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 #include <linux/blkdev.h>
  21
  22 #include "bmap.h"
  23 #include "dir.h"
  24 #include "gfs2.h"
  25 #include "incore.h"
  26 #include "inode.h"
  27 #include "glock.h"
  28 #include "log.h"
  29 #include "lops.h"
  30 #include "meta_io.h"
  31 #include "recovery.h"
  32 #include "rgrp.h"
  33 #include "trans.h"
  34 #include "util.h"
  35 #include "trace_gfs2.h"
  36
  37 /**
  38  * gfs2_pin - Pin a buffer in memory
  39  * @sdp: The superblock
  40  * @bh: The buffer to be pinned
  41  *
  42  * The log lock must be held when calling this function
  43  */
  44 void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
  45 {
  46         struct gfs2_bufdata *bd;
  47
  48         BUG_ON(!current->journal_info);
  49
  50         clear_buffer_dirty(bh);
  51         if (test_set_buffer_pinned(bh))
  52                 gfs2_assert_withdraw(sdp, 0);
  53         if (!buffer_uptodate(bh))
  54                 gfs2_io_error_bh_wd(sdp, bh);
  55         bd = bh->b_private;
  56         /* If this buffer is in the AIL and it has already been written
  57          * to in-place disk block, remove it from the AIL.
  58          */
  59         spin_lock(&sdp->sd_ail_lock);
  60         if (bd->bd_tr)
  61                 list_move(&bd->bd_ail_st_list, &bd->bd_tr->tr_ail2_list);
  62         spin_unlock(&sdp->sd_ail_lock);
  63         get_bh(bh);
  64         atomic_inc(&sdp->sd_log_pinned);
  65         trace_gfs2_pin(bd, 1);
  66 }
  67
  68 static bool buffer_is_rgrp(const struct gfs2_bufdata *bd)
  69 {
  70         return bd->bd_gl->gl_name.ln_type == LM_TYPE_RGRP;
  71 }
  72
  73 static void maybe_release_space(struct gfs2_bufdata *bd)
  74 {
  75         struct gfs2_glock *gl = bd->bd_gl;
  76         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
  77         struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
  78         unsigned int index = bd->bd_bh->b_blocknr - gl->gl_name.ln_number;
  79         struct gfs2_bitmap *bi = rgd->rd_bits + index;
  80
  81         if (bi->bi_clone == NULL)
  82                 return;
  83         if (sdp->sd_args.ar_discard)
  84                 gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi, 1, NULL);
  85         memcpy(bi->bi_clone + bi->bi_offset,
  86                bd->bd_bh->b_data + bi->bi_offset, bi->bi_bytes);
  87         clear_bit(GBF_FULL, &bi->bi_flags);
  88         rgd->rd_free_clone = rgd->rd_free;
  89         rgd->rd_extfail_pt = rgd->rd_free;
  90 }
  91
  92 /**
  93  * gfs2_unpin - Unpin a buffer
  94  * @sdp: the filesystem the buffer belongs to
  95  * @bh: The buffer to unpin
  96  * @ai:
  97  * @flags: The inode dirty flags
  98  *
  99  */
 100
 101 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
 102                        struct gfs2_trans *tr)
 103 {
 104         struct gfs2_bufdata *bd = bh->b_private;
 105
 106         BUG_ON(!buffer_uptodate(bh));
 107         BUG_ON(!buffer_pinned(bh));
 108
 109         lock_buffer(bh);
 110         mark_buffer_dirty(bh);
 111         clear_buffer_pinned(bh);
 112
 113         if (buffer_is_rgrp(bd))
 114                 maybe_release_space(bd);
 115
 116         spin_lock(&sdp->sd_ail_lock);
 117         if (bd->bd_tr) {
 118                 list_del(&bd->bd_ail_st_list);
 119                 brelse(bh);
 120         } else {
 121                 struct gfs2_glock *gl = bd->bd_gl;
 122                 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
 123                 atomic_inc(&gl->gl_ail_count);
 124         }
 125         bd->bd_tr = tr;
 126         list_add(&bd->bd_ail_st_list, &tr->tr_ail1_list);
 127         spin_unlock(&sdp->sd_ail_lock);
 128
 129         clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 130         trace_gfs2_pin(bd, 0);
 131         unlock_buffer(bh);
 132         atomic_dec(&sdp->sd_log_pinned);
 133 }
 134
 135 static void gfs2_log_incr_head(struct gfs2_sbd *sdp)
 136 {
 137         BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) &&
 138                (sdp->sd_log_flush_head != sdp->sd_log_head));
 139
 140         if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks)
 141                 sdp->sd_log_flush_head = 0;
 142 }
 143
 144 u64 gfs2_log_bmap(struct gfs2_sbd *sdp)
 145 {
 146         unsigned int lbn = sdp->sd_log_flush_head;
 147         struct gfs2_journal_extent *je;
 148         u64 block;
 149
 150         list_for_each_entry(je, &sdp->sd_jdesc->extent_list, list) {
 151                 if ((lbn >= je->lblock) && (lbn < (je->lblock + je->blocks))) {
 152                         block = je->dblock + lbn - je->lblock;
 153                         gfs2_log_incr_head(sdp);
 154                         return block;
 155                 }
 156         }
 157
 158         return -1;
 159 }
 160
 161 /**
 162  * gfs2_end_log_write_bh - end log write of pagecache data with buffers
 163  * @sdp: The superblock
 164  * @bvec: The bio_vec
 165  * @error: The i/o status
 166  *
 167  * This finds the relevant buffers and unlocks them and sets the
 168  * error flag according to the status of the i/o request. This is
 169  * used when the log is writing data which has an in-place version
 170  * that is pinned in the pagecache.
 171  */
 172
 173 static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp,
 174                                   struct bio_vec *bvec,
 175                                   blk_status_t error)
 176 {
 177         struct buffer_head *bh, *next;
 178         struct page *page = bvec->bv_page;
 179         unsigned size;
 180
 181         bh = page_buffers(page);
 182         size = bvec->bv_len;
 183         while (bh_offset(bh) < bvec->bv_offset)
 184                 bh = bh->b_this_page;
 185         do {
 186                 if (error)
 187                         mark_buffer_write_io_error(bh);
 188                 unlock_buffer(bh);
 189                 next = bh->b_this_page;
 190                 size -= bh->b_size;
 191                 brelse(bh);
 192                 bh = next;
 193         } while(bh && size);
 194 }
 195
 196 /**
 197  * gfs2_end_log_write - end of i/o to the log
 198  * @bio: The bio
 199  *
 200  * Each bio_vec contains either data from the pagecache or data
 201  * relating to the log itself. Here we iterate over the bio_vec
 202  * array, processing both kinds of data.
 203  *
 204  */
 205
 206 static void gfs2_end_log_write(struct bio *bio)
 207 {
 208         struct gfs2_sbd *sdp = bio->bi_private;
 209         struct bio_vec *bvec;
 210         struct page *page;
 211         int i;
 212         struct bvec_iter_all iter_all;
 213
 214         if (bio->bi_status) {
 215                 fs_err(sdp, "Error %d writing to journal, jid=%u\n",
 216                        bio->bi_status, sdp->sd_jdesc->jd_jid);
 217                 wake_up(&sdp->sd_logd_waitq);
 218         }
 219
 220         bio_for_each_segment_all(bvec, bio, i, iter_all) {
 221                 page = bvec->bv_page;
 222                 if (page_has_buffers(page))
 223                         gfs2_end_log_write_bh(sdp, bvec, bio->bi_status);
 224                 else
 225                         mempool_free(page, gfs2_page_pool);
 226         }
 227
 228         bio_put(bio);
 229         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 230                 wake_up(&sdp->sd_log_flush_wait);
 231 }
 232
 233 /**
 234  * gfs2_log_submit_bio - Submit any pending log bio
 235  * @biop: Address of the bio pointer
 236  * @opf: REQ_OP | op_flags
 237  *
 238  * Submit any pending part-built or full bio to the block device. If
 239  * there is no pending bio, then this is a no-op.
 240  */
 241
 242 void gfs2_log_submit_bio(struct bio **biop, int opf)
 243 {
 244         struct bio *bio = *biop;
 245         if (bio) {
 246                 struct gfs2_sbd *sdp = bio->bi_private;
 247                 atomic_inc(&sdp->sd_log_in_flight);
 248                 bio->bi_opf = opf;
 249                 submit_bio(bio);
 250                 *biop = NULL;
 251         }
 252 }
 253
 254 /**
 255  * gfs2_log_alloc_bio - Allocate a bio
 256  * @sdp: The super block
 257  * @blkno: The device block number we want to write to
 258  * @end_io: The bi_end_io callback
 259  *
 260  * Allocate a new bio, initialize it with the given parameters and return it.
 261  *
 262  * Returns: The newly allocated bio
 263  */
 264
 265 static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno,
 266                                       bio_end_io_t *end_io)
 267 {
 268         struct super_block *sb = sdp->sd_vfs;
 269         struct bio *bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
 270
 271         bio->bi_iter.bi_sector = blkno * (sb->s_blocksize >> 9);
 272         bio_set_dev(bio, sb->s_bdev);
 273         bio->bi_end_io = end_io;
 274         bio->bi_private = sdp;
 275
 276         return bio;
 277 }
 278
 279 /**
 280  * gfs2_log_get_bio - Get cached log bio, or allocate a new one
 281  * @sdp: The super block
 282  * @blkno: The device block number we want to write to
 283  * @bio: The bio to get or allocate
 284  * @op: REQ_OP
 285  * @end_io: The bi_end_io callback
 286  * @flush: Always flush the current bio and allocate a new one?
 287  *
 288  * If there is a cached bio, then if the next block number is sequential
 289  * with the previous one, return it, otherwise flush the bio to the
 290  * device. If there is no cached bio, or we just flushed it, then
 291  * allocate a new one.
 292  *
 293  * Returns: The bio to use for log writes
 294  */
 295
 296 static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno,
 297                                     struct bio **biop, int op,
 298                                     bio_end_io_t *end_io, bool flush)
 299 {
 300         struct bio *bio = *biop;
 301
 302         if (bio) {
 303                 u64 nblk;
 304
 305                 nblk = bio_end_sector(bio);
 306                 nblk >>= sdp->sd_fsb2bb_shift;
 307                 if (blkno == nblk && !flush)
 308                         return bio;
 309                 gfs2_log_submit_bio(biop, op);
 310         }
 311
 312         *biop = gfs2_log_alloc_bio(sdp, blkno, end_io);
 313         return *biop;
 314 }
 315
 316 /**
 317  * gfs2_log_write - write to log
 318  * @sdp: the filesystem
 319  * @page: the page to write
 320  * @size: the size of the data to write
 321  * @offset: the offset within the page
 322  * @blkno: block number of the log entry
 323  *
 324  * Try and add the page segment to the current bio. If that fails,
 325  * submit the current bio to the device and create a new one, and
 326  * then add the page segment to that.
 327  */
 328
 329 void gfs2_log_write(struct gfs2_sbd *sdp, struct page *page,
 330                     unsigned size, unsigned offset, u64 blkno)
 331 {
 332         struct bio *bio;
 333         int ret;
 334
 335         bio = gfs2_log_get_bio(sdp, blkno, &sdp->sd_log_bio, REQ_OP_WRITE,
 336                                gfs2_end_log_write, false);
 337         ret = bio_add_page(bio, page, size, offset);
 338         if (ret == 0) {
 339                 bio = gfs2_log_get_bio(sdp, blkno, &sdp->sd_log_bio,
 340                                        REQ_OP_WRITE, gfs2_end_log_write, true);
 341                 ret = bio_add_page(bio, page, size, offset);
 342                 WARN_ON(ret == 0);
 343         }
 344 }
 345
 346 /**
 347  * gfs2_log_write_bh - write a buffer's content to the log
 348  * @sdp: The super block
 349  * @bh: The buffer pointing to the in-place location
 350  *
 351  * This writes the content of the buffer to the next available location
 352  * in the log. The buffer will be unlocked once the i/o to the log has
 353  * completed.
 354  */
 355
 356 static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh)
 357 {
 358         gfs2_log_write(sdp, bh->b_page, bh->b_size, bh_offset(bh),
 359                        gfs2_log_bmap(sdp));
 360 }
 361
 362 /**
 363  * gfs2_log_write_page - write one block stored in a page, into the log
 364  * @sdp: The superblock
 365  * @page: The struct page
 366  *
 367  * This writes the first block-sized part of the page into the log. Note
 368  * that the page must have been allocated from the gfs2_page_pool mempool
 369  * and that after this has been called, ownership has been transferred and
 370  * the page may be freed at any time.
 371  */
 372
 373 void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
 374 {
 375         struct super_block *sb = sdp->sd_vfs;
 376         gfs2_log_write(sdp, page, sb->s_blocksize, 0,
 377                        gfs2_log_bmap(sdp));
 378 }
 379
 380 /**
 381  * gfs2_end_log_read - end I/O callback for reads from the log
 382  * @bio: The bio
 383  *
 384  * Simply unlock the pages in the bio. The main thread will wait on them and
 385  * process them in order as necessary.
 386  */
 387
 388 static void gfs2_end_log_read(struct bio *bio)
 389 {
 390         struct page *page;
 391         struct bio_vec *bvec;
 392         int i;
 393         struct bvec_iter_all iter_all;
 394
 395         bio_for_each_segment_all(bvec, bio, i, iter_all) {
 396                 page = bvec->bv_page;
 397                 if (bio->bi_status) {
 398                         int err = blk_status_to_errno(bio->bi_status);
 399
 400                         SetPageError(page);
 401                         mapping_set_error(page->mapping, err);
 402                 }
 403                 unlock_page(page);
 404         }
 405
 406         bio_put(bio);
 407 }
 408
 409 /**
 410  * gfs2_jhead_pg_srch - Look for the journal head in a given page.
 411  * @jd: The journal descriptor
 412  * @page: The page to look in
 413  *
 414  * Returns: 1 if found, 0 otherwise.
 415  */
 416
 417 static bool gfs2_jhead_pg_srch(struct gfs2_jdesc *jd,
 418                               struct gfs2_log_header_host *head,
 419                               struct page *page)
 420 {
 421         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 422         struct gfs2_log_header_host uninitialized_var(lh);
 423         void *kaddr = kmap_atomic(page);
 424         unsigned int offset;
 425         bool ret = false;
 426
 427         for (offset = 0; offset < PAGE_SIZE; offset += sdp->sd_sb.sb_bsize) {
 428                 if (!__get_log_header(sdp, kaddr + offset, 0, &lh)) {
 429                         if (lh.lh_sequence > head->lh_sequence)
 430                                 *head = lh;
 431                         else {
 432                                 ret = true;
 433                                 break;
 434                         }
 435                 }
 436         }
 437         kunmap_atomic(kaddr);
 438         return ret;
 439 }
 440
 441 /**
 442  * gfs2_jhead_process_page - Search/cleanup a page
 443  * @jd: The journal descriptor
 444  * @index: Index of the page to look into
 445  * @done: If set, perform only cleanup, else search and set if found.
 446  *
 447  * Find the page with 'index' in the journal's mapping. Search the page for
 448  * the journal head if requested (cleanup == false). Release refs on the
 449  * page so the page cache can reclaim it (put_page() twice). We grabbed a
 450  * reference on this page two times, first when we did a find_or_create_page()
 451  * to obtain the page to add it to the bio and second when we do a
 452  * find_get_page() here to get the page to wait on while I/O on it is being
 453  * completed.
 454  * This function is also used to free up a page we might've grabbed but not
 455  * used. Maybe we added it to a bio, but not submitted it for I/O. Or we
 456  * submitted the I/O, but we already found the jhead so we only need to drop
 457  * our references to the page.
 458  */
 459
 460 static void gfs2_jhead_process_page(struct gfs2_jdesc *jd, unsigned long index,
 461                                     struct gfs2_log_header_host *head,
 462                                     bool *done)
 463 {
 464         struct page *page;
 465
 466         page = find_get_page(jd->jd_inode->i_mapping, index);
 467         wait_on_page_locked(page);
 468
 469         if (PageError(page))
 470                 *done = true;
 471
 472         if (!*done)
 473                 *done = gfs2_jhead_pg_srch(jd, head, page);
 474
 475         put_page(page); /* Once for find_get_page */
 476         put_page(page); /* Once more for find_or_create_page */
 477 }
 478
 479 /**
 480  * gfs2_find_jhead - find the head of a log
 481  * @jd: The journal descriptor
 482  * @head: The log descriptor for the head of the log is returned here
 483  *
 484  * Do a search of a journal by reading it in large chunks using bios and find
 485  * the valid log entry with the highest sequence number.  (i.e. the log head)
 486  *
 487  * Returns: 0 on success, errno otherwise
 488  */
 489 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head,
 490                     bool keep_cache)
 491 {
 492         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 493         struct address_space *mapping = jd->jd_inode->i_mapping;
 494         unsigned int block = 0, blocks_submitted = 0, blocks_read = 0;
 495         unsigned int bsize = sdp->sd_sb.sb_bsize;
 496         unsigned int bsize_shift = sdp->sd_sb.sb_bsize_shift;
 497         unsigned int shift = PAGE_SHIFT - bsize_shift;
 498         unsigned int readhead_blocks = BIO_MAX_PAGES << shift;
 499         struct gfs2_journal_extent *je;
 500         int sz, ret = 0;
 501         struct bio *bio = NULL;
 502         struct page *page = NULL;
 503         bool done = false;
 504         errseq_t since;
 505
 506         memset(head, 0, sizeof(*head));
 507         if (list_empty(&jd->extent_list))
 508                 gfs2_map_journal_extents(sdp, jd);
 509
 510         since = filemap_sample_wb_err(mapping);
 511         list_for_each_entry(je, &jd->extent_list, list) {
 512                 for (; block < je->lblock + je->blocks; block++) {
 513                         u64 dblock;
 514
 515                         if (!page) {
 516                                 page = find_or_create_page(mapping,
 517                                                 block >> shift, GFP_NOFS);
 518                                 if (!page) {
 519                                         ret = -ENOMEM;
 520                                         done = true;
 521                                         goto out;
 522                                 }
 523                         }
 524
 525                         if (bio) {
 526                                 unsigned int off;
 527
 528                                 off = (block << bsize_shift) & ~PAGE_MASK;
 529                                 sz = bio_add_page(bio, page, bsize, off);
 530                                 if (sz == bsize) { /* block added */
 531                                         if (off + bsize == PAGE_SIZE) {
 532                                                 page = NULL;
 533                                                 goto page_added;
 534                                         }
 535                                         continue;
 536                                 }
 537                                 blocks_submitted = block + 1;
 538                                 submit_bio(bio);
 539                                 bio = NULL;
 540                         }
 541
 542                         dblock = je->dblock + (block - je->lblock);
 543                         bio = gfs2_log_alloc_bio(sdp, dblock, gfs2_end_log_read);
 544                         bio->bi_opf = REQ_OP_READ;
 545                         sz = bio_add_page(bio, page, bsize, 0);
 546                         gfs2_assert_warn(sdp, sz == bsize);
 547                         if (bsize == PAGE_SIZE)
 548                                 page = NULL;
 549
 550 page_added:
 551                         if (blocks_submitted < blocks_read + readhead_blocks) {
 552                                 /* Keep at least one bio in flight */
 553                                 continue;
 554                         }
 555
 556                         gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done);
 557                         blocks_read += PAGE_SIZE >> bsize_shift;
 558                         if (done)
 559                                 goto out;  /* found */
 560                 }
 561         }
 562
 563 out:
 564         if (bio)
 565                 submit_bio(bio);
 566         while (blocks_read < block) {
 567                 gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done);
 568                 blocks_read += PAGE_SIZE >> bsize_shift;
 569         }
 570
 571         if (!ret)
 572                 ret = filemap_check_wb_err(mapping, since);
 573
 574         if (!keep_cache)
 575                 truncate_inode_pages(mapping, 0);
 576
 577         return ret;
 578 }
 579
 580 static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
 581                                       u32 ld_length, u32 ld_data1)
 582 {
 583         struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 584         struct gfs2_log_descriptor *ld = page_address(page);
 585         clear_page(ld);
 586         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 587         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 588         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 589         ld->ld_type = cpu_to_be32(ld_type);
 590         ld->ld_length = cpu_to_be32(ld_length);
 591         ld->ld_data1 = cpu_to_be32(ld_data1);
 592         ld->ld_data2 = 0;
 593         return page;
 594 }
 595
 596 static void gfs2_check_magic(struct buffer_head *bh)
 597 {
 598         void *kaddr;
 599         __be32 *ptr;
 600
 601         clear_buffer_escaped(bh);
 602         kaddr = kmap_atomic(bh->b_page);
 603         ptr = kaddr + bh_offset(bh);
 604         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 605                 set_buffer_escaped(bh);
 606         kunmap_atomic(kaddr);
 607 }
 608
 609 static int blocknr_cmp(void *priv, struct list_head *a, struct list_head *b)
 610 {
 611         struct gfs2_bufdata *bda, *bdb;
 612
 613         bda = list_entry(a, struct gfs2_bufdata, bd_list);
 614         bdb = list_entry(b, struct gfs2_bufdata, bd_list);
 615
 616         if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
 617                 return -1;
 618         if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
 619                 return 1;
 620         return 0;
 621 }
 622
 623 static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
 624                                 unsigned int total, struct list_head *blist,
 625                                 bool is_databuf)
 626 {
 627         struct gfs2_log_descriptor *ld;
 628         struct gfs2_bufdata *bd1 = NULL, *bd2;
 629         struct page *page;
 630         unsigned int num;
 631         unsigned n;
 632         __be64 *ptr;
 633
 634         gfs2_log_lock(sdp);
 635         list_sort(NULL, blist, blocknr_cmp);
 636         bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
 637         while(total) {
 638                 num = total;
 639                 if (total > limit)
 640                         num = limit;
 641                 gfs2_log_unlock(sdp);
 642                 page = gfs2_get_log_desc(sdp,
 643                                          is_databuf ? GFS2_LOG_DESC_JDATA :
 644                                          GFS2_LOG_DESC_METADATA, num + 1, num);
 645                 ld = page_address(page);
 646                 gfs2_log_lock(sdp);
 647                 ptr = (__be64 *)(ld + 1);
 648
 649                 n = 0;
 650                 list_for_each_entry_continue(bd1, blist, bd_list) {
 651                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 652                         if (is_databuf) {
 653                                 gfs2_check_magic(bd1->bd_bh);
 654                                 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
 655                         }
 656                         if (++n >= num)
 657                                 break;
 658                 }
 659
 660                 gfs2_log_unlock(sdp);
 661                 gfs2_log_write_page(sdp, page);
 662                 gfs2_log_lock(sdp);
 663
 664                 n = 0;
 665                 list_for_each_entry_continue(bd2, blist, bd_list) {
 666                         get_bh(bd2->bd_bh);
 667                         gfs2_log_unlock(sdp);
 668                         lock_buffer(bd2->bd_bh);
 669
 670                         if (buffer_escaped(bd2->bd_bh)) {
 671                                 void *kaddr;
 672                                 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 673                                 ptr = page_address(page);
 674                                 kaddr = kmap_atomic(bd2->bd_bh->b_page);
 675                                 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh),
 676                                        bd2->bd_bh->b_size);
 677                                 kunmap_atomic(kaddr);
 678                                 *(__be32 *)ptr = 0;
 679                                 clear_buffer_escaped(bd2->bd_bh);
 680                                 unlock_buffer(bd2->bd_bh);
 681                                 brelse(bd2->bd_bh);
 682                                 gfs2_log_write_page(sdp, page);
 683                         } else {
 684                                 gfs2_log_write_bh(sdp, bd2->bd_bh);
 685                         }
 686                         gfs2_log_lock(sdp);
 687                         if (++n >= num)
 688                                 break;
 689                 }
 690
 691                 BUG_ON(total < num);
 692                 total -= num;
 693         }
 694         gfs2_log_unlock(sdp);
 695 }
 696
 697 static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 698 {
 699         unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
 700         unsigned int nbuf;
 701         if (tr == NULL)
 702                 return;
 703         nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
 704         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0);
 705 }
 706
 707 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 708 {
 709         struct list_head *head;
 710         struct gfs2_bufdata *bd;
 711
 712         if (tr == NULL)
 713                 return;
 714
 715         head = &tr->tr_buf;
 716         while (!list_empty(head)) {
 717                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 718                 list_del_init(&bd->bd_list);
 719                 gfs2_unpin(sdp, bd->bd_bh, tr);
 720         }
 721 }
 722
 723 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 724                                struct gfs2_log_header_host *head, int pass)
 725 {
 726         if (pass != 0)
 727                 return;
 728
 729         jd->jd_found_blocks = 0;
 730         jd->jd_replayed_blocks = 0;
 731 }
 732
 733 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
 734                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 735                                 int pass)
 736 {
 737         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 738         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 739         struct gfs2_glock *gl = ip->i_gl;
 740         unsigned int blks = be32_to_cpu(ld->ld_data1);
 741         struct buffer_head *bh_log, *bh_ip;
 742         u64 blkno;
 743         int error = 0;
 744
 745         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 746                 return 0;
 747
 748         gfs2_replay_incr_blk(jd, &start);
 749
 750         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 751                 blkno = be64_to_cpu(*ptr++);
 752
 753                 jd->jd_found_blocks++;
 754
 755                 if (gfs2_revoke_check(jd, blkno, start))
 756                         continue;
 757
 758                 error = gfs2_replay_read_block(jd, start, &bh_log);
 759                 if (error)
 760                         return error;
 761
 762                 bh_ip = gfs2_meta_new(gl, blkno);
 763                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 764
 765                 if (gfs2_meta_check(sdp, bh_ip))
 766                         error = -EIO;
 767                 else
 768                         mark_buffer_dirty(bh_ip);
 769
 770                 brelse(bh_log);
 771                 brelse(bh_ip);
 772
 773                 if (error)
 774                         break;
 775
 776                 jd->jd_replayed_blocks++;
 777         }
 778
 779         return error;
 780 }
 781
 782 /**
 783  * gfs2_meta_sync - Sync all buffers associated with a glock
 784  * @gl: The glock
 785  *
 786  */
 787
 788 static void gfs2_meta_sync(struct gfs2_glock *gl)
 789 {
 790         struct address_space *mapping = gfs2_glock2aspace(gl);
 791         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 792         int error;
 793
 794         if (mapping == NULL)
 795                 mapping = &sdp->sd_aspace;
 796
 797         filemap_fdatawrite(mapping);
 798         error = filemap_fdatawait(mapping);
 799
 800         if (error)
 801                 gfs2_io_error(gl->gl_name.ln_sbd);
 802 }
 803
 804 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 805 {
 806         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 807         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 808
 809         if (error) {
 810                 gfs2_meta_sync(ip->i_gl);
 811                 return;
 812         }
 813         if (pass != 1)
 814                 return;
 815
 816         gfs2_meta_sync(ip->i_gl);
 817
 818         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 819                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
 820 }
 821
 822 static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 823 {
 824         struct gfs2_meta_header *mh;
 825         unsigned int offset;
 826         struct list_head *head = &sdp->sd_log_revokes;
 827         struct gfs2_bufdata *bd;
 828         struct page *page;
 829         unsigned int length;
 830
 831         gfs2_write_revokes(sdp);
 832         if (!sdp->sd_log_num_revoke)
 833                 return;
 834
 835         length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(u64));
 836         page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
 837         offset = sizeof(struct gfs2_log_descriptor);
 838
 839         list_for_each_entry(bd, head, bd_list) {
 840                 sdp->sd_log_num_revoke--;
 841
 842                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 843
 844                         gfs2_log_write_page(sdp, page);
 845                         page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 846                         mh = page_address(page);
 847                         clear_page(mh);
 848                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 849                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 850                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 851                         offset = sizeof(struct gfs2_meta_header);
 852                 }
 853
 854                 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
 855                 offset += sizeof(u64);
 856         }
 857         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 858
 859         gfs2_log_write_page(sdp, page);
 860 }
 861
 862 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 863 {
 864         struct list_head *head = &sdp->sd_log_revokes;
 865         struct gfs2_bufdata *bd, *tmp;
 866
 867         /*
 868          * Glocks can be referenced repeatedly on the revoke list, but the list
 869          * only holds one reference.  All glocks on the list will have the
 870          * GLF_REVOKES flag set initially.
 871          */
 872
 873         list_for_each_entry_safe(bd, tmp, head, bd_list) {
 874                 struct gfs2_glock *gl = bd->bd_gl;
 875
 876                 if (test_bit(GLF_REVOKES, &gl->gl_flags)) {
 877                         /* Keep each glock on the list exactly once. */
 878                         clear_bit(GLF_REVOKES, &gl->gl_flags);
 879                         continue;
 880                 }
 881                 list_del(&bd->bd_list);
 882                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 883         }
 884         list_for_each_entry_safe(bd, tmp, head, bd_list) {
 885                 struct gfs2_glock *gl = bd->bd_gl;
 886
 887                 list_del(&bd->bd_list);
 888                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 889                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
 890                 gfs2_glock_queue_put(gl);
 891         }
 892         /* the list is empty now */
 893 }
 894
 895 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 896                                   struct gfs2_log_header_host *head, int pass)
 897 {
 898         if (pass != 0)
 899                 return;
 900
 901         jd->jd_found_revokes = 0;
 902         jd->jd_replay_tail = head->lh_tail;
 903 }
 904
 905 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
 906                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 907                                    int pass)
 908 {
 909         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 910         unsigned int blks = be32_to_cpu(ld->ld_length);
 911         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 912         struct buffer_head *bh;
 913         unsigned int offset;
 914         u64 blkno;
 915         int first = 1;
 916         int error;
 917
 918         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 919                 return 0;
 920
 921         offset = sizeof(struct gfs2_log_descriptor);
 922
 923         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 924                 error = gfs2_replay_read_block(jd, start, &bh);
 925                 if (error)
 926                         return error;
 927
 928                 if (!first)
 929                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 930
 931                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 932                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 933
 934                         error = gfs2_revoke_add(jd, blkno, start);
 935                         if (error < 0) {
 936                                 brelse(bh);
 937                                 return error;
 938                         }
 939                         else if (error)
 940                                 jd->jd_found_revokes++;
 941
 942                         if (!--revokes)
 943                                 break;
 944                         offset += sizeof(u64);
 945                 }
 946
 947                 brelse(bh);
 948                 offset = sizeof(struct gfs2_meta_header);
 949                 first = 0;
 950         }
 951
 952         return 0;
 953 }
 954
 955 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 956 {
 957         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 958
 959         if (error) {
 960                 gfs2_revoke_clean(jd);
 961                 return;
 962         }
 963         if (pass != 1)
 964                 return;
 965
 966         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 967                 jd->jd_jid, jd->jd_found_revokes);
 968
 969         gfs2_revoke_clean(jd);
 970 }
 971
 972 /**
 973  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 974  *
 975  */
 976
 977 static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 978 {
 979         unsigned int limit = databuf_limit(sdp);
 980         unsigned int nbuf;
 981         if (tr == NULL)
 982                 return;
 983         nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
 984         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1);
 985 }
 986
 987 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
 988                                     struct gfs2_log_descriptor *ld,
 989                                     __be64 *ptr, int pass)
 990 {
 991         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 992         struct gfs2_glock *gl = ip->i_gl;
 993         unsigned int blks = be32_to_cpu(ld->ld_data1);
 994         struct buffer_head *bh_log, *bh_ip;
 995         u64 blkno;
 996         u64 esc;
 997         int error = 0;
 998
 999         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
1000                 return 0;
1001
1002         gfs2_replay_incr_blk(jd, &start);
1003         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
1004                 blkno = be64_to_cpu(*ptr++);
1005                 esc = be64_to_cpu(*ptr++);
1006
1007                 jd->jd_found_blocks++;
1008
1009                 if (gfs2_revoke_check(jd, blkno, start))
1010                         continue;
1011
1012                 error = gfs2_replay_read_block(jd, start, &bh_log);
1013                 if (error)
1014                         return error;
1015
1016                 bh_ip = gfs2_meta_new(gl, blkno);
1017                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
1018
1019                 /* Unescape */
1020                 if (esc) {
1021                         __be32 *eptr = (__be32 *)bh_ip->b_data;
1022                         *eptr = cpu_to_be32(GFS2_MAGIC);
1023                 }
1024                 mark_buffer_dirty(bh_ip);
1025
1026                 brelse(bh_log);
1027                 brelse(bh_ip);
1028
1029                 jd->jd_replayed_blocks++;
1030         }
1031
1032         return error;
1033 }
1034
1035 /* FIXME: sort out accounting for log blocks etc. */
1036
1037 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
1038 {
1039         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1040         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
1041
1042         if (error) {
1043                 gfs2_meta_sync(ip->i_gl);
1044                 return;
1045         }
1046         if (pass != 1)
1047                 return;
1048
1049         /* data sync? */
1050         gfs2_meta_sync(ip->i_gl);
1051
1052         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
1053                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
1054 }
1055
1056 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1057 {
1058         struct list_head *head;
1059         struct gfs2_bufdata *bd;
1060
1061         if (tr == NULL)
1062                 return;
1063
1064         head = &tr->tr_databuf;
1065         while (!list_empty(head)) {
1066                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
1067                 list_del_init(&bd->bd_list);
1068                 gfs2_unpin(sdp, bd->bd_bh, tr);
1069         }
1070 }
1071
1072
1073 static const struct gfs2_log_operations gfs2_buf_lops = {
1074         .lo_before_commit = buf_lo_before_commit,
1075         .lo_after_commit = buf_lo_after_commit,
1076         .lo_before_scan = buf_lo_before_scan,
1077         .lo_scan_elements = buf_lo_scan_elements,
1078         .lo_after_scan = buf_lo_after_scan,
1079         .lo_name = "buf",
1080 };
1081
1082 static const struct gfs2_log_operations gfs2_revoke_lops = {
1083         .lo_before_commit = revoke_lo_before_commit,
1084         .lo_after_commit = revoke_lo_after_commit,
1085         .lo_before_scan = revoke_lo_before_scan,
1086         .lo_scan_elements = revoke_lo_scan_elements,
1087         .lo_after_scan = revoke_lo_after_scan,
1088         .lo_name = "revoke",
1089 };
1090
1091 static const struct gfs2_log_operations gfs2_databuf_lops = {
1092         .lo_before_commit = databuf_lo_before_commit,
1093         .lo_after_commit = databuf_lo_after_commit,
1094         .lo_scan_elements = databuf_lo_scan_elements,
1095         .lo_after_scan = databuf_lo_after_scan,
1096         .lo_name = "databuf",
1097 };
1098
1099 const struct gfs2_log_operations *gfs2_log_ops[] = {
1100         &gfs2_databuf_lops,
1101         &gfs2_buf_lops,
1102         &gfs2_revoke_lops,
1103         NULL,
1104 };
1105