xref: /openbmc/linux/fs/xfs/libxfs/xfs_bmap.c (revision 264e3509)
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_rtbitmap.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 struct kmem_cache		*xfs_bmap_intent_cache;
41 
42 /*
43  * Miscellaneous helper functions
44  */
45 
46 /*
47  * Compute and fill in the value of the maximum depth of a bmap btree
48  * in this filesystem.  Done once, during mount.
49  */
50 void
51 xfs_bmap_compute_maxlevels(
52 	xfs_mount_t	*mp,		/* file system mount structure */
53 	int		whichfork)	/* data or attr fork */
54 {
55 	uint64_t	maxblocks;	/* max blocks at this level */
56 	xfs_extnum_t	maxleafents;	/* max leaf entries possible */
57 	int		level;		/* btree level */
58 	int		maxrootrecs;	/* max records in root block */
59 	int		minleafrecs;	/* min records in leaf block */
60 	int		minnoderecs;	/* min records in node block */
61 	int		sz;		/* root block size */
62 
63 	/*
64 	 * The maximum number of extents in a fork, hence the maximum number of
65 	 * leaf entries, is controlled by the size of the on-disk extent count.
66 	 *
67 	 * Note that we can no longer assume that if we are in ATTR1 that the
68 	 * fork offset of all the inodes will be
69 	 * (xfs_default_attroffset(ip) >> 3) because we could have mounted with
70 	 * ATTR2 and then mounted back with ATTR1, keeping the i_forkoff's fixed
71 	 * but probably at various positions. Therefore, for both ATTR1 and
72 	 * ATTR2 we have to assume the worst case scenario of a minimum size
73 	 * available.
74 	 */
75 	maxleafents = xfs_iext_max_nextents(xfs_has_large_extent_counts(mp),
76 				whichfork);
77 	if (whichfork == XFS_DATA_FORK)
78 		sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
79 	else
80 		sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
81 
82 	maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
83 	minleafrecs = mp->m_bmap_dmnr[0];
84 	minnoderecs = mp->m_bmap_dmnr[1];
85 	maxblocks = howmany_64(maxleafents, minleafrecs);
86 	for (level = 1; maxblocks > 1; level++) {
87 		if (maxblocks <= maxrootrecs)
88 			maxblocks = 1;
89 		else
90 			maxblocks = howmany_64(maxblocks, minnoderecs);
91 	}
92 	mp->m_bm_maxlevels[whichfork] = level;
93 	ASSERT(mp->m_bm_maxlevels[whichfork] <= xfs_bmbt_maxlevels_ondisk());
94 }
95 
96 unsigned int
97 xfs_bmap_compute_attr_offset(
98 	struct xfs_mount	*mp)
99 {
100 	if (mp->m_sb.sb_inodesize == 256)
101 		return XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
102 	return XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
103 }
104 
105 STATIC int				/* error */
106 xfs_bmbt_lookup_eq(
107 	struct xfs_btree_cur	*cur,
108 	struct xfs_bmbt_irec	*irec,
109 	int			*stat)	/* success/failure */
110 {
111 	cur->bc_rec.b = *irec;
112 	return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
113 }
114 
115 STATIC int				/* error */
116 xfs_bmbt_lookup_first(
117 	struct xfs_btree_cur	*cur,
118 	int			*stat)	/* success/failure */
119 {
120 	cur->bc_rec.b.br_startoff = 0;
121 	cur->bc_rec.b.br_startblock = 0;
122 	cur->bc_rec.b.br_blockcount = 0;
123 	return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
124 }
125 
126 /*
127  * Check if the inode needs to be converted to btree format.
128  */
129 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
130 {
131 	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
132 
133 	return whichfork != XFS_COW_FORK &&
134 		ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
135 		ifp->if_nextents > XFS_IFORK_MAXEXT(ip, whichfork);
136 }
137 
138 /*
139  * Check if the inode should be converted to extent format.
140  */
141 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
142 {
143 	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
144 
145 	return whichfork != XFS_COW_FORK &&
146 		ifp->if_format == XFS_DINODE_FMT_BTREE &&
147 		ifp->if_nextents <= XFS_IFORK_MAXEXT(ip, whichfork);
148 }
149 
150 /*
151  * Update the record referred to by cur to the value given by irec
152  * This either works (return 0) or gets an EFSCORRUPTED error.
153  */
154 STATIC int
155 xfs_bmbt_update(
156 	struct xfs_btree_cur	*cur,
157 	struct xfs_bmbt_irec	*irec)
158 {
159 	union xfs_btree_rec	rec;
160 
161 	xfs_bmbt_disk_set_all(&rec.bmbt, irec);
162 	return xfs_btree_update(cur, &rec);
163 }
164 
165 /*
166  * Compute the worst-case number of indirect blocks that will be used
167  * for ip's delayed extent of length "len".
168  */
169 STATIC xfs_filblks_t
170 xfs_bmap_worst_indlen(
171 	xfs_inode_t	*ip,		/* incore inode pointer */
172 	xfs_filblks_t	len)		/* delayed extent length */
173 {
174 	int		level;		/* btree level number */
175 	int		maxrecs;	/* maximum record count at this level */
176 	xfs_mount_t	*mp;		/* mount structure */
177 	xfs_filblks_t	rval;		/* return value */
178 
179 	mp = ip->i_mount;
180 	maxrecs = mp->m_bmap_dmxr[0];
181 	for (level = 0, rval = 0;
182 	     level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
183 	     level++) {
184 		len += maxrecs - 1;
185 		do_div(len, maxrecs);
186 		rval += len;
187 		if (len == 1)
188 			return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
189 				level - 1;
190 		if (level == 0)
191 			maxrecs = mp->m_bmap_dmxr[1];
192 	}
193 	return rval;
194 }
195 
196 /*
197  * Calculate the default attribute fork offset for newly created inodes.
198  */
199 uint
200 xfs_default_attroffset(
201 	struct xfs_inode	*ip)
202 {
203 	if (ip->i_df.if_format == XFS_DINODE_FMT_DEV)
204 		return roundup(sizeof(xfs_dev_t), 8);
205 	return M_IGEO(ip->i_mount)->attr_fork_offset;
206 }
207 
208 /*
209  * Helper routine to reset inode i_forkoff field when switching attribute fork
210  * from local to extent format - we reset it where possible to make space
211  * available for inline data fork extents.
212  */
213 STATIC void
214 xfs_bmap_forkoff_reset(
215 	xfs_inode_t	*ip,
216 	int		whichfork)
217 {
218 	if (whichfork == XFS_ATTR_FORK &&
219 	    ip->i_df.if_format != XFS_DINODE_FMT_DEV &&
220 	    ip->i_df.if_format != XFS_DINODE_FMT_BTREE) {
221 		uint	dfl_forkoff = xfs_default_attroffset(ip) >> 3;
222 
223 		if (dfl_forkoff > ip->i_forkoff)
224 			ip->i_forkoff = dfl_forkoff;
225 	}
226 }
227 
228 #ifdef DEBUG
229 STATIC struct xfs_buf *
230 xfs_bmap_get_bp(
231 	struct xfs_btree_cur	*cur,
232 	xfs_fsblock_t		bno)
233 {
234 	struct xfs_log_item	*lip;
235 	int			i;
236 
237 	if (!cur)
238 		return NULL;
239 
240 	for (i = 0; i < cur->bc_maxlevels; i++) {
241 		if (!cur->bc_levels[i].bp)
242 			break;
243 		if (xfs_buf_daddr(cur->bc_levels[i].bp) == bno)
244 			return cur->bc_levels[i].bp;
245 	}
246 
247 	/* Chase down all the log items to see if the bp is there */
248 	list_for_each_entry(lip, &cur->bc_tp->t_items, li_trans) {
249 		struct xfs_buf_log_item	*bip = (struct xfs_buf_log_item *)lip;
250 
251 		if (bip->bli_item.li_type == XFS_LI_BUF &&
252 		    xfs_buf_daddr(bip->bli_buf) == bno)
253 			return bip->bli_buf;
254 	}
255 
256 	return NULL;
257 }
258 
259 STATIC void
260 xfs_check_block(
261 	struct xfs_btree_block	*block,
262 	xfs_mount_t		*mp,
263 	int			root,
264 	short			sz)
265 {
266 	int			i, j, dmxr;
267 	__be64			*pp, *thispa;	/* pointer to block address */
268 	xfs_bmbt_key_t		*prevp, *keyp;
269 
270 	ASSERT(be16_to_cpu(block->bb_level) > 0);
271 
272 	prevp = NULL;
273 	for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
274 		dmxr = mp->m_bmap_dmxr[0];
275 		keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
276 
277 		if (prevp) {
278 			ASSERT(be64_to_cpu(prevp->br_startoff) <
279 			       be64_to_cpu(keyp->br_startoff));
280 		}
281 		prevp = keyp;
282 
283 		/*
284 		 * Compare the block numbers to see if there are dups.
285 		 */
286 		if (root)
287 			pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
288 		else
289 			pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
290 
291 		for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
292 			if (root)
293 				thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
294 			else
295 				thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
296 			if (*thispa == *pp) {
297 				xfs_warn(mp, "%s: thispa(%d) == pp(%d) %lld",
298 					__func__, j, i,
299 					(unsigned long long)be64_to_cpu(*thispa));
300 				xfs_err(mp, "%s: ptrs are equal in node\n",
301 					__func__);
302 				xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
303 			}
304 		}
305 	}
306 }
307 
308 /*
309  * Check that the extents for the inode ip are in the right order in all
310  * btree leaves. THis becomes prohibitively expensive for large extent count
311  * files, so don't bother with inodes that have more than 10,000 extents in
312  * them. The btree record ordering checks will still be done, so for such large
313  * bmapbt constructs that is going to catch most corruptions.
314  */
315 STATIC void
316 xfs_bmap_check_leaf_extents(
317 	struct xfs_btree_cur	*cur,	/* btree cursor or null */
318 	xfs_inode_t		*ip,		/* incore inode pointer */
319 	int			whichfork)	/* data or attr fork */
320 {
321 	struct xfs_mount	*mp = ip->i_mount;
322 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
323 	struct xfs_btree_block	*block;	/* current btree block */
324 	xfs_fsblock_t		bno;	/* block # of "block" */
325 	struct xfs_buf		*bp;	/* buffer for "block" */
326 	int			error;	/* error return value */
327 	xfs_extnum_t		i=0, j;	/* index into the extents list */
328 	int			level;	/* btree level, for checking */
329 	__be64			*pp;	/* pointer to block address */
330 	xfs_bmbt_rec_t		*ep;	/* pointer to current extent */
331 	xfs_bmbt_rec_t		last = {0, 0}; /* last extent in prev block */
332 	xfs_bmbt_rec_t		*nextp;	/* pointer to next extent */
333 	int			bp_release = 0;
334 
335 	if (ifp->if_format != XFS_DINODE_FMT_BTREE)
336 		return;
337 
338 	/* skip large extent count inodes */
339 	if (ip->i_df.if_nextents > 10000)
340 		return;
341 
342 	bno = NULLFSBLOCK;
343 	block = ifp->if_broot;
344 	/*
345 	 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
346 	 */
347 	level = be16_to_cpu(block->bb_level);
348 	ASSERT(level > 0);
349 	xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
350 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
351 	bno = be64_to_cpu(*pp);
352 
353 	ASSERT(bno != NULLFSBLOCK);
354 	ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
355 	ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
356 
357 	/*
358 	 * Go down the tree until leaf level is reached, following the first
359 	 * pointer (leftmost) at each level.
360 	 */
361 	while (level-- > 0) {
362 		/* See if buf is in cur first */
363 		bp_release = 0;
364 		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
365 		if (!bp) {
366 			bp_release = 1;
367 			error = xfs_btree_read_bufl(mp, NULL, bno, &bp,
368 						XFS_BMAP_BTREE_REF,
369 						&xfs_bmbt_buf_ops);
370 			if (error)
371 				goto error_norelse;
372 		}
373 		block = XFS_BUF_TO_BLOCK(bp);
374 		if (level == 0)
375 			break;
376 
377 		/*
378 		 * Check this block for basic sanity (increasing keys and
379 		 * no duplicate blocks).
380 		 */
381 
382 		xfs_check_block(block, mp, 0, 0);
383 		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
384 		bno = be64_to_cpu(*pp);
385 		if (XFS_IS_CORRUPT(mp, !xfs_verify_fsbno(mp, bno))) {
386 			error = -EFSCORRUPTED;
387 			goto error0;
388 		}
389 		if (bp_release) {
390 			bp_release = 0;
391 			xfs_trans_brelse(NULL, bp);
392 		}
393 	}
394 
395 	/*
396 	 * Here with bp and block set to the leftmost leaf node in the tree.
397 	 */
398 	i = 0;
399 
400 	/*
401 	 * Loop over all leaf nodes checking that all extents are in the right order.
402 	 */
403 	for (;;) {
404 		xfs_fsblock_t	nextbno;
405 		xfs_extnum_t	num_recs;
406 
407 
408 		num_recs = xfs_btree_get_numrecs(block);
409 
410 		/*
411 		 * Read-ahead the next leaf block, if any.
412 		 */
413 
414 		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
415 
416 		/*
417 		 * Check all the extents to make sure they are OK.
418 		 * If we had a previous block, the last entry should
419 		 * conform with the first entry in this one.
420 		 */
421 
422 		ep = XFS_BMBT_REC_ADDR(mp, block, 1);
423 		if (i) {
424 			ASSERT(xfs_bmbt_disk_get_startoff(&last) +
425 			       xfs_bmbt_disk_get_blockcount(&last) <=
426 			       xfs_bmbt_disk_get_startoff(ep));
427 		}
428 		for (j = 1; j < num_recs; j++) {
429 			nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
430 			ASSERT(xfs_bmbt_disk_get_startoff(ep) +
431 			       xfs_bmbt_disk_get_blockcount(ep) <=
432 			       xfs_bmbt_disk_get_startoff(nextp));
433 			ep = nextp;
434 		}
435 
436 		last = *ep;
437 		i += num_recs;
438 		if (bp_release) {
439 			bp_release = 0;
440 			xfs_trans_brelse(NULL, bp);
441 		}
442 		bno = nextbno;
443 		/*
444 		 * If we've reached the end, stop.
445 		 */
446 		if (bno == NULLFSBLOCK)
447 			break;
448 
449 		bp_release = 0;
450 		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
451 		if (!bp) {
452 			bp_release = 1;
453 			error = xfs_btree_read_bufl(mp, NULL, bno, &bp,
454 						XFS_BMAP_BTREE_REF,
455 						&xfs_bmbt_buf_ops);
456 			if (error)
457 				goto error_norelse;
458 		}
459 		block = XFS_BUF_TO_BLOCK(bp);
460 	}
461 
462 	return;
463 
464 error0:
465 	xfs_warn(mp, "%s: at error0", __func__);
466 	if (bp_release)
467 		xfs_trans_brelse(NULL, bp);
468 error_norelse:
469 	xfs_warn(mp, "%s: BAD after btree leaves for %llu extents",
470 		__func__, i);
471 	xfs_err(mp, "%s: CORRUPTED BTREE OR SOMETHING", __func__);
472 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
473 	return;
474 }
475 
476 /*
477  * Validate that the bmbt_irecs being returned from bmapi are valid
478  * given the caller's original parameters.  Specifically check the
479  * ranges of the returned irecs to ensure that they only extend beyond
480  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
481  */
482 STATIC void
483 xfs_bmap_validate_ret(
484 	xfs_fileoff_t		bno,
485 	xfs_filblks_t		len,
486 	uint32_t		flags,
487 	xfs_bmbt_irec_t		*mval,
488 	int			nmap,
489 	int			ret_nmap)
490 {
491 	int			i;		/* index to map values */
492 
493 	ASSERT(ret_nmap <= nmap);
494 
495 	for (i = 0; i < ret_nmap; i++) {
496 		ASSERT(mval[i].br_blockcount > 0);
497 		if (!(flags & XFS_BMAPI_ENTIRE)) {
498 			ASSERT(mval[i].br_startoff >= bno);
499 			ASSERT(mval[i].br_blockcount <= len);
500 			ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
501 			       bno + len);
502 		} else {
503 			ASSERT(mval[i].br_startoff < bno + len);
504 			ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
505 			       bno);
506 		}
507 		ASSERT(i == 0 ||
508 		       mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
509 		       mval[i].br_startoff);
510 		ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
511 		       mval[i].br_startblock != HOLESTARTBLOCK);
512 		ASSERT(mval[i].br_state == XFS_EXT_NORM ||
513 		       mval[i].br_state == XFS_EXT_UNWRITTEN);
514 	}
515 }
516 
517 #else
518 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)		do { } while (0)
519 #define	xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)	do { } while (0)
520 #endif /* DEBUG */
521 
522 /*
523  * Inode fork format manipulation functions
524  */
525 
526 /*
527  * Convert the inode format to extent format if it currently is in btree format,
528  * but the extent list is small enough that it fits into the extent format.
529  *
530  * Since the extents are already in-core, all we have to do is give up the space
531  * for the btree root and pitch the leaf block.
532  */
533 STATIC int				/* error */
534 xfs_bmap_btree_to_extents(
535 	struct xfs_trans	*tp,	/* transaction pointer */
536 	struct xfs_inode	*ip,	/* incore inode pointer */
537 	struct xfs_btree_cur	*cur,	/* btree cursor */
538 	int			*logflagsp, /* inode logging flags */
539 	int			whichfork)  /* data or attr fork */
540 {
541 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
542 	struct xfs_mount	*mp = ip->i_mount;
543 	struct xfs_btree_block	*rblock = ifp->if_broot;
544 	struct xfs_btree_block	*cblock;/* child btree block */
545 	xfs_fsblock_t		cbno;	/* child block number */
546 	struct xfs_buf		*cbp;	/* child block's buffer */
547 	int			error;	/* error return value */
548 	__be64			*pp;	/* ptr to block address */
549 	struct xfs_owner_info	oinfo;
550 
551 	/* check if we actually need the extent format first: */
552 	if (!xfs_bmap_wants_extents(ip, whichfork))
553 		return 0;
554 
555 	ASSERT(cur);
556 	ASSERT(whichfork != XFS_COW_FORK);
557 	ASSERT(ifp->if_format == XFS_DINODE_FMT_BTREE);
558 	ASSERT(be16_to_cpu(rblock->bb_level) == 1);
559 	ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
560 	ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
561 
562 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
563 	cbno = be64_to_cpu(*pp);
564 #ifdef DEBUG
565 	if (XFS_IS_CORRUPT(cur->bc_mp, !xfs_btree_check_lptr(cur, cbno, 1)))
566 		return -EFSCORRUPTED;
567 #endif
568 	error = xfs_btree_read_bufl(mp, tp, cbno, &cbp, XFS_BMAP_BTREE_REF,
569 				&xfs_bmbt_buf_ops);
570 	if (error)
571 		return error;
572 	cblock = XFS_BUF_TO_BLOCK(cbp);
573 	if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
574 		return error;
575 
576 	xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
577 	error = xfs_free_extent_later(cur->bc_tp, cbno, 1, &oinfo,
578 			XFS_AG_RESV_NONE);
579 	if (error)
580 		return error;
581 
582 	ip->i_nblocks--;
583 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
584 	xfs_trans_binval(tp, cbp);
585 	if (cur->bc_levels[0].bp == cbp)
586 		cur->bc_levels[0].bp = NULL;
587 	xfs_iroot_realloc(ip, -1, whichfork);
588 	ASSERT(ifp->if_broot == NULL);
589 	ifp->if_format = XFS_DINODE_FMT_EXTENTS;
590 	*logflagsp |= XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
591 	return 0;
592 }
593 
594 /*
595  * Convert an extents-format file into a btree-format file.
596  * The new file will have a root block (in the inode) and a single child block.
597  */
598 STATIC int					/* error */
599 xfs_bmap_extents_to_btree(
600 	struct xfs_trans	*tp,		/* transaction pointer */
601 	struct xfs_inode	*ip,		/* incore inode pointer */
602 	struct xfs_btree_cur	**curp,		/* cursor returned to caller */
603 	int			wasdel,		/* converting a delayed alloc */
604 	int			*logflagsp,	/* inode logging flags */
605 	int			whichfork)	/* data or attr fork */
606 {
607 	struct xfs_btree_block	*ablock;	/* allocated (child) bt block */
608 	struct xfs_buf		*abp;		/* buffer for ablock */
609 	struct xfs_alloc_arg	args;		/* allocation arguments */
610 	struct xfs_bmbt_rec	*arp;		/* child record pointer */
611 	struct xfs_btree_block	*block;		/* btree root block */
612 	struct xfs_btree_cur	*cur;		/* bmap btree cursor */
613 	int			error;		/* error return value */
614 	struct xfs_ifork	*ifp;		/* inode fork pointer */
615 	struct xfs_bmbt_key	*kp;		/* root block key pointer */
616 	struct xfs_mount	*mp;		/* mount structure */
617 	xfs_bmbt_ptr_t		*pp;		/* root block address pointer */
618 	struct xfs_iext_cursor	icur;
619 	struct xfs_bmbt_irec	rec;
620 	xfs_extnum_t		cnt = 0;
621 
622 	mp = ip->i_mount;
623 	ASSERT(whichfork != XFS_COW_FORK);
624 	ifp = xfs_ifork_ptr(ip, whichfork);
625 	ASSERT(ifp->if_format == XFS_DINODE_FMT_EXTENTS);
626 
627 	/*
628 	 * Make space in the inode incore. This needs to be undone if we fail
629 	 * to expand the root.
630 	 */
631 	xfs_iroot_realloc(ip, 1, whichfork);
632 
633 	/*
634 	 * Fill in the root.
635 	 */
636 	block = ifp->if_broot;
637 	xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
638 				 XFS_BTNUM_BMAP, 1, 1, ip->i_ino,
639 				 XFS_BTREE_LONG_PTRS);
640 	/*
641 	 * Need a cursor.  Can't allocate until bb_level is filled in.
642 	 */
643 	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
644 	cur->bc_ino.flags = wasdel ? XFS_BTCUR_BMBT_WASDEL : 0;
645 	/*
646 	 * Convert to a btree with two levels, one record in root.
647 	 */
648 	ifp->if_format = XFS_DINODE_FMT_BTREE;
649 	memset(&args, 0, sizeof(args));
650 	args.tp = tp;
651 	args.mp = mp;
652 	xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork);
653 
654 	args.minlen = args.maxlen = args.prod = 1;
655 	args.wasdel = wasdel;
656 	*logflagsp = 0;
657 	error = xfs_alloc_vextent_start_ag(&args,
658 				XFS_INO_TO_FSB(mp, ip->i_ino));
659 	if (error)
660 		goto out_root_realloc;
661 
662 	/*
663 	 * Allocation can't fail, the space was reserved.
664 	 */
665 	if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
666 		error = -ENOSPC;
667 		goto out_root_realloc;
668 	}
669 
670 	cur->bc_ino.allocated++;
671 	ip->i_nblocks++;
672 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
673 	error = xfs_trans_get_buf(tp, mp->m_ddev_targp,
674 			XFS_FSB_TO_DADDR(mp, args.fsbno),
675 			mp->m_bsize, 0, &abp);
676 	if (error)
677 		goto out_unreserve_dquot;
678 
679 	/*
680 	 * Fill in the child block.
681 	 */
682 	abp->b_ops = &xfs_bmbt_buf_ops;
683 	ablock = XFS_BUF_TO_BLOCK(abp);
684 	xfs_btree_init_block_int(mp, ablock, xfs_buf_daddr(abp),
685 				XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
686 				XFS_BTREE_LONG_PTRS);
687 
688 	for_each_xfs_iext(ifp, &icur, &rec) {
689 		if (isnullstartblock(rec.br_startblock))
690 			continue;
691 		arp = XFS_BMBT_REC_ADDR(mp, ablock, 1 + cnt);
692 		xfs_bmbt_disk_set_all(arp, &rec);
693 		cnt++;
694 	}
695 	ASSERT(cnt == ifp->if_nextents);
696 	xfs_btree_set_numrecs(ablock, cnt);
697 
698 	/*
699 	 * Fill in the root key and pointer.
700 	 */
701 	kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
702 	arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
703 	kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
704 	pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
705 						be16_to_cpu(block->bb_level)));
706 	*pp = cpu_to_be64(args.fsbno);
707 
708 	/*
709 	 * Do all this logging at the end so that
710 	 * the root is at the right level.
711 	 */
712 	xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
713 	xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
714 	ASSERT(*curp == NULL);
715 	*curp = cur;
716 	*logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
717 	return 0;
718 
719 out_unreserve_dquot:
720 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
721 out_root_realloc:
722 	xfs_iroot_realloc(ip, -1, whichfork);
723 	ifp->if_format = XFS_DINODE_FMT_EXTENTS;
724 	ASSERT(ifp->if_broot == NULL);
725 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
726 
727 	return error;
728 }
729 
730 /*
731  * Convert a local file to an extents file.
732  * This code is out of bounds for data forks of regular files,
733  * since the file data needs to get logged so things will stay consistent.
734  * (The bmap-level manipulations are ok, though).
735  */
736 void
737 xfs_bmap_local_to_extents_empty(
738 	struct xfs_trans	*tp,
739 	struct xfs_inode	*ip,
740 	int			whichfork)
741 {
742 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
743 
744 	ASSERT(whichfork != XFS_COW_FORK);
745 	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
746 	ASSERT(ifp->if_bytes == 0);
747 	ASSERT(ifp->if_nextents == 0);
748 
749 	xfs_bmap_forkoff_reset(ip, whichfork);
750 	ifp->if_u1.if_root = NULL;
751 	ifp->if_height = 0;
752 	ifp->if_format = XFS_DINODE_FMT_EXTENTS;
753 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
754 }
755 
756 
757 STATIC int				/* error */
758 xfs_bmap_local_to_extents(
759 	xfs_trans_t	*tp,		/* transaction pointer */
760 	xfs_inode_t	*ip,		/* incore inode pointer */
761 	xfs_extlen_t	total,		/* total blocks needed by transaction */
762 	int		*logflagsp,	/* inode logging flags */
763 	int		whichfork,
764 	void		(*init_fn)(struct xfs_trans *tp,
765 				   struct xfs_buf *bp,
766 				   struct xfs_inode *ip,
767 				   struct xfs_ifork *ifp))
768 {
769 	int		error = 0;
770 	int		flags;		/* logging flags returned */
771 	struct xfs_ifork *ifp;		/* inode fork pointer */
772 	xfs_alloc_arg_t	args;		/* allocation arguments */
773 	struct xfs_buf	*bp;		/* buffer for extent block */
774 	struct xfs_bmbt_irec rec;
775 	struct xfs_iext_cursor icur;
776 
777 	/*
778 	 * We don't want to deal with the case of keeping inode data inline yet.
779 	 * So sending the data fork of a regular inode is invalid.
780 	 */
781 	ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
782 	ifp = xfs_ifork_ptr(ip, whichfork);
783 	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
784 
785 	if (!ifp->if_bytes) {
786 		xfs_bmap_local_to_extents_empty(tp, ip, whichfork);
787 		flags = XFS_ILOG_CORE;
788 		goto done;
789 	}
790 
791 	flags = 0;
792 	error = 0;
793 	memset(&args, 0, sizeof(args));
794 	args.tp = tp;
795 	args.mp = ip->i_mount;
796 	args.total = total;
797 	args.minlen = args.maxlen = args.prod = 1;
798 	xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0);
799 
800 	/*
801 	 * Allocate a block.  We know we need only one, since the
802 	 * file currently fits in an inode.
803 	 */
804 	args.total = total;
805 	args.minlen = args.maxlen = args.prod = 1;
806 	error = xfs_alloc_vextent_start_ag(&args,
807 			XFS_INO_TO_FSB(args.mp, ip->i_ino));
808 	if (error)
809 		goto done;
810 
811 	/* Can't fail, the space was reserved. */
812 	ASSERT(args.fsbno != NULLFSBLOCK);
813 	ASSERT(args.len == 1);
814 	error = xfs_trans_get_buf(tp, args.mp->m_ddev_targp,
815 			XFS_FSB_TO_DADDR(args.mp, args.fsbno),
816 			args.mp->m_bsize, 0, &bp);
817 	if (error)
818 		goto done;
819 
820 	/*
821 	 * Initialize the block, copy the data and log the remote buffer.
822 	 *
823 	 * The callout is responsible for logging because the remote format
824 	 * might differ from the local format and thus we don't know how much to
825 	 * log here. Note that init_fn must also set the buffer log item type
826 	 * correctly.
827 	 */
828 	init_fn(tp, bp, ip, ifp);
829 
830 	/* account for the change in fork size */
831 	xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
832 	xfs_bmap_local_to_extents_empty(tp, ip, whichfork);
833 	flags |= XFS_ILOG_CORE;
834 
835 	ifp->if_u1.if_root = NULL;
836 	ifp->if_height = 0;
837 
838 	rec.br_startoff = 0;
839 	rec.br_startblock = args.fsbno;
840 	rec.br_blockcount = 1;
841 	rec.br_state = XFS_EXT_NORM;
842 	xfs_iext_first(ifp, &icur);
843 	xfs_iext_insert(ip, &icur, &rec, 0);
844 
845 	ifp->if_nextents = 1;
846 	ip->i_nblocks = 1;
847 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
848 	flags |= xfs_ilog_fext(whichfork);
849 
850 done:
851 	*logflagsp = flags;
852 	return error;
853 }
854 
855 /*
856  * Called from xfs_bmap_add_attrfork to handle btree format files.
857  */
858 STATIC int					/* error */
859 xfs_bmap_add_attrfork_btree(
860 	xfs_trans_t		*tp,		/* transaction pointer */
861 	xfs_inode_t		*ip,		/* incore inode pointer */
862 	int			*flags)		/* inode logging flags */
863 {
864 	struct xfs_btree_block	*block = ip->i_df.if_broot;
865 	struct xfs_btree_cur	*cur;		/* btree cursor */
866 	int			error;		/* error return value */
867 	xfs_mount_t		*mp;		/* file system mount struct */
868 	int			stat;		/* newroot status */
869 
870 	mp = ip->i_mount;
871 
872 	if (XFS_BMAP_BMDR_SPACE(block) <= xfs_inode_data_fork_size(ip))
873 		*flags |= XFS_ILOG_DBROOT;
874 	else {
875 		cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
876 		error = xfs_bmbt_lookup_first(cur, &stat);
877 		if (error)
878 			goto error0;
879 		/* must be at least one entry */
880 		if (XFS_IS_CORRUPT(mp, stat != 1)) {
881 			error = -EFSCORRUPTED;
882 			goto error0;
883 		}
884 		if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
885 			goto error0;
886 		if (stat == 0) {
887 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
888 			return -ENOSPC;
889 		}
890 		cur->bc_ino.allocated = 0;
891 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
892 	}
893 	return 0;
894 error0:
895 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
896 	return error;
897 }
898 
899 /*
900  * Called from xfs_bmap_add_attrfork to handle extents format files.
901  */
902 STATIC int					/* error */
903 xfs_bmap_add_attrfork_extents(
904 	struct xfs_trans	*tp,		/* transaction pointer */
905 	struct xfs_inode	*ip,		/* incore inode pointer */
906 	int			*flags)		/* inode logging flags */
907 {
908 	struct xfs_btree_cur	*cur;		/* bmap btree cursor */
909 	int			error;		/* error return value */
910 
911 	if (ip->i_df.if_nextents * sizeof(struct xfs_bmbt_rec) <=
912 	    xfs_inode_data_fork_size(ip))
913 		return 0;
914 	cur = NULL;
915 	error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, flags,
916 					  XFS_DATA_FORK);
917 	if (cur) {
918 		cur->bc_ino.allocated = 0;
919 		xfs_btree_del_cursor(cur, error);
920 	}
921 	return error;
922 }
923 
924 /*
925  * Called from xfs_bmap_add_attrfork to handle local format files. Each
926  * different data fork content type needs a different callout to do the
927  * conversion. Some are basic and only require special block initialisation
928  * callouts for the data formating, others (directories) are so specialised they
929  * handle everything themselves.
930  *
931  * XXX (dgc): investigate whether directory conversion can use the generic
932  * formatting callout. It should be possible - it's just a very complex
933  * formatter.
934  */
935 STATIC int					/* error */
936 xfs_bmap_add_attrfork_local(
937 	struct xfs_trans	*tp,		/* transaction pointer */
938 	struct xfs_inode	*ip,		/* incore inode pointer */
939 	int			*flags)		/* inode logging flags */
940 {
941 	struct xfs_da_args	dargs;		/* args for dir/attr code */
942 
943 	if (ip->i_df.if_bytes <= xfs_inode_data_fork_size(ip))
944 		return 0;
945 
946 	if (S_ISDIR(VFS_I(ip)->i_mode)) {
947 		memset(&dargs, 0, sizeof(dargs));
948 		dargs.geo = ip->i_mount->m_dir_geo;
949 		dargs.dp = ip;
950 		dargs.total = dargs.geo->fsbcount;
951 		dargs.whichfork = XFS_DATA_FORK;
952 		dargs.trans = tp;
953 		return xfs_dir2_sf_to_block(&dargs);
954 	}
955 
956 	if (S_ISLNK(VFS_I(ip)->i_mode))
957 		return xfs_bmap_local_to_extents(tp, ip, 1, flags,
958 						 XFS_DATA_FORK,
959 						 xfs_symlink_local_to_remote);
960 
961 	/* should only be called for types that support local format data */
962 	ASSERT(0);
963 	return -EFSCORRUPTED;
964 }
965 
966 /*
967  * Set an inode attr fork offset based on the format of the data fork.
968  */
969 static int
970 xfs_bmap_set_attrforkoff(
971 	struct xfs_inode	*ip,
972 	int			size,
973 	int			*version)
974 {
975 	int			default_size = xfs_default_attroffset(ip) >> 3;
976 
977 	switch (ip->i_df.if_format) {
978 	case XFS_DINODE_FMT_DEV:
979 		ip->i_forkoff = default_size;
980 		break;
981 	case XFS_DINODE_FMT_LOCAL:
982 	case XFS_DINODE_FMT_EXTENTS:
983 	case XFS_DINODE_FMT_BTREE:
984 		ip->i_forkoff = xfs_attr_shortform_bytesfit(ip, size);
985 		if (!ip->i_forkoff)
986 			ip->i_forkoff = default_size;
987 		else if (xfs_has_attr2(ip->i_mount) && version)
988 			*version = 2;
989 		break;
990 	default:
991 		ASSERT(0);
992 		return -EINVAL;
993 	}
994 
995 	return 0;
996 }
997 
998 /*
999  * Convert inode from non-attributed to attributed.
1000  * Must not be in a transaction, ip must not be locked.
1001  */
1002 int						/* error code */
1003 xfs_bmap_add_attrfork(
1004 	xfs_inode_t		*ip,		/* incore inode pointer */
1005 	int			size,		/* space new attribute needs */
1006 	int			rsvd)		/* xact may use reserved blks */
1007 {
1008 	xfs_mount_t		*mp;		/* mount structure */
1009 	xfs_trans_t		*tp;		/* transaction pointer */
1010 	int			blks;		/* space reservation */
1011 	int			version = 1;	/* superblock attr version */
1012 	int			logflags;	/* logging flags */
1013 	int			error;		/* error return value */
1014 
1015 	ASSERT(xfs_inode_has_attr_fork(ip) == 0);
1016 
1017 	mp = ip->i_mount;
1018 	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1019 
1020 	blks = XFS_ADDAFORK_SPACE_RES(mp);
1021 
1022 	error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_addafork, blks, 0,
1023 			rsvd, &tp);
1024 	if (error)
1025 		return error;
1026 	if (xfs_inode_has_attr_fork(ip))
1027 		goto trans_cancel;
1028 
1029 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1030 	error = xfs_bmap_set_attrforkoff(ip, size, &version);
1031 	if (error)
1032 		goto trans_cancel;
1033 
1034 	xfs_ifork_init_attr(ip, XFS_DINODE_FMT_EXTENTS, 0);
1035 	logflags = 0;
1036 	switch (ip->i_df.if_format) {
1037 	case XFS_DINODE_FMT_LOCAL:
1038 		error = xfs_bmap_add_attrfork_local(tp, ip, &logflags);
1039 		break;
1040 	case XFS_DINODE_FMT_EXTENTS:
1041 		error = xfs_bmap_add_attrfork_extents(tp, ip, &logflags);
1042 		break;
1043 	case XFS_DINODE_FMT_BTREE:
1044 		error = xfs_bmap_add_attrfork_btree(tp, ip, &logflags);
1045 		break;
1046 	default:
1047 		error = 0;
1048 		break;
1049 	}
1050 	if (logflags)
1051 		xfs_trans_log_inode(tp, ip, logflags);
1052 	if (error)
1053 		goto trans_cancel;
1054 	if (!xfs_has_attr(mp) ||
1055 	   (!xfs_has_attr2(mp) && version == 2)) {
1056 		bool log_sb = false;
1057 
1058 		spin_lock(&mp->m_sb_lock);
1059 		if (!xfs_has_attr(mp)) {
1060 			xfs_add_attr(mp);
1061 			log_sb = true;
1062 		}
1063 		if (!xfs_has_attr2(mp) && version == 2) {
1064 			xfs_add_attr2(mp);
1065 			log_sb = true;
1066 		}
1067 		spin_unlock(&mp->m_sb_lock);
1068 		if (log_sb)
1069 			xfs_log_sb(tp);
1070 	}
1071 
1072 	error = xfs_trans_commit(tp);
1073 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1074 	return error;
1075 
1076 trans_cancel:
1077 	xfs_trans_cancel(tp);
1078 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1079 	return error;
1080 }
1081 
1082 /*
1083  * Internal and external extent tree search functions.
1084  */
1085 
1086 struct xfs_iread_state {
1087 	struct xfs_iext_cursor	icur;
1088 	xfs_extnum_t		loaded;
1089 };
1090 
1091 int
1092 xfs_bmap_complain_bad_rec(
1093 	struct xfs_inode		*ip,
1094 	int				whichfork,
1095 	xfs_failaddr_t			fa,
1096 	const struct xfs_bmbt_irec	*irec)
1097 {
1098 	struct xfs_mount		*mp = ip->i_mount;
1099 	const char			*forkname;
1100 
1101 	switch (whichfork) {
1102 	case XFS_DATA_FORK:	forkname = "data"; break;
1103 	case XFS_ATTR_FORK:	forkname = "attr"; break;
1104 	case XFS_COW_FORK:	forkname = "CoW"; break;
1105 	default:		forkname = "???"; break;
1106 	}
1107 
1108 	xfs_warn(mp,
1109  "Bmap BTree record corruption in inode 0x%llx %s fork detected at %pS!",
1110 				ip->i_ino, forkname, fa);
1111 	xfs_warn(mp,
1112 		"Offset 0x%llx, start block 0x%llx, block count 0x%llx state 0x%x",
1113 		irec->br_startoff, irec->br_startblock, irec->br_blockcount,
1114 		irec->br_state);
1115 
1116 	return -EFSCORRUPTED;
1117 }
1118 
1119 /* Stuff every bmbt record from this block into the incore extent map. */
1120 static int
1121 xfs_iread_bmbt_block(
1122 	struct xfs_btree_cur	*cur,
1123 	int			level,
1124 	void			*priv)
1125 {
1126 	struct xfs_iread_state	*ir = priv;
1127 	struct xfs_mount	*mp = cur->bc_mp;
1128 	struct xfs_inode	*ip = cur->bc_ino.ip;
1129 	struct xfs_btree_block	*block;
1130 	struct xfs_buf		*bp;
1131 	struct xfs_bmbt_rec	*frp;
1132 	xfs_extnum_t		num_recs;
1133 	xfs_extnum_t		j;
1134 	int			whichfork = cur->bc_ino.whichfork;
1135 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
1136 
1137 	block = xfs_btree_get_block(cur, level, &bp);
1138 
1139 	/* Abort if we find more records than nextents. */
1140 	num_recs = xfs_btree_get_numrecs(block);
1141 	if (unlikely(ir->loaded + num_recs > ifp->if_nextents)) {
1142 		xfs_warn(ip->i_mount, "corrupt dinode %llu, (btree extents).",
1143 				(unsigned long long)ip->i_ino);
1144 		xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, block,
1145 				sizeof(*block), __this_address);
1146 		return -EFSCORRUPTED;
1147 	}
1148 
1149 	/* Copy records into the incore cache. */
1150 	frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1151 	for (j = 0; j < num_recs; j++, frp++, ir->loaded++) {
1152 		struct xfs_bmbt_irec	new;
1153 		xfs_failaddr_t		fa;
1154 
1155 		xfs_bmbt_disk_get_all(frp, &new);
1156 		fa = xfs_bmap_validate_extent(ip, whichfork, &new);
1157 		if (fa) {
1158 			xfs_inode_verifier_error(ip, -EFSCORRUPTED,
1159 					"xfs_iread_extents(2)", frp,
1160 					sizeof(*frp), fa);
1161 			return xfs_bmap_complain_bad_rec(ip, whichfork, fa,
1162 					&new);
1163 		}
1164 		xfs_iext_insert(ip, &ir->icur, &new,
1165 				xfs_bmap_fork_to_state(whichfork));
1166 		trace_xfs_read_extent(ip, &ir->icur,
1167 				xfs_bmap_fork_to_state(whichfork), _THIS_IP_);
1168 		xfs_iext_next(ifp, &ir->icur);
1169 	}
1170 
1171 	return 0;
1172 }
1173 
1174 /*
1175  * Read in extents from a btree-format inode.
1176  */
1177 int
1178 xfs_iread_extents(
1179 	struct xfs_trans	*tp,
1180 	struct xfs_inode	*ip,
1181 	int			whichfork)
1182 {
1183 	struct xfs_iread_state	ir;
1184 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
1185 	struct xfs_mount	*mp = ip->i_mount;
1186 	struct xfs_btree_cur	*cur;
1187 	int			error;
1188 
1189 	if (!xfs_need_iread_extents(ifp))
1190 		return 0;
1191 
1192 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1193 
1194 	ir.loaded = 0;
1195 	xfs_iext_first(ifp, &ir.icur);
1196 	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
1197 	error = xfs_btree_visit_blocks(cur, xfs_iread_bmbt_block,
1198 			XFS_BTREE_VISIT_RECORDS, &ir);
1199 	xfs_btree_del_cursor(cur, error);
1200 	if (error)
1201 		goto out;
1202 
1203 	if (XFS_IS_CORRUPT(mp, ir.loaded != ifp->if_nextents)) {
1204 		error = -EFSCORRUPTED;
1205 		goto out;
1206 	}
1207 	ASSERT(ir.loaded == xfs_iext_count(ifp));
1208 	/*
1209 	 * Use release semantics so that we can use acquire semantics in
1210 	 * xfs_need_iread_extents and be guaranteed to see a valid mapping tree
1211 	 * after that load.
1212 	 */
1213 	smp_store_release(&ifp->if_needextents, 0);
1214 	return 0;
1215 out:
1216 	xfs_iext_destroy(ifp);
1217 	return error;
1218 }
1219 
1220 /*
1221  * Returns the relative block number of the first unused block(s) in the given
1222  * fork with at least "len" logically contiguous blocks free.  This is the
1223  * lowest-address hole if the fork has holes, else the first block past the end
1224  * of fork.  Return 0 if the fork is currently local (in-inode).
1225  */
1226 int						/* error */
1227 xfs_bmap_first_unused(
1228 	struct xfs_trans	*tp,		/* transaction pointer */
1229 	struct xfs_inode	*ip,		/* incore inode */
1230 	xfs_extlen_t		len,		/* size of hole to find */
1231 	xfs_fileoff_t		*first_unused,	/* unused block */
1232 	int			whichfork)	/* data or attr fork */
1233 {
1234 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
1235 	struct xfs_bmbt_irec	got;
1236 	struct xfs_iext_cursor	icur;
1237 	xfs_fileoff_t		lastaddr = 0;
1238 	xfs_fileoff_t		lowest, max;
1239 	int			error;
1240 
1241 	if (ifp->if_format == XFS_DINODE_FMT_LOCAL) {
1242 		*first_unused = 0;
1243 		return 0;
1244 	}
1245 
1246 	ASSERT(xfs_ifork_has_extents(ifp));
1247 
1248 	error = xfs_iread_extents(tp, ip, whichfork);
1249 	if (error)
1250 		return error;
1251 
1252 	lowest = max = *first_unused;
1253 	for_each_xfs_iext(ifp, &icur, &got) {
1254 		/*
1255 		 * See if the hole before this extent will work.
1256 		 */
1257 		if (got.br_startoff >= lowest + len &&
1258 		    got.br_startoff - max >= len)
1259 			break;
1260 		lastaddr = got.br_startoff + got.br_blockcount;
1261 		max = XFS_FILEOFF_MAX(lastaddr, lowest);
1262 	}
1263 
1264 	*first_unused = max;
1265 	return 0;
1266 }
1267 
1268 /*
1269  * Returns the file-relative block number of the last block - 1 before
1270  * last_block (input value) in the file.
1271  * This is not based on i_size, it is based on the extent records.
1272  * Returns 0 for local files, as they do not have extent records.
1273  */
1274 int						/* error */
1275 xfs_bmap_last_before(
1276 	struct xfs_trans	*tp,		/* transaction pointer */
1277 	struct xfs_inode	*ip,		/* incore inode */
1278 	xfs_fileoff_t		*last_block,	/* last block */
1279 	int			whichfork)	/* data or attr fork */
1280 {
1281 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
1282 	struct xfs_bmbt_irec	got;
1283 	struct xfs_iext_cursor	icur;
1284 	int			error;
1285 
1286 	switch (ifp->if_format) {
1287 	case XFS_DINODE_FMT_LOCAL:
1288 		*last_block = 0;
1289 		return 0;
1290 	case XFS_DINODE_FMT_BTREE:
1291 	case XFS_DINODE_FMT_EXTENTS:
1292 		break;
1293 	default:
1294 		ASSERT(0);
1295 		return -EFSCORRUPTED;
1296 	}
1297 
1298 	error = xfs_iread_extents(tp, ip, whichfork);
1299 	if (error)
1300 		return error;
1301 
1302 	if (!xfs_iext_lookup_extent_before(ip, ifp, last_block, &icur, &got))
1303 		*last_block = 0;
1304 	return 0;
1305 }
1306 
1307 int
1308 xfs_bmap_last_extent(
1309 	struct xfs_trans	*tp,
1310 	struct xfs_inode	*ip,
1311 	int			whichfork,
1312 	struct xfs_bmbt_irec	*rec,
1313 	int			*is_empty)
1314 {
1315 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
1316 	struct xfs_iext_cursor	icur;
1317 	int			error;
1318 
1319 	error = xfs_iread_extents(tp, ip, whichfork);
1320 	if (error)
1321 		return error;
1322 
1323 	xfs_iext_last(ifp, &icur);
1324 	if (!xfs_iext_get_extent(ifp, &icur, rec))
1325 		*is_empty = 1;
1326 	else
1327 		*is_empty = 0;
1328 	return 0;
1329 }
1330 
1331 /*
1332  * Check the last inode extent to determine whether this allocation will result
1333  * in blocks being allocated at the end of the file. When we allocate new data
1334  * blocks at the end of the file which do not start at the previous data block,
1335  * we will try to align the new blocks at stripe unit boundaries.
1336  *
1337  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1338  * at, or past the EOF.
1339  */
1340 STATIC int
1341 xfs_bmap_isaeof(
1342 	struct xfs_bmalloca	*bma,
1343 	int			whichfork)
1344 {
1345 	struct xfs_bmbt_irec	rec;
1346 	int			is_empty;
1347 	int			error;
1348 
1349 	bma->aeof = false;
1350 	error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1351 				     &is_empty);
1352 	if (error)
1353 		return error;
1354 
1355 	if (is_empty) {
1356 		bma->aeof = true;
1357 		return 0;
1358 	}
1359 
1360 	/*
1361 	 * Check if we are allocation or past the last extent, or at least into
1362 	 * the last delayed allocated extent.
1363 	 */
1364 	bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1365 		(bma->offset >= rec.br_startoff &&
1366 		 isnullstartblock(rec.br_startblock));
1367 	return 0;
1368 }
1369 
1370 /*
1371  * Returns the file-relative block number of the first block past eof in
1372  * the file.  This is not based on i_size, it is based on the extent records.
1373  * Returns 0 for local files, as they do not have extent records.
1374  */
1375 int
1376 xfs_bmap_last_offset(
1377 	struct xfs_inode	*ip,
1378 	xfs_fileoff_t		*last_block,
1379 	int			whichfork)
1380 {
1381 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
1382 	struct xfs_bmbt_irec	rec;
1383 	int			is_empty;
1384 	int			error;
1385 
1386 	*last_block = 0;
1387 
1388 	if (ifp->if_format == XFS_DINODE_FMT_LOCAL)
1389 		return 0;
1390 
1391 	if (XFS_IS_CORRUPT(ip->i_mount, !xfs_ifork_has_extents(ifp)))
1392 		return -EFSCORRUPTED;
1393 
1394 	error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1395 	if (error || is_empty)
1396 		return error;
1397 
1398 	*last_block = rec.br_startoff + rec.br_blockcount;
1399 	return 0;
1400 }
1401 
1402 /*
1403  * Extent tree manipulation functions used during allocation.
1404  */
1405 
1406 /*
1407  * Convert a delayed allocation to a real allocation.
1408  */
1409 STATIC int				/* error */
1410 xfs_bmap_add_extent_delay_real(
1411 	struct xfs_bmalloca	*bma,
1412 	int			whichfork)
1413 {
1414 	struct xfs_mount	*mp = bma->ip->i_mount;
1415 	struct xfs_ifork	*ifp = xfs_ifork_ptr(bma->ip, whichfork);
1416 	struct xfs_bmbt_irec	*new = &bma->got;
1417 	int			error;	/* error return value */
1418 	int			i;	/* temp state */
1419 	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
1420 	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
1421 					/* left is 0, right is 1, prev is 2 */
1422 	int			rval=0;	/* return value (logging flags) */
1423 	uint32_t		state = xfs_bmap_fork_to_state(whichfork);
1424 	xfs_filblks_t		da_new; /* new count del alloc blocks used */
1425 	xfs_filblks_t		da_old; /* old count del alloc blocks used */
1426 	xfs_filblks_t		temp=0;	/* value for da_new calculations */
1427 	int			tmp_rval;	/* partial logging flags */
1428 	struct xfs_bmbt_irec	old;
1429 
1430 	ASSERT(whichfork != XFS_ATTR_FORK);
1431 	ASSERT(!isnullstartblock(new->br_startblock));
1432 	ASSERT(!bma->cur ||
1433 	       (bma->cur->bc_ino.flags & XFS_BTCUR_BMBT_WASDEL));
1434 
1435 	XFS_STATS_INC(mp, xs_add_exlist);
1436 
1437 #define	LEFT		r[0]
1438 #define	RIGHT		r[1]
1439 #define	PREV		r[2]
1440 
1441 	/*
1442 	 * Set up a bunch of variables to make the tests simpler.
1443 	 */
1444 	xfs_iext_get_extent(ifp, &bma->icur, &PREV);
1445 	new_endoff = new->br_startoff + new->br_blockcount;
1446 	ASSERT(isnullstartblock(PREV.br_startblock));
1447 	ASSERT(PREV.br_startoff <= new->br_startoff);
1448 	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1449 
1450 	da_old = startblockval(PREV.br_startblock);
1451 	da_new = 0;
1452 
1453 	/*
1454 	 * Set flags determining what part of the previous delayed allocation
1455 	 * extent is being replaced by a real allocation.
1456 	 */
1457 	if (PREV.br_startoff == new->br_startoff)
1458 		state |= BMAP_LEFT_FILLING;
1459 	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1460 		state |= BMAP_RIGHT_FILLING;
1461 
1462 	/*
1463 	 * Check and set flags if this segment has a left neighbor.
1464 	 * Don't set contiguous if the combined extent would be too large.
1465 	 */
1466 	if (xfs_iext_peek_prev_extent(ifp, &bma->icur, &LEFT)) {
1467 		state |= BMAP_LEFT_VALID;
1468 		if (isnullstartblock(LEFT.br_startblock))
1469 			state |= BMAP_LEFT_DELAY;
1470 	}
1471 
1472 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1473 	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1474 	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1475 	    LEFT.br_state == new->br_state &&
1476 	    LEFT.br_blockcount + new->br_blockcount <= XFS_MAX_BMBT_EXTLEN)
1477 		state |= BMAP_LEFT_CONTIG;
1478 
1479 	/*
1480 	 * Check and set flags if this segment has a right neighbor.
1481 	 * Don't set contiguous if the combined extent would be too large.
1482 	 * Also check for all-three-contiguous being too large.
1483 	 */
1484 	if (xfs_iext_peek_next_extent(ifp, &bma->icur, &RIGHT)) {
1485 		state |= BMAP_RIGHT_VALID;
1486 		if (isnullstartblock(RIGHT.br_startblock))
1487 			state |= BMAP_RIGHT_DELAY;
1488 	}
1489 
1490 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1491 	    new_endoff == RIGHT.br_startoff &&
1492 	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1493 	    new->br_state == RIGHT.br_state &&
1494 	    new->br_blockcount + RIGHT.br_blockcount <= XFS_MAX_BMBT_EXTLEN &&
1495 	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1496 		       BMAP_RIGHT_FILLING)) !=
1497 		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1498 		       BMAP_RIGHT_FILLING) ||
1499 	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1500 			<= XFS_MAX_BMBT_EXTLEN))
1501 		state |= BMAP_RIGHT_CONTIG;
1502 
1503 	error = 0;
1504 	/*
1505 	 * Switch out based on the FILLING and CONTIG state bits.
1506 	 */
1507 	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1508 			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1509 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1510 	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1511 		/*
1512 		 * Filling in all of a previously delayed allocation extent.
1513 		 * The left and right neighbors are both contiguous with new.
1514 		 */
1515 		LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
1516 
1517 		xfs_iext_remove(bma->ip, &bma->icur, state);
1518 		xfs_iext_remove(bma->ip, &bma->icur, state);
1519 		xfs_iext_prev(ifp, &bma->icur);
1520 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1521 		ifp->if_nextents--;
1522 
1523 		if (bma->cur == NULL)
1524 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1525 		else {
1526 			rval = XFS_ILOG_CORE;
1527 			error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1528 			if (error)
1529 				goto done;
1530 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1531 				error = -EFSCORRUPTED;
1532 				goto done;
1533 			}
1534 			error = xfs_btree_delete(bma->cur, &i);
1535 			if (error)
1536 				goto done;
1537 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1538 				error = -EFSCORRUPTED;
1539 				goto done;
1540 			}
1541 			error = xfs_btree_decrement(bma->cur, 0, &i);
1542 			if (error)
1543 				goto done;
1544 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1545 				error = -EFSCORRUPTED;
1546 				goto done;
1547 			}
1548 			error = xfs_bmbt_update(bma->cur, &LEFT);
1549 			if (error)
1550 				goto done;
1551 		}
1552 		break;
1553 
1554 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1555 		/*
1556 		 * Filling in all of a previously delayed allocation extent.
1557 		 * The left neighbor is contiguous, the right is not.
1558 		 */
1559 		old = LEFT;
1560 		LEFT.br_blockcount += PREV.br_blockcount;
1561 
1562 		xfs_iext_remove(bma->ip, &bma->icur, state);
1563 		xfs_iext_prev(ifp, &bma->icur);
1564 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1565 
1566 		if (bma->cur == NULL)
1567 			rval = XFS_ILOG_DEXT;
1568 		else {
1569 			rval = 0;
1570 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &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_RIGHT_CONTIG:
1584 		/*
1585 		 * Filling in all of a previously delayed allocation extent.
1586 		 * The right neighbor is contiguous, the left is not. Take care
1587 		 * with delay -> unwritten extent allocation here because the
1588 		 * delalloc record we are overwriting is always written.
1589 		 */
1590 		PREV.br_startblock = new->br_startblock;
1591 		PREV.br_blockcount += RIGHT.br_blockcount;
1592 		PREV.br_state = new->br_state;
1593 
1594 		xfs_iext_next(ifp, &bma->icur);
1595 		xfs_iext_remove(bma->ip, &bma->icur, state);
1596 		xfs_iext_prev(ifp, &bma->icur);
1597 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1598 
1599 		if (bma->cur == NULL)
1600 			rval = XFS_ILOG_DEXT;
1601 		else {
1602 			rval = 0;
1603 			error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1604 			if (error)
1605 				goto done;
1606 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1607 				error = -EFSCORRUPTED;
1608 				goto done;
1609 			}
1610 			error = xfs_bmbt_update(bma->cur, &PREV);
1611 			if (error)
1612 				goto done;
1613 		}
1614 		break;
1615 
1616 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1617 		/*
1618 		 * Filling in all of a previously delayed allocation extent.
1619 		 * Neither the left nor right neighbors are contiguous with
1620 		 * the new one.
1621 		 */
1622 		PREV.br_startblock = new->br_startblock;
1623 		PREV.br_state = new->br_state;
1624 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1625 		ifp->if_nextents++;
1626 
1627 		if (bma->cur == NULL)
1628 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1629 		else {
1630 			rval = XFS_ILOG_CORE;
1631 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1632 			if (error)
1633 				goto done;
1634 			if (XFS_IS_CORRUPT(mp, i != 0)) {
1635 				error = -EFSCORRUPTED;
1636 				goto done;
1637 			}
1638 			error = xfs_btree_insert(bma->cur, &i);
1639 			if (error)
1640 				goto done;
1641 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1642 				error = -EFSCORRUPTED;
1643 				goto done;
1644 			}
1645 		}
1646 		break;
1647 
1648 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1649 		/*
1650 		 * Filling in the first part of a previous delayed allocation.
1651 		 * The left neighbor is contiguous.
1652 		 */
1653 		old = LEFT;
1654 		temp = PREV.br_blockcount - new->br_blockcount;
1655 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1656 				startblockval(PREV.br_startblock));
1657 
1658 		LEFT.br_blockcount += new->br_blockcount;
1659 
1660 		PREV.br_blockcount = temp;
1661 		PREV.br_startoff += new->br_blockcount;
1662 		PREV.br_startblock = nullstartblock(da_new);
1663 
1664 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1665 		xfs_iext_prev(ifp, &bma->icur);
1666 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1667 
1668 		if (bma->cur == NULL)
1669 			rval = XFS_ILOG_DEXT;
1670 		else {
1671 			rval = 0;
1672 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1673 			if (error)
1674 				goto done;
1675 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1676 				error = -EFSCORRUPTED;
1677 				goto done;
1678 			}
1679 			error = xfs_bmbt_update(bma->cur, &LEFT);
1680 			if (error)
1681 				goto done;
1682 		}
1683 		break;
1684 
1685 	case BMAP_LEFT_FILLING:
1686 		/*
1687 		 * Filling in the first part of a previous delayed allocation.
1688 		 * The left neighbor is not contiguous.
1689 		 */
1690 		xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1691 		ifp->if_nextents++;
1692 
1693 		if (bma->cur == NULL)
1694 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1695 		else {
1696 			rval = XFS_ILOG_CORE;
1697 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1698 			if (error)
1699 				goto done;
1700 			if (XFS_IS_CORRUPT(mp, i != 0)) {
1701 				error = -EFSCORRUPTED;
1702 				goto done;
1703 			}
1704 			error = xfs_btree_insert(bma->cur, &i);
1705 			if (error)
1706 				goto done;
1707 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1708 				error = -EFSCORRUPTED;
1709 				goto done;
1710 			}
1711 		}
1712 
1713 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1714 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1715 					&bma->cur, 1, &tmp_rval, whichfork);
1716 			rval |= tmp_rval;
1717 			if (error)
1718 				goto done;
1719 		}
1720 
1721 		temp = PREV.br_blockcount - new->br_blockcount;
1722 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1723 			startblockval(PREV.br_startblock) -
1724 			(bma->cur ? bma->cur->bc_ino.allocated : 0));
1725 
1726 		PREV.br_startoff = new_endoff;
1727 		PREV.br_blockcount = temp;
1728 		PREV.br_startblock = nullstartblock(da_new);
1729 		xfs_iext_next(ifp, &bma->icur);
1730 		xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1731 		xfs_iext_prev(ifp, &bma->icur);
1732 		break;
1733 
1734 	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1735 		/*
1736 		 * Filling in the last part of a previous delayed allocation.
1737 		 * The right neighbor is contiguous with the new allocation.
1738 		 */
1739 		old = RIGHT;
1740 		RIGHT.br_startoff = new->br_startoff;
1741 		RIGHT.br_startblock = new->br_startblock;
1742 		RIGHT.br_blockcount += new->br_blockcount;
1743 
1744 		if (bma->cur == NULL)
1745 			rval = XFS_ILOG_DEXT;
1746 		else {
1747 			rval = 0;
1748 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1749 			if (error)
1750 				goto done;
1751 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1752 				error = -EFSCORRUPTED;
1753 				goto done;
1754 			}
1755 			error = xfs_bmbt_update(bma->cur, &RIGHT);
1756 			if (error)
1757 				goto done;
1758 		}
1759 
1760 		temp = PREV.br_blockcount - new->br_blockcount;
1761 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1762 			startblockval(PREV.br_startblock));
1763 
1764 		PREV.br_blockcount = temp;
1765 		PREV.br_startblock = nullstartblock(da_new);
1766 
1767 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1768 		xfs_iext_next(ifp, &bma->icur);
1769 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &RIGHT);
1770 		break;
1771 
1772 	case BMAP_RIGHT_FILLING:
1773 		/*
1774 		 * Filling in the last part of a previous delayed allocation.
1775 		 * The right neighbor is not contiguous.
1776 		 */
1777 		xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1778 		ifp->if_nextents++;
1779 
1780 		if (bma->cur == NULL)
1781 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1782 		else {
1783 			rval = XFS_ILOG_CORE;
1784 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1785 			if (error)
1786 				goto done;
1787 			if (XFS_IS_CORRUPT(mp, i != 0)) {
1788 				error = -EFSCORRUPTED;
1789 				goto done;
1790 			}
1791 			error = xfs_btree_insert(bma->cur, &i);
1792 			if (error)
1793 				goto done;
1794 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1795 				error = -EFSCORRUPTED;
1796 				goto done;
1797 			}
1798 		}
1799 
1800 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1801 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1802 				&bma->cur, 1, &tmp_rval, whichfork);
1803 			rval |= tmp_rval;
1804 			if (error)
1805 				goto done;
1806 		}
1807 
1808 		temp = PREV.br_blockcount - new->br_blockcount;
1809 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1810 			startblockval(PREV.br_startblock) -
1811 			(bma->cur ? bma->cur->bc_ino.allocated : 0));
1812 
1813 		PREV.br_startblock = nullstartblock(da_new);
1814 		PREV.br_blockcount = temp;
1815 		xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1816 		xfs_iext_next(ifp, &bma->icur);
1817 		break;
1818 
1819 	case 0:
1820 		/*
1821 		 * Filling in the middle part of a previous delayed allocation.
1822 		 * Contiguity is impossible here.
1823 		 * This case is avoided almost all the time.
1824 		 *
1825 		 * We start with a delayed allocation:
1826 		 *
1827 		 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1828 		 *  PREV @ idx
1829 		 *
1830 	         * and we are allocating:
1831 		 *                     +rrrrrrrrrrrrrrrrr+
1832 		 *			      new
1833 		 *
1834 		 * and we set it up for insertion as:
1835 		 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1836 		 *                            new
1837 		 *  PREV @ idx          LEFT              RIGHT
1838 		 *                      inserted at idx + 1
1839 		 */
1840 		old = PREV;
1841 
1842 		/* LEFT is the new middle */
1843 		LEFT = *new;
1844 
1845 		/* RIGHT is the new right */
1846 		RIGHT.br_state = PREV.br_state;
1847 		RIGHT.br_startoff = new_endoff;
1848 		RIGHT.br_blockcount =
1849 			PREV.br_startoff + PREV.br_blockcount - new_endoff;
1850 		RIGHT.br_startblock =
1851 			nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1852 					RIGHT.br_blockcount));
1853 
1854 		/* truncate PREV */
1855 		PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
1856 		PREV.br_startblock =
1857 			nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1858 					PREV.br_blockcount));
1859 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1860 
1861 		xfs_iext_next(ifp, &bma->icur);
1862 		xfs_iext_insert(bma->ip, &bma->icur, &RIGHT, state);
1863 		xfs_iext_insert(bma->ip, &bma->icur, &LEFT, state);
1864 		ifp->if_nextents++;
1865 
1866 		if (bma->cur == NULL)
1867 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1868 		else {
1869 			rval = XFS_ILOG_CORE;
1870 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1871 			if (error)
1872 				goto done;
1873 			if (XFS_IS_CORRUPT(mp, i != 0)) {
1874 				error = -EFSCORRUPTED;
1875 				goto done;
1876 			}
1877 			error = xfs_btree_insert(bma->cur, &i);
1878 			if (error)
1879 				goto done;
1880 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1881 				error = -EFSCORRUPTED;
1882 				goto done;
1883 			}
1884 		}
1885 
1886 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1887 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1888 					&bma->cur, 1, &tmp_rval, whichfork);
1889 			rval |= tmp_rval;
1890 			if (error)
1891 				goto done;
1892 		}
1893 
1894 		da_new = startblockval(PREV.br_startblock) +
1895 			 startblockval(RIGHT.br_startblock);
1896 		break;
1897 
1898 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1899 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1900 	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1901 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1902 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1903 	case BMAP_LEFT_CONTIG:
1904 	case BMAP_RIGHT_CONTIG:
1905 		/*
1906 		 * These cases are all impossible.
1907 		 */
1908 		ASSERT(0);
1909 	}
1910 
1911 	/* add reverse mapping unless caller opted out */
1912 	if (!(bma->flags & XFS_BMAPI_NORMAP))
1913 		xfs_rmap_map_extent(bma->tp, bma->ip, whichfork, new);
1914 
1915 	/* convert to a btree if necessary */
1916 	if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1917 		int	tmp_logflags;	/* partial log flag return val */
1918 
1919 		ASSERT(bma->cur == NULL);
1920 		error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1921 				&bma->cur, da_old > 0, &tmp_logflags,
1922 				whichfork);
1923 		bma->logflags |= tmp_logflags;
1924 		if (error)
1925 			goto done;
1926 	}
1927 
1928 	if (da_new != da_old)
1929 		xfs_mod_delalloc(mp, (int64_t)da_new - da_old);
1930 
1931 	if (bma->cur) {
1932 		da_new += bma->cur->bc_ino.allocated;
1933 		bma->cur->bc_ino.allocated = 0;
1934 	}
1935 
1936 	/* adjust for changes in reserved delayed indirect blocks */
1937 	if (da_new != da_old) {
1938 		ASSERT(state == 0 || da_new < da_old);
1939 		error = xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new),
1940 				false);
1941 	}
1942 
1943 	xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
1944 done:
1945 	if (whichfork != XFS_COW_FORK)
1946 		bma->logflags |= rval;
1947 	return error;
1948 #undef	LEFT
1949 #undef	RIGHT
1950 #undef	PREV
1951 }
1952 
1953 /*
1954  * Convert an unwritten allocation to a real allocation or vice versa.
1955  */
1956 int					/* error */
1957 xfs_bmap_add_extent_unwritten_real(
1958 	struct xfs_trans	*tp,
1959 	xfs_inode_t		*ip,	/* incore inode pointer */
1960 	int			whichfork,
1961 	struct xfs_iext_cursor	*icur,
1962 	struct xfs_btree_cur	**curp,	/* if *curp is null, not a btree */
1963 	xfs_bmbt_irec_t		*new,	/* new data to add to file extents */
1964 	int			*logflagsp) /* inode logging flags */
1965 {
1966 	struct xfs_btree_cur	*cur;	/* btree cursor */
1967 	int			error;	/* error return value */
1968 	int			i;	/* temp state */
1969 	struct xfs_ifork	*ifp;	/* inode fork pointer */
1970 	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
1971 	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
1972 					/* left is 0, right is 1, prev is 2 */
1973 	int			rval=0;	/* return value (logging flags) */
1974 	uint32_t		state = xfs_bmap_fork_to_state(whichfork);
1975 	struct xfs_mount	*mp = ip->i_mount;
1976 	struct xfs_bmbt_irec	old;
1977 
1978 	*logflagsp = 0;
1979 
1980 	cur = *curp;
1981 	ifp = xfs_ifork_ptr(ip, whichfork);
1982 
1983 	ASSERT(!isnullstartblock(new->br_startblock));
1984 
1985 	XFS_STATS_INC(mp, xs_add_exlist);
1986 
1987 #define	LEFT		r[0]
1988 #define	RIGHT		r[1]
1989 #define	PREV		r[2]
1990 
1991 	/*
1992 	 * Set up a bunch of variables to make the tests simpler.
1993 	 */
1994 	error = 0;
1995 	xfs_iext_get_extent(ifp, icur, &PREV);
1996 	ASSERT(new->br_state != PREV.br_state);
1997 	new_endoff = new->br_startoff + new->br_blockcount;
1998 	ASSERT(PREV.br_startoff <= new->br_startoff);
1999 	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2000 
2001 	/*
2002 	 * Set flags determining what part of the previous oldext allocation
2003 	 * extent is being replaced by a newext allocation.
2004 	 */
2005 	if (PREV.br_startoff == new->br_startoff)
2006 		state |= BMAP_LEFT_FILLING;
2007 	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2008 		state |= BMAP_RIGHT_FILLING;
2009 
2010 	/*
2011 	 * Check and set flags if this segment has a left neighbor.
2012 	 * Don't set contiguous if the combined extent would be too large.
2013 	 */
2014 	if (xfs_iext_peek_prev_extent(ifp, icur, &LEFT)) {
2015 		state |= BMAP_LEFT_VALID;
2016 		if (isnullstartblock(LEFT.br_startblock))
2017 			state |= BMAP_LEFT_DELAY;
2018 	}
2019 
2020 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2021 	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2022 	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2023 	    LEFT.br_state == new->br_state &&
2024 	    LEFT.br_blockcount + new->br_blockcount <= XFS_MAX_BMBT_EXTLEN)
2025 		state |= BMAP_LEFT_CONTIG;
2026 
2027 	/*
2028 	 * Check and set flags if this segment has a right neighbor.
2029 	 * Don't set contiguous if the combined extent would be too large.
2030 	 * Also check for all-three-contiguous being too large.
2031 	 */
2032 	if (xfs_iext_peek_next_extent(ifp, icur, &RIGHT)) {
2033 		state |= BMAP_RIGHT_VALID;
2034 		if (isnullstartblock(RIGHT.br_startblock))
2035 			state |= BMAP_RIGHT_DELAY;
2036 	}
2037 
2038 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2039 	    new_endoff == RIGHT.br_startoff &&
2040 	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2041 	    new->br_state == RIGHT.br_state &&
2042 	    new->br_blockcount + RIGHT.br_blockcount <= XFS_MAX_BMBT_EXTLEN &&
2043 	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2044 		       BMAP_RIGHT_FILLING)) !=
2045 		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2046 		       BMAP_RIGHT_FILLING) ||
2047 	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2048 			<= XFS_MAX_BMBT_EXTLEN))
2049 		state |= BMAP_RIGHT_CONTIG;
2050 
2051 	/*
2052 	 * Switch out based on the FILLING and CONTIG state bits.
2053 	 */
2054 	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2055 			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2056 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2057 	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2058 		/*
2059 		 * Setting all of a previous oldext extent to newext.
2060 		 * The left and right neighbors are both contiguous with new.
2061 		 */
2062 		LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
2063 
2064 		xfs_iext_remove(ip, icur, state);
2065 		xfs_iext_remove(ip, icur, state);
2066 		xfs_iext_prev(ifp, icur);
2067 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2068 		ifp->if_nextents -= 2;
2069 		if (cur == NULL)
2070 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2071 		else {
2072 			rval = XFS_ILOG_CORE;
2073 			error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2074 			if (error)
2075 				goto done;
2076 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2077 				error = -EFSCORRUPTED;
2078 				goto done;
2079 			}
2080 			if ((error = xfs_btree_delete(cur, &i)))
2081 				goto done;
2082 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2083 				error = -EFSCORRUPTED;
2084 				goto done;
2085 			}
2086 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2087 				goto done;
2088 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2089 				error = -EFSCORRUPTED;
2090 				goto done;
2091 			}
2092 			if ((error = xfs_btree_delete(cur, &i)))
2093 				goto done;
2094 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2095 				error = -EFSCORRUPTED;
2096 				goto done;
2097 			}
2098 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2099 				goto done;
2100 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2101 				error = -EFSCORRUPTED;
2102 				goto done;
2103 			}
2104 			error = xfs_bmbt_update(cur, &LEFT);
2105 			if (error)
2106 				goto done;
2107 		}
2108 		break;
2109 
2110 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2111 		/*
2112 		 * Setting all of a previous oldext extent to newext.
2113 		 * The left neighbor is contiguous, the right is not.
2114 		 */
2115 		LEFT.br_blockcount += PREV.br_blockcount;
2116 
2117 		xfs_iext_remove(ip, icur, state);
2118 		xfs_iext_prev(ifp, icur);
2119 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2120 		ifp->if_nextents--;
2121 		if (cur == NULL)
2122 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2123 		else {
2124 			rval = XFS_ILOG_CORE;
2125 			error = xfs_bmbt_lookup_eq(cur, &PREV, &i);
2126 			if (error)
2127 				goto done;
2128 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2129 				error = -EFSCORRUPTED;
2130 				goto done;
2131 			}
2132 			if ((error = xfs_btree_delete(cur, &i)))
2133 				goto done;
2134 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2135 				error = -EFSCORRUPTED;
2136 				goto done;
2137 			}
2138 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2139 				goto done;
2140 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2141 				error = -EFSCORRUPTED;
2142 				goto done;
2143 			}
2144 			error = xfs_bmbt_update(cur, &LEFT);
2145 			if (error)
2146 				goto done;
2147 		}
2148 		break;
2149 
2150 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2151 		/*
2152 		 * Setting all of a previous oldext extent to newext.
2153 		 * The right neighbor is contiguous, the left is not.
2154 		 */
2155 		PREV.br_blockcount += RIGHT.br_blockcount;
2156 		PREV.br_state = new->br_state;
2157 
2158 		xfs_iext_next(ifp, icur);
2159 		xfs_iext_remove(ip, icur, state);
2160 		xfs_iext_prev(ifp, icur);
2161 		xfs_iext_update_extent(ip, state, icur, &PREV);
2162 		ifp->if_nextents--;
2163 
2164 		if (cur == NULL)
2165 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2166 		else {
2167 			rval = XFS_ILOG_CORE;
2168 			error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2169 			if (error)
2170 				goto done;
2171 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2172 				error = -EFSCORRUPTED;
2173 				goto done;
2174 			}
2175 			if ((error = xfs_btree_delete(cur, &i)))
2176 				goto done;
2177 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2178 				error = -EFSCORRUPTED;
2179 				goto done;
2180 			}
2181 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2182 				goto done;
2183 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2184 				error = -EFSCORRUPTED;
2185 				goto done;
2186 			}
2187 			error = xfs_bmbt_update(cur, &PREV);
2188 			if (error)
2189 				goto done;
2190 		}
2191 		break;
2192 
2193 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2194 		/*
2195 		 * Setting all of a previous oldext extent to newext.
2196 		 * Neither the left nor right neighbors are contiguous with
2197 		 * the new one.
2198 		 */
2199 		PREV.br_state = new->br_state;
2200 		xfs_iext_update_extent(ip, state, icur, &PREV);
2201 
2202 		if (cur == NULL)
2203 			rval = XFS_ILOG_DEXT;
2204 		else {
2205 			rval = 0;
2206 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2207 			if (error)
2208 				goto done;
2209 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2210 				error = -EFSCORRUPTED;
2211 				goto done;
2212 			}
2213 			error = xfs_bmbt_update(cur, &PREV);
2214 			if (error)
2215 				goto done;
2216 		}
2217 		break;
2218 
2219 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2220 		/*
2221 		 * Setting the first part of a previous oldext extent to newext.
2222 		 * The left neighbor is contiguous.
2223 		 */
2224 		LEFT.br_blockcount += new->br_blockcount;
2225 
2226 		old = PREV;
2227 		PREV.br_startoff += new->br_blockcount;
2228 		PREV.br_startblock += new->br_blockcount;
2229 		PREV.br_blockcount -= new->br_blockcount;
2230 
2231 		xfs_iext_update_extent(ip, state, icur, &PREV);
2232 		xfs_iext_prev(ifp, icur);
2233 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2234 
2235 		if (cur == NULL)
2236 			rval = XFS_ILOG_DEXT;
2237 		else {
2238 			rval = 0;
2239 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2240 			if (error)
2241 				goto done;
2242 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2243 				error = -EFSCORRUPTED;
2244 				goto done;
2245 			}
2246 			error = xfs_bmbt_update(cur, &PREV);
2247 			if (error)
2248 				goto done;
2249 			error = xfs_btree_decrement(cur, 0, &i);
2250 			if (error)
2251 				goto done;
2252 			error = xfs_bmbt_update(cur, &LEFT);
2253 			if (error)
2254 				goto done;
2255 		}
2256 		break;
2257 
2258 	case BMAP_LEFT_FILLING:
2259 		/*
2260 		 * Setting the first part of a previous oldext extent to newext.
2261 		 * The left neighbor is not contiguous.
2262 		 */
2263 		old = PREV;
2264 		PREV.br_startoff += new->br_blockcount;
2265 		PREV.br_startblock += new->br_blockcount;
2266 		PREV.br_blockcount -= new->br_blockcount;
2267 
2268 		xfs_iext_update_extent(ip, state, icur, &PREV);
2269 		xfs_iext_insert(ip, icur, new, state);
2270 		ifp->if_nextents++;
2271 
2272 		if (cur == NULL)
2273 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2274 		else {
2275 			rval = XFS_ILOG_CORE;
2276 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2277 			if (error)
2278 				goto done;
2279 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2280 				error = -EFSCORRUPTED;
2281 				goto done;
2282 			}
2283 			error = xfs_bmbt_update(cur, &PREV);
2284 			if (error)
2285 				goto done;
2286 			cur->bc_rec.b = *new;
2287 			if ((error = xfs_btree_insert(cur, &i)))
2288 				goto done;
2289 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2290 				error = -EFSCORRUPTED;
2291 				goto done;
2292 			}
2293 		}
2294 		break;
2295 
2296 	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2297 		/*
2298 		 * Setting the last part of a previous oldext extent to newext.
2299 		 * The right neighbor is contiguous with the new allocation.
2300 		 */
2301 		old = PREV;
2302 		PREV.br_blockcount -= new->br_blockcount;
2303 
2304 		RIGHT.br_startoff = new->br_startoff;
2305 		RIGHT.br_startblock = new->br_startblock;
2306 		RIGHT.br_blockcount += new->br_blockcount;
2307 
2308 		xfs_iext_update_extent(ip, state, icur, &PREV);
2309 		xfs_iext_next(ifp, icur);
2310 		xfs_iext_update_extent(ip, state, icur, &RIGHT);
2311 
2312 		if (cur == NULL)
2313 			rval = XFS_ILOG_DEXT;
2314 		else {
2315 			rval = 0;
2316 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2317 			if (error)
2318 				goto done;
2319 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2320 				error = -EFSCORRUPTED;
2321 				goto done;
2322 			}
2323 			error = xfs_bmbt_update(cur, &PREV);
2324 			if (error)
2325 				goto done;
2326 			error = xfs_btree_increment(cur, 0, &i);
2327 			if (error)
2328 				goto done;
2329 			error = xfs_bmbt_update(cur, &RIGHT);
2330 			if (error)
2331 				goto done;
2332 		}
2333 		break;
2334 
2335 	case BMAP_RIGHT_FILLING:
2336 		/*
2337 		 * Setting the last part of a previous oldext extent to newext.
2338 		 * The right neighbor is not contiguous.
2339 		 */
2340 		old = PREV;
2341 		PREV.br_blockcount -= new->br_blockcount;
2342 
2343 		xfs_iext_update_extent(ip, state, icur, &PREV);
2344 		xfs_iext_next(ifp, icur);
2345 		xfs_iext_insert(ip, icur, new, state);
2346 		ifp->if_nextents++;
2347 
2348 		if (cur == NULL)
2349 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2350 		else {
2351 			rval = XFS_ILOG_CORE;
2352 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2353 			if (error)
2354 				goto done;
2355 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2356 				error = -EFSCORRUPTED;
2357 				goto done;
2358 			}
2359 			error = xfs_bmbt_update(cur, &PREV);
2360 			if (error)
2361 				goto done;
2362 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2363 			if (error)
2364 				goto done;
2365 			if (XFS_IS_CORRUPT(mp, i != 0)) {
2366 				error = -EFSCORRUPTED;
2367 				goto done;
2368 			}
2369 			if ((error = xfs_btree_insert(cur, &i)))
2370 				goto done;
2371 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2372 				error = -EFSCORRUPTED;
2373 				goto done;
2374 			}
2375 		}
2376 		break;
2377 
2378 	case 0:
2379 		/*
2380 		 * Setting the middle part of a previous oldext extent to
2381 		 * newext.  Contiguity is impossible here.
2382 		 * One extent becomes three extents.
2383 		 */
2384 		old = PREV;
2385 		PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
2386 
2387 		r[0] = *new;
2388 		r[1].br_startoff = new_endoff;
2389 		r[1].br_blockcount =
2390 			old.br_startoff + old.br_blockcount - new_endoff;
2391 		r[1].br_startblock = new->br_startblock + new->br_blockcount;
2392 		r[1].br_state = PREV.br_state;
2393 
2394 		xfs_iext_update_extent(ip, state, icur, &PREV);
2395 		xfs_iext_next(ifp, icur);
2396 		xfs_iext_insert(ip, icur, &r[1], state);
2397 		xfs_iext_insert(ip, icur, &r[0], state);
2398 		ifp->if_nextents += 2;
2399 
2400 		if (cur == NULL)
2401 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2402 		else {
2403 			rval = XFS_ILOG_CORE;
2404 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2405 			if (error)
2406 				goto done;
2407 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2408 				error = -EFSCORRUPTED;
2409 				goto done;
2410 			}
2411 			/* new right extent - oldext */
2412 			error = xfs_bmbt_update(cur, &r[1]);
2413 			if (error)
2414 				goto done;
2415 			/* new left extent - oldext */
2416 			cur->bc_rec.b = PREV;
2417 			if ((error = xfs_btree_insert(cur, &i)))
2418 				goto done;
2419 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2420 				error = -EFSCORRUPTED;
2421 				goto done;
2422 			}
2423 			/*
2424 			 * Reset the cursor to the position of the new extent
2425 			 * we are about to insert as we can't trust it after
2426 			 * the previous insert.
2427 			 */
2428 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2429 			if (error)
2430 				goto done;
2431 			if (XFS_IS_CORRUPT(mp, i != 0)) {
2432 				error = -EFSCORRUPTED;
2433 				goto done;
2434 			}
2435 			/* new middle extent - newext */
2436 			if ((error = xfs_btree_insert(cur, &i)))
2437 				goto done;
2438 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2439 				error = -EFSCORRUPTED;
2440 				goto done;
2441 			}
2442 		}
2443 		break;
2444 
2445 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2446 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2447 	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2448 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2449 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2450 	case BMAP_LEFT_CONTIG:
2451 	case BMAP_RIGHT_CONTIG:
2452 		/*
2453 		 * These cases are all impossible.
2454 		 */
2455 		ASSERT(0);
2456 	}
2457 
2458 	/* update reverse mappings */
2459 	xfs_rmap_convert_extent(mp, tp, ip, whichfork, new);
2460 
2461 	/* convert to a btree if necessary */
2462 	if (xfs_bmap_needs_btree(ip, whichfork)) {
2463 		int	tmp_logflags;	/* partial log flag return val */
2464 
2465 		ASSERT(cur == NULL);
2466 		error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
2467 				&tmp_logflags, whichfork);
2468 		*logflagsp |= tmp_logflags;
2469 		if (error)
2470 			goto done;
2471 	}
2472 
2473 	/* clear out the allocated field, done with it now in any case. */
2474 	if (cur) {
2475 		cur->bc_ino.allocated = 0;
2476 		*curp = cur;
2477 	}
2478 
2479 	xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
2480 done:
2481 	*logflagsp |= rval;
2482 	return error;
2483 #undef	LEFT
2484 #undef	RIGHT
2485 #undef	PREV
2486 }
2487 
2488 /*
2489  * Convert a hole to a delayed allocation.
2490  */
2491 STATIC void
2492 xfs_bmap_add_extent_hole_delay(
2493 	xfs_inode_t		*ip,	/* incore inode pointer */
2494 	int			whichfork,
2495 	struct xfs_iext_cursor	*icur,
2496 	xfs_bmbt_irec_t		*new)	/* new data to add to file extents */
2497 {
2498 	struct xfs_ifork	*ifp;	/* inode fork pointer */
2499 	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2500 	xfs_filblks_t		newlen=0;	/* new indirect size */
2501 	xfs_filblks_t		oldlen=0;	/* old indirect size */
2502 	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2503 	uint32_t		state = xfs_bmap_fork_to_state(whichfork);
2504 	xfs_filblks_t		temp;	 /* temp for indirect calculations */
2505 
2506 	ifp = xfs_ifork_ptr(ip, whichfork);
2507 	ASSERT(isnullstartblock(new->br_startblock));
2508 
2509 	/*
2510 	 * Check and set flags if this segment has a left neighbor
2511 	 */
2512 	if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2513 		state |= BMAP_LEFT_VALID;
2514 		if (isnullstartblock(left.br_startblock))
2515 			state |= BMAP_LEFT_DELAY;
2516 	}
2517 
2518 	/*
2519 	 * Check and set flags if the current (right) segment exists.
2520 	 * If it doesn't exist, we're converting the hole at end-of-file.
2521 	 */
2522 	if (xfs_iext_get_extent(ifp, icur, &right)) {
2523 		state |= BMAP_RIGHT_VALID;
2524 		if (isnullstartblock(right.br_startblock))
2525 			state |= BMAP_RIGHT_DELAY;
2526 	}
2527 
2528 	/*
2529 	 * Set contiguity flags on the left and right neighbors.
2530 	 * Don't let extents get too large, even if the pieces are contiguous.
2531 	 */
2532 	if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2533 	    left.br_startoff + left.br_blockcount == new->br_startoff &&
2534 	    left.br_blockcount + new->br_blockcount <= XFS_MAX_BMBT_EXTLEN)
2535 		state |= BMAP_LEFT_CONTIG;
2536 
2537 	if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2538 	    new->br_startoff + new->br_blockcount == right.br_startoff &&
2539 	    new->br_blockcount + right.br_blockcount <= XFS_MAX_BMBT_EXTLEN &&
2540 	    (!(state & BMAP_LEFT_CONTIG) ||
2541 	     (left.br_blockcount + new->br_blockcount +
2542 	      right.br_blockcount <= XFS_MAX_BMBT_EXTLEN)))
2543 		state |= BMAP_RIGHT_CONTIG;
2544 
2545 	/*
2546 	 * Switch out based on the contiguity flags.
2547 	 */
2548 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2549 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2550 		/*
2551 		 * New allocation is contiguous with delayed allocations
2552 		 * on the left and on the right.
2553 		 * Merge all three into a single extent record.
2554 		 */
2555 		temp = left.br_blockcount + new->br_blockcount +
2556 			right.br_blockcount;
2557 
2558 		oldlen = startblockval(left.br_startblock) +
2559 			startblockval(new->br_startblock) +
2560 			startblockval(right.br_startblock);
2561 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2562 					 oldlen);
2563 		left.br_startblock = nullstartblock(newlen);
2564 		left.br_blockcount = temp;
2565 
2566 		xfs_iext_remove(ip, icur, state);
2567 		xfs_iext_prev(ifp, icur);
2568 		xfs_iext_update_extent(ip, state, icur, &left);
2569 		break;
2570 
2571 	case BMAP_LEFT_CONTIG:
2572 		/*
2573 		 * New allocation is contiguous with a delayed allocation
2574 		 * on the left.
2575 		 * Merge the new allocation with the left neighbor.
2576 		 */
2577 		temp = left.br_blockcount + new->br_blockcount;
2578 
2579 		oldlen = startblockval(left.br_startblock) +
2580 			startblockval(new->br_startblock);
2581 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2582 					 oldlen);
2583 		left.br_blockcount = temp;
2584 		left.br_startblock = nullstartblock(newlen);
2585 
2586 		xfs_iext_prev(ifp, icur);
2587 		xfs_iext_update_extent(ip, state, icur, &left);
2588 		break;
2589 
2590 	case BMAP_RIGHT_CONTIG:
2591 		/*
2592 		 * New allocation is contiguous with a delayed allocation
2593 		 * on the right.
2594 		 * Merge the new allocation with the right neighbor.
2595 		 */
2596 		temp = new->br_blockcount + right.br_blockcount;
2597 		oldlen = startblockval(new->br_startblock) +
2598 			startblockval(right.br_startblock);
2599 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2600 					 oldlen);
2601 		right.br_startoff = new->br_startoff;
2602 		right.br_startblock = nullstartblock(newlen);
2603 		right.br_blockcount = temp;
2604 		xfs_iext_update_extent(ip, state, icur, &right);
2605 		break;
2606 
2607 	case 0:
2608 		/*
2609 		 * New allocation is not contiguous with another
2610 		 * delayed allocation.
2611 		 * Insert a new entry.
2612 		 */
2613 		oldlen = newlen = 0;
2614 		xfs_iext_insert(ip, icur, new, state);
2615 		break;
2616 	}
2617 	if (oldlen != newlen) {
2618 		ASSERT(oldlen > newlen);
2619 		xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2620 				 false);
2621 		/*
2622 		 * Nothing to do for disk quota accounting here.
2623 		 */
2624 		xfs_mod_delalloc(ip->i_mount, (int64_t)newlen - oldlen);
2625 	}
2626 }
2627 
2628 /*
2629  * Convert a hole to a real allocation.
2630  */
2631 STATIC int				/* error */
2632 xfs_bmap_add_extent_hole_real(
2633 	struct xfs_trans	*tp,
2634 	struct xfs_inode	*ip,
2635 	int			whichfork,
2636 	struct xfs_iext_cursor	*icur,
2637 	struct xfs_btree_cur	**curp,
2638 	struct xfs_bmbt_irec	*new,
2639 	int			*logflagsp,
2640 	uint32_t		flags)
2641 {
2642 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
2643 	struct xfs_mount	*mp = ip->i_mount;
2644 	struct xfs_btree_cur	*cur = *curp;
2645 	int			error;	/* error return value */
2646 	int			i;	/* temp state */
2647 	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2648 	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2649 	int			rval=0;	/* return value (logging flags) */
2650 	uint32_t		state = xfs_bmap_fork_to_state(whichfork);
2651 	struct xfs_bmbt_irec	old;
2652 
2653 	ASSERT(!isnullstartblock(new->br_startblock));
2654 	ASSERT(!cur || !(cur->bc_ino.flags & XFS_BTCUR_BMBT_WASDEL));
2655 
2656 	XFS_STATS_INC(mp, xs_add_exlist);
2657 
2658 	/*
2659 	 * Check and set flags if this segment has a left neighbor.
2660 	 */
2661 	if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2662 		state |= BMAP_LEFT_VALID;
2663 		if (isnullstartblock(left.br_startblock))
2664 			state |= BMAP_LEFT_DELAY;
2665 	}
2666 
2667 	/*
2668 	 * Check and set flags if this segment has a current value.
2669 	 * Not true if we're inserting into the "hole" at eof.
2670 	 */
2671 	if (xfs_iext_get_extent(ifp, icur, &right)) {
2672 		state |= BMAP_RIGHT_VALID;
2673 		if (isnullstartblock(right.br_startblock))
2674 			state |= BMAP_RIGHT_DELAY;
2675 	}
2676 
2677 	/*
2678 	 * We're inserting a real allocation between "left" and "right".
2679 	 * Set the contiguity flags.  Don't let extents get too large.
2680 	 */
2681 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2682 	    left.br_startoff + left.br_blockcount == new->br_startoff &&
2683 	    left.br_startblock + left.br_blockcount == new->br_startblock &&
2684 	    left.br_state == new->br_state &&
2685 	    left.br_blockcount + new->br_blockcount <= XFS_MAX_BMBT_EXTLEN)
2686 		state |= BMAP_LEFT_CONTIG;
2687 
2688 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2689 	    new->br_startoff + new->br_blockcount == right.br_startoff &&
2690 	    new->br_startblock + new->br_blockcount == right.br_startblock &&
2691 	    new->br_state == right.br_state &&
2692 	    new->br_blockcount + right.br_blockcount <= XFS_MAX_BMBT_EXTLEN &&
2693 	    (!(state & BMAP_LEFT_CONTIG) ||
2694 	     left.br_blockcount + new->br_blockcount +
2695 	     right.br_blockcount <= XFS_MAX_BMBT_EXTLEN))
2696 		state |= BMAP_RIGHT_CONTIG;
2697 
2698 	error = 0;
2699 	/*
2700 	 * Select which case we're in here, and implement it.
2701 	 */
2702 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2703 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2704 		/*
2705 		 * New allocation is contiguous with real allocations on the
2706 		 * left and on the right.
2707 		 * Merge all three into a single extent record.
2708 		 */
2709 		left.br_blockcount += new->br_blockcount + right.br_blockcount;
2710 
2711 		xfs_iext_remove(ip, icur, state);
2712 		xfs_iext_prev(ifp, icur);
2713 		xfs_iext_update_extent(ip, state, icur, &left);
2714 		ifp->if_nextents--;
2715 
2716 		if (cur == NULL) {
2717 			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2718 		} else {
2719 			rval = XFS_ILOG_CORE;
2720 			error = xfs_bmbt_lookup_eq(cur, &right, &i);
2721 			if (error)
2722 				goto done;
2723 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2724 				error = -EFSCORRUPTED;
2725 				goto done;
2726 			}
2727 			error = xfs_btree_delete(cur, &i);
2728 			if (error)
2729 				goto done;
2730 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2731 				error = -EFSCORRUPTED;
2732 				goto done;
2733 			}
2734 			error = xfs_btree_decrement(cur, 0, &i);
2735 			if (error)
2736 				goto done;
2737 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2738 				error = -EFSCORRUPTED;
2739 				goto done;
2740 			}
2741 			error = xfs_bmbt_update(cur, &left);
2742 			if (error)
2743 				goto done;
2744 		}
2745 		break;
2746 
2747 	case BMAP_LEFT_CONTIG:
2748 		/*
2749 		 * New allocation is contiguous with a real allocation
2750 		 * on the left.
2751 		 * Merge the new allocation with the left neighbor.
2752 		 */
2753 		old = left;
2754 		left.br_blockcount += new->br_blockcount;
2755 
2756 		xfs_iext_prev(ifp, icur);
2757 		xfs_iext_update_extent(ip, state, icur, &left);
2758 
2759 		if (cur == NULL) {
2760 			rval = xfs_ilog_fext(whichfork);
2761 		} else {
2762 			rval = 0;
2763 			error = xfs_bmbt_lookup_eq(cur, &old, &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_RIGHT_CONTIG:
2777 		/*
2778 		 * New allocation is contiguous with a real allocation
2779 		 * on the right.
2780 		 * Merge the new allocation with the right neighbor.
2781 		 */
2782 		old = right;
2783 
2784 		right.br_startoff = new->br_startoff;
2785 		right.br_startblock = new->br_startblock;
2786 		right.br_blockcount += new->br_blockcount;
2787 		xfs_iext_update_extent(ip, state, icur, &right);
2788 
2789 		if (cur == NULL) {
2790 			rval = xfs_ilog_fext(whichfork);
2791 		} else {
2792 			rval = 0;
2793 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2794 			if (error)
2795 				goto done;
2796 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2797 				error = -EFSCORRUPTED;
2798 				goto done;
2799 			}
2800 			error = xfs_bmbt_update(cur, &right);
2801 			if (error)
2802 				goto done;
2803 		}
2804 		break;
2805 
2806 	case 0:
2807 		/*
2808 		 * New allocation is not contiguous with another
2809 		 * real allocation.
2810 		 * Insert a new entry.
2811 		 */
2812 		xfs_iext_insert(ip, icur, new, state);
2813 		ifp->if_nextents++;
2814 
2815 		if (cur == NULL) {
2816 			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2817 		} else {
2818 			rval = XFS_ILOG_CORE;
2819 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2820 			if (error)
2821 				goto done;
2822 			if (XFS_IS_CORRUPT(mp, i != 0)) {
2823 				error = -EFSCORRUPTED;
2824 				goto done;
2825 			}
2826 			error = xfs_btree_insert(cur, &i);
2827 			if (error)
2828 				goto done;
2829 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2830 				error = -EFSCORRUPTED;
2831 				goto done;
2832 			}
2833 		}
2834 		break;
2835 	}
2836 
2837 	/* add reverse mapping unless caller opted out */
2838 	if (!(flags & XFS_BMAPI_NORMAP))
2839 		xfs_rmap_map_extent(tp, ip, whichfork, new);
2840 
2841 	/* convert to a btree if necessary */
2842 	if (xfs_bmap_needs_btree(ip, whichfork)) {
2843 		int	tmp_logflags;	/* partial log flag return val */
2844 
2845 		ASSERT(cur == NULL);
2846 		error = xfs_bmap_extents_to_btree(tp, ip, curp, 0,
2847 				&tmp_logflags, whichfork);
2848 		*logflagsp |= tmp_logflags;
2849 		cur = *curp;
2850 		if (error)
2851 			goto done;
2852 	}
2853 
2854 	/* clear out the allocated field, done with it now in any case. */
2855 	if (cur)
2856 		cur->bc_ino.allocated = 0;
2857 
2858 	xfs_bmap_check_leaf_extents(cur, ip, whichfork);
2859 done:
2860 	*logflagsp |= rval;
2861 	return error;
2862 }
2863 
2864 /*
2865  * Functions used in the extent read, allocate and remove paths
2866  */
2867 
2868 /*
2869  * Adjust the size of the new extent based on i_extsize and rt extsize.
2870  */
2871 int
2872 xfs_bmap_extsize_align(
2873 	xfs_mount_t	*mp,
2874 	xfs_bmbt_irec_t	*gotp,		/* next extent pointer */
2875 	xfs_bmbt_irec_t	*prevp,		/* previous extent pointer */
2876 	xfs_extlen_t	extsz,		/* align to this extent size */
2877 	int		rt,		/* is this a realtime inode? */
2878 	int		eof,		/* is extent at end-of-file? */
2879 	int		delay,		/* creating delalloc extent? */
2880 	int		convert,	/* overwriting unwritten extent? */
2881 	xfs_fileoff_t	*offp,		/* in/out: aligned offset */
2882 	xfs_extlen_t	*lenp)		/* in/out: aligned length */
2883 {
2884 	xfs_fileoff_t	orig_off;	/* original offset */
2885 	xfs_extlen_t	orig_alen;	/* original length */
2886 	xfs_fileoff_t	orig_end;	/* original off+len */
2887 	xfs_fileoff_t	nexto;		/* next file offset */
2888 	xfs_fileoff_t	prevo;		/* previous file offset */
2889 	xfs_fileoff_t	align_off;	/* temp for offset */
2890 	xfs_extlen_t	align_alen;	/* temp for length */
2891 	xfs_extlen_t	temp;		/* temp for calculations */
2892 
2893 	if (convert)
2894 		return 0;
2895 
2896 	orig_off = align_off = *offp;
2897 	orig_alen = align_alen = *lenp;
2898 	orig_end = orig_off + orig_alen;
2899 
2900 	/*
2901 	 * If this request overlaps an existing extent, then don't
2902 	 * attempt to perform any additional alignment.
2903 	 */
2904 	if (!delay && !eof &&
2905 	    (orig_off >= gotp->br_startoff) &&
2906 	    (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2907 		return 0;
2908 	}
2909 
2910 	/*
2911 	 * If the file offset is unaligned vs. the extent size
2912 	 * we need to align it.  This will be possible unless
2913 	 * the file was previously written with a kernel that didn't
2914 	 * perform this alignment, or if a truncate shot us in the
2915 	 * foot.
2916 	 */
2917 	div_u64_rem(orig_off, extsz, &temp);
2918 	if (temp) {
2919 		align_alen += temp;
2920 		align_off -= temp;
2921 	}
2922 
2923 	/* Same adjustment for the end of the requested area. */
2924 	temp = (align_alen % extsz);
2925 	if (temp)
2926 		align_alen += extsz - temp;
2927 
2928 	/*
2929 	 * For large extent hint sizes, the aligned extent might be larger than
2930 	 * XFS_BMBT_MAX_EXTLEN. In that case, reduce the size by an extsz so
2931 	 * that it pulls the length back under XFS_BMBT_MAX_EXTLEN. The outer
2932 	 * allocation loops handle short allocation just fine, so it is safe to
2933 	 * do this. We only want to do it when we are forced to, though, because
2934 	 * it means more allocation operations are required.
2935 	 */
2936 	while (align_alen > XFS_MAX_BMBT_EXTLEN)
2937 		align_alen -= extsz;
2938 	ASSERT(align_alen <= XFS_MAX_BMBT_EXTLEN);
2939 
2940 	/*
2941 	 * If the previous block overlaps with this proposed allocation
2942 	 * then move the start forward without adjusting the length.
2943 	 */
2944 	if (prevp->br_startoff != NULLFILEOFF) {
2945 		if (prevp->br_startblock == HOLESTARTBLOCK)
2946 			prevo = prevp->br_startoff;
2947 		else
2948 			prevo = prevp->br_startoff + prevp->br_blockcount;
2949 	} else
2950 		prevo = 0;
2951 	if (align_off != orig_off && align_off < prevo)
2952 		align_off = prevo;
2953 	/*
2954 	 * If the next block overlaps with this proposed allocation
2955 	 * then move the start back without adjusting the length,
2956 	 * but not before offset 0.
2957 	 * This may of course make the start overlap previous block,
2958 	 * and if we hit the offset 0 limit then the next block
2959 	 * can still overlap too.
2960 	 */
2961 	if (!eof && gotp->br_startoff != NULLFILEOFF) {
2962 		if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2963 		    (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2964 			nexto = gotp->br_startoff + gotp->br_blockcount;
2965 		else
2966 			nexto = gotp->br_startoff;
2967 	} else
2968 		nexto = NULLFILEOFF;
2969 	if (!eof &&
2970 	    align_off + align_alen != orig_end &&
2971 	    align_off + align_alen > nexto)
2972 		align_off = nexto > align_alen ? nexto - align_alen : 0;
2973 	/*
2974 	 * If we're now overlapping the next or previous extent that
2975 	 * means we can't fit an extsz piece in this hole.  Just move
2976 	 * the start forward to the first valid spot and set
2977 	 * the length so we hit the end.
2978 	 */
2979 	if (align_off != orig_off && align_off < prevo)
2980 		align_off = prevo;
2981 	if (align_off + align_alen != orig_end &&
2982 	    align_off + align_alen > nexto &&
2983 	    nexto != NULLFILEOFF) {
2984 		ASSERT(nexto > prevo);
2985 		align_alen = nexto - align_off;
2986 	}
2987 
2988 	/*
2989 	 * If realtime, and the result isn't a multiple of the realtime
2990 	 * extent size we need to remove blocks until it is.
2991 	 */
2992 	if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2993 		/*
2994 		 * We're not covering the original request, or
2995 		 * we won't be able to once we fix the length.
2996 		 */
2997 		if (orig_off < align_off ||
2998 		    orig_end > align_off + align_alen ||
2999 		    align_alen - temp < orig_alen)
3000 			return -EINVAL;
3001 		/*
3002 		 * Try to fix it by moving the start up.
3003 		 */
3004 		if (align_off + temp <= orig_off) {
3005 			align_alen -= temp;
3006 			align_off += temp;
3007 		}
3008 		/*
3009 		 * Try to fix it by moving the end in.
3010 		 */
3011 		else if (align_off + align_alen - temp >= orig_end)
3012 			align_alen -= temp;
3013 		/*
3014 		 * Set the start to the minimum then trim the length.
3015 		 */
3016 		else {
3017 			align_alen -= orig_off - align_off;
3018 			align_off = orig_off;
3019 			align_alen -= align_alen % mp->m_sb.sb_rextsize;
3020 		}
3021 		/*
3022 		 * Result doesn't cover the request, fail it.
3023 		 */
3024 		if (orig_off < align_off || orig_end > align_off + align_alen)
3025 			return -EINVAL;
3026 	} else {
3027 		ASSERT(orig_off >= align_off);
3028 		/* see XFS_BMBT_MAX_EXTLEN handling above */
3029 		ASSERT(orig_end <= align_off + align_alen ||
3030 		       align_alen + extsz > XFS_MAX_BMBT_EXTLEN);
3031 	}
3032 
3033 #ifdef DEBUG
3034 	if (!eof && gotp->br_startoff != NULLFILEOFF)
3035 		ASSERT(align_off + align_alen <= gotp->br_startoff);
3036 	if (prevp->br_startoff != NULLFILEOFF)
3037 		ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3038 #endif
3039 
3040 	*lenp = align_alen;
3041 	*offp = align_off;
3042 	return 0;
3043 }
3044 
3045 #define XFS_ALLOC_GAP_UNITS	4
3046 
3047 void
3048 xfs_bmap_adjacent(
3049 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3050 {
3051 	xfs_fsblock_t	adjust;		/* adjustment to block numbers */
3052 	xfs_mount_t	*mp;		/* mount point structure */
3053 	int		rt;		/* true if inode is realtime */
3054 
3055 #define	ISVALID(x,y)	\
3056 	(rt ? \
3057 		(x) < mp->m_sb.sb_rblocks : \
3058 		XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3059 		XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3060 		XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3061 
3062 	mp = ap->ip->i_mount;
3063 	rt = XFS_IS_REALTIME_INODE(ap->ip) &&
3064 		(ap->datatype & XFS_ALLOC_USERDATA);
3065 	/*
3066 	 * If allocating at eof, and there's a previous real block,
3067 	 * try to use its last block as our starting point.
3068 	 */
3069 	if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3070 	    !isnullstartblock(ap->prev.br_startblock) &&
3071 	    ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3072 		    ap->prev.br_startblock)) {
3073 		ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3074 		/*
3075 		 * Adjust for the gap between prevp and us.
3076 		 */
3077 		adjust = ap->offset -
3078 			(ap->prev.br_startoff + ap->prev.br_blockcount);
3079 		if (adjust &&
3080 		    ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3081 			ap->blkno += adjust;
3082 	}
3083 	/*
3084 	 * If not at eof, then compare the two neighbor blocks.
3085 	 * Figure out whether either one gives us a good starting point,
3086 	 * and pick the better one.
3087 	 */
3088 	else if (!ap->eof) {
3089 		xfs_fsblock_t	gotbno;		/* right side block number */
3090 		xfs_fsblock_t	gotdiff=0;	/* right side difference */
3091 		xfs_fsblock_t	prevbno;	/* left side block number */
3092 		xfs_fsblock_t	prevdiff=0;	/* left side difference */
3093 
3094 		/*
3095 		 * If there's a previous (left) block, select a requested
3096 		 * start block based on it.
3097 		 */
3098 		if (ap->prev.br_startoff != NULLFILEOFF &&
3099 		    !isnullstartblock(ap->prev.br_startblock) &&
3100 		    (prevbno = ap->prev.br_startblock +
3101 			       ap->prev.br_blockcount) &&
3102 		    ISVALID(prevbno, ap->prev.br_startblock)) {
3103 			/*
3104 			 * Calculate gap to end of previous block.
3105 			 */
3106 			adjust = prevdiff = ap->offset -
3107 				(ap->prev.br_startoff +
3108 				 ap->prev.br_blockcount);
3109 			/*
3110 			 * Figure the startblock based on the previous block's
3111 			 * end and the gap size.
3112 			 * Heuristic!
3113 			 * If the gap is large relative to the piece we're
3114 			 * allocating, or using it gives us an invalid block
3115 			 * number, then just use the end of the previous block.
3116 			 */
3117 			if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3118 			    ISVALID(prevbno + prevdiff,
3119 				    ap->prev.br_startblock))
3120 				prevbno += adjust;
3121 			else
3122 				prevdiff += adjust;
3123 		}
3124 		/*
3125 		 * No previous block or can't follow it, just default.
3126 		 */
3127 		else
3128 			prevbno = NULLFSBLOCK;
3129 		/*
3130 		 * If there's a following (right) block, select a requested
3131 		 * start block based on it.
3132 		 */
3133 		if (!isnullstartblock(ap->got.br_startblock)) {
3134 			/*
3135 			 * Calculate gap to start of next block.
3136 			 */
3137 			adjust = gotdiff = ap->got.br_startoff - ap->offset;
3138 			/*
3139 			 * Figure the startblock based on the next block's
3140 			 * start and the gap size.
3141 			 */
3142 			gotbno = ap->got.br_startblock;
3143 			/*
3144 			 * Heuristic!
3145 			 * If the gap is large relative to the piece we're
3146 			 * allocating, or using it gives us an invalid block
3147 			 * number, then just use the start of the next block
3148 			 * offset by our length.
3149 			 */
3150 			if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3151 			    ISVALID(gotbno - gotdiff, gotbno))
3152 				gotbno -= adjust;
3153 			else if (ISVALID(gotbno - ap->length, gotbno)) {
3154 				gotbno -= ap->length;
3155 				gotdiff += adjust - ap->length;
3156 			} else
3157 				gotdiff += adjust;
3158 		}
3159 		/*
3160 		 * No next block, just default.
3161 		 */
3162 		else
3163 			gotbno = NULLFSBLOCK;
3164 		/*
3165 		 * If both valid, pick the better one, else the only good
3166 		 * one, else ap->blkno is already set (to 0 or the inode block).
3167 		 */
3168 		if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3169 			ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3170 		else if (prevbno != NULLFSBLOCK)
3171 			ap->blkno = prevbno;
3172 		else if (gotbno != NULLFSBLOCK)
3173 			ap->blkno = gotbno;
3174 	}
3175 #undef ISVALID
3176 }
3177 
3178 int
3179 xfs_bmap_longest_free_extent(
3180 	struct xfs_perag	*pag,
3181 	struct xfs_trans	*tp,
3182 	xfs_extlen_t		*blen)
3183 {
3184 	xfs_extlen_t		longest;
3185 	int			error = 0;
3186 
3187 	if (!xfs_perag_initialised_agf(pag)) {
3188 		error = xfs_alloc_read_agf(pag, tp, XFS_ALLOC_FLAG_TRYLOCK,
3189 				NULL);
3190 		if (error)
3191 			return error;
3192 	}
3193 
3194 	longest = xfs_alloc_longest_free_extent(pag,
3195 				xfs_alloc_min_freelist(pag->pag_mount, pag),
3196 				xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
3197 	if (*blen < longest)
3198 		*blen = longest;
3199 
3200 	return 0;
3201 }
3202 
3203 static xfs_extlen_t
3204 xfs_bmap_select_minlen(
3205 	struct xfs_bmalloca	*ap,
3206 	struct xfs_alloc_arg	*args,
3207 	xfs_extlen_t		blen)
3208 {
3209 
3210 	/*
3211 	 * Since we used XFS_ALLOC_FLAG_TRYLOCK in _longest_free_extent(), it is
3212 	 * possible that there is enough contiguous free space for this request.
3213 	 */
3214 	if (blen < ap->minlen)
3215 		return ap->minlen;
3216 
3217 	/*
3218 	 * If the best seen length is less than the request length,
3219 	 * use the best as the minimum, otherwise we've got the maxlen we
3220 	 * were asked for.
3221 	 */
3222 	if (blen < args->maxlen)
3223 		return blen;
3224 	return args->maxlen;
3225 }
3226 
3227 static int
3228 xfs_bmap_btalloc_select_lengths(
3229 	struct xfs_bmalloca	*ap,
3230 	struct xfs_alloc_arg	*args,
3231 	xfs_extlen_t		*blen)
3232 {
3233 	struct xfs_mount	*mp = args->mp;
3234 	struct xfs_perag	*pag;
3235 	xfs_agnumber_t		agno, startag;
3236 	int			error = 0;
3237 
3238 	if (ap->tp->t_flags & XFS_TRANS_LOWMODE) {
3239 		args->total = ap->minlen;
3240 		args->minlen = ap->minlen;
3241 		return 0;
3242 	}
3243 
3244 	args->total = ap->total;
3245 	startag = XFS_FSB_TO_AGNO(mp, ap->blkno);
3246 	if (startag == NULLAGNUMBER)
3247 		startag = 0;
3248 
3249 	*blen = 0;
3250 	for_each_perag_wrap(mp, startag, agno, pag) {
3251 		error = xfs_bmap_longest_free_extent(pag, args->tp, blen);
3252 		if (error && error != -EAGAIN)
3253 			break;
3254 		error = 0;
3255 		if (*blen >= args->maxlen)
3256 			break;
3257 	}
3258 	if (pag)
3259 		xfs_perag_rele(pag);
3260 
3261 	args->minlen = xfs_bmap_select_minlen(ap, args, *blen);
3262 	return error;
3263 }
3264 
3265 /* Update all inode and quota accounting for the allocation we just did. */
3266 static void
3267 xfs_bmap_btalloc_accounting(
3268 	struct xfs_bmalloca	*ap,
3269 	struct xfs_alloc_arg	*args)
3270 {
3271 	if (ap->flags & XFS_BMAPI_COWFORK) {
3272 		/*
3273 		 * COW fork blocks are in-core only and thus are treated as
3274 		 * in-core quota reservation (like delalloc blocks) even when
3275 		 * converted to real blocks. The quota reservation is not
3276 		 * accounted to disk until blocks are remapped to the data
3277 		 * fork. So if these blocks were previously delalloc, we
3278 		 * already have quota reservation and there's nothing to do
3279 		 * yet.
3280 		 */
3281 		if (ap->wasdel) {
3282 			xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len);
3283 			return;
3284 		}
3285 
3286 		/*
3287 		 * Otherwise, we've allocated blocks in a hole. The transaction
3288 		 * has acquired in-core quota reservation for this extent.
3289 		 * Rather than account these as real blocks, however, we reduce
3290 		 * the transaction quota reservation based on the allocation.
3291 		 * This essentially transfers the transaction quota reservation
3292 		 * to that of a delalloc extent.
3293 		 */
3294 		ap->ip->i_delayed_blks += args->len;
3295 		xfs_trans_mod_dquot_byino(ap->tp, ap->ip, XFS_TRANS_DQ_RES_BLKS,
3296 				-(long)args->len);
3297 		return;
3298 	}
3299 
3300 	/* data/attr fork only */
3301 	ap->ip->i_nblocks += args->len;
3302 	xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3303 	if (ap->wasdel) {
3304 		ap->ip->i_delayed_blks -= args->len;
3305 		xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len);
3306 	}
3307 	xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3308 		ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT : XFS_TRANS_DQ_BCOUNT,
3309 		args->len);
3310 }
3311 
3312 static int
3313 xfs_bmap_compute_alignments(
3314 	struct xfs_bmalloca	*ap,
3315 	struct xfs_alloc_arg	*args)
3316 {
3317 	struct xfs_mount	*mp = args->mp;
3318 	xfs_extlen_t		align = 0; /* minimum allocation alignment */
3319 	int			stripe_align = 0;
3320 
3321 	/* stripe alignment for allocation is determined by mount parameters */
3322 	if (mp->m_swidth && xfs_has_swalloc(mp))
3323 		stripe_align = mp->m_swidth;
3324 	else if (mp->m_dalign)
3325 		stripe_align = mp->m_dalign;
3326 
3327 	if (ap->flags & XFS_BMAPI_COWFORK)
3328 		align = xfs_get_cowextsz_hint(ap->ip);
3329 	else if (ap->datatype & XFS_ALLOC_USERDATA)
3330 		align = xfs_get_extsz_hint(ap->ip);
3331 	if (align) {
3332 		if (xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, align, 0,
3333 					ap->eof, 0, ap->conv, &ap->offset,
3334 					&ap->length))
3335 			ASSERT(0);
3336 		ASSERT(ap->length);
3337 	}
3338 
3339 	/* apply extent size hints if obtained earlier */
3340 	if (align) {
3341 		args->prod = align;
3342 		div_u64_rem(ap->offset, args->prod, &args->mod);
3343 		if (args->mod)
3344 			args->mod = args->prod - args->mod;
3345 	} else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3346 		args->prod = 1;
3347 		args->mod = 0;
3348 	} else {
3349 		args->prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3350 		div_u64_rem(ap->offset, args->prod, &args->mod);
3351 		if (args->mod)
3352 			args->mod = args->prod - args->mod;
3353 	}
3354 
3355 	return stripe_align;
3356 }
3357 
3358 static void
3359 xfs_bmap_process_allocated_extent(
3360 	struct xfs_bmalloca	*ap,
3361 	struct xfs_alloc_arg	*args,
3362 	xfs_fileoff_t		orig_offset,
3363 	xfs_extlen_t		orig_length)
3364 {
3365 	ap->blkno = args->fsbno;
3366 	ap->length = args->len;
3367 	/*
3368 	 * If the extent size hint is active, we tried to round the
3369 	 * caller's allocation request offset down to extsz and the
3370 	 * length up to another extsz boundary.  If we found a free
3371 	 * extent we mapped it in starting at this new offset.  If the
3372 	 * newly mapped space isn't long enough to cover any of the
3373 	 * range of offsets that was originally requested, move the
3374 	 * mapping up so that we can fill as much of the caller's
3375 	 * original request as possible.  Free space is apparently
3376 	 * very fragmented so we're unlikely to be able to satisfy the
3377 	 * hints anyway.
3378 	 */
3379 	if (ap->length <= orig_length)
3380 		ap->offset = orig_offset;
3381 	else if (ap->offset + ap->length < orig_offset + orig_length)
3382 		ap->offset = orig_offset + orig_length - ap->length;
3383 	xfs_bmap_btalloc_accounting(ap, args);
3384 }
3385 
3386 #ifdef DEBUG
3387 static int
3388 xfs_bmap_exact_minlen_extent_alloc(
3389 	struct xfs_bmalloca	*ap)
3390 {
3391 	struct xfs_mount	*mp = ap->ip->i_mount;
3392 	struct xfs_alloc_arg	args = { .tp = ap->tp, .mp = mp };
3393 	xfs_fileoff_t		orig_offset;
3394 	xfs_extlen_t		orig_length;
3395 	int			error;
3396 
3397 	ASSERT(ap->length);
3398 
3399 	if (ap->minlen != 1) {
3400 		ap->blkno = NULLFSBLOCK;
3401 		ap->length = 0;
3402 		return 0;
3403 	}
3404 
3405 	orig_offset = ap->offset;
3406 	orig_length = ap->length;
3407 
3408 	args.alloc_minlen_only = 1;
3409 
3410 	xfs_bmap_compute_alignments(ap, &args);
3411 
3412 	/*
3413 	 * Unlike the longest extent available in an AG, we don't track
3414 	 * the length of an AG's shortest extent.
3415 	 * XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT is a debug only knob and
3416 	 * hence we can afford to start traversing from the 0th AG since
3417 	 * we need not be concerned about a drop in performance in
3418 	 * "debug only" code paths.
3419 	 */
3420 	ap->blkno = XFS_AGB_TO_FSB(mp, 0, 0);
3421 
3422 	args.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
3423 	args.minlen = args.maxlen = ap->minlen;
3424 	args.total = ap->total;
3425 
3426 	args.alignment = 1;
3427 	args.minalignslop = 0;
3428 
3429 	args.minleft = ap->minleft;
3430 	args.wasdel = ap->wasdel;
3431 	args.resv = XFS_AG_RESV_NONE;
3432 	args.datatype = ap->datatype;
3433 
3434 	error = xfs_alloc_vextent_first_ag(&args, ap->blkno);
3435 	if (error)
3436 		return error;
3437 
3438 	if (args.fsbno != NULLFSBLOCK) {
3439 		xfs_bmap_process_allocated_extent(ap, &args, orig_offset,
3440 			orig_length);
3441 	} else {
3442 		ap->blkno = NULLFSBLOCK;
3443 		ap->length = 0;
3444 	}
3445 
3446 	return 0;
3447 }
3448 #else
3449 
3450 #define xfs_bmap_exact_minlen_extent_alloc(bma) (-EFSCORRUPTED)
3451 
3452 #endif
3453 
3454 /*
3455  * If we are not low on available data blocks and we are allocating at
3456  * EOF, optimise allocation for contiguous file extension and/or stripe
3457  * alignment of the new extent.
3458  *
3459  * NOTE: ap->aeof is only set if the allocation length is >= the
3460  * stripe unit and the allocation offset is at the end of file.
3461  */
3462 static int
3463 xfs_bmap_btalloc_at_eof(
3464 	struct xfs_bmalloca	*ap,
3465 	struct xfs_alloc_arg	*args,
3466 	xfs_extlen_t		blen,
3467 	int			stripe_align,
3468 	bool			ag_only)
3469 {
3470 	struct xfs_mount	*mp = args->mp;
3471 	struct xfs_perag	*caller_pag = args->pag;
3472 	int			error;
3473 
3474 	/*
3475 	 * If there are already extents in the file, try an exact EOF block
3476 	 * allocation to extend the file as a contiguous extent. If that fails,
3477 	 * or it's the first allocation in a file, just try for a stripe aligned
3478 	 * allocation.
3479 	 */
3480 	if (ap->offset) {
3481 		xfs_extlen_t	nextminlen = 0;
3482 
3483 		/*
3484 		 * Compute the minlen+alignment for the next case.  Set slop so
3485 		 * that the value of minlen+alignment+slop doesn't go up between
3486 		 * the calls.
3487 		 */
3488 		args->alignment = 1;
3489 		if (blen > stripe_align && blen <= args->maxlen)
3490 			nextminlen = blen - stripe_align;
3491 		else
3492 			nextminlen = args->minlen;
3493 		if (nextminlen + stripe_align > args->minlen + 1)
3494 			args->minalignslop = nextminlen + stripe_align -
3495 					args->minlen - 1;
3496 		else
3497 			args->minalignslop = 0;
3498 
3499 		if (!caller_pag)
3500 			args->pag = xfs_perag_get(mp, XFS_FSB_TO_AGNO(mp, ap->blkno));
3501 		error = xfs_alloc_vextent_exact_bno(args, ap->blkno);
3502 		if (!caller_pag) {
3503 			xfs_perag_put(args->pag);
3504 			args->pag = NULL;
3505 		}
3506 		if (error)
3507 			return error;
3508 
3509 		if (args->fsbno != NULLFSBLOCK)
3510 			return 0;
3511 		/*
3512 		 * Exact allocation failed. Reset to try an aligned allocation
3513 		 * according to the original allocation specification.
3514 		 */
3515 		args->alignment = stripe_align;
3516 		args->minlen = nextminlen;
3517 		args->minalignslop = 0;
3518 	} else {
3519 		/*
3520 		 * Adjust minlen to try and preserve alignment if we
3521 		 * can't guarantee an aligned maxlen extent.
3522 		 */
3523 		args->alignment = stripe_align;
3524 		if (blen > args->alignment &&
3525 		    blen <= args->maxlen + args->alignment)
3526 			args->minlen = blen - args->alignment;
3527 		args->minalignslop = 0;
3528 	}
3529 
3530 	if (ag_only) {
3531 		error = xfs_alloc_vextent_near_bno(args, ap->blkno);
3532 	} else {
3533 		args->pag = NULL;
3534 		error = xfs_alloc_vextent_start_ag(args, ap->blkno);
3535 		ASSERT(args->pag == NULL);
3536 		args->pag = caller_pag;
3537 	}
3538 	if (error)
3539 		return error;
3540 
3541 	if (args->fsbno != NULLFSBLOCK)
3542 		return 0;
3543 
3544 	/*
3545 	 * Allocation failed, so turn return the allocation args to their
3546 	 * original non-aligned state so the caller can proceed on allocation
3547 	 * failure as if this function was never called.
3548 	 */
3549 	args->alignment = 1;
3550 	return 0;
3551 }
3552 
3553 /*
3554  * We have failed multiple allocation attempts so now are in a low space
3555  * allocation situation. Try a locality first full filesystem minimum length
3556  * allocation whilst still maintaining necessary total block reservation
3557  * requirements.
3558  *
3559  * If that fails, we are now critically low on space, so perform a last resort
3560  * allocation attempt: no reserve, no locality, blocking, minimum length, full
3561  * filesystem free space scan. We also indicate to future allocations in this
3562  * transaction that we are critically low on space so they don't waste time on
3563  * allocation modes that are unlikely to succeed.
3564  */
3565 int
3566 xfs_bmap_btalloc_low_space(
3567 	struct xfs_bmalloca	*ap,
3568 	struct xfs_alloc_arg	*args)
3569 {
3570 	int			error;
3571 
3572 	if (args->minlen > ap->minlen) {
3573 		args->minlen = ap->minlen;
3574 		error = xfs_alloc_vextent_start_ag(args, ap->blkno);
3575 		if (error || args->fsbno != NULLFSBLOCK)
3576 			return error;
3577 	}
3578 
3579 	/* Last ditch attempt before failure is declared. */
3580 	args->total = ap->minlen;
3581 	error = xfs_alloc_vextent_first_ag(args, 0);
3582 	if (error)
3583 		return error;
3584 	ap->tp->t_flags |= XFS_TRANS_LOWMODE;
3585 	return 0;
3586 }
3587 
3588 static int
3589 xfs_bmap_btalloc_filestreams(
3590 	struct xfs_bmalloca	*ap,
3591 	struct xfs_alloc_arg	*args,
3592 	int			stripe_align)
3593 {
3594 	xfs_extlen_t		blen = 0;
3595 	int			error = 0;
3596 
3597 
3598 	error = xfs_filestream_select_ag(ap, args, &blen);
3599 	if (error)
3600 		return error;
3601 	ASSERT(args->pag);
3602 
3603 	/*
3604 	 * If we are in low space mode, then optimal allocation will fail so
3605 	 * prepare for minimal allocation and jump to the low space algorithm
3606 	 * immediately.
3607 	 */
3608 	if (ap->tp->t_flags & XFS_TRANS_LOWMODE) {
3609 		args->minlen = ap->minlen;
3610 		ASSERT(args->fsbno == NULLFSBLOCK);
3611 		goto out_low_space;
3612 	}
3613 
3614 	args->minlen = xfs_bmap_select_minlen(ap, args, blen);
3615 	if (ap->aeof)
3616 		error = xfs_bmap_btalloc_at_eof(ap, args, blen, stripe_align,
3617 				true);
3618 
3619 	if (!error && args->fsbno == NULLFSBLOCK)
3620 		error = xfs_alloc_vextent_near_bno(args, ap->blkno);
3621 
3622 out_low_space:
3623 	/*
3624 	 * We are now done with the perag reference for the filestreams
3625 	 * association provided by xfs_filestream_select_ag(). Release it now as
3626 	 * we've either succeeded, had a fatal error or we are out of space and
3627 	 * need to do a full filesystem scan for free space which will take it's
3628 	 * own references.
3629 	 */
3630 	xfs_perag_rele(args->pag);
3631 	args->pag = NULL;
3632 	if (error || args->fsbno != NULLFSBLOCK)
3633 		return error;
3634 
3635 	return xfs_bmap_btalloc_low_space(ap, args);
3636 }
3637 
3638 static int
3639 xfs_bmap_btalloc_best_length(
3640 	struct xfs_bmalloca	*ap,
3641 	struct xfs_alloc_arg	*args,
3642 	int			stripe_align)
3643 {
3644 	xfs_extlen_t		blen = 0;
3645 	int			error;
3646 
3647 	ap->blkno = XFS_INO_TO_FSB(args->mp, ap->ip->i_ino);
3648 	xfs_bmap_adjacent(ap);
3649 
3650 	/*
3651 	 * Search for an allocation group with a single extent large enough for
3652 	 * the request.  If one isn't found, then adjust the minimum allocation
3653 	 * size to the largest space found.
3654 	 */
3655 	error = xfs_bmap_btalloc_select_lengths(ap, args, &blen);
3656 	if (error)
3657 		return error;
3658 
3659 	/*
3660 	 * Don't attempt optimal EOF allocation if previous allocations barely
3661 	 * succeeded due to being near ENOSPC. It is highly unlikely we'll get
3662 	 * optimal or even aligned allocations in this case, so don't waste time
3663 	 * trying.
3664 	 */
3665 	if (ap->aeof && !(ap->tp->t_flags & XFS_TRANS_LOWMODE)) {
3666 		error = xfs_bmap_btalloc_at_eof(ap, args, blen, stripe_align,
3667 				false);
3668 		if (error || args->fsbno != NULLFSBLOCK)
3669 			return error;
3670 	}
3671 
3672 	error = xfs_alloc_vextent_start_ag(args, ap->blkno);
3673 	if (error || args->fsbno != NULLFSBLOCK)
3674 		return error;
3675 
3676 	return xfs_bmap_btalloc_low_space(ap, args);
3677 }
3678 
3679 static int
3680 xfs_bmap_btalloc(
3681 	struct xfs_bmalloca	*ap)
3682 {
3683 	struct xfs_mount	*mp = ap->ip->i_mount;
3684 	struct xfs_alloc_arg	args = {
3685 		.tp		= ap->tp,
3686 		.mp		= mp,
3687 		.fsbno		= NULLFSBLOCK,
3688 		.oinfo		= XFS_RMAP_OINFO_SKIP_UPDATE,
3689 		.minleft	= ap->minleft,
3690 		.wasdel		= ap->wasdel,
3691 		.resv		= XFS_AG_RESV_NONE,
3692 		.datatype	= ap->datatype,
3693 		.alignment	= 1,
3694 		.minalignslop	= 0,
3695 	};
3696 	xfs_fileoff_t		orig_offset;
3697 	xfs_extlen_t		orig_length;
3698 	int			error;
3699 	int			stripe_align;
3700 
3701 	ASSERT(ap->length);
3702 	orig_offset = ap->offset;
3703 	orig_length = ap->length;
3704 
3705 	stripe_align = xfs_bmap_compute_alignments(ap, &args);
3706 
3707 	/* Trim the allocation back to the maximum an AG can fit. */
3708 	args.maxlen = min(ap->length, mp->m_ag_max_usable);
3709 
3710 	if ((ap->datatype & XFS_ALLOC_USERDATA) &&
3711 	    xfs_inode_is_filestream(ap->ip))
3712 		error = xfs_bmap_btalloc_filestreams(ap, &args, stripe_align);
3713 	else
3714 		error = xfs_bmap_btalloc_best_length(ap, &args, stripe_align);
3715 	if (error)
3716 		return error;
3717 
3718 	if (args.fsbno != NULLFSBLOCK) {
3719 		xfs_bmap_process_allocated_extent(ap, &args, orig_offset,
3720 			orig_length);
3721 	} else {
3722 		ap->blkno = NULLFSBLOCK;
3723 		ap->length = 0;
3724 	}
3725 	return 0;
3726 }
3727 
3728 /* Trim extent to fit a logical block range. */
3729 void
3730 xfs_trim_extent(
3731 	struct xfs_bmbt_irec	*irec,
3732 	xfs_fileoff_t		bno,
3733 	xfs_filblks_t		len)
3734 {
3735 	xfs_fileoff_t		distance;
3736 	xfs_fileoff_t		end = bno + len;
3737 
3738 	if (irec->br_startoff + irec->br_blockcount <= bno ||
3739 	    irec->br_startoff >= end) {
3740 		irec->br_blockcount = 0;
3741 		return;
3742 	}
3743 
3744 	if (irec->br_startoff < bno) {
3745 		distance = bno - irec->br_startoff;
3746 		if (isnullstartblock(irec->br_startblock))
3747 			irec->br_startblock = DELAYSTARTBLOCK;
3748 		if (irec->br_startblock != DELAYSTARTBLOCK &&
3749 		    irec->br_startblock != HOLESTARTBLOCK)
3750 			irec->br_startblock += distance;
3751 		irec->br_startoff += distance;
3752 		irec->br_blockcount -= distance;
3753 	}
3754 
3755 	if (end < irec->br_startoff + irec->br_blockcount) {
3756 		distance = irec->br_startoff + irec->br_blockcount - end;
3757 		irec->br_blockcount -= distance;
3758 	}
3759 }
3760 
3761 /*
3762  * Trim the returned map to the required bounds
3763  */
3764 STATIC void
3765 xfs_bmapi_trim_map(
3766 	struct xfs_bmbt_irec	*mval,
3767 	struct xfs_bmbt_irec	*got,
3768 	xfs_fileoff_t		*bno,
3769 	xfs_filblks_t		len,
3770 	xfs_fileoff_t		obno,
3771 	xfs_fileoff_t		end,
3772 	int			n,
3773 	uint32_t		flags)
3774 {
3775 	if ((flags & XFS_BMAPI_ENTIRE) ||
3776 	    got->br_startoff + got->br_blockcount <= obno) {
3777 		*mval = *got;
3778 		if (isnullstartblock(got->br_startblock))
3779 			mval->br_startblock = DELAYSTARTBLOCK;
3780 		return;
3781 	}
3782 
3783 	if (obno > *bno)
3784 		*bno = obno;
3785 	ASSERT((*bno >= obno) || (n == 0));
3786 	ASSERT(*bno < end);
3787 	mval->br_startoff = *bno;
3788 	if (isnullstartblock(got->br_startblock))
3789 		mval->br_startblock = DELAYSTARTBLOCK;
3790 	else
3791 		mval->br_startblock = got->br_startblock +
3792 					(*bno - got->br_startoff);
3793 	/*
3794 	 * Return the minimum of what we got and what we asked for for
3795 	 * the length.  We can use the len variable here because it is
3796 	 * modified below and we could have been there before coming
3797 	 * here if the first part of the allocation didn't overlap what
3798 	 * was asked for.
3799 	 */
3800 	mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3801 			got->br_blockcount - (*bno - got->br_startoff));
3802 	mval->br_state = got->br_state;
3803 	ASSERT(mval->br_blockcount <= len);
3804 	return;
3805 }
3806 
3807 /*
3808  * Update and validate the extent map to return
3809  */
3810 STATIC void
3811 xfs_bmapi_update_map(
3812 	struct xfs_bmbt_irec	**map,
3813 	xfs_fileoff_t		*bno,
3814 	xfs_filblks_t		*len,
3815 	xfs_fileoff_t		obno,
3816 	xfs_fileoff_t		end,
3817 	int			*n,
3818 	uint32_t		flags)
3819 {
3820 	xfs_bmbt_irec_t	*mval = *map;
3821 
3822 	ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3823 	       ((mval->br_startoff + mval->br_blockcount) <= end));
3824 	ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3825 	       (mval->br_startoff < obno));
3826 
3827 	*bno = mval->br_startoff + mval->br_blockcount;
3828 	*len = end - *bno;
3829 	if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3830 		/* update previous map with new information */
3831 		ASSERT(mval->br_startblock == mval[-1].br_startblock);
3832 		ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3833 		ASSERT(mval->br_state == mval[-1].br_state);
3834 		mval[-1].br_blockcount = mval->br_blockcount;
3835 		mval[-1].br_state = mval->br_state;
3836 	} else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3837 		   mval[-1].br_startblock != DELAYSTARTBLOCK &&
3838 		   mval[-1].br_startblock != HOLESTARTBLOCK &&
3839 		   mval->br_startblock == mval[-1].br_startblock +
3840 					  mval[-1].br_blockcount &&
3841 		   mval[-1].br_state == mval->br_state) {
3842 		ASSERT(mval->br_startoff ==
3843 		       mval[-1].br_startoff + mval[-1].br_blockcount);
3844 		mval[-1].br_blockcount += mval->br_blockcount;
3845 	} else if (*n > 0 &&
3846 		   mval->br_startblock == DELAYSTARTBLOCK &&
3847 		   mval[-1].br_startblock == DELAYSTARTBLOCK &&
3848 		   mval->br_startoff ==
3849 		   mval[-1].br_startoff + mval[-1].br_blockcount) {
3850 		mval[-1].br_blockcount += mval->br_blockcount;
3851 		mval[-1].br_state = mval->br_state;
3852 	} else if (!((*n == 0) &&
3853 		     ((mval->br_startoff + mval->br_blockcount) <=
3854 		      obno))) {
3855 		mval++;
3856 		(*n)++;
3857 	}
3858 	*map = mval;
3859 }
3860 
3861 /*
3862  * Map file blocks to filesystem blocks without allocation.
3863  */
3864 int
3865 xfs_bmapi_read(
3866 	struct xfs_inode	*ip,
3867 	xfs_fileoff_t		bno,
3868 	xfs_filblks_t		len,
3869 	struct xfs_bmbt_irec	*mval,
3870 	int			*nmap,
3871 	uint32_t		flags)
3872 {
3873 	struct xfs_mount	*mp = ip->i_mount;
3874 	int			whichfork = xfs_bmapi_whichfork(flags);
3875 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
3876 	struct xfs_bmbt_irec	got;
3877 	xfs_fileoff_t		obno;
3878 	xfs_fileoff_t		end;
3879 	struct xfs_iext_cursor	icur;
3880 	int			error;
3881 	bool			eof = false;
3882 	int			n = 0;
3883 
3884 	ASSERT(*nmap >= 1);
3885 	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_ENTIRE)));
3886 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
3887 
3888 	if (WARN_ON_ONCE(!ifp))
3889 		return -EFSCORRUPTED;
3890 
3891 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
3892 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT))
3893 		return -EFSCORRUPTED;
3894 
3895 	if (xfs_is_shutdown(mp))
3896 		return -EIO;
3897 
3898 	XFS_STATS_INC(mp, xs_blk_mapr);
3899 
3900 	error = xfs_iread_extents(NULL, ip, whichfork);
3901 	if (error)
3902 		return error;
3903 
3904 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got))
3905 		eof = true;
3906 	end = bno + len;
3907 	obno = bno;
3908 
3909 	while (bno < end && n < *nmap) {
3910 		/* Reading past eof, act as though there's a hole up to end. */
3911 		if (eof)
3912 			got.br_startoff = end;
3913 		if (got.br_startoff > bno) {
3914 			/* Reading in a hole.  */
3915 			mval->br_startoff = bno;
3916 			mval->br_startblock = HOLESTARTBLOCK;
3917 			mval->br_blockcount =
3918 				XFS_FILBLKS_MIN(len, got.br_startoff - bno);
3919 			mval->br_state = XFS_EXT_NORM;
3920 			bno += mval->br_blockcount;
3921 			len -= mval->br_blockcount;
3922 			mval++;
3923 			n++;
3924 			continue;
3925 		}
3926 
3927 		/* set up the extent map to return. */
3928 		xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
3929 		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
3930 
3931 		/* If we're done, stop now. */
3932 		if (bno >= end || n >= *nmap)
3933 			break;
3934 
3935 		/* Else go on to the next record. */
3936 		if (!xfs_iext_next_extent(ifp, &icur, &got))
3937 			eof = true;
3938 	}
3939 	*nmap = n;
3940 	return 0;
3941 }
3942 
3943 /*
3944  * Add a delayed allocation extent to an inode. Blocks are reserved from the
3945  * global pool and the extent inserted into the inode in-core extent tree.
3946  *
3947  * On entry, got refers to the first extent beyond the offset of the extent to
3948  * allocate or eof is specified if no such extent exists. On return, got refers
3949  * to the extent record that was inserted to the inode fork.
3950  *
3951  * Note that the allocated extent may have been merged with contiguous extents
3952  * during insertion into the inode fork. Thus, got does not reflect the current
3953  * state of the inode fork on return. If necessary, the caller can use lastx to
3954  * look up the updated record in the inode fork.
3955  */
3956 int
3957 xfs_bmapi_reserve_delalloc(
3958 	struct xfs_inode	*ip,
3959 	int			whichfork,
3960 	xfs_fileoff_t		off,
3961 	xfs_filblks_t		len,
3962 	xfs_filblks_t		prealloc,
3963 	struct xfs_bmbt_irec	*got,
3964 	struct xfs_iext_cursor	*icur,
3965 	int			eof)
3966 {
3967 	struct xfs_mount	*mp = ip->i_mount;
3968 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
3969 	xfs_extlen_t		alen;
3970 	xfs_extlen_t		indlen;
3971 	int			error;
3972 	xfs_fileoff_t		aoff = off;
3973 
3974 	/*
3975 	 * Cap the alloc length. Keep track of prealloc so we know whether to
3976 	 * tag the inode before we return.
3977 	 */
3978 	alen = XFS_FILBLKS_MIN(len + prealloc, XFS_MAX_BMBT_EXTLEN);
3979 	if (!eof)
3980 		alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
3981 	if (prealloc && alen >= len)
3982 		prealloc = alen - len;
3983 
3984 	/* Figure out the extent size, adjust alen */
3985 	if (whichfork == XFS_COW_FORK) {
3986 		struct xfs_bmbt_irec	prev;
3987 		xfs_extlen_t		extsz = xfs_get_cowextsz_hint(ip);
3988 
3989 		if (!xfs_iext_peek_prev_extent(ifp, icur, &prev))
3990 			prev.br_startoff = NULLFILEOFF;
3991 
3992 		error = xfs_bmap_extsize_align(mp, got, &prev, extsz, 0, eof,
3993 					       1, 0, &aoff, &alen);
3994 		ASSERT(!error);
3995 	}
3996 
3997 	/*
3998 	 * Make a transaction-less quota reservation for delayed allocation
3999 	 * blocks.  This number gets adjusted later.  We return if we haven't
4000 	 * allocated blocks already inside this loop.
4001 	 */
4002 	error = xfs_quota_reserve_blkres(ip, alen);
4003 	if (error)
4004 		return error;
4005 
4006 	/*
4007 	 * Split changing sb for alen and indlen since they could be coming
4008 	 * from different places.
4009 	 */
4010 	indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4011 	ASSERT(indlen > 0);
4012 
4013 	error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
4014 	if (error)
4015 		goto out_unreserve_quota;
4016 
4017 	error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
4018 	if (error)
4019 		goto out_unreserve_blocks;
4020 
4021 
4022 	ip->i_delayed_blks += alen;
4023 	xfs_mod_delalloc(ip->i_mount, alen + indlen);
4024 
4025 	got->br_startoff = aoff;
4026 	got->br_startblock = nullstartblock(indlen);
4027 	got->br_blockcount = alen;
4028 	got->br_state = XFS_EXT_NORM;
4029 
4030 	xfs_bmap_add_extent_hole_delay(ip, whichfork, icur, got);
4031 
4032 	/*
4033 	 * Tag the inode if blocks were preallocated. Note that COW fork
4034 	 * preallocation can occur at the start or end of the extent, even when
4035 	 * prealloc == 0, so we must also check the aligned offset and length.
4036 	 */
4037 	if (whichfork == XFS_DATA_FORK && prealloc)
4038 		xfs_inode_set_eofblocks_tag(ip);
4039 	if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len))
4040 		xfs_inode_set_cowblocks_tag(ip);
4041 
4042 	return 0;
4043 
4044 out_unreserve_blocks:
4045 	xfs_mod_fdblocks(mp, alen, false);
4046 out_unreserve_quota:
4047 	if (XFS_IS_QUOTA_ON(mp))
4048 		xfs_quota_unreserve_blkres(ip, alen);
4049 	return error;
4050 }
4051 
4052 static int
4053 xfs_bmap_alloc_userdata(
4054 	struct xfs_bmalloca	*bma)
4055 {
4056 	struct xfs_mount	*mp = bma->ip->i_mount;
4057 	int			whichfork = xfs_bmapi_whichfork(bma->flags);
4058 	int			error;
4059 
4060 	/*
4061 	 * Set the data type being allocated. For the data fork, the first data
4062 	 * in the file is treated differently to all other allocations. For the
4063 	 * attribute fork, we only need to ensure the allocated range is not on
4064 	 * the busy list.
4065 	 */
4066 	bma->datatype = XFS_ALLOC_NOBUSY;
4067 	if (whichfork == XFS_DATA_FORK || whichfork == XFS_COW_FORK) {
4068 		bma->datatype |= XFS_ALLOC_USERDATA;
4069 		if (bma->offset == 0)
4070 			bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA;
4071 
4072 		if (mp->m_dalign && bma->length >= mp->m_dalign) {
4073 			error = xfs_bmap_isaeof(bma, whichfork);
4074 			if (error)
4075 				return error;
4076 		}
4077 
4078 		if (XFS_IS_REALTIME_INODE(bma->ip))
4079 			return xfs_bmap_rtalloc(bma);
4080 	}
4081 
4082 	if (unlikely(XFS_TEST_ERROR(false, mp,
4083 			XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT)))
4084 		return xfs_bmap_exact_minlen_extent_alloc(bma);
4085 
4086 	return xfs_bmap_btalloc(bma);
4087 }
4088 
4089 static int
4090 xfs_bmapi_allocate(
4091 	struct xfs_bmalloca	*bma)
4092 {
4093 	struct xfs_mount	*mp = bma->ip->i_mount;
4094 	int			whichfork = xfs_bmapi_whichfork(bma->flags);
4095 	struct xfs_ifork	*ifp = xfs_ifork_ptr(bma->ip, whichfork);
4096 	int			tmp_logflags = 0;
4097 	int			error;
4098 
4099 	ASSERT(bma->length > 0);
4100 
4101 	/*
4102 	 * For the wasdelay case, we could also just allocate the stuff asked
4103 	 * for in this bmap call but that wouldn't be as good.
4104 	 */
4105 	if (bma->wasdel) {
4106 		bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4107 		bma->offset = bma->got.br_startoff;
4108 		if (!xfs_iext_peek_prev_extent(ifp, &bma->icur, &bma->prev))
4109 			bma->prev.br_startoff = NULLFILEOFF;
4110 	} else {
4111 		bma->length = XFS_FILBLKS_MIN(bma->length, XFS_MAX_BMBT_EXTLEN);
4112 		if (!bma->eof)
4113 			bma->length = XFS_FILBLKS_MIN(bma->length,
4114 					bma->got.br_startoff - bma->offset);
4115 	}
4116 
4117 	if (bma->flags & XFS_BMAPI_CONTIG)
4118 		bma->minlen = bma->length;
4119 	else
4120 		bma->minlen = 1;
4121 
4122 	if (bma->flags & XFS_BMAPI_METADATA) {
4123 		if (unlikely(XFS_TEST_ERROR(false, mp,
4124 				XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT)))
4125 			error = xfs_bmap_exact_minlen_extent_alloc(bma);
4126 		else
4127 			error = xfs_bmap_btalloc(bma);
4128 	} else {
4129 		error = xfs_bmap_alloc_userdata(bma);
4130 	}
4131 	if (error || bma->blkno == NULLFSBLOCK)
4132 		return error;
4133 
4134 	if (bma->flags & XFS_BMAPI_ZERO) {
4135 		error = xfs_zero_extent(bma->ip, bma->blkno, bma->length);
4136 		if (error)
4137 			return error;
4138 	}
4139 
4140 	if (ifp->if_format == XFS_DINODE_FMT_BTREE && !bma->cur)
4141 		bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4142 	/*
4143 	 * Bump the number of extents we've allocated
4144 	 * in this call.
4145 	 */
4146 	bma->nallocs++;
4147 
4148 	if (bma->cur)
4149 		bma->cur->bc_ino.flags =
4150 			bma->wasdel ? XFS_BTCUR_BMBT_WASDEL : 0;
4151 
4152 	bma->got.br_startoff = bma->offset;
4153 	bma->got.br_startblock = bma->blkno;
4154 	bma->got.br_blockcount = bma->length;
4155 	bma->got.br_state = XFS_EXT_NORM;
4156 
4157 	if (bma->flags & XFS_BMAPI_PREALLOC)
4158 		bma->got.br_state = XFS_EXT_UNWRITTEN;
4159 
4160 	if (bma->wasdel)
4161 		error = xfs_bmap_add_extent_delay_real(bma, whichfork);
4162 	else
4163 		error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip,
4164 				whichfork, &bma->icur, &bma->cur, &bma->got,
4165 				&bma->logflags, bma->flags);
4166 
4167 	bma->logflags |= tmp_logflags;
4168 	if (error)
4169 		return error;
4170 
4171 	/*
4172 	 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4173 	 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4174 	 * the neighbouring ones.
4175 	 */
4176 	xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4177 
4178 	ASSERT(bma->got.br_startoff <= bma->offset);
4179 	ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4180 	       bma->offset + bma->length);
4181 	ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4182 	       bma->got.br_state == XFS_EXT_UNWRITTEN);
4183 	return 0;
4184 }
4185 
4186 STATIC int
4187 xfs_bmapi_convert_unwritten(
4188 	struct xfs_bmalloca	*bma,
4189 	struct xfs_bmbt_irec	*mval,
4190 	xfs_filblks_t		len,
4191 	uint32_t		flags)
4192 {
4193 	int			whichfork = xfs_bmapi_whichfork(flags);
4194 	struct xfs_ifork	*ifp = xfs_ifork_ptr(bma->ip, whichfork);
4195 	int			tmp_logflags = 0;
4196 	int			error;
4197 
4198 	/* check if we need to do unwritten->real conversion */
4199 	if (mval->br_state == XFS_EXT_UNWRITTEN &&
4200 	    (flags & XFS_BMAPI_PREALLOC))
4201 		return 0;
4202 
4203 	/* check if we need to do real->unwritten conversion */
4204 	if (mval->br_state == XFS_EXT_NORM &&
4205 	    (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4206 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4207 		return 0;
4208 
4209 	/*
4210 	 * Modify (by adding) the state flag, if writing.
4211 	 */
4212 	ASSERT(mval->br_blockcount <= len);
4213 	if (ifp->if_format == XFS_DINODE_FMT_BTREE && !bma->cur) {
4214 		bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4215 					bma->ip, whichfork);
4216 	}
4217 	mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4218 				? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4219 
4220 	/*
4221 	 * Before insertion into the bmbt, zero the range being converted
4222 	 * if required.
4223 	 */
4224 	if (flags & XFS_BMAPI_ZERO) {
4225 		error = xfs_zero_extent(bma->ip, mval->br_startblock,
4226 					mval->br_blockcount);
4227 		if (error)
4228 			return error;
4229 	}
4230 
4231 	error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork,
4232 			&bma->icur, &bma->cur, mval, &tmp_logflags);
4233 	/*
4234 	 * Log the inode core unconditionally in the unwritten extent conversion
4235 	 * path because the conversion might not have done so (e.g., if the
4236 	 * extent count hasn't changed). We need to make sure the inode is dirty
4237 	 * in the transaction for the sake of fsync(), even if nothing has
4238 	 * changed, because fsync() will not force the log for this transaction
4239 	 * unless it sees the inode pinned.
4240 	 *
4241 	 * Note: If we're only converting cow fork extents, there aren't
4242 	 * any on-disk updates to make, so we don't need to log anything.
4243 	 */
4244 	if (whichfork != XFS_COW_FORK)
4245 		bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4246 	if (error)
4247 		return error;
4248 
4249 	/*
4250 	 * Update our extent pointer, given that
4251 	 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4252 	 * of the neighbouring ones.
4253 	 */
4254 	xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4255 
4256 	/*
4257 	 * We may have combined previously unwritten space with written space,
4258 	 * so generate another request.
4259 	 */
4260 	if (mval->br_blockcount < len)
4261 		return -EAGAIN;
4262 	return 0;
4263 }
4264 
4265 xfs_extlen_t
4266 xfs_bmapi_minleft(
4267 	struct xfs_trans	*tp,
4268 	struct xfs_inode	*ip,
4269 	int			fork)
4270 {
4271 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, fork);
4272 
4273 	if (tp && tp->t_highest_agno != NULLAGNUMBER)
4274 		return 0;
4275 	if (ifp->if_format != XFS_DINODE_FMT_BTREE)
4276 		return 1;
4277 	return be16_to_cpu(ifp->if_broot->bb_level) + 1;
4278 }
4279 
4280 /*
4281  * Log whatever the flags say, even if error.  Otherwise we might miss detecting
4282  * a case where the data is changed, there's an error, and it's not logged so we
4283  * don't shutdown when we should.  Don't bother logging extents/btree changes if
4284  * we converted to the other format.
4285  */
4286 static void
4287 xfs_bmapi_finish(
4288 	struct xfs_bmalloca	*bma,
4289 	int			whichfork,
4290 	int			error)
4291 {
4292 	struct xfs_ifork	*ifp = xfs_ifork_ptr(bma->ip, whichfork);
4293 
4294 	if ((bma->logflags & xfs_ilog_fext(whichfork)) &&
4295 	    ifp->if_format != XFS_DINODE_FMT_EXTENTS)
4296 		bma->logflags &= ~xfs_ilog_fext(whichfork);
4297 	else if ((bma->logflags & xfs_ilog_fbroot(whichfork)) &&
4298 		 ifp->if_format != XFS_DINODE_FMT_BTREE)
4299 		bma->logflags &= ~xfs_ilog_fbroot(whichfork);
4300 
4301 	if (bma->logflags)
4302 		xfs_trans_log_inode(bma->tp, bma->ip, bma->logflags);
4303 	if (bma->cur)
4304 		xfs_btree_del_cursor(bma->cur, error);
4305 }
4306 
4307 /*
4308  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4309  * extent state if necessary.  Details behaviour is controlled by the flags
4310  * parameter.  Only allocates blocks from a single allocation group, to avoid
4311  * locking problems.
4312  */
4313 int
4314 xfs_bmapi_write(
4315 	struct xfs_trans	*tp,		/* transaction pointer */
4316 	struct xfs_inode	*ip,		/* incore inode */
4317 	xfs_fileoff_t		bno,		/* starting file offs. mapped */
4318 	xfs_filblks_t		len,		/* length to map in file */
4319 	uint32_t		flags,		/* XFS_BMAPI_... */
4320 	xfs_extlen_t		total,		/* total blocks needed */
4321 	struct xfs_bmbt_irec	*mval,		/* output: map values */
4322 	int			*nmap)		/* i/o: mval size/count */
4323 {
4324 	struct xfs_bmalloca	bma = {
4325 		.tp		= tp,
4326 		.ip		= ip,
4327 		.total		= total,
4328 	};
4329 	struct xfs_mount	*mp = ip->i_mount;
4330 	int			whichfork = xfs_bmapi_whichfork(flags);
4331 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
4332 	xfs_fileoff_t		end;		/* end of mapped file region */
4333 	bool			eof = false;	/* after the end of extents */
4334 	int			error;		/* error return */
4335 	int			n;		/* current extent index */
4336 	xfs_fileoff_t		obno;		/* old block number (offset) */
4337 
4338 #ifdef DEBUG
4339 	xfs_fileoff_t		orig_bno;	/* original block number value */
4340 	int			orig_flags;	/* original flags arg value */
4341 	xfs_filblks_t		orig_len;	/* original value of len arg */
4342 	struct xfs_bmbt_irec	*orig_mval;	/* original value of mval */
4343 	int			orig_nmap;	/* original value of *nmap */
4344 
4345 	orig_bno = bno;
4346 	orig_len = len;
4347 	orig_flags = flags;
4348 	orig_mval = mval;
4349 	orig_nmap = *nmap;
4350 #endif
4351 
4352 	ASSERT(*nmap >= 1);
4353 	ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4354 	ASSERT(tp != NULL);
4355 	ASSERT(len > 0);
4356 	ASSERT(ifp->if_format != XFS_DINODE_FMT_LOCAL);
4357 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4358 	ASSERT(!(flags & XFS_BMAPI_REMAP));
4359 
4360 	/* zeroing is for currently only for data extents, not metadata */
4361 	ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4362 			(XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4363 	/*
4364 	 * we can allocate unwritten extents or pre-zero allocated blocks,
4365 	 * but it makes no sense to do both at once. This would result in
4366 	 * zeroing the unwritten extent twice, but it still being an
4367 	 * unwritten extent....
4368 	 */
4369 	ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4370 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4371 
4372 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
4373 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
4374 		return -EFSCORRUPTED;
4375 	}
4376 
4377 	if (xfs_is_shutdown(mp))
4378 		return -EIO;
4379 
4380 	XFS_STATS_INC(mp, xs_blk_mapw);
4381 
4382 	error = xfs_iread_extents(tp, ip, whichfork);
4383 	if (error)
4384 		goto error0;
4385 
4386 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.icur, &bma.got))
4387 		eof = true;
4388 	if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4389 		bma.prev.br_startoff = NULLFILEOFF;
4390 	bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4391 
4392 	n = 0;
4393 	end = bno + len;
4394 	obno = bno;
4395 	while (bno < end && n < *nmap) {
4396 		bool			need_alloc = false, wasdelay = false;
4397 
4398 		/* in hole or beyond EOF? */
4399 		if (eof || bma.got.br_startoff > bno) {
4400 			/*
4401 			 * CoW fork conversions should /never/ hit EOF or
4402 			 * holes.  There should always be something for us
4403 			 * to work on.
4404 			 */
4405 			ASSERT(!((flags & XFS_BMAPI_CONVERT) &&
4406 			         (flags & XFS_BMAPI_COWFORK)));
4407 
4408 			need_alloc = true;
4409 		} else if (isnullstartblock(bma.got.br_startblock)) {
4410 			wasdelay = true;
4411 		}
4412 
4413 		/*
4414 		 * First, deal with the hole before the allocated space
4415 		 * that we found, if any.
4416 		 */
4417 		if (need_alloc || wasdelay) {
4418 			bma.eof = eof;
4419 			bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4420 			bma.wasdel = wasdelay;
4421 			bma.offset = bno;
4422 			bma.flags = flags;
4423 
4424 			/*
4425 			 * There's a 32/64 bit type mismatch between the
4426 			 * allocation length request (which can be 64 bits in
4427 			 * length) and the bma length request, which is
4428 			 * xfs_extlen_t and therefore 32 bits. Hence we have to
4429 			 * check for 32-bit overflows and handle them here.
4430 			 */
4431 			if (len > (xfs_filblks_t)XFS_MAX_BMBT_EXTLEN)
4432 				bma.length = XFS_MAX_BMBT_EXTLEN;
4433 			else
4434 				bma.length = len;
4435 
4436 			ASSERT(len > 0);
4437 			ASSERT(bma.length > 0);
4438 			error = xfs_bmapi_allocate(&bma);
4439 			if (error)
4440 				goto error0;
4441 			if (bma.blkno == NULLFSBLOCK)
4442 				break;
4443 
4444 			/*
4445 			 * If this is a CoW allocation, record the data in
4446 			 * the refcount btree for orphan recovery.
4447 			 */
4448 			if (whichfork == XFS_COW_FORK)
4449 				xfs_refcount_alloc_cow_extent(tp, bma.blkno,
4450 						bma.length);
4451 		}
4452 
4453 		/* Deal with the allocated space we found.  */
4454 		xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4455 							end, n, flags);
4456 
4457 		/* Execute unwritten extent conversion if necessary */
4458 		error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4459 		if (error == -EAGAIN)
4460 			continue;
4461 		if (error)
4462 			goto error0;
4463 
4464 		/* update the extent map to return */
4465 		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4466 
4467 		/*
4468 		 * If we're done, stop now.  Stop when we've allocated
4469 		 * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4470 		 * the transaction may get too big.
4471 		 */
4472 		if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4473 			break;
4474 
4475 		/* Else go on to the next record. */
4476 		bma.prev = bma.got;
4477 		if (!xfs_iext_next_extent(ifp, &bma.icur, &bma.got))
4478 			eof = true;
4479 	}
4480 	*nmap = n;
4481 
4482 	error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4483 			whichfork);
4484 	if (error)
4485 		goto error0;
4486 
4487 	ASSERT(ifp->if_format != XFS_DINODE_FMT_BTREE ||
4488 	       ifp->if_nextents > XFS_IFORK_MAXEXT(ip, whichfork));
4489 	xfs_bmapi_finish(&bma, whichfork, 0);
4490 	xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4491 		orig_nmap, *nmap);
4492 	return 0;
4493 error0:
4494 	xfs_bmapi_finish(&bma, whichfork, error);
4495 	return error;
4496 }
4497 
4498 /*
4499  * Convert an existing delalloc extent to real blocks based on file offset. This
4500  * attempts to allocate the entire delalloc extent and may require multiple
4501  * invocations to allocate the target offset if a large enough physical extent
4502  * is not available.
4503  */
4504 int
4505 xfs_bmapi_convert_delalloc(
4506 	struct xfs_inode	*ip,
4507 	int			whichfork,
4508 	xfs_off_t		offset,
4509 	struct iomap		*iomap,
4510 	unsigned int		*seq)
4511 {
4512 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
4513 	struct xfs_mount	*mp = ip->i_mount;
4514 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
4515 	struct xfs_bmalloca	bma = { NULL };
4516 	uint16_t		flags = 0;
4517 	struct xfs_trans	*tp;
4518 	int			error;
4519 
4520 	if (whichfork == XFS_COW_FORK)
4521 		flags |= IOMAP_F_SHARED;
4522 
4523 	/*
4524 	 * Space for the extent and indirect blocks was reserved when the
4525 	 * delalloc extent was created so there's no need to do so here.
4526 	 */
4527 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0,
4528 				XFS_TRANS_RESERVE, &tp);
4529 	if (error)
4530 		return error;
4531 
4532 	xfs_ilock(ip, XFS_ILOCK_EXCL);
4533 	xfs_trans_ijoin(tp, ip, 0);
4534 
4535 	error = xfs_iext_count_may_overflow(ip, whichfork,
4536 			XFS_IEXT_ADD_NOSPLIT_CNT);
4537 	if (error == -EFBIG)
4538 		error = xfs_iext_count_upgrade(tp, ip,
4539 				XFS_IEXT_ADD_NOSPLIT_CNT);
4540 	if (error)
4541 		goto out_trans_cancel;
4542 
4543 	if (!xfs_iext_lookup_extent(ip, ifp, offset_fsb, &bma.icur, &bma.got) ||
4544 	    bma.got.br_startoff > offset_fsb) {
4545 		/*
4546 		 * No extent found in the range we are trying to convert.  This
4547 		 * should only happen for the COW fork, where another thread
4548 		 * might have moved the extent to the data fork in the meantime.
4549 		 */
4550 		WARN_ON_ONCE(whichfork != XFS_COW_FORK);
4551 		error = -EAGAIN;
4552 		goto out_trans_cancel;
4553 	}
4554 
4555 	/*
4556 	 * If we find a real extent here we raced with another thread converting
4557 	 * the extent.  Just return the real extent at this offset.
4558 	 */
4559 	if (!isnullstartblock(bma.got.br_startblock)) {
4560 		xfs_bmbt_to_iomap(ip, iomap, &bma.got, 0, flags,
4561 				xfs_iomap_inode_sequence(ip, flags));
4562 		*seq = READ_ONCE(ifp->if_seq);
4563 		goto out_trans_cancel;
4564 	}
4565 
4566 	bma.tp = tp;
4567 	bma.ip = ip;
4568 	bma.wasdel = true;
4569 	bma.offset = bma.got.br_startoff;
4570 	bma.length = max_t(xfs_filblks_t, bma.got.br_blockcount,
4571 			XFS_MAX_BMBT_EXTLEN);
4572 	bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4573 
4574 	/*
4575 	 * When we're converting the delalloc reservations backing dirty pages
4576 	 * in the page cache, we must be careful about how we create the new
4577 	 * extents:
4578 	 *
4579 	 * New CoW fork extents are created unwritten, turned into real extents
4580 	 * when we're about to write the data to disk, and mapped into the data
4581 	 * fork after the write finishes.  End of story.
4582 	 *
4583 	 * New data fork extents must be mapped in as unwritten and converted
4584 	 * to real extents after the write succeeds to avoid exposing stale
4585 	 * disk contents if we crash.
4586 	 */
4587 	bma.flags = XFS_BMAPI_PREALLOC;
4588 	if (whichfork == XFS_COW_FORK)
4589 		bma.flags |= XFS_BMAPI_COWFORK;
4590 
4591 	if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4592 		bma.prev.br_startoff = NULLFILEOFF;
4593 
4594 	error = xfs_bmapi_allocate(&bma);
4595 	if (error)
4596 		goto out_finish;
4597 
4598 	error = -ENOSPC;
4599 	if (WARN_ON_ONCE(bma.blkno == NULLFSBLOCK))
4600 		goto out_finish;
4601 	error = -EFSCORRUPTED;
4602 	if (WARN_ON_ONCE(!xfs_valid_startblock(ip, bma.got.br_startblock)))
4603 		goto out_finish;
4604 
4605 	XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, bma.length));
4606 	XFS_STATS_INC(mp, xs_xstrat_quick);
4607 
4608 	ASSERT(!isnullstartblock(bma.got.br_startblock));
4609 	xfs_bmbt_to_iomap(ip, iomap, &bma.got, 0, flags,
4610 				xfs_iomap_inode_sequence(ip, flags));
4611 	*seq = READ_ONCE(ifp->if_seq);
4612 
4613 	if (whichfork == XFS_COW_FORK)
4614 		xfs_refcount_alloc_cow_extent(tp, bma.blkno, bma.length);
4615 
4616 	error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4617 			whichfork);
4618 	if (error)
4619 		goto out_finish;
4620 
4621 	xfs_bmapi_finish(&bma, whichfork, 0);
4622 	error = xfs_trans_commit(tp);
4623 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
4624 	return error;
4625 
4626 out_finish:
4627 	xfs_bmapi_finish(&bma, whichfork, error);
4628 out_trans_cancel:
4629 	xfs_trans_cancel(tp);
4630 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
4631 	return error;
4632 }
4633 
4634 int
4635 xfs_bmapi_remap(
4636 	struct xfs_trans	*tp,
4637 	struct xfs_inode	*ip,
4638 	xfs_fileoff_t		bno,
4639 	xfs_filblks_t		len,
4640 	xfs_fsblock_t		startblock,
4641 	uint32_t		flags)
4642 {
4643 	struct xfs_mount	*mp = ip->i_mount;
4644 	struct xfs_ifork	*ifp;
4645 	struct xfs_btree_cur	*cur = NULL;
4646 	struct xfs_bmbt_irec	got;
4647 	struct xfs_iext_cursor	icur;
4648 	int			whichfork = xfs_bmapi_whichfork(flags);
4649 	int			logflags = 0, error;
4650 
4651 	ifp = xfs_ifork_ptr(ip, whichfork);
4652 	ASSERT(len > 0);
4653 	ASSERT(len <= (xfs_filblks_t)XFS_MAX_BMBT_EXTLEN);
4654 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4655 	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC |
4656 			   XFS_BMAPI_NORMAP)));
4657 	ASSERT((flags & (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC)) !=
4658 			(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC));
4659 
4660 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
4661 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
4662 		return -EFSCORRUPTED;
4663 	}
4664 
4665 	if (xfs_is_shutdown(mp))
4666 		return -EIO;
4667 
4668 	error = xfs_iread_extents(tp, ip, whichfork);
4669 	if (error)
4670 		return error;
4671 
4672 	if (xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
4673 		/* make sure we only reflink into a hole. */
4674 		ASSERT(got.br_startoff > bno);
4675 		ASSERT(got.br_startoff - bno >= len);
4676 	}
4677 
4678 	ip->i_nblocks += len;
4679 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4680 
4681 	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
4682 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
4683 		cur->bc_ino.flags = 0;
4684 	}
4685 
4686 	got.br_startoff = bno;
4687 	got.br_startblock = startblock;
4688 	got.br_blockcount = len;
4689 	if (flags & XFS_BMAPI_PREALLOC)
4690 		got.br_state = XFS_EXT_UNWRITTEN;
4691 	else
4692 		got.br_state = XFS_EXT_NORM;
4693 
4694 	error = xfs_bmap_add_extent_hole_real(tp, ip, whichfork, &icur,
4695 			&cur, &got, &logflags, flags);
4696 	if (error)
4697 		goto error0;
4698 
4699 	error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags, whichfork);
4700 
4701 error0:
4702 	if (ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS)
4703 		logflags &= ~XFS_ILOG_DEXT;
4704 	else if (ip->i_df.if_format != XFS_DINODE_FMT_BTREE)
4705 		logflags &= ~XFS_ILOG_DBROOT;
4706 
4707 	if (logflags)
4708 		xfs_trans_log_inode(tp, ip, logflags);
4709 	if (cur)
4710 		xfs_btree_del_cursor(cur, error);
4711 	return error;
4712 }
4713 
4714 /*
4715  * When a delalloc extent is split (e.g., due to a hole punch), the original
4716  * indlen reservation must be shared across the two new extents that are left
4717  * behind.
4718  *
4719  * Given the original reservation and the worst case indlen for the two new
4720  * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4721  * reservation fairly across the two new extents. If necessary, steal available
4722  * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4723  * ores == 1). The number of stolen blocks is returned. The availability and
4724  * subsequent accounting of stolen blocks is the responsibility of the caller.
4725  */
4726 static xfs_filblks_t
4727 xfs_bmap_split_indlen(
4728 	xfs_filblks_t			ores,		/* original res. */
4729 	xfs_filblks_t			*indlen1,	/* ext1 worst indlen */
4730 	xfs_filblks_t			*indlen2,	/* ext2 worst indlen */
4731 	xfs_filblks_t			avail)		/* stealable blocks */
4732 {
4733 	xfs_filblks_t			len1 = *indlen1;
4734 	xfs_filblks_t			len2 = *indlen2;
4735 	xfs_filblks_t			nres = len1 + len2; /* new total res. */
4736 	xfs_filblks_t			stolen = 0;
4737 	xfs_filblks_t			resfactor;
4738 
4739 	/*
4740 	 * Steal as many blocks as we can to try and satisfy the worst case
4741 	 * indlen for both new extents.
4742 	 */
4743 	if (ores < nres && avail)
4744 		stolen = XFS_FILBLKS_MIN(nres - ores, avail);
4745 	ores += stolen;
4746 
4747 	 /* nothing else to do if we've satisfied the new reservation */
4748 	if (ores >= nres)
4749 		return stolen;
4750 
4751 	/*
4752 	 * We can't meet the total required reservation for the two extents.
4753 	 * Calculate the percent of the overall shortage between both extents
4754 	 * and apply this percentage to each of the requested indlen values.
4755 	 * This distributes the shortage fairly and reduces the chances that one
4756 	 * of the two extents is left with nothing when extents are repeatedly
4757 	 * split.
4758 	 */
4759 	resfactor = (ores * 100);
4760 	do_div(resfactor, nres);
4761 	len1 *= resfactor;
4762 	do_div(len1, 100);
4763 	len2 *= resfactor;
4764 	do_div(len2, 100);
4765 	ASSERT(len1 + len2 <= ores);
4766 	ASSERT(len1 < *indlen1 && len2 < *indlen2);
4767 
4768 	/*
4769 	 * Hand out the remainder to each extent. If one of the two reservations
4770 	 * is zero, we want to make sure that one gets a block first. The loop
4771 	 * below starts with len1, so hand len2 a block right off the bat if it
4772 	 * is zero.
4773 	 */
4774 	ores -= (len1 + len2);
4775 	ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores);
4776 	if (ores && !len2 && *indlen2) {
4777 		len2++;
4778 		ores--;
4779 	}
4780 	while (ores) {
4781 		if (len1 < *indlen1) {
4782 			len1++;
4783 			ores--;
4784 		}
4785 		if (!ores)
4786 			break;
4787 		if (len2 < *indlen2) {
4788 			len2++;
4789 			ores--;
4790 		}
4791 	}
4792 
4793 	*indlen1 = len1;
4794 	*indlen2 = len2;
4795 
4796 	return stolen;
4797 }
4798 
4799 int
4800 xfs_bmap_del_extent_delay(
4801 	struct xfs_inode	*ip,
4802 	int			whichfork,
4803 	struct xfs_iext_cursor	*icur,
4804 	struct xfs_bmbt_irec	*got,
4805 	struct xfs_bmbt_irec	*del)
4806 {
4807 	struct xfs_mount	*mp = ip->i_mount;
4808 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
4809 	struct xfs_bmbt_irec	new;
4810 	int64_t			da_old, da_new, da_diff = 0;
4811 	xfs_fileoff_t		del_endoff, got_endoff;
4812 	xfs_filblks_t		got_indlen, new_indlen, stolen;
4813 	uint32_t		state = xfs_bmap_fork_to_state(whichfork);
4814 	int			error = 0;
4815 	bool			isrt;
4816 
4817 	XFS_STATS_INC(mp, xs_del_exlist);
4818 
4819 	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4820 	del_endoff = del->br_startoff + del->br_blockcount;
4821 	got_endoff = got->br_startoff + got->br_blockcount;
4822 	da_old = startblockval(got->br_startblock);
4823 	da_new = 0;
4824 
4825 	ASSERT(del->br_blockcount > 0);
4826 	ASSERT(got->br_startoff <= del->br_startoff);
4827 	ASSERT(got_endoff >= del_endoff);
4828 
4829 	if (isrt) {
4830 		uint64_t	rtexts = del->br_blockcount;
4831 
4832 		do_div(rtexts, mp->m_sb.sb_rextsize);
4833 		xfs_mod_frextents(mp, rtexts);
4834 	}
4835 
4836 	/*
4837 	 * Update the inode delalloc counter now and wait to update the
4838 	 * sb counters as we might have to borrow some blocks for the
4839 	 * indirect block accounting.
4840 	 */
4841 	ASSERT(!isrt);
4842 	error = xfs_quota_unreserve_blkres(ip, del->br_blockcount);
4843 	if (error)
4844 		return error;
4845 	ip->i_delayed_blks -= del->br_blockcount;
4846 
4847 	if (got->br_startoff == del->br_startoff)
4848 		state |= BMAP_LEFT_FILLING;
4849 	if (got_endoff == del_endoff)
4850 		state |= BMAP_RIGHT_FILLING;
4851 
4852 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4853 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4854 		/*
4855 		 * Matches the whole extent.  Delete the entry.
4856 		 */
4857 		xfs_iext_remove(ip, icur, state);
4858 		xfs_iext_prev(ifp, icur);
4859 		break;
4860 	case BMAP_LEFT_FILLING:
4861 		/*
4862 		 * Deleting the first part of the extent.
4863 		 */
4864 		got->br_startoff = del_endoff;
4865 		got->br_blockcount -= del->br_blockcount;
4866 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4867 				got->br_blockcount), da_old);
4868 		got->br_startblock = nullstartblock((int)da_new);
4869 		xfs_iext_update_extent(ip, state, icur, got);
4870 		break;
4871 	case BMAP_RIGHT_FILLING:
4872 		/*
4873 		 * Deleting the last part of the extent.
4874 		 */
4875 		got->br_blockcount = got->br_blockcount - del->br_blockcount;
4876 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4877 				got->br_blockcount), da_old);
4878 		got->br_startblock = nullstartblock((int)da_new);
4879 		xfs_iext_update_extent(ip, state, icur, got);
4880 		break;
4881 	case 0:
4882 		/*
4883 		 * Deleting the middle of the extent.
4884 		 *
4885 		 * Distribute the original indlen reservation across the two new
4886 		 * extents.  Steal blocks from the deleted extent if necessary.
4887 		 * Stealing blocks simply fudges the fdblocks accounting below.
4888 		 * Warn if either of the new indlen reservations is zero as this
4889 		 * can lead to delalloc problems.
4890 		 */
4891 		got->br_blockcount = del->br_startoff - got->br_startoff;
4892 		got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount);
4893 
4894 		new.br_blockcount = got_endoff - del_endoff;
4895 		new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount);
4896 
4897 		WARN_ON_ONCE(!got_indlen || !new_indlen);
4898 		stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen,
4899 						       del->br_blockcount);
4900 
4901 		got->br_startblock = nullstartblock((int)got_indlen);
4902 
4903 		new.br_startoff = del_endoff;
4904 		new.br_state = got->br_state;
4905 		new.br_startblock = nullstartblock((int)new_indlen);
4906 
4907 		xfs_iext_update_extent(ip, state, icur, got);
4908 		xfs_iext_next(ifp, icur);
4909 		xfs_iext_insert(ip, icur, &new, state);
4910 
4911 		da_new = got_indlen + new_indlen - stolen;
4912 		del->br_blockcount -= stolen;
4913 		break;
4914 	}
4915 
4916 	ASSERT(da_old >= da_new);
4917 	da_diff = da_old - da_new;
4918 	if (!isrt)
4919 		da_diff += del->br_blockcount;
4920 	if (da_diff) {
4921 		xfs_mod_fdblocks(mp, da_diff, false);
4922 		xfs_mod_delalloc(mp, -da_diff);
4923 	}
4924 	return error;
4925 }
4926 
4927 void
4928 xfs_bmap_del_extent_cow(
4929 	struct xfs_inode	*ip,
4930 	struct xfs_iext_cursor	*icur,
4931 	struct xfs_bmbt_irec	*got,
4932 	struct xfs_bmbt_irec	*del)
4933 {
4934 	struct xfs_mount	*mp = ip->i_mount;
4935 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, XFS_COW_FORK);
4936 	struct xfs_bmbt_irec	new;
4937 	xfs_fileoff_t		del_endoff, got_endoff;
4938 	uint32_t		state = BMAP_COWFORK;
4939 
4940 	XFS_STATS_INC(mp, xs_del_exlist);
4941 
4942 	del_endoff = del->br_startoff + del->br_blockcount;
4943 	got_endoff = got->br_startoff + got->br_blockcount;
4944 
4945 	ASSERT(del->br_blockcount > 0);
4946 	ASSERT(got->br_startoff <= del->br_startoff);
4947 	ASSERT(got_endoff >= del_endoff);
4948 	ASSERT(!isnullstartblock(got->br_startblock));
4949 
4950 	if (got->br_startoff == del->br_startoff)
4951 		state |= BMAP_LEFT_FILLING;
4952 	if (got_endoff == del_endoff)
4953 		state |= BMAP_RIGHT_FILLING;
4954 
4955 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4956 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4957 		/*
4958 		 * Matches the whole extent.  Delete the entry.
4959 		 */
4960 		xfs_iext_remove(ip, icur, state);
4961 		xfs_iext_prev(ifp, icur);
4962 		break;
4963 	case BMAP_LEFT_FILLING:
4964 		/*
4965 		 * Deleting the first part of the extent.
4966 		 */
4967 		got->br_startoff = del_endoff;
4968 		got->br_blockcount -= del->br_blockcount;
4969 		got->br_startblock = del->br_startblock + del->br_blockcount;
4970 		xfs_iext_update_extent(ip, state, icur, got);
4971 		break;
4972 	case BMAP_RIGHT_FILLING:
4973 		/*
4974 		 * Deleting the last part of the extent.
4975 		 */
4976 		got->br_blockcount -= del->br_blockcount;
4977 		xfs_iext_update_extent(ip, state, icur, got);
4978 		break;
4979 	case 0:
4980 		/*
4981 		 * Deleting the middle of the extent.
4982 		 */
4983 		got->br_blockcount = del->br_startoff - got->br_startoff;
4984 
4985 		new.br_startoff = del_endoff;
4986 		new.br_blockcount = got_endoff - del_endoff;
4987 		new.br_state = got->br_state;
4988 		new.br_startblock = del->br_startblock + del->br_blockcount;
4989 
4990 		xfs_iext_update_extent(ip, state, icur, got);
4991 		xfs_iext_next(ifp, icur);
4992 		xfs_iext_insert(ip, icur, &new, state);
4993 		break;
4994 	}
4995 	ip->i_delayed_blks -= del->br_blockcount;
4996 }
4997 
4998 /*
4999  * Called by xfs_bmapi to update file extent records and the btree
5000  * after removing space.
5001  */
5002 STATIC int				/* error */
5003 xfs_bmap_del_extent_real(
5004 	xfs_inode_t		*ip,	/* incore inode pointer */
5005 	xfs_trans_t		*tp,	/* current transaction pointer */
5006 	struct xfs_iext_cursor	*icur,
5007 	struct xfs_btree_cur	*cur,	/* if null, not a btree */
5008 	xfs_bmbt_irec_t		*del,	/* data to remove from extents */
5009 	int			*logflagsp, /* inode logging flags */
5010 	int			whichfork, /* data or attr fork */
5011 	uint32_t		bflags)	/* bmapi flags */
5012 {
5013 	xfs_fsblock_t		del_endblock=0;	/* first block past del */
5014 	xfs_fileoff_t		del_endoff;	/* first offset past del */
5015 	int			do_fx;	/* free extent at end of routine */
5016 	int			error;	/* error return value */
5017 	struct xfs_bmbt_irec	got;	/* current extent entry */
5018 	xfs_fileoff_t		got_endoff;	/* first offset past got */
5019 	int			i;	/* temp state */
5020 	struct xfs_ifork	*ifp;	/* inode fork pointer */
5021 	xfs_mount_t		*mp;	/* mount structure */
5022 	xfs_filblks_t		nblks;	/* quota/sb block count */
5023 	xfs_bmbt_irec_t		new;	/* new record to be inserted */
5024 	/* REFERENCED */
5025 	uint			qfield;	/* quota field to update */
5026 	uint32_t		state = xfs_bmap_fork_to_state(whichfork);
5027 	struct xfs_bmbt_irec	old;
5028 
5029 	*logflagsp = 0;
5030 
5031 	mp = ip->i_mount;
5032 	XFS_STATS_INC(mp, xs_del_exlist);
5033 
5034 	ifp = xfs_ifork_ptr(ip, whichfork);
5035 	ASSERT(del->br_blockcount > 0);
5036 	xfs_iext_get_extent(ifp, icur, &got);
5037 	ASSERT(got.br_startoff <= del->br_startoff);
5038 	del_endoff = del->br_startoff + del->br_blockcount;
5039 	got_endoff = got.br_startoff + got.br_blockcount;
5040 	ASSERT(got_endoff >= del_endoff);
5041 	ASSERT(!isnullstartblock(got.br_startblock));
5042 	qfield = 0;
5043 
5044 	/*
5045 	 * If it's the case where the directory code is running with no block
5046 	 * reservation, and the deleted block is in the middle of its extent,
5047 	 * and the resulting insert of an extent would cause transformation to
5048 	 * btree format, then reject it.  The calling code will then swap blocks
5049 	 * around instead.  We have to do this now, rather than waiting for the
5050 	 * conversion to btree format, since the transaction will be dirty then.
5051 	 */
5052 	if (tp->t_blk_res == 0 &&
5053 	    ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
5054 	    ifp->if_nextents >= XFS_IFORK_MAXEXT(ip, whichfork) &&
5055 	    del->br_startoff > got.br_startoff && del_endoff < got_endoff)
5056 		return -ENOSPC;
5057 
5058 	*logflagsp = XFS_ILOG_CORE;
5059 	if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
5060 		if (!(bflags & XFS_BMAPI_REMAP)) {
5061 			error = xfs_rtfree_blocks(tp, del->br_startblock,
5062 					del->br_blockcount);
5063 			if (error)
5064 				return error;
5065 		}
5066 
5067 		do_fx = 0;
5068 		qfield = XFS_TRANS_DQ_RTBCOUNT;
5069 	} else {
5070 		do_fx = 1;
5071 		qfield = XFS_TRANS_DQ_BCOUNT;
5072 	}
5073 	nblks = del->br_blockcount;
5074 
5075 	del_endblock = del->br_startblock + del->br_blockcount;
5076 	if (cur) {
5077 		error = xfs_bmbt_lookup_eq(cur, &got, &i);
5078 		if (error)
5079 			return error;
5080 		if (XFS_IS_CORRUPT(mp, i != 1))
5081 			return -EFSCORRUPTED;
5082 	}
5083 
5084 	if (got.br_startoff == del->br_startoff)
5085 		state |= BMAP_LEFT_FILLING;
5086 	if (got_endoff == del_endoff)
5087 		state |= BMAP_RIGHT_FILLING;
5088 
5089 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
5090 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
5091 		/*
5092 		 * Matches the whole extent.  Delete the entry.
5093 		 */
5094 		xfs_iext_remove(ip, icur, state);
5095 		xfs_iext_prev(ifp, icur);
5096 		ifp->if_nextents--;
5097 
5098 		*logflagsp |= XFS_ILOG_CORE;
5099 		if (!cur) {
5100 			*logflagsp |= xfs_ilog_fext(whichfork);
5101 			break;
5102 		}
5103 		if ((error = xfs_btree_delete(cur, &i)))
5104 			return error;
5105 		if (XFS_IS_CORRUPT(mp, i != 1))
5106 			return -EFSCORRUPTED;
5107 		break;
5108 	case BMAP_LEFT_FILLING:
5109 		/*
5110 		 * Deleting the first part of the extent.
5111 		 */
5112 		got.br_startoff = del_endoff;
5113 		got.br_startblock = del_endblock;
5114 		got.br_blockcount -= del->br_blockcount;
5115 		xfs_iext_update_extent(ip, state, icur, &got);
5116 		if (!cur) {
5117 			*logflagsp |= xfs_ilog_fext(whichfork);
5118 			break;
5119 		}
5120 		error = xfs_bmbt_update(cur, &got);
5121 		if (error)
5122 			return error;
5123 		break;
5124 	case BMAP_RIGHT_FILLING:
5125 		/*
5126 		 * Deleting the last part of the extent.
5127 		 */
5128 		got.br_blockcount -= del->br_blockcount;
5129 		xfs_iext_update_extent(ip, state, icur, &got);
5130 		if (!cur) {
5131 			*logflagsp |= xfs_ilog_fext(whichfork);
5132 			break;
5133 		}
5134 		error = xfs_bmbt_update(cur, &got);
5135 		if (error)
5136 			return error;
5137 		break;
5138 	case 0:
5139 		/*
5140 		 * Deleting the middle of the extent.
5141 		 */
5142 
5143 		old = got;
5144 
5145 		got.br_blockcount = del->br_startoff - got.br_startoff;
5146 		xfs_iext_update_extent(ip, state, icur, &got);
5147 
5148 		new.br_startoff = del_endoff;
5149 		new.br_blockcount = got_endoff - del_endoff;
5150 		new.br_state = got.br_state;
5151 		new.br_startblock = del_endblock;
5152 
5153 		*logflagsp |= XFS_ILOG_CORE;
5154 		if (cur) {
5155 			error = xfs_bmbt_update(cur, &got);
5156 			if (error)
5157 				return error;
5158 			error = xfs_btree_increment(cur, 0, &i);
5159 			if (error)
5160 				return error;
5161 			cur->bc_rec.b = new;
5162 			error = xfs_btree_insert(cur, &i);
5163 			if (error && error != -ENOSPC)
5164 				return error;
5165 			/*
5166 			 * If get no-space back from btree insert, it tried a
5167 			 * split, and we have a zero block reservation.  Fix up
5168 			 * our state and return the error.
5169 			 */
5170 			if (error == -ENOSPC) {
5171 				/*
5172 				 * Reset the cursor, don't trust it after any
5173 				 * insert operation.
5174 				 */
5175 				error = xfs_bmbt_lookup_eq(cur, &got, &i);
5176 				if (error)
5177 					return error;
5178 				if (XFS_IS_CORRUPT(mp, i != 1))
5179 					return -EFSCORRUPTED;
5180 				/*
5181 				 * Update the btree record back
5182 				 * to the original value.
5183 				 */
5184 				error = xfs_bmbt_update(cur, &old);
5185 				if (error)
5186 					return error;
5187 				/*
5188 				 * Reset the extent record back
5189 				 * to the original value.
5190 				 */
5191 				xfs_iext_update_extent(ip, state, icur, &old);
5192 				*logflagsp = 0;
5193 				return -ENOSPC;
5194 			}
5195 			if (XFS_IS_CORRUPT(mp, i != 1))
5196 				return -EFSCORRUPTED;
5197 		} else
5198 			*logflagsp |= xfs_ilog_fext(whichfork);
5199 
5200 		ifp->if_nextents++;
5201 		xfs_iext_next(ifp, icur);
5202 		xfs_iext_insert(ip, icur, &new, state);
5203 		break;
5204 	}
5205 
5206 	/* remove reverse mapping */
5207 	xfs_rmap_unmap_extent(tp, ip, whichfork, del);
5208 
5209 	/*
5210 	 * If we need to, add to list of extents to delete.
5211 	 */
5212 	if (do_fx && !(bflags & XFS_BMAPI_REMAP)) {
5213 		if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
5214 			xfs_refcount_decrease_extent(tp, del);
5215 		} else {
5216 			error = __xfs_free_extent_later(tp, del->br_startblock,
5217 					del->br_blockcount, NULL,
5218 					XFS_AG_RESV_NONE,
5219 					((bflags & XFS_BMAPI_NODISCARD) ||
5220 					del->br_state == XFS_EXT_UNWRITTEN));
5221 			if (error)
5222 				return error;
5223 		}
5224 	}
5225 
5226 	/*
5227 	 * Adjust inode # blocks in the file.
5228 	 */
5229 	if (nblks)
5230 		ip->i_nblocks -= nblks;
5231 	/*
5232 	 * Adjust quota data.
5233 	 */
5234 	if (qfield && !(bflags & XFS_BMAPI_REMAP))
5235 		xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5236 
5237 	return 0;
5238 }
5239 
5240 /*
5241  * Unmap (remove) blocks from a file.
5242  * If nexts is nonzero then the number of extents to remove is limited to
5243  * that value.  If not all extents in the block range can be removed then
5244  * *done is set.
5245  */
5246 int						/* error */
5247 __xfs_bunmapi(
5248 	struct xfs_trans	*tp,		/* transaction pointer */
5249 	struct xfs_inode	*ip,		/* incore inode */
5250 	xfs_fileoff_t		start,		/* first file offset deleted */
5251 	xfs_filblks_t		*rlen,		/* i/o: amount remaining */
5252 	uint32_t		flags,		/* misc flags */
5253 	xfs_extnum_t		nexts)		/* number of extents max */
5254 {
5255 	struct xfs_btree_cur	*cur;		/* bmap btree cursor */
5256 	struct xfs_bmbt_irec	del;		/* extent being deleted */
5257 	int			error;		/* error return value */
5258 	xfs_extnum_t		extno;		/* extent number in list */
5259 	struct xfs_bmbt_irec	got;		/* current extent record */
5260 	struct xfs_ifork	*ifp;		/* inode fork pointer */
5261 	int			isrt;		/* freeing in rt area */
5262 	int			logflags;	/* transaction logging flags */
5263 	xfs_extlen_t		mod;		/* rt extent offset */
5264 	struct xfs_mount	*mp = ip->i_mount;
5265 	int			tmp_logflags;	/* partial logging flags */
5266 	int			wasdel;		/* was a delayed alloc extent */
5267 	int			whichfork;	/* data or attribute fork */
5268 	xfs_fsblock_t		sum;
5269 	xfs_filblks_t		len = *rlen;	/* length to unmap in file */
5270 	xfs_fileoff_t		end;
5271 	struct xfs_iext_cursor	icur;
5272 	bool			done = false;
5273 
5274 	trace_xfs_bunmap(ip, start, len, flags, _RET_IP_);
5275 
5276 	whichfork = xfs_bmapi_whichfork(flags);
5277 	ASSERT(whichfork != XFS_COW_FORK);
5278 	ifp = xfs_ifork_ptr(ip, whichfork);
5279 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)))
5280 		return -EFSCORRUPTED;
5281 	if (xfs_is_shutdown(mp))
5282 		return -EIO;
5283 
5284 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5285 	ASSERT(len > 0);
5286 	ASSERT(nexts >= 0);
5287 
5288 	error = xfs_iread_extents(tp, ip, whichfork);
5289 	if (error)
5290 		return error;
5291 
5292 	if (xfs_iext_count(ifp) == 0) {
5293 		*rlen = 0;
5294 		return 0;
5295 	}
5296 	XFS_STATS_INC(mp, xs_blk_unmap);
5297 	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5298 	end = start + len;
5299 
5300 	if (!xfs_iext_lookup_extent_before(ip, ifp, &end, &icur, &got)) {
5301 		*rlen = 0;
5302 		return 0;
5303 	}
5304 	end--;
5305 
5306 	logflags = 0;
5307 	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
5308 		ASSERT(ifp->if_format == XFS_DINODE_FMT_BTREE);
5309 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5310 		cur->bc_ino.flags = 0;
5311 	} else
5312 		cur = NULL;
5313 
5314 	if (isrt) {
5315 		/*
5316 		 * Synchronize by locking the bitmap inode.
5317 		 */
5318 		xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
5319 		xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5320 		xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
5321 		xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
5322 	}
5323 
5324 	extno = 0;
5325 	while (end != (xfs_fileoff_t)-1 && end >= start &&
5326 	       (nexts == 0 || extno < nexts)) {
5327 		/*
5328 		 * Is the found extent after a hole in which end lives?
5329 		 * Just back up to the previous extent, if so.
5330 		 */
5331 		if (got.br_startoff > end &&
5332 		    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5333 			done = true;
5334 			break;
5335 		}
5336 		/*
5337 		 * Is the last block of this extent before the range
5338 		 * we're supposed to delete?  If so, we're done.
5339 		 */
5340 		end = XFS_FILEOFF_MIN(end,
5341 			got.br_startoff + got.br_blockcount - 1);
5342 		if (end < start)
5343 			break;
5344 		/*
5345 		 * Then deal with the (possibly delayed) allocated space
5346 		 * we found.
5347 		 */
5348 		del = got;
5349 		wasdel = isnullstartblock(del.br_startblock);
5350 
5351 		if (got.br_startoff < start) {
5352 			del.br_startoff = start;
5353 			del.br_blockcount -= start - got.br_startoff;
5354 			if (!wasdel)
5355 				del.br_startblock += start - got.br_startoff;
5356 		}
5357 		if (del.br_startoff + del.br_blockcount > end + 1)
5358 			del.br_blockcount = end + 1 - del.br_startoff;
5359 
5360 		if (!isrt)
5361 			goto delete;
5362 
5363 		sum = del.br_startblock + del.br_blockcount;
5364 		div_u64_rem(sum, mp->m_sb.sb_rextsize, &mod);
5365 		if (mod) {
5366 			/*
5367 			 * Realtime extent not lined up at the end.
5368 			 * The extent could have been split into written
5369 			 * and unwritten pieces, or we could just be
5370 			 * unmapping part of it.  But we can't really
5371 			 * get rid of part of a realtime extent.
5372 			 */
5373 			if (del.br_state == XFS_EXT_UNWRITTEN) {
5374 				/*
5375 				 * This piece is unwritten, or we're not
5376 				 * using unwritten extents.  Skip over it.
5377 				 */
5378 				ASSERT(end >= mod);
5379 				end -= mod > del.br_blockcount ?
5380 					del.br_blockcount : mod;
5381 				if (end < got.br_startoff &&
5382 				    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5383 					done = true;
5384 					break;
5385 				}
5386 				continue;
5387 			}
5388 			/*
5389 			 * It's written, turn it unwritten.
5390 			 * This is better than zeroing it.
5391 			 */
5392 			ASSERT(del.br_state == XFS_EXT_NORM);
5393 			ASSERT(tp->t_blk_res > 0);
5394 			/*
5395 			 * If this spans a realtime extent boundary,
5396 			 * chop it back to the start of the one we end at.
5397 			 */
5398 			if (del.br_blockcount > mod) {
5399 				del.br_startoff += del.br_blockcount - mod;
5400 				del.br_startblock += del.br_blockcount - mod;
5401 				del.br_blockcount = mod;
5402 			}
5403 			del.br_state = XFS_EXT_UNWRITTEN;
5404 			error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5405 					whichfork, &icur, &cur, &del,
5406 					&logflags);
5407 			if (error)
5408 				goto error0;
5409 			goto nodelete;
5410 		}
5411 		div_u64_rem(del.br_startblock, mp->m_sb.sb_rextsize, &mod);
5412 		if (mod) {
5413 			xfs_extlen_t off = mp->m_sb.sb_rextsize - mod;
5414 
5415 			/*
5416 			 * Realtime extent is lined up at the end but not
5417 			 * at the front.  We'll get rid of full extents if
5418 			 * we can.
5419 			 */
5420 			if (del.br_blockcount > off) {
5421 				del.br_blockcount -= off;
5422 				del.br_startoff += off;
5423 				del.br_startblock += off;
5424 			} else if (del.br_startoff == start &&
5425 				   (del.br_state == XFS_EXT_UNWRITTEN ||
5426 				    tp->t_blk_res == 0)) {
5427 				/*
5428 				 * Can't make it unwritten.  There isn't
5429 				 * a full extent here so just skip it.
5430 				 */
5431 				ASSERT(end >= del.br_blockcount);
5432 				end -= del.br_blockcount;
5433 				if (got.br_startoff > end &&
5434 				    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5435 					done = true;
5436 					break;
5437 				}
5438 				continue;
5439 			} else if (del.br_state == XFS_EXT_UNWRITTEN) {
5440 				struct xfs_bmbt_irec	prev;
5441 				xfs_fileoff_t		unwrite_start;
5442 
5443 				/*
5444 				 * This one is already unwritten.
5445 				 * It must have a written left neighbor.
5446 				 * Unwrite the killed part of that one and
5447 				 * try again.
5448 				 */
5449 				if (!xfs_iext_prev_extent(ifp, &icur, &prev))
5450 					ASSERT(0);
5451 				ASSERT(prev.br_state == XFS_EXT_NORM);
5452 				ASSERT(!isnullstartblock(prev.br_startblock));
5453 				ASSERT(del.br_startblock ==
5454 				       prev.br_startblock + prev.br_blockcount);
5455 				unwrite_start = max3(start,
5456 						     del.br_startoff - mod,
5457 						     prev.br_startoff);
5458 				mod = unwrite_start - prev.br_startoff;
5459 				prev.br_startoff = unwrite_start;
5460 				prev.br_startblock += mod;
5461 				prev.br_blockcount -= mod;
5462 				prev.br_state = XFS_EXT_UNWRITTEN;
5463 				error = xfs_bmap_add_extent_unwritten_real(tp,
5464 						ip, whichfork, &icur, &cur,
5465 						&prev, &logflags);
5466 				if (error)
5467 					goto error0;
5468 				goto nodelete;
5469 			} else {
5470 				ASSERT(del.br_state == XFS_EXT_NORM);
5471 				del.br_state = XFS_EXT_UNWRITTEN;
5472 				error = xfs_bmap_add_extent_unwritten_real(tp,
5473 						ip, whichfork, &icur, &cur,
5474 						&del, &logflags);
5475 				if (error)
5476 					goto error0;
5477 				goto nodelete;
5478 			}
5479 		}
5480 
5481 delete:
5482 		if (wasdel) {
5483 			error = xfs_bmap_del_extent_delay(ip, whichfork, &icur,
5484 					&got, &del);
5485 		} else {
5486 			error = xfs_bmap_del_extent_real(ip, tp, &icur, cur,
5487 					&del, &tmp_logflags, whichfork,
5488 					flags);
5489 			logflags |= tmp_logflags;
5490 		}
5491 
5492 		if (error)
5493 			goto error0;
5494 
5495 		end = del.br_startoff - 1;
5496 nodelete:
5497 		/*
5498 		 * If not done go on to the next (previous) record.
5499 		 */
5500 		if (end != (xfs_fileoff_t)-1 && end >= start) {
5501 			if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5502 			    (got.br_startoff > end &&
5503 			     !xfs_iext_prev_extent(ifp, &icur, &got))) {
5504 				done = true;
5505 				break;
5506 			}
5507 			extno++;
5508 		}
5509 	}
5510 	if (done || end == (xfs_fileoff_t)-1 || end < start)
5511 		*rlen = 0;
5512 	else
5513 		*rlen = end - start + 1;
5514 
5515 	/*
5516 	 * Convert to a btree if necessary.
5517 	 */
5518 	if (xfs_bmap_needs_btree(ip, whichfork)) {
5519 		ASSERT(cur == NULL);
5520 		error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
5521 				&tmp_logflags, whichfork);
5522 		logflags |= tmp_logflags;
5523 	} else {
5524 		error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags,
5525 			whichfork);
5526 	}
5527 
5528 error0:
5529 	/*
5530 	 * Log everything.  Do this after conversion, there's no point in
5531 	 * logging the extent records if we've converted to btree format.
5532 	 */
5533 	if ((logflags & xfs_ilog_fext(whichfork)) &&
5534 	    ifp->if_format != XFS_DINODE_FMT_EXTENTS)
5535 		logflags &= ~xfs_ilog_fext(whichfork);
5536 	else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5537 		 ifp->if_format != XFS_DINODE_FMT_BTREE)
5538 		logflags &= ~xfs_ilog_fbroot(whichfork);
5539 	/*
5540 	 * Log inode even in the error case, if the transaction
5541 	 * is dirty we'll need to shut down the filesystem.
5542 	 */
5543 	if (logflags)
5544 		xfs_trans_log_inode(tp, ip, logflags);
5545 	if (cur) {
5546 		if (!error)
5547 			cur->bc_ino.allocated = 0;
5548 		xfs_btree_del_cursor(cur, error);
5549 	}
5550 	return error;
5551 }
5552 
5553 /* Unmap a range of a file. */
5554 int
5555 xfs_bunmapi(
5556 	xfs_trans_t		*tp,
5557 	struct xfs_inode	*ip,
5558 	xfs_fileoff_t		bno,
5559 	xfs_filblks_t		len,
5560 	uint32_t		flags,
5561 	xfs_extnum_t		nexts,
5562 	int			*done)
5563 {
5564 	int			error;
5565 
5566 	error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts);
5567 	*done = (len == 0);
5568 	return error;
5569 }
5570 
5571 /*
5572  * Determine whether an extent shift can be accomplished by a merge with the
5573  * extent that precedes the target hole of the shift.
5574  */
5575 STATIC bool
5576 xfs_bmse_can_merge(
5577 	struct xfs_bmbt_irec	*left,	/* preceding extent */
5578 	struct xfs_bmbt_irec	*got,	/* current extent to shift */
5579 	xfs_fileoff_t		shift)	/* shift fsb */
5580 {
5581 	xfs_fileoff_t		startoff;
5582 
5583 	startoff = got->br_startoff - shift;
5584 
5585 	/*
5586 	 * The extent, once shifted, must be adjacent in-file and on-disk with
5587 	 * the preceding extent.
5588 	 */
5589 	if ((left->br_startoff + left->br_blockcount != startoff) ||
5590 	    (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5591 	    (left->br_state != got->br_state) ||
5592 	    (left->br_blockcount + got->br_blockcount > XFS_MAX_BMBT_EXTLEN))
5593 		return false;
5594 
5595 	return true;
5596 }
5597 
5598 /*
5599  * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5600  * hole in the file. If an extent shift would result in the extent being fully
5601  * adjacent to the extent that currently precedes the hole, we can merge with
5602  * the preceding extent rather than do the shift.
5603  *
5604  * This function assumes the caller has verified a shift-by-merge is possible
5605  * with the provided extents via xfs_bmse_can_merge().
5606  */
5607 STATIC int
5608 xfs_bmse_merge(
5609 	struct xfs_trans		*tp,
5610 	struct xfs_inode		*ip,
5611 	int				whichfork,
5612 	xfs_fileoff_t			shift,		/* shift fsb */
5613 	struct xfs_iext_cursor		*icur,
5614 	struct xfs_bmbt_irec		*got,		/* extent to shift */
5615 	struct xfs_bmbt_irec		*left,		/* preceding extent */
5616 	struct xfs_btree_cur		*cur,
5617 	int				*logflags)	/* output */
5618 {
5619 	struct xfs_ifork		*ifp = xfs_ifork_ptr(ip, whichfork);
5620 	struct xfs_bmbt_irec		new;
5621 	xfs_filblks_t			blockcount;
5622 	int				error, i;
5623 	struct xfs_mount		*mp = ip->i_mount;
5624 
5625 	blockcount = left->br_blockcount + got->br_blockcount;
5626 
5627 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5628 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5629 	ASSERT(xfs_bmse_can_merge(left, got, shift));
5630 
5631 	new = *left;
5632 	new.br_blockcount = blockcount;
5633 
5634 	/*
5635 	 * Update the on-disk extent count, the btree if necessary and log the
5636 	 * inode.
5637 	 */
5638 	ifp->if_nextents--;
5639 	*logflags |= XFS_ILOG_CORE;
5640 	if (!cur) {
5641 		*logflags |= XFS_ILOG_DEXT;
5642 		goto done;
5643 	}
5644 
5645 	/* lookup and remove the extent to merge */
5646 	error = xfs_bmbt_lookup_eq(cur, got, &i);
5647 	if (error)
5648 		return error;
5649 	if (XFS_IS_CORRUPT(mp, i != 1))
5650 		return -EFSCORRUPTED;
5651 
5652 	error = xfs_btree_delete(cur, &i);
5653 	if (error)
5654 		return error;
5655 	if (XFS_IS_CORRUPT(mp, i != 1))
5656 		return -EFSCORRUPTED;
5657 
5658 	/* lookup and update size of the previous extent */
5659 	error = xfs_bmbt_lookup_eq(cur, left, &i);
5660 	if (error)
5661 		return error;
5662 	if (XFS_IS_CORRUPT(mp, i != 1))
5663 		return -EFSCORRUPTED;
5664 
5665 	error = xfs_bmbt_update(cur, &new);
5666 	if (error)
5667 		return error;
5668 
5669 	/* change to extent format if required after extent removal */
5670 	error = xfs_bmap_btree_to_extents(tp, ip, cur, logflags, whichfork);
5671 	if (error)
5672 		return error;
5673 
5674 done:
5675 	xfs_iext_remove(ip, icur, 0);
5676 	xfs_iext_prev(ifp, icur);
5677 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5678 			&new);
5679 
5680 	/* update reverse mapping. rmap functions merge the rmaps for us */
5681 	xfs_rmap_unmap_extent(tp, ip, whichfork, got);
5682 	memcpy(&new, got, sizeof(new));
5683 	new.br_startoff = left->br_startoff + left->br_blockcount;
5684 	xfs_rmap_map_extent(tp, ip, whichfork, &new);
5685 	return 0;
5686 }
5687 
5688 static int
5689 xfs_bmap_shift_update_extent(
5690 	struct xfs_trans	*tp,
5691 	struct xfs_inode	*ip,
5692 	int			whichfork,
5693 	struct xfs_iext_cursor	*icur,
5694 	struct xfs_bmbt_irec	*got,
5695 	struct xfs_btree_cur	*cur,
5696 	int			*logflags,
5697 	xfs_fileoff_t		startoff)
5698 {
5699 	struct xfs_mount	*mp = ip->i_mount;
5700 	struct xfs_bmbt_irec	prev = *got;
5701 	int			error, i;
5702 
5703 	*logflags |= XFS_ILOG_CORE;
5704 
5705 	got->br_startoff = startoff;
5706 
5707 	if (cur) {
5708 		error = xfs_bmbt_lookup_eq(cur, &prev, &i);
5709 		if (error)
5710 			return error;
5711 		if (XFS_IS_CORRUPT(mp, i != 1))
5712 			return -EFSCORRUPTED;
5713 
5714 		error = xfs_bmbt_update(cur, got);
5715 		if (error)
5716 			return error;
5717 	} else {
5718 		*logflags |= XFS_ILOG_DEXT;
5719 	}
5720 
5721 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5722 			got);
5723 
5724 	/* update reverse mapping */
5725 	xfs_rmap_unmap_extent(tp, ip, whichfork, &prev);
5726 	xfs_rmap_map_extent(tp, ip, whichfork, got);
5727 	return 0;
5728 }
5729 
5730 int
5731 xfs_bmap_collapse_extents(
5732 	struct xfs_trans	*tp,
5733 	struct xfs_inode	*ip,
5734 	xfs_fileoff_t		*next_fsb,
5735 	xfs_fileoff_t		offset_shift_fsb,
5736 	bool			*done)
5737 {
5738 	int			whichfork = XFS_DATA_FORK;
5739 	struct xfs_mount	*mp = ip->i_mount;
5740 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
5741 	struct xfs_btree_cur	*cur = NULL;
5742 	struct xfs_bmbt_irec	got, prev;
5743 	struct xfs_iext_cursor	icur;
5744 	xfs_fileoff_t		new_startoff;
5745 	int			error = 0;
5746 	int			logflags = 0;
5747 
5748 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
5749 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
5750 		return -EFSCORRUPTED;
5751 	}
5752 
5753 	if (xfs_is_shutdown(mp))
5754 		return -EIO;
5755 
5756 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL));
5757 
5758 	error = xfs_iread_extents(tp, ip, whichfork);
5759 	if (error)
5760 		return error;
5761 
5762 	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
5763 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5764 		cur->bc_ino.flags = 0;
5765 	}
5766 
5767 	if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5768 		*done = true;
5769 		goto del_cursor;
5770 	}
5771 	if (XFS_IS_CORRUPT(mp, isnullstartblock(got.br_startblock))) {
5772 		error = -EFSCORRUPTED;
5773 		goto del_cursor;
5774 	}
5775 
5776 	new_startoff = got.br_startoff - offset_shift_fsb;
5777 	if (xfs_iext_peek_prev_extent(ifp, &icur, &prev)) {
5778 		if (new_startoff < prev.br_startoff + prev.br_blockcount) {
5779 			error = -EINVAL;
5780 			goto del_cursor;
5781 		}
5782 
5783 		if (xfs_bmse_can_merge(&prev, &got, offset_shift_fsb)) {
5784 			error = xfs_bmse_merge(tp, ip, whichfork,
5785 					offset_shift_fsb, &icur, &got, &prev,
5786 					cur, &logflags);
5787 			if (error)
5788 				goto del_cursor;
5789 			goto done;
5790 		}
5791 	} else {
5792 		if (got.br_startoff < offset_shift_fsb) {
5793 			error = -EINVAL;
5794 			goto del_cursor;
5795 		}
5796 	}
5797 
5798 	error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got,
5799 			cur, &logflags, new_startoff);
5800 	if (error)
5801 		goto del_cursor;
5802 
5803 done:
5804 	if (!xfs_iext_next_extent(ifp, &icur, &got)) {
5805 		*done = true;
5806 		goto del_cursor;
5807 	}
5808 
5809 	*next_fsb = got.br_startoff;
5810 del_cursor:
5811 	if (cur)
5812 		xfs_btree_del_cursor(cur, error);
5813 	if (logflags)
5814 		xfs_trans_log_inode(tp, ip, logflags);
5815 	return error;
5816 }
5817 
5818 /* Make sure we won't be right-shifting an extent past the maximum bound. */
5819 int
5820 xfs_bmap_can_insert_extents(
5821 	struct xfs_inode	*ip,
5822 	xfs_fileoff_t		off,
5823 	xfs_fileoff_t		shift)
5824 {
5825 	struct xfs_bmbt_irec	got;
5826 	int			is_empty;
5827 	int			error = 0;
5828 
5829 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5830 
5831 	if (xfs_is_shutdown(ip->i_mount))
5832 		return -EIO;
5833 
5834 	xfs_ilock(ip, XFS_ILOCK_EXCL);
5835 	error = xfs_bmap_last_extent(NULL, ip, XFS_DATA_FORK, &got, &is_empty);
5836 	if (!error && !is_empty && got.br_startoff >= off &&
5837 	    ((got.br_startoff + shift) & BMBT_STARTOFF_MASK) < got.br_startoff)
5838 		error = -EINVAL;
5839 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
5840 
5841 	return error;
5842 }
5843 
5844 int
5845 xfs_bmap_insert_extents(
5846 	struct xfs_trans	*tp,
5847 	struct xfs_inode	*ip,
5848 	xfs_fileoff_t		*next_fsb,
5849 	xfs_fileoff_t		offset_shift_fsb,
5850 	bool			*done,
5851 	xfs_fileoff_t		stop_fsb)
5852 {
5853 	int			whichfork = XFS_DATA_FORK;
5854 	struct xfs_mount	*mp = ip->i_mount;
5855 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
5856 	struct xfs_btree_cur	*cur = NULL;
5857 	struct xfs_bmbt_irec	got, next;
5858 	struct xfs_iext_cursor	icur;
5859 	xfs_fileoff_t		new_startoff;
5860 	int			error = 0;
5861 	int			logflags = 0;
5862 
5863 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
5864 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
5865 		return -EFSCORRUPTED;
5866 	}
5867 
5868 	if (xfs_is_shutdown(mp))
5869 		return -EIO;
5870 
5871 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL));
5872 
5873 	error = xfs_iread_extents(tp, ip, whichfork);
5874 	if (error)
5875 		return error;
5876 
5877 	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
5878 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5879 		cur->bc_ino.flags = 0;
5880 	}
5881 
5882 	if (*next_fsb == NULLFSBLOCK) {
5883 		xfs_iext_last(ifp, &icur);
5884 		if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5885 		    stop_fsb > got.br_startoff) {
5886 			*done = true;
5887 			goto del_cursor;
5888 		}
5889 	} else {
5890 		if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5891 			*done = true;
5892 			goto del_cursor;
5893 		}
5894 	}
5895 	if (XFS_IS_CORRUPT(mp, isnullstartblock(got.br_startblock))) {
5896 		error = -EFSCORRUPTED;
5897 		goto del_cursor;
5898 	}
5899 
5900 	if (XFS_IS_CORRUPT(mp, stop_fsb > got.br_startoff)) {
5901 		error = -EFSCORRUPTED;
5902 		goto del_cursor;
5903 	}
5904 
5905 	new_startoff = got.br_startoff + offset_shift_fsb;
5906 	if (xfs_iext_peek_next_extent(ifp, &icur, &next)) {
5907 		if (new_startoff + got.br_blockcount > next.br_startoff) {
5908 			error = -EINVAL;
5909 			goto del_cursor;
5910 		}
5911 
5912 		/*
5913 		 * Unlike a left shift (which involves a hole punch), a right
5914 		 * shift does not modify extent neighbors in any way.  We should
5915 		 * never find mergeable extents in this scenario.  Check anyways
5916 		 * and warn if we encounter two extents that could be one.
5917 		 */
5918 		if (xfs_bmse_can_merge(&got, &next, offset_shift_fsb))
5919 			WARN_ON_ONCE(1);
5920 	}
5921 
5922 	error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got,
5923 			cur, &logflags, new_startoff);
5924 	if (error)
5925 		goto del_cursor;
5926 
5927 	if (!xfs_iext_prev_extent(ifp, &icur, &got) ||
5928 	    stop_fsb >= got.br_startoff + got.br_blockcount) {
5929 		*done = true;
5930 		goto del_cursor;
5931 	}
5932 
5933 	*next_fsb = got.br_startoff;
5934 del_cursor:
5935 	if (cur)
5936 		xfs_btree_del_cursor(cur, error);
5937 	if (logflags)
5938 		xfs_trans_log_inode(tp, ip, logflags);
5939 	return error;
5940 }
5941 
5942 /*
5943  * Splits an extent into two extents at split_fsb block such that it is the
5944  * first block of the current_ext. @ext is a target extent to be split.
5945  * @split_fsb is a block where the extents is split.  If split_fsb lies in a
5946  * hole or the first block of extents, just return 0.
5947  */
5948 int
5949 xfs_bmap_split_extent(
5950 	struct xfs_trans	*tp,
5951 	struct xfs_inode	*ip,
5952 	xfs_fileoff_t		split_fsb)
5953 {
5954 	int				whichfork = XFS_DATA_FORK;
5955 	struct xfs_ifork		*ifp = xfs_ifork_ptr(ip, whichfork);
5956 	struct xfs_btree_cur		*cur = NULL;
5957 	struct xfs_bmbt_irec		got;
5958 	struct xfs_bmbt_irec		new; /* split extent */
5959 	struct xfs_mount		*mp = ip->i_mount;
5960 	xfs_fsblock_t			gotblkcnt; /* new block count for got */
5961 	struct xfs_iext_cursor		icur;
5962 	int				error = 0;
5963 	int				logflags = 0;
5964 	int				i = 0;
5965 
5966 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
5967 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
5968 		return -EFSCORRUPTED;
5969 	}
5970 
5971 	if (xfs_is_shutdown(mp))
5972 		return -EIO;
5973 
5974 	/* Read in all the extents */
5975 	error = xfs_iread_extents(tp, ip, whichfork);
5976 	if (error)
5977 		return error;
5978 
5979 	/*
5980 	 * If there are not extents, or split_fsb lies in a hole we are done.
5981 	 */
5982 	if (!xfs_iext_lookup_extent(ip, ifp, split_fsb, &icur, &got) ||
5983 	    got.br_startoff >= split_fsb)
5984 		return 0;
5985 
5986 	gotblkcnt = split_fsb - got.br_startoff;
5987 	new.br_startoff = split_fsb;
5988 	new.br_startblock = got.br_startblock + gotblkcnt;
5989 	new.br_blockcount = got.br_blockcount - gotblkcnt;
5990 	new.br_state = got.br_state;
5991 
5992 	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
5993 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5994 		cur->bc_ino.flags = 0;
5995 		error = xfs_bmbt_lookup_eq(cur, &got, &i);
5996 		if (error)
5997 			goto del_cursor;
5998 		if (XFS_IS_CORRUPT(mp, i != 1)) {
5999 			error = -EFSCORRUPTED;
6000 			goto del_cursor;
6001 		}
6002 	}
6003 
6004 	got.br_blockcount = gotblkcnt;
6005 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), &icur,
6006 			&got);
6007 
6008 	logflags = XFS_ILOG_CORE;
6009 	if (cur) {
6010 		error = xfs_bmbt_update(cur, &got);
6011 		if (error)
6012 			goto del_cursor;
6013 	} else
6014 		logflags |= XFS_ILOG_DEXT;
6015 
6016 	/* Add new extent */
6017 	xfs_iext_next(ifp, &icur);
6018 	xfs_iext_insert(ip, &icur, &new, 0);
6019 	ifp->if_nextents++;
6020 
6021 	if (cur) {
6022 		error = xfs_bmbt_lookup_eq(cur, &new, &i);
6023 		if (error)
6024 			goto del_cursor;
6025 		if (XFS_IS_CORRUPT(mp, i != 0)) {
6026 			error = -EFSCORRUPTED;
6027 			goto del_cursor;
6028 		}
6029 		error = xfs_btree_insert(cur, &i);
6030 		if (error)
6031 			goto del_cursor;
6032 		if (XFS_IS_CORRUPT(mp, i != 1)) {
6033 			error = -EFSCORRUPTED;
6034 			goto del_cursor;
6035 		}
6036 	}
6037 
6038 	/*
6039 	 * Convert to a btree if necessary.
6040 	 */
6041 	if (xfs_bmap_needs_btree(ip, whichfork)) {
6042 		int tmp_logflags; /* partial log flag return val */
6043 
6044 		ASSERT(cur == NULL);
6045 		error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
6046 				&tmp_logflags, whichfork);
6047 		logflags |= tmp_logflags;
6048 	}
6049 
6050 del_cursor:
6051 	if (cur) {
6052 		cur->bc_ino.allocated = 0;
6053 		xfs_btree_del_cursor(cur, error);
6054 	}
6055 
6056 	if (logflags)
6057 		xfs_trans_log_inode(tp, ip, logflags);
6058 	return error;
6059 }
6060 
6061 /* Deferred mapping is only for real extents in the data fork. */
6062 static bool
6063 xfs_bmap_is_update_needed(
6064 	struct xfs_bmbt_irec	*bmap)
6065 {
6066 	return  bmap->br_startblock != HOLESTARTBLOCK &&
6067 		bmap->br_startblock != DELAYSTARTBLOCK;
6068 }
6069 
6070 /* Record a bmap intent. */
6071 static int
6072 __xfs_bmap_add(
6073 	struct xfs_trans		*tp,
6074 	enum xfs_bmap_intent_type	type,
6075 	struct xfs_inode		*ip,
6076 	int				whichfork,
6077 	struct xfs_bmbt_irec		*bmap)
6078 {
6079 	struct xfs_bmap_intent		*bi;
6080 
6081 	trace_xfs_bmap_defer(tp->t_mountp,
6082 			XFS_FSB_TO_AGNO(tp->t_mountp, bmap->br_startblock),
6083 			type,
6084 			XFS_FSB_TO_AGBNO(tp->t_mountp, bmap->br_startblock),
6085 			ip->i_ino, whichfork,
6086 			bmap->br_startoff,
6087 			bmap->br_blockcount,
6088 			bmap->br_state);
6089 
6090 	bi = kmem_cache_alloc(xfs_bmap_intent_cache, GFP_NOFS | __GFP_NOFAIL);
6091 	INIT_LIST_HEAD(&bi->bi_list);
6092 	bi->bi_type = type;
6093 	bi->bi_owner = ip;
6094 	bi->bi_whichfork = whichfork;
6095 	bi->bi_bmap = *bmap;
6096 
6097 	xfs_bmap_update_get_group(tp->t_mountp, bi);
6098 	xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list);
6099 	return 0;
6100 }
6101 
6102 /* Map an extent into a file. */
6103 void
6104 xfs_bmap_map_extent(
6105 	struct xfs_trans	*tp,
6106 	struct xfs_inode	*ip,
6107 	struct xfs_bmbt_irec	*PREV)
6108 {
6109 	if (!xfs_bmap_is_update_needed(PREV))
6110 		return;
6111 
6112 	__xfs_bmap_add(tp, XFS_BMAP_MAP, ip, XFS_DATA_FORK, PREV);
6113 }
6114 
6115 /* Unmap an extent out of a file. */
6116 void
6117 xfs_bmap_unmap_extent(
6118 	struct xfs_trans	*tp,
6119 	struct xfs_inode	*ip,
6120 	struct xfs_bmbt_irec	*PREV)
6121 {
6122 	if (!xfs_bmap_is_update_needed(PREV))
6123 		return;
6124 
6125 	__xfs_bmap_add(tp, XFS_BMAP_UNMAP, ip, XFS_DATA_FORK, PREV);
6126 }
6127 
6128 /*
6129  * Process one of the deferred bmap operations.  We pass back the
6130  * btree cursor to maintain our lock on the bmapbt between calls.
6131  */
6132 int
6133 xfs_bmap_finish_one(
6134 	struct xfs_trans		*tp,
6135 	struct xfs_bmap_intent		*bi)
6136 {
6137 	struct xfs_bmbt_irec		*bmap = &bi->bi_bmap;
6138 	int				error = 0;
6139 
6140 	ASSERT(tp->t_highest_agno == NULLAGNUMBER);
6141 
6142 	trace_xfs_bmap_deferred(tp->t_mountp,
6143 			XFS_FSB_TO_AGNO(tp->t_mountp, bmap->br_startblock),
6144 			bi->bi_type,
6145 			XFS_FSB_TO_AGBNO(tp->t_mountp, bmap->br_startblock),
6146 			bi->bi_owner->i_ino, bi->bi_whichfork,
6147 			bmap->br_startoff, bmap->br_blockcount,
6148 			bmap->br_state);
6149 
6150 	if (WARN_ON_ONCE(bi->bi_whichfork != XFS_DATA_FORK))
6151 		return -EFSCORRUPTED;
6152 
6153 	if (XFS_TEST_ERROR(false, tp->t_mountp,
6154 			XFS_ERRTAG_BMAP_FINISH_ONE))
6155 		return -EIO;
6156 
6157 	switch (bi->bi_type) {
6158 	case XFS_BMAP_MAP:
6159 		error = xfs_bmapi_remap(tp, bi->bi_owner, bmap->br_startoff,
6160 				bmap->br_blockcount, bmap->br_startblock, 0);
6161 		bmap->br_blockcount = 0;
6162 		break;
6163 	case XFS_BMAP_UNMAP:
6164 		error = __xfs_bunmapi(tp, bi->bi_owner, bmap->br_startoff,
6165 				&bmap->br_blockcount, XFS_BMAPI_REMAP, 1);
6166 		break;
6167 	default:
6168 		ASSERT(0);
6169 		error = -EFSCORRUPTED;
6170 	}
6171 
6172 	return error;
6173 }
6174 
6175 /* Check that an inode's extent does not have invalid flags or bad ranges. */
6176 xfs_failaddr_t
6177 xfs_bmap_validate_extent(
6178 	struct xfs_inode	*ip,
6179 	int			whichfork,
6180 	struct xfs_bmbt_irec	*irec)
6181 {
6182 	struct xfs_mount	*mp = ip->i_mount;
6183 
6184 	if (!xfs_verify_fileext(mp, irec->br_startoff, irec->br_blockcount))
6185 		return __this_address;
6186 
6187 	if (XFS_IS_REALTIME_INODE(ip) && whichfork == XFS_DATA_FORK) {
6188 		if (!xfs_verify_rtext(mp, irec->br_startblock,
6189 					  irec->br_blockcount))
6190 			return __this_address;
6191 	} else {
6192 		if (!xfs_verify_fsbext(mp, irec->br_startblock,
6193 					   irec->br_blockcount))
6194 			return __this_address;
6195 	}
6196 	if (irec->br_state != XFS_EXT_NORM && whichfork != XFS_DATA_FORK)
6197 		return __this_address;
6198 	return NULL;
6199 }
6200 
6201 int __init
6202 xfs_bmap_intent_init_cache(void)
6203 {
6204 	xfs_bmap_intent_cache = kmem_cache_create("xfs_bmap_intent",
6205 			sizeof(struct xfs_bmap_intent),
6206 			0, 0, NULL);
6207 
6208 	return xfs_bmap_intent_cache != NULL ? 0 : -ENOMEM;
6209 }
6210 
6211 void
6212 xfs_bmap_intent_destroy_cache(void)
6213 {
6214 	kmem_cache_destroy(xfs_bmap_intent_cache);
6215 	xfs_bmap_intent_cache = NULL;
6216 }
6217