Linux Audio

Check our new training course

Loading...
v4.17
 
   1/*
   2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   3 * Copyright (c) 2016 Christoph Hellwig.
   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 <linux/iomap.h>
  20#include "xfs.h"
  21#include "xfs_fs.h"
  22#include "xfs_shared.h"
  23#include "xfs_format.h"
  24#include "xfs_log_format.h"
  25#include "xfs_trans_resv.h"
  26#include "xfs_mount.h"
  27#include "xfs_defer.h"
  28#include "xfs_inode.h"
  29#include "xfs_btree.h"
  30#include "xfs_bmap_btree.h"
  31#include "xfs_bmap.h"
  32#include "xfs_bmap_util.h"
  33#include "xfs_errortag.h"
  34#include "xfs_error.h"
  35#include "xfs_trans.h"
  36#include "xfs_trans_space.h"
  37#include "xfs_inode_item.h"
  38#include "xfs_iomap.h"
  39#include "xfs_trace.h"
  40#include "xfs_icache.h"
  41#include "xfs_quota.h"
 
  42#include "xfs_dquot_item.h"
  43#include "xfs_dquot.h"
  44#include "xfs_reflink.h"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  45
 
 
 
  46
  47#define XFS_WRITEIO_ALIGN(mp,off)	(((off) >> mp->m_writeio_log) \
  48						<< mp->m_writeio_log)
 
  49
  50void
  51xfs_bmbt_to_iomap(
  52	struct xfs_inode	*ip,
  53	struct iomap		*iomap,
  54	struct xfs_bmbt_irec	*imap)
 
 
 
  55{
  56	struct xfs_mount	*mp = ip->i_mount;
 
 
 
 
 
 
  57
  58	if (imap->br_startblock == HOLESTARTBLOCK) {
  59		iomap->addr = IOMAP_NULL_ADDR;
  60		iomap->type = IOMAP_HOLE;
  61	} else if (imap->br_startblock == DELAYSTARTBLOCK) {
 
  62		iomap->addr = IOMAP_NULL_ADDR;
  63		iomap->type = IOMAP_DELALLOC;
  64	} else {
  65		iomap->addr = BBTOB(xfs_fsb_to_db(ip, imap->br_startblock));
 
 
 
 
 
  66		if (imap->br_state == XFS_EXT_UNWRITTEN)
  67			iomap->type = IOMAP_UNWRITTEN;
  68		else
  69			iomap->type = IOMAP_MAPPED;
 
 
 
 
 
 
 
 
 
  70	}
  71	iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff);
  72	iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount);
  73	iomap->bdev = xfs_find_bdev_for_inode(VFS_I(ip));
  74	iomap->dax_dev = xfs_find_daxdev_for_inode(VFS_I(ip));
 
 
 
 
 
 
 
 
 
 
 
  75}
  76
  77xfs_extlen_t
  78xfs_eof_alignment(
  79	struct xfs_inode	*ip,
  80	xfs_extlen_t		extsize)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  81{
  82	struct xfs_mount	*mp = ip->i_mount;
  83	xfs_extlen_t		align = 0;
  84
  85	if (!XFS_IS_REALTIME_INODE(ip)) {
  86		/*
  87		 * Round up the allocation request to a stripe unit
  88		 * (m_dalign) boundary if the file size is >= stripe unit
  89		 * size, and we are allocating past the allocation eof.
  90		 *
  91		 * If mounted with the "-o swalloc" option the alignment is
  92		 * increased from the strip unit size to the stripe width.
  93		 */
  94		if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
  95			align = mp->m_swidth;
  96		else if (mp->m_dalign)
  97			align = mp->m_dalign;
  98
  99		if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
 100			align = 0;
 101	}
 102
 103	/*
 104	 * Always round up the allocation request to an extent boundary
 105	 * (when file on a real-time subvolume or has di_extsize hint).
 106	 */
 107	if (extsize) {
 108		if (align)
 109			align = roundup_64(align, extsize);
 110		else
 111			align = extsize;
 112	}
 113
 114	return align;
 115}
 116
 117STATIC int
 
 
 
 
 118xfs_iomap_eof_align_last_fsb(
 119	struct xfs_inode	*ip,
 120	xfs_extlen_t		extsize,
 121	xfs_fileoff_t		*last_fsb)
 122{
 123	xfs_extlen_t		align = xfs_eof_alignment(ip, extsize);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 124
 125	if (align) {
 126		xfs_fileoff_t	new_last_fsb = roundup_64(*last_fsb, align);
 127		int		eof, error;
 128
 129		error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
 130		if (error)
 131			return error;
 132		if (eof)
 133			*last_fsb = new_last_fsb;
 134	}
 135	return 0;
 136}
 137
 138STATIC int
 139xfs_alert_fsblock_zero(
 140	xfs_inode_t	*ip,
 141	xfs_bmbt_irec_t	*imap)
 142{
 143	xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
 144			"Access to block zero in inode %llu "
 145			"start_block: %llx start_off: %llx "
 146			"blkcnt: %llx extent-state: %x",
 147		(unsigned long long)ip->i_ino,
 148		(unsigned long long)imap->br_startblock,
 149		(unsigned long long)imap->br_startoff,
 150		(unsigned long long)imap->br_blockcount,
 151		imap->br_state);
 152	return -EFSCORRUPTED;
 153}
 154
 155int
 156xfs_iomap_write_direct(
 157	xfs_inode_t	*ip,
 158	xfs_off_t	offset,
 159	size_t		count,
 160	xfs_bmbt_irec_t *imap,
 161	int		nmaps)
 
 162{
 163	xfs_mount_t	*mp = ip->i_mount;
 164	xfs_fileoff_t	offset_fsb;
 165	xfs_fileoff_t	last_fsb;
 166	xfs_filblks_t	count_fsb, resaligned;
 167	xfs_fsblock_t	firstfsb;
 168	xfs_extlen_t	extsz;
 169	int		nimaps;
 170	int		quota_flag;
 171	int		rt;
 172	xfs_trans_t	*tp;
 173	struct xfs_defer_ops dfops;
 174	uint		qblocks, resblks, resrtextents;
 175	int		error;
 176	int		lockmode;
 177	int		bmapi_flags = XFS_BMAPI_PREALLOC;
 178	uint		tflags = 0;
 179
 180	rt = XFS_IS_REALTIME_INODE(ip);
 181	extsz = xfs_get_extsz_hint(ip);
 182	lockmode = XFS_ILOCK_SHARED;	/* locked by caller */
 183
 184	ASSERT(xfs_isilocked(ip, lockmode));
 185
 186	offset_fsb = XFS_B_TO_FSBT(mp, offset);
 187	last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
 188	if ((offset + count) > XFS_ISIZE(ip)) {
 189		/*
 190		 * Assert that the in-core extent list is present since this can
 191		 * call xfs_iread_extents() and we only have the ilock shared.
 192		 * This should be safe because the lock was held around a bmapi
 193		 * call in the caller and we only need it to access the in-core
 194		 * list.
 195		 */
 196		ASSERT(XFS_IFORK_PTR(ip, XFS_DATA_FORK)->if_flags &
 197								XFS_IFEXTENTS);
 198		error = xfs_iomap_eof_align_last_fsb(ip, extsz, &last_fsb);
 199		if (error)
 200			goto out_unlock;
 201	} else {
 202		if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
 203			last_fsb = MIN(last_fsb, (xfs_fileoff_t)
 204					imap->br_blockcount +
 205					imap->br_startoff);
 206	}
 207	count_fsb = last_fsb - offset_fsb;
 208	ASSERT(count_fsb > 0);
 209	resaligned = xfs_aligned_fsb_count(offset_fsb, count_fsb, extsz);
 210
 211	if (unlikely(rt)) {
 212		resrtextents = qblocks = resaligned;
 213		resrtextents /= mp->m_sb.sb_rextsize;
 214		resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
 215		quota_flag = XFS_QMOPT_RES_RTBLKS;
 216	} else {
 217		resrtextents = 0;
 218		resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
 219		quota_flag = XFS_QMOPT_RES_REGBLKS;
 220	}
 221
 222	/*
 223	 * Drop the shared lock acquired by the caller, attach the dquot if
 224	 * necessary and move on to transaction setup.
 225	 */
 226	xfs_iunlock(ip, lockmode);
 227	error = xfs_qm_dqattach(ip, 0);
 228	if (error)
 229		return error;
 230
 231	/*
 232	 * For DAX, we do not allocate unwritten extents, but instead we zero
 233	 * the block before we commit the transaction.  Ideally we'd like to do
 234	 * this outside the transaction context, but if we commit and then crash
 235	 * we may not have zeroed the blocks and this will be exposed on
 236	 * recovery of the allocation. Hence we must zero before commit.
 237	 *
 238	 * Further, if we are mapping unwritten extents here, we need to zero
 239	 * and convert them to written so that we don't need an unwritten extent
 240	 * callback for DAX. This also means that we need to be able to dip into
 241	 * the reserve block pool for bmbt block allocation if there is no space
 242	 * left but we need to do unwritten extent conversion.
 243	 */
 244	if (IS_DAX(VFS_I(ip))) {
 245		bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
 246		if (imap->br_state == XFS_EXT_UNWRITTEN) {
 247			tflags |= XFS_TRANS_RESERVE;
 248			resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
 
 249		}
 250	}
 251	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, resrtextents,
 252			tflags, &tp);
 
 253	if (error)
 254		return error;
 255
 256	lockmode = XFS_ILOCK_EXCL;
 257	xfs_ilock(ip, lockmode);
 258
 259	error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
 260	if (error)
 261		goto out_trans_cancel;
 262
 263	xfs_trans_ijoin(tp, ip, 0);
 264
 265	/*
 266	 * From this point onwards we overwrite the imap pointer that the
 267	 * caller gave to us.
 268	 */
 269	xfs_defer_init(&dfops, &firstfsb);
 270	nimaps = 1;
 271	error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
 272				bmapi_flags, &firstfsb, resblks, imap,
 273				&nimaps, &dfops);
 274	if (error)
 275		goto out_bmap_cancel;
 276
 277	/*
 278	 * Complete the transaction
 279	 */
 280	error = xfs_defer_finish(&tp, &dfops);
 281	if (error)
 282		goto out_bmap_cancel;
 283
 284	error = xfs_trans_commit(tp);
 285	if (error)
 286		goto out_unlock;
 287
 288	/*
 289	 * Copy any maps to caller's array and return any error.
 290	 */
 291	if (nimaps == 0) {
 292		error = -ENOSPC;
 293		goto out_unlock;
 294	}
 295
 296	if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
 297		error = xfs_alert_fsblock_zero(ip, imap);
 
 298
 299out_unlock:
 300	xfs_iunlock(ip, lockmode);
 
 301	return error;
 302
 303out_bmap_cancel:
 304	xfs_defer_cancel(&dfops);
 305	xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
 306out_trans_cancel:
 307	xfs_trans_cancel(tp);
 308	goto out_unlock;
 309}
 310
 311STATIC bool
 312xfs_quota_need_throttle(
 313	struct xfs_inode *ip,
 314	int type,
 315	xfs_fsblock_t alloc_blocks)
 316{
 317	struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
 
 
 318
 319	if (!dq || !xfs_this_quota_on(ip->i_mount, type))
 320		return false;
 321
 
 
 
 
 
 
 
 
 322	/* no hi watermark, no throttle */
 323	if (!dq->q_prealloc_hi_wmark)
 324		return false;
 325
 326	/* under the lo watermark, no throttle */
 327	if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark)
 328		return false;
 329
 330	return true;
 331}
 332
 333STATIC void
 334xfs_quota_calc_throttle(
 335	struct xfs_inode *ip,
 336	int type,
 337	xfs_fsblock_t *qblocks,
 338	int *qshift,
 339	int64_t	*qfreesp)
 340{
 341	int64_t freesp;
 342	int shift = 0;
 343	struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
 
 
 
 
 
 
 
 
 
 
 
 
 
 344
 345	/* no dq, or over hi wmark, squash the prealloc completely */
 346	if (!dq || dq->q_res_bcount >= dq->q_prealloc_hi_wmark) {
 347		*qblocks = 0;
 348		*qfreesp = 0;
 349		return;
 350	}
 351
 352	freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount;
 353	if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
 354		shift = 2;
 355		if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
 356			shift += 2;
 357		if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
 358			shift += 2;
 359	}
 360
 361	if (freesp < *qfreesp)
 362		*qfreesp = freesp;
 363
 364	/* only overwrite the throttle values if we are more aggressive */
 365	if ((freesp >> shift) < (*qblocks >> *qshift)) {
 366		*qblocks = freesp;
 367		*qshift = shift;
 368	}
 369}
 370
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 371/*
 372 * If we are doing a write at the end of the file and there are no allocations
 373 * past this one, then extend the allocation out to the file system's write
 374 * iosize.
 375 *
 376 * If we don't have a user specified preallocation size, dynamically increase
 377 * the preallocation size as the size of the file grows.  Cap the maximum size
 378 * at a single extent or less if the filesystem is near full. The closer the
 379 * filesystem is to full, the smaller the maximum prealocation.
 380 *
 381 * As an exception we don't do any preallocation at all if the file is smaller
 382 * than the minimum preallocation and we are using the default dynamic
 383 * preallocation scheme, as it is likely this is the only write to the file that
 384 * is going to be done.
 385 *
 386 * We clean up any extra space left over when the file is closed in
 387 * xfs_inactive().
 388 */
 389STATIC xfs_fsblock_t
 390xfs_iomap_prealloc_size(
 391	struct xfs_inode	*ip,
 
 392	loff_t			offset,
 393	loff_t			count,
 394	struct xfs_iext_cursor	*icur)
 395{
 
 
 396	struct xfs_mount	*mp = ip->i_mount;
 397	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
 398	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
 399	struct xfs_bmbt_irec	prev;
 400	int			shift = 0;
 401	int64_t			freesp;
 402	xfs_fsblock_t		qblocks;
 403	int			qshift = 0;
 404	xfs_fsblock_t		alloc_blocks = 0;
 
 
 
 405
 406	if (offset + count <= XFS_ISIZE(ip))
 407		return 0;
 408
 409	if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) &&
 410	    (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks)))
 
 
 411		return 0;
 412
 413	/*
 414	 * If an explicit allocsize is set, the file is small, or we
 415	 * are writing behind a hole, then use the minimum prealloc:
 416	 */
 417	if ((mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) ||
 418	    XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign) ||
 419	    !xfs_iext_peek_prev_extent(ifp, icur, &prev) ||
 420	    prev.br_startoff + prev.br_blockcount < offset_fsb)
 421		return mp->m_writeio_blocks;
 422
 423	/*
 424	 * Determine the initial size of the preallocation. We are beyond the
 425	 * current EOF here, but we need to take into account whether this is
 426	 * a sparse write or an extending write when determining the
 427	 * preallocation size.  Hence we need to look up the extent that ends
 428	 * at the current write offset and use the result to determine the
 429	 * preallocation size.
 430	 *
 431	 * If the extent is a hole, then preallocation is essentially disabled.
 432	 * Otherwise we take the size of the preceding data extent as the basis
 433	 * for the preallocation size. If the size of the extent is greater than
 434	 * half the maximum extent length, then use the current offset as the
 435	 * basis. This ensures that for large files the preallocation size
 436	 * always extends to MAXEXTLEN rather than falling short due to things
 437	 * like stripe unit/width alignment of real extents.
 
 
 
 
 
 
 
 438	 */
 439	if (prev.br_blockcount <= (MAXEXTLEN >> 1))
 440		alloc_blocks = prev.br_blockcount << 1;
 441	else
 442		alloc_blocks = XFS_B_TO_FSB(mp, offset);
 443	if (!alloc_blocks)
 444		goto check_writeio;
 445	qblocks = alloc_blocks;
 446
 447	/*
 448	 * MAXEXTLEN is not a power of two value but we round the prealloc down
 449	 * to the nearest power of two value after throttling. To prevent the
 450	 * round down from unconditionally reducing the maximum supported prealloc
 451	 * size, we round up first, apply appropriate throttling, round down and
 452	 * cap the value to MAXEXTLEN.
 453	 */
 454	alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN),
 455				       alloc_blocks);
 456
 457	freesp = percpu_counter_read_positive(&mp->m_fdblocks);
 458	if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
 459		shift = 2;
 460		if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
 461			shift++;
 462		if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
 463			shift++;
 464		if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
 465			shift++;
 466		if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
 467			shift++;
 468	}
 469
 470	/*
 471	 * Check each quota to cap the prealloc size, provide a shift value to
 472	 * throttle with and adjust amount of available space.
 473	 */
 474	if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks))
 475		xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift,
 476					&freesp);
 477	if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks))
 478		xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift,
 479					&freesp);
 480	if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks))
 481		xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift,
 482					&freesp);
 483
 484	/*
 485	 * The final prealloc size is set to the minimum of free space available
 486	 * in each of the quotas and the overall filesystem.
 487	 *
 488	 * The shift throttle value is set to the maximum value as determined by
 489	 * the global low free space values and per-quota low free space values.
 490	 */
 491	alloc_blocks = MIN(alloc_blocks, qblocks);
 492	shift = MAX(shift, qshift);
 493
 494	if (shift)
 495		alloc_blocks >>= shift;
 496	/*
 497	 * rounddown_pow_of_two() returns an undefined result if we pass in
 498	 * alloc_blocks = 0.
 499	 */
 500	if (alloc_blocks)
 501		alloc_blocks = rounddown_pow_of_two(alloc_blocks);
 502	if (alloc_blocks > MAXEXTLEN)
 503		alloc_blocks = MAXEXTLEN;
 504
 505	/*
 506	 * If we are still trying to allocate more space than is
 507	 * available, squash the prealloc hard. This can happen if we
 508	 * have a large file on a small filesystem and the above
 509	 * lowspace thresholds are smaller than MAXEXTLEN.
 510	 */
 511	while (alloc_blocks && alloc_blocks >= freesp)
 512		alloc_blocks >>= 4;
 513check_writeio:
 514	if (alloc_blocks < mp->m_writeio_blocks)
 515		alloc_blocks = mp->m_writeio_blocks;
 516	trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
 517				      mp->m_writeio_blocks);
 518	return alloc_blocks;
 519}
 520
 521static int
 522xfs_file_iomap_begin_delay(
 523	struct inode		*inode,
 524	loff_t			offset,
 525	loff_t			count,
 526	struct iomap		*iomap)
 527{
 528	struct xfs_inode	*ip = XFS_I(inode);
 529	struct xfs_mount	*mp = ip->i_mount;
 530	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
 531	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
 532	xfs_fileoff_t		maxbytes_fsb =
 533		XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
 534	xfs_fileoff_t		end_fsb;
 535	int			error = 0, eof = 0;
 536	struct xfs_bmbt_irec	got;
 537	struct xfs_iext_cursor	icur;
 538	xfs_fsblock_t		prealloc_blocks = 0;
 539
 540	ASSERT(!XFS_IS_REALTIME_INODE(ip));
 541	ASSERT(!xfs_get_extsz_hint(ip));
 542
 543	xfs_ilock(ip, XFS_ILOCK_EXCL);
 544
 545	if (unlikely(XFS_TEST_ERROR(
 546	    (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
 547	     XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
 548	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
 549		XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
 550		error = -EFSCORRUPTED;
 551		goto out_unlock;
 552	}
 553
 554	XFS_STATS_INC(mp, xs_blk_mapw);
 555
 556	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
 557		error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
 558		if (error)
 559			goto out_unlock;
 560	}
 561
 562	eof = !xfs_iext_lookup_extent(ip, ifp, offset_fsb, &icur, &got);
 563	if (!eof && got.br_startoff <= offset_fsb) {
 564		if (xfs_is_reflink_inode(ip)) {
 565			bool		shared;
 566
 567			end_fsb = min(XFS_B_TO_FSB(mp, offset + count),
 568					maxbytes_fsb);
 569			xfs_trim_extent(&got, offset_fsb, end_fsb - offset_fsb);
 570			error = xfs_reflink_reserve_cow(ip, &got, &shared);
 571			if (error)
 572				goto out_unlock;
 573		}
 574
 575		trace_xfs_iomap_found(ip, offset, count, 0, &got);
 576		goto done;
 577	}
 578
 579	error = xfs_qm_dqattach_locked(ip, 0);
 580	if (error)
 581		goto out_unlock;
 582
 583	/*
 584	 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES pages
 585	 * to keep the chunks of work done where somewhat symmetric with the
 586	 * work writeback does. This is a completely arbitrary number pulled
 587	 * out of thin air as a best guess for initial testing.
 588	 *
 589	 * Note that the values needs to be less than 32-bits wide until
 590	 * the lower level functions are updated.
 591	 */
 592	count = min_t(loff_t, count, 1024 * PAGE_SIZE);
 593	end_fsb = min(XFS_B_TO_FSB(mp, offset + count), maxbytes_fsb);
 594
 595	if (eof) {
 596		prealloc_blocks = xfs_iomap_prealloc_size(ip, offset, count,
 597				&icur);
 598		if (prealloc_blocks) {
 599			xfs_extlen_t	align;
 600			xfs_off_t	end_offset;
 601			xfs_fileoff_t	p_end_fsb;
 602
 603			end_offset = XFS_WRITEIO_ALIGN(mp, offset + count - 1);
 604			p_end_fsb = XFS_B_TO_FSBT(mp, end_offset) +
 605					prealloc_blocks;
 606
 607			align = xfs_eof_alignment(ip, 0);
 608			if (align)
 609				p_end_fsb = roundup_64(p_end_fsb, align);
 610
 611			p_end_fsb = min(p_end_fsb, maxbytes_fsb);
 612			ASSERT(p_end_fsb > offset_fsb);
 613			prealloc_blocks = p_end_fsb - end_fsb;
 614		}
 615	}
 616
 617retry:
 618	error = xfs_bmapi_reserve_delalloc(ip, XFS_DATA_FORK, offset_fsb,
 619			end_fsb - offset_fsb, prealloc_blocks, &got, &icur,
 620			eof);
 621	switch (error) {
 622	case 0:
 623		break;
 624	case -ENOSPC:
 625	case -EDQUOT:
 626		/* retry without any preallocation */
 627		trace_xfs_delalloc_enospc(ip, offset, count);
 628		if (prealloc_blocks) {
 629			prealloc_blocks = 0;
 630			goto retry;
 631		}
 632		/*FALLTHRU*/
 633	default:
 634		goto out_unlock;
 635	}
 636
 637	/*
 638	 * Flag newly allocated delalloc blocks with IOMAP_F_NEW so we punch
 639	 * them out if the write happens to fail.
 640	 */
 641	iomap->flags = IOMAP_F_NEW;
 642	trace_xfs_iomap_alloc(ip, offset, count, 0, &got);
 643done:
 644	if (isnullstartblock(got.br_startblock))
 645		got.br_startblock = DELAYSTARTBLOCK;
 646
 647	if (!got.br_startblock) {
 648		error = xfs_alert_fsblock_zero(ip, &got);
 649		if (error)
 650			goto out_unlock;
 651	}
 652
 653	xfs_bmbt_to_iomap(ip, iomap, &got);
 654
 655out_unlock:
 656	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 657	return error;
 658}
 659
 660/*
 661 * Pass in a delayed allocate extent, convert it to real extents;
 662 * return to the caller the extent we create which maps on top of
 663 * the originating callers request.
 664 *
 665 * Called without a lock on the inode.
 666 *
 667 * We no longer bother to look at the incoming map - all we have to
 668 * guarantee is that whatever we allocate fills the required range.
 669 */
 670int
 671xfs_iomap_write_allocate(
 672	xfs_inode_t	*ip,
 673	int		whichfork,
 674	xfs_off_t	offset,
 675	xfs_bmbt_irec_t *imap)
 676{
 677	xfs_mount_t	*mp = ip->i_mount;
 678	xfs_fileoff_t	offset_fsb, last_block;
 679	xfs_fileoff_t	end_fsb, map_start_fsb;
 680	xfs_fsblock_t	first_block;
 681	struct xfs_defer_ops	dfops;
 682	xfs_filblks_t	count_fsb;
 683	xfs_trans_t	*tp;
 684	int		nimaps;
 685	int		error = 0;
 686	int		flags = XFS_BMAPI_DELALLOC;
 687	int		nres;
 688
 689	if (whichfork == XFS_COW_FORK)
 690		flags |= XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC;
 691
 692	/*
 693	 * Make sure that the dquots are there.
 694	 */
 695	error = xfs_qm_dqattach(ip, 0);
 696	if (error)
 697		return error;
 698
 699	offset_fsb = XFS_B_TO_FSBT(mp, offset);
 700	count_fsb = imap->br_blockcount;
 701	map_start_fsb = imap->br_startoff;
 702
 703	XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
 704
 705	while (count_fsb != 0) {
 706		/*
 707		 * Set up a transaction with which to allocate the
 708		 * backing store for the file.  Do allocations in a
 709		 * loop until we get some space in the range we are
 710		 * interested in.  The other space that might be allocated
 711		 * is in the delayed allocation extent on which we sit
 712		 * but before our buffer starts.
 713		 */
 714		nimaps = 0;
 715		while (nimaps == 0) {
 716			nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
 717			/*
 718			 * We have already reserved space for the extent and any
 719			 * indirect blocks when creating the delalloc extent,
 720			 * there is no need to reserve space in this transaction
 721			 * again.
 722			 */
 723			error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0,
 724					0, XFS_TRANS_RESERVE, &tp);
 725			if (error)
 726				return error;
 727
 728			xfs_ilock(ip, XFS_ILOCK_EXCL);
 729			xfs_trans_ijoin(tp, ip, 0);
 730
 731			xfs_defer_init(&dfops, &first_block);
 732
 733			/*
 734			 * it is possible that the extents have changed since
 735			 * we did the read call as we dropped the ilock for a
 736			 * while. We have to be careful about truncates or hole
 737			 * punchs here - we are not allowed to allocate
 738			 * non-delalloc blocks here.
 739			 *
 740			 * The only protection against truncation is the pages
 741			 * for the range we are being asked to convert are
 742			 * locked and hence a truncate will block on them
 743			 * first.
 744			 *
 745			 * As a result, if we go beyond the range we really
 746			 * need and hit an delalloc extent boundary followed by
 747			 * a hole while we have excess blocks in the map, we
 748			 * will fill the hole incorrectly and overrun the
 749			 * transaction reservation.
 750			 *
 751			 * Using a single map prevents this as we are forced to
 752			 * check each map we look for overlap with the desired
 753			 * range and abort as soon as we find it. Also, given
 754			 * that we only return a single map, having one beyond
 755			 * what we can return is probably a bit silly.
 756			 *
 757			 * We also need to check that we don't go beyond EOF;
 758			 * this is a truncate optimisation as a truncate sets
 759			 * the new file size before block on the pages we
 760			 * currently have locked under writeback. Because they
 761			 * are about to be tossed, we don't need to write them
 762			 * back....
 763			 */
 764			nimaps = 1;
 765			end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
 766			error = xfs_bmap_last_offset(ip, &last_block,
 767							XFS_DATA_FORK);
 768			if (error)
 769				goto trans_cancel;
 770
 771			last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
 772			if ((map_start_fsb + count_fsb) > last_block) {
 773				count_fsb = last_block - map_start_fsb;
 774				if (count_fsb == 0) {
 775					error = -EAGAIN;
 776					goto trans_cancel;
 777				}
 778			}
 779
 780			/*
 781			 * From this point onwards we overwrite the imap
 782			 * pointer that the caller gave to us.
 783			 */
 784			error = xfs_bmapi_write(tp, ip, map_start_fsb,
 785						count_fsb, flags, &first_block,
 786						nres, imap, &nimaps,
 787						&dfops);
 788			if (error)
 789				goto trans_cancel;
 790
 791			error = xfs_defer_finish(&tp, &dfops);
 792			if (error)
 793				goto trans_cancel;
 794
 795			error = xfs_trans_commit(tp);
 796			if (error)
 797				goto error0;
 798
 799			xfs_iunlock(ip, XFS_ILOCK_EXCL);
 800		}
 801
 802		/*
 803		 * See if we were able to allocate an extent that
 804		 * covers at least part of the callers request
 805		 */
 806		if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
 807			return xfs_alert_fsblock_zero(ip, imap);
 808
 809		if ((offset_fsb >= imap->br_startoff) &&
 810		    (offset_fsb < (imap->br_startoff +
 811				   imap->br_blockcount))) {
 812			XFS_STATS_INC(mp, xs_xstrat_quick);
 813			return 0;
 814		}
 815
 816		/*
 817		 * So far we have not mapped the requested part of the
 818		 * file, just surrounding data, try again.
 819		 */
 820		count_fsb -= imap->br_blockcount;
 821		map_start_fsb = imap->br_startoff + imap->br_blockcount;
 822	}
 823
 824trans_cancel:
 825	xfs_defer_cancel(&dfops);
 826	xfs_trans_cancel(tp);
 827error0:
 828	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 829	return error;
 830}
 831
 832int
 833xfs_iomap_write_unwritten(
 834	xfs_inode_t	*ip,
 835	xfs_off_t	offset,
 836	xfs_off_t	count,
 837	bool		update_isize)
 838{
 839	xfs_mount_t	*mp = ip->i_mount;
 840	xfs_fileoff_t	offset_fsb;
 841	xfs_filblks_t	count_fsb;
 842	xfs_filblks_t	numblks_fsb;
 843	xfs_fsblock_t	firstfsb;
 844	int		nimaps;
 845	xfs_trans_t	*tp;
 846	xfs_bmbt_irec_t imap;
 847	struct xfs_defer_ops dfops;
 848	struct inode	*inode = VFS_I(ip);
 849	xfs_fsize_t	i_size;
 850	uint		resblks;
 851	int		error;
 852
 853	trace_xfs_unwritten_convert(ip, offset, count);
 854
 855	offset_fsb = XFS_B_TO_FSBT(mp, offset);
 856	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
 857	count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
 858
 859	/*
 860	 * Reserve enough blocks in this transaction for two complete extent
 861	 * btree splits.  We may be converting the middle part of an unwritten
 862	 * extent and in this case we will insert two new extents in the btree
 863	 * each of which could cause a full split.
 864	 *
 865	 * This reservation amount will be used in the first call to
 866	 * xfs_bmbt_split() to select an AG with enough space to satisfy the
 867	 * rest of the operation.
 868	 */
 869	resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
 870
 
 
 
 
 
 871	do {
 872		/*
 873		 * Set up a transaction to convert the range of extents
 874		 * from unwritten to real. Do allocations in a loop until
 875		 * we have covered the range passed in.
 876		 *
 877		 * Note that we can't risk to recursing back into the filesystem
 878		 * here as we might be asked to write out the same inode that we
 879		 * complete here and might deadlock on the iolock.
 880		 */
 881		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0,
 882				XFS_TRANS_RESERVE | XFS_TRANS_NOFS, &tp);
 883		if (error)
 884			return error;
 885
 886		xfs_ilock(ip, XFS_ILOCK_EXCL);
 887		xfs_trans_ijoin(tp, ip, 0);
 
 
 888
 889		/*
 890		 * Modify the unwritten extent state of the buffer.
 891		 */
 892		xfs_defer_init(&dfops, &firstfsb);
 893		nimaps = 1;
 894		error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
 895					XFS_BMAPI_CONVERT, &firstfsb, resblks,
 896					&imap, &nimaps, &dfops);
 897		if (error)
 898			goto error_on_bmapi_transaction;
 899
 900		/*
 901		 * Log the updated inode size as we go.  We have to be careful
 902		 * to only log it up to the actual write offset if it is
 903		 * halfway into a block.
 904		 */
 905		i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
 906		if (i_size > offset + count)
 907			i_size = offset + count;
 908		if (update_isize && i_size > i_size_read(inode))
 909			i_size_write(inode, i_size);
 910		i_size = xfs_new_eof(ip, i_size);
 911		if (i_size) {
 912			ip->i_d.di_size = i_size;
 913			xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 914		}
 915
 916		error = xfs_defer_finish(&tp, &dfops);
 917		if (error)
 918			goto error_on_bmapi_transaction;
 919
 920		error = xfs_trans_commit(tp);
 921		xfs_iunlock(ip, XFS_ILOCK_EXCL);
 922		if (error)
 923			return error;
 924
 925		if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
 
 926			return xfs_alert_fsblock_zero(ip, &imap);
 
 927
 928		if ((numblks_fsb = imap.br_blockcount) == 0) {
 929			/*
 930			 * The numblks_fsb value should always get
 931			 * smaller, otherwise the loop is stuck.
 932			 */
 933			ASSERT(imap.br_blockcount);
 934			break;
 935		}
 936		offset_fsb += numblks_fsb;
 937		count_fsb -= numblks_fsb;
 938	} while (count_fsb > 0);
 939
 940	return 0;
 941
 942error_on_bmapi_transaction:
 943	xfs_defer_cancel(&dfops);
 944	xfs_trans_cancel(tp);
 945	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 946	return error;
 947}
 948
 949static inline bool imap_needs_alloc(struct inode *inode,
 950		struct xfs_bmbt_irec *imap, int nimaps)
 
 
 
 
 951{
 952	return !nimaps ||
 953		imap->br_startblock == HOLESTARTBLOCK ||
 954		imap->br_startblock == DELAYSTARTBLOCK ||
 955		(IS_DAX(inode) && imap->br_state == XFS_EXT_UNWRITTEN);
 
 
 
 
 
 
 
 956}
 957
 958static inline bool needs_cow_for_zeroing(struct xfs_bmbt_irec *imap, int nimaps)
 
 
 
 
 
 959{
 960	return nimaps &&
 961		imap->br_startblock != HOLESTARTBLOCK &&
 962		imap->br_state != XFS_EXT_UNWRITTEN;
 
 
 
 
 
 
 
 
 
 963}
 964
 965static inline bool need_excl_ilock(struct xfs_inode *ip, unsigned flags)
 
 
 
 
 
 
 
 
 
 
 
 966{
 967	/*
 968	 * COW writes may allocate delalloc space or convert unwritten COW
 969	 * extents, so we need to make sure to take the lock exclusively here.
 970	 */
 971	if (xfs_is_reflink_inode(ip) && (flags & (IOMAP_WRITE | IOMAP_ZERO)))
 972		return true;
 
 
 
 
 973
 974	/*
 975	 * Extents not yet cached requires exclusive access, don't block.
 976	 * This is an opencoded xfs_ilock_data_map_shared() to cater for the
 977	 * non-blocking behaviour.
 978	 */
 979	if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
 980	    !(ip->i_df.if_flags & XFS_IFEXTENTS))
 981		return true;
 982	return false;
 
 
 
 
 
 
 
 
 
 983}
 984
 985static int
 986xfs_file_iomap_begin(
 987	struct inode		*inode,
 988	loff_t			offset,
 989	loff_t			length,
 990	unsigned		flags,
 991	struct iomap		*iomap)
 
 992{
 993	struct xfs_inode	*ip = XFS_I(inode);
 994	struct xfs_mount	*mp = ip->i_mount;
 995	struct xfs_bmbt_irec	imap;
 996	xfs_fileoff_t		offset_fsb, end_fsb;
 
 997	int			nimaps = 1, error = 0;
 998	bool			shared = false, trimmed = false;
 999	unsigned		lockmode;
 
 
 
 
1000
1001	if (XFS_FORCED_SHUTDOWN(mp))
1002		return -EIO;
1003
1004	if (((flags & (IOMAP_WRITE | IOMAP_DIRECT)) == IOMAP_WRITE) &&
1005			!IS_DAX(inode) && !xfs_get_extsz_hint(ip)) {
1006		/* Reserve delalloc blocks for regular writeback. */
1007		return xfs_file_iomap_begin_delay(inode, offset, length, iomap);
1008	}
 
 
1009
1010	if (need_excl_ilock(ip, flags))
 
 
 
 
1011		lockmode = XFS_ILOCK_EXCL;
1012	else
1013		lockmode = XFS_ILOCK_SHARED;
1014
1015	if (flags & IOMAP_NOWAIT) {
1016		if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
1017			return -EAGAIN;
1018		if (!xfs_ilock_nowait(ip, lockmode))
1019			return -EAGAIN;
1020	} else {
1021		xfs_ilock(ip, lockmode);
1022	}
1023
1024	ASSERT(offset <= mp->m_super->s_maxbytes);
1025	if (offset > mp->m_super->s_maxbytes - length)
1026		length = mp->m_super->s_maxbytes - offset;
1027	offset_fsb = XFS_B_TO_FSBT(mp, offset);
1028	end_fsb = XFS_B_TO_FSB(mp, offset + length);
 
 
 
 
1029
1030	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1031			       &nimaps, 0);
1032	if (error)
1033		goto out_unlock;
1034
1035	if (flags & IOMAP_REPORT) {
1036		/* Trim the mapping to the nearest shared extent boundary. */
1037		error = xfs_reflink_trim_around_shared(ip, &imap, &shared,
1038				&trimmed);
 
 
 
 
 
1039		if (error)
1040			goto out_unlock;
 
 
 
 
1041	}
1042
1043	if (xfs_is_reflink_inode(ip) &&
1044	    ((flags & IOMAP_WRITE) ||
1045	     ((flags & IOMAP_ZERO) && needs_cow_for_zeroing(&imap, nimaps)))) {
1046		if (flags & IOMAP_DIRECT) {
1047			/*
1048			 * A reflinked inode will result in CoW alloc.
1049			 * FIXME: It could still overwrite on unshared extents
1050			 * and not need allocation.
1051			 */
1052			if (flags & IOMAP_NOWAIT) {
1053				error = -EAGAIN;
1054				goto out_unlock;
1055			}
1056			/* may drop and re-acquire the ilock */
1057			error = xfs_reflink_allocate_cow(ip, &imap, &shared,
1058					&lockmode);
1059			if (error)
1060				goto out_unlock;
1061		} else {
1062			error = xfs_reflink_reserve_cow(ip, &imap, &shared);
1063			if (error)
1064				goto out_unlock;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1065		}
 
1066
1067		end_fsb = imap.br_startoff + imap.br_blockcount;
1068		length = XFS_FSB_TO_B(mp, end_fsb) - offset;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1069	}
1070
1071	if ((flags & IOMAP_WRITE) && imap_needs_alloc(inode, &imap, nimaps)) {
1072		/*
1073		 * If nowait is set bail since we are going to make
1074		 * allocations.
 
1075		 */
1076		if (flags & IOMAP_NOWAIT) {
1077			error = -EAGAIN;
 
 
 
 
 
 
 
 
 
 
1078			goto out_unlock;
 
 
 
 
 
 
1079		}
 
 
 
 
 
 
 
 
1080		/*
1081		 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES
1082		 * pages to keep the chunks of work done where somewhat symmetric
1083		 * with the work writeback does. This is a completely arbitrary
1084		 * number pulled out of thin air as a best guess for initial
1085		 * testing.
1086		 *
1087		 * Note that the values needs to be less than 32-bits wide until
1088		 * the lower level functions are updated.
1089		 */
1090		length = min_t(loff_t, length, 1024 * PAGE_SIZE);
 
 
 
 
 
 
 
1091		/*
1092		 * xfs_iomap_write_direct() expects the shared lock. It
1093		 * is unlocked on return.
1094		 */
1095		if (lockmode == XFS_ILOCK_EXCL)
1096			xfs_ilock_demote(ip, lockmode);
1097		error = xfs_iomap_write_direct(ip, offset, length, &imap,
1098				nimaps);
1099		if (error)
1100			return error;
 
 
 
 
 
 
1101
1102		iomap->flags = IOMAP_F_NEW;
1103		trace_xfs_iomap_alloc(ip, offset, length, 0, &imap);
1104	} else {
1105		ASSERT(nimaps);
1106
1107		xfs_iunlock(ip, lockmode);
1108		trace_xfs_iomap_found(ip, offset, length, 0, &imap);
 
 
 
 
 
 
 
1109	}
1110
1111	if (xfs_ipincount(ip) && (ip->i_itemp->ili_fsync_fields
1112				& ~XFS_ILOG_TIMESTAMP))
1113		iomap->flags |= IOMAP_F_DIRTY;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1114
1115	xfs_bmbt_to_iomap(ip, iomap, &imap);
 
 
 
 
 
 
 
 
 
 
 
 
1116
1117	if (shared)
1118		iomap->flags |= IOMAP_F_SHARED;
1119	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1120out_unlock:
1121	xfs_iunlock(ip, lockmode);
1122	return error;
1123}
1124
 
 
 
 
 
 
 
 
 
 
 
 
 
1125static int
1126xfs_file_iomap_end_delalloc(
1127	struct xfs_inode	*ip,
1128	loff_t			offset,
1129	loff_t			length,
1130	ssize_t			written,
 
1131	struct iomap		*iomap)
1132{
1133	struct xfs_mount	*mp = ip->i_mount;
1134	xfs_fileoff_t		start_fsb;
1135	xfs_fileoff_t		end_fsb;
1136	int			error = 0;
1137
1138	/*
1139	 * Behave as if the write failed if drop writes is enabled. Set the NEW
1140	 * flag to force delalloc cleanup.
1141	 */
1142	if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_DROP_WRITES)) {
1143		iomap->flags |= IOMAP_F_NEW;
1144		written = 0;
1145	}
1146
1147	/*
1148	 * start_fsb refers to the first unused block after a short write. If
1149	 * nothing was written, round offset down to point at the first block in
1150	 * the range.
1151	 */
1152	if (unlikely(!written))
1153		start_fsb = XFS_B_TO_FSBT(mp, offset);
1154	else
1155		start_fsb = XFS_B_TO_FSB(mp, offset + written);
1156	end_fsb = XFS_B_TO_FSB(mp, offset + length);
1157
1158	/*
1159	 * Trim delalloc blocks if they were allocated by this write and we
1160	 * didn't manage to write the whole range.
1161	 *
1162	 * We don't need to care about racing delalloc as we hold i_mutex
1163	 * across the reserve/allocate/unreserve calls. If there are delalloc
1164	 * blocks in the range, they are ours.
1165	 */
1166	if ((iomap->flags & IOMAP_F_NEW) && start_fsb < end_fsb) {
1167		truncate_pagecache_range(VFS_I(ip), XFS_FSB_TO_B(mp, start_fsb),
1168					 XFS_FSB_TO_B(mp, end_fsb) - 1);
1169
1170		xfs_ilock(ip, XFS_ILOCK_EXCL);
1171		error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1172					       end_fsb - start_fsb);
1173		xfs_iunlock(ip, XFS_ILOCK_EXCL);
1174
1175		if (error && !XFS_FORCED_SHUTDOWN(mp)) {
1176			xfs_alert(mp, "%s: unable to clean up ino %lld",
1177				__func__, ip->i_ino);
1178			return error;
1179		}
 
 
 
 
 
1180	}
1181
1182	return 0;
1183}
1184
 
 
 
 
 
1185static int
1186xfs_file_iomap_end(
1187	struct inode		*inode,
1188	loff_t			offset,
1189	loff_t			length,
1190	ssize_t			written,
1191	unsigned		flags,
1192	struct iomap		*iomap)
 
1193{
1194	if ((flags & IOMAP_WRITE) && iomap->type == IOMAP_DELALLOC)
1195		return xfs_file_iomap_end_delalloc(XFS_I(inode), offset,
1196				length, written, iomap);
1197	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1198}
1199
1200const struct iomap_ops xfs_iomap_ops = {
1201	.iomap_begin		= xfs_file_iomap_begin,
1202	.iomap_end		= xfs_file_iomap_end,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1203};
1204
1205static int
1206xfs_xattr_iomap_begin(
1207	struct inode		*inode,
1208	loff_t			offset,
1209	loff_t			length,
1210	unsigned		flags,
1211	struct iomap		*iomap)
 
1212{
1213	struct xfs_inode	*ip = XFS_I(inode);
1214	struct xfs_mount	*mp = ip->i_mount;
1215	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
1216	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, offset + length);
1217	struct xfs_bmbt_irec	imap;
1218	int			nimaps = 1, error = 0;
1219	unsigned		lockmode;
 
1220
1221	if (XFS_FORCED_SHUTDOWN(mp))
1222		return -EIO;
1223
1224	lockmode = xfs_ilock_attr_map_shared(ip);
1225
1226	/* if there are no attribute fork or extents, return ENOENT */
1227	if (!XFS_IFORK_Q(ip) || !ip->i_d.di_anextents) {
1228		error = -ENOENT;
1229		goto out_unlock;
1230	}
1231
1232	ASSERT(ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL);
1233	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1234			       &nimaps, XFS_BMAPI_ATTRFORK);
1235out_unlock:
1236	xfs_iunlock(ip, lockmode);
1237
1238	if (!error) {
1239		ASSERT(nimaps);
1240		xfs_bmbt_to_iomap(ip, iomap, &imap);
1241	}
1242
1243	return error;
 
 
 
1244}
1245
1246const struct iomap_ops xfs_xattr_iomap_ops = {
1247	.iomap_begin		= xfs_xattr_iomap_begin,
1248};
v6.13.7
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   4 * Copyright (c) 2016-2018 Christoph Hellwig.
   5 * All Rights Reserved.
 
 
 
 
 
 
 
 
 
 
 
 
 
   6 */
 
   7#include "xfs.h"
   8#include "xfs_fs.h"
   9#include "xfs_shared.h"
  10#include "xfs_format.h"
  11#include "xfs_log_format.h"
  12#include "xfs_trans_resv.h"
  13#include "xfs_mount.h"
 
  14#include "xfs_inode.h"
  15#include "xfs_btree.h"
  16#include "xfs_bmap_btree.h"
  17#include "xfs_bmap.h"
  18#include "xfs_bmap_util.h"
  19#include "xfs_errortag.h"
  20#include "xfs_error.h"
  21#include "xfs_trans.h"
  22#include "xfs_trans_space.h"
  23#include "xfs_inode_item.h"
  24#include "xfs_iomap.h"
  25#include "xfs_trace.h"
 
  26#include "xfs_quota.h"
  27#include "xfs_rtgroup.h"
  28#include "xfs_dquot_item.h"
  29#include "xfs_dquot.h"
  30#include "xfs_reflink.h"
  31#include "xfs_health.h"
  32#include "xfs_rtbitmap.h"
  33
  34#define XFS_ALLOC_ALIGN(mp, off) \
  35	(((off) >> mp->m_allocsize_log) << mp->m_allocsize_log)
  36
  37static int
  38xfs_alert_fsblock_zero(
  39	xfs_inode_t	*ip,
  40	xfs_bmbt_irec_t	*imap)
  41{
  42	xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
  43			"Access to block zero in inode %llu "
  44			"start_block: %llx start_off: %llx "
  45			"blkcnt: %llx extent-state: %x",
  46		(unsigned long long)ip->i_ino,
  47		(unsigned long long)imap->br_startblock,
  48		(unsigned long long)imap->br_startoff,
  49		(unsigned long long)imap->br_blockcount,
  50		imap->br_state);
  51	xfs_bmap_mark_sick(ip, XFS_DATA_FORK);
  52	return -EFSCORRUPTED;
  53}
  54
  55u64
  56xfs_iomap_inode_sequence(
  57	struct xfs_inode	*ip,
  58	u16			iomap_flags)
  59{
  60	u64			cookie = 0;
  61
  62	if (iomap_flags & IOMAP_F_XATTR)
  63		return READ_ONCE(ip->i_af.if_seq);
  64	if ((iomap_flags & IOMAP_F_SHARED) && ip->i_cowfp)
  65		cookie = (u64)READ_ONCE(ip->i_cowfp->if_seq) << 32;
  66	return cookie | READ_ONCE(ip->i_df.if_seq);
  67}
  68
  69/*
  70 * Check that the iomap passed to us is still valid for the given offset and
  71 * length.
  72 */
  73static bool
  74xfs_iomap_valid(
  75	struct inode		*inode,
  76	const struct iomap	*iomap)
  77{
  78	struct xfs_inode	*ip = XFS_I(inode);
  79
  80	if (iomap->validity_cookie !=
  81			xfs_iomap_inode_sequence(ip, iomap->flags)) {
  82		trace_xfs_iomap_invalid(ip, iomap);
  83		return false;
  84	}
  85
  86	XFS_ERRORTAG_DELAY(ip->i_mount, XFS_ERRTAG_WRITE_DELAY_MS);
  87	return true;
  88}
  89
  90static const struct iomap_folio_ops xfs_iomap_folio_ops = {
  91	.iomap_valid		= xfs_iomap_valid,
  92};
  93
  94int
  95xfs_bmbt_to_iomap(
  96	struct xfs_inode	*ip,
  97	struct iomap		*iomap,
  98	struct xfs_bmbt_irec	*imap,
  99	unsigned int		mapping_flags,
 100	u16			iomap_flags,
 101	u64			sequence_cookie)
 102{
 103	struct xfs_mount	*mp = ip->i_mount;
 104	struct xfs_buftarg	*target = xfs_inode_buftarg(ip);
 105
 106	if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock))) {
 107		xfs_bmap_mark_sick(ip, XFS_DATA_FORK);
 108		return xfs_alert_fsblock_zero(ip, imap);
 109	}
 110
 111	if (imap->br_startblock == HOLESTARTBLOCK) {
 112		iomap->addr = IOMAP_NULL_ADDR;
 113		iomap->type = IOMAP_HOLE;
 114	} else if (imap->br_startblock == DELAYSTARTBLOCK ||
 115		   isnullstartblock(imap->br_startblock)) {
 116		iomap->addr = IOMAP_NULL_ADDR;
 117		iomap->type = IOMAP_DELALLOC;
 118	} else {
 119		xfs_daddr_t	daddr = xfs_fsb_to_db(ip, imap->br_startblock);
 120
 121		iomap->addr = BBTOB(daddr);
 122		if (mapping_flags & IOMAP_DAX)
 123			iomap->addr += target->bt_dax_part_off;
 124
 125		if (imap->br_state == XFS_EXT_UNWRITTEN)
 126			iomap->type = IOMAP_UNWRITTEN;
 127		else
 128			iomap->type = IOMAP_MAPPED;
 129
 130		/*
 131		 * Mark iomaps starting at the first sector of a RTG as merge
 132		 * boundary so that each I/O completions is contained to a
 133		 * single RTG.
 134		 */
 135		if (XFS_IS_REALTIME_INODE(ip) && xfs_has_rtgroups(mp) &&
 136		    xfs_rtbno_is_group_start(mp, imap->br_startblock))
 137			iomap->flags |= IOMAP_F_BOUNDARY;
 138	}
 139	iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff);
 140	iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount);
 141	if (mapping_flags & IOMAP_DAX)
 142		iomap->dax_dev = target->bt_daxdev;
 143	else
 144		iomap->bdev = target->bt_bdev;
 145	iomap->flags = iomap_flags;
 146
 147	if (xfs_ipincount(ip) &&
 148	    (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
 149		iomap->flags |= IOMAP_F_DIRTY;
 150
 151	iomap->validity_cookie = sequence_cookie;
 152	iomap->folio_ops = &xfs_iomap_folio_ops;
 153	return 0;
 154}
 155
 156static void
 157xfs_hole_to_iomap(
 158	struct xfs_inode	*ip,
 159	struct iomap		*iomap,
 160	xfs_fileoff_t		offset_fsb,
 161	xfs_fileoff_t		end_fsb)
 162{
 163	struct xfs_buftarg	*target = xfs_inode_buftarg(ip);
 164
 165	iomap->addr = IOMAP_NULL_ADDR;
 166	iomap->type = IOMAP_HOLE;
 167	iomap->offset = XFS_FSB_TO_B(ip->i_mount, offset_fsb);
 168	iomap->length = XFS_FSB_TO_B(ip->i_mount, end_fsb - offset_fsb);
 169	iomap->bdev = target->bt_bdev;
 170	iomap->dax_dev = target->bt_daxdev;
 171}
 172
 173static inline xfs_fileoff_t
 174xfs_iomap_end_fsb(
 175	struct xfs_mount	*mp,
 176	loff_t			offset,
 177	loff_t			count)
 178{
 179	ASSERT(offset <= mp->m_super->s_maxbytes);
 180	return min(XFS_B_TO_FSB(mp, offset + count),
 181		   XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes));
 182}
 183
 184static xfs_extlen_t
 185xfs_eof_alignment(
 186	struct xfs_inode	*ip)
 187{
 188	struct xfs_mount	*mp = ip->i_mount;
 189	xfs_extlen_t		align = 0;
 190
 191	if (!XFS_IS_REALTIME_INODE(ip)) {
 192		/*
 193		 * Round up the allocation request to a stripe unit
 194		 * (m_dalign) boundary if the file size is >= stripe unit
 195		 * size, and we are allocating past the allocation eof.
 196		 *
 197		 * If mounted with the "-o swalloc" option the alignment is
 198		 * increased from the strip unit size to the stripe width.
 199		 */
 200		if (mp->m_swidth && xfs_has_swalloc(mp))
 201			align = mp->m_swidth;
 202		else if (mp->m_dalign)
 203			align = mp->m_dalign;
 204
 205		if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
 206			align = 0;
 207	}
 208
 
 
 
 
 
 
 
 
 
 
 
 209	return align;
 210}
 211
 212/*
 213 * Check if last_fsb is outside the last extent, and if so grow it to the next
 214 * stripe unit boundary.
 215 */
 216xfs_fileoff_t
 217xfs_iomap_eof_align_last_fsb(
 218	struct xfs_inode	*ip,
 219	xfs_fileoff_t		end_fsb)
 
 220{
 221	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, XFS_DATA_FORK);
 222	xfs_extlen_t		extsz = xfs_get_extsz_hint(ip);
 223	xfs_extlen_t		align = xfs_eof_alignment(ip);
 224	struct xfs_bmbt_irec	irec;
 225	struct xfs_iext_cursor	icur;
 226
 227	ASSERT(!xfs_need_iread_extents(ifp));
 228
 229	/*
 230	 * Always round up the allocation request to the extent hint boundary.
 231	 */
 232	if (extsz) {
 233		if (align)
 234			align = roundup_64(align, extsz);
 235		else
 236			align = extsz;
 237	}
 238
 239	if (align) {
 240		xfs_fileoff_t	aligned_end_fsb = roundup_64(end_fsb, align);
 
 241
 242		xfs_iext_last(ifp, &icur);
 243		if (!xfs_iext_get_extent(ifp, &icur, &irec) ||
 244		    aligned_end_fsb >= irec.br_startoff + irec.br_blockcount)
 245			return aligned_end_fsb;
 
 246	}
 
 
 247
 248	return end_fsb;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 249}
 250
 251int
 252xfs_iomap_write_direct(
 253	struct xfs_inode	*ip,
 254	xfs_fileoff_t		offset_fsb,
 255	xfs_fileoff_t		count_fsb,
 256	unsigned int		flags,
 257	struct xfs_bmbt_irec	*imap,
 258	u64			*seq)
 259{
 260	struct xfs_mount	*mp = ip->i_mount;
 261	struct xfs_trans	*tp;
 262	xfs_filblks_t		resaligned;
 263	int			nimaps;
 264	unsigned int		dblocks, rblocks;
 265	bool			force = false;
 266	int			error;
 267	int			bmapi_flags = XFS_BMAPI_PREALLOC;
 268	int			nr_exts = XFS_IEXT_ADD_NOSPLIT_CNT;
 
 
 
 
 
 
 
 
 
 
 
 
 
 269
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 270	ASSERT(count_fsb > 0);
 
 271
 272	resaligned = xfs_aligned_fsb_count(offset_fsb, count_fsb,
 273					   xfs_get_extsz_hint(ip));
 274	if (unlikely(XFS_IS_REALTIME_INODE(ip))) {
 275		dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
 276		rblocks = resaligned;
 277	} else {
 278		dblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
 279		rblocks = 0;
 
 280	}
 281
 282	error = xfs_qm_dqattach(ip);
 
 
 
 
 
 283	if (error)
 284		return error;
 285
 286	/*
 287	 * For DAX, we do not allocate unwritten extents, but instead we zero
 288	 * the block before we commit the transaction.  Ideally we'd like to do
 289	 * this outside the transaction context, but if we commit and then crash
 290	 * we may not have zeroed the blocks and this will be exposed on
 291	 * recovery of the allocation. Hence we must zero before commit.
 292	 *
 293	 * Further, if we are mapping unwritten extents here, we need to zero
 294	 * and convert them to written so that we don't need an unwritten extent
 295	 * callback for DAX. This also means that we need to be able to dip into
 296	 * the reserve block pool for bmbt block allocation if there is no space
 297	 * left but we need to do unwritten extent conversion.
 298	 */
 299	if (flags & IOMAP_DAX) {
 300		bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
 301		if (imap->br_state == XFS_EXT_UNWRITTEN) {
 302			force = true;
 303			nr_exts = XFS_IEXT_WRITE_UNWRITTEN_CNT;
 304			dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
 305		}
 306	}
 307
 308	error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, dblocks,
 309			rblocks, force, &tp);
 310	if (error)
 311		return error;
 312
 313	error = xfs_iext_count_extend(tp, ip, XFS_DATA_FORK, nr_exts);
 
 
 
 314	if (error)
 315		goto out_trans_cancel;
 316
 
 
 317	/*
 318	 * From this point onwards we overwrite the imap pointer that the
 319	 * caller gave to us.
 320	 */
 
 321	nimaps = 1;
 322	error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flags, 0,
 323				imap, &nimaps);
 
 324	if (error)
 325		goto out_trans_cancel;
 326
 327	/*
 328	 * Complete the transaction
 329	 */
 
 
 
 
 330	error = xfs_trans_commit(tp);
 331	if (error)
 332		goto out_unlock;
 333
 334	if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock))) {
 335		xfs_bmap_mark_sick(ip, XFS_DATA_FORK);
 
 
 
 
 
 
 
 336		error = xfs_alert_fsblock_zero(ip, imap);
 337	}
 338
 339out_unlock:
 340	*seq = xfs_iomap_inode_sequence(ip, 0);
 341	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 342	return error;
 343
 
 
 
 344out_trans_cancel:
 345	xfs_trans_cancel(tp);
 346	goto out_unlock;
 347}
 348
 349STATIC bool
 350xfs_quota_need_throttle(
 351	struct xfs_inode	*ip,
 352	xfs_dqtype_t		type,
 353	xfs_fsblock_t		alloc_blocks)
 354{
 355	struct xfs_dquot	*dq = xfs_inode_dquot(ip, type);
 356	struct xfs_dquot_res	*res;
 357	struct xfs_dquot_pre	*pre;
 358
 359	if (!dq || !xfs_this_quota_on(ip->i_mount, type))
 360		return false;
 361
 362	if (XFS_IS_REALTIME_INODE(ip)) {
 363		res = &dq->q_rtb;
 364		pre = &dq->q_rtb_prealloc;
 365	} else {
 366		res = &dq->q_blk;
 367		pre = &dq->q_blk_prealloc;
 368	}
 369
 370	/* no hi watermark, no throttle */
 371	if (!pre->q_prealloc_hi_wmark)
 372		return false;
 373
 374	/* under the lo watermark, no throttle */
 375	if (res->reserved + alloc_blocks < pre->q_prealloc_lo_wmark)
 376		return false;
 377
 378	return true;
 379}
 380
 381STATIC void
 382xfs_quota_calc_throttle(
 383	struct xfs_inode	*ip,
 384	xfs_dqtype_t		type,
 385	xfs_fsblock_t		*qblocks,
 386	int			*qshift,
 387	int64_t			*qfreesp)
 388{
 389	struct xfs_dquot	*dq = xfs_inode_dquot(ip, type);
 390	struct xfs_dquot_res	*res;
 391	struct xfs_dquot_pre	*pre;
 392	int64_t			freesp;
 393	int			shift = 0;
 394
 395	if (!dq) {
 396		res = NULL;
 397		pre = NULL;
 398	} else if (XFS_IS_REALTIME_INODE(ip)) {
 399		res = &dq->q_rtb;
 400		pre = &dq->q_rtb_prealloc;
 401	} else {
 402		res = &dq->q_blk;
 403		pre = &dq->q_blk_prealloc;
 404	}
 405
 406	/* no dq, or over hi wmark, squash the prealloc completely */
 407	if (!res || res->reserved >= pre->q_prealloc_hi_wmark) {
 408		*qblocks = 0;
 409		*qfreesp = 0;
 410		return;
 411	}
 412
 413	freesp = pre->q_prealloc_hi_wmark - res->reserved;
 414	if (freesp < pre->q_low_space[XFS_QLOWSP_5_PCNT]) {
 415		shift = 2;
 416		if (freesp < pre->q_low_space[XFS_QLOWSP_3_PCNT])
 417			shift += 2;
 418		if (freesp < pre->q_low_space[XFS_QLOWSP_1_PCNT])
 419			shift += 2;
 420	}
 421
 422	if (freesp < *qfreesp)
 423		*qfreesp = freesp;
 424
 425	/* only overwrite the throttle values if we are more aggressive */
 426	if ((freesp >> shift) < (*qblocks >> *qshift)) {
 427		*qblocks = freesp;
 428		*qshift = shift;
 429	}
 430}
 431
 432static int64_t
 433xfs_iomap_freesp(
 434	struct percpu_counter	*counter,
 435	uint64_t		low_space[XFS_LOWSP_MAX],
 436	int			*shift)
 437{
 438	int64_t			freesp;
 439
 440	freesp = percpu_counter_read_positive(counter);
 441	if (freesp < low_space[XFS_LOWSP_5_PCNT]) {
 442		*shift = 2;
 443		if (freesp < low_space[XFS_LOWSP_4_PCNT])
 444			(*shift)++;
 445		if (freesp < low_space[XFS_LOWSP_3_PCNT])
 446			(*shift)++;
 447		if (freesp < low_space[XFS_LOWSP_2_PCNT])
 448			(*shift)++;
 449		if (freesp < low_space[XFS_LOWSP_1_PCNT])
 450			(*shift)++;
 451	}
 452	return freesp;
 453}
 454
 455/*
 
 
 
 
 456 * If we don't have a user specified preallocation size, dynamically increase
 457 * the preallocation size as the size of the file grows.  Cap the maximum size
 458 * at a single extent or less if the filesystem is near full. The closer the
 459 * filesystem is to being full, the smaller the maximum preallocation.
 
 
 
 
 
 
 
 
 460 */
 461STATIC xfs_fsblock_t
 462xfs_iomap_prealloc_size(
 463	struct xfs_inode	*ip,
 464	int			whichfork,
 465	loff_t			offset,
 466	loff_t			count,
 467	struct xfs_iext_cursor	*icur)
 468{
 469	struct xfs_iext_cursor	ncur = *icur;
 470	struct xfs_bmbt_irec	prev, got;
 471	struct xfs_mount	*mp = ip->i_mount;
 472	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
 473	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
 
 
 474	int64_t			freesp;
 475	xfs_fsblock_t		qblocks;
 
 476	xfs_fsblock_t		alloc_blocks = 0;
 477	xfs_extlen_t		plen;
 478	int			shift = 0;
 479	int			qshift = 0;
 480
 481	/*
 482	 * As an exception we don't do any preallocation at all if the file is
 483	 * smaller than the minimum preallocation and we are using the default
 484	 * dynamic preallocation scheme, as it is likely this is the only write
 485	 * to the file that is going to be done.
 486	 */
 487	if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_allocsize_blocks))
 488		return 0;
 489
 490	/*
 491	 * Use the minimum preallocation size for small files or if we are
 492	 * writing right after a hole.
 493	 */
 494	if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign) ||
 495	    !xfs_iext_prev_extent(ifp, &ncur, &prev) ||
 
 496	    prev.br_startoff + prev.br_blockcount < offset_fsb)
 497		return mp->m_allocsize_blocks;
 498
 499	/*
 500	 * Take the size of the preceding data extents as the basis for the
 501	 * preallocation size. Note that we don't care if the previous extents
 502	 * are written or not.
 503	 */
 504	plen = prev.br_blockcount;
 505	while (xfs_iext_prev_extent(ifp, &ncur, &got)) {
 506		if (plen > XFS_MAX_BMBT_EXTLEN / 2 ||
 507		    isnullstartblock(got.br_startblock) ||
 508		    got.br_startoff + got.br_blockcount != prev.br_startoff ||
 509		    got.br_startblock + got.br_blockcount != prev.br_startblock)
 510			break;
 511		plen += got.br_blockcount;
 512		prev = got;
 513	}
 514
 515	/*
 516	 * If the size of the extents is greater than half the maximum extent
 517	 * length, then use the current offset as the basis.  This ensures that
 518	 * for large files the preallocation size always extends to
 519	 * XFS_BMBT_MAX_EXTLEN rather than falling short due to things like stripe
 520	 * unit/width alignment of real extents.
 521	 */
 522	alloc_blocks = plen * 2;
 523	if (alloc_blocks > XFS_MAX_BMBT_EXTLEN)
 
 524		alloc_blocks = XFS_B_TO_FSB(mp, offset);
 
 
 525	qblocks = alloc_blocks;
 526
 527	/*
 528	 * XFS_BMBT_MAX_EXTLEN is not a power of two value but we round the prealloc
 529	 * down to the nearest power of two value after throttling. To prevent
 530	 * the round down from unconditionally reducing the maximum supported
 531	 * prealloc size, we round up first, apply appropriate throttling, round
 532	 * down and cap the value to XFS_BMBT_MAX_EXTLEN.
 533	 */
 534	alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(XFS_MAX_BMBT_EXTLEN),
 535				       alloc_blocks);
 536
 537	if (unlikely(XFS_IS_REALTIME_INODE(ip)))
 538		freesp = xfs_rtbxlen_to_blen(mp,
 539				xfs_iomap_freesp(&mp->m_frextents,
 540					mp->m_low_rtexts, &shift));
 541	else
 542		freesp = xfs_iomap_freesp(&mp->m_fdblocks, mp->m_low_space,
 543				&shift);
 
 
 
 
 
 544
 545	/*
 546	 * Check each quota to cap the prealloc size, provide a shift value to
 547	 * throttle with and adjust amount of available space.
 548	 */
 549	if (xfs_quota_need_throttle(ip, XFS_DQTYPE_USER, alloc_blocks))
 550		xfs_quota_calc_throttle(ip, XFS_DQTYPE_USER, &qblocks, &qshift,
 551					&freesp);
 552	if (xfs_quota_need_throttle(ip, XFS_DQTYPE_GROUP, alloc_blocks))
 553		xfs_quota_calc_throttle(ip, XFS_DQTYPE_GROUP, &qblocks, &qshift,
 554					&freesp);
 555	if (xfs_quota_need_throttle(ip, XFS_DQTYPE_PROJ, alloc_blocks))
 556		xfs_quota_calc_throttle(ip, XFS_DQTYPE_PROJ, &qblocks, &qshift,
 557					&freesp);
 558
 559	/*
 560	 * The final prealloc size is set to the minimum of free space available
 561	 * in each of the quotas and the overall filesystem.
 562	 *
 563	 * The shift throttle value is set to the maximum value as determined by
 564	 * the global low free space values and per-quota low free space values.
 565	 */
 566	alloc_blocks = min(alloc_blocks, qblocks);
 567	shift = max(shift, qshift);
 568
 569	if (shift)
 570		alloc_blocks >>= shift;
 571	/*
 572	 * rounddown_pow_of_two() returns an undefined result if we pass in
 573	 * alloc_blocks = 0.
 574	 */
 575	if (alloc_blocks)
 576		alloc_blocks = rounddown_pow_of_two(alloc_blocks);
 577	if (alloc_blocks > XFS_MAX_BMBT_EXTLEN)
 578		alloc_blocks = XFS_MAX_BMBT_EXTLEN;
 579
 580	/*
 581	 * If we are still trying to allocate more space than is
 582	 * available, squash the prealloc hard. This can happen if we
 583	 * have a large file on a small filesystem and the above
 584	 * lowspace thresholds are smaller than XFS_BMBT_MAX_EXTLEN.
 585	 */
 586	while (alloc_blocks && alloc_blocks >= freesp)
 587		alloc_blocks >>= 4;
 588	if (alloc_blocks < mp->m_allocsize_blocks)
 589		alloc_blocks = mp->m_allocsize_blocks;
 
 590	trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
 591				      mp->m_allocsize_blocks);
 592	return alloc_blocks;
 593}
 594
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 595int
 596xfs_iomap_write_unwritten(
 597	xfs_inode_t	*ip,
 598	xfs_off_t	offset,
 599	xfs_off_t	count,
 600	bool		update_isize)
 601{
 602	xfs_mount_t	*mp = ip->i_mount;
 603	xfs_fileoff_t	offset_fsb;
 604	xfs_filblks_t	count_fsb;
 605	xfs_filblks_t	numblks_fsb;
 
 606	int		nimaps;
 607	xfs_trans_t	*tp;
 608	xfs_bmbt_irec_t imap;
 
 609	struct inode	*inode = VFS_I(ip);
 610	xfs_fsize_t	i_size;
 611	uint		resblks;
 612	int		error;
 613
 614	trace_xfs_unwritten_convert(ip, offset, count);
 615
 616	offset_fsb = XFS_B_TO_FSBT(mp, offset);
 617	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
 618	count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
 619
 620	/*
 621	 * Reserve enough blocks in this transaction for two complete extent
 622	 * btree splits.  We may be converting the middle part of an unwritten
 623	 * extent and in this case we will insert two new extents in the btree
 624	 * each of which could cause a full split.
 625	 *
 626	 * This reservation amount will be used in the first call to
 627	 * xfs_bmbt_split() to select an AG with enough space to satisfy the
 628	 * rest of the operation.
 629	 */
 630	resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
 631
 632	/* Attach dquots so that bmbt splits are accounted correctly. */
 633	error = xfs_qm_dqattach(ip);
 634	if (error)
 635		return error;
 636
 637	do {
 638		/*
 639		 * Set up a transaction to convert the range of extents
 640		 * from unwritten to real. Do allocations in a loop until
 641		 * we have covered the range passed in.
 642		 *
 643		 * Note that we can't risk to recursing back into the filesystem
 644		 * here as we might be asked to write out the same inode that we
 645		 * complete here and might deadlock on the iolock.
 646		 */
 647		error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, resblks,
 648				0, true, &tp);
 649		if (error)
 650			return error;
 651
 652		error = xfs_iext_count_extend(tp, ip, XFS_DATA_FORK,
 653				XFS_IEXT_WRITE_UNWRITTEN_CNT);
 654		if (error)
 655			goto error_on_bmapi_transaction;
 656
 657		/*
 658		 * Modify the unwritten extent state of the buffer.
 659		 */
 
 660		nimaps = 1;
 661		error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
 662					XFS_BMAPI_CONVERT, resblks, &imap,
 663					&nimaps);
 664		if (error)
 665			goto error_on_bmapi_transaction;
 666
 667		/*
 668		 * Log the updated inode size as we go.  We have to be careful
 669		 * to only log it up to the actual write offset if it is
 670		 * halfway into a block.
 671		 */
 672		i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
 673		if (i_size > offset + count)
 674			i_size = offset + count;
 675		if (update_isize && i_size > i_size_read(inode))
 676			i_size_write(inode, i_size);
 677		i_size = xfs_new_eof(ip, i_size);
 678		if (i_size) {
 679			ip->i_disk_size = i_size;
 680			xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 681		}
 682
 
 
 
 
 683		error = xfs_trans_commit(tp);
 684		xfs_iunlock(ip, XFS_ILOCK_EXCL);
 685		if (error)
 686			return error;
 687
 688		if (unlikely(!xfs_valid_startblock(ip, imap.br_startblock))) {
 689			xfs_bmap_mark_sick(ip, XFS_DATA_FORK);
 690			return xfs_alert_fsblock_zero(ip, &imap);
 691		}
 692
 693		if ((numblks_fsb = imap.br_blockcount) == 0) {
 694			/*
 695			 * The numblks_fsb value should always get
 696			 * smaller, otherwise the loop is stuck.
 697			 */
 698			ASSERT(imap.br_blockcount);
 699			break;
 700		}
 701		offset_fsb += numblks_fsb;
 702		count_fsb -= numblks_fsb;
 703	} while (count_fsb > 0);
 704
 705	return 0;
 706
 707error_on_bmapi_transaction:
 
 708	xfs_trans_cancel(tp);
 709	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 710	return error;
 711}
 712
 713static inline bool
 714imap_needs_alloc(
 715	struct inode		*inode,
 716	unsigned		flags,
 717	struct xfs_bmbt_irec	*imap,
 718	int			nimaps)
 719{
 720	/* don't allocate blocks when just zeroing */
 721	if (flags & IOMAP_ZERO)
 722		return false;
 723	if (!nimaps ||
 724	    imap->br_startblock == HOLESTARTBLOCK ||
 725	    imap->br_startblock == DELAYSTARTBLOCK)
 726		return true;
 727	/* we convert unwritten extents before copying the data for DAX */
 728	if ((flags & IOMAP_DAX) && imap->br_state == XFS_EXT_UNWRITTEN)
 729		return true;
 730	return false;
 731}
 732
 733static inline bool
 734imap_needs_cow(
 735	struct xfs_inode	*ip,
 736	unsigned int		flags,
 737	struct xfs_bmbt_irec	*imap,
 738	int			nimaps)
 739{
 740	if (!xfs_is_cow_inode(ip))
 741		return false;
 742
 743	/* when zeroing we don't have to COW holes or unwritten extents */
 744	if (flags & (IOMAP_UNSHARE | IOMAP_ZERO)) {
 745		if (!nimaps ||
 746		    imap->br_startblock == HOLESTARTBLOCK ||
 747		    imap->br_state == XFS_EXT_UNWRITTEN)
 748			return false;
 749	}
 750
 751	return true;
 752}
 753
 754/*
 755 * Extents not yet cached requires exclusive access, don't block for
 756 * IOMAP_NOWAIT.
 757 *
 758 * This is basically an opencoded xfs_ilock_data_map_shared() call, but with
 759 * support for IOMAP_NOWAIT.
 760 */
 761static int
 762xfs_ilock_for_iomap(
 763	struct xfs_inode	*ip,
 764	unsigned		flags,
 765	unsigned		*lockmode)
 766{
 767	if (flags & IOMAP_NOWAIT) {
 768		if (xfs_need_iread_extents(&ip->i_df))
 769			return -EAGAIN;
 770		if (!xfs_ilock_nowait(ip, *lockmode))
 771			return -EAGAIN;
 772	} else {
 773		if (xfs_need_iread_extents(&ip->i_df))
 774			*lockmode = XFS_ILOCK_EXCL;
 775		xfs_ilock(ip, *lockmode);
 776	}
 777
 778	return 0;
 779}
 780
 781/*
 782 * Check that the imap we are going to return to the caller spans the entire
 783 * range that the caller requested for the IO.
 784 */
 785static bool
 786imap_spans_range(
 787	struct xfs_bmbt_irec	*imap,
 788	xfs_fileoff_t		offset_fsb,
 789	xfs_fileoff_t		end_fsb)
 790{
 791	if (imap->br_startoff > offset_fsb)
 792		return false;
 793	if (imap->br_startoff + imap->br_blockcount < end_fsb)
 794		return false;
 795	return true;
 796}
 797
 798static int
 799xfs_direct_write_iomap_begin(
 800	struct inode		*inode,
 801	loff_t			offset,
 802	loff_t			length,
 803	unsigned		flags,
 804	struct iomap		*iomap,
 805	struct iomap		*srcmap)
 806{
 807	struct xfs_inode	*ip = XFS_I(inode);
 808	struct xfs_mount	*mp = ip->i_mount;
 809	struct xfs_bmbt_irec	imap, cmap;
 810	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
 811	xfs_fileoff_t		end_fsb = xfs_iomap_end_fsb(mp, offset, length);
 812	int			nimaps = 1, error = 0;
 813	bool			shared = false;
 814	u16			iomap_flags = 0;
 815	unsigned int		lockmode;
 816	u64			seq;
 817
 818	ASSERT(flags & (IOMAP_WRITE | IOMAP_ZERO));
 819
 820	if (xfs_is_shutdown(mp))
 821		return -EIO;
 822
 823	/*
 824	 * Writes that span EOF might trigger an IO size update on completion,
 825	 * so consider them to be dirty for the purposes of O_DSYNC even if
 826	 * there is no other metadata changes pending or have been made here.
 827	 */
 828	if (offset + length > i_size_read(inode))
 829		iomap_flags |= IOMAP_F_DIRTY;
 830
 831	/*
 832	 * COW writes may allocate delalloc space or convert unwritten COW
 833	 * extents, so we need to make sure to take the lock exclusively here.
 834	 */
 835	if (xfs_is_cow_inode(ip))
 836		lockmode = XFS_ILOCK_EXCL;
 837	else
 838		lockmode = XFS_ILOCK_SHARED;
 839
 840relock:
 841	error = xfs_ilock_for_iomap(ip, flags, &lockmode);
 842	if (error)
 843		return error;
 
 
 
 
 844
 845	/*
 846	 * The reflink iflag could have changed since the earlier unlocked
 847	 * check, check if it again and relock if needed.
 848	 */
 849	if (xfs_is_cow_inode(ip) && lockmode == XFS_ILOCK_SHARED) {
 850		xfs_iunlock(ip, lockmode);
 851		lockmode = XFS_ILOCK_EXCL;
 852		goto relock;
 853	}
 854
 855	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
 856			       &nimaps, 0);
 857	if (error)
 858		goto out_unlock;
 859
 860	if (imap_needs_cow(ip, flags, &imap, nimaps)) {
 861		error = -EAGAIN;
 862		if (flags & IOMAP_NOWAIT)
 863			goto out_unlock;
 864
 865		/* may drop and re-acquire the ilock */
 866		error = xfs_reflink_allocate_cow(ip, &imap, &cmap, &shared,
 867				&lockmode,
 868				(flags & IOMAP_DIRECT) || IS_DAX(inode));
 869		if (error)
 870			goto out_unlock;
 871		if (shared)
 872			goto out_found_cow;
 873		end_fsb = imap.br_startoff + imap.br_blockcount;
 874		length = XFS_FSB_TO_B(mp, end_fsb) - offset;
 875	}
 876
 877	if (imap_needs_alloc(inode, flags, &imap, nimaps))
 878		goto allocate_blocks;
 879
 880	/*
 881	 * NOWAIT and OVERWRITE I/O needs to span the entire requested I/O with
 882	 * a single map so that we avoid partial IO failures due to the rest of
 883	 * the I/O range not covered by this map triggering an EAGAIN condition
 884	 * when it is subsequently mapped and aborting the I/O.
 885	 */
 886	if (flags & (IOMAP_NOWAIT | IOMAP_OVERWRITE_ONLY)) {
 887		error = -EAGAIN;
 888		if (!imap_spans_range(&imap, offset_fsb, end_fsb))
 889			goto out_unlock;
 890	}
 891
 892	/*
 893	 * For overwrite only I/O, we cannot convert unwritten extents without
 894	 * requiring sub-block zeroing.  This can only be done under an
 895	 * exclusive IOLOCK, hence return -EAGAIN if this is not a written
 896	 * extent to tell the caller to try again.
 897	 */
 898	if (flags & IOMAP_OVERWRITE_ONLY) {
 899		error = -EAGAIN;
 900		if (imap.br_state != XFS_EXT_NORM &&
 901	            ((offset | length) & mp->m_blockmask))
 902			goto out_unlock;
 903	}
 904
 905	seq = xfs_iomap_inode_sequence(ip, iomap_flags);
 906	xfs_iunlock(ip, lockmode);
 907	trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
 908	return xfs_bmbt_to_iomap(ip, iomap, &imap, flags, iomap_flags, seq);
 909
 910allocate_blocks:
 911	error = -EAGAIN;
 912	if (flags & (IOMAP_NOWAIT | IOMAP_OVERWRITE_ONLY))
 913		goto out_unlock;
 914
 915	/*
 916	 * We cap the maximum length we map to a sane size  to keep the chunks
 917	 * of work done where somewhat symmetric with the work writeback does.
 918	 * This is a completely arbitrary number pulled out of thin air as a
 919	 * best guess for initial testing.
 920	 *
 921	 * Note that the values needs to be less than 32-bits wide until the
 922	 * lower level functions are updated.
 923	 */
 924	length = min_t(loff_t, length, 1024 * PAGE_SIZE);
 925	end_fsb = xfs_iomap_end_fsb(mp, offset, length);
 926
 927	if (offset + length > XFS_ISIZE(ip))
 928		end_fsb = xfs_iomap_eof_align_last_fsb(ip, end_fsb);
 929	else if (nimaps && imap.br_startblock == HOLESTARTBLOCK)
 930		end_fsb = min(end_fsb, imap.br_startoff + imap.br_blockcount);
 931	xfs_iunlock(ip, lockmode);
 932
 933	error = xfs_iomap_write_direct(ip, offset_fsb, end_fsb - offset_fsb,
 934			flags, &imap, &seq);
 935	if (error)
 936		return error;
 937
 938	trace_xfs_iomap_alloc(ip, offset, length, XFS_DATA_FORK, &imap);
 939	return xfs_bmbt_to_iomap(ip, iomap, &imap, flags,
 940				 iomap_flags | IOMAP_F_NEW, seq);
 941
 942out_found_cow:
 943	length = XFS_FSB_TO_B(mp, cmap.br_startoff + cmap.br_blockcount);
 944	trace_xfs_iomap_found(ip, offset, length - offset, XFS_COW_FORK, &cmap);
 945	if (imap.br_startblock != HOLESTARTBLOCK) {
 946		seq = xfs_iomap_inode_sequence(ip, 0);
 947		error = xfs_bmbt_to_iomap(ip, srcmap, &imap, flags, 0, seq);
 948		if (error)
 949			goto out_unlock;
 950	}
 951	seq = xfs_iomap_inode_sequence(ip, IOMAP_F_SHARED);
 952	xfs_iunlock(ip, lockmode);
 953	return xfs_bmbt_to_iomap(ip, iomap, &cmap, flags, IOMAP_F_SHARED, seq);
 954
 955out_unlock:
 956	if (lockmode)
 957		xfs_iunlock(ip, lockmode);
 958	return error;
 959}
 960
 961const struct iomap_ops xfs_direct_write_iomap_ops = {
 962	.iomap_begin		= xfs_direct_write_iomap_begin,
 963};
 964
 965static int
 966xfs_dax_write_iomap_end(
 967	struct inode		*inode,
 968	loff_t			pos,
 969	loff_t			length,
 970	ssize_t			written,
 971	unsigned		flags,
 972	struct iomap		*iomap)
 973{
 974	struct xfs_inode	*ip = XFS_I(inode);
 975
 976	if (!xfs_is_cow_inode(ip))
 977		return 0;
 978
 979	if (!written)
 980		return xfs_reflink_cancel_cow_range(ip, pos, length, true);
 981
 982	return xfs_reflink_end_cow(ip, pos, written);
 983}
 984
 985const struct iomap_ops xfs_dax_write_iomap_ops = {
 986	.iomap_begin	= xfs_direct_write_iomap_begin,
 987	.iomap_end	= xfs_dax_write_iomap_end,
 988};
 989
 990static int
 991xfs_buffered_write_iomap_begin(
 992	struct inode		*inode,
 993	loff_t			offset,
 994	loff_t			count,
 995	unsigned		flags,
 996	struct iomap		*iomap,
 997	struct iomap		*srcmap)
 998{
 999	struct xfs_inode	*ip = XFS_I(inode);
1000	struct xfs_mount	*mp = ip->i_mount;
1001	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
1002	xfs_fileoff_t		end_fsb = xfs_iomap_end_fsb(mp, offset, count);
1003	struct xfs_bmbt_irec	imap, cmap;
1004	struct xfs_iext_cursor	icur, ccur;
1005	xfs_fsblock_t		prealloc_blocks = 0;
1006	bool			eof = false, cow_eof = false, shared = false;
1007	int			allocfork = XFS_DATA_FORK;
1008	int			error = 0;
1009	unsigned int		lockmode = XFS_ILOCK_EXCL;
1010	unsigned int		iomap_flags = 0;
1011	u64			seq;
1012
1013	if (xfs_is_shutdown(mp))
1014		return -EIO;
1015
1016	/* we can't use delayed allocations when using extent size hints */
1017	if (xfs_get_extsz_hint(ip))
1018		return xfs_direct_write_iomap_begin(inode, offset, count,
1019				flags, iomap, srcmap);
1020
1021	error = xfs_qm_dqattach(ip);
1022	if (error)
1023		return error;
1024
1025	error = xfs_ilock_for_iomap(ip, flags, &lockmode);
1026	if (error)
1027		return error;
1028
1029	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(&ip->i_df)) ||
1030	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
1031		xfs_bmap_mark_sick(ip, XFS_DATA_FORK);
1032		error = -EFSCORRUPTED;
1033		goto out_unlock;
1034	}
1035
1036	XFS_STATS_INC(mp, xs_blk_mapw);
1037
1038	error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
1039	if (error)
1040		goto out_unlock;
1041
1042	/*
1043	 * Search the data fork first to look up our source mapping.  We
1044	 * always need the data fork map, as we have to return it to the
1045	 * iomap code so that the higher level write code can read data in to
1046	 * perform read-modify-write cycles for unaligned writes.
1047	 */
1048	eof = !xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap);
1049	if (eof)
1050		imap.br_startoff = end_fsb; /* fake hole until the end */
1051
1052	/* We never need to allocate blocks for zeroing or unsharing a hole. */
1053	if ((flags & (IOMAP_UNSHARE | IOMAP_ZERO)) &&
1054	    imap.br_startoff > offset_fsb) {
1055		xfs_hole_to_iomap(ip, iomap, offset_fsb, imap.br_startoff);
1056		goto out_unlock;
1057	}
1058
1059	/*
1060	 * For zeroing, trim a delalloc extent that extends beyond the EOF
1061	 * block.  If it starts beyond the EOF block, convert it to an
1062	 * unwritten extent.
1063	 */
1064	if ((flags & IOMAP_ZERO) && imap.br_startoff <= offset_fsb &&
1065	    isnullstartblock(imap.br_startblock)) {
1066		xfs_fileoff_t eof_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
1067
1068		if (offset_fsb >= eof_fsb)
1069			goto convert_delay;
1070		if (end_fsb > eof_fsb) {
1071			end_fsb = eof_fsb;
1072			xfs_trim_extent(&imap, offset_fsb,
1073					end_fsb - offset_fsb);
1074		}
1075	}
1076
1077	/*
1078	 * Search the COW fork extent list even if we did not find a data fork
1079	 * extent.  This serves two purposes: first this implements the
1080	 * speculative preallocation using cowextsize, so that we also unshare
1081	 * block adjacent to shared blocks instead of just the shared blocks
1082	 * themselves.  Second the lookup in the extent list is generally faster
1083	 * than going out to the shared extent tree.
1084	 */
1085	if (xfs_is_cow_inode(ip)) {
1086		if (!ip->i_cowfp) {
1087			ASSERT(!xfs_is_reflink_inode(ip));
1088			xfs_ifork_init_cow(ip);
1089		}
1090		cow_eof = !xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb,
1091				&ccur, &cmap);
1092		if (!cow_eof && cmap.br_startoff <= offset_fsb) {
1093			trace_xfs_reflink_cow_found(ip, &cmap);
1094			goto found_cow;
1095		}
1096	}
1097
1098	if (imap.br_startoff <= offset_fsb) {
1099		/*
1100		 * For reflink files we may need a delalloc reservation when
1101		 * overwriting shared extents.   This includes zeroing of
1102		 * existing extents that contain data.
1103		 */
1104		if (!xfs_is_cow_inode(ip) ||
1105		    ((flags & IOMAP_ZERO) && imap.br_state != XFS_EXT_NORM)) {
1106			trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
1107					&imap);
1108			goto found_imap;
1109		}
1110
1111		xfs_trim_extent(&imap, offset_fsb, end_fsb - offset_fsb);
1112
1113		/* Trim the mapping to the nearest shared extent boundary. */
1114		error = xfs_bmap_trim_cow(ip, &imap, &shared);
1115		if (error)
1116			goto out_unlock;
1117
1118		/* Not shared?  Just report the (potentially capped) extent. */
1119		if (!shared) {
1120			trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
1121					&imap);
1122			goto found_imap;
1123		}
1124
1125		/*
1126		 * Fork all the shared blocks from our write offset until the
1127		 * end of the extent.
1128		 */
1129		allocfork = XFS_COW_FORK;
1130		end_fsb = imap.br_startoff + imap.br_blockcount;
1131	} else {
1132		/*
1133		 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES
1134		 * pages to keep the chunks of work done where somewhat
1135		 * symmetric with the work writeback does.  This is a completely
1136		 * arbitrary number pulled out of thin air.
 
1137		 *
1138		 * Note that the values needs to be less than 32-bits wide until
1139		 * the lower level functions are updated.
1140		 */
1141		count = min_t(loff_t, count, 1024 * PAGE_SIZE);
1142		end_fsb = xfs_iomap_end_fsb(mp, offset, count);
1143
1144		if (xfs_is_always_cow_inode(ip))
1145			allocfork = XFS_COW_FORK;
1146	}
1147
1148	if (eof && offset + count > XFS_ISIZE(ip)) {
1149		/*
1150		 * Determine the initial size of the preallocation.
1151		 * We clean up any extra preallocation when the file is closed.
1152		 */
1153		if (xfs_has_allocsize(mp))
1154			prealloc_blocks = mp->m_allocsize_blocks;
1155		else if (allocfork == XFS_DATA_FORK)
1156			prealloc_blocks = xfs_iomap_prealloc_size(ip, allocfork,
1157						offset, count, &icur);
1158		else
1159			prealloc_blocks = xfs_iomap_prealloc_size(ip, allocfork,
1160						offset, count, &ccur);
1161		if (prealloc_blocks) {
1162			xfs_extlen_t	align;
1163			xfs_off_t	end_offset;
1164			xfs_fileoff_t	p_end_fsb;
1165
1166			end_offset = XFS_ALLOC_ALIGN(mp, offset + count - 1);
1167			p_end_fsb = XFS_B_TO_FSBT(mp, end_offset) +
1168					prealloc_blocks;
 
1169
1170			align = xfs_eof_alignment(ip);
1171			if (align)
1172				p_end_fsb = roundup_64(p_end_fsb, align);
1173
1174			p_end_fsb = min(p_end_fsb,
1175				XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes));
1176			ASSERT(p_end_fsb > offset_fsb);
1177			prealloc_blocks = p_end_fsb - end_fsb;
1178		}
1179	}
1180
1181	/*
1182	 * Flag newly allocated delalloc blocks with IOMAP_F_NEW so we punch
1183	 * them out if the write happens to fail.
1184	 */
1185	iomap_flags |= IOMAP_F_NEW;
1186	if (allocfork == XFS_COW_FORK) {
1187		error = xfs_bmapi_reserve_delalloc(ip, allocfork, offset_fsb,
1188				end_fsb - offset_fsb, prealloc_blocks, &cmap,
1189				&ccur, cow_eof);
1190		if (error)
1191			goto out_unlock;
1192
1193		trace_xfs_iomap_alloc(ip, offset, count, allocfork, &cmap);
1194		goto found_cow;
1195	}
1196
1197	error = xfs_bmapi_reserve_delalloc(ip, allocfork, offset_fsb,
1198			end_fsb - offset_fsb, prealloc_blocks, &imap, &icur,
1199			eof);
1200	if (error)
1201		goto out_unlock;
1202
1203	trace_xfs_iomap_alloc(ip, offset, count, allocfork, &imap);
1204found_imap:
1205	seq = xfs_iomap_inode_sequence(ip, iomap_flags);
1206	xfs_iunlock(ip, lockmode);
1207	return xfs_bmbt_to_iomap(ip, iomap, &imap, flags, iomap_flags, seq);
1208
1209convert_delay:
1210	xfs_iunlock(ip, lockmode);
1211	truncate_pagecache(inode, offset);
1212	error = xfs_bmapi_convert_delalloc(ip, XFS_DATA_FORK, offset,
1213					   iomap, NULL);
1214	if (error)
1215		return error;
1216
1217	trace_xfs_iomap_alloc(ip, offset, count, XFS_DATA_FORK, &imap);
 
1218	return 0;
1219
1220found_cow:
1221	if (imap.br_startoff <= offset_fsb) {
1222		error = xfs_bmbt_to_iomap(ip, srcmap, &imap, flags, 0,
1223				xfs_iomap_inode_sequence(ip, 0));
1224		if (error)
1225			goto out_unlock;
1226	} else {
1227		xfs_trim_extent(&cmap, offset_fsb,
1228				imap.br_startoff - offset_fsb);
1229	}
1230
1231	iomap_flags |= IOMAP_F_SHARED;
1232	seq = xfs_iomap_inode_sequence(ip, iomap_flags);
1233	xfs_iunlock(ip, lockmode);
1234	return xfs_bmbt_to_iomap(ip, iomap, &cmap, flags, iomap_flags, seq);
1235
1236out_unlock:
1237	xfs_iunlock(ip, lockmode);
1238	return error;
1239}
1240
1241static void
1242xfs_buffered_write_delalloc_punch(
1243	struct inode		*inode,
1244	loff_t			offset,
1245	loff_t			length,
1246	struct iomap		*iomap)
1247{
1248	xfs_bmap_punch_delalloc_range(XFS_I(inode),
1249			(iomap->flags & IOMAP_F_SHARED) ?
1250				XFS_COW_FORK : XFS_DATA_FORK,
1251			offset, offset + length);
1252}
1253
1254static int
1255xfs_buffered_write_iomap_end(
1256	struct inode		*inode,
1257	loff_t			offset,
1258	loff_t			length,
1259	ssize_t			written,
1260	unsigned		flags,
1261	struct iomap		*iomap)
1262{
1263	loff_t			start_byte, end_byte;
 
 
 
1264
1265	/* If we didn't reserve the blocks, we're not allowed to punch them. */
1266	if (iomap->type != IOMAP_DELALLOC || !(iomap->flags & IOMAP_F_NEW))
1267		return 0;
 
 
 
 
 
1268
1269	/*
1270	 * iomap_page_mkwrite() will never fail in a way that requires delalloc
1271	 * extents that it allocated to be revoked.  Hence never try to release
1272	 * them here.
1273	 */
1274	if (flags & IOMAP_FAULT)
1275		return 0;
 
 
 
1276
1277	/* Nothing to do if we've written the entire delalloc extent */
1278	start_byte = iomap_last_written_block(inode, offset, written);
1279	end_byte = round_up(offset + length, i_blocksize(inode));
1280	if (start_byte >= end_byte)
1281		return 0;
 
 
 
 
 
 
 
 
 
 
 
1282
1283	/* For zeroing operations the callers already hold invalidate_lock. */
1284	if (flags & (IOMAP_UNSHARE | IOMAP_ZERO)) {
1285		rwsem_assert_held_write(&inode->i_mapping->invalidate_lock);
1286		iomap_write_delalloc_release(inode, start_byte, end_byte, flags,
1287				iomap, xfs_buffered_write_delalloc_punch);
1288	} else {
1289		filemap_invalidate_lock(inode->i_mapping);
1290		iomap_write_delalloc_release(inode, start_byte, end_byte, flags,
1291				iomap, xfs_buffered_write_delalloc_punch);
1292		filemap_invalidate_unlock(inode->i_mapping);
1293	}
1294
1295	return 0;
1296}
1297
1298const struct iomap_ops xfs_buffered_write_iomap_ops = {
1299	.iomap_begin		= xfs_buffered_write_iomap_begin,
1300	.iomap_end		= xfs_buffered_write_iomap_end,
1301};
1302
1303static int
1304xfs_read_iomap_begin(
1305	struct inode		*inode,
1306	loff_t			offset,
1307	loff_t			length,
 
1308	unsigned		flags,
1309	struct iomap		*iomap,
1310	struct iomap		*srcmap)
1311{
1312	struct xfs_inode	*ip = XFS_I(inode);
1313	struct xfs_mount	*mp = ip->i_mount;
1314	struct xfs_bmbt_irec	imap;
1315	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
1316	xfs_fileoff_t		end_fsb = xfs_iomap_end_fsb(mp, offset, length);
1317	int			nimaps = 1, error = 0;
1318	bool			shared = false;
1319	unsigned int		lockmode = XFS_ILOCK_SHARED;
1320	u64			seq;
1321
1322	ASSERT(!(flags & (IOMAP_WRITE | IOMAP_ZERO)));
1323
1324	if (xfs_is_shutdown(mp))
1325		return -EIO;
1326
1327	error = xfs_ilock_for_iomap(ip, flags, &lockmode);
1328	if (error)
1329		return error;
1330	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1331			       &nimaps, 0);
1332	if (!error && ((flags & IOMAP_REPORT) || IS_DAX(inode)))
1333		error = xfs_reflink_trim_around_shared(ip, &imap, &shared);
1334	seq = xfs_iomap_inode_sequence(ip, shared ? IOMAP_F_SHARED : 0);
1335	xfs_iunlock(ip, lockmode);
1336
1337	if (error)
1338		return error;
1339	trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
1340	return xfs_bmbt_to_iomap(ip, iomap, &imap, flags,
1341				 shared ? IOMAP_F_SHARED : 0, seq);
1342}
1343
1344const struct iomap_ops xfs_read_iomap_ops = {
1345	.iomap_begin		= xfs_read_iomap_begin,
1346};
1347
1348static int
1349xfs_seek_iomap_begin(
1350	struct inode		*inode,
1351	loff_t			offset,
1352	loff_t			length,
1353	unsigned		flags,
1354	struct iomap		*iomap,
1355	struct iomap		*srcmap)
1356{
1357	struct xfs_inode	*ip = XFS_I(inode);
1358	struct xfs_mount	*mp = ip->i_mount;
1359	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
1360	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, offset + length);
1361	xfs_fileoff_t		cow_fsb = NULLFILEOFF, data_fsb = NULLFILEOFF;
1362	struct xfs_iext_cursor	icur;
1363	struct xfs_bmbt_irec	imap, cmap;
1364	int			error = 0;
1365	unsigned		lockmode;
1366	u64			seq;
1367
1368	if (xfs_is_shutdown(mp))
1369		return -EIO;
1370
1371	lockmode = xfs_ilock_data_map_shared(ip);
1372	error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
1373	if (error)
1374		goto out_unlock;
1375
1376	if (xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap)) {
1377		/*
1378		 * If we found a data extent we are done.
1379		 */
1380		if (imap.br_startoff <= offset_fsb)
1381			goto done;
1382		data_fsb = imap.br_startoff;
1383	} else {
1384		/*
1385		 * Fake a hole until the end of the file.
1386		 */
1387		data_fsb = xfs_iomap_end_fsb(mp, offset, length);
1388	}
1389
1390	/*
1391	 * If a COW fork extent covers the hole, report it - capped to the next
1392	 * data fork extent:
1393	 */
1394	if (xfs_inode_has_cow_data(ip) &&
1395	    xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &cmap))
1396		cow_fsb = cmap.br_startoff;
1397	if (cow_fsb != NULLFILEOFF && cow_fsb <= offset_fsb) {
1398		if (data_fsb < cow_fsb + cmap.br_blockcount)
1399			end_fsb = min(end_fsb, data_fsb);
1400		xfs_trim_extent(&cmap, offset_fsb, end_fsb - offset_fsb);
1401		seq = xfs_iomap_inode_sequence(ip, IOMAP_F_SHARED);
1402		error = xfs_bmbt_to_iomap(ip, iomap, &cmap, flags,
1403				IOMAP_F_SHARED, seq);
1404		/*
1405		 * This is a COW extent, so we must probe the page cache
1406		 * because there could be dirty page cache being backed
1407		 * by this extent.
1408		 */
1409		iomap->type = IOMAP_UNWRITTEN;
1410		goto out_unlock;
1411	}
1412
1413	/*
1414	 * Else report a hole, capped to the next found data or COW extent.
1415	 */
1416	if (cow_fsb != NULLFILEOFF && cow_fsb < data_fsb)
1417		imap.br_blockcount = cow_fsb - offset_fsb;
1418	else
1419		imap.br_blockcount = data_fsb - offset_fsb;
1420	imap.br_startoff = offset_fsb;
1421	imap.br_startblock = HOLESTARTBLOCK;
1422	imap.br_state = XFS_EXT_NORM;
1423done:
1424	seq = xfs_iomap_inode_sequence(ip, 0);
1425	xfs_trim_extent(&imap, offset_fsb, end_fsb - offset_fsb);
1426	error = xfs_bmbt_to_iomap(ip, iomap, &imap, flags, 0, seq);
1427out_unlock:
1428	xfs_iunlock(ip, lockmode);
1429	return error;
1430}
1431
1432const struct iomap_ops xfs_seek_iomap_ops = {
1433	.iomap_begin		= xfs_seek_iomap_begin,
1434};
1435
1436static int
1437xfs_xattr_iomap_begin(
1438	struct inode		*inode,
1439	loff_t			offset,
1440	loff_t			length,
1441	unsigned		flags,
1442	struct iomap		*iomap,
1443	struct iomap		*srcmap)
1444{
1445	struct xfs_inode	*ip = XFS_I(inode);
1446	struct xfs_mount	*mp = ip->i_mount;
1447	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
1448	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, offset + length);
1449	struct xfs_bmbt_irec	imap;
1450	int			nimaps = 1, error = 0;
1451	unsigned		lockmode;
1452	int			seq;
1453
1454	if (xfs_is_shutdown(mp))
1455		return -EIO;
1456
1457	lockmode = xfs_ilock_attr_map_shared(ip);
1458
1459	/* if there are no attribute fork or extents, return ENOENT */
1460	if (!xfs_inode_has_attr_fork(ip) || !ip->i_af.if_nextents) {
1461		error = -ENOENT;
1462		goto out_unlock;
1463	}
1464
1465	ASSERT(ip->i_af.if_format != XFS_DINODE_FMT_LOCAL);
1466	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1467			       &nimaps, XFS_BMAPI_ATTRFORK);
1468out_unlock:
 
1469
1470	seq = xfs_iomap_inode_sequence(ip, IOMAP_F_XATTR);
1471	xfs_iunlock(ip, lockmode);
 
 
1472
1473	if (error)
1474		return error;
1475	ASSERT(nimaps);
1476	return xfs_bmbt_to_iomap(ip, iomap, &imap, flags, IOMAP_F_XATTR, seq);
1477}
1478
1479const struct iomap_ops xfs_xattr_iomap_ops = {
1480	.iomap_begin		= xfs_xattr_iomap_begin,
1481};
1482
1483int
1484xfs_zero_range(
1485	struct xfs_inode	*ip,
1486	loff_t			pos,
1487	loff_t			len,
1488	bool			*did_zero)
1489{
1490	struct inode		*inode = VFS_I(ip);
1491
1492	xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL);
1493
1494	if (IS_DAX(inode))
1495		return dax_zero_range(inode, pos, len, did_zero,
1496				      &xfs_dax_write_iomap_ops);
1497	return iomap_zero_range(inode, pos, len, did_zero,
1498				&xfs_buffered_write_iomap_ops);
1499}
1500
1501int
1502xfs_truncate_page(
1503	struct xfs_inode	*ip,
1504	loff_t			pos,
1505	bool			*did_zero)
1506{
1507	struct inode		*inode = VFS_I(ip);
1508
1509	if (IS_DAX(inode))
1510		return dax_truncate_page(inode, pos, did_zero,
1511					&xfs_dax_write_iomap_ops);
1512	return iomap_truncate_page(inode, pos, did_zero,
1513				   &xfs_buffered_write_iomap_ops);
1514}