1/*
   2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   3 * Copyright (c) 2012 Red Hat, Inc.
   4 * All Rights Reserved.
   5 *
   6 * This program is free software; you can redistribute it and/or
   7 * modify it under the terms of the GNU General Public License as
   8 * published by the Free Software Foundation.
   9 *
  10 * This program is distributed in the hope that it would be useful,
  11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13 * GNU General Public License for more details.
  14 *
  15 * You should have received a copy of the GNU General Public License
  16 * along with this program; if not, write the Free Software Foundation,
  17 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  18 */
  19#include "xfs.h"
  20#include "xfs_fs.h"
  21#include "xfs_shared.h"
  22#include "xfs_format.h"
  23#include "xfs_log_format.h"
  24#include "xfs_trans_resv.h"
  25#include "xfs_bit.h"
  26#include "xfs_mount.h"
  27#include "xfs_da_format.h"
  28#include "xfs_defer.h"
  29#include "xfs_inode.h"
  30#include "xfs_btree.h"
  31#include "xfs_trans.h"
  32#include "xfs_extfree_item.h"
  33#include "xfs_alloc.h"
  34#include "xfs_bmap.h"
  35#include "xfs_bmap_util.h"
  36#include "xfs_bmap_btree.h"
  37#include "xfs_rtalloc.h"
  38#include "xfs_error.h"
  39#include "xfs_quota.h"
  40#include "xfs_trans_space.h"
  41#include "xfs_trace.h"
  42#include "xfs_icache.h"
  43#include "xfs_log.h"
  44#include "xfs_rmap_btree.h"
  45#include "xfs_iomap.h"
  46#include "xfs_reflink.h"
  47#include "xfs_refcount.h"
  48
  49/* Kernel only BMAP related definitions and functions */
  50
  51/*
  52 * Convert the given file system block to a disk block.  We have to treat it
  53 * differently based on whether the file is a real time file or not, because the
  54 * bmap code does.
  55 */
  56xfs_daddr_t
  57xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb)
  58{
  59	return (XFS_IS_REALTIME_INODE(ip) ? \
  60		 (xfs_daddr_t)XFS_FSB_TO_BB((ip)->i_mount, (fsb)) : \
  61		 XFS_FSB_TO_DADDR((ip)->i_mount, (fsb)));
  62}
  63
  64/*
  65 * Routine to zero an extent on disk allocated to the specific inode.
  66 *
  67 * The VFS functions take a linearised filesystem block offset, so we have to
  68 * convert the sparse xfs fsb to the right format first.
  69 * VFS types are real funky, too.
  70 */
  71int
  72xfs_zero_extent(
  73	struct xfs_inode *ip,
  74	xfs_fsblock_t	start_fsb,
  75	xfs_off_t	count_fsb)
  76{
  77	struct xfs_mount *mp = ip->i_mount;
  78	xfs_daddr_t	sector = xfs_fsb_to_db(ip, start_fsb);
  79	sector_t	block = XFS_BB_TO_FSBT(mp, sector);
  80
  81	return blkdev_issue_zeroout(xfs_find_bdev_for_inode(VFS_I(ip)),
  82		block << (mp->m_super->s_blocksize_bits - 9),
  83		count_fsb << (mp->m_super->s_blocksize_bits - 9),
  84		GFP_NOFS, 0);
  85}
  86
  87#ifdef CONFIG_XFS_RT
  88int
  89xfs_bmap_rtalloc(
  90	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
  91{
  92	int		error;		/* error return value */
  93	xfs_mount_t	*mp;		/* mount point structure */
  94	xfs_extlen_t	prod = 0;	/* product factor for allocators */
  95	xfs_extlen_t	ralen = 0;	/* realtime allocation length */
  96	xfs_extlen_t	align;		/* minimum allocation alignment */
  97	xfs_rtblock_t	rtb;
  98
  99	mp = ap->ip->i_mount;
 100	align = xfs_get_extsz_hint(ap->ip);
 101	prod = align / mp->m_sb.sb_rextsize;
 102	error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
 103					align, 1, ap->eof, 0,
 104					ap->conv, &ap->offset, &ap->length);
 105	if (error)
 106		return error;
 107	ASSERT(ap->length);
 108	ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
 109
 110	/*
 111	 * If the offset & length are not perfectly aligned
 112	 * then kill prod, it will just get us in trouble.
 113	 */
 114	if (do_mod(ap->offset, align) || ap->length % align)
 115		prod = 1;
 116	/*
 117	 * Set ralen to be the actual requested length in rtextents.
 118	 */
 119	ralen = ap->length / mp->m_sb.sb_rextsize;
 120	/*
 121	 * If the old value was close enough to MAXEXTLEN that
 122	 * we rounded up to it, cut it back so it's valid again.
 123	 * Note that if it's a really large request (bigger than
 124	 * MAXEXTLEN), we don't hear about that number, and can't
 125	 * adjust the starting point to match it.
 126	 */
 127	if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
 128		ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
 129
 130	/*
 131	 * Lock out modifications to both the RT bitmap and summary inodes
 132	 */
 133	xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
 134	xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
 135	xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
 136	xfs_trans_ijoin(ap->tp, mp->m_rsumip, XFS_ILOCK_EXCL);
 137
 138	/*
 139	 * If it's an allocation to an empty file at offset 0,
 140	 * pick an extent that will space things out in the rt area.
 141	 */
 142	if (ap->eof && ap->offset == 0) {
 143		xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
 144
 145		error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
 146		if (error)
 147			return error;
 148		ap->blkno = rtx * mp->m_sb.sb_rextsize;
 149	} else {
 150		ap->blkno = 0;
 151	}
 152
 153	xfs_bmap_adjacent(ap);
 154
 155	/*
 156	 * Realtime allocation, done through xfs_rtallocate_extent.
 157	 */
 158	do_div(ap->blkno, mp->m_sb.sb_rextsize);
 159	rtb = ap->blkno;
 160	ap->length = ralen;
 161	error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
 162				&ralen, ap->wasdel, prod, &rtb);
 163	if (error)
 164		return error;
 165
 166	ap->blkno = rtb;
 167	if (ap->blkno != NULLFSBLOCK) {
 168		ap->blkno *= mp->m_sb.sb_rextsize;
 169		ralen *= mp->m_sb.sb_rextsize;
 170		ap->length = ralen;
 171		ap->ip->i_d.di_nblocks += ralen;
 172		xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
 173		if (ap->wasdel)
 174			ap->ip->i_delayed_blks -= ralen;
 175		/*
 176		 * Adjust the disk quota also. This was reserved
 177		 * earlier.
 178		 */
 179		xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
 180			ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
 181					XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
 182
 183		/* Zero the extent if we were asked to do so */
 184		if (ap->datatype & XFS_ALLOC_USERDATA_ZERO) {
 185			error = xfs_zero_extent(ap->ip, ap->blkno, ap->length);
 186			if (error)
 187				return error;
 188		}
 189	} else {
 190		ap->length = 0;
 191	}
 192	return 0;
 193}
 194#endif /* CONFIG_XFS_RT */
 195
 196/*
 197 * Check if the endoff is outside the last extent. If so the caller will grow
 198 * the allocation to a stripe unit boundary.  All offsets are considered outside
 199 * the end of file for an empty fork, so 1 is returned in *eof in that case.
 200 */
 201int
 202xfs_bmap_eof(
 203	struct xfs_inode	*ip,
 204	xfs_fileoff_t		endoff,
 205	int			whichfork,
 206	int			*eof)
 207{
 208	struct xfs_bmbt_irec	rec;
 209	int			error;
 210
 211	error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
 212	if (error || *eof)
 213		return error;
 214
 215	*eof = endoff >= rec.br_startoff + rec.br_blockcount;
 216	return 0;
 
 
 217}
 218
 219/*
 220 * Extent tree block counting routines.
 221 */
 222
 223/*
 224 * Count leaf blocks given a range of extent records.  Delayed allocation
 225 * extents are not counted towards the totals.
 226 */
 227xfs_extnum_t
 228xfs_bmap_count_leaves(
 229	struct xfs_ifork	*ifp,
 230	xfs_filblks_t		*count)
 231{
 232	struct xfs_iext_cursor	icur;
 233	struct xfs_bmbt_irec	got;
 234	xfs_extnum_t		numrecs = 0;
 235
 236	for_each_xfs_iext(ifp, &icur, &got) {
 237		if (!isnullstartblock(got.br_startblock)) {
 238			*count += got.br_blockcount;
 239			numrecs++;
 240		}
 241	}
 242
 243	return numrecs;
 244}
 245
 246/*
 247 * Count leaf blocks given a range of extent records originally
 248 * in btree format.
 249 */
 250STATIC void
 251xfs_bmap_disk_count_leaves(
 252	struct xfs_mount	*mp,
 253	struct xfs_btree_block	*block,
 254	int			numrecs,
 255	xfs_filblks_t		*count)
 256{
 257	int		b;
 258	xfs_bmbt_rec_t	*frp;
 259
 260	for (b = 1; b <= numrecs; b++) {
 261		frp = XFS_BMBT_REC_ADDR(mp, block, b);
 262		*count += xfs_bmbt_disk_get_blockcount(frp);
 263	}
 264}
 265
 266/*
 267 * Recursively walks each level of a btree
 268 * to count total fsblocks in use.
 269 */
 270STATIC int
 271xfs_bmap_count_tree(
 272	struct xfs_mount	*mp,
 273	struct xfs_trans	*tp,
 274	struct xfs_ifork	*ifp,
 275	xfs_fsblock_t		blockno,
 276	int			levelin,
 277	xfs_extnum_t		*nextents,
 278	xfs_filblks_t		*count)
 279{
 280	int			error;
 281	struct xfs_buf		*bp, *nbp;
 282	int			level = levelin;
 283	__be64			*pp;
 284	xfs_fsblock_t           bno = blockno;
 285	xfs_fsblock_t		nextbno;
 286	struct xfs_btree_block	*block, *nextblock;
 287	int			numrecs;
 288
 289	error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF,
 290						&xfs_bmbt_buf_ops);
 291	if (error)
 292		return error;
 293	*count += 1;
 294	block = XFS_BUF_TO_BLOCK(bp);
 295
 296	if (--level) {
 297		/* Not at node above leaves, count this level of nodes */
 298		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
 299		while (nextbno != NULLFSBLOCK) {
 300			error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp,
 301						XFS_BMAP_BTREE_REF,
 302						&xfs_bmbt_buf_ops);
 303			if (error)
 304				return error;
 305			*count += 1;
 306			nextblock = XFS_BUF_TO_BLOCK(nbp);
 307			nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
 308			xfs_trans_brelse(tp, nbp);
 309		}
 310
 311		/* Dive to the next level */
 312		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
 313		bno = be64_to_cpu(*pp);
 314		error = xfs_bmap_count_tree(mp, tp, ifp, bno, level, nextents,
 315				count);
 316		if (error) {
 317			xfs_trans_brelse(tp, bp);
 318			XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
 319					 XFS_ERRLEVEL_LOW, mp);
 320			return -EFSCORRUPTED;
 321		}
 322		xfs_trans_brelse(tp, bp);
 323	} else {
 324		/* count all level 1 nodes and their leaves */
 325		for (;;) {
 326			nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
 327			numrecs = be16_to_cpu(block->bb_numrecs);
 328			(*nextents) += numrecs;
 329			xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
 330			xfs_trans_brelse(tp, bp);
 331			if (nextbno == NULLFSBLOCK)
 332				break;
 333			bno = nextbno;
 334			error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
 335						XFS_BMAP_BTREE_REF,
 336						&xfs_bmbt_buf_ops);
 337			if (error)
 338				return error;
 339			*count += 1;
 340			block = XFS_BUF_TO_BLOCK(bp);
 341		}
 342	}
 343	return 0;
 344}
 345
 346/*
 347 * Count fsblocks of the given fork.  Delayed allocation extents are
 348 * not counted towards the totals.
 349 */
 350int
 351xfs_bmap_count_blocks(
 352	struct xfs_trans	*tp,
 353	struct xfs_inode	*ip,
 354	int			whichfork,
 355	xfs_extnum_t		*nextents,
 356	xfs_filblks_t		*count)
 357{
 358	struct xfs_mount	*mp;	/* file system mount structure */
 359	__be64			*pp;	/* pointer to block address */
 360	struct xfs_btree_block	*block;	/* current btree block */
 361	struct xfs_ifork	*ifp;	/* fork structure */
 362	xfs_fsblock_t		bno;	/* block # of "block" */
 363	int			level;	/* btree level, for checking */
 364	int			error;
 365
 366	bno = NULLFSBLOCK;
 367	mp = ip->i_mount;
 368	*nextents = 0;
 369	*count = 0;
 370	ifp = XFS_IFORK_PTR(ip, whichfork);
 371	if (!ifp)
 372		return 0;
 373
 374	switch (XFS_IFORK_FORMAT(ip, whichfork)) {
 375	case XFS_DINODE_FMT_EXTENTS:
 376		*nextents = xfs_bmap_count_leaves(ifp, count);
 377		return 0;
 378	case XFS_DINODE_FMT_BTREE:
 379		if (!(ifp->if_flags & XFS_IFEXTENTS)) {
 380			error = xfs_iread_extents(tp, ip, whichfork);
 381			if (error)
 382				return error;
 383		}
 
 
 
 
 384
 385		/*
 386		 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
 
 
 387		 */
 388		block = ifp->if_broot;
 389		level = be16_to_cpu(block->bb_level);
 390		ASSERT(level > 0);
 391		pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
 392		bno = be64_to_cpu(*pp);
 393		ASSERT(bno != NULLFSBLOCK);
 394		ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
 395		ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
 396
 397		error = xfs_bmap_count_tree(mp, tp, ifp, bno, level,
 398				nextents, count);
 399		if (error) {
 400			XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)",
 401					XFS_ERRLEVEL_LOW, mp);
 402			return -EFSCORRUPTED;
 403		}
 404		return 0;
 405	}
 406
 407	return 0;
 408}
 409
 410static int
 411xfs_getbmap_report_one(
 412	struct xfs_inode	*ip,
 413	struct getbmapx		*bmv,
 414	struct kgetbmap		*out,
 415	int64_t			bmv_end,
 416	struct xfs_bmbt_irec	*got)
 417{
 418	struct kgetbmap		*p = out + bmv->bmv_entries;
 419	bool			shared = false, trimmed = false;
 420	int			error;
 421
 422	error = xfs_reflink_trim_around_shared(ip, got, &shared, &trimmed);
 423	if (error)
 424		return error;
 425
 426	if (isnullstartblock(got->br_startblock) ||
 427	    got->br_startblock == DELAYSTARTBLOCK) {
 428		/*
 429		 * Delalloc extents that start beyond EOF can occur due to
 430		 * speculative EOF allocation when the delalloc extent is larger
 431		 * than the largest freespace extent at conversion time.  These
 432		 * extents cannot be converted by data writeback, so can exist
 433		 * here even if we are not supposed to be finding delalloc
 434		 * extents.
 435		 */
 436		if (got->br_startoff < XFS_B_TO_FSB(ip->i_mount, XFS_ISIZE(ip)))
 437			ASSERT((bmv->bmv_iflags & BMV_IF_DELALLOC) != 0);
 438
 439		p->bmv_oflags |= BMV_OF_DELALLOC;
 440		p->bmv_block = -2;
 441	} else {
 442		p->bmv_block = xfs_fsb_to_db(ip, got->br_startblock);
 443	}
 444
 445	if (got->br_state == XFS_EXT_UNWRITTEN &&
 446	    (bmv->bmv_iflags & BMV_IF_PREALLOC))
 447		p->bmv_oflags |= BMV_OF_PREALLOC;
 448
 449	if (shared)
 450		p->bmv_oflags |= BMV_OF_SHARED;
 451
 452	p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, got->br_startoff);
 453	p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, got->br_blockcount);
 454
 455	bmv->bmv_offset = p->bmv_offset + p->bmv_length;
 456	bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset);
 457	bmv->bmv_entries++;
 458	return 0;
 459}
 460
 461static void
 462xfs_getbmap_report_hole(
 463	struct xfs_inode	*ip,
 464	struct getbmapx		*bmv,
 465	struct kgetbmap		*out,
 466	int64_t			bmv_end,
 467	xfs_fileoff_t		bno,
 468	xfs_fileoff_t		end)
 469{
 470	struct kgetbmap		*p = out + bmv->bmv_entries;
 471
 472	if (bmv->bmv_iflags & BMV_IF_NO_HOLES)
 473		return;
 474
 475	p->bmv_block = -1;
 476	p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, bno);
 477	p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, end - bno);
 478
 479	bmv->bmv_offset = p->bmv_offset + p->bmv_length;
 480	bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset);
 481	bmv->bmv_entries++;
 482}
 483
 484static inline bool
 485xfs_getbmap_full(
 486	struct getbmapx		*bmv)
 487{
 488	return bmv->bmv_length == 0 || bmv->bmv_entries >= bmv->bmv_count - 1;
 489}
 490
 491static bool
 492xfs_getbmap_next_rec(
 493	struct xfs_bmbt_irec	*rec,
 494	xfs_fileoff_t		total_end)
 495{
 496	xfs_fileoff_t		end = rec->br_startoff + rec->br_blockcount;
 497
 498	if (end == total_end)
 499		return false;
 500
 501	rec->br_startoff += rec->br_blockcount;
 502	if (!isnullstartblock(rec->br_startblock) &&
 503	    rec->br_startblock != DELAYSTARTBLOCK)
 504		rec->br_startblock += rec->br_blockcount;
 505	rec->br_blockcount = total_end - end;
 506	return true;
 507}
 508
 509/*
 510 * Get inode's extents as described in bmv, and format for output.
 511 * Calls formatter to fill the user's buffer until all extents
 512 * are mapped, until the passed-in bmv->bmv_count slots have
 513 * been filled, or until the formatter short-circuits the loop,
 514 * if it is tracking filled-in extents on its own.
 515 */
 516int						/* error code */
 517xfs_getbmap(
 518	struct xfs_inode	*ip,
 519	struct getbmapx		*bmv,		/* user bmap structure */
 520	struct kgetbmap		*out)
 521{
 522	struct xfs_mount	*mp = ip->i_mount;
 523	int			iflags = bmv->bmv_iflags;
 524	int			whichfork, lock, error = 0;
 525	int64_t			bmv_end, max_len;
 526	xfs_fileoff_t		bno, first_bno;
 527	struct xfs_ifork	*ifp;
 528	struct xfs_bmbt_irec	got, rec;
 529	xfs_filblks_t		len;
 530	struct xfs_iext_cursor	icur;
 531
 532	if (bmv->bmv_iflags & ~BMV_IF_VALID)
 533		return -EINVAL;
 534#ifndef DEBUG
 535	/* Only allow CoW fork queries if we're debugging. */
 536	if (iflags & BMV_IF_COWFORK)
 537		return -EINVAL;
 538#endif
 539	if ((iflags & BMV_IF_ATTRFORK) && (iflags & BMV_IF_COWFORK))
 540		return -EINVAL;
 541
 542	if (bmv->bmv_length < -1)
 543		return -EINVAL;
 544	bmv->bmv_entries = 0;
 545	if (bmv->bmv_length == 0)
 546		return 0;
 547
 548	if (iflags & BMV_IF_ATTRFORK)
 549		whichfork = XFS_ATTR_FORK;
 550	else if (iflags & BMV_IF_COWFORK)
 551		whichfork = XFS_COW_FORK;
 552	else
 553		whichfork = XFS_DATA_FORK;
 554	ifp = XFS_IFORK_PTR(ip, whichfork);
 555
 556	xfs_ilock(ip, XFS_IOLOCK_SHARED);
 557	switch (whichfork) {
 558	case XFS_ATTR_FORK:
 559		if (!XFS_IFORK_Q(ip))
 560			goto out_unlock_iolock;
 
 561
 562		max_len = 1LL << 32;
 563		lock = xfs_ilock_attr_map_shared(ip);
 564		break;
 565	case XFS_COW_FORK:
 
 
 
 566		/* No CoW fork? Just return */
 567		if (!ifp)
 568			goto out_unlock_iolock;
 569
 570		if (xfs_get_cowextsz_hint(ip))
 571			max_len = mp->m_super->s_maxbytes;
 572		else
 573			max_len = XFS_ISIZE(ip);
 574
 575		lock = XFS_ILOCK_SHARED;
 576		xfs_ilock(ip, lock);
 577		break;
 578	case XFS_DATA_FORK:
 579		if (!(iflags & BMV_IF_DELALLOC) &&
 580		    (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) {
 581			error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
 582			if (error)
 583				goto out_unlock_iolock;
 584
 585			/*
 586			 * Even after flushing the inode, there can still be
 587			 * delalloc blocks on the inode beyond EOF due to
 588			 * speculative preallocation.  These are not removed
 589			 * until the release function is called or the inode
 590			 * is inactivated.  Hence we cannot assert here that
 591			 * ip->i_delayed_blks == 0.
 592			 */
 593		}
 594
 595		if (xfs_get_extsz_hint(ip) ||
 596		    (ip->i_d.di_flags &
 597		     (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))
 598			max_len = mp->m_super->s_maxbytes;
 599		else
 600			max_len = XFS_ISIZE(ip);
 601
 602		lock = xfs_ilock_data_map_shared(ip);
 603		break;
 604	}
 605
 606	switch (XFS_IFORK_FORMAT(ip, whichfork)) {
 
 
 607	case XFS_DINODE_FMT_EXTENTS:
 608	case XFS_DINODE_FMT_BTREE:
 609		break;
 610	case XFS_DINODE_FMT_LOCAL:
 611		/* Local format inode forks report no extents. */
 612		goto out_unlock_ilock;
 613	default:
 614		error = -EINVAL;
 615		goto out_unlock_ilock;
 616	}
 617
 618	if (bmv->bmv_length == -1) {
 619		max_len = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, max_len));
 620		bmv->bmv_length = max(0LL, max_len - bmv->bmv_offset);
 621	}
 622
 623	bmv_end = bmv->bmv_offset + bmv->bmv_length;
 624
 625	first_bno = bno = XFS_BB_TO_FSBT(mp, bmv->bmv_offset);
 626	len = XFS_BB_TO_FSB(mp, bmv->bmv_length);
 627
 628	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
 629		error = xfs_iread_extents(NULL, ip, whichfork);
 630		if (error)
 631			goto out_unlock_ilock;
 632	}
 633
 634	if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
 635		/*
 636		 * Report a whole-file hole if the delalloc flag is set to
 637		 * stay compatible with the old implementation.
 638		 */
 639		if (iflags & BMV_IF_DELALLOC)
 640			xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
 641					XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
 642		goto out_unlock_ilock;
 643	}
 644
 645	while (!xfs_getbmap_full(bmv)) {
 646		xfs_trim_extent(&got, first_bno, len);
 647
 648		/*
 649		 * Report an entry for a hole if this extent doesn't directly
 650		 * follow the previous one.
 651		 */
 652		if (got.br_startoff > bno) {
 653			xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
 654					got.br_startoff);
 655			if (xfs_getbmap_full(bmv))
 656				break;
 657		}
 658
 659		/*
 660		 * In order to report shared extents accurately, we report each
 661		 * distinct shared / unshared part of a single bmbt record with
 662		 * an individual getbmapx record.
 663		 */
 664		bno = got.br_startoff + got.br_blockcount;
 665		rec = got;
 666		do {
 667			error = xfs_getbmap_report_one(ip, bmv, out, bmv_end,
 668					&rec);
 669			if (error || xfs_getbmap_full(bmv))
 670				goto out_unlock_ilock;
 671		} while (xfs_getbmap_next_rec(&rec, bno));
 672
 673		if (!xfs_iext_next_extent(ifp, &icur, &got)) {
 674			xfs_fileoff_t	end = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
 675
 676			out[bmv->bmv_entries - 1].bmv_oflags |= BMV_OF_LAST;
 
 
 677
 678			if (whichfork != XFS_ATTR_FORK && bno < end &&
 679			    !xfs_getbmap_full(bmv)) {
 680				xfs_getbmap_report_hole(ip, bmv, out, bmv_end,
 681						bno, end);
 682			}
 683			break;
 684		}
 685
 686		if (bno >= first_bno + len)
 687			break;
 688	}
 689
 690out_unlock_ilock:
 691	xfs_iunlock(ip, lock);
 692out_unlock_iolock:
 693	xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 694	return error;
 695}
 696
 697/*
 698 * dead simple method of punching delalyed allocation blocks from a range in
 699 * the inode. Walks a block at a time so will be slow, but is only executed in
 700 * rare error cases so the overhead is not critical. This will always punch out
 701 * both the start and end blocks, even if the ranges only partially overlap
 702 * them, so it is up to the caller to ensure that partial blocks are not
 703 * passed in.
 704 */
 705int
 706xfs_bmap_punch_delalloc_range(
 707	struct xfs_inode	*ip,
 708	xfs_fileoff_t		start_fsb,
 709	xfs_fileoff_t		length)
 710{
 711	xfs_fileoff_t		remaining = length;
 
 
 
 
 
 712	int			error = 0;
 713
 714	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 715
 716	do {
 717		int		done;
 718		xfs_bmbt_irec_t	imap;
 719		int		nimaps = 1;
 720		xfs_fsblock_t	firstblock;
 721		struct xfs_defer_ops dfops;
 
 722
 723		/*
 724		 * Map the range first and check that it is a delalloc extent
 725		 * before trying to unmap the range. Otherwise we will be
 726		 * trying to remove a real extent (which requires a
 727		 * transaction) or a hole, which is probably a bad idea...
 728		 */
 729		error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
 730				       XFS_BMAPI_ENTIRE);
 731
 732		if (error) {
 733			/* something screwed, just bail */
 734			if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
 735				xfs_alert(ip->i_mount,
 736			"Failed delalloc mapping lookup ino %lld fsb %lld.",
 737						ip->i_ino, start_fsb);
 738			}
 739			break;
 740		}
 741		if (!nimaps) {
 742			/* nothing there */
 743			goto next_block;
 744		}
 745		if (imap.br_startblock != DELAYSTARTBLOCK) {
 746			/* been converted, ignore */
 747			goto next_block;
 748		}
 749		WARN_ON(imap.br_blockcount == 0);
 750
 751		/*
 752		 * Note: while we initialise the firstblock/dfops pair, they
 753		 * should never be used because blocks should never be
 754		 * allocated or freed for a delalloc extent and hence we need
 755		 * don't cancel or finish them after the xfs_bunmapi() call.
 756		 */
 757		xfs_defer_init(&dfops, &firstblock);
 758		error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
 759					&dfops, &done);
 760		if (error)
 761			break;
 
 762
 763		ASSERT(!xfs_defer_has_unfinished_work(&dfops));
 764next_block:
 765		start_fsb++;
 766		remaining--;
 767	} while(remaining > 0);
 768
 769	return error;
 770}
 771
 772/*
 773 * Test whether it is appropriate to check an inode for and free post EOF
 774 * blocks. The 'force' parameter determines whether we should also consider
 775 * regular files that are marked preallocated or append-only.
 776 */
 777bool
 778xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
 
 
 779{
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 780	/* prealloc/delalloc exists only on regular files */
 781	if (!S_ISREG(VFS_I(ip)->i_mode))
 782		return false;
 783
 784	/*
 785	 * Zero sized files with no cached pages and delalloc blocks will not
 786	 * have speculative prealloc/delalloc blocks to remove.
 787	 */
 788	if (VFS_I(ip)->i_size == 0 &&
 789	    VFS_I(ip)->i_mapping->nrpages == 0 &&
 790	    ip->i_delayed_blks == 0)
 791		return false;
 792
 793	/* If we haven't read in the extent list, then don't do it now. */
 794	if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
 795		return false;
 796
 797	/*
 798	 * Do not free real preallocated or append-only files unless the file
 799	 * has delalloc blocks and we are forced to remove them.
 800	 */
 801	if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
 802		if (!force || ip->i_delayed_blks == 0)
 803			return false;
 804
 805	return true;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 806}
 807
 808/*
 809 * This is called to free any blocks beyond eof. The caller must hold
 810 * IOLOCK_EXCL unless we are in the inode reclaim path and have the only
 811 * reference to the inode.
 812 */
 813int
 814xfs_free_eofblocks(
 815	struct xfs_inode	*ip)
 816{
 817	struct xfs_trans	*tp;
 818	int			error;
 819	xfs_fileoff_t		end_fsb;
 820	xfs_fileoff_t		last_fsb;
 821	xfs_filblks_t		map_len;
 822	int			nimaps;
 823	struct xfs_bmbt_irec	imap;
 824	struct xfs_mount	*mp = ip->i_mount;
 
 825
 826	/*
 827	 * Figure out if there are any blocks beyond the end
 828	 * of the file.  If not, then there is nothing to do.
 829	 */
 830	end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
 831	last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
 832	if (last_fsb <= end_fsb)
 833		return 0;
 834	map_len = last_fsb - end_fsb;
 835
 836	nimaps = 1;
 837	xfs_ilock(ip, XFS_ILOCK_SHARED);
 838	error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
 839	xfs_iunlock(ip, XFS_ILOCK_SHARED);
 840
 841	/*
 842	 * If there are blocks after the end of file, truncate the file to its
 843	 * current size to free them up.
 844	 */
 845	if (!error && (nimaps != 0) &&
 846	    (imap.br_startblock != HOLESTARTBLOCK ||
 847	     ip->i_delayed_blks)) {
 848		/*
 849		 * Attach the dquots to the inode up front.
 850		 */
 851		error = xfs_qm_dqattach(ip, 0);
 852		if (error)
 853			return error;
 854
 855		/* wait on dio to ensure i_size has settled */
 856		inode_dio_wait(VFS_I(ip));
 857
 858		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0,
 859				&tp);
 860		if (error) {
 861			ASSERT(XFS_FORCED_SHUTDOWN(mp));
 862			return error;
 863		}
 
 
 
 
 864
 865		xfs_ilock(ip, XFS_ILOCK_EXCL);
 866		xfs_trans_ijoin(tp, ip, 0);
 
 867
 868		/*
 869		 * Do not update the on-disk file size.  If we update the
 870		 * on-disk file size and then the system crashes before the
 871		 * contents of the file are flushed to disk then the files
 872		 * may be full of holes (ie NULL files bug).
 873		 */
 874		error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK,
 875					      XFS_ISIZE(ip));
 876		if (error) {
 877			/*
 878			 * If we get an error at this point we simply don't
 879			 * bother truncating the file.
 880			 */
 881			xfs_trans_cancel(tp);
 882		} else {
 883			error = xfs_trans_commit(tp);
 884			if (!error)
 885				xfs_inode_clear_eofblocks_tag(ip);
 886		}
 887
 888		xfs_iunlock(ip, XFS_ILOCK_EXCL);
 889	}
 
 
 
 
 
 
 890	return error;
 891}
 892
 893int
 894xfs_alloc_file_space(
 895	struct xfs_inode	*ip,
 896	xfs_off_t		offset,
 897	xfs_off_t		len,
 898	int			alloc_type)
 899{
 900	xfs_mount_t		*mp = ip->i_mount;
 901	xfs_off_t		count;
 902	xfs_filblks_t		allocated_fsb;
 903	xfs_filblks_t		allocatesize_fsb;
 904	xfs_extlen_t		extsz, temp;
 905	xfs_fileoff_t		startoffset_fsb;
 906	xfs_fsblock_t		firstfsb;
 907	int			nimaps;
 908	int			quota_flag;
 909	int			rt;
 910	xfs_trans_t		*tp;
 911	xfs_bmbt_irec_t		imaps[1], *imapp;
 912	struct xfs_defer_ops	dfops;
 913	uint			qblocks, resblks, resrtextents;
 914	int			error;
 915
 916	trace_xfs_alloc_file_space(ip);
 917
 918	if (XFS_FORCED_SHUTDOWN(mp))
 919		return -EIO;
 920
 921	error = xfs_qm_dqattach(ip, 0);
 922	if (error)
 923		return error;
 924
 925	if (len <= 0)
 926		return -EINVAL;
 927
 928	rt = XFS_IS_REALTIME_INODE(ip);
 929	extsz = xfs_get_extsz_hint(ip);
 930
 931	count = len;
 932	imapp = &imaps[0];
 933	nimaps = 1;
 934	startoffset_fsb	= XFS_B_TO_FSBT(mp, offset);
 935	allocatesize_fsb = XFS_B_TO_FSB(mp, count);
 
 936
 937	/*
 938	 * Allocate file space until done or until there is an error
 939	 */
 940	while (allocatesize_fsb && !error) {
 941		xfs_fileoff_t	s, e;
 
 
 942
 943		/*
 944		 * Determine space reservations for data/realtime.
 945		 */
 946		if (unlikely(extsz)) {
 947			s = startoffset_fsb;
 948			do_div(s, extsz);
 949			s *= extsz;
 950			e = startoffset_fsb + allocatesize_fsb;
 951			if ((temp = do_mod(startoffset_fsb, extsz)))
 
 952				e += temp;
 953			if ((temp = do_mod(e, extsz)))
 
 954				e += extsz - temp;
 955		} else {
 956			s = 0;
 957			e = allocatesize_fsb;
 958		}
 959
 960		/*
 961		 * The transaction reservation is limited to a 32-bit block
 962		 * count, hence we need to limit the number of blocks we are
 963		 * trying to reserve to avoid an overflow. We can't allocate
 964		 * more than @nimaps extents, and an extent is limited on disk
 965		 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
 
 966		 */
 967		resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
 
 968		if (unlikely(rt)) {
 969			resrtextents = qblocks = resblks;
 970			resrtextents /= mp->m_sb.sb_rextsize;
 971			resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
 972			quota_flag = XFS_QMOPT_RES_RTBLKS;
 973		} else {
 974			resrtextents = 0;
 975			resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
 976			quota_flag = XFS_QMOPT_RES_REGBLKS;
 977		}
 978
 979		/*
 980		 * Allocate and setup the transaction.
 981		 */
 982		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks,
 983				resrtextents, 0, &tp);
 984
 985		/*
 986		 * Check for running out of space
 987		 */
 988		if (error) {
 989			/*
 990			 * Free the transaction structure.
 991			 */
 992			ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
 993			break;
 994		}
 995		xfs_ilock(ip, XFS_ILOCK_EXCL);
 996		error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
 997						      0, quota_flag);
 998		if (error)
 999			goto error1;
1000
1001		xfs_trans_ijoin(tp, ip, 0);
 
 
 
 
 
 
1002
1003		xfs_defer_init(&dfops, &firstfsb);
1004		error = xfs_bmapi_write(tp, ip, startoffset_fsb,
1005					allocatesize_fsb, alloc_type, &firstfsb,
1006					resblks, imapp, &nimaps, &dfops);
1007		if (error)
1008			goto error0;
1009
1010		/*
1011		 * Complete the transaction
1012		 */
1013		error = xfs_defer_finish(&tp, &dfops);
1014		if (error)
1015			goto error0;
1016
1017		error = xfs_trans_commit(tp);
1018		xfs_iunlock(ip, XFS_ILOCK_EXCL);
1019		if (error)
1020			break;
1021
1022		allocated_fsb = imapp->br_blockcount;
1023
1024		if (nimaps == 0) {
1025			error = -ENOSPC;
1026			break;
 
 
 
 
 
 
 
1027		}
1028
1029		startoffset_fsb += allocated_fsb;
1030		allocatesize_fsb -= allocated_fsb;
1031	}
1032
1033	return error;
1034
1035error0:	/* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1036	xfs_defer_cancel(&dfops);
1037	xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
1038
1039error1:	/* Just cancel transaction */
1040	xfs_trans_cancel(tp);
1041	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1042	return error;
1043}
1044
1045static int
1046xfs_unmap_extent(
1047	struct xfs_inode	*ip,
1048	xfs_fileoff_t		startoffset_fsb,
1049	xfs_filblks_t		len_fsb,
1050	int			*done)
1051{
1052	struct xfs_mount	*mp = ip->i_mount;
1053	struct xfs_trans	*tp;
1054	struct xfs_defer_ops	dfops;
1055	xfs_fsblock_t		firstfsb;
1056	uint			resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1057	int			error;
1058
1059	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
1060	if (error) {
1061		ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1062		return error;
1063	}
1064
1065	xfs_ilock(ip, XFS_ILOCK_EXCL);
1066	error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot, ip->i_gdquot,
1067			ip->i_pdquot, resblks, 0, XFS_QMOPT_RES_REGBLKS);
 
1068	if (error)
1069		goto out_trans_cancel;
1070
1071	xfs_trans_ijoin(tp, ip, 0);
1072
1073	xfs_defer_init(&dfops, &firstfsb);
1074	error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, &firstfsb,
1075			&dfops, done);
1076	if (error)
1077		goto out_bmap_cancel;
1078
1079	xfs_defer_ijoin(&dfops, ip);
1080	error = xfs_defer_finish(&tp, &dfops);
1081	if (error)
1082		goto out_bmap_cancel;
1083
1084	error = xfs_trans_commit(tp);
1085out_unlock:
1086	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1087	return error;
1088
1089out_bmap_cancel:
1090	xfs_defer_cancel(&dfops);
1091out_trans_cancel:
1092	xfs_trans_cancel(tp);
1093	goto out_unlock;
1094}
1095
1096static int
1097xfs_adjust_extent_unmap_boundaries(
1098	struct xfs_inode	*ip,
1099	xfs_fileoff_t		*startoffset_fsb,
1100	xfs_fileoff_t		*endoffset_fsb)
1101{
1102	struct xfs_mount	*mp = ip->i_mount;
1103	struct xfs_bmbt_irec	imap;
1104	int			nimap, error;
1105	xfs_extlen_t		mod = 0;
1106
1107	nimap = 1;
1108	error = xfs_bmapi_read(ip, *startoffset_fsb, 1, &imap, &nimap, 0);
1109	if (error)
1110		return error;
1111
1112	if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1113		ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1114		mod = do_mod(imap.br_startblock, mp->m_sb.sb_rextsize);
1115		if (mod)
1116			*startoffset_fsb += mp->m_sb.sb_rextsize - mod;
1117	}
1118
1119	nimap = 1;
1120	error = xfs_bmapi_read(ip, *endoffset_fsb - 1, 1, &imap, &nimap, 0);
1121	if (error)
1122		return error;
1123
1124	if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1125		ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1126		mod++;
1127		if (mod && mod != mp->m_sb.sb_rextsize)
1128			*endoffset_fsb -= mod;
1129	}
1130
1131	return 0;
1132}
1133
1134static int
1135xfs_flush_unmap_range(
1136	struct xfs_inode	*ip,
1137	xfs_off_t		offset,
1138	xfs_off_t		len)
1139{
1140	struct xfs_mount	*mp = ip->i_mount;
1141	struct inode		*inode = VFS_I(ip);
1142	xfs_off_t		rounding, start, end;
1143	int			error;
1144
1145	/* wait for the completion of any pending DIOs */
1146	inode_dio_wait(inode);
1147
1148	rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_SIZE);
1149	start = round_down(offset, rounding);
1150	end = round_up(offset + len, rounding) - 1;
1151
1152	error = filemap_write_and_wait_range(inode->i_mapping, start, end);
1153	if (error)
1154		return error;
1155	truncate_pagecache_range(inode, start, end);
1156	return 0;
1157}
1158
1159int
1160xfs_free_file_space(
1161	struct xfs_inode	*ip,
1162	xfs_off_t		offset,
1163	xfs_off_t		len)
1164{
1165	struct xfs_mount	*mp = ip->i_mount;
1166	xfs_fileoff_t		startoffset_fsb;
1167	xfs_fileoff_t		endoffset_fsb;
1168	int			done = 0, error;
1169
1170	trace_xfs_free_file_space(ip);
1171
1172	error = xfs_qm_dqattach(ip, 0);
1173	if (error)
1174		return error;
1175
1176	if (len <= 0)	/* if nothing being freed */
1177		return 0;
1178
1179	error = xfs_flush_unmap_range(ip, offset, len);
1180	if (error)
1181		return error;
1182
1183	startoffset_fsb = XFS_B_TO_FSB(mp, offset);
1184	endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
1185
1186	/*
1187	 * Need to zero the stuff we're not freeing, on disk.  If it's a RT file
1188	 * and we can't use unwritten extents then we actually need to ensure
1189	 * to zero the whole extent, otherwise we just need to take of block
1190	 * boundaries, and xfs_bunmapi will handle the rest.
1191	 */
1192	if (XFS_IS_REALTIME_INODE(ip) &&
1193	    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
1194		error = xfs_adjust_extent_unmap_boundaries(ip, &startoffset_fsb,
1195				&endoffset_fsb);
1196		if (error)
1197			return error;
1198	}
1199
 
 
 
1200	if (endoffset_fsb > startoffset_fsb) {
1201		while (!done) {
1202			error = xfs_unmap_extent(ip, startoffset_fsb,
1203					endoffset_fsb - startoffset_fsb, &done);
1204			if (error)
1205				return error;
1206		}
1207	}
1208
1209	/*
1210	 * Now that we've unmap all full blocks we'll have to zero out any
1211	 * partial block at the beginning and/or end.  iomap_zero_range is smart
1212	 * enough to skip any holes, including those we just created, but we
1213	 * must take care not to zero beyond EOF and enlarge i_size.
1214	 */
1215	if (offset >= XFS_ISIZE(ip))
1216		return 0;
1217	if (offset + len > XFS_ISIZE(ip))
1218		len = XFS_ISIZE(ip) - offset;
1219	return iomap_zero_range(VFS_I(ip), offset, len, NULL, &xfs_iomap_ops);
1220}
1221
1222/*
1223 * Preallocate and zero a range of a file. This mechanism has the allocation
1224 * semantics of fallocate and in addition converts data in the range to zeroes.
1225 */
1226int
1227xfs_zero_file_space(
1228	struct xfs_inode	*ip,
1229	xfs_off_t		offset,
1230	xfs_off_t		len)
1231{
1232	struct xfs_mount	*mp = ip->i_mount;
1233	uint			blksize;
1234	int			error;
1235
1236	trace_xfs_zero_file_space(ip);
1237
1238	blksize = 1 << mp->m_sb.sb_blocklog;
1239
1240	/*
1241	 * Punch a hole and prealloc the range. We use hole punch rather than
1242	 * unwritten extent conversion for two reasons:
1243	 *
1244	 * 1.) Hole punch handles partial block zeroing for us.
1245	 *
1246	 * 2.) If prealloc returns ENOSPC, the file range is still zero-valued
1247	 * by virtue of the hole punch.
1248	 */
1249	error = xfs_free_file_space(ip, offset, len);
1250	if (error)
1251		goto out;
1252
1253	error = xfs_alloc_file_space(ip, round_down(offset, blksize),
1254				     round_up(offset + len, blksize) -
1255				     round_down(offset, blksize),
1256				     XFS_BMAPI_PREALLOC);
1257out:
1258	return error;
1259
1260}
1261
1262static int
1263xfs_prepare_shift(
1264	struct xfs_inode	*ip,
1265	loff_t			offset)
1266{
 
1267	int			error;
1268
1269	/*
1270	 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
1271	 * into the accessible region of the file.
1272	 */
1273	if (xfs_can_free_eofblocks(ip, true)) {
1274		error = xfs_free_eofblocks(ip);
1275		if (error)
1276			return error;
1277	}
1278
1279	/*
 
 
 
 
 
 
 
 
 
 
 
1280	 * Writeback and invalidate cache for the remainder of the file as we're
1281	 * about to shift down every extent from offset to EOF.
1282	 */
1283	error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, offset, -1);
1284	if (error)
1285		return error;
1286	error = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
1287					offset >> PAGE_SHIFT, -1);
1288	if (error)
1289		return error;
1290
1291	/*
1292	 * Clean out anything hanging around in the cow fork now that
1293	 * we've flushed all the dirty data out to disk to avoid having
1294	 * CoW extents at the wrong offsets.
1295	 */
1296	if (xfs_is_reflink_inode(ip)) {
1297		error = xfs_reflink_cancel_cow_range(ip, offset, NULLFILEOFF,
1298				true);
1299		if (error)
1300			return error;
1301	}
1302
1303	return 0;
1304}
1305
1306/*
1307 * xfs_collapse_file_space()
1308 *	This routine frees disk space and shift extent for the given file.
1309 *	The first thing we do is to free data blocks in the specified range
1310 *	by calling xfs_free_file_space(). It would also sync dirty data
1311 *	and invalidate page cache over the region on which collapse range
1312 *	is working. And Shift extent records to the left to cover a hole.
1313 * RETURNS:
1314 *	0 on success
1315 *	errno on error
1316 *
1317 */
1318int
1319xfs_collapse_file_space(
1320	struct xfs_inode	*ip,
1321	xfs_off_t		offset,
1322	xfs_off_t		len)
1323{
1324	struct xfs_mount	*mp = ip->i_mount;
1325	struct xfs_trans	*tp;
1326	int			error;
1327	struct xfs_defer_ops	dfops;
1328	xfs_fsblock_t		first_block;
1329	xfs_fileoff_t		next_fsb = XFS_B_TO_FSB(mp, offset + len);
1330	xfs_fileoff_t		shift_fsb = XFS_B_TO_FSB(mp, len);
1331	uint			resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1332	bool			done = false;
1333
1334	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1335	ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
1336
1337	trace_xfs_collapse_file_space(ip);
1338
1339	error = xfs_free_file_space(ip, offset, len);
1340	if (error)
1341		return error;
1342
1343	error = xfs_prepare_shift(ip, offset);
1344	if (error)
1345		return error;
1346
1347	while (!error && !done) {
1348		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0,
1349					&tp);
1350		if (error)
1351			break;
1352
1353		xfs_ilock(ip, XFS_ILOCK_EXCL);
1354		error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot,
1355				ip->i_gdquot, ip->i_pdquot, resblks, 0,
1356				XFS_QMOPT_RES_REGBLKS);
1357		if (error)
1358			goto out_trans_cancel;
1359		xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1360
1361		xfs_defer_init(&dfops, &first_block);
1362		error = xfs_bmap_collapse_extents(tp, ip, &next_fsb, shift_fsb,
1363				&done, &first_block, &dfops);
1364		if (error)
1365			goto out_bmap_cancel;
 
 
1366
1367		error = xfs_defer_finish(&tp, &dfops);
 
1368		if (error)
1369			goto out_bmap_cancel;
1370		error = xfs_trans_commit(tp);
1371	}
1372
 
 
1373	return error;
1374
1375out_bmap_cancel:
1376	xfs_defer_cancel(&dfops);
1377out_trans_cancel:
1378	xfs_trans_cancel(tp);
 
1379	return error;
1380}
1381
1382/*
1383 * xfs_insert_file_space()
1384 *	This routine create hole space by shifting extents for the given file.
1385 *	The first thing we do is to sync dirty data and invalidate page cache
1386 *	over the region on which insert range is working. And split an extent
1387 *	to two extents at given offset by calling xfs_bmap_split_extent.
1388 *	And shift all extent records which are laying between [offset,
1389 *	last allocated extent] to the right to reserve hole range.
1390 * RETURNS:
1391 *	0 on success
1392 *	errno on error
1393 */
1394int
1395xfs_insert_file_space(
1396	struct xfs_inode	*ip,
1397	loff_t			offset,
1398	loff_t			len)
1399{
1400	struct xfs_mount	*mp = ip->i_mount;
1401	struct xfs_trans	*tp;
1402	int			error;
1403	struct xfs_defer_ops	dfops;
1404	xfs_fsblock_t		first_block;
1405	xfs_fileoff_t		stop_fsb = XFS_B_TO_FSB(mp, offset);
1406	xfs_fileoff_t		next_fsb = NULLFSBLOCK;
1407	xfs_fileoff_t		shift_fsb = XFS_B_TO_FSB(mp, len);
1408	bool			done = false;
1409
1410	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1411	ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
1412
1413	trace_xfs_insert_file_space(ip);
1414
 
 
 
 
1415	error = xfs_prepare_shift(ip, offset);
1416	if (error)
1417		return error;
1418
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1419	/*
1420	 * The extent shifting code works on extent granularity. So, if stop_fsb
1421	 * is not the starting block of extent, we need to split the extent at
1422	 * stop_fsb.
1423	 */
1424	error = xfs_bmap_split_extent(ip, stop_fsb);
1425	if (error)
1426		return error;
1427
1428	while (!error && !done) {
1429		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0,
1430					&tp);
1431		if (error)
1432			break;
1433
1434		xfs_ilock(ip, XFS_ILOCK_EXCL);
1435		xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1436		xfs_defer_init(&dfops, &first_block);
1437		error = xfs_bmap_insert_extents(tp, ip, &next_fsb, shift_fsb,
1438				&done, stop_fsb, &first_block, &dfops);
1439		if (error)
1440			goto out_bmap_cancel;
1441
1442		error = xfs_defer_finish(&tp, &dfops);
1443		if (error)
1444			goto out_bmap_cancel;
1445		error = xfs_trans_commit(tp);
1446	}
1447
 
 
1448	return error;
1449
1450out_bmap_cancel:
1451	xfs_defer_cancel(&dfops);
1452	xfs_trans_cancel(tp);
 
1453	return error;
1454}
1455
1456/*
1457 * We need to check that the format of the data fork in the temporary inode is
1458 * valid for the target inode before doing the swap. This is not a problem with
1459 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1460 * data fork depending on the space the attribute fork is taking so we can get
1461 * invalid formats on the target inode.
1462 *
1463 * E.g. target has space for 7 extents in extent format, temp inode only has
1464 * space for 6.  If we defragment down to 7 extents, then the tmp format is a
1465 * btree, but when swapped it needs to be in extent format. Hence we can't just
1466 * blindly swap data forks on attr2 filesystems.
1467 *
1468 * Note that we check the swap in both directions so that we don't end up with
1469 * a corrupt temporary inode, either.
1470 *
1471 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1472 * inode will prevent this situation from occurring, so all we do here is
1473 * reject and log the attempt. basically we are putting the responsibility on
1474 * userspace to get this right.
1475 */
1476static int
1477xfs_swap_extents_check_format(
1478	struct xfs_inode	*ip,	/* target inode */
1479	struct xfs_inode	*tip)	/* tmp inode */
1480{
 
 
 
 
 
 
 
 
 
1481
1482	/* Should never get a local format */
1483	if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
1484	    tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
1485		return -EINVAL;
1486
1487	/*
1488	 * if the target inode has less extents that then temporary inode then
1489	 * why did userspace call us?
1490	 */
1491	if (ip->i_d.di_nextents < tip->i_d.di_nextents)
1492		return -EINVAL;
1493
1494	/*
1495	 * If we have to use the (expensive) rmap swap method, we can
1496	 * handle any number of extents and any format.
1497	 */
1498	if (xfs_sb_version_hasrmapbt(&ip->i_mount->m_sb))
1499		return 0;
1500
1501	/*
1502	 * if the target inode is in extent form and the temp inode is in btree
1503	 * form then we will end up with the target inode in the wrong format
1504	 * as we already know there are less extents in the temp inode.
1505	 */
1506	if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1507	    tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1508		return -EINVAL;
1509
1510	/* Check temp in extent form to max in target */
1511	if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1512	    XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) >
1513			XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1514		return -EINVAL;
1515
1516	/* Check target in extent form to max in temp */
1517	if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1518	    XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) >
1519			XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1520		return -EINVAL;
1521
1522	/*
1523	 * If we are in a btree format, check that the temp root block will fit
1524	 * in the target and that it has enough extents to be in btree format
1525	 * in the target.
1526	 *
1527	 * Note that we have to be careful to allow btree->extent conversions
1528	 * (a common defrag case) which will occur when the temp inode is in
1529	 * extent format...
1530	 */
1531	if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1532		if (XFS_IFORK_Q(ip) &&
1533		    XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip))
1534			return -EINVAL;
1535		if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <=
1536		    XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1537			return -EINVAL;
1538	}
1539
1540	/* Reciprocal target->temp btree format checks */
1541	if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1542		if (XFS_IFORK_Q(tip) &&
1543		    XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
1544			return -EINVAL;
1545		if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <=
1546		    XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1547			return -EINVAL;
1548	}
1549
1550	return 0;
1551}
1552
1553static int
1554xfs_swap_extent_flush(
1555	struct xfs_inode	*ip)
1556{
1557	int	error;
1558
1559	error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1560	if (error)
1561		return error;
1562	truncate_pagecache_range(VFS_I(ip), 0, -1);
1563
1564	/* Verify O_DIRECT for ftmp */
1565	if (VFS_I(ip)->i_mapping->nrpages)
1566		return -EINVAL;
1567	return 0;
1568}
1569
1570/*
1571 * Move extents from one file to another, when rmap is enabled.
1572 */
1573STATIC int
1574xfs_swap_extent_rmap(
1575	struct xfs_trans		**tpp,
1576	struct xfs_inode		*ip,
1577	struct xfs_inode		*tip)
1578{
 
1579	struct xfs_bmbt_irec		irec;
1580	struct xfs_bmbt_irec		uirec;
1581	struct xfs_bmbt_irec		tirec;
1582	xfs_fileoff_t			offset_fsb;
1583	xfs_fileoff_t			end_fsb;
1584	xfs_filblks_t			count_fsb;
1585	xfs_fsblock_t			firstfsb;
1586	struct xfs_defer_ops		dfops;
1587	int				error;
1588	xfs_filblks_t			ilen;
1589	xfs_filblks_t			rlen;
1590	int				nimaps;
1591	uint64_t			tip_flags2;
1592
1593	/*
1594	 * If the source file has shared blocks, we must flag the donor
1595	 * file as having shared blocks so that we get the shared-block
1596	 * rmap functions when we go to fix up the rmaps.  The flags
1597	 * will be switch for reals later.
1598	 */
1599	tip_flags2 = tip->i_d.di_flags2;
1600	if (ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)
1601		tip->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK;
1602
1603	offset_fsb = 0;
1604	end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip)));
1605	count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
1606
1607	while (count_fsb) {
1608		/* Read extent from the donor file */
1609		nimaps = 1;
1610		error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec,
1611				&nimaps, 0);
1612		if (error)
1613			goto out;
1614		ASSERT(nimaps == 1);
1615		ASSERT(tirec.br_startblock != DELAYSTARTBLOCK);
1616
1617		trace_xfs_swap_extent_rmap_remap(tip, &tirec);
1618		ilen = tirec.br_blockcount;
1619
1620		/* Unmap the old blocks in the source file. */
1621		while (tirec.br_blockcount) {
1622			xfs_defer_init(&dfops, &firstfsb);
1623			trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec);
1624
1625			/* Read extent from the source file */
1626			nimaps = 1;
1627			error = xfs_bmapi_read(ip, tirec.br_startoff,
1628					tirec.br_blockcount, &irec,
1629					&nimaps, 0);
1630			if (error)
1631				goto out_defer;
1632			ASSERT(nimaps == 1);
1633			ASSERT(tirec.br_startoff == irec.br_startoff);
1634			trace_xfs_swap_extent_rmap_remap_piece(ip, &irec);
1635
1636			/* Trim the extent. */
1637			uirec = tirec;
1638			uirec.br_blockcount = rlen = min_t(xfs_filblks_t,
1639					tirec.br_blockcount,
1640					irec.br_blockcount);
1641			trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec);
1642
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1643			/* Remove the mapping from the donor file. */
1644			error = xfs_bmap_unmap_extent((*tpp)->t_mountp, &dfops,
1645					tip, &uirec);
1646			if (error)
1647				goto out_defer;
1648
1649			/* Remove the mapping from the source file. */
1650			error = xfs_bmap_unmap_extent((*tpp)->t_mountp, &dfops,
1651					ip, &irec);
1652			if (error)
1653				goto out_defer;
1654
1655			/* Map the donor file's blocks into the source file. */
1656			error = xfs_bmap_map_extent((*tpp)->t_mountp, &dfops,
1657					ip, &uirec);
1658			if (error)
1659				goto out_defer;
1660
1661			/* Map the source file's blocks into the donor file. */
1662			error = xfs_bmap_map_extent((*tpp)->t_mountp, &dfops,
1663					tip, &irec);
1664			if (error)
1665				goto out_defer;
1666
1667			xfs_defer_ijoin(&dfops, ip);
1668			error = xfs_defer_finish(tpp, &dfops);
1669			if (error)
1670				goto out_defer;
1671
1672			tirec.br_startoff += rlen;
1673			if (tirec.br_startblock != HOLESTARTBLOCK &&
1674			    tirec.br_startblock != DELAYSTARTBLOCK)
1675				tirec.br_startblock += rlen;
1676			tirec.br_blockcount -= rlen;
1677		}
1678
1679		/* Roll on... */
1680		count_fsb -= ilen;
1681		offset_fsb += ilen;
1682	}
1683
1684	tip->i_d.di_flags2 = tip_flags2;
1685	return 0;
1686
1687out_defer:
1688	xfs_defer_cancel(&dfops);
1689out:
1690	trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_);
1691	tip->i_d.di_flags2 = tip_flags2;
1692	return error;
1693}
1694
1695/* Swap the extents of two files by swapping data forks. */
1696STATIC int
1697xfs_swap_extent_forks(
1698	struct xfs_trans	*tp,
1699	struct xfs_inode	*ip,
1700	struct xfs_inode	*tip,
1701	int			*src_log_flags,
1702	int			*target_log_flags)
1703{
1704	struct xfs_ifork	tempifp, *ifp, *tifp;
1705	xfs_filblks_t		aforkblks = 0;
1706	xfs_filblks_t		taforkblks = 0;
1707	xfs_extnum_t		junk;
1708	uint64_t		tmp;
1709	int			error;
1710
1711	/*
1712	 * Count the number of extended attribute blocks
1713	 */
1714	if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
1715	     (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1716		error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &junk,
1717				&aforkblks);
1718		if (error)
1719			return error;
1720	}
1721	if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
1722	     (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1723		error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK, &junk,
1724				&taforkblks);
1725		if (error)
1726			return error;
1727	}
1728
1729	/*
1730	 * Btree format (v3) inodes have the inode number stamped in the bmbt
1731	 * block headers. We can't start changing the bmbt blocks until the
1732	 * inode owner change is logged so recovery does the right thing in the
1733	 * event of a crash. Set the owner change log flags now and leave the
1734	 * bmbt scan as the last step.
1735	 */
1736	if (ip->i_d.di_version == 3 &&
1737	    ip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1738		(*target_log_flags) |= XFS_ILOG_DOWNER;
1739	if (tip->i_d.di_version == 3 &&
1740	    tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1741		(*src_log_flags) |= XFS_ILOG_DOWNER;
1742
1743	/*
1744	 * Swap the data forks of the inodes
1745	 */
1746	ifp = &ip->i_df;
1747	tifp = &tip->i_df;
1748	tempifp = *ifp;		/* struct copy */
1749	*ifp = *tifp;		/* struct copy */
1750	*tifp = tempifp;	/* struct copy */
1751
1752	/*
1753	 * Fix the on-disk inode values
1754	 */
1755	tmp = (uint64_t)ip->i_d.di_nblocks;
1756	ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
1757	tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
1758
1759	tmp = (uint64_t) ip->i_d.di_nextents;
1760	ip->i_d.di_nextents = tip->i_d.di_nextents;
1761	tip->i_d.di_nextents = tmp;
1762
1763	tmp = (uint64_t) ip->i_d.di_format;
1764	ip->i_d.di_format = tip->i_d.di_format;
1765	tip->i_d.di_format = tmp;
1766
1767	/*
1768	 * The extents in the source inode could still contain speculative
1769	 * preallocation beyond EOF (e.g. the file is open but not modified
1770	 * while defrag is in progress). In that case, we need to copy over the
1771	 * number of delalloc blocks the data fork in the source inode is
1772	 * tracking beyond EOF so that when the fork is truncated away when the
1773	 * temporary inode is unlinked we don't underrun the i_delayed_blks
1774	 * counter on that inode.
1775	 */
1776	ASSERT(tip->i_delayed_blks == 0);
1777	tip->i_delayed_blks = ip->i_delayed_blks;
1778	ip->i_delayed_blks = 0;
1779
1780	switch (ip->i_d.di_format) {
1781	case XFS_DINODE_FMT_EXTENTS:
1782		(*src_log_flags) |= XFS_ILOG_DEXT;
1783		break;
1784	case XFS_DINODE_FMT_BTREE:
1785		ASSERT(ip->i_d.di_version < 3 ||
1786		       (*src_log_flags & XFS_ILOG_DOWNER));
1787		(*src_log_flags) |= XFS_ILOG_DBROOT;
1788		break;
1789	}
1790
1791	switch (tip->i_d.di_format) {
1792	case XFS_DINODE_FMT_EXTENTS:
1793		(*target_log_flags) |= XFS_ILOG_DEXT;
1794		break;
1795	case XFS_DINODE_FMT_BTREE:
1796		(*target_log_flags) |= XFS_ILOG_DBROOT;
1797		ASSERT(tip->i_d.di_version < 3 ||
1798		       (*target_log_flags & XFS_ILOG_DOWNER));
1799		break;
1800	}
1801
1802	return 0;
1803}
1804
1805/*
1806 * Fix up the owners of the bmbt blocks to refer to the current inode. The
1807 * change owner scan attempts to order all modified buffers in the current
1808 * transaction. In the event of ordered buffer failure, the offending buffer is
1809 * physically logged as a fallback and the scan returns -EAGAIN. We must roll
1810 * the transaction in this case to replenish the fallback log reservation and
1811 * restart the scan. This process repeats until the scan completes.
1812 */
1813static int
1814xfs_swap_change_owner(
1815	struct xfs_trans	**tpp,
1816	struct xfs_inode	*ip,
1817	struct xfs_inode	*tmpip)
1818{
1819	int			error;
1820	struct xfs_trans	*tp = *tpp;
1821
1822	do {
1823		error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK, ip->i_ino,
1824					      NULL);
1825		/* success or fatal error */
1826		if (error != -EAGAIN)
1827			break;
1828
1829		error = xfs_trans_roll(tpp);
1830		if (error)
1831			break;
1832		tp = *tpp;
1833
1834		/*
1835		 * Redirty both inodes so they can relog and keep the log tail
1836		 * moving forward.
1837		 */
1838		xfs_trans_ijoin(tp, ip, 0);
1839		xfs_trans_ijoin(tp, tmpip, 0);
1840		xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1841		xfs_trans_log_inode(tp, tmpip, XFS_ILOG_CORE);
1842	} while (true);
1843
1844	return error;
1845}
1846
1847int
1848xfs_swap_extents(
1849	struct xfs_inode	*ip,	/* target inode */
1850	struct xfs_inode	*tip,	/* tmp inode */
1851	struct xfs_swapext	*sxp)
1852{
1853	struct xfs_mount	*mp = ip->i_mount;
1854	struct xfs_trans	*tp;
1855	struct xfs_bstat	*sbp = &sxp->sx_stat;
1856	int			src_log_flags, target_log_flags;
1857	int			error = 0;
1858	int			lock_flags;
1859	struct xfs_ifork	*cowfp;
1860	uint64_t		f;
1861	int			resblks = 0;
 
 
1862
1863	/*
1864	 * Lock the inodes against other IO, page faults and truncate to
1865	 * begin with.  Then we can ensure the inodes are flushed and have no
1866	 * page cache safely. Once we have done this we can take the ilocks and
1867	 * do the rest of the checks.
1868	 */
1869	lock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1870	lock_flags = XFS_MMAPLOCK_EXCL;
1871	xfs_lock_two_inodes(ip, XFS_MMAPLOCK_EXCL, tip, XFS_MMAPLOCK_EXCL);
1872
1873	/* Verify that both files have the same format */
1874	if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) {
1875		error = -EINVAL;
1876		goto out_unlock;
1877	}
1878
1879	/* Verify both files are either real-time or non-realtime */
1880	if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1881		error = -EINVAL;
1882		goto out_unlock;
1883	}
1884
 
 
 
 
 
 
 
 
1885	error = xfs_swap_extent_flush(ip);
1886	if (error)
1887		goto out_unlock;
1888	error = xfs_swap_extent_flush(tip);
1889	if (error)
1890		goto out_unlock;
1891
 
 
 
 
 
 
1892	/*
1893	 * Extent "swapping" with rmap requires a permanent reservation and
1894	 * a block reservation because it's really just a remap operation
1895	 * performed with log redo items!
1896	 */
1897	if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
1898		int		w	= XFS_DATA_FORK;
1899		uint32_t	ipnext	= XFS_IFORK_NEXTENTS(ip, w);
1900		uint32_t	tipnext	= XFS_IFORK_NEXTENTS(tip, w);
1901
1902		/*
1903		 * Conceptually this shouldn't affect the shape of either bmbt,
1904		 * but since we atomically move extents one by one, we reserve
1905		 * enough space to rebuild both trees.
1906		 */
1907		resblks = XFS_SWAP_RMAP_SPACE_RES(mp, ipnext, w);
1908		resblks +=  XFS_SWAP_RMAP_SPACE_RES(mp, tipnext, w);
1909
1910		/*
1911		 * Handle the corner case where either inode might straddle the
1912		 * btree format boundary. If so, the inode could bounce between
1913		 * btree <-> extent format on unmap -> remap cycles, freeing and
1914		 * allocating a bmapbt block each time.
 
1915		 */
1916		if (ipnext == (XFS_IFORK_MAXEXT(ip, w) + 1))
1917			resblks += XFS_IFORK_MAXEXT(ip, w);
1918		if (tipnext == (XFS_IFORK_MAXEXT(tip, w) + 1))
1919			resblks += XFS_IFORK_MAXEXT(tip, w);
1920	}
1921	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
 
1922	if (error)
1923		goto out_unlock;
1924
1925	/*
1926	 * Lock and join the inodes to the tansaction so that transaction commit
1927	 * or cancel will unlock the inodes from this point onwards.
1928	 */
1929	xfs_lock_two_inodes(ip, XFS_ILOCK_EXCL, tip, XFS_ILOCK_EXCL);
1930	lock_flags |= XFS_ILOCK_EXCL;
1931	xfs_trans_ijoin(tp, ip, 0);
1932	xfs_trans_ijoin(tp, tip, 0);
1933
1934
1935	/* Verify all data are being swapped */
1936	if (sxp->sx_offset != 0 ||
1937	    sxp->sx_length != ip->i_d.di_size ||
1938	    sxp->sx_length != tip->i_d.di_size) {
1939		error = -EFAULT;
1940		goto out_trans_cancel;
1941	}
1942
1943	trace_xfs_swap_extent_before(ip, 0);
1944	trace_xfs_swap_extent_before(tip, 1);
1945
1946	/* check inode formats now that data is flushed */
1947	error = xfs_swap_extents_check_format(ip, tip);
1948	if (error) {
1949		xfs_notice(mp,
1950		    "%s: inode 0x%llx format is incompatible for exchanging.",
1951				__func__, ip->i_ino);
1952		goto out_trans_cancel;
1953	}
1954
1955	/*
1956	 * Compare the current change & modify times with that
1957	 * passed in.  If they differ, we abort this swap.
1958	 * This is the mechanism used to ensure the calling
1959	 * process that the file was not changed out from
1960	 * under it.
1961	 */
1962	if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
1963	    (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
1964	    (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
1965	    (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
 
 
1966		error = -EBUSY;
1967		goto out_trans_cancel;
1968	}
1969
1970	/*
1971	 * Note the trickiness in setting the log flags - we set the owner log
1972	 * flag on the opposite inode (i.e. the inode we are setting the new
1973	 * owner to be) because once we swap the forks and log that, log
1974	 * recovery is going to see the fork as owned by the swapped inode,
1975	 * not the pre-swapped inodes.
1976	 */
1977	src_log_flags = XFS_ILOG_CORE;
1978	target_log_flags = XFS_ILOG_CORE;
1979
1980	if (xfs_sb_version_hasrmapbt(&mp->m_sb))
1981		error = xfs_swap_extent_rmap(&tp, ip, tip);
1982	else
1983		error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags,
1984				&target_log_flags);
1985	if (error)
1986		goto out_trans_cancel;
1987
1988	/* Do we have to swap reflink flags? */
1989	if ((ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK) ^
1990	    (tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)) {
1991		f = ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
1992		ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1993		ip->i_d.di_flags2 |= tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
1994		tip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1995		tip->i_d.di_flags2 |= f & XFS_DIFLAG2_REFLINK;
1996	}
1997
1998	/* Swap the cow forks. */
1999	if (xfs_sb_version_hasreflink(&mp->m_sb)) {
2000		xfs_extnum_t	extnum;
2001
2002		ASSERT(ip->i_cformat == XFS_DINODE_FMT_EXTENTS);
2003		ASSERT(tip->i_cformat == XFS_DINODE_FMT_EXTENTS);
2004
2005		extnum = ip->i_cnextents;
2006		ip->i_cnextents = tip->i_cnextents;
2007		tip->i_cnextents = extnum;
2008
2009		cowfp = ip->i_cowfp;
2010		ip->i_cowfp = tip->i_cowfp;
2011		tip->i_cowfp = cowfp;
2012
2013		if (ip->i_cowfp && ip->i_cowfp->if_bytes)
2014			xfs_inode_set_cowblocks_tag(ip);
2015		else
2016			xfs_inode_clear_cowblocks_tag(ip);
2017		if (tip->i_cowfp && tip->i_cowfp->if_bytes)
2018			xfs_inode_set_cowblocks_tag(tip);
2019		else
2020			xfs_inode_clear_cowblocks_tag(tip);
2021	}
2022
2023	xfs_trans_log_inode(tp, ip,  src_log_flags);
2024	xfs_trans_log_inode(tp, tip, target_log_flags);
2025
2026	/*
2027	 * The extent forks have been swapped, but crc=1,rmapbt=0 filesystems
2028	 * have inode number owner values in the bmbt blocks that still refer to
2029	 * the old inode. Scan each bmbt to fix up the owner values with the
2030	 * inode number of the current inode.
2031	 */
2032	if (src_log_flags & XFS_ILOG_DOWNER) {
2033		error = xfs_swap_change_owner(&tp, ip, tip);
2034		if (error)
2035			goto out_trans_cancel;
2036	}
2037	if (target_log_flags & XFS_ILOG_DOWNER) {
2038		error = xfs_swap_change_owner(&tp, tip, ip);
2039		if (error)
2040			goto out_trans_cancel;
2041	}
2042
2043	/*
2044	 * If this is a synchronous mount, make sure that the
2045	 * transaction goes to disk before returning to the user.
2046	 */
2047	if (mp->m_flags & XFS_MOUNT_WSYNC)
2048		xfs_trans_set_sync(tp);
2049
2050	error = xfs_trans_commit(tp);
2051
2052	trace_xfs_swap_extent_after(ip, 0);
2053	trace_xfs_swap_extent_after(tip, 1);
2054
 
 
 
2055out_unlock:
2056	xfs_iunlock(ip, lock_flags);
2057	xfs_iunlock(tip, lock_flags);
2058	unlock_two_nondirectories(VFS_I(ip), VFS_I(tip));
2059	return error;
2060
2061out_trans_cancel:
2062	xfs_trans_cancel(tp);
2063	goto out_unlock;
2064}