xref: /linux/fs/xfs/libxfs/xfs_bmap.c (revision bb5b94f5bbe75470912b70fb08880fc5273aa62d)
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.h"
35 #include "xfs_ag_resv.h"
36 #include "xfs_refcount.h"
37 #include "xfs_icache.h"
38 #include "xfs_iomap.h"
39 
40 
41 kmem_zone_t		*xfs_bmap_free_item_zone;
42 
43 /*
44  * Miscellaneous helper functions
45  */
46 
47 /*
48  * Compute and fill in the value of the maximum depth of a bmap btree
49  * in this filesystem.  Done once, during mount.
50  */
51 void
52 xfs_bmap_compute_maxlevels(
53 	xfs_mount_t	*mp,		/* file system mount structure */
54 	int		whichfork)	/* data or attr fork */
55 {
56 	int		level;		/* btree level */
57 	uint		maxblocks;	/* max blocks at this level */
58 	uint		maxleafents;	/* max leaf entries possible */
59 	int		maxrootrecs;	/* max records in root block */
60 	int		minleafrecs;	/* min records in leaf block */
61 	int		minnoderecs;	/* min records in node block */
62 	int		sz;		/* root block size */
63 
64 	/*
65 	 * The maximum number of extents in a file, hence the maximum number of
66 	 * leaf entries, is controlled by the size of the on-disk extent count,
67 	 * either a signed 32-bit number for the data fork, or a signed 16-bit
68 	 * number for the attr fork.
69 	 *
70 	 * Note that we can no longer assume that if we are in ATTR1 that the
71 	 * fork offset of all the inodes will be
72 	 * (xfs_default_attroffset(ip) >> 3) because we could have mounted with
73 	 * ATTR2 and then mounted back with ATTR1, keeping the i_forkoff's fixed
74 	 * but probably at various positions. Therefore, for both ATTR1 and
75 	 * ATTR2 we have to assume the worst case scenario of a minimum size
76 	 * available.
77 	 */
78 	if (whichfork == XFS_DATA_FORK) {
79 		maxleafents = MAXEXTNUM;
80 		sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
81 	} else {
82 		maxleafents = MAXAEXTNUM;
83 		sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
84 	}
85 	maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
86 	minleafrecs = mp->m_bmap_dmnr[0];
87 	minnoderecs = mp->m_bmap_dmnr[1];
88 	maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
89 	for (level = 1; maxblocks > 1; level++) {
90 		if (maxblocks <= maxrootrecs)
91 			maxblocks = 1;
92 		else
93 			maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
94 	}
95 	mp->m_bm_maxlevels[whichfork] = level;
96 }
97 
98 unsigned int
99 xfs_bmap_compute_attr_offset(
100 	struct xfs_mount	*mp)
101 {
102 	if (mp->m_sb.sb_inodesize == 256)
103 		return XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
104 	return XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
105 }
106 
107 STATIC int				/* error */
108 xfs_bmbt_lookup_eq(
109 	struct xfs_btree_cur	*cur,
110 	struct xfs_bmbt_irec	*irec,
111 	int			*stat)	/* success/failure */
112 {
113 	cur->bc_rec.b = *irec;
114 	return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
115 }
116 
117 STATIC int				/* error */
118 xfs_bmbt_lookup_first(
119 	struct xfs_btree_cur	*cur,
120 	int			*stat)	/* success/failure */
121 {
122 	cur->bc_rec.b.br_startoff = 0;
123 	cur->bc_rec.b.br_startblock = 0;
124 	cur->bc_rec.b.br_blockcount = 0;
125 	return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
126 }
127 
128 /*
129  * Check if the inode needs to be converted to btree format.
130  */
131 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
132 {
133 	struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
134 
135 	return whichfork != XFS_COW_FORK &&
136 		ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
137 		ifp->if_nextents > XFS_IFORK_MAXEXT(ip, whichfork);
138 }
139 
140 /*
141  * Check if the inode should be converted to extent format.
142  */
143 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
144 {
145 	struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
146 
147 	return whichfork != XFS_COW_FORK &&
148 		ifp->if_format == XFS_DINODE_FMT_BTREE &&
149 		ifp->if_nextents <= XFS_IFORK_MAXEXT(ip, whichfork);
150 }
151 
152 /*
153  * Update the record referred to by cur to the value given by irec
154  * This either works (return 0) or gets an EFSCORRUPTED error.
155  */
156 STATIC int
157 xfs_bmbt_update(
158 	struct xfs_btree_cur	*cur,
159 	struct xfs_bmbt_irec	*irec)
160 {
161 	union xfs_btree_rec	rec;
162 
163 	xfs_bmbt_disk_set_all(&rec.bmbt, irec);
164 	return xfs_btree_update(cur, &rec);
165 }
166 
167 /*
168  * Compute the worst-case number of indirect blocks that will be used
169  * for ip's delayed extent of length "len".
170  */
171 STATIC xfs_filblks_t
172 xfs_bmap_worst_indlen(
173 	xfs_inode_t	*ip,		/* incore inode pointer */
174 	xfs_filblks_t	len)		/* delayed extent length */
175 {
176 	int		level;		/* btree level number */
177 	int		maxrecs;	/* maximum record count at this level */
178 	xfs_mount_t	*mp;		/* mount structure */
179 	xfs_filblks_t	rval;		/* return value */
180 
181 	mp = ip->i_mount;
182 	maxrecs = mp->m_bmap_dmxr[0];
183 	for (level = 0, rval = 0;
184 	     level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
185 	     level++) {
186 		len += maxrecs - 1;
187 		do_div(len, maxrecs);
188 		rval += len;
189 		if (len == 1)
190 			return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
191 				level - 1;
192 		if (level == 0)
193 			maxrecs = mp->m_bmap_dmxr[1];
194 	}
195 	return rval;
196 }
197 
198 /*
199  * Calculate the default attribute fork offset for newly created inodes.
200  */
201 uint
202 xfs_default_attroffset(
203 	struct xfs_inode	*ip)
204 {
205 	if (ip->i_df.if_format == XFS_DINODE_FMT_DEV)
206 		return roundup(sizeof(xfs_dev_t), 8);
207 	return M_IGEO(ip->i_mount)->attr_fork_offset;
208 }
209 
210 /*
211  * Helper routine to reset inode i_forkoff field when switching attribute fork
212  * from local to extent format - we reset it where possible to make space
213  * available for inline data fork extents.
214  */
215 STATIC void
216 xfs_bmap_forkoff_reset(
217 	xfs_inode_t	*ip,
218 	int		whichfork)
219 {
220 	if (whichfork == XFS_ATTR_FORK &&
221 	    ip->i_df.if_format != XFS_DINODE_FMT_DEV &&
222 	    ip->i_df.if_format != XFS_DINODE_FMT_BTREE) {
223 		uint	dfl_forkoff = xfs_default_attroffset(ip) >> 3;
224 
225 		if (dfl_forkoff > ip->i_forkoff)
226 			ip->i_forkoff = dfl_forkoff;
227 	}
228 }
229 
230 #ifdef DEBUG
231 STATIC struct xfs_buf *
232 xfs_bmap_get_bp(
233 	struct xfs_btree_cur	*cur,
234 	xfs_fsblock_t		bno)
235 {
236 	struct xfs_log_item	*lip;
237 	int			i;
238 
239 	if (!cur)
240 		return NULL;
241 
242 	for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
243 		if (!cur->bc_bufs[i])
244 			break;
245 		if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
246 			return cur->bc_bufs[i];
247 	}
248 
249 	/* Chase down all the log items to see if the bp is there */
250 	list_for_each_entry(lip, &cur->bc_tp->t_items, li_trans) {
251 		struct xfs_buf_log_item	*bip = (struct xfs_buf_log_item *)lip;
252 
253 		if (bip->bli_item.li_type == XFS_LI_BUF &&
254 		    XFS_BUF_ADDR(bip->bli_buf) == bno)
255 			return bip->bli_buf;
256 	}
257 
258 	return NULL;
259 }
260 
261 STATIC void
262 xfs_check_block(
263 	struct xfs_btree_block	*block,
264 	xfs_mount_t		*mp,
265 	int			root,
266 	short			sz)
267 {
268 	int			i, j, dmxr;
269 	__be64			*pp, *thispa;	/* pointer to block address */
270 	xfs_bmbt_key_t		*prevp, *keyp;
271 
272 	ASSERT(be16_to_cpu(block->bb_level) > 0);
273 
274 	prevp = NULL;
275 	for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
276 		dmxr = mp->m_bmap_dmxr[0];
277 		keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
278 
279 		if (prevp) {
280 			ASSERT(be64_to_cpu(prevp->br_startoff) <
281 			       be64_to_cpu(keyp->br_startoff));
282 		}
283 		prevp = keyp;
284 
285 		/*
286 		 * Compare the block numbers to see if there are dups.
287 		 */
288 		if (root)
289 			pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
290 		else
291 			pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
292 
293 		for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
294 			if (root)
295 				thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
296 			else
297 				thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
298 			if (*thispa == *pp) {
299 				xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
300 					__func__, j, i,
301 					(unsigned long long)be64_to_cpu(*thispa));
302 				xfs_err(mp, "%s: ptrs are equal in node\n",
303 					__func__);
304 				xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
305 			}
306 		}
307 	}
308 }
309 
310 /*
311  * Check that the extents for the inode ip are in the right order in all
312  * btree leaves. THis becomes prohibitively expensive for large extent count
313  * files, so don't bother with inodes that have more than 10,000 extents in
314  * them. The btree record ordering checks will still be done, so for such large
315  * bmapbt constructs that is going to catch most corruptions.
316  */
317 STATIC void
318 xfs_bmap_check_leaf_extents(
319 	xfs_btree_cur_t		*cur,	/* btree cursor or null */
320 	xfs_inode_t		*ip,		/* incore inode pointer */
321 	int			whichfork)	/* data or attr fork */
322 {
323 	struct xfs_mount	*mp = ip->i_mount;
324 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
325 	struct xfs_btree_block	*block;	/* current btree block */
326 	xfs_fsblock_t		bno;	/* block # of "block" */
327 	struct xfs_buf		*bp;	/* buffer for "block" */
328 	int			error;	/* error return value */
329 	xfs_extnum_t		i=0, j;	/* index into the extents list */
330 	int			level;	/* btree level, for checking */
331 	__be64			*pp;	/* pointer to block address */
332 	xfs_bmbt_rec_t		*ep;	/* pointer to current extent */
333 	xfs_bmbt_rec_t		last = {0, 0}; /* last extent in prev block */
334 	xfs_bmbt_rec_t		*nextp;	/* pointer to next extent */
335 	int			bp_release = 0;
336 
337 	if (ifp->if_format != XFS_DINODE_FMT_BTREE)
338 		return;
339 
340 	/* skip large extent count inodes */
341 	if (ip->i_df.if_nextents > 10000)
342 		return;
343 
344 	bno = NULLFSBLOCK;
345 	block = ifp->if_broot;
346 	/*
347 	 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
348 	 */
349 	level = be16_to_cpu(block->bb_level);
350 	ASSERT(level > 0);
351 	xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
352 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
353 	bno = be64_to_cpu(*pp);
354 
355 	ASSERT(bno != NULLFSBLOCK);
356 	ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
357 	ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
358 
359 	/*
360 	 * Go down the tree until leaf level is reached, following the first
361 	 * pointer (leftmost) at each level.
362 	 */
363 	while (level-- > 0) {
364 		/* See if buf is in cur first */
365 		bp_release = 0;
366 		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
367 		if (!bp) {
368 			bp_release = 1;
369 			error = xfs_btree_read_bufl(mp, NULL, bno, &bp,
370 						XFS_BMAP_BTREE_REF,
371 						&xfs_bmbt_buf_ops);
372 			if (error)
373 				goto error_norelse;
374 		}
375 		block = XFS_BUF_TO_BLOCK(bp);
376 		if (level == 0)
377 			break;
378 
379 		/*
380 		 * Check this block for basic sanity (increasing keys and
381 		 * no duplicate blocks).
382 		 */
383 
384 		xfs_check_block(block, mp, 0, 0);
385 		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
386 		bno = be64_to_cpu(*pp);
387 		if (XFS_IS_CORRUPT(mp, !xfs_verify_fsbno(mp, bno))) {
388 			error = -EFSCORRUPTED;
389 			goto error0;
390 		}
391 		if (bp_release) {
392 			bp_release = 0;
393 			xfs_trans_brelse(NULL, bp);
394 		}
395 	}
396 
397 	/*
398 	 * Here with bp and block set to the leftmost leaf node in the tree.
399 	 */
400 	i = 0;
401 
402 	/*
403 	 * Loop over all leaf nodes checking that all extents are in the right order.
404 	 */
405 	for (;;) {
406 		xfs_fsblock_t	nextbno;
407 		xfs_extnum_t	num_recs;
408 
409 
410 		num_recs = xfs_btree_get_numrecs(block);
411 
412 		/*
413 		 * Read-ahead the next leaf block, if any.
414 		 */
415 
416 		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
417 
418 		/*
419 		 * Check all the extents to make sure they are OK.
420 		 * If we had a previous block, the last entry should
421 		 * conform with the first entry in this one.
422 		 */
423 
424 		ep = XFS_BMBT_REC_ADDR(mp, block, 1);
425 		if (i) {
426 			ASSERT(xfs_bmbt_disk_get_startoff(&last) +
427 			       xfs_bmbt_disk_get_blockcount(&last) <=
428 			       xfs_bmbt_disk_get_startoff(ep));
429 		}
430 		for (j = 1; j < num_recs; j++) {
431 			nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
432 			ASSERT(xfs_bmbt_disk_get_startoff(ep) +
433 			       xfs_bmbt_disk_get_blockcount(ep) <=
434 			       xfs_bmbt_disk_get_startoff(nextp));
435 			ep = nextp;
436 		}
437 
438 		last = *ep;
439 		i += num_recs;
440 		if (bp_release) {
441 			bp_release = 0;
442 			xfs_trans_brelse(NULL, bp);
443 		}
444 		bno = nextbno;
445 		/*
446 		 * If we've reached the end, stop.
447 		 */
448 		if (bno == NULLFSBLOCK)
449 			break;
450 
451 		bp_release = 0;
452 		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
453 		if (!bp) {
454 			bp_release = 1;
455 			error = xfs_btree_read_bufl(mp, NULL, bno, &bp,
456 						XFS_BMAP_BTREE_REF,
457 						&xfs_bmbt_buf_ops);
458 			if (error)
459 				goto error_norelse;
460 		}
461 		block = XFS_BUF_TO_BLOCK(bp);
462 	}
463 
464 	return;
465 
466 error0:
467 	xfs_warn(mp, "%s: at error0", __func__);
468 	if (bp_release)
469 		xfs_trans_brelse(NULL, bp);
470 error_norelse:
471 	xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
472 		__func__, i);
473 	xfs_err(mp, "%s: CORRUPTED BTREE OR SOMETHING", __func__);
474 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
475 	return;
476 }
477 
478 /*
479  * Validate that the bmbt_irecs being returned from bmapi are valid
480  * given the caller's original parameters.  Specifically check the
481  * ranges of the returned irecs to ensure that they only extend beyond
482  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
483  */
484 STATIC void
485 xfs_bmap_validate_ret(
486 	xfs_fileoff_t		bno,
487 	xfs_filblks_t		len,
488 	int			flags,
489 	xfs_bmbt_irec_t		*mval,
490 	int			nmap,
491 	int			ret_nmap)
492 {
493 	int			i;		/* index to map values */
494 
495 	ASSERT(ret_nmap <= nmap);
496 
497 	for (i = 0; i < ret_nmap; i++) {
498 		ASSERT(mval[i].br_blockcount > 0);
499 		if (!(flags & XFS_BMAPI_ENTIRE)) {
500 			ASSERT(mval[i].br_startoff >= bno);
501 			ASSERT(mval[i].br_blockcount <= len);
502 			ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
503 			       bno + len);
504 		} else {
505 			ASSERT(mval[i].br_startoff < bno + len);
506 			ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
507 			       bno);
508 		}
509 		ASSERT(i == 0 ||
510 		       mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
511 		       mval[i].br_startoff);
512 		ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
513 		       mval[i].br_startblock != HOLESTARTBLOCK);
514 		ASSERT(mval[i].br_state == XFS_EXT_NORM ||
515 		       mval[i].br_state == XFS_EXT_UNWRITTEN);
516 	}
517 }
518 
519 #else
520 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)		do { } while (0)
521 #define	xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)	do { } while (0)
522 #endif /* DEBUG */
523 
524 /*
525  * bmap free list manipulation functions
526  */
527 
528 /*
529  * Add the extent to the list of extents to be free at transaction end.
530  * The list is maintained sorted (by block number).
531  */
532 void
533 __xfs_bmap_add_free(
534 	struct xfs_trans		*tp,
535 	xfs_fsblock_t			bno,
536 	xfs_filblks_t			len,
537 	const struct xfs_owner_info	*oinfo,
538 	bool				skip_discard)
539 {
540 	struct xfs_extent_free_item	*new;		/* new element */
541 #ifdef DEBUG
542 	struct xfs_mount		*mp = tp->t_mountp;
543 	xfs_agnumber_t			agno;
544 	xfs_agblock_t			agbno;
545 
546 	ASSERT(bno != NULLFSBLOCK);
547 	ASSERT(len > 0);
548 	ASSERT(len <= MAXEXTLEN);
549 	ASSERT(!isnullstartblock(bno));
550 	agno = XFS_FSB_TO_AGNO(mp, bno);
551 	agbno = XFS_FSB_TO_AGBNO(mp, bno);
552 	ASSERT(agno < mp->m_sb.sb_agcount);
553 	ASSERT(agbno < mp->m_sb.sb_agblocks);
554 	ASSERT(len < mp->m_sb.sb_agblocks);
555 	ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
556 #endif
557 	ASSERT(xfs_bmap_free_item_zone != NULL);
558 
559 	new = kmem_cache_alloc(xfs_bmap_free_item_zone,
560 			       GFP_KERNEL | __GFP_NOFAIL);
561 	new->xefi_startblock = bno;
562 	new->xefi_blockcount = (xfs_extlen_t)len;
563 	if (oinfo)
564 		new->xefi_oinfo = *oinfo;
565 	else
566 		new->xefi_oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
567 	new->xefi_skip_discard = skip_discard;
568 	trace_xfs_bmap_free_defer(tp->t_mountp,
569 			XFS_FSB_TO_AGNO(tp->t_mountp, bno), 0,
570 			XFS_FSB_TO_AGBNO(tp->t_mountp, bno), len);
571 	xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list);
572 }
573 
574 /*
575  * Inode fork format manipulation functions
576  */
577 
578 /*
579  * Convert the inode format to extent format if it currently is in btree format,
580  * but the extent list is small enough that it fits into the extent format.
581  *
582  * Since the extents are already in-core, all we have to do is give up the space
583  * for the btree root and pitch the leaf block.
584  */
585 STATIC int				/* error */
586 xfs_bmap_btree_to_extents(
587 	struct xfs_trans	*tp,	/* transaction pointer */
588 	struct xfs_inode	*ip,	/* incore inode pointer */
589 	struct xfs_btree_cur	*cur,	/* btree cursor */
590 	int			*logflagsp, /* inode logging flags */
591 	int			whichfork)  /* data or attr fork */
592 {
593 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
594 	struct xfs_mount	*mp = ip->i_mount;
595 	struct xfs_btree_block	*rblock = ifp->if_broot;
596 	struct xfs_btree_block	*cblock;/* child btree block */
597 	xfs_fsblock_t		cbno;	/* child block number */
598 	struct xfs_buf		*cbp;	/* child block's buffer */
599 	int			error;	/* error return value */
600 	__be64			*pp;	/* ptr to block address */
601 	struct xfs_owner_info	oinfo;
602 
603 	/* check if we actually need the extent format first: */
604 	if (!xfs_bmap_wants_extents(ip, whichfork))
605 		return 0;
606 
607 	ASSERT(cur);
608 	ASSERT(whichfork != XFS_COW_FORK);
609 	ASSERT(ifp->if_format == XFS_DINODE_FMT_BTREE);
610 	ASSERT(be16_to_cpu(rblock->bb_level) == 1);
611 	ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
612 	ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
613 
614 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
615 	cbno = be64_to_cpu(*pp);
616 #ifdef DEBUG
617 	if (XFS_IS_CORRUPT(cur->bc_mp, !xfs_btree_check_lptr(cur, cbno, 1)))
618 		return -EFSCORRUPTED;
619 #endif
620 	error = xfs_btree_read_bufl(mp, tp, cbno, &cbp, XFS_BMAP_BTREE_REF,
621 				&xfs_bmbt_buf_ops);
622 	if (error)
623 		return error;
624 	cblock = XFS_BUF_TO_BLOCK(cbp);
625 	if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
626 		return error;
627 	xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
628 	xfs_bmap_add_free(cur->bc_tp, cbno, 1, &oinfo);
629 	ip->i_nblocks--;
630 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
631 	xfs_trans_binval(tp, cbp);
632 	if (cur->bc_bufs[0] == cbp)
633 		cur->bc_bufs[0] = NULL;
634 	xfs_iroot_realloc(ip, -1, whichfork);
635 	ASSERT(ifp->if_broot == NULL);
636 	ifp->if_format = XFS_DINODE_FMT_EXTENTS;
637 	*logflagsp |= XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
638 	return 0;
639 }
640 
641 /*
642  * Convert an extents-format file into a btree-format file.
643  * The new file will have a root block (in the inode) and a single child block.
644  */
645 STATIC int					/* error */
646 xfs_bmap_extents_to_btree(
647 	struct xfs_trans	*tp,		/* transaction pointer */
648 	struct xfs_inode	*ip,		/* incore inode pointer */
649 	struct xfs_btree_cur	**curp,		/* cursor returned to caller */
650 	int			wasdel,		/* converting a delayed alloc */
651 	int			*logflagsp,	/* inode logging flags */
652 	int			whichfork)	/* data or attr fork */
653 {
654 	struct xfs_btree_block	*ablock;	/* allocated (child) bt block */
655 	struct xfs_buf		*abp;		/* buffer for ablock */
656 	struct xfs_alloc_arg	args;		/* allocation arguments */
657 	struct xfs_bmbt_rec	*arp;		/* child record pointer */
658 	struct xfs_btree_block	*block;		/* btree root block */
659 	struct xfs_btree_cur	*cur;		/* bmap btree cursor */
660 	int			error;		/* error return value */
661 	struct xfs_ifork	*ifp;		/* inode fork pointer */
662 	struct xfs_bmbt_key	*kp;		/* root block key pointer */
663 	struct xfs_mount	*mp;		/* mount structure */
664 	xfs_bmbt_ptr_t		*pp;		/* root block address pointer */
665 	struct xfs_iext_cursor	icur;
666 	struct xfs_bmbt_irec	rec;
667 	xfs_extnum_t		cnt = 0;
668 
669 	mp = ip->i_mount;
670 	ASSERT(whichfork != XFS_COW_FORK);
671 	ifp = XFS_IFORK_PTR(ip, whichfork);
672 	ASSERT(ifp->if_format == XFS_DINODE_FMT_EXTENTS);
673 
674 	/*
675 	 * Make space in the inode incore. This needs to be undone if we fail
676 	 * to expand the root.
677 	 */
678 	xfs_iroot_realloc(ip, 1, whichfork);
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_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_u1.if_root = NULL;
809 	ifp->if_height = 0;
810 	ifp->if_format = XFS_DINODE_FMT_EXTENTS;
811 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
812 }
813 
814 
815 STATIC int				/* error */
816 xfs_bmap_local_to_extents(
817 	xfs_trans_t	*tp,		/* transaction pointer */
818 	xfs_inode_t	*ip,		/* incore inode pointer */
819 	xfs_extlen_t	total,		/* total blocks needed by transaction */
820 	int		*logflagsp,	/* inode logging flags */
821 	int		whichfork,
822 	void		(*init_fn)(struct xfs_trans *tp,
823 				   struct xfs_buf *bp,
824 				   struct xfs_inode *ip,
825 				   struct xfs_ifork *ifp))
826 {
827 	int		error = 0;
828 	int		flags;		/* logging flags returned */
829 	struct xfs_ifork *ifp;		/* inode fork pointer */
830 	xfs_alloc_arg_t	args;		/* allocation arguments */
831 	struct xfs_buf	*bp;		/* buffer for extent block */
832 	struct xfs_bmbt_irec rec;
833 	struct xfs_iext_cursor icur;
834 
835 	/*
836 	 * We don't want to deal with the case of keeping inode data inline yet.
837 	 * So sending the data fork of a regular inode is invalid.
838 	 */
839 	ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
840 	ifp = XFS_IFORK_PTR(ip, whichfork);
841 	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
842 
843 	if (!ifp->if_bytes) {
844 		xfs_bmap_local_to_extents_empty(tp, ip, whichfork);
845 		flags = XFS_ILOG_CORE;
846 		goto done;
847 	}
848 
849 	flags = 0;
850 	error = 0;
851 	memset(&args, 0, sizeof(args));
852 	args.tp = tp;
853 	args.mp = ip->i_mount;
854 	xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0);
855 	/*
856 	 * Allocate a block.  We know we need only one, since the
857 	 * file currently fits in an inode.
858 	 */
859 	if (tp->t_firstblock == NULLFSBLOCK) {
860 		args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
861 		args.type = XFS_ALLOCTYPE_START_BNO;
862 	} else {
863 		args.fsbno = tp->t_firstblock;
864 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
865 	}
866 	args.total = total;
867 	args.minlen = args.maxlen = args.prod = 1;
868 	error = xfs_alloc_vextent(&args);
869 	if (error)
870 		goto done;
871 
872 	/* Can't fail, the space was reserved. */
873 	ASSERT(args.fsbno != NULLFSBLOCK);
874 	ASSERT(args.len == 1);
875 	tp->t_firstblock = args.fsbno;
876 	error = xfs_trans_get_buf(tp, args.mp->m_ddev_targp,
877 			XFS_FSB_TO_DADDR(args.mp, args.fsbno),
878 			args.mp->m_bsize, 0, &bp);
879 	if (error)
880 		goto done;
881 
882 	/*
883 	 * Initialize the block, copy the data and log the remote buffer.
884 	 *
885 	 * The callout is responsible for logging because the remote format
886 	 * might differ from the local format and thus we don't know how much to
887 	 * log here. Note that init_fn must also set the buffer log item type
888 	 * correctly.
889 	 */
890 	init_fn(tp, bp, ip, ifp);
891 
892 	/* account for the change in fork size */
893 	xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
894 	xfs_bmap_local_to_extents_empty(tp, ip, whichfork);
895 	flags |= XFS_ILOG_CORE;
896 
897 	ifp->if_u1.if_root = NULL;
898 	ifp->if_height = 0;
899 
900 	rec.br_startoff = 0;
901 	rec.br_startblock = args.fsbno;
902 	rec.br_blockcount = 1;
903 	rec.br_state = XFS_EXT_NORM;
904 	xfs_iext_first(ifp, &icur);
905 	xfs_iext_insert(ip, &icur, &rec, 0);
906 
907 	ifp->if_nextents = 1;
908 	ip->i_nblocks = 1;
909 	xfs_trans_mod_dquot_byino(tp, ip,
910 		XFS_TRANS_DQ_BCOUNT, 1L);
911 	flags |= xfs_ilog_fext(whichfork);
912 
913 done:
914 	*logflagsp = flags;
915 	return error;
916 }
917 
918 /*
919  * Called from xfs_bmap_add_attrfork to handle btree format files.
920  */
921 STATIC int					/* error */
922 xfs_bmap_add_attrfork_btree(
923 	xfs_trans_t		*tp,		/* transaction pointer */
924 	xfs_inode_t		*ip,		/* incore inode pointer */
925 	int			*flags)		/* inode logging flags */
926 {
927 	struct xfs_btree_block	*block = ip->i_df.if_broot;
928 	xfs_btree_cur_t		*cur;		/* btree cursor */
929 	int			error;		/* error return value */
930 	xfs_mount_t		*mp;		/* file system mount struct */
931 	int			stat;		/* newroot status */
932 
933 	mp = ip->i_mount;
934 
935 	if (XFS_BMAP_BMDR_SPACE(block) <= XFS_IFORK_DSIZE(ip))
936 		*flags |= XFS_ILOG_DBROOT;
937 	else {
938 		cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
939 		error = xfs_bmbt_lookup_first(cur, &stat);
940 		if (error)
941 			goto error0;
942 		/* must be at least one entry */
943 		if (XFS_IS_CORRUPT(mp, stat != 1)) {
944 			error = -EFSCORRUPTED;
945 			goto error0;
946 		}
947 		if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
948 			goto error0;
949 		if (stat == 0) {
950 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
951 			return -ENOSPC;
952 		}
953 		cur->bc_ino.allocated = 0;
954 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
955 	}
956 	return 0;
957 error0:
958 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
959 	return error;
960 }
961 
962 /*
963  * Called from xfs_bmap_add_attrfork to handle extents format files.
964  */
965 STATIC int					/* error */
966 xfs_bmap_add_attrfork_extents(
967 	struct xfs_trans	*tp,		/* transaction pointer */
968 	struct xfs_inode	*ip,		/* incore inode pointer */
969 	int			*flags)		/* inode logging flags */
970 {
971 	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
972 	int			error;		/* error return value */
973 
974 	if (ip->i_df.if_nextents * sizeof(struct xfs_bmbt_rec) <=
975 	    XFS_IFORK_DSIZE(ip))
976 		return 0;
977 	cur = NULL;
978 	error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, flags,
979 					  XFS_DATA_FORK);
980 	if (cur) {
981 		cur->bc_ino.allocated = 0;
982 		xfs_btree_del_cursor(cur, error);
983 	}
984 	return error;
985 }
986 
987 /*
988  * Called from xfs_bmap_add_attrfork to handle local format files. Each
989  * different data fork content type needs a different callout to do the
990  * conversion. Some are basic and only require special block initialisation
991  * callouts for the data formating, others (directories) are so specialised they
992  * handle everything themselves.
993  *
994  * XXX (dgc): investigate whether directory conversion can use the generic
995  * formatting callout. It should be possible - it's just a very complex
996  * formatter.
997  */
998 STATIC int					/* error */
999 xfs_bmap_add_attrfork_local(
1000 	struct xfs_trans	*tp,		/* transaction pointer */
1001 	struct xfs_inode	*ip,		/* incore inode pointer */
1002 	int			*flags)		/* inode logging flags */
1003 {
1004 	struct xfs_da_args	dargs;		/* args for dir/attr code */
1005 
1006 	if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1007 		return 0;
1008 
1009 	if (S_ISDIR(VFS_I(ip)->i_mode)) {
1010 		memset(&dargs, 0, sizeof(dargs));
1011 		dargs.geo = ip->i_mount->m_dir_geo;
1012 		dargs.dp = ip;
1013 		dargs.total = dargs.geo->fsbcount;
1014 		dargs.whichfork = XFS_DATA_FORK;
1015 		dargs.trans = tp;
1016 		return xfs_dir2_sf_to_block(&dargs);
1017 	}
1018 
1019 	if (S_ISLNK(VFS_I(ip)->i_mode))
1020 		return xfs_bmap_local_to_extents(tp, ip, 1, flags,
1021 						 XFS_DATA_FORK,
1022 						 xfs_symlink_local_to_remote);
1023 
1024 	/* should only be called for types that support local format data */
1025 	ASSERT(0);
1026 	return -EFSCORRUPTED;
1027 }
1028 
1029 /*
1030  * Set an inode attr fork offset based on the format of the data fork.
1031  */
1032 static int
1033 xfs_bmap_set_attrforkoff(
1034 	struct xfs_inode	*ip,
1035 	int			size,
1036 	int			*version)
1037 {
1038 	int			default_size = xfs_default_attroffset(ip) >> 3;
1039 
1040 	switch (ip->i_df.if_format) {
1041 	case XFS_DINODE_FMT_DEV:
1042 		ip->i_forkoff = default_size;
1043 		break;
1044 	case XFS_DINODE_FMT_LOCAL:
1045 	case XFS_DINODE_FMT_EXTENTS:
1046 	case XFS_DINODE_FMT_BTREE:
1047 		ip->i_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1048 		if (!ip->i_forkoff)
1049 			ip->i_forkoff = default_size;
1050 		else if ((ip->i_mount->m_flags & XFS_MOUNT_ATTR2) && version)
1051 			*version = 2;
1052 		break;
1053 	default:
1054 		ASSERT(0);
1055 		return -EINVAL;
1056 	}
1057 
1058 	return 0;
1059 }
1060 
1061 /*
1062  * Convert inode from non-attributed to attributed.
1063  * Must not be in a transaction, ip must not be locked.
1064  */
1065 int						/* error code */
1066 xfs_bmap_add_attrfork(
1067 	xfs_inode_t		*ip,		/* incore inode pointer */
1068 	int			size,		/* space new attribute needs */
1069 	int			rsvd)		/* xact may use reserved blks */
1070 {
1071 	xfs_mount_t		*mp;		/* mount structure */
1072 	xfs_trans_t		*tp;		/* transaction pointer */
1073 	int			blks;		/* space reservation */
1074 	int			version = 1;	/* superblock attr version */
1075 	int			logflags;	/* logging flags */
1076 	int			error;		/* error return value */
1077 
1078 	ASSERT(XFS_IFORK_Q(ip) == 0);
1079 
1080 	mp = ip->i_mount;
1081 	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1082 
1083 	blks = XFS_ADDAFORK_SPACE_RES(mp);
1084 
1085 	error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_addafork, blks, 0,
1086 			rsvd, &tp);
1087 	if (error)
1088 		return error;
1089 	if (XFS_IFORK_Q(ip))
1090 		goto trans_cancel;
1091 
1092 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1093 	error = xfs_bmap_set_attrforkoff(ip, size, &version);
1094 	if (error)
1095 		goto trans_cancel;
1096 	ASSERT(ip->i_afp == NULL);
1097 
1098 	ip->i_afp = xfs_ifork_alloc(XFS_DINODE_FMT_EXTENTS, 0);
1099 	logflags = 0;
1100 	switch (ip->i_df.if_format) {
1101 	case XFS_DINODE_FMT_LOCAL:
1102 		error = xfs_bmap_add_attrfork_local(tp, ip, &logflags);
1103 		break;
1104 	case XFS_DINODE_FMT_EXTENTS:
1105 		error = xfs_bmap_add_attrfork_extents(tp, ip, &logflags);
1106 		break;
1107 	case XFS_DINODE_FMT_BTREE:
1108 		error = xfs_bmap_add_attrfork_btree(tp, ip, &logflags);
1109 		break;
1110 	default:
1111 		error = 0;
1112 		break;
1113 	}
1114 	if (logflags)
1115 		xfs_trans_log_inode(tp, ip, logflags);
1116 	if (error)
1117 		goto trans_cancel;
1118 	if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1119 	   (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1120 		bool log_sb = false;
1121 
1122 		spin_lock(&mp->m_sb_lock);
1123 		if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1124 			xfs_sb_version_addattr(&mp->m_sb);
1125 			log_sb = true;
1126 		}
1127 		if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1128 			xfs_sb_version_addattr2(&mp->m_sb);
1129 			log_sb = true;
1130 		}
1131 		spin_unlock(&mp->m_sb_lock);
1132 		if (log_sb)
1133 			xfs_log_sb(tp);
1134 	}
1135 
1136 	error = xfs_trans_commit(tp);
1137 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1138 	return error;
1139 
1140 trans_cancel:
1141 	xfs_trans_cancel(tp);
1142 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1143 	return error;
1144 }
1145 
1146 /*
1147  * Internal and external extent tree search functions.
1148  */
1149 
1150 struct xfs_iread_state {
1151 	struct xfs_iext_cursor	icur;
1152 	xfs_extnum_t		loaded;
1153 };
1154 
1155 /* Stuff every bmbt record from this block into the incore extent map. */
1156 static int
1157 xfs_iread_bmbt_block(
1158 	struct xfs_btree_cur	*cur,
1159 	int			level,
1160 	void			*priv)
1161 {
1162 	struct xfs_iread_state	*ir = priv;
1163 	struct xfs_mount	*mp = cur->bc_mp;
1164 	struct xfs_inode	*ip = cur->bc_ino.ip;
1165 	struct xfs_btree_block	*block;
1166 	struct xfs_buf		*bp;
1167 	struct xfs_bmbt_rec	*frp;
1168 	xfs_extnum_t		num_recs;
1169 	xfs_extnum_t		j;
1170 	int			whichfork = cur->bc_ino.whichfork;
1171 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1172 
1173 	block = xfs_btree_get_block(cur, level, &bp);
1174 
1175 	/* Abort if we find more records than nextents. */
1176 	num_recs = xfs_btree_get_numrecs(block);
1177 	if (unlikely(ir->loaded + num_recs > ifp->if_nextents)) {
1178 		xfs_warn(ip->i_mount, "corrupt dinode %llu, (btree extents).",
1179 				(unsigned long long)ip->i_ino);
1180 		xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, block,
1181 				sizeof(*block), __this_address);
1182 		return -EFSCORRUPTED;
1183 	}
1184 
1185 	/* Copy records into the incore cache. */
1186 	frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1187 	for (j = 0; j < num_recs; j++, frp++, ir->loaded++) {
1188 		struct xfs_bmbt_irec	new;
1189 		xfs_failaddr_t		fa;
1190 
1191 		xfs_bmbt_disk_get_all(frp, &new);
1192 		fa = xfs_bmap_validate_extent(ip, whichfork, &new);
1193 		if (fa) {
1194 			xfs_inode_verifier_error(ip, -EFSCORRUPTED,
1195 					"xfs_iread_extents(2)", frp,
1196 					sizeof(*frp), fa);
1197 			return -EFSCORRUPTED;
1198 		}
1199 		xfs_iext_insert(ip, &ir->icur, &new,
1200 				xfs_bmap_fork_to_state(whichfork));
1201 		trace_xfs_read_extent(ip, &ir->icur,
1202 				xfs_bmap_fork_to_state(whichfork), _THIS_IP_);
1203 		xfs_iext_next(ifp, &ir->icur);
1204 	}
1205 
1206 	return 0;
1207 }
1208 
1209 /*
1210  * Read in extents from a btree-format inode.
1211  */
1212 int
1213 xfs_iread_extents(
1214 	struct xfs_trans	*tp,
1215 	struct xfs_inode	*ip,
1216 	int			whichfork)
1217 {
1218 	struct xfs_iread_state	ir;
1219 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1220 	struct xfs_mount	*mp = ip->i_mount;
1221 	struct xfs_btree_cur	*cur;
1222 	int			error;
1223 
1224 	if (!xfs_need_iread_extents(ifp))
1225 		return 0;
1226 
1227 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1228 
1229 	ir.loaded = 0;
1230 	xfs_iext_first(ifp, &ir.icur);
1231 	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
1232 	error = xfs_btree_visit_blocks(cur, xfs_iread_bmbt_block,
1233 			XFS_BTREE_VISIT_RECORDS, &ir);
1234 	xfs_btree_del_cursor(cur, error);
1235 	if (error)
1236 		goto out;
1237 
1238 	if (XFS_IS_CORRUPT(mp, ir.loaded != ifp->if_nextents)) {
1239 		error = -EFSCORRUPTED;
1240 		goto out;
1241 	}
1242 	ASSERT(ir.loaded == xfs_iext_count(ifp));
1243 	return 0;
1244 out:
1245 	xfs_iext_destroy(ifp);
1246 	return error;
1247 }
1248 
1249 /*
1250  * Returns the relative block number of the first unused block(s) in the given
1251  * fork with at least "len" logically contiguous blocks free.  This is the
1252  * lowest-address hole if the fork has holes, else the first block past the end
1253  * of fork.  Return 0 if the fork is currently local (in-inode).
1254  */
1255 int						/* error */
1256 xfs_bmap_first_unused(
1257 	struct xfs_trans	*tp,		/* transaction pointer */
1258 	struct xfs_inode	*ip,		/* incore inode */
1259 	xfs_extlen_t		len,		/* size of hole to find */
1260 	xfs_fileoff_t		*first_unused,	/* unused block */
1261 	int			whichfork)	/* data or attr fork */
1262 {
1263 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1264 	struct xfs_bmbt_irec	got;
1265 	struct xfs_iext_cursor	icur;
1266 	xfs_fileoff_t		lastaddr = 0;
1267 	xfs_fileoff_t		lowest, max;
1268 	int			error;
1269 
1270 	if (ifp->if_format == XFS_DINODE_FMT_LOCAL) {
1271 		*first_unused = 0;
1272 		return 0;
1273 	}
1274 
1275 	ASSERT(xfs_ifork_has_extents(ifp));
1276 
1277 	error = xfs_iread_extents(tp, ip, whichfork);
1278 	if (error)
1279 		return error;
1280 
1281 	lowest = max = *first_unused;
1282 	for_each_xfs_iext(ifp, &icur, &got) {
1283 		/*
1284 		 * See if the hole before this extent will work.
1285 		 */
1286 		if (got.br_startoff >= lowest + len &&
1287 		    got.br_startoff - max >= len)
1288 			break;
1289 		lastaddr = got.br_startoff + got.br_blockcount;
1290 		max = XFS_FILEOFF_MAX(lastaddr, lowest);
1291 	}
1292 
1293 	*first_unused = max;
1294 	return 0;
1295 }
1296 
1297 /*
1298  * Returns the file-relative block number of the last block - 1 before
1299  * last_block (input value) in the file.
1300  * This is not based on i_size, it is based on the extent records.
1301  * Returns 0 for local files, as they do not have extent records.
1302  */
1303 int						/* error */
1304 xfs_bmap_last_before(
1305 	struct xfs_trans	*tp,		/* transaction pointer */
1306 	struct xfs_inode	*ip,		/* incore inode */
1307 	xfs_fileoff_t		*last_block,	/* last block */
1308 	int			whichfork)	/* data or attr fork */
1309 {
1310 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1311 	struct xfs_bmbt_irec	got;
1312 	struct xfs_iext_cursor	icur;
1313 	int			error;
1314 
1315 	switch (ifp->if_format) {
1316 	case XFS_DINODE_FMT_LOCAL:
1317 		*last_block = 0;
1318 		return 0;
1319 	case XFS_DINODE_FMT_BTREE:
1320 	case XFS_DINODE_FMT_EXTENTS:
1321 		break;
1322 	default:
1323 		ASSERT(0);
1324 		return -EFSCORRUPTED;
1325 	}
1326 
1327 	error = xfs_iread_extents(tp, ip, whichfork);
1328 	if (error)
1329 		return error;
1330 
1331 	if (!xfs_iext_lookup_extent_before(ip, ifp, last_block, &icur, &got))
1332 		*last_block = 0;
1333 	return 0;
1334 }
1335 
1336 int
1337 xfs_bmap_last_extent(
1338 	struct xfs_trans	*tp,
1339 	struct xfs_inode	*ip,
1340 	int			whichfork,
1341 	struct xfs_bmbt_irec	*rec,
1342 	int			*is_empty)
1343 {
1344 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1345 	struct xfs_iext_cursor	icur;
1346 	int			error;
1347 
1348 	error = xfs_iread_extents(tp, ip, whichfork);
1349 	if (error)
1350 		return error;
1351 
1352 	xfs_iext_last(ifp, &icur);
1353 	if (!xfs_iext_get_extent(ifp, &icur, rec))
1354 		*is_empty = 1;
1355 	else
1356 		*is_empty = 0;
1357 	return 0;
1358 }
1359 
1360 /*
1361  * Check the last inode extent to determine whether this allocation will result
1362  * in blocks being allocated at the end of the file. When we allocate new data
1363  * blocks at the end of the file which do not start at the previous data block,
1364  * we will try to align the new blocks at stripe unit boundaries.
1365  *
1366  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1367  * at, or past the EOF.
1368  */
1369 STATIC int
1370 xfs_bmap_isaeof(
1371 	struct xfs_bmalloca	*bma,
1372 	int			whichfork)
1373 {
1374 	struct xfs_bmbt_irec	rec;
1375 	int			is_empty;
1376 	int			error;
1377 
1378 	bma->aeof = false;
1379 	error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1380 				     &is_empty);
1381 	if (error)
1382 		return error;
1383 
1384 	if (is_empty) {
1385 		bma->aeof = true;
1386 		return 0;
1387 	}
1388 
1389 	/*
1390 	 * Check if we are allocation or past the last extent, or at least into
1391 	 * the last delayed allocated extent.
1392 	 */
1393 	bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1394 		(bma->offset >= rec.br_startoff &&
1395 		 isnullstartblock(rec.br_startblock));
1396 	return 0;
1397 }
1398 
1399 /*
1400  * Returns the file-relative block number of the first block past eof in
1401  * the file.  This is not based on i_size, it is based on the extent records.
1402  * Returns 0 for local files, as they do not have extent records.
1403  */
1404 int
1405 xfs_bmap_last_offset(
1406 	struct xfs_inode	*ip,
1407 	xfs_fileoff_t		*last_block,
1408 	int			whichfork)
1409 {
1410 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1411 	struct xfs_bmbt_irec	rec;
1412 	int			is_empty;
1413 	int			error;
1414 
1415 	*last_block = 0;
1416 
1417 	if (ifp->if_format == XFS_DINODE_FMT_LOCAL)
1418 		return 0;
1419 
1420 	if (XFS_IS_CORRUPT(ip->i_mount, !xfs_ifork_has_extents(ifp)))
1421 		return -EFSCORRUPTED;
1422 
1423 	error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1424 	if (error || is_empty)
1425 		return error;
1426 
1427 	*last_block = rec.br_startoff + rec.br_blockcount;
1428 	return 0;
1429 }
1430 
1431 /*
1432  * Extent tree manipulation functions used during allocation.
1433  */
1434 
1435 /*
1436  * Convert a delayed allocation to a real allocation.
1437  */
1438 STATIC int				/* error */
1439 xfs_bmap_add_extent_delay_real(
1440 	struct xfs_bmalloca	*bma,
1441 	int			whichfork)
1442 {
1443 	struct xfs_mount	*mp = bma->ip->i_mount;
1444 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1445 	struct xfs_bmbt_irec	*new = &bma->got;
1446 	int			error;	/* error return value */
1447 	int			i;	/* temp state */
1448 	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
1449 	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
1450 					/* left is 0, right is 1, prev is 2 */
1451 	int			rval=0;	/* return value (logging flags) */
1452 	int			state = xfs_bmap_fork_to_state(whichfork);
1453 	xfs_filblks_t		da_new; /* new count del alloc blocks used */
1454 	xfs_filblks_t		da_old; /* old count del alloc blocks used */
1455 	xfs_filblks_t		temp=0;	/* value for da_new calculations */
1456 	int			tmp_rval;	/* partial logging flags */
1457 	struct xfs_bmbt_irec	old;
1458 
1459 	ASSERT(whichfork != XFS_ATTR_FORK);
1460 	ASSERT(!isnullstartblock(new->br_startblock));
1461 	ASSERT(!bma->cur ||
1462 	       (bma->cur->bc_ino.flags & XFS_BTCUR_BMBT_WASDEL));
1463 
1464 	XFS_STATS_INC(mp, xs_add_exlist);
1465 
1466 #define	LEFT		r[0]
1467 #define	RIGHT		r[1]
1468 #define	PREV		r[2]
1469 
1470 	/*
1471 	 * Set up a bunch of variables to make the tests simpler.
1472 	 */
1473 	xfs_iext_get_extent(ifp, &bma->icur, &PREV);
1474 	new_endoff = new->br_startoff + new->br_blockcount;
1475 	ASSERT(isnullstartblock(PREV.br_startblock));
1476 	ASSERT(PREV.br_startoff <= new->br_startoff);
1477 	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1478 
1479 	da_old = startblockval(PREV.br_startblock);
1480 	da_new = 0;
1481 
1482 	/*
1483 	 * Set flags determining what part of the previous delayed allocation
1484 	 * extent is being replaced by a real allocation.
1485 	 */
1486 	if (PREV.br_startoff == new->br_startoff)
1487 		state |= BMAP_LEFT_FILLING;
1488 	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1489 		state |= BMAP_RIGHT_FILLING;
1490 
1491 	/*
1492 	 * Check and set flags if this segment has a left neighbor.
1493 	 * Don't set contiguous if the combined extent would be too large.
1494 	 */
1495 	if (xfs_iext_peek_prev_extent(ifp, &bma->icur, &LEFT)) {
1496 		state |= BMAP_LEFT_VALID;
1497 		if (isnullstartblock(LEFT.br_startblock))
1498 			state |= BMAP_LEFT_DELAY;
1499 	}
1500 
1501 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1502 	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1503 	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1504 	    LEFT.br_state == new->br_state &&
1505 	    LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1506 		state |= BMAP_LEFT_CONTIG;
1507 
1508 	/*
1509 	 * Check and set flags if this segment has a right neighbor.
1510 	 * Don't set contiguous if the combined extent would be too large.
1511 	 * Also check for all-three-contiguous being too large.
1512 	 */
1513 	if (xfs_iext_peek_next_extent(ifp, &bma->icur, &RIGHT)) {
1514 		state |= BMAP_RIGHT_VALID;
1515 		if (isnullstartblock(RIGHT.br_startblock))
1516 			state |= BMAP_RIGHT_DELAY;
1517 	}
1518 
1519 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1520 	    new_endoff == RIGHT.br_startoff &&
1521 	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1522 	    new->br_state == RIGHT.br_state &&
1523 	    new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1524 	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1525 		       BMAP_RIGHT_FILLING)) !=
1526 		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1527 		       BMAP_RIGHT_FILLING) ||
1528 	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1529 			<= MAXEXTLEN))
1530 		state |= BMAP_RIGHT_CONTIG;
1531 
1532 	error = 0;
1533 	/*
1534 	 * Switch out based on the FILLING and CONTIG state bits.
1535 	 */
1536 	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1537 			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1538 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1539 	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1540 		/*
1541 		 * Filling in all of a previously delayed allocation extent.
1542 		 * The left and right neighbors are both contiguous with new.
1543 		 */
1544 		LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
1545 
1546 		xfs_iext_remove(bma->ip, &bma->icur, state);
1547 		xfs_iext_remove(bma->ip, &bma->icur, state);
1548 		xfs_iext_prev(ifp, &bma->icur);
1549 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1550 		ifp->if_nextents--;
1551 
1552 		if (bma->cur == NULL)
1553 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1554 		else {
1555 			rval = XFS_ILOG_CORE;
1556 			error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1557 			if (error)
1558 				goto done;
1559 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1560 				error = -EFSCORRUPTED;
1561 				goto done;
1562 			}
1563 			error = xfs_btree_delete(bma->cur, &i);
1564 			if (error)
1565 				goto done;
1566 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1567 				error = -EFSCORRUPTED;
1568 				goto done;
1569 			}
1570 			error = xfs_btree_decrement(bma->cur, 0, &i);
1571 			if (error)
1572 				goto done;
1573 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1574 				error = -EFSCORRUPTED;
1575 				goto done;
1576 			}
1577 			error = xfs_bmbt_update(bma->cur, &LEFT);
1578 			if (error)
1579 				goto done;
1580 		}
1581 		break;
1582 
1583 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1584 		/*
1585 		 * Filling in all of a previously delayed allocation extent.
1586 		 * The left neighbor is contiguous, the right is not.
1587 		 */
1588 		old = LEFT;
1589 		LEFT.br_blockcount += PREV.br_blockcount;
1590 
1591 		xfs_iext_remove(bma->ip, &bma->icur, state);
1592 		xfs_iext_prev(ifp, &bma->icur);
1593 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1594 
1595 		if (bma->cur == NULL)
1596 			rval = XFS_ILOG_DEXT;
1597 		else {
1598 			rval = 0;
1599 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1600 			if (error)
1601 				goto done;
1602 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1603 				error = -EFSCORRUPTED;
1604 				goto done;
1605 			}
1606 			error = xfs_bmbt_update(bma->cur, &LEFT);
1607 			if (error)
1608 				goto done;
1609 		}
1610 		break;
1611 
1612 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1613 		/*
1614 		 * Filling in all of a previously delayed allocation extent.
1615 		 * The right neighbor is contiguous, the left is not. Take care
1616 		 * with delay -> unwritten extent allocation here because the
1617 		 * delalloc record we are overwriting is always written.
1618 		 */
1619 		PREV.br_startblock = new->br_startblock;
1620 		PREV.br_blockcount += RIGHT.br_blockcount;
1621 		PREV.br_state = new->br_state;
1622 
1623 		xfs_iext_next(ifp, &bma->icur);
1624 		xfs_iext_remove(bma->ip, &bma->icur, state);
1625 		xfs_iext_prev(ifp, &bma->icur);
1626 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1627 
1628 		if (bma->cur == NULL)
1629 			rval = XFS_ILOG_DEXT;
1630 		else {
1631 			rval = 0;
1632 			error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1633 			if (error)
1634 				goto done;
1635 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1636 				error = -EFSCORRUPTED;
1637 				goto done;
1638 			}
1639 			error = xfs_bmbt_update(bma->cur, &PREV);
1640 			if (error)
1641 				goto done;
1642 		}
1643 		break;
1644 
1645 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1646 		/*
1647 		 * Filling in all of a previously delayed allocation extent.
1648 		 * Neither the left nor right neighbors are contiguous with
1649 		 * the new one.
1650 		 */
1651 		PREV.br_startblock = new->br_startblock;
1652 		PREV.br_state = new->br_state;
1653 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1654 		ifp->if_nextents++;
1655 
1656 		if (bma->cur == NULL)
1657 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1658 		else {
1659 			rval = XFS_ILOG_CORE;
1660 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1661 			if (error)
1662 				goto done;
1663 			if (XFS_IS_CORRUPT(mp, i != 0)) {
1664 				error = -EFSCORRUPTED;
1665 				goto done;
1666 			}
1667 			error = xfs_btree_insert(bma->cur, &i);
1668 			if (error)
1669 				goto done;
1670 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1671 				error = -EFSCORRUPTED;
1672 				goto done;
1673 			}
1674 		}
1675 		break;
1676 
1677 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1678 		/*
1679 		 * Filling in the first part of a previous delayed allocation.
1680 		 * The left neighbor is contiguous.
1681 		 */
1682 		old = LEFT;
1683 		temp = PREV.br_blockcount - new->br_blockcount;
1684 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1685 				startblockval(PREV.br_startblock));
1686 
1687 		LEFT.br_blockcount += new->br_blockcount;
1688 
1689 		PREV.br_blockcount = temp;
1690 		PREV.br_startoff += new->br_blockcount;
1691 		PREV.br_startblock = nullstartblock(da_new);
1692 
1693 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1694 		xfs_iext_prev(ifp, &bma->icur);
1695 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1696 
1697 		if (bma->cur == NULL)
1698 			rval = XFS_ILOG_DEXT;
1699 		else {
1700 			rval = 0;
1701 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1702 			if (error)
1703 				goto done;
1704 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1705 				error = -EFSCORRUPTED;
1706 				goto done;
1707 			}
1708 			error = xfs_bmbt_update(bma->cur, &LEFT);
1709 			if (error)
1710 				goto done;
1711 		}
1712 		break;
1713 
1714 	case BMAP_LEFT_FILLING:
1715 		/*
1716 		 * Filling in the first part of a previous delayed allocation.
1717 		 * The left neighbor is not contiguous.
1718 		 */
1719 		xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1720 		ifp->if_nextents++;
1721 
1722 		if (bma->cur == NULL)
1723 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1724 		else {
1725 			rval = XFS_ILOG_CORE;
1726 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1727 			if (error)
1728 				goto done;
1729 			if (XFS_IS_CORRUPT(mp, i != 0)) {
1730 				error = -EFSCORRUPTED;
1731 				goto done;
1732 			}
1733 			error = xfs_btree_insert(bma->cur, &i);
1734 			if (error)
1735 				goto done;
1736 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1737 				error = -EFSCORRUPTED;
1738 				goto done;
1739 			}
1740 		}
1741 
1742 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1743 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1744 					&bma->cur, 1, &tmp_rval, whichfork);
1745 			rval |= tmp_rval;
1746 			if (error)
1747 				goto done;
1748 		}
1749 
1750 		temp = PREV.br_blockcount - new->br_blockcount;
1751 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1752 			startblockval(PREV.br_startblock) -
1753 			(bma->cur ? bma->cur->bc_ino.allocated : 0));
1754 
1755 		PREV.br_startoff = new_endoff;
1756 		PREV.br_blockcount = temp;
1757 		PREV.br_startblock = nullstartblock(da_new);
1758 		xfs_iext_next(ifp, &bma->icur);
1759 		xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1760 		xfs_iext_prev(ifp, &bma->icur);
1761 		break;
1762 
1763 	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1764 		/*
1765 		 * Filling in the last part of a previous delayed allocation.
1766 		 * The right neighbor is contiguous with the new allocation.
1767 		 */
1768 		old = RIGHT;
1769 		RIGHT.br_startoff = new->br_startoff;
1770 		RIGHT.br_startblock = new->br_startblock;
1771 		RIGHT.br_blockcount += new->br_blockcount;
1772 
1773 		if (bma->cur == NULL)
1774 			rval = XFS_ILOG_DEXT;
1775 		else {
1776 			rval = 0;
1777 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1778 			if (error)
1779 				goto done;
1780 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1781 				error = -EFSCORRUPTED;
1782 				goto done;
1783 			}
1784 			error = xfs_bmbt_update(bma->cur, &RIGHT);
1785 			if (error)
1786 				goto done;
1787 		}
1788 
1789 		temp = PREV.br_blockcount - new->br_blockcount;
1790 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1791 			startblockval(PREV.br_startblock));
1792 
1793 		PREV.br_blockcount = temp;
1794 		PREV.br_startblock = nullstartblock(da_new);
1795 
1796 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1797 		xfs_iext_next(ifp, &bma->icur);
1798 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &RIGHT);
1799 		break;
1800 
1801 	case BMAP_RIGHT_FILLING:
1802 		/*
1803 		 * Filling in the last part of a previous delayed allocation.
1804 		 * The right neighbor is not contiguous.
1805 		 */
1806 		xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1807 		ifp->if_nextents++;
1808 
1809 		if (bma->cur == NULL)
1810 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1811 		else {
1812 			rval = XFS_ILOG_CORE;
1813 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1814 			if (error)
1815 				goto done;
1816 			if (XFS_IS_CORRUPT(mp, i != 0)) {
1817 				error = -EFSCORRUPTED;
1818 				goto done;
1819 			}
1820 			error = xfs_btree_insert(bma->cur, &i);
1821 			if (error)
1822 				goto done;
1823 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1824 				error = -EFSCORRUPTED;
1825 				goto done;
1826 			}
1827 		}
1828 
1829 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1830 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1831 				&bma->cur, 1, &tmp_rval, whichfork);
1832 			rval |= tmp_rval;
1833 			if (error)
1834 				goto done;
1835 		}
1836 
1837 		temp = PREV.br_blockcount - new->br_blockcount;
1838 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1839 			startblockval(PREV.br_startblock) -
1840 			(bma->cur ? bma->cur->bc_ino.allocated : 0));
1841 
1842 		PREV.br_startblock = nullstartblock(da_new);
1843 		PREV.br_blockcount = temp;
1844 		xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1845 		xfs_iext_next(ifp, &bma->icur);
1846 		break;
1847 
1848 	case 0:
1849 		/*
1850 		 * Filling in the middle part of a previous delayed allocation.
1851 		 * Contiguity is impossible here.
1852 		 * This case is avoided almost all the time.
1853 		 *
1854 		 * We start with a delayed allocation:
1855 		 *
1856 		 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1857 		 *  PREV @ idx
1858 		 *
1859 	         * and we are allocating:
1860 		 *                     +rrrrrrrrrrrrrrrrr+
1861 		 *			      new
1862 		 *
1863 		 * and we set it up for insertion as:
1864 		 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1865 		 *                            new
1866 		 *  PREV @ idx          LEFT              RIGHT
1867 		 *                      inserted at idx + 1
1868 		 */
1869 		old = PREV;
1870 
1871 		/* LEFT is the new middle */
1872 		LEFT = *new;
1873 
1874 		/* RIGHT is the new right */
1875 		RIGHT.br_state = PREV.br_state;
1876 		RIGHT.br_startoff = new_endoff;
1877 		RIGHT.br_blockcount =
1878 			PREV.br_startoff + PREV.br_blockcount - new_endoff;
1879 		RIGHT.br_startblock =
1880 			nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1881 					RIGHT.br_blockcount));
1882 
1883 		/* truncate PREV */
1884 		PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
1885 		PREV.br_startblock =
1886 			nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1887 					PREV.br_blockcount));
1888 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1889 
1890 		xfs_iext_next(ifp, &bma->icur);
1891 		xfs_iext_insert(bma->ip, &bma->icur, &RIGHT, state);
1892 		xfs_iext_insert(bma->ip, &bma->icur, &LEFT, state);
1893 		ifp->if_nextents++;
1894 
1895 		if (bma->cur == NULL)
1896 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1897 		else {
1898 			rval = XFS_ILOG_CORE;
1899 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1900 			if (error)
1901 				goto done;
1902 			if (XFS_IS_CORRUPT(mp, i != 0)) {
1903 				error = -EFSCORRUPTED;
1904 				goto done;
1905 			}
1906 			error = xfs_btree_insert(bma->cur, &i);
1907 			if (error)
1908 				goto done;
1909 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1910 				error = -EFSCORRUPTED;
1911 				goto done;
1912 			}
1913 		}
1914 
1915 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1916 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1917 					&bma->cur, 1, &tmp_rval, whichfork);
1918 			rval |= tmp_rval;
1919 			if (error)
1920 				goto done;
1921 		}
1922 
1923 		da_new = startblockval(PREV.br_startblock) +
1924 			 startblockval(RIGHT.br_startblock);
1925 		break;
1926 
1927 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1928 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1929 	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1930 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1931 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1932 	case BMAP_LEFT_CONTIG:
1933 	case BMAP_RIGHT_CONTIG:
1934 		/*
1935 		 * These cases are all impossible.
1936 		 */
1937 		ASSERT(0);
1938 	}
1939 
1940 	/* add reverse mapping unless caller opted out */
1941 	if (!(bma->flags & XFS_BMAPI_NORMAP))
1942 		xfs_rmap_map_extent(bma->tp, bma->ip, whichfork, new);
1943 
1944 	/* convert to a btree if necessary */
1945 	if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1946 		int	tmp_logflags;	/* partial log flag return val */
1947 
1948 		ASSERT(bma->cur == NULL);
1949 		error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1950 				&bma->cur, da_old > 0, &tmp_logflags,
1951 				whichfork);
1952 		bma->logflags |= tmp_logflags;
1953 		if (error)
1954 			goto done;
1955 	}
1956 
1957 	if (da_new != da_old)
1958 		xfs_mod_delalloc(mp, (int64_t)da_new - da_old);
1959 
1960 	if (bma->cur) {
1961 		da_new += bma->cur->bc_ino.allocated;
1962 		bma->cur->bc_ino.allocated = 0;
1963 	}
1964 
1965 	/* adjust for changes in reserved delayed indirect blocks */
1966 	if (da_new != da_old) {
1967 		ASSERT(state == 0 || da_new < da_old);
1968 		error = xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new),
1969 				false);
1970 	}
1971 
1972 	xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
1973 done:
1974 	if (whichfork != XFS_COW_FORK)
1975 		bma->logflags |= rval;
1976 	return error;
1977 #undef	LEFT
1978 #undef	RIGHT
1979 #undef	PREV
1980 }
1981 
1982 /*
1983  * Convert an unwritten allocation to a real allocation or vice versa.
1984  */
1985 int					/* error */
1986 xfs_bmap_add_extent_unwritten_real(
1987 	struct xfs_trans	*tp,
1988 	xfs_inode_t		*ip,	/* incore inode pointer */
1989 	int			whichfork,
1990 	struct xfs_iext_cursor	*icur,
1991 	xfs_btree_cur_t		**curp,	/* if *curp is null, not a btree */
1992 	xfs_bmbt_irec_t		*new,	/* new data to add to file extents */
1993 	int			*logflagsp) /* inode logging flags */
1994 {
1995 	xfs_btree_cur_t		*cur;	/* btree cursor */
1996 	int			error;	/* error return value */
1997 	int			i;	/* temp state */
1998 	struct xfs_ifork	*ifp;	/* inode fork pointer */
1999 	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
2000 	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
2001 					/* left is 0, right is 1, prev is 2 */
2002 	int			rval=0;	/* return value (logging flags) */
2003 	int			state = xfs_bmap_fork_to_state(whichfork);
2004 	struct xfs_mount	*mp = ip->i_mount;
2005 	struct xfs_bmbt_irec	old;
2006 
2007 	*logflagsp = 0;
2008 
2009 	cur = *curp;
2010 	ifp = XFS_IFORK_PTR(ip, whichfork);
2011 
2012 	ASSERT(!isnullstartblock(new->br_startblock));
2013 
2014 	XFS_STATS_INC(mp, xs_add_exlist);
2015 
2016 #define	LEFT		r[0]
2017 #define	RIGHT		r[1]
2018 #define	PREV		r[2]
2019 
2020 	/*
2021 	 * Set up a bunch of variables to make the tests simpler.
2022 	 */
2023 	error = 0;
2024 	xfs_iext_get_extent(ifp, icur, &PREV);
2025 	ASSERT(new->br_state != PREV.br_state);
2026 	new_endoff = new->br_startoff + new->br_blockcount;
2027 	ASSERT(PREV.br_startoff <= new->br_startoff);
2028 	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2029 
2030 	/*
2031 	 * Set flags determining what part of the previous oldext allocation
2032 	 * extent is being replaced by a newext allocation.
2033 	 */
2034 	if (PREV.br_startoff == new->br_startoff)
2035 		state |= BMAP_LEFT_FILLING;
2036 	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2037 		state |= BMAP_RIGHT_FILLING;
2038 
2039 	/*
2040 	 * Check and set flags if this segment has a left neighbor.
2041 	 * Don't set contiguous if the combined extent would be too large.
2042 	 */
2043 	if (xfs_iext_peek_prev_extent(ifp, icur, &LEFT)) {
2044 		state |= BMAP_LEFT_VALID;
2045 		if (isnullstartblock(LEFT.br_startblock))
2046 			state |= BMAP_LEFT_DELAY;
2047 	}
2048 
2049 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2050 	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2051 	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2052 	    LEFT.br_state == new->br_state &&
2053 	    LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2054 		state |= BMAP_LEFT_CONTIG;
2055 
2056 	/*
2057 	 * Check and set flags if this segment has a right neighbor.
2058 	 * Don't set contiguous if the combined extent would be too large.
2059 	 * Also check for all-three-contiguous being too large.
2060 	 */
2061 	if (xfs_iext_peek_next_extent(ifp, icur, &RIGHT)) {
2062 		state |= BMAP_RIGHT_VALID;
2063 		if (isnullstartblock(RIGHT.br_startblock))
2064 			state |= BMAP_RIGHT_DELAY;
2065 	}
2066 
2067 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2068 	    new_endoff == RIGHT.br_startoff &&
2069 	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2070 	    new->br_state == RIGHT.br_state &&
2071 	    new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2072 	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2073 		       BMAP_RIGHT_FILLING)) !=
2074 		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2075 		       BMAP_RIGHT_FILLING) ||
2076 	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2077 			<= MAXEXTLEN))
2078 		state |= BMAP_RIGHT_CONTIG;
2079 
2080 	/*
2081 	 * Switch out based on the FILLING and CONTIG state bits.
2082 	 */
2083 	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2084 			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2085 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2086 	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2087 		/*
2088 		 * Setting all of a previous oldext extent to newext.
2089 		 * The left and right neighbors are both contiguous with new.
2090 		 */
2091 		LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
2092 
2093 		xfs_iext_remove(ip, icur, state);
2094 		xfs_iext_remove(ip, icur, state);
2095 		xfs_iext_prev(ifp, icur);
2096 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2097 		ifp->if_nextents -= 2;
2098 		if (cur == NULL)
2099 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2100 		else {
2101 			rval = XFS_ILOG_CORE;
2102 			error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2103 			if (error)
2104 				goto done;
2105 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2106 				error = -EFSCORRUPTED;
2107 				goto done;
2108 			}
2109 			if ((error = xfs_btree_delete(cur, &i)))
2110 				goto done;
2111 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2112 				error = -EFSCORRUPTED;
2113 				goto done;
2114 			}
2115 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2116 				goto done;
2117 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2118 				error = -EFSCORRUPTED;
2119 				goto done;
2120 			}
2121 			if ((error = xfs_btree_delete(cur, &i)))
2122 				goto done;
2123 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2124 				error = -EFSCORRUPTED;
2125 				goto done;
2126 			}
2127 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2128 				goto done;
2129 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2130 				error = -EFSCORRUPTED;
2131 				goto done;
2132 			}
2133 			error = xfs_bmbt_update(cur, &LEFT);
2134 			if (error)
2135 				goto done;
2136 		}
2137 		break;
2138 
2139 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2140 		/*
2141 		 * Setting all of a previous oldext extent to newext.
2142 		 * The left neighbor is contiguous, the right is not.
2143 		 */
2144 		LEFT.br_blockcount += PREV.br_blockcount;
2145 
2146 		xfs_iext_remove(ip, icur, state);
2147 		xfs_iext_prev(ifp, icur);
2148 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2149 		ifp->if_nextents--;
2150 		if (cur == NULL)
2151 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2152 		else {
2153 			rval = XFS_ILOG_CORE;
2154 			error = xfs_bmbt_lookup_eq(cur, &PREV, &i);
2155 			if (error)
2156 				goto done;
2157 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2158 				error = -EFSCORRUPTED;
2159 				goto done;
2160 			}
2161 			if ((error = xfs_btree_delete(cur, &i)))
2162 				goto done;
2163 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2164 				error = -EFSCORRUPTED;
2165 				goto done;
2166 			}
2167 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2168 				goto done;
2169 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2170 				error = -EFSCORRUPTED;
2171 				goto done;
2172 			}
2173 			error = xfs_bmbt_update(cur, &LEFT);
2174 			if (error)
2175 				goto done;
2176 		}
2177 		break;
2178 
2179 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2180 		/*
2181 		 * Setting all of a previous oldext extent to newext.
2182 		 * The right neighbor is contiguous, the left is not.
2183 		 */
2184 		PREV.br_blockcount += RIGHT.br_blockcount;
2185 		PREV.br_state = new->br_state;
2186 
2187 		xfs_iext_next(ifp, icur);
2188 		xfs_iext_remove(ip, icur, state);
2189 		xfs_iext_prev(ifp, icur);
2190 		xfs_iext_update_extent(ip, state, icur, &PREV);
2191 		ifp->if_nextents--;
2192 
2193 		if (cur == NULL)
2194 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2195 		else {
2196 			rval = XFS_ILOG_CORE;
2197 			error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2198 			if (error)
2199 				goto done;
2200 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2201 				error = -EFSCORRUPTED;
2202 				goto done;
2203 			}
2204 			if ((error = xfs_btree_delete(cur, &i)))
2205 				goto done;
2206 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2207 				error = -EFSCORRUPTED;
2208 				goto done;
2209 			}
2210 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2211 				goto done;
2212 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2213 				error = -EFSCORRUPTED;
2214 				goto done;
2215 			}
2216 			error = xfs_bmbt_update(cur, &PREV);
2217 			if (error)
2218 				goto done;
2219 		}
2220 		break;
2221 
2222 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2223 		/*
2224 		 * Setting all of a previous oldext extent to newext.
2225 		 * Neither the left nor right neighbors are contiguous with
2226 		 * the new one.
2227 		 */
2228 		PREV.br_state = new->br_state;
2229 		xfs_iext_update_extent(ip, state, icur, &PREV);
2230 
2231 		if (cur == NULL)
2232 			rval = XFS_ILOG_DEXT;
2233 		else {
2234 			rval = 0;
2235 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2236 			if (error)
2237 				goto done;
2238 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2239 				error = -EFSCORRUPTED;
2240 				goto done;
2241 			}
2242 			error = xfs_bmbt_update(cur, &PREV);
2243 			if (error)
2244 				goto done;
2245 		}
2246 		break;
2247 
2248 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2249 		/*
2250 		 * Setting the first part of a previous oldext extent to newext.
2251 		 * The left neighbor is contiguous.
2252 		 */
2253 		LEFT.br_blockcount += new->br_blockcount;
2254 
2255 		old = PREV;
2256 		PREV.br_startoff += new->br_blockcount;
2257 		PREV.br_startblock += new->br_blockcount;
2258 		PREV.br_blockcount -= new->br_blockcount;
2259 
2260 		xfs_iext_update_extent(ip, state, icur, &PREV);
2261 		xfs_iext_prev(ifp, icur);
2262 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2263 
2264 		if (cur == NULL)
2265 			rval = XFS_ILOG_DEXT;
2266 		else {
2267 			rval = 0;
2268 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2269 			if (error)
2270 				goto done;
2271 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2272 				error = -EFSCORRUPTED;
2273 				goto done;
2274 			}
2275 			error = xfs_bmbt_update(cur, &PREV);
2276 			if (error)
2277 				goto done;
2278 			error = xfs_btree_decrement(cur, 0, &i);
2279 			if (error)
2280 				goto done;
2281 			error = xfs_bmbt_update(cur, &LEFT);
2282 			if (error)
2283 				goto done;
2284 		}
2285 		break;
2286 
2287 	case BMAP_LEFT_FILLING:
2288 		/*
2289 		 * Setting the first part of a previous oldext extent to newext.
2290 		 * The left neighbor is not contiguous.
2291 		 */
2292 		old = PREV;
2293 		PREV.br_startoff += new->br_blockcount;
2294 		PREV.br_startblock += new->br_blockcount;
2295 		PREV.br_blockcount -= new->br_blockcount;
2296 
2297 		xfs_iext_update_extent(ip, state, icur, &PREV);
2298 		xfs_iext_insert(ip, icur, new, state);
2299 		ifp->if_nextents++;
2300 
2301 		if (cur == NULL)
2302 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2303 		else {
2304 			rval = XFS_ILOG_CORE;
2305 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2306 			if (error)
2307 				goto done;
2308 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2309 				error = -EFSCORRUPTED;
2310 				goto done;
2311 			}
2312 			error = xfs_bmbt_update(cur, &PREV);
2313 			if (error)
2314 				goto done;
2315 			cur->bc_rec.b = *new;
2316 			if ((error = xfs_btree_insert(cur, &i)))
2317 				goto done;
2318 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2319 				error = -EFSCORRUPTED;
2320 				goto done;
2321 			}
2322 		}
2323 		break;
2324 
2325 	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2326 		/*
2327 		 * Setting the last part of a previous oldext extent to newext.
2328 		 * The right neighbor is contiguous with the new allocation.
2329 		 */
2330 		old = PREV;
2331 		PREV.br_blockcount -= new->br_blockcount;
2332 
2333 		RIGHT.br_startoff = new->br_startoff;
2334 		RIGHT.br_startblock = new->br_startblock;
2335 		RIGHT.br_blockcount += new->br_blockcount;
2336 
2337 		xfs_iext_update_extent(ip, state, icur, &PREV);
2338 		xfs_iext_next(ifp, icur);
2339 		xfs_iext_update_extent(ip, state, icur, &RIGHT);
2340 
2341 		if (cur == NULL)
2342 			rval = XFS_ILOG_DEXT;
2343 		else {
2344 			rval = 0;
2345 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2346 			if (error)
2347 				goto done;
2348 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2349 				error = -EFSCORRUPTED;
2350 				goto done;
2351 			}
2352 			error = xfs_bmbt_update(cur, &PREV);
2353 			if (error)
2354 				goto done;
2355 			error = xfs_btree_increment(cur, 0, &i);
2356 			if (error)
2357 				goto done;
2358 			error = xfs_bmbt_update(cur, &RIGHT);
2359 			if (error)
2360 				goto done;
2361 		}
2362 		break;
2363 
2364 	case BMAP_RIGHT_FILLING:
2365 		/*
2366 		 * Setting the last part of a previous oldext extent to newext.
2367 		 * The right neighbor is not contiguous.
2368 		 */
2369 		old = PREV;
2370 		PREV.br_blockcount -= new->br_blockcount;
2371 
2372 		xfs_iext_update_extent(ip, state, icur, &PREV);
2373 		xfs_iext_next(ifp, icur);
2374 		xfs_iext_insert(ip, icur, new, state);
2375 		ifp->if_nextents++;
2376 
2377 		if (cur == NULL)
2378 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2379 		else {
2380 			rval = XFS_ILOG_CORE;
2381 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2382 			if (error)
2383 				goto done;
2384 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2385 				error = -EFSCORRUPTED;
2386 				goto done;
2387 			}
2388 			error = xfs_bmbt_update(cur, &PREV);
2389 			if (error)
2390 				goto done;
2391 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2392 			if (error)
2393 				goto done;
2394 			if (XFS_IS_CORRUPT(mp, i != 0)) {
2395 				error = -EFSCORRUPTED;
2396 				goto done;
2397 			}
2398 			if ((error = xfs_btree_insert(cur, &i)))
2399 				goto done;
2400 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2401 				error = -EFSCORRUPTED;
2402 				goto done;
2403 			}
2404 		}
2405 		break;
2406 
2407 	case 0:
2408 		/*
2409 		 * Setting the middle part of a previous oldext extent to
2410 		 * newext.  Contiguity is impossible here.
2411 		 * One extent becomes three extents.
2412 		 */
2413 		old = PREV;
2414 		PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
2415 
2416 		r[0] = *new;
2417 		r[1].br_startoff = new_endoff;
2418 		r[1].br_blockcount =
2419 			old.br_startoff + old.br_blockcount - new_endoff;
2420 		r[1].br_startblock = new->br_startblock + new->br_blockcount;
2421 		r[1].br_state = PREV.br_state;
2422 
2423 		xfs_iext_update_extent(ip, state, icur, &PREV);
2424 		xfs_iext_next(ifp, icur);
2425 		xfs_iext_insert(ip, icur, &r[1], state);
2426 		xfs_iext_insert(ip, icur, &r[0], state);
2427 		ifp->if_nextents += 2;
2428 
2429 		if (cur == NULL)
2430 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2431 		else {
2432 			rval = XFS_ILOG_CORE;
2433 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2434 			if (error)
2435 				goto done;
2436 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2437 				error = -EFSCORRUPTED;
2438 				goto done;
2439 			}
2440 			/* new right extent - oldext */
2441 			error = xfs_bmbt_update(cur, &r[1]);
2442 			if (error)
2443 				goto done;
2444 			/* new left extent - oldext */
2445 			cur->bc_rec.b = PREV;
2446 			if ((error = xfs_btree_insert(cur, &i)))
2447 				goto done;
2448 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2449 				error = -EFSCORRUPTED;
2450 				goto done;
2451 			}
2452 			/*
2453 			 * Reset the cursor to the position of the new extent
2454 			 * we are about to insert as we can't trust it after
2455 			 * the previous insert.
2456 			 */
2457 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2458 			if (error)
2459 				goto done;
2460 			if (XFS_IS_CORRUPT(mp, i != 0)) {
2461 				error = -EFSCORRUPTED;
2462 				goto done;
2463 			}
2464 			/* new middle extent - newext */
2465 			if ((error = xfs_btree_insert(cur, &i)))
2466 				goto done;
2467 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2468 				error = -EFSCORRUPTED;
2469 				goto done;
2470 			}
2471 		}
2472 		break;
2473 
2474 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2475 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2476 	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2477 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2478 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2479 	case BMAP_LEFT_CONTIG:
2480 	case BMAP_RIGHT_CONTIG:
2481 		/*
2482 		 * These cases are all impossible.
2483 		 */
2484 		ASSERT(0);
2485 	}
2486 
2487 	/* update reverse mappings */
2488 	xfs_rmap_convert_extent(mp, tp, ip, whichfork, new);
2489 
2490 	/* convert to a btree if necessary */
2491 	if (xfs_bmap_needs_btree(ip, whichfork)) {
2492 		int	tmp_logflags;	/* partial log flag return val */
2493 
2494 		ASSERT(cur == NULL);
2495 		error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
2496 				&tmp_logflags, whichfork);
2497 		*logflagsp |= tmp_logflags;
2498 		if (error)
2499 			goto done;
2500 	}
2501 
2502 	/* clear out the allocated field, done with it now in any case. */
2503 	if (cur) {
2504 		cur->bc_ino.allocated = 0;
2505 		*curp = cur;
2506 	}
2507 
2508 	xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
2509 done:
2510 	*logflagsp |= rval;
2511 	return error;
2512 #undef	LEFT
2513 #undef	RIGHT
2514 #undef	PREV
2515 }
2516 
2517 /*
2518  * Convert a hole to a delayed allocation.
2519  */
2520 STATIC void
2521 xfs_bmap_add_extent_hole_delay(
2522 	xfs_inode_t		*ip,	/* incore inode pointer */
2523 	int			whichfork,
2524 	struct xfs_iext_cursor	*icur,
2525 	xfs_bmbt_irec_t		*new)	/* new data to add to file extents */
2526 {
2527 	struct xfs_ifork	*ifp;	/* inode fork pointer */
2528 	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2529 	xfs_filblks_t		newlen=0;	/* new indirect size */
2530 	xfs_filblks_t		oldlen=0;	/* old indirect size */
2531 	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2532 	int			state = xfs_bmap_fork_to_state(whichfork);
2533 	xfs_filblks_t		temp;	 /* temp for indirect calculations */
2534 
2535 	ifp = XFS_IFORK_PTR(ip, whichfork);
2536 	ASSERT(isnullstartblock(new->br_startblock));
2537 
2538 	/*
2539 	 * Check and set flags if this segment has a left neighbor
2540 	 */
2541 	if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2542 		state |= BMAP_LEFT_VALID;
2543 		if (isnullstartblock(left.br_startblock))
2544 			state |= BMAP_LEFT_DELAY;
2545 	}
2546 
2547 	/*
2548 	 * Check and set flags if the current (right) segment exists.
2549 	 * If it doesn't exist, we're converting the hole at end-of-file.
2550 	 */
2551 	if (xfs_iext_get_extent(ifp, icur, &right)) {
2552 		state |= BMAP_RIGHT_VALID;
2553 		if (isnullstartblock(right.br_startblock))
2554 			state |= BMAP_RIGHT_DELAY;
2555 	}
2556 
2557 	/*
2558 	 * Set contiguity flags on the left and right neighbors.
2559 	 * Don't let extents get too large, even if the pieces are contiguous.
2560 	 */
2561 	if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2562 	    left.br_startoff + left.br_blockcount == new->br_startoff &&
2563 	    left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2564 		state |= BMAP_LEFT_CONTIG;
2565 
2566 	if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2567 	    new->br_startoff + new->br_blockcount == right.br_startoff &&
2568 	    new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2569 	    (!(state & BMAP_LEFT_CONTIG) ||
2570 	     (left.br_blockcount + new->br_blockcount +
2571 	      right.br_blockcount <= MAXEXTLEN)))
2572 		state |= BMAP_RIGHT_CONTIG;
2573 
2574 	/*
2575 	 * Switch out based on the contiguity flags.
2576 	 */
2577 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2578 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2579 		/*
2580 		 * New allocation is contiguous with delayed allocations
2581 		 * on the left and on the right.
2582 		 * Merge all three into a single extent record.
2583 		 */
2584 		temp = left.br_blockcount + new->br_blockcount +
2585 			right.br_blockcount;
2586 
2587 		oldlen = startblockval(left.br_startblock) +
2588 			startblockval(new->br_startblock) +
2589 			startblockval(right.br_startblock);
2590 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2591 					 oldlen);
2592 		left.br_startblock = nullstartblock(newlen);
2593 		left.br_blockcount = temp;
2594 
2595 		xfs_iext_remove(ip, icur, state);
2596 		xfs_iext_prev(ifp, icur);
2597 		xfs_iext_update_extent(ip, state, icur, &left);
2598 		break;
2599 
2600 	case BMAP_LEFT_CONTIG:
2601 		/*
2602 		 * New allocation is contiguous with a delayed allocation
2603 		 * on the left.
2604 		 * Merge the new allocation with the left neighbor.
2605 		 */
2606 		temp = left.br_blockcount + new->br_blockcount;
2607 
2608 		oldlen = startblockval(left.br_startblock) +
2609 			startblockval(new->br_startblock);
2610 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2611 					 oldlen);
2612 		left.br_blockcount = temp;
2613 		left.br_startblock = nullstartblock(newlen);
2614 
2615 		xfs_iext_prev(ifp, icur);
2616 		xfs_iext_update_extent(ip, state, icur, &left);
2617 		break;
2618 
2619 	case BMAP_RIGHT_CONTIG:
2620 		/*
2621 		 * New allocation is contiguous with a delayed allocation
2622 		 * on the right.
2623 		 * Merge the new allocation with the right neighbor.
2624 		 */
2625 		temp = new->br_blockcount + right.br_blockcount;
2626 		oldlen = startblockval(new->br_startblock) +
2627 			startblockval(right.br_startblock);
2628 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2629 					 oldlen);
2630 		right.br_startoff = new->br_startoff;
2631 		right.br_startblock = nullstartblock(newlen);
2632 		right.br_blockcount = temp;
2633 		xfs_iext_update_extent(ip, state, icur, &right);
2634 		break;
2635 
2636 	case 0:
2637 		/*
2638 		 * New allocation is not contiguous with another
2639 		 * delayed allocation.
2640 		 * Insert a new entry.
2641 		 */
2642 		oldlen = newlen = 0;
2643 		xfs_iext_insert(ip, icur, new, state);
2644 		break;
2645 	}
2646 	if (oldlen != newlen) {
2647 		ASSERT(oldlen > newlen);
2648 		xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2649 				 false);
2650 		/*
2651 		 * Nothing to do for disk quota accounting here.
2652 		 */
2653 		xfs_mod_delalloc(ip->i_mount, (int64_t)newlen - oldlen);
2654 	}
2655 }
2656 
2657 /*
2658  * Convert a hole to a real allocation.
2659  */
2660 STATIC int				/* error */
2661 xfs_bmap_add_extent_hole_real(
2662 	struct xfs_trans	*tp,
2663 	struct xfs_inode	*ip,
2664 	int			whichfork,
2665 	struct xfs_iext_cursor	*icur,
2666 	struct xfs_btree_cur	**curp,
2667 	struct xfs_bmbt_irec	*new,
2668 	int			*logflagsp,
2669 	int			flags)
2670 {
2671 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
2672 	struct xfs_mount	*mp = ip->i_mount;
2673 	struct xfs_btree_cur	*cur = *curp;
2674 	int			error;	/* error return value */
2675 	int			i;	/* temp state */
2676 	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2677 	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2678 	int			rval=0;	/* return value (logging flags) */
2679 	int			state = xfs_bmap_fork_to_state(whichfork);
2680 	struct xfs_bmbt_irec	old;
2681 
2682 	ASSERT(!isnullstartblock(new->br_startblock));
2683 	ASSERT(!cur || !(cur->bc_ino.flags & XFS_BTCUR_BMBT_WASDEL));
2684 
2685 	XFS_STATS_INC(mp, xs_add_exlist);
2686 
2687 	/*
2688 	 * Check and set flags if this segment has a left neighbor.
2689 	 */
2690 	if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2691 		state |= BMAP_LEFT_VALID;
2692 		if (isnullstartblock(left.br_startblock))
2693 			state |= BMAP_LEFT_DELAY;
2694 	}
2695 
2696 	/*
2697 	 * Check and set flags if this segment has a current value.
2698 	 * Not true if we're inserting into the "hole" at eof.
2699 	 */
2700 	if (xfs_iext_get_extent(ifp, icur, &right)) {
2701 		state |= BMAP_RIGHT_VALID;
2702 		if (isnullstartblock(right.br_startblock))
2703 			state |= BMAP_RIGHT_DELAY;
2704 	}
2705 
2706 	/*
2707 	 * We're inserting a real allocation between "left" and "right".
2708 	 * Set the contiguity flags.  Don't let extents get too large.
2709 	 */
2710 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2711 	    left.br_startoff + left.br_blockcount == new->br_startoff &&
2712 	    left.br_startblock + left.br_blockcount == new->br_startblock &&
2713 	    left.br_state == new->br_state &&
2714 	    left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2715 		state |= BMAP_LEFT_CONTIG;
2716 
2717 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2718 	    new->br_startoff + new->br_blockcount == right.br_startoff &&
2719 	    new->br_startblock + new->br_blockcount == right.br_startblock &&
2720 	    new->br_state == right.br_state &&
2721 	    new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2722 	    (!(state & BMAP_LEFT_CONTIG) ||
2723 	     left.br_blockcount + new->br_blockcount +
2724 	     right.br_blockcount <= MAXEXTLEN))
2725 		state |= BMAP_RIGHT_CONTIG;
2726 
2727 	error = 0;
2728 	/*
2729 	 * Select which case we're in here, and implement it.
2730 	 */
2731 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2732 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2733 		/*
2734 		 * New allocation is contiguous with real allocations on the
2735 		 * left and on the right.
2736 		 * Merge all three into a single extent record.
2737 		 */
2738 		left.br_blockcount += new->br_blockcount + right.br_blockcount;
2739 
2740 		xfs_iext_remove(ip, icur, state);
2741 		xfs_iext_prev(ifp, icur);
2742 		xfs_iext_update_extent(ip, state, icur, &left);
2743 		ifp->if_nextents--;
2744 
2745 		if (cur == NULL) {
2746 			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2747 		} else {
2748 			rval = XFS_ILOG_CORE;
2749 			error = xfs_bmbt_lookup_eq(cur, &right, &i);
2750 			if (error)
2751 				goto done;
2752 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2753 				error = -EFSCORRUPTED;
2754 				goto done;
2755 			}
2756 			error = xfs_btree_delete(cur, &i);
2757 			if (error)
2758 				goto done;
2759 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2760 				error = -EFSCORRUPTED;
2761 				goto done;
2762 			}
2763 			error = xfs_btree_decrement(cur, 0, &i);
2764 			if (error)
2765 				goto done;
2766 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2767 				error = -EFSCORRUPTED;
2768 				goto done;
2769 			}
2770 			error = xfs_bmbt_update(cur, &left);
2771 			if (error)
2772 				goto done;
2773 		}
2774 		break;
2775 
2776 	case BMAP_LEFT_CONTIG:
2777 		/*
2778 		 * New allocation is contiguous with a real allocation
2779 		 * on the left.
2780 		 * Merge the new allocation with the left neighbor.
2781 		 */
2782 		old = left;
2783 		left.br_blockcount += new->br_blockcount;
2784 
2785 		xfs_iext_prev(ifp, icur);
2786 		xfs_iext_update_extent(ip, state, icur, &left);
2787 
2788 		if (cur == NULL) {
2789 			rval = xfs_ilog_fext(whichfork);
2790 		} else {
2791 			rval = 0;
2792 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2793 			if (error)
2794 				goto done;
2795 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2796 				error = -EFSCORRUPTED;
2797 				goto done;
2798 			}
2799 			error = xfs_bmbt_update(cur, &left);
2800 			if (error)
2801 				goto done;
2802 		}
2803 		break;
2804 
2805 	case BMAP_RIGHT_CONTIG:
2806 		/*
2807 		 * New allocation is contiguous with a real allocation
2808 		 * on the right.
2809 		 * Merge the new allocation with the right neighbor.
2810 		 */
2811 		old = right;
2812 
2813 		right.br_startoff = new->br_startoff;
2814 		right.br_startblock = new->br_startblock;
2815 		right.br_blockcount += new->br_blockcount;
2816 		xfs_iext_update_extent(ip, state, icur, &right);
2817 
2818 		if (cur == NULL) {
2819 			rval = xfs_ilog_fext(whichfork);
2820 		} else {
2821 			rval = 0;
2822 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2823 			if (error)
2824 				goto done;
2825 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2826 				error = -EFSCORRUPTED;
2827 				goto done;
2828 			}
2829 			error = xfs_bmbt_update(cur, &right);
2830 			if (error)
2831 				goto done;
2832 		}
2833 		break;
2834 
2835 	case 0:
2836 		/*
2837 		 * New allocation is not contiguous with another
2838 		 * real allocation.
2839 		 * Insert a new entry.
2840 		 */
2841 		xfs_iext_insert(ip, icur, new, state);
2842 		ifp->if_nextents++;
2843 
2844 		if (cur == NULL) {
2845 			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2846 		} else {
2847 			rval = XFS_ILOG_CORE;
2848 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2849 			if (error)
2850 				goto done;
2851 			if (XFS_IS_CORRUPT(mp, i != 0)) {
2852 				error = -EFSCORRUPTED;
2853 				goto done;
2854 			}
2855 			error = xfs_btree_insert(cur, &i);
2856 			if (error)
2857 				goto done;
2858 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2859 				error = -EFSCORRUPTED;
2860 				goto done;
2861 			}
2862 		}
2863 		break;
2864 	}
2865 
2866 	/* add reverse mapping unless caller opted out */
2867 	if (!(flags & XFS_BMAPI_NORMAP))
2868 		xfs_rmap_map_extent(tp, ip, whichfork, new);
2869 
2870 	/* convert to a btree if necessary */
2871 	if (xfs_bmap_needs_btree(ip, whichfork)) {
2872 		int	tmp_logflags;	/* partial log flag return val */
2873 
2874 		ASSERT(cur == NULL);
2875 		error = xfs_bmap_extents_to_btree(tp, ip, curp, 0,
2876 				&tmp_logflags, whichfork);
2877 		*logflagsp |= tmp_logflags;
2878 		cur = *curp;
2879 		if (error)
2880 			goto done;
2881 	}
2882 
2883 	/* clear out the allocated field, done with it now in any case. */
2884 	if (cur)
2885 		cur->bc_ino.allocated = 0;
2886 
2887 	xfs_bmap_check_leaf_extents(cur, ip, whichfork);
2888 done:
2889 	*logflagsp |= rval;
2890 	return error;
2891 }
2892 
2893 /*
2894  * Functions used in the extent read, allocate and remove paths
2895  */
2896 
2897 /*
2898  * Adjust the size of the new extent based on i_extsize and rt extsize.
2899  */
2900 int
2901 xfs_bmap_extsize_align(
2902 	xfs_mount_t	*mp,
2903 	xfs_bmbt_irec_t	*gotp,		/* next extent pointer */
2904 	xfs_bmbt_irec_t	*prevp,		/* previous extent pointer */
2905 	xfs_extlen_t	extsz,		/* align to this extent size */
2906 	int		rt,		/* is this a realtime inode? */
2907 	int		eof,		/* is extent at end-of-file? */
2908 	int		delay,		/* creating delalloc extent? */
2909 	int		convert,	/* overwriting unwritten extent? */
2910 	xfs_fileoff_t	*offp,		/* in/out: aligned offset */
2911 	xfs_extlen_t	*lenp)		/* in/out: aligned length */
2912 {
2913 	xfs_fileoff_t	orig_off;	/* original offset */
2914 	xfs_extlen_t	orig_alen;	/* original length */
2915 	xfs_fileoff_t	orig_end;	/* original off+len */
2916 	xfs_fileoff_t	nexto;		/* next file offset */
2917 	xfs_fileoff_t	prevo;		/* previous file offset */
2918 	xfs_fileoff_t	align_off;	/* temp for offset */
2919 	xfs_extlen_t	align_alen;	/* temp for length */
2920 	xfs_extlen_t	temp;		/* temp for calculations */
2921 
2922 	if (convert)
2923 		return 0;
2924 
2925 	orig_off = align_off = *offp;
2926 	orig_alen = align_alen = *lenp;
2927 	orig_end = orig_off + orig_alen;
2928 
2929 	/*
2930 	 * If this request overlaps an existing extent, then don't
2931 	 * attempt to perform any additional alignment.
2932 	 */
2933 	if (!delay && !eof &&
2934 	    (orig_off >= gotp->br_startoff) &&
2935 	    (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2936 		return 0;
2937 	}
2938 
2939 	/*
2940 	 * If the file offset is unaligned vs. the extent size
2941 	 * we need to align it.  This will be possible unless
2942 	 * the file was previously written with a kernel that didn't
2943 	 * perform this alignment, or if a truncate shot us in the
2944 	 * foot.
2945 	 */
2946 	div_u64_rem(orig_off, extsz, &temp);
2947 	if (temp) {
2948 		align_alen += temp;
2949 		align_off -= temp;
2950 	}
2951 
2952 	/* Same adjustment for the end of the requested area. */
2953 	temp = (align_alen % extsz);
2954 	if (temp)
2955 		align_alen += extsz - temp;
2956 
2957 	/*
2958 	 * For large extent hint sizes, the aligned extent might be larger than
2959 	 * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
2960 	 * the length back under MAXEXTLEN. The outer allocation loops handle
2961 	 * short allocation just fine, so it is safe to do this. We only want to
2962 	 * do it when we are forced to, though, because it means more allocation
2963 	 * operations are required.
2964 	 */
2965 	while (align_alen > MAXEXTLEN)
2966 		align_alen -= extsz;
2967 	ASSERT(align_alen <= MAXEXTLEN);
2968 
2969 	/*
2970 	 * If the previous block overlaps with this proposed allocation
2971 	 * then move the start forward without adjusting the length.
2972 	 */
2973 	if (prevp->br_startoff != NULLFILEOFF) {
2974 		if (prevp->br_startblock == HOLESTARTBLOCK)
2975 			prevo = prevp->br_startoff;
2976 		else
2977 			prevo = prevp->br_startoff + prevp->br_blockcount;
2978 	} else
2979 		prevo = 0;
2980 	if (align_off != orig_off && align_off < prevo)
2981 		align_off = prevo;
2982 	/*
2983 	 * If the next block overlaps with this proposed allocation
2984 	 * then move the start back without adjusting the length,
2985 	 * but not before offset 0.
2986 	 * This may of course make the start overlap previous block,
2987 	 * and if we hit the offset 0 limit then the next block
2988 	 * can still overlap too.
2989 	 */
2990 	if (!eof && gotp->br_startoff != NULLFILEOFF) {
2991 		if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2992 		    (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2993 			nexto = gotp->br_startoff + gotp->br_blockcount;
2994 		else
2995 			nexto = gotp->br_startoff;
2996 	} else
2997 		nexto = NULLFILEOFF;
2998 	if (!eof &&
2999 	    align_off + align_alen != orig_end &&
3000 	    align_off + align_alen > nexto)
3001 		align_off = nexto > align_alen ? nexto - align_alen : 0;
3002 	/*
3003 	 * If we're now overlapping the next or previous extent that
3004 	 * means we can't fit an extsz piece in this hole.  Just move
3005 	 * the start forward to the first valid spot and set
3006 	 * the length so we hit the end.
3007 	 */
3008 	if (align_off != orig_off && align_off < prevo)
3009 		align_off = prevo;
3010 	if (align_off + align_alen != orig_end &&
3011 	    align_off + align_alen > nexto &&
3012 	    nexto != NULLFILEOFF) {
3013 		ASSERT(nexto > prevo);
3014 		align_alen = nexto - align_off;
3015 	}
3016 
3017 	/*
3018 	 * If realtime, and the result isn't a multiple of the realtime
3019 	 * extent size we need to remove blocks until it is.
3020 	 */
3021 	if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3022 		/*
3023 		 * We're not covering the original request, or
3024 		 * we won't be able to once we fix the length.
3025 		 */
3026 		if (orig_off < align_off ||
3027 		    orig_end > align_off + align_alen ||
3028 		    align_alen - temp < orig_alen)
3029 			return -EINVAL;
3030 		/*
3031 		 * Try to fix it by moving the start up.
3032 		 */
3033 		if (align_off + temp <= orig_off) {
3034 			align_alen -= temp;
3035 			align_off += temp;
3036 		}
3037 		/*
3038 		 * Try to fix it by moving the end in.
3039 		 */
3040 		else if (align_off + align_alen - temp >= orig_end)
3041 			align_alen -= temp;
3042 		/*
3043 		 * Set the start to the minimum then trim the length.
3044 		 */
3045 		else {
3046 			align_alen -= orig_off - align_off;
3047 			align_off = orig_off;
3048 			align_alen -= align_alen % mp->m_sb.sb_rextsize;
3049 		}
3050 		/*
3051 		 * Result doesn't cover the request, fail it.
3052 		 */
3053 		if (orig_off < align_off || orig_end > align_off + align_alen)
3054 			return -EINVAL;
3055 	} else {
3056 		ASSERT(orig_off >= align_off);
3057 		/* see MAXEXTLEN handling above */
3058 		ASSERT(orig_end <= align_off + align_alen ||
3059 		       align_alen + extsz > MAXEXTLEN);
3060 	}
3061 
3062 #ifdef DEBUG
3063 	if (!eof && gotp->br_startoff != NULLFILEOFF)
3064 		ASSERT(align_off + align_alen <= gotp->br_startoff);
3065 	if (prevp->br_startoff != NULLFILEOFF)
3066 		ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3067 #endif
3068 
3069 	*lenp = align_alen;
3070 	*offp = align_off;
3071 	return 0;
3072 }
3073 
3074 #define XFS_ALLOC_GAP_UNITS	4
3075 
3076 void
3077 xfs_bmap_adjacent(
3078 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3079 {
3080 	xfs_fsblock_t	adjust;		/* adjustment to block numbers */
3081 	xfs_agnumber_t	fb_agno;	/* ag number of ap->firstblock */
3082 	xfs_mount_t	*mp;		/* mount point structure */
3083 	int		nullfb;		/* true if ap->firstblock isn't set */
3084 	int		rt;		/* true if inode is realtime */
3085 
3086 #define	ISVALID(x,y)	\
3087 	(rt ? \
3088 		(x) < mp->m_sb.sb_rblocks : \
3089 		XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3090 		XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3091 		XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3092 
3093 	mp = ap->ip->i_mount;
3094 	nullfb = ap->tp->t_firstblock == NULLFSBLOCK;
3095 	rt = XFS_IS_REALTIME_INODE(ap->ip) &&
3096 		(ap->datatype & XFS_ALLOC_USERDATA);
3097 	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp,
3098 							ap->tp->t_firstblock);
3099 	/*
3100 	 * If allocating at eof, and there's a previous real block,
3101 	 * try to use its last block as our starting point.
3102 	 */
3103 	if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3104 	    !isnullstartblock(ap->prev.br_startblock) &&
3105 	    ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3106 		    ap->prev.br_startblock)) {
3107 		ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3108 		/*
3109 		 * Adjust for the gap between prevp and us.
3110 		 */
3111 		adjust = ap->offset -
3112 			(ap->prev.br_startoff + ap->prev.br_blockcount);
3113 		if (adjust &&
3114 		    ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3115 			ap->blkno += adjust;
3116 	}
3117 	/*
3118 	 * If not at eof, then compare the two neighbor blocks.
3119 	 * Figure out whether either one gives us a good starting point,
3120 	 * and pick the better one.
3121 	 */
3122 	else if (!ap->eof) {
3123 		xfs_fsblock_t	gotbno;		/* right side block number */
3124 		xfs_fsblock_t	gotdiff=0;	/* right side difference */
3125 		xfs_fsblock_t	prevbno;	/* left side block number */
3126 		xfs_fsblock_t	prevdiff=0;	/* left side difference */
3127 
3128 		/*
3129 		 * If there's a previous (left) block, select a requested
3130 		 * start block based on it.
3131 		 */
3132 		if (ap->prev.br_startoff != NULLFILEOFF &&
3133 		    !isnullstartblock(ap->prev.br_startblock) &&
3134 		    (prevbno = ap->prev.br_startblock +
3135 			       ap->prev.br_blockcount) &&
3136 		    ISVALID(prevbno, ap->prev.br_startblock)) {
3137 			/*
3138 			 * Calculate gap to end of previous block.
3139 			 */
3140 			adjust = prevdiff = ap->offset -
3141 				(ap->prev.br_startoff +
3142 				 ap->prev.br_blockcount);
3143 			/*
3144 			 * Figure the startblock based on the previous block's
3145 			 * end and the gap size.
3146 			 * Heuristic!
3147 			 * If the gap is large relative to the piece we're
3148 			 * allocating, or using it gives us an invalid block
3149 			 * number, then just use the end of the previous block.
3150 			 */
3151 			if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3152 			    ISVALID(prevbno + prevdiff,
3153 				    ap->prev.br_startblock))
3154 				prevbno += adjust;
3155 			else
3156 				prevdiff += adjust;
3157 			/*
3158 			 * If the firstblock forbids it, can't use it,
3159 			 * must use default.
3160 			 */
3161 			if (!rt && !nullfb &&
3162 			    XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3163 				prevbno = NULLFSBLOCK;
3164 		}
3165 		/*
3166 		 * No previous block or can't follow it, just default.
3167 		 */
3168 		else
3169 			prevbno = NULLFSBLOCK;
3170 		/*
3171 		 * If there's a following (right) block, select a requested
3172 		 * start block based on it.
3173 		 */
3174 		if (!isnullstartblock(ap->got.br_startblock)) {
3175 			/*
3176 			 * Calculate gap to start of next block.
3177 			 */
3178 			adjust = gotdiff = ap->got.br_startoff - ap->offset;
3179 			/*
3180 			 * Figure the startblock based on the next block's
3181 			 * start and the gap size.
3182 			 */
3183 			gotbno = ap->got.br_startblock;
3184 			/*
3185 			 * Heuristic!
3186 			 * If the gap is large relative to the piece we're
3187 			 * allocating, or using it gives us an invalid block
3188 			 * number, then just use the start of the next block
3189 			 * offset by our length.
3190 			 */
3191 			if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3192 			    ISVALID(gotbno - gotdiff, gotbno))
3193 				gotbno -= adjust;
3194 			else if (ISVALID(gotbno - ap->length, gotbno)) {
3195 				gotbno -= ap->length;
3196 				gotdiff += adjust - ap->length;
3197 			} else
3198 				gotdiff += adjust;
3199 			/*
3200 			 * If the firstblock forbids it, can't use it,
3201 			 * must use default.
3202 			 */
3203 			if (!rt && !nullfb &&
3204 			    XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3205 				gotbno = NULLFSBLOCK;
3206 		}
3207 		/*
3208 		 * No next block, just default.
3209 		 */
3210 		else
3211 			gotbno = NULLFSBLOCK;
3212 		/*
3213 		 * If both valid, pick the better one, else the only good
3214 		 * one, else ap->blkno is already set (to 0 or the inode block).
3215 		 */
3216 		if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3217 			ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3218 		else if (prevbno != NULLFSBLOCK)
3219 			ap->blkno = prevbno;
3220 		else if (gotbno != NULLFSBLOCK)
3221 			ap->blkno = gotbno;
3222 	}
3223 #undef ISVALID
3224 }
3225 
3226 static int
3227 xfs_bmap_longest_free_extent(
3228 	struct xfs_trans	*tp,
3229 	xfs_agnumber_t		ag,
3230 	xfs_extlen_t		*blen,
3231 	int			*notinit)
3232 {
3233 	struct xfs_mount	*mp = tp->t_mountp;
3234 	struct xfs_perag	*pag;
3235 	xfs_extlen_t		longest;
3236 	int			error = 0;
3237 
3238 	pag = xfs_perag_get(mp, ag);
3239 	if (!pag->pagf_init) {
3240 		error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3241 		if (error) {
3242 			/* Couldn't lock the AGF, so skip this AG. */
3243 			if (error == -EAGAIN) {
3244 				*notinit = 1;
3245 				error = 0;
3246 			}
3247 			goto out;
3248 		}
3249 	}
3250 
3251 	longest = xfs_alloc_longest_free_extent(pag,
3252 				xfs_alloc_min_freelist(mp, pag),
3253 				xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
3254 	if (*blen < longest)
3255 		*blen = longest;
3256 
3257 out:
3258 	xfs_perag_put(pag);
3259 	return error;
3260 }
3261 
3262 static void
3263 xfs_bmap_select_minlen(
3264 	struct xfs_bmalloca	*ap,
3265 	struct xfs_alloc_arg	*args,
3266 	xfs_extlen_t		*blen,
3267 	int			notinit)
3268 {
3269 	if (notinit || *blen < ap->minlen) {
3270 		/*
3271 		 * Since we did a BUF_TRYLOCK above, it is possible that
3272 		 * there is space for this request.
3273 		 */
3274 		args->minlen = ap->minlen;
3275 	} else if (*blen < args->maxlen) {
3276 		/*
3277 		 * If the best seen length is less than the request length,
3278 		 * use the best as the minimum.
3279 		 */
3280 		args->minlen = *blen;
3281 	} else {
3282 		/*
3283 		 * Otherwise we've seen an extent as big as maxlen, use that
3284 		 * as the minimum.
3285 		 */
3286 		args->minlen = args->maxlen;
3287 	}
3288 }
3289 
3290 STATIC int
3291 xfs_bmap_btalloc_nullfb(
3292 	struct xfs_bmalloca	*ap,
3293 	struct xfs_alloc_arg	*args,
3294 	xfs_extlen_t		*blen)
3295 {
3296 	struct xfs_mount	*mp = ap->ip->i_mount;
3297 	xfs_agnumber_t		ag, startag;
3298 	int			notinit = 0;
3299 	int			error;
3300 
3301 	args->type = XFS_ALLOCTYPE_START_BNO;
3302 	args->total = ap->total;
3303 
3304 	startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3305 	if (startag == NULLAGNUMBER)
3306 		startag = ag = 0;
3307 
3308 	while (*blen < args->maxlen) {
3309 		error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3310 						     &notinit);
3311 		if (error)
3312 			return error;
3313 
3314 		if (++ag == mp->m_sb.sb_agcount)
3315 			ag = 0;
3316 		if (ag == startag)
3317 			break;
3318 	}
3319 
3320 	xfs_bmap_select_minlen(ap, args, blen, notinit);
3321 	return 0;
3322 }
3323 
3324 STATIC int
3325 xfs_bmap_btalloc_filestreams(
3326 	struct xfs_bmalloca	*ap,
3327 	struct xfs_alloc_arg	*args,
3328 	xfs_extlen_t		*blen)
3329 {
3330 	struct xfs_mount	*mp = ap->ip->i_mount;
3331 	xfs_agnumber_t		ag;
3332 	int			notinit = 0;
3333 	int			error;
3334 
3335 	args->type = XFS_ALLOCTYPE_NEAR_BNO;
3336 	args->total = ap->total;
3337 
3338 	ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3339 	if (ag == NULLAGNUMBER)
3340 		ag = 0;
3341 
3342 	error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3343 	if (error)
3344 		return error;
3345 
3346 	if (*blen < args->maxlen) {
3347 		error = xfs_filestream_new_ag(ap, &ag);
3348 		if (error)
3349 			return error;
3350 
3351 		error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3352 						     &notinit);
3353 		if (error)
3354 			return error;
3355 
3356 	}
3357 
3358 	xfs_bmap_select_minlen(ap, args, blen, notinit);
3359 
3360 	/*
3361 	 * Set the failure fallback case to look in the selected AG as stream
3362 	 * may have moved.
3363 	 */
3364 	ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3365 	return 0;
3366 }
3367 
3368 /* Update all inode and quota accounting for the allocation we just did. */
3369 static void
3370 xfs_bmap_btalloc_accounting(
3371 	struct xfs_bmalloca	*ap,
3372 	struct xfs_alloc_arg	*args)
3373 {
3374 	if (ap->flags & XFS_BMAPI_COWFORK) {
3375 		/*
3376 		 * COW fork blocks are in-core only and thus are treated as
3377 		 * in-core quota reservation (like delalloc blocks) even when
3378 		 * converted to real blocks. The quota reservation is not
3379 		 * accounted to disk until blocks are remapped to the data
3380 		 * fork. So if these blocks were previously delalloc, we
3381 		 * already have quota reservation and there's nothing to do
3382 		 * yet.
3383 		 */
3384 		if (ap->wasdel) {
3385 			xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len);
3386 			return;
3387 		}
3388 
3389 		/*
3390 		 * Otherwise, we've allocated blocks in a hole. The transaction
3391 		 * has acquired in-core quota reservation for this extent.
3392 		 * Rather than account these as real blocks, however, we reduce
3393 		 * the transaction quota reservation based on the allocation.
3394 		 * This essentially transfers the transaction quota reservation
3395 		 * to that of a delalloc extent.
3396 		 */
3397 		ap->ip->i_delayed_blks += args->len;
3398 		xfs_trans_mod_dquot_byino(ap->tp, ap->ip, XFS_TRANS_DQ_RES_BLKS,
3399 				-(long)args->len);
3400 		return;
3401 	}
3402 
3403 	/* data/attr fork only */
3404 	ap->ip->i_nblocks += args->len;
3405 	xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3406 	if (ap->wasdel) {
3407 		ap->ip->i_delayed_blks -= args->len;
3408 		xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len);
3409 	}
3410 	xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3411 		ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT : XFS_TRANS_DQ_BCOUNT,
3412 		args->len);
3413 }
3414 
3415 static int
3416 xfs_bmap_compute_alignments(
3417 	struct xfs_bmalloca	*ap,
3418 	struct xfs_alloc_arg	*args)
3419 {
3420 	struct xfs_mount	*mp = args->mp;
3421 	xfs_extlen_t		align = 0; /* minimum allocation alignment */
3422 	int			stripe_align = 0;
3423 
3424 	/* stripe alignment for allocation is determined by mount parameters */
3425 	if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3426 		stripe_align = mp->m_swidth;
3427 	else if (mp->m_dalign)
3428 		stripe_align = mp->m_dalign;
3429 
3430 	if (ap->flags & XFS_BMAPI_COWFORK)
3431 		align = xfs_get_cowextsz_hint(ap->ip);
3432 	else if (ap->datatype & XFS_ALLOC_USERDATA)
3433 		align = xfs_get_extsz_hint(ap->ip);
3434 	if (align) {
3435 		if (xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, align, 0,
3436 					ap->eof, 0, ap->conv, &ap->offset,
3437 					&ap->length))
3438 			ASSERT(0);
3439 		ASSERT(ap->length);
3440 	}
3441 
3442 	/* apply extent size hints if obtained earlier */
3443 	if (align) {
3444 		args->prod = align;
3445 		div_u64_rem(ap->offset, args->prod, &args->mod);
3446 		if (args->mod)
3447 			args->mod = args->prod - args->mod;
3448 	} else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3449 		args->prod = 1;
3450 		args->mod = 0;
3451 	} else {
3452 		args->prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3453 		div_u64_rem(ap->offset, args->prod, &args->mod);
3454 		if (args->mod)
3455 			args->mod = args->prod - args->mod;
3456 	}
3457 
3458 	return stripe_align;
3459 }
3460 
3461 static void
3462 xfs_bmap_process_allocated_extent(
3463 	struct xfs_bmalloca	*ap,
3464 	struct xfs_alloc_arg	*args,
3465 	xfs_fileoff_t		orig_offset,
3466 	xfs_extlen_t		orig_length)
3467 {
3468 	int			nullfb;
3469 
3470 	nullfb = ap->tp->t_firstblock == NULLFSBLOCK;
3471 
3472 	/*
3473 	 * check the allocation happened at the same or higher AG than
3474 	 * the first block that was allocated.
3475 	 */
3476 	ASSERT(nullfb ||
3477 		XFS_FSB_TO_AGNO(args->mp, ap->tp->t_firstblock) <=
3478 		XFS_FSB_TO_AGNO(args->mp, args->fsbno));
3479 
3480 	ap->blkno = args->fsbno;
3481 	if (nullfb)
3482 		ap->tp->t_firstblock = args->fsbno;
3483 	ap->length = args->len;
3484 	/*
3485 	 * If the extent size hint is active, we tried to round the
3486 	 * caller's allocation request offset down to extsz and the
3487 	 * length up to another extsz boundary.  If we found a free
3488 	 * extent we mapped it in starting at this new offset.  If the
3489 	 * newly mapped space isn't long enough to cover any of the
3490 	 * range of offsets that was originally requested, move the
3491 	 * mapping up so that we can fill as much of the caller's
3492 	 * original request as possible.  Free space is apparently
3493 	 * very fragmented so we're unlikely to be able to satisfy the
3494 	 * hints anyway.
3495 	 */
3496 	if (ap->length <= orig_length)
3497 		ap->offset = orig_offset;
3498 	else if (ap->offset + ap->length < orig_offset + orig_length)
3499 		ap->offset = orig_offset + orig_length - ap->length;
3500 	xfs_bmap_btalloc_accounting(ap, args);
3501 }
3502 
3503 #ifdef DEBUG
3504 static int
3505 xfs_bmap_exact_minlen_extent_alloc(
3506 	struct xfs_bmalloca	*ap)
3507 {
3508 	struct xfs_mount	*mp = ap->ip->i_mount;
3509 	struct xfs_alloc_arg	args = { .tp = ap->tp, .mp = mp };
3510 	xfs_fileoff_t		orig_offset;
3511 	xfs_extlen_t		orig_length;
3512 	int			error;
3513 
3514 	ASSERT(ap->length);
3515 
3516 	if (ap->minlen != 1) {
3517 		ap->blkno = NULLFSBLOCK;
3518 		ap->length = 0;
3519 		return 0;
3520 	}
3521 
3522 	orig_offset = ap->offset;
3523 	orig_length = ap->length;
3524 
3525 	args.alloc_minlen_only = 1;
3526 
3527 	xfs_bmap_compute_alignments(ap, &args);
3528 
3529 	if (ap->tp->t_firstblock == NULLFSBLOCK) {
3530 		/*
3531 		 * Unlike the longest extent available in an AG, we don't track
3532 		 * the length of an AG's shortest extent.
3533 		 * XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT is a debug only knob and
3534 		 * hence we can afford to start traversing from the 0th AG since
3535 		 * we need not be concerned about a drop in performance in
3536 		 * "debug only" code paths.
3537 		 */
3538 		ap->blkno = XFS_AGB_TO_FSB(mp, 0, 0);
3539 	} else {
3540 		ap->blkno = ap->tp->t_firstblock;
3541 	}
3542 
3543 	args.fsbno = ap->blkno;
3544 	args.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
3545 	args.type = XFS_ALLOCTYPE_FIRST_AG;
3546 	args.minlen = args.maxlen = ap->minlen;
3547 	args.total = ap->total;
3548 
3549 	args.alignment = 1;
3550 	args.minalignslop = 0;
3551 
3552 	args.minleft = ap->minleft;
3553 	args.wasdel = ap->wasdel;
3554 	args.resv = XFS_AG_RESV_NONE;
3555 	args.datatype = ap->datatype;
3556 
3557 	error = xfs_alloc_vextent(&args);
3558 	if (error)
3559 		return error;
3560 
3561 	if (args.fsbno != NULLFSBLOCK) {
3562 		xfs_bmap_process_allocated_extent(ap, &args, orig_offset,
3563 			orig_length);
3564 	} else {
3565 		ap->blkno = NULLFSBLOCK;
3566 		ap->length = 0;
3567 	}
3568 
3569 	return 0;
3570 }
3571 #else
3572 
3573 #define xfs_bmap_exact_minlen_extent_alloc(bma) (-EFSCORRUPTED)
3574 
3575 #endif
3576 
3577 STATIC int
3578 xfs_bmap_btalloc(
3579 	struct xfs_bmalloca	*ap)
3580 {
3581 	struct xfs_mount	*mp = ap->ip->i_mount;
3582 	struct xfs_alloc_arg	args = { .tp = ap->tp, .mp = mp };
3583 	xfs_alloctype_t		atype = 0;
3584 	xfs_agnumber_t		fb_agno;	/* ag number of ap->firstblock */
3585 	xfs_agnumber_t		ag;
3586 	xfs_fileoff_t		orig_offset;
3587 	xfs_extlen_t		orig_length;
3588 	xfs_extlen_t		blen;
3589 	xfs_extlen_t		nextminlen = 0;
3590 	int			nullfb; /* true if ap->firstblock isn't set */
3591 	int			isaligned;
3592 	int			tryagain;
3593 	int			error;
3594 	int			stripe_align;
3595 
3596 	ASSERT(ap->length);
3597 	orig_offset = ap->offset;
3598 	orig_length = ap->length;
3599 
3600 	stripe_align = xfs_bmap_compute_alignments(ap, &args);
3601 
3602 	nullfb = ap->tp->t_firstblock == NULLFSBLOCK;
3603 	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp,
3604 							ap->tp->t_firstblock);
3605 	if (nullfb) {
3606 		if ((ap->datatype & XFS_ALLOC_USERDATA) &&
3607 		    xfs_inode_is_filestream(ap->ip)) {
3608 			ag = xfs_filestream_lookup_ag(ap->ip);
3609 			ag = (ag != NULLAGNUMBER) ? ag : 0;
3610 			ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3611 		} else {
3612 			ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3613 		}
3614 	} else
3615 		ap->blkno = ap->tp->t_firstblock;
3616 
3617 	xfs_bmap_adjacent(ap);
3618 
3619 	/*
3620 	 * If allowed, use ap->blkno; otherwise must use firstblock since
3621 	 * it's in the right allocation group.
3622 	 */
3623 	if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3624 		;
3625 	else
3626 		ap->blkno = ap->tp->t_firstblock;
3627 	/*
3628 	 * Normal allocation, done through xfs_alloc_vextent.
3629 	 */
3630 	tryagain = isaligned = 0;
3631 	args.fsbno = ap->blkno;
3632 	args.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
3633 
3634 	/* Trim the allocation back to the maximum an AG can fit. */
3635 	args.maxlen = min(ap->length, mp->m_ag_max_usable);
3636 	blen = 0;
3637 	if (nullfb) {
3638 		/*
3639 		 * Search for an allocation group with a single extent large
3640 		 * enough for the request.  If one isn't found, then adjust
3641 		 * the minimum allocation size to the largest space found.
3642 		 */
3643 		if ((ap->datatype & XFS_ALLOC_USERDATA) &&
3644 		    xfs_inode_is_filestream(ap->ip))
3645 			error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3646 		else
3647 			error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3648 		if (error)
3649 			return error;
3650 	} else if (ap->tp->t_flags & XFS_TRANS_LOWMODE) {
3651 		if (xfs_inode_is_filestream(ap->ip))
3652 			args.type = XFS_ALLOCTYPE_FIRST_AG;
3653 		else
3654 			args.type = XFS_ALLOCTYPE_START_BNO;
3655 		args.total = args.minlen = ap->minlen;
3656 	} else {
3657 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
3658 		args.total = ap->total;
3659 		args.minlen = ap->minlen;
3660 	}
3661 
3662 	/*
3663 	 * If we are not low on available data blocks, and the underlying
3664 	 * logical volume manager is a stripe, and the file offset is zero then
3665 	 * try to allocate data blocks on stripe unit boundary. NOTE: ap->aeof
3666 	 * is only set if the allocation length is >= the stripe unit and the
3667 	 * allocation offset is at the end of file.
3668 	 */
3669 	if (!(ap->tp->t_flags & XFS_TRANS_LOWMODE) && ap->aeof) {
3670 		if (!ap->offset) {
3671 			args.alignment = stripe_align;
3672 			atype = args.type;
3673 			isaligned = 1;
3674 			/*
3675 			 * Adjust minlen to try and preserve alignment if we
3676 			 * can't guarantee an aligned maxlen extent.
3677 			 */
3678 			if (blen > args.alignment &&
3679 			    blen <= args.maxlen + args.alignment)
3680 				args.minlen = blen - args.alignment;
3681 			args.minalignslop = 0;
3682 		} else {
3683 			/*
3684 			 * First try an exact bno allocation.
3685 			 * If it fails then do a near or start bno
3686 			 * allocation with alignment turned on.
3687 			 */
3688 			atype = args.type;
3689 			tryagain = 1;
3690 			args.type = XFS_ALLOCTYPE_THIS_BNO;
3691 			args.alignment = 1;
3692 			/*
3693 			 * Compute the minlen+alignment for the
3694 			 * next case.  Set slop so that the value
3695 			 * of minlen+alignment+slop doesn't go up
3696 			 * between the calls.
3697 			 */
3698 			if (blen > stripe_align && blen <= args.maxlen)
3699 				nextminlen = blen - stripe_align;
3700 			else
3701 				nextminlen = args.minlen;
3702 			if (nextminlen + stripe_align > args.minlen + 1)
3703 				args.minalignslop =
3704 					nextminlen + stripe_align -
3705 					args.minlen - 1;
3706 			else
3707 				args.minalignslop = 0;
3708 		}
3709 	} else {
3710 		args.alignment = 1;
3711 		args.minalignslop = 0;
3712 	}
3713 	args.minleft = ap->minleft;
3714 	args.wasdel = ap->wasdel;
3715 	args.resv = XFS_AG_RESV_NONE;
3716 	args.datatype = ap->datatype;
3717 
3718 	error = xfs_alloc_vextent(&args);
3719 	if (error)
3720 		return error;
3721 
3722 	if (tryagain && args.fsbno == NULLFSBLOCK) {
3723 		/*
3724 		 * Exact allocation failed. Now try with alignment
3725 		 * turned on.
3726 		 */
3727 		args.type = atype;
3728 		args.fsbno = ap->blkno;
3729 		args.alignment = stripe_align;
3730 		args.minlen = nextminlen;
3731 		args.minalignslop = 0;
3732 		isaligned = 1;
3733 		if ((error = xfs_alloc_vextent(&args)))
3734 			return error;
3735 	}
3736 	if (isaligned && args.fsbno == NULLFSBLOCK) {
3737 		/*
3738 		 * allocation failed, so turn off alignment and
3739 		 * try again.
3740 		 */
3741 		args.type = atype;
3742 		args.fsbno = ap->blkno;
3743 		args.alignment = 0;
3744 		if ((error = xfs_alloc_vextent(&args)))
3745 			return error;
3746 	}
3747 	if (args.fsbno == NULLFSBLOCK && nullfb &&
3748 	    args.minlen > ap->minlen) {
3749 		args.minlen = ap->minlen;
3750 		args.type = XFS_ALLOCTYPE_START_BNO;
3751 		args.fsbno = ap->blkno;
3752 		if ((error = xfs_alloc_vextent(&args)))
3753 			return error;
3754 	}
3755 	if (args.fsbno == NULLFSBLOCK && nullfb) {
3756 		args.fsbno = 0;
3757 		args.type = XFS_ALLOCTYPE_FIRST_AG;
3758 		args.total = ap->minlen;
3759 		if ((error = xfs_alloc_vextent(&args)))
3760 			return error;
3761 		ap->tp->t_flags |= XFS_TRANS_LOWMODE;
3762 	}
3763 
3764 	if (args.fsbno != NULLFSBLOCK) {
3765 		xfs_bmap_process_allocated_extent(ap, &args, orig_offset,
3766 			orig_length);
3767 	} else {
3768 		ap->blkno = NULLFSBLOCK;
3769 		ap->length = 0;
3770 	}
3771 	return 0;
3772 }
3773 
3774 /* Trim extent to fit a logical block range. */
3775 void
3776 xfs_trim_extent(
3777 	struct xfs_bmbt_irec	*irec,
3778 	xfs_fileoff_t		bno,
3779 	xfs_filblks_t		len)
3780 {
3781 	xfs_fileoff_t		distance;
3782 	xfs_fileoff_t		end = bno + len;
3783 
3784 	if (irec->br_startoff + irec->br_blockcount <= bno ||
3785 	    irec->br_startoff >= end) {
3786 		irec->br_blockcount = 0;
3787 		return;
3788 	}
3789 
3790 	if (irec->br_startoff < bno) {
3791 		distance = bno - irec->br_startoff;
3792 		if (isnullstartblock(irec->br_startblock))
3793 			irec->br_startblock = DELAYSTARTBLOCK;
3794 		if (irec->br_startblock != DELAYSTARTBLOCK &&
3795 		    irec->br_startblock != HOLESTARTBLOCK)
3796 			irec->br_startblock += distance;
3797 		irec->br_startoff += distance;
3798 		irec->br_blockcount -= distance;
3799 	}
3800 
3801 	if (end < irec->br_startoff + irec->br_blockcount) {
3802 		distance = irec->br_startoff + irec->br_blockcount - end;
3803 		irec->br_blockcount -= distance;
3804 	}
3805 }
3806 
3807 /*
3808  * Trim the returned map to the required bounds
3809  */
3810 STATIC void
3811 xfs_bmapi_trim_map(
3812 	struct xfs_bmbt_irec	*mval,
3813 	struct xfs_bmbt_irec	*got,
3814 	xfs_fileoff_t		*bno,
3815 	xfs_filblks_t		len,
3816 	xfs_fileoff_t		obno,
3817 	xfs_fileoff_t		end,
3818 	int			n,
3819 	int			flags)
3820 {
3821 	if ((flags & XFS_BMAPI_ENTIRE) ||
3822 	    got->br_startoff + got->br_blockcount <= obno) {
3823 		*mval = *got;
3824 		if (isnullstartblock(got->br_startblock))
3825 			mval->br_startblock = DELAYSTARTBLOCK;
3826 		return;
3827 	}
3828 
3829 	if (obno > *bno)
3830 		*bno = obno;
3831 	ASSERT((*bno >= obno) || (n == 0));
3832 	ASSERT(*bno < end);
3833 	mval->br_startoff = *bno;
3834 	if (isnullstartblock(got->br_startblock))
3835 		mval->br_startblock = DELAYSTARTBLOCK;
3836 	else
3837 		mval->br_startblock = got->br_startblock +
3838 					(*bno - got->br_startoff);
3839 	/*
3840 	 * Return the minimum of what we got and what we asked for for
3841 	 * the length.  We can use the len variable here because it is
3842 	 * modified below and we could have been there before coming
3843 	 * here if the first part of the allocation didn't overlap what
3844 	 * was asked for.
3845 	 */
3846 	mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3847 			got->br_blockcount - (*bno - got->br_startoff));
3848 	mval->br_state = got->br_state;
3849 	ASSERT(mval->br_blockcount <= len);
3850 	return;
3851 }
3852 
3853 /*
3854  * Update and validate the extent map to return
3855  */
3856 STATIC void
3857 xfs_bmapi_update_map(
3858 	struct xfs_bmbt_irec	**map,
3859 	xfs_fileoff_t		*bno,
3860 	xfs_filblks_t		*len,
3861 	xfs_fileoff_t		obno,
3862 	xfs_fileoff_t		end,
3863 	int			*n,
3864 	int			flags)
3865 {
3866 	xfs_bmbt_irec_t	*mval = *map;
3867 
3868 	ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3869 	       ((mval->br_startoff + mval->br_blockcount) <= end));
3870 	ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3871 	       (mval->br_startoff < obno));
3872 
3873 	*bno = mval->br_startoff + mval->br_blockcount;
3874 	*len = end - *bno;
3875 	if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3876 		/* update previous map with new information */
3877 		ASSERT(mval->br_startblock == mval[-1].br_startblock);
3878 		ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3879 		ASSERT(mval->br_state == mval[-1].br_state);
3880 		mval[-1].br_blockcount = mval->br_blockcount;
3881 		mval[-1].br_state = mval->br_state;
3882 	} else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3883 		   mval[-1].br_startblock != DELAYSTARTBLOCK &&
3884 		   mval[-1].br_startblock != HOLESTARTBLOCK &&
3885 		   mval->br_startblock == mval[-1].br_startblock +
3886 					  mval[-1].br_blockcount &&
3887 		   mval[-1].br_state == mval->br_state) {
3888 		ASSERT(mval->br_startoff ==
3889 		       mval[-1].br_startoff + mval[-1].br_blockcount);
3890 		mval[-1].br_blockcount += mval->br_blockcount;
3891 	} else if (*n > 0 &&
3892 		   mval->br_startblock == DELAYSTARTBLOCK &&
3893 		   mval[-1].br_startblock == DELAYSTARTBLOCK &&
3894 		   mval->br_startoff ==
3895 		   mval[-1].br_startoff + mval[-1].br_blockcount) {
3896 		mval[-1].br_blockcount += mval->br_blockcount;
3897 		mval[-1].br_state = mval->br_state;
3898 	} else if (!((*n == 0) &&
3899 		     ((mval->br_startoff + mval->br_blockcount) <=
3900 		      obno))) {
3901 		mval++;
3902 		(*n)++;
3903 	}
3904 	*map = mval;
3905 }
3906 
3907 /*
3908  * Map file blocks to filesystem blocks without allocation.
3909  */
3910 int
3911 xfs_bmapi_read(
3912 	struct xfs_inode	*ip,
3913 	xfs_fileoff_t		bno,
3914 	xfs_filblks_t		len,
3915 	struct xfs_bmbt_irec	*mval,
3916 	int			*nmap,
3917 	int			flags)
3918 {
3919 	struct xfs_mount	*mp = ip->i_mount;
3920 	int			whichfork = xfs_bmapi_whichfork(flags);
3921 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
3922 	struct xfs_bmbt_irec	got;
3923 	xfs_fileoff_t		obno;
3924 	xfs_fileoff_t		end;
3925 	struct xfs_iext_cursor	icur;
3926 	int			error;
3927 	bool			eof = false;
3928 	int			n = 0;
3929 
3930 	ASSERT(*nmap >= 1);
3931 	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_ENTIRE)));
3932 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
3933 
3934 	if (WARN_ON_ONCE(!ifp))
3935 		return -EFSCORRUPTED;
3936 
3937 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
3938 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT))
3939 		return -EFSCORRUPTED;
3940 
3941 	if (XFS_FORCED_SHUTDOWN(mp))
3942 		return -EIO;
3943 
3944 	XFS_STATS_INC(mp, xs_blk_mapr);
3945 
3946 	error = xfs_iread_extents(NULL, ip, whichfork);
3947 	if (error)
3948 		return error;
3949 
3950 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got))
3951 		eof = true;
3952 	end = bno + len;
3953 	obno = bno;
3954 
3955 	while (bno < end && n < *nmap) {
3956 		/* Reading past eof, act as though there's a hole up to end. */
3957 		if (eof)
3958 			got.br_startoff = end;
3959 		if (got.br_startoff > bno) {
3960 			/* Reading in a hole.  */
3961 			mval->br_startoff = bno;
3962 			mval->br_startblock = HOLESTARTBLOCK;
3963 			mval->br_blockcount =
3964 				XFS_FILBLKS_MIN(len, got.br_startoff - bno);
3965 			mval->br_state = XFS_EXT_NORM;
3966 			bno += mval->br_blockcount;
3967 			len -= mval->br_blockcount;
3968 			mval++;
3969 			n++;
3970 			continue;
3971 		}
3972 
3973 		/* set up the extent map to return. */
3974 		xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
3975 		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
3976 
3977 		/* If we're done, stop now. */
3978 		if (bno >= end || n >= *nmap)
3979 			break;
3980 
3981 		/* Else go on to the next record. */
3982 		if (!xfs_iext_next_extent(ifp, &icur, &got))
3983 			eof = true;
3984 	}
3985 	*nmap = n;
3986 	return 0;
3987 }
3988 
3989 /*
3990  * Add a delayed allocation extent to an inode. Blocks are reserved from the
3991  * global pool and the extent inserted into the inode in-core extent tree.
3992  *
3993  * On entry, got refers to the first extent beyond the offset of the extent to
3994  * allocate or eof is specified if no such extent exists. On return, got refers
3995  * to the extent record that was inserted to the inode fork.
3996  *
3997  * Note that the allocated extent may have been merged with contiguous extents
3998  * during insertion into the inode fork. Thus, got does not reflect the current
3999  * state of the inode fork on return. If necessary, the caller can use lastx to
4000  * look up the updated record in the inode fork.
4001  */
4002 int
4003 xfs_bmapi_reserve_delalloc(
4004 	struct xfs_inode	*ip,
4005 	int			whichfork,
4006 	xfs_fileoff_t		off,
4007 	xfs_filblks_t		len,
4008 	xfs_filblks_t		prealloc,
4009 	struct xfs_bmbt_irec	*got,
4010 	struct xfs_iext_cursor	*icur,
4011 	int			eof)
4012 {
4013 	struct xfs_mount	*mp = ip->i_mount;
4014 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
4015 	xfs_extlen_t		alen;
4016 	xfs_extlen_t		indlen;
4017 	int			error;
4018 	xfs_fileoff_t		aoff = off;
4019 
4020 	/*
4021 	 * Cap the alloc length. Keep track of prealloc so we know whether to
4022 	 * tag the inode before we return.
4023 	 */
4024 	alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN);
4025 	if (!eof)
4026 		alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4027 	if (prealloc && alen >= len)
4028 		prealloc = alen - len;
4029 
4030 	/* Figure out the extent size, adjust alen */
4031 	if (whichfork == XFS_COW_FORK) {
4032 		struct xfs_bmbt_irec	prev;
4033 		xfs_extlen_t		extsz = xfs_get_cowextsz_hint(ip);
4034 
4035 		if (!xfs_iext_peek_prev_extent(ifp, icur, &prev))
4036 			prev.br_startoff = NULLFILEOFF;
4037 
4038 		error = xfs_bmap_extsize_align(mp, got, &prev, extsz, 0, eof,
4039 					       1, 0, &aoff, &alen);
4040 		ASSERT(!error);
4041 	}
4042 
4043 	/*
4044 	 * Make a transaction-less quota reservation for delayed allocation
4045 	 * blocks.  This number gets adjusted later.  We return if we haven't
4046 	 * allocated blocks already inside this loop.
4047 	 */
4048 	error = xfs_quota_reserve_blkres(ip, alen);
4049 	if (error)
4050 		return error;
4051 
4052 	/*
4053 	 * Split changing sb for alen and indlen since they could be coming
4054 	 * from different places.
4055 	 */
4056 	indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4057 	ASSERT(indlen > 0);
4058 
4059 	error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
4060 	if (error)
4061 		goto out_unreserve_quota;
4062 
4063 	error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
4064 	if (error)
4065 		goto out_unreserve_blocks;
4066 
4067 
4068 	ip->i_delayed_blks += alen;
4069 	xfs_mod_delalloc(ip->i_mount, alen + indlen);
4070 
4071 	got->br_startoff = aoff;
4072 	got->br_startblock = nullstartblock(indlen);
4073 	got->br_blockcount = alen;
4074 	got->br_state = XFS_EXT_NORM;
4075 
4076 	xfs_bmap_add_extent_hole_delay(ip, whichfork, icur, got);
4077 
4078 	/*
4079 	 * Tag the inode if blocks were preallocated. Note that COW fork
4080 	 * preallocation can occur at the start or end of the extent, even when
4081 	 * prealloc == 0, so we must also check the aligned offset and length.
4082 	 */
4083 	if (whichfork == XFS_DATA_FORK && prealloc)
4084 		xfs_inode_set_eofblocks_tag(ip);
4085 	if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len))
4086 		xfs_inode_set_cowblocks_tag(ip);
4087 
4088 	return 0;
4089 
4090 out_unreserve_blocks:
4091 	xfs_mod_fdblocks(mp, alen, false);
4092 out_unreserve_quota:
4093 	if (XFS_IS_QUOTA_ON(mp))
4094 		xfs_quota_unreserve_blkres(ip, alen);
4095 	return error;
4096 }
4097 
4098 static int
4099 xfs_bmap_alloc_userdata(
4100 	struct xfs_bmalloca	*bma)
4101 {
4102 	struct xfs_mount	*mp = bma->ip->i_mount;
4103 	int			whichfork = xfs_bmapi_whichfork(bma->flags);
4104 	int			error;
4105 
4106 	/*
4107 	 * Set the data type being allocated. For the data fork, the first data
4108 	 * in the file is treated differently to all other allocations. For the
4109 	 * attribute fork, we only need to ensure the allocated range is not on
4110 	 * the busy list.
4111 	 */
4112 	bma->datatype = XFS_ALLOC_NOBUSY;
4113 	if (whichfork == XFS_DATA_FORK) {
4114 		bma->datatype |= XFS_ALLOC_USERDATA;
4115 		if (bma->offset == 0)
4116 			bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA;
4117 
4118 		if (mp->m_dalign && bma->length >= mp->m_dalign) {
4119 			error = xfs_bmap_isaeof(bma, whichfork);
4120 			if (error)
4121 				return error;
4122 		}
4123 
4124 		if (XFS_IS_REALTIME_INODE(bma->ip))
4125 			return xfs_bmap_rtalloc(bma);
4126 	}
4127 
4128 	if (unlikely(XFS_TEST_ERROR(false, mp,
4129 			XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT)))
4130 		return xfs_bmap_exact_minlen_extent_alloc(bma);
4131 
4132 	return xfs_bmap_btalloc(bma);
4133 }
4134 
4135 static int
4136 xfs_bmapi_allocate(
4137 	struct xfs_bmalloca	*bma)
4138 {
4139 	struct xfs_mount	*mp = bma->ip->i_mount;
4140 	int			whichfork = xfs_bmapi_whichfork(bma->flags);
4141 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4142 	int			tmp_logflags = 0;
4143 	int			error;
4144 
4145 	ASSERT(bma->length > 0);
4146 
4147 	/*
4148 	 * For the wasdelay case, we could also just allocate the stuff asked
4149 	 * for in this bmap call but that wouldn't be as good.
4150 	 */
4151 	if (bma->wasdel) {
4152 		bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4153 		bma->offset = bma->got.br_startoff;
4154 		if (!xfs_iext_peek_prev_extent(ifp, &bma->icur, &bma->prev))
4155 			bma->prev.br_startoff = NULLFILEOFF;
4156 	} else {
4157 		bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4158 		if (!bma->eof)
4159 			bma->length = XFS_FILBLKS_MIN(bma->length,
4160 					bma->got.br_startoff - bma->offset);
4161 	}
4162 
4163 	if (bma->flags & XFS_BMAPI_CONTIG)
4164 		bma->minlen = bma->length;
4165 	else
4166 		bma->minlen = 1;
4167 
4168 	if (bma->flags & XFS_BMAPI_METADATA) {
4169 		if (unlikely(XFS_TEST_ERROR(false, mp,
4170 				XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT)))
4171 			error = xfs_bmap_exact_minlen_extent_alloc(bma);
4172 		else
4173 			error = xfs_bmap_btalloc(bma);
4174 	} else {
4175 		error = xfs_bmap_alloc_userdata(bma);
4176 	}
4177 	if (error || bma->blkno == NULLFSBLOCK)
4178 		return error;
4179 
4180 	if (bma->flags & XFS_BMAPI_ZERO) {
4181 		error = xfs_zero_extent(bma->ip, bma->blkno, bma->length);
4182 		if (error)
4183 			return error;
4184 	}
4185 
4186 	if (ifp->if_format == XFS_DINODE_FMT_BTREE && !bma->cur)
4187 		bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4188 	/*
4189 	 * Bump the number of extents we've allocated
4190 	 * in this call.
4191 	 */
4192 	bma->nallocs++;
4193 
4194 	if (bma->cur)
4195 		bma->cur->bc_ino.flags =
4196 			bma->wasdel ? XFS_BTCUR_BMBT_WASDEL : 0;
4197 
4198 	bma->got.br_startoff = bma->offset;
4199 	bma->got.br_startblock = bma->blkno;
4200 	bma->got.br_blockcount = bma->length;
4201 	bma->got.br_state = XFS_EXT_NORM;
4202 
4203 	if (bma->flags & XFS_BMAPI_PREALLOC)
4204 		bma->got.br_state = XFS_EXT_UNWRITTEN;
4205 
4206 	if (bma->wasdel)
4207 		error = xfs_bmap_add_extent_delay_real(bma, whichfork);
4208 	else
4209 		error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip,
4210 				whichfork, &bma->icur, &bma->cur, &bma->got,
4211 				&bma->logflags, bma->flags);
4212 
4213 	bma->logflags |= tmp_logflags;
4214 	if (error)
4215 		return error;
4216 
4217 	/*
4218 	 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4219 	 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4220 	 * the neighbouring ones.
4221 	 */
4222 	xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4223 
4224 	ASSERT(bma->got.br_startoff <= bma->offset);
4225 	ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4226 	       bma->offset + bma->length);
4227 	ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4228 	       bma->got.br_state == XFS_EXT_UNWRITTEN);
4229 	return 0;
4230 }
4231 
4232 STATIC int
4233 xfs_bmapi_convert_unwritten(
4234 	struct xfs_bmalloca	*bma,
4235 	struct xfs_bmbt_irec	*mval,
4236 	xfs_filblks_t		len,
4237 	int			flags)
4238 {
4239 	int			whichfork = xfs_bmapi_whichfork(flags);
4240 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4241 	int			tmp_logflags = 0;
4242 	int			error;
4243 
4244 	/* check if we need to do unwritten->real conversion */
4245 	if (mval->br_state == XFS_EXT_UNWRITTEN &&
4246 	    (flags & XFS_BMAPI_PREALLOC))
4247 		return 0;
4248 
4249 	/* check if we need to do real->unwritten conversion */
4250 	if (mval->br_state == XFS_EXT_NORM &&
4251 	    (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4252 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4253 		return 0;
4254 
4255 	/*
4256 	 * Modify (by adding) the state flag, if writing.
4257 	 */
4258 	ASSERT(mval->br_blockcount <= len);
4259 	if (ifp->if_format == XFS_DINODE_FMT_BTREE && !bma->cur) {
4260 		bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4261 					bma->ip, whichfork);
4262 	}
4263 	mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4264 				? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4265 
4266 	/*
4267 	 * Before insertion into the bmbt, zero the range being converted
4268 	 * if required.
4269 	 */
4270 	if (flags & XFS_BMAPI_ZERO) {
4271 		error = xfs_zero_extent(bma->ip, mval->br_startblock,
4272 					mval->br_blockcount);
4273 		if (error)
4274 			return error;
4275 	}
4276 
4277 	error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork,
4278 			&bma->icur, &bma->cur, mval, &tmp_logflags);
4279 	/*
4280 	 * Log the inode core unconditionally in the unwritten extent conversion
4281 	 * path because the conversion might not have done so (e.g., if the
4282 	 * extent count hasn't changed). We need to make sure the inode is dirty
4283 	 * in the transaction for the sake of fsync(), even if nothing has
4284 	 * changed, because fsync() will not force the log for this transaction
4285 	 * unless it sees the inode pinned.
4286 	 *
4287 	 * Note: If we're only converting cow fork extents, there aren't
4288 	 * any on-disk updates to make, so we don't need to log anything.
4289 	 */
4290 	if (whichfork != XFS_COW_FORK)
4291 		bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4292 	if (error)
4293 		return error;
4294 
4295 	/*
4296 	 * Update our extent pointer, given that
4297 	 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4298 	 * of the neighbouring ones.
4299 	 */
4300 	xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4301 
4302 	/*
4303 	 * We may have combined previously unwritten space with written space,
4304 	 * so generate another request.
4305 	 */
4306 	if (mval->br_blockcount < len)
4307 		return -EAGAIN;
4308 	return 0;
4309 }
4310 
4311 static inline xfs_extlen_t
4312 xfs_bmapi_minleft(
4313 	struct xfs_trans	*tp,
4314 	struct xfs_inode	*ip,
4315 	int			fork)
4316 {
4317 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, fork);
4318 
4319 	if (tp && tp->t_firstblock != NULLFSBLOCK)
4320 		return 0;
4321 	if (ifp->if_format != XFS_DINODE_FMT_BTREE)
4322 		return 1;
4323 	return be16_to_cpu(ifp->if_broot->bb_level) + 1;
4324 }
4325 
4326 /*
4327  * Log whatever the flags say, even if error.  Otherwise we might miss detecting
4328  * a case where the data is changed, there's an error, and it's not logged so we
4329  * don't shutdown when we should.  Don't bother logging extents/btree changes if
4330  * we converted to the other format.
4331  */
4332 static void
4333 xfs_bmapi_finish(
4334 	struct xfs_bmalloca	*bma,
4335 	int			whichfork,
4336 	int			error)
4337 {
4338 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4339 
4340 	if ((bma->logflags & xfs_ilog_fext(whichfork)) &&
4341 	    ifp->if_format != XFS_DINODE_FMT_EXTENTS)
4342 		bma->logflags &= ~xfs_ilog_fext(whichfork);
4343 	else if ((bma->logflags & xfs_ilog_fbroot(whichfork)) &&
4344 		 ifp->if_format != XFS_DINODE_FMT_BTREE)
4345 		bma->logflags &= ~xfs_ilog_fbroot(whichfork);
4346 
4347 	if (bma->logflags)
4348 		xfs_trans_log_inode(bma->tp, bma->ip, bma->logflags);
4349 	if (bma->cur)
4350 		xfs_btree_del_cursor(bma->cur, error);
4351 }
4352 
4353 /*
4354  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4355  * extent state if necessary.  Details behaviour is controlled by the flags
4356  * parameter.  Only allocates blocks from a single allocation group, to avoid
4357  * locking problems.
4358  */
4359 int
4360 xfs_bmapi_write(
4361 	struct xfs_trans	*tp,		/* transaction pointer */
4362 	struct xfs_inode	*ip,		/* incore inode */
4363 	xfs_fileoff_t		bno,		/* starting file offs. mapped */
4364 	xfs_filblks_t		len,		/* length to map in file */
4365 	int			flags,		/* XFS_BMAPI_... */
4366 	xfs_extlen_t		total,		/* total blocks needed */
4367 	struct xfs_bmbt_irec	*mval,		/* output: map values */
4368 	int			*nmap)		/* i/o: mval size/count */
4369 {
4370 	struct xfs_bmalloca	bma = {
4371 		.tp		= tp,
4372 		.ip		= ip,
4373 		.total		= total,
4374 	};
4375 	struct xfs_mount	*mp = ip->i_mount;
4376 	int			whichfork = xfs_bmapi_whichfork(flags);
4377 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
4378 	xfs_fileoff_t		end;		/* end of mapped file region */
4379 	bool			eof = false;	/* after the end of extents */
4380 	int			error;		/* error return */
4381 	int			n;		/* current extent index */
4382 	xfs_fileoff_t		obno;		/* old block number (offset) */
4383 
4384 #ifdef DEBUG
4385 	xfs_fileoff_t		orig_bno;	/* original block number value */
4386 	int			orig_flags;	/* original flags arg value */
4387 	xfs_filblks_t		orig_len;	/* original value of len arg */
4388 	struct xfs_bmbt_irec	*orig_mval;	/* original value of mval */
4389 	int			orig_nmap;	/* original value of *nmap */
4390 
4391 	orig_bno = bno;
4392 	orig_len = len;
4393 	orig_flags = flags;
4394 	orig_mval = mval;
4395 	orig_nmap = *nmap;
4396 #endif
4397 
4398 	ASSERT(*nmap >= 1);
4399 	ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4400 	ASSERT(tp != NULL);
4401 	ASSERT(len > 0);
4402 	ASSERT(ifp->if_format != XFS_DINODE_FMT_LOCAL);
4403 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4404 	ASSERT(!(flags & XFS_BMAPI_REMAP));
4405 
4406 	/* zeroing is for currently only for data extents, not metadata */
4407 	ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4408 			(XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4409 	/*
4410 	 * we can allocate unwritten extents or pre-zero allocated blocks,
4411 	 * but it makes no sense to do both at once. This would result in
4412 	 * zeroing the unwritten extent twice, but it still being an
4413 	 * unwritten extent....
4414 	 */
4415 	ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4416 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4417 
4418 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
4419 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
4420 		return -EFSCORRUPTED;
4421 	}
4422 
4423 	if (XFS_FORCED_SHUTDOWN(mp))
4424 		return -EIO;
4425 
4426 	XFS_STATS_INC(mp, xs_blk_mapw);
4427 
4428 	error = xfs_iread_extents(tp, ip, whichfork);
4429 	if (error)
4430 		goto error0;
4431 
4432 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.icur, &bma.got))
4433 		eof = true;
4434 	if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4435 		bma.prev.br_startoff = NULLFILEOFF;
4436 	bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4437 
4438 	n = 0;
4439 	end = bno + len;
4440 	obno = bno;
4441 	while (bno < end && n < *nmap) {
4442 		bool			need_alloc = false, wasdelay = false;
4443 
4444 		/* in hole or beyond EOF? */
4445 		if (eof || bma.got.br_startoff > bno) {
4446 			/*
4447 			 * CoW fork conversions should /never/ hit EOF or
4448 			 * holes.  There should always be something for us
4449 			 * to work on.
4450 			 */
4451 			ASSERT(!((flags & XFS_BMAPI_CONVERT) &&
4452 			         (flags & XFS_BMAPI_COWFORK)));
4453 
4454 			need_alloc = true;
4455 		} else if (isnullstartblock(bma.got.br_startblock)) {
4456 			wasdelay = true;
4457 		}
4458 
4459 		/*
4460 		 * First, deal with the hole before the allocated space
4461 		 * that we found, if any.
4462 		 */
4463 		if (need_alloc || wasdelay) {
4464 			bma.eof = eof;
4465 			bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4466 			bma.wasdel = wasdelay;
4467 			bma.offset = bno;
4468 			bma.flags = flags;
4469 
4470 			/*
4471 			 * There's a 32/64 bit type mismatch between the
4472 			 * allocation length request (which can be 64 bits in
4473 			 * length) and the bma length request, which is
4474 			 * xfs_extlen_t and therefore 32 bits. Hence we have to
4475 			 * check for 32-bit overflows and handle them here.
4476 			 */
4477 			if (len > (xfs_filblks_t)MAXEXTLEN)
4478 				bma.length = MAXEXTLEN;
4479 			else
4480 				bma.length = len;
4481 
4482 			ASSERT(len > 0);
4483 			ASSERT(bma.length > 0);
4484 			error = xfs_bmapi_allocate(&bma);
4485 			if (error)
4486 				goto error0;
4487 			if (bma.blkno == NULLFSBLOCK)
4488 				break;
4489 
4490 			/*
4491 			 * If this is a CoW allocation, record the data in
4492 			 * the refcount btree for orphan recovery.
4493 			 */
4494 			if (whichfork == XFS_COW_FORK)
4495 				xfs_refcount_alloc_cow_extent(tp, bma.blkno,
4496 						bma.length);
4497 		}
4498 
4499 		/* Deal with the allocated space we found.  */
4500 		xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4501 							end, n, flags);
4502 
4503 		/* Execute unwritten extent conversion if necessary */
4504 		error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4505 		if (error == -EAGAIN)
4506 			continue;
4507 		if (error)
4508 			goto error0;
4509 
4510 		/* update the extent map to return */
4511 		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4512 
4513 		/*
4514 		 * If we're done, stop now.  Stop when we've allocated
4515 		 * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4516 		 * the transaction may get too big.
4517 		 */
4518 		if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4519 			break;
4520 
4521 		/* Else go on to the next record. */
4522 		bma.prev = bma.got;
4523 		if (!xfs_iext_next_extent(ifp, &bma.icur, &bma.got))
4524 			eof = true;
4525 	}
4526 	*nmap = n;
4527 
4528 	error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4529 			whichfork);
4530 	if (error)
4531 		goto error0;
4532 
4533 	ASSERT(ifp->if_format != XFS_DINODE_FMT_BTREE ||
4534 	       ifp->if_nextents > XFS_IFORK_MAXEXT(ip, whichfork));
4535 	xfs_bmapi_finish(&bma, whichfork, 0);
4536 	xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4537 		orig_nmap, *nmap);
4538 	return 0;
4539 error0:
4540 	xfs_bmapi_finish(&bma, whichfork, error);
4541 	return error;
4542 }
4543 
4544 /*
4545  * Convert an existing delalloc extent to real blocks based on file offset. This
4546  * attempts to allocate the entire delalloc extent and may require multiple
4547  * invocations to allocate the target offset if a large enough physical extent
4548  * is not available.
4549  */
4550 int
4551 xfs_bmapi_convert_delalloc(
4552 	struct xfs_inode	*ip,
4553 	int			whichfork,
4554 	xfs_off_t		offset,
4555 	struct iomap		*iomap,
4556 	unsigned int		*seq)
4557 {
4558 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
4559 	struct xfs_mount	*mp = ip->i_mount;
4560 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
4561 	struct xfs_bmalloca	bma = { NULL };
4562 	uint16_t		flags = 0;
4563 	struct xfs_trans	*tp;
4564 	int			error;
4565 
4566 	if (whichfork == XFS_COW_FORK)
4567 		flags |= IOMAP_F_SHARED;
4568 
4569 	/*
4570 	 * Space for the extent and indirect blocks was reserved when the
4571 	 * delalloc extent was created so there's no need to do so here.
4572 	 */
4573 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0,
4574 				XFS_TRANS_RESERVE, &tp);
4575 	if (error)
4576 		return error;
4577 
4578 	xfs_ilock(ip, XFS_ILOCK_EXCL);
4579 
4580 	error = xfs_iext_count_may_overflow(ip, whichfork,
4581 			XFS_IEXT_ADD_NOSPLIT_CNT);
4582 	if (error)
4583 		goto out_trans_cancel;
4584 
4585 	xfs_trans_ijoin(tp, ip, 0);
4586 
4587 	if (!xfs_iext_lookup_extent(ip, ifp, offset_fsb, &bma.icur, &bma.got) ||
4588 	    bma.got.br_startoff > offset_fsb) {
4589 		/*
4590 		 * No extent found in the range we are trying to convert.  This
4591 		 * should only happen for the COW fork, where another thread
4592 		 * might have moved the extent to the data fork in the meantime.
4593 		 */
4594 		WARN_ON_ONCE(whichfork != XFS_COW_FORK);
4595 		error = -EAGAIN;
4596 		goto out_trans_cancel;
4597 	}
4598 
4599 	/*
4600 	 * If we find a real extent here we raced with another thread converting
4601 	 * the extent.  Just return the real extent at this offset.
4602 	 */
4603 	if (!isnullstartblock(bma.got.br_startblock)) {
4604 		xfs_bmbt_to_iomap(ip, iomap, &bma.got, flags);
4605 		*seq = READ_ONCE(ifp->if_seq);
4606 		goto out_trans_cancel;
4607 	}
4608 
4609 	bma.tp = tp;
4610 	bma.ip = ip;
4611 	bma.wasdel = true;
4612 	bma.offset = bma.got.br_startoff;
4613 	bma.length = max_t(xfs_filblks_t, bma.got.br_blockcount, MAXEXTLEN);
4614 	bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4615 
4616 	/*
4617 	 * When we're converting the delalloc reservations backing dirty pages
4618 	 * in the page cache, we must be careful about how we create the new
4619 	 * extents:
4620 	 *
4621 	 * New CoW fork extents are created unwritten, turned into real extents
4622 	 * when we're about to write the data to disk, and mapped into the data
4623 	 * fork after the write finishes.  End of story.
4624 	 *
4625 	 * New data fork extents must be mapped in as unwritten and converted
4626 	 * to real extents after the write succeeds to avoid exposing stale
4627 	 * disk contents if we crash.
4628 	 */
4629 	bma.flags = XFS_BMAPI_PREALLOC;
4630 	if (whichfork == XFS_COW_FORK)
4631 		bma.flags |= XFS_BMAPI_COWFORK;
4632 
4633 	if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4634 		bma.prev.br_startoff = NULLFILEOFF;
4635 
4636 	error = xfs_bmapi_allocate(&bma);
4637 	if (error)
4638 		goto out_finish;
4639 
4640 	error = -ENOSPC;
4641 	if (WARN_ON_ONCE(bma.blkno == NULLFSBLOCK))
4642 		goto out_finish;
4643 	error = -EFSCORRUPTED;
4644 	if (WARN_ON_ONCE(!xfs_valid_startblock(ip, bma.got.br_startblock)))
4645 		goto out_finish;
4646 
4647 	XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, bma.length));
4648 	XFS_STATS_INC(mp, xs_xstrat_quick);
4649 
4650 	ASSERT(!isnullstartblock(bma.got.br_startblock));
4651 	xfs_bmbt_to_iomap(ip, iomap, &bma.got, flags);
4652 	*seq = READ_ONCE(ifp->if_seq);
4653 
4654 	if (whichfork == XFS_COW_FORK)
4655 		xfs_refcount_alloc_cow_extent(tp, bma.blkno, bma.length);
4656 
4657 	error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4658 			whichfork);
4659 	if (error)
4660 		goto out_finish;
4661 
4662 	xfs_bmapi_finish(&bma, whichfork, 0);
4663 	error = xfs_trans_commit(tp);
4664 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
4665 	return error;
4666 
4667 out_finish:
4668 	xfs_bmapi_finish(&bma, whichfork, error);
4669 out_trans_cancel:
4670 	xfs_trans_cancel(tp);
4671 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
4672 	return error;
4673 }
4674 
4675 int
4676 xfs_bmapi_remap(
4677 	struct xfs_trans	*tp,
4678 	struct xfs_inode	*ip,
4679 	xfs_fileoff_t		bno,
4680 	xfs_filblks_t		len,
4681 	xfs_fsblock_t		startblock,
4682 	int			flags)
4683 {
4684 	struct xfs_mount	*mp = ip->i_mount;
4685 	struct xfs_ifork	*ifp;
4686 	struct xfs_btree_cur	*cur = NULL;
4687 	struct xfs_bmbt_irec	got;
4688 	struct xfs_iext_cursor	icur;
4689 	int			whichfork = xfs_bmapi_whichfork(flags);
4690 	int			logflags = 0, error;
4691 
4692 	ifp = XFS_IFORK_PTR(ip, whichfork);
4693 	ASSERT(len > 0);
4694 	ASSERT(len <= (xfs_filblks_t)MAXEXTLEN);
4695 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4696 	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC |
4697 			   XFS_BMAPI_NORMAP)));
4698 	ASSERT((flags & (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC)) !=
4699 			(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC));
4700 
4701 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
4702 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
4703 		return -EFSCORRUPTED;
4704 	}
4705 
4706 	if (XFS_FORCED_SHUTDOWN(mp))
4707 		return -EIO;
4708 
4709 	error = xfs_iread_extents(tp, ip, whichfork);
4710 	if (error)
4711 		return error;
4712 
4713 	if (xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
4714 		/* make sure we only reflink into a hole. */
4715 		ASSERT(got.br_startoff > bno);
4716 		ASSERT(got.br_startoff - bno >= len);
4717 	}
4718 
4719 	ip->i_nblocks += len;
4720 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4721 
4722 	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
4723 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
4724 		cur->bc_ino.flags = 0;
4725 	}
4726 
4727 	got.br_startoff = bno;
4728 	got.br_startblock = startblock;
4729 	got.br_blockcount = len;
4730 	if (flags & XFS_BMAPI_PREALLOC)
4731 		got.br_state = XFS_EXT_UNWRITTEN;
4732 	else
4733 		got.br_state = XFS_EXT_NORM;
4734 
4735 	error = xfs_bmap_add_extent_hole_real(tp, ip, whichfork, &icur,
4736 			&cur, &got, &logflags, flags);
4737 	if (error)
4738 		goto error0;
4739 
4740 	error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags, whichfork);
4741 
4742 error0:
4743 	if (ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS)
4744 		logflags &= ~XFS_ILOG_DEXT;
4745 	else if (ip->i_df.if_format != XFS_DINODE_FMT_BTREE)
4746 		logflags &= ~XFS_ILOG_DBROOT;
4747 
4748 	if (logflags)
4749 		xfs_trans_log_inode(tp, ip, logflags);
4750 	if (cur)
4751 		xfs_btree_del_cursor(cur, error);
4752 	return error;
4753 }
4754 
4755 /*
4756  * When a delalloc extent is split (e.g., due to a hole punch), the original
4757  * indlen reservation must be shared across the two new extents that are left
4758  * behind.
4759  *
4760  * Given the original reservation and the worst case indlen for the two new
4761  * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4762  * reservation fairly across the two new extents. If necessary, steal available
4763  * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4764  * ores == 1). The number of stolen blocks is returned. The availability and
4765  * subsequent accounting of stolen blocks is the responsibility of the caller.
4766  */
4767 static xfs_filblks_t
4768 xfs_bmap_split_indlen(
4769 	xfs_filblks_t			ores,		/* original res. */
4770 	xfs_filblks_t			*indlen1,	/* ext1 worst indlen */
4771 	xfs_filblks_t			*indlen2,	/* ext2 worst indlen */
4772 	xfs_filblks_t			avail)		/* stealable blocks */
4773 {
4774 	xfs_filblks_t			len1 = *indlen1;
4775 	xfs_filblks_t			len2 = *indlen2;
4776 	xfs_filblks_t			nres = len1 + len2; /* new total res. */
4777 	xfs_filblks_t			stolen = 0;
4778 	xfs_filblks_t			resfactor;
4779 
4780 	/*
4781 	 * Steal as many blocks as we can to try and satisfy the worst case
4782 	 * indlen for both new extents.
4783 	 */
4784 	if (ores < nres && avail)
4785 		stolen = XFS_FILBLKS_MIN(nres - ores, avail);
4786 	ores += stolen;
4787 
4788 	 /* nothing else to do if we've satisfied the new reservation */
4789 	if (ores >= nres)
4790 		return stolen;
4791 
4792 	/*
4793 	 * We can't meet the total required reservation for the two extents.
4794 	 * Calculate the percent of the overall shortage between both extents
4795 	 * and apply this percentage to each of the requested indlen values.
4796 	 * This distributes the shortage fairly and reduces the chances that one
4797 	 * of the two extents is left with nothing when extents are repeatedly
4798 	 * split.
4799 	 */
4800 	resfactor = (ores * 100);
4801 	do_div(resfactor, nres);
4802 	len1 *= resfactor;
4803 	do_div(len1, 100);
4804 	len2 *= resfactor;
4805 	do_div(len2, 100);
4806 	ASSERT(len1 + len2 <= ores);
4807 	ASSERT(len1 < *indlen1 && len2 < *indlen2);
4808 
4809 	/*
4810 	 * Hand out the remainder to each extent. If one of the two reservations
4811 	 * is zero, we want to make sure that one gets a block first. The loop
4812 	 * below starts with len1, so hand len2 a block right off the bat if it
4813 	 * is zero.
4814 	 */
4815 	ores -= (len1 + len2);
4816 	ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores);
4817 	if (ores && !len2 && *indlen2) {
4818 		len2++;
4819 		ores--;
4820 	}
4821 	while (ores) {
4822 		if (len1 < *indlen1) {
4823 			len1++;
4824 			ores--;
4825 		}
4826 		if (!ores)
4827 			break;
4828 		if (len2 < *indlen2) {
4829 			len2++;
4830 			ores--;
4831 		}
4832 	}
4833 
4834 	*indlen1 = len1;
4835 	*indlen2 = len2;
4836 
4837 	return stolen;
4838 }
4839 
4840 int
4841 xfs_bmap_del_extent_delay(
4842 	struct xfs_inode	*ip,
4843 	int			whichfork,
4844 	struct xfs_iext_cursor	*icur,
4845 	struct xfs_bmbt_irec	*got,
4846 	struct xfs_bmbt_irec	*del)
4847 {
4848 	struct xfs_mount	*mp = ip->i_mount;
4849 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
4850 	struct xfs_bmbt_irec	new;
4851 	int64_t			da_old, da_new, da_diff = 0;
4852 	xfs_fileoff_t		del_endoff, got_endoff;
4853 	xfs_filblks_t		got_indlen, new_indlen, stolen;
4854 	int			state = xfs_bmap_fork_to_state(whichfork);
4855 	int			error = 0;
4856 	bool			isrt;
4857 
4858 	XFS_STATS_INC(mp, xs_del_exlist);
4859 
4860 	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4861 	del_endoff = del->br_startoff + del->br_blockcount;
4862 	got_endoff = got->br_startoff + got->br_blockcount;
4863 	da_old = startblockval(got->br_startblock);
4864 	da_new = 0;
4865 
4866 	ASSERT(del->br_blockcount > 0);
4867 	ASSERT(got->br_startoff <= del->br_startoff);
4868 	ASSERT(got_endoff >= del_endoff);
4869 
4870 	if (isrt) {
4871 		uint64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount);
4872 
4873 		do_div(rtexts, mp->m_sb.sb_rextsize);
4874 		xfs_mod_frextents(mp, rtexts);
4875 	}
4876 
4877 	/*
4878 	 * Update the inode delalloc counter now and wait to update the
4879 	 * sb counters as we might have to borrow some blocks for the
4880 	 * indirect block accounting.
4881 	 */
4882 	ASSERT(!isrt);
4883 	error = xfs_quota_unreserve_blkres(ip, del->br_blockcount);
4884 	if (error)
4885 		return error;
4886 	ip->i_delayed_blks -= del->br_blockcount;
4887 
4888 	if (got->br_startoff == del->br_startoff)
4889 		state |= BMAP_LEFT_FILLING;
4890 	if (got_endoff == del_endoff)
4891 		state |= BMAP_RIGHT_FILLING;
4892 
4893 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4894 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4895 		/*
4896 		 * Matches the whole extent.  Delete the entry.
4897 		 */
4898 		xfs_iext_remove(ip, icur, state);
4899 		xfs_iext_prev(ifp, icur);
4900 		break;
4901 	case BMAP_LEFT_FILLING:
4902 		/*
4903 		 * Deleting the first part of the extent.
4904 		 */
4905 		got->br_startoff = del_endoff;
4906 		got->br_blockcount -= del->br_blockcount;
4907 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4908 				got->br_blockcount), da_old);
4909 		got->br_startblock = nullstartblock((int)da_new);
4910 		xfs_iext_update_extent(ip, state, icur, got);
4911 		break;
4912 	case BMAP_RIGHT_FILLING:
4913 		/*
4914 		 * Deleting the last part of the extent.
4915 		 */
4916 		got->br_blockcount = got->br_blockcount - del->br_blockcount;
4917 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4918 				got->br_blockcount), da_old);
4919 		got->br_startblock = nullstartblock((int)da_new);
4920 		xfs_iext_update_extent(ip, state, icur, got);
4921 		break;
4922 	case 0:
4923 		/*
4924 		 * Deleting the middle of the extent.
4925 		 *
4926 		 * Distribute the original indlen reservation across the two new
4927 		 * extents.  Steal blocks from the deleted extent if necessary.
4928 		 * Stealing blocks simply fudges the fdblocks accounting below.
4929 		 * Warn if either of the new indlen reservations is zero as this
4930 		 * can lead to delalloc problems.
4931 		 */
4932 		got->br_blockcount = del->br_startoff - got->br_startoff;
4933 		got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount);
4934 
4935 		new.br_blockcount = got_endoff - del_endoff;
4936 		new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount);
4937 
4938 		WARN_ON_ONCE(!got_indlen || !new_indlen);
4939 		stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen,
4940 						       del->br_blockcount);
4941 
4942 		got->br_startblock = nullstartblock((int)got_indlen);
4943 
4944 		new.br_startoff = del_endoff;
4945 		new.br_state = got->br_state;
4946 		new.br_startblock = nullstartblock((int)new_indlen);
4947 
4948 		xfs_iext_update_extent(ip, state, icur, got);
4949 		xfs_iext_next(ifp, icur);
4950 		xfs_iext_insert(ip, icur, &new, state);
4951 
4952 		da_new = got_indlen + new_indlen - stolen;
4953 		del->br_blockcount -= stolen;
4954 		break;
4955 	}
4956 
4957 	ASSERT(da_old >= da_new);
4958 	da_diff = da_old - da_new;
4959 	if (!isrt)
4960 		da_diff += del->br_blockcount;
4961 	if (da_diff) {
4962 		xfs_mod_fdblocks(mp, da_diff, false);
4963 		xfs_mod_delalloc(mp, -da_diff);
4964 	}
4965 	return error;
4966 }
4967 
4968 void
4969 xfs_bmap_del_extent_cow(
4970 	struct xfs_inode	*ip,
4971 	struct xfs_iext_cursor	*icur,
4972 	struct xfs_bmbt_irec	*got,
4973 	struct xfs_bmbt_irec	*del)
4974 {
4975 	struct xfs_mount	*mp = ip->i_mount;
4976 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
4977 	struct xfs_bmbt_irec	new;
4978 	xfs_fileoff_t		del_endoff, got_endoff;
4979 	int			state = BMAP_COWFORK;
4980 
4981 	XFS_STATS_INC(mp, xs_del_exlist);
4982 
4983 	del_endoff = del->br_startoff + del->br_blockcount;
4984 	got_endoff = got->br_startoff + got->br_blockcount;
4985 
4986 	ASSERT(del->br_blockcount > 0);
4987 	ASSERT(got->br_startoff <= del->br_startoff);
4988 	ASSERT(got_endoff >= del_endoff);
4989 	ASSERT(!isnullstartblock(got->br_startblock));
4990 
4991 	if (got->br_startoff == del->br_startoff)
4992 		state |= BMAP_LEFT_FILLING;
4993 	if (got_endoff == del_endoff)
4994 		state |= BMAP_RIGHT_FILLING;
4995 
4996 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4997 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4998 		/*
4999 		 * Matches the whole extent.  Delete the entry.
5000 		 */
5001 		xfs_iext_remove(ip, icur, state);
5002 		xfs_iext_prev(ifp, icur);
5003 		break;
5004 	case BMAP_LEFT_FILLING:
5005 		/*
5006 		 * Deleting the first part of the extent.
5007 		 */
5008 		got->br_startoff = del_endoff;
5009 		got->br_blockcount -= del->br_blockcount;
5010 		got->br_startblock = del->br_startblock + del->br_blockcount;
5011 		xfs_iext_update_extent(ip, state, icur, got);
5012 		break;
5013 	case BMAP_RIGHT_FILLING:
5014 		/*
5015 		 * Deleting the last part of the extent.
5016 		 */
5017 		got->br_blockcount -= del->br_blockcount;
5018 		xfs_iext_update_extent(ip, state, icur, got);
5019 		break;
5020 	case 0:
5021 		/*
5022 		 * Deleting the middle of the extent.
5023 		 */
5024 		got->br_blockcount = del->br_startoff - got->br_startoff;
5025 
5026 		new.br_startoff = del_endoff;
5027 		new.br_blockcount = got_endoff - del_endoff;
5028 		new.br_state = got->br_state;
5029 		new.br_startblock = del->br_startblock + del->br_blockcount;
5030 
5031 		xfs_iext_update_extent(ip, state, icur, got);
5032 		xfs_iext_next(ifp, icur);
5033 		xfs_iext_insert(ip, icur, &new, state);
5034 		break;
5035 	}
5036 	ip->i_delayed_blks -= del->br_blockcount;
5037 }
5038 
5039 /*
5040  * Called by xfs_bmapi to update file extent records and the btree
5041  * after removing space.
5042  */
5043 STATIC int				/* error */
5044 xfs_bmap_del_extent_real(
5045 	xfs_inode_t		*ip,	/* incore inode pointer */
5046 	xfs_trans_t		*tp,	/* current transaction pointer */
5047 	struct xfs_iext_cursor	*icur,
5048 	xfs_btree_cur_t		*cur,	/* if null, not a btree */
5049 	xfs_bmbt_irec_t		*del,	/* data to remove from extents */
5050 	int			*logflagsp, /* inode logging flags */
5051 	int			whichfork, /* data or attr fork */
5052 	int			bflags)	/* bmapi flags */
5053 {
5054 	xfs_fsblock_t		del_endblock=0;	/* first block past del */
5055 	xfs_fileoff_t		del_endoff;	/* first offset past del */
5056 	int			do_fx;	/* free extent at end of routine */
5057 	int			error;	/* error return value */
5058 	int			flags = 0;/* inode logging flags */
5059 	struct xfs_bmbt_irec	got;	/* current extent entry */
5060 	xfs_fileoff_t		got_endoff;	/* first offset past got */
5061 	int			i;	/* temp state */
5062 	struct xfs_ifork	*ifp;	/* inode fork pointer */
5063 	xfs_mount_t		*mp;	/* mount structure */
5064 	xfs_filblks_t		nblks;	/* quota/sb block count */
5065 	xfs_bmbt_irec_t		new;	/* new record to be inserted */
5066 	/* REFERENCED */
5067 	uint			qfield;	/* quota field to update */
5068 	int			state = xfs_bmap_fork_to_state(whichfork);
5069 	struct xfs_bmbt_irec	old;
5070 
5071 	mp = ip->i_mount;
5072 	XFS_STATS_INC(mp, xs_del_exlist);
5073 
5074 	ifp = XFS_IFORK_PTR(ip, whichfork);
5075 	ASSERT(del->br_blockcount > 0);
5076 	xfs_iext_get_extent(ifp, icur, &got);
5077 	ASSERT(got.br_startoff <= del->br_startoff);
5078 	del_endoff = del->br_startoff + del->br_blockcount;
5079 	got_endoff = got.br_startoff + got.br_blockcount;
5080 	ASSERT(got_endoff >= del_endoff);
5081 	ASSERT(!isnullstartblock(got.br_startblock));
5082 	qfield = 0;
5083 	error = 0;
5084 
5085 	/*
5086 	 * If it's the case where the directory code is running with no block
5087 	 * reservation, and the deleted block is in the middle of its extent,
5088 	 * and the resulting insert of an extent would cause transformation to
5089 	 * btree format, then reject it.  The calling code will then swap blocks
5090 	 * around instead.  We have to do this now, rather than waiting for the
5091 	 * conversion to btree format, since the transaction will be dirty then.
5092 	 */
5093 	if (tp->t_blk_res == 0 &&
5094 	    ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
5095 	    ifp->if_nextents >= XFS_IFORK_MAXEXT(ip, whichfork) &&
5096 	    del->br_startoff > got.br_startoff && del_endoff < got_endoff)
5097 		return -ENOSPC;
5098 
5099 	flags = XFS_ILOG_CORE;
5100 	if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
5101 		xfs_filblks_t	len;
5102 		xfs_extlen_t	mod;
5103 
5104 		len = div_u64_rem(del->br_blockcount, mp->m_sb.sb_rextsize,
5105 				  &mod);
5106 		ASSERT(mod == 0);
5107 
5108 		if (!(bflags & XFS_BMAPI_REMAP)) {
5109 			xfs_fsblock_t	bno;
5110 
5111 			bno = div_u64_rem(del->br_startblock,
5112 					mp->m_sb.sb_rextsize, &mod);
5113 			ASSERT(mod == 0);
5114 
5115 			error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
5116 			if (error)
5117 				goto done;
5118 		}
5119 
5120 		do_fx = 0;
5121 		nblks = len * mp->m_sb.sb_rextsize;
5122 		qfield = XFS_TRANS_DQ_RTBCOUNT;
5123 	} else {
5124 		do_fx = 1;
5125 		nblks = del->br_blockcount;
5126 		qfield = XFS_TRANS_DQ_BCOUNT;
5127 	}
5128 
5129 	del_endblock = del->br_startblock + del->br_blockcount;
5130 	if (cur) {
5131 		error = xfs_bmbt_lookup_eq(cur, &got, &i);
5132 		if (error)
5133 			goto done;
5134 		if (XFS_IS_CORRUPT(mp, i != 1)) {
5135 			error = -EFSCORRUPTED;
5136 			goto done;
5137 		}
5138 	}
5139 
5140 	if (got.br_startoff == del->br_startoff)
5141 		state |= BMAP_LEFT_FILLING;
5142 	if (got_endoff == del_endoff)
5143 		state |= BMAP_RIGHT_FILLING;
5144 
5145 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
5146 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
5147 		/*
5148 		 * Matches the whole extent.  Delete the entry.
5149 		 */
5150 		xfs_iext_remove(ip, icur, state);
5151 		xfs_iext_prev(ifp, icur);
5152 		ifp->if_nextents--;
5153 
5154 		flags |= XFS_ILOG_CORE;
5155 		if (!cur) {
5156 			flags |= xfs_ilog_fext(whichfork);
5157 			break;
5158 		}
5159 		if ((error = xfs_btree_delete(cur, &i)))
5160 			goto done;
5161 		if (XFS_IS_CORRUPT(mp, i != 1)) {
5162 			error = -EFSCORRUPTED;
5163 			goto done;
5164 		}
5165 		break;
5166 	case BMAP_LEFT_FILLING:
5167 		/*
5168 		 * Deleting the first part of the extent.
5169 		 */
5170 		got.br_startoff = del_endoff;
5171 		got.br_startblock = del_endblock;
5172 		got.br_blockcount -= del->br_blockcount;
5173 		xfs_iext_update_extent(ip, state, icur, &got);
5174 		if (!cur) {
5175 			flags |= xfs_ilog_fext(whichfork);
5176 			break;
5177 		}
5178 		error = xfs_bmbt_update(cur, &got);
5179 		if (error)
5180 			goto done;
5181 		break;
5182 	case BMAP_RIGHT_FILLING:
5183 		/*
5184 		 * Deleting the last part of the extent.
5185 		 */
5186 		got.br_blockcount -= del->br_blockcount;
5187 		xfs_iext_update_extent(ip, state, icur, &got);
5188 		if (!cur) {
5189 			flags |= xfs_ilog_fext(whichfork);
5190 			break;
5191 		}
5192 		error = xfs_bmbt_update(cur, &got);
5193 		if (error)
5194 			goto done;
5195 		break;
5196 	case 0:
5197 		/*
5198 		 * Deleting the middle of the extent.
5199 		 */
5200 
5201 		/*
5202 		 * For directories, -ENOSPC is returned since a directory entry
5203 		 * remove operation must not fail due to low extent count
5204 		 * availability. -ENOSPC will be handled by higher layers of XFS
5205 		 * by letting the corresponding empty Data/Free blocks to linger
5206 		 * until a future remove operation. Dabtree blocks would be
5207 		 * swapped with the last block in the leaf space and then the
5208 		 * new last block will be unmapped.
5209 		 *
5210 		 * The above logic also applies to the source directory entry of
5211 		 * a rename operation.
5212 		 */
5213 		error = xfs_iext_count_may_overflow(ip, whichfork, 1);
5214 		if (error) {
5215 			ASSERT(S_ISDIR(VFS_I(ip)->i_mode) &&
5216 				whichfork == XFS_DATA_FORK);
5217 			error = -ENOSPC;
5218 			goto done;
5219 		}
5220 
5221 		old = got;
5222 
5223 		got.br_blockcount = del->br_startoff - got.br_startoff;
5224 		xfs_iext_update_extent(ip, state, icur, &got);
5225 
5226 		new.br_startoff = del_endoff;
5227 		new.br_blockcount = got_endoff - del_endoff;
5228 		new.br_state = got.br_state;
5229 		new.br_startblock = del_endblock;
5230 
5231 		flags |= XFS_ILOG_CORE;
5232 		if (cur) {
5233 			error = xfs_bmbt_update(cur, &got);
5234 			if (error)
5235 				goto done;
5236 			error = xfs_btree_increment(cur, 0, &i);
5237 			if (error)
5238 				goto done;
5239 			cur->bc_rec.b = new;
5240 			error = xfs_btree_insert(cur, &i);
5241 			if (error && error != -ENOSPC)
5242 				goto done;
5243 			/*
5244 			 * If get no-space back from btree insert, it tried a
5245 			 * split, and we have a zero block reservation.  Fix up
5246 			 * our state and return the error.
5247 			 */
5248 			if (error == -ENOSPC) {
5249 				/*
5250 				 * Reset the cursor, don't trust it after any
5251 				 * insert operation.
5252 				 */
5253 				error = xfs_bmbt_lookup_eq(cur, &got, &i);
5254 				if (error)
5255 					goto done;
5256 				if (XFS_IS_CORRUPT(mp, i != 1)) {
5257 					error = -EFSCORRUPTED;
5258 					goto done;
5259 				}
5260 				/*
5261 				 * Update the btree record back
5262 				 * to the original value.
5263 				 */
5264 				error = xfs_bmbt_update(cur, &old);
5265 				if (error)
5266 					goto done;
5267 				/*
5268 				 * Reset the extent record back
5269 				 * to the original value.
5270 				 */
5271 				xfs_iext_update_extent(ip, state, icur, &old);
5272 				flags = 0;
5273 				error = -ENOSPC;
5274 				goto done;
5275 			}
5276 			if (XFS_IS_CORRUPT(mp, i != 1)) {
5277 				error = -EFSCORRUPTED;
5278 				goto done;
5279 			}
5280 		} else
5281 			flags |= xfs_ilog_fext(whichfork);
5282 
5283 		ifp->if_nextents++;
5284 		xfs_iext_next(ifp, icur);
5285 		xfs_iext_insert(ip, icur, &new, state);
5286 		break;
5287 	}
5288 
5289 	/* remove reverse mapping */
5290 	xfs_rmap_unmap_extent(tp, ip, whichfork, del);
5291 
5292 	/*
5293 	 * If we need to, add to list of extents to delete.
5294 	 */
5295 	if (do_fx && !(bflags & XFS_BMAPI_REMAP)) {
5296 		if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
5297 			xfs_refcount_decrease_extent(tp, del);
5298 		} else {
5299 			__xfs_bmap_add_free(tp, del->br_startblock,
5300 					del->br_blockcount, NULL,
5301 					(bflags & XFS_BMAPI_NODISCARD) ||
5302 					del->br_state == XFS_EXT_UNWRITTEN);
5303 		}
5304 	}
5305 
5306 	/*
5307 	 * Adjust inode # blocks in the file.
5308 	 */
5309 	if (nblks)
5310 		ip->i_nblocks -= nblks;
5311 	/*
5312 	 * Adjust quota data.
5313 	 */
5314 	if (qfield && !(bflags & XFS_BMAPI_REMAP))
5315 		xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5316 
5317 done:
5318 	*logflagsp = flags;
5319 	return error;
5320 }
5321 
5322 /*
5323  * Unmap (remove) blocks from a file.
5324  * If nexts is nonzero then the number of extents to remove is limited to
5325  * that value.  If not all extents in the block range can be removed then
5326  * *done is set.
5327  */
5328 int						/* error */
5329 __xfs_bunmapi(
5330 	struct xfs_trans	*tp,		/* transaction pointer */
5331 	struct xfs_inode	*ip,		/* incore inode */
5332 	xfs_fileoff_t		start,		/* first file offset deleted */
5333 	xfs_filblks_t		*rlen,		/* i/o: amount remaining */
5334 	int			flags,		/* misc flags */
5335 	xfs_extnum_t		nexts)		/* number of extents max */
5336 {
5337 	struct xfs_btree_cur	*cur;		/* bmap btree cursor */
5338 	struct xfs_bmbt_irec	del;		/* extent being deleted */
5339 	int			error;		/* error return value */
5340 	xfs_extnum_t		extno;		/* extent number in list */
5341 	struct xfs_bmbt_irec	got;		/* current extent record */
5342 	struct xfs_ifork	*ifp;		/* inode fork pointer */
5343 	int			isrt;		/* freeing in rt area */
5344 	int			logflags;	/* transaction logging flags */
5345 	xfs_extlen_t		mod;		/* rt extent offset */
5346 	struct xfs_mount	*mp = ip->i_mount;
5347 	int			tmp_logflags;	/* partial logging flags */
5348 	int			wasdel;		/* was a delayed alloc extent */
5349 	int			whichfork;	/* data or attribute fork */
5350 	xfs_fsblock_t		sum;
5351 	xfs_filblks_t		len = *rlen;	/* length to unmap in file */
5352 	xfs_fileoff_t		max_len;
5353 	xfs_fileoff_t		end;
5354 	struct xfs_iext_cursor	icur;
5355 	bool			done = false;
5356 
5357 	trace_xfs_bunmap(ip, start, len, flags, _RET_IP_);
5358 
5359 	whichfork = xfs_bmapi_whichfork(flags);
5360 	ASSERT(whichfork != XFS_COW_FORK);
5361 	ifp = XFS_IFORK_PTR(ip, whichfork);
5362 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)))
5363 		return -EFSCORRUPTED;
5364 	if (XFS_FORCED_SHUTDOWN(mp))
5365 		return -EIO;
5366 
5367 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5368 	ASSERT(len > 0);
5369 	ASSERT(nexts >= 0);
5370 
5371 	/*
5372 	 * Guesstimate how many blocks we can unmap without running the risk of
5373 	 * blowing out the transaction with a mix of EFIs and reflink
5374 	 * adjustments.
5375 	 */
5376 	if (tp && xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK)
5377 		max_len = min(len, xfs_refcount_max_unmap(tp->t_log_res));
5378 	else
5379 		max_len = len;
5380 
5381 	error = xfs_iread_extents(tp, ip, whichfork);
5382 	if (error)
5383 		return error;
5384 
5385 	if (xfs_iext_count(ifp) == 0) {
5386 		*rlen = 0;
5387 		return 0;
5388 	}
5389 	XFS_STATS_INC(mp, xs_blk_unmap);
5390 	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5391 	end = start + len;
5392 
5393 	if (!xfs_iext_lookup_extent_before(ip, ifp, &end, &icur, &got)) {
5394 		*rlen = 0;
5395 		return 0;
5396 	}
5397 	end--;
5398 
5399 	logflags = 0;
5400 	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
5401 		ASSERT(ifp->if_format == XFS_DINODE_FMT_BTREE);
5402 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5403 		cur->bc_ino.flags = 0;
5404 	} else
5405 		cur = NULL;
5406 
5407 	if (isrt) {
5408 		/*
5409 		 * Synchronize by locking the bitmap inode.
5410 		 */
5411 		xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
5412 		xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5413 		xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
5414 		xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
5415 	}
5416 
5417 	extno = 0;
5418 	while (end != (xfs_fileoff_t)-1 && end >= start &&
5419 	       (nexts == 0 || extno < nexts) && max_len > 0) {
5420 		/*
5421 		 * Is the found extent after a hole in which end lives?
5422 		 * Just back up to the previous extent, if so.
5423 		 */
5424 		if (got.br_startoff > end &&
5425 		    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5426 			done = true;
5427 			break;
5428 		}
5429 		/*
5430 		 * Is the last block of this extent before the range
5431 		 * we're supposed to delete?  If so, we're done.
5432 		 */
5433 		end = XFS_FILEOFF_MIN(end,
5434 			got.br_startoff + got.br_blockcount - 1);
5435 		if (end < start)
5436 			break;
5437 		/*
5438 		 * Then deal with the (possibly delayed) allocated space
5439 		 * we found.
5440 		 */
5441 		del = got;
5442 		wasdel = isnullstartblock(del.br_startblock);
5443 
5444 		if (got.br_startoff < start) {
5445 			del.br_startoff = start;
5446 			del.br_blockcount -= start - got.br_startoff;
5447 			if (!wasdel)
5448 				del.br_startblock += start - got.br_startoff;
5449 		}
5450 		if (del.br_startoff + del.br_blockcount > end + 1)
5451 			del.br_blockcount = end + 1 - del.br_startoff;
5452 
5453 		/* How much can we safely unmap? */
5454 		if (max_len < del.br_blockcount) {
5455 			del.br_startoff += del.br_blockcount - max_len;
5456 			if (!wasdel)
5457 				del.br_startblock += del.br_blockcount - max_len;
5458 			del.br_blockcount = max_len;
5459 		}
5460 
5461 		if (!isrt)
5462 			goto delete;
5463 
5464 		sum = del.br_startblock + del.br_blockcount;
5465 		div_u64_rem(sum, mp->m_sb.sb_rextsize, &mod);
5466 		if (mod) {
5467 			/*
5468 			 * Realtime extent not lined up at the end.
5469 			 * The extent could have been split into written
5470 			 * and unwritten pieces, or we could just be
5471 			 * unmapping part of it.  But we can't really
5472 			 * get rid of part of a realtime extent.
5473 			 */
5474 			if (del.br_state == XFS_EXT_UNWRITTEN) {
5475 				/*
5476 				 * This piece is unwritten, or we're not
5477 				 * using unwritten extents.  Skip over it.
5478 				 */
5479 				ASSERT(end >= mod);
5480 				end -= mod > del.br_blockcount ?
5481 					del.br_blockcount : mod;
5482 				if (end < got.br_startoff &&
5483 				    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5484 					done = true;
5485 					break;
5486 				}
5487 				continue;
5488 			}
5489 			/*
5490 			 * It's written, turn it unwritten.
5491 			 * This is better than zeroing it.
5492 			 */
5493 			ASSERT(del.br_state == XFS_EXT_NORM);
5494 			ASSERT(tp->t_blk_res > 0);
5495 			/*
5496 			 * If this spans a realtime extent boundary,
5497 			 * chop it back to the start of the one we end at.
5498 			 */
5499 			if (del.br_blockcount > mod) {
5500 				del.br_startoff += del.br_blockcount - mod;
5501 				del.br_startblock += del.br_blockcount - mod;
5502 				del.br_blockcount = mod;
5503 			}
5504 			del.br_state = XFS_EXT_UNWRITTEN;
5505 			error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5506 					whichfork, &icur, &cur, &del,
5507 					&logflags);
5508 			if (error)
5509 				goto error0;
5510 			goto nodelete;
5511 		}
5512 		div_u64_rem(del.br_startblock, mp->m_sb.sb_rextsize, &mod);
5513 		if (mod) {
5514 			xfs_extlen_t off = mp->m_sb.sb_rextsize - mod;
5515 
5516 			/*
5517 			 * Realtime extent is lined up at the end but not
5518 			 * at the front.  We'll get rid of full extents if
5519 			 * we can.
5520 			 */
5521 			if (del.br_blockcount > off) {
5522 				del.br_blockcount -= off;
5523 				del.br_startoff += off;
5524 				del.br_startblock += off;
5525 			} else if (del.br_startoff == start &&
5526 				   (del.br_state == XFS_EXT_UNWRITTEN ||
5527 				    tp->t_blk_res == 0)) {
5528 				/*
5529 				 * Can't make it unwritten.  There isn't
5530 				 * a full extent here so just skip it.
5531 				 */
5532 				ASSERT(end >= del.br_blockcount);
5533 				end -= del.br_blockcount;
5534 				if (got.br_startoff > end &&
5535 				    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5536 					done = true;
5537 					break;
5538 				}
5539 				continue;
5540 			} else if (del.br_state == XFS_EXT_UNWRITTEN) {
5541 				struct xfs_bmbt_irec	prev;
5542 				xfs_fileoff_t		unwrite_start;
5543 
5544 				/*
5545 				 * This one is already unwritten.
5546 				 * It must have a written left neighbor.
5547 				 * Unwrite the killed part of that one and
5548 				 * try again.
5549 				 */
5550 				if (!xfs_iext_prev_extent(ifp, &icur, &prev))
5551 					ASSERT(0);
5552 				ASSERT(prev.br_state == XFS_EXT_NORM);
5553 				ASSERT(!isnullstartblock(prev.br_startblock));
5554 				ASSERT(del.br_startblock ==
5555 				       prev.br_startblock + prev.br_blockcount);
5556 				unwrite_start = max3(start,
5557 						     del.br_startoff - mod,
5558 						     prev.br_startoff);
5559 				mod = unwrite_start - prev.br_startoff;
5560 				prev.br_startoff = unwrite_start;
5561 				prev.br_startblock += mod;
5562 				prev.br_blockcount -= mod;
5563 				prev.br_state = XFS_EXT_UNWRITTEN;
5564 				error = xfs_bmap_add_extent_unwritten_real(tp,
5565 						ip, whichfork, &icur, &cur,
5566 						&prev, &logflags);
5567 				if (error)
5568 					goto error0;
5569 				goto nodelete;
5570 			} else {
5571 				ASSERT(del.br_state == XFS_EXT_NORM);
5572 				del.br_state = XFS_EXT_UNWRITTEN;
5573 				error = xfs_bmap_add_extent_unwritten_real(tp,
5574 						ip, whichfork, &icur, &cur,
5575 						&del, &logflags);
5576 				if (error)
5577 					goto error0;
5578 				goto nodelete;
5579 			}
5580 		}
5581 
5582 delete:
5583 		if (wasdel) {
5584 			error = xfs_bmap_del_extent_delay(ip, whichfork, &icur,
5585 					&got, &del);
5586 		} else {
5587 			error = xfs_bmap_del_extent_real(ip, tp, &icur, cur,
5588 					&del, &tmp_logflags, whichfork,
5589 					flags);
5590 			logflags |= tmp_logflags;
5591 		}
5592 
5593 		if (error)
5594 			goto error0;
5595 
5596 		max_len -= del.br_blockcount;
5597 		end = del.br_startoff - 1;
5598 nodelete:
5599 		/*
5600 		 * If not done go on to the next (previous) record.
5601 		 */
5602 		if (end != (xfs_fileoff_t)-1 && end >= start) {
5603 			if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5604 			    (got.br_startoff > end &&
5605 			     !xfs_iext_prev_extent(ifp, &icur, &got))) {
5606 				done = true;
5607 				break;
5608 			}
5609 			extno++;
5610 		}
5611 	}
5612 	if (done || end == (xfs_fileoff_t)-1 || end < start)
5613 		*rlen = 0;
5614 	else
5615 		*rlen = end - start + 1;
5616 
5617 	/*
5618 	 * Convert to a btree if necessary.
5619 	 */
5620 	if (xfs_bmap_needs_btree(ip, whichfork)) {
5621 		ASSERT(cur == NULL);
5622 		error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
5623 				&tmp_logflags, whichfork);
5624 		logflags |= tmp_logflags;
5625 	} else {
5626 		error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags,
5627 			whichfork);
5628 	}
5629 
5630 error0:
5631 	/*
5632 	 * Log everything.  Do this after conversion, there's no point in
5633 	 * logging the extent records if we've converted to btree format.
5634 	 */
5635 	if ((logflags & xfs_ilog_fext(whichfork)) &&
5636 	    ifp->if_format != XFS_DINODE_FMT_EXTENTS)
5637 		logflags &= ~xfs_ilog_fext(whichfork);
5638 	else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5639 		 ifp->if_format != XFS_DINODE_FMT_BTREE)
5640 		logflags &= ~xfs_ilog_fbroot(whichfork);
5641 	/*
5642 	 * Log inode even in the error case, if the transaction
5643 	 * is dirty we'll need to shut down the filesystem.
5644 	 */
5645 	if (logflags)
5646 		xfs_trans_log_inode(tp, ip, logflags);
5647 	if (cur) {
5648 		if (!error)
5649 			cur->bc_ino.allocated = 0;
5650 		xfs_btree_del_cursor(cur, error);
5651 	}
5652 	return error;
5653 }
5654 
5655 /* Unmap a range of a file. */
5656 int
5657 xfs_bunmapi(
5658 	xfs_trans_t		*tp,
5659 	struct xfs_inode	*ip,
5660 	xfs_fileoff_t		bno,
5661 	xfs_filblks_t		len,
5662 	int			flags,
5663 	xfs_extnum_t		nexts,
5664 	int			*done)
5665 {
5666 	int			error;
5667 
5668 	error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts);
5669 	*done = (len == 0);
5670 	return error;
5671 }
5672 
5673 /*
5674  * Determine whether an extent shift can be accomplished by a merge with the
5675  * extent that precedes the target hole of the shift.
5676  */
5677 STATIC bool
5678 xfs_bmse_can_merge(
5679 	struct xfs_bmbt_irec	*left,	/* preceding extent */
5680 	struct xfs_bmbt_irec	*got,	/* current extent to shift */
5681 	xfs_fileoff_t		shift)	/* shift fsb */
5682 {
5683 	xfs_fileoff_t		startoff;
5684 
5685 	startoff = got->br_startoff - shift;
5686 
5687 	/*
5688 	 * The extent, once shifted, must be adjacent in-file and on-disk with
5689 	 * the preceding extent.
5690 	 */
5691 	if ((left->br_startoff + left->br_blockcount != startoff) ||
5692 	    (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5693 	    (left->br_state != got->br_state) ||
5694 	    (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5695 		return false;
5696 
5697 	return true;
5698 }
5699 
5700 /*
5701  * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5702  * hole in the file. If an extent shift would result in the extent being fully
5703  * adjacent to the extent that currently precedes the hole, we can merge with
5704  * the preceding extent rather than do the shift.
5705  *
5706  * This function assumes the caller has verified a shift-by-merge is possible
5707  * with the provided extents via xfs_bmse_can_merge().
5708  */
5709 STATIC int
5710 xfs_bmse_merge(
5711 	struct xfs_trans		*tp,
5712 	struct xfs_inode		*ip,
5713 	int				whichfork,
5714 	xfs_fileoff_t			shift,		/* shift fsb */
5715 	struct xfs_iext_cursor		*icur,
5716 	struct xfs_bmbt_irec		*got,		/* extent to shift */
5717 	struct xfs_bmbt_irec		*left,		/* preceding extent */
5718 	struct xfs_btree_cur		*cur,
5719 	int				*logflags)	/* output */
5720 {
5721 	struct xfs_ifork		*ifp = XFS_IFORK_PTR(ip, whichfork);
5722 	struct xfs_bmbt_irec		new;
5723 	xfs_filblks_t			blockcount;
5724 	int				error, i;
5725 	struct xfs_mount		*mp = ip->i_mount;
5726 
5727 	blockcount = left->br_blockcount + got->br_blockcount;
5728 
5729 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5730 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5731 	ASSERT(xfs_bmse_can_merge(left, got, shift));
5732 
5733 	new = *left;
5734 	new.br_blockcount = blockcount;
5735 
5736 	/*
5737 	 * Update the on-disk extent count, the btree if necessary and log the
5738 	 * inode.
5739 	 */
5740 	ifp->if_nextents--;
5741 	*logflags |= XFS_ILOG_CORE;
5742 	if (!cur) {
5743 		*logflags |= XFS_ILOG_DEXT;
5744 		goto done;
5745 	}
5746 
5747 	/* lookup and remove the extent to merge */
5748 	error = xfs_bmbt_lookup_eq(cur, got, &i);
5749 	if (error)
5750 		return error;
5751 	if (XFS_IS_CORRUPT(mp, i != 1))
5752 		return -EFSCORRUPTED;
5753 
5754 	error = xfs_btree_delete(cur, &i);
5755 	if (error)
5756 		return error;
5757 	if (XFS_IS_CORRUPT(mp, i != 1))
5758 		return -EFSCORRUPTED;
5759 
5760 	/* lookup and update size of the previous extent */
5761 	error = xfs_bmbt_lookup_eq(cur, left, &i);
5762 	if (error)
5763 		return error;
5764 	if (XFS_IS_CORRUPT(mp, i != 1))
5765 		return -EFSCORRUPTED;
5766 
5767 	error = xfs_bmbt_update(cur, &new);
5768 	if (error)
5769 		return error;
5770 
5771 	/* change to extent format if required after extent removal */
5772 	error = xfs_bmap_btree_to_extents(tp, ip, cur, logflags, whichfork);
5773 	if (error)
5774 		return error;
5775 
5776 done:
5777 	xfs_iext_remove(ip, icur, 0);
5778 	xfs_iext_prev(ifp, icur);
5779 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5780 			&new);
5781 
5782 	/* update reverse mapping. rmap functions merge the rmaps for us */
5783 	xfs_rmap_unmap_extent(tp, ip, whichfork, got);
5784 	memcpy(&new, got, sizeof(new));
5785 	new.br_startoff = left->br_startoff + left->br_blockcount;
5786 	xfs_rmap_map_extent(tp, ip, whichfork, &new);
5787 	return 0;
5788 }
5789 
5790 static int
5791 xfs_bmap_shift_update_extent(
5792 	struct xfs_trans	*tp,
5793 	struct xfs_inode	*ip,
5794 	int			whichfork,
5795 	struct xfs_iext_cursor	*icur,
5796 	struct xfs_bmbt_irec	*got,
5797 	struct xfs_btree_cur	*cur,
5798 	int			*logflags,
5799 	xfs_fileoff_t		startoff)
5800 {
5801 	struct xfs_mount	*mp = ip->i_mount;
5802 	struct xfs_bmbt_irec	prev = *got;
5803 	int			error, i;
5804 
5805 	*logflags |= XFS_ILOG_CORE;
5806 
5807 	got->br_startoff = startoff;
5808 
5809 	if (cur) {
5810 		error = xfs_bmbt_lookup_eq(cur, &prev, &i);
5811 		if (error)
5812 			return error;
5813 		if (XFS_IS_CORRUPT(mp, i != 1))
5814 			return -EFSCORRUPTED;
5815 
5816 		error = xfs_bmbt_update(cur, got);
5817 		if (error)
5818 			return error;
5819 	} else {
5820 		*logflags |= XFS_ILOG_DEXT;
5821 	}
5822 
5823 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5824 			got);
5825 
5826 	/* update reverse mapping */
5827 	xfs_rmap_unmap_extent(tp, ip, whichfork, &prev);
5828 	xfs_rmap_map_extent(tp, ip, whichfork, got);
5829 	return 0;
5830 }
5831 
5832 int
5833 xfs_bmap_collapse_extents(
5834 	struct xfs_trans	*tp,
5835 	struct xfs_inode	*ip,
5836 	xfs_fileoff_t		*next_fsb,
5837 	xfs_fileoff_t		offset_shift_fsb,
5838 	bool			*done)
5839 {
5840 	int			whichfork = XFS_DATA_FORK;
5841 	struct xfs_mount	*mp = ip->i_mount;
5842 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
5843 	struct xfs_btree_cur	*cur = NULL;
5844 	struct xfs_bmbt_irec	got, prev;
5845 	struct xfs_iext_cursor	icur;
5846 	xfs_fileoff_t		new_startoff;
5847 	int			error = 0;
5848 	int			logflags = 0;
5849 
5850 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
5851 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
5852 		return -EFSCORRUPTED;
5853 	}
5854 
5855 	if (XFS_FORCED_SHUTDOWN(mp))
5856 		return -EIO;
5857 
5858 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL));
5859 
5860 	error = xfs_iread_extents(tp, ip, whichfork);
5861 	if (error)
5862 		return error;
5863 
5864 	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
5865 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5866 		cur->bc_ino.flags = 0;
5867 	}
5868 
5869 	if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5870 		*done = true;
5871 		goto del_cursor;
5872 	}
5873 	if (XFS_IS_CORRUPT(mp, isnullstartblock(got.br_startblock))) {
5874 		error = -EFSCORRUPTED;
5875 		goto del_cursor;
5876 	}
5877 
5878 	new_startoff = got.br_startoff - offset_shift_fsb;
5879 	if (xfs_iext_peek_prev_extent(ifp, &icur, &prev)) {
5880 		if (new_startoff < prev.br_startoff + prev.br_blockcount) {
5881 			error = -EINVAL;
5882 			goto del_cursor;
5883 		}
5884 
5885 		if (xfs_bmse_can_merge(&prev, &got, offset_shift_fsb)) {
5886 			error = xfs_bmse_merge(tp, ip, whichfork,
5887 					offset_shift_fsb, &icur, &got, &prev,
5888 					cur, &logflags);
5889 			if (error)
5890 				goto del_cursor;
5891 			goto done;
5892 		}
5893 	} else {
5894 		if (got.br_startoff < offset_shift_fsb) {
5895 			error = -EINVAL;
5896 			goto del_cursor;
5897 		}
5898 	}
5899 
5900 	error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got,
5901 			cur, &logflags, new_startoff);
5902 	if (error)
5903 		goto del_cursor;
5904 
5905 done:
5906 	if (!xfs_iext_next_extent(ifp, &icur, &got)) {
5907 		*done = true;
5908 		goto del_cursor;
5909 	}
5910 
5911 	*next_fsb = got.br_startoff;
5912 del_cursor:
5913 	if (cur)
5914 		xfs_btree_del_cursor(cur, error);
5915 	if (logflags)
5916 		xfs_trans_log_inode(tp, ip, logflags);
5917 	return error;
5918 }
5919 
5920 /* Make sure we won't be right-shifting an extent past the maximum bound. */
5921 int
5922 xfs_bmap_can_insert_extents(
5923 	struct xfs_inode	*ip,
5924 	xfs_fileoff_t		off,
5925 	xfs_fileoff_t		shift)
5926 {
5927 	struct xfs_bmbt_irec	got;
5928 	int			is_empty;
5929 	int			error = 0;
5930 
5931 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5932 
5933 	if (XFS_FORCED_SHUTDOWN(ip->i_mount))
5934 		return -EIO;
5935 
5936 	xfs_ilock(ip, XFS_ILOCK_EXCL);
5937 	error = xfs_bmap_last_extent(NULL, ip, XFS_DATA_FORK, &got, &is_empty);
5938 	if (!error && !is_empty && got.br_startoff >= off &&
5939 	    ((got.br_startoff + shift) & BMBT_STARTOFF_MASK) < got.br_startoff)
5940 		error = -EINVAL;
5941 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
5942 
5943 	return error;
5944 }
5945 
5946 int
5947 xfs_bmap_insert_extents(
5948 	struct xfs_trans	*tp,
5949 	struct xfs_inode	*ip,
5950 	xfs_fileoff_t		*next_fsb,
5951 	xfs_fileoff_t		offset_shift_fsb,
5952 	bool			*done,
5953 	xfs_fileoff_t		stop_fsb)
5954 {
5955 	int			whichfork = XFS_DATA_FORK;
5956 	struct xfs_mount	*mp = ip->i_mount;
5957 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
5958 	struct xfs_btree_cur	*cur = NULL;
5959 	struct xfs_bmbt_irec	got, next;
5960 	struct xfs_iext_cursor	icur;
5961 	xfs_fileoff_t		new_startoff;
5962 	int			error = 0;
5963 	int			logflags = 0;
5964 
5965 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
5966 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
5967 		return -EFSCORRUPTED;
5968 	}
5969 
5970 	if (XFS_FORCED_SHUTDOWN(mp))
5971 		return -EIO;
5972 
5973 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL));
5974 
5975 	error = xfs_iread_extents(tp, ip, whichfork);
5976 	if (error)
5977 		return error;
5978 
5979 	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
5980 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5981 		cur->bc_ino.flags = 0;
5982 	}
5983 
5984 	if (*next_fsb == NULLFSBLOCK) {
5985 		xfs_iext_last(ifp, &icur);
5986 		if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5987 		    stop_fsb > got.br_startoff) {
5988 			*done = true;
5989 			goto del_cursor;
5990 		}
5991 	} else {
5992 		if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5993 			*done = true;
5994 			goto del_cursor;
5995 		}
5996 	}
5997 	if (XFS_IS_CORRUPT(mp, isnullstartblock(got.br_startblock))) {
5998 		error = -EFSCORRUPTED;
5999 		goto del_cursor;
6000 	}
6001 
6002 	if (XFS_IS_CORRUPT(mp, stop_fsb > got.br_startoff)) {
6003 		error = -EFSCORRUPTED;
6004 		goto del_cursor;
6005 	}
6006 
6007 	new_startoff = got.br_startoff + offset_shift_fsb;
6008 	if (xfs_iext_peek_next_extent(ifp, &icur, &next)) {
6009 		if (new_startoff + got.br_blockcount > next.br_startoff) {
6010 			error = -EINVAL;
6011 			goto del_cursor;
6012 		}
6013 
6014 		/*
6015 		 * Unlike a left shift (which involves a hole punch), a right
6016 		 * shift does not modify extent neighbors in any way.  We should
6017 		 * never find mergeable extents in this scenario.  Check anyways
6018 		 * and warn if we encounter two extents that could be one.
6019 		 */
6020 		if (xfs_bmse_can_merge(&got, &next, offset_shift_fsb))
6021 			WARN_ON_ONCE(1);
6022 	}
6023 
6024 	error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got,
6025 			cur, &logflags, new_startoff);
6026 	if (error)
6027 		goto del_cursor;
6028 
6029 	if (!xfs_iext_prev_extent(ifp, &icur, &got) ||
6030 	    stop_fsb >= got.br_startoff + got.br_blockcount) {
6031 		*done = true;
6032 		goto del_cursor;
6033 	}
6034 
6035 	*next_fsb = got.br_startoff;
6036 del_cursor:
6037 	if (cur)
6038 		xfs_btree_del_cursor(cur, error);
6039 	if (logflags)
6040 		xfs_trans_log_inode(tp, ip, logflags);
6041 	return error;
6042 }
6043 
6044 /*
6045  * Splits an extent into two extents at split_fsb block such that it is the
6046  * first block of the current_ext. @ext is a target extent to be split.
6047  * @split_fsb is a block where the extents is split.  If split_fsb lies in a
6048  * hole or the first block of extents, just return 0.
6049  */
6050 int
6051 xfs_bmap_split_extent(
6052 	struct xfs_trans	*tp,
6053 	struct xfs_inode	*ip,
6054 	xfs_fileoff_t		split_fsb)
6055 {
6056 	int				whichfork = XFS_DATA_FORK;
6057 	struct xfs_ifork		*ifp = XFS_IFORK_PTR(ip, whichfork);
6058 	struct xfs_btree_cur		*cur = NULL;
6059 	struct xfs_bmbt_irec		got;
6060 	struct xfs_bmbt_irec		new; /* split extent */
6061 	struct xfs_mount		*mp = ip->i_mount;
6062 	xfs_fsblock_t			gotblkcnt; /* new block count for got */
6063 	struct xfs_iext_cursor		icur;
6064 	int				error = 0;
6065 	int				logflags = 0;
6066 	int				i = 0;
6067 
6068 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
6069 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
6070 		return -EFSCORRUPTED;
6071 	}
6072 
6073 	if (XFS_FORCED_SHUTDOWN(mp))
6074 		return -EIO;
6075 
6076 	/* Read in all the extents */
6077 	error = xfs_iread_extents(tp, ip, whichfork);
6078 	if (error)
6079 		return error;
6080 
6081 	/*
6082 	 * If there are not extents, or split_fsb lies in a hole we are done.
6083 	 */
6084 	if (!xfs_iext_lookup_extent(ip, ifp, split_fsb, &icur, &got) ||
6085 	    got.br_startoff >= split_fsb)
6086 		return 0;
6087 
6088 	gotblkcnt = split_fsb - got.br_startoff;
6089 	new.br_startoff = split_fsb;
6090 	new.br_startblock = got.br_startblock + gotblkcnt;
6091 	new.br_blockcount = got.br_blockcount - gotblkcnt;
6092 	new.br_state = got.br_state;
6093 
6094 	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
6095 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6096 		cur->bc_ino.flags = 0;
6097 		error = xfs_bmbt_lookup_eq(cur, &got, &i);
6098 		if (error)
6099 			goto del_cursor;
6100 		if (XFS_IS_CORRUPT(mp, i != 1)) {
6101 			error = -EFSCORRUPTED;
6102 			goto del_cursor;
6103 		}
6104 	}
6105 
6106 	got.br_blockcount = gotblkcnt;
6107 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), &icur,
6108 			&got);
6109 
6110 	logflags = XFS_ILOG_CORE;
6111 	if (cur) {
6112 		error = xfs_bmbt_update(cur, &got);
6113 		if (error)
6114 			goto del_cursor;
6115 	} else
6116 		logflags |= XFS_ILOG_DEXT;
6117 
6118 	/* Add new extent */
6119 	xfs_iext_next(ifp, &icur);
6120 	xfs_iext_insert(ip, &icur, &new, 0);
6121 	ifp->if_nextents++;
6122 
6123 	if (cur) {
6124 		error = xfs_bmbt_lookup_eq(cur, &new, &i);
6125 		if (error)
6126 			goto del_cursor;
6127 		if (XFS_IS_CORRUPT(mp, i != 0)) {
6128 			error = -EFSCORRUPTED;
6129 			goto del_cursor;
6130 		}
6131 		error = xfs_btree_insert(cur, &i);
6132 		if (error)
6133 			goto del_cursor;
6134 		if (XFS_IS_CORRUPT(mp, i != 1)) {
6135 			error = -EFSCORRUPTED;
6136 			goto del_cursor;
6137 		}
6138 	}
6139 
6140 	/*
6141 	 * Convert to a btree if necessary.
6142 	 */
6143 	if (xfs_bmap_needs_btree(ip, whichfork)) {
6144 		int tmp_logflags; /* partial log flag return val */
6145 
6146 		ASSERT(cur == NULL);
6147 		error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
6148 				&tmp_logflags, whichfork);
6149 		logflags |= tmp_logflags;
6150 	}
6151 
6152 del_cursor:
6153 	if (cur) {
6154 		cur->bc_ino.allocated = 0;
6155 		xfs_btree_del_cursor(cur, error);
6156 	}
6157 
6158 	if (logflags)
6159 		xfs_trans_log_inode(tp, ip, logflags);
6160 	return error;
6161 }
6162 
6163 /* Deferred mapping is only for real extents in the data fork. */
6164 static bool
6165 xfs_bmap_is_update_needed(
6166 	struct xfs_bmbt_irec	*bmap)
6167 {
6168 	return  bmap->br_startblock != HOLESTARTBLOCK &&
6169 		bmap->br_startblock != DELAYSTARTBLOCK;
6170 }
6171 
6172 /* Record a bmap intent. */
6173 static int
6174 __xfs_bmap_add(
6175 	struct xfs_trans		*tp,
6176 	enum xfs_bmap_intent_type	type,
6177 	struct xfs_inode		*ip,
6178 	int				whichfork,
6179 	struct xfs_bmbt_irec		*bmap)
6180 {
6181 	struct xfs_bmap_intent		*bi;
6182 
6183 	trace_xfs_bmap_defer(tp->t_mountp,
6184 			XFS_FSB_TO_AGNO(tp->t_mountp, bmap->br_startblock),
6185 			type,
6186 			XFS_FSB_TO_AGBNO(tp->t_mountp, bmap->br_startblock),
6187 			ip->i_ino, whichfork,
6188 			bmap->br_startoff,
6189 			bmap->br_blockcount,
6190 			bmap->br_state);
6191 
6192 	bi = kmem_alloc(sizeof(struct xfs_bmap_intent), KM_NOFS);
6193 	INIT_LIST_HEAD(&bi->bi_list);
6194 	bi->bi_type = type;
6195 	bi->bi_owner = ip;
6196 	bi->bi_whichfork = whichfork;
6197 	bi->bi_bmap = *bmap;
6198 
6199 	xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list);
6200 	return 0;
6201 }
6202 
6203 /* Map an extent into a file. */
6204 void
6205 xfs_bmap_map_extent(
6206 	struct xfs_trans	*tp,
6207 	struct xfs_inode	*ip,
6208 	struct xfs_bmbt_irec	*PREV)
6209 {
6210 	if (!xfs_bmap_is_update_needed(PREV))
6211 		return;
6212 
6213 	__xfs_bmap_add(tp, XFS_BMAP_MAP, ip, XFS_DATA_FORK, PREV);
6214 }
6215 
6216 /* Unmap an extent out of a file. */
6217 void
6218 xfs_bmap_unmap_extent(
6219 	struct xfs_trans	*tp,
6220 	struct xfs_inode	*ip,
6221 	struct xfs_bmbt_irec	*PREV)
6222 {
6223 	if (!xfs_bmap_is_update_needed(PREV))
6224 		return;
6225 
6226 	__xfs_bmap_add(tp, XFS_BMAP_UNMAP, ip, XFS_DATA_FORK, PREV);
6227 }
6228 
6229 /*
6230  * Process one of the deferred bmap operations.  We pass back the
6231  * btree cursor to maintain our lock on the bmapbt between calls.
6232  */
6233 int
6234 xfs_bmap_finish_one(
6235 	struct xfs_trans		*tp,
6236 	struct xfs_inode		*ip,
6237 	enum xfs_bmap_intent_type	type,
6238 	int				whichfork,
6239 	xfs_fileoff_t			startoff,
6240 	xfs_fsblock_t			startblock,
6241 	xfs_filblks_t			*blockcount,
6242 	xfs_exntst_t			state)
6243 {
6244 	int				error = 0;
6245 
6246 	ASSERT(tp->t_firstblock == NULLFSBLOCK);
6247 
6248 	trace_xfs_bmap_deferred(tp->t_mountp,
6249 			XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type,
6250 			XFS_FSB_TO_AGBNO(tp->t_mountp, startblock),
6251 			ip->i_ino, whichfork, startoff, *blockcount, state);
6252 
6253 	if (WARN_ON_ONCE(whichfork != XFS_DATA_FORK))
6254 		return -EFSCORRUPTED;
6255 
6256 	if (XFS_TEST_ERROR(false, tp->t_mountp,
6257 			XFS_ERRTAG_BMAP_FINISH_ONE))
6258 		return -EIO;
6259 
6260 	switch (type) {
6261 	case XFS_BMAP_MAP:
6262 		error = xfs_bmapi_remap(tp, ip, startoff, *blockcount,
6263 				startblock, 0);
6264 		*blockcount = 0;
6265 		break;
6266 	case XFS_BMAP_UNMAP:
6267 		error = __xfs_bunmapi(tp, ip, startoff, blockcount,
6268 				XFS_BMAPI_REMAP, 1);
6269 		break;
6270 	default:
6271 		ASSERT(0);
6272 		error = -EFSCORRUPTED;
6273 	}
6274 
6275 	return error;
6276 }
6277 
6278 /* Check that an inode's extent does not have invalid flags or bad ranges. */
6279 xfs_failaddr_t
6280 xfs_bmap_validate_extent(
6281 	struct xfs_inode	*ip,
6282 	int			whichfork,
6283 	struct xfs_bmbt_irec	*irec)
6284 {
6285 	struct xfs_mount	*mp = ip->i_mount;
6286 
6287 	if (!xfs_verify_fileext(mp, irec->br_startoff, irec->br_blockcount))
6288 		return __this_address;
6289 
6290 	if (XFS_IS_REALTIME_INODE(ip) && whichfork == XFS_DATA_FORK) {
6291 		if (!xfs_verify_rtext(mp, irec->br_startblock,
6292 					  irec->br_blockcount))
6293 			return __this_address;
6294 	} else {
6295 		if (!xfs_verify_fsbext(mp, irec->br_startblock,
6296 					   irec->br_blockcount))
6297 			return __this_address;
6298 	}
6299 	if (irec->br_state != XFS_EXT_NORM && whichfork != XFS_DATA_FORK)
6300 		return __this_address;
6301 	return NULL;
6302 }
6303