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         struct bvec_iter_all iter_all;
 211
 212         if (bio->bi_status) {
 213                 fs_err(sdp, "Error %d writing to journal, jid=%u\n",
 214                        bio->bi_status, sdp->sd_jdesc->jd_jid);
 215                 wake_up(&sdp->sd_logd_waitq);
 216         }
 217
 218         bio_for_each_segment_all(bvec, bio, iter_all) {
 219                 page = bvec->bv_page;
 220                 if (page_has_buffers(page))
 221                         gfs2_end_log_write_bh(sdp, bvec, bio->bi_status);
 222                 else
 223                         mempool_free(page, gfs2_page_pool);
 224         }
 225
 226         bio_put(bio);
 227         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 228                 wake_up(&sdp->sd_log_flush_wait);
 229 }
 230
 231 /**
 232  * gfs2_log_submit_bio - Submit any pending log bio
 233  * @biop: Address of the bio pointer
 234  * @op: REQ_OP
 235  * @op_flags: req_flag_bits
 236  *
 237  * Submit any pending part-built or full bio to the block device. If
 238  * there is no pending bio, then this is a no-op.
 239  */
 240
 241 void gfs2_log_submit_bio(struct bio **biop, int op, int op_flags)
 242 {
 243         struct bio *bio = *biop;
 244         if (bio) {
 245                 struct gfs2_sbd *sdp = bio->bi_private;
 246                 atomic_inc(&sdp->sd_log_in_flight);
 247                 bio_set_op_attrs(bio, op, op_flags);
 248                 submit_bio(bio);
 249                 *biop = NULL;
 250         }
 251 }
 252
 253 /**
 254  * gfs2_log_alloc_bio - Allocate a bio
 255  * @sdp: The super block
 256  * @blkno: The device block number we want to write to
 257  * @end_io: The bi_end_io callback
 258  *
 259  * Allocate a new bio, initialize it with the given parameters and return it.
 260  *
 261  * Returns: The newly allocated bio
 262  */
 263
 264 static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno,
 265                                       bio_end_io_t *end_io)
 266 {
 267         struct super_block *sb = sdp->sd_vfs;
 268         struct bio *bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
 269
 270         bio->bi_iter.bi_sector = blkno * (sb->s_blocksize >> 9);
 271         bio_set_dev(bio, sb->s_bdev);
 272         bio->bi_end_io = end_io;
 273         bio->bi_private = sdp;
 274
 275         return bio;
 276 }
 277
 278 /**
 279  * gfs2_log_get_bio - Get cached log bio, or allocate a new one
 280  * @sdp: The super block
 281  * @blkno: The device block number we want to write to
 282  * @bio: The bio to get or allocate
 283  * @op: REQ_OP
 284  * @end_io: The bi_end_io callback
 285  * @flush: Always flush the current bio and allocate a new one?
 286  *
 287  * If there is a cached bio, then if the next block number is sequential
 288  * with the previous one, return it, otherwise flush the bio to the
 289  * device. If there is no cached bio, or we just flushed it, then
 290  * allocate a new one.
 291  *
 292  * Returns: The bio to use for log writes
 293  */
 294
 295 static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno,
 296                                     struct bio **biop, int op,
 297                                     bio_end_io_t *end_io, bool flush)
 298 {
 299         struct bio *bio = *biop;
 300
 301         if (bio) {
 302                 u64 nblk;
 303
 304                 nblk = bio_end_sector(bio);
 305                 nblk >>= sdp->sd_fsb2bb_shift;
 306                 if (blkno == nblk && !flush)
 307                         return bio;
 308                 gfs2_log_submit_bio(biop, op, 0);
 309         }
 310
 311         *biop = gfs2_log_alloc_bio(sdp, blkno, end_io);
 312         return *biop;
 313 }
 314
 315 /**
 316  * gfs2_log_write - write to log
 317  * @sdp: the filesystem
 318  * @page: the page to write
 319  * @size: the size of the data to write
 320  * @offset: the offset within the page
 321  * @blkno: block number of the log entry
 322  *
 323  * Try and add the page segment to the current bio. If that fails,
 324  * submit the current bio to the device and create a new one, and
 325  * then add the page segment to that.
 326  */
 327
 328 void gfs2_log_write(struct gfs2_sbd *sdp, struct page *page,
 329                     unsigned size, unsigned offset, u64 blkno)
 330 {
 331         struct bio *bio;
 332         int ret;
 333
 334         bio = gfs2_log_get_bio(sdp, blkno, &sdp->sd_log_bio, REQ_OP_WRITE,
 335                                gfs2_end_log_write, false);
 336         ret = bio_add_page(bio, page, size, offset);
 337         if (ret == 0) {
 338                 bio = gfs2_log_get_bio(sdp, blkno, &sdp->sd_log_bio,
 339                                        REQ_OP_WRITE, gfs2_end_log_write, true);
 340                 ret = bio_add_page(bio, page, size, offset);
 341                 WARN_ON(ret == 0);
 342         }
 343 }
 344
 345 /**
 346  * gfs2_log_write_bh - write a buffer's content to the log
 347  * @sdp: The super block
 348  * @bh: The buffer pointing to the in-place location
 349  *
 350  * This writes the content of the buffer to the next available location
 351  * in the log. The buffer will be unlocked once the i/o to the log has
 352  * completed.
 353  */
 354
 355 static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh)
 356 {
 357         gfs2_log_write(sdp, bh->b_page, bh->b_size, bh_offset(bh),
 358                        gfs2_log_bmap(sdp));
 359 }
 360
 361 /**
 362  * gfs2_log_write_page - write one block stored in a page, into the log
 363  * @sdp: The superblock
 364  * @page: The struct page
 365  *
 366  * This writes the first block-sized part of the page into the log. Note
 367  * that the page must have been allocated from the gfs2_page_pool mempool
 368  * and that after this has been called, ownership has been transferred and
 369  * the page may be freed at any time.
 370  */
 371
 372 void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
 373 {
 374         struct super_block *sb = sdp->sd_vfs;
 375         gfs2_log_write(sdp, page, sb->s_blocksize, 0,
 376                        gfs2_log_bmap(sdp));
 377 }
 378
 379 static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
 380                                       u32 ld_length, u32 ld_data1)
 381 {
 382         struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 383         struct gfs2_log_descriptor *ld = page_address(page);
 384         clear_page(ld);
 385         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 386         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 387         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 388         ld->ld_type = cpu_to_be32(ld_type);
 389         ld->ld_length = cpu_to_be32(ld_length);
 390         ld->ld_data1 = cpu_to_be32(ld_data1);
 391         ld->ld_data2 = 0;
 392         return page;
 393 }
 394
 395 static void gfs2_check_magic(struct buffer_head *bh)
 396 {
 397         void *kaddr;
 398         __be32 *ptr;
 399
 400         clear_buffer_escaped(bh);
 401         kaddr = kmap_atomic(bh->b_page);
 402         ptr = kaddr + bh_offset(bh);
 403         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 404                 set_buffer_escaped(bh);
 405         kunmap_atomic(kaddr);
 406 }
 407
 408 static int blocknr_cmp(void *priv, struct list_head *a, struct list_head *b)
 409 {
 410         struct gfs2_bufdata *bda, *bdb;
 411
 412         bda = list_entry(a, struct gfs2_bufdata, bd_list);
 413         bdb = list_entry(b, struct gfs2_bufdata, bd_list);
 414
 415         if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
 416                 return -1;
 417         if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
 418                 return 1;
 419         return 0;
 420 }
 421
 422 static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
 423                                 unsigned int total, struct list_head *blist,
 424                                 bool is_databuf)
 425 {
 426         struct gfs2_log_descriptor *ld;
 427         struct gfs2_bufdata *bd1 = NULL, *bd2;
 428         struct page *page;
 429         unsigned int num;
 430         unsigned n;
 431         __be64 *ptr;
 432
 433         gfs2_log_lock(sdp);
 434         list_sort(NULL, blist, blocknr_cmp);
 435         bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
 436         while(total) {
 437                 num = total;
 438                 if (total > limit)
 439                         num = limit;
 440                 gfs2_log_unlock(sdp);
 441                 page = gfs2_get_log_desc(sdp,
 442                                          is_databuf ? GFS2_LOG_DESC_JDATA :
 443                                          GFS2_LOG_DESC_METADATA, num + 1, num);
 444                 ld = page_address(page);
 445                 gfs2_log_lock(sdp);
 446                 ptr = (__be64 *)(ld + 1);
 447
 448                 n = 0;
 449                 list_for_each_entry_continue(bd1, blist, bd_list) {
 450                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 451                         if (is_databuf) {
 452                                 gfs2_check_magic(bd1->bd_bh);
 453                                 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
 454                         }
 455                         if (++n >= num)
 456                                 break;
 457                 }
 458
 459                 gfs2_log_unlock(sdp);
 460                 gfs2_log_write_page(sdp, page);
 461                 gfs2_log_lock(sdp);
 462
 463                 n = 0;
 464                 list_for_each_entry_continue(bd2, blist, bd_list) {
 465                         get_bh(bd2->bd_bh);
 466                         gfs2_log_unlock(sdp);
 467                         lock_buffer(bd2->bd_bh);
 468
 469                         if (buffer_escaped(bd2->bd_bh)) {
 470                                 void *kaddr;
 471                                 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 472                                 ptr = page_address(page);
 473                                 kaddr = kmap_atomic(bd2->bd_bh->b_page);
 474                                 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh),
 475                                        bd2->bd_bh->b_size);
 476                                 kunmap_atomic(kaddr);
 477                                 *(__be32 *)ptr = 0;
 478                                 clear_buffer_escaped(bd2->bd_bh);
 479                                 unlock_buffer(bd2->bd_bh);
 480                                 brelse(bd2->bd_bh);
 481                                 gfs2_log_write_page(sdp, page);
 482                         } else {
 483                                 gfs2_log_write_bh(sdp, bd2->bd_bh);
 484                         }
 485                         gfs2_log_lock(sdp);
 486                         if (++n >= num)
 487                                 break;
 488                 }
 489
 490                 BUG_ON(total < num);
 491                 total -= num;
 492         }
 493         gfs2_log_unlock(sdp);
 494 }
 495
 496 static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 497 {
 498         unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
 499         unsigned int nbuf;
 500         if (tr == NULL)
 501                 return;
 502         nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
 503         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0);
 504 }
 505
 506 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 507 {
 508         struct list_head *head;
 509         struct gfs2_bufdata *bd;
 510
 511         if (tr == NULL)
 512                 return;
 513
 514         head = &tr->tr_buf;
 515         while (!list_empty(head)) {
 516                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 517                 list_del_init(&bd->bd_list);
 518                 gfs2_unpin(sdp, bd->bd_bh, tr);
 519         }
 520 }
 521
 522 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 523                                struct gfs2_log_header_host *head, int pass)
 524 {
 525         if (pass != 0)
 526                 return;
 527
 528         jd->jd_found_blocks = 0;
 529         jd->jd_replayed_blocks = 0;
 530 }
 531
 532 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 533                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 534                                 int pass)
 535 {
 536         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 537         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 538         struct gfs2_glock *gl = ip->i_gl;
 539         unsigned int blks = be32_to_cpu(ld->ld_data1);
 540         struct buffer_head *bh_log, *bh_ip;
 541         u64 blkno;
 542         int error = 0;
 543
 544         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 545                 return 0;
 546
 547         gfs2_replay_incr_blk(jd, &start);
 548
 549         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 550                 blkno = be64_to_cpu(*ptr++);
 551
 552                 jd->jd_found_blocks++;
 553
 554                 if (gfs2_revoke_check(jd, blkno, start))
 555                         continue;
 556
 557                 error = gfs2_replay_read_block(jd, start, &bh_log);
 558                 if (error)
 559                         return error;
 560
 561                 bh_ip = gfs2_meta_new(gl, blkno);
 562                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 563
 564                 if (gfs2_meta_check(sdp, bh_ip))
 565                         error = -EIO;
 566                 else
 567                         mark_buffer_dirty(bh_ip);
 568
 569                 brelse(bh_log);
 570                 brelse(bh_ip);
 571
 572                 if (error)
 573                         break;
 574
 575                 jd->jd_replayed_blocks++;
 576         }
 577
 578         return error;
 579 }
 580
 581 /**
 582  * gfs2_meta_sync - Sync all buffers associated with a glock
 583  * @gl: The glock
 584  *
 585  */
 586
 587 static void gfs2_meta_sync(struct gfs2_glock *gl)
 588 {
 589         struct address_space *mapping = gfs2_glock2aspace(gl);
 590         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 591         int error;
 592
 593         if (mapping == NULL)
 594                 mapping = &sdp->sd_aspace;
 595
 596         filemap_fdatawrite(mapping);
 597         error = filemap_fdatawait(mapping);
 598
 599         if (error)
 600                 gfs2_io_error(gl->gl_name.ln_sbd);
 601 }
 602
 603 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 604 {
 605         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 606         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 607
 608         if (error) {
 609                 gfs2_meta_sync(ip->i_gl);
 610                 return;
 611         }
 612         if (pass != 1)
 613                 return;
 614
 615         gfs2_meta_sync(ip->i_gl);
 616
 617         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 618                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
 619 }
 620
 621 static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 622 {
 623         struct gfs2_meta_header *mh;
 624         unsigned int offset;
 625         struct list_head *head = &sdp->sd_log_le_revoke;
 626         struct gfs2_bufdata *bd;
 627         struct page *page;
 628         unsigned int length;
 629
 630         gfs2_write_revokes(sdp);
 631         if (!sdp->sd_log_num_revoke)
 632                 return;
 633
 634         length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(u64));
 635         page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
 636         offset = sizeof(struct gfs2_log_descriptor);
 637
 638         list_for_each_entry(bd, head, bd_list) {
 639                 sdp->sd_log_num_revoke--;
 640
 641                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 642
 643                         gfs2_log_write_page(sdp, page);
 644                         page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 645                         mh = page_address(page);
 646                         clear_page(mh);
 647                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 648                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 649                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 650                         offset = sizeof(struct gfs2_meta_header);
 651                 }
 652
 653                 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
 654                 offset += sizeof(u64);
 655         }
 656         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 657
 658         gfs2_log_write_page(sdp, page);
 659 }
 660
 661 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 662 {
 663         struct list_head *head = &sdp->sd_log_le_revoke;
 664         struct gfs2_bufdata *bd;
 665         struct gfs2_glock *gl;
 666
 667         while (!list_empty(head)) {
 668                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 669                 list_del_init(&bd->bd_list);
 670                 gl = bd->bd_gl;
 671                 atomic_dec(&gl->gl_revokes);
 672                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
 673                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 674         }
 675 }
 676
 677 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 678                                   struct gfs2_log_header_host *head, int pass)
 679 {
 680         if (pass != 0)
 681                 return;
 682
 683         jd->jd_found_revokes = 0;
 684         jd->jd_replay_tail = head->lh_tail;
 685 }
 686
 687 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 688                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 689                                    int pass)
 690 {
 691         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 692         unsigned int blks = be32_to_cpu(ld->ld_length);
 693         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 694         struct buffer_head *bh;
 695         unsigned int offset;
 696         u64 blkno;
 697         int first = 1;
 698         int error;
 699
 700         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 701                 return 0;
 702
 703         offset = sizeof(struct gfs2_log_descriptor);
 704
 705         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 706                 error = gfs2_replay_read_block(jd, start, &bh);
 707                 if (error)
 708                         return error;
 709
 710                 if (!first)
 711                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 712
 713                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 714                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 715
 716                         error = gfs2_revoke_add(jd, blkno, start);
 717                         if (error < 0) {
 718                                 brelse(bh);
 719                                 return error;
 720                         }
 721                         else if (error)
 722                                 jd->jd_found_revokes++;
 723
 724                         if (!--revokes)
 725                                 break;
 726                         offset += sizeof(u64);
 727                 }
 728
 729                 brelse(bh);
 730                 offset = sizeof(struct gfs2_meta_header);
 731                 first = 0;
 732         }
 733
 734         return 0;
 735 }
 736
 737 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 738 {
 739         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 740
 741         if (error) {
 742                 gfs2_revoke_clean(jd);
 743                 return;
 744         }
 745         if (pass != 1)
 746                 return;
 747
 748         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 749                 jd->jd_jid, jd->jd_found_revokes);
 750
 751         gfs2_revoke_clean(jd);
 752 }
 753
 754 /**
 755  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 756  *
 757  */
 758
 759 static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 760 {
 761         unsigned int limit = databuf_limit(sdp);
 762         unsigned int nbuf;
 763         if (tr == NULL)
 764                 return;
 765         nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
 766         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1);
 767 }
 768
 769 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 770                                     struct gfs2_log_descriptor *ld,
 771                                     __be64 *ptr, int pass)
 772 {
 773         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 774         struct gfs2_glock *gl = ip->i_gl;
 775         unsigned int blks = be32_to_cpu(ld->ld_data1);
 776         struct buffer_head *bh_log, *bh_ip;
 777         u64 blkno;
 778         u64 esc;
 779         int error = 0;
 780
 781         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
 782                 return 0;
 783
 784         gfs2_replay_incr_blk(jd, &start);
 785         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 786                 blkno = be64_to_cpu(*ptr++);
 787                 esc = be64_to_cpu(*ptr++);
 788
 789                 jd->jd_found_blocks++;
 790
 791                 if (gfs2_revoke_check(jd, blkno, start))
 792                         continue;
 793
 794                 error = gfs2_replay_read_block(jd, start, &bh_log);
 795                 if (error)
 796                         return error;
 797
 798                 bh_ip = gfs2_meta_new(gl, blkno);
 799                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 800
 801                 /* Unescape */
 802                 if (esc) {
 803                         __be32 *eptr = (__be32 *)bh_ip->b_data;
 804                         *eptr = cpu_to_be32(GFS2_MAGIC);
 805                 }
 806                 mark_buffer_dirty(bh_ip);
 807
 808                 brelse(bh_log);
 809                 brelse(bh_ip);
 810
 811                 jd->jd_replayed_blocks++;
 812         }
 813
 814         return error;
 815 }
 816
 817 /* FIXME: sort out accounting for log blocks etc. */
 818
 819 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 820 {
 821         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 822         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 823
 824         if (error) {
 825                 gfs2_meta_sync(ip->i_gl);
 826                 return;
 827         }
 828         if (pass != 1)
 829                 return;
 830
 831         /* data sync? */
 832         gfs2_meta_sync(ip->i_gl);
 833
 834         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
 835                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
 836 }
 837
 838 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 839 {
 840         struct list_head *head;
 841         struct gfs2_bufdata *bd;
 842
 843         if (tr == NULL)
 844                 return;
 845
 846         head = &tr->tr_databuf;
 847         while (!list_empty(head)) {
 848                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 849                 list_del_init(&bd->bd_list);
 850                 gfs2_unpin(sdp, bd->bd_bh, tr);
 851         }
 852 }
 853
 854
 855 const struct gfs2_log_operations gfs2_buf_lops = {
 856         .lo_before_commit = buf_lo_before_commit,
 857         .lo_after_commit = buf_lo_after_commit,
 858         .lo_before_scan = buf_lo_before_scan,
 859         .lo_scan_elements = buf_lo_scan_elements,
 860         .lo_after_scan = buf_lo_after_scan,
 861         .lo_name = "buf",
 862 };
 863
 864 const struct gfs2_log_operations gfs2_revoke_lops = {
 865         .lo_before_commit = revoke_lo_before_commit,
 866         .lo_after_commit = revoke_lo_after_commit,
 867         .lo_before_scan = revoke_lo_before_scan,
 868         .lo_scan_elements = revoke_lo_scan_elements,
 869         .lo_after_scan = revoke_lo_after_scan,
 870         .lo_name = "revoke",
 871 };
 872
 873 const struct gfs2_log_operations gfs2_databuf_lops = {
 874         .lo_before_commit = databuf_lo_before_commit,
 875         .lo_after_commit = databuf_lo_after_commit,
 876         .lo_scan_elements = databuf_lo_scan_elements,
 877         .lo_after_scan = databuf_lo_after_scan,
 878         .lo_name = "databuf",
 879 };
 880
 881 const struct gfs2_log_operations *gfs2_log_ops[] = {
 882         &gfs2_databuf_lops,
 883         &gfs2_buf_lops,
 884         &gfs2_revoke_lops,
 885         NULL,
 886 };
 887