fs/gfs2/lops.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   4  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
   5  */
   6
   7 #include <linux/sched.h>
   8 #include <linux/slab.h>
   9 #include <linux/spinlock.h>
  10 #include <linux/completion.h>
  11 #include <linux/buffer_head.h>
  12 #include <linux/mempool.h>
  13 #include <linux/gfs2_ondisk.h>
  14 #include <linux/bio.h>
  15 #include <linux/fs.h>
  16 #include <linux/list_sort.h>
  17 #include <linux/blkdev.h>
  18
  19 #include "bmap.h"
  20 #include "dir.h"
  21 #include "gfs2.h"
  22 #include "incore.h"
  23 #include "inode.h"
  24 #include "glock.h"
  25 #include "glops.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         rgrp_lock_local(rgd);
  80         if (bi->bi_clone == NULL)
  81                 goto out;
  82         if (sdp->sd_args.ar_discard)
  83                 gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi, 1, NULL);
  84         memcpy(bi->bi_clone + bi->bi_offset,
  85                bd->bd_bh->b_data + bi->bi_offset, bi->bi_bytes);
  86         clear_bit(GBF_FULL, &bi->bi_flags);
  87         rgd->rd_free_clone = rgd->rd_free;
  88         BUG_ON(rgd->rd_free_clone < rgd->rd_reserved);
  89         rgd->rd_extfail_pt = rgd->rd_free;
  90
  91 out:
  92         rgrp_unlock_local(rgd);
  93 }
  94
  95 /**
  96  * gfs2_unpin - Unpin a buffer
  97  * @sdp: the filesystem the buffer belongs to
  98  * @bh: The buffer to unpin
  99  * @tr: The system transaction being flushed
 100  */
 101
 102 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
 103                        struct gfs2_trans *tr)
 104 {
 105         struct gfs2_bufdata *bd = bh->b_private;
 106
 107         BUG_ON(!buffer_uptodate(bh));
 108         BUG_ON(!buffer_pinned(bh));
 109
 110         lock_buffer(bh);
 111         mark_buffer_dirty(bh);
 112         clear_buffer_pinned(bh);
 113
 114         if (buffer_is_rgrp(bd))
 115                 maybe_release_space(bd);
 116
 117         spin_lock(&sdp->sd_ail_lock);
 118         if (bd->bd_tr) {
 119                 list_del(&bd->bd_ail_st_list);
 120                 brelse(bh);
 121         } else {
 122                 struct gfs2_glock *gl = bd->bd_gl;
 123                 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
 124                 atomic_inc(&gl->gl_ail_count);
 125         }
 126         bd->bd_tr = tr;
 127         list_add(&bd->bd_ail_st_list, &tr->tr_ail1_list);
 128         spin_unlock(&sdp->sd_ail_lock);
 129
 130         clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 131         trace_gfs2_pin(bd, 0);
 132         unlock_buffer(bh);
 133         atomic_dec(&sdp->sd_log_pinned);
 134 }
 135
 136 void gfs2_log_incr_head(struct gfs2_sbd *sdp)
 137 {
 138         BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) &&
 139                (sdp->sd_log_flush_head != sdp->sd_log_head));
 140
 141         if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks)
 142                 sdp->sd_log_flush_head = 0;
 143 }
 144
 145 u64 gfs2_log_bmap(struct gfs2_jdesc *jd, unsigned int lblock)
 146 {
 147         struct gfs2_journal_extent *je;
 148
 149         list_for_each_entry(je, &jd->extent_list, list) {
 150                 if (lblock >= je->lblock && lblock < je->lblock + je->blocks)
 151                         return je->dblock + lblock - je->lblock;
 152         }
 153
 154         return -1;
 155 }
 156
 157 /**
 158  * gfs2_end_log_write_bh - end log write of pagecache data with buffers
 159  * @sdp: The superblock
 160  * @bvec: The bio_vec
 161  * @error: The i/o status
 162  *
 163  * This finds the relevant buffers and unlocks them and sets the
 164  * error flag according to the status of the i/o request. This is
 165  * used when the log is writing data which has an in-place version
 166  * that is pinned in the pagecache.
 167  */
 168
 169 static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp,
 170                                   struct bio_vec *bvec,
 171                                   blk_status_t error)
 172 {
 173         struct buffer_head *bh, *next;
 174         struct page *page = bvec->bv_page;
 175         unsigned size;
 176
 177         bh = page_buffers(page);
 178         size = bvec->bv_len;
 179         while (bh_offset(bh) < bvec->bv_offset)
 180                 bh = bh->b_this_page;
 181         do {
 182                 if (error)
 183                         mark_buffer_write_io_error(bh);
 184                 unlock_buffer(bh);
 185                 next = bh->b_this_page;
 186                 size -= bh->b_size;
 187                 brelse(bh);
 188                 bh = next;
 189         } while(bh && size);
 190 }
 191
 192 /**
 193  * gfs2_end_log_write - end of i/o to the log
 194  * @bio: The bio
 195  *
 196  * Each bio_vec contains either data from the pagecache or data
 197  * relating to the log itself. Here we iterate over the bio_vec
 198  * array, processing both kinds of data.
 199  *
 200  */
 201
 202 static void gfs2_end_log_write(struct bio *bio)
 203 {
 204         struct gfs2_sbd *sdp = bio->bi_private;
 205         struct bio_vec *bvec;
 206         struct page *page;
 207         struct bvec_iter_all iter_all;
 208
 209         if (bio->bi_status) {
 210                 if (!cmpxchg(&sdp->sd_log_error, 0, (int)bio->bi_status))
 211                         fs_err(sdp, "Error %d writing to journal, jid=%u\n",
 212                                bio->bi_status, sdp->sd_jdesc->jd_jid);
 213                 gfs2_withdraw_delayed(sdp);
 214                 /* prevent more writes to the journal */
 215                 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 216                 wake_up(&sdp->sd_logd_waitq);
 217         }
 218
 219         bio_for_each_segment_all(bvec, bio, 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  * @opf: REQ_OP | op_flags
 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, blk_opf_t opf)
 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->bi_opf = opf;
 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(sb->s_bdev, BIO_MAX_VECS, 0, GFP_NOIO);
 269
 270         bio->bi_iter.bi_sector = blkno << sdp->sd_fsb2bb_shift;
 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  * @biop: 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, enum req_op 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);
 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  * @jd: The journal descriptor
 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 gfs2_jdesc *jd,
 329                     struct page *page, unsigned size, unsigned offset,
 330                     u64 blkno)
 331 {
 332         struct bio *bio;
 333         int ret;
 334
 335         bio = gfs2_log_get_bio(sdp, blkno, &jd->jd_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, &jd->jd_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         u64 dblock;
 359
 360         dblock = gfs2_log_bmap(sdp->sd_jdesc, sdp->sd_log_flush_head);
 361         gfs2_log_incr_head(sdp);
 362         gfs2_log_write(sdp, sdp->sd_jdesc, bh->b_page, bh->b_size,
 363                        bh_offset(bh), dblock);
 364 }
 365
 366 /**
 367  * gfs2_log_write_page - write one block stored in a page, into the log
 368  * @sdp: The superblock
 369  * @page: The struct page
 370  *
 371  * This writes the first block-sized part of the page into the log. Note
 372  * that the page must have been allocated from the gfs2_page_pool mempool
 373  * and that after this has been called, ownership has been transferred and
 374  * the page may be freed at any time.
 375  */
 376
 377 static void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
 378 {
 379         struct super_block *sb = sdp->sd_vfs;
 380         u64 dblock;
 381
 382         dblock = gfs2_log_bmap(sdp->sd_jdesc, sdp->sd_log_flush_head);
 383         gfs2_log_incr_head(sdp);
 384         gfs2_log_write(sdp, sdp->sd_jdesc, page, sb->s_blocksize, 0, dblock);
 385 }
 386
 387 /**
 388  * gfs2_end_log_read - end I/O callback for reads from the log
 389  * @bio: The bio
 390  *
 391  * Simply unlock the pages in the bio. The main thread will wait on them and
 392  * process them in order as necessary.
 393  */
 394 static void gfs2_end_log_read(struct bio *bio)
 395 {
 396         int error = blk_status_to_errno(bio->bi_status);
 397         struct folio_iter fi;
 398
 399         bio_for_each_folio_all(fi, bio) {
 400                 /* We're abusing wb_err to get the error to gfs2_find_jhead */
 401                 filemap_set_wb_err(fi.folio->mapping, error);
 402                 folio_end_read(fi.folio, !error);
 403         }
 404
 405         bio_put(bio);
 406 }
 407
 408 /**
 409  * gfs2_jhead_pg_srch - Look for the journal head in a given page.
 410  * @jd: The journal descriptor
 411  * @head: The journal head to start from
 412  * @page: The page to look in
 413  *
 414  * Returns: 1 if found, 0 otherwise.
 415  */
 416
 417 static bool gfs2_jhead_pg_srch(struct gfs2_jdesc *jd,
 418                               struct gfs2_log_header_host *head,
 419                               struct page *page)
 420 {
 421         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 422         struct gfs2_log_header_host lh;
 423         void *kaddr;
 424         unsigned int offset;
 425         bool ret = false;
 426
 427         kaddr = kmap_local_page(page);
 428         for (offset = 0; offset < PAGE_SIZE; offset += sdp->sd_sb.sb_bsize) {
 429                 if (!__get_log_header(sdp, kaddr + offset, 0, &lh)) {
 430                         if (lh.lh_sequence >= head->lh_sequence)
 431                                 *head = lh;
 432                         else {
 433                                 ret = true;
 434                                 break;
 435                         }
 436                 }
 437         }
 438         kunmap_local(kaddr);
 439         return ret;
 440 }
 441
 442 /**
 443  * gfs2_jhead_process_page - Search/cleanup a page
 444  * @jd: The journal descriptor
 445  * @index: Index of the page to look into
 446  * @head: The journal head to start from
 447  * @done: If set, perform only cleanup, else search and set if found.
 448  *
 449  * Find the folio with 'index' in the journal's mapping. Search the folio for
 450  * the journal head if requested (cleanup == false). Release refs on the
 451  * folio so the page cache can reclaim it. We grabbed a
 452  * reference on this folio twice, first when we did a grab_cache_page()
 453  * to obtain the folio to add it to the bio and second when we do a
 454  * filemap_get_folio() here to get the folio to wait on while I/O on it is being
 455  * completed.
 456  * This function is also used to free up a folio we might've grabbed but not
 457  * used. Maybe we added it to a bio, but not submitted it for I/O. Or we
 458  * submitted the I/O, but we already found the jhead so we only need to drop
 459  * our references to the folio.
 460  */
 461
 462 static void gfs2_jhead_process_page(struct gfs2_jdesc *jd, unsigned long index,
 463                                     struct gfs2_log_header_host *head,
 464                                     bool *done)
 465 {
 466         struct folio *folio;
 467
 468         folio = filemap_get_folio(jd->jd_inode->i_mapping, index);
 469
 470         folio_wait_locked(folio);
 471         if (!folio_test_uptodate(folio))
 472                 *done = true;
 473
 474         if (!*done)
 475                 *done = gfs2_jhead_pg_srch(jd, head, &folio->page);
 476
 477         /* filemap_get_folio() and the earlier grab_cache_page() */
 478         folio_put_refs(folio, 2);
 479 }
 480
 481 static struct bio *gfs2_chain_bio(struct bio *prev, unsigned int nr_iovecs)
 482 {
 483         struct bio *new;
 484
 485         new = bio_alloc(prev->bi_bdev, nr_iovecs, prev->bi_opf, GFP_NOIO);
 486         bio_clone_blkg_association(new, prev);
 487         new->bi_iter.bi_sector = bio_end_sector(prev);
 488         bio_chain(new, prev);
 489         submit_bio(prev);
 490         return new;
 491 }
 492
 493 /**
 494  * gfs2_find_jhead - find the head of a log
 495  * @jd: The journal descriptor
 496  * @head: The log descriptor for the head of the log is returned here
 497  * @keep_cache: If set inode pages will not be truncated
 498  *
 499  * Do a search of a journal by reading it in large chunks using bios and find
 500  * the valid log entry with the highest sequence number.  (i.e. the log head)
 501  *
 502  * Returns: 0 on success, errno otherwise
 503  */
 504 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head,
 505                     bool keep_cache)
 506 {
 507         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 508         struct address_space *mapping = jd->jd_inode->i_mapping;
 509         unsigned int block = 0, blocks_submitted = 0, blocks_read = 0;
 510         unsigned int bsize = sdp->sd_sb.sb_bsize, off;
 511         unsigned int bsize_shift = sdp->sd_sb.sb_bsize_shift;
 512         unsigned int shift = PAGE_SHIFT - bsize_shift;
 513         unsigned int max_blocks = 2 * 1024 * 1024 >> bsize_shift;
 514         struct gfs2_journal_extent *je;
 515         int sz, ret = 0;
 516         struct bio *bio = NULL;
 517         struct page *page = NULL;
 518         bool done = false;
 519         errseq_t since;
 520
 521         memset(head, 0, sizeof(*head));
 522         if (list_empty(&jd->extent_list))
 523                 gfs2_map_journal_extents(sdp, jd);
 524
 525         since = filemap_sample_wb_err(mapping);
 526         list_for_each_entry(je, &jd->extent_list, list) {
 527                 u64 dblock = je->dblock;
 528
 529                 for (; block < je->lblock + je->blocks; block++, dblock++) {
 530                         if (!page) {
 531                                 page = grab_cache_page(mapping, block >> shift);
 532                                 if (!page) {
 533                                         ret = -ENOMEM;
 534                                         done = true;
 535                                         goto out;
 536                                 }
 537                                 off = 0;
 538                         }
 539
 540                         if (bio && (off || block < blocks_submitted + max_blocks)) {
 541                                 sector_t sector = dblock << sdp->sd_fsb2bb_shift;
 542
 543                                 if (bio_end_sector(bio) == sector) {
 544                                         sz = bio_add_page(bio, page, bsize, off);
 545                                         if (sz == bsize)
 546                                                 goto block_added;
 547                                 }
 548                                 if (off) {
 549                                         unsigned int blocks =
 550                                                 (PAGE_SIZE - off) >> bsize_shift;
 551
 552                                         bio = gfs2_chain_bio(bio, blocks);
 553                                         goto add_block_to_new_bio;
 554                                 }
 555                         }
 556
 557                         if (bio) {
 558                                 blocks_submitted = block;
 559                                 submit_bio(bio);
 560                         }
 561
 562                         bio = gfs2_log_alloc_bio(sdp, dblock, gfs2_end_log_read);
 563                         bio->bi_opf = REQ_OP_READ;
 564 add_block_to_new_bio:
 565                         sz = bio_add_page(bio, page, bsize, off);
 566                         BUG_ON(sz != bsize);
 567 block_added:
 568                         off += bsize;
 569                         if (off == PAGE_SIZE)
 570                                 page = NULL;
 571                         if (blocks_submitted <= blocks_read + max_blocks) {
 572                                 /* Keep at least one bio in flight */
 573                                 continue;
 574                         }
 575
 576                         gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done);
 577                         blocks_read += PAGE_SIZE >> bsize_shift;
 578                         if (done)
 579                                 goto out;  /* found */
 580                 }
 581         }
 582
 583 out:
 584         if (bio)
 585                 submit_bio(bio);
 586         while (blocks_read < block) {
 587                 gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done);
 588                 blocks_read += PAGE_SIZE >> bsize_shift;
 589         }
 590
 591         if (!ret)
 592                 ret = filemap_check_wb_err(mapping, since);
 593
 594         if (!keep_cache)
 595                 truncate_inode_pages(mapping, 0);
 596
 597         return ret;
 598 }
 599
 600 static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
 601                                       u32 ld_length, u32 ld_data1)
 602 {
 603         struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 604         struct gfs2_log_descriptor *ld = page_address(page);
 605         clear_page(ld);
 606         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 607         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 608         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 609         ld->ld_type = cpu_to_be32(ld_type);
 610         ld->ld_length = cpu_to_be32(ld_length);
 611         ld->ld_data1 = cpu_to_be32(ld_data1);
 612         ld->ld_data2 = 0;
 613         return page;
 614 }
 615
 616 static void gfs2_check_magic(struct buffer_head *bh)
 617 {
 618         void *kaddr;
 619         __be32 *ptr;
 620
 621         clear_buffer_escaped(bh);
 622         kaddr = kmap_local_page(bh->b_page);
 623         ptr = kaddr + bh_offset(bh);
 624         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 625                 set_buffer_escaped(bh);
 626         kunmap_local(kaddr);
 627 }
 628
 629 static int blocknr_cmp(void *priv, const struct list_head *a,
 630                        const struct list_head *b)
 631 {
 632         struct gfs2_bufdata *bda, *bdb;
 633
 634         bda = list_entry(a, struct gfs2_bufdata, bd_list);
 635         bdb = list_entry(b, struct gfs2_bufdata, bd_list);
 636
 637         if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
 638                 return -1;
 639         if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
 640                 return 1;
 641         return 0;
 642 }
 643
 644 static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
 645                                 unsigned int total, struct list_head *blist,
 646                                 bool is_databuf)
 647 {
 648         struct gfs2_log_descriptor *ld;
 649         struct gfs2_bufdata *bd1 = NULL, *bd2;
 650         struct page *page;
 651         unsigned int num;
 652         unsigned n;
 653         __be64 *ptr;
 654
 655         gfs2_log_lock(sdp);
 656         list_sort(NULL, blist, blocknr_cmp);
 657         bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
 658         while(total) {
 659                 num = total;
 660                 if (total > limit)
 661                         num = limit;
 662                 gfs2_log_unlock(sdp);
 663                 page = gfs2_get_log_desc(sdp,
 664                                          is_databuf ? GFS2_LOG_DESC_JDATA :
 665                                          GFS2_LOG_DESC_METADATA, num + 1, num);
 666                 ld = page_address(page);
 667                 gfs2_log_lock(sdp);
 668                 ptr = (__be64 *)(ld + 1);
 669
 670                 n = 0;
 671                 list_for_each_entry_continue(bd1, blist, bd_list) {
 672                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 673                         if (is_databuf) {
 674                                 gfs2_check_magic(bd1->bd_bh);
 675                                 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
 676                         }
 677                         if (++n >= num)
 678                                 break;
 679                 }
 680
 681                 gfs2_log_unlock(sdp);
 682                 gfs2_log_write_page(sdp, page);
 683                 gfs2_log_lock(sdp);
 684
 685                 n = 0;
 686                 list_for_each_entry_continue(bd2, blist, bd_list) {
 687                         get_bh(bd2->bd_bh);
 688                         gfs2_log_unlock(sdp);
 689                         lock_buffer(bd2->bd_bh);
 690
 691                         if (buffer_escaped(bd2->bd_bh)) {
 692                                 void *p;
 693
 694                                 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 695                                 p = page_address(page);
 696                                 memcpy_from_page(p, page, bh_offset(bd2->bd_bh), bd2->bd_bh->b_size);
 697                                 *(__be32 *)p = 0;
 698                                 clear_buffer_escaped(bd2->bd_bh);
 699                                 unlock_buffer(bd2->bd_bh);
 700                                 brelse(bd2->bd_bh);
 701                                 gfs2_log_write_page(sdp, page);
 702                         } else {
 703                                 gfs2_log_write_bh(sdp, bd2->bd_bh);
 704                         }
 705                         gfs2_log_lock(sdp);
 706                         if (++n >= num)
 707                                 break;
 708                 }
 709
 710                 BUG_ON(total < num);
 711                 total -= num;
 712         }
 713         gfs2_log_unlock(sdp);
 714 }
 715
 716 static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 717 {
 718         unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
 719         unsigned int nbuf;
 720         if (tr == NULL)
 721                 return;
 722         nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
 723         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0);
 724 }
 725
 726 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 727 {
 728         struct list_head *head;
 729         struct gfs2_bufdata *bd;
 730
 731         if (tr == NULL)
 732                 return;
 733
 734         head = &tr->tr_buf;
 735         while (!list_empty(head)) {
 736                 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
 737                 list_del_init(&bd->bd_list);
 738                 gfs2_unpin(sdp, bd->bd_bh, tr);
 739         }
 740 }
 741
 742 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 743                                struct gfs2_log_header_host *head, int pass)
 744 {
 745         if (pass != 0)
 746                 return;
 747
 748         jd->jd_found_blocks = 0;
 749         jd->jd_replayed_blocks = 0;
 750 }
 751
 752 #define obsolete_rgrp_replay \
 753 "Replaying 0x%llx from jid=%d/0x%llx but we already have a bh!\n"
 754 #define obsolete_rgrp_replay2 \
 755 "busy:%d, pinned:%d rg_gen:0x%llx, j_gen:0x%llx\n"
 756
 757 static void obsolete_rgrp(struct gfs2_jdesc *jd, struct buffer_head *bh_log,
 758                           u64 blkno)
 759 {
 760         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 761         struct gfs2_rgrpd *rgd;
 762         struct gfs2_rgrp *jrgd = (struct gfs2_rgrp *)bh_log->b_data;
 763
 764         rgd = gfs2_blk2rgrpd(sdp, blkno, false);
 765         if (rgd && rgd->rd_addr == blkno &&
 766             rgd->rd_bits && rgd->rd_bits->bi_bh) {
 767                 fs_info(sdp, obsolete_rgrp_replay, (unsigned long long)blkno,
 768                         jd->jd_jid, bh_log->b_blocknr);
 769                 fs_info(sdp, obsolete_rgrp_replay2,
 770                         buffer_busy(rgd->rd_bits->bi_bh) ? 1 : 0,
 771                         buffer_pinned(rgd->rd_bits->bi_bh),
 772                         rgd->rd_igeneration,
 773                         be64_to_cpu(jrgd->rg_igeneration));
 774                 gfs2_dump_glock(NULL, rgd->rd_gl, true);
 775         }
 776 }
 777
 778 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
 779                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 780                                 int pass)
 781 {
 782         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 783         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 784         struct gfs2_glock *gl = ip->i_gl;
 785         unsigned int blks = be32_to_cpu(ld->ld_data1);
 786         struct buffer_head *bh_log, *bh_ip;
 787         u64 blkno;
 788         int error = 0;
 789
 790         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 791                 return 0;
 792
 793         gfs2_replay_incr_blk(jd, &start);
 794
 795         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 796                 blkno = be64_to_cpu(*ptr++);
 797
 798                 jd->jd_found_blocks++;
 799
 800                 if (gfs2_revoke_check(jd, blkno, start))
 801                         continue;
 802
 803                 error = gfs2_replay_read_block(jd, start, &bh_log);
 804                 if (error)
 805                         return error;
 806
 807                 bh_ip = gfs2_meta_new(gl, blkno);
 808                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 809
 810                 if (gfs2_meta_check(sdp, bh_ip))
 811                         error = -EIO;
 812                 else {
 813                         struct gfs2_meta_header *mh =
 814                                 (struct gfs2_meta_header *)bh_ip->b_data;
 815
 816                         if (mh->mh_type == cpu_to_be32(GFS2_METATYPE_RG))
 817                                 obsolete_rgrp(jd, bh_log, blkno);
 818
 819                         mark_buffer_dirty(bh_ip);
 820                 }
 821                 brelse(bh_log);
 822                 brelse(bh_ip);
 823
 824                 if (error)
 825                         break;
 826
 827                 jd->jd_replayed_blocks++;
 828         }
 829
 830         return error;
 831 }
 832
 833 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 834 {
 835         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 836         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 837
 838         if (error) {
 839                 gfs2_inode_metasync(ip->i_gl);
 840                 return;
 841         }
 842         if (pass != 1)
 843                 return;
 844
 845         gfs2_inode_metasync(ip->i_gl);
 846
 847         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 848                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
 849 }
 850
 851 static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 852 {
 853         struct gfs2_meta_header *mh;
 854         unsigned int offset;
 855         struct list_head *head = &sdp->sd_log_revokes;
 856         struct gfs2_bufdata *bd;
 857         struct page *page;
 858         unsigned int length;
 859
 860         gfs2_flush_revokes(sdp);
 861         if (!sdp->sd_log_num_revoke)
 862                 return;
 863
 864         length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke);
 865         page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
 866         offset = sizeof(struct gfs2_log_descriptor);
 867
 868         list_for_each_entry(bd, head, bd_list) {
 869                 sdp->sd_log_num_revoke--;
 870
 871                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 872                         gfs2_log_write_page(sdp, page);
 873                         page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 874                         mh = page_address(page);
 875                         clear_page(mh);
 876                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 877                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 878                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 879                         offset = sizeof(struct gfs2_meta_header);
 880                 }
 881
 882                 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
 883                 offset += sizeof(u64);
 884         }
 885         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 886
 887         gfs2_log_write_page(sdp, page);
 888 }
 889
 890 void gfs2_drain_revokes(struct gfs2_sbd *sdp)
 891 {
 892         struct list_head *head = &sdp->sd_log_revokes;
 893         struct gfs2_bufdata *bd;
 894         struct gfs2_glock *gl;
 895
 896         while (!list_empty(head)) {
 897                 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
 898                 list_del_init(&bd->bd_list);
 899                 gl = bd->bd_gl;
 900                 gfs2_glock_remove_revoke(gl);
 901                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 902         }
 903 }
 904
 905 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 906 {
 907         gfs2_drain_revokes(sdp);
 908 }
 909
 910 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 911                                   struct gfs2_log_header_host *head, int pass)
 912 {
 913         if (pass != 0)
 914                 return;
 915
 916         jd->jd_found_revokes = 0;
 917         jd->jd_replay_tail = head->lh_tail;
 918 }
 919
 920 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
 921                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 922                                    int pass)
 923 {
 924         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 925         unsigned int blks = be32_to_cpu(ld->ld_length);
 926         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 927         struct buffer_head *bh;
 928         unsigned int offset;
 929         u64 blkno;
 930         int first = 1;
 931         int error;
 932
 933         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 934                 return 0;
 935
 936         offset = sizeof(struct gfs2_log_descriptor);
 937
 938         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 939                 error = gfs2_replay_read_block(jd, start, &bh);
 940                 if (error)
 941                         return error;
 942
 943                 if (!first)
 944                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 945
 946                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 947                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 948
 949                         error = gfs2_revoke_add(jd, blkno, start);
 950                         if (error < 0) {
 951                                 brelse(bh);
 952                                 return error;
 953                         }
 954                         else if (error)
 955                                 jd->jd_found_revokes++;
 956
 957                         if (!--revokes)
 958                                 break;
 959                         offset += sizeof(u64);
 960                 }
 961
 962                 brelse(bh);
 963                 offset = sizeof(struct gfs2_meta_header);
 964                 first = 0;
 965         }
 966
 967         return 0;
 968 }
 969
 970 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 971 {
 972         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 973
 974         if (error) {
 975                 gfs2_revoke_clean(jd);
 976                 return;
 977         }
 978         if (pass != 1)
 979                 return;
 980
 981         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 982                 jd->jd_jid, jd->jd_found_revokes);
 983
 984         gfs2_revoke_clean(jd);
 985 }
 986
 987 /**
 988  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 989  * @sdp: The filesystem
 990  * @tr: The system transaction being flushed
 991  */
 992
 993 static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 994 {
 995         unsigned int limit = databuf_limit(sdp);
 996         unsigned int nbuf;
 997         if (tr == NULL)
 998                 return;
 999         nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
1000         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1);
1001 }
1002
1003 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
1004                                     struct gfs2_log_descriptor *ld,
1005                                     __be64 *ptr, int pass)
1006 {
1007         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1008         struct gfs2_glock *gl = ip->i_gl;
1009         unsigned int blks = be32_to_cpu(ld->ld_data1);
1010         struct buffer_head *bh_log, *bh_ip;
1011         u64 blkno;
1012         u64 esc;
1013         int error = 0;
1014
1015         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
1016                 return 0;
1017
1018         gfs2_replay_incr_blk(jd, &start);
1019         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
1020                 blkno = be64_to_cpu(*ptr++);
1021                 esc = be64_to_cpu(*ptr++);
1022
1023                 jd->jd_found_blocks++;
1024
1025                 if (gfs2_revoke_check(jd, blkno, start))
1026                         continue;
1027
1028                 error = gfs2_replay_read_block(jd, start, &bh_log);
1029                 if (error)
1030                         return error;
1031
1032                 bh_ip = gfs2_meta_new(gl, blkno);
1033                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
1034
1035                 /* Unescape */
1036                 if (esc) {
1037                         __be32 *eptr = (__be32 *)bh_ip->b_data;
1038                         *eptr = cpu_to_be32(GFS2_MAGIC);
1039                 }
1040                 mark_buffer_dirty(bh_ip);
1041
1042                 brelse(bh_log);
1043                 brelse(bh_ip);
1044
1045                 jd->jd_replayed_blocks++;
1046         }
1047
1048         return error;
1049 }
1050
1051 /* FIXME: sort out accounting for log blocks etc. */
1052
1053 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
1054 {
1055         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1056         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
1057
1058         if (error) {
1059                 gfs2_inode_metasync(ip->i_gl);
1060                 return;
1061         }
1062         if (pass != 1)
1063                 return;
1064
1065         /* data sync? */
1066         gfs2_inode_metasync(ip->i_gl);
1067
1068         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
1069                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
1070 }
1071
1072 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1073 {
1074         struct list_head *head;
1075         struct gfs2_bufdata *bd;
1076
1077         if (tr == NULL)
1078                 return;
1079
1080         head = &tr->tr_databuf;
1081         while (!list_empty(head)) {
1082                 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
1083                 list_del_init(&bd->bd_list);
1084                 gfs2_unpin(sdp, bd->bd_bh, tr);
1085         }
1086 }
1087
1088
1089 static const struct gfs2_log_operations gfs2_buf_lops = {
1090         .lo_before_commit = buf_lo_before_commit,
1091         .lo_after_commit = buf_lo_after_commit,
1092         .lo_before_scan = buf_lo_before_scan,
1093         .lo_scan_elements = buf_lo_scan_elements,
1094         .lo_after_scan = buf_lo_after_scan,
1095         .lo_name = "buf",
1096 };
1097
1098 static const struct gfs2_log_operations gfs2_revoke_lops = {
1099         .lo_before_commit = revoke_lo_before_commit,
1100         .lo_after_commit = revoke_lo_after_commit,
1101         .lo_before_scan = revoke_lo_before_scan,
1102         .lo_scan_elements = revoke_lo_scan_elements,
1103         .lo_after_scan = revoke_lo_after_scan,
1104         .lo_name = "revoke",
1105 };
1106
1107 static const struct gfs2_log_operations gfs2_databuf_lops = {
1108         .lo_before_commit = databuf_lo_before_commit,
1109         .lo_after_commit = databuf_lo_after_commit,
1110         .lo_scan_elements = databuf_lo_scan_elements,
1111         .lo_after_scan = databuf_lo_after_scan,
1112         .lo_name = "databuf",
1113 };
1114
1115 const struct gfs2_log_operations *gfs2_log_ops[] = {
1116         &gfs2_databuf_lops,
1117         &gfs2_buf_lops,
1118         &gfs2_revoke_lops,
1119         NULL,
1120 };
1121