log.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   4 * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
   5 */
   6
   7#include <linux/sched.h>
   8#include <linux/slab.h>
   9#include <linux/spinlock.h>
  10#include <linux/completion.h>
  11#include <linux/buffer_head.h>
  12#include <linux/gfs2_ondisk.h>
  13#include <linux/crc32.h>
  14#include <linux/crc32c.h>
  15#include <linux/delay.h>
  16#include <linux/kthread.h>
  17#include <linux/freezer.h>
  18#include <linux/bio.h>
  19#include <linux/blkdev.h>
  20#include <linux/writeback.h>
  21#include <linux/list_sort.h>
  22
  23#include "gfs2.h"
  24#include "incore.h"
  25#include "bmap.h"
  26#include "glock.h"
  27#include "log.h"
  28#include "lops.h"
  29#include "meta_io.h"
  30#include "util.h"
  31#include "dir.h"
  32#include "trace_gfs2.h"
  33#include "trans.h"
  34#include "aops.h"
  35
  36static void gfs2_log_shutdown(struct gfs2_sbd *sdp);
  37
  38/**
  39 * gfs2_struct2blk - compute stuff
  40 * @sdp: the filesystem
  41 * @nstruct: the number of structures
  42 *
  43 * Compute the number of log descriptor blocks needed to hold a certain number
  44 * of structures of a certain size.
  45 *
  46 * Returns: the number of blocks needed (minimum is always 1)
  47 */
  48
  49unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct)
  50{
  51	unsigned int blks;
  52	unsigned int first, second;
  53
  54	/* The initial struct gfs2_log_descriptor block */
  55	blks = 1;
  56	first = sdp->sd_ldptrs;
  57
  58	if (nstruct > first) {
  59		/* Subsequent struct gfs2_meta_header blocks */
  60		second = sdp->sd_inptrs;
  61		blks += DIV_ROUND_UP(nstruct - first, second);
  62	}
  63
  64	return blks;
  65}
  66
  67/**
  68 * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters
  69 * @bd: The gfs2_bufdata to remove
  70 *
  71 * The ail lock _must_ be held when calling this function
  72 *
  73 */
  74
  75void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
  76{
  77	bd->bd_tr = NULL;
  78	list_del_init(&bd->bd_ail_st_list);
  79	list_del_init(&bd->bd_ail_gl_list);
  80	atomic_dec(&bd->bd_gl->gl_ail_count);
  81	brelse(bd->bd_bh);
  82}
  83
  84/**
  85 * gfs2_ail1_start_one - Start I/O on a transaction
  86 * @sdp: The superblock
  87 * @wbc: The writeback control structure
  88 * @tr: The transaction to start I/O on
  89 * @plug: The block plug currently active
  90 */
  91
  92static int gfs2_ail1_start_one(struct gfs2_sbd *sdp,
  93			       struct writeback_control *wbc,
  94			       struct gfs2_trans *tr, struct blk_plug *plug)
  95__releases(&sdp->sd_ail_lock)
  96__acquires(&sdp->sd_ail_lock)
  97{
  98	struct gfs2_glock *gl = NULL;
  99	struct address_space *mapping;
 100	struct gfs2_bufdata *bd, *s;
 101	struct buffer_head *bh;
 102	int ret = 0;
 103
 104	list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) {
 105		bh = bd->bd_bh;
 106
 107		gfs2_assert(sdp, bd->bd_tr == tr);
 108
 109		if (!buffer_busy(bh)) {
 110			if (buffer_uptodate(bh)) {
 111				list_move(&bd->bd_ail_st_list,
 112					  &tr->tr_ail2_list);
 113				continue;
 114			}
 115			if (!cmpxchg(&sdp->sd_log_error, 0, -EIO))
 116				gfs2_io_error_bh(sdp, bh);
 117		}
 118
 119		if (gfs2_withdrawn(sdp)) {
 120			gfs2_remove_from_ail(bd);
 121			continue;
 122		}
 123		if (!buffer_dirty(bh))
 124			continue;
 125		if (gl == bd->bd_gl)
 126			continue;
 127		gl = bd->bd_gl;
 128		list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list);
 129		mapping = bh->b_folio->mapping;
 130		if (!mapping)
 131			continue;
 132		spin_unlock(&sdp->sd_ail_lock);
 133		BUG_ON(GFS2_SB(mapping->host) != sdp);
 134		if (gfs2_is_jdata(GFS2_I(mapping->host)))
 135			ret = gfs2_jdata_writeback(mapping, wbc);
 136		else
 137			ret = mapping->a_ops->writepages(mapping, wbc);
 138		if (need_resched()) {
 139			blk_finish_plug(plug);
 140			cond_resched();
 141			blk_start_plug(plug);
 142		}
 143		spin_lock(&sdp->sd_ail_lock);
 144		if (ret == -ENODATA) /* if a jdata write into a new hole */
 145			ret = 0; /* ignore it */
 146		mapping_set_error(mapping, ret);
 147		if (ret || wbc->nr_to_write <= 0)
 148			break;
 149		return -EBUSY;
 150	}
 151
 152	return ret;
 153}
 154
 155static void dump_ail_list(struct gfs2_sbd *sdp)
 156{
 157	struct gfs2_trans *tr;
 158	struct gfs2_bufdata *bd;
 159	struct buffer_head *bh;
 160
 161	list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
 162		list_for_each_entry_reverse(bd, &tr->tr_ail1_list,
 163					    bd_ail_st_list) {
 164			bh = bd->bd_bh;
 165			fs_err(sdp, "bd %p: blk:0x%llx bh=%p ", bd,
 166			       (unsigned long long)bd->bd_blkno, bh);
 167			if (!bh) {
 168				fs_err(sdp, "\n");
 169				continue;
 170			}
 171			fs_err(sdp, "0x%llx up2:%d dirt:%d lkd:%d req:%d "
 172			       "map:%d new:%d ar:%d aw:%d delay:%d "
 173			       "io err:%d unwritten:%d dfr:%d pin:%d esc:%d\n",
 174			       (unsigned long long)bh->b_blocknr,
 175			       buffer_uptodate(bh), buffer_dirty(bh),
 176			       buffer_locked(bh), buffer_req(bh),
 177			       buffer_mapped(bh), buffer_new(bh),
 178			       buffer_async_read(bh), buffer_async_write(bh),
 179			       buffer_delay(bh), buffer_write_io_error(bh),
 180			       buffer_unwritten(bh),
 181			       buffer_defer_completion(bh),
 182			       buffer_pinned(bh), buffer_escaped(bh));
 183		}
 184	}
 185}
 186
 187/**
 188 * gfs2_ail1_flush - start writeback of some ail1 entries 
 189 * @sdp: The super block
 190 * @wbc: The writeback control structure
 191 *
 192 * Writes back some ail1 entries, according to the limits in the
 193 * writeback control structure
 194 */
 195
 196void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc)
 197{
 198	struct list_head *head = &sdp->sd_ail1_list;
 199	struct gfs2_trans *tr;
 200	struct blk_plug plug;
 201	int ret;
 202	unsigned long flush_start = jiffies;
 203
 204	trace_gfs2_ail_flush(sdp, wbc, 1);
 205	blk_start_plug(&plug);
 206	spin_lock(&sdp->sd_ail_lock);
 207restart:
 208	ret = 0;
 209	if (time_after(jiffies, flush_start + (HZ * 600))) {
 210		fs_err(sdp, "Error: In %s for ten minutes! t=%d\n",
 211		       __func__, current->journal_info ? 1 : 0);
 212		dump_ail_list(sdp);
 213		goto out;
 214	}
 215	list_for_each_entry_reverse(tr, head, tr_list) {
 216		if (wbc->nr_to_write <= 0)
 217			break;
 218		ret = gfs2_ail1_start_one(sdp, wbc, tr, &plug);
 219		if (ret) {
 220			if (ret == -EBUSY)
 221				goto restart;
 222			break;
 223		}
 224	}
 225out:
 226	spin_unlock(&sdp->sd_ail_lock);
 227	blk_finish_plug(&plug);
 228	if (ret) {
 229		gfs2_lm(sdp, "gfs2_ail1_start_one returned: %d\n", ret);
 230		gfs2_withdraw(sdp);
 231	}
 232	trace_gfs2_ail_flush(sdp, wbc, 0);
 233}
 234
 235/**
 236 * gfs2_ail1_start - start writeback of all ail1 entries
 237 * @sdp: The superblock
 238 */
 239
 240static void gfs2_ail1_start(struct gfs2_sbd *sdp)
 241{
 242	struct writeback_control wbc = {
 243		.sync_mode = WB_SYNC_NONE,
 244		.nr_to_write = LONG_MAX,
 245		.range_start = 0,
 246		.range_end = LLONG_MAX,
 247	};
 248
 249	return gfs2_ail1_flush(sdp, &wbc);
 250}
 251
 252static void gfs2_log_update_flush_tail(struct gfs2_sbd *sdp)
 253{
 254	unsigned int new_flush_tail = sdp->sd_log_head;
 255	struct gfs2_trans *tr;
 256
 257	if (!list_empty(&sdp->sd_ail1_list)) {
 258		tr = list_last_entry(&sdp->sd_ail1_list,
 259				     struct gfs2_trans, tr_list);
 260		new_flush_tail = tr->tr_first;
 261	}
 262	sdp->sd_log_flush_tail = new_flush_tail;
 263}
 264
 265static void gfs2_log_update_head(struct gfs2_sbd *sdp)
 266{
 267	unsigned int new_head = sdp->sd_log_flush_head;
 268
 269	if (sdp->sd_log_flush_tail == sdp->sd_log_head)
 270		sdp->sd_log_flush_tail = new_head;
 271	sdp->sd_log_head = new_head;
 272}
 273
 274/*
 275 * gfs2_ail_empty_tr - empty one of the ail lists of a transaction
 276 */
 277
 278static void gfs2_ail_empty_tr(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
 279			      struct list_head *head)
 280{
 281	struct gfs2_bufdata *bd;
 282
 283	while (!list_empty(head)) {
 284		bd = list_first_entry(head, struct gfs2_bufdata,
 285				      bd_ail_st_list);
 286		gfs2_assert(sdp, bd->bd_tr == tr);
 287		gfs2_remove_from_ail(bd);
 288	}
 289}
 290
 291/**
 292 * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
 293 * @sdp: the filesystem
 294 * @tr: the transaction
 295 * @max_revokes: If nonzero, issue revokes for the bd items for written buffers
 296 *
 297 * returns: the transaction's count of remaining active items
 298 */
 299
 300static int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
 301				int *max_revokes)
 302{
 303	struct gfs2_bufdata *bd, *s;
 304	struct buffer_head *bh;
 305	int active_count = 0;
 306
 307	list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list,
 308					 bd_ail_st_list) {
 309		bh = bd->bd_bh;
 310		gfs2_assert(sdp, bd->bd_tr == tr);
 311		/*
 312		 * If another process flagged an io error, e.g. writing to the
 313		 * journal, error all other bhs and move them off the ail1 to
 314		 * prevent a tight loop when unmount tries to flush ail1,
 315		 * regardless of whether they're still busy. If no outside
 316		 * errors were found and the buffer is busy, move to the next.
 317		 * If the ail buffer is not busy and caught an error, flag it
 318		 * for others.
 319		 */
 320		if (!sdp->sd_log_error && buffer_busy(bh)) {
 321			active_count++;
 322			continue;
 323		}
 324		if (!buffer_uptodate(bh) &&
 325		    !cmpxchg(&sdp->sd_log_error, 0, -EIO))
 326			gfs2_io_error_bh(sdp, bh);
 327		/*
 328		 * If we have space for revokes and the bd is no longer on any
 329		 * buf list, we can just add a revoke for it immediately and
 330		 * avoid having to put it on the ail2 list, where it would need
 331		 * to be revoked later.
 332		 */
 333		if (*max_revokes && list_empty(&bd->bd_list)) {
 334			gfs2_add_revoke(sdp, bd);
 335			(*max_revokes)--;
 336			continue;
 337		}
 338		list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
 339	}
 340	return active_count;
 341}
 342
 343/**
 344 * gfs2_ail1_empty - Try to empty the ail1 lists
 345 * @sdp: The superblock
 346 * @max_revokes: If non-zero, add revokes where appropriate
 347 *
 348 * Tries to empty the ail1 lists, starting with the oldest first.
 349 * Returns %true if the ail1 list is now empty.
 350 */
 351
 352static bool gfs2_ail1_empty(struct gfs2_sbd *sdp, int max_revokes)
 353{
 354	struct gfs2_trans *tr, *s;
 355	int oldest_tr = 1;
 356	bool empty;
 357
 358	spin_lock(&sdp->sd_ail_lock);
 359	list_for_each_entry_safe_reverse(tr, s, &sdp->sd_ail1_list, tr_list) {
 360		if (!gfs2_ail1_empty_one(sdp, tr, &max_revokes) && oldest_tr)
 361			list_move(&tr->tr_list, &sdp->sd_ail2_list);
 362		else
 363			oldest_tr = 0;
 364	}
 365	gfs2_log_update_flush_tail(sdp);
 366	empty = list_empty(&sdp->sd_ail1_list);
 367	spin_unlock(&sdp->sd_ail_lock);
 368
 369	return empty;
 370}
 371
 372static void gfs2_ail1_wait(struct gfs2_sbd *sdp)
 373{
 374	struct gfs2_trans *tr;
 375	struct gfs2_bufdata *bd;
 376	struct buffer_head *bh;
 377
 378	spin_lock(&sdp->sd_ail_lock);
 379	list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
 380		list_for_each_entry(bd, &tr->tr_ail1_list, bd_ail_st_list) {
 381			bh = bd->bd_bh;
 382			if (!buffer_locked(bh))
 383				continue;
 384			get_bh(bh);
 385			spin_unlock(&sdp->sd_ail_lock);
 386			wait_on_buffer(bh);
 387			brelse(bh);
 388			return;
 389		}
 390	}
 391	spin_unlock(&sdp->sd_ail_lock);
 392}
 393
 394static void __ail2_empty(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 395{
 396	gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
 397	list_del(&tr->tr_list);
 398	gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list));
 399	gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list));
 400	gfs2_trans_free(sdp, tr);
 401}
 402
 403static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
 404{
 405	struct list_head *ail2_list = &sdp->sd_ail2_list;
 406	unsigned int old_tail = sdp->sd_log_tail;
 407	struct gfs2_trans *tr, *safe;
 408
 409	spin_lock(&sdp->sd_ail_lock);
 410	if (old_tail <= new_tail) {
 411		list_for_each_entry_safe(tr, safe, ail2_list, tr_list) {
 412			if (old_tail <= tr->tr_first && tr->tr_first < new_tail)
 413				__ail2_empty(sdp, tr);
 414		}
 415	} else {
 416		list_for_each_entry_safe(tr, safe, ail2_list, tr_list) {
 417			if (old_tail <= tr->tr_first || tr->tr_first < new_tail)
 418				__ail2_empty(sdp, tr);
 419		}
 420	}
 421	spin_unlock(&sdp->sd_ail_lock);
 422}
 423
 424/**
 425 * gfs2_log_is_empty - Check if the log is empty
 426 * @sdp: The GFS2 superblock
 427 */
 428
 429bool gfs2_log_is_empty(struct gfs2_sbd *sdp) {
 430	return atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks;
 431}
 432
 433static bool __gfs2_log_try_reserve_revokes(struct gfs2_sbd *sdp, unsigned int revokes)
 434{
 435	unsigned int available;
 436
 437	available = atomic_read(&sdp->sd_log_revokes_available);
 438	while (available >= revokes) {
 439		if (atomic_try_cmpxchg(&sdp->sd_log_revokes_available,
 440				       &available, available - revokes))
 441			return true;
 442	}
 443	return false;
 444}
 445
 446/**
 447 * gfs2_log_release_revokes - Release a given number of revokes
 448 * @sdp: The GFS2 superblock
 449 * @revokes: The number of revokes to release
 450 *
 451 * sdp->sd_log_flush_lock must be held.
 452 */
 453void gfs2_log_release_revokes(struct gfs2_sbd *sdp, unsigned int revokes)
 454{
 455	if (revokes)
 456		atomic_add(revokes, &sdp->sd_log_revokes_available);
 457}
 458
 459/**
 460 * gfs2_log_release - Release a given number of log blocks
 461 * @sdp: The GFS2 superblock
 462 * @blks: The number of blocks
 463 *
 464 */
 465
 466void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
 467{
 468	atomic_add(blks, &sdp->sd_log_blks_free);
 469	trace_gfs2_log_blocks(sdp, blks);
 470	gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
 471				  sdp->sd_jdesc->jd_blocks);
 472	if (atomic_read(&sdp->sd_log_blks_needed))
 473		wake_up(&sdp->sd_log_waitq);
 474}
 475
 476/**
 477 * __gfs2_log_try_reserve - Try to make a log reservation
 478 * @sdp: The GFS2 superblock
 479 * @blks: The number of blocks to reserve
 480 * @taboo_blks: The number of blocks to leave free
 481 *
 482 * Try to do the same as __gfs2_log_reserve(), but fail if no more log
 483 * space is immediately available.
 484 */
 485static bool __gfs2_log_try_reserve(struct gfs2_sbd *sdp, unsigned int blks,
 486				   unsigned int taboo_blks)
 487{
 488	unsigned wanted = blks + taboo_blks;
 489	unsigned int free_blocks;
 490
 491	free_blocks = atomic_read(&sdp->sd_log_blks_free);
 492	while (free_blocks >= wanted) {
 493		if (atomic_try_cmpxchg(&sdp->sd_log_blks_free, &free_blocks,
 494				       free_blocks - blks)) {
 495			trace_gfs2_log_blocks(sdp, -blks);
 496			return true;
 497		}
 498	}
 499	return false;
 500}
 501
 502/**
 503 * __gfs2_log_reserve - Make a log reservation
 504 * @sdp: The GFS2 superblock
 505 * @blks: The number of blocks to reserve
 506 * @taboo_blks: The number of blocks to leave free
 507 *
 508 * @taboo_blks is set to 0 for logd, and to GFS2_LOG_FLUSH_MIN_BLOCKS
 509 * for all other processes.  This ensures that when the log is almost full,
 510 * logd will still be able to call gfs2_log_flush one more time  without
 511 * blocking, which will advance the tail and make some more log space
 512 * available.
 513 *
 514 * We no longer flush the log here, instead we wake up logd to do that
 515 * for us. To avoid the thundering herd and to ensure that we deal fairly
 516 * with queued waiters, we use an exclusive wait. This means that when we
 517 * get woken with enough journal space to get our reservation, we need to
 518 * wake the next waiter on the list.
 519 */
 520
 521static void __gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks,
 522			       unsigned int taboo_blks)
 523{
 524	unsigned wanted = blks + taboo_blks;
 525	unsigned int free_blocks;
 526
 527	atomic_add(blks, &sdp->sd_log_blks_needed);
 528	for (;;) {
 529		if (current != sdp->sd_logd_process)
 530			wake_up(&sdp->sd_logd_waitq);
 531		io_wait_event(sdp->sd_log_waitq,
 532			(free_blocks = atomic_read(&sdp->sd_log_blks_free),
 533			 free_blocks >= wanted));
 534		do {
 535			if (atomic_try_cmpxchg(&sdp->sd_log_blks_free,
 536					       &free_blocks,
 537					       free_blocks - blks))
 538				goto reserved;
 539		} while (free_blocks >= wanted);
 540	}
 541
 542reserved:
 543	trace_gfs2_log_blocks(sdp, -blks);
 544	if (atomic_sub_return(blks, &sdp->sd_log_blks_needed))
 545		wake_up(&sdp->sd_log_waitq);
 546}
 547
 548/**
 549 * gfs2_log_try_reserve - Try to make a log reservation
 550 * @sdp: The GFS2 superblock
 551 * @tr: The transaction
 552 * @extra_revokes: The number of additional revokes reserved (output)
 553 *
 554 * This is similar to gfs2_log_reserve, but sdp->sd_log_flush_lock must be
 555 * held for correct revoke accounting.
 556 */
 557
 558bool gfs2_log_try_reserve(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
 559			  unsigned int *extra_revokes)
 560{
 561	unsigned int blks = tr->tr_reserved;
 562	unsigned int revokes = tr->tr_revokes;
 563	unsigned int revoke_blks = 0;
 564
 565	*extra_revokes = 0;
 566	if (revokes && !__gfs2_log_try_reserve_revokes(sdp, revokes)) {
 567		revoke_blks = DIV_ROUND_UP(revokes, sdp->sd_inptrs);
 568		*extra_revokes = revoke_blks * sdp->sd_inptrs - revokes;
 569		blks += revoke_blks;
 570	}
 571	if (!blks)
 572		return true;
 573	if (__gfs2_log_try_reserve(sdp, blks, GFS2_LOG_FLUSH_MIN_BLOCKS))
 574		return true;
 575	if (!revoke_blks)
 576		gfs2_log_release_revokes(sdp, revokes);
 577	return false;
 578}
 579
 580/**
 581 * gfs2_log_reserve - Make a log reservation
 582 * @sdp: The GFS2 superblock
 583 * @tr: The transaction
 584 * @extra_revokes: The number of additional revokes reserved (output)
 585 *
 586 * sdp->sd_log_flush_lock must not be held.
 587 */
 588
 589void gfs2_log_reserve(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
 590		      unsigned int *extra_revokes)
 591{
 592	unsigned int blks = tr->tr_reserved;
 593	unsigned int revokes = tr->tr_revokes;
 594	unsigned int revoke_blks;
 595
 596	*extra_revokes = 0;
 597	if (revokes) {
 598		revoke_blks = DIV_ROUND_UP(revokes, sdp->sd_inptrs);
 599		*extra_revokes = revoke_blks * sdp->sd_inptrs - revokes;
 600		blks += revoke_blks;
 601	}
 602	__gfs2_log_reserve(sdp, blks, GFS2_LOG_FLUSH_MIN_BLOCKS);
 603}
 604
 605/**
 606 * log_distance - Compute distance between two journal blocks
 607 * @sdp: The GFS2 superblock
 608 * @newer: The most recent journal block of the pair
 609 * @older: The older journal block of the pair
 610 *
 611 *   Compute the distance (in the journal direction) between two
 612 *   blocks in the journal
 613 *
 614 * Returns: the distance in blocks
 615 */
 616
 617static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
 618					unsigned int older)
 619{
 620	int dist;
 621
 622	dist = newer - older;
 623	if (dist < 0)
 624		dist += sdp->sd_jdesc->jd_blocks;
 625
 626	return dist;
 627}
 628
 629/**
 630 * calc_reserved - Calculate the number of blocks to keep reserved
 631 * @sdp: The GFS2 superblock
 632 *
 633 * This is complex.  We need to reserve room for all our currently used
 634 * metadata blocks (e.g. normal file I/O rewriting file time stamps) and
 635 * all our journaled data blocks for journaled files (e.g. files in the
 636 * meta_fs like rindex, or files for which chattr +j was done.)
 637 * If we don't reserve enough space, corruption will follow.
 638 *
 639 * We can have metadata blocks and jdata blocks in the same journal.  Each
 640 * type gets its own log descriptor, for which we need to reserve a block.
 641 * In fact, each type has the potential for needing more than one log descriptor
 642 * in cases where we have more blocks than will fit in a log descriptor.
 643 * Metadata journal entries take up half the space of journaled buffer entries.
 644 *
 645 * Also, we need to reserve blocks for revoke journal entries and one for an
 646 * overall header for the lot.
 647 *
 648 * Returns: the number of blocks reserved
 649 */
 650static unsigned int calc_reserved(struct gfs2_sbd *sdp)
 651{
 652	unsigned int reserved = GFS2_LOG_FLUSH_MIN_BLOCKS;
 653	unsigned int blocks;
 654	struct gfs2_trans *tr = sdp->sd_log_tr;
 655
 656	if (tr) {
 657		blocks = tr->tr_num_buf_new - tr->tr_num_buf_rm;
 658		reserved += blocks + DIV_ROUND_UP(blocks, buf_limit(sdp));
 659		blocks = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
 660		reserved += blocks + DIV_ROUND_UP(blocks, databuf_limit(sdp));
 661	}
 662	return reserved;
 663}
 664
 665static void log_pull_tail(struct gfs2_sbd *sdp)
 666{
 667	unsigned int new_tail = sdp->sd_log_flush_tail;
 668	unsigned int dist;
 669
 670	if (new_tail == sdp->sd_log_tail)
 671		return;
 672	dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
 673	ail2_empty(sdp, new_tail);
 674	gfs2_log_release(sdp, dist);
 675	sdp->sd_log_tail = new_tail;
 676}
 677
 678
 679void log_flush_wait(struct gfs2_sbd *sdp)
 680{
 681	DEFINE_WAIT(wait);
 682
 683	if (atomic_read(&sdp->sd_log_in_flight)) {
 684		do {
 685			prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
 686					TASK_UNINTERRUPTIBLE);
 687			if (atomic_read(&sdp->sd_log_in_flight))
 688				io_schedule();
 689		} while(atomic_read(&sdp->sd_log_in_flight));
 690		finish_wait(&sdp->sd_log_flush_wait, &wait);
 691	}
 692}
 693
 694static int ip_cmp(void *priv, const struct list_head *a, const struct list_head *b)
 695{
 696	struct gfs2_inode *ipa, *ipb;
 697
 698	ipa = list_entry(a, struct gfs2_inode, i_ordered);
 699	ipb = list_entry(b, struct gfs2_inode, i_ordered);
 700
 701	if (ipa->i_no_addr < ipb->i_no_addr)
 702		return -1;
 703	if (ipa->i_no_addr > ipb->i_no_addr)
 704		return 1;
 705	return 0;
 706}
 707
 708static void __ordered_del_inode(struct gfs2_inode *ip)
 709{
 710	if (!list_empty(&ip->i_ordered))
 711		list_del_init(&ip->i_ordered);
 712}
 713
 714static void gfs2_ordered_write(struct gfs2_sbd *sdp)
 715{
 716	struct gfs2_inode *ip;
 717	LIST_HEAD(written);
 718
 719	spin_lock(&sdp->sd_ordered_lock);
 720	list_sort(NULL, &sdp->sd_log_ordered, &ip_cmp);
 721	while (!list_empty(&sdp->sd_log_ordered)) {
 722		ip = list_first_entry(&sdp->sd_log_ordered, struct gfs2_inode, i_ordered);
 723		if (ip->i_inode.i_mapping->nrpages == 0) {
 724			__ordered_del_inode(ip);
 725			continue;
 726		}
 727		list_move(&ip->i_ordered, &written);
 728		spin_unlock(&sdp->sd_ordered_lock);
 729		filemap_fdatawrite(ip->i_inode.i_mapping);
 730		spin_lock(&sdp->sd_ordered_lock);
 731	}
 732	list_splice(&written, &sdp->sd_log_ordered);
 733	spin_unlock(&sdp->sd_ordered_lock);
 734}
 735
 736static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
 737{
 738	struct gfs2_inode *ip;
 739
 740	spin_lock(&sdp->sd_ordered_lock);
 741	while (!list_empty(&sdp->sd_log_ordered)) {
 742		ip = list_first_entry(&sdp->sd_log_ordered, struct gfs2_inode, i_ordered);
 743		__ordered_del_inode(ip);
 744		if (ip->i_inode.i_mapping->nrpages == 0)
 745			continue;
 746		spin_unlock(&sdp->sd_ordered_lock);
 747		filemap_fdatawait(ip->i_inode.i_mapping);
 748		spin_lock(&sdp->sd_ordered_lock);
 749	}
 750	spin_unlock(&sdp->sd_ordered_lock);
 751}
 752
 753void gfs2_ordered_del_inode(struct gfs2_inode *ip)
 754{
 755	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 756
 757	spin_lock(&sdp->sd_ordered_lock);
 758	__ordered_del_inode(ip);
 759	spin_unlock(&sdp->sd_ordered_lock);
 760}
 761
 762void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
 763{
 764	struct buffer_head *bh = bd->bd_bh;
 765	struct gfs2_glock *gl = bd->bd_gl;
 766
 767	sdp->sd_log_num_revoke++;
 768	if (atomic_inc_return(&gl->gl_revokes) == 1)
 769		gfs2_glock_hold(gl);
 770	bh->b_private = NULL;
 771	bd->bd_blkno = bh->b_blocknr;
 772	gfs2_remove_from_ail(bd); /* drops ref on bh */
 773	bd->bd_bh = NULL;
 774	set_bit(GLF_LFLUSH, &gl->gl_flags);
 775	list_add(&bd->bd_list, &sdp->sd_log_revokes);
 776}
 777
 778void gfs2_glock_remove_revoke(struct gfs2_glock *gl)
 779{
 780	if (atomic_dec_return(&gl->gl_revokes) == 0) {
 781		clear_bit(GLF_LFLUSH, &gl->gl_flags);
 782		gfs2_glock_put_async(gl);
 783	}
 784}
 785
 786/**
 787 * gfs2_flush_revokes - Add as many revokes to the system transaction as we can
 788 * @sdp: The GFS2 superblock
 789 *
 790 * Our usual strategy is to defer writing revokes as much as we can in the hope
 791 * that we'll eventually overwrite the journal, which will make those revokes
 792 * go away.  This changes when we flush the log: at that point, there will
 793 * likely be some left-over space in the last revoke block of that transaction.
 794 * We can fill that space with additional revokes for blocks that have already
 795 * been written back.  This will basically come at no cost now, and will save
 796 * us from having to keep track of those blocks on the AIL2 list later.
 797 */
 798void gfs2_flush_revokes(struct gfs2_sbd *sdp)
 799{
 800	/* number of revokes we still have room for */
 801	unsigned int max_revokes = atomic_read(&sdp->sd_log_revokes_available);
 802
 803	gfs2_log_lock(sdp);
 804	gfs2_ail1_empty(sdp, max_revokes);
 805	gfs2_log_unlock(sdp);
 806}
 807
 808/**
 809 * gfs2_write_log_header - Write a journal log header buffer at lblock
 810 * @sdp: The GFS2 superblock
 811 * @jd: journal descriptor of the journal to which we are writing
 812 * @seq: sequence number
 813 * @tail: tail of the log
 814 * @lblock: value for lh_blkno (block number relative to start of journal)
 815 * @flags: log header flags GFS2_LOG_HEAD_*
 816 * @op_flags: flags to pass to the bio
 817 *
 818 * Returns: the initialized log buffer descriptor
 819 */
 820
 821void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
 822			   u64 seq, u32 tail, u32 lblock, u32 flags,
 823			   blk_opf_t op_flags)
 824{
 825	struct gfs2_log_header *lh;
 826	u32 hash, crc;
 827	struct page *page;
 828	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 829	struct timespec64 tv;
 830	struct super_block *sb = sdp->sd_vfs;
 831	u64 dblock;
 832
 833	if (gfs2_withdrawn(sdp))
 834		return;
 835
 836	page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 837	lh = page_address(page);
 838	clear_page(lh);
 839
 840	lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 841	lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
 842	lh->lh_header.__pad0 = cpu_to_be64(0);
 843	lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
 844	lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
 845	lh->lh_sequence = cpu_to_be64(seq);
 846	lh->lh_flags = cpu_to_be32(flags);
 847	lh->lh_tail = cpu_to_be32(tail);
 848	lh->lh_blkno = cpu_to_be32(lblock);
 849	hash = ~crc32(~0, lh, LH_V1_SIZE);
 850	lh->lh_hash = cpu_to_be32(hash);
 851
 852	ktime_get_coarse_real_ts64(&tv);
 853	lh->lh_nsec = cpu_to_be32(tv.tv_nsec);
 854	lh->lh_sec = cpu_to_be64(tv.tv_sec);
 855	if (!list_empty(&jd->extent_list))
 856		dblock = gfs2_log_bmap(jd, lblock);
 857	else {
 858		unsigned int extlen;
 859		int ret;
 860
 861		extlen = 1;
 862		ret = gfs2_get_extent(jd->jd_inode, lblock, &dblock, &extlen);
 863		if (gfs2_assert_withdraw(sdp, ret == 0))
 864			return;
 865	}
 866	lh->lh_addr = cpu_to_be64(dblock);
 867	lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr);
 868
 869	/* We may only write local statfs, quota, etc., when writing to our
 870	   own journal. The values are left 0 when recovering a journal
 871	   different from our own. */
 872	if (!(flags & GFS2_LOG_HEAD_RECOVERY)) {
 873		lh->lh_statfs_addr =
 874			cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr);
 875		lh->lh_quota_addr =
 876			cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr);
 877
 878		spin_lock(&sdp->sd_statfs_spin);
 879		lh->lh_local_total = cpu_to_be64(l_sc->sc_total);
 880		lh->lh_local_free = cpu_to_be64(l_sc->sc_free);
 881		lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes);
 882		spin_unlock(&sdp->sd_statfs_spin);
 883	}
 884
 885	BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE);
 886
 887	crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
 888		     sb->s_blocksize - LH_V1_SIZE - 4);
 889	lh->lh_crc = cpu_to_be32(crc);
 890
 891	gfs2_log_write(sdp, jd, page, sb->s_blocksize, 0, dblock,
 892		       REQ_OP_WRITE | op_flags);
 893	gfs2_log_submit_write(&jd->jd_log_bio);
 894}
 895
 896/**
 897 * log_write_header - Get and initialize a journal header buffer
 898 * @sdp: The GFS2 superblock
 899 * @flags: The log header flags, including log header origin
 900 *
 901 * Returns: the initialized log buffer descriptor
 902 */
 903
 904static void log_write_header(struct gfs2_sbd *sdp, u32 flags)
 905{
 906	blk_opf_t op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC;
 907	struct super_block *sb = sdp->sd_vfs;
 908
 909	gfs2_assert_withdraw(sdp, sb->s_writers.frozen != SB_FREEZE_COMPLETE);
 910
 911	if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) {
 912		gfs2_ordered_wait(sdp);
 913		log_flush_wait(sdp);
 914		op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
 915	}
 916	sdp->sd_log_idle = (sdp->sd_log_flush_tail == sdp->sd_log_flush_head);
 917	gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++,
 918			      sdp->sd_log_flush_tail, sdp->sd_log_flush_head,
 919			      flags, op_flags);
 920	gfs2_log_incr_head(sdp);
 921	log_flush_wait(sdp);
 922	log_pull_tail(sdp);
 923	gfs2_log_update_head(sdp);
 924}
 925
 926/**
 927 * gfs2_ail_drain - drain the ail lists after a withdraw
 928 * @sdp: Pointer to GFS2 superblock
 929 */
 930void gfs2_ail_drain(struct gfs2_sbd *sdp)
 931{
 932	struct gfs2_trans *tr;
 933
 934	spin_lock(&sdp->sd_ail_lock);
 935	/*
 936	 * For transactions on the sd_ail1_list we need to drain both the
 937	 * ail1 and ail2 lists. That's because function gfs2_ail1_start_one
 938	 * (temporarily) moves items from its tr_ail1 list to tr_ail2 list
 939	 * before revokes are sent for that block. Items on the sd_ail2_list
 940	 * should have already gotten beyond that point, so no need.
 941	 */
 942	while (!list_empty(&sdp->sd_ail1_list)) {
 943		tr = list_first_entry(&sdp->sd_ail1_list, struct gfs2_trans,
 944				      tr_list);
 945		gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail1_list);
 946		gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
 947		list_del(&tr->tr_list);
 948		gfs2_trans_free(sdp, tr);
 949	}
 950	while (!list_empty(&sdp->sd_ail2_list)) {
 951		tr = list_first_entry(&sdp->sd_ail2_list, struct gfs2_trans,
 952				      tr_list);
 953		gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
 954		list_del(&tr->tr_list);
 955		gfs2_trans_free(sdp, tr);
 956	}
 957	gfs2_drain_revokes(sdp);
 958	spin_unlock(&sdp->sd_ail_lock);
 959}
 960
 961/**
 962 * empty_ail1_list - try to start IO and empty the ail1 list
 963 * @sdp: Pointer to GFS2 superblock
 964 */
 965static void empty_ail1_list(struct gfs2_sbd *sdp)
 966{
 967	unsigned long start = jiffies;
 968	bool empty = false;
 969
 970	while (!empty) {
 971		if (time_after(jiffies, start + (HZ * 600))) {
 972			fs_err(sdp, "Error: In %s for 10 minutes! t=%d\n",
 973			       __func__, current->journal_info ? 1 : 0);
 974			dump_ail_list(sdp);
 975			return;
 976		}
 977		gfs2_ail1_start(sdp);
 978		gfs2_ail1_wait(sdp);
 979		empty = gfs2_ail1_empty(sdp, 0);
 980
 981		if (gfs2_withdrawn(sdp))
 982			break;
 983	}
 984}
 985
 986/**
 987 * trans_drain - drain the buf and databuf queue for a failed transaction
 988 * @tr: the transaction to drain
 989 *
 990 * When this is called, we're taking an error exit for a log write that failed
 991 * but since we bypassed the after_commit functions, we need to remove the
 992 * items from the buf and databuf queue.
 993 */
 994static void trans_drain(struct gfs2_trans *tr)
 995{
 996	struct gfs2_bufdata *bd;
 997	struct list_head *head;
 998
 999	if (!tr)
1000		return;
1001
1002	head = &tr->tr_buf;
1003	while (!list_empty(head)) {
1004		bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
1005		list_del_init(&bd->bd_list);
1006		if (!list_empty(&bd->bd_ail_st_list))
1007			gfs2_remove_from_ail(bd);
1008		kmem_cache_free(gfs2_bufdata_cachep, bd);
1009	}
1010	head = &tr->tr_databuf;
1011	while (!list_empty(head)) {
1012		bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
1013		list_del_init(&bd->bd_list);
1014		if (!list_empty(&bd->bd_ail_st_list))
1015			gfs2_remove_from_ail(bd);
1016		kmem_cache_free(gfs2_bufdata_cachep, bd);
1017	}
1018}
1019
1020/**
1021 * gfs2_log_flush - flush incore transaction(s)
1022 * @sdp: The filesystem
1023 * @gl: The glock structure to flush.  If NULL, flush the whole incore log
1024 * @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags
1025 *
1026 */
1027
1028void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags)
1029{
1030	struct gfs2_trans *tr = NULL;
1031	unsigned int reserved_blocks = 0, used_blocks = 0;
1032	bool frozen = test_bit(SDF_FROZEN, &sdp->sd_flags);
1033	unsigned int first_log_head;
1034	unsigned int reserved_revokes = 0;
1035
1036	down_write(&sdp->sd_log_flush_lock);
1037	trace_gfs2_log_flush(sdp, 1, flags);
1038
1039repeat:
1040	/*
1041	 * Do this check while holding the log_flush_lock to prevent new
1042	 * buffers from being added to the ail via gfs2_pin()
1043	 */
1044	if (gfs2_withdrawn(sdp) ||
1045	    !test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
1046		goto out;
1047
1048	/* Log might have been flushed while we waited for the flush lock */
1049	if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags))
1050		goto out;
1051
1052	first_log_head = sdp->sd_log_head;
1053	sdp->sd_log_flush_head = first_log_head;
1054
1055	tr = sdp->sd_log_tr;
1056	if (tr || sdp->sd_log_num_revoke) {
1057		if (reserved_blocks)
1058			gfs2_log_release(sdp, reserved_blocks);
1059		reserved_blocks = sdp->sd_log_blks_reserved;
1060		reserved_revokes = sdp->sd_log_num_revoke;
1061		if (tr) {
1062			sdp->sd_log_tr = NULL;
1063			tr->tr_first = first_log_head;
1064			if (unlikely(frozen)) {
1065				if (gfs2_assert_withdraw(sdp,
1066				       !tr->tr_num_buf_new && !tr->tr_num_databuf_new))
1067					goto out_withdraw;
1068			}
1069		}
1070	} else if (!reserved_blocks) {
1071		unsigned int taboo_blocks = GFS2_LOG_FLUSH_MIN_BLOCKS;
1072
1073		reserved_blocks = GFS2_LOG_FLUSH_MIN_BLOCKS;
1074		if (current == sdp->sd_logd_process)
1075			taboo_blocks = 0;
1076
1077		if (!__gfs2_log_try_reserve(sdp, reserved_blocks, taboo_blocks)) {
1078			up_write(&sdp->sd_log_flush_lock);
1079			__gfs2_log_reserve(sdp, reserved_blocks, taboo_blocks);
1080			down_write(&sdp->sd_log_flush_lock);
1081			goto repeat;
1082		}
1083		BUG_ON(sdp->sd_log_num_revoke);
1084	}
1085
1086	if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN)
1087		clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
1088
1089	if (unlikely(frozen))
1090		if (gfs2_assert_withdraw(sdp, !reserved_revokes))
1091			goto out_withdraw;
1092
1093	gfs2_ordered_write(sdp);
1094	if (gfs2_withdrawn(sdp))
1095		goto out_withdraw;
1096	lops_before_commit(sdp, tr);
1097	if (gfs2_withdrawn(sdp))
1098		goto out_withdraw;
1099	if (sdp->sd_jdesc)
1100		gfs2_log_submit_write(&sdp->sd_jdesc->jd_log_bio);
1101	if (gfs2_withdrawn(sdp))
1102		goto out_withdraw;
1103
1104	if (sdp->sd_log_head != sdp->sd_log_flush_head) {
1105		log_write_header(sdp, flags);
1106	} else if (sdp->sd_log_tail != sdp->sd_log_flush_tail && !sdp->sd_log_idle) {
1107		log_write_header(sdp, flags);
1108	}
1109	if (gfs2_withdrawn(sdp))
1110		goto out_withdraw;
1111	lops_after_commit(sdp, tr);
1112
1113	gfs2_log_lock(sdp);
1114	sdp->sd_log_blks_reserved = 0;
1115
1116	spin_lock(&sdp->sd_ail_lock);
1117	if (tr && !list_empty(&tr->tr_ail1_list)) {
1118		list_add(&tr->tr_list, &sdp->sd_ail1_list);
1119		tr = NULL;
1120	}
1121	spin_unlock(&sdp->sd_ail_lock);
1122	gfs2_log_unlock(sdp);
1123
1124	if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) {
1125		if (!sdp->sd_log_idle) {
1126			empty_ail1_list(sdp);
1127			if (gfs2_withdrawn(sdp))
1128				goto out_withdraw;
1129			log_write_header(sdp, flags);
1130		}
1131		if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
1132			     GFS2_LOG_HEAD_FLUSH_FREEZE))
1133			gfs2_log_shutdown(sdp);
1134	}
1135
1136out_end:
1137	used_blocks = log_distance(sdp, sdp->sd_log_flush_head, first_log_head);
1138	reserved_revokes += atomic_read(&sdp->sd_log_revokes_available);
1139	atomic_set(&sdp->sd_log_revokes_available, sdp->sd_ldptrs);
1140	gfs2_assert_withdraw(sdp, reserved_revokes % sdp->sd_inptrs == sdp->sd_ldptrs);
1141	if (reserved_revokes > sdp->sd_ldptrs)
1142		reserved_blocks += (reserved_revokes - sdp->sd_ldptrs) / sdp->sd_inptrs;
1143out:
1144	if (used_blocks != reserved_blocks) {
1145		gfs2_assert_withdraw(sdp, used_blocks < reserved_blocks);
1146		gfs2_log_release(sdp, reserved_blocks - used_blocks);
1147	}
1148	up_write(&sdp->sd_log_flush_lock);
1149	gfs2_trans_free(sdp, tr);
1150	trace_gfs2_log_flush(sdp, 0, flags);
1151	return;
1152
1153out_withdraw:
1154	trans_drain(tr);
1155	/**
1156	 * If the tr_list is empty, we're withdrawing during a log
1157	 * flush that targets a transaction, but the transaction was
1158	 * never queued onto any of the ail lists. Here we add it to
1159	 * ail1 just so that ail_drain() will find and free it.
1160	 */
1161	spin_lock(&sdp->sd_ail_lock);
1162	if (tr && list_empty(&tr->tr_list))
1163		list_add(&tr->tr_list, &sdp->sd_ail1_list);
1164	spin_unlock(&sdp->sd_ail_lock);
1165	tr = NULL;
1166	goto out_end;
1167}
1168
1169/**
1170 * gfs2_merge_trans - Merge a new transaction into a cached transaction
1171 * @sdp: the filesystem
1172 * @new: New transaction to be merged
1173 */
1174
1175static void gfs2_merge_trans(struct gfs2_sbd *sdp, struct gfs2_trans *new)
1176{
1177	struct gfs2_trans *old = sdp->sd_log_tr;
1178
1179	WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags));
1180
1181	old->tr_num_buf_new	+= new->tr_num_buf_new;
1182	old->tr_num_databuf_new	+= new->tr_num_databuf_new;
1183	old->tr_num_buf_rm	+= new->tr_num_buf_rm;
1184	old->tr_num_databuf_rm	+= new->tr_num_databuf_rm;
1185	old->tr_revokes		+= new->tr_revokes;
1186	old->tr_num_revoke	+= new->tr_num_revoke;
1187
1188	list_splice_tail_init(&new->tr_databuf, &old->tr_databuf);
1189	list_splice_tail_init(&new->tr_buf, &old->tr_buf);
1190
1191	spin_lock(&sdp->sd_ail_lock);
1192	list_splice_tail_init(&new->tr_ail1_list, &old->tr_ail1_list);
1193	list_splice_tail_init(&new->tr_ail2_list, &old->tr_ail2_list);
1194	spin_unlock(&sdp->sd_ail_lock);
1195}
1196
1197static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1198{
1199	unsigned int reserved;
1200	unsigned int unused;
1201	unsigned int maxres;
1202
1203	gfs2_log_lock(sdp);
1204
1205	if (sdp->sd_log_tr) {
1206		gfs2_merge_trans(sdp, tr);
1207	} else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) {
1208		gfs2_assert_withdraw(sdp, !test_bit(TR_ONSTACK, &tr->tr_flags));
1209		sdp->sd_log_tr = tr;
1210		set_bit(TR_ATTACHED, &tr->tr_flags);
1211	}
1212
1213	reserved = calc_reserved(sdp);
1214	maxres = sdp->sd_log_blks_reserved + tr->tr_reserved;
1215	gfs2_assert_withdraw(sdp, maxres >= reserved);
1216	unused = maxres - reserved;
1217	if (unused)
1218		gfs2_log_release(sdp, unused);
1219	sdp->sd_log_blks_reserved = reserved;
1220
1221	gfs2_log_unlock(sdp);
1222}
1223
1224static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
1225{
1226	return atomic_read(&sdp->sd_log_pinned) +
1227	       atomic_read(&sdp->sd_log_blks_needed) >=
1228	       atomic_read(&sdp->sd_log_thresh1);
1229}
1230
1231static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
1232{
1233	return sdp->sd_jdesc->jd_blocks -
1234	       atomic_read(&sdp->sd_log_blks_free) +
1235	       atomic_read(&sdp->sd_log_blks_needed) >=
1236	       atomic_read(&sdp->sd_log_thresh2);
1237}
1238
1239/**
1240 * gfs2_log_commit - Commit a transaction to the log
1241 * @sdp: the filesystem
1242 * @tr: the transaction
1243 *
1244 * We wake up gfs2_logd if the number of pinned blocks exceed thresh1
1245 * or the total number of used blocks (pinned blocks plus AIL blocks)
1246 * is greater than thresh2.
1247 *
1248 * At mount time thresh1 is 2/5ths of journal size, thresh2 is 4/5ths of
1249 * journal size.
1250 *
1251 * Returns: errno
1252 */
1253
1254void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1255{
1256	log_refund(sdp, tr);
1257
1258	if (gfs2_ail_flush_reqd(sdp) || gfs2_jrnl_flush_reqd(sdp))
1259		wake_up(&sdp->sd_logd_waitq);
1260}
1261
1262/**
1263 * gfs2_log_shutdown - write a shutdown header into a journal
1264 * @sdp: the filesystem
1265 *
1266 */
1267
1268static void gfs2_log_shutdown(struct gfs2_sbd *sdp)
1269{
1270	gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
1271	gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
1272	gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
1273
1274	log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN);
1275	log_pull_tail(sdp);
1276
1277	gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
1278	gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
1279}
1280
1281/**
1282 * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
1283 * @data: Pointer to GFS2 superblock
1284 *
1285 * Also, periodically check to make sure that we're using the most recent
1286 * journal index.
1287 */
1288
1289int gfs2_logd(void *data)
1290{
1291	struct gfs2_sbd *sdp = data;
1292	unsigned long t = 1;
1293
1294	set_freezable();
1295	while (!kthread_should_stop()) {
1296		if (gfs2_withdrawn(sdp))
1297			break;
1298
1299		if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
1300			gfs2_ail1_empty(sdp, 0);
1301			gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1302						  GFS2_LFC_LOGD_JFLUSH_REQD);
1303		}
1304
1305		if (test_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags) ||
1306		    gfs2_ail_flush_reqd(sdp)) {
1307			clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags);
1308			gfs2_ail1_start(sdp);
1309			gfs2_ail1_wait(sdp);
1310			gfs2_ail1_empty(sdp, 0);
1311			gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1312						  GFS2_LFC_LOGD_AIL_FLUSH_REQD);
1313		}
1314
1315		t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
1316
1317		t = wait_event_freezable_timeout(sdp->sd_logd_waitq,
1318				test_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags) ||
1319				gfs2_ail_flush_reqd(sdp) ||
1320				gfs2_jrnl_flush_reqd(sdp) ||
1321				gfs2_withdrawn(sdp) ||
1322				kthread_should_stop(),
1323				t);
1324	}
1325
1326	return 0;
1327}
1328