PageRenderTime 9ms CodeModel.GetById 2ms app.highlight 105ms RepoModel.GetById 1ms app.codeStats 0ms

/fs/xfs/xfs_bmap_util.c

http://github.com/torvalds/linux
C | 1777 lines | 1139 code | 231 blank | 407 comment | 254 complexity | 1e93b3e9340316a68b4a33ee2d5b2d6c MD5 | raw file
   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 uninitialized_var(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 (XFS_IFORK_FORMAT(ip, whichfork)) {
 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 (XFS_IFORK_FORMAT(ip, whichfork)) {
 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		/*
 804		 * Allocate and setup the transaction.
 805		 */
 806		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks,
 807				resrtextents, 0, &tp);
 808
 809		/*
 810		 * Check for running out of space
 811		 */
 812		if (error) {
 813			/*
 814			 * Free the transaction structure.
 815			 */
 816			ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
 817			break;
 818		}
 819		xfs_ilock(ip, XFS_ILOCK_EXCL);
 820		error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
 821						      0, quota_flag);
 822		if (error)
 823			goto error1;
 824
 825		xfs_trans_ijoin(tp, ip, 0);
 826
 827		error = xfs_bmapi_write(tp, ip, startoffset_fsb,
 828					allocatesize_fsb, alloc_type, 0, imapp,
 829					&nimaps);
 830		if (error)
 831			goto error0;
 832
 833		/*
 834		 * Complete the transaction
 835		 */
 836		error = xfs_trans_commit(tp);
 837		xfs_iunlock(ip, XFS_ILOCK_EXCL);
 838		if (error)
 839			break;
 840
 841		allocated_fsb = imapp->br_blockcount;
 842
 843		if (nimaps == 0) {
 844			error = -ENOSPC;
 845			break;
 846		}
 847
 848		startoffset_fsb += allocated_fsb;
 849		allocatesize_fsb -= allocated_fsb;
 850	}
 851
 852	return error;
 853
 854error0:	/* unlock inode, unreserve quota blocks, cancel trans */
 855	xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
 856
 857error1:	/* Just cancel transaction */
 858	xfs_trans_cancel(tp);
 859	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 860	return error;
 861}
 862
 863static int
 864xfs_unmap_extent(
 865	struct xfs_inode	*ip,
 866	xfs_fileoff_t		startoffset_fsb,
 867	xfs_filblks_t		len_fsb,
 868	int			*done)
 869{
 870	struct xfs_mount	*mp = ip->i_mount;
 871	struct xfs_trans	*tp;
 872	uint			resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
 873	int			error;
 874
 875	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
 876	if (error) {
 877		ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
 878		return error;
 879	}
 880
 881	xfs_ilock(ip, XFS_ILOCK_EXCL);
 882	error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot, ip->i_gdquot,
 883			ip->i_pdquot, resblks, 0, XFS_QMOPT_RES_REGBLKS);
 884	if (error)
 885		goto out_trans_cancel;
 886
 887	xfs_trans_ijoin(tp, ip, 0);
 888
 889	error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, done);
 890	if (error)
 891		goto out_trans_cancel;
 892
 893	error = xfs_trans_commit(tp);
 894out_unlock:
 895	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 896	return error;
 897
 898out_trans_cancel:
 899	xfs_trans_cancel(tp);
 900	goto out_unlock;
 901}
 902
 903/* Caller must first wait for the completion of any pending DIOs if required. */
 904int
 905xfs_flush_unmap_range(
 906	struct xfs_inode	*ip,
 907	xfs_off_t		offset,
 908	xfs_off_t		len)
 909{
 910	struct xfs_mount	*mp = ip->i_mount;
 911	struct inode		*inode = VFS_I(ip);
 912	xfs_off_t		rounding, start, end;
 913	int			error;
 914
 915	rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_SIZE);
 916	start = round_down(offset, rounding);
 917	end = round_up(offset + len, rounding) - 1;
 918
 919	error = filemap_write_and_wait_range(inode->i_mapping, start, end);
 920	if (error)
 921		return error;
 922	truncate_pagecache_range(inode, start, end);
 923	return 0;
 924}
 925
 926int
 927xfs_free_file_space(
 928	struct xfs_inode	*ip,
 929	xfs_off_t		offset,
 930	xfs_off_t		len)
 931{
 932	struct xfs_mount	*mp = ip->i_mount;
 933	xfs_fileoff_t		startoffset_fsb;
 934	xfs_fileoff_t		endoffset_fsb;
 935	int			done = 0, error;
 936
 937	trace_xfs_free_file_space(ip);
 938
 939	error = xfs_qm_dqattach(ip);
 940	if (error)
 941		return error;
 942
 943	if (len <= 0)	/* if nothing being freed */
 944		return 0;
 945
 946	startoffset_fsb = XFS_B_TO_FSB(mp, offset);
 947	endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
 948
 949	/*
 950	 * Need to zero the stuff we're not freeing, on disk.
 951	 */
 952	if (endoffset_fsb > startoffset_fsb) {
 953		while (!done) {
 954			error = xfs_unmap_extent(ip, startoffset_fsb,
 955					endoffset_fsb - startoffset_fsb, &done);
 956			if (error)
 957				return error;
 958		}
 959	}
 960
 961	/*
 962	 * Now that we've unmap all full blocks we'll have to zero out any
 963	 * partial block at the beginning and/or end.  iomap_zero_range is smart
 964	 * enough to skip any holes, including those we just created, but we
 965	 * must take care not to zero beyond EOF and enlarge i_size.
 966	 */
 967	if (offset >= XFS_ISIZE(ip))
 968		return 0;
 969	if (offset + len > XFS_ISIZE(ip))
 970		len = XFS_ISIZE(ip) - offset;
 971	error = iomap_zero_range(VFS_I(ip), offset, len, NULL,
 972			&xfs_buffered_write_iomap_ops);
 973	if (error)
 974		return error;
 975
 976	/*
 977	 * If we zeroed right up to EOF and EOF straddles a page boundary we
 978	 * must make sure that the post-EOF area is also zeroed because the
 979	 * page could be mmap'd and iomap_zero_range doesn't do that for us.
 980	 * Writeback of the eof page will do this, albeit clumsily.
 981	 */
 982	if (offset + len >= XFS_ISIZE(ip) && offset_in_page(offset + len) > 0) {
 983		error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
 984				round_down(offset + len, PAGE_SIZE), LLONG_MAX);
 985	}
 986
 987	return error;
 988}
 989
 990static int
 991xfs_prepare_shift(
 992	struct xfs_inode	*ip,
 993	loff_t			offset)
 994{
 995	struct xfs_mount	*mp = ip->i_mount;
 996	int			error;
 997
 998	/*
 999	 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
1000	 * into the accessible region of the file.
1001	 */
1002	if (xfs_can_free_eofblocks(ip, true)) {
1003		error = xfs_free_eofblocks(ip);
1004		if (error)
1005			return error;
1006	}
1007
1008	/*
1009	 * Shift operations must stabilize the start block offset boundary along
1010	 * with the full range of the operation. If we don't, a COW writeback
1011	 * completion could race with an insert, front merge with the start
1012	 * extent (after split) during the shift and corrupt the file. Start
1013	 * with the block just prior to the start to stabilize the boundary.
1014	 */
1015	offset = round_down(offset, 1 << mp->m_sb.sb_blocklog);
1016	if (offset)
1017		offset -= (1 << mp->m_sb.sb_blocklog);
1018
1019	/*
1020	 * Writeback and invalidate cache for the remainder of the file as we're
1021	 * about to shift down every extent from offset to EOF.
1022	 */
1023	error = xfs_flush_unmap_range(ip, offset, XFS_ISIZE(ip));
1024	if (error)
1025		return error;
1026
1027	/*
1028	 * Clean out anything hanging around in the cow fork now that
1029	 * we've flushed all the dirty data out to disk to avoid having
1030	 * CoW extents at the wrong offsets.
1031	 */
1032	if (xfs_inode_has_cow_data(ip)) {
1033		error = xfs_reflink_cancel_cow_range(ip, offset, NULLFILEOFF,
1034				true);
1035		if (error)
1036			return error;
1037	}
1038
1039	return 0;
1040}
1041
1042/*
1043 * xfs_collapse_file_space()
1044 *	This routine frees disk space and shift extent for the given file.
1045 *	The first thing we do is to free data blocks in the specified range
1046 *	by calling xfs_free_file_space(). It would also sync dirty data
1047 *	and invalidate page cache over the region on which collapse range
1048 *	is working. And Shift extent records to the left to cover a hole.
1049 * RETURNS:
1050 *	0 on success
1051 *	errno on error
1052 *
1053 */
1054int
1055xfs_collapse_file_space(
1056	struct xfs_inode	*ip,
1057	xfs_off_t		offset,
1058	xfs_off_t		len)
1059{
1060	struct xfs_mount	*mp = ip->i_mount;
1061	struct xfs_trans	*tp;
1062	int			error;
1063	xfs_fileoff_t		next_fsb = XFS_B_TO_FSB(mp, offset + len);
1064	xfs_fileoff_t		shift_fsb = XFS_B_TO_FSB(mp, len);
1065	bool			done = false;
1066
1067	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1068	ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
1069
1070	trace_xfs_collapse_file_space(ip);
1071
1072	error = xfs_free_file_space(ip, offset, len);
1073	if (error)
1074		return error;
1075
1076	error = xfs_prepare_shift(ip, offset);
1077	if (error)
1078		return error;
1079
1080	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
1081	if (error)
1082		return error;
1083
1084	xfs_ilock(ip, XFS_ILOCK_EXCL);
1085	xfs_trans_ijoin(tp, ip, 0);
1086
1087	while (!done) {
1088		error = xfs_bmap_collapse_extents(tp, ip, &next_fsb, shift_fsb,
1089				&done);
1090		if (error)
1091			goto out_trans_cancel;
1092		if (done)
1093			break;
1094
1095		/* finish any deferred frees and roll the transaction */
1096		error = xfs_defer_finish(&tp);
1097		if (error)
1098			goto out_trans_cancel;
1099	}
1100
1101	error = xfs_trans_commit(tp);
1102	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1103	return error;
1104
1105out_trans_cancel:
1106	xfs_trans_cancel(tp);
1107	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1108	return error;
1109}
1110
1111/*
1112 * xfs_insert_file_space()
1113 *	This routine create hole space by shifting extents for the given file.
1114 *	The first thing we do is to sync dirty data and invalidate page cache
1115 *	over the region on which insert range is working. And split an extent
1116 *	to two extents at given offset by calling xfs_bmap_split_extent.
1117 *	And shift all extent records which are laying between [offset,
1118 *	last allocated extent] to the right to reserve hole range.
1119 * RETURNS:
1120 *	0 on success
1121 *	errno on error
1122 */
1123int
1124xfs_insert_file_space(
1125	struct xfs_inode	*ip,
1126	loff_t			offset,
1127	loff_t			len)
1128{
1129	struct xfs_mount	*mp = ip->i_mount;
1130	struct xfs_trans	*tp;
1131	int			error;
1132	xfs_fileoff_t		stop_fsb = XFS_B_TO_FSB(mp, offset);
1133	xfs_fileoff_t		next_fsb = NULLFSBLOCK;
1134	xfs_fileoff_t		shift_fsb = XFS_B_TO_FSB(mp, len);
1135	bool			done = false;
1136
1137	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1138	ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
1139
1140	trace_xfs_insert_file_space(ip);
1141
1142	error = xfs_bmap_can_insert_extents(ip, stop_fsb, shift_fsb);
1143	if (error)
1144		return error;
1145
1146	error = xfs_prepare_shift(ip, offset);
1147	if (error)
1148		return error;
1149
1150	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
1151			XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
1152	if (error)
1153		return error;
1154
1155	xfs_ilock(ip, XFS_ILOCK_EXCL);
1156	xfs_trans_ijoin(tp, ip, 0);
1157
1158	/*
1159	 * The extent shifting code works on extent granularity. So, if stop_fsb
1160	 * is not the starting block of extent, we need to split the extent at
1161	 * stop_fsb.
1162	 */
1163	error = xfs_bmap_split_extent(tp, ip, stop_fsb);
1164	if (error)
1165		goto out_trans_cancel;
1166
1167	do {
1168		error = xfs_trans_roll_inode(&tp, ip);
1169		if (error)
1170			goto out_trans_cancel;
1171
1172		error = xfs_bmap_insert_extents(tp, ip, &next_fsb, shift_fsb,
1173				&done, stop_fsb);
1174		if (error)
1175			goto out_trans_cancel;
1176	} while (!done);
1177
1178	error = xfs_trans_commit(tp);
1179	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1180	return error;
1181
1182out_trans_cancel:
1183	xfs_trans_cancel(tp);
1184	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1185	return error;
1186}
1187
1188/*
1189 * We need to check that the format of the data fork in the temporary inode is
1190 * valid for the target inode before doing the swap. This is not a problem with
1191 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1192 * data fork depending on the space the attribute fork is taking so we can get
1193 * invalid formats on the target inode.
1194 *
1195 * E.g. target has space for 7 extents in extent format, temp inode only has
1196 * space for 6.  If we defragment down to 7 extents, then the tmp format is a
1197 * btree, but when swapped it needs to be in extent format. Hence we can't just
1198 * blindly swap data forks on attr2 filesystems.
1199 *
1200 * Note that we check the swap in both directions so that we don't end up with
1201 * a corrupt temporary inode, either.
1202 *
1203 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1204 * inode will prevent this situation from occurring, so all we do here is
1205 * reject and log the attempt. basically we are putting the responsibility on
1206 * userspace to get this right.
1207 */
1208static int
1209xfs_swap_extents_check_format(
1210	struct xfs_inode	*ip,	/* target inode */
1211	struct xfs_inode	*tip)	/* tmp inode */
1212{
1213
1214	/* Should never get a local format */
1215	if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
1216	    tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
1217		return -EINVAL;
1218
1219	/*
1220	 * if the target inode has less extents that then temporary inode then
1221	 * why did userspace call us?
1222	 */
1223	if (ip->i_d.di_nextents < tip->i_d.di_nextents)
1224		return -EINVAL;
1225
1226	/*
1227	 * If we have to use the (expensive) rmap swap method, we can
1228	 * handle any number of extents and any format.
1229	 */
1230	if (xfs_sb_version_hasrmapbt(&ip->i_mount->m_sb))
1231		return 0;
1232
1233	/*
1234	 * if the target inode is in extent form and the temp inode is in btree
1235	 * form then we will end up with the target inode in the wrong format
1236	 * as we already know there are less extents in the temp inode.
1237	 */
1238	if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1239	    tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1240		return -EINVAL;
1241
1242	/* Check temp in extent form to max in target */
1243	if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1244	    XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) >
1245			XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1246		return -EINVAL;
1247
1248	/* Check target in extent form to max in temp */
1249	if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1250	    XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) >
1251			XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1252		return -EINVAL;
1253
1254	/*
1255	 * If we are in a btree format, check that the temp root block will fit
1256	 * in the target and that it has enough extents to be in btree format
1257	 * in the target.
1258	 *
1259	 * Note that we have to be careful to allow btree->extent conversions
1260	 * (a common defrag case) which will occur when the temp inode is in
1261	 * extent format...
1262	 */
1263	if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1264		if (XFS_IFORK_Q(ip) &&
1265		    XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip))
1266			return -EINVAL;
1267		if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <=
1268		    XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1269			return -EINVAL;
1270	}
1271
1272	/* Reciprocal target->temp btree format checks */
1273	if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1274		if (XFS_IFORK_Q(tip) &&
1275		    XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
1276			return -EINVAL;
1277		if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <=
1278		    XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1279			return -EINVAL;
1280	}
1281
1282	return 0;
1283}
1284
1285static int
1286xfs_swap_extent_flush(
1287	struct xfs_inode	*ip)
1288{
1289	int	error;
1290
1291	error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1292	if (error)
1293		return error;
1294	truncate_pagecache_range(VFS_I(ip), 0, -1);
1295
1296	/* Verify O_DIRECT for ftmp */
1297	if (VFS_I(ip)->i_mapping->nrpages)
1298		return -EINVAL;
1299	return 0;
1300}
1301
1302/*
1303 * Move extents from one file to another, when rmap is enabled.
1304 */
1305STATIC int
1306xfs_swap_extent_rmap(
1307	struct xfs_trans		**tpp,
1308	struct xfs_inode		*ip,
1309	struct xfs_inode		*tip)
1310{
1311	struct xfs_trans		*tp = *tpp;
1312	struct xfs_bmbt_irec		irec;
1313	struct xfs_bmbt_irec		uirec;
1314	struct xfs_bmbt_irec		tirec;
1315	xfs_fileoff_t			offset_fsb;
1316	xfs_fileoff_t			end_fsb;
1317	xfs_filblks_t			count_fsb;
1318	int				error;
1319	xfs_filblks_t			ilen;
1320	xfs_filblks_t			rlen;
1321	int				nimaps;
1322	uint64_t			tip_flags2;
1323
1324	/*
1325	 * If the source file has shared blocks, we must flag the donor
1326	 * file as having shared blocks so that we get the shared-block
1327	 * rmap functions when we go to fix up the rmaps.  The flags
1328	 * will be switch for reals later.
1329	 */
1330	tip_flags2 = tip->i_d.di_flags2;
1331	if (ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)
1332		tip->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK;
1333
1334	offset_fsb = 0;
1335	end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip)));
1336	count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
1337
1338	while (count_fsb) {
1339		/* Read extent from the donor file */
1340		nimaps = 1;
1341		error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec,
1342				&nimaps, 0);
1343		if (error)
1344			goto out;
1345		ASSERT(nimaps == 1);
1346		ASSERT(tirec.br_startblock != DELAYSTARTBLOCK);
1347
1348		trace_xfs_swap_extent_rmap_remap(tip, &tirec);
1349		ilen = tirec.br_blockcount;
1350
1351		/* Unmap the old blocks in the source file. */
1352		while (tirec.br_blockcount) {
1353			ASSERT(tp->t_firstblock == NULLFSBLOCK);
1354			trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec);
1355
1356			/* Read extent from the source file */
1357			nimaps = 1;
1358			error = xfs_bmapi_read(ip, tirec.br_startoff,
1359					tirec.br_blockcount, &irec,
1360					&nimaps, 0);
1361			if (error)
1362				goto out;
1363			ASSERT(nimaps == 1);
1364			ASSERT(tirec.br_startoff == irec.br_startoff);
1365			trace_xfs_swap_extent_rmap_remap_piece(ip, &irec);
1366
1367			/* Trim the extent. */
1368			uirec = tirec;
1369			uirec.br_blockcount = rlen = min_t(xfs_filblks_t,
1370					tirec.br_blockcount,
1371					irec.br_blockcount);
1372			trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec);
1373
1374			/* Remove the mapping from the donor file. */
1375			xfs_bmap_unmap_extent(tp, tip, &uirec);
1376
1377			/* Remove the mapping from the source file. */
1378			xfs_bmap_unmap_extent(tp, ip, &irec);
1379
1380			/* Map the donor file's blocks into the source file. */
1381			xfs_bmap_map_extent(tp, ip, &uirec);
1382
1383			/* Map the source file's blocks into the donor file. */
1384			xfs_bmap_map_extent(tp, tip, &irec);
1385
1386			error = xfs_defer_finish(tpp);
1387			tp = *tpp;
1388			if (error)
1389				goto out;
1390
1391			tirec.br_startoff += rlen;
1392			if (tirec.br_startblock != HOLESTARTBLOCK &&
1393			    tirec.br_startblock != DELAYSTARTBLOCK)
1394				tirec.br_startblock += rlen;
1395			tirec.br_blockcount -= rlen;
1396		}
1397
1398		/* Roll on... */
1399		count_fsb -= ilen;
1400		offset_fsb += ilen;
1401	}
1402
1403	tip->i_d.di_flags2 = tip_flags2;
1404	return 0;
1405
1406out:
1407	trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_);
1408	tip->i_d.di_flags2 = tip_flags2;
1409	return error;
1410}
1411
1412/* Swap the extents of two files by swapping data forks. */
1413STATIC int
1414xfs_swap_extent_forks(
1415	struct xfs_trans	*tp,
1416	struct xfs_inode	*ip,
1417	struct xfs_inode	*tip,
1418	int			*src_log_flags,
1419	int			*target_log_flags)
1420{
1421	xfs_filblks_t		aforkblks = 0;
1422	xfs_filblks_t		taforkblks = 0;
1423	xfs_extnum_t		junk;
1424	uint64_t		tmp;
1425	int			error;
1426
1427	/*
1428	 * Count the number of extended attribute blocks
1429	 */
1430	if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
1431	     (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1432		error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &junk,
1433				&aforkblks);
1434		if (error)
1435			return error;
1436	}
1437	if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
1438	     (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1439		error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK, &junk,
1440				&taforkblks);
1441		if (error)
1442			return error;
1443	}
1444
1445	/*
1446	 * Btree format (v3) inodes have the inode number stamped in the bmbt
1447	 * block headers. We can't start changing the bmbt blocks until the
1448	 * inode owner change is logged so recovery does the right thing in the
1449	 * event of a crash. Set the owner change log flags now and leave the
1450	 * bmbt scan as the last step.
1451	 */
1452	if (xfs_sb_version_has_v3inode(&ip->i_mount->m_sb)) {
1453		if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1454			(*target_log_flags) |= XFS_ILOG_DOWNER;
1455		if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1456			(*src_log_flags) |= XFS_ILOG_DOWNER;
1457	}
1458
1459	/*
1460	 * Swap the data forks of the inodes
1461	 */
1462	swap(ip->i_df, tip->i_df);
1463
1464	/*
1465	 * Fix the on-disk inode values
1466	 */
1467	tmp = (uint64_t)ip->i_d.di_nblocks;
1468	ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
1469	tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
1470
1471	swap(ip->i_d.di_nextents, tip->i_d.di_nextents);
1472	swap(ip->i_d.di_format, tip->i_d.di_format);
1473
1474	/*
1475	 * The extents in the source inode could still contain speculative
1476	 * preallocation beyond EOF (e.g. the file is open but not modified
1477	 * while defrag is in progress). In that case, we need to copy over the
1478	 * number of delalloc blocks the data fork in the source inode is
1479	 * tracking beyond EOF so that when the fork is truncated away when the
1480	 * temporary inode is unlinked we don't underrun the i_delayed_blks
1481	 * counter on that inode.
1482	 */
1483	ASSERT(tip->i_delayed_blks == 0);
1484	tip->i_delayed_blks = ip->i_delayed_blks;
1485	ip->i_delayed_blks = 0;
1486
1487	switch (ip->i_d.di_format) {
1488	case XFS_DINODE_FMT_EXTENTS:
1489		(*src_log_flags) |= XFS_ILOG_DEXT;
1490		break;
1491	case XFS_DINODE_FMT_BTREE:
1492		ASSERT(!xfs_sb_version_has_v3inode(&ip->i_mount->m_sb) ||
1493		       (*src_log_flags & XFS_ILOG_DOWNER));
1494		(*src_log_flags) |= XFS_ILOG_DBROOT;
1495		break;
1496	}
1497
1498	switch (tip->i_d.di_format) {
1499	case XFS_DINODE_FMT_EXTENTS:
1500		(*target_log_flags) |= XFS_ILOG_DEXT;
1501		break;
1502	case XFS_DINODE_FMT_BTREE:
1503		(*target_log_flags) |= XFS_ILOG_DBROOT;
1504		ASSERT(!xfs_sb_version_has_v3inode(&ip->i_mount->m_sb) ||
1505		       (*target_log_flags & XFS_ILOG_DOWNER));
1506		break;
1507	}
1508
1509	return 0;
1510}
1511
1512/*
1513 * Fix up the owners of the bmbt blocks to refer to the current inode. The
1514 * change owner scan attempts to order all modified buffers in the current
1515 * transaction. In the event of ordered buffer failure, the offending buffer is
1516 * physically logged as a fallback and the scan returns -EAGAIN. We must roll
1517 * the transaction in this case to replenish the fallback log reservation and
1518 * restart the scan. This process repeats until the scan completes.
1519 */
1520static int
1521xfs_swap_change_owner(
1522	struct xfs_trans	**tpp,
1523	struct xfs_inode	*ip,
1524	struct xfs_inode	*tmpip)
1525{
1526	int			error;
1527	struct xfs_trans	*tp = *tpp;
1528
1529	do {
1530		error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK, ip->i_ino,
1531					      NULL);
1532		/* success or fatal error */
1533		if (error != -EAGAIN)
1534			break;
1535
1536		error = xfs_trans_roll(tpp);
1537		if (error)
1538			break;
1539		tp = *tpp;
1540
1541		/*
1542		 * Redirty both inodes so they can relog and keep the log tail
1543		 * moving forward.
1544		 */
1545		xfs_trans_ijoin(tp, ip, 0);
1546		xfs_trans_ijoin(tp, tmpip, 0);
1547		xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1548		xfs_trans_log_inode(tp, tmpip, XFS_ILOG_CORE);
1549	} while (true);
1550
1551	return error;
1552}
1553
1554int
1555xfs_swap_extents(
1556	struct xfs_inode	*ip,	/* target inode */
1557	struct xfs_inode	*tip,	/* tmp inode */
1558	struct xfs_swapext	*sxp)
1559{
1560	struct xfs_mount	*mp = ip->i_mount;
1561	struct xfs_trans	*tp;
1562	struct xfs_bstat	*sbp = &sxp->sx_stat;
1563	int			src_log_flags, target_log_flags;
1564	int			error = 0;
1565	int			lock_flags;
1566	uint64_t		f;
1567	int			resblks = 0;
1568
1569	/*
1570	 * Lock the inodes against other IO, page faults and truncate to
1571	 * begin with.  Then we can ensure the inodes are flushed and have no
1572	 * page cache safely. Once we have done this we can take the ilocks and
1573	 * do the rest of the checks.
1574	 */
1575	lock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1576	lock_flags = XFS_MMAPLOCK_EXCL;
1577	xfs_lock_two_inodes(ip, XFS_MMAPLOCK_EXCL, tip, XFS_MMAPLOCK_EXCL);
1578
1579	/* Verify that both files have the same format */
1580	if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) {
1581		error = -EINVAL;
1582		goto out_unlock;
1583	}
1584
1585	/* Verify both files are either real-time or non-realtime */
1586	if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1587		error = -EINVAL;
1588		goto out_unlock;
1589	}
1590
1591	error = xfs_qm_dqattach(ip);
1592	if (error)
1593		goto out_unlock;
1594
1595	error = xfs_qm_dqattach(tip);
1596	if (error)
1597		goto out_unlock;
1598
1599	error = xfs_swap_extent_flush(ip);
1600	if (error)
1601		goto out_unlock;
1602	error = xfs_swap_extent_flush(tip);
1603	if (error)
1604		goto out_unlock;
1605
1606	if (xfs_inode_has_cow_data(tip)) {
1607		error = xfs_reflink_cancel_cow_range(tip, 0, NULLFILEOFF, true);
1608		if (error)
1609			return error;
1610	}
1611
1612	/*
1613	 * Extent "swapping" with rmap requires a permanent reservation and
1614	 * a block reservation because it's really just a remap operation
1615	 * performed with log redo items!
1616	 */
1617	if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
1618		int		w	= XFS_DATA_FORK;
1619		uint32_t	ipnext	= XFS_IFORK_NEXTENTS(ip, w);
1620		uint32_t	tipnext	= XFS_IFORK_NEXTENTS(tip, w);
1621
1622		/*
1623		 * Conceptually this shouldn't affect the shape of either bmbt,
1624		 * but since we atomically move extents one by one, we reserve
1625		 * enough space to rebuild both trees.
1626		 */
1627		resblks = XFS_SWAP_RMAP_SPACE_RES(mp, ipnext, w);
1628		resblks +=  XFS_SWAP_RMAP_SPACE_RES(mp, tipnext, w);
1629
1630		/*
1631		 * Handle the corner case where either inode might straddle the
1632		 * btree format boundary. If so, the inode could bounce between
1633		 * btree <-> extent format on unmap -> remap cycles, freeing and
1634		 * allocating a bmapbt block each time.
1635		 */
1636		if (ipnext == (XFS_IFORK_MAXEXT(ip, w) + 1))
1637			resblks += XFS_IFORK_MAXEXT(ip, w);
1638		if (tipnext == (XFS_IFORK_MAXEXT(tip, w) + 1))
1639			resblks += XFS_IFORK_MAXEXT(tip, w);
1640	}
1641	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
1642	if (error)
1643		goto out_unlock;
1644
1645	/*
1646	 * Lock and join the inodes to the tansaction so that transaction commit
1647	 * or cancel will unlock the inodes from this point onwards.
1648	 */
1649	xfs_lock_two_inodes(ip, XFS_ILOCK_EXCL, tip, XFS_ILOCK_EXCL);
1650	lock_flags |= XFS_ILOCK_EXCL;
1651	xfs_trans_ijoin(tp, ip, 0);
1652	xfs_trans_ijoin(tp, tip, 0);
1653
1654
1655	/* Verify all data are being swapped */
1656	if (sxp->sx_offset != 0 ||
1657	    sxp->sx_length != ip->i_d.di_size ||
1658	    sxp->sx_length != tip->i_d.di_size) {
1659		error = -EFAULT;
1660		goto out_trans_cancel;
1661	}
1662
1663	trace_xfs_swap_extent_before(ip, 0);
1664	trace_xfs_swap_extent_before(tip, 1);
1665
1666	/* check inode formats now that data is flushed */
1667	error = xfs_swap_extents_check_format(ip, tip);
1668	if (error) {
1669		xfs_notice(mp,
1670		    "%s: inode 0x%llx format is incompatible for exchanging.",
1671				__func__, ip->i_ino);
1672		goto out_trans_cancel;
1673	}
1674
1675	/*
1676	 * Compare the current change & modify times with that
1677	 * passed in.  If they differ, we abort this swap.
1678	 * This is the mechanism used to ensure the calling
1679	 * process that the file was not changed out from
1680	 * under it.
1681	 */
1682	if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
1683	    (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
1684	    (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
1685	    (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
1686		error = -EBUSY;
1687		goto out_trans_cancel;
1688	}
1689
1690	/*
1691	 * Note the trickiness in setting the log flags - we set the owner log
1692	 * flag on the opposite inode (i.e. the inode we are setting the new
1693	 * owner to be) because once we swap the forks and log that, log
1694	 * recovery is going to see the fork as owned by the swapped inode,
1695	 * not the pre-swapped inodes.
1696	 */
1697	src_log_flags = XFS_ILOG_CORE;
1698	target_log_flags = XFS_ILOG_CORE;
1699
1700	if (xfs_sb_version_hasrmapbt(&mp->m_sb))
1701		error = xfs_swap_extent_rmap(&tp, ip, tip);
1702	else
1703		error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags,
1704				&target_log_flags);
1705	if (error)
1706		goto out_trans_cancel;
1707
1708	/* Do we have to swap reflink flags? */
1709	if ((ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK) ^
1710	    (tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)) {
1711		f = ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
1712		ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1713		ip->i_d.di_flags2 |= tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
1714		tip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1715		tip->i_d.di_flags2 |= f & XFS_DIFLAG2_REFLINK;
1716	}
1717
1718	/* Swap the cow forks. */
1719	if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1720		ASSERT(ip->i_cformat == XFS_DINODE_FMT_EXTENTS);
1721		ASSERT(tip->i_cformat == XFS_DINODE_FMT_EXTENTS);
1722
1723		swap(ip->i_cnextents, tip->i_cnextents);
1724		swap(ip->i_cowfp, tip->i_cowfp);
1725
1726		if (ip->i_cowfp && ip->i_cowfp->if_bytes)
1727			xfs_inode_set_cowblocks_tag(ip);
1728		else
1729			xfs_inode_clear_cowblocks_tag(ip);
1730		if (tip->i_cowfp && tip->i_cowfp->if_bytes)
1731			xfs_inode_set_cowblocks_tag(tip);
1732		else
1733			xfs_inode_clear_cowblocks_tag(tip);
1734	}
1735
1736	xfs_trans_log_inode(tp, ip,  src_log_flags);
1737	xfs_trans_log_inode(tp, tip, target_log_flags);
1738
1739	/*
1740	 * The extent forks have been swapped, but crc=1,rmapbt=0 filesystems
1741	 * have inode number owner values in the bmbt blocks that still refer to
1742	 * the old inode. Scan each bmbt to fix up the owner values with the
1743	 * inode number of the current inode.
1744	 */
1745	if (src_log_flags & XFS_ILOG_DOWNER) {
1746		error = xfs_swap_change_owner(&tp, ip, tip);
1747		if (error)
1748			goto out_trans_cancel;
1749	}
1750	if (target_log_flags & XFS_ILOG_DOWNER) {
1751		error = xfs_swap_change_owner(&tp, tip, ip);
1752		if (error)
1753			goto out_trans_cancel;
1754	}
1755
1756	/*
1757	 * If this is a synchronous mount, make sure that the
1758	 * transaction goes to disk before returning to the user.
1759	 */
1760	if (mp->m_flags & XFS_MOUNT_WSYNC)
1761		xfs_trans_set_sync(tp);
1762
1763	error = xfs_trans_commit(tp);
1764
1765	trace_xfs_swap_extent_after(ip, 0);
1766	trace_xfs_swap_extent_after(tip, 1);
1767
1768out_unlock:
1769	xfs_iunlock(ip, lock_flags);
1770	xfs_iunlock(tip, lock_flags);
1771	unlock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1772	return error;
1773
1774out_trans_cancel:
1775	xfs_trans_cancel(tp);
1776	goto out_unlock;
1777}