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 = gfs2_glock2rgrp(gl);
  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 }
 140
 141 static u64 gfs2_log_bmap(struct gfs2_sbd *sdp)
 142 {
 143         unsigned int lbn = sdp->sd_log_flush_head;
 144         struct gfs2_journal_extent *je;
 145         u64 block;
 146
 147         list_for_each_entry(je, &sdp->sd_jdesc->extent_list, list) {
 148                 if ((lbn >= je->lblock) && (lbn < (je->lblock + je->blocks))) {
 149                         block = je->dblock + lbn - je->lblock;
 150                         gfs2_log_incr_head(sdp);
 151                         return block;
 152                 }
 153         }
 154
 155         return -1;
 156 }
 157
 158 /**
 159  * gfs2_end_log_write_bh - end log write of pagecache data with buffers
 160  * @sdp: The superblock
 161  * @bvec: The bio_vec
 162  * @error: The i/o status
 163  *
 164  * This finds the relavent buffers and unlocks then and sets the
 165  * error flag according to the status of the i/o request. This is
 166  * used when the log is writing data which has an in-place version
 167  * that is pinned in the pagecache.
 168  */
 169
 170 static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp, struct bio_vec *bvec,
 171                                   blk_status_t error)
 172 {
 173         struct buffer_head *bh, *next;
 174         struct page *page = bvec->bv_page;
 175         unsigned size;
 176
 177         bh = page_buffers(page);
 178         size = bvec->bv_len;
 179         while (bh_offset(bh) < bvec->bv_offset)
 180                 bh = bh->b_this_page;
 181         do {
 182                 if (error)
 183                         mark_buffer_write_io_error(bh);
 184                 unlock_buffer(bh);
 185                 next = bh->b_this_page;
 186                 size -= bh->b_size;
 187                 brelse(bh);
 188                 bh = next;
 189         } while(bh && size);
 190 }
 191
 192 /**
 193  * gfs2_end_log_write - end of i/o to the log
 194  * @bio: The bio
 195  * @error: Status of i/o request
 196  *
 197  * Each bio_vec contains either data from the pagecache or data
 198  * relating to the log itself. Here we iterate over the bio_vec
 199  * array, processing both kinds of data.
 200  *
 201  */
 202
 203 static void gfs2_end_log_write(struct bio *bio)
 204 {
 205         struct gfs2_sbd *sdp = bio->bi_private;
 206         struct bio_vec *bvec;
 207         struct page *page;
 208         int i;
 209
 210         if (bio->bi_status) {
 211                 fs_err(sdp, "Error %d writing to journal, jid=%u\n",
 212                        bio->bi_status, sdp->sd_jdesc->jd_jid);
 213                 wake_up(&sdp->sd_logd_waitq);
 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, bio->bi_status);
 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  * @op: REQ_OP
 233  * @op_flags: req_flag_bits
 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 op, int op_flags)
 240 {
 241         if (sdp->sd_log_bio) {
 242                 atomic_inc(&sdp->sd_log_in_flight);
 243                 bio_set_op_attrs(sdp->sd_log_bio, op, op_flags);
 244                 submit_bio(sdp->sd_log_bio);
 245                 sdp->sd_log_bio = NULL;
 246         }
 247 }
 248
 249 /**
 250  * gfs2_log_alloc_bio - Allocate a new bio for log writing
 251  * @sdp: The superblock
 252  * @blkno: The next device block number we want to write to
 253  *
 254  * This should never be called when there is a cached bio in the
 255  * super block. When it returns, there will be a cached bio in the
 256  * super block which will have as many bio_vecs as the device is
 257  * happy to handle.
 258  *
 259  * Returns: Newly allocated bio
 260  */
 261
 262 static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno)
 263 {
 264         struct super_block *sb = sdp->sd_vfs;
 265         struct bio *bio;
 266
 267         BUG_ON(sdp->sd_log_bio);
 268
 269         bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
 270         bio->bi_iter.bi_sector = blkno * (sb->s_blocksize >> 9);
 271         bio_set_dev(bio, sb->s_bdev);
 272         bio->bi_end_io = gfs2_end_log_write;
 273         bio->bi_private = sdp;
 274
 275         sdp->sd_log_bio = bio;
 276
 277         return bio;
 278 }
 279
 280 /**
 281  * gfs2_log_get_bio - Get cached log bio, or allocate a new one
 282  * @sdp: The superblock
 283  * @blkno: The device block number we want to write to
 284  *
 285  * If there is a cached bio, then if the next block number is sequential
 286  * with the previous one, return it, otherwise flush the bio to the
 287  * device. If there is not a cached bio, or we just flushed it, then
 288  * allocate a new one.
 289  *
 290  * Returns: The bio to use for log writes
 291  */
 292
 293 static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno)
 294 {
 295         struct bio *bio = sdp->sd_log_bio;
 296         u64 nblk;
 297
 298         if (bio) {
 299                 nblk = bio_end_sector(bio);
 300                 nblk >>= sdp->sd_fsb2bb_shift;
 301                 if (blkno == nblk)
 302                         return bio;
 303                 gfs2_log_flush_bio(sdp, REQ_OP_WRITE, 0);
 304         }
 305
 306         return gfs2_log_alloc_bio(sdp, blkno);
 307 }
 308
 309
 310 /**
 311  * gfs2_log_write - write to log
 312  * @sdp: the filesystem
 313  * @page: the page to write
 314  * @size: the size of the data to write
 315  * @offset: the offset within the page
 316  *
 317  * Try and add the page segment to the current bio. If that fails,
 318  * submit the current bio to the device and create a new one, and
 319  * then add the page segment to that.
 320  */
 321
 322 static void gfs2_log_write(struct gfs2_sbd *sdp, struct page *page,
 323                            unsigned size, unsigned offset)
 324 {
 325         u64 blkno = gfs2_log_bmap(sdp);
 326         struct bio *bio;
 327         int ret;
 328
 329         bio = gfs2_log_get_bio(sdp, blkno);
 330         ret = bio_add_page(bio, page, size, offset);
 331         if (ret == 0) {
 332                 gfs2_log_flush_bio(sdp, REQ_OP_WRITE, 0);
 333                 bio = gfs2_log_alloc_bio(sdp, blkno);
 334                 ret = bio_add_page(bio, page, size, offset);
 335                 WARN_ON(ret == 0);
 336         }
 337 }
 338
 339 /**
 340  * gfs2_log_write_bh - write a buffer's content to the log
 341  * @sdp: The super block
 342  * @bh: The buffer pointing to the in-place location
 343  *
 344  * This writes the content of the buffer to the next available location
 345  * in the log. The buffer will be unlocked once the i/o to the log has
 346  * completed.
 347  */
 348
 349 static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh)
 350 {
 351         gfs2_log_write(sdp, bh->b_page, bh->b_size, bh_offset(bh));
 352 }
 353
 354 /**
 355  * gfs2_log_write_page - write one block stored in a page, into the log
 356  * @sdp: The superblock
 357  * @page: The struct page
 358  *
 359  * This writes the first block-sized part of the page into the log. Note
 360  * that the page must have been allocated from the gfs2_page_pool mempool
 361  * and that after this has been called, ownership has been transferred and
 362  * the page may be freed at any time.
 363  */
 364
 365 void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
 366 {
 367         struct super_block *sb = sdp->sd_vfs;
 368         gfs2_log_write(sdp, page, sb->s_blocksize, 0);
 369 }
 370
 371 static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
 372                                       u32 ld_length, u32 ld_data1)
 373 {
 374         struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 375         struct gfs2_log_descriptor *ld = page_address(page);
 376         clear_page(ld);
 377         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 378         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 379         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 380         ld->ld_type = cpu_to_be32(ld_type);
 381         ld->ld_length = cpu_to_be32(ld_length);
 382         ld->ld_data1 = cpu_to_be32(ld_data1);
 383         ld->ld_data2 = 0;
 384         return page;
 385 }
 386
 387 static void gfs2_check_magic(struct buffer_head *bh)
 388 {
 389         void *kaddr;
 390         __be32 *ptr;
 391
 392         clear_buffer_escaped(bh);
 393         kaddr = kmap_atomic(bh->b_page);
 394         ptr = kaddr + bh_offset(bh);
 395         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 396                 set_buffer_escaped(bh);
 397         kunmap_atomic(kaddr);
 398 }
 399
 400 static int blocknr_cmp(void *priv, struct list_head *a, struct list_head *b)
 401 {
 402         struct gfs2_bufdata *bda, *bdb;
 403
 404         bda = list_entry(a, struct gfs2_bufdata, bd_list);
 405         bdb = list_entry(b, struct gfs2_bufdata, bd_list);
 406
 407         if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
 408                 return -1;
 409         if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
 410                 return 1;
 411         return 0;
 412 }
 413
 414 static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
 415                                 unsigned int total, struct list_head *blist,
 416                                 bool is_databuf)
 417 {
 418         struct gfs2_log_descriptor *ld;
 419         struct gfs2_bufdata *bd1 = NULL, *bd2;
 420         struct page *page;
 421         unsigned int num;
 422         unsigned n;
 423         __be64 *ptr;
 424
 425         gfs2_log_lock(sdp);
 426         list_sort(NULL, blist, blocknr_cmp);
 427         bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
 428         while(total) {
 429                 num = total;
 430                 if (total > limit)
 431                         num = limit;
 432                 gfs2_log_unlock(sdp);
 433                 page = gfs2_get_log_desc(sdp,
 434                                          is_databuf ? GFS2_LOG_DESC_JDATA :
 435                                          GFS2_LOG_DESC_METADATA, num + 1, num);
 436                 ld = page_address(page);
 437                 gfs2_log_lock(sdp);
 438                 ptr = (__be64 *)(ld + 1);
 439
 440                 n = 0;
 441                 list_for_each_entry_continue(bd1, blist, bd_list) {
 442                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 443                         if (is_databuf) {
 444                                 gfs2_check_magic(bd1->bd_bh);
 445                                 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
 446                         }
 447                         if (++n >= num)
 448                                 break;
 449                 }
 450
 451                 gfs2_log_unlock(sdp);
 452                 gfs2_log_write_page(sdp, page);
 453                 gfs2_log_lock(sdp);
 454
 455                 n = 0;
 456                 list_for_each_entry_continue(bd2, blist, bd_list) {
 457                         get_bh(bd2->bd_bh);
 458                         gfs2_log_unlock(sdp);
 459                         lock_buffer(bd2->bd_bh);
 460
 461                         if (buffer_escaped(bd2->bd_bh)) {
 462                                 void *kaddr;
 463                                 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 464                                 ptr = page_address(page);
 465                                 kaddr = kmap_atomic(bd2->bd_bh->b_page);
 466                                 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh),
 467                                        bd2->bd_bh->b_size);
 468                                 kunmap_atomic(kaddr);
 469                                 *(__be32 *)ptr = 0;
 470                                 clear_buffer_escaped(bd2->bd_bh);
 471                                 unlock_buffer(bd2->bd_bh);
 472                                 brelse(bd2->bd_bh);
 473                                 gfs2_log_write_page(sdp, page);
 474                         } else {
 475                                 gfs2_log_write_bh(sdp, bd2->bd_bh);
 476                         }
 477                         gfs2_log_lock(sdp);
 478                         if (++n >= num)
 479                                 break;
 480                 }
 481
 482                 BUG_ON(total < num);
 483                 total -= num;
 484         }
 485         gfs2_log_unlock(sdp);
 486 }
 487
 488 static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 489 {
 490         unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
 491         unsigned int nbuf;
 492         if (tr == NULL)
 493                 return;
 494         nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
 495         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0);
 496 }
 497
 498 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 499 {
 500         struct list_head *head;
 501         struct gfs2_bufdata *bd;
 502
 503         if (tr == NULL)
 504                 return;
 505
 506         head = &tr->tr_buf;
 507         while (!list_empty(head)) {
 508                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 509                 list_del_init(&bd->bd_list);
 510                 gfs2_unpin(sdp, bd->bd_bh, tr);
 511         }
 512 }
 513
 514 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 515                                struct gfs2_log_header_host *head, int pass)
 516 {
 517         if (pass != 0)
 518                 return;
 519
 520         jd->jd_found_blocks = 0;
 521         jd->jd_replayed_blocks = 0;
 522 }
 523
 524 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 525                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 526                                 int pass)
 527 {
 528         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 529         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 530         struct gfs2_glock *gl = ip->i_gl;
 531         unsigned int blks = be32_to_cpu(ld->ld_data1);
 532         struct buffer_head *bh_log, *bh_ip;
 533         u64 blkno;
 534         int error = 0;
 535
 536         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 537                 return 0;
 538
 539         gfs2_replay_incr_blk(jd, &start);
 540
 541         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 542                 blkno = be64_to_cpu(*ptr++);
 543
 544                 jd->jd_found_blocks++;
 545
 546                 if (gfs2_revoke_check(jd, blkno, start))
 547                         continue;
 548
 549                 error = gfs2_replay_read_block(jd, start, &bh_log);
 550                 if (error)
 551                         return error;
 552
 553                 bh_ip = gfs2_meta_new(gl, blkno);
 554                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 555
 556                 if (gfs2_meta_check(sdp, bh_ip))
 557                         error = -EIO;
 558                 else
 559                         mark_buffer_dirty(bh_ip);
 560
 561                 brelse(bh_log);
 562                 brelse(bh_ip);
 563
 564                 if (error)
 565                         break;
 566
 567                 jd->jd_replayed_blocks++;
 568         }
 569
 570         return error;
 571 }
 572
 573 /**
 574  * gfs2_meta_sync - Sync all buffers associated with a glock
 575  * @gl: The glock
 576  *
 577  */
 578
 579 static void gfs2_meta_sync(struct gfs2_glock *gl)
 580 {
 581         struct address_space *mapping = gfs2_glock2aspace(gl);
 582         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 583         int error;
 584
 585         if (mapping == NULL)
 586                 mapping = &sdp->sd_aspace;
 587
 588         filemap_fdatawrite(mapping);
 589         error = filemap_fdatawait(mapping);
 590
 591         if (error)
 592                 gfs2_io_error(gl->gl_name.ln_sbd);
 593 }
 594
 595 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 596 {
 597         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 598         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 599
 600         if (error) {
 601                 gfs2_meta_sync(ip->i_gl);
 602                 return;
 603         }
 604         if (pass != 1)
 605                 return;
 606
 607         gfs2_meta_sync(ip->i_gl);
 608
 609         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 610                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
 611 }
 612
 613 static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 614 {
 615         struct gfs2_meta_header *mh;
 616         unsigned int offset;
 617         struct list_head *head = &sdp->sd_log_le_revoke;
 618         struct gfs2_bufdata *bd;
 619         struct page *page;
 620         unsigned int length;
 621
 622         gfs2_write_revokes(sdp);
 623         if (!sdp->sd_log_num_revoke)
 624                 return;
 625
 626         length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(u64));
 627         page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
 628         offset = sizeof(struct gfs2_log_descriptor);
 629
 630         list_for_each_entry(bd, head, bd_list) {
 631                 sdp->sd_log_num_revoke--;
 632
 633                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 634
 635                         gfs2_log_write_page(sdp, page);
 636                         page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 637                         mh = page_address(page);
 638                         clear_page(mh);
 639                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 640                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 641                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 642                         offset = sizeof(struct gfs2_meta_header);
 643                 }
 644
 645                 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
 646                 offset += sizeof(u64);
 647         }
 648         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 649
 650         gfs2_log_write_page(sdp, page);
 651 }
 652
 653 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 654 {
 655         struct list_head *head = &sdp->sd_log_le_revoke;
 656         struct gfs2_bufdata *bd;
 657         struct gfs2_glock *gl;
 658
 659         while (!list_empty(head)) {
 660                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 661                 list_del_init(&bd->bd_list);
 662                 gl = bd->bd_gl;
 663                 atomic_dec(&gl->gl_revokes);
 664                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
 665                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 666         }
 667 }
 668
 669 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 670                                   struct gfs2_log_header_host *head, int pass)
 671 {
 672         if (pass != 0)
 673                 return;
 674
 675         jd->jd_found_revokes = 0;
 676         jd->jd_replay_tail = head->lh_tail;
 677 }
 678
 679 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 680                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 681                                    int pass)
 682 {
 683         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 684         unsigned int blks = be32_to_cpu(ld->ld_length);
 685         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 686         struct buffer_head *bh;
 687         unsigned int offset;
 688         u64 blkno;
 689         int first = 1;
 690         int error;
 691
 692         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 693                 return 0;
 694
 695         offset = sizeof(struct gfs2_log_descriptor);
 696
 697         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 698                 error = gfs2_replay_read_block(jd, start, &bh);
 699                 if (error)
 700                         return error;
 701
 702                 if (!first)
 703                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 704
 705                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 706                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 707
 708                         error = gfs2_revoke_add(jd, blkno, start);
 709                         if (error < 0) {
 710                                 brelse(bh);
 711                                 return error;
 712                         }
 713                         else if (error)
 714                                 jd->jd_found_revokes++;
 715
 716                         if (!--revokes)
 717                                 break;
 718                         offset += sizeof(u64);
 719                 }
 720
 721                 brelse(bh);
 722                 offset = sizeof(struct gfs2_meta_header);
 723                 first = 0;
 724         }
 725
 726         return 0;
 727 }
 728
 729 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 730 {
 731         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 732
 733         if (error) {
 734                 gfs2_revoke_clean(jd);
 735                 return;
 736         }
 737         if (pass != 1)
 738                 return;
 739
 740         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 741                 jd->jd_jid, jd->jd_found_revokes);
 742
 743         gfs2_revoke_clean(jd);
 744 }
 745
 746 /**
 747  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 748  *
 749  */
 750
 751 static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 752 {
 753         unsigned int limit = databuf_limit(sdp);
 754         unsigned int nbuf;
 755         if (tr == NULL)
 756                 return;
 757         nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
 758         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1);
 759 }
 760
 761 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 762                                     struct gfs2_log_descriptor *ld,
 763                                     __be64 *ptr, int pass)
 764 {
 765         struct gfs2_inode *ip = GFS2_I(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(jd, &start);
 777         for (; blks; gfs2_replay_incr_blk(jd, &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