xref: /openbmc/linux/fs/xfs/libxfs/xfs_bmap.c (revision 8730046c)
1 /*
2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_bit.h"
25 #include "xfs_sb.h"
26 #include "xfs_mount.h"
27 #include "xfs_defer.h"
28 #include "xfs_da_format.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_dir2.h"
31 #include "xfs_inode.h"
32 #include "xfs_btree.h"
33 #include "xfs_trans.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_extfree_item.h"
36 #include "xfs_alloc.h"
37 #include "xfs_bmap.h"
38 #include "xfs_bmap_util.h"
39 #include "xfs_bmap_btree.h"
40 #include "xfs_rtalloc.h"
41 #include "xfs_error.h"
42 #include "xfs_quota.h"
43 #include "xfs_trans_space.h"
44 #include "xfs_buf_item.h"
45 #include "xfs_trace.h"
46 #include "xfs_symlink.h"
47 #include "xfs_attr_leaf.h"
48 #include "xfs_filestream.h"
49 #include "xfs_rmap.h"
50 #include "xfs_ag_resv.h"
51 #include "xfs_refcount.h"
52 #include "xfs_rmap_btree.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 	xfs_fileoff_t		off,
117 	xfs_fsblock_t		bno,
118 	xfs_filblks_t		len,
119 	int			*stat)	/* success/failure */
120 {
121 	cur->bc_rec.b.br_startoff = off;
122 	cur->bc_rec.b.br_startblock = bno;
123 	cur->bc_rec.b.br_blockcount = len;
124 	return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
125 }
126 
127 STATIC int				/* error */
128 xfs_bmbt_lookup_ge(
129 	struct xfs_btree_cur	*cur,
130 	xfs_fileoff_t		off,
131 	xfs_fsblock_t		bno,
132 	xfs_filblks_t		len,
133 	int			*stat)	/* success/failure */
134 {
135 	cur->bc_rec.b.br_startoff = off;
136 	cur->bc_rec.b.br_startblock = bno;
137 	cur->bc_rec.b.br_blockcount = len;
138 	return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
139 }
140 
141 /*
142  * Check if the inode needs to be converted to btree format.
143  */
144 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
145 {
146 	return whichfork != XFS_COW_FORK &&
147 		XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
148 		XFS_IFORK_NEXTENTS(ip, whichfork) >
149 			XFS_IFORK_MAXEXT(ip, whichfork);
150 }
151 
152 /*
153  * Check if the inode should be converted to extent format.
154  */
155 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
156 {
157 	return whichfork != XFS_COW_FORK &&
158 		XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
159 		XFS_IFORK_NEXTENTS(ip, whichfork) <=
160 			XFS_IFORK_MAXEXT(ip, whichfork);
161 }
162 
163 /*
164  * Update the record referred to by cur to the value given
165  * by [off, bno, len, state].
166  * This either works (return 0) or gets an EFSCORRUPTED error.
167  */
168 STATIC int
169 xfs_bmbt_update(
170 	struct xfs_btree_cur	*cur,
171 	xfs_fileoff_t		off,
172 	xfs_fsblock_t		bno,
173 	xfs_filblks_t		len,
174 	xfs_exntst_t		state)
175 {
176 	union xfs_btree_rec	rec;
177 
178 	xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
179 	return xfs_btree_update(cur, &rec);
180 }
181 
182 /*
183  * Compute the worst-case number of indirect blocks that will be used
184  * for ip's delayed extent of length "len".
185  */
186 STATIC xfs_filblks_t
187 xfs_bmap_worst_indlen(
188 	xfs_inode_t	*ip,		/* incore inode pointer */
189 	xfs_filblks_t	len)		/* delayed extent length */
190 {
191 	int		level;		/* btree level number */
192 	int		maxrecs;	/* maximum record count at this level */
193 	xfs_mount_t	*mp;		/* mount structure */
194 	xfs_filblks_t	rval;		/* return value */
195 	xfs_filblks_t   orig_len;
196 
197 	mp = ip->i_mount;
198 
199 	/* Calculate the worst-case size of the bmbt. */
200 	orig_len = len;
201 	maxrecs = mp->m_bmap_dmxr[0];
202 	for (level = 0, rval = 0;
203 	     level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
204 	     level++) {
205 		len += maxrecs - 1;
206 		do_div(len, maxrecs);
207 		rval += len;
208 		if (len == 1) {
209 			rval += XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
210 				level - 1;
211 			break;
212 		}
213 		if (level == 0)
214 			maxrecs = mp->m_bmap_dmxr[1];
215 	}
216 
217 	/* Calculate the worst-case size of the rmapbt. */
218 	if (xfs_sb_version_hasrmapbt(&mp->m_sb))
219 		rval += 1 + xfs_rmapbt_calc_size(mp, orig_len) +
220 				mp->m_rmap_maxlevels;
221 
222 	return rval;
223 }
224 
225 /*
226  * Calculate the default attribute fork offset for newly created inodes.
227  */
228 uint
229 xfs_default_attroffset(
230 	struct xfs_inode	*ip)
231 {
232 	struct xfs_mount	*mp = ip->i_mount;
233 	uint			offset;
234 
235 	if (mp->m_sb.sb_inodesize == 256) {
236 		offset = XFS_LITINO(mp, ip->i_d.di_version) -
237 				XFS_BMDR_SPACE_CALC(MINABTPTRS);
238 	} else {
239 		offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
240 	}
241 
242 	ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
243 	return offset;
244 }
245 
246 /*
247  * Helper routine to reset inode di_forkoff field when switching
248  * attribute fork from local to extent format - we reset it where
249  * possible to make space available for inline data fork extents.
250  */
251 STATIC void
252 xfs_bmap_forkoff_reset(
253 	xfs_inode_t	*ip,
254 	int		whichfork)
255 {
256 	if (whichfork == XFS_ATTR_FORK &&
257 	    ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
258 	    ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
259 	    ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
260 		uint	dfl_forkoff = xfs_default_attroffset(ip) >> 3;
261 
262 		if (dfl_forkoff > ip->i_d.di_forkoff)
263 			ip->i_d.di_forkoff = dfl_forkoff;
264 	}
265 }
266 
267 #ifdef DEBUG
268 STATIC struct xfs_buf *
269 xfs_bmap_get_bp(
270 	struct xfs_btree_cur	*cur,
271 	xfs_fsblock_t		bno)
272 {
273 	struct xfs_log_item_desc *lidp;
274 	int			i;
275 
276 	if (!cur)
277 		return NULL;
278 
279 	for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
280 		if (!cur->bc_bufs[i])
281 			break;
282 		if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
283 			return cur->bc_bufs[i];
284 	}
285 
286 	/* Chase down all the log items to see if the bp is there */
287 	list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
288 		struct xfs_buf_log_item	*bip;
289 		bip = (struct xfs_buf_log_item *)lidp->lid_item;
290 		if (bip->bli_item.li_type == XFS_LI_BUF &&
291 		    XFS_BUF_ADDR(bip->bli_buf) == bno)
292 			return bip->bli_buf;
293 	}
294 
295 	return NULL;
296 }
297 
298 STATIC void
299 xfs_check_block(
300 	struct xfs_btree_block	*block,
301 	xfs_mount_t		*mp,
302 	int			root,
303 	short			sz)
304 {
305 	int			i, j, dmxr;
306 	__be64			*pp, *thispa;	/* pointer to block address */
307 	xfs_bmbt_key_t		*prevp, *keyp;
308 
309 	ASSERT(be16_to_cpu(block->bb_level) > 0);
310 
311 	prevp = NULL;
312 	for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
313 		dmxr = mp->m_bmap_dmxr[0];
314 		keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
315 
316 		if (prevp) {
317 			ASSERT(be64_to_cpu(prevp->br_startoff) <
318 			       be64_to_cpu(keyp->br_startoff));
319 		}
320 		prevp = keyp;
321 
322 		/*
323 		 * Compare the block numbers to see if there are dups.
324 		 */
325 		if (root)
326 			pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
327 		else
328 			pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
329 
330 		for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
331 			if (root)
332 				thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
333 			else
334 				thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
335 			if (*thispa == *pp) {
336 				xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
337 					__func__, j, i,
338 					(unsigned long long)be64_to_cpu(*thispa));
339 				panic("%s: ptrs are equal in node\n",
340 					__func__);
341 			}
342 		}
343 	}
344 }
345 
346 /*
347  * Check that the extents for the inode ip are in the right order in all
348  * btree leaves. THis becomes prohibitively expensive for large extent count
349  * files, so don't bother with inodes that have more than 10,000 extents in
350  * them. The btree record ordering checks will still be done, so for such large
351  * bmapbt constructs that is going to catch most corruptions.
352  */
353 STATIC void
354 xfs_bmap_check_leaf_extents(
355 	xfs_btree_cur_t		*cur,	/* btree cursor or null */
356 	xfs_inode_t		*ip,		/* incore inode pointer */
357 	int			whichfork)	/* data or attr fork */
358 {
359 	struct xfs_btree_block	*block;	/* current btree block */
360 	xfs_fsblock_t		bno;	/* block # of "block" */
361 	xfs_buf_t		*bp;	/* buffer for "block" */
362 	int			error;	/* error return value */
363 	xfs_extnum_t		i=0, j;	/* index into the extents list */
364 	xfs_ifork_t		*ifp;	/* fork structure */
365 	int			level;	/* btree level, for checking */
366 	xfs_mount_t		*mp;	/* file system mount structure */
367 	__be64			*pp;	/* pointer to block address */
368 	xfs_bmbt_rec_t		*ep;	/* pointer to current extent */
369 	xfs_bmbt_rec_t		last = {0, 0}; /* last extent in prev block */
370 	xfs_bmbt_rec_t		*nextp;	/* pointer to next extent */
371 	int			bp_release = 0;
372 
373 	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
374 		return;
375 	}
376 
377 	/* skip large extent count inodes */
378 	if (ip->i_d.di_nextents > 10000)
379 		return;
380 
381 	bno = NULLFSBLOCK;
382 	mp = ip->i_mount;
383 	ifp = XFS_IFORK_PTR(ip, whichfork);
384 	block = ifp->if_broot;
385 	/*
386 	 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
387 	 */
388 	level = be16_to_cpu(block->bb_level);
389 	ASSERT(level > 0);
390 	xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
391 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
392 	bno = be64_to_cpu(*pp);
393 
394 	ASSERT(bno != NULLFSBLOCK);
395 	ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
396 	ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
397 
398 	/*
399 	 * Go down the tree until leaf level is reached, following the first
400 	 * pointer (leftmost) at each level.
401 	 */
402 	while (level-- > 0) {
403 		/* See if buf is in cur first */
404 		bp_release = 0;
405 		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
406 		if (!bp) {
407 			bp_release = 1;
408 			error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
409 						XFS_BMAP_BTREE_REF,
410 						&xfs_bmbt_buf_ops);
411 			if (error)
412 				goto error_norelse;
413 		}
414 		block = XFS_BUF_TO_BLOCK(bp);
415 		if (level == 0)
416 			break;
417 
418 		/*
419 		 * Check this block for basic sanity (increasing keys and
420 		 * no duplicate blocks).
421 		 */
422 
423 		xfs_check_block(block, mp, 0, 0);
424 		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
425 		bno = be64_to_cpu(*pp);
426 		XFS_WANT_CORRUPTED_GOTO(mp,
427 					XFS_FSB_SANITY_CHECK(mp, bno), error0);
428 		if (bp_release) {
429 			bp_release = 0;
430 			xfs_trans_brelse(NULL, bp);
431 		}
432 	}
433 
434 	/*
435 	 * Here with bp and block set to the leftmost leaf node in the tree.
436 	 */
437 	i = 0;
438 
439 	/*
440 	 * Loop over all leaf nodes checking that all extents are in the right order.
441 	 */
442 	for (;;) {
443 		xfs_fsblock_t	nextbno;
444 		xfs_extnum_t	num_recs;
445 
446 
447 		num_recs = xfs_btree_get_numrecs(block);
448 
449 		/*
450 		 * Read-ahead the next leaf block, if any.
451 		 */
452 
453 		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
454 
455 		/*
456 		 * Check all the extents to make sure they are OK.
457 		 * If we had a previous block, the last entry should
458 		 * conform with the first entry in this one.
459 		 */
460 
461 		ep = XFS_BMBT_REC_ADDR(mp, block, 1);
462 		if (i) {
463 			ASSERT(xfs_bmbt_disk_get_startoff(&last) +
464 			       xfs_bmbt_disk_get_blockcount(&last) <=
465 			       xfs_bmbt_disk_get_startoff(ep));
466 		}
467 		for (j = 1; j < num_recs; j++) {
468 			nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
469 			ASSERT(xfs_bmbt_disk_get_startoff(ep) +
470 			       xfs_bmbt_disk_get_blockcount(ep) <=
471 			       xfs_bmbt_disk_get_startoff(nextp));
472 			ep = nextp;
473 		}
474 
475 		last = *ep;
476 		i += num_recs;
477 		if (bp_release) {
478 			bp_release = 0;
479 			xfs_trans_brelse(NULL, bp);
480 		}
481 		bno = nextbno;
482 		/*
483 		 * If we've reached the end, stop.
484 		 */
485 		if (bno == NULLFSBLOCK)
486 			break;
487 
488 		bp_release = 0;
489 		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
490 		if (!bp) {
491 			bp_release = 1;
492 			error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
493 						XFS_BMAP_BTREE_REF,
494 						&xfs_bmbt_buf_ops);
495 			if (error)
496 				goto error_norelse;
497 		}
498 		block = XFS_BUF_TO_BLOCK(bp);
499 	}
500 
501 	return;
502 
503 error0:
504 	xfs_warn(mp, "%s: at error0", __func__);
505 	if (bp_release)
506 		xfs_trans_brelse(NULL, bp);
507 error_norelse:
508 	xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
509 		__func__, i);
510 	panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
511 	return;
512 }
513 
514 /*
515  * Add bmap trace insert entries for all the contents of the extent records.
516  */
517 void
518 xfs_bmap_trace_exlist(
519 	xfs_inode_t	*ip,		/* incore inode pointer */
520 	xfs_extnum_t	cnt,		/* count of entries in the list */
521 	int		whichfork,	/* data or attr or cow fork */
522 	unsigned long	caller_ip)
523 {
524 	xfs_extnum_t	idx;		/* extent record index */
525 	xfs_ifork_t	*ifp;		/* inode fork pointer */
526 	int		state = 0;
527 
528 	if (whichfork == XFS_ATTR_FORK)
529 		state |= BMAP_ATTRFORK;
530 	else if (whichfork == XFS_COW_FORK)
531 		state |= BMAP_COWFORK;
532 
533 	ifp = XFS_IFORK_PTR(ip, whichfork);
534 	ASSERT(cnt == xfs_iext_count(ifp));
535 	for (idx = 0; idx < cnt; idx++)
536 		trace_xfs_extlist(ip, idx, state, caller_ip);
537 }
538 
539 /*
540  * Validate that the bmbt_irecs being returned from bmapi are valid
541  * given the caller's original parameters.  Specifically check the
542  * ranges of the returned irecs to ensure that they only extend beyond
543  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
544  */
545 STATIC void
546 xfs_bmap_validate_ret(
547 	xfs_fileoff_t		bno,
548 	xfs_filblks_t		len,
549 	int			flags,
550 	xfs_bmbt_irec_t		*mval,
551 	int			nmap,
552 	int			ret_nmap)
553 {
554 	int			i;		/* index to map values */
555 
556 	ASSERT(ret_nmap <= nmap);
557 
558 	for (i = 0; i < ret_nmap; i++) {
559 		ASSERT(mval[i].br_blockcount > 0);
560 		if (!(flags & XFS_BMAPI_ENTIRE)) {
561 			ASSERT(mval[i].br_startoff >= bno);
562 			ASSERT(mval[i].br_blockcount <= len);
563 			ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
564 			       bno + len);
565 		} else {
566 			ASSERT(mval[i].br_startoff < bno + len);
567 			ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
568 			       bno);
569 		}
570 		ASSERT(i == 0 ||
571 		       mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
572 		       mval[i].br_startoff);
573 		ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
574 		       mval[i].br_startblock != HOLESTARTBLOCK);
575 		ASSERT(mval[i].br_state == XFS_EXT_NORM ||
576 		       mval[i].br_state == XFS_EXT_UNWRITTEN);
577 	}
578 }
579 
580 #else
581 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)		do { } while (0)
582 #define	xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
583 #endif /* DEBUG */
584 
585 /*
586  * bmap free list manipulation functions
587  */
588 
589 /*
590  * Add the extent to the list of extents to be free at transaction end.
591  * The list is maintained sorted (by block number).
592  */
593 void
594 xfs_bmap_add_free(
595 	struct xfs_mount		*mp,
596 	struct xfs_defer_ops		*dfops,
597 	xfs_fsblock_t			bno,
598 	xfs_filblks_t			len,
599 	struct xfs_owner_info		*oinfo)
600 {
601 	struct xfs_extent_free_item	*new;		/* new element */
602 #ifdef DEBUG
603 	xfs_agnumber_t		agno;
604 	xfs_agblock_t		agbno;
605 
606 	ASSERT(bno != NULLFSBLOCK);
607 	ASSERT(len > 0);
608 	ASSERT(len <= MAXEXTLEN);
609 	ASSERT(!isnullstartblock(bno));
610 	agno = XFS_FSB_TO_AGNO(mp, bno);
611 	agbno = XFS_FSB_TO_AGBNO(mp, bno);
612 	ASSERT(agno < mp->m_sb.sb_agcount);
613 	ASSERT(agbno < mp->m_sb.sb_agblocks);
614 	ASSERT(len < mp->m_sb.sb_agblocks);
615 	ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
616 #endif
617 	ASSERT(xfs_bmap_free_item_zone != NULL);
618 
619 	new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
620 	new->xefi_startblock = bno;
621 	new->xefi_blockcount = (xfs_extlen_t)len;
622 	if (oinfo)
623 		new->xefi_oinfo = *oinfo;
624 	else
625 		xfs_rmap_skip_owner_update(&new->xefi_oinfo);
626 	trace_xfs_bmap_free_defer(mp, XFS_FSB_TO_AGNO(mp, bno), 0,
627 			XFS_FSB_TO_AGBNO(mp, bno), len);
628 	xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list);
629 }
630 
631 /*
632  * Inode fork format manipulation functions
633  */
634 
635 /*
636  * Transform a btree format file with only one leaf node, where the
637  * extents list will fit in the inode, into an extents format file.
638  * Since the file extents are already in-core, all we have to do is
639  * give up the space for the btree root and pitch the leaf block.
640  */
641 STATIC int				/* error */
642 xfs_bmap_btree_to_extents(
643 	xfs_trans_t		*tp,	/* transaction pointer */
644 	xfs_inode_t		*ip,	/* incore inode pointer */
645 	xfs_btree_cur_t		*cur,	/* btree cursor */
646 	int			*logflagsp, /* inode logging flags */
647 	int			whichfork)  /* data or attr fork */
648 {
649 	/* REFERENCED */
650 	struct xfs_btree_block	*cblock;/* child btree block */
651 	xfs_fsblock_t		cbno;	/* child block number */
652 	xfs_buf_t		*cbp;	/* child block's buffer */
653 	int			error;	/* error return value */
654 	xfs_ifork_t		*ifp;	/* inode fork data */
655 	xfs_mount_t		*mp;	/* mount point structure */
656 	__be64			*pp;	/* ptr to block address */
657 	struct xfs_btree_block	*rblock;/* root btree block */
658 	struct xfs_owner_info	oinfo;
659 
660 	mp = ip->i_mount;
661 	ifp = XFS_IFORK_PTR(ip, whichfork);
662 	ASSERT(whichfork != XFS_COW_FORK);
663 	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
664 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
665 	rblock = ifp->if_broot;
666 	ASSERT(be16_to_cpu(rblock->bb_level) == 1);
667 	ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
668 	ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
669 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
670 	cbno = be64_to_cpu(*pp);
671 	*logflagsp = 0;
672 #ifdef DEBUG
673 	if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
674 		return error;
675 #endif
676 	error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
677 				&xfs_bmbt_buf_ops);
678 	if (error)
679 		return error;
680 	cblock = XFS_BUF_TO_BLOCK(cbp);
681 	if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
682 		return error;
683 	xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
684 	xfs_bmap_add_free(mp, cur->bc_private.b.dfops, cbno, 1, &oinfo);
685 	ip->i_d.di_nblocks--;
686 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
687 	xfs_trans_binval(tp, cbp);
688 	if (cur->bc_bufs[0] == cbp)
689 		cur->bc_bufs[0] = NULL;
690 	xfs_iroot_realloc(ip, -1, whichfork);
691 	ASSERT(ifp->if_broot == NULL);
692 	ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
693 	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
694 	*logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
695 	return 0;
696 }
697 
698 /*
699  * Convert an extents-format file into a btree-format file.
700  * The new file will have a root block (in the inode) and a single child block.
701  */
702 STATIC int					/* error */
703 xfs_bmap_extents_to_btree(
704 	xfs_trans_t		*tp,		/* transaction pointer */
705 	xfs_inode_t		*ip,		/* incore inode pointer */
706 	xfs_fsblock_t		*firstblock,	/* first-block-allocated */
707 	struct xfs_defer_ops	*dfops,		/* blocks freed in xaction */
708 	xfs_btree_cur_t		**curp,		/* cursor returned to caller */
709 	int			wasdel,		/* converting a delayed alloc */
710 	int			*logflagsp,	/* inode logging flags */
711 	int			whichfork)	/* data or attr fork */
712 {
713 	struct xfs_btree_block	*ablock;	/* allocated (child) bt block */
714 	xfs_buf_t		*abp;		/* buffer for ablock */
715 	xfs_alloc_arg_t		args;		/* allocation arguments */
716 	xfs_bmbt_rec_t		*arp;		/* child record pointer */
717 	struct xfs_btree_block	*block;		/* btree root block */
718 	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
719 	xfs_bmbt_rec_host_t	*ep;		/* extent record pointer */
720 	int			error;		/* error return value */
721 	xfs_extnum_t		i, cnt;		/* extent record index */
722 	xfs_ifork_t		*ifp;		/* inode fork pointer */
723 	xfs_bmbt_key_t		*kp;		/* root block key pointer */
724 	xfs_mount_t		*mp;		/* mount structure */
725 	xfs_extnum_t		nextents;	/* number of file extents */
726 	xfs_bmbt_ptr_t		*pp;		/* root block address pointer */
727 
728 	mp = ip->i_mount;
729 	ASSERT(whichfork != XFS_COW_FORK);
730 	ifp = XFS_IFORK_PTR(ip, whichfork);
731 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
732 
733 	/*
734 	 * Make space in the inode incore.
735 	 */
736 	xfs_iroot_realloc(ip, 1, whichfork);
737 	ifp->if_flags |= XFS_IFBROOT;
738 
739 	/*
740 	 * Fill in the root.
741 	 */
742 	block = ifp->if_broot;
743 	if (xfs_sb_version_hascrc(&mp->m_sb))
744 		xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
745 				 XFS_BMAP_CRC_MAGIC, 1, 1, ip->i_ino,
746 				 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
747 	else
748 		xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
749 				 XFS_BMAP_MAGIC, 1, 1, ip->i_ino,
750 				 XFS_BTREE_LONG_PTRS);
751 
752 	/*
753 	 * Need a cursor.  Can't allocate until bb_level is filled in.
754 	 */
755 	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
756 	cur->bc_private.b.firstblock = *firstblock;
757 	cur->bc_private.b.dfops = dfops;
758 	cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
759 	/*
760 	 * Convert to a btree with two levels, one record in root.
761 	 */
762 	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
763 	memset(&args, 0, sizeof(args));
764 	args.tp = tp;
765 	args.mp = mp;
766 	xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork);
767 	args.firstblock = *firstblock;
768 	if (*firstblock == NULLFSBLOCK) {
769 		args.type = XFS_ALLOCTYPE_START_BNO;
770 		args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
771 	} else if (dfops->dop_low) {
772 try_another_ag:
773 		args.type = XFS_ALLOCTYPE_START_BNO;
774 		args.fsbno = *firstblock;
775 	} else {
776 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
777 		args.fsbno = *firstblock;
778 	}
779 	args.minlen = args.maxlen = args.prod = 1;
780 	args.wasdel = wasdel;
781 	*logflagsp = 0;
782 	if ((error = xfs_alloc_vextent(&args))) {
783 		xfs_iroot_realloc(ip, -1, whichfork);
784 		xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
785 		return error;
786 	}
787 
788 	/*
789 	 * During a CoW operation, the allocation and bmbt updates occur in
790 	 * different transactions.  The mapping code tries to put new bmbt
791 	 * blocks near extents being mapped, but the only way to guarantee this
792 	 * is if the alloc and the mapping happen in a single transaction that
793 	 * has a block reservation.  That isn't the case here, so if we run out
794 	 * of space we'll try again with another AG.
795 	 */
796 	if (xfs_sb_version_hasreflink(&cur->bc_mp->m_sb) &&
797 	    args.fsbno == NULLFSBLOCK &&
798 	    args.type == XFS_ALLOCTYPE_NEAR_BNO) {
799 		dfops->dop_low = true;
800 		goto try_another_ag;
801 	}
802 	/*
803 	 * Allocation can't fail, the space was reserved.
804 	 */
805 	ASSERT(args.fsbno != NULLFSBLOCK);
806 	ASSERT(*firstblock == NULLFSBLOCK ||
807 	       args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
808 	       (dfops->dop_low &&
809 		args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
810 	*firstblock = cur->bc_private.b.firstblock = args.fsbno;
811 	cur->bc_private.b.allocated++;
812 	ip->i_d.di_nblocks++;
813 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
814 	abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
815 	/*
816 	 * Fill in the child block.
817 	 */
818 	abp->b_ops = &xfs_bmbt_buf_ops;
819 	ablock = XFS_BUF_TO_BLOCK(abp);
820 	if (xfs_sb_version_hascrc(&mp->m_sb))
821 		xfs_btree_init_block_int(mp, ablock, abp->b_bn,
822 				XFS_BMAP_CRC_MAGIC, 0, 0, ip->i_ino,
823 				XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
824 	else
825 		xfs_btree_init_block_int(mp, ablock, abp->b_bn,
826 				XFS_BMAP_MAGIC, 0, 0, ip->i_ino,
827 				XFS_BTREE_LONG_PTRS);
828 
829 	arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
830 	nextents =  xfs_iext_count(ifp);
831 	for (cnt = i = 0; i < nextents; i++) {
832 		ep = xfs_iext_get_ext(ifp, i);
833 		if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
834 			arp->l0 = cpu_to_be64(ep->l0);
835 			arp->l1 = cpu_to_be64(ep->l1);
836 			arp++; cnt++;
837 		}
838 	}
839 	ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
840 	xfs_btree_set_numrecs(ablock, cnt);
841 
842 	/*
843 	 * Fill in the root key and pointer.
844 	 */
845 	kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
846 	arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
847 	kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
848 	pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
849 						be16_to_cpu(block->bb_level)));
850 	*pp = cpu_to_be64(args.fsbno);
851 
852 	/*
853 	 * Do all this logging at the end so that
854 	 * the root is at the right level.
855 	 */
856 	xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
857 	xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
858 	ASSERT(*curp == NULL);
859 	*curp = cur;
860 	*logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
861 	return 0;
862 }
863 
864 /*
865  * Convert a local file to an extents file.
866  * This code is out of bounds for data forks of regular files,
867  * since the file data needs to get logged so things will stay consistent.
868  * (The bmap-level manipulations are ok, though).
869  */
870 void
871 xfs_bmap_local_to_extents_empty(
872 	struct xfs_inode	*ip,
873 	int			whichfork)
874 {
875 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
876 
877 	ASSERT(whichfork != XFS_COW_FORK);
878 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
879 	ASSERT(ifp->if_bytes == 0);
880 	ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
881 
882 	xfs_bmap_forkoff_reset(ip, whichfork);
883 	ifp->if_flags &= ~XFS_IFINLINE;
884 	ifp->if_flags |= XFS_IFEXTENTS;
885 	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
886 }
887 
888 
889 STATIC int				/* error */
890 xfs_bmap_local_to_extents(
891 	xfs_trans_t	*tp,		/* transaction pointer */
892 	xfs_inode_t	*ip,		/* incore inode pointer */
893 	xfs_fsblock_t	*firstblock,	/* first block allocated in xaction */
894 	xfs_extlen_t	total,		/* total blocks needed by transaction */
895 	int		*logflagsp,	/* inode logging flags */
896 	int		whichfork,
897 	void		(*init_fn)(struct xfs_trans *tp,
898 				   struct xfs_buf *bp,
899 				   struct xfs_inode *ip,
900 				   struct xfs_ifork *ifp))
901 {
902 	int		error = 0;
903 	int		flags;		/* logging flags returned */
904 	xfs_ifork_t	*ifp;		/* inode fork pointer */
905 	xfs_alloc_arg_t	args;		/* allocation arguments */
906 	xfs_buf_t	*bp;		/* buffer for extent block */
907 	xfs_bmbt_rec_host_t *ep;	/* extent record pointer */
908 
909 	/*
910 	 * We don't want to deal with the case of keeping inode data inline yet.
911 	 * So sending the data fork of a regular inode is invalid.
912 	 */
913 	ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
914 	ifp = XFS_IFORK_PTR(ip, whichfork);
915 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
916 
917 	if (!ifp->if_bytes) {
918 		xfs_bmap_local_to_extents_empty(ip, whichfork);
919 		flags = XFS_ILOG_CORE;
920 		goto done;
921 	}
922 
923 	flags = 0;
924 	error = 0;
925 	ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) ==
926 								XFS_IFINLINE);
927 	memset(&args, 0, sizeof(args));
928 	args.tp = tp;
929 	args.mp = ip->i_mount;
930 	xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0);
931 	args.firstblock = *firstblock;
932 	/*
933 	 * Allocate a block.  We know we need only one, since the
934 	 * file currently fits in an inode.
935 	 */
936 	if (*firstblock == NULLFSBLOCK) {
937 try_another_ag:
938 		args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
939 		args.type = XFS_ALLOCTYPE_START_BNO;
940 	} else {
941 		args.fsbno = *firstblock;
942 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
943 	}
944 	args.total = total;
945 	args.minlen = args.maxlen = args.prod = 1;
946 	error = xfs_alloc_vextent(&args);
947 	if (error)
948 		goto done;
949 
950 	/*
951 	 * During a CoW operation, the allocation and bmbt updates occur in
952 	 * different transactions.  The mapping code tries to put new bmbt
953 	 * blocks near extents being mapped, but the only way to guarantee this
954 	 * is if the alloc and the mapping happen in a single transaction that
955 	 * has a block reservation.  That isn't the case here, so if we run out
956 	 * of space we'll try again with another AG.
957 	 */
958 	if (xfs_sb_version_hasreflink(&ip->i_mount->m_sb) &&
959 	    args.fsbno == NULLFSBLOCK &&
960 	    args.type == XFS_ALLOCTYPE_NEAR_BNO) {
961 		goto try_another_ag;
962 	}
963 	/* Can't fail, the space was reserved. */
964 	ASSERT(args.fsbno != NULLFSBLOCK);
965 	ASSERT(args.len == 1);
966 	*firstblock = args.fsbno;
967 	bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
968 
969 	/*
970 	 * Initialize the block, copy the data and log the remote buffer.
971 	 *
972 	 * The callout is responsible for logging because the remote format
973 	 * might differ from the local format and thus we don't know how much to
974 	 * log here. Note that init_fn must also set the buffer log item type
975 	 * correctly.
976 	 */
977 	init_fn(tp, bp, ip, ifp);
978 
979 	/* account for the change in fork size */
980 	xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
981 	xfs_bmap_local_to_extents_empty(ip, whichfork);
982 	flags |= XFS_ILOG_CORE;
983 
984 	xfs_iext_add(ifp, 0, 1);
985 	ep = xfs_iext_get_ext(ifp, 0);
986 	xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
987 	trace_xfs_bmap_post_update(ip, 0,
988 			whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
989 			_THIS_IP_);
990 	XFS_IFORK_NEXT_SET(ip, whichfork, 1);
991 	ip->i_d.di_nblocks = 1;
992 	xfs_trans_mod_dquot_byino(tp, ip,
993 		XFS_TRANS_DQ_BCOUNT, 1L);
994 	flags |= xfs_ilog_fext(whichfork);
995 
996 done:
997 	*logflagsp = flags;
998 	return error;
999 }
1000 
1001 /*
1002  * Called from xfs_bmap_add_attrfork to handle btree format files.
1003  */
1004 STATIC int					/* error */
1005 xfs_bmap_add_attrfork_btree(
1006 	xfs_trans_t		*tp,		/* transaction pointer */
1007 	xfs_inode_t		*ip,		/* incore inode pointer */
1008 	xfs_fsblock_t		*firstblock,	/* first block allocated */
1009 	struct xfs_defer_ops	*dfops,		/* blocks to free at commit */
1010 	int			*flags)		/* inode logging flags */
1011 {
1012 	xfs_btree_cur_t		*cur;		/* btree cursor */
1013 	int			error;		/* error return value */
1014 	xfs_mount_t		*mp;		/* file system mount struct */
1015 	int			stat;		/* newroot status */
1016 
1017 	mp = ip->i_mount;
1018 	if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
1019 		*flags |= XFS_ILOG_DBROOT;
1020 	else {
1021 		cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
1022 		cur->bc_private.b.dfops = dfops;
1023 		cur->bc_private.b.firstblock = *firstblock;
1024 		if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
1025 			goto error0;
1026 		/* must be at least one entry */
1027 		XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0);
1028 		if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
1029 			goto error0;
1030 		if (stat == 0) {
1031 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1032 			return -ENOSPC;
1033 		}
1034 		*firstblock = cur->bc_private.b.firstblock;
1035 		cur->bc_private.b.allocated = 0;
1036 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1037 	}
1038 	return 0;
1039 error0:
1040 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1041 	return error;
1042 }
1043 
1044 /*
1045  * Called from xfs_bmap_add_attrfork to handle extents format files.
1046  */
1047 STATIC int					/* error */
1048 xfs_bmap_add_attrfork_extents(
1049 	xfs_trans_t		*tp,		/* transaction pointer */
1050 	xfs_inode_t		*ip,		/* incore inode pointer */
1051 	xfs_fsblock_t		*firstblock,	/* first block allocated */
1052 	struct xfs_defer_ops	*dfops,		/* blocks to free at commit */
1053 	int			*flags)		/* inode logging flags */
1054 {
1055 	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
1056 	int			error;		/* error return value */
1057 
1058 	if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1059 		return 0;
1060 	cur = NULL;
1061 	error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops, &cur, 0,
1062 		flags, XFS_DATA_FORK);
1063 	if (cur) {
1064 		cur->bc_private.b.allocated = 0;
1065 		xfs_btree_del_cursor(cur,
1066 			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1067 	}
1068 	return error;
1069 }
1070 
1071 /*
1072  * Called from xfs_bmap_add_attrfork to handle local format files. Each
1073  * different data fork content type needs a different callout to do the
1074  * conversion. Some are basic and only require special block initialisation
1075  * callouts for the data formating, others (directories) are so specialised they
1076  * handle everything themselves.
1077  *
1078  * XXX (dgc): investigate whether directory conversion can use the generic
1079  * formatting callout. It should be possible - it's just a very complex
1080  * formatter.
1081  */
1082 STATIC int					/* error */
1083 xfs_bmap_add_attrfork_local(
1084 	xfs_trans_t		*tp,		/* transaction pointer */
1085 	xfs_inode_t		*ip,		/* incore inode pointer */
1086 	xfs_fsblock_t		*firstblock,	/* first block allocated */
1087 	struct xfs_defer_ops	*dfops,		/* blocks to free at commit */
1088 	int			*flags)		/* inode logging flags */
1089 {
1090 	xfs_da_args_t		dargs;		/* args for dir/attr code */
1091 
1092 	if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1093 		return 0;
1094 
1095 	if (S_ISDIR(VFS_I(ip)->i_mode)) {
1096 		memset(&dargs, 0, sizeof(dargs));
1097 		dargs.geo = ip->i_mount->m_dir_geo;
1098 		dargs.dp = ip;
1099 		dargs.firstblock = firstblock;
1100 		dargs.dfops = dfops;
1101 		dargs.total = dargs.geo->fsbcount;
1102 		dargs.whichfork = XFS_DATA_FORK;
1103 		dargs.trans = tp;
1104 		return xfs_dir2_sf_to_block(&dargs);
1105 	}
1106 
1107 	if (S_ISLNK(VFS_I(ip)->i_mode))
1108 		return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1109 						 flags, XFS_DATA_FORK,
1110 						 xfs_symlink_local_to_remote);
1111 
1112 	/* should only be called for types that support local format data */
1113 	ASSERT(0);
1114 	return -EFSCORRUPTED;
1115 }
1116 
1117 /*
1118  * Convert inode from non-attributed to attributed.
1119  * Must not be in a transaction, ip must not be locked.
1120  */
1121 int						/* error code */
1122 xfs_bmap_add_attrfork(
1123 	xfs_inode_t		*ip,		/* incore inode pointer */
1124 	int			size,		/* space new attribute needs */
1125 	int			rsvd)		/* xact may use reserved blks */
1126 {
1127 	xfs_fsblock_t		firstblock;	/* 1st block/ag allocated */
1128 	struct xfs_defer_ops	dfops;		/* freed extent records */
1129 	xfs_mount_t		*mp;		/* mount structure */
1130 	xfs_trans_t		*tp;		/* transaction pointer */
1131 	int			blks;		/* space reservation */
1132 	int			version = 1;	/* superblock attr version */
1133 	int			logflags;	/* logging flags */
1134 	int			error;		/* error return value */
1135 
1136 	ASSERT(XFS_IFORK_Q(ip) == 0);
1137 
1138 	mp = ip->i_mount;
1139 	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1140 
1141 	blks = XFS_ADDAFORK_SPACE_RES(mp);
1142 
1143 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_addafork, blks, 0,
1144 			rsvd ? XFS_TRANS_RESERVE : 0, &tp);
1145 	if (error)
1146 		return error;
1147 
1148 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1149 	error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1150 			XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1151 			XFS_QMOPT_RES_REGBLKS);
1152 	if (error)
1153 		goto trans_cancel;
1154 	if (XFS_IFORK_Q(ip))
1155 		goto trans_cancel;
1156 	if (ip->i_d.di_anextents != 0) {
1157 		error = -EFSCORRUPTED;
1158 		goto trans_cancel;
1159 	}
1160 	if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1161 		/*
1162 		 * For inodes coming from pre-6.2 filesystems.
1163 		 */
1164 		ASSERT(ip->i_d.di_aformat == 0);
1165 		ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1166 	}
1167 
1168 	xfs_trans_ijoin(tp, ip, 0);
1169 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1170 
1171 	switch (ip->i_d.di_format) {
1172 	case XFS_DINODE_FMT_DEV:
1173 		ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1174 		break;
1175 	case XFS_DINODE_FMT_UUID:
1176 		ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1177 		break;
1178 	case XFS_DINODE_FMT_LOCAL:
1179 	case XFS_DINODE_FMT_EXTENTS:
1180 	case XFS_DINODE_FMT_BTREE:
1181 		ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1182 		if (!ip->i_d.di_forkoff)
1183 			ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1184 		else if (mp->m_flags & XFS_MOUNT_ATTR2)
1185 			version = 2;
1186 		break;
1187 	default:
1188 		ASSERT(0);
1189 		error = -EINVAL;
1190 		goto trans_cancel;
1191 	}
1192 
1193 	ASSERT(ip->i_afp == NULL);
1194 	ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1195 	ip->i_afp->if_flags = XFS_IFEXTENTS;
1196 	logflags = 0;
1197 	xfs_defer_init(&dfops, &firstblock);
1198 	switch (ip->i_d.di_format) {
1199 	case XFS_DINODE_FMT_LOCAL:
1200 		error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &dfops,
1201 			&logflags);
1202 		break;
1203 	case XFS_DINODE_FMT_EXTENTS:
1204 		error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1205 			&dfops, &logflags);
1206 		break;
1207 	case XFS_DINODE_FMT_BTREE:
1208 		error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &dfops,
1209 			&logflags);
1210 		break;
1211 	default:
1212 		error = 0;
1213 		break;
1214 	}
1215 	if (logflags)
1216 		xfs_trans_log_inode(tp, ip, logflags);
1217 	if (error)
1218 		goto bmap_cancel;
1219 	if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1220 	   (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1221 		bool log_sb = false;
1222 
1223 		spin_lock(&mp->m_sb_lock);
1224 		if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1225 			xfs_sb_version_addattr(&mp->m_sb);
1226 			log_sb = true;
1227 		}
1228 		if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1229 			xfs_sb_version_addattr2(&mp->m_sb);
1230 			log_sb = true;
1231 		}
1232 		spin_unlock(&mp->m_sb_lock);
1233 		if (log_sb)
1234 			xfs_log_sb(tp);
1235 	}
1236 
1237 	error = xfs_defer_finish(&tp, &dfops, NULL);
1238 	if (error)
1239 		goto bmap_cancel;
1240 	error = xfs_trans_commit(tp);
1241 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1242 	return error;
1243 
1244 bmap_cancel:
1245 	xfs_defer_cancel(&dfops);
1246 trans_cancel:
1247 	xfs_trans_cancel(tp);
1248 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1249 	return error;
1250 }
1251 
1252 /*
1253  * Internal and external extent tree search functions.
1254  */
1255 
1256 /*
1257  * Read in the extents to if_extents.
1258  * All inode fields are set up by caller, we just traverse the btree
1259  * and copy the records in. If the file system cannot contain unwritten
1260  * extents, the records are checked for no "state" flags.
1261  */
1262 int					/* error */
1263 xfs_bmap_read_extents(
1264 	xfs_trans_t		*tp,	/* transaction pointer */
1265 	xfs_inode_t		*ip,	/* incore inode */
1266 	int			whichfork) /* data or attr fork */
1267 {
1268 	struct xfs_btree_block	*block;	/* current btree block */
1269 	xfs_fsblock_t		bno;	/* block # of "block" */
1270 	xfs_buf_t		*bp;	/* buffer for "block" */
1271 	int			error;	/* error return value */
1272 	xfs_exntfmt_t		exntf;	/* XFS_EXTFMT_NOSTATE, if checking */
1273 	xfs_extnum_t		i, j;	/* index into the extents list */
1274 	xfs_ifork_t		*ifp;	/* fork structure */
1275 	int			level;	/* btree level, for checking */
1276 	xfs_mount_t		*mp;	/* file system mount structure */
1277 	__be64			*pp;	/* pointer to block address */
1278 	/* REFERENCED */
1279 	xfs_extnum_t		room;	/* number of entries there's room for */
1280 
1281 	bno = NULLFSBLOCK;
1282 	mp = ip->i_mount;
1283 	ifp = XFS_IFORK_PTR(ip, whichfork);
1284 	exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
1285 					XFS_EXTFMT_INODE(ip);
1286 	block = ifp->if_broot;
1287 	/*
1288 	 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1289 	 */
1290 	level = be16_to_cpu(block->bb_level);
1291 	ASSERT(level > 0);
1292 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1293 	bno = be64_to_cpu(*pp);
1294 	ASSERT(bno != NULLFSBLOCK);
1295 	ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
1296 	ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
1297 	/*
1298 	 * Go down the tree until leaf level is reached, following the first
1299 	 * pointer (leftmost) at each level.
1300 	 */
1301 	while (level-- > 0) {
1302 		error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1303 				XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1304 		if (error)
1305 			return error;
1306 		block = XFS_BUF_TO_BLOCK(bp);
1307 		if (level == 0)
1308 			break;
1309 		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1310 		bno = be64_to_cpu(*pp);
1311 		XFS_WANT_CORRUPTED_GOTO(mp,
1312 			XFS_FSB_SANITY_CHECK(mp, bno), error0);
1313 		xfs_trans_brelse(tp, bp);
1314 	}
1315 	/*
1316 	 * Here with bp and block set to the leftmost leaf node in the tree.
1317 	 */
1318 	room = xfs_iext_count(ifp);
1319 	i = 0;
1320 	/*
1321 	 * Loop over all leaf nodes.  Copy information to the extent records.
1322 	 */
1323 	for (;;) {
1324 		xfs_bmbt_rec_t	*frp;
1325 		xfs_fsblock_t	nextbno;
1326 		xfs_extnum_t	num_recs;
1327 		xfs_extnum_t	start;
1328 
1329 		num_recs = xfs_btree_get_numrecs(block);
1330 		if (unlikely(i + num_recs > room)) {
1331 			ASSERT(i + num_recs <= room);
1332 			xfs_warn(ip->i_mount,
1333 				"corrupt dinode %Lu, (btree extents).",
1334 				(unsigned long long) ip->i_ino);
1335 			XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1336 				XFS_ERRLEVEL_LOW, ip->i_mount, block);
1337 			goto error0;
1338 		}
1339 		/*
1340 		 * Read-ahead the next leaf block, if any.
1341 		 */
1342 		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1343 		if (nextbno != NULLFSBLOCK)
1344 			xfs_btree_reada_bufl(mp, nextbno, 1,
1345 					     &xfs_bmbt_buf_ops);
1346 		/*
1347 		 * Copy records into the extent records.
1348 		 */
1349 		frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1350 		start = i;
1351 		for (j = 0; j < num_recs; j++, i++, frp++) {
1352 			xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1353 			trp->l0 = be64_to_cpu(frp->l0);
1354 			trp->l1 = be64_to_cpu(frp->l1);
1355 		}
1356 		if (exntf == XFS_EXTFMT_NOSTATE) {
1357 			/*
1358 			 * Check all attribute bmap btree records and
1359 			 * any "older" data bmap btree records for a
1360 			 * set bit in the "extent flag" position.
1361 			 */
1362 			if (unlikely(xfs_check_nostate_extents(ifp,
1363 					start, num_recs))) {
1364 				XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1365 						 XFS_ERRLEVEL_LOW,
1366 						 ip->i_mount);
1367 				goto error0;
1368 			}
1369 		}
1370 		xfs_trans_brelse(tp, bp);
1371 		bno = nextbno;
1372 		/*
1373 		 * If we've reached the end, stop.
1374 		 */
1375 		if (bno == NULLFSBLOCK)
1376 			break;
1377 		error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1378 				XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1379 		if (error)
1380 			return error;
1381 		block = XFS_BUF_TO_BLOCK(bp);
1382 	}
1383 	if (i != XFS_IFORK_NEXTENTS(ip, whichfork))
1384 		return -EFSCORRUPTED;
1385 	ASSERT(i == xfs_iext_count(ifp));
1386 	XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1387 	return 0;
1388 error0:
1389 	xfs_trans_brelse(tp, bp);
1390 	return -EFSCORRUPTED;
1391 }
1392 
1393 /*
1394  * Returns the file-relative block number of the first unused block(s)
1395  * in the file with at least "len" logically contiguous blocks free.
1396  * This is the lowest-address hole if the file has holes, else the first block
1397  * past the end of file.
1398  * Return 0 if the file is currently local (in-inode).
1399  */
1400 int						/* error */
1401 xfs_bmap_first_unused(
1402 	xfs_trans_t	*tp,			/* transaction pointer */
1403 	xfs_inode_t	*ip,			/* incore inode */
1404 	xfs_extlen_t	len,			/* size of hole to find */
1405 	xfs_fileoff_t	*first_unused,		/* unused block */
1406 	int		whichfork)		/* data or attr fork */
1407 {
1408 	int		error;			/* error return value */
1409 	int		idx;			/* extent record index */
1410 	xfs_ifork_t	*ifp;			/* inode fork pointer */
1411 	xfs_fileoff_t	lastaddr;		/* last block number seen */
1412 	xfs_fileoff_t	lowest;			/* lowest useful block */
1413 	xfs_fileoff_t	max;			/* starting useful block */
1414 	xfs_fileoff_t	off;			/* offset for this block */
1415 	xfs_extnum_t	nextents;		/* number of extent entries */
1416 
1417 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1418 	       XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1419 	       XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1420 	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1421 		*first_unused = 0;
1422 		return 0;
1423 	}
1424 	ifp = XFS_IFORK_PTR(ip, whichfork);
1425 	if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1426 	    (error = xfs_iread_extents(tp, ip, whichfork)))
1427 		return error;
1428 	lowest = *first_unused;
1429 	nextents = xfs_iext_count(ifp);
1430 	for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1431 		xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
1432 		off = xfs_bmbt_get_startoff(ep);
1433 		/*
1434 		 * See if the hole before this extent will work.
1435 		 */
1436 		if (off >= lowest + len && off - max >= len) {
1437 			*first_unused = max;
1438 			return 0;
1439 		}
1440 		lastaddr = off + xfs_bmbt_get_blockcount(ep);
1441 		max = XFS_FILEOFF_MAX(lastaddr, lowest);
1442 	}
1443 	*first_unused = max;
1444 	return 0;
1445 }
1446 
1447 /*
1448  * Returns the file-relative block number of the last block - 1 before
1449  * last_block (input value) in the file.
1450  * This is not based on i_size, it is based on the extent records.
1451  * Returns 0 for local files, as they do not have extent records.
1452  */
1453 int						/* error */
1454 xfs_bmap_last_before(
1455 	struct xfs_trans	*tp,		/* transaction pointer */
1456 	struct xfs_inode	*ip,		/* incore inode */
1457 	xfs_fileoff_t		*last_block,	/* last block */
1458 	int			whichfork)	/* data or attr fork */
1459 {
1460 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1461 	struct xfs_bmbt_irec	got;
1462 	xfs_extnum_t		idx;
1463 	int			error;
1464 
1465 	switch (XFS_IFORK_FORMAT(ip, whichfork)) {
1466 	case XFS_DINODE_FMT_LOCAL:
1467 		*last_block = 0;
1468 		return 0;
1469 	case XFS_DINODE_FMT_BTREE:
1470 	case XFS_DINODE_FMT_EXTENTS:
1471 		break;
1472 	default:
1473 		return -EIO;
1474 	}
1475 
1476 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1477 		error = xfs_iread_extents(tp, ip, whichfork);
1478 		if (error)
1479 			return error;
1480 	}
1481 
1482 	if (xfs_iext_lookup_extent(ip, ifp, *last_block - 1, &idx, &got)) {
1483 		if (got.br_startoff <= *last_block - 1)
1484 			return 0;
1485 	}
1486 
1487 	if (xfs_iext_get_extent(ifp, idx - 1, &got)) {
1488 		*last_block = got.br_startoff + got.br_blockcount;
1489 		return 0;
1490 	}
1491 
1492 	*last_block = 0;
1493 	return 0;
1494 }
1495 
1496 int
1497 xfs_bmap_last_extent(
1498 	struct xfs_trans	*tp,
1499 	struct xfs_inode	*ip,
1500 	int			whichfork,
1501 	struct xfs_bmbt_irec	*rec,
1502 	int			*is_empty)
1503 {
1504 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1505 	int			error;
1506 	int			nextents;
1507 
1508 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1509 		error = xfs_iread_extents(tp, ip, whichfork);
1510 		if (error)
1511 			return error;
1512 	}
1513 
1514 	nextents = xfs_iext_count(ifp);
1515 	if (nextents == 0) {
1516 		*is_empty = 1;
1517 		return 0;
1518 	}
1519 
1520 	xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1521 	*is_empty = 0;
1522 	return 0;
1523 }
1524 
1525 /*
1526  * Check the last inode extent to determine whether this allocation will result
1527  * in blocks being allocated at the end of the file. When we allocate new data
1528  * blocks at the end of the file which do not start at the previous data block,
1529  * we will try to align the new blocks at stripe unit boundaries.
1530  *
1531  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1532  * at, or past the EOF.
1533  */
1534 STATIC int
1535 xfs_bmap_isaeof(
1536 	struct xfs_bmalloca	*bma,
1537 	int			whichfork)
1538 {
1539 	struct xfs_bmbt_irec	rec;
1540 	int			is_empty;
1541 	int			error;
1542 
1543 	bma->aeof = 0;
1544 	error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1545 				     &is_empty);
1546 	if (error)
1547 		return error;
1548 
1549 	if (is_empty) {
1550 		bma->aeof = 1;
1551 		return 0;
1552 	}
1553 
1554 	/*
1555 	 * Check if we are allocation or past the last extent, or at least into
1556 	 * the last delayed allocated extent.
1557 	 */
1558 	bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1559 		(bma->offset >= rec.br_startoff &&
1560 		 isnullstartblock(rec.br_startblock));
1561 	return 0;
1562 }
1563 
1564 /*
1565  * Returns the file-relative block number of the first block past eof in
1566  * the file.  This is not based on i_size, it is based on the extent records.
1567  * Returns 0 for local files, as they do not have extent records.
1568  */
1569 int
1570 xfs_bmap_last_offset(
1571 	struct xfs_inode	*ip,
1572 	xfs_fileoff_t		*last_block,
1573 	int			whichfork)
1574 {
1575 	struct xfs_bmbt_irec	rec;
1576 	int			is_empty;
1577 	int			error;
1578 
1579 	*last_block = 0;
1580 
1581 	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1582 		return 0;
1583 
1584 	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1585 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1586 	       return -EIO;
1587 
1588 	error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1589 	if (error || is_empty)
1590 		return error;
1591 
1592 	*last_block = rec.br_startoff + rec.br_blockcount;
1593 	return 0;
1594 }
1595 
1596 /*
1597  * Returns whether the selected fork of the inode has exactly one
1598  * block or not.  For the data fork we check this matches di_size,
1599  * implying the file's range is 0..bsize-1.
1600  */
1601 int					/* 1=>1 block, 0=>otherwise */
1602 xfs_bmap_one_block(
1603 	xfs_inode_t	*ip,		/* incore inode */
1604 	int		whichfork)	/* data or attr fork */
1605 {
1606 	xfs_bmbt_rec_host_t *ep;	/* ptr to fork's extent */
1607 	xfs_ifork_t	*ifp;		/* inode fork pointer */
1608 	int		rval;		/* return value */
1609 	xfs_bmbt_irec_t	s;		/* internal version of extent */
1610 
1611 #ifndef DEBUG
1612 	if (whichfork == XFS_DATA_FORK)
1613 		return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1614 #endif	/* !DEBUG */
1615 	if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1616 		return 0;
1617 	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1618 		return 0;
1619 	ifp = XFS_IFORK_PTR(ip, whichfork);
1620 	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1621 	ep = xfs_iext_get_ext(ifp, 0);
1622 	xfs_bmbt_get_all(ep, &s);
1623 	rval = s.br_startoff == 0 && s.br_blockcount == 1;
1624 	if (rval && whichfork == XFS_DATA_FORK)
1625 		ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1626 	return rval;
1627 }
1628 
1629 /*
1630  * Extent tree manipulation functions used during allocation.
1631  */
1632 
1633 /*
1634  * Convert a delayed allocation to a real allocation.
1635  */
1636 STATIC int				/* error */
1637 xfs_bmap_add_extent_delay_real(
1638 	struct xfs_bmalloca	*bma,
1639 	int			whichfork)
1640 {
1641 	struct xfs_bmbt_irec	*new = &bma->got;
1642 	int			diff;	/* temp value */
1643 	xfs_bmbt_rec_host_t	*ep;	/* extent entry for idx */
1644 	int			error;	/* error return value */
1645 	int			i;	/* temp state */
1646 	xfs_ifork_t		*ifp;	/* inode fork pointer */
1647 	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
1648 	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
1649 					/* left is 0, right is 1, prev is 2 */
1650 	int			rval=0;	/* return value (logging flags) */
1651 	int			state = 0;/* state bits, accessed thru macros */
1652 	xfs_filblks_t		da_new; /* new count del alloc blocks used */
1653 	xfs_filblks_t		da_old; /* old count del alloc blocks used */
1654 	xfs_filblks_t		temp=0;	/* value for da_new calculations */
1655 	xfs_filblks_t		temp2=0;/* value for da_new calculations */
1656 	int			tmp_rval;	/* partial logging flags */
1657 	struct xfs_mount	*mp;
1658 	xfs_extnum_t		*nextents;
1659 
1660 	mp = bma->ip->i_mount;
1661 	ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1662 	ASSERT(whichfork != XFS_ATTR_FORK);
1663 	nextents = (whichfork == XFS_COW_FORK ? &bma->ip->i_cnextents :
1664 						&bma->ip->i_d.di_nextents);
1665 
1666 	ASSERT(bma->idx >= 0);
1667 	ASSERT(bma->idx <= xfs_iext_count(ifp));
1668 	ASSERT(!isnullstartblock(new->br_startblock));
1669 	ASSERT(!bma->cur ||
1670 	       (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1671 
1672 	XFS_STATS_INC(mp, xs_add_exlist);
1673 
1674 #define	LEFT		r[0]
1675 #define	RIGHT		r[1]
1676 #define	PREV		r[2]
1677 
1678 	if (whichfork == XFS_COW_FORK)
1679 		state |= BMAP_COWFORK;
1680 
1681 	/*
1682 	 * Set up a bunch of variables to make the tests simpler.
1683 	 */
1684 	ep = xfs_iext_get_ext(ifp, bma->idx);
1685 	xfs_bmbt_get_all(ep, &PREV);
1686 	new_endoff = new->br_startoff + new->br_blockcount;
1687 	ASSERT(PREV.br_startoff <= new->br_startoff);
1688 	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1689 
1690 	da_old = startblockval(PREV.br_startblock);
1691 	da_new = 0;
1692 
1693 	/*
1694 	 * Set flags determining what part of the previous delayed allocation
1695 	 * extent is being replaced by a real allocation.
1696 	 */
1697 	if (PREV.br_startoff == new->br_startoff)
1698 		state |= BMAP_LEFT_FILLING;
1699 	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1700 		state |= BMAP_RIGHT_FILLING;
1701 
1702 	/*
1703 	 * Check and set flags if this segment has a left neighbor.
1704 	 * Don't set contiguous if the combined extent would be too large.
1705 	 */
1706 	if (bma->idx > 0) {
1707 		state |= BMAP_LEFT_VALID;
1708 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
1709 
1710 		if (isnullstartblock(LEFT.br_startblock))
1711 			state |= BMAP_LEFT_DELAY;
1712 	}
1713 
1714 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1715 	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1716 	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1717 	    LEFT.br_state == new->br_state &&
1718 	    LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1719 		state |= BMAP_LEFT_CONTIG;
1720 
1721 	/*
1722 	 * Check and set flags if this segment has a right neighbor.
1723 	 * Don't set contiguous if the combined extent would be too large.
1724 	 * Also check for all-three-contiguous being too large.
1725 	 */
1726 	if (bma->idx < xfs_iext_count(ifp) - 1) {
1727 		state |= BMAP_RIGHT_VALID;
1728 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
1729 
1730 		if (isnullstartblock(RIGHT.br_startblock))
1731 			state |= BMAP_RIGHT_DELAY;
1732 	}
1733 
1734 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1735 	    new_endoff == RIGHT.br_startoff &&
1736 	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1737 	    new->br_state == RIGHT.br_state &&
1738 	    new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1739 	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1740 		       BMAP_RIGHT_FILLING)) !=
1741 		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1742 		       BMAP_RIGHT_FILLING) ||
1743 	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1744 			<= MAXEXTLEN))
1745 		state |= BMAP_RIGHT_CONTIG;
1746 
1747 	error = 0;
1748 	/*
1749 	 * Switch out based on the FILLING and CONTIG state bits.
1750 	 */
1751 	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1752 			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1753 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1754 	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1755 		/*
1756 		 * Filling in all of a previously delayed allocation extent.
1757 		 * The left and right neighbors are both contiguous with new.
1758 		 */
1759 		bma->idx--;
1760 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1761 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1762 			LEFT.br_blockcount + PREV.br_blockcount +
1763 			RIGHT.br_blockcount);
1764 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1765 
1766 		xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1767 		(*nextents)--;
1768 		if (bma->cur == NULL)
1769 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1770 		else {
1771 			rval = XFS_ILOG_CORE;
1772 			error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1773 					RIGHT.br_startblock,
1774 					RIGHT.br_blockcount, &i);
1775 			if (error)
1776 				goto done;
1777 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1778 			error = xfs_btree_delete(bma->cur, &i);
1779 			if (error)
1780 				goto done;
1781 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1782 			error = xfs_btree_decrement(bma->cur, 0, &i);
1783 			if (error)
1784 				goto done;
1785 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1786 			error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1787 					LEFT.br_startblock,
1788 					LEFT.br_blockcount +
1789 					PREV.br_blockcount +
1790 					RIGHT.br_blockcount, LEFT.br_state);
1791 			if (error)
1792 				goto done;
1793 		}
1794 		break;
1795 
1796 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1797 		/*
1798 		 * Filling in all of a previously delayed allocation extent.
1799 		 * The left neighbor is contiguous, the right is not.
1800 		 */
1801 		bma->idx--;
1802 
1803 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1804 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1805 			LEFT.br_blockcount + PREV.br_blockcount);
1806 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1807 
1808 		xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1809 		if (bma->cur == NULL)
1810 			rval = XFS_ILOG_DEXT;
1811 		else {
1812 			rval = 0;
1813 			error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1814 					LEFT.br_startblock, LEFT.br_blockcount,
1815 					&i);
1816 			if (error)
1817 				goto done;
1818 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1819 			error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1820 					LEFT.br_startblock,
1821 					LEFT.br_blockcount +
1822 					PREV.br_blockcount, LEFT.br_state);
1823 			if (error)
1824 				goto done;
1825 		}
1826 		break;
1827 
1828 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1829 		/*
1830 		 * Filling in all of a previously delayed allocation extent.
1831 		 * The right neighbor is contiguous, the left is not.
1832 		 */
1833 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1834 		xfs_bmbt_set_startblock(ep, new->br_startblock);
1835 		xfs_bmbt_set_blockcount(ep,
1836 			PREV.br_blockcount + RIGHT.br_blockcount);
1837 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1838 
1839 		xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1840 		if (bma->cur == NULL)
1841 			rval = XFS_ILOG_DEXT;
1842 		else {
1843 			rval = 0;
1844 			error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1845 					RIGHT.br_startblock,
1846 					RIGHT.br_blockcount, &i);
1847 			if (error)
1848 				goto done;
1849 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1850 			error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1851 					new->br_startblock,
1852 					PREV.br_blockcount +
1853 					RIGHT.br_blockcount, PREV.br_state);
1854 			if (error)
1855 				goto done;
1856 		}
1857 		break;
1858 
1859 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1860 		/*
1861 		 * Filling in all of a previously delayed allocation extent.
1862 		 * Neither the left nor right neighbors are contiguous with
1863 		 * the new one.
1864 		 */
1865 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1866 		xfs_bmbt_set_startblock(ep, new->br_startblock);
1867 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1868 
1869 		(*nextents)++;
1870 		if (bma->cur == NULL)
1871 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1872 		else {
1873 			rval = XFS_ILOG_CORE;
1874 			error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1875 					new->br_startblock, new->br_blockcount,
1876 					&i);
1877 			if (error)
1878 				goto done;
1879 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1880 			bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1881 			error = xfs_btree_insert(bma->cur, &i);
1882 			if (error)
1883 				goto done;
1884 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1885 		}
1886 		break;
1887 
1888 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1889 		/*
1890 		 * Filling in the first part of a previous delayed allocation.
1891 		 * The left neighbor is contiguous.
1892 		 */
1893 		trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1894 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1895 			LEFT.br_blockcount + new->br_blockcount);
1896 		xfs_bmbt_set_startoff(ep,
1897 			PREV.br_startoff + new->br_blockcount);
1898 		trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1899 
1900 		temp = PREV.br_blockcount - new->br_blockcount;
1901 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1902 		xfs_bmbt_set_blockcount(ep, temp);
1903 		if (bma->cur == NULL)
1904 			rval = XFS_ILOG_DEXT;
1905 		else {
1906 			rval = 0;
1907 			error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1908 					LEFT.br_startblock, LEFT.br_blockcount,
1909 					&i);
1910 			if (error)
1911 				goto done;
1912 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1913 			error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1914 					LEFT.br_startblock,
1915 					LEFT.br_blockcount +
1916 					new->br_blockcount,
1917 					LEFT.br_state);
1918 			if (error)
1919 				goto done;
1920 		}
1921 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1922 			startblockval(PREV.br_startblock));
1923 		xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1924 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1925 
1926 		bma->idx--;
1927 		break;
1928 
1929 	case BMAP_LEFT_FILLING:
1930 		/*
1931 		 * Filling in the first part of a previous delayed allocation.
1932 		 * The left neighbor is not contiguous.
1933 		 */
1934 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1935 		xfs_bmbt_set_startoff(ep, new_endoff);
1936 		temp = PREV.br_blockcount - new->br_blockcount;
1937 		xfs_bmbt_set_blockcount(ep, temp);
1938 		xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
1939 		(*nextents)++;
1940 		if (bma->cur == NULL)
1941 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1942 		else {
1943 			rval = XFS_ILOG_CORE;
1944 			error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1945 					new->br_startblock, new->br_blockcount,
1946 					&i);
1947 			if (error)
1948 				goto done;
1949 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1950 			bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1951 			error = xfs_btree_insert(bma->cur, &i);
1952 			if (error)
1953 				goto done;
1954 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1955 		}
1956 
1957 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1958 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1959 					bma->firstblock, bma->dfops,
1960 					&bma->cur, 1, &tmp_rval, whichfork);
1961 			rval |= tmp_rval;
1962 			if (error)
1963 				goto done;
1964 		}
1965 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1966 			startblockval(PREV.br_startblock) -
1967 			(bma->cur ? bma->cur->bc_private.b.allocated : 0));
1968 		ep = xfs_iext_get_ext(ifp, bma->idx + 1);
1969 		xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1970 		trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1971 		break;
1972 
1973 	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1974 		/*
1975 		 * Filling in the last part of a previous delayed allocation.
1976 		 * The right neighbor is contiguous with the new allocation.
1977 		 */
1978 		temp = PREV.br_blockcount - new->br_blockcount;
1979 		trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1980 		xfs_bmbt_set_blockcount(ep, temp);
1981 		xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
1982 			new->br_startoff, new->br_startblock,
1983 			new->br_blockcount + RIGHT.br_blockcount,
1984 			RIGHT.br_state);
1985 		trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1986 		if (bma->cur == NULL)
1987 			rval = XFS_ILOG_DEXT;
1988 		else {
1989 			rval = 0;
1990 			error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1991 					RIGHT.br_startblock,
1992 					RIGHT.br_blockcount, &i);
1993 			if (error)
1994 				goto done;
1995 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1996 			error = xfs_bmbt_update(bma->cur, new->br_startoff,
1997 					new->br_startblock,
1998 					new->br_blockcount +
1999 					RIGHT.br_blockcount,
2000 					RIGHT.br_state);
2001 			if (error)
2002 				goto done;
2003 		}
2004 
2005 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2006 			startblockval(PREV.br_startblock));
2007 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2008 		xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2009 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2010 
2011 		bma->idx++;
2012 		break;
2013 
2014 	case BMAP_RIGHT_FILLING:
2015 		/*
2016 		 * Filling in the last part of a previous delayed allocation.
2017 		 * The right neighbor is not contiguous.
2018 		 */
2019 		temp = PREV.br_blockcount - new->br_blockcount;
2020 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2021 		xfs_bmbt_set_blockcount(ep, temp);
2022 		xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
2023 		(*nextents)++;
2024 		if (bma->cur == NULL)
2025 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2026 		else {
2027 			rval = XFS_ILOG_CORE;
2028 			error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2029 					new->br_startblock, new->br_blockcount,
2030 					&i);
2031 			if (error)
2032 				goto done;
2033 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2034 			bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2035 			error = xfs_btree_insert(bma->cur, &i);
2036 			if (error)
2037 				goto done;
2038 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2039 		}
2040 
2041 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2042 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2043 				bma->firstblock, bma->dfops, &bma->cur, 1,
2044 				&tmp_rval, whichfork);
2045 			rval |= tmp_rval;
2046 			if (error)
2047 				goto done;
2048 		}
2049 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2050 			startblockval(PREV.br_startblock) -
2051 			(bma->cur ? bma->cur->bc_private.b.allocated : 0));
2052 		ep = xfs_iext_get_ext(ifp, bma->idx);
2053 		xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2054 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2055 
2056 		bma->idx++;
2057 		break;
2058 
2059 	case 0:
2060 		/*
2061 		 * Filling in the middle part of a previous delayed allocation.
2062 		 * Contiguity is impossible here.
2063 		 * This case is avoided almost all the time.
2064 		 *
2065 		 * We start with a delayed allocation:
2066 		 *
2067 		 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2068 		 *  PREV @ idx
2069 		 *
2070 	         * and we are allocating:
2071 		 *                     +rrrrrrrrrrrrrrrrr+
2072 		 *			      new
2073 		 *
2074 		 * and we set it up for insertion as:
2075 		 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2076 		 *                            new
2077 		 *  PREV @ idx          LEFT              RIGHT
2078 		 *                      inserted at idx + 1
2079 		 */
2080 		temp = new->br_startoff - PREV.br_startoff;
2081 		temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2082 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2083 		xfs_bmbt_set_blockcount(ep, temp);	/* truncate PREV */
2084 		LEFT = *new;
2085 		RIGHT.br_state = PREV.br_state;
2086 		RIGHT.br_startblock = nullstartblock(
2087 				(int)xfs_bmap_worst_indlen(bma->ip, temp2));
2088 		RIGHT.br_startoff = new_endoff;
2089 		RIGHT.br_blockcount = temp2;
2090 		/* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2091 		xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2092 		(*nextents)++;
2093 		if (bma->cur == NULL)
2094 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2095 		else {
2096 			rval = XFS_ILOG_CORE;
2097 			error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2098 					new->br_startblock, new->br_blockcount,
2099 					&i);
2100 			if (error)
2101 				goto done;
2102 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2103 			bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2104 			error = xfs_btree_insert(bma->cur, &i);
2105 			if (error)
2106 				goto done;
2107 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2108 		}
2109 
2110 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2111 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2112 					bma->firstblock, bma->dfops, &bma->cur,
2113 					1, &tmp_rval, whichfork);
2114 			rval |= tmp_rval;
2115 			if (error)
2116 				goto done;
2117 		}
2118 		temp = xfs_bmap_worst_indlen(bma->ip, temp);
2119 		temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2120 		diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
2121 			(bma->cur ? bma->cur->bc_private.b.allocated : 0));
2122 		if (diff > 0) {
2123 			error = xfs_mod_fdblocks(bma->ip->i_mount,
2124 						 -((int64_t)diff), false);
2125 			ASSERT(!error);
2126 			if (error)
2127 				goto done;
2128 		}
2129 
2130 		ep = xfs_iext_get_ext(ifp, bma->idx);
2131 		xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2132 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2133 		trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2134 		xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2135 			nullstartblock((int)temp2));
2136 		trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2137 
2138 		bma->idx++;
2139 		da_new = temp + temp2;
2140 		break;
2141 
2142 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2143 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2144 	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2145 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2146 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2147 	case BMAP_LEFT_CONTIG:
2148 	case BMAP_RIGHT_CONTIG:
2149 		/*
2150 		 * These cases are all impossible.
2151 		 */
2152 		ASSERT(0);
2153 	}
2154 
2155 	/* add reverse mapping */
2156 	error = xfs_rmap_map_extent(mp, bma->dfops, bma->ip, whichfork, new);
2157 	if (error)
2158 		goto done;
2159 
2160 	/* convert to a btree if necessary */
2161 	if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2162 		int	tmp_logflags;	/* partial log flag return val */
2163 
2164 		ASSERT(bma->cur == NULL);
2165 		error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2166 				bma->firstblock, bma->dfops, &bma->cur,
2167 				da_old > 0, &tmp_logflags, whichfork);
2168 		bma->logflags |= tmp_logflags;
2169 		if (error)
2170 			goto done;
2171 	}
2172 
2173 	/* adjust for changes in reserved delayed indirect blocks */
2174 	if (da_old || da_new) {
2175 		temp = da_new;
2176 		if (bma->cur)
2177 			temp += bma->cur->bc_private.b.allocated;
2178 		ASSERT(temp <= da_old);
2179 		if (temp < da_old)
2180 			xfs_mod_fdblocks(bma->ip->i_mount,
2181 					(int64_t)(da_old - temp), false);
2182 	}
2183 
2184 	/* clear out the allocated field, done with it now in any case. */
2185 	if (bma->cur)
2186 		bma->cur->bc_private.b.allocated = 0;
2187 
2188 	xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
2189 done:
2190 	if (whichfork != XFS_COW_FORK)
2191 		bma->logflags |= rval;
2192 	return error;
2193 #undef	LEFT
2194 #undef	RIGHT
2195 #undef	PREV
2196 }
2197 
2198 /*
2199  * Convert an unwritten allocation to a real allocation or vice versa.
2200  */
2201 STATIC int				/* error */
2202 xfs_bmap_add_extent_unwritten_real(
2203 	struct xfs_trans	*tp,
2204 	xfs_inode_t		*ip,	/* incore inode pointer */
2205 	xfs_extnum_t		*idx,	/* extent number to update/insert */
2206 	xfs_btree_cur_t		**curp,	/* if *curp is null, not a btree */
2207 	xfs_bmbt_irec_t		*new,	/* new data to add to file extents */
2208 	xfs_fsblock_t		*first,	/* pointer to firstblock variable */
2209 	struct xfs_defer_ops	*dfops,	/* list of extents to be freed */
2210 	int			*logflagsp) /* inode logging flags */
2211 {
2212 	xfs_btree_cur_t		*cur;	/* btree cursor */
2213 	xfs_bmbt_rec_host_t	*ep;	/* extent entry for idx */
2214 	int			error;	/* error return value */
2215 	int			i;	/* temp state */
2216 	xfs_ifork_t		*ifp;	/* inode fork pointer */
2217 	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
2218 	xfs_exntst_t		newext;	/* new extent state */
2219 	xfs_exntst_t		oldext;	/* old extent state */
2220 	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
2221 					/* left is 0, right is 1, prev is 2 */
2222 	int			rval=0;	/* return value (logging flags) */
2223 	int			state = 0;/* state bits, accessed thru macros */
2224 	struct xfs_mount	*mp = tp->t_mountp;
2225 
2226 	*logflagsp = 0;
2227 
2228 	cur = *curp;
2229 	ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2230 
2231 	ASSERT(*idx >= 0);
2232 	ASSERT(*idx <= xfs_iext_count(ifp));
2233 	ASSERT(!isnullstartblock(new->br_startblock));
2234 
2235 	XFS_STATS_INC(mp, xs_add_exlist);
2236 
2237 #define	LEFT		r[0]
2238 #define	RIGHT		r[1]
2239 #define	PREV		r[2]
2240 
2241 	/*
2242 	 * Set up a bunch of variables to make the tests simpler.
2243 	 */
2244 	error = 0;
2245 	ep = xfs_iext_get_ext(ifp, *idx);
2246 	xfs_bmbt_get_all(ep, &PREV);
2247 	newext = new->br_state;
2248 	oldext = (newext == XFS_EXT_UNWRITTEN) ?
2249 		XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2250 	ASSERT(PREV.br_state == oldext);
2251 	new_endoff = new->br_startoff + new->br_blockcount;
2252 	ASSERT(PREV.br_startoff <= new->br_startoff);
2253 	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2254 
2255 	/*
2256 	 * Set flags determining what part of the previous oldext allocation
2257 	 * extent is being replaced by a newext allocation.
2258 	 */
2259 	if (PREV.br_startoff == new->br_startoff)
2260 		state |= BMAP_LEFT_FILLING;
2261 	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2262 		state |= BMAP_RIGHT_FILLING;
2263 
2264 	/*
2265 	 * Check and set flags if this segment has a left neighbor.
2266 	 * Don't set contiguous if the combined extent would be too large.
2267 	 */
2268 	if (*idx > 0) {
2269 		state |= BMAP_LEFT_VALID;
2270 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2271 
2272 		if (isnullstartblock(LEFT.br_startblock))
2273 			state |= BMAP_LEFT_DELAY;
2274 	}
2275 
2276 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2277 	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2278 	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2279 	    LEFT.br_state == newext &&
2280 	    LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2281 		state |= BMAP_LEFT_CONTIG;
2282 
2283 	/*
2284 	 * Check and set flags if this segment has a right neighbor.
2285 	 * Don't set contiguous if the combined extent would be too large.
2286 	 * Also check for all-three-contiguous being too large.
2287 	 */
2288 	if (*idx < xfs_iext_count(&ip->i_df) - 1) {
2289 		state |= BMAP_RIGHT_VALID;
2290 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2291 		if (isnullstartblock(RIGHT.br_startblock))
2292 			state |= BMAP_RIGHT_DELAY;
2293 	}
2294 
2295 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2296 	    new_endoff == RIGHT.br_startoff &&
2297 	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2298 	    newext == RIGHT.br_state &&
2299 	    new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2300 	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2301 		       BMAP_RIGHT_FILLING)) !=
2302 		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2303 		       BMAP_RIGHT_FILLING) ||
2304 	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2305 			<= MAXEXTLEN))
2306 		state |= BMAP_RIGHT_CONTIG;
2307 
2308 	/*
2309 	 * Switch out based on the FILLING and CONTIG state bits.
2310 	 */
2311 	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2312 			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2313 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2314 	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2315 		/*
2316 		 * Setting all of a previous oldext extent to newext.
2317 		 * The left and right neighbors are both contiguous with new.
2318 		 */
2319 		--*idx;
2320 
2321 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2322 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2323 			LEFT.br_blockcount + PREV.br_blockcount +
2324 			RIGHT.br_blockcount);
2325 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2326 
2327 		xfs_iext_remove(ip, *idx + 1, 2, state);
2328 		ip->i_d.di_nextents -= 2;
2329 		if (cur == NULL)
2330 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2331 		else {
2332 			rval = XFS_ILOG_CORE;
2333 			if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2334 					RIGHT.br_startblock,
2335 					RIGHT.br_blockcount, &i)))
2336 				goto done;
2337 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2338 			if ((error = xfs_btree_delete(cur, &i)))
2339 				goto done;
2340 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2341 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2342 				goto done;
2343 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2344 			if ((error = xfs_btree_delete(cur, &i)))
2345 				goto done;
2346 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2347 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2348 				goto done;
2349 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2350 			if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2351 				LEFT.br_startblock,
2352 				LEFT.br_blockcount + PREV.br_blockcount +
2353 				RIGHT.br_blockcount, LEFT.br_state)))
2354 				goto done;
2355 		}
2356 		break;
2357 
2358 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2359 		/*
2360 		 * Setting all of a previous oldext extent to newext.
2361 		 * The left neighbor is contiguous, the right is not.
2362 		 */
2363 		--*idx;
2364 
2365 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2366 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2367 			LEFT.br_blockcount + PREV.br_blockcount);
2368 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2369 
2370 		xfs_iext_remove(ip, *idx + 1, 1, state);
2371 		ip->i_d.di_nextents--;
2372 		if (cur == NULL)
2373 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2374 		else {
2375 			rval = XFS_ILOG_CORE;
2376 			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2377 					PREV.br_startblock, PREV.br_blockcount,
2378 					&i)))
2379 				goto done;
2380 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2381 			if ((error = xfs_btree_delete(cur, &i)))
2382 				goto done;
2383 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2384 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2385 				goto done;
2386 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2387 			if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2388 				LEFT.br_startblock,
2389 				LEFT.br_blockcount + PREV.br_blockcount,
2390 				LEFT.br_state)))
2391 				goto done;
2392 		}
2393 		break;
2394 
2395 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2396 		/*
2397 		 * Setting all of a previous oldext extent to newext.
2398 		 * The right neighbor is contiguous, the left is not.
2399 		 */
2400 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2401 		xfs_bmbt_set_blockcount(ep,
2402 			PREV.br_blockcount + RIGHT.br_blockcount);
2403 		xfs_bmbt_set_state(ep, newext);
2404 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2405 		xfs_iext_remove(ip, *idx + 1, 1, state);
2406 		ip->i_d.di_nextents--;
2407 		if (cur == NULL)
2408 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2409 		else {
2410 			rval = XFS_ILOG_CORE;
2411 			if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2412 					RIGHT.br_startblock,
2413 					RIGHT.br_blockcount, &i)))
2414 				goto done;
2415 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2416 			if ((error = xfs_btree_delete(cur, &i)))
2417 				goto done;
2418 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2419 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2420 				goto done;
2421 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2422 			if ((error = xfs_bmbt_update(cur, new->br_startoff,
2423 				new->br_startblock,
2424 				new->br_blockcount + RIGHT.br_blockcount,
2425 				newext)))
2426 				goto done;
2427 		}
2428 		break;
2429 
2430 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2431 		/*
2432 		 * Setting all of a previous oldext extent to newext.
2433 		 * Neither the left nor right neighbors are contiguous with
2434 		 * the new one.
2435 		 */
2436 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2437 		xfs_bmbt_set_state(ep, newext);
2438 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2439 
2440 		if (cur == NULL)
2441 			rval = XFS_ILOG_DEXT;
2442 		else {
2443 			rval = 0;
2444 			if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2445 					new->br_startblock, new->br_blockcount,
2446 					&i)))
2447 				goto done;
2448 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2449 			if ((error = xfs_bmbt_update(cur, new->br_startoff,
2450 				new->br_startblock, new->br_blockcount,
2451 				newext)))
2452 				goto done;
2453 		}
2454 		break;
2455 
2456 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2457 		/*
2458 		 * Setting the first part of a previous oldext extent to newext.
2459 		 * The left neighbor is contiguous.
2460 		 */
2461 		trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2462 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2463 			LEFT.br_blockcount + new->br_blockcount);
2464 		xfs_bmbt_set_startoff(ep,
2465 			PREV.br_startoff + new->br_blockcount);
2466 		trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2467 
2468 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2469 		xfs_bmbt_set_startblock(ep,
2470 			new->br_startblock + new->br_blockcount);
2471 		xfs_bmbt_set_blockcount(ep,
2472 			PREV.br_blockcount - new->br_blockcount);
2473 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2474 
2475 		--*idx;
2476 
2477 		if (cur == NULL)
2478 			rval = XFS_ILOG_DEXT;
2479 		else {
2480 			rval = 0;
2481 			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2482 					PREV.br_startblock, PREV.br_blockcount,
2483 					&i)))
2484 				goto done;
2485 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2486 			if ((error = xfs_bmbt_update(cur,
2487 				PREV.br_startoff + new->br_blockcount,
2488 				PREV.br_startblock + new->br_blockcount,
2489 				PREV.br_blockcount - new->br_blockcount,
2490 				oldext)))
2491 				goto done;
2492 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2493 				goto done;
2494 			error = xfs_bmbt_update(cur, LEFT.br_startoff,
2495 				LEFT.br_startblock,
2496 				LEFT.br_blockcount + new->br_blockcount,
2497 				LEFT.br_state);
2498 			if (error)
2499 				goto done;
2500 		}
2501 		break;
2502 
2503 	case BMAP_LEFT_FILLING:
2504 		/*
2505 		 * Setting the first part of a previous oldext extent to newext.
2506 		 * The left neighbor is not contiguous.
2507 		 */
2508 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2509 		ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2510 		xfs_bmbt_set_startoff(ep, new_endoff);
2511 		xfs_bmbt_set_blockcount(ep,
2512 			PREV.br_blockcount - new->br_blockcount);
2513 		xfs_bmbt_set_startblock(ep,
2514 			new->br_startblock + new->br_blockcount);
2515 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2516 
2517 		xfs_iext_insert(ip, *idx, 1, new, state);
2518 		ip->i_d.di_nextents++;
2519 		if (cur == NULL)
2520 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2521 		else {
2522 			rval = XFS_ILOG_CORE;
2523 			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2524 					PREV.br_startblock, PREV.br_blockcount,
2525 					&i)))
2526 				goto done;
2527 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2528 			if ((error = xfs_bmbt_update(cur,
2529 				PREV.br_startoff + new->br_blockcount,
2530 				PREV.br_startblock + new->br_blockcount,
2531 				PREV.br_blockcount - new->br_blockcount,
2532 				oldext)))
2533 				goto done;
2534 			cur->bc_rec.b = *new;
2535 			if ((error = xfs_btree_insert(cur, &i)))
2536 				goto done;
2537 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2538 		}
2539 		break;
2540 
2541 	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2542 		/*
2543 		 * Setting the last part of a previous oldext extent to newext.
2544 		 * The right neighbor is contiguous with the new allocation.
2545 		 */
2546 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2547 		xfs_bmbt_set_blockcount(ep,
2548 			PREV.br_blockcount - new->br_blockcount);
2549 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2550 
2551 		++*idx;
2552 
2553 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2554 		xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2555 			new->br_startoff, new->br_startblock,
2556 			new->br_blockcount + RIGHT.br_blockcount, newext);
2557 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2558 
2559 		if (cur == NULL)
2560 			rval = XFS_ILOG_DEXT;
2561 		else {
2562 			rval = 0;
2563 			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2564 					PREV.br_startblock,
2565 					PREV.br_blockcount, &i)))
2566 				goto done;
2567 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2568 			if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2569 				PREV.br_startblock,
2570 				PREV.br_blockcount - new->br_blockcount,
2571 				oldext)))
2572 				goto done;
2573 			if ((error = xfs_btree_increment(cur, 0, &i)))
2574 				goto done;
2575 			if ((error = xfs_bmbt_update(cur, new->br_startoff,
2576 				new->br_startblock,
2577 				new->br_blockcount + RIGHT.br_blockcount,
2578 				newext)))
2579 				goto done;
2580 		}
2581 		break;
2582 
2583 	case BMAP_RIGHT_FILLING:
2584 		/*
2585 		 * Setting the last part of a previous oldext extent to newext.
2586 		 * The right neighbor is not contiguous.
2587 		 */
2588 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2589 		xfs_bmbt_set_blockcount(ep,
2590 			PREV.br_blockcount - new->br_blockcount);
2591 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2592 
2593 		++*idx;
2594 		xfs_iext_insert(ip, *idx, 1, new, state);
2595 
2596 		ip->i_d.di_nextents++;
2597 		if (cur == NULL)
2598 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2599 		else {
2600 			rval = XFS_ILOG_CORE;
2601 			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2602 					PREV.br_startblock, PREV.br_blockcount,
2603 					&i)))
2604 				goto done;
2605 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2606 			if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2607 				PREV.br_startblock,
2608 				PREV.br_blockcount - new->br_blockcount,
2609 				oldext)))
2610 				goto done;
2611 			if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2612 					new->br_startblock, new->br_blockcount,
2613 					&i)))
2614 				goto done;
2615 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2616 			cur->bc_rec.b.br_state = XFS_EXT_NORM;
2617 			if ((error = xfs_btree_insert(cur, &i)))
2618 				goto done;
2619 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2620 		}
2621 		break;
2622 
2623 	case 0:
2624 		/*
2625 		 * Setting the middle part of a previous oldext extent to
2626 		 * newext.  Contiguity is impossible here.
2627 		 * One extent becomes three extents.
2628 		 */
2629 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2630 		xfs_bmbt_set_blockcount(ep,
2631 			new->br_startoff - PREV.br_startoff);
2632 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2633 
2634 		r[0] = *new;
2635 		r[1].br_startoff = new_endoff;
2636 		r[1].br_blockcount =
2637 			PREV.br_startoff + PREV.br_blockcount - new_endoff;
2638 		r[1].br_startblock = new->br_startblock + new->br_blockcount;
2639 		r[1].br_state = oldext;
2640 
2641 		++*idx;
2642 		xfs_iext_insert(ip, *idx, 2, &r[0], state);
2643 
2644 		ip->i_d.di_nextents += 2;
2645 		if (cur == NULL)
2646 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2647 		else {
2648 			rval = XFS_ILOG_CORE;
2649 			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2650 					PREV.br_startblock, PREV.br_blockcount,
2651 					&i)))
2652 				goto done;
2653 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2654 			/* new right extent - oldext */
2655 			if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2656 				r[1].br_startblock, r[1].br_blockcount,
2657 				r[1].br_state)))
2658 				goto done;
2659 			/* new left extent - oldext */
2660 			cur->bc_rec.b = PREV;
2661 			cur->bc_rec.b.br_blockcount =
2662 				new->br_startoff - PREV.br_startoff;
2663 			if ((error = xfs_btree_insert(cur, &i)))
2664 				goto done;
2665 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2666 			/*
2667 			 * Reset the cursor to the position of the new extent
2668 			 * we are about to insert as we can't trust it after
2669 			 * the previous insert.
2670 			 */
2671 			if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2672 					new->br_startblock, new->br_blockcount,
2673 					&i)))
2674 				goto done;
2675 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2676 			/* new middle extent - newext */
2677 			cur->bc_rec.b.br_state = new->br_state;
2678 			if ((error = xfs_btree_insert(cur, &i)))
2679 				goto done;
2680 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2681 		}
2682 		break;
2683 
2684 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2685 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2686 	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2687 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2688 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2689 	case BMAP_LEFT_CONTIG:
2690 	case BMAP_RIGHT_CONTIG:
2691 		/*
2692 		 * These cases are all impossible.
2693 		 */
2694 		ASSERT(0);
2695 	}
2696 
2697 	/* update reverse mappings */
2698 	error = xfs_rmap_convert_extent(mp, dfops, ip, XFS_DATA_FORK, new);
2699 	if (error)
2700 		goto done;
2701 
2702 	/* convert to a btree if necessary */
2703 	if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
2704 		int	tmp_logflags;	/* partial log flag return val */
2705 
2706 		ASSERT(cur == NULL);
2707 		error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, &cur,
2708 				0, &tmp_logflags, XFS_DATA_FORK);
2709 		*logflagsp |= tmp_logflags;
2710 		if (error)
2711 			goto done;
2712 	}
2713 
2714 	/* clear out the allocated field, done with it now in any case. */
2715 	if (cur) {
2716 		cur->bc_private.b.allocated = 0;
2717 		*curp = cur;
2718 	}
2719 
2720 	xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
2721 done:
2722 	*logflagsp |= rval;
2723 	return error;
2724 #undef	LEFT
2725 #undef	RIGHT
2726 #undef	PREV
2727 }
2728 
2729 /*
2730  * Convert a hole to a delayed allocation.
2731  */
2732 STATIC void
2733 xfs_bmap_add_extent_hole_delay(
2734 	xfs_inode_t		*ip,	/* incore inode pointer */
2735 	int			whichfork,
2736 	xfs_extnum_t		*idx,	/* extent number to update/insert */
2737 	xfs_bmbt_irec_t		*new)	/* new data to add to file extents */
2738 {
2739 	xfs_ifork_t		*ifp;	/* inode fork pointer */
2740 	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2741 	xfs_filblks_t		newlen=0;	/* new indirect size */
2742 	xfs_filblks_t		oldlen=0;	/* old indirect size */
2743 	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2744 	int			state;  /* state bits, accessed thru macros */
2745 	xfs_filblks_t		temp=0;	/* temp for indirect calculations */
2746 
2747 	ifp = XFS_IFORK_PTR(ip, whichfork);
2748 	state = 0;
2749 	if (whichfork == XFS_COW_FORK)
2750 		state |= BMAP_COWFORK;
2751 	ASSERT(isnullstartblock(new->br_startblock));
2752 
2753 	/*
2754 	 * Check and set flags if this segment has a left neighbor
2755 	 */
2756 	if (*idx > 0) {
2757 		state |= BMAP_LEFT_VALID;
2758 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2759 
2760 		if (isnullstartblock(left.br_startblock))
2761 			state |= BMAP_LEFT_DELAY;
2762 	}
2763 
2764 	/*
2765 	 * Check and set flags if the current (right) segment exists.
2766 	 * If it doesn't exist, we're converting the hole at end-of-file.
2767 	 */
2768 	if (*idx < xfs_iext_count(ifp)) {
2769 		state |= BMAP_RIGHT_VALID;
2770 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2771 
2772 		if (isnullstartblock(right.br_startblock))
2773 			state |= BMAP_RIGHT_DELAY;
2774 	}
2775 
2776 	/*
2777 	 * Set contiguity flags on the left and right neighbors.
2778 	 * Don't let extents get too large, even if the pieces are contiguous.
2779 	 */
2780 	if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2781 	    left.br_startoff + left.br_blockcount == new->br_startoff &&
2782 	    left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2783 		state |= BMAP_LEFT_CONTIG;
2784 
2785 	if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2786 	    new->br_startoff + new->br_blockcount == right.br_startoff &&
2787 	    new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2788 	    (!(state & BMAP_LEFT_CONTIG) ||
2789 	     (left.br_blockcount + new->br_blockcount +
2790 	      right.br_blockcount <= MAXEXTLEN)))
2791 		state |= BMAP_RIGHT_CONTIG;
2792 
2793 	/*
2794 	 * Switch out based on the contiguity flags.
2795 	 */
2796 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2797 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2798 		/*
2799 		 * New allocation is contiguous with delayed allocations
2800 		 * on the left and on the right.
2801 		 * Merge all three into a single extent record.
2802 		 */
2803 		--*idx;
2804 		temp = left.br_blockcount + new->br_blockcount +
2805 			right.br_blockcount;
2806 
2807 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2808 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2809 		oldlen = startblockval(left.br_startblock) +
2810 			startblockval(new->br_startblock) +
2811 			startblockval(right.br_startblock);
2812 		newlen = xfs_bmap_worst_indlen(ip, temp);
2813 		xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2814 			nullstartblock((int)newlen));
2815 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2816 
2817 		xfs_iext_remove(ip, *idx + 1, 1, state);
2818 		break;
2819 
2820 	case BMAP_LEFT_CONTIG:
2821 		/*
2822 		 * New allocation is contiguous with a delayed allocation
2823 		 * on the left.
2824 		 * Merge the new allocation with the left neighbor.
2825 		 */
2826 		--*idx;
2827 		temp = left.br_blockcount + new->br_blockcount;
2828 
2829 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2830 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2831 		oldlen = startblockval(left.br_startblock) +
2832 			startblockval(new->br_startblock);
2833 		newlen = xfs_bmap_worst_indlen(ip, temp);
2834 		xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2835 			nullstartblock((int)newlen));
2836 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2837 		break;
2838 
2839 	case BMAP_RIGHT_CONTIG:
2840 		/*
2841 		 * New allocation is contiguous with a delayed allocation
2842 		 * on the right.
2843 		 * Merge the new allocation with the right neighbor.
2844 		 */
2845 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2846 		temp = new->br_blockcount + right.br_blockcount;
2847 		oldlen = startblockval(new->br_startblock) +
2848 			startblockval(right.br_startblock);
2849 		newlen = xfs_bmap_worst_indlen(ip, temp);
2850 		xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2851 			new->br_startoff,
2852 			nullstartblock((int)newlen), temp, right.br_state);
2853 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2854 		break;
2855 
2856 	case 0:
2857 		/*
2858 		 * New allocation is not contiguous with another
2859 		 * delayed allocation.
2860 		 * Insert a new entry.
2861 		 */
2862 		oldlen = newlen = 0;
2863 		xfs_iext_insert(ip, *idx, 1, new, state);
2864 		break;
2865 	}
2866 	if (oldlen != newlen) {
2867 		ASSERT(oldlen > newlen);
2868 		xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2869 				 false);
2870 		/*
2871 		 * Nothing to do for disk quota accounting here.
2872 		 */
2873 	}
2874 }
2875 
2876 /*
2877  * Convert a hole to a real allocation.
2878  */
2879 STATIC int				/* error */
2880 xfs_bmap_add_extent_hole_real(
2881 	struct xfs_bmalloca	*bma,
2882 	int			whichfork)
2883 {
2884 	struct xfs_bmbt_irec	*new = &bma->got;
2885 	int			error;	/* error return value */
2886 	int			i;	/* temp state */
2887 	xfs_ifork_t		*ifp;	/* inode fork pointer */
2888 	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2889 	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2890 	int			rval=0;	/* return value (logging flags) */
2891 	int			state;	/* state bits, accessed thru macros */
2892 	struct xfs_mount	*mp;
2893 
2894 	mp = bma->ip->i_mount;
2895 	ifp = XFS_IFORK_PTR(bma->ip, whichfork);
2896 
2897 	ASSERT(bma->idx >= 0);
2898 	ASSERT(bma->idx <= xfs_iext_count(ifp));
2899 	ASSERT(!isnullstartblock(new->br_startblock));
2900 	ASSERT(!bma->cur ||
2901 	       !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2902 	ASSERT(whichfork != XFS_COW_FORK);
2903 
2904 	XFS_STATS_INC(mp, xs_add_exlist);
2905 
2906 	state = 0;
2907 	if (whichfork == XFS_ATTR_FORK)
2908 		state |= BMAP_ATTRFORK;
2909 
2910 	/*
2911 	 * Check and set flags if this segment has a left neighbor.
2912 	 */
2913 	if (bma->idx > 0) {
2914 		state |= BMAP_LEFT_VALID;
2915 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
2916 		if (isnullstartblock(left.br_startblock))
2917 			state |= BMAP_LEFT_DELAY;
2918 	}
2919 
2920 	/*
2921 	 * Check and set flags if this segment has a current value.
2922 	 * Not true if we're inserting into the "hole" at eof.
2923 	 */
2924 	if (bma->idx < xfs_iext_count(ifp)) {
2925 		state |= BMAP_RIGHT_VALID;
2926 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
2927 		if (isnullstartblock(right.br_startblock))
2928 			state |= BMAP_RIGHT_DELAY;
2929 	}
2930 
2931 	/*
2932 	 * We're inserting a real allocation between "left" and "right".
2933 	 * Set the contiguity flags.  Don't let extents get too large.
2934 	 */
2935 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2936 	    left.br_startoff + left.br_blockcount == new->br_startoff &&
2937 	    left.br_startblock + left.br_blockcount == new->br_startblock &&
2938 	    left.br_state == new->br_state &&
2939 	    left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2940 		state |= BMAP_LEFT_CONTIG;
2941 
2942 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2943 	    new->br_startoff + new->br_blockcount == right.br_startoff &&
2944 	    new->br_startblock + new->br_blockcount == right.br_startblock &&
2945 	    new->br_state == right.br_state &&
2946 	    new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2947 	    (!(state & BMAP_LEFT_CONTIG) ||
2948 	     left.br_blockcount + new->br_blockcount +
2949 	     right.br_blockcount <= MAXEXTLEN))
2950 		state |= BMAP_RIGHT_CONTIG;
2951 
2952 	error = 0;
2953 	/*
2954 	 * Select which case we're in here, and implement it.
2955 	 */
2956 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2957 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2958 		/*
2959 		 * New allocation is contiguous with real allocations on the
2960 		 * left and on the right.
2961 		 * Merge all three into a single extent record.
2962 		 */
2963 		--bma->idx;
2964 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2965 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
2966 			left.br_blockcount + new->br_blockcount +
2967 			right.br_blockcount);
2968 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2969 
2970 		xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
2971 
2972 		XFS_IFORK_NEXT_SET(bma->ip, whichfork,
2973 			XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
2974 		if (bma->cur == NULL) {
2975 			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2976 		} else {
2977 			rval = XFS_ILOG_CORE;
2978 			error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
2979 					right.br_startblock, right.br_blockcount,
2980 					&i);
2981 			if (error)
2982 				goto done;
2983 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2984 			error = xfs_btree_delete(bma->cur, &i);
2985 			if (error)
2986 				goto done;
2987 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2988 			error = xfs_btree_decrement(bma->cur, 0, &i);
2989 			if (error)
2990 				goto done;
2991 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2992 			error = xfs_bmbt_update(bma->cur, left.br_startoff,
2993 					left.br_startblock,
2994 					left.br_blockcount +
2995 						new->br_blockcount +
2996 						right.br_blockcount,
2997 					left.br_state);
2998 			if (error)
2999 				goto done;
3000 		}
3001 		break;
3002 
3003 	case BMAP_LEFT_CONTIG:
3004 		/*
3005 		 * New allocation is contiguous with a real allocation
3006 		 * on the left.
3007 		 * Merge the new allocation with the left neighbor.
3008 		 */
3009 		--bma->idx;
3010 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3011 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3012 			left.br_blockcount + new->br_blockcount);
3013 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3014 
3015 		if (bma->cur == NULL) {
3016 			rval = xfs_ilog_fext(whichfork);
3017 		} else {
3018 			rval = 0;
3019 			error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
3020 					left.br_startblock, left.br_blockcount,
3021 					&i);
3022 			if (error)
3023 				goto done;
3024 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3025 			error = xfs_bmbt_update(bma->cur, left.br_startoff,
3026 					left.br_startblock,
3027 					left.br_blockcount +
3028 						new->br_blockcount,
3029 					left.br_state);
3030 			if (error)
3031 				goto done;
3032 		}
3033 		break;
3034 
3035 	case BMAP_RIGHT_CONTIG:
3036 		/*
3037 		 * New allocation is contiguous with a real allocation
3038 		 * on the right.
3039 		 * Merge the new allocation with the right neighbor.
3040 		 */
3041 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3042 		xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
3043 			new->br_startoff, new->br_startblock,
3044 			new->br_blockcount + right.br_blockcount,
3045 			right.br_state);
3046 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3047 
3048 		if (bma->cur == NULL) {
3049 			rval = xfs_ilog_fext(whichfork);
3050 		} else {
3051 			rval = 0;
3052 			error = xfs_bmbt_lookup_eq(bma->cur,
3053 					right.br_startoff,
3054 					right.br_startblock,
3055 					right.br_blockcount, &i);
3056 			if (error)
3057 				goto done;
3058 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3059 			error = xfs_bmbt_update(bma->cur, new->br_startoff,
3060 					new->br_startblock,
3061 					new->br_blockcount +
3062 						right.br_blockcount,
3063 					right.br_state);
3064 			if (error)
3065 				goto done;
3066 		}
3067 		break;
3068 
3069 	case 0:
3070 		/*
3071 		 * New allocation is not contiguous with another
3072 		 * real allocation.
3073 		 * Insert a new entry.
3074 		 */
3075 		xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
3076 		XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3077 			XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
3078 		if (bma->cur == NULL) {
3079 			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3080 		} else {
3081 			rval = XFS_ILOG_CORE;
3082 			error = xfs_bmbt_lookup_eq(bma->cur,
3083 					new->br_startoff,
3084 					new->br_startblock,
3085 					new->br_blockcount, &i);
3086 			if (error)
3087 				goto done;
3088 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
3089 			bma->cur->bc_rec.b.br_state = new->br_state;
3090 			error = xfs_btree_insert(bma->cur, &i);
3091 			if (error)
3092 				goto done;
3093 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3094 		}
3095 		break;
3096 	}
3097 
3098 	/* add reverse mapping */
3099 	error = xfs_rmap_map_extent(mp, bma->dfops, bma->ip, whichfork, new);
3100 	if (error)
3101 		goto done;
3102 
3103 	/* convert to a btree if necessary */
3104 	if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
3105 		int	tmp_logflags;	/* partial log flag return val */
3106 
3107 		ASSERT(bma->cur == NULL);
3108 		error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
3109 				bma->firstblock, bma->dfops, &bma->cur,
3110 				0, &tmp_logflags, whichfork);
3111 		bma->logflags |= tmp_logflags;
3112 		if (error)
3113 			goto done;
3114 	}
3115 
3116 	/* clear out the allocated field, done with it now in any case. */
3117 	if (bma->cur)
3118 		bma->cur->bc_private.b.allocated = 0;
3119 
3120 	xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
3121 done:
3122 	bma->logflags |= rval;
3123 	return error;
3124 }
3125 
3126 /*
3127  * Functions used in the extent read, allocate and remove paths
3128  */
3129 
3130 /*
3131  * Adjust the size of the new extent based on di_extsize and rt extsize.
3132  */
3133 int
3134 xfs_bmap_extsize_align(
3135 	xfs_mount_t	*mp,
3136 	xfs_bmbt_irec_t	*gotp,		/* next extent pointer */
3137 	xfs_bmbt_irec_t	*prevp,		/* previous extent pointer */
3138 	xfs_extlen_t	extsz,		/* align to this extent size */
3139 	int		rt,		/* is this a realtime inode? */
3140 	int		eof,		/* is extent at end-of-file? */
3141 	int		delay,		/* creating delalloc extent? */
3142 	int		convert,	/* overwriting unwritten extent? */
3143 	xfs_fileoff_t	*offp,		/* in/out: aligned offset */
3144 	xfs_extlen_t	*lenp)		/* in/out: aligned length */
3145 {
3146 	xfs_fileoff_t	orig_off;	/* original offset */
3147 	xfs_extlen_t	orig_alen;	/* original length */
3148 	xfs_fileoff_t	orig_end;	/* original off+len */
3149 	xfs_fileoff_t	nexto;		/* next file offset */
3150 	xfs_fileoff_t	prevo;		/* previous file offset */
3151 	xfs_fileoff_t	align_off;	/* temp for offset */
3152 	xfs_extlen_t	align_alen;	/* temp for length */
3153 	xfs_extlen_t	temp;		/* temp for calculations */
3154 
3155 	if (convert)
3156 		return 0;
3157 
3158 	orig_off = align_off = *offp;
3159 	orig_alen = align_alen = *lenp;
3160 	orig_end = orig_off + orig_alen;
3161 
3162 	/*
3163 	 * If this request overlaps an existing extent, then don't
3164 	 * attempt to perform any additional alignment.
3165 	 */
3166 	if (!delay && !eof &&
3167 	    (orig_off >= gotp->br_startoff) &&
3168 	    (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3169 		return 0;
3170 	}
3171 
3172 	/*
3173 	 * If the file offset is unaligned vs. the extent size
3174 	 * we need to align it.  This will be possible unless
3175 	 * the file was previously written with a kernel that didn't
3176 	 * perform this alignment, or if a truncate shot us in the
3177 	 * foot.
3178 	 */
3179 	temp = do_mod(orig_off, extsz);
3180 	if (temp) {
3181 		align_alen += temp;
3182 		align_off -= temp;
3183 	}
3184 
3185 	/* Same adjustment for the end of the requested area. */
3186 	temp = (align_alen % extsz);
3187 	if (temp)
3188 		align_alen += extsz - temp;
3189 
3190 	/*
3191 	 * For large extent hint sizes, the aligned extent might be larger than
3192 	 * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
3193 	 * the length back under MAXEXTLEN. The outer allocation loops handle
3194 	 * short allocation just fine, so it is safe to do this. We only want to
3195 	 * do it when we are forced to, though, because it means more allocation
3196 	 * operations are required.
3197 	 */
3198 	while (align_alen > MAXEXTLEN)
3199 		align_alen -= extsz;
3200 	ASSERT(align_alen <= MAXEXTLEN);
3201 
3202 	/*
3203 	 * If the previous block overlaps with this proposed allocation
3204 	 * then move the start forward without adjusting the length.
3205 	 */
3206 	if (prevp->br_startoff != NULLFILEOFF) {
3207 		if (prevp->br_startblock == HOLESTARTBLOCK)
3208 			prevo = prevp->br_startoff;
3209 		else
3210 			prevo = prevp->br_startoff + prevp->br_blockcount;
3211 	} else
3212 		prevo = 0;
3213 	if (align_off != orig_off && align_off < prevo)
3214 		align_off = prevo;
3215 	/*
3216 	 * If the next block overlaps with this proposed allocation
3217 	 * then move the start back without adjusting the length,
3218 	 * but not before offset 0.
3219 	 * This may of course make the start overlap previous block,
3220 	 * and if we hit the offset 0 limit then the next block
3221 	 * can still overlap too.
3222 	 */
3223 	if (!eof && gotp->br_startoff != NULLFILEOFF) {
3224 		if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3225 		    (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3226 			nexto = gotp->br_startoff + gotp->br_blockcount;
3227 		else
3228 			nexto = gotp->br_startoff;
3229 	} else
3230 		nexto = NULLFILEOFF;
3231 	if (!eof &&
3232 	    align_off + align_alen != orig_end &&
3233 	    align_off + align_alen > nexto)
3234 		align_off = nexto > align_alen ? nexto - align_alen : 0;
3235 	/*
3236 	 * If we're now overlapping the next or previous extent that
3237 	 * means we can't fit an extsz piece in this hole.  Just move
3238 	 * the start forward to the first valid spot and set
3239 	 * the length so we hit the end.
3240 	 */
3241 	if (align_off != orig_off && align_off < prevo)
3242 		align_off = prevo;
3243 	if (align_off + align_alen != orig_end &&
3244 	    align_off + align_alen > nexto &&
3245 	    nexto != NULLFILEOFF) {
3246 		ASSERT(nexto > prevo);
3247 		align_alen = nexto - align_off;
3248 	}
3249 
3250 	/*
3251 	 * If realtime, and the result isn't a multiple of the realtime
3252 	 * extent size we need to remove blocks until it is.
3253 	 */
3254 	if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3255 		/*
3256 		 * We're not covering the original request, or
3257 		 * we won't be able to once we fix the length.
3258 		 */
3259 		if (orig_off < align_off ||
3260 		    orig_end > align_off + align_alen ||
3261 		    align_alen - temp < orig_alen)
3262 			return -EINVAL;
3263 		/*
3264 		 * Try to fix it by moving the start up.
3265 		 */
3266 		if (align_off + temp <= orig_off) {
3267 			align_alen -= temp;
3268 			align_off += temp;
3269 		}
3270 		/*
3271 		 * Try to fix it by moving the end in.
3272 		 */
3273 		else if (align_off + align_alen - temp >= orig_end)
3274 			align_alen -= temp;
3275 		/*
3276 		 * Set the start to the minimum then trim the length.
3277 		 */
3278 		else {
3279 			align_alen -= orig_off - align_off;
3280 			align_off = orig_off;
3281 			align_alen -= align_alen % mp->m_sb.sb_rextsize;
3282 		}
3283 		/*
3284 		 * Result doesn't cover the request, fail it.
3285 		 */
3286 		if (orig_off < align_off || orig_end > align_off + align_alen)
3287 			return -EINVAL;
3288 	} else {
3289 		ASSERT(orig_off >= align_off);
3290 		/* see MAXEXTLEN handling above */
3291 		ASSERT(orig_end <= align_off + align_alen ||
3292 		       align_alen + extsz > MAXEXTLEN);
3293 	}
3294 
3295 #ifdef DEBUG
3296 	if (!eof && gotp->br_startoff != NULLFILEOFF)
3297 		ASSERT(align_off + align_alen <= gotp->br_startoff);
3298 	if (prevp->br_startoff != NULLFILEOFF)
3299 		ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3300 #endif
3301 
3302 	*lenp = align_alen;
3303 	*offp = align_off;
3304 	return 0;
3305 }
3306 
3307 #define XFS_ALLOC_GAP_UNITS	4
3308 
3309 void
3310 xfs_bmap_adjacent(
3311 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3312 {
3313 	xfs_fsblock_t	adjust;		/* adjustment to block numbers */
3314 	xfs_agnumber_t	fb_agno;	/* ag number of ap->firstblock */
3315 	xfs_mount_t	*mp;		/* mount point structure */
3316 	int		nullfb;		/* true if ap->firstblock isn't set */
3317 	int		rt;		/* true if inode is realtime */
3318 
3319 #define	ISVALID(x,y)	\
3320 	(rt ? \
3321 		(x) < mp->m_sb.sb_rblocks : \
3322 		XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3323 		XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3324 		XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3325 
3326 	mp = ap->ip->i_mount;
3327 	nullfb = *ap->firstblock == NULLFSBLOCK;
3328 	rt = XFS_IS_REALTIME_INODE(ap->ip) &&
3329 		xfs_alloc_is_userdata(ap->datatype);
3330 	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3331 	/*
3332 	 * If allocating at eof, and there's a previous real block,
3333 	 * try to use its last block as our starting point.
3334 	 */
3335 	if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3336 	    !isnullstartblock(ap->prev.br_startblock) &&
3337 	    ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3338 		    ap->prev.br_startblock)) {
3339 		ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3340 		/*
3341 		 * Adjust for the gap between prevp and us.
3342 		 */
3343 		adjust = ap->offset -
3344 			(ap->prev.br_startoff + ap->prev.br_blockcount);
3345 		if (adjust &&
3346 		    ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3347 			ap->blkno += adjust;
3348 	}
3349 	/*
3350 	 * If not at eof, then compare the two neighbor blocks.
3351 	 * Figure out whether either one gives us a good starting point,
3352 	 * and pick the better one.
3353 	 */
3354 	else if (!ap->eof) {
3355 		xfs_fsblock_t	gotbno;		/* right side block number */
3356 		xfs_fsblock_t	gotdiff=0;	/* right side difference */
3357 		xfs_fsblock_t	prevbno;	/* left side block number */
3358 		xfs_fsblock_t	prevdiff=0;	/* left side difference */
3359 
3360 		/*
3361 		 * If there's a previous (left) block, select a requested
3362 		 * start block based on it.
3363 		 */
3364 		if (ap->prev.br_startoff != NULLFILEOFF &&
3365 		    !isnullstartblock(ap->prev.br_startblock) &&
3366 		    (prevbno = ap->prev.br_startblock +
3367 			       ap->prev.br_blockcount) &&
3368 		    ISVALID(prevbno, ap->prev.br_startblock)) {
3369 			/*
3370 			 * Calculate gap to end of previous block.
3371 			 */
3372 			adjust = prevdiff = ap->offset -
3373 				(ap->prev.br_startoff +
3374 				 ap->prev.br_blockcount);
3375 			/*
3376 			 * Figure the startblock based on the previous block's
3377 			 * end and the gap size.
3378 			 * Heuristic!
3379 			 * If the gap is large relative to the piece we're
3380 			 * allocating, or using it gives us an invalid block
3381 			 * number, then just use the end of the previous block.
3382 			 */
3383 			if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3384 			    ISVALID(prevbno + prevdiff,
3385 				    ap->prev.br_startblock))
3386 				prevbno += adjust;
3387 			else
3388 				prevdiff += adjust;
3389 			/*
3390 			 * If the firstblock forbids it, can't use it,
3391 			 * must use default.
3392 			 */
3393 			if (!rt && !nullfb &&
3394 			    XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3395 				prevbno = NULLFSBLOCK;
3396 		}
3397 		/*
3398 		 * No previous block or can't follow it, just default.
3399 		 */
3400 		else
3401 			prevbno = NULLFSBLOCK;
3402 		/*
3403 		 * If there's a following (right) block, select a requested
3404 		 * start block based on it.
3405 		 */
3406 		if (!isnullstartblock(ap->got.br_startblock)) {
3407 			/*
3408 			 * Calculate gap to start of next block.
3409 			 */
3410 			adjust = gotdiff = ap->got.br_startoff - ap->offset;
3411 			/*
3412 			 * Figure the startblock based on the next block's
3413 			 * start and the gap size.
3414 			 */
3415 			gotbno = ap->got.br_startblock;
3416 			/*
3417 			 * Heuristic!
3418 			 * If the gap is large relative to the piece we're
3419 			 * allocating, or using it gives us an invalid block
3420 			 * number, then just use the start of the next block
3421 			 * offset by our length.
3422 			 */
3423 			if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3424 			    ISVALID(gotbno - gotdiff, gotbno))
3425 				gotbno -= adjust;
3426 			else if (ISVALID(gotbno - ap->length, gotbno)) {
3427 				gotbno -= ap->length;
3428 				gotdiff += adjust - ap->length;
3429 			} else
3430 				gotdiff += adjust;
3431 			/*
3432 			 * If the firstblock forbids it, can't use it,
3433 			 * must use default.
3434 			 */
3435 			if (!rt && !nullfb &&
3436 			    XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3437 				gotbno = NULLFSBLOCK;
3438 		}
3439 		/*
3440 		 * No next block, just default.
3441 		 */
3442 		else
3443 			gotbno = NULLFSBLOCK;
3444 		/*
3445 		 * If both valid, pick the better one, else the only good
3446 		 * one, else ap->blkno is already set (to 0 or the inode block).
3447 		 */
3448 		if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3449 			ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3450 		else if (prevbno != NULLFSBLOCK)
3451 			ap->blkno = prevbno;
3452 		else if (gotbno != NULLFSBLOCK)
3453 			ap->blkno = gotbno;
3454 	}
3455 #undef ISVALID
3456 }
3457 
3458 static int
3459 xfs_bmap_longest_free_extent(
3460 	struct xfs_trans	*tp,
3461 	xfs_agnumber_t		ag,
3462 	xfs_extlen_t		*blen,
3463 	int			*notinit)
3464 {
3465 	struct xfs_mount	*mp = tp->t_mountp;
3466 	struct xfs_perag	*pag;
3467 	xfs_extlen_t		longest;
3468 	int			error = 0;
3469 
3470 	pag = xfs_perag_get(mp, ag);
3471 	if (!pag->pagf_init) {
3472 		error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3473 		if (error)
3474 			goto out;
3475 
3476 		if (!pag->pagf_init) {
3477 			*notinit = 1;
3478 			goto out;
3479 		}
3480 	}
3481 
3482 	longest = xfs_alloc_longest_free_extent(mp, pag,
3483 				xfs_alloc_min_freelist(mp, pag),
3484 				xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
3485 	if (*blen < longest)
3486 		*blen = longest;
3487 
3488 out:
3489 	xfs_perag_put(pag);
3490 	return error;
3491 }
3492 
3493 static void
3494 xfs_bmap_select_minlen(
3495 	struct xfs_bmalloca	*ap,
3496 	struct xfs_alloc_arg	*args,
3497 	xfs_extlen_t		*blen,
3498 	int			notinit)
3499 {
3500 	if (notinit || *blen < ap->minlen) {
3501 		/*
3502 		 * Since we did a BUF_TRYLOCK above, it is possible that
3503 		 * there is space for this request.
3504 		 */
3505 		args->minlen = ap->minlen;
3506 	} else if (*blen < args->maxlen) {
3507 		/*
3508 		 * If the best seen length is less than the request length,
3509 		 * use the best as the minimum.
3510 		 */
3511 		args->minlen = *blen;
3512 	} else {
3513 		/*
3514 		 * Otherwise we've seen an extent as big as maxlen, use that
3515 		 * as the minimum.
3516 		 */
3517 		args->minlen = args->maxlen;
3518 	}
3519 }
3520 
3521 STATIC int
3522 xfs_bmap_btalloc_nullfb(
3523 	struct xfs_bmalloca	*ap,
3524 	struct xfs_alloc_arg	*args,
3525 	xfs_extlen_t		*blen)
3526 {
3527 	struct xfs_mount	*mp = ap->ip->i_mount;
3528 	xfs_agnumber_t		ag, startag;
3529 	int			notinit = 0;
3530 	int			error;
3531 
3532 	args->type = XFS_ALLOCTYPE_START_BNO;
3533 	args->total = ap->total;
3534 
3535 	startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3536 	if (startag == NULLAGNUMBER)
3537 		startag = ag = 0;
3538 
3539 	while (*blen < args->maxlen) {
3540 		error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3541 						     &notinit);
3542 		if (error)
3543 			return error;
3544 
3545 		if (++ag == mp->m_sb.sb_agcount)
3546 			ag = 0;
3547 		if (ag == startag)
3548 			break;
3549 	}
3550 
3551 	xfs_bmap_select_minlen(ap, args, blen, notinit);
3552 	return 0;
3553 }
3554 
3555 STATIC int
3556 xfs_bmap_btalloc_filestreams(
3557 	struct xfs_bmalloca	*ap,
3558 	struct xfs_alloc_arg	*args,
3559 	xfs_extlen_t		*blen)
3560 {
3561 	struct xfs_mount	*mp = ap->ip->i_mount;
3562 	xfs_agnumber_t		ag;
3563 	int			notinit = 0;
3564 	int			error;
3565 
3566 	args->type = XFS_ALLOCTYPE_NEAR_BNO;
3567 	args->total = ap->total;
3568 
3569 	ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3570 	if (ag == NULLAGNUMBER)
3571 		ag = 0;
3572 
3573 	error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3574 	if (error)
3575 		return error;
3576 
3577 	if (*blen < args->maxlen) {
3578 		error = xfs_filestream_new_ag(ap, &ag);
3579 		if (error)
3580 			return error;
3581 
3582 		error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3583 						     &notinit);
3584 		if (error)
3585 			return error;
3586 
3587 	}
3588 
3589 	xfs_bmap_select_minlen(ap, args, blen, notinit);
3590 
3591 	/*
3592 	 * Set the failure fallback case to look in the selected AG as stream
3593 	 * may have moved.
3594 	 */
3595 	ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3596 	return 0;
3597 }
3598 
3599 STATIC int
3600 xfs_bmap_btalloc(
3601 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3602 {
3603 	xfs_mount_t	*mp;		/* mount point structure */
3604 	xfs_alloctype_t	atype = 0;	/* type for allocation routines */
3605 	xfs_extlen_t	align = 0;	/* minimum allocation alignment */
3606 	xfs_agnumber_t	fb_agno;	/* ag number of ap->firstblock */
3607 	xfs_agnumber_t	ag;
3608 	xfs_alloc_arg_t	args;
3609 	xfs_extlen_t	blen;
3610 	xfs_extlen_t	nextminlen = 0;
3611 	int		nullfb;		/* true if ap->firstblock isn't set */
3612 	int		isaligned;
3613 	int		tryagain;
3614 	int		error;
3615 	int		stripe_align;
3616 
3617 	ASSERT(ap->length);
3618 
3619 	mp = ap->ip->i_mount;
3620 
3621 	/* stripe alignment for allocation is determined by mount parameters */
3622 	stripe_align = 0;
3623 	if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3624 		stripe_align = mp->m_swidth;
3625 	else if (mp->m_dalign)
3626 		stripe_align = mp->m_dalign;
3627 
3628 	if (ap->flags & XFS_BMAPI_COWFORK)
3629 		align = xfs_get_cowextsz_hint(ap->ip);
3630 	else if (xfs_alloc_is_userdata(ap->datatype))
3631 		align = xfs_get_extsz_hint(ap->ip);
3632 	if (unlikely(align)) {
3633 		error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3634 						align, 0, ap->eof, 0, ap->conv,
3635 						&ap->offset, &ap->length);
3636 		ASSERT(!error);
3637 		ASSERT(ap->length);
3638 	}
3639 
3640 
3641 	nullfb = *ap->firstblock == NULLFSBLOCK;
3642 	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3643 	if (nullfb) {
3644 		if (xfs_alloc_is_userdata(ap->datatype) &&
3645 		    xfs_inode_is_filestream(ap->ip)) {
3646 			ag = xfs_filestream_lookup_ag(ap->ip);
3647 			ag = (ag != NULLAGNUMBER) ? ag : 0;
3648 			ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3649 		} else {
3650 			ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3651 		}
3652 	} else
3653 		ap->blkno = *ap->firstblock;
3654 
3655 	xfs_bmap_adjacent(ap);
3656 
3657 	/*
3658 	 * If allowed, use ap->blkno; otherwise must use firstblock since
3659 	 * it's in the right allocation group.
3660 	 */
3661 	if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3662 		;
3663 	else
3664 		ap->blkno = *ap->firstblock;
3665 	/*
3666 	 * Normal allocation, done through xfs_alloc_vextent.
3667 	 */
3668 	tryagain = isaligned = 0;
3669 	memset(&args, 0, sizeof(args));
3670 	args.tp = ap->tp;
3671 	args.mp = mp;
3672 	args.fsbno = ap->blkno;
3673 	xfs_rmap_skip_owner_update(&args.oinfo);
3674 
3675 	/* Trim the allocation back to the maximum an AG can fit. */
3676 	args.maxlen = MIN(ap->length, mp->m_ag_max_usable);
3677 	args.firstblock = *ap->firstblock;
3678 	blen = 0;
3679 	if (nullfb) {
3680 		/*
3681 		 * Search for an allocation group with a single extent large
3682 		 * enough for the request.  If one isn't found, then adjust
3683 		 * the minimum allocation size to the largest space found.
3684 		 */
3685 		if (xfs_alloc_is_userdata(ap->datatype) &&
3686 		    xfs_inode_is_filestream(ap->ip))
3687 			error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3688 		else
3689 			error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3690 		if (error)
3691 			return error;
3692 	} else if (ap->dfops->dop_low) {
3693 		if (xfs_inode_is_filestream(ap->ip))
3694 			args.type = XFS_ALLOCTYPE_FIRST_AG;
3695 		else
3696 			args.type = XFS_ALLOCTYPE_START_BNO;
3697 		args.total = args.minlen = ap->minlen;
3698 	} else {
3699 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
3700 		args.total = ap->total;
3701 		args.minlen = ap->minlen;
3702 	}
3703 	/* apply extent size hints if obtained earlier */
3704 	if (unlikely(align)) {
3705 		args.prod = align;
3706 		if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3707 			args.mod = (xfs_extlen_t)(args.prod - args.mod);
3708 	} else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3709 		args.prod = 1;
3710 		args.mod = 0;
3711 	} else {
3712 		args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3713 		if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3714 			args.mod = (xfs_extlen_t)(args.prod - args.mod);
3715 	}
3716 	/*
3717 	 * If we are not low on available data blocks, and the
3718 	 * underlying logical volume manager is a stripe, and
3719 	 * the file offset is zero then try to allocate data
3720 	 * blocks on stripe unit boundary.
3721 	 * NOTE: ap->aeof is only set if the allocation length
3722 	 * is >= the stripe unit and the allocation offset is
3723 	 * at the end of file.
3724 	 */
3725 	if (!ap->dfops->dop_low && ap->aeof) {
3726 		if (!ap->offset) {
3727 			args.alignment = stripe_align;
3728 			atype = args.type;
3729 			isaligned = 1;
3730 			/*
3731 			 * Adjust for alignment
3732 			 */
3733 			if (blen > args.alignment && blen <= args.maxlen)
3734 				args.minlen = blen - args.alignment;
3735 			args.minalignslop = 0;
3736 		} else {
3737 			/*
3738 			 * First try an exact bno allocation.
3739 			 * If it fails then do a near or start bno
3740 			 * allocation with alignment turned on.
3741 			 */
3742 			atype = args.type;
3743 			tryagain = 1;
3744 			args.type = XFS_ALLOCTYPE_THIS_BNO;
3745 			args.alignment = 1;
3746 			/*
3747 			 * Compute the minlen+alignment for the
3748 			 * next case.  Set slop so that the value
3749 			 * of minlen+alignment+slop doesn't go up
3750 			 * between the calls.
3751 			 */
3752 			if (blen > stripe_align && blen <= args.maxlen)
3753 				nextminlen = blen - stripe_align;
3754 			else
3755 				nextminlen = args.minlen;
3756 			if (nextminlen + stripe_align > args.minlen + 1)
3757 				args.minalignslop =
3758 					nextminlen + stripe_align -
3759 					args.minlen - 1;
3760 			else
3761 				args.minalignslop = 0;
3762 		}
3763 	} else {
3764 		args.alignment = 1;
3765 		args.minalignslop = 0;
3766 	}
3767 	args.minleft = ap->minleft;
3768 	args.wasdel = ap->wasdel;
3769 	args.resv = XFS_AG_RESV_NONE;
3770 	args.datatype = ap->datatype;
3771 	if (ap->datatype & XFS_ALLOC_USERDATA_ZERO)
3772 		args.ip = ap->ip;
3773 
3774 	error = xfs_alloc_vextent(&args);
3775 	if (error)
3776 		return error;
3777 
3778 	if (tryagain && args.fsbno == NULLFSBLOCK) {
3779 		/*
3780 		 * Exact allocation failed. Now try with alignment
3781 		 * turned on.
3782 		 */
3783 		args.type = atype;
3784 		args.fsbno = ap->blkno;
3785 		args.alignment = stripe_align;
3786 		args.minlen = nextminlen;
3787 		args.minalignslop = 0;
3788 		isaligned = 1;
3789 		if ((error = xfs_alloc_vextent(&args)))
3790 			return error;
3791 	}
3792 	if (isaligned && args.fsbno == NULLFSBLOCK) {
3793 		/*
3794 		 * allocation failed, so turn off alignment and
3795 		 * try again.
3796 		 */
3797 		args.type = atype;
3798 		args.fsbno = ap->blkno;
3799 		args.alignment = 0;
3800 		if ((error = xfs_alloc_vextent(&args)))
3801 			return error;
3802 	}
3803 	if (args.fsbno == NULLFSBLOCK && nullfb &&
3804 	    args.minlen > ap->minlen) {
3805 		args.minlen = ap->minlen;
3806 		args.type = XFS_ALLOCTYPE_START_BNO;
3807 		args.fsbno = ap->blkno;
3808 		if ((error = xfs_alloc_vextent(&args)))
3809 			return error;
3810 	}
3811 	if (args.fsbno == NULLFSBLOCK && nullfb) {
3812 		args.fsbno = 0;
3813 		args.type = XFS_ALLOCTYPE_FIRST_AG;
3814 		args.total = ap->minlen;
3815 		if ((error = xfs_alloc_vextent(&args)))
3816 			return error;
3817 		ap->dfops->dop_low = true;
3818 	}
3819 	if (args.fsbno != NULLFSBLOCK) {
3820 		/*
3821 		 * check the allocation happened at the same or higher AG than
3822 		 * the first block that was allocated.
3823 		 */
3824 		ASSERT(*ap->firstblock == NULLFSBLOCK ||
3825 		       XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
3826 		       XFS_FSB_TO_AGNO(mp, args.fsbno) ||
3827 		       (ap->dfops->dop_low &&
3828 			XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
3829 			XFS_FSB_TO_AGNO(mp, args.fsbno)));
3830 
3831 		ap->blkno = args.fsbno;
3832 		if (*ap->firstblock == NULLFSBLOCK)
3833 			*ap->firstblock = args.fsbno;
3834 		ASSERT(nullfb || fb_agno == args.agno ||
3835 		       (ap->dfops->dop_low && fb_agno < args.agno));
3836 		ap->length = args.len;
3837 		if (!(ap->flags & XFS_BMAPI_COWFORK))
3838 			ap->ip->i_d.di_nblocks += args.len;
3839 		xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3840 		if (ap->wasdel)
3841 			ap->ip->i_delayed_blks -= args.len;
3842 		/*
3843 		 * Adjust the disk quota also. This was reserved
3844 		 * earlier.
3845 		 */
3846 		xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3847 			ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3848 					XFS_TRANS_DQ_BCOUNT,
3849 			(long) args.len);
3850 	} else {
3851 		ap->blkno = NULLFSBLOCK;
3852 		ap->length = 0;
3853 	}
3854 	return 0;
3855 }
3856 
3857 /*
3858  * For a remap operation, just "allocate" an extent at the address that the
3859  * caller passed in, and ensure that the AGFL is the right size.  The caller
3860  * will then map the "allocated" extent into the file somewhere.
3861  */
3862 STATIC int
3863 xfs_bmap_remap_alloc(
3864 	struct xfs_bmalloca	*ap)
3865 {
3866 	struct xfs_trans	*tp = ap->tp;
3867 	struct xfs_mount	*mp = tp->t_mountp;
3868 	xfs_agblock_t		bno;
3869 	struct xfs_alloc_arg	args;
3870 	int			error;
3871 
3872 	/*
3873 	 * validate that the block number is legal - the enables us to detect
3874 	 * and handle a silent filesystem corruption rather than crashing.
3875 	 */
3876 	memset(&args, 0, sizeof(struct xfs_alloc_arg));
3877 	args.tp = ap->tp;
3878 	args.mp = ap->tp->t_mountp;
3879 	bno = *ap->firstblock;
3880 	args.agno = XFS_FSB_TO_AGNO(mp, bno);
3881 	args.agbno = XFS_FSB_TO_AGBNO(mp, bno);
3882 	if (args.agno >= mp->m_sb.sb_agcount ||
3883 	    args.agbno >= mp->m_sb.sb_agblocks)
3884 		return -EFSCORRUPTED;
3885 
3886 	/* "Allocate" the extent from the range we passed in. */
3887 	trace_xfs_bmap_remap_alloc(ap->ip, *ap->firstblock, ap->length);
3888 	ap->blkno = bno;
3889 	ap->ip->i_d.di_nblocks += ap->length;
3890 	xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3891 
3892 	/* Fix the freelist, like a real allocator does. */
3893 	args.datatype = ap->datatype;
3894 	args.pag = xfs_perag_get(args.mp, args.agno);
3895 	ASSERT(args.pag);
3896 
3897 	/*
3898 	 * The freelist fixing code will decline the allocation if
3899 	 * the size and shape of the free space doesn't allow for
3900 	 * allocating the extent and updating all the metadata that
3901 	 * happens during an allocation.  We're remapping, not
3902 	 * allocating, so skip that check by pretending to be freeing.
3903 	 */
3904 	error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING);
3905 	xfs_perag_put(args.pag);
3906 	if (error)
3907 		trace_xfs_bmap_remap_alloc_error(ap->ip, error, _RET_IP_);
3908 	return error;
3909 }
3910 
3911 /*
3912  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3913  * It figures out where to ask the underlying allocator to put the new extent.
3914  */
3915 STATIC int
3916 xfs_bmap_alloc(
3917 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3918 {
3919 	if (ap->flags & XFS_BMAPI_REMAP)
3920 		return xfs_bmap_remap_alloc(ap);
3921 	if (XFS_IS_REALTIME_INODE(ap->ip) &&
3922 	    xfs_alloc_is_userdata(ap->datatype))
3923 		return xfs_bmap_rtalloc(ap);
3924 	return xfs_bmap_btalloc(ap);
3925 }
3926 
3927 /* Trim extent to fit a logical block range. */
3928 void
3929 xfs_trim_extent(
3930 	struct xfs_bmbt_irec	*irec,
3931 	xfs_fileoff_t		bno,
3932 	xfs_filblks_t		len)
3933 {
3934 	xfs_fileoff_t		distance;
3935 	xfs_fileoff_t		end = bno + len;
3936 
3937 	if (irec->br_startoff + irec->br_blockcount <= bno ||
3938 	    irec->br_startoff >= end) {
3939 		irec->br_blockcount = 0;
3940 		return;
3941 	}
3942 
3943 	if (irec->br_startoff < bno) {
3944 		distance = bno - irec->br_startoff;
3945 		if (isnullstartblock(irec->br_startblock))
3946 			irec->br_startblock = DELAYSTARTBLOCK;
3947 		if (irec->br_startblock != DELAYSTARTBLOCK &&
3948 		    irec->br_startblock != HOLESTARTBLOCK)
3949 			irec->br_startblock += distance;
3950 		irec->br_startoff += distance;
3951 		irec->br_blockcount -= distance;
3952 	}
3953 
3954 	if (end < irec->br_startoff + irec->br_blockcount) {
3955 		distance = irec->br_startoff + irec->br_blockcount - end;
3956 		irec->br_blockcount -= distance;
3957 	}
3958 }
3959 
3960 /*
3961  * Trim the returned map to the required bounds
3962  */
3963 STATIC void
3964 xfs_bmapi_trim_map(
3965 	struct xfs_bmbt_irec	*mval,
3966 	struct xfs_bmbt_irec	*got,
3967 	xfs_fileoff_t		*bno,
3968 	xfs_filblks_t		len,
3969 	xfs_fileoff_t		obno,
3970 	xfs_fileoff_t		end,
3971 	int			n,
3972 	int			flags)
3973 {
3974 	if ((flags & XFS_BMAPI_ENTIRE) ||
3975 	    got->br_startoff + got->br_blockcount <= obno) {
3976 		*mval = *got;
3977 		if (isnullstartblock(got->br_startblock))
3978 			mval->br_startblock = DELAYSTARTBLOCK;
3979 		return;
3980 	}
3981 
3982 	if (obno > *bno)
3983 		*bno = obno;
3984 	ASSERT((*bno >= obno) || (n == 0));
3985 	ASSERT(*bno < end);
3986 	mval->br_startoff = *bno;
3987 	if (isnullstartblock(got->br_startblock))
3988 		mval->br_startblock = DELAYSTARTBLOCK;
3989 	else
3990 		mval->br_startblock = got->br_startblock +
3991 					(*bno - got->br_startoff);
3992 	/*
3993 	 * Return the minimum of what we got and what we asked for for
3994 	 * the length.  We can use the len variable here because it is
3995 	 * modified below and we could have been there before coming
3996 	 * here if the first part of the allocation didn't overlap what
3997 	 * was asked for.
3998 	 */
3999 	mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
4000 			got->br_blockcount - (*bno - got->br_startoff));
4001 	mval->br_state = got->br_state;
4002 	ASSERT(mval->br_blockcount <= len);
4003 	return;
4004 }
4005 
4006 /*
4007  * Update and validate the extent map to return
4008  */
4009 STATIC void
4010 xfs_bmapi_update_map(
4011 	struct xfs_bmbt_irec	**map,
4012 	xfs_fileoff_t		*bno,
4013 	xfs_filblks_t		*len,
4014 	xfs_fileoff_t		obno,
4015 	xfs_fileoff_t		end,
4016 	int			*n,
4017 	int			flags)
4018 {
4019 	xfs_bmbt_irec_t	*mval = *map;
4020 
4021 	ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4022 	       ((mval->br_startoff + mval->br_blockcount) <= end));
4023 	ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
4024 	       (mval->br_startoff < obno));
4025 
4026 	*bno = mval->br_startoff + mval->br_blockcount;
4027 	*len = end - *bno;
4028 	if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4029 		/* update previous map with new information */
4030 		ASSERT(mval->br_startblock == mval[-1].br_startblock);
4031 		ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4032 		ASSERT(mval->br_state == mval[-1].br_state);
4033 		mval[-1].br_blockcount = mval->br_blockcount;
4034 		mval[-1].br_state = mval->br_state;
4035 	} else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4036 		   mval[-1].br_startblock != DELAYSTARTBLOCK &&
4037 		   mval[-1].br_startblock != HOLESTARTBLOCK &&
4038 		   mval->br_startblock == mval[-1].br_startblock +
4039 					  mval[-1].br_blockcount &&
4040 		   ((flags & XFS_BMAPI_IGSTATE) ||
4041 			mval[-1].br_state == mval->br_state)) {
4042 		ASSERT(mval->br_startoff ==
4043 		       mval[-1].br_startoff + mval[-1].br_blockcount);
4044 		mval[-1].br_blockcount += mval->br_blockcount;
4045 	} else if (*n > 0 &&
4046 		   mval->br_startblock == DELAYSTARTBLOCK &&
4047 		   mval[-1].br_startblock == DELAYSTARTBLOCK &&
4048 		   mval->br_startoff ==
4049 		   mval[-1].br_startoff + mval[-1].br_blockcount) {
4050 		mval[-1].br_blockcount += mval->br_blockcount;
4051 		mval[-1].br_state = mval->br_state;
4052 	} else if (!((*n == 0) &&
4053 		     ((mval->br_startoff + mval->br_blockcount) <=
4054 		      obno))) {
4055 		mval++;
4056 		(*n)++;
4057 	}
4058 	*map = mval;
4059 }
4060 
4061 /*
4062  * Map file blocks to filesystem blocks without allocation.
4063  */
4064 int
4065 xfs_bmapi_read(
4066 	struct xfs_inode	*ip,
4067 	xfs_fileoff_t		bno,
4068 	xfs_filblks_t		len,
4069 	struct xfs_bmbt_irec	*mval,
4070 	int			*nmap,
4071 	int			flags)
4072 {
4073 	struct xfs_mount	*mp = ip->i_mount;
4074 	struct xfs_ifork	*ifp;
4075 	struct xfs_bmbt_irec	got;
4076 	xfs_fileoff_t		obno;
4077 	xfs_fileoff_t		end;
4078 	xfs_extnum_t		idx;
4079 	int			error;
4080 	bool			eof = false;
4081 	int			n = 0;
4082 	int			whichfork = xfs_bmapi_whichfork(flags);
4083 
4084 	ASSERT(*nmap >= 1);
4085 	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4086 			   XFS_BMAPI_IGSTATE|XFS_BMAPI_COWFORK)));
4087 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
4088 
4089 	if (unlikely(XFS_TEST_ERROR(
4090 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4091 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4092 	     mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4093 		XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4094 		return -EFSCORRUPTED;
4095 	}
4096 
4097 	if (XFS_FORCED_SHUTDOWN(mp))
4098 		return -EIO;
4099 
4100 	XFS_STATS_INC(mp, xs_blk_mapr);
4101 
4102 	ifp = XFS_IFORK_PTR(ip, whichfork);
4103 
4104 	/* No CoW fork?  Return a hole. */
4105 	if (whichfork == XFS_COW_FORK && !ifp) {
4106 		mval->br_startoff = bno;
4107 		mval->br_startblock = HOLESTARTBLOCK;
4108 		mval->br_blockcount = len;
4109 		mval->br_state = XFS_EXT_NORM;
4110 		*nmap = 1;
4111 		return 0;
4112 	}
4113 
4114 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4115 		error = xfs_iread_extents(NULL, ip, whichfork);
4116 		if (error)
4117 			return error;
4118 	}
4119 
4120 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &idx, &got))
4121 		eof = true;
4122 	end = bno + len;
4123 	obno = bno;
4124 
4125 	while (bno < end && n < *nmap) {
4126 		/* Reading past eof, act as though there's a hole up to end. */
4127 		if (eof)
4128 			got.br_startoff = end;
4129 		if (got.br_startoff > bno) {
4130 			/* Reading in a hole.  */
4131 			mval->br_startoff = bno;
4132 			mval->br_startblock = HOLESTARTBLOCK;
4133 			mval->br_blockcount =
4134 				XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4135 			mval->br_state = XFS_EXT_NORM;
4136 			bno += mval->br_blockcount;
4137 			len -= mval->br_blockcount;
4138 			mval++;
4139 			n++;
4140 			continue;
4141 		}
4142 
4143 		/* set up the extent map to return. */
4144 		xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4145 		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4146 
4147 		/* If we're done, stop now. */
4148 		if (bno >= end || n >= *nmap)
4149 			break;
4150 
4151 		/* Else go on to the next record. */
4152 		if (!xfs_iext_get_extent(ifp, ++idx, &got))
4153 			eof = true;
4154 	}
4155 	*nmap = n;
4156 	return 0;
4157 }
4158 
4159 int
4160 xfs_bmapi_reserve_delalloc(
4161 	struct xfs_inode	*ip,
4162 	int			whichfork,
4163 	xfs_fileoff_t		off,
4164 	xfs_filblks_t		len,
4165 	xfs_filblks_t		prealloc,
4166 	struct xfs_bmbt_irec	*got,
4167 	xfs_extnum_t		*lastx,
4168 	int			eof)
4169 {
4170 	struct xfs_mount	*mp = ip->i_mount;
4171 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
4172 	xfs_extlen_t		alen;
4173 	xfs_extlen_t		indlen;
4174 	char			rt = XFS_IS_REALTIME_INODE(ip);
4175 	xfs_extlen_t		extsz;
4176 	int			error;
4177 	xfs_fileoff_t		aoff = off;
4178 
4179 	/*
4180 	 * Cap the alloc length. Keep track of prealloc so we know whether to
4181 	 * tag the inode before we return.
4182 	 */
4183 	alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN);
4184 	if (!eof)
4185 		alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4186 	if (prealloc && alen >= len)
4187 		prealloc = alen - len;
4188 
4189 	/* Figure out the extent size, adjust alen */
4190 	if (whichfork == XFS_COW_FORK)
4191 		extsz = xfs_get_cowextsz_hint(ip);
4192 	else
4193 		extsz = xfs_get_extsz_hint(ip);
4194 	if (extsz) {
4195 		struct xfs_bmbt_irec	prev;
4196 
4197 		if (!xfs_iext_get_extent(ifp, *lastx - 1, &prev))
4198 			prev.br_startoff = NULLFILEOFF;
4199 
4200 		error = xfs_bmap_extsize_align(mp, got, &prev, extsz, rt, eof,
4201 					       1, 0, &aoff, &alen);
4202 		ASSERT(!error);
4203 	}
4204 
4205 	if (rt)
4206 		extsz = alen / mp->m_sb.sb_rextsize;
4207 
4208 	/*
4209 	 * Make a transaction-less quota reservation for delayed allocation
4210 	 * blocks.  This number gets adjusted later.  We return if we haven't
4211 	 * allocated blocks already inside this loop.
4212 	 */
4213 	error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4214 			rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4215 	if (error)
4216 		return error;
4217 
4218 	/*
4219 	 * Split changing sb for alen and indlen since they could be coming
4220 	 * from different places.
4221 	 */
4222 	indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4223 	ASSERT(indlen > 0);
4224 
4225 	if (rt) {
4226 		error = xfs_mod_frextents(mp, -((int64_t)extsz));
4227 	} else {
4228 		error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
4229 	}
4230 
4231 	if (error)
4232 		goto out_unreserve_quota;
4233 
4234 	error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
4235 	if (error)
4236 		goto out_unreserve_blocks;
4237 
4238 
4239 	ip->i_delayed_blks += alen;
4240 
4241 	got->br_startoff = aoff;
4242 	got->br_startblock = nullstartblock(indlen);
4243 	got->br_blockcount = alen;
4244 	got->br_state = XFS_EXT_NORM;
4245 	xfs_bmap_add_extent_hole_delay(ip, whichfork, lastx, got);
4246 
4247 	/*
4248 	 * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4249 	 * might have merged it into one of the neighbouring ones.
4250 	 */
4251 	xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4252 
4253 	/*
4254 	 * Tag the inode if blocks were preallocated. Note that COW fork
4255 	 * preallocation can occur at the start or end of the extent, even when
4256 	 * prealloc == 0, so we must also check the aligned offset and length.
4257 	 */
4258 	if (whichfork == XFS_DATA_FORK && prealloc)
4259 		xfs_inode_set_eofblocks_tag(ip);
4260 	if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len))
4261 		xfs_inode_set_cowblocks_tag(ip);
4262 
4263 	ASSERT(got->br_startoff <= aoff);
4264 	ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4265 	ASSERT(isnullstartblock(got->br_startblock));
4266 	ASSERT(got->br_state == XFS_EXT_NORM);
4267 	return 0;
4268 
4269 out_unreserve_blocks:
4270 	if (rt)
4271 		xfs_mod_frextents(mp, extsz);
4272 	else
4273 		xfs_mod_fdblocks(mp, alen, false);
4274 out_unreserve_quota:
4275 	if (XFS_IS_QUOTA_ON(mp))
4276 		xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4277 				XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4278 	return error;
4279 }
4280 
4281 static int
4282 xfs_bmapi_allocate(
4283 	struct xfs_bmalloca	*bma)
4284 {
4285 	struct xfs_mount	*mp = bma->ip->i_mount;
4286 	int			whichfork = xfs_bmapi_whichfork(bma->flags);
4287 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4288 	int			tmp_logflags = 0;
4289 	int			error;
4290 
4291 	ASSERT(bma->length > 0);
4292 
4293 	/*
4294 	 * For the wasdelay case, we could also just allocate the stuff asked
4295 	 * for in this bmap call but that wouldn't be as good.
4296 	 */
4297 	if (bma->wasdel) {
4298 		bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4299 		bma->offset = bma->got.br_startoff;
4300 		if (bma->idx) {
4301 			xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4302 					 &bma->prev);
4303 		}
4304 	} else {
4305 		bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4306 		if (!bma->eof)
4307 			bma->length = XFS_FILBLKS_MIN(bma->length,
4308 					bma->got.br_startoff - bma->offset);
4309 	}
4310 
4311 	/*
4312 	 * Set the data type being allocated. For the data fork, the first data
4313 	 * in the file is treated differently to all other allocations. For the
4314 	 * attribute fork, we only need to ensure the allocated range is not on
4315 	 * the busy list.
4316 	 */
4317 	if (!(bma->flags & XFS_BMAPI_METADATA)) {
4318 		bma->datatype = XFS_ALLOC_NOBUSY;
4319 		if (whichfork == XFS_DATA_FORK) {
4320 			if (bma->offset == 0)
4321 				bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA;
4322 			else
4323 				bma->datatype |= XFS_ALLOC_USERDATA;
4324 		}
4325 		if (bma->flags & XFS_BMAPI_ZERO)
4326 			bma->datatype |= XFS_ALLOC_USERDATA_ZERO;
4327 	}
4328 
4329 	bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4330 
4331 	/*
4332 	 * Only want to do the alignment at the eof if it is userdata and
4333 	 * allocation length is larger than a stripe unit.
4334 	 */
4335 	if (mp->m_dalign && bma->length >= mp->m_dalign &&
4336 	    !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4337 		error = xfs_bmap_isaeof(bma, whichfork);
4338 		if (error)
4339 			return error;
4340 	}
4341 
4342 	error = xfs_bmap_alloc(bma);
4343 	if (error)
4344 		return error;
4345 
4346 	if (bma->cur)
4347 		bma->cur->bc_private.b.firstblock = *bma->firstblock;
4348 	if (bma->blkno == NULLFSBLOCK)
4349 		return 0;
4350 	if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4351 		bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4352 		bma->cur->bc_private.b.firstblock = *bma->firstblock;
4353 		bma->cur->bc_private.b.dfops = bma->dfops;
4354 	}
4355 	/*
4356 	 * Bump the number of extents we've allocated
4357 	 * in this call.
4358 	 */
4359 	bma->nallocs++;
4360 
4361 	if (bma->cur)
4362 		bma->cur->bc_private.b.flags =
4363 			bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4364 
4365 	bma->got.br_startoff = bma->offset;
4366 	bma->got.br_startblock = bma->blkno;
4367 	bma->got.br_blockcount = bma->length;
4368 	bma->got.br_state = XFS_EXT_NORM;
4369 
4370 	/*
4371 	 * A wasdelay extent has been initialized, so shouldn't be flagged
4372 	 * as unwritten.
4373 	 */
4374 	if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4375 	    xfs_sb_version_hasextflgbit(&mp->m_sb))
4376 		bma->got.br_state = XFS_EXT_UNWRITTEN;
4377 
4378 	if (bma->wasdel)
4379 		error = xfs_bmap_add_extent_delay_real(bma, whichfork);
4380 	else
4381 		error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4382 
4383 	bma->logflags |= tmp_logflags;
4384 	if (error)
4385 		return error;
4386 
4387 	/*
4388 	 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4389 	 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4390 	 * the neighbouring ones.
4391 	 */
4392 	xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4393 
4394 	ASSERT(bma->got.br_startoff <= bma->offset);
4395 	ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4396 	       bma->offset + bma->length);
4397 	ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4398 	       bma->got.br_state == XFS_EXT_UNWRITTEN);
4399 	return 0;
4400 }
4401 
4402 STATIC int
4403 xfs_bmapi_convert_unwritten(
4404 	struct xfs_bmalloca	*bma,
4405 	struct xfs_bmbt_irec	*mval,
4406 	xfs_filblks_t		len,
4407 	int			flags)
4408 {
4409 	int			whichfork = xfs_bmapi_whichfork(flags);
4410 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4411 	int			tmp_logflags = 0;
4412 	int			error;
4413 
4414 	/* check if we need to do unwritten->real conversion */
4415 	if (mval->br_state == XFS_EXT_UNWRITTEN &&
4416 	    (flags & XFS_BMAPI_PREALLOC))
4417 		return 0;
4418 
4419 	/* check if we need to do real->unwritten conversion */
4420 	if (mval->br_state == XFS_EXT_NORM &&
4421 	    (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4422 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4423 		return 0;
4424 
4425 	ASSERT(whichfork != XFS_COW_FORK);
4426 
4427 	/*
4428 	 * Modify (by adding) the state flag, if writing.
4429 	 */
4430 	ASSERT(mval->br_blockcount <= len);
4431 	if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4432 		bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4433 					bma->ip, whichfork);
4434 		bma->cur->bc_private.b.firstblock = *bma->firstblock;
4435 		bma->cur->bc_private.b.dfops = bma->dfops;
4436 	}
4437 	mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4438 				? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4439 
4440 	/*
4441 	 * Before insertion into the bmbt, zero the range being converted
4442 	 * if required.
4443 	 */
4444 	if (flags & XFS_BMAPI_ZERO) {
4445 		error = xfs_zero_extent(bma->ip, mval->br_startblock,
4446 					mval->br_blockcount);
4447 		if (error)
4448 			return error;
4449 	}
4450 
4451 	error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4452 			&bma->cur, mval, bma->firstblock, bma->dfops,
4453 			&tmp_logflags);
4454 	/*
4455 	 * Log the inode core unconditionally in the unwritten extent conversion
4456 	 * path because the conversion might not have done so (e.g., if the
4457 	 * extent count hasn't changed). We need to make sure the inode is dirty
4458 	 * in the transaction for the sake of fsync(), even if nothing has
4459 	 * changed, because fsync() will not force the log for this transaction
4460 	 * unless it sees the inode pinned.
4461 	 */
4462 	bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4463 	if (error)
4464 		return error;
4465 
4466 	/*
4467 	 * Update our extent pointer, given that
4468 	 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4469 	 * of the neighbouring ones.
4470 	 */
4471 	xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4472 
4473 	/*
4474 	 * We may have combined previously unwritten space with written space,
4475 	 * so generate another request.
4476 	 */
4477 	if (mval->br_blockcount < len)
4478 		return -EAGAIN;
4479 	return 0;
4480 }
4481 
4482 /*
4483  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4484  * extent state if necessary.  Details behaviour is controlled by the flags
4485  * parameter.  Only allocates blocks from a single allocation group, to avoid
4486  * locking problems.
4487  *
4488  * The returned value in "firstblock" from the first call in a transaction
4489  * must be remembered and presented to subsequent calls in "firstblock".
4490  * An upper bound for the number of blocks to be allocated is supplied to
4491  * the first call in "total"; if no allocation group has that many free
4492  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4493  */
4494 int
4495 xfs_bmapi_write(
4496 	struct xfs_trans	*tp,		/* transaction pointer */
4497 	struct xfs_inode	*ip,		/* incore inode */
4498 	xfs_fileoff_t		bno,		/* starting file offs. mapped */
4499 	xfs_filblks_t		len,		/* length to map in file */
4500 	int			flags,		/* XFS_BMAPI_... */
4501 	xfs_fsblock_t		*firstblock,	/* first allocated block
4502 						   controls a.g. for allocs */
4503 	xfs_extlen_t		total,		/* total blocks needed */
4504 	struct xfs_bmbt_irec	*mval,		/* output: map values */
4505 	int			*nmap,		/* i/o: mval size/count */
4506 	struct xfs_defer_ops	*dfops)		/* i/o: list extents to free */
4507 {
4508 	struct xfs_mount	*mp = ip->i_mount;
4509 	struct xfs_ifork	*ifp;
4510 	struct xfs_bmalloca	bma = { NULL };	/* args for xfs_bmap_alloc */
4511 	xfs_fileoff_t		end;		/* end of mapped file region */
4512 	bool			eof = false;	/* after the end of extents */
4513 	int			error;		/* error return */
4514 	int			n;		/* current extent index */
4515 	xfs_fileoff_t		obno;		/* old block number (offset) */
4516 	int			whichfork;	/* data or attr fork */
4517 	char			inhole;		/* current location is hole in file */
4518 	char			wasdelay;	/* old extent was delayed */
4519 
4520 #ifdef DEBUG
4521 	xfs_fileoff_t		orig_bno;	/* original block number value */
4522 	int			orig_flags;	/* original flags arg value */
4523 	xfs_filblks_t		orig_len;	/* original value of len arg */
4524 	struct xfs_bmbt_irec	*orig_mval;	/* original value of mval */
4525 	int			orig_nmap;	/* original value of *nmap */
4526 
4527 	orig_bno = bno;
4528 	orig_len = len;
4529 	orig_flags = flags;
4530 	orig_mval = mval;
4531 	orig_nmap = *nmap;
4532 #endif
4533 	whichfork = xfs_bmapi_whichfork(flags);
4534 
4535 	ASSERT(*nmap >= 1);
4536 	ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4537 	ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4538 	ASSERT(tp != NULL);
4539 	ASSERT(len > 0);
4540 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4541 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4542 	ASSERT(!(flags & XFS_BMAPI_REMAP) || whichfork == XFS_DATA_FORK);
4543 	ASSERT(!(flags & XFS_BMAPI_PREALLOC) || !(flags & XFS_BMAPI_REMAP));
4544 	ASSERT(!(flags & XFS_BMAPI_CONVERT) || !(flags & XFS_BMAPI_REMAP));
4545 	ASSERT(!(flags & XFS_BMAPI_PREALLOC) || whichfork != XFS_COW_FORK);
4546 	ASSERT(!(flags & XFS_BMAPI_CONVERT) || whichfork != XFS_COW_FORK);
4547 
4548 	/* zeroing is for currently only for data extents, not metadata */
4549 	ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4550 			(XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4551 	/*
4552 	 * we can allocate unwritten extents or pre-zero allocated blocks,
4553 	 * but it makes no sense to do both at once. This would result in
4554 	 * zeroing the unwritten extent twice, but it still being an
4555 	 * unwritten extent....
4556 	 */
4557 	ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4558 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4559 
4560 	if (unlikely(XFS_TEST_ERROR(
4561 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4562 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4563 	     mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4564 		XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4565 		return -EFSCORRUPTED;
4566 	}
4567 
4568 	if (XFS_FORCED_SHUTDOWN(mp))
4569 		return -EIO;
4570 
4571 	ifp = XFS_IFORK_PTR(ip, whichfork);
4572 
4573 	XFS_STATS_INC(mp, xs_blk_mapw);
4574 
4575 	if (*firstblock == NULLFSBLOCK) {
4576 		if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4577 			bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4578 		else
4579 			bma.minleft = 1;
4580 	} else {
4581 		bma.minleft = 0;
4582 	}
4583 
4584 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4585 		error = xfs_iread_extents(tp, ip, whichfork);
4586 		if (error)
4587 			goto error0;
4588 	}
4589 
4590 	n = 0;
4591 	end = bno + len;
4592 	obno = bno;
4593 
4594 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.idx, &bma.got))
4595 		eof = true;
4596 	if (!xfs_iext_get_extent(ifp, bma.idx - 1, &bma.prev))
4597 		bma.prev.br_startoff = NULLFILEOFF;
4598 	bma.tp = tp;
4599 	bma.ip = ip;
4600 	bma.total = total;
4601 	bma.datatype = 0;
4602 	bma.dfops = dfops;
4603 	bma.firstblock = firstblock;
4604 
4605 	while (bno < end && n < *nmap) {
4606 		inhole = eof || bma.got.br_startoff > bno;
4607 		wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4608 
4609 		/*
4610 		 * Make sure we only reflink into a hole.
4611 		 */
4612 		if (flags & XFS_BMAPI_REMAP)
4613 			ASSERT(inhole);
4614 		if (flags & XFS_BMAPI_COWFORK)
4615 			ASSERT(!inhole);
4616 
4617 		/*
4618 		 * First, deal with the hole before the allocated space
4619 		 * that we found, if any.
4620 		 */
4621 		if (inhole || wasdelay) {
4622 			bma.eof = eof;
4623 			bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4624 			bma.wasdel = wasdelay;
4625 			bma.offset = bno;
4626 			bma.flags = flags;
4627 
4628 			/*
4629 			 * There's a 32/64 bit type mismatch between the
4630 			 * allocation length request (which can be 64 bits in
4631 			 * length) and the bma length request, which is
4632 			 * xfs_extlen_t and therefore 32 bits. Hence we have to
4633 			 * check for 32-bit overflows and handle them here.
4634 			 */
4635 			if (len > (xfs_filblks_t)MAXEXTLEN)
4636 				bma.length = MAXEXTLEN;
4637 			else
4638 				bma.length = len;
4639 
4640 			ASSERT(len > 0);
4641 			ASSERT(bma.length > 0);
4642 			error = xfs_bmapi_allocate(&bma);
4643 			if (error)
4644 				goto error0;
4645 			if (bma.blkno == NULLFSBLOCK)
4646 				break;
4647 
4648 			/*
4649 			 * If this is a CoW allocation, record the data in
4650 			 * the refcount btree for orphan recovery.
4651 			 */
4652 			if (whichfork == XFS_COW_FORK) {
4653 				error = xfs_refcount_alloc_cow_extent(mp, dfops,
4654 						bma.blkno, bma.length);
4655 				if (error)
4656 					goto error0;
4657 			}
4658 		}
4659 
4660 		/* Deal with the allocated space we found.  */
4661 		xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4662 							end, n, flags);
4663 
4664 		/* Execute unwritten extent conversion if necessary */
4665 		error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4666 		if (error == -EAGAIN)
4667 			continue;
4668 		if (error)
4669 			goto error0;
4670 
4671 		/* update the extent map to return */
4672 		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4673 
4674 		/*
4675 		 * If we're done, stop now.  Stop when we've allocated
4676 		 * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4677 		 * the transaction may get too big.
4678 		 */
4679 		if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4680 			break;
4681 
4682 		/* Else go on to the next record. */
4683 		bma.prev = bma.got;
4684 		if (!xfs_iext_get_extent(ifp, ++bma.idx, &bma.got))
4685 			eof = true;
4686 	}
4687 	*nmap = n;
4688 
4689 	/*
4690 	 * Transform from btree to extents, give it cur.
4691 	 */
4692 	if (xfs_bmap_wants_extents(ip, whichfork)) {
4693 		int		tmp_logflags = 0;
4694 
4695 		ASSERT(bma.cur);
4696 		error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4697 			&tmp_logflags, whichfork);
4698 		bma.logflags |= tmp_logflags;
4699 		if (error)
4700 			goto error0;
4701 	}
4702 
4703 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4704 	       XFS_IFORK_NEXTENTS(ip, whichfork) >
4705 		XFS_IFORK_MAXEXT(ip, whichfork));
4706 	error = 0;
4707 error0:
4708 	/*
4709 	 * Log everything.  Do this after conversion, there's no point in
4710 	 * logging the extent records if we've converted to btree format.
4711 	 */
4712 	if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4713 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4714 		bma.logflags &= ~xfs_ilog_fext(whichfork);
4715 	else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4716 		 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4717 		bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4718 	/*
4719 	 * Log whatever the flags say, even if error.  Otherwise we might miss
4720 	 * detecting a case where the data is changed, there's an error,
4721 	 * and it's not logged so we don't shutdown when we should.
4722 	 */
4723 	if (bma.logflags)
4724 		xfs_trans_log_inode(tp, ip, bma.logflags);
4725 
4726 	if (bma.cur) {
4727 		if (!error) {
4728 			ASSERT(*firstblock == NULLFSBLOCK ||
4729 			       XFS_FSB_TO_AGNO(mp, *firstblock) ==
4730 			       XFS_FSB_TO_AGNO(mp,
4731 				       bma.cur->bc_private.b.firstblock) ||
4732 			       (dfops->dop_low &&
4733 				XFS_FSB_TO_AGNO(mp, *firstblock) <
4734 				XFS_FSB_TO_AGNO(mp,
4735 					bma.cur->bc_private.b.firstblock)));
4736 			*firstblock = bma.cur->bc_private.b.firstblock;
4737 		}
4738 		xfs_btree_del_cursor(bma.cur,
4739 			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4740 	}
4741 	if (!error)
4742 		xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4743 			orig_nmap, *nmap);
4744 	return error;
4745 }
4746 
4747 /*
4748  * When a delalloc extent is split (e.g., due to a hole punch), the original
4749  * indlen reservation must be shared across the two new extents that are left
4750  * behind.
4751  *
4752  * Given the original reservation and the worst case indlen for the two new
4753  * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4754  * reservation fairly across the two new extents. If necessary, steal available
4755  * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4756  * ores == 1). The number of stolen blocks is returned. The availability and
4757  * subsequent accounting of stolen blocks is the responsibility of the caller.
4758  */
4759 static xfs_filblks_t
4760 xfs_bmap_split_indlen(
4761 	xfs_filblks_t			ores,		/* original res. */
4762 	xfs_filblks_t			*indlen1,	/* ext1 worst indlen */
4763 	xfs_filblks_t			*indlen2,	/* ext2 worst indlen */
4764 	xfs_filblks_t			avail)		/* stealable blocks */
4765 {
4766 	xfs_filblks_t			len1 = *indlen1;
4767 	xfs_filblks_t			len2 = *indlen2;
4768 	xfs_filblks_t			nres = len1 + len2; /* new total res. */
4769 	xfs_filblks_t			stolen = 0;
4770 
4771 	/*
4772 	 * Steal as many blocks as we can to try and satisfy the worst case
4773 	 * indlen for both new extents.
4774 	 */
4775 	while (nres > ores && avail) {
4776 		nres--;
4777 		avail--;
4778 		stolen++;
4779 	}
4780 
4781 	/*
4782 	 * The only blocks available are those reserved for the original
4783 	 * extent and what we can steal from the extent being removed.
4784 	 * If this still isn't enough to satisfy the combined
4785 	 * requirements for the two new extents, skim blocks off of each
4786 	 * of the new reservations until they match what is available.
4787 	 */
4788 	while (nres > ores) {
4789 		if (len1) {
4790 			len1--;
4791 			nres--;
4792 		}
4793 		if (nres == ores)
4794 			break;
4795 		if (len2) {
4796 			len2--;
4797 			nres--;
4798 		}
4799 	}
4800 
4801 	*indlen1 = len1;
4802 	*indlen2 = len2;
4803 
4804 	return stolen;
4805 }
4806 
4807 int
4808 xfs_bmap_del_extent_delay(
4809 	struct xfs_inode	*ip,
4810 	int			whichfork,
4811 	xfs_extnum_t		*idx,
4812 	struct xfs_bmbt_irec	*got,
4813 	struct xfs_bmbt_irec	*del)
4814 {
4815 	struct xfs_mount	*mp = ip->i_mount;
4816 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
4817 	struct xfs_bmbt_irec	new;
4818 	int64_t			da_old, da_new, da_diff = 0;
4819 	xfs_fileoff_t		del_endoff, got_endoff;
4820 	xfs_filblks_t		got_indlen, new_indlen, stolen;
4821 	int			error = 0, state = 0;
4822 	bool			isrt;
4823 
4824 	XFS_STATS_INC(mp, xs_del_exlist);
4825 
4826 	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4827 	del_endoff = del->br_startoff + del->br_blockcount;
4828 	got_endoff = got->br_startoff + got->br_blockcount;
4829 	da_old = startblockval(got->br_startblock);
4830 	da_new = 0;
4831 
4832 	ASSERT(*idx >= 0);
4833 	ASSERT(*idx <= xfs_iext_count(ifp));
4834 	ASSERT(del->br_blockcount > 0);
4835 	ASSERT(got->br_startoff <= del->br_startoff);
4836 	ASSERT(got_endoff >= del_endoff);
4837 
4838 	if (isrt) {
4839 		int64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount);
4840 
4841 		do_div(rtexts, mp->m_sb.sb_rextsize);
4842 		xfs_mod_frextents(mp, rtexts);
4843 	}
4844 
4845 	/*
4846 	 * Update the inode delalloc counter now and wait to update the
4847 	 * sb counters as we might have to borrow some blocks for the
4848 	 * indirect block accounting.
4849 	 */
4850 	error = xfs_trans_reserve_quota_nblks(NULL, ip,
4851 			-((long)del->br_blockcount), 0,
4852 			isrt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4853 	if (error)
4854 		return error;
4855 	ip->i_delayed_blks -= del->br_blockcount;
4856 
4857 	if (whichfork == XFS_COW_FORK)
4858 		state |= BMAP_COWFORK;
4859 
4860 	if (got->br_startoff == del->br_startoff)
4861 		state |= BMAP_LEFT_CONTIG;
4862 	if (got_endoff == del_endoff)
4863 		state |= BMAP_RIGHT_CONTIG;
4864 
4865 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
4866 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
4867 		/*
4868 		 * Matches the whole extent.  Delete the entry.
4869 		 */
4870 		xfs_iext_remove(ip, *idx, 1, state);
4871 		--*idx;
4872 		break;
4873 	case BMAP_LEFT_CONTIG:
4874 		/*
4875 		 * Deleting the first part of the extent.
4876 		 */
4877 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4878 		got->br_startoff = del_endoff;
4879 		got->br_blockcount -= del->br_blockcount;
4880 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4881 				got->br_blockcount), da_old);
4882 		got->br_startblock = nullstartblock((int)da_new);
4883 		xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
4884 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4885 		break;
4886 	case BMAP_RIGHT_CONTIG:
4887 		/*
4888 		 * Deleting the last part of the extent.
4889 		 */
4890 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4891 		got->br_blockcount = got->br_blockcount - del->br_blockcount;
4892 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4893 				got->br_blockcount), da_old);
4894 		got->br_startblock = nullstartblock((int)da_new);
4895 		xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
4896 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4897 		break;
4898 	case 0:
4899 		/*
4900 		 * Deleting the middle of the extent.
4901 		 *
4902 		 * Distribute the original indlen reservation across the two new
4903 		 * extents.  Steal blocks from the deleted extent if necessary.
4904 		 * Stealing blocks simply fudges the fdblocks accounting below.
4905 		 * Warn if either of the new indlen reservations is zero as this
4906 		 * can lead to delalloc problems.
4907 		 */
4908 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4909 
4910 		got->br_blockcount = del->br_startoff - got->br_startoff;
4911 		got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount);
4912 
4913 		new.br_blockcount = got_endoff - del_endoff;
4914 		new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount);
4915 
4916 		WARN_ON_ONCE(!got_indlen || !new_indlen);
4917 		stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen,
4918 						       del->br_blockcount);
4919 
4920 		got->br_startblock = nullstartblock((int)got_indlen);
4921 		xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
4922 		trace_xfs_bmap_post_update(ip, *idx, 0, _THIS_IP_);
4923 
4924 		new.br_startoff = del_endoff;
4925 		new.br_state = got->br_state;
4926 		new.br_startblock = nullstartblock((int)new_indlen);
4927 
4928 		++*idx;
4929 		xfs_iext_insert(ip, *idx, 1, &new, state);
4930 
4931 		da_new = got_indlen + new_indlen - stolen;
4932 		del->br_blockcount -= stolen;
4933 		break;
4934 	}
4935 
4936 	ASSERT(da_old >= da_new);
4937 	da_diff = da_old - da_new;
4938 	if (!isrt)
4939 		da_diff += del->br_blockcount;
4940 	if (da_diff)
4941 		xfs_mod_fdblocks(mp, da_diff, false);
4942 	return error;
4943 }
4944 
4945 void
4946 xfs_bmap_del_extent_cow(
4947 	struct xfs_inode	*ip,
4948 	xfs_extnum_t		*idx,
4949 	struct xfs_bmbt_irec	*got,
4950 	struct xfs_bmbt_irec	*del)
4951 {
4952 	struct xfs_mount	*mp = ip->i_mount;
4953 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
4954 	struct xfs_bmbt_irec	new;
4955 	xfs_fileoff_t		del_endoff, got_endoff;
4956 	int			state = BMAP_COWFORK;
4957 
4958 	XFS_STATS_INC(mp, xs_del_exlist);
4959 
4960 	del_endoff = del->br_startoff + del->br_blockcount;
4961 	got_endoff = got->br_startoff + got->br_blockcount;
4962 
4963 	ASSERT(*idx >= 0);
4964 	ASSERT(*idx <= xfs_iext_count(ifp));
4965 	ASSERT(del->br_blockcount > 0);
4966 	ASSERT(got->br_startoff <= del->br_startoff);
4967 	ASSERT(got_endoff >= del_endoff);
4968 	ASSERT(!isnullstartblock(got->br_startblock));
4969 
4970 	if (got->br_startoff == del->br_startoff)
4971 		state |= BMAP_LEFT_CONTIG;
4972 	if (got_endoff == del_endoff)
4973 		state |= BMAP_RIGHT_CONTIG;
4974 
4975 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
4976 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
4977 		/*
4978 		 * Matches the whole extent.  Delete the entry.
4979 		 */
4980 		xfs_iext_remove(ip, *idx, 1, state);
4981 		--*idx;
4982 		break;
4983 	case BMAP_LEFT_CONTIG:
4984 		/*
4985 		 * Deleting the first part of the extent.
4986 		 */
4987 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4988 		got->br_startoff = del_endoff;
4989 		got->br_blockcount -= del->br_blockcount;
4990 		got->br_startblock = del->br_startblock + del->br_blockcount;
4991 		xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
4992 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4993 		break;
4994 	case BMAP_RIGHT_CONTIG:
4995 		/*
4996 		 * Deleting the last part of the extent.
4997 		 */
4998 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4999 		got->br_blockcount -= del->br_blockcount;
5000 		xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
5001 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5002 		break;
5003 	case 0:
5004 		/*
5005 		 * Deleting the middle of the extent.
5006 		 */
5007 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5008 		got->br_blockcount = del->br_startoff - got->br_startoff;
5009 		xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
5010 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5011 
5012 		new.br_startoff = del_endoff;
5013 		new.br_blockcount = got_endoff - del_endoff;
5014 		new.br_state = got->br_state;
5015 		new.br_startblock = del->br_startblock + del->br_blockcount;
5016 
5017 		++*idx;
5018 		xfs_iext_insert(ip, *idx, 1, &new, state);
5019 		break;
5020 	}
5021 }
5022 
5023 /*
5024  * Called by xfs_bmapi to update file extent records and the btree
5025  * after removing space (or undoing a delayed allocation).
5026  */
5027 STATIC int				/* error */
5028 xfs_bmap_del_extent(
5029 	xfs_inode_t		*ip,	/* incore inode pointer */
5030 	xfs_trans_t		*tp,	/* current transaction pointer */
5031 	xfs_extnum_t		*idx,	/* extent number to update/delete */
5032 	struct xfs_defer_ops	*dfops,	/* list of extents to be freed */
5033 	xfs_btree_cur_t		*cur,	/* if null, not a btree */
5034 	xfs_bmbt_irec_t		*del,	/* data to remove from extents */
5035 	int			*logflagsp, /* inode logging flags */
5036 	int			whichfork, /* data or attr fork */
5037 	int			bflags)	/* bmapi flags */
5038 {
5039 	xfs_filblks_t		da_new;	/* new delay-alloc indirect blocks */
5040 	xfs_filblks_t		da_old;	/* old delay-alloc indirect blocks */
5041 	xfs_fsblock_t		del_endblock=0;	/* first block past del */
5042 	xfs_fileoff_t		del_endoff;	/* first offset past del */
5043 	int			delay;	/* current block is delayed allocated */
5044 	int			do_fx;	/* free extent at end of routine */
5045 	xfs_bmbt_rec_host_t	*ep;	/* current extent entry pointer */
5046 	int			error;	/* error return value */
5047 	int			flags;	/* inode logging flags */
5048 	xfs_bmbt_irec_t		got;	/* current extent entry */
5049 	xfs_fileoff_t		got_endoff;	/* first offset past got */
5050 	int			i;	/* temp state */
5051 	xfs_ifork_t		*ifp;	/* inode fork pointer */
5052 	xfs_mount_t		*mp;	/* mount structure */
5053 	xfs_filblks_t		nblks;	/* quota/sb block count */
5054 	xfs_bmbt_irec_t		new;	/* new record to be inserted */
5055 	/* REFERENCED */
5056 	uint			qfield;	/* quota field to update */
5057 	xfs_filblks_t		temp;	/* for indirect length calculations */
5058 	xfs_filblks_t		temp2;	/* for indirect length calculations */
5059 	int			state = 0;
5060 
5061 	mp = ip->i_mount;
5062 	XFS_STATS_INC(mp, xs_del_exlist);
5063 
5064 	if (whichfork == XFS_ATTR_FORK)
5065 		state |= BMAP_ATTRFORK;
5066 	else if (whichfork == XFS_COW_FORK)
5067 		state |= BMAP_COWFORK;
5068 
5069 	ifp = XFS_IFORK_PTR(ip, whichfork);
5070 	ASSERT((*idx >= 0) && (*idx < xfs_iext_count(ifp)));
5071 	ASSERT(del->br_blockcount > 0);
5072 	ep = xfs_iext_get_ext(ifp, *idx);
5073 	xfs_bmbt_get_all(ep, &got);
5074 	ASSERT(got.br_startoff <= del->br_startoff);
5075 	del_endoff = del->br_startoff + del->br_blockcount;
5076 	got_endoff = got.br_startoff + got.br_blockcount;
5077 	ASSERT(got_endoff >= del_endoff);
5078 	delay = isnullstartblock(got.br_startblock);
5079 	ASSERT(isnullstartblock(del->br_startblock) == delay);
5080 	flags = 0;
5081 	qfield = 0;
5082 	error = 0;
5083 	/*
5084 	 * If deleting a real allocation, must free up the disk space.
5085 	 */
5086 	if (!delay) {
5087 		flags = XFS_ILOG_CORE;
5088 		/*
5089 		 * Realtime allocation.  Free it and record di_nblocks update.
5090 		 */
5091 		if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
5092 			xfs_fsblock_t	bno;
5093 			xfs_filblks_t	len;
5094 
5095 			ASSERT(do_mod(del->br_blockcount,
5096 				      mp->m_sb.sb_rextsize) == 0);
5097 			ASSERT(do_mod(del->br_startblock,
5098 				      mp->m_sb.sb_rextsize) == 0);
5099 			bno = del->br_startblock;
5100 			len = del->br_blockcount;
5101 			do_div(bno, mp->m_sb.sb_rextsize);
5102 			do_div(len, mp->m_sb.sb_rextsize);
5103 			error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
5104 			if (error)
5105 				goto done;
5106 			do_fx = 0;
5107 			nblks = len * mp->m_sb.sb_rextsize;
5108 			qfield = XFS_TRANS_DQ_RTBCOUNT;
5109 		}
5110 		/*
5111 		 * Ordinary allocation.
5112 		 */
5113 		else {
5114 			do_fx = 1;
5115 			nblks = del->br_blockcount;
5116 			qfield = XFS_TRANS_DQ_BCOUNT;
5117 		}
5118 		/*
5119 		 * Set up del_endblock and cur for later.
5120 		 */
5121 		del_endblock = del->br_startblock + del->br_blockcount;
5122 		if (cur) {
5123 			if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
5124 					got.br_startblock, got.br_blockcount,
5125 					&i)))
5126 				goto done;
5127 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5128 		}
5129 		da_old = da_new = 0;
5130 	} else {
5131 		da_old = startblockval(got.br_startblock);
5132 		da_new = 0;
5133 		nblks = 0;
5134 		do_fx = 0;
5135 	}
5136 
5137 	/*
5138 	 * Set flag value to use in switch statement.
5139 	 * Left-contig is 2, right-contig is 1.
5140 	 */
5141 	switch (((got.br_startoff == del->br_startoff) << 1) |
5142 		(got_endoff == del_endoff)) {
5143 	case 3:
5144 		/*
5145 		 * Matches the whole extent.  Delete the entry.
5146 		 */
5147 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5148 		xfs_iext_remove(ip, *idx, 1,
5149 				whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
5150 		--*idx;
5151 		if (delay)
5152 			break;
5153 
5154 		XFS_IFORK_NEXT_SET(ip, whichfork,
5155 			XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5156 		flags |= XFS_ILOG_CORE;
5157 		if (!cur) {
5158 			flags |= xfs_ilog_fext(whichfork);
5159 			break;
5160 		}
5161 		if ((error = xfs_btree_delete(cur, &i)))
5162 			goto done;
5163 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5164 		break;
5165 
5166 	case 2:
5167 		/*
5168 		 * Deleting the first part of the extent.
5169 		 */
5170 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5171 		xfs_bmbt_set_startoff(ep, del_endoff);
5172 		temp = got.br_blockcount - del->br_blockcount;
5173 		xfs_bmbt_set_blockcount(ep, temp);
5174 		if (delay) {
5175 			temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
5176 				da_old);
5177 			xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5178 			trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5179 			da_new = temp;
5180 			break;
5181 		}
5182 		xfs_bmbt_set_startblock(ep, del_endblock);
5183 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5184 		if (!cur) {
5185 			flags |= xfs_ilog_fext(whichfork);
5186 			break;
5187 		}
5188 		if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
5189 				got.br_blockcount - del->br_blockcount,
5190 				got.br_state)))
5191 			goto done;
5192 		break;
5193 
5194 	case 1:
5195 		/*
5196 		 * Deleting the last part of the extent.
5197 		 */
5198 		temp = got.br_blockcount - del->br_blockcount;
5199 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5200 		xfs_bmbt_set_blockcount(ep, temp);
5201 		if (delay) {
5202 			temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
5203 				da_old);
5204 			xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5205 			trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5206 			da_new = temp;
5207 			break;
5208 		}
5209 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5210 		if (!cur) {
5211 			flags |= xfs_ilog_fext(whichfork);
5212 			break;
5213 		}
5214 		if ((error = xfs_bmbt_update(cur, got.br_startoff,
5215 				got.br_startblock,
5216 				got.br_blockcount - del->br_blockcount,
5217 				got.br_state)))
5218 			goto done;
5219 		break;
5220 
5221 	case 0:
5222 		/*
5223 		 * Deleting the middle of the extent.
5224 		 */
5225 		temp = del->br_startoff - got.br_startoff;
5226 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5227 		xfs_bmbt_set_blockcount(ep, temp);
5228 		new.br_startoff = del_endoff;
5229 		temp2 = got_endoff - del_endoff;
5230 		new.br_blockcount = temp2;
5231 		new.br_state = got.br_state;
5232 		if (!delay) {
5233 			new.br_startblock = del_endblock;
5234 			flags |= XFS_ILOG_CORE;
5235 			if (cur) {
5236 				if ((error = xfs_bmbt_update(cur,
5237 						got.br_startoff,
5238 						got.br_startblock, temp,
5239 						got.br_state)))
5240 					goto done;
5241 				if ((error = xfs_btree_increment(cur, 0, &i)))
5242 					goto done;
5243 				cur->bc_rec.b = new;
5244 				error = xfs_btree_insert(cur, &i);
5245 				if (error && error != -ENOSPC)
5246 					goto done;
5247 				/*
5248 				 * If get no-space back from btree insert,
5249 				 * it tried a split, and we have a zero
5250 				 * block reservation.
5251 				 * Fix up our state and return the error.
5252 				 */
5253 				if (error == -ENOSPC) {
5254 					/*
5255 					 * Reset the cursor, don't trust
5256 					 * it after any insert operation.
5257 					 */
5258 					if ((error = xfs_bmbt_lookup_eq(cur,
5259 							got.br_startoff,
5260 							got.br_startblock,
5261 							temp, &i)))
5262 						goto done;
5263 					XFS_WANT_CORRUPTED_GOTO(mp,
5264 								i == 1, done);
5265 					/*
5266 					 * Update the btree record back
5267 					 * to the original value.
5268 					 */
5269 					if ((error = xfs_bmbt_update(cur,
5270 							got.br_startoff,
5271 							got.br_startblock,
5272 							got.br_blockcount,
5273 							got.br_state)))
5274 						goto done;
5275 					/*
5276 					 * Reset the extent record back
5277 					 * to the original value.
5278 					 */
5279 					xfs_bmbt_set_blockcount(ep,
5280 						got.br_blockcount);
5281 					flags = 0;
5282 					error = -ENOSPC;
5283 					goto done;
5284 				}
5285 				XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5286 			} else
5287 				flags |= xfs_ilog_fext(whichfork);
5288 			XFS_IFORK_NEXT_SET(ip, whichfork,
5289 				XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5290 		} else {
5291 			xfs_filblks_t	stolen;
5292 			ASSERT(whichfork == XFS_DATA_FORK);
5293 
5294 			/*
5295 			 * Distribute the original indlen reservation across the
5296 			 * two new extents. Steal blocks from the deleted extent
5297 			 * if necessary. Stealing blocks simply fudges the
5298 			 * fdblocks accounting in xfs_bunmapi().
5299 			 */
5300 			temp = xfs_bmap_worst_indlen(ip, got.br_blockcount);
5301 			temp2 = xfs_bmap_worst_indlen(ip, new.br_blockcount);
5302 			stolen = xfs_bmap_split_indlen(da_old, &temp, &temp2,
5303 						       del->br_blockcount);
5304 			da_new = temp + temp2 - stolen;
5305 			del->br_blockcount -= stolen;
5306 
5307 			/*
5308 			 * Set the reservation for each extent. Warn if either
5309 			 * is zero as this can lead to delalloc problems.
5310 			 */
5311 			WARN_ON_ONCE(!temp || !temp2);
5312 			xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5313 			new.br_startblock = nullstartblock((int)temp2);
5314 		}
5315 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5316 		xfs_iext_insert(ip, *idx + 1, 1, &new, state);
5317 		++*idx;
5318 		break;
5319 	}
5320 
5321 	/* remove reverse mapping */
5322 	if (!delay) {
5323 		error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, del);
5324 		if (error)
5325 			goto done;
5326 	}
5327 
5328 	/*
5329 	 * If we need to, add to list of extents to delete.
5330 	 */
5331 	if (do_fx && !(bflags & XFS_BMAPI_REMAP)) {
5332 		if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
5333 			error = xfs_refcount_decrease_extent(mp, dfops, del);
5334 			if (error)
5335 				goto done;
5336 		} else
5337 			xfs_bmap_add_free(mp, dfops, del->br_startblock,
5338 					del->br_blockcount, NULL);
5339 	}
5340 
5341 	/*
5342 	 * Adjust inode # blocks in the file.
5343 	 */
5344 	if (nblks)
5345 		ip->i_d.di_nblocks -= nblks;
5346 	/*
5347 	 * Adjust quota data.
5348 	 */
5349 	if (qfield && !(bflags & XFS_BMAPI_REMAP))
5350 		xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5351 
5352 	/*
5353 	 * Account for change in delayed indirect blocks.
5354 	 * Nothing to do for disk quota accounting here.
5355 	 */
5356 	ASSERT(da_old >= da_new);
5357 	if (da_old > da_new)
5358 		xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new), false);
5359 done:
5360 	*logflagsp = flags;
5361 	return error;
5362 }
5363 
5364 /*
5365  * Unmap (remove) blocks from a file.
5366  * If nexts is nonzero then the number of extents to remove is limited to
5367  * that value.  If not all extents in the block range can be removed then
5368  * *done is set.
5369  */
5370 int						/* error */
5371 __xfs_bunmapi(
5372 	xfs_trans_t		*tp,		/* transaction pointer */
5373 	struct xfs_inode	*ip,		/* incore inode */
5374 	xfs_fileoff_t		bno,		/* starting offset to unmap */
5375 	xfs_filblks_t		*rlen,		/* i/o: amount remaining */
5376 	int			flags,		/* misc flags */
5377 	xfs_extnum_t		nexts,		/* number of extents max */
5378 	xfs_fsblock_t		*firstblock,	/* first allocated block
5379 						   controls a.g. for allocs */
5380 	struct xfs_defer_ops	*dfops)		/* i/o: deferred updates */
5381 {
5382 	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
5383 	xfs_bmbt_irec_t		del;		/* extent being deleted */
5384 	int			error;		/* error return value */
5385 	xfs_extnum_t		extno;		/* extent number in list */
5386 	xfs_bmbt_irec_t		got;		/* current extent record */
5387 	xfs_ifork_t		*ifp;		/* inode fork pointer */
5388 	int			isrt;		/* freeing in rt area */
5389 	xfs_extnum_t		lastx;		/* last extent index used */
5390 	int			logflags;	/* transaction logging flags */
5391 	xfs_extlen_t		mod;		/* rt extent offset */
5392 	xfs_mount_t		*mp;		/* mount structure */
5393 	xfs_fileoff_t		start;		/* first file offset deleted */
5394 	int			tmp_logflags;	/* partial logging flags */
5395 	int			wasdel;		/* was a delayed alloc extent */
5396 	int			whichfork;	/* data or attribute fork */
5397 	xfs_fsblock_t		sum;
5398 	xfs_filblks_t		len = *rlen;	/* length to unmap in file */
5399 
5400 	trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5401 
5402 	whichfork = xfs_bmapi_whichfork(flags);
5403 	ASSERT(whichfork != XFS_COW_FORK);
5404 	ifp = XFS_IFORK_PTR(ip, whichfork);
5405 	if (unlikely(
5406 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5407 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5408 		XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5409 				 ip->i_mount);
5410 		return -EFSCORRUPTED;
5411 	}
5412 	mp = ip->i_mount;
5413 	if (XFS_FORCED_SHUTDOWN(mp))
5414 		return -EIO;
5415 
5416 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5417 	ASSERT(len > 0);
5418 	ASSERT(nexts >= 0);
5419 
5420 	if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5421 	    (error = xfs_iread_extents(tp, ip, whichfork)))
5422 		return error;
5423 	if (xfs_iext_count(ifp) == 0) {
5424 		*rlen = 0;
5425 		return 0;
5426 	}
5427 	XFS_STATS_INC(mp, xs_blk_unmap);
5428 	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5429 	start = bno;
5430 	bno = start + len - 1;
5431 
5432 	/*
5433 	 * Check to see if the given block number is past the end of the
5434 	 * file, back up to the last block if so...
5435 	 */
5436 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &lastx, &got)) {
5437 		ASSERT(lastx > 0);
5438 		xfs_iext_get_extent(ifp, --lastx, &got);
5439 		bno = got.br_startoff + got.br_blockcount - 1;
5440 	}
5441 
5442 	logflags = 0;
5443 	if (ifp->if_flags & XFS_IFBROOT) {
5444 		ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5445 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5446 		cur->bc_private.b.firstblock = *firstblock;
5447 		cur->bc_private.b.dfops = dfops;
5448 		cur->bc_private.b.flags = 0;
5449 	} else
5450 		cur = NULL;
5451 
5452 	if (isrt) {
5453 		/*
5454 		 * Synchronize by locking the bitmap inode.
5455 		 */
5456 		xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
5457 		xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5458 		xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
5459 		xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
5460 	}
5461 
5462 	extno = 0;
5463 	while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5464 	       (nexts == 0 || extno < nexts)) {
5465 		/*
5466 		 * Is the found extent after a hole in which bno lives?
5467 		 * Just back up to the previous extent, if so.
5468 		 */
5469 		if (got.br_startoff > bno) {
5470 			if (--lastx < 0)
5471 				break;
5472 			xfs_iext_get_extent(ifp, lastx, &got);
5473 		}
5474 		/*
5475 		 * Is the last block of this extent before the range
5476 		 * we're supposed to delete?  If so, we're done.
5477 		 */
5478 		bno = XFS_FILEOFF_MIN(bno,
5479 			got.br_startoff + got.br_blockcount - 1);
5480 		if (bno < start)
5481 			break;
5482 		/*
5483 		 * Then deal with the (possibly delayed) allocated space
5484 		 * we found.
5485 		 */
5486 		del = got;
5487 		wasdel = isnullstartblock(del.br_startblock);
5488 		if (got.br_startoff < start) {
5489 			del.br_startoff = start;
5490 			del.br_blockcount -= start - got.br_startoff;
5491 			if (!wasdel)
5492 				del.br_startblock += start - got.br_startoff;
5493 		}
5494 		if (del.br_startoff + del.br_blockcount > bno + 1)
5495 			del.br_blockcount = bno + 1 - del.br_startoff;
5496 		sum = del.br_startblock + del.br_blockcount;
5497 		if (isrt &&
5498 		    (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5499 			/*
5500 			 * Realtime extent not lined up at the end.
5501 			 * The extent could have been split into written
5502 			 * and unwritten pieces, or we could just be
5503 			 * unmapping part of it.  But we can't really
5504 			 * get rid of part of a realtime extent.
5505 			 */
5506 			if (del.br_state == XFS_EXT_UNWRITTEN ||
5507 			    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5508 				/*
5509 				 * This piece is unwritten, or we're not
5510 				 * using unwritten extents.  Skip over it.
5511 				 */
5512 				ASSERT(bno >= mod);
5513 				bno -= mod > del.br_blockcount ?
5514 					del.br_blockcount : mod;
5515 				if (bno < got.br_startoff) {
5516 					if (--lastx >= 0)
5517 						xfs_bmbt_get_all(xfs_iext_get_ext(
5518 							ifp, lastx), &got);
5519 				}
5520 				continue;
5521 			}
5522 			/*
5523 			 * It's written, turn it unwritten.
5524 			 * This is better than zeroing it.
5525 			 */
5526 			ASSERT(del.br_state == XFS_EXT_NORM);
5527 			ASSERT(tp->t_blk_res > 0);
5528 			/*
5529 			 * If this spans a realtime extent boundary,
5530 			 * chop it back to the start of the one we end at.
5531 			 */
5532 			if (del.br_blockcount > mod) {
5533 				del.br_startoff += del.br_blockcount - mod;
5534 				del.br_startblock += del.br_blockcount - mod;
5535 				del.br_blockcount = mod;
5536 			}
5537 			del.br_state = XFS_EXT_UNWRITTEN;
5538 			error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5539 					&lastx, &cur, &del, firstblock, dfops,
5540 					&logflags);
5541 			if (error)
5542 				goto error0;
5543 			goto nodelete;
5544 		}
5545 		if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5546 			/*
5547 			 * Realtime extent is lined up at the end but not
5548 			 * at the front.  We'll get rid of full extents if
5549 			 * we can.
5550 			 */
5551 			mod = mp->m_sb.sb_rextsize - mod;
5552 			if (del.br_blockcount > mod) {
5553 				del.br_blockcount -= mod;
5554 				del.br_startoff += mod;
5555 				del.br_startblock += mod;
5556 			} else if ((del.br_startoff == start &&
5557 				    (del.br_state == XFS_EXT_UNWRITTEN ||
5558 				     tp->t_blk_res == 0)) ||
5559 				   !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5560 				/*
5561 				 * Can't make it unwritten.  There isn't
5562 				 * a full extent here so just skip it.
5563 				 */
5564 				ASSERT(bno >= del.br_blockcount);
5565 				bno -= del.br_blockcount;
5566 				if (got.br_startoff > bno && --lastx >= 0)
5567 					xfs_iext_get_extent(ifp, lastx, &got);
5568 				continue;
5569 			} else if (del.br_state == XFS_EXT_UNWRITTEN) {
5570 				struct xfs_bmbt_irec	prev;
5571 
5572 				/*
5573 				 * This one is already unwritten.
5574 				 * It must have a written left neighbor.
5575 				 * Unwrite the killed part of that one and
5576 				 * try again.
5577 				 */
5578 				ASSERT(lastx > 0);
5579 				xfs_iext_get_extent(ifp, lastx - 1, &prev);
5580 				ASSERT(prev.br_state == XFS_EXT_NORM);
5581 				ASSERT(!isnullstartblock(prev.br_startblock));
5582 				ASSERT(del.br_startblock ==
5583 				       prev.br_startblock + prev.br_blockcount);
5584 				if (prev.br_startoff < start) {
5585 					mod = start - prev.br_startoff;
5586 					prev.br_blockcount -= mod;
5587 					prev.br_startblock += mod;
5588 					prev.br_startoff = start;
5589 				}
5590 				prev.br_state = XFS_EXT_UNWRITTEN;
5591 				lastx--;
5592 				error = xfs_bmap_add_extent_unwritten_real(tp,
5593 						ip, &lastx, &cur, &prev,
5594 						firstblock, dfops, &logflags);
5595 				if (error)
5596 					goto error0;
5597 				goto nodelete;
5598 			} else {
5599 				ASSERT(del.br_state == XFS_EXT_NORM);
5600 				del.br_state = XFS_EXT_UNWRITTEN;
5601 				error = xfs_bmap_add_extent_unwritten_real(tp,
5602 						ip, &lastx, &cur, &del,
5603 						firstblock, dfops, &logflags);
5604 				if (error)
5605 					goto error0;
5606 				goto nodelete;
5607 			}
5608 		}
5609 
5610 		/*
5611 		 * If it's the case where the directory code is running
5612 		 * with no block reservation, and the deleted block is in
5613 		 * the middle of its extent, and the resulting insert
5614 		 * of an extent would cause transformation to btree format,
5615 		 * then reject it.  The calling code will then swap
5616 		 * blocks around instead.
5617 		 * We have to do this now, rather than waiting for the
5618 		 * conversion to btree format, since the transaction
5619 		 * will be dirty.
5620 		 */
5621 		if (!wasdel && tp->t_blk_res == 0 &&
5622 		    XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5623 		    XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5624 			XFS_IFORK_MAXEXT(ip, whichfork) &&
5625 		    del.br_startoff > got.br_startoff &&
5626 		    del.br_startoff + del.br_blockcount <
5627 		    got.br_startoff + got.br_blockcount) {
5628 			error = -ENOSPC;
5629 			goto error0;
5630 		}
5631 
5632 		/*
5633 		 * Unreserve quota and update realtime free space, if
5634 		 * appropriate. If delayed allocation, update the inode delalloc
5635 		 * counter now and wait to update the sb counters as
5636 		 * xfs_bmap_del_extent() might need to borrow some blocks.
5637 		 */
5638 		if (wasdel) {
5639 			ASSERT(startblockval(del.br_startblock) > 0);
5640 			if (isrt) {
5641 				xfs_filblks_t rtexts;
5642 
5643 				rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5644 				do_div(rtexts, mp->m_sb.sb_rextsize);
5645 				xfs_mod_frextents(mp, (int64_t)rtexts);
5646 				(void)xfs_trans_reserve_quota_nblks(NULL,
5647 					ip, -((long)del.br_blockcount), 0,
5648 					XFS_QMOPT_RES_RTBLKS);
5649 			} else {
5650 				(void)xfs_trans_reserve_quota_nblks(NULL,
5651 					ip, -((long)del.br_blockcount), 0,
5652 					XFS_QMOPT_RES_REGBLKS);
5653 			}
5654 			ip->i_delayed_blks -= del.br_blockcount;
5655 			if (cur)
5656 				cur->bc_private.b.flags |=
5657 					XFS_BTCUR_BPRV_WASDEL;
5658 		} else if (cur)
5659 			cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5660 
5661 		error = xfs_bmap_del_extent(ip, tp, &lastx, dfops, cur, &del,
5662 				&tmp_logflags, whichfork, flags);
5663 		logflags |= tmp_logflags;
5664 		if (error)
5665 			goto error0;
5666 
5667 		if (!isrt && wasdel)
5668 			xfs_mod_fdblocks(mp, (int64_t)del.br_blockcount, false);
5669 
5670 		bno = del.br_startoff - 1;
5671 nodelete:
5672 		/*
5673 		 * If not done go on to the next (previous) record.
5674 		 */
5675 		if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5676 			if (lastx >= 0) {
5677 				xfs_iext_get_extent(ifp, lastx, &got);
5678 				if (got.br_startoff > bno && --lastx >= 0)
5679 					xfs_iext_get_extent(ifp, lastx, &got);
5680 			}
5681 			extno++;
5682 		}
5683 	}
5684 	if (bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0)
5685 		*rlen = 0;
5686 	else
5687 		*rlen = bno - start + 1;
5688 
5689 	/*
5690 	 * Convert to a btree if necessary.
5691 	 */
5692 	if (xfs_bmap_needs_btree(ip, whichfork)) {
5693 		ASSERT(cur == NULL);
5694 		error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops,
5695 			&cur, 0, &tmp_logflags, whichfork);
5696 		logflags |= tmp_logflags;
5697 		if (error)
5698 			goto error0;
5699 	}
5700 	/*
5701 	 * transform from btree to extents, give it cur
5702 	 */
5703 	else if (xfs_bmap_wants_extents(ip, whichfork)) {
5704 		ASSERT(cur != NULL);
5705 		error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5706 			whichfork);
5707 		logflags |= tmp_logflags;
5708 		if (error)
5709 			goto error0;
5710 	}
5711 	/*
5712 	 * transform from extents to local?
5713 	 */
5714 	error = 0;
5715 error0:
5716 	/*
5717 	 * Log everything.  Do this after conversion, there's no point in
5718 	 * logging the extent records if we've converted to btree format.
5719 	 */
5720 	if ((logflags & xfs_ilog_fext(whichfork)) &&
5721 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5722 		logflags &= ~xfs_ilog_fext(whichfork);
5723 	else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5724 		 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5725 		logflags &= ~xfs_ilog_fbroot(whichfork);
5726 	/*
5727 	 * Log inode even in the error case, if the transaction
5728 	 * is dirty we'll need to shut down the filesystem.
5729 	 */
5730 	if (logflags)
5731 		xfs_trans_log_inode(tp, ip, logflags);
5732 	if (cur) {
5733 		if (!error) {
5734 			*firstblock = cur->bc_private.b.firstblock;
5735 			cur->bc_private.b.allocated = 0;
5736 		}
5737 		xfs_btree_del_cursor(cur,
5738 			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5739 	}
5740 	return error;
5741 }
5742 
5743 /* Unmap a range of a file. */
5744 int
5745 xfs_bunmapi(
5746 	xfs_trans_t		*tp,
5747 	struct xfs_inode	*ip,
5748 	xfs_fileoff_t		bno,
5749 	xfs_filblks_t		len,
5750 	int			flags,
5751 	xfs_extnum_t		nexts,
5752 	xfs_fsblock_t		*firstblock,
5753 	struct xfs_defer_ops	*dfops,
5754 	int			*done)
5755 {
5756 	int			error;
5757 
5758 	error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts, firstblock,
5759 			dfops);
5760 	*done = (len == 0);
5761 	return error;
5762 }
5763 
5764 /*
5765  * Determine whether an extent shift can be accomplished by a merge with the
5766  * extent that precedes the target hole of the shift.
5767  */
5768 STATIC bool
5769 xfs_bmse_can_merge(
5770 	struct xfs_bmbt_irec	*left,	/* preceding extent */
5771 	struct xfs_bmbt_irec	*got,	/* current extent to shift */
5772 	xfs_fileoff_t		shift)	/* shift fsb */
5773 {
5774 	xfs_fileoff_t		startoff;
5775 
5776 	startoff = got->br_startoff - shift;
5777 
5778 	/*
5779 	 * The extent, once shifted, must be adjacent in-file and on-disk with
5780 	 * the preceding extent.
5781 	 */
5782 	if ((left->br_startoff + left->br_blockcount != startoff) ||
5783 	    (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5784 	    (left->br_state != got->br_state) ||
5785 	    (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5786 		return false;
5787 
5788 	return true;
5789 }
5790 
5791 /*
5792  * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5793  * hole in the file. If an extent shift would result in the extent being fully
5794  * adjacent to the extent that currently precedes the hole, we can merge with
5795  * the preceding extent rather than do the shift.
5796  *
5797  * This function assumes the caller has verified a shift-by-merge is possible
5798  * with the provided extents via xfs_bmse_can_merge().
5799  */
5800 STATIC int
5801 xfs_bmse_merge(
5802 	struct xfs_inode		*ip,
5803 	int				whichfork,
5804 	xfs_fileoff_t			shift,		/* shift fsb */
5805 	int				current_ext,	/* idx of gotp */
5806 	struct xfs_bmbt_rec_host	*gotp,		/* extent to shift */
5807 	struct xfs_bmbt_rec_host	*leftp,		/* preceding extent */
5808 	struct xfs_btree_cur		*cur,
5809 	int				*logflags)	/* output */
5810 {
5811 	struct xfs_bmbt_irec		got;
5812 	struct xfs_bmbt_irec		left;
5813 	xfs_filblks_t			blockcount;
5814 	int				error, i;
5815 	struct xfs_mount		*mp = ip->i_mount;
5816 
5817 	xfs_bmbt_get_all(gotp, &got);
5818 	xfs_bmbt_get_all(leftp, &left);
5819 	blockcount = left.br_blockcount + got.br_blockcount;
5820 
5821 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5822 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5823 	ASSERT(xfs_bmse_can_merge(&left, &got, shift));
5824 
5825 	/*
5826 	 * Merge the in-core extents. Note that the host record pointers and
5827 	 * current_ext index are invalid once the extent has been removed via
5828 	 * xfs_iext_remove().
5829 	 */
5830 	xfs_bmbt_set_blockcount(leftp, blockcount);
5831 	xfs_iext_remove(ip, current_ext, 1, 0);
5832 
5833 	/*
5834 	 * Update the on-disk extent count, the btree if necessary and log the
5835 	 * inode.
5836 	 */
5837 	XFS_IFORK_NEXT_SET(ip, whichfork,
5838 			   XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5839 	*logflags |= XFS_ILOG_CORE;
5840 	if (!cur) {
5841 		*logflags |= XFS_ILOG_DEXT;
5842 		return 0;
5843 	}
5844 
5845 	/* lookup and remove the extent to merge */
5846 	error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
5847 				   got.br_blockcount, &i);
5848 	if (error)
5849 		return error;
5850 	XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5851 
5852 	error = xfs_btree_delete(cur, &i);
5853 	if (error)
5854 		return error;
5855 	XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5856 
5857 	/* lookup and update size of the previous extent */
5858 	error = xfs_bmbt_lookup_eq(cur, left.br_startoff, left.br_startblock,
5859 				   left.br_blockcount, &i);
5860 	if (error)
5861 		return error;
5862 	XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5863 
5864 	left.br_blockcount = blockcount;
5865 
5866 	return xfs_bmbt_update(cur, left.br_startoff, left.br_startblock,
5867 			       left.br_blockcount, left.br_state);
5868 }
5869 
5870 /*
5871  * Shift a single extent.
5872  */
5873 STATIC int
5874 xfs_bmse_shift_one(
5875 	struct xfs_inode		*ip,
5876 	int				whichfork,
5877 	xfs_fileoff_t			offset_shift_fsb,
5878 	int				*current_ext,
5879 	struct xfs_bmbt_rec_host	*gotp,
5880 	struct xfs_btree_cur		*cur,
5881 	int				*logflags,
5882 	enum shift_direction		direction,
5883 	struct xfs_defer_ops		*dfops)
5884 {
5885 	struct xfs_ifork		*ifp;
5886 	struct xfs_mount		*mp;
5887 	xfs_fileoff_t			startoff;
5888 	struct xfs_bmbt_rec_host	*adj_irecp;
5889 	struct xfs_bmbt_irec		got;
5890 	struct xfs_bmbt_irec		adj_irec;
5891 	int				error;
5892 	int				i;
5893 	int				total_extents;
5894 
5895 	mp = ip->i_mount;
5896 	ifp = XFS_IFORK_PTR(ip, whichfork);
5897 	total_extents = xfs_iext_count(ifp);
5898 
5899 	xfs_bmbt_get_all(gotp, &got);
5900 
5901 	/* delalloc extents should be prevented by caller */
5902 	XFS_WANT_CORRUPTED_RETURN(mp, !isnullstartblock(got.br_startblock));
5903 
5904 	if (direction == SHIFT_LEFT) {
5905 		startoff = got.br_startoff - offset_shift_fsb;
5906 
5907 		/*
5908 		 * Check for merge if we've got an extent to the left,
5909 		 * otherwise make sure there's enough room at the start
5910 		 * of the file for the shift.
5911 		 */
5912 		if (!*current_ext) {
5913 			if (got.br_startoff < offset_shift_fsb)
5914 				return -EINVAL;
5915 			goto update_current_ext;
5916 		}
5917 		/*
5918 		 * grab the left extent and check for a large
5919 		 * enough hole.
5920 		 */
5921 		adj_irecp = xfs_iext_get_ext(ifp, *current_ext - 1);
5922 		xfs_bmbt_get_all(adj_irecp, &adj_irec);
5923 
5924 		if (startoff <
5925 		    adj_irec.br_startoff + adj_irec.br_blockcount)
5926 			return -EINVAL;
5927 
5928 		/* check whether to merge the extent or shift it down */
5929 		if (xfs_bmse_can_merge(&adj_irec, &got,
5930 				       offset_shift_fsb)) {
5931 			error = xfs_bmse_merge(ip, whichfork, offset_shift_fsb,
5932 					       *current_ext, gotp, adj_irecp,
5933 					       cur, logflags);
5934 			if (error)
5935 				return error;
5936 			adj_irec = got;
5937 			goto update_rmap;
5938 		}
5939 	} else {
5940 		startoff = got.br_startoff + offset_shift_fsb;
5941 		/* nothing to move if this is the last extent */
5942 		if (*current_ext >= (total_extents - 1))
5943 			goto update_current_ext;
5944 		/*
5945 		 * If this is not the last extent in the file, make sure there
5946 		 * is enough room between current extent and next extent for
5947 		 * accommodating the shift.
5948 		 */
5949 		adj_irecp = xfs_iext_get_ext(ifp, *current_ext + 1);
5950 		xfs_bmbt_get_all(adj_irecp, &adj_irec);
5951 		if (startoff + got.br_blockcount > adj_irec.br_startoff)
5952 			return -EINVAL;
5953 		/*
5954 		 * Unlike a left shift (which involves a hole punch),
5955 		 * a right shift does not modify extent neighbors
5956 		 * in any way. We should never find mergeable extents
5957 		 * in this scenario. Check anyways and warn if we
5958 		 * encounter two extents that could be one.
5959 		 */
5960 		if (xfs_bmse_can_merge(&got, &adj_irec, offset_shift_fsb))
5961 			WARN_ON_ONCE(1);
5962 	}
5963 	/*
5964 	 * Increment the extent index for the next iteration, update the start
5965 	 * offset of the in-core extent and update the btree if applicable.
5966 	 */
5967 update_current_ext:
5968 	if (direction == SHIFT_LEFT)
5969 		(*current_ext)++;
5970 	else
5971 		(*current_ext)--;
5972 	xfs_bmbt_set_startoff(gotp, startoff);
5973 	*logflags |= XFS_ILOG_CORE;
5974 	adj_irec = got;
5975 	if (!cur) {
5976 		*logflags |= XFS_ILOG_DEXT;
5977 		goto update_rmap;
5978 	}
5979 
5980 	error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
5981 				   got.br_blockcount, &i);
5982 	if (error)
5983 		return error;
5984 	XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5985 
5986 	got.br_startoff = startoff;
5987 	error = xfs_bmbt_update(cur, got.br_startoff, got.br_startblock,
5988 			got.br_blockcount, got.br_state);
5989 	if (error)
5990 		return error;
5991 
5992 update_rmap:
5993 	/* update reverse mapping */
5994 	error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, &adj_irec);
5995 	if (error)
5996 		return error;
5997 	adj_irec.br_startoff = startoff;
5998 	return xfs_rmap_map_extent(mp, dfops, ip, whichfork, &adj_irec);
5999 }
6000 
6001 /*
6002  * Shift extent records to the left/right to cover/create a hole.
6003  *
6004  * The maximum number of extents to be shifted in a single operation is
6005  * @num_exts. @stop_fsb specifies the file offset at which to stop shift and the
6006  * file offset where we've left off is returned in @next_fsb. @offset_shift_fsb
6007  * is the length by which each extent is shifted. If there is no hole to shift
6008  * the extents into, this will be considered invalid operation and we abort
6009  * immediately.
6010  */
6011 int
6012 xfs_bmap_shift_extents(
6013 	struct xfs_trans	*tp,
6014 	struct xfs_inode	*ip,
6015 	xfs_fileoff_t		*next_fsb,
6016 	xfs_fileoff_t		offset_shift_fsb,
6017 	int			*done,
6018 	xfs_fileoff_t		stop_fsb,
6019 	xfs_fsblock_t		*firstblock,
6020 	struct xfs_defer_ops	*dfops,
6021 	enum shift_direction	direction,
6022 	int			num_exts)
6023 {
6024 	struct xfs_btree_cur		*cur = NULL;
6025 	struct xfs_bmbt_rec_host	*gotp;
6026 	struct xfs_bmbt_irec            got;
6027 	struct xfs_mount		*mp = ip->i_mount;
6028 	struct xfs_ifork		*ifp;
6029 	xfs_extnum_t			nexts = 0;
6030 	xfs_extnum_t			current_ext;
6031 	xfs_extnum_t			total_extents;
6032 	xfs_extnum_t			stop_extent;
6033 	int				error = 0;
6034 	int				whichfork = XFS_DATA_FORK;
6035 	int				logflags = 0;
6036 
6037 	if (unlikely(XFS_TEST_ERROR(
6038 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
6039 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
6040 	     mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
6041 		XFS_ERROR_REPORT("xfs_bmap_shift_extents",
6042 				 XFS_ERRLEVEL_LOW, mp);
6043 		return -EFSCORRUPTED;
6044 	}
6045 
6046 	if (XFS_FORCED_SHUTDOWN(mp))
6047 		return -EIO;
6048 
6049 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
6050 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6051 	ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT);
6052 	ASSERT(*next_fsb != NULLFSBLOCK || direction == SHIFT_RIGHT);
6053 
6054 	ifp = XFS_IFORK_PTR(ip, whichfork);
6055 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
6056 		/* Read in all the extents */
6057 		error = xfs_iread_extents(tp, ip, whichfork);
6058 		if (error)
6059 			return error;
6060 	}
6061 
6062 	if (ifp->if_flags & XFS_IFBROOT) {
6063 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6064 		cur->bc_private.b.firstblock = *firstblock;
6065 		cur->bc_private.b.dfops = dfops;
6066 		cur->bc_private.b.flags = 0;
6067 	}
6068 
6069 	/*
6070 	 * There may be delalloc extents in the data fork before the range we
6071 	 * are collapsing out, so we cannot use the count of real extents here.
6072 	 * Instead we have to calculate it from the incore fork.
6073 	 */
6074 	total_extents = xfs_iext_count(ifp);
6075 	if (total_extents == 0) {
6076 		*done = 1;
6077 		goto del_cursor;
6078 	}
6079 
6080 	/*
6081 	 * In case of first right shift, we need to initialize next_fsb
6082 	 */
6083 	if (*next_fsb == NULLFSBLOCK) {
6084 		gotp = xfs_iext_get_ext(ifp, total_extents - 1);
6085 		xfs_bmbt_get_all(gotp, &got);
6086 		*next_fsb = got.br_startoff;
6087 		if (stop_fsb > *next_fsb) {
6088 			*done = 1;
6089 			goto del_cursor;
6090 		}
6091 	}
6092 
6093 	/* Lookup the extent index at which we have to stop */
6094 	if (direction == SHIFT_RIGHT) {
6095 		gotp = xfs_iext_bno_to_ext(ifp, stop_fsb, &stop_extent);
6096 		/* Make stop_extent exclusive of shift range */
6097 		stop_extent--;
6098 	} else
6099 		stop_extent = total_extents;
6100 
6101 	/*
6102 	 * Look up the extent index for the fsb where we start shifting. We can
6103 	 * henceforth iterate with current_ext as extent list changes are locked
6104 	 * out via ilock.
6105 	 *
6106 	 * gotp can be null in 2 cases: 1) if there are no extents or 2)
6107 	 * *next_fsb lies in a hole beyond which there are no extents. Either
6108 	 * way, we are done.
6109 	 */
6110 	gotp = xfs_iext_bno_to_ext(ifp, *next_fsb, &current_ext);
6111 	if (!gotp) {
6112 		*done = 1;
6113 		goto del_cursor;
6114 	}
6115 
6116 	/* some sanity checking before we finally start shifting extents */
6117 	if ((direction == SHIFT_LEFT && current_ext >= stop_extent) ||
6118 	     (direction == SHIFT_RIGHT && current_ext <= stop_extent)) {
6119 		error = -EIO;
6120 		goto del_cursor;
6121 	}
6122 
6123 	while (nexts++ < num_exts) {
6124 		error = xfs_bmse_shift_one(ip, whichfork, offset_shift_fsb,
6125 					   &current_ext, gotp, cur, &logflags,
6126 					   direction, dfops);
6127 		if (error)
6128 			goto del_cursor;
6129 		/*
6130 		 * If there was an extent merge during the shift, the extent
6131 		 * count can change. Update the total and grade the next record.
6132 		 */
6133 		if (direction == SHIFT_LEFT) {
6134 			total_extents = xfs_iext_count(ifp);
6135 			stop_extent = total_extents;
6136 		}
6137 
6138 		if (current_ext == stop_extent) {
6139 			*done = 1;
6140 			*next_fsb = NULLFSBLOCK;
6141 			break;
6142 		}
6143 		gotp = xfs_iext_get_ext(ifp, current_ext);
6144 	}
6145 
6146 	if (!*done) {
6147 		xfs_bmbt_get_all(gotp, &got);
6148 		*next_fsb = got.br_startoff;
6149 	}
6150 
6151 del_cursor:
6152 	if (cur)
6153 		xfs_btree_del_cursor(cur,
6154 			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
6155 
6156 	if (logflags)
6157 		xfs_trans_log_inode(tp, ip, logflags);
6158 
6159 	return error;
6160 }
6161 
6162 /*
6163  * Splits an extent into two extents at split_fsb block such that it is
6164  * the first block of the current_ext. @current_ext is a target extent
6165  * to be split. @split_fsb is a block where the extents is split.
6166  * If split_fsb lies in a hole or the first block of extents, just return 0.
6167  */
6168 STATIC int
6169 xfs_bmap_split_extent_at(
6170 	struct xfs_trans	*tp,
6171 	struct xfs_inode	*ip,
6172 	xfs_fileoff_t		split_fsb,
6173 	xfs_fsblock_t		*firstfsb,
6174 	struct xfs_defer_ops	*dfops)
6175 {
6176 	int				whichfork = XFS_DATA_FORK;
6177 	struct xfs_btree_cur		*cur = NULL;
6178 	struct xfs_bmbt_rec_host	*gotp;
6179 	struct xfs_bmbt_irec		got;
6180 	struct xfs_bmbt_irec		new; /* split extent */
6181 	struct xfs_mount		*mp = ip->i_mount;
6182 	struct xfs_ifork		*ifp;
6183 	xfs_fsblock_t			gotblkcnt; /* new block count for got */
6184 	xfs_extnum_t			current_ext;
6185 	int				error = 0;
6186 	int				logflags = 0;
6187 	int				i = 0;
6188 
6189 	if (unlikely(XFS_TEST_ERROR(
6190 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
6191 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
6192 	     mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
6193 		XFS_ERROR_REPORT("xfs_bmap_split_extent_at",
6194 				 XFS_ERRLEVEL_LOW, mp);
6195 		return -EFSCORRUPTED;
6196 	}
6197 
6198 	if (XFS_FORCED_SHUTDOWN(mp))
6199 		return -EIO;
6200 
6201 	ifp = XFS_IFORK_PTR(ip, whichfork);
6202 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
6203 		/* Read in all the extents */
6204 		error = xfs_iread_extents(tp, ip, whichfork);
6205 		if (error)
6206 			return error;
6207 	}
6208 
6209 	/*
6210 	 * gotp can be null in 2 cases: 1) if there are no extents
6211 	 * or 2) split_fsb lies in a hole beyond which there are
6212 	 * no extents. Either way, we are done.
6213 	 */
6214 	gotp = xfs_iext_bno_to_ext(ifp, split_fsb, &current_ext);
6215 	if (!gotp)
6216 		return 0;
6217 
6218 	xfs_bmbt_get_all(gotp, &got);
6219 
6220 	/*
6221 	 * Check split_fsb lies in a hole or the start boundary offset
6222 	 * of the extent.
6223 	 */
6224 	if (got.br_startoff >= split_fsb)
6225 		return 0;
6226 
6227 	gotblkcnt = split_fsb - got.br_startoff;
6228 	new.br_startoff = split_fsb;
6229 	new.br_startblock = got.br_startblock + gotblkcnt;
6230 	new.br_blockcount = got.br_blockcount - gotblkcnt;
6231 	new.br_state = got.br_state;
6232 
6233 	if (ifp->if_flags & XFS_IFBROOT) {
6234 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6235 		cur->bc_private.b.firstblock = *firstfsb;
6236 		cur->bc_private.b.dfops = dfops;
6237 		cur->bc_private.b.flags = 0;
6238 		error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
6239 				got.br_startblock,
6240 				got.br_blockcount,
6241 				&i);
6242 		if (error)
6243 			goto del_cursor;
6244 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
6245 	}
6246 
6247 	xfs_bmbt_set_blockcount(gotp, gotblkcnt);
6248 	got.br_blockcount = gotblkcnt;
6249 
6250 	logflags = XFS_ILOG_CORE;
6251 	if (cur) {
6252 		error = xfs_bmbt_update(cur, got.br_startoff,
6253 				got.br_startblock,
6254 				got.br_blockcount,
6255 				got.br_state);
6256 		if (error)
6257 			goto del_cursor;
6258 	} else
6259 		logflags |= XFS_ILOG_DEXT;
6260 
6261 	/* Add new extent */
6262 	current_ext++;
6263 	xfs_iext_insert(ip, current_ext, 1, &new, 0);
6264 	XFS_IFORK_NEXT_SET(ip, whichfork,
6265 			   XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
6266 
6267 	if (cur) {
6268 		error = xfs_bmbt_lookup_eq(cur, new.br_startoff,
6269 				new.br_startblock, new.br_blockcount,
6270 				&i);
6271 		if (error)
6272 			goto del_cursor;
6273 		XFS_WANT_CORRUPTED_GOTO(mp, i == 0, del_cursor);
6274 		cur->bc_rec.b.br_state = new.br_state;
6275 
6276 		error = xfs_btree_insert(cur, &i);
6277 		if (error)
6278 			goto del_cursor;
6279 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
6280 	}
6281 
6282 	/*
6283 	 * Convert to a btree if necessary.
6284 	 */
6285 	if (xfs_bmap_needs_btree(ip, whichfork)) {
6286 		int tmp_logflags; /* partial log flag return val */
6287 
6288 		ASSERT(cur == NULL);
6289 		error = xfs_bmap_extents_to_btree(tp, ip, firstfsb, dfops,
6290 				&cur, 0, &tmp_logflags, whichfork);
6291 		logflags |= tmp_logflags;
6292 	}
6293 
6294 del_cursor:
6295 	if (cur) {
6296 		cur->bc_private.b.allocated = 0;
6297 		xfs_btree_del_cursor(cur,
6298 				error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
6299 	}
6300 
6301 	if (logflags)
6302 		xfs_trans_log_inode(tp, ip, logflags);
6303 	return error;
6304 }
6305 
6306 int
6307 xfs_bmap_split_extent(
6308 	struct xfs_inode        *ip,
6309 	xfs_fileoff_t           split_fsb)
6310 {
6311 	struct xfs_mount        *mp = ip->i_mount;
6312 	struct xfs_trans        *tp;
6313 	struct xfs_defer_ops    dfops;
6314 	xfs_fsblock_t           firstfsb;
6315 	int                     error;
6316 
6317 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
6318 			XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
6319 	if (error)
6320 		return error;
6321 
6322 	xfs_ilock(ip, XFS_ILOCK_EXCL);
6323 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
6324 
6325 	xfs_defer_init(&dfops, &firstfsb);
6326 
6327 	error = xfs_bmap_split_extent_at(tp, ip, split_fsb,
6328 			&firstfsb, &dfops);
6329 	if (error)
6330 		goto out;
6331 
6332 	error = xfs_defer_finish(&tp, &dfops, NULL);
6333 	if (error)
6334 		goto out;
6335 
6336 	return xfs_trans_commit(tp);
6337 
6338 out:
6339 	xfs_defer_cancel(&dfops);
6340 	xfs_trans_cancel(tp);
6341 	return error;
6342 }
6343 
6344 /* Deferred mapping is only for real extents in the data fork. */
6345 static bool
6346 xfs_bmap_is_update_needed(
6347 	struct xfs_bmbt_irec	*bmap)
6348 {
6349 	return  bmap->br_startblock != HOLESTARTBLOCK &&
6350 		bmap->br_startblock != DELAYSTARTBLOCK;
6351 }
6352 
6353 /* Record a bmap intent. */
6354 static int
6355 __xfs_bmap_add(
6356 	struct xfs_mount		*mp,
6357 	struct xfs_defer_ops		*dfops,
6358 	enum xfs_bmap_intent_type	type,
6359 	struct xfs_inode		*ip,
6360 	int				whichfork,
6361 	struct xfs_bmbt_irec		*bmap)
6362 {
6363 	int				error;
6364 	struct xfs_bmap_intent		*bi;
6365 
6366 	trace_xfs_bmap_defer(mp,
6367 			XFS_FSB_TO_AGNO(mp, bmap->br_startblock),
6368 			type,
6369 			XFS_FSB_TO_AGBNO(mp, bmap->br_startblock),
6370 			ip->i_ino, whichfork,
6371 			bmap->br_startoff,
6372 			bmap->br_blockcount,
6373 			bmap->br_state);
6374 
6375 	bi = kmem_alloc(sizeof(struct xfs_bmap_intent), KM_SLEEP | KM_NOFS);
6376 	INIT_LIST_HEAD(&bi->bi_list);
6377 	bi->bi_type = type;
6378 	bi->bi_owner = ip;
6379 	bi->bi_whichfork = whichfork;
6380 	bi->bi_bmap = *bmap;
6381 
6382 	error = xfs_defer_join(dfops, bi->bi_owner);
6383 	if (error) {
6384 		kmem_free(bi);
6385 		return error;
6386 	}
6387 
6388 	xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list);
6389 	return 0;
6390 }
6391 
6392 /* Map an extent into a file. */
6393 int
6394 xfs_bmap_map_extent(
6395 	struct xfs_mount	*mp,
6396 	struct xfs_defer_ops	*dfops,
6397 	struct xfs_inode	*ip,
6398 	struct xfs_bmbt_irec	*PREV)
6399 {
6400 	if (!xfs_bmap_is_update_needed(PREV))
6401 		return 0;
6402 
6403 	return __xfs_bmap_add(mp, dfops, XFS_BMAP_MAP, ip,
6404 			XFS_DATA_FORK, PREV);
6405 }
6406 
6407 /* Unmap an extent out of a file. */
6408 int
6409 xfs_bmap_unmap_extent(
6410 	struct xfs_mount	*mp,
6411 	struct xfs_defer_ops	*dfops,
6412 	struct xfs_inode	*ip,
6413 	struct xfs_bmbt_irec	*PREV)
6414 {
6415 	if (!xfs_bmap_is_update_needed(PREV))
6416 		return 0;
6417 
6418 	return __xfs_bmap_add(mp, dfops, XFS_BMAP_UNMAP, ip,
6419 			XFS_DATA_FORK, PREV);
6420 }
6421 
6422 /*
6423  * Process one of the deferred bmap operations.  We pass back the
6424  * btree cursor to maintain our lock on the bmapbt between calls.
6425  */
6426 int
6427 xfs_bmap_finish_one(
6428 	struct xfs_trans		*tp,
6429 	struct xfs_defer_ops		*dfops,
6430 	struct xfs_inode		*ip,
6431 	enum xfs_bmap_intent_type	type,
6432 	int				whichfork,
6433 	xfs_fileoff_t			startoff,
6434 	xfs_fsblock_t			startblock,
6435 	xfs_filblks_t			blockcount,
6436 	xfs_exntst_t			state)
6437 {
6438 	struct xfs_bmbt_irec		bmap;
6439 	int				nimaps = 1;
6440 	xfs_fsblock_t			firstfsb;
6441 	int				flags = XFS_BMAPI_REMAP;
6442 	int				done;
6443 	int				error = 0;
6444 
6445 	bmap.br_startblock = startblock;
6446 	bmap.br_startoff = startoff;
6447 	bmap.br_blockcount = blockcount;
6448 	bmap.br_state = state;
6449 
6450 	trace_xfs_bmap_deferred(tp->t_mountp,
6451 			XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type,
6452 			XFS_FSB_TO_AGBNO(tp->t_mountp, startblock),
6453 			ip->i_ino, whichfork, startoff, blockcount, state);
6454 
6455 	if (whichfork != XFS_DATA_FORK && whichfork != XFS_ATTR_FORK)
6456 		return -EFSCORRUPTED;
6457 	if (whichfork == XFS_ATTR_FORK)
6458 		flags |= XFS_BMAPI_ATTRFORK;
6459 
6460 	if (XFS_TEST_ERROR(false, tp->t_mountp,
6461 			XFS_ERRTAG_BMAP_FINISH_ONE,
6462 			XFS_RANDOM_BMAP_FINISH_ONE))
6463 		return -EIO;
6464 
6465 	switch (type) {
6466 	case XFS_BMAP_MAP:
6467 		firstfsb = bmap.br_startblock;
6468 		error = xfs_bmapi_write(tp, ip, bmap.br_startoff,
6469 					bmap.br_blockcount, flags, &firstfsb,
6470 					bmap.br_blockcount, &bmap, &nimaps,
6471 					dfops);
6472 		break;
6473 	case XFS_BMAP_UNMAP:
6474 		error = xfs_bunmapi(tp, ip, bmap.br_startoff,
6475 				bmap.br_blockcount, flags, 1, &firstfsb,
6476 				dfops, &done);
6477 		ASSERT(done);
6478 		break;
6479 	default:
6480 		ASSERT(0);
6481 		error = -EFSCORRUPTED;
6482 	}
6483 
6484 	return error;
6485 }
6486