fs/gfs2/lops.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2006 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/mempool.h>
  16 #include <linux/gfs2_ondisk.h>
  17 #include <linux/bio.h>
  18 #include <linux/fs.h>
  19 #include <linux/list_sort.h>
  20
  21 #include "gfs2.h"
  22 #include "incore.h"
  23 #include "inode.h"
  24 #include "glock.h"
  25 #include "log.h"
  26 #include "lops.h"
  27 #include "meta_io.h"
  28 #include "recovery.h"
  29 #include "rgrp.h"
  30 #include "trans.h"
  31 #include "util.h"
  32 #include "trace_gfs2.h"
  33
  34 /**
  35  * gfs2_pin - Pin a buffer in memory
  36  * @sdp: The superblock
  37  * @bh: The buffer to be pinned
  38  *
  39  * The log lock must be held when calling this function
  40  */
  41 void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
  42 {
  43         struct gfs2_bufdata *bd;
  44
  45         BUG_ON(!current->journal_info);
  46
  47         clear_buffer_dirty(bh);
  48         if (test_set_buffer_pinned(bh))
  49                 gfs2_assert_withdraw(sdp, 0);
  50         if (!buffer_uptodate(bh))
  51                 gfs2_io_error_bh(sdp, bh);
  52         bd = bh->b_private;
  53         /* If this buffer is in the AIL and it has already been written
  54          * to in-place disk block, remove it from the AIL.
  55          */
  56         spin_lock(&sdp->sd_ail_lock);
  57         if (bd->bd_tr)
  58                 list_move(&bd->bd_ail_st_list, &bd->bd_tr->tr_ail2_list);
  59         spin_unlock(&sdp->sd_ail_lock);
  60         get_bh(bh);
  61         atomic_inc(&sdp->sd_log_pinned);
  62         trace_gfs2_pin(bd, 1);
  63 }
  64
  65 static bool buffer_is_rgrp(const struct gfs2_bufdata *bd)
  66 {
  67         return bd->bd_gl->gl_name.ln_type == LM_TYPE_RGRP;
  68 }
  69
  70 static void maybe_release_space(struct gfs2_bufdata *bd)
  71 {
  72         struct gfs2_glock *gl = bd->bd_gl;
  73         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
  74         struct gfs2_rgrpd *rgd = gl->gl_object;
  75         unsigned int index = bd->bd_bh->b_blocknr - gl->gl_name.ln_number;
  76         struct gfs2_bitmap *bi = rgd->rd_bits + index;
  77
  78         if (bi->bi_clone == NULL)
  79                 return;
  80         if (sdp->sd_args.ar_discard)
  81                 gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi, 1, NULL);
  82         memcpy(bi->bi_clone + bi->bi_offset,
  83                bd->bd_bh->b_data + bi->bi_offset, bi->bi_len);
  84         clear_bit(GBF_FULL, &bi->bi_flags);
  85         rgd->rd_free_clone = rgd->rd_free;
  86         rgd->rd_extfail_pt = rgd->rd_free;
  87 }
  88
  89 /**
  90  * gfs2_unpin - Unpin a buffer
  91  * @sdp: the filesystem the buffer belongs to
  92  * @bh: The buffer to unpin
  93  * @ai:
  94  * @flags: The inode dirty flags
  95  *
  96  */
  97
  98 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
  99                        struct gfs2_trans *tr)
 100 {
 101         struct gfs2_bufdata *bd = bh->b_private;
 102
 103         BUG_ON(!buffer_uptodate(bh));
 104         BUG_ON(!buffer_pinned(bh));
 105
 106         lock_buffer(bh);
 107         mark_buffer_dirty(bh);
 108         clear_buffer_pinned(bh);
 109
 110         if (buffer_is_rgrp(bd))
 111                 maybe_release_space(bd);
 112
 113         spin_lock(&sdp->sd_ail_lock);
 114         if (bd->bd_tr) {
 115                 list_del(&bd->bd_ail_st_list);
 116                 brelse(bh);
 117         } else {
 118                 struct gfs2_glock *gl = bd->bd_gl;
 119                 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
 120                 atomic_inc(&gl->gl_ail_count);
 121         }
 122         bd->bd_tr = tr;
 123         list_add(&bd->bd_ail_st_list, &tr->tr_ail1_list);
 124         spin_unlock(&sdp->sd_ail_lock);
 125
 126         clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 127         trace_gfs2_pin(bd, 0);
 128         unlock_buffer(bh);
 129         atomic_dec(&sdp->sd_log_pinned);
 130 }
 131
 132 static void gfs2_log_incr_head(struct gfs2_sbd *sdp)
 133 {
 134         BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) &&
 135                (sdp->sd_log_flush_head != sdp->sd_log_head));
 136
 137         if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
 138                 sdp->sd_log_flush_head = 0;
 139                 sdp->sd_log_flush_wrapped = 1;
 140         }
 141 }
 142
 143 static u64 gfs2_log_bmap(struct gfs2_sbd *sdp)
 144 {
 145         unsigned int lbn = sdp->sd_log_flush_head;
 146         struct gfs2_journal_extent *je;
 147         u64 block;
 148
 149         list_for_each_entry(je, &sdp->sd_jdesc->extent_list, list) {
 150                 if ((lbn >= je->lblock) && (lbn < (je->lblock + je->blocks))) {
 151                         block = je->dblock + lbn - je->lblock;
 152                         gfs2_log_incr_head(sdp);
 153                         return block;
 154                 }
 155         }
 156
 157         return -1;
 158 }
 159
 160 /**
 161  * gfs2_end_log_write_bh - end log write of pagecache data with buffers
 162  * @sdp: The superblock
 163  * @bvec: The bio_vec
 164  * @error: The i/o status
 165  *
 166  * This finds the relavent buffers and unlocks then and sets the
 167  * error flag according to the status of the i/o request. This is
 168  * used when the log is writing data which has an in-place version
 169  * that is pinned in the pagecache.
 170  */
 171
 172 static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp, struct bio_vec *bvec,
 173                                   int error)
 174 {
 175         struct buffer_head *bh, *next;
 176         struct page *page = bvec->bv_page;
 177         unsigned size;
 178
 179         bh = page_buffers(page);
 180         size = bvec->bv_len;
 181         while (bh_offset(bh) < bvec->bv_offset)
 182                 bh = bh->b_this_page;
 183         do {
 184                 if (error)
 185                         set_buffer_write_io_error(bh);
 186                 unlock_buffer(bh);
 187                 next = bh->b_this_page;
 188                 size -= bh->b_size;
 189                 brelse(bh);
 190                 bh = next;
 191         } while(bh && size);
 192 }
 193
 194 /**
 195  * gfs2_end_log_write - end of i/o to the log
 196  * @bio: The bio
 197  * @error: Status of i/o request
 198  *
 199  * Each bio_vec contains either data from the pagecache or data
 200  * relating to the log itself. Here we iterate over the bio_vec
 201  * array, processing both kinds of data.
 202  *
 203  */
 204
 205 static void gfs2_end_log_write(struct bio *bio)
 206 {
 207         struct gfs2_sbd *sdp = bio->bi_private;
 208         struct bio_vec *bvec;
 209         struct page *page;
 210         int i;
 211
 212         if (bio->bi_error) {
 213                 sdp->sd_log_error = bio->bi_error;
 214                 fs_err(sdp, "Error %d writing to log\n", bio->bi_error);
 215         }
 216
 217         bio_for_each_segment_all(bvec, bio, i) {
 218                 page = bvec->bv_page;
 219                 if (page_has_buffers(page))
 220                         gfs2_end_log_write_bh(sdp, bvec, bio->bi_error);
 221                 else
 222                         mempool_free(page, gfs2_page_pool);
 223         }
 224
 225         bio_put(bio);
 226         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 227                 wake_up(&sdp->sd_log_flush_wait);
 228 }
 229
 230 /**
 231  * gfs2_log_flush_bio - Submit any pending log bio
 232  * @sdp: The superblock
 233  * @rw: The rw flags
 234  *
 235  * Submit any pending part-built or full bio to the block device. If
 236  * there is no pending bio, then this is a no-op.
 237  */
 238
 239 void gfs2_log_flush_bio(struct gfs2_sbd *sdp, int rw)
 240 {
 241         if (sdp->sd_log_bio) {
 242                 atomic_inc(&sdp->sd_log_in_flight);
 243                 submit_bio(rw, sdp->sd_log_bio);
 244                 sdp->sd_log_bio = NULL;
 245         }
 246 }
 247
 248 /**
 249  * gfs2_log_alloc_bio - Allocate a new bio for log writing
 250  * @sdp: The superblock
 251  * @blkno: The next device block number we want to write to
 252  *
 253  * This should never be called when there is a cached bio in the
 254  * super block. When it returns, there will be a cached bio in the
 255  * super block which will have as many bio_vecs as the device is
 256  * happy to handle.
 257  *
 258  * Returns: Newly allocated bio
 259  */
 260
 261 static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno)
 262 {
 263         struct super_block *sb = sdp->sd_vfs;
 264         struct bio *bio;
 265
 266         BUG_ON(sdp->sd_log_bio);
 267
 268         bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
 269         bio->bi_iter.bi_sector = blkno * (sb->s_blocksize >> 9);
 270         bio->bi_bdev = sb->s_bdev;
 271         bio->bi_end_io = gfs2_end_log_write;
 272         bio->bi_private = sdp;
 273
 274         sdp->sd_log_bio = bio;
 275
 276         return bio;
 277 }
 278
 279 /**
 280  * gfs2_log_get_bio - Get cached log bio, or allocate a new one
 281  * @sdp: The superblock
 282  * @blkno: The device block number we want to write to
 283  *
 284  * If there is a cached bio, then if the next block number is sequential
 285  * with the previous one, return it, otherwise flush the bio to the
 286  * device. If there is not a cached bio, or we just flushed it, then
 287  * allocate a new one.
 288  *
 289  * Returns: The bio to use for log writes
 290  */
 291
 292 static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno)
 293 {
 294         struct bio *bio = sdp->sd_log_bio;
 295         u64 nblk;
 296
 297         if (bio) {
 298                 nblk = bio_end_sector(bio);
 299                 nblk >>= sdp->sd_fsb2bb_shift;
 300                 if (blkno == nblk)
 301                         return bio;
 302                 gfs2_log_flush_bio(sdp, WRITE);
 303         }
 304
 305         return gfs2_log_alloc_bio(sdp, blkno);
 306 }
 307
 308
 309 /**
 310  * gfs2_log_write - write to log
 311  * @sdp: the filesystem
 312  * @page: the page to write
 313  * @size: the size of the data to write
 314  * @offset: the offset within the page
 315  *
 316  * Try and add the page segment to the current bio. If that fails,
 317  * submit the current bio to the device and create a new one, and
 318  * then add the page segment to that.
 319  */
 320
 321 static void gfs2_log_write(struct gfs2_sbd *sdp, struct page *page,
 322                            unsigned size, unsigned offset)
 323 {
 324         u64 blkno = gfs2_log_bmap(sdp);
 325         struct bio *bio;
 326         int ret;
 327
 328         bio = gfs2_log_get_bio(sdp, blkno);
 329         ret = bio_add_page(bio, page, size, offset);
 330         if (ret == 0) {
 331                 gfs2_log_flush_bio(sdp, WRITE);
 332                 bio = gfs2_log_alloc_bio(sdp, blkno);
 333                 ret = bio_add_page(bio, page, size, offset);
 334                 WARN_ON(ret == 0);
 335         }
 336 }
 337
 338 /**
 339  * gfs2_log_write_bh - write a buffer's content to the log
 340  * @sdp: The super block
 341  * @bh: The buffer pointing to the in-place location
 342  *
 343  * This writes the content of the buffer to the next available location
 344  * in the log. The buffer will be unlocked once the i/o to the log has
 345  * completed.
 346  */
 347
 348 static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh)
 349 {
 350         gfs2_log_write(sdp, bh->b_page, bh->b_size, bh_offset(bh));
 351 }
 352
 353 /**
 354  * gfs2_log_write_page - write one block stored in a page, into the log
 355  * @sdp: The superblock
 356  * @page: The struct page
 357  *
 358  * This writes the first block-sized part of the page into the log. Note
 359  * that the page must have been allocated from the gfs2_page_pool mempool
 360  * and that after this has been called, ownership has been transferred and
 361  * the page may be freed at any time.
 362  */
 363
 364 void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
 365 {
 366         struct super_block *sb = sdp->sd_vfs;
 367         gfs2_log_write(sdp, page, sb->s_blocksize, 0);
 368 }
 369
 370 static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
 371                                       u32 ld_length, u32 ld_data1)
 372 {
 373         struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 374         struct gfs2_log_descriptor *ld = page_address(page);
 375         clear_page(ld);
 376         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 377         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 378         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 379         ld->ld_type = cpu_to_be32(ld_type);
 380         ld->ld_length = cpu_to_be32(ld_length);
 381         ld->ld_data1 = cpu_to_be32(ld_data1);
 382         ld->ld_data2 = 0;
 383         return page;
 384 }
 385
 386 static void gfs2_check_magic(struct buffer_head *bh)
 387 {
 388         void *kaddr;
 389         __be32 *ptr;
 390
 391         clear_buffer_escaped(bh);
 392         kaddr = kmap_atomic(bh->b_page);
 393         ptr = kaddr + bh_offset(bh);
 394         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 395                 set_buffer_escaped(bh);
 396         kunmap_atomic(kaddr);
 397 }
 398
 399 static int blocknr_cmp(void *priv, struct list_head *a, struct list_head *b)
 400 {
 401         struct gfs2_bufdata *bda, *bdb;
 402
 403         bda = list_entry(a, struct gfs2_bufdata, bd_list);
 404         bdb = list_entry(b, struct gfs2_bufdata, bd_list);
 405
 406         if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
 407                 return -1;
 408         if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
 409                 return 1;
 410         return 0;
 411 }
 412
 413 static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
 414                                 unsigned int total, struct list_head *blist,
 415                                 bool is_databuf)
 416 {
 417         struct gfs2_log_descriptor *ld;
 418         struct gfs2_bufdata *bd1 = NULL, *bd2;
 419         struct page *page;
 420         unsigned int num;
 421         unsigned n;
 422         __be64 *ptr;
 423
 424         gfs2_log_lock(sdp);
 425         list_sort(NULL, blist, blocknr_cmp);
 426         bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
 427         while(total) {
 428                 num = total;
 429                 if (total > limit)
 430                         num = limit;
 431                 gfs2_log_unlock(sdp);
 432                 page = gfs2_get_log_desc(sdp,
 433                                          is_databuf ? GFS2_LOG_DESC_JDATA :
 434                                          GFS2_LOG_DESC_METADATA, num + 1, num);
 435                 ld = page_address(page);
 436                 gfs2_log_lock(sdp);
 437                 ptr = (__be64 *)(ld + 1);
 438
 439                 n = 0;
 440                 list_for_each_entry_continue(bd1, blist, bd_list) {
 441                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 442                         if (is_databuf) {
 443                                 gfs2_check_magic(bd1->bd_bh);
 444                                 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
 445                         }
 446                         if (++n >= num)
 447                                 break;
 448                 }
 449
 450                 gfs2_log_unlock(sdp);
 451                 gfs2_log_write_page(sdp, page);
 452                 gfs2_log_lock(sdp);
 453
 454                 n = 0;
 455                 list_for_each_entry_continue(bd2, blist, bd_list) {
 456                         get_bh(bd2->bd_bh);
 457                         gfs2_log_unlock(sdp);
 458                         lock_buffer(bd2->bd_bh);
 459
 460                         if (buffer_escaped(bd2->bd_bh)) {
 461                                 void *kaddr;
 462                                 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 463                                 ptr = page_address(page);
 464                                 kaddr = kmap_atomic(bd2->bd_bh->b_page);
 465                                 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh),
 466                                        bd2->bd_bh->b_size);
 467                                 kunmap_atomic(kaddr);
 468                                 *(__be32 *)ptr = 0;
 469                                 clear_buffer_escaped(bd2->bd_bh);
 470                                 unlock_buffer(bd2->bd_bh);
 471                                 brelse(bd2->bd_bh);
 472                                 gfs2_log_write_page(sdp, page);
 473                         } else {
 474                                 gfs2_log_write_bh(sdp, bd2->bd_bh);
 475                         }
 476                         gfs2_log_lock(sdp);
 477                         if (++n >= num)
 478                                 break;
 479                 }
 480
 481                 BUG_ON(total < num);
 482                 total -= num;
 483         }
 484         gfs2_log_unlock(sdp);
 485 }
 486
 487 static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 488 {
 489         unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
 490         unsigned int nbuf;
 491         if (tr == NULL)
 492                 return;
 493         nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
 494         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0);
 495 }
 496
 497 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 498 {
 499         struct list_head *head;
 500         struct gfs2_bufdata *bd;
 501
 502         if (tr == NULL)
 503                 return;
 504
 505         head = &tr->tr_buf;
 506         while (!list_empty(head)) {
 507                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 508                 list_del_init(&bd->bd_list);
 509                 gfs2_unpin(sdp, bd->bd_bh, tr);
 510         }
 511 }
 512
 513 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 514                                struct gfs2_log_header_host *head, int pass)
 515 {
 516         if (pass != 0)
 517                 return;
 518
 519         jd->jd_found_blocks = 0;
 520         jd->jd_replayed_blocks = 0;
 521 }
 522
 523 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 524                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 525                                 int pass)
 526 {
 527         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 528         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 529         struct gfs2_glock *gl = ip->i_gl;
 530         unsigned int blks = be32_to_cpu(ld->ld_data1);
 531         struct buffer_head *bh_log, *bh_ip;
 532         u64 blkno;
 533         int error = 0;
 534
 535         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 536                 return 0;
 537
 538         gfs2_replay_incr_blk(sdp, &start);
 539
 540         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 541                 blkno = be64_to_cpu(*ptr++);
 542
 543                 jd->jd_found_blocks++;
 544
 545                 if (gfs2_revoke_check(jd, blkno, start))
 546                         continue;
 547
 548                 error = gfs2_replay_read_block(jd, start, &bh_log);
 549                 if (error)
 550                         return error;
 551
 552                 bh_ip = gfs2_meta_new(gl, blkno);
 553                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 554
 555                 if (gfs2_meta_check(sdp, bh_ip))
 556                         error = -EIO;
 557                 else
 558                         mark_buffer_dirty(bh_ip);
 559
 560                 brelse(bh_log);
 561                 brelse(bh_ip);
 562
 563                 if (error)
 564                         break;
 565
 566                 jd->jd_replayed_blocks++;
 567         }
 568
 569         return error;
 570 }
 571
 572 /**
 573  * gfs2_meta_sync - Sync all buffers associated with a glock
 574  * @gl: The glock
 575  *
 576  */
 577
 578 static void gfs2_meta_sync(struct gfs2_glock *gl)
 579 {
 580         struct address_space *mapping = gfs2_glock2aspace(gl);
 581         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 582         int error;
 583
 584         if (mapping == NULL)
 585                 mapping = &sdp->sd_aspace;
 586
 587         filemap_fdatawrite(mapping);
 588         error = filemap_fdatawait(mapping);
 589
 590         if (error)
 591                 gfs2_io_error(gl->gl_name.ln_sbd);
 592 }
 593
 594 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 595 {
 596         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 597         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 598
 599         if (error) {
 600                 gfs2_meta_sync(ip->i_gl);
 601                 return;
 602         }
 603         if (pass != 1)
 604                 return;
 605
 606         gfs2_meta_sync(ip->i_gl);
 607
 608         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 609                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
 610 }
 611
 612 static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 613 {
 614         struct gfs2_meta_header *mh;
 615         unsigned int offset;
 616         struct list_head *head = &sdp->sd_log_le_revoke;
 617         struct gfs2_bufdata *bd;
 618         struct page *page;
 619         unsigned int length;
 620
 621         gfs2_write_revokes(sdp);
 622         if (!sdp->sd_log_num_revoke)
 623                 return;
 624
 625         length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(u64));
 626         page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
 627         offset = sizeof(struct gfs2_log_descriptor);
 628
 629         list_for_each_entry(bd, head, bd_list) {
 630                 sdp->sd_log_num_revoke--;
 631
 632                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 633
 634                         gfs2_log_write_page(sdp, page);
 635                         page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 636                         mh = page_address(page);
 637                         clear_page(mh);
 638                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 639                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 640                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 641                         offset = sizeof(struct gfs2_meta_header);
 642                 }
 643
 644                 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
 645                 offset += sizeof(u64);
 646         }
 647         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 648
 649         gfs2_log_write_page(sdp, page);
 650 }
 651
 652 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 653 {
 654         struct list_head *head = &sdp->sd_log_le_revoke;
 655         struct gfs2_bufdata *bd;
 656         struct gfs2_glock *gl;
 657
 658         while (!list_empty(head)) {
 659                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 660                 list_del_init(&bd->bd_list);
 661                 gl = bd->bd_gl;
 662                 atomic_dec(&gl->gl_revokes);
 663                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
 664                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 665         }
 666 }
 667
 668 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 669                                   struct gfs2_log_header_host *head, int pass)
 670 {
 671         if (pass != 0)
 672                 return;
 673
 674         jd->jd_found_revokes = 0;
 675         jd->jd_replay_tail = head->lh_tail;
 676 }
 677
 678 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 679                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 680                                    int pass)
 681 {
 682         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 683         unsigned int blks = be32_to_cpu(ld->ld_length);
 684         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 685         struct buffer_head *bh;
 686         unsigned int offset;
 687         u64 blkno;
 688         int first = 1;
 689         int error;
 690
 691         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 692                 return 0;
 693
 694         offset = sizeof(struct gfs2_log_descriptor);
 695
 696         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 697                 error = gfs2_replay_read_block(jd, start, &bh);
 698                 if (error)
 699                         return error;
 700
 701                 if (!first)
 702                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 703
 704                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 705                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 706
 707                         error = gfs2_revoke_add(jd, blkno, start);
 708                         if (error < 0) {
 709                                 brelse(bh);
 710                                 return error;
 711                         }
 712                         else if (error)
 713                                 jd->jd_found_revokes++;
 714
 715                         if (!--revokes)
 716                                 break;
 717                         offset += sizeof(u64);
 718                 }
 719
 720                 brelse(bh);
 721                 offset = sizeof(struct gfs2_meta_header);
 722                 first = 0;
 723         }
 724
 725         return 0;
 726 }
 727
 728 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 729 {
 730         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 731
 732         if (error) {
 733                 gfs2_revoke_clean(jd);
 734                 return;
 735         }
 736         if (pass != 1)
 737                 return;
 738
 739         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 740                 jd->jd_jid, jd->jd_found_revokes);
 741
 742         gfs2_revoke_clean(jd);
 743 }
 744
 745 /**
 746  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 747  *
 748  */
 749
 750 static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 751 {
 752         unsigned int limit = databuf_limit(sdp);
 753         unsigned int nbuf;
 754         if (tr == NULL)
 755                 return;
 756         nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
 757         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1);
 758 }
 759
 760 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 761                                     struct gfs2_log_descriptor *ld,
 762                                     __be64 *ptr, int pass)
 763 {
 764         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 765         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 766         struct gfs2_glock *gl = ip->i_gl;
 767         unsigned int blks = be32_to_cpu(ld->ld_data1);
 768         struct buffer_head *bh_log, *bh_ip;
 769         u64 blkno;
 770         u64 esc;
 771         int error = 0;
 772
 773         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
 774                 return 0;
 775
 776         gfs2_replay_incr_blk(sdp, &start);
 777         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 778                 blkno = be64_to_cpu(*ptr++);
 779                 esc = be64_to_cpu(*ptr++);
 780
 781                 jd->jd_found_blocks++;
 782
 783                 if (gfs2_revoke_check(jd, blkno, start))
 784                         continue;
 785
 786                 error = gfs2_replay_read_block(jd, start, &bh_log);
 787                 if (error)
 788                         return error;
 789
 790                 bh_ip = gfs2_meta_new(gl, blkno);
 791                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 792
 793                 /* Unescape */
 794                 if (esc) {
 795                         __be32 *eptr = (__be32 *)bh_ip->b_data;
 796                         *eptr = cpu_to_be32(GFS2_MAGIC);
 797                 }
 798                 mark_buffer_dirty(bh_ip);
 799
 800                 brelse(bh_log);
 801                 brelse(bh_ip);
 802
 803                 jd->jd_replayed_blocks++;
 804         }
 805
 806         return error;
 807 }
 808
 809 /* FIXME: sort out accounting for log blocks etc. */
 810
 811 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 812 {
 813         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 814         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 815
 816         if (error) {
 817                 gfs2_meta_sync(ip->i_gl);
 818                 return;
 819         }
 820         if (pass != 1)
 821                 return;
 822
 823         /* data sync? */
 824         gfs2_meta_sync(ip->i_gl);
 825
 826         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
 827                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
 828 }
 829
 830 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 831 {
 832         struct list_head *head;
 833         struct gfs2_bufdata *bd;
 834
 835         if (tr == NULL)
 836                 return;
 837
 838         head = &tr->tr_databuf;
 839         while (!list_empty(head)) {
 840                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 841                 list_del_init(&bd->bd_list);
 842                 gfs2_unpin(sdp, bd->bd_bh, tr);
 843         }
 844 }
 845
 846
 847 const struct gfs2_log_operations gfs2_buf_lops = {
 848         .lo_before_commit = buf_lo_before_commit,
 849         .lo_after_commit = buf_lo_after_commit,
 850         .lo_before_scan = buf_lo_before_scan,
 851         .lo_scan_elements = buf_lo_scan_elements,
 852         .lo_after_scan = buf_lo_after_scan,
 853         .lo_name = "buf",
 854 };
 855
 856 const struct gfs2_log_operations gfs2_revoke_lops = {
 857         .lo_before_commit = revoke_lo_before_commit,
 858         .lo_after_commit = revoke_lo_after_commit,
 859         .lo_before_scan = revoke_lo_before_scan,
 860         .lo_scan_elements = revoke_lo_scan_elements,
 861         .lo_after_scan = revoke_lo_after_scan,
 862         .lo_name = "revoke",
 863 };
 864
 865 const struct gfs2_log_operations gfs2_databuf_lops = {
 866         .lo_before_commit = databuf_lo_before_commit,
 867         .lo_after_commit = databuf_lo_after_commit,
 868         .lo_scan_elements = databuf_lo_scan_elements,
 869         .lo_after_scan = databuf_lo_after_scan,
 870         .lo_name = "databuf",
 871 };
 872
 873 const struct gfs2_log_operations *gfs2_log_ops[] = {
 874         &gfs2_databuf_lops,
 875         &gfs2_buf_lops,
 876         &gfs2_revoke_lops,
 877         NULL,
 878 };
 879