got.br_startoff <= del->br_startoff);
4918 	del_endoff = del->br_startoff + del->br_blockcount;
4919 	got_endoff = got.br_startoff + got.br_blockcount;
4920 	ASSERT(got_endoff >= del_endoff);
4921 	ASSERT(!isnullstartblock(got.br_startblock));
4922 	qfield = 0;
4923 	error = 0;
4924 
4925 	/*
4926 	 * If it's the case where the directory code is running with no block
4927 	 * reservation, and the deleted block is in the middle of its extent,
4928 	 * and the resulting insert of an extent would cause transformation to
4929 	 * btree format, then reject it.  The calling code will then swap blocks
4930 	 * around instead.  We have to do this now, rather than waiting for the
4931 	 * conversion to btree format, since the transaction will be dirty then.
4932 	 */
4933 	if (tp->t_blk_res == 0 &&
4934 	    XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
4935 	    XFS_IFORK_NEXTENTS(ip, whichfork) >=
4936 			XFS_IFORK_MAXEXT(ip, whichfork) &&
4937 	    del->br_startoff > got.br_startoff && del_endoff < got_endoff)
4938 		return -ENOSPC;
4939 
4940 	flags = XFS_ILOG_CORE;
4941 	if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
4942 		xfs_fsblock_t	bno;
4943 		xfs_filblks_t	len;
4944 
4945 		ASSERT(do_mod(del->br_blockcount, mp->m_sb.sb_rextsize) == 0);
4946 		ASSERT(do_mod(del->br_startblock, mp->m_sb.sb_rextsize) == 0);
4947 		bno = del->br_startblock;
4948 		len = del->br_blockcount;
4949 		do_div(bno, mp->m_sb.sb_rextsize);
4950 		do_div(len, mp->m_sb.sb_rextsize);
4951 		error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
4952 		if (error)
4953 			goto done;
4954 		do_fx = 0;
4955 		nblks = len * mp->m_sb.sb_rextsize;
4956 		qfield = XFS_TRANS_DQ_RTBCOUNT;
4957 	} else {
4958 		do_fx = 1;
4959 		nblks = del->br_blockcount;
4960 		qfield = XFS_TRANS_DQ_BCOUNT;
4961 	}
4962 
4963 	del_endblock = del->br_startblock + del->br_blockcount;
4964 	if (cur) {
4965 		error = xfs_bmbt_lookup_eq(cur, &got, &i);
4966 		if (error)
4967 			goto done;
4968 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
4969 	}
4970 
4971 	if (got.br_startoff == del->br_startoff)
4972 		state |= BMAP_LEFT_FILLING;
4973 	if (got_endoff == del_endoff)
4974 		state |= BMAP_RIGHT_FILLING;
4975 
4976 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4977 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4978 		/*
4979 		 * Matches the whole extent.  Delete the entry.
4980 		 */
4981 		xfs_iext_remove(ip, icur, state);
4982 		xfs_iext_prev(ifp, icur);
4983 		XFS_IFORK_NEXT_SET(ip, whichfork,
4984 			XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
4985 		flags |= XFS_ILOG_CORE;
4986 		if (!cur) {
4987 			flags |= xfs_ilog_fext(whichfork);
4988 			break;
4989 		}
4990 		if ((error = xfs_btree_delete(cur, &i)))
4991 			goto done;
4992 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
4993 		break;
4994 	case BMAP_LEFT_FILLING:
4995 		/*
4996 		 * Deleting the first part of the extent.
4997 		 */
4998 		got.br_startoff = del_endoff;
4999 		got.br_startblock = del_endblock;
5000 		got.br_blockcount -= del->br_blockcount;
5001 		xfs_iext_update_extent(ip, state, icur, &got);
5002 		if (!cur) {
5003 			flags |= xfs_ilog_fext(whichfork);
5004 			break;
5005 		}
5006 		error = xfs_bmbt_update(cur, &got);
5007 		if (error)
5008 			goto done;
5009 		break;
5010 	case BMAP_RIGHT_FILLING:
5011 		/*
5012 		 * Deleting the last part of the extent.
5013 		 */
5014 		got.br_blockcount -= del->br_blockcount;
5015 		xfs_iext_update_extent(ip, state, icur, &got);
5016 		if (!cur) {
5017 			flags |= xfs_ilog_fext(whichfork);
5018 			break;
5019 		}
5020 		error = xfs_bmbt_update(cur, &got);
5021 		if (error)
5022 			goto done;
5023 		break;
5024 	case 0:
5025 		/*
5026 		 * Deleting the middle of the extent.
5027 		 */
5028 		old = got;
5029 
5030 		got.br_blockcount = del->br_startoff - got.br_startoff;
5031 		xfs_iext_update_extent(ip, state, icur, &got);
5032 
5033 		new.br_startoff = del_endoff;
5034 		new.br_blockcount = got_endoff - del_endoff;
5035 		new.br_state = got.br_state;
5036 		new.br_startblock = del_endblock;
5037 
5038 		flags |= XFS_ILOG_CORE;
5039 		if (cur) {
5040 			error = xfs_bmbt_update(cur, &got);
5041 			if (error)
5042 				goto done;
5043 			error = xfs_btree_increment(cur, 0, &i);
5044 			if (error)
5045 				goto done;
5046 			cur->bc_rec.b = new;
5047 			error = xfs_btree_insert(cur, &i);
5048 			if (error && error != -ENOSPC)
5049 				goto done;
5050 			/*
5051 			 * If get no-space back from btree insert, it tried a
5052 			 * split, and we have a zero block reservation.  Fix up
5053 			 * our state and return the error.
5054 			 */
5055 			if (error == -ENOSPC) {
5056 				/*
5057 				 * Reset the cursor, don't trust it after any
5058 				 * insert operation.
5059 				 */
5060 				error = xfs_bmbt_lookup_eq(cur, &got, &i);
5061 				if (error)
5062 					goto done;
5063 				XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5064 				/*
5065 				 * Update the btree record back
5066 				 * to the original value.
5067 				 */
5068 				error = xfs_bmbt_update(cur, &old);
5069 				if (error)
5070 					goto done;
5071 				/*
5072 				 * Reset the extent record back
5073 				 * to the original value.
5074 				 */
5075 				xfs_iext_update_extent(ip, state, icur, &old);
5076 				flags = 0;
5077 				error = -ENOSPC;
5078 				goto done;
5079 			}
5080 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5081 		} else
5082 			flags |= xfs_ilog_fext(whichfork);
5083 		XFS_IFORK_NEXT_SET(ip, whichfork,
5084 			XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5085 		xfs_iext_next(ifp, icur);
5086 		xfs_iext_insert(ip, icur, &new, state);
5087 		break;
5088 	}
5089 
5090 	/* remove reverse mapping */
5091 	error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, del);
5092 	if (error)
5093 		goto done;
5094 
5095 	/*
5096 	 * If we need to, add to list of extents to delete.
5097 	 */
5098 	if (do_fx && !(bflags & XFS_BMAPI_REMAP)) {
5099 		if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
5100 			error = xfs_refcount_decrease_extent(mp, dfops, del);
5101 			if (error)
5102 				goto done;
5103 		} else
5104 			xfs_bmap_add_free(mp, dfops, del->br_startblock,
5105 					del->br_blockcount, NULL);
5106 	}
5107 
5108 	/*
5109 	 * Adjust inode # blocks in the file.
5110 	 */
5111 	if (nblks)
5112 		ip->i_d.di_nblocks -= nblks;
5113 	/*
5114 	 * Adjust quota data.
5115 	 */
5116 	if (qfield && !(bflags & XFS_BMAPI_REMAP))
5117 		xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5118 
5119 done:
5120 	*logflagsp = flags;
5121 	return error;
5122 }
5123 
5124 /*
5125  * Unmap (remove) blocks from a file.
5126  * If nexts is nonzero then the number of extents to remove is limited to
5127  * that value.  If not all extents in the block range can be removed then
5128  * *done is set.
5129  */
5130 int						/* error */
5131 __xfs_bunmapi(
5132 	xfs_trans_t		*tp,		/* transaction pointer */
5133 	struct xfs_inode	*ip,		/* incore inode */
5134 	xfs_fileoff_t		start,		/* first file offset deleted */
5135 	xfs_filblks_t		*rlen,		/* i/o: amount remaining */
5136 	int			flags,		/* misc flags */
5137 	xfs_extnum_t		nexts,		/* number of extents max */
5138 	xfs_fsblock_t		*firstblock,	/* first allocated block
5139 						   controls a.g. for allocs */
5140 	struct xfs_defer_ops	*dfops)		/* i/o: deferred updates */
5141 {
5142 	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
5143 	xfs_bmbt_irec_t		del;		/* extent being deleted */
5144 	int			error;		/* error return value */
5145 	xfs_extnum_t		extno;		/* extent number in list */
5146 	xfs_bmbt_irec_t		got;		/* current extent record */
5147 	xfs_ifork_t		*ifp;		/* inode fork pointer */
5148 	int			isrt;		/* freeing in rt area */
5149 	int			logflags;	/* transaction logging flags */
5150 	xfs_extlen_t		mod;		/* rt extent offset */
5151 	xfs_mount_t		*mp;		/* mount structure */
5152 	int			tmp_logflags;	/* partial logging flags */
5153 	int			wasdel;		/* was a delayed alloc extent */
5154 	int			whichfork;	/* data or attribute fork */
5155 	xfs_fsblock_t		sum;
5156 	xfs_filblks_t		len = *rlen;	/* length to unmap in file */
5157 	xfs_fileoff_t		max_len;
5158 	xfs_agnumber_t		prev_agno = NULLAGNUMBER, agno;
5159 	xfs_fileoff_t		end;
5160 	struct xfs_iext_cursor	icur;
5161 	bool			done = false;
5162 
5163 	trace_xfs_bunmap(ip, start, len, flags, _RET_IP_);
5164 
5165 	whichfork = xfs_bmapi_whichfork(flags);
5166 	ASSERT(whichfork != XFS_COW_FORK);
5167 	ifp = XFS_IFORK_PTR(ip, whichfork);
5168 	if (unlikely(
5169 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5170 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5171 		XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5172 				 ip->i_mount);
5173 		return -EFSCORRUPTED;
5174 	}
5175 	mp = ip->i_mount;
5176 	if (XFS_FORCED_SHUTDOWN(mp))
5177 		return -EIO;
5178 
5179 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5180 	ASSERT(len > 0);
5181 	ASSERT(nexts >= 0);
5182 
5183 	/*
5184 	 * Guesstimate how many blocks we can unmap without running the risk of
5185 	 * blowing out the transaction with a mix of EFIs and reflink
5186 	 * adjustments.
5187 	 */
5188 	if (tp && xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK)
5189 		max_len = min(len, xfs_refcount_max_unmap(tp->t_log_res));
5190 	else
5191 		max_len = len;
5192 
5193 	if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5194 	    (error = xfs_iread_extents(tp, ip, whichfork)))
5195 		return error;
5196 	if (xfs_iext_count(ifp) == 0) {
5197 		*rlen = 0;
5198 		return 0;
5199 	}
5200 	XFS_STATS_INC(mp, xs_blk_unmap);
5201 	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5202 	end = start + len;
5203 
5204 	if (!xfs_iext_lookup_extent_before(ip, ifp, &end, &icur, &got)) {
5205 		*rlen = 0;
5206 		return 0;
5207 	}
5208 	end--;
5209 
5210 	logflags = 0;
5211 	if (ifp->if_flags & XFS_IFBROOT) {
5212 		ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5213 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5214 		cur->bc_private.b.firstblock = *firstblock;
5215 		cur->bc_private.b.dfops = dfops;
5216 		cur->bc_private.b.flags = 0;
5217 	} else
5218 		cur = NULL;
5219 
5220 	if (isrt) {
5221 		/*
5222 		 * Synchronize by locking the bitmap inode.
5223 		 */
5224 		xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
5225 		xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5226 		xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
5227 		xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
5228 	}
5229 
5230 	extno = 0;
5231 	while (end != (xfs_fileoff_t)-1 && end >= start &&
5232 	       (nexts == 0 || extno < nexts) && max_len > 0) {
5233 		/*
5234 		 * Is the found extent after a hole in which end lives?
5235 		 * Just back up to the previous extent, if so.
5236 		 */
5237 		if (got.br_startoff > end &&
5238 		    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5239 			done = true;
5240 			break;
5241 		}
5242 		/*
5243 		 * Is the last block of this extent before the range
5244 		 * we're supposed to delete?  If so, we're done.
5245 		 */
5246 		end = XFS_FILEOFF_MIN(end,
5247 			got.br_startoff + got.br_blockcount - 1);
5248 		if (end < start)
5249 			break;
5250 		/*
5251 		 * Then deal with the (possibly delayed) allocated space
5252 		 * we found.
5253 		 */
5254 		del = got;
5255 		wasdel = isnullstartblock(del.br_startblock);
5256 
5257 		/*
5258 		 * Make sure we don't touch multiple AGF headers out of order
5259 		 * in a single transaction, as that could cause AB-BA deadlocks.
5260 		 */
5261 		if (!wasdel) {
5262 			agno = XFS_FSB_TO_AGNO(mp, del.br_startblock);
5263 			if (prev_agno != NULLAGNUMBER && prev_agno > agno)
5264 				break;
5265 			prev_agno = agno;
5266 		}
5267 		if (got.br_startoff < start) {
5268 			del.br_startoff = start;
5269 			del.br_blockcount -= start - got.br_startoff;
5270 			if (!wasdel)
5271 				del.br_startblock += start - got.br_startoff;
5272 		}
5273 		if (del.br_startoff + del.br_blockcount > end + 1)
5274 			del.br_blockcount = end + 1 - del.br_startoff;
5275 
5276 		/* How much can we safely unmap? */
5277 		if (max_len < del.br_blockcount) {
5278 			del.br_startoff += del.br_blockcount - max_len;
5279 			if (!wasdel)
5280 				del.br_startblock += del.br_blockcount - max_len;
5281 			del.br_blockcount = max_len;
5282 		}
5283 
5284 		sum = del.br_startblock + del.br_blockcount;
5285 		if (isrt &&
5286 		    (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5287 			/*
5288 			 * Realtime extent not lined up at the end.
5289 			 * The extent could have been split into written
5290 			 * and unwritten pieces, or we could just be
5291 			 * unmapping part of it.  But we can't really
5292 			 * get rid of part of a realtime extent.
5293 			 */
5294 			if (del.br_state == XFS_EXT_UNWRITTEN ||
5295 			    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5296 				/*
5297 				 * This piece is unwritten, or we're not
5298 				 * using unwritten extents.  Skip over it.
5299 				 */
5300 				ASSERT(end >= mod);
5301 				end -= mod > del.br_blockcount ?
5302 					del.br_blockcount : mod;
5303 				if (end < got.br_startoff &&
5304 				    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5305 					done = true;
5306 					break;
5307 				}
5308 				continue;
5309 			}
5310 			/*
5311 			 * It's written, turn it unwritten.
5312 			 * This is better than zeroing it.
5313 			 */
5314 			ASSERT(del.br_state == XFS_EXT_NORM);
5315 			ASSERT(tp->t_blk_res > 0);
5316 			/*
5317 			 * If this spans a realtime extent boundary,
5318 			 * chop it back to the start of the one we end at.
5319 			 */
5320 			if (del.br_blockcount > mod) {
5321 				del.br_startoff += del.br_blockcount - mod;
5322 				del.br_startblock += del.br_blockcount - mod;
5323 				del.br_blockcount = mod;
5324 			}
5325 			del.br_state = XFS_EXT_UNWRITTEN;
5326 			error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5327 					whichfork, &icur, &cur, &del,
5328 					firstblock, dfops, &logflags);
5329 			if (error)
5330 				goto error0;
5331 			goto nodelete;
5332 		}
5333 		if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5334 			/*
5335 			 * Realtime extent is lined up at the end but not
5336 			 * at the front.  We'll get rid of full extents if
5337 			 * we can.
5338 			 */
5339 			mod = mp->m_sb.sb_rextsize - mod;
5340 			if (del.br_blockcount > mod) {
5341 				del.br_blockcount -= mod;
5342 				del.br_startoff += mod;
5343 				del.br_startblock += mod;
5344 			} else if ((del.br_startoff == start &&
5345 				    (del.br_state == XFS_EXT_UNWRITTEN ||
5346 				     tp->t_blk_res == 0)) ||
5347 				   !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5348 				/*
5349 				 * Can't make it unwritten.  There isn't
5350 				 * a full extent here so just skip it.
5351 				 */
5352 				ASSERT(end >= del.br_blockcount);
5353 				end -= del.br_blockcount;
5354 				if (got.br_startoff > end &&
5355 				    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5356 					done = true;
5357 					break;
5358 				}
5359 				continue;
5360 			} else if (del.br_state == XFS_EXT_UNWRITTEN) {
5361 				struct xfs_bmbt_irec	prev;
5362 
5363 				/*
5364 				 * This one is already unwritten.
5365 				 * It must have a written left neighbor.
5366 				 * Unwrite the killed part of that one and
5367 				 * try again.
5368 				 */
5369 				if (!xfs_iext_prev_extent(ifp, &icur, &prev))
5370 					ASSERT(0);
5371 				ASSERT(prev.br_state == XFS_EXT_NORM);
5372 				ASSERT(!isnullstartblock(prev.br_startblock));
5373 				ASSERT(del.br_startblock ==
5374 				       prev.br_startblock + prev.br_blockcount);
5375 				if (prev.br_startoff < start) {
5376 					mod = start - prev.br_startoff;
5377 					prev.br_blockcount -= mod;
5378 					prev.br_startblock += mod;
5379 					prev.br_startoff = start;
5380 				}
5381 				prev.br_state = XFS_EXT_UNWRITTEN;
5382 				error = xfs_bmap_add_extent_unwritten_real(tp,
5383 						ip, whichfork, &icur, &cur,
5384 						&prev, firstblock, dfops,
5385 						&logflags);
5386 				if (error)
5387 					goto error0;
5388 				goto nodelete;
5389 			} else {
5390 				ASSERT(del.br_state == XFS_EXT_NORM);
5391 				del.br_state = XFS_EXT_UNWRITTEN;
5392 				error = xfs_bmap_add_extent_unwritten_real(tp,
5393 						ip, whichfork, &icur, &cur,
5394 						&del, firstblock, dfops,
5395 						&logflags);
5396 				if (error)
5397 					goto error0;
5398 				goto nodelete;
5399 			}
5400 		}
5401 
5402 		if (wasdel) {
5403 			error = xfs_bmap_del_extent_delay(ip, whichfork, &icur,
5404 					&got, &del);
5405 		} else {
5406 			error = xfs_bmap_del_extent_real(ip, tp, &icur, dfops,
5407 					cur, &del, &tmp_logflags, whichfork,
5408 					flags);
5409 			logflags |= tmp_logflags;
5410 		}
5411 
5412 		if (error)
5413 			goto error0;
5414 
5415 		max_len -= del.br_blockcount;
5416 		end = del.br_startoff - 1;
5417 nodelete:
5418 		/*
5419 		 * If not done go on to the next (previous) record.
5420 		 */
5421 		if (end != (xfs_fileoff_t)-1 && end >= start) {
5422 			if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5423 			    (got.br_startoff > end &&
5424 			     !xfs_iext_prev_extent(ifp, &icur, &got))) {
5425 				done = true;
5426 				break;
5427 			}
5428 			extno++;
5429 		}
5430 	}
5431 	if (done || end == (xfs_fileoff_t)-1 || end < start)
5432 		*rlen = 0;
5433 	else
5434 		*rlen = end - start + 1;
5435 
5436 	/*
5437 	 * Convert to a btree if necessary.
5438 	 */
5439 	if (xfs_bmap_needs_btree(ip, whichfork)) {
5440 		ASSERT(cur == NULL);
5441 		error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops,
5442 			&cur, 0, &tmp_logflags, whichfork);
5443 		logflags |= tmp_logflags;
5444 		if (error)
5445 			goto error0;
5446 	}
5447 	/*
5448 	 * transform from btree to extents, give it cur
5449 	 */
5450 	else if (xfs_bmap_wants_extents(ip, whichfork)) {
5451 		ASSERT(cur != NULL);
5452 		error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5453 			whichfork);
5454 		logflags |= tmp_logflags;
5455 		if (error)
5456 			goto error0;
5457 	}
5458 	/*
5459 	 * transform from extents to local?
5460 	 */
5461 	error = 0;
5462 error0:
5463 	/*
5464 	 * Log everything.  Do this after conversion, there's no point in
5465 	 * logging the extent records if we've converted to btree format.
5466 	 */
5467 	if ((logflags & xfs_ilog_fext(whichfork)) &&
5468 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5469 		logflags &= ~xfs_ilog_fext(whichfork);
5470 	else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5471 		 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5472 		logflags &= ~xfs_ilog_fbroot(whichfork);
5473 	/*
5474 	 * Log inode even in the error case, if the transaction
5475 	 * is dirty we'll need to shut down the filesystem.
5476 	 */
5477 	if (logflags)
5478 		xfs_trans_log_inode(tp, ip, logflags);
5479 	if (cur) {
5480 		if (!error) {
5481 			*firstblock = cur->bc_private.b.firstblock;
5482 			cur->bc_private.b.allocated = 0;
5483 		}
5484 		xfs_btree_del_cursor(cur,
5485 			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5486 	}
5487 	return error;
5488 }
5489 
5490 /* Unmap a range of a file. */
5491 int
5492 xfs_bunmapi(
5493 	xfs_trans_t		*tp,
5494 	struct xfs_inode	*ip,
5495 	xfs_fileoff_t		bno,
5496 	xfs_filblks_t		len,
5497 	int			flags,
5498 	xfs_extnum_t		nexts,
5499 	xfs_fsblock_t		*firstblock,
5500 	struct xfs_defer_ops	*dfops,
5501 	int			*done)
5502 {
5503 	int			error;
5504 
5505 	error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts, firstblock,
5506 			dfops);
5507 	*done = (len == 0);
5508 	return error;
5509 }
5510 
5511 /*
5512  * Determine whether an extent shift can be accomplished by a merge with the
5513  * extent that precedes the target hole of the shift.
5514  */
5515 STATIC bool
5516 xfs_bmse_can_merge(
5517 	struct xfs_bmbt_irec	*left,	/* preceding extent */
5518 	struct xfs_bmbt_irec	*got,	/* current extent to shift */
5519 	xfs_fileoff_t		shift)	/* shift fsb */
5520 {
5521 	xfs_fileoff_t		startoff;
5522 
5523 	startoff = got->br_startoff - shift;
5524 
5525 	/*
5526 	 * The extent, once shifted, must be adjacent in-file and on-disk with
5527 	 * the preceding extent.
5528 	 */
5529 	if ((left->br_startoff + left->br_blockcount != startoff) ||
5530 	    (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5531 	    (left->br_state != got->br_state) ||
5532 	    (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5533 		return false;
5534 
5535 	return true;
5536 }
5537 
5538 /*
5539  * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5540  * hole in the file. If an extent shift would result in the extent being fully
5541  * adjacent to the extent that currently precedes the hole, we can merge with
5542  * the preceding extent rather than do the shift.
5543  *
5544  * This function assumes the caller has verified a shift-by-merge is possible
5545  * with the provided extents via xfs_bmse_can_merge().
5546  */
5547 STATIC int
5548 xfs_bmse_merge(
5549 	struct xfs_inode		*ip,
5550 	int				whichfork,
5551 	xfs_fileoff_t			shift,		/* shift fsb */
5552 	struct xfs_iext_cursor		*icur,
5553 	struct xfs_bmbt_irec		*got,		/* extent to shift */
5554 	struct xfs_bmbt_irec		*left,		/* preceding extent */
5555 	struct xfs_btree_cur		*cur,
5556 	int				*logflags,	/* output */
5557 	struct xfs_defer_ops		*dfops)
5558 {
5559 	struct xfs_bmbt_irec		new;
5560 	xfs_filblks_t			blockcount;
5561 	int				error, i;
5562 	struct xfs_mount		*mp = ip->i_mount;
5563 
5564 	blockcount = left->br_blockcount + got->br_blockcount;
5565 
5566 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5567 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5568 	ASSERT(xfs_bmse_can_merge(left, got, shift));
5569 
5570 	new = *left;
5571 	new.br_blockcount = blockcount;
5572 
5573 	/*
5574 	 * Update the on-disk extent count, the btree if necessary and log the
5575 	 * inode.
5576 	 */
5577 	XFS_IFORK_NEXT_SET(ip, whichfork,
5578 			   XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5579 	*logflags |= XFS_ILOG_CORE;
5580 	if (!cur) {
5581 		*logflags |= XFS_ILOG_DEXT;
5582 		goto done;
5583 	}
5584 
5585 	/* lookup and remove the extent to merge */
5586 	error = xfs_bmbt_lookup_eq(cur, got, &i);
5587 	if (error)
5588 		return error;
5589 	XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5590 
5591 	error = xfs_btree_delete(cur, &i);
5592 	if (error)
5593 		return error;
5594 	XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5595 
5596 	/* lookup and update size of the previous extent */
5597 	error = xfs_bmbt_lookup_eq(cur, left, &i);
5598 	if (error)
5599 		return error;
5600 	XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5601 
5602 	error = xfs_bmbt_update(cur, &new);
5603 	if (error)
5604 		return error;
5605 
5606 done:
5607 	xfs_iext_remove(ip, icur, 0);
5608 	xfs_iext_prev(XFS_IFORK_PTR(ip, whichfork), icur);
5609 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5610 			&new);
5611 
5612 	/* update reverse mapping. rmap functions merge the rmaps for us */
5613 	error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, got);
5614 	if (error)
5615 		return error;
5616 	memcpy(&new, got, sizeof(new));
5617 	new.br_startoff = left->br_startoff + left->br_blockcount;
5618 	return xfs_rmap_map_extent(mp, dfops, ip, whichfork, &new);
5619 }
5620 
5621 static int
5622 xfs_bmap_shift_update_extent(
5623 	struct xfs_inode	*ip,
5624 	int			whichfork,
5625 	struct xfs_iext_cursor	*icur,
5626 	struct xfs_bmbt_irec	*got,
5627 	struct xfs_btree_cur	*cur,
5628 	int			*logflags,
5629 	struct xfs_defer_ops	*dfops,
5630 	xfs_fileoff_t		startoff)
5631 {
5632 	struct xfs_mount	*mp = ip->i_mount;
5633 	struct xfs_bmbt_irec	prev = *got;
5634 	int			error, i;
5635 
5636 	*logflags |= XFS_ILOG_CORE;
5637 
5638 	got->br_startoff = startoff;
5639 
5640 	if (cur) {
5641 		error = xfs_bmbt_lookup_eq(cur, &prev, &i);
5642 		if (error)
5643 			return error;
5644 		XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5645 
5646 		error = xfs_bmbt_update(cur, got);
5647 		if (error)
5648 			return error;
5649 	} else {
5650 		*logflags |= XFS_ILOG_DEXT;
5651 	}
5652 
5653 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5654 			got);
5655 
5656 	/* update reverse mapping */
5657 	error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, &prev);
5658 	if (error)
5659 		return error;
5660 	return xfs_rmap_map_extent(mp, dfops, ip, whichfork, got);
5661 }
5662 
5663 int
5664 xfs_bmap_collapse_extents(
5665 	struct xfs_trans	*tp,
5666 	struct xfs_inode	*ip,
5667 	xfs_fileoff_t		*next_fsb,
5668 	xfs_fileoff_t		offset_shift_fsb,
5669 	bool			*done,
5670 	xfs_fsblock_t		*firstblock,
5671 	struct xfs_defer_ops	*dfops)
5672 {
5673 	int			whichfork = XFS_DATA_FORK;
5674 	struct xfs_mount	*mp = ip->i_mount;
5675 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
5676 	struct xfs_btree_cur	*cur = NULL;
5677 	struct xfs_bmbt_irec	got, prev;
5678 	struct xfs_iext_cursor	icur;
5679 	xfs_fileoff_t		new_startoff;
5680 	int			error = 0;
5681 	int			logflags = 0;
5682 
5683 	if (unlikely(XFS_TEST_ERROR(
5684 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5685 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5686 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
5687 		XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
5688 		return -EFSCORRUPTED;
5689 	}
5690 
5691 	if (XFS_FORCED_SHUTDOWN(mp))
5692 		return -EIO;
5693 
5694 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL));
5695 
5696 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5697 		error = xfs_iread_extents(tp, ip, whichfork);
5698 		if (error)
5699 			return error;
5700 	}
5701 
5702 	if (ifp->if_flags & XFS_IFBROOT) {
5703 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5704 		cur->bc_private.b.firstblock = *firstblock;
5705 		cur->bc_private.b.dfops = dfops;
5706 		cur->bc_private.b.flags = 0;
5707 	}
5708 
5709 	if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5710 		*done = true;
5711 		goto del_cursor;
5712 	}
5713 	XFS_WANT_CORRUPTED_GOTO(mp, !isnullstartblock(got.br_startblock),
5714 				del_cursor);
5715 
5716 	new_startoff = got.br_startoff - offset_shift_fsb;
5717 	if (xfs_iext_peek_prev_extent(ifp, &icur, &prev)) {
5718 		if (new_startoff < prev.br_startoff + prev.br_blockcount) {
5719 			error = -EINVAL;
5720 			goto del_cursor;
5721 		}
5722 
5723 		if (xfs_bmse_can_merge(&prev, &got, offset_shift_fsb)) {
5724 			error = xfs_bmse_merge(ip, whichfork, offset_shift_fsb,
5725 					&icur, &got, &prev, cur, &logflags,
5726 					dfops);
5727 			if (error)
5728 				goto del_cursor;
5729 			goto done;
5730 		}
5731 	} else {
5732 		if (got.br_startoff < offset_shift_fsb) {
5733 			error = -EINVAL;
5734 			goto del_cursor;
5735 		}
5736 	}
5737 
5738 	error = xfs_bmap_shift_update_extent(ip, whichfork, &icur, &got, cur,
5739 			&logflags, dfops, new_startoff);
5740 	if (error)
5741 		goto del_cursor;
5742 
5743 done:
5744 	if (!xfs_iext_next_extent(ifp, &icur, &got)) {
5745 		*done = true;
5746 		goto del_cursor;
5747 	}
5748 
5749 	*next_fsb = got.br_startoff;
5750 del_cursor:
5751 	if (cur)
5752 		xfs_btree_del_cursor(cur,
5753 			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5754 	if (logflags)
5755 		xfs_trans_log_inode(tp, ip, logflags);
5756 	return error;
5757 }
5758 
5759 int
5760 xfs_bmap_insert_extents(
5761 	struct xfs_trans	*tp,
5762 	struct xfs_inode	*ip,
5763 	xfs_fileoff_t		*next_fsb,
5764 	xfs_fileoff_t		offset_shift_fsb,
5765 	bool			*done,
5766 	xfs_fileoff_t		stop_fsb,
5767 	xfs_fsblock_t		*firstblock,
5768 	struct xfs_defer_ops	*dfops)
5769 {
5770 	int			whichfork = XFS_DATA_FORK;
5771 	struct xfs_mount	*mp = ip->i_mount;
5772 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
5773 	struct xfs_btree_cur	*cur = NULL;
5774 	struct xfs_bmbt_irec	got, next;
5775 	struct xfs_iext_cursor	icur;
5776 	xfs_fileoff_t		new_startoff;
5777 	int			error = 0;
5778 	int			logflags = 0;
5779 
5780 	if (unlikely(XFS_TEST_ERROR(
5781 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5782 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5783 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
5784 		XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
5785 		return -EFSCORRUPTED;
5786 	}
5787 
5788 	if (XFS_FORCED_SHUTDOWN(mp))
5789 		return -EIO;
5790 
5791 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL));
5792 
5793 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5794 		error = xfs_iread_extents(tp, ip, whichfork);
5795 		if (error)
5796 			return error;
5797 	}
5798 
5799 	if (ifp->if_flags & XFS_IFBROOT) {
5800 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5801 		cur->bc_private.b.firstblock = *firstblock;
5802 		cur->bc_private.b.dfops = dfops;
5803 		cur->bc_private.b.flags = 0;
5804 	}
5805 
5806 	if (*next_fsb == NULLFSBLOCK) {
5807 		xfs_iext_last(ifp, &icur);
5808 		if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5809 		    stop_fsb > got.br_startoff) {
5810 			*done = true;
5811 			goto del_cursor;
5812 		}
5813 	} else {
5814 		if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5815 			*done = true;
5816 			goto del_cursor;
5817 		}
5818 	}
5819 	XFS_WANT_CORRUPTED_GOTO(mp, !isnullstartblock(got.br_startblock),
5820 				del_cursor);
5821 
5822 	if (stop_fsb >= got.br_startoff + got.br_blockcount) {
5823 		error = -EIO;
5824 		goto del_cursor;
5825 	}
5826 
5827 	new_startoff = got.br_startoff + offset_shift_fsb;
5828 	if (xfs_iext_peek_next_extent(ifp, &icur, &next)) {
5829 		if (new_startoff + got.br_blockcount > next.br_startoff) {
5830 			error = -EINVAL;
5831 			goto del_cursor;
5832 		}
5833 
5834 		/*
5835 		 * Unlike a left shift (which involves a hole punch), a right
5836 		 * shift does not modify extent neighbors in any way.  We should
5837 		 * never find mergeable extents in this scenario.  Check anyways
5838 		 * and warn if we encounter two extents that could be one.
5839 		 */
5840 		if (xfs_bmse_can_merge(&got, &next, offset_shift_fsb))
5841 			WARN_ON_ONCE(1);
5842 	}
5843 
5844 	error = xfs_bmap_shift_update_extent(ip, whichfork, &icur, &got, cur,
5845 			&logflags, dfops, new_startoff);
5846 	if (error)
5847 		goto del_cursor;
5848 
5849 	if (!xfs_iext_prev_extent(ifp, &icur, &got) ||
5850 	    stop_fsb >= got.br_startoff + got.br_blockcount) {
5851 		*done = true;
5852 		goto del_cursor;
5853 	}
5854 
5855 	*next_fsb = got.br_startoff;
5856 del_cursor:
5857 	if (cur)
5858 		xfs_btree_del_cursor(cur,
5859 			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5860 	if (logflags)
5861 		xfs_trans_log_inode(tp, ip, logflags);
5862 	return error;
5863 }
5864 
5865 /*
5866  * Splits an extent into two extents at split_fsb block such that it is the
5867  * first block of the current_ext. @ext is a target extent to be split.
5868  * @split_fsb is a block where the extents is split.  If split_fsb lies in a
5869  * hole or the first block of extents, just return 0.
5870  */
5871 STATIC int
5872 xfs_bmap_split_extent_at(
5873 	struct xfs_trans	*tp,
5874 	struct xfs_inode	*ip,
5875 	xfs_fileoff_t		split_fsb,
5876 	xfs_fsblock_t		*firstfsb,
5877 	struct xfs_defer_ops	*dfops)
5878 {
5879 	int				whichfork = XFS_DATA_FORK;
5880 	struct xfs_btree_cur		*cur = NULL;
5881 	struct xfs_bmbt_irec		got;
5882 	struct xfs_bmbt_irec		new; /* split extent */
5883 	struct xfs_mount		*mp = ip->i_mount;
5884 	struct xfs_ifork		*ifp;
5885 	xfs_fsblock_t			gotblkcnt; /* new block count for got */
5886 	struct xfs_iext_cursor		icur;
5887 	int				error = 0;
5888 	int				logflags = 0;
5889 	int				i = 0;
5890 
5891 	if (unlikely(XFS_TEST_ERROR(
5892 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5893 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5894 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
5895 		XFS_ERROR_REPORT("xfs_bmap_split_extent_at",
5896 				 XFS_ERRLEVEL_LOW, mp);
5897 		return -EFSCORRUPTED;
5898 	}
5899 
5900 	if (XFS_FORCED_SHUTDOWN(mp))
5901 		return -EIO;
5902 
5903 	ifp = XFS_IFORK_PTR(ip, whichfork);
5904 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5905 		/* Read in all the extents */
5906 		error = xfs_iread_extents(tp, ip, whichfork);
5907 		if (error)
5908 			return error;
5909 	}
5910 
5911 	/*
5912 	 * If there are not extents, or split_fsb lies in a hole we are done.
5913 	 */
5914 	if (!xfs_iext_lookup_extent(ip, ifp, split_fsb, &icur, &got) ||
5915 	    got.br_startoff >= split_fsb)
5916 		return 0;
5917 
5918 	gotblkcnt = split_fsb - got.br_startoff;
5919 	new.br_startoff = split_fsb;
5920 	new.br_startblock = got.br_startblock + gotblkcnt;
5921 	new.br_blockcount = got.br_blockcount - gotblkcnt;
5922 	new.br_state = got.br_state;
5923 
5924 	if (ifp->if_flags & XFS_IFBROOT) {
5925 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5926 		cur->bc_private.b.firstblock = *firstfsb;
5927 		cur->bc_private.b.dfops = dfops;
5928 		cur->bc_private.b.flags = 0;
5929 		error = xfs_bmbt_lookup_eq(cur, &got, &i);
5930 		if (error)
5931 			goto del_cursor;
5932 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
5933 	}
5934 
5935 	got.br_blockcount = gotblkcnt;
5936 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), &icur,
5937 			&got);
5938 
5939 	logflags = XFS_ILOG_CORE;
5940 	if (cur) {
5941 		error = xfs_bmbt_update(cur, &got);
5942 		if (error)
5943 			goto del_cursor;
5944 	} else
5945 		logflags |= XFS_ILOG_DEXT;
5946 
5947 	/* Add new extent */
5948 	xfs_iext_next(ifp, &icur);
5949 	xfs_iext_insert(ip, &icur, &new, 0);
5950 	XFS_IFORK_NEXT_SET(ip, whichfork,
5951 			   XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5952 
5953 	if (cur) {
5954 		error = xfs_bmbt_lookup_eq(cur, &new, &i);
5955 		if (error)
5956 			goto del_cursor;
5957 		XFS_WANT_CORRUPTED_GOTO(mp, i == 0, del_cursor);
5958 		error = xfs_btree_insert(cur, &i);
5959 		if (error)
5960 			goto del_cursor;
5961 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
5962 	}
5963 
5964 	/*
5965 	 * Convert to a btree if necessary.
5966 	 */
5967 	if (xfs_bmap_needs_btree(ip, whichfork)) {
5968 		int tmp_logflags; /* partial log flag return val */
5969 
5970 		ASSERT(cur == NULL);
5971 		error = xfs_bmap_extents_to_btree(tp, ip, firstfsb, dfops,
5972 				&cur, 0, &tmp_logflags, whichfork);
5973 		logflags |= tmp_logflags;
5974 	}
5975 
5976 del_cursor:
5977 	if (cur) {
5978 		cur->bc_private.b.allocated = 0;
5979 		xfs_btree_del_cursor(cur,
5980 				error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5981 	}
5982 
5983 	if (logflags)
5984 		xfs_trans_log_inode(tp, ip, logflags);
5985 	return error;
5986 }
5987 
5988 int
5989 xfs_bmap_split_extent(
5990 	struct xfs_inode        *ip,
5991 	xfs_fileoff_t           split_fsb)
5992 {
5993 	struct xfs_mount        *mp = ip->i_mount;
5994 	struct xfs_trans        *tp;
5995 	struct xfs_defer_ops    dfops;
5996 	xfs_fsblock_t           firstfsb;
5997 	int                     error;
5998 
5999 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
6000 			XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
6001 	if (error)
6002 		return error;
6003 
6004 	xfs_ilock(ip, XFS_ILOCK_EXCL);
6005 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
6006 
6007 	xfs_defer_init(&dfops, &firstfsb);
6008 
6009 	error = xfs_bmap_split_extent_at(tp, ip, split_fsb,
6010 			&firstfsb, &dfops);
6011 	if (error)
6012 		goto out;
6013 
6014 	error = xfs_defer_finish(&tp, &dfops);
6015 	if (error)
6016 		goto out;
6017 
6018 	return xfs_trans_commit(tp);
6019 
6020 out:
6021 	xfs_defer_cancel(&dfops);
6022 	xfs_trans_cancel(tp);
6023 	return error;
6024 }
6025 
6026 /* Deferred mapping is only for real extents in the data fork. */
6027 static bool
6028 xfs_bmap_is_update_needed(
6029 	struct xfs_bmbt_irec	*bmap)
6030 {
6031 	return  bmap->br_startblock != HOLESTARTBLOCK &&
6032 		bmap->br_startblock != DELAYSTARTBLOCK;
6033 }
6034 
6035 /* Record a bmap intent. */
6036 static int
6037 __xfs_bmap_add(
6038 	struct xfs_mount		*mp,
6039 	struct xfs_defer_ops		*dfops,
6040 	enum xfs_bmap_intent_type	type,
6041 	struct xfs_inode		*ip,
6042 	int				whichfork,
6043 	struct xfs_bmbt_irec		*bmap)
6044 {
6045 	int				error;
6046 	struct xfs_bmap_intent		*bi;
6047 
6048 	trace_xfs_bmap_defer(mp,
6049 			XFS_FSB_TO_AGNO(mp, bmap->br_startblock),
6050 			type,
6051 			XFS_FSB_TO_AGBNO(mp, bmap->br_startblock),
6052 			ip->i_ino, whichfork,
6053 			bmap->br_startoff,
6054 			bmap->br_blockcount,
6055 			bmap->br_state);
6056 
6057 	bi = kmem_alloc(sizeof(struct xfs_bmap_intent), KM_SLEEP | KM_NOFS);
6058 	INIT_LIST_HEAD(&bi->bi_list);
6059 	bi->bi_type = type;
6060 	bi->bi_owner = ip;
6061 	bi->bi_whichfork = whichfork;
6062 	bi->bi_bmap = *bmap;
6063 
6064 	error = xfs_defer_ijoin(dfops, bi->bi_owner);
6065 	if (error) {
6066 		kmem_free(bi);
6067 		return error;
6068 	}
6069 
6070 	xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list);
6071 	return 0;
6072 }
6073 
6074 /* Map an extent into a file. */
6075 int
6076 xfs_bmap_map_extent(
6077 	struct xfs_mount	*mp,
6078 	struct xfs_defer_ops	*dfops,
6079 	struct xfs_inode	*ip,
6080 	struct xfs_bmbt_irec	*PREV)
6081 {
6082 	if (!xfs_bmap_is_update_needed(PREV))
6083 		return 0;
6084 
6085 	return __xfs_bmap_add(mp, dfops, XFS_BMAP_MAP, ip,
6086 			XFS_DATA_FORK, PREV);
6087 }
6088 
6089 /* Unmap an extent out of a file. */
6090 int
6091 xfs_bmap_unmap_extent(
6092 	struct xfs_mount	*mp,
6093 	struct xfs_defer_ops	*dfops,
6094 	struct xfs_inode	*ip,
6095 	struct xfs_bmbt_irec	*PREV)
6096 {
6097 	if (!xfs_bmap_is_update_needed(PREV))
6098 		return 0;
6099 
6100 	return __xfs_bmap_add(mp, dfops, XFS_BMAP_UNMAP, ip,
6101 			XFS_DATA_FORK, PREV);
6102 }
6103 
6104 /*
6105  * Process one of the deferred bmap operations.  We pass back the
6106  * btree cursor to maintain our lock on the bmapbt between calls.
6107  */
6108 int
6109 xfs_bmap_finish_one(
6110 	struct xfs_trans		*tp,
6111 	struct xfs_defer_ops		*dfops,
6112 	struct xfs_inode		*ip,
6113 	enum xfs_bmap_intent_type	type,
6114 	int				whichfork,
6115 	xfs_fileoff_t			startoff,
6116 	xfs_fsblock_t			startblock,
6117 	xfs_filblks_t			*blockcount,
6118 	xfs_exntst_t			state)
6119 {
6120 	xfs_fsblock_t			firstfsb;
6121 	int				error = 0;
6122 
6123 	/*
6124 	 * firstfsb is tied to the transaction lifetime and is used to
6125 	 * ensure correct AG locking order and schedule work item
6126 	 * continuations.  XFS_BUI_MAX_FAST_EXTENTS (== 1) restricts us
6127 	 * to only making one bmap call per transaction, so it should
6128 	 * be safe to have it as a local variable here.
6129 	 */
6130 	firstfsb = NULLFSBLOCK;
6131 
6132 	trace_xfs_bmap_deferred(tp->t_mountp,
6133 			XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type,
6134 			XFS_FSB_TO_AGBNO(tp->t_mountp, startblock),
6135 			ip->i_ino, whichfork, startoff, *blockcount, state);
6136 
6137 	if (WARN_ON_ONCE(whichfork != XFS_DATA_FORK))
6138 		return -EFSCORRUPTED;
6139 
6140 	if (XFS_TEST_ERROR(false, tp->t_mountp,
6141 			XFS_ERRTAG_BMAP_FINISH_ONE))
6142 		return -EIO;
6143 
6144 	switch (type) {
6145 	case XFS_BMAP_MAP:
6146 		error = xfs_bmapi_remap(tp, ip, startoff, *blockcount,
6147 				startblock, dfops);
6148 		*blockcount = 0;
6149 		break;
6150 	case XFS_BMAP_UNMAP:
6151 		error = __xfs_bunmapi(tp, ip, startoff, blockcount,
6152 				XFS_BMAPI_REMAP, 1, &firstfsb, dfops);
6153 		break;
6154 	default:
6155 		ASSERT(0);
6156 		error = -EFSCORRUPTED;
6157 	}
6158 
6159 	return error;
6160 }
6161 
6162 /* Check that an inode's extent does not have invalid flags or bad ranges. */
6163 xfs_failaddr_t
6164 xfs_bmap_validate_extent(
6165 	struct xfs_inode	*ip,
6166 	int			whichfork,
6167 	struct xfs_bmbt_irec	*irec)
6168 {
6169 	struct xfs_mount	*mp = ip->i_mount;
6170 	xfs_fsblock_t		endfsb;
6171 	bool			isrt;
6172 
6173 	isrt = XFS_IS_REALTIME_INODE(ip);
6174 	endfsb = irec->br_startblock + irec->br_blockcount - 1;
6175 	if (isrt) {
6176 		if (!xfs_verify_rtbno(mp, irec->br_startblock))
6177 			return __this_address;
6178 		if (!xfs_verify_rtbno(mp, endfsb))
6179 			return __this_address;
6180 	} else {
6181 		if (!xfs_verify_fsbno(mp, irec->br_startblock))
6182 			return __this_address;
6183 		if (!xfs_verify_fsbno(mp, endfsb))
6184 			return __this_address;
6185 		if (XFS_FSB_TO_AGNO(mp, irec->br_startblock) !=
6186 		    XFS_FSB_TO_AGNO(mp, endfsb))
6187 			return __this_address;
6188 	}
6189 	if (irec->br_state != XFS_EXT_NORM) {
6190 		if (whichfork != XFS_DATA_FORK)
6191 			return __this_address;
6192 		if (!xfs_sb_version_hasextflgbit(&mp->m_sb))
6193 			return __this_address;
6194 	}
6195 	return NULL;
6196 }
6197