xref: /openbmc/linux/fs/xfs/xfs_bmap_util.c (revision 4da722ca)
1 /*
2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3  * Copyright (c) 2012 Red Hat, Inc.
4  * All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License as
8  * published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it would be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write the Free Software Foundation,
17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
18  */
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_shared.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
25 #include "xfs_bit.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_defer.h"
29 #include "xfs_inode.h"
30 #include "xfs_btree.h"
31 #include "xfs_trans.h"
32 #include "xfs_extfree_item.h"
33 #include "xfs_alloc.h"
34 #include "xfs_bmap.h"
35 #include "xfs_bmap_util.h"
36 #include "xfs_bmap_btree.h"
37 #include "xfs_rtalloc.h"
38 #include "xfs_error.h"
39 #include "xfs_quota.h"
40 #include "xfs_trans_space.h"
41 #include "xfs_trace.h"
42 #include "xfs_icache.h"
43 #include "xfs_log.h"
44 #include "xfs_rmap_btree.h"
45 #include "xfs_iomap.h"
46 #include "xfs_reflink.h"
47 #include "xfs_refcount.h"
48 
49 /* Kernel only BMAP related definitions and functions */
50 
51 /*
52  * Convert the given file system block to a disk block.  We have to treat it
53  * differently based on whether the file is a real time file or not, because the
54  * bmap code does.
55  */
56 xfs_daddr_t
57 xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb)
58 {
59 	return (XFS_IS_REALTIME_INODE(ip) ? \
60 		 (xfs_daddr_t)XFS_FSB_TO_BB((ip)->i_mount, (fsb)) : \
61 		 XFS_FSB_TO_DADDR((ip)->i_mount, (fsb)));
62 }
63 
64 /*
65  * Routine to zero an extent on disk allocated to the specific inode.
66  *
67  * The VFS functions take a linearised filesystem block offset, so we have to
68  * convert the sparse xfs fsb to the right format first.
69  * VFS types are real funky, too.
70  */
71 int
72 xfs_zero_extent(
73 	struct xfs_inode *ip,
74 	xfs_fsblock_t	start_fsb,
75 	xfs_off_t	count_fsb)
76 {
77 	struct xfs_mount *mp = ip->i_mount;
78 	xfs_daddr_t	sector = xfs_fsb_to_db(ip, start_fsb);
79 	sector_t	block = XFS_BB_TO_FSBT(mp, sector);
80 
81 	return blkdev_issue_zeroout(xfs_find_bdev_for_inode(VFS_I(ip)),
82 		block << (mp->m_super->s_blocksize_bits - 9),
83 		count_fsb << (mp->m_super->s_blocksize_bits - 9),
84 		GFP_NOFS, 0);
85 }
86 
87 int
88 xfs_bmap_rtalloc(
89 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
90 {
91 	int		error;		/* error return value */
92 	xfs_mount_t	*mp;		/* mount point structure */
93 	xfs_extlen_t	prod = 0;	/* product factor for allocators */
94 	xfs_extlen_t	ralen = 0;	/* realtime allocation length */
95 	xfs_extlen_t	align;		/* minimum allocation alignment */
96 	xfs_rtblock_t	rtb;
97 
98 	mp = ap->ip->i_mount;
99 	align = xfs_get_extsz_hint(ap->ip);
100 	prod = align / mp->m_sb.sb_rextsize;
101 	error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
102 					align, 1, ap->eof, 0,
103 					ap->conv, &ap->offset, &ap->length);
104 	if (error)
105 		return error;
106 	ASSERT(ap->length);
107 	ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
108 
109 	/*
110 	 * If the offset & length are not perfectly aligned
111 	 * then kill prod, it will just get us in trouble.
112 	 */
113 	if (do_mod(ap->offset, align) || ap->length % align)
114 		prod = 1;
115 	/*
116 	 * Set ralen to be the actual requested length in rtextents.
117 	 */
118 	ralen = ap->length / mp->m_sb.sb_rextsize;
119 	/*
120 	 * If the old value was close enough to MAXEXTLEN that
121 	 * we rounded up to it, cut it back so it's valid again.
122 	 * Note that if it's a really large request (bigger than
123 	 * MAXEXTLEN), we don't hear about that number, and can't
124 	 * adjust the starting point to match it.
125 	 */
126 	if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
127 		ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
128 
129 	/*
130 	 * Lock out modifications to both the RT bitmap and summary inodes
131 	 */
132 	xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
133 	xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
134 	xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
135 	xfs_trans_ijoin(ap->tp, mp->m_rsumip, XFS_ILOCK_EXCL);
136 
137 	/*
138 	 * If it's an allocation to an empty file at offset 0,
139 	 * pick an extent that will space things out in the rt area.
140 	 */
141 	if (ap->eof && ap->offset == 0) {
142 		xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
143 
144 		error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
145 		if (error)
146 			return error;
147 		ap->blkno = rtx * mp->m_sb.sb_rextsize;
148 	} else {
149 		ap->blkno = 0;
150 	}
151 
152 	xfs_bmap_adjacent(ap);
153 
154 	/*
155 	 * Realtime allocation, done through xfs_rtallocate_extent.
156 	 */
157 	do_div(ap->blkno, mp->m_sb.sb_rextsize);
158 	rtb = ap->blkno;
159 	ap->length = ralen;
160 	error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
161 				&ralen, ap->wasdel, prod, &rtb);
162 	if (error)
163 		return error;
164 
165 	ap->blkno = rtb;
166 	if (ap->blkno != NULLFSBLOCK) {
167 		ap->blkno *= mp->m_sb.sb_rextsize;
168 		ralen *= mp->m_sb.sb_rextsize;
169 		ap->length = ralen;
170 		ap->ip->i_d.di_nblocks += ralen;
171 		xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
172 		if (ap->wasdel)
173 			ap->ip->i_delayed_blks -= ralen;
174 		/*
175 		 * Adjust the disk quota also. This was reserved
176 		 * earlier.
177 		 */
178 		xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
179 			ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
180 					XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
181 
182 		/* Zero the extent if we were asked to do so */
183 		if (ap->datatype & XFS_ALLOC_USERDATA_ZERO) {
184 			error = xfs_zero_extent(ap->ip, ap->blkno, ap->length);
185 			if (error)
186 				return error;
187 		}
188 	} else {
189 		ap->length = 0;
190 	}
191 	return 0;
192 }
193 
194 /*
195  * Check if the endoff is outside the last extent. If so the caller will grow
196  * the allocation to a stripe unit boundary.  All offsets are considered outside
197  * the end of file for an empty fork, so 1 is returned in *eof in that case.
198  */
199 int
200 xfs_bmap_eof(
201 	struct xfs_inode	*ip,
202 	xfs_fileoff_t		endoff,
203 	int			whichfork,
204 	int			*eof)
205 {
206 	struct xfs_bmbt_irec	rec;
207 	int			error;
208 
209 	error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
210 	if (error || *eof)
211 		return error;
212 
213 	*eof = endoff >= rec.br_startoff + rec.br_blockcount;
214 	return 0;
215 }
216 
217 /*
218  * Extent tree block counting routines.
219  */
220 
221 /*
222  * Count leaf blocks given a range of extent records.  Delayed allocation
223  * extents are not counted towards the totals.
224  */
225 STATIC void
226 xfs_bmap_count_leaves(
227 	struct xfs_ifork	*ifp,
228 	xfs_extnum_t		*numrecs,
229 	xfs_filblks_t		*count)
230 {
231 	xfs_extnum_t		i;
232 	xfs_extnum_t		nr_exts = xfs_iext_count(ifp);
233 
234 	for (i = 0; i < nr_exts; i++) {
235 		xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, i);
236 		if (!isnullstartblock(xfs_bmbt_get_startblock(frp))) {
237 			(*numrecs)++;
238 			*count += xfs_bmbt_get_blockcount(frp);
239 		}
240 	}
241 }
242 
243 /*
244  * Count leaf blocks given a range of extent records originally
245  * in btree format.
246  */
247 STATIC void
248 xfs_bmap_disk_count_leaves(
249 	struct xfs_mount	*mp,
250 	struct xfs_btree_block	*block,
251 	int			numrecs,
252 	xfs_filblks_t		*count)
253 {
254 	int		b;
255 	xfs_bmbt_rec_t	*frp;
256 
257 	for (b = 1; b <= numrecs; b++) {
258 		frp = XFS_BMBT_REC_ADDR(mp, block, b);
259 		*count += xfs_bmbt_disk_get_blockcount(frp);
260 	}
261 }
262 
263 /*
264  * Recursively walks each level of a btree
265  * to count total fsblocks in use.
266  */
267 STATIC int
268 xfs_bmap_count_tree(
269 	struct xfs_mount	*mp,
270 	struct xfs_trans	*tp,
271 	struct xfs_ifork	*ifp,
272 	xfs_fsblock_t		blockno,
273 	int			levelin,
274 	xfs_extnum_t		*nextents,
275 	xfs_filblks_t		*count)
276 {
277 	int			error;
278 	struct xfs_buf		*bp, *nbp;
279 	int			level = levelin;
280 	__be64			*pp;
281 	xfs_fsblock_t           bno = blockno;
282 	xfs_fsblock_t		nextbno;
283 	struct xfs_btree_block	*block, *nextblock;
284 	int			numrecs;
285 
286 	error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF,
287 						&xfs_bmbt_buf_ops);
288 	if (error)
289 		return error;
290 	*count += 1;
291 	block = XFS_BUF_TO_BLOCK(bp);
292 
293 	if (--level) {
294 		/* Not at node above leaves, count this level of nodes */
295 		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
296 		while (nextbno != NULLFSBLOCK) {
297 			error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp,
298 						XFS_BMAP_BTREE_REF,
299 						&xfs_bmbt_buf_ops);
300 			if (error)
301 				return error;
302 			*count += 1;
303 			nextblock = XFS_BUF_TO_BLOCK(nbp);
304 			nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
305 			xfs_trans_brelse(tp, nbp);
306 		}
307 
308 		/* Dive to the next level */
309 		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
310 		bno = be64_to_cpu(*pp);
311 		error = xfs_bmap_count_tree(mp, tp, ifp, bno, level, nextents,
312 				count);
313 		if (error) {
314 			xfs_trans_brelse(tp, bp);
315 			XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
316 					 XFS_ERRLEVEL_LOW, mp);
317 			return -EFSCORRUPTED;
318 		}
319 		xfs_trans_brelse(tp, bp);
320 	} else {
321 		/* count all level 1 nodes and their leaves */
322 		for (;;) {
323 			nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
324 			numrecs = be16_to_cpu(block->bb_numrecs);
325 			(*nextents) += numrecs;
326 			xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
327 			xfs_trans_brelse(tp, bp);
328 			if (nextbno == NULLFSBLOCK)
329 				break;
330 			bno = nextbno;
331 			error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
332 						XFS_BMAP_BTREE_REF,
333 						&xfs_bmbt_buf_ops);
334 			if (error)
335 				return error;
336 			*count += 1;
337 			block = XFS_BUF_TO_BLOCK(bp);
338 		}
339 	}
340 	return 0;
341 }
342 
343 /*
344  * Count fsblocks of the given fork.  Delayed allocation extents are
345  * not counted towards the totals.
346  */
347 int
348 xfs_bmap_count_blocks(
349 	struct xfs_trans	*tp,
350 	struct xfs_inode	*ip,
351 	int			whichfork,
352 	xfs_extnum_t		*nextents,
353 	xfs_filblks_t		*count)
354 {
355 	struct xfs_mount	*mp;	/* file system mount structure */
356 	__be64			*pp;	/* pointer to block address */
357 	struct xfs_btree_block	*block;	/* current btree block */
358 	struct xfs_ifork	*ifp;	/* fork structure */
359 	xfs_fsblock_t		bno;	/* block # of "block" */
360 	int			level;	/* btree level, for checking */
361 	int			error;
362 
363 	bno = NULLFSBLOCK;
364 	mp = ip->i_mount;
365 	*nextents = 0;
366 	*count = 0;
367 	ifp = XFS_IFORK_PTR(ip, whichfork);
368 	if (!ifp)
369 		return 0;
370 
371 	switch (XFS_IFORK_FORMAT(ip, whichfork)) {
372 	case XFS_DINODE_FMT_EXTENTS:
373 		xfs_bmap_count_leaves(ifp, nextents, count);
374 		return 0;
375 	case XFS_DINODE_FMT_BTREE:
376 		if (!(ifp->if_flags & XFS_IFEXTENTS)) {
377 			error = xfs_iread_extents(tp, ip, whichfork);
378 			if (error)
379 				return error;
380 		}
381 
382 		/*
383 		 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
384 		 */
385 		block = ifp->if_broot;
386 		level = be16_to_cpu(block->bb_level);
387 		ASSERT(level > 0);
388 		pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
389 		bno = be64_to_cpu(*pp);
390 		ASSERT(bno != NULLFSBLOCK);
391 		ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
392 		ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
393 
394 		error = xfs_bmap_count_tree(mp, tp, ifp, bno, level,
395 				nextents, count);
396 		if (error) {
397 			XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)",
398 					XFS_ERRLEVEL_LOW, mp);
399 			return -EFSCORRUPTED;
400 		}
401 		return 0;
402 	}
403 
404 	return 0;
405 }
406 
407 /*
408  * returns 1 for success, 0 if we failed to map the extent.
409  */
410 STATIC int
411 xfs_getbmapx_fix_eof_hole(
412 	xfs_inode_t		*ip,		/* xfs incore inode pointer */
413 	int			whichfork,
414 	struct getbmapx		*out,		/* output structure */
415 	int			prealloced,	/* this is a file with
416 						 * preallocated data space */
417 	int64_t			end,		/* last block requested */
418 	xfs_fsblock_t		startblock,
419 	bool			moretocome)
420 {
421 	int64_t			fixlen;
422 	xfs_mount_t		*mp;		/* file system mount point */
423 	xfs_ifork_t		*ifp;		/* inode fork pointer */
424 	xfs_extnum_t		lastx;		/* last extent pointer */
425 	xfs_fileoff_t		fileblock;
426 
427 	if (startblock == HOLESTARTBLOCK) {
428 		mp = ip->i_mount;
429 		out->bmv_block = -1;
430 		fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
431 		fixlen -= out->bmv_offset;
432 		if (prealloced && out->bmv_offset + out->bmv_length == end) {
433 			/* Came to hole at EOF. Trim it. */
434 			if (fixlen <= 0)
435 				return 0;
436 			out->bmv_length = fixlen;
437 		}
438 	} else {
439 		if (startblock == DELAYSTARTBLOCK)
440 			out->bmv_block = -2;
441 		else
442 			out->bmv_block = xfs_fsb_to_db(ip, startblock);
443 		fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
444 		ifp = XFS_IFORK_PTR(ip, whichfork);
445 		if (!moretocome &&
446 		    xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
447 		   (lastx == xfs_iext_count(ifp) - 1))
448 			out->bmv_oflags |= BMV_OF_LAST;
449 	}
450 
451 	return 1;
452 }
453 
454 /* Adjust the reported bmap around shared/unshared extent transitions. */
455 STATIC int
456 xfs_getbmap_adjust_shared(
457 	struct xfs_inode		*ip,
458 	int				whichfork,
459 	struct xfs_bmbt_irec		*map,
460 	struct getbmapx			*out,
461 	struct xfs_bmbt_irec		*next_map)
462 {
463 	struct xfs_mount		*mp = ip->i_mount;
464 	xfs_agnumber_t			agno;
465 	xfs_agblock_t			agbno;
466 	xfs_agblock_t			ebno;
467 	xfs_extlen_t			elen;
468 	xfs_extlen_t			nlen;
469 	int				error;
470 
471 	next_map->br_startblock = NULLFSBLOCK;
472 	next_map->br_startoff = NULLFILEOFF;
473 	next_map->br_blockcount = 0;
474 
475 	/* Only written data blocks can be shared. */
476 	if (!xfs_is_reflink_inode(ip) ||
477 	    whichfork != XFS_DATA_FORK ||
478 	    !xfs_bmap_is_real_extent(map))
479 		return 0;
480 
481 	agno = XFS_FSB_TO_AGNO(mp, map->br_startblock);
482 	agbno = XFS_FSB_TO_AGBNO(mp, map->br_startblock);
483 	error = xfs_reflink_find_shared(mp, NULL, agno, agbno,
484 			map->br_blockcount, &ebno, &elen, true);
485 	if (error)
486 		return error;
487 
488 	if (ebno == NULLAGBLOCK) {
489 		/* No shared blocks at all. */
490 		return 0;
491 	} else if (agbno == ebno) {
492 		/*
493 		 * Shared extent at (agbno, elen).  Shrink the reported
494 		 * extent length and prepare to move the start of map[i]
495 		 * to agbno+elen, with the aim of (re)formatting the new
496 		 * map[i] the next time through the inner loop.
497 		 */
498 		out->bmv_length = XFS_FSB_TO_BB(mp, elen);
499 		out->bmv_oflags |= BMV_OF_SHARED;
500 		if (elen != map->br_blockcount) {
501 			*next_map = *map;
502 			next_map->br_startblock += elen;
503 			next_map->br_startoff += elen;
504 			next_map->br_blockcount -= elen;
505 		}
506 		map->br_blockcount -= elen;
507 	} else {
508 		/*
509 		 * There's an unshared extent (agbno, ebno - agbno)
510 		 * followed by shared extent at (ebno, elen).  Shrink
511 		 * the reported extent length to cover only the unshared
512 		 * extent and prepare to move up the start of map[i] to
513 		 * ebno, with the aim of (re)formatting the new map[i]
514 		 * the next time through the inner loop.
515 		 */
516 		*next_map = *map;
517 		nlen = ebno - agbno;
518 		out->bmv_length = XFS_FSB_TO_BB(mp, nlen);
519 		next_map->br_startblock += nlen;
520 		next_map->br_startoff += nlen;
521 		next_map->br_blockcount -= nlen;
522 		map->br_blockcount -= nlen;
523 	}
524 
525 	return 0;
526 }
527 
528 /*
529  * Get inode's extents as described in bmv, and format for output.
530  * Calls formatter to fill the user's buffer until all extents
531  * are mapped, until the passed-in bmv->bmv_count slots have
532  * been filled, or until the formatter short-circuits the loop,
533  * if it is tracking filled-in extents on its own.
534  */
535 int						/* error code */
536 xfs_getbmap(
537 	xfs_inode_t		*ip,
538 	struct getbmapx		*bmv,		/* user bmap structure */
539 	xfs_bmap_format_t	formatter,	/* format to user */
540 	void			*arg)		/* formatter arg */
541 {
542 	int64_t			bmvend;		/* last block requested */
543 	int			error = 0;	/* return value */
544 	int64_t			fixlen;		/* length for -1 case */
545 	int			i;		/* extent number */
546 	int			lock;		/* lock state */
547 	xfs_bmbt_irec_t		*map;		/* buffer for user's data */
548 	xfs_mount_t		*mp;		/* file system mount point */
549 	int			nex;		/* # of user extents can do */
550 	int			subnex;		/* # of bmapi's can do */
551 	int			nmap;		/* number of map entries */
552 	struct getbmapx		*out;		/* output structure */
553 	int			whichfork;	/* data or attr fork */
554 	int			prealloced;	/* this is a file with
555 						 * preallocated data space */
556 	int			iflags;		/* interface flags */
557 	int			bmapi_flags;	/* flags for xfs_bmapi */
558 	int			cur_ext = 0;
559 	struct xfs_bmbt_irec	inject_map;
560 
561 	mp = ip->i_mount;
562 	iflags = bmv->bmv_iflags;
563 
564 #ifndef DEBUG
565 	/* Only allow CoW fork queries if we're debugging. */
566 	if (iflags & BMV_IF_COWFORK)
567 		return -EINVAL;
568 #endif
569 	if ((iflags & BMV_IF_ATTRFORK) && (iflags & BMV_IF_COWFORK))
570 		return -EINVAL;
571 
572 	if (iflags & BMV_IF_ATTRFORK)
573 		whichfork = XFS_ATTR_FORK;
574 	else if (iflags & BMV_IF_COWFORK)
575 		whichfork = XFS_COW_FORK;
576 	else
577 		whichfork = XFS_DATA_FORK;
578 
579 	switch (whichfork) {
580 	case XFS_ATTR_FORK:
581 		if (XFS_IFORK_Q(ip)) {
582 			if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
583 			    ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
584 			    ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
585 				return -EINVAL;
586 		} else if (unlikely(
587 			   ip->i_d.di_aformat != 0 &&
588 			   ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
589 			XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
590 					 ip->i_mount);
591 			return -EFSCORRUPTED;
592 		}
593 
594 		prealloced = 0;
595 		fixlen = 1LL << 32;
596 		break;
597 	case XFS_COW_FORK:
598 		if (ip->i_cformat != XFS_DINODE_FMT_EXTENTS)
599 			return -EINVAL;
600 
601 		if (xfs_get_cowextsz_hint(ip)) {
602 			prealloced = 1;
603 			fixlen = mp->m_super->s_maxbytes;
604 		} else {
605 			prealloced = 0;
606 			fixlen = XFS_ISIZE(ip);
607 		}
608 		break;
609 	default:
610 		/* Local format data forks report no extents. */
611 		if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL) {
612 			bmv->bmv_entries = 0;
613 			return 0;
614 		}
615 		if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
616 		    ip->i_d.di_format != XFS_DINODE_FMT_BTREE)
617 			return -EINVAL;
618 
619 		if (xfs_get_extsz_hint(ip) ||
620 		    ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
621 			prealloced = 1;
622 			fixlen = mp->m_super->s_maxbytes;
623 		} else {
624 			prealloced = 0;
625 			fixlen = XFS_ISIZE(ip);
626 		}
627 		break;
628 	}
629 
630 	if (bmv->bmv_length == -1) {
631 		fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
632 		bmv->bmv_length =
633 			max_t(int64_t, fixlen - bmv->bmv_offset, 0);
634 	} else if (bmv->bmv_length == 0) {
635 		bmv->bmv_entries = 0;
636 		return 0;
637 	} else if (bmv->bmv_length < 0) {
638 		return -EINVAL;
639 	}
640 
641 	nex = bmv->bmv_count - 1;
642 	if (nex <= 0)
643 		return -EINVAL;
644 	bmvend = bmv->bmv_offset + bmv->bmv_length;
645 
646 
647 	if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
648 		return -ENOMEM;
649 	out = kmem_zalloc_large(bmv->bmv_count * sizeof(struct getbmapx), 0);
650 	if (!out)
651 		return -ENOMEM;
652 
653 	xfs_ilock(ip, XFS_IOLOCK_SHARED);
654 	switch (whichfork) {
655 	case XFS_DATA_FORK:
656 		if (!(iflags & BMV_IF_DELALLOC) &&
657 		    (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) {
658 			error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
659 			if (error)
660 				goto out_unlock_iolock;
661 
662 			/*
663 			 * Even after flushing the inode, there can still be
664 			 * delalloc blocks on the inode beyond EOF due to
665 			 * speculative preallocation.  These are not removed
666 			 * until the release function is called or the inode
667 			 * is inactivated.  Hence we cannot assert here that
668 			 * ip->i_delayed_blks == 0.
669 			 */
670 		}
671 
672 		lock = xfs_ilock_data_map_shared(ip);
673 		break;
674 	case XFS_COW_FORK:
675 		lock = XFS_ILOCK_SHARED;
676 		xfs_ilock(ip, lock);
677 		break;
678 	case XFS_ATTR_FORK:
679 		lock = xfs_ilock_attr_map_shared(ip);
680 		break;
681 	}
682 
683 	/*
684 	 * Don't let nex be bigger than the number of extents
685 	 * we can have assuming alternating holes and real extents.
686 	 */
687 	if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
688 		nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
689 
690 	bmapi_flags = xfs_bmapi_aflag(whichfork);
691 	if (!(iflags & BMV_IF_PREALLOC))
692 		bmapi_flags |= XFS_BMAPI_IGSTATE;
693 
694 	/*
695 	 * Allocate enough space to handle "subnex" maps at a time.
696 	 */
697 	error = -ENOMEM;
698 	subnex = 16;
699 	map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
700 	if (!map)
701 		goto out_unlock_ilock;
702 
703 	bmv->bmv_entries = 0;
704 
705 	if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
706 	    (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
707 		error = 0;
708 		goto out_free_map;
709 	}
710 
711 	do {
712 		nmap = (nex> subnex) ? subnex : nex;
713 		error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
714 				       XFS_BB_TO_FSB(mp, bmv->bmv_length),
715 				       map, &nmap, bmapi_flags);
716 		if (error)
717 			goto out_free_map;
718 		ASSERT(nmap <= subnex);
719 
720 		for (i = 0; i < nmap && bmv->bmv_length &&
721 				cur_ext < bmv->bmv_count - 1; i++) {
722 			out[cur_ext].bmv_oflags = 0;
723 			if (map[i].br_state == XFS_EXT_UNWRITTEN)
724 				out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
725 			else if (map[i].br_startblock == DELAYSTARTBLOCK)
726 				out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
727 			out[cur_ext].bmv_offset =
728 				XFS_FSB_TO_BB(mp, map[i].br_startoff);
729 			out[cur_ext].bmv_length =
730 				XFS_FSB_TO_BB(mp, map[i].br_blockcount);
731 			out[cur_ext].bmv_unused1 = 0;
732 			out[cur_ext].bmv_unused2 = 0;
733 
734 			/*
735 			 * delayed allocation extents that start beyond EOF can
736 			 * occur due to speculative EOF allocation when the
737 			 * delalloc extent is larger than the largest freespace
738 			 * extent at conversion time. These extents cannot be
739 			 * converted by data writeback, so can exist here even
740 			 * if we are not supposed to be finding delalloc
741 			 * extents.
742 			 */
743 			if (map[i].br_startblock == DELAYSTARTBLOCK &&
744 			    map[i].br_startoff < XFS_B_TO_FSB(mp, XFS_ISIZE(ip)))
745 				ASSERT((iflags & BMV_IF_DELALLOC) != 0);
746 
747                         if (map[i].br_startblock == HOLESTARTBLOCK &&
748 			    whichfork == XFS_ATTR_FORK) {
749 				/* came to the end of attribute fork */
750 				out[cur_ext].bmv_oflags |= BMV_OF_LAST;
751 				goto out_free_map;
752 			}
753 
754 			/* Is this a shared block? */
755 			error = xfs_getbmap_adjust_shared(ip, whichfork,
756 					&map[i], &out[cur_ext], &inject_map);
757 			if (error)
758 				goto out_free_map;
759 
760 			if (!xfs_getbmapx_fix_eof_hole(ip, whichfork,
761 					&out[cur_ext], prealloced, bmvend,
762 					map[i].br_startblock,
763 					inject_map.br_startblock != NULLFSBLOCK))
764 				goto out_free_map;
765 
766 			bmv->bmv_offset =
767 				out[cur_ext].bmv_offset +
768 				out[cur_ext].bmv_length;
769 			bmv->bmv_length =
770 				max_t(int64_t, 0, bmvend - bmv->bmv_offset);
771 
772 			/*
773 			 * In case we don't want to return the hole,
774 			 * don't increase cur_ext so that we can reuse
775 			 * it in the next loop.
776 			 */
777 			if ((iflags & BMV_IF_NO_HOLES) &&
778 			    map[i].br_startblock == HOLESTARTBLOCK) {
779 				memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
780 				continue;
781 			}
782 
783 			/*
784 			 * In order to report shared extents accurately,
785 			 * we report each distinct shared/unshared part
786 			 * of a single bmbt record using multiple bmap
787 			 * extents.  To make that happen, we iterate the
788 			 * same map array item multiple times, each
789 			 * time trimming out the subextent that we just
790 			 * reported.
791 			 *
792 			 * Because of this, we must check the out array
793 			 * index (cur_ext) directly against bmv_count-1
794 			 * to avoid overflows.
795 			 */
796 			if (inject_map.br_startblock != NULLFSBLOCK) {
797 				map[i] = inject_map;
798 				i--;
799 			}
800 			bmv->bmv_entries++;
801 			cur_ext++;
802 		}
803 	} while (nmap && bmv->bmv_length && cur_ext < bmv->bmv_count - 1);
804 
805  out_free_map:
806 	kmem_free(map);
807  out_unlock_ilock:
808 	xfs_iunlock(ip, lock);
809  out_unlock_iolock:
810 	xfs_iunlock(ip, XFS_IOLOCK_SHARED);
811 
812 	for (i = 0; i < cur_ext; i++) {
813 		/* format results & advance arg */
814 		error = formatter(&arg, &out[i]);
815 		if (error)
816 			break;
817 	}
818 
819 	kmem_free(out);
820 	return error;
821 }
822 
823 /*
824  * dead simple method of punching delalyed allocation blocks from a range in
825  * the inode. Walks a block at a time so will be slow, but is only executed in
826  * rare error cases so the overhead is not critical. This will always punch out
827  * both the start and end blocks, even if the ranges only partially overlap
828  * them, so it is up to the caller to ensure that partial blocks are not
829  * passed in.
830  */
831 int
832 xfs_bmap_punch_delalloc_range(
833 	struct xfs_inode	*ip,
834 	xfs_fileoff_t		start_fsb,
835 	xfs_fileoff_t		length)
836 {
837 	xfs_fileoff_t		remaining = length;
838 	int			error = 0;
839 
840 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
841 
842 	do {
843 		int		done;
844 		xfs_bmbt_irec_t	imap;
845 		int		nimaps = 1;
846 		xfs_fsblock_t	firstblock;
847 		struct xfs_defer_ops dfops;
848 
849 		/*
850 		 * Map the range first and check that it is a delalloc extent
851 		 * before trying to unmap the range. Otherwise we will be
852 		 * trying to remove a real extent (which requires a
853 		 * transaction) or a hole, which is probably a bad idea...
854 		 */
855 		error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
856 				       XFS_BMAPI_ENTIRE);
857 
858 		if (error) {
859 			/* something screwed, just bail */
860 			if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
861 				xfs_alert(ip->i_mount,
862 			"Failed delalloc mapping lookup ino %lld fsb %lld.",
863 						ip->i_ino, start_fsb);
864 			}
865 			break;
866 		}
867 		if (!nimaps) {
868 			/* nothing there */
869 			goto next_block;
870 		}
871 		if (imap.br_startblock != DELAYSTARTBLOCK) {
872 			/* been converted, ignore */
873 			goto next_block;
874 		}
875 		WARN_ON(imap.br_blockcount == 0);
876 
877 		/*
878 		 * Note: while we initialise the firstblock/dfops pair, they
879 		 * should never be used because blocks should never be
880 		 * allocated or freed for a delalloc extent and hence we need
881 		 * don't cancel or finish them after the xfs_bunmapi() call.
882 		 */
883 		xfs_defer_init(&dfops, &firstblock);
884 		error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
885 					&dfops, &done);
886 		if (error)
887 			break;
888 
889 		ASSERT(!xfs_defer_has_unfinished_work(&dfops));
890 next_block:
891 		start_fsb++;
892 		remaining--;
893 	} while(remaining > 0);
894 
895 	return error;
896 }
897 
898 /*
899  * Test whether it is appropriate to check an inode for and free post EOF
900  * blocks. The 'force' parameter determines whether we should also consider
901  * regular files that are marked preallocated or append-only.
902  */
903 bool
904 xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
905 {
906 	/* prealloc/delalloc exists only on regular files */
907 	if (!S_ISREG(VFS_I(ip)->i_mode))
908 		return false;
909 
910 	/*
911 	 * Zero sized files with no cached pages and delalloc blocks will not
912 	 * have speculative prealloc/delalloc blocks to remove.
913 	 */
914 	if (VFS_I(ip)->i_size == 0 &&
915 	    VFS_I(ip)->i_mapping->nrpages == 0 &&
916 	    ip->i_delayed_blks == 0)
917 		return false;
918 
919 	/* If we haven't read in the extent list, then don't do it now. */
920 	if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
921 		return false;
922 
923 	/*
924 	 * Do not free real preallocated or append-only files unless the file
925 	 * has delalloc blocks and we are forced to remove them.
926 	 */
927 	if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
928 		if (!force || ip->i_delayed_blks == 0)
929 			return false;
930 
931 	return true;
932 }
933 
934 /*
935  * This is called to free any blocks beyond eof. The caller must hold
936  * IOLOCK_EXCL unless we are in the inode reclaim path and have the only
937  * reference to the inode.
938  */
939 int
940 xfs_free_eofblocks(
941 	struct xfs_inode	*ip)
942 {
943 	struct xfs_trans	*tp;
944 	int			error;
945 	xfs_fileoff_t		end_fsb;
946 	xfs_fileoff_t		last_fsb;
947 	xfs_filblks_t		map_len;
948 	int			nimaps;
949 	struct xfs_bmbt_irec	imap;
950 	struct xfs_mount	*mp = ip->i_mount;
951 
952 	/*
953 	 * Figure out if there are any blocks beyond the end
954 	 * of the file.  If not, then there is nothing to do.
955 	 */
956 	end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
957 	last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
958 	if (last_fsb <= end_fsb)
959 		return 0;
960 	map_len = last_fsb - end_fsb;
961 
962 	nimaps = 1;
963 	xfs_ilock(ip, XFS_ILOCK_SHARED);
964 	error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
965 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
966 
967 	/*
968 	 * If there are blocks after the end of file, truncate the file to its
969 	 * current size to free them up.
970 	 */
971 	if (!error && (nimaps != 0) &&
972 	    (imap.br_startblock != HOLESTARTBLOCK ||
973 	     ip->i_delayed_blks)) {
974 		/*
975 		 * Attach the dquots to the inode up front.
976 		 */
977 		error = xfs_qm_dqattach(ip, 0);
978 		if (error)
979 			return error;
980 
981 		/* wait on dio to ensure i_size has settled */
982 		inode_dio_wait(VFS_I(ip));
983 
984 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0,
985 				&tp);
986 		if (error) {
987 			ASSERT(XFS_FORCED_SHUTDOWN(mp));
988 			return error;
989 		}
990 
991 		xfs_ilock(ip, XFS_ILOCK_EXCL);
992 		xfs_trans_ijoin(tp, ip, 0);
993 
994 		/*
995 		 * Do not update the on-disk file size.  If we update the
996 		 * on-disk file size and then the system crashes before the
997 		 * contents of the file are flushed to disk then the files
998 		 * may be full of holes (ie NULL files bug).
999 		 */
1000 		error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK,
1001 					      XFS_ISIZE(ip));
1002 		if (error) {
1003 			/*
1004 			 * If we get an error at this point we simply don't
1005 			 * bother truncating the file.
1006 			 */
1007 			xfs_trans_cancel(tp);
1008 		} else {
1009 			error = xfs_trans_commit(tp);
1010 			if (!error)
1011 				xfs_inode_clear_eofblocks_tag(ip);
1012 		}
1013 
1014 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
1015 	}
1016 	return error;
1017 }
1018 
1019 int
1020 xfs_alloc_file_space(
1021 	struct xfs_inode	*ip,
1022 	xfs_off_t		offset,
1023 	xfs_off_t		len,
1024 	int			alloc_type)
1025 {
1026 	xfs_mount_t		*mp = ip->i_mount;
1027 	xfs_off_t		count;
1028 	xfs_filblks_t		allocated_fsb;
1029 	xfs_filblks_t		allocatesize_fsb;
1030 	xfs_extlen_t		extsz, temp;
1031 	xfs_fileoff_t		startoffset_fsb;
1032 	xfs_fsblock_t		firstfsb;
1033 	int			nimaps;
1034 	int			quota_flag;
1035 	int			rt;
1036 	xfs_trans_t		*tp;
1037 	xfs_bmbt_irec_t		imaps[1], *imapp;
1038 	struct xfs_defer_ops	dfops;
1039 	uint			qblocks, resblks, resrtextents;
1040 	int			error;
1041 
1042 	trace_xfs_alloc_file_space(ip);
1043 
1044 	if (XFS_FORCED_SHUTDOWN(mp))
1045 		return -EIO;
1046 
1047 	error = xfs_qm_dqattach(ip, 0);
1048 	if (error)
1049 		return error;
1050 
1051 	if (len <= 0)
1052 		return -EINVAL;
1053 
1054 	rt = XFS_IS_REALTIME_INODE(ip);
1055 	extsz = xfs_get_extsz_hint(ip);
1056 
1057 	count = len;
1058 	imapp = &imaps[0];
1059 	nimaps = 1;
1060 	startoffset_fsb	= XFS_B_TO_FSBT(mp, offset);
1061 	allocatesize_fsb = XFS_B_TO_FSB(mp, count);
1062 
1063 	/*
1064 	 * Allocate file space until done or until there is an error
1065 	 */
1066 	while (allocatesize_fsb && !error) {
1067 		xfs_fileoff_t	s, e;
1068 
1069 		/*
1070 		 * Determine space reservations for data/realtime.
1071 		 */
1072 		if (unlikely(extsz)) {
1073 			s = startoffset_fsb;
1074 			do_div(s, extsz);
1075 			s *= extsz;
1076 			e = startoffset_fsb + allocatesize_fsb;
1077 			if ((temp = do_mod(startoffset_fsb, extsz)))
1078 				e += temp;
1079 			if ((temp = do_mod(e, extsz)))
1080 				e += extsz - temp;
1081 		} else {
1082 			s = 0;
1083 			e = allocatesize_fsb;
1084 		}
1085 
1086 		/*
1087 		 * The transaction reservation is limited to a 32-bit block
1088 		 * count, hence we need to limit the number of blocks we are
1089 		 * trying to reserve to avoid an overflow. We can't allocate
1090 		 * more than @nimaps extents, and an extent is limited on disk
1091 		 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
1092 		 */
1093 		resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
1094 		if (unlikely(rt)) {
1095 			resrtextents = qblocks = resblks;
1096 			resrtextents /= mp->m_sb.sb_rextsize;
1097 			resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1098 			quota_flag = XFS_QMOPT_RES_RTBLKS;
1099 		} else {
1100 			resrtextents = 0;
1101 			resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
1102 			quota_flag = XFS_QMOPT_RES_REGBLKS;
1103 		}
1104 
1105 		/*
1106 		 * Allocate and setup the transaction.
1107 		 */
1108 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks,
1109 				resrtextents, 0, &tp);
1110 
1111 		/*
1112 		 * Check for running out of space
1113 		 */
1114 		if (error) {
1115 			/*
1116 			 * Free the transaction structure.
1117 			 */
1118 			ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1119 			break;
1120 		}
1121 		xfs_ilock(ip, XFS_ILOCK_EXCL);
1122 		error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
1123 						      0, quota_flag);
1124 		if (error)
1125 			goto error1;
1126 
1127 		xfs_trans_ijoin(tp, ip, 0);
1128 
1129 		xfs_defer_init(&dfops, &firstfsb);
1130 		error = xfs_bmapi_write(tp, ip, startoffset_fsb,
1131 					allocatesize_fsb, alloc_type, &firstfsb,
1132 					resblks, imapp, &nimaps, &dfops);
1133 		if (error)
1134 			goto error0;
1135 
1136 		/*
1137 		 * Complete the transaction
1138 		 */
1139 		error = xfs_defer_finish(&tp, &dfops, NULL);
1140 		if (error)
1141 			goto error0;
1142 
1143 		error = xfs_trans_commit(tp);
1144 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
1145 		if (error)
1146 			break;
1147 
1148 		allocated_fsb = imapp->br_blockcount;
1149 
1150 		if (nimaps == 0) {
1151 			error = -ENOSPC;
1152 			break;
1153 		}
1154 
1155 		startoffset_fsb += allocated_fsb;
1156 		allocatesize_fsb -= allocated_fsb;
1157 	}
1158 
1159 	return error;
1160 
1161 error0:	/* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1162 	xfs_defer_cancel(&dfops);
1163 	xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
1164 
1165 error1:	/* Just cancel transaction */
1166 	xfs_trans_cancel(tp);
1167 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1168 	return error;
1169 }
1170 
1171 static int
1172 xfs_unmap_extent(
1173 	struct xfs_inode	*ip,
1174 	xfs_fileoff_t		startoffset_fsb,
1175 	xfs_filblks_t		len_fsb,
1176 	int			*done)
1177 {
1178 	struct xfs_mount	*mp = ip->i_mount;
1179 	struct xfs_trans	*tp;
1180 	struct xfs_defer_ops	dfops;
1181 	xfs_fsblock_t		firstfsb;
1182 	uint			resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1183 	int			error;
1184 
1185 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
1186 	if (error) {
1187 		ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1188 		return error;
1189 	}
1190 
1191 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1192 	error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot, ip->i_gdquot,
1193 			ip->i_pdquot, resblks, 0, XFS_QMOPT_RES_REGBLKS);
1194 	if (error)
1195 		goto out_trans_cancel;
1196 
1197 	xfs_trans_ijoin(tp, ip, 0);
1198 
1199 	xfs_defer_init(&dfops, &firstfsb);
1200 	error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, &firstfsb,
1201 			&dfops, done);
1202 	if (error)
1203 		goto out_bmap_cancel;
1204 
1205 	error = xfs_defer_finish(&tp, &dfops, ip);
1206 	if (error)
1207 		goto out_bmap_cancel;
1208 
1209 	error = xfs_trans_commit(tp);
1210 out_unlock:
1211 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1212 	return error;
1213 
1214 out_bmap_cancel:
1215 	xfs_defer_cancel(&dfops);
1216 out_trans_cancel:
1217 	xfs_trans_cancel(tp);
1218 	goto out_unlock;
1219 }
1220 
1221 static int
1222 xfs_adjust_extent_unmap_boundaries(
1223 	struct xfs_inode	*ip,
1224 	xfs_fileoff_t		*startoffset_fsb,
1225 	xfs_fileoff_t		*endoffset_fsb)
1226 {
1227 	struct xfs_mount	*mp = ip->i_mount;
1228 	struct xfs_bmbt_irec	imap;
1229 	int			nimap, error;
1230 	xfs_extlen_t		mod = 0;
1231 
1232 	nimap = 1;
1233 	error = xfs_bmapi_read(ip, *startoffset_fsb, 1, &imap, &nimap, 0);
1234 	if (error)
1235 		return error;
1236 
1237 	if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1238 		ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1239 		mod = do_mod(imap.br_startblock, mp->m_sb.sb_rextsize);
1240 		if (mod)
1241 			*startoffset_fsb += mp->m_sb.sb_rextsize - mod;
1242 	}
1243 
1244 	nimap = 1;
1245 	error = xfs_bmapi_read(ip, *endoffset_fsb - 1, 1, &imap, &nimap, 0);
1246 	if (error)
1247 		return error;
1248 
1249 	if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1250 		ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1251 		mod++;
1252 		if (mod && mod != mp->m_sb.sb_rextsize)
1253 			*endoffset_fsb -= mod;
1254 	}
1255 
1256 	return 0;
1257 }
1258 
1259 static int
1260 xfs_flush_unmap_range(
1261 	struct xfs_inode	*ip,
1262 	xfs_off_t		offset,
1263 	xfs_off_t		len)
1264 {
1265 	struct xfs_mount	*mp = ip->i_mount;
1266 	struct inode		*inode = VFS_I(ip);
1267 	xfs_off_t		rounding, start, end;
1268 	int			error;
1269 
1270 	/* wait for the completion of any pending DIOs */
1271 	inode_dio_wait(inode);
1272 
1273 	rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_SIZE);
1274 	start = round_down(offset, rounding);
1275 	end = round_up(offset + len, rounding) - 1;
1276 
1277 	error = filemap_write_and_wait_range(inode->i_mapping, start, end);
1278 	if (error)
1279 		return error;
1280 	truncate_pagecache_range(inode, start, end);
1281 	return 0;
1282 }
1283 
1284 int
1285 xfs_free_file_space(
1286 	struct xfs_inode	*ip,
1287 	xfs_off_t		offset,
1288 	xfs_off_t		len)
1289 {
1290 	struct xfs_mount	*mp = ip->i_mount;
1291 	xfs_fileoff_t		startoffset_fsb;
1292 	xfs_fileoff_t		endoffset_fsb;
1293 	int			done = 0, error;
1294 
1295 	trace_xfs_free_file_space(ip);
1296 
1297 	error = xfs_qm_dqattach(ip, 0);
1298 	if (error)
1299 		return error;
1300 
1301 	if (len <= 0)	/* if nothing being freed */
1302 		return 0;
1303 
1304 	error = xfs_flush_unmap_range(ip, offset, len);
1305 	if (error)
1306 		return error;
1307 
1308 	startoffset_fsb = XFS_B_TO_FSB(mp, offset);
1309 	endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
1310 
1311 	/*
1312 	 * Need to zero the stuff we're not freeing, on disk.  If it's a RT file
1313 	 * and we can't use unwritten extents then we actually need to ensure
1314 	 * to zero the whole extent, otherwise we just need to take of block
1315 	 * boundaries, and xfs_bunmapi will handle the rest.
1316 	 */
1317 	if (XFS_IS_REALTIME_INODE(ip) &&
1318 	    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
1319 		error = xfs_adjust_extent_unmap_boundaries(ip, &startoffset_fsb,
1320 				&endoffset_fsb);
1321 		if (error)
1322 			return error;
1323 	}
1324 
1325 	if (endoffset_fsb > startoffset_fsb) {
1326 		while (!done) {
1327 			error = xfs_unmap_extent(ip, startoffset_fsb,
1328 					endoffset_fsb - startoffset_fsb, &done);
1329 			if (error)
1330 				return error;
1331 		}
1332 	}
1333 
1334 	/*
1335 	 * Now that we've unmap all full blocks we'll have to zero out any
1336 	 * partial block at the beginning and/or end.  xfs_zero_range is
1337 	 * smart enough to skip any holes, including those we just created,
1338 	 * but we must take care not to zero beyond EOF and enlarge i_size.
1339 	 */
1340 
1341 	if (offset >= XFS_ISIZE(ip))
1342 		return 0;
1343 
1344 	if (offset + len > XFS_ISIZE(ip))
1345 		len = XFS_ISIZE(ip) - offset;
1346 
1347 	return xfs_zero_range(ip, offset, len, NULL);
1348 }
1349 
1350 /*
1351  * Preallocate and zero a range of a file. This mechanism has the allocation
1352  * semantics of fallocate and in addition converts data in the range to zeroes.
1353  */
1354 int
1355 xfs_zero_file_space(
1356 	struct xfs_inode	*ip,
1357 	xfs_off_t		offset,
1358 	xfs_off_t		len)
1359 {
1360 	struct xfs_mount	*mp = ip->i_mount;
1361 	uint			blksize;
1362 	int			error;
1363 
1364 	trace_xfs_zero_file_space(ip);
1365 
1366 	blksize = 1 << mp->m_sb.sb_blocklog;
1367 
1368 	/*
1369 	 * Punch a hole and prealloc the range. We use hole punch rather than
1370 	 * unwritten extent conversion for two reasons:
1371 	 *
1372 	 * 1.) Hole punch handles partial block zeroing for us.
1373 	 *
1374 	 * 2.) If prealloc returns ENOSPC, the file range is still zero-valued
1375 	 * by virtue of the hole punch.
1376 	 */
1377 	error = xfs_free_file_space(ip, offset, len);
1378 	if (error)
1379 		goto out;
1380 
1381 	error = xfs_alloc_file_space(ip, round_down(offset, blksize),
1382 				     round_up(offset + len, blksize) -
1383 				     round_down(offset, blksize),
1384 				     XFS_BMAPI_PREALLOC);
1385 out:
1386 	return error;
1387 
1388 }
1389 
1390 /*
1391  * @next_fsb will keep track of the extent currently undergoing shift.
1392  * @stop_fsb will keep track of the extent at which we have to stop.
1393  * If we are shifting left, we will start with block (offset + len) and
1394  * shift each extent till last extent.
1395  * If we are shifting right, we will start with last extent inside file space
1396  * and continue until we reach the block corresponding to offset.
1397  */
1398 static int
1399 xfs_shift_file_space(
1400 	struct xfs_inode        *ip,
1401 	xfs_off_t               offset,
1402 	xfs_off_t               len,
1403 	enum shift_direction	direction)
1404 {
1405 	int			done = 0;
1406 	struct xfs_mount	*mp = ip->i_mount;
1407 	struct xfs_trans	*tp;
1408 	int			error;
1409 	struct xfs_defer_ops	dfops;
1410 	xfs_fsblock_t		first_block;
1411 	xfs_fileoff_t		stop_fsb;
1412 	xfs_fileoff_t		next_fsb;
1413 	xfs_fileoff_t		shift_fsb;
1414 	uint			resblks;
1415 
1416 	ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT);
1417 
1418 	if (direction == SHIFT_LEFT) {
1419 		/*
1420 		 * Reserve blocks to cover potential extent merges after left
1421 		 * shift operations.
1422 		 */
1423 		resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1424 		next_fsb = XFS_B_TO_FSB(mp, offset + len);
1425 		stop_fsb = XFS_B_TO_FSB(mp, VFS_I(ip)->i_size);
1426 	} else {
1427 		/*
1428 		 * If right shift, delegate the work of initialization of
1429 		 * next_fsb to xfs_bmap_shift_extent as it has ilock held.
1430 		 */
1431 		resblks = 0;
1432 		next_fsb = NULLFSBLOCK;
1433 		stop_fsb = XFS_B_TO_FSB(mp, offset);
1434 	}
1435 
1436 	shift_fsb = XFS_B_TO_FSB(mp, len);
1437 
1438 	/*
1439 	 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
1440 	 * into the accessible region of the file.
1441 	 */
1442 	if (xfs_can_free_eofblocks(ip, true)) {
1443 		error = xfs_free_eofblocks(ip);
1444 		if (error)
1445 			return error;
1446 	}
1447 
1448 	/*
1449 	 * Writeback and invalidate cache for the remainder of the file as we're
1450 	 * about to shift down every extent from offset to EOF.
1451 	 */
1452 	error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1453 					     offset, -1);
1454 	if (error)
1455 		return error;
1456 	error = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
1457 					offset >> PAGE_SHIFT, -1);
1458 	if (error)
1459 		return error;
1460 
1461 	/*
1462 	 * The extent shiting code works on extent granularity. So, if
1463 	 * stop_fsb is not the starting block of extent, we need to split
1464 	 * the extent at stop_fsb.
1465 	 */
1466 	if (direction == SHIFT_RIGHT) {
1467 		error = xfs_bmap_split_extent(ip, stop_fsb);
1468 		if (error)
1469 			return error;
1470 	}
1471 
1472 	while (!error && !done) {
1473 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0,
1474 					&tp);
1475 		if (error)
1476 			break;
1477 
1478 		xfs_ilock(ip, XFS_ILOCK_EXCL);
1479 		error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot,
1480 				ip->i_gdquot, ip->i_pdquot, resblks, 0,
1481 				XFS_QMOPT_RES_REGBLKS);
1482 		if (error)
1483 			goto out_trans_cancel;
1484 
1485 		xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1486 
1487 		xfs_defer_init(&dfops, &first_block);
1488 
1489 		/*
1490 		 * We are using the write transaction in which max 2 bmbt
1491 		 * updates are allowed
1492 		 */
1493 		error = xfs_bmap_shift_extents(tp, ip, &next_fsb, shift_fsb,
1494 				&done, stop_fsb, &first_block, &dfops,
1495 				direction, XFS_BMAP_MAX_SHIFT_EXTENTS);
1496 		if (error)
1497 			goto out_bmap_cancel;
1498 
1499 		error = xfs_defer_finish(&tp, &dfops, NULL);
1500 		if (error)
1501 			goto out_bmap_cancel;
1502 
1503 		error = xfs_trans_commit(tp);
1504 	}
1505 
1506 	return error;
1507 
1508 out_bmap_cancel:
1509 	xfs_defer_cancel(&dfops);
1510 out_trans_cancel:
1511 	xfs_trans_cancel(tp);
1512 	return error;
1513 }
1514 
1515 /*
1516  * xfs_collapse_file_space()
1517  *	This routine frees disk space and shift extent for the given file.
1518  *	The first thing we do is to free data blocks in the specified range
1519  *	by calling xfs_free_file_space(). It would also sync dirty data
1520  *	and invalidate page cache over the region on which collapse range
1521  *	is working. And Shift extent records to the left to cover a hole.
1522  * RETURNS:
1523  *	0 on success
1524  *	errno on error
1525  *
1526  */
1527 int
1528 xfs_collapse_file_space(
1529 	struct xfs_inode	*ip,
1530 	xfs_off_t		offset,
1531 	xfs_off_t		len)
1532 {
1533 	int error;
1534 
1535 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1536 	trace_xfs_collapse_file_space(ip);
1537 
1538 	error = xfs_free_file_space(ip, offset, len);
1539 	if (error)
1540 		return error;
1541 
1542 	return xfs_shift_file_space(ip, offset, len, SHIFT_LEFT);
1543 }
1544 
1545 /*
1546  * xfs_insert_file_space()
1547  *	This routine create hole space by shifting extents for the given file.
1548  *	The first thing we do is to sync dirty data and invalidate page cache
1549  *	over the region on which insert range is working. And split an extent
1550  *	to two extents at given offset by calling xfs_bmap_split_extent.
1551  *	And shift all extent records which are laying between [offset,
1552  *	last allocated extent] to the right to reserve hole range.
1553  * RETURNS:
1554  *	0 on success
1555  *	errno on error
1556  */
1557 int
1558 xfs_insert_file_space(
1559 	struct xfs_inode	*ip,
1560 	loff_t			offset,
1561 	loff_t			len)
1562 {
1563 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1564 	trace_xfs_insert_file_space(ip);
1565 
1566 	return xfs_shift_file_space(ip, offset, len, SHIFT_RIGHT);
1567 }
1568 
1569 /*
1570  * We need to check that the format of the data fork in the temporary inode is
1571  * valid for the target inode before doing the swap. This is not a problem with
1572  * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1573  * data fork depending on the space the attribute fork is taking so we can get
1574  * invalid formats on the target inode.
1575  *
1576  * E.g. target has space for 7 extents in extent format, temp inode only has
1577  * space for 6.  If we defragment down to 7 extents, then the tmp format is a
1578  * btree, but when swapped it needs to be in extent format. Hence we can't just
1579  * blindly swap data forks on attr2 filesystems.
1580  *
1581  * Note that we check the swap in both directions so that we don't end up with
1582  * a corrupt temporary inode, either.
1583  *
1584  * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1585  * inode will prevent this situation from occurring, so all we do here is
1586  * reject and log the attempt. basically we are putting the responsibility on
1587  * userspace to get this right.
1588  */
1589 static int
1590 xfs_swap_extents_check_format(
1591 	struct xfs_inode	*ip,	/* target inode */
1592 	struct xfs_inode	*tip)	/* tmp inode */
1593 {
1594 
1595 	/* Should never get a local format */
1596 	if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
1597 	    tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
1598 		return -EINVAL;
1599 
1600 	/*
1601 	 * if the target inode has less extents that then temporary inode then
1602 	 * why did userspace call us?
1603 	 */
1604 	if (ip->i_d.di_nextents < tip->i_d.di_nextents)
1605 		return -EINVAL;
1606 
1607 	/*
1608 	 * If we have to use the (expensive) rmap swap method, we can
1609 	 * handle any number of extents and any format.
1610 	 */
1611 	if (xfs_sb_version_hasrmapbt(&ip->i_mount->m_sb))
1612 		return 0;
1613 
1614 	/*
1615 	 * if the target inode is in extent form and the temp inode is in btree
1616 	 * form then we will end up with the target inode in the wrong format
1617 	 * as we already know there are less extents in the temp inode.
1618 	 */
1619 	if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1620 	    tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1621 		return -EINVAL;
1622 
1623 	/* Check temp in extent form to max in target */
1624 	if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1625 	    XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) >
1626 			XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1627 		return -EINVAL;
1628 
1629 	/* Check target in extent form to max in temp */
1630 	if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1631 	    XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) >
1632 			XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1633 		return -EINVAL;
1634 
1635 	/*
1636 	 * If we are in a btree format, check that the temp root block will fit
1637 	 * in the target and that it has enough extents to be in btree format
1638 	 * in the target.
1639 	 *
1640 	 * Note that we have to be careful to allow btree->extent conversions
1641 	 * (a common defrag case) which will occur when the temp inode is in
1642 	 * extent format...
1643 	 */
1644 	if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1645 		if (XFS_IFORK_Q(ip) &&
1646 		    XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip))
1647 			return -EINVAL;
1648 		if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <=
1649 		    XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1650 			return -EINVAL;
1651 	}
1652 
1653 	/* Reciprocal target->temp btree format checks */
1654 	if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1655 		if (XFS_IFORK_Q(tip) &&
1656 		    XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
1657 			return -EINVAL;
1658 		if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <=
1659 		    XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1660 			return -EINVAL;
1661 	}
1662 
1663 	return 0;
1664 }
1665 
1666 static int
1667 xfs_swap_extent_flush(
1668 	struct xfs_inode	*ip)
1669 {
1670 	int	error;
1671 
1672 	error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1673 	if (error)
1674 		return error;
1675 	truncate_pagecache_range(VFS_I(ip), 0, -1);
1676 
1677 	/* Verify O_DIRECT for ftmp */
1678 	if (VFS_I(ip)->i_mapping->nrpages)
1679 		return -EINVAL;
1680 	return 0;
1681 }
1682 
1683 /*
1684  * Move extents from one file to another, when rmap is enabled.
1685  */
1686 STATIC int
1687 xfs_swap_extent_rmap(
1688 	struct xfs_trans		**tpp,
1689 	struct xfs_inode		*ip,
1690 	struct xfs_inode		*tip)
1691 {
1692 	struct xfs_bmbt_irec		irec;
1693 	struct xfs_bmbt_irec		uirec;
1694 	struct xfs_bmbt_irec		tirec;
1695 	xfs_fileoff_t			offset_fsb;
1696 	xfs_fileoff_t			end_fsb;
1697 	xfs_filblks_t			count_fsb;
1698 	xfs_fsblock_t			firstfsb;
1699 	struct xfs_defer_ops		dfops;
1700 	int				error;
1701 	xfs_filblks_t			ilen;
1702 	xfs_filblks_t			rlen;
1703 	int				nimaps;
1704 	uint64_t			tip_flags2;
1705 
1706 	/*
1707 	 * If the source file has shared blocks, we must flag the donor
1708 	 * file as having shared blocks so that we get the shared-block
1709 	 * rmap functions when we go to fix up the rmaps.  The flags
1710 	 * will be switch for reals later.
1711 	 */
1712 	tip_flags2 = tip->i_d.di_flags2;
1713 	if (ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)
1714 		tip->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK;
1715 
1716 	offset_fsb = 0;
1717 	end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip)));
1718 	count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
1719 
1720 	while (count_fsb) {
1721 		/* Read extent from the donor file */
1722 		nimaps = 1;
1723 		error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec,
1724 				&nimaps, 0);
1725 		if (error)
1726 			goto out;
1727 		ASSERT(nimaps == 1);
1728 		ASSERT(tirec.br_startblock != DELAYSTARTBLOCK);
1729 
1730 		trace_xfs_swap_extent_rmap_remap(tip, &tirec);
1731 		ilen = tirec.br_blockcount;
1732 
1733 		/* Unmap the old blocks in the source file. */
1734 		while (tirec.br_blockcount) {
1735 			xfs_defer_init(&dfops, &firstfsb);
1736 			trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec);
1737 
1738 			/* Read extent from the source file */
1739 			nimaps = 1;
1740 			error = xfs_bmapi_read(ip, tirec.br_startoff,
1741 					tirec.br_blockcount, &irec,
1742 					&nimaps, 0);
1743 			if (error)
1744 				goto out_defer;
1745 			ASSERT(nimaps == 1);
1746 			ASSERT(tirec.br_startoff == irec.br_startoff);
1747 			trace_xfs_swap_extent_rmap_remap_piece(ip, &irec);
1748 
1749 			/* Trim the extent. */
1750 			uirec = tirec;
1751 			uirec.br_blockcount = rlen = min_t(xfs_filblks_t,
1752 					tirec.br_blockcount,
1753 					irec.br_blockcount);
1754 			trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec);
1755 
1756 			/* Remove the mapping from the donor file. */
1757 			error = xfs_bmap_unmap_extent((*tpp)->t_mountp, &dfops,
1758 					tip, &uirec);
1759 			if (error)
1760 				goto out_defer;
1761 
1762 			/* Remove the mapping from the source file. */
1763 			error = xfs_bmap_unmap_extent((*tpp)->t_mountp, &dfops,
1764 					ip, &irec);
1765 			if (error)
1766 				goto out_defer;
1767 
1768 			/* Map the donor file's blocks into the source file. */
1769 			error = xfs_bmap_map_extent((*tpp)->t_mountp, &dfops,
1770 					ip, &uirec);
1771 			if (error)
1772 				goto out_defer;
1773 
1774 			/* Map the source file's blocks into the donor file. */
1775 			error = xfs_bmap_map_extent((*tpp)->t_mountp, &dfops,
1776 					tip, &irec);
1777 			if (error)
1778 				goto out_defer;
1779 
1780 			error = xfs_defer_finish(tpp, &dfops, ip);
1781 			if (error)
1782 				goto out_defer;
1783 
1784 			tirec.br_startoff += rlen;
1785 			if (tirec.br_startblock != HOLESTARTBLOCK &&
1786 			    tirec.br_startblock != DELAYSTARTBLOCK)
1787 				tirec.br_startblock += rlen;
1788 			tirec.br_blockcount -= rlen;
1789 		}
1790 
1791 		/* Roll on... */
1792 		count_fsb -= ilen;
1793 		offset_fsb += ilen;
1794 	}
1795 
1796 	tip->i_d.di_flags2 = tip_flags2;
1797 	return 0;
1798 
1799 out_defer:
1800 	xfs_defer_cancel(&dfops);
1801 out:
1802 	trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_);
1803 	tip->i_d.di_flags2 = tip_flags2;
1804 	return error;
1805 }
1806 
1807 /* Swap the extents of two files by swapping data forks. */
1808 STATIC int
1809 xfs_swap_extent_forks(
1810 	struct xfs_trans	*tp,
1811 	struct xfs_inode	*ip,
1812 	struct xfs_inode	*tip,
1813 	int			*src_log_flags,
1814 	int			*target_log_flags)
1815 {
1816 	struct xfs_ifork	tempifp, *ifp, *tifp;
1817 	xfs_filblks_t		aforkblks = 0;
1818 	xfs_filblks_t		taforkblks = 0;
1819 	xfs_extnum_t		junk;
1820 	xfs_extnum_t		nextents;
1821 	uint64_t		tmp;
1822 	int			error;
1823 
1824 	/*
1825 	 * Count the number of extended attribute blocks
1826 	 */
1827 	if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
1828 	     (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1829 		error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &junk,
1830 				&aforkblks);
1831 		if (error)
1832 			return error;
1833 	}
1834 	if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
1835 	     (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1836 		error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK, &junk,
1837 				&taforkblks);
1838 		if (error)
1839 			return error;
1840 	}
1841 
1842 	/*
1843 	 * Before we've swapped the forks, lets set the owners of the forks
1844 	 * appropriately. We have to do this as we are demand paging the btree
1845 	 * buffers, and so the validation done on read will expect the owner
1846 	 * field to be correctly set. Once we change the owners, we can swap the
1847 	 * inode forks.
1848 	 */
1849 	if (ip->i_d.di_version == 3 &&
1850 	    ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1851 		(*target_log_flags) |= XFS_ILOG_DOWNER;
1852 		error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK,
1853 					      tip->i_ino, NULL);
1854 		if (error)
1855 			return error;
1856 	}
1857 
1858 	if (tip->i_d.di_version == 3 &&
1859 	    tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1860 		(*src_log_flags) |= XFS_ILOG_DOWNER;
1861 		error = xfs_bmbt_change_owner(tp, tip, XFS_DATA_FORK,
1862 					      ip->i_ino, NULL);
1863 		if (error)
1864 			return error;
1865 	}
1866 
1867 	/*
1868 	 * Swap the data forks of the inodes
1869 	 */
1870 	ifp = &ip->i_df;
1871 	tifp = &tip->i_df;
1872 	tempifp = *ifp;		/* struct copy */
1873 	*ifp = *tifp;		/* struct copy */
1874 	*tifp = tempifp;	/* struct copy */
1875 
1876 	/*
1877 	 * Fix the on-disk inode values
1878 	 */
1879 	tmp = (uint64_t)ip->i_d.di_nblocks;
1880 	ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
1881 	tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
1882 
1883 	tmp = (uint64_t) ip->i_d.di_nextents;
1884 	ip->i_d.di_nextents = tip->i_d.di_nextents;
1885 	tip->i_d.di_nextents = tmp;
1886 
1887 	tmp = (uint64_t) ip->i_d.di_format;
1888 	ip->i_d.di_format = tip->i_d.di_format;
1889 	tip->i_d.di_format = tmp;
1890 
1891 	/*
1892 	 * The extents in the source inode could still contain speculative
1893 	 * preallocation beyond EOF (e.g. the file is open but not modified
1894 	 * while defrag is in progress). In that case, we need to copy over the
1895 	 * number of delalloc blocks the data fork in the source inode is
1896 	 * tracking beyond EOF so that when the fork is truncated away when the
1897 	 * temporary inode is unlinked we don't underrun the i_delayed_blks
1898 	 * counter on that inode.
1899 	 */
1900 	ASSERT(tip->i_delayed_blks == 0);
1901 	tip->i_delayed_blks = ip->i_delayed_blks;
1902 	ip->i_delayed_blks = 0;
1903 
1904 	switch (ip->i_d.di_format) {
1905 	case XFS_DINODE_FMT_EXTENTS:
1906 		/*
1907 		 * If the extents fit in the inode, fix the pointer.  Otherwise
1908 		 * it's already NULL or pointing to the extent.
1909 		 */
1910 		nextents = xfs_iext_count(&ip->i_df);
1911 		if (nextents <= XFS_INLINE_EXTS)
1912 			ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
1913 		(*src_log_flags) |= XFS_ILOG_DEXT;
1914 		break;
1915 	case XFS_DINODE_FMT_BTREE:
1916 		ASSERT(ip->i_d.di_version < 3 ||
1917 		       (*src_log_flags & XFS_ILOG_DOWNER));
1918 		(*src_log_flags) |= XFS_ILOG_DBROOT;
1919 		break;
1920 	}
1921 
1922 	switch (tip->i_d.di_format) {
1923 	case XFS_DINODE_FMT_EXTENTS:
1924 		/*
1925 		 * If the extents fit in the inode, fix the pointer.  Otherwise
1926 		 * it's already NULL or pointing to the extent.
1927 		 */
1928 		nextents = xfs_iext_count(&tip->i_df);
1929 		if (nextents <= XFS_INLINE_EXTS)
1930 			tifp->if_u1.if_extents = tifp->if_u2.if_inline_ext;
1931 		(*target_log_flags) |= XFS_ILOG_DEXT;
1932 		break;
1933 	case XFS_DINODE_FMT_BTREE:
1934 		(*target_log_flags) |= XFS_ILOG_DBROOT;
1935 		ASSERT(tip->i_d.di_version < 3 ||
1936 		       (*target_log_flags & XFS_ILOG_DOWNER));
1937 		break;
1938 	}
1939 
1940 	return 0;
1941 }
1942 
1943 int
1944 xfs_swap_extents(
1945 	struct xfs_inode	*ip,	/* target inode */
1946 	struct xfs_inode	*tip,	/* tmp inode */
1947 	struct xfs_swapext	*sxp)
1948 {
1949 	struct xfs_mount	*mp = ip->i_mount;
1950 	struct xfs_trans	*tp;
1951 	struct xfs_bstat	*sbp = &sxp->sx_stat;
1952 	int			src_log_flags, target_log_flags;
1953 	int			error = 0;
1954 	int			lock_flags;
1955 	struct xfs_ifork	*cowfp;
1956 	uint64_t		f;
1957 	int			resblks;
1958 
1959 	/*
1960 	 * Lock the inodes against other IO, page faults and truncate to
1961 	 * begin with.  Then we can ensure the inodes are flushed and have no
1962 	 * page cache safely. Once we have done this we can take the ilocks and
1963 	 * do the rest of the checks.
1964 	 */
1965 	lock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1966 	lock_flags = XFS_MMAPLOCK_EXCL;
1967 	xfs_lock_two_inodes(ip, tip, XFS_MMAPLOCK_EXCL);
1968 
1969 	/* Verify that both files have the same format */
1970 	if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) {
1971 		error = -EINVAL;
1972 		goto out_unlock;
1973 	}
1974 
1975 	/* Verify both files are either real-time or non-realtime */
1976 	if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1977 		error = -EINVAL;
1978 		goto out_unlock;
1979 	}
1980 
1981 	error = xfs_swap_extent_flush(ip);
1982 	if (error)
1983 		goto out_unlock;
1984 	error = xfs_swap_extent_flush(tip);
1985 	if (error)
1986 		goto out_unlock;
1987 
1988 	/*
1989 	 * Extent "swapping" with rmap requires a permanent reservation and
1990 	 * a block reservation because it's really just a remap operation
1991 	 * performed with log redo items!
1992 	 */
1993 	if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
1994 		/*
1995 		 * Conceptually this shouldn't affect the shape of either
1996 		 * bmbt, but since we atomically move extents one by one,
1997 		 * we reserve enough space to rebuild both trees.
1998 		 */
1999 		resblks = XFS_SWAP_RMAP_SPACE_RES(mp,
2000 				XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK),
2001 				XFS_DATA_FORK) +
2002 			  XFS_SWAP_RMAP_SPACE_RES(mp,
2003 				XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK),
2004 				XFS_DATA_FORK);
2005 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks,
2006 				0, 0, &tp);
2007 	} else
2008 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0,
2009 				0, 0, &tp);
2010 	if (error)
2011 		goto out_unlock;
2012 
2013 	/*
2014 	 * Lock and join the inodes to the tansaction so that transaction commit
2015 	 * or cancel will unlock the inodes from this point onwards.
2016 	 */
2017 	xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
2018 	lock_flags |= XFS_ILOCK_EXCL;
2019 	xfs_trans_ijoin(tp, ip, 0);
2020 	xfs_trans_ijoin(tp, tip, 0);
2021 
2022 
2023 	/* Verify all data are being swapped */
2024 	if (sxp->sx_offset != 0 ||
2025 	    sxp->sx_length != ip->i_d.di_size ||
2026 	    sxp->sx_length != tip->i_d.di_size) {
2027 		error = -EFAULT;
2028 		goto out_trans_cancel;
2029 	}
2030 
2031 	trace_xfs_swap_extent_before(ip, 0);
2032 	trace_xfs_swap_extent_before(tip, 1);
2033 
2034 	/* check inode formats now that data is flushed */
2035 	error = xfs_swap_extents_check_format(ip, tip);
2036 	if (error) {
2037 		xfs_notice(mp,
2038 		    "%s: inode 0x%llx format is incompatible for exchanging.",
2039 				__func__, ip->i_ino);
2040 		goto out_trans_cancel;
2041 	}
2042 
2043 	/*
2044 	 * Compare the current change & modify times with that
2045 	 * passed in.  If they differ, we abort this swap.
2046 	 * This is the mechanism used to ensure the calling
2047 	 * process that the file was not changed out from
2048 	 * under it.
2049 	 */
2050 	if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
2051 	    (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
2052 	    (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
2053 	    (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
2054 		error = -EBUSY;
2055 		goto out_trans_cancel;
2056 	}
2057 
2058 	/*
2059 	 * Note the trickiness in setting the log flags - we set the owner log
2060 	 * flag on the opposite inode (i.e. the inode we are setting the new
2061 	 * owner to be) because once we swap the forks and log that, log
2062 	 * recovery is going to see the fork as owned by the swapped inode,
2063 	 * not the pre-swapped inodes.
2064 	 */
2065 	src_log_flags = XFS_ILOG_CORE;
2066 	target_log_flags = XFS_ILOG_CORE;
2067 
2068 	if (xfs_sb_version_hasrmapbt(&mp->m_sb))
2069 		error = xfs_swap_extent_rmap(&tp, ip, tip);
2070 	else
2071 		error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags,
2072 				&target_log_flags);
2073 	if (error)
2074 		goto out_trans_cancel;
2075 
2076 	/* Do we have to swap reflink flags? */
2077 	if ((ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK) ^
2078 	    (tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)) {
2079 		f = ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
2080 		ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
2081 		ip->i_d.di_flags2 |= tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
2082 		tip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
2083 		tip->i_d.di_flags2 |= f & XFS_DIFLAG2_REFLINK;
2084 		cowfp = ip->i_cowfp;
2085 		ip->i_cowfp = tip->i_cowfp;
2086 		tip->i_cowfp = cowfp;
2087 		xfs_inode_set_cowblocks_tag(ip);
2088 		xfs_inode_set_cowblocks_tag(tip);
2089 	}
2090 
2091 	xfs_trans_log_inode(tp, ip,  src_log_flags);
2092 	xfs_trans_log_inode(tp, tip, target_log_flags);
2093 
2094 	/*
2095 	 * If this is a synchronous mount, make sure that the
2096 	 * transaction goes to disk before returning to the user.
2097 	 */
2098 	if (mp->m_flags & XFS_MOUNT_WSYNC)
2099 		xfs_trans_set_sync(tp);
2100 
2101 	error = xfs_trans_commit(tp);
2102 
2103 	trace_xfs_swap_extent_after(ip, 0);
2104 	trace_xfs_swap_extent_after(tip, 1);
2105 
2106 out_unlock:
2107 	xfs_iunlock(ip, lock_flags);
2108 	xfs_iunlock(tip, lock_flags);
2109 	unlock_two_nondirectories(VFS_I(ip), VFS_I(tip));
2110 	return error;
2111 
2112 out_trans_cancel:
2113 	xfs_trans_cancel(tp);
2114 	goto out_unlock;
2115 }
2116