xref: /openbmc/linux/fs/xfs/libxfs/xfs_bmap.c (revision 6396bb221514d2876fd6dc0aa2a1f240d99b37bb)
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	*lip;
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(lip, &cur->bc_tp->t_items, li_trans) {
264 		struct xfs_buf_log_item	*bip = (struct xfs_buf_log_item *)lip;
265 
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 				xfs_err(mp, "%s: ptrs are equal in node\n",
316 					__func__);
317 				xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
318 			}
319 		}
320 	}
321 }
322 
323 /*
324  * Check that the extents for the inode ip are in the right order in all
325  * btree leaves. THis becomes prohibitively expensive for large extent count
326  * files, so don't bother with inodes that have more than 10,000 extents in
327  * them. The btree record ordering checks will still be done, so for such large
328  * bmapbt constructs that is going to catch most corruptions.
329  */
330 STATIC void
331 xfs_bmap_check_leaf_extents(
332 	xfs_btree_cur_t		*cur,	/* btree cursor or null */
333 	xfs_inode_t		*ip,		/* incore inode pointer */
334 	int			whichfork)	/* data or attr fork */
335 {
336 	struct xfs_btree_block	*block;	/* current btree block */
337 	xfs_fsblock_t		bno;	/* block # of "block" */
338 	xfs_buf_t		*bp;	/* buffer for "block" */
339 	int			error;	/* error return value */
340 	xfs_extnum_t		i=0, j;	/* index into the extents list */
341 	xfs_ifork_t		*ifp;	/* fork structure */
342 	int			level;	/* btree level, for checking */
343 	xfs_mount_t		*mp;	/* file system mount structure */
344 	__be64			*pp;	/* pointer to block address */
345 	xfs_bmbt_rec_t		*ep;	/* pointer to current extent */
346 	xfs_bmbt_rec_t		last = {0, 0}; /* last extent in prev block */
347 	xfs_bmbt_rec_t		*nextp;	/* pointer to next extent */
348 	int			bp_release = 0;
349 
350 	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
351 		return;
352 	}
353 
354 	/* skip large extent count inodes */
355 	if (ip->i_d.di_nextents > 10000)
356 		return;
357 
358 	bno = NULLFSBLOCK;
359 	mp = ip->i_mount;
360 	ifp = XFS_IFORK_PTR(ip, whichfork);
361 	block = ifp->if_broot;
362 	/*
363 	 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
364 	 */
365 	level = be16_to_cpu(block->bb_level);
366 	ASSERT(level > 0);
367 	xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
368 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
369 	bno = be64_to_cpu(*pp);
370 
371 	ASSERT(bno != NULLFSBLOCK);
372 	ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
373 	ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
374 
375 	/*
376 	 * Go down the tree until leaf level is reached, following the first
377 	 * pointer (leftmost) at each level.
378 	 */
379 	while (level-- > 0) {
380 		/* See if buf is in cur first */
381 		bp_release = 0;
382 		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
383 		if (!bp) {
384 			bp_release = 1;
385 			error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
386 						XFS_BMAP_BTREE_REF,
387 						&xfs_bmbt_buf_ops);
388 			if (error)
389 				goto error_norelse;
390 		}
391 		block = XFS_BUF_TO_BLOCK(bp);
392 		if (level == 0)
393 			break;
394 
395 		/*
396 		 * Check this block for basic sanity (increasing keys and
397 		 * no duplicate blocks).
398 		 */
399 
400 		xfs_check_block(block, mp, 0, 0);
401 		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
402 		bno = be64_to_cpu(*pp);
403 		XFS_WANT_CORRUPTED_GOTO(mp,
404 					xfs_verify_fsbno(mp, bno), error0);
405 		if (bp_release) {
406 			bp_release = 0;
407 			xfs_trans_brelse(NULL, bp);
408 		}
409 	}
410 
411 	/*
412 	 * Here with bp and block set to the leftmost leaf node in the tree.
413 	 */
414 	i = 0;
415 
416 	/*
417 	 * Loop over all leaf nodes checking that all extents are in the right order.
418 	 */
419 	for (;;) {
420 		xfs_fsblock_t	nextbno;
421 		xfs_extnum_t	num_recs;
422 
423 
424 		num_recs = xfs_btree_get_numrecs(block);
425 
426 		/*
427 		 * Read-ahead the next leaf block, if any.
428 		 */
429 
430 		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
431 
432 		/*
433 		 * Check all the extents to make sure they are OK.
434 		 * If we had a previous block, the last entry should
435 		 * conform with the first entry in this one.
436 		 */
437 
438 		ep = XFS_BMBT_REC_ADDR(mp, block, 1);
439 		if (i) {
440 			ASSERT(xfs_bmbt_disk_get_startoff(&last) +
441 			       xfs_bmbt_disk_get_blockcount(&last) <=
442 			       xfs_bmbt_disk_get_startoff(ep));
443 		}
444 		for (j = 1; j < num_recs; j++) {
445 			nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
446 			ASSERT(xfs_bmbt_disk_get_startoff(ep) +
447 			       xfs_bmbt_disk_get_blockcount(ep) <=
448 			       xfs_bmbt_disk_get_startoff(nextp));
449 			ep = nextp;
450 		}
451 
452 		last = *ep;
453 		i += num_recs;
454 		if (bp_release) {
455 			bp_release = 0;
456 			xfs_trans_brelse(NULL, bp);
457 		}
458 		bno = nextbno;
459 		/*
460 		 * If we've reached the end, stop.
461 		 */
462 		if (bno == NULLFSBLOCK)
463 			break;
464 
465 		bp_release = 0;
466 		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
467 		if (!bp) {
468 			bp_release = 1;
469 			error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
470 						XFS_BMAP_BTREE_REF,
471 						&xfs_bmbt_buf_ops);
472 			if (error)
473 				goto error_norelse;
474 		}
475 		block = XFS_BUF_TO_BLOCK(bp);
476 	}
477 
478 	return;
479 
480 error0:
481 	xfs_warn(mp, "%s: at error0", __func__);
482 	if (bp_release)
483 		xfs_trans_brelse(NULL, bp);
484 error_norelse:
485 	xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
486 		__func__, i);
487 	xfs_err(mp, "%s: CORRUPTED BTREE OR SOMETHING", __func__);
488 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
489 	return;
490 }
491 
492 /*
493  * Validate that the bmbt_irecs being returned from bmapi are valid
494  * given the caller's original parameters.  Specifically check the
495  * ranges of the returned irecs to ensure that they only extend beyond
496  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
497  */
498 STATIC void
499 xfs_bmap_validate_ret(
500 	xfs_fileoff_t		bno,
501 	xfs_filblks_t		len,
502 	int			flags,
503 	xfs_bmbt_irec_t		*mval,
504 	int			nmap,
505 	int			ret_nmap)
506 {
507 	int			i;		/* index to map values */
508 
509 	ASSERT(ret_nmap <= nmap);
510 
511 	for (i = 0; i < ret_nmap; i++) {
512 		ASSERT(mval[i].br_blockcount > 0);
513 		if (!(flags & XFS_BMAPI_ENTIRE)) {
514 			ASSERT(mval[i].br_startoff >= bno);
515 			ASSERT(mval[i].br_blockcount <= len);
516 			ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
517 			       bno + len);
518 		} else {
519 			ASSERT(mval[i].br_startoff < bno + len);
520 			ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
521 			       bno);
522 		}
523 		ASSERT(i == 0 ||
524 		       mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
525 		       mval[i].br_startoff);
526 		ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
527 		       mval[i].br_startblock != HOLESTARTBLOCK);
528 		ASSERT(mval[i].br_state == XFS_EXT_NORM ||
529 		       mval[i].br_state == XFS_EXT_UNWRITTEN);
530 	}
531 }
532 
533 #else
534 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)		do { } while (0)
535 #define	xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)	do { } while (0)
536 #endif /* DEBUG */
537 
538 /*
539  * bmap free list manipulation functions
540  */
541 
542 /*
543  * Add the extent to the list of extents to be free at transaction end.
544  * The list is maintained sorted (by block number).
545  */
546 void
547 __xfs_bmap_add_free(
548 	struct xfs_mount		*mp,
549 	struct xfs_defer_ops		*dfops,
550 	xfs_fsblock_t			bno,
551 	xfs_filblks_t			len,
552 	struct xfs_owner_info		*oinfo,
553 	bool				skip_discard)
554 {
555 	struct xfs_extent_free_item	*new;		/* new element */
556 #ifdef DEBUG
557 	xfs_agnumber_t		agno;
558 	xfs_agblock_t		agbno;
559 
560 	ASSERT(bno != NULLFSBLOCK);
561 	ASSERT(len > 0);
562 	ASSERT(len <= MAXEXTLEN);
563 	ASSERT(!isnullstartblock(bno));
564 	agno = XFS_FSB_TO_AGNO(mp, bno);
565 	agbno = XFS_FSB_TO_AGBNO(mp, bno);
566 	ASSERT(agno < mp->m_sb.sb_agcount);
567 	ASSERT(agbno < mp->m_sb.sb_agblocks);
568 	ASSERT(len < mp->m_sb.sb_agblocks);
569 	ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
570 #endif
571 	ASSERT(xfs_bmap_free_item_zone != NULL);
572 
573 	new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
574 	new->xefi_startblock = bno;
575 	new->xefi_blockcount = (xfs_extlen_t)len;
576 	if (oinfo)
577 		new->xefi_oinfo = *oinfo;
578 	else
579 		xfs_rmap_skip_owner_update(&new->xefi_oinfo);
580 	new->xefi_skip_discard = skip_discard;
581 	trace_xfs_bmap_free_defer(mp, XFS_FSB_TO_AGNO(mp, bno), 0,
582 			XFS_FSB_TO_AGBNO(mp, bno), len);
583 	xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list);
584 }
585 
586 /*
587  * Inode fork format manipulation functions
588  */
589 
590 /*
591  * Transform a btree format file with only one leaf node, where the
592  * extents list will fit in the inode, into an extents format file.
593  * Since the file extents are already in-core, all we have to do is
594  * give up the space for the btree root and pitch the leaf block.
595  */
596 STATIC int				/* error */
597 xfs_bmap_btree_to_extents(
598 	xfs_trans_t		*tp,	/* transaction pointer */
599 	xfs_inode_t		*ip,	/* incore inode pointer */
600 	xfs_btree_cur_t		*cur,	/* btree cursor */
601 	int			*logflagsp, /* inode logging flags */
602 	int			whichfork)  /* data or attr fork */
603 {
604 	/* REFERENCED */
605 	struct xfs_btree_block	*cblock;/* child btree block */
606 	xfs_fsblock_t		cbno;	/* child block number */
607 	xfs_buf_t		*cbp;	/* child block's buffer */
608 	int			error;	/* error return value */
609 	xfs_ifork_t		*ifp;	/* inode fork data */
610 	xfs_mount_t		*mp;	/* mount point structure */
611 	__be64			*pp;	/* ptr to block address */
612 	struct xfs_btree_block	*rblock;/* root btree block */
613 	struct xfs_owner_info	oinfo;
614 
615 	mp = ip->i_mount;
616 	ifp = XFS_IFORK_PTR(ip, whichfork);
617 	ASSERT(whichfork != XFS_COW_FORK);
618 	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
619 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
620 	rblock = ifp->if_broot;
621 	ASSERT(be16_to_cpu(rblock->bb_level) == 1);
622 	ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
623 	ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
624 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
625 	cbno = be64_to_cpu(*pp);
626 	*logflagsp = 0;
627 #ifdef DEBUG
628 	XFS_WANT_CORRUPTED_RETURN(cur->bc_mp,
629 			xfs_btree_check_lptr(cur, cbno, 1));
630 #endif
631 	error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
632 				&xfs_bmbt_buf_ops);
633 	if (error)
634 		return error;
635 	cblock = XFS_BUF_TO_BLOCK(cbp);
636 	if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
637 		return error;
638 	xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
639 	xfs_bmap_add_free(mp, cur->bc_private.b.dfops, cbno, 1, &oinfo);
640 	ip->i_d.di_nblocks--;
641 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
642 	xfs_trans_binval(tp, cbp);
643 	if (cur->bc_bufs[0] == cbp)
644 		cur->bc_bufs[0] = NULL;
645 	xfs_iroot_realloc(ip, -1, whichfork);
646 	ASSERT(ifp->if_broot == NULL);
647 	ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
648 	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
649 	*logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
650 	return 0;
651 }
652 
653 /*
654  * Convert an extents-format file into a btree-format file.
655  * The new file will have a root block (in the inode) and a single child block.
656  */
657 STATIC int					/* error */
658 xfs_bmap_extents_to_btree(
659 	xfs_trans_t		*tp,		/* transaction pointer */
660 	xfs_inode_t		*ip,		/* incore inode pointer */
661 	xfs_fsblock_t		*firstblock,	/* first-block-allocated */
662 	struct xfs_defer_ops	*dfops,		/* blocks freed in xaction */
663 	xfs_btree_cur_t		**curp,		/* cursor returned to caller */
664 	int			wasdel,		/* converting a delayed alloc */
665 	int			*logflagsp,	/* inode logging flags */
666 	int			whichfork)	/* data or attr fork */
667 {
668 	struct xfs_btree_block	*ablock;	/* allocated (child) bt block */
669 	xfs_buf_t		*abp;		/* buffer for ablock */
670 	xfs_alloc_arg_t		args;		/* allocation arguments */
671 	xfs_bmbt_rec_t		*arp;		/* child record pointer */
672 	struct xfs_btree_block	*block;		/* btree root block */
673 	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
674 	int			error;		/* error return value */
675 	xfs_ifork_t		*ifp;		/* inode fork pointer */
676 	xfs_bmbt_key_t		*kp;		/* root block key pointer */
677 	xfs_mount_t		*mp;		/* mount structure */
678 	xfs_bmbt_ptr_t		*pp;		/* root block address pointer */
679 	struct xfs_iext_cursor	icur;
680 	struct xfs_bmbt_irec	rec;
681 	xfs_extnum_t		cnt = 0;
682 
683 	mp = ip->i_mount;
684 	ASSERT(whichfork != XFS_COW_FORK);
685 	ifp = XFS_IFORK_PTR(ip, whichfork);
686 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
687 
688 	/*
689 	 * Make space in the inode incore.
690 	 */
691 	xfs_iroot_realloc(ip, 1, whichfork);
692 	ifp->if_flags |= XFS_IFBROOT;
693 
694 	/*
695 	 * Fill in the root.
696 	 */
697 	block = ifp->if_broot;
698 	xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
699 				 XFS_BTNUM_BMAP, 1, 1, ip->i_ino,
700 				 XFS_BTREE_LONG_PTRS);
701 	/*
702 	 * Need a cursor.  Can't allocate until bb_level is filled in.
703 	 */
704 	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
705 	cur->bc_private.b.firstblock = *firstblock;
706 	cur->bc_private.b.dfops = dfops;
707 	cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
708 	/*
709 	 * Convert to a btree with two levels, one record in root.
710 	 */
711 	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
712 	memset(&args, 0, sizeof(args));
713 	args.tp = tp;
714 	args.mp = mp;
715 	xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork);
716 	args.firstblock = *firstblock;
717 	if (*firstblock == NULLFSBLOCK) {
718 		args.type = XFS_ALLOCTYPE_START_BNO;
719 		args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
720 	} else if (dfops->dop_low) {
721 		args.type = XFS_ALLOCTYPE_START_BNO;
722 		args.fsbno = *firstblock;
723 	} else {
724 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
725 		args.fsbno = *firstblock;
726 	}
727 	args.minlen = args.maxlen = args.prod = 1;
728 	args.wasdel = wasdel;
729 	*logflagsp = 0;
730 	if ((error = xfs_alloc_vextent(&args))) {
731 		xfs_iroot_realloc(ip, -1, whichfork);
732 		ASSERT(ifp->if_broot == NULL);
733 		XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
734 		xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
735 		return error;
736 	}
737 
738 	if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
739 		xfs_iroot_realloc(ip, -1, whichfork);
740 		ASSERT(ifp->if_broot == NULL);
741 		XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
742 		xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
743 		return -ENOSPC;
744 	}
745 	/*
746 	 * Allocation can't fail, the space was reserved.
747 	 */
748 	ASSERT(*firstblock == NULLFSBLOCK ||
749 	       args.agno >= XFS_FSB_TO_AGNO(mp, *firstblock));
750 	*firstblock = cur->bc_private.b.firstblock = args.fsbno;
751 	cur->bc_private.b.allocated++;
752 	ip->i_d.di_nblocks++;
753 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
754 	abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
755 	/*
756 	 * Fill in the child block.
757 	 */
758 	abp->b_ops = &xfs_bmbt_buf_ops;
759 	ablock = XFS_BUF_TO_BLOCK(abp);
760 	xfs_btree_init_block_int(mp, ablock, abp->b_bn,
761 				XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
762 				XFS_BTREE_LONG_PTRS);
763 
764 	for_each_xfs_iext(ifp, &icur, &rec) {
765 		if (isnullstartblock(rec.br_startblock))
766 			continue;
767 		arp = XFS_BMBT_REC_ADDR(mp, ablock, 1 + cnt);
768 		xfs_bmbt_disk_set_all(arp, &rec);
769 		cnt++;
770 	}
771 	ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
772 	xfs_btree_set_numrecs(ablock, cnt);
773 
774 	/*
775 	 * Fill in the root key and pointer.
776 	 */
777 	kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
778 	arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
779 	kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
780 	pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
781 						be16_to_cpu(block->bb_level)));
782 	*pp = cpu_to_be64(args.fsbno);
783 
784 	/*
785 	 * Do all this logging at the end so that
786 	 * the root is at the right level.
787 	 */
788 	xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
789 	xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
790 	ASSERT(*curp == NULL);
791 	*curp = cur;
792 	*logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
793 	return 0;
794 }
795 
796 /*
797  * Convert a local file to an extents file.
798  * This code is out of bounds for data forks of regular files,
799  * since the file data needs to get logged so things will stay consistent.
800  * (The bmap-level manipulations are ok, though).
801  */
802 void
803 xfs_bmap_local_to_extents_empty(
804 	struct xfs_inode	*ip,
805 	int			whichfork)
806 {
807 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
808 
809 	ASSERT(whichfork != XFS_COW_FORK);
810 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
811 	ASSERT(ifp->if_bytes == 0);
812 	ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
813 
814 	xfs_bmap_forkoff_reset(ip, whichfork);
815 	ifp->if_flags &= ~XFS_IFINLINE;
816 	ifp->if_flags |= XFS_IFEXTENTS;
817 	ifp->if_u1.if_root = NULL;
818 	ifp->if_height = 0;
819 	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
820 }
821 
822 
823 STATIC int				/* error */
824 xfs_bmap_local_to_extents(
825 	xfs_trans_t	*tp,		/* transaction pointer */
826 	xfs_inode_t	*ip,		/* incore inode pointer */
827 	xfs_fsblock_t	*firstblock,	/* first block allocated in xaction */
828 	xfs_extlen_t	total,		/* total blocks needed by transaction */
829 	int		*logflagsp,	/* inode logging flags */
830 	int		whichfork,
831 	void		(*init_fn)(struct xfs_trans *tp,
832 				   struct xfs_buf *bp,
833 				   struct xfs_inode *ip,
834 				   struct xfs_ifork *ifp))
835 {
836 	int		error = 0;
837 	int		flags;		/* logging flags returned */
838 	xfs_ifork_t	*ifp;		/* inode fork pointer */
839 	xfs_alloc_arg_t	args;		/* allocation arguments */
840 	xfs_buf_t	*bp;		/* buffer for extent block */
841 	struct xfs_bmbt_irec rec;
842 	struct xfs_iext_cursor icur;
843 
844 	/*
845 	 * We don't want to deal with the case of keeping inode data inline yet.
846 	 * So sending the data fork of a regular inode is invalid.
847 	 */
848 	ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
849 	ifp = XFS_IFORK_PTR(ip, whichfork);
850 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
851 
852 	if (!ifp->if_bytes) {
853 		xfs_bmap_local_to_extents_empty(ip, whichfork);
854 		flags = XFS_ILOG_CORE;
855 		goto done;
856 	}
857 
858 	flags = 0;
859 	error = 0;
860 	ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS)) == XFS_IFINLINE);
861 	memset(&args, 0, sizeof(args));
862 	args.tp = tp;
863 	args.mp = ip->i_mount;
864 	xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0);
865 	args.firstblock = *firstblock;
866 	/*
867 	 * Allocate a block.  We know we need only one, since the
868 	 * file currently fits in an inode.
869 	 */
870 	if (*firstblock == NULLFSBLOCK) {
871 		args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
872 		args.type = XFS_ALLOCTYPE_START_BNO;
873 	} else {
874 		args.fsbno = *firstblock;
875 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
876 	}
877 	args.total = total;
878 	args.minlen = args.maxlen = args.prod = 1;
879 	error = xfs_alloc_vextent(&args);
880 	if (error)
881 		goto done;
882 
883 	/* Can't fail, the space was reserved. */
884 	ASSERT(args.fsbno != NULLFSBLOCK);
885 	ASSERT(args.len == 1);
886 	*firstblock = args.fsbno;
887 	bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
888 
889 	/*
890 	 * Initialize the block, copy the data and log the remote buffer.
891 	 *
892 	 * The callout is responsible for logging because the remote format
893 	 * might differ from the local format and thus we don't know how much to
894 	 * log here. Note that init_fn must also set the buffer log item type
895 	 * correctly.
896 	 */
897 	init_fn(tp, bp, ip, ifp);
898 
899 	/* account for the change in fork size */
900 	xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
901 	xfs_bmap_local_to_extents_empty(ip, whichfork);
902 	flags |= XFS_ILOG_CORE;
903 
904 	ifp->if_u1.if_root = NULL;
905 	ifp->if_height = 0;
906 
907 	rec.br_startoff = 0;
908 	rec.br_startblock = args.fsbno;
909 	rec.br_blockcount = 1;
910 	rec.br_state = XFS_EXT_NORM;
911 	xfs_iext_first(ifp, &icur);
912 	xfs_iext_insert(ip, &icur, &rec, 0);
913 
914 	XFS_IFORK_NEXT_SET(ip, whichfork, 1);
915 	ip->i_d.di_nblocks = 1;
916 	xfs_trans_mod_dquot_byino(tp, ip,
917 		XFS_TRANS_DQ_BCOUNT, 1L);
918 	flags |= xfs_ilog_fext(whichfork);
919 
920 done:
921 	*logflagsp = flags;
922 	return error;
923 }
924 
925 /*
926  * Called from xfs_bmap_add_attrfork to handle btree format files.
927  */
928 STATIC int					/* error */
929 xfs_bmap_add_attrfork_btree(
930 	xfs_trans_t		*tp,		/* transaction pointer */
931 	xfs_inode_t		*ip,		/* incore inode pointer */
932 	xfs_fsblock_t		*firstblock,	/* first block allocated */
933 	struct xfs_defer_ops	*dfops,		/* blocks to free at commit */
934 	int			*flags)		/* inode logging flags */
935 {
936 	xfs_btree_cur_t		*cur;		/* btree cursor */
937 	int			error;		/* error return value */
938 	xfs_mount_t		*mp;		/* file system mount struct */
939 	int			stat;		/* newroot status */
940 
941 	mp = ip->i_mount;
942 	if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
943 		*flags |= XFS_ILOG_DBROOT;
944 	else {
945 		cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
946 		cur->bc_private.b.dfops = dfops;
947 		cur->bc_private.b.firstblock = *firstblock;
948 		error = xfs_bmbt_lookup_first(cur, &stat);
949 		if (error)
950 			goto error0;
951 		/* must be at least one entry */
952 		XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0);
953 		if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
954 			goto error0;
955 		if (stat == 0) {
956 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
957 			return -ENOSPC;
958 		}
959 		*firstblock = cur->bc_private.b.firstblock;
960 		cur->bc_private.b.allocated = 0;
961 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
962 	}
963 	return 0;
964 error0:
965 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
966 	return error;
967 }
968 
969 /*
970  * Called from xfs_bmap_add_attrfork to handle extents format files.
971  */
972 STATIC int					/* error */
973 xfs_bmap_add_attrfork_extents(
974 	xfs_trans_t		*tp,		/* transaction pointer */
975 	xfs_inode_t		*ip,		/* incore inode pointer */
976 	xfs_fsblock_t		*firstblock,	/* first block allocated */
977 	struct xfs_defer_ops	*dfops,		/* blocks to free at commit */
978 	int			*flags)		/* inode logging flags */
979 {
980 	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
981 	int			error;		/* error return value */
982 
983 	if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
984 		return 0;
985 	cur = NULL;
986 	error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops, &cur, 0,
987 		flags, XFS_DATA_FORK);
988 	if (cur) {
989 		cur->bc_private.b.allocated = 0;
990 		xfs_btree_del_cursor(cur,
991 			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
992 	}
993 	return error;
994 }
995 
996 /*
997  * Called from xfs_bmap_add_attrfork to handle local format files. Each
998  * different data fork content type needs a different callout to do the
999  * conversion. Some are basic and only require special block initialisation
1000  * callouts for the data formating, others (directories) are so specialised they
1001  * handle everything themselves.
1002  *
1003  * XXX (dgc): investigate whether directory conversion can use the generic
1004  * formatting callout. It should be possible - it's just a very complex
1005  * formatter.
1006  */
1007 STATIC int					/* error */
1008 xfs_bmap_add_attrfork_local(
1009 	xfs_trans_t		*tp,		/* transaction pointer */
1010 	xfs_inode_t		*ip,		/* incore inode pointer */
1011 	xfs_fsblock_t		*firstblock,	/* first block allocated */
1012 	struct xfs_defer_ops	*dfops,		/* blocks to free at commit */
1013 	int			*flags)		/* inode logging flags */
1014 {
1015 	xfs_da_args_t		dargs;		/* args for dir/attr code */
1016 
1017 	if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1018 		return 0;
1019 
1020 	if (S_ISDIR(VFS_I(ip)->i_mode)) {
1021 		memset(&dargs, 0, sizeof(dargs));
1022 		dargs.geo = ip->i_mount->m_dir_geo;
1023 		dargs.dp = ip;
1024 		dargs.firstblock = firstblock;
1025 		dargs.dfops = dfops;
1026 		dargs.total = dargs.geo->fsbcount;
1027 		dargs.whichfork = XFS_DATA_FORK;
1028 		dargs.trans = tp;
1029 		return xfs_dir2_sf_to_block(&dargs);
1030 	}
1031 
1032 	if (S_ISLNK(VFS_I(ip)->i_mode))
1033 		return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1034 						 flags, XFS_DATA_FORK,
1035 						 xfs_symlink_local_to_remote);
1036 
1037 	/* should only be called for types that support local format data */
1038 	ASSERT(0);
1039 	return -EFSCORRUPTED;
1040 }
1041 
1042 /*
1043  * Convert inode from non-attributed to attributed.
1044  * Must not be in a transaction, ip must not be locked.
1045  */
1046 int						/* error code */
1047 xfs_bmap_add_attrfork(
1048 	xfs_inode_t		*ip,		/* incore inode pointer */
1049 	int			size,		/* space new attribute needs */
1050 	int			rsvd)		/* xact may use reserved blks */
1051 {
1052 	xfs_fsblock_t		firstblock;	/* 1st block/ag allocated */
1053 	struct xfs_defer_ops	dfops;		/* freed extent records */
1054 	xfs_mount_t		*mp;		/* mount structure */
1055 	xfs_trans_t		*tp;		/* transaction pointer */
1056 	int			blks;		/* space reservation */
1057 	int			version = 1;	/* superblock attr version */
1058 	int			logflags;	/* logging flags */
1059 	int			error;		/* error return value */
1060 
1061 	ASSERT(XFS_IFORK_Q(ip) == 0);
1062 
1063 	mp = ip->i_mount;
1064 	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1065 
1066 	blks = XFS_ADDAFORK_SPACE_RES(mp);
1067 
1068 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_addafork, blks, 0,
1069 			rsvd ? XFS_TRANS_RESERVE : 0, &tp);
1070 	if (error)
1071 		return error;
1072 
1073 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1074 	error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1075 			XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1076 			XFS_QMOPT_RES_REGBLKS);
1077 	if (error)
1078 		goto trans_cancel;
1079 	if (XFS_IFORK_Q(ip))
1080 		goto trans_cancel;
1081 	if (ip->i_d.di_anextents != 0) {
1082 		error = -EFSCORRUPTED;
1083 		goto trans_cancel;
1084 	}
1085 	if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1086 		/*
1087 		 * For inodes coming from pre-6.2 filesystems.
1088 		 */
1089 		ASSERT(ip->i_d.di_aformat == 0);
1090 		ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1091 	}
1092 
1093 	xfs_trans_ijoin(tp, ip, 0);
1094 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1095 
1096 	switch (ip->i_d.di_format) {
1097 	case XFS_DINODE_FMT_DEV:
1098 		ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1099 		break;
1100 	case XFS_DINODE_FMT_LOCAL:
1101 	case XFS_DINODE_FMT_EXTENTS:
1102 	case XFS_DINODE_FMT_BTREE:
1103 		ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1104 		if (!ip->i_d.di_forkoff)
1105 			ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1106 		else if (mp->m_flags & XFS_MOUNT_ATTR2)
1107 			version = 2;
1108 		break;
1109 	default:
1110 		ASSERT(0);
1111 		error = -EINVAL;
1112 		goto trans_cancel;
1113 	}
1114 
1115 	ASSERT(ip->i_afp == NULL);
1116 	ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1117 	ip->i_afp->if_flags = XFS_IFEXTENTS;
1118 	logflags = 0;
1119 	xfs_defer_init(&dfops, &firstblock);
1120 	switch (ip->i_d.di_format) {
1121 	case XFS_DINODE_FMT_LOCAL:
1122 		error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &dfops,
1123 			&logflags);
1124 		break;
1125 	case XFS_DINODE_FMT_EXTENTS:
1126 		error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1127 			&dfops, &logflags);
1128 		break;
1129 	case XFS_DINODE_FMT_BTREE:
1130 		error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &dfops,
1131 			&logflags);
1132 		break;
1133 	default:
1134 		error = 0;
1135 		break;
1136 	}
1137 	if (logflags)
1138 		xfs_trans_log_inode(tp, ip, logflags);
1139 	if (error)
1140 		goto bmap_cancel;
1141 	if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1142 	   (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1143 		bool log_sb = false;
1144 
1145 		spin_lock(&mp->m_sb_lock);
1146 		if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1147 			xfs_sb_version_addattr(&mp->m_sb);
1148 			log_sb = true;
1149 		}
1150 		if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1151 			xfs_sb_version_addattr2(&mp->m_sb);
1152 			log_sb = true;
1153 		}
1154 		spin_unlock(&mp->m_sb_lock);
1155 		if (log_sb)
1156 			xfs_log_sb(tp);
1157 	}
1158 
1159 	error = xfs_defer_finish(&tp, &dfops);
1160 	if (error)
1161 		goto bmap_cancel;
1162 	error = xfs_trans_commit(tp);
1163 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1164 	return error;
1165 
1166 bmap_cancel:
1167 	xfs_defer_cancel(&dfops);
1168 trans_cancel:
1169 	xfs_trans_cancel(tp);
1170 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1171 	return error;
1172 }
1173 
1174 /*
1175  * Internal and external extent tree search functions.
1176  */
1177 
1178 /*
1179  * Read in extents from a btree-format inode.
1180  */
1181 int
1182 xfs_iread_extents(
1183 	struct xfs_trans	*tp,
1184 	struct xfs_inode	*ip,
1185 	int			whichfork)
1186 {
1187 	struct xfs_mount	*mp = ip->i_mount;
1188 	int			state = xfs_bmap_fork_to_state(whichfork);
1189 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1190 	xfs_extnum_t		nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
1191 	struct xfs_btree_block	*block = ifp->if_broot;
1192 	struct xfs_iext_cursor	icur;
1193 	struct xfs_bmbt_irec	new;
1194 	xfs_fsblock_t		bno;
1195 	struct xfs_buf		*bp;
1196 	xfs_extnum_t		i, j;
1197 	int			level;
1198 	__be64			*pp;
1199 	int			error;
1200 
1201 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1202 
1203 	if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1204 		XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
1205 		return -EFSCORRUPTED;
1206 	}
1207 
1208 	/*
1209 	 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1210 	 */
1211 	level = be16_to_cpu(block->bb_level);
1212 	ASSERT(level > 0);
1213 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1214 	bno = be64_to_cpu(*pp);
1215 
1216 	/*
1217 	 * Go down the tree until leaf level is reached, following the first
1218 	 * pointer (leftmost) at each level.
1219 	 */
1220 	while (level-- > 0) {
1221 		error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1222 				XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1223 		if (error)
1224 			goto out;
1225 		block = XFS_BUF_TO_BLOCK(bp);
1226 		if (level == 0)
1227 			break;
1228 		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1229 		bno = be64_to_cpu(*pp);
1230 		XFS_WANT_CORRUPTED_GOTO(mp,
1231 			xfs_verify_fsbno(mp, bno), out_brelse);
1232 		xfs_trans_brelse(tp, bp);
1233 	}
1234 
1235 	/*
1236 	 * Here with bp and block set to the leftmost leaf node in the tree.
1237 	 */
1238 	i = 0;
1239 	xfs_iext_first(ifp, &icur);
1240 
1241 	/*
1242 	 * Loop over all leaf nodes.  Copy information to the extent records.
1243 	 */
1244 	for (;;) {
1245 		xfs_bmbt_rec_t	*frp;
1246 		xfs_fsblock_t	nextbno;
1247 		xfs_extnum_t	num_recs;
1248 
1249 		num_recs = xfs_btree_get_numrecs(block);
1250 		if (unlikely(i + num_recs > nextents)) {
1251 			ASSERT(i + num_recs <= nextents);
1252 			xfs_warn(ip->i_mount,
1253 				"corrupt dinode %Lu, (btree extents).",
1254 				(unsigned long long) ip->i_ino);
1255 			xfs_inode_verifier_error(ip, -EFSCORRUPTED,
1256 					__func__, block, sizeof(*block),
1257 					__this_address);
1258 			error = -EFSCORRUPTED;
1259 			goto out_brelse;
1260 		}
1261 		/*
1262 		 * Read-ahead the next leaf block, if any.
1263 		 */
1264 		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1265 		if (nextbno != NULLFSBLOCK)
1266 			xfs_btree_reada_bufl(mp, nextbno, 1,
1267 					     &xfs_bmbt_buf_ops);
1268 		/*
1269 		 * Copy records into the extent records.
1270 		 */
1271 		frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1272 		for (j = 0; j < num_recs; j++, frp++, i++) {
1273 			xfs_failaddr_t	fa;
1274 
1275 			xfs_bmbt_disk_get_all(frp, &new);
1276 			fa = xfs_bmap_validate_extent(ip, whichfork, &new);
1277 			if (fa) {
1278 				error = -EFSCORRUPTED;
1279 				xfs_inode_verifier_error(ip, error,
1280 						"xfs_iread_extents(2)",
1281 						frp, sizeof(*frp), fa);
1282 				goto out_brelse;
1283 			}
1284 			xfs_iext_insert(ip, &icur, &new, state);
1285 			trace_xfs_read_extent(ip, &icur, state, _THIS_IP_);
1286 			xfs_iext_next(ifp, &icur);
1287 		}
1288 		xfs_trans_brelse(tp, bp);
1289 		bno = nextbno;
1290 		/*
1291 		 * If we've reached the end, stop.
1292 		 */
1293 		if (bno == NULLFSBLOCK)
1294 			break;
1295 		error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1296 				XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1297 		if (error)
1298 			goto out;
1299 		block = XFS_BUF_TO_BLOCK(bp);
1300 	}
1301 
1302 	if (i != XFS_IFORK_NEXTENTS(ip, whichfork)) {
1303 		error = -EFSCORRUPTED;
1304 		goto out;
1305 	}
1306 	ASSERT(i == xfs_iext_count(ifp));
1307 
1308 	ifp->if_flags |= XFS_IFEXTENTS;
1309 	return 0;
1310 
1311 out_brelse:
1312 	xfs_trans_brelse(tp, bp);
1313 out:
1314 	xfs_iext_destroy(ifp);
1315 	return error;
1316 }
1317 
1318 /*
1319  * Returns the relative block number of the first unused block(s) in the given
1320  * fork with at least "len" logically contiguous blocks free.  This is the
1321  * lowest-address hole if the fork has holes, else the first block past the end
1322  * of fork.  Return 0 if the fork is currently local (in-inode).
1323  */
1324 int						/* error */
1325 xfs_bmap_first_unused(
1326 	struct xfs_trans	*tp,		/* transaction pointer */
1327 	struct xfs_inode	*ip,		/* incore inode */
1328 	xfs_extlen_t		len,		/* size of hole to find */
1329 	xfs_fileoff_t		*first_unused,	/* unused block */
1330 	int			whichfork)	/* data or attr fork */
1331 {
1332 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1333 	struct xfs_bmbt_irec	got;
1334 	struct xfs_iext_cursor	icur;
1335 	xfs_fileoff_t		lastaddr = 0;
1336 	xfs_fileoff_t		lowest, max;
1337 	int			error;
1338 
1339 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1340 	       XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1341 	       XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1342 
1343 	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1344 		*first_unused = 0;
1345 		return 0;
1346 	}
1347 
1348 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1349 		error = xfs_iread_extents(tp, ip, whichfork);
1350 		if (error)
1351 			return error;
1352 	}
1353 
1354 	lowest = max = *first_unused;
1355 	for_each_xfs_iext(ifp, &icur, &got) {
1356 		/*
1357 		 * See if the hole before this extent will work.
1358 		 */
1359 		if (got.br_startoff >= lowest + len &&
1360 		    got.br_startoff - max >= len)
1361 			break;
1362 		lastaddr = got.br_startoff + got.br_blockcount;
1363 		max = XFS_FILEOFF_MAX(lastaddr, lowest);
1364 	}
1365 
1366 	*first_unused = max;
1367 	return 0;
1368 }
1369 
1370 /*
1371  * Returns the file-relative block number of the last block - 1 before
1372  * last_block (input value) in the file.
1373  * This is not based on i_size, it is based on the extent records.
1374  * Returns 0 for local files, as they do not have extent records.
1375  */
1376 int						/* error */
1377 xfs_bmap_last_before(
1378 	struct xfs_trans	*tp,		/* transaction pointer */
1379 	struct xfs_inode	*ip,		/* incore inode */
1380 	xfs_fileoff_t		*last_block,	/* last block */
1381 	int			whichfork)	/* data or attr fork */
1382 {
1383 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1384 	struct xfs_bmbt_irec	got;
1385 	struct xfs_iext_cursor	icur;
1386 	int			error;
1387 
1388 	switch (XFS_IFORK_FORMAT(ip, whichfork)) {
1389 	case XFS_DINODE_FMT_LOCAL:
1390 		*last_block = 0;
1391 		return 0;
1392 	case XFS_DINODE_FMT_BTREE:
1393 	case XFS_DINODE_FMT_EXTENTS:
1394 		break;
1395 	default:
1396 		return -EIO;
1397 	}
1398 
1399 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1400 		error = xfs_iread_extents(tp, ip, whichfork);
1401 		if (error)
1402 			return error;
1403 	}
1404 
1405 	if (!xfs_iext_lookup_extent_before(ip, ifp, last_block, &icur, &got))
1406 		*last_block = 0;
1407 	return 0;
1408 }
1409 
1410 int
1411 xfs_bmap_last_extent(
1412 	struct xfs_trans	*tp,
1413 	struct xfs_inode	*ip,
1414 	int			whichfork,
1415 	struct xfs_bmbt_irec	*rec,
1416 	int			*is_empty)
1417 {
1418 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1419 	struct xfs_iext_cursor	icur;
1420 	int			error;
1421 
1422 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1423 		error = xfs_iread_extents(tp, ip, whichfork);
1424 		if (error)
1425 			return error;
1426 	}
1427 
1428 	xfs_iext_last(ifp, &icur);
1429 	if (!xfs_iext_get_extent(ifp, &icur, rec))
1430 		*is_empty = 1;
1431 	else
1432 		*is_empty = 0;
1433 	return 0;
1434 }
1435 
1436 /*
1437  * Check the last inode extent to determine whether this allocation will result
1438  * in blocks being allocated at the end of the file. When we allocate new data
1439  * blocks at the end of the file which do not start at the previous data block,
1440  * we will try to align the new blocks at stripe unit boundaries.
1441  *
1442  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1443  * at, or past the EOF.
1444  */
1445 STATIC int
1446 xfs_bmap_isaeof(
1447 	struct xfs_bmalloca	*bma,
1448 	int			whichfork)
1449 {
1450 	struct xfs_bmbt_irec	rec;
1451 	int			is_empty;
1452 	int			error;
1453 
1454 	bma->aeof = false;
1455 	error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1456 				     &is_empty);
1457 	if (error)
1458 		return error;
1459 
1460 	if (is_empty) {
1461 		bma->aeof = true;
1462 		return 0;
1463 	}
1464 
1465 	/*
1466 	 * Check if we are allocation or past the last extent, or at least into
1467 	 * the last delayed allocated extent.
1468 	 */
1469 	bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1470 		(bma->offset >= rec.br_startoff &&
1471 		 isnullstartblock(rec.br_startblock));
1472 	return 0;
1473 }
1474 
1475 /*
1476  * Returns the file-relative block number of the first block past eof in
1477  * the file.  This is not based on i_size, it is based on the extent records.
1478  * Returns 0 for local files, as they do not have extent records.
1479  */
1480 int
1481 xfs_bmap_last_offset(
1482 	struct xfs_inode	*ip,
1483 	xfs_fileoff_t		*last_block,
1484 	int			whichfork)
1485 {
1486 	struct xfs_bmbt_irec	rec;
1487 	int			is_empty;
1488 	int			error;
1489 
1490 	*last_block = 0;
1491 
1492 	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1493 		return 0;
1494 
1495 	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1496 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1497 	       return -EIO;
1498 
1499 	error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1500 	if (error || is_empty)
1501 		return error;
1502 
1503 	*last_block = rec.br_startoff + rec.br_blockcount;
1504 	return 0;
1505 }
1506 
1507 /*
1508  * Returns whether the selected fork of the inode has exactly one
1509  * block or not.  For the data fork we check this matches di_size,
1510  * implying the file's range is 0..bsize-1.
1511  */
1512 int					/* 1=>1 block, 0=>otherwise */
1513 xfs_bmap_one_block(
1514 	xfs_inode_t	*ip,		/* incore inode */
1515 	int		whichfork)	/* data or attr fork */
1516 {
1517 	xfs_ifork_t	*ifp;		/* inode fork pointer */
1518 	int		rval;		/* return value */
1519 	xfs_bmbt_irec_t	s;		/* internal version of extent */
1520 	struct xfs_iext_cursor icur;
1521 
1522 #ifndef DEBUG
1523 	if (whichfork == XFS_DATA_FORK)
1524 		return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1525 #endif	/* !DEBUG */
1526 	if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1527 		return 0;
1528 	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1529 		return 0;
1530 	ifp = XFS_IFORK_PTR(ip, whichfork);
1531 	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1532 	xfs_iext_first(ifp, &icur);
1533 	xfs_iext_get_extent(ifp, &icur, &s);
1534 	rval = s.br_startoff == 0 && s.br_blockcount == 1;
1535 	if (rval && whichfork == XFS_DATA_FORK)
1536 		ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1537 	return rval;
1538 }
1539 
1540 /*
1541  * Extent tree manipulation functions used during allocation.
1542  */
1543 
1544 /*
1545  * Convert a delayed allocation to a real allocation.
1546  */
1547 STATIC int				/* error */
1548 xfs_bmap_add_extent_delay_real(
1549 	struct xfs_bmalloca	*bma,
1550 	int			whichfork)
1551 {
1552 	struct xfs_bmbt_irec	*new = &bma->got;
1553 	int			error;	/* error return value */
1554 	int			i;	/* temp state */
1555 	xfs_ifork_t		*ifp;	/* inode fork pointer */
1556 	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
1557 	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
1558 					/* left is 0, right is 1, prev is 2 */
1559 	int			rval=0;	/* return value (logging flags) */
1560 	int			state = xfs_bmap_fork_to_state(whichfork);
1561 	xfs_filblks_t		da_new; /* new count del alloc blocks used */
1562 	xfs_filblks_t		da_old; /* old count del alloc blocks used */
1563 	xfs_filblks_t		temp=0;	/* value for da_new calculations */
1564 	int			tmp_rval;	/* partial logging flags */
1565 	struct xfs_mount	*mp;
1566 	xfs_extnum_t		*nextents;
1567 	struct xfs_bmbt_irec	old;
1568 
1569 	mp = bma->ip->i_mount;
1570 	ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1571 	ASSERT(whichfork != XFS_ATTR_FORK);
1572 	nextents = (whichfork == XFS_COW_FORK ? &bma->ip->i_cnextents :
1573 						&bma->ip->i_d.di_nextents);
1574 
1575 	ASSERT(!isnullstartblock(new->br_startblock));
1576 	ASSERT(!bma->cur ||
1577 	       (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1578 
1579 	XFS_STATS_INC(mp, xs_add_exlist);
1580 
1581 #define	LEFT		r[0]
1582 #define	RIGHT		r[1]
1583 #define	PREV		r[2]
1584 
1585 	/*
1586 	 * Set up a bunch of variables to make the tests simpler.
1587 	 */
1588 	xfs_iext_get_extent(ifp, &bma->icur, &PREV);
1589 	new_endoff = new->br_startoff + new->br_blockcount;
1590 	ASSERT(isnullstartblock(PREV.br_startblock));
1591 	ASSERT(PREV.br_startoff <= new->br_startoff);
1592 	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1593 
1594 	da_old = startblockval(PREV.br_startblock);
1595 	da_new = 0;
1596 
1597 	/*
1598 	 * Set flags determining what part of the previous delayed allocation
1599 	 * extent is being replaced by a real allocation.
1600 	 */
1601 	if (PREV.br_startoff == new->br_startoff)
1602 		state |= BMAP_LEFT_FILLING;
1603 	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1604 		state |= BMAP_RIGHT_FILLING;
1605 
1606 	/*
1607 	 * Check and set flags if this segment has a left neighbor.
1608 	 * Don't set contiguous if the combined extent would be too large.
1609 	 */
1610 	if (xfs_iext_peek_prev_extent(ifp, &bma->icur, &LEFT)) {
1611 		state |= BMAP_LEFT_VALID;
1612 		if (isnullstartblock(LEFT.br_startblock))
1613 			state |= BMAP_LEFT_DELAY;
1614 	}
1615 
1616 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1617 	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1618 	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1619 	    LEFT.br_state == new->br_state &&
1620 	    LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1621 		state |= BMAP_LEFT_CONTIG;
1622 
1623 	/*
1624 	 * Check and set flags if this segment has a right neighbor.
1625 	 * Don't set contiguous if the combined extent would be too large.
1626 	 * Also check for all-three-contiguous being too large.
1627 	 */
1628 	if (xfs_iext_peek_next_extent(ifp, &bma->icur, &RIGHT)) {
1629 		state |= BMAP_RIGHT_VALID;
1630 		if (isnullstartblock(RIGHT.br_startblock))
1631 			state |= BMAP_RIGHT_DELAY;
1632 	}
1633 
1634 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1635 	    new_endoff == RIGHT.br_startoff &&
1636 	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1637 	    new->br_state == RIGHT.br_state &&
1638 	    new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1639 	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1640 		       BMAP_RIGHT_FILLING)) !=
1641 		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1642 		       BMAP_RIGHT_FILLING) ||
1643 	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1644 			<= MAXEXTLEN))
1645 		state |= BMAP_RIGHT_CONTIG;
1646 
1647 	error = 0;
1648 	/*
1649 	 * Switch out based on the FILLING and CONTIG state bits.
1650 	 */
1651 	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1652 			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1653 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1654 	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1655 		/*
1656 		 * Filling in all of a previously delayed allocation extent.
1657 		 * The left and right neighbors are both contiguous with new.
1658 		 */
1659 		LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
1660 
1661 		xfs_iext_remove(bma->ip, &bma->icur, state);
1662 		xfs_iext_remove(bma->ip, &bma->icur, state);
1663 		xfs_iext_prev(ifp, &bma->icur);
1664 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1665 		(*nextents)--;
1666 
1667 		if (bma->cur == NULL)
1668 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1669 		else {
1670 			rval = XFS_ILOG_CORE;
1671 			error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1672 			if (error)
1673 				goto done;
1674 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1675 			error = xfs_btree_delete(bma->cur, &i);
1676 			if (error)
1677 				goto done;
1678 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1679 			error = xfs_btree_decrement(bma->cur, 0, &i);
1680 			if (error)
1681 				goto done;
1682 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1683 			error = xfs_bmbt_update(bma->cur, &LEFT);
1684 			if (error)
1685 				goto done;
1686 		}
1687 		break;
1688 
1689 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1690 		/*
1691 		 * Filling in all of a previously delayed allocation extent.
1692 		 * The left neighbor is contiguous, the right is not.
1693 		 */
1694 		old = LEFT;
1695 		LEFT.br_blockcount += PREV.br_blockcount;
1696 
1697 		xfs_iext_remove(bma->ip, &bma->icur, state);
1698 		xfs_iext_prev(ifp, &bma->icur);
1699 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1700 
1701 		if (bma->cur == NULL)
1702 			rval = XFS_ILOG_DEXT;
1703 		else {
1704 			rval = 0;
1705 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1706 			if (error)
1707 				goto done;
1708 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1709 			error = xfs_bmbt_update(bma->cur, &LEFT);
1710 			if (error)
1711 				goto done;
1712 		}
1713 		break;
1714 
1715 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1716 		/*
1717 		 * Filling in all of a previously delayed allocation extent.
1718 		 * The right neighbor is contiguous, the left is not.
1719 		 */
1720 		PREV.br_startblock = new->br_startblock;
1721 		PREV.br_blockcount += RIGHT.br_blockcount;
1722 
1723 		xfs_iext_next(ifp, &bma->icur);
1724 		xfs_iext_remove(bma->ip, &bma->icur, state);
1725 		xfs_iext_prev(ifp, &bma->icur);
1726 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1727 
1728 		if (bma->cur == NULL)
1729 			rval = XFS_ILOG_DEXT;
1730 		else {
1731 			rval = 0;
1732 			error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1733 			if (error)
1734 				goto done;
1735 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1736 			error = xfs_bmbt_update(bma->cur, &PREV);
1737 			if (error)
1738 				goto done;
1739 		}
1740 		break;
1741 
1742 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1743 		/*
1744 		 * Filling in all of a previously delayed allocation extent.
1745 		 * Neither the left nor right neighbors are contiguous with
1746 		 * the new one.
1747 		 */
1748 		PREV.br_startblock = new->br_startblock;
1749 		PREV.br_state = new->br_state;
1750 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1751 
1752 		(*nextents)++;
1753 		if (bma->cur == NULL)
1754 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1755 		else {
1756 			rval = XFS_ILOG_CORE;
1757 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1758 			if (error)
1759 				goto done;
1760 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1761 			error = xfs_btree_insert(bma->cur, &i);
1762 			if (error)
1763 				goto done;
1764 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1765 		}
1766 		break;
1767 
1768 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1769 		/*
1770 		 * Filling in the first part of a previous delayed allocation.
1771 		 * The left neighbor is contiguous.
1772 		 */
1773 		old = LEFT;
1774 		temp = PREV.br_blockcount - new->br_blockcount;
1775 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1776 				startblockval(PREV.br_startblock));
1777 
1778 		LEFT.br_blockcount += new->br_blockcount;
1779 
1780 		PREV.br_blockcount = temp;
1781 		PREV.br_startoff += new->br_blockcount;
1782 		PREV.br_startblock = nullstartblock(da_new);
1783 
1784 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1785 		xfs_iext_prev(ifp, &bma->icur);
1786 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1787 
1788 		if (bma->cur == NULL)
1789 			rval = XFS_ILOG_DEXT;
1790 		else {
1791 			rval = 0;
1792 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1793 			if (error)
1794 				goto done;
1795 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1796 			error = xfs_bmbt_update(bma->cur, &LEFT);
1797 			if (error)
1798 				goto done;
1799 		}
1800 		break;
1801 
1802 	case BMAP_LEFT_FILLING:
1803 		/*
1804 		 * Filling in the first part of a previous delayed allocation.
1805 		 * The left neighbor is not contiguous.
1806 		 */
1807 		xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1808 		(*nextents)++;
1809 		if (bma->cur == NULL)
1810 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1811 		else {
1812 			rval = XFS_ILOG_CORE;
1813 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1814 			if (error)
1815 				goto done;
1816 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1817 			error = xfs_btree_insert(bma->cur, &i);
1818 			if (error)
1819 				goto done;
1820 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1821 		}
1822 
1823 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1824 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1825 					bma->firstblock, bma->dfops,
1826 					&bma->cur, 1, &tmp_rval, whichfork);
1827 			rval |= tmp_rval;
1828 			if (error)
1829 				goto done;
1830 		}
1831 
1832 		temp = PREV.br_blockcount - new->br_blockcount;
1833 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1834 			startblockval(PREV.br_startblock) -
1835 			(bma->cur ? bma->cur->bc_private.b.allocated : 0));
1836 
1837 		PREV.br_startoff = new_endoff;
1838 		PREV.br_blockcount = temp;
1839 		PREV.br_startblock = nullstartblock(da_new);
1840 		xfs_iext_next(ifp, &bma->icur);
1841 		xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1842 		xfs_iext_prev(ifp, &bma->icur);
1843 		break;
1844 
1845 	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1846 		/*
1847 		 * Filling in the last part of a previous delayed allocation.
1848 		 * The right neighbor is contiguous with the new allocation.
1849 		 */
1850 		old = RIGHT;
1851 		RIGHT.br_startoff = new->br_startoff;
1852 		RIGHT.br_startblock = new->br_startblock;
1853 		RIGHT.br_blockcount += new->br_blockcount;
1854 
1855 		if (bma->cur == NULL)
1856 			rval = XFS_ILOG_DEXT;
1857 		else {
1858 			rval = 0;
1859 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1860 			if (error)
1861 				goto done;
1862 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1863 			error = xfs_bmbt_update(bma->cur, &RIGHT);
1864 			if (error)
1865 				goto done;
1866 		}
1867 
1868 		temp = PREV.br_blockcount - new->br_blockcount;
1869 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1870 			startblockval(PREV.br_startblock));
1871 
1872 		PREV.br_blockcount = temp;
1873 		PREV.br_startblock = nullstartblock(da_new);
1874 
1875 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1876 		xfs_iext_next(ifp, &bma->icur);
1877 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &RIGHT);
1878 		break;
1879 
1880 	case BMAP_RIGHT_FILLING:
1881 		/*
1882 		 * Filling in the last part of a previous delayed allocation.
1883 		 * The right neighbor is not contiguous.
1884 		 */
1885 		xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1886 		(*nextents)++;
1887 		if (bma->cur == NULL)
1888 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1889 		else {
1890 			rval = XFS_ILOG_CORE;
1891 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1892 			if (error)
1893 				goto done;
1894 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1895 			error = xfs_btree_insert(bma->cur, &i);
1896 			if (error)
1897 				goto done;
1898 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1899 		}
1900 
1901 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1902 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1903 				bma->firstblock, bma->dfops, &bma->cur, 1,
1904 				&tmp_rval, whichfork);
1905 			rval |= tmp_rval;
1906 			if (error)
1907 				goto done;
1908 		}
1909 
1910 		temp = PREV.br_blockcount - new->br_blockcount;
1911 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1912 			startblockval(PREV.br_startblock) -
1913 			(bma->cur ? bma->cur->bc_private.b.allocated : 0));
1914 
1915 		PREV.br_startblock = nullstartblock(da_new);
1916 		PREV.br_blockcount = temp;
1917 		xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1918 		xfs_iext_next(ifp, &bma->icur);
1919 		break;
1920 
1921 	case 0:
1922 		/*
1923 		 * Filling in the middle part of a previous delayed allocation.
1924 		 * Contiguity is impossible here.
1925 		 * This case is avoided almost all the time.
1926 		 *
1927 		 * We start with a delayed allocation:
1928 		 *
1929 		 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1930 		 *  PREV @ idx
1931 		 *
1932 	         * and we are allocating:
1933 		 *                     +rrrrrrrrrrrrrrrrr+
1934 		 *			      new
1935 		 *
1936 		 * and we set it up for insertion as:
1937 		 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1938 		 *                            new
1939 		 *  PREV @ idx          LEFT              RIGHT
1940 		 *                      inserted at idx + 1
1941 		 */
1942 		old = PREV;
1943 
1944 		/* LEFT is the new middle */
1945 		LEFT = *new;
1946 
1947 		/* RIGHT is the new right */
1948 		RIGHT.br_state = PREV.br_state;
1949 		RIGHT.br_startoff = new_endoff;
1950 		RIGHT.br_blockcount =
1951 			PREV.br_startoff + PREV.br_blockcount - new_endoff;
1952 		RIGHT.br_startblock =
1953 			nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1954 					RIGHT.br_blockcount));
1955 
1956 		/* truncate PREV */
1957 		PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
1958 		PREV.br_startblock =
1959 			nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1960 					PREV.br_blockcount));
1961 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1962 
1963 		xfs_iext_next(ifp, &bma->icur);
1964 		xfs_iext_insert(bma->ip, &bma->icur, &RIGHT, state);
1965 		xfs_iext_insert(bma->ip, &bma->icur, &LEFT, state);
1966 		(*nextents)++;
1967 
1968 		if (bma->cur == NULL)
1969 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1970 		else {
1971 			rval = XFS_ILOG_CORE;
1972 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1973 			if (error)
1974 				goto done;
1975 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1976 			error = xfs_btree_insert(bma->cur, &i);
1977 			if (error)
1978 				goto done;
1979 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1980 		}
1981 
1982 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1983 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1984 					bma->firstblock, bma->dfops, &bma->cur,
1985 					1, &tmp_rval, whichfork);
1986 			rval |= tmp_rval;
1987 			if (error)
1988 				goto done;
1989 		}
1990 
1991 		da_new = startblockval(PREV.br_startblock) +
1992 			 startblockval(RIGHT.br_startblock);
1993 		break;
1994 
1995 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1996 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1997 	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1998 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1999 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2000 	case BMAP_LEFT_CONTIG:
2001 	case BMAP_RIGHT_CONTIG:
2002 		/*
2003 		 * These cases are all impossible.
2004 		 */
2005 		ASSERT(0);
2006 	}
2007 
2008 	/* add reverse mapping unless caller opted out */
2009 	if (!(bma->flags & XFS_BMAPI_NORMAP)) {
2010 		error = xfs_rmap_map_extent(mp, bma->dfops, bma->ip,
2011 				whichfork, new);
2012 		if (error)
2013 			goto done;
2014 	}
2015 
2016 	/* convert to a btree if necessary */
2017 	if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2018 		int	tmp_logflags;	/* partial log flag return val */
2019 
2020 		ASSERT(bma->cur == NULL);
2021 		error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2022 				bma->firstblock, bma->dfops, &bma->cur,
2023 				da_old > 0, &tmp_logflags, whichfork);
2024 		bma->logflags |= tmp_logflags;
2025 		if (error)
2026 			goto done;
2027 	}
2028 
2029 	if (bma->cur) {
2030 		da_new += bma->cur->bc_private.b.allocated;
2031 		bma->cur->bc_private.b.allocated = 0;
2032 	}
2033 
2034 	/* adjust for changes in reserved delayed indirect blocks */
2035 	if (da_new != da_old) {
2036 		ASSERT(state == 0 || da_new < da_old);
2037 		error = xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new),
2038 				false);
2039 	}
2040 
2041 	xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
2042 done:
2043 	if (whichfork != XFS_COW_FORK)
2044 		bma->logflags |= rval;
2045 	return error;
2046 #undef	LEFT
2047 #undef	RIGHT
2048 #undef	PREV
2049 }
2050 
2051 /*
2052  * Convert an unwritten allocation to a real allocation or vice versa.
2053  */
2054 STATIC int				/* error */
2055 xfs_bmap_add_extent_unwritten_real(
2056 	struct xfs_trans	*tp,
2057 	xfs_inode_t		*ip,	/* incore inode pointer */
2058 	int			whichfork,
2059 	struct xfs_iext_cursor	*icur,
2060 	xfs_btree_cur_t		**curp,	/* if *curp is null, not a btree */
2061 	xfs_bmbt_irec_t		*new,	/* new data to add to file extents */
2062 	xfs_fsblock_t		*first,	/* pointer to firstblock variable */
2063 	struct xfs_defer_ops	*dfops,	/* list of extents to be freed */
2064 	int			*logflagsp) /* inode logging flags */
2065 {
2066 	xfs_btree_cur_t		*cur;	/* btree cursor */
2067 	int			error;	/* error return value */
2068 	int			i;	/* temp state */
2069 	xfs_ifork_t		*ifp;	/* inode fork pointer */
2070 	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
2071 	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
2072 					/* left is 0, right is 1, prev is 2 */
2073 	int			rval=0;	/* return value (logging flags) */
2074 	int			state = xfs_bmap_fork_to_state(whichfork);
2075 	struct xfs_mount	*mp = ip->i_mount;
2076 	struct xfs_bmbt_irec	old;
2077 
2078 	*logflagsp = 0;
2079 
2080 	cur = *curp;
2081 	ifp = XFS_IFORK_PTR(ip, whichfork);
2082 
2083 	ASSERT(!isnullstartblock(new->br_startblock));
2084 
2085 	XFS_STATS_INC(mp, xs_add_exlist);
2086 
2087 #define	LEFT		r[0]
2088 #define	RIGHT		r[1]
2089 #define	PREV		r[2]
2090 
2091 	/*
2092 	 * Set up a bunch of variables to make the tests simpler.
2093 	 */
2094 	error = 0;
2095 	xfs_iext_get_extent(ifp, icur, &PREV);
2096 	ASSERT(new->br_state != PREV.br_state);
2097 	new_endoff = new->br_startoff + new->br_blockcount;
2098 	ASSERT(PREV.br_startoff <= new->br_startoff);
2099 	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2100 
2101 	/*
2102 	 * Set flags determining what part of the previous oldext allocation
2103 	 * extent is being replaced by a newext allocation.
2104 	 */
2105 	if (PREV.br_startoff == new->br_startoff)
2106 		state |= BMAP_LEFT_FILLING;
2107 	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2108 		state |= BMAP_RIGHT_FILLING;
2109 
2110 	/*
2111 	 * Check and set flags if this segment has a left neighbor.
2112 	 * Don't set contiguous if the combined extent would be too large.
2113 	 */
2114 	if (xfs_iext_peek_prev_extent(ifp, icur, &LEFT)) {
2115 		state |= BMAP_LEFT_VALID;
2116 		if (isnullstartblock(LEFT.br_startblock))
2117 			state |= BMAP_LEFT_DELAY;
2118 	}
2119 
2120 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2121 	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2122 	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2123 	    LEFT.br_state == new->br_state &&
2124 	    LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2125 		state |= BMAP_LEFT_CONTIG;
2126 
2127 	/*
2128 	 * Check and set flags if this segment has a right neighbor.
2129 	 * Don't set contiguous if the combined extent would be too large.
2130 	 * Also check for all-three-contiguous being too large.
2131 	 */
2132 	if (xfs_iext_peek_next_extent(ifp, icur, &RIGHT)) {
2133 		state |= BMAP_RIGHT_VALID;
2134 		if (isnullstartblock(RIGHT.br_startblock))
2135 			state |= BMAP_RIGHT_DELAY;
2136 	}
2137 
2138 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2139 	    new_endoff == RIGHT.br_startoff &&
2140 	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2141 	    new->br_state == RIGHT.br_state &&
2142 	    new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2143 	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2144 		       BMAP_RIGHT_FILLING)) !=
2145 		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2146 		       BMAP_RIGHT_FILLING) ||
2147 	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2148 			<= MAXEXTLEN))
2149 		state |= BMAP_RIGHT_CONTIG;
2150 
2151 	/*
2152 	 * Switch out based on the FILLING and CONTIG state bits.
2153 	 */
2154 	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2155 			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2156 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2157 	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2158 		/*
2159 		 * Setting all of a previous oldext extent to newext.
2160 		 * The left and right neighbors are both contiguous with new.
2161 		 */
2162 		LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
2163 
2164 		xfs_iext_remove(ip, icur, state);
2165 		xfs_iext_remove(ip, icur, state);
2166 		xfs_iext_prev(ifp, icur);
2167 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2168 		XFS_IFORK_NEXT_SET(ip, whichfork,
2169 				XFS_IFORK_NEXTENTS(ip, whichfork) - 2);
2170 		if (cur == NULL)
2171 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2172 		else {
2173 			rval = XFS_ILOG_CORE;
2174 			error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2175 			if (error)
2176 				goto done;
2177 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2178 			if ((error = xfs_btree_delete(cur, &i)))
2179 				goto done;
2180 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2181 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2182 				goto done;
2183 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2184 			if ((error = xfs_btree_delete(cur, &i)))
2185 				goto done;
2186 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2187 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2188 				goto done;
2189 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2190 			error = xfs_bmbt_update(cur, &LEFT);
2191 			if (error)
2192 				goto done;
2193 		}
2194 		break;
2195 
2196 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2197 		/*
2198 		 * Setting all of a previous oldext extent to newext.
2199 		 * The left neighbor is contiguous, the right is not.
2200 		 */
2201 		LEFT.br_blockcount += PREV.br_blockcount;
2202 
2203 		xfs_iext_remove(ip, icur, state);
2204 		xfs_iext_prev(ifp, icur);
2205 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2206 		XFS_IFORK_NEXT_SET(ip, whichfork,
2207 				XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2208 		if (cur == NULL)
2209 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2210 		else {
2211 			rval = XFS_ILOG_CORE;
2212 			error = xfs_bmbt_lookup_eq(cur, &PREV, &i);
2213 			if (error)
2214 				goto done;
2215 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2216 			if ((error = xfs_btree_delete(cur, &i)))
2217 				goto done;
2218 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2219 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2220 				goto done;
2221 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2222 			error = xfs_bmbt_update(cur, &LEFT);
2223 			if (error)
2224 				goto done;
2225 		}
2226 		break;
2227 
2228 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2229 		/*
2230 		 * Setting all of a previous oldext extent to newext.
2231 		 * The right neighbor is contiguous, the left is not.
2232 		 */
2233 		PREV.br_blockcount += RIGHT.br_blockcount;
2234 		PREV.br_state = new->br_state;
2235 
2236 		xfs_iext_next(ifp, icur);
2237 		xfs_iext_remove(ip, icur, state);
2238 		xfs_iext_prev(ifp, icur);
2239 		xfs_iext_update_extent(ip, state, icur, &PREV);
2240 
2241 		XFS_IFORK_NEXT_SET(ip, whichfork,
2242 				XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2243 		if (cur == NULL)
2244 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2245 		else {
2246 			rval = XFS_ILOG_CORE;
2247 			error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2248 			if (error)
2249 				goto done;
2250 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2251 			if ((error = xfs_btree_delete(cur, &i)))
2252 				goto done;
2253 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2254 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2255 				goto done;
2256 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2257 			error = xfs_bmbt_update(cur, &PREV);
2258 			if (error)
2259 				goto done;
2260 		}
2261 		break;
2262 
2263 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2264 		/*
2265 		 * Setting all of a previous oldext extent to newext.
2266 		 * Neither the left nor right neighbors are contiguous with
2267 		 * the new one.
2268 		 */
2269 		PREV.br_state = new->br_state;
2270 		xfs_iext_update_extent(ip, state, icur, &PREV);
2271 
2272 		if (cur == NULL)
2273 			rval = XFS_ILOG_DEXT;
2274 		else {
2275 			rval = 0;
2276 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2277 			if (error)
2278 				goto done;
2279 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2280 			error = xfs_bmbt_update(cur, &PREV);
2281 			if (error)
2282 				goto done;
2283 		}
2284 		break;
2285 
2286 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2287 		/*
2288 		 * Setting the first part of a previous oldext extent to newext.
2289 		 * The left neighbor is contiguous.
2290 		 */
2291 		LEFT.br_blockcount += new->br_blockcount;
2292 
2293 		old = PREV;
2294 		PREV.br_startoff += new->br_blockcount;
2295 		PREV.br_startblock += new->br_blockcount;
2296 		PREV.br_blockcount -= new->br_blockcount;
2297 
2298 		xfs_iext_update_extent(ip, state, icur, &PREV);
2299 		xfs_iext_prev(ifp, icur);
2300 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2301 
2302 		if (cur == NULL)
2303 			rval = XFS_ILOG_DEXT;
2304 		else {
2305 			rval = 0;
2306 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2307 			if (error)
2308 				goto done;
2309 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2310 			error = xfs_bmbt_update(cur, &PREV);
2311 			if (error)
2312 				goto done;
2313 			error = xfs_btree_decrement(cur, 0, &i);
2314 			if (error)
2315 				goto done;
2316 			error = xfs_bmbt_update(cur, &LEFT);
2317 			if (error)
2318 				goto done;
2319 		}
2320 		break;
2321 
2322 	case BMAP_LEFT_FILLING:
2323 		/*
2324 		 * Setting the first part of a previous oldext extent to newext.
2325 		 * The left neighbor is not contiguous.
2326 		 */
2327 		old = PREV;
2328 		PREV.br_startoff += new->br_blockcount;
2329 		PREV.br_startblock += new->br_blockcount;
2330 		PREV.br_blockcount -= new->br_blockcount;
2331 
2332 		xfs_iext_update_extent(ip, state, icur, &PREV);
2333 		xfs_iext_insert(ip, icur, new, state);
2334 		XFS_IFORK_NEXT_SET(ip, whichfork,
2335 				XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2336 		if (cur == NULL)
2337 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2338 		else {
2339 			rval = XFS_ILOG_CORE;
2340 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2341 			if (error)
2342 				goto done;
2343 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2344 			error = xfs_bmbt_update(cur, &PREV);
2345 			if (error)
2346 				goto done;
2347 			cur->bc_rec.b = *new;
2348 			if ((error = xfs_btree_insert(cur, &i)))
2349 				goto done;
2350 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2351 		}
2352 		break;
2353 
2354 	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2355 		/*
2356 		 * Setting the last part of a previous oldext extent to newext.
2357 		 * The right neighbor is contiguous with the new allocation.
2358 		 */
2359 		old = PREV;
2360 		PREV.br_blockcount -= new->br_blockcount;
2361 
2362 		RIGHT.br_startoff = new->br_startoff;
2363 		RIGHT.br_startblock = new->br_startblock;
2364 		RIGHT.br_blockcount += new->br_blockcount;
2365 
2366 		xfs_iext_update_extent(ip, state, icur, &PREV);
2367 		xfs_iext_next(ifp, icur);
2368 		xfs_iext_update_extent(ip, state, icur, &RIGHT);
2369 
2370 		if (cur == NULL)
2371 			rval = XFS_ILOG_DEXT;
2372 		else {
2373 			rval = 0;
2374 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2375 			if (error)
2376 				goto done;
2377 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2378 			error = xfs_bmbt_update(cur, &PREV);
2379 			if (error)
2380 				goto done;
2381 			error = xfs_btree_increment(cur, 0, &i);
2382 			if (error)
2383 				goto done;
2384 			error = xfs_bmbt_update(cur, &RIGHT);
2385 			if (error)
2386 				goto done;
2387 		}
2388 		break;
2389 
2390 	case BMAP_RIGHT_FILLING:
2391 		/*
2392 		 * Setting the last part of a previous oldext extent to newext.
2393 		 * The right neighbor is not contiguous.
2394 		 */
2395 		old = PREV;
2396 		PREV.br_blockcount -= new->br_blockcount;
2397 
2398 		xfs_iext_update_extent(ip, state, icur, &PREV);
2399 		xfs_iext_next(ifp, icur);
2400 		xfs_iext_insert(ip, icur, new, state);
2401 
2402 		XFS_IFORK_NEXT_SET(ip, whichfork,
2403 				XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2404 		if (cur == NULL)
2405 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2406 		else {
2407 			rval = XFS_ILOG_CORE;
2408 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2409 			if (error)
2410 				goto done;
2411 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2412 			error = xfs_bmbt_update(cur, &PREV);
2413 			if (error)
2414 				goto done;
2415 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2416 			if (error)
2417 				goto done;
2418 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2419 			if ((error = xfs_btree_insert(cur, &i)))
2420 				goto done;
2421 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2422 		}
2423 		break;
2424 
2425 	case 0:
2426 		/*
2427 		 * Setting the middle part of a previous oldext extent to
2428 		 * newext.  Contiguity is impossible here.
2429 		 * One extent becomes three extents.
2430 		 */
2431 		old = PREV;
2432 		PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
2433 
2434 		r[0] = *new;
2435 		r[1].br_startoff = new_endoff;
2436 		r[1].br_blockcount =
2437 			old.br_startoff + old.br_blockcount - new_endoff;
2438 		r[1].br_startblock = new->br_startblock + new->br_blockcount;
2439 		r[1].br_state = PREV.br_state;
2440 
2441 		xfs_iext_update_extent(ip, state, icur, &PREV);
2442 		xfs_iext_next(ifp, icur);
2443 		xfs_iext_insert(ip, icur, &r[1], state);
2444 		xfs_iext_insert(ip, icur, &r[0], state);
2445 
2446 		XFS_IFORK_NEXT_SET(ip, whichfork,
2447 				XFS_IFORK_NEXTENTS(ip, whichfork) + 2);
2448 		if (cur == NULL)
2449 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2450 		else {
2451 			rval = XFS_ILOG_CORE;
2452 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2453 			if (error)
2454 				goto done;
2455 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2456 			/* new right extent - oldext */
2457 			error = xfs_bmbt_update(cur, &r[1]);
2458 			if (error)
2459 				goto done;
2460 			/* new left extent - oldext */
2461 			cur->bc_rec.b = PREV;
2462 			if ((error = xfs_btree_insert(cur, &i)))
2463 				goto done;
2464 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2465 			/*
2466 			 * Reset the cursor to the position of the new extent
2467 			 * we are about to insert as we can't trust it after
2468 			 * the previous insert.
2469 			 */
2470 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2471 			if (error)
2472 				goto done;
2473 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2474 			/* new middle extent - newext */
2475 			if ((error = xfs_btree_insert(cur, &i)))
2476 				goto done;
2477 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2478 		}
2479 		break;
2480 
2481 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2482 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2483 	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2484 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2485 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2486 	case BMAP_LEFT_CONTIG:
2487 	case BMAP_RIGHT_CONTIG:
2488 		/*
2489 		 * These cases are all impossible.
2490 		 */
2491 		ASSERT(0);
2492 	}
2493 
2494 	/* update reverse mappings */
2495 	error = xfs_rmap_convert_extent(mp, dfops, ip, whichfork, new);
2496 	if (error)
2497 		goto done;
2498 
2499 	/* convert to a btree if necessary */
2500 	if (xfs_bmap_needs_btree(ip, whichfork)) {
2501 		int	tmp_logflags;	/* partial log flag return val */
2502 
2503 		ASSERT(cur == NULL);
2504 		error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, &cur,
2505 				0, &tmp_logflags, whichfork);
2506 		*logflagsp |= tmp_logflags;
2507 		if (error)
2508 			goto done;
2509 	}
2510 
2511 	/* clear out the allocated field, done with it now in any case. */
2512 	if (cur) {
2513 		cur->bc_private.b.allocated = 0;
2514 		*curp = cur;
2515 	}
2516 
2517 	xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
2518 done:
2519 	*logflagsp |= rval;
2520 	return error;
2521 #undef	LEFT
2522 #undef	RIGHT
2523 #undef	PREV
2524 }
2525 
2526 /*
2527  * Convert a hole to a delayed allocation.
2528  */
2529 STATIC void
2530 xfs_bmap_add_extent_hole_delay(
2531 	xfs_inode_t		*ip,	/* incore inode pointer */
2532 	int			whichfork,
2533 	struct xfs_iext_cursor	*icur,
2534 	xfs_bmbt_irec_t		*new)	/* new data to add to file extents */
2535 {
2536 	xfs_ifork_t		*ifp;	/* inode fork pointer */
2537 	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2538 	xfs_filblks_t		newlen=0;	/* new indirect size */
2539 	xfs_filblks_t		oldlen=0;	/* old indirect size */
2540 	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2541 	int			state = xfs_bmap_fork_to_state(whichfork);
2542 	xfs_filblks_t		temp;	 /* temp for indirect calculations */
2543 
2544 	ifp = XFS_IFORK_PTR(ip, whichfork);
2545 	ASSERT(isnullstartblock(new->br_startblock));
2546 
2547 	/*
2548 	 * Check and set flags if this segment has a left neighbor
2549 	 */
2550 	if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2551 		state |= BMAP_LEFT_VALID;
2552 		if (isnullstartblock(left.br_startblock))
2553 			state |= BMAP_LEFT_DELAY;
2554 	}
2555 
2556 	/*
2557 	 * Check and set flags if the current (right) segment exists.
2558 	 * If it doesn't exist, we're converting the hole at end-of-file.
2559 	 */
2560 	if (xfs_iext_get_extent(ifp, icur, &right)) {
2561 		state |= BMAP_RIGHT_VALID;
2562 		if (isnullstartblock(right.br_startblock))
2563 			state |= BMAP_RIGHT_DELAY;
2564 	}
2565 
2566 	/*
2567 	 * Set contiguity flags on the left and right neighbors.
2568 	 * Don't let extents get too large, even if the pieces are contiguous.
2569 	 */
2570 	if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2571 	    left.br_startoff + left.br_blockcount == new->br_startoff &&
2572 	    left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2573 		state |= BMAP_LEFT_CONTIG;
2574 
2575 	if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2576 	    new->br_startoff + new->br_blockcount == right.br_startoff &&
2577 	    new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2578 	    (!(state & BMAP_LEFT_CONTIG) ||
2579 	     (left.br_blockcount + new->br_blockcount +
2580 	      right.br_blockcount <= MAXEXTLEN)))
2581 		state |= BMAP_RIGHT_CONTIG;
2582 
2583 	/*
2584 	 * Switch out based on the contiguity flags.
2585 	 */
2586 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2587 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2588 		/*
2589 		 * New allocation is contiguous with delayed allocations
2590 		 * on the left and on the right.
2591 		 * Merge all three into a single extent record.
2592 		 */
2593 		temp = left.br_blockcount + new->br_blockcount +
2594 			right.br_blockcount;
2595 
2596 		oldlen = startblockval(left.br_startblock) +
2597 			startblockval(new->br_startblock) +
2598 			startblockval(right.br_startblock);
2599 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2600 					 oldlen);
2601 		left.br_startblock = nullstartblock(newlen);
2602 		left.br_blockcount = temp;
2603 
2604 		xfs_iext_remove(ip, icur, state);
2605 		xfs_iext_prev(ifp, icur);
2606 		xfs_iext_update_extent(ip, state, icur, &left);
2607 		break;
2608 
2609 	case BMAP_LEFT_CONTIG:
2610 		/*
2611 		 * New allocation is contiguous with a delayed allocation
2612 		 * on the left.
2613 		 * Merge the new allocation with the left neighbor.
2614 		 */
2615 		temp = left.br_blockcount + new->br_blockcount;
2616 
2617 		oldlen = startblockval(left.br_startblock) +
2618 			startblockval(new->br_startblock);
2619 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2620 					 oldlen);
2621 		left.br_blockcount = temp;
2622 		left.br_startblock = nullstartblock(newlen);
2623 
2624 		xfs_iext_prev(ifp, icur);
2625 		xfs_iext_update_extent(ip, state, icur, &left);
2626 		break;
2627 
2628 	case BMAP_RIGHT_CONTIG:
2629 		/*
2630 		 * New allocation is contiguous with a delayed allocation
2631 		 * on the right.
2632 		 * Merge the new allocation with the right neighbor.
2633 		 */
2634 		temp = new->br_blockcount + right.br_blockcount;
2635 		oldlen = startblockval(new->br_startblock) +
2636 			startblockval(right.br_startblock);
2637 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2638 					 oldlen);
2639 		right.br_startoff = new->br_startoff;
2640 		right.br_startblock = nullstartblock(newlen);
2641 		right.br_blockcount = temp;
2642 		xfs_iext_update_extent(ip, state, icur, &right);
2643 		break;
2644 
2645 	case 0:
2646 		/*
2647 		 * New allocation is not contiguous with another
2648 		 * delayed allocation.
2649 		 * Insert a new entry.
2650 		 */
2651 		oldlen = newlen = 0;
2652 		xfs_iext_insert(ip, icur, new, state);
2653 		break;
2654 	}
2655 	if (oldlen != newlen) {
2656 		ASSERT(oldlen > newlen);
2657 		xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2658 				 false);
2659 		/*
2660 		 * Nothing to do for disk quota accounting here.
2661 		 */
2662 	}
2663 }
2664 
2665 /*
2666  * Convert a hole to a real allocation.
2667  */
2668 STATIC int				/* error */
2669 xfs_bmap_add_extent_hole_real(
2670 	struct xfs_trans	*tp,
2671 	struct xfs_inode	*ip,
2672 	int			whichfork,
2673 	struct xfs_iext_cursor	*icur,
2674 	struct xfs_btree_cur	**curp,
2675 	struct xfs_bmbt_irec	*new,
2676 	xfs_fsblock_t		*first,
2677 	struct xfs_defer_ops	*dfops,
2678 	int			*logflagsp,
2679 	int			flags)
2680 {
2681 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
2682 	struct xfs_mount	*mp = ip->i_mount;
2683 	struct xfs_btree_cur	*cur = *curp;
2684 	int			error;	/* error return value */
2685 	int			i;	/* temp state */
2686 	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2687 	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2688 	int			rval=0;	/* return value (logging flags) */
2689 	int			state = xfs_bmap_fork_to_state(whichfork);
2690 	struct xfs_bmbt_irec	old;
2691 
2692 	ASSERT(!isnullstartblock(new->br_startblock));
2693 	ASSERT(!cur || !(cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2694 
2695 	XFS_STATS_INC(mp, xs_add_exlist);
2696 
2697 	/*
2698 	 * Check and set flags if this segment has a left neighbor.
2699 	 */
2700 	if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2701 		state |= BMAP_LEFT_VALID;
2702 		if (isnullstartblock(left.br_startblock))
2703 			state |= BMAP_LEFT_DELAY;
2704 	}
2705 
2706 	/*
2707 	 * Check and set flags if this segment has a current value.
2708 	 * Not true if we're inserting into the "hole" at eof.
2709 	 */
2710 	if (xfs_iext_get_extent(ifp, icur, &right)) {
2711 		state |= BMAP_RIGHT_VALID;
2712 		if (isnullstartblock(right.br_startblock))
2713 			state |= BMAP_RIGHT_DELAY;
2714 	}
2715 
2716 	/*
2717 	 * We're inserting a real allocation between "left" and "right".
2718 	 * Set the contiguity flags.  Don't let extents get too large.
2719 	 */
2720 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2721 	    left.br_startoff + left.br_blockcount == new->br_startoff &&
2722 	    left.br_startblock + left.br_blockcount == new->br_startblock &&
2723 	    left.br_state == new->br_state &&
2724 	    left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2725 		state |= BMAP_LEFT_CONTIG;
2726 
2727 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2728 	    new->br_startoff + new->br_blockcount == right.br_startoff &&
2729 	    new->br_startblock + new->br_blockcount == right.br_startblock &&
2730 	    new->br_state == right.br_state &&
2731 	    new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2732 	    (!(state & BMAP_LEFT_CONTIG) ||
2733 	     left.br_blockcount + new->br_blockcount +
2734 	     right.br_blockcount <= MAXEXTLEN))
2735 		state |= BMAP_RIGHT_CONTIG;
2736 
2737 	error = 0;
2738 	/*
2739 	 * Select which case we're in here, and implement it.
2740 	 */
2741 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2742 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2743 		/*
2744 		 * New allocation is contiguous with real allocations on the
2745 		 * left and on the right.
2746 		 * Merge all three into a single extent record.
2747 		 */
2748 		left.br_blockcount += new->br_blockcount + right.br_blockcount;
2749 
2750 		xfs_iext_remove(ip, icur, state);
2751 		xfs_iext_prev(ifp, icur);
2752 		xfs_iext_update_extent(ip, state, icur, &left);
2753 
2754 		XFS_IFORK_NEXT_SET(ip, whichfork,
2755 			XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2756 		if (cur == NULL) {
2757 			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2758 		} else {
2759 			rval = XFS_ILOG_CORE;
2760 			error = xfs_bmbt_lookup_eq(cur, &right, &i);
2761 			if (error)
2762 				goto done;
2763 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2764 			error = xfs_btree_delete(cur, &i);
2765 			if (error)
2766 				goto done;
2767 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2768 			error = xfs_btree_decrement(cur, 0, &i);
2769 			if (error)
2770 				goto done;
2771 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2772 			error = xfs_bmbt_update(cur, &left);
2773 			if (error)
2774 				goto done;
2775 		}
2776 		break;
2777 
2778 	case BMAP_LEFT_CONTIG:
2779 		/*
2780 		 * New allocation is contiguous with a real allocation
2781 		 * on the left.
2782 		 * Merge the new allocation with the left neighbor.
2783 		 */
2784 		old = left;
2785 		left.br_blockcount += new->br_blockcount;
2786 
2787 		xfs_iext_prev(ifp, icur);
2788 		xfs_iext_update_extent(ip, state, icur, &left);
2789 
2790 		if (cur == NULL) {
2791 			rval = xfs_ilog_fext(whichfork);
2792 		} else {
2793 			rval = 0;
2794 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2795 			if (error)
2796 				goto done;
2797 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2798 			error = xfs_bmbt_update(cur, &left);
2799 			if (error)
2800 				goto done;
2801 		}
2802 		break;
2803 
2804 	case BMAP_RIGHT_CONTIG:
2805 		/*
2806 		 * New allocation is contiguous with a real allocation
2807 		 * on the right.
2808 		 * Merge the new allocation with the right neighbor.
2809 		 */
2810 		old = right;
2811 
2812 		right.br_startoff = new->br_startoff;
2813 		right.br_startblock = new->br_startblock;
2814 		right.br_blockcount += new->br_blockcount;
2815 		xfs_iext_update_extent(ip, state, icur, &right);
2816 
2817 		if (cur == NULL) {
2818 			rval = xfs_ilog_fext(whichfork);
2819 		} else {
2820 			rval = 0;
2821 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2822 			if (error)
2823 				goto done;
2824 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2825 			error = xfs_bmbt_update(cur, &right);
2826 			if (error)
2827 				goto done;
2828 		}
2829 		break;
2830 
2831 	case 0:
2832 		/*
2833 		 * New allocation is not contiguous with another
2834 		 * real allocation.
2835 		 * Insert a new entry.
2836 		 */
2837 		xfs_iext_insert(ip, icur, new, state);
2838 		XFS_IFORK_NEXT_SET(ip, whichfork,
2839 			XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2840 		if (cur == NULL) {
2841 			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2842 		} else {
2843 			rval = XFS_ILOG_CORE;
2844 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2845 			if (error)
2846 				goto done;
2847 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2848 			error = xfs_btree_insert(cur, &i);
2849 			if (error)
2850 				goto done;
2851 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2852 		}
2853 		break;
2854 	}
2855 
2856 	/* add reverse mapping unless caller opted out */
2857 	if (!(flags & XFS_BMAPI_NORMAP)) {
2858 		error = xfs_rmap_map_extent(mp, dfops, ip, whichfork, new);
2859 		if (error)
2860 			goto done;
2861 	}
2862 
2863 	/* convert to a btree if necessary */
2864 	if (xfs_bmap_needs_btree(ip, whichfork)) {
2865 		int	tmp_logflags;	/* partial log flag return val */
2866 
2867 		ASSERT(cur == NULL);
2868 		error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, curp,
2869 				0, &tmp_logflags, whichfork);
2870 		*logflagsp |= tmp_logflags;
2871 		cur = *curp;
2872 		if (error)
2873 			goto done;
2874 	}
2875 
2876 	/* clear out the allocated field, done with it now in any case. */
2877 	if (cur)
2878 		cur->bc_private.b.allocated = 0;
2879 
2880 	xfs_bmap_check_leaf_extents(cur, ip, whichfork);
2881 done:
2882 	*logflagsp |= rval;
2883 	return error;
2884 }
2885 
2886 /*
2887  * Functions used in the extent read, allocate and remove paths
2888  */
2889 
2890 /*
2891  * Adjust the size of the new extent based on di_extsize and rt extsize.
2892  */
2893 int
2894 xfs_bmap_extsize_align(
2895 	xfs_mount_t	*mp,
2896 	xfs_bmbt_irec_t	*gotp,		/* next extent pointer */
2897 	xfs_bmbt_irec_t	*prevp,		/* previous extent pointer */
2898 	xfs_extlen_t	extsz,		/* align to this extent size */
2899 	int		rt,		/* is this a realtime inode? */
2900 	int		eof,		/* is extent at end-of-file? */
2901 	int		delay,		/* creating delalloc extent? */
2902 	int		convert,	/* overwriting unwritten extent? */
2903 	xfs_fileoff_t	*offp,		/* in/out: aligned offset */
2904 	xfs_extlen_t	*lenp)		/* in/out: aligned length */
2905 {
2906 	xfs_fileoff_t	orig_off;	/* original offset */
2907 	xfs_extlen_t	orig_alen;	/* original length */
2908 	xfs_fileoff_t	orig_end;	/* original off+len */
2909 	xfs_fileoff_t	nexto;		/* next file offset */
2910 	xfs_fileoff_t	prevo;		/* previous file offset */
2911 	xfs_fileoff_t	align_off;	/* temp for offset */
2912 	xfs_extlen_t	align_alen;	/* temp for length */
2913 	xfs_extlen_t	temp;		/* temp for calculations */
2914 
2915 	if (convert)
2916 		return 0;
2917 
2918 	orig_off = align_off = *offp;
2919 	orig_alen = align_alen = *lenp;
2920 	orig_end = orig_off + orig_alen;
2921 
2922 	/*
2923 	 * If this request overlaps an existing extent, then don't
2924 	 * attempt to perform any additional alignment.
2925 	 */
2926 	if (!delay && !eof &&
2927 	    (orig_off >= gotp->br_startoff) &&
2928 	    (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2929 		return 0;
2930 	}
2931 
2932 	/*
2933 	 * If the file offset is unaligned vs. the extent size
2934 	 * we need to align it.  This will be possible unless
2935 	 * the file was previously written with a kernel that didn't
2936 	 * perform this alignment, or if a truncate shot us in the
2937 	 * foot.
2938 	 */
2939 	temp = do_mod(orig_off, extsz);
2940 	if (temp) {
2941 		align_alen += temp;
2942 		align_off -= temp;
2943 	}
2944 
2945 	/* Same adjustment for the end of the requested area. */
2946 	temp = (align_alen % extsz);
2947 	if (temp)
2948 		align_alen += extsz - temp;
2949 
2950 	/*
2951 	 * For large extent hint sizes, the aligned extent might be larger than
2952 	 * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
2953 	 * the length back under MAXEXTLEN. The outer allocation loops handle
2954 	 * short allocation just fine, so it is safe to do this. We only want to
2955 	 * do it when we are forced to, though, because it means more allocation
2956 	 * operations are required.
2957 	 */
2958 	while (align_alen > MAXEXTLEN)
2959 		align_alen -= extsz;
2960 	ASSERT(align_alen <= MAXEXTLEN);
2961 
2962 	/*
2963 	 * If the previous block overlaps with this proposed allocation
2964 	 * then move the start forward without adjusting the length.
2965 	 */
2966 	if (prevp->br_startoff != NULLFILEOFF) {
2967 		if (prevp->br_startblock == HOLESTARTBLOCK)
2968 			prevo = prevp->br_startoff;
2969 		else
2970 			prevo = prevp->br_startoff + prevp->br_blockcount;
2971 	} else
2972 		prevo = 0;
2973 	if (align_off != orig_off && align_off < prevo)
2974 		align_off = prevo;
2975 	/*
2976 	 * If the next block overlaps with this proposed allocation
2977 	 * then move the start back without adjusting the length,
2978 	 * but not before offset 0.
2979 	 * This may of course make the start overlap previous block,
2980 	 * and if we hit the offset 0 limit then the next block
2981 	 * can still overlap too.
2982 	 */
2983 	if (!eof && gotp->br_startoff != NULLFILEOFF) {
2984 		if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2985 		    (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2986 			nexto = gotp->br_startoff + gotp->br_blockcount;
2987 		else
2988 			nexto = gotp->br_startoff;
2989 	} else
2990 		nexto = NULLFILEOFF;
2991 	if (!eof &&
2992 	    align_off + align_alen != orig_end &&
2993 	    align_off + align_alen > nexto)
2994 		align_off = nexto > align_alen ? nexto - align_alen : 0;
2995 	/*
2996 	 * If we're now overlapping the next or previous extent that
2997 	 * means we can't fit an extsz piece in this hole.  Just move
2998 	 * the start forward to the first valid spot and set
2999 	 * the length so we hit the end.
3000 	 */
3001 	if (align_off != orig_off && align_off < prevo)
3002 		align_off = prevo;
3003 	if (align_off + align_alen != orig_end &&
3004 	    align_off + align_alen > nexto &&
3005 	    nexto != NULLFILEOFF) {
3006 		ASSERT(nexto > prevo);
3007 		align_alen = nexto - align_off;
3008 	}
3009 
3010 	/*
3011 	 * If realtime, and the result isn't a multiple of the realtime
3012 	 * extent size we need to remove blocks until it is.
3013 	 */
3014 	if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3015 		/*
3016 		 * We're not covering the original request, or
3017 		 * we won't be able to once we fix the length.
3018 		 */
3019 		if (orig_off < align_off ||
3020 		    orig_end > align_off + align_alen ||
3021 		    align_alen - temp < orig_alen)
3022 			return -EINVAL;
3023 		/*
3024 		 * Try to fix it by moving the start up.
3025 		 */
3026 		if (align_off + temp <= orig_off) {
3027 			align_alen -= temp;
3028 			align_off += temp;
3029 		}
3030 		/*
3031 		 * Try to fix it by moving the end in.
3032 		 */
3033 		else if (align_off + align_alen - temp >= orig_end)
3034 			align_alen -= temp;
3035 		/*
3036 		 * Set the start to the minimum then trim the length.
3037 		 */
3038 		else {
3039 			align_alen -= orig_off - align_off;
3040 			align_off = orig_off;
3041 			align_alen -= align_alen % mp->m_sb.sb_rextsize;
3042 		}
3043 		/*
3044 		 * Result doesn't cover the request, fail it.
3045 		 */
3046 		if (orig_off < align_off || orig_end > align_off + align_alen)
3047 			return -EINVAL;
3048 	} else {
3049 		ASSERT(orig_off >= align_off);
3050 		/* see MAXEXTLEN handling above */
3051 		ASSERT(orig_end <= align_off + align_alen ||
3052 		       align_alen + extsz > MAXEXTLEN);
3053 	}
3054 
3055 #ifdef DEBUG
3056 	if (!eof && gotp->br_startoff != NULLFILEOFF)
3057 		ASSERT(align_off + align_alen <= gotp->br_startoff);
3058 	if (prevp->br_startoff != NULLFILEOFF)
3059 		ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3060 #endif
3061 
3062 	*lenp = align_alen;
3063 	*offp = align_off;
3064 	return 0;
3065 }
3066 
3067 #define XFS_ALLOC_GAP_UNITS	4
3068 
3069 void
3070 xfs_bmap_adjacent(
3071 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3072 {
3073 	xfs_fsblock_t	adjust;		/* adjustment to block numbers */
3074 	xfs_agnumber_t	fb_agno;	/* ag number of ap->firstblock */
3075 	xfs_mount_t	*mp;		/* mount point structure */
3076 	int		nullfb;		/* true if ap->firstblock isn't set */
3077 	int		rt;		/* true if inode is realtime */
3078 
3079 #define	ISVALID(x,y)	\
3080 	(rt ? \
3081 		(x) < mp->m_sb.sb_rblocks : \
3082 		XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3083 		XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3084 		XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3085 
3086 	mp = ap->ip->i_mount;
3087 	nullfb = *ap->firstblock == NULLFSBLOCK;
3088 	rt = XFS_IS_REALTIME_INODE(ap->ip) &&
3089 		xfs_alloc_is_userdata(ap->datatype);
3090 	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3091 	/*
3092 	 * If allocating at eof, and there's a previous real block,
3093 	 * try to use its last block as our starting point.
3094 	 */
3095 	if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3096 	    !isnullstartblock(ap->prev.br_startblock) &&
3097 	    ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3098 		    ap->prev.br_startblock)) {
3099 		ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3100 		/*
3101 		 * Adjust for the gap between prevp and us.
3102 		 */
3103 		adjust = ap->offset -
3104 			(ap->prev.br_startoff + ap->prev.br_blockcount);
3105 		if (adjust &&
3106 		    ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3107 			ap->blkno += adjust;
3108 	}
3109 	/*
3110 	 * If not at eof, then compare the two neighbor blocks.
3111 	 * Figure out whether either one gives us a good starting point,
3112 	 * and pick the better one.
3113 	 */
3114 	else if (!ap->eof) {
3115 		xfs_fsblock_t	gotbno;		/* right side block number */
3116 		xfs_fsblock_t	gotdiff=0;	/* right side difference */
3117 		xfs_fsblock_t	prevbno;	/* left side block number */
3118 		xfs_fsblock_t	prevdiff=0;	/* left side difference */
3119 
3120 		/*
3121 		 * If there's a previous (left) block, select a requested
3122 		 * start block based on it.
3123 		 */
3124 		if (ap->prev.br_startoff != NULLFILEOFF &&
3125 		    !isnullstartblock(ap->prev.br_startblock) &&
3126 		    (prevbno = ap->prev.br_startblock +
3127 			       ap->prev.br_blockcount) &&
3128 		    ISVALID(prevbno, ap->prev.br_startblock)) {
3129 			/*
3130 			 * Calculate gap to end of previous block.
3131 			 */
3132 			adjust = prevdiff = ap->offset -
3133 				(ap->prev.br_startoff +
3134 				 ap->prev.br_blockcount);
3135 			/*
3136 			 * Figure the startblock based on the previous block's
3137 			 * end and the gap size.
3138 			 * Heuristic!
3139 			 * If the gap is large relative to the piece we're
3140 			 * allocating, or using it gives us an invalid block
3141 			 * number, then just use the end of the previous block.
3142 			 */
3143 			if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3144 			    ISVALID(prevbno + prevdiff,
3145 				    ap->prev.br_startblock))
3146 				prevbno += adjust;
3147 			else
3148 				prevdiff += adjust;
3149 			/*
3150 			 * If the firstblock forbids it, can't use it,
3151 			 * must use default.
3152 			 */
3153 			if (!rt && !nullfb &&
3154 			    XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3155 				prevbno = NULLFSBLOCK;
3156 		}
3157 		/*
3158 		 * No previous block or can't follow it, just default.
3159 		 */
3160 		else
3161 			prevbno = NULLFSBLOCK;
3162 		/*
3163 		 * If there's a following (right) block, select a requested
3164 		 * start block based on it.
3165 		 */
3166 		if (!isnullstartblock(ap->got.br_startblock)) {
3167 			/*
3168 			 * Calculate gap to start of next block.
3169 			 */
3170 			adjust = gotdiff = ap->got.br_startoff - ap->offset;
3171 			/*
3172 			 * Figure the startblock based on the next block's
3173 			 * start and the gap size.
3174 			 */
3175 			gotbno = ap->got.br_startblock;
3176 			/*
3177 			 * Heuristic!
3178 			 * If the gap is large relative to the piece we're
3179 			 * allocating, or using it gives us an invalid block
3180 			 * number, then just use the start of the next block
3181 			 * offset by our length.
3182 			 */
3183 			if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3184 			    ISVALID(gotbno - gotdiff, gotbno))
3185 				gotbno -= adjust;
3186 			else if (ISVALID(gotbno - ap->length, gotbno)) {
3187 				gotbno -= ap->length;
3188 				gotdiff += adjust - ap->length;
3189 			} else
3190 				gotdiff += adjust;
3191 			/*
3192 			 * If the firstblock forbids it, can't use it,
3193 			 * must use default.
3194 			 */
3195 			if (!rt && !nullfb &&
3196 			    XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3197 				gotbno = NULLFSBLOCK;
3198 		}
3199 		/*
3200 		 * No next block, just default.
3201 		 */
3202 		else
3203 			gotbno = NULLFSBLOCK;
3204 		/*
3205 		 * If both valid, pick the better one, else the only good
3206 		 * one, else ap->blkno is already set (to 0 or the inode block).
3207 		 */
3208 		if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3209 			ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3210 		else if (prevbno != NULLFSBLOCK)
3211 			ap->blkno = prevbno;
3212 		else if (gotbno != NULLFSBLOCK)
3213 			ap->blkno = gotbno;
3214 	}
3215 #undef ISVALID
3216 }
3217 
3218 static int
3219 xfs_bmap_longest_free_extent(
3220 	struct xfs_trans	*tp,
3221 	xfs_agnumber_t		ag,
3222 	xfs_extlen_t		*blen,
3223 	int			*notinit)
3224 {
3225 	struct xfs_mount	*mp = tp->t_mountp;
3226 	struct xfs_perag	*pag;
3227 	xfs_extlen_t		longest;
3228 	int			error = 0;
3229 
3230 	pag = xfs_perag_get(mp, ag);
3231 	if (!pag->pagf_init) {
3232 		error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3233 		if (error)
3234 			goto out;
3235 
3236 		if (!pag->pagf_init) {
3237 			*notinit = 1;
3238 			goto out;
3239 		}
3240 	}
3241 
3242 	longest = xfs_alloc_longest_free_extent(pag,
3243 				xfs_alloc_min_freelist(mp, pag),
3244 				xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
3245 	if (*blen < longest)
3246 		*blen = longest;
3247 
3248 out:
3249 	xfs_perag_put(pag);
3250 	return error;
3251 }
3252 
3253 static void
3254 xfs_bmap_select_minlen(
3255 	struct xfs_bmalloca	*ap,
3256 	struct xfs_alloc_arg	*args,
3257 	xfs_extlen_t		*blen,
3258 	int			notinit)
3259 {
3260 	if (notinit || *blen < ap->minlen) {
3261 		/*
3262 		 * Since we did a BUF_TRYLOCK above, it is possible that
3263 		 * there is space for this request.
3264 		 */
3265 		args->minlen = ap->minlen;
3266 	} else if (*blen < args->maxlen) {
3267 		/*
3268 		 * If the best seen length is less than the request length,
3269 		 * use the best as the minimum.
3270 		 */
3271 		args->minlen = *blen;
3272 	} else {
3273 		/*
3274 		 * Otherwise we've seen an extent as big as maxlen, use that
3275 		 * as the minimum.
3276 		 */
3277 		args->minlen = args->maxlen;
3278 	}
3279 }
3280 
3281 STATIC int
3282 xfs_bmap_btalloc_nullfb(
3283 	struct xfs_bmalloca	*ap,
3284 	struct xfs_alloc_arg	*args,
3285 	xfs_extlen_t		*blen)
3286 {
3287 	struct xfs_mount	*mp = ap->ip->i_mount;
3288 	xfs_agnumber_t		ag, startag;
3289 	int			notinit = 0;
3290 	int			error;
3291 
3292 	args->type = XFS_ALLOCTYPE_START_BNO;
3293 	args->total = ap->total;
3294 
3295 	startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3296 	if (startag == NULLAGNUMBER)
3297 		startag = ag = 0;
3298 
3299 	while (*blen < args->maxlen) {
3300 		error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3301 						     &notinit);
3302 		if (error)
3303 			return error;
3304 
3305 		if (++ag == mp->m_sb.sb_agcount)
3306 			ag = 0;
3307 		if (ag == startag)
3308 			break;
3309 	}
3310 
3311 	xfs_bmap_select_minlen(ap, args, blen, notinit);
3312 	return 0;
3313 }
3314 
3315 STATIC int
3316 xfs_bmap_btalloc_filestreams(
3317 	struct xfs_bmalloca	*ap,
3318 	struct xfs_alloc_arg	*args,
3319 	xfs_extlen_t		*blen)
3320 {
3321 	struct xfs_mount	*mp = ap->ip->i_mount;
3322 	xfs_agnumber_t		ag;
3323 	int			notinit = 0;
3324 	int			error;
3325 
3326 	args->type = XFS_ALLOCTYPE_NEAR_BNO;
3327 	args->total = ap->total;
3328 
3329 	ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3330 	if (ag == NULLAGNUMBER)
3331 		ag = 0;
3332 
3333 	error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3334 	if (error)
3335 		return error;
3336 
3337 	if (*blen < args->maxlen) {
3338 		error = xfs_filestream_new_ag(ap, &ag);
3339 		if (error)
3340 			return error;
3341 
3342 		error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3343 						     &notinit);
3344 		if (error)
3345 			return error;
3346 
3347 	}
3348 
3349 	xfs_bmap_select_minlen(ap, args, blen, notinit);
3350 
3351 	/*
3352 	 * Set the failure fallback case to look in the selected AG as stream
3353 	 * may have moved.
3354 	 */
3355 	ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3356 	return 0;
3357 }
3358 
3359 /* Update all inode and quota accounting for the allocation we just did. */
3360 static void
3361 xfs_bmap_btalloc_accounting(
3362 	struct xfs_bmalloca	*ap,
3363 	struct xfs_alloc_arg	*args)
3364 {
3365 	if (ap->flags & XFS_BMAPI_COWFORK) {
3366 		/*
3367 		 * COW fork blocks are in-core only and thus are treated as
3368 		 * in-core quota reservation (like delalloc blocks) even when
3369 		 * converted to real blocks. The quota reservation is not
3370 		 * accounted to disk until blocks are remapped to the data
3371 		 * fork. So if these blocks were previously delalloc, we
3372 		 * already have quota reservation and there's nothing to do
3373 		 * yet.
3374 		 */
3375 		if (ap->wasdel)
3376 			return;
3377 
3378 		/*
3379 		 * Otherwise, we've allocated blocks in a hole. The transaction
3380 		 * has acquired in-core quota reservation for this extent.
3381 		 * Rather than account these as real blocks, however, we reduce
3382 		 * the transaction quota reservation based on the allocation.
3383 		 * This essentially transfers the transaction quota reservation
3384 		 * to that of a delalloc extent.
3385 		 */
3386 		ap->ip->i_delayed_blks += args->len;
3387 		xfs_trans_mod_dquot_byino(ap->tp, ap->ip, XFS_TRANS_DQ_RES_BLKS,
3388 				-(long)args->len);
3389 		return;
3390 	}
3391 
3392 	/* data/attr fork only */
3393 	ap->ip->i_d.di_nblocks += args->len;
3394 	xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3395 	if (ap->wasdel)
3396 		ap->ip->i_delayed_blks -= args->len;
3397 	xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3398 		ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT : XFS_TRANS_DQ_BCOUNT,
3399 		args->len);
3400 }
3401 
3402 STATIC int
3403 xfs_bmap_btalloc(
3404 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3405 {
3406 	xfs_mount_t	*mp;		/* mount point structure */
3407 	xfs_alloctype_t	atype = 0;	/* type for allocation routines */
3408 	xfs_extlen_t	align = 0;	/* minimum allocation alignment */
3409 	xfs_agnumber_t	fb_agno;	/* ag number of ap->firstblock */
3410 	xfs_agnumber_t	ag;
3411 	xfs_alloc_arg_t	args;
3412 	xfs_fileoff_t	orig_offset;
3413 	xfs_extlen_t	orig_length;
3414 	xfs_extlen_t	blen;
3415 	xfs_extlen_t	nextminlen = 0;
3416 	int		nullfb;		/* true if ap->firstblock isn't set */
3417 	int		isaligned;
3418 	int		tryagain;
3419 	int		error;
3420 	int		stripe_align;
3421 
3422 	ASSERT(ap->length);
3423 	orig_offset = ap->offset;
3424 	orig_length = ap->length;
3425 
3426 	mp = ap->ip->i_mount;
3427 
3428 	/* stripe alignment for allocation is determined by mount parameters */
3429 	stripe_align = 0;
3430 	if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3431 		stripe_align = mp->m_swidth;
3432 	else if (mp->m_dalign)
3433 		stripe_align = mp->m_dalign;
3434 
3435 	if (ap->flags & XFS_BMAPI_COWFORK)
3436 		align = xfs_get_cowextsz_hint(ap->ip);
3437 	else if (xfs_alloc_is_userdata(ap->datatype))
3438 		align = xfs_get_extsz_hint(ap->ip);
3439 	if (align) {
3440 		error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3441 						align, 0, ap->eof, 0, ap->conv,
3442 						&ap->offset, &ap->length);
3443 		ASSERT(!error);
3444 		ASSERT(ap->length);
3445 	}
3446 
3447 
3448 	nullfb = *ap->firstblock == NULLFSBLOCK;
3449 	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3450 	if (nullfb) {
3451 		if (xfs_alloc_is_userdata(ap->datatype) &&
3452 		    xfs_inode_is_filestream(ap->ip)) {
3453 			ag = xfs_filestream_lookup_ag(ap->ip);
3454 			ag = (ag != NULLAGNUMBER) ? ag : 0;
3455 			ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3456 		} else {
3457 			ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3458 		}
3459 	} else
3460 		ap->blkno = *ap->firstblock;
3461 
3462 	xfs_bmap_adjacent(ap);
3463 
3464 	/*
3465 	 * If allowed, use ap->blkno; otherwise must use firstblock since
3466 	 * it's in the right allocation group.
3467 	 */
3468 	if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3469 		;
3470 	else
3471 		ap->blkno = *ap->firstblock;
3472 	/*
3473 	 * Normal allocation, done through xfs_alloc_vextent.
3474 	 */
3475 	tryagain = isaligned = 0;
3476 	memset(&args, 0, sizeof(args));
3477 	args.tp = ap->tp;
3478 	args.mp = mp;
3479 	args.fsbno = ap->blkno;
3480 	xfs_rmap_skip_owner_update(&args.oinfo);
3481 
3482 	/* Trim the allocation back to the maximum an AG can fit. */
3483 	args.maxlen = MIN(ap->length, mp->m_ag_max_usable);
3484 	args.firstblock = *ap->firstblock;
3485 	blen = 0;
3486 	if (nullfb) {
3487 		/*
3488 		 * Search for an allocation group with a single extent large
3489 		 * enough for the request.  If one isn't found, then adjust
3490 		 * the minimum allocation size to the largest space found.
3491 		 */
3492 		if (xfs_alloc_is_userdata(ap->datatype) &&
3493 		    xfs_inode_is_filestream(ap->ip))
3494 			error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3495 		else
3496 			error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3497 		if (error)
3498 			return error;
3499 	} else if (ap->dfops->dop_low) {
3500 		if (xfs_inode_is_filestream(ap->ip))
3501 			args.type = XFS_ALLOCTYPE_FIRST_AG;
3502 		else
3503 			args.type = XFS_ALLOCTYPE_START_BNO;
3504 		args.total = args.minlen = ap->minlen;
3505 	} else {
3506 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
3507 		args.total = ap->total;
3508 		args.minlen = ap->minlen;
3509 	}
3510 	/* apply extent size hints if obtained earlier */
3511 	if (align) {
3512 		args.prod = align;
3513 		if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3514 			args.mod = (xfs_extlen_t)(args.prod - args.mod);
3515 	} else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3516 		args.prod = 1;
3517 		args.mod = 0;
3518 	} else {
3519 		args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3520 		if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3521 			args.mod = (xfs_extlen_t)(args.prod - args.mod);
3522 	}
3523 	/*
3524 	 * If we are not low on available data blocks, and the
3525 	 * underlying logical volume manager is a stripe, and
3526 	 * the file offset is zero then try to allocate data
3527 	 * blocks on stripe unit boundary.
3528 	 * NOTE: ap->aeof is only set if the allocation length
3529 	 * is >= the stripe unit and the allocation offset is
3530 	 * at the end of file.
3531 	 */
3532 	if (!ap->dfops->dop_low && ap->aeof) {
3533 		if (!ap->offset) {
3534 			args.alignment = stripe_align;
3535 			atype = args.type;
3536 			isaligned = 1;
3537 			/*
3538 			 * Adjust for alignment
3539 			 */
3540 			if (blen > args.alignment && blen <= args.maxlen)
3541 				args.minlen = blen - args.alignment;
3542 			args.minalignslop = 0;
3543 		} else {
3544 			/*
3545 			 * First try an exact bno allocation.
3546 			 * If it fails then do a near or start bno
3547 			 * allocation with alignment turned on.
3548 			 */
3549 			atype = args.type;
3550 			tryagain = 1;
3551 			args.type = XFS_ALLOCTYPE_THIS_BNO;
3552 			args.alignment = 1;
3553 			/*
3554 			 * Compute the minlen+alignment for the
3555 			 * next case.  Set slop so that the value
3556 			 * of minlen+alignment+slop doesn't go up
3557 			 * between the calls.
3558 			 */
3559 			if (blen > stripe_align && blen <= args.maxlen)
3560 				nextminlen = blen - stripe_align;
3561 			else
3562 				nextminlen = args.minlen;
3563 			if (nextminlen + stripe_align > args.minlen + 1)
3564 				args.minalignslop =
3565 					nextminlen + stripe_align -
3566 					args.minlen - 1;
3567 			else
3568 				args.minalignslop = 0;
3569 		}
3570 	} else {
3571 		args.alignment = 1;
3572 		args.minalignslop = 0;
3573 	}
3574 	args.minleft = ap->minleft;
3575 	args.wasdel = ap->wasdel;
3576 	args.resv = XFS_AG_RESV_NONE;
3577 	args.datatype = ap->datatype;
3578 	if (ap->datatype & XFS_ALLOC_USERDATA_ZERO)
3579 		args.ip = ap->ip;
3580 
3581 	error = xfs_alloc_vextent(&args);
3582 	if (error)
3583 		return error;
3584 
3585 	if (tryagain && args.fsbno == NULLFSBLOCK) {
3586 		/*
3587 		 * Exact allocation failed. Now try with alignment
3588 		 * turned on.
3589 		 */
3590 		args.type = atype;
3591 		args.fsbno = ap->blkno;
3592 		args.alignment = stripe_align;
3593 		args.minlen = nextminlen;
3594 		args.minalignslop = 0;
3595 		isaligned = 1;
3596 		if ((error = xfs_alloc_vextent(&args)))
3597 			return error;
3598 	}
3599 	if (isaligned && args.fsbno == NULLFSBLOCK) {
3600 		/*
3601 		 * allocation failed, so turn off alignment and
3602 		 * try again.
3603 		 */
3604 		args.type = atype;
3605 		args.fsbno = ap->blkno;
3606 		args.alignment = 0;
3607 		if ((error = xfs_alloc_vextent(&args)))
3608 			return error;
3609 	}
3610 	if (args.fsbno == NULLFSBLOCK && nullfb &&
3611 	    args.minlen > ap->minlen) {
3612 		args.minlen = ap->minlen;
3613 		args.type = XFS_ALLOCTYPE_START_BNO;
3614 		args.fsbno = ap->blkno;
3615 		if ((error = xfs_alloc_vextent(&args)))
3616 			return error;
3617 	}
3618 	if (args.fsbno == NULLFSBLOCK && nullfb) {
3619 		args.fsbno = 0;
3620 		args.type = XFS_ALLOCTYPE_FIRST_AG;
3621 		args.total = ap->minlen;
3622 		if ((error = xfs_alloc_vextent(&args)))
3623 			return error;
3624 		ap->dfops->dop_low = true;
3625 	}
3626 	if (args.fsbno != NULLFSBLOCK) {
3627 		/*
3628 		 * check the allocation happened at the same or higher AG than
3629 		 * the first block that was allocated.
3630 		 */
3631 		ASSERT(*ap->firstblock == NULLFSBLOCK ||
3632 		       XFS_FSB_TO_AGNO(mp, *ap->firstblock) <=
3633 		       XFS_FSB_TO_AGNO(mp, args.fsbno));
3634 
3635 		ap->blkno = args.fsbno;
3636 		if (*ap->firstblock == NULLFSBLOCK)
3637 			*ap->firstblock = args.fsbno;
3638 		ASSERT(nullfb || fb_agno <= args.agno);
3639 		ap->length = args.len;
3640 		/*
3641 		 * If the extent size hint is active, we tried to round the
3642 		 * caller's allocation request offset down to extsz and the
3643 		 * length up to another extsz boundary.  If we found a free
3644 		 * extent we mapped it in starting at this new offset.  If the
3645 		 * newly mapped space isn't long enough to cover any of the
3646 		 * range of offsets that was originally requested, move the
3647 		 * mapping up so that we can fill as much of the caller's
3648 		 * original request as possible.  Free space is apparently
3649 		 * very fragmented so we're unlikely to be able to satisfy the
3650 		 * hints anyway.
3651 		 */
3652 		if (ap->length <= orig_length)
3653 			ap->offset = orig_offset;
3654 		else if (ap->offset + ap->length < orig_offset + orig_length)
3655 			ap->offset = orig_offset + orig_length - ap->length;
3656 		xfs_bmap_btalloc_accounting(ap, &args);
3657 	} else {
3658 		ap->blkno = NULLFSBLOCK;
3659 		ap->length = 0;
3660 	}
3661 	return 0;
3662 }
3663 
3664 /*
3665  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3666  * It figures out where to ask the underlying allocator to put the new extent.
3667  */
3668 STATIC int
3669 xfs_bmap_alloc(
3670 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3671 {
3672 	if (XFS_IS_REALTIME_INODE(ap->ip) &&
3673 	    xfs_alloc_is_userdata(ap->datatype))
3674 		return xfs_bmap_rtalloc(ap);
3675 	return xfs_bmap_btalloc(ap);
3676 }
3677 
3678 /* Trim extent to fit a logical block range. */
3679 void
3680 xfs_trim_extent(
3681 	struct xfs_bmbt_irec	*irec,
3682 	xfs_fileoff_t		bno,
3683 	xfs_filblks_t		len)
3684 {
3685 	xfs_fileoff_t		distance;
3686 	xfs_fileoff_t		end = bno + len;
3687 
3688 	if (irec->br_startoff + irec->br_blockcount <= bno ||
3689 	    irec->br_startoff >= end) {
3690 		irec->br_blockcount = 0;
3691 		return;
3692 	}
3693 
3694 	if (irec->br_startoff < bno) {
3695 		distance = bno - irec->br_startoff;
3696 		if (isnullstartblock(irec->br_startblock))
3697 			irec->br_startblock = DELAYSTARTBLOCK;
3698 		if (irec->br_startblock != DELAYSTARTBLOCK &&
3699 		    irec->br_startblock != HOLESTARTBLOCK)
3700 			irec->br_startblock += distance;
3701 		irec->br_startoff += distance;
3702 		irec->br_blockcount -= distance;
3703 	}
3704 
3705 	if (end < irec->br_startoff + irec->br_blockcount) {
3706 		distance = irec->br_startoff + irec->br_blockcount - end;
3707 		irec->br_blockcount -= distance;
3708 	}
3709 }
3710 
3711 /* trim extent to within eof */
3712 void
3713 xfs_trim_extent_eof(
3714 	struct xfs_bmbt_irec	*irec,
3715 	struct xfs_inode	*ip)
3716 
3717 {
3718 	xfs_trim_extent(irec, 0, XFS_B_TO_FSB(ip->i_mount,
3719 					      i_size_read(VFS_I(ip))));
3720 }
3721 
3722 /*
3723  * Trim the returned map to the required bounds
3724  */
3725 STATIC void
3726 xfs_bmapi_trim_map(
3727 	struct xfs_bmbt_irec	*mval,
3728 	struct xfs_bmbt_irec	*got,
3729 	xfs_fileoff_t		*bno,
3730 	xfs_filblks_t		len,
3731 	xfs_fileoff_t		obno,
3732 	xfs_fileoff_t		end,
3733 	int			n,
3734 	int			flags)
3735 {
3736 	if ((flags & XFS_BMAPI_ENTIRE) ||
3737 	    got->br_startoff + got->br_blockcount <= obno) {
3738 		*mval = *got;
3739 		if (isnullstartblock(got->br_startblock))
3740 			mval->br_startblock = DELAYSTARTBLOCK;
3741 		return;
3742 	}
3743 
3744 	if (obno > *bno)
3745 		*bno = obno;
3746 	ASSERT((*bno >= obno) || (n == 0));
3747 	ASSERT(*bno < end);
3748 	mval->br_startoff = *bno;
3749 	if (isnullstartblock(got->br_startblock))
3750 		mval->br_startblock = DELAYSTARTBLOCK;
3751 	else
3752 		mval->br_startblock = got->br_startblock +
3753 					(*bno - got->br_startoff);
3754 	/*
3755 	 * Return the minimum of what we got and what we asked for for
3756 	 * the length.  We can use the len variable here because it is
3757 	 * modified below and we could have been there before coming
3758 	 * here if the first part of the allocation didn't overlap what
3759 	 * was asked for.
3760 	 */
3761 	mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3762 			got->br_blockcount - (*bno - got->br_startoff));
3763 	mval->br_state = got->br_state;
3764 	ASSERT(mval->br_blockcount <= len);
3765 	return;
3766 }
3767 
3768 /*
3769  * Update and validate the extent map to return
3770  */
3771 STATIC void
3772 xfs_bmapi_update_map(
3773 	struct xfs_bmbt_irec	**map,
3774 	xfs_fileoff_t		*bno,
3775 	xfs_filblks_t		*len,
3776 	xfs_fileoff_t		obno,
3777 	xfs_fileoff_t		end,
3778 	int			*n,
3779 	int			flags)
3780 {
3781 	xfs_bmbt_irec_t	*mval = *map;
3782 
3783 	ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3784 	       ((mval->br_startoff + mval->br_blockcount) <= end));
3785 	ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3786 	       (mval->br_startoff < obno));
3787 
3788 	*bno = mval->br_startoff + mval->br_blockcount;
3789 	*len = end - *bno;
3790 	if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3791 		/* update previous map with new information */
3792 		ASSERT(mval->br_startblock == mval[-1].br_startblock);
3793 		ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3794 		ASSERT(mval->br_state == mval[-1].br_state);
3795 		mval[-1].br_blockcount = mval->br_blockcount;
3796 		mval[-1].br_state = mval->br_state;
3797 	} else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3798 		   mval[-1].br_startblock != DELAYSTARTBLOCK &&
3799 		   mval[-1].br_startblock != HOLESTARTBLOCK &&
3800 		   mval->br_startblock == mval[-1].br_startblock +
3801 					  mval[-1].br_blockcount &&
3802 		   ((flags & XFS_BMAPI_IGSTATE) ||
3803 			mval[-1].br_state == mval->br_state)) {
3804 		ASSERT(mval->br_startoff ==
3805 		       mval[-1].br_startoff + mval[-1].br_blockcount);
3806 		mval[-1].br_blockcount += mval->br_blockcount;
3807 	} else if (*n > 0 &&
3808 		   mval->br_startblock == DELAYSTARTBLOCK &&
3809 		   mval[-1].br_startblock == DELAYSTARTBLOCK &&
3810 		   mval->br_startoff ==
3811 		   mval[-1].br_startoff + mval[-1].br_blockcount) {
3812 		mval[-1].br_blockcount += mval->br_blockcount;
3813 		mval[-1].br_state = mval->br_state;
3814 	} else if (!((*n == 0) &&
3815 		     ((mval->br_startoff + mval->br_blockcount) <=
3816 		      obno))) {
3817 		mval++;
3818 		(*n)++;
3819 	}
3820 	*map = mval;
3821 }
3822 
3823 /*
3824  * Map file blocks to filesystem blocks without allocation.
3825  */
3826 int
3827 xfs_bmapi_read(
3828 	struct xfs_inode	*ip,
3829 	xfs_fileoff_t		bno,
3830 	xfs_filblks_t		len,
3831 	struct xfs_bmbt_irec	*mval,
3832 	int			*nmap,
3833 	int			flags)
3834 {
3835 	struct xfs_mount	*mp = ip->i_mount;
3836 	struct xfs_ifork	*ifp;
3837 	struct xfs_bmbt_irec	got;
3838 	xfs_fileoff_t		obno;
3839 	xfs_fileoff_t		end;
3840 	struct xfs_iext_cursor	icur;
3841 	int			error;
3842 	bool			eof = false;
3843 	int			n = 0;
3844 	int			whichfork = xfs_bmapi_whichfork(flags);
3845 
3846 	ASSERT(*nmap >= 1);
3847 	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
3848 			   XFS_BMAPI_IGSTATE|XFS_BMAPI_COWFORK)));
3849 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
3850 
3851 	if (unlikely(XFS_TEST_ERROR(
3852 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3853 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
3854 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
3855 		XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
3856 		return -EFSCORRUPTED;
3857 	}
3858 
3859 	if (XFS_FORCED_SHUTDOWN(mp))
3860 		return -EIO;
3861 
3862 	XFS_STATS_INC(mp, xs_blk_mapr);
3863 
3864 	ifp = XFS_IFORK_PTR(ip, whichfork);
3865 
3866 	/* No CoW fork?  Return a hole. */
3867 	if (whichfork == XFS_COW_FORK && !ifp) {
3868 		mval->br_startoff = bno;
3869 		mval->br_startblock = HOLESTARTBLOCK;
3870 		mval->br_blockcount = len;
3871 		mval->br_state = XFS_EXT_NORM;
3872 		*nmap = 1;
3873 		return 0;
3874 	}
3875 
3876 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
3877 		error = xfs_iread_extents(NULL, ip, whichfork);
3878 		if (error)
3879 			return error;
3880 	}
3881 
3882 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got))
3883 		eof = true;
3884 	end = bno + len;
3885 	obno = bno;
3886 
3887 	while (bno < end && n < *nmap) {
3888 		/* Reading past eof, act as though there's a hole up to end. */
3889 		if (eof)
3890 			got.br_startoff = end;
3891 		if (got.br_startoff > bno) {
3892 			/* Reading in a hole.  */
3893 			mval->br_startoff = bno;
3894 			mval->br_startblock = HOLESTARTBLOCK;
3895 			mval->br_blockcount =
3896 				XFS_FILBLKS_MIN(len, got.br_startoff - bno);
3897 			mval->br_state = XFS_EXT_NORM;
3898 			bno += mval->br_blockcount;
3899 			len -= mval->br_blockcount;
3900 			mval++;
3901 			n++;
3902 			continue;
3903 		}
3904 
3905 		/* set up the extent map to return. */
3906 		xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
3907 		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
3908 
3909 		/* If we're done, stop now. */
3910 		if (bno >= end || n >= *nmap)
3911 			break;
3912 
3913 		/* Else go on to the next record. */
3914 		if (!xfs_iext_next_extent(ifp, &icur, &got))
3915 			eof = true;
3916 	}
3917 	*nmap = n;
3918 	return 0;
3919 }
3920 
3921 /*
3922  * Add a delayed allocation extent to an inode. Blocks are reserved from the
3923  * global pool and the extent inserted into the inode in-core extent tree.
3924  *
3925  * On entry, got refers to the first extent beyond the offset of the extent to
3926  * allocate or eof is specified if no such extent exists. On return, got refers
3927  * to the extent record that was inserted to the inode fork.
3928  *
3929  * Note that the allocated extent may have been merged with contiguous extents
3930  * during insertion into the inode fork. Thus, got does not reflect the current
3931  * state of the inode fork on return. If necessary, the caller can use lastx to
3932  * look up the updated record in the inode fork.
3933  */
3934 int
3935 xfs_bmapi_reserve_delalloc(
3936 	struct xfs_inode	*ip,
3937 	int			whichfork,
3938 	xfs_fileoff_t		off,
3939 	xfs_filblks_t		len,
3940 	xfs_filblks_t		prealloc,
3941 	struct xfs_bmbt_irec	*got,
3942 	struct xfs_iext_cursor	*icur,
3943 	int			eof)
3944 {
3945 	struct xfs_mount	*mp = ip->i_mount;
3946 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
3947 	xfs_extlen_t		alen;
3948 	xfs_extlen_t		indlen;
3949 	int			error;
3950 	xfs_fileoff_t		aoff = off;
3951 
3952 	/*
3953 	 * Cap the alloc length. Keep track of prealloc so we know whether to
3954 	 * tag the inode before we return.
3955 	 */
3956 	alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN);
3957 	if (!eof)
3958 		alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
3959 	if (prealloc && alen >= len)
3960 		prealloc = alen - len;
3961 
3962 	/* Figure out the extent size, adjust alen */
3963 	if (whichfork == XFS_COW_FORK) {
3964 		struct xfs_bmbt_irec	prev;
3965 		xfs_extlen_t		extsz = xfs_get_cowextsz_hint(ip);
3966 
3967 		if (!xfs_iext_peek_prev_extent(ifp, icur, &prev))
3968 			prev.br_startoff = NULLFILEOFF;
3969 
3970 		error = xfs_bmap_extsize_align(mp, got, &prev, extsz, 0, eof,
3971 					       1, 0, &aoff, &alen);
3972 		ASSERT(!error);
3973 	}
3974 
3975 	/*
3976 	 * Make a transaction-less quota reservation for delayed allocation
3977 	 * blocks.  This number gets adjusted later.  We return if we haven't
3978 	 * allocated blocks already inside this loop.
3979 	 */
3980 	error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
3981 						XFS_QMOPT_RES_REGBLKS);
3982 	if (error)
3983 		return error;
3984 
3985 	/*
3986 	 * Split changing sb for alen and indlen since they could be coming
3987 	 * from different places.
3988 	 */
3989 	indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
3990 	ASSERT(indlen > 0);
3991 
3992 	error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
3993 	if (error)
3994 		goto out_unreserve_quota;
3995 
3996 	error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
3997 	if (error)
3998 		goto out_unreserve_blocks;
3999 
4000 
4001 	ip->i_delayed_blks += alen;
4002 
4003 	got->br_startoff = aoff;
4004 	got->br_startblock = nullstartblock(indlen);
4005 	got->br_blockcount = alen;
4006 	got->br_state = XFS_EXT_NORM;
4007 
4008 	xfs_bmap_add_extent_hole_delay(ip, whichfork, icur, got);
4009 
4010 	/*
4011 	 * Tag the inode if blocks were preallocated. Note that COW fork
4012 	 * preallocation can occur at the start or end of the extent, even when
4013 	 * prealloc == 0, so we must also check the aligned offset and length.
4014 	 */
4015 	if (whichfork == XFS_DATA_FORK && prealloc)
4016 		xfs_inode_set_eofblocks_tag(ip);
4017 	if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len))
4018 		xfs_inode_set_cowblocks_tag(ip);
4019 
4020 	return 0;
4021 
4022 out_unreserve_blocks:
4023 	xfs_mod_fdblocks(mp, alen, false);
4024 out_unreserve_quota:
4025 	if (XFS_IS_QUOTA_ON(mp))
4026 		xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0,
4027 						XFS_QMOPT_RES_REGBLKS);
4028 	return error;
4029 }
4030 
4031 static int
4032 xfs_bmapi_allocate(
4033 	struct xfs_bmalloca	*bma)
4034 {
4035 	struct xfs_mount	*mp = bma->ip->i_mount;
4036 	int			whichfork = xfs_bmapi_whichfork(bma->flags);
4037 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4038 	int			tmp_logflags = 0;
4039 	int			error;
4040 
4041 	ASSERT(bma->length > 0);
4042 
4043 	/*
4044 	 * For the wasdelay case, we could also just allocate the stuff asked
4045 	 * for in this bmap call but that wouldn't be as good.
4046 	 */
4047 	if (bma->wasdel) {
4048 		bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4049 		bma->offset = bma->got.br_startoff;
4050 		xfs_iext_peek_prev_extent(ifp, &bma->icur, &bma->prev);
4051 	} else {
4052 		bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4053 		if (!bma->eof)
4054 			bma->length = XFS_FILBLKS_MIN(bma->length,
4055 					bma->got.br_startoff - bma->offset);
4056 	}
4057 
4058 	/*
4059 	 * Set the data type being allocated. For the data fork, the first data
4060 	 * in the file is treated differently to all other allocations. For the
4061 	 * attribute fork, we only need to ensure the allocated range is not on
4062 	 * the busy list.
4063 	 */
4064 	if (!(bma->flags & XFS_BMAPI_METADATA)) {
4065 		bma->datatype = XFS_ALLOC_NOBUSY;
4066 		if (whichfork == XFS_DATA_FORK) {
4067 			if (bma->offset == 0)
4068 				bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA;
4069 			else
4070 				bma->datatype |= XFS_ALLOC_USERDATA;
4071 		}
4072 		if (bma->flags & XFS_BMAPI_ZERO)
4073 			bma->datatype |= XFS_ALLOC_USERDATA_ZERO;
4074 	}
4075 
4076 	bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4077 
4078 	/*
4079 	 * Only want to do the alignment at the eof if it is userdata and
4080 	 * allocation length is larger than a stripe unit.
4081 	 */
4082 	if (mp->m_dalign && bma->length >= mp->m_dalign &&
4083 	    !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4084 		error = xfs_bmap_isaeof(bma, whichfork);
4085 		if (error)
4086 			return error;
4087 	}
4088 
4089 	error = xfs_bmap_alloc(bma);
4090 	if (error)
4091 		return error;
4092 
4093 	if (bma->cur)
4094 		bma->cur->bc_private.b.firstblock = *bma->firstblock;
4095 	if (bma->blkno == NULLFSBLOCK)
4096 		return 0;
4097 	if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4098 		bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4099 		bma->cur->bc_private.b.firstblock = *bma->firstblock;
4100 		bma->cur->bc_private.b.dfops = bma->dfops;
4101 	}
4102 	/*
4103 	 * Bump the number of extents we've allocated
4104 	 * in this call.
4105 	 */
4106 	bma->nallocs++;
4107 
4108 	if (bma->cur)
4109 		bma->cur->bc_private.b.flags =
4110 			bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4111 
4112 	bma->got.br_startoff = bma->offset;
4113 	bma->got.br_startblock = bma->blkno;
4114 	bma->got.br_blockcount = bma->length;
4115 	bma->got.br_state = XFS_EXT_NORM;
4116 
4117 	/*
4118 	 * In the data fork, a wasdelay extent has been initialized, so
4119 	 * shouldn't be flagged as unwritten.
4120 	 *
4121 	 * For the cow fork, however, we convert delalloc reservations
4122 	 * (extents allocated for speculative preallocation) to
4123 	 * allocated unwritten extents, and only convert the unwritten
4124 	 * extents to real extents when we're about to write the data.
4125 	 */
4126 	if ((!bma->wasdel || (bma->flags & XFS_BMAPI_COWFORK)) &&
4127 	    (bma->flags & XFS_BMAPI_PREALLOC) &&
4128 	    xfs_sb_version_hasextflgbit(&mp->m_sb))
4129 		bma->got.br_state = XFS_EXT_UNWRITTEN;
4130 
4131 	if (bma->wasdel)
4132 		error = xfs_bmap_add_extent_delay_real(bma, whichfork);
4133 	else
4134 		error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip,
4135 				whichfork, &bma->icur, &bma->cur, &bma->got,
4136 				bma->firstblock, bma->dfops, &bma->logflags,
4137 				bma->flags);
4138 
4139 	bma->logflags |= tmp_logflags;
4140 	if (error)
4141 		return error;
4142 
4143 	/*
4144 	 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4145 	 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4146 	 * the neighbouring ones.
4147 	 */
4148 	xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4149 
4150 	ASSERT(bma->got.br_startoff <= bma->offset);
4151 	ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4152 	       bma->offset + bma->length);
4153 	ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4154 	       bma->got.br_state == XFS_EXT_UNWRITTEN);
4155 	return 0;
4156 }
4157 
4158 STATIC int
4159 xfs_bmapi_convert_unwritten(
4160 	struct xfs_bmalloca	*bma,
4161 	struct xfs_bmbt_irec	*mval,
4162 	xfs_filblks_t		len,
4163 	int			flags)
4164 {
4165 	int			whichfork = xfs_bmapi_whichfork(flags);
4166 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4167 	int			tmp_logflags = 0;
4168 	int			error;
4169 
4170 	/* check if we need to do unwritten->real conversion */
4171 	if (mval->br_state == XFS_EXT_UNWRITTEN &&
4172 	    (flags & XFS_BMAPI_PREALLOC))
4173 		return 0;
4174 
4175 	/* check if we need to do real->unwritten conversion */
4176 	if (mval->br_state == XFS_EXT_NORM &&
4177 	    (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4178 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4179 		return 0;
4180 
4181 	/*
4182 	 * Modify (by adding) the state flag, if writing.
4183 	 */
4184 	ASSERT(mval->br_blockcount <= len);
4185 	if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4186 		bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4187 					bma->ip, whichfork);
4188 		bma->cur->bc_private.b.firstblock = *bma->firstblock;
4189 		bma->cur->bc_private.b.dfops = bma->dfops;
4190 	}
4191 	mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4192 				? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4193 
4194 	/*
4195 	 * Before insertion into the bmbt, zero the range being converted
4196 	 * if required.
4197 	 */
4198 	if (flags & XFS_BMAPI_ZERO) {
4199 		error = xfs_zero_extent(bma->ip, mval->br_startblock,
4200 					mval->br_blockcount);
4201 		if (error)
4202 			return error;
4203 	}
4204 
4205 	error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork,
4206 			&bma->icur, &bma->cur, mval, bma->firstblock,
4207 			bma->dfops, &tmp_logflags);
4208 	/*
4209 	 * Log the inode core unconditionally in the unwritten extent conversion
4210 	 * path because the conversion might not have done so (e.g., if the
4211 	 * extent count hasn't changed). We need to make sure the inode is dirty
4212 	 * in the transaction for the sake of fsync(), even if nothing has
4213 	 * changed, because fsync() will not force the log for this transaction
4214 	 * unless it sees the inode pinned.
4215 	 *
4216 	 * Note: If we're only converting cow fork extents, there aren't
4217 	 * any on-disk updates to make, so we don't need to log anything.
4218 	 */
4219 	if (whichfork != XFS_COW_FORK)
4220 		bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4221 	if (error)
4222 		return error;
4223 
4224 	/*
4225 	 * Update our extent pointer, given that
4226 	 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4227 	 * of the neighbouring ones.
4228 	 */
4229 	xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4230 
4231 	/*
4232 	 * We may have combined previously unwritten space with written space,
4233 	 * so generate another request.
4234 	 */
4235 	if (mval->br_blockcount < len)
4236 		return -EAGAIN;
4237 	return 0;
4238 }
4239 
4240 /*
4241  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4242  * extent state if necessary.  Details behaviour is controlled by the flags
4243  * parameter.  Only allocates blocks from a single allocation group, to avoid
4244  * locking problems.
4245  *
4246  * The returned value in "firstblock" from the first call in a transaction
4247  * must be remembered and presented to subsequent calls in "firstblock".
4248  * An upper bound for the number of blocks to be allocated is supplied to
4249  * the first call in "total"; if no allocation group has that many free
4250  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4251  */
4252 int
4253 xfs_bmapi_write(
4254 	struct xfs_trans	*tp,		/* transaction pointer */
4255 	struct xfs_inode	*ip,		/* incore inode */
4256 	xfs_fileoff_t		bno,		/* starting file offs. mapped */
4257 	xfs_filblks_t		len,		/* length to map in file */
4258 	int			flags,		/* XFS_BMAPI_... */
4259 	xfs_fsblock_t		*firstblock,	/* first allocated block
4260 						   controls a.g. for allocs */
4261 	xfs_extlen_t		total,		/* total blocks needed */
4262 	struct xfs_bmbt_irec	*mval,		/* output: map values */
4263 	int			*nmap,		/* i/o: mval size/count */
4264 	struct xfs_defer_ops	*dfops)		/* i/o: list extents to free */
4265 {
4266 	struct xfs_mount	*mp = ip->i_mount;
4267 	struct xfs_ifork	*ifp;
4268 	struct xfs_bmalloca	bma = { NULL };	/* args for xfs_bmap_alloc */
4269 	xfs_fileoff_t		end;		/* end of mapped file region */
4270 	bool			eof = false;	/* after the end of extents */
4271 	int			error;		/* error return */
4272 	int			n;		/* current extent index */
4273 	xfs_fileoff_t		obno;		/* old block number (offset) */
4274 	int			whichfork;	/* data or attr fork */
4275 
4276 #ifdef DEBUG
4277 	xfs_fileoff_t		orig_bno;	/* original block number value */
4278 	int			orig_flags;	/* original flags arg value */
4279 	xfs_filblks_t		orig_len;	/* original value of len arg */
4280 	struct xfs_bmbt_irec	*orig_mval;	/* original value of mval */
4281 	int			orig_nmap;	/* original value of *nmap */
4282 
4283 	orig_bno = bno;
4284 	orig_len = len;
4285 	orig_flags = flags;
4286 	orig_mval = mval;
4287 	orig_nmap = *nmap;
4288 #endif
4289 	whichfork = xfs_bmapi_whichfork(flags);
4290 
4291 	ASSERT(*nmap >= 1);
4292 	ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4293 	ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4294 	ASSERT(tp != NULL ||
4295 	       (flags & (XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK)) ==
4296 			(XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK));
4297 	ASSERT(len > 0);
4298 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4299 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4300 	ASSERT(!(flags & XFS_BMAPI_REMAP));
4301 
4302 	/* zeroing is for currently only for data extents, not metadata */
4303 	ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4304 			(XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4305 	/*
4306 	 * we can allocate unwritten extents or pre-zero allocated blocks,
4307 	 * but it makes no sense to do both at once. This would result in
4308 	 * zeroing the unwritten extent twice, but it still being an
4309 	 * unwritten extent....
4310 	 */
4311 	ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4312 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4313 
4314 	if (unlikely(XFS_TEST_ERROR(
4315 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4316 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4317 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
4318 		XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4319 		return -EFSCORRUPTED;
4320 	}
4321 
4322 	if (XFS_FORCED_SHUTDOWN(mp))
4323 		return -EIO;
4324 
4325 	ifp = XFS_IFORK_PTR(ip, whichfork);
4326 
4327 	XFS_STATS_INC(mp, xs_blk_mapw);
4328 
4329 	if (*firstblock == NULLFSBLOCK) {
4330 		if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4331 			bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4332 		else
4333 			bma.minleft = 1;
4334 	} else {
4335 		bma.minleft = 0;
4336 	}
4337 
4338 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4339 		error = xfs_iread_extents(tp, ip, whichfork);
4340 		if (error)
4341 			goto error0;
4342 	}
4343 
4344 	n = 0;
4345 	end = bno + len;
4346 	obno = bno;
4347 
4348 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.icur, &bma.got))
4349 		eof = true;
4350 	if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4351 		bma.prev.br_startoff = NULLFILEOFF;
4352 	bma.tp = tp;
4353 	bma.ip = ip;
4354 	bma.total = total;
4355 	bma.datatype = 0;
4356 	bma.dfops = dfops;
4357 	bma.firstblock = firstblock;
4358 
4359 	while (bno < end && n < *nmap) {
4360 		bool			need_alloc = false, wasdelay = false;
4361 
4362 		/* in hole or beyond EOF? */
4363 		if (eof || bma.got.br_startoff > bno) {
4364 			/*
4365 			 * CoW fork conversions should /never/ hit EOF or
4366 			 * holes.  There should always be something for us
4367 			 * to work on.
4368 			 */
4369 			ASSERT(!((flags & XFS_BMAPI_CONVERT) &&
4370 			         (flags & XFS_BMAPI_COWFORK)));
4371 
4372 			if (flags & XFS_BMAPI_DELALLOC) {
4373 				/*
4374 				 * For the COW fork we can reasonably get a
4375 				 * request for converting an extent that races
4376 				 * with other threads already having converted
4377 				 * part of it, as there converting COW to
4378 				 * regular blocks is not protected using the
4379 				 * IOLOCK.
4380 				 */
4381 				ASSERT(flags & XFS_BMAPI_COWFORK);
4382 				if (!(flags & XFS_BMAPI_COWFORK)) {
4383 					error = -EIO;
4384 					goto error0;
4385 				}
4386 
4387 				if (eof || bno >= end)
4388 					break;
4389 			} else {
4390 				need_alloc = true;
4391 			}
4392 		} else if (isnullstartblock(bma.got.br_startblock)) {
4393 			wasdelay = true;
4394 		}
4395 
4396 		/*
4397 		 * First, deal with the hole before the allocated space
4398 		 * that we found, if any.
4399 		 */
4400 		if ((need_alloc || wasdelay) &&
4401 		    !(flags & XFS_BMAPI_CONVERT_ONLY)) {
4402 			bma.eof = eof;
4403 			bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4404 			bma.wasdel = wasdelay;
4405 			bma.offset = bno;
4406 			bma.flags = flags;
4407 
4408 			/*
4409 			 * There's a 32/64 bit type mismatch between the
4410 			 * allocation length request (which can be 64 bits in
4411 			 * length) and the bma length request, which is
4412 			 * xfs_extlen_t and therefore 32 bits. Hence we have to
4413 			 * check for 32-bit overflows and handle them here.
4414 			 */
4415 			if (len > (xfs_filblks_t)MAXEXTLEN)
4416 				bma.length = MAXEXTLEN;
4417 			else
4418 				bma.length = len;
4419 
4420 			ASSERT(len > 0);
4421 			ASSERT(bma.length > 0);
4422 			error = xfs_bmapi_allocate(&bma);
4423 			if (error)
4424 				goto error0;
4425 			if (bma.blkno == NULLFSBLOCK)
4426 				break;
4427 
4428 			/*
4429 			 * If this is a CoW allocation, record the data in
4430 			 * the refcount btree for orphan recovery.
4431 			 */
4432 			if (whichfork == XFS_COW_FORK) {
4433 				error = xfs_refcount_alloc_cow_extent(mp, dfops,
4434 						bma.blkno, bma.length);
4435 				if (error)
4436 					goto error0;
4437 			}
4438 		}
4439 
4440 		/* Deal with the allocated space we found.  */
4441 		xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4442 							end, n, flags);
4443 
4444 		/* Execute unwritten extent conversion if necessary */
4445 		error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4446 		if (error == -EAGAIN)
4447 			continue;
4448 		if (error)
4449 			goto error0;
4450 
4451 		/* update the extent map to return */
4452 		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4453 
4454 		/*
4455 		 * If we're done, stop now.  Stop when we've allocated
4456 		 * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4457 		 * the transaction may get too big.
4458 		 */
4459 		if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4460 			break;
4461 
4462 		/* Else go on to the next record. */
4463 		bma.prev = bma.got;
4464 		if (!xfs_iext_next_extent(ifp, &bma.icur, &bma.got))
4465 			eof = true;
4466 	}
4467 	*nmap = n;
4468 
4469 	/*
4470 	 * Transform from btree to extents, give it cur.
4471 	 */
4472 	if (xfs_bmap_wants_extents(ip, whichfork)) {
4473 		int		tmp_logflags = 0;
4474 
4475 		ASSERT(bma.cur);
4476 		error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4477 			&tmp_logflags, whichfork);
4478 		bma.logflags |= tmp_logflags;
4479 		if (error)
4480 			goto error0;
4481 	}
4482 
4483 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4484 	       XFS_IFORK_NEXTENTS(ip, whichfork) >
4485 		XFS_IFORK_MAXEXT(ip, whichfork));
4486 	error = 0;
4487 error0:
4488 	/*
4489 	 * Log everything.  Do this after conversion, there's no point in
4490 	 * logging the extent records if we've converted to btree format.
4491 	 */
4492 	if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4493 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4494 		bma.logflags &= ~xfs_ilog_fext(whichfork);
4495 	else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4496 		 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4497 		bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4498 	/*
4499 	 * Log whatever the flags say, even if error.  Otherwise we might miss
4500 	 * detecting a case where the data is changed, there's an error,
4501 	 * and it's not logged so we don't shutdown when we should.
4502 	 */
4503 	if (bma.logflags)
4504 		xfs_trans_log_inode(tp, ip, bma.logflags);
4505 
4506 	if (bma.cur) {
4507 		if (!error) {
4508 			ASSERT(*firstblock == NULLFSBLOCK ||
4509 			       XFS_FSB_TO_AGNO(mp, *firstblock) <=
4510 			       XFS_FSB_TO_AGNO(mp,
4511 				       bma.cur->bc_private.b.firstblock));
4512 			*firstblock = bma.cur->bc_private.b.firstblock;
4513 		}
4514 		xfs_btree_del_cursor(bma.cur,
4515 			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4516 	}
4517 	if (!error)
4518 		xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4519 			orig_nmap, *nmap);
4520 	return error;
4521 }
4522 
4523 int
4524 xfs_bmapi_remap(
4525 	struct xfs_trans	*tp,
4526 	struct xfs_inode	*ip,
4527 	xfs_fileoff_t		bno,
4528 	xfs_filblks_t		len,
4529 	xfs_fsblock_t		startblock,
4530 	struct xfs_defer_ops	*dfops,
4531 	int			flags)
4532 {
4533 	struct xfs_mount	*mp = ip->i_mount;
4534 	struct xfs_ifork	*ifp;
4535 	struct xfs_btree_cur	*cur = NULL;
4536 	xfs_fsblock_t		firstblock = NULLFSBLOCK;
4537 	struct xfs_bmbt_irec	got;
4538 	struct xfs_iext_cursor	icur;
4539 	int			whichfork = xfs_bmapi_whichfork(flags);
4540 	int			logflags = 0, error;
4541 
4542 	ifp = XFS_IFORK_PTR(ip, whichfork);
4543 	ASSERT(len > 0);
4544 	ASSERT(len <= (xfs_filblks_t)MAXEXTLEN);
4545 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4546 	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC |
4547 			   XFS_BMAPI_NORMAP)));
4548 	ASSERT((flags & (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC)) !=
4549 			(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC));
4550 
4551 	if (unlikely(XFS_TEST_ERROR(
4552 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4553 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4554 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
4555 		XFS_ERROR_REPORT("xfs_bmapi_remap", XFS_ERRLEVEL_LOW, mp);
4556 		return -EFSCORRUPTED;
4557 	}
4558 
4559 	if (XFS_FORCED_SHUTDOWN(mp))
4560 		return -EIO;
4561 
4562 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4563 		error = xfs_iread_extents(tp, ip, whichfork);
4564 		if (error)
4565 			return error;
4566 	}
4567 
4568 	if (xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
4569 		/* make sure we only reflink into a hole. */
4570 		ASSERT(got.br_startoff > bno);
4571 		ASSERT(got.br_startoff - bno >= len);
4572 	}
4573 
4574 	ip->i_d.di_nblocks += len;
4575 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4576 
4577 	if (ifp->if_flags & XFS_IFBROOT) {
4578 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
4579 		cur->bc_private.b.firstblock = firstblock;
4580 		cur->bc_private.b.dfops = dfops;
4581 		cur->bc_private.b.flags = 0;
4582 	}
4583 
4584 	got.br_startoff = bno;
4585 	got.br_startblock = startblock;
4586 	got.br_blockcount = len;
4587 	if (flags & XFS_BMAPI_PREALLOC)
4588 		got.br_state = XFS_EXT_UNWRITTEN;
4589 	else
4590 		got.br_state = XFS_EXT_NORM;
4591 
4592 	error = xfs_bmap_add_extent_hole_real(tp, ip, whichfork, &icur,
4593 			&cur, &got, &firstblock, dfops, &logflags, flags);
4594 	if (error)
4595 		goto error0;
4596 
4597 	if (xfs_bmap_wants_extents(ip, whichfork)) {
4598 		int		tmp_logflags = 0;
4599 
4600 		error = xfs_bmap_btree_to_extents(tp, ip, cur,
4601 			&tmp_logflags, whichfork);
4602 		logflags |= tmp_logflags;
4603 	}
4604 
4605 error0:
4606 	if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS)
4607 		logflags &= ~XFS_ILOG_DEXT;
4608 	else if (ip->i_d.di_format != XFS_DINODE_FMT_BTREE)
4609 		logflags &= ~XFS_ILOG_DBROOT;
4610 
4611 	if (logflags)
4612 		xfs_trans_log_inode(tp, ip, logflags);
4613 	if (cur) {
4614 		xfs_btree_del_cursor(cur,
4615 				error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4616 	}
4617 	return error;
4618 }
4619 
4620 /*
4621  * When a delalloc extent is split (e.g., due to a hole punch), the original
4622  * indlen reservation must be shared across the two new extents that are left
4623  * behind.
4624  *
4625  * Given the original reservation and the worst case indlen for the two new
4626  * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4627  * reservation fairly across the two new extents. If necessary, steal available
4628  * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4629  * ores == 1). The number of stolen blocks is returned. The availability and
4630  * subsequent accounting of stolen blocks is the responsibility of the caller.
4631  */
4632 static xfs_filblks_t
4633 xfs_bmap_split_indlen(
4634 	xfs_filblks_t			ores,		/* original res. */
4635 	xfs_filblks_t			*indlen1,	/* ext1 worst indlen */
4636 	xfs_filblks_t			*indlen2,	/* ext2 worst indlen */
4637 	xfs_filblks_t			avail)		/* stealable blocks */
4638 {
4639 	xfs_filblks_t			len1 = *indlen1;
4640 	xfs_filblks_t			len2 = *indlen2;
4641 	xfs_filblks_t			nres = len1 + len2; /* new total res. */
4642 	xfs_filblks_t			stolen = 0;
4643 	xfs_filblks_t			resfactor;
4644 
4645 	/*
4646 	 * Steal as many blocks as we can to try and satisfy the worst case
4647 	 * indlen for both new extents.
4648 	 */
4649 	if (ores < nres && avail)
4650 		stolen = XFS_FILBLKS_MIN(nres - ores, avail);
4651 	ores += stolen;
4652 
4653 	 /* nothing else to do if we've satisfied the new reservation */
4654 	if (ores >= nres)
4655 		return stolen;
4656 
4657 	/*
4658 	 * We can't meet the total required reservation for the two extents.
4659 	 * Calculate the percent of the overall shortage between both extents
4660 	 * and apply this percentage to each of the requested indlen values.
4661 	 * This distributes the shortage fairly and reduces the chances that one
4662 	 * of the two extents is left with nothing when extents are repeatedly
4663 	 * split.
4664 	 */
4665 	resfactor = (ores * 100);
4666 	do_div(resfactor, nres);
4667 	len1 *= resfactor;
4668 	do_div(len1, 100);
4669 	len2 *= resfactor;
4670 	do_div(len2, 100);
4671 	ASSERT(len1 + len2 <= ores);
4672 	ASSERT(len1 < *indlen1 && len2 < *indlen2);
4673 
4674 	/*
4675 	 * Hand out the remainder to each extent. If one of the two reservations
4676 	 * is zero, we want to make sure that one gets a block first. The loop
4677 	 * below starts with len1, so hand len2 a block right off the bat if it
4678 	 * is zero.
4679 	 */
4680 	ores -= (len1 + len2);
4681 	ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores);
4682 	if (ores && !len2 && *indlen2) {
4683 		len2++;
4684 		ores--;
4685 	}
4686 	while (ores) {
4687 		if (len1 < *indlen1) {
4688 			len1++;
4689 			ores--;
4690 		}
4691 		if (!ores)
4692 			break;
4693 		if (len2 < *indlen2) {
4694 			len2++;
4695 			ores--;
4696 		}
4697 	}
4698 
4699 	*indlen1 = len1;
4700 	*indlen2 = len2;
4701 
4702 	return stolen;
4703 }
4704 
4705 int
4706 xfs_bmap_del_extent_delay(
4707 	struct xfs_inode	*ip,
4708 	int			whichfork,
4709 	struct xfs_iext_cursor	*icur,
4710 	struct xfs_bmbt_irec	*got,
4711 	struct xfs_bmbt_irec	*del)
4712 {
4713 	struct xfs_mount	*mp = ip->i_mount;
4714 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
4715 	struct xfs_bmbt_irec	new;
4716 	int64_t			da_old, da_new, da_diff = 0;
4717 	xfs_fileoff_t		del_endoff, got_endoff;
4718 	xfs_filblks_t		got_indlen, new_indlen, stolen;
4719 	int			state = xfs_bmap_fork_to_state(whichfork);
4720 	int			error = 0;
4721 	bool			isrt;
4722 
4723 	XFS_STATS_INC(mp, xs_del_exlist);
4724 
4725 	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4726 	del_endoff = del->br_startoff + del->br_blockcount;
4727 	got_endoff = got->br_startoff + got->br_blockcount;
4728 	da_old = startblockval(got->br_startblock);
4729 	da_new = 0;
4730 
4731 	ASSERT(del->br_blockcount > 0);
4732 	ASSERT(got->br_startoff <= del->br_startoff);
4733 	ASSERT(got_endoff >= del_endoff);
4734 
4735 	if (isrt) {
4736 		uint64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount);
4737 
4738 		do_div(rtexts, mp->m_sb.sb_rextsize);
4739 		xfs_mod_frextents(mp, rtexts);
4740 	}
4741 
4742 	/*
4743 	 * Update the inode delalloc counter now and wait to update the
4744 	 * sb counters as we might have to borrow some blocks for the
4745 	 * indirect block accounting.
4746 	 */
4747 	error = xfs_trans_reserve_quota_nblks(NULL, ip,
4748 			-((long)del->br_blockcount), 0,
4749 			isrt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4750 	if (error)
4751 		return error;
4752 	ip->i_delayed_blks -= del->br_blockcount;
4753 
4754 	if (got->br_startoff == del->br_startoff)
4755 		state |= BMAP_LEFT_FILLING;
4756 	if (got_endoff == del_endoff)
4757 		state |= BMAP_RIGHT_FILLING;
4758 
4759 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4760 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4761 		/*
4762 		 * Matches the whole extent.  Delete the entry.
4763 		 */
4764 		xfs_iext_remove(ip, icur, state);
4765 		xfs_iext_prev(ifp, icur);
4766 		break;
4767 	case BMAP_LEFT_FILLING:
4768 		/*
4769 		 * Deleting the first part of the extent.
4770 		 */
4771 		got->br_startoff = del_endoff;
4772 		got->br_blockcount -= del->br_blockcount;
4773 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4774 				got->br_blockcount), da_old);
4775 		got->br_startblock = nullstartblock((int)da_new);
4776 		xfs_iext_update_extent(ip, state, icur, got);
4777 		break;
4778 	case BMAP_RIGHT_FILLING:
4779 		/*
4780 		 * Deleting the last part of the extent.
4781 		 */
4782 		got->br_blockcount = got->br_blockcount - del->br_blockcount;
4783 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4784 				got->br_blockcount), da_old);
4785 		got->br_startblock = nullstartblock((int)da_new);
4786 		xfs_iext_update_extent(ip, state, icur, got);
4787 		break;
4788 	case 0:
4789 		/*
4790 		 * Deleting the middle of the extent.
4791 		 *
4792 		 * Distribute the original indlen reservation across the two new
4793 		 * extents.  Steal blocks from the deleted extent if necessary.
4794 		 * Stealing blocks simply fudges the fdblocks accounting below.
4795 		 * Warn if either of the new indlen reservations is zero as this
4796 		 * can lead to delalloc problems.
4797 		 */
4798 		got->br_blockcount = del->br_startoff - got->br_startoff;
4799 		got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount);
4800 
4801 		new.br_blockcount = got_endoff - del_endoff;
4802 		new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount);
4803 
4804 		WARN_ON_ONCE(!got_indlen || !new_indlen);
4805 		stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen,
4806 						       del->br_blockcount);
4807 
4808 		got->br_startblock = nullstartblock((int)got_indlen);
4809 
4810 		new.br_startoff = del_endoff;
4811 		new.br_state = got->br_state;
4812 		new.br_startblock = nullstartblock((int)new_indlen);
4813 
4814 		xfs_iext_update_extent(ip, state, icur, got);
4815 		xfs_iext_next(ifp, icur);
4816 		xfs_iext_insert(ip, icur, &new, state);
4817 
4818 		da_new = got_indlen + new_indlen - stolen;
4819 		del->br_blockcount -= stolen;
4820 		break;
4821 	}
4822 
4823 	ASSERT(da_old >= da_new);
4824 	da_diff = da_old - da_new;
4825 	if (!isrt)
4826 		da_diff += del->br_blockcount;
4827 	if (da_diff)
4828 		xfs_mod_fdblocks(mp, da_diff, false);
4829 	return error;
4830 }
4831 
4832 void
4833 xfs_bmap_del_extent_cow(
4834 	struct xfs_inode	*ip,
4835 	struct xfs_iext_cursor	*icur,
4836 	struct xfs_bmbt_irec	*got,
4837 	struct xfs_bmbt_irec	*del)
4838 {
4839 	struct xfs_mount	*mp = ip->i_mount;
4840 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
4841 	struct xfs_bmbt_irec	new;
4842 	xfs_fileoff_t		del_endoff, got_endoff;
4843 	int			state = BMAP_COWFORK;
4844 
4845 	XFS_STATS_INC(mp, xs_del_exlist);
4846 
4847 	del_endoff = del->br_startoff + del->br_blockcount;
4848 	got_endoff = got->br_startoff + got->br_blockcount;
4849 
4850 	ASSERT(del->br_blockcount > 0);
4851 	ASSERT(got->br_startoff <= del->br_startoff);
4852 	ASSERT(got_endoff >= del_endoff);
4853 	ASSERT(!isnullstartblock(got->br_startblock));
4854 
4855 	if (got->br_startoff == del->br_startoff)
4856 		state |= BMAP_LEFT_FILLING;
4857 	if (got_endoff == del_endoff)
4858 		state |= BMAP_RIGHT_FILLING;
4859 
4860 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4861 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4862 		/*
4863 		 * Matches the whole extent.  Delete the entry.
4864 		 */
4865 		xfs_iext_remove(ip, icur, state);
4866 		xfs_iext_prev(ifp, icur);
4867 		break;
4868 	case BMAP_LEFT_FILLING:
4869 		/*
4870 		 * Deleting the first part of the extent.
4871 		 */
4872 		got->br_startoff = del_endoff;
4873 		got->br_blockcount -= del->br_blockcount;
4874 		got->br_startblock = del->br_startblock + del->br_blockcount;
4875 		xfs_iext_update_extent(ip, state, icur, got);
4876 		break;
4877 	case BMAP_RIGHT_FILLING:
4878 		/*
4879 		 * Deleting the last part of the extent.
4880 		 */
4881 		got->br_blockcount -= del->br_blockcount;
4882 		xfs_iext_update_extent(ip, state, icur, got);
4883 		break;
4884 	case 0:
4885 		/*
4886 		 * Deleting the middle of the extent.
4887 		 */
4888 		got->br_blockcount = del->br_startoff - got->br_startoff;
4889 
4890 		new.br_startoff = del_endoff;
4891 		new.br_blockcount = got_endoff - del_endoff;
4892 		new.br_state = got->br_state;
4893 		new.br_startblock = del->br_startblock + del->br_blockcount;
4894 
4895 		xfs_iext_update_extent(ip, state, icur, got);
4896 		xfs_iext_next(ifp, icur);
4897 		xfs_iext_insert(ip, icur, &new, state);
4898 		break;
4899 	}
4900 	ip->i_delayed_blks -= del->br_blockcount;
4901 }
4902 
4903 /*
4904  * Called by xfs_bmapi to update file extent records and the btree
4905  * after removing space.
4906  */
4907 STATIC int				/* error */
4908 xfs_bmap_del_extent_real(
4909 	xfs_inode_t		*ip,	/* incore inode pointer */
4910 	xfs_trans_t		*tp,	/* current transaction pointer */
4911 	struct xfs_iext_cursor	*icur,
4912 	struct xfs_defer_ops	*dfops,	/* list of extents to be freed */
4913 	xfs_btree_cur_t		*cur,	/* if null, not a btree */
4914 	xfs_bmbt_irec_t		*del,	/* data to remove from extents */
4915 	int			*logflagsp, /* inode logging flags */
4916 	int			whichfork, /* data or attr fork */
4917 	int			bflags)	/* bmapi flags */
4918 {
4919 	xfs_fsblock_t		del_endblock=0;	/* first block past del */
4920 	xfs_fileoff_t		del_endoff;	/* first offset past del */
4921 	int			do_fx;	/* free extent at end of routine */
4922 	int			error;	/* error return value */
4923 	int			flags = 0;/* inode logging flags */
4924 	struct xfs_bmbt_irec	got;	/* current extent entry */
4925 	xfs_fileoff_t		got_endoff;	/* first offset past got */
4926 	int			i;	/* temp state */
4927 	xfs_ifork_t		*ifp;	/* inode fork pointer */
4928 	xfs_mount_t		*mp;	/* mount structure */
4929 	xfs_filblks_t		nblks;	/* quota/sb block count */
4930 	xfs_bmbt_irec_t		new;	/* new record to be inserted */
4931 	/* REFERENCED */
4932 	uint			qfield;	/* quota field to update */
4933 	int			state = xfs_bmap_fork_to_state(whichfork);
4934 	struct xfs_bmbt_irec	old;
4935 
4936 	mp = ip->i_mount;
4937 	XFS_STATS_INC(mp, xs_del_exlist);
4938 
4939 	ifp = XFS_IFORK_PTR(ip, whichfork);
4940 	ASSERT(del->br_blockcount > 0);
4941 	xfs_iext_get_extent(ifp, icur, &got);
4942 	ASSERT(got.br_startoff <= del->br_startoff);
4943 	del_endoff = del->br_startoff + del->br_blockcount;
4944 	got_endoff = got.br_startoff + got.br_blockcount;
4945 	ASSERT(got_endoff >= del_endoff);
4946 	ASSERT(!isnullstartblock(got.br_startblock));
4947 	qfield = 0;
4948 	error = 0;
4949 
4950 	/*
4951 	 * If it's the case where the directory code is running with no block
4952 	 * reservation, and the deleted block is in the middle of its extent,
4953 	 * and the resulting insert of an extent would cause transformation to
4954 	 * btree format, then reject it.  The calling code will then swap blocks
4955 	 * around instead.  We have to do this now, rather than waiting for the
4956 	 * conversion to btree format, since the transaction will be dirty then.
4957 	 */
4958 	if (tp->t_blk_res == 0 &&
4959 	    XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
4960 	    XFS_IFORK_NEXTENTS(ip, whichfork) >=
4961 			XFS_IFORK_MAXEXT(ip, whichfork) &&
4962 	    del->br_startoff > got.br_startoff && del_endoff < got_endoff)
4963 		return -ENOSPC;
4964 
4965 	flags = XFS_ILOG_CORE;
4966 	if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
4967 		xfs_fsblock_t	bno;
4968 		xfs_filblks_t	len;
4969 
4970 		ASSERT(do_mod(del->br_blockcount, mp->m_sb.sb_rextsize) == 0);
4971 		ASSERT(do_mod(del->br_startblock, mp->m_sb.sb_rextsize) == 0);
4972 		bno = del->br_startblock;
4973 		len = del->br_blockcount;
4974 		do_div(bno, mp->m_sb.sb_rextsize);
4975 		do_div(len, mp->m_sb.sb_rextsize);
4976 		error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
4977 		if (error)
4978 			goto done;
4979 		do_fx = 0;
4980 		nblks = len * mp->m_sb.sb_rextsize;
4981 		qfield = XFS_TRANS_DQ_RTBCOUNT;
4982 	} else {
4983 		do_fx = 1;
4984 		nblks = del->br_blockcount;
4985 		qfield = XFS_TRANS_DQ_BCOUNT;
4986 	}
4987 
4988 	del_endblock = del->br_startblock + del->br_blockcount;
4989 	if (cur) {
4990 		error = xfs_bmbt_lookup_eq(cur, &got, &i);
4991 		if (error)
4992 			goto done;
4993 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
4994 	}
4995 
4996 	if (got.br_startoff == del->br_startoff)
4997 		state |= BMAP_LEFT_FILLING;
4998 	if (got_endoff == del_endoff)
4999 		state |= BMAP_RIGHT_FILLING;
5000 
5001 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
5002 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
5003 		/*
5004 		 * Matches the whole extent.  Delete the entry.
5005 		 */
5006 		xfs_iext_remove(ip, icur, state);
5007 		xfs_iext_prev(ifp, icur);
5008 		XFS_IFORK_NEXT_SET(ip, whichfork,
5009 			XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5010 		flags |= XFS_ILOG_CORE;
5011 		if (!cur) {
5012 			flags |= xfs_ilog_fext(whichfork);
5013 			break;
5014 		}
5015 		if ((error = xfs_btree_delete(cur, &i)))
5016 			goto done;
5017 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5018 		break;
5019 	case BMAP_LEFT_FILLING:
5020 		/*
5021 		 * Deleting the first part of the extent.
5022 		 */
5023 		got.br_startoff = del_endoff;
5024 		got.br_startblock = del_endblock;
5025 		got.br_blockcount -= del->br_blockcount;
5026 		xfs_iext_update_extent(ip, state, icur, &got);
5027 		if (!cur) {
5028 			flags |= xfs_ilog_fext(whichfork);
5029 			break;
5030 		}
5031 		error = xfs_bmbt_update(cur, &got);
5032 		if (error)
5033 			goto done;
5034 		break;
5035 	case BMAP_RIGHT_FILLING:
5036 		/*
5037 		 * Deleting the last part of the extent.
5038 		 */
5039 		got.br_blockcount -= del->br_blockcount;
5040 		xfs_iext_update_extent(ip, state, icur, &got);
5041 		if (!cur) {
5042 			flags |= xfs_ilog_fext(whichfork);
5043 			break;
5044 		}
5045 		error = xfs_bmbt_update(cur, &got);
5046 		if (error)
5047 			goto done;
5048 		break;
5049 	case 0:
5050 		/*
5051 		 * Deleting the middle of the extent.
5052 		 */
5053 		old = got;
5054 
5055 		got.br_blockcount = del->br_startoff - got.br_startoff;
5056 		xfs_iext_update_extent(ip, state, icur, &got);
5057 
5058 		new.br_startoff = del_endoff;
5059 		new.br_blockcount = got_endoff - del_endoff;
5060 		new.br_state = got.br_state;
5061 		new.br_startblock = del_endblock;
5062 
5063 		flags |= XFS_ILOG_CORE;
5064 		if (cur) {
5065 			error = xfs_bmbt_update(cur, &got);
5066 			if (error)
5067 				goto done;
5068 			error = xfs_btree_increment(cur, 0, &i);
5069 			if (error)
5070 				goto done;
5071 			cur->bc_rec.b = new;
5072 			error = xfs_btree_insert(cur, &i);
5073 			if (error && error != -ENOSPC)
5074 				goto done;
5075 			/*
5076 			 * If get no-space back from btree insert, it tried a
5077 			 * split, and we have a zero block reservation.  Fix up
5078 			 * our state and return the error.
5079 			 */
5080 			if (error == -ENOSPC) {
5081 				/*
5082 				 * Reset the cursor, don't trust it after any
5083 				 * insert operation.
5084 				 */
5085 				error = xfs_bmbt_lookup_eq(cur, &got, &i);
5086 				if (error)
5087 					goto done;
5088 				XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5089 				/*
5090 				 * Update the btree record back
5091 				 * to the original value.
5092 				 */
5093 				error = xfs_bmbt_update(cur, &old);
5094 				if (error)
5095 					goto done;
5096 				/*
5097 				 * Reset the extent record back
5098 				 * to the original value.
5099 				 */
5100 				xfs_iext_update_extent(ip, state, icur, &old);
5101 				flags = 0;
5102 				error = -ENOSPC;
5103 				goto done;
5104 			}
5105 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5106 		} else
5107 			flags |= xfs_ilog_fext(whichfork);
5108 		XFS_IFORK_NEXT_SET(ip, whichfork,
5109 			XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5110 		xfs_iext_next(ifp, icur);
5111 		xfs_iext_insert(ip, icur, &new, state);
5112 		break;
5113 	}
5114 
5115 	/* remove reverse mapping */
5116 	error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, del);
5117 	if (error)
5118 		goto done;
5119 
5120 	/*
5121 	 * If we need to, add to list of extents to delete.
5122 	 */
5123 	if (do_fx && !(bflags & XFS_BMAPI_REMAP)) {
5124 		if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
5125 			error = xfs_refcount_decrease_extent(mp, dfops, del);
5126 			if (error)
5127 				goto done;
5128 		} else {
5129 			__xfs_bmap_add_free(mp, dfops, del->br_startblock,
5130 					del->br_blockcount, NULL,
5131 					(bflags & XFS_BMAPI_NODISCARD) ||
5132 					del->br_state == XFS_EXT_UNWRITTEN);
5133 		}
5134 	}
5135 
5136 	/*
5137 	 * Adjust inode # blocks in the file.
5138 	 */
5139 	if (nblks)
5140 		ip->i_d.di_nblocks -= nblks;
5141 	/*
5142 	 * Adjust quota data.
5143 	 */
5144 	if (qfield && !(bflags & XFS_BMAPI_REMAP))
5145 		xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5146 
5147 done:
5148 	*logflagsp = flags;
5149 	return error;
5150 }
5151 
5152 /*
5153  * Unmap (remove) blocks from a file.
5154  * If nexts is nonzero then the number of extents to remove is limited to
5155  * that value.  If not all extents in the block range can be removed then
5156  * *done is set.
5157  */
5158 int						/* error */
5159 __xfs_bunmapi(
5160 	xfs_trans_t		*tp,		/* transaction pointer */
5161 	struct xfs_inode	*ip,		/* incore inode */
5162 	xfs_fileoff_t		start,		/* first file offset deleted */
5163 	xfs_filblks_t		*rlen,		/* i/o: amount remaining */
5164 	int			flags,		/* misc flags */
5165 	xfs_extnum_t		nexts,		/* number of extents max */
5166 	xfs_fsblock_t		*firstblock,	/* first allocated block
5167 						   controls a.g. for allocs */
5168 	struct xfs_defer_ops	*dfops)		/* i/o: deferred updates */
5169 {
5170 	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
5171 	xfs_bmbt_irec_t		del;		/* extent being deleted */
5172 	int			error;		/* error return value */
5173 	xfs_extnum_t		extno;		/* extent number in list */
5174 	xfs_bmbt_irec_t		got;		/* current extent record */
5175 	xfs_ifork_t		*ifp;		/* inode fork pointer */
5176 	int			isrt;		/* freeing in rt area */
5177 	int			logflags;	/* transaction logging flags */
5178 	xfs_extlen_t		mod;		/* rt extent offset */
5179 	xfs_mount_t		*mp;		/* mount structure */
5180 	int			tmp_logflags;	/* partial logging flags */
5181 	int			wasdel;		/* was a delayed alloc extent */
5182 	int			whichfork;	/* data or attribute fork */
5183 	xfs_fsblock_t		sum;
5184 	xfs_filblks_t		len = *rlen;	/* length to unmap in file */
5185 	xfs_fileoff_t		max_len;
5186 	xfs_agnumber_t		prev_agno = NULLAGNUMBER, agno;
5187 	xfs_fileoff_t		end;
5188 	struct xfs_iext_cursor	icur;
5189 	bool			done = false;
5190 
5191 	trace_xfs_bunmap(ip, start, len, flags, _RET_IP_);
5192 
5193 	whichfork = xfs_bmapi_whichfork(flags);
5194 	ASSERT(whichfork != XFS_COW_FORK);
5195 	ifp = XFS_IFORK_PTR(ip, whichfork);
5196 	if (unlikely(
5197 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5198 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5199 		XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5200 				 ip->i_mount);
5201 		return -EFSCORRUPTED;
5202 	}
5203 	mp = ip->i_mount;
5204 	if (XFS_FORCED_SHUTDOWN(mp))
5205 		return -EIO;
5206 
5207 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5208 	ASSERT(len > 0);
5209 	ASSERT(nexts >= 0);
5210 
5211 	/*
5212 	 * Guesstimate how many blocks we can unmap without running the risk of
5213 	 * blowing out the transaction with a mix of EFIs and reflink
5214 	 * adjustments.
5215 	 */
5216 	if (tp && xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK)
5217 		max_len = min(len, xfs_refcount_max_unmap(tp->t_log_res));
5218 	else
5219 		max_len = len;
5220 
5221 	if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5222 	    (error = xfs_iread_extents(tp, ip, whichfork)))
5223 		return error;
5224 	if (xfs_iext_count(ifp) == 0) {
5225 		*rlen = 0;
5226 		return 0;
5227 	}
5228 	XFS_STATS_INC(mp, xs_blk_unmap);
5229 	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5230 	end = start + len;
5231 
5232 	if (!xfs_iext_lookup_extent_before(ip, ifp, &end, &icur, &got)) {
5233 		*rlen = 0;
5234 		return 0;
5235 	}
5236 	end--;
5237 
5238 	logflags = 0;
5239 	if (ifp->if_flags & XFS_IFBROOT) {
5240 		ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5241 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5242 		cur->bc_private.b.firstblock = *firstblock;
5243 		cur->bc_private.b.dfops = dfops;
5244 		cur->bc_private.b.flags = 0;
5245 	} else
5246 		cur = NULL;
5247 
5248 	if (isrt) {
5249 		/*
5250 		 * Synchronize by locking the bitmap inode.
5251 		 */
5252 		xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
5253 		xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5254 		xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
5255 		xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
5256 	}
5257 
5258 	extno = 0;
5259 	while (end != (xfs_fileoff_t)-1 && end >= start &&
5260 	       (nexts == 0 || extno < nexts) && max_len > 0) {
5261 		/*
5262 		 * Is the found extent after a hole in which end lives?
5263 		 * Just back up to the previous extent, if so.
5264 		 */
5265 		if (got.br_startoff > end &&
5266 		    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5267 			done = true;
5268 			break;
5269 		}
5270 		/*
5271 		 * Is the last block of this extent before the range
5272 		 * we're supposed to delete?  If so, we're done.
5273 		 */
5274 		end = XFS_FILEOFF_MIN(end,
5275 			got.br_startoff + got.br_blockcount - 1);
5276 		if (end < start)
5277 			break;
5278 		/*
5279 		 * Then deal with the (possibly delayed) allocated space
5280 		 * we found.
5281 		 */
5282 		del = got;
5283 		wasdel = isnullstartblock(del.br_startblock);
5284 
5285 		/*
5286 		 * Make sure we don't touch multiple AGF headers out of order
5287 		 * in a single transaction, as that could cause AB-BA deadlocks.
5288 		 */
5289 		if (!wasdel) {
5290 			agno = XFS_FSB_TO_AGNO(mp, del.br_startblock);
5291 			if (prev_agno != NULLAGNUMBER && prev_agno > agno)
5292 				break;
5293 			prev_agno = agno;
5294 		}
5295 		if (got.br_startoff < start) {
5296 			del.br_startoff = start;
5297 			del.br_blockcount -= start - got.br_startoff;
5298 			if (!wasdel)
5299 				del.br_startblock += start - got.br_startoff;
5300 		}
5301 		if (del.br_startoff + del.br_blockcount > end + 1)
5302 			del.br_blockcount = end + 1 - del.br_startoff;
5303 
5304 		/* How much can we safely unmap? */
5305 		if (max_len < del.br_blockcount) {
5306 			del.br_startoff += del.br_blockcount - max_len;
5307 			if (!wasdel)
5308 				del.br_startblock += del.br_blockcount - max_len;
5309 			del.br_blockcount = max_len;
5310 		}
5311 
5312 		sum = del.br_startblock + del.br_blockcount;
5313 		if (isrt &&
5314 		    (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5315 			/*
5316 			 * Realtime extent not lined up at the end.
5317 			 * The extent could have been split into written
5318 			 * and unwritten pieces, or we could just be
5319 			 * unmapping part of it.  But we can't really
5320 			 * get rid of part of a realtime extent.
5321 			 */
5322 			if (del.br_state == XFS_EXT_UNWRITTEN ||
5323 			    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5324 				/*
5325 				 * This piece is unwritten, or we're not
5326 				 * using unwritten extents.  Skip over it.
5327 				 */
5328 				ASSERT(end >= mod);
5329 				end -= mod > del.br_blockcount ?
5330 					del.br_blockcount : mod;
5331 				if (end < got.br_startoff &&
5332 				    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5333 					done = true;
5334 					break;
5335 				}
5336 				continue;
5337 			}
5338 			/*
5339 			 * It's written, turn it unwritten.
5340 			 * This is better than zeroing it.
5341 			 */
5342 			ASSERT(del.br_state == XFS_EXT_NORM);
5343 			ASSERT(tp->t_blk_res > 0);
5344 			/*
5345 			 * If this spans a realtime extent boundary,
5346 			 * chop it back to the start of the one we end at.
5347 			 */
5348 			if (del.br_blockcount > mod) {
5349 				del.br_startoff += del.br_blockcount - mod;
5350 				del.br_startblock += del.br_blockcount - mod;
5351 				del.br_blockcount = mod;
5352 			}
5353 			del.br_state = XFS_EXT_UNWRITTEN;
5354 			error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5355 					whichfork, &icur, &cur, &del,
5356 					firstblock, dfops, &logflags);
5357 			if (error)
5358 				goto error0;
5359 			goto nodelete;
5360 		}
5361 		if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5362 			/*
5363 			 * Realtime extent is lined up at the end but not
5364 			 * at the front.  We'll get rid of full extents if
5365 			 * we can.
5366 			 */
5367 			mod = mp->m_sb.sb_rextsize - mod;
5368 			if (del.br_blockcount > mod) {
5369 				del.br_blockcount -= mod;
5370 				del.br_startoff += mod;
5371 				del.br_startblock += mod;
5372 			} else if ((del.br_startoff == start &&
5373 				    (del.br_state == XFS_EXT_UNWRITTEN ||
5374 				     tp->t_blk_res == 0)) ||
5375 				   !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5376 				/*
5377 				 * Can't make it unwritten.  There isn't
5378 				 * a full extent here so just skip it.
5379 				 */
5380 				ASSERT(end >= del.br_blockcount);
5381 				end -= del.br_blockcount;
5382 				if (got.br_startoff > end &&
5383 				    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5384 					done = true;
5385 					break;
5386 				}
5387 				continue;
5388 			} else if (del.br_state == XFS_EXT_UNWRITTEN) {
5389 				struct xfs_bmbt_irec	prev;
5390 
5391 				/*
5392 				 * This one is already unwritten.
5393 				 * It must have a written left neighbor.
5394 				 * Unwrite the killed part of that one and
5395 				 * try again.
5396 				 */
5397 				if (!xfs_iext_prev_extent(ifp, &icur, &prev))
5398 					ASSERT(0);
5399 				ASSERT(prev.br_state == XFS_EXT_NORM);
5400 				ASSERT(!isnullstartblock(prev.br_startblock));
5401 				ASSERT(del.br_startblock ==
5402 				       prev.br_startblock + prev.br_blockcount);
5403 				if (prev.br_startoff < start) {
5404 					mod = start - prev.br_startoff;
5405 					prev.br_blockcount -= mod;
5406 					prev.br_startblock += mod;
5407 					prev.br_startoff = start;
5408 				}
5409 				prev.br_state = XFS_EXT_UNWRITTEN;
5410 				error = xfs_bmap_add_extent_unwritten_real(tp,
5411 						ip, whichfork, &icur, &cur,
5412 						&prev, firstblock, dfops,
5413 						&logflags);
5414 				if (error)
5415 					goto error0;
5416 				goto nodelete;
5417 			} else {
5418 				ASSERT(del.br_state == XFS_EXT_NORM);
5419 				del.br_state = XFS_EXT_UNWRITTEN;
5420 				error = xfs_bmap_add_extent_unwritten_real(tp,
5421 						ip, whichfork, &icur, &cur,
5422 						&del, firstblock, dfops,
5423 						&logflags);
5424 				if (error)
5425 					goto error0;
5426 				goto nodelete;
5427 			}
5428 		}
5429 
5430 		if (wasdel) {
5431 			error = xfs_bmap_del_extent_delay(ip, whichfork, &icur,
5432 					&got, &del);
5433 		} else {
5434 			error = xfs_bmap_del_extent_real(ip, tp, &icur, dfops,
5435 					cur, &del, &tmp_logflags, whichfork,
5436 					flags);
5437 			logflags |= tmp_logflags;
5438 		}
5439 
5440 		if (error)
5441 			goto error0;
5442 
5443 		max_len -= del.br_blockcount;
5444 		end = del.br_startoff - 1;
5445 nodelete:
5446 		/*
5447 		 * If not done go on to the next (previous) record.
5448 		 */
5449 		if (end != (xfs_fileoff_t)-1 && end >= start) {
5450 			if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5451 			    (got.br_startoff > end &&
5452 			     !xfs_iext_prev_extent(ifp, &icur, &got))) {
5453 				done = true;
5454 				break;
5455 			}
5456 			extno++;
5457 		}
5458 	}
5459 	if (done || end == (xfs_fileoff_t)-1 || end < start)
5460 		*rlen = 0;
5461 	else
5462 		*rlen = end - start + 1;
5463 
5464 	/*
5465 	 * Convert to a btree if necessary.
5466 	 */
5467 	if (xfs_bmap_needs_btree(ip, whichfork)) {
5468 		ASSERT(cur == NULL);
5469 		error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops,
5470 			&cur, 0, &tmp_logflags, whichfork);
5471 		logflags |= tmp_logflags;
5472 		if (error)
5473 			goto error0;
5474 	}
5475 	/*
5476 	 * transform from btree to extents, give it cur
5477 	 */
5478 	else if (xfs_bmap_wants_extents(ip, whichfork)) {
5479 		ASSERT(cur != NULL);
5480 		error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5481 			whichfork);
5482 		logflags |= tmp_logflags;
5483 		if (error)
5484 			goto error0;
5485 	}
5486 	/*
5487 	 * transform from extents to local?
5488 	 */
5489 	error = 0;
5490 error0:
5491 	/*
5492 	 * Log everything.  Do this after conversion, there's no point in
5493 	 * logging the extent records if we've converted to btree format.
5494 	 */
5495 	if ((logflags & xfs_ilog_fext(whichfork)) &&
5496 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5497 		logflags &= ~xfs_ilog_fext(whichfork);
5498 	else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5499 		 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5500 		logflags &= ~xfs_ilog_fbroot(whichfork);
5501 	/*
5502 	 * Log inode even in the error case, if the transaction
5503 	 * is dirty we'll need to shut down the filesystem.
5504 	 */
5505 	if (logflags)
5506 		xfs_trans_log_inode(tp, ip, logflags);
5507 	if (cur) {
5508 		if (!error) {
5509 			*firstblock = cur->bc_private.b.firstblock;
5510 			cur->bc_private.b.allocated = 0;
5511 		}
5512 		xfs_btree_del_cursor(cur,
5513 			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5514 	}
5515 	return error;
5516 }
5517 
5518 /* Unmap a range of a file. */
5519 int
5520 xfs_bunmapi(
5521 	xfs_trans_t		*tp,
5522 	struct xfs_inode	*ip,
5523 	xfs_fileoff_t		bno,
5524 	xfs_filblks_t		len,
5525 	int			flags,
5526 	xfs_extnum_t		nexts,
5527 	xfs_fsblock_t		*firstblock,
5528 	struct xfs_defer_ops	*dfops,
5529 	int			*done)
5530 {
5531 	int			error;
5532 
5533 	error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts, firstblock,
5534 			dfops);
5535 	*done = (len == 0);
5536 	return error;
5537 }
5538 
5539 /*
5540  * Determine whether an extent shift can be accomplished by a merge with the
5541  * extent that precedes the target hole of the shift.
5542  */
5543 STATIC bool
5544 xfs_bmse_can_merge(
5545 	struct xfs_bmbt_irec	*left,	/* preceding extent */
5546 	struct xfs_bmbt_irec	*got,	/* current extent to shift */
5547 	xfs_fileoff_t		shift)	/* shift fsb */
5548 {
5549 	xfs_fileoff_t		startoff;
5550 
5551 	startoff = got->br_startoff - shift;
5552 
5553 	/*
5554 	 * The extent, once shifted, must be adjacent in-file and on-disk with
5555 	 * the preceding extent.
5556 	 */
5557 	if ((left->br_startoff + left->br_blockcount != startoff) ||
5558 	    (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5559 	    (left->br_state != got->br_state) ||
5560 	    (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5561 		return false;
5562 
5563 	return true;
5564 }
5565 
5566 /*
5567  * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5568  * hole in the file. If an extent shift would result in the extent being fully
5569  * adjacent to the extent that currently precedes the hole, we can merge with
5570  * the preceding extent rather than do the shift.
5571  *
5572  * This function assumes the caller has verified a shift-by-merge is possible
5573  * with the provided extents via xfs_bmse_can_merge().
5574  */
5575 STATIC int
5576 xfs_bmse_merge(
5577 	struct xfs_inode		*ip,
5578 	int				whichfork,
5579 	xfs_fileoff_t			shift,		/* shift fsb */
5580 	struct xfs_iext_cursor		*icur,
5581 	struct xfs_bmbt_irec		*got,		/* extent to shift */
5582 	struct xfs_bmbt_irec		*left,		/* preceding extent */
5583 	struct xfs_btree_cur		*cur,
5584 	int				*logflags,	/* output */
5585 	struct xfs_defer_ops		*dfops)
5586 {
5587 	struct xfs_bmbt_irec		new;
5588 	xfs_filblks_t			blockcount;
5589 	int				error, i;
5590 	struct xfs_mount		*mp = ip->i_mount;
5591 
5592 	blockcount = left->br_blockcount + got->br_blockcount;
5593 
5594 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5595 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5596 	ASSERT(xfs_bmse_can_merge(left, got, shift));
5597 
5598 	new = *left;
5599 	new.br_blockcount = blockcount;
5600 
5601 	/*
5602 	 * Update the on-disk extent count, the btree if necessary and log the
5603 	 * inode.
5604 	 */
5605 	XFS_IFORK_NEXT_SET(ip, whichfork,
5606 			   XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5607 	*logflags |= XFS_ILOG_CORE;
5608 	if (!cur) {
5609 		*logflags |= XFS_ILOG_DEXT;
5610 		goto done;
5611 	}
5612 
5613 	/* lookup and remove the extent to merge */
5614 	error = xfs_bmbt_lookup_eq(cur, got, &i);
5615 	if (error)
5616 		return error;
5617 	XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5618 
5619 	error = xfs_btree_delete(cur, &i);
5620 	if (error)
5621 		return error;
5622 	XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5623 
5624 	/* lookup and update size of the previous extent */
5625 	error = xfs_bmbt_lookup_eq(cur, left, &i);
5626 	if (error)
5627 		return error;
5628 	XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5629 
5630 	error = xfs_bmbt_update(cur, &new);
5631 	if (error)
5632 		return error;
5633 
5634 done:
5635 	xfs_iext_remove(ip, icur, 0);
5636 	xfs_iext_prev(XFS_IFORK_PTR(ip, whichfork), icur);
5637 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5638 			&new);
5639 
5640 	/* update reverse mapping. rmap functions merge the rmaps for us */
5641 	error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, got);
5642 	if (error)
5643 		return error;
5644 	memcpy(&new, got, sizeof(new));
5645 	new.br_startoff = left->br_startoff + left->br_blockcount;
5646 	return xfs_rmap_map_extent(mp, dfops, ip, whichfork, &new);
5647 }
5648 
5649 static int
5650 xfs_bmap_shift_update_extent(
5651 	struct xfs_inode	*ip,
5652 	int			whichfork,
5653 	struct xfs_iext_cursor	*icur,
5654 	struct xfs_bmbt_irec	*got,
5655 	struct xfs_btree_cur	*cur,
5656 	int			*logflags,
5657 	struct xfs_defer_ops	*dfops,
5658 	xfs_fileoff_t		startoff)
5659 {
5660 	struct xfs_mount	*mp = ip->i_mount;
5661 	struct xfs_bmbt_irec	prev = *got;
5662 	int			error, i;
5663 
5664 	*logflags |= XFS_ILOG_CORE;
5665 
5666 	got->br_startoff = startoff;
5667 
5668 	if (cur) {
5669 		error = xfs_bmbt_lookup_eq(cur, &prev, &i);
5670 		if (error)
5671 			return error;
5672 		XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5673 
5674 		error = xfs_bmbt_update(cur, got);
5675 		if (error)
5676 			return error;
5677 	} else {
5678 		*logflags |= XFS_ILOG_DEXT;
5679 	}
5680 
5681 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5682 			got);
5683 
5684 	/* update reverse mapping */
5685 	error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, &prev);
5686 	if (error)
5687 		return error;
5688 	return xfs_rmap_map_extent(mp, dfops, ip, whichfork, got);
5689 }
5690 
5691 int
5692 xfs_bmap_collapse_extents(
5693 	struct xfs_trans	*tp,
5694 	struct xfs_inode	*ip,
5695 	xfs_fileoff_t		*next_fsb,
5696 	xfs_fileoff_t		offset_shift_fsb,
5697 	bool			*done,
5698 	xfs_fsblock_t		*firstblock,
5699 	struct xfs_defer_ops	*dfops)
5700 {
5701 	int			whichfork = XFS_DATA_FORK;
5702 	struct xfs_mount	*mp = ip->i_mount;
5703 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
5704 	struct xfs_btree_cur	*cur = NULL;
5705 	struct xfs_bmbt_irec	got, prev;
5706 	struct xfs_iext_cursor	icur;
5707 	xfs_fileoff_t		new_startoff;
5708 	int			error = 0;
5709 	int			logflags = 0;
5710 
5711 	if (unlikely(XFS_TEST_ERROR(
5712 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5713 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5714 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
5715 		XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
5716 		return -EFSCORRUPTED;
5717 	}
5718 
5719 	if (XFS_FORCED_SHUTDOWN(mp))
5720 		return -EIO;
5721 
5722 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL));
5723 
5724 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5725 		error = xfs_iread_extents(tp, ip, whichfork);
5726 		if (error)
5727 			return error;
5728 	}
5729 
5730 	if (ifp->if_flags & XFS_IFBROOT) {
5731 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5732 		cur->bc_private.b.firstblock = *firstblock;
5733 		cur->bc_private.b.dfops = dfops;
5734 		cur->bc_private.b.flags = 0;
5735 	}
5736 
5737 	if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5738 		*done = true;
5739 		goto del_cursor;
5740 	}
5741 	XFS_WANT_CORRUPTED_GOTO(mp, !isnullstartblock(got.br_startblock),
5742 				del_cursor);
5743 
5744 	new_startoff = got.br_startoff - offset_shift_fsb;
5745 	if (xfs_iext_peek_prev_extent(ifp, &icur, &prev)) {
5746 		if (new_startoff < prev.br_startoff + prev.br_blockcount) {
5747 			error = -EINVAL;
5748 			goto del_cursor;
5749 		}
5750 
5751 		if (xfs_bmse_can_merge(&prev, &got, offset_shift_fsb)) {
5752 			error = xfs_bmse_merge(ip, whichfork, offset_shift_fsb,
5753 					&icur, &got, &prev, cur, &logflags,
5754 					dfops);
5755 			if (error)
5756 				goto del_cursor;
5757 			goto done;
5758 		}
5759 	} else {
5760 		if (got.br_startoff < offset_shift_fsb) {
5761 			error = -EINVAL;
5762 			goto del_cursor;
5763 		}
5764 	}
5765 
5766 	error = xfs_bmap_shift_update_extent(ip, whichfork, &icur, &got, cur,
5767 			&logflags, dfops, new_startoff);
5768 	if (error)
5769 		goto del_cursor;
5770 
5771 done:
5772 	if (!xfs_iext_next_extent(ifp, &icur, &got)) {
5773 		*done = true;
5774 		goto del_cursor;
5775 	}
5776 
5777 	*next_fsb = got.br_startoff;
5778 del_cursor:
5779 	if (cur)
5780 		xfs_btree_del_cursor(cur,
5781 			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5782 	if (logflags)
5783 		xfs_trans_log_inode(tp, ip, logflags);
5784 	return error;
5785 }
5786 
5787 int
5788 xfs_bmap_insert_extents(
5789 	struct xfs_trans	*tp,
5790 	struct xfs_inode	*ip,
5791 	xfs_fileoff_t		*next_fsb,
5792 	xfs_fileoff_t		offset_shift_fsb,
5793 	bool			*done,
5794 	xfs_fileoff_t		stop_fsb,
5795 	xfs_fsblock_t		*firstblock,
5796 	struct xfs_defer_ops	*dfops)
5797 {
5798 	int			whichfork = XFS_DATA_FORK;
5799 	struct xfs_mount	*mp = ip->i_mount;
5800 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
5801 	struct xfs_btree_cur	*cur = NULL;
5802 	struct xfs_bmbt_irec	got, next;
5803 	struct xfs_iext_cursor	icur;
5804 	xfs_fileoff_t		new_startoff;
5805 	int			error = 0;
5806 	int			logflags = 0;
5807 
5808 	if (unlikely(XFS_TEST_ERROR(
5809 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5810 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5811 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
5812 		XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
5813 		return -EFSCORRUPTED;
5814 	}
5815 
5816 	if (XFS_FORCED_SHUTDOWN(mp))
5817 		return -EIO;
5818 
5819 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL));
5820 
5821 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5822 		error = xfs_iread_extents(tp, ip, whichfork);
5823 		if (error)
5824 			return error;
5825 	}
5826 
5827 	if (ifp->if_flags & XFS_IFBROOT) {
5828 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5829 		cur->bc_private.b.firstblock = *firstblock;
5830 		cur->bc_private.b.dfops = dfops;
5831 		cur->bc_private.b.flags = 0;
5832 	}
5833 
5834 	if (*next_fsb == NULLFSBLOCK) {
5835 		xfs_iext_last(ifp, &icur);
5836 		if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5837 		    stop_fsb > got.br_startoff) {
5838 			*done = true;
5839 			goto del_cursor;
5840 		}
5841 	} else {
5842 		if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5843 			*done = true;
5844 			goto del_cursor;
5845 		}
5846 	}
5847 	XFS_WANT_CORRUPTED_GOTO(mp, !isnullstartblock(got.br_startblock),
5848 				del_cursor);
5849 
5850 	if (stop_fsb >= got.br_startoff + got.br_blockcount) {
5851 		error = -EIO;
5852 		goto del_cursor;
5853 	}
5854 
5855 	new_startoff = got.br_startoff + offset_shift_fsb;
5856 	if (xfs_iext_peek_next_extent(ifp, &icur, &next)) {
5857 		if (new_startoff + got.br_blockcount > next.br_startoff) {
5858 			error = -EINVAL;
5859 			goto del_cursor;
5860 		}
5861 
5862 		/*
5863 		 * Unlike a left shift (which involves a hole punch), a right
5864 		 * shift does not modify extent neighbors in any way.  We should
5865 		 * never find mergeable extents in this scenario.  Check anyways
5866 		 * and warn if we encounter two extents that could be one.
5867 		 */
5868 		if (xfs_bmse_can_merge(&got, &next, offset_shift_fsb))
5869 			WARN_ON_ONCE(1);
5870 	}
5871 
5872 	error = xfs_bmap_shift_update_extent(ip, whichfork, &icur, &got, cur,
5873 			&logflags, dfops, new_startoff);
5874 	if (error)
5875 		goto del_cursor;
5876 
5877 	if (!xfs_iext_prev_extent(ifp, &icur, &got) ||
5878 	    stop_fsb >= got.br_startoff + got.br_blockcount) {
5879 		*done = true;
5880 		goto del_cursor;
5881 	}
5882 
5883 	*next_fsb = got.br_startoff;
5884 del_cursor:
5885 	if (cur)
5886 		xfs_btree_del_cursor(cur,
5887 			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5888 	if (logflags)
5889 		xfs_trans_log_inode(tp, ip, logflags);
5890 	return error;
5891 }
5892 
5893 /*
5894  * Splits an extent into two extents at split_fsb block such that it is the
5895  * first block of the current_ext. @ext is a target extent to be split.
5896  * @split_fsb is a block where the extents is split.  If split_fsb lies in a
5897  * hole or the first block of extents, just return 0.
5898  */
5899 STATIC int
5900 xfs_bmap_split_extent_at(
5901 	struct xfs_trans	*tp,
5902 	struct xfs_inode	*ip,
5903 	xfs_fileoff_t		split_fsb,
5904 	xfs_fsblock_t		*firstfsb,
5905 	struct xfs_defer_ops	*dfops)
5906 {
5907 	int				whichfork = XFS_DATA_FORK;
5908 	struct xfs_btree_cur		*cur = NULL;
5909 	struct xfs_bmbt_irec		got;
5910 	struct xfs_bmbt_irec		new; /* split extent */
5911 	struct xfs_mount		*mp = ip->i_mount;
5912 	struct xfs_ifork		*ifp;
5913 	xfs_fsblock_t			gotblkcnt; /* new block count for got */
5914 	struct xfs_iext_cursor		icur;
5915 	int				error = 0;
5916 	int				logflags = 0;
5917 	int				i = 0;
5918 
5919 	if (unlikely(XFS_TEST_ERROR(
5920 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5921 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5922 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
5923 		XFS_ERROR_REPORT("xfs_bmap_split_extent_at",
5924 				 XFS_ERRLEVEL_LOW, mp);
5925 		return -EFSCORRUPTED;
5926 	}
5927 
5928 	if (XFS_FORCED_SHUTDOWN(mp))
5929 		return -EIO;
5930 
5931 	ifp = XFS_IFORK_PTR(ip, whichfork);
5932 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5933 		/* Read in all the extents */
5934 		error = xfs_iread_extents(tp, ip, whichfork);
5935 		if (error)
5936 			return error;
5937 	}
5938 
5939 	/*
5940 	 * If there are not extents, or split_fsb lies in a hole we are done.
5941 	 */
5942 	if (!xfs_iext_lookup_extent(ip, ifp, split_fsb, &icur, &got) ||
5943 	    got.br_startoff >= split_fsb)
5944 		return 0;
5945 
5946 	gotblkcnt = split_fsb - got.br_startoff;
5947 	new.br_startoff = split_fsb;
5948 	new.br_startblock = got.br_startblock + gotblkcnt;
5949 	new.br_blockcount = got.br_blockcount - gotblkcnt;
5950 	new.br_state = got.br_state;
5951 
5952 	if (ifp->if_flags & XFS_IFBROOT) {
5953 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5954 		cur->bc_private.b.firstblock = *firstfsb;
5955 		cur->bc_private.b.dfops = dfops;
5956 		cur->bc_private.b.flags = 0;
5957 		error = xfs_bmbt_lookup_eq(cur, &got, &i);
5958 		if (error)
5959 			goto del_cursor;
5960 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
5961 	}
5962 
5963 	got.br_blockcount = gotblkcnt;
5964 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), &icur,
5965 			&got);
5966 
5967 	logflags = XFS_ILOG_CORE;
5968 	if (cur) {
5969 		error = xfs_bmbt_update(cur, &got);
5970 		if (error)
5971 			goto del_cursor;
5972 	} else
5973 		logflags |= XFS_ILOG_DEXT;
5974 
5975 	/* Add new extent */
5976 	xfs_iext_next(ifp, &icur);
5977 	xfs_iext_insert(ip, &icur, &new, 0);
5978 	XFS_IFORK_NEXT_SET(ip, whichfork,
5979 			   XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5980 
5981 	if (cur) {
5982 		error = xfs_bmbt_lookup_eq(cur, &new, &i);
5983 		if (error)
5984 			goto del_cursor;
5985 		XFS_WANT_CORRUPTED_GOTO(mp, i == 0, del_cursor);
5986 		error = xfs_btree_insert(cur, &i);
5987 		if (error)
5988 			goto del_cursor;
5989 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
5990 	}
5991 
5992 	/*
5993 	 * Convert to a btree if necessary.
5994 	 */
5995 	if (xfs_bmap_needs_btree(ip, whichfork)) {
5996 		int tmp_logflags; /* partial log flag return val */
5997 
5998 		ASSERT(cur == NULL);
5999 		error = xfs_bmap_extents_to_btree(tp, ip, firstfsb, dfops,
6000 				&cur, 0, &tmp_logflags, whichfork);
6001 		logflags |= tmp_logflags;
6002 	}
6003 
6004 del_cursor:
6005 	if (cur) {
6006 		cur->bc_private.b.allocated = 0;
6007 		xfs_btree_del_cursor(cur,
6008 				error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
6009 	}
6010 
6011 	if (logflags)
6012 		xfs_trans_log_inode(tp, ip, logflags);
6013 	return error;
6014 }
6015 
6016 int
6017 xfs_bmap_split_extent(
6018 	struct xfs_inode        *ip,
6019 	xfs_fileoff_t           split_fsb)
6020 {
6021 	struct xfs_mount        *mp = ip->i_mount;
6022 	struct xfs_trans        *tp;
6023 	struct xfs_defer_ops    dfops;
6024 	xfs_fsblock_t           firstfsb;
6025 	int                     error;
6026 
6027 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
6028 			XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
6029 	if (error)
6030 		return error;
6031 
6032 	xfs_ilock(ip, XFS_ILOCK_EXCL);
6033 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
6034 
6035 	xfs_defer_init(&dfops, &firstfsb);
6036 
6037 	error = xfs_bmap_split_extent_at(tp, ip, split_fsb,
6038 			&firstfsb, &dfops);
6039 	if (error)
6040 		goto out;
6041 
6042 	error = xfs_defer_finish(&tp, &dfops);
6043 	if (error)
6044 		goto out;
6045 
6046 	return xfs_trans_commit(tp);
6047 
6048 out:
6049 	xfs_defer_cancel(&dfops);
6050 	xfs_trans_cancel(tp);
6051 	return error;
6052 }
6053 
6054 /* Deferred mapping is only for real extents in the data fork. */
6055 static bool
6056 xfs_bmap_is_update_needed(
6057 	struct xfs_bmbt_irec	*bmap)
6058 {
6059 	return  bmap->br_startblock != HOLESTARTBLOCK &&
6060 		bmap->br_startblock != DELAYSTARTBLOCK;
6061 }
6062 
6063 /* Record a bmap intent. */
6064 static int
6065 __xfs_bmap_add(
6066 	struct xfs_mount		*mp,
6067 	struct xfs_defer_ops		*dfops,
6068 	enum xfs_bmap_intent_type	type,
6069 	struct xfs_inode		*ip,
6070 	int				whichfork,
6071 	struct xfs_bmbt_irec		*bmap)
6072 {
6073 	int				error;
6074 	struct xfs_bmap_intent		*bi;
6075 
6076 	trace_xfs_bmap_defer(mp,
6077 			XFS_FSB_TO_AGNO(mp, bmap->br_startblock),
6078 			type,
6079 			XFS_FSB_TO_AGBNO(mp, bmap->br_startblock),
6080 			ip->i_ino, whichfork,
6081 			bmap->br_startoff,
6082 			bmap->br_blockcount,
6083 			bmap->br_state);
6084 
6085 	bi = kmem_alloc(sizeof(struct xfs_bmap_intent), KM_SLEEP | KM_NOFS);
6086 	INIT_LIST_HEAD(&bi->bi_list);
6087 	bi->bi_type = type;
6088 	bi->bi_owner = ip;
6089 	bi->bi_whichfork = whichfork;
6090 	bi->bi_bmap = *bmap;
6091 
6092 	error = xfs_defer_ijoin(dfops, bi->bi_owner);
6093 	if (error) {
6094 		kmem_free(bi);
6095 		return error;
6096 	}
6097 
6098 	xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list);
6099 	return 0;
6100 }
6101 
6102 /* Map an extent into a file. */
6103 int
6104 xfs_bmap_map_extent(
6105 	struct xfs_mount	*mp,
6106 	struct xfs_defer_ops	*dfops,
6107 	struct xfs_inode	*ip,
6108 	struct xfs_bmbt_irec	*PREV)
6109 {
6110 	if (!xfs_bmap_is_update_needed(PREV))
6111 		return 0;
6112 
6113 	return __xfs_bmap_add(mp, dfops, XFS_BMAP_MAP, ip,
6114 			XFS_DATA_FORK, PREV);
6115 }
6116 
6117 /* Unmap an extent out of a file. */
6118 int
6119 xfs_bmap_unmap_extent(
6120 	struct xfs_mount	*mp,
6121 	struct xfs_defer_ops	*dfops,
6122 	struct xfs_inode	*ip,
6123 	struct xfs_bmbt_irec	*PREV)
6124 {
6125 	if (!xfs_bmap_is_update_needed(PREV))
6126 		return 0;
6127 
6128 	return __xfs_bmap_add(mp, dfops, XFS_BMAP_UNMAP, ip,
6129 			XFS_DATA_FORK, PREV);
6130 }
6131 
6132 /*
6133  * Process one of the deferred bmap operations.  We pass back the
6134  * btree cursor to maintain our lock on the bmapbt between calls.
6135  */
6136 int
6137 xfs_bmap_finish_one(
6138 	struct xfs_trans		*tp,
6139 	struct xfs_defer_ops		*dfops,
6140 	struct xfs_inode		*ip,
6141 	enum xfs_bmap_intent_type	type,
6142 	int				whichfork,
6143 	xfs_fileoff_t			startoff,
6144 	xfs_fsblock_t			startblock,
6145 	xfs_filblks_t			*blockcount,
6146 	xfs_exntst_t			state)
6147 {
6148 	xfs_fsblock_t			firstfsb;
6149 	int				error = 0;
6150 
6151 	/*
6152 	 * firstfsb is tied to the transaction lifetime and is used to
6153 	 * ensure correct AG locking order and schedule work item
6154 	 * continuations.  XFS_BUI_MAX_FAST_EXTENTS (== 1) restricts us
6155 	 * to only making one bmap call per transaction, so it should
6156 	 * be safe to have it as a local variable here.
6157 	 */
6158 	firstfsb = NULLFSBLOCK;
6159 
6160 	trace_xfs_bmap_deferred(tp->t_mountp,
6161 			XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type,
6162 			XFS_FSB_TO_AGBNO(tp->t_mountp, startblock),
6163 			ip->i_ino, whichfork, startoff, *blockcount, state);
6164 
6165 	if (WARN_ON_ONCE(whichfork != XFS_DATA_FORK))
6166 		return -EFSCORRUPTED;
6167 
6168 	if (XFS_TEST_ERROR(false, tp->t_mountp,
6169 			XFS_ERRTAG_BMAP_FINISH_ONE))
6170 		return -EIO;
6171 
6172 	switch (type) {
6173 	case XFS_BMAP_MAP:
6174 		error = xfs_bmapi_remap(tp, ip, startoff, *blockcount,
6175 				startblock, dfops, 0);
6176 		*blockcount = 0;
6177 		break;
6178 	case XFS_BMAP_UNMAP:
6179 		error = __xfs_bunmapi(tp, ip, startoff, blockcount,
6180 				XFS_BMAPI_REMAP, 1, &firstfsb, dfops);
6181 		break;
6182 	default:
6183 		ASSERT(0);
6184 		error = -EFSCORRUPTED;
6185 	}
6186 
6187 	return error;
6188 }
6189 
6190 /* Check that an inode's extent does not have invalid flags or bad ranges. */
6191 xfs_failaddr_t
6192 xfs_bmap_validate_extent(
6193 	struct xfs_inode	*ip,
6194 	int			whichfork,
6195 	struct xfs_bmbt_irec	*irec)
6196 {
6197 	struct xfs_mount	*mp = ip->i_mount;
6198 	xfs_fsblock_t		endfsb;
6199 	bool			isrt;
6200 
6201 	isrt = XFS_IS_REALTIME_INODE(ip);
6202 	endfsb = irec->br_startblock + irec->br_blockcount - 1;
6203 	if (isrt) {
6204 		if (!xfs_verify_rtbno(mp, irec->br_startblock))
6205 			return __this_address;
6206 		if (!xfs_verify_rtbno(mp, endfsb))
6207 			return __this_address;
6208 	} else {
6209 		if (!xfs_verify_fsbno(mp, irec->br_startblock))
6210 			return __this_address;
6211 		if (!xfs_verify_fsbno(mp, endfsb))
6212 			return __this_address;
6213 		if (XFS_FSB_TO_AGNO(mp, irec->br_startblock) !=
6214 		    XFS_FSB_TO_AGNO(mp, endfsb))
6215 			return __this_address;
6216 	}
6217 	if (irec->br_state != XFS_EXT_NORM) {
6218 		if (whichfork != XFS_DATA_FORK)
6219 			return __this_address;
6220 		if (!xfs_sb_version_hasextflgbit(&mp->m_sb))
6221 			return __this_address;
6222 	}
6223 	return NULL;
6224 }
6225