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_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 == 0)
  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 }
  87
  88 /**
  89  * gfs2_unpin - Unpin a buffer
  90  * @sdp: the filesystem the buffer belongs to
  91  * @bh: The buffer to unpin
  92  * @ai:
  93  * @flags: The inode dirty flags
  94  *
  95  */
  96
  97 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
  98                        struct gfs2_trans *tr)
  99 {
 100         struct gfs2_bufdata *bd = bh->b_private;
 101
 102         BUG_ON(!buffer_uptodate(bh));
 103         BUG_ON(!buffer_pinned(bh));
 104
 105         lock_buffer(bh);
 106         mark_buffer_dirty(bh);
 107         clear_buffer_pinned(bh);
 108
 109         if (buffer_is_rgrp(bd))
 110                 maybe_release_space(bd);
 111
 112         spin_lock(&sdp->sd_ail_lock);
 113         if (bd->bd_tr) {
 114                 list_del(&bd->bd_ail_st_list);
 115                 brelse(bh);
 116         } else {
 117                 struct gfs2_glock *gl = bd->bd_gl;
 118                 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
 119                 atomic_inc(&gl->gl_ail_count);
 120         }
 121         bd->bd_tr = tr;
 122         list_add(&bd->bd_ail_st_list, &tr->tr_ail1_list);
 123         spin_unlock(&sdp->sd_ail_lock);
 124
 125         clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 126         trace_gfs2_pin(bd, 0);
 127         unlock_buffer(bh);
 128         atomic_dec(&sdp->sd_log_pinned);
 129 }
 130
 131 static void gfs2_log_incr_head(struct gfs2_sbd *sdp)
 132 {
 133         BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) &&
 134                (sdp->sd_log_flush_head != sdp->sd_log_head));
 135
 136         if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
 137                 sdp->sd_log_flush_head = 0;
 138                 sdp->sd_log_flush_wrapped = 1;
 139         }
 140 }
 141
 142 static u64 gfs2_log_bmap(struct gfs2_sbd *sdp)
 143 {
 144         unsigned int lbn = sdp->sd_log_flush_head;
 145         struct gfs2_journal_extent *je;
 146         u64 block;
 147
 148         list_for_each_entry(je, &sdp->sd_jdesc->extent_list, extent_list) {
 149                 if (lbn >= je->lblock && lbn < je->lblock + je->blocks) {
 150                         block = je->dblock + lbn - je->lblock;
 151                         gfs2_log_incr_head(sdp);
 152                         return block;
 153                 }
 154         }
 155
 156         return -1;
 157 }
 158
 159 /**
 160  * gfs2_end_log_write_bh - end log write of pagecache data with buffers
 161  * @sdp: The superblock
 162  * @bvec: The bio_vec
 163  * @error: The i/o status
 164  *
 165  * This finds the relavent buffers and unlocks then and sets the
 166  * error flag according to the status of the i/o request. This is
 167  * used when the log is writing data which has an in-place version
 168  * that is pinned in the pagecache.
 169  */
 170
 171 static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp, struct bio_vec *bvec,
 172                                   int error)
 173 {
 174         struct buffer_head *bh, *next;
 175         struct page *page = bvec->bv_page;
 176         unsigned size;
 177
 178         bh = page_buffers(page);
 179         size = bvec->bv_len;
 180         while (bh_offset(bh) < bvec->bv_offset)
 181                 bh = bh->b_this_page;
 182         do {
 183                 if (error)
 184                         set_buffer_write_io_error(bh);
 185                 unlock_buffer(bh);
 186                 next = bh->b_this_page;
 187                 size -= bh->b_size;
 188                 brelse(bh);
 189                 bh = next;
 190         } while(bh && size);
 191 }
 192
 193 /**
 194  * gfs2_end_log_write - end of i/o to the log
 195  * @bio: The bio
 196  * @error: Status of i/o request
 197  *
 198  * Each bio_vec contains either data from the pagecache or data
 199  * relating to the log itself. Here we iterate over the bio_vec
 200  * array, processing both kinds of data.
 201  *
 202  */
 203
 204 static void gfs2_end_log_write(struct bio *bio, int error)
 205 {
 206         struct gfs2_sbd *sdp = bio->bi_private;
 207         struct bio_vec *bvec;
 208         struct page *page;
 209         int i;
 210
 211         if (error) {
 212                 sdp->sd_log_error = error;
 213                 fs_err(sdp, "Error %d writing to log\n", error);
 214         }
 215
 216         bio_for_each_segment_all(bvec, bio, i) {
 217                 page = bvec->bv_page;
 218                 if (page_has_buffers(page))
 219                         gfs2_end_log_write_bh(sdp, bvec, error);
 220                 else
 221                         mempool_free(page, gfs2_page_pool);
 222         }
 223
 224         bio_put(bio);
 225         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 226                 wake_up(&sdp->sd_log_flush_wait);
 227 }
 228
 229 /**
 230  * gfs2_log_flush_bio - Submit any pending log bio
 231  * @sdp: The superblock
 232  * @rw: The rw flags
 233  *
 234  * Submit any pending part-built or full bio to the block device. If
 235  * there is no pending bio, then this is a no-op.
 236  */
 237
 238 void gfs2_log_flush_bio(struct gfs2_sbd *sdp, int rw)
 239 {
 240         if (sdp->sd_log_bio) {
 241                 atomic_inc(&sdp->sd_log_in_flight);
 242                 submit_bio(rw, sdp->sd_log_bio);
 243                 sdp->sd_log_bio = NULL;
 244         }
 245 }
 246
 247 /**
 248  * gfs2_log_alloc_bio - Allocate a new bio for log writing
 249  * @sdp: The superblock
 250  * @blkno: The next device block number we want to write to
 251  *
 252  * This should never be called when there is a cached bio in the
 253  * super block. When it returns, there will be a cached bio in the
 254  * super block which will have as many bio_vecs as the device is
 255  * happy to handle.
 256  *
 257  * Returns: Newly allocated bio
 258  */
 259
 260 static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno)
 261 {
 262         struct super_block *sb = sdp->sd_vfs;
 263         unsigned nrvecs = bio_get_nr_vecs(sb->s_bdev);
 264         struct bio *bio;
 265
 266         BUG_ON(sdp->sd_log_bio);
 267
 268         while (1) {
 269                 bio = bio_alloc(GFP_NOIO, nrvecs);
 270                 if (likely(bio))
 271                         break;
 272                 nrvecs = max(nrvecs/2, 1U);
 273         }
 274
 275         bio->bi_sector = blkno * (sb->s_blocksize >> 9);
 276         bio->bi_bdev = sb->s_bdev;
 277         bio->bi_end_io = gfs2_end_log_write;
 278         bio->bi_private = sdp;
 279
 280         sdp->sd_log_bio = bio;
 281
 282         return bio;
 283 }
 284
 285 /**
 286  * gfs2_log_get_bio - Get cached log bio, or allocate a new one
 287  * @sdp: The superblock
 288  * @blkno: The device block number we want to write to
 289  *
 290  * If there is a cached bio, then if the next block number is sequential
 291  * with the previous one, return it, otherwise flush the bio to the
 292  * device. If there is not a cached bio, or we just flushed it, then
 293  * allocate a new one.
 294  *
 295  * Returns: The bio to use for log writes
 296  */
 297
 298 static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno)
 299 {
 300         struct bio *bio = sdp->sd_log_bio;
 301         u64 nblk;
 302
 303         if (bio) {
 304                 nblk = bio_end_sector(bio);
 305                 nblk >>= sdp->sd_fsb2bb_shift;
 306                 if (blkno == nblk)
 307                         return bio;
 308                 gfs2_log_flush_bio(sdp, WRITE);
 309         }
 310
 311         return gfs2_log_alloc_bio(sdp, blkno);
 312 }
 313
 314
 315 /**
 316  * gfs2_log_write - write to log
 317  * @sdp: the filesystem
 318  * @page: the page to write
 319  * @size: the size of the data to write
 320  * @offset: the offset within the page
 321  *
 322  * Try and add the page segment to the current bio. If that fails,
 323  * submit the current bio to the device and create a new one, and
 324  * then add the page segment to that.
 325  */
 326
 327 static void gfs2_log_write(struct gfs2_sbd *sdp, struct page *page,
 328                            unsigned size, unsigned offset)
 329 {
 330         u64 blkno = gfs2_log_bmap(sdp);
 331         struct bio *bio;
 332         int ret;
 333
 334         bio = gfs2_log_get_bio(sdp, blkno);
 335         ret = bio_add_page(bio, page, size, offset);
 336         if (ret == 0) {
 337                 gfs2_log_flush_bio(sdp, WRITE);
 338                 bio = gfs2_log_alloc_bio(sdp, blkno);
 339                 ret = bio_add_page(bio, page, size, offset);
 340                 WARN_ON(ret == 0);
 341         }
 342 }
 343
 344 /**
 345  * gfs2_log_write_bh - write a buffer's content to the log
 346  * @sdp: The super block
 347  * @bh: The buffer pointing to the in-place location
 348  *
 349  * This writes the content of the buffer to the next available location
 350  * in the log. The buffer will be unlocked once the i/o to the log has
 351  * completed.
 352  */
 353
 354 static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh)
 355 {
 356         gfs2_log_write(sdp, bh->b_page, bh->b_size, bh_offset(bh));
 357 }
 358
 359 /**
 360  * gfs2_log_write_page - write one block stored in a page, into the log
 361  * @sdp: The superblock
 362  * @page: The struct page
 363  *
 364  * This writes the first block-sized part of the page into the log. Note
 365  * that the page must have been allocated from the gfs2_page_pool mempool
 366  * and that after this has been called, ownership has been transferred and
 367  * the page may be freed at any time.
 368  */
 369
 370 void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
 371 {
 372         struct super_block *sb = sdp->sd_vfs;
 373         gfs2_log_write(sdp, page, sb->s_blocksize, 0);
 374 }
 375
 376 static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
 377                                       u32 ld_length, u32 ld_data1)
 378 {
 379         struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 380         struct gfs2_log_descriptor *ld = page_address(page);
 381         clear_page(ld);
 382         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 383         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 384         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 385         ld->ld_type = cpu_to_be32(ld_type);
 386         ld->ld_length = cpu_to_be32(ld_length);
 387         ld->ld_data1 = cpu_to_be32(ld_data1);
 388         ld->ld_data2 = 0;
 389         return page;
 390 }
 391
 392 static void gfs2_check_magic(struct buffer_head *bh)
 393 {
 394         void *kaddr;
 395         __be32 *ptr;
 396
 397         clear_buffer_escaped(bh);
 398         kaddr = kmap_atomic(bh->b_page);
 399         ptr = kaddr + bh_offset(bh);
 400         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 401                 set_buffer_escaped(bh);
 402         kunmap_atomic(kaddr);
 403 }
 404
 405 static int blocknr_cmp(void *priv, struct list_head *a, struct list_head *b)
 406 {
 407         struct gfs2_bufdata *bda, *bdb;
 408
 409         bda = list_entry(a, struct gfs2_bufdata, bd_list);
 410         bdb = list_entry(b, struct gfs2_bufdata, bd_list);
 411
 412         if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
 413                 return -1;
 414         if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
 415                 return 1;
 416         return 0;
 417 }
 418
 419 static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
 420                                 unsigned int total, struct list_head *blist,
 421                                 bool is_databuf)
 422 {
 423         struct gfs2_log_descriptor *ld;
 424         struct gfs2_bufdata *bd1 = NULL, *bd2;
 425         struct page *page;
 426         unsigned int num;
 427         unsigned n;
 428         __be64 *ptr;
 429
 430         gfs2_log_lock(sdp);
 431         list_sort(NULL, blist, blocknr_cmp);
 432         bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
 433         while(total) {
 434                 num = total;
 435                 if (total > limit)
 436                         num = limit;
 437                 gfs2_log_unlock(sdp);
 438                 page = gfs2_get_log_desc(sdp,
 439                                          is_databuf ? GFS2_LOG_DESC_JDATA :
 440                                          GFS2_LOG_DESC_METADATA, num + 1, num);
 441                 ld = page_address(page);
 442                 gfs2_log_lock(sdp);
 443                 ptr = (__be64 *)(ld + 1);
 444
 445                 n = 0;
 446                 list_for_each_entry_continue(bd1, blist, bd_list) {
 447                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 448                         if (is_databuf) {
 449                                 gfs2_check_magic(bd1->bd_bh);
 450                                 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
 451                         }
 452                         if (++n >= num)
 453                                 break;
 454                 }
 455
 456                 gfs2_log_unlock(sdp);
 457                 gfs2_log_write_page(sdp, page);
 458                 gfs2_log_lock(sdp);
 459
 460                 n = 0;
 461                 list_for_each_entry_continue(bd2, blist, bd_list) {
 462                         get_bh(bd2->bd_bh);
 463                         gfs2_log_unlock(sdp);
 464                         lock_buffer(bd2->bd_bh);
 465
 466                         if (buffer_escaped(bd2->bd_bh)) {
 467                                 void *kaddr;
 468                                 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 469                                 ptr = page_address(page);
 470                                 kaddr = kmap_atomic(bd2->bd_bh->b_page);
 471                                 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh),
 472                                        bd2->bd_bh->b_size);
 473                                 kunmap_atomic(kaddr);
 474                                 *(__be32 *)ptr = 0;
 475                                 clear_buffer_escaped(bd2->bd_bh);
 476                                 unlock_buffer(bd2->bd_bh);
 477                                 brelse(bd2->bd_bh);
 478                                 gfs2_log_write_page(sdp, page);
 479                         } else {
 480                                 gfs2_log_write_bh(sdp, bd2->bd_bh);
 481                         }
 482                         gfs2_log_lock(sdp);
 483                         if (++n >= num)
 484                                 break;
 485                 }
 486
 487                 BUG_ON(total < num);
 488                 total -= num;
 489         }
 490         gfs2_log_unlock(sdp);
 491 }
 492
 493 static void buf_lo_before_commit(struct gfs2_sbd *sdp)
 494 {
 495         unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
 496
 497         gfs2_before_commit(sdp, limit, sdp->sd_log_num_buf,
 498                            &sdp->sd_log_le_buf, 0);
 499 }
 500
 501 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 502 {
 503         struct list_head *head = &sdp->sd_log_le_buf;
 504         struct gfs2_bufdata *bd;
 505
 506         if (tr == NULL) {
 507                 gfs2_assert(sdp, list_empty(head));
 508                 return;
 509         }
 510
 511         while (!list_empty(head)) {
 512                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 513                 list_del_init(&bd->bd_list);
 514                 sdp->sd_log_num_buf--;
 515
 516                 gfs2_unpin(sdp, bd->bd_bh, tr);
 517         }
 518         gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
 519 }
 520
 521 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 522                                struct gfs2_log_header_host *head, int pass)
 523 {
 524         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 525
 526         if (pass != 0)
 527                 return;
 528
 529         sdp->sd_found_blocks = 0;
 530         sdp->sd_replayed_blocks = 0;
 531 }
 532
 533 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 534                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 535                                 int pass)
 536 {
 537         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 538         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 539         struct gfs2_glock *gl = ip->i_gl;
 540         unsigned int blks = be32_to_cpu(ld->ld_data1);
 541         struct buffer_head *bh_log, *bh_ip;
 542         u64 blkno;
 543         int error = 0;
 544
 545         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 546                 return 0;
 547
 548         gfs2_replay_incr_blk(sdp, &start);
 549
 550         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 551                 blkno = be64_to_cpu(*ptr++);
 552
 553                 sdp->sd_found_blocks++;
 554
 555                 if (gfs2_revoke_check(sdp, blkno, start))
 556                         continue;
 557
 558                 error = gfs2_replay_read_block(jd, start, &bh_log);
 559                 if (error)
 560                         return error;
 561
 562                 bh_ip = gfs2_meta_new(gl, blkno);
 563                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 564
 565                 if (gfs2_meta_check(sdp, bh_ip))
 566                         error = -EIO;
 567                 else
 568                         mark_buffer_dirty(bh_ip);
 569
 570                 brelse(bh_log);
 571                 brelse(bh_ip);
 572
 573                 if (error)
 574                         break;
 575
 576                 sdp->sd_replayed_blocks++;
 577         }
 578
 579         return error;
 580 }
 581
 582 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 583 {
 584         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 585         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 586
 587         if (error) {
 588                 gfs2_meta_sync(ip->i_gl);
 589                 return;
 590         }
 591         if (pass != 1)
 592                 return;
 593
 594         gfs2_meta_sync(ip->i_gl);
 595
 596         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 597                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 598 }
 599
 600 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
 601 {
 602         struct gfs2_meta_header *mh;
 603         unsigned int offset;
 604         struct list_head *head = &sdp->sd_log_le_revoke;
 605         struct gfs2_bufdata *bd;
 606         struct page *page;
 607         unsigned int length;
 608
 609         gfs2_write_revokes(sdp);
 610         if (!sdp->sd_log_num_revoke)
 611                 return;
 612
 613         length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(u64));
 614         page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
 615         offset = sizeof(struct gfs2_log_descriptor);
 616
 617         list_for_each_entry(bd, head, bd_list) {
 618                 sdp->sd_log_num_revoke--;
 619
 620                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 621
 622                         gfs2_log_write_page(sdp, page);
 623                         page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 624                         mh = page_address(page);
 625                         clear_page(mh);
 626                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 627                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 628                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 629                         offset = sizeof(struct gfs2_meta_header);
 630                 }
 631
 632                 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
 633                 offset += sizeof(u64);
 634         }
 635         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 636
 637         gfs2_log_write_page(sdp, page);
 638 }
 639
 640 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 641 {
 642         struct list_head *head = &sdp->sd_log_le_revoke;
 643         struct gfs2_bufdata *bd;
 644         struct gfs2_glock *gl;
 645
 646         while (!list_empty(head)) {
 647                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 648                 list_del_init(&bd->bd_list);
 649                 gl = bd->bd_gl;
 650                 atomic_dec(&gl->gl_revokes);
 651                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
 652                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 653         }
 654 }
 655
 656 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 657                                   struct gfs2_log_header_host *head, int pass)
 658 {
 659         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 660
 661         if (pass != 0)
 662                 return;
 663
 664         sdp->sd_found_revokes = 0;
 665         sdp->sd_replay_tail = head->lh_tail;
 666 }
 667
 668 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 669                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 670                                    int pass)
 671 {
 672         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 673         unsigned int blks = be32_to_cpu(ld->ld_length);
 674         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 675         struct buffer_head *bh;
 676         unsigned int offset;
 677         u64 blkno;
 678         int first = 1;
 679         int error;
 680
 681         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 682                 return 0;
 683
 684         offset = sizeof(struct gfs2_log_descriptor);
 685
 686         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 687                 error = gfs2_replay_read_block(jd, start, &bh);
 688                 if (error)
 689                         return error;
 690
 691                 if (!first)
 692                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 693
 694                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 695                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 696
 697                         error = gfs2_revoke_add(sdp, blkno, start);
 698                         if (error < 0) {
 699                                 brelse(bh);
 700                                 return error;
 701                         }
 702                         else if (error)
 703                                 sdp->sd_found_revokes++;
 704
 705                         if (!--revokes)
 706                                 break;
 707                         offset += sizeof(u64);
 708                 }
 709
 710                 brelse(bh);
 711                 offset = sizeof(struct gfs2_meta_header);
 712                 first = 0;
 713         }
 714
 715         return 0;
 716 }
 717
 718 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 719 {
 720         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 721
 722         if (error) {
 723                 gfs2_revoke_clean(sdp);
 724                 return;
 725         }
 726         if (pass != 1)
 727                 return;
 728
 729         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 730                 jd->jd_jid, sdp->sd_found_revokes);
 731
 732         gfs2_revoke_clean(sdp);
 733 }
 734
 735 /**
 736  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 737  *
 738  */
 739
 740 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
 741 {
 742         unsigned int limit = buf_limit(sdp) / 2;
 743
 744         gfs2_before_commit(sdp, limit, sdp->sd_log_num_databuf,
 745                            &sdp->sd_log_le_databuf, 1);
 746 }
 747
 748 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 749                                     struct gfs2_log_descriptor *ld,
 750                                     __be64 *ptr, int pass)
 751 {
 752         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 753         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 754         struct gfs2_glock *gl = ip->i_gl;
 755         unsigned int blks = be32_to_cpu(ld->ld_data1);
 756         struct buffer_head *bh_log, *bh_ip;
 757         u64 blkno;
 758         u64 esc;
 759         int error = 0;
 760
 761         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
 762                 return 0;
 763
 764         gfs2_replay_incr_blk(sdp, &start);
 765         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 766                 blkno = be64_to_cpu(*ptr++);
 767                 esc = be64_to_cpu(*ptr++);
 768
 769                 sdp->sd_found_blocks++;
 770
 771                 if (gfs2_revoke_check(sdp, blkno, start))
 772                         continue;
 773
 774                 error = gfs2_replay_read_block(jd, start, &bh_log);
 775                 if (error)
 776                         return error;
 777
 778                 bh_ip = gfs2_meta_new(gl, blkno);
 779                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 780
 781                 /* Unescape */
 782                 if (esc) {
 783                         __be32 *eptr = (__be32 *)bh_ip->b_data;
 784                         *eptr = cpu_to_be32(GFS2_MAGIC);
 785                 }
 786                 mark_buffer_dirty(bh_ip);
 787
 788                 brelse(bh_log);
 789                 brelse(bh_ip);
 790
 791                 sdp->sd_replayed_blocks++;
 792         }
 793
 794         return error;
 795 }
 796
 797 /* FIXME: sort out accounting for log blocks etc. */
 798
 799 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 800 {
 801         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 802         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 803
 804         if (error) {
 805                 gfs2_meta_sync(ip->i_gl);
 806                 return;
 807         }
 808         if (pass != 1)
 809                 return;
 810
 811         /* data sync? */
 812         gfs2_meta_sync(ip->i_gl);
 813
 814         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
 815                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 816 }
 817
 818 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 819 {
 820         struct list_head *head = &sdp->sd_log_le_databuf;
 821         struct gfs2_bufdata *bd;
 822
 823         if (tr == NULL) {
 824                 gfs2_assert(sdp, list_empty(head));
 825                 return;
 826         }
 827
 828         while (!list_empty(head)) {
 829                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 830                 list_del_init(&bd->bd_list);
 831                 sdp->sd_log_num_databuf--;
 832                 gfs2_unpin(sdp, bd->bd_bh, tr);
 833         }
 834         gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
 835 }
 836
 837
 838 const struct gfs2_log_operations gfs2_buf_lops = {
 839         .lo_before_commit = buf_lo_before_commit,
 840         .lo_after_commit = buf_lo_after_commit,
 841         .lo_before_scan = buf_lo_before_scan,
 842         .lo_scan_elements = buf_lo_scan_elements,
 843         .lo_after_scan = buf_lo_after_scan,
 844         .lo_name = "buf",
 845 };
 846
 847 const struct gfs2_log_operations gfs2_revoke_lops = {
 848         .lo_before_commit = revoke_lo_before_commit,
 849         .lo_after_commit = revoke_lo_after_commit,
 850         .lo_before_scan = revoke_lo_before_scan,
 851         .lo_scan_elements = revoke_lo_scan_elements,
 852         .lo_after_scan = revoke_lo_after_scan,
 853         .lo_name = "revoke",
 854 };
 855
 856 const struct gfs2_log_operations gfs2_databuf_lops = {
 857         .lo_before_commit = databuf_lo_before_commit,
 858         .lo_after_commit = databuf_lo_after_commit,
 859         .lo_scan_elements = databuf_lo_scan_elements,
 860         .lo_after_scan = databuf_lo_after_scan,
 861         .lo_name = "databuf",
 862 };
 863
 864 const struct gfs2_log_operations *gfs2_log_ops[] = {
 865         &gfs2_databuf_lops,
 866         &gfs2_buf_lops,
 867         &gfs2_revoke_lops,
 868         NULL,
 869 };
 870