fs/gfs2/log.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License version 2.
   8  */
   9
  10 #include <linux/sched.h>
  11 #include <linux/slab.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/completion.h>
  14 #include <linux/buffer_head.h>
  15 #include <linux/gfs2_ondisk.h>
  16 #include <linux/crc32.h>
  17 #include <linux/crc32c.h>
  18 #include <linux/delay.h>
  19 #include <linux/kthread.h>
  20 #include <linux/freezer.h>
  21 #include <linux/bio.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/writeback.h>
  24 #include <linux/list_sort.h>
  25
  26 #include "gfs2.h"
  27 #include "incore.h"
  28 #include "bmap.h"
  29 #include "glock.h"
  30 #include "log.h"
  31 #include "lops.h"
  32 #include "meta_io.h"
  33 #include "util.h"
  34 #include "dir.h"
  35 #include "trace_gfs2.h"
  36
  37 /**
  38  * gfs2_struct2blk - compute stuff
  39  * @sdp: the filesystem
  40  * @nstruct: the number of structures
  41  * @ssize: the size of the 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
  49 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct,
  50                              unsigned int ssize)
  51 {
  52         unsigned int blks;
  53         unsigned int first, second;
  54
  55         blks = 1;
  56         first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize;
  57
  58         if (nstruct > first) {
  59                 second = (sdp->sd_sb.sb_bsize -
  60                           sizeof(struct gfs2_meta_header)) / ssize;
  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  * @mapping: The associated mapping (maybe NULL)
  70  * @bd: The gfs2_bufdata to remove
  71  *
  72  * The ail lock _must_ be held when calling this function
  73  *
  74  */
  75
  76 static void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
  77 {
  78         bd->bd_tr = NULL;
  79         list_del_init(&bd->bd_ail_st_list);
  80         list_del_init(&bd->bd_ail_gl_list);
  81         atomic_dec(&bd->bd_gl->gl_ail_count);
  82         brelse(bd->bd_bh);
  83 }
  84
  85 /**
  86  * gfs2_ail1_start_one - Start I/O on a part of the AIL
  87  * @sdp: the filesystem
  88  * @wbc: The writeback control structure
  89  * @ai: The ail structure
  90  *
  91  */
  92
  93 static int gfs2_ail1_start_one(struct gfs2_sbd *sdp,
  94                                struct writeback_control *wbc,
  95                                struct gfs2_trans *tr,
  96                                bool *withdraw)
  97 __releases(&sdp->sd_ail_lock)
  98 __acquires(&sdp->sd_ail_lock)
  99 {
 100         struct gfs2_glock *gl = NULL;
 101         struct address_space *mapping;
 102         struct gfs2_bufdata *bd, *s;
 103         struct buffer_head *bh;
 104
 105         list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) {
 106                 bh = bd->bd_bh;
 107
 108                 gfs2_assert(sdp, bd->bd_tr == tr);
 109
 110                 if (!buffer_busy(bh)) {
 111                         if (!buffer_uptodate(bh) &&
 112                             !test_and_set_bit(SDF_AIL1_IO_ERROR,
 113                                               &sdp->sd_flags)) {
 114                                 gfs2_io_error_bh(sdp, bh);
 115                                 *withdraw = true;
 116                         }
 117                         list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
 118                         continue;
 119                 }
 120
 121                 if (!buffer_dirty(bh))
 122                         continue;
 123                 if (gl == bd->bd_gl)
 124                         continue;
 125                 gl = bd->bd_gl;
 126                 list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list);
 127                 mapping = bh->b_page->mapping;
 128                 if (!mapping)
 129                         continue;
 130                 spin_unlock(&sdp->sd_ail_lock);
 131                 generic_writepages(mapping, wbc);
 132                 spin_lock(&sdp->sd_ail_lock);
 133                 if (wbc->nr_to_write <= 0)
 134                         break;
 135                 return 1;
 136         }
 137
 138         return 0;
 139 }
 140
 141
 142 /**
 143  * gfs2_ail1_flush - start writeback of some ail1 entries
 144  * @sdp: The super block
 145  * @wbc: The writeback control structure
 146  *
 147  * Writes back some ail1 entries, according to the limits in the
 148  * writeback control structure
 149  */
 150
 151 void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc)
 152 {
 153         struct list_head *head = &sdp->sd_ail1_list;
 154         struct gfs2_trans *tr;
 155         struct blk_plug plug;
 156         bool withdraw = false;
 157
 158         trace_gfs2_ail_flush(sdp, wbc, 1);
 159         blk_start_plug(&plug);
 160         spin_lock(&sdp->sd_ail_lock);
 161 restart:
 162         list_for_each_entry_reverse(tr, head, tr_list) {
 163                 if (wbc->nr_to_write <= 0)
 164                         break;
 165                 if (gfs2_ail1_start_one(sdp, wbc, tr, &withdraw))
 166                         goto restart;
 167         }
 168         spin_unlock(&sdp->sd_ail_lock);
 169         blk_finish_plug(&plug);
 170         if (withdraw)
 171                 gfs2_lm_withdraw(sdp, NULL);
 172         trace_gfs2_ail_flush(sdp, wbc, 0);
 173 }
 174
 175 /**
 176  * gfs2_ail1_start - start writeback of all ail1 entries
 177  * @sdp: The superblock
 178  */
 179
 180 static void gfs2_ail1_start(struct gfs2_sbd *sdp)
 181 {
 182         struct writeback_control wbc = {
 183                 .sync_mode = WB_SYNC_NONE,
 184                 .nr_to_write = LONG_MAX,
 185                 .range_start = 0,
 186                 .range_end = LLONG_MAX,
 187         };
 188
 189         return gfs2_ail1_flush(sdp, &wbc);
 190 }
 191
 192 /**
 193  * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
 194  * @sdp: the filesystem
 195  * @ai: the AIL entry
 196  *
 197  */
 198
 199 static void gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
 200                                 bool *withdraw)
 201 {
 202         struct gfs2_bufdata *bd, *s;
 203         struct buffer_head *bh;
 204
 205         list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list,
 206                                          bd_ail_st_list) {
 207                 bh = bd->bd_bh;
 208                 gfs2_assert(sdp, bd->bd_tr == tr);
 209                 if (buffer_busy(bh))
 210                         continue;
 211                 if (!buffer_uptodate(bh) &&
 212                     !test_and_set_bit(SDF_AIL1_IO_ERROR, &sdp->sd_flags)) {
 213                         gfs2_io_error_bh(sdp, bh);
 214                         *withdraw = true;
 215                 }
 216                 list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
 217         }
 218 }
 219
 220 /**
 221  * gfs2_ail1_empty - Try to empty the ail1 lists
 222  * @sdp: The superblock
 223  *
 224  * Tries to empty the ail1 lists, starting with the oldest first
 225  */
 226
 227 static int gfs2_ail1_empty(struct gfs2_sbd *sdp)
 228 {
 229         struct gfs2_trans *tr, *s;
 230         int oldest_tr = 1;
 231         int ret;
 232         bool withdraw = false;
 233
 234         spin_lock(&sdp->sd_ail_lock);
 235         list_for_each_entry_safe_reverse(tr, s, &sdp->sd_ail1_list, tr_list) {
 236                 gfs2_ail1_empty_one(sdp, tr, &withdraw);
 237                 if (list_empty(&tr->tr_ail1_list) && oldest_tr)
 238                         list_move(&tr->tr_list, &sdp->sd_ail2_list);
 239                 else
 240                         oldest_tr = 0;
 241         }
 242         ret = list_empty(&sdp->sd_ail1_list);
 243         spin_unlock(&sdp->sd_ail_lock);
 244
 245         if (withdraw)
 246                 gfs2_lm_withdraw(sdp, "fatal: I/O error(s)\n");
 247
 248         return ret;
 249 }
 250
 251 static void gfs2_ail1_wait(struct gfs2_sbd *sdp)
 252 {
 253         struct gfs2_trans *tr;
 254         struct gfs2_bufdata *bd;
 255         struct buffer_head *bh;
 256
 257         spin_lock(&sdp->sd_ail_lock);
 258         list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
 259                 list_for_each_entry(bd, &tr->tr_ail1_list, bd_ail_st_list) {
 260                         bh = bd->bd_bh;
 261                         if (!buffer_locked(bh))
 262                                 continue;
 263                         get_bh(bh);
 264                         spin_unlock(&sdp->sd_ail_lock);
 265                         wait_on_buffer(bh);
 266                         brelse(bh);
 267                         return;
 268                 }
 269         }
 270         spin_unlock(&sdp->sd_ail_lock);
 271 }
 272
 273 /**
 274  * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
 275  * @sdp: the filesystem
 276  * @ai: the AIL entry
 277  *
 278  */
 279
 280 static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 281 {
 282         struct list_head *head = &tr->tr_ail2_list;
 283         struct gfs2_bufdata *bd;
 284
 285         while (!list_empty(head)) {
 286                 bd = list_entry(head->prev, struct gfs2_bufdata,
 287                                 bd_ail_st_list);
 288                 gfs2_assert(sdp, bd->bd_tr == tr);
 289                 gfs2_remove_from_ail(bd);
 290         }
 291 }
 292
 293 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
 294 {
 295         struct gfs2_trans *tr, *safe;
 296         unsigned int old_tail = sdp->sd_log_tail;
 297         int wrap = (new_tail < old_tail);
 298         int a, b, rm;
 299
 300         spin_lock(&sdp->sd_ail_lock);
 301
 302         list_for_each_entry_safe(tr, safe, &sdp->sd_ail2_list, tr_list) {
 303                 a = (old_tail <= tr->tr_first);
 304                 b = (tr->tr_first < new_tail);
 305                 rm = (wrap) ? (a || b) : (a && b);
 306                 if (!rm)
 307                         continue;
 308
 309                 gfs2_ail2_empty_one(sdp, tr);
 310                 list_del(&tr->tr_list);
 311                 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list));
 312                 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list));
 313                 kfree(tr);
 314         }
 315
 316         spin_unlock(&sdp->sd_ail_lock);
 317 }
 318
 319 /**
 320  * gfs2_log_release - Release a given number of log blocks
 321  * @sdp: The GFS2 superblock
 322  * @blks: The number of blocks
 323  *
 324  */
 325
 326 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
 327 {
 328
 329         atomic_add(blks, &sdp->sd_log_blks_free);
 330         trace_gfs2_log_blocks(sdp, blks);
 331         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
 332                                   sdp->sd_jdesc->jd_blocks);
 333         up_read(&sdp->sd_log_flush_lock);
 334 }
 335
 336 /**
 337  * gfs2_log_reserve - Make a log reservation
 338  * @sdp: The GFS2 superblock
 339  * @blks: The number of blocks to reserve
 340  *
 341  * Note that we never give out the last few blocks of the journal. Thats
 342  * due to the fact that there is a small number of header blocks
 343  * associated with each log flush. The exact number can't be known until
 344  * flush time, so we ensure that we have just enough free blocks at all
 345  * times to avoid running out during a log flush.
 346  *
 347  * We no longer flush the log here, instead we wake up logd to do that
 348  * for us. To avoid the thundering herd and to ensure that we deal fairly
 349  * with queued waiters, we use an exclusive wait. This means that when we
 350  * get woken with enough journal space to get our reservation, we need to
 351  * wake the next waiter on the list.
 352  *
 353  * Returns: errno
 354  */
 355
 356 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
 357 {
 358         int ret = 0;
 359         unsigned reserved_blks = 7 * (4096 / sdp->sd_vfs->s_blocksize);
 360         unsigned wanted = blks + reserved_blks;
 361         DEFINE_WAIT(wait);
 362         int did_wait = 0;
 363         unsigned int free_blocks;
 364
 365         if (gfs2_assert_warn(sdp, blks) ||
 366             gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
 367                 return -EINVAL;
 368         atomic_add(blks, &sdp->sd_log_blks_needed);
 369 retry:
 370         free_blocks = atomic_read(&sdp->sd_log_blks_free);
 371         if (unlikely(free_blocks <= wanted)) {
 372                 do {
 373                         prepare_to_wait_exclusive(&sdp->sd_log_waitq, &wait,
 374                                         TASK_UNINTERRUPTIBLE);
 375                         wake_up(&sdp->sd_logd_waitq);
 376                         did_wait = 1;
 377                         if (atomic_read(&sdp->sd_log_blks_free) <= wanted)
 378                                 io_schedule();
 379                         free_blocks = atomic_read(&sdp->sd_log_blks_free);
 380                 } while(free_blocks <= wanted);
 381                 finish_wait(&sdp->sd_log_waitq, &wait);
 382         }
 383         atomic_inc(&sdp->sd_reserving_log);
 384         if (atomic_cmpxchg(&sdp->sd_log_blks_free, free_blocks,
 385                                 free_blocks - blks) != free_blocks) {
 386                 if (atomic_dec_and_test(&sdp->sd_reserving_log))
 387                         wake_up(&sdp->sd_reserving_log_wait);
 388                 goto retry;
 389         }
 390         atomic_sub(blks, &sdp->sd_log_blks_needed);
 391         trace_gfs2_log_blocks(sdp, -blks);
 392
 393         /*
 394          * If we waited, then so might others, wake them up _after_ we get
 395          * our share of the log.
 396          */
 397         if (unlikely(did_wait))
 398                 wake_up(&sdp->sd_log_waitq);
 399
 400         down_read(&sdp->sd_log_flush_lock);
 401         if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) {
 402                 gfs2_log_release(sdp, blks);
 403                 ret = -EROFS;
 404         }
 405         if (atomic_dec_and_test(&sdp->sd_reserving_log))
 406                 wake_up(&sdp->sd_reserving_log_wait);
 407         return ret;
 408 }
 409
 410 /**
 411  * log_distance - Compute distance between two journal blocks
 412  * @sdp: The GFS2 superblock
 413  * @newer: The most recent journal block of the pair
 414  * @older: The older journal block of the pair
 415  *
 416  *   Compute the distance (in the journal direction) between two
 417  *   blocks in the journal
 418  *
 419  * Returns: the distance in blocks
 420  */
 421
 422 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
 423                                         unsigned int older)
 424 {
 425         int dist;
 426
 427         dist = newer - older;
 428         if (dist < 0)
 429                 dist += sdp->sd_jdesc->jd_blocks;
 430
 431         return dist;
 432 }
 433
 434 /**
 435  * calc_reserved - Calculate the number of blocks to reserve when
 436  *                 refunding a transaction's unused buffers.
 437  * @sdp: The GFS2 superblock
 438  *
 439  * This is complex.  We need to reserve room for all our currently used
 440  * metadata buffers (e.g. normal file I/O rewriting file time stamps) and
 441  * all our journaled data buffers for journaled files (e.g. files in the
 442  * meta_fs like rindex, or files for which chattr +j was done.)
 443  * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush
 444  * will count it as free space (sd_log_blks_free) and corruption will follow.
 445  *
 446  * We can have metadata bufs and jdata bufs in the same journal.  So each
 447  * type gets its own log header, for which we need to reserve a block.
 448  * In fact, each type has the potential for needing more than one header
 449  * in cases where we have more buffers than will fit on a journal page.
 450  * Metadata journal entries take up half the space of journaled buffer entries.
 451  * Thus, metadata entries have buf_limit (502) and journaled buffers have
 452  * databuf_limit (251) before they cause a wrap around.
 453  *
 454  * Also, we need to reserve blocks for revoke journal entries and one for an
 455  * overall header for the lot.
 456  *
 457  * Returns: the number of blocks reserved
 458  */
 459 static unsigned int calc_reserved(struct gfs2_sbd *sdp)
 460 {
 461         unsigned int reserved = 0;
 462         unsigned int mbuf;
 463         unsigned int dbuf;
 464         struct gfs2_trans *tr = sdp->sd_log_tr;
 465
 466         if (tr) {
 467                 mbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
 468                 dbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
 469                 reserved = mbuf + dbuf;
 470                 /* Account for header blocks */
 471                 reserved += DIV_ROUND_UP(mbuf, buf_limit(sdp));
 472                 reserved += DIV_ROUND_UP(dbuf, databuf_limit(sdp));
 473         }
 474
 475         if (sdp->sd_log_commited_revoke > 0)
 476                 reserved += gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
 477                                           sizeof(u64));
 478         /* One for the overall header */
 479         if (reserved)
 480                 reserved++;
 481         return reserved;
 482 }
 483
 484 static unsigned int current_tail(struct gfs2_sbd *sdp)
 485 {
 486         struct gfs2_trans *tr;
 487         unsigned int tail;
 488
 489         spin_lock(&sdp->sd_ail_lock);
 490
 491         if (list_empty(&sdp->sd_ail1_list)) {
 492                 tail = sdp->sd_log_head;
 493         } else {
 494                 tr = list_entry(sdp->sd_ail1_list.prev, struct gfs2_trans,
 495                                 tr_list);
 496                 tail = tr->tr_first;
 497         }
 498
 499         spin_unlock(&sdp->sd_ail_lock);
 500
 501         return tail;
 502 }
 503
 504 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
 505 {
 506         unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
 507
 508         ail2_empty(sdp, new_tail);
 509
 510         atomic_add(dist, &sdp->sd_log_blks_free);
 511         trace_gfs2_log_blocks(sdp, dist);
 512         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
 513                              sdp->sd_jdesc->jd_blocks);
 514
 515         sdp->sd_log_tail = new_tail;
 516 }
 517
 518
 519 static void log_flush_wait(struct gfs2_sbd *sdp)
 520 {
 521         DEFINE_WAIT(wait);
 522
 523         if (atomic_read(&sdp->sd_log_in_flight)) {
 524                 do {
 525                         prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
 526                                         TASK_UNINTERRUPTIBLE);
 527                         if (atomic_read(&sdp->sd_log_in_flight))
 528                                 io_schedule();
 529                 } while(atomic_read(&sdp->sd_log_in_flight));
 530                 finish_wait(&sdp->sd_log_flush_wait, &wait);
 531         }
 532 }
 533
 534 static int ip_cmp(void *priv, struct list_head *a, struct list_head *b)
 535 {
 536         struct gfs2_inode *ipa, *ipb;
 537
 538         ipa = list_entry(a, struct gfs2_inode, i_ordered);
 539         ipb = list_entry(b, struct gfs2_inode, i_ordered);
 540
 541         if (ipa->i_no_addr < ipb->i_no_addr)
 542                 return -1;
 543         if (ipa->i_no_addr > ipb->i_no_addr)
 544                 return 1;
 545         return 0;
 546 }
 547
 548 static void gfs2_ordered_write(struct gfs2_sbd *sdp)
 549 {
 550         struct gfs2_inode *ip;
 551         LIST_HEAD(written);
 552
 553         spin_lock(&sdp->sd_ordered_lock);
 554         list_sort(NULL, &sdp->sd_log_le_ordered, &ip_cmp);
 555         while (!list_empty(&sdp->sd_log_le_ordered)) {
 556                 ip = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_inode, i_ordered);
 557                 if (ip->i_inode.i_mapping->nrpages == 0) {
 558                         test_and_clear_bit(GIF_ORDERED, &ip->i_flags);
 559                         list_del(&ip->i_ordered);
 560                         continue;
 561                 }
 562                 list_move(&ip->i_ordered, &written);
 563                 spin_unlock(&sdp->sd_ordered_lock);
 564                 filemap_fdatawrite(ip->i_inode.i_mapping);
 565                 spin_lock(&sdp->sd_ordered_lock);
 566         }
 567         list_splice(&written, &sdp->sd_log_le_ordered);
 568         spin_unlock(&sdp->sd_ordered_lock);
 569 }
 570
 571 static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
 572 {
 573         struct gfs2_inode *ip;
 574
 575         spin_lock(&sdp->sd_ordered_lock);
 576         while (!list_empty(&sdp->sd_log_le_ordered)) {
 577                 ip = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_inode, i_ordered);
 578                 list_del(&ip->i_ordered);
 579                 WARN_ON(!test_and_clear_bit(GIF_ORDERED, &ip->i_flags));
 580                 if (ip->i_inode.i_mapping->nrpages == 0)
 581                         continue;
 582                 spin_unlock(&sdp->sd_ordered_lock);
 583                 filemap_fdatawait(ip->i_inode.i_mapping);
 584                 spin_lock(&sdp->sd_ordered_lock);
 585         }
 586         spin_unlock(&sdp->sd_ordered_lock);
 587 }
 588
 589 void gfs2_ordered_del_inode(struct gfs2_inode *ip)
 590 {
 591         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 592
 593         spin_lock(&sdp->sd_ordered_lock);
 594         if (test_and_clear_bit(GIF_ORDERED, &ip->i_flags))
 595                 list_del(&ip->i_ordered);
 596         spin_unlock(&sdp->sd_ordered_lock);
 597 }
 598
 599 void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
 600 {
 601         struct buffer_head *bh = bd->bd_bh;
 602         struct gfs2_glock *gl = bd->bd_gl;
 603
 604         bh->b_private = NULL;
 605         bd->bd_blkno = bh->b_blocknr;
 606         gfs2_remove_from_ail(bd); /* drops ref on bh */
 607         bd->bd_bh = NULL;
 608         sdp->sd_log_num_revoke++;
 609         if (!test_bit(GLF_REVOKES, &gl->gl_flags)) {
 610                 set_bit(GLF_REVOKES, &gl->gl_flags);
 611                 gfs2_glock_hold(gl);
 612         }
 613         set_bit(GLF_LFLUSH, &gl->gl_flags);
 614         list_add(&bd->bd_list, &sdp->sd_log_le_revoke);
 615 }
 616
 617 void gfs2_write_revokes(struct gfs2_sbd *sdp)
 618 {
 619         struct gfs2_trans *tr;
 620         struct gfs2_bufdata *bd, *tmp;
 621         int have_revokes = 0;
 622         int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
 623
 624         gfs2_ail1_empty(sdp);
 625         spin_lock(&sdp->sd_ail_lock);
 626         list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
 627                 list_for_each_entry(bd, &tr->tr_ail2_list, bd_ail_st_list) {
 628                         if (list_empty(&bd->bd_list)) {
 629                                 have_revokes = 1;
 630                                 goto done;
 631                         }
 632                 }
 633         }
 634 done:
 635         spin_unlock(&sdp->sd_ail_lock);
 636         if (have_revokes == 0)
 637                 return;
 638         while (sdp->sd_log_num_revoke > max_revokes)
 639                 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
 640         max_revokes -= sdp->sd_log_num_revoke;
 641         if (!sdp->sd_log_num_revoke) {
 642                 atomic_dec(&sdp->sd_log_blks_free);
 643                 /* If no blocks have been reserved, we need to also
 644                  * reserve a block for the header */
 645                 if (!sdp->sd_log_blks_reserved)
 646                         atomic_dec(&sdp->sd_log_blks_free);
 647         }
 648         gfs2_log_lock(sdp);
 649         spin_lock(&sdp->sd_ail_lock);
 650         list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
 651                 list_for_each_entry_safe(bd, tmp, &tr->tr_ail2_list, bd_ail_st_list) {
 652                         if (max_revokes == 0)
 653                                 goto out_of_blocks;
 654                         if (!list_empty(&bd->bd_list))
 655                                 continue;
 656                         gfs2_add_revoke(sdp, bd);
 657                         max_revokes--;
 658                 }
 659         }
 660 out_of_blocks:
 661         spin_unlock(&sdp->sd_ail_lock);
 662         gfs2_log_unlock(sdp);
 663
 664         if (!sdp->sd_log_num_revoke) {
 665                 atomic_inc(&sdp->sd_log_blks_free);
 666                 if (!sdp->sd_log_blks_reserved)
 667                         atomic_inc(&sdp->sd_log_blks_free);
 668         }
 669 }
 670
 671 /**
 672  * gfs2_write_log_header - Write a journal log header buffer at lblock
 673  * @sdp: The GFS2 superblock
 674  * @jd: journal descriptor of the journal to which we are writing
 675  * @seq: sequence number
 676  * @tail: tail of the log
 677  * @lblock: value for lh_blkno (block number relative to start of journal)
 678  * @flags: log header flags GFS2_LOG_HEAD_*
 679  * @op_flags: flags to pass to the bio
 680  *
 681  * Returns: the initialized log buffer descriptor
 682  */
 683
 684 void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
 685                            u64 seq, u32 tail, u32 lblock, u32 flags,
 686                            int op_flags)
 687 {
 688         struct gfs2_log_header *lh;
 689         u32 hash, crc;
 690         struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 691         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 692         struct timespec64 tv;
 693         struct super_block *sb = sdp->sd_vfs;
 694         u64 dblock;
 695
 696         lh = page_address(page);
 697         clear_page(lh);
 698
 699         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 700         lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
 701         lh->lh_header.__pad0 = cpu_to_be64(0);
 702         lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
 703         lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
 704         lh->lh_sequence = cpu_to_be64(seq);
 705         lh->lh_flags = cpu_to_be32(flags);
 706         lh->lh_tail = cpu_to_be32(tail);
 707         lh->lh_blkno = cpu_to_be32(lblock);
 708         hash = ~crc32(~0, lh, LH_V1_SIZE);
 709         lh->lh_hash = cpu_to_be32(hash);
 710
 711         ktime_get_coarse_real_ts64(&tv);
 712         lh->lh_nsec = cpu_to_be32(tv.tv_nsec);
 713         lh->lh_sec = cpu_to_be64(tv.tv_sec);
 714         if (!list_empty(&jd->extent_list))
 715                 dblock = gfs2_log_bmap(sdp);
 716         else {
 717                 int ret = gfs2_lblk_to_dblk(jd->jd_inode, lblock, &dblock);
 718                 if (gfs2_assert_withdraw(sdp, ret == 0))
 719                         return;
 720         }
 721         lh->lh_addr = cpu_to_be64(dblock);
 722         lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr);
 723
 724         /* We may only write local statfs, quota, etc., when writing to our
 725            own journal. The values are left 0 when recovering a journal
 726            different from our own. */
 727         if (!(flags & GFS2_LOG_HEAD_RECOVERY)) {
 728                 lh->lh_statfs_addr =
 729                         cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr);
 730                 lh->lh_quota_addr =
 731                         cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr);
 732
 733                 spin_lock(&sdp->sd_statfs_spin);
 734                 lh->lh_local_total = cpu_to_be64(l_sc->sc_total);
 735                 lh->lh_local_free = cpu_to_be64(l_sc->sc_free);
 736                 lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes);
 737                 spin_unlock(&sdp->sd_statfs_spin);
 738         }
 739
 740         BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE);
 741
 742         crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
 743                      sb->s_blocksize - LH_V1_SIZE - 4);
 744         lh->lh_crc = cpu_to_be32(crc);
 745
 746         gfs2_log_write(sdp, page, sb->s_blocksize, 0, dblock);
 747         gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE, op_flags);
 748         log_flush_wait(sdp);
 749 }
 750
 751 /**
 752  * log_write_header - Get and initialize a journal header buffer
 753  * @sdp: The GFS2 superblock
 754  * @flags: The log header flags, including log header origin
 755  *
 756  * Returns: the initialized log buffer descriptor
 757  */
 758
 759 static void log_write_header(struct gfs2_sbd *sdp, u32 flags)
 760 {
 761         unsigned int tail;
 762         int op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC;
 763         enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
 764
 765         gfs2_assert_withdraw(sdp, (state != SFS_FROZEN));
 766         tail = current_tail(sdp);
 767
 768         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) {
 769                 gfs2_ordered_wait(sdp);
 770                 log_flush_wait(sdp);
 771                 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
 772         }
 773         sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
 774         gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++, tail,
 775                               sdp->sd_log_flush_head, flags, op_flags);
 776
 777         if (sdp->sd_log_tail != tail)
 778                 log_pull_tail(sdp, tail);
 779 }
 780
 781 /**
 782  * gfs2_log_flush - flush incore transaction(s)
 783  * @sdp: the filesystem
 784  * @gl: The glock structure to flush.  If NULL, flush the whole incore log
 785  * @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags
 786  *
 787  */
 788
 789 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags)
 790 {
 791         struct gfs2_trans *tr;
 792         enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
 793
 794         down_write(&sdp->sd_log_flush_lock);
 795
 796         /* Log might have been flushed while we waited for the flush lock */
 797         if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) {
 798                 up_write(&sdp->sd_log_flush_lock);
 799                 return;
 800         }
 801         trace_gfs2_log_flush(sdp, 1, flags);
 802
 803         if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN)
 804                 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 805
 806         sdp->sd_log_flush_head = sdp->sd_log_head;
 807         tr = sdp->sd_log_tr;
 808         if (tr) {
 809                 sdp->sd_log_tr = NULL;
 810                 INIT_LIST_HEAD(&tr->tr_ail1_list);
 811                 INIT_LIST_HEAD(&tr->tr_ail2_list);
 812                 tr->tr_first = sdp->sd_log_flush_head;
 813                 if (unlikely (state == SFS_FROZEN))
 814                         gfs2_assert_withdraw(sdp, !tr->tr_num_buf_new && !tr->tr_num_databuf_new);
 815         }
 816
 817         if (unlikely(state == SFS_FROZEN))
 818                 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 819         gfs2_assert_withdraw(sdp,
 820                         sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
 821
 822         gfs2_ordered_write(sdp);
 823         lops_before_commit(sdp, tr);
 824         gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE, 0);
 825
 826         if (sdp->sd_log_head != sdp->sd_log_flush_head) {
 827                 log_flush_wait(sdp);
 828                 log_write_header(sdp, flags);
 829         } else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
 830                 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
 831                 trace_gfs2_log_blocks(sdp, -1);
 832                 log_write_header(sdp, flags);
 833         }
 834         lops_after_commit(sdp, tr);
 835
 836         gfs2_log_lock(sdp);
 837         sdp->sd_log_head = sdp->sd_log_flush_head;
 838         sdp->sd_log_blks_reserved = 0;
 839         sdp->sd_log_commited_revoke = 0;
 840
 841         spin_lock(&sdp->sd_ail_lock);
 842         if (tr && !list_empty(&tr->tr_ail1_list)) {
 843                 list_add(&tr->tr_list, &sdp->sd_ail1_list);
 844                 tr = NULL;
 845         }
 846         spin_unlock(&sdp->sd_ail_lock);
 847         gfs2_log_unlock(sdp);
 848
 849         if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) {
 850                 if (!sdp->sd_log_idle) {
 851                         for (;;) {
 852                                 gfs2_ail1_start(sdp);
 853                                 gfs2_ail1_wait(sdp);
 854                                 if (gfs2_ail1_empty(sdp))
 855                                         break;
 856                         }
 857                         atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
 858                         trace_gfs2_log_blocks(sdp, -1);
 859                         log_write_header(sdp, flags);
 860                         sdp->sd_log_head = sdp->sd_log_flush_head;
 861                 }
 862                 if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
 863                              GFS2_LOG_HEAD_FLUSH_FREEZE))
 864                         gfs2_log_shutdown(sdp);
 865                 if (flags & GFS2_LOG_HEAD_FLUSH_FREEZE)
 866                         atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
 867         }
 868
 869         trace_gfs2_log_flush(sdp, 0, flags);
 870         up_write(&sdp->sd_log_flush_lock);
 871
 872         kfree(tr);
 873 }
 874
 875 /**
 876  * gfs2_merge_trans - Merge a new transaction into a cached transaction
 877  * @old: Original transaction to be expanded
 878  * @new: New transaction to be merged
 879  */
 880
 881 static void gfs2_merge_trans(struct gfs2_trans *old, struct gfs2_trans *new)
 882 {
 883         WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags));
 884
 885         old->tr_num_buf_new     += new->tr_num_buf_new;
 886         old->tr_num_databuf_new += new->tr_num_databuf_new;
 887         old->tr_num_buf_rm      += new->tr_num_buf_rm;
 888         old->tr_num_databuf_rm  += new->tr_num_databuf_rm;
 889         old->tr_num_revoke      += new->tr_num_revoke;
 890         old->tr_num_revoke_rm   += new->tr_num_revoke_rm;
 891
 892         list_splice_tail_init(&new->tr_databuf, &old->tr_databuf);
 893         list_splice_tail_init(&new->tr_buf, &old->tr_buf);
 894 }
 895
 896 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 897 {
 898         unsigned int reserved;
 899         unsigned int unused;
 900         unsigned int maxres;
 901
 902         gfs2_log_lock(sdp);
 903
 904         if (sdp->sd_log_tr) {
 905                 gfs2_merge_trans(sdp->sd_log_tr, tr);
 906         } else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) {
 907                 gfs2_assert_withdraw(sdp, test_bit(TR_ALLOCED, &tr->tr_flags));
 908                 sdp->sd_log_tr = tr;
 909                 set_bit(TR_ATTACHED, &tr->tr_flags);
 910         }
 911
 912         sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
 913         reserved = calc_reserved(sdp);
 914         maxres = sdp->sd_log_blks_reserved + tr->tr_reserved;
 915         gfs2_assert_withdraw(sdp, maxres >= reserved);
 916         unused = maxres - reserved;
 917         atomic_add(unused, &sdp->sd_log_blks_free);
 918         trace_gfs2_log_blocks(sdp, unused);
 919         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
 920                              sdp->sd_jdesc->jd_blocks);
 921         sdp->sd_log_blks_reserved = reserved;
 922
 923         gfs2_log_unlock(sdp);
 924 }
 925
 926 /**
 927  * gfs2_log_commit - Commit a transaction to the log
 928  * @sdp: the filesystem
 929  * @tr: the transaction
 930  *
 931  * We wake up gfs2_logd if the number of pinned blocks exceed thresh1
 932  * or the total number of used blocks (pinned blocks plus AIL blocks)
 933  * is greater than thresh2.
 934  *
 935  * At mount time thresh1 is 1/3rd of journal size, thresh2 is 2/3rd of
 936  * journal size.
 937  *
 938  * Returns: errno
 939  */
 940
 941 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 942 {
 943         log_refund(sdp, tr);
 944
 945         if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) ||
 946             ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) >
 947             atomic_read(&sdp->sd_log_thresh2)))
 948                 wake_up(&sdp->sd_logd_waitq);
 949 }
 950
 951 /**
 952  * gfs2_log_shutdown - write a shutdown header into a journal
 953  * @sdp: the filesystem
 954  *
 955  */
 956
 957 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
 958 {
 959         gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
 960         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 961         gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
 962
 963         sdp->sd_log_flush_head = sdp->sd_log_head;
 964
 965         log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN);
 966
 967         gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
 968         gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
 969
 970         sdp->sd_log_head = sdp->sd_log_flush_head;
 971         sdp->sd_log_tail = sdp->sd_log_head;
 972 }
 973
 974 static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
 975 {
 976         return (atomic_read(&sdp->sd_log_pinned) +
 977                 atomic_read(&sdp->sd_log_blks_needed) >=
 978                 atomic_read(&sdp->sd_log_thresh1));
 979 }
 980
 981 static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
 982 {
 983         unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free);
 984
 985         if (test_and_clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags))
 986                 return 1;
 987
 988         return used_blocks + atomic_read(&sdp->sd_log_blks_needed) >=
 989                 atomic_read(&sdp->sd_log_thresh2);
 990 }
 991
 992 /**
 993  * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
 994  * @sdp: Pointer to GFS2 superblock
 995  *
 996  * Also, periodically check to make sure that we're using the most recent
 997  * journal index.
 998  */
 999
1000 int gfs2_logd(void *data)
1001 {
1002         struct gfs2_sbd *sdp = data;
1003         unsigned long t = 1;
1004         DEFINE_WAIT(wait);
1005         bool did_flush;
1006
1007         while (!kthread_should_stop()) {
1008
1009                 /* Check for errors writing to the journal */
1010                 if (sdp->sd_log_error) {
1011                         gfs2_lm_withdraw(sdp,
1012                                          "GFS2: fsid=%s: error %d: "
1013                                          "withdrawing the file system to "
1014                                          "prevent further damage.\n",
1015                                          sdp->sd_fsname, sdp->sd_log_error);
1016                 }
1017
1018                 did_flush = false;
1019                 if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
1020                         gfs2_ail1_empty(sdp);
1021                         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1022                                        GFS2_LFC_LOGD_JFLUSH_REQD);
1023                         did_flush = true;
1024                 }
1025
1026                 if (gfs2_ail_flush_reqd(sdp)) {
1027                         gfs2_ail1_start(sdp);
1028                         gfs2_ail1_wait(sdp);
1029                         gfs2_ail1_empty(sdp);
1030                         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1031                                        GFS2_LFC_LOGD_AIL_FLUSH_REQD);
1032                         did_flush = true;
1033                 }
1034
1035                 if (!gfs2_ail_flush_reqd(sdp) || did_flush)
1036                         wake_up(&sdp->sd_log_waitq);
1037
1038                 t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
1039
1040                 try_to_freeze();
1041
1042                 do {
1043                         prepare_to_wait(&sdp->sd_logd_waitq, &wait,
1044                                         TASK_INTERRUPTIBLE);
1045                         if (!gfs2_ail_flush_reqd(sdp) &&
1046                             !gfs2_jrnl_flush_reqd(sdp) &&
1047                             !kthread_should_stop())
1048                                 t = schedule_timeout(t);
1049                 } while(t && !gfs2_ail_flush_reqd(sdp) &&
1050                         !gfs2_jrnl_flush_reqd(sdp) &&
1051                         !kthread_should_stop());
1052                 finish_wait(&sdp->sd_logd_waitq, &wait);
1053         }
1054
1055         return 0;
1056 }
1057