xref: /openbmc/linux/fs/xfs/libxfs/xfs_bmap.c (revision 7cc39531)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4  * All Rights Reserved.
5  */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_bit.h"
13 #include "xfs_sb.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_dir2.h"
17 #include "xfs_inode.h"
18 #include "xfs_btree.h"
19 #include "xfs_trans.h"
20 #include "xfs_alloc.h"
21 #include "xfs_bmap.h"
22 #include "xfs_bmap_util.h"
23 #include "xfs_bmap_btree.h"
24 #include "xfs_rtalloc.h"
25 #include "xfs_errortag.h"
26 #include "xfs_error.h"
27 #include "xfs_quota.h"
28 #include "xfs_trans_space.h"
29 #include "xfs_buf_item.h"
30 #include "xfs_trace.h"
31 #include "xfs_attr_leaf.h"
32 #include "xfs_filestream.h"
33 #include "xfs_rmap.h"
34 #include "xfs_ag.h"
35 #include "xfs_ag_resv.h"
36 #include "xfs_refcount.h"
37 #include "xfs_icache.h"
38 #include "xfs_iomap.h"
39 
40 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) %Ld",
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 	xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
576 	xfs_free_extent_later(cur->bc_tp, cbno, 1, &oinfo);
577 	ip->i_nblocks--;
578 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
579 	xfs_trans_binval(tp, cbp);
580 	if (cur->bc_levels[0].bp == cbp)
581 		cur->bc_levels[0].bp = NULL;
582 	xfs_iroot_realloc(ip, -1, whichfork);
583 	ASSERT(ifp->if_broot == NULL);
584 	ifp->if_format = XFS_DINODE_FMT_EXTENTS;
585 	*logflagsp |= XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
586 	return 0;
587 }
588 
589 /*
590  * Convert an extents-format file into a btree-format file.
591  * The new file will have a root block (in the inode) and a single child block.
592  */
593 STATIC int					/* error */
594 xfs_bmap_extents_to_btree(
595 	struct xfs_trans	*tp,		/* transaction pointer */
596 	struct xfs_inode	*ip,		/* incore inode pointer */
597 	struct xfs_btree_cur	**curp,		/* cursor returned to caller */
598 	int			wasdel,		/* converting a delayed alloc */
599 	int			*logflagsp,	/* inode logging flags */
600 	int			whichfork)	/* data or attr fork */
601 {
602 	struct xfs_btree_block	*ablock;	/* allocated (child) bt block */
603 	struct xfs_buf		*abp;		/* buffer for ablock */
604 	struct xfs_alloc_arg	args;		/* allocation arguments */
605 	struct xfs_bmbt_rec	*arp;		/* child record pointer */
606 	struct xfs_btree_block	*block;		/* btree root block */
607 	struct xfs_btree_cur	*cur;		/* bmap btree cursor */
608 	int			error;		/* error return value */
609 	struct xfs_ifork	*ifp;		/* inode fork pointer */
610 	struct xfs_bmbt_key	*kp;		/* root block key pointer */
611 	struct xfs_mount	*mp;		/* mount structure */
612 	xfs_bmbt_ptr_t		*pp;		/* root block address pointer */
613 	struct xfs_iext_cursor	icur;
614 	struct xfs_bmbt_irec	rec;
615 	xfs_extnum_t		cnt = 0;
616 
617 	mp = ip->i_mount;
618 	ASSERT(whichfork != XFS_COW_FORK);
619 	ifp = xfs_ifork_ptr(ip, whichfork);
620 	ASSERT(ifp->if_format == XFS_DINODE_FMT_EXTENTS);
621 
622 	/*
623 	 * Make space in the inode incore. This needs to be undone if we fail
624 	 * to expand the root.
625 	 */
626 	xfs_iroot_realloc(ip, 1, whichfork);
627 
628 	/*
629 	 * Fill in the root.
630 	 */
631 	block = ifp->if_broot;
632 	xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
633 				 XFS_BTNUM_BMAP, 1, 1, ip->i_ino,
634 				 XFS_BTREE_LONG_PTRS);
635 	/*
636 	 * Need a cursor.  Can't allocate until bb_level is filled in.
637 	 */
638 	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
639 	cur->bc_ino.flags = wasdel ? XFS_BTCUR_BMBT_WASDEL : 0;
640 	/*
641 	 * Convert to a btree with two levels, one record in root.
642 	 */
643 	ifp->if_format = XFS_DINODE_FMT_BTREE;
644 	memset(&args, 0, sizeof(args));
645 	args.tp = tp;
646 	args.mp = mp;
647 	xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork);
648 	if (tp->t_firstblock == NULLFSBLOCK) {
649 		args.type = XFS_ALLOCTYPE_START_BNO;
650 		args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
651 	} else if (tp->t_flags & XFS_TRANS_LOWMODE) {
652 		args.type = XFS_ALLOCTYPE_START_BNO;
653 		args.fsbno = tp->t_firstblock;
654 	} else {
655 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
656 		args.fsbno = tp->t_firstblock;
657 	}
658 	args.minlen = args.maxlen = args.prod = 1;
659 	args.wasdel = wasdel;
660 	*logflagsp = 0;
661 	error = xfs_alloc_vextent(&args);
662 	if (error)
663 		goto out_root_realloc;
664 
665 	if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
666 		error = -ENOSPC;
667 		goto out_root_realloc;
668 	}
669 
670 	/*
671 	 * Allocation can't fail, the space was reserved.
672 	 */
673 	ASSERT(tp->t_firstblock == NULLFSBLOCK ||
674 	       args.agno >= XFS_FSB_TO_AGNO(mp, tp->t_firstblock));
675 	tp->t_firstblock = args.fsbno;
676 	cur->bc_ino.allocated++;
677 	ip->i_nblocks++;
678 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
679 	error = xfs_trans_get_buf(tp, mp->m_ddev_targp,
680 			XFS_FSB_TO_DADDR(mp, args.fsbno),
681 			mp->m_bsize, 0, &abp);
682 	if (error)
683 		goto out_unreserve_dquot;
684 
685 	/*
686 	 * Fill in the child block.
687 	 */
688 	abp->b_ops = &xfs_bmbt_buf_ops;
689 	ablock = XFS_BUF_TO_BLOCK(abp);
690 	xfs_btree_init_block_int(mp, ablock, xfs_buf_daddr(abp),
691 				XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
692 				XFS_BTREE_LONG_PTRS);
693 
694 	for_each_xfs_iext(ifp, &icur, &rec) {
695 		if (isnullstartblock(rec.br_startblock))
696 			continue;
697 		arp = XFS_BMBT_REC_ADDR(mp, ablock, 1 + cnt);
698 		xfs_bmbt_disk_set_all(arp, &rec);
699 		cnt++;
700 	}
701 	ASSERT(cnt == ifp->if_nextents);
702 	xfs_btree_set_numrecs(ablock, cnt);
703 
704 	/*
705 	 * Fill in the root key and pointer.
706 	 */
707 	kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
708 	arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
709 	kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
710 	pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
711 						be16_to_cpu(block->bb_level)));
712 	*pp = cpu_to_be64(args.fsbno);
713 
714 	/*
715 	 * Do all this logging at the end so that
716 	 * the root is at the right level.
717 	 */
718 	xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
719 	xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
720 	ASSERT(*curp == NULL);
721 	*curp = cur;
722 	*logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
723 	return 0;
724 
725 out_unreserve_dquot:
726 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
727 out_root_realloc:
728 	xfs_iroot_realloc(ip, -1, whichfork);
729 	ifp->if_format = XFS_DINODE_FMT_EXTENTS;
730 	ASSERT(ifp->if_broot == NULL);
731 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
732 
733 	return error;
734 }
735 
736 /*
737  * Convert a local file to an extents file.
738  * This code is out of bounds for data forks of regular files,
739  * since the file data needs to get logged so things will stay consistent.
740  * (The bmap-level manipulations are ok, though).
741  */
742 void
743 xfs_bmap_local_to_extents_empty(
744 	struct xfs_trans	*tp,
745 	struct xfs_inode	*ip,
746 	int			whichfork)
747 {
748 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
749 
750 	ASSERT(whichfork != XFS_COW_FORK);
751 	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
752 	ASSERT(ifp->if_bytes == 0);
753 	ASSERT(ifp->if_nextents == 0);
754 
755 	xfs_bmap_forkoff_reset(ip, whichfork);
756 	ifp->if_u1.if_root = NULL;
757 	ifp->if_height = 0;
758 	ifp->if_format = XFS_DINODE_FMT_EXTENTS;
759 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
760 }
761 
762 
763 STATIC int				/* error */
764 xfs_bmap_local_to_extents(
765 	xfs_trans_t	*tp,		/* transaction pointer */
766 	xfs_inode_t	*ip,		/* incore inode pointer */
767 	xfs_extlen_t	total,		/* total blocks needed by transaction */
768 	int		*logflagsp,	/* inode logging flags */
769 	int		whichfork,
770 	void		(*init_fn)(struct xfs_trans *tp,
771 				   struct xfs_buf *bp,
772 				   struct xfs_inode *ip,
773 				   struct xfs_ifork *ifp))
774 {
775 	int		error = 0;
776 	int		flags;		/* logging flags returned */
777 	struct xfs_ifork *ifp;		/* inode fork pointer */
778 	xfs_alloc_arg_t	args;		/* allocation arguments */
779 	struct xfs_buf	*bp;		/* buffer for extent block */
780 	struct xfs_bmbt_irec rec;
781 	struct xfs_iext_cursor icur;
782 
783 	/*
784 	 * We don't want to deal with the case of keeping inode data inline yet.
785 	 * So sending the data fork of a regular inode is invalid.
786 	 */
787 	ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
788 	ifp = xfs_ifork_ptr(ip, whichfork);
789 	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
790 
791 	if (!ifp->if_bytes) {
792 		xfs_bmap_local_to_extents_empty(tp, ip, whichfork);
793 		flags = XFS_ILOG_CORE;
794 		goto done;
795 	}
796 
797 	flags = 0;
798 	error = 0;
799 	memset(&args, 0, sizeof(args));
800 	args.tp = tp;
801 	args.mp = ip->i_mount;
802 	xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0);
803 	/*
804 	 * Allocate a block.  We know we need only one, since the
805 	 * file currently fits in an inode.
806 	 */
807 	if (tp->t_firstblock == NULLFSBLOCK) {
808 		args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
809 		args.type = XFS_ALLOCTYPE_START_BNO;
810 	} else {
811 		args.fsbno = tp->t_firstblock;
812 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
813 	}
814 	args.total = total;
815 	args.minlen = args.maxlen = args.prod = 1;
816 	error = xfs_alloc_vextent(&args);
817 	if (error)
818 		goto done;
819 
820 	/* Can't fail, the space was reserved. */
821 	ASSERT(args.fsbno != NULLFSBLOCK);
822 	ASSERT(args.len == 1);
823 	tp->t_firstblock = args.fsbno;
824 	error = xfs_trans_get_buf(tp, args.mp->m_ddev_targp,
825 			XFS_FSB_TO_DADDR(args.mp, args.fsbno),
826 			args.mp->m_bsize, 0, &bp);
827 	if (error)
828 		goto done;
829 
830 	/*
831 	 * Initialize the block, copy the data and log the remote buffer.
832 	 *
833 	 * The callout is responsible for logging because the remote format
834 	 * might differ from the local format and thus we don't know how much to
835 	 * log here. Note that init_fn must also set the buffer log item type
836 	 * correctly.
837 	 */
838 	init_fn(tp, bp, ip, ifp);
839 
840 	/* account for the change in fork size */
841 	xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
842 	xfs_bmap_local_to_extents_empty(tp, ip, whichfork);
843 	flags |= XFS_ILOG_CORE;
844 
845 	ifp->if_u1.if_root = NULL;
846 	ifp->if_height = 0;
847 
848 	rec.br_startoff = 0;
849 	rec.br_startblock = args.fsbno;
850 	rec.br_blockcount = 1;
851 	rec.br_state = XFS_EXT_NORM;
852 	xfs_iext_first(ifp, &icur);
853 	xfs_iext_insert(ip, &icur, &rec, 0);
854 
855 	ifp->if_nextents = 1;
856 	ip->i_nblocks = 1;
857 	xfs_trans_mod_dquot_byino(tp, ip,
858 		XFS_TRANS_DQ_BCOUNT, 1L);
859 	flags |= xfs_ilog_fext(whichfork);
860 
861 done:
862 	*logflagsp = flags;
863 	return error;
864 }
865 
866 /*
867  * Called from xfs_bmap_add_attrfork to handle btree format files.
868  */
869 STATIC int					/* error */
870 xfs_bmap_add_attrfork_btree(
871 	xfs_trans_t		*tp,		/* transaction pointer */
872 	xfs_inode_t		*ip,		/* incore inode pointer */
873 	int			*flags)		/* inode logging flags */
874 {
875 	struct xfs_btree_block	*block = ip->i_df.if_broot;
876 	struct xfs_btree_cur	*cur;		/* btree cursor */
877 	int			error;		/* error return value */
878 	xfs_mount_t		*mp;		/* file system mount struct */
879 	int			stat;		/* newroot status */
880 
881 	mp = ip->i_mount;
882 
883 	if (XFS_BMAP_BMDR_SPACE(block) <= xfs_inode_data_fork_size(ip))
884 		*flags |= XFS_ILOG_DBROOT;
885 	else {
886 		cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
887 		error = xfs_bmbt_lookup_first(cur, &stat);
888 		if (error)
889 			goto error0;
890 		/* must be at least one entry */
891 		if (XFS_IS_CORRUPT(mp, stat != 1)) {
892 			error = -EFSCORRUPTED;
893 			goto error0;
894 		}
895 		if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
896 			goto error0;
897 		if (stat == 0) {
898 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
899 			return -ENOSPC;
900 		}
901 		cur->bc_ino.allocated = 0;
902 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
903 	}
904 	return 0;
905 error0:
906 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
907 	return error;
908 }
909 
910 /*
911  * Called from xfs_bmap_add_attrfork to handle extents format files.
912  */
913 STATIC int					/* error */
914 xfs_bmap_add_attrfork_extents(
915 	struct xfs_trans	*tp,		/* transaction pointer */
916 	struct xfs_inode	*ip,		/* incore inode pointer */
917 	int			*flags)		/* inode logging flags */
918 {
919 	struct xfs_btree_cur	*cur;		/* bmap btree cursor */
920 	int			error;		/* error return value */
921 
922 	if (ip->i_df.if_nextents * sizeof(struct xfs_bmbt_rec) <=
923 	    xfs_inode_data_fork_size(ip))
924 		return 0;
925 	cur = NULL;
926 	error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, flags,
927 					  XFS_DATA_FORK);
928 	if (cur) {
929 		cur->bc_ino.allocated = 0;
930 		xfs_btree_del_cursor(cur, error);
931 	}
932 	return error;
933 }
934 
935 /*
936  * Called from xfs_bmap_add_attrfork to handle local format files. Each
937  * different data fork content type needs a different callout to do the
938  * conversion. Some are basic and only require special block initialisation
939  * callouts for the data formating, others (directories) are so specialised they
940  * handle everything themselves.
941  *
942  * XXX (dgc): investigate whether directory conversion can use the generic
943  * formatting callout. It should be possible - it's just a very complex
944  * formatter.
945  */
946 STATIC int					/* error */
947 xfs_bmap_add_attrfork_local(
948 	struct xfs_trans	*tp,		/* transaction pointer */
949 	struct xfs_inode	*ip,		/* incore inode pointer */
950 	int			*flags)		/* inode logging flags */
951 {
952 	struct xfs_da_args	dargs;		/* args for dir/attr code */
953 
954 	if (ip->i_df.if_bytes <= xfs_inode_data_fork_size(ip))
955 		return 0;
956 
957 	if (S_ISDIR(VFS_I(ip)->i_mode)) {
958 		memset(&dargs, 0, sizeof(dargs));
959 		dargs.geo = ip->i_mount->m_dir_geo;
960 		dargs.dp = ip;
961 		dargs.total = dargs.geo->fsbcount;
962 		dargs.whichfork = XFS_DATA_FORK;
963 		dargs.trans = tp;
964 		return xfs_dir2_sf_to_block(&dargs);
965 	}
966 
967 	if (S_ISLNK(VFS_I(ip)->i_mode))
968 		return xfs_bmap_local_to_extents(tp, ip, 1, flags,
969 						 XFS_DATA_FORK,
970 						 xfs_symlink_local_to_remote);
971 
972 	/* should only be called for types that support local format data */
973 	ASSERT(0);
974 	return -EFSCORRUPTED;
975 }
976 
977 /*
978  * Set an inode attr fork offset based on the format of the data fork.
979  */
980 static int
981 xfs_bmap_set_attrforkoff(
982 	struct xfs_inode	*ip,
983 	int			size,
984 	int			*version)
985 {
986 	int			default_size = xfs_default_attroffset(ip) >> 3;
987 
988 	switch (ip->i_df.if_format) {
989 	case XFS_DINODE_FMT_DEV:
990 		ip->i_forkoff = default_size;
991 		break;
992 	case XFS_DINODE_FMT_LOCAL:
993 	case XFS_DINODE_FMT_EXTENTS:
994 	case XFS_DINODE_FMT_BTREE:
995 		ip->i_forkoff = xfs_attr_shortform_bytesfit(ip, size);
996 		if (!ip->i_forkoff)
997 			ip->i_forkoff = default_size;
998 		else if (xfs_has_attr2(ip->i_mount) && version)
999 			*version = 2;
1000 		break;
1001 	default:
1002 		ASSERT(0);
1003 		return -EINVAL;
1004 	}
1005 
1006 	return 0;
1007 }
1008 
1009 /*
1010  * Convert inode from non-attributed to attributed.
1011  * Must not be in a transaction, ip must not be locked.
1012  */
1013 int						/* error code */
1014 xfs_bmap_add_attrfork(
1015 	xfs_inode_t		*ip,		/* incore inode pointer */
1016 	int			size,		/* space new attribute needs */
1017 	int			rsvd)		/* xact may use reserved blks */
1018 {
1019 	xfs_mount_t		*mp;		/* mount structure */
1020 	xfs_trans_t		*tp;		/* transaction pointer */
1021 	int			blks;		/* space reservation */
1022 	int			version = 1;	/* superblock attr version */
1023 	int			logflags;	/* logging flags */
1024 	int			error;		/* error return value */
1025 
1026 	ASSERT(xfs_inode_has_attr_fork(ip) == 0);
1027 
1028 	mp = ip->i_mount;
1029 	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1030 
1031 	blks = XFS_ADDAFORK_SPACE_RES(mp);
1032 
1033 	error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_addafork, blks, 0,
1034 			rsvd, &tp);
1035 	if (error)
1036 		return error;
1037 	if (xfs_inode_has_attr_fork(ip))
1038 		goto trans_cancel;
1039 
1040 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1041 	error = xfs_bmap_set_attrforkoff(ip, size, &version);
1042 	if (error)
1043 		goto trans_cancel;
1044 
1045 	xfs_ifork_init_attr(ip, XFS_DINODE_FMT_EXTENTS, 0);
1046 	logflags = 0;
1047 	switch (ip->i_df.if_format) {
1048 	case XFS_DINODE_FMT_LOCAL:
1049 		error = xfs_bmap_add_attrfork_local(tp, ip, &logflags);
1050 		break;
1051 	case XFS_DINODE_FMT_EXTENTS:
1052 		error = xfs_bmap_add_attrfork_extents(tp, ip, &logflags);
1053 		break;
1054 	case XFS_DINODE_FMT_BTREE:
1055 		error = xfs_bmap_add_attrfork_btree(tp, ip, &logflags);
1056 		break;
1057 	default:
1058 		error = 0;
1059 		break;
1060 	}
1061 	if (logflags)
1062 		xfs_trans_log_inode(tp, ip, logflags);
1063 	if (error)
1064 		goto trans_cancel;
1065 	if (!xfs_has_attr(mp) ||
1066 	   (!xfs_has_attr2(mp) && version == 2)) {
1067 		bool log_sb = false;
1068 
1069 		spin_lock(&mp->m_sb_lock);
1070 		if (!xfs_has_attr(mp)) {
1071 			xfs_add_attr(mp);
1072 			log_sb = true;
1073 		}
1074 		if (!xfs_has_attr2(mp) && version == 2) {
1075 			xfs_add_attr2(mp);
1076 			log_sb = true;
1077 		}
1078 		spin_unlock(&mp->m_sb_lock);
1079 		if (log_sb)
1080 			xfs_log_sb(tp);
1081 	}
1082 
1083 	error = xfs_trans_commit(tp);
1084 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1085 	return error;
1086 
1087 trans_cancel:
1088 	xfs_trans_cancel(tp);
1089 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1090 	return error;
1091 }
1092 
1093 /*
1094  * Internal and external extent tree search functions.
1095  */
1096 
1097 struct xfs_iread_state {
1098 	struct xfs_iext_cursor	icur;
1099 	xfs_extnum_t		loaded;
1100 };
1101 
1102 /* Stuff every bmbt record from this block into the incore extent map. */
1103 static int
1104 xfs_iread_bmbt_block(
1105 	struct xfs_btree_cur	*cur,
1106 	int			level,
1107 	void			*priv)
1108 {
1109 	struct xfs_iread_state	*ir = priv;
1110 	struct xfs_mount	*mp = cur->bc_mp;
1111 	struct xfs_inode	*ip = cur->bc_ino.ip;
1112 	struct xfs_btree_block	*block;
1113 	struct xfs_buf		*bp;
1114 	struct xfs_bmbt_rec	*frp;
1115 	xfs_extnum_t		num_recs;
1116 	xfs_extnum_t		j;
1117 	int			whichfork = cur->bc_ino.whichfork;
1118 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
1119 
1120 	block = xfs_btree_get_block(cur, level, &bp);
1121 
1122 	/* Abort if we find more records than nextents. */
1123 	num_recs = xfs_btree_get_numrecs(block);
1124 	if (unlikely(ir->loaded + num_recs > ifp->if_nextents)) {
1125 		xfs_warn(ip->i_mount, "corrupt dinode %llu, (btree extents).",
1126 				(unsigned long long)ip->i_ino);
1127 		xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, block,
1128 				sizeof(*block), __this_address);
1129 		return -EFSCORRUPTED;
1130 	}
1131 
1132 	/* Copy records into the incore cache. */
1133 	frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1134 	for (j = 0; j < num_recs; j++, frp++, ir->loaded++) {
1135 		struct xfs_bmbt_irec	new;
1136 		xfs_failaddr_t		fa;
1137 
1138 		xfs_bmbt_disk_get_all(frp, &new);
1139 		fa = xfs_bmap_validate_extent(ip, whichfork, &new);
1140 		if (fa) {
1141 			xfs_inode_verifier_error(ip, -EFSCORRUPTED,
1142 					"xfs_iread_extents(2)", frp,
1143 					sizeof(*frp), fa);
1144 			return -EFSCORRUPTED;
1145 		}
1146 		xfs_iext_insert(ip, &ir->icur, &new,
1147 				xfs_bmap_fork_to_state(whichfork));
1148 		trace_xfs_read_extent(ip, &ir->icur,
1149 				xfs_bmap_fork_to_state(whichfork), _THIS_IP_);
1150 		xfs_iext_next(ifp, &ir->icur);
1151 	}
1152 
1153 	return 0;
1154 }
1155 
1156 /*
1157  * Read in extents from a btree-format inode.
1158  */
1159 int
1160 xfs_iread_extents(
1161 	struct xfs_trans	*tp,
1162 	struct xfs_inode	*ip,
1163 	int			whichfork)
1164 {
1165 	struct xfs_iread_state	ir;
1166 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
1167 	struct xfs_mount	*mp = ip->i_mount;
1168 	struct xfs_btree_cur	*cur;
1169 	int			error;
1170 
1171 	if (!xfs_need_iread_extents(ifp))
1172 		return 0;
1173 
1174 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1175 
1176 	ir.loaded = 0;
1177 	xfs_iext_first(ifp, &ir.icur);
1178 	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
1179 	error = xfs_btree_visit_blocks(cur, xfs_iread_bmbt_block,
1180 			XFS_BTREE_VISIT_RECORDS, &ir);
1181 	xfs_btree_del_cursor(cur, error);
1182 	if (error)
1183 		goto out;
1184 
1185 	if (XFS_IS_CORRUPT(mp, ir.loaded != ifp->if_nextents)) {
1186 		error = -EFSCORRUPTED;
1187 		goto out;
1188 	}
1189 	ASSERT(ir.loaded == xfs_iext_count(ifp));
1190 	return 0;
1191 out:
1192 	xfs_iext_destroy(ifp);
1193 	return error;
1194 }
1195 
1196 /*
1197  * Returns the relative block number of the first unused block(s) in the given
1198  * fork with at least "len" logically contiguous blocks free.  This is the
1199  * lowest-address hole if the fork has holes, else the first block past the end
1200  * of fork.  Return 0 if the fork is currently local (in-inode).
1201  */
1202 int						/* error */
1203 xfs_bmap_first_unused(
1204 	struct xfs_trans	*tp,		/* transaction pointer */
1205 	struct xfs_inode	*ip,		/* incore inode */
1206 	xfs_extlen_t		len,		/* size of hole to find */
1207 	xfs_fileoff_t		*first_unused,	/* unused block */
1208 	int			whichfork)	/* data or attr fork */
1209 {
1210 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
1211 	struct xfs_bmbt_irec	got;
1212 	struct xfs_iext_cursor	icur;
1213 	xfs_fileoff_t		lastaddr = 0;
1214 	xfs_fileoff_t		lowest, max;
1215 	int			error;
1216 
1217 	if (ifp->if_format == XFS_DINODE_FMT_LOCAL) {
1218 		*first_unused = 0;
1219 		return 0;
1220 	}
1221 
1222 	ASSERT(xfs_ifork_has_extents(ifp));
1223 
1224 	error = xfs_iread_extents(tp, ip, whichfork);
1225 	if (error)
1226 		return error;
1227 
1228 	lowest = max = *first_unused;
1229 	for_each_xfs_iext(ifp, &icur, &got) {
1230 		/*
1231 		 * See if the hole before this extent will work.
1232 		 */
1233 		if (got.br_startoff >= lowest + len &&
1234 		    got.br_startoff - max >= len)
1235 			break;
1236 		lastaddr = got.br_startoff + got.br_blockcount;
1237 		max = XFS_FILEOFF_MAX(lastaddr, lowest);
1238 	}
1239 
1240 	*first_unused = max;
1241 	return 0;
1242 }
1243 
1244 /*
1245  * Returns the file-relative block number of the last block - 1 before
1246  * last_block (input value) in the file.
1247  * This is not based on i_size, it is based on the extent records.
1248  * Returns 0 for local files, as they do not have extent records.
1249  */
1250 int						/* error */
1251 xfs_bmap_last_before(
1252 	struct xfs_trans	*tp,		/* transaction pointer */
1253 	struct xfs_inode	*ip,		/* incore inode */
1254 	xfs_fileoff_t		*last_block,	/* last block */
1255 	int			whichfork)	/* data or attr fork */
1256 {
1257 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
1258 	struct xfs_bmbt_irec	got;
1259 	struct xfs_iext_cursor	icur;
1260 	int			error;
1261 
1262 	switch (ifp->if_format) {
1263 	case XFS_DINODE_FMT_LOCAL:
1264 		*last_block = 0;
1265 		return 0;
1266 	case XFS_DINODE_FMT_BTREE:
1267 	case XFS_DINODE_FMT_EXTENTS:
1268 		break;
1269 	default:
1270 		ASSERT(0);
1271 		return -EFSCORRUPTED;
1272 	}
1273 
1274 	error = xfs_iread_extents(tp, ip, whichfork);
1275 	if (error)
1276 		return error;
1277 
1278 	if (!xfs_iext_lookup_extent_before(ip, ifp, last_block, &icur, &got))
1279 		*last_block = 0;
1280 	return 0;
1281 }
1282 
1283 int
1284 xfs_bmap_last_extent(
1285 	struct xfs_trans	*tp,
1286 	struct xfs_inode	*ip,
1287 	int			whichfork,
1288 	struct xfs_bmbt_irec	*rec,
1289 	int			*is_empty)
1290 {
1291 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
1292 	struct xfs_iext_cursor	icur;
1293 	int			error;
1294 
1295 	error = xfs_iread_extents(tp, ip, whichfork);
1296 	if (error)
1297 		return error;
1298 
1299 	xfs_iext_last(ifp, &icur);
1300 	if (!xfs_iext_get_extent(ifp, &icur, rec))
1301 		*is_empty = 1;
1302 	else
1303 		*is_empty = 0;
1304 	return 0;
1305 }
1306 
1307 /*
1308  * Check the last inode extent to determine whether this allocation will result
1309  * in blocks being allocated at the end of the file. When we allocate new data
1310  * blocks at the end of the file which do not start at the previous data block,
1311  * we will try to align the new blocks at stripe unit boundaries.
1312  *
1313  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1314  * at, or past the EOF.
1315  */
1316 STATIC int
1317 xfs_bmap_isaeof(
1318 	struct xfs_bmalloca	*bma,
1319 	int			whichfork)
1320 {
1321 	struct xfs_bmbt_irec	rec;
1322 	int			is_empty;
1323 	int			error;
1324 
1325 	bma->aeof = false;
1326 	error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1327 				     &is_empty);
1328 	if (error)
1329 		return error;
1330 
1331 	if (is_empty) {
1332 		bma->aeof = true;
1333 		return 0;
1334 	}
1335 
1336 	/*
1337 	 * Check if we are allocation or past the last extent, or at least into
1338 	 * the last delayed allocated extent.
1339 	 */
1340 	bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1341 		(bma->offset >= rec.br_startoff &&
1342 		 isnullstartblock(rec.br_startblock));
1343 	return 0;
1344 }
1345 
1346 /*
1347  * Returns the file-relative block number of the first block past eof in
1348  * the file.  This is not based on i_size, it is based on the extent records.
1349  * Returns 0 for local files, as they do not have extent records.
1350  */
1351 int
1352 xfs_bmap_last_offset(
1353 	struct xfs_inode	*ip,
1354 	xfs_fileoff_t		*last_block,
1355 	int			whichfork)
1356 {
1357 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
1358 	struct xfs_bmbt_irec	rec;
1359 	int			is_empty;
1360 	int			error;
1361 
1362 	*last_block = 0;
1363 
1364 	if (ifp->if_format == XFS_DINODE_FMT_LOCAL)
1365 		return 0;
1366 
1367 	if (XFS_IS_CORRUPT(ip->i_mount, !xfs_ifork_has_extents(ifp)))
1368 		return -EFSCORRUPTED;
1369 
1370 	error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1371 	if (error || is_empty)
1372 		return error;
1373 
1374 	*last_block = rec.br_startoff + rec.br_blockcount;
1375 	return 0;
1376 }
1377 
1378 /*
1379  * Extent tree manipulation functions used during allocation.
1380  */
1381 
1382 /*
1383  * Convert a delayed allocation to a real allocation.
1384  */
1385 STATIC int				/* error */
1386 xfs_bmap_add_extent_delay_real(
1387 	struct xfs_bmalloca	*bma,
1388 	int			whichfork)
1389 {
1390 	struct xfs_mount	*mp = bma->ip->i_mount;
1391 	struct xfs_ifork	*ifp = xfs_ifork_ptr(bma->ip, whichfork);
1392 	struct xfs_bmbt_irec	*new = &bma->got;
1393 	int			error;	/* error return value */
1394 	int			i;	/* temp state */
1395 	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
1396 	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
1397 					/* left is 0, right is 1, prev is 2 */
1398 	int			rval=0;	/* return value (logging flags) */
1399 	uint32_t		state = xfs_bmap_fork_to_state(whichfork);
1400 	xfs_filblks_t		da_new; /* new count del alloc blocks used */
1401 	xfs_filblks_t		da_old; /* old count del alloc blocks used */
1402 	xfs_filblks_t		temp=0;	/* value for da_new calculations */
1403 	int			tmp_rval;	/* partial logging flags */
1404 	struct xfs_bmbt_irec	old;
1405 
1406 	ASSERT(whichfork != XFS_ATTR_FORK);
1407 	ASSERT(!isnullstartblock(new->br_startblock));
1408 	ASSERT(!bma->cur ||
1409 	       (bma->cur->bc_ino.flags & XFS_BTCUR_BMBT_WASDEL));
1410 
1411 	XFS_STATS_INC(mp, xs_add_exlist);
1412 
1413 #define	LEFT		r[0]
1414 #define	RIGHT		r[1]
1415 #define	PREV		r[2]
1416 
1417 	/*
1418 	 * Set up a bunch of variables to make the tests simpler.
1419 	 */
1420 	xfs_iext_get_extent(ifp, &bma->icur, &PREV);
1421 	new_endoff = new->br_startoff + new->br_blockcount;
1422 	ASSERT(isnullstartblock(PREV.br_startblock));
1423 	ASSERT(PREV.br_startoff <= new->br_startoff);
1424 	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1425 
1426 	da_old = startblockval(PREV.br_startblock);
1427 	da_new = 0;
1428 
1429 	/*
1430 	 * Set flags determining what part of the previous delayed allocation
1431 	 * extent is being replaced by a real allocation.
1432 	 */
1433 	if (PREV.br_startoff == new->br_startoff)
1434 		state |= BMAP_LEFT_FILLING;
1435 	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1436 		state |= BMAP_RIGHT_FILLING;
1437 
1438 	/*
1439 	 * Check and set flags if this segment has a left neighbor.
1440 	 * Don't set contiguous if the combined extent would be too large.
1441 	 */
1442 	if (xfs_iext_peek_prev_extent(ifp, &bma->icur, &LEFT)) {
1443 		state |= BMAP_LEFT_VALID;
1444 		if (isnullstartblock(LEFT.br_startblock))
1445 			state |= BMAP_LEFT_DELAY;
1446 	}
1447 
1448 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1449 	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1450 	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1451 	    LEFT.br_state == new->br_state &&
1452 	    LEFT.br_blockcount + new->br_blockcount <= XFS_MAX_BMBT_EXTLEN)
1453 		state |= BMAP_LEFT_CONTIG;
1454 
1455 	/*
1456 	 * Check and set flags if this segment has a right neighbor.
1457 	 * Don't set contiguous if the combined extent would be too large.
1458 	 * Also check for all-three-contiguous being too large.
1459 	 */
1460 	if (xfs_iext_peek_next_extent(ifp, &bma->icur, &RIGHT)) {
1461 		state |= BMAP_RIGHT_VALID;
1462 		if (isnullstartblock(RIGHT.br_startblock))
1463 			state |= BMAP_RIGHT_DELAY;
1464 	}
1465 
1466 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1467 	    new_endoff == RIGHT.br_startoff &&
1468 	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1469 	    new->br_state == RIGHT.br_state &&
1470 	    new->br_blockcount + RIGHT.br_blockcount <= XFS_MAX_BMBT_EXTLEN &&
1471 	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1472 		       BMAP_RIGHT_FILLING)) !=
1473 		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1474 		       BMAP_RIGHT_FILLING) ||
1475 	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1476 			<= XFS_MAX_BMBT_EXTLEN))
1477 		state |= BMAP_RIGHT_CONTIG;
1478 
1479 	error = 0;
1480 	/*
1481 	 * Switch out based on the FILLING and CONTIG state bits.
1482 	 */
1483 	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1484 			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1485 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1486 	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1487 		/*
1488 		 * Filling in all of a previously delayed allocation extent.
1489 		 * The left and right neighbors are both contiguous with new.
1490 		 */
1491 		LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
1492 
1493 		xfs_iext_remove(bma->ip, &bma->icur, state);
1494 		xfs_iext_remove(bma->ip, &bma->icur, state);
1495 		xfs_iext_prev(ifp, &bma->icur);
1496 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1497 		ifp->if_nextents--;
1498 
1499 		if (bma->cur == NULL)
1500 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1501 		else {
1502 			rval = XFS_ILOG_CORE;
1503 			error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1504 			if (error)
1505 				goto done;
1506 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1507 				error = -EFSCORRUPTED;
1508 				goto done;
1509 			}
1510 			error = xfs_btree_delete(bma->cur, &i);
1511 			if (error)
1512 				goto done;
1513 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1514 				error = -EFSCORRUPTED;
1515 				goto done;
1516 			}
1517 			error = xfs_btree_decrement(bma->cur, 0, &i);
1518 			if (error)
1519 				goto done;
1520 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1521 				error = -EFSCORRUPTED;
1522 				goto done;
1523 			}
1524 			error = xfs_bmbt_update(bma->cur, &LEFT);
1525 			if (error)
1526 				goto done;
1527 		}
1528 		break;
1529 
1530 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1531 		/*
1532 		 * Filling in all of a previously delayed allocation extent.
1533 		 * The left neighbor is contiguous, the right is not.
1534 		 */
1535 		old = LEFT;
1536 		LEFT.br_blockcount += PREV.br_blockcount;
1537 
1538 		xfs_iext_remove(bma->ip, &bma->icur, state);
1539 		xfs_iext_prev(ifp, &bma->icur);
1540 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1541 
1542 		if (bma->cur == NULL)
1543 			rval = XFS_ILOG_DEXT;
1544 		else {
1545 			rval = 0;
1546 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1547 			if (error)
1548 				goto done;
1549 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1550 				error = -EFSCORRUPTED;
1551 				goto done;
1552 			}
1553 			error = xfs_bmbt_update(bma->cur, &LEFT);
1554 			if (error)
1555 				goto done;
1556 		}
1557 		break;
1558 
1559 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1560 		/*
1561 		 * Filling in all of a previously delayed allocation extent.
1562 		 * The right neighbor is contiguous, the left is not. Take care
1563 		 * with delay -> unwritten extent allocation here because the
1564 		 * delalloc record we are overwriting is always written.
1565 		 */
1566 		PREV.br_startblock = new->br_startblock;
1567 		PREV.br_blockcount += RIGHT.br_blockcount;
1568 		PREV.br_state = new->br_state;
1569 
1570 		xfs_iext_next(ifp, &bma->icur);
1571 		xfs_iext_remove(bma->ip, &bma->icur, state);
1572 		xfs_iext_prev(ifp, &bma->icur);
1573 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1574 
1575 		if (bma->cur == NULL)
1576 			rval = XFS_ILOG_DEXT;
1577 		else {
1578 			rval = 0;
1579 			error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1580 			if (error)
1581 				goto done;
1582 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1583 				error = -EFSCORRUPTED;
1584 				goto done;
1585 			}
1586 			error = xfs_bmbt_update(bma->cur, &PREV);
1587 			if (error)
1588 				goto done;
1589 		}
1590 		break;
1591 
1592 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1593 		/*
1594 		 * Filling in all of a previously delayed allocation extent.
1595 		 * Neither the left nor right neighbors are contiguous with
1596 		 * the new one.
1597 		 */
1598 		PREV.br_startblock = new->br_startblock;
1599 		PREV.br_state = new->br_state;
1600 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1601 		ifp->if_nextents++;
1602 
1603 		if (bma->cur == NULL)
1604 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1605 		else {
1606 			rval = XFS_ILOG_CORE;
1607 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1608 			if (error)
1609 				goto done;
1610 			if (XFS_IS_CORRUPT(mp, i != 0)) {
1611 				error = -EFSCORRUPTED;
1612 				goto done;
1613 			}
1614 			error = xfs_btree_insert(bma->cur, &i);
1615 			if (error)
1616 				goto done;
1617 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1618 				error = -EFSCORRUPTED;
1619 				goto done;
1620 			}
1621 		}
1622 		break;
1623 
1624 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1625 		/*
1626 		 * Filling in the first part of a previous delayed allocation.
1627 		 * The left neighbor is contiguous.
1628 		 */
1629 		old = LEFT;
1630 		temp = PREV.br_blockcount - new->br_blockcount;
1631 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1632 				startblockval(PREV.br_startblock));
1633 
1634 		LEFT.br_blockcount += new->br_blockcount;
1635 
1636 		PREV.br_blockcount = temp;
1637 		PREV.br_startoff += new->br_blockcount;
1638 		PREV.br_startblock = nullstartblock(da_new);
1639 
1640 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1641 		xfs_iext_prev(ifp, &bma->icur);
1642 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1643 
1644 		if (bma->cur == NULL)
1645 			rval = XFS_ILOG_DEXT;
1646 		else {
1647 			rval = 0;
1648 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1649 			if (error)
1650 				goto done;
1651 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1652 				error = -EFSCORRUPTED;
1653 				goto done;
1654 			}
1655 			error = xfs_bmbt_update(bma->cur, &LEFT);
1656 			if (error)
1657 				goto done;
1658 		}
1659 		break;
1660 
1661 	case BMAP_LEFT_FILLING:
1662 		/*
1663 		 * Filling in the first part of a previous delayed allocation.
1664 		 * The left neighbor is not contiguous.
1665 		 */
1666 		xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1667 		ifp->if_nextents++;
1668 
1669 		if (bma->cur == NULL)
1670 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1671 		else {
1672 			rval = XFS_ILOG_CORE;
1673 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1674 			if (error)
1675 				goto done;
1676 			if (XFS_IS_CORRUPT(mp, i != 0)) {
1677 				error = -EFSCORRUPTED;
1678 				goto done;
1679 			}
1680 			error = xfs_btree_insert(bma->cur, &i);
1681 			if (error)
1682 				goto done;
1683 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1684 				error = -EFSCORRUPTED;
1685 				goto done;
1686 			}
1687 		}
1688 
1689 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1690 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1691 					&bma->cur, 1, &tmp_rval, whichfork);
1692 			rval |= tmp_rval;
1693 			if (error)
1694 				goto done;
1695 		}
1696 
1697 		temp = PREV.br_blockcount - new->br_blockcount;
1698 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1699 			startblockval(PREV.br_startblock) -
1700 			(bma->cur ? bma->cur->bc_ino.allocated : 0));
1701 
1702 		PREV.br_startoff = new_endoff;
1703 		PREV.br_blockcount = temp;
1704 		PREV.br_startblock = nullstartblock(da_new);
1705 		xfs_iext_next(ifp, &bma->icur);
1706 		xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1707 		xfs_iext_prev(ifp, &bma->icur);
1708 		break;
1709 
1710 	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1711 		/*
1712 		 * Filling in the last part of a previous delayed allocation.
1713 		 * The right neighbor is contiguous with the new allocation.
1714 		 */
1715 		old = RIGHT;
1716 		RIGHT.br_startoff = new->br_startoff;
1717 		RIGHT.br_startblock = new->br_startblock;
1718 		RIGHT.br_blockcount += new->br_blockcount;
1719 
1720 		if (bma->cur == NULL)
1721 			rval = XFS_ILOG_DEXT;
1722 		else {
1723 			rval = 0;
1724 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1725 			if (error)
1726 				goto done;
1727 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1728 				error = -EFSCORRUPTED;
1729 				goto done;
1730 			}
1731 			error = xfs_bmbt_update(bma->cur, &RIGHT);
1732 			if (error)
1733 				goto done;
1734 		}
1735 
1736 		temp = PREV.br_blockcount - new->br_blockcount;
1737 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1738 			startblockval(PREV.br_startblock));
1739 
1740 		PREV.br_blockcount = temp;
1741 		PREV.br_startblock = nullstartblock(da_new);
1742 
1743 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1744 		xfs_iext_next(ifp, &bma->icur);
1745 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &RIGHT);
1746 		break;
1747 
1748 	case BMAP_RIGHT_FILLING:
1749 		/*
1750 		 * Filling in the last part of a previous delayed allocation.
1751 		 * The right neighbor is not contiguous.
1752 		 */
1753 		xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1754 		ifp->if_nextents++;
1755 
1756 		if (bma->cur == NULL)
1757 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1758 		else {
1759 			rval = XFS_ILOG_CORE;
1760 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1761 			if (error)
1762 				goto done;
1763 			if (XFS_IS_CORRUPT(mp, i != 0)) {
1764 				error = -EFSCORRUPTED;
1765 				goto done;
1766 			}
1767 			error = xfs_btree_insert(bma->cur, &i);
1768 			if (error)
1769 				goto done;
1770 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1771 				error = -EFSCORRUPTED;
1772 				goto done;
1773 			}
1774 		}
1775 
1776 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1777 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1778 				&bma->cur, 1, &tmp_rval, whichfork);
1779 			rval |= tmp_rval;
1780 			if (error)
1781 				goto done;
1782 		}
1783 
1784 		temp = PREV.br_blockcount - new->br_blockcount;
1785 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1786 			startblockval(PREV.br_startblock) -
1787 			(bma->cur ? bma->cur->bc_ino.allocated : 0));
1788 
1789 		PREV.br_startblock = nullstartblock(da_new);
1790 		PREV.br_blockcount = temp;
1791 		xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1792 		xfs_iext_next(ifp, &bma->icur);
1793 		break;
1794 
1795 	case 0:
1796 		/*
1797 		 * Filling in the middle part of a previous delayed allocation.
1798 		 * Contiguity is impossible here.
1799 		 * This case is avoided almost all the time.
1800 		 *
1801 		 * We start with a delayed allocation:
1802 		 *
1803 		 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1804 		 *  PREV @ idx
1805 		 *
1806 	         * and we are allocating:
1807 		 *                     +rrrrrrrrrrrrrrrrr+
1808 		 *			      new
1809 		 *
1810 		 * and we set it up for insertion as:
1811 		 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1812 		 *                            new
1813 		 *  PREV @ idx          LEFT              RIGHT
1814 		 *                      inserted at idx + 1
1815 		 */
1816 		old = PREV;
1817 
1818 		/* LEFT is the new middle */
1819 		LEFT = *new;
1820 
1821 		/* RIGHT is the new right */
1822 		RIGHT.br_state = PREV.br_state;
1823 		RIGHT.br_startoff = new_endoff;
1824 		RIGHT.br_blockcount =
1825 			PREV.br_startoff + PREV.br_blockcount - new_endoff;
1826 		RIGHT.br_startblock =
1827 			nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1828 					RIGHT.br_blockcount));
1829 
1830 		/* truncate PREV */
1831 		PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
1832 		PREV.br_startblock =
1833 			nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1834 					PREV.br_blockcount));
1835 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1836 
1837 		xfs_iext_next(ifp, &bma->icur);
1838 		xfs_iext_insert(bma->ip, &bma->icur, &RIGHT, state);
1839 		xfs_iext_insert(bma->ip, &bma->icur, &LEFT, state);
1840 		ifp->if_nextents++;
1841 
1842 		if (bma->cur == NULL)
1843 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1844 		else {
1845 			rval = XFS_ILOG_CORE;
1846 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1847 			if (error)
1848 				goto done;
1849 			if (XFS_IS_CORRUPT(mp, i != 0)) {
1850 				error = -EFSCORRUPTED;
1851 				goto done;
1852 			}
1853 			error = xfs_btree_insert(bma->cur, &i);
1854 			if (error)
1855 				goto done;
1856 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1857 				error = -EFSCORRUPTED;
1858 				goto done;
1859 			}
1860 		}
1861 
1862 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1863 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1864 					&bma->cur, 1, &tmp_rval, whichfork);
1865 			rval |= tmp_rval;
1866 			if (error)
1867 				goto done;
1868 		}
1869 
1870 		da_new = startblockval(PREV.br_startblock) +
1871 			 startblockval(RIGHT.br_startblock);
1872 		break;
1873 
1874 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1875 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1876 	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1877 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1878 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1879 	case BMAP_LEFT_CONTIG:
1880 	case BMAP_RIGHT_CONTIG:
1881 		/*
1882 		 * These cases are all impossible.
1883 		 */
1884 		ASSERT(0);
1885 	}
1886 
1887 	/* add reverse mapping unless caller opted out */
1888 	if (!(bma->flags & XFS_BMAPI_NORMAP))
1889 		xfs_rmap_map_extent(bma->tp, bma->ip, whichfork, new);
1890 
1891 	/* convert to a btree if necessary */
1892 	if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1893 		int	tmp_logflags;	/* partial log flag return val */
1894 
1895 		ASSERT(bma->cur == NULL);
1896 		error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1897 				&bma->cur, da_old > 0, &tmp_logflags,
1898 				whichfork);
1899 		bma->logflags |= tmp_logflags;
1900 		if (error)
1901 			goto done;
1902 	}
1903 
1904 	if (da_new != da_old)
1905 		xfs_mod_delalloc(mp, (int64_t)da_new - da_old);
1906 
1907 	if (bma->cur) {
1908 		da_new += bma->cur->bc_ino.allocated;
1909 		bma->cur->bc_ino.allocated = 0;
1910 	}
1911 
1912 	/* adjust for changes in reserved delayed indirect blocks */
1913 	if (da_new != da_old) {
1914 		ASSERT(state == 0 || da_new < da_old);
1915 		error = xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new),
1916 				false);
1917 	}
1918 
1919 	xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
1920 done:
1921 	if (whichfork != XFS_COW_FORK)
1922 		bma->logflags |= rval;
1923 	return error;
1924 #undef	LEFT
1925 #undef	RIGHT
1926 #undef	PREV
1927 }
1928 
1929 /*
1930  * Convert an unwritten allocation to a real allocation or vice versa.
1931  */
1932 int					/* error */
1933 xfs_bmap_add_extent_unwritten_real(
1934 	struct xfs_trans	*tp,
1935 	xfs_inode_t		*ip,	/* incore inode pointer */
1936 	int			whichfork,
1937 	struct xfs_iext_cursor	*icur,
1938 	struct xfs_btree_cur	**curp,	/* if *curp is null, not a btree */
1939 	xfs_bmbt_irec_t		*new,	/* new data to add to file extents */
1940 	int			*logflagsp) /* inode logging flags */
1941 {
1942 	struct xfs_btree_cur	*cur;	/* btree cursor */
1943 	int			error;	/* error return value */
1944 	int			i;	/* temp state */
1945 	struct xfs_ifork	*ifp;	/* inode fork pointer */
1946 	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
1947 	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
1948 					/* left is 0, right is 1, prev is 2 */
1949 	int			rval=0;	/* return value (logging flags) */
1950 	uint32_t		state = xfs_bmap_fork_to_state(whichfork);
1951 	struct xfs_mount	*mp = ip->i_mount;
1952 	struct xfs_bmbt_irec	old;
1953 
1954 	*logflagsp = 0;
1955 
1956 	cur = *curp;
1957 	ifp = xfs_ifork_ptr(ip, whichfork);
1958 
1959 	ASSERT(!isnullstartblock(new->br_startblock));
1960 
1961 	XFS_STATS_INC(mp, xs_add_exlist);
1962 
1963 #define	LEFT		r[0]
1964 #define	RIGHT		r[1]
1965 #define	PREV		r[2]
1966 
1967 	/*
1968 	 * Set up a bunch of variables to make the tests simpler.
1969 	 */
1970 	error = 0;
1971 	xfs_iext_get_extent(ifp, icur, &PREV);
1972 	ASSERT(new->br_state != PREV.br_state);
1973 	new_endoff = new->br_startoff + new->br_blockcount;
1974 	ASSERT(PREV.br_startoff <= new->br_startoff);
1975 	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1976 
1977 	/*
1978 	 * Set flags determining what part of the previous oldext allocation
1979 	 * extent is being replaced by a newext allocation.
1980 	 */
1981 	if (PREV.br_startoff == new->br_startoff)
1982 		state |= BMAP_LEFT_FILLING;
1983 	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1984 		state |= BMAP_RIGHT_FILLING;
1985 
1986 	/*
1987 	 * Check and set flags if this segment has a left neighbor.
1988 	 * Don't set contiguous if the combined extent would be too large.
1989 	 */
1990 	if (xfs_iext_peek_prev_extent(ifp, icur, &LEFT)) {
1991 		state |= BMAP_LEFT_VALID;
1992 		if (isnullstartblock(LEFT.br_startblock))
1993 			state |= BMAP_LEFT_DELAY;
1994 	}
1995 
1996 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1997 	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1998 	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1999 	    LEFT.br_state == new->br_state &&
2000 	    LEFT.br_blockcount + new->br_blockcount <= XFS_MAX_BMBT_EXTLEN)
2001 		state |= BMAP_LEFT_CONTIG;
2002 
2003 	/*
2004 	 * Check and set flags if this segment has a right neighbor.
2005 	 * Don't set contiguous if the combined extent would be too large.
2006 	 * Also check for all-three-contiguous being too large.
2007 	 */
2008 	if (xfs_iext_peek_next_extent(ifp, icur, &RIGHT)) {
2009 		state |= BMAP_RIGHT_VALID;
2010 		if (isnullstartblock(RIGHT.br_startblock))
2011 			state |= BMAP_RIGHT_DELAY;
2012 	}
2013 
2014 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2015 	    new_endoff == RIGHT.br_startoff &&
2016 	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2017 	    new->br_state == RIGHT.br_state &&
2018 	    new->br_blockcount + RIGHT.br_blockcount <= XFS_MAX_BMBT_EXTLEN &&
2019 	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2020 		       BMAP_RIGHT_FILLING)) !=
2021 		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2022 		       BMAP_RIGHT_FILLING) ||
2023 	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2024 			<= XFS_MAX_BMBT_EXTLEN))
2025 		state |= BMAP_RIGHT_CONTIG;
2026 
2027 	/*
2028 	 * Switch out based on the FILLING and CONTIG state bits.
2029 	 */
2030 	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2031 			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2032 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2033 	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2034 		/*
2035 		 * Setting all of a previous oldext extent to newext.
2036 		 * The left and right neighbors are both contiguous with new.
2037 		 */
2038 		LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
2039 
2040 		xfs_iext_remove(ip, icur, state);
2041 		xfs_iext_remove(ip, icur, state);
2042 		xfs_iext_prev(ifp, icur);
2043 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2044 		ifp->if_nextents -= 2;
2045 		if (cur == NULL)
2046 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2047 		else {
2048 			rval = XFS_ILOG_CORE;
2049 			error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2050 			if (error)
2051 				goto done;
2052 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2053 				error = -EFSCORRUPTED;
2054 				goto done;
2055 			}
2056 			if ((error = xfs_btree_delete(cur, &i)))
2057 				goto done;
2058 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2059 				error = -EFSCORRUPTED;
2060 				goto done;
2061 			}
2062 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2063 				goto done;
2064 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2065 				error = -EFSCORRUPTED;
2066 				goto done;
2067 			}
2068 			if ((error = xfs_btree_delete(cur, &i)))
2069 				goto done;
2070 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2071 				error = -EFSCORRUPTED;
2072 				goto done;
2073 			}
2074 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2075 				goto done;
2076 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2077 				error = -EFSCORRUPTED;
2078 				goto done;
2079 			}
2080 			error = xfs_bmbt_update(cur, &LEFT);
2081 			if (error)
2082 				goto done;
2083 		}
2084 		break;
2085 
2086 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2087 		/*
2088 		 * Setting all of a previous oldext extent to newext.
2089 		 * The left neighbor is contiguous, the right is not.
2090 		 */
2091 		LEFT.br_blockcount += PREV.br_blockcount;
2092 
2093 		xfs_iext_remove(ip, icur, state);
2094 		xfs_iext_prev(ifp, icur);
2095 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2096 		ifp->if_nextents--;
2097 		if (cur == NULL)
2098 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2099 		else {
2100 			rval = XFS_ILOG_CORE;
2101 			error = xfs_bmbt_lookup_eq(cur, &PREV, &i);
2102 			if (error)
2103 				goto done;
2104 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2105 				error = -EFSCORRUPTED;
2106 				goto done;
2107 			}
2108 			if ((error = xfs_btree_delete(cur, &i)))
2109 				goto done;
2110 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2111 				error = -EFSCORRUPTED;
2112 				goto done;
2113 			}
2114 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2115 				goto done;
2116 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2117 				error = -EFSCORRUPTED;
2118 				goto done;
2119 			}
2120 			error = xfs_bmbt_update(cur, &LEFT);
2121 			if (error)
2122 				goto done;
2123 		}
2124 		break;
2125 
2126 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2127 		/*
2128 		 * Setting all of a previous oldext extent to newext.
2129 		 * The right neighbor is contiguous, the left is not.
2130 		 */
2131 		PREV.br_blockcount += RIGHT.br_blockcount;
2132 		PREV.br_state = new->br_state;
2133 
2134 		xfs_iext_next(ifp, icur);
2135 		xfs_iext_remove(ip, icur, state);
2136 		xfs_iext_prev(ifp, icur);
2137 		xfs_iext_update_extent(ip, state, icur, &PREV);
2138 		ifp->if_nextents--;
2139 
2140 		if (cur == NULL)
2141 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2142 		else {
2143 			rval = XFS_ILOG_CORE;
2144 			error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2145 			if (error)
2146 				goto done;
2147 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2148 				error = -EFSCORRUPTED;
2149 				goto done;
2150 			}
2151 			if ((error = xfs_btree_delete(cur, &i)))
2152 				goto done;
2153 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2154 				error = -EFSCORRUPTED;
2155 				goto done;
2156 			}
2157 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2158 				goto done;
2159 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2160 				error = -EFSCORRUPTED;
2161 				goto done;
2162 			}
2163 			error = xfs_bmbt_update(cur, &PREV);
2164 			if (error)
2165 				goto done;
2166 		}
2167 		break;
2168 
2169 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2170 		/*
2171 		 * Setting all of a previous oldext extent to newext.
2172 		 * Neither the left nor right neighbors are contiguous with
2173 		 * the new one.
2174 		 */
2175 		PREV.br_state = new->br_state;
2176 		xfs_iext_update_extent(ip, state, icur, &PREV);
2177 
2178 		if (cur == NULL)
2179 			rval = XFS_ILOG_DEXT;
2180 		else {
2181 			rval = 0;
2182 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2183 			if (error)
2184 				goto done;
2185 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2186 				error = -EFSCORRUPTED;
2187 				goto done;
2188 			}
2189 			error = xfs_bmbt_update(cur, &PREV);
2190 			if (error)
2191 				goto done;
2192 		}
2193 		break;
2194 
2195 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2196 		/*
2197 		 * Setting the first part of a previous oldext extent to newext.
2198 		 * The left neighbor is contiguous.
2199 		 */
2200 		LEFT.br_blockcount += new->br_blockcount;
2201 
2202 		old = PREV;
2203 		PREV.br_startoff += new->br_blockcount;
2204 		PREV.br_startblock += new->br_blockcount;
2205 		PREV.br_blockcount -= new->br_blockcount;
2206 
2207 		xfs_iext_update_extent(ip, state, icur, &PREV);
2208 		xfs_iext_prev(ifp, icur);
2209 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2210 
2211 		if (cur == NULL)
2212 			rval = XFS_ILOG_DEXT;
2213 		else {
2214 			rval = 0;
2215 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2216 			if (error)
2217 				goto done;
2218 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2219 				error = -EFSCORRUPTED;
2220 				goto done;
2221 			}
2222 			error = xfs_bmbt_update(cur, &PREV);
2223 			if (error)
2224 				goto done;
2225 			error = xfs_btree_decrement(cur, 0, &i);
2226 			if (error)
2227 				goto done;
2228 			error = xfs_bmbt_update(cur, &LEFT);
2229 			if (error)
2230 				goto done;
2231 		}
2232 		break;
2233 
2234 	case BMAP_LEFT_FILLING:
2235 		/*
2236 		 * Setting the first part of a previous oldext extent to newext.
2237 		 * The left neighbor is not contiguous.
2238 		 */
2239 		old = PREV;
2240 		PREV.br_startoff += new->br_blockcount;
2241 		PREV.br_startblock += new->br_blockcount;
2242 		PREV.br_blockcount -= new->br_blockcount;
2243 
2244 		xfs_iext_update_extent(ip, state, icur, &PREV);
2245 		xfs_iext_insert(ip, icur, new, state);
2246 		ifp->if_nextents++;
2247 
2248 		if (cur == NULL)
2249 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2250 		else {
2251 			rval = XFS_ILOG_CORE;
2252 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2253 			if (error)
2254 				goto done;
2255 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2256 				error = -EFSCORRUPTED;
2257 				goto done;
2258 			}
2259 			error = xfs_bmbt_update(cur, &PREV);
2260 			if (error)
2261 				goto done;
2262 			cur->bc_rec.b = *new;
2263 			if ((error = xfs_btree_insert(cur, &i)))
2264 				goto done;
2265 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2266 				error = -EFSCORRUPTED;
2267 				goto done;
2268 			}
2269 		}
2270 		break;
2271 
2272 	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2273 		/*
2274 		 * Setting the last part of a previous oldext extent to newext.
2275 		 * The right neighbor is contiguous with the new allocation.
2276 		 */
2277 		old = PREV;
2278 		PREV.br_blockcount -= new->br_blockcount;
2279 
2280 		RIGHT.br_startoff = new->br_startoff;
2281 		RIGHT.br_startblock = new->br_startblock;
2282 		RIGHT.br_blockcount += new->br_blockcount;
2283 
2284 		xfs_iext_update_extent(ip, state, icur, &PREV);
2285 		xfs_iext_next(ifp, icur);
2286 		xfs_iext_update_extent(ip, state, icur, &RIGHT);
2287 
2288 		if (cur == NULL)
2289 			rval = XFS_ILOG_DEXT;
2290 		else {
2291 			rval = 0;
2292 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2293 			if (error)
2294 				goto done;
2295 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2296 				error = -EFSCORRUPTED;
2297 				goto done;
2298 			}
2299 			error = xfs_bmbt_update(cur, &PREV);
2300 			if (error)
2301 				goto done;
2302 			error = xfs_btree_increment(cur, 0, &i);
2303 			if (error)
2304 				goto done;
2305 			error = xfs_bmbt_update(cur, &RIGHT);
2306 			if (error)
2307 				goto done;
2308 		}
2309 		break;
2310 
2311 	case BMAP_RIGHT_FILLING:
2312 		/*
2313 		 * Setting the last part of a previous oldext extent to newext.
2314 		 * The right neighbor is not contiguous.
2315 		 */
2316 		old = PREV;
2317 		PREV.br_blockcount -= new->br_blockcount;
2318 
2319 		xfs_iext_update_extent(ip, state, icur, &PREV);
2320 		xfs_iext_next(ifp, icur);
2321 		xfs_iext_insert(ip, icur, new, state);
2322 		ifp->if_nextents++;
2323 
2324 		if (cur == NULL)
2325 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2326 		else {
2327 			rval = XFS_ILOG_CORE;
2328 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2329 			if (error)
2330 				goto done;
2331 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2332 				error = -EFSCORRUPTED;
2333 				goto done;
2334 			}
2335 			error = xfs_bmbt_update(cur, &PREV);
2336 			if (error)
2337 				goto done;
2338 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2339 			if (error)
2340 				goto done;
2341 			if (XFS_IS_CORRUPT(mp, i != 0)) {
2342 				error = -EFSCORRUPTED;
2343 				goto done;
2344 			}
2345 			if ((error = xfs_btree_insert(cur, &i)))
2346 				goto done;
2347 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2348 				error = -EFSCORRUPTED;
2349 				goto done;
2350 			}
2351 		}
2352 		break;
2353 
2354 	case 0:
2355 		/*
2356 		 * Setting the middle part of a previous oldext extent to
2357 		 * newext.  Contiguity is impossible here.
2358 		 * One extent becomes three extents.
2359 		 */
2360 		old = PREV;
2361 		PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
2362 
2363 		r[0] = *new;
2364 		r[1].br_startoff = new_endoff;
2365 		r[1].br_blockcount =
2366 			old.br_startoff + old.br_blockcount - new_endoff;
2367 		r[1].br_startblock = new->br_startblock + new->br_blockcount;
2368 		r[1].br_state = PREV.br_state;
2369 
2370 		xfs_iext_update_extent(ip, state, icur, &PREV);
2371 		xfs_iext_next(ifp, icur);
2372 		xfs_iext_insert(ip, icur, &r[1], state);
2373 		xfs_iext_insert(ip, icur, &r[0], state);
2374 		ifp->if_nextents += 2;
2375 
2376 		if (cur == NULL)
2377 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2378 		else {
2379 			rval = XFS_ILOG_CORE;
2380 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2381 			if (error)
2382 				goto done;
2383 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2384 				error = -EFSCORRUPTED;
2385 				goto done;
2386 			}
2387 			/* new right extent - oldext */
2388 			error = xfs_bmbt_update(cur, &r[1]);
2389 			if (error)
2390 				goto done;
2391 			/* new left extent - oldext */
2392 			cur->bc_rec.b = PREV;
2393 			if ((error = xfs_btree_insert(cur, &i)))
2394 				goto done;
2395 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2396 				error = -EFSCORRUPTED;
2397 				goto done;
2398 			}
2399 			/*
2400 			 * Reset the cursor to the position of the new extent
2401 			 * we are about to insert as we can't trust it after
2402 			 * the previous insert.
2403 			 */
2404 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2405 			if (error)
2406 				goto done;
2407 			if (XFS_IS_CORRUPT(mp, i != 0)) {
2408 				error = -EFSCORRUPTED;
2409 				goto done;
2410 			}
2411 			/* new middle extent - newext */
2412 			if ((error = xfs_btree_insert(cur, &i)))
2413 				goto done;
2414 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2415 				error = -EFSCORRUPTED;
2416 				goto done;
2417 			}
2418 		}
2419 		break;
2420 
2421 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2422 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2423 	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2424 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2425 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2426 	case BMAP_LEFT_CONTIG:
2427 	case BMAP_RIGHT_CONTIG:
2428 		/*
2429 		 * These cases are all impossible.
2430 		 */
2431 		ASSERT(0);
2432 	}
2433 
2434 	/* update reverse mappings */
2435 	xfs_rmap_convert_extent(mp, tp, ip, whichfork, new);
2436 
2437 	/* convert to a btree if necessary */
2438 	if (xfs_bmap_needs_btree(ip, whichfork)) {
2439 		int	tmp_logflags;	/* partial log flag return val */
2440 
2441 		ASSERT(cur == NULL);
2442 		error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
2443 				&tmp_logflags, whichfork);
2444 		*logflagsp |= tmp_logflags;
2445 		if (error)
2446 			goto done;
2447 	}
2448 
2449 	/* clear out the allocated field, done with it now in any case. */
2450 	if (cur) {
2451 		cur->bc_ino.allocated = 0;
2452 		*curp = cur;
2453 	}
2454 
2455 	xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
2456 done:
2457 	*logflagsp |= rval;
2458 	return error;
2459 #undef	LEFT
2460 #undef	RIGHT
2461 #undef	PREV
2462 }
2463 
2464 /*
2465  * Convert a hole to a delayed allocation.
2466  */
2467 STATIC void
2468 xfs_bmap_add_extent_hole_delay(
2469 	xfs_inode_t		*ip,	/* incore inode pointer */
2470 	int			whichfork,
2471 	struct xfs_iext_cursor	*icur,
2472 	xfs_bmbt_irec_t		*new)	/* new data to add to file extents */
2473 {
2474 	struct xfs_ifork	*ifp;	/* inode fork pointer */
2475 	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2476 	xfs_filblks_t		newlen=0;	/* new indirect size */
2477 	xfs_filblks_t		oldlen=0;	/* old indirect size */
2478 	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2479 	uint32_t		state = xfs_bmap_fork_to_state(whichfork);
2480 	xfs_filblks_t		temp;	 /* temp for indirect calculations */
2481 
2482 	ifp = xfs_ifork_ptr(ip, whichfork);
2483 	ASSERT(isnullstartblock(new->br_startblock));
2484 
2485 	/*
2486 	 * Check and set flags if this segment has a left neighbor
2487 	 */
2488 	if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2489 		state |= BMAP_LEFT_VALID;
2490 		if (isnullstartblock(left.br_startblock))
2491 			state |= BMAP_LEFT_DELAY;
2492 	}
2493 
2494 	/*
2495 	 * Check and set flags if the current (right) segment exists.
2496 	 * If it doesn't exist, we're converting the hole at end-of-file.
2497 	 */
2498 	if (xfs_iext_get_extent(ifp, icur, &right)) {
2499 		state |= BMAP_RIGHT_VALID;
2500 		if (isnullstartblock(right.br_startblock))
2501 			state |= BMAP_RIGHT_DELAY;
2502 	}
2503 
2504 	/*
2505 	 * Set contiguity flags on the left and right neighbors.
2506 	 * Don't let extents get too large, even if the pieces are contiguous.
2507 	 */
2508 	if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2509 	    left.br_startoff + left.br_blockcount == new->br_startoff &&
2510 	    left.br_blockcount + new->br_blockcount <= XFS_MAX_BMBT_EXTLEN)
2511 		state |= BMAP_LEFT_CONTIG;
2512 
2513 	if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2514 	    new->br_startoff + new->br_blockcount == right.br_startoff &&
2515 	    new->br_blockcount + right.br_blockcount <= XFS_MAX_BMBT_EXTLEN &&
2516 	    (!(state & BMAP_LEFT_CONTIG) ||
2517 	     (left.br_blockcount + new->br_blockcount +
2518 	      right.br_blockcount <= XFS_MAX_BMBT_EXTLEN)))
2519 		state |= BMAP_RIGHT_CONTIG;
2520 
2521 	/*
2522 	 * Switch out based on the contiguity flags.
2523 	 */
2524 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2525 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2526 		/*
2527 		 * New allocation is contiguous with delayed allocations
2528 		 * on the left and on the right.
2529 		 * Merge all three into a single extent record.
2530 		 */
2531 		temp = left.br_blockcount + new->br_blockcount +
2532 			right.br_blockcount;
2533 
2534 		oldlen = startblockval(left.br_startblock) +
2535 			startblockval(new->br_startblock) +
2536 			startblockval(right.br_startblock);
2537 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2538 					 oldlen);
2539 		left.br_startblock = nullstartblock(newlen);
2540 		left.br_blockcount = temp;
2541 
2542 		xfs_iext_remove(ip, icur, state);
2543 		xfs_iext_prev(ifp, icur);
2544 		xfs_iext_update_extent(ip, state, icur, &left);
2545 		break;
2546 
2547 	case BMAP_LEFT_CONTIG:
2548 		/*
2549 		 * New allocation is contiguous with a delayed allocation
2550 		 * on the left.
2551 		 * Merge the new allocation with the left neighbor.
2552 		 */
2553 		temp = left.br_blockcount + new->br_blockcount;
2554 
2555 		oldlen = startblockval(left.br_startblock) +
2556 			startblockval(new->br_startblock);
2557 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2558 					 oldlen);
2559 		left.br_blockcount = temp;
2560 		left.br_startblock = nullstartblock(newlen);
2561 
2562 		xfs_iext_prev(ifp, icur);
2563 		xfs_iext_update_extent(ip, state, icur, &left);
2564 		break;
2565 
2566 	case BMAP_RIGHT_CONTIG:
2567 		/*
2568 		 * New allocation is contiguous with a delayed allocation
2569 		 * on the right.
2570 		 * Merge the new allocation with the right neighbor.
2571 		 */
2572 		temp = new->br_blockcount + right.br_blockcount;
2573 		oldlen = startblockval(new->br_startblock) +
2574 			startblockval(right.br_startblock);
2575 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2576 					 oldlen);
2577 		right.br_startoff = new->br_startoff;
2578 		right.br_startblock = nullstartblock(newlen);
2579 		right.br_blockcount = temp;
2580 		xfs_iext_update_extent(ip, state, icur, &right);
2581 		break;
2582 
2583 	case 0:
2584 		/*
2585 		 * New allocation is not contiguous with another
2586 		 * delayed allocation.
2587 		 * Insert a new entry.
2588 		 */
2589 		oldlen = newlen = 0;
2590 		xfs_iext_insert(ip, icur, new, state);
2591 		break;
2592 	}
2593 	if (oldlen != newlen) {
2594 		ASSERT(oldlen > newlen);
2595 		xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2596 				 false);
2597 		/*
2598 		 * Nothing to do for disk quota accounting here.
2599 		 */
2600 		xfs_mod_delalloc(ip->i_mount, (int64_t)newlen - oldlen);
2601 	}
2602 }
2603 
2604 /*
2605  * Convert a hole to a real allocation.
2606  */
2607 STATIC int				/* error */
2608 xfs_bmap_add_extent_hole_real(
2609 	struct xfs_trans	*tp,
2610 	struct xfs_inode	*ip,
2611 	int			whichfork,
2612 	struct xfs_iext_cursor	*icur,
2613 	struct xfs_btree_cur	**curp,
2614 	struct xfs_bmbt_irec	*new,
2615 	int			*logflagsp,
2616 	uint32_t		flags)
2617 {
2618 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
2619 	struct xfs_mount	*mp = ip->i_mount;
2620 	struct xfs_btree_cur	*cur = *curp;
2621 	int			error;	/* error return value */
2622 	int			i;	/* temp state */
2623 	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2624 	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2625 	int			rval=0;	/* return value (logging flags) */
2626 	uint32_t		state = xfs_bmap_fork_to_state(whichfork);
2627 	struct xfs_bmbt_irec	old;
2628 
2629 	ASSERT(!isnullstartblock(new->br_startblock));
2630 	ASSERT(!cur || !(cur->bc_ino.flags & XFS_BTCUR_BMBT_WASDEL));
2631 
2632 	XFS_STATS_INC(mp, xs_add_exlist);
2633 
2634 	/*
2635 	 * Check and set flags if this segment has a left neighbor.
2636 	 */
2637 	if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2638 		state |= BMAP_LEFT_VALID;
2639 		if (isnullstartblock(left.br_startblock))
2640 			state |= BMAP_LEFT_DELAY;
2641 	}
2642 
2643 	/*
2644 	 * Check and set flags if this segment has a current value.
2645 	 * Not true if we're inserting into the "hole" at eof.
2646 	 */
2647 	if (xfs_iext_get_extent(ifp, icur, &right)) {
2648 		state |= BMAP_RIGHT_VALID;
2649 		if (isnullstartblock(right.br_startblock))
2650 			state |= BMAP_RIGHT_DELAY;
2651 	}
2652 
2653 	/*
2654 	 * We're inserting a real allocation between "left" and "right".
2655 	 * Set the contiguity flags.  Don't let extents get too large.
2656 	 */
2657 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2658 	    left.br_startoff + left.br_blockcount == new->br_startoff &&
2659 	    left.br_startblock + left.br_blockcount == new->br_startblock &&
2660 	    left.br_state == new->br_state &&
2661 	    left.br_blockcount + new->br_blockcount <= XFS_MAX_BMBT_EXTLEN)
2662 		state |= BMAP_LEFT_CONTIG;
2663 
2664 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2665 	    new->br_startoff + new->br_blockcount == right.br_startoff &&
2666 	    new->br_startblock + new->br_blockcount == right.br_startblock &&
2667 	    new->br_state == right.br_state &&
2668 	    new->br_blockcount + right.br_blockcount <= XFS_MAX_BMBT_EXTLEN &&
2669 	    (!(state & BMAP_LEFT_CONTIG) ||
2670 	     left.br_blockcount + new->br_blockcount +
2671 	     right.br_blockcount <= XFS_MAX_BMBT_EXTLEN))
2672 		state |= BMAP_RIGHT_CONTIG;
2673 
2674 	error = 0;
2675 	/*
2676 	 * Select which case we're in here, and implement it.
2677 	 */
2678 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2679 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2680 		/*
2681 		 * New allocation is contiguous with real allocations on the
2682 		 * left and on the right.
2683 		 * Merge all three into a single extent record.
2684 		 */
2685 		left.br_blockcount += new->br_blockcount + right.br_blockcount;
2686 
2687 		xfs_iext_remove(ip, icur, state);
2688 		xfs_iext_prev(ifp, icur);
2689 		xfs_iext_update_extent(ip, state, icur, &left);
2690 		ifp->if_nextents--;
2691 
2692 		if (cur == NULL) {
2693 			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2694 		} else {
2695 			rval = XFS_ILOG_CORE;
2696 			error = xfs_bmbt_lookup_eq(cur, &right, &i);
2697 			if (error)
2698 				goto done;
2699 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2700 				error = -EFSCORRUPTED;
2701 				goto done;
2702 			}
2703 			error = xfs_btree_delete(cur, &i);
2704 			if (error)
2705 				goto done;
2706 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2707 				error = -EFSCORRUPTED;
2708 				goto done;
2709 			}
2710 			error = xfs_btree_decrement(cur, 0, &i);
2711 			if (error)
2712 				goto done;
2713 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2714 				error = -EFSCORRUPTED;
2715 				goto done;
2716 			}
2717 			error = xfs_bmbt_update(cur, &left);
2718 			if (error)
2719 				goto done;
2720 		}
2721 		break;
2722 
2723 	case BMAP_LEFT_CONTIG:
2724 		/*
2725 		 * New allocation is contiguous with a real allocation
2726 		 * on the left.
2727 		 * Merge the new allocation with the left neighbor.
2728 		 */
2729 		old = left;
2730 		left.br_blockcount += new->br_blockcount;
2731 
2732 		xfs_iext_prev(ifp, icur);
2733 		xfs_iext_update_extent(ip, state, icur, &left);
2734 
2735 		if (cur == NULL) {
2736 			rval = xfs_ilog_fext(whichfork);
2737 		} else {
2738 			rval = 0;
2739 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2740 			if (error)
2741 				goto done;
2742 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2743 				error = -EFSCORRUPTED;
2744 				goto done;
2745 			}
2746 			error = xfs_bmbt_update(cur, &left);
2747 			if (error)
2748 				goto done;
2749 		}
2750 		break;
2751 
2752 	case BMAP_RIGHT_CONTIG:
2753 		/*
2754 		 * New allocation is contiguous with a real allocation
2755 		 * on the right.
2756 		 * Merge the new allocation with the right neighbor.
2757 		 */
2758 		old = right;
2759 
2760 		right.br_startoff = new->br_startoff;
2761 		right.br_startblock = new->br_startblock;
2762 		right.br_blockcount += new->br_blockcount;
2763 		xfs_iext_update_extent(ip, state, icur, &right);
2764 
2765 		if (cur == NULL) {
2766 			rval = xfs_ilog_fext(whichfork);
2767 		} else {
2768 			rval = 0;
2769 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2770 			if (error)
2771 				goto done;
2772 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2773 				error = -EFSCORRUPTED;
2774 				goto done;
2775 			}
2776 			error = xfs_bmbt_update(cur, &right);
2777 			if (error)
2778 				goto done;
2779 		}
2780 		break;
2781 
2782 	case 0:
2783 		/*
2784 		 * New allocation is not contiguous with another
2785 		 * real allocation.
2786 		 * Insert a new entry.
2787 		 */
2788 		xfs_iext_insert(ip, icur, new, state);
2789 		ifp->if_nextents++;
2790 
2791 		if (cur == NULL) {
2792 			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2793 		} else {
2794 			rval = XFS_ILOG_CORE;
2795 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2796 			if (error)
2797 				goto done;
2798 			if (XFS_IS_CORRUPT(mp, i != 0)) {
2799 				error = -EFSCORRUPTED;
2800 				goto done;
2801 			}
2802 			error = xfs_btree_insert(cur, &i);
2803 			if (error)
2804 				goto done;
2805 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2806 				error = -EFSCORRUPTED;
2807 				goto done;
2808 			}
2809 		}
2810 		break;
2811 	}
2812 
2813 	/* add reverse mapping unless caller opted out */
2814 	if (!(flags & XFS_BMAPI_NORMAP))
2815 		xfs_rmap_map_extent(tp, ip, whichfork, new);
2816 
2817 	/* convert to a btree if necessary */
2818 	if (xfs_bmap_needs_btree(ip, whichfork)) {
2819 		int	tmp_logflags;	/* partial log flag return val */
2820 
2821 		ASSERT(cur == NULL);
2822 		error = xfs_bmap_extents_to_btree(tp, ip, curp, 0,
2823 				&tmp_logflags, whichfork);
2824 		*logflagsp |= tmp_logflags;
2825 		cur = *curp;
2826 		if (error)
2827 			goto done;
2828 	}
2829 
2830 	/* clear out the allocated field, done with it now in any case. */
2831 	if (cur)
2832 		cur->bc_ino.allocated = 0;
2833 
2834 	xfs_bmap_check_leaf_extents(cur, ip, whichfork);
2835 done:
2836 	*logflagsp |= rval;
2837 	return error;
2838 }
2839 
2840 /*
2841  * Functions used in the extent read, allocate and remove paths
2842  */
2843 
2844 /*
2845  * Adjust the size of the new extent based on i_extsize and rt extsize.
2846  */
2847 int
2848 xfs_bmap_extsize_align(
2849 	xfs_mount_t	*mp,
2850 	xfs_bmbt_irec_t	*gotp,		/* next extent pointer */
2851 	xfs_bmbt_irec_t	*prevp,		/* previous extent pointer */
2852 	xfs_extlen_t	extsz,		/* align to this extent size */
2853 	int		rt,		/* is this a realtime inode? */
2854 	int		eof,		/* is extent at end-of-file? */
2855 	int		delay,		/* creating delalloc extent? */
2856 	int		convert,	/* overwriting unwritten extent? */
2857 	xfs_fileoff_t	*offp,		/* in/out: aligned offset */
2858 	xfs_extlen_t	*lenp)		/* in/out: aligned length */
2859 {
2860 	xfs_fileoff_t	orig_off;	/* original offset */
2861 	xfs_extlen_t	orig_alen;	/* original length */
2862 	xfs_fileoff_t	orig_end;	/* original off+len */
2863 	xfs_fileoff_t	nexto;		/* next file offset */
2864 	xfs_fileoff_t	prevo;		/* previous file offset */
2865 	xfs_fileoff_t	align_off;	/* temp for offset */
2866 	xfs_extlen_t	align_alen;	/* temp for length */
2867 	xfs_extlen_t	temp;		/* temp for calculations */
2868 
2869 	if (convert)
2870 		return 0;
2871 
2872 	orig_off = align_off = *offp;
2873 	orig_alen = align_alen = *lenp;
2874 	orig_end = orig_off + orig_alen;
2875 
2876 	/*
2877 	 * If this request overlaps an existing extent, then don't
2878 	 * attempt to perform any additional alignment.
2879 	 */
2880 	if (!delay && !eof &&
2881 	    (orig_off >= gotp->br_startoff) &&
2882 	    (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2883 		return 0;
2884 	}
2885 
2886 	/*
2887 	 * If the file offset is unaligned vs. the extent size
2888 	 * we need to align it.  This will be possible unless
2889 	 * the file was previously written with a kernel that didn't
2890 	 * perform this alignment, or if a truncate shot us in the
2891 	 * foot.
2892 	 */
2893 	div_u64_rem(orig_off, extsz, &temp);
2894 	if (temp) {
2895 		align_alen += temp;
2896 		align_off -= temp;
2897 	}
2898 
2899 	/* Same adjustment for the end of the requested area. */
2900 	temp = (align_alen % extsz);
2901 	if (temp)
2902 		align_alen += extsz - temp;
2903 
2904 	/*
2905 	 * For large extent hint sizes, the aligned extent might be larger than
2906 	 * XFS_BMBT_MAX_EXTLEN. In that case, reduce the size by an extsz so
2907 	 * that it pulls the length back under XFS_BMBT_MAX_EXTLEN. The outer
2908 	 * allocation loops handle short allocation just fine, so it is safe to
2909 	 * do this. We only want to do it when we are forced to, though, because
2910 	 * it means more allocation operations are required.
2911 	 */
2912 	while (align_alen > XFS_MAX_BMBT_EXTLEN)
2913 		align_alen -= extsz;
2914 	ASSERT(align_alen <= XFS_MAX_BMBT_EXTLEN);
2915 
2916 	/*
2917 	 * If the previous block overlaps with this proposed allocation
2918 	 * then move the start forward without adjusting the length.
2919 	 */
2920 	if (prevp->br_startoff != NULLFILEOFF) {
2921 		if (prevp->br_startblock == HOLESTARTBLOCK)
2922 			prevo = prevp->br_startoff;
2923 		else
2924 			prevo = prevp->br_startoff + prevp->br_blockcount;
2925 	} else
2926 		prevo = 0;
2927 	if (align_off != orig_off && align_off < prevo)
2928 		align_off = prevo;
2929 	/*
2930 	 * If the next block overlaps with this proposed allocation
2931 	 * then move the start back without adjusting the length,
2932 	 * but not before offset 0.
2933 	 * This may of course make the start overlap previous block,
2934 	 * and if we hit the offset 0 limit then the next block
2935 	 * can still overlap too.
2936 	 */
2937 	if (!eof && gotp->br_startoff != NULLFILEOFF) {
2938 		if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2939 		    (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2940 			nexto = gotp->br_startoff + gotp->br_blockcount;
2941 		else
2942 			nexto = gotp->br_startoff;
2943 	} else
2944 		nexto = NULLFILEOFF;
2945 	if (!eof &&
2946 	    align_off + align_alen != orig_end &&
2947 	    align_off + align_alen > nexto)
2948 		align_off = nexto > align_alen ? nexto - align_alen : 0;
2949 	/*
2950 	 * If we're now overlapping the next or previous extent that
2951 	 * means we can't fit an extsz piece in this hole.  Just move
2952 	 * the start forward to the first valid spot and set
2953 	 * the length so we hit the end.
2954 	 */
2955 	if (align_off != orig_off && align_off < prevo)
2956 		align_off = prevo;
2957 	if (align_off + align_alen != orig_end &&
2958 	    align_off + align_alen > nexto &&
2959 	    nexto != NULLFILEOFF) {
2960 		ASSERT(nexto > prevo);
2961 		align_alen = nexto - align_off;
2962 	}
2963 
2964 	/*
2965 	 * If realtime, and the result isn't a multiple of the realtime
2966 	 * extent size we need to remove blocks until it is.
2967 	 */
2968 	if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2969 		/*
2970 		 * We're not covering the original request, or
2971 		 * we won't be able to once we fix the length.
2972 		 */
2973 		if (orig_off < align_off ||
2974 		    orig_end > align_off + align_alen ||
2975 		    align_alen - temp < orig_alen)
2976 			return -EINVAL;
2977 		/*
2978 		 * Try to fix it by moving the start up.
2979 		 */
2980 		if (align_off + temp <= orig_off) {
2981 			align_alen -= temp;
2982 			align_off += temp;
2983 		}
2984 		/*
2985 		 * Try to fix it by moving the end in.
2986 		 */
2987 		else if (align_off + align_alen - temp >= orig_end)
2988 			align_alen -= temp;
2989 		/*
2990 		 * Set the start to the minimum then trim the length.
2991 		 */
2992 		else {
2993 			align_alen -= orig_off - align_off;
2994 			align_off = orig_off;
2995 			align_alen -= align_alen % mp->m_sb.sb_rextsize;
2996 		}
2997 		/*
2998 		 * Result doesn't cover the request, fail it.
2999 		 */
3000 		if (orig_off < align_off || orig_end > align_off + align_alen)
3001 			return -EINVAL;
3002 	} else {
3003 		ASSERT(orig_off >= align_off);
3004 		/* see XFS_BMBT_MAX_EXTLEN handling above */
3005 		ASSERT(orig_end <= align_off + align_alen ||
3006 		       align_alen + extsz > XFS_MAX_BMBT_EXTLEN);
3007 	}
3008 
3009 #ifdef DEBUG
3010 	if (!eof && gotp->br_startoff != NULLFILEOFF)
3011 		ASSERT(align_off + align_alen <= gotp->br_startoff);
3012 	if (prevp->br_startoff != NULLFILEOFF)
3013 		ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3014 #endif
3015 
3016 	*lenp = align_alen;
3017 	*offp = align_off;
3018 	return 0;
3019 }
3020 
3021 #define XFS_ALLOC_GAP_UNITS	4
3022 
3023 void
3024 xfs_bmap_adjacent(
3025 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3026 {
3027 	xfs_fsblock_t	adjust;		/* adjustment to block numbers */
3028 	xfs_agnumber_t	fb_agno;	/* ag number of ap->firstblock */
3029 	xfs_mount_t	*mp;		/* mount point structure */
3030 	int		nullfb;		/* true if ap->firstblock isn't set */
3031 	int		rt;		/* true if inode is realtime */
3032 
3033 #define	ISVALID(x,y)	\
3034 	(rt ? \
3035 		(x) < mp->m_sb.sb_rblocks : \
3036 		XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3037 		XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3038 		XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3039 
3040 	mp = ap->ip->i_mount;
3041 	nullfb = ap->tp->t_firstblock == NULLFSBLOCK;
3042 	rt = XFS_IS_REALTIME_INODE(ap->ip) &&
3043 		(ap->datatype & XFS_ALLOC_USERDATA);
3044 	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp,
3045 							ap->tp->t_firstblock);
3046 	/*
3047 	 * If allocating at eof, and there's a previous real block,
3048 	 * try to use its last block as our starting point.
3049 	 */
3050 	if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3051 	    !isnullstartblock(ap->prev.br_startblock) &&
3052 	    ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3053 		    ap->prev.br_startblock)) {
3054 		ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3055 		/*
3056 		 * Adjust for the gap between prevp and us.
3057 		 */
3058 		adjust = ap->offset -
3059 			(ap->prev.br_startoff + ap->prev.br_blockcount);
3060 		if (adjust &&
3061 		    ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3062 			ap->blkno += adjust;
3063 	}
3064 	/*
3065 	 * If not at eof, then compare the two neighbor blocks.
3066 	 * Figure out whether either one gives us a good starting point,
3067 	 * and pick the better one.
3068 	 */
3069 	else if (!ap->eof) {
3070 		xfs_fsblock_t	gotbno;		/* right side block number */
3071 		xfs_fsblock_t	gotdiff=0;	/* right side difference */
3072 		xfs_fsblock_t	prevbno;	/* left side block number */
3073 		xfs_fsblock_t	prevdiff=0;	/* left side difference */
3074 
3075 		/*
3076 		 * If there's a previous (left) block, select a requested
3077 		 * start block based on it.
3078 		 */
3079 		if (ap->prev.br_startoff != NULLFILEOFF &&
3080 		    !isnullstartblock(ap->prev.br_startblock) &&
3081 		    (prevbno = ap->prev.br_startblock +
3082 			       ap->prev.br_blockcount) &&
3083 		    ISVALID(prevbno, ap->prev.br_startblock)) {
3084 			/*
3085 			 * Calculate gap to end of previous block.
3086 			 */
3087 			adjust = prevdiff = ap->offset -
3088 				(ap->prev.br_startoff +
3089 				 ap->prev.br_blockcount);
3090 			/*
3091 			 * Figure the startblock based on the previous block's
3092 			 * end and the gap size.
3093 			 * Heuristic!
3094 			 * If the gap is large relative to the piece we're
3095 			 * allocating, or using it gives us an invalid block
3096 			 * number, then just use the end of the previous block.
3097 			 */
3098 			if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3099 			    ISVALID(prevbno + prevdiff,
3100 				    ap->prev.br_startblock))
3101 				prevbno += adjust;
3102 			else
3103 				prevdiff += adjust;
3104 			/*
3105 			 * If the firstblock forbids it, can't use it,
3106 			 * must use default.
3107 			 */
3108 			if (!rt && !nullfb &&
3109 			    XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3110 				prevbno = NULLFSBLOCK;
3111 		}
3112 		/*
3113 		 * No previous block or can't follow it, just default.
3114 		 */
3115 		else
3116 			prevbno = NULLFSBLOCK;
3117 		/*
3118 		 * If there's a following (right) block, select a requested
3119 		 * start block based on it.
3120 		 */
3121 		if (!isnullstartblock(ap->got.br_startblock)) {
3122 			/*
3123 			 * Calculate gap to start of next block.
3124 			 */
3125 			adjust = gotdiff = ap->got.br_startoff - ap->offset;
3126 			/*
3127 			 * Figure the startblock based on the next block's
3128 			 * start and the gap size.
3129 			 */
3130 			gotbno = ap->got.br_startblock;
3131 			/*
3132 			 * Heuristic!
3133 			 * If the gap is large relative to the piece we're
3134 			 * allocating, or using it gives us an invalid block
3135 			 * number, then just use the start of the next block
3136 			 * offset by our length.
3137 			 */
3138 			if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3139 			    ISVALID(gotbno - gotdiff, gotbno))
3140 				gotbno -= adjust;
3141 			else if (ISVALID(gotbno - ap->length, gotbno)) {
3142 				gotbno -= ap->length;
3143 				gotdiff += adjust - ap->length;
3144 			} else
3145 				gotdiff += adjust;
3146 			/*
3147 			 * If the firstblock forbids it, can't use it,
3148 			 * must use default.
3149 			 */
3150 			if (!rt && !nullfb &&
3151 			    XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3152 				gotbno = NULLFSBLOCK;
3153 		}
3154 		/*
3155 		 * No next block, just default.
3156 		 */
3157 		else
3158 			gotbno = NULLFSBLOCK;
3159 		/*
3160 		 * If both valid, pick the better one, else the only good
3161 		 * one, else ap->blkno is already set (to 0 or the inode block).
3162 		 */
3163 		if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3164 			ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3165 		else if (prevbno != NULLFSBLOCK)
3166 			ap->blkno = prevbno;
3167 		else if (gotbno != NULLFSBLOCK)
3168 			ap->blkno = gotbno;
3169 	}
3170 #undef ISVALID
3171 }
3172 
3173 static int
3174 xfs_bmap_longest_free_extent(
3175 	struct xfs_trans	*tp,
3176 	xfs_agnumber_t		ag,
3177 	xfs_extlen_t		*blen,
3178 	int			*notinit)
3179 {
3180 	struct xfs_mount	*mp = tp->t_mountp;
3181 	struct xfs_perag	*pag;
3182 	xfs_extlen_t		longest;
3183 	int			error = 0;
3184 
3185 	pag = xfs_perag_get(mp, ag);
3186 	if (!pag->pagf_init) {
3187 		error = xfs_alloc_read_agf(pag, tp, XFS_ALLOC_FLAG_TRYLOCK,
3188 				NULL);
3189 		if (error) {
3190 			/* Couldn't lock the AGF, so skip this AG. */
3191 			if (error == -EAGAIN) {
3192 				*notinit = 1;
3193 				error = 0;
3194 			}
3195 			goto out;
3196 		}
3197 	}
3198 
3199 	longest = xfs_alloc_longest_free_extent(pag,
3200 				xfs_alloc_min_freelist(mp, pag),
3201 				xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
3202 	if (*blen < longest)
3203 		*blen = longest;
3204 
3205 out:
3206 	xfs_perag_put(pag);
3207 	return error;
3208 }
3209 
3210 static void
3211 xfs_bmap_select_minlen(
3212 	struct xfs_bmalloca	*ap,
3213 	struct xfs_alloc_arg	*args,
3214 	xfs_extlen_t		*blen,
3215 	int			notinit)
3216 {
3217 	if (notinit || *blen < ap->minlen) {
3218 		/*
3219 		 * Since we did a BUF_TRYLOCK above, it is possible that
3220 		 * there is space for this request.
3221 		 */
3222 		args->minlen = ap->minlen;
3223 	} else if (*blen < args->maxlen) {
3224 		/*
3225 		 * If the best seen length is less than the request length,
3226 		 * use the best as the minimum.
3227 		 */
3228 		args->minlen = *blen;
3229 	} else {
3230 		/*
3231 		 * Otherwise we've seen an extent as big as maxlen, use that
3232 		 * as the minimum.
3233 		 */
3234 		args->minlen = args->maxlen;
3235 	}
3236 }
3237 
3238 STATIC int
3239 xfs_bmap_btalloc_nullfb(
3240 	struct xfs_bmalloca	*ap,
3241 	struct xfs_alloc_arg	*args,
3242 	xfs_extlen_t		*blen)
3243 {
3244 	struct xfs_mount	*mp = ap->ip->i_mount;
3245 	xfs_agnumber_t		ag, startag;
3246 	int			notinit = 0;
3247 	int			error;
3248 
3249 	args->type = XFS_ALLOCTYPE_START_BNO;
3250 	args->total = ap->total;
3251 
3252 	startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3253 	if (startag == NULLAGNUMBER)
3254 		startag = ag = 0;
3255 
3256 	while (*blen < args->maxlen) {
3257 		error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3258 						     &notinit);
3259 		if (error)
3260 			return error;
3261 
3262 		if (++ag == mp->m_sb.sb_agcount)
3263 			ag = 0;
3264 		if (ag == startag)
3265 			break;
3266 	}
3267 
3268 	xfs_bmap_select_minlen(ap, args, blen, notinit);
3269 	return 0;
3270 }
3271 
3272 STATIC int
3273 xfs_bmap_btalloc_filestreams(
3274 	struct xfs_bmalloca	*ap,
3275 	struct xfs_alloc_arg	*args,
3276 	xfs_extlen_t		*blen)
3277 {
3278 	struct xfs_mount	*mp = ap->ip->i_mount;
3279 	xfs_agnumber_t		ag;
3280 	int			notinit = 0;
3281 	int			error;
3282 
3283 	args->type = XFS_ALLOCTYPE_NEAR_BNO;
3284 	args->total = ap->total;
3285 
3286 	ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3287 	if (ag == NULLAGNUMBER)
3288 		ag = 0;
3289 
3290 	error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3291 	if (error)
3292 		return error;
3293 
3294 	if (*blen < args->maxlen) {
3295 		error = xfs_filestream_new_ag(ap, &ag);
3296 		if (error)
3297 			return error;
3298 
3299 		error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3300 						     &notinit);
3301 		if (error)
3302 			return error;
3303 
3304 	}
3305 
3306 	xfs_bmap_select_minlen(ap, args, blen, notinit);
3307 
3308 	/*
3309 	 * Set the failure fallback case to look in the selected AG as stream
3310 	 * may have moved.
3311 	 */
3312 	ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3313 	return 0;
3314 }
3315 
3316 /* Update all inode and quota accounting for the allocation we just did. */
3317 static void
3318 xfs_bmap_btalloc_accounting(
3319 	struct xfs_bmalloca	*ap,
3320 	struct xfs_alloc_arg	*args)
3321 {
3322 	if (ap->flags & XFS_BMAPI_COWFORK) {
3323 		/*
3324 		 * COW fork blocks are in-core only and thus are treated as
3325 		 * in-core quota reservation (like delalloc blocks) even when
3326 		 * converted to real blocks. The quota reservation is not
3327 		 * accounted to disk until blocks are remapped to the data
3328 		 * fork. So if these blocks were previously delalloc, we
3329 		 * already have quota reservation and there's nothing to do
3330 		 * yet.
3331 		 */
3332 		if (ap->wasdel) {
3333 			xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len);
3334 			return;
3335 		}
3336 
3337 		/*
3338 		 * Otherwise, we've allocated blocks in a hole. The transaction
3339 		 * has acquired in-core quota reservation for this extent.
3340 		 * Rather than account these as real blocks, however, we reduce
3341 		 * the transaction quota reservation based on the allocation.
3342 		 * This essentially transfers the transaction quota reservation
3343 		 * to that of a delalloc extent.
3344 		 */
3345 		ap->ip->i_delayed_blks += args->len;
3346 		xfs_trans_mod_dquot_byino(ap->tp, ap->ip, XFS_TRANS_DQ_RES_BLKS,
3347 				-(long)args->len);
3348 		return;
3349 	}
3350 
3351 	/* data/attr fork only */
3352 	ap->ip->i_nblocks += args->len;
3353 	xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3354 	if (ap->wasdel) {
3355 		ap->ip->i_delayed_blks -= args->len;
3356 		xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len);
3357 	}
3358 	xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3359 		ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT : XFS_TRANS_DQ_BCOUNT,
3360 		args->len);
3361 }
3362 
3363 static int
3364 xfs_bmap_compute_alignments(
3365 	struct xfs_bmalloca	*ap,
3366 	struct xfs_alloc_arg	*args)
3367 {
3368 	struct xfs_mount	*mp = args->mp;
3369 	xfs_extlen_t		align = 0; /* minimum allocation alignment */
3370 	int			stripe_align = 0;
3371 
3372 	/* stripe alignment for allocation is determined by mount parameters */
3373 	if (mp->m_swidth && xfs_has_swalloc(mp))
3374 		stripe_align = mp->m_swidth;
3375 	else if (mp->m_dalign)
3376 		stripe_align = mp->m_dalign;
3377 
3378 	if (ap->flags & XFS_BMAPI_COWFORK)
3379 		align = xfs_get_cowextsz_hint(ap->ip);
3380 	else if (ap->datatype & XFS_ALLOC_USERDATA)
3381 		align = xfs_get_extsz_hint(ap->ip);
3382 	if (align) {
3383 		if (xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, align, 0,
3384 					ap->eof, 0, ap->conv, &ap->offset,
3385 					&ap->length))
3386 			ASSERT(0);
3387 		ASSERT(ap->length);
3388 	}
3389 
3390 	/* apply extent size hints if obtained earlier */
3391 	if (align) {
3392 		args->prod = align;
3393 		div_u64_rem(ap->offset, args->prod, &args->mod);
3394 		if (args->mod)
3395 			args->mod = args->prod - args->mod;
3396 	} else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3397 		args->prod = 1;
3398 		args->mod = 0;
3399 	} else {
3400 		args->prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3401 		div_u64_rem(ap->offset, args->prod, &args->mod);
3402 		if (args->mod)
3403 			args->mod = args->prod - args->mod;
3404 	}
3405 
3406 	return stripe_align;
3407 }
3408 
3409 static void
3410 xfs_bmap_process_allocated_extent(
3411 	struct xfs_bmalloca	*ap,
3412 	struct xfs_alloc_arg	*args,
3413 	xfs_fileoff_t		orig_offset,
3414 	xfs_extlen_t		orig_length)
3415 {
3416 	int			nullfb;
3417 
3418 	nullfb = ap->tp->t_firstblock == NULLFSBLOCK;
3419 
3420 	/*
3421 	 * check the allocation happened at the same or higher AG than
3422 	 * the first block that was allocated.
3423 	 */
3424 	ASSERT(nullfb ||
3425 		XFS_FSB_TO_AGNO(args->mp, ap->tp->t_firstblock) <=
3426 		XFS_FSB_TO_AGNO(args->mp, args->fsbno));
3427 
3428 	ap->blkno = args->fsbno;
3429 	if (nullfb)
3430 		ap->tp->t_firstblock = args->fsbno;
3431 	ap->length = args->len;
3432 	/*
3433 	 * If the extent size hint is active, we tried to round the
3434 	 * caller's allocation request offset down to extsz and the
3435 	 * length up to another extsz boundary.  If we found a free
3436 	 * extent we mapped it in starting at this new offset.  If the
3437 	 * newly mapped space isn't long enough to cover any of the
3438 	 * range of offsets that was originally requested, move the
3439 	 * mapping up so that we can fill as much of the caller's
3440 	 * original request as possible.  Free space is apparently
3441 	 * very fragmented so we're unlikely to be able to satisfy the
3442 	 * hints anyway.
3443 	 */
3444 	if (ap->length <= orig_length)
3445 		ap->offset = orig_offset;
3446 	else if (ap->offset + ap->length < orig_offset + orig_length)
3447 		ap->offset = orig_offset + orig_length - ap->length;
3448 	xfs_bmap_btalloc_accounting(ap, args);
3449 }
3450 
3451 #ifdef DEBUG
3452 static int
3453 xfs_bmap_exact_minlen_extent_alloc(
3454 	struct xfs_bmalloca	*ap)
3455 {
3456 	struct xfs_mount	*mp = ap->ip->i_mount;
3457 	struct xfs_alloc_arg	args = { .tp = ap->tp, .mp = mp };
3458 	xfs_fileoff_t		orig_offset;
3459 	xfs_extlen_t		orig_length;
3460 	int			error;
3461 
3462 	ASSERT(ap->length);
3463 
3464 	if (ap->minlen != 1) {
3465 		ap->blkno = NULLFSBLOCK;
3466 		ap->length = 0;
3467 		return 0;
3468 	}
3469 
3470 	orig_offset = ap->offset;
3471 	orig_length = ap->length;
3472 
3473 	args.alloc_minlen_only = 1;
3474 
3475 	xfs_bmap_compute_alignments(ap, &args);
3476 
3477 	if (ap->tp->t_firstblock == NULLFSBLOCK) {
3478 		/*
3479 		 * Unlike the longest extent available in an AG, we don't track
3480 		 * the length of an AG's shortest extent.
3481 		 * XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT is a debug only knob and
3482 		 * hence we can afford to start traversing from the 0th AG since
3483 		 * we need not be concerned about a drop in performance in
3484 		 * "debug only" code paths.
3485 		 */
3486 		ap->blkno = XFS_AGB_TO_FSB(mp, 0, 0);
3487 	} else {
3488 		ap->blkno = ap->tp->t_firstblock;
3489 	}
3490 
3491 	args.fsbno = ap->blkno;
3492 	args.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
3493 	args.type = XFS_ALLOCTYPE_FIRST_AG;
3494 	args.minlen = args.maxlen = ap->minlen;
3495 	args.total = ap->total;
3496 
3497 	args.alignment = 1;
3498 	args.minalignslop = 0;
3499 
3500 	args.minleft = ap->minleft;
3501 	args.wasdel = ap->wasdel;
3502 	args.resv = XFS_AG_RESV_NONE;
3503 	args.datatype = ap->datatype;
3504 
3505 	error = xfs_alloc_vextent(&args);
3506 	if (error)
3507 		return error;
3508 
3509 	if (args.fsbno != NULLFSBLOCK) {
3510 		xfs_bmap_process_allocated_extent(ap, &args, orig_offset,
3511 			orig_length);
3512 	} else {
3513 		ap->blkno = NULLFSBLOCK;
3514 		ap->length = 0;
3515 	}
3516 
3517 	return 0;
3518 }
3519 #else
3520 
3521 #define xfs_bmap_exact_minlen_extent_alloc(bma) (-EFSCORRUPTED)
3522 
3523 #endif
3524 
3525 STATIC int
3526 xfs_bmap_btalloc(
3527 	struct xfs_bmalloca	*ap)
3528 {
3529 	struct xfs_mount	*mp = ap->ip->i_mount;
3530 	struct xfs_alloc_arg	args = { .tp = ap->tp, .mp = mp };
3531 	xfs_alloctype_t		atype = 0;
3532 	xfs_agnumber_t		fb_agno;	/* ag number of ap->firstblock */
3533 	xfs_agnumber_t		ag;
3534 	xfs_fileoff_t		orig_offset;
3535 	xfs_extlen_t		orig_length;
3536 	xfs_extlen_t		blen;
3537 	xfs_extlen_t		nextminlen = 0;
3538 	int			nullfb; /* true if ap->firstblock isn't set */
3539 	int			isaligned;
3540 	int			tryagain;
3541 	int			error;
3542 	int			stripe_align;
3543 
3544 	ASSERT(ap->length);
3545 	orig_offset = ap->offset;
3546 	orig_length = ap->length;
3547 
3548 	stripe_align = xfs_bmap_compute_alignments(ap, &args);
3549 
3550 	nullfb = ap->tp->t_firstblock == NULLFSBLOCK;
3551 	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp,
3552 							ap->tp->t_firstblock);
3553 	if (nullfb) {
3554 		if ((ap->datatype & XFS_ALLOC_USERDATA) &&
3555 		    xfs_inode_is_filestream(ap->ip)) {
3556 			ag = xfs_filestream_lookup_ag(ap->ip);
3557 			ag = (ag != NULLAGNUMBER) ? ag : 0;
3558 			ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3559 		} else {
3560 			ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3561 		}
3562 	} else
3563 		ap->blkno = ap->tp->t_firstblock;
3564 
3565 	xfs_bmap_adjacent(ap);
3566 
3567 	/*
3568 	 * If allowed, use ap->blkno; otherwise must use firstblock since
3569 	 * it's in the right allocation group.
3570 	 */
3571 	if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3572 		;
3573 	else
3574 		ap->blkno = ap->tp->t_firstblock;
3575 	/*
3576 	 * Normal allocation, done through xfs_alloc_vextent.
3577 	 */
3578 	tryagain = isaligned = 0;
3579 	args.fsbno = ap->blkno;
3580 	args.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
3581 
3582 	/* Trim the allocation back to the maximum an AG can fit. */
3583 	args.maxlen = min(ap->length, mp->m_ag_max_usable);
3584 	blen = 0;
3585 	if (nullfb) {
3586 		/*
3587 		 * Search for an allocation group with a single extent large
3588 		 * enough for the request.  If one isn't found, then adjust
3589 		 * the minimum allocation size to the largest space found.
3590 		 */
3591 		if ((ap->datatype & XFS_ALLOC_USERDATA) &&
3592 		    xfs_inode_is_filestream(ap->ip))
3593 			error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3594 		else
3595 			error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3596 		if (error)
3597 			return error;
3598 	} else if (ap->tp->t_flags & XFS_TRANS_LOWMODE) {
3599 		if (xfs_inode_is_filestream(ap->ip))
3600 			args.type = XFS_ALLOCTYPE_FIRST_AG;
3601 		else
3602 			args.type = XFS_ALLOCTYPE_START_BNO;
3603 		args.total = args.minlen = ap->minlen;
3604 	} else {
3605 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
3606 		args.total = ap->total;
3607 		args.minlen = ap->minlen;
3608 	}
3609 
3610 	/*
3611 	 * If we are not low on available data blocks, and the underlying
3612 	 * logical volume manager is a stripe, and the file offset is zero then
3613 	 * try to allocate data blocks on stripe unit boundary. NOTE: ap->aeof
3614 	 * is only set if the allocation length is >= the stripe unit and the
3615 	 * allocation offset is at the end of file.
3616 	 */
3617 	if (!(ap->tp->t_flags & XFS_TRANS_LOWMODE) && ap->aeof) {
3618 		if (!ap->offset) {
3619 			args.alignment = stripe_align;
3620 			atype = args.type;
3621 			isaligned = 1;
3622 			/*
3623 			 * Adjust minlen to try and preserve alignment if we
3624 			 * can't guarantee an aligned maxlen extent.
3625 			 */
3626 			if (blen > args.alignment &&
3627 			    blen <= args.maxlen + args.alignment)
3628 				args.minlen = blen - args.alignment;
3629 			args.minalignslop = 0;
3630 		} else {
3631 			/*
3632 			 * First try an exact bno allocation.
3633 			 * If it fails then do a near or start bno
3634 			 * allocation with alignment turned on.
3635 			 */
3636 			atype = args.type;
3637 			tryagain = 1;
3638 			args.type = XFS_ALLOCTYPE_THIS_BNO;
3639 			args.alignment = 1;
3640 			/*
3641 			 * Compute the minlen+alignment for the
3642 			 * next case.  Set slop so that the value
3643 			 * of minlen+alignment+slop doesn't go up
3644 			 * between the calls.
3645 			 */
3646 			if (blen > stripe_align && blen <= args.maxlen)
3647 				nextminlen = blen - stripe_align;
3648 			else
3649 				nextminlen = args.minlen;
3650 			if (nextminlen + stripe_align > args.minlen + 1)
3651 				args.minalignslop =
3652 					nextminlen + stripe_align -
3653 					args.minlen - 1;
3654 			else
3655 				args.minalignslop = 0;
3656 		}
3657 	} else {
3658 		args.alignment = 1;
3659 		args.minalignslop = 0;
3660 	}
3661 	args.minleft = ap->minleft;
3662 	args.wasdel = ap->wasdel;
3663 	args.resv = XFS_AG_RESV_NONE;
3664 	args.datatype = ap->datatype;
3665 
3666 	error = xfs_alloc_vextent(&args);
3667 	if (error)
3668 		return error;
3669 
3670 	if (tryagain && args.fsbno == NULLFSBLOCK) {
3671 		/*
3672 		 * Exact allocation failed. Now try with alignment
3673 		 * turned on.
3674 		 */
3675 		args.type = atype;
3676 		args.fsbno = ap->blkno;
3677 		args.alignment = stripe_align;
3678 		args.minlen = nextminlen;
3679 		args.minalignslop = 0;
3680 		isaligned = 1;
3681 		if ((error = xfs_alloc_vextent(&args)))
3682 			return error;
3683 	}
3684 	if (isaligned && args.fsbno == NULLFSBLOCK) {
3685 		/*
3686 		 * allocation failed, so turn off alignment and
3687 		 * try again.
3688 		 */
3689 		args.type = atype;
3690 		args.fsbno = ap->blkno;
3691 		args.alignment = 0;
3692 		if ((error = xfs_alloc_vextent(&args)))
3693 			return error;
3694 	}
3695 	if (args.fsbno == NULLFSBLOCK && nullfb &&
3696 	    args.minlen > ap->minlen) {
3697 		args.minlen = ap->minlen;
3698 		args.type = XFS_ALLOCTYPE_START_BNO;
3699 		args.fsbno = ap->blkno;
3700 		if ((error = xfs_alloc_vextent(&args)))
3701 			return error;
3702 	}
3703 	if (args.fsbno == NULLFSBLOCK && nullfb) {
3704 		args.fsbno = 0;
3705 		args.type = XFS_ALLOCTYPE_FIRST_AG;
3706 		args.total = ap->minlen;
3707 		if ((error = xfs_alloc_vextent(&args)))
3708 			return error;
3709 		ap->tp->t_flags |= XFS_TRANS_LOWMODE;
3710 	}
3711 
3712 	if (args.fsbno != NULLFSBLOCK) {
3713 		xfs_bmap_process_allocated_extent(ap, &args, orig_offset,
3714 			orig_length);
3715 	} else {
3716 		ap->blkno = NULLFSBLOCK;
3717 		ap->length = 0;
3718 	}
3719 	return 0;
3720 }
3721 
3722 /* Trim extent to fit a logical block range. */
3723 void
3724 xfs_trim_extent(
3725 	struct xfs_bmbt_irec	*irec,
3726 	xfs_fileoff_t		bno,
3727 	xfs_filblks_t		len)
3728 {
3729 	xfs_fileoff_t		distance;
3730 	xfs_fileoff_t		end = bno + len;
3731 
3732 	if (irec->br_startoff + irec->br_blockcount <= bno ||
3733 	    irec->br_startoff >= end) {
3734 		irec->br_blockcount = 0;
3735 		return;
3736 	}
3737 
3738 	if (irec->br_startoff < bno) {
3739 		distance = bno - irec->br_startoff;
3740 		if (isnullstartblock(irec->br_startblock))
3741 			irec->br_startblock = DELAYSTARTBLOCK;
3742 		if (irec->br_startblock != DELAYSTARTBLOCK &&
3743 		    irec->br_startblock != HOLESTARTBLOCK)
3744 			irec->br_startblock += distance;
3745 		irec->br_startoff += distance;
3746 		irec->br_blockcount -= distance;
3747 	}
3748 
3749 	if (end < irec->br_startoff + irec->br_blockcount) {
3750 		distance = irec->br_startoff + irec->br_blockcount - end;
3751 		irec->br_blockcount -= distance;
3752 	}
3753 }
3754 
3755 /*
3756  * Trim the returned map to the required bounds
3757  */
3758 STATIC void
3759 xfs_bmapi_trim_map(
3760 	struct xfs_bmbt_irec	*mval,
3761 	struct xfs_bmbt_irec	*got,
3762 	xfs_fileoff_t		*bno,
3763 	xfs_filblks_t		len,
3764 	xfs_fileoff_t		obno,
3765 	xfs_fileoff_t		end,
3766 	int			n,
3767 	uint32_t		flags)
3768 {
3769 	if ((flags & XFS_BMAPI_ENTIRE) ||
3770 	    got->br_startoff + got->br_blockcount <= obno) {
3771 		*mval = *got;
3772 		if (isnullstartblock(got->br_startblock))
3773 			mval->br_startblock = DELAYSTARTBLOCK;
3774 		return;
3775 	}
3776 
3777 	if (obno > *bno)
3778 		*bno = obno;
3779 	ASSERT((*bno >= obno) || (n == 0));
3780 	ASSERT(*bno < end);
3781 	mval->br_startoff = *bno;
3782 	if (isnullstartblock(got->br_startblock))
3783 		mval->br_startblock = DELAYSTARTBLOCK;
3784 	else
3785 		mval->br_startblock = got->br_startblock +
3786 					(*bno - got->br_startoff);
3787 	/*
3788 	 * Return the minimum of what we got and what we asked for for
3789 	 * the length.  We can use the len variable here because it is
3790 	 * modified below and we could have been there before coming
3791 	 * here if the first part of the allocation didn't overlap what
3792 	 * was asked for.
3793 	 */
3794 	mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3795 			got->br_blockcount - (*bno - got->br_startoff));
3796 	mval->br_state = got->br_state;
3797 	ASSERT(mval->br_blockcount <= len);
3798 	return;
3799 }
3800 
3801 /*
3802  * Update and validate the extent map to return
3803  */
3804 STATIC void
3805 xfs_bmapi_update_map(
3806 	struct xfs_bmbt_irec	**map,
3807 	xfs_fileoff_t		*bno,
3808 	xfs_filblks_t		*len,
3809 	xfs_fileoff_t		obno,
3810 	xfs_fileoff_t		end,
3811 	int			*n,
3812 	uint32_t		flags)
3813 {
3814 	xfs_bmbt_irec_t	*mval = *map;
3815 
3816 	ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3817 	       ((mval->br_startoff + mval->br_blockcount) <= end));
3818 	ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3819 	       (mval->br_startoff < obno));
3820 
3821 	*bno = mval->br_startoff + mval->br_blockcount;
3822 	*len = end - *bno;
3823 	if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3824 		/* update previous map with new information */
3825 		ASSERT(mval->br_startblock == mval[-1].br_startblock);
3826 		ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3827 		ASSERT(mval->br_state == mval[-1].br_state);
3828 		mval[-1].br_blockcount = mval->br_blockcount;
3829 		mval[-1].br_state = mval->br_state;
3830 	} else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3831 		   mval[-1].br_startblock != DELAYSTARTBLOCK &&
3832 		   mval[-1].br_startblock != HOLESTARTBLOCK &&
3833 		   mval->br_startblock == mval[-1].br_startblock +
3834 					  mval[-1].br_blockcount &&
3835 		   mval[-1].br_state == mval->br_state) {
3836 		ASSERT(mval->br_startoff ==
3837 		       mval[-1].br_startoff + mval[-1].br_blockcount);
3838 		mval[-1].br_blockcount += mval->br_blockcount;
3839 	} else if (*n > 0 &&
3840 		   mval->br_startblock == DELAYSTARTBLOCK &&
3841 		   mval[-1].br_startblock == DELAYSTARTBLOCK &&
3842 		   mval->br_startoff ==
3843 		   mval[-1].br_startoff + mval[-1].br_blockcount) {
3844 		mval[-1].br_blockcount += mval->br_blockcount;
3845 		mval[-1].br_state = mval->br_state;
3846 	} else if (!((*n == 0) &&
3847 		     ((mval->br_startoff + mval->br_blockcount) <=
3848 		      obno))) {
3849 		mval++;
3850 		(*n)++;
3851 	}
3852 	*map = mval;
3853 }
3854 
3855 /*
3856  * Map file blocks to filesystem blocks without allocation.
3857  */
3858 int
3859 xfs_bmapi_read(
3860 	struct xfs_inode	*ip,
3861 	xfs_fileoff_t		bno,
3862 	xfs_filblks_t		len,
3863 	struct xfs_bmbt_irec	*mval,
3864 	int			*nmap,
3865 	uint32_t		flags)
3866 {
3867 	struct xfs_mount	*mp = ip->i_mount;
3868 	int			whichfork = xfs_bmapi_whichfork(flags);
3869 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
3870 	struct xfs_bmbt_irec	got;
3871 	xfs_fileoff_t		obno;
3872 	xfs_fileoff_t		end;
3873 	struct xfs_iext_cursor	icur;
3874 	int			error;
3875 	bool			eof = false;
3876 	int			n = 0;
3877 
3878 	ASSERT(*nmap >= 1);
3879 	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_ENTIRE)));
3880 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
3881 
3882 	if (WARN_ON_ONCE(!ifp))
3883 		return -EFSCORRUPTED;
3884 
3885 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
3886 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT))
3887 		return -EFSCORRUPTED;
3888 
3889 	if (xfs_is_shutdown(mp))
3890 		return -EIO;
3891 
3892 	XFS_STATS_INC(mp, xs_blk_mapr);
3893 
3894 	error = xfs_iread_extents(NULL, ip, whichfork);
3895 	if (error)
3896 		return error;
3897 
3898 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got))
3899 		eof = true;
3900 	end = bno + len;
3901 	obno = bno;
3902 
3903 	while (bno < end && n < *nmap) {
3904 		/* Reading past eof, act as though there's a hole up to end. */
3905 		if (eof)
3906 			got.br_startoff = end;
3907 		if (got.br_startoff > bno) {
3908 			/* Reading in a hole.  */
3909 			mval->br_startoff = bno;
3910 			mval->br_startblock = HOLESTARTBLOCK;
3911 			mval->br_blockcount =
3912 				XFS_FILBLKS_MIN(len, got.br_startoff - bno);
3913 			mval->br_state = XFS_EXT_NORM;
3914 			bno += mval->br_blockcount;
3915 			len -= mval->br_blockcount;
3916 			mval++;
3917 			n++;
3918 			continue;
3919 		}
3920 
3921 		/* set up the extent map to return. */
3922 		xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
3923 		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
3924 
3925 		/* If we're done, stop now. */
3926 		if (bno >= end || n >= *nmap)
3927 			break;
3928 
3929 		/* Else go on to the next record. */
3930 		if (!xfs_iext_next_extent(ifp, &icur, &got))
3931 			eof = true;
3932 	}
3933 	*nmap = n;
3934 	return 0;
3935 }
3936 
3937 /*
3938  * Add a delayed allocation extent to an inode. Blocks are reserved from the
3939  * global pool and the extent inserted into the inode in-core extent tree.
3940  *
3941  * On entry, got refers to the first extent beyond the offset of the extent to
3942  * allocate or eof is specified if no such extent exists. On return, got refers
3943  * to the extent record that was inserted to the inode fork.
3944  *
3945  * Note that the allocated extent may have been merged with contiguous extents
3946  * during insertion into the inode fork. Thus, got does not reflect the current
3947  * state of the inode fork on return. If necessary, the caller can use lastx to
3948  * look up the updated record in the inode fork.
3949  */
3950 int
3951 xfs_bmapi_reserve_delalloc(
3952 	struct xfs_inode	*ip,
3953 	int			whichfork,
3954 	xfs_fileoff_t		off,
3955 	xfs_filblks_t		len,
3956 	xfs_filblks_t		prealloc,
3957 	struct xfs_bmbt_irec	*got,
3958 	struct xfs_iext_cursor	*icur,
3959 	int			eof)
3960 {
3961 	struct xfs_mount	*mp = ip->i_mount;
3962 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
3963 	xfs_extlen_t		alen;
3964 	xfs_extlen_t		indlen;
3965 	int			error;
3966 	xfs_fileoff_t		aoff = off;
3967 
3968 	/*
3969 	 * Cap the alloc length. Keep track of prealloc so we know whether to
3970 	 * tag the inode before we return.
3971 	 */
3972 	alen = XFS_FILBLKS_MIN(len + prealloc, XFS_MAX_BMBT_EXTLEN);
3973 	if (!eof)
3974 		alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
3975 	if (prealloc && alen >= len)
3976 		prealloc = alen - len;
3977 
3978 	/* Figure out the extent size, adjust alen */
3979 	if (whichfork == XFS_COW_FORK) {
3980 		struct xfs_bmbt_irec	prev;
3981 		xfs_extlen_t		extsz = xfs_get_cowextsz_hint(ip);
3982 
3983 		if (!xfs_iext_peek_prev_extent(ifp, icur, &prev))
3984 			prev.br_startoff = NULLFILEOFF;
3985 
3986 		error = xfs_bmap_extsize_align(mp, got, &prev, extsz, 0, eof,
3987 					       1, 0, &aoff, &alen);
3988 		ASSERT(!error);
3989 	}
3990 
3991 	/*
3992 	 * Make a transaction-less quota reservation for delayed allocation
3993 	 * blocks.  This number gets adjusted later.  We return if we haven't
3994 	 * allocated blocks already inside this loop.
3995 	 */
3996 	error = xfs_quota_reserve_blkres(ip, alen);
3997 	if (error)
3998 		return error;
3999 
4000 	/*
4001 	 * Split changing sb for alen and indlen since they could be coming
4002 	 * from different places.
4003 	 */
4004 	indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4005 	ASSERT(indlen > 0);
4006 
4007 	error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
4008 	if (error)
4009 		goto out_unreserve_quota;
4010 
4011 	error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
4012 	if (error)
4013 		goto out_unreserve_blocks;
4014 
4015 
4016 	ip->i_delayed_blks += alen;
4017 	xfs_mod_delalloc(ip->i_mount, alen + indlen);
4018 
4019 	got->br_startoff = aoff;
4020 	got->br_startblock = nullstartblock(indlen);
4021 	got->br_blockcount = alen;
4022 	got->br_state = XFS_EXT_NORM;
4023 
4024 	xfs_bmap_add_extent_hole_delay(ip, whichfork, icur, got);
4025 
4026 	/*
4027 	 * Tag the inode if blocks were preallocated. Note that COW fork
4028 	 * preallocation can occur at the start or end of the extent, even when
4029 	 * prealloc == 0, so we must also check the aligned offset and length.
4030 	 */
4031 	if (whichfork == XFS_DATA_FORK && prealloc)
4032 		xfs_inode_set_eofblocks_tag(ip);
4033 	if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len))
4034 		xfs_inode_set_cowblocks_tag(ip);
4035 
4036 	return 0;
4037 
4038 out_unreserve_blocks:
4039 	xfs_mod_fdblocks(mp, alen, false);
4040 out_unreserve_quota:
4041 	if (XFS_IS_QUOTA_ON(mp))
4042 		xfs_quota_unreserve_blkres(ip, alen);
4043 	return error;
4044 }
4045 
4046 static int
4047 xfs_bmap_alloc_userdata(
4048 	struct xfs_bmalloca	*bma)
4049 {
4050 	struct xfs_mount	*mp = bma->ip->i_mount;
4051 	int			whichfork = xfs_bmapi_whichfork(bma->flags);
4052 	int			error;
4053 
4054 	/*
4055 	 * Set the data type being allocated. For the data fork, the first data
4056 	 * in the file is treated differently to all other allocations. For the
4057 	 * attribute fork, we only need to ensure the allocated range is not on
4058 	 * the busy list.
4059 	 */
4060 	bma->datatype = XFS_ALLOC_NOBUSY;
4061 	if (whichfork == XFS_DATA_FORK) {
4062 		bma->datatype |= XFS_ALLOC_USERDATA;
4063 		if (bma->offset == 0)
4064 			bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA;
4065 
4066 		if (mp->m_dalign && bma->length >= mp->m_dalign) {
4067 			error = xfs_bmap_isaeof(bma, whichfork);
4068 			if (error)
4069 				return error;
4070 		}
4071 
4072 		if (XFS_IS_REALTIME_INODE(bma->ip))
4073 			return xfs_bmap_rtalloc(bma);
4074 	}
4075 
4076 	if (unlikely(XFS_TEST_ERROR(false, mp,
4077 			XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT)))
4078 		return xfs_bmap_exact_minlen_extent_alloc(bma);
4079 
4080 	return xfs_bmap_btalloc(bma);
4081 }
4082 
4083 static int
4084 xfs_bmapi_allocate(
4085 	struct xfs_bmalloca	*bma)
4086 {
4087 	struct xfs_mount	*mp = bma->ip->i_mount;
4088 	int			whichfork = xfs_bmapi_whichfork(bma->flags);
4089 	struct xfs_ifork	*ifp = xfs_ifork_ptr(bma->ip, whichfork);
4090 	int			tmp_logflags = 0;
4091 	int			error;
4092 
4093 	ASSERT(bma->length > 0);
4094 
4095 	/*
4096 	 * For the wasdelay case, we could also just allocate the stuff asked
4097 	 * for in this bmap call but that wouldn't be as good.
4098 	 */
4099 	if (bma->wasdel) {
4100 		bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4101 		bma->offset = bma->got.br_startoff;
4102 		if (!xfs_iext_peek_prev_extent(ifp, &bma->icur, &bma->prev))
4103 			bma->prev.br_startoff = NULLFILEOFF;
4104 	} else {
4105 		bma->length = XFS_FILBLKS_MIN(bma->length, XFS_MAX_BMBT_EXTLEN);
4106 		if (!bma->eof)
4107 			bma->length = XFS_FILBLKS_MIN(bma->length,
4108 					bma->got.br_startoff - bma->offset);
4109 	}
4110 
4111 	if (bma->flags & XFS_BMAPI_CONTIG)
4112 		bma->minlen = bma->length;
4113 	else
4114 		bma->minlen = 1;
4115 
4116 	if (bma->flags & XFS_BMAPI_METADATA) {
4117 		if (unlikely(XFS_TEST_ERROR(false, mp,
4118 				XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT)))
4119 			error = xfs_bmap_exact_minlen_extent_alloc(bma);
4120 		else
4121 			error = xfs_bmap_btalloc(bma);
4122 	} else {
4123 		error = xfs_bmap_alloc_userdata(bma);
4124 	}
4125 	if (error || bma->blkno == NULLFSBLOCK)
4126 		return error;
4127 
4128 	if (bma->flags & XFS_BMAPI_ZERO) {
4129 		error = xfs_zero_extent(bma->ip, bma->blkno, bma->length);
4130 		if (error)
4131 			return error;
4132 	}
4133 
4134 	if (ifp->if_format == XFS_DINODE_FMT_BTREE && !bma->cur)
4135 		bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4136 	/*
4137 	 * Bump the number of extents we've allocated
4138 	 * in this call.
4139 	 */
4140 	bma->nallocs++;
4141 
4142 	if (bma->cur)
4143 		bma->cur->bc_ino.flags =
4144 			bma->wasdel ? XFS_BTCUR_BMBT_WASDEL : 0;
4145 
4146 	bma->got.br_startoff = bma->offset;
4147 	bma->got.br_startblock = bma->blkno;
4148 	bma->got.br_blockcount = bma->length;
4149 	bma->got.br_state = XFS_EXT_NORM;
4150 
4151 	if (bma->flags & XFS_BMAPI_PREALLOC)
4152 		bma->got.br_state = XFS_EXT_UNWRITTEN;
4153 
4154 	if (bma->wasdel)
4155 		error = xfs_bmap_add_extent_delay_real(bma, whichfork);
4156 	else
4157 		error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip,
4158 				whichfork, &bma->icur, &bma->cur, &bma->got,
4159 				&bma->logflags, bma->flags);
4160 
4161 	bma->logflags |= tmp_logflags;
4162 	if (error)
4163 		return error;
4164 
4165 	/*
4166 	 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4167 	 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4168 	 * the neighbouring ones.
4169 	 */
4170 	xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4171 
4172 	ASSERT(bma->got.br_startoff <= bma->offset);
4173 	ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4174 	       bma->offset + bma->length);
4175 	ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4176 	       bma->got.br_state == XFS_EXT_UNWRITTEN);
4177 	return 0;
4178 }
4179 
4180 STATIC int
4181 xfs_bmapi_convert_unwritten(
4182 	struct xfs_bmalloca	*bma,
4183 	struct xfs_bmbt_irec	*mval,
4184 	xfs_filblks_t		len,
4185 	uint32_t		flags)
4186 {
4187 	int			whichfork = xfs_bmapi_whichfork(flags);
4188 	struct xfs_ifork	*ifp = xfs_ifork_ptr(bma->ip, whichfork);
4189 	int			tmp_logflags = 0;
4190 	int			error;
4191 
4192 	/* check if we need to do unwritten->real conversion */
4193 	if (mval->br_state == XFS_EXT_UNWRITTEN &&
4194 	    (flags & XFS_BMAPI_PREALLOC))
4195 		return 0;
4196 
4197 	/* check if we need to do real->unwritten conversion */
4198 	if (mval->br_state == XFS_EXT_NORM &&
4199 	    (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4200 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4201 		return 0;
4202 
4203 	/*
4204 	 * Modify (by adding) the state flag, if writing.
4205 	 */
4206 	ASSERT(mval->br_blockcount <= len);
4207 	if (ifp->if_format == XFS_DINODE_FMT_BTREE && !bma->cur) {
4208 		bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4209 					bma->ip, whichfork);
4210 	}
4211 	mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4212 				? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4213 
4214 	/*
4215 	 * Before insertion into the bmbt, zero the range being converted
4216 	 * if required.
4217 	 */
4218 	if (flags & XFS_BMAPI_ZERO) {
4219 		error = xfs_zero_extent(bma->ip, mval->br_startblock,
4220 					mval->br_blockcount);
4221 		if (error)
4222 			return error;
4223 	}
4224 
4225 	error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork,
4226 			&bma->icur, &bma->cur, mval, &tmp_logflags);
4227 	/*
4228 	 * Log the inode core unconditionally in the unwritten extent conversion
4229 	 * path because the conversion might not have done so (e.g., if the
4230 	 * extent count hasn't changed). We need to make sure the inode is dirty
4231 	 * in the transaction for the sake of fsync(), even if nothing has
4232 	 * changed, because fsync() will not force the log for this transaction
4233 	 * unless it sees the inode pinned.
4234 	 *
4235 	 * Note: If we're only converting cow fork extents, there aren't
4236 	 * any on-disk updates to make, so we don't need to log anything.
4237 	 */
4238 	if (whichfork != XFS_COW_FORK)
4239 		bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4240 	if (error)
4241 		return error;
4242 
4243 	/*
4244 	 * Update our extent pointer, given that
4245 	 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4246 	 * of the neighbouring ones.
4247 	 */
4248 	xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4249 
4250 	/*
4251 	 * We may have combined previously unwritten space with written space,
4252 	 * so generate another request.
4253 	 */
4254 	if (mval->br_blockcount < len)
4255 		return -EAGAIN;
4256 	return 0;
4257 }
4258 
4259 static inline xfs_extlen_t
4260 xfs_bmapi_minleft(
4261 	struct xfs_trans	*tp,
4262 	struct xfs_inode	*ip,
4263 	int			fork)
4264 {
4265 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, fork);
4266 
4267 	if (tp && tp->t_firstblock != NULLFSBLOCK)
4268 		return 0;
4269 	if (ifp->if_format != XFS_DINODE_FMT_BTREE)
4270 		return 1;
4271 	return be16_to_cpu(ifp->if_broot->bb_level) + 1;
4272 }
4273 
4274 /*
4275  * Log whatever the flags say, even if error.  Otherwise we might miss detecting
4276  * a case where the data is changed, there's an error, and it's not logged so we
4277  * don't shutdown when we should.  Don't bother logging extents/btree changes if
4278  * we converted to the other format.
4279  */
4280 static void
4281 xfs_bmapi_finish(
4282 	struct xfs_bmalloca	*bma,
4283 	int			whichfork,
4284 	int			error)
4285 {
4286 	struct xfs_ifork	*ifp = xfs_ifork_ptr(bma->ip, whichfork);
4287 
4288 	if ((bma->logflags & xfs_ilog_fext(whichfork)) &&
4289 	    ifp->if_format != XFS_DINODE_FMT_EXTENTS)
4290 		bma->logflags &= ~xfs_ilog_fext(whichfork);
4291 	else if ((bma->logflags & xfs_ilog_fbroot(whichfork)) &&
4292 		 ifp->if_format != XFS_DINODE_FMT_BTREE)
4293 		bma->logflags &= ~xfs_ilog_fbroot(whichfork);
4294 
4295 	if (bma->logflags)
4296 		xfs_trans_log_inode(bma->tp, bma->ip, bma->logflags);
4297 	if (bma->cur)
4298 		xfs_btree_del_cursor(bma->cur, error);
4299 }
4300 
4301 /*
4302  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4303  * extent state if necessary.  Details behaviour is controlled by the flags
4304  * parameter.  Only allocates blocks from a single allocation group, to avoid
4305  * locking problems.
4306  */
4307 int
4308 xfs_bmapi_write(
4309 	struct xfs_trans	*tp,		/* transaction pointer */
4310 	struct xfs_inode	*ip,		/* incore inode */
4311 	xfs_fileoff_t		bno,		/* starting file offs. mapped */
4312 	xfs_filblks_t		len,		/* length to map in file */
4313 	uint32_t		flags,		/* XFS_BMAPI_... */
4314 	xfs_extlen_t		total,		/* total blocks needed */
4315 	struct xfs_bmbt_irec	*mval,		/* output: map values */
4316 	int			*nmap)		/* i/o: mval size/count */
4317 {
4318 	struct xfs_bmalloca	bma = {
4319 		.tp		= tp,
4320 		.ip		= ip,
4321 		.total		= total,
4322 	};
4323 	struct xfs_mount	*mp = ip->i_mount;
4324 	int			whichfork = xfs_bmapi_whichfork(flags);
4325 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
4326 	xfs_fileoff_t		end;		/* end of mapped file region */
4327 	bool			eof = false;	/* after the end of extents */
4328 	int			error;		/* error return */
4329 	int			n;		/* current extent index */
4330 	xfs_fileoff_t		obno;		/* old block number (offset) */
4331 
4332 #ifdef DEBUG
4333 	xfs_fileoff_t		orig_bno;	/* original block number value */
4334 	int			orig_flags;	/* original flags arg value */
4335 	xfs_filblks_t		orig_len;	/* original value of len arg */
4336 	struct xfs_bmbt_irec	*orig_mval;	/* original value of mval */
4337 	int			orig_nmap;	/* original value of *nmap */
4338 
4339 	orig_bno = bno;
4340 	orig_len = len;
4341 	orig_flags = flags;
4342 	orig_mval = mval;
4343 	orig_nmap = *nmap;
4344 #endif
4345 
4346 	ASSERT(*nmap >= 1);
4347 	ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4348 	ASSERT(tp != NULL);
4349 	ASSERT(len > 0);
4350 	ASSERT(ifp->if_format != XFS_DINODE_FMT_LOCAL);
4351 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4352 	ASSERT(!(flags & XFS_BMAPI_REMAP));
4353 
4354 	/* zeroing is for currently only for data extents, not metadata */
4355 	ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4356 			(XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4357 	/*
4358 	 * we can allocate unwritten extents or pre-zero allocated blocks,
4359 	 * but it makes no sense to do both at once. This would result in
4360 	 * zeroing the unwritten extent twice, but it still being an
4361 	 * unwritten extent....
4362 	 */
4363 	ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4364 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4365 
4366 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
4367 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
4368 		return -EFSCORRUPTED;
4369 	}
4370 
4371 	if (xfs_is_shutdown(mp))
4372 		return -EIO;
4373 
4374 	XFS_STATS_INC(mp, xs_blk_mapw);
4375 
4376 	error = xfs_iread_extents(tp, ip, whichfork);
4377 	if (error)
4378 		goto error0;
4379 
4380 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.icur, &bma.got))
4381 		eof = true;
4382 	if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4383 		bma.prev.br_startoff = NULLFILEOFF;
4384 	bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4385 
4386 	n = 0;
4387 	end = bno + len;
4388 	obno = bno;
4389 	while (bno < end && n < *nmap) {
4390 		bool			need_alloc = false, wasdelay = false;
4391 
4392 		/* in hole or beyond EOF? */
4393 		if (eof || bma.got.br_startoff > bno) {
4394 			/*
4395 			 * CoW fork conversions should /never/ hit EOF or
4396 			 * holes.  There should always be something for us
4397 			 * to work on.
4398 			 */
4399 			ASSERT(!((flags & XFS_BMAPI_CONVERT) &&
4400 			         (flags & XFS_BMAPI_COWFORK)));
4401 
4402 			need_alloc = true;
4403 		} else if (isnullstartblock(bma.got.br_startblock)) {
4404 			wasdelay = true;
4405 		}
4406 
4407 		/*
4408 		 * First, deal with the hole before the allocated space
4409 		 * that we found, if any.
4410 		 */
4411 		if (need_alloc || wasdelay) {
4412 			bma.eof = eof;
4413 			bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4414 			bma.wasdel = wasdelay;
4415 			bma.offset = bno;
4416 			bma.flags = flags;
4417 
4418 			/*
4419 			 * There's a 32/64 bit type mismatch between the
4420 			 * allocation length request (which can be 64 bits in
4421 			 * length) and the bma length request, which is
4422 			 * xfs_extlen_t and therefore 32 bits. Hence we have to
4423 			 * check for 32-bit overflows and handle them here.
4424 			 */
4425 			if (len > (xfs_filblks_t)XFS_MAX_BMBT_EXTLEN)
4426 				bma.length = XFS_MAX_BMBT_EXTLEN;
4427 			else
4428 				bma.length = len;
4429 
4430 			ASSERT(len > 0);
4431 			ASSERT(bma.length > 0);
4432 			error = xfs_bmapi_allocate(&bma);
4433 			if (error)
4434 				goto error0;
4435 			if (bma.blkno == NULLFSBLOCK)
4436 				break;
4437 
4438 			/*
4439 			 * If this is a CoW allocation, record the data in
4440 			 * the refcount btree for orphan recovery.
4441 			 */
4442 			if (whichfork == XFS_COW_FORK)
4443 				xfs_refcount_alloc_cow_extent(tp, bma.blkno,
4444 						bma.length);
4445 		}
4446 
4447 		/* Deal with the allocated space we found.  */
4448 		xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4449 							end, n, flags);
4450 
4451 		/* Execute unwritten extent conversion if necessary */
4452 		error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4453 		if (error == -EAGAIN)
4454 			continue;
4455 		if (error)
4456 			goto error0;
4457 
4458 		/* update the extent map to return */
4459 		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4460 
4461 		/*
4462 		 * If we're done, stop now.  Stop when we've allocated
4463 		 * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4464 		 * the transaction may get too big.
4465 		 */
4466 		if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4467 			break;
4468 
4469 		/* Else go on to the next record. */
4470 		bma.prev = bma.got;
4471 		if (!xfs_iext_next_extent(ifp, &bma.icur, &bma.got))
4472 			eof = true;
4473 	}
4474 	*nmap = n;
4475 
4476 	error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4477 			whichfork);
4478 	if (error)
4479 		goto error0;
4480 
4481 	ASSERT(ifp->if_format != XFS_DINODE_FMT_BTREE ||
4482 	       ifp->if_nextents > XFS_IFORK_MAXEXT(ip, whichfork));
4483 	xfs_bmapi_finish(&bma, whichfork, 0);
4484 	xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4485 		orig_nmap, *nmap);
4486 	return 0;
4487 error0:
4488 	xfs_bmapi_finish(&bma, whichfork, error);
4489 	return error;
4490 }
4491 
4492 /*
4493  * Convert an existing delalloc extent to real blocks based on file offset. This
4494  * attempts to allocate the entire delalloc extent and may require multiple
4495  * invocations to allocate the target offset if a large enough physical extent
4496  * is not available.
4497  */
4498 int
4499 xfs_bmapi_convert_delalloc(
4500 	struct xfs_inode	*ip,
4501 	int			whichfork,
4502 	xfs_off_t		offset,
4503 	struct iomap		*iomap,
4504 	unsigned int		*seq)
4505 {
4506 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
4507 	struct xfs_mount	*mp = ip->i_mount;
4508 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
4509 	struct xfs_bmalloca	bma = { NULL };
4510 	uint16_t		flags = 0;
4511 	struct xfs_trans	*tp;
4512 	int			error;
4513 
4514 	if (whichfork == XFS_COW_FORK)
4515 		flags |= IOMAP_F_SHARED;
4516 
4517 	/*
4518 	 * Space for the extent and indirect blocks was reserved when the
4519 	 * delalloc extent was created so there's no need to do so here.
4520 	 */
4521 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0,
4522 				XFS_TRANS_RESERVE, &tp);
4523 	if (error)
4524 		return error;
4525 
4526 	xfs_ilock(ip, XFS_ILOCK_EXCL);
4527 	xfs_trans_ijoin(tp, ip, 0);
4528 
4529 	error = xfs_iext_count_may_overflow(ip, whichfork,
4530 			XFS_IEXT_ADD_NOSPLIT_CNT);
4531 	if (error == -EFBIG)
4532 		error = xfs_iext_count_upgrade(tp, ip,
4533 				XFS_IEXT_ADD_NOSPLIT_CNT);
4534 	if (error)
4535 		goto out_trans_cancel;
4536 
4537 	if (!xfs_iext_lookup_extent(ip, ifp, offset_fsb, &bma.icur, &bma.got) ||
4538 	    bma.got.br_startoff > offset_fsb) {
4539 		/*
4540 		 * No extent found in the range we are trying to convert.  This
4541 		 * should only happen for the COW fork, where another thread
4542 		 * might have moved the extent to the data fork in the meantime.
4543 		 */
4544 		WARN_ON_ONCE(whichfork != XFS_COW_FORK);
4545 		error = -EAGAIN;
4546 		goto out_trans_cancel;
4547 	}
4548 
4549 	/*
4550 	 * If we find a real extent here we raced with another thread converting
4551 	 * the extent.  Just return the real extent at this offset.
4552 	 */
4553 	if (!isnullstartblock(bma.got.br_startblock)) {
4554 		xfs_bmbt_to_iomap(ip, iomap, &bma.got, 0, flags);
4555 		*seq = READ_ONCE(ifp->if_seq);
4556 		goto out_trans_cancel;
4557 	}
4558 
4559 	bma.tp = tp;
4560 	bma.ip = ip;
4561 	bma.wasdel = true;
4562 	bma.offset = bma.got.br_startoff;
4563 	bma.length = max_t(xfs_filblks_t, bma.got.br_blockcount,
4564 			XFS_MAX_BMBT_EXTLEN);
4565 	bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4566 
4567 	/*
4568 	 * When we're converting the delalloc reservations backing dirty pages
4569 	 * in the page cache, we must be careful about how we create the new
4570 	 * extents:
4571 	 *
4572 	 * New CoW fork extents are created unwritten, turned into real extents
4573 	 * when we're about to write the data to disk, and mapped into the data
4574 	 * fork after the write finishes.  End of story.
4575 	 *
4576 	 * New data fork extents must be mapped in as unwritten and converted
4577 	 * to real extents after the write succeeds to avoid exposing stale
4578 	 * disk contents if we crash.
4579 	 */
4580 	bma.flags = XFS_BMAPI_PREALLOC;
4581 	if (whichfork == XFS_COW_FORK)
4582 		bma.flags |= XFS_BMAPI_COWFORK;
4583 
4584 	if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4585 		bma.prev.br_startoff = NULLFILEOFF;
4586 
4587 	error = xfs_bmapi_allocate(&bma);
4588 	if (error)
4589 		goto out_finish;
4590 
4591 	error = -ENOSPC;
4592 	if (WARN_ON_ONCE(bma.blkno == NULLFSBLOCK))
4593 		goto out_finish;
4594 	error = -EFSCORRUPTED;
4595 	if (WARN_ON_ONCE(!xfs_valid_startblock(ip, bma.got.br_startblock)))
4596 		goto out_finish;
4597 
4598 	XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, bma.length));
4599 	XFS_STATS_INC(mp, xs_xstrat_quick);
4600 
4601 	ASSERT(!isnullstartblock(bma.got.br_startblock));
4602 	xfs_bmbt_to_iomap(ip, iomap, &bma.got, 0, flags);
4603 	*seq = READ_ONCE(ifp->if_seq);
4604 
4605 	if (whichfork == XFS_COW_FORK)
4606 		xfs_refcount_alloc_cow_extent(tp, bma.blkno, bma.length);
4607 
4608 	error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4609 			whichfork);
4610 	if (error)
4611 		goto out_finish;
4612 
4613 	xfs_bmapi_finish(&bma, whichfork, 0);
4614 	error = xfs_trans_commit(tp);
4615 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
4616 	return error;
4617 
4618 out_finish:
4619 	xfs_bmapi_finish(&bma, whichfork, error);
4620 out_trans_cancel:
4621 	xfs_trans_cancel(tp);
4622 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
4623 	return error;
4624 }
4625 
4626 int
4627 xfs_bmapi_remap(
4628 	struct xfs_trans	*tp,
4629 	struct xfs_inode	*ip,
4630 	xfs_fileoff_t		bno,
4631 	xfs_filblks_t		len,
4632 	xfs_fsblock_t		startblock,
4633 	uint32_t		flags)
4634 {
4635 	struct xfs_mount	*mp = ip->i_mount;
4636 	struct xfs_ifork	*ifp;
4637 	struct xfs_btree_cur	*cur = NULL;
4638 	struct xfs_bmbt_irec	got;
4639 	struct xfs_iext_cursor	icur;
4640 	int			whichfork = xfs_bmapi_whichfork(flags);
4641 	int			logflags = 0, error;
4642 
4643 	ifp = xfs_ifork_ptr(ip, whichfork);
4644 	ASSERT(len > 0);
4645 	ASSERT(len <= (xfs_filblks_t)XFS_MAX_BMBT_EXTLEN);
4646 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4647 	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC |
4648 			   XFS_BMAPI_NORMAP)));
4649 	ASSERT((flags & (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC)) !=
4650 			(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC));
4651 
4652 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
4653 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
4654 		return -EFSCORRUPTED;
4655 	}
4656 
4657 	if (xfs_is_shutdown(mp))
4658 		return -EIO;
4659 
4660 	error = xfs_iread_extents(tp, ip, whichfork);
4661 	if (error)
4662 		return error;
4663 
4664 	if (xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
4665 		/* make sure we only reflink into a hole. */
4666 		ASSERT(got.br_startoff > bno);
4667 		ASSERT(got.br_startoff - bno >= len);
4668 	}
4669 
4670 	ip->i_nblocks += len;
4671 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4672 
4673 	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
4674 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
4675 		cur->bc_ino.flags = 0;
4676 	}
4677 
4678 	got.br_startoff = bno;
4679 	got.br_startblock = startblock;
4680 	got.br_blockcount = len;
4681 	if (flags & XFS_BMAPI_PREALLOC)
4682 		got.br_state = XFS_EXT_UNWRITTEN;
4683 	else
4684 		got.br_state = XFS_EXT_NORM;
4685 
4686 	error = xfs_bmap_add_extent_hole_real(tp, ip, whichfork, &icur,
4687 			&cur, &got, &logflags, flags);
4688 	if (error)
4689 		goto error0;
4690 
4691 	error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags, whichfork);
4692 
4693 error0:
4694 	if (ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS)
4695 		logflags &= ~XFS_ILOG_DEXT;
4696 	else if (ip->i_df.if_format != XFS_DINODE_FMT_BTREE)
4697 		logflags &= ~XFS_ILOG_DBROOT;
4698 
4699 	if (logflags)
4700 		xfs_trans_log_inode(tp, ip, logflags);
4701 	if (cur)
4702 		xfs_btree_del_cursor(cur, error);
4703 	return error;
4704 }
4705 
4706 /*
4707  * When a delalloc extent is split (e.g., due to a hole punch), the original
4708  * indlen reservation must be shared across the two new extents that are left
4709  * behind.
4710  *
4711  * Given the original reservation and the worst case indlen for the two new
4712  * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4713  * reservation fairly across the two new extents. If necessary, steal available
4714  * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4715  * ores == 1). The number of stolen blocks is returned. The availability and
4716  * subsequent accounting of stolen blocks is the responsibility of the caller.
4717  */
4718 static xfs_filblks_t
4719 xfs_bmap_split_indlen(
4720 	xfs_filblks_t			ores,		/* original res. */
4721 	xfs_filblks_t			*indlen1,	/* ext1 worst indlen */
4722 	xfs_filblks_t			*indlen2,	/* ext2 worst indlen */
4723 	xfs_filblks_t			avail)		/* stealable blocks */
4724 {
4725 	xfs_filblks_t			len1 = *indlen1;
4726 	xfs_filblks_t			len2 = *indlen2;
4727 	xfs_filblks_t			nres = len1 + len2; /* new total res. */
4728 	xfs_filblks_t			stolen = 0;
4729 	xfs_filblks_t			resfactor;
4730 
4731 	/*
4732 	 * Steal as many blocks as we can to try and satisfy the worst case
4733 	 * indlen for both new extents.
4734 	 */
4735 	if (ores < nres && avail)
4736 		stolen = XFS_FILBLKS_MIN(nres - ores, avail);
4737 	ores += stolen;
4738 
4739 	 /* nothing else to do if we've satisfied the new reservation */
4740 	if (ores >= nres)
4741 		return stolen;
4742 
4743 	/*
4744 	 * We can't meet the total required reservation for the two extents.
4745 	 * Calculate the percent of the overall shortage between both extents
4746 	 * and apply this percentage to each of the requested indlen values.
4747 	 * This distributes the shortage fairly and reduces the chances that one
4748 	 * of the two extents is left with nothing when extents are repeatedly
4749 	 * split.
4750 	 */
4751 	resfactor = (ores * 100);
4752 	do_div(resfactor, nres);
4753 	len1 *= resfactor;
4754 	do_div(len1, 100);
4755 	len2 *= resfactor;
4756 	do_div(len2, 100);
4757 	ASSERT(len1 + len2 <= ores);
4758 	ASSERT(len1 < *indlen1 && len2 < *indlen2);
4759 
4760 	/*
4761 	 * Hand out the remainder to each extent. If one of the two reservations
4762 	 * is zero, we want to make sure that one gets a block first. The loop
4763 	 * below starts with len1, so hand len2 a block right off the bat if it
4764 	 * is zero.
4765 	 */
4766 	ores -= (len1 + len2);
4767 	ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores);
4768 	if (ores && !len2 && *indlen2) {
4769 		len2++;
4770 		ores--;
4771 	}
4772 	while (ores) {
4773 		if (len1 < *indlen1) {
4774 			len1++;
4775 			ores--;
4776 		}
4777 		if (!ores)
4778 			break;
4779 		if (len2 < *indlen2) {
4780 			len2++;
4781 			ores--;
4782 		}
4783 	}
4784 
4785 	*indlen1 = len1;
4786 	*indlen2 = len2;
4787 
4788 	return stolen;
4789 }
4790 
4791 int
4792 xfs_bmap_del_extent_delay(
4793 	struct xfs_inode	*ip,
4794 	int			whichfork,
4795 	struct xfs_iext_cursor	*icur,
4796 	struct xfs_bmbt_irec	*got,
4797 	struct xfs_bmbt_irec	*del)
4798 {
4799 	struct xfs_mount	*mp = ip->i_mount;
4800 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
4801 	struct xfs_bmbt_irec	new;
4802 	int64_t			da_old, da_new, da_diff = 0;
4803 	xfs_fileoff_t		del_endoff, got_endoff;
4804 	xfs_filblks_t		got_indlen, new_indlen, stolen;
4805 	uint32_t		state = xfs_bmap_fork_to_state(whichfork);
4806 	int			error = 0;
4807 	bool			isrt;
4808 
4809 	XFS_STATS_INC(mp, xs_del_exlist);
4810 
4811 	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4812 	del_endoff = del->br_startoff + del->br_blockcount;
4813 	got_endoff = got->br_startoff + got->br_blockcount;
4814 	da_old = startblockval(got->br_startblock);
4815 	da_new = 0;
4816 
4817 	ASSERT(del->br_blockcount > 0);
4818 	ASSERT(got->br_startoff <= del->br_startoff);
4819 	ASSERT(got_endoff >= del_endoff);
4820 
4821 	if (isrt) {
4822 		uint64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount);
4823 
4824 		do_div(rtexts, mp->m_sb.sb_rextsize);
4825 		xfs_mod_frextents(mp, rtexts);
4826 	}
4827 
4828 	/*
4829 	 * Update the inode delalloc counter now and wait to update the
4830 	 * sb counters as we might have to borrow some blocks for the
4831 	 * indirect block accounting.
4832 	 */
4833 	ASSERT(!isrt);
4834 	error = xfs_quota_unreserve_blkres(ip, del->br_blockcount);
4835 	if (error)
4836 		return error;
4837 	ip->i_delayed_blks -= del->br_blockcount;
4838 
4839 	if (got->br_startoff == del->br_startoff)
4840 		state |= BMAP_LEFT_FILLING;
4841 	if (got_endoff == del_endoff)
4842 		state |= BMAP_RIGHT_FILLING;
4843 
4844 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4845 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4846 		/*
4847 		 * Matches the whole extent.  Delete the entry.
4848 		 */
4849 		xfs_iext_remove(ip, icur, state);
4850 		xfs_iext_prev(ifp, icur);
4851 		break;
4852 	case BMAP_LEFT_FILLING:
4853 		/*
4854 		 * Deleting the first part of the extent.
4855 		 */
4856 		got->br_startoff = del_endoff;
4857 		got->br_blockcount -= del->br_blockcount;
4858 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4859 				got->br_blockcount), da_old);
4860 		got->br_startblock = nullstartblock((int)da_new);
4861 		xfs_iext_update_extent(ip, state, icur, got);
4862 		break;
4863 	case BMAP_RIGHT_FILLING:
4864 		/*
4865 		 * Deleting the last part of the extent.
4866 		 */
4867 		got->br_blockcount = got->br_blockcount - del->br_blockcount;
4868 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4869 				got->br_blockcount), da_old);
4870 		got->br_startblock = nullstartblock((int)da_new);
4871 		xfs_iext_update_extent(ip, state, icur, got);
4872 		break;
4873 	case 0:
4874 		/*
4875 		 * Deleting the middle of the extent.
4876 		 *
4877 		 * Distribute the original indlen reservation across the two new
4878 		 * extents.  Steal blocks from the deleted extent if necessary.
4879 		 * Stealing blocks simply fudges the fdblocks accounting below.
4880 		 * Warn if either of the new indlen reservations is zero as this
4881 		 * can lead to delalloc problems.
4882 		 */
4883 		got->br_blockcount = del->br_startoff - got->br_startoff;
4884 		got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount);
4885 
4886 		new.br_blockcount = got_endoff - del_endoff;
4887 		new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount);
4888 
4889 		WARN_ON_ONCE(!got_indlen || !new_indlen);
4890 		stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen,
4891 						       del->br_blockcount);
4892 
4893 		got->br_startblock = nullstartblock((int)got_indlen);
4894 
4895 		new.br_startoff = del_endoff;
4896 		new.br_state = got->br_state;
4897 		new.br_startblock = nullstartblock((int)new_indlen);
4898 
4899 		xfs_iext_update_extent(ip, state, icur, got);
4900 		xfs_iext_next(ifp, icur);
4901 		xfs_iext_insert(ip, icur, &new, state);
4902 
4903 		da_new = got_indlen + new_indlen - stolen;
4904 		del->br_blockcount -= stolen;
4905 		break;
4906 	}
4907 
4908 	ASSERT(da_old >= da_new);
4909 	da_diff = da_old - da_new;
4910 	if (!isrt)
4911 		da_diff += del->br_blockcount;
4912 	if (da_diff) {
4913 		xfs_mod_fdblocks(mp, da_diff, false);
4914 		xfs_mod_delalloc(mp, -da_diff);
4915 	}
4916 	return error;
4917 }
4918 
4919 void
4920 xfs_bmap_del_extent_cow(
4921 	struct xfs_inode	*ip,
4922 	struct xfs_iext_cursor	*icur,
4923 	struct xfs_bmbt_irec	*got,
4924 	struct xfs_bmbt_irec	*del)
4925 {
4926 	struct xfs_mount	*mp = ip->i_mount;
4927 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, XFS_COW_FORK);
4928 	struct xfs_bmbt_irec	new;
4929 	xfs_fileoff_t		del_endoff, got_endoff;
4930 	uint32_t		state = BMAP_COWFORK;
4931 
4932 	XFS_STATS_INC(mp, xs_del_exlist);
4933 
4934 	del_endoff = del->br_startoff + del->br_blockcount;
4935 	got_endoff = got->br_startoff + got->br_blockcount;
4936 
4937 	ASSERT(del->br_blockcount > 0);
4938 	ASSERT(got->br_startoff <= del->br_startoff);
4939 	ASSERT(got_endoff >= del_endoff);
4940 	ASSERT(!isnullstartblock(got->br_startblock));
4941 
4942 	if (got->br_startoff == del->br_startoff)
4943 		state |= BMAP_LEFT_FILLING;
4944 	if (got_endoff == del_endoff)
4945 		state |= BMAP_RIGHT_FILLING;
4946 
4947 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4948 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4949 		/*
4950 		 * Matches the whole extent.  Delete the entry.
4951 		 */
4952 		xfs_iext_remove(ip, icur, state);
4953 		xfs_iext_prev(ifp, icur);
4954 		break;
4955 	case BMAP_LEFT_FILLING:
4956 		/*
4957 		 * Deleting the first part of the extent.
4958 		 */
4959 		got->br_startoff = del_endoff;
4960 		got->br_blockcount -= del->br_blockcount;
4961 		got->br_startblock = del->br_startblock + del->br_blockcount;
4962 		xfs_iext_update_extent(ip, state, icur, got);
4963 		break;
4964 	case BMAP_RIGHT_FILLING:
4965 		/*
4966 		 * Deleting the last part of the extent.
4967 		 */
4968 		got->br_blockcount -= del->br_blockcount;
4969 		xfs_iext_update_extent(ip, state, icur, got);
4970 		break;
4971 	case 0:
4972 		/*
4973 		 * Deleting the middle of the extent.
4974 		 */
4975 		got->br_blockcount = del->br_startoff - got->br_startoff;
4976 
4977 		new.br_startoff = del_endoff;
4978 		new.br_blockcount = got_endoff - del_endoff;
4979 		new.br_state = got->br_state;
4980 		new.br_startblock = del->br_startblock + del->br_blockcount;
4981 
4982 		xfs_iext_update_extent(ip, state, icur, got);
4983 		xfs_iext_next(ifp, icur);
4984 		xfs_iext_insert(ip, icur, &new, state);
4985 		break;
4986 	}
4987 	ip->i_delayed_blks -= del->br_blockcount;
4988 }
4989 
4990 /*
4991  * Called by xfs_bmapi to update file extent records and the btree
4992  * after removing space.
4993  */
4994 STATIC int				/* error */
4995 xfs_bmap_del_extent_real(
4996 	xfs_inode_t		*ip,	/* incore inode pointer */
4997 	xfs_trans_t		*tp,	/* current transaction pointer */
4998 	struct xfs_iext_cursor	*icur,
4999 	struct xfs_btree_cur	*cur,	/* if null, not a btree */
5000 	xfs_bmbt_irec_t		*del,	/* data to remove from extents */
5001 	int			*logflagsp, /* inode logging flags */
5002 	int			whichfork, /* data or attr fork */
5003 	uint32_t		bflags)	/* bmapi flags */
5004 {
5005 	xfs_fsblock_t		del_endblock=0;	/* first block past del */
5006 	xfs_fileoff_t		del_endoff;	/* first offset past del */
5007 	int			do_fx;	/* free extent at end of routine */
5008 	int			error;	/* error return value */
5009 	int			flags = 0;/* inode logging flags */
5010 	struct xfs_bmbt_irec	got;	/* current extent entry */
5011 	xfs_fileoff_t		got_endoff;	/* first offset past got */
5012 	int			i;	/* temp state */
5013 	struct xfs_ifork	*ifp;	/* inode fork pointer */
5014 	xfs_mount_t		*mp;	/* mount structure */
5015 	xfs_filblks_t		nblks;	/* quota/sb block count */
5016 	xfs_bmbt_irec_t		new;	/* new record to be inserted */
5017 	/* REFERENCED */
5018 	uint			qfield;	/* quota field to update */
5019 	uint32_t		state = xfs_bmap_fork_to_state(whichfork);
5020 	struct xfs_bmbt_irec	old;
5021 
5022 	mp = ip->i_mount;
5023 	XFS_STATS_INC(mp, xs_del_exlist);
5024 
5025 	ifp = xfs_ifork_ptr(ip, whichfork);
5026 	ASSERT(del->br_blockcount > 0);
5027 	xfs_iext_get_extent(ifp, icur, &got);
5028 	ASSERT(got.br_startoff <= del->br_startoff);
5029 	del_endoff = del->br_startoff + del->br_blockcount;
5030 	got_endoff = got.br_startoff + got.br_blockcount;
5031 	ASSERT(got_endoff >= del_endoff);
5032 	ASSERT(!isnullstartblock(got.br_startblock));
5033 	qfield = 0;
5034 	error = 0;
5035 
5036 	/*
5037 	 * If it's the case where the directory code is running with no block
5038 	 * reservation, and the deleted block is in the middle of its extent,
5039 	 * and the resulting insert of an extent would cause transformation to
5040 	 * btree format, then reject it.  The calling code will then swap blocks
5041 	 * around instead.  We have to do this now, rather than waiting for the
5042 	 * conversion to btree format, since the transaction will be dirty then.
5043 	 */
5044 	if (tp->t_blk_res == 0 &&
5045 	    ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
5046 	    ifp->if_nextents >= XFS_IFORK_MAXEXT(ip, whichfork) &&
5047 	    del->br_startoff > got.br_startoff && del_endoff < got_endoff)
5048 		return -ENOSPC;
5049 
5050 	flags = XFS_ILOG_CORE;
5051 	if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
5052 		xfs_filblks_t	len;
5053 		xfs_extlen_t	mod;
5054 
5055 		len = div_u64_rem(del->br_blockcount, mp->m_sb.sb_rextsize,
5056 				  &mod);
5057 		ASSERT(mod == 0);
5058 
5059 		if (!(bflags & XFS_BMAPI_REMAP)) {
5060 			xfs_fsblock_t	bno;
5061 
5062 			bno = div_u64_rem(del->br_startblock,
5063 					mp->m_sb.sb_rextsize, &mod);
5064 			ASSERT(mod == 0);
5065 
5066 			error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
5067 			if (error)
5068 				goto done;
5069 		}
5070 
5071 		do_fx = 0;
5072 		nblks = len * mp->m_sb.sb_rextsize;
5073 		qfield = XFS_TRANS_DQ_RTBCOUNT;
5074 	} else {
5075 		do_fx = 1;
5076 		nblks = del->br_blockcount;
5077 		qfield = XFS_TRANS_DQ_BCOUNT;
5078 	}
5079 
5080 	del_endblock = del->br_startblock + del->br_blockcount;
5081 	if (cur) {
5082 		error = xfs_bmbt_lookup_eq(cur, &got, &i);
5083 		if (error)
5084 			goto done;
5085 		if (XFS_IS_CORRUPT(mp, i != 1)) {
5086 			error = -EFSCORRUPTED;
5087 			goto done;
5088 		}
5089 	}
5090 
5091 	if (got.br_startoff == del->br_startoff)
5092 		state |= BMAP_LEFT_FILLING;
5093 	if (got_endoff == del_endoff)
5094 		state |= BMAP_RIGHT_FILLING;
5095 
5096 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
5097 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
5098 		/*
5099 		 * Matches the whole extent.  Delete the entry.
5100 		 */
5101 		xfs_iext_remove(ip, icur, state);
5102 		xfs_iext_prev(ifp, icur);
5103 		ifp->if_nextents--;
5104 
5105 		flags |= XFS_ILOG_CORE;
5106 		if (!cur) {
5107 			flags |= xfs_ilog_fext(whichfork);
5108 			break;
5109 		}
5110 		if ((error = xfs_btree_delete(cur, &i)))
5111 			goto done;
5112 		if (XFS_IS_CORRUPT(mp, i != 1)) {
5113 			error = -EFSCORRUPTED;
5114 			goto done;
5115 		}
5116 		break;
5117 	case BMAP_LEFT_FILLING:
5118 		/*
5119 		 * Deleting the first part of the extent.
5120 		 */
5121 		got.br_startoff = del_endoff;
5122 		got.br_startblock = del_endblock;
5123 		got.br_blockcount -= del->br_blockcount;
5124 		xfs_iext_update_extent(ip, state, icur, &got);
5125 		if (!cur) {
5126 			flags |= xfs_ilog_fext(whichfork);
5127 			break;
5128 		}
5129 		error = xfs_bmbt_update(cur, &got);
5130 		if (error)
5131 			goto done;
5132 		break;
5133 	case BMAP_RIGHT_FILLING:
5134 		/*
5135 		 * Deleting the last part of the extent.
5136 		 */
5137 		got.br_blockcount -= del->br_blockcount;
5138 		xfs_iext_update_extent(ip, state, icur, &got);
5139 		if (!cur) {
5140 			flags |= xfs_ilog_fext(whichfork);
5141 			break;
5142 		}
5143 		error = xfs_bmbt_update(cur, &got);
5144 		if (error)
5145 			goto done;
5146 		break;
5147 	case 0:
5148 		/*
5149 		 * Deleting the middle of the extent.
5150 		 */
5151 
5152 		old = got;
5153 
5154 		got.br_blockcount = del->br_startoff - got.br_startoff;
5155 		xfs_iext_update_extent(ip, state, icur, &got);
5156 
5157 		new.br_startoff = del_endoff;
5158 		new.br_blockcount = got_endoff - del_endoff;
5159 		new.br_state = got.br_state;
5160 		new.br_startblock = del_endblock;
5161 
5162 		flags |= XFS_ILOG_CORE;
5163 		if (cur) {
5164 			error = xfs_bmbt_update(cur, &got);
5165 			if (error)
5166 				goto done;
5167 			error = xfs_btree_increment(cur, 0, &i);
5168 			if (error)
5169 				goto done;
5170 			cur->bc_rec.b = new;
5171 			error = xfs_btree_insert(cur, &i);
5172 			if (error && error != -ENOSPC)
5173 				goto done;
5174 			/*
5175 			 * If get no-space back from btree insert, it tried a
5176 			 * split, and we have a zero block reservation.  Fix up
5177 			 * our state and return the error.
5178 			 */
5179 			if (error == -ENOSPC) {
5180 				/*
5181 				 * Reset the cursor, don't trust it after any
5182 				 * insert operation.
5183 				 */
5184 				error = xfs_bmbt_lookup_eq(cur, &got, &i);
5185 				if (error)
5186 					goto done;
5187 				if (XFS_IS_CORRUPT(mp, i != 1)) {
5188 					error = -EFSCORRUPTED;
5189 					goto done;
5190 				}
5191 				/*
5192 				 * Update the btree record back
5193 				 * to the original value.
5194 				 */
5195 				error = xfs_bmbt_update(cur, &old);
5196 				if (error)
5197 					goto done;
5198 				/*
5199 				 * Reset the extent record back
5200 				 * to the original value.
5201 				 */
5202 				xfs_iext_update_extent(ip, state, icur, &old);
5203 				flags = 0;
5204 				error = -ENOSPC;
5205 				goto done;
5206 			}
5207 			if (XFS_IS_CORRUPT(mp, i != 1)) {
5208 				error = -EFSCORRUPTED;
5209 				goto done;
5210 			}
5211 		} else
5212 			flags |= xfs_ilog_fext(whichfork);
5213 
5214 		ifp->if_nextents++;
5215 		xfs_iext_next(ifp, icur);
5216 		xfs_iext_insert(ip, icur, &new, state);
5217 		break;
5218 	}
5219 
5220 	/* remove reverse mapping */
5221 	xfs_rmap_unmap_extent(tp, ip, whichfork, del);
5222 
5223 	/*
5224 	 * If we need to, add to list of extents to delete.
5225 	 */
5226 	if (do_fx && !(bflags & XFS_BMAPI_REMAP)) {
5227 		if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
5228 			xfs_refcount_decrease_extent(tp, del);
5229 		} else {
5230 			__xfs_free_extent_later(tp, del->br_startblock,
5231 					del->br_blockcount, NULL,
5232 					(bflags & XFS_BMAPI_NODISCARD) ||
5233 					del->br_state == XFS_EXT_UNWRITTEN);
5234 		}
5235 	}
5236 
5237 	/*
5238 	 * Adjust inode # blocks in the file.
5239 	 */
5240 	if (nblks)
5241 		ip->i_nblocks -= nblks;
5242 	/*
5243 	 * Adjust quota data.
5244 	 */
5245 	if (qfield && !(bflags & XFS_BMAPI_REMAP))
5246 		xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5247 
5248 done:
5249 	*logflagsp = flags;
5250 	return error;
5251 }
5252 
5253 /*
5254  * Unmap (remove) blocks from a file.
5255  * If nexts is nonzero then the number of extents to remove is limited to
5256  * that value.  If not all extents in the block range can be removed then
5257  * *done is set.
5258  */
5259 int						/* error */
5260 __xfs_bunmapi(
5261 	struct xfs_trans	*tp,		/* transaction pointer */
5262 	struct xfs_inode	*ip,		/* incore inode */
5263 	xfs_fileoff_t		start,		/* first file offset deleted */
5264 	xfs_filblks_t		*rlen,		/* i/o: amount remaining */
5265 	uint32_t		flags,		/* misc flags */
5266 	xfs_extnum_t		nexts)		/* number of extents max */
5267 {
5268 	struct xfs_btree_cur	*cur;		/* bmap btree cursor */
5269 	struct xfs_bmbt_irec	del;		/* extent being deleted */
5270 	int			error;		/* error return value */
5271 	xfs_extnum_t		extno;		/* extent number in list */
5272 	struct xfs_bmbt_irec	got;		/* current extent record */
5273 	struct xfs_ifork	*ifp;		/* inode fork pointer */
5274 	int			isrt;		/* freeing in rt area */
5275 	int			logflags;	/* transaction logging flags */
5276 	xfs_extlen_t		mod;		/* rt extent offset */
5277 	struct xfs_mount	*mp = ip->i_mount;
5278 	int			tmp_logflags;	/* partial logging flags */
5279 	int			wasdel;		/* was a delayed alloc extent */
5280 	int			whichfork;	/* data or attribute fork */
5281 	xfs_fsblock_t		sum;
5282 	xfs_filblks_t		len = *rlen;	/* length to unmap in file */
5283 	xfs_fileoff_t		end;
5284 	struct xfs_iext_cursor	icur;
5285 	bool			done = false;
5286 
5287 	trace_xfs_bunmap(ip, start, len, flags, _RET_IP_);
5288 
5289 	whichfork = xfs_bmapi_whichfork(flags);
5290 	ASSERT(whichfork != XFS_COW_FORK);
5291 	ifp = xfs_ifork_ptr(ip, whichfork);
5292 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)))
5293 		return -EFSCORRUPTED;
5294 	if (xfs_is_shutdown(mp))
5295 		return -EIO;
5296 
5297 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5298 	ASSERT(len > 0);
5299 	ASSERT(nexts >= 0);
5300 
5301 	error = xfs_iread_extents(tp, ip, whichfork);
5302 	if (error)
5303 		return error;
5304 
5305 	if (xfs_iext_count(ifp) == 0) {
5306 		*rlen = 0;
5307 		return 0;
5308 	}
5309 	XFS_STATS_INC(mp, xs_blk_unmap);
5310 	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5311 	end = start + len;
5312 
5313 	if (!xfs_iext_lookup_extent_before(ip, ifp, &end, &icur, &got)) {
5314 		*rlen = 0;
5315 		return 0;
5316 	}
5317 	end--;
5318 
5319 	logflags = 0;
5320 	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
5321 		ASSERT(ifp->if_format == XFS_DINODE_FMT_BTREE);
5322 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5323 		cur->bc_ino.flags = 0;
5324 	} else
5325 		cur = NULL;
5326 
5327 	if (isrt) {
5328 		/*
5329 		 * Synchronize by locking the bitmap inode.
5330 		 */
5331 		xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
5332 		xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5333 		xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
5334 		xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
5335 	}
5336 
5337 	extno = 0;
5338 	while (end != (xfs_fileoff_t)-1 && end >= start &&
5339 	       (nexts == 0 || extno < nexts)) {
5340 		/*
5341 		 * Is the found extent after a hole in which end lives?
5342 		 * Just back up to the previous extent, if so.
5343 		 */
5344 		if (got.br_startoff > end &&
5345 		    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5346 			done = true;
5347 			break;
5348 		}
5349 		/*
5350 		 * Is the last block of this extent before the range
5351 		 * we're supposed to delete?  If so, we're done.
5352 		 */
5353 		end = XFS_FILEOFF_MIN(end,
5354 			got.br_startoff + got.br_blockcount - 1);
5355 		if (end < start)
5356 			break;
5357 		/*
5358 		 * Then deal with the (possibly delayed) allocated space
5359 		 * we found.
5360 		 */
5361 		del = got;
5362 		wasdel = isnullstartblock(del.br_startblock);
5363 
5364 		if (got.br_startoff < start) {
5365 			del.br_startoff = start;
5366 			del.br_blockcount -= start - got.br_startoff;
5367 			if (!wasdel)
5368 				del.br_startblock += start - got.br_startoff;
5369 		}
5370 		if (del.br_startoff + del.br_blockcount > end + 1)
5371 			del.br_blockcount = end + 1 - del.br_startoff;
5372 
5373 		if (!isrt)
5374 			goto delete;
5375 
5376 		sum = del.br_startblock + del.br_blockcount;
5377 		div_u64_rem(sum, mp->m_sb.sb_rextsize, &mod);
5378 		if (mod) {
5379 			/*
5380 			 * Realtime extent not lined up at the end.
5381 			 * The extent could have been split into written
5382 			 * and unwritten pieces, or we could just be
5383 			 * unmapping part of it.  But we can't really
5384 			 * get rid of part of a realtime extent.
5385 			 */
5386 			if (del.br_state == XFS_EXT_UNWRITTEN) {
5387 				/*
5388 				 * This piece is unwritten, or we're not
5389 				 * using unwritten extents.  Skip over it.
5390 				 */
5391 				ASSERT(end >= mod);
5392 				end -= mod > del.br_blockcount ?
5393 					del.br_blockcount : mod;
5394 				if (end < got.br_startoff &&
5395 				    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5396 					done = true;
5397 					break;
5398 				}
5399 				continue;
5400 			}
5401 			/*
5402 			 * It's written, turn it unwritten.
5403 			 * This is better than zeroing it.
5404 			 */
5405 			ASSERT(del.br_state == XFS_EXT_NORM);
5406 			ASSERT(tp->t_blk_res > 0);
5407 			/*
5408 			 * If this spans a realtime extent boundary,
5409 			 * chop it back to the start of the one we end at.
5410 			 */
5411 			if (del.br_blockcount > mod) {
5412 				del.br_startoff += del.br_blockcount - mod;
5413 				del.br_startblock += del.br_blockcount - mod;
5414 				del.br_blockcount = mod;
5415 			}
5416 			del.br_state = XFS_EXT_UNWRITTEN;
5417 			error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5418 					whichfork, &icur, &cur, &del,
5419 					&logflags);
5420 			if (error)
5421 				goto error0;
5422 			goto nodelete;
5423 		}
5424 		div_u64_rem(del.br_startblock, mp->m_sb.sb_rextsize, &mod);
5425 		if (mod) {
5426 			xfs_extlen_t off = mp->m_sb.sb_rextsize - mod;
5427 
5428 			/*
5429 			 * Realtime extent is lined up at the end but not
5430 			 * at the front.  We'll get rid of full extents if
5431 			 * we can.
5432 			 */
5433 			if (del.br_blockcount > off) {
5434 				del.br_blockcount -= off;
5435 				del.br_startoff += off;
5436 				del.br_startblock += off;
5437 			} else if (del.br_startoff == start &&
5438 				   (del.br_state == XFS_EXT_UNWRITTEN ||
5439 				    tp->t_blk_res == 0)) {
5440 				/*
5441 				 * Can't make it unwritten.  There isn't
5442 				 * a full extent here so just skip it.
5443 				 */
5444 				ASSERT(end >= del.br_blockcount);
5445 				end -= del.br_blockcount;
5446 				if (got.br_startoff > end &&
5447 				    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5448 					done = true;
5449 					break;
5450 				}
5451 				continue;
5452 			} else if (del.br_state == XFS_EXT_UNWRITTEN) {
5453 				struct xfs_bmbt_irec	prev;
5454 				xfs_fileoff_t		unwrite_start;
5455 
5456 				/*
5457 				 * This one is already unwritten.
5458 				 * It must have a written left neighbor.
5459 				 * Unwrite the killed part of that one and
5460 				 * try again.
5461 				 */
5462 				if (!xfs_iext_prev_extent(ifp, &icur, &prev))
5463 					ASSERT(0);
5464 				ASSERT(prev.br_state == XFS_EXT_NORM);
5465 				ASSERT(!isnullstartblock(prev.br_startblock));
5466 				ASSERT(del.br_startblock ==
5467 				       prev.br_startblock + prev.br_blockcount);
5468 				unwrite_start = max3(start,
5469 						     del.br_startoff - mod,
5470 						     prev.br_startoff);
5471 				mod = unwrite_start - prev.br_startoff;
5472 				prev.br_startoff = unwrite_start;
5473 				prev.br_startblock += mod;
5474 				prev.br_blockcount -= mod;
5475 				prev.br_state = XFS_EXT_UNWRITTEN;
5476 				error = xfs_bmap_add_extent_unwritten_real(tp,
5477 						ip, whichfork, &icur, &cur,
5478 						&prev, &logflags);
5479 				if (error)
5480 					goto error0;
5481 				goto nodelete;
5482 			} else {
5483 				ASSERT(del.br_state == XFS_EXT_NORM);
5484 				del.br_state = XFS_EXT_UNWRITTEN;
5485 				error = xfs_bmap_add_extent_unwritten_real(tp,
5486 						ip, whichfork, &icur, &cur,
5487 						&del, &logflags);
5488 				if (error)
5489 					goto error0;
5490 				goto nodelete;
5491 			}
5492 		}
5493 
5494 delete:
5495 		if (wasdel) {
5496 			error = xfs_bmap_del_extent_delay(ip, whichfork, &icur,
5497 					&got, &del);
5498 		} else {
5499 			error = xfs_bmap_del_extent_real(ip, tp, &icur, cur,
5500 					&del, &tmp_logflags, whichfork,
5501 					flags);
5502 			logflags |= tmp_logflags;
5503 		}
5504 
5505 		if (error)
5506 			goto error0;
5507 
5508 		end = del.br_startoff - 1;
5509 nodelete:
5510 		/*
5511 		 * If not done go on to the next (previous) record.
5512 		 */
5513 		if (end != (xfs_fileoff_t)-1 && end >= start) {
5514 			if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5515 			    (got.br_startoff > end &&
5516 			     !xfs_iext_prev_extent(ifp, &icur, &got))) {
5517 				done = true;
5518 				break;
5519 			}
5520 			extno++;
5521 		}
5522 	}
5523 	if (done || end == (xfs_fileoff_t)-1 || end < start)
5524 		*rlen = 0;
5525 	else
5526 		*rlen = end - start + 1;
5527 
5528 	/*
5529 	 * Convert to a btree if necessary.
5530 	 */
5531 	if (xfs_bmap_needs_btree(ip, whichfork)) {
5532 		ASSERT(cur == NULL);
5533 		error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
5534 				&tmp_logflags, whichfork);
5535 		logflags |= tmp_logflags;
5536 	} else {
5537 		error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags,
5538 			whichfork);
5539 	}
5540 
5541 error0:
5542 	/*
5543 	 * Log everything.  Do this after conversion, there's no point in
5544 	 * logging the extent records if we've converted to btree format.
5545 	 */
5546 	if ((logflags & xfs_ilog_fext(whichfork)) &&
5547 	    ifp->if_format != XFS_DINODE_FMT_EXTENTS)
5548 		logflags &= ~xfs_ilog_fext(whichfork);
5549 	else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5550 		 ifp->if_format != XFS_DINODE_FMT_BTREE)
5551 		logflags &= ~xfs_ilog_fbroot(whichfork);
5552 	/*
5553 	 * Log inode even in the error case, if the transaction
5554 	 * is dirty we'll need to shut down the filesystem.
5555 	 */
5556 	if (logflags)
5557 		xfs_trans_log_inode(tp, ip, logflags);
5558 	if (cur) {
5559 		if (!error)
5560 			cur->bc_ino.allocated = 0;
5561 		xfs_btree_del_cursor(cur, error);
5562 	}
5563 	return error;
5564 }
5565 
5566 /* Unmap a range of a file. */
5567 int
5568 xfs_bunmapi(
5569 	xfs_trans_t		*tp,
5570 	struct xfs_inode	*ip,
5571 	xfs_fileoff_t		bno,
5572 	xfs_filblks_t		len,
5573 	uint32_t		flags,
5574 	xfs_extnum_t		nexts,
5575 	int			*done)
5576 {
5577 	int			error;
5578 
5579 	error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts);
5580 	*done = (len == 0);
5581 	return error;
5582 }
5583 
5584 /*
5585  * Determine whether an extent shift can be accomplished by a merge with the
5586  * extent that precedes the target hole of the shift.
5587  */
5588 STATIC bool
5589 xfs_bmse_can_merge(
5590 	struct xfs_bmbt_irec	*left,	/* preceding extent */
5591 	struct xfs_bmbt_irec	*got,	/* current extent to shift */
5592 	xfs_fileoff_t		shift)	/* shift fsb */
5593 {
5594 	xfs_fileoff_t		startoff;
5595 
5596 	startoff = got->br_startoff - shift;
5597 
5598 	/*
5599 	 * The extent, once shifted, must be adjacent in-file and on-disk with
5600 	 * the preceding extent.
5601 	 */
5602 	if ((left->br_startoff + left->br_blockcount != startoff) ||
5603 	    (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5604 	    (left->br_state != got->br_state) ||
5605 	    (left->br_blockcount + got->br_blockcount > XFS_MAX_BMBT_EXTLEN))
5606 		return false;
5607 
5608 	return true;
5609 }
5610 
5611 /*
5612  * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5613  * hole in the file. If an extent shift would result in the extent being fully
5614  * adjacent to the extent that currently precedes the hole, we can merge with
5615  * the preceding extent rather than do the shift.
5616  *
5617  * This function assumes the caller has verified a shift-by-merge is possible
5618  * with the provided extents via xfs_bmse_can_merge().
5619  */
5620 STATIC int
5621 xfs_bmse_merge(
5622 	struct xfs_trans		*tp,
5623 	struct xfs_inode		*ip,
5624 	int				whichfork,
5625 	xfs_fileoff_t			shift,		/* shift fsb */
5626 	struct xfs_iext_cursor		*icur,
5627 	struct xfs_bmbt_irec		*got,		/* extent to shift */
5628 	struct xfs_bmbt_irec		*left,		/* preceding extent */
5629 	struct xfs_btree_cur		*cur,
5630 	int				*logflags)	/* output */
5631 {
5632 	struct xfs_ifork		*ifp = xfs_ifork_ptr(ip, whichfork);
5633 	struct xfs_bmbt_irec		new;
5634 	xfs_filblks_t			blockcount;
5635 	int				error, i;
5636 	struct xfs_mount		*mp = ip->i_mount;
5637 
5638 	blockcount = left->br_blockcount + got->br_blockcount;
5639 
5640 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5641 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5642 	ASSERT(xfs_bmse_can_merge(left, got, shift));
5643 
5644 	new = *left;
5645 	new.br_blockcount = blockcount;
5646 
5647 	/*
5648 	 * Update the on-disk extent count, the btree if necessary and log the
5649 	 * inode.
5650 	 */
5651 	ifp->if_nextents--;
5652 	*logflags |= XFS_ILOG_CORE;
5653 	if (!cur) {
5654 		*logflags |= XFS_ILOG_DEXT;
5655 		goto done;
5656 	}
5657 
5658 	/* lookup and remove the extent to merge */
5659 	error = xfs_bmbt_lookup_eq(cur, got, &i);
5660 	if (error)
5661 		return error;
5662 	if (XFS_IS_CORRUPT(mp, i != 1))
5663 		return -EFSCORRUPTED;
5664 
5665 	error = xfs_btree_delete(cur, &i);
5666 	if (error)
5667 		return error;
5668 	if (XFS_IS_CORRUPT(mp, i != 1))
5669 		return -EFSCORRUPTED;
5670 
5671 	/* lookup and update size of the previous extent */
5672 	error = xfs_bmbt_lookup_eq(cur, left, &i);
5673 	if (error)
5674 		return error;
5675 	if (XFS_IS_CORRUPT(mp, i != 1))
5676 		return -EFSCORRUPTED;
5677 
5678 	error = xfs_bmbt_update(cur, &new);
5679 	if (error)
5680 		return error;
5681 
5682 	/* change to extent format if required after extent removal */
5683 	error = xfs_bmap_btree_to_extents(tp, ip, cur, logflags, whichfork);
5684 	if (error)
5685 		return error;
5686 
5687 done:
5688 	xfs_iext_remove(ip, icur, 0);
5689 	xfs_iext_prev(ifp, icur);
5690 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5691 			&new);
5692 
5693 	/* update reverse mapping. rmap functions merge the rmaps for us */
5694 	xfs_rmap_unmap_extent(tp, ip, whichfork, got);
5695 	memcpy(&new, got, sizeof(new));
5696 	new.br_startoff = left->br_startoff + left->br_blockcount;
5697 	xfs_rmap_map_extent(tp, ip, whichfork, &new);
5698 	return 0;
5699 }
5700 
5701 static int
5702 xfs_bmap_shift_update_extent(
5703 	struct xfs_trans	*tp,
5704 	struct xfs_inode	*ip,
5705 	int			whichfork,
5706 	struct xfs_iext_cursor	*icur,
5707 	struct xfs_bmbt_irec	*got,
5708 	struct xfs_btree_cur	*cur,
5709 	int			*logflags,
5710 	xfs_fileoff_t		startoff)
5711 {
5712 	struct xfs_mount	*mp = ip->i_mount;
5713 	struct xfs_bmbt_irec	prev = *got;
5714 	int			error, i;
5715 
5716 	*logflags |= XFS_ILOG_CORE;
5717 
5718 	got->br_startoff = startoff;
5719 
5720 	if (cur) {
5721 		error = xfs_bmbt_lookup_eq(cur, &prev, &i);
5722 		if (error)
5723 			return error;
5724 		if (XFS_IS_CORRUPT(mp, i != 1))
5725 			return -EFSCORRUPTED;
5726 
5727 		error = xfs_bmbt_update(cur, got);
5728 		if (error)
5729 			return error;
5730 	} else {
5731 		*logflags |= XFS_ILOG_DEXT;
5732 	}
5733 
5734 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5735 			got);
5736 
5737 	/* update reverse mapping */
5738 	xfs_rmap_unmap_extent(tp, ip, whichfork, &prev);
5739 	xfs_rmap_map_extent(tp, ip, whichfork, got);
5740 	return 0;
5741 }
5742 
5743 int
5744 xfs_bmap_collapse_extents(
5745 	struct xfs_trans	*tp,
5746 	struct xfs_inode	*ip,
5747 	xfs_fileoff_t		*next_fsb,
5748 	xfs_fileoff_t		offset_shift_fsb,
5749 	bool			*done)
5750 {
5751 	int			whichfork = XFS_DATA_FORK;
5752 	struct xfs_mount	*mp = ip->i_mount;
5753 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
5754 	struct xfs_btree_cur	*cur = NULL;
5755 	struct xfs_bmbt_irec	got, prev;
5756 	struct xfs_iext_cursor	icur;
5757 	xfs_fileoff_t		new_startoff;
5758 	int			error = 0;
5759 	int			logflags = 0;
5760 
5761 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
5762 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
5763 		return -EFSCORRUPTED;
5764 	}
5765 
5766 	if (xfs_is_shutdown(mp))
5767 		return -EIO;
5768 
5769 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL));
5770 
5771 	error = xfs_iread_extents(tp, ip, whichfork);
5772 	if (error)
5773 		return error;
5774 
5775 	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
5776 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5777 		cur->bc_ino.flags = 0;
5778 	}
5779 
5780 	if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5781 		*done = true;
5782 		goto del_cursor;
5783 	}
5784 	if (XFS_IS_CORRUPT(mp, isnullstartblock(got.br_startblock))) {
5785 		error = -EFSCORRUPTED;
5786 		goto del_cursor;
5787 	}
5788 
5789 	new_startoff = got.br_startoff - offset_shift_fsb;
5790 	if (xfs_iext_peek_prev_extent(ifp, &icur, &prev)) {
5791 		if (new_startoff < prev.br_startoff + prev.br_blockcount) {
5792 			error = -EINVAL;
5793 			goto del_cursor;
5794 		}
5795 
5796 		if (xfs_bmse_can_merge(&prev, &got, offset_shift_fsb)) {
5797 			error = xfs_bmse_merge(tp, ip, whichfork,
5798 					offset_shift_fsb, &icur, &got, &prev,
5799 					cur, &logflags);
5800 			if (error)
5801 				goto del_cursor;
5802 			goto done;
5803 		}
5804 	} else {
5805 		if (got.br_startoff < offset_shift_fsb) {
5806 			error = -EINVAL;
5807 			goto del_cursor;
5808 		}
5809 	}
5810 
5811 	error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got,
5812 			cur, &logflags, new_startoff);
5813 	if (error)
5814 		goto del_cursor;
5815 
5816 done:
5817 	if (!xfs_iext_next_extent(ifp, &icur, &got)) {
5818 		*done = true;
5819 		goto del_cursor;
5820 	}
5821 
5822 	*next_fsb = got.br_startoff;
5823 del_cursor:
5824 	if (cur)
5825 		xfs_btree_del_cursor(cur, error);
5826 	if (logflags)
5827 		xfs_trans_log_inode(tp, ip, logflags);
5828 	return error;
5829 }
5830 
5831 /* Make sure we won't be right-shifting an extent past the maximum bound. */
5832 int
5833 xfs_bmap_can_insert_extents(
5834 	struct xfs_inode	*ip,
5835 	xfs_fileoff_t		off,
5836 	xfs_fileoff_t		shift)
5837 {
5838 	struct xfs_bmbt_irec	got;
5839 	int			is_empty;
5840 	int			error = 0;
5841 
5842 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5843 
5844 	if (xfs_is_shutdown(ip->i_mount))
5845 		return -EIO;
5846 
5847 	xfs_ilock(ip, XFS_ILOCK_EXCL);
5848 	error = xfs_bmap_last_extent(NULL, ip, XFS_DATA_FORK, &got, &is_empty);
5849 	if (!error && !is_empty && got.br_startoff >= off &&
5850 	    ((got.br_startoff + shift) & BMBT_STARTOFF_MASK) < got.br_startoff)
5851 		error = -EINVAL;
5852 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
5853 
5854 	return error;
5855 }
5856 
5857 int
5858 xfs_bmap_insert_extents(
5859 	struct xfs_trans	*tp,
5860 	struct xfs_inode	*ip,
5861 	xfs_fileoff_t		*next_fsb,
5862 	xfs_fileoff_t		offset_shift_fsb,
5863 	bool			*done,
5864 	xfs_fileoff_t		stop_fsb)
5865 {
5866 	int			whichfork = XFS_DATA_FORK;
5867 	struct xfs_mount	*mp = ip->i_mount;
5868 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
5869 	struct xfs_btree_cur	*cur = NULL;
5870 	struct xfs_bmbt_irec	got, next;
5871 	struct xfs_iext_cursor	icur;
5872 	xfs_fileoff_t		new_startoff;
5873 	int			error = 0;
5874 	int			logflags = 0;
5875 
5876 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
5877 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
5878 		return -EFSCORRUPTED;
5879 	}
5880 
5881 	if (xfs_is_shutdown(mp))
5882 		return -EIO;
5883 
5884 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL));
5885 
5886 	error = xfs_iread_extents(tp, ip, whichfork);
5887 	if (error)
5888 		return error;
5889 
5890 	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
5891 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5892 		cur->bc_ino.flags = 0;
5893 	}
5894 
5895 	if (*next_fsb == NULLFSBLOCK) {
5896 		xfs_iext_last(ifp, &icur);
5897 		if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5898 		    stop_fsb > got.br_startoff) {
5899 			*done = true;
5900 			goto del_cursor;
5901 		}
5902 	} else {
5903 		if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5904 			*done = true;
5905 			goto del_cursor;
5906 		}
5907 	}
5908 	if (XFS_IS_CORRUPT(mp, isnullstartblock(got.br_startblock))) {
5909 		error = -EFSCORRUPTED;
5910 		goto del_cursor;
5911 	}
5912 
5913 	if (XFS_IS_CORRUPT(mp, stop_fsb > got.br_startoff)) {
5914 		error = -EFSCORRUPTED;
5915 		goto del_cursor;
5916 	}
5917 
5918 	new_startoff = got.br_startoff + offset_shift_fsb;
5919 	if (xfs_iext_peek_next_extent(ifp, &icur, &next)) {
5920 		if (new_startoff + got.br_blockcount > next.br_startoff) {
5921 			error = -EINVAL;
5922 			goto del_cursor;
5923 		}
5924 
5925 		/*
5926 		 * Unlike a left shift (which involves a hole punch), a right
5927 		 * shift does not modify extent neighbors in any way.  We should
5928 		 * never find mergeable extents in this scenario.  Check anyways
5929 		 * and warn if we encounter two extents that could be one.
5930 		 */
5931 		if (xfs_bmse_can_merge(&got, &next, offset_shift_fsb))
5932 			WARN_ON_ONCE(1);
5933 	}
5934 
5935 	error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got,
5936 			cur, &logflags, new_startoff);
5937 	if (error)
5938 		goto del_cursor;
5939 
5940 	if (!xfs_iext_prev_extent(ifp, &icur, &got) ||
5941 	    stop_fsb >= got.br_startoff + got.br_blockcount) {
5942 		*done = true;
5943 		goto del_cursor;
5944 	}
5945 
5946 	*next_fsb = got.br_startoff;
5947 del_cursor:
5948 	if (cur)
5949 		xfs_btree_del_cursor(cur, error);
5950 	if (logflags)
5951 		xfs_trans_log_inode(tp, ip, logflags);
5952 	return error;
5953 }
5954 
5955 /*
5956  * Splits an extent into two extents at split_fsb block such that it is the
5957  * first block of the current_ext. @ext is a target extent to be split.
5958  * @split_fsb is a block where the extents is split.  If split_fsb lies in a
5959  * hole or the first block of extents, just return 0.
5960  */
5961 int
5962 xfs_bmap_split_extent(
5963 	struct xfs_trans	*tp,
5964 	struct xfs_inode	*ip,
5965 	xfs_fileoff_t		split_fsb)
5966 {
5967 	int				whichfork = XFS_DATA_FORK;
5968 	struct xfs_ifork		*ifp = xfs_ifork_ptr(ip, whichfork);
5969 	struct xfs_btree_cur		*cur = NULL;
5970 	struct xfs_bmbt_irec		got;
5971 	struct xfs_bmbt_irec		new; /* split extent */
5972 	struct xfs_mount		*mp = ip->i_mount;
5973 	xfs_fsblock_t			gotblkcnt; /* new block count for got */
5974 	struct xfs_iext_cursor		icur;
5975 	int				error = 0;
5976 	int				logflags = 0;
5977 	int				i = 0;
5978 
5979 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
5980 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
5981 		return -EFSCORRUPTED;
5982 	}
5983 
5984 	if (xfs_is_shutdown(mp))
5985 		return -EIO;
5986 
5987 	/* Read in all the extents */
5988 	error = xfs_iread_extents(tp, ip, whichfork);
5989 	if (error)
5990 		return error;
5991 
5992 	/*
5993 	 * If there are not extents, or split_fsb lies in a hole we are done.
5994 	 */
5995 	if (!xfs_iext_lookup_extent(ip, ifp, split_fsb, &icur, &got) ||
5996 	    got.br_startoff >= split_fsb)
5997 		return 0;
5998 
5999 	gotblkcnt = split_fsb - got.br_startoff;
6000 	new.br_startoff = split_fsb;
6001 	new.br_startblock = got.br_startblock + gotblkcnt;
6002 	new.br_blockcount = got.br_blockcount - gotblkcnt;
6003 	new.br_state = got.br_state;
6004 
6005 	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
6006 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6007 		cur->bc_ino.flags = 0;
6008 		error = xfs_bmbt_lookup_eq(cur, &got, &i);
6009 		if (error)
6010 			goto del_cursor;
6011 		if (XFS_IS_CORRUPT(mp, i != 1)) {
6012 			error = -EFSCORRUPTED;
6013 			goto del_cursor;
6014 		}
6015 	}
6016 
6017 	got.br_blockcount = gotblkcnt;
6018 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), &icur,
6019 			&got);
6020 
6021 	logflags = XFS_ILOG_CORE;
6022 	if (cur) {
6023 		error = xfs_bmbt_update(cur, &got);
6024 		if (error)
6025 			goto del_cursor;
6026 	} else
6027 		logflags |= XFS_ILOG_DEXT;
6028 
6029 	/* Add new extent */
6030 	xfs_iext_next(ifp, &icur);
6031 	xfs_iext_insert(ip, &icur, &new, 0);
6032 	ifp->if_nextents++;
6033 
6034 	if (cur) {
6035 		error = xfs_bmbt_lookup_eq(cur, &new, &i);
6036 		if (error)
6037 			goto del_cursor;
6038 		if (XFS_IS_CORRUPT(mp, i != 0)) {
6039 			error = -EFSCORRUPTED;
6040 			goto del_cursor;
6041 		}
6042 		error = xfs_btree_insert(cur, &i);
6043 		if (error)
6044 			goto del_cursor;
6045 		if (XFS_IS_CORRUPT(mp, i != 1)) {
6046 			error = -EFSCORRUPTED;
6047 			goto del_cursor;
6048 		}
6049 	}
6050 
6051 	/*
6052 	 * Convert to a btree if necessary.
6053 	 */
6054 	if (xfs_bmap_needs_btree(ip, whichfork)) {
6055 		int tmp_logflags; /* partial log flag return val */
6056 
6057 		ASSERT(cur == NULL);
6058 		error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
6059 				&tmp_logflags, whichfork);
6060 		logflags |= tmp_logflags;
6061 	}
6062 
6063 del_cursor:
6064 	if (cur) {
6065 		cur->bc_ino.allocated = 0;
6066 		xfs_btree_del_cursor(cur, error);
6067 	}
6068 
6069 	if (logflags)
6070 		xfs_trans_log_inode(tp, ip, logflags);
6071 	return error;
6072 }
6073 
6074 /* Deferred mapping is only for real extents in the data fork. */
6075 static bool
6076 xfs_bmap_is_update_needed(
6077 	struct xfs_bmbt_irec	*bmap)
6078 {
6079 	return  bmap->br_startblock != HOLESTARTBLOCK &&
6080 		bmap->br_startblock != DELAYSTARTBLOCK;
6081 }
6082 
6083 /* Record a bmap intent. */
6084 static int
6085 __xfs_bmap_add(
6086 	struct xfs_trans		*tp,
6087 	enum xfs_bmap_intent_type	type,
6088 	struct xfs_inode		*ip,
6089 	int				whichfork,
6090 	struct xfs_bmbt_irec		*bmap)
6091 {
6092 	struct xfs_bmap_intent		*bi;
6093 
6094 	trace_xfs_bmap_defer(tp->t_mountp,
6095 			XFS_FSB_TO_AGNO(tp->t_mountp, bmap->br_startblock),
6096 			type,
6097 			XFS_FSB_TO_AGBNO(tp->t_mountp, bmap->br_startblock),
6098 			ip->i_ino, whichfork,
6099 			bmap->br_startoff,
6100 			bmap->br_blockcount,
6101 			bmap->br_state);
6102 
6103 	bi = kmem_cache_alloc(xfs_bmap_intent_cache, GFP_NOFS | __GFP_NOFAIL);
6104 	INIT_LIST_HEAD(&bi->bi_list);
6105 	bi->bi_type = type;
6106 	bi->bi_owner = ip;
6107 	bi->bi_whichfork = whichfork;
6108 	bi->bi_bmap = *bmap;
6109 
6110 	xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list);
6111 	return 0;
6112 }
6113 
6114 /* Map an extent into a file. */
6115 void
6116 xfs_bmap_map_extent(
6117 	struct xfs_trans	*tp,
6118 	struct xfs_inode	*ip,
6119 	struct xfs_bmbt_irec	*PREV)
6120 {
6121 	if (!xfs_bmap_is_update_needed(PREV))
6122 		return;
6123 
6124 	__xfs_bmap_add(tp, XFS_BMAP_MAP, ip, XFS_DATA_FORK, PREV);
6125 }
6126 
6127 /* Unmap an extent out of a file. */
6128 void
6129 xfs_bmap_unmap_extent(
6130 	struct xfs_trans	*tp,
6131 	struct xfs_inode	*ip,
6132 	struct xfs_bmbt_irec	*PREV)
6133 {
6134 	if (!xfs_bmap_is_update_needed(PREV))
6135 		return;
6136 
6137 	__xfs_bmap_add(tp, XFS_BMAP_UNMAP, ip, XFS_DATA_FORK, PREV);
6138 }
6139 
6140 /*
6141  * Process one of the deferred bmap operations.  We pass back the
6142  * btree cursor to maintain our lock on the bmapbt between calls.
6143  */
6144 int
6145 xfs_bmap_finish_one(
6146 	struct xfs_trans		*tp,
6147 	struct xfs_inode		*ip,
6148 	enum xfs_bmap_intent_type	type,
6149 	int				whichfork,
6150 	xfs_fileoff_t			startoff,
6151 	xfs_fsblock_t			startblock,
6152 	xfs_filblks_t			*blockcount,
6153 	xfs_exntst_t			state)
6154 {
6155 	int				error = 0;
6156 
6157 	ASSERT(tp->t_firstblock == NULLFSBLOCK);
6158 
6159 	trace_xfs_bmap_deferred(tp->t_mountp,
6160 			XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type,
6161 			XFS_FSB_TO_AGBNO(tp->t_mountp, startblock),
6162 			ip->i_ino, whichfork, startoff, *blockcount, state);
6163 
6164 	if (WARN_ON_ONCE(whichfork != XFS_DATA_FORK))
6165 		return -EFSCORRUPTED;
6166 
6167 	if (XFS_TEST_ERROR(false, tp->t_mountp,
6168 			XFS_ERRTAG_BMAP_FINISH_ONE))
6169 		return -EIO;
6170 
6171 	switch (type) {
6172 	case XFS_BMAP_MAP:
6173 		error = xfs_bmapi_remap(tp, ip, startoff, *blockcount,
6174 				startblock, 0);
6175 		*blockcount = 0;
6176 		break;
6177 	case XFS_BMAP_UNMAP:
6178 		error = __xfs_bunmapi(tp, ip, startoff, blockcount,
6179 				XFS_BMAPI_REMAP, 1);
6180 		break;
6181 	default:
6182 		ASSERT(0);
6183 		error = -EFSCORRUPTED;
6184 	}
6185 
6186 	return error;
6187 }
6188 
6189 /* Check that an inode's extent does not have invalid flags or bad ranges. */
6190 xfs_failaddr_t
6191 xfs_bmap_validate_extent(
6192 	struct xfs_inode	*ip,
6193 	int			whichfork,
6194 	struct xfs_bmbt_irec	*irec)
6195 {
6196 	struct xfs_mount	*mp = ip->i_mount;
6197 
6198 	if (!xfs_verify_fileext(mp, irec->br_startoff, irec->br_blockcount))
6199 		return __this_address;
6200 
6201 	if (XFS_IS_REALTIME_INODE(ip) && whichfork == XFS_DATA_FORK) {
6202 		if (!xfs_verify_rtext(mp, irec->br_startblock,
6203 					  irec->br_blockcount))
6204 			return __this_address;
6205 	} else {
6206 		if (!xfs_verify_fsbext(mp, irec->br_startblock,
6207 					   irec->br_blockcount))
6208 			return __this_address;
6209 	}
6210 	if (irec->br_state != XFS_EXT_NORM && whichfork != XFS_DATA_FORK)
6211 		return __this_address;
6212 	return NULL;
6213 }
6214 
6215 int __init
6216 xfs_bmap_intent_init_cache(void)
6217 {
6218 	xfs_bmap_intent_cache = kmem_cache_create("xfs_bmap_intent",
6219 			sizeof(struct xfs_bmap_intent),
6220 			0, 0, NULL);
6221 
6222 	return xfs_bmap_intent_cache != NULL ? 0 : -ENOMEM;
6223 }
6224 
6225 void
6226 xfs_bmap_intent_destroy_cache(void)
6227 {
6228 	kmem_cache_destroy(xfs_bmap_intent_cache);
6229 	xfs_bmap_intent_cache = NULL;
6230 }
6231