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,
 172                                   struct bio_vec *bvec,
 173                                   blk_status_t error)
 174 {
 175         struct buffer_head *bh, *next;
 176         struct page *page = bvec->bv_page;
 177         unsigned size;
 178
 179         bh = page_buffers(page);
 180         size = bvec->bv_len;
 181         while (bh_offset(bh) < bvec->bv_offset)
 182                 bh = bh->b_this_page;
 183         do {
 184                 if (error)
 185                         mark_buffer_write_io_error(bh);
 186                 unlock_buffer(bh);
 187                 next = bh->b_this_page;
 188                 size -= bh->b_size;
 189                 brelse(bh);
 190                 bh = next;
 191         } while(bh && size);
 192 }
 193
 194 /**
 195  * gfs2_end_log_write - end of i/o to the log
 196  * @bio: The bio
 197  * @error: Status of i/o request
 198  *
 199  * Each bio_vec contains either data from the pagecache or data
 200  * relating to the log itself. Here we iterate over the bio_vec
 201  * array, processing both kinds of data.
 202  *
 203  */
 204
 205 static void gfs2_end_log_write(struct bio *bio)
 206 {
 207         struct gfs2_sbd *sdp = bio->bi_private;
 208         struct bio_vec *bvec;
 209         struct page *page;
 210         int i;
 211         struct bvec_iter_all iter_all;
 212
 213         if (bio->bi_status) {
 214                 fs_err(sdp, "Error %d writing to journal, jid=%u\n",
 215                        bio->bi_status, sdp->sd_jdesc->jd_jid);
 216                 wake_up(&sdp->sd_logd_waitq);
 217         }
 218
 219         bio_for_each_segment_all(bvec, bio, i, iter_all) {
 220                 page = bvec->bv_page;
 221                 if (page_has_buffers(page))
 222                         gfs2_end_log_write_bh(sdp, bvec, bio->bi_status);
 223                 else
 224                         mempool_free(page, gfs2_page_pool);
 225         }
 226
 227         bio_put(bio);
 228         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 229                 wake_up(&sdp->sd_log_flush_wait);
 230 }
 231
 232 /**
 233  * gfs2_log_submit_bio - Submit any pending log bio
 234  * @biop: Address of the bio pointer
 235  * @op: REQ_OP
 236  * @op_flags: req_flag_bits
 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 op, int op_flags)
 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_set_op_attrs(bio, op, op_flags);
 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, 0);
 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 static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
 381                                       u32 ld_length, u32 ld_data1)
 382 {
 383         struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 384         struct gfs2_log_descriptor *ld = page_address(page);
 385         clear_page(ld);
 386         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 387         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 388         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 389         ld->ld_type = cpu_to_be32(ld_type);
 390         ld->ld_length = cpu_to_be32(ld_length);
 391         ld->ld_data1 = cpu_to_be32(ld_data1);
 392         ld->ld_data2 = 0;
 393         return page;
 394 }
 395
 396 static void gfs2_check_magic(struct buffer_head *bh)
 397 {
 398         void *kaddr;
 399         __be32 *ptr;
 400
 401         clear_buffer_escaped(bh);
 402         kaddr = kmap_atomic(bh->b_page);
 403         ptr = kaddr + bh_offset(bh);
 404         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 405                 set_buffer_escaped(bh);
 406         kunmap_atomic(kaddr);
 407 }
 408
 409 static int blocknr_cmp(void *priv, struct list_head *a, struct list_head *b)
 410 {
 411         struct gfs2_bufdata *bda, *bdb;
 412
 413         bda = list_entry(a, struct gfs2_bufdata, bd_list);
 414         bdb = list_entry(b, struct gfs2_bufdata, bd_list);
 415
 416         if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
 417                 return -1;
 418         if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
 419                 return 1;
 420         return 0;
 421 }
 422
 423 static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
 424                                 unsigned int total, struct list_head *blist,
 425                                 bool is_databuf)
 426 {
 427         struct gfs2_log_descriptor *ld;
 428         struct gfs2_bufdata *bd1 = NULL, *bd2;
 429         struct page *page;
 430         unsigned int num;
 431         unsigned n;
 432         __be64 *ptr;
 433
 434         gfs2_log_lock(sdp);
 435         list_sort(NULL, blist, blocknr_cmp);
 436         bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
 437         while(total) {
 438                 num = total;
 439                 if (total > limit)
 440                         num = limit;
 441                 gfs2_log_unlock(sdp);
 442                 page = gfs2_get_log_desc(sdp,
 443                                          is_databuf ? GFS2_LOG_DESC_JDATA :
 444                                          GFS2_LOG_DESC_METADATA, num + 1, num);
 445                 ld = page_address(page);
 446                 gfs2_log_lock(sdp);
 447                 ptr = (__be64 *)(ld + 1);
 448
 449                 n = 0;
 450                 list_for_each_entry_continue(bd1, blist, bd_list) {
 451                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 452                         if (is_databuf) {
 453                                 gfs2_check_magic(bd1->bd_bh);
 454                                 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
 455                         }
 456                         if (++n >= num)
 457                                 break;
 458                 }
 459
 460                 gfs2_log_unlock(sdp);
 461                 gfs2_log_write_page(sdp, page);
 462                 gfs2_log_lock(sdp);
 463
 464                 n = 0;
 465                 list_for_each_entry_continue(bd2, blist, bd_list) {
 466                         get_bh(bd2->bd_bh);
 467                         gfs2_log_unlock(sdp);
 468                         lock_buffer(bd2->bd_bh);
 469
 470                         if (buffer_escaped(bd2->bd_bh)) {
 471                                 void *kaddr;
 472                                 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 473                                 ptr = page_address(page);
 474                                 kaddr = kmap_atomic(bd2->bd_bh->b_page);
 475                                 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh),
 476                                        bd2->bd_bh->b_size);
 477                                 kunmap_atomic(kaddr);
 478                                 *(__be32 *)ptr = 0;
 479                                 clear_buffer_escaped(bd2->bd_bh);
 480                                 unlock_buffer(bd2->bd_bh);
 481                                 brelse(bd2->bd_bh);
 482                                 gfs2_log_write_page(sdp, page);
 483                         } else {
 484                                 gfs2_log_write_bh(sdp, bd2->bd_bh);
 485                         }
 486                         gfs2_log_lock(sdp);
 487                         if (++n >= num)
 488                                 break;
 489                 }
 490
 491                 BUG_ON(total < num);
 492                 total -= num;
 493         }
 494         gfs2_log_unlock(sdp);
 495 }
 496
 497 static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 498 {
 499         unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
 500         unsigned int nbuf;
 501         if (tr == NULL)
 502                 return;
 503         nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
 504         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0);
 505 }
 506
 507 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 508 {
 509         struct list_head *head;
 510         struct gfs2_bufdata *bd;
 511
 512         if (tr == NULL)
 513                 return;
 514
 515         head = &tr->tr_buf;
 516         while (!list_empty(head)) {
 517                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 518                 list_del_init(&bd->bd_list);
 519                 gfs2_unpin(sdp, bd->bd_bh, tr);
 520         }
 521 }
 522
 523 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 524                                struct gfs2_log_header_host *head, int pass)
 525 {
 526         if (pass != 0)
 527                 return;
 528
 529         jd->jd_found_blocks = 0;
 530         jd->jd_replayed_blocks = 0;
 531 }
 532
 533 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 534                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 535                                 int pass)
 536 {
 537         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 538         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 539         struct gfs2_glock *gl = ip->i_gl;
 540         unsigned int blks = be32_to_cpu(ld->ld_data1);
 541         struct buffer_head *bh_log, *bh_ip;
 542         u64 blkno;
 543         int error = 0;
 544
 545         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 546                 return 0;
 547
 548         gfs2_replay_incr_blk(jd, &start);
 549
 550         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 551                 blkno = be64_to_cpu(*ptr++);
 552
 553                 jd->jd_found_blocks++;
 554
 555                 if (gfs2_revoke_check(jd, blkno, start))
 556                         continue;
 557
 558                 error = gfs2_replay_read_block(jd, start, &bh_log);
 559                 if (error)
 560                         return error;
 561
 562                 bh_ip = gfs2_meta_new(gl, blkno);
 563                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 564
 565                 if (gfs2_meta_check(sdp, bh_ip))
 566                         error = -EIO;
 567                 else
 568                         mark_buffer_dirty(bh_ip);
 569
 570                 brelse(bh_log);
 571                 brelse(bh_ip);
 572
 573                 if (error)
 574                         break;
 575
 576                 jd->jd_replayed_blocks++;
 577         }
 578
 579         return error;
 580 }
 581
 582 /**
 583  * gfs2_meta_sync - Sync all buffers associated with a glock
 584  * @gl: The glock
 585  *
 586  */
 587
 588 static void gfs2_meta_sync(struct gfs2_glock *gl)
 589 {
 590         struct address_space *mapping = gfs2_glock2aspace(gl);
 591         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 592         int error;
 593
 594         if (mapping == NULL)
 595                 mapping = &sdp->sd_aspace;
 596
 597         filemap_fdatawrite(mapping);
 598         error = filemap_fdatawait(mapping);
 599
 600         if (error)
 601                 gfs2_io_error(gl->gl_name.ln_sbd);
 602 }
 603
 604 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 605 {
 606         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 607         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 608
 609         if (error) {
 610                 gfs2_meta_sync(ip->i_gl);
 611                 return;
 612         }
 613         if (pass != 1)
 614                 return;
 615
 616         gfs2_meta_sync(ip->i_gl);
 617
 618         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 619                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
 620 }
 621
 622 static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 623 {
 624         struct gfs2_meta_header *mh;
 625         unsigned int offset;
 626         struct list_head *head = &sdp->sd_log_le_revoke;
 627         struct gfs2_bufdata *bd;
 628         struct page *page;
 629         unsigned int length;
 630
 631         gfs2_write_revokes(sdp);
 632         if (!sdp->sd_log_num_revoke)
 633                 return;
 634
 635         length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(u64));
 636         page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
 637         offset = sizeof(struct gfs2_log_descriptor);
 638
 639         list_for_each_entry(bd, head, bd_list) {
 640                 sdp->sd_log_num_revoke--;
 641
 642                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 643
 644                         gfs2_log_write_page(sdp, page);
 645                         page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 646                         mh = page_address(page);
 647                         clear_page(mh);
 648                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 649                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 650                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 651                         offset = sizeof(struct gfs2_meta_header);
 652                 }
 653
 654                 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
 655                 offset += sizeof(u64);
 656         }
 657         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 658
 659         gfs2_log_write_page(sdp, page);
 660 }
 661
 662 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 663 {
 664         struct list_head *head = &sdp->sd_log_le_revoke;
 665         struct gfs2_bufdata *bd;
 666         struct gfs2_glock *gl;
 667
 668         while (!list_empty(head)) {
 669                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 670                 list_del_init(&bd->bd_list);
 671                 gl = bd->bd_gl;
 672                 atomic_dec(&gl->gl_revokes);
 673                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
 674                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 675         }
 676 }
 677
 678 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 679                                   struct gfs2_log_header_host *head, int pass)
 680 {
 681         if (pass != 0)
 682                 return;
 683
 684         jd->jd_found_revokes = 0;
 685         jd->jd_replay_tail = head->lh_tail;
 686 }
 687
 688 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 689                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 690                                    int pass)
 691 {
 692         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 693         unsigned int blks = be32_to_cpu(ld->ld_length);
 694         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 695         struct buffer_head *bh;
 696         unsigned int offset;
 697         u64 blkno;
 698         int first = 1;
 699         int error;
 700
 701         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 702                 return 0;
 703
 704         offset = sizeof(struct gfs2_log_descriptor);
 705
 706         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 707                 error = gfs2_replay_read_block(jd, start, &bh);
 708                 if (error)
 709                         return error;
 710
 711                 if (!first)
 712                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 713
 714                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 715                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 716
 717                         error = gfs2_revoke_add(jd, blkno, start);
 718                         if (error < 0) {
 719                                 brelse(bh);
 720                                 return error;
 721                         }
 722                         else if (error)
 723                                 jd->jd_found_revokes++;
 724
 725                         if (!--revokes)
 726                                 break;
 727                         offset += sizeof(u64);
 728                 }
 729
 730                 brelse(bh);
 731                 offset = sizeof(struct gfs2_meta_header);
 732                 first = 0;
 733         }
 734
 735         return 0;
 736 }
 737
 738 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 739 {
 740         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 741
 742         if (error) {
 743                 gfs2_revoke_clean(jd);
 744                 return;
 745         }
 746         if (pass != 1)
 747                 return;
 748
 749         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 750                 jd->jd_jid, jd->jd_found_revokes);
 751
 752         gfs2_revoke_clean(jd);
 753 }
 754
 755 /**
 756  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 757  *
 758  */
 759
 760 static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 761 {
 762         unsigned int limit = databuf_limit(sdp);
 763         unsigned int nbuf;
 764         if (tr == NULL)
 765                 return;
 766         nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
 767         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1);
 768 }
 769
 770 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 771                                     struct gfs2_log_descriptor *ld,
 772                                     __be64 *ptr, int pass)
 773 {
 774         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 775         struct gfs2_glock *gl = ip->i_gl;
 776         unsigned int blks = be32_to_cpu(ld->ld_data1);
 777         struct buffer_head *bh_log, *bh_ip;
 778         u64 blkno;
 779         u64 esc;
 780         int error = 0;
 781
 782         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
 783                 return 0;
 784
 785         gfs2_replay_incr_blk(jd, &start);
 786         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 787                 blkno = be64_to_cpu(*ptr++);
 788                 esc = be64_to_cpu(*ptr++);
 789
 790                 jd->jd_found_blocks++;
 791
 792                 if (gfs2_revoke_check(jd, blkno, start))
 793                         continue;
 794
 795                 error = gfs2_replay_read_block(jd, start, &bh_log);
 796                 if (error)
 797                         return error;
 798
 799                 bh_ip = gfs2_meta_new(gl, blkno);
 800                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 801
 802                 /* Unescape */
 803                 if (esc) {
 804                         __be32 *eptr = (__be32 *)bh_ip->b_data;
 805                         *eptr = cpu_to_be32(GFS2_MAGIC);
 806                 }
 807                 mark_buffer_dirty(bh_ip);
 808
 809                 brelse(bh_log);
 810                 brelse(bh_ip);
 811
 812                 jd->jd_replayed_blocks++;
 813         }
 814
 815         return error;
 816 }
 817
 818 /* FIXME: sort out accounting for log blocks etc. */
 819
 820 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 821 {
 822         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 823         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 824
 825         if (error) {
 826                 gfs2_meta_sync(ip->i_gl);
 827                 return;
 828         }
 829         if (pass != 1)
 830                 return;
 831
 832         /* data sync? */
 833         gfs2_meta_sync(ip->i_gl);
 834
 835         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
 836                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
 837 }
 838
 839 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 840 {
 841         struct list_head *head;
 842         struct gfs2_bufdata *bd;
 843
 844         if (tr == NULL)
 845                 return;
 846
 847         head = &tr->tr_databuf;
 848         while (!list_empty(head)) {
 849                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 850                 list_del_init(&bd->bd_list);
 851                 gfs2_unpin(sdp, bd->bd_bh, tr);
 852         }
 853 }
 854
 855
 856 const struct gfs2_log_operations gfs2_buf_lops = {
 857         .lo_before_commit = buf_lo_before_commit,
 858         .lo_after_commit = buf_lo_after_commit,
 859         .lo_before_scan = buf_lo_before_scan,
 860         .lo_scan_elements = buf_lo_scan_elements,
 861         .lo_after_scan = buf_lo_after_scan,
 862         .lo_name = "buf",
 863 };
 864
 865 const struct gfs2_log_operations gfs2_revoke_lops = {
 866         .lo_before_commit = revoke_lo_before_commit,
 867         .lo_after_commit = revoke_lo_after_commit,
 868         .lo_before_scan = revoke_lo_before_scan,
 869         .lo_scan_elements = revoke_lo_scan_elements,
 870         .lo_after_scan = revoke_lo_after_scan,
 871         .lo_name = "revoke",
 872 };
 873
 874 const struct gfs2_log_operations gfs2_databuf_lops = {
 875         .lo_before_commit = databuf_lo_before_commit,
 876         .lo_after_commit = databuf_lo_after_commit,
 877         .lo_scan_elements = databuf_lo_scan_elements,
 878         .lo_after_scan = databuf_lo_after_scan,
 879         .lo_name = "databuf",
 880 };
 881
 882 const struct gfs2_log_operations *gfs2_log_ops[] = {
 883         &gfs2_databuf_lops,
 884         &gfs2_buf_lops,
 885         &gfs2_revoke_lops,
 886         NULL,
 887 };
 888