xref: /openbmc/linux/fs/xfs/libxfs/xfs_bmap.c (revision 67559900)
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 	int		level;		/* btree level */
56 	uint		maxblocks;	/* max blocks at this level */
57 	uint		maxleafents;	/* max leaf entries possible */
58 	int		maxrootrecs;	/* max records in root block */
59 	int		minleafrecs;	/* min records in leaf block */
60 	int		minnoderecs;	/* min records in node block */
61 	int		sz;		/* root block size */
62 
63 	/*
64 	 * The maximum number of extents in a file, hence the maximum number of
65 	 * leaf entries, is controlled by the size of the on-disk extent count,
66 	 * either a signed 32-bit number for the data fork, or a signed 16-bit
67 	 * number for the attr fork.
68 	 *
69 	 * Note that we can no longer assume that if we are in ATTR1 that the
70 	 * fork offset of all the inodes will be
71 	 * (xfs_default_attroffset(ip) >> 3) because we could have mounted with
72 	 * ATTR2 and then mounted back with ATTR1, keeping the i_forkoff's fixed
73 	 * but probably at various positions. Therefore, for both ATTR1 and
74 	 * ATTR2 we have to assume the worst case scenario of a minimum size
75 	 * available.
76 	 */
77 	if (whichfork == XFS_DATA_FORK) {
78 		maxleafents = MAXEXTNUM;
79 		sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
80 	} else {
81 		maxleafents = MAXAEXTNUM;
82 		sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
83 	}
84 	maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
85 	minleafrecs = mp->m_bmap_dmnr[0];
86 	minnoderecs = mp->m_bmap_dmnr[1];
87 	maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
88 	for (level = 1; maxblocks > 1; level++) {
89 		if (maxblocks <= maxrootrecs)
90 			maxblocks = 1;
91 		else
92 			maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
93 	}
94 	mp->m_bm_maxlevels[whichfork] = level;
95 	ASSERT(mp->m_bm_maxlevels[whichfork] <= xfs_bmbt_maxlevels_ondisk());
96 }
97 
98 unsigned int
99 xfs_bmap_compute_attr_offset(
100 	struct xfs_mount	*mp)
101 {
102 	if (mp->m_sb.sb_inodesize == 256)
103 		return XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
104 	return XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
105 }
106 
107 STATIC int				/* error */
108 xfs_bmbt_lookup_eq(
109 	struct xfs_btree_cur	*cur,
110 	struct xfs_bmbt_irec	*irec,
111 	int			*stat)	/* success/failure */
112 {
113 	cur->bc_rec.b = *irec;
114 	return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
115 }
116 
117 STATIC int				/* error */
118 xfs_bmbt_lookup_first(
119 	struct xfs_btree_cur	*cur,
120 	int			*stat)	/* success/failure */
121 {
122 	cur->bc_rec.b.br_startoff = 0;
123 	cur->bc_rec.b.br_startblock = 0;
124 	cur->bc_rec.b.br_blockcount = 0;
125 	return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
126 }
127 
128 /*
129  * Check if the inode needs to be converted to btree format.
130  */
131 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
132 {
133 	struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
134 
135 	return whichfork != XFS_COW_FORK &&
136 		ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
137 		ifp->if_nextents > XFS_IFORK_MAXEXT(ip, whichfork);
138 }
139 
140 /*
141  * Check if the inode should be converted to extent format.
142  */
143 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
144 {
145 	struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
146 
147 	return whichfork != XFS_COW_FORK &&
148 		ifp->if_format == XFS_DINODE_FMT_BTREE &&
149 		ifp->if_nextents <= XFS_IFORK_MAXEXT(ip, whichfork);
150 }
151 
152 /*
153  * Update the record referred to by cur to the value given by irec
154  * This either works (return 0) or gets an EFSCORRUPTED error.
155  */
156 STATIC int
157 xfs_bmbt_update(
158 	struct xfs_btree_cur	*cur,
159 	struct xfs_bmbt_irec	*irec)
160 {
161 	union xfs_btree_rec	rec;
162 
163 	xfs_bmbt_disk_set_all(&rec.bmbt, irec);
164 	return xfs_btree_update(cur, &rec);
165 }
166 
167 /*
168  * Compute the worst-case number of indirect blocks that will be used
169  * for ip's delayed extent of length "len".
170  */
171 STATIC xfs_filblks_t
172 xfs_bmap_worst_indlen(
173 	xfs_inode_t	*ip,		/* incore inode pointer */
174 	xfs_filblks_t	len)		/* delayed extent length */
175 {
176 	int		level;		/* btree level number */
177 	int		maxrecs;	/* maximum record count at this level */
178 	xfs_mount_t	*mp;		/* mount structure */
179 	xfs_filblks_t	rval;		/* return value */
180 
181 	mp = ip->i_mount;
182 	maxrecs = mp->m_bmap_dmxr[0];
183 	for (level = 0, rval = 0;
184 	     level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
185 	     level++) {
186 		len += maxrecs - 1;
187 		do_div(len, maxrecs);
188 		rval += len;
189 		if (len == 1)
190 			return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
191 				level - 1;
192 		if (level == 0)
193 			maxrecs = mp->m_bmap_dmxr[1];
194 	}
195 	return rval;
196 }
197 
198 /*
199  * Calculate the default attribute fork offset for newly created inodes.
200  */
201 uint
202 xfs_default_attroffset(
203 	struct xfs_inode	*ip)
204 {
205 	if (ip->i_df.if_format == XFS_DINODE_FMT_DEV)
206 		return roundup(sizeof(xfs_dev_t), 8);
207 	return M_IGEO(ip->i_mount)->attr_fork_offset;
208 }
209 
210 /*
211  * Helper routine to reset inode i_forkoff field when switching attribute fork
212  * from local to extent format - we reset it where possible to make space
213  * available for inline data fork extents.
214  */
215 STATIC void
216 xfs_bmap_forkoff_reset(
217 	xfs_inode_t	*ip,
218 	int		whichfork)
219 {
220 	if (whichfork == XFS_ATTR_FORK &&
221 	    ip->i_df.if_format != XFS_DINODE_FMT_DEV &&
222 	    ip->i_df.if_format != XFS_DINODE_FMT_BTREE) {
223 		uint	dfl_forkoff = xfs_default_attroffset(ip) >> 3;
224 
225 		if (dfl_forkoff > ip->i_forkoff)
226 			ip->i_forkoff = dfl_forkoff;
227 	}
228 }
229 
230 #ifdef DEBUG
231 STATIC struct xfs_buf *
232 xfs_bmap_get_bp(
233 	struct xfs_btree_cur	*cur,
234 	xfs_fsblock_t		bno)
235 {
236 	struct xfs_log_item	*lip;
237 	int			i;
238 
239 	if (!cur)
240 		return NULL;
241 
242 	for (i = 0; i < cur->bc_maxlevels; i++) {
243 		if (!cur->bc_levels[i].bp)
244 			break;
245 		if (xfs_buf_daddr(cur->bc_levels[i].bp) == bno)
246 			return cur->bc_levels[i].bp;
247 	}
248 
249 	/* Chase down all the log items to see if the bp is there */
250 	list_for_each_entry(lip, &cur->bc_tp->t_items, li_trans) {
251 		struct xfs_buf_log_item	*bip = (struct xfs_buf_log_item *)lip;
252 
253 		if (bip->bli_item.li_type == XFS_LI_BUF &&
254 		    xfs_buf_daddr(bip->bli_buf) == bno)
255 			return bip->bli_buf;
256 	}
257 
258 	return NULL;
259 }
260 
261 STATIC void
262 xfs_check_block(
263 	struct xfs_btree_block	*block,
264 	xfs_mount_t		*mp,
265 	int			root,
266 	short			sz)
267 {
268 	int			i, j, dmxr;
269 	__be64			*pp, *thispa;	/* pointer to block address */
270 	xfs_bmbt_key_t		*prevp, *keyp;
271 
272 	ASSERT(be16_to_cpu(block->bb_level) > 0);
273 
274 	prevp = NULL;
275 	for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
276 		dmxr = mp->m_bmap_dmxr[0];
277 		keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
278 
279 		if (prevp) {
280 			ASSERT(be64_to_cpu(prevp->br_startoff) <
281 			       be64_to_cpu(keyp->br_startoff));
282 		}
283 		prevp = keyp;
284 
285 		/*
286 		 * Compare the block numbers to see if there are dups.
287 		 */
288 		if (root)
289 			pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
290 		else
291 			pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
292 
293 		for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
294 			if (root)
295 				thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
296 			else
297 				thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
298 			if (*thispa == *pp) {
299 				xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
300 					__func__, j, i,
301 					(unsigned long long)be64_to_cpu(*thispa));
302 				xfs_err(mp, "%s: ptrs are equal in node\n",
303 					__func__);
304 				xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
305 			}
306 		}
307 	}
308 }
309 
310 /*
311  * Check that the extents for the inode ip are in the right order in all
312  * btree leaves. THis becomes prohibitively expensive for large extent count
313  * files, so don't bother with inodes that have more than 10,000 extents in
314  * them. The btree record ordering checks will still be done, so for such large
315  * bmapbt constructs that is going to catch most corruptions.
316  */
317 STATIC void
318 xfs_bmap_check_leaf_extents(
319 	struct xfs_btree_cur	*cur,	/* btree cursor or null */
320 	xfs_inode_t		*ip,		/* incore inode pointer */
321 	int			whichfork)	/* data or attr fork */
322 {
323 	struct xfs_mount	*mp = ip->i_mount;
324 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
325 	struct xfs_btree_block	*block;	/* current btree block */
326 	xfs_fsblock_t		bno;	/* block # of "block" */
327 	struct xfs_buf		*bp;	/* buffer for "block" */
328 	int			error;	/* error return value */
329 	xfs_extnum_t		i=0, j;	/* index into the extents list */
330 	int			level;	/* btree level, for checking */
331 	__be64			*pp;	/* pointer to block address */
332 	xfs_bmbt_rec_t		*ep;	/* pointer to current extent */
333 	xfs_bmbt_rec_t		last = {0, 0}; /* last extent in prev block */
334 	xfs_bmbt_rec_t		*nextp;	/* pointer to next extent */
335 	int			bp_release = 0;
336 
337 	if (ifp->if_format != XFS_DINODE_FMT_BTREE)
338 		return;
339 
340 	/* skip large extent count inodes */
341 	if (ip->i_df.if_nextents > 10000)
342 		return;
343 
344 	bno = NULLFSBLOCK;
345 	block = ifp->if_broot;
346 	/*
347 	 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
348 	 */
349 	level = be16_to_cpu(block->bb_level);
350 	ASSERT(level > 0);
351 	xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
352 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
353 	bno = be64_to_cpu(*pp);
354 
355 	ASSERT(bno != NULLFSBLOCK);
356 	ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
357 	ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
358 
359 	/*
360 	 * Go down the tree until leaf level is reached, following the first
361 	 * pointer (leftmost) at each level.
362 	 */
363 	while (level-- > 0) {
364 		/* See if buf is in cur first */
365 		bp_release = 0;
366 		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
367 		if (!bp) {
368 			bp_release = 1;
369 			error = xfs_btree_read_bufl(mp, NULL, bno, &bp,
370 						XFS_BMAP_BTREE_REF,
371 						&xfs_bmbt_buf_ops);
372 			if (error)
373 				goto error_norelse;
374 		}
375 		block = XFS_BUF_TO_BLOCK(bp);
376 		if (level == 0)
377 			break;
378 
379 		/*
380 		 * Check this block for basic sanity (increasing keys and
381 		 * no duplicate blocks).
382 		 */
383 
384 		xfs_check_block(block, mp, 0, 0);
385 		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
386 		bno = be64_to_cpu(*pp);
387 		if (XFS_IS_CORRUPT(mp, !xfs_verify_fsbno(mp, bno))) {
388 			error = -EFSCORRUPTED;
389 			goto error0;
390 		}
391 		if (bp_release) {
392 			bp_release = 0;
393 			xfs_trans_brelse(NULL, bp);
394 		}
395 	}
396 
397 	/*
398 	 * Here with bp and block set to the leftmost leaf node in the tree.
399 	 */
400 	i = 0;
401 
402 	/*
403 	 * Loop over all leaf nodes checking that all extents are in the right order.
404 	 */
405 	for (;;) {
406 		xfs_fsblock_t	nextbno;
407 		xfs_extnum_t	num_recs;
408 
409 
410 		num_recs = xfs_btree_get_numrecs(block);
411 
412 		/*
413 		 * Read-ahead the next leaf block, if any.
414 		 */
415 
416 		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
417 
418 		/*
419 		 * Check all the extents to make sure they are OK.
420 		 * If we had a previous block, the last entry should
421 		 * conform with the first entry in this one.
422 		 */
423 
424 		ep = XFS_BMBT_REC_ADDR(mp, block, 1);
425 		if (i) {
426 			ASSERT(xfs_bmbt_disk_get_startoff(&last) +
427 			       xfs_bmbt_disk_get_blockcount(&last) <=
428 			       xfs_bmbt_disk_get_startoff(ep));
429 		}
430 		for (j = 1; j < num_recs; j++) {
431 			nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
432 			ASSERT(xfs_bmbt_disk_get_startoff(ep) +
433 			       xfs_bmbt_disk_get_blockcount(ep) <=
434 			       xfs_bmbt_disk_get_startoff(nextp));
435 			ep = nextp;
436 		}
437 
438 		last = *ep;
439 		i += num_recs;
440 		if (bp_release) {
441 			bp_release = 0;
442 			xfs_trans_brelse(NULL, bp);
443 		}
444 		bno = nextbno;
445 		/*
446 		 * If we've reached the end, stop.
447 		 */
448 		if (bno == NULLFSBLOCK)
449 			break;
450 
451 		bp_release = 0;
452 		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
453 		if (!bp) {
454 			bp_release = 1;
455 			error = xfs_btree_read_bufl(mp, NULL, bno, &bp,
456 						XFS_BMAP_BTREE_REF,
457 						&xfs_bmbt_buf_ops);
458 			if (error)
459 				goto error_norelse;
460 		}
461 		block = XFS_BUF_TO_BLOCK(bp);
462 	}
463 
464 	return;
465 
466 error0:
467 	xfs_warn(mp, "%s: at error0", __func__);
468 	if (bp_release)
469 		xfs_trans_brelse(NULL, bp);
470 error_norelse:
471 	xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
472 		__func__, i);
473 	xfs_err(mp, "%s: CORRUPTED BTREE OR SOMETHING", __func__);
474 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
475 	return;
476 }
477 
478 /*
479  * Validate that the bmbt_irecs being returned from bmapi are valid
480  * given the caller's original parameters.  Specifically check the
481  * ranges of the returned irecs to ensure that they only extend beyond
482  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
483  */
484 STATIC void
485 xfs_bmap_validate_ret(
486 	xfs_fileoff_t		bno,
487 	xfs_filblks_t		len,
488 	int			flags,
489 	xfs_bmbt_irec_t		*mval,
490 	int			nmap,
491 	int			ret_nmap)
492 {
493 	int			i;		/* index to map values */
494 
495 	ASSERT(ret_nmap <= nmap);
496 
497 	for (i = 0; i < ret_nmap; i++) {
498 		ASSERT(mval[i].br_blockcount > 0);
499 		if (!(flags & XFS_BMAPI_ENTIRE)) {
500 			ASSERT(mval[i].br_startoff >= bno);
501 			ASSERT(mval[i].br_blockcount <= len);
502 			ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
503 			       bno + len);
504 		} else {
505 			ASSERT(mval[i].br_startoff < bno + len);
506 			ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
507 			       bno);
508 		}
509 		ASSERT(i == 0 ||
510 		       mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
511 		       mval[i].br_startoff);
512 		ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
513 		       mval[i].br_startblock != HOLESTARTBLOCK);
514 		ASSERT(mval[i].br_state == XFS_EXT_NORM ||
515 		       mval[i].br_state == XFS_EXT_UNWRITTEN);
516 	}
517 }
518 
519 #else
520 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)		do { } while (0)
521 #define	xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)	do { } while (0)
522 #endif /* DEBUG */
523 
524 /*
525  * Inode fork format manipulation functions
526  */
527 
528 /*
529  * Convert the inode format to extent format if it currently is in btree format,
530  * but the extent list is small enough that it fits into the extent format.
531  *
532  * Since the extents are already in-core, all we have to do is give up the space
533  * for the btree root and pitch the leaf block.
534  */
535 STATIC int				/* error */
536 xfs_bmap_btree_to_extents(
537 	struct xfs_trans	*tp,	/* transaction pointer */
538 	struct xfs_inode	*ip,	/* incore inode pointer */
539 	struct xfs_btree_cur	*cur,	/* btree cursor */
540 	int			*logflagsp, /* inode logging flags */
541 	int			whichfork)  /* data or attr fork */
542 {
543 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
544 	struct xfs_mount	*mp = ip->i_mount;
545 	struct xfs_btree_block	*rblock = ifp->if_broot;
546 	struct xfs_btree_block	*cblock;/* child btree block */
547 	xfs_fsblock_t		cbno;	/* child block number */
548 	struct xfs_buf		*cbp;	/* child block's buffer */
549 	int			error;	/* error return value */
550 	__be64			*pp;	/* ptr to block address */
551 	struct xfs_owner_info	oinfo;
552 
553 	/* check if we actually need the extent format first: */
554 	if (!xfs_bmap_wants_extents(ip, whichfork))
555 		return 0;
556 
557 	ASSERT(cur);
558 	ASSERT(whichfork != XFS_COW_FORK);
559 	ASSERT(ifp->if_format == XFS_DINODE_FMT_BTREE);
560 	ASSERT(be16_to_cpu(rblock->bb_level) == 1);
561 	ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
562 	ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
563 
564 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
565 	cbno = be64_to_cpu(*pp);
566 #ifdef DEBUG
567 	if (XFS_IS_CORRUPT(cur->bc_mp, !xfs_btree_check_lptr(cur, cbno, 1)))
568 		return -EFSCORRUPTED;
569 #endif
570 	error = xfs_btree_read_bufl(mp, tp, cbno, &cbp, XFS_BMAP_BTREE_REF,
571 				&xfs_bmbt_buf_ops);
572 	if (error)
573 		return error;
574 	cblock = XFS_BUF_TO_BLOCK(cbp);
575 	if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
576 		return error;
577 	xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
578 	xfs_free_extent_later(cur->bc_tp, cbno, 1, &oinfo);
579 	ip->i_nblocks--;
580 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
581 	xfs_trans_binval(tp, cbp);
582 	if (cur->bc_levels[0].bp == cbp)
583 		cur->bc_levels[0].bp = NULL;
584 	xfs_iroot_realloc(ip, -1, whichfork);
585 	ASSERT(ifp->if_broot == NULL);
586 	ifp->if_format = XFS_DINODE_FMT_EXTENTS;
587 	*logflagsp |= XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
588 	return 0;
589 }
590 
591 /*
592  * Convert an extents-format file into a btree-format file.
593  * The new file will have a root block (in the inode) and a single child block.
594  */
595 STATIC int					/* error */
596 xfs_bmap_extents_to_btree(
597 	struct xfs_trans	*tp,		/* transaction pointer */
598 	struct xfs_inode	*ip,		/* incore inode pointer */
599 	struct xfs_btree_cur	**curp,		/* cursor returned to caller */
600 	int			wasdel,		/* converting a delayed alloc */
601 	int			*logflagsp,	/* inode logging flags */
602 	int			whichfork)	/* data or attr fork */
603 {
604 	struct xfs_btree_block	*ablock;	/* allocated (child) bt block */
605 	struct xfs_buf		*abp;		/* buffer for ablock */
606 	struct xfs_alloc_arg	args;		/* allocation arguments */
607 	struct xfs_bmbt_rec	*arp;		/* child record pointer */
608 	struct xfs_btree_block	*block;		/* btree root block */
609 	struct xfs_btree_cur	*cur;		/* bmap btree cursor */
610 	int			error;		/* error return value */
611 	struct xfs_ifork	*ifp;		/* inode fork pointer */
612 	struct xfs_bmbt_key	*kp;		/* root block key pointer */
613 	struct xfs_mount	*mp;		/* mount structure */
614 	xfs_bmbt_ptr_t		*pp;		/* root block address pointer */
615 	struct xfs_iext_cursor	icur;
616 	struct xfs_bmbt_irec	rec;
617 	xfs_extnum_t		cnt = 0;
618 
619 	mp = ip->i_mount;
620 	ASSERT(whichfork != XFS_COW_FORK);
621 	ifp = XFS_IFORK_PTR(ip, whichfork);
622 	ASSERT(ifp->if_format == XFS_DINODE_FMT_EXTENTS);
623 
624 	/*
625 	 * Make space in the inode incore. This needs to be undone if we fail
626 	 * to expand the root.
627 	 */
628 	xfs_iroot_realloc(ip, 1, whichfork);
629 
630 	/*
631 	 * Fill in the root.
632 	 */
633 	block = ifp->if_broot;
634 	xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
635 				 XFS_BTNUM_BMAP, 1, 1, ip->i_ino,
636 				 XFS_BTREE_LONG_PTRS);
637 	/*
638 	 * Need a cursor.  Can't allocate until bb_level is filled in.
639 	 */
640 	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
641 	cur->bc_ino.flags = wasdel ? XFS_BTCUR_BMBT_WASDEL : 0;
642 	/*
643 	 * Convert to a btree with two levels, one record in root.
644 	 */
645 	ifp->if_format = XFS_DINODE_FMT_BTREE;
646 	memset(&args, 0, sizeof(args));
647 	args.tp = tp;
648 	args.mp = mp;
649 	xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork);
650 	if (tp->t_firstblock == NULLFSBLOCK) {
651 		args.type = XFS_ALLOCTYPE_START_BNO;
652 		args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
653 	} else if (tp->t_flags & XFS_TRANS_LOWMODE) {
654 		args.type = XFS_ALLOCTYPE_START_BNO;
655 		args.fsbno = tp->t_firstblock;
656 	} else {
657 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
658 		args.fsbno = tp->t_firstblock;
659 	}
660 	args.minlen = args.maxlen = args.prod = 1;
661 	args.wasdel = wasdel;
662 	*logflagsp = 0;
663 	error = xfs_alloc_vextent(&args);
664 	if (error)
665 		goto out_root_realloc;
666 
667 	if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
668 		error = -ENOSPC;
669 		goto out_root_realloc;
670 	}
671 
672 	/*
673 	 * Allocation can't fail, the space was reserved.
674 	 */
675 	ASSERT(tp->t_firstblock == NULLFSBLOCK ||
676 	       args.agno >= XFS_FSB_TO_AGNO(mp, tp->t_firstblock));
677 	tp->t_firstblock = args.fsbno;
678 	cur->bc_ino.allocated++;
679 	ip->i_nblocks++;
680 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
681 	error = xfs_trans_get_buf(tp, mp->m_ddev_targp,
682 			XFS_FSB_TO_DADDR(mp, args.fsbno),
683 			mp->m_bsize, 0, &abp);
684 	if (error)
685 		goto out_unreserve_dquot;
686 
687 	/*
688 	 * Fill in the child block.
689 	 */
690 	abp->b_ops = &xfs_bmbt_buf_ops;
691 	ablock = XFS_BUF_TO_BLOCK(abp);
692 	xfs_btree_init_block_int(mp, ablock, xfs_buf_daddr(abp),
693 				XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
694 				XFS_BTREE_LONG_PTRS);
695 
696 	for_each_xfs_iext(ifp, &icur, &rec) {
697 		if (isnullstartblock(rec.br_startblock))
698 			continue;
699 		arp = XFS_BMBT_REC_ADDR(mp, ablock, 1 + cnt);
700 		xfs_bmbt_disk_set_all(arp, &rec);
701 		cnt++;
702 	}
703 	ASSERT(cnt == ifp->if_nextents);
704 	xfs_btree_set_numrecs(ablock, cnt);
705 
706 	/*
707 	 * Fill in the root key and pointer.
708 	 */
709 	kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
710 	arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
711 	kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
712 	pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
713 						be16_to_cpu(block->bb_level)));
714 	*pp = cpu_to_be64(args.fsbno);
715 
716 	/*
717 	 * Do all this logging at the end so that
718 	 * the root is at the right level.
719 	 */
720 	xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
721 	xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
722 	ASSERT(*curp == NULL);
723 	*curp = cur;
724 	*logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
725 	return 0;
726 
727 out_unreserve_dquot:
728 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
729 out_root_realloc:
730 	xfs_iroot_realloc(ip, -1, whichfork);
731 	ifp->if_format = XFS_DINODE_FMT_EXTENTS;
732 	ASSERT(ifp->if_broot == NULL);
733 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
734 
735 	return error;
736 }
737 
738 /*
739  * Convert a local file to an extents file.
740  * This code is out of bounds for data forks of regular files,
741  * since the file data needs to get logged so things will stay consistent.
742  * (The bmap-level manipulations are ok, though).
743  */
744 void
745 xfs_bmap_local_to_extents_empty(
746 	struct xfs_trans	*tp,
747 	struct xfs_inode	*ip,
748 	int			whichfork)
749 {
750 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
751 
752 	ASSERT(whichfork != XFS_COW_FORK);
753 	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
754 	ASSERT(ifp->if_bytes == 0);
755 	ASSERT(ifp->if_nextents == 0);
756 
757 	xfs_bmap_forkoff_reset(ip, whichfork);
758 	ifp->if_u1.if_root = NULL;
759 	ifp->if_height = 0;
760 	ifp->if_format = XFS_DINODE_FMT_EXTENTS;
761 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
762 }
763 
764 
765 STATIC int				/* error */
766 xfs_bmap_local_to_extents(
767 	xfs_trans_t	*tp,		/* transaction pointer */
768 	xfs_inode_t	*ip,		/* incore inode pointer */
769 	xfs_extlen_t	total,		/* total blocks needed by transaction */
770 	int		*logflagsp,	/* inode logging flags */
771 	int		whichfork,
772 	void		(*init_fn)(struct xfs_trans *tp,
773 				   struct xfs_buf *bp,
774 				   struct xfs_inode *ip,
775 				   struct xfs_ifork *ifp))
776 {
777 	int		error = 0;
778 	int		flags;		/* logging flags returned */
779 	struct xfs_ifork *ifp;		/* inode fork pointer */
780 	xfs_alloc_arg_t	args;		/* allocation arguments */
781 	struct xfs_buf	*bp;		/* buffer for extent block */
782 	struct xfs_bmbt_irec rec;
783 	struct xfs_iext_cursor icur;
784 
785 	/*
786 	 * We don't want to deal with the case of keeping inode data inline yet.
787 	 * So sending the data fork of a regular inode is invalid.
788 	 */
789 	ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
790 	ifp = XFS_IFORK_PTR(ip, whichfork);
791 	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
792 
793 	if (!ifp->if_bytes) {
794 		xfs_bmap_local_to_extents_empty(tp, ip, whichfork);
795 		flags = XFS_ILOG_CORE;
796 		goto done;
797 	}
798 
799 	flags = 0;
800 	error = 0;
801 	memset(&args, 0, sizeof(args));
802 	args.tp = tp;
803 	args.mp = ip->i_mount;
804 	xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0);
805 	/*
806 	 * Allocate a block.  We know we need only one, since the
807 	 * file currently fits in an inode.
808 	 */
809 	if (tp->t_firstblock == NULLFSBLOCK) {
810 		args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
811 		args.type = XFS_ALLOCTYPE_START_BNO;
812 	} else {
813 		args.fsbno = tp->t_firstblock;
814 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
815 	}
816 	args.total = total;
817 	args.minlen = args.maxlen = args.prod = 1;
818 	error = xfs_alloc_vextent(&args);
819 	if (error)
820 		goto done;
821 
822 	/* Can't fail, the space was reserved. */
823 	ASSERT(args.fsbno != NULLFSBLOCK);
824 	ASSERT(args.len == 1);
825 	tp->t_firstblock = args.fsbno;
826 	error = xfs_trans_get_buf(tp, args.mp->m_ddev_targp,
827 			XFS_FSB_TO_DADDR(args.mp, args.fsbno),
828 			args.mp->m_bsize, 0, &bp);
829 	if (error)
830 		goto done;
831 
832 	/*
833 	 * Initialize the block, copy the data and log the remote buffer.
834 	 *
835 	 * The callout is responsible for logging because the remote format
836 	 * might differ from the local format and thus we don't know how much to
837 	 * log here. Note that init_fn must also set the buffer log item type
838 	 * correctly.
839 	 */
840 	init_fn(tp, bp, ip, ifp);
841 
842 	/* account for the change in fork size */
843 	xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
844 	xfs_bmap_local_to_extents_empty(tp, ip, whichfork);
845 	flags |= XFS_ILOG_CORE;
846 
847 	ifp->if_u1.if_root = NULL;
848 	ifp->if_height = 0;
849 
850 	rec.br_startoff = 0;
851 	rec.br_startblock = args.fsbno;
852 	rec.br_blockcount = 1;
853 	rec.br_state = XFS_EXT_NORM;
854 	xfs_iext_first(ifp, &icur);
855 	xfs_iext_insert(ip, &icur, &rec, 0);
856 
857 	ifp->if_nextents = 1;
858 	ip->i_nblocks = 1;
859 	xfs_trans_mod_dquot_byino(tp, ip,
860 		XFS_TRANS_DQ_BCOUNT, 1L);
861 	flags |= xfs_ilog_fext(whichfork);
862 
863 done:
864 	*logflagsp = flags;
865 	return error;
866 }
867 
868 /*
869  * Called from xfs_bmap_add_attrfork to handle btree format files.
870  */
871 STATIC int					/* error */
872 xfs_bmap_add_attrfork_btree(
873 	xfs_trans_t		*tp,		/* transaction pointer */
874 	xfs_inode_t		*ip,		/* incore inode pointer */
875 	int			*flags)		/* inode logging flags */
876 {
877 	struct xfs_btree_block	*block = ip->i_df.if_broot;
878 	struct xfs_btree_cur	*cur;		/* btree cursor */
879 	int			error;		/* error return value */
880 	xfs_mount_t		*mp;		/* file system mount struct */
881 	int			stat;		/* newroot status */
882 
883 	mp = ip->i_mount;
884 
885 	if (XFS_BMAP_BMDR_SPACE(block) <= XFS_IFORK_DSIZE(ip))
886 		*flags |= XFS_ILOG_DBROOT;
887 	else {
888 		cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
889 		error = xfs_bmbt_lookup_first(cur, &stat);
890 		if (error)
891 			goto error0;
892 		/* must be at least one entry */
893 		if (XFS_IS_CORRUPT(mp, stat != 1)) {
894 			error = -EFSCORRUPTED;
895 			goto error0;
896 		}
897 		if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
898 			goto error0;
899 		if (stat == 0) {
900 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
901 			return -ENOSPC;
902 		}
903 		cur->bc_ino.allocated = 0;
904 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
905 	}
906 	return 0;
907 error0:
908 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
909 	return error;
910 }
911 
912 /*
913  * Called from xfs_bmap_add_attrfork to handle extents format files.
914  */
915 STATIC int					/* error */
916 xfs_bmap_add_attrfork_extents(
917 	struct xfs_trans	*tp,		/* transaction pointer */
918 	struct xfs_inode	*ip,		/* incore inode pointer */
919 	int			*flags)		/* inode logging flags */
920 {
921 	struct xfs_btree_cur	*cur;		/* bmap btree cursor */
922 	int			error;		/* error return value */
923 
924 	if (ip->i_df.if_nextents * sizeof(struct xfs_bmbt_rec) <=
925 	    XFS_IFORK_DSIZE(ip))
926 		return 0;
927 	cur = NULL;
928 	error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, flags,
929 					  XFS_DATA_FORK);
930 	if (cur) {
931 		cur->bc_ino.allocated = 0;
932 		xfs_btree_del_cursor(cur, error);
933 	}
934 	return error;
935 }
936 
937 /*
938  * Called from xfs_bmap_add_attrfork to handle local format files. Each
939  * different data fork content type needs a different callout to do the
940  * conversion. Some are basic and only require special block initialisation
941  * callouts for the data formating, others (directories) are so specialised they
942  * handle everything themselves.
943  *
944  * XXX (dgc): investigate whether directory conversion can use the generic
945  * formatting callout. It should be possible - it's just a very complex
946  * formatter.
947  */
948 STATIC int					/* error */
949 xfs_bmap_add_attrfork_local(
950 	struct xfs_trans	*tp,		/* transaction pointer */
951 	struct xfs_inode	*ip,		/* incore inode pointer */
952 	int			*flags)		/* inode logging flags */
953 {
954 	struct xfs_da_args	dargs;		/* args for dir/attr code */
955 
956 	if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
957 		return 0;
958 
959 	if (S_ISDIR(VFS_I(ip)->i_mode)) {
960 		memset(&dargs, 0, sizeof(dargs));
961 		dargs.geo = ip->i_mount->m_dir_geo;
962 		dargs.dp = ip;
963 		dargs.total = dargs.geo->fsbcount;
964 		dargs.whichfork = XFS_DATA_FORK;
965 		dargs.trans = tp;
966 		return xfs_dir2_sf_to_block(&dargs);
967 	}
968 
969 	if (S_ISLNK(VFS_I(ip)->i_mode))
970 		return xfs_bmap_local_to_extents(tp, ip, 1, flags,
971 						 XFS_DATA_FORK,
972 						 xfs_symlink_local_to_remote);
973 
974 	/* should only be called for types that support local format data */
975 	ASSERT(0);
976 	return -EFSCORRUPTED;
977 }
978 
979 /*
980  * Set an inode attr fork offset based on the format of the data fork.
981  */
982 static int
983 xfs_bmap_set_attrforkoff(
984 	struct xfs_inode	*ip,
985 	int			size,
986 	int			*version)
987 {
988 	int			default_size = xfs_default_attroffset(ip) >> 3;
989 
990 	switch (ip->i_df.if_format) {
991 	case XFS_DINODE_FMT_DEV:
992 		ip->i_forkoff = default_size;
993 		break;
994 	case XFS_DINODE_FMT_LOCAL:
995 	case XFS_DINODE_FMT_EXTENTS:
996 	case XFS_DINODE_FMT_BTREE:
997 		ip->i_forkoff = xfs_attr_shortform_bytesfit(ip, size);
998 		if (!ip->i_forkoff)
999 			ip->i_forkoff = default_size;
1000 		else if (xfs_has_attr2(ip->i_mount) && version)
1001 			*version = 2;
1002 		break;
1003 	default:
1004 		ASSERT(0);
1005 		return -EINVAL;
1006 	}
1007 
1008 	return 0;
1009 }
1010 
1011 /*
1012  * Convert inode from non-attributed to attributed.
1013  * Must not be in a transaction, ip must not be locked.
1014  */
1015 int						/* error code */
1016 xfs_bmap_add_attrfork(
1017 	xfs_inode_t		*ip,		/* incore inode pointer */
1018 	int			size,		/* space new attribute needs */
1019 	int			rsvd)		/* xact may use reserved blks */
1020 {
1021 	xfs_mount_t		*mp;		/* mount structure */
1022 	xfs_trans_t		*tp;		/* transaction pointer */
1023 	int			blks;		/* space reservation */
1024 	int			version = 1;	/* superblock attr version */
1025 	int			logflags;	/* logging flags */
1026 	int			error;		/* error return value */
1027 
1028 	ASSERT(XFS_IFORK_Q(ip) == 0);
1029 
1030 	mp = ip->i_mount;
1031 	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1032 
1033 	blks = XFS_ADDAFORK_SPACE_RES(mp);
1034 
1035 	error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_addafork, blks, 0,
1036 			rsvd, &tp);
1037 	if (error)
1038 		return error;
1039 	if (XFS_IFORK_Q(ip))
1040 		goto trans_cancel;
1041 
1042 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1043 	error = xfs_bmap_set_attrforkoff(ip, size, &version);
1044 	if (error)
1045 		goto trans_cancel;
1046 	ASSERT(ip->i_afp == NULL);
1047 
1048 	ip->i_afp = xfs_ifork_alloc(XFS_DINODE_FMT_EXTENTS, 0);
1049 	logflags = 0;
1050 	switch (ip->i_df.if_format) {
1051 	case XFS_DINODE_FMT_LOCAL:
1052 		error = xfs_bmap_add_attrfork_local(tp, ip, &logflags);
1053 		break;
1054 	case XFS_DINODE_FMT_EXTENTS:
1055 		error = xfs_bmap_add_attrfork_extents(tp, ip, &logflags);
1056 		break;
1057 	case XFS_DINODE_FMT_BTREE:
1058 		error = xfs_bmap_add_attrfork_btree(tp, ip, &logflags);
1059 		break;
1060 	default:
1061 		error = 0;
1062 		break;
1063 	}
1064 	if (logflags)
1065 		xfs_trans_log_inode(tp, ip, logflags);
1066 	if (error)
1067 		goto trans_cancel;
1068 	if (!xfs_has_attr(mp) ||
1069 	   (!xfs_has_attr2(mp) && version == 2)) {
1070 		bool log_sb = false;
1071 
1072 		spin_lock(&mp->m_sb_lock);
1073 		if (!xfs_has_attr(mp)) {
1074 			xfs_add_attr(mp);
1075 			log_sb = true;
1076 		}
1077 		if (!xfs_has_attr2(mp) && version == 2) {
1078 			xfs_add_attr2(mp);
1079 			log_sb = true;
1080 		}
1081 		spin_unlock(&mp->m_sb_lock);
1082 		if (log_sb)
1083 			xfs_log_sb(tp);
1084 	}
1085 
1086 	error = xfs_trans_commit(tp);
1087 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1088 	return error;
1089 
1090 trans_cancel:
1091 	xfs_trans_cancel(tp);
1092 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1093 	return error;
1094 }
1095 
1096 /*
1097  * Internal and external extent tree search functions.
1098  */
1099 
1100 struct xfs_iread_state {
1101 	struct xfs_iext_cursor	icur;
1102 	xfs_extnum_t		loaded;
1103 };
1104 
1105 /* Stuff every bmbt record from this block into the incore extent map. */
1106 static int
1107 xfs_iread_bmbt_block(
1108 	struct xfs_btree_cur	*cur,
1109 	int			level,
1110 	void			*priv)
1111 {
1112 	struct xfs_iread_state	*ir = priv;
1113 	struct xfs_mount	*mp = cur->bc_mp;
1114 	struct xfs_inode	*ip = cur->bc_ino.ip;
1115 	struct xfs_btree_block	*block;
1116 	struct xfs_buf		*bp;
1117 	struct xfs_bmbt_rec	*frp;
1118 	xfs_extnum_t		num_recs;
1119 	xfs_extnum_t		j;
1120 	int			whichfork = cur->bc_ino.whichfork;
1121 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1122 
1123 	block = xfs_btree_get_block(cur, level, &bp);
1124 
1125 	/* Abort if we find more records than nextents. */
1126 	num_recs = xfs_btree_get_numrecs(block);
1127 	if (unlikely(ir->loaded + num_recs > ifp->if_nextents)) {
1128 		xfs_warn(ip->i_mount, "corrupt dinode %llu, (btree extents).",
1129 				(unsigned long long)ip->i_ino);
1130 		xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, block,
1131 				sizeof(*block), __this_address);
1132 		return -EFSCORRUPTED;
1133 	}
1134 
1135 	/* Copy records into the incore cache. */
1136 	frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1137 	for (j = 0; j < num_recs; j++, frp++, ir->loaded++) {
1138 		struct xfs_bmbt_irec	new;
1139 		xfs_failaddr_t		fa;
1140 
1141 		xfs_bmbt_disk_get_all(frp, &new);
1142 		fa = xfs_bmap_validate_extent(ip, whichfork, &new);
1143 		if (fa) {
1144 			xfs_inode_verifier_error(ip, -EFSCORRUPTED,
1145 					"xfs_iread_extents(2)", frp,
1146 					sizeof(*frp), fa);
1147 			return -EFSCORRUPTED;
1148 		}
1149 		xfs_iext_insert(ip, &ir->icur, &new,
1150 				xfs_bmap_fork_to_state(whichfork));
1151 		trace_xfs_read_extent(ip, &ir->icur,
1152 				xfs_bmap_fork_to_state(whichfork), _THIS_IP_);
1153 		xfs_iext_next(ifp, &ir->icur);
1154 	}
1155 
1156 	return 0;
1157 }
1158 
1159 /*
1160  * Read in extents from a btree-format inode.
1161  */
1162 int
1163 xfs_iread_extents(
1164 	struct xfs_trans	*tp,
1165 	struct xfs_inode	*ip,
1166 	int			whichfork)
1167 {
1168 	struct xfs_iread_state	ir;
1169 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1170 	struct xfs_mount	*mp = ip->i_mount;
1171 	struct xfs_btree_cur	*cur;
1172 	int			error;
1173 
1174 	if (!xfs_need_iread_extents(ifp))
1175 		return 0;
1176 
1177 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1178 
1179 	ir.loaded = 0;
1180 	xfs_iext_first(ifp, &ir.icur);
1181 	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
1182 	error = xfs_btree_visit_blocks(cur, xfs_iread_bmbt_block,
1183 			XFS_BTREE_VISIT_RECORDS, &ir);
1184 	xfs_btree_del_cursor(cur, error);
1185 	if (error)
1186 		goto out;
1187 
1188 	if (XFS_IS_CORRUPT(mp, ir.loaded != ifp->if_nextents)) {
1189 		error = -EFSCORRUPTED;
1190 		goto out;
1191 	}
1192 	ASSERT(ir.loaded == xfs_iext_count(ifp));
1193 	return 0;
1194 out:
1195 	xfs_iext_destroy(ifp);
1196 	return error;
1197 }
1198 
1199 /*
1200  * Returns the relative block number of the first unused block(s) in the given
1201  * fork with at least "len" logically contiguous blocks free.  This is the
1202  * lowest-address hole if the fork has holes, else the first block past the end
1203  * of fork.  Return 0 if the fork is currently local (in-inode).
1204  */
1205 int						/* error */
1206 xfs_bmap_first_unused(
1207 	struct xfs_trans	*tp,		/* transaction pointer */
1208 	struct xfs_inode	*ip,		/* incore inode */
1209 	xfs_extlen_t		len,		/* size of hole to find */
1210 	xfs_fileoff_t		*first_unused,	/* unused block */
1211 	int			whichfork)	/* data or attr fork */
1212 {
1213 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1214 	struct xfs_bmbt_irec	got;
1215 	struct xfs_iext_cursor	icur;
1216 	xfs_fileoff_t		lastaddr = 0;
1217 	xfs_fileoff_t		lowest, max;
1218 	int			error;
1219 
1220 	if (ifp->if_format == XFS_DINODE_FMT_LOCAL) {
1221 		*first_unused = 0;
1222 		return 0;
1223 	}
1224 
1225 	ASSERT(xfs_ifork_has_extents(ifp));
1226 
1227 	error = xfs_iread_extents(tp, ip, whichfork);
1228 	if (error)
1229 		return error;
1230 
1231 	lowest = max = *first_unused;
1232 	for_each_xfs_iext(ifp, &icur, &got) {
1233 		/*
1234 		 * See if the hole before this extent will work.
1235 		 */
1236 		if (got.br_startoff >= lowest + len &&
1237 		    got.br_startoff - max >= len)
1238 			break;
1239 		lastaddr = got.br_startoff + got.br_blockcount;
1240 		max = XFS_FILEOFF_MAX(lastaddr, lowest);
1241 	}
1242 
1243 	*first_unused = max;
1244 	return 0;
1245 }
1246 
1247 /*
1248  * Returns the file-relative block number of the last block - 1 before
1249  * last_block (input value) in the file.
1250  * This is not based on i_size, it is based on the extent records.
1251  * Returns 0 for local files, as they do not have extent records.
1252  */
1253 int						/* error */
1254 xfs_bmap_last_before(
1255 	struct xfs_trans	*tp,		/* transaction pointer */
1256 	struct xfs_inode	*ip,		/* incore inode */
1257 	xfs_fileoff_t		*last_block,	/* last block */
1258 	int			whichfork)	/* data or attr fork */
1259 {
1260 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1261 	struct xfs_bmbt_irec	got;
1262 	struct xfs_iext_cursor	icur;
1263 	int			error;
1264 
1265 	switch (ifp->if_format) {
1266 	case XFS_DINODE_FMT_LOCAL:
1267 		*last_block = 0;
1268 		return 0;
1269 	case XFS_DINODE_FMT_BTREE:
1270 	case XFS_DINODE_FMT_EXTENTS:
1271 		break;
1272 	default:
1273 		ASSERT(0);
1274 		return -EFSCORRUPTED;
1275 	}
1276 
1277 	error = xfs_iread_extents(tp, ip, whichfork);
1278 	if (error)
1279 		return error;
1280 
1281 	if (!xfs_iext_lookup_extent_before(ip, ifp, last_block, &icur, &got))
1282 		*last_block = 0;
1283 	return 0;
1284 }
1285 
1286 int
1287 xfs_bmap_last_extent(
1288 	struct xfs_trans	*tp,
1289 	struct xfs_inode	*ip,
1290 	int			whichfork,
1291 	struct xfs_bmbt_irec	*rec,
1292 	int			*is_empty)
1293 {
1294 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1295 	struct xfs_iext_cursor	icur;
1296 	int			error;
1297 
1298 	error = xfs_iread_extents(tp, ip, whichfork);
1299 	if (error)
1300 		return error;
1301 
1302 	xfs_iext_last(ifp, &icur);
1303 	if (!xfs_iext_get_extent(ifp, &icur, rec))
1304 		*is_empty = 1;
1305 	else
1306 		*is_empty = 0;
1307 	return 0;
1308 }
1309 
1310 /*
1311  * Check the last inode extent to determine whether this allocation will result
1312  * in blocks being allocated at the end of the file. When we allocate new data
1313  * blocks at the end of the file which do not start at the previous data block,
1314  * we will try to align the new blocks at stripe unit boundaries.
1315  *
1316  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1317  * at, or past the EOF.
1318  */
1319 STATIC int
1320 xfs_bmap_isaeof(
1321 	struct xfs_bmalloca	*bma,
1322 	int			whichfork)
1323 {
1324 	struct xfs_bmbt_irec	rec;
1325 	int			is_empty;
1326 	int			error;
1327 
1328 	bma->aeof = false;
1329 	error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1330 				     &is_empty);
1331 	if (error)
1332 		return error;
1333 
1334 	if (is_empty) {
1335 		bma->aeof = true;
1336 		return 0;
1337 	}
1338 
1339 	/*
1340 	 * Check if we are allocation or past the last extent, or at least into
1341 	 * the last delayed allocated extent.
1342 	 */
1343 	bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1344 		(bma->offset >= rec.br_startoff &&
1345 		 isnullstartblock(rec.br_startblock));
1346 	return 0;
1347 }
1348 
1349 /*
1350  * Returns the file-relative block number of the first block past eof in
1351  * the file.  This is not based on i_size, it is based on the extent records.
1352  * Returns 0 for local files, as they do not have extent records.
1353  */
1354 int
1355 xfs_bmap_last_offset(
1356 	struct xfs_inode	*ip,
1357 	xfs_fileoff_t		*last_block,
1358 	int			whichfork)
1359 {
1360 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1361 	struct xfs_bmbt_irec	rec;
1362 	int			is_empty;
1363 	int			error;
1364 
1365 	*last_block = 0;
1366 
1367 	if (ifp->if_format == XFS_DINODE_FMT_LOCAL)
1368 		return 0;
1369 
1370 	if (XFS_IS_CORRUPT(ip->i_mount, !xfs_ifork_has_extents(ifp)))
1371 		return -EFSCORRUPTED;
1372 
1373 	error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1374 	if (error || is_empty)
1375 		return error;
1376 
1377 	*last_block = rec.br_startoff + rec.br_blockcount;
1378 	return 0;
1379 }
1380 
1381 /*
1382  * Extent tree manipulation functions used during allocation.
1383  */
1384 
1385 /*
1386  * Convert a delayed allocation to a real allocation.
1387  */
1388 STATIC int				/* error */
1389 xfs_bmap_add_extent_delay_real(
1390 	struct xfs_bmalloca	*bma,
1391 	int			whichfork)
1392 {
1393 	struct xfs_mount	*mp = bma->ip->i_mount;
1394 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1395 	struct xfs_bmbt_irec	*new = &bma->got;
1396 	int			error;	/* error return value */
1397 	int			i;	/* temp state */
1398 	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
1399 	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
1400 					/* left is 0, right is 1, prev is 2 */
1401 	int			rval=0;	/* return value (logging flags) */
1402 	int			state = xfs_bmap_fork_to_state(whichfork);
1403 	xfs_filblks_t		da_new; /* new count del alloc blocks used */
1404 	xfs_filblks_t		da_old; /* old count del alloc blocks used */
1405 	xfs_filblks_t		temp=0;	/* value for da_new calculations */
1406 	int			tmp_rval;	/* partial logging flags */
1407 	struct xfs_bmbt_irec	old;
1408 
1409 	ASSERT(whichfork != XFS_ATTR_FORK);
1410 	ASSERT(!isnullstartblock(new->br_startblock));
1411 	ASSERT(!bma->cur ||
1412 	       (bma->cur->bc_ino.flags & XFS_BTCUR_BMBT_WASDEL));
1413 
1414 	XFS_STATS_INC(mp, xs_add_exlist);
1415 
1416 #define	LEFT		r[0]
1417 #define	RIGHT		r[1]
1418 #define	PREV		r[2]
1419 
1420 	/*
1421 	 * Set up a bunch of variables to make the tests simpler.
1422 	 */
1423 	xfs_iext_get_extent(ifp, &bma->icur, &PREV);
1424 	new_endoff = new->br_startoff + new->br_blockcount;
1425 	ASSERT(isnullstartblock(PREV.br_startblock));
1426 	ASSERT(PREV.br_startoff <= new->br_startoff);
1427 	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1428 
1429 	da_old = startblockval(PREV.br_startblock);
1430 	da_new = 0;
1431 
1432 	/*
1433 	 * Set flags determining what part of the previous delayed allocation
1434 	 * extent is being replaced by a real allocation.
1435 	 */
1436 	if (PREV.br_startoff == new->br_startoff)
1437 		state |= BMAP_LEFT_FILLING;
1438 	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1439 		state |= BMAP_RIGHT_FILLING;
1440 
1441 	/*
1442 	 * Check and set flags if this segment has a left neighbor.
1443 	 * Don't set contiguous if the combined extent would be too large.
1444 	 */
1445 	if (xfs_iext_peek_prev_extent(ifp, &bma->icur, &LEFT)) {
1446 		state |= BMAP_LEFT_VALID;
1447 		if (isnullstartblock(LEFT.br_startblock))
1448 			state |= BMAP_LEFT_DELAY;
1449 	}
1450 
1451 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1452 	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1453 	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1454 	    LEFT.br_state == new->br_state &&
1455 	    LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1456 		state |= BMAP_LEFT_CONTIG;
1457 
1458 	/*
1459 	 * Check and set flags if this segment has a right neighbor.
1460 	 * Don't set contiguous if the combined extent would be too large.
1461 	 * Also check for all-three-contiguous being too large.
1462 	 */
1463 	if (xfs_iext_peek_next_extent(ifp, &bma->icur, &RIGHT)) {
1464 		state |= BMAP_RIGHT_VALID;
1465 		if (isnullstartblock(RIGHT.br_startblock))
1466 			state |= BMAP_RIGHT_DELAY;
1467 	}
1468 
1469 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1470 	    new_endoff == RIGHT.br_startoff &&
1471 	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1472 	    new->br_state == RIGHT.br_state &&
1473 	    new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1474 	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1475 		       BMAP_RIGHT_FILLING)) !=
1476 		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1477 		       BMAP_RIGHT_FILLING) ||
1478 	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1479 			<= MAXEXTLEN))
1480 		state |= BMAP_RIGHT_CONTIG;
1481 
1482 	error = 0;
1483 	/*
1484 	 * Switch out based on the FILLING and CONTIG state bits.
1485 	 */
1486 	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1487 			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1488 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1489 	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1490 		/*
1491 		 * Filling in all of a previously delayed allocation extent.
1492 		 * The left and right neighbors are both contiguous with new.
1493 		 */
1494 		LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
1495 
1496 		xfs_iext_remove(bma->ip, &bma->icur, state);
1497 		xfs_iext_remove(bma->ip, &bma->icur, state);
1498 		xfs_iext_prev(ifp, &bma->icur);
1499 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1500 		ifp->if_nextents--;
1501 
1502 		if (bma->cur == NULL)
1503 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1504 		else {
1505 			rval = XFS_ILOG_CORE;
1506 			error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1507 			if (error)
1508 				goto done;
1509 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1510 				error = -EFSCORRUPTED;
1511 				goto done;
1512 			}
1513 			error = xfs_btree_delete(bma->cur, &i);
1514 			if (error)
1515 				goto done;
1516 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1517 				error = -EFSCORRUPTED;
1518 				goto done;
1519 			}
1520 			error = xfs_btree_decrement(bma->cur, 0, &i);
1521 			if (error)
1522 				goto done;
1523 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1524 				error = -EFSCORRUPTED;
1525 				goto done;
1526 			}
1527 			error = xfs_bmbt_update(bma->cur, &LEFT);
1528 			if (error)
1529 				goto done;
1530 		}
1531 		break;
1532 
1533 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1534 		/*
1535 		 * Filling in all of a previously delayed allocation extent.
1536 		 * The left neighbor is contiguous, the right is not.
1537 		 */
1538 		old = LEFT;
1539 		LEFT.br_blockcount += PREV.br_blockcount;
1540 
1541 		xfs_iext_remove(bma->ip, &bma->icur, state);
1542 		xfs_iext_prev(ifp, &bma->icur);
1543 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1544 
1545 		if (bma->cur == NULL)
1546 			rval = XFS_ILOG_DEXT;
1547 		else {
1548 			rval = 0;
1549 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1550 			if (error)
1551 				goto done;
1552 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1553 				error = -EFSCORRUPTED;
1554 				goto done;
1555 			}
1556 			error = xfs_bmbt_update(bma->cur, &LEFT);
1557 			if (error)
1558 				goto done;
1559 		}
1560 		break;
1561 
1562 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1563 		/*
1564 		 * Filling in all of a previously delayed allocation extent.
1565 		 * The right neighbor is contiguous, the left is not. Take care
1566 		 * with delay -> unwritten extent allocation here because the
1567 		 * delalloc record we are overwriting is always written.
1568 		 */
1569 		PREV.br_startblock = new->br_startblock;
1570 		PREV.br_blockcount += RIGHT.br_blockcount;
1571 		PREV.br_state = new->br_state;
1572 
1573 		xfs_iext_next(ifp, &bma->icur);
1574 		xfs_iext_remove(bma->ip, &bma->icur, state);
1575 		xfs_iext_prev(ifp, &bma->icur);
1576 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1577 
1578 		if (bma->cur == NULL)
1579 			rval = XFS_ILOG_DEXT;
1580 		else {
1581 			rval = 0;
1582 			error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1583 			if (error)
1584 				goto done;
1585 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1586 				error = -EFSCORRUPTED;
1587 				goto done;
1588 			}
1589 			error = xfs_bmbt_update(bma->cur, &PREV);
1590 			if (error)
1591 				goto done;
1592 		}
1593 		break;
1594 
1595 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1596 		/*
1597 		 * Filling in all of a previously delayed allocation extent.
1598 		 * Neither the left nor right neighbors are contiguous with
1599 		 * the new one.
1600 		 */
1601 		PREV.br_startblock = new->br_startblock;
1602 		PREV.br_state = new->br_state;
1603 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1604 		ifp->if_nextents++;
1605 
1606 		if (bma->cur == NULL)
1607 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1608 		else {
1609 			rval = XFS_ILOG_CORE;
1610 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1611 			if (error)
1612 				goto done;
1613 			if (XFS_IS_CORRUPT(mp, i != 0)) {
1614 				error = -EFSCORRUPTED;
1615 				goto done;
1616 			}
1617 			error = xfs_btree_insert(bma->cur, &i);
1618 			if (error)
1619 				goto done;
1620 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1621 				error = -EFSCORRUPTED;
1622 				goto done;
1623 			}
1624 		}
1625 		break;
1626 
1627 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1628 		/*
1629 		 * Filling in the first part of a previous delayed allocation.
1630 		 * The left neighbor is contiguous.
1631 		 */
1632 		old = LEFT;
1633 		temp = PREV.br_blockcount - new->br_blockcount;
1634 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1635 				startblockval(PREV.br_startblock));
1636 
1637 		LEFT.br_blockcount += new->br_blockcount;
1638 
1639 		PREV.br_blockcount = temp;
1640 		PREV.br_startoff += new->br_blockcount;
1641 		PREV.br_startblock = nullstartblock(da_new);
1642 
1643 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1644 		xfs_iext_prev(ifp, &bma->icur);
1645 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1646 
1647 		if (bma->cur == NULL)
1648 			rval = XFS_ILOG_DEXT;
1649 		else {
1650 			rval = 0;
1651 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1652 			if (error)
1653 				goto done;
1654 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1655 				error = -EFSCORRUPTED;
1656 				goto done;
1657 			}
1658 			error = xfs_bmbt_update(bma->cur, &LEFT);
1659 			if (error)
1660 				goto done;
1661 		}
1662 		break;
1663 
1664 	case BMAP_LEFT_FILLING:
1665 		/*
1666 		 * Filling in the first part of a previous delayed allocation.
1667 		 * The left neighbor is not contiguous.
1668 		 */
1669 		xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1670 		ifp->if_nextents++;
1671 
1672 		if (bma->cur == NULL)
1673 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1674 		else {
1675 			rval = XFS_ILOG_CORE;
1676 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1677 			if (error)
1678 				goto done;
1679 			if (XFS_IS_CORRUPT(mp, i != 0)) {
1680 				error = -EFSCORRUPTED;
1681 				goto done;
1682 			}
1683 			error = xfs_btree_insert(bma->cur, &i);
1684 			if (error)
1685 				goto done;
1686 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1687 				error = -EFSCORRUPTED;
1688 				goto done;
1689 			}
1690 		}
1691 
1692 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1693 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1694 					&bma->cur, 1, &tmp_rval, whichfork);
1695 			rval |= tmp_rval;
1696 			if (error)
1697 				goto done;
1698 		}
1699 
1700 		temp = PREV.br_blockcount - new->br_blockcount;
1701 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1702 			startblockval(PREV.br_startblock) -
1703 			(bma->cur ? bma->cur->bc_ino.allocated : 0));
1704 
1705 		PREV.br_startoff = new_endoff;
1706 		PREV.br_blockcount = temp;
1707 		PREV.br_startblock = nullstartblock(da_new);
1708 		xfs_iext_next(ifp, &bma->icur);
1709 		xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1710 		xfs_iext_prev(ifp, &bma->icur);
1711 		break;
1712 
1713 	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1714 		/*
1715 		 * Filling in the last part of a previous delayed allocation.
1716 		 * The right neighbor is contiguous with the new allocation.
1717 		 */
1718 		old = RIGHT;
1719 		RIGHT.br_startoff = new->br_startoff;
1720 		RIGHT.br_startblock = new->br_startblock;
1721 		RIGHT.br_blockcount += new->br_blockcount;
1722 
1723 		if (bma->cur == NULL)
1724 			rval = XFS_ILOG_DEXT;
1725 		else {
1726 			rval = 0;
1727 			error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1728 			if (error)
1729 				goto done;
1730 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1731 				error = -EFSCORRUPTED;
1732 				goto done;
1733 			}
1734 			error = xfs_bmbt_update(bma->cur, &RIGHT);
1735 			if (error)
1736 				goto done;
1737 		}
1738 
1739 		temp = PREV.br_blockcount - new->br_blockcount;
1740 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1741 			startblockval(PREV.br_startblock));
1742 
1743 		PREV.br_blockcount = temp;
1744 		PREV.br_startblock = nullstartblock(da_new);
1745 
1746 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1747 		xfs_iext_next(ifp, &bma->icur);
1748 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &RIGHT);
1749 		break;
1750 
1751 	case BMAP_RIGHT_FILLING:
1752 		/*
1753 		 * Filling in the last part of a previous delayed allocation.
1754 		 * The right neighbor is not contiguous.
1755 		 */
1756 		xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1757 		ifp->if_nextents++;
1758 
1759 		if (bma->cur == NULL)
1760 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1761 		else {
1762 			rval = XFS_ILOG_CORE;
1763 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1764 			if (error)
1765 				goto done;
1766 			if (XFS_IS_CORRUPT(mp, i != 0)) {
1767 				error = -EFSCORRUPTED;
1768 				goto done;
1769 			}
1770 			error = xfs_btree_insert(bma->cur, &i);
1771 			if (error)
1772 				goto done;
1773 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1774 				error = -EFSCORRUPTED;
1775 				goto done;
1776 			}
1777 		}
1778 
1779 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1780 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1781 				&bma->cur, 1, &tmp_rval, whichfork);
1782 			rval |= tmp_rval;
1783 			if (error)
1784 				goto done;
1785 		}
1786 
1787 		temp = PREV.br_blockcount - new->br_blockcount;
1788 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1789 			startblockval(PREV.br_startblock) -
1790 			(bma->cur ? bma->cur->bc_ino.allocated : 0));
1791 
1792 		PREV.br_startblock = nullstartblock(da_new);
1793 		PREV.br_blockcount = temp;
1794 		xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1795 		xfs_iext_next(ifp, &bma->icur);
1796 		break;
1797 
1798 	case 0:
1799 		/*
1800 		 * Filling in the middle part of a previous delayed allocation.
1801 		 * Contiguity is impossible here.
1802 		 * This case is avoided almost all the time.
1803 		 *
1804 		 * We start with a delayed allocation:
1805 		 *
1806 		 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1807 		 *  PREV @ idx
1808 		 *
1809 	         * and we are allocating:
1810 		 *                     +rrrrrrrrrrrrrrrrr+
1811 		 *			      new
1812 		 *
1813 		 * and we set it up for insertion as:
1814 		 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1815 		 *                            new
1816 		 *  PREV @ idx          LEFT              RIGHT
1817 		 *                      inserted at idx + 1
1818 		 */
1819 		old = PREV;
1820 
1821 		/* LEFT is the new middle */
1822 		LEFT = *new;
1823 
1824 		/* RIGHT is the new right */
1825 		RIGHT.br_state = PREV.br_state;
1826 		RIGHT.br_startoff = new_endoff;
1827 		RIGHT.br_blockcount =
1828 			PREV.br_startoff + PREV.br_blockcount - new_endoff;
1829 		RIGHT.br_startblock =
1830 			nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1831 					RIGHT.br_blockcount));
1832 
1833 		/* truncate PREV */
1834 		PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
1835 		PREV.br_startblock =
1836 			nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1837 					PREV.br_blockcount));
1838 		xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1839 
1840 		xfs_iext_next(ifp, &bma->icur);
1841 		xfs_iext_insert(bma->ip, &bma->icur, &RIGHT, state);
1842 		xfs_iext_insert(bma->ip, &bma->icur, &LEFT, state);
1843 		ifp->if_nextents++;
1844 
1845 		if (bma->cur == NULL)
1846 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1847 		else {
1848 			rval = XFS_ILOG_CORE;
1849 			error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1850 			if (error)
1851 				goto done;
1852 			if (XFS_IS_CORRUPT(mp, i != 0)) {
1853 				error = -EFSCORRUPTED;
1854 				goto done;
1855 			}
1856 			error = xfs_btree_insert(bma->cur, &i);
1857 			if (error)
1858 				goto done;
1859 			if (XFS_IS_CORRUPT(mp, i != 1)) {
1860 				error = -EFSCORRUPTED;
1861 				goto done;
1862 			}
1863 		}
1864 
1865 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1866 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1867 					&bma->cur, 1, &tmp_rval, whichfork);
1868 			rval |= tmp_rval;
1869 			if (error)
1870 				goto done;
1871 		}
1872 
1873 		da_new = startblockval(PREV.br_startblock) +
1874 			 startblockval(RIGHT.br_startblock);
1875 		break;
1876 
1877 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1878 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1879 	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1880 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1881 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1882 	case BMAP_LEFT_CONTIG:
1883 	case BMAP_RIGHT_CONTIG:
1884 		/*
1885 		 * These cases are all impossible.
1886 		 */
1887 		ASSERT(0);
1888 	}
1889 
1890 	/* add reverse mapping unless caller opted out */
1891 	if (!(bma->flags & XFS_BMAPI_NORMAP))
1892 		xfs_rmap_map_extent(bma->tp, bma->ip, whichfork, new);
1893 
1894 	/* convert to a btree if necessary */
1895 	if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1896 		int	tmp_logflags;	/* partial log flag return val */
1897 
1898 		ASSERT(bma->cur == NULL);
1899 		error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1900 				&bma->cur, da_old > 0, &tmp_logflags,
1901 				whichfork);
1902 		bma->logflags |= tmp_logflags;
1903 		if (error)
1904 			goto done;
1905 	}
1906 
1907 	if (da_new != da_old)
1908 		xfs_mod_delalloc(mp, (int64_t)da_new - da_old);
1909 
1910 	if (bma->cur) {
1911 		da_new += bma->cur->bc_ino.allocated;
1912 		bma->cur->bc_ino.allocated = 0;
1913 	}
1914 
1915 	/* adjust for changes in reserved delayed indirect blocks */
1916 	if (da_new != da_old) {
1917 		ASSERT(state == 0 || da_new < da_old);
1918 		error = xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new),
1919 				false);
1920 	}
1921 
1922 	xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
1923 done:
1924 	if (whichfork != XFS_COW_FORK)
1925 		bma->logflags |= rval;
1926 	return error;
1927 #undef	LEFT
1928 #undef	RIGHT
1929 #undef	PREV
1930 }
1931 
1932 /*
1933  * Convert an unwritten allocation to a real allocation or vice versa.
1934  */
1935 int					/* error */
1936 xfs_bmap_add_extent_unwritten_real(
1937 	struct xfs_trans	*tp,
1938 	xfs_inode_t		*ip,	/* incore inode pointer */
1939 	int			whichfork,
1940 	struct xfs_iext_cursor	*icur,
1941 	struct xfs_btree_cur	**curp,	/* if *curp is null, not a btree */
1942 	xfs_bmbt_irec_t		*new,	/* new data to add to file extents */
1943 	int			*logflagsp) /* inode logging flags */
1944 {
1945 	struct xfs_btree_cur	*cur;	/* btree cursor */
1946 	int			error;	/* error return value */
1947 	int			i;	/* temp state */
1948 	struct xfs_ifork	*ifp;	/* inode fork pointer */
1949 	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
1950 	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
1951 					/* left is 0, right is 1, prev is 2 */
1952 	int			rval=0;	/* return value (logging flags) */
1953 	int			state = xfs_bmap_fork_to_state(whichfork);
1954 	struct xfs_mount	*mp = ip->i_mount;
1955 	struct xfs_bmbt_irec	old;
1956 
1957 	*logflagsp = 0;
1958 
1959 	cur = *curp;
1960 	ifp = XFS_IFORK_PTR(ip, whichfork);
1961 
1962 	ASSERT(!isnullstartblock(new->br_startblock));
1963 
1964 	XFS_STATS_INC(mp, xs_add_exlist);
1965 
1966 #define	LEFT		r[0]
1967 #define	RIGHT		r[1]
1968 #define	PREV		r[2]
1969 
1970 	/*
1971 	 * Set up a bunch of variables to make the tests simpler.
1972 	 */
1973 	error = 0;
1974 	xfs_iext_get_extent(ifp, icur, &PREV);
1975 	ASSERT(new->br_state != PREV.br_state);
1976 	new_endoff = new->br_startoff + new->br_blockcount;
1977 	ASSERT(PREV.br_startoff <= new->br_startoff);
1978 	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1979 
1980 	/*
1981 	 * Set flags determining what part of the previous oldext allocation
1982 	 * extent is being replaced by a newext allocation.
1983 	 */
1984 	if (PREV.br_startoff == new->br_startoff)
1985 		state |= BMAP_LEFT_FILLING;
1986 	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1987 		state |= BMAP_RIGHT_FILLING;
1988 
1989 	/*
1990 	 * Check and set flags if this segment has a left neighbor.
1991 	 * Don't set contiguous if the combined extent would be too large.
1992 	 */
1993 	if (xfs_iext_peek_prev_extent(ifp, icur, &LEFT)) {
1994 		state |= BMAP_LEFT_VALID;
1995 		if (isnullstartblock(LEFT.br_startblock))
1996 			state |= BMAP_LEFT_DELAY;
1997 	}
1998 
1999 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2000 	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2001 	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2002 	    LEFT.br_state == new->br_state &&
2003 	    LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2004 		state |= BMAP_LEFT_CONTIG;
2005 
2006 	/*
2007 	 * Check and set flags if this segment has a right neighbor.
2008 	 * Don't set contiguous if the combined extent would be too large.
2009 	 * Also check for all-three-contiguous being too large.
2010 	 */
2011 	if (xfs_iext_peek_next_extent(ifp, icur, &RIGHT)) {
2012 		state |= BMAP_RIGHT_VALID;
2013 		if (isnullstartblock(RIGHT.br_startblock))
2014 			state |= BMAP_RIGHT_DELAY;
2015 	}
2016 
2017 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2018 	    new_endoff == RIGHT.br_startoff &&
2019 	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2020 	    new->br_state == RIGHT.br_state &&
2021 	    new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2022 	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2023 		       BMAP_RIGHT_FILLING)) !=
2024 		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2025 		       BMAP_RIGHT_FILLING) ||
2026 	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2027 			<= MAXEXTLEN))
2028 		state |= BMAP_RIGHT_CONTIG;
2029 
2030 	/*
2031 	 * Switch out based on the FILLING and CONTIG state bits.
2032 	 */
2033 	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2034 			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2035 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2036 	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2037 		/*
2038 		 * Setting all of a previous oldext extent to newext.
2039 		 * The left and right neighbors are both contiguous with new.
2040 		 */
2041 		LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
2042 
2043 		xfs_iext_remove(ip, icur, state);
2044 		xfs_iext_remove(ip, icur, state);
2045 		xfs_iext_prev(ifp, icur);
2046 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2047 		ifp->if_nextents -= 2;
2048 		if (cur == NULL)
2049 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2050 		else {
2051 			rval = XFS_ILOG_CORE;
2052 			error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2053 			if (error)
2054 				goto done;
2055 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2056 				error = -EFSCORRUPTED;
2057 				goto done;
2058 			}
2059 			if ((error = xfs_btree_delete(cur, &i)))
2060 				goto done;
2061 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2062 				error = -EFSCORRUPTED;
2063 				goto done;
2064 			}
2065 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2066 				goto done;
2067 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2068 				error = -EFSCORRUPTED;
2069 				goto done;
2070 			}
2071 			if ((error = xfs_btree_delete(cur, &i)))
2072 				goto done;
2073 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2074 				error = -EFSCORRUPTED;
2075 				goto done;
2076 			}
2077 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2078 				goto done;
2079 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2080 				error = -EFSCORRUPTED;
2081 				goto done;
2082 			}
2083 			error = xfs_bmbt_update(cur, &LEFT);
2084 			if (error)
2085 				goto done;
2086 		}
2087 		break;
2088 
2089 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2090 		/*
2091 		 * Setting all of a previous oldext extent to newext.
2092 		 * The left neighbor is contiguous, the right is not.
2093 		 */
2094 		LEFT.br_blockcount += PREV.br_blockcount;
2095 
2096 		xfs_iext_remove(ip, icur, state);
2097 		xfs_iext_prev(ifp, icur);
2098 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2099 		ifp->if_nextents--;
2100 		if (cur == NULL)
2101 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2102 		else {
2103 			rval = XFS_ILOG_CORE;
2104 			error = xfs_bmbt_lookup_eq(cur, &PREV, &i);
2105 			if (error)
2106 				goto done;
2107 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2108 				error = -EFSCORRUPTED;
2109 				goto done;
2110 			}
2111 			if ((error = xfs_btree_delete(cur, &i)))
2112 				goto done;
2113 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2114 				error = -EFSCORRUPTED;
2115 				goto done;
2116 			}
2117 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2118 				goto done;
2119 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2120 				error = -EFSCORRUPTED;
2121 				goto done;
2122 			}
2123 			error = xfs_bmbt_update(cur, &LEFT);
2124 			if (error)
2125 				goto done;
2126 		}
2127 		break;
2128 
2129 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2130 		/*
2131 		 * Setting all of a previous oldext extent to newext.
2132 		 * The right neighbor is contiguous, the left is not.
2133 		 */
2134 		PREV.br_blockcount += RIGHT.br_blockcount;
2135 		PREV.br_state = new->br_state;
2136 
2137 		xfs_iext_next(ifp, icur);
2138 		xfs_iext_remove(ip, icur, state);
2139 		xfs_iext_prev(ifp, icur);
2140 		xfs_iext_update_extent(ip, state, icur, &PREV);
2141 		ifp->if_nextents--;
2142 
2143 		if (cur == NULL)
2144 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2145 		else {
2146 			rval = XFS_ILOG_CORE;
2147 			error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2148 			if (error)
2149 				goto done;
2150 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2151 				error = -EFSCORRUPTED;
2152 				goto done;
2153 			}
2154 			if ((error = xfs_btree_delete(cur, &i)))
2155 				goto done;
2156 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2157 				error = -EFSCORRUPTED;
2158 				goto done;
2159 			}
2160 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2161 				goto done;
2162 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2163 				error = -EFSCORRUPTED;
2164 				goto done;
2165 			}
2166 			error = xfs_bmbt_update(cur, &PREV);
2167 			if (error)
2168 				goto done;
2169 		}
2170 		break;
2171 
2172 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2173 		/*
2174 		 * Setting all of a previous oldext extent to newext.
2175 		 * Neither the left nor right neighbors are contiguous with
2176 		 * the new one.
2177 		 */
2178 		PREV.br_state = new->br_state;
2179 		xfs_iext_update_extent(ip, state, icur, &PREV);
2180 
2181 		if (cur == NULL)
2182 			rval = XFS_ILOG_DEXT;
2183 		else {
2184 			rval = 0;
2185 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2186 			if (error)
2187 				goto done;
2188 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2189 				error = -EFSCORRUPTED;
2190 				goto done;
2191 			}
2192 			error = xfs_bmbt_update(cur, &PREV);
2193 			if (error)
2194 				goto done;
2195 		}
2196 		break;
2197 
2198 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2199 		/*
2200 		 * Setting the first part of a previous oldext extent to newext.
2201 		 * The left neighbor is contiguous.
2202 		 */
2203 		LEFT.br_blockcount += new->br_blockcount;
2204 
2205 		old = PREV;
2206 		PREV.br_startoff += new->br_blockcount;
2207 		PREV.br_startblock += new->br_blockcount;
2208 		PREV.br_blockcount -= new->br_blockcount;
2209 
2210 		xfs_iext_update_extent(ip, state, icur, &PREV);
2211 		xfs_iext_prev(ifp, icur);
2212 		xfs_iext_update_extent(ip, state, icur, &LEFT);
2213 
2214 		if (cur == NULL)
2215 			rval = XFS_ILOG_DEXT;
2216 		else {
2217 			rval = 0;
2218 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2219 			if (error)
2220 				goto done;
2221 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2222 				error = -EFSCORRUPTED;
2223 				goto done;
2224 			}
2225 			error = xfs_bmbt_update(cur, &PREV);
2226 			if (error)
2227 				goto done;
2228 			error = xfs_btree_decrement(cur, 0, &i);
2229 			if (error)
2230 				goto done;
2231 			error = xfs_bmbt_update(cur, &LEFT);
2232 			if (error)
2233 				goto done;
2234 		}
2235 		break;
2236 
2237 	case BMAP_LEFT_FILLING:
2238 		/*
2239 		 * Setting the first part of a previous oldext extent to newext.
2240 		 * The left neighbor is not contiguous.
2241 		 */
2242 		old = PREV;
2243 		PREV.br_startoff += new->br_blockcount;
2244 		PREV.br_startblock += new->br_blockcount;
2245 		PREV.br_blockcount -= new->br_blockcount;
2246 
2247 		xfs_iext_update_extent(ip, state, icur, &PREV);
2248 		xfs_iext_insert(ip, icur, new, state);
2249 		ifp->if_nextents++;
2250 
2251 		if (cur == NULL)
2252 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2253 		else {
2254 			rval = XFS_ILOG_CORE;
2255 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2256 			if (error)
2257 				goto done;
2258 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2259 				error = -EFSCORRUPTED;
2260 				goto done;
2261 			}
2262 			error = xfs_bmbt_update(cur, &PREV);
2263 			if (error)
2264 				goto done;
2265 			cur->bc_rec.b = *new;
2266 			if ((error = xfs_btree_insert(cur, &i)))
2267 				goto done;
2268 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2269 				error = -EFSCORRUPTED;
2270 				goto done;
2271 			}
2272 		}
2273 		break;
2274 
2275 	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2276 		/*
2277 		 * Setting the last part of a previous oldext extent to newext.
2278 		 * The right neighbor is contiguous with the new allocation.
2279 		 */
2280 		old = PREV;
2281 		PREV.br_blockcount -= new->br_blockcount;
2282 
2283 		RIGHT.br_startoff = new->br_startoff;
2284 		RIGHT.br_startblock = new->br_startblock;
2285 		RIGHT.br_blockcount += new->br_blockcount;
2286 
2287 		xfs_iext_update_extent(ip, state, icur, &PREV);
2288 		xfs_iext_next(ifp, icur);
2289 		xfs_iext_update_extent(ip, state, icur, &RIGHT);
2290 
2291 		if (cur == NULL)
2292 			rval = XFS_ILOG_DEXT;
2293 		else {
2294 			rval = 0;
2295 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2296 			if (error)
2297 				goto done;
2298 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2299 				error = -EFSCORRUPTED;
2300 				goto done;
2301 			}
2302 			error = xfs_bmbt_update(cur, &PREV);
2303 			if (error)
2304 				goto done;
2305 			error = xfs_btree_increment(cur, 0, &i);
2306 			if (error)
2307 				goto done;
2308 			error = xfs_bmbt_update(cur, &RIGHT);
2309 			if (error)
2310 				goto done;
2311 		}
2312 		break;
2313 
2314 	case BMAP_RIGHT_FILLING:
2315 		/*
2316 		 * Setting the last part of a previous oldext extent to newext.
2317 		 * The right neighbor is not contiguous.
2318 		 */
2319 		old = PREV;
2320 		PREV.br_blockcount -= new->br_blockcount;
2321 
2322 		xfs_iext_update_extent(ip, state, icur, &PREV);
2323 		xfs_iext_next(ifp, icur);
2324 		xfs_iext_insert(ip, icur, new, state);
2325 		ifp->if_nextents++;
2326 
2327 		if (cur == NULL)
2328 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2329 		else {
2330 			rval = XFS_ILOG_CORE;
2331 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2332 			if (error)
2333 				goto done;
2334 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2335 				error = -EFSCORRUPTED;
2336 				goto done;
2337 			}
2338 			error = xfs_bmbt_update(cur, &PREV);
2339 			if (error)
2340 				goto done;
2341 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2342 			if (error)
2343 				goto done;
2344 			if (XFS_IS_CORRUPT(mp, i != 0)) {
2345 				error = -EFSCORRUPTED;
2346 				goto done;
2347 			}
2348 			if ((error = xfs_btree_insert(cur, &i)))
2349 				goto done;
2350 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2351 				error = -EFSCORRUPTED;
2352 				goto done;
2353 			}
2354 		}
2355 		break;
2356 
2357 	case 0:
2358 		/*
2359 		 * Setting the middle part of a previous oldext extent to
2360 		 * newext.  Contiguity is impossible here.
2361 		 * One extent becomes three extents.
2362 		 */
2363 		old = PREV;
2364 		PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
2365 
2366 		r[0] = *new;
2367 		r[1].br_startoff = new_endoff;
2368 		r[1].br_blockcount =
2369 			old.br_startoff + old.br_blockcount - new_endoff;
2370 		r[1].br_startblock = new->br_startblock + new->br_blockcount;
2371 		r[1].br_state = PREV.br_state;
2372 
2373 		xfs_iext_update_extent(ip, state, icur, &PREV);
2374 		xfs_iext_next(ifp, icur);
2375 		xfs_iext_insert(ip, icur, &r[1], state);
2376 		xfs_iext_insert(ip, icur, &r[0], state);
2377 		ifp->if_nextents += 2;
2378 
2379 		if (cur == NULL)
2380 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2381 		else {
2382 			rval = XFS_ILOG_CORE;
2383 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2384 			if (error)
2385 				goto done;
2386 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2387 				error = -EFSCORRUPTED;
2388 				goto done;
2389 			}
2390 			/* new right extent - oldext */
2391 			error = xfs_bmbt_update(cur, &r[1]);
2392 			if (error)
2393 				goto done;
2394 			/* new left extent - oldext */
2395 			cur->bc_rec.b = PREV;
2396 			if ((error = xfs_btree_insert(cur, &i)))
2397 				goto done;
2398 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2399 				error = -EFSCORRUPTED;
2400 				goto done;
2401 			}
2402 			/*
2403 			 * Reset the cursor to the position of the new extent
2404 			 * we are about to insert as we can't trust it after
2405 			 * the previous insert.
2406 			 */
2407 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2408 			if (error)
2409 				goto done;
2410 			if (XFS_IS_CORRUPT(mp, i != 0)) {
2411 				error = -EFSCORRUPTED;
2412 				goto done;
2413 			}
2414 			/* new middle extent - newext */
2415 			if ((error = xfs_btree_insert(cur, &i)))
2416 				goto done;
2417 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2418 				error = -EFSCORRUPTED;
2419 				goto done;
2420 			}
2421 		}
2422 		break;
2423 
2424 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2425 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2426 	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2427 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2428 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2429 	case BMAP_LEFT_CONTIG:
2430 	case BMAP_RIGHT_CONTIG:
2431 		/*
2432 		 * These cases are all impossible.
2433 		 */
2434 		ASSERT(0);
2435 	}
2436 
2437 	/* update reverse mappings */
2438 	xfs_rmap_convert_extent(mp, tp, ip, whichfork, new);
2439 
2440 	/* convert to a btree if necessary */
2441 	if (xfs_bmap_needs_btree(ip, whichfork)) {
2442 		int	tmp_logflags;	/* partial log flag return val */
2443 
2444 		ASSERT(cur == NULL);
2445 		error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
2446 				&tmp_logflags, whichfork);
2447 		*logflagsp |= tmp_logflags;
2448 		if (error)
2449 			goto done;
2450 	}
2451 
2452 	/* clear out the allocated field, done with it now in any case. */
2453 	if (cur) {
2454 		cur->bc_ino.allocated = 0;
2455 		*curp = cur;
2456 	}
2457 
2458 	xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
2459 done:
2460 	*logflagsp |= rval;
2461 	return error;
2462 #undef	LEFT
2463 #undef	RIGHT
2464 #undef	PREV
2465 }
2466 
2467 /*
2468  * Convert a hole to a delayed allocation.
2469  */
2470 STATIC void
2471 xfs_bmap_add_extent_hole_delay(
2472 	xfs_inode_t		*ip,	/* incore inode pointer */
2473 	int			whichfork,
2474 	struct xfs_iext_cursor	*icur,
2475 	xfs_bmbt_irec_t		*new)	/* new data to add to file extents */
2476 {
2477 	struct xfs_ifork	*ifp;	/* inode fork pointer */
2478 	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2479 	xfs_filblks_t		newlen=0;	/* new indirect size */
2480 	xfs_filblks_t		oldlen=0;	/* old indirect size */
2481 	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2482 	int			state = xfs_bmap_fork_to_state(whichfork);
2483 	xfs_filblks_t		temp;	 /* temp for indirect calculations */
2484 
2485 	ifp = XFS_IFORK_PTR(ip, whichfork);
2486 	ASSERT(isnullstartblock(new->br_startblock));
2487 
2488 	/*
2489 	 * Check and set flags if this segment has a left neighbor
2490 	 */
2491 	if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2492 		state |= BMAP_LEFT_VALID;
2493 		if (isnullstartblock(left.br_startblock))
2494 			state |= BMAP_LEFT_DELAY;
2495 	}
2496 
2497 	/*
2498 	 * Check and set flags if the current (right) segment exists.
2499 	 * If it doesn't exist, we're converting the hole at end-of-file.
2500 	 */
2501 	if (xfs_iext_get_extent(ifp, icur, &right)) {
2502 		state |= BMAP_RIGHT_VALID;
2503 		if (isnullstartblock(right.br_startblock))
2504 			state |= BMAP_RIGHT_DELAY;
2505 	}
2506 
2507 	/*
2508 	 * Set contiguity flags on the left and right neighbors.
2509 	 * Don't let extents get too large, even if the pieces are contiguous.
2510 	 */
2511 	if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2512 	    left.br_startoff + left.br_blockcount == new->br_startoff &&
2513 	    left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2514 		state |= BMAP_LEFT_CONTIG;
2515 
2516 	if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2517 	    new->br_startoff + new->br_blockcount == right.br_startoff &&
2518 	    new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2519 	    (!(state & BMAP_LEFT_CONTIG) ||
2520 	     (left.br_blockcount + new->br_blockcount +
2521 	      right.br_blockcount <= MAXEXTLEN)))
2522 		state |= BMAP_RIGHT_CONTIG;
2523 
2524 	/*
2525 	 * Switch out based on the contiguity flags.
2526 	 */
2527 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2528 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2529 		/*
2530 		 * New allocation is contiguous with delayed allocations
2531 		 * on the left and on the right.
2532 		 * Merge all three into a single extent record.
2533 		 */
2534 		temp = left.br_blockcount + new->br_blockcount +
2535 			right.br_blockcount;
2536 
2537 		oldlen = startblockval(left.br_startblock) +
2538 			startblockval(new->br_startblock) +
2539 			startblockval(right.br_startblock);
2540 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2541 					 oldlen);
2542 		left.br_startblock = nullstartblock(newlen);
2543 		left.br_blockcount = temp;
2544 
2545 		xfs_iext_remove(ip, icur, state);
2546 		xfs_iext_prev(ifp, icur);
2547 		xfs_iext_update_extent(ip, state, icur, &left);
2548 		break;
2549 
2550 	case BMAP_LEFT_CONTIG:
2551 		/*
2552 		 * New allocation is contiguous with a delayed allocation
2553 		 * on the left.
2554 		 * Merge the new allocation with the left neighbor.
2555 		 */
2556 		temp = left.br_blockcount + new->br_blockcount;
2557 
2558 		oldlen = startblockval(left.br_startblock) +
2559 			startblockval(new->br_startblock);
2560 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2561 					 oldlen);
2562 		left.br_blockcount = temp;
2563 		left.br_startblock = nullstartblock(newlen);
2564 
2565 		xfs_iext_prev(ifp, icur);
2566 		xfs_iext_update_extent(ip, state, icur, &left);
2567 		break;
2568 
2569 	case BMAP_RIGHT_CONTIG:
2570 		/*
2571 		 * New allocation is contiguous with a delayed allocation
2572 		 * on the right.
2573 		 * Merge the new allocation with the right neighbor.
2574 		 */
2575 		temp = new->br_blockcount + right.br_blockcount;
2576 		oldlen = startblockval(new->br_startblock) +
2577 			startblockval(right.br_startblock);
2578 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2579 					 oldlen);
2580 		right.br_startoff = new->br_startoff;
2581 		right.br_startblock = nullstartblock(newlen);
2582 		right.br_blockcount = temp;
2583 		xfs_iext_update_extent(ip, state, icur, &right);
2584 		break;
2585 
2586 	case 0:
2587 		/*
2588 		 * New allocation is not contiguous with another
2589 		 * delayed allocation.
2590 		 * Insert a new entry.
2591 		 */
2592 		oldlen = newlen = 0;
2593 		xfs_iext_insert(ip, icur, new, state);
2594 		break;
2595 	}
2596 	if (oldlen != newlen) {
2597 		ASSERT(oldlen > newlen);
2598 		xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2599 				 false);
2600 		/*
2601 		 * Nothing to do for disk quota accounting here.
2602 		 */
2603 		xfs_mod_delalloc(ip->i_mount, (int64_t)newlen - oldlen);
2604 	}
2605 }
2606 
2607 /*
2608  * Convert a hole to a real allocation.
2609  */
2610 STATIC int				/* error */
2611 xfs_bmap_add_extent_hole_real(
2612 	struct xfs_trans	*tp,
2613 	struct xfs_inode	*ip,
2614 	int			whichfork,
2615 	struct xfs_iext_cursor	*icur,
2616 	struct xfs_btree_cur	**curp,
2617 	struct xfs_bmbt_irec	*new,
2618 	int			*logflagsp,
2619 	int			flags)
2620 {
2621 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
2622 	struct xfs_mount	*mp = ip->i_mount;
2623 	struct xfs_btree_cur	*cur = *curp;
2624 	int			error;	/* error return value */
2625 	int			i;	/* temp state */
2626 	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2627 	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2628 	int			rval=0;	/* return value (logging flags) */
2629 	int			state = xfs_bmap_fork_to_state(whichfork);
2630 	struct xfs_bmbt_irec	old;
2631 
2632 	ASSERT(!isnullstartblock(new->br_startblock));
2633 	ASSERT(!cur || !(cur->bc_ino.flags & XFS_BTCUR_BMBT_WASDEL));
2634 
2635 	XFS_STATS_INC(mp, xs_add_exlist);
2636 
2637 	/*
2638 	 * Check and set flags if this segment has a left neighbor.
2639 	 */
2640 	if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2641 		state |= BMAP_LEFT_VALID;
2642 		if (isnullstartblock(left.br_startblock))
2643 			state |= BMAP_LEFT_DELAY;
2644 	}
2645 
2646 	/*
2647 	 * Check and set flags if this segment has a current value.
2648 	 * Not true if we're inserting into the "hole" at eof.
2649 	 */
2650 	if (xfs_iext_get_extent(ifp, icur, &right)) {
2651 		state |= BMAP_RIGHT_VALID;
2652 		if (isnullstartblock(right.br_startblock))
2653 			state |= BMAP_RIGHT_DELAY;
2654 	}
2655 
2656 	/*
2657 	 * We're inserting a real allocation between "left" and "right".
2658 	 * Set the contiguity flags.  Don't let extents get too large.
2659 	 */
2660 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2661 	    left.br_startoff + left.br_blockcount == new->br_startoff &&
2662 	    left.br_startblock + left.br_blockcount == new->br_startblock &&
2663 	    left.br_state == new->br_state &&
2664 	    left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2665 		state |= BMAP_LEFT_CONTIG;
2666 
2667 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2668 	    new->br_startoff + new->br_blockcount == right.br_startoff &&
2669 	    new->br_startblock + new->br_blockcount == right.br_startblock &&
2670 	    new->br_state == right.br_state &&
2671 	    new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2672 	    (!(state & BMAP_LEFT_CONTIG) ||
2673 	     left.br_blockcount + new->br_blockcount +
2674 	     right.br_blockcount <= MAXEXTLEN))
2675 		state |= BMAP_RIGHT_CONTIG;
2676 
2677 	error = 0;
2678 	/*
2679 	 * Select which case we're in here, and implement it.
2680 	 */
2681 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2682 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2683 		/*
2684 		 * New allocation is contiguous with real allocations on the
2685 		 * left and on the right.
2686 		 * Merge all three into a single extent record.
2687 		 */
2688 		left.br_blockcount += new->br_blockcount + right.br_blockcount;
2689 
2690 		xfs_iext_remove(ip, icur, state);
2691 		xfs_iext_prev(ifp, icur);
2692 		xfs_iext_update_extent(ip, state, icur, &left);
2693 		ifp->if_nextents--;
2694 
2695 		if (cur == NULL) {
2696 			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2697 		} else {
2698 			rval = XFS_ILOG_CORE;
2699 			error = xfs_bmbt_lookup_eq(cur, &right, &i);
2700 			if (error)
2701 				goto done;
2702 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2703 				error = -EFSCORRUPTED;
2704 				goto done;
2705 			}
2706 			error = xfs_btree_delete(cur, &i);
2707 			if (error)
2708 				goto done;
2709 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2710 				error = -EFSCORRUPTED;
2711 				goto done;
2712 			}
2713 			error = xfs_btree_decrement(cur, 0, &i);
2714 			if (error)
2715 				goto done;
2716 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2717 				error = -EFSCORRUPTED;
2718 				goto done;
2719 			}
2720 			error = xfs_bmbt_update(cur, &left);
2721 			if (error)
2722 				goto done;
2723 		}
2724 		break;
2725 
2726 	case BMAP_LEFT_CONTIG:
2727 		/*
2728 		 * New allocation is contiguous with a real allocation
2729 		 * on the left.
2730 		 * Merge the new allocation with the left neighbor.
2731 		 */
2732 		old = left;
2733 		left.br_blockcount += new->br_blockcount;
2734 
2735 		xfs_iext_prev(ifp, icur);
2736 		xfs_iext_update_extent(ip, state, icur, &left);
2737 
2738 		if (cur == NULL) {
2739 			rval = xfs_ilog_fext(whichfork);
2740 		} else {
2741 			rval = 0;
2742 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2743 			if (error)
2744 				goto done;
2745 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2746 				error = -EFSCORRUPTED;
2747 				goto done;
2748 			}
2749 			error = xfs_bmbt_update(cur, &left);
2750 			if (error)
2751 				goto done;
2752 		}
2753 		break;
2754 
2755 	case BMAP_RIGHT_CONTIG:
2756 		/*
2757 		 * New allocation is contiguous with a real allocation
2758 		 * on the right.
2759 		 * Merge the new allocation with the right neighbor.
2760 		 */
2761 		old = right;
2762 
2763 		right.br_startoff = new->br_startoff;
2764 		right.br_startblock = new->br_startblock;
2765 		right.br_blockcount += new->br_blockcount;
2766 		xfs_iext_update_extent(ip, state, icur, &right);
2767 
2768 		if (cur == NULL) {
2769 			rval = xfs_ilog_fext(whichfork);
2770 		} else {
2771 			rval = 0;
2772 			error = xfs_bmbt_lookup_eq(cur, &old, &i);
2773 			if (error)
2774 				goto done;
2775 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2776 				error = -EFSCORRUPTED;
2777 				goto done;
2778 			}
2779 			error = xfs_bmbt_update(cur, &right);
2780 			if (error)
2781 				goto done;
2782 		}
2783 		break;
2784 
2785 	case 0:
2786 		/*
2787 		 * New allocation is not contiguous with another
2788 		 * real allocation.
2789 		 * Insert a new entry.
2790 		 */
2791 		xfs_iext_insert(ip, icur, new, state);
2792 		ifp->if_nextents++;
2793 
2794 		if (cur == NULL) {
2795 			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2796 		} else {
2797 			rval = XFS_ILOG_CORE;
2798 			error = xfs_bmbt_lookup_eq(cur, new, &i);
2799 			if (error)
2800 				goto done;
2801 			if (XFS_IS_CORRUPT(mp, i != 0)) {
2802 				error = -EFSCORRUPTED;
2803 				goto done;
2804 			}
2805 			error = xfs_btree_insert(cur, &i);
2806 			if (error)
2807 				goto done;
2808 			if (XFS_IS_CORRUPT(mp, i != 1)) {
2809 				error = -EFSCORRUPTED;
2810 				goto done;
2811 			}
2812 		}
2813 		break;
2814 	}
2815 
2816 	/* add reverse mapping unless caller opted out */
2817 	if (!(flags & XFS_BMAPI_NORMAP))
2818 		xfs_rmap_map_extent(tp, ip, whichfork, new);
2819 
2820 	/* convert to a btree if necessary */
2821 	if (xfs_bmap_needs_btree(ip, whichfork)) {
2822 		int	tmp_logflags;	/* partial log flag return val */
2823 
2824 		ASSERT(cur == NULL);
2825 		error = xfs_bmap_extents_to_btree(tp, ip, curp, 0,
2826 				&tmp_logflags, whichfork);
2827 		*logflagsp |= tmp_logflags;
2828 		cur = *curp;
2829 		if (error)
2830 			goto done;
2831 	}
2832 
2833 	/* clear out the allocated field, done with it now in any case. */
2834 	if (cur)
2835 		cur->bc_ino.allocated = 0;
2836 
2837 	xfs_bmap_check_leaf_extents(cur, ip, whichfork);
2838 done:
2839 	*logflagsp |= rval;
2840 	return error;
2841 }
2842 
2843 /*
2844  * Functions used in the extent read, allocate and remove paths
2845  */
2846 
2847 /*
2848  * Adjust the size of the new extent based on i_extsize and rt extsize.
2849  */
2850 int
2851 xfs_bmap_extsize_align(
2852 	xfs_mount_t	*mp,
2853 	xfs_bmbt_irec_t	*gotp,		/* next extent pointer */
2854 	xfs_bmbt_irec_t	*prevp,		/* previous extent pointer */
2855 	xfs_extlen_t	extsz,		/* align to this extent size */
2856 	int		rt,		/* is this a realtime inode? */
2857 	int		eof,		/* is extent at end-of-file? */
2858 	int		delay,		/* creating delalloc extent? */
2859 	int		convert,	/* overwriting unwritten extent? */
2860 	xfs_fileoff_t	*offp,		/* in/out: aligned offset */
2861 	xfs_extlen_t	*lenp)		/* in/out: aligned length */
2862 {
2863 	xfs_fileoff_t	orig_off;	/* original offset */
2864 	xfs_extlen_t	orig_alen;	/* original length */
2865 	xfs_fileoff_t	orig_end;	/* original off+len */
2866 	xfs_fileoff_t	nexto;		/* next file offset */
2867 	xfs_fileoff_t	prevo;		/* previous file offset */
2868 	xfs_fileoff_t	align_off;	/* temp for offset */
2869 	xfs_extlen_t	align_alen;	/* temp for length */
2870 	xfs_extlen_t	temp;		/* temp for calculations */
2871 
2872 	if (convert)
2873 		return 0;
2874 
2875 	orig_off = align_off = *offp;
2876 	orig_alen = align_alen = *lenp;
2877 	orig_end = orig_off + orig_alen;
2878 
2879 	/*
2880 	 * If this request overlaps an existing extent, then don't
2881 	 * attempt to perform any additional alignment.
2882 	 */
2883 	if (!delay && !eof &&
2884 	    (orig_off >= gotp->br_startoff) &&
2885 	    (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2886 		return 0;
2887 	}
2888 
2889 	/*
2890 	 * If the file offset is unaligned vs. the extent size
2891 	 * we need to align it.  This will be possible unless
2892 	 * the file was previously written with a kernel that didn't
2893 	 * perform this alignment, or if a truncate shot us in the
2894 	 * foot.
2895 	 */
2896 	div_u64_rem(orig_off, extsz, &temp);
2897 	if (temp) {
2898 		align_alen += temp;
2899 		align_off -= temp;
2900 	}
2901 
2902 	/* Same adjustment for the end of the requested area. */
2903 	temp = (align_alen % extsz);
2904 	if (temp)
2905 		align_alen += extsz - temp;
2906 
2907 	/*
2908 	 * For large extent hint sizes, the aligned extent might be larger than
2909 	 * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
2910 	 * the length back under MAXEXTLEN. The outer allocation loops handle
2911 	 * short allocation just fine, so it is safe to do this. We only want to
2912 	 * do it when we are forced to, though, because it means more allocation
2913 	 * operations are required.
2914 	 */
2915 	while (align_alen > MAXEXTLEN)
2916 		align_alen -= extsz;
2917 	ASSERT(align_alen <= MAXEXTLEN);
2918 
2919 	/*
2920 	 * If the previous block overlaps with this proposed allocation
2921 	 * then move the start forward without adjusting the length.
2922 	 */
2923 	if (prevp->br_startoff != NULLFILEOFF) {
2924 		if (prevp->br_startblock == HOLESTARTBLOCK)
2925 			prevo = prevp->br_startoff;
2926 		else
2927 			prevo = prevp->br_startoff + prevp->br_blockcount;
2928 	} else
2929 		prevo = 0;
2930 	if (align_off != orig_off && align_off < prevo)
2931 		align_off = prevo;
2932 	/*
2933 	 * If the next block overlaps with this proposed allocation
2934 	 * then move the start back without adjusting the length,
2935 	 * but not before offset 0.
2936 	 * This may of course make the start overlap previous block,
2937 	 * and if we hit the offset 0 limit then the next block
2938 	 * can still overlap too.
2939 	 */
2940 	if (!eof && gotp->br_startoff != NULLFILEOFF) {
2941 		if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2942 		    (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2943 			nexto = gotp->br_startoff + gotp->br_blockcount;
2944 		else
2945 			nexto = gotp->br_startoff;
2946 	} else
2947 		nexto = NULLFILEOFF;
2948 	if (!eof &&
2949 	    align_off + align_alen != orig_end &&
2950 	    align_off + align_alen > nexto)
2951 		align_off = nexto > align_alen ? nexto - align_alen : 0;
2952 	/*
2953 	 * If we're now overlapping the next or previous extent that
2954 	 * means we can't fit an extsz piece in this hole.  Just move
2955 	 * the start forward to the first valid spot and set
2956 	 * the length so we hit the end.
2957 	 */
2958 	if (align_off != orig_off && align_off < prevo)
2959 		align_off = prevo;
2960 	if (align_off + align_alen != orig_end &&
2961 	    align_off + align_alen > nexto &&
2962 	    nexto != NULLFILEOFF) {
2963 		ASSERT(nexto > prevo);
2964 		align_alen = nexto - align_off;
2965 	}
2966 
2967 	/*
2968 	 * If realtime, and the result isn't a multiple of the realtime
2969 	 * extent size we need to remove blocks until it is.
2970 	 */
2971 	if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2972 		/*
2973 		 * We're not covering the original request, or
2974 		 * we won't be able to once we fix the length.
2975 		 */
2976 		if (orig_off < align_off ||
2977 		    orig_end > align_off + align_alen ||
2978 		    align_alen - temp < orig_alen)
2979 			return -EINVAL;
2980 		/*
2981 		 * Try to fix it by moving the start up.
2982 		 */
2983 		if (align_off + temp <= orig_off) {
2984 			align_alen -= temp;
2985 			align_off += temp;
2986 		}
2987 		/*
2988 		 * Try to fix it by moving the end in.
2989 		 */
2990 		else if (align_off + align_alen - temp >= orig_end)
2991 			align_alen -= temp;
2992 		/*
2993 		 * Set the start to the minimum then trim the length.
2994 		 */
2995 		else {
2996 			align_alen -= orig_off - align_off;
2997 			align_off = orig_off;
2998 			align_alen -= align_alen % mp->m_sb.sb_rextsize;
2999 		}
3000 		/*
3001 		 * Result doesn't cover the request, fail it.
3002 		 */
3003 		if (orig_off < align_off || orig_end > align_off + align_alen)
3004 			return -EINVAL;
3005 	} else {
3006 		ASSERT(orig_off >= align_off);
3007 		/* see MAXEXTLEN handling above */
3008 		ASSERT(orig_end <= align_off + align_alen ||
3009 		       align_alen + extsz > MAXEXTLEN);
3010 	}
3011 
3012 #ifdef DEBUG
3013 	if (!eof && gotp->br_startoff != NULLFILEOFF)
3014 		ASSERT(align_off + align_alen <= gotp->br_startoff);
3015 	if (prevp->br_startoff != NULLFILEOFF)
3016 		ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3017 #endif
3018 
3019 	*lenp = align_alen;
3020 	*offp = align_off;
3021 	return 0;
3022 }
3023 
3024 #define XFS_ALLOC_GAP_UNITS	4
3025 
3026 void
3027 xfs_bmap_adjacent(
3028 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3029 {
3030 	xfs_fsblock_t	adjust;		/* adjustment to block numbers */
3031 	xfs_agnumber_t	fb_agno;	/* ag number of ap->firstblock */
3032 	xfs_mount_t	*mp;		/* mount point structure */
3033 	int		nullfb;		/* true if ap->firstblock isn't set */
3034 	int		rt;		/* true if inode is realtime */
3035 
3036 #define	ISVALID(x,y)	\
3037 	(rt ? \
3038 		(x) < mp->m_sb.sb_rblocks : \
3039 		XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3040 		XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3041 		XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3042 
3043 	mp = ap->ip->i_mount;
3044 	nullfb = ap->tp->t_firstblock == NULLFSBLOCK;
3045 	rt = XFS_IS_REALTIME_INODE(ap->ip) &&
3046 		(ap->datatype & XFS_ALLOC_USERDATA);
3047 	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp,
3048 							ap->tp->t_firstblock);
3049 	/*
3050 	 * If allocating at eof, and there's a previous real block,
3051 	 * try to use its last block as our starting point.
3052 	 */
3053 	if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3054 	    !isnullstartblock(ap->prev.br_startblock) &&
3055 	    ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3056 		    ap->prev.br_startblock)) {
3057 		ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3058 		/*
3059 		 * Adjust for the gap between prevp and us.
3060 		 */
3061 		adjust = ap->offset -
3062 			(ap->prev.br_startoff + ap->prev.br_blockcount);
3063 		if (adjust &&
3064 		    ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3065 			ap->blkno += adjust;
3066 	}
3067 	/*
3068 	 * If not at eof, then compare the two neighbor blocks.
3069 	 * Figure out whether either one gives us a good starting point,
3070 	 * and pick the better one.
3071 	 */
3072 	else if (!ap->eof) {
3073 		xfs_fsblock_t	gotbno;		/* right side block number */
3074 		xfs_fsblock_t	gotdiff=0;	/* right side difference */
3075 		xfs_fsblock_t	prevbno;	/* left side block number */
3076 		xfs_fsblock_t	prevdiff=0;	/* left side difference */
3077 
3078 		/*
3079 		 * If there's a previous (left) block, select a requested
3080 		 * start block based on it.
3081 		 */
3082 		if (ap->prev.br_startoff != NULLFILEOFF &&
3083 		    !isnullstartblock(ap->prev.br_startblock) &&
3084 		    (prevbno = ap->prev.br_startblock +
3085 			       ap->prev.br_blockcount) &&
3086 		    ISVALID(prevbno, ap->prev.br_startblock)) {
3087 			/*
3088 			 * Calculate gap to end of previous block.
3089 			 */
3090 			adjust = prevdiff = ap->offset -
3091 				(ap->prev.br_startoff +
3092 				 ap->prev.br_blockcount);
3093 			/*
3094 			 * Figure the startblock based on the previous block's
3095 			 * end and the gap size.
3096 			 * Heuristic!
3097 			 * If the gap is large relative to the piece we're
3098 			 * allocating, or using it gives us an invalid block
3099 			 * number, then just use the end of the previous block.
3100 			 */
3101 			if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3102 			    ISVALID(prevbno + prevdiff,
3103 				    ap->prev.br_startblock))
3104 				prevbno += adjust;
3105 			else
3106 				prevdiff += adjust;
3107 			/*
3108 			 * If the firstblock forbids it, can't use it,
3109 			 * must use default.
3110 			 */
3111 			if (!rt && !nullfb &&
3112 			    XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3113 				prevbno = NULLFSBLOCK;
3114 		}
3115 		/*
3116 		 * No previous block or can't follow it, just default.
3117 		 */
3118 		else
3119 			prevbno = NULLFSBLOCK;
3120 		/*
3121 		 * If there's a following (right) block, select a requested
3122 		 * start block based on it.
3123 		 */
3124 		if (!isnullstartblock(ap->got.br_startblock)) {
3125 			/*
3126 			 * Calculate gap to start of next block.
3127 			 */
3128 			adjust = gotdiff = ap->got.br_startoff - ap->offset;
3129 			/*
3130 			 * Figure the startblock based on the next block's
3131 			 * start and the gap size.
3132 			 */
3133 			gotbno = ap->got.br_startblock;
3134 			/*
3135 			 * Heuristic!
3136 			 * If the gap is large relative to the piece we're
3137 			 * allocating, or using it gives us an invalid block
3138 			 * number, then just use the start of the next block
3139 			 * offset by our length.
3140 			 */
3141 			if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3142 			    ISVALID(gotbno - gotdiff, gotbno))
3143 				gotbno -= adjust;
3144 			else if (ISVALID(gotbno - ap->length, gotbno)) {
3145 				gotbno -= ap->length;
3146 				gotdiff += adjust - ap->length;
3147 			} else
3148 				gotdiff += adjust;
3149 			/*
3150 			 * If the firstblock forbids it, can't use it,
3151 			 * must use default.
3152 			 */
3153 			if (!rt && !nullfb &&
3154 			    XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3155 				gotbno = NULLFSBLOCK;
3156 		}
3157 		/*
3158 		 * No next block, just default.
3159 		 */
3160 		else
3161 			gotbno = NULLFSBLOCK;
3162 		/*
3163 		 * If both valid, pick the better one, else the only good
3164 		 * one, else ap->blkno is already set (to 0 or the inode block).
3165 		 */
3166 		if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3167 			ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3168 		else if (prevbno != NULLFSBLOCK)
3169 			ap->blkno = prevbno;
3170 		else if (gotbno != NULLFSBLOCK)
3171 			ap->blkno = gotbno;
3172 	}
3173 #undef ISVALID
3174 }
3175 
3176 static int
3177 xfs_bmap_longest_free_extent(
3178 	struct xfs_trans	*tp,
3179 	xfs_agnumber_t		ag,
3180 	xfs_extlen_t		*blen,
3181 	int			*notinit)
3182 {
3183 	struct xfs_mount	*mp = tp->t_mountp;
3184 	struct xfs_perag	*pag;
3185 	xfs_extlen_t		longest;
3186 	int			error = 0;
3187 
3188 	pag = xfs_perag_get(mp, ag);
3189 	if (!pag->pagf_init) {
3190 		error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3191 		if (error) {
3192 			/* Couldn't lock the AGF, so skip this AG. */
3193 			if (error == -EAGAIN) {
3194 				*notinit = 1;
3195 				error = 0;
3196 			}
3197 			goto out;
3198 		}
3199 	}
3200 
3201 	longest = xfs_alloc_longest_free_extent(pag,
3202 				xfs_alloc_min_freelist(mp, pag),
3203 				xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
3204 	if (*blen < longest)
3205 		*blen = longest;
3206 
3207 out:
3208 	xfs_perag_put(pag);
3209 	return error;
3210 }
3211 
3212 static void
3213 xfs_bmap_select_minlen(
3214 	struct xfs_bmalloca	*ap,
3215 	struct xfs_alloc_arg	*args,
3216 	xfs_extlen_t		*blen,
3217 	int			notinit)
3218 {
3219 	if (notinit || *blen < ap->minlen) {
3220 		/*
3221 		 * Since we did a BUF_TRYLOCK above, it is possible that
3222 		 * there is space for this request.
3223 		 */
3224 		args->minlen = ap->minlen;
3225 	} else if (*blen < args->maxlen) {
3226 		/*
3227 		 * If the best seen length is less than the request length,
3228 		 * use the best as the minimum.
3229 		 */
3230 		args->minlen = *blen;
3231 	} else {
3232 		/*
3233 		 * Otherwise we've seen an extent as big as maxlen, use that
3234 		 * as the minimum.
3235 		 */
3236 		args->minlen = args->maxlen;
3237 	}
3238 }
3239 
3240 STATIC int
3241 xfs_bmap_btalloc_nullfb(
3242 	struct xfs_bmalloca	*ap,
3243 	struct xfs_alloc_arg	*args,
3244 	xfs_extlen_t		*blen)
3245 {
3246 	struct xfs_mount	*mp = ap->ip->i_mount;
3247 	xfs_agnumber_t		ag, startag;
3248 	int			notinit = 0;
3249 	int			error;
3250 
3251 	args->type = XFS_ALLOCTYPE_START_BNO;
3252 	args->total = ap->total;
3253 
3254 	startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3255 	if (startag == NULLAGNUMBER)
3256 		startag = ag = 0;
3257 
3258 	while (*blen < args->maxlen) {
3259 		error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3260 						     &notinit);
3261 		if (error)
3262 			return error;
3263 
3264 		if (++ag == mp->m_sb.sb_agcount)
3265 			ag = 0;
3266 		if (ag == startag)
3267 			break;
3268 	}
3269 
3270 	xfs_bmap_select_minlen(ap, args, blen, notinit);
3271 	return 0;
3272 }
3273 
3274 STATIC int
3275 xfs_bmap_btalloc_filestreams(
3276 	struct xfs_bmalloca	*ap,
3277 	struct xfs_alloc_arg	*args,
3278 	xfs_extlen_t		*blen)
3279 {
3280 	struct xfs_mount	*mp = ap->ip->i_mount;
3281 	xfs_agnumber_t		ag;
3282 	int			notinit = 0;
3283 	int			error;
3284 
3285 	args->type = XFS_ALLOCTYPE_NEAR_BNO;
3286 	args->total = ap->total;
3287 
3288 	ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3289 	if (ag == NULLAGNUMBER)
3290 		ag = 0;
3291 
3292 	error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3293 	if (error)
3294 		return error;
3295 
3296 	if (*blen < args->maxlen) {
3297 		error = xfs_filestream_new_ag(ap, &ag);
3298 		if (error)
3299 			return error;
3300 
3301 		error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3302 						     &notinit);
3303 		if (error)
3304 			return error;
3305 
3306 	}
3307 
3308 	xfs_bmap_select_minlen(ap, args, blen, notinit);
3309 
3310 	/*
3311 	 * Set the failure fallback case to look in the selected AG as stream
3312 	 * may have moved.
3313 	 */
3314 	ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3315 	return 0;
3316 }
3317 
3318 /* Update all inode and quota accounting for the allocation we just did. */
3319 static void
3320 xfs_bmap_btalloc_accounting(
3321 	struct xfs_bmalloca	*ap,
3322 	struct xfs_alloc_arg	*args)
3323 {
3324 	if (ap->flags & XFS_BMAPI_COWFORK) {
3325 		/*
3326 		 * COW fork blocks are in-core only and thus are treated as
3327 		 * in-core quota reservation (like delalloc blocks) even when
3328 		 * converted to real blocks. The quota reservation is not
3329 		 * accounted to disk until blocks are remapped to the data
3330 		 * fork. So if these blocks were previously delalloc, we
3331 		 * already have quota reservation and there's nothing to do
3332 		 * yet.
3333 		 */
3334 		if (ap->wasdel) {
3335 			xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len);
3336 			return;
3337 		}
3338 
3339 		/*
3340 		 * Otherwise, we've allocated blocks in a hole. The transaction
3341 		 * has acquired in-core quota reservation for this extent.
3342 		 * Rather than account these as real blocks, however, we reduce
3343 		 * the transaction quota reservation based on the allocation.
3344 		 * This essentially transfers the transaction quota reservation
3345 		 * to that of a delalloc extent.
3346 		 */
3347 		ap->ip->i_delayed_blks += args->len;
3348 		xfs_trans_mod_dquot_byino(ap->tp, ap->ip, XFS_TRANS_DQ_RES_BLKS,
3349 				-(long)args->len);
3350 		return;
3351 	}
3352 
3353 	/* data/attr fork only */
3354 	ap->ip->i_nblocks += args->len;
3355 	xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3356 	if (ap->wasdel) {
3357 		ap->ip->i_delayed_blks -= args->len;
3358 		xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len);
3359 	}
3360 	xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3361 		ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT : XFS_TRANS_DQ_BCOUNT,
3362 		args->len);
3363 }
3364 
3365 static int
3366 xfs_bmap_compute_alignments(
3367 	struct xfs_bmalloca	*ap,
3368 	struct xfs_alloc_arg	*args)
3369 {
3370 	struct xfs_mount	*mp = args->mp;
3371 	xfs_extlen_t		align = 0; /* minimum allocation alignment */
3372 	int			stripe_align = 0;
3373 
3374 	/* stripe alignment for allocation is determined by mount parameters */
3375 	if (mp->m_swidth && xfs_has_swalloc(mp))
3376 		stripe_align = mp->m_swidth;
3377 	else if (mp->m_dalign)
3378 		stripe_align = mp->m_dalign;
3379 
3380 	if (ap->flags & XFS_BMAPI_COWFORK)
3381 		align = xfs_get_cowextsz_hint(ap->ip);
3382 	else if (ap->datatype & XFS_ALLOC_USERDATA)
3383 		align = xfs_get_extsz_hint(ap->ip);
3384 	if (align) {
3385 		if (xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, align, 0,
3386 					ap->eof, 0, ap->conv, &ap->offset,
3387 					&ap->length))
3388 			ASSERT(0);
3389 		ASSERT(ap->length);
3390 	}
3391 
3392 	/* apply extent size hints if obtained earlier */
3393 	if (align) {
3394 		args->prod = align;
3395 		div_u64_rem(ap->offset, args->prod, &args->mod);
3396 		if (args->mod)
3397 			args->mod = args->prod - args->mod;
3398 	} else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3399 		args->prod = 1;
3400 		args->mod = 0;
3401 	} else {
3402 		args->prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3403 		div_u64_rem(ap->offset, args->prod, &args->mod);
3404 		if (args->mod)
3405 			args->mod = args->prod - args->mod;
3406 	}
3407 
3408 	return stripe_align;
3409 }
3410 
3411 static void
3412 xfs_bmap_process_allocated_extent(
3413 	struct xfs_bmalloca	*ap,
3414 	struct xfs_alloc_arg	*args,
3415 	xfs_fileoff_t		orig_offset,
3416 	xfs_extlen_t		orig_length)
3417 {
3418 	int			nullfb;
3419 
3420 	nullfb = ap->tp->t_firstblock == NULLFSBLOCK;
3421 
3422 	/*
3423 	 * check the allocation happened at the same or higher AG than
3424 	 * the first block that was allocated.
3425 	 */
3426 	ASSERT(nullfb ||
3427 		XFS_FSB_TO_AGNO(args->mp, ap->tp->t_firstblock) <=
3428 		XFS_FSB_TO_AGNO(args->mp, args->fsbno));
3429 
3430 	ap->blkno = args->fsbno;
3431 	if (nullfb)
3432 		ap->tp->t_firstblock = args->fsbno;
3433 	ap->length = args->len;
3434 	/*
3435 	 * If the extent size hint is active, we tried to round the
3436 	 * caller's allocation request offset down to extsz and the
3437 	 * length up to another extsz boundary.  If we found a free
3438 	 * extent we mapped it in starting at this new offset.  If the
3439 	 * newly mapped space isn't long enough to cover any of the
3440 	 * range of offsets that was originally requested, move the
3441 	 * mapping up so that we can fill as much of the caller's
3442 	 * original request as possible.  Free space is apparently
3443 	 * very fragmented so we're unlikely to be able to satisfy the
3444 	 * hints anyway.
3445 	 */
3446 	if (ap->length <= orig_length)
3447 		ap->offset = orig_offset;
3448 	else if (ap->offset + ap->length < orig_offset + orig_length)
3449 		ap->offset = orig_offset + orig_length - ap->length;
3450 	xfs_bmap_btalloc_accounting(ap, args);
3451 }
3452 
3453 #ifdef DEBUG
3454 static int
3455 xfs_bmap_exact_minlen_extent_alloc(
3456 	struct xfs_bmalloca	*ap)
3457 {
3458 	struct xfs_mount	*mp = ap->ip->i_mount;
3459 	struct xfs_alloc_arg	args = { .tp = ap->tp, .mp = mp };
3460 	xfs_fileoff_t		orig_offset;
3461 	xfs_extlen_t		orig_length;
3462 	int			error;
3463 
3464 	ASSERT(ap->length);
3465 
3466 	if (ap->minlen != 1) {
3467 		ap->blkno = NULLFSBLOCK;
3468 		ap->length = 0;
3469 		return 0;
3470 	}
3471 
3472 	orig_offset = ap->offset;
3473 	orig_length = ap->length;
3474 
3475 	args.alloc_minlen_only = 1;
3476 
3477 	xfs_bmap_compute_alignments(ap, &args);
3478 
3479 	if (ap->tp->t_firstblock == NULLFSBLOCK) {
3480 		/*
3481 		 * Unlike the longest extent available in an AG, we don't track
3482 		 * the length of an AG's shortest extent.
3483 		 * XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT is a debug only knob and
3484 		 * hence we can afford to start traversing from the 0th AG since
3485 		 * we need not be concerned about a drop in performance in
3486 		 * "debug only" code paths.
3487 		 */
3488 		ap->blkno = XFS_AGB_TO_FSB(mp, 0, 0);
3489 	} else {
3490 		ap->blkno = ap->tp->t_firstblock;
3491 	}
3492 
3493 	args.fsbno = ap->blkno;
3494 	args.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
3495 	args.type = XFS_ALLOCTYPE_FIRST_AG;
3496 	args.minlen = args.maxlen = ap->minlen;
3497 	args.total = ap->total;
3498 
3499 	args.alignment = 1;
3500 	args.minalignslop = 0;
3501 
3502 	args.minleft = ap->minleft;
3503 	args.wasdel = ap->wasdel;
3504 	args.resv = XFS_AG_RESV_NONE;
3505 	args.datatype = ap->datatype;
3506 
3507 	error = xfs_alloc_vextent(&args);
3508 	if (error)
3509 		return error;
3510 
3511 	if (args.fsbno != NULLFSBLOCK) {
3512 		xfs_bmap_process_allocated_extent(ap, &args, orig_offset,
3513 			orig_length);
3514 	} else {
3515 		ap->blkno = NULLFSBLOCK;
3516 		ap->length = 0;
3517 	}
3518 
3519 	return 0;
3520 }
3521 #else
3522 
3523 #define xfs_bmap_exact_minlen_extent_alloc(bma) (-EFSCORRUPTED)
3524 
3525 #endif
3526 
3527 STATIC int
3528 xfs_bmap_btalloc(
3529 	struct xfs_bmalloca	*ap)
3530 {
3531 	struct xfs_mount	*mp = ap->ip->i_mount;
3532 	struct xfs_alloc_arg	args = { .tp = ap->tp, .mp = mp };
3533 	xfs_alloctype_t		atype = 0;
3534 	xfs_agnumber_t		fb_agno;	/* ag number of ap->firstblock */
3535 	xfs_agnumber_t		ag;
3536 	xfs_fileoff_t		orig_offset;
3537 	xfs_extlen_t		orig_length;
3538 	xfs_extlen_t		blen;
3539 	xfs_extlen_t		nextminlen = 0;
3540 	int			nullfb; /* true if ap->firstblock isn't set */
3541 	int			isaligned;
3542 	int			tryagain;
3543 	int			error;
3544 	int			stripe_align;
3545 
3546 	ASSERT(ap->length);
3547 	orig_offset = ap->offset;
3548 	orig_length = ap->length;
3549 
3550 	stripe_align = xfs_bmap_compute_alignments(ap, &args);
3551 
3552 	nullfb = ap->tp->t_firstblock == NULLFSBLOCK;
3553 	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp,
3554 							ap->tp->t_firstblock);
3555 	if (nullfb) {
3556 		if ((ap->datatype & XFS_ALLOC_USERDATA) &&
3557 		    xfs_inode_is_filestream(ap->ip)) {
3558 			ag = xfs_filestream_lookup_ag(ap->ip);
3559 			ag = (ag != NULLAGNUMBER) ? ag : 0;
3560 			ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3561 		} else {
3562 			ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3563 		}
3564 	} else
3565 		ap->blkno = ap->tp->t_firstblock;
3566 
3567 	xfs_bmap_adjacent(ap);
3568 
3569 	/*
3570 	 * If allowed, use ap->blkno; otherwise must use firstblock since
3571 	 * it's in the right allocation group.
3572 	 */
3573 	if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3574 		;
3575 	else
3576 		ap->blkno = ap->tp->t_firstblock;
3577 	/*
3578 	 * Normal allocation, done through xfs_alloc_vextent.
3579 	 */
3580 	tryagain = isaligned = 0;
3581 	args.fsbno = ap->blkno;
3582 	args.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
3583 
3584 	/* Trim the allocation back to the maximum an AG can fit. */
3585 	args.maxlen = min(ap->length, mp->m_ag_max_usable);
3586 	blen = 0;
3587 	if (nullfb) {
3588 		/*
3589 		 * Search for an allocation group with a single extent large
3590 		 * enough for the request.  If one isn't found, then adjust
3591 		 * the minimum allocation size to the largest space found.
3592 		 */
3593 		if ((ap->datatype & XFS_ALLOC_USERDATA) &&
3594 		    xfs_inode_is_filestream(ap->ip))
3595 			error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3596 		else
3597 			error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3598 		if (error)
3599 			return error;
3600 	} else if (ap->tp->t_flags & XFS_TRANS_LOWMODE) {
3601 		if (xfs_inode_is_filestream(ap->ip))
3602 			args.type = XFS_ALLOCTYPE_FIRST_AG;
3603 		else
3604 			args.type = XFS_ALLOCTYPE_START_BNO;
3605 		args.total = args.minlen = ap->minlen;
3606 	} else {
3607 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
3608 		args.total = ap->total;
3609 		args.minlen = ap->minlen;
3610 	}
3611 
3612 	/*
3613 	 * If we are not low on available data blocks, and the underlying
3614 	 * logical volume manager is a stripe, and the file offset is zero then
3615 	 * try to allocate data blocks on stripe unit boundary. NOTE: ap->aeof
3616 	 * is only set if the allocation length is >= the stripe unit and the
3617 	 * allocation offset is at the end of file.
3618 	 */
3619 	if (!(ap->tp->t_flags & XFS_TRANS_LOWMODE) && ap->aeof) {
3620 		if (!ap->offset) {
3621 			args.alignment = stripe_align;
3622 			atype = args.type;
3623 			isaligned = 1;
3624 			/*
3625 			 * Adjust minlen to try and preserve alignment if we
3626 			 * can't guarantee an aligned maxlen extent.
3627 			 */
3628 			if (blen > args.alignment &&
3629 			    blen <= args.maxlen + args.alignment)
3630 				args.minlen = blen - args.alignment;
3631 			args.minalignslop = 0;
3632 		} else {
3633 			/*
3634 			 * First try an exact bno allocation.
3635 			 * If it fails then do a near or start bno
3636 			 * allocation with alignment turned on.
3637 			 */
3638 			atype = args.type;
3639 			tryagain = 1;
3640 			args.type = XFS_ALLOCTYPE_THIS_BNO;
3641 			args.alignment = 1;
3642 			/*
3643 			 * Compute the minlen+alignment for the
3644 			 * next case.  Set slop so that the value
3645 			 * of minlen+alignment+slop doesn't go up
3646 			 * between the calls.
3647 			 */
3648 			if (blen > stripe_align && blen <= args.maxlen)
3649 				nextminlen = blen - stripe_align;
3650 			else
3651 				nextminlen = args.minlen;
3652 			if (nextminlen + stripe_align > args.minlen + 1)
3653 				args.minalignslop =
3654 					nextminlen + stripe_align -
3655 					args.minlen - 1;
3656 			else
3657 				args.minalignslop = 0;
3658 		}
3659 	} else {
3660 		args.alignment = 1;
3661 		args.minalignslop = 0;
3662 	}
3663 	args.minleft = ap->minleft;
3664 	args.wasdel = ap->wasdel;
3665 	args.resv = XFS_AG_RESV_NONE;
3666 	args.datatype = ap->datatype;
3667 
3668 	error = xfs_alloc_vextent(&args);
3669 	if (error)
3670 		return error;
3671 
3672 	if (tryagain && args.fsbno == NULLFSBLOCK) {
3673 		/*
3674 		 * Exact allocation failed. Now try with alignment
3675 		 * turned on.
3676 		 */
3677 		args.type = atype;
3678 		args.fsbno = ap->blkno;
3679 		args.alignment = stripe_align;
3680 		args.minlen = nextminlen;
3681 		args.minalignslop = 0;
3682 		isaligned = 1;
3683 		if ((error = xfs_alloc_vextent(&args)))
3684 			return error;
3685 	}
3686 	if (isaligned && args.fsbno == NULLFSBLOCK) {
3687 		/*
3688 		 * allocation failed, so turn off alignment and
3689 		 * try again.
3690 		 */
3691 		args.type = atype;
3692 		args.fsbno = ap->blkno;
3693 		args.alignment = 0;
3694 		if ((error = xfs_alloc_vextent(&args)))
3695 			return error;
3696 	}
3697 	if (args.fsbno == NULLFSBLOCK && nullfb &&
3698 	    args.minlen > ap->minlen) {
3699 		args.minlen = ap->minlen;
3700 		args.type = XFS_ALLOCTYPE_START_BNO;
3701 		args.fsbno = ap->blkno;
3702 		if ((error = xfs_alloc_vextent(&args)))
3703 			return error;
3704 	}
3705 	if (args.fsbno == NULLFSBLOCK && nullfb) {
3706 		args.fsbno = 0;
3707 		args.type = XFS_ALLOCTYPE_FIRST_AG;
3708 		args.total = ap->minlen;
3709 		if ((error = xfs_alloc_vextent(&args)))
3710 			return error;
3711 		ap->tp->t_flags |= XFS_TRANS_LOWMODE;
3712 	}
3713 
3714 	if (args.fsbno != NULLFSBLOCK) {
3715 		xfs_bmap_process_allocated_extent(ap, &args, orig_offset,
3716 			orig_length);
3717 	} else {
3718 		ap->blkno = NULLFSBLOCK;
3719 		ap->length = 0;
3720 	}
3721 	return 0;
3722 }
3723 
3724 /* Trim extent to fit a logical block range. */
3725 void
3726 xfs_trim_extent(
3727 	struct xfs_bmbt_irec	*irec,
3728 	xfs_fileoff_t		bno,
3729 	xfs_filblks_t		len)
3730 {
3731 	xfs_fileoff_t		distance;
3732 	xfs_fileoff_t		end = bno + len;
3733 
3734 	if (irec->br_startoff + irec->br_blockcount <= bno ||
3735 	    irec->br_startoff >= end) {
3736 		irec->br_blockcount = 0;
3737 		return;
3738 	}
3739 
3740 	if (irec->br_startoff < bno) {
3741 		distance = bno - irec->br_startoff;
3742 		if (isnullstartblock(irec->br_startblock))
3743 			irec->br_startblock = DELAYSTARTBLOCK;
3744 		if (irec->br_startblock != DELAYSTARTBLOCK &&
3745 		    irec->br_startblock != HOLESTARTBLOCK)
3746 			irec->br_startblock += distance;
3747 		irec->br_startoff += distance;
3748 		irec->br_blockcount -= distance;
3749 	}
3750 
3751 	if (end < irec->br_startoff + irec->br_blockcount) {
3752 		distance = irec->br_startoff + irec->br_blockcount - end;
3753 		irec->br_blockcount -= distance;
3754 	}
3755 }
3756 
3757 /*
3758  * Trim the returned map to the required bounds
3759  */
3760 STATIC void
3761 xfs_bmapi_trim_map(
3762 	struct xfs_bmbt_irec	*mval,
3763 	struct xfs_bmbt_irec	*got,
3764 	xfs_fileoff_t		*bno,
3765 	xfs_filblks_t		len,
3766 	xfs_fileoff_t		obno,
3767 	xfs_fileoff_t		end,
3768 	int			n,
3769 	int			flags)
3770 {
3771 	if ((flags & XFS_BMAPI_ENTIRE) ||
3772 	    got->br_startoff + got->br_blockcount <= obno) {
3773 		*mval = *got;
3774 		if (isnullstartblock(got->br_startblock))
3775 			mval->br_startblock = DELAYSTARTBLOCK;
3776 		return;
3777 	}
3778 
3779 	if (obno > *bno)
3780 		*bno = obno;
3781 	ASSERT((*bno >= obno) || (n == 0));
3782 	ASSERT(*bno < end);
3783 	mval->br_startoff = *bno;
3784 	if (isnullstartblock(got->br_startblock))
3785 		mval->br_startblock = DELAYSTARTBLOCK;
3786 	else
3787 		mval->br_startblock = got->br_startblock +
3788 					(*bno - got->br_startoff);
3789 	/*
3790 	 * Return the minimum of what we got and what we asked for for
3791 	 * the length.  We can use the len variable here because it is
3792 	 * modified below and we could have been there before coming
3793 	 * here if the first part of the allocation didn't overlap what
3794 	 * was asked for.
3795 	 */
3796 	mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3797 			got->br_blockcount - (*bno - got->br_startoff));
3798 	mval->br_state = got->br_state;
3799 	ASSERT(mval->br_blockcount <= len);
3800 	return;
3801 }
3802 
3803 /*
3804  * Update and validate the extent map to return
3805  */
3806 STATIC void
3807 xfs_bmapi_update_map(
3808 	struct xfs_bmbt_irec	**map,
3809 	xfs_fileoff_t		*bno,
3810 	xfs_filblks_t		*len,
3811 	xfs_fileoff_t		obno,
3812 	xfs_fileoff_t		end,
3813 	int			*n,
3814 	int			flags)
3815 {
3816 	xfs_bmbt_irec_t	*mval = *map;
3817 
3818 	ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3819 	       ((mval->br_startoff + mval->br_blockcount) <= end));
3820 	ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3821 	       (mval->br_startoff < obno));
3822 
3823 	*bno = mval->br_startoff + mval->br_blockcount;
3824 	*len = end - *bno;
3825 	if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3826 		/* update previous map with new information */
3827 		ASSERT(mval->br_startblock == mval[-1].br_startblock);
3828 		ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3829 		ASSERT(mval->br_state == mval[-1].br_state);
3830 		mval[-1].br_blockcount = mval->br_blockcount;
3831 		mval[-1].br_state = mval->br_state;
3832 	} else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3833 		   mval[-1].br_startblock != DELAYSTARTBLOCK &&
3834 		   mval[-1].br_startblock != HOLESTARTBLOCK &&
3835 		   mval->br_startblock == mval[-1].br_startblock +
3836 					  mval[-1].br_blockcount &&
3837 		   mval[-1].br_state == mval->br_state) {
3838 		ASSERT(mval->br_startoff ==
3839 		       mval[-1].br_startoff + mval[-1].br_blockcount);
3840 		mval[-1].br_blockcount += mval->br_blockcount;
3841 	} else if (*n > 0 &&
3842 		   mval->br_startblock == DELAYSTARTBLOCK &&
3843 		   mval[-1].br_startblock == DELAYSTARTBLOCK &&
3844 		   mval->br_startoff ==
3845 		   mval[-1].br_startoff + mval[-1].br_blockcount) {
3846 		mval[-1].br_blockcount += mval->br_blockcount;
3847 		mval[-1].br_state = mval->br_state;
3848 	} else if (!((*n == 0) &&
3849 		     ((mval->br_startoff + mval->br_blockcount) <=
3850 		      obno))) {
3851 		mval++;
3852 		(*n)++;
3853 	}
3854 	*map = mval;
3855 }
3856 
3857 /*
3858  * Map file blocks to filesystem blocks without allocation.
3859  */
3860 int
3861 xfs_bmapi_read(
3862 	struct xfs_inode	*ip,
3863 	xfs_fileoff_t		bno,
3864 	xfs_filblks_t		len,
3865 	struct xfs_bmbt_irec	*mval,
3866 	int			*nmap,
3867 	int			flags)
3868 {
3869 	struct xfs_mount	*mp = ip->i_mount;
3870 	int			whichfork = xfs_bmapi_whichfork(flags);
3871 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
3872 	struct xfs_bmbt_irec	got;
3873 	xfs_fileoff_t		obno;
3874 	xfs_fileoff_t		end;
3875 	struct xfs_iext_cursor	icur;
3876 	int			error;
3877 	bool			eof = false;
3878 	int			n = 0;
3879 
3880 	ASSERT(*nmap >= 1);
3881 	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_ENTIRE)));
3882 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
3883 
3884 	if (WARN_ON_ONCE(!ifp))
3885 		return -EFSCORRUPTED;
3886 
3887 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
3888 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT))
3889 		return -EFSCORRUPTED;
3890 
3891 	if (xfs_is_shutdown(mp))
3892 		return -EIO;
3893 
3894 	XFS_STATS_INC(mp, xs_blk_mapr);
3895 
3896 	error = xfs_iread_extents(NULL, ip, whichfork);
3897 	if (error)
3898 		return error;
3899 
3900 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got))
3901 		eof = true;
3902 	end = bno + len;
3903 	obno = bno;
3904 
3905 	while (bno < end && n < *nmap) {
3906 		/* Reading past eof, act as though there's a hole up to end. */
3907 		if (eof)
3908 			got.br_startoff = end;
3909 		if (got.br_startoff > bno) {
3910 			/* Reading in a hole.  */
3911 			mval->br_startoff = bno;
3912 			mval->br_startblock = HOLESTARTBLOCK;
3913 			mval->br_blockcount =
3914 				XFS_FILBLKS_MIN(len, got.br_startoff - bno);
3915 			mval->br_state = XFS_EXT_NORM;
3916 			bno += mval->br_blockcount;
3917 			len -= mval->br_blockcount;
3918 			mval++;
3919 			n++;
3920 			continue;
3921 		}
3922 
3923 		/* set up the extent map to return. */
3924 		xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
3925 		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
3926 
3927 		/* If we're done, stop now. */
3928 		if (bno >= end || n >= *nmap)
3929 			break;
3930 
3931 		/* Else go on to the next record. */
3932 		if (!xfs_iext_next_extent(ifp, &icur, &got))
3933 			eof = true;
3934 	}
3935 	*nmap = n;
3936 	return 0;
3937 }
3938 
3939 /*
3940  * Add a delayed allocation extent to an inode. Blocks are reserved from the
3941  * global pool and the extent inserted into the inode in-core extent tree.
3942  *
3943  * On entry, got refers to the first extent beyond the offset of the extent to
3944  * allocate or eof is specified if no such extent exists. On return, got refers
3945  * to the extent record that was inserted to the inode fork.
3946  *
3947  * Note that the allocated extent may have been merged with contiguous extents
3948  * during insertion into the inode fork. Thus, got does not reflect the current
3949  * state of the inode fork on return. If necessary, the caller can use lastx to
3950  * look up the updated record in the inode fork.
3951  */
3952 int
3953 xfs_bmapi_reserve_delalloc(
3954 	struct xfs_inode	*ip,
3955 	int			whichfork,
3956 	xfs_fileoff_t		off,
3957 	xfs_filblks_t		len,
3958 	xfs_filblks_t		prealloc,
3959 	struct xfs_bmbt_irec	*got,
3960 	struct xfs_iext_cursor	*icur,
3961 	int			eof)
3962 {
3963 	struct xfs_mount	*mp = ip->i_mount;
3964 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
3965 	xfs_extlen_t		alen;
3966 	xfs_extlen_t		indlen;
3967 	int			error;
3968 	xfs_fileoff_t		aoff = off;
3969 
3970 	/*
3971 	 * Cap the alloc length. Keep track of prealloc so we know whether to
3972 	 * tag the inode before we return.
3973 	 */
3974 	alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN);
3975 	if (!eof)
3976 		alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
3977 	if (prealloc && alen >= len)
3978 		prealloc = alen - len;
3979 
3980 	/* Figure out the extent size, adjust alen */
3981 	if (whichfork == XFS_COW_FORK) {
3982 		struct xfs_bmbt_irec	prev;
3983 		xfs_extlen_t		extsz = xfs_get_cowextsz_hint(ip);
3984 
3985 		if (!xfs_iext_peek_prev_extent(ifp, icur, &prev))
3986 			prev.br_startoff = NULLFILEOFF;
3987 
3988 		error = xfs_bmap_extsize_align(mp, got, &prev, extsz, 0, eof,
3989 					       1, 0, &aoff, &alen);
3990 		ASSERT(!error);
3991 	}
3992 
3993 	/*
3994 	 * Make a transaction-less quota reservation for delayed allocation
3995 	 * blocks.  This number gets adjusted later.  We return if we haven't
3996 	 * allocated blocks already inside this loop.
3997 	 */
3998 	error = xfs_quota_reserve_blkres(ip, alen);
3999 	if (error)
4000 		return error;
4001 
4002 	/*
4003 	 * Split changing sb for alen and indlen since they could be coming
4004 	 * from different places.
4005 	 */
4006 	indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4007 	ASSERT(indlen > 0);
4008 
4009 	error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
4010 	if (error)
4011 		goto out_unreserve_quota;
4012 
4013 	error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
4014 	if (error)
4015 		goto out_unreserve_blocks;
4016 
4017 
4018 	ip->i_delayed_blks += alen;
4019 	xfs_mod_delalloc(ip->i_mount, alen + indlen);
4020 
4021 	got->br_startoff = aoff;
4022 	got->br_startblock = nullstartblock(indlen);
4023 	got->br_blockcount = alen;
4024 	got->br_state = XFS_EXT_NORM;
4025 
4026 	xfs_bmap_add_extent_hole_delay(ip, whichfork, icur, got);
4027 
4028 	/*
4029 	 * Tag the inode if blocks were preallocated. Note that COW fork
4030 	 * preallocation can occur at the start or end of the extent, even when
4031 	 * prealloc == 0, so we must also check the aligned offset and length.
4032 	 */
4033 	if (whichfork == XFS_DATA_FORK && prealloc)
4034 		xfs_inode_set_eofblocks_tag(ip);
4035 	if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len))
4036 		xfs_inode_set_cowblocks_tag(ip);
4037 
4038 	return 0;
4039 
4040 out_unreserve_blocks:
4041 	xfs_mod_fdblocks(mp, alen, false);
4042 out_unreserve_quota:
4043 	if (XFS_IS_QUOTA_ON(mp))
4044 		xfs_quota_unreserve_blkres(ip, alen);
4045 	return error;
4046 }
4047 
4048 static int
4049 xfs_bmap_alloc_userdata(
4050 	struct xfs_bmalloca	*bma)
4051 {
4052 	struct xfs_mount	*mp = bma->ip->i_mount;
4053 	int			whichfork = xfs_bmapi_whichfork(bma->flags);
4054 	int			error;
4055 
4056 	/*
4057 	 * Set the data type being allocated. For the data fork, the first data
4058 	 * in the file is treated differently to all other allocations. For the
4059 	 * attribute fork, we only need to ensure the allocated range is not on
4060 	 * the busy list.
4061 	 */
4062 	bma->datatype = XFS_ALLOC_NOBUSY;
4063 	if (whichfork == XFS_DATA_FORK) {
4064 		bma->datatype |= XFS_ALLOC_USERDATA;
4065 		if (bma->offset == 0)
4066 			bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA;
4067 
4068 		if (mp->m_dalign && bma->length >= mp->m_dalign) {
4069 			error = xfs_bmap_isaeof(bma, whichfork);
4070 			if (error)
4071 				return error;
4072 		}
4073 
4074 		if (XFS_IS_REALTIME_INODE(bma->ip))
4075 			return xfs_bmap_rtalloc(bma);
4076 	}
4077 
4078 	if (unlikely(XFS_TEST_ERROR(false, mp,
4079 			XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT)))
4080 		return xfs_bmap_exact_minlen_extent_alloc(bma);
4081 
4082 	return xfs_bmap_btalloc(bma);
4083 }
4084 
4085 static int
4086 xfs_bmapi_allocate(
4087 	struct xfs_bmalloca	*bma)
4088 {
4089 	struct xfs_mount	*mp = bma->ip->i_mount;
4090 	int			whichfork = xfs_bmapi_whichfork(bma->flags);
4091 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4092 	int			tmp_logflags = 0;
4093 	int			error;
4094 
4095 	ASSERT(bma->length > 0);
4096 
4097 	/*
4098 	 * For the wasdelay case, we could also just allocate the stuff asked
4099 	 * for in this bmap call but that wouldn't be as good.
4100 	 */
4101 	if (bma->wasdel) {
4102 		bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4103 		bma->offset = bma->got.br_startoff;
4104 		if (!xfs_iext_peek_prev_extent(ifp, &bma->icur, &bma->prev))
4105 			bma->prev.br_startoff = NULLFILEOFF;
4106 	} else {
4107 		bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4108 		if (!bma->eof)
4109 			bma->length = XFS_FILBLKS_MIN(bma->length,
4110 					bma->got.br_startoff - bma->offset);
4111 	}
4112 
4113 	if (bma->flags & XFS_BMAPI_CONTIG)
4114 		bma->minlen = bma->length;
4115 	else
4116 		bma->minlen = 1;
4117 
4118 	if (bma->flags & XFS_BMAPI_METADATA) {
4119 		if (unlikely(XFS_TEST_ERROR(false, mp,
4120 				XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT)))
4121 			error = xfs_bmap_exact_minlen_extent_alloc(bma);
4122 		else
4123 			error = xfs_bmap_btalloc(bma);
4124 	} else {
4125 		error = xfs_bmap_alloc_userdata(bma);
4126 	}
4127 	if (error || bma->blkno == NULLFSBLOCK)
4128 		return error;
4129 
4130 	if (bma->flags & XFS_BMAPI_ZERO) {
4131 		error = xfs_zero_extent(bma->ip, bma->blkno, bma->length);
4132 		if (error)
4133 			return error;
4134 	}
4135 
4136 	if (ifp->if_format == XFS_DINODE_FMT_BTREE && !bma->cur)
4137 		bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4138 	/*
4139 	 * Bump the number of extents we've allocated
4140 	 * in this call.
4141 	 */
4142 	bma->nallocs++;
4143 
4144 	if (bma->cur)
4145 		bma->cur->bc_ino.flags =
4146 			bma->wasdel ? XFS_BTCUR_BMBT_WASDEL : 0;
4147 
4148 	bma->got.br_startoff = bma->offset;
4149 	bma->got.br_startblock = bma->blkno;
4150 	bma->got.br_blockcount = bma->length;
4151 	bma->got.br_state = XFS_EXT_NORM;
4152 
4153 	if (bma->flags & XFS_BMAPI_PREALLOC)
4154 		bma->got.br_state = XFS_EXT_UNWRITTEN;
4155 
4156 	if (bma->wasdel)
4157 		error = xfs_bmap_add_extent_delay_real(bma, whichfork);
4158 	else
4159 		error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip,
4160 				whichfork, &bma->icur, &bma->cur, &bma->got,
4161 				&bma->logflags, bma->flags);
4162 
4163 	bma->logflags |= tmp_logflags;
4164 	if (error)
4165 		return error;
4166 
4167 	/*
4168 	 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4169 	 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4170 	 * the neighbouring ones.
4171 	 */
4172 	xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4173 
4174 	ASSERT(bma->got.br_startoff <= bma->offset);
4175 	ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4176 	       bma->offset + bma->length);
4177 	ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4178 	       bma->got.br_state == XFS_EXT_UNWRITTEN);
4179 	return 0;
4180 }
4181 
4182 STATIC int
4183 xfs_bmapi_convert_unwritten(
4184 	struct xfs_bmalloca	*bma,
4185 	struct xfs_bmbt_irec	*mval,
4186 	xfs_filblks_t		len,
4187 	int			flags)
4188 {
4189 	int			whichfork = xfs_bmapi_whichfork(flags);
4190 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4191 	int			tmp_logflags = 0;
4192 	int			error;
4193 
4194 	/* check if we need to do unwritten->real conversion */
4195 	if (mval->br_state == XFS_EXT_UNWRITTEN &&
4196 	    (flags & XFS_BMAPI_PREALLOC))
4197 		return 0;
4198 
4199 	/* check if we need to do real->unwritten conversion */
4200 	if (mval->br_state == XFS_EXT_NORM &&
4201 	    (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4202 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4203 		return 0;
4204 
4205 	/*
4206 	 * Modify (by adding) the state flag, if writing.
4207 	 */
4208 	ASSERT(mval->br_blockcount <= len);
4209 	if (ifp->if_format == XFS_DINODE_FMT_BTREE && !bma->cur) {
4210 		bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4211 					bma->ip, whichfork);
4212 	}
4213 	mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4214 				? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4215 
4216 	/*
4217 	 * Before insertion into the bmbt, zero the range being converted
4218 	 * if required.
4219 	 */
4220 	if (flags & XFS_BMAPI_ZERO) {
4221 		error = xfs_zero_extent(bma->ip, mval->br_startblock,
4222 					mval->br_blockcount);
4223 		if (error)
4224 			return error;
4225 	}
4226 
4227 	error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork,
4228 			&bma->icur, &bma->cur, mval, &tmp_logflags);
4229 	/*
4230 	 * Log the inode core unconditionally in the unwritten extent conversion
4231 	 * path because the conversion might not have done so (e.g., if the
4232 	 * extent count hasn't changed). We need to make sure the inode is dirty
4233 	 * in the transaction for the sake of fsync(), even if nothing has
4234 	 * changed, because fsync() will not force the log for this transaction
4235 	 * unless it sees the inode pinned.
4236 	 *
4237 	 * Note: If we're only converting cow fork extents, there aren't
4238 	 * any on-disk updates to make, so we don't need to log anything.
4239 	 */
4240 	if (whichfork != XFS_COW_FORK)
4241 		bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4242 	if (error)
4243 		return error;
4244 
4245 	/*
4246 	 * Update our extent pointer, given that
4247 	 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4248 	 * of the neighbouring ones.
4249 	 */
4250 	xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4251 
4252 	/*
4253 	 * We may have combined previously unwritten space with written space,
4254 	 * so generate another request.
4255 	 */
4256 	if (mval->br_blockcount < len)
4257 		return -EAGAIN;
4258 	return 0;
4259 }
4260 
4261 static inline xfs_extlen_t
4262 xfs_bmapi_minleft(
4263 	struct xfs_trans	*tp,
4264 	struct xfs_inode	*ip,
4265 	int			fork)
4266 {
4267 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, fork);
4268 
4269 	if (tp && tp->t_firstblock != NULLFSBLOCK)
4270 		return 0;
4271 	if (ifp->if_format != XFS_DINODE_FMT_BTREE)
4272 		return 1;
4273 	return be16_to_cpu(ifp->if_broot->bb_level) + 1;
4274 }
4275 
4276 /*
4277  * Log whatever the flags say, even if error.  Otherwise we might miss detecting
4278  * a case where the data is changed, there's an error, and it's not logged so we
4279  * don't shutdown when we should.  Don't bother logging extents/btree changes if
4280  * we converted to the other format.
4281  */
4282 static void
4283 xfs_bmapi_finish(
4284 	struct xfs_bmalloca	*bma,
4285 	int			whichfork,
4286 	int			error)
4287 {
4288 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4289 
4290 	if ((bma->logflags & xfs_ilog_fext(whichfork)) &&
4291 	    ifp->if_format != XFS_DINODE_FMT_EXTENTS)
4292 		bma->logflags &= ~xfs_ilog_fext(whichfork);
4293 	else if ((bma->logflags & xfs_ilog_fbroot(whichfork)) &&
4294 		 ifp->if_format != XFS_DINODE_FMT_BTREE)
4295 		bma->logflags &= ~xfs_ilog_fbroot(whichfork);
4296 
4297 	if (bma->logflags)
4298 		xfs_trans_log_inode(bma->tp, bma->ip, bma->logflags);
4299 	if (bma->cur)
4300 		xfs_btree_del_cursor(bma->cur, error);
4301 }
4302 
4303 /*
4304  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4305  * extent state if necessary.  Details behaviour is controlled by the flags
4306  * parameter.  Only allocates blocks from a single allocation group, to avoid
4307  * locking problems.
4308  */
4309 int
4310 xfs_bmapi_write(
4311 	struct xfs_trans	*tp,		/* transaction pointer */
4312 	struct xfs_inode	*ip,		/* incore inode */
4313 	xfs_fileoff_t		bno,		/* starting file offs. mapped */
4314 	xfs_filblks_t		len,		/* length to map in file */
4315 	int			flags,		/* XFS_BMAPI_... */
4316 	xfs_extlen_t		total,		/* total blocks needed */
4317 	struct xfs_bmbt_irec	*mval,		/* output: map values */
4318 	int			*nmap)		/* i/o: mval size/count */
4319 {
4320 	struct xfs_bmalloca	bma = {
4321 		.tp		= tp,
4322 		.ip		= ip,
4323 		.total		= total,
4324 	};
4325 	struct xfs_mount	*mp = ip->i_mount;
4326 	int			whichfork = xfs_bmapi_whichfork(flags);
4327 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
4328 	xfs_fileoff_t		end;		/* end of mapped file region */
4329 	bool			eof = false;	/* after the end of extents */
4330 	int			error;		/* error return */
4331 	int			n;		/* current extent index */
4332 	xfs_fileoff_t		obno;		/* old block number (offset) */
4333 
4334 #ifdef DEBUG
4335 	xfs_fileoff_t		orig_bno;	/* original block number value */
4336 	int			orig_flags;	/* original flags arg value */
4337 	xfs_filblks_t		orig_len;	/* original value of len arg */
4338 	struct xfs_bmbt_irec	*orig_mval;	/* original value of mval */
4339 	int			orig_nmap;	/* original value of *nmap */
4340 
4341 	orig_bno = bno;
4342 	orig_len = len;
4343 	orig_flags = flags;
4344 	orig_mval = mval;
4345 	orig_nmap = *nmap;
4346 #endif
4347 
4348 	ASSERT(*nmap >= 1);
4349 	ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4350 	ASSERT(tp != NULL);
4351 	ASSERT(len > 0);
4352 	ASSERT(ifp->if_format != XFS_DINODE_FMT_LOCAL);
4353 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4354 	ASSERT(!(flags & XFS_BMAPI_REMAP));
4355 
4356 	/* zeroing is for currently only for data extents, not metadata */
4357 	ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4358 			(XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4359 	/*
4360 	 * we can allocate unwritten extents or pre-zero allocated blocks,
4361 	 * but it makes no sense to do both at once. This would result in
4362 	 * zeroing the unwritten extent twice, but it still being an
4363 	 * unwritten extent....
4364 	 */
4365 	ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4366 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4367 
4368 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
4369 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
4370 		return -EFSCORRUPTED;
4371 	}
4372 
4373 	if (xfs_is_shutdown(mp))
4374 		return -EIO;
4375 
4376 	XFS_STATS_INC(mp, xs_blk_mapw);
4377 
4378 	error = xfs_iread_extents(tp, ip, whichfork);
4379 	if (error)
4380 		goto error0;
4381 
4382 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.icur, &bma.got))
4383 		eof = true;
4384 	if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4385 		bma.prev.br_startoff = NULLFILEOFF;
4386 	bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4387 
4388 	n = 0;
4389 	end = bno + len;
4390 	obno = bno;
4391 	while (bno < end && n < *nmap) {
4392 		bool			need_alloc = false, wasdelay = false;
4393 
4394 		/* in hole or beyond EOF? */
4395 		if (eof || bma.got.br_startoff > bno) {
4396 			/*
4397 			 * CoW fork conversions should /never/ hit EOF or
4398 			 * holes.  There should always be something for us
4399 			 * to work on.
4400 			 */
4401 			ASSERT(!((flags & XFS_BMAPI_CONVERT) &&
4402 			         (flags & XFS_BMAPI_COWFORK)));
4403 
4404 			need_alloc = true;
4405 		} else if (isnullstartblock(bma.got.br_startblock)) {
4406 			wasdelay = true;
4407 		}
4408 
4409 		/*
4410 		 * First, deal with the hole before the allocated space
4411 		 * that we found, if any.
4412 		 */
4413 		if (need_alloc || wasdelay) {
4414 			bma.eof = eof;
4415 			bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4416 			bma.wasdel = wasdelay;
4417 			bma.offset = bno;
4418 			bma.flags = flags;
4419 
4420 			/*
4421 			 * There's a 32/64 bit type mismatch between the
4422 			 * allocation length request (which can be 64 bits in
4423 			 * length) and the bma length request, which is
4424 			 * xfs_extlen_t and therefore 32 bits. Hence we have to
4425 			 * check for 32-bit overflows and handle them here.
4426 			 */
4427 			if (len > (xfs_filblks_t)MAXEXTLEN)
4428 				bma.length = MAXEXTLEN;
4429 			else
4430 				bma.length = len;
4431 
4432 			ASSERT(len > 0);
4433 			ASSERT(bma.length > 0);
4434 			error = xfs_bmapi_allocate(&bma);
4435 			if (error)
4436 				goto error0;
4437 			if (bma.blkno == NULLFSBLOCK)
4438 				break;
4439 
4440 			/*
4441 			 * If this is a CoW allocation, record the data in
4442 			 * the refcount btree for orphan recovery.
4443 			 */
4444 			if (whichfork == XFS_COW_FORK)
4445 				xfs_refcount_alloc_cow_extent(tp, bma.blkno,
4446 						bma.length);
4447 		}
4448 
4449 		/* Deal with the allocated space we found.  */
4450 		xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4451 							end, n, flags);
4452 
4453 		/* Execute unwritten extent conversion if necessary */
4454 		error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4455 		if (error == -EAGAIN)
4456 			continue;
4457 		if (error)
4458 			goto error0;
4459 
4460 		/* update the extent map to return */
4461 		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4462 
4463 		/*
4464 		 * If we're done, stop now.  Stop when we've allocated
4465 		 * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4466 		 * the transaction may get too big.
4467 		 */
4468 		if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4469 			break;
4470 
4471 		/* Else go on to the next record. */
4472 		bma.prev = bma.got;
4473 		if (!xfs_iext_next_extent(ifp, &bma.icur, &bma.got))
4474 			eof = true;
4475 	}
4476 	*nmap = n;
4477 
4478 	error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4479 			whichfork);
4480 	if (error)
4481 		goto error0;
4482 
4483 	ASSERT(ifp->if_format != XFS_DINODE_FMT_BTREE ||
4484 	       ifp->if_nextents > XFS_IFORK_MAXEXT(ip, whichfork));
4485 	xfs_bmapi_finish(&bma, whichfork, 0);
4486 	xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4487 		orig_nmap, *nmap);
4488 	return 0;
4489 error0:
4490 	xfs_bmapi_finish(&bma, whichfork, error);
4491 	return error;
4492 }
4493 
4494 /*
4495  * Convert an existing delalloc extent to real blocks based on file offset. This
4496  * attempts to allocate the entire delalloc extent and may require multiple
4497  * invocations to allocate the target offset if a large enough physical extent
4498  * is not available.
4499  */
4500 int
4501 xfs_bmapi_convert_delalloc(
4502 	struct xfs_inode	*ip,
4503 	int			whichfork,
4504 	xfs_off_t		offset,
4505 	struct iomap		*iomap,
4506 	unsigned int		*seq)
4507 {
4508 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
4509 	struct xfs_mount	*mp = ip->i_mount;
4510 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
4511 	struct xfs_bmalloca	bma = { NULL };
4512 	uint16_t		flags = 0;
4513 	struct xfs_trans	*tp;
4514 	int			error;
4515 
4516 	if (whichfork == XFS_COW_FORK)
4517 		flags |= IOMAP_F_SHARED;
4518 
4519 	/*
4520 	 * Space for the extent and indirect blocks was reserved when the
4521 	 * delalloc extent was created so there's no need to do so here.
4522 	 */
4523 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0,
4524 				XFS_TRANS_RESERVE, &tp);
4525 	if (error)
4526 		return error;
4527 
4528 	xfs_ilock(ip, XFS_ILOCK_EXCL);
4529 
4530 	error = xfs_iext_count_may_overflow(ip, whichfork,
4531 			XFS_IEXT_ADD_NOSPLIT_CNT);
4532 	if (error)
4533 		goto out_trans_cancel;
4534 
4535 	xfs_trans_ijoin(tp, ip, 0);
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, 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, MAXEXTLEN);
4564 	bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4565 
4566 	/*
4567 	 * When we're converting the delalloc reservations backing dirty pages
4568 	 * in the page cache, we must be careful about how we create the new
4569 	 * extents:
4570 	 *
4571 	 * New CoW fork extents are created unwritten, turned into real extents
4572 	 * when we're about to write the data to disk, and mapped into the data
4573 	 * fork after the write finishes.  End of story.
4574 	 *
4575 	 * New data fork extents must be mapped in as unwritten and converted
4576 	 * to real extents after the write succeeds to avoid exposing stale
4577 	 * disk contents if we crash.
4578 	 */
4579 	bma.flags = XFS_BMAPI_PREALLOC;
4580 	if (whichfork == XFS_COW_FORK)
4581 		bma.flags |= XFS_BMAPI_COWFORK;
4582 
4583 	if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4584 		bma.prev.br_startoff = NULLFILEOFF;
4585 
4586 	error = xfs_bmapi_allocate(&bma);
4587 	if (error)
4588 		goto out_finish;
4589 
4590 	error = -ENOSPC;
4591 	if (WARN_ON_ONCE(bma.blkno == NULLFSBLOCK))
4592 		goto out_finish;
4593 	error = -EFSCORRUPTED;
4594 	if (WARN_ON_ONCE(!xfs_valid_startblock(ip, bma.got.br_startblock)))
4595 		goto out_finish;
4596 
4597 	XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, bma.length));
4598 	XFS_STATS_INC(mp, xs_xstrat_quick);
4599 
4600 	ASSERT(!isnullstartblock(bma.got.br_startblock));
4601 	xfs_bmbt_to_iomap(ip, iomap, &bma.got, flags);
4602 	*seq = READ_ONCE(ifp->if_seq);
4603 
4604 	if (whichfork == XFS_COW_FORK)
4605 		xfs_refcount_alloc_cow_extent(tp, bma.blkno, bma.length);
4606 
4607 	error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4608 			whichfork);
4609 	if (error)
4610 		goto out_finish;
4611 
4612 	xfs_bmapi_finish(&bma, whichfork, 0);
4613 	error = xfs_trans_commit(tp);
4614 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
4615 	return error;
4616 
4617 out_finish:
4618 	xfs_bmapi_finish(&bma, whichfork, error);
4619 out_trans_cancel:
4620 	xfs_trans_cancel(tp);
4621 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
4622 	return error;
4623 }
4624 
4625 int
4626 xfs_bmapi_remap(
4627 	struct xfs_trans	*tp,
4628 	struct xfs_inode	*ip,
4629 	xfs_fileoff_t		bno,
4630 	xfs_filblks_t		len,
4631 	xfs_fsblock_t		startblock,
4632 	int			flags)
4633 {
4634 	struct xfs_mount	*mp = ip->i_mount;
4635 	struct xfs_ifork	*ifp;
4636 	struct xfs_btree_cur	*cur = NULL;
4637 	struct xfs_bmbt_irec	got;
4638 	struct xfs_iext_cursor	icur;
4639 	int			whichfork = xfs_bmapi_whichfork(flags);
4640 	int			logflags = 0, error;
4641 
4642 	ifp = XFS_IFORK_PTR(ip, whichfork);
4643 	ASSERT(len > 0);
4644 	ASSERT(len <= (xfs_filblks_t)MAXEXTLEN);
4645 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4646 	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC |
4647 			   XFS_BMAPI_NORMAP)));
4648 	ASSERT((flags & (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC)) !=
4649 			(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC));
4650 
4651 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
4652 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
4653 		return -EFSCORRUPTED;
4654 	}
4655 
4656 	if (xfs_is_shutdown(mp))
4657 		return -EIO;
4658 
4659 	error = xfs_iread_extents(tp, ip, whichfork);
4660 	if (error)
4661 		return error;
4662 
4663 	if (xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
4664 		/* make sure we only reflink into a hole. */
4665 		ASSERT(got.br_startoff > bno);
4666 		ASSERT(got.br_startoff - bno >= len);
4667 	}
4668 
4669 	ip->i_nblocks += len;
4670 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4671 
4672 	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
4673 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
4674 		cur->bc_ino.flags = 0;
4675 	}
4676 
4677 	got.br_startoff = bno;
4678 	got.br_startblock = startblock;
4679 	got.br_blockcount = len;
4680 	if (flags & XFS_BMAPI_PREALLOC)
4681 		got.br_state = XFS_EXT_UNWRITTEN;
4682 	else
4683 		got.br_state = XFS_EXT_NORM;
4684 
4685 	error = xfs_bmap_add_extent_hole_real(tp, ip, whichfork, &icur,
4686 			&cur, &got, &logflags, flags);
4687 	if (error)
4688 		goto error0;
4689 
4690 	error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags, whichfork);
4691 
4692 error0:
4693 	if (ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS)
4694 		logflags &= ~XFS_ILOG_DEXT;
4695 	else if (ip->i_df.if_format != XFS_DINODE_FMT_BTREE)
4696 		logflags &= ~XFS_ILOG_DBROOT;
4697 
4698 	if (logflags)
4699 		xfs_trans_log_inode(tp, ip, logflags);
4700 	if (cur)
4701 		xfs_btree_del_cursor(cur, error);
4702 	return error;
4703 }
4704 
4705 /*
4706  * When a delalloc extent is split (e.g., due to a hole punch), the original
4707  * indlen reservation must be shared across the two new extents that are left
4708  * behind.
4709  *
4710  * Given the original reservation and the worst case indlen for the two new
4711  * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4712  * reservation fairly across the two new extents. If necessary, steal available
4713  * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4714  * ores == 1). The number of stolen blocks is returned. The availability and
4715  * subsequent accounting of stolen blocks is the responsibility of the caller.
4716  */
4717 static xfs_filblks_t
4718 xfs_bmap_split_indlen(
4719 	xfs_filblks_t			ores,		/* original res. */
4720 	xfs_filblks_t			*indlen1,	/* ext1 worst indlen */
4721 	xfs_filblks_t			*indlen2,	/* ext2 worst indlen */
4722 	xfs_filblks_t			avail)		/* stealable blocks */
4723 {
4724 	xfs_filblks_t			len1 = *indlen1;
4725 	xfs_filblks_t			len2 = *indlen2;
4726 	xfs_filblks_t			nres = len1 + len2; /* new total res. */
4727 	xfs_filblks_t			stolen = 0;
4728 	xfs_filblks_t			resfactor;
4729 
4730 	/*
4731 	 * Steal as many blocks as we can to try and satisfy the worst case
4732 	 * indlen for both new extents.
4733 	 */
4734 	if (ores < nres && avail)
4735 		stolen = XFS_FILBLKS_MIN(nres - ores, avail);
4736 	ores += stolen;
4737 
4738 	 /* nothing else to do if we've satisfied the new reservation */
4739 	if (ores >= nres)
4740 		return stolen;
4741 
4742 	/*
4743 	 * We can't meet the total required reservation for the two extents.
4744 	 * Calculate the percent of the overall shortage between both extents
4745 	 * and apply this percentage to each of the requested indlen values.
4746 	 * This distributes the shortage fairly and reduces the chances that one
4747 	 * of the two extents is left with nothing when extents are repeatedly
4748 	 * split.
4749 	 */
4750 	resfactor = (ores * 100);
4751 	do_div(resfactor, nres);
4752 	len1 *= resfactor;
4753 	do_div(len1, 100);
4754 	len2 *= resfactor;
4755 	do_div(len2, 100);
4756 	ASSERT(len1 + len2 <= ores);
4757 	ASSERT(len1 < *indlen1 && len2 < *indlen2);
4758 
4759 	/*
4760 	 * Hand out the remainder to each extent. If one of the two reservations
4761 	 * is zero, we want to make sure that one gets a block first. The loop
4762 	 * below starts with len1, so hand len2 a block right off the bat if it
4763 	 * is zero.
4764 	 */
4765 	ores -= (len1 + len2);
4766 	ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores);
4767 	if (ores && !len2 && *indlen2) {
4768 		len2++;
4769 		ores--;
4770 	}
4771 	while (ores) {
4772 		if (len1 < *indlen1) {
4773 			len1++;
4774 			ores--;
4775 		}
4776 		if (!ores)
4777 			break;
4778 		if (len2 < *indlen2) {
4779 			len2++;
4780 			ores--;
4781 		}
4782 	}
4783 
4784 	*indlen1 = len1;
4785 	*indlen2 = len2;
4786 
4787 	return stolen;
4788 }
4789 
4790 int
4791 xfs_bmap_del_extent_delay(
4792 	struct xfs_inode	*ip,
4793 	int			whichfork,
4794 	struct xfs_iext_cursor	*icur,
4795 	struct xfs_bmbt_irec	*got,
4796 	struct xfs_bmbt_irec	*del)
4797 {
4798 	struct xfs_mount	*mp = ip->i_mount;
4799 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
4800 	struct xfs_bmbt_irec	new;
4801 	int64_t			da_old, da_new, da_diff = 0;
4802 	xfs_fileoff_t		del_endoff, got_endoff;
4803 	xfs_filblks_t		got_indlen, new_indlen, stolen;
4804 	int			state = xfs_bmap_fork_to_state(whichfork);
4805 	int			error = 0;
4806 	bool			isrt;
4807 
4808 	XFS_STATS_INC(mp, xs_del_exlist);
4809 
4810 	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4811 	del_endoff = del->br_startoff + del->br_blockcount;
4812 	got_endoff = got->br_startoff + got->br_blockcount;
4813 	da_old = startblockval(got->br_startblock);
4814 	da_new = 0;
4815 
4816 	ASSERT(del->br_blockcount > 0);
4817 	ASSERT(got->br_startoff <= del->br_startoff);
4818 	ASSERT(got_endoff >= del_endoff);
4819 
4820 	if (isrt) {
4821 		uint64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount);
4822 
4823 		do_div(rtexts, mp->m_sb.sb_rextsize);
4824 		xfs_mod_frextents(mp, rtexts);
4825 	}
4826 
4827 	/*
4828 	 * Update the inode delalloc counter now and wait to update the
4829 	 * sb counters as we might have to borrow some blocks for the
4830 	 * indirect block accounting.
4831 	 */
4832 	ASSERT(!isrt);
4833 	error = xfs_quota_unreserve_blkres(ip, del->br_blockcount);
4834 	if (error)
4835 		return error;
4836 	ip->i_delayed_blks -= del->br_blockcount;
4837 
4838 	if (got->br_startoff == del->br_startoff)
4839 		state |= BMAP_LEFT_FILLING;
4840 	if (got_endoff == del_endoff)
4841 		state |= BMAP_RIGHT_FILLING;
4842 
4843 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4844 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4845 		/*
4846 		 * Matches the whole extent.  Delete the entry.
4847 		 */
4848 		xfs_iext_remove(ip, icur, state);
4849 		xfs_iext_prev(ifp, icur);
4850 		break;
4851 	case BMAP_LEFT_FILLING:
4852 		/*
4853 		 * Deleting the first part of the extent.
4854 		 */
4855 		got->br_startoff = del_endoff;
4856 		got->br_blockcount -= del->br_blockcount;
4857 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4858 				got->br_blockcount), da_old);
4859 		got->br_startblock = nullstartblock((int)da_new);
4860 		xfs_iext_update_extent(ip, state, icur, got);
4861 		break;
4862 	case BMAP_RIGHT_FILLING:
4863 		/*
4864 		 * Deleting the last part of the extent.
4865 		 */
4866 		got->br_blockcount = got->br_blockcount - del->br_blockcount;
4867 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4868 				got->br_blockcount), da_old);
4869 		got->br_startblock = nullstartblock((int)da_new);
4870 		xfs_iext_update_extent(ip, state, icur, got);
4871 		break;
4872 	case 0:
4873 		/*
4874 		 * Deleting the middle of the extent.
4875 		 *
4876 		 * Distribute the original indlen reservation across the two new
4877 		 * extents.  Steal blocks from the deleted extent if necessary.
4878 		 * Stealing blocks simply fudges the fdblocks accounting below.
4879 		 * Warn if either of the new indlen reservations is zero as this
4880 		 * can lead to delalloc problems.
4881 		 */
4882 		got->br_blockcount = del->br_startoff - got->br_startoff;
4883 		got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount);
4884 
4885 		new.br_blockcount = got_endoff - del_endoff;
4886 		new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount);
4887 
4888 		WARN_ON_ONCE(!got_indlen || !new_indlen);
4889 		stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen,
4890 						       del->br_blockcount);
4891 
4892 		got->br_startblock = nullstartblock((int)got_indlen);
4893 
4894 		new.br_startoff = del_endoff;
4895 		new.br_state = got->br_state;
4896 		new.br_startblock = nullstartblock((int)new_indlen);
4897 
4898 		xfs_iext_update_extent(ip, state, icur, got);
4899 		xfs_iext_next(ifp, icur);
4900 		xfs_iext_insert(ip, icur, &new, state);
4901 
4902 		da_new = got_indlen + new_indlen - stolen;
4903 		del->br_blockcount -= stolen;
4904 		break;
4905 	}
4906 
4907 	ASSERT(da_old >= da_new);
4908 	da_diff = da_old - da_new;
4909 	if (!isrt)
4910 		da_diff += del->br_blockcount;
4911 	if (da_diff) {
4912 		xfs_mod_fdblocks(mp, da_diff, false);
4913 		xfs_mod_delalloc(mp, -da_diff);
4914 	}
4915 	return error;
4916 }
4917 
4918 void
4919 xfs_bmap_del_extent_cow(
4920 	struct xfs_inode	*ip,
4921 	struct xfs_iext_cursor	*icur,
4922 	struct xfs_bmbt_irec	*got,
4923 	struct xfs_bmbt_irec	*del)
4924 {
4925 	struct xfs_mount	*mp = ip->i_mount;
4926 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
4927 	struct xfs_bmbt_irec	new;
4928 	xfs_fileoff_t		del_endoff, got_endoff;
4929 	int			state = BMAP_COWFORK;
4930 
4931 	XFS_STATS_INC(mp, xs_del_exlist);
4932 
4933 	del_endoff = del->br_startoff + del->br_blockcount;
4934 	got_endoff = got->br_startoff + got->br_blockcount;
4935 
4936 	ASSERT(del->br_blockcount > 0);
4937 	ASSERT(got->br_startoff <= del->br_startoff);
4938 	ASSERT(got_endoff >= del_endoff);
4939 	ASSERT(!isnullstartblock(got->br_startblock));
4940 
4941 	if (got->br_startoff == del->br_startoff)
4942 		state |= BMAP_LEFT_FILLING;
4943 	if (got_endoff == del_endoff)
4944 		state |= BMAP_RIGHT_FILLING;
4945 
4946 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4947 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4948 		/*
4949 		 * Matches the whole extent.  Delete the entry.
4950 		 */
4951 		xfs_iext_remove(ip, icur, state);
4952 		xfs_iext_prev(ifp, icur);
4953 		break;
4954 	case BMAP_LEFT_FILLING:
4955 		/*
4956 		 * Deleting the first part of the extent.
4957 		 */
4958 		got->br_startoff = del_endoff;
4959 		got->br_blockcount -= del->br_blockcount;
4960 		got->br_startblock = del->br_startblock + del->br_blockcount;
4961 		xfs_iext_update_extent(ip, state, icur, got);
4962 		break;
4963 	case BMAP_RIGHT_FILLING:
4964 		/*
4965 		 * Deleting the last part of the extent.
4966 		 */
4967 		got->br_blockcount -= del->br_blockcount;
4968 		xfs_iext_update_extent(ip, state, icur, got);
4969 		break;
4970 	case 0:
4971 		/*
4972 		 * Deleting the middle of the extent.
4973 		 */
4974 		got->br_blockcount = del->br_startoff - got->br_startoff;
4975 
4976 		new.br_startoff = del_endoff;
4977 		new.br_blockcount = got_endoff - del_endoff;
4978 		new.br_state = got->br_state;
4979 		new.br_startblock = del->br_startblock + del->br_blockcount;
4980 
4981 		xfs_iext_update_extent(ip, state, icur, got);
4982 		xfs_iext_next(ifp, icur);
4983 		xfs_iext_insert(ip, icur, &new, state);
4984 		break;
4985 	}
4986 	ip->i_delayed_blks -= del->br_blockcount;
4987 }
4988 
4989 /*
4990  * Called by xfs_bmapi to update file extent records and the btree
4991  * after removing space.
4992  */
4993 STATIC int				/* error */
4994 xfs_bmap_del_extent_real(
4995 	xfs_inode_t		*ip,	/* incore inode pointer */
4996 	xfs_trans_t		*tp,	/* current transaction pointer */
4997 	struct xfs_iext_cursor	*icur,
4998 	struct xfs_btree_cur	*cur,	/* if null, not a btree */
4999 	xfs_bmbt_irec_t		*del,	/* data to remove from extents */
5000 	int			*logflagsp, /* inode logging flags */
5001 	int			whichfork, /* data or attr fork */
5002 	int			bflags)	/* bmapi flags */
5003 {
5004 	xfs_fsblock_t		del_endblock=0;	/* first block past del */
5005 	xfs_fileoff_t		del_endoff;	/* first offset past del */
5006 	int			do_fx;	/* free extent at end of routine */
5007 	int			error;	/* error return value */
5008 	int			flags = 0;/* inode logging flags */
5009 	struct xfs_bmbt_irec	got;	/* current extent entry */
5010 	xfs_fileoff_t		got_endoff;	/* first offset past got */
5011 	int			i;	/* temp state */
5012 	struct xfs_ifork	*ifp;	/* inode fork pointer */
5013 	xfs_mount_t		*mp;	/* mount structure */
5014 	xfs_filblks_t		nblks;	/* quota/sb block count */
5015 	xfs_bmbt_irec_t		new;	/* new record to be inserted */
5016 	/* REFERENCED */
5017 	uint			qfield;	/* quota field to update */
5018 	int			state = xfs_bmap_fork_to_state(whichfork);
5019 	struct xfs_bmbt_irec	old;
5020 
5021 	mp = ip->i_mount;
5022 	XFS_STATS_INC(mp, xs_del_exlist);
5023 
5024 	ifp = XFS_IFORK_PTR(ip, whichfork);
5025 	ASSERT(del->br_blockcount > 0);
5026 	xfs_iext_get_extent(ifp, icur, &got);
5027 	ASSERT(got.br_startoff <= del->br_startoff);
5028 	del_endoff = del->br_startoff + del->br_blockcount;
5029 	got_endoff = got.br_startoff + got.br_blockcount;
5030 	ASSERT(got_endoff >= del_endoff);
5031 	ASSERT(!isnullstartblock(got.br_startblock));
5032 	qfield = 0;
5033 	error = 0;
5034 
5035 	/*
5036 	 * If it's the case where the directory code is running with no block
5037 	 * reservation, and the deleted block is in the middle of its extent,
5038 	 * and the resulting insert of an extent would cause transformation to
5039 	 * btree format, then reject it.  The calling code will then swap blocks
5040 	 * around instead.  We have to do this now, rather than waiting for the
5041 	 * conversion to btree format, since the transaction will be dirty then.
5042 	 */
5043 	if (tp->t_blk_res == 0 &&
5044 	    ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
5045 	    ifp->if_nextents >= XFS_IFORK_MAXEXT(ip, whichfork) &&
5046 	    del->br_startoff > got.br_startoff && del_endoff < got_endoff)
5047 		return -ENOSPC;
5048 
5049 	flags = XFS_ILOG_CORE;
5050 	if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
5051 		xfs_filblks_t	len;
5052 		xfs_extlen_t	mod;
5053 
5054 		len = div_u64_rem(del->br_blockcount, mp->m_sb.sb_rextsize,
5055 				  &mod);
5056 		ASSERT(mod == 0);
5057 
5058 		if (!(bflags & XFS_BMAPI_REMAP)) {
5059 			xfs_fsblock_t	bno;
5060 
5061 			bno = div_u64_rem(del->br_startblock,
5062 					mp->m_sb.sb_rextsize, &mod);
5063 			ASSERT(mod == 0);
5064 
5065 			error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
5066 			if (error)
5067 				goto done;
5068 		}
5069 
5070 		do_fx = 0;
5071 		nblks = len * mp->m_sb.sb_rextsize;
5072 		qfield = XFS_TRANS_DQ_RTBCOUNT;
5073 	} else {
5074 		do_fx = 1;
5075 		nblks = del->br_blockcount;
5076 		qfield = XFS_TRANS_DQ_BCOUNT;
5077 	}
5078 
5079 	del_endblock = del->br_startblock + del->br_blockcount;
5080 	if (cur) {
5081 		error = xfs_bmbt_lookup_eq(cur, &got, &i);
5082 		if (error)
5083 			goto done;
5084 		if (XFS_IS_CORRUPT(mp, i != 1)) {
5085 			error = -EFSCORRUPTED;
5086 			goto done;
5087 		}
5088 	}
5089 
5090 	if (got.br_startoff == del->br_startoff)
5091 		state |= BMAP_LEFT_FILLING;
5092 	if (got_endoff == del_endoff)
5093 		state |= BMAP_RIGHT_FILLING;
5094 
5095 	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
5096 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
5097 		/*
5098 		 * Matches the whole extent.  Delete the entry.
5099 		 */
5100 		xfs_iext_remove(ip, icur, state);
5101 		xfs_iext_prev(ifp, icur);
5102 		ifp->if_nextents--;
5103 
5104 		flags |= XFS_ILOG_CORE;
5105 		if (!cur) {
5106 			flags |= xfs_ilog_fext(whichfork);
5107 			break;
5108 		}
5109 		if ((error = xfs_btree_delete(cur, &i)))
5110 			goto done;
5111 		if (XFS_IS_CORRUPT(mp, i != 1)) {
5112 			error = -EFSCORRUPTED;
5113 			goto done;
5114 		}
5115 		break;
5116 	case BMAP_LEFT_FILLING:
5117 		/*
5118 		 * Deleting the first part of the extent.
5119 		 */
5120 		got.br_startoff = del_endoff;
5121 		got.br_startblock = del_endblock;
5122 		got.br_blockcount -= del->br_blockcount;
5123 		xfs_iext_update_extent(ip, state, icur, &got);
5124 		if (!cur) {
5125 			flags |= xfs_ilog_fext(whichfork);
5126 			break;
5127 		}
5128 		error = xfs_bmbt_update(cur, &got);
5129 		if (error)
5130 			goto done;
5131 		break;
5132 	case BMAP_RIGHT_FILLING:
5133 		/*
5134 		 * Deleting the last part of the extent.
5135 		 */
5136 		got.br_blockcount -= del->br_blockcount;
5137 		xfs_iext_update_extent(ip, state, icur, &got);
5138 		if (!cur) {
5139 			flags |= xfs_ilog_fext(whichfork);
5140 			break;
5141 		}
5142 		error = xfs_bmbt_update(cur, &got);
5143 		if (error)
5144 			goto done;
5145 		break;
5146 	case 0:
5147 		/*
5148 		 * Deleting the middle of the extent.
5149 		 */
5150 
5151 		/*
5152 		 * For directories, -ENOSPC is returned since a directory entry
5153 		 * remove operation must not fail due to low extent count
5154 		 * availability. -ENOSPC will be handled by higher layers of XFS
5155 		 * by letting the corresponding empty Data/Free blocks to linger
5156 		 * until a future remove operation. Dabtree blocks would be
5157 		 * swapped with the last block in the leaf space and then the
5158 		 * new last block will be unmapped.
5159 		 *
5160 		 * The above logic also applies to the source directory entry of
5161 		 * a rename operation.
5162 		 */
5163 		error = xfs_iext_count_may_overflow(ip, whichfork, 1);
5164 		if (error) {
5165 			ASSERT(S_ISDIR(VFS_I(ip)->i_mode) &&
5166 				whichfork == XFS_DATA_FORK);
5167 			error = -ENOSPC;
5168 			goto done;
5169 		}
5170 
5171 		old = got;
5172 
5173 		got.br_blockcount = del->br_startoff - got.br_startoff;
5174 		xfs_iext_update_extent(ip, state, icur, &got);
5175 
5176 		new.br_startoff = del_endoff;
5177 		new.br_blockcount = got_endoff - del_endoff;
5178 		new.br_state = got.br_state;
5179 		new.br_startblock = del_endblock;
5180 
5181 		flags |= XFS_ILOG_CORE;
5182 		if (cur) {
5183 			error = xfs_bmbt_update(cur, &got);
5184 			if (error)
5185 				goto done;
5186 			error = xfs_btree_increment(cur, 0, &i);
5187 			if (error)
5188 				goto done;
5189 			cur->bc_rec.b = new;
5190 			error = xfs_btree_insert(cur, &i);
5191 			if (error && error != -ENOSPC)
5192 				goto done;
5193 			/*
5194 			 * If get no-space back from btree insert, it tried a
5195 			 * split, and we have a zero block reservation.  Fix up
5196 			 * our state and return the error.
5197 			 */
5198 			if (error == -ENOSPC) {
5199 				/*
5200 				 * Reset the cursor, don't trust it after any
5201 				 * insert operation.
5202 				 */
5203 				error = xfs_bmbt_lookup_eq(cur, &got, &i);
5204 				if (error)
5205 					goto done;
5206 				if (XFS_IS_CORRUPT(mp, i != 1)) {
5207 					error = -EFSCORRUPTED;
5208 					goto done;
5209 				}
5210 				/*
5211 				 * Update the btree record back
5212 				 * to the original value.
5213 				 */
5214 				error = xfs_bmbt_update(cur, &old);
5215 				if (error)
5216 					goto done;
5217 				/*
5218 				 * Reset the extent record back
5219 				 * to the original value.
5220 				 */
5221 				xfs_iext_update_extent(ip, state, icur, &old);
5222 				flags = 0;
5223 				error = -ENOSPC;
5224 				goto done;
5225 			}
5226 			if (XFS_IS_CORRUPT(mp, i != 1)) {
5227 				error = -EFSCORRUPTED;
5228 				goto done;
5229 			}
5230 		} else
5231 			flags |= xfs_ilog_fext(whichfork);
5232 
5233 		ifp->if_nextents++;
5234 		xfs_iext_next(ifp, icur);
5235 		xfs_iext_insert(ip, icur, &new, state);
5236 		break;
5237 	}
5238 
5239 	/* remove reverse mapping */
5240 	xfs_rmap_unmap_extent(tp, ip, whichfork, del);
5241 
5242 	/*
5243 	 * If we need to, add to list of extents to delete.
5244 	 */
5245 	if (do_fx && !(bflags & XFS_BMAPI_REMAP)) {
5246 		if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
5247 			xfs_refcount_decrease_extent(tp, del);
5248 		} else {
5249 			__xfs_free_extent_later(tp, del->br_startblock,
5250 					del->br_blockcount, NULL,
5251 					(bflags & XFS_BMAPI_NODISCARD) ||
5252 					del->br_state == XFS_EXT_UNWRITTEN);
5253 		}
5254 	}
5255 
5256 	/*
5257 	 * Adjust inode # blocks in the file.
5258 	 */
5259 	if (nblks)
5260 		ip->i_nblocks -= nblks;
5261 	/*
5262 	 * Adjust quota data.
5263 	 */
5264 	if (qfield && !(bflags & XFS_BMAPI_REMAP))
5265 		xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5266 
5267 done:
5268 	*logflagsp = flags;
5269 	return error;
5270 }
5271 
5272 /*
5273  * Unmap (remove) blocks from a file.
5274  * If nexts is nonzero then the number of extents to remove is limited to
5275  * that value.  If not all extents in the block range can be removed then
5276  * *done is set.
5277  */
5278 int						/* error */
5279 __xfs_bunmapi(
5280 	struct xfs_trans	*tp,		/* transaction pointer */
5281 	struct xfs_inode	*ip,		/* incore inode */
5282 	xfs_fileoff_t		start,		/* first file offset deleted */
5283 	xfs_filblks_t		*rlen,		/* i/o: amount remaining */
5284 	int			flags,		/* misc flags */
5285 	xfs_extnum_t		nexts)		/* number of extents max */
5286 {
5287 	struct xfs_btree_cur	*cur;		/* bmap btree cursor */
5288 	struct xfs_bmbt_irec	del;		/* extent being deleted */
5289 	int			error;		/* error return value */
5290 	xfs_extnum_t		extno;		/* extent number in list */
5291 	struct xfs_bmbt_irec	got;		/* current extent record */
5292 	struct xfs_ifork	*ifp;		/* inode fork pointer */
5293 	int			isrt;		/* freeing in rt area */
5294 	int			logflags;	/* transaction logging flags */
5295 	xfs_extlen_t		mod;		/* rt extent offset */
5296 	struct xfs_mount	*mp = ip->i_mount;
5297 	int			tmp_logflags;	/* partial logging flags */
5298 	int			wasdel;		/* was a delayed alloc extent */
5299 	int			whichfork;	/* data or attribute fork */
5300 	xfs_fsblock_t		sum;
5301 	xfs_filblks_t		len = *rlen;	/* length to unmap in file */
5302 	xfs_fileoff_t		max_len;
5303 	xfs_fileoff_t		end;
5304 	struct xfs_iext_cursor	icur;
5305 	bool			done = false;
5306 
5307 	trace_xfs_bunmap(ip, start, len, flags, _RET_IP_);
5308 
5309 	whichfork = xfs_bmapi_whichfork(flags);
5310 	ASSERT(whichfork != XFS_COW_FORK);
5311 	ifp = XFS_IFORK_PTR(ip, whichfork);
5312 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)))
5313 		return -EFSCORRUPTED;
5314 	if (xfs_is_shutdown(mp))
5315 		return -EIO;
5316 
5317 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5318 	ASSERT(len > 0);
5319 	ASSERT(nexts >= 0);
5320 
5321 	/*
5322 	 * Guesstimate how many blocks we can unmap without running the risk of
5323 	 * blowing out the transaction with a mix of EFIs and reflink
5324 	 * adjustments.
5325 	 */
5326 	if (tp && xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK)
5327 		max_len = min(len, xfs_refcount_max_unmap(tp->t_log_res));
5328 	else
5329 		max_len = len;
5330 
5331 	error = xfs_iread_extents(tp, ip, whichfork);
5332 	if (error)
5333 		return error;
5334 
5335 	if (xfs_iext_count(ifp) == 0) {
5336 		*rlen = 0;
5337 		return 0;
5338 	}
5339 	XFS_STATS_INC(mp, xs_blk_unmap);
5340 	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5341 	end = start + len;
5342 
5343 	if (!xfs_iext_lookup_extent_before(ip, ifp, &end, &icur, &got)) {
5344 		*rlen = 0;
5345 		return 0;
5346 	}
5347 	end--;
5348 
5349 	logflags = 0;
5350 	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
5351 		ASSERT(ifp->if_format == XFS_DINODE_FMT_BTREE);
5352 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5353 		cur->bc_ino.flags = 0;
5354 	} else
5355 		cur = NULL;
5356 
5357 	if (isrt) {
5358 		/*
5359 		 * Synchronize by locking the bitmap inode.
5360 		 */
5361 		xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
5362 		xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5363 		xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
5364 		xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
5365 	}
5366 
5367 	extno = 0;
5368 	while (end != (xfs_fileoff_t)-1 && end >= start &&
5369 	       (nexts == 0 || extno < nexts) && max_len > 0) {
5370 		/*
5371 		 * Is the found extent after a hole in which end lives?
5372 		 * Just back up to the previous extent, if so.
5373 		 */
5374 		if (got.br_startoff > end &&
5375 		    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5376 			done = true;
5377 			break;
5378 		}
5379 		/*
5380 		 * Is the last block of this extent before the range
5381 		 * we're supposed to delete?  If so, we're done.
5382 		 */
5383 		end = XFS_FILEOFF_MIN(end,
5384 			got.br_startoff + got.br_blockcount - 1);
5385 		if (end < start)
5386 			break;
5387 		/*
5388 		 * Then deal with the (possibly delayed) allocated space
5389 		 * we found.
5390 		 */
5391 		del = got;
5392 		wasdel = isnullstartblock(del.br_startblock);
5393 
5394 		if (got.br_startoff < start) {
5395 			del.br_startoff = start;
5396 			del.br_blockcount -= start - got.br_startoff;
5397 			if (!wasdel)
5398 				del.br_startblock += start - got.br_startoff;
5399 		}
5400 		if (del.br_startoff + del.br_blockcount > end + 1)
5401 			del.br_blockcount = end + 1 - del.br_startoff;
5402 
5403 		/* How much can we safely unmap? */
5404 		if (max_len < del.br_blockcount) {
5405 			del.br_startoff += del.br_blockcount - max_len;
5406 			if (!wasdel)
5407 				del.br_startblock += del.br_blockcount - max_len;
5408 			del.br_blockcount = max_len;
5409 		}
5410 
5411 		if (!isrt)
5412 			goto delete;
5413 
5414 		sum = del.br_startblock + del.br_blockcount;
5415 		div_u64_rem(sum, mp->m_sb.sb_rextsize, &mod);
5416 		if (mod) {
5417 			/*
5418 			 * Realtime extent not lined up at the end.
5419 			 * The extent could have been split into written
5420 			 * and unwritten pieces, or we could just be
5421 			 * unmapping part of it.  But we can't really
5422 			 * get rid of part of a realtime extent.
5423 			 */
5424 			if (del.br_state == XFS_EXT_UNWRITTEN) {
5425 				/*
5426 				 * This piece is unwritten, or we're not
5427 				 * using unwritten extents.  Skip over it.
5428 				 */
5429 				ASSERT(end >= mod);
5430 				end -= mod > del.br_blockcount ?
5431 					del.br_blockcount : mod;
5432 				if (end < got.br_startoff &&
5433 				    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5434 					done = true;
5435 					break;
5436 				}
5437 				continue;
5438 			}
5439 			/*
5440 			 * It's written, turn it unwritten.
5441 			 * This is better than zeroing it.
5442 			 */
5443 			ASSERT(del.br_state == XFS_EXT_NORM);
5444 			ASSERT(tp->t_blk_res > 0);
5445 			/*
5446 			 * If this spans a realtime extent boundary,
5447 			 * chop it back to the start of the one we end at.
5448 			 */
5449 			if (del.br_blockcount > mod) {
5450 				del.br_startoff += del.br_blockcount - mod;
5451 				del.br_startblock += del.br_blockcount - mod;
5452 				del.br_blockcount = mod;
5453 			}
5454 			del.br_state = XFS_EXT_UNWRITTEN;
5455 			error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5456 					whichfork, &icur, &cur, &del,
5457 					&logflags);
5458 			if (error)
5459 				goto error0;
5460 			goto nodelete;
5461 		}
5462 		div_u64_rem(del.br_startblock, mp->m_sb.sb_rextsize, &mod);
5463 		if (mod) {
5464 			xfs_extlen_t off = mp->m_sb.sb_rextsize - mod;
5465 
5466 			/*
5467 			 * Realtime extent is lined up at the end but not
5468 			 * at the front.  We'll get rid of full extents if
5469 			 * we can.
5470 			 */
5471 			if (del.br_blockcount > off) {
5472 				del.br_blockcount -= off;
5473 				del.br_startoff += off;
5474 				del.br_startblock += off;
5475 			} else if (del.br_startoff == start &&
5476 				   (del.br_state == XFS_EXT_UNWRITTEN ||
5477 				    tp->t_blk_res == 0)) {
5478 				/*
5479 				 * Can't make it unwritten.  There isn't
5480 				 * a full extent here so just skip it.
5481 				 */
5482 				ASSERT(end >= del.br_blockcount);
5483 				end -= del.br_blockcount;
5484 				if (got.br_startoff > end &&
5485 				    !xfs_iext_prev_extent(ifp, &icur, &got)) {
5486 					done = true;
5487 					break;
5488 				}
5489 				continue;
5490 			} else if (del.br_state == XFS_EXT_UNWRITTEN) {
5491 				struct xfs_bmbt_irec	prev;
5492 				xfs_fileoff_t		unwrite_start;
5493 
5494 				/*
5495 				 * This one is already unwritten.
5496 				 * It must have a written left neighbor.
5497 				 * Unwrite the killed part of that one and
5498 				 * try again.
5499 				 */
5500 				if (!xfs_iext_prev_extent(ifp, &icur, &prev))
5501 					ASSERT(0);
5502 				ASSERT(prev.br_state == XFS_EXT_NORM);
5503 				ASSERT(!isnullstartblock(prev.br_startblock));
5504 				ASSERT(del.br_startblock ==
5505 				       prev.br_startblock + prev.br_blockcount);
5506 				unwrite_start = max3(start,
5507 						     del.br_startoff - mod,
5508 						     prev.br_startoff);
5509 				mod = unwrite_start - prev.br_startoff;
5510 				prev.br_startoff = unwrite_start;
5511 				prev.br_startblock += mod;
5512 				prev.br_blockcount -= mod;
5513 				prev.br_state = XFS_EXT_UNWRITTEN;
5514 				error = xfs_bmap_add_extent_unwritten_real(tp,
5515 						ip, whichfork, &icur, &cur,
5516 						&prev, &logflags);
5517 				if (error)
5518 					goto error0;
5519 				goto nodelete;
5520 			} else {
5521 				ASSERT(del.br_state == XFS_EXT_NORM);
5522 				del.br_state = XFS_EXT_UNWRITTEN;
5523 				error = xfs_bmap_add_extent_unwritten_real(tp,
5524 						ip, whichfork, &icur, &cur,
5525 						&del, &logflags);
5526 				if (error)
5527 					goto error0;
5528 				goto nodelete;
5529 			}
5530 		}
5531 
5532 delete:
5533 		if (wasdel) {
5534 			error = xfs_bmap_del_extent_delay(ip, whichfork, &icur,
5535 					&got, &del);
5536 		} else {
5537 			error = xfs_bmap_del_extent_real(ip, tp, &icur, cur,
5538 					&del, &tmp_logflags, whichfork,
5539 					flags);
5540 			logflags |= tmp_logflags;
5541 		}
5542 
5543 		if (error)
5544 			goto error0;
5545 
5546 		max_len -= del.br_blockcount;
5547 		end = del.br_startoff - 1;
5548 nodelete:
5549 		/*
5550 		 * If not done go on to the next (previous) record.
5551 		 */
5552 		if (end != (xfs_fileoff_t)-1 && end >= start) {
5553 			if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5554 			    (got.br_startoff > end &&
5555 			     !xfs_iext_prev_extent(ifp, &icur, &got))) {
5556 				done = true;
5557 				break;
5558 			}
5559 			extno++;
5560 		}
5561 	}
5562 	if (done || end == (xfs_fileoff_t)-1 || end < start)
5563 		*rlen = 0;
5564 	else
5565 		*rlen = end - start + 1;
5566 
5567 	/*
5568 	 * Convert to a btree if necessary.
5569 	 */
5570 	if (xfs_bmap_needs_btree(ip, whichfork)) {
5571 		ASSERT(cur == NULL);
5572 		error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
5573 				&tmp_logflags, whichfork);
5574 		logflags |= tmp_logflags;
5575 	} else {
5576 		error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags,
5577 			whichfork);
5578 	}
5579 
5580 error0:
5581 	/*
5582 	 * Log everything.  Do this after conversion, there's no point in
5583 	 * logging the extent records if we've converted to btree format.
5584 	 */
5585 	if ((logflags & xfs_ilog_fext(whichfork)) &&
5586 	    ifp->if_format != XFS_DINODE_FMT_EXTENTS)
5587 		logflags &= ~xfs_ilog_fext(whichfork);
5588 	else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5589 		 ifp->if_format != XFS_DINODE_FMT_BTREE)
5590 		logflags &= ~xfs_ilog_fbroot(whichfork);
5591 	/*
5592 	 * Log inode even in the error case, if the transaction
5593 	 * is dirty we'll need to shut down the filesystem.
5594 	 */
5595 	if (logflags)
5596 		xfs_trans_log_inode(tp, ip, logflags);
5597 	if (cur) {
5598 		if (!error)
5599 			cur->bc_ino.allocated = 0;
5600 		xfs_btree_del_cursor(cur, error);
5601 	}
5602 	return error;
5603 }
5604 
5605 /* Unmap a range of a file. */
5606 int
5607 xfs_bunmapi(
5608 	xfs_trans_t		*tp,
5609 	struct xfs_inode	*ip,
5610 	xfs_fileoff_t		bno,
5611 	xfs_filblks_t		len,
5612 	int			flags,
5613 	xfs_extnum_t		nexts,
5614 	int			*done)
5615 {
5616 	int			error;
5617 
5618 	error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts);
5619 	*done = (len == 0);
5620 	return error;
5621 }
5622 
5623 /*
5624  * Determine whether an extent shift can be accomplished by a merge with the
5625  * extent that precedes the target hole of the shift.
5626  */
5627 STATIC bool
5628 xfs_bmse_can_merge(
5629 	struct xfs_bmbt_irec	*left,	/* preceding extent */
5630 	struct xfs_bmbt_irec	*got,	/* current extent to shift */
5631 	xfs_fileoff_t		shift)	/* shift fsb */
5632 {
5633 	xfs_fileoff_t		startoff;
5634 
5635 	startoff = got->br_startoff - shift;
5636 
5637 	/*
5638 	 * The extent, once shifted, must be adjacent in-file and on-disk with
5639 	 * the preceding extent.
5640 	 */
5641 	if ((left->br_startoff + left->br_blockcount != startoff) ||
5642 	    (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5643 	    (left->br_state != got->br_state) ||
5644 	    (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5645 		return false;
5646 
5647 	return true;
5648 }
5649 
5650 /*
5651  * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5652  * hole in the file. If an extent shift would result in the extent being fully
5653  * adjacent to the extent that currently precedes the hole, we can merge with
5654  * the preceding extent rather than do the shift.
5655  *
5656  * This function assumes the caller has verified a shift-by-merge is possible
5657  * with the provided extents via xfs_bmse_can_merge().
5658  */
5659 STATIC int
5660 xfs_bmse_merge(
5661 	struct xfs_trans		*tp,
5662 	struct xfs_inode		*ip,
5663 	int				whichfork,
5664 	xfs_fileoff_t			shift,		/* shift fsb */
5665 	struct xfs_iext_cursor		*icur,
5666 	struct xfs_bmbt_irec		*got,		/* extent to shift */
5667 	struct xfs_bmbt_irec		*left,		/* preceding extent */
5668 	struct xfs_btree_cur		*cur,
5669 	int				*logflags)	/* output */
5670 {
5671 	struct xfs_ifork		*ifp = XFS_IFORK_PTR(ip, whichfork);
5672 	struct xfs_bmbt_irec		new;
5673 	xfs_filblks_t			blockcount;
5674 	int				error, i;
5675 	struct xfs_mount		*mp = ip->i_mount;
5676 
5677 	blockcount = left->br_blockcount + got->br_blockcount;
5678 
5679 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5680 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5681 	ASSERT(xfs_bmse_can_merge(left, got, shift));
5682 
5683 	new = *left;
5684 	new.br_blockcount = blockcount;
5685 
5686 	/*
5687 	 * Update the on-disk extent count, the btree if necessary and log the
5688 	 * inode.
5689 	 */
5690 	ifp->if_nextents--;
5691 	*logflags |= XFS_ILOG_CORE;
5692 	if (!cur) {
5693 		*logflags |= XFS_ILOG_DEXT;
5694 		goto done;
5695 	}
5696 
5697 	/* lookup and remove the extent to merge */
5698 	error = xfs_bmbt_lookup_eq(cur, got, &i);
5699 	if (error)
5700 		return error;
5701 	if (XFS_IS_CORRUPT(mp, i != 1))
5702 		return -EFSCORRUPTED;
5703 
5704 	error = xfs_btree_delete(cur, &i);
5705 	if (error)
5706 		return error;
5707 	if (XFS_IS_CORRUPT(mp, i != 1))
5708 		return -EFSCORRUPTED;
5709 
5710 	/* lookup and update size of the previous extent */
5711 	error = xfs_bmbt_lookup_eq(cur, left, &i);
5712 	if (error)
5713 		return error;
5714 	if (XFS_IS_CORRUPT(mp, i != 1))
5715 		return -EFSCORRUPTED;
5716 
5717 	error = xfs_bmbt_update(cur, &new);
5718 	if (error)
5719 		return error;
5720 
5721 	/* change to extent format if required after extent removal */
5722 	error = xfs_bmap_btree_to_extents(tp, ip, cur, logflags, whichfork);
5723 	if (error)
5724 		return error;
5725 
5726 done:
5727 	xfs_iext_remove(ip, icur, 0);
5728 	xfs_iext_prev(ifp, icur);
5729 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5730 			&new);
5731 
5732 	/* update reverse mapping. rmap functions merge the rmaps for us */
5733 	xfs_rmap_unmap_extent(tp, ip, whichfork, got);
5734 	memcpy(&new, got, sizeof(new));
5735 	new.br_startoff = left->br_startoff + left->br_blockcount;
5736 	xfs_rmap_map_extent(tp, ip, whichfork, &new);
5737 	return 0;
5738 }
5739 
5740 static int
5741 xfs_bmap_shift_update_extent(
5742 	struct xfs_trans	*tp,
5743 	struct xfs_inode	*ip,
5744 	int			whichfork,
5745 	struct xfs_iext_cursor	*icur,
5746 	struct xfs_bmbt_irec	*got,
5747 	struct xfs_btree_cur	*cur,
5748 	int			*logflags,
5749 	xfs_fileoff_t		startoff)
5750 {
5751 	struct xfs_mount	*mp = ip->i_mount;
5752 	struct xfs_bmbt_irec	prev = *got;
5753 	int			error, i;
5754 
5755 	*logflags |= XFS_ILOG_CORE;
5756 
5757 	got->br_startoff = startoff;
5758 
5759 	if (cur) {
5760 		error = xfs_bmbt_lookup_eq(cur, &prev, &i);
5761 		if (error)
5762 			return error;
5763 		if (XFS_IS_CORRUPT(mp, i != 1))
5764 			return -EFSCORRUPTED;
5765 
5766 		error = xfs_bmbt_update(cur, got);
5767 		if (error)
5768 			return error;
5769 	} else {
5770 		*logflags |= XFS_ILOG_DEXT;
5771 	}
5772 
5773 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5774 			got);
5775 
5776 	/* update reverse mapping */
5777 	xfs_rmap_unmap_extent(tp, ip, whichfork, &prev);
5778 	xfs_rmap_map_extent(tp, ip, whichfork, got);
5779 	return 0;
5780 }
5781 
5782 int
5783 xfs_bmap_collapse_extents(
5784 	struct xfs_trans	*tp,
5785 	struct xfs_inode	*ip,
5786 	xfs_fileoff_t		*next_fsb,
5787 	xfs_fileoff_t		offset_shift_fsb,
5788 	bool			*done)
5789 {
5790 	int			whichfork = XFS_DATA_FORK;
5791 	struct xfs_mount	*mp = ip->i_mount;
5792 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
5793 	struct xfs_btree_cur	*cur = NULL;
5794 	struct xfs_bmbt_irec	got, prev;
5795 	struct xfs_iext_cursor	icur;
5796 	xfs_fileoff_t		new_startoff;
5797 	int			error = 0;
5798 	int			logflags = 0;
5799 
5800 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
5801 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
5802 		return -EFSCORRUPTED;
5803 	}
5804 
5805 	if (xfs_is_shutdown(mp))
5806 		return -EIO;
5807 
5808 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL));
5809 
5810 	error = xfs_iread_extents(tp, ip, whichfork);
5811 	if (error)
5812 		return error;
5813 
5814 	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
5815 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5816 		cur->bc_ino.flags = 0;
5817 	}
5818 
5819 	if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5820 		*done = true;
5821 		goto del_cursor;
5822 	}
5823 	if (XFS_IS_CORRUPT(mp, isnullstartblock(got.br_startblock))) {
5824 		error = -EFSCORRUPTED;
5825 		goto del_cursor;
5826 	}
5827 
5828 	new_startoff = got.br_startoff - offset_shift_fsb;
5829 	if (xfs_iext_peek_prev_extent(ifp, &icur, &prev)) {
5830 		if (new_startoff < prev.br_startoff + prev.br_blockcount) {
5831 			error = -EINVAL;
5832 			goto del_cursor;
5833 		}
5834 
5835 		if (xfs_bmse_can_merge(&prev, &got, offset_shift_fsb)) {
5836 			error = xfs_bmse_merge(tp, ip, whichfork,
5837 					offset_shift_fsb, &icur, &got, &prev,
5838 					cur, &logflags);
5839 			if (error)
5840 				goto del_cursor;
5841 			goto done;
5842 		}
5843 	} else {
5844 		if (got.br_startoff < offset_shift_fsb) {
5845 			error = -EINVAL;
5846 			goto del_cursor;
5847 		}
5848 	}
5849 
5850 	error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got,
5851 			cur, &logflags, new_startoff);
5852 	if (error)
5853 		goto del_cursor;
5854 
5855 done:
5856 	if (!xfs_iext_next_extent(ifp, &icur, &got)) {
5857 		*done = true;
5858 		goto del_cursor;
5859 	}
5860 
5861 	*next_fsb = got.br_startoff;
5862 del_cursor:
5863 	if (cur)
5864 		xfs_btree_del_cursor(cur, error);
5865 	if (logflags)
5866 		xfs_trans_log_inode(tp, ip, logflags);
5867 	return error;
5868 }
5869 
5870 /* Make sure we won't be right-shifting an extent past the maximum bound. */
5871 int
5872 xfs_bmap_can_insert_extents(
5873 	struct xfs_inode	*ip,
5874 	xfs_fileoff_t		off,
5875 	xfs_fileoff_t		shift)
5876 {
5877 	struct xfs_bmbt_irec	got;
5878 	int			is_empty;
5879 	int			error = 0;
5880 
5881 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5882 
5883 	if (xfs_is_shutdown(ip->i_mount))
5884 		return -EIO;
5885 
5886 	xfs_ilock(ip, XFS_ILOCK_EXCL);
5887 	error = xfs_bmap_last_extent(NULL, ip, XFS_DATA_FORK, &got, &is_empty);
5888 	if (!error && !is_empty && got.br_startoff >= off &&
5889 	    ((got.br_startoff + shift) & BMBT_STARTOFF_MASK) < got.br_startoff)
5890 		error = -EINVAL;
5891 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
5892 
5893 	return error;
5894 }
5895 
5896 int
5897 xfs_bmap_insert_extents(
5898 	struct xfs_trans	*tp,
5899 	struct xfs_inode	*ip,
5900 	xfs_fileoff_t		*next_fsb,
5901 	xfs_fileoff_t		offset_shift_fsb,
5902 	bool			*done,
5903 	xfs_fileoff_t		stop_fsb)
5904 {
5905 	int			whichfork = XFS_DATA_FORK;
5906 	struct xfs_mount	*mp = ip->i_mount;
5907 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
5908 	struct xfs_btree_cur	*cur = NULL;
5909 	struct xfs_bmbt_irec	got, next;
5910 	struct xfs_iext_cursor	icur;
5911 	xfs_fileoff_t		new_startoff;
5912 	int			error = 0;
5913 	int			logflags = 0;
5914 
5915 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
5916 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
5917 		return -EFSCORRUPTED;
5918 	}
5919 
5920 	if (xfs_is_shutdown(mp))
5921 		return -EIO;
5922 
5923 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL));
5924 
5925 	error = xfs_iread_extents(tp, ip, whichfork);
5926 	if (error)
5927 		return error;
5928 
5929 	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
5930 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5931 		cur->bc_ino.flags = 0;
5932 	}
5933 
5934 	if (*next_fsb == NULLFSBLOCK) {
5935 		xfs_iext_last(ifp, &icur);
5936 		if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5937 		    stop_fsb > got.br_startoff) {
5938 			*done = true;
5939 			goto del_cursor;
5940 		}
5941 	} else {
5942 		if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5943 			*done = true;
5944 			goto del_cursor;
5945 		}
5946 	}
5947 	if (XFS_IS_CORRUPT(mp, isnullstartblock(got.br_startblock))) {
5948 		error = -EFSCORRUPTED;
5949 		goto del_cursor;
5950 	}
5951 
5952 	if (XFS_IS_CORRUPT(mp, stop_fsb > got.br_startoff)) {
5953 		error = -EFSCORRUPTED;
5954 		goto del_cursor;
5955 	}
5956 
5957 	new_startoff = got.br_startoff + offset_shift_fsb;
5958 	if (xfs_iext_peek_next_extent(ifp, &icur, &next)) {
5959 		if (new_startoff + got.br_blockcount > next.br_startoff) {
5960 			error = -EINVAL;
5961 			goto del_cursor;
5962 		}
5963 
5964 		/*
5965 		 * Unlike a left shift (which involves a hole punch), a right
5966 		 * shift does not modify extent neighbors in any way.  We should
5967 		 * never find mergeable extents in this scenario.  Check anyways
5968 		 * and warn if we encounter two extents that could be one.
5969 		 */
5970 		if (xfs_bmse_can_merge(&got, &next, offset_shift_fsb))
5971 			WARN_ON_ONCE(1);
5972 	}
5973 
5974 	error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got,
5975 			cur, &logflags, new_startoff);
5976 	if (error)
5977 		goto del_cursor;
5978 
5979 	if (!xfs_iext_prev_extent(ifp, &icur, &got) ||
5980 	    stop_fsb >= got.br_startoff + got.br_blockcount) {
5981 		*done = true;
5982 		goto del_cursor;
5983 	}
5984 
5985 	*next_fsb = got.br_startoff;
5986 del_cursor:
5987 	if (cur)
5988 		xfs_btree_del_cursor(cur, error);
5989 	if (logflags)
5990 		xfs_trans_log_inode(tp, ip, logflags);
5991 	return error;
5992 }
5993 
5994 /*
5995  * Splits an extent into two extents at split_fsb block such that it is the
5996  * first block of the current_ext. @ext is a target extent to be split.
5997  * @split_fsb is a block where the extents is split.  If split_fsb lies in a
5998  * hole or the first block of extents, just return 0.
5999  */
6000 int
6001 xfs_bmap_split_extent(
6002 	struct xfs_trans	*tp,
6003 	struct xfs_inode	*ip,
6004 	xfs_fileoff_t		split_fsb)
6005 {
6006 	int				whichfork = XFS_DATA_FORK;
6007 	struct xfs_ifork		*ifp = XFS_IFORK_PTR(ip, whichfork);
6008 	struct xfs_btree_cur		*cur = NULL;
6009 	struct xfs_bmbt_irec		got;
6010 	struct xfs_bmbt_irec		new; /* split extent */
6011 	struct xfs_mount		*mp = ip->i_mount;
6012 	xfs_fsblock_t			gotblkcnt; /* new block count for got */
6013 	struct xfs_iext_cursor		icur;
6014 	int				error = 0;
6015 	int				logflags = 0;
6016 	int				i = 0;
6017 
6018 	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
6019 	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
6020 		return -EFSCORRUPTED;
6021 	}
6022 
6023 	if (xfs_is_shutdown(mp))
6024 		return -EIO;
6025 
6026 	/* Read in all the extents */
6027 	error = xfs_iread_extents(tp, ip, whichfork);
6028 	if (error)
6029 		return error;
6030 
6031 	/*
6032 	 * If there are not extents, or split_fsb lies in a hole we are done.
6033 	 */
6034 	if (!xfs_iext_lookup_extent(ip, ifp, split_fsb, &icur, &got) ||
6035 	    got.br_startoff >= split_fsb)
6036 		return 0;
6037 
6038 	gotblkcnt = split_fsb - got.br_startoff;
6039 	new.br_startoff = split_fsb;
6040 	new.br_startblock = got.br_startblock + gotblkcnt;
6041 	new.br_blockcount = got.br_blockcount - gotblkcnt;
6042 	new.br_state = got.br_state;
6043 
6044 	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
6045 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6046 		cur->bc_ino.flags = 0;
6047 		error = xfs_bmbt_lookup_eq(cur, &got, &i);
6048 		if (error)
6049 			goto del_cursor;
6050 		if (XFS_IS_CORRUPT(mp, i != 1)) {
6051 			error = -EFSCORRUPTED;
6052 			goto del_cursor;
6053 		}
6054 	}
6055 
6056 	got.br_blockcount = gotblkcnt;
6057 	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), &icur,
6058 			&got);
6059 
6060 	logflags = XFS_ILOG_CORE;
6061 	if (cur) {
6062 		error = xfs_bmbt_update(cur, &got);
6063 		if (error)
6064 			goto del_cursor;
6065 	} else
6066 		logflags |= XFS_ILOG_DEXT;
6067 
6068 	/* Add new extent */
6069 	xfs_iext_next(ifp, &icur);
6070 	xfs_iext_insert(ip, &icur, &new, 0);
6071 	ifp->if_nextents++;
6072 
6073 	if (cur) {
6074 		error = xfs_bmbt_lookup_eq(cur, &new, &i);
6075 		if (error)
6076 			goto del_cursor;
6077 		if (XFS_IS_CORRUPT(mp, i != 0)) {
6078 			error = -EFSCORRUPTED;
6079 			goto del_cursor;
6080 		}
6081 		error = xfs_btree_insert(cur, &i);
6082 		if (error)
6083 			goto del_cursor;
6084 		if (XFS_IS_CORRUPT(mp, i != 1)) {
6085 			error = -EFSCORRUPTED;
6086 			goto del_cursor;
6087 		}
6088 	}
6089 
6090 	/*
6091 	 * Convert to a btree if necessary.
6092 	 */
6093 	if (xfs_bmap_needs_btree(ip, whichfork)) {
6094 		int tmp_logflags; /* partial log flag return val */
6095 
6096 		ASSERT(cur == NULL);
6097 		error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
6098 				&tmp_logflags, whichfork);
6099 		logflags |= tmp_logflags;
6100 	}
6101 
6102 del_cursor:
6103 	if (cur) {
6104 		cur->bc_ino.allocated = 0;
6105 		xfs_btree_del_cursor(cur, error);
6106 	}
6107 
6108 	if (logflags)
6109 		xfs_trans_log_inode(tp, ip, logflags);
6110 	return error;
6111 }
6112 
6113 /* Deferred mapping is only for real extents in the data fork. */
6114 static bool
6115 xfs_bmap_is_update_needed(
6116 	struct xfs_bmbt_irec	*bmap)
6117 {
6118 	return  bmap->br_startblock != HOLESTARTBLOCK &&
6119 		bmap->br_startblock != DELAYSTARTBLOCK;
6120 }
6121 
6122 /* Record a bmap intent. */
6123 static int
6124 __xfs_bmap_add(
6125 	struct xfs_trans		*tp,
6126 	enum xfs_bmap_intent_type	type,
6127 	struct xfs_inode		*ip,
6128 	int				whichfork,
6129 	struct xfs_bmbt_irec		*bmap)
6130 {
6131 	struct xfs_bmap_intent		*bi;
6132 
6133 	trace_xfs_bmap_defer(tp->t_mountp,
6134 			XFS_FSB_TO_AGNO(tp->t_mountp, bmap->br_startblock),
6135 			type,
6136 			XFS_FSB_TO_AGBNO(tp->t_mountp, bmap->br_startblock),
6137 			ip->i_ino, whichfork,
6138 			bmap->br_startoff,
6139 			bmap->br_blockcount,
6140 			bmap->br_state);
6141 
6142 	bi = kmem_cache_alloc(xfs_bmap_intent_cache, GFP_NOFS | __GFP_NOFAIL);
6143 	INIT_LIST_HEAD(&bi->bi_list);
6144 	bi->bi_type = type;
6145 	bi->bi_owner = ip;
6146 	bi->bi_whichfork = whichfork;
6147 	bi->bi_bmap = *bmap;
6148 
6149 	xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list);
6150 	return 0;
6151 }
6152 
6153 /* Map an extent into a file. */
6154 void
6155 xfs_bmap_map_extent(
6156 	struct xfs_trans	*tp,
6157 	struct xfs_inode	*ip,
6158 	struct xfs_bmbt_irec	*PREV)
6159 {
6160 	if (!xfs_bmap_is_update_needed(PREV))
6161 		return;
6162 
6163 	__xfs_bmap_add(tp, XFS_BMAP_MAP, ip, XFS_DATA_FORK, PREV);
6164 }
6165 
6166 /* Unmap an extent out of a file. */
6167 void
6168 xfs_bmap_unmap_extent(
6169 	struct xfs_trans	*tp,
6170 	struct xfs_inode	*ip,
6171 	struct xfs_bmbt_irec	*PREV)
6172 {
6173 	if (!xfs_bmap_is_update_needed(PREV))
6174 		return;
6175 
6176 	__xfs_bmap_add(tp, XFS_BMAP_UNMAP, ip, XFS_DATA_FORK, PREV);
6177 }
6178 
6179 /*
6180  * Process one of the deferred bmap operations.  We pass back the
6181  * btree cursor to maintain our lock on the bmapbt between calls.
6182  */
6183 int
6184 xfs_bmap_finish_one(
6185 	struct xfs_trans		*tp,
6186 	struct xfs_inode		*ip,
6187 	enum xfs_bmap_intent_type	type,
6188 	int				whichfork,
6189 	xfs_fileoff_t			startoff,
6190 	xfs_fsblock_t			startblock,
6191 	xfs_filblks_t			*blockcount,
6192 	xfs_exntst_t			state)
6193 {
6194 	int				error = 0;
6195 
6196 	ASSERT(tp->t_firstblock == NULLFSBLOCK);
6197 
6198 	trace_xfs_bmap_deferred(tp->t_mountp,
6199 			XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type,
6200 			XFS_FSB_TO_AGBNO(tp->t_mountp, startblock),
6201 			ip->i_ino, whichfork, startoff, *blockcount, state);
6202 
6203 	if (WARN_ON_ONCE(whichfork != XFS_DATA_FORK))
6204 		return -EFSCORRUPTED;
6205 
6206 	if (XFS_TEST_ERROR(false, tp->t_mountp,
6207 			XFS_ERRTAG_BMAP_FINISH_ONE))
6208 		return -EIO;
6209 
6210 	switch (type) {
6211 	case XFS_BMAP_MAP:
6212 		error = xfs_bmapi_remap(tp, ip, startoff, *blockcount,
6213 				startblock, 0);
6214 		*blockcount = 0;
6215 		break;
6216 	case XFS_BMAP_UNMAP:
6217 		error = __xfs_bunmapi(tp, ip, startoff, blockcount,
6218 				XFS_BMAPI_REMAP, 1);
6219 		break;
6220 	default:
6221 		ASSERT(0);
6222 		error = -EFSCORRUPTED;
6223 	}
6224 
6225 	return error;
6226 }
6227 
6228 /* Check that an inode's extent does not have invalid flags or bad ranges. */
6229 xfs_failaddr_t
6230 xfs_bmap_validate_extent(
6231 	struct xfs_inode	*ip,
6232 	int			whichfork,
6233 	struct xfs_bmbt_irec	*irec)
6234 {
6235 	struct xfs_mount	*mp = ip->i_mount;
6236 
6237 	if (!xfs_verify_fileext(mp, irec->br_startoff, irec->br_blockcount))
6238 		return __this_address;
6239 
6240 	if (XFS_IS_REALTIME_INODE(ip) && whichfork == XFS_DATA_FORK) {
6241 		if (!xfs_verify_rtext(mp, irec->br_startblock,
6242 					  irec->br_blockcount))
6243 			return __this_address;
6244 	} else {
6245 		if (!xfs_verify_fsbext(mp, irec->br_startblock,
6246 					   irec->br_blockcount))
6247 			return __this_address;
6248 	}
6249 	if (irec->br_state != XFS_EXT_NORM && whichfork != XFS_DATA_FORK)
6250 		return __this_address;
6251 	return NULL;
6252 }
6253 
6254 int __init
6255 xfs_bmap_intent_init_cache(void)
6256 {
6257 	xfs_bmap_intent_cache = kmem_cache_create("xfs_bmap_intent",
6258 			sizeof(struct xfs_bmap_intent),
6259 			0, 0, NULL);
6260 
6261 	return xfs_bmap_intent_cache != NULL ? 0 : -ENOMEM;
6262 }
6263 
6264 void
6265 xfs_bmap_intent_destroy_cache(void)
6266 {
6267 	kmem_cache_destroy(xfs_bmap_intent_cache);
6268 	xfs_bmap_intent_cache = NULL;
6269 }
6270