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 "dir.h"
  22 #include "gfs2.h"
  23 #include "incore.h"
  24 #include "inode.h"
  25 #include "glock.h"
  26 #include "log.h"
  27 #include "lops.h"
  28 #include "meta_io.h"
  29 #include "recovery.h"
  30 #include "rgrp.h"
  31 #include "trans.h"
  32 #include "util.h"
  33 #include "trace_gfs2.h"
  34
  35 /**
  36  * gfs2_pin - Pin a buffer in memory
  37  * @sdp: The superblock
  38  * @bh: The buffer to be pinned
  39  *
  40  * The log lock must be held when calling this function
  41  */
  42 void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
  43 {
  44         struct gfs2_bufdata *bd;
  45
  46         BUG_ON(!current->journal_info);
  47
  48         clear_buffer_dirty(bh);
  49         if (test_set_buffer_pinned(bh))
  50                 gfs2_assert_withdraw(sdp, 0);
  51         if (!buffer_uptodate(bh))
  52                 gfs2_io_error_bh_wd(sdp, bh);
  53         bd = bh->b_private;
  54         /* If this buffer is in the AIL and it has already been written
  55          * to in-place disk block, remove it from the AIL.
  56          */
  57         spin_lock(&sdp->sd_ail_lock);
  58         if (bd->bd_tr)
  59                 list_move(&bd->bd_ail_st_list, &bd->bd_tr->tr_ail2_list);
  60         spin_unlock(&sdp->sd_ail_lock);
  61         get_bh(bh);
  62         atomic_inc(&sdp->sd_log_pinned);
  63         trace_gfs2_pin(bd, 1);
  64 }
  65
  66 static bool buffer_is_rgrp(const struct gfs2_bufdata *bd)
  67 {
  68         return bd->bd_gl->gl_name.ln_type == LM_TYPE_RGRP;
  69 }
  70
  71 static void maybe_release_space(struct gfs2_bufdata *bd)
  72 {
  73         struct gfs2_glock *gl = bd->bd_gl;
  74         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
  75         struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
  76         unsigned int index = bd->bd_bh->b_blocknr - gl->gl_name.ln_number;
  77         struct gfs2_bitmap *bi = rgd->rd_bits + index;
  78
  79         if (bi->bi_clone == NULL)
  80                 return;
  81         if (sdp->sd_args.ar_discard)
  82                 gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi, 1, NULL);
  83         memcpy(bi->bi_clone + bi->bi_offset,
  84                bd->bd_bh->b_data + bi->bi_offset, bi->bi_bytes);
  85         clear_bit(GBF_FULL, &bi->bi_flags);
  86         rgd->rd_free_clone = rgd->rd_free;
  87         rgd->rd_extfail_pt = rgd->rd_free;
  88 }
  89
  90 /**
  91  * gfs2_unpin - Unpin a buffer
  92  * @sdp: the filesystem the buffer belongs to
  93  * @bh: The buffer to unpin
  94  * @ai:
  95  * @flags: The inode dirty flags
  96  *
  97  */
  98
  99 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
 100                        struct gfs2_trans *tr)
 101 {
 102         struct gfs2_bufdata *bd = bh->b_private;
 103
 104         BUG_ON(!buffer_uptodate(bh));
 105         BUG_ON(!buffer_pinned(bh));
 106
 107         lock_buffer(bh);
 108         mark_buffer_dirty(bh);
 109         clear_buffer_pinned(bh);
 110
 111         if (buffer_is_rgrp(bd))
 112                 maybe_release_space(bd);
 113
 114         spin_lock(&sdp->sd_ail_lock);
 115         if (bd->bd_tr) {
 116                 list_del(&bd->bd_ail_st_list);
 117                 brelse(bh);
 118         } else {
 119                 struct gfs2_glock *gl = bd->bd_gl;
 120                 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
 121                 atomic_inc(&gl->gl_ail_count);
 122         }
 123         bd->bd_tr = tr;
 124         list_add(&bd->bd_ail_st_list, &tr->tr_ail1_list);
 125         spin_unlock(&sdp->sd_ail_lock);
 126
 127         clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 128         trace_gfs2_pin(bd, 0);
 129         unlock_buffer(bh);
 130         atomic_dec(&sdp->sd_log_pinned);
 131 }
 132
 133 static void gfs2_log_incr_head(struct gfs2_sbd *sdp)
 134 {
 135         BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) &&
 136                (sdp->sd_log_flush_head != sdp->sd_log_head));
 137
 138         if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks)
 139                 sdp->sd_log_flush_head = 0;
 140 }
 141
 142 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, 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 relevant buffers and unlocks them 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                                   blk_status_t 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                         mark_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)
 205 {
 206         struct gfs2_sbd *sdp = bio->bi_private;
 207         struct bio_vec *bvec;
 208         struct page *page;
 209         int i;
 210
 211         if (bio->bi_status) {
 212                 fs_err(sdp, "Error %d writing to journal, jid=%u\n",
 213                        bio->bi_status, sdp->sd_jdesc->jd_jid);
 214                 wake_up(&sdp->sd_logd_waitq);
 215         }
 216
 217         bio_for_each_segment_all(bvec, bio, i) {
 218                 page = bvec->bv_page;
 219                 if (page_has_buffers(page))
 220                         gfs2_end_log_write_bh(sdp, bvec, bio->bi_status);
 221                 else
 222                         mempool_free(page, gfs2_page_pool);
 223         }
 224
 225         bio_put(bio);
 226         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 227                 wake_up(&sdp->sd_log_flush_wait);
 228 }
 229
 230 /**
 231  * gfs2_log_submit_bio - Submit any pending log bio
 232  * @biop: Address of the bio pointer
 233  * @op: REQ_OP
 234  * @op_flags: req_flag_bits
 235  *
 236  * Submit any pending part-built or full bio to the block device. If
 237  * there is no pending bio, then this is a no-op.
 238  */
 239
 240 void gfs2_log_submit_bio(struct bio **biop, int op, int op_flags)
 241 {
 242         struct bio *bio = *biop;
 243         if (bio) {
 244                 struct gfs2_sbd *sdp = bio->bi_private;
 245                 atomic_inc(&sdp->sd_log_in_flight);
 246                 bio_set_op_attrs(bio, op, op_flags);
 247                 submit_bio(bio);
 248                 *biop = NULL;
 249         }
 250 }
 251
 252 /**
 253  * gfs2_log_alloc_bio - Allocate a bio
 254  * @sdp: The super block
 255  * @blkno: The device block number we want to write to
 256  * @end_io: The bi_end_io callback
 257  *
 258  * Allocate a new bio, initialize it with the given parameters and return it.
 259  *
 260  * Returns: The newly allocated bio
 261  */
 262
 263 static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno,
 264                                       bio_end_io_t *end_io)
 265 {
 266         struct super_block *sb = sdp->sd_vfs;
 267         struct bio *bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
 268
 269         bio->bi_iter.bi_sector = blkno * (sb->s_blocksize >> 9);
 270         bio_set_dev(bio, sb->s_bdev);
 271         bio->bi_end_io = end_io;
 272         bio->bi_private = sdp;
 273
 274         return bio;
 275 }
 276
 277 /**
 278  * gfs2_log_get_bio - Get cached log bio, or allocate a new one
 279  * @sdp: The super block
 280  * @blkno: The device block number we want to write to
 281  * @bio: The bio to get or allocate
 282  * @op: REQ_OP
 283  * @end_io: The bi_end_io callback
 284  * @flush: Always flush the current bio and allocate a new one?
 285  *
 286  * If there is a cached bio, then if the next block number is sequential
 287  * with the previous one, return it, otherwise flush the bio to the
 288  * device. If there is no cached bio, or we just flushed it, then
 289  * allocate a new one.
 290  *
 291  * Returns: The bio to use for log writes
 292  */
 293
 294 static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno,
 295                                     struct bio **biop, int op,
 296                                     bio_end_io_t *end_io, bool flush)
 297 {
 298         struct bio *bio = *biop;
 299
 300         if (bio) {
 301                 u64 nblk;
 302
 303                 nblk = bio_end_sector(bio);
 304                 nblk >>= sdp->sd_fsb2bb_shift;
 305                 if (blkno == nblk && !flush)
 306                         return bio;
 307                 gfs2_log_submit_bio(biop, op, 0);
 308         }
 309
 310         *biop = gfs2_log_alloc_bio(sdp, blkno, end_io);
 311         return *biop;
 312 }
 313
 314 /**
 315  * gfs2_log_write - write to log
 316  * @sdp: the filesystem
 317  * @page: the page to write
 318  * @size: the size of the data to write
 319  * @offset: the offset within the page
 320  * @blkno: block number of the log entry
 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 void gfs2_log_write(struct gfs2_sbd *sdp, struct page *page,
 328                     unsigned size, unsigned offset, u64 blkno)
 329 {
 330         struct bio *bio;
 331         int ret;
 332
 333         bio = gfs2_log_get_bio(sdp, blkno, &sdp->sd_log_bio, REQ_OP_WRITE,
 334                                gfs2_end_log_write, false);
 335         ret = bio_add_page(bio, page, size, offset);
 336         if (ret == 0) {
 337                 bio = gfs2_log_get_bio(sdp, blkno, &sdp->sd_log_bio,
 338                                        REQ_OP_WRITE, gfs2_end_log_write, true);
 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                        gfs2_log_bmap(sdp));
 358 }
 359
 360 /**
 361  * gfs2_log_write_page - write one block stored in a page, into the log
 362  * @sdp: The superblock
 363  * @page: The struct page
 364  *
 365  * This writes the first block-sized part of the page into the log. Note
 366  * that the page must have been allocated from the gfs2_page_pool mempool
 367  * and that after this has been called, ownership has been transferred and
 368  * the page may be freed at any time.
 369  */
 370
 371 void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
 372 {
 373         struct super_block *sb = sdp->sd_vfs;
 374         gfs2_log_write(sdp, page, sb->s_blocksize, 0,
 375                        gfs2_log_bmap(sdp));
 376 }
 377
 378 static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
 379                                       u32 ld_length, u32 ld_data1)
 380 {
 381         struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 382         struct gfs2_log_descriptor *ld = page_address(page);
 383         clear_page(ld);
 384         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 385         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 386         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 387         ld->ld_type = cpu_to_be32(ld_type);
 388         ld->ld_length = cpu_to_be32(ld_length);
 389         ld->ld_data1 = cpu_to_be32(ld_data1);
 390         ld->ld_data2 = 0;
 391         return page;
 392 }
 393
 394 static void gfs2_check_magic(struct buffer_head *bh)
 395 {
 396         void *kaddr;
 397         __be32 *ptr;
 398
 399         clear_buffer_escaped(bh);
 400         kaddr = kmap_atomic(bh->b_page);
 401         ptr = kaddr + bh_offset(bh);
 402         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 403                 set_buffer_escaped(bh);
 404         kunmap_atomic(kaddr);
 405 }
 406
 407 static int blocknr_cmp(void *priv, struct list_head *a, struct list_head *b)
 408 {
 409         struct gfs2_bufdata *bda, *bdb;
 410
 411         bda = list_entry(a, struct gfs2_bufdata, bd_list);
 412         bdb = list_entry(b, struct gfs2_bufdata, bd_list);
 413
 414         if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
 415                 return -1;
 416         if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
 417                 return 1;
 418         return 0;
 419 }
 420
 421 static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
 422                                 unsigned int total, struct list_head *blist,
 423                                 bool is_databuf)
 424 {
 425         struct gfs2_log_descriptor *ld;
 426         struct gfs2_bufdata *bd1 = NULL, *bd2;
 427         struct page *page;
 428         unsigned int num;
 429         unsigned n;
 430         __be64 *ptr;
 431
 432         gfs2_log_lock(sdp);
 433         list_sort(NULL, blist, blocknr_cmp);
 434         bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
 435         while(total) {
 436                 num = total;
 437                 if (total > limit)
 438                         num = limit;
 439                 gfs2_log_unlock(sdp);
 440                 page = gfs2_get_log_desc(sdp,
 441                                          is_databuf ? GFS2_LOG_DESC_JDATA :
 442                                          GFS2_LOG_DESC_METADATA, num + 1, num);
 443                 ld = page_address(page);
 444                 gfs2_log_lock(sdp);
 445                 ptr = (__be64 *)(ld + 1);
 446
 447                 n = 0;
 448                 list_for_each_entry_continue(bd1, blist, bd_list) {
 449                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 450                         if (is_databuf) {
 451                                 gfs2_check_magic(bd1->bd_bh);
 452                                 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
 453                         }
 454                         if (++n >= num)
 455                                 break;
 456                 }
 457
 458                 gfs2_log_unlock(sdp);
 459                 gfs2_log_write_page(sdp, page);
 460                 gfs2_log_lock(sdp);
 461
 462                 n = 0;
 463                 list_for_each_entry_continue(bd2, blist, bd_list) {
 464                         get_bh(bd2->bd_bh);
 465                         gfs2_log_unlock(sdp);
 466                         lock_buffer(bd2->bd_bh);
 467
 468                         if (buffer_escaped(bd2->bd_bh)) {
 469                                 void *kaddr;
 470                                 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 471                                 ptr = page_address(page);
 472                                 kaddr = kmap_atomic(bd2->bd_bh->b_page);
 473                                 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh),
 474                                        bd2->bd_bh->b_size);
 475                                 kunmap_atomic(kaddr);
 476                                 *(__be32 *)ptr = 0;
 477                                 clear_buffer_escaped(bd2->bd_bh);
 478                                 unlock_buffer(bd2->bd_bh);
 479                                 brelse(bd2->bd_bh);
 480                                 gfs2_log_write_page(sdp, page);
 481                         } else {
 482                                 gfs2_log_write_bh(sdp, bd2->bd_bh);
 483                         }
 484                         gfs2_log_lock(sdp);
 485                         if (++n >= num)
 486                                 break;
 487                 }
 488
 489                 BUG_ON(total < num);
 490                 total -= num;
 491         }
 492         gfs2_log_unlock(sdp);
 493 }
 494
 495 static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 496 {
 497         unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
 498         unsigned int nbuf;
 499         if (tr == NULL)
 500                 return;
 501         nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
 502         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0);
 503 }
 504
 505 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 506 {
 507         struct list_head *head;
 508         struct gfs2_bufdata *bd;
 509
 510         if (tr == NULL)
 511                 return;
 512
 513         head = &tr->tr_buf;
 514         while (!list_empty(head)) {
 515                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 516                 list_del_init(&bd->bd_list);
 517                 gfs2_unpin(sdp, bd->bd_bh, tr);
 518         }
 519 }
 520
 521 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 522                                struct gfs2_log_header_host *head, int pass)
 523 {
 524         if (pass != 0)
 525                 return;
 526
 527         jd->jd_found_blocks = 0;
 528         jd->jd_replayed_blocks = 0;
 529 }
 530
 531 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 532                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 533                                 int pass)
 534 {
 535         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 536         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 537         struct gfs2_glock *gl = ip->i_gl;
 538         unsigned int blks = be32_to_cpu(ld->ld_data1);
 539         struct buffer_head *bh_log, *bh_ip;
 540         u64 blkno;
 541         int error = 0;
 542
 543         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 544                 return 0;
 545
 546         gfs2_replay_incr_blk(jd, &start);
 547
 548         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 549                 blkno = be64_to_cpu(*ptr++);
 550
 551                 jd->jd_found_blocks++;
 552
 553                 if (gfs2_revoke_check(jd, blkno, start))
 554                         continue;
 555
 556                 error = gfs2_replay_read_block(jd, start, &bh_log);
 557                 if (error)
 558                         return error;
 559
 560                 bh_ip = gfs2_meta_new(gl, blkno);
 561                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 562
 563                 if (gfs2_meta_check(sdp, bh_ip))
 564                         error = -EIO;
 565                 else
 566                         mark_buffer_dirty(bh_ip);
 567
 568                 brelse(bh_log);
 569                 brelse(bh_ip);
 570
 571                 if (error)
 572                         break;
 573
 574                 jd->jd_replayed_blocks++;
 575         }
 576
 577         return error;
 578 }
 579
 580 /**
 581  * gfs2_meta_sync - Sync all buffers associated with a glock
 582  * @gl: The glock
 583  *
 584  */
 585
 586 static void gfs2_meta_sync(struct gfs2_glock *gl)
 587 {
 588         struct address_space *mapping = gfs2_glock2aspace(gl);
 589         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 590         int error;
 591
 592         if (mapping == NULL)
 593                 mapping = &sdp->sd_aspace;
 594
 595         filemap_fdatawrite(mapping);
 596         error = filemap_fdatawait(mapping);
 597
 598         if (error)
 599                 gfs2_io_error(gl->gl_name.ln_sbd);
 600 }
 601
 602 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 603 {
 604         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 605         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 606
 607         if (error) {
 608                 gfs2_meta_sync(ip->i_gl);
 609                 return;
 610         }
 611         if (pass != 1)
 612                 return;
 613
 614         gfs2_meta_sync(ip->i_gl);
 615
 616         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 617                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
 618 }
 619
 620 static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 621 {
 622         struct gfs2_meta_header *mh;
 623         unsigned int offset;
 624         struct list_head *head = &sdp->sd_log_le_revoke;
 625         struct gfs2_bufdata *bd;
 626         struct page *page;
 627         unsigned int length;
 628
 629         gfs2_write_revokes(sdp);
 630         if (!sdp->sd_log_num_revoke)
 631                 return;
 632
 633         length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(u64));
 634         page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
 635         offset = sizeof(struct gfs2_log_descriptor);
 636
 637         list_for_each_entry(bd, head, bd_list) {
 638                 sdp->sd_log_num_revoke--;
 639
 640                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 641
 642                         gfs2_log_write_page(sdp, page);
 643                         page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 644                         mh = page_address(page);
 645                         clear_page(mh);
 646                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 647                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 648                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 649                         offset = sizeof(struct gfs2_meta_header);
 650                 }
 651
 652                 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
 653                 offset += sizeof(u64);
 654         }
 655         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 656
 657         gfs2_log_write_page(sdp, page);
 658 }
 659
 660 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 661 {
 662         struct list_head *head = &sdp->sd_log_le_revoke;
 663         struct gfs2_bufdata *bd;
 664         struct gfs2_glock *gl;
 665
 666         while (!list_empty(head)) {
 667                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 668                 list_del_init(&bd->bd_list);
 669                 gl = bd->bd_gl;
 670                 atomic_dec(&gl->gl_revokes);
 671                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
 672                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 673         }
 674 }
 675
 676 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 677                                   struct gfs2_log_header_host *head, int pass)
 678 {
 679         if (pass != 0)
 680                 return;
 681
 682         jd->jd_found_revokes = 0;
 683         jd->jd_replay_tail = head->lh_tail;
 684 }
 685
 686 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 687                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 688                                    int pass)
 689 {
 690         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 691         unsigned int blks = be32_to_cpu(ld->ld_length);
 692         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 693         struct buffer_head *bh;
 694         unsigned int offset;
 695         u64 blkno;
 696         int first = 1;
 697         int error;
 698
 699         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 700                 return 0;
 701
 702         offset = sizeof(struct gfs2_log_descriptor);
 703
 704         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 705                 error = gfs2_replay_read_block(jd, start, &bh);
 706                 if (error)
 707                         return error;
 708
 709                 if (!first)
 710                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 711
 712                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 713                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 714
 715                         error = gfs2_revoke_add(jd, blkno, start);
 716                         if (error < 0) {
 717                                 brelse(bh);
 718                                 return error;
 719                         }
 720                         else if (error)
 721                                 jd->jd_found_revokes++;
 722
 723                         if (!--revokes)
 724                                 break;
 725                         offset += sizeof(u64);
 726                 }
 727
 728                 brelse(bh);
 729                 offset = sizeof(struct gfs2_meta_header);
 730                 first = 0;
 731         }
 732
 733         return 0;
 734 }
 735
 736 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 737 {
 738         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 739
 740         if (error) {
 741                 gfs2_revoke_clean(jd);
 742                 return;
 743         }
 744         if (pass != 1)
 745                 return;
 746
 747         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 748                 jd->jd_jid, jd->jd_found_revokes);
 749
 750         gfs2_revoke_clean(jd);
 751 }
 752
 753 /**
 754  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 755  *
 756  */
 757
 758 static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 759 {
 760         unsigned int limit = databuf_limit(sdp);
 761         unsigned int nbuf;
 762         if (tr == NULL)
 763                 return;
 764         nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
 765         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1);
 766 }
 767
 768 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 769                                     struct gfs2_log_descriptor *ld,
 770                                     __be64 *ptr, int pass)
 771 {
 772         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 773         struct gfs2_glock *gl = ip->i_gl;
 774         unsigned int blks = be32_to_cpu(ld->ld_data1);
 775         struct buffer_head *bh_log, *bh_ip;
 776         u64 blkno;
 777         u64 esc;
 778         int error = 0;
 779
 780         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
 781                 return 0;
 782
 783         gfs2_replay_incr_blk(jd, &start);
 784         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 785                 blkno = be64_to_cpu(*ptr++);
 786                 esc = be64_to_cpu(*ptr++);
 787
 788                 jd->jd_found_blocks++;
 789
 790                 if (gfs2_revoke_check(jd, blkno, start))
 791                         continue;
 792
 793                 error = gfs2_replay_read_block(jd, start, &bh_log);
 794                 if (error)
 795                         return error;
 796
 797                 bh_ip = gfs2_meta_new(gl, blkno);
 798                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 799
 800                 /* Unescape */
 801                 if (esc) {
 802                         __be32 *eptr = (__be32 *)bh_ip->b_data;
 803                         *eptr = cpu_to_be32(GFS2_MAGIC);
 804                 }
 805                 mark_buffer_dirty(bh_ip);
 806
 807                 brelse(bh_log);
 808                 brelse(bh_ip);
 809
 810                 jd->jd_replayed_blocks++;
 811         }
 812
 813         return error;
 814 }
 815
 816 /* FIXME: sort out accounting for log blocks etc. */
 817
 818 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 819 {
 820         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 821         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 822
 823         if (error) {
 824                 gfs2_meta_sync(ip->i_gl);
 825                 return;
 826         }
 827         if (pass != 1)
 828                 return;
 829
 830         /* data sync? */
 831         gfs2_meta_sync(ip->i_gl);
 832
 833         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
 834                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
 835 }
 836
 837 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 838 {
 839         struct list_head *head;
 840         struct gfs2_bufdata *bd;
 841
 842         if (tr == NULL)
 843                 return;
 844
 845         head = &tr->tr_databuf;
 846         while (!list_empty(head)) {
 847                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 848                 list_del_init(&bd->bd_list);
 849                 gfs2_unpin(sdp, bd->bd_bh, tr);
 850         }
 851 }
 852
 853
 854 const struct gfs2_log_operations gfs2_buf_lops = {
 855         .lo_before_commit = buf_lo_before_commit,
 856         .lo_after_commit = buf_lo_after_commit,
 857         .lo_before_scan = buf_lo_before_scan,
 858         .lo_scan_elements = buf_lo_scan_elements,
 859         .lo_after_scan = buf_lo_after_scan,
 860         .lo_name = "buf",
 861 };
 862
 863 const struct gfs2_log_operations gfs2_revoke_lops = {
 864         .lo_before_commit = revoke_lo_before_commit,
 865         .lo_after_commit = revoke_lo_after_commit,
 866         .lo_before_scan = revoke_lo_before_scan,
 867         .lo_scan_elements = revoke_lo_scan_elements,
 868         .lo_after_scan = revoke_lo_after_scan,
 869         .lo_name = "revoke",
 870 };
 871
 872 const struct gfs2_log_operations gfs2_databuf_lops = {
 873         .lo_before_commit = databuf_lo_before_commit,
 874         .lo_after_commit = databuf_lo_after_commit,
 875         .lo_scan_elements = databuf_lo_scan_elements,
 876         .lo_after_scan = databuf_lo_after_scan,
 877         .lo_name = "databuf",
 878 };
 879
 880 const struct gfs2_log_operations *gfs2_log_ops[] = {
 881         &gfs2_databuf_lops,
 882         &gfs2_buf_lops,
 883         &gfs2_revoke_lops,
 884         NULL,
 885 };
 886