xref: /kernel/linux/linux-5.10/fs/xfs/xfs_bmap_util.c (revision 8c2ecf20)
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
32/* Kernel only BMAP related definitions and functions */
33
34/*
35 * Convert the given file system block to a disk block.  We have to treat it
36 * differently based on whether the file is a real time file or not, because the
37 * bmap code does.
38 */
39xfs_daddr_t
40xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb)
41{
42	if (XFS_IS_REALTIME_INODE(ip))
43		return XFS_FSB_TO_BB(ip->i_mount, fsb);
44	return XFS_FSB_TO_DADDR(ip->i_mount, fsb);
45}
46
47/*
48 * Routine to zero an extent on disk allocated to the specific inode.
49 *
50 * The VFS functions take a linearised filesystem block offset, so we have to
51 * convert the sparse xfs fsb to the right format first.
52 * VFS types are real funky, too.
53 */
54int
55xfs_zero_extent(
56	struct xfs_inode	*ip,
57	xfs_fsblock_t		start_fsb,
58	xfs_off_t		count_fsb)
59{
60	struct xfs_mount	*mp = ip->i_mount;
61	struct xfs_buftarg	*target = xfs_inode_buftarg(ip);
62	xfs_daddr_t		sector = xfs_fsb_to_db(ip, start_fsb);
63	sector_t		block = XFS_BB_TO_FSBT(mp, sector);
64
65	return blkdev_issue_zeroout(target->bt_bdev,
66		block << (mp->m_super->s_blocksize_bits - 9),
67		count_fsb << (mp->m_super->s_blocksize_bits - 9),
68		GFP_NOFS, 0);
69}
70
71#ifdef CONFIG_XFS_RT
72int
73xfs_bmap_rtalloc(
74	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
75{
76	int		error;		/* error return value */
77	xfs_mount_t	*mp;		/* mount point structure */
78	xfs_extlen_t	prod = 0;	/* product factor for allocators */
79	xfs_extlen_t	mod = 0;	/* product factor for allocators */
80	xfs_extlen_t	ralen = 0;	/* realtime allocation length */
81	xfs_extlen_t	align;		/* minimum allocation alignment */
82	xfs_rtblock_t	rtb;
83
84	mp = ap->ip->i_mount;
85	align = xfs_get_extsz_hint(ap->ip);
86	prod = align / mp->m_sb.sb_rextsize;
87	error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
88					align, 1, ap->eof, 0,
89					ap->conv, &ap->offset, &ap->length);
90	if (error)
91		return error;
92	ASSERT(ap->length);
93	ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
94
95	/*
96	 * If the offset & length are not perfectly aligned
97	 * then kill prod, it will just get us in trouble.
98	 */
99	div_u64_rem(ap->offset, align, &mod);
100	if (mod || ap->length % align)
101		prod = 1;
102	/*
103	 * Set ralen to be the actual requested length in rtextents.
104	 */
105	ralen = ap->length / mp->m_sb.sb_rextsize;
106	/*
107	 * If the old value was close enough to MAXEXTLEN that
108	 * we rounded up to it, cut it back so it's valid again.
109	 * Note that if it's a really large request (bigger than
110	 * MAXEXTLEN), we don't hear about that number, and can't
111	 * adjust the starting point to match it.
112	 */
113	if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
114		ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
115
116	/*
117	 * Lock out modifications to both the RT bitmap and summary inodes
118	 */
119	xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
120	xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
121	xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
122	xfs_trans_ijoin(ap->tp, mp->m_rsumip, XFS_ILOCK_EXCL);
123
124	/*
125	 * If it's an allocation to an empty file at offset 0,
126	 * pick an extent that will space things out in the rt area.
127	 */
128	if (ap->eof && ap->offset == 0) {
129		xfs_rtblock_t rtx; /* realtime extent no */
130
131		error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
132		if (error)
133			return error;
134		ap->blkno = rtx * mp->m_sb.sb_rextsize;
135	} else {
136		ap->blkno = 0;
137	}
138
139	xfs_bmap_adjacent(ap);
140
141	/*
142	 * Realtime allocation, done through xfs_rtallocate_extent.
143	 */
144	do_div(ap->blkno, mp->m_sb.sb_rextsize);
145	rtb = ap->blkno;
146	ap->length = ralen;
147	error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
148				&ralen, ap->wasdel, prod, &rtb);
149	if (error)
150		return error;
151
152	ap->blkno = rtb;
153	if (ap->blkno != NULLFSBLOCK) {
154		ap->blkno *= mp->m_sb.sb_rextsize;
155		ralen *= mp->m_sb.sb_rextsize;
156		ap->length = ralen;
157		ap->ip->i_d.di_nblocks += ralen;
158		xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
159		if (ap->wasdel)
160			ap->ip->i_delayed_blks -= ralen;
161		/*
162		 * Adjust the disk quota also. This was reserved
163		 * earlier.
164		 */
165		xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
166			ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
167					XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
168	} else {
169		ap->length = 0;
170	}
171	return 0;
172}
173#endif /* CONFIG_XFS_RT */
174
175/*
176 * Extent tree block counting routines.
177 */
178
179/*
180 * Count leaf blocks given a range of extent records.  Delayed allocation
181 * extents are not counted towards the totals.
182 */
183xfs_extnum_t
184xfs_bmap_count_leaves(
185	struct xfs_ifork	*ifp,
186	xfs_filblks_t		*count)
187{
188	struct xfs_iext_cursor	icur;
189	struct xfs_bmbt_irec	got;
190	xfs_extnum_t		numrecs = 0;
191
192	for_each_xfs_iext(ifp, &icur, &got) {
193		if (!isnullstartblock(got.br_startblock)) {
194			*count += got.br_blockcount;
195			numrecs++;
196		}
197	}
198
199	return numrecs;
200}
201
202/*
203 * Count fsblocks of the given fork.  Delayed allocation extents are
204 * not counted towards the totals.
205 */
206int
207xfs_bmap_count_blocks(
208	struct xfs_trans	*tp,
209	struct xfs_inode	*ip,
210	int			whichfork,
211	xfs_extnum_t		*nextents,
212	xfs_filblks_t		*count)
213{
214	struct xfs_mount	*mp = ip->i_mount;
215	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
216	struct xfs_btree_cur	*cur;
217	xfs_extlen_t		btblocks = 0;
218	int			error;
219
220	*nextents = 0;
221	*count = 0;
222
223	if (!ifp)
224		return 0;
225
226	switch (ifp->if_format) {
227	case XFS_DINODE_FMT_BTREE:
228		if (!(ifp->if_flags & XFS_IFEXTENTS)) {
229			error = xfs_iread_extents(tp, ip, whichfork);
230			if (error)
231				return error;
232		}
233
234		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
235		error = xfs_btree_count_blocks(cur, &btblocks);
236		xfs_btree_del_cursor(cur, error);
237		if (error)
238			return error;
239
240		/*
241		 * xfs_btree_count_blocks includes the root block contained in
242		 * the inode fork in @btblocks, so subtract one because we're
243		 * only interested in allocated disk blocks.
244		 */
245		*count += btblocks - 1;
246
247		/* fall through */
248	case XFS_DINODE_FMT_EXTENTS:
249		*nextents = xfs_bmap_count_leaves(ifp, count);
250		break;
251	}
252
253	return 0;
254}
255
256static int
257xfs_getbmap_report_one(
258	struct xfs_inode	*ip,
259	struct getbmapx		*bmv,
260	struct kgetbmap		*out,
261	int64_t			bmv_end,
262	struct xfs_bmbt_irec	*got)
263{
264	struct kgetbmap		*p = out + bmv->bmv_entries;
265	bool			shared = false;
266	int			error;
267
268	error = xfs_reflink_trim_around_shared(ip, got, &shared);
269	if (error)
270		return error;
271
272	if (isnullstartblock(got->br_startblock) ||
273	    got->br_startblock == DELAYSTARTBLOCK) {
274		/*
275		 * Delalloc extents that start beyond EOF can occur due to
276		 * speculative EOF allocation when the delalloc extent is larger
277		 * than the largest freespace extent at conversion time.  These
278		 * extents cannot be converted by data writeback, so can exist
279		 * here even if we are not supposed to be finding delalloc
280		 * extents.
281		 */
282		if (got->br_startoff < XFS_B_TO_FSB(ip->i_mount, XFS_ISIZE(ip)))
283			ASSERT((bmv->bmv_iflags & BMV_IF_DELALLOC) != 0);
284
285		p->bmv_oflags |= BMV_OF_DELALLOC;
286		p->bmv_block = -2;
287	} else {
288		p->bmv_block = xfs_fsb_to_db(ip, got->br_startblock);
289	}
290
291	if (got->br_state == XFS_EXT_UNWRITTEN &&
292	    (bmv->bmv_iflags & BMV_IF_PREALLOC))
293		p->bmv_oflags |= BMV_OF_PREALLOC;
294
295	if (shared)
296		p->bmv_oflags |= BMV_OF_SHARED;
297
298	p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, got->br_startoff);
299	p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, got->br_blockcount);
300
301	bmv->bmv_offset = p->bmv_offset + p->bmv_length;
302	bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset);
303	bmv->bmv_entries++;
304	return 0;
305}
306
307static void
308xfs_getbmap_report_hole(
309	struct xfs_inode	*ip,
310	struct getbmapx		*bmv,
311	struct kgetbmap		*out,
312	int64_t			bmv_end,
313	xfs_fileoff_t		bno,
314	xfs_fileoff_t		end)
315{
316	struct kgetbmap		*p = out + bmv->bmv_entries;
317
318	if (bmv->bmv_iflags & BMV_IF_NO_HOLES)
319		return;
320
321	p->bmv_block = -1;
322	p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, bno);
323	p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, end - bno);
324
325	bmv->bmv_offset = p->bmv_offset + p->bmv_length;
326	bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset);
327	bmv->bmv_entries++;
328}
329
330static inline bool
331xfs_getbmap_full(
332	struct getbmapx		*bmv)
333{
334	return bmv->bmv_length == 0 || bmv->bmv_entries >= bmv->bmv_count - 1;
335}
336
337static bool
338xfs_getbmap_next_rec(
339	struct xfs_bmbt_irec	*rec,
340	xfs_fileoff_t		total_end)
341{
342	xfs_fileoff_t		end = rec->br_startoff + rec->br_blockcount;
343
344	if (end == total_end)
345		return false;
346
347	rec->br_startoff += rec->br_blockcount;
348	if (!isnullstartblock(rec->br_startblock) &&
349	    rec->br_startblock != DELAYSTARTBLOCK)
350		rec->br_startblock += rec->br_blockcount;
351	rec->br_blockcount = total_end - end;
352	return true;
353}
354
355/*
356 * Get inode's extents as described in bmv, and format for output.
357 * Calls formatter to fill the user's buffer until all extents
358 * are mapped, until the passed-in bmv->bmv_count slots have
359 * been filled, or until the formatter short-circuits the loop,
360 * if it is tracking filled-in extents on its own.
361 */
362int						/* error code */
363xfs_getbmap(
364	struct xfs_inode	*ip,
365	struct getbmapx		*bmv,		/* user bmap structure */
366	struct kgetbmap		*out)
367{
368	struct xfs_mount	*mp = ip->i_mount;
369	int			iflags = bmv->bmv_iflags;
370	int			whichfork, lock, error = 0;
371	int64_t			bmv_end, max_len;
372	xfs_fileoff_t		bno, first_bno;
373	struct xfs_ifork	*ifp;
374	struct xfs_bmbt_irec	got, rec;
375	xfs_filblks_t		len;
376	struct xfs_iext_cursor	icur;
377
378	if (bmv->bmv_iflags & ~BMV_IF_VALID)
379		return -EINVAL;
380#ifndef DEBUG
381	/* Only allow CoW fork queries if we're debugging. */
382	if (iflags & BMV_IF_COWFORK)
383		return -EINVAL;
384#endif
385	if ((iflags & BMV_IF_ATTRFORK) && (iflags & BMV_IF_COWFORK))
386		return -EINVAL;
387
388	if (bmv->bmv_length < -1)
389		return -EINVAL;
390	bmv->bmv_entries = 0;
391	if (bmv->bmv_length == 0)
392		return 0;
393
394	if (iflags & BMV_IF_ATTRFORK)
395		whichfork = XFS_ATTR_FORK;
396	else if (iflags & BMV_IF_COWFORK)
397		whichfork = XFS_COW_FORK;
398	else
399		whichfork = XFS_DATA_FORK;
400	ifp = XFS_IFORK_PTR(ip, whichfork);
401
402	xfs_ilock(ip, XFS_IOLOCK_SHARED);
403	switch (whichfork) {
404	case XFS_ATTR_FORK:
405		if (!XFS_IFORK_Q(ip))
406			goto out_unlock_iolock;
407
408		max_len = 1LL << 32;
409		lock = xfs_ilock_attr_map_shared(ip);
410		break;
411	case XFS_COW_FORK:
412		/* No CoW fork? Just return */
413		if (!ifp)
414			goto out_unlock_iolock;
415
416		if (xfs_get_cowextsz_hint(ip))
417			max_len = mp->m_super->s_maxbytes;
418		else
419			max_len = XFS_ISIZE(ip);
420
421		lock = XFS_ILOCK_SHARED;
422		xfs_ilock(ip, lock);
423		break;
424	case XFS_DATA_FORK:
425		if (!(iflags & BMV_IF_DELALLOC) &&
426		    (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) {
427			error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
428			if (error)
429				goto out_unlock_iolock;
430
431			/*
432			 * Even after flushing the inode, there can still be
433			 * delalloc blocks on the inode beyond EOF due to
434			 * speculative preallocation.  These are not removed
435			 * until the release function is called or the inode
436			 * is inactivated.  Hence we cannot assert here that
437			 * ip->i_delayed_blks == 0.
438			 */
439		}
440
441		if (xfs_get_extsz_hint(ip) ||
442		    (ip->i_d.di_flags &
443		     (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))
444			max_len = mp->m_super->s_maxbytes;
445		else
446			max_len = XFS_ISIZE(ip);
447
448		lock = xfs_ilock_data_map_shared(ip);
449		break;
450	}
451
452	switch (ifp->if_format) {
453	case XFS_DINODE_FMT_EXTENTS:
454	case XFS_DINODE_FMT_BTREE:
455		break;
456	case XFS_DINODE_FMT_LOCAL:
457		/* Local format inode forks report no extents. */
458		goto out_unlock_ilock;
459	default:
460		error = -EINVAL;
461		goto out_unlock_ilock;
462	}
463
464	if (bmv->bmv_length == -1) {
465		max_len = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, max_len));
466		bmv->bmv_length = max(0LL, max_len - bmv->bmv_offset);
467	}
468
469	bmv_end = bmv->bmv_offset + bmv->bmv_length;
470
471	first_bno = bno = XFS_BB_TO_FSBT(mp, bmv->bmv_offset);
472	len = XFS_BB_TO_FSB(mp, bmv->bmv_length);
473
474	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
475		error = xfs_iread_extents(NULL, ip, whichfork);
476		if (error)
477			goto out_unlock_ilock;
478	}
479
480	if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
481		/*
482		 * Report a whole-file hole if the delalloc flag is set to
483		 * stay compatible with the old implementation.
484		 */
485		if (iflags & BMV_IF_DELALLOC)
486			xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
487					XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
488		goto out_unlock_ilock;
489	}
490
491	while (!xfs_getbmap_full(bmv)) {
492		xfs_trim_extent(&got, first_bno, len);
493
494		/*
495		 * Report an entry for a hole if this extent doesn't directly
496		 * follow the previous one.
497		 */
498		if (got.br_startoff > bno) {
499			xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
500					got.br_startoff);
501			if (xfs_getbmap_full(bmv))
502				break;
503		}
504
505		/*
506		 * In order to report shared extents accurately, we report each
507		 * distinct shared / unshared part of a single bmbt record with
508		 * an individual getbmapx record.
509		 */
510		bno = got.br_startoff + got.br_blockcount;
511		rec = got;
512		do {
513			error = xfs_getbmap_report_one(ip, bmv, out, bmv_end,
514					&rec);
515			if (error || xfs_getbmap_full(bmv))
516				goto out_unlock_ilock;
517		} while (xfs_getbmap_next_rec(&rec, bno));
518
519		if (!xfs_iext_next_extent(ifp, &icur, &got)) {
520			xfs_fileoff_t	end = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
521
522			out[bmv->bmv_entries - 1].bmv_oflags |= BMV_OF_LAST;
523
524			if (whichfork != XFS_ATTR_FORK && bno < end &&
525			    !xfs_getbmap_full(bmv)) {
526				xfs_getbmap_report_hole(ip, bmv, out, bmv_end,
527						bno, end);
528			}
529			break;
530		}
531
532		if (bno >= first_bno + len)
533			break;
534	}
535
536out_unlock_ilock:
537	xfs_iunlock(ip, lock);
538out_unlock_iolock:
539	xfs_iunlock(ip, XFS_IOLOCK_SHARED);
540	return error;
541}
542
543/*
544 * Dead simple method of punching delalyed allocation blocks from a range in
545 * the inode.  This will always punch out both the start and end blocks, even
546 * if the ranges only partially overlap them, so it is up to the caller to
547 * ensure that partial blocks are not passed in.
548 */
549int
550xfs_bmap_punch_delalloc_range(
551	struct xfs_inode	*ip,
552	xfs_fileoff_t		start_fsb,
553	xfs_fileoff_t		length)
554{
555	struct xfs_ifork	*ifp = &ip->i_df;
556	xfs_fileoff_t		end_fsb = start_fsb + length;
557	struct xfs_bmbt_irec	got, del;
558	struct xfs_iext_cursor	icur;
559	int			error = 0;
560
561	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
562
563	xfs_ilock(ip, XFS_ILOCK_EXCL);
564	if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got))
565		goto out_unlock;
566
567	while (got.br_startoff + got.br_blockcount > start_fsb) {
568		del = got;
569		xfs_trim_extent(&del, start_fsb, length);
570
571		/*
572		 * A delete can push the cursor forward. Step back to the
573		 * previous extent on non-delalloc or extents outside the
574		 * target range.
575		 */
576		if (!del.br_blockcount ||
577		    !isnullstartblock(del.br_startblock)) {
578			if (!xfs_iext_prev_extent(ifp, &icur, &got))
579				break;
580			continue;
581		}
582
583		error = xfs_bmap_del_extent_delay(ip, XFS_DATA_FORK, &icur,
584						  &got, &del);
585		if (error || !xfs_iext_get_extent(ifp, &icur, &got))
586			break;
587	}
588
589out_unlock:
590	xfs_iunlock(ip, XFS_ILOCK_EXCL);
591	return error;
592}
593
594/*
595 * Test whether it is appropriate to check an inode for and free post EOF
596 * blocks. The 'force' parameter determines whether we should also consider
597 * regular files that are marked preallocated or append-only.
598 */
599bool
600xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
601{
602	/* prealloc/delalloc exists only on regular files */
603	if (!S_ISREG(VFS_I(ip)->i_mode))
604		return false;
605
606	/*
607	 * Zero sized files with no cached pages and delalloc blocks will not
608	 * have speculative prealloc/delalloc blocks to remove.
609	 */
610	if (VFS_I(ip)->i_size == 0 &&
611	    VFS_I(ip)->i_mapping->nrpages == 0 &&
612	    ip->i_delayed_blks == 0)
613		return false;
614
615	/* If we haven't read in the extent list, then don't do it now. */
616	if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
617		return false;
618
619	/*
620	 * Do not free real preallocated or append-only files unless the file
621	 * has delalloc blocks and we are forced to remove them.
622	 */
623	if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
624		if (!force || ip->i_delayed_blks == 0)
625			return false;
626
627	return true;
628}
629
630/*
631 * This is called to free any blocks beyond eof. The caller must hold
632 * IOLOCK_EXCL unless we are in the inode reclaim path and have the only
633 * reference to the inode.
634 */
635int
636xfs_free_eofblocks(
637	struct xfs_inode	*ip)
638{
639	struct xfs_trans	*tp;
640	int			error;
641	xfs_fileoff_t		end_fsb;
642	xfs_fileoff_t		last_fsb;
643	xfs_filblks_t		map_len;
644	int			nimaps;
645	struct xfs_bmbt_irec	imap;
646	struct xfs_mount	*mp = ip->i_mount;
647
648	/*
649	 * Figure out if there are any blocks beyond the end
650	 * of the file.  If not, then there is nothing to do.
651	 */
652	end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
653	last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
654	if (last_fsb <= end_fsb)
655		return 0;
656	map_len = last_fsb - end_fsb;
657
658	nimaps = 1;
659	xfs_ilock(ip, XFS_ILOCK_SHARED);
660	error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
661	xfs_iunlock(ip, XFS_ILOCK_SHARED);
662
663	/*
664	 * If there are blocks after the end of file, truncate the file to its
665	 * current size to free them up.
666	 */
667	if (!error && (nimaps != 0) &&
668	    (imap.br_startblock != HOLESTARTBLOCK ||
669	     ip->i_delayed_blks)) {
670		/*
671		 * Attach the dquots to the inode up front.
672		 */
673		error = xfs_qm_dqattach(ip);
674		if (error)
675			return error;
676
677		/* wait on dio to ensure i_size has settled */
678		inode_dio_wait(VFS_I(ip));
679
680		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0,
681				&tp);
682		if (error) {
683			ASSERT(XFS_FORCED_SHUTDOWN(mp));
684			return error;
685		}
686
687		xfs_ilock(ip, XFS_ILOCK_EXCL);
688		xfs_trans_ijoin(tp, ip, 0);
689
690		/*
691		 * Do not update the on-disk file size.  If we update the
692		 * on-disk file size and then the system crashes before the
693		 * contents of the file are flushed to disk then the files
694		 * may be full of holes (ie NULL files bug).
695		 */
696		error = xfs_itruncate_extents_flags(&tp, ip, XFS_DATA_FORK,
697					XFS_ISIZE(ip), XFS_BMAPI_NODISCARD);
698		if (error) {
699			/*
700			 * If we get an error at this point we simply don't
701			 * bother truncating the file.
702			 */
703			xfs_trans_cancel(tp);
704		} else {
705			error = xfs_trans_commit(tp);
706			if (!error)
707				xfs_inode_clear_eofblocks_tag(ip);
708		}
709
710		xfs_iunlock(ip, XFS_ILOCK_EXCL);
711	}
712	return error;
713}
714
715int
716xfs_alloc_file_space(
717	struct xfs_inode	*ip,
718	xfs_off_t		offset,
719	xfs_off_t		len,
720	int			alloc_type)
721{
722	xfs_mount_t		*mp = ip->i_mount;
723	xfs_off_t		count;
724	xfs_filblks_t		allocated_fsb;
725	xfs_filblks_t		allocatesize_fsb;
726	xfs_extlen_t		extsz, temp;
727	xfs_fileoff_t		startoffset_fsb;
728	xfs_fileoff_t		endoffset_fsb;
729	int			nimaps;
730	int			quota_flag;
731	int			rt;
732	xfs_trans_t		*tp;
733	xfs_bmbt_irec_t		imaps[1], *imapp;
734	uint			qblocks, resblks, resrtextents;
735	int			error;
736
737	trace_xfs_alloc_file_space(ip);
738
739	if (XFS_FORCED_SHUTDOWN(mp))
740		return -EIO;
741
742	error = xfs_qm_dqattach(ip);
743	if (error)
744		return error;
745
746	if (len <= 0)
747		return -EINVAL;
748
749	rt = XFS_IS_REALTIME_INODE(ip);
750	extsz = xfs_get_extsz_hint(ip);
751
752	count = len;
753	imapp = &imaps[0];
754	nimaps = 1;
755	startoffset_fsb	= XFS_B_TO_FSBT(mp, offset);
756	endoffset_fsb = XFS_B_TO_FSB(mp, offset + count);
757	allocatesize_fsb = endoffset_fsb - startoffset_fsb;
758
759	/*
760	 * Allocate file space until done or until there is an error
761	 */
762	while (allocatesize_fsb && !error) {
763		xfs_fileoff_t	s, e;
764
765		/*
766		 * Determine space reservations for data/realtime.
767		 */
768		if (unlikely(extsz)) {
769			s = startoffset_fsb;
770			do_div(s, extsz);
771			s *= extsz;
772			e = startoffset_fsb + allocatesize_fsb;
773			div_u64_rem(startoffset_fsb, extsz, &temp);
774			if (temp)
775				e += temp;
776			div_u64_rem(e, extsz, &temp);
777			if (temp)
778				e += extsz - temp;
779		} else {
780			s = 0;
781			e = allocatesize_fsb;
782		}
783
784		/*
785		 * The transaction reservation is limited to a 32-bit block
786		 * count, hence we need to limit the number of blocks we are
787		 * trying to reserve to avoid an overflow. We can't allocate
788		 * more than @nimaps extents, and an extent is limited on disk
789		 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
790		 */
791		resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
792		if (unlikely(rt)) {
793			resrtextents = qblocks = resblks;
794			resrtextents /= mp->m_sb.sb_rextsize;
795			resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
796			quota_flag = XFS_QMOPT_RES_RTBLKS;
797		} else {
798			resrtextents = 0;
799			resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
800			quota_flag = XFS_QMOPT_RES_REGBLKS;
801		}
802
803		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks,
804				resrtextents, 0, &tp);
805
806		/*
807		 * Check for running out of space
808		 */
809		if (error) {
810			/*
811			 * Free the transaction structure.
812			 */
813			ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
814			break;
815		}
816		xfs_ilock(ip, XFS_ILOCK_EXCL);
817		error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
818						      0, quota_flag);
819		if (error)
820			goto error1;
821
822		xfs_trans_ijoin(tp, ip, 0);
823
824		error = xfs_bmapi_write(tp, ip, startoffset_fsb,
825					allocatesize_fsb, alloc_type, 0, imapp,
826					&nimaps);
827		if (error)
828			goto error0;
829
830		ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
831		xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
832
833		error = xfs_trans_commit(tp);
834		xfs_iunlock(ip, XFS_ILOCK_EXCL);
835		if (error)
836			break;
837
838		allocated_fsb = imapp->br_blockcount;
839
840		if (nimaps == 0) {
841			error = -ENOSPC;
842			break;
843		}
844
845		startoffset_fsb += allocated_fsb;
846		allocatesize_fsb -= allocated_fsb;
847	}
848
849	return error;
850
851error0:	/* unlock inode, unreserve quota blocks, cancel trans */
852	xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
853
854error1:	/* Just cancel transaction */
855	xfs_trans_cancel(tp);
856	xfs_iunlock(ip, XFS_ILOCK_EXCL);
857	return error;
858}
859
860static int
861xfs_unmap_extent(
862	struct xfs_inode	*ip,
863	xfs_fileoff_t		startoffset_fsb,
864	xfs_filblks_t		len_fsb,
865	int			*done)
866{
867	struct xfs_mount	*mp = ip->i_mount;
868	struct xfs_trans	*tp;
869	uint			resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
870	int			error;
871
872	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
873	if (error) {
874		ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
875		return error;
876	}
877
878	xfs_ilock(ip, XFS_ILOCK_EXCL);
879	error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot, ip->i_gdquot,
880			ip->i_pdquot, resblks, 0, XFS_QMOPT_RES_REGBLKS);
881	if (error)
882		goto out_trans_cancel;
883
884	xfs_trans_ijoin(tp, ip, 0);
885
886	error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, done);
887	if (error)
888		goto out_trans_cancel;
889
890	error = xfs_trans_commit(tp);
891out_unlock:
892	xfs_iunlock(ip, XFS_ILOCK_EXCL);
893	return error;
894
895out_trans_cancel:
896	xfs_trans_cancel(tp);
897	goto out_unlock;
898}
899
900/* Caller must first wait for the completion of any pending DIOs if required. */
901int
902xfs_flush_unmap_range(
903	struct xfs_inode	*ip,
904	xfs_off_t		offset,
905	xfs_off_t		len)
906{
907	struct xfs_mount	*mp = ip->i_mount;
908	struct inode		*inode = VFS_I(ip);
909	xfs_off_t		rounding, start, end;
910	int			error;
911
912	rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_SIZE);
913	start = round_down(offset, rounding);
914	end = round_up(offset + len, rounding) - 1;
915
916	error = filemap_write_and_wait_range(inode->i_mapping, start, end);
917	if (error)
918		return error;
919	truncate_pagecache_range(inode, start, end);
920	return 0;
921}
922
923int
924xfs_free_file_space(
925	struct xfs_inode	*ip,
926	xfs_off_t		offset,
927	xfs_off_t		len)
928{
929	struct xfs_mount	*mp = ip->i_mount;
930	xfs_fileoff_t		startoffset_fsb;
931	xfs_fileoff_t		endoffset_fsb;
932	int			done = 0, error;
933
934	trace_xfs_free_file_space(ip);
935
936	error = xfs_qm_dqattach(ip);
937	if (error)
938		return error;
939
940	if (len <= 0)	/* if nothing being freed */
941		return 0;
942
943	startoffset_fsb = XFS_B_TO_FSB(mp, offset);
944	endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
945
946	/* We can only free complete realtime extents. */
947	if (XFS_IS_REALTIME_INODE(ip) && mp->m_sb.sb_rextsize > 1) {
948		startoffset_fsb = roundup_64(startoffset_fsb,
949					     mp->m_sb.sb_rextsize);
950		endoffset_fsb = rounddown_64(endoffset_fsb,
951					     mp->m_sb.sb_rextsize);
952	}
953
954	/*
955	 * Need to zero the stuff we're not freeing, on disk.
956	 */
957	if (endoffset_fsb > startoffset_fsb) {
958		while (!done) {
959			error = xfs_unmap_extent(ip, startoffset_fsb,
960					endoffset_fsb - startoffset_fsb, &done);
961			if (error)
962				return error;
963		}
964	}
965
966	/*
967	 * Now that we've unmap all full blocks we'll have to zero out any
968	 * partial block at the beginning and/or end.  iomap_zero_range is smart
969	 * enough to skip any holes, including those we just created, but we
970	 * must take care not to zero beyond EOF and enlarge i_size.
971	 */
972	if (offset >= XFS_ISIZE(ip))
973		return 0;
974	if (offset + len > XFS_ISIZE(ip))
975		len = XFS_ISIZE(ip) - offset;
976	error = iomap_zero_range(VFS_I(ip), offset, len, NULL,
977			&xfs_buffered_write_iomap_ops);
978	if (error)
979		return error;
980
981	/*
982	 * If we zeroed right up to EOF and EOF straddles a page boundary we
983	 * must make sure that the post-EOF area is also zeroed because the
984	 * page could be mmap'd and iomap_zero_range doesn't do that for us.
985	 * Writeback of the eof page will do this, albeit clumsily.
986	 */
987	if (offset + len >= XFS_ISIZE(ip) && offset_in_page(offset + len) > 0) {
988		error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
989				round_down(offset + len, PAGE_SIZE), LLONG_MAX);
990	}
991
992	return error;
993}
994
995static int
996xfs_prepare_shift(
997	struct xfs_inode	*ip,
998	loff_t			offset)
999{
1000	struct xfs_mount	*mp = ip->i_mount;
1001	int			error;
1002
1003	/*
1004	 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
1005	 * into the accessible region of the file.
1006	 */
1007	if (xfs_can_free_eofblocks(ip, true)) {
1008		error = xfs_free_eofblocks(ip);
1009		if (error)
1010			return error;
1011	}
1012
1013	/*
1014	 * Shift operations must stabilize the start block offset boundary along
1015	 * with the full range of the operation. If we don't, a COW writeback
1016	 * completion could race with an insert, front merge with the start
1017	 * extent (after split) during the shift and corrupt the file. Start
1018	 * with the block just prior to the start to stabilize the boundary.
1019	 */
1020	offset = round_down(offset, 1 << mp->m_sb.sb_blocklog);
1021	if (offset)
1022		offset -= (1 << mp->m_sb.sb_blocklog);
1023
1024	/*
1025	 * Writeback and invalidate cache for the remainder of the file as we're
1026	 * about to shift down every extent from offset to EOF.
1027	 */
1028	error = xfs_flush_unmap_range(ip, offset, XFS_ISIZE(ip));
1029	if (error)
1030		return error;
1031
1032	/*
1033	 * Clean out anything hanging around in the cow fork now that
1034	 * we've flushed all the dirty data out to disk to avoid having
1035	 * CoW extents at the wrong offsets.
1036	 */
1037	if (xfs_inode_has_cow_data(ip)) {
1038		error = xfs_reflink_cancel_cow_range(ip, offset, NULLFILEOFF,
1039				true);
1040		if (error)
1041			return error;
1042	}
1043
1044	return 0;
1045}
1046
1047/*
1048 * xfs_collapse_file_space()
1049 *	This routine frees disk space and shift extent for the given file.
1050 *	The first thing we do is to free data blocks in the specified range
1051 *	by calling xfs_free_file_space(). It would also sync dirty data
1052 *	and invalidate page cache over the region on which collapse range
1053 *	is working. And Shift extent records to the left to cover a hole.
1054 * RETURNS:
1055 *	0 on success
1056 *	errno on error
1057 *
1058 */
1059int
1060xfs_collapse_file_space(
1061	struct xfs_inode	*ip,
1062	xfs_off_t		offset,
1063	xfs_off_t		len)
1064{
1065	struct xfs_mount	*mp = ip->i_mount;
1066	struct xfs_trans	*tp;
1067	int			error;
1068	xfs_fileoff_t		next_fsb = XFS_B_TO_FSB(mp, offset + len);
1069	xfs_fileoff_t		shift_fsb = XFS_B_TO_FSB(mp, len);
1070	bool			done = false;
1071
1072	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1073	ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
1074
1075	trace_xfs_collapse_file_space(ip);
1076
1077	error = xfs_free_file_space(ip, offset, len);
1078	if (error)
1079		return error;
1080
1081	error = xfs_prepare_shift(ip, offset);
1082	if (error)
1083		return error;
1084
1085	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
1086	if (error)
1087		return error;
1088
1089	xfs_ilock(ip, XFS_ILOCK_EXCL);
1090	xfs_trans_ijoin(tp, ip, 0);
1091
1092	while (!done) {
1093		error = xfs_bmap_collapse_extents(tp, ip, &next_fsb, shift_fsb,
1094				&done);
1095		if (error)
1096			goto out_trans_cancel;
1097		if (done)
1098			break;
1099
1100		/* finish any deferred frees and roll the transaction */
1101		error = xfs_defer_finish(&tp);
1102		if (error)
1103			goto out_trans_cancel;
1104	}
1105
1106	error = xfs_trans_commit(tp);
1107	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1108	return error;
1109
1110out_trans_cancel:
1111	xfs_trans_cancel(tp);
1112	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1113	return error;
1114}
1115
1116/*
1117 * xfs_insert_file_space()
1118 *	This routine create hole space by shifting extents for the given file.
1119 *	The first thing we do is to sync dirty data and invalidate page cache
1120 *	over the region on which insert range is working. And split an extent
1121 *	to two extents at given offset by calling xfs_bmap_split_extent.
1122 *	And shift all extent records which are laying between [offset,
1123 *	last allocated extent] to the right to reserve hole range.
1124 * RETURNS:
1125 *	0 on success
1126 *	errno on error
1127 */
1128int
1129xfs_insert_file_space(
1130	struct xfs_inode	*ip,
1131	loff_t			offset,
1132	loff_t			len)
1133{
1134	struct xfs_mount	*mp = ip->i_mount;
1135	struct xfs_trans	*tp;
1136	int			error;
1137	xfs_fileoff_t		stop_fsb = XFS_B_TO_FSB(mp, offset);
1138	xfs_fileoff_t		next_fsb = NULLFSBLOCK;
1139	xfs_fileoff_t		shift_fsb = XFS_B_TO_FSB(mp, len);
1140	bool			done = false;
1141
1142	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1143	ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
1144
1145	trace_xfs_insert_file_space(ip);
1146
1147	error = xfs_bmap_can_insert_extents(ip, stop_fsb, shift_fsb);
1148	if (error)
1149		return error;
1150
1151	error = xfs_prepare_shift(ip, offset);
1152	if (error)
1153		return error;
1154
1155	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
1156			XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
1157	if (error)
1158		return error;
1159
1160	xfs_ilock(ip, XFS_ILOCK_EXCL);
1161	xfs_trans_ijoin(tp, ip, 0);
1162
1163	/*
1164	 * The extent shifting code works on extent granularity. So, if stop_fsb
1165	 * is not the starting block of extent, we need to split the extent at
1166	 * stop_fsb.
1167	 */
1168	error = xfs_bmap_split_extent(tp, ip, stop_fsb);
1169	if (error)
1170		goto out_trans_cancel;
1171
1172	do {
1173		error = xfs_defer_finish(&tp);
1174		if (error)
1175			goto out_trans_cancel;
1176
1177		error = xfs_bmap_insert_extents(tp, ip, &next_fsb, shift_fsb,
1178				&done, stop_fsb);
1179		if (error)
1180			goto out_trans_cancel;
1181	} while (!done);
1182
1183	error = xfs_trans_commit(tp);
1184	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1185	return error;
1186
1187out_trans_cancel:
1188	xfs_trans_cancel(tp);
1189	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1190	return error;
1191}
1192
1193/*
1194 * We need to check that the format of the data fork in the temporary inode is
1195 * valid for the target inode before doing the swap. This is not a problem with
1196 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1197 * data fork depending on the space the attribute fork is taking so we can get
1198 * invalid formats on the target inode.
1199 *
1200 * E.g. target has space for 7 extents in extent format, temp inode only has
1201 * space for 6.  If we defragment down to 7 extents, then the tmp format is a
1202 * btree, but when swapped it needs to be in extent format. Hence we can't just
1203 * blindly swap data forks on attr2 filesystems.
1204 *
1205 * Note that we check the swap in both directions so that we don't end up with
1206 * a corrupt temporary inode, either.
1207 *
1208 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1209 * inode will prevent this situation from occurring, so all we do here is
1210 * reject and log the attempt. basically we are putting the responsibility on
1211 * userspace to get this right.
1212 */
1213static int
1214xfs_swap_extents_check_format(
1215	struct xfs_inode	*ip,	/* target inode */
1216	struct xfs_inode	*tip)	/* tmp inode */
1217{
1218	struct xfs_ifork	*ifp = &ip->i_df;
1219	struct xfs_ifork	*tifp = &tip->i_df;
1220
1221	/* User/group/project quota ids must match if quotas are enforced. */
1222	if (XFS_IS_QUOTA_ON(ip->i_mount) &&
1223	    (!uid_eq(VFS_I(ip)->i_uid, VFS_I(tip)->i_uid) ||
1224	     !gid_eq(VFS_I(ip)->i_gid, VFS_I(tip)->i_gid) ||
1225	     ip->i_d.di_projid != tip->i_d.di_projid))
1226		return -EINVAL;
1227
1228	/* Should never get a local format */
1229	if (ifp->if_format == XFS_DINODE_FMT_LOCAL ||
1230	    tifp->if_format == XFS_DINODE_FMT_LOCAL)
1231		return -EINVAL;
1232
1233	/*
1234	 * if the target inode has less extents that then temporary inode then
1235	 * why did userspace call us?
1236	 */
1237	if (ifp->if_nextents < tifp->if_nextents)
1238		return -EINVAL;
1239
1240	/*
1241	 * If we have to use the (expensive) rmap swap method, we can
1242	 * handle any number of extents and any format.
1243	 */
1244	if (xfs_sb_version_hasrmapbt(&ip->i_mount->m_sb))
1245		return 0;
1246
1247	/*
1248	 * if the target inode is in extent form and the temp inode is in btree
1249	 * form then we will end up with the target inode in the wrong format
1250	 * as we already know there are less extents in the temp inode.
1251	 */
1252	if (ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1253	    tifp->if_format == XFS_DINODE_FMT_BTREE)
1254		return -EINVAL;
1255
1256	/* Check temp in extent form to max in target */
1257	if (tifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1258	    tifp->if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1259		return -EINVAL;
1260
1261	/* Check target in extent form to max in temp */
1262	if (ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1263	    ifp->if_nextents > XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1264		return -EINVAL;
1265
1266	/*
1267	 * If we are in a btree format, check that the temp root block will fit
1268	 * in the target and that it has enough extents to be in btree format
1269	 * in the target.
1270	 *
1271	 * Note that we have to be careful to allow btree->extent conversions
1272	 * (a common defrag case) which will occur when the temp inode is in
1273	 * extent format...
1274	 */
1275	if (tifp->if_format == XFS_DINODE_FMT_BTREE) {
1276		if (XFS_IFORK_Q(ip) &&
1277		    XFS_BMAP_BMDR_SPACE(tifp->if_broot) > XFS_IFORK_BOFF(ip))
1278			return -EINVAL;
1279		if (tifp->if_nextents <= XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1280			return -EINVAL;
1281	}
1282
1283	/* Reciprocal target->temp btree format checks */
1284	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
1285		if (XFS_IFORK_Q(tip) &&
1286		    XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
1287			return -EINVAL;
1288		if (ifp->if_nextents <= XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1289			return -EINVAL;
1290	}
1291
1292	return 0;
1293}
1294
1295static int
1296xfs_swap_extent_flush(
1297	struct xfs_inode	*ip)
1298{
1299	int	error;
1300
1301	error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1302	if (error)
1303		return error;
1304	truncate_pagecache_range(VFS_I(ip), 0, -1);
1305
1306	/* Verify O_DIRECT for ftmp */
1307	if (VFS_I(ip)->i_mapping->nrpages)
1308		return -EINVAL;
1309	return 0;
1310}
1311
1312/*
1313 * Move extents from one file to another, when rmap is enabled.
1314 */
1315STATIC int
1316xfs_swap_extent_rmap(
1317	struct xfs_trans		**tpp,
1318	struct xfs_inode		*ip,
1319	struct xfs_inode		*tip)
1320{
1321	struct xfs_trans		*tp = *tpp;
1322	struct xfs_bmbt_irec		irec;
1323	struct xfs_bmbt_irec		uirec;
1324	struct xfs_bmbt_irec		tirec;
1325	xfs_fileoff_t			offset_fsb;
1326	xfs_fileoff_t			end_fsb;
1327	xfs_filblks_t			count_fsb;
1328	int				error;
1329	xfs_filblks_t			ilen;
1330	xfs_filblks_t			rlen;
1331	int				nimaps;
1332	uint64_t			tip_flags2;
1333
1334	/*
1335	 * If the source file has shared blocks, we must flag the donor
1336	 * file as having shared blocks so that we get the shared-block
1337	 * rmap functions when we go to fix up the rmaps.  The flags
1338	 * will be switch for reals later.
1339	 */
1340	tip_flags2 = tip->i_d.di_flags2;
1341	if (ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)
1342		tip->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK;
1343
1344	offset_fsb = 0;
1345	end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip)));
1346	count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
1347
1348	while (count_fsb) {
1349		/* Read extent from the donor file */
1350		nimaps = 1;
1351		error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec,
1352				&nimaps, 0);
1353		if (error)
1354			goto out;
1355		ASSERT(nimaps == 1);
1356		ASSERT(tirec.br_startblock != DELAYSTARTBLOCK);
1357
1358		trace_xfs_swap_extent_rmap_remap(tip, &tirec);
1359		ilen = tirec.br_blockcount;
1360
1361		/* Unmap the old blocks in the source file. */
1362		while (tirec.br_blockcount) {
1363			ASSERT(tp->t_firstblock == NULLFSBLOCK);
1364			trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec);
1365
1366			/* Read extent from the source file */
1367			nimaps = 1;
1368			error = xfs_bmapi_read(ip, tirec.br_startoff,
1369					tirec.br_blockcount, &irec,
1370					&nimaps, 0);
1371			if (error)
1372				goto out;
1373			ASSERT(nimaps == 1);
1374			ASSERT(tirec.br_startoff == irec.br_startoff);
1375			trace_xfs_swap_extent_rmap_remap_piece(ip, &irec);
1376
1377			/* Trim the extent. */
1378			uirec = tirec;
1379			uirec.br_blockcount = rlen = min_t(xfs_filblks_t,
1380					tirec.br_blockcount,
1381					irec.br_blockcount);
1382			trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec);
1383
1384			/* Remove the mapping from the donor file. */
1385			xfs_bmap_unmap_extent(tp, tip, &uirec);
1386
1387			/* Remove the mapping from the source file. */
1388			xfs_bmap_unmap_extent(tp, ip, &irec);
1389
1390			/* Map the donor file's blocks into the source file. */
1391			xfs_bmap_map_extent(tp, ip, &uirec);
1392
1393			/* Map the source file's blocks into the donor file. */
1394			xfs_bmap_map_extent(tp, tip, &irec);
1395
1396			error = xfs_defer_finish(tpp);
1397			tp = *tpp;
1398			if (error)
1399				goto out;
1400
1401			tirec.br_startoff += rlen;
1402			if (tirec.br_startblock != HOLESTARTBLOCK &&
1403			    tirec.br_startblock != DELAYSTARTBLOCK)
1404				tirec.br_startblock += rlen;
1405			tirec.br_blockcount -= rlen;
1406		}
1407
1408		/* Roll on... */
1409		count_fsb -= ilen;
1410		offset_fsb += ilen;
1411	}
1412
1413	tip->i_d.di_flags2 = tip_flags2;
1414	return 0;
1415
1416out:
1417	trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_);
1418	tip->i_d.di_flags2 = tip_flags2;
1419	return error;
1420}
1421
1422/* Swap the extents of two files by swapping data forks. */
1423STATIC int
1424xfs_swap_extent_forks(
1425	struct xfs_trans	*tp,
1426	struct xfs_inode	*ip,
1427	struct xfs_inode	*tip,
1428	int			*src_log_flags,
1429	int			*target_log_flags)
1430{
1431	xfs_filblks_t		aforkblks = 0;
1432	xfs_filblks_t		taforkblks = 0;
1433	xfs_extnum_t		junk;
1434	uint64_t		tmp;
1435	int			error;
1436
1437	/*
1438	 * Count the number of extended attribute blocks
1439	 */
1440	if (XFS_IFORK_Q(ip) && ip->i_afp->if_nextents > 0 &&
1441	    ip->i_afp->if_format != XFS_DINODE_FMT_LOCAL) {
1442		error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &junk,
1443				&aforkblks);
1444		if (error)
1445			return error;
1446	}
1447	if (XFS_IFORK_Q(tip) && tip->i_afp->if_nextents > 0 &&
1448	    tip->i_afp->if_format != XFS_DINODE_FMT_LOCAL) {
1449		error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK, &junk,
1450				&taforkblks);
1451		if (error)
1452			return error;
1453	}
1454
1455	/*
1456	 * Btree format (v3) inodes have the inode number stamped in the bmbt
1457	 * block headers. We can't start changing the bmbt blocks until the
1458	 * inode owner change is logged so recovery does the right thing in the
1459	 * event of a crash. Set the owner change log flags now and leave the
1460	 * bmbt scan as the last step.
1461	 */
1462	if (xfs_sb_version_has_v3inode(&ip->i_mount->m_sb)) {
1463		if (ip->i_df.if_format == XFS_DINODE_FMT_BTREE)
1464			(*target_log_flags) |= XFS_ILOG_DOWNER;
1465		if (tip->i_df.if_format == XFS_DINODE_FMT_BTREE)
1466			(*src_log_flags) |= XFS_ILOG_DOWNER;
1467	}
1468
1469	/*
1470	 * Swap the data forks of the inodes
1471	 */
1472	swap(ip->i_df, tip->i_df);
1473
1474	/*
1475	 * Fix the on-disk inode values
1476	 */
1477	tmp = (uint64_t)ip->i_d.di_nblocks;
1478	ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
1479	tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
1480
1481	/*
1482	 * The extents in the source inode could still contain speculative
1483	 * preallocation beyond EOF (e.g. the file is open but not modified
1484	 * while defrag is in progress). In that case, we need to copy over the
1485	 * number of delalloc blocks the data fork in the source inode is
1486	 * tracking beyond EOF so that when the fork is truncated away when the
1487	 * temporary inode is unlinked we don't underrun the i_delayed_blks
1488	 * counter on that inode.
1489	 */
1490	ASSERT(tip->i_delayed_blks == 0);
1491	tip->i_delayed_blks = ip->i_delayed_blks;
1492	ip->i_delayed_blks = 0;
1493
1494	switch (ip->i_df.if_format) {
1495	case XFS_DINODE_FMT_EXTENTS:
1496		(*src_log_flags) |= XFS_ILOG_DEXT;
1497		break;
1498	case XFS_DINODE_FMT_BTREE:
1499		ASSERT(!xfs_sb_version_has_v3inode(&ip->i_mount->m_sb) ||
1500		       (*src_log_flags & XFS_ILOG_DOWNER));
1501		(*src_log_flags) |= XFS_ILOG_DBROOT;
1502		break;
1503	}
1504
1505	switch (tip->i_df.if_format) {
1506	case XFS_DINODE_FMT_EXTENTS:
1507		(*target_log_flags) |= XFS_ILOG_DEXT;
1508		break;
1509	case XFS_DINODE_FMT_BTREE:
1510		(*target_log_flags) |= XFS_ILOG_DBROOT;
1511		ASSERT(!xfs_sb_version_has_v3inode(&ip->i_mount->m_sb) ||
1512		       (*target_log_flags & XFS_ILOG_DOWNER));
1513		break;
1514	}
1515
1516	return 0;
1517}
1518
1519/*
1520 * Fix up the owners of the bmbt blocks to refer to the current inode. The
1521 * change owner scan attempts to order all modified buffers in the current
1522 * transaction. In the event of ordered buffer failure, the offending buffer is
1523 * physically logged as a fallback and the scan returns -EAGAIN. We must roll
1524 * the transaction in this case to replenish the fallback log reservation and
1525 * restart the scan. This process repeats until the scan completes.
1526 */
1527static int
1528xfs_swap_change_owner(
1529	struct xfs_trans	**tpp,
1530	struct xfs_inode	*ip,
1531	struct xfs_inode	*tmpip)
1532{
1533	int			error;
1534	struct xfs_trans	*tp = *tpp;
1535
1536	do {
1537		error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK, ip->i_ino,
1538					      NULL);
1539		/* success or fatal error */
1540		if (error != -EAGAIN)
1541			break;
1542
1543		error = xfs_trans_roll(tpp);
1544		if (error)
1545			break;
1546		tp = *tpp;
1547
1548		/*
1549		 * Redirty both inodes so they can relog and keep the log tail
1550		 * moving forward.
1551		 */
1552		xfs_trans_ijoin(tp, ip, 0);
1553		xfs_trans_ijoin(tp, tmpip, 0);
1554		xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1555		xfs_trans_log_inode(tp, tmpip, XFS_ILOG_CORE);
1556	} while (true);
1557
1558	return error;
1559}
1560
1561int
1562xfs_swap_extents(
1563	struct xfs_inode	*ip,	/* target inode */
1564	struct xfs_inode	*tip,	/* tmp inode */
1565	struct xfs_swapext	*sxp)
1566{
1567	struct xfs_mount	*mp = ip->i_mount;
1568	struct xfs_trans	*tp;
1569	struct xfs_bstat	*sbp = &sxp->sx_stat;
1570	int			src_log_flags, target_log_flags;
1571	int			error = 0;
1572	int			lock_flags;
1573	uint64_t		f;
1574	int			resblks = 0;
1575	unsigned int		flags = 0;
1576
1577	/*
1578	 * Lock the inodes against other IO, page faults and truncate to
1579	 * begin with.  Then we can ensure the inodes are flushed and have no
1580	 * page cache safely. Once we have done this we can take the ilocks and
1581	 * do the rest of the checks.
1582	 */
1583	lock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1584	lock_flags = XFS_MMAPLOCK_EXCL;
1585	xfs_lock_two_inodes(ip, XFS_MMAPLOCK_EXCL, tip, XFS_MMAPLOCK_EXCL);
1586
1587	/* Verify that both files have the same format */
1588	if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) {
1589		error = -EINVAL;
1590		goto out_unlock;
1591	}
1592
1593	/* Verify both files are either real-time or non-realtime */
1594	if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1595		error = -EINVAL;
1596		goto out_unlock;
1597	}
1598
1599	error = xfs_qm_dqattach(ip);
1600	if (error)
1601		goto out_unlock;
1602
1603	error = xfs_qm_dqattach(tip);
1604	if (error)
1605		goto out_unlock;
1606
1607	error = xfs_swap_extent_flush(ip);
1608	if (error)
1609		goto out_unlock;
1610	error = xfs_swap_extent_flush(tip);
1611	if (error)
1612		goto out_unlock;
1613
1614	if (xfs_inode_has_cow_data(tip)) {
1615		error = xfs_reflink_cancel_cow_range(tip, 0, NULLFILEOFF, true);
1616		if (error)
1617			goto out_unlock;
1618	}
1619
1620	/*
1621	 * Extent "swapping" with rmap requires a permanent reservation and
1622	 * a block reservation because it's really just a remap operation
1623	 * performed with log redo items!
1624	 */
1625	if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
1626		int		w = XFS_DATA_FORK;
1627		uint32_t	ipnext = ip->i_df.if_nextents;
1628		uint32_t	tipnext	= tip->i_df.if_nextents;
1629
1630		/*
1631		 * Conceptually this shouldn't affect the shape of either bmbt,
1632		 * but since we atomically move extents one by one, we reserve
1633		 * enough space to rebuild both trees.
1634		 */
1635		resblks = XFS_SWAP_RMAP_SPACE_RES(mp, ipnext, w);
1636		resblks +=  XFS_SWAP_RMAP_SPACE_RES(mp, tipnext, w);
1637
1638		/*
1639		 * If either inode straddles a bmapbt block allocation boundary,
1640		 * the rmapbt algorithm triggers repeated allocs and frees as
1641		 * extents are remapped. This can exhaust the block reservation
1642		 * prematurely and cause shutdown. Return freed blocks to the
1643		 * transaction reservation to counter this behavior.
1644		 */
1645		flags |= XFS_TRANS_RES_FDBLKS;
1646	}
1647	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, flags,
1648				&tp);
1649	if (error)
1650		goto out_unlock;
1651
1652	/*
1653	 * Lock and join the inodes to the tansaction so that transaction commit
1654	 * or cancel will unlock the inodes from this point onwards.
1655	 */
1656	xfs_lock_two_inodes(ip, XFS_ILOCK_EXCL, tip, XFS_ILOCK_EXCL);
1657	lock_flags |= XFS_ILOCK_EXCL;
1658	xfs_trans_ijoin(tp, ip, 0);
1659	xfs_trans_ijoin(tp, tip, 0);
1660
1661
1662	/* Verify all data are being swapped */
1663	if (sxp->sx_offset != 0 ||
1664	    sxp->sx_length != ip->i_d.di_size ||
1665	    sxp->sx_length != tip->i_d.di_size) {
1666		error = -EFAULT;
1667		goto out_trans_cancel;
1668	}
1669
1670	trace_xfs_swap_extent_before(ip, 0);
1671	trace_xfs_swap_extent_before(tip, 1);
1672
1673	/* check inode formats now that data is flushed */
1674	error = xfs_swap_extents_check_format(ip, tip);
1675	if (error) {
1676		xfs_notice(mp,
1677		    "%s: inode 0x%llx format is incompatible for exchanging.",
1678				__func__, ip->i_ino);
1679		goto out_trans_cancel;
1680	}
1681
1682	/*
1683	 * Compare the current change & modify times with that
1684	 * passed in.  If they differ, we abort this swap.
1685	 * This is the mechanism used to ensure the calling
1686	 * process that the file was not changed out from
1687	 * under it.
1688	 */
1689	if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
1690	    (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
1691	    (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
1692	    (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
1693		error = -EBUSY;
1694		goto out_trans_cancel;
1695	}
1696
1697	/*
1698	 * Note the trickiness in setting the log flags - we set the owner log
1699	 * flag on the opposite inode (i.e. the inode we are setting the new
1700	 * owner to be) because once we swap the forks and log that, log
1701	 * recovery is going to see the fork as owned by the swapped inode,
1702	 * not the pre-swapped inodes.
1703	 */
1704	src_log_flags = XFS_ILOG_CORE;
1705	target_log_flags = XFS_ILOG_CORE;
1706
1707	if (xfs_sb_version_hasrmapbt(&mp->m_sb))
1708		error = xfs_swap_extent_rmap(&tp, ip, tip);
1709	else
1710		error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags,
1711				&target_log_flags);
1712	if (error)
1713		goto out_trans_cancel;
1714
1715	/* Do we have to swap reflink flags? */
1716	if ((ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK) ^
1717	    (tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)) {
1718		f = ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
1719		ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1720		ip->i_d.di_flags2 |= tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
1721		tip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1722		tip->i_d.di_flags2 |= f & XFS_DIFLAG2_REFLINK;
1723	}
1724
1725	/* Swap the cow forks. */
1726	if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1727		ASSERT(!ip->i_cowfp ||
1728		       ip->i_cowfp->if_format == XFS_DINODE_FMT_EXTENTS);
1729		ASSERT(!tip->i_cowfp ||
1730		       tip->i_cowfp->if_format == XFS_DINODE_FMT_EXTENTS);
1731
1732		swap(ip->i_cowfp, tip->i_cowfp);
1733
1734		if (ip->i_cowfp && ip->i_cowfp->if_bytes)
1735			xfs_inode_set_cowblocks_tag(ip);
1736		else
1737			xfs_inode_clear_cowblocks_tag(ip);
1738		if (tip->i_cowfp && tip->i_cowfp->if_bytes)
1739			xfs_inode_set_cowblocks_tag(tip);
1740		else
1741			xfs_inode_clear_cowblocks_tag(tip);
1742	}
1743
1744	xfs_trans_log_inode(tp, ip,  src_log_flags);
1745	xfs_trans_log_inode(tp, tip, target_log_flags);
1746
1747	/*
1748	 * The extent forks have been swapped, but crc=1,rmapbt=0 filesystems
1749	 * have inode number owner values in the bmbt blocks that still refer to
1750	 * the old inode. Scan each bmbt to fix up the owner values with the
1751	 * inode number of the current inode.
1752	 */
1753	if (src_log_flags & XFS_ILOG_DOWNER) {
1754		error = xfs_swap_change_owner(&tp, ip, tip);
1755		if (error)
1756			goto out_trans_cancel;
1757	}
1758	if (target_log_flags & XFS_ILOG_DOWNER) {
1759		error = xfs_swap_change_owner(&tp, tip, ip);
1760		if (error)
1761			goto out_trans_cancel;
1762	}
1763
1764	/*
1765	 * If this is a synchronous mount, make sure that the
1766	 * transaction goes to disk before returning to the user.
1767	 */
1768	if (mp->m_flags & XFS_MOUNT_WSYNC)
1769		xfs_trans_set_sync(tp);
1770
1771	error = xfs_trans_commit(tp);
1772
1773	trace_xfs_swap_extent_after(ip, 0);
1774	trace_xfs_swap_extent_after(tip, 1);
1775
1776out_unlock:
1777	xfs_iunlock(ip, lock_flags);
1778	xfs_iunlock(tip, lock_flags);
1779	unlock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1780	return error;
1781
1782out_trans_cancel:
1783	xfs_trans_cancel(tp);
1784	goto out_unlock;
1785}
1786