fs/gfs2/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
  35 static void gfs2_log_shutdown(struct gfs2_sbd *sdp);
  36
  37 /**
  38  * gfs2_struct2blk - compute stuff
  39  * @sdp: the filesystem
  40  * @nstruct: the number of structures
  41  *
  42  * Compute the number of log descriptor blocks needed to hold a certain number
  43  * of structures of a certain size.
  44  *
  45  * Returns: the number of blocks needed (minimum is always 1)
  46  */
  47
  48 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct)
  49 {
  50         unsigned int blks;
  51         unsigned int first, second;
  52
  53         blks = 1;
  54         first = sdp->sd_ldptrs;
  55
  56         if (nstruct > first) {
  57                 second = sdp->sd_inptrs;
  58                 blks += DIV_ROUND_UP(nstruct - first, second);
  59         }
  60
  61         return blks;
  62 }
  63
  64 /**
  65  * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters
  66  * @mapping: The associated mapping (maybe NULL)
  67  * @bd: The gfs2_bufdata to remove
  68  *
  69  * The ail lock _must_ be held when calling this function
  70  *
  71  */
  72
  73 void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
  74 {
  75         bd->bd_tr = NULL;
  76         list_del_init(&bd->bd_ail_st_list);
  77         list_del_init(&bd->bd_ail_gl_list);
  78         atomic_dec(&bd->bd_gl->gl_ail_count);
  79         brelse(bd->bd_bh);
  80 }
  81
  82 /**
  83  * gfs2_ail1_start_one - Start I/O on a part of the AIL
  84  * @sdp: the filesystem
  85  * @wbc: The writeback control structure
  86  * @ai: The ail structure
  87  *
  88  */
  89
  90 static int gfs2_ail1_start_one(struct gfs2_sbd *sdp,
  91                                struct writeback_control *wbc,
  92                                struct gfs2_trans *tr)
  93 __releases(&sdp->sd_ail_lock)
  94 __acquires(&sdp->sd_ail_lock)
  95 {
  96         struct gfs2_glock *gl = NULL;
  97         struct address_space *mapping;
  98         struct gfs2_bufdata *bd, *s;
  99         struct buffer_head *bh;
 100         int ret = 0;
 101
 102         list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) {
 103                 bh = bd->bd_bh;
 104
 105                 gfs2_assert(sdp, bd->bd_tr == tr);
 106
 107                 if (!buffer_busy(bh)) {
 108                         if (buffer_uptodate(bh)) {
 109                                 list_move(&bd->bd_ail_st_list,
 110                                           &tr->tr_ail2_list);
 111                                 continue;
 112                         }
 113                         if (!cmpxchg(&sdp->sd_log_error, 0, -EIO)) {
 114                                 gfs2_io_error_bh(sdp, bh);
 115                                 gfs2_withdraw_delayed(sdp);
 116                         }
 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_page->mapping;
 130                 if (!mapping)
 131                         continue;
 132                 spin_unlock(&sdp->sd_ail_lock);
 133                 ret = generic_writepages(mapping, wbc);
 134                 spin_lock(&sdp->sd_ail_lock);
 135                 if (ret == -ENODATA) /* if a jdata write into a new hole */
 136                         ret = 0; /* ignore it */
 137                 if (ret || wbc->nr_to_write <= 0)
 138                         break;
 139                 return -EBUSY;
 140         }
 141
 142         return ret;
 143 }
 144
 145 static void dump_ail_list(struct gfs2_sbd *sdp)
 146 {
 147         struct gfs2_trans *tr;
 148         struct gfs2_bufdata *bd;
 149         struct buffer_head *bh;
 150
 151         list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
 152                 list_for_each_entry_reverse(bd, &tr->tr_ail1_list,
 153                                             bd_ail_st_list) {
 154                         bh = bd->bd_bh;
 155                         fs_err(sdp, "bd %p: blk:0x%llx bh=%p ", bd,
 156                                (unsigned long long)bd->bd_blkno, bh);
 157                         if (!bh) {
 158                                 fs_err(sdp, "\n");
 159                                 continue;
 160                         }
 161                         fs_err(sdp, "0x%llx up2:%d dirt:%d lkd:%d req:%d "
 162                                "map:%d new:%d ar:%d aw:%d delay:%d "
 163                                "io err:%d unwritten:%d dfr:%d pin:%d esc:%d\n",
 164                                (unsigned long long)bh->b_blocknr,
 165                                buffer_uptodate(bh), buffer_dirty(bh),
 166                                buffer_locked(bh), buffer_req(bh),
 167                                buffer_mapped(bh), buffer_new(bh),
 168                                buffer_async_read(bh), buffer_async_write(bh),
 169                                buffer_delay(bh), buffer_write_io_error(bh),
 170                                buffer_unwritten(bh),
 171                                buffer_defer_completion(bh),
 172                                buffer_pinned(bh), buffer_escaped(bh));
 173                 }
 174         }
 175 }
 176
 177 /**
 178  * gfs2_ail1_flush - start writeback of some ail1 entries
 179  * @sdp: The super block
 180  * @wbc: The writeback control structure
 181  *
 182  * Writes back some ail1 entries, according to the limits in the
 183  * writeback control structure
 184  */
 185
 186 void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc)
 187 {
 188         struct list_head *head = &sdp->sd_ail1_list;
 189         struct gfs2_trans *tr;
 190         struct blk_plug plug;
 191         int ret;
 192         unsigned long flush_start = jiffies;
 193
 194         trace_gfs2_ail_flush(sdp, wbc, 1);
 195         blk_start_plug(&plug);
 196         spin_lock(&sdp->sd_ail_lock);
 197 restart:
 198         ret = 0;
 199         if (time_after(jiffies, flush_start + (HZ * 600))) {
 200                 fs_err(sdp, "Error: In %s for ten minutes! t=%d\n",
 201                        __func__, current->journal_info ? 1 : 0);
 202                 dump_ail_list(sdp);
 203                 goto out;
 204         }
 205         list_for_each_entry_reverse(tr, head, tr_list) {
 206                 if (wbc->nr_to_write <= 0)
 207                         break;
 208                 ret = gfs2_ail1_start_one(sdp, wbc, tr);
 209                 if (ret) {
 210                         if (ret == -EBUSY)
 211                                 goto restart;
 212                         break;
 213                 }
 214         }
 215 out:
 216         spin_unlock(&sdp->sd_ail_lock);
 217         blk_finish_plug(&plug);
 218         if (ret) {
 219                 gfs2_lm(sdp, "gfs2_ail1_start_one (generic_writepages) "
 220                         "returned: %d\n", ret);
 221                 gfs2_withdraw(sdp);
 222         }
 223         trace_gfs2_ail_flush(sdp, wbc, 0);
 224 }
 225
 226 /**
 227  * gfs2_ail1_start - start writeback of all ail1 entries
 228  * @sdp: The superblock
 229  */
 230
 231 static void gfs2_ail1_start(struct gfs2_sbd *sdp)
 232 {
 233         struct writeback_control wbc = {
 234                 .sync_mode = WB_SYNC_NONE,
 235                 .nr_to_write = LONG_MAX,
 236                 .range_start = 0,
 237                 .range_end = LLONG_MAX,
 238         };
 239
 240         return gfs2_ail1_flush(sdp, &wbc);
 241 }
 242
 243 /**
 244  * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
 245  * @sdp: the filesystem
 246  * @tr: the transaction
 247  * @max_revokes: If nonzero, issue revokes for the bd items for written buffers
 248  *
 249  * returns: the transaction's count of remaining active items
 250  */
 251
 252 static int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
 253                                 int *max_revokes)
 254 {
 255         struct gfs2_bufdata *bd, *s;
 256         struct buffer_head *bh;
 257         int active_count = 0;
 258
 259         list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list,
 260                                          bd_ail_st_list) {
 261                 bh = bd->bd_bh;
 262                 gfs2_assert(sdp, bd->bd_tr == tr);
 263                 /*
 264                  * If another process flagged an io error, e.g. writing to the
 265                  * journal, error all other bhs and move them off the ail1 to
 266                  * prevent a tight loop when unmount tries to flush ail1,
 267                  * regardless of whether they're still busy. If no outside
 268                  * errors were found and the buffer is busy, move to the next.
 269                  * If the ail buffer is not busy and caught an error, flag it
 270                  * for others.
 271                  */
 272                 if (!sdp->sd_log_error && buffer_busy(bh)) {
 273                         active_count++;
 274                         continue;
 275                 }
 276                 if (!buffer_uptodate(bh) &&
 277                     !cmpxchg(&sdp->sd_log_error, 0, -EIO)) {
 278                         gfs2_io_error_bh(sdp, bh);
 279                         gfs2_withdraw_delayed(sdp);
 280                 }
 281                 /*
 282                  * If we have space for revokes and the bd is no longer on any
 283                  * buf list, we can just add a revoke for it immediately and
 284                  * avoid having to put it on the ail2 list, where it would need
 285                  * to be revoked later.
 286                  */
 287                 if (*max_revokes && list_empty(&bd->bd_list)) {
 288                         gfs2_add_revoke(sdp, bd);
 289                         (*max_revokes)--;
 290                         continue;
 291                 }
 292                 list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
 293         }
 294         return active_count;
 295 }
 296
 297 /**
 298  * gfs2_ail1_empty - Try to empty the ail1 lists
 299  * @sdp: The superblock
 300  * @max_revokes: If non-zero, add revokes where appropriate
 301  *
 302  * Tries to empty the ail1 lists, starting with the oldest first
 303  */
 304
 305 static int gfs2_ail1_empty(struct gfs2_sbd *sdp, int max_revokes)
 306 {
 307         struct gfs2_trans *tr, *s;
 308         int oldest_tr = 1;
 309         int ret;
 310
 311         spin_lock(&sdp->sd_ail_lock);
 312         list_for_each_entry_safe_reverse(tr, s, &sdp->sd_ail1_list, tr_list) {
 313                 if (!gfs2_ail1_empty_one(sdp, tr, &max_revokes) && oldest_tr)
 314                         list_move(&tr->tr_list, &sdp->sd_ail2_list);
 315                 else
 316                         oldest_tr = 0;
 317         }
 318         ret = list_empty(&sdp->sd_ail1_list);
 319         spin_unlock(&sdp->sd_ail_lock);
 320
 321         if (test_bit(SDF_WITHDRAWING, &sdp->sd_flags)) {
 322                 gfs2_lm(sdp, "fatal: I/O error(s)\n");
 323                 gfs2_withdraw(sdp);
 324         }
 325
 326         return ret;
 327 }
 328
 329 static void gfs2_ail1_wait(struct gfs2_sbd *sdp)
 330 {
 331         struct gfs2_trans *tr;
 332         struct gfs2_bufdata *bd;
 333         struct buffer_head *bh;
 334
 335         spin_lock(&sdp->sd_ail_lock);
 336         list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
 337                 list_for_each_entry(bd, &tr->tr_ail1_list, bd_ail_st_list) {
 338                         bh = bd->bd_bh;
 339                         if (!buffer_locked(bh))
 340                                 continue;
 341                         get_bh(bh);
 342                         spin_unlock(&sdp->sd_ail_lock);
 343                         wait_on_buffer(bh);
 344                         brelse(bh);
 345                         return;
 346                 }
 347         }
 348         spin_unlock(&sdp->sd_ail_lock);
 349 }
 350
 351 /**
 352  * gfs2_ail_empty_tr - empty one of the ail lists for a transaction
 353  */
 354
 355 static void gfs2_ail_empty_tr(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
 356                               struct list_head *head)
 357 {
 358         struct gfs2_bufdata *bd;
 359
 360         while (!list_empty(head)) {
 361                 bd = list_first_entry(head, struct gfs2_bufdata,
 362                                       bd_ail_st_list);
 363                 gfs2_assert(sdp, bd->bd_tr == tr);
 364                 gfs2_remove_from_ail(bd);
 365         }
 366 }
 367
 368 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
 369 {
 370         struct gfs2_trans *tr, *safe;
 371         unsigned int old_tail = sdp->sd_log_tail;
 372         int wrap = (new_tail < old_tail);
 373         int a, b, rm;
 374
 375         spin_lock(&sdp->sd_ail_lock);
 376
 377         list_for_each_entry_safe(tr, safe, &sdp->sd_ail2_list, tr_list) {
 378                 a = (old_tail <= tr->tr_first);
 379                 b = (tr->tr_first < new_tail);
 380                 rm = (wrap) ? (a || b) : (a && b);
 381                 if (!rm)
 382                         continue;
 383
 384                 gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
 385                 list_del(&tr->tr_list);
 386                 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list));
 387                 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list));
 388                 gfs2_trans_free(sdp, tr);
 389         }
 390
 391         spin_unlock(&sdp->sd_ail_lock);
 392 }
 393
 394 /**
 395  * gfs2_log_release - Release a given number of log blocks
 396  * @sdp: The GFS2 superblock
 397  * @blks: The number of blocks
 398  *
 399  */
 400
 401 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
 402 {
 403
 404         atomic_add(blks, &sdp->sd_log_blks_free);
 405         trace_gfs2_log_blocks(sdp, blks);
 406         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
 407                                   sdp->sd_jdesc->jd_blocks);
 408         up_read(&sdp->sd_log_flush_lock);
 409 }
 410
 411 /**
 412  * gfs2_log_reserve - Make a log reservation
 413  * @sdp: The GFS2 superblock
 414  * @blks: The number of blocks to reserve
 415  *
 416  * Note that we never give out the last few blocks of the journal. Thats
 417  * due to the fact that there is a small number of header blocks
 418  * associated with each log flush. The exact number can't be known until
 419  * flush time, so we ensure that we have just enough free blocks at all
 420  * times to avoid running out during a log flush.
 421  *
 422  * We no longer flush the log here, instead we wake up logd to do that
 423  * for us. To avoid the thundering herd and to ensure that we deal fairly
 424  * with queued waiters, we use an exclusive wait. This means that when we
 425  * get woken with enough journal space to get our reservation, we need to
 426  * wake the next waiter on the list.
 427  *
 428  * Returns: errno
 429  */
 430
 431 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
 432 {
 433         int ret = 0;
 434         unsigned reserved_blks = 7 * (4096 / sdp->sd_vfs->s_blocksize);
 435         unsigned wanted = blks + reserved_blks;
 436         DEFINE_WAIT(wait);
 437         int did_wait = 0;
 438         unsigned int free_blocks;
 439
 440         if (gfs2_assert_warn(sdp, blks) ||
 441             gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
 442                 return -EINVAL;
 443         atomic_add(blks, &sdp->sd_log_blks_needed);
 444 retry:
 445         free_blocks = atomic_read(&sdp->sd_log_blks_free);
 446         if (unlikely(free_blocks <= wanted)) {
 447                 do {
 448                         prepare_to_wait_exclusive(&sdp->sd_log_waitq, &wait,
 449                                         TASK_UNINTERRUPTIBLE);
 450                         wake_up(&sdp->sd_logd_waitq);
 451                         did_wait = 1;
 452                         if (atomic_read(&sdp->sd_log_blks_free) <= wanted)
 453                                 io_schedule();
 454                         free_blocks = atomic_read(&sdp->sd_log_blks_free);
 455                 } while(free_blocks <= wanted);
 456                 finish_wait(&sdp->sd_log_waitq, &wait);
 457         }
 458         atomic_inc(&sdp->sd_reserving_log);
 459         if (atomic_cmpxchg(&sdp->sd_log_blks_free, free_blocks,
 460                                 free_blocks - blks) != free_blocks) {
 461                 if (atomic_dec_and_test(&sdp->sd_reserving_log))
 462                         wake_up(&sdp->sd_reserving_log_wait);
 463                 goto retry;
 464         }
 465         atomic_sub(blks, &sdp->sd_log_blks_needed);
 466         trace_gfs2_log_blocks(sdp, -blks);
 467
 468         /*
 469          * If we waited, then so might others, wake them up _after_ we get
 470          * our share of the log.
 471          */
 472         if (unlikely(did_wait))
 473                 wake_up(&sdp->sd_log_waitq);
 474
 475         down_read(&sdp->sd_log_flush_lock);
 476         if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) {
 477                 gfs2_log_release(sdp, blks);
 478                 ret = -EROFS;
 479         }
 480         if (atomic_dec_and_test(&sdp->sd_reserving_log))
 481                 wake_up(&sdp->sd_reserving_log_wait);
 482         return ret;
 483 }
 484
 485 /**
 486  * log_distance - Compute distance between two journal blocks
 487  * @sdp: The GFS2 superblock
 488  * @newer: The most recent journal block of the pair
 489  * @older: The older journal block of the pair
 490  *
 491  *   Compute the distance (in the journal direction) between two
 492  *   blocks in the journal
 493  *
 494  * Returns: the distance in blocks
 495  */
 496
 497 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
 498                                         unsigned int older)
 499 {
 500         int dist;
 501
 502         dist = newer - older;
 503         if (dist < 0)
 504                 dist += sdp->sd_jdesc->jd_blocks;
 505
 506         return dist;
 507 }
 508
 509 /**
 510  * calc_reserved - Calculate the number of blocks to reserve when
 511  *                 refunding a transaction's unused buffers.
 512  * @sdp: The GFS2 superblock
 513  *
 514  * This is complex.  We need to reserve room for all our currently used
 515  * metadata buffers (e.g. normal file I/O rewriting file time stamps) and
 516  * all our journaled data buffers for journaled files (e.g. files in the
 517  * meta_fs like rindex, or files for which chattr +j was done.)
 518  * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush
 519  * will count it as free space (sd_log_blks_free) and corruption will follow.
 520  *
 521  * We can have metadata bufs and jdata bufs in the same journal.  So each
 522  * type gets its own log header, for which we need to reserve a block.
 523  * In fact, each type has the potential for needing more than one header
 524  * in cases where we have more buffers than will fit on a journal page.
 525  * Metadata journal entries take up half the space of journaled buffer entries.
 526  * Thus, metadata entries have buf_limit (502) and journaled buffers have
 527  * databuf_limit (251) before they cause a wrap around.
 528  *
 529  * Also, we need to reserve blocks for revoke journal entries and one for an
 530  * overall header for the lot.
 531  *
 532  * Returns: the number of blocks reserved
 533  */
 534 static unsigned int calc_reserved(struct gfs2_sbd *sdp)
 535 {
 536         unsigned int reserved = 0;
 537         unsigned int mbuf;
 538         unsigned int dbuf;
 539         struct gfs2_trans *tr = sdp->sd_log_tr;
 540
 541         if (tr) {
 542                 mbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
 543                 dbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
 544                 reserved = mbuf + dbuf;
 545                 /* Account for header blocks */
 546                 reserved += DIV_ROUND_UP(mbuf, buf_limit(sdp));
 547                 reserved += DIV_ROUND_UP(dbuf, databuf_limit(sdp));
 548         }
 549
 550         if (sdp->sd_log_committed_revoke > 0)
 551                 reserved += gfs2_struct2blk(sdp, sdp->sd_log_committed_revoke);
 552         /* One for the overall header */
 553         if (reserved)
 554                 reserved++;
 555         return reserved;
 556 }
 557
 558 static unsigned int current_tail(struct gfs2_sbd *sdp)
 559 {
 560         struct gfs2_trans *tr;
 561         unsigned int tail;
 562
 563         spin_lock(&sdp->sd_ail_lock);
 564
 565         if (list_empty(&sdp->sd_ail1_list)) {
 566                 tail = sdp->sd_log_head;
 567         } else {
 568                 tr = list_last_entry(&sdp->sd_ail1_list, struct gfs2_trans,
 569                                 tr_list);
 570                 tail = tr->tr_first;
 571         }
 572
 573         spin_unlock(&sdp->sd_ail_lock);
 574
 575         return tail;
 576 }
 577
 578 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
 579 {
 580         unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
 581
 582         ail2_empty(sdp, new_tail);
 583
 584         atomic_add(dist, &sdp->sd_log_blks_free);
 585         trace_gfs2_log_blocks(sdp, dist);
 586         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
 587                              sdp->sd_jdesc->jd_blocks);
 588
 589         sdp->sd_log_tail = new_tail;
 590 }
 591
 592
 593 void log_flush_wait(struct gfs2_sbd *sdp)
 594 {
 595         DEFINE_WAIT(wait);
 596
 597         if (atomic_read(&sdp->sd_log_in_flight)) {
 598                 do {
 599                         prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
 600                                         TASK_UNINTERRUPTIBLE);
 601                         if (atomic_read(&sdp->sd_log_in_flight))
 602                                 io_schedule();
 603                 } while(atomic_read(&sdp->sd_log_in_flight));
 604                 finish_wait(&sdp->sd_log_flush_wait, &wait);
 605         }
 606 }
 607
 608 static int ip_cmp(void *priv, const struct list_head *a, const struct list_head *b)
 609 {
 610         struct gfs2_inode *ipa, *ipb;
 611
 612         ipa = list_entry(a, struct gfs2_inode, i_ordered);
 613         ipb = list_entry(b, struct gfs2_inode, i_ordered);
 614
 615         if (ipa->i_no_addr < ipb->i_no_addr)
 616                 return -1;
 617         if (ipa->i_no_addr > ipb->i_no_addr)
 618                 return 1;
 619         return 0;
 620 }
 621
 622 static void __ordered_del_inode(struct gfs2_inode *ip)
 623 {
 624         if (!list_empty(&ip->i_ordered))
 625                 list_del_init(&ip->i_ordered);
 626 }
 627
 628 static void gfs2_ordered_write(struct gfs2_sbd *sdp)
 629 {
 630         struct gfs2_inode *ip;
 631         LIST_HEAD(written);
 632
 633         spin_lock(&sdp->sd_ordered_lock);
 634         list_sort(NULL, &sdp->sd_log_ordered, &ip_cmp);
 635         while (!list_empty(&sdp->sd_log_ordered)) {
 636                 ip = list_first_entry(&sdp->sd_log_ordered, struct gfs2_inode, i_ordered);
 637                 if (ip->i_inode.i_mapping->nrpages == 0) {
 638                         __ordered_del_inode(ip);
 639                         continue;
 640                 }
 641                 list_move(&ip->i_ordered, &written);
 642                 spin_unlock(&sdp->sd_ordered_lock);
 643                 filemap_fdatawrite(ip->i_inode.i_mapping);
 644                 spin_lock(&sdp->sd_ordered_lock);
 645         }
 646         list_splice(&written, &sdp->sd_log_ordered);
 647         spin_unlock(&sdp->sd_ordered_lock);
 648 }
 649
 650 static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
 651 {
 652         struct gfs2_inode *ip;
 653
 654         spin_lock(&sdp->sd_ordered_lock);
 655         while (!list_empty(&sdp->sd_log_ordered)) {
 656                 ip = list_first_entry(&sdp->sd_log_ordered, struct gfs2_inode, i_ordered);
 657                 __ordered_del_inode(ip);
 658                 if (ip->i_inode.i_mapping->nrpages == 0)
 659                         continue;
 660                 spin_unlock(&sdp->sd_ordered_lock);
 661                 filemap_fdatawait(ip->i_inode.i_mapping);
 662                 spin_lock(&sdp->sd_ordered_lock);
 663         }
 664         spin_unlock(&sdp->sd_ordered_lock);
 665 }
 666
 667 void gfs2_ordered_del_inode(struct gfs2_inode *ip)
 668 {
 669         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 670
 671         spin_lock(&sdp->sd_ordered_lock);
 672         __ordered_del_inode(ip);
 673         spin_unlock(&sdp->sd_ordered_lock);
 674 }
 675
 676 void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
 677 {
 678         struct buffer_head *bh = bd->bd_bh;
 679         struct gfs2_glock *gl = bd->bd_gl;
 680
 681         sdp->sd_log_num_revoke++;
 682         if (atomic_inc_return(&gl->gl_revokes) == 1)
 683                 gfs2_glock_hold(gl);
 684         bh->b_private = NULL;
 685         bd->bd_blkno = bh->b_blocknr;
 686         gfs2_remove_from_ail(bd); /* drops ref on bh */
 687         bd->bd_bh = NULL;
 688         set_bit(GLF_LFLUSH, &gl->gl_flags);
 689         list_add(&bd->bd_list, &sdp->sd_log_revokes);
 690 }
 691
 692 void gfs2_glock_remove_revoke(struct gfs2_glock *gl)
 693 {
 694         if (atomic_dec_return(&gl->gl_revokes) == 0) {
 695                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
 696                 gfs2_glock_queue_put(gl);
 697         }
 698 }
 699
 700 /**
 701  * gfs2_write_revokes - Add as many revokes to the system transaction as we can
 702  * @sdp: The GFS2 superblock
 703  *
 704  * Our usual strategy is to defer writing revokes as much as we can in the hope
 705  * that we'll eventually overwrite the journal, which will make those revokes
 706  * go away.  This changes when we flush the log: at that point, there will
 707  * likely be some left-over space in the last revoke block of that transaction.
 708  * We can fill that space with additional revokes for blocks that have already
 709  * been written back.  This will basically come at no cost now, and will save
 710  * us from having to keep track of those blocks on the AIL2 list later.
 711  */
 712 void gfs2_write_revokes(struct gfs2_sbd *sdp)
 713 {
 714         /* number of revokes we still have room for */
 715         int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
 716
 717         gfs2_log_lock(sdp);
 718         while (sdp->sd_log_num_revoke > max_revokes)
 719                 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
 720         max_revokes -= sdp->sd_log_num_revoke;
 721         if (!sdp->sd_log_num_revoke) {
 722                 atomic_dec(&sdp->sd_log_blks_free);
 723                 /* If no blocks have been reserved, we need to also
 724                  * reserve a block for the header */
 725                 if (!sdp->sd_log_blks_reserved) {
 726                         atomic_dec(&sdp->sd_log_blks_free);
 727                         trace_gfs2_log_blocks(sdp, -2);
 728                 } else {
 729                         trace_gfs2_log_blocks(sdp, -1);
 730                 }
 731         }
 732         gfs2_ail1_empty(sdp, max_revokes);
 733         gfs2_log_unlock(sdp);
 734
 735         if (!sdp->sd_log_num_revoke) {
 736                 atomic_inc(&sdp->sd_log_blks_free);
 737                 if (!sdp->sd_log_blks_reserved) {
 738                         atomic_inc(&sdp->sd_log_blks_free);
 739                         trace_gfs2_log_blocks(sdp, 2);
 740                 } else {
 741                         trace_gfs2_log_blocks(sdp, 1);
 742                 }
 743         }
 744 }
 745
 746 /**
 747  * gfs2_write_log_header - Write a journal log header buffer at lblock
 748  * @sdp: The GFS2 superblock
 749  * @jd: journal descriptor of the journal to which we are writing
 750  * @seq: sequence number
 751  * @tail: tail of the log
 752  * @lblock: value for lh_blkno (block number relative to start of journal)
 753  * @flags: log header flags GFS2_LOG_HEAD_*
 754  * @op_flags: flags to pass to the bio
 755  *
 756  * Returns: the initialized log buffer descriptor
 757  */
 758
 759 void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
 760                            u64 seq, u32 tail, u32 lblock, u32 flags,
 761                            int op_flags)
 762 {
 763         struct gfs2_log_header *lh;
 764         u32 hash, crc;
 765         struct page *page;
 766         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 767         struct timespec64 tv;
 768         struct super_block *sb = sdp->sd_vfs;
 769         u64 dblock;
 770
 771         if (gfs2_withdrawn(sdp))
 772                 goto out;
 773
 774         page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 775         lh = page_address(page);
 776         clear_page(lh);
 777
 778         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 779         lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
 780         lh->lh_header.__pad0 = cpu_to_be64(0);
 781         lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
 782         lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
 783         lh->lh_sequence = cpu_to_be64(seq);
 784         lh->lh_flags = cpu_to_be32(flags);
 785         lh->lh_tail = cpu_to_be32(tail);
 786         lh->lh_blkno = cpu_to_be32(lblock);
 787         hash = ~crc32(~0, lh, LH_V1_SIZE);
 788         lh->lh_hash = cpu_to_be32(hash);
 789
 790         ktime_get_coarse_real_ts64(&tv);
 791         lh->lh_nsec = cpu_to_be32(tv.tv_nsec);
 792         lh->lh_sec = cpu_to_be64(tv.tv_sec);
 793         if (!list_empty(&jd->extent_list))
 794                 dblock = gfs2_log_bmap(jd, lblock);
 795         else {
 796                 int ret = gfs2_lblk_to_dblk(jd->jd_inode, lblock, &dblock);
 797                 if (gfs2_assert_withdraw(sdp, ret == 0))
 798                         return;
 799         }
 800         lh->lh_addr = cpu_to_be64(dblock);
 801         lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr);
 802
 803         /* We may only write local statfs, quota, etc., when writing to our
 804            own journal. The values are left 0 when recovering a journal
 805            different from our own. */
 806         if (!(flags & GFS2_LOG_HEAD_RECOVERY)) {
 807                 lh->lh_statfs_addr =
 808                         cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr);
 809                 lh->lh_quota_addr =
 810                         cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr);
 811
 812                 spin_lock(&sdp->sd_statfs_spin);
 813                 lh->lh_local_total = cpu_to_be64(l_sc->sc_total);
 814                 lh->lh_local_free = cpu_to_be64(l_sc->sc_free);
 815                 lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes);
 816                 spin_unlock(&sdp->sd_statfs_spin);
 817         }
 818
 819         BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE);
 820
 821         crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
 822                      sb->s_blocksize - LH_V1_SIZE - 4);
 823         lh->lh_crc = cpu_to_be32(crc);
 824
 825         gfs2_log_write(sdp, page, sb->s_blocksize, 0, dblock);
 826         gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE | op_flags);
 827 out:
 828         log_flush_wait(sdp);
 829 }
 830
 831 /**
 832  * log_write_header - Get and initialize a journal header buffer
 833  * @sdp: The GFS2 superblock
 834  * @flags: The log header flags, including log header origin
 835  *
 836  * Returns: the initialized log buffer descriptor
 837  */
 838
 839 static void log_write_header(struct gfs2_sbd *sdp, u32 flags)
 840 {
 841         unsigned int tail;
 842         int op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC;
 843         enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
 844
 845         gfs2_assert_withdraw(sdp, (state != SFS_FROZEN));
 846         tail = current_tail(sdp);
 847
 848         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) {
 849                 gfs2_ordered_wait(sdp);
 850                 log_flush_wait(sdp);
 851                 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
 852         }
 853         sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
 854         gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++, tail,
 855                               sdp->sd_log_flush_head, flags, op_flags);
 856         gfs2_log_incr_head(sdp);
 857
 858         if (sdp->sd_log_tail != tail)
 859                 log_pull_tail(sdp, tail);
 860 }
 861
 862 /**
 863  * ail_drain - drain the ail lists after a withdraw
 864  * @sdp: Pointer to GFS2 superblock
 865  */
 866 static void ail_drain(struct gfs2_sbd *sdp)
 867 {
 868         struct gfs2_trans *tr;
 869
 870         spin_lock(&sdp->sd_ail_lock);
 871         /*
 872          * For transactions on the sd_ail1_list we need to drain both the
 873          * ail1 and ail2 lists. That's because function gfs2_ail1_start_one
 874          * (temporarily) moves items from its tr_ail1 list to tr_ail2 list
 875          * before revokes are sent for that block. Items on the sd_ail2_list
 876          * should have already gotten beyond that point, so no need.
 877          */
 878         while (!list_empty(&sdp->sd_ail1_list)) {
 879                 tr = list_first_entry(&sdp->sd_ail1_list, struct gfs2_trans,
 880                                       tr_list);
 881                 gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail1_list);
 882                 gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
 883                 list_del(&tr->tr_list);
 884                 gfs2_trans_free(sdp, tr);
 885         }
 886         while (!list_empty(&sdp->sd_ail2_list)) {
 887                 tr = list_first_entry(&sdp->sd_ail2_list, struct gfs2_trans,
 888                                       tr_list);
 889                 gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
 890                 list_del(&tr->tr_list);
 891                 gfs2_trans_free(sdp, tr);
 892         }
 893         spin_unlock(&sdp->sd_ail_lock);
 894 }
 895
 896 /**
 897  * empty_ail1_list - try to start IO and empty the ail1 list
 898  * @sdp: Pointer to GFS2 superblock
 899  */
 900 static void empty_ail1_list(struct gfs2_sbd *sdp)
 901 {
 902         unsigned long start = jiffies;
 903
 904         for (;;) {
 905                 if (time_after(jiffies, start + (HZ * 600))) {
 906                         fs_err(sdp, "Error: In %s for 10 minutes! t=%d\n",
 907                                __func__, current->journal_info ? 1 : 0);
 908                         dump_ail_list(sdp);
 909                         return;
 910                 }
 911                 gfs2_ail1_start(sdp);
 912                 gfs2_ail1_wait(sdp);
 913                 if (gfs2_ail1_empty(sdp, 0))
 914                         return;
 915         }
 916 }
 917
 918 /**
 919  * trans_drain - drain the buf and databuf queue for a failed transaction
 920  * @tr: the transaction to drain
 921  *
 922  * When this is called, we're taking an error exit for a log write that failed
 923  * but since we bypassed the after_commit functions, we need to remove the
 924  * items from the buf and databuf queue.
 925  */
 926 static void trans_drain(struct gfs2_trans *tr)
 927 {
 928         struct gfs2_bufdata *bd;
 929         struct list_head *head;
 930
 931         if (!tr)
 932                 return;
 933
 934         head = &tr->tr_buf;
 935         while (!list_empty(head)) {
 936                 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
 937                 list_del_init(&bd->bd_list);
 938                 if (!list_empty(&bd->bd_ail_st_list))
 939                         gfs2_remove_from_ail(bd);
 940                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 941         }
 942         head = &tr->tr_databuf;
 943         while (!list_empty(head)) {
 944                 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
 945                 list_del_init(&bd->bd_list);
 946                 if (!list_empty(&bd->bd_ail_st_list))
 947                         gfs2_remove_from_ail(bd);
 948                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 949         }
 950 }
 951
 952 /**
 953  * gfs2_log_flush - flush incore transaction(s)
 954  * @sdp: the filesystem
 955  * @gl: The glock structure to flush.  If NULL, flush the whole incore log
 956  * @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags
 957  *
 958  */
 959
 960 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags)
 961 {
 962         struct gfs2_trans *tr = NULL;
 963         enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
 964
 965         down_write(&sdp->sd_log_flush_lock);
 966
 967         /*
 968          * Do this check while holding the log_flush_lock to prevent new
 969          * buffers from being added to the ail via gfs2_pin()
 970          */
 971         if (gfs2_withdrawn(sdp))
 972                 goto out;
 973
 974         /* Log might have been flushed while we waited for the flush lock */
 975         if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags))
 976                 goto out;
 977         trace_gfs2_log_flush(sdp, 1, flags);
 978
 979         if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN)
 980                 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 981
 982         sdp->sd_log_flush_head = sdp->sd_log_head;
 983         tr = sdp->sd_log_tr;
 984         if (tr) {
 985                 sdp->sd_log_tr = NULL;
 986                 tr->tr_first = sdp->sd_log_flush_head;
 987                 if (unlikely (state == SFS_FROZEN))
 988                         if (gfs2_assert_withdraw_delayed(sdp,
 989                                !tr->tr_num_buf_new && !tr->tr_num_databuf_new))
 990                                 goto out_withdraw;
 991         }
 992
 993         if (unlikely(state == SFS_FROZEN))
 994                 if (gfs2_assert_withdraw_delayed(sdp, !sdp->sd_log_num_revoke))
 995                         goto out_withdraw;
 996         if (gfs2_assert_withdraw_delayed(sdp,
 997                         sdp->sd_log_num_revoke == sdp->sd_log_committed_revoke))
 998                 goto out_withdraw;
 999
1000         gfs2_ordered_write(sdp);
1001         if (gfs2_withdrawn(sdp))
1002                 goto out_withdraw;
1003         lops_before_commit(sdp, tr);
1004         if (gfs2_withdrawn(sdp))
1005                 goto out_withdraw;
1006         gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE);
1007         if (gfs2_withdrawn(sdp))
1008                 goto out_withdraw;
1009
1010         if (sdp->sd_log_head != sdp->sd_log_flush_head) {
1011                 log_flush_wait(sdp);
1012                 log_write_header(sdp, flags);
1013         } else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
1014                 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
1015                 trace_gfs2_log_blocks(sdp, -1);
1016                 log_write_header(sdp, flags);
1017         }
1018         if (gfs2_withdrawn(sdp))
1019                 goto out_withdraw;
1020         lops_after_commit(sdp, tr);
1021
1022         gfs2_log_lock(sdp);
1023         sdp->sd_log_head = sdp->sd_log_flush_head;
1024         sdp->sd_log_blks_reserved = 0;
1025         sdp->sd_log_committed_revoke = 0;
1026
1027         spin_lock(&sdp->sd_ail_lock);
1028         if (tr && !list_empty(&tr->tr_ail1_list)) {
1029                 list_add(&tr->tr_list, &sdp->sd_ail1_list);
1030                 tr = NULL;
1031         }
1032         spin_unlock(&sdp->sd_ail_lock);
1033         gfs2_log_unlock(sdp);
1034
1035         if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) {
1036                 if (!sdp->sd_log_idle) {
1037                         empty_ail1_list(sdp);
1038                         if (gfs2_withdrawn(sdp))
1039                                 goto out_withdraw;
1040                         atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
1041                         trace_gfs2_log_blocks(sdp, -1);
1042                         log_write_header(sdp, flags);
1043                         sdp->sd_log_head = sdp->sd_log_flush_head;
1044                 }
1045                 if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
1046                              GFS2_LOG_HEAD_FLUSH_FREEZE))
1047                         gfs2_log_shutdown(sdp);
1048                 if (flags & GFS2_LOG_HEAD_FLUSH_FREEZE)
1049                         atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
1050         }
1051
1052 out_end:
1053         trace_gfs2_log_flush(sdp, 0, flags);
1054 out:
1055         up_write(&sdp->sd_log_flush_lock);
1056         gfs2_trans_free(sdp, tr);
1057         if (gfs2_withdrawing(sdp))
1058                 gfs2_withdraw(sdp);
1059         return;
1060
1061 out_withdraw:
1062         trans_drain(tr);
1063         /**
1064          * If the tr_list is empty, we're withdrawing during a log
1065          * flush that targets a transaction, but the transaction was
1066          * never queued onto any of the ail lists. Here we add it to
1067          * ail1 just so that ail_drain() will find and free it.
1068          */
1069         spin_lock(&sdp->sd_ail_lock);
1070         if (tr && list_empty(&tr->tr_list))
1071                 list_add(&tr->tr_list, &sdp->sd_ail1_list);
1072         spin_unlock(&sdp->sd_ail_lock);
1073         ail_drain(sdp); /* frees all transactions */
1074         tr = NULL;
1075         goto out_end;
1076 }
1077
1078 /**
1079  * gfs2_merge_trans - Merge a new transaction into a cached transaction
1080  * @old: Original transaction to be expanded
1081  * @new: New transaction to be merged
1082  */
1083
1084 static void gfs2_merge_trans(struct gfs2_sbd *sdp, struct gfs2_trans *new)
1085 {
1086         struct gfs2_trans *old = sdp->sd_log_tr;
1087
1088         WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags));
1089
1090         old->tr_num_buf_new     += new->tr_num_buf_new;
1091         old->tr_num_databuf_new += new->tr_num_databuf_new;
1092         old->tr_num_buf_rm      += new->tr_num_buf_rm;
1093         old->tr_num_databuf_rm  += new->tr_num_databuf_rm;
1094         old->tr_num_revoke      += new->tr_num_revoke;
1095         old->tr_num_revoke_rm   += new->tr_num_revoke_rm;
1096
1097         list_splice_tail_init(&new->tr_databuf, &old->tr_databuf);
1098         list_splice_tail_init(&new->tr_buf, &old->tr_buf);
1099
1100         spin_lock(&sdp->sd_ail_lock);
1101         list_splice_tail_init(&new->tr_ail1_list, &old->tr_ail1_list);
1102         list_splice_tail_init(&new->tr_ail2_list, &old->tr_ail2_list);
1103         spin_unlock(&sdp->sd_ail_lock);
1104 }
1105
1106 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1107 {
1108         unsigned int reserved;
1109         unsigned int unused;
1110         unsigned int maxres;
1111
1112         gfs2_log_lock(sdp);
1113
1114         if (sdp->sd_log_tr) {
1115                 gfs2_merge_trans(sdp, tr);
1116         } else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) {
1117                 gfs2_assert_withdraw(sdp, test_bit(TR_ALLOCED, &tr->tr_flags));
1118                 sdp->sd_log_tr = tr;
1119                 set_bit(TR_ATTACHED, &tr->tr_flags);
1120         }
1121
1122         sdp->sd_log_committed_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
1123         reserved = calc_reserved(sdp);
1124         maxres = sdp->sd_log_blks_reserved + tr->tr_reserved;
1125         gfs2_assert_withdraw(sdp, maxres >= reserved);
1126         unused = maxres - reserved;
1127         atomic_add(unused, &sdp->sd_log_blks_free);
1128         trace_gfs2_log_blocks(sdp, unused);
1129         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
1130                              sdp->sd_jdesc->jd_blocks);
1131         sdp->sd_log_blks_reserved = reserved;
1132
1133         gfs2_log_unlock(sdp);
1134 }
1135
1136 /**
1137  * gfs2_log_commit - Commit a transaction to the log
1138  * @sdp: the filesystem
1139  * @tr: the transaction
1140  *
1141  * We wake up gfs2_logd if the number of pinned blocks exceed thresh1
1142  * or the total number of used blocks (pinned blocks plus AIL blocks)
1143  * is greater than thresh2.
1144  *
1145  * At mount time thresh1 is 2/5ths of journal size, thresh2 is 4/5ths of
1146  * journal size.
1147  *
1148  * Returns: errno
1149  */
1150
1151 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1152 {
1153         log_refund(sdp, tr);
1154
1155         if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) ||
1156             ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) >
1157             atomic_read(&sdp->sd_log_thresh2)))
1158                 wake_up(&sdp->sd_logd_waitq);
1159 }
1160
1161 /**
1162  * gfs2_log_shutdown - write a shutdown header into a journal
1163  * @sdp: the filesystem
1164  *
1165  */
1166
1167 static void gfs2_log_shutdown(struct gfs2_sbd *sdp)
1168 {
1169         gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
1170         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
1171         gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
1172
1173         sdp->sd_log_flush_head = sdp->sd_log_head;
1174
1175         log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN);
1176
1177         gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
1178         gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
1179
1180         sdp->sd_log_head = sdp->sd_log_flush_head;
1181         sdp->sd_log_tail = sdp->sd_log_head;
1182 }
1183
1184 static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
1185 {
1186         return (atomic_read(&sdp->sd_log_pinned) +
1187                 atomic_read(&sdp->sd_log_blks_needed) >=
1188                 atomic_read(&sdp->sd_log_thresh1));
1189 }
1190
1191 static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
1192 {
1193         unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free);
1194
1195         if (test_and_clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags))
1196                 return 1;
1197
1198         return used_blocks + atomic_read(&sdp->sd_log_blks_needed) >=
1199                 atomic_read(&sdp->sd_log_thresh2);
1200 }
1201
1202 /**
1203  * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
1204  * @sdp: Pointer to GFS2 superblock
1205  *
1206  * Also, periodically check to make sure that we're using the most recent
1207  * journal index.
1208  */
1209
1210 int gfs2_logd(void *data)
1211 {
1212         struct gfs2_sbd *sdp = data;
1213         unsigned long t = 1;
1214         DEFINE_WAIT(wait);
1215         bool did_flush;
1216
1217         while (!kthread_should_stop()) {
1218
1219                 if (gfs2_withdrawn(sdp)) {
1220                         msleep_interruptible(HZ);
1221                         continue;
1222                 }
1223                 /* Check for errors writing to the journal */
1224                 if (sdp->sd_log_error) {
1225                         gfs2_lm(sdp,
1226                                 "GFS2: fsid=%s: error %d: "
1227                                 "withdrawing the file system to "
1228                                 "prevent further damage.\n",
1229                                 sdp->sd_fsname, sdp->sd_log_error);
1230                         gfs2_withdraw(sdp);
1231                         continue;
1232                 }
1233
1234                 did_flush = false;
1235                 if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
1236                         gfs2_ail1_empty(sdp, 0);
1237                         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1238                                        GFS2_LFC_LOGD_JFLUSH_REQD);
1239                         did_flush = true;
1240                 }
1241
1242                 if (gfs2_ail_flush_reqd(sdp)) {
1243                         gfs2_ail1_start(sdp);
1244                         gfs2_ail1_wait(sdp);
1245                         gfs2_ail1_empty(sdp, 0);
1246                         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1247                                        GFS2_LFC_LOGD_AIL_FLUSH_REQD);
1248                         did_flush = true;
1249                 }
1250
1251                 if (!gfs2_ail_flush_reqd(sdp) || did_flush)
1252                         wake_up(&sdp->sd_log_waitq);
1253
1254                 t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
1255
1256                 try_to_freeze();
1257
1258                 do {
1259                         prepare_to_wait(&sdp->sd_logd_waitq, &wait,
1260                                         TASK_INTERRUPTIBLE);
1261                         if (!gfs2_ail_flush_reqd(sdp) &&
1262                             !gfs2_jrnl_flush_reqd(sdp) &&
1263                             !kthread_should_stop())
1264                                 t = schedule_timeout(t);
1265                 } while(t && !gfs2_ail_flush_reqd(sdp) &&
1266                         !gfs2_jrnl_flush_reqd(sdp) &&
1267                         !kthread_should_stop());
1268                 finish_wait(&sdp->sd_logd_waitq, &wait);
1269         }
1270
1271         return 0;
1272 }
1273