1// SPDX-License-Identifier: GPL-2.0+
   2/*
   3 * Copyright (C) 2016 Oracle.  All Rights Reserved.
   4 * Author: Darrick J. Wong <darrick.wong@oracle.com>
   5 */
   6#include "xfs.h"
   7#include "xfs_fs.h"
   8#include "xfs_shared.h"
   9#include "xfs_format.h"
  10#include "xfs_log_format.h"
  11#include "xfs_trans_resv.h"
  12#include "xfs_mount.h"
  13#include "xfs_defer.h"
  14#include "xfs_inode.h"
  15#include "xfs_trans.h"
  16#include "xfs_bmap.h"
  17#include "xfs_bmap_util.h"
  18#include "xfs_trace.h"
  19#include "xfs_icache.h"
  20#include "xfs_btree.h"
  21#include "xfs_refcount_btree.h"
  22#include "xfs_refcount.h"
  23#include "xfs_bmap_btree.h"
  24#include "xfs_trans_space.h"
  25#include "xfs_bit.h"
  26#include "xfs_alloc.h"
  27#include "xfs_quota.h"
  28#include "xfs_reflink.h"
  29#include "xfs_iomap.h"
  30#include "xfs_ag.h"
  31#include "xfs_ag_resv.h"
  32#include "xfs_health.h"
  33
  34/*
  35 * Copy on Write of Shared Blocks
  36 *
  37 * XFS must preserve "the usual" file semantics even when two files share
  38 * the same physical blocks.  This means that a write to one file must not
  39 * alter the blocks in a different file; the way that we'll do that is
  40 * through the use of a copy-on-write mechanism.  At a high level, that
  41 * means that when we want to write to a shared block, we allocate a new
  42 * block, write the data to the new block, and if that succeeds we map the
  43 * new block into the file.
  44 *
  45 * XFS provides a "delayed allocation" mechanism that defers the allocation
  46 * of disk blocks to dirty-but-not-yet-mapped file blocks as long as
  47 * possible.  This reduces fragmentation by enabling the filesystem to ask
  48 * for bigger chunks less often, which is exactly what we want for CoW.
  49 *
  50 * The delalloc mechanism begins when the kernel wants to make a block
  51 * writable (write_begin or page_mkwrite).  If the offset is not mapped, we
  52 * create a delalloc mapping, which is a regular in-core extent, but without
  53 * a real startblock.  (For delalloc mappings, the startblock encodes both
  54 * a flag that this is a delalloc mapping, and a worst-case estimate of how
  55 * many blocks might be required to put the mapping into the BMBT.)  delalloc
  56 * mappings are a reservation against the free space in the filesystem;
  57 * adjacent mappings can also be combined into fewer larger mappings.
  58 *
  59 * As an optimization, the CoW extent size hint (cowextsz) creates
  60 * outsized aligned delalloc reservations in the hope of landing out of
  61 * order nearby CoW writes in a single extent on disk, thereby reducing
  62 * fragmentation and improving future performance.
  63 *
  64 * D: --RRRRRRSSSRRRRRRRR--- (data fork)
  65 * C: ------DDDDDDD--------- (CoW fork)
  66 *
  67 * When dirty pages are being written out (typically in writepage), the
  68 * delalloc reservations are converted into unwritten mappings by
  69 * allocating blocks and replacing the delalloc mapping with real ones.
  70 * A delalloc mapping can be replaced by several unwritten ones if the
  71 * free space is fragmented.
  72 *
  73 * D: --RRRRRRSSSRRRRRRRR---
  74 * C: ------UUUUUUU---------
  75 *
  76 * We want to adapt the delalloc mechanism for copy-on-write, since the
  77 * write paths are similar.  The first two steps (creating the reservation
  78 * and allocating the blocks) are exactly the same as delalloc except that
  79 * the mappings must be stored in a separate CoW fork because we do not want
  80 * to disturb the mapping in the data fork until we're sure that the write
  81 * succeeded.  IO completion in this case is the process of removing the old
  82 * mapping from the data fork and moving the new mapping from the CoW fork to
  83 * the data fork.  This will be discussed shortly.
  84 *
  85 * For now, unaligned directio writes will be bounced back to the page cache.
  86 * Block-aligned directio writes will use the same mechanism as buffered
  87 * writes.
  88 *
  89 * Just prior to submitting the actual disk write requests, we convert
  90 * the extents representing the range of the file actually being written
  91 * (as opposed to extra pieces created for the cowextsize hint) to real
  92 * extents.  This will become important in the next step:
  93 *
  94 * D: --RRRRRRSSSRRRRRRRR---
  95 * C: ------UUrrUUU---------
  96 *
  97 * CoW remapping must be done after the data block write completes,
  98 * because we don't want to destroy the old data fork map until we're sure
  99 * the new block has been written.  Since the new mappings are kept in a
 100 * separate fork, we can simply iterate these mappings to find the ones
 101 * that cover the file blocks that we just CoW'd.  For each extent, simply
 102 * unmap the corresponding range in the data fork, map the new range into
 103 * the data fork, and remove the extent from the CoW fork.  Because of
 104 * the presence of the cowextsize hint, however, we must be careful
 105 * only to remap the blocks that we've actually written out --  we must
 106 * never remap delalloc reservations nor CoW staging blocks that have
 107 * yet to be written.  This corresponds exactly to the real extents in
 108 * the CoW fork:
 109 *
 110 * D: --RRRRRRrrSRRRRRRRR---
 111 * C: ------UU--UUU---------
 112 *
 113 * Since the remapping operation can be applied to an arbitrary file
 114 * range, we record the need for the remap step as a flag in the ioend
 115 * instead of declaring a new IO type.  This is required for direct io
 116 * because we only have ioend for the whole dio, and we have to be able to
 117 * remember the presence of unwritten blocks and CoW blocks with a single
 118 * ioend structure.  Better yet, the more ground we can cover with one
 119 * ioend, the better.
 120 */
 121
 122/*
 123 * Given an AG extent, find the lowest-numbered run of shared blocks
 124 * within that range and return the range in fbno/flen.  If
 125 * find_end_of_shared is true, return the longest contiguous extent of
 126 * shared blocks.  If there are no shared extents, fbno and flen will
 127 * be set to NULLAGBLOCK and 0, respectively.
 128 */
 129static int
 130xfs_reflink_find_shared(
 131	struct xfs_perag	*pag,
 132	struct xfs_trans	*tp,
 
 133	xfs_agblock_t		agbno,
 134	xfs_extlen_t		aglen,
 135	xfs_agblock_t		*fbno,
 136	xfs_extlen_t		*flen,
 137	bool			find_end_of_shared)
 138{
 139	struct xfs_buf		*agbp;
 140	struct xfs_btree_cur	*cur;
 141	int			error;
 142
 143	error = xfs_alloc_read_agf(pag, tp, 0, &agbp);
 144	if (error)
 145		return error;
 146
 147	cur = xfs_refcountbt_init_cursor(pag_mount(pag), tp, agbp, pag);
 148
 149	error = xfs_refcount_find_shared(cur, agbno, aglen, fbno, flen,
 150			find_end_of_shared);
 151
 152	xfs_btree_del_cursor(cur, error);
 153
 154	xfs_trans_brelse(tp, agbp);
 155	return error;
 156}
 157
 158/*
 159 * Trim the mapping to the next block where there's a change in the
 160 * shared/unshared status.  More specifically, this means that we
 161 * find the lowest-numbered extent of shared blocks that coincides with
 162 * the given block mapping.  If the shared extent overlaps the start of
 163 * the mapping, trim the mapping to the end of the shared extent.  If
 164 * the shared region intersects the mapping, trim the mapping to the
 165 * start of the shared extent.  If there are no shared regions that
 166 * overlap, just return the original extent.
 167 */
 168int
 169xfs_reflink_trim_around_shared(
 170	struct xfs_inode	*ip,
 171	struct xfs_bmbt_irec	*irec,
 172	bool			*shared)
 173{
 174	struct xfs_mount	*mp = ip->i_mount;
 175	struct xfs_perag	*pag;
 176	xfs_agblock_t		agbno;
 177	xfs_extlen_t		aglen;
 178	xfs_agblock_t		fbno;
 179	xfs_extlen_t		flen;
 180	int			error = 0;
 181
 182	/* Holes, unwritten, and delalloc extents cannot be shared */
 183	if (!xfs_is_cow_inode(ip) || !xfs_bmap_is_written_extent(irec)) {
 184		*shared = false;
 185		return 0;
 186	}
 187
 188	trace_xfs_reflink_trim_around_shared(ip, irec);
 189
 190	pag = xfs_perag_get(mp, XFS_FSB_TO_AGNO(mp, irec->br_startblock));
 191	agbno = XFS_FSB_TO_AGBNO(mp, irec->br_startblock);
 192	aglen = irec->br_blockcount;
 193
 194	error = xfs_reflink_find_shared(pag, NULL, agbno, aglen, &fbno, &flen,
 195			true);
 196	xfs_perag_put(pag);
 197	if (error)
 198		return error;
 199
 200	*shared = false;
 201	if (fbno == NULLAGBLOCK) {
 202		/* No shared blocks at all. */
 203		return 0;
 204	}
 205
 206	if (fbno == agbno) {
 207		/*
 208		 * The start of this extent is shared.  Truncate the
 209		 * mapping at the end of the shared region so that a
 210		 * subsequent iteration starts at the start of the
 211		 * unshared region.
 212		 */
 213		irec->br_blockcount = flen;
 214		*shared = true;
 215		return 0;
 
 
 
 
 
 
 
 
 
 216	}
 217
 218	/*
 219	 * There's a shared extent midway through this extent.
 220	 * Truncate the mapping at the start of the shared
 221	 * extent so that a subsequent iteration starts at the
 222	 * start of the shared region.
 223	 */
 224	irec->br_blockcount = fbno - agbno;
 225	return 0;
 226}
 227
 228int
 229xfs_bmap_trim_cow(
 230	struct xfs_inode	*ip,
 231	struct xfs_bmbt_irec	*imap,
 232	bool			*shared)
 233{
 234	/* We can't update any real extents in always COW mode. */
 235	if (xfs_is_always_cow_inode(ip) &&
 236	    !isnullstartblock(imap->br_startblock)) {
 237		*shared = true;
 238		return 0;
 239	}
 240
 241	/* Trim the mapping to the nearest shared extent boundary. */
 242	return xfs_reflink_trim_around_shared(ip, imap, shared);
 243}
 244
 245static int
 246xfs_reflink_convert_cow_locked(
 247	struct xfs_inode	*ip,
 248	xfs_fileoff_t		offset_fsb,
 249	xfs_filblks_t		count_fsb)
 250{
 251	struct xfs_iext_cursor	icur;
 252	struct xfs_bmbt_irec	got;
 253	struct xfs_btree_cur	*dummy_cur = NULL;
 254	int			dummy_logflags;
 255	int			error = 0;
 256
 257	if (!xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &got))
 258		return 0;
 259
 260	do {
 261		if (got.br_startoff >= offset_fsb + count_fsb)
 262			break;
 263		if (got.br_state == XFS_EXT_NORM)
 264			continue;
 265		if (WARN_ON_ONCE(isnullstartblock(got.br_startblock)))
 266			return -EIO;
 267
 268		xfs_trim_extent(&got, offset_fsb, count_fsb);
 269		if (!got.br_blockcount)
 270			continue;
 271
 272		got.br_state = XFS_EXT_NORM;
 273		error = xfs_bmap_add_extent_unwritten_real(NULL, ip,
 274				XFS_COW_FORK, &icur, &dummy_cur, &got,
 275				&dummy_logflags);
 276		if (error)
 277			return error;
 278	} while (xfs_iext_next_extent(ip->i_cowfp, &icur, &got));
 279
 280	return error;
 281}
 282
 283/* Convert all of the unwritten CoW extents in a file's range to real ones. */
 284int
 285xfs_reflink_convert_cow(
 286	struct xfs_inode	*ip,
 287	xfs_off_t		offset,
 288	xfs_off_t		count)
 289{
 290	struct xfs_mount	*mp = ip->i_mount;
 291	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
 292	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, offset + count);
 293	xfs_filblks_t		count_fsb = end_fsb - offset_fsb;
 294	int			error;
 295
 296	ASSERT(count != 0);
 297
 298	xfs_ilock(ip, XFS_ILOCK_EXCL);
 299	error = xfs_reflink_convert_cow_locked(ip, offset_fsb, count_fsb);
 300	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 301	return error;
 302}
 303
 304/*
 305 * Find the extent that maps the given range in the COW fork. Even if the extent
 306 * is not shared we might have a preallocation for it in the COW fork. If so we
 307 * use it that rather than trigger a new allocation.
 308 */
 309static int
 310xfs_find_trim_cow_extent(
 311	struct xfs_inode	*ip,
 312	struct xfs_bmbt_irec	*imap,
 313	struct xfs_bmbt_irec	*cmap,
 314	bool			*shared,
 315	bool			*found)
 316{
 317	xfs_fileoff_t		offset_fsb = imap->br_startoff;
 318	xfs_filblks_t		count_fsb = imap->br_blockcount;
 319	struct xfs_iext_cursor	icur;
 320
 321	*found = false;
 322
 323	/*
 324	 * If we don't find an overlapping extent, trim the range we need to
 325	 * allocate to fit the hole we found.
 326	 */
 327	if (!xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, cmap))
 328		cmap->br_startoff = offset_fsb + count_fsb;
 329	if (cmap->br_startoff > offset_fsb) {
 330		xfs_trim_extent(imap, imap->br_startoff,
 331				cmap->br_startoff - imap->br_startoff);
 332		return xfs_bmap_trim_cow(ip, imap, shared);
 333	}
 334
 335	*shared = true;
 336	if (isnullstartblock(cmap->br_startblock)) {
 337		xfs_trim_extent(imap, cmap->br_startoff, cmap->br_blockcount);
 338		return 0;
 339	}
 340
 341	/* real extent found - no need to allocate */
 342	xfs_trim_extent(cmap, offset_fsb, count_fsb);
 343	*found = true;
 344	return 0;
 345}
 346
 347static int
 348xfs_reflink_convert_unwritten(
 349	struct xfs_inode	*ip,
 350	struct xfs_bmbt_irec	*imap,
 351	struct xfs_bmbt_irec	*cmap,
 352	bool			convert_now)
 353{
 354	xfs_fileoff_t		offset_fsb = imap->br_startoff;
 355	xfs_filblks_t		count_fsb = imap->br_blockcount;
 356	int			error;
 357
 358	/*
 359	 * cmap might larger than imap due to cowextsize hint.
 360	 */
 361	xfs_trim_extent(cmap, offset_fsb, count_fsb);
 362
 363	/*
 364	 * COW fork extents are supposed to remain unwritten until we're ready
 365	 * to initiate a disk write.  For direct I/O we are going to write the
 366	 * data and need the conversion, but for buffered writes we're done.
 367	 */
 368	if (!convert_now || cmap->br_state == XFS_EXT_NORM)
 369		return 0;
 370
 371	trace_xfs_reflink_convert_cow(ip, cmap);
 372
 373	error = xfs_reflink_convert_cow_locked(ip, offset_fsb, count_fsb);
 374	if (!error)
 375		cmap->br_state = XFS_EXT_NORM;
 376
 377	return error;
 378}
 379
 380static int
 381xfs_reflink_fill_cow_hole(
 382	struct xfs_inode	*ip,
 383	struct xfs_bmbt_irec	*imap,
 384	struct xfs_bmbt_irec	*cmap,
 385	bool			*shared,
 386	uint			*lockmode,
 387	bool			convert_now)
 388{
 389	struct xfs_mount	*mp = ip->i_mount;
 
 
 390	struct xfs_trans	*tp;
 391	xfs_filblks_t		resaligned;
 392	xfs_extlen_t		resblks;
 393	int			nimaps;
 394	int			error;
 395	bool			found;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 396
 397	resaligned = xfs_aligned_fsb_count(imap->br_startoff,
 398		imap->br_blockcount, xfs_get_cowextsz_hint(ip));
 399	resblks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
 400
 401	xfs_iunlock(ip, *lockmode);
 402	*lockmode = 0;
 
 
 403
 404	error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, resblks, 0,
 405			false, &tp);
 406	if (error)
 407		return error;
 408
 409	*lockmode = XFS_ILOCK_EXCL;
 
 
 410
 
 
 
 411	error = xfs_find_trim_cow_extent(ip, imap, cmap, shared, &found);
 412	if (error || !*shared)
 413		goto out_trans_cancel;
 414
 415	if (found) {
 416		xfs_trans_cancel(tp);
 417		goto convert;
 418	}
 419
 
 
 
 
 
 
 
 420	/* Allocate the entire reservation as unwritten blocks. */
 421	nimaps = 1;
 422	error = xfs_bmapi_write(tp, ip, imap->br_startoff, imap->br_blockcount,
 423			XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC, 0, cmap,
 424			&nimaps);
 425	if (error)
 426		goto out_trans_cancel;
 427
 428	xfs_inode_set_cowblocks_tag(ip);
 429	error = xfs_trans_commit(tp);
 430	if (error)
 431		return error;
 432
 
 
 
 
 
 
 433convert:
 434	return xfs_reflink_convert_unwritten(ip, imap, cmap, convert_now);
 435
 436out_trans_cancel:
 437	xfs_trans_cancel(tp);
 438	return error;
 439}
 440
 441static int
 442xfs_reflink_fill_delalloc(
 443	struct xfs_inode	*ip,
 444	struct xfs_bmbt_irec	*imap,
 445	struct xfs_bmbt_irec	*cmap,
 446	bool			*shared,
 447	uint			*lockmode,
 448	bool			convert_now)
 449{
 450	struct xfs_mount	*mp = ip->i_mount;
 451	struct xfs_trans	*tp;
 452	int			nimaps;
 453	int			error;
 454	bool			found;
 455
 456	do {
 457		xfs_iunlock(ip, *lockmode);
 458		*lockmode = 0;
 459
 460		error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, 0, 0,
 461				false, &tp);
 462		if (error)
 463			return error;
 464
 465		*lockmode = XFS_ILOCK_EXCL;
 466
 467		error = xfs_find_trim_cow_extent(ip, imap, cmap, shared,
 468				&found);
 469		if (error || !*shared)
 470			goto out_trans_cancel;
 471
 472		if (found) {
 473			xfs_trans_cancel(tp);
 474			break;
 475		}
 476
 477		ASSERT(isnullstartblock(cmap->br_startblock) ||
 478		       cmap->br_startblock == DELAYSTARTBLOCK);
 479
 480		/*
 481		 * Replace delalloc reservation with an unwritten extent.
 482		 */
 483		nimaps = 1;
 484		error = xfs_bmapi_write(tp, ip, cmap->br_startoff,
 485				cmap->br_blockcount,
 486				XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC, 0,
 487				cmap, &nimaps);
 488		if (error)
 489			goto out_trans_cancel;
 490
 491		xfs_inode_set_cowblocks_tag(ip);
 492		error = xfs_trans_commit(tp);
 493		if (error)
 494			return error;
 495	} while (cmap->br_startoff + cmap->br_blockcount <= imap->br_startoff);
 496
 497	return xfs_reflink_convert_unwritten(ip, imap, cmap, convert_now);
 498
 
 
 
 499out_trans_cancel:
 500	xfs_trans_cancel(tp);
 501	return error;
 502}
 503
 504/* Allocate all CoW reservations covering a range of blocks in a file. */
 505int
 506xfs_reflink_allocate_cow(
 507	struct xfs_inode	*ip,
 508	struct xfs_bmbt_irec	*imap,
 509	struct xfs_bmbt_irec	*cmap,
 510	bool			*shared,
 511	uint			*lockmode,
 512	bool			convert_now)
 513{
 514	int			error;
 515	bool			found;
 516
 517	xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
 518	if (!ip->i_cowfp) {
 519		ASSERT(!xfs_is_reflink_inode(ip));
 520		xfs_ifork_init_cow(ip);
 521	}
 522
 523	error = xfs_find_trim_cow_extent(ip, imap, cmap, shared, &found);
 524	if (error || !*shared)
 525		return error;
 526
 527	/* CoW fork has a real extent */
 528	if (found)
 529		return xfs_reflink_convert_unwritten(ip, imap, cmap,
 530				convert_now);
 531
 532	/*
 533	 * CoW fork does not have an extent and data extent is shared.
 534	 * Allocate a real extent in the CoW fork.
 535	 */
 536	if (cmap->br_startoff > imap->br_startoff)
 537		return xfs_reflink_fill_cow_hole(ip, imap, cmap, shared,
 538				lockmode, convert_now);
 539
 540	/*
 541	 * CoW fork has a delalloc reservation. Replace it with a real extent.
 542	 * There may or may not be a data fork mapping.
 543	 */
 544	if (isnullstartblock(cmap->br_startblock) ||
 545	    cmap->br_startblock == DELAYSTARTBLOCK)
 546		return xfs_reflink_fill_delalloc(ip, imap, cmap, shared,
 547				lockmode, convert_now);
 548
 549	/* Shouldn't get here. */
 550	ASSERT(0);
 551	return -EFSCORRUPTED;
 552}
 553
 554/*
 555 * Cancel CoW reservations for some block range of an inode.
 556 *
 557 * If cancel_real is true this function cancels all COW fork extents for the
 558 * inode; if cancel_real is false, real extents are not cleared.
 559 *
 560 * Caller must have already joined the inode to the current transaction. The
 561 * inode will be joined to the transaction returned to the caller.
 562 */
 563int
 564xfs_reflink_cancel_cow_blocks(
 565	struct xfs_inode		*ip,
 566	struct xfs_trans		**tpp,
 567	xfs_fileoff_t			offset_fsb,
 568	xfs_fileoff_t			end_fsb,
 569	bool				cancel_real)
 570{
 571	struct xfs_ifork		*ifp = xfs_ifork_ptr(ip, XFS_COW_FORK);
 572	struct xfs_bmbt_irec		got, del;
 573	struct xfs_iext_cursor		icur;
 574	int				error = 0;
 575
 576	if (!xfs_inode_has_cow_data(ip))
 577		return 0;
 578	if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got))
 579		return 0;
 580
 581	/* Walk backwards until we're out of the I/O range... */
 582	while (got.br_startoff + got.br_blockcount > offset_fsb) {
 583		del = got;
 584		xfs_trim_extent(&del, offset_fsb, end_fsb - offset_fsb);
 585
 586		/* Extent delete may have bumped ext forward */
 587		if (!del.br_blockcount) {
 588			xfs_iext_prev(ifp, &icur);
 589			goto next_extent;
 590		}
 591
 592		trace_xfs_reflink_cancel_cow(ip, &del);
 593
 594		if (isnullstartblock(del.br_startblock)) {
 595			xfs_bmap_del_extent_delay(ip, XFS_COW_FORK, &icur, &got,
 596					&del);
 
 
 597		} else if (del.br_state == XFS_EXT_UNWRITTEN || cancel_real) {
 598			ASSERT((*tpp)->t_highest_agno == NULLAGNUMBER);
 599
 600			/* Free the CoW orphan record. */
 601			xfs_refcount_free_cow_extent(*tpp, del.br_startblock,
 602					del.br_blockcount);
 603
 604			error = xfs_free_extent_later(*tpp, del.br_startblock,
 605					del.br_blockcount, NULL,
 606					XFS_AG_RESV_NONE, 0);
 607			if (error)
 608				break;
 609
 610			/* Roll the transaction */
 611			error = xfs_defer_finish(tpp);
 612			if (error)
 613				break;
 614
 615			/* Remove the mapping from the CoW fork. */
 616			xfs_bmap_del_extent_cow(ip, &icur, &got, &del);
 617
 618			/* Remove the quota reservation */
 619			xfs_quota_unreserve_blkres(ip, del.br_blockcount);
 
 
 
 
 620		} else {
 621			/* Didn't do anything, push cursor back. */
 622			xfs_iext_prev(ifp, &icur);
 623		}
 624next_extent:
 625		if (!xfs_iext_get_extent(ifp, &icur, &got))
 626			break;
 627	}
 628
 629	/* clear tag if cow fork is emptied */
 630	if (!ifp->if_bytes)
 631		xfs_inode_clear_cowblocks_tag(ip);
 632	return error;
 633}
 634
 635/*
 636 * Cancel CoW reservations for some byte range of an inode.
 637 *
 638 * If cancel_real is true this function cancels all COW fork extents for the
 639 * inode; if cancel_real is false, real extents are not cleared.
 640 */
 641int
 642xfs_reflink_cancel_cow_range(
 643	struct xfs_inode	*ip,
 644	xfs_off_t		offset,
 645	xfs_off_t		count,
 646	bool			cancel_real)
 647{
 648	struct xfs_trans	*tp;
 649	xfs_fileoff_t		offset_fsb;
 650	xfs_fileoff_t		end_fsb;
 651	int			error;
 652
 653	trace_xfs_reflink_cancel_cow_range(ip, offset, count);
 654	ASSERT(ip->i_cowfp);
 655
 656	offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
 657	if (count == NULLFILEOFF)
 658		end_fsb = NULLFILEOFF;
 659	else
 660		end_fsb = XFS_B_TO_FSB(ip->i_mount, offset + count);
 661
 662	/* Start a rolling transaction to remove the mappings */
 663	error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_write,
 664			0, 0, 0, &tp);
 665	if (error)
 666		goto out;
 667
 668	xfs_ilock(ip, XFS_ILOCK_EXCL);
 669	xfs_trans_ijoin(tp, ip, 0);
 670
 671	/* Scrape out the old CoW reservations */
 672	error = xfs_reflink_cancel_cow_blocks(ip, &tp, offset_fsb, end_fsb,
 673			cancel_real);
 674	if (error)
 675		goto out_cancel;
 676
 677	error = xfs_trans_commit(tp);
 678
 679	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 680	return error;
 681
 682out_cancel:
 683	xfs_trans_cancel(tp);
 684	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 685out:
 686	trace_xfs_reflink_cancel_cow_range_error(ip, error, _RET_IP_);
 687	return error;
 688}
 689
 690/*
 691 * Remap part of the CoW fork into the data fork.
 692 *
 693 * We aim to remap the range starting at @offset_fsb and ending at @end_fsb
 694 * into the data fork; this function will remap what it can (at the end of the
 695 * range) and update @end_fsb appropriately.  Each remap gets its own
 696 * transaction because we can end up merging and splitting bmbt blocks for
 697 * every remap operation and we'd like to keep the block reservation
 698 * requirements as low as possible.
 699 */
 700STATIC int
 701xfs_reflink_end_cow_extent(
 702	struct xfs_inode	*ip,
 703	xfs_fileoff_t		*offset_fsb,
 704	xfs_fileoff_t		end_fsb)
 705{
 
 706	struct xfs_iext_cursor	icur;
 707	struct xfs_bmbt_irec	got, del, data;
 708	struct xfs_mount	*mp = ip->i_mount;
 709	struct xfs_trans	*tp;
 710	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, XFS_COW_FORK);
 
 711	unsigned int		resblks;
 712	int			nmaps;
 713	int			error;
 714
 
 
 
 
 
 
 715	resblks = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
 716	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0,
 717			XFS_TRANS_RESERVE, &tp);
 718	if (error)
 719		return error;
 720
 721	/*
 722	 * Lock the inode.  We have to ijoin without automatic unlock because
 723	 * the lead transaction is the refcountbt record deletion; the data
 724	 * fork update follows as a deferred log item.
 725	 */
 726	xfs_ilock(ip, XFS_ILOCK_EXCL);
 727	xfs_trans_ijoin(tp, ip, 0);
 728
 729	/*
 730	 * In case of racing, overlapping AIO writes no COW extents might be
 731	 * left by the time I/O completes for the loser of the race.  In that
 732	 * case we are done.
 733	 */
 734	if (!xfs_iext_lookup_extent(ip, ifp, *offset_fsb, &icur, &got) ||
 735	    got.br_startoff >= end_fsb) {
 736		*offset_fsb = end_fsb;
 737		goto out_cancel;
 738	}
 739
 740	/*
 741	 * Only remap real extents that contain data.  With AIO, speculative
 742	 * preallocations can leak into the range we are called upon, and we
 743	 * need to skip them.  Preserve @got for the eventual CoW fork
 744	 * deletion; from now on @del represents the mapping that we're
 745	 * actually remapping.
 746	 */
 747	while (!xfs_bmap_is_written_extent(&got)) {
 748		if (!xfs_iext_next_extent(ifp, &icur, &got) ||
 749		    got.br_startoff >= end_fsb) {
 750			*offset_fsb = end_fsb;
 751			goto out_cancel;
 752		}
 753	}
 754	del = got;
 755	xfs_trim_extent(&del, *offset_fsb, end_fsb - *offset_fsb);
 756
 757	error = xfs_iext_count_extend(tp, ip, XFS_DATA_FORK,
 758			XFS_IEXT_REFLINK_END_COW_CNT);
 759	if (error)
 
 
 
 
 
 
 760		goto out_cancel;
 
 761
 762	/* Grab the corresponding mapping in the data fork. */
 763	nmaps = 1;
 764	error = xfs_bmapi_read(ip, del.br_startoff, del.br_blockcount, &data,
 765			&nmaps, 0);
 766	if (error)
 767		goto out_cancel;
 768
 769	/* We can only remap the smaller of the two extent sizes. */
 770	data.br_blockcount = min(data.br_blockcount, del.br_blockcount);
 771	del.br_blockcount = data.br_blockcount;
 772
 773	trace_xfs_reflink_cow_remap_from(ip, &del);
 774	trace_xfs_reflink_cow_remap_to(ip, &data);
 775
 776	if (xfs_bmap_is_real_extent(&data)) {
 777		/*
 778		 * If the extent we're remapping is backed by storage (written
 779		 * or not), unmap the extent and drop its refcount.
 780		 */
 781		xfs_bmap_unmap_extent(tp, ip, XFS_DATA_FORK, &data);
 782		xfs_refcount_decrease_extent(tp, &data);
 783		xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT,
 784				-data.br_blockcount);
 785	} else if (data.br_startblock == DELAYSTARTBLOCK) {
 786		int		done;
 787
 788		/*
 789		 * If the extent we're remapping is a delalloc reservation,
 790		 * we can use the regular bunmapi function to release the
 791		 * incore state.  Dropping the delalloc reservation takes care
 792		 * of the quota reservation for us.
 793		 */
 794		error = xfs_bunmapi(NULL, ip, data.br_startoff,
 795				data.br_blockcount, 0, 1, &done);
 796		if (error)
 797			goto out_cancel;
 798		ASSERT(done);
 799	}
 800
 801	/* Free the CoW orphan record. */
 802	xfs_refcount_free_cow_extent(tp, del.br_startblock, del.br_blockcount);
 803
 804	/* Map the new blocks into the data fork. */
 805	xfs_bmap_map_extent(tp, ip, XFS_DATA_FORK, &del);
 806
 807	/* Charge this new data fork mapping to the on-disk quota. */
 808	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_DELBCOUNT,
 809			(long)del.br_blockcount);
 810
 811	/* Remove the mapping from the CoW fork. */
 812	xfs_bmap_del_extent_cow(ip, &icur, &got, &del);
 813
 814	error = xfs_trans_commit(tp);
 815	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 816	if (error)
 817		return error;
 818
 819	/* Update the caller about how much progress we made. */
 820	*offset_fsb = del.br_startoff + del.br_blockcount;
 821	return 0;
 822
 823out_cancel:
 824	xfs_trans_cancel(tp);
 825	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 826	return error;
 827}
 828
 829/*
 830 * Remap parts of a file's data fork after a successful CoW.
 831 */
 832int
 833xfs_reflink_end_cow(
 834	struct xfs_inode		*ip,
 835	xfs_off_t			offset,
 836	xfs_off_t			count)
 837{
 838	xfs_fileoff_t			offset_fsb;
 839	xfs_fileoff_t			end_fsb;
 840	int				error = 0;
 841
 842	trace_xfs_reflink_end_cow(ip, offset, count);
 843
 844	offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
 845	end_fsb = XFS_B_TO_FSB(ip->i_mount, offset + count);
 846
 847	/*
 848	 * Walk forwards until we've remapped the I/O range.  The loop function
 849	 * repeatedly cycles the ILOCK to allocate one transaction per remapped
 850	 * extent.
 851	 *
 852	 * If we're being called by writeback then the pages will still
 853	 * have PageWriteback set, which prevents races with reflink remapping
 854	 * and truncate.  Reflink remapping prevents races with writeback by
 855	 * taking the iolock and mmaplock before flushing the pages and
 856	 * remapping, which means there won't be any further writeback or page
 857	 * cache dirtying until the reflink completes.
 858	 *
 859	 * We should never have two threads issuing writeback for the same file
 860	 * region.  There are also have post-eof checks in the writeback
 861	 * preparation code so that we don't bother writing out pages that are
 862	 * about to be truncated.
 863	 *
 864	 * If we're being called as part of directio write completion, the dio
 865	 * count is still elevated, which reflink and truncate will wait for.
 866	 * Reflink remapping takes the iolock and mmaplock and waits for
 867	 * pending dio to finish, which should prevent any directio until the
 868	 * remap completes.  Multiple concurrent directio writes to the same
 869	 * region are handled by end_cow processing only occurring for the
 870	 * threads which succeed; the outcome of multiple overlapping direct
 871	 * writes is not well defined anyway.
 872	 *
 873	 * It's possible that a buffered write and a direct write could collide
 874	 * here (the buffered write stumbles in after the dio flushes and
 875	 * invalidates the page cache and immediately queues writeback), but we
 876	 * have never supported this 100%.  If either disk write succeeds the
 877	 * blocks will be remapped.
 878	 */
 879	while (end_fsb > offset_fsb && !error)
 880		error = xfs_reflink_end_cow_extent(ip, &offset_fsb, end_fsb);
 881
 882	if (error)
 883		trace_xfs_reflink_end_cow_error(ip, error, _RET_IP_);
 884	return error;
 885}
 886
 887/*
 888 * Free all CoW staging blocks that are still referenced by the ondisk refcount
 889 * metadata.  The ondisk metadata does not track which inode created the
 890 * staging extent, so callers must ensure that there are no cached inodes with
 891 * live CoW staging extents.
 892 */
 893int
 894xfs_reflink_recover_cow(
 895	struct xfs_mount	*mp)
 896{
 897	struct xfs_perag	*pag = NULL;
 898	int			error = 0;
 899
 900	if (!xfs_has_reflink(mp))
 901		return 0;
 902
 903	while ((pag = xfs_perag_next(mp, pag))) {
 904		error = xfs_refcount_recover_cow_leftovers(mp, pag);
 905		if (error) {
 906			xfs_perag_rele(pag);
 907			break;
 908		}
 909	}
 910
 911	return error;
 912}
 913
 914/*
 915 * Reflinking (Block) Ranges of Two Files Together
 916 *
 917 * First, ensure that the reflink flag is set on both inodes.  The flag is an
 918 * optimization to avoid unnecessary refcount btree lookups in the write path.
 919 *
 920 * Now we can iteratively remap the range of extents (and holes) in src to the
 921 * corresponding ranges in dest.  Let drange and srange denote the ranges of
 922 * logical blocks in dest and src touched by the reflink operation.
 923 *
 924 * While the length of drange is greater than zero,
 925 *    - Read src's bmbt at the start of srange ("imap")
 926 *    - If imap doesn't exist, make imap appear to start at the end of srange
 927 *      with zero length.
 928 *    - If imap starts before srange, advance imap to start at srange.
 929 *    - If imap goes beyond srange, truncate imap to end at the end of srange.
 930 *    - Punch (imap start - srange start + imap len) blocks from dest at
 931 *      offset (drange start).
 932 *    - If imap points to a real range of pblks,
 933 *         > Increase the refcount of the imap's pblks
 934 *         > Map imap's pblks into dest at the offset
 935 *           (drange start + imap start - srange start)
 936 *    - Advance drange and srange by (imap start - srange start + imap len)
 937 *
 938 * Finally, if the reflink made dest longer, update both the in-core and
 939 * on-disk file sizes.
 940 *
 941 * ASCII Art Demonstration:
 942 *
 943 * Let's say we want to reflink this source file:
 944 *
 945 * ----SSSSSSS-SSSSS----SSSSSS (src file)
 946 *   <-------------------->
 947 *
 948 * into this destination file:
 949 *
 950 * --DDDDDDDDDDDDDDDDDDD--DDD (dest file)
 951 *        <-------------------->
 952 * '-' means a hole, and 'S' and 'D' are written blocks in the src and dest.
 953 * Observe that the range has different logical offsets in either file.
 954 *
 955 * Consider that the first extent in the source file doesn't line up with our
 956 * reflink range.  Unmapping  and remapping are separate operations, so we can
 957 * unmap more blocks from the destination file than we remap.
 958 *
 959 * ----SSSSSSS-SSSSS----SSSSSS
 960 *   <------->
 961 * --DDDDD---------DDDDD--DDD
 962 *        <------->
 963 *
 964 * Now remap the source extent into the destination file:
 965 *
 966 * ----SSSSSSS-SSSSS----SSSSSS
 967 *   <------->
 968 * --DDDDD--SSSSSSSDDDDD--DDD
 969 *        <------->
 970 *
 971 * Do likewise with the second hole and extent in our range.  Holes in the
 972 * unmap range don't affect our operation.
 973 *
 974 * ----SSSSSSS-SSSSS----SSSSSS
 975 *            <---->
 976 * --DDDDD--SSSSSSS-SSSSS-DDD
 977 *                 <---->
 978 *
 979 * Finally, unmap and remap part of the third extent.  This will increase the
 980 * size of the destination file.
 981 *
 982 * ----SSSSSSS-SSSSS----SSSSSS
 983 *                  <----->
 984 * --DDDDD--SSSSSSS-SSSSS----SSS
 985 *                       <----->
 986 *
 987 * Once we update the destination file's i_size, we're done.
 988 */
 989
 990/*
 991 * Ensure the reflink bit is set in both inodes.
 992 */
 993STATIC int
 994xfs_reflink_set_inode_flag(
 995	struct xfs_inode	*src,
 996	struct xfs_inode	*dest)
 997{
 998	struct xfs_mount	*mp = src->i_mount;
 999	int			error;
1000	struct xfs_trans	*tp;
1001
1002	if (xfs_is_reflink_inode(src) && xfs_is_reflink_inode(dest))
1003		return 0;
1004
1005	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
1006	if (error)
1007		goto out_error;
1008
1009	/* Lock both files against IO */
1010	if (src->i_ino == dest->i_ino)
1011		xfs_ilock(src, XFS_ILOCK_EXCL);
1012	else
1013		xfs_lock_two_inodes(src, XFS_ILOCK_EXCL, dest, XFS_ILOCK_EXCL);
1014
1015	if (!xfs_is_reflink_inode(src)) {
1016		trace_xfs_reflink_set_inode_flag(src);
1017		xfs_trans_ijoin(tp, src, XFS_ILOCK_EXCL);
1018		src->i_diflags2 |= XFS_DIFLAG2_REFLINK;
1019		xfs_trans_log_inode(tp, src, XFS_ILOG_CORE);
1020		xfs_ifork_init_cow(src);
1021	} else
1022		xfs_iunlock(src, XFS_ILOCK_EXCL);
1023
1024	if (src->i_ino == dest->i_ino)
1025		goto commit_flags;
1026
1027	if (!xfs_is_reflink_inode(dest)) {
1028		trace_xfs_reflink_set_inode_flag(dest);
1029		xfs_trans_ijoin(tp, dest, XFS_ILOCK_EXCL);
1030		dest->i_diflags2 |= XFS_DIFLAG2_REFLINK;
1031		xfs_trans_log_inode(tp, dest, XFS_ILOG_CORE);
1032		xfs_ifork_init_cow(dest);
1033	} else
1034		xfs_iunlock(dest, XFS_ILOCK_EXCL);
1035
1036commit_flags:
1037	error = xfs_trans_commit(tp);
1038	if (error)
1039		goto out_error;
1040	return error;
1041
1042out_error:
1043	trace_xfs_reflink_set_inode_flag_error(dest, error, _RET_IP_);
1044	return error;
1045}
1046
1047/*
1048 * Update destination inode size & cowextsize hint, if necessary.
1049 */
1050int
1051xfs_reflink_update_dest(
1052	struct xfs_inode	*dest,
1053	xfs_off_t		newlen,
1054	xfs_extlen_t		cowextsize,
1055	unsigned int		remap_flags)
1056{
1057	struct xfs_mount	*mp = dest->i_mount;
1058	struct xfs_trans	*tp;
1059	int			error;
1060
1061	if (newlen <= i_size_read(VFS_I(dest)) && cowextsize == 0)
1062		return 0;
1063
1064	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
1065	if (error)
1066		goto out_error;
1067
1068	xfs_ilock(dest, XFS_ILOCK_EXCL);
1069	xfs_trans_ijoin(tp, dest, XFS_ILOCK_EXCL);
1070
1071	if (newlen > i_size_read(VFS_I(dest))) {
1072		trace_xfs_reflink_update_inode_size(dest, newlen);
1073		i_size_write(VFS_I(dest), newlen);
1074		dest->i_disk_size = newlen;
1075	}
1076
1077	if (cowextsize) {
1078		dest->i_cowextsize = cowextsize;
1079		dest->i_diflags2 |= XFS_DIFLAG2_COWEXTSIZE;
1080	}
1081
1082	xfs_trans_log_inode(tp, dest, XFS_ILOG_CORE);
1083
1084	error = xfs_trans_commit(tp);
1085	if (error)
1086		goto out_error;
1087	return error;
1088
1089out_error:
1090	trace_xfs_reflink_update_inode_size_error(dest, error, _RET_IP_);
1091	return error;
1092}
1093
1094/*
1095 * Do we have enough reserve in this AG to handle a reflink?  The refcount
1096 * btree already reserved all the space it needs, but the rmap btree can grow
1097 * infinitely, so we won't allow more reflinks when the AG is down to the
1098 * btree reserves.
1099 */
1100static int
1101xfs_reflink_ag_has_free_space(
1102	struct xfs_mount	*mp,
1103	xfs_agnumber_t		agno)
1104{
1105	struct xfs_perag	*pag;
1106	int			error = 0;
1107
1108	if (!xfs_has_rmapbt(mp))
1109		return 0;
1110
1111	pag = xfs_perag_get(mp, agno);
1112	if (xfs_ag_resv_critical(pag, XFS_AG_RESV_RMAPBT) ||
1113	    xfs_ag_resv_critical(pag, XFS_AG_RESV_METADATA))
1114		error = -ENOSPC;
1115	xfs_perag_put(pag);
1116	return error;
1117}
1118
1119/*
1120 * Remap the given extent into the file.  The dmap blockcount will be set to
1121 * the number of blocks that were actually remapped.
1122 */
1123STATIC int
1124xfs_reflink_remap_extent(
1125	struct xfs_inode	*ip,
1126	struct xfs_bmbt_irec	*dmap,
1127	xfs_off_t		new_isize)
1128{
1129	struct xfs_bmbt_irec	smap;
1130	struct xfs_mount	*mp = ip->i_mount;
1131	struct xfs_trans	*tp;
1132	xfs_off_t		newlen;
1133	int64_t			qdelta = 0;
1134	unsigned int		resblks;
1135	bool			quota_reserved = true;
1136	bool			smap_real;
1137	bool			dmap_written = xfs_bmap_is_written_extent(dmap);
1138	int			iext_delta = 0;
1139	int			nimaps;
1140	int			error;
1141
1142	/*
1143	 * Start a rolling transaction to switch the mappings.
1144	 *
1145	 * Adding a written extent to the extent map can cause a bmbt split,
1146	 * and removing a mapped extent from the extent can cause a bmbt split.
1147	 * The two operations cannot both cause a split since they operate on
1148	 * the same index in the bmap btree, so we only need a reservation for
1149	 * one bmbt split if either thing is happening.  However, we haven't
1150	 * locked the inode yet, so we reserve assuming this is the case.
1151	 *
1152	 * The first allocation call tries to reserve enough space to handle
1153	 * mapping dmap into a sparse part of the file plus the bmbt split.  We
1154	 * haven't locked the inode or read the existing mapping yet, so we do
1155	 * not know for sure that we need the space.  This should succeed most
1156	 * of the time.
1157	 *
1158	 * If the first attempt fails, try again but reserving only enough
1159	 * space to handle a bmbt split.  This is the hard minimum requirement,
1160	 * and we revisit quota reservations later when we know more about what
1161	 * we're remapping.
1162	 */
1163	resblks = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
1164	error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write,
1165			resblks + dmap->br_blockcount, 0, false, &tp);
1166	if (error == -EDQUOT || error == -ENOSPC) {
1167		quota_reserved = false;
1168		error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write,
1169				resblks, 0, false, &tp);
1170	}
1171	if (error)
1172		goto out;
1173
 
 
 
1174	/*
1175	 * Read what's currently mapped in the destination file into smap.
1176	 * If smap isn't a hole, we will have to remove it before we can add
1177	 * dmap to the destination file.
1178	 */
1179	nimaps = 1;
1180	error = xfs_bmapi_read(ip, dmap->br_startoff, dmap->br_blockcount,
1181			&smap, &nimaps, 0);
1182	if (error)
1183		goto out_cancel;
1184	ASSERT(nimaps == 1 && smap.br_startoff == dmap->br_startoff);
1185	smap_real = xfs_bmap_is_real_extent(&smap);
1186
1187	/*
1188	 * We can only remap as many blocks as the smaller of the two extent
1189	 * maps, because we can only remap one extent at a time.
1190	 */
1191	dmap->br_blockcount = min(dmap->br_blockcount, smap.br_blockcount);
1192	ASSERT(dmap->br_blockcount == smap.br_blockcount);
1193
1194	trace_xfs_reflink_remap_extent_dest(ip, &smap);
1195
1196	/*
1197	 * Two extents mapped to the same physical block must not have
1198	 * different states; that's filesystem corruption.  Move on to the next
1199	 * extent if they're both holes or both the same physical extent.
1200	 */
1201	if (dmap->br_startblock == smap.br_startblock) {
1202		if (dmap->br_state != smap.br_state) {
1203			xfs_bmap_mark_sick(ip, XFS_DATA_FORK);
1204			error = -EFSCORRUPTED;
1205		}
1206		goto out_cancel;
1207	}
1208
1209	/* If both extents are unwritten, leave them alone. */
1210	if (dmap->br_state == XFS_EXT_UNWRITTEN &&
1211	    smap.br_state == XFS_EXT_UNWRITTEN)
1212		goto out_cancel;
1213
1214	/* No reflinking if the AG of the dest mapping is low on space. */
1215	if (dmap_written) {
1216		error = xfs_reflink_ag_has_free_space(mp,
1217				XFS_FSB_TO_AGNO(mp, dmap->br_startblock));
1218		if (error)
1219			goto out_cancel;
1220	}
1221
1222	/*
1223	 * Increase quota reservation if we think the quota block counter for
1224	 * this file could increase.
1225	 *
 
 
 
 
 
 
1226	 * If we are mapping a written extent into the file, we need to have
1227	 * enough quota block count reservation to handle the blocks in that
1228	 * extent.  We log only the delta to the quota block counts, so if the
1229	 * extent we're unmapping also has blocks allocated to it, we don't
1230	 * need a quota reservation for the extent itself.
1231	 *
1232	 * Note that if we're replacing a delalloc reservation with a written
1233	 * extent, we have to take the full quota reservation because removing
1234	 * the delalloc reservation gives the block count back to the quota
1235	 * count.  This is suboptimal, but the VFS flushed the dest range
1236	 * before we started.  That should have removed all the delalloc
1237	 * reservations, but we code defensively.
1238	 *
1239	 * xfs_trans_alloc_inode above already tried to grab an even larger
1240	 * quota reservation, and kicked off a blockgc scan if it couldn't.
1241	 * If we can't get a potentially smaller quota reservation now, we're
1242	 * done.
1243	 */
1244	if (!quota_reserved && !smap_real && dmap_written) {
1245		error = xfs_trans_reserve_quota_nblks(tp, ip,
1246				dmap->br_blockcount, 0, false);
 
 
 
 
 
1247		if (error)
1248			goto out_cancel;
1249	}
1250
1251	if (smap_real)
1252		++iext_delta;
1253
1254	if (dmap_written)
1255		++iext_delta;
1256
1257	error = xfs_iext_count_extend(tp, ip, XFS_DATA_FORK, iext_delta);
1258	if (error)
1259		goto out_cancel;
1260
1261	if (smap_real) {
1262		/*
1263		 * If the extent we're unmapping is backed by storage (written
1264		 * or not), unmap the extent and drop its refcount.
1265		 */
1266		xfs_bmap_unmap_extent(tp, ip, XFS_DATA_FORK, &smap);
1267		xfs_refcount_decrease_extent(tp, &smap);
1268		qdelta -= smap.br_blockcount;
1269	} else if (smap.br_startblock == DELAYSTARTBLOCK) {
1270		int		done;
1271
1272		/*
1273		 * If the extent we're unmapping is a delalloc reservation,
1274		 * we can use the regular bunmapi function to release the
1275		 * incore state.  Dropping the delalloc reservation takes care
1276		 * of the quota reservation for us.
1277		 */
1278		error = xfs_bunmapi(NULL, ip, smap.br_startoff,
1279				smap.br_blockcount, 0, 1, &done);
1280		if (error)
1281			goto out_cancel;
1282		ASSERT(done);
1283	}
1284
1285	/*
1286	 * If the extent we're sharing is backed by written storage, increase
1287	 * its refcount and map it into the file.
1288	 */
1289	if (dmap_written) {
1290		xfs_refcount_increase_extent(tp, dmap);
1291		xfs_bmap_map_extent(tp, ip, XFS_DATA_FORK, dmap);
1292		qdelta += dmap->br_blockcount;
1293	}
1294
1295	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, qdelta);
1296
1297	/* Update dest isize if needed. */
1298	newlen = XFS_FSB_TO_B(mp, dmap->br_startoff + dmap->br_blockcount);
1299	newlen = min_t(xfs_off_t, newlen, new_isize);
1300	if (newlen > i_size_read(VFS_I(ip))) {
1301		trace_xfs_reflink_update_inode_size(ip, newlen);
1302		i_size_write(VFS_I(ip), newlen);
1303		ip->i_disk_size = newlen;
1304		xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1305	}
1306
1307	/* Commit everything and unlock. */
1308	error = xfs_trans_commit(tp);
1309	goto out_unlock;
1310
1311out_cancel:
1312	xfs_trans_cancel(tp);
1313out_unlock:
1314	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1315out:
1316	if (error)
1317		trace_xfs_reflink_remap_extent_error(ip, error, _RET_IP_);
1318	return error;
1319}
1320
1321/* Remap a range of one file to the other. */
1322int
1323xfs_reflink_remap_blocks(
1324	struct xfs_inode	*src,
1325	loff_t			pos_in,
1326	struct xfs_inode	*dest,
1327	loff_t			pos_out,
1328	loff_t			remap_len,
1329	loff_t			*remapped)
1330{
1331	struct xfs_bmbt_irec	imap;
1332	struct xfs_mount	*mp = src->i_mount;
1333	xfs_fileoff_t		srcoff = XFS_B_TO_FSBT(mp, pos_in);
1334	xfs_fileoff_t		destoff = XFS_B_TO_FSBT(mp, pos_out);
1335	xfs_filblks_t		len;
1336	xfs_filblks_t		remapped_len = 0;
1337	xfs_off_t		new_isize = pos_out + remap_len;
1338	int			nimaps;
1339	int			error = 0;
1340
1341	len = min_t(xfs_filblks_t, XFS_B_TO_FSB(mp, remap_len),
1342			XFS_MAX_FILEOFF);
1343
1344	trace_xfs_reflink_remap_blocks(src, srcoff, len, dest, destoff);
1345
1346	while (len > 0) {
1347		unsigned int	lock_mode;
1348
1349		/* Read extent from the source file */
1350		nimaps = 1;
1351		lock_mode = xfs_ilock_data_map_shared(src);
1352		error = xfs_bmapi_read(src, srcoff, len, &imap, &nimaps, 0);
1353		xfs_iunlock(src, lock_mode);
1354		if (error)
1355			break;
1356		/*
1357		 * The caller supposedly flushed all dirty pages in the source
1358		 * file range, which means that writeback should have allocated
1359		 * or deleted all delalloc reservations in that range.  If we
1360		 * find one, that's a good sign that something is seriously
1361		 * wrong here.
1362		 */
1363		ASSERT(nimaps == 1 && imap.br_startoff == srcoff);
1364		if (imap.br_startblock == DELAYSTARTBLOCK) {
1365			ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1366			xfs_bmap_mark_sick(src, XFS_DATA_FORK);
1367			error = -EFSCORRUPTED;
1368			break;
1369		}
1370
1371		trace_xfs_reflink_remap_extent_src(src, &imap);
1372
1373		/* Remap into the destination file at the given offset. */
1374		imap.br_startoff = destoff;
1375		error = xfs_reflink_remap_extent(dest, &imap, new_isize);
1376		if (error)
1377			break;
1378
1379		if (fatal_signal_pending(current)) {
1380			error = -EINTR;
1381			break;
1382		}
1383
1384		/* Advance drange/srange */
1385		srcoff += imap.br_blockcount;
1386		destoff += imap.br_blockcount;
1387		len -= imap.br_blockcount;
1388		remapped_len += imap.br_blockcount;
1389		cond_resched();
1390	}
1391
1392	if (error)
1393		trace_xfs_reflink_remap_blocks_error(dest, error, _RET_IP_);
1394	*remapped = min_t(loff_t, remap_len,
1395			  XFS_FSB_TO_B(src->i_mount, remapped_len));
1396	return error;
1397}
1398
1399/*
1400 * If we're reflinking to a point past the destination file's EOF, we must
1401 * zero any speculative post-EOF preallocations that sit between the old EOF
1402 * and the destination file offset.
1403 */
1404static int
1405xfs_reflink_zero_posteof(
1406	struct xfs_inode	*ip,
1407	loff_t			pos)
1408{
1409	loff_t			isize = i_size_read(VFS_I(ip));
1410
1411	if (pos <= isize)
1412		return 0;
1413
1414	trace_xfs_zero_eof(ip, isize, pos - isize);
1415	return xfs_zero_range(ip, isize, pos - isize, NULL);
 
1416}
1417
1418/*
1419 * Prepare two files for range cloning.  Upon a successful return both inodes
1420 * will have the iolock and mmaplock held, the page cache of the out file will
1421 * be truncated, and any leases on the out file will have been broken.  This
1422 * function borrows heavily from xfs_file_aio_write_checks.
1423 *
1424 * The VFS allows partial EOF blocks to "match" for dedupe even though it hasn't
1425 * checked that the bytes beyond EOF physically match. Hence we cannot use the
1426 * EOF block in the source dedupe range because it's not a complete block match,
1427 * hence can introduce a corruption into the file that has it's block replaced.
1428 *
1429 * In similar fashion, the VFS file cloning also allows partial EOF blocks to be
1430 * "block aligned" for the purposes of cloning entire files.  However, if the
1431 * source file range includes the EOF block and it lands within the existing EOF
1432 * of the destination file, then we can expose stale data from beyond the source
1433 * file EOF in the destination file.
1434 *
1435 * XFS doesn't support partial block sharing, so in both cases we have check
1436 * these cases ourselves. For dedupe, we can simply round the length to dedupe
1437 * down to the previous whole block and ignore the partial EOF block. While this
1438 * means we can't dedupe the last block of a file, this is an acceptible
1439 * tradeoff for simplicity on implementation.
1440 *
1441 * For cloning, we want to share the partial EOF block if it is also the new EOF
1442 * block of the destination file. If the partial EOF block lies inside the
1443 * existing destination EOF, then we have to abort the clone to avoid exposing
1444 * stale data in the destination file. Hence we reject these clone attempts with
1445 * -EINVAL in this case.
1446 */
1447int
1448xfs_reflink_remap_prep(
1449	struct file		*file_in,
1450	loff_t			pos_in,
1451	struct file		*file_out,
1452	loff_t			pos_out,
1453	loff_t			*len,
1454	unsigned int		remap_flags)
1455{
1456	struct inode		*inode_in = file_inode(file_in);
1457	struct xfs_inode	*src = XFS_I(inode_in);
1458	struct inode		*inode_out = file_inode(file_out);
1459	struct xfs_inode	*dest = XFS_I(inode_out);
1460	int			ret;
1461
1462	/* Lock both files against IO */
1463	ret = xfs_ilock2_io_mmap(src, dest);
1464	if (ret)
1465		return ret;
1466
1467	/* Check file eligibility and prepare for block sharing. */
1468	ret = -EINVAL;
1469	/* Don't reflink realtime inodes */
1470	if (XFS_IS_REALTIME_INODE(src) || XFS_IS_REALTIME_INODE(dest))
1471		goto out_unlock;
1472
1473	/* Don't share DAX file data with non-DAX file. */
1474	if (IS_DAX(inode_in) != IS_DAX(inode_out))
1475		goto out_unlock;
1476
1477	if (!IS_DAX(inode_in))
1478		ret = generic_remap_file_range_prep(file_in, pos_in, file_out,
1479				pos_out, len, remap_flags);
1480	else
1481		ret = dax_remap_file_range_prep(file_in, pos_in, file_out,
1482				pos_out, len, remap_flags, &xfs_read_iomap_ops);
1483	if (ret || *len == 0)
1484		goto out_unlock;
1485
1486	/* Attach dquots to dest inode before changing block map */
1487	ret = xfs_qm_dqattach(dest);
1488	if (ret)
1489		goto out_unlock;
1490
1491	/*
1492	 * Zero existing post-eof speculative preallocations in the destination
1493	 * file.
1494	 */
1495	ret = xfs_reflink_zero_posteof(dest, pos_out);
1496	if (ret)
1497		goto out_unlock;
1498
1499	/* Set flags and remap blocks. */
1500	ret = xfs_reflink_set_inode_flag(src, dest);
1501	if (ret)
1502		goto out_unlock;
1503
1504	/*
1505	 * If pos_out > EOF, we may have dirtied blocks between EOF and
1506	 * pos_out. In that case, we need to extend the flush and unmap to cover
1507	 * from EOF to the end of the copy length.
1508	 */
1509	if (pos_out > XFS_ISIZE(dest)) {
1510		loff_t	flen = *len + (pos_out - XFS_ISIZE(dest));
1511		ret = xfs_flush_unmap_range(dest, XFS_ISIZE(dest), flen);
1512	} else {
1513		ret = xfs_flush_unmap_range(dest, pos_out, *len);
1514	}
1515	if (ret)
1516		goto out_unlock;
1517
1518	xfs_iflags_set(src, XFS_IREMAPPING);
1519	if (inode_in != inode_out)
1520		xfs_ilock_demote(src, XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL);
1521
1522	return 0;
1523out_unlock:
1524	xfs_iunlock2_io_mmap(src, dest);
1525	return ret;
1526}
1527
1528/* Does this inode need the reflink flag? */
1529int
1530xfs_reflink_inode_has_shared_extents(
1531	struct xfs_trans		*tp,
1532	struct xfs_inode		*ip,
1533	bool				*has_shared)
1534{
1535	struct xfs_bmbt_irec		got;
1536	struct xfs_mount		*mp = ip->i_mount;
1537	struct xfs_ifork		*ifp;
 
 
 
 
 
1538	struct xfs_iext_cursor		icur;
1539	bool				found;
1540	int				error;
1541
1542	ifp = xfs_ifork_ptr(ip, XFS_DATA_FORK);
1543	error = xfs_iread_extents(tp, ip, XFS_DATA_FORK);
1544	if (error)
1545		return error;
 
 
1546
1547	*has_shared = false;
1548	found = xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got);
1549	while (found) {
1550		struct xfs_perag	*pag;
1551		xfs_agblock_t		agbno;
1552		xfs_extlen_t		aglen;
1553		xfs_agblock_t		rbno;
1554		xfs_extlen_t		rlen;
1555
1556		if (isnullstartblock(got.br_startblock) ||
1557		    got.br_state != XFS_EXT_NORM)
1558			goto next;
1559
1560		pag = xfs_perag_get(mp, XFS_FSB_TO_AGNO(mp, got.br_startblock));
1561		agbno = XFS_FSB_TO_AGBNO(mp, got.br_startblock);
1562		aglen = got.br_blockcount;
1563		error = xfs_reflink_find_shared(pag, tp, agbno, aglen,
 
1564				&rbno, &rlen, false);
1565		xfs_perag_put(pag);
1566		if (error)
1567			return error;
1568
1569		/* Is there still a shared block here? */
1570		if (rbno != NULLAGBLOCK) {
1571			*has_shared = true;
1572			return 0;
1573		}
1574next:
1575		found = xfs_iext_next_extent(ifp, &icur, &got);
1576	}
1577
1578	return 0;
1579}
1580
1581/*
1582 * Clear the inode reflink flag if there are no shared extents.
1583 *
1584 * The caller is responsible for joining the inode to the transaction passed in.
1585 * The inode will be joined to the transaction that is returned to the caller.
1586 */
1587int
1588xfs_reflink_clear_inode_flag(
1589	struct xfs_inode	*ip,
1590	struct xfs_trans	**tpp)
1591{
1592	bool			needs_flag;
1593	int			error = 0;
1594
1595	ASSERT(xfs_is_reflink_inode(ip));
1596
1597	if (!xfs_can_free_cowblocks(ip))
1598		return 0;
1599
1600	error = xfs_reflink_inode_has_shared_extents(*tpp, ip, &needs_flag);
1601	if (error || needs_flag)
1602		return error;
1603
1604	/*
1605	 * We didn't find any shared blocks so turn off the reflink flag.
1606	 * First, get rid of any leftover CoW mappings.
1607	 */
1608	error = xfs_reflink_cancel_cow_blocks(ip, tpp, 0, XFS_MAX_FILEOFF,
1609			true);
1610	if (error)
1611		return error;
1612
1613	/* Clear the inode flag. */
1614	trace_xfs_reflink_unset_inode_flag(ip);
1615	ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1616	xfs_inode_clear_cowblocks_tag(ip);
1617	xfs_trans_log_inode(*tpp, ip, XFS_ILOG_CORE);
1618
1619	return error;
1620}
1621
1622/*
1623 * Clear the inode reflink flag if there are no shared extents and the size
1624 * hasn't changed.
1625 */
1626STATIC int
1627xfs_reflink_try_clear_inode_flag(
1628	struct xfs_inode	*ip)
1629{
1630	struct xfs_mount	*mp = ip->i_mount;
1631	struct xfs_trans	*tp;
1632	int			error = 0;
1633
1634	/* Start a rolling transaction to remove the mappings */
1635	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
1636	if (error)
1637		return error;
1638
1639	xfs_ilock(ip, XFS_ILOCK_EXCL);
1640	xfs_trans_ijoin(tp, ip, 0);
1641
1642	error = xfs_reflink_clear_inode_flag(ip, &tp);
1643	if (error)
1644		goto cancel;
1645
1646	error = xfs_trans_commit(tp);
1647	if (error)
1648		goto out;
1649
1650	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1651	return 0;
1652cancel:
1653	xfs_trans_cancel(tp);
1654out:
1655	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1656	return error;
1657}
1658
1659/*
1660 * Pre-COW all shared blocks within a given byte range of a file and turn off
1661 * the reflink flag if we unshare all of the file's blocks.
1662 */
1663int
1664xfs_reflink_unshare(
1665	struct xfs_inode	*ip,
1666	xfs_off_t		offset,
1667	xfs_off_t		len)
1668{
1669	struct inode		*inode = VFS_I(ip);
1670	int			error;
1671
1672	if (!xfs_is_reflink_inode(ip))
1673		return 0;
1674
1675	trace_xfs_reflink_unshare(ip, offset, len);
1676
1677	inode_dio_wait(inode);
1678
1679	if (IS_DAX(inode))
1680		error = dax_file_unshare(inode, offset, len,
1681				&xfs_dax_write_iomap_ops);
1682	else
1683		error = iomap_file_unshare(inode, offset, len,
1684				&xfs_buffered_write_iomap_ops);
1685	if (error)
1686		goto out;
1687
1688	error = filemap_write_and_wait_range(inode->i_mapping, offset,
1689			offset + len - 1);
1690	if (error)
1691		goto out;
1692
1693	/* Turn off the reflink flag if possible. */
1694	error = xfs_reflink_try_clear_inode_flag(ip);
1695	if (error)
1696		goto out;
1697	return 0;
1698
1699out:
1700	trace_xfs_reflink_unshare_error(ip, error, _RET_IP_);
1701	return error;
1702}