Linux Audio

Check our new training course

Linux debugging, profiling, tracing and performance analysis training

Mar 24-27, 2025, special US time zones
Register
Loading...
v4.17
   1/*
   2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
   3 * Copyright (C) 2010 Red Hat, Inc.
   4 * All Rights Reserved.
   5 *
   6 * This program is free software; you can redistribute it and/or
   7 * modify it under the terms of the GNU General Public License as
   8 * published by the Free Software Foundation.
   9 *
  10 * This program is distributed in the hope that it would be useful,
  11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13 * GNU General Public License for more details.
  14 *
  15 * You should have received a copy of the GNU General Public License
  16 * along with this program; if not, write the Free Software Foundation,
  17 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  18 */
  19#include "xfs.h"
  20#include "xfs_fs.h"
  21#include "xfs_shared.h"
  22#include "xfs_format.h"
  23#include "xfs_log_format.h"
  24#include "xfs_trans_resv.h"
  25#include "xfs_mount.h"
  26#include "xfs_inode.h"
  27#include "xfs_extent_busy.h"
  28#include "xfs_quota.h"
  29#include "xfs_trans.h"
  30#include "xfs_trans_priv.h"
  31#include "xfs_log.h"
  32#include "xfs_trace.h"
  33#include "xfs_error.h"
  34
  35kmem_zone_t	*xfs_trans_zone;
  36kmem_zone_t	*xfs_log_item_desc_zone;
  37
  38#if defined(CONFIG_TRACEPOINTS)
  39static void
  40xfs_trans_trace_reservations(
  41	struct xfs_mount	*mp)
  42{
  43	struct xfs_trans_res	resv;
  44	struct xfs_trans_res	*res;
  45	struct xfs_trans_res	*end_res;
  46	int			i;
  47
  48	res = (struct xfs_trans_res *)M_RES(mp);
  49	end_res = (struct xfs_trans_res *)(M_RES(mp) + 1);
  50	for (i = 0; res < end_res; i++, res++)
  51		trace_xfs_trans_resv_calc(mp, i, res);
  52	xfs_log_get_max_trans_res(mp, &resv);
  53	trace_xfs_trans_resv_calc(mp, -1, &resv);
  54}
  55#else
  56# define xfs_trans_trace_reservations(mp)
  57#endif
  58
  59/*
  60 * Initialize the precomputed transaction reservation values
  61 * in the mount structure.
  62 */
  63void
  64xfs_trans_init(
  65	struct xfs_mount	*mp)
  66{
  67	xfs_trans_resv_calc(mp, M_RES(mp));
  68	xfs_trans_trace_reservations(mp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  69}
  70
  71/*
  72 * Free the transaction structure.  If there is more clean up
  73 * to do when the structure is freed, add it here.
  74 */
  75STATIC void
  76xfs_trans_free(
  77	struct xfs_trans	*tp)
  78{
  79	xfs_extent_busy_sort(&tp->t_busy);
  80	xfs_extent_busy_clear(tp->t_mountp, &tp->t_busy, false);
  81
  82	atomic_dec(&tp->t_mountp->m_active_trans);
  83	if (!(tp->t_flags & XFS_TRANS_NO_WRITECOUNT))
  84		sb_end_intwrite(tp->t_mountp->m_super);
  85	xfs_trans_free_dqinfo(tp);
  86	kmem_zone_free(xfs_trans_zone, tp);
  87}
  88
  89/*
  90 * This is called to create a new transaction which will share the
  91 * permanent log reservation of the given transaction.  The remaining
  92 * unused block and rt extent reservations are also inherited.  This
  93 * implies that the original transaction is no longer allowed to allocate
  94 * blocks.  Locks and log items, however, are no inherited.  They must
  95 * be added to the new transaction explicitly.
  96 */
  97STATIC xfs_trans_t *
  98xfs_trans_dup(
  99	xfs_trans_t	*tp)
 100{
 101	xfs_trans_t	*ntp;
 102
 103	ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
 104
 105	/*
 106	 * Initialize the new transaction structure.
 107	 */
 108	ntp->t_magic = XFS_TRANS_HEADER_MAGIC;
 
 109	ntp->t_mountp = tp->t_mountp;
 110	INIT_LIST_HEAD(&ntp->t_items);
 111	INIT_LIST_HEAD(&ntp->t_busy);
 112
 113	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
 114	ASSERT(tp->t_ticket != NULL);
 115
 116	ntp->t_flags = XFS_TRANS_PERM_LOG_RES |
 117		       (tp->t_flags & XFS_TRANS_RESERVE) |
 118		       (tp->t_flags & XFS_TRANS_NO_WRITECOUNT);
 119	/* We gave our writer reference to the new transaction */
 120	tp->t_flags |= XFS_TRANS_NO_WRITECOUNT;
 121	ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket);
 122
 123	ASSERT(tp->t_blk_res >= tp->t_blk_res_used);
 124	ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
 125	tp->t_blk_res = tp->t_blk_res_used;
 126
 127	ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
 128	tp->t_rtx_res = tp->t_rtx_res_used;
 129	ntp->t_pflags = tp->t_pflags;
 130
 131	xfs_trans_dup_dqinfo(tp, ntp);
 132
 133	atomic_inc(&tp->t_mountp->m_active_trans);
 134	return ntp;
 135}
 136
 137/*
 138 * This is called to reserve free disk blocks and log space for the
 139 * given transaction.  This must be done before allocating any resources
 140 * within the transaction.
 141 *
 142 * This will return ENOSPC if there are not enough blocks available.
 143 * It will sleep waiting for available log space.
 144 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
 145 * is used by long running transactions.  If any one of the reservations
 146 * fails then they will all be backed out.
 147 *
 148 * This does not do quota reservations. That typically is done by the
 149 * caller afterwards.
 150 */
 151static int
 152xfs_trans_reserve(
 153	struct xfs_trans	*tp,
 154	struct xfs_trans_res	*resp,
 155	uint			blocks,
 156	uint			rtextents)
 157{
 158	int		error = 0;
 159	bool		rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
 160
 161	/* Mark this thread as being in a transaction */
 162	current_set_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
 163
 164	/*
 165	 * Attempt to reserve the needed disk blocks by decrementing
 166	 * the number needed from the number available.  This will
 167	 * fail if the count would go below zero.
 168	 */
 169	if (blocks > 0) {
 170		error = xfs_mod_fdblocks(tp->t_mountp, -((int64_t)blocks), rsvd);
 171		if (error != 0) {
 172			current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
 173			return -ENOSPC;
 174		}
 175		tp->t_blk_res += blocks;
 176	}
 177
 178	/*
 179	 * Reserve the log space needed for this transaction.
 180	 */
 181	if (resp->tr_logres > 0) {
 182		bool	permanent = false;
 183
 184		ASSERT(tp->t_log_res == 0 ||
 185		       tp->t_log_res == resp->tr_logres);
 186		ASSERT(tp->t_log_count == 0 ||
 187		       tp->t_log_count == resp->tr_logcount);
 188
 189		if (resp->tr_logflags & XFS_TRANS_PERM_LOG_RES) {
 190			tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
 191			permanent = true;
 192		} else {
 193			ASSERT(tp->t_ticket == NULL);
 194			ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
 195		}
 196
 197		if (tp->t_ticket != NULL) {
 198			ASSERT(resp->tr_logflags & XFS_TRANS_PERM_LOG_RES);
 199			error = xfs_log_regrant(tp->t_mountp, tp->t_ticket);
 200		} else {
 201			error = xfs_log_reserve(tp->t_mountp,
 202						resp->tr_logres,
 203						resp->tr_logcount,
 204						&tp->t_ticket, XFS_TRANSACTION,
 205						permanent);
 206		}
 207
 208		if (error)
 209			goto undo_blocks;
 210
 211		tp->t_log_res = resp->tr_logres;
 212		tp->t_log_count = resp->tr_logcount;
 213	}
 214
 215	/*
 216	 * Attempt to reserve the needed realtime extents by decrementing
 217	 * the number needed from the number available.  This will
 218	 * fail if the count would go below zero.
 219	 */
 220	if (rtextents > 0) {
 221		error = xfs_mod_frextents(tp->t_mountp, -((int64_t)rtextents));
 222		if (error) {
 223			error = -ENOSPC;
 224			goto undo_log;
 225		}
 226		tp->t_rtx_res += rtextents;
 227	}
 228
 229	return 0;
 230
 231	/*
 232	 * Error cases jump to one of these labels to undo any
 233	 * reservations which have already been performed.
 234	 */
 235undo_log:
 236	if (resp->tr_logres > 0) {
 237		xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, false);
 238		tp->t_ticket = NULL;
 239		tp->t_log_res = 0;
 240		tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
 241	}
 242
 243undo_blocks:
 244	if (blocks > 0) {
 245		xfs_mod_fdblocks(tp->t_mountp, (int64_t)blocks, rsvd);
 246		tp->t_blk_res = 0;
 247	}
 248
 249	current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
 250
 251	return error;
 252}
 253
 254int
 255xfs_trans_alloc(
 256	struct xfs_mount	*mp,
 257	struct xfs_trans_res	*resp,
 258	uint			blocks,
 259	uint			rtextents,
 260	uint			flags,
 261	struct xfs_trans	**tpp)
 262{
 263	struct xfs_trans	*tp;
 264	int			error;
 265
 266	if (!(flags & XFS_TRANS_NO_WRITECOUNT))
 267		sb_start_intwrite(mp->m_super);
 268
 269	WARN_ON(mp->m_super->s_writers.frozen == SB_FREEZE_COMPLETE);
 270	atomic_inc(&mp->m_active_trans);
 271
 272	tp = kmem_zone_zalloc(xfs_trans_zone,
 273		(flags & XFS_TRANS_NOFS) ? KM_NOFS : KM_SLEEP);
 274	tp->t_magic = XFS_TRANS_HEADER_MAGIC;
 275	tp->t_flags = flags;
 276	tp->t_mountp = mp;
 277	INIT_LIST_HEAD(&tp->t_items);
 278	INIT_LIST_HEAD(&tp->t_busy);
 279
 280	error = xfs_trans_reserve(tp, resp, blocks, rtextents);
 281	if (error) {
 282		xfs_trans_cancel(tp);
 283		return error;
 284	}
 285
 286	*tpp = tp;
 287	return 0;
 288}
 289
 290/*
 291 * Create an empty transaction with no reservation.  This is a defensive
 292 * mechanism for routines that query metadata without actually modifying
 293 * them -- if the metadata being queried is somehow cross-linked (think a
 294 * btree block pointer that points higher in the tree), we risk deadlock.
 295 * However, blocks grabbed as part of a transaction can be re-grabbed.
 296 * The verifiers will notice the corrupt block and the operation will fail
 297 * back to userspace without deadlocking.
 298 *
 299 * Note the zero-length reservation; this transaction MUST be cancelled
 300 * without any dirty data.
 301 */
 302int
 303xfs_trans_alloc_empty(
 304	struct xfs_mount		*mp,
 305	struct xfs_trans		**tpp)
 306{
 307	struct xfs_trans_res		resv = {0};
 308
 309	return xfs_trans_alloc(mp, &resv, 0, 0, XFS_TRANS_NO_WRITECOUNT, tpp);
 310}
 311
 312/*
 313 * Record the indicated change to the given field for application
 314 * to the file system's superblock when the transaction commits.
 315 * For now, just store the change in the transaction structure.
 316 *
 317 * Mark the transaction structure to indicate that the superblock
 318 * needs to be updated before committing.
 319 *
 320 * Because we may not be keeping track of allocated/free inodes and
 321 * used filesystem blocks in the superblock, we do not mark the
 322 * superblock dirty in this transaction if we modify these fields.
 323 * We still need to update the transaction deltas so that they get
 324 * applied to the incore superblock, but we don't want them to
 325 * cause the superblock to get locked and logged if these are the
 326 * only fields in the superblock that the transaction modifies.
 327 */
 328void
 329xfs_trans_mod_sb(
 330	xfs_trans_t	*tp,
 331	uint		field,
 332	int64_t		delta)
 333{
 334	uint32_t	flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
 335	xfs_mount_t	*mp = tp->t_mountp;
 336
 337	switch (field) {
 338	case XFS_TRANS_SB_ICOUNT:
 339		tp->t_icount_delta += delta;
 340		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
 341			flags &= ~XFS_TRANS_SB_DIRTY;
 342		break;
 343	case XFS_TRANS_SB_IFREE:
 344		tp->t_ifree_delta += delta;
 345		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
 346			flags &= ~XFS_TRANS_SB_DIRTY;
 347		break;
 348	case XFS_TRANS_SB_FDBLOCKS:
 349		/*
 350		 * Track the number of blocks allocated in the transaction.
 351		 * Make sure it does not exceed the number reserved. If so,
 352		 * shutdown as this can lead to accounting inconsistency.
 353		 */
 354		if (delta < 0) {
 355			tp->t_blk_res_used += (uint)-delta;
 356			if (tp->t_blk_res_used > tp->t_blk_res)
 357				xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
 358		}
 359		tp->t_fdblocks_delta += delta;
 360		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
 361			flags &= ~XFS_TRANS_SB_DIRTY;
 362		break;
 363	case XFS_TRANS_SB_RES_FDBLOCKS:
 364		/*
 365		 * The allocation has already been applied to the
 366		 * in-core superblock's counter.  This should only
 367		 * be applied to the on-disk superblock.
 368		 */
 
 369		tp->t_res_fdblocks_delta += delta;
 370		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
 371			flags &= ~XFS_TRANS_SB_DIRTY;
 372		break;
 373	case XFS_TRANS_SB_FREXTENTS:
 374		/*
 375		 * Track the number of blocks allocated in the
 376		 * transaction.  Make sure it does not exceed the
 377		 * number reserved.
 378		 */
 379		if (delta < 0) {
 380			tp->t_rtx_res_used += (uint)-delta;
 381			ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
 382		}
 383		tp->t_frextents_delta += delta;
 384		break;
 385	case XFS_TRANS_SB_RES_FREXTENTS:
 386		/*
 387		 * The allocation has already been applied to the
 388		 * in-core superblock's counter.  This should only
 389		 * be applied to the on-disk superblock.
 390		 */
 391		ASSERT(delta < 0);
 392		tp->t_res_frextents_delta += delta;
 393		break;
 394	case XFS_TRANS_SB_DBLOCKS:
 395		ASSERT(delta > 0);
 396		tp->t_dblocks_delta += delta;
 397		break;
 398	case XFS_TRANS_SB_AGCOUNT:
 399		ASSERT(delta > 0);
 400		tp->t_agcount_delta += delta;
 401		break;
 402	case XFS_TRANS_SB_IMAXPCT:
 403		tp->t_imaxpct_delta += delta;
 404		break;
 405	case XFS_TRANS_SB_REXTSIZE:
 406		tp->t_rextsize_delta += delta;
 407		break;
 408	case XFS_TRANS_SB_RBMBLOCKS:
 409		tp->t_rbmblocks_delta += delta;
 410		break;
 411	case XFS_TRANS_SB_RBLOCKS:
 412		tp->t_rblocks_delta += delta;
 413		break;
 414	case XFS_TRANS_SB_REXTENTS:
 415		tp->t_rextents_delta += delta;
 416		break;
 417	case XFS_TRANS_SB_REXTSLOG:
 418		tp->t_rextslog_delta += delta;
 419		break;
 420	default:
 421		ASSERT(0);
 422		return;
 423	}
 424
 425	tp->t_flags |= flags;
 426}
 427
 428/*
 429 * xfs_trans_apply_sb_deltas() is called from the commit code
 430 * to bring the superblock buffer into the current transaction
 431 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
 432 *
 433 * For now we just look at each field allowed to change and change
 434 * it if necessary.
 435 */
 436STATIC void
 437xfs_trans_apply_sb_deltas(
 438	xfs_trans_t	*tp)
 439{
 440	xfs_dsb_t	*sbp;
 441	xfs_buf_t	*bp;
 442	int		whole = 0;
 443
 444	bp = xfs_trans_getsb(tp, tp->t_mountp, 0);
 445	sbp = XFS_BUF_TO_SBP(bp);
 446
 447	/*
 448	 * Check that superblock mods match the mods made to AGF counters.
 449	 */
 450	ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
 451	       (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
 452		tp->t_ag_btree_delta));
 453
 454	/*
 455	 * Only update the superblock counters if we are logging them
 456	 */
 457	if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) {
 458		if (tp->t_icount_delta)
 459			be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
 460		if (tp->t_ifree_delta)
 461			be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
 462		if (tp->t_fdblocks_delta)
 463			be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
 464		if (tp->t_res_fdblocks_delta)
 465			be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
 466	}
 467
 468	if (tp->t_frextents_delta)
 469		be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta);
 470	if (tp->t_res_frextents_delta)
 471		be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta);
 472
 473	if (tp->t_dblocks_delta) {
 474		be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
 475		whole = 1;
 476	}
 477	if (tp->t_agcount_delta) {
 478		be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
 479		whole = 1;
 480	}
 481	if (tp->t_imaxpct_delta) {
 482		sbp->sb_imax_pct += tp->t_imaxpct_delta;
 483		whole = 1;
 484	}
 485	if (tp->t_rextsize_delta) {
 486		be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
 487		whole = 1;
 488	}
 489	if (tp->t_rbmblocks_delta) {
 490		be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
 491		whole = 1;
 492	}
 493	if (tp->t_rblocks_delta) {
 494		be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
 495		whole = 1;
 496	}
 497	if (tp->t_rextents_delta) {
 498		be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
 499		whole = 1;
 500	}
 501	if (tp->t_rextslog_delta) {
 502		sbp->sb_rextslog += tp->t_rextslog_delta;
 503		whole = 1;
 504	}
 505
 506	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
 507	if (whole)
 508		/*
 509		 * Log the whole thing, the fields are noncontiguous.
 510		 */
 511		xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1);
 512	else
 513		/*
 514		 * Since all the modifiable fields are contiguous, we
 515		 * can get away with this.
 516		 */
 517		xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount),
 518				  offsetof(xfs_dsb_t, sb_frextents) +
 519				  sizeof(sbp->sb_frextents) - 1);
 520}
 521
 522STATIC int
 523xfs_sb_mod8(
 524	uint8_t			*field,
 525	int8_t			delta)
 526{
 527	int8_t			counter = *field;
 528
 529	counter += delta;
 530	if (counter < 0) {
 531		ASSERT(0);
 532		return -EINVAL;
 533	}
 534	*field = counter;
 535	return 0;
 536}
 537
 538STATIC int
 539xfs_sb_mod32(
 540	uint32_t		*field,
 541	int32_t			delta)
 542{
 543	int32_t			counter = *field;
 544
 545	counter += delta;
 546	if (counter < 0) {
 547		ASSERT(0);
 548		return -EINVAL;
 549	}
 550	*field = counter;
 551	return 0;
 552}
 553
 554STATIC int
 555xfs_sb_mod64(
 556	uint64_t		*field,
 557	int64_t			delta)
 558{
 559	int64_t			counter = *field;
 560
 561	counter += delta;
 562	if (counter < 0) {
 563		ASSERT(0);
 564		return -EINVAL;
 565	}
 566	*field = counter;
 567	return 0;
 568}
 569
 570/*
 571 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations
 572 * and apply superblock counter changes to the in-core superblock.  The
 573 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
 574 * applied to the in-core superblock.  The idea is that that has already been
 575 * done.
 576 *
 577 * If we are not logging superblock counters, then the inode allocated/free and
 578 * used block counts are not updated in the on disk superblock. In this case,
 579 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
 580 * still need to update the incore superblock with the changes.
 581 */
 582void
 583xfs_trans_unreserve_and_mod_sb(
 584	struct xfs_trans	*tp)
 585{
 586	struct xfs_mount	*mp = tp->t_mountp;
 587	bool			rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
 588	int64_t			blkdelta = 0;
 589	int64_t			rtxdelta = 0;
 590	int64_t			idelta = 0;
 591	int64_t			ifreedelta = 0;
 592	int			error;
 593
 594	/* calculate deltas */
 595	if (tp->t_blk_res > 0)
 596		blkdelta = tp->t_blk_res;
 597	if ((tp->t_fdblocks_delta != 0) &&
 598	    (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
 599	     (tp->t_flags & XFS_TRANS_SB_DIRTY)))
 600	        blkdelta += tp->t_fdblocks_delta;
 601
 602	if (tp->t_rtx_res > 0)
 603		rtxdelta = tp->t_rtx_res;
 604	if ((tp->t_frextents_delta != 0) &&
 605	    (tp->t_flags & XFS_TRANS_SB_DIRTY))
 606		rtxdelta += tp->t_frextents_delta;
 607
 608	if (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
 609	     (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
 610		idelta = tp->t_icount_delta;
 611		ifreedelta = tp->t_ifree_delta;
 612	}
 613
 614	/* apply the per-cpu counters */
 615	if (blkdelta) {
 616		error = xfs_mod_fdblocks(mp, blkdelta, rsvd);
 617		if (error)
 618			goto out;
 619	}
 620
 621	if (idelta) {
 622		error = xfs_mod_icount(mp, idelta);
 623		if (error)
 624			goto out_undo_fdblocks;
 625	}
 626
 627	if (ifreedelta) {
 628		error = xfs_mod_ifree(mp, ifreedelta);
 629		if (error)
 630			goto out_undo_icount;
 631	}
 632
 633	if (rtxdelta == 0 && !(tp->t_flags & XFS_TRANS_SB_DIRTY))
 634		return;
 635
 636	/* apply remaining deltas */
 637	spin_lock(&mp->m_sb_lock);
 638	if (rtxdelta) {
 639		error = xfs_sb_mod64(&mp->m_sb.sb_frextents, rtxdelta);
 640		if (error)
 641			goto out_undo_ifree;
 642	}
 643
 644	if (tp->t_dblocks_delta != 0) {
 645		error = xfs_sb_mod64(&mp->m_sb.sb_dblocks, tp->t_dblocks_delta);
 646		if (error)
 647			goto out_undo_frextents;
 648	}
 649	if (tp->t_agcount_delta != 0) {
 650		error = xfs_sb_mod32(&mp->m_sb.sb_agcount, tp->t_agcount_delta);
 651		if (error)
 652			goto out_undo_dblocks;
 653	}
 654	if (tp->t_imaxpct_delta != 0) {
 655		error = xfs_sb_mod8(&mp->m_sb.sb_imax_pct, tp->t_imaxpct_delta);
 656		if (error)
 657			goto out_undo_agcount;
 658	}
 659	if (tp->t_rextsize_delta != 0) {
 660		error = xfs_sb_mod32(&mp->m_sb.sb_rextsize,
 661				     tp->t_rextsize_delta);
 662		if (error)
 663			goto out_undo_imaxpct;
 664	}
 665	if (tp->t_rbmblocks_delta != 0) {
 666		error = xfs_sb_mod32(&mp->m_sb.sb_rbmblocks,
 667				     tp->t_rbmblocks_delta);
 668		if (error)
 669			goto out_undo_rextsize;
 670	}
 671	if (tp->t_rblocks_delta != 0) {
 672		error = xfs_sb_mod64(&mp->m_sb.sb_rblocks, tp->t_rblocks_delta);
 673		if (error)
 674			goto out_undo_rbmblocks;
 675	}
 676	if (tp->t_rextents_delta != 0) {
 677		error = xfs_sb_mod64(&mp->m_sb.sb_rextents,
 678				     tp->t_rextents_delta);
 679		if (error)
 680			goto out_undo_rblocks;
 681	}
 682	if (tp->t_rextslog_delta != 0) {
 683		error = xfs_sb_mod8(&mp->m_sb.sb_rextslog,
 684				     tp->t_rextslog_delta);
 685		if (error)
 686			goto out_undo_rextents;
 687	}
 688	spin_unlock(&mp->m_sb_lock);
 689	return;
 690
 691out_undo_rextents:
 692	if (tp->t_rextents_delta)
 693		xfs_sb_mod64(&mp->m_sb.sb_rextents, -tp->t_rextents_delta);
 694out_undo_rblocks:
 695	if (tp->t_rblocks_delta)
 696		xfs_sb_mod64(&mp->m_sb.sb_rblocks, -tp->t_rblocks_delta);
 697out_undo_rbmblocks:
 698	if (tp->t_rbmblocks_delta)
 699		xfs_sb_mod32(&mp->m_sb.sb_rbmblocks, -tp->t_rbmblocks_delta);
 700out_undo_rextsize:
 701	if (tp->t_rextsize_delta)
 702		xfs_sb_mod32(&mp->m_sb.sb_rextsize, -tp->t_rextsize_delta);
 703out_undo_imaxpct:
 704	if (tp->t_rextsize_delta)
 705		xfs_sb_mod8(&mp->m_sb.sb_imax_pct, -tp->t_imaxpct_delta);
 706out_undo_agcount:
 707	if (tp->t_agcount_delta)
 708		xfs_sb_mod32(&mp->m_sb.sb_agcount, -tp->t_agcount_delta);
 709out_undo_dblocks:
 710	if (tp->t_dblocks_delta)
 711		xfs_sb_mod64(&mp->m_sb.sb_dblocks, -tp->t_dblocks_delta);
 712out_undo_frextents:
 713	if (rtxdelta)
 714		xfs_sb_mod64(&mp->m_sb.sb_frextents, -rtxdelta);
 715out_undo_ifree:
 716	spin_unlock(&mp->m_sb_lock);
 717	if (ifreedelta)
 718		xfs_mod_ifree(mp, -ifreedelta);
 719out_undo_icount:
 720	if (idelta)
 721		xfs_mod_icount(mp, -idelta);
 722out_undo_fdblocks:
 723	if (blkdelta)
 724		xfs_mod_fdblocks(mp, -blkdelta, rsvd);
 725out:
 726	ASSERT(error == 0);
 727	return;
 728}
 729
 730/*
 731 * Add the given log item to the transaction's list of log items.
 732 *
 733 * The log item will now point to its new descriptor with its li_desc field.
 734 */
 735void
 736xfs_trans_add_item(
 737	struct xfs_trans	*tp,
 738	struct xfs_log_item	*lip)
 739{
 740	struct xfs_log_item_desc *lidp;
 741
 742	ASSERT(lip->li_mountp == tp->t_mountp);
 743	ASSERT(lip->li_ailp == tp->t_mountp->m_ail);
 744
 745	lidp = kmem_zone_zalloc(xfs_log_item_desc_zone, KM_SLEEP | KM_NOFS);
 746
 747	lidp->lid_item = lip;
 748	lidp->lid_flags = 0;
 749	list_add_tail(&lidp->lid_trans, &tp->t_items);
 750
 751	lip->li_desc = lidp;
 752}
 753
 754STATIC void
 755xfs_trans_free_item_desc(
 756	struct xfs_log_item_desc *lidp)
 757{
 758	list_del_init(&lidp->lid_trans);
 759	kmem_zone_free(xfs_log_item_desc_zone, lidp);
 760}
 761
 762/*
 763 * Unlink and free the given descriptor.
 764 */
 765void
 766xfs_trans_del_item(
 767	struct xfs_log_item	*lip)
 768{
 769	xfs_trans_free_item_desc(lip->li_desc);
 770	lip->li_desc = NULL;
 771}
 772
 773/*
 774 * Unlock all of the items of a transaction and free all the descriptors
 775 * of that transaction.
 776 */
 777void
 778xfs_trans_free_items(
 779	struct xfs_trans	*tp,
 780	xfs_lsn_t		commit_lsn,
 781	bool			abort)
 782{
 783	struct xfs_log_item_desc *lidp, *next;
 784
 785	list_for_each_entry_safe(lidp, next, &tp->t_items, lid_trans) {
 786		struct xfs_log_item	*lip = lidp->lid_item;
 787
 788		lip->li_desc = NULL;
 789
 790		if (commit_lsn != NULLCOMMITLSN)
 791			lip->li_ops->iop_committing(lip, commit_lsn);
 792		if (abort)
 793			lip->li_flags |= XFS_LI_ABORTED;
 794		lip->li_ops->iop_unlock(lip);
 795
 796		xfs_trans_free_item_desc(lidp);
 797	}
 798}
 799
 800static inline void
 801xfs_log_item_batch_insert(
 802	struct xfs_ail		*ailp,
 803	struct xfs_ail_cursor	*cur,
 804	struct xfs_log_item	**log_items,
 805	int			nr_items,
 806	xfs_lsn_t		commit_lsn)
 807{
 808	int	i;
 809
 810	spin_lock(&ailp->ail_lock);
 811	/* xfs_trans_ail_update_bulk drops ailp->ail_lock */
 812	xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn);
 813
 814	for (i = 0; i < nr_items; i++) {
 815		struct xfs_log_item *lip = log_items[i];
 816
 817		lip->li_ops->iop_unpin(lip, 0);
 818	}
 819}
 820
 821/*
 822 * Bulk operation version of xfs_trans_committed that takes a log vector of
 823 * items to insert into the AIL. This uses bulk AIL insertion techniques to
 824 * minimise lock traffic.
 825 *
 826 * If we are called with the aborted flag set, it is because a log write during
 827 * a CIL checkpoint commit has failed. In this case, all the items in the
 828 * checkpoint have already gone through iop_commited and iop_unlock, which
 829 * means that checkpoint commit abort handling is treated exactly the same
 830 * as an iclog write error even though we haven't started any IO yet. Hence in
 831 * this case all we need to do is iop_committed processing, followed by an
 832 * iop_unpin(aborted) call.
 833 *
 834 * The AIL cursor is used to optimise the insert process. If commit_lsn is not
 835 * at the end of the AIL, the insert cursor avoids the need to walk
 836 * the AIL to find the insertion point on every xfs_log_item_batch_insert()
 837 * call. This saves a lot of needless list walking and is a net win, even
 838 * though it slightly increases that amount of AIL lock traffic to set it up
 839 * and tear it down.
 840 */
 841void
 842xfs_trans_committed_bulk(
 843	struct xfs_ail		*ailp,
 844	struct xfs_log_vec	*log_vector,
 845	xfs_lsn_t		commit_lsn,
 846	int			aborted)
 847{
 848#define LOG_ITEM_BATCH_SIZE	32
 849	struct xfs_log_item	*log_items[LOG_ITEM_BATCH_SIZE];
 850	struct xfs_log_vec	*lv;
 851	struct xfs_ail_cursor	cur;
 852	int			i = 0;
 853
 854	spin_lock(&ailp->ail_lock);
 855	xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn);
 856	spin_unlock(&ailp->ail_lock);
 857
 858	/* unpin all the log items */
 859	for (lv = log_vector; lv; lv = lv->lv_next ) {
 860		struct xfs_log_item	*lip = lv->lv_item;
 861		xfs_lsn_t		item_lsn;
 862
 863		if (aborted)
 864			lip->li_flags |= XFS_LI_ABORTED;
 865		item_lsn = lip->li_ops->iop_committed(lip, commit_lsn);
 866
 867		/* item_lsn of -1 means the item needs no further processing */
 868		if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
 869			continue;
 870
 871		/*
 872		 * if we are aborting the operation, no point in inserting the
 873		 * object into the AIL as we are in a shutdown situation.
 874		 */
 875		if (aborted) {
 876			ASSERT(XFS_FORCED_SHUTDOWN(ailp->ail_mount));
 877			lip->li_ops->iop_unpin(lip, 1);
 878			continue;
 879		}
 880
 881		if (item_lsn != commit_lsn) {
 882
 883			/*
 884			 * Not a bulk update option due to unusual item_lsn.
 885			 * Push into AIL immediately, rechecking the lsn once
 886			 * we have the ail lock. Then unpin the item. This does
 887			 * not affect the AIL cursor the bulk insert path is
 888			 * using.
 889			 */
 890			spin_lock(&ailp->ail_lock);
 891			if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
 892				xfs_trans_ail_update(ailp, lip, item_lsn);
 893			else
 894				spin_unlock(&ailp->ail_lock);
 895			lip->li_ops->iop_unpin(lip, 0);
 896			continue;
 897		}
 898
 899		/* Item is a candidate for bulk AIL insert.  */
 900		log_items[i++] = lv->lv_item;
 901		if (i >= LOG_ITEM_BATCH_SIZE) {
 902			xfs_log_item_batch_insert(ailp, &cur, log_items,
 903					LOG_ITEM_BATCH_SIZE, commit_lsn);
 904			i = 0;
 905		}
 906	}
 907
 908	/* make sure we insert the remainder! */
 909	if (i)
 910		xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn);
 911
 912	spin_lock(&ailp->ail_lock);
 913	xfs_trans_ail_cursor_done(&cur);
 914	spin_unlock(&ailp->ail_lock);
 915}
 916
 917/*
 918 * Commit the given transaction to the log.
 919 *
 920 * XFS disk error handling mechanism is not based on a typical
 921 * transaction abort mechanism. Logically after the filesystem
 922 * gets marked 'SHUTDOWN', we can't let any new transactions
 923 * be durable - ie. committed to disk - because some metadata might
 924 * be inconsistent. In such cases, this returns an error, and the
 925 * caller may assume that all locked objects joined to the transaction
 926 * have already been unlocked as if the commit had succeeded.
 927 * Do not reference the transaction structure after this call.
 928 */
 929static int
 930__xfs_trans_commit(
 931	struct xfs_trans	*tp,
 932	bool			regrant)
 933{
 934	struct xfs_mount	*mp = tp->t_mountp;
 935	xfs_lsn_t		commit_lsn = -1;
 936	int			error = 0;
 937	int			sync = tp->t_flags & XFS_TRANS_SYNC;
 938
 939	/*
 940	 * If there is nothing to be logged by the transaction,
 941	 * then unlock all of the items associated with the
 942	 * transaction and free the transaction structure.
 943	 * Also make sure to return any reserved blocks to
 944	 * the free pool.
 945	 */
 946	if (!(tp->t_flags & XFS_TRANS_DIRTY))
 947		goto out_unreserve;
 948
 949	if (XFS_FORCED_SHUTDOWN(mp)) {
 950		error = -EIO;
 951		goto out_unreserve;
 952	}
 953
 954	ASSERT(tp->t_ticket != NULL);
 955
 956	/*
 957	 * If we need to update the superblock, then do it now.
 958	 */
 959	if (tp->t_flags & XFS_TRANS_SB_DIRTY)
 960		xfs_trans_apply_sb_deltas(tp);
 961	xfs_trans_apply_dquot_deltas(tp);
 962
 963	xfs_log_commit_cil(mp, tp, &commit_lsn, regrant);
 964
 965	current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
 966	xfs_trans_free(tp);
 967
 968	/*
 969	 * If the transaction needs to be synchronous, then force the
 970	 * log out now and wait for it.
 971	 */
 972	if (sync) {
 973		error = xfs_log_force_lsn(mp, commit_lsn, XFS_LOG_SYNC, NULL);
 974		XFS_STATS_INC(mp, xs_trans_sync);
 975	} else {
 976		XFS_STATS_INC(mp, xs_trans_async);
 977	}
 978
 979	return error;
 980
 981out_unreserve:
 982	xfs_trans_unreserve_and_mod_sb(tp);
 983
 984	/*
 985	 * It is indeed possible for the transaction to be not dirty but
 986	 * the dqinfo portion to be.  All that means is that we have some
 987	 * (non-persistent) quota reservations that need to be unreserved.
 988	 */
 989	xfs_trans_unreserve_and_mod_dquots(tp);
 990	if (tp->t_ticket) {
 991		commit_lsn = xfs_log_done(mp, tp->t_ticket, NULL, regrant);
 992		if (commit_lsn == -1 && !error)
 993			error = -EIO;
 994	}
 995	current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
 996	xfs_trans_free_items(tp, NULLCOMMITLSN, !!error);
 997	xfs_trans_free(tp);
 998
 999	XFS_STATS_INC(mp, xs_trans_empty);
1000	return error;
1001}
1002
1003int
1004xfs_trans_commit(
1005	struct xfs_trans	*tp)
1006{
1007	return __xfs_trans_commit(tp, false);
1008}
1009
1010/*
1011 * Unlock all of the transaction's items and free the transaction.
1012 * The transaction must not have modified any of its items, because
1013 * there is no way to restore them to their previous state.
1014 *
1015 * If the transaction has made a log reservation, make sure to release
1016 * it as well.
1017 */
1018void
1019xfs_trans_cancel(
1020	struct xfs_trans	*tp)
1021{
1022	struct xfs_mount	*mp = tp->t_mountp;
1023	bool			dirty = (tp->t_flags & XFS_TRANS_DIRTY);
1024
1025	/*
1026	 * See if the caller is relying on us to shut down the
1027	 * filesystem.  This happens in paths where we detect
1028	 * corruption and decide to give up.
1029	 */
1030	if (dirty && !XFS_FORCED_SHUTDOWN(mp)) {
1031		XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
1032		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1033	}
1034#ifdef DEBUG
1035	if (!dirty && !XFS_FORCED_SHUTDOWN(mp)) {
1036		struct xfs_log_item_desc *lidp;
1037
1038		list_for_each_entry(lidp, &tp->t_items, lid_trans)
1039			ASSERT(!(lidp->lid_item->li_type == XFS_LI_EFD));
1040	}
1041#endif
1042	xfs_trans_unreserve_and_mod_sb(tp);
1043	xfs_trans_unreserve_and_mod_dquots(tp);
1044
1045	if (tp->t_ticket)
1046		xfs_log_done(mp, tp->t_ticket, NULL, false);
1047
1048	/* mark this thread as no longer being in a transaction */
1049	current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
1050
1051	xfs_trans_free_items(tp, NULLCOMMITLSN, dirty);
1052	xfs_trans_free(tp);
1053}
1054
1055/*
1056 * Roll from one trans in the sequence of PERMANENT transactions to
1057 * the next: permanent transactions are only flushed out when
1058 * committed with xfs_trans_commit(), but we still want as soon
1059 * as possible to let chunks of it go to the log. So we commit the
1060 * chunk we've been working on and get a new transaction to continue.
1061 */
1062int
1063xfs_trans_roll(
1064	struct xfs_trans	**tpp)
 
 
1065{
1066	struct xfs_trans	*trans = *tpp;
1067	struct xfs_trans_res	tres;
1068	int			error;
1069
 
 
 
 
 
 
 
 
 
1070	/*
1071	 * Copy the critical parameters from one trans to the next.
1072	 */
1073	tres.tr_logres = trans->t_log_res;
1074	tres.tr_logcount = trans->t_log_count;
1075
1076	*tpp = xfs_trans_dup(trans);
1077
1078	/*
1079	 * Commit the current transaction.
1080	 * If this commit failed, then it'd just unlock those items that
1081	 * are not marked ihold. That also means that a filesystem shutdown
1082	 * is in progress. The caller takes the responsibility to cancel
1083	 * the duplicate transaction that gets returned.
1084	 */
1085	error = __xfs_trans_commit(trans, true);
1086	if (error)
1087		return error;
1088
 
 
 
1089	/*
1090	 * Reserve space in the log for the next transaction.
1091	 * This also pushes items in the "AIL", the list of logged items,
1092	 * out to disk if they are taking up space at the tail of the log
1093	 * that we want to use.  This requires that either nothing be locked
1094	 * across this call, or that anything that is locked be logged in
1095	 * the prior and the next transactions.
1096	 */
1097	tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1098	return xfs_trans_reserve(*tpp, &tres, 0, 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1099}
v4.6
   1/*
   2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
   3 * Copyright (C) 2010 Red Hat, Inc.
   4 * All Rights Reserved.
   5 *
   6 * This program is free software; you can redistribute it and/or
   7 * modify it under the terms of the GNU General Public License as
   8 * published by the Free Software Foundation.
   9 *
  10 * This program is distributed in the hope that it would be useful,
  11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13 * GNU General Public License for more details.
  14 *
  15 * You should have received a copy of the GNU General Public License
  16 * along with this program; if not, write the Free Software Foundation,
  17 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  18 */
  19#include "xfs.h"
  20#include "xfs_fs.h"
  21#include "xfs_shared.h"
  22#include "xfs_format.h"
  23#include "xfs_log_format.h"
  24#include "xfs_trans_resv.h"
  25#include "xfs_mount.h"
  26#include "xfs_inode.h"
  27#include "xfs_extent_busy.h"
  28#include "xfs_quota.h"
  29#include "xfs_trans.h"
  30#include "xfs_trans_priv.h"
  31#include "xfs_log.h"
  32#include "xfs_trace.h"
  33#include "xfs_error.h"
  34
  35kmem_zone_t	*xfs_trans_zone;
  36kmem_zone_t	*xfs_log_item_desc_zone;
  37
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  38/*
  39 * Initialize the precomputed transaction reservation values
  40 * in the mount structure.
  41 */
  42void
  43xfs_trans_init(
  44	struct xfs_mount	*mp)
  45{
  46	xfs_trans_resv_calc(mp, M_RES(mp));
  47}
  48
  49/*
  50 * This routine is called to allocate a transaction structure.
  51 * The type parameter indicates the type of the transaction.  These
  52 * are enumerated in xfs_trans.h.
  53 *
  54 * Dynamically allocate the transaction structure from the transaction
  55 * zone, initialize it, and return it to the caller.
  56 */
  57xfs_trans_t *
  58xfs_trans_alloc(
  59	xfs_mount_t	*mp,
  60	uint		type)
  61{
  62	xfs_trans_t     *tp;
  63
  64	sb_start_intwrite(mp->m_super);
  65	tp = _xfs_trans_alloc(mp, type, KM_SLEEP);
  66	tp->t_flags |= XFS_TRANS_FREEZE_PROT;
  67	return tp;
  68}
  69
  70xfs_trans_t *
  71_xfs_trans_alloc(
  72	xfs_mount_t	*mp,
  73	uint		type,
  74	xfs_km_flags_t	memflags)
  75{
  76	xfs_trans_t	*tp;
  77
  78	WARN_ON(mp->m_super->s_writers.frozen == SB_FREEZE_COMPLETE);
  79	atomic_inc(&mp->m_active_trans);
  80
  81	tp = kmem_zone_zalloc(xfs_trans_zone, memflags);
  82	tp->t_magic = XFS_TRANS_HEADER_MAGIC;
  83	tp->t_type = type;
  84	tp->t_mountp = mp;
  85	INIT_LIST_HEAD(&tp->t_items);
  86	INIT_LIST_HEAD(&tp->t_busy);
  87	return tp;
  88}
  89
  90/*
  91 * Free the transaction structure.  If there is more clean up
  92 * to do when the structure is freed, add it here.
  93 */
  94STATIC void
  95xfs_trans_free(
  96	struct xfs_trans	*tp)
  97{
  98	xfs_extent_busy_sort(&tp->t_busy);
  99	xfs_extent_busy_clear(tp->t_mountp, &tp->t_busy, false);
 100
 101	atomic_dec(&tp->t_mountp->m_active_trans);
 102	if (tp->t_flags & XFS_TRANS_FREEZE_PROT)
 103		sb_end_intwrite(tp->t_mountp->m_super);
 104	xfs_trans_free_dqinfo(tp);
 105	kmem_zone_free(xfs_trans_zone, tp);
 106}
 107
 108/*
 109 * This is called to create a new transaction which will share the
 110 * permanent log reservation of the given transaction.  The remaining
 111 * unused block and rt extent reservations are also inherited.  This
 112 * implies that the original transaction is no longer allowed to allocate
 113 * blocks.  Locks and log items, however, are no inherited.  They must
 114 * be added to the new transaction explicitly.
 115 */
 116STATIC xfs_trans_t *
 117xfs_trans_dup(
 118	xfs_trans_t	*tp)
 119{
 120	xfs_trans_t	*ntp;
 121
 122	ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
 123
 124	/*
 125	 * Initialize the new transaction structure.
 126	 */
 127	ntp->t_magic = XFS_TRANS_HEADER_MAGIC;
 128	ntp->t_type = tp->t_type;
 129	ntp->t_mountp = tp->t_mountp;
 130	INIT_LIST_HEAD(&ntp->t_items);
 131	INIT_LIST_HEAD(&ntp->t_busy);
 132
 133	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
 134	ASSERT(tp->t_ticket != NULL);
 135
 136	ntp->t_flags = XFS_TRANS_PERM_LOG_RES |
 137		       (tp->t_flags & XFS_TRANS_RESERVE) |
 138		       (tp->t_flags & XFS_TRANS_FREEZE_PROT);
 139	/* We gave our writer reference to the new transaction */
 140	tp->t_flags &= ~XFS_TRANS_FREEZE_PROT;
 141	ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket);
 
 
 142	ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
 143	tp->t_blk_res = tp->t_blk_res_used;
 
 144	ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
 145	tp->t_rtx_res = tp->t_rtx_res_used;
 146	ntp->t_pflags = tp->t_pflags;
 147
 148	xfs_trans_dup_dqinfo(tp, ntp);
 149
 150	atomic_inc(&tp->t_mountp->m_active_trans);
 151	return ntp;
 152}
 153
 154/*
 155 * This is called to reserve free disk blocks and log space for the
 156 * given transaction.  This must be done before allocating any resources
 157 * within the transaction.
 158 *
 159 * This will return ENOSPC if there are not enough blocks available.
 160 * It will sleep waiting for available log space.
 161 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
 162 * is used by long running transactions.  If any one of the reservations
 163 * fails then they will all be backed out.
 164 *
 165 * This does not do quota reservations. That typically is done by the
 166 * caller afterwards.
 167 */
 168int
 169xfs_trans_reserve(
 170	struct xfs_trans	*tp,
 171	struct xfs_trans_res	*resp,
 172	uint			blocks,
 173	uint			rtextents)
 174{
 175	int		error = 0;
 176	bool		rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
 177
 178	/* Mark this thread as being in a transaction */
 179	current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
 180
 181	/*
 182	 * Attempt to reserve the needed disk blocks by decrementing
 183	 * the number needed from the number available.  This will
 184	 * fail if the count would go below zero.
 185	 */
 186	if (blocks > 0) {
 187		error = xfs_mod_fdblocks(tp->t_mountp, -((int64_t)blocks), rsvd);
 188		if (error != 0) {
 189			current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
 190			return -ENOSPC;
 191		}
 192		tp->t_blk_res += blocks;
 193	}
 194
 195	/*
 196	 * Reserve the log space needed for this transaction.
 197	 */
 198	if (resp->tr_logres > 0) {
 199		bool	permanent = false;
 200
 201		ASSERT(tp->t_log_res == 0 ||
 202		       tp->t_log_res == resp->tr_logres);
 203		ASSERT(tp->t_log_count == 0 ||
 204		       tp->t_log_count == resp->tr_logcount);
 205
 206		if (resp->tr_logflags & XFS_TRANS_PERM_LOG_RES) {
 207			tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
 208			permanent = true;
 209		} else {
 210			ASSERT(tp->t_ticket == NULL);
 211			ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
 212		}
 213
 214		if (tp->t_ticket != NULL) {
 215			ASSERT(resp->tr_logflags & XFS_TRANS_PERM_LOG_RES);
 216			error = xfs_log_regrant(tp->t_mountp, tp->t_ticket);
 217		} else {
 218			error = xfs_log_reserve(tp->t_mountp,
 219						resp->tr_logres,
 220						resp->tr_logcount,
 221						&tp->t_ticket, XFS_TRANSACTION,
 222						permanent, tp->t_type);
 223		}
 224
 225		if (error)
 226			goto undo_blocks;
 227
 228		tp->t_log_res = resp->tr_logres;
 229		tp->t_log_count = resp->tr_logcount;
 230	}
 231
 232	/*
 233	 * Attempt to reserve the needed realtime extents by decrementing
 234	 * the number needed from the number available.  This will
 235	 * fail if the count would go below zero.
 236	 */
 237	if (rtextents > 0) {
 238		error = xfs_mod_frextents(tp->t_mountp, -((int64_t)rtextents));
 239		if (error) {
 240			error = -ENOSPC;
 241			goto undo_log;
 242		}
 243		tp->t_rtx_res += rtextents;
 244	}
 245
 246	return 0;
 247
 248	/*
 249	 * Error cases jump to one of these labels to undo any
 250	 * reservations which have already been performed.
 251	 */
 252undo_log:
 253	if (resp->tr_logres > 0) {
 254		xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, false);
 255		tp->t_ticket = NULL;
 256		tp->t_log_res = 0;
 257		tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
 258	}
 259
 260undo_blocks:
 261	if (blocks > 0) {
 262		xfs_mod_fdblocks(tp->t_mountp, -((int64_t)blocks), rsvd);
 263		tp->t_blk_res = 0;
 264	}
 265
 266	current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
 267
 268	return error;
 269}
 270
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 271/*
 272 * Record the indicated change to the given field for application
 273 * to the file system's superblock when the transaction commits.
 274 * For now, just store the change in the transaction structure.
 275 *
 276 * Mark the transaction structure to indicate that the superblock
 277 * needs to be updated before committing.
 278 *
 279 * Because we may not be keeping track of allocated/free inodes and
 280 * used filesystem blocks in the superblock, we do not mark the
 281 * superblock dirty in this transaction if we modify these fields.
 282 * We still need to update the transaction deltas so that they get
 283 * applied to the incore superblock, but we don't want them to
 284 * cause the superblock to get locked and logged if these are the
 285 * only fields in the superblock that the transaction modifies.
 286 */
 287void
 288xfs_trans_mod_sb(
 289	xfs_trans_t	*tp,
 290	uint		field,
 291	int64_t		delta)
 292{
 293	uint32_t	flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
 294	xfs_mount_t	*mp = tp->t_mountp;
 295
 296	switch (field) {
 297	case XFS_TRANS_SB_ICOUNT:
 298		tp->t_icount_delta += delta;
 299		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
 300			flags &= ~XFS_TRANS_SB_DIRTY;
 301		break;
 302	case XFS_TRANS_SB_IFREE:
 303		tp->t_ifree_delta += delta;
 304		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
 305			flags &= ~XFS_TRANS_SB_DIRTY;
 306		break;
 307	case XFS_TRANS_SB_FDBLOCKS:
 308		/*
 309		 * Track the number of blocks allocated in the
 310		 * transaction.  Make sure it does not exceed the
 311		 * number reserved.
 312		 */
 313		if (delta < 0) {
 314			tp->t_blk_res_used += (uint)-delta;
 315			ASSERT(tp->t_blk_res_used <= tp->t_blk_res);
 
 316		}
 317		tp->t_fdblocks_delta += delta;
 318		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
 319			flags &= ~XFS_TRANS_SB_DIRTY;
 320		break;
 321	case XFS_TRANS_SB_RES_FDBLOCKS:
 322		/*
 323		 * The allocation has already been applied to the
 324		 * in-core superblock's counter.  This should only
 325		 * be applied to the on-disk superblock.
 326		 */
 327		ASSERT(delta < 0);
 328		tp->t_res_fdblocks_delta += delta;
 329		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
 330			flags &= ~XFS_TRANS_SB_DIRTY;
 331		break;
 332	case XFS_TRANS_SB_FREXTENTS:
 333		/*
 334		 * Track the number of blocks allocated in the
 335		 * transaction.  Make sure it does not exceed the
 336		 * number reserved.
 337		 */
 338		if (delta < 0) {
 339			tp->t_rtx_res_used += (uint)-delta;
 340			ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
 341		}
 342		tp->t_frextents_delta += delta;
 343		break;
 344	case XFS_TRANS_SB_RES_FREXTENTS:
 345		/*
 346		 * The allocation has already been applied to the
 347		 * in-core superblock's counter.  This should only
 348		 * be applied to the on-disk superblock.
 349		 */
 350		ASSERT(delta < 0);
 351		tp->t_res_frextents_delta += delta;
 352		break;
 353	case XFS_TRANS_SB_DBLOCKS:
 354		ASSERT(delta > 0);
 355		tp->t_dblocks_delta += delta;
 356		break;
 357	case XFS_TRANS_SB_AGCOUNT:
 358		ASSERT(delta > 0);
 359		tp->t_agcount_delta += delta;
 360		break;
 361	case XFS_TRANS_SB_IMAXPCT:
 362		tp->t_imaxpct_delta += delta;
 363		break;
 364	case XFS_TRANS_SB_REXTSIZE:
 365		tp->t_rextsize_delta += delta;
 366		break;
 367	case XFS_TRANS_SB_RBMBLOCKS:
 368		tp->t_rbmblocks_delta += delta;
 369		break;
 370	case XFS_TRANS_SB_RBLOCKS:
 371		tp->t_rblocks_delta += delta;
 372		break;
 373	case XFS_TRANS_SB_REXTENTS:
 374		tp->t_rextents_delta += delta;
 375		break;
 376	case XFS_TRANS_SB_REXTSLOG:
 377		tp->t_rextslog_delta += delta;
 378		break;
 379	default:
 380		ASSERT(0);
 381		return;
 382	}
 383
 384	tp->t_flags |= flags;
 385}
 386
 387/*
 388 * xfs_trans_apply_sb_deltas() is called from the commit code
 389 * to bring the superblock buffer into the current transaction
 390 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
 391 *
 392 * For now we just look at each field allowed to change and change
 393 * it if necessary.
 394 */
 395STATIC void
 396xfs_trans_apply_sb_deltas(
 397	xfs_trans_t	*tp)
 398{
 399	xfs_dsb_t	*sbp;
 400	xfs_buf_t	*bp;
 401	int		whole = 0;
 402
 403	bp = xfs_trans_getsb(tp, tp->t_mountp, 0);
 404	sbp = XFS_BUF_TO_SBP(bp);
 405
 406	/*
 407	 * Check that superblock mods match the mods made to AGF counters.
 408	 */
 409	ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
 410	       (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
 411		tp->t_ag_btree_delta));
 412
 413	/*
 414	 * Only update the superblock counters if we are logging them
 415	 */
 416	if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) {
 417		if (tp->t_icount_delta)
 418			be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
 419		if (tp->t_ifree_delta)
 420			be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
 421		if (tp->t_fdblocks_delta)
 422			be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
 423		if (tp->t_res_fdblocks_delta)
 424			be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
 425	}
 426
 427	if (tp->t_frextents_delta)
 428		be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta);
 429	if (tp->t_res_frextents_delta)
 430		be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta);
 431
 432	if (tp->t_dblocks_delta) {
 433		be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
 434		whole = 1;
 435	}
 436	if (tp->t_agcount_delta) {
 437		be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
 438		whole = 1;
 439	}
 440	if (tp->t_imaxpct_delta) {
 441		sbp->sb_imax_pct += tp->t_imaxpct_delta;
 442		whole = 1;
 443	}
 444	if (tp->t_rextsize_delta) {
 445		be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
 446		whole = 1;
 447	}
 448	if (tp->t_rbmblocks_delta) {
 449		be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
 450		whole = 1;
 451	}
 452	if (tp->t_rblocks_delta) {
 453		be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
 454		whole = 1;
 455	}
 456	if (tp->t_rextents_delta) {
 457		be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
 458		whole = 1;
 459	}
 460	if (tp->t_rextslog_delta) {
 461		sbp->sb_rextslog += tp->t_rextslog_delta;
 462		whole = 1;
 463	}
 464
 465	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
 466	if (whole)
 467		/*
 468		 * Log the whole thing, the fields are noncontiguous.
 469		 */
 470		xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1);
 471	else
 472		/*
 473		 * Since all the modifiable fields are contiguous, we
 474		 * can get away with this.
 475		 */
 476		xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount),
 477				  offsetof(xfs_dsb_t, sb_frextents) +
 478				  sizeof(sbp->sb_frextents) - 1);
 479}
 480
 481STATIC int
 482xfs_sb_mod8(
 483	uint8_t			*field,
 484	int8_t			delta)
 485{
 486	int8_t			counter = *field;
 487
 488	counter += delta;
 489	if (counter < 0) {
 490		ASSERT(0);
 491		return -EINVAL;
 492	}
 493	*field = counter;
 494	return 0;
 495}
 496
 497STATIC int
 498xfs_sb_mod32(
 499	uint32_t		*field,
 500	int32_t			delta)
 501{
 502	int32_t			counter = *field;
 503
 504	counter += delta;
 505	if (counter < 0) {
 506		ASSERT(0);
 507		return -EINVAL;
 508	}
 509	*field = counter;
 510	return 0;
 511}
 512
 513STATIC int
 514xfs_sb_mod64(
 515	uint64_t		*field,
 516	int64_t			delta)
 517{
 518	int64_t			counter = *field;
 519
 520	counter += delta;
 521	if (counter < 0) {
 522		ASSERT(0);
 523		return -EINVAL;
 524	}
 525	*field = counter;
 526	return 0;
 527}
 528
 529/*
 530 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations
 531 * and apply superblock counter changes to the in-core superblock.  The
 532 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
 533 * applied to the in-core superblock.  The idea is that that has already been
 534 * done.
 535 *
 536 * If we are not logging superblock counters, then the inode allocated/free and
 537 * used block counts are not updated in the on disk superblock. In this case,
 538 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
 539 * still need to update the incore superblock with the changes.
 540 */
 541void
 542xfs_trans_unreserve_and_mod_sb(
 543	struct xfs_trans	*tp)
 544{
 545	struct xfs_mount	*mp = tp->t_mountp;
 546	bool			rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
 547	int64_t			blkdelta = 0;
 548	int64_t			rtxdelta = 0;
 549	int64_t			idelta = 0;
 550	int64_t			ifreedelta = 0;
 551	int			error;
 552
 553	/* calculate deltas */
 554	if (tp->t_blk_res > 0)
 555		blkdelta = tp->t_blk_res;
 556	if ((tp->t_fdblocks_delta != 0) &&
 557	    (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
 558	     (tp->t_flags & XFS_TRANS_SB_DIRTY)))
 559	        blkdelta += tp->t_fdblocks_delta;
 560
 561	if (tp->t_rtx_res > 0)
 562		rtxdelta = tp->t_rtx_res;
 563	if ((tp->t_frextents_delta != 0) &&
 564	    (tp->t_flags & XFS_TRANS_SB_DIRTY))
 565		rtxdelta += tp->t_frextents_delta;
 566
 567	if (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
 568	     (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
 569		idelta = tp->t_icount_delta;
 570		ifreedelta = tp->t_ifree_delta;
 571	}
 572
 573	/* apply the per-cpu counters */
 574	if (blkdelta) {
 575		error = xfs_mod_fdblocks(mp, blkdelta, rsvd);
 576		if (error)
 577			goto out;
 578	}
 579
 580	if (idelta) {
 581		error = xfs_mod_icount(mp, idelta);
 582		if (error)
 583			goto out_undo_fdblocks;
 584	}
 585
 586	if (ifreedelta) {
 587		error = xfs_mod_ifree(mp, ifreedelta);
 588		if (error)
 589			goto out_undo_icount;
 590	}
 591
 592	if (rtxdelta == 0 && !(tp->t_flags & XFS_TRANS_SB_DIRTY))
 593		return;
 594
 595	/* apply remaining deltas */
 596	spin_lock(&mp->m_sb_lock);
 597	if (rtxdelta) {
 598		error = xfs_sb_mod64(&mp->m_sb.sb_frextents, rtxdelta);
 599		if (error)
 600			goto out_undo_ifree;
 601	}
 602
 603	if (tp->t_dblocks_delta != 0) {
 604		error = xfs_sb_mod64(&mp->m_sb.sb_dblocks, tp->t_dblocks_delta);
 605		if (error)
 606			goto out_undo_frextents;
 607	}
 608	if (tp->t_agcount_delta != 0) {
 609		error = xfs_sb_mod32(&mp->m_sb.sb_agcount, tp->t_agcount_delta);
 610		if (error)
 611			goto out_undo_dblocks;
 612	}
 613	if (tp->t_imaxpct_delta != 0) {
 614		error = xfs_sb_mod8(&mp->m_sb.sb_imax_pct, tp->t_imaxpct_delta);
 615		if (error)
 616			goto out_undo_agcount;
 617	}
 618	if (tp->t_rextsize_delta != 0) {
 619		error = xfs_sb_mod32(&mp->m_sb.sb_rextsize,
 620				     tp->t_rextsize_delta);
 621		if (error)
 622			goto out_undo_imaxpct;
 623	}
 624	if (tp->t_rbmblocks_delta != 0) {
 625		error = xfs_sb_mod32(&mp->m_sb.sb_rbmblocks,
 626				     tp->t_rbmblocks_delta);
 627		if (error)
 628			goto out_undo_rextsize;
 629	}
 630	if (tp->t_rblocks_delta != 0) {
 631		error = xfs_sb_mod64(&mp->m_sb.sb_rblocks, tp->t_rblocks_delta);
 632		if (error)
 633			goto out_undo_rbmblocks;
 634	}
 635	if (tp->t_rextents_delta != 0) {
 636		error = xfs_sb_mod64(&mp->m_sb.sb_rextents,
 637				     tp->t_rextents_delta);
 638		if (error)
 639			goto out_undo_rblocks;
 640	}
 641	if (tp->t_rextslog_delta != 0) {
 642		error = xfs_sb_mod8(&mp->m_sb.sb_rextslog,
 643				     tp->t_rextslog_delta);
 644		if (error)
 645			goto out_undo_rextents;
 646	}
 647	spin_unlock(&mp->m_sb_lock);
 648	return;
 649
 650out_undo_rextents:
 651	if (tp->t_rextents_delta)
 652		xfs_sb_mod64(&mp->m_sb.sb_rextents, -tp->t_rextents_delta);
 653out_undo_rblocks:
 654	if (tp->t_rblocks_delta)
 655		xfs_sb_mod64(&mp->m_sb.sb_rblocks, -tp->t_rblocks_delta);
 656out_undo_rbmblocks:
 657	if (tp->t_rbmblocks_delta)
 658		xfs_sb_mod32(&mp->m_sb.sb_rbmblocks, -tp->t_rbmblocks_delta);
 659out_undo_rextsize:
 660	if (tp->t_rextsize_delta)
 661		xfs_sb_mod32(&mp->m_sb.sb_rextsize, -tp->t_rextsize_delta);
 662out_undo_imaxpct:
 663	if (tp->t_rextsize_delta)
 664		xfs_sb_mod8(&mp->m_sb.sb_imax_pct, -tp->t_imaxpct_delta);
 665out_undo_agcount:
 666	if (tp->t_agcount_delta)
 667		xfs_sb_mod32(&mp->m_sb.sb_agcount, -tp->t_agcount_delta);
 668out_undo_dblocks:
 669	if (tp->t_dblocks_delta)
 670		xfs_sb_mod64(&mp->m_sb.sb_dblocks, -tp->t_dblocks_delta);
 671out_undo_frextents:
 672	if (rtxdelta)
 673		xfs_sb_mod64(&mp->m_sb.sb_frextents, -rtxdelta);
 674out_undo_ifree:
 675	spin_unlock(&mp->m_sb_lock);
 676	if (ifreedelta)
 677		xfs_mod_ifree(mp, -ifreedelta);
 678out_undo_icount:
 679	if (idelta)
 680		xfs_mod_icount(mp, -idelta);
 681out_undo_fdblocks:
 682	if (blkdelta)
 683		xfs_mod_fdblocks(mp, -blkdelta, rsvd);
 684out:
 685	ASSERT(error == 0);
 686	return;
 687}
 688
 689/*
 690 * Add the given log item to the transaction's list of log items.
 691 *
 692 * The log item will now point to its new descriptor with its li_desc field.
 693 */
 694void
 695xfs_trans_add_item(
 696	struct xfs_trans	*tp,
 697	struct xfs_log_item	*lip)
 698{
 699	struct xfs_log_item_desc *lidp;
 700
 701	ASSERT(lip->li_mountp == tp->t_mountp);
 702	ASSERT(lip->li_ailp == tp->t_mountp->m_ail);
 703
 704	lidp = kmem_zone_zalloc(xfs_log_item_desc_zone, KM_SLEEP | KM_NOFS);
 705
 706	lidp->lid_item = lip;
 707	lidp->lid_flags = 0;
 708	list_add_tail(&lidp->lid_trans, &tp->t_items);
 709
 710	lip->li_desc = lidp;
 711}
 712
 713STATIC void
 714xfs_trans_free_item_desc(
 715	struct xfs_log_item_desc *lidp)
 716{
 717	list_del_init(&lidp->lid_trans);
 718	kmem_zone_free(xfs_log_item_desc_zone, lidp);
 719}
 720
 721/*
 722 * Unlink and free the given descriptor.
 723 */
 724void
 725xfs_trans_del_item(
 726	struct xfs_log_item	*lip)
 727{
 728	xfs_trans_free_item_desc(lip->li_desc);
 729	lip->li_desc = NULL;
 730}
 731
 732/*
 733 * Unlock all of the items of a transaction and free all the descriptors
 734 * of that transaction.
 735 */
 736void
 737xfs_trans_free_items(
 738	struct xfs_trans	*tp,
 739	xfs_lsn_t		commit_lsn,
 740	bool			abort)
 741{
 742	struct xfs_log_item_desc *lidp, *next;
 743
 744	list_for_each_entry_safe(lidp, next, &tp->t_items, lid_trans) {
 745		struct xfs_log_item	*lip = lidp->lid_item;
 746
 747		lip->li_desc = NULL;
 748
 749		if (commit_lsn != NULLCOMMITLSN)
 750			lip->li_ops->iop_committing(lip, commit_lsn);
 751		if (abort)
 752			lip->li_flags |= XFS_LI_ABORTED;
 753		lip->li_ops->iop_unlock(lip);
 754
 755		xfs_trans_free_item_desc(lidp);
 756	}
 757}
 758
 759static inline void
 760xfs_log_item_batch_insert(
 761	struct xfs_ail		*ailp,
 762	struct xfs_ail_cursor	*cur,
 763	struct xfs_log_item	**log_items,
 764	int			nr_items,
 765	xfs_lsn_t		commit_lsn)
 766{
 767	int	i;
 768
 769	spin_lock(&ailp->xa_lock);
 770	/* xfs_trans_ail_update_bulk drops ailp->xa_lock */
 771	xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn);
 772
 773	for (i = 0; i < nr_items; i++) {
 774		struct xfs_log_item *lip = log_items[i];
 775
 776		lip->li_ops->iop_unpin(lip, 0);
 777	}
 778}
 779
 780/*
 781 * Bulk operation version of xfs_trans_committed that takes a log vector of
 782 * items to insert into the AIL. This uses bulk AIL insertion techniques to
 783 * minimise lock traffic.
 784 *
 785 * If we are called with the aborted flag set, it is because a log write during
 786 * a CIL checkpoint commit has failed. In this case, all the items in the
 787 * checkpoint have already gone through iop_commited and iop_unlock, which
 788 * means that checkpoint commit abort handling is treated exactly the same
 789 * as an iclog write error even though we haven't started any IO yet. Hence in
 790 * this case all we need to do is iop_committed processing, followed by an
 791 * iop_unpin(aborted) call.
 792 *
 793 * The AIL cursor is used to optimise the insert process. If commit_lsn is not
 794 * at the end of the AIL, the insert cursor avoids the need to walk
 795 * the AIL to find the insertion point on every xfs_log_item_batch_insert()
 796 * call. This saves a lot of needless list walking and is a net win, even
 797 * though it slightly increases that amount of AIL lock traffic to set it up
 798 * and tear it down.
 799 */
 800void
 801xfs_trans_committed_bulk(
 802	struct xfs_ail		*ailp,
 803	struct xfs_log_vec	*log_vector,
 804	xfs_lsn_t		commit_lsn,
 805	int			aborted)
 806{
 807#define LOG_ITEM_BATCH_SIZE	32
 808	struct xfs_log_item	*log_items[LOG_ITEM_BATCH_SIZE];
 809	struct xfs_log_vec	*lv;
 810	struct xfs_ail_cursor	cur;
 811	int			i = 0;
 812
 813	spin_lock(&ailp->xa_lock);
 814	xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn);
 815	spin_unlock(&ailp->xa_lock);
 816
 817	/* unpin all the log items */
 818	for (lv = log_vector; lv; lv = lv->lv_next ) {
 819		struct xfs_log_item	*lip = lv->lv_item;
 820		xfs_lsn_t		item_lsn;
 821
 822		if (aborted)
 823			lip->li_flags |= XFS_LI_ABORTED;
 824		item_lsn = lip->li_ops->iop_committed(lip, commit_lsn);
 825
 826		/* item_lsn of -1 means the item needs no further processing */
 827		if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
 828			continue;
 829
 830		/*
 831		 * if we are aborting the operation, no point in inserting the
 832		 * object into the AIL as we are in a shutdown situation.
 833		 */
 834		if (aborted) {
 835			ASSERT(XFS_FORCED_SHUTDOWN(ailp->xa_mount));
 836			lip->li_ops->iop_unpin(lip, 1);
 837			continue;
 838		}
 839
 840		if (item_lsn != commit_lsn) {
 841
 842			/*
 843			 * Not a bulk update option due to unusual item_lsn.
 844			 * Push into AIL immediately, rechecking the lsn once
 845			 * we have the ail lock. Then unpin the item. This does
 846			 * not affect the AIL cursor the bulk insert path is
 847			 * using.
 848			 */
 849			spin_lock(&ailp->xa_lock);
 850			if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
 851				xfs_trans_ail_update(ailp, lip, item_lsn);
 852			else
 853				spin_unlock(&ailp->xa_lock);
 854			lip->li_ops->iop_unpin(lip, 0);
 855			continue;
 856		}
 857
 858		/* Item is a candidate for bulk AIL insert.  */
 859		log_items[i++] = lv->lv_item;
 860		if (i >= LOG_ITEM_BATCH_SIZE) {
 861			xfs_log_item_batch_insert(ailp, &cur, log_items,
 862					LOG_ITEM_BATCH_SIZE, commit_lsn);
 863			i = 0;
 864		}
 865	}
 866
 867	/* make sure we insert the remainder! */
 868	if (i)
 869		xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn);
 870
 871	spin_lock(&ailp->xa_lock);
 872	xfs_trans_ail_cursor_done(&cur);
 873	spin_unlock(&ailp->xa_lock);
 874}
 875
 876/*
 877 * Commit the given transaction to the log.
 878 *
 879 * XFS disk error handling mechanism is not based on a typical
 880 * transaction abort mechanism. Logically after the filesystem
 881 * gets marked 'SHUTDOWN', we can't let any new transactions
 882 * be durable - ie. committed to disk - because some metadata might
 883 * be inconsistent. In such cases, this returns an error, and the
 884 * caller may assume that all locked objects joined to the transaction
 885 * have already been unlocked as if the commit had succeeded.
 886 * Do not reference the transaction structure after this call.
 887 */
 888static int
 889__xfs_trans_commit(
 890	struct xfs_trans	*tp,
 891	bool			regrant)
 892{
 893	struct xfs_mount	*mp = tp->t_mountp;
 894	xfs_lsn_t		commit_lsn = -1;
 895	int			error = 0;
 896	int			sync = tp->t_flags & XFS_TRANS_SYNC;
 897
 898	/*
 899	 * If there is nothing to be logged by the transaction,
 900	 * then unlock all of the items associated with the
 901	 * transaction and free the transaction structure.
 902	 * Also make sure to return any reserved blocks to
 903	 * the free pool.
 904	 */
 905	if (!(tp->t_flags & XFS_TRANS_DIRTY))
 906		goto out_unreserve;
 907
 908	if (XFS_FORCED_SHUTDOWN(mp)) {
 909		error = -EIO;
 910		goto out_unreserve;
 911	}
 912
 913	ASSERT(tp->t_ticket != NULL);
 914
 915	/*
 916	 * If we need to update the superblock, then do it now.
 917	 */
 918	if (tp->t_flags & XFS_TRANS_SB_DIRTY)
 919		xfs_trans_apply_sb_deltas(tp);
 920	xfs_trans_apply_dquot_deltas(tp);
 921
 922	xfs_log_commit_cil(mp, tp, &commit_lsn, regrant);
 923
 924	current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
 925	xfs_trans_free(tp);
 926
 927	/*
 928	 * If the transaction needs to be synchronous, then force the
 929	 * log out now and wait for it.
 930	 */
 931	if (sync) {
 932		error = _xfs_log_force_lsn(mp, commit_lsn, XFS_LOG_SYNC, NULL);
 933		XFS_STATS_INC(mp, xs_trans_sync);
 934	} else {
 935		XFS_STATS_INC(mp, xs_trans_async);
 936	}
 937
 938	return error;
 939
 940out_unreserve:
 941	xfs_trans_unreserve_and_mod_sb(tp);
 942
 943	/*
 944	 * It is indeed possible for the transaction to be not dirty but
 945	 * the dqinfo portion to be.  All that means is that we have some
 946	 * (non-persistent) quota reservations that need to be unreserved.
 947	 */
 948	xfs_trans_unreserve_and_mod_dquots(tp);
 949	if (tp->t_ticket) {
 950		commit_lsn = xfs_log_done(mp, tp->t_ticket, NULL, regrant);
 951		if (commit_lsn == -1 && !error)
 952			error = -EIO;
 953	}
 954	current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
 955	xfs_trans_free_items(tp, NULLCOMMITLSN, !!error);
 956	xfs_trans_free(tp);
 957
 958	XFS_STATS_INC(mp, xs_trans_empty);
 959	return error;
 960}
 961
 962int
 963xfs_trans_commit(
 964	struct xfs_trans	*tp)
 965{
 966	return __xfs_trans_commit(tp, false);
 967}
 968
 969/*
 970 * Unlock all of the transaction's items and free the transaction.
 971 * The transaction must not have modified any of its items, because
 972 * there is no way to restore them to their previous state.
 973 *
 974 * If the transaction has made a log reservation, make sure to release
 975 * it as well.
 976 */
 977void
 978xfs_trans_cancel(
 979	struct xfs_trans	*tp)
 980{
 981	struct xfs_mount	*mp = tp->t_mountp;
 982	bool			dirty = (tp->t_flags & XFS_TRANS_DIRTY);
 983
 984	/*
 985	 * See if the caller is relying on us to shut down the
 986	 * filesystem.  This happens in paths where we detect
 987	 * corruption and decide to give up.
 988	 */
 989	if (dirty && !XFS_FORCED_SHUTDOWN(mp)) {
 990		XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
 991		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
 992	}
 993#ifdef DEBUG
 994	if (!dirty && !XFS_FORCED_SHUTDOWN(mp)) {
 995		struct xfs_log_item_desc *lidp;
 996
 997		list_for_each_entry(lidp, &tp->t_items, lid_trans)
 998			ASSERT(!(lidp->lid_item->li_type == XFS_LI_EFD));
 999	}
1000#endif
1001	xfs_trans_unreserve_and_mod_sb(tp);
1002	xfs_trans_unreserve_and_mod_dquots(tp);
1003
1004	if (tp->t_ticket)
1005		xfs_log_done(mp, tp->t_ticket, NULL, false);
1006
1007	/* mark this thread as no longer being in a transaction */
1008	current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1009
1010	xfs_trans_free_items(tp, NULLCOMMITLSN, dirty);
1011	xfs_trans_free(tp);
1012}
1013
1014/*
1015 * Roll from one trans in the sequence of PERMANENT transactions to
1016 * the next: permanent transactions are only flushed out when
1017 * committed with xfs_trans_commit(), but we still want as soon
1018 * as possible to let chunks of it go to the log. So we commit the
1019 * chunk we've been working on and get a new transaction to continue.
1020 */
1021int
1022__xfs_trans_roll(
1023	struct xfs_trans	**tpp,
1024	struct xfs_inode	*dp,
1025	int			*committed)
1026{
1027	struct xfs_trans	*trans;
1028	struct xfs_trans_res	tres;
1029	int			error;
1030
1031	*committed = 0;
1032
1033	/*
1034	 * Ensure that the inode is always logged.
1035	 */
1036	trans = *tpp;
1037	if (dp)
1038		xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
1039
1040	/*
1041	 * Copy the critical parameters from one trans to the next.
1042	 */
1043	tres.tr_logres = trans->t_log_res;
1044	tres.tr_logcount = trans->t_log_count;
 
1045	*tpp = xfs_trans_dup(trans);
1046
1047	/*
1048	 * Commit the current transaction.
1049	 * If this commit failed, then it'd just unlock those items that
1050	 * are not marked ihold. That also means that a filesystem shutdown
1051	 * is in progress. The caller takes the responsibility to cancel
1052	 * the duplicate transaction that gets returned.
1053	 */
1054	error = __xfs_trans_commit(trans, true);
1055	if (error)
1056		return error;
1057
1058	*committed = 1;
1059	trans = *tpp;
1060
1061	/*
1062	 * Reserve space in the log for th next transaction.
1063	 * This also pushes items in the "AIL", the list of logged items,
1064	 * out to disk if they are taking up space at the tail of the log
1065	 * that we want to use.  This requires that either nothing be locked
1066	 * across this call, or that anything that is locked be logged in
1067	 * the prior and the next transactions.
1068	 */
1069	tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1070	error = xfs_trans_reserve(trans, &tres, 0, 0);
1071	/*
1072	 *  Ensure that the inode is in the new transaction and locked.
1073	 */
1074	if (error)
1075		return error;
1076
1077	if (dp)
1078		xfs_trans_ijoin(trans, dp, 0);
1079	return 0;
1080}
1081
1082int
1083xfs_trans_roll(
1084	struct xfs_trans	**tpp,
1085	struct xfs_inode	*dp)
1086{
1087	int			committed;
1088	return __xfs_trans_roll(tpp, dp, &committed);
1089}