xref: /openbmc/linux/fs/xfs/libxfs/xfs_bmap.c (revision 11788d9b)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4  * All Rights Reserved.
5  */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_bit.h"
13 #include "xfs_sb.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_dir2.h"
17 #include "xfs_inode.h"
18 #include "xfs_btree.h"
19 #include "xfs_trans.h"
20 #include "xfs_alloc.h"
21 #include "xfs_bmap.h"
22 #include "xfs_bmap_util.h"
23 #include "xfs_bmap_btree.h"
24 #include "xfs_rtalloc.h"
25 #include "xfs_errortag.h"
26 #include "xfs_error.h"
27 #include "xfs_quota.h"
28 #include "xfs_trans_space.h"
29 #include "xfs_buf_item.h"
30 #include "xfs_trace.h"
31 #include "xfs_attr_leaf.h"
32 #include "xfs_filestream.h"
33 #include "xfs_rmap.h"
34 #include "xfs_ag_resv.h"
35 #include "xfs_refcount.h"
36 #include "xfs_icache.h"
37 #include "xfs_iomap.h"
38 
39 
40 kmem_zone_t		*xfs_bmap_free_item_zone;
41 
42 /*
43  * Miscellaneous helper functions
44  */
45 
46 /*
47  * Compute and fill in the value of the maximum depth of a bmap btree
48  * in this filesystem.  Done once, during mount.
49  */
50 void
51 xfs_bmap_compute_maxlevels(
52 	xfs_mount_t	*mp,		/* file system mount structure */
53 	int		whichfork)	/* data or attr fork */
54 {
55 	int		level;		/* btree level */
56 	uint		maxblocks;	/* max blocks at this level */
57 	uint		maxleafents;	/* max leaf entries possible */
58 	int		maxrootrecs;	/* max records in root block */
59 	int		minleafrecs;	/* min records in leaf block */
60 	int		minnoderecs;	/* min records in node block */
61 	int		sz;		/* root block size */
62 
63 	/*
64 	 * The maximum number of extents in a file, hence the maximum number of
65 	 * leaf entries, is controlled by the size of the on-disk extent count,
66 	 * either a signed 32-bit number for the data fork, or a signed 16-bit
67 	 * number for the attr fork.
68 	 *
69 	 * Note that we can no longer assume that if we are in ATTR1 that
70 	 * the fork offset of all the inodes will be
71 	 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
72 	 * with ATTR2 and then mounted back with ATTR1, keeping the
73 	 * di_forkoff's fixed but probably at various positions. Therefore,
74 	 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
75 	 * of a minimum size available.
76 	 */
77 	if (whichfork == XFS_DATA_FORK) {
78 		maxleafents = MAXEXTNUM;
79 		sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
80 	} else {
81 		maxleafents = MAXAEXTNUM;
82 		sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
83 	}
84 	maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
85 	minleafrecs = mp->m_bmap_dmnr[0];
86 	minnoderecs = mp->m_bmap_dmnr[1];
87 	maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
88 	for (level = 1; maxblocks > 1; level++) {
89 		if (maxblocks <= maxrootrecs)
90 			maxblocks = 1;
91 		else
92 			maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
93 	}
94 	mp->m_bm_maxlevels[whichfork] = level;
95 }
96 
97 STATIC int				/* error */
98 xfs_bmbt_lookup_eq(
99 	struct xfs_btree_cur	*cur,
100 	struct xfs_bmbt_irec	*irec,
101 	int			*stat)	/* success/failure */
102 {
103 	cur->bc_rec.b = *irec;
104 	return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
105 }
106 
107 STATIC int				/* error */
108 xfs_bmbt_lookup_first(
109 	struct xfs_btree_cur	*cur,
110 	int			*stat)	/* success/failure */
111 {
112 	cur->bc_rec.b.br_startoff = 0;
113 	cur->bc_rec.b.br_startblock = 0;
114 	cur->bc_rec.b.br_blockcount = 0;
115 	return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
116 }
117 
118 /*
119  * Check if the inode needs to be converted to btree format.
120  */
121 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
122 {
123 	struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
124 
125 	return whichfork != XFS_COW_FORK &&
126 		ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
127 		ifp->if_nextents > XFS_IFORK_MAXEXT(ip, whichfork);
128 }
129 
130 /*
131  * Check if the inode should be converted to extent format.
132  */
133 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
134 {
135 	struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
136 
137 	return whichfork != XFS_COW_FORK &&
138 		ifp->if_format == XFS_DINODE_FMT_BTREE &&
139 		ifp->if_nextents <= XFS_IFORK_MAXEXT(ip, whichfork);
140 }
141 
142 /*
143  * Update the record referred to by cur to the value given by irec
144  * This either works (return 0) or gets an EFSCORRUPTED error.
145  */
146 STATIC int
147 xfs_bmbt_update(
148 	struct xfs_btree_cur	*cur,
149 	struct xfs_bmbt_irec	*irec)
150 {
151 	union xfs_btree_rec	rec;
152 
153 	xfs_bmbt_disk_set_all(&rec.bmbt, irec);
154 	return xfs_btree_update(cur, &rec);
155 }
156 
157 /*
158  * Compute the worst-case number of indirect blocks that will be used
159  * for ip's delayed extent of length "len".
160  */
161 STATIC xfs_filblks_t
162 xfs_bmap_worst_indlen(
163 	xfs_inode_t	*ip,		/* incore inode pointer */
164 	xfs_filblks_t	len)		/* delayed extent length */
165 {
166 	int		level;		/* btree level number */
167 	int		maxrecs;	/* maximum record count at this level */
168 	xfs_mount_t	*mp;		/* mount structure */
169 	xfs_filblks_t	rval;		/* return value */
170 
171 	mp = ip->i_mount;
172 	maxrecs = mp->m_bmap_dmxr[0];
173 	for (level = 0, rval = 0;
174 	     level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
175 	     level++) {
176 		len += maxrecs - 1;
177 		do_div(len, maxrecs);
178 		rval += len;
179 		if (len == 1)
180 			return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
181 				level - 1;
182 		if (level == 0)
183 			maxrecs = mp->m_bmap_dmxr[1];
184 	}
185 	return rval;
186 }
187 
188 /*
189  * Calculate the default attribute fork offset for newly created inodes.
190  */
191 uint
192 xfs_default_attroffset(
193 	struct xfs_inode	*ip)
194 {
195 	struct xfs_mount	*mp = ip->i_mount;
196 	uint			offset;
197 
198 	if (mp->m_sb.sb_inodesize == 256)
199 		offset = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
200 	else
201 		offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
202 
203 	ASSERT(offset < XFS_LITINO(mp));
204 	return offset;
205 }
206 
207 /*
208  * Helper routine to reset inode di_forkoff field when switching
209  * attribute fork from local to extent format - we reset it where
210  * possible to make space available for inline data fork extents.
211  */
212 STATIC void
213 xfs_bmap_forkoff_reset(
214 	xfs_inode_t	*ip,
215 	int		whichfork)
216 {
217 	if (whichfork == XFS_ATTR_FORK &&
218 	    ip->i_df.if_format != XFS_DINODE_FMT_DEV &&
219 	    ip->i_df.if_format != XFS_DINODE_FMT_BTREE) {
220 		uint	dfl_forkoff = xfs_default_attroffset(ip) >> 3;
221 
222 		if (dfl_forkoff > ip->i_d.di_forkoff)
223 			ip->i_d.di_forkoff = dfl_forkoff;
224 	}
225 }
226 
227 #ifdef DEBUG
228 STATIC struct xfs_buf *
229 xfs_bmap_get_bp(
230 	struct xfs_btree_cur	*cur,
231 	xfs_fsblock_t		bno)
232 {
233 	struct xfs_log_item	*lip;
234 	int			i;
235 
236 	if (!cur)
237 		return NULL;
238 
239 	for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
240 		if (!cur->bc_bufs[i])
241 			break;
242 		if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
243 			return cur->bc_bufs[i];
244 	}
245 
246 	/* Chase down all the log items to see if the bp is there */
247 	list_for_each_entry(lip, &cur->bc_tp->t_items, li_trans) {
248 		struct xfs_buf_log_item	*bip = (struct xfs_buf_log_item *)lip;
249 
250 		if (bip->bli_item.li_type == XFS_LI_BUF &&
251 		    XFS_BUF_ADDR(bip->bli_buf) == bno)
252 			return bip->bli_buf;
253 	}
254 
255 	return NULL;
256 }
257 
258 STATIC void
259 xfs_check_block(
260 	struct xfs_btree_block	*block,
261 	xfs_mount_t		*mp,
262 	int			root,
263 	short			sz)
264 {
265 	int			i, j, dmxr;
266 	__be64			*pp, *thispa;	/* pointer to block address */
267 	xfs_bmbt_key_t		*prevp, *keyp;
268 
269 	ASSERT(be16_to_cpu(block->bb_level) > 0);
270 
271 	prevp = NULL;
272 	for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
273 		dmxr = mp->m_bmap_dmxr[0];
274 		keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
275 
276 		if (prevp) {
277 			ASSERT(be64_to_cpu(prevp->br_startoff) <
278 			       be64_to_cpu(keyp->br_startoff));
279 		}
280 		prevp = keyp;
281 
282 		/*
283 		 * Compare the block numbers to see if there are dups.
284 		 */
285 		if (root)
286 			pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
287 		else
288 			pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
289 
290 		for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
291 			if (root)
292 				thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
293 			else
294 				thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
295 			if (*thispa == *pp) {
296 				xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
297 					__func__, j, i,
298 					(unsigned long long)be64_to_cpu(*thispa));
299 				xfs_err(mp, "%s: ptrs are equal in node\n",
300 					__func__);
301 				xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
302 			}
303 		}
304 	}
305 }
306 
307 /*
308  * Check that the extents for the inode ip are in the right order in all
309  * btree leaves. THis becomes prohibitively expensive for large extent count
310  * files, so don't bother with inodes that have more than 10,000 extents in
311  * them. The btree record ordering checks will still be done, so for such large
312  * bmapbt constructs that is going to catch most corruptions.
313  */
314 STATIC void
315 xfs_bmap_check_leaf_extents(
316 	xfs_btree_cur_t		*cur,	/* btree cursor or null */
317 	xfs_inode_t		*ip,		/* incore inode pointer */
318 	int			whichfork)	/* data or attr fork */
319 {
320 	struct xfs_mount	*mp = ip->i_mount;
321 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
322 	struct xfs_btree_block	*block;	/* current btree block */
323 	xfs_fsblock_t		bno;	/* block # of "block" */
324 	xfs_buf_t		*bp;	/* buffer for "block" */
325 	int			error;	/* error return value */
326 	xfs_extnum_t		i=0, j;	/* index into the extents list */
327 	int			level;	/* btree level, for checking */
328 	__be64			*pp;	/* pointer to block address */
329 	xfs_bmbt_rec_t		*ep;	/* pointer to current extent */
330 	xfs_bmbt_rec_t		last = {0, 0}; /* last extent in prev block */
331 	xfs_bmbt_rec_t		*nextp;	/* pointer to next extent */
332 	int			bp_release = 0;
333 
334 	if (ifp->if_format != XFS_DINODE_FMT_BTREE)
335 		return;
336 
337 	/* skip large extent count inodes */
338 	if (ip->i_df.if_nextents > 10000)
339 		return;
340 
341 	bno = NULLFSBLOCK;
342 	block = ifp->if_broot;
343 	/*
344 	 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
345 	 */
346 	level = be16_to_cpu(block->bb_level);
347 	ASSERT(level > 0);
348 	xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
349 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
350 	bno = be64_to_cpu(*pp);
351 
352 	ASSERT(bno != NULLFSBLOCK);
353 	ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
354 	ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
355 
356 	/*
357 	 * Go down the tree until leaf level is reached, following the first
358 	 * pointer (leftmost) at each level.
359 	 */
360 	while (level-- > 0) {
361 		/* See if buf is in cur first */
362 		bp_release = 0;
363 		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
364 		if (!bp) {
365 			bp_release = 1;
366 			error = xfs_btree_read_bufl(mp, NULL, bno, &bp,
367 						XFS_BMAP_BTREE_REF,
368 						&xfs_bmbt_buf_ops);
369 			if (error)
370 				goto error_norelse;
371 		}
372 		block = XFS_BUF_TO_BLOCK(bp);
373 		if (level == 0)
374 			break;
375 
376 		/*
377 		 * Check this block for basic sanity (increasing keys and
378 		 * no duplicate blocks).
379 		 */
380 
381 		xfs_check_block(block, mp, 0, 0);
382 		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
383 		bno = be64_to_cpu(*pp);
384 		if (XFS_IS_CORRUPT(mp, !xfs_verify_fsbno(mp, bno))) {
385 			error = -EFSCORRUPTED;
386 			goto error0;
387 		}
388 		if (bp_release) {
389 			bp_release = 0;
390 			xfs_trans_brelse(NULL, bp);
391 		}
392 	}
393 
394 	/*
395 	 * Here with bp and block set to the leftmost leaf node in the tree.
396 	 */
397 	i = 0;
398 
399 	/*
400 	 * Loop over all leaf nodes checking that all extents are in the right order.
401 	 */
402 	for (;;) {
403 		xfs_fsblock_t	nextbno;
404 		xfs_extnum_t	num_recs;
405 
406 
407 		num_recs = xfs_btree_get_numrecs(block);
408 
409 		/*
410 		 * Read-ahead the next leaf block, if any.
411 		 */
412 
413 		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
414 
415 		/*
416 		 * Check all the extents to make sure they are OK.
417 		 * If we had a previous block, the last entry should
418 		 * conform with the first entry in this one.
419 		 */
420 
421 		ep = XFS_BMBT_REC_ADDR(mp, block, 1);
422 		if (i) {
423 			ASSERT(xfs_bmbt_disk_get_startoff(&last) +
424 			       xfs_bmbt_disk_get_blockcount(&last) <=
425 			       xfs_bmbt_disk_get_startoff(ep));
426 		}
427 		for (j = 1; j < num_recs; j++) {
428 			nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
429 			ASSERT(xfs_bmbt_disk_get_startoff(ep) +
430 			       xfs_bmbt_disk_get_blockcount(ep) <=
431 			       xfs_bmbt_disk_get_startoff(nextp));
432 			ep = nextp;
433 		}
434 
435 		last = *ep;
436 		i += num_recs;
437 		if (bp_release) {
438 			bp_release = 0;
439 			xfs_trans_brelse(NULL, bp);
440 		}
441 		bno = nextbno;
442 		/*
443 		 * If we've reached the end, stop.
444 		 */
445 		if (bno == NULLFSBLOCK)
446 			break;
447 
448 		bp_release = 0;
449 		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
450 		if (!bp) {
451 			bp_release = 1;
452 			error = xfs_btree_read_bufl(mp, NULL, bno, &bp,
453 						XFS_BMAP_BTREE_REF,
454 						&xfs_bmbt_buf_ops);
455 			if (error)
456 				goto error_norelse;
457 		}
458 		block = XFS_BUF_TO_BLOCK(bp);
459 	}
460 
461 	return;
462 
463 error0:
464 	xfs_warn(mp, "%s: at error0", __func__);
465 	if (bp_release)
466 		xfs_trans_brelse(NULL, bp);
467 error_norelse:
468 	xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
469 		__func__, i);
470 	xfs_err(mp, "%s: CORRUPTED BTREE OR SOMETHING", __func__);
471 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
472 	return;
473 }
474 
475 /*
476  * Validate that the bmbt_irecs being returned from bmapi are valid
477  * given the caller's original parameters.  Specifically check the
478  * ranges of the returned irecs to ensure that they only extend beyond
479  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
480  */
481 STATIC void
482 xfs_bmap_validate_ret(
483 	xfs_fileoff_t		bno,
484 	xfs_filblks_t		len,
485 	int			flags,
486 	xfs_bmbt_irec_t		*mval,
487 	int			nmap,
488 	int			ret_nmap)
489 {
490 	int			i;		/* index to map values */
491 
492 	ASSERT(ret_nmap <= nmap);
493 
494 	for (i = 0; i < ret_nmap; i++) {
495 		ASSERT(mval[i].br_blockcount > 0);
496 		if (!(flags & XFS_BMAPI_ENTIRE)) {
497 			ASSERT(mval[i].br_startoff >= bno);
498 			ASSERT(mval[i].br_blockcount <= len);
499 			ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
500 			       bno + len);
501 		} else {
502 			ASSERT(mval[i].br_startoff < bno + len);
503 			ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
504 			       bno);
505 		}
506 		ASSERT(i == 0 ||
507 		       mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
508 		       mval[i].br_startoff);
509 		ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
510 		       mval[i].br_startblock != HOLESTARTBLOCK);
511 		ASSERT(mval[i].br_state == XFS_EXT_NORM ||
512 		       mval[i].br_state == XFS_EXT_UNWRITTEN);
513 	}
514 }
515 
516 #else
517 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)		do { } while (0)
518 #define	xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)	do { } while (0)
519 #endif /* DEBUG */
520 
521 /*
522  * bmap free list manipulation functions
523  */
524 
525 /*
526  * Add the extent to the list of extents to be free at transaction end.
527  * The list is maintained sorted (by block number).
528  */
529 void
530 __xfs_bmap_add_free(
531 	struct xfs_trans		*tp,
532 	xfs_fsblock_t			bno,
533 	xfs_filblks_t			len,
534 	const struct xfs_owner_info	*oinfo,
535 	bool				skip_discard)
536 {
537 	struct xfs_extent_free_item	*new;		/* new element */
538 #ifdef DEBUG
539 	struct xfs_mount		*mp = tp->t_mountp;
540 	xfs_agnumber_t			agno;
541 	xfs_agblock_t			agbno;
542 
543 	ASSERT(bno != NULLFSBLOCK);
544 	ASSERT(len > 0);
545 	ASSERT(len <= MAXEXTLEN);
546 	ASSERT(!isnullstartblock(bno));
547 	agno = XFS_FSB_TO_AGNO(mp, bno);
548 	agbno = XFS_FSB_TO_AGBNO(mp, bno);
549 	ASSERT(agno < mp->m_sb.sb_agcount);
550 	ASSERT(agbno < mp->m_sb.sb_agblocks);
551 	ASSERT(len < mp->m_sb.sb_agblocks);
552 	ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
553 #endif
554 	ASSERT(xfs_bmap_free_item_zone != NULL);
555 
556 	new = kmem_cache_alloc(xfs_bmap_free_item_zone,
557 			       GFP_KERNEL | __GFP_NOFAIL);
558 	new->xefi_startblock = bno;
559 	new->xefi_blockcount = (xfs_extlen_t)len;
560 	if (oinfo)
561 		new->xefi_oinfo = *oinfo;
562 	else
563 		new->xefi_oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
564 	new->xefi_skip_discard = skip_discard;
565 	trace_xfs_bmap_free_defer(tp->t_mountp,
566 			XFS_FSB_TO_AGNO(tp->t_mountp, bno), 0,
567 			XFS_FSB_TO_AGBNO(tp->t_mountp, bno), len);
568 	xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list);
569 }
570 
571 /*
572  * Inode fork format manipulation functions
573  */
574 
575 /*
576  * Convert the inode format to extent format if it currently is in btree format,
577  * but the extent list is small enough that it fits into the extent format.
578  *
579  * Since the extents are already in-core, all we have to do is give up the space
580  * for the btree root and pitch the leaf block.
581  */
582 STATIC int				/* error */
583 xfs_bmap_btree_to_extents(
584 	struct xfs_trans	*tp,	/* transaction pointer */
585 	struct xfs_inode	*ip,	/* incore inode pointer */
586 	struct xfs_btree_cur	*cur,	/* btree cursor */
587 	int			*logflagsp, /* inode logging flags */
588 	int			whichfork)  /* data or attr fork */
589 {
590 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
591 	struct xfs_mount	*mp = ip->i_mount;
592 	struct xfs_btree_block	*rblock = ifp->if_broot;
593 	struct xfs_btree_block	*cblock;/* child btree block */
594 	xfs_fsblock_t		cbno;	/* child block number */
595 	xfs_buf_t		*cbp;	/* child block's buffer */
596 	int			error;	/* error return value */
597 	__be64			*pp;	/* ptr to block address */
598 	struct xfs_owner_info	oinfo;
599 
600 	/* check if we actually need the extent format first: */
601 	if (!xfs_bmap_wants_extents(ip, whichfork))
602 		return 0;
603 
604 	ASSERT(cur);
605 	ASSERT(whichfork != XFS_COW_FORK);
606 	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
607 	ASSERT(ifp->if_format == XFS_DINODE_FMT_BTREE);
608 	ASSERT(be16_to_cpu(rblock->bb_level) == 1);
609 	ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
610 	ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
611 
612 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
613 	cbno = be64_to_cpu(*pp);
614 #ifdef DEBUG
615 	if (XFS_IS_CORRUPT(cur->bc_mp, !xfs_btree_check_lptr(cur, cbno, 1)))
616 		return -EFSCORRUPTED;
617 #endif
618 	error = xfs_btree_read_bufl(mp, tp, cbno, &cbp, XFS_BMAP_BTREE_REF,
619 				&xfs_bmbt_buf_ops);
620 	if (error)
621 		return error;
622 	cblock = XFS_BUF_TO_BLOCK(cbp);
623 	if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
624 		return error;
625 	xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
626 	xfs_bmap_add_free(cur->bc_tp, cbno, 1, &oinfo);
627 	ip->i_d.di_nblocks--;
628 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
629 	xfs_trans_binval(tp, cbp);
630 	if (cur->bc_bufs[0] == cbp)
631 		cur->bc_bufs[0] = NULL;
632 	xfs_iroot_realloc(ip, -1, whichfork);
633 	ASSERT(ifp->if_broot == NULL);
634 	ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
635 	ifp->if_format = XFS_DINODE_FMT_EXTENTS;
636 	*logflagsp |= XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
637 	return 0;
638 }
639 
640 /*
641  * Convert an extents-format file into a btree-format file.
642  * The new file will have a root block (in the inode) and a single child block.
643  */
644 STATIC int					/* error */
645 xfs_bmap_extents_to_btree(
646 	struct xfs_trans	*tp,		/* transaction pointer */
647 	struct xfs_inode	*ip,		/* incore inode pointer */
648 	struct xfs_btree_cur	**curp,		/* cursor returned to caller */
649 	int			wasdel,		/* converting a delayed alloc */
650 	int			*logflagsp,	/* inode logging flags */
651 	int			whichfork)	/* data or attr fork */
652 {
653 	struct xfs_btree_block	*ablock;	/* allocated (child) bt block */
654 	struct xfs_buf		*abp;		/* buffer for ablock */
655 	struct xfs_alloc_arg	args;		/* allocation arguments */
656 	struct xfs_bmbt_rec	*arp;		/* child record pointer */
657 	struct xfs_btree_block	*block;		/* btree root block */
658 	struct xfs_btree_cur	*cur;		/* bmap btree cursor */
659 	int			error;		/* error return value */
660 	struct xfs_ifork	*ifp;		/* inode fork pointer */
661 	struct xfs_bmbt_key	*kp;		/* root block key pointer */
662 	struct xfs_mount	*mp;		/* mount structure */
663 	xfs_bmbt_ptr_t		*pp;		/* root block address pointer */
664 	struct xfs_iext_cursor	icur;
665 	struct xfs_bmbt_irec	rec;
666 	xfs_extnum_t		cnt = 0;
667 
668 	mp = ip->i_mount;
669 	ASSERT(whichfork != XFS_COW_FORK);
670 	ifp = XFS_IFORK_PTR(ip, whichfork);
671 	ASSERT(ifp->if_format == XFS_DINODE_FMT_EXTENTS);
672 
673 	/*
674 	 * Make space in the inode incore. This needs to be undone if we fail
675 	 * to expand the root.
676 	 */
677 	xfs_iroot_realloc(ip, 1, whichfork);
678 	ifp->if_flags |= XFS_IFBROOT;
679 
680 	/*
681 	 * Fill in the root.
682 	 */
683 	block = ifp->if_broot;
684 	xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
685 				 XFS_BTNUM_BMAP, 1, 1, ip->i_ino,
686 				 XFS_BTREE_LONG_PTRS);
687 	/*
688 	 * Need a cursor.  Can't allocate until bb_level is filled in.
689 	 */
690 	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
691 	cur->bc_ino.flags = wasdel ? XFS_BTCUR_BMBT_WASDEL : 0;
692 	/*
693 	 * Convert to a btree with two levels, one record in root.
694 	 */
695 	ifp->if_format = XFS_DINODE_FMT_BTREE;
696 	memset(&args, 0, sizeof(args));
697 	args.tp = tp;
698 	args.mp = mp;
699 	xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork);
700 	if (tp->t_firstblock == NULLFSBLOCK) {
701 		args.type = XFS_ALLOCTYPE_START_BNO;
702 		args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
703 	} else if (tp->t_flags & XFS_TRANS_LOWMODE) {
704 		args.type = XFS_ALLOCTYPE_START_BNO;
705 		args.fsbno = tp->t_firstblock;
706 	} else {
707 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
708 		args.fsbno = tp->t_firstblock;
709 	}
710 	args.minlen = args.maxlen = args.prod = 1;
711 	args.wasdel = wasdel;
712 	*logflagsp = 0;
713 	error = xfs_alloc_vextent(&args);
714 	if (error)
715 		goto out_root_realloc;
716 
717 	if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
718 		error = -ENOSPC;
719 		goto out_root_realloc;
720 	}
721 
722 	/*
723 	 * Allocation can't fail, the space was reserved.
724 	 */
725 	ASSERT(tp->t_firstblock == NULLFSBLOCK ||
726 	       args.agno >= XFS_FSB_TO_AGNO(mp, tp->t_firstblock));
727 	tp->t_firstblock = args.fsbno;
728 	cur->bc_ino.allocated++;
729 	ip->i_d.di_nblocks++;
730 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
731 	error = xfs_trans_get_buf(tp, mp->m_ddev_targp,
732 			XFS_FSB_TO_DADDR(mp, args.fsbno),
733 			mp->m_bsize, 0, &abp);
734 	if (error)
735 		goto out_unreserve_dquot;
736 
737 	/*
738 	 * Fill in the child block.
739 	 */
740 	abp->b_ops = &xfs_bmbt_buf_ops;
741 	ablock = XFS_BUF_TO_BLOCK(abp);
742 	xfs_btree_init_block_int(mp, ablock, abp->b_bn,
743 				XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
744 				XFS_BTREE_LONG_PTRS);
745 
746 	for_each_xfs_iext(ifp, &icur, &rec) {
747 		if (isnullstartblock(rec.br_startblock))
748 			continue;
749 		arp = XFS_BMBT_REC_ADDR(mp, ablock, 1 + cnt);
750 		xfs_bmbt_disk_set_all(arp, &rec);
751 		cnt++;
752 	}
753 	ASSERT(cnt == ifp->if_nextents);
754 	xfs_btree_set_numrecs(ablock, cnt);
755 
756 	/*
757 	 * Fill in the root key and pointer.
758 	 */
759 	kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
760 	arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
761 	kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
762 	pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
763 						be16_to_cpu(block->bb_level)));
764 	*pp = cpu_to_be64(args.fsbno);
765 
766 	/*
767 	 * Do all this logging at the end so that
768 	 * the root is at the right level.
769 	 */
770 	xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
771 	xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
772 	ASSERT(*curp == NULL);
773 	*curp = cur;
774 	*logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
775 	return 0;
776 
777 out_unreserve_dquot:
778 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
779 out_root_realloc:
780 	xfs_iroot_realloc(ip, -1, whichfork);
781 	ifp->if_format = XFS_DINODE_FMT_EXTENTS;
782 	ASSERT(ifp->if_broot == NULL);
783 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
784 
785 	return error;
786 }
787 
788 /*
789  * Convert a local file to an extents file.
790  * This code is out of bounds for data forks of regular files,
791  * since the file data needs to get logged so things will stay consistent.
792  * (The bmap-level manipulations are ok, though).
793  */
794 void
795 xfs_bmap_local_to_extents_empty(
796 	struct xfs_trans	*tp,
797 	struct xfs_inode	*ip,
798 	int			whichfork)
799 {
800 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
801 
802 	ASSERT(whichfork != XFS_COW_FORK);
803 	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
804 	ASSERT(ifp->if_bytes == 0);
805 	ASSERT(ifp->if_nextents == 0);
806 
807 	xfs_bmap_forkoff_reset(ip, whichfork);
808 	ifp->if_flags &= ~XFS_IFINLINE;
809 	ifp->if_flags |= XFS_IFEXTENTS;
810 	ifp->if_u1.if_root = NULL;
811 	ifp->if_height = 0;
812 	ifp->if_format = XFS_DINODE_FMT_EXTENTS;
813 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
814 }
815 
816 
817 STATIC int				/* error */
818 xfs_bmap_local_to_extents(
819 	xfs_trans_t	*tp,		/* transaction pointer */
820 	xfs_inode_t	*ip,		/* incore inode pointer */
821 	xfs_extlen_t	total,		/* total blocks needed by transaction */
822 	int		*logflagsp,	/* inode logging flags */
823 	int		whichfork,
824 	void		(*init_fn)(struct xfs_trans *tp,
825 				   struct xfs_buf *bp,
826 				   struct xfs_inode *ip,
827 				   struct xfs_ifork *ifp))
828 {
829 	int		error = 0;
830 	int		flags;		/* logging flags returned */
831 	struct xfs_ifork *ifp;		/* inode fork pointer */
832 	xfs_alloc_arg_t	args;		/* allocation arguments */
833 	xfs_buf_t	*bp;		/* buffer for extent block */
834 	struct xfs_bmbt_irec rec;
835 	struct xfs_iext_cursor icur;
836 
837 	/*
838 	 * We don't want to deal with the case of keeping inode data inline yet.
839 	 * So sending the data fork of a regular inode is invalid.
840 	 */
841 	ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
842 	ifp = XFS_IFORK_PTR(ip, whichfork);
843 	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
844 
845 	if (!ifp->if_bytes) {
846 		xfs_bmap_local_to_extents_empty(tp, ip, whichfork);
847 		flags = XFS_ILOG_CORE;
848 		goto done;
849 	}
850 
851 	flags = 0;
852 	error = 0;
853 	ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS)) == XFS_IFINLINE);
854 	memset(&args, 0, sizeof(args));
855 	args.tp = tp;
856 	args.mp = ip->i_mount;
857 	xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0);
858 	/*
859 	 * Allocate a block.  We know we need only one, since the
860 	 * file currently fits in an inode.
861 	 */
862 	if (tp->t_firstblock == NULLFSBLOCK) {
863 		args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
864 		args.type = XFS_ALLOCTYPE_START_BNO;
865 	} else {
866 		args.fsbno = tp->t_firstblock;
867 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
868 	}
869 	args.total = total;
870 	args.minlen = args.maxlen = args.prod = 1;
871 	error = xfs_alloc_vextent(&args);
872 	if (error)
873 		goto done;
874 
875 	/* Can't fail, the space was reserved. */
876 	ASSERT(args.fsbno != NULLFSBLOCK);
877 	ASSERT(args.len == 1);
878 	tp->t_firstblock = args.fsbno;
879 	error = xfs_trans_get_buf(tp, args.mp->m_ddev_targp,
880 			XFS_FSB_TO_DADDR(args.mp, args.fsbno),
881 			args.mp->m_bsize, 0, &bp);
882 	if (error)
883 		goto done;
884 
885 	/*
886 	 * Initialize the block, copy the data and log the remote buffer.
887 	 *
888 	 * The callout is responsible for logging because the remote format
889 	 * might differ from the local format and thus we don't know how much to
890 	 * log here. Note that init_fn must also set the buffer log item type
891 	 * correctly.
892 	 */
893 	init_fn(tp, bp, ip, ifp);
894 
895 	/* account for the change in fork size */
896 	xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
897 	xfs_bmap_local_to_extents_empty(tp, ip, whichfork);
898 	flags |= XFS_ILOG_CORE;
899 
900 	ifp->if_u1.if_root = NULL;
901 	ifp->if_height = 0;
902 
903 	rec.br_startoff = 0;
904 	rec.br_startblock = args.fsbno;
905 	rec.br_blockcount = 1;
906 	rec.br_state = XFS_EXT_NORM;
907 	xfs_iext_first(ifp, &icur);
908 	xfs_iext_insert(ip, &icur, &rec, 0);
909 
910 	ifp->if_nextents = 1;
911 	ip->i_d.di_nblocks = 1;
912 	xfs_trans_mod_dquot_byino(tp, ip,
913 		XFS_TRANS_DQ_BCOUNT, 1L);
914 	flags |= xfs_ilog_fext(whichfork);
915 
916 done:
917 	*logflagsp = flags;
918 	return error;
919 }
920 
921 /*
922  * Called from xfs_bmap_add_attrfork to handle btree format files.
923  */
924 STATIC int					/* error */
925 xfs_bmap_add_attrfork_btree(
926 	xfs_trans_t		*tp,		/* transaction pointer */
927 	xfs_inode_t		*ip,		/* incore inode pointer */
928 	int			*flags)		/* inode logging flags */
929 {
930 	xfs_btree_cur_t		*cur;		/* btree cursor */
931 	int			error;		/* error return value */
932 	xfs_mount_t		*mp;		/* file system mount struct */
933 	int			stat;		/* newroot status */
934 
935 	mp = ip->i_mount;
936 	if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
937 		*flags |= XFS_ILOG_DBROOT;
938 	else {
939 		cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
940 		error = xfs_bmbt_lookup_first(cur, &stat);
941 		if (error)
942 			goto error0;
943 		/* must be at least one entry */
944 		if (XFS_IS_CORRUPT(mp, stat != 1)) {
945 			error = -EFSCORRUPTED;
946 			goto error0;
947 		}
948 		if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
949 			goto error0;
950 		if (stat == 0) {
951 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
952 			return -ENOSPC;
953 		}
954 		cur->bc_ino.allocated = 0;
955 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
956 	}
957 	return 0;
958 error0:
959 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
960 	return error;
961 }
962 
963 /*
964  * Called from xfs_bmap_add_attrfork to handle extents format files.
965  */
966 STATIC int					/* error */
967 xfs_bmap_add_attrfork_extents(
968 	struct xfs_trans	*tp,		/* transaction pointer */
969 	struct xfs_inode	*ip,		/* incore inode pointer */
970 	int			*flags)		/* inode logging flags */
971 {
972 	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
973 	int			error;		/* error return value */
974 
975 	if (ip->i_df.if_nextents * sizeof(struct xfs_bmbt_rec) <=
976 	    XFS_IFORK_DSIZE(ip))
977 		return 0;
978 	cur = NULL;
979 	error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, flags,
980 					  XFS_DATA_FORK);
981 	if (cur) {
982 		cur->bc_ino.allocated = 0;
983 		xfs_btree_del_cursor(cur, error);
984 	}
985 	return error;
986 }
987 
988 /*
989  * Called from xfs_bmap_add_attrfork to handle local format files. Each
990  * different data fork content type needs a different callout to do the
991  * conversion. Some are basic and only require special block initialisation
992  * callouts for the data formating, others (directories) are so specialised they
993  * handle everything themselves.
994  *
995  * XXX (dgc): investigate whether directory conversion can use the generic
996  * formatting callout. It should be possible - it's just a very complex
997  * formatter.
998  */
999 STATIC int					/* error */
1000 xfs_bmap_add_attrfork_local(
1001 	struct xfs_trans	*tp,		/* transaction pointer */
1002 	struct xfs_inode	*ip,		/* incore inode pointer */
1003 	int			*flags)		/* inode logging flags */
1004 {
1005 	struct xfs_da_args	dargs;		/* args for dir/attr code */
1006 
1007 	if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1008 		return 0;
1009 
1010 	if (S_ISDIR(VFS_I(ip)->i_mode)) {
1011 		memset(&dargs, 0, sizeof(dargs));
1012 		dargs.geo = ip->i_mount->m_dir_geo;
1013 		dargs.dp = ip;
1014 		dargs.total = dargs.geo->fsbcount;
1015 		dargs.whichfork = XFS_DATA_FORK;
1016 		dargs.trans = tp;
1017 		return xfs_dir2_sf_to_block(&dargs);
1018 	}
1019 
1020 	if (S_ISLNK(VFS_I(ip)->i_mode))
1021 		return xfs_bmap_local_to_extents(tp, ip, 1, flags,
1022 						 XFS_DATA_FORK,
1023 						 xfs_symlink_local_to_remote);
1024 
1025 	/* should only be called for types that support local format data */
1026 	ASSERT(0);
1027 	return -EFSCORRUPTED;
1028 }
1029 
1030 /* Set an inode attr fork off based on the format */
1031 int
1032 xfs_bmap_set_attrforkoff(
1033 	struct xfs_inode	*ip,
1034 	int			size,
1035 	int			*version)
1036 {
1037 	switch (ip->i_df.if_format) {
1038 	case XFS_DINODE_FMT_DEV:
1039 		ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1040 		break;
1041 	case XFS_DINODE_FMT_LOCAL:
1042 	case XFS_DINODE_FMT_EXTENTS:
1043 	case XFS_DINODE_FMT_BTREE:
1044 		ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1045 		if (!ip->i_d.di_forkoff)
1046 			ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1047 		else if ((ip->i_mount->m_flags & XFS_MOUNT_ATTR2) && version)
1048 			*version = 2;
1049 		break;
1050 	default:
1051 		ASSERT(0);
1052 		return -EINVAL;
1053 	}
1054 
1055 	return 0;
1056 }
1057 
1058 /*
1059  * Convert inode from non-attributed to attributed.
1060  * Must not be in a transaction, ip must not be locked.
1061  */
1062 int						/* error code */
1063 xfs_bmap_add_attrfork(
1064 	xfs_inode_t		*ip,		/* incore inode pointer */
1065 	int			size,		/* space new attribute needs */
1066 	int			rsvd)		/* xact may use reserved blks */
1067 {
1068 	xfs_mount_t		*mp;		/* mount structure */
1069 	xfs_trans_t		*tp;		/* transaction pointer */
1070 	int			blks;		/* space reservation */
1071 	int			version = 1;	/* superblock attr version */
1072 	int			logflags;	/* logging flags */
1073 	int			error;		/* error return value */
1074 
1075 	ASSERT(XFS_IFORK_Q(ip) == 0);
1076 
1077 	mp = ip->i_mount;
1078 	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1079 
1080 	blks = XFS_ADDAFORK_SPACE_RES(mp);
1081 
1082 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_addafork, blks, 0,
1083 			rsvd ? XFS_TRANS_RESERVE : 0, &tp);
1084 	if (error)
1085 		return error;
1086 
1087 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1088 	error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1089 			XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1090 			XFS_QMOPT_RES_REGBLKS);
1091 	if (error)
1092 		goto trans_cancel;
1093 	if (XFS_IFORK_Q(ip))
1094 		goto trans_cancel;
1095 
1096 	xfs_trans_ijoin(tp, ip, 0);
1097 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1098 	error = xfs_bmap_set_attrforkoff(ip, size, &version);
1099 	if (error)
1100 		goto trans_cancel;
1101 	ASSERT(ip->i_afp == NULL);
1102 
1103 	ip->i_afp = kmem_cache_zalloc(xfs_ifork_zone,
1104 				      GFP_KERNEL | __GFP_NOFAIL);
1105 
1106 	ip->i_afp->if_format = XFS_DINODE_FMT_EXTENTS;
1107 	ip->i_afp->if_flags = XFS_IFEXTENTS;
1108 	logflags = 0;
1109 	switch (ip->i_df.if_format) {
1110 	case XFS_DINODE_FMT_LOCAL:
1111 		error = xfs_bmap_add_attrfork_local(tp, ip, &logflags);
1112 		break;
1113 	case XFS_DINODE_FMT_EXTENTS:
1114 		error = xfs_bmap_add_attrfork_extents(tp, ip, &logflags);
1115 		break;
1116 	case XFS_DINODE_FMT_BTREE:
1117 		error = xfs_bmap_add_attrfork_btree(tp, ip, &logflags);
1118 		break;
1119 	default:
1120 		error = 0;
1121 		break;
1122 	}
1123 	if (logflags)
1124 		xfs_trans_log_inode(tp, ip, logflags);
1125 	if (error)
1126 		goto trans_cancel;
1127 	if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1128 	   (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1129 		bool log_sb = false;
1130 
1131 		spin_lock(&mp->m_sb_lock);
1132 		if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1133 			xfs_sb_version_addattr(&mp->m_sb);
1134 			log_sb = true;
1135 		}
1136 		if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1137 			xfs_sb_version_addattr2(&mp->m_sb);
1138 			log_sb = true;
1139 		}
1140 		spin_unlock(&mp->m_sb_lock);
1141 		if (log_sb)
1142 			xfs_log_sb(tp);
1143 	}
1144 
1145 	error = xfs_trans_commit(tp);
1146 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1147 	return error;
1148 
1149 trans_cancel:
1150 	xfs_trans_cancel(tp);
1151 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1152 	return error;
1153 }
1154 
1155 /*
1156  * Internal and external extent tree search functions.
1157  */
1158 
1159 struct xfs_iread_state {
1160 	struct xfs_iext_cursor	icur;
1161 	xfs_extnum_t		loaded;
1162 };
1163 
1164 /* Stuff every bmbt record from this block into the incore extent map. */
1165 static int
1166 xfs_iread_bmbt_block(
1167 	struct xfs_btree_cur	*cur,
1168 	int			level,
1169 	void			*priv)
1170 {
1171 	struct xfs_iread_state	*ir = priv;
1172 	struct xfs_mount	*mp = cur->bc_mp;
1173 	struct xfs_inode	*ip = cur->bc_ino.ip;
1174 	struct xfs_btree_block	*block;
1175 	struct xfs_buf		*bp;
1176 	struct xfs_bmbt_rec	*frp;
1177 	xfs_extnum_t		num_recs;
1178 	xfs_extnum_t		j;
1179 	int			whichfork = cur->bc_ino.whichfork;
1180 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1181 
1182 	block = xfs_btree_get_block(cur, level, &bp);
1183 
1184 	/* Abort if we find more records than nextents. */
1185 	num_recs = xfs_btree_get_numrecs(block);
1186 	if (unlikely(ir->loaded + num_recs > ifp->if_nextents)) {
1187 		xfs_warn(ip->i_mount, "corrupt dinode %llu, (btree extents).",
1188 				(unsigned long long)ip->i_ino);
1189 		xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, block,
1190 				sizeof(*block), __this_address);
1191 		return -EFSCORRUPTED;
1192 	}
1193 
1194 	/* Copy records into the incore cache. */
1195 	frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1196 	for (j = 0; j < num_recs; j++, frp++, ir->loaded++) {
1197 		struct xfs_bmbt_irec	new;
1198 		xfs_failaddr_t		fa;
1199 
1200 		xfs_bmbt_disk_get_all(frp, &new);
1201 		fa = xfs_bmap_validate_extent(ip, whichfork, &new);
1202 		if (fa) {
1203 			xfs_inode_verifier_error(ip, -EFSCORRUPTED,
1204 					"xfs_iread_extents(2)", frp,
1205 					sizeof(*frp), fa);
1206 			return -EFSCORRUPTED;
1207 		}
1208 		xfs_iext_insert(ip, &ir->icur, &new,
1209 				xfs_bmap_fork_to_state(whichfork));
1210 		trace_xfs_read_extent(ip, &ir->icur,
1211 				xfs_bmap_fork_to_state(whichfork), _THIS_IP_);
1212 		xfs_iext_next(ifp, &ir->icur);
1213 	}
1214 
1215 	return 0;
1216 }
1217 
1218 /*
1219  * Read in extents from a btree-format inode.
1220  */
1221 int
1222 xfs_iread_extents(
1223 	struct xfs_trans	*tp,
1224 	struct xfs_inode	*ip,
1225 	int			whichfork)
1226 {
1227 	struct xfs_iread_state	ir;
1228 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1229 	struct xfs_mount	*mp = ip->i_mount;
1230 	struct xfs_btree_cur	*cur;
1231 	int			error;
1232 
1233 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1234 
1235 	if (XFS_IS_CORRUPT(mp, ifp->if_format != XFS_DINODE_FMT_BTREE)) {
1236 		error = -EFSCORRUPTED;
1237 		goto out;
1238 	}
1239 
1240 	ir.loaded = 0;
1241 	xfs_iext_first(ifp, &ir.icur);
1242 	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
1243 	error = xfs_btree_visit_blocks(cur, xfs_iread_bmbt_block,
1244 			XFS_BTREE_VISIT_RECORDS, &ir);
1245 	xfs_btree_del_cursor(cur, error);
1246 	if (error)
1247 		goto out;
1248 
1249 	if (XFS_IS_CORRUPT(mp, ir.loaded != ifp->if_nextents)) {
1250 		error = -EFSCORRUPTED;
1251 		goto out;
1252 	}
1253 	ASSERT(ir.loaded == xfs_iext_count(ifp));
1254 
1255 	ifp->if_flags |= XFS_IFEXTENTS;
1256 	return 0;
1257 out:
1258 	xfs_iext_destroy(ifp);
1259 	return error;
1260 }
1261 
1262 /*
1263  * Returns the relative block number of the first unused block(s) in the given
1264  * fork with at least "len" logically contiguous blocks free.  This is the
1265  * lowest-address hole if the fork has holes, else the first block past the end
1266  * of fork.  Return 0 if the fork is currently local (in-inode).
1267  */
1268 int						/* error */
1269 xfs_bmap_first_unused(
1270 	struct xfs_trans	*tp,		/* transaction pointer */
1271 	struct xfs_inode	*ip,		/* incore inode */
1272 	xfs_extlen_t		len,		/* size of hole to find */
1273 	xfs_fileoff_t		*first_unused,	/* unused block */
1274 	int			whichfork)	/* data or attr fork */
1275 {
1276 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1277 	struct xfs_bmbt_irec	got;
1278 	struct xfs_iext_cursor	icur;
1279 	xfs_fileoff_t		lastaddr = 0;
1280 	xfs_fileoff_t		lowest, max;
1281 	int			error;
1282 
1283 	if (ifp->if_format == XFS_DINODE_FMT_LOCAL) {
1284 		*first_unused = 0;
1285 		return 0;
1286 	}
1287 
1288 	ASSERT(xfs_ifork_has_extents(ifp));
1289 
1290 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1291 		error = xfs_iread_extents(tp, ip, whichfork);
1292 		if (error)
1293 			return error;
1294 	}
1295 
1296 	lowest = max = *first_unused;
1297 	for_each_xfs_iext(ifp, &icur, &got) {
1298 		/*
1299 		 * See if the hole before this extent will work.
1300 		 */
1301 		if (got.br_startoff >= lowest + len &&
1302 		    got.br_startoff - max >= len)
1303 			break;
1304 		lastaddr = got.br_startoff + got.br_blockcount;
1305 		max = XFS_FILEOFF_MAX(lastaddr, lowest);
1306 	}
1307 
1308 	*first_unused = max;
1309 	return 0;
1310 }
1311 
1312 /*
1313  * Returns the file-relative block number of the last block - 1 before
1314  * last_block (input value) in the file.
1315  * This is not based on i_size, it is based on the extent records.
1316  * Returns 0 for local files, as they do not have extent records.
1317  */
1318 int						/* error */
1319 xfs_bmap_last_before(
1320 	struct xfs_trans	*tp,		/* transaction pointer */
1321 	struct xfs_inode	*ip,		/* incore inode */
1322 	xfs_fileoff_t		*last_block,	/* last block */
1323 	int			whichfork)	/* data or attr fork */
1324 {
1325 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1326 	struct xfs_bmbt_irec	got;
1327 	struct xfs_iext_cursor	icur;
1328 	int			error;
1329 
1330 	switch (ifp->if_format) {
1331 	case XFS_DINODE_FMT_LOCAL:
1332 		*last_block = 0;
1333 		return 0;
1334 	case XFS_DINODE_FMT_BTREE:
1335 	case XFS_DINODE_FMT_EXTENTS:
1336 		break;
1337 	default:
1338 		ASSERT(0);
1339 		return -EFSCORRUPTED;
1340 	}
1341 
1342 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1343 		error = xfs_iread_extents(tp, ip, whichfork);
1344 		if (error)
1345 			return error;
1346 	}
1347 
1348 	if (!xfs_iext_lookup_extent_before(ip, ifp, last_block, &icur, &got))
1349 		*last_block = 0;
1350 	return 0;
1351 }
1352 
1353 int
1354 xfs_bmap_last_extent(
1355 	struct xfs_trans	*tp,
1356 	struct xfs_inode	*ip,
1357 	int			whichfork,
1358 	struct xfs_bmbt_irec	*rec,
1359 	int			*is_empty)
1360 {
1361 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1362 	struct xfs_iext_cursor	icur;
1363 	int			error;
1364 
1365 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1366 		error = xfs_iread_extents(tp, ip, whichfork);
1367 		if (error)
1368 			return error;
1369 	}
1370 
1371 	xfs_iext_last(ifp, &icur);
1372 	if (!xfs_iext_get_extent(ifp, &icur, rec))
1373 		*is_empty = 1;
1374 	else
1375 		*is_empty = 0;
1376 	return 0;
1377 }
1378 
1379 /*
1380  * Check the last inode extent to determine whether this allocation will result
1381  * in blocks being allocated at the end of the file. When we allocate new data
1382  * blocks at the end of the file which do not start at the previous data block,
1383  * we will try to align the new blocks at stripe unit boundaries.
1384  *
1385  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1386  * at, or past the EOF.
1387  */
1388 STATIC int
1389 xfs_bmap_isaeof(
1390 	struct xfs_bmalloca	*bma,
1391 	int			whichfork)
1392 {
1393 	struct xfs_bmbt_irec	rec;
1394 	int			is_empty;
1395 	int			error;
1396 
1397 	bma->aeof = false;
1398 	error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1399 				     &is_empty);
1400 	if (error)
1401 		return error;
1402 
1403 	if (is_empty) {
1404 		bma->aeof = true;
1405 		return 0;
1406 	}
1407 
1408 	/*
1409 	 * Check if we are allocation or past the last extent, or at least into
1410 	 * the last delayed allocated extent.
1411 	 */
1412 	bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1413 		(bma->offset >= rec.br_startoff &&
1414 		 isnullstartblock(rec.br_startblock));
1415 	return 0;
1416 }
1417 
1418 /*
1419  * Returns the file-relative block number of the first block past eof in
1420  * the file.  This is not based on i_size, it is based on the extent records.
1421  * Returns 0 for local files, as they do not have extent records.
1422  */
1423 int
1424 xfs_bmap_last_offset(
1425 	struct xfs_inode	*ip,
1426 	xfs_fileoff_t		*last_block,
1427 	int			whichfork)
1428 {
1429 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1430 	struct xfs_bmbt_irec	rec;
1431 	int			is_empty;
1432 	int			error;
1433 
1434 	*last_block = 0;
1435 
1436 	if (ifp->if_format == XFS_DINODE_FMT_LOCAL)
1437 		return 0;
1438 
1439 	if (XFS_IS_CORRUPT(ip->i_mount, !xfs_ifork_has_extents(ifp)))
1440 		return -EFSCORRUPTED;
1441 
1442 	error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1443 	if (error || is_empty)
1444 		return error;
1445 
1446 	*last_block = rec.br_startoff + rec.br_blockcount;
1447 	return 0;
1448 }
1449 
1450 /*
1451  * Returns whether the selected fork of the inode has exactly one
1452  * block or not.  For the data fork we check this matches di_size,
1453  * implying the file's range is 0..bsize-1.
1454  */
1455 int					/* 1=>1 block, 0=>otherwise */
1456 xfs_bmap_one_block(
1457 	struct xfs_inode	*ip,		/* incore inode */
1458 	int			whichfork)	/* data or attr fork */
1459 {
1460 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1461 	int			rval;		/* return value */
1462 	struct xfs_bmbt_irec	s;		/* internal version of extent */
1463 	struct xfs_iext_cursor icur;
1464 
1465 #ifndef DEBUG
1466 	if (whichfork == XFS_DATA_FORK)
1467 		return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1468 #endif	/* !DEBUG */
1469 	if (ifp->if_nextents != 1)
1470 		return 0;
1471 	if (ifp->if_format != XFS_DINODE_FMT_EXTENTS)
1472 		return 0;
1473 	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1474 	xfs_iext_first(ifp, &icur);
1475 	xfs_iext_get_extent(ifp, &icur, &s);
1476 	rval = s.br_startoff == 0 && s.br_blockcount == 1;
1477 	if (rval && whichfork == XFS_DATA_FORK)
1478 		ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1479 	return rval;
1480 }
1481 
1482 /*
1483  * Extent tree manipulation functions used during allocation.
1484  */
1485 
1486 /*
1487  * Convert a delayed allocation to a real allocation.
1488  */
1489 STATIC int				/* error */
1490 xfs_bmap_add_extent_delay_real(
1491 	struct xfs_bmalloca	*bma,
1492 	int			whichfork)
1493 {
1494 	struct xfs_mount	*mp = bma->ip->i_mount;
1495 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1496 	struct xfs_bmbt_irec	*new = &bma->got;
1497 	int			error;	/* error return value */
1498 	int			i;	/* temp state */
1499 	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
1500 	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
1501 					/* left is 0, right is 1, prev is 2 */
1502 	int			rval=0;	/* return value (logging flags) */
1503 	int			state = xfs_bmap_fork_to_state(whichfork);
1504 	xfs_filblks_t		da_new; /* new count del alloc blocks used */
1505 	xfs_filblks_t		da_old; /* old count del alloc blocks used */
1506 	xfs_filblks_t		temp=0;	/* value for da_new calculations */
1507 	int			tmp_rval;	/* partial logging flags */
1508 	struct xfs_bmbt_irec	old;
1509 
1510 	ASSERT(whichfork != XFS_ATTR_FORK);
1511 	ASSERT(!isnullstartblock(new->br_startblock));
1512 	ASSERT(!bma->cur ||
1513 	       (bma->cur->bc_ino.flags & XFS_BTCUR_BMBT_WASDEL));
1514 
1515 	XFS_STATS_INC(mp, xs_add_exlist);
1516 
1517 #define	LEFT		r[0]
1518 #define	RIGHT		r[1]
1519 #define	PREV		r[2]
1520 
1521 	/*
1522 	 * Set up a bunch of variables to make the tests simpler.
1523 	 */
1524 	xfs_iext_get_extent(ifp, &bma->icur, &PREV);
1525 	new_endoff = new->br_startoff + new->br_blockcount;
1526 	ASSERT(isnullstartblock(PREV.br_startblock));
1527 	ASSERT(PREV.br_startoff <= new->br_startoff);
1528 	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1529 
1530 	da_old = startblockval(PREV.br_startblock);
1531 	da_new = 0;
1532 
1533 	/*
1534 	 * Set flags determining what part of the previous delayed allocation
1535 	 * extent is being replaced by a real allocation.
1536 	 */
1537 	if (PREV.br_startoff == new->br_startoff)
1538 		state |= BMAP_LEFT_FILLING;
1539 	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1540 		state |= BMAP_RIGHT_FILLING;
1541 
1542 	/*
1543 	 * Check and set flags if this segment has a left neighbor.
1544 	 * Don't set contiguous if the combined extent would be too large.
1545 	 */
1546 	if (xfs_iext_peek_prev_extent(ifp, &bma->icur, &LEFT)) {
1547 		state |= BMAP_LEFT_VALID;
1548 		if (isnullstartblock(LEFT.br_startblock))
1549 			state |= BMAP_LEFT_DELAY;
1550 	}
1551 
1552 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1553 	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1554 	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1555 	    LEFT.br_state == new->br_state &&
1556 	    LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1557 		state |= BMAP_LEFT_CONTIG;
1558 
1559 	/*
1560 	 * Check and set flags if this segment has a right neighbor.
1561 	 * Don't set contiguous if the combined extent would be too large.
1562 	 * Also check for all-three-contiguous being too large.
1563 	 */
1564 	if (xfs_iext_peek_next_extent(ifp, &bma->icur, &RIGHT)) {
1565 		state |= BMAP_RIGHT_VALID;
1566 		if (isnullstartblock(RIGHT.br_startblock))
1567 			state |= BMAP_RIGHT_DELAY;
1568 	}
1569 
1570 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1571 	    new_endoff == RIGHT.br_startoff &&
1572 	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1573 	    new->br_state == RIGHT.br_state &&
1574 	    new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1575 	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1576 		       BMAP_RIGHT_FILLING)) !=
1577 		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1578 		       BMAP_RIGHT_FILLING) ||
1579 	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1580 			<= MAXEXTLEN))
1581 		state |= BMAP_RIGHT_CONTIG;
1582 
1583 	error = 0;
1584 	/*
1585 	 * Switch out based on the FILLING and CONTIG state bits.
1586 	 */
1587 	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1588 			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1589 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1590 	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1591 		/*
1592 		 * Filling in all of a previously delayed allocation extent.
1593 		 * The left and right neighbors are both contiguous with new.
1594 		 */
1595 		LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
1596 
1597 		xfs_iext_remove(bma->ip, &bma->icur, state);
1598 		xfs_iext_remove(bma->ip, &bma->icur, state);
1599 		xfs_iext_prev(ifp, &bma->icur);
1600 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1601 		ifp->if_nextents--;
1602 
1603 		if (bma->cur == NULL)
1604 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1605 		else {
1606 			rval = XFS_ILOG_CORE;
1607 			error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1608 			if (error)
1609 				goto done;
1610 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1611 				error = -EFSCORRUPTED;
1612 				goto done;
1613 			}
1614 			error = xfs_btree_delete(bma->cur, &i);
1615 			if (error)
1616 				goto done;
1617 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1618 				error = -EFSCORRUPTED;
1619 				goto done;
1620 			}
1621 			error = xfs_btree_decrement(bma->cur, 0, &i);
1622 			if (error)
1623 				goto done;
1624 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1625 				error = -EFSCORRUPTED;
1626 				goto done;
1627 			}
1628 			error = xfs_bmbt_update(bma->cur, &LEFT);
1629 			if (error)
1630 				goto done;
1631 		}
1632 		break;
1633 
1634 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1635 		/*
1636 		 * Filling in all of a previously delayed allocation extent.
1637 		 * The left neighbor is contiguous, the right is not.
1638 		 */
1639 		old = LEFT;
1640 		LEFT.br_blockcount += PREV.br_blockcount;
1641 
1642 		xfs_iext_remove(bma->ip, &bma->icur, state);
1643 		xfs_iext_prev(ifp, &bma->icur);
1644 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1645 
1646 		if (bma->cur == NULL)
1647 			rval = XFS_ILOG_DEXT;
1648 		else {
1649 			rval = 0;
1650 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1651 			if (error)
1652 				goto done;
1653 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1654 				error = -EFSCORRUPTED;
1655 				goto done;
1656 			}
1657 			error = xfs_bmbt_update(bma->cur, &LEFT);
1658 			if (error)
1659 				goto done;
1660 		}
1661 		break;
1662 
1663 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1664 		/*
1665 		 * Filling in all of a previously delayed allocation extent.
1666 		 * The right neighbor is contiguous, the left is not. Take care
1667 		 * with delay -> unwritten extent allocation here because the
1668 		 * delalloc record we are overwriting is always written.
1669 		 */
1670 		PREV.br_startblock = new->br_startblock;
1671 		PREV.br_blockcount += RIGHT.br_blockcount;
1672 		PREV.br_state = new->br_state;
1673 
1674 		xfs_iext_next(ifp, &bma->icur);
1675 		xfs_iext_remove(bma->ip, &bma->icur, state);
1676 		xfs_iext_prev(ifp, &bma->icur);
1677 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1678 
1679 		if (bma->cur == NULL)
1680 			rval = XFS_ILOG_DEXT;
1681 		else {
1682 			rval = 0;
1683 			error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1684 			if (error)
1685 				goto done;
1686 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1687 				error = -EFSCORRUPTED;
1688 				goto done;
1689 			}
1690 			error = xfs_bmbt_update(bma->cur, &PREV);
1691 			if (error)
1692 				goto done;
1693 		}
1694 		break;
1695 
1696 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1697 		/*
1698 		 * Filling in all of a previously delayed allocation extent.
1699 		 * Neither the left nor right neighbors are contiguous with
1700 		 * the new one.
1701 		 */
1702 		PREV.br_startblock = new->br_startblock;
1703 		PREV.br_state = new->br_state;
1704 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1705 		ifp->if_nextents++;
1706 
1707 		if (bma->cur == NULL)
1708 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1709 		else {
1710 			rval = XFS_ILOG_CORE;
1711 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1712 			if (error)
1713 				goto done;
1714 			if (XFS_IS_CORRUPT(mp, i != 0)) {
1715 				error = -EFSCORRUPTED;
1716 				goto done;
1717 			}
1718 			error = xfs_btree_insert(bma->cur, &i);
1719 			if (error)
1720 				goto done;
1721 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1722 				error = -EFSCORRUPTED;
1723 				goto done;
1724 			}
1725 		}
1726 		break;
1727 
1728 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1729 		/*
1730 		 * Filling in the first part of a previous delayed allocation.
1731 		 * The left neighbor is contiguous.
1732 		 */
1733 		old = LEFT;
1734 		temp = PREV.br_blockcount - new->br_blockcount;
1735 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1736 				startblockval(PREV.br_startblock));
1737 
1738 		LEFT.br_blockcount += new->br_blockcount;
1739 
1740 		PREV.br_blockcount = temp;
1741 		PREV.br_startoff += new->br_blockcount;
1742 		PREV.br_startblock = nullstartblock(da_new);
1743 
1744 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1745 		xfs_iext_prev(ifp, &bma->icur);
1746 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1747 
1748 		if (bma->cur == NULL)
1749 			rval = XFS_ILOG_DEXT;
1750 		else {
1751 			rval = 0;
1752 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1753 			if (error)
1754 				goto done;
1755 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1756 				error = -EFSCORRUPTED;
1757 				goto done;
1758 			}
1759 			error = xfs_bmbt_update(bma->cur, &LEFT);
1760 			if (error)
1761 				goto done;
1762 		}
1763 		break;
1764 
1765 	case BMAP_LEFT_FILLING:
1766 		/*
1767 		 * Filling in the first part of a previous delayed allocation.
1768 		 * The left neighbor is not contiguous.
1769 		 */
1770 		xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1771 		ifp->if_nextents++;
1772 
1773 		if (bma->cur == NULL)
1774 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1775 		else {
1776 			rval = XFS_ILOG_CORE;
1777 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1778 			if (error)
1779 				goto done;
1780 			if (XFS_IS_CORRUPT(mp, i != 0)) {
1781 				error = -EFSCORRUPTED;
1782 				goto done;
1783 			}
1784 			error = xfs_btree_insert(bma->cur, &i);
1785 			if (error)
1786 				goto done;
1787 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1788 				error = -EFSCORRUPTED;
1789 				goto done;
1790 			}
1791 		}
1792 
1793 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1794 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1795 					&bma->cur, 1, &tmp_rval, whichfork);
1796 			rval |= tmp_rval;
1797 			if (error)
1798 				goto done;
1799 		}
1800 
1801 		temp = PREV.br_blockcount - new->br_blockcount;
1802 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1803 			startblockval(PREV.br_startblock) -
1804 			(bma->cur ? bma->cur->bc_ino.allocated : 0));
1805 
1806 		PREV.br_startoff = new_endoff;
1807 		PREV.br_blockcount = temp;
1808 		PREV.br_startblock = nullstartblock(da_new);
1809 		xfs_iext_next(ifp, &bma->icur);
1810 		xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1811 		xfs_iext_prev(ifp, &bma->icur);
1812 		break;
1813 
1814 	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1815 		/*
1816 		 * Filling in the last part of a previous delayed allocation.
1817 		 * The right neighbor is contiguous with the new allocation.
1818 		 */
1819 		old = RIGHT;
1820 		RIGHT.br_startoff = new->br_startoff;
1821 		RIGHT.br_startblock = new->br_startblock;
1822 		RIGHT.br_blockcount += new->br_blockcount;
1823 
1824 		if (bma->cur == NULL)
1825 			rval = XFS_ILOG_DEXT;
1826 		else {
1827 			rval = 0;
1828 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1829 			if (error)
1830 				goto done;
1831 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1832 				error = -EFSCORRUPTED;
1833 				goto done;
1834 			}
1835 			error = xfs_bmbt_update(bma->cur, &RIGHT);
1836 			if (error)
1837 				goto done;
1838 		}
1839 
1840 		temp = PREV.br_blockcount - new->br_blockcount;
1841 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1842 			startblockval(PREV.br_startblock));
1843 
1844 		PREV.br_blockcount = temp;
1845 		PREV.br_startblock = nullstartblock(da_new);
1846 
1847 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1848 		xfs_iext_next(ifp, &bma->icur);
1849 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &RIGHT);
1850 		break;
1851 
1852 	case BMAP_RIGHT_FILLING:
1853 		/*
1854 		 * Filling in the last part of a previous delayed allocation.
1855 		 * The right neighbor is not contiguous.
1856 		 */
1857 		xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1858 		ifp->if_nextents++;
1859 
1860 		if (bma->cur == NULL)
1861 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1862 		else {
1863 			rval = XFS_ILOG_CORE;
1864 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1865 			if (error)
1866 				goto done;
1867 			if (XFS_IS_CORRUPT(mp, i != 0)) {
1868 				error = -EFSCORRUPTED;
1869 				goto done;
1870 			}
1871 			error = xfs_btree_insert(bma->cur, &i);
1872 			if (error)
1873 				goto done;
1874 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1875 				error = -EFSCORRUPTED;
1876 				goto done;
1877 			}
1878 		}
1879 
1880 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1881 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1882 				&bma->cur, 1, &tmp_rval, whichfork);
1883 			rval |= tmp_rval;
1884 			if (error)
1885 				goto done;
1886 		}
1887 
1888 		temp = PREV.br_blockcount - new->br_blockcount;
1889 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1890 			startblockval(PREV.br_startblock) -
1891 			(bma->cur ? bma->cur->bc_ino.allocated : 0));
1892 
1893 		PREV.br_startblock = nullstartblock(da_new);
1894 		PREV.br_blockcount = temp;
1895 		xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1896 		xfs_iext_next(ifp, &bma->icur);
1897 		break;
1898 
1899 	case 0:
1900 		/*
1901 		 * Filling in the middle part of a previous delayed allocation.
1902 		 * Contiguity is impossible here.
1903 		 * This case is avoided almost all the time.
1904 		 *
1905 		 * We start with a delayed allocation:
1906 		 *
1907 		 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1908 		 *  PREV @ idx
1909 		 *
1910 	         * and we are allocating:
1911 		 *                     +rrrrrrrrrrrrrrrrr+
1912 		 *			      new
1913 		 *
1914 		 * and we set it up for insertion as:
1915 		 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1916 		 *                            new
1917 		 *  PREV @ idx          LEFT              RIGHT
1918 		 *                      inserted at idx + 1
1919 		 */
1920 		old = PREV;
1921 
1922 		/* LEFT is the new middle */
1923 		LEFT = *new;
1924 
1925 		/* RIGHT is the new right */
1926 		RIGHT.br_state = PREV.br_state;
1927 		RIGHT.br_startoff = new_endoff;
1928 		RIGHT.br_blockcount =
1929 			PREV.br_startoff + PREV.br_blockcount - new_endoff;
1930 		RIGHT.br_startblock =
1931 			nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1932 					RIGHT.br_blockcount));
1933 
1934 		/* truncate PREV */
1935 		PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
1936 		PREV.br_startblock =
1937 			nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1938 					PREV.br_blockcount));
1939 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1940 
1941 		xfs_iext_next(ifp, &bma->icur);
1942 		xfs_iext_insert(bma->ip, &bma->icur, &RIGHT, state);
1943 		xfs_iext_insert(bma->ip, &bma->icur, &LEFT, state);
1944 		ifp->if_nextents++;
1945 
1946 		if (bma->cur == NULL)
1947 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1948 		else {
1949 			rval = XFS_ILOG_CORE;
1950 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1951 			if (error)
1952 				goto done;
1953 			if (XFS_IS_CORRUPT(mp, i != 0)) {
1954 				error = -EFSCORRUPTED;
1955 				goto done;
1956 			}
1957 			error = xfs_btree_insert(bma->cur, &i);
1958 			if (error)
1959 				goto done;
1960 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1961 				error = -EFSCORRUPTED;
1962 				goto done;
1963 			}
1964 		}
1965 
1966 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1967 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1968 					&bma->cur, 1, &tmp_rval, whichfork);
1969 			rval |= tmp_rval;
1970 			if (error)
1971 				goto done;
1972 		}
1973 
1974 		da_new = startblockval(PREV.br_startblock) +
1975 			 startblockval(RIGHT.br_startblock);
1976 		break;
1977 
1978 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1979 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1980 	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1981 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1982 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1983 	case BMAP_LEFT_CONTIG:
1984 	case BMAP_RIGHT_CONTIG:
1985 		/*
1986 		 * These cases are all impossible.
1987 		 */
1988 		ASSERT(0);
1989 	}
1990 
1991 	/* add reverse mapping unless caller opted out */
1992 	if (!(bma->flags & XFS_BMAPI_NORMAP))
1993 		xfs_rmap_map_extent(bma->tp, bma->ip, whichfork, new);
1994 
1995 	/* convert to a btree if necessary */
1996 	if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1997 		int	tmp_logflags;	/* partial log flag return val */
1998 
1999 		ASSERT(bma->cur == NULL);
2000 		error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2001 				&bma->cur, da_old > 0, &tmp_logflags,
2002 				whichfork);
2003 		bma->logflags |= tmp_logflags;
2004 		if (error)
2005 			goto done;
2006 	}
2007 
2008 	if (da_new != da_old)
2009 		xfs_mod_delalloc(mp, (int64_t)da_new - da_old);
2010 
2011 	if (bma->cur) {
2012 		da_new += bma->cur->bc_ino.allocated;
2013 		bma->cur->bc_ino.allocated = 0;
2014 	}
2015 
2016 	/* adjust for changes in reserved delayed indirect blocks */
2017 	if (da_new != da_old) {
2018 		ASSERT(state == 0 || da_new < da_old);
2019 		error = xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new),
2020 				false);
2021 	}
2022 
2023 	xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
2024 done:
2025 	if (whichfork != XFS_COW_FORK)
2026 		bma->logflags |= rval;
2027 	return error;
2028 #undef	LEFT
2029 #undef	RIGHT
2030 #undef	PREV
2031 }
2032 
2033 /*
2034  * Convert an unwritten allocation to a real allocation or vice versa.
2035  */
2036 int					/* error */
2037 xfs_bmap_add_extent_unwritten_real(
2038 	struct xfs_trans	*tp,
2039 	xfs_inode_t		*ip,	/* incore inode pointer */
2040 	int			whichfork,
2041 	struct xfs_iext_cursor	*icur,
2042 	xfs_btree_cur_t		**curp,	/* if *curp is null, not a btree */
2043 	xfs_bmbt_irec_t		*new,	/* new data to add to file extents */
2044 	int			*logflagsp) /* inode logging flags */
2045 {
2046 	xfs_btree_cur_t		*cur;	/* btree cursor */
2047 	int			error;	/* error return value */
2048 	int			i;	/* temp state */
2049 	struct xfs_ifork	*ifp;	/* inode fork pointer */
2050 	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
2051 	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
2052 					/* left is 0, right is 1, prev is 2 */
2053 	int			rval=0;	/* return value (logging flags) */
2054 	int			state = xfs_bmap_fork_to_state(whichfork);
2055 	struct xfs_mount	*mp = ip->i_mount;
2056 	struct xfs_bmbt_irec	old;
2057 
2058 	*logflagsp = 0;
2059 
2060 	cur = *curp;
2061 	ifp = XFS_IFORK_PTR(ip, whichfork);
2062 
2063 	ASSERT(!isnullstartblock(new->br_startblock));
2064 
2065 	XFS_STATS_INC(mp, xs_add_exlist);
2066 
2067 #define	LEFT		r[0]
2068 #define	RIGHT		r[1]
2069 #define	PREV		r[2]
2070 
2071 	/*
2072 	 * Set up a bunch of variables to make the tests simpler.
2073 	 */
2074 	error = 0;
2075 	xfs_iext_get_extent(ifp, icur, &PREV);
2076 	ASSERT(new->br_state != PREV.br_state);
2077 	new_endoff = new->br_startoff + new->br_blockcount;
2078 	ASSERT(PREV.br_startoff <= new->br_startoff);
2079 	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2080 
2081 	/*
2082 	 * Set flags determining what part of the previous oldext allocation
2083 	 * extent is being replaced by a newext allocation.
2084 	 */
2085 	if (PREV.br_startoff == new->br_startoff)
2086 		state |= BMAP_LEFT_FILLING;
2087 	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2088 		state |= BMAP_RIGHT_FILLING;
2089 
2090 	/*
2091 	 * Check and set flags if this segment has a left neighbor.
2092 	 * Don't set contiguous if the combined extent would be too large.
2093 	 */
2094 	if (xfs_iext_peek_prev_extent(ifp, icur, &LEFT)) {
2095 		state |= BMAP_LEFT_VALID;
2096 		if (isnullstartblock(LEFT.br_startblock))
2097 			state |= BMAP_LEFT_DELAY;
2098 	}
2099 
2100 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2101 	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2102 	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2103 	    LEFT.br_state == new->br_state &&
2104 	    LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2105 		state |= BMAP_LEFT_CONTIG;
2106 
2107 	/*
2108 	 * Check and set flags if this segment has a right neighbor.
2109 	 * Don't set contiguous if the combined extent would be too large.
2110 	 * Also check for all-three-contiguous being too large.
2111 	 */
2112 	if (xfs_iext_peek_next_extent(ifp, icur, &RIGHT)) {
2113 		state |= BMAP_RIGHT_VALID;
2114 		if (isnullstartblock(RIGHT.br_startblock))
2115 			state |= BMAP_RIGHT_DELAY;
2116 	}
2117 
2118 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2119 	    new_endoff == RIGHT.br_startoff &&
2120 	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2121 	    new->br_state == RIGHT.br_state &&
2122 	    new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2123 	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2124 		       BMAP_RIGHT_FILLING)) !=
2125 		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2126 		       BMAP_RIGHT_FILLING) ||
2127 	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2128 			<= MAXEXTLEN))
2129 		state |= BMAP_RIGHT_CONTIG;
2130 
2131 	/*
2132 	 * Switch out based on the FILLING and CONTIG state bits.
2133 	 */
2134 	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2135 			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2136 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2137 	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2138 		/*
2139 		 * Setting all of a previous oldext extent to newext.
2140 		 * The left and right neighbors are both contiguous with new.
2141 		 */
2142 		LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
2143 
2144 		xfs_iext_remove(ip, icur, state);
2145 		xfs_iext_remove(ip, icur, state);
2146 		xfs_iext_prev(ifp, icur);
2147 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2148 		ifp->if_nextents -= 2;
2149 		if (cur == NULL)
2150 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2151 		else {
2152 			rval = XFS_ILOG_CORE;
2153 			error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2154 			if (error)
2155 				goto done;
2156 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2157 				error = -EFSCORRUPTED;
2158 				goto done;
2159 			}
2160 			if ((error = xfs_btree_delete(cur, &i)))
2161 				goto done;
2162 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2163 				error = -EFSCORRUPTED;
2164 				goto done;
2165 			}
2166 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2167 				goto done;
2168 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2169 				error = -EFSCORRUPTED;
2170 				goto done;
2171 			}
2172 			if ((error = xfs_btree_delete(cur, &i)))
2173 				goto done;
2174 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2175 				error = -EFSCORRUPTED;
2176 				goto done;
2177 			}
2178 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2179 				goto done;
2180 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2181 				error = -EFSCORRUPTED;
2182 				goto done;
2183 			}
2184 			error = xfs_bmbt_update(cur, &LEFT);
2185 			if (error)
2186 				goto done;
2187 		}
2188 		break;
2189 
2190 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2191 		/*
2192 		 * Setting all of a previous oldext extent to newext.
2193 		 * The left neighbor is contiguous, the right is not.
2194 		 */
2195 		LEFT.br_blockcount += PREV.br_blockcount;
2196 
2197 		xfs_iext_remove(ip, icur, state);
2198 		xfs_iext_prev(ifp, icur);
2199 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2200 		ifp->if_nextents--;
2201 		if (cur == NULL)
2202 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2203 		else {
2204 			rval = XFS_ILOG_CORE;
2205 			error = xfs_bmbt_lookup_eq(cur, &PREV, &i);
2206 			if (error)
2207 				goto done;
2208 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2209 				error = -EFSCORRUPTED;
2210 				goto done;
2211 			}
2212 			if ((error = xfs_btree_delete(cur, &i)))
2213 				goto done;
2214 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2215 				error = -EFSCORRUPTED;
2216 				goto done;
2217 			}
2218 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2219 				goto done;
2220 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2221 				error = -EFSCORRUPTED;
2222 				goto done;
2223 			}
2224 			error = xfs_bmbt_update(cur, &LEFT);
2225 			if (error)
2226 				goto done;
2227 		}
2228 		break;
2229 
2230 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2231 		/*
2232 		 * Setting all of a previous oldext extent to newext.
2233 		 * The right neighbor is contiguous, the left is not.
2234 		 */
2235 		PREV.br_blockcount += RIGHT.br_blockcount;
2236 		PREV.br_state = new->br_state;
2237 
2238 		xfs_iext_next(ifp, icur);
2239 		xfs_iext_remove(ip, icur, state);
2240 		xfs_iext_prev(ifp, icur);
2241 		xfs_iext_update_extent(ip, state, icur, &PREV);
2242 		ifp->if_nextents--;
2243 
2244 		if (cur == NULL)
2245 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2246 		else {
2247 			rval = XFS_ILOG_CORE;
2248 			error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2249 			if (error)
2250 				goto done;
2251 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2252 				error = -EFSCORRUPTED;
2253 				goto done;
2254 			}
2255 			if ((error = xfs_btree_delete(cur, &i)))
2256 				goto done;
2257 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2258 				error = -EFSCORRUPTED;
2259 				goto done;
2260 			}
2261 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2262 				goto done;
2263 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2264 				error = -EFSCORRUPTED;
2265 				goto done;
2266 			}
2267 			error = xfs_bmbt_update(cur, &PREV);
2268 			if (error)
2269 				goto done;
2270 		}
2271 		break;
2272 
2273 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2274 		/*
2275 		 * Setting all of a previous oldext extent to newext.
2276 		 * Neither the left nor right neighbors are contiguous with
2277 		 * the new one.
2278 		 */
2279 		PREV.br_state = new->br_state;
2280 		xfs_iext_update_extent(ip, state, icur, &PREV);
2281 
2282 		if (cur == NULL)
2283 			rval = XFS_ILOG_DEXT;
2284 		else {
2285 			rval = 0;
2286 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2287 			if (error)
2288 				goto done;
2289 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2290 				error = -EFSCORRUPTED;
2291 				goto done;
2292 			}
2293 			error = xfs_bmbt_update(cur, &PREV);
2294 			if (error)
2295 				goto done;
2296 		}
2297 		break;
2298 
2299 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2300 		/*
2301 		 * Setting the first part of a previous oldext extent to newext.
2302 		 * The left neighbor is contiguous.
2303 		 */
2304 		LEFT.br_blockcount += new->br_blockcount;
2305 
2306 		old = PREV;
2307 		PREV.br_startoff += new->br_blockcount;
2308 		PREV.br_startblock += new->br_blockcount;
2309 		PREV.br_blockcount -= new->br_blockcount;
2310 
2311 		xfs_iext_update_extent(ip, state, icur, &PREV);
2312 		xfs_iext_prev(ifp, icur);
2313 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2314 
2315 		if (cur == NULL)
2316 			rval = XFS_ILOG_DEXT;
2317 		else {
2318 			rval = 0;
2319 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2320 			if (error)
2321 				goto done;
2322 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2323 				error = -EFSCORRUPTED;
2324 				goto done;
2325 			}
2326 			error = xfs_bmbt_update(cur, &PREV);
2327 			if (error)
2328 				goto done;
2329 			error = xfs_btree_decrement(cur, 0, &i);
2330 			if (error)
2331 				goto done;
2332 			error = xfs_bmbt_update(cur, &LEFT);
2333 			if (error)
2334 				goto done;
2335 		}
2336 		break;
2337 
2338 	case BMAP_LEFT_FILLING:
2339 		/*
2340 		 * Setting the first part of a previous oldext extent to newext.
2341 		 * The left neighbor is not contiguous.
2342 		 */
2343 		old = PREV;
2344 		PREV.br_startoff += new->br_blockcount;
2345 		PREV.br_startblock += new->br_blockcount;
2346 		PREV.br_blockcount -= new->br_blockcount;
2347 
2348 		xfs_iext_update_extent(ip, state, icur, &PREV);
2349 		xfs_iext_insert(ip, icur, new, state);
2350 		ifp->if_nextents++;
2351 
2352 		if (cur == NULL)
2353 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2354 		else {
2355 			rval = XFS_ILOG_CORE;
2356 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2357 			if (error)
2358 				goto done;
2359 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2360 				error = -EFSCORRUPTED;
2361 				goto done;
2362 			}
2363 			error = xfs_bmbt_update(cur, &PREV);
2364 			if (error)
2365 				goto done;
2366 			cur->bc_rec.b = *new;
2367 			if ((error = xfs_btree_insert(cur, &i)))
2368 				goto done;
2369 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2370 				error = -EFSCORRUPTED;
2371 				goto done;
2372 			}
2373 		}
2374 		break;
2375 
2376 	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2377 		/*
2378 		 * Setting the last part of a previous oldext extent to newext.
2379 		 * The right neighbor is contiguous with the new allocation.
2380 		 */
2381 		old = PREV;
2382 		PREV.br_blockcount -= new->br_blockcount;
2383 
2384 		RIGHT.br_startoff = new->br_startoff;
2385 		RIGHT.br_startblock = new->br_startblock;
2386 		RIGHT.br_blockcount += new->br_blockcount;
2387 
2388 		xfs_iext_update_extent(ip, state, icur, &PREV);
2389 		xfs_iext_next(ifp, icur);
2390 		xfs_iext_update_extent(ip, state, icur, &RIGHT);
2391 
2392 		if (cur == NULL)
2393 			rval = XFS_ILOG_DEXT;
2394 		else {
2395 			rval = 0;
2396 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2397 			if (error)
2398 				goto done;
2399 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2400 				error = -EFSCORRUPTED;
2401 				goto done;
2402 			}
2403 			error = xfs_bmbt_update(cur, &PREV);
2404 			if (error)
2405 				goto done;
2406 			error = xfs_btree_increment(cur, 0, &i);
2407 			if (error)
2408 				goto done;
2409 			error = xfs_bmbt_update(cur, &RIGHT);
2410 			if (error)
2411 				goto done;
2412 		}
2413 		break;
2414 
2415 	case BMAP_RIGHT_FILLING:
2416 		/*
2417 		 * Setting the last part of a previous oldext extent to newext.
2418 		 * The right neighbor is not contiguous.
2419 		 */
2420 		old = PREV;
2421 		PREV.br_blockcount -= new->br_blockcount;
2422 
2423 		xfs_iext_update_extent(ip, state, icur, &PREV);
2424 		xfs_iext_next(ifp, icur);
2425 		xfs_iext_insert(ip, icur, new, state);
2426 		ifp->if_nextents++;
2427 
2428 		if (cur == NULL)
2429 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2430 		else {
2431 			rval = XFS_ILOG_CORE;
2432 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2433 			if (error)
2434 				goto done;
2435 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2436 				error = -EFSCORRUPTED;
2437 				goto done;
2438 			}
2439 			error = xfs_bmbt_update(cur, &PREV);
2440 			if (error)
2441 				goto done;
2442 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2443 			if (error)
2444 				goto done;
2445 			if (XFS_IS_CORRUPT(mp, i != 0)) {
2446 				error = -EFSCORRUPTED;
2447 				goto done;
2448 			}
2449 			if ((error = xfs_btree_insert(cur, &i)))
2450 				goto done;
2451 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2452 				error = -EFSCORRUPTED;
2453 				goto done;
2454 			}
2455 		}
2456 		break;
2457 
2458 	case 0:
2459 		/*
2460 		 * Setting the middle part of a previous oldext extent to
2461 		 * newext.  Contiguity is impossible here.
2462 		 * One extent becomes three extents.
2463 		 */
2464 		old = PREV;
2465 		PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
2466 
2467 		r[0] = *new;
2468 		r[1].br_startoff = new_endoff;
2469 		r[1].br_blockcount =
2470 			old.br_startoff + old.br_blockcount - new_endoff;
2471 		r[1].br_startblock = new->br_startblock + new->br_blockcount;
2472 		r[1].br_state = PREV.br_state;
2473 
2474 		xfs_iext_update_extent(ip, state, icur, &PREV);
2475 		xfs_iext_next(ifp, icur);
2476 		xfs_iext_insert(ip, icur, &r[1], state);
2477 		xfs_iext_insert(ip, icur, &r[0], state);
2478 		ifp->if_nextents += 2;
2479 
2480 		if (cur == NULL)
2481 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2482 		else {
2483 			rval = XFS_ILOG_CORE;
2484 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2485 			if (error)
2486 				goto done;
2487 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2488 				error = -EFSCORRUPTED;
2489 				goto done;
2490 			}
2491 			/* new right extent - oldext */
2492 			error = xfs_bmbt_update(cur, &r[1]);
2493 			if (error)
2494 				goto done;
2495 			/* new left extent - oldext */
2496 			cur->bc_rec.b = PREV;
2497 			if ((error = xfs_btree_insert(cur, &i)))
2498 				goto done;
2499 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2500 				error = -EFSCORRUPTED;
2501 				goto done;
2502 			}
2503 			/*
2504 			 * Reset the cursor to the position of the new extent
2505 			 * we are about to insert as we can't trust it after
2506 			 * the previous insert.
2507 			 */
2508 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2509 			if (error)
2510 				goto done;
2511 			if (XFS_IS_CORRUPT(mp, i != 0)) {
2512 				error = -EFSCORRUPTED;
2513 				goto done;
2514 			}
2515 			/* new middle extent - newext */
2516 			if ((error = xfs_btree_insert(cur, &i)))
2517 				goto done;
2518 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2519 				error = -EFSCORRUPTED;
2520 				goto done;
2521 			}
2522 		}
2523 		break;
2524 
2525 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2526 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2527 	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2528 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2529 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2530 	case BMAP_LEFT_CONTIG:
2531 	case BMAP_RIGHT_CONTIG:
2532 		/*
2533 		 * These cases are all impossible.
2534 		 */
2535 		ASSERT(0);
2536 	}
2537 
2538 	/* update reverse mappings */
2539 	xfs_rmap_convert_extent(mp, tp, ip, whichfork, new);
2540 
2541 	/* convert to a btree if necessary */
2542 	if (xfs_bmap_needs_btree(ip, whichfork)) {
2543 		int	tmp_logflags;	/* partial log flag return val */
2544 
2545 		ASSERT(cur == NULL);
2546 		error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
2547 				&tmp_logflags, whichfork);
2548 		*logflagsp |= tmp_logflags;
2549 		if (error)
2550 			goto done;
2551 	}
2552 
2553 	/* clear out the allocated field, done with it now in any case. */
2554 	if (cur) {
2555 		cur->bc_ino.allocated = 0;
2556 		*curp = cur;
2557 	}
2558 
2559 	xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
2560 done:
2561 	*logflagsp |= rval;
2562 	return error;
2563 #undef	LEFT
2564 #undef	RIGHT
2565 #undef	PREV
2566 }
2567 
2568 /*
2569  * Convert a hole to a delayed allocation.
2570  */
2571 STATIC void
2572 xfs_bmap_add_extent_hole_delay(
2573 	xfs_inode_t		*ip,	/* incore inode pointer */
2574 	int			whichfork,
2575 	struct xfs_iext_cursor	*icur,
2576 	xfs_bmbt_irec_t		*new)	/* new data to add to file extents */
2577 {
2578 	struct xfs_ifork	*ifp;	/* inode fork pointer */
2579 	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2580 	xfs_filblks_t		newlen=0;	/* new indirect size */
2581 	xfs_filblks_t		oldlen=0;	/* old indirect size */
2582 	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2583 	int			state = xfs_bmap_fork_to_state(whichfork);
2584 	xfs_filblks_t		temp;	 /* temp for indirect calculations */
2585 
2586 	ifp = XFS_IFORK_PTR(ip, whichfork);
2587 	ASSERT(isnullstartblock(new->br_startblock));
2588 
2589 	/*
2590 	 * Check and set flags if this segment has a left neighbor
2591 	 */
2592 	if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2593 		state |= BMAP_LEFT_VALID;
2594 		if (isnullstartblock(left.br_startblock))
2595 			state |= BMAP_LEFT_DELAY;
2596 	}
2597 
2598 	/*
2599 	 * Check and set flags if the current (right) segment exists.
2600 	 * If it doesn't exist, we're converting the hole at end-of-file.
2601 	 */
2602 	if (xfs_iext_get_extent(ifp, icur, &right)) {
2603 		state |= BMAP_RIGHT_VALID;
2604 		if (isnullstartblock(right.br_startblock))
2605 			state |= BMAP_RIGHT_DELAY;
2606 	}
2607 
2608 	/*
2609 	 * Set contiguity flags on the left and right neighbors.
2610 	 * Don't let extents get too large, even if the pieces are contiguous.
2611 	 */
2612 	if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2613 	    left.br_startoff + left.br_blockcount == new->br_startoff &&
2614 	    left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2615 		state |= BMAP_LEFT_CONTIG;
2616 
2617 	if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2618 	    new->br_startoff + new->br_blockcount == right.br_startoff &&
2619 	    new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2620 	    (!(state & BMAP_LEFT_CONTIG) ||
2621 	     (left.br_blockcount + new->br_blockcount +
2622 	      right.br_blockcount <= MAXEXTLEN)))
2623 		state |= BMAP_RIGHT_CONTIG;
2624 
2625 	/*
2626 	 * Switch out based on the contiguity flags.
2627 	 */
2628 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2629 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2630 		/*
2631 		 * New allocation is contiguous with delayed allocations
2632 		 * on the left and on the right.
2633 		 * Merge all three into a single extent record.
2634 		 */
2635 		temp = left.br_blockcount + new->br_blockcount +
2636 			right.br_blockcount;
2637 
2638 		oldlen = startblockval(left.br_startblock) +
2639 			startblockval(new->br_startblock) +
2640 			startblockval(right.br_startblock);
2641 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2642 					 oldlen);
2643 		left.br_startblock = nullstartblock(newlen);
2644 		left.br_blockcount = temp;
2645 
2646 		xfs_iext_remove(ip, icur, state);
2647 		xfs_iext_prev(ifp, icur);
2648 		xfs_iext_update_extent(ip, state, icur, &left);
2649 		break;
2650 
2651 	case BMAP_LEFT_CONTIG:
2652 		/*
2653 		 * New allocation is contiguous with a delayed allocation
2654 		 * on the left.
2655 		 * Merge the new allocation with the left neighbor.
2656 		 */
2657 		temp = left.br_blockcount + new->br_blockcount;
2658 
2659 		oldlen = startblockval(left.br_startblock) +
2660 			startblockval(new->br_startblock);
2661 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2662 					 oldlen);
2663 		left.br_blockcount = temp;
2664 		left.br_startblock = nullstartblock(newlen);
2665 
2666 		xfs_iext_prev(ifp, icur);
2667 		xfs_iext_update_extent(ip, state, icur, &left);
2668 		break;
2669 
2670 	case BMAP_RIGHT_CONTIG:
2671 		/*
2672 		 * New allocation is contiguous with a delayed allocation
2673 		 * on the right.
2674 		 * Merge the new allocation with the right neighbor.
2675 		 */
2676 		temp = new->br_blockcount + right.br_blockcount;
2677 		oldlen = startblockval(new->br_startblock) +
2678 			startblockval(right.br_startblock);
2679 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2680 					 oldlen);
2681 		right.br_startoff = new->br_startoff;
2682 		right.br_startblock = nullstartblock(newlen);
2683 		right.br_blockcount = temp;
2684 		xfs_iext_update_extent(ip, state, icur, &right);
2685 		break;
2686 
2687 	case 0:
2688 		/*
2689 		 * New allocation is not contiguous with another
2690 		 * delayed allocation.
2691 		 * Insert a new entry.
2692 		 */
2693 		oldlen = newlen = 0;
2694 		xfs_iext_insert(ip, icur, new, state);
2695 		break;
2696 	}
2697 	if (oldlen != newlen) {
2698 		ASSERT(oldlen > newlen);
2699 		xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2700 				 false);
2701 		/*
2702 		 * Nothing to do for disk quota accounting here.
2703 		 */
2704 		xfs_mod_delalloc(ip->i_mount, (int64_t)newlen - oldlen);
2705 	}
2706 }
2707 
2708 /*
2709  * Convert a hole to a real allocation.
2710  */
2711 STATIC int				/* error */
2712 xfs_bmap_add_extent_hole_real(
2713 	struct xfs_trans	*tp,
2714 	struct xfs_inode	*ip,
2715 	int			whichfork,
2716 	struct xfs_iext_cursor	*icur,
2717 	struct xfs_btree_cur	**curp,
2718 	struct xfs_bmbt_irec	*new,
2719 	int			*logflagsp,
2720 	int			flags)
2721 {
2722 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
2723 	struct xfs_mount	*mp = ip->i_mount;
2724 	struct xfs_btree_cur	*cur = *curp;
2725 	int			error;	/* error return value */
2726 	int			i;	/* temp state */
2727 	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2728 	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2729 	int			rval=0;	/* return value (logging flags) */
2730 	int			state = xfs_bmap_fork_to_state(whichfork);
2731 	struct xfs_bmbt_irec	old;
2732 
2733 	ASSERT(!isnullstartblock(new->br_startblock));
2734 	ASSERT(!cur || !(cur->bc_ino.flags & XFS_BTCUR_BMBT_WASDEL));
2735 
2736 	XFS_STATS_INC(mp, xs_add_exlist);
2737 
2738 	/*
2739 	 * Check and set flags if this segment has a left neighbor.
2740 	 */
2741 	if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2742 		state |= BMAP_LEFT_VALID;
2743 		if (isnullstartblock(left.br_startblock))
2744 			state |= BMAP_LEFT_DELAY;
2745 	}
2746 
2747 	/*
2748 	 * Check and set flags if this segment has a current value.
2749 	 * Not true if we're inserting into the "hole" at eof.
2750 	 */
2751 	if (xfs_iext_get_extent(ifp, icur, &right)) {
2752 		state |= BMAP_RIGHT_VALID;
2753 		if (isnullstartblock(right.br_startblock))
2754 			state |= BMAP_RIGHT_DELAY;
2755 	}
2756 
2757 	/*
2758 	 * We're inserting a real allocation between "left" and "right".
2759 	 * Set the contiguity flags.  Don't let extents get too large.
2760 	 */
2761 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2762 	    left.br_startoff + left.br_blockcount == new->br_startoff &&
2763 	    left.br_startblock + left.br_blockcount == new->br_startblock &&
2764 	    left.br_state == new->br_state &&
2765 	    left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2766 		state |= BMAP_LEFT_CONTIG;
2767 
2768 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2769 	    new->br_startoff + new->br_blockcount == right.br_startoff &&
2770 	    new->br_startblock + new->br_blockcount == right.br_startblock &&
2771 	    new->br_state == right.br_state &&
2772 	    new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2773 	    (!(state & BMAP_LEFT_CONTIG) ||
2774 	     left.br_blockcount + new->br_blockcount +
2775 	     right.br_blockcount <= MAXEXTLEN))
2776 		state |= BMAP_RIGHT_CONTIG;
2777 
2778 	error = 0;
2779 	/*
2780 	 * Select which case we're in here, and implement it.
2781 	 */
2782 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2783 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2784 		/*
2785 		 * New allocation is contiguous with real allocations on the
2786 		 * left and on the right.
2787 		 * Merge all three into a single extent record.
2788 		 */
2789 		left.br_blockcount += new->br_blockcount + right.br_blockcount;
2790 
2791 		xfs_iext_remove(ip, icur, state);
2792 		xfs_iext_prev(ifp, icur);
2793 		xfs_iext_update_extent(ip, state, icur, &left);
2794 		ifp->if_nextents--;
2795 
2796 		if (cur == NULL) {
2797 			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2798 		} else {
2799 			rval = XFS_ILOG_CORE;
2800 			error = xfs_bmbt_lookup_eq(cur, &right, &i);
2801 			if (error)
2802 				goto done;
2803 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2804 				error = -EFSCORRUPTED;
2805 				goto done;
2806 			}
2807 			error = xfs_btree_delete(cur, &i);
2808 			if (error)
2809 				goto done;
2810 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2811 				error = -EFSCORRUPTED;
2812 				goto done;
2813 			}
2814 			error = xfs_btree_decrement(cur, 0, &i);
2815 			if (error)
2816 				goto done;
2817 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2818 				error = -EFSCORRUPTED;
2819 				goto done;
2820 			}
2821 			error = xfs_bmbt_update(cur, &left);
2822 			if (error)
2823 				goto done;
2824 		}
2825 		break;
2826 
2827 	case BMAP_LEFT_CONTIG:
2828 		/*
2829 		 * New allocation is contiguous with a real allocation
2830 		 * on the left.
2831 		 * Merge the new allocation with the left neighbor.
2832 		 */
2833 		old = left;
2834 		left.br_blockcount += new->br_blockcount;
2835 
2836 		xfs_iext_prev(ifp, icur);
2837 		xfs_iext_update_extent(ip, state, icur, &left);
2838 
2839 		if (cur == NULL) {
2840 			rval = xfs_ilog_fext(whichfork);
2841 		} else {
2842 			rval = 0;
2843 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2844 			if (error)
2845 				goto done;
2846 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2847 				error = -EFSCORRUPTED;
2848 				goto done;
2849 			}
2850 			error = xfs_bmbt_update(cur, &left);
2851 			if (error)
2852 				goto done;
2853 		}
2854 		break;
2855 
2856 	case BMAP_RIGHT_CONTIG:
2857 		/*
2858 		 * New allocation is contiguous with a real allocation
2859 		 * on the right.
2860 		 * Merge the new allocation with the right neighbor.
2861 		 */
2862 		old = right;
2863 
2864 		right.br_startoff = new->br_startoff;
2865 		right.br_startblock = new->br_startblock;
2866 		right.br_blockcount += new->br_blockcount;
2867 		xfs_iext_update_extent(ip, state, icur, &right);
2868 
2869 		if (cur == NULL) {
2870 			rval = xfs_ilog_fext(whichfork);
2871 		} else {
2872 			rval = 0;
2873 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2874 			if (error)
2875 				goto done;
2876 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2877 				error = -EFSCORRUPTED;
2878 				goto done;
2879 			}
2880 			error = xfs_bmbt_update(cur, &right);
2881 			if (error)
2882 				goto done;
2883 		}
2884 		break;
2885 
2886 	case 0:
2887 		/*
2888 		 * New allocation is not contiguous with another
2889 		 * real allocation.
2890 		 * Insert a new entry.
2891 		 */
2892 		xfs_iext_insert(ip, icur, new, state);
2893 		ifp->if_nextents++;
2894 
2895 		if (cur == NULL) {
2896 			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2897 		} else {
2898 			rval = XFS_ILOG_CORE;
2899 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2900 			if (error)
2901 				goto done;
2902 			if (XFS_IS_CORRUPT(mp, i != 0)) {
2903 				error = -EFSCORRUPTED;
2904 				goto done;
2905 			}
2906 			error = xfs_btree_insert(cur, &i);
2907 			if (error)
2908 				goto done;
2909 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2910 				error = -EFSCORRUPTED;
2911 				goto done;
2912 			}
2913 		}
2914 		break;
2915 	}
2916 
2917 	/* add reverse mapping unless caller opted out */
2918 	if (!(flags & XFS_BMAPI_NORMAP))
2919 		xfs_rmap_map_extent(tp, ip, whichfork, new);
2920 
2921 	/* convert to a btree if necessary */
2922 	if (xfs_bmap_needs_btree(ip, whichfork)) {
2923 		int	tmp_logflags;	/* partial log flag return val */
2924 
2925 		ASSERT(cur == NULL);
2926 		error = xfs_bmap_extents_to_btree(tp, ip, curp, 0,
2927 				&tmp_logflags, whichfork);
2928 		*logflagsp |= tmp_logflags;
2929 		cur = *curp;
2930 		if (error)
2931 			goto done;
2932 	}
2933 
2934 	/* clear out the allocated field, done with it now in any case. */
2935 	if (cur)
2936 		cur->bc_ino.allocated = 0;
2937 
2938 	xfs_bmap_check_leaf_extents(cur, ip, whichfork);
2939 done:
2940 	*logflagsp |= rval;
2941 	return error;
2942 }
2943 
2944 /*
2945  * Functions used in the extent read, allocate and remove paths
2946  */
2947 
2948 /*
2949  * Adjust the size of the new extent based on di_extsize and rt extsize.
2950  */
2951 int
2952 xfs_bmap_extsize_align(
2953 	xfs_mount_t	*mp,
2954 	xfs_bmbt_irec_t	*gotp,		/* next extent pointer */
2955 	xfs_bmbt_irec_t	*prevp,		/* previous extent pointer */
2956 	xfs_extlen_t	extsz,		/* align to this extent size */
2957 	int		rt,		/* is this a realtime inode? */
2958 	int		eof,		/* is extent at end-of-file? */
2959 	int		delay,		/* creating delalloc extent? */
2960 	int		convert,	/* overwriting unwritten extent? */
2961 	xfs_fileoff_t	*offp,		/* in/out: aligned offset */
2962 	xfs_extlen_t	*lenp)		/* in/out: aligned length */
2963 {
2964 	xfs_fileoff_t	orig_off;	/* original offset */
2965 	xfs_extlen_t	orig_alen;	/* original length */
2966 	xfs_fileoff_t	orig_end;	/* original off+len */
2967 	xfs_fileoff_t	nexto;		/* next file offset */
2968 	xfs_fileoff_t	prevo;		/* previous file offset */
2969 	xfs_fileoff_t	align_off;	/* temp for offset */
2970 	xfs_extlen_t	align_alen;	/* temp for length */
2971 	xfs_extlen_t	temp;		/* temp for calculations */
2972 
2973 	if (convert)
2974 		return 0;
2975 
2976 	orig_off = align_off = *offp;
2977 	orig_alen = align_alen = *lenp;
2978 	orig_end = orig_off + orig_alen;
2979 
2980 	/*
2981 	 * If this request overlaps an existing extent, then don't
2982 	 * attempt to perform any additional alignment.
2983 	 */
2984 	if (!delay && !eof &&
2985 	    (orig_off >= gotp->br_startoff) &&
2986 	    (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2987 		return 0;
2988 	}
2989 
2990 	/*
2991 	 * If the file offset is unaligned vs. the extent size
2992 	 * we need to align it.  This will be possible unless
2993 	 * the file was previously written with a kernel that didn't
2994 	 * perform this alignment, or if a truncate shot us in the
2995 	 * foot.
2996 	 */
2997 	div_u64_rem(orig_off, extsz, &temp);
2998 	if (temp) {
2999 		align_alen += temp;
3000 		align_off -= temp;
3001 	}
3002 
3003 	/* Same adjustment for the end of the requested area. */
3004 	temp = (align_alen % extsz);
3005 	if (temp)
3006 		align_alen += extsz - temp;
3007 
3008 	/*
3009 	 * For large extent hint sizes, the aligned extent might be larger than
3010 	 * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
3011 	 * the length back under MAXEXTLEN. The outer allocation loops handle
3012 	 * short allocation just fine, so it is safe to do this. We only want to
3013 	 * do it when we are forced to, though, because it means more allocation
3014 	 * operations are required.
3015 	 */
3016 	while (align_alen > MAXEXTLEN)
3017 		align_alen -= extsz;
3018 	ASSERT(align_alen <= MAXEXTLEN);
3019 
3020 	/*
3021 	 * If the previous block overlaps with this proposed allocation
3022 	 * then move the start forward without adjusting the length.
3023 	 */
3024 	if (prevp->br_startoff != NULLFILEOFF) {
3025 		if (prevp->br_startblock == HOLESTARTBLOCK)
3026 			prevo = prevp->br_startoff;
3027 		else
3028 			prevo = prevp->br_startoff + prevp->br_blockcount;
3029 	} else
3030 		prevo = 0;
3031 	if (align_off != orig_off && align_off < prevo)
3032 		align_off = prevo;
3033 	/*
3034 	 * If the next block overlaps with this proposed allocation
3035 	 * then move the start back without adjusting the length,
3036 	 * but not before offset 0.
3037 	 * This may of course make the start overlap previous block,
3038 	 * and if we hit the offset 0 limit then the next block
3039 	 * can still overlap too.
3040 	 */
3041 	if (!eof && gotp->br_startoff != NULLFILEOFF) {
3042 		if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3043 		    (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3044 			nexto = gotp->br_startoff + gotp->br_blockcount;
3045 		else
3046 			nexto = gotp->br_startoff;
3047 	} else
3048 		nexto = NULLFILEOFF;
3049 	if (!eof &&
3050 	    align_off + align_alen != orig_end &&
3051 	    align_off + align_alen > nexto)
3052 		align_off = nexto > align_alen ? nexto - align_alen : 0;
3053 	/*
3054 	 * If we're now overlapping the next or previous extent that
3055 	 * means we can't fit an extsz piece in this hole.  Just move
3056 	 * the start forward to the first valid spot and set
3057 	 * the length so we hit the end.
3058 	 */
3059 	if (align_off != orig_off && align_off < prevo)
3060 		align_off = prevo;
3061 	if (align_off + align_alen != orig_end &&
3062 	    align_off + align_alen > nexto &&
3063 	    nexto != NULLFILEOFF) {
3064 		ASSERT(nexto > prevo);
3065 		align_alen = nexto - align_off;
3066 	}
3067 
3068 	/*
3069 	 * If realtime, and the result isn't a multiple of the realtime
3070 	 * extent size we need to remove blocks until it is.
3071 	 */
3072 	if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3073 		/*
3074 		 * We're not covering the original request, or
3075 		 * we won't be able to once we fix the length.
3076 		 */
3077 		if (orig_off < align_off ||
3078 		    orig_end > align_off + align_alen ||
3079 		    align_alen - temp < orig_alen)
3080 			return -EINVAL;
3081 		/*
3082 		 * Try to fix it by moving the start up.
3083 		 */
3084 		if (align_off + temp <= orig_off) {
3085 			align_alen -= temp;
3086 			align_off += temp;
3087 		}
3088 		/*
3089 		 * Try to fix it by moving the end in.
3090 		 */
3091 		else if (align_off + align_alen - temp >= orig_end)
3092 			align_alen -= temp;
3093 		/*
3094 		 * Set the start to the minimum then trim the length.
3095 		 */
3096 		else {
3097 			align_alen -= orig_off - align_off;
3098 			align_off = orig_off;
3099 			align_alen -= align_alen % mp->m_sb.sb_rextsize;
3100 		}
3101 		/*
3102 		 * Result doesn't cover the request, fail it.
3103 		 */
3104 		if (orig_off < align_off || orig_end > align_off + align_alen)
3105 			return -EINVAL;
3106 	} else {
3107 		ASSERT(orig_off >= align_off);
3108 		/* see MAXEXTLEN handling above */
3109 		ASSERT(orig_end <= align_off + align_alen ||
3110 		       align_alen + extsz > MAXEXTLEN);
3111 	}
3112 
3113 #ifdef DEBUG
3114 	if (!eof && gotp->br_startoff != NULLFILEOFF)
3115 		ASSERT(align_off + align_alen <= gotp->br_startoff);
3116 	if (prevp->br_startoff != NULLFILEOFF)
3117 		ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3118 #endif
3119 
3120 	*lenp = align_alen;
3121 	*offp = align_off;
3122 	return 0;
3123 }
3124 
3125 #define XFS_ALLOC_GAP_UNITS	4
3126 
3127 void
3128 xfs_bmap_adjacent(
3129 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3130 {
3131 	xfs_fsblock_t	adjust;		/* adjustment to block numbers */
3132 	xfs_agnumber_t	fb_agno;	/* ag number of ap->firstblock */
3133 	xfs_mount_t	*mp;		/* mount point structure */
3134 	int		nullfb;		/* true if ap->firstblock isn't set */
3135 	int		rt;		/* true if inode is realtime */
3136 
3137 #define	ISVALID(x,y)	\
3138 	(rt ? \
3139 		(x) < mp->m_sb.sb_rblocks : \
3140 		XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3141 		XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3142 		XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3143 
3144 	mp = ap->ip->i_mount;
3145 	nullfb = ap->tp->t_firstblock == NULLFSBLOCK;
3146 	rt = XFS_IS_REALTIME_INODE(ap->ip) &&
3147 		(ap->datatype & XFS_ALLOC_USERDATA);
3148 	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp,
3149 							ap->tp->t_firstblock);
3150 	/*
3151 	 * If allocating at eof, and there's a previous real block,
3152 	 * try to use its last block as our starting point.
3153 	 */
3154 	if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3155 	    !isnullstartblock(ap->prev.br_startblock) &&
3156 	    ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3157 		    ap->prev.br_startblock)) {
3158 		ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3159 		/*
3160 		 * Adjust for the gap between prevp and us.
3161 		 */
3162 		adjust = ap->offset -
3163 			(ap->prev.br_startoff + ap->prev.br_blockcount);
3164 		if (adjust &&
3165 		    ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3166 			ap->blkno += adjust;
3167 	}
3168 	/*
3169 	 * If not at eof, then compare the two neighbor blocks.
3170 	 * Figure out whether either one gives us a good starting point,
3171 	 * and pick the better one.
3172 	 */
3173 	else if (!ap->eof) {
3174 		xfs_fsblock_t	gotbno;		/* right side block number */
3175 		xfs_fsblock_t	gotdiff=0;	/* right side difference */
3176 		xfs_fsblock_t	prevbno;	/* left side block number */
3177 		xfs_fsblock_t	prevdiff=0;	/* left side difference */
3178 
3179 		/*
3180 		 * If there's a previous (left) block, select a requested
3181 		 * start block based on it.
3182 		 */
3183 		if (ap->prev.br_startoff != NULLFILEOFF &&
3184 		    !isnullstartblock(ap->prev.br_startblock) &&
3185 		    (prevbno = ap->prev.br_startblock +
3186 			       ap->prev.br_blockcount) &&
3187 		    ISVALID(prevbno, ap->prev.br_startblock)) {
3188 			/*
3189 			 * Calculate gap to end of previous block.
3190 			 */
3191 			adjust = prevdiff = ap->offset -
3192 				(ap->prev.br_startoff +
3193 				 ap->prev.br_blockcount);
3194 			/*
3195 			 * Figure the startblock based on the previous block's
3196 			 * end and the gap size.
3197 			 * Heuristic!
3198 			 * If the gap is large relative to the piece we're
3199 			 * allocating, or using it gives us an invalid block
3200 			 * number, then just use the end of the previous block.
3201 			 */
3202 			if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3203 			    ISVALID(prevbno + prevdiff,
3204 				    ap->prev.br_startblock))
3205 				prevbno += adjust;
3206 			else
3207 				prevdiff += adjust;
3208 			/*
3209 			 * If the firstblock forbids it, can't use it,
3210 			 * must use default.
3211 			 */
3212 			if (!rt && !nullfb &&
3213 			    XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3214 				prevbno = NULLFSBLOCK;
3215 		}
3216 		/*
3217 		 * No previous block or can't follow it, just default.
3218 		 */
3219 		else
3220 			prevbno = NULLFSBLOCK;
3221 		/*
3222 		 * If there's a following (right) block, select a requested
3223 		 * start block based on it.
3224 		 */
3225 		if (!isnullstartblock(ap->got.br_startblock)) {
3226 			/*
3227 			 * Calculate gap to start of next block.
3228 			 */
3229 			adjust = gotdiff = ap->got.br_startoff - ap->offset;
3230 			/*
3231 			 * Figure the startblock based on the next block's
3232 			 * start and the gap size.
3233 			 */
3234 			gotbno = ap->got.br_startblock;
3235 			/*
3236 			 * Heuristic!
3237 			 * If the gap is large relative to the piece we're
3238 			 * allocating, or using it gives us an invalid block
3239 			 * number, then just use the start of the next block
3240 			 * offset by our length.
3241 			 */
3242 			if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3243 			    ISVALID(gotbno - gotdiff, gotbno))
3244 				gotbno -= adjust;
3245 			else if (ISVALID(gotbno - ap->length, gotbno)) {
3246 				gotbno -= ap->length;
3247 				gotdiff += adjust - ap->length;
3248 			} else
3249 				gotdiff += adjust;
3250 			/*
3251 			 * If the firstblock forbids it, can't use it,
3252 			 * must use default.
3253 			 */
3254 			if (!rt && !nullfb &&
3255 			    XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3256 				gotbno = NULLFSBLOCK;
3257 		}
3258 		/*
3259 		 * No next block, just default.
3260 		 */
3261 		else
3262 			gotbno = NULLFSBLOCK;
3263 		/*
3264 		 * If both valid, pick the better one, else the only good
3265 		 * one, else ap->blkno is already set (to 0 or the inode block).
3266 		 */
3267 		if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3268 			ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3269 		else if (prevbno != NULLFSBLOCK)
3270 			ap->blkno = prevbno;
3271 		else if (gotbno != NULLFSBLOCK)
3272 			ap->blkno = gotbno;
3273 	}
3274 #undef ISVALID
3275 }
3276 
3277 static int
3278 xfs_bmap_longest_free_extent(
3279 	struct xfs_trans	*tp,
3280 	xfs_agnumber_t		ag,
3281 	xfs_extlen_t		*blen,
3282 	int			*notinit)
3283 {
3284 	struct xfs_mount	*mp = tp->t_mountp;
3285 	struct xfs_perag	*pag;
3286 	xfs_extlen_t		longest;
3287 	int			error = 0;
3288 
3289 	pag = xfs_perag_get(mp, ag);
3290 	if (!pag->pagf_init) {
3291 		error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3292 		if (error) {
3293 			/* Couldn't lock the AGF, so skip this AG. */
3294 			if (error == -EAGAIN) {
3295 				*notinit = 1;
3296 				error = 0;
3297 			}
3298 			goto out;
3299 		}
3300 	}
3301 
3302 	longest = xfs_alloc_longest_free_extent(pag,
3303 				xfs_alloc_min_freelist(mp, pag),
3304 				xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
3305 	if (*blen < longest)
3306 		*blen = longest;
3307 
3308 out:
3309 	xfs_perag_put(pag);
3310 	return error;
3311 }
3312 
3313 static void
3314 xfs_bmap_select_minlen(
3315 	struct xfs_bmalloca	*ap,
3316 	struct xfs_alloc_arg	*args,
3317 	xfs_extlen_t		*blen,
3318 	int			notinit)
3319 {
3320 	if (notinit || *blen < ap->minlen) {
3321 		/*
3322 		 * Since we did a BUF_TRYLOCK above, it is possible that
3323 		 * there is space for this request.
3324 		 */
3325 		args->minlen = ap->minlen;
3326 	} else if (*blen < args->maxlen) {
3327 		/*
3328 		 * If the best seen length is less than the request length,
3329 		 * use the best as the minimum.
3330 		 */
3331 		args->minlen = *blen;
3332 	} else {
3333 		/*
3334 		 * Otherwise we've seen an extent as big as maxlen, use that
3335 		 * as the minimum.
3336 		 */
3337 		args->minlen = args->maxlen;
3338 	}
3339 }
3340 
3341 STATIC int
3342 xfs_bmap_btalloc_nullfb(
3343 	struct xfs_bmalloca	*ap,
3344 	struct xfs_alloc_arg	*args,
3345 	xfs_extlen_t		*blen)
3346 {
3347 	struct xfs_mount	*mp = ap->ip->i_mount;
3348 	xfs_agnumber_t		ag, startag;
3349 	int			notinit = 0;
3350 	int			error;
3351 
3352 	args->type = XFS_ALLOCTYPE_START_BNO;
3353 	args->total = ap->total;
3354 
3355 	startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3356 	if (startag == NULLAGNUMBER)
3357 		startag = ag = 0;
3358 
3359 	while (*blen < args->maxlen) {
3360 		error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3361 						     &notinit);
3362 		if (error)
3363 			return error;
3364 
3365 		if (++ag == mp->m_sb.sb_agcount)
3366 			ag = 0;
3367 		if (ag == startag)
3368 			break;
3369 	}
3370 
3371 	xfs_bmap_select_minlen(ap, args, blen, notinit);
3372 	return 0;
3373 }
3374 
3375 STATIC int
3376 xfs_bmap_btalloc_filestreams(
3377 	struct xfs_bmalloca	*ap,
3378 	struct xfs_alloc_arg	*args,
3379 	xfs_extlen_t		*blen)
3380 {
3381 	struct xfs_mount	*mp = ap->ip->i_mount;
3382 	xfs_agnumber_t		ag;
3383 	int			notinit = 0;
3384 	int			error;
3385 
3386 	args->type = XFS_ALLOCTYPE_NEAR_BNO;
3387 	args->total = ap->total;
3388 
3389 	ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3390 	if (ag == NULLAGNUMBER)
3391 		ag = 0;
3392 
3393 	error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3394 	if (error)
3395 		return error;
3396 
3397 	if (*blen < args->maxlen) {
3398 		error = xfs_filestream_new_ag(ap, &ag);
3399 		if (error)
3400 			return error;
3401 
3402 		error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3403 						     &notinit);
3404 		if (error)
3405 			return error;
3406 
3407 	}
3408 
3409 	xfs_bmap_select_minlen(ap, args, blen, notinit);
3410 
3411 	/*
3412 	 * Set the failure fallback case to look in the selected AG as stream
3413 	 * may have moved.
3414 	 */
3415 	ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3416 	return 0;
3417 }
3418 
3419 /* Update all inode and quota accounting for the allocation we just did. */
3420 static void
3421 xfs_bmap_btalloc_accounting(
3422 	struct xfs_bmalloca	*ap,
3423 	struct xfs_alloc_arg	*args)
3424 {
3425 	if (ap->flags & XFS_BMAPI_COWFORK) {
3426 		/*
3427 		 * COW fork blocks are in-core only and thus are treated as
3428 		 * in-core quota reservation (like delalloc blocks) even when
3429 		 * converted to real blocks. The quota reservation is not
3430 		 * accounted to disk until blocks are remapped to the data
3431 		 * fork. So if these blocks were previously delalloc, we
3432 		 * already have quota reservation and there's nothing to do
3433 		 * yet.
3434 		 */
3435 		if (ap->wasdel) {
3436 			xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len);
3437 			return;
3438 		}
3439 
3440 		/*
3441 		 * Otherwise, we've allocated blocks in a hole. The transaction
3442 		 * has acquired in-core quota reservation for this extent.
3443 		 * Rather than account these as real blocks, however, we reduce
3444 		 * the transaction quota reservation based on the allocation.
3445 		 * This essentially transfers the transaction quota reservation
3446 		 * to that of a delalloc extent.
3447 		 */
3448 		ap->ip->i_delayed_blks += args->len;
3449 		xfs_trans_mod_dquot_byino(ap->tp, ap->ip, XFS_TRANS_DQ_RES_BLKS,
3450 				-(long)args->len);
3451 		return;
3452 	}
3453 
3454 	/* data/attr fork only */
3455 	ap->ip->i_d.di_nblocks += args->len;
3456 	xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3457 	if (ap->wasdel) {
3458 		ap->ip->i_delayed_blks -= args->len;
3459 		xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len);
3460 	}
3461 	xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3462 		ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT : XFS_TRANS_DQ_BCOUNT,
3463 		args->len);
3464 }
3465 
3466 STATIC int
3467 xfs_bmap_btalloc(
3468 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3469 {
3470 	xfs_mount_t	*mp;		/* mount point structure */
3471 	xfs_alloctype_t	atype = 0;	/* type for allocation routines */
3472 	xfs_extlen_t	align = 0;	/* minimum allocation alignment */
3473 	xfs_agnumber_t	fb_agno;	/* ag number of ap->firstblock */
3474 	xfs_agnumber_t	ag;
3475 	xfs_alloc_arg_t	args;
3476 	xfs_fileoff_t	orig_offset;
3477 	xfs_extlen_t	orig_length;
3478 	xfs_extlen_t	blen;
3479 	xfs_extlen_t	nextminlen = 0;
3480 	int		nullfb;		/* true if ap->firstblock isn't set */
3481 	int		isaligned;
3482 	int		tryagain;
3483 	int		error;
3484 	int		stripe_align;
3485 
3486 	ASSERT(ap->length);
3487 	orig_offset = ap->offset;
3488 	orig_length = ap->length;
3489 
3490 	mp = ap->ip->i_mount;
3491 
3492 	/* stripe alignment for allocation is determined by mount parameters */
3493 	stripe_align = 0;
3494 	if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3495 		stripe_align = mp->m_swidth;
3496 	else if (mp->m_dalign)
3497 		stripe_align = mp->m_dalign;
3498 
3499 	if (ap->flags & XFS_BMAPI_COWFORK)
3500 		align = xfs_get_cowextsz_hint(ap->ip);
3501 	else if (ap->datatype & XFS_ALLOC_USERDATA)
3502 		align = xfs_get_extsz_hint(ap->ip);
3503 	if (align) {
3504 		error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3505 						align, 0, ap->eof, 0, ap->conv,
3506 						&ap->offset, &ap->length);
3507 		ASSERT(!error);
3508 		ASSERT(ap->length);
3509 	}
3510 
3511 
3512 	nullfb = ap->tp->t_firstblock == NULLFSBLOCK;
3513 	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp,
3514 							ap->tp->t_firstblock);
3515 	if (nullfb) {
3516 		if ((ap->datatype & XFS_ALLOC_USERDATA) &&
3517 		    xfs_inode_is_filestream(ap->ip)) {
3518 			ag = xfs_filestream_lookup_ag(ap->ip);
3519 			ag = (ag != NULLAGNUMBER) ? ag : 0;
3520 			ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3521 		} else {
3522 			ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3523 		}
3524 	} else
3525 		ap->blkno = ap->tp->t_firstblock;
3526 
3527 	xfs_bmap_adjacent(ap);
3528 
3529 	/*
3530 	 * If allowed, use ap->blkno; otherwise must use firstblock since
3531 	 * it's in the right allocation group.
3532 	 */
3533 	if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3534 		;
3535 	else
3536 		ap->blkno = ap->tp->t_firstblock;
3537 	/*
3538 	 * Normal allocation, done through xfs_alloc_vextent.
3539 	 */
3540 	tryagain = isaligned = 0;
3541 	memset(&args, 0, sizeof(args));
3542 	args.tp = ap->tp;
3543 	args.mp = mp;
3544 	args.fsbno = ap->blkno;
3545 	args.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
3546 
3547 	/* Trim the allocation back to the maximum an AG can fit. */
3548 	args.maxlen = min(ap->length, mp->m_ag_max_usable);
3549 	blen = 0;
3550 	if (nullfb) {
3551 		/*
3552 		 * Search for an allocation group with a single extent large
3553 		 * enough for the request.  If one isn't found, then adjust
3554 		 * the minimum allocation size to the largest space found.
3555 		 */
3556 		if ((ap->datatype & XFS_ALLOC_USERDATA) &&
3557 		    xfs_inode_is_filestream(ap->ip))
3558 			error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3559 		else
3560 			error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3561 		if (error)
3562 			return error;
3563 	} else if (ap->tp->t_flags & XFS_TRANS_LOWMODE) {
3564 		if (xfs_inode_is_filestream(ap->ip))
3565 			args.type = XFS_ALLOCTYPE_FIRST_AG;
3566 		else
3567 			args.type = XFS_ALLOCTYPE_START_BNO;
3568 		args.total = args.minlen = ap->minlen;
3569 	} else {
3570 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
3571 		args.total = ap->total;
3572 		args.minlen = ap->minlen;
3573 	}
3574 	/* apply extent size hints if obtained earlier */
3575 	if (align) {
3576 		args.prod = align;
3577 		div_u64_rem(ap->offset, args.prod, &args.mod);
3578 		if (args.mod)
3579 			args.mod = args.prod - args.mod;
3580 	} else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3581 		args.prod = 1;
3582 		args.mod = 0;
3583 	} else {
3584 		args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3585 		div_u64_rem(ap->offset, args.prod, &args.mod);
3586 		if (args.mod)
3587 			args.mod = args.prod - args.mod;
3588 	}
3589 	/*
3590 	 * If we are not low on available data blocks, and the underlying
3591 	 * logical volume manager is a stripe, and the file offset is zero then
3592 	 * try to allocate data blocks on stripe unit boundary. NOTE: ap->aeof
3593 	 * is only set if the allocation length is >= the stripe unit and the
3594 	 * allocation offset is at the end of file.
3595 	 */
3596 	if (!(ap->tp->t_flags & XFS_TRANS_LOWMODE) && ap->aeof) {
3597 		if (!ap->offset) {
3598 			args.alignment = stripe_align;
3599 			atype = args.type;
3600 			isaligned = 1;
3601 			/*
3602 			 * Adjust minlen to try and preserve alignment if we
3603 			 * can't guarantee an aligned maxlen extent.
3604 			 */
3605 			if (blen > args.alignment &&
3606 			    blen <= args.maxlen + args.alignment)
3607 				args.minlen = blen - args.alignment;
3608 			args.minalignslop = 0;
3609 		} else {
3610 			/*
3611 			 * First try an exact bno allocation.
3612 			 * If it fails then do a near or start bno
3613 			 * allocation with alignment turned on.
3614 			 */
3615 			atype = args.type;
3616 			tryagain = 1;
3617 			args.type = XFS_ALLOCTYPE_THIS_BNO;
3618 			args.alignment = 1;
3619 			/*
3620 			 * Compute the minlen+alignment for the
3621 			 * next case.  Set slop so that the value
3622 			 * of minlen+alignment+slop doesn't go up
3623 			 * between the calls.
3624 			 */
3625 			if (blen > stripe_align && blen <= args.maxlen)
3626 				nextminlen = blen - stripe_align;
3627 			else
3628 				nextminlen = args.minlen;
3629 			if (nextminlen + stripe_align > args.minlen + 1)
3630 				args.minalignslop =
3631 					nextminlen + stripe_align -
3632 					args.minlen - 1;
3633 			else
3634 				args.minalignslop = 0;
3635 		}
3636 	} else {
3637 		args.alignment = 1;
3638 		args.minalignslop = 0;
3639 	}
3640 	args.minleft = ap->minleft;
3641 	args.wasdel = ap->wasdel;
3642 	args.resv = XFS_AG_RESV_NONE;
3643 	args.datatype = ap->datatype;
3644 
3645 	error = xfs_alloc_vextent(&args);
3646 	if (error)
3647 		return error;
3648 
3649 	if (tryagain && args.fsbno == NULLFSBLOCK) {
3650 		/*
3651 		 * Exact allocation failed. Now try with alignment
3652 		 * turned on.
3653 		 */
3654 		args.type = atype;
3655 		args.fsbno = ap->blkno;
3656 		args.alignment = stripe_align;
3657 		args.minlen = nextminlen;
3658 		args.minalignslop = 0;
3659 		isaligned = 1;
3660 		if ((error = xfs_alloc_vextent(&args)))
3661 			return error;
3662 	}
3663 	if (isaligned && args.fsbno == NULLFSBLOCK) {
3664 		/*
3665 		 * allocation failed, so turn off alignment and
3666 		 * try again.
3667 		 */
3668 		args.type = atype;
3669 		args.fsbno = ap->blkno;
3670 		args.alignment = 0;
3671 		if ((error = xfs_alloc_vextent(&args)))
3672 			return error;
3673 	}
3674 	if (args.fsbno == NULLFSBLOCK && nullfb &&
3675 	    args.minlen > ap->minlen) {
3676 		args.minlen = ap->minlen;
3677 		args.type = XFS_ALLOCTYPE_START_BNO;
3678 		args.fsbno = ap->blkno;
3679 		if ((error = xfs_alloc_vextent(&args)))
3680 			return error;
3681 	}
3682 	if (args.fsbno == NULLFSBLOCK && nullfb) {
3683 		args.fsbno = 0;
3684 		args.type = XFS_ALLOCTYPE_FIRST_AG;
3685 		args.total = ap->minlen;
3686 		if ((error = xfs_alloc_vextent(&args)))
3687 			return error;
3688 		ap->tp->t_flags |= XFS_TRANS_LOWMODE;
3689 	}
3690 	if (args.fsbno != NULLFSBLOCK) {
3691 		/*
3692 		 * check the allocation happened at the same or higher AG than
3693 		 * the first block that was allocated.
3694 		 */
3695 		ASSERT(ap->tp->t_firstblock == NULLFSBLOCK ||
3696 		       XFS_FSB_TO_AGNO(mp, ap->tp->t_firstblock) <=
3697 		       XFS_FSB_TO_AGNO(mp, args.fsbno));
3698 
3699 		ap->blkno = args.fsbno;
3700 		if (ap->tp->t_firstblock == NULLFSBLOCK)
3701 			ap->tp->t_firstblock = args.fsbno;
3702 		ASSERT(nullfb || fb_agno <= args.agno);
3703 		ap->length = args.len;
3704 		/*
3705 		 * If the extent size hint is active, we tried to round the
3706 		 * caller's allocation request offset down to extsz and the
3707 		 * length up to another extsz boundary.  If we found a free
3708 		 * extent we mapped it in starting at this new offset.  If the
3709 		 * newly mapped space isn't long enough to cover any of the
3710 		 * range of offsets that was originally requested, move the
3711 		 * mapping up so that we can fill as much of the caller's
3712 		 * original request as possible.  Free space is apparently
3713 		 * very fragmented so we're unlikely to be able to satisfy the
3714 		 * hints anyway.
3715 		 */
3716 		if (ap->length <= orig_length)
3717 			ap->offset = orig_offset;
3718 		else if (ap->offset + ap->length < orig_offset + orig_length)
3719 			ap->offset = orig_offset + orig_length - ap->length;
3720 		xfs_bmap_btalloc_accounting(ap, &args);
3721 	} else {
3722 		ap->blkno = NULLFSBLOCK;
3723 		ap->length = 0;
3724 	}
3725 	return 0;
3726 }
3727 
3728 /* Trim extent to fit a logical block range. */
3729 void
3730 xfs_trim_extent(
3731 	struct xfs_bmbt_irec	*irec,
3732 	xfs_fileoff_t		bno,
3733 	xfs_filblks_t		len)
3734 {
3735 	xfs_fileoff_t		distance;
3736 	xfs_fileoff_t		end = bno + len;
3737 
3738 	if (irec->br_startoff + irec->br_blockcount <= bno ||
3739 	    irec->br_startoff >= end) {
3740 		irec->br_blockcount = 0;
3741 		return;
3742 	}
3743 
3744 	if (irec->br_startoff < bno) {
3745 		distance = bno - irec->br_startoff;
3746 		if (isnullstartblock(irec->br_startblock))
3747 			irec->br_startblock = DELAYSTARTBLOCK;
3748 		if (irec->br_startblock != DELAYSTARTBLOCK &&
3749 		    irec->br_startblock != HOLESTARTBLOCK)
3750 			irec->br_startblock += distance;
3751 		irec->br_startoff += distance;
3752 		irec->br_blockcount -= distance;
3753 	}
3754 
3755 	if (end < irec->br_startoff + irec->br_blockcount) {
3756 		distance = irec->br_startoff + irec->br_blockcount - end;
3757 		irec->br_blockcount -= distance;
3758 	}
3759 }
3760 
3761 /*
3762  * Trim the returned map to the required bounds
3763  */
3764 STATIC void
3765 xfs_bmapi_trim_map(
3766 	struct xfs_bmbt_irec	*mval,
3767 	struct xfs_bmbt_irec	*got,
3768 	xfs_fileoff_t		*bno,
3769 	xfs_filblks_t		len,
3770 	xfs_fileoff_t		obno,
3771 	xfs_fileoff_t		end,
3772 	int			n,
3773 	int			flags)
3774 {
3775 	if ((flags & XFS_BMAPI_ENTIRE) ||
3776 	    got->br_startoff + got->br_blockcount <= obno) {
3777 		*mval = *got;
3778 		if (isnullstartblock(got->br_startblock))
3779 			mval->br_startblock = DELAYSTARTBLOCK;
3780 		return;
3781 	}
3782 
3783 	if (obno > *bno)
3784 		*bno = obno;
3785 	ASSERT((*bno >= obno) || (n == 0));
3786 	ASSERT(*bno < end);
3787 	mval->br_startoff = *bno;
3788 	if (isnullstartblock(got->br_startblock))
3789 		mval->br_startblock = DELAYSTARTBLOCK;
3790 	else
3791 		mval->br_startblock = got->br_startblock +
3792 					(*bno - got->br_startoff);
3793 	/*
3794 	 * Return the minimum of what we got and what we asked for for
3795 	 * the length.  We can use the len variable here because it is
3796 	 * modified below and we could have been there before coming
3797 	 * here if the first part of the allocation didn't overlap what
3798 	 * was asked for.
3799 	 */
3800 	mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3801 			got->br_blockcount - (*bno - got->br_startoff));
3802 	mval->br_state = got->br_state;
3803 	ASSERT(mval->br_blockcount <= len);
3804 	return;
3805 }
3806 
3807 /*
3808  * Update and validate the extent map to return
3809  */
3810 STATIC void
3811 xfs_bmapi_update_map(
3812 	struct xfs_bmbt_irec	**map,
3813 	xfs_fileoff_t		*bno,
3814 	xfs_filblks_t		*len,
3815 	xfs_fileoff_t		obno,
3816 	xfs_fileoff_t		end,
3817 	int			*n,
3818 	int			flags)
3819 {
3820 	xfs_bmbt_irec_t	*mval = *map;
3821 
3822 	ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3823 	       ((mval->br_startoff + mval->br_blockcount) <= end));
3824 	ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3825 	       (mval->br_startoff < obno));
3826 
3827 	*bno = mval->br_startoff + mval->br_blockcount;
3828 	*len = end - *bno;
3829 	if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3830 		/* update previous map with new information */
3831 		ASSERT(mval->br_startblock == mval[-1].br_startblock);
3832 		ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3833 		ASSERT(mval->br_state == mval[-1].br_state);
3834 		mval[-1].br_blockcount = mval->br_blockcount;
3835 		mval[-1].br_state = mval->br_state;
3836 	} else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3837 		   mval[-1].br_startblock != DELAYSTARTBLOCK &&
3838 		   mval[-1].br_startblock != HOLESTARTBLOCK &&
3839 		   mval->br_startblock == mval[-1].br_startblock +
3840 					  mval[-1].br_blockcount &&
3841 		   mval[-1].br_state == mval->br_state) {
3842 		ASSERT(mval->br_startoff ==
3843 		       mval[-1].br_startoff + mval[-1].br_blockcount);
3844 		mval[-1].br_blockcount += mval->br_blockcount;
3845 	} else if (*n > 0 &&
3846 		   mval->br_startblock == DELAYSTARTBLOCK &&
3847 		   mval[-1].br_startblock == DELAYSTARTBLOCK &&
3848 		   mval->br_startoff ==
3849 		   mval[-1].br_startoff + mval[-1].br_blockcount) {
3850 		mval[-1].br_blockcount += mval->br_blockcount;
3851 		mval[-1].br_state = mval->br_state;
3852 	} else if (!((*n == 0) &&
3853 		     ((mval->br_startoff + mval->br_blockcount) <=
3854 		      obno))) {
3855 		mval++;
3856 		(*n)++;
3857 	}
3858 	*map = mval;
3859 }
3860 
3861 /*
3862  * Map file blocks to filesystem blocks without allocation.
3863  */
3864 int
3865 xfs_bmapi_read(
3866 	struct xfs_inode	*ip,
3867 	xfs_fileoff_t		bno,
3868 	xfs_filblks_t		len,
3869 	struct xfs_bmbt_irec	*mval,
3870 	int			*nmap,
3871 	int			flags)
3872 {
3873 	struct xfs_mount	*mp = ip->i_mount;
3874 	int			whichfork = xfs_bmapi_whichfork(flags);
3875 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
3876 	struct xfs_bmbt_irec	got;
3877 	xfs_fileoff_t		obno;
3878 	xfs_fileoff_t		end;
3879 	struct xfs_iext_cursor	icur;
3880 	int			error;
3881 	bool			eof = false;
3882 	int			n = 0;
3883 
3884 	ASSERT(*nmap >= 1);
3885 	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_ENTIRE)));
3886 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
3887 
3888 	if (WARN_ON_ONCE(!ifp))
3889 		return -EFSCORRUPTED;
3890 
3891 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
3892 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT))
3893 		return -EFSCORRUPTED;
3894 
3895 	if (XFS_FORCED_SHUTDOWN(mp))
3896 		return -EIO;
3897 
3898 	XFS_STATS_INC(mp, xs_blk_mapr);
3899 
3900 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
3901 		error = xfs_iread_extents(NULL, ip, whichfork);
3902 		if (error)
3903 			return error;
3904 	}
3905 
3906 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got))
3907 		eof = true;
3908 	end = bno + len;
3909 	obno = bno;
3910 
3911 	while (bno < end && n < *nmap) {
3912 		/* Reading past eof, act as though there's a hole up to end. */
3913 		if (eof)
3914 			got.br_startoff = end;
3915 		if (got.br_startoff > bno) {
3916 			/* Reading in a hole.  */
3917 			mval->br_startoff = bno;
3918 			mval->br_startblock = HOLESTARTBLOCK;
3919 			mval->br_blockcount =
3920 				XFS_FILBLKS_MIN(len, got.br_startoff - bno);
3921 			mval->br_state = XFS_EXT_NORM;
3922 			bno += mval->br_blockcount;
3923 			len -= mval->br_blockcount;
3924 			mval++;
3925 			n++;
3926 			continue;
3927 		}
3928 
3929 		/* set up the extent map to return. */
3930 		xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
3931 		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
3932 
3933 		/* If we're done, stop now. */
3934 		if (bno >= end || n >= *nmap)
3935 			break;
3936 
3937 		/* Else go on to the next record. */
3938 		if (!xfs_iext_next_extent(ifp, &icur, &got))
3939 			eof = true;
3940 	}
3941 	*nmap = n;
3942 	return 0;
3943 }
3944 
3945 /*
3946  * Add a delayed allocation extent to an inode. Blocks are reserved from the
3947  * global pool and the extent inserted into the inode in-core extent tree.
3948  *
3949  * On entry, got refers to the first extent beyond the offset of the extent to
3950  * allocate or eof is specified if no such extent exists. On return, got refers
3951  * to the extent record that was inserted to the inode fork.
3952  *
3953  * Note that the allocated extent may have been merged with contiguous extents
3954  * during insertion into the inode fork. Thus, got does not reflect the current
3955  * state of the inode fork on return. If necessary, the caller can use lastx to
3956  * look up the updated record in the inode fork.
3957  */
3958 int
3959 xfs_bmapi_reserve_delalloc(
3960 	struct xfs_inode	*ip,
3961 	int			whichfork,
3962 	xfs_fileoff_t		off,
3963 	xfs_filblks_t		len,
3964 	xfs_filblks_t		prealloc,
3965 	struct xfs_bmbt_irec	*got,
3966 	struct xfs_iext_cursor	*icur,
3967 	int			eof)
3968 {
3969 	struct xfs_mount	*mp = ip->i_mount;
3970 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
3971 	xfs_extlen_t		alen;
3972 	xfs_extlen_t		indlen;
3973 	int			error;
3974 	xfs_fileoff_t		aoff = off;
3975 
3976 	/*
3977 	 * Cap the alloc length. Keep track of prealloc so we know whether to
3978 	 * tag the inode before we return.
3979 	 */
3980 	alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN);
3981 	if (!eof)
3982 		alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
3983 	if (prealloc && alen >= len)
3984 		prealloc = alen - len;
3985 
3986 	/* Figure out the extent size, adjust alen */
3987 	if (whichfork == XFS_COW_FORK) {
3988 		struct xfs_bmbt_irec	prev;
3989 		xfs_extlen_t		extsz = xfs_get_cowextsz_hint(ip);
3990 
3991 		if (!xfs_iext_peek_prev_extent(ifp, icur, &prev))
3992 			prev.br_startoff = NULLFILEOFF;
3993 
3994 		error = xfs_bmap_extsize_align(mp, got, &prev, extsz, 0, eof,
3995 					       1, 0, &aoff, &alen);
3996 		ASSERT(!error);
3997 	}
3998 
3999 	/*
4000 	 * Make a transaction-less quota reservation for delayed allocation
4001 	 * blocks.  This number gets adjusted later.  We return if we haven't
4002 	 * allocated blocks already inside this loop.
4003 	 */
4004 	error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4005 						XFS_QMOPT_RES_REGBLKS);
4006 	if (error)
4007 		return error;
4008 
4009 	/*
4010 	 * Split changing sb for alen and indlen since they could be coming
4011 	 * from different places.
4012 	 */
4013 	indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4014 	ASSERT(indlen > 0);
4015 
4016 	error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
4017 	if (error)
4018 		goto out_unreserve_quota;
4019 
4020 	error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
4021 	if (error)
4022 		goto out_unreserve_blocks;
4023 
4024 
4025 	ip->i_delayed_blks += alen;
4026 	xfs_mod_delalloc(ip->i_mount, alen + indlen);
4027 
4028 	got->br_startoff = aoff;
4029 	got->br_startblock = nullstartblock(indlen);
4030 	got->br_blockcount = alen;
4031 	got->br_state = XFS_EXT_NORM;
4032 
4033 	xfs_bmap_add_extent_hole_delay(ip, whichfork, icur, got);
4034 
4035 	/*
4036 	 * Tag the inode if blocks were preallocated. Note that COW fork
4037 	 * preallocation can occur at the start or end of the extent, even when
4038 	 * prealloc == 0, so we must also check the aligned offset and length.
4039 	 */
4040 	if (whichfork == XFS_DATA_FORK && prealloc)
4041 		xfs_inode_set_eofblocks_tag(ip);
4042 	if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len))
4043 		xfs_inode_set_cowblocks_tag(ip);
4044 
4045 	return 0;
4046 
4047 out_unreserve_blocks:
4048 	xfs_mod_fdblocks(mp, alen, false);
4049 out_unreserve_quota:
4050 	if (XFS_IS_QUOTA_ON(mp))
4051 		xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0,
4052 						XFS_QMOPT_RES_REGBLKS);
4053 	return error;
4054 }
4055 
4056 static int
4057 xfs_bmap_alloc_userdata(
4058 	struct xfs_bmalloca	*bma)
4059 {
4060 	struct xfs_mount	*mp = bma->ip->i_mount;
4061 	int			whichfork = xfs_bmapi_whichfork(bma->flags);
4062 	int			error;
4063 
4064 	/*
4065 	 * Set the data type being allocated. For the data fork, the first data
4066 	 * in the file is treated differently to all other allocations. For the
4067 	 * attribute fork, we only need to ensure the allocated range is not on
4068 	 * the busy list.
4069 	 */
4070 	bma->datatype = XFS_ALLOC_NOBUSY;
4071 	if (whichfork == XFS_DATA_FORK) {
4072 		bma->datatype |= XFS_ALLOC_USERDATA;
4073 		if (bma->offset == 0)
4074 			bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA;
4075 
4076 		if (mp->m_dalign && bma->length >= mp->m_dalign) {
4077 			error = xfs_bmap_isaeof(bma, whichfork);
4078 			if (error)
4079 				return error;
4080 		}
4081 
4082 		if (XFS_IS_REALTIME_INODE(bma->ip))
4083 			return xfs_bmap_rtalloc(bma);
4084 	}
4085 
4086 	return xfs_bmap_btalloc(bma);
4087 }
4088 
4089 static int
4090 xfs_bmapi_allocate(
4091 	struct xfs_bmalloca	*bma)
4092 {
4093 	struct xfs_mount	*mp = bma->ip->i_mount;
4094 	int			whichfork = xfs_bmapi_whichfork(bma->flags);
4095 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4096 	int			tmp_logflags = 0;
4097 	int			error;
4098 
4099 	ASSERT(bma->length > 0);
4100 
4101 	/*
4102 	 * For the wasdelay case, we could also just allocate the stuff asked
4103 	 * for in this bmap call but that wouldn't be as good.
4104 	 */
4105 	if (bma->wasdel) {
4106 		bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4107 		bma->offset = bma->got.br_startoff;
4108 		if (!xfs_iext_peek_prev_extent(ifp, &bma->icur, &bma->prev))
4109 			bma->prev.br_startoff = NULLFILEOFF;
4110 	} else {
4111 		bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4112 		if (!bma->eof)
4113 			bma->length = XFS_FILBLKS_MIN(bma->length,
4114 					bma->got.br_startoff - bma->offset);
4115 	}
4116 
4117 	if (bma->flags & XFS_BMAPI_CONTIG)
4118 		bma->minlen = bma->length;
4119 	else
4120 		bma->minlen = 1;
4121 
4122 	if (bma->flags & XFS_BMAPI_METADATA)
4123 		error = xfs_bmap_btalloc(bma);
4124 	else
4125 		error = xfs_bmap_alloc_userdata(bma);
4126 	if (error || bma->blkno == NULLFSBLOCK)
4127 		return error;
4128 
4129 	if (bma->flags & XFS_BMAPI_ZERO) {
4130 		error = xfs_zero_extent(bma->ip, bma->blkno, bma->length);
4131 		if (error)
4132 			return error;
4133 	}
4134 
4135 	if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur)
4136 		bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4137 	/*
4138 	 * Bump the number of extents we've allocated
4139 	 * in this call.
4140 	 */
4141 	bma->nallocs++;
4142 
4143 	if (bma->cur)
4144 		bma->cur->bc_ino.flags =
4145 			bma->wasdel ? XFS_BTCUR_BMBT_WASDEL : 0;
4146 
4147 	bma->got.br_startoff = bma->offset;
4148 	bma->got.br_startblock = bma->blkno;
4149 	bma->got.br_blockcount = bma->length;
4150 	bma->got.br_state = XFS_EXT_NORM;
4151 
4152 	if (bma->flags & XFS_BMAPI_PREALLOC)
4153 		bma->got.br_state = XFS_EXT_UNWRITTEN;
4154 
4155 	if (bma->wasdel)
4156 		error = xfs_bmap_add_extent_delay_real(bma, whichfork);
4157 	else
4158 		error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip,
4159 				whichfork, &bma->icur, &bma->cur, &bma->got,
4160 				&bma->logflags, bma->flags);
4161 
4162 	bma->logflags |= tmp_logflags;
4163 	if (error)
4164 		return error;
4165 
4166 	/*
4167 	 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4168 	 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4169 	 * the neighbouring ones.
4170 	 */
4171 	xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4172 
4173 	ASSERT(bma->got.br_startoff <= bma->offset);
4174 	ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4175 	       bma->offset + bma->length);
4176 	ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4177 	       bma->got.br_state == XFS_EXT_UNWRITTEN);
4178 	return 0;
4179 }
4180 
4181 STATIC int
4182 xfs_bmapi_convert_unwritten(
4183 	struct xfs_bmalloca	*bma,
4184 	struct xfs_bmbt_irec	*mval,
4185 	xfs_filblks_t		len,
4186 	int			flags)
4187 {
4188 	int			whichfork = xfs_bmapi_whichfork(flags);
4189 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4190 	int			tmp_logflags = 0;
4191 	int			error;
4192 
4193 	/* check if we need to do unwritten->real conversion */
4194 	if (mval->br_state == XFS_EXT_UNWRITTEN &&
4195 	    (flags & XFS_BMAPI_PREALLOC))
4196 		return 0;
4197 
4198 	/* check if we need to do real->unwritten conversion */
4199 	if (mval->br_state == XFS_EXT_NORM &&
4200 	    (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4201 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4202 		return 0;
4203 
4204 	/*
4205 	 * Modify (by adding) the state flag, if writing.
4206 	 */
4207 	ASSERT(mval->br_blockcount <= len);
4208 	if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4209 		bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4210 					bma->ip, whichfork);
4211 	}
4212 	mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4213 				? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4214 
4215 	/*
4216 	 * Before insertion into the bmbt, zero the range being converted
4217 	 * if required.
4218 	 */
4219 	if (flags & XFS_BMAPI_ZERO) {
4220 		error = xfs_zero_extent(bma->ip, mval->br_startblock,
4221 					mval->br_blockcount);
4222 		if (error)
4223 			return error;
4224 	}
4225 
4226 	error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork,
4227 			&bma->icur, &bma->cur, mval, &tmp_logflags);
4228 	/*
4229 	 * Log the inode core unconditionally in the unwritten extent conversion
4230 	 * path because the conversion might not have done so (e.g., if the
4231 	 * extent count hasn't changed). We need to make sure the inode is dirty
4232 	 * in the transaction for the sake of fsync(), even if nothing has
4233 	 * changed, because fsync() will not force the log for this transaction
4234 	 * unless it sees the inode pinned.
4235 	 *
4236 	 * Note: If we're only converting cow fork extents, there aren't
4237 	 * any on-disk updates to make, so we don't need to log anything.
4238 	 */
4239 	if (whichfork != XFS_COW_FORK)
4240 		bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4241 	if (error)
4242 		return error;
4243 
4244 	/*
4245 	 * Update our extent pointer, given that
4246 	 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4247 	 * of the neighbouring ones.
4248 	 */
4249 	xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4250 
4251 	/*
4252 	 * We may have combined previously unwritten space with written space,
4253 	 * so generate another request.
4254 	 */
4255 	if (mval->br_blockcount < len)
4256 		return -EAGAIN;
4257 	return 0;
4258 }
4259 
4260 static inline xfs_extlen_t
4261 xfs_bmapi_minleft(
4262 	struct xfs_trans	*tp,
4263 	struct xfs_inode	*ip,
4264 	int			fork)
4265 {
4266 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, fork);
4267 
4268 	if (tp && tp->t_firstblock != NULLFSBLOCK)
4269 		return 0;
4270 	if (ifp->if_format != XFS_DINODE_FMT_BTREE)
4271 		return 1;
4272 	return be16_to_cpu(ifp->if_broot->bb_level) + 1;
4273 }
4274 
4275 /*
4276  * Log whatever the flags say, even if error.  Otherwise we might miss detecting
4277  * a case where the data is changed, there's an error, and it's not logged so we
4278  * don't shutdown when we should.  Don't bother logging extents/btree changes if
4279  * we converted to the other format.
4280  */
4281 static void
4282 xfs_bmapi_finish(
4283 	struct xfs_bmalloca	*bma,
4284 	int			whichfork,
4285 	int			error)
4286 {
4287 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4288 
4289 	if ((bma->logflags & xfs_ilog_fext(whichfork)) &&
4290 	    ifp->if_format != XFS_DINODE_FMT_EXTENTS)
4291 		bma->logflags &= ~xfs_ilog_fext(whichfork);
4292 	else if ((bma->logflags & xfs_ilog_fbroot(whichfork)) &&
4293 		 ifp->if_format != XFS_DINODE_FMT_BTREE)
4294 		bma->logflags &= ~xfs_ilog_fbroot(whichfork);
4295 
4296 	if (bma->logflags)
4297 		xfs_trans_log_inode(bma->tp, bma->ip, bma->logflags);
4298 	if (bma->cur)
4299 		xfs_btree_del_cursor(bma->cur, error);
4300 }
4301 
4302 /*
4303  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4304  * extent state if necessary.  Details behaviour is controlled by the flags
4305  * parameter.  Only allocates blocks from a single allocation group, to avoid
4306  * locking problems.
4307  */
4308 int
4309 xfs_bmapi_write(
4310 	struct xfs_trans	*tp,		/* transaction pointer */
4311 	struct xfs_inode	*ip,		/* incore inode */
4312 	xfs_fileoff_t		bno,		/* starting file offs. mapped */
4313 	xfs_filblks_t		len,		/* length to map in file */
4314 	int			flags,		/* XFS_BMAPI_... */
4315 	xfs_extlen_t		total,		/* total blocks needed */
4316 	struct xfs_bmbt_irec	*mval,		/* output: map values */
4317 	int			*nmap)		/* i/o: mval size/count */
4318 {
4319 	struct xfs_bmalloca	bma = {
4320 		.tp		= tp,
4321 		.ip		= ip,
4322 		.total		= total,
4323 	};
4324 	struct xfs_mount	*mp = ip->i_mount;
4325 	int			whichfork = xfs_bmapi_whichfork(flags);
4326 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
4327 	xfs_fileoff_t		end;		/* end of mapped file region */
4328 	bool			eof = false;	/* after the end of extents */
4329 	int			error;		/* error return */
4330 	int			n;		/* current extent index */
4331 	xfs_fileoff_t		obno;		/* old block number (offset) */
4332 
4333 #ifdef DEBUG
4334 	xfs_fileoff_t		orig_bno;	/* original block number value */
4335 	int			orig_flags;	/* original flags arg value */
4336 	xfs_filblks_t		orig_len;	/* original value of len arg */
4337 	struct xfs_bmbt_irec	*orig_mval;	/* original value of mval */
4338 	int			orig_nmap;	/* original value of *nmap */
4339 
4340 	orig_bno = bno;
4341 	orig_len = len;
4342 	orig_flags = flags;
4343 	orig_mval = mval;
4344 	orig_nmap = *nmap;
4345 #endif
4346 
4347 	ASSERT(*nmap >= 1);
4348 	ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4349 	ASSERT(tp != NULL);
4350 	ASSERT(len > 0);
4351 	ASSERT(ifp->if_format != XFS_DINODE_FMT_LOCAL);
4352 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4353 	ASSERT(!(flags & XFS_BMAPI_REMAP));
4354 
4355 	/* zeroing is for currently only for data extents, not metadata */
4356 	ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4357 			(XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4358 	/*
4359 	 * we can allocate unwritten extents or pre-zero allocated blocks,
4360 	 * but it makes no sense to do both at once. This would result in
4361 	 * zeroing the unwritten extent twice, but it still being an
4362 	 * unwritten extent....
4363 	 */
4364 	ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4365 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4366 
4367 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
4368 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
4369 		return -EFSCORRUPTED;
4370 	}
4371 
4372 	if (XFS_FORCED_SHUTDOWN(mp))
4373 		return -EIO;
4374 
4375 	XFS_STATS_INC(mp, xs_blk_mapw);
4376 
4377 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4378 		error = xfs_iread_extents(tp, ip, whichfork);
4379 		if (error)
4380 			goto error0;
4381 	}
4382 
4383 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.icur, &bma.got))
4384 		eof = true;
4385 	if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4386 		bma.prev.br_startoff = NULLFILEOFF;
4387 	bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4388 
4389 	n = 0;
4390 	end = bno + len;
4391 	obno = bno;
4392 	while (bno < end && n < *nmap) {
4393 		bool			need_alloc = false, wasdelay = false;
4394 
4395 		/* in hole or beyond EOF? */
4396 		if (eof || bma.got.br_startoff > bno) {
4397 			/*
4398 			 * CoW fork conversions should /never/ hit EOF or
4399 			 * holes.  There should always be something for us
4400 			 * to work on.
4401 			 */
4402 			ASSERT(!((flags & XFS_BMAPI_CONVERT) &&
4403 			         (flags & XFS_BMAPI_COWFORK)));
4404 
4405 			need_alloc = true;
4406 		} else if (isnullstartblock(bma.got.br_startblock)) {
4407 			wasdelay = true;
4408 		}
4409 
4410 		/*
4411 		 * First, deal with the hole before the allocated space
4412 		 * that we found, if any.
4413 		 */
4414 		if (need_alloc || wasdelay) {
4415 			bma.eof = eof;
4416 			bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4417 			bma.wasdel = wasdelay;
4418 			bma.offset = bno;
4419 			bma.flags = flags;
4420 
4421 			/*
4422 			 * There's a 32/64 bit type mismatch between the
4423 			 * allocation length request (which can be 64 bits in
4424 			 * length) and the bma length request, which is
4425 			 * xfs_extlen_t and therefore 32 bits. Hence we have to
4426 			 * check for 32-bit overflows and handle them here.
4427 			 */
4428 			if (len > (xfs_filblks_t)MAXEXTLEN)
4429 				bma.length = MAXEXTLEN;
4430 			else
4431 				bma.length = len;
4432 
4433 			ASSERT(len > 0);
4434 			ASSERT(bma.length > 0);
4435 			error = xfs_bmapi_allocate(&bma);
4436 			if (error)
4437 				goto error0;
4438 			if (bma.blkno == NULLFSBLOCK)
4439 				break;
4440 
4441 			/*
4442 			 * If this is a CoW allocation, record the data in
4443 			 * the refcount btree for orphan recovery.
4444 			 */
4445 			if (whichfork == XFS_COW_FORK)
4446 				xfs_refcount_alloc_cow_extent(tp, bma.blkno,
4447 						bma.length);
4448 		}
4449 
4450 		/* Deal with the allocated space we found.  */
4451 		xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4452 							end, n, flags);
4453 
4454 		/* Execute unwritten extent conversion if necessary */
4455 		error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4456 		if (error == -EAGAIN)
4457 			continue;
4458 		if (error)
4459 			goto error0;
4460 
4461 		/* update the extent map to return */
4462 		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4463 
4464 		/*
4465 		 * If we're done, stop now.  Stop when we've allocated
4466 		 * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4467 		 * the transaction may get too big.
4468 		 */
4469 		if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4470 			break;
4471 
4472 		/* Else go on to the next record. */
4473 		bma.prev = bma.got;
4474 		if (!xfs_iext_next_extent(ifp, &bma.icur, &bma.got))
4475 			eof = true;
4476 	}
4477 	*nmap = n;
4478 
4479 	error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4480 			whichfork);
4481 	if (error)
4482 		goto error0;
4483 
4484 	ASSERT(ifp->if_format != XFS_DINODE_FMT_BTREE ||
4485 	       ifp->if_nextents > XFS_IFORK_MAXEXT(ip, whichfork));
4486 	xfs_bmapi_finish(&bma, whichfork, 0);
4487 	xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4488 		orig_nmap, *nmap);
4489 	return 0;
4490 error0:
4491 	xfs_bmapi_finish(&bma, whichfork, error);
4492 	return error;
4493 }
4494 
4495 /*
4496  * Convert an existing delalloc extent to real blocks based on file offset. This
4497  * attempts to allocate the entire delalloc extent and may require multiple
4498  * invocations to allocate the target offset if a large enough physical extent
4499  * is not available.
4500  */
4501 int
4502 xfs_bmapi_convert_delalloc(
4503 	struct xfs_inode	*ip,
4504 	int			whichfork,
4505 	xfs_off_t		offset,
4506 	struct iomap		*iomap,
4507 	unsigned int		*seq)
4508 {
4509 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
4510 	struct xfs_mount	*mp = ip->i_mount;
4511 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
4512 	struct xfs_bmalloca	bma = { NULL };
4513 	uint16_t		flags = 0;
4514 	struct xfs_trans	*tp;
4515 	int			error;
4516 
4517 	if (whichfork == XFS_COW_FORK)
4518 		flags |= IOMAP_F_SHARED;
4519 
4520 	/*
4521 	 * Space for the extent and indirect blocks was reserved when the
4522 	 * delalloc extent was created so there's no need to do so here.
4523 	 */
4524 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0,
4525 				XFS_TRANS_RESERVE, &tp);
4526 	if (error)
4527 		return error;
4528 
4529 	xfs_ilock(ip, XFS_ILOCK_EXCL);
4530 	xfs_trans_ijoin(tp, ip, 0);
4531 
4532 	if (!xfs_iext_lookup_extent(ip, ifp, offset_fsb, &bma.icur, &bma.got) ||
4533 	    bma.got.br_startoff > offset_fsb) {
4534 		/*
4535 		 * No extent found in the range we are trying to convert.  This
4536 		 * should only happen for the COW fork, where another thread
4537 		 * might have moved the extent to the data fork in the meantime.
4538 		 */
4539 		WARN_ON_ONCE(whichfork != XFS_COW_FORK);
4540 		error = -EAGAIN;
4541 		goto out_trans_cancel;
4542 	}
4543 
4544 	/*
4545 	 * If we find a real extent here we raced with another thread converting
4546 	 * the extent.  Just return the real extent at this offset.
4547 	 */
4548 	if (!isnullstartblock(bma.got.br_startblock)) {
4549 		xfs_bmbt_to_iomap(ip, iomap, &bma.got, flags);
4550 		*seq = READ_ONCE(ifp->if_seq);
4551 		goto out_trans_cancel;
4552 	}
4553 
4554 	bma.tp = tp;
4555 	bma.ip = ip;
4556 	bma.wasdel = true;
4557 	bma.offset = bma.got.br_startoff;
4558 	bma.length = max_t(xfs_filblks_t, bma.got.br_blockcount, MAXEXTLEN);
4559 	bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4560 
4561 	/*
4562 	 * When we're converting the delalloc reservations backing dirty pages
4563 	 * in the page cache, we must be careful about how we create the new
4564 	 * extents:
4565 	 *
4566 	 * New CoW fork extents are created unwritten, turned into real extents
4567 	 * when we're about to write the data to disk, and mapped into the data
4568 	 * fork after the write finishes.  End of story.
4569 	 *
4570 	 * New data fork extents must be mapped in as unwritten and converted
4571 	 * to real extents after the write succeeds to avoid exposing stale
4572 	 * disk contents if we crash.
4573 	 */
4574 	bma.flags = XFS_BMAPI_PREALLOC;
4575 	if (whichfork == XFS_COW_FORK)
4576 		bma.flags |= XFS_BMAPI_COWFORK;
4577 
4578 	if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4579 		bma.prev.br_startoff = NULLFILEOFF;
4580 
4581 	error = xfs_bmapi_allocate(&bma);
4582 	if (error)
4583 		goto out_finish;
4584 
4585 	error = -ENOSPC;
4586 	if (WARN_ON_ONCE(bma.blkno == NULLFSBLOCK))
4587 		goto out_finish;
4588 	error = -EFSCORRUPTED;
4589 	if (WARN_ON_ONCE(!xfs_valid_startblock(ip, bma.got.br_startblock)))
4590 		goto out_finish;
4591 
4592 	XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, bma.length));
4593 	XFS_STATS_INC(mp, xs_xstrat_quick);
4594 
4595 	ASSERT(!isnullstartblock(bma.got.br_startblock));
4596 	xfs_bmbt_to_iomap(ip, iomap, &bma.got, flags);
4597 	*seq = READ_ONCE(ifp->if_seq);
4598 
4599 	if (whichfork == XFS_COW_FORK)
4600 		xfs_refcount_alloc_cow_extent(tp, bma.blkno, bma.length);
4601 
4602 	error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4603 			whichfork);
4604 	if (error)
4605 		goto out_finish;
4606 
4607 	xfs_bmapi_finish(&bma, whichfork, 0);
4608 	error = xfs_trans_commit(tp);
4609 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
4610 	return error;
4611 
4612 out_finish:
4613 	xfs_bmapi_finish(&bma, whichfork, error);
4614 out_trans_cancel:
4615 	xfs_trans_cancel(tp);
4616 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
4617 	return error;
4618 }
4619 
4620 int
4621 xfs_bmapi_remap(
4622 	struct xfs_trans	*tp,
4623 	struct xfs_inode	*ip,
4624 	xfs_fileoff_t		bno,
4625 	xfs_filblks_t		len,
4626 	xfs_fsblock_t		startblock,
4627 	int			flags)
4628 {
4629 	struct xfs_mount	*mp = ip->i_mount;
4630 	struct xfs_ifork	*ifp;
4631 	struct xfs_btree_cur	*cur = NULL;
4632 	struct xfs_bmbt_irec	got;
4633 	struct xfs_iext_cursor	icur;
4634 	int			whichfork = xfs_bmapi_whichfork(flags);
4635 	int			logflags = 0, error;
4636 
4637 	ifp = XFS_IFORK_PTR(ip, whichfork);
4638 	ASSERT(len > 0);
4639 	ASSERT(len <= (xfs_filblks_t)MAXEXTLEN);
4640 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4641 	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC |
4642 			   XFS_BMAPI_NORMAP)));
4643 	ASSERT((flags & (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC)) !=
4644 			(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC));
4645 
4646 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
4647 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
4648 		return -EFSCORRUPTED;
4649 	}
4650 
4651 	if (XFS_FORCED_SHUTDOWN(mp))
4652 		return -EIO;
4653 
4654 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4655 		error = xfs_iread_extents(tp, ip, whichfork);
4656 		if (error)
4657 			return error;
4658 	}
4659 
4660 	if (xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
4661 		/* make sure we only reflink into a hole. */
4662 		ASSERT(got.br_startoff > bno);
4663 		ASSERT(got.br_startoff - bno >= len);
4664 	}
4665 
4666 	ip->i_d.di_nblocks += len;
4667 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4668 
4669 	if (ifp->if_flags & XFS_IFBROOT) {
4670 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
4671 		cur->bc_ino.flags = 0;
4672 	}
4673 
4674 	got.br_startoff = bno;
4675 	got.br_startblock = startblock;
4676 	got.br_blockcount = len;
4677 	if (flags & XFS_BMAPI_PREALLOC)
4678 		got.br_state = XFS_EXT_UNWRITTEN;
4679 	else
4680 		got.br_state = XFS_EXT_NORM;
4681 
4682 	error = xfs_bmap_add_extent_hole_real(tp, ip, whichfork, &icur,
4683 			&cur, &got, &logflags, flags);
4684 	if (error)
4685 		goto error0;
4686 
4687 	error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags, whichfork);
4688 
4689 error0:
4690 	if (ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS)
4691 		logflags &= ~XFS_ILOG_DEXT;
4692 	else if (ip->i_df.if_format != XFS_DINODE_FMT_BTREE)
4693 		logflags &= ~XFS_ILOG_DBROOT;
4694 
4695 	if (logflags)
4696 		xfs_trans_log_inode(tp, ip, logflags);
4697 	if (cur)
4698 		xfs_btree_del_cursor(cur, error);
4699 	return error;
4700 }
4701 
4702 /*
4703  * When a delalloc extent is split (e.g., due to a hole punch), the original
4704  * indlen reservation must be shared across the two new extents that are left
4705  * behind.
4706  *
4707  * Given the original reservation and the worst case indlen for the two new
4708  * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4709  * reservation fairly across the two new extents. If necessary, steal available
4710  * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4711  * ores == 1). The number of stolen blocks is returned. The availability and
4712  * subsequent accounting of stolen blocks is the responsibility of the caller.
4713  */
4714 static xfs_filblks_t
4715 xfs_bmap_split_indlen(
4716 	xfs_filblks_t			ores,		/* original res. */
4717 	xfs_filblks_t			*indlen1,	/* ext1 worst indlen */
4718 	xfs_filblks_t			*indlen2,	/* ext2 worst indlen */
4719 	xfs_filblks_t			avail)		/* stealable blocks */
4720 {
4721 	xfs_filblks_t			len1 = *indlen1;
4722 	xfs_filblks_t			len2 = *indlen2;
4723 	xfs_filblks_t			nres = len1 + len2; /* new total res. */
4724 	xfs_filblks_t			stolen = 0;
4725 	xfs_filblks_t			resfactor;
4726 
4727 	/*
4728 	 * Steal as many blocks as we can to try and satisfy the worst case
4729 	 * indlen for both new extents.
4730 	 */
4731 	if (ores < nres && avail)
4732 		stolen = XFS_FILBLKS_MIN(nres - ores, avail);
4733 	ores += stolen;
4734 
4735 	 /* nothing else to do if we've satisfied the new reservation */
4736 	if (ores >= nres)
4737 		return stolen;
4738 
4739 	/*
4740 	 * We can't meet the total required reservation for the two extents.
4741 	 * Calculate the percent of the overall shortage between both extents
4742 	 * and apply this percentage to each of the requested indlen values.
4743 	 * This distributes the shortage fairly and reduces the chances that one
4744 	 * of the two extents is left with nothing when extents are repeatedly
4745 	 * split.
4746 	 */
4747 	resfactor = (ores * 100);
4748 	do_div(resfactor, nres);
4749 	len1 *= resfactor;
4750 	do_div(len1, 100);
4751 	len2 *= resfactor;
4752 	do_div(len2, 100);
4753 	ASSERT(len1 + len2 <= ores);
4754 	ASSERT(len1 < *indlen1 && len2 < *indlen2);
4755 
4756 	/*
4757 	 * Hand out the remainder to each extent. If one of the two reservations
4758 	 * is zero, we want to make sure that one gets a block first. The loop
4759 	 * below starts with len1, so hand len2 a block right off the bat if it
4760 	 * is zero.
4761 	 */
4762 	ores -= (len1 + len2);
4763 	ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores);
4764 	if (ores && !len2 && *indlen2) {
4765 		len2++;
4766 		ores--;
4767 	}
4768 	while (ores) {
4769 		if (len1 < *indlen1) {
4770 			len1++;
4771 			ores--;
4772 		}
4773 		if (!ores)
4774 			break;
4775 		if (len2 < *indlen2) {
4776 			len2++;
4777 			ores--;
4778 		}
4779 	}
4780 
4781 	*indlen1 = len1;
4782 	*indlen2 = len2;
4783 
4784 	return stolen;
4785 }
4786 
4787 int
4788 xfs_bmap_del_extent_delay(
4789 	struct xfs_inode	*ip,
4790 	int			whichfork,
4791 	struct xfs_iext_cursor	*icur,
4792 	struct xfs_bmbt_irec	*got,
4793 	struct xfs_bmbt_irec	*del)
4794 {
4795 	struct xfs_mount	*mp = ip->i_mount;
4796 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
4797 	struct xfs_bmbt_irec	new;
4798 	int64_t			da_old, da_new, da_diff = 0;
4799 	xfs_fileoff_t		del_endoff, got_endoff;
4800 	xfs_filblks_t		got_indlen, new_indlen, stolen;
4801 	int			state = xfs_bmap_fork_to_state(whichfork);
4802 	int			error = 0;
4803 	bool			isrt;
4804 
4805 	XFS_STATS_INC(mp, xs_del_exlist);
4806 
4807 	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4808 	del_endoff = del->br_startoff + del->br_blockcount;
4809 	got_endoff = got->br_startoff + got->br_blockcount;
4810 	da_old = startblockval(got->br_startblock);
4811 	da_new = 0;
4812 
4813 	ASSERT(del->br_blockcount > 0);
4814 	ASSERT(got->br_startoff <= del->br_startoff);
4815 	ASSERT(got_endoff >= del_endoff);
4816 
4817 	if (isrt) {
4818 		uint64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount);
4819 
4820 		do_div(rtexts, mp->m_sb.sb_rextsize);
4821 		xfs_mod_frextents(mp, rtexts);
4822 	}
4823 
4824 	/*
4825 	 * Update the inode delalloc counter now and wait to update the
4826 	 * sb counters as we might have to borrow some blocks for the
4827 	 * indirect block accounting.
4828 	 */
4829 	error = xfs_trans_reserve_quota_nblks(NULL, ip,
4830 			-((long)del->br_blockcount), 0,
4831 			isrt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4832 	if (error)
4833 		return error;
4834 	ip->i_delayed_blks -= del->br_blockcount;
4835 
4836 	if (got->br_startoff == del->br_startoff)
4837 		state |= BMAP_LEFT_FILLING;
4838 	if (got_endoff == del_endoff)
4839 		state |= BMAP_RIGHT_FILLING;
4840 
4841 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4842 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4843 		/*
4844 		 * Matches the whole extent.  Delete the entry.
4845 		 */
4846 		xfs_iext_remove(ip, icur, state);
4847 		xfs_iext_prev(ifp, icur);
4848 		break;
4849 	case BMAP_LEFT_FILLING:
4850 		/*
4851 		 * Deleting the first part of the extent.
4852 		 */
4853 		got->br_startoff = del_endoff;
4854 		got->br_blockcount -= del->br_blockcount;
4855 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4856 				got->br_blockcount), da_old);
4857 		got->br_startblock = nullstartblock((int)da_new);
4858 		xfs_iext_update_extent(ip, state, icur, got);
4859 		break;
4860 	case BMAP_RIGHT_FILLING:
4861 		/*
4862 		 * Deleting the last part of the extent.
4863 		 */
4864 		got->br_blockcount = got->br_blockcount - del->br_blockcount;
4865 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4866 				got->br_blockcount), da_old);
4867 		got->br_startblock = nullstartblock((int)da_new);
4868 		xfs_iext_update_extent(ip, state, icur, got);
4869 		break;
4870 	case 0:
4871 		/*
4872 		 * Deleting the middle of the extent.
4873 		 *
4874 		 * Distribute the original indlen reservation across the two new
4875 		 * extents.  Steal blocks from the deleted extent if necessary.
4876 		 * Stealing blocks simply fudges the fdblocks accounting below.
4877 		 * Warn if either of the new indlen reservations is zero as this
4878 		 * can lead to delalloc problems.
4879 		 */
4880 		got->br_blockcount = del->br_startoff - got->br_startoff;
4881 		got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount);
4882 
4883 		new.br_blockcount = got_endoff - del_endoff;
4884 		new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount);
4885 
4886 		WARN_ON_ONCE(!got_indlen || !new_indlen);
4887 		stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen,
4888 						       del->br_blockcount);
4889 
4890 		got->br_startblock = nullstartblock((int)got_indlen);
4891 
4892 		new.br_startoff = del_endoff;
4893 		new.br_state = got->br_state;
4894 		new.br_startblock = nullstartblock((int)new_indlen);
4895 
4896 		xfs_iext_update_extent(ip, state, icur, got);
4897 		xfs_iext_next(ifp, icur);
4898 		xfs_iext_insert(ip, icur, &new, state);
4899 
4900 		da_new = got_indlen + new_indlen - stolen;
4901 		del->br_blockcount -= stolen;
4902 		break;
4903 	}
4904 
4905 	ASSERT(da_old >= da_new);
4906 	da_diff = da_old - da_new;
4907 	if (!isrt)
4908 		da_diff += del->br_blockcount;
4909 	if (da_diff) {
4910 		xfs_mod_fdblocks(mp, da_diff, false);
4911 		xfs_mod_delalloc(mp, -da_diff);
4912 	}
4913 	return error;
4914 }
4915 
4916 void
4917 xfs_bmap_del_extent_cow(
4918 	struct xfs_inode	*ip,
4919 	struct xfs_iext_cursor	*icur,
4920 	struct xfs_bmbt_irec	*got,
4921 	struct xfs_bmbt_irec	*del)
4922 {
4923 	struct xfs_mount	*mp = ip->i_mount;
4924 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
4925 	struct xfs_bmbt_irec	new;
4926 	xfs_fileoff_t		del_endoff, got_endoff;
4927 	int			state = BMAP_COWFORK;
4928 
4929 	XFS_STATS_INC(mp, xs_del_exlist);
4930 
4931 	del_endoff = del->br_startoff + del->br_blockcount;
4932 	got_endoff = got->br_startoff + got->br_blockcount;
4933 
4934 	ASSERT(del->br_blockcount > 0);
4935 	ASSERT(got->br_startoff <= del->br_startoff);
4936 	ASSERT(got_endoff >= del_endoff);
4937 	ASSERT(!isnullstartblock(got->br_startblock));
4938 
4939 	if (got->br_startoff == del->br_startoff)
4940 		state |= BMAP_LEFT_FILLING;
4941 	if (got_endoff == del_endoff)
4942 		state |= BMAP_RIGHT_FILLING;
4943 
4944 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4945 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4946 		/*
4947 		 * Matches the whole extent.  Delete the entry.
4948 		 */
4949 		xfs_iext_remove(ip, icur, state);
4950 		xfs_iext_prev(ifp, icur);
4951 		break;
4952 	case BMAP_LEFT_FILLING:
4953 		/*
4954 		 * Deleting the first part of the extent.
4955 		 */
4956 		got->br_startoff = del_endoff;
4957 		got->br_blockcount -= del->br_blockcount;
4958 		got->br_startblock = del->br_startblock + del->br_blockcount;
4959 		xfs_iext_update_extent(ip, state, icur, got);
4960 		break;
4961 	case BMAP_RIGHT_FILLING:
4962 		/*
4963 		 * Deleting the last part of the extent.
4964 		 */
4965 		got->br_blockcount -= del->br_blockcount;
4966 		xfs_iext_update_extent(ip, state, icur, got);
4967 		break;
4968 	case 0:
4969 		/*
4970 		 * Deleting the middle of the extent.
4971 		 */
4972 		got->br_blockcount = del->br_startoff - got->br_startoff;
4973 
4974 		new.br_startoff = del_endoff;
4975 		new.br_blockcount = got_endoff - del_endoff;
4976 		new.br_state = got->br_state;
4977 		new.br_startblock = del->br_startblock + del->br_blockcount;
4978 
4979 		xfs_iext_update_extent(ip, state, icur, got);
4980 		xfs_iext_next(ifp, icur);
4981 		xfs_iext_insert(ip, icur, &new, state);
4982 		break;
4983 	}
4984 	ip->i_delayed_blks -= del->br_blockcount;
4985 }
4986 
4987 /*
4988  * Called by xfs_bmapi to update file extent records and the btree
4989  * after removing space.
4990  */
4991 STATIC int				/* error */
4992 xfs_bmap_del_extent_real(
4993 	xfs_inode_t		*ip,	/* incore inode pointer */
4994 	xfs_trans_t		*tp,	/* current transaction pointer */
4995 	struct xfs_iext_cursor	*icur,
4996 	xfs_btree_cur_t		*cur,	/* if null, not a btree */
4997 	xfs_bmbt_irec_t		*del,	/* data to remove from extents */
4998 	int			*logflagsp, /* inode logging flags */
4999 	int			whichfork, /* data or attr fork */
5000 	int			bflags)	/* bmapi flags */
5001 {
5002 	xfs_fsblock_t		del_endblock=0;	/* first block past del */
5003 	xfs_fileoff_t		del_endoff;	/* first offset past del */
5004 	int			do_fx;	/* free extent at end of routine */
5005 	int			error;	/* error return value */
5006 	int			flags = 0;/* inode logging flags */
5007 	struct xfs_bmbt_irec	got;	/* current extent entry */
5008 	xfs_fileoff_t		got_endoff;	/* first offset past got */
5009 	int			i;	/* temp state */
5010 	struct xfs_ifork	*ifp;	/* inode fork pointer */
5011 	xfs_mount_t		*mp;	/* mount structure */
5012 	xfs_filblks_t		nblks;	/* quota/sb block count */
5013 	xfs_bmbt_irec_t		new;	/* new record to be inserted */
5014 	/* REFERENCED */
5015 	uint			qfield;	/* quota field to update */
5016 	int			state = xfs_bmap_fork_to_state(whichfork);
5017 	struct xfs_bmbt_irec	old;
5018 
5019 	mp = ip->i_mount;
5020 	XFS_STATS_INC(mp, xs_del_exlist);
5021 
5022 	ifp = XFS_IFORK_PTR(ip, whichfork);
5023 	ASSERT(del->br_blockcount > 0);
5024 	xfs_iext_get_extent(ifp, icur, &got);
5025 	ASSERT(got.br_startoff <= del->br_startoff);
5026 	del_endoff = del->br_startoff + del->br_blockcount;
5027 	got_endoff = got.br_startoff + got.br_blockcount;
5028 	ASSERT(got_endoff >= del_endoff);
5029 	ASSERT(!isnullstartblock(got.br_startblock));
5030 	qfield = 0;
5031 	error = 0;
5032 
5033 	/*
5034 	 * If it's the case where the directory code is running with no block
5035 	 * reservation, and the deleted block is in the middle of its extent,
5036 	 * and the resulting insert of an extent would cause transformation to
5037 	 * btree format, then reject it.  The calling code will then swap blocks
5038 	 * around instead.  We have to do this now, rather than waiting for the
5039 	 * conversion to btree format, since the transaction will be dirty then.
5040 	 */
5041 	if (tp->t_blk_res == 0 &&
5042 	    ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
5043 	    ifp->if_nextents >= XFS_IFORK_MAXEXT(ip, whichfork) &&
5044 	    del->br_startoff > got.br_startoff && del_endoff < got_endoff)
5045 		return -ENOSPC;
5046 
5047 	flags = XFS_ILOG_CORE;
5048 	if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
5049 		xfs_fsblock_t	bno;
5050 		xfs_filblks_t	len;
5051 		xfs_extlen_t	mod;
5052 
5053 		bno = div_u64_rem(del->br_startblock, mp->m_sb.sb_rextsize,
5054 				  &mod);
5055 		ASSERT(mod == 0);
5056 		len = div_u64_rem(del->br_blockcount, mp->m_sb.sb_rextsize,
5057 				  &mod);
5058 		ASSERT(mod == 0);
5059 
5060 		error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
5061 		if (error)
5062 			goto done;
5063 		do_fx = 0;
5064 		nblks = len * mp->m_sb.sb_rextsize;
5065 		qfield = XFS_TRANS_DQ_RTBCOUNT;
5066 	} else {
5067 		do_fx = 1;
5068 		nblks = del->br_blockcount;
5069 		qfield = XFS_TRANS_DQ_BCOUNT;
5070 	}
5071 
5072 	del_endblock = del->br_startblock + del->br_blockcount;
5073 	if (cur) {
5074 		error = xfs_bmbt_lookup_eq(cur, &got, &i);
5075 		if (error)
5076 			goto done;
5077 		if (XFS_IS_CORRUPT(mp, i != 1)) {
5078 			error = -EFSCORRUPTED;
5079 			goto done;
5080 		}
5081 	}
5082 
5083 	if (got.br_startoff == del->br_startoff)
5084 		state |= BMAP_LEFT_FILLING;
5085 	if (got_endoff == del_endoff)
5086 		state |= BMAP_RIGHT_FILLING;
5087 
5088 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
5089 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
5090 		/*
5091 		 * Matches the whole extent.  Delete the entry.
5092 		 */
5093 		xfs_iext_remove(ip, icur, state);
5094 		xfs_iext_prev(ifp, icur);
5095 		ifp->if_nextents--;
5096 
5097 		flags |= XFS_ILOG_CORE;
5098 		if (!cur) {
5099 			flags |= xfs_ilog_fext(whichfork);
5100 			break;
5101 		}
5102 		if ((error = xfs_btree_delete(cur, &i)))
5103 			goto done;
5104 		if (XFS_IS_CORRUPT(mp, i != 1)) {
5105 			error = -EFSCORRUPTED;
5106 			goto done;
5107 		}
5108 		break;
5109 	case BMAP_LEFT_FILLING:
5110 		/*
5111 		 * Deleting the first part of the extent.
5112 		 */
5113 		got.br_startoff = del_endoff;
5114 		got.br_startblock = del_endblock;
5115 		got.br_blockcount -= del->br_blockcount;
5116 		xfs_iext_update_extent(ip, state, icur, &got);
5117 		if (!cur) {
5118 			flags |= xfs_ilog_fext(whichfork);
5119 			break;
5120 		}
5121 		error = xfs_bmbt_update(cur, &got);
5122 		if (error)
5123 			goto done;
5124 		break;
5125 	case BMAP_RIGHT_FILLING:
5126 		/*
5127 		 * Deleting the last part of the extent.
5128 		 */
5129 		got.br_blockcount -= del->br_blockcount;
5130 		xfs_iext_update_extent(ip, state, icur, &got);
5131 		if (!cur) {
5132 			flags |= xfs_ilog_fext(whichfork);
5133 			break;
5134 		}
5135 		error = xfs_bmbt_update(cur, &got);
5136 		if (error)
5137 			goto done;
5138 		break;
5139 	case 0:
5140 		/*
5141 		 * Deleting the middle of the extent.
5142 		 */
5143 		old = got;
5144 
5145 		got.br_blockcount = del->br_startoff - got.br_startoff;
5146 		xfs_iext_update_extent(ip, state, icur, &got);
5147 
5148 		new.br_startoff = del_endoff;
5149 		new.br_blockcount = got_endoff - del_endoff;
5150 		new.br_state = got.br_state;
5151 		new.br_startblock = del_endblock;
5152 
5153 		flags |= XFS_ILOG_CORE;
5154 		if (cur) {
5155 			error = xfs_bmbt_update(cur, &got);
5156 			if (error)
5157 				goto done;
5158 			error = xfs_btree_increment(cur, 0, &i);
5159 			if (error)
5160 				goto done;
5161 			cur->bc_rec.b = new;
5162 			error = xfs_btree_insert(cur, &i);
5163 			if (error && error != -ENOSPC)
5164 				goto done;
5165 			/*
5166 			 * If get no-space back from btree insert, it tried a
5167 			 * split, and we have a zero block reservation.  Fix up
5168 			 * our state and return the error.
5169 			 */
5170 			if (error == -ENOSPC) {
5171 				/*
5172 				 * Reset the cursor, don't trust it after any
5173 				 * insert operation.
5174 				 */
5175 				error = xfs_bmbt_lookup_eq(cur, &got, &i);
5176 				if (error)
5177 					goto done;
5178 				if (XFS_IS_CORRUPT(mp, i != 1)) {
5179 					error = -EFSCORRUPTED;
5180 					goto done;
5181 				}
5182 				/*
5183 				 * Update the btree record back
5184 				 * to the original value.
5185 				 */
5186 				error = xfs_bmbt_update(cur, &old);
5187 				if (error)
5188 					goto done;
5189 				/*
5190 				 * Reset the extent record back
5191 				 * to the original value.
5192 				 */
5193 				xfs_iext_update_extent(ip, state, icur, &old);
5194 				flags = 0;
5195 				error = -ENOSPC;
5196 				goto done;
5197 			}
5198 			if (XFS_IS_CORRUPT(mp, i != 1)) {
5199 				error = -EFSCORRUPTED;
5200 				goto done;
5201 			}
5202 		} else
5203 			flags |= xfs_ilog_fext(whichfork);
5204 
5205 		ifp->if_nextents++;
5206 		xfs_iext_next(ifp, icur);
5207 		xfs_iext_insert(ip, icur, &new, state);
5208 		break;
5209 	}
5210 
5211 	/* remove reverse mapping */
5212 	xfs_rmap_unmap_extent(tp, ip, whichfork, del);
5213 
5214 	/*
5215 	 * If we need to, add to list of extents to delete.
5216 	 */
5217 	if (do_fx && !(bflags & XFS_BMAPI_REMAP)) {
5218 		if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
5219 			xfs_refcount_decrease_extent(tp, del);
5220 		} else {
5221 			__xfs_bmap_add_free(tp, del->br_startblock,
5222 					del->br_blockcount, NULL,
5223 					(bflags & XFS_BMAPI_NODISCARD) ||
5224 					del->br_state == XFS_EXT_UNWRITTEN);
5225 		}
5226 	}
5227 
5228 	/*
5229 	 * Adjust inode # blocks in the file.
5230 	 */
5231 	if (nblks)
5232 		ip->i_d.di_nblocks -= nblks;
5233 	/*
5234 	 * Adjust quota data.
5235 	 */
5236 	if (qfield && !(bflags & XFS_BMAPI_REMAP))
5237 		xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5238 
5239 done:
5240 	*logflagsp = flags;
5241 	return error;
5242 }
5243 
5244 /*
5245  * Unmap (remove) blocks from a file.
5246  * If nexts is nonzero then the number of extents to remove is limited to
5247  * that value.  If not all extents in the block range can be removed then
5248  * *done is set.
5249  */
5250 int						/* error */
5251 __xfs_bunmapi(
5252 	struct xfs_trans	*tp,		/* transaction pointer */
5253 	struct xfs_inode	*ip,		/* incore inode */
5254 	xfs_fileoff_t		start,		/* first file offset deleted */
5255 	xfs_filblks_t		*rlen,		/* i/o: amount remaining */
5256 	int			flags,		/* misc flags */
5257 	xfs_extnum_t		nexts)		/* number of extents max */
5258 {
5259 	struct xfs_btree_cur	*cur;		/* bmap btree cursor */
5260 	struct xfs_bmbt_irec	del;		/* extent being deleted */
5261 	int			error;		/* error return value */
5262 	xfs_extnum_t		extno;		/* extent number in list */
5263 	struct xfs_bmbt_irec	got;		/* current extent record */
5264 	struct xfs_ifork	*ifp;		/* inode fork pointer */
5265 	int			isrt;		/* freeing in rt area */
5266 	int			logflags;	/* transaction logging flags */
5267 	xfs_extlen_t		mod;		/* rt extent offset */
5268 	struct xfs_mount	*mp = ip->i_mount;
5269 	int			tmp_logflags;	/* partial logging flags */
5270 	int			wasdel;		/* was a delayed alloc extent */
5271 	int			whichfork;	/* data or attribute fork */
5272 	xfs_fsblock_t		sum;
5273 	xfs_filblks_t		len = *rlen;	/* length to unmap in file */
5274 	xfs_fileoff_t		max_len;
5275 	xfs_agnumber_t		prev_agno = NULLAGNUMBER, agno;
5276 	xfs_fileoff_t		end;
5277 	struct xfs_iext_cursor	icur;
5278 	bool			done = false;
5279 
5280 	trace_xfs_bunmap(ip, start, len, flags, _RET_IP_);
5281 
5282 	whichfork = xfs_bmapi_whichfork(flags);
5283 	ASSERT(whichfork != XFS_COW_FORK);
5284 	ifp = XFS_IFORK_PTR(ip, whichfork);
5285 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)))
5286 		return -EFSCORRUPTED;
5287 	if (XFS_FORCED_SHUTDOWN(mp))
5288 		return -EIO;
5289 
5290 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5291 	ASSERT(len > 0);
5292 	ASSERT(nexts >= 0);
5293 
5294 	/*
5295 	 * Guesstimate how many blocks we can unmap without running the risk of
5296 	 * blowing out the transaction with a mix of EFIs and reflink
5297 	 * adjustments.
5298 	 */
5299 	if (tp && xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK)
5300 		max_len = min(len, xfs_refcount_max_unmap(tp->t_log_res));
5301 	else
5302 		max_len = len;
5303 
5304 	if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5305 	    (error = xfs_iread_extents(tp, ip, whichfork)))
5306 		return error;
5307 	if (xfs_iext_count(ifp) == 0) {
5308 		*rlen = 0;
5309 		return 0;
5310 	}
5311 	XFS_STATS_INC(mp, xs_blk_unmap);
5312 	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5313 	end = start + len;
5314 
5315 	if (!xfs_iext_lookup_extent_before(ip, ifp, &end, &icur, &got)) {
5316 		*rlen = 0;
5317 		return 0;
5318 	}
5319 	end--;
5320 
5321 	logflags = 0;
5322 	if (ifp->if_flags & XFS_IFBROOT) {
5323 		ASSERT(ifp->if_format == XFS_DINODE_FMT_BTREE);
5324 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5325 		cur->bc_ino.flags = 0;
5326 	} else
5327 		cur = NULL;
5328 
5329 	if (isrt) {
5330 		/*
5331 		 * Synchronize by locking the bitmap inode.
5332 		 */
5333 		xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
5334 		xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5335 		xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
5336 		xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
5337 	}
5338 
5339 	extno = 0;
5340 	while (end != (xfs_fileoff_t)-1 && end >= start &&
5341 	       (nexts == 0 || extno < nexts) && max_len > 0) {
5342 		/*
5343 		 * Is the found extent after a hole in which end lives?
5344 		 * Just back up to the previous extent, if so.
5345 		 */
5346 		if (got.br_startoff > end &&
5347 		    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5348 			done = true;
5349 			break;
5350 		}
5351 		/*
5352 		 * Is the last block of this extent before the range
5353 		 * we're supposed to delete?  If so, we're done.
5354 		 */
5355 		end = XFS_FILEOFF_MIN(end,
5356 			got.br_startoff + got.br_blockcount - 1);
5357 		if (end < start)
5358 			break;
5359 		/*
5360 		 * Then deal with the (possibly delayed) allocated space
5361 		 * we found.
5362 		 */
5363 		del = got;
5364 		wasdel = isnullstartblock(del.br_startblock);
5365 
5366 		/*
5367 		 * Make sure we don't touch multiple AGF headers out of order
5368 		 * in a single transaction, as that could cause AB-BA deadlocks.
5369 		 */
5370 		if (!wasdel && !isrt) {
5371 			agno = XFS_FSB_TO_AGNO(mp, del.br_startblock);
5372 			if (prev_agno != NULLAGNUMBER && prev_agno > agno)
5373 				break;
5374 			prev_agno = agno;
5375 		}
5376 		if (got.br_startoff < start) {
5377 			del.br_startoff = start;
5378 			del.br_blockcount -= start - got.br_startoff;
5379 			if (!wasdel)
5380 				del.br_startblock += start - got.br_startoff;
5381 		}
5382 		if (del.br_startoff + del.br_blockcount > end + 1)
5383 			del.br_blockcount = end + 1 - del.br_startoff;
5384 
5385 		/* How much can we safely unmap? */
5386 		if (max_len < del.br_blockcount) {
5387 			del.br_startoff += del.br_blockcount - max_len;
5388 			if (!wasdel)
5389 				del.br_startblock += del.br_blockcount - max_len;
5390 			del.br_blockcount = max_len;
5391 		}
5392 
5393 		if (!isrt)
5394 			goto delete;
5395 
5396 		sum = del.br_startblock + del.br_blockcount;
5397 		div_u64_rem(sum, mp->m_sb.sb_rextsize, &mod);
5398 		if (mod) {
5399 			/*
5400 			 * Realtime extent not lined up at the end.
5401 			 * The extent could have been split into written
5402 			 * and unwritten pieces, or we could just be
5403 			 * unmapping part of it.  But we can't really
5404 			 * get rid of part of a realtime extent.
5405 			 */
5406 			if (del.br_state == XFS_EXT_UNWRITTEN) {
5407 				/*
5408 				 * This piece is unwritten, or we're not
5409 				 * using unwritten extents.  Skip over it.
5410 				 */
5411 				ASSERT(end >= mod);
5412 				end -= mod > del.br_blockcount ?
5413 					del.br_blockcount : mod;
5414 				if (end < got.br_startoff &&
5415 				    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5416 					done = true;
5417 					break;
5418 				}
5419 				continue;
5420 			}
5421 			/*
5422 			 * It's written, turn it unwritten.
5423 			 * This is better than zeroing it.
5424 			 */
5425 			ASSERT(del.br_state == XFS_EXT_NORM);
5426 			ASSERT(tp->t_blk_res > 0);
5427 			/*
5428 			 * If this spans a realtime extent boundary,
5429 			 * chop it back to the start of the one we end at.
5430 			 */
5431 			if (del.br_blockcount > mod) {
5432 				del.br_startoff += del.br_blockcount - mod;
5433 				del.br_startblock += del.br_blockcount - mod;
5434 				del.br_blockcount = mod;
5435 			}
5436 			del.br_state = XFS_EXT_UNWRITTEN;
5437 			error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5438 					whichfork, &icur, &cur, &del,
5439 					&logflags);
5440 			if (error)
5441 				goto error0;
5442 			goto nodelete;
5443 		}
5444 		div_u64_rem(del.br_startblock, mp->m_sb.sb_rextsize, &mod);
5445 		if (mod) {
5446 			xfs_extlen_t off = mp->m_sb.sb_rextsize - mod;
5447 
5448 			/*
5449 			 * Realtime extent is lined up at the end but not
5450 			 * at the front.  We'll get rid of full extents if
5451 			 * we can.
5452 			 */
5453 			if (del.br_blockcount > off) {
5454 				del.br_blockcount -= off;
5455 				del.br_startoff += off;
5456 				del.br_startblock += off;
5457 			} else if (del.br_startoff == start &&
5458 				   (del.br_state == XFS_EXT_UNWRITTEN ||
5459 				    tp->t_blk_res == 0)) {
5460 				/*
5461 				 * Can't make it unwritten.  There isn't
5462 				 * a full extent here so just skip it.
5463 				 */
5464 				ASSERT(end >= del.br_blockcount);
5465 				end -= del.br_blockcount;
5466 				if (got.br_startoff > end &&
5467 				    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5468 					done = true;
5469 					break;
5470 				}
5471 				continue;
5472 			} else if (del.br_state == XFS_EXT_UNWRITTEN) {
5473 				struct xfs_bmbt_irec	prev;
5474 				xfs_fileoff_t		unwrite_start;
5475 
5476 				/*
5477 				 * This one is already unwritten.
5478 				 * It must have a written left neighbor.
5479 				 * Unwrite the killed part of that one and
5480 				 * try again.
5481 				 */
5482 				if (!xfs_iext_prev_extent(ifp, &icur, &prev))
5483 					ASSERT(0);
5484 				ASSERT(prev.br_state == XFS_EXT_NORM);
5485 				ASSERT(!isnullstartblock(prev.br_startblock));
5486 				ASSERT(del.br_startblock ==
5487 				       prev.br_startblock + prev.br_blockcount);
5488 				unwrite_start = max3(start,
5489 						     del.br_startoff - mod,
5490 						     prev.br_startoff);
5491 				mod = unwrite_start - prev.br_startoff;
5492 				prev.br_startoff = unwrite_start;
5493 				prev.br_startblock += mod;
5494 				prev.br_blockcount -= mod;
5495 				prev.br_state = XFS_EXT_UNWRITTEN;
5496 				error = xfs_bmap_add_extent_unwritten_real(tp,
5497 						ip, whichfork, &icur, &cur,
5498 						&prev, &logflags);
5499 				if (error)
5500 					goto error0;
5501 				goto nodelete;
5502 			} else {
5503 				ASSERT(del.br_state == XFS_EXT_NORM);
5504 				del.br_state = XFS_EXT_UNWRITTEN;
5505 				error = xfs_bmap_add_extent_unwritten_real(tp,
5506 						ip, whichfork, &icur, &cur,
5507 						&del, &logflags);
5508 				if (error)
5509 					goto error0;
5510 				goto nodelete;
5511 			}
5512 		}
5513 
5514 delete:
5515 		if (wasdel) {
5516 			error = xfs_bmap_del_extent_delay(ip, whichfork, &icur,
5517 					&got, &del);
5518 		} else {
5519 			error = xfs_bmap_del_extent_real(ip, tp, &icur, cur,
5520 					&del, &tmp_logflags, whichfork,
5521 					flags);
5522 			logflags |= tmp_logflags;
5523 		}
5524 
5525 		if (error)
5526 			goto error0;
5527 
5528 		max_len -= del.br_blockcount;
5529 		end = del.br_startoff - 1;
5530 nodelete:
5531 		/*
5532 		 * If not done go on to the next (previous) record.
5533 		 */
5534 		if (end != (xfs_fileoff_t)-1 && end >= start) {
5535 			if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5536 			    (got.br_startoff > end &&
5537 			     !xfs_iext_prev_extent(ifp, &icur, &got))) {
5538 				done = true;
5539 				break;
5540 			}
5541 			extno++;
5542 		}
5543 	}
5544 	if (done || end == (xfs_fileoff_t)-1 || end < start)
5545 		*rlen = 0;
5546 	else
5547 		*rlen = end - start + 1;
5548 
5549 	/*
5550 	 * Convert to a btree if necessary.
5551 	 */
5552 	if (xfs_bmap_needs_btree(ip, whichfork)) {
5553 		ASSERT(cur == NULL);
5554 		error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
5555 				&tmp_logflags, whichfork);
5556 		logflags |= tmp_logflags;
5557 	} else {
5558 		error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags,
5559 			whichfork);
5560 	}
5561 
5562 error0:
5563 	/*
5564 	 * Log everything.  Do this after conversion, there's no point in
5565 	 * logging the extent records if we've converted to btree format.
5566 	 */
5567 	if ((logflags & xfs_ilog_fext(whichfork)) &&
5568 	    ifp->if_format != XFS_DINODE_FMT_EXTENTS)
5569 		logflags &= ~xfs_ilog_fext(whichfork);
5570 	else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5571 		 ifp->if_format != XFS_DINODE_FMT_BTREE)
5572 		logflags &= ~xfs_ilog_fbroot(whichfork);
5573 	/*
5574 	 * Log inode even in the error case, if the transaction
5575 	 * is dirty we'll need to shut down the filesystem.
5576 	 */
5577 	if (logflags)
5578 		xfs_trans_log_inode(tp, ip, logflags);
5579 	if (cur) {
5580 		if (!error)
5581 			cur->bc_ino.allocated = 0;
5582 		xfs_btree_del_cursor(cur, error);
5583 	}
5584 	return error;
5585 }
5586 
5587 /* Unmap a range of a file. */
5588 int
5589 xfs_bunmapi(
5590 	xfs_trans_t		*tp,
5591 	struct xfs_inode	*ip,
5592 	xfs_fileoff_t		bno,
5593 	xfs_filblks_t		len,
5594 	int			flags,
5595 	xfs_extnum_t		nexts,
5596 	int			*done)
5597 {
5598 	int			error;
5599 
5600 	error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts);
5601 	*done = (len == 0);
5602 	return error;
5603 }
5604 
5605 /*
5606  * Determine whether an extent shift can be accomplished by a merge with the
5607  * extent that precedes the target hole of the shift.
5608  */
5609 STATIC bool
5610 xfs_bmse_can_merge(
5611 	struct xfs_bmbt_irec	*left,	/* preceding extent */
5612 	struct xfs_bmbt_irec	*got,	/* current extent to shift */
5613 	xfs_fileoff_t		shift)	/* shift fsb */
5614 {
5615 	xfs_fileoff_t		startoff;
5616 
5617 	startoff = got->br_startoff - shift;
5618 
5619 	/*
5620 	 * The extent, once shifted, must be adjacent in-file and on-disk with
5621 	 * the preceding extent.
5622 	 */
5623 	if ((left->br_startoff + left->br_blockcount != startoff) ||
5624 	    (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5625 	    (left->br_state != got->br_state) ||
5626 	    (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5627 		return false;
5628 
5629 	return true;
5630 }
5631 
5632 /*
5633  * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5634  * hole in the file. If an extent shift would result in the extent being fully
5635  * adjacent to the extent that currently precedes the hole, we can merge with
5636  * the preceding extent rather than do the shift.
5637  *
5638  * This function assumes the caller has verified a shift-by-merge is possible
5639  * with the provided extents via xfs_bmse_can_merge().
5640  */
5641 STATIC int
5642 xfs_bmse_merge(
5643 	struct xfs_trans		*tp,
5644 	struct xfs_inode		*ip,
5645 	int				whichfork,
5646 	xfs_fileoff_t			shift,		/* shift fsb */
5647 	struct xfs_iext_cursor		*icur,
5648 	struct xfs_bmbt_irec		*got,		/* extent to shift */
5649 	struct xfs_bmbt_irec		*left,		/* preceding extent */
5650 	struct xfs_btree_cur		*cur,
5651 	int				*logflags)	/* output */
5652 {
5653 	struct xfs_ifork		*ifp = XFS_IFORK_PTR(ip, whichfork);
5654 	struct xfs_bmbt_irec		new;
5655 	xfs_filblks_t			blockcount;
5656 	int				error, i;
5657 	struct xfs_mount		*mp = ip->i_mount;
5658 
5659 	blockcount = left->br_blockcount + got->br_blockcount;
5660 
5661 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5662 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5663 	ASSERT(xfs_bmse_can_merge(left, got, shift));
5664 
5665 	new = *left;
5666 	new.br_blockcount = blockcount;
5667 
5668 	/*
5669 	 * Update the on-disk extent count, the btree if necessary and log the
5670 	 * inode.
5671 	 */
5672 	ifp->if_nextents--;
5673 	*logflags |= XFS_ILOG_CORE;
5674 	if (!cur) {
5675 		*logflags |= XFS_ILOG_DEXT;
5676 		goto done;
5677 	}
5678 
5679 	/* lookup and remove the extent to merge */
5680 	error = xfs_bmbt_lookup_eq(cur, got, &i);
5681 	if (error)
5682 		return error;
5683 	if (XFS_IS_CORRUPT(mp, i != 1))
5684 		return -EFSCORRUPTED;
5685 
5686 	error = xfs_btree_delete(cur, &i);
5687 	if (error)
5688 		return error;
5689 	if (XFS_IS_CORRUPT(mp, i != 1))
5690 		return -EFSCORRUPTED;
5691 
5692 	/* lookup and update size of the previous extent */
5693 	error = xfs_bmbt_lookup_eq(cur, left, &i);
5694 	if (error)
5695 		return error;
5696 	if (XFS_IS_CORRUPT(mp, i != 1))
5697 		return -EFSCORRUPTED;
5698 
5699 	error = xfs_bmbt_update(cur, &new);
5700 	if (error)
5701 		return error;
5702 
5703 	/* change to extent format if required after extent removal */
5704 	error = xfs_bmap_btree_to_extents(tp, ip, cur, logflags, whichfork);
5705 	if (error)
5706 		return error;
5707 
5708 done:
5709 	xfs_iext_remove(ip, icur, 0);
5710 	xfs_iext_prev(ifp, icur);
5711 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5712 			&new);
5713 
5714 	/* update reverse mapping. rmap functions merge the rmaps for us */
5715 	xfs_rmap_unmap_extent(tp, ip, whichfork, got);
5716 	memcpy(&new, got, sizeof(new));
5717 	new.br_startoff = left->br_startoff + left->br_blockcount;
5718 	xfs_rmap_map_extent(tp, ip, whichfork, &new);
5719 	return 0;
5720 }
5721 
5722 static int
5723 xfs_bmap_shift_update_extent(
5724 	struct xfs_trans	*tp,
5725 	struct xfs_inode	*ip,
5726 	int			whichfork,
5727 	struct xfs_iext_cursor	*icur,
5728 	struct xfs_bmbt_irec	*got,
5729 	struct xfs_btree_cur	*cur,
5730 	int			*logflags,
5731 	xfs_fileoff_t		startoff)
5732 {
5733 	struct xfs_mount	*mp = ip->i_mount;
5734 	struct xfs_bmbt_irec	prev = *got;
5735 	int			error, i;
5736 
5737 	*logflags |= XFS_ILOG_CORE;
5738 
5739 	got->br_startoff = startoff;
5740 
5741 	if (cur) {
5742 		error = xfs_bmbt_lookup_eq(cur, &prev, &i);
5743 		if (error)
5744 			return error;
5745 		if (XFS_IS_CORRUPT(mp, i != 1))
5746 			return -EFSCORRUPTED;
5747 
5748 		error = xfs_bmbt_update(cur, got);
5749 		if (error)
5750 			return error;
5751 	} else {
5752 		*logflags |= XFS_ILOG_DEXT;
5753 	}
5754 
5755 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5756 			got);
5757 
5758 	/* update reverse mapping */
5759 	xfs_rmap_unmap_extent(tp, ip, whichfork, &prev);
5760 	xfs_rmap_map_extent(tp, ip, whichfork, got);
5761 	return 0;
5762 }
5763 
5764 int
5765 xfs_bmap_collapse_extents(
5766 	struct xfs_trans	*tp,
5767 	struct xfs_inode	*ip,
5768 	xfs_fileoff_t		*next_fsb,
5769 	xfs_fileoff_t		offset_shift_fsb,
5770 	bool			*done)
5771 {
5772 	int			whichfork = XFS_DATA_FORK;
5773 	struct xfs_mount	*mp = ip->i_mount;
5774 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
5775 	struct xfs_btree_cur	*cur = NULL;
5776 	struct xfs_bmbt_irec	got, prev;
5777 	struct xfs_iext_cursor	icur;
5778 	xfs_fileoff_t		new_startoff;
5779 	int			error = 0;
5780 	int			logflags = 0;
5781 
5782 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
5783 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
5784 		return -EFSCORRUPTED;
5785 	}
5786 
5787 	if (XFS_FORCED_SHUTDOWN(mp))
5788 		return -EIO;
5789 
5790 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL));
5791 
5792 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5793 		error = xfs_iread_extents(tp, ip, whichfork);
5794 		if (error)
5795 			return error;
5796 	}
5797 
5798 	if (ifp->if_flags & XFS_IFBROOT) {
5799 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5800 		cur->bc_ino.flags = 0;
5801 	}
5802 
5803 	if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5804 		*done = true;
5805 		goto del_cursor;
5806 	}
5807 	if (XFS_IS_CORRUPT(mp, isnullstartblock(got.br_startblock))) {
5808 		error = -EFSCORRUPTED;
5809 		goto del_cursor;
5810 	}
5811 
5812 	new_startoff = got.br_startoff - offset_shift_fsb;
5813 	if (xfs_iext_peek_prev_extent(ifp, &icur, &prev)) {
5814 		if (new_startoff < prev.br_startoff + prev.br_blockcount) {
5815 			error = -EINVAL;
5816 			goto del_cursor;
5817 		}
5818 
5819 		if (xfs_bmse_can_merge(&prev, &got, offset_shift_fsb)) {
5820 			error = xfs_bmse_merge(tp, ip, whichfork,
5821 					offset_shift_fsb, &icur, &got, &prev,
5822 					cur, &logflags);
5823 			if (error)
5824 				goto del_cursor;
5825 			goto done;
5826 		}
5827 	} else {
5828 		if (got.br_startoff < offset_shift_fsb) {
5829 			error = -EINVAL;
5830 			goto del_cursor;
5831 		}
5832 	}
5833 
5834 	error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got,
5835 			cur, &logflags, new_startoff);
5836 	if (error)
5837 		goto del_cursor;
5838 
5839 done:
5840 	if (!xfs_iext_next_extent(ifp, &icur, &got)) {
5841 		*done = true;
5842 		goto del_cursor;
5843 	}
5844 
5845 	*next_fsb = got.br_startoff;
5846 del_cursor:
5847 	if (cur)
5848 		xfs_btree_del_cursor(cur, error);
5849 	if (logflags)
5850 		xfs_trans_log_inode(tp, ip, logflags);
5851 	return error;
5852 }
5853 
5854 /* Make sure we won't be right-shifting an extent past the maximum bound. */
5855 int
5856 xfs_bmap_can_insert_extents(
5857 	struct xfs_inode	*ip,
5858 	xfs_fileoff_t		off,
5859 	xfs_fileoff_t		shift)
5860 {
5861 	struct xfs_bmbt_irec	got;
5862 	int			is_empty;
5863 	int			error = 0;
5864 
5865 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5866 
5867 	if (XFS_FORCED_SHUTDOWN(ip->i_mount))
5868 		return -EIO;
5869 
5870 	xfs_ilock(ip, XFS_ILOCK_EXCL);
5871 	error = xfs_bmap_last_extent(NULL, ip, XFS_DATA_FORK, &got, &is_empty);
5872 	if (!error && !is_empty && got.br_startoff >= off &&
5873 	    ((got.br_startoff + shift) & BMBT_STARTOFF_MASK) < got.br_startoff)
5874 		error = -EINVAL;
5875 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
5876 
5877 	return error;
5878 }
5879 
5880 int
5881 xfs_bmap_insert_extents(
5882 	struct xfs_trans	*tp,
5883 	struct xfs_inode	*ip,
5884 	xfs_fileoff_t		*next_fsb,
5885 	xfs_fileoff_t		offset_shift_fsb,
5886 	bool			*done,
5887 	xfs_fileoff_t		stop_fsb)
5888 {
5889 	int			whichfork = XFS_DATA_FORK;
5890 	struct xfs_mount	*mp = ip->i_mount;
5891 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
5892 	struct xfs_btree_cur	*cur = NULL;
5893 	struct xfs_bmbt_irec	got, next;
5894 	struct xfs_iext_cursor	icur;
5895 	xfs_fileoff_t		new_startoff;
5896 	int			error = 0;
5897 	int			logflags = 0;
5898 
5899 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
5900 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
5901 		return -EFSCORRUPTED;
5902 	}
5903 
5904 	if (XFS_FORCED_SHUTDOWN(mp))
5905 		return -EIO;
5906 
5907 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL));
5908 
5909 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5910 		error = xfs_iread_extents(tp, ip, whichfork);
5911 		if (error)
5912 			return error;
5913 	}
5914 
5915 	if (ifp->if_flags & XFS_IFBROOT) {
5916 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5917 		cur->bc_ino.flags = 0;
5918 	}
5919 
5920 	if (*next_fsb == NULLFSBLOCK) {
5921 		xfs_iext_last(ifp, &icur);
5922 		if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5923 		    stop_fsb > got.br_startoff) {
5924 			*done = true;
5925 			goto del_cursor;
5926 		}
5927 	} else {
5928 		if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5929 			*done = true;
5930 			goto del_cursor;
5931 		}
5932 	}
5933 	if (XFS_IS_CORRUPT(mp, isnullstartblock(got.br_startblock))) {
5934 		error = -EFSCORRUPTED;
5935 		goto del_cursor;
5936 	}
5937 
5938 	if (XFS_IS_CORRUPT(mp, stop_fsb > got.br_startoff)) {
5939 		error = -EFSCORRUPTED;
5940 		goto del_cursor;
5941 	}
5942 
5943 	new_startoff = got.br_startoff + offset_shift_fsb;
5944 	if (xfs_iext_peek_next_extent(ifp, &icur, &next)) {
5945 		if (new_startoff + got.br_blockcount > next.br_startoff) {
5946 			error = -EINVAL;
5947 			goto del_cursor;
5948 		}
5949 
5950 		/*
5951 		 * Unlike a left shift (which involves a hole punch), a right
5952 		 * shift does not modify extent neighbors in any way.  We should
5953 		 * never find mergeable extents in this scenario.  Check anyways
5954 		 * and warn if we encounter two extents that could be one.
5955 		 */
5956 		if (xfs_bmse_can_merge(&got, &next, offset_shift_fsb))
5957 			WARN_ON_ONCE(1);
5958 	}
5959 
5960 	error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got,
5961 			cur, &logflags, new_startoff);
5962 	if (error)
5963 		goto del_cursor;
5964 
5965 	if (!xfs_iext_prev_extent(ifp, &icur, &got) ||
5966 	    stop_fsb >= got.br_startoff + got.br_blockcount) {
5967 		*done = true;
5968 		goto del_cursor;
5969 	}
5970 
5971 	*next_fsb = got.br_startoff;
5972 del_cursor:
5973 	if (cur)
5974 		xfs_btree_del_cursor(cur, error);
5975 	if (logflags)
5976 		xfs_trans_log_inode(tp, ip, logflags);
5977 	return error;
5978 }
5979 
5980 /*
5981  * Splits an extent into two extents at split_fsb block such that it is the
5982  * first block of the current_ext. @ext is a target extent to be split.
5983  * @split_fsb is a block where the extents is split.  If split_fsb lies in a
5984  * hole or the first block of extents, just return 0.
5985  */
5986 int
5987 xfs_bmap_split_extent(
5988 	struct xfs_trans	*tp,
5989 	struct xfs_inode	*ip,
5990 	xfs_fileoff_t		split_fsb)
5991 {
5992 	int				whichfork = XFS_DATA_FORK;
5993 	struct xfs_ifork		*ifp = XFS_IFORK_PTR(ip, whichfork);
5994 	struct xfs_btree_cur		*cur = NULL;
5995 	struct xfs_bmbt_irec		got;
5996 	struct xfs_bmbt_irec		new; /* split extent */
5997 	struct xfs_mount		*mp = ip->i_mount;
5998 	xfs_fsblock_t			gotblkcnt; /* new block count for got */
5999 	struct xfs_iext_cursor		icur;
6000 	int				error = 0;
6001 	int				logflags = 0;
6002 	int				i = 0;
6003 
6004 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
6005 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
6006 		return -EFSCORRUPTED;
6007 	}
6008 
6009 	if (XFS_FORCED_SHUTDOWN(mp))
6010 		return -EIO;
6011 
6012 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
6013 		/* Read in all the extents */
6014 		error = xfs_iread_extents(tp, ip, whichfork);
6015 		if (error)
6016 			return error;
6017 	}
6018 
6019 	/*
6020 	 * If there are not extents, or split_fsb lies in a hole we are done.
6021 	 */
6022 	if (!xfs_iext_lookup_extent(ip, ifp, split_fsb, &icur, &got) ||
6023 	    got.br_startoff >= split_fsb)
6024 		return 0;
6025 
6026 	gotblkcnt = split_fsb - got.br_startoff;
6027 	new.br_startoff = split_fsb;
6028 	new.br_startblock = got.br_startblock + gotblkcnt;
6029 	new.br_blockcount = got.br_blockcount - gotblkcnt;
6030 	new.br_state = got.br_state;
6031 
6032 	if (ifp->if_flags & XFS_IFBROOT) {
6033 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6034 		cur->bc_ino.flags = 0;
6035 		error = xfs_bmbt_lookup_eq(cur, &got, &i);
6036 		if (error)
6037 			goto del_cursor;
6038 		if (XFS_IS_CORRUPT(mp, i != 1)) {
6039 			error = -EFSCORRUPTED;
6040 			goto del_cursor;
6041 		}
6042 	}
6043 
6044 	got.br_blockcount = gotblkcnt;
6045 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), &icur,
6046 			&got);
6047 
6048 	logflags = XFS_ILOG_CORE;
6049 	if (cur) {
6050 		error = xfs_bmbt_update(cur, &got);
6051 		if (error)
6052 			goto del_cursor;
6053 	} else
6054 		logflags |= XFS_ILOG_DEXT;
6055 
6056 	/* Add new extent */
6057 	xfs_iext_next(ifp, &icur);
6058 	xfs_iext_insert(ip, &icur, &new, 0);
6059 	ifp->if_nextents++;
6060 
6061 	if (cur) {
6062 		error = xfs_bmbt_lookup_eq(cur, &new, &i);
6063 		if (error)
6064 			goto del_cursor;
6065 		if (XFS_IS_CORRUPT(mp, i != 0)) {
6066 			error = -EFSCORRUPTED;
6067 			goto del_cursor;
6068 		}
6069 		error = xfs_btree_insert(cur, &i);
6070 		if (error)
6071 			goto del_cursor;
6072 		if (XFS_IS_CORRUPT(mp, i != 1)) {
6073 			error = -EFSCORRUPTED;
6074 			goto del_cursor;
6075 		}
6076 	}
6077 
6078 	/*
6079 	 * Convert to a btree if necessary.
6080 	 */
6081 	if (xfs_bmap_needs_btree(ip, whichfork)) {
6082 		int tmp_logflags; /* partial log flag return val */
6083 
6084 		ASSERT(cur == NULL);
6085 		error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
6086 				&tmp_logflags, whichfork);
6087 		logflags |= tmp_logflags;
6088 	}
6089 
6090 del_cursor:
6091 	if (cur) {
6092 		cur->bc_ino.allocated = 0;
6093 		xfs_btree_del_cursor(cur, error);
6094 	}
6095 
6096 	if (logflags)
6097 		xfs_trans_log_inode(tp, ip, logflags);
6098 	return error;
6099 }
6100 
6101 /* Deferred mapping is only for real extents in the data fork. */
6102 static bool
6103 xfs_bmap_is_update_needed(
6104 	struct xfs_bmbt_irec	*bmap)
6105 {
6106 	return  bmap->br_startblock != HOLESTARTBLOCK &&
6107 		bmap->br_startblock != DELAYSTARTBLOCK;
6108 }
6109 
6110 /* Record a bmap intent. */
6111 static int
6112 __xfs_bmap_add(
6113 	struct xfs_trans		*tp,
6114 	enum xfs_bmap_intent_type	type,
6115 	struct xfs_inode		*ip,
6116 	int				whichfork,
6117 	struct xfs_bmbt_irec		*bmap)
6118 {
6119 	struct xfs_bmap_intent		*bi;
6120 
6121 	trace_xfs_bmap_defer(tp->t_mountp,
6122 			XFS_FSB_TO_AGNO(tp->t_mountp, bmap->br_startblock),
6123 			type,
6124 			XFS_FSB_TO_AGBNO(tp->t_mountp, bmap->br_startblock),
6125 			ip->i_ino, whichfork,
6126 			bmap->br_startoff,
6127 			bmap->br_blockcount,
6128 			bmap->br_state);
6129 
6130 	bi = kmem_alloc(sizeof(struct xfs_bmap_intent), KM_NOFS);
6131 	INIT_LIST_HEAD(&bi->bi_list);
6132 	bi->bi_type = type;
6133 	bi->bi_owner = ip;
6134 	bi->bi_whichfork = whichfork;
6135 	bi->bi_bmap = *bmap;
6136 
6137 	xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list);
6138 	return 0;
6139 }
6140 
6141 /* Map an extent into a file. */
6142 void
6143 xfs_bmap_map_extent(
6144 	struct xfs_trans	*tp,
6145 	struct xfs_inode	*ip,
6146 	struct xfs_bmbt_irec	*PREV)
6147 {
6148 	if (!xfs_bmap_is_update_needed(PREV))
6149 		return;
6150 
6151 	__xfs_bmap_add(tp, XFS_BMAP_MAP, ip, XFS_DATA_FORK, PREV);
6152 }
6153 
6154 /* Unmap an extent out of a file. */
6155 void
6156 xfs_bmap_unmap_extent(
6157 	struct xfs_trans	*tp,
6158 	struct xfs_inode	*ip,
6159 	struct xfs_bmbt_irec	*PREV)
6160 {
6161 	if (!xfs_bmap_is_update_needed(PREV))
6162 		return;
6163 
6164 	__xfs_bmap_add(tp, XFS_BMAP_UNMAP, ip, XFS_DATA_FORK, PREV);
6165 }
6166 
6167 /*
6168  * Process one of the deferred bmap operations.  We pass back the
6169  * btree cursor to maintain our lock on the bmapbt between calls.
6170  */
6171 int
6172 xfs_bmap_finish_one(
6173 	struct xfs_trans		*tp,
6174 	struct xfs_inode		*ip,
6175 	enum xfs_bmap_intent_type	type,
6176 	int				whichfork,
6177 	xfs_fileoff_t			startoff,
6178 	xfs_fsblock_t			startblock,
6179 	xfs_filblks_t			*blockcount,
6180 	xfs_exntst_t			state)
6181 {
6182 	int				error = 0;
6183 
6184 	ASSERT(tp->t_firstblock == NULLFSBLOCK);
6185 
6186 	trace_xfs_bmap_deferred(tp->t_mountp,
6187 			XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type,
6188 			XFS_FSB_TO_AGBNO(tp->t_mountp, startblock),
6189 			ip->i_ino, whichfork, startoff, *blockcount, state);
6190 
6191 	if (WARN_ON_ONCE(whichfork != XFS_DATA_FORK))
6192 		return -EFSCORRUPTED;
6193 
6194 	if (XFS_TEST_ERROR(false, tp->t_mountp,
6195 			XFS_ERRTAG_BMAP_FINISH_ONE))
6196 		return -EIO;
6197 
6198 	switch (type) {
6199 	case XFS_BMAP_MAP:
6200 		error = xfs_bmapi_remap(tp, ip, startoff, *blockcount,
6201 				startblock, 0);
6202 		*blockcount = 0;
6203 		break;
6204 	case XFS_BMAP_UNMAP:
6205 		error = __xfs_bunmapi(tp, ip, startoff, blockcount,
6206 				XFS_BMAPI_REMAP, 1);
6207 		break;
6208 	default:
6209 		ASSERT(0);
6210 		error = -EFSCORRUPTED;
6211 	}
6212 
6213 	return error;
6214 }
6215 
6216 /* Check that an inode's extent does not have invalid flags or bad ranges. */
6217 xfs_failaddr_t
6218 xfs_bmap_validate_extent(
6219 	struct xfs_inode	*ip,
6220 	int			whichfork,
6221 	struct xfs_bmbt_irec	*irec)
6222 {
6223 	struct xfs_mount	*mp = ip->i_mount;
6224 	xfs_fsblock_t		endfsb;
6225 	bool			isrt;
6226 
6227 	isrt = XFS_IS_REALTIME_INODE(ip);
6228 	endfsb = irec->br_startblock + irec->br_blockcount - 1;
6229 	if (isrt) {
6230 		if (!xfs_verify_rtbno(mp, irec->br_startblock))
6231 			return __this_address;
6232 		if (!xfs_verify_rtbno(mp, endfsb))
6233 			return __this_address;
6234 	} else {
6235 		if (!xfs_verify_fsbno(mp, irec->br_startblock))
6236 			return __this_address;
6237 		if (!xfs_verify_fsbno(mp, endfsb))
6238 			return __this_address;
6239 		if (XFS_FSB_TO_AGNO(mp, irec->br_startblock) !=
6240 		    XFS_FSB_TO_AGNO(mp, endfsb))
6241 			return __this_address;
6242 	}
6243 	if (irec->br_state != XFS_EXT_NORM && whichfork != XFS_DATA_FORK)
6244 		return __this_address;
6245 	return NULL;
6246 }
6247