xfs_trans.c 38.5 KB
Newer Older
L
Linus Torvalds 已提交
1
/*
2 3
 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
 * All Rights Reserved.
L
Linus Torvalds 已提交
4
 *
5 6
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License as
L
Linus Torvalds 已提交
7 8
 * published by the Free Software Foundation.
 *
9 10 11 12
 * This program is distributed in the hope that it would be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
L
Linus Torvalds 已提交
13
 *
14 15 16
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write the Free Software Foundation,
 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
L
Linus Torvalds 已提交
17 18
 */
#include "xfs.h"
19
#include "xfs_fs.h"
L
Linus Torvalds 已提交
20
#include "xfs_types.h"
21
#include "xfs_bit.h"
L
Linus Torvalds 已提交
22
#include "xfs_log.h"
23
#include "xfs_inum.h"
L
Linus Torvalds 已提交
24 25 26 27 28 29 30
#include "xfs_trans.h"
#include "xfs_sb.h"
#include "xfs_ag.h"
#include "xfs_dir2.h"
#include "xfs_dmapi.h"
#include "xfs_mount.h"
#include "xfs_error.h"
31
#include "xfs_da_btree.h"
L
Linus Torvalds 已提交
32
#include "xfs_bmap_btree.h"
33
#include "xfs_alloc_btree.h"
L
Linus Torvalds 已提交
34 35
#include "xfs_ialloc_btree.h"
#include "xfs_dir2_sf.h"
36
#include "xfs_attr_sf.h"
L
Linus Torvalds 已提交
37 38
#include "xfs_dinode.h"
#include "xfs_inode.h"
39 40 41
#include "xfs_btree.h"
#include "xfs_ialloc.h"
#include "xfs_alloc.h"
L
Linus Torvalds 已提交
42 43
#include "xfs_bmap.h"
#include "xfs_quota.h"
44
#include "xfs_trans_priv.h"
L
Linus Torvalds 已提交
45
#include "xfs_trans_space.h"
46
#include "xfs_inode_item.h"
L
Linus Torvalds 已提交
47 48 49 50 51 52 53 54 55 56


STATIC void	xfs_trans_apply_sb_deltas(xfs_trans_t *);
STATIC uint	xfs_trans_count_vecs(xfs_trans_t *);
STATIC void	xfs_trans_fill_vecs(xfs_trans_t *, xfs_log_iovec_t *);
STATIC void	xfs_trans_uncommit(xfs_trans_t *, uint);
STATIC void	xfs_trans_committed(xfs_trans_t *, int);
STATIC void	xfs_trans_chunk_committed(xfs_log_item_chunk_t *, xfs_lsn_t, int);
STATIC void	xfs_trans_free(xfs_trans_t *);

57
kmem_zone_t	*xfs_trans_zone;
L
Linus Torvalds 已提交
58 59


60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85
/*
 * Reservation functions here avoid a huge stack in xfs_trans_init
 * due to register overflow from temporaries in the calculations.
 */

STATIC uint
xfs_calc_write_reservation(xfs_mount_t *mp)
{
	return XFS_CALC_WRITE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
}

STATIC uint
xfs_calc_itruncate_reservation(xfs_mount_t *mp)
{
	return XFS_CALC_ITRUNCATE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
}

STATIC uint
xfs_calc_rename_reservation(xfs_mount_t *mp)
{
	return XFS_CALC_RENAME_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
}

STATIC uint
xfs_calc_link_reservation(xfs_mount_t *mp)
{
86
	return XFS_CALC_LINK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190
}

STATIC uint
xfs_calc_remove_reservation(xfs_mount_t *mp)
{
	return XFS_CALC_REMOVE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
}

STATIC uint
xfs_calc_symlink_reservation(xfs_mount_t *mp)
{
	return XFS_CALC_SYMLINK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
}

STATIC uint
xfs_calc_create_reservation(xfs_mount_t *mp)
{
	return XFS_CALC_CREATE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
}

STATIC uint
xfs_calc_mkdir_reservation(xfs_mount_t *mp)
{
	return XFS_CALC_MKDIR_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
}

STATIC uint
xfs_calc_ifree_reservation(xfs_mount_t *mp)
{
	return XFS_CALC_IFREE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
}

STATIC uint
xfs_calc_ichange_reservation(xfs_mount_t *mp)
{
	return XFS_CALC_ICHANGE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
}

STATIC uint
xfs_calc_growdata_reservation(xfs_mount_t *mp)
{
	return XFS_CALC_GROWDATA_LOG_RES(mp);
}

STATIC uint
xfs_calc_growrtalloc_reservation(xfs_mount_t *mp)
{
	return XFS_CALC_GROWRTALLOC_LOG_RES(mp);
}

STATIC uint
xfs_calc_growrtzero_reservation(xfs_mount_t *mp)
{
	return XFS_CALC_GROWRTZERO_LOG_RES(mp);
}

STATIC uint
xfs_calc_growrtfree_reservation(xfs_mount_t *mp)
{
	return XFS_CALC_GROWRTFREE_LOG_RES(mp);
}

STATIC uint
xfs_calc_swrite_reservation(xfs_mount_t *mp)
{
	return XFS_CALC_SWRITE_LOG_RES(mp);
}

STATIC uint
xfs_calc_writeid_reservation(xfs_mount_t *mp)
{
	return XFS_CALC_WRITEID_LOG_RES(mp);
}

STATIC uint
xfs_calc_addafork_reservation(xfs_mount_t *mp)
{
	return XFS_CALC_ADDAFORK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
}

STATIC uint
xfs_calc_attrinval_reservation(xfs_mount_t *mp)
{
	return XFS_CALC_ATTRINVAL_LOG_RES(mp);
}

STATIC uint
xfs_calc_attrset_reservation(xfs_mount_t *mp)
{
	return XFS_CALC_ATTRSET_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
}

STATIC uint
xfs_calc_attrrm_reservation(xfs_mount_t *mp)
{
	return XFS_CALC_ATTRRM_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
}

STATIC uint
xfs_calc_clear_agi_bucket_reservation(xfs_mount_t *mp)
{
	return XFS_CALC_CLEAR_AGI_BUCKET_LOG_RES(mp);
}

L
Linus Torvalds 已提交
191 192 193 194 195 196 197 198 199 200 201
/*
 * Initialize the precomputed transaction reservation values
 * in the mount structure.
 */
void
xfs_trans_init(
	xfs_mount_t	*mp)
{
	xfs_trans_reservations_t	*resp;

	resp = &(mp->m_reservations);
202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222
	resp->tr_write = xfs_calc_write_reservation(mp);
	resp->tr_itruncate = xfs_calc_itruncate_reservation(mp);
	resp->tr_rename = xfs_calc_rename_reservation(mp);
	resp->tr_link = xfs_calc_link_reservation(mp);
	resp->tr_remove = xfs_calc_remove_reservation(mp);
	resp->tr_symlink = xfs_calc_symlink_reservation(mp);
	resp->tr_create = xfs_calc_create_reservation(mp);
	resp->tr_mkdir = xfs_calc_mkdir_reservation(mp);
	resp->tr_ifree = xfs_calc_ifree_reservation(mp);
	resp->tr_ichange = xfs_calc_ichange_reservation(mp);
	resp->tr_growdata = xfs_calc_growdata_reservation(mp);
	resp->tr_swrite = xfs_calc_swrite_reservation(mp);
	resp->tr_writeid = xfs_calc_writeid_reservation(mp);
	resp->tr_addafork = xfs_calc_addafork_reservation(mp);
	resp->tr_attrinval = xfs_calc_attrinval_reservation(mp);
	resp->tr_attrset = xfs_calc_attrset_reservation(mp);
	resp->tr_attrrm = xfs_calc_attrrm_reservation(mp);
	resp->tr_clearagi = xfs_calc_clear_agi_bucket_reservation(mp);
	resp->tr_growrtalloc = xfs_calc_growrtalloc_reservation(mp);
	resp->tr_growrtzero = xfs_calc_growrtzero_reservation(mp);
	resp->tr_growrtfree = xfs_calc_growrtfree_reservation(mp);
L
Linus Torvalds 已提交
223 224 225 226 227 228 229 230 231 232 233 234 235 236 237
}

/*
 * This routine is called to allocate a transaction structure.
 * The type parameter indicates the type of the transaction.  These
 * are enumerated in xfs_trans.h.
 *
 * Dynamically allocate the transaction structure from the transaction
 * zone, initialize it, and return it to the caller.
 */
xfs_trans_t *
xfs_trans_alloc(
	xfs_mount_t	*mp,
	uint		type)
{
C
Christoph Hellwig 已提交
238
	xfs_wait_for_freeze(mp, SB_FREEZE_TRANS);
239
	return _xfs_trans_alloc(mp, type, KM_SLEEP);
L
Linus Torvalds 已提交
240 241 242 243 244
}

xfs_trans_t *
_xfs_trans_alloc(
	xfs_mount_t	*mp,
245 246
	uint		type,
	uint		memflags)
L
Linus Torvalds 已提交
247 248 249
{
	xfs_trans_t	*tp;

250
	atomic_inc(&mp->m_active_trans);
L
Linus Torvalds 已提交
251

252
	tp = kmem_zone_zalloc(xfs_trans_zone, memflags);
L
Linus Torvalds 已提交
253 254 255 256 257
	tp->t_magic = XFS_TRANS_MAGIC;
	tp->t_type = type;
	tp->t_mountp = mp;
	tp->t_items_free = XFS_LIC_NUM_SLOTS;
	tp->t_busy_free = XFS_LBC_NUM_SLOTS;
258
	xfs_lic_init(&(tp->t_items));
L
Linus Torvalds 已提交
259
	XFS_LBC_INIT(&(tp->t_busy));
260
	return tp;
L
Linus Torvalds 已提交
261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286
}

/*
 * This is called to create a new transaction which will share the
 * permanent log reservation of the given transaction.  The remaining
 * unused block and rt extent reservations are also inherited.  This
 * implies that the original transaction is no longer allowed to allocate
 * blocks.  Locks and log items, however, are no inherited.  They must
 * be added to the new transaction explicitly.
 */
xfs_trans_t *
xfs_trans_dup(
	xfs_trans_t	*tp)
{
	xfs_trans_t	*ntp;

	ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);

	/*
	 * Initialize the new transaction structure.
	 */
	ntp->t_magic = XFS_TRANS_MAGIC;
	ntp->t_type = tp->t_type;
	ntp->t_mountp = tp->t_mountp;
	ntp->t_items_free = XFS_LIC_NUM_SLOTS;
	ntp->t_busy_free = XFS_LBC_NUM_SLOTS;
287
	xfs_lic_init(&(ntp->t_items));
L
Linus Torvalds 已提交
288 289 290 291
	XFS_LBC_INIT(&(ntp->t_busy));

	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
	ASSERT(tp->t_ticket != NULL);
292

L
Linus Torvalds 已提交
293
	ntp->t_flags = XFS_TRANS_PERM_LOG_RES | (tp->t_flags & XFS_TRANS_RESERVE);
294
	ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket);
L
Linus Torvalds 已提交
295 296 297 298
	ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
	tp->t_blk_res = tp->t_blk_res_used;
	ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
	tp->t_rtx_res = tp->t_rtx_res_used;
299
	ntp->t_pflags = tp->t_pflags;
L
Linus Torvalds 已提交
300

C
Christoph Hellwig 已提交
301
	xfs_trans_dup_dqinfo(tp, ntp);
L
Linus Torvalds 已提交
302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330

	atomic_inc(&tp->t_mountp->m_active_trans);
	return ntp;
}

/*
 * This is called to reserve free disk blocks and log space for the
 * given transaction.  This must be done before allocating any resources
 * within the transaction.
 *
 * This will return ENOSPC if there are not enough blocks available.
 * It will sleep waiting for available log space.
 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
 * is used by long running transactions.  If any one of the reservations
 * fails then they will all be backed out.
 *
 * This does not do quota reservations. That typically is done by the
 * caller afterwards.
 */
int
xfs_trans_reserve(
	xfs_trans_t	*tp,
	uint		blocks,
	uint		logspace,
	uint		rtextents,
	uint		flags,
	uint		logcount)
{
	int		log_flags;
331 332
	int		error = 0;
	int		rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
L
Linus Torvalds 已提交
333 334

	/* Mark this thread as being in a transaction */
335
	current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
L
Linus Torvalds 已提交
336 337 338 339 340 341 342 343

	/*
	 * Attempt to reserve the needed disk blocks by decrementing
	 * the number needed from the number available.  This will
	 * fail if the count would go below zero.
	 */
	if (blocks > 0) {
		error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS,
344
					  -((int64_t)blocks), rsvd);
L
Linus Torvalds 已提交
345
		if (error != 0) {
346
			current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
L
Linus Torvalds 已提交
347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369
			return (XFS_ERROR(ENOSPC));
		}
		tp->t_blk_res += blocks;
	}

	/*
	 * Reserve the log space needed for this transaction.
	 */
	if (logspace > 0) {
		ASSERT((tp->t_log_res == 0) || (tp->t_log_res == logspace));
		ASSERT((tp->t_log_count == 0) ||
			(tp->t_log_count == logcount));
		if (flags & XFS_TRANS_PERM_LOG_RES) {
			log_flags = XFS_LOG_PERM_RESERV;
			tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
		} else {
			ASSERT(tp->t_ticket == NULL);
			ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
			log_flags = 0;
		}

		error = xfs_log_reserve(tp->t_mountp, logspace, logcount,
					&tp->t_ticket,
370
					XFS_TRANSACTION, log_flags, tp->t_type);
L
Linus Torvalds 已提交
371 372 373 374 375 376 377 378 379 380 381 382 383 384
		if (error) {
			goto undo_blocks;
		}
		tp->t_log_res = logspace;
		tp->t_log_count = logcount;
	}

	/*
	 * Attempt to reserve the needed realtime extents by decrementing
	 * the number needed from the number available.  This will
	 * fail if the count would go below zero.
	 */
	if (rtextents > 0) {
		error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FREXTENTS,
385
					  -((int64_t)rtextents), rsvd);
L
Linus Torvalds 已提交
386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414
		if (error) {
			error = XFS_ERROR(ENOSPC);
			goto undo_log;
		}
		tp->t_rtx_res += rtextents;
	}

	return 0;

	/*
	 * Error cases jump to one of these labels to undo any
	 * reservations which have already been performed.
	 */
undo_log:
	if (logspace > 0) {
		if (flags & XFS_TRANS_PERM_LOG_RES) {
			log_flags = XFS_LOG_REL_PERM_RESERV;
		} else {
			log_flags = 0;
		}
		xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags);
		tp->t_ticket = NULL;
		tp->t_log_res = 0;
		tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
	}

undo_blocks:
	if (blocks > 0) {
		(void) xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS,
415
					 (int64_t)blocks, rsvd);
L
Linus Torvalds 已提交
416 417 418
		tp->t_blk_res = 0;
	}

419
	current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
L
Linus Torvalds 已提交
420

421
	return error;
L
Linus Torvalds 已提交
422 423 424 425 426 427 428 429 430 431
}


/*
 * Record the indicated change to the given field for application
 * to the file system's superblock when the transaction commits.
 * For now, just store the change in the transaction structure.
 *
 * Mark the transaction structure to indicate that the superblock
 * needs to be updated before committing.
D
David Chinner 已提交
432 433 434 435 436 437 438 439
 *
 * Because we may not be keeping track of allocated/free inodes and
 * used filesystem blocks in the superblock, we do not mark the
 * superblock dirty in this transaction if we modify these fields.
 * We still need to update the transaction deltas so that they get
 * applied to the incore superblock, but we don't want them to
 * cause the superblock to get locked and logged if these are the
 * only fields in the superblock that the transaction modifies.
L
Linus Torvalds 已提交
440 441 442 443 444
 */
void
xfs_trans_mod_sb(
	xfs_trans_t	*tp,
	uint		field,
445
	int64_t		delta)
L
Linus Torvalds 已提交
446
{
D
David Chinner 已提交
447 448
	uint32_t	flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
	xfs_mount_t	*mp = tp->t_mountp;
L
Linus Torvalds 已提交
449 450 451 452

	switch (field) {
	case XFS_TRANS_SB_ICOUNT:
		tp->t_icount_delta += delta;
D
David Chinner 已提交
453 454
		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
			flags &= ~XFS_TRANS_SB_DIRTY;
L
Linus Torvalds 已提交
455 456 457
		break;
	case XFS_TRANS_SB_IFREE:
		tp->t_ifree_delta += delta;
D
David Chinner 已提交
458 459
		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
			flags &= ~XFS_TRANS_SB_DIRTY;
L
Linus Torvalds 已提交
460 461 462 463 464 465 466 467 468 469 470 471
		break;
	case XFS_TRANS_SB_FDBLOCKS:
		/*
		 * Track the number of blocks allocated in the
		 * transaction.  Make sure it does not exceed the
		 * number reserved.
		 */
		if (delta < 0) {
			tp->t_blk_res_used += (uint)-delta;
			ASSERT(tp->t_blk_res_used <= tp->t_blk_res);
		}
		tp->t_fdblocks_delta += delta;
D
David Chinner 已提交
472 473
		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
			flags &= ~XFS_TRANS_SB_DIRTY;
L
Linus Torvalds 已提交
474 475 476 477 478 479 480 481 482
		break;
	case XFS_TRANS_SB_RES_FDBLOCKS:
		/*
		 * The allocation has already been applied to the
		 * in-core superblock's counter.  This should only
		 * be applied to the on-disk superblock.
		 */
		ASSERT(delta < 0);
		tp->t_res_fdblocks_delta += delta;
D
David Chinner 已提交
483 484
		if (xfs_sb_version_haslazysbcount(&mp->m_sb))
			flags &= ~XFS_TRANS_SB_DIRTY;
L
Linus Torvalds 已提交
485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500
		break;
	case XFS_TRANS_SB_FREXTENTS:
		/*
		 * Track the number of blocks allocated in the
		 * transaction.  Make sure it does not exceed the
		 * number reserved.
		 */
		if (delta < 0) {
			tp->t_rtx_res_used += (uint)-delta;
			ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
		}
		tp->t_frextents_delta += delta;
		break;
	case XFS_TRANS_SB_RES_FREXTENTS:
		/*
		 * The allocation has already been applied to the
501
		 * in-core superblock's counter.  This should only
L
Linus Torvalds 已提交
502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537
		 * be applied to the on-disk superblock.
		 */
		ASSERT(delta < 0);
		tp->t_res_frextents_delta += delta;
		break;
	case XFS_TRANS_SB_DBLOCKS:
		ASSERT(delta > 0);
		tp->t_dblocks_delta += delta;
		break;
	case XFS_TRANS_SB_AGCOUNT:
		ASSERT(delta > 0);
		tp->t_agcount_delta += delta;
		break;
	case XFS_TRANS_SB_IMAXPCT:
		tp->t_imaxpct_delta += delta;
		break;
	case XFS_TRANS_SB_REXTSIZE:
		tp->t_rextsize_delta += delta;
		break;
	case XFS_TRANS_SB_RBMBLOCKS:
		tp->t_rbmblocks_delta += delta;
		break;
	case XFS_TRANS_SB_RBLOCKS:
		tp->t_rblocks_delta += delta;
		break;
	case XFS_TRANS_SB_REXTENTS:
		tp->t_rextents_delta += delta;
		break;
	case XFS_TRANS_SB_REXTSLOG:
		tp->t_rextslog_delta += delta;
		break;
	default:
		ASSERT(0);
		return;
	}

538
	tp->t_flags |= flags;
L
Linus Torvalds 已提交
539 540 541 542 543 544 545 546 547 548 549 550 551 552
}

/*
 * xfs_trans_apply_sb_deltas() is called from the commit code
 * to bring the superblock buffer into the current transaction
 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
 *
 * For now we just look at each field allowed to change and change
 * it if necessary.
 */
STATIC void
xfs_trans_apply_sb_deltas(
	xfs_trans_t	*tp)
{
553
	xfs_dsb_t	*sbp;
L
Linus Torvalds 已提交
554 555 556 557 558 559 560 561 562 563 564 565 566
	xfs_buf_t	*bp;
	int		whole = 0;

	bp = xfs_trans_getsb(tp, tp->t_mountp, 0);
	sbp = XFS_BUF_TO_SBP(bp);

	/*
	 * Check that superblock mods match the mods made to AGF counters.
	 */
	ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
	       (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
		tp->t_ag_btree_delta));

D
David Chinner 已提交
567 568 569 570
	/*
	 * Only update the superblock counters if we are logging them
	 */
	if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) {
571
		if (tp->t_icount_delta)
572
			be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
573
		if (tp->t_ifree_delta)
574
			be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
575
		if (tp->t_fdblocks_delta)
576
			be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
577
		if (tp->t_res_fdblocks_delta)
578
			be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
L
Linus Torvalds 已提交
579 580
	}

581
	if (tp->t_frextents_delta)
582
		be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta);
583
	if (tp->t_res_frextents_delta)
584
		be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta);
585 586

	if (tp->t_dblocks_delta) {
587
		be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
L
Linus Torvalds 已提交
588 589
		whole = 1;
	}
590
	if (tp->t_agcount_delta) {
591
		be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
L
Linus Torvalds 已提交
592 593
		whole = 1;
	}
594 595
	if (tp->t_imaxpct_delta) {
		sbp->sb_imax_pct += tp->t_imaxpct_delta;
L
Linus Torvalds 已提交
596 597
		whole = 1;
	}
598
	if (tp->t_rextsize_delta) {
599
		be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
L
Linus Torvalds 已提交
600 601
		whole = 1;
	}
602
	if (tp->t_rbmblocks_delta) {
603
		be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
L
Linus Torvalds 已提交
604 605
		whole = 1;
	}
606
	if (tp->t_rblocks_delta) {
607
		be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
L
Linus Torvalds 已提交
608 609
		whole = 1;
	}
610
	if (tp->t_rextents_delta) {
611
		be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
L
Linus Torvalds 已提交
612 613
		whole = 1;
	}
614 615
	if (tp->t_rextslog_delta) {
		sbp->sb_rextslog += tp->t_rextslog_delta;
L
Linus Torvalds 已提交
616 617 618 619 620
		whole = 1;
	}

	if (whole)
		/*
621
		 * Log the whole thing, the fields are noncontiguous.
L
Linus Torvalds 已提交
622
		 */
623
		xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1);
L
Linus Torvalds 已提交
624 625 626 627 628
	else
		/*
		 * Since all the modifiable fields are contiguous, we
		 * can get away with this.
		 */
629 630
		xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount),
				  offsetof(xfs_dsb_t, sb_frextents) +
L
Linus Torvalds 已提交
631 632 633 634
				  sizeof(sbp->sb_frextents) - 1);
}

/*
635 636 637 638 639
 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations
 * and apply superblock counter changes to the in-core superblock.  The
 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
 * applied to the in-core superblock.  The idea is that that has already been
 * done.
L
Linus Torvalds 已提交
640 641
 *
 * This is done efficiently with a single call to xfs_mod_incore_sb_batch().
642 643 644 645 646 647 648 649 650 651
 * However, we have to ensure that we only modify each superblock field only
 * once because the application of the delta values may not be atomic. That can
 * lead to ENOSPC races occurring if we have two separate modifcations of the
 * free space counter to put back the entire reservation and then take away
 * what we used.
 *
 * If we are not logging superblock counters, then the inode allocated/free and
 * used block counts are not updated in the on disk superblock. In this case,
 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
 * still need to update the incore superblock with the changes.
L
Linus Torvalds 已提交
652
 */
653
STATIC void
L
Linus Torvalds 已提交
654 655 656 657 658
xfs_trans_unreserve_and_mod_sb(
	xfs_trans_t	*tp)
{
	xfs_mod_sb_t	msb[14];	/* If you add cases, add entries */
	xfs_mod_sb_t	*msbp;
D
David Chinner 已提交
659
	xfs_mount_t	*mp = tp->t_mountp;
L
Linus Torvalds 已提交
660 661 662
	/* REFERENCED */
	int		error;
	int		rsvd;
663 664
	int64_t		blkdelta = 0;
	int64_t		rtxdelta = 0;
L
Linus Torvalds 已提交
665 666 667 668

	msbp = msb;
	rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;

669 670 671 672 673 674 675 676 677 678
	/* calculate free blocks delta */
	if (tp->t_blk_res > 0)
		blkdelta = tp->t_blk_res;

	if ((tp->t_fdblocks_delta != 0) &&
	    (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
	     (tp->t_flags & XFS_TRANS_SB_DIRTY)))
	        blkdelta += tp->t_fdblocks_delta;

	if (blkdelta != 0) {
L
Linus Torvalds 已提交
679
		msbp->msb_field = XFS_SBS_FDBLOCKS;
680
		msbp->msb_delta = blkdelta;
L
Linus Torvalds 已提交
681 682 683
		msbp++;
	}

684 685 686 687 688 689 690 691 692
	/* calculate free realtime extents delta */
	if (tp->t_rtx_res > 0)
		rtxdelta = tp->t_rtx_res;

	if ((tp->t_frextents_delta != 0) &&
	    (tp->t_flags & XFS_TRANS_SB_DIRTY))
		rtxdelta += tp->t_frextents_delta;

	if (rtxdelta != 0) {
L
Linus Torvalds 已提交
693
		msbp->msb_field = XFS_SBS_FREXTENTS;
694
		msbp->msb_delta = rtxdelta;
L
Linus Torvalds 已提交
695 696 697
		msbp++;
	}

698 699
	/* apply remaining deltas */

D
David Chinner 已提交
700 701
	if (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
	     (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
L
Linus Torvalds 已提交
702 703
		if (tp->t_icount_delta != 0) {
			msbp->msb_field = XFS_SBS_ICOUNT;
704
			msbp->msb_delta = tp->t_icount_delta;
L
Linus Torvalds 已提交
705 706 707 708
			msbp++;
		}
		if (tp->t_ifree_delta != 0) {
			msbp->msb_field = XFS_SBS_IFREE;
709
			msbp->msb_delta = tp->t_ifree_delta;
L
Linus Torvalds 已提交
710 711
			msbp++;
		}
D
David Chinner 已提交
712 713 714
	}

	if (tp->t_flags & XFS_TRANS_SB_DIRTY) {
L
Linus Torvalds 已提交
715 716
		if (tp->t_dblocks_delta != 0) {
			msbp->msb_field = XFS_SBS_DBLOCKS;
717
			msbp->msb_delta = tp->t_dblocks_delta;
L
Linus Torvalds 已提交
718 719 720 721
			msbp++;
		}
		if (tp->t_agcount_delta != 0) {
			msbp->msb_field = XFS_SBS_AGCOUNT;
722
			msbp->msb_delta = tp->t_agcount_delta;
L
Linus Torvalds 已提交
723 724 725 726
			msbp++;
		}
		if (tp->t_imaxpct_delta != 0) {
			msbp->msb_field = XFS_SBS_IMAX_PCT;
727
			msbp->msb_delta = tp->t_imaxpct_delta;
L
Linus Torvalds 已提交
728 729 730 731
			msbp++;
		}
		if (tp->t_rextsize_delta != 0) {
			msbp->msb_field = XFS_SBS_REXTSIZE;
732
			msbp->msb_delta = tp->t_rextsize_delta;
L
Linus Torvalds 已提交
733 734 735 736
			msbp++;
		}
		if (tp->t_rbmblocks_delta != 0) {
			msbp->msb_field = XFS_SBS_RBMBLOCKS;
737
			msbp->msb_delta = tp->t_rbmblocks_delta;
L
Linus Torvalds 已提交
738 739 740 741
			msbp++;
		}
		if (tp->t_rblocks_delta != 0) {
			msbp->msb_field = XFS_SBS_RBLOCKS;
742
			msbp->msb_delta = tp->t_rblocks_delta;
L
Linus Torvalds 已提交
743 744 745 746
			msbp++;
		}
		if (tp->t_rextents_delta != 0) {
			msbp->msb_field = XFS_SBS_REXTENTS;
747
			msbp->msb_delta = tp->t_rextents_delta;
L
Linus Torvalds 已提交
748 749 750 751
			msbp++;
		}
		if (tp->t_rextslog_delta != 0) {
			msbp->msb_field = XFS_SBS_REXTSLOG;
752
			msbp->msb_delta = tp->t_rextslog_delta;
L
Linus Torvalds 已提交
753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783
			msbp++;
		}
	}

	/*
	 * If we need to change anything, do it.
	 */
	if (msbp > msb) {
		error = xfs_mod_incore_sb_batch(tp->t_mountp, msb,
			(uint)(msbp - msb), rsvd);
		ASSERT(error == 0);
	}
}


/*
 * xfs_trans_commit
 *
 * Commit the given transaction to the log a/synchronously.
 *
 * XFS disk error handling mechanism is not based on a typical
 * transaction abort mechanism. Logically after the filesystem
 * gets marked 'SHUTDOWN', we can't let any new transactions
 * be durable - ie. committed to disk - because some metadata might
 * be inconsistent. In such cases, this returns an error, and the
 * caller may assume that all locked objects joined to the transaction
 * have already been unlocked as if the commit had succeeded.
 * Do not reference the transaction structure after this call.
 */
 /*ARGSUSED*/
int
784
_xfs_trans_commit(
L
Linus Torvalds 已提交
785 786
	xfs_trans_t	*tp,
	uint		flags,
787
	int		*log_flushed)
L
Linus Torvalds 已提交
788 789 790 791 792 793 794 795 796 797 798
{
	xfs_log_iovec_t		*log_vector;
	int			nvec;
	xfs_mount_t		*mp;
	xfs_lsn_t		commit_lsn;
	/* REFERENCED */
	int			error;
	int			log_flags;
	int			sync;
#define	XFS_TRANS_LOGVEC_COUNT	16
	xfs_log_iovec_t		log_vector_fast[XFS_TRANS_LOGVEC_COUNT];
799
	struct xlog_in_core	*commit_iclog;
L
Linus Torvalds 已提交
800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832
	int			shutdown;

	commit_lsn = -1;

	/*
	 * Determine whether this commit is releasing a permanent
	 * log reservation or not.
	 */
	if (flags & XFS_TRANS_RELEASE_LOG_RES) {
		ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
		log_flags = XFS_LOG_REL_PERM_RESERV;
	} else {
		log_flags = 0;
	}
	mp = tp->t_mountp;

	/*
	 * If there is nothing to be logged by the transaction,
	 * then unlock all of the items associated with the
	 * transaction and free the transaction structure.
	 * Also make sure to return any reserved blocks to
	 * the free pool.
	 */
shut_us_down:
	shutdown = XFS_FORCED_SHUTDOWN(mp) ? EIO : 0;
	if (!(tp->t_flags & XFS_TRANS_DIRTY) || shutdown) {
		xfs_trans_unreserve_and_mod_sb(tp);
		/*
		 * It is indeed possible for the transaction to be
		 * not dirty but the dqinfo portion to be. All that
		 * means is that we have some (non-persistent) quota
		 * reservations that need to be unreserved.
		 */
C
Christoph Hellwig 已提交
833
		xfs_trans_unreserve_and_mod_dquots(tp);
L
Linus Torvalds 已提交
834 835 836 837 838 839
		if (tp->t_ticket) {
			commit_lsn = xfs_log_done(mp, tp->t_ticket,
							NULL, log_flags);
			if (commit_lsn == -1 && !shutdown)
				shutdown = XFS_ERROR(EIO);
		}
840
		current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
L
Linus Torvalds 已提交
841 842 843 844 845 846 847 848 849 850 851
		xfs_trans_free_items(tp, shutdown? XFS_TRANS_ABORT : 0);
		xfs_trans_free_busy(tp);
		xfs_trans_free(tp);
		XFS_STATS_INC(xs_trans_empty);
		return (shutdown);
	}
	ASSERT(tp->t_ticket != NULL);

	/*
	 * If we need to update the superblock, then do it now.
	 */
C
Christoph Hellwig 已提交
852
	if (tp->t_flags & XFS_TRANS_SB_DIRTY)
L
Linus Torvalds 已提交
853
		xfs_trans_apply_sb_deltas(tp);
C
Christoph Hellwig 已提交
854
	xfs_trans_apply_dquot_deltas(tp);
L
Linus Torvalds 已提交
855 856 857 858 859 860 861 862 863

	/*
	 * Ask each log item how many log_vector entries it will
	 * need so we can figure out how many to allocate.
	 * Try to avoid the kmem_alloc() call in the common case
	 * by using a vector from the stack when it fits.
	 */
	nvec = xfs_trans_count_vecs(tp);
	if (nvec == 0) {
864
		xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
L
Linus Torvalds 已提交
865
		goto shut_us_down;
866
	} else if (nvec <= XFS_TRANS_LOGVEC_COUNT) {
L
Linus Torvalds 已提交
867 868 869 870 871 872 873 874 875 876 877 878 879
		log_vector = log_vector_fast;
	} else {
		log_vector = (xfs_log_iovec_t *)kmem_alloc(nvec *
						   sizeof(xfs_log_iovec_t),
						   KM_SLEEP);
	}

	/*
	 * Fill in the log_vector and pin the logged items, and
	 * then write the transaction to the log.
	 */
	xfs_trans_fill_vecs(tp, log_vector);

880
	error = xfs_log_write(mp, log_vector, nvec, tp->t_ticket, &(tp->t_lsn));
L
Linus Torvalds 已提交
881 882

	/*
883 884 885
	 * The transaction is committed incore here, and can go out to disk
	 * at any time after this call.  However, all the items associated
	 * with the transaction are still locked and pinned in memory.
L
Linus Torvalds 已提交
886 887 888 889 890
	 */
	commit_lsn = xfs_log_done(mp, tp->t_ticket, &commit_iclog, log_flags);

	tp->t_commit_lsn = commit_lsn;
	if (nvec > XFS_TRANS_LOGVEC_COUNT) {
891
		kmem_free(log_vector);
L
Linus Torvalds 已提交
892 893 894 895 896 897 898
	}

	/*
	 * If we got a log write error. Unpin the logitems that we
	 * had pinned, clean up, free trans structure, and return error.
	 */
	if (error || commit_lsn == -1) {
899
		current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
L
Linus Torvalds 已提交
900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940
		xfs_trans_uncommit(tp, flags|XFS_TRANS_ABORT);
		return XFS_ERROR(EIO);
	}

	/*
	 * Once the transaction has committed, unused
	 * reservations need to be released and changes to
	 * the superblock need to be reflected in the in-core
	 * version.  Do that now.
	 */
	xfs_trans_unreserve_and_mod_sb(tp);

	sync = tp->t_flags & XFS_TRANS_SYNC;

	/*
	 * Tell the LM to call the transaction completion routine
	 * when the log write with LSN commit_lsn completes (e.g.
	 * when the transaction commit really hits the on-disk log).
	 * After this call we cannot reference tp, because the call
	 * can happen at any time and the call will free the transaction
	 * structure pointed to by tp.  The only case where we call
	 * the completion routine (xfs_trans_committed) directly is
	 * if the log is turned off on a debug kernel or we're
	 * running in simulation mode (the log is explicitly turned
	 * off).
	 */
	tp->t_logcb.cb_func = (void(*)(void*, int))xfs_trans_committed;
	tp->t_logcb.cb_arg = tp;

	/*
	 * We need to pass the iclog buffer which was used for the
	 * transaction commit record into this function, and attach
	 * the callback to it. The callback must be attached before
	 * the items are unlocked to avoid racing with other threads
	 * waiting for an item to unlock.
	 */
	shutdown = xfs_log_notify(mp, commit_iclog, &(tp->t_logcb));

	/*
	 * Mark this thread as no longer being in a transaction
	 */
941
	current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
L
Linus Torvalds 已提交
942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982

	/*
	 * Once all the items of the transaction have been copied
	 * to the in core log and the callback is attached, the
	 * items can be unlocked.
	 *
	 * This will free descriptors pointing to items which were
	 * not logged since there is nothing more to do with them.
	 * For items which were logged, we will keep pointers to them
	 * so they can be unpinned after the transaction commits to disk.
	 * This will also stamp each modified meta-data item with
	 * the commit lsn of this transaction for dependency tracking
	 * purposes.
	 */
	xfs_trans_unlock_items(tp, commit_lsn);

	/*
	 * If we detected a log error earlier, finish committing
	 * the transaction now (unpin log items, etc).
	 *
	 * Order is critical here, to avoid using the transaction
	 * pointer after its been freed (by xfs_trans_committed
	 * either here now, or as a callback).  We cannot do this
	 * step inside xfs_log_notify as was done earlier because
	 * of this issue.
	 */
	if (shutdown)
		xfs_trans_committed(tp, XFS_LI_ABORTED);

	/*
	 * Now that the xfs_trans_committed callback has been attached,
	 * and the items are released we can finally allow the iclog to
	 * go to disk.
	 */
	error = xfs_log_release_iclog(mp, commit_iclog);

	/*
	 * If the transaction needs to be synchronous, then force the
	 * log out now and wait for it.
	 */
	if (sync) {
983
		if (!error) {
984 985
			error = _xfs_log_force_lsn(mp, commit_lsn,
				      XFS_LOG_SYNC, log_flushed);
986
		}
L
Linus Torvalds 已提交
987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057
		XFS_STATS_INC(xs_trans_sync);
	} else {
		XFS_STATS_INC(xs_trans_async);
	}

	return (error);
}


/*
 * Total up the number of log iovecs needed to commit this
 * transaction.  The transaction itself needs one for the
 * transaction header.  Ask each dirty item in turn how many
 * it needs to get the total.
 */
STATIC uint
xfs_trans_count_vecs(
	xfs_trans_t	*tp)
{
	int			nvecs;
	xfs_log_item_desc_t	*lidp;

	nvecs = 1;
	lidp = xfs_trans_first_item(tp);
	ASSERT(lidp != NULL);

	/* In the non-debug case we need to start bailing out if we
	 * didn't find a log_item here, return zero and let trans_commit
	 * deal with it.
	 */
	if (lidp == NULL)
		return 0;

	while (lidp != NULL) {
		/*
		 * Skip items which aren't dirty in this transaction.
		 */
		if (!(lidp->lid_flags & XFS_LID_DIRTY)) {
			lidp = xfs_trans_next_item(tp, lidp);
			continue;
		}
		lidp->lid_size = IOP_SIZE(lidp->lid_item);
		nvecs += lidp->lid_size;
		lidp = xfs_trans_next_item(tp, lidp);
	}

	return nvecs;
}

/*
 * Called from the trans_commit code when we notice that
 * the filesystem is in the middle of a forced shutdown.
 */
STATIC void
xfs_trans_uncommit(
	xfs_trans_t	*tp,
	uint		flags)
{
	xfs_log_item_desc_t	*lidp;

	for (lidp = xfs_trans_first_item(tp);
	     lidp != NULL;
	     lidp = xfs_trans_next_item(tp, lidp)) {
		/*
		 * Unpin all but those that aren't dirty.
		 */
		if (lidp->lid_flags & XFS_LID_DIRTY)
			IOP_UNPIN_REMOVE(lidp->lid_item, tp);
	}

	xfs_trans_unreserve_and_mod_sb(tp);
C
Christoph Hellwig 已提交
1058
	xfs_trans_unreserve_and_mod_dquots(tp);
L
Linus Torvalds 已提交
1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122

	xfs_trans_free_items(tp, flags);
	xfs_trans_free_busy(tp);
	xfs_trans_free(tp);
}

/*
 * Fill in the vector with pointers to data to be logged
 * by this transaction.  The transaction header takes
 * the first vector, and then each dirty item takes the
 * number of vectors it indicated it needed in xfs_trans_count_vecs().
 *
 * As each item fills in the entries it needs, also pin the item
 * so that it cannot be flushed out until the log write completes.
 */
STATIC void
xfs_trans_fill_vecs(
	xfs_trans_t		*tp,
	xfs_log_iovec_t		*log_vector)
{
	xfs_log_item_desc_t	*lidp;
	xfs_log_iovec_t		*vecp;
	uint			nitems;

	/*
	 * Skip over the entry for the transaction header, we'll
	 * fill that in at the end.
	 */
	vecp = log_vector + 1;		/* pointer arithmetic */

	nitems = 0;
	lidp = xfs_trans_first_item(tp);
	ASSERT(lidp != NULL);
	while (lidp != NULL) {
		/*
		 * Skip items which aren't dirty in this transaction.
		 */
		if (!(lidp->lid_flags & XFS_LID_DIRTY)) {
			lidp = xfs_trans_next_item(tp, lidp);
			continue;
		}
		/*
		 * The item may be marked dirty but not log anything.
		 * This can be used to get called when a transaction
		 * is committed.
		 */
		if (lidp->lid_size) {
			nitems++;
		}
		IOP_FORMAT(lidp->lid_item, vecp);
		vecp += lidp->lid_size;		/* pointer arithmetic */
		IOP_PIN(lidp->lid_item);
		lidp = xfs_trans_next_item(tp, lidp);
	}

	/*
	 * Now that we've counted the number of items in this
	 * transaction, fill in the transaction header.
	 */
	tp->t_header.th_magic = XFS_TRANS_HEADER_MAGIC;
	tp->t_header.th_type = tp->t_type;
	tp->t_header.th_num_items = nitems;
	log_vector->i_addr = (xfs_caddr_t)&tp->t_header;
	log_vector->i_len = sizeof(xfs_trans_header_t);
C
Christoph Hellwig 已提交
1123
	log_vector->i_type = XLOG_REG_TYPE_TRANSHDR;
L
Linus Torvalds 已提交
1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146
}


/*
 * Unlock all of the transaction's items and free the transaction.
 * The transaction must not have modified any of its items, because
 * there is no way to restore them to their previous state.
 *
 * If the transaction has made a log reservation, make sure to release
 * it as well.
 */
void
xfs_trans_cancel(
	xfs_trans_t		*tp,
	int			flags)
{
	int			log_flags;
#ifdef DEBUG
	xfs_log_item_chunk_t	*licp;
	xfs_log_item_desc_t	*lidp;
	xfs_log_item_t		*lip;
	int			i;
#endif
1147
	xfs_mount_t		*mp = tp->t_mountp;
L
Linus Torvalds 已提交
1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159

	/*
	 * See if the caller is being too lazy to figure out if
	 * the transaction really needs an abort.
	 */
	if ((flags & XFS_TRANS_ABORT) && !(tp->t_flags & XFS_TRANS_DIRTY))
		flags &= ~XFS_TRANS_ABORT;
	/*
	 * See if the caller is relying on us to shut down the
	 * filesystem.  This happens in paths where we detect
	 * corruption and decide to give up.
	 */
1160
	if ((tp->t_flags & XFS_TRANS_DIRTY) && !XFS_FORCED_SHUTDOWN(mp)) {
1161
		XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
1162
		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1163
	}
L
Linus Torvalds 已提交
1164 1165 1166 1167 1168 1169
#ifdef DEBUG
	if (!(flags & XFS_TRANS_ABORT)) {
		licp = &(tp->t_items);
		while (licp != NULL) {
			lidp = licp->lic_descs;
			for (i = 0; i < licp->lic_unused; i++, lidp++) {
1170
				if (xfs_lic_isfree(licp, i)) {
L
Linus Torvalds 已提交
1171 1172 1173 1174
					continue;
				}

				lip = lidp->lid_item;
1175
				if (!XFS_FORCED_SHUTDOWN(mp))
L
Linus Torvalds 已提交
1176 1177 1178 1179 1180 1181 1182
					ASSERT(!(lip->li_type == XFS_LI_EFD));
			}
			licp = licp->lic_next;
		}
	}
#endif
	xfs_trans_unreserve_and_mod_sb(tp);
C
Christoph Hellwig 已提交
1183
	xfs_trans_unreserve_and_mod_dquots(tp);
L
Linus Torvalds 已提交
1184 1185 1186 1187 1188 1189 1190 1191

	if (tp->t_ticket) {
		if (flags & XFS_TRANS_RELEASE_LOG_RES) {
			ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
			log_flags = XFS_LOG_REL_PERM_RESERV;
		} else {
			log_flags = 0;
		}
1192
		xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
L
Linus Torvalds 已提交
1193 1194 1195
	}

	/* mark this thread as no longer being in a transaction */
1196
	current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
L
Linus Torvalds 已提交
1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212

	xfs_trans_free_items(tp, flags);
	xfs_trans_free_busy(tp);
	xfs_trans_free(tp);
}


/*
 * Free the transaction structure.  If there is more clean up
 * to do when the structure is freed, add it here.
 */
STATIC void
xfs_trans_free(
	xfs_trans_t	*tp)
{
	atomic_dec(&tp->t_mountp->m_active_trans);
C
Christoph Hellwig 已提交
1213
	xfs_trans_free_dqinfo(tp);
L
Linus Torvalds 已提交
1214 1215 1216
	kmem_zone_free(xfs_trans_zone, tp);
}

1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258
/*
 * Roll from one trans in the sequence of PERMANENT transactions to
 * the next: permanent transactions are only flushed out when
 * committed with XFS_TRANS_RELEASE_LOG_RES, but we still want as soon
 * as possible to let chunks of it go to the log. So we commit the
 * chunk we've been working on and get a new transaction to continue.
 */
int
xfs_trans_roll(
	struct xfs_trans	**tpp,
	struct xfs_inode	*dp)
{
	struct xfs_trans	*trans;
	unsigned int		logres, count;
	int			error;

	/*
	 * Ensure that the inode is always logged.
	 */
	trans = *tpp;
	xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);

	/*
	 * Copy the critical parameters from one trans to the next.
	 */
	logres = trans->t_log_res;
	count = trans->t_log_count;
	*tpp = xfs_trans_dup(trans);

	/*
	 * Commit the current transaction.
	 * If this commit failed, then it'd just unlock those items that
	 * are not marked ihold. That also means that a filesystem shutdown
	 * is in progress. The caller takes the responsibility to cancel
	 * the duplicate transaction that gets returned.
	 */
	error = xfs_trans_commit(trans, 0);
	if (error)
		return (error);

	trans = *tpp;

1259 1260 1261 1262 1263 1264 1265
	/*
	 * transaction commit worked ok so we can drop the extra ticket
	 * reference that we gained in xfs_trans_dup()
	 */
	xfs_log_ticket_put(trans->t_ticket);


1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285
	/*
	 * Reserve space in the log for th next transaction.
	 * This also pushes items in the "AIL", the list of logged items,
	 * out to disk if they are taking up space at the tail of the log
	 * that we want to use.  This requires that either nothing be locked
	 * across this call, or that anything that is locked be logged in
	 * the prior and the next transactions.
	 */
	error = xfs_trans_reserve(trans, 0, logres, 0,
				  XFS_TRANS_PERM_LOG_RES, count);
	/*
	 *  Ensure that the inode is in the new transaction and locked.
	 */
	if (error)
		return error;

	xfs_trans_ijoin(trans, dp, XFS_ILOCK_EXCL);
	xfs_trans_ihold(trans, dp);
	return 0;
}
L
Linus Torvalds 已提交
1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322

/*
 * THIS SHOULD BE REWRITTEN TO USE xfs_trans_next_item().
 *
 * This is typically called by the LM when a transaction has been fully
 * committed to disk.  It needs to unpin the items which have
 * been logged by the transaction and update their positions
 * in the AIL if necessary.
 * This also gets called when the transactions didn't get written out
 * because of an I/O error. Abortflag & XFS_LI_ABORTED is set then.
 *
 * Call xfs_trans_chunk_committed() to process the items in
 * each chunk.
 */
STATIC void
xfs_trans_committed(
	xfs_trans_t	*tp,
	int		abortflag)
{
	xfs_log_item_chunk_t	*licp;
	xfs_log_item_chunk_t	*next_licp;
	xfs_log_busy_chunk_t	*lbcp;
	xfs_log_busy_slot_t	*lbsp;
	int			i;

	/*
	 * Call the transaction's completion callback if there
	 * is one.
	 */
	if (tp->t_callback != NULL) {
		tp->t_callback(tp, tp->t_callarg);
	}

	/*
	 * Special case the chunk embedded in the transaction.
	 */
	licp = &(tp->t_items);
1323
	if (!(xfs_lic_are_all_free(licp))) {
L
Linus Torvalds 已提交
1324 1325 1326 1327 1328 1329 1330 1331
		xfs_trans_chunk_committed(licp, tp->t_lsn, abortflag);
	}

	/*
	 * Process the items in each chunk in turn.
	 */
	licp = licp->lic_next;
	while (licp != NULL) {
1332
		ASSERT(!xfs_lic_are_all_free(licp));
L
Linus Torvalds 已提交
1333 1334
		xfs_trans_chunk_committed(licp, tp->t_lsn, abortflag);
		next_licp = licp->lic_next;
1335
		kmem_free(licp);
L
Linus Torvalds 已提交
1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393
		licp = next_licp;
	}

	/*
	 * Clear all the per-AG busy list items listed in this transaction
	 */
	lbcp = &tp->t_busy;
	while (lbcp != NULL) {
		for (i = 0, lbsp = lbcp->lbc_busy; i < lbcp->lbc_unused; i++, lbsp++) {
			if (!XFS_LBC_ISFREE(lbcp, i)) {
				xfs_alloc_clear_busy(tp, lbsp->lbc_ag,
						     lbsp->lbc_idx);
			}
		}
		lbcp = lbcp->lbc_next;
	}
	xfs_trans_free_busy(tp);

	/*
	 * That's it for the transaction structure.  Free it.
	 */
	xfs_trans_free(tp);
}

/*
 * This is called to perform the commit processing for each
 * item described by the given chunk.
 *
 * The commit processing consists of unlocking items which were
 * held locked with the SYNC_UNLOCK attribute, calling the committed
 * routine of each logged item, updating the item's position in the AIL
 * if necessary, and unpinning each item.  If the committed routine
 * returns -1, then do nothing further with the item because it
 * may have been freed.
 *
 * Since items are unlocked when they are copied to the incore
 * log, it is possible for two transactions to be completing
 * and manipulating the same item simultaneously.  The AIL lock
 * will protect the lsn field of each item.  The value of this
 * field can never go backwards.
 *
 * We unpin the items after repositioning them in the AIL, because
 * otherwise they could be immediately flushed and we'd have to race
 * with the flusher trying to pull the item from the AIL as we add it.
 */
STATIC void
xfs_trans_chunk_committed(
	xfs_log_item_chunk_t	*licp,
	xfs_lsn_t		lsn,
	int			aborted)
{
	xfs_log_item_desc_t	*lidp;
	xfs_log_item_t		*lip;
	xfs_lsn_t		item_lsn;
	int			i;

	lidp = licp->lic_descs;
	for (i = 0; i < licp->lic_unused; i++, lidp++) {
1394 1395
		struct xfs_ail		*ailp;

1396
		if (xfs_lic_isfree(licp, i)) {
L
Linus Torvalds 已提交
1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431
			continue;
		}

		lip = lidp->lid_item;
		if (aborted)
			lip->li_flags |= XFS_LI_ABORTED;

		/*
		 * Send in the ABORTED flag to the COMMITTED routine
		 * so that it knows whether the transaction was aborted
		 * or not.
		 */
		item_lsn = IOP_COMMITTED(lip, lsn);

		/*
		 * If the committed routine returns -1, make
		 * no more references to the item.
		 */
		if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0) {
			continue;
		}

		/*
		 * If the returned lsn is greater than what it
		 * contained before, update the location of the
		 * item in the AIL.  If it is not, then do nothing.
		 * Items can never move backwards in the AIL.
		 *
		 * While the new lsn should usually be greater, it
		 * is possible that a later transaction completing
		 * simultaneously with an earlier one using the
		 * same item could complete first with a higher lsn.
		 * This would cause the earlier transaction to fail
		 * the test below.
		 */
1432 1433
		ailp = lip->li_ailp;
		spin_lock(&ailp->xa_lock);
L
Linus Torvalds 已提交
1434 1435 1436 1437 1438 1439
		if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0) {
			/*
			 * This will set the item's lsn to item_lsn
			 * and update the position of the item in
			 * the AIL.
			 *
1440
			 * xfs_trans_ail_update() drops the AIL lock.
L
Linus Torvalds 已提交
1441
			 */
1442
			xfs_trans_ail_update(ailp, lip, item_lsn);
L
Linus Torvalds 已提交
1443
		} else {
1444
			spin_unlock(&ailp->xa_lock);
L
Linus Torvalds 已提交
1445 1446 1447 1448 1449 1450 1451 1452 1453
		}

		/*
		 * Now that we've repositioned the item in the AIL,
		 * unpin it so it can be flushed. Pass information
		 * about buffer stale state down from the log item
		 * flags, if anyone else stales the buffer we do not
		 * want to pay any attention to it.
		 */
1454
		IOP_UNPIN(lip);
L
Linus Torvalds 已提交
1455 1456
	}
}