xref: /linux/fs/xfs/xfs_bmap_util.c (revision 35219bc5c71f4197c8bd10297597de797c1eece5)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4  * Copyright (c) 2012 Red Hat, Inc.
5  * All Rights Reserved.
6  */
7 #include "xfs.h"
8 #include "xfs_fs.h"
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_bit.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_inode.h"
17 #include "xfs_btree.h"
18 #include "xfs_trans.h"
19 #include "xfs_alloc.h"
20 #include "xfs_bmap.h"
21 #include "xfs_bmap_util.h"
22 #include "xfs_bmap_btree.h"
23 #include "xfs_rtalloc.h"
24 #include "xfs_error.h"
25 #include "xfs_quota.h"
26 #include "xfs_trans_space.h"
27 #include "xfs_trace.h"
28 #include "xfs_icache.h"
29 #include "xfs_iomap.h"
30 #include "xfs_reflink.h"
31 #include "xfs_rtbitmap.h"
32 
33 /* Kernel only BMAP related definitions and functions */
34 
35 /*
36  * Convert the given file system block to a disk block.  We have to treat it
37  * differently based on whether the file is a real time file or not, because the
38  * bmap code does.
39  */
40 xfs_daddr_t
41 xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb)
42 {
43 	if (XFS_IS_REALTIME_INODE(ip))
44 		return XFS_FSB_TO_BB(ip->i_mount, fsb);
45 	return XFS_FSB_TO_DADDR(ip->i_mount, fsb);
46 }
47 
48 /*
49  * Routine to zero an extent on disk allocated to the specific inode.
50  *
51  * The VFS functions take a linearised filesystem block offset, so we have to
52  * convert the sparse xfs fsb to the right format first.
53  * VFS types are real funky, too.
54  */
55 int
56 xfs_zero_extent(
57 	struct xfs_inode	*ip,
58 	xfs_fsblock_t		start_fsb,
59 	xfs_off_t		count_fsb)
60 {
61 	struct xfs_mount	*mp = ip->i_mount;
62 	struct xfs_buftarg	*target = xfs_inode_buftarg(ip);
63 	xfs_daddr_t		sector = xfs_fsb_to_db(ip, start_fsb);
64 	sector_t		block = XFS_BB_TO_FSBT(mp, sector);
65 
66 	return blkdev_issue_zeroout(target->bt_bdev,
67 		block << (mp->m_super->s_blocksize_bits - 9),
68 		count_fsb << (mp->m_super->s_blocksize_bits - 9),
69 		GFP_KERNEL, 0);
70 }
71 
72 /*
73  * Extent tree block counting routines.
74  */
75 
76 /*
77  * Count leaf blocks given a range of extent records.  Delayed allocation
78  * extents are not counted towards the totals.
79  */
80 xfs_extnum_t
81 xfs_bmap_count_leaves(
82 	struct xfs_ifork	*ifp,
83 	xfs_filblks_t		*count)
84 {
85 	struct xfs_iext_cursor	icur;
86 	struct xfs_bmbt_irec	got;
87 	xfs_extnum_t		numrecs = 0;
88 
89 	for_each_xfs_iext(ifp, &icur, &got) {
90 		if (!isnullstartblock(got.br_startblock)) {
91 			*count += got.br_blockcount;
92 			numrecs++;
93 		}
94 	}
95 
96 	return numrecs;
97 }
98 
99 /*
100  * Count fsblocks of the given fork.  Delayed allocation extents are
101  * not counted towards the totals.
102  */
103 int
104 xfs_bmap_count_blocks(
105 	struct xfs_trans	*tp,
106 	struct xfs_inode	*ip,
107 	int			whichfork,
108 	xfs_extnum_t		*nextents,
109 	xfs_filblks_t		*count)
110 {
111 	struct xfs_mount	*mp = ip->i_mount;
112 	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
113 	struct xfs_btree_cur	*cur;
114 	xfs_extlen_t		btblocks = 0;
115 	int			error;
116 
117 	*nextents = 0;
118 	*count = 0;
119 
120 	if (!ifp)
121 		return 0;
122 
123 	switch (ifp->if_format) {
124 	case XFS_DINODE_FMT_BTREE:
125 		error = xfs_iread_extents(tp, ip, whichfork);
126 		if (error)
127 			return error;
128 
129 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
130 		error = xfs_btree_count_blocks(cur, &btblocks);
131 		xfs_btree_del_cursor(cur, error);
132 		if (error)
133 			return error;
134 
135 		/*
136 		 * xfs_btree_count_blocks includes the root block contained in
137 		 * the inode fork in @btblocks, so subtract one because we're
138 		 * only interested in allocated disk blocks.
139 		 */
140 		*count += btblocks - 1;
141 
142 		fallthrough;
143 	case XFS_DINODE_FMT_EXTENTS:
144 		*nextents = xfs_bmap_count_leaves(ifp, count);
145 		break;
146 	}
147 
148 	return 0;
149 }
150 
151 static int
152 xfs_getbmap_report_one(
153 	struct xfs_inode	*ip,
154 	struct getbmapx		*bmv,
155 	struct kgetbmap		*out,
156 	int64_t			bmv_end,
157 	struct xfs_bmbt_irec	*got)
158 {
159 	struct kgetbmap		*p = out + bmv->bmv_entries;
160 	bool			shared = false;
161 	int			error;
162 
163 	error = xfs_reflink_trim_around_shared(ip, got, &shared);
164 	if (error)
165 		return error;
166 
167 	if (isnullstartblock(got->br_startblock) ||
168 	    got->br_startblock == DELAYSTARTBLOCK) {
169 		/*
170 		 * Take the flush completion as being a point-in-time snapshot
171 		 * where there are no delalloc extents, and if any new ones
172 		 * have been created racily, just skip them as being 'after'
173 		 * the flush and so don't get reported.
174 		 */
175 		if (!(bmv->bmv_iflags & BMV_IF_DELALLOC))
176 			return 0;
177 
178 		p->bmv_oflags |= BMV_OF_DELALLOC;
179 		p->bmv_block = -2;
180 	} else {
181 		p->bmv_block = xfs_fsb_to_db(ip, got->br_startblock);
182 	}
183 
184 	if (got->br_state == XFS_EXT_UNWRITTEN &&
185 	    (bmv->bmv_iflags & BMV_IF_PREALLOC))
186 		p->bmv_oflags |= BMV_OF_PREALLOC;
187 
188 	if (shared)
189 		p->bmv_oflags |= BMV_OF_SHARED;
190 
191 	p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, got->br_startoff);
192 	p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, got->br_blockcount);
193 
194 	bmv->bmv_offset = p->bmv_offset + p->bmv_length;
195 	bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset);
196 	bmv->bmv_entries++;
197 	return 0;
198 }
199 
200 static void
201 xfs_getbmap_report_hole(
202 	struct xfs_inode	*ip,
203 	struct getbmapx		*bmv,
204 	struct kgetbmap		*out,
205 	int64_t			bmv_end,
206 	xfs_fileoff_t		bno,
207 	xfs_fileoff_t		end)
208 {
209 	struct kgetbmap		*p = out + bmv->bmv_entries;
210 
211 	if (bmv->bmv_iflags & BMV_IF_NO_HOLES)
212 		return;
213 
214 	p->bmv_block = -1;
215 	p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, bno);
216 	p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, end - bno);
217 
218 	bmv->bmv_offset = p->bmv_offset + p->bmv_length;
219 	bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset);
220 	bmv->bmv_entries++;
221 }
222 
223 static inline bool
224 xfs_getbmap_full(
225 	struct getbmapx		*bmv)
226 {
227 	return bmv->bmv_length == 0 || bmv->bmv_entries >= bmv->bmv_count - 1;
228 }
229 
230 static bool
231 xfs_getbmap_next_rec(
232 	struct xfs_bmbt_irec	*rec,
233 	xfs_fileoff_t		total_end)
234 {
235 	xfs_fileoff_t		end = rec->br_startoff + rec->br_blockcount;
236 
237 	if (end == total_end)
238 		return false;
239 
240 	rec->br_startoff += rec->br_blockcount;
241 	if (!isnullstartblock(rec->br_startblock) &&
242 	    rec->br_startblock != DELAYSTARTBLOCK)
243 		rec->br_startblock += rec->br_blockcount;
244 	rec->br_blockcount = total_end - end;
245 	return true;
246 }
247 
248 /*
249  * Get inode's extents as described in bmv, and format for output.
250  * Calls formatter to fill the user's buffer until all extents
251  * are mapped, until the passed-in bmv->bmv_count slots have
252  * been filled, or until the formatter short-circuits the loop,
253  * if it is tracking filled-in extents on its own.
254  */
255 int						/* error code */
256 xfs_getbmap(
257 	struct xfs_inode	*ip,
258 	struct getbmapx		*bmv,		/* user bmap structure */
259 	struct kgetbmap		*out)
260 {
261 	struct xfs_mount	*mp = ip->i_mount;
262 	int			iflags = bmv->bmv_iflags;
263 	int			whichfork, lock, error = 0;
264 	int64_t			bmv_end, max_len;
265 	xfs_fileoff_t		bno, first_bno;
266 	struct xfs_ifork	*ifp;
267 	struct xfs_bmbt_irec	got, rec;
268 	xfs_filblks_t		len;
269 	struct xfs_iext_cursor	icur;
270 
271 	if (bmv->bmv_iflags & ~BMV_IF_VALID)
272 		return -EINVAL;
273 #ifndef DEBUG
274 	/* Only allow CoW fork queries if we're debugging. */
275 	if (iflags & BMV_IF_COWFORK)
276 		return -EINVAL;
277 #endif
278 	if ((iflags & BMV_IF_ATTRFORK) && (iflags & BMV_IF_COWFORK))
279 		return -EINVAL;
280 
281 	if (bmv->bmv_length < -1)
282 		return -EINVAL;
283 	bmv->bmv_entries = 0;
284 	if (bmv->bmv_length == 0)
285 		return 0;
286 
287 	if (iflags & BMV_IF_ATTRFORK)
288 		whichfork = XFS_ATTR_FORK;
289 	else if (iflags & BMV_IF_COWFORK)
290 		whichfork = XFS_COW_FORK;
291 	else
292 		whichfork = XFS_DATA_FORK;
293 
294 	xfs_ilock(ip, XFS_IOLOCK_SHARED);
295 	switch (whichfork) {
296 	case XFS_ATTR_FORK:
297 		lock = xfs_ilock_attr_map_shared(ip);
298 		if (!xfs_inode_has_attr_fork(ip))
299 			goto out_unlock_ilock;
300 
301 		max_len = 1LL << 32;
302 		break;
303 	case XFS_COW_FORK:
304 		lock = XFS_ILOCK_SHARED;
305 		xfs_ilock(ip, lock);
306 
307 		/* No CoW fork? Just return */
308 		if (!xfs_ifork_ptr(ip, whichfork))
309 			goto out_unlock_ilock;
310 
311 		if (xfs_get_cowextsz_hint(ip))
312 			max_len = mp->m_super->s_maxbytes;
313 		else
314 			max_len = XFS_ISIZE(ip);
315 		break;
316 	case XFS_DATA_FORK:
317 		if (!(iflags & BMV_IF_DELALLOC) &&
318 		    (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_disk_size)) {
319 			error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
320 			if (error)
321 				goto out_unlock_iolock;
322 
323 			/*
324 			 * Even after flushing the inode, there can still be
325 			 * delalloc blocks on the inode beyond EOF due to
326 			 * speculative preallocation.  These are not removed
327 			 * until the release function is called or the inode
328 			 * is inactivated.  Hence we cannot assert here that
329 			 * ip->i_delayed_blks == 0.
330 			 */
331 		}
332 
333 		if (xfs_get_extsz_hint(ip) ||
334 		    (ip->i_diflags &
335 		     (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))
336 			max_len = mp->m_super->s_maxbytes;
337 		else
338 			max_len = XFS_ISIZE(ip);
339 
340 		lock = xfs_ilock_data_map_shared(ip);
341 		break;
342 	}
343 
344 	ifp = xfs_ifork_ptr(ip, whichfork);
345 
346 	switch (ifp->if_format) {
347 	case XFS_DINODE_FMT_EXTENTS:
348 	case XFS_DINODE_FMT_BTREE:
349 		break;
350 	case XFS_DINODE_FMT_LOCAL:
351 		/* Local format inode forks report no extents. */
352 		goto out_unlock_ilock;
353 	default:
354 		error = -EINVAL;
355 		goto out_unlock_ilock;
356 	}
357 
358 	if (bmv->bmv_length == -1) {
359 		max_len = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, max_len));
360 		bmv->bmv_length = max(0LL, max_len - bmv->bmv_offset);
361 	}
362 
363 	bmv_end = bmv->bmv_offset + bmv->bmv_length;
364 
365 	first_bno = bno = XFS_BB_TO_FSBT(mp, bmv->bmv_offset);
366 	len = XFS_BB_TO_FSB(mp, bmv->bmv_length);
367 
368 	error = xfs_iread_extents(NULL, ip, whichfork);
369 	if (error)
370 		goto out_unlock_ilock;
371 
372 	if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
373 		/*
374 		 * Report a whole-file hole if the delalloc flag is set to
375 		 * stay compatible with the old implementation.
376 		 */
377 		if (iflags & BMV_IF_DELALLOC)
378 			xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
379 					XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
380 		goto out_unlock_ilock;
381 	}
382 
383 	while (!xfs_getbmap_full(bmv)) {
384 		xfs_trim_extent(&got, first_bno, len);
385 
386 		/*
387 		 * Report an entry for a hole if this extent doesn't directly
388 		 * follow the previous one.
389 		 */
390 		if (got.br_startoff > bno) {
391 			xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
392 					got.br_startoff);
393 			if (xfs_getbmap_full(bmv))
394 				break;
395 		}
396 
397 		/*
398 		 * In order to report shared extents accurately, we report each
399 		 * distinct shared / unshared part of a single bmbt record with
400 		 * an individual getbmapx record.
401 		 */
402 		bno = got.br_startoff + got.br_blockcount;
403 		rec = got;
404 		do {
405 			error = xfs_getbmap_report_one(ip, bmv, out, bmv_end,
406 					&rec);
407 			if (error || xfs_getbmap_full(bmv))
408 				goto out_unlock_ilock;
409 		} while (xfs_getbmap_next_rec(&rec, bno));
410 
411 		if (!xfs_iext_next_extent(ifp, &icur, &got)) {
412 			xfs_fileoff_t	end = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
413 
414 			if (bmv->bmv_entries > 0)
415 				out[bmv->bmv_entries - 1].bmv_oflags |=
416 								BMV_OF_LAST;
417 
418 			if (whichfork != XFS_ATTR_FORK && bno < end &&
419 			    !xfs_getbmap_full(bmv)) {
420 				xfs_getbmap_report_hole(ip, bmv, out, bmv_end,
421 						bno, end);
422 			}
423 			break;
424 		}
425 
426 		if (bno >= first_bno + len)
427 			break;
428 	}
429 
430 out_unlock_ilock:
431 	xfs_iunlock(ip, lock);
432 out_unlock_iolock:
433 	xfs_iunlock(ip, XFS_IOLOCK_SHARED);
434 	return error;
435 }
436 
437 /*
438  * Dead simple method of punching delalyed allocation blocks from a range in
439  * the inode.  This will always punch out both the start and end blocks, even
440  * if the ranges only partially overlap them, so it is up to the caller to
441  * ensure that partial blocks are not passed in.
442  */
443 void
444 xfs_bmap_punch_delalloc_range(
445 	struct xfs_inode	*ip,
446 	xfs_off_t		start_byte,
447 	xfs_off_t		end_byte)
448 {
449 	struct xfs_mount	*mp = ip->i_mount;
450 	struct xfs_ifork	*ifp = &ip->i_df;
451 	xfs_fileoff_t		start_fsb = XFS_B_TO_FSBT(mp, start_byte);
452 	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, end_byte);
453 	struct xfs_bmbt_irec	got, del;
454 	struct xfs_iext_cursor	icur;
455 
456 	ASSERT(!xfs_need_iread_extents(ifp));
457 
458 	xfs_ilock(ip, XFS_ILOCK_EXCL);
459 	if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got))
460 		goto out_unlock;
461 
462 	while (got.br_startoff + got.br_blockcount > start_fsb) {
463 		del = got;
464 		xfs_trim_extent(&del, start_fsb, end_fsb - start_fsb);
465 
466 		/*
467 		 * A delete can push the cursor forward. Step back to the
468 		 * previous extent on non-delalloc or extents outside the
469 		 * target range.
470 		 */
471 		if (!del.br_blockcount ||
472 		    !isnullstartblock(del.br_startblock)) {
473 			if (!xfs_iext_prev_extent(ifp, &icur, &got))
474 				break;
475 			continue;
476 		}
477 
478 		xfs_bmap_del_extent_delay(ip, XFS_DATA_FORK, &icur, &got, &del);
479 		if (!xfs_iext_get_extent(ifp, &icur, &got))
480 			break;
481 	}
482 
483 out_unlock:
484 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
485 }
486 
487 /*
488  * Test whether it is appropriate to check an inode for and free post EOF
489  * blocks.
490  */
491 bool
492 xfs_can_free_eofblocks(
493 	struct xfs_inode	*ip)
494 {
495 	struct xfs_bmbt_irec	imap;
496 	struct xfs_mount	*mp = ip->i_mount;
497 	xfs_fileoff_t		end_fsb;
498 	xfs_fileoff_t		last_fsb;
499 	int			nimaps = 1;
500 	int			error;
501 
502 	/*
503 	 * Caller must either hold the exclusive io lock; or be inactivating
504 	 * the inode, which guarantees there are no other users of the inode.
505 	 */
506 	if (!(VFS_I(ip)->i_state & I_FREEING))
507 		xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL);
508 
509 	/* prealloc/delalloc exists only on regular files */
510 	if (!S_ISREG(VFS_I(ip)->i_mode))
511 		return false;
512 
513 	/*
514 	 * Zero sized files with no cached pages and delalloc blocks will not
515 	 * have speculative prealloc/delalloc blocks to remove.
516 	 */
517 	if (VFS_I(ip)->i_size == 0 &&
518 	    VFS_I(ip)->i_mapping->nrpages == 0 &&
519 	    ip->i_delayed_blks == 0)
520 		return false;
521 
522 	/* If we haven't read in the extent list, then don't do it now. */
523 	if (xfs_need_iread_extents(&ip->i_df))
524 		return false;
525 
526 	/*
527 	 * Only free real extents for inodes with persistent preallocations or
528 	 * the append-only flag.
529 	 */
530 	if (ip->i_diflags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
531 		if (ip->i_delayed_blks == 0)
532 			return false;
533 
534 	/*
535 	 * Do not try to free post-EOF blocks if EOF is beyond the end of the
536 	 * range supported by the page cache, because the truncation will loop
537 	 * forever.
538 	 */
539 	end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
540 	if (xfs_inode_has_bigrtalloc(ip))
541 		end_fsb = xfs_rtb_roundup_rtx(mp, end_fsb);
542 	last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
543 	if (last_fsb <= end_fsb)
544 		return false;
545 
546 	/*
547 	 * Look up the mapping for the first block past EOF.  If we can't find
548 	 * it, there's nothing to free.
549 	 */
550 	xfs_ilock(ip, XFS_ILOCK_SHARED);
551 	error = xfs_bmapi_read(ip, end_fsb, last_fsb - end_fsb, &imap, &nimaps,
552 			0);
553 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
554 	if (error || nimaps == 0)
555 		return false;
556 
557 	/*
558 	 * If there's a real mapping there or there are delayed allocation
559 	 * reservations, then we have post-EOF blocks to try to free.
560 	 */
561 	return imap.br_startblock != HOLESTARTBLOCK || ip->i_delayed_blks;
562 }
563 
564 /*
565  * This is called to free any blocks beyond eof. The caller must hold
566  * IOLOCK_EXCL unless we are in the inode reclaim path and have the only
567  * reference to the inode.
568  */
569 int
570 xfs_free_eofblocks(
571 	struct xfs_inode	*ip)
572 {
573 	struct xfs_trans	*tp;
574 	struct xfs_mount	*mp = ip->i_mount;
575 	int			error;
576 
577 	/* Attach the dquots to the inode up front. */
578 	error = xfs_qm_dqattach(ip);
579 	if (error)
580 		return error;
581 
582 	/* Wait on dio to ensure i_size has settled. */
583 	inode_dio_wait(VFS_I(ip));
584 
585 	/*
586 	 * For preallocated files only free delayed allocations.
587 	 *
588 	 * Note that this means we also leave speculative preallocations in
589 	 * place for preallocated files.
590 	 */
591 	if (ip->i_diflags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) {
592 		if (ip->i_delayed_blks) {
593 			xfs_bmap_punch_delalloc_range(ip,
594 				round_up(XFS_ISIZE(ip), mp->m_sb.sb_blocksize),
595 				LLONG_MAX);
596 		}
597 		xfs_inode_clear_eofblocks_tag(ip);
598 		return 0;
599 	}
600 
601 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
602 	if (error) {
603 		ASSERT(xfs_is_shutdown(mp));
604 		return error;
605 	}
606 
607 	xfs_ilock(ip, XFS_ILOCK_EXCL);
608 	xfs_trans_ijoin(tp, ip, 0);
609 
610 	/*
611 	 * Do not update the on-disk file size.  If we update the on-disk file
612 	 * size and then the system crashes before the contents of the file are
613 	 * flushed to disk then the files may be full of holes (ie NULL files
614 	 * bug).
615 	 */
616 	error = xfs_itruncate_extents_flags(&tp, ip, XFS_DATA_FORK,
617 				XFS_ISIZE(ip), XFS_BMAPI_NODISCARD);
618 	if (error)
619 		goto err_cancel;
620 
621 	error = xfs_trans_commit(tp);
622 	if (error)
623 		goto out_unlock;
624 
625 	xfs_inode_clear_eofblocks_tag(ip);
626 	goto out_unlock;
627 
628 err_cancel:
629 	/*
630 	 * If we get an error at this point we simply don't
631 	 * bother truncating the file.
632 	 */
633 	xfs_trans_cancel(tp);
634 out_unlock:
635 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
636 	return error;
637 }
638 
639 int
640 xfs_alloc_file_space(
641 	struct xfs_inode	*ip,
642 	xfs_off_t		offset,
643 	xfs_off_t		len)
644 {
645 	xfs_mount_t		*mp = ip->i_mount;
646 	xfs_off_t		count;
647 	xfs_filblks_t		allocatesize_fsb;
648 	xfs_extlen_t		extsz, temp;
649 	xfs_fileoff_t		startoffset_fsb;
650 	xfs_fileoff_t		endoffset_fsb;
651 	int			rt;
652 	xfs_trans_t		*tp;
653 	xfs_bmbt_irec_t		imaps[1], *imapp;
654 	int			error;
655 
656 	if (xfs_is_always_cow_inode(ip))
657 		return 0;
658 
659 	trace_xfs_alloc_file_space(ip);
660 
661 	if (xfs_is_shutdown(mp))
662 		return -EIO;
663 
664 	error = xfs_qm_dqattach(ip);
665 	if (error)
666 		return error;
667 
668 	if (len <= 0)
669 		return -EINVAL;
670 
671 	rt = XFS_IS_REALTIME_INODE(ip);
672 	extsz = xfs_get_extsz_hint(ip);
673 
674 	count = len;
675 	imapp = &imaps[0];
676 	startoffset_fsb	= XFS_B_TO_FSBT(mp, offset);
677 	endoffset_fsb = XFS_B_TO_FSB(mp, offset + count);
678 	allocatesize_fsb = endoffset_fsb - startoffset_fsb;
679 
680 	/*
681 	 * Allocate file space until done or until there is an error
682 	 */
683 	while (allocatesize_fsb && !error) {
684 		xfs_fileoff_t	s, e;
685 		unsigned int	dblocks, rblocks, resblks;
686 		int		nimaps = 1;
687 
688 		/*
689 		 * Determine space reservations for data/realtime.
690 		 */
691 		if (unlikely(extsz)) {
692 			s = startoffset_fsb;
693 			do_div(s, extsz);
694 			s *= extsz;
695 			e = startoffset_fsb + allocatesize_fsb;
696 			div_u64_rem(startoffset_fsb, extsz, &temp);
697 			if (temp)
698 				e += temp;
699 			div_u64_rem(e, extsz, &temp);
700 			if (temp)
701 				e += extsz - temp;
702 		} else {
703 			s = 0;
704 			e = allocatesize_fsb;
705 		}
706 
707 		/*
708 		 * The transaction reservation is limited to a 32-bit block
709 		 * count, hence we need to limit the number of blocks we are
710 		 * trying to reserve to avoid an overflow. We can't allocate
711 		 * more than @nimaps extents, and an extent is limited on disk
712 		 * to XFS_BMBT_MAX_EXTLEN (21 bits), so use that to enforce the
713 		 * limit.
714 		 */
715 		resblks = min_t(xfs_fileoff_t, (e - s),
716 				(XFS_MAX_BMBT_EXTLEN * nimaps));
717 		if (unlikely(rt)) {
718 			dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
719 			rblocks = resblks;
720 		} else {
721 			dblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
722 			rblocks = 0;
723 		}
724 
725 		error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write,
726 				dblocks, rblocks, false, &tp);
727 		if (error)
728 			break;
729 
730 		error = xfs_iext_count_extend(tp, ip, XFS_DATA_FORK,
731 				XFS_IEXT_ADD_NOSPLIT_CNT);
732 		if (error)
733 			goto error;
734 
735 		/*
736 		 * If the allocator cannot find a single free extent large
737 		 * enough to cover the start block of the requested range,
738 		 * xfs_bmapi_write will return -ENOSR.
739 		 *
740 		 * In that case we simply need to keep looping with the same
741 		 * startoffset_fsb so that one of the following allocations
742 		 * will eventually reach the requested range.
743 		 */
744 		error = xfs_bmapi_write(tp, ip, startoffset_fsb,
745 				allocatesize_fsb, XFS_BMAPI_PREALLOC, 0, imapp,
746 				&nimaps);
747 		if (error) {
748 			if (error != -ENOSR)
749 				goto error;
750 			error = 0;
751 		} else {
752 			startoffset_fsb += imapp->br_blockcount;
753 			allocatesize_fsb -= imapp->br_blockcount;
754 		}
755 
756 		ip->i_diflags |= XFS_DIFLAG_PREALLOC;
757 		xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
758 
759 		error = xfs_trans_commit(tp);
760 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
761 	}
762 
763 	return error;
764 
765 error:
766 	xfs_trans_cancel(tp);
767 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
768 	return error;
769 }
770 
771 static int
772 xfs_unmap_extent(
773 	struct xfs_inode	*ip,
774 	xfs_fileoff_t		startoffset_fsb,
775 	xfs_filblks_t		len_fsb,
776 	int			*done)
777 {
778 	struct xfs_mount	*mp = ip->i_mount;
779 	struct xfs_trans	*tp;
780 	uint			resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
781 	int			error;
782 
783 	error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, resblks, 0,
784 			false, &tp);
785 	if (error)
786 		return error;
787 
788 	error = xfs_iext_count_extend(tp, ip, XFS_DATA_FORK,
789 			XFS_IEXT_PUNCH_HOLE_CNT);
790 	if (error)
791 		goto out_trans_cancel;
792 
793 	error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, done);
794 	if (error)
795 		goto out_trans_cancel;
796 
797 	error = xfs_trans_commit(tp);
798 out_unlock:
799 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
800 	return error;
801 
802 out_trans_cancel:
803 	xfs_trans_cancel(tp);
804 	goto out_unlock;
805 }
806 
807 /* Caller must first wait for the completion of any pending DIOs if required. */
808 int
809 xfs_flush_unmap_range(
810 	struct xfs_inode	*ip,
811 	xfs_off_t		offset,
812 	xfs_off_t		len)
813 {
814 	struct inode		*inode = VFS_I(ip);
815 	xfs_off_t		rounding, start, end;
816 	int			error;
817 
818 	/*
819 	 * Make sure we extend the flush out to extent alignment
820 	 * boundaries so any extent range overlapping the start/end
821 	 * of the modification we are about to do is clean and idle.
822 	 */
823 	rounding = max_t(xfs_off_t, xfs_inode_alloc_unitsize(ip), PAGE_SIZE);
824 	start = rounddown_64(offset, rounding);
825 	end = roundup_64(offset + len, rounding) - 1;
826 
827 	error = filemap_write_and_wait_range(inode->i_mapping, start, end);
828 	if (error)
829 		return error;
830 	truncate_pagecache_range(inode, start, end);
831 	return 0;
832 }
833 
834 int
835 xfs_free_file_space(
836 	struct xfs_inode	*ip,
837 	xfs_off_t		offset,
838 	xfs_off_t		len)
839 {
840 	struct xfs_mount	*mp = ip->i_mount;
841 	xfs_fileoff_t		startoffset_fsb;
842 	xfs_fileoff_t		endoffset_fsb;
843 	int			done = 0, error;
844 
845 	trace_xfs_free_file_space(ip);
846 
847 	error = xfs_qm_dqattach(ip);
848 	if (error)
849 		return error;
850 
851 	if (len <= 0)	/* if nothing being freed */
852 		return 0;
853 
854 	/*
855 	 * Now AIO and DIO has drained we flush and (if necessary) invalidate
856 	 * the cached range over the first operation we are about to run.
857 	 */
858 	error = xfs_flush_unmap_range(ip, offset, len);
859 	if (error)
860 		return error;
861 
862 	startoffset_fsb = XFS_B_TO_FSB(mp, offset);
863 	endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
864 
865 	/* We can only free complete realtime extents. */
866 	if (xfs_inode_has_bigrtalloc(ip)) {
867 		startoffset_fsb = xfs_rtb_roundup_rtx(mp, startoffset_fsb);
868 		endoffset_fsb = xfs_rtb_rounddown_rtx(mp, endoffset_fsb);
869 	}
870 
871 	/*
872 	 * Need to zero the stuff we're not freeing, on disk.
873 	 */
874 	if (endoffset_fsb > startoffset_fsb) {
875 		while (!done) {
876 			error = xfs_unmap_extent(ip, startoffset_fsb,
877 					endoffset_fsb - startoffset_fsb, &done);
878 			if (error)
879 				return error;
880 		}
881 	}
882 
883 	/*
884 	 * Now that we've unmap all full blocks we'll have to zero out any
885 	 * partial block at the beginning and/or end.  xfs_zero_range is smart
886 	 * enough to skip any holes, including those we just created, but we
887 	 * must take care not to zero beyond EOF and enlarge i_size.
888 	 */
889 	if (offset >= XFS_ISIZE(ip))
890 		return 0;
891 	if (offset + len > XFS_ISIZE(ip))
892 		len = XFS_ISIZE(ip) - offset;
893 	error = xfs_zero_range(ip, offset, len, NULL);
894 	if (error)
895 		return error;
896 
897 	/*
898 	 * If we zeroed right up to EOF and EOF straddles a page boundary we
899 	 * must make sure that the post-EOF area is also zeroed because the
900 	 * page could be mmap'd and xfs_zero_range doesn't do that for us.
901 	 * Writeback of the eof page will do this, albeit clumsily.
902 	 */
903 	if (offset + len >= XFS_ISIZE(ip) && offset_in_page(offset + len) > 0) {
904 		error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
905 				round_down(offset + len, PAGE_SIZE), LLONG_MAX);
906 	}
907 
908 	return error;
909 }
910 
911 static int
912 xfs_prepare_shift(
913 	struct xfs_inode	*ip,
914 	loff_t			offset)
915 {
916 	unsigned int		rounding;
917 	int			error;
918 
919 	/*
920 	 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
921 	 * into the accessible region of the file.
922 	 */
923 	if (xfs_can_free_eofblocks(ip)) {
924 		error = xfs_free_eofblocks(ip);
925 		if (error)
926 			return error;
927 	}
928 
929 	/*
930 	 * Shift operations must stabilize the start block offset boundary along
931 	 * with the full range of the operation. If we don't, a COW writeback
932 	 * completion could race with an insert, front merge with the start
933 	 * extent (after split) during the shift and corrupt the file. Start
934 	 * with the allocation unit just prior to the start to stabilize the
935 	 * boundary.
936 	 */
937 	rounding = xfs_inode_alloc_unitsize(ip);
938 	offset = rounddown_64(offset, rounding);
939 	if (offset)
940 		offset -= rounding;
941 
942 	/*
943 	 * Writeback and invalidate cache for the remainder of the file as we're
944 	 * about to shift down every extent from offset to EOF.
945 	 */
946 	error = xfs_flush_unmap_range(ip, offset, XFS_ISIZE(ip));
947 	if (error)
948 		return error;
949 
950 	/*
951 	 * Clean out anything hanging around in the cow fork now that
952 	 * we've flushed all the dirty data out to disk to avoid having
953 	 * CoW extents at the wrong offsets.
954 	 */
955 	if (xfs_inode_has_cow_data(ip)) {
956 		error = xfs_reflink_cancel_cow_range(ip, offset, NULLFILEOFF,
957 				true);
958 		if (error)
959 			return error;
960 	}
961 
962 	return 0;
963 }
964 
965 /*
966  * xfs_collapse_file_space()
967  *	This routine frees disk space and shift extent for the given file.
968  *	The first thing we do is to free data blocks in the specified range
969  *	by calling xfs_free_file_space(). It would also sync dirty data
970  *	and invalidate page cache over the region on which collapse range
971  *	is working. And Shift extent records to the left to cover a hole.
972  * RETURNS:
973  *	0 on success
974  *	errno on error
975  *
976  */
977 int
978 xfs_collapse_file_space(
979 	struct xfs_inode	*ip,
980 	xfs_off_t		offset,
981 	xfs_off_t		len)
982 {
983 	struct xfs_mount	*mp = ip->i_mount;
984 	struct xfs_trans	*tp;
985 	int			error;
986 	xfs_fileoff_t		next_fsb = XFS_B_TO_FSB(mp, offset + len);
987 	xfs_fileoff_t		shift_fsb = XFS_B_TO_FSB(mp, len);
988 	bool			done = false;
989 
990 	xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL);
991 
992 	trace_xfs_collapse_file_space(ip);
993 
994 	error = xfs_free_file_space(ip, offset, len);
995 	if (error)
996 		return error;
997 
998 	error = xfs_prepare_shift(ip, offset);
999 	if (error)
1000 		return error;
1001 
1002 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
1003 	if (error)
1004 		return error;
1005 
1006 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1007 	xfs_trans_ijoin(tp, ip, 0);
1008 
1009 	while (!done) {
1010 		error = xfs_bmap_collapse_extents(tp, ip, &next_fsb, shift_fsb,
1011 				&done);
1012 		if (error)
1013 			goto out_trans_cancel;
1014 		if (done)
1015 			break;
1016 
1017 		/* finish any deferred frees and roll the transaction */
1018 		error = xfs_defer_finish(&tp);
1019 		if (error)
1020 			goto out_trans_cancel;
1021 	}
1022 
1023 	error = xfs_trans_commit(tp);
1024 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1025 	return error;
1026 
1027 out_trans_cancel:
1028 	xfs_trans_cancel(tp);
1029 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1030 	return error;
1031 }
1032 
1033 /*
1034  * xfs_insert_file_space()
1035  *	This routine create hole space by shifting extents for the given file.
1036  *	The first thing we do is to sync dirty data and invalidate page cache
1037  *	over the region on which insert range is working. And split an extent
1038  *	to two extents at given offset by calling xfs_bmap_split_extent.
1039  *	And shift all extent records which are laying between [offset,
1040  *	last allocated extent] to the right to reserve hole range.
1041  * RETURNS:
1042  *	0 on success
1043  *	errno on error
1044  */
1045 int
1046 xfs_insert_file_space(
1047 	struct xfs_inode	*ip,
1048 	loff_t			offset,
1049 	loff_t			len)
1050 {
1051 	struct xfs_mount	*mp = ip->i_mount;
1052 	struct xfs_trans	*tp;
1053 	int			error;
1054 	xfs_fileoff_t		stop_fsb = XFS_B_TO_FSB(mp, offset);
1055 	xfs_fileoff_t		next_fsb = NULLFSBLOCK;
1056 	xfs_fileoff_t		shift_fsb = XFS_B_TO_FSB(mp, len);
1057 	bool			done = false;
1058 
1059 	xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL);
1060 
1061 	trace_xfs_insert_file_space(ip);
1062 
1063 	error = xfs_bmap_can_insert_extents(ip, stop_fsb, shift_fsb);
1064 	if (error)
1065 		return error;
1066 
1067 	error = xfs_prepare_shift(ip, offset);
1068 	if (error)
1069 		return error;
1070 
1071 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
1072 			XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
1073 	if (error)
1074 		return error;
1075 
1076 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1077 	xfs_trans_ijoin(tp, ip, 0);
1078 
1079 	error = xfs_iext_count_extend(tp, ip, XFS_DATA_FORK,
1080 			XFS_IEXT_PUNCH_HOLE_CNT);
1081 	if (error)
1082 		goto out_trans_cancel;
1083 
1084 	/*
1085 	 * The extent shifting code works on extent granularity. So, if stop_fsb
1086 	 * is not the starting block of extent, we need to split the extent at
1087 	 * stop_fsb.
1088 	 */
1089 	error = xfs_bmap_split_extent(tp, ip, stop_fsb);
1090 	if (error)
1091 		goto out_trans_cancel;
1092 
1093 	do {
1094 		error = xfs_defer_finish(&tp);
1095 		if (error)
1096 			goto out_trans_cancel;
1097 
1098 		error = xfs_bmap_insert_extents(tp, ip, &next_fsb, shift_fsb,
1099 				&done, stop_fsb);
1100 		if (error)
1101 			goto out_trans_cancel;
1102 	} while (!done);
1103 
1104 	error = xfs_trans_commit(tp);
1105 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1106 	return error;
1107 
1108 out_trans_cancel:
1109 	xfs_trans_cancel(tp);
1110 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1111 	return error;
1112 }
1113 
1114 /*
1115  * We need to check that the format of the data fork in the temporary inode is
1116  * valid for the target inode before doing the swap. This is not a problem with
1117  * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1118  * data fork depending on the space the attribute fork is taking so we can get
1119  * invalid formats on the target inode.
1120  *
1121  * E.g. target has space for 7 extents in extent format, temp inode only has
1122  * space for 6.  If we defragment down to 7 extents, then the tmp format is a
1123  * btree, but when swapped it needs to be in extent format. Hence we can't just
1124  * blindly swap data forks on attr2 filesystems.
1125  *
1126  * Note that we check the swap in both directions so that we don't end up with
1127  * a corrupt temporary inode, either.
1128  *
1129  * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1130  * inode will prevent this situation from occurring, so all we do here is
1131  * reject and log the attempt. basically we are putting the responsibility on
1132  * userspace to get this right.
1133  */
1134 static int
1135 xfs_swap_extents_check_format(
1136 	struct xfs_inode	*ip,	/* target inode */
1137 	struct xfs_inode	*tip)	/* tmp inode */
1138 {
1139 	struct xfs_ifork	*ifp = &ip->i_df;
1140 	struct xfs_ifork	*tifp = &tip->i_df;
1141 
1142 	/* User/group/project quota ids must match if quotas are enforced. */
1143 	if (XFS_IS_QUOTA_ON(ip->i_mount) &&
1144 	    (!uid_eq(VFS_I(ip)->i_uid, VFS_I(tip)->i_uid) ||
1145 	     !gid_eq(VFS_I(ip)->i_gid, VFS_I(tip)->i_gid) ||
1146 	     ip->i_projid != tip->i_projid))
1147 		return -EINVAL;
1148 
1149 	/* Should never get a local format */
1150 	if (ifp->if_format == XFS_DINODE_FMT_LOCAL ||
1151 	    tifp->if_format == XFS_DINODE_FMT_LOCAL)
1152 		return -EINVAL;
1153 
1154 	/*
1155 	 * if the target inode has less extents that then temporary inode then
1156 	 * why did userspace call us?
1157 	 */
1158 	if (ifp->if_nextents < tifp->if_nextents)
1159 		return -EINVAL;
1160 
1161 	/*
1162 	 * If we have to use the (expensive) rmap swap method, we can
1163 	 * handle any number of extents and any format.
1164 	 */
1165 	if (xfs_has_rmapbt(ip->i_mount))
1166 		return 0;
1167 
1168 	/*
1169 	 * if the target inode is in extent form and the temp inode is in btree
1170 	 * form then we will end up with the target inode in the wrong format
1171 	 * as we already know there are less extents in the temp inode.
1172 	 */
1173 	if (ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1174 	    tifp->if_format == XFS_DINODE_FMT_BTREE)
1175 		return -EINVAL;
1176 
1177 	/* Check temp in extent form to max in target */
1178 	if (tifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1179 	    tifp->if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1180 		return -EINVAL;
1181 
1182 	/* Check target in extent form to max in temp */
1183 	if (ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1184 	    ifp->if_nextents > XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1185 		return -EINVAL;
1186 
1187 	/*
1188 	 * If we are in a btree format, check that the temp root block will fit
1189 	 * in the target and that it has enough extents to be in btree format
1190 	 * in the target.
1191 	 *
1192 	 * Note that we have to be careful to allow btree->extent conversions
1193 	 * (a common defrag case) which will occur when the temp inode is in
1194 	 * extent format...
1195 	 */
1196 	if (tifp->if_format == XFS_DINODE_FMT_BTREE) {
1197 		if (xfs_inode_has_attr_fork(ip) &&
1198 		    XFS_BMAP_BMDR_SPACE(tifp->if_broot) > xfs_inode_fork_boff(ip))
1199 			return -EINVAL;
1200 		if (tifp->if_nextents <= XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1201 			return -EINVAL;
1202 	}
1203 
1204 	/* Reciprocal target->temp btree format checks */
1205 	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
1206 		if (xfs_inode_has_attr_fork(tip) &&
1207 		    XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > xfs_inode_fork_boff(tip))
1208 			return -EINVAL;
1209 		if (ifp->if_nextents <= XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1210 			return -EINVAL;
1211 	}
1212 
1213 	return 0;
1214 }
1215 
1216 static int
1217 xfs_swap_extent_flush(
1218 	struct xfs_inode	*ip)
1219 {
1220 	int	error;
1221 
1222 	error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1223 	if (error)
1224 		return error;
1225 	truncate_pagecache_range(VFS_I(ip), 0, -1);
1226 
1227 	/* Verify O_DIRECT for ftmp */
1228 	if (VFS_I(ip)->i_mapping->nrpages)
1229 		return -EINVAL;
1230 	return 0;
1231 }
1232 
1233 /*
1234  * Move extents from one file to another, when rmap is enabled.
1235  */
1236 STATIC int
1237 xfs_swap_extent_rmap(
1238 	struct xfs_trans		**tpp,
1239 	struct xfs_inode		*ip,
1240 	struct xfs_inode		*tip)
1241 {
1242 	struct xfs_trans		*tp = *tpp;
1243 	struct xfs_bmbt_irec		irec;
1244 	struct xfs_bmbt_irec		uirec;
1245 	struct xfs_bmbt_irec		tirec;
1246 	xfs_fileoff_t			offset_fsb;
1247 	xfs_fileoff_t			end_fsb;
1248 	xfs_filblks_t			count_fsb;
1249 	int				error;
1250 	xfs_filblks_t			ilen;
1251 	xfs_filblks_t			rlen;
1252 	int				nimaps;
1253 	uint64_t			tip_flags2;
1254 
1255 	/*
1256 	 * If the source file has shared blocks, we must flag the donor
1257 	 * file as having shared blocks so that we get the shared-block
1258 	 * rmap functions when we go to fix up the rmaps.  The flags
1259 	 * will be switch for reals later.
1260 	 */
1261 	tip_flags2 = tip->i_diflags2;
1262 	if (ip->i_diflags2 & XFS_DIFLAG2_REFLINK)
1263 		tip->i_diflags2 |= XFS_DIFLAG2_REFLINK;
1264 
1265 	offset_fsb = 0;
1266 	end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip)));
1267 	count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
1268 
1269 	while (count_fsb) {
1270 		/* Read extent from the donor file */
1271 		nimaps = 1;
1272 		error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec,
1273 				&nimaps, 0);
1274 		if (error)
1275 			goto out;
1276 		ASSERT(nimaps == 1);
1277 		ASSERT(tirec.br_startblock != DELAYSTARTBLOCK);
1278 
1279 		trace_xfs_swap_extent_rmap_remap(tip, &tirec);
1280 		ilen = tirec.br_blockcount;
1281 
1282 		/* Unmap the old blocks in the source file. */
1283 		while (tirec.br_blockcount) {
1284 			ASSERT(tp->t_highest_agno == NULLAGNUMBER);
1285 			trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec);
1286 
1287 			/* Read extent from the source file */
1288 			nimaps = 1;
1289 			error = xfs_bmapi_read(ip, tirec.br_startoff,
1290 					tirec.br_blockcount, &irec,
1291 					&nimaps, 0);
1292 			if (error)
1293 				goto out;
1294 			ASSERT(nimaps == 1);
1295 			ASSERT(tirec.br_startoff == irec.br_startoff);
1296 			trace_xfs_swap_extent_rmap_remap_piece(ip, &irec);
1297 
1298 			/* Trim the extent. */
1299 			uirec = tirec;
1300 			uirec.br_blockcount = rlen = min_t(xfs_filblks_t,
1301 					tirec.br_blockcount,
1302 					irec.br_blockcount);
1303 			trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec);
1304 
1305 			if (xfs_bmap_is_real_extent(&uirec)) {
1306 				error = xfs_iext_count_extend(tp, ip,
1307 						XFS_DATA_FORK,
1308 						XFS_IEXT_SWAP_RMAP_CNT);
1309 				if (error)
1310 					goto out;
1311 			}
1312 
1313 			if (xfs_bmap_is_real_extent(&irec)) {
1314 				error = xfs_iext_count_extend(tp, tip,
1315 						XFS_DATA_FORK,
1316 						XFS_IEXT_SWAP_RMAP_CNT);
1317 				if (error)
1318 					goto out;
1319 			}
1320 
1321 			/* Remove the mapping from the donor file. */
1322 			xfs_bmap_unmap_extent(tp, tip, XFS_DATA_FORK, &uirec);
1323 
1324 			/* Remove the mapping from the source file. */
1325 			xfs_bmap_unmap_extent(tp, ip, XFS_DATA_FORK, &irec);
1326 
1327 			/* Map the donor file's blocks into the source file. */
1328 			xfs_bmap_map_extent(tp, ip, XFS_DATA_FORK, &uirec);
1329 
1330 			/* Map the source file's blocks into the donor file. */
1331 			xfs_bmap_map_extent(tp, tip, XFS_DATA_FORK, &irec);
1332 
1333 			error = xfs_defer_finish(tpp);
1334 			tp = *tpp;
1335 			if (error)
1336 				goto out;
1337 
1338 			tirec.br_startoff += rlen;
1339 			if (tirec.br_startblock != HOLESTARTBLOCK &&
1340 			    tirec.br_startblock != DELAYSTARTBLOCK)
1341 				tirec.br_startblock += rlen;
1342 			tirec.br_blockcount -= rlen;
1343 		}
1344 
1345 		/* Roll on... */
1346 		count_fsb -= ilen;
1347 		offset_fsb += ilen;
1348 	}
1349 
1350 	tip->i_diflags2 = tip_flags2;
1351 	return 0;
1352 
1353 out:
1354 	trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_);
1355 	tip->i_diflags2 = tip_flags2;
1356 	return error;
1357 }
1358 
1359 /* Swap the extents of two files by swapping data forks. */
1360 STATIC int
1361 xfs_swap_extent_forks(
1362 	struct xfs_trans	*tp,
1363 	struct xfs_inode	*ip,
1364 	struct xfs_inode	*tip,
1365 	int			*src_log_flags,
1366 	int			*target_log_flags)
1367 {
1368 	xfs_filblks_t		aforkblks = 0;
1369 	xfs_filblks_t		taforkblks = 0;
1370 	xfs_extnum_t		junk;
1371 	uint64_t		tmp;
1372 	int			error;
1373 
1374 	/*
1375 	 * Count the number of extended attribute blocks
1376 	 */
1377 	if (xfs_inode_has_attr_fork(ip) && ip->i_af.if_nextents > 0 &&
1378 	    ip->i_af.if_format != XFS_DINODE_FMT_LOCAL) {
1379 		error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &junk,
1380 				&aforkblks);
1381 		if (error)
1382 			return error;
1383 	}
1384 	if (xfs_inode_has_attr_fork(tip) && tip->i_af.if_nextents > 0 &&
1385 	    tip->i_af.if_format != XFS_DINODE_FMT_LOCAL) {
1386 		error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK, &junk,
1387 				&taforkblks);
1388 		if (error)
1389 			return error;
1390 	}
1391 
1392 	/*
1393 	 * Btree format (v3) inodes have the inode number stamped in the bmbt
1394 	 * block headers. We can't start changing the bmbt blocks until the
1395 	 * inode owner change is logged so recovery does the right thing in the
1396 	 * event of a crash. Set the owner change log flags now and leave the
1397 	 * bmbt scan as the last step.
1398 	 */
1399 	if (xfs_has_v3inodes(ip->i_mount)) {
1400 		if (ip->i_df.if_format == XFS_DINODE_FMT_BTREE)
1401 			(*target_log_flags) |= XFS_ILOG_DOWNER;
1402 		if (tip->i_df.if_format == XFS_DINODE_FMT_BTREE)
1403 			(*src_log_flags) |= XFS_ILOG_DOWNER;
1404 	}
1405 
1406 	/*
1407 	 * Swap the data forks of the inodes
1408 	 */
1409 	swap(ip->i_df, tip->i_df);
1410 
1411 	/*
1412 	 * Fix the on-disk inode values
1413 	 */
1414 	tmp = (uint64_t)ip->i_nblocks;
1415 	ip->i_nblocks = tip->i_nblocks - taforkblks + aforkblks;
1416 	tip->i_nblocks = tmp + taforkblks - aforkblks;
1417 
1418 	/*
1419 	 * The extents in the source inode could still contain speculative
1420 	 * preallocation beyond EOF (e.g. the file is open but not modified
1421 	 * while defrag is in progress). In that case, we need to copy over the
1422 	 * number of delalloc blocks the data fork in the source inode is
1423 	 * tracking beyond EOF so that when the fork is truncated away when the
1424 	 * temporary inode is unlinked we don't underrun the i_delayed_blks
1425 	 * counter on that inode.
1426 	 */
1427 	ASSERT(tip->i_delayed_blks == 0);
1428 	tip->i_delayed_blks = ip->i_delayed_blks;
1429 	ip->i_delayed_blks = 0;
1430 
1431 	switch (ip->i_df.if_format) {
1432 	case XFS_DINODE_FMT_EXTENTS:
1433 		(*src_log_flags) |= XFS_ILOG_DEXT;
1434 		break;
1435 	case XFS_DINODE_FMT_BTREE:
1436 		ASSERT(!xfs_has_v3inodes(ip->i_mount) ||
1437 		       (*src_log_flags & XFS_ILOG_DOWNER));
1438 		(*src_log_flags) |= XFS_ILOG_DBROOT;
1439 		break;
1440 	}
1441 
1442 	switch (tip->i_df.if_format) {
1443 	case XFS_DINODE_FMT_EXTENTS:
1444 		(*target_log_flags) |= XFS_ILOG_DEXT;
1445 		break;
1446 	case XFS_DINODE_FMT_BTREE:
1447 		(*target_log_flags) |= XFS_ILOG_DBROOT;
1448 		ASSERT(!xfs_has_v3inodes(ip->i_mount) ||
1449 		       (*target_log_flags & XFS_ILOG_DOWNER));
1450 		break;
1451 	}
1452 
1453 	return 0;
1454 }
1455 
1456 /*
1457  * Fix up the owners of the bmbt blocks to refer to the current inode. The
1458  * change owner scan attempts to order all modified buffers in the current
1459  * transaction. In the event of ordered buffer failure, the offending buffer is
1460  * physically logged as a fallback and the scan returns -EAGAIN. We must roll
1461  * the transaction in this case to replenish the fallback log reservation and
1462  * restart the scan. This process repeats until the scan completes.
1463  */
1464 static int
1465 xfs_swap_change_owner(
1466 	struct xfs_trans	**tpp,
1467 	struct xfs_inode	*ip,
1468 	struct xfs_inode	*tmpip)
1469 {
1470 	int			error;
1471 	struct xfs_trans	*tp = *tpp;
1472 
1473 	do {
1474 		error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK, ip->i_ino,
1475 					      NULL);
1476 		/* success or fatal error */
1477 		if (error != -EAGAIN)
1478 			break;
1479 
1480 		error = xfs_trans_roll(tpp);
1481 		if (error)
1482 			break;
1483 		tp = *tpp;
1484 
1485 		/*
1486 		 * Redirty both inodes so they can relog and keep the log tail
1487 		 * moving forward.
1488 		 */
1489 		xfs_trans_ijoin(tp, ip, 0);
1490 		xfs_trans_ijoin(tp, tmpip, 0);
1491 		xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1492 		xfs_trans_log_inode(tp, tmpip, XFS_ILOG_CORE);
1493 	} while (true);
1494 
1495 	return error;
1496 }
1497 
1498 int
1499 xfs_swap_extents(
1500 	struct xfs_inode	*ip,	/* target inode */
1501 	struct xfs_inode	*tip,	/* tmp inode */
1502 	struct xfs_swapext	*sxp)
1503 {
1504 	struct xfs_mount	*mp = ip->i_mount;
1505 	struct xfs_trans	*tp;
1506 	struct xfs_bstat	*sbp = &sxp->sx_stat;
1507 	int			src_log_flags, target_log_flags;
1508 	int			error = 0;
1509 	uint64_t		f;
1510 	int			resblks = 0;
1511 	unsigned int		flags = 0;
1512 	struct timespec64	ctime, mtime;
1513 
1514 	/*
1515 	 * Lock the inodes against other IO, page faults and truncate to
1516 	 * begin with.  Then we can ensure the inodes are flushed and have no
1517 	 * page cache safely. Once we have done this we can take the ilocks and
1518 	 * do the rest of the checks.
1519 	 */
1520 	lock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1521 	filemap_invalidate_lock_two(VFS_I(ip)->i_mapping,
1522 				    VFS_I(tip)->i_mapping);
1523 
1524 	/* Verify that both files have the same format */
1525 	if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) {
1526 		error = -EINVAL;
1527 		goto out_unlock;
1528 	}
1529 
1530 	/* Verify both files are either real-time or non-realtime */
1531 	if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1532 		error = -EINVAL;
1533 		goto out_unlock;
1534 	}
1535 
1536 	error = xfs_qm_dqattach(ip);
1537 	if (error)
1538 		goto out_unlock;
1539 
1540 	error = xfs_qm_dqattach(tip);
1541 	if (error)
1542 		goto out_unlock;
1543 
1544 	error = xfs_swap_extent_flush(ip);
1545 	if (error)
1546 		goto out_unlock;
1547 	error = xfs_swap_extent_flush(tip);
1548 	if (error)
1549 		goto out_unlock;
1550 
1551 	if (xfs_inode_has_cow_data(tip)) {
1552 		error = xfs_reflink_cancel_cow_range(tip, 0, NULLFILEOFF, true);
1553 		if (error)
1554 			goto out_unlock;
1555 	}
1556 
1557 	/*
1558 	 * Extent "swapping" with rmap requires a permanent reservation and
1559 	 * a block reservation because it's really just a remap operation
1560 	 * performed with log redo items!
1561 	 */
1562 	if (xfs_has_rmapbt(mp)) {
1563 		int		w = XFS_DATA_FORK;
1564 		uint32_t	ipnext = ip->i_df.if_nextents;
1565 		uint32_t	tipnext	= tip->i_df.if_nextents;
1566 
1567 		/*
1568 		 * Conceptually this shouldn't affect the shape of either bmbt,
1569 		 * but since we atomically move extents one by one, we reserve
1570 		 * enough space to rebuild both trees.
1571 		 */
1572 		resblks = XFS_SWAP_RMAP_SPACE_RES(mp, ipnext, w);
1573 		resblks +=  XFS_SWAP_RMAP_SPACE_RES(mp, tipnext, w);
1574 
1575 		/*
1576 		 * If either inode straddles a bmapbt block allocation boundary,
1577 		 * the rmapbt algorithm triggers repeated allocs and frees as
1578 		 * extents are remapped. This can exhaust the block reservation
1579 		 * prematurely and cause shutdown. Return freed blocks to the
1580 		 * transaction reservation to counter this behavior.
1581 		 */
1582 		flags |= XFS_TRANS_RES_FDBLKS;
1583 	}
1584 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, flags,
1585 				&tp);
1586 	if (error)
1587 		goto out_unlock;
1588 
1589 	/*
1590 	 * Lock and join the inodes to the tansaction so that transaction commit
1591 	 * or cancel will unlock the inodes from this point onwards.
1592 	 */
1593 	xfs_lock_two_inodes(ip, XFS_ILOCK_EXCL, tip, XFS_ILOCK_EXCL);
1594 	xfs_trans_ijoin(tp, ip, 0);
1595 	xfs_trans_ijoin(tp, tip, 0);
1596 
1597 
1598 	/* Verify all data are being swapped */
1599 	if (sxp->sx_offset != 0 ||
1600 	    sxp->sx_length != ip->i_disk_size ||
1601 	    sxp->sx_length != tip->i_disk_size) {
1602 		error = -EFAULT;
1603 		goto out_trans_cancel;
1604 	}
1605 
1606 	trace_xfs_swap_extent_before(ip, 0);
1607 	trace_xfs_swap_extent_before(tip, 1);
1608 
1609 	/* check inode formats now that data is flushed */
1610 	error = xfs_swap_extents_check_format(ip, tip);
1611 	if (error) {
1612 		xfs_notice(mp,
1613 		    "%s: inode 0x%llx format is incompatible for exchanging.",
1614 				__func__, ip->i_ino);
1615 		goto out_trans_cancel;
1616 	}
1617 
1618 	/*
1619 	 * Compare the current change & modify times with that
1620 	 * passed in.  If they differ, we abort this swap.
1621 	 * This is the mechanism used to ensure the calling
1622 	 * process that the file was not changed out from
1623 	 * under it.
1624 	 */
1625 	ctime = inode_get_ctime(VFS_I(ip));
1626 	mtime = inode_get_mtime(VFS_I(ip));
1627 	if ((sbp->bs_ctime.tv_sec != ctime.tv_sec) ||
1628 	    (sbp->bs_ctime.tv_nsec != ctime.tv_nsec) ||
1629 	    (sbp->bs_mtime.tv_sec != mtime.tv_sec) ||
1630 	    (sbp->bs_mtime.tv_nsec != mtime.tv_nsec)) {
1631 		error = -EBUSY;
1632 		goto out_trans_cancel;
1633 	}
1634 
1635 	/*
1636 	 * Note the trickiness in setting the log flags - we set the owner log
1637 	 * flag on the opposite inode (i.e. the inode we are setting the new
1638 	 * owner to be) because once we swap the forks and log that, log
1639 	 * recovery is going to see the fork as owned by the swapped inode,
1640 	 * not the pre-swapped inodes.
1641 	 */
1642 	src_log_flags = XFS_ILOG_CORE;
1643 	target_log_flags = XFS_ILOG_CORE;
1644 
1645 	if (xfs_has_rmapbt(mp))
1646 		error = xfs_swap_extent_rmap(&tp, ip, tip);
1647 	else
1648 		error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags,
1649 				&target_log_flags);
1650 	if (error)
1651 		goto out_trans_cancel;
1652 
1653 	/* Do we have to swap reflink flags? */
1654 	if ((ip->i_diflags2 & XFS_DIFLAG2_REFLINK) ^
1655 	    (tip->i_diflags2 & XFS_DIFLAG2_REFLINK)) {
1656 		f = ip->i_diflags2 & XFS_DIFLAG2_REFLINK;
1657 		ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1658 		ip->i_diflags2 |= tip->i_diflags2 & XFS_DIFLAG2_REFLINK;
1659 		tip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1660 		tip->i_diflags2 |= f & XFS_DIFLAG2_REFLINK;
1661 	}
1662 
1663 	/* Swap the cow forks. */
1664 	if (xfs_has_reflink(mp)) {
1665 		ASSERT(!ip->i_cowfp ||
1666 		       ip->i_cowfp->if_format == XFS_DINODE_FMT_EXTENTS);
1667 		ASSERT(!tip->i_cowfp ||
1668 		       tip->i_cowfp->if_format == XFS_DINODE_FMT_EXTENTS);
1669 
1670 		swap(ip->i_cowfp, tip->i_cowfp);
1671 
1672 		if (ip->i_cowfp && ip->i_cowfp->if_bytes)
1673 			xfs_inode_set_cowblocks_tag(ip);
1674 		else
1675 			xfs_inode_clear_cowblocks_tag(ip);
1676 		if (tip->i_cowfp && tip->i_cowfp->if_bytes)
1677 			xfs_inode_set_cowblocks_tag(tip);
1678 		else
1679 			xfs_inode_clear_cowblocks_tag(tip);
1680 	}
1681 
1682 	xfs_trans_log_inode(tp, ip,  src_log_flags);
1683 	xfs_trans_log_inode(tp, tip, target_log_flags);
1684 
1685 	/*
1686 	 * The extent forks have been swapped, but crc=1,rmapbt=0 filesystems
1687 	 * have inode number owner values in the bmbt blocks that still refer to
1688 	 * the old inode. Scan each bmbt to fix up the owner values with the
1689 	 * inode number of the current inode.
1690 	 */
1691 	if (src_log_flags & XFS_ILOG_DOWNER) {
1692 		error = xfs_swap_change_owner(&tp, ip, tip);
1693 		if (error)
1694 			goto out_trans_cancel;
1695 	}
1696 	if (target_log_flags & XFS_ILOG_DOWNER) {
1697 		error = xfs_swap_change_owner(&tp, tip, ip);
1698 		if (error)
1699 			goto out_trans_cancel;
1700 	}
1701 
1702 	/*
1703 	 * If this is a synchronous mount, make sure that the
1704 	 * transaction goes to disk before returning to the user.
1705 	 */
1706 	if (xfs_has_wsync(mp))
1707 		xfs_trans_set_sync(tp);
1708 
1709 	error = xfs_trans_commit(tp);
1710 
1711 	trace_xfs_swap_extent_after(ip, 0);
1712 	trace_xfs_swap_extent_after(tip, 1);
1713 
1714 out_unlock_ilock:
1715 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1716 	xfs_iunlock(tip, XFS_ILOCK_EXCL);
1717 out_unlock:
1718 	filemap_invalidate_unlock_two(VFS_I(ip)->i_mapping,
1719 				      VFS_I(tip)->i_mapping);
1720 	unlock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1721 	return error;
1722 
1723 out_trans_cancel:
1724 	xfs_trans_cancel(tp);
1725 	goto out_unlock_ilock;
1726 }
1727