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 "log.h"
  26 #include "lops.h"
  27 #include "meta_io.h"
  28 #include "recovery.h"
  29 #include "rgrp.h"
  30 #include "trans.h"
  31 #include "util.h"
  32 #include "trace_gfs2.h"
  33
  34 /**
  35  * gfs2_pin - Pin a buffer in memory
  36  * @sdp: The superblock
  37  * @bh: The buffer to be pinned
  38  *
  39  * The log lock must be held when calling this function
  40  */
  41 void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
  42 {
  43         struct gfs2_bufdata *bd;
  44
  45         BUG_ON(!current->journal_info);
  46
  47         clear_buffer_dirty(bh);
  48         if (test_set_buffer_pinned(bh))
  49                 gfs2_assert_withdraw(sdp, 0);
  50         if (!buffer_uptodate(bh))
  51                 gfs2_io_error_bh_wd(sdp, bh);
  52         bd = bh->b_private;
  53         /* If this buffer is in the AIL and it has already been written
  54          * to in-place disk block, remove it from the AIL.
  55          */
  56         spin_lock(&sdp->sd_ail_lock);
  57         if (bd->bd_tr)
  58                 list_move(&bd->bd_ail_st_list, &bd->bd_tr->tr_ail2_list);
  59         spin_unlock(&sdp->sd_ail_lock);
  60         get_bh(bh);
  61         atomic_inc(&sdp->sd_log_pinned);
  62         trace_gfs2_pin(bd, 1);
  63 }
  64
  65 static bool buffer_is_rgrp(const struct gfs2_bufdata *bd)
  66 {
  67         return bd->bd_gl->gl_name.ln_type == LM_TYPE_RGRP;
  68 }
  69
  70 static void maybe_release_space(struct gfs2_bufdata *bd)
  71 {
  72         struct gfs2_glock *gl = bd->bd_gl;
  73         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
  74         struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
  75         unsigned int index = bd->bd_bh->b_blocknr - gl->gl_name.ln_number;
  76         struct gfs2_bitmap *bi = rgd->rd_bits + index;
  77
  78         if (bi->bi_clone == NULL)
  79                 return;
  80         if (sdp->sd_args.ar_discard)
  81                 gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi, 1, NULL);
  82         memcpy(bi->bi_clone + bi->bi_offset,
  83                bd->bd_bh->b_data + bi->bi_offset, bi->bi_bytes);
  84         clear_bit(GBF_FULL, &bi->bi_flags);
  85         rgd->rd_free_clone = rgd->rd_free;
  86         rgd->rd_extfail_pt = rgd->rd_free;
  87 }
  88
  89 /**
  90  * gfs2_unpin - Unpin a buffer
  91  * @sdp: the filesystem the buffer belongs to
  92  * @bh: The buffer to unpin
  93  * @ai:
  94  * @flags: The inode dirty flags
  95  *
  96  */
  97
  98 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
  99                        struct gfs2_trans *tr)
 100 {
 101         struct gfs2_bufdata *bd = bh->b_private;
 102
 103         BUG_ON(!buffer_uptodate(bh));
 104         BUG_ON(!buffer_pinned(bh));
 105
 106         lock_buffer(bh);
 107         mark_buffer_dirty(bh);
 108         clear_buffer_pinned(bh);
 109
 110         if (buffer_is_rgrp(bd))
 111                 maybe_release_space(bd);
 112
 113         spin_lock(&sdp->sd_ail_lock);
 114         if (bd->bd_tr) {
 115                 list_del(&bd->bd_ail_st_list);
 116                 brelse(bh);
 117         } else {
 118                 struct gfs2_glock *gl = bd->bd_gl;
 119                 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
 120                 atomic_inc(&gl->gl_ail_count);
 121         }
 122         bd->bd_tr = tr;
 123         list_add(&bd->bd_ail_st_list, &tr->tr_ail1_list);
 124         spin_unlock(&sdp->sd_ail_lock);
 125
 126         clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 127         trace_gfs2_pin(bd, 0);
 128         unlock_buffer(bh);
 129         atomic_dec(&sdp->sd_log_pinned);
 130 }
 131
 132 void gfs2_log_incr_head(struct gfs2_sbd *sdp)
 133 {
 134         BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) &&
 135                (sdp->sd_log_flush_head != sdp->sd_log_head));
 136
 137         if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks)
 138                 sdp->sd_log_flush_head = 0;
 139 }
 140
 141 u64 gfs2_log_bmap(struct gfs2_jdesc *jd, unsigned int lblock)
 142 {
 143         struct gfs2_journal_extent *je;
 144
 145         list_for_each_entry(je, &jd->extent_list, list) {
 146                 if (lblock >= je->lblock && lblock < je->lblock + je->blocks)
 147                         return je->dblock + lblock - je->lblock;
 148         }
 149
 150         return -1;
 151 }
 152
 153 /**
 154  * gfs2_end_log_write_bh - end log write of pagecache data with buffers
 155  * @sdp: The superblock
 156  * @bvec: The bio_vec
 157  * @error: The i/o status
 158  *
 159  * This finds the relevant buffers and unlocks them and sets the
 160  * error flag according to the status of the i/o request. This is
 161  * used when the log is writing data which has an in-place version
 162  * that is pinned in the pagecache.
 163  */
 164
 165 static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp,
 166                                   struct bio_vec *bvec,
 167                                   blk_status_t error)
 168 {
 169         struct buffer_head *bh, *next;
 170         struct page *page = bvec->bv_page;
 171         unsigned size;
 172
 173         bh = page_buffers(page);
 174         size = bvec->bv_len;
 175         while (bh_offset(bh) < bvec->bv_offset)
 176                 bh = bh->b_this_page;
 177         do {
 178                 if (error)
 179                         mark_buffer_write_io_error(bh);
 180                 unlock_buffer(bh);
 181                 next = bh->b_this_page;
 182                 size -= bh->b_size;
 183                 brelse(bh);
 184                 bh = next;
 185         } while(bh && size);
 186 }
 187
 188 /**
 189  * gfs2_end_log_write - end of i/o to the log
 190  * @bio: The bio
 191  *
 192  * Each bio_vec contains either data from the pagecache or data
 193  * relating to the log itself. Here we iterate over the bio_vec
 194  * array, processing both kinds of data.
 195  *
 196  */
 197
 198 static void gfs2_end_log_write(struct bio *bio)
 199 {
 200         struct gfs2_sbd *sdp = bio->bi_private;
 201         struct bio_vec *bvec;
 202         struct page *page;
 203         struct bvec_iter_all iter_all;
 204
 205         if (bio->bi_status) {
 206                 fs_err(sdp, "Error %d writing to journal, jid=%u\n",
 207                        bio->bi_status, sdp->sd_jdesc->jd_jid);
 208                 wake_up(&sdp->sd_logd_waitq);
 209         }
 210
 211         bio_for_each_segment_all(bvec, bio, iter_all) {
 212                 page = bvec->bv_page;
 213                 if (page_has_buffers(page))
 214                         gfs2_end_log_write_bh(sdp, bvec, bio->bi_status);
 215                 else
 216                         mempool_free(page, gfs2_page_pool);
 217         }
 218
 219         bio_put(bio);
 220         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 221                 wake_up(&sdp->sd_log_flush_wait);
 222 }
 223
 224 /**
 225  * gfs2_log_submit_bio - Submit any pending log bio
 226  * @biop: Address of the bio pointer
 227  * @opf: REQ_OP | op_flags
 228  *
 229  * Submit any pending part-built or full bio to the block device. If
 230  * there is no pending bio, then this is a no-op.
 231  */
 232
 233 void gfs2_log_submit_bio(struct bio **biop, int opf)
 234 {
 235         struct bio *bio = *biop;
 236         if (bio) {
 237                 struct gfs2_sbd *sdp = bio->bi_private;
 238                 atomic_inc(&sdp->sd_log_in_flight);
 239                 bio->bi_opf = opf;
 240                 submit_bio(bio);
 241                 *biop = NULL;
 242         }
 243 }
 244
 245 /**
 246  * gfs2_log_alloc_bio - Allocate a bio
 247  * @sdp: The super block
 248  * @blkno: The device block number we want to write to
 249  * @end_io: The bi_end_io callback
 250  *
 251  * Allocate a new bio, initialize it with the given parameters and return it.
 252  *
 253  * Returns: The newly allocated bio
 254  */
 255
 256 static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno,
 257                                       bio_end_io_t *end_io)
 258 {
 259         struct super_block *sb = sdp->sd_vfs;
 260         struct bio *bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
 261
 262         bio->bi_iter.bi_sector = blkno * (sb->s_blocksize >> 9);
 263         bio_set_dev(bio, sb->s_bdev);
 264         bio->bi_end_io = end_io;
 265         bio->bi_private = sdp;
 266
 267         return bio;
 268 }
 269
 270 /**
 271  * gfs2_log_get_bio - Get cached log bio, or allocate a new one
 272  * @sdp: The super block
 273  * @blkno: The device block number we want to write to
 274  * @bio: The bio to get or allocate
 275  * @op: REQ_OP
 276  * @end_io: The bi_end_io callback
 277  * @flush: Always flush the current bio and allocate a new one?
 278  *
 279  * If there is a cached bio, then if the next block number is sequential
 280  * with the previous one, return it, otherwise flush the bio to the
 281  * device. If there is no cached bio, or we just flushed it, then
 282  * allocate a new one.
 283  *
 284  * Returns: The bio to use for log writes
 285  */
 286
 287 static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno,
 288                                     struct bio **biop, int op,
 289                                     bio_end_io_t *end_io, bool flush)
 290 {
 291         struct bio *bio = *biop;
 292
 293         if (bio) {
 294                 u64 nblk;
 295
 296                 nblk = bio_end_sector(bio);
 297                 nblk >>= sdp->sd_fsb2bb_shift;
 298                 if (blkno == nblk && !flush)
 299                         return bio;
 300                 gfs2_log_submit_bio(biop, op);
 301         }
 302
 303         *biop = gfs2_log_alloc_bio(sdp, blkno, end_io);
 304         return *biop;
 305 }
 306
 307 /**
 308  * gfs2_log_write - write to log
 309  * @sdp: the filesystem
 310  * @page: the page to write
 311  * @size: the size of the data to write
 312  * @offset: the offset within the page
 313  * @blkno: block number of the log entry
 314  *
 315  * Try and add the page segment to the current bio. If that fails,
 316  * submit the current bio to the device and create a new one, and
 317  * then add the page segment to that.
 318  */
 319
 320 void gfs2_log_write(struct gfs2_sbd *sdp, struct page *page,
 321                     unsigned size, unsigned offset, u64 blkno)
 322 {
 323         struct bio *bio;
 324         int ret;
 325
 326         bio = gfs2_log_get_bio(sdp, blkno, &sdp->sd_log_bio, REQ_OP_WRITE,
 327                                gfs2_end_log_write, false);
 328         ret = bio_add_page(bio, page, size, offset);
 329         if (ret == 0) {
 330                 bio = gfs2_log_get_bio(sdp, blkno, &sdp->sd_log_bio,
 331                                        REQ_OP_WRITE, gfs2_end_log_write, true);
 332                 ret = bio_add_page(bio, page, size, offset);
 333                 WARN_ON(ret == 0);
 334         }
 335 }
 336
 337 /**
 338  * gfs2_log_write_bh - write a buffer's content to the log
 339  * @sdp: The super block
 340  * @bh: The buffer pointing to the in-place location
 341  *
 342  * This writes the content of the buffer to the next available location
 343  * in the log. The buffer will be unlocked once the i/o to the log has
 344  * completed.
 345  */
 346
 347 static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh)
 348 {
 349         u64 dblock;
 350
 351         dblock = gfs2_log_bmap(sdp->sd_jdesc, sdp->sd_log_flush_head);
 352         gfs2_log_incr_head(sdp);
 353         gfs2_log_write(sdp, bh->b_page, bh->b_size, bh_offset(bh), dblock);
 354 }
 355
 356 /**
 357  * gfs2_log_write_page - write one block stored in a page, into the log
 358  * @sdp: The superblock
 359  * @page: The struct page
 360  *
 361  * This writes the first block-sized part of the page into the log. Note
 362  * that the page must have been allocated from the gfs2_page_pool mempool
 363  * and that after this has been called, ownership has been transferred and
 364  * the page may be freed at any time.
 365  */
 366
 367 void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
 368 {
 369         struct super_block *sb = sdp->sd_vfs;
 370         u64 dblock;
 371
 372         dblock = gfs2_log_bmap(sdp->sd_jdesc, sdp->sd_log_flush_head);
 373         gfs2_log_incr_head(sdp);
 374         gfs2_log_write(sdp, page, sb->s_blocksize, 0, dblock);
 375 }
 376
 377 /**
 378  * gfs2_end_log_read - end I/O callback for reads from the log
 379  * @bio: The bio
 380  *
 381  * Simply unlock the pages in the bio. The main thread will wait on them and
 382  * process them in order as necessary.
 383  */
 384
 385 static void gfs2_end_log_read(struct bio *bio)
 386 {
 387         struct page *page;
 388         struct bio_vec *bvec;
 389         struct bvec_iter_all iter_all;
 390
 391         bio_for_each_segment_all(bvec, bio, iter_all) {
 392                 page = bvec->bv_page;
 393                 if (bio->bi_status) {
 394                         int err = blk_status_to_errno(bio->bi_status);
 395
 396                         SetPageError(page);
 397                         mapping_set_error(page->mapping, err);
 398                 }
 399                 unlock_page(page);
 400         }
 401
 402         bio_put(bio);
 403 }
 404
 405 /**
 406  * gfs2_jhead_pg_srch - Look for the journal head in a given page.
 407  * @jd: The journal descriptor
 408  * @page: The page to look in
 409  *
 410  * Returns: 1 if found, 0 otherwise.
 411  */
 412
 413 static bool gfs2_jhead_pg_srch(struct gfs2_jdesc *jd,
 414                               struct gfs2_log_header_host *head,
 415                               struct page *page)
 416 {
 417         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 418         struct gfs2_log_header_host uninitialized_var(lh);
 419         void *kaddr = kmap_atomic(page);
 420         unsigned int offset;
 421         bool ret = false;
 422
 423         for (offset = 0; offset < PAGE_SIZE; offset += sdp->sd_sb.sb_bsize) {
 424                 if (!__get_log_header(sdp, kaddr + offset, 0, &lh)) {
 425                         if (lh.lh_sequence > head->lh_sequence)
 426                                 *head = lh;
 427                         else {
 428                                 ret = true;
 429                                 break;
 430                         }
 431                 }
 432         }
 433         kunmap_atomic(kaddr);
 434         return ret;
 435 }
 436
 437 /**
 438  * gfs2_jhead_process_page - Search/cleanup a page
 439  * @jd: The journal descriptor
 440  * @index: Index of the page to look into
 441  * @done: If set, perform only cleanup, else search and set if found.
 442  *
 443  * Find the page with 'index' in the journal's mapping. Search the page for
 444  * the journal head if requested (cleanup == false). Release refs on the
 445  * page so the page cache can reclaim it (put_page() twice). We grabbed a
 446  * reference on this page two times, first when we did a find_or_create_page()
 447  * to obtain the page to add it to the bio and second when we do a
 448  * find_get_page() here to get the page to wait on while I/O on it is being
 449  * completed.
 450  * This function is also used to free up a page we might've grabbed but not
 451  * used. Maybe we added it to a bio, but not submitted it for I/O. Or we
 452  * submitted the I/O, but we already found the jhead so we only need to drop
 453  * our references to the page.
 454  */
 455
 456 static void gfs2_jhead_process_page(struct gfs2_jdesc *jd, unsigned long index,
 457                                     struct gfs2_log_header_host *head,
 458                                     bool *done)
 459 {
 460         struct page *page;
 461
 462         page = find_get_page(jd->jd_inode->i_mapping, index);
 463         wait_on_page_locked(page);
 464
 465         if (PageError(page))
 466                 *done = true;
 467
 468         if (!*done)
 469                 *done = gfs2_jhead_pg_srch(jd, head, page);
 470
 471         put_page(page); /* Once for find_get_page */
 472         put_page(page); /* Once more for find_or_create_page */
 473 }
 474
 475 /**
 476  * gfs2_find_jhead - find the head of a log
 477  * @jd: The journal descriptor
 478  * @head: The log descriptor for the head of the log is returned here
 479  *
 480  * Do a search of a journal by reading it in large chunks using bios and find
 481  * the valid log entry with the highest sequence number.  (i.e. the log head)
 482  *
 483  * Returns: 0 on success, errno otherwise
 484  */
 485 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head,
 486                     bool keep_cache)
 487 {
 488         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 489         struct address_space *mapping = jd->jd_inode->i_mapping;
 490         unsigned int block = 0, blocks_submitted = 0, blocks_read = 0;
 491         unsigned int bsize = sdp->sd_sb.sb_bsize;
 492         unsigned int bsize_shift = sdp->sd_sb.sb_bsize_shift;
 493         unsigned int shift = PAGE_SHIFT - bsize_shift;
 494         unsigned int readhead_blocks = BIO_MAX_PAGES << shift;
 495         struct gfs2_journal_extent *je;
 496         int sz, ret = 0;
 497         struct bio *bio = NULL;
 498         struct page *page = NULL;
 499         bool done = false;
 500         errseq_t since;
 501
 502         memset(head, 0, sizeof(*head));
 503         if (list_empty(&jd->extent_list))
 504                 gfs2_map_journal_extents(sdp, jd);
 505
 506         since = filemap_sample_wb_err(mapping);
 507         list_for_each_entry(je, &jd->extent_list, list) {
 508                 for (; block < je->lblock + je->blocks; block++) {
 509                         u64 dblock;
 510
 511                         if (!page) {
 512                                 page = find_or_create_page(mapping,
 513                                                 block >> shift, GFP_NOFS);
 514                                 if (!page) {
 515                                         ret = -ENOMEM;
 516                                         done = true;
 517                                         goto out;
 518                                 }
 519                         }
 520
 521                         if (bio) {
 522                                 unsigned int off;
 523
 524                                 off = (block << bsize_shift) & ~PAGE_MASK;
 525                                 sz = bio_add_page(bio, page, bsize, off);
 526                                 if (sz == bsize) { /* block added */
 527                                         if (off + bsize == PAGE_SIZE) {
 528                                                 page = NULL;
 529                                                 goto page_added;
 530                                         }
 531                                         continue;
 532                                 }
 533                                 blocks_submitted = block + 1;
 534                                 submit_bio(bio);
 535                                 bio = NULL;
 536                         }
 537
 538                         dblock = je->dblock + (block - je->lblock);
 539                         bio = gfs2_log_alloc_bio(sdp, dblock, gfs2_end_log_read);
 540                         bio->bi_opf = REQ_OP_READ;
 541                         sz = bio_add_page(bio, page, bsize, 0);
 542                         gfs2_assert_warn(sdp, sz == bsize);
 543                         if (bsize == PAGE_SIZE)
 544                                 page = NULL;
 545
 546 page_added:
 547                         if (blocks_submitted < blocks_read + readhead_blocks) {
 548                                 /* Keep at least one bio in flight */
 549                                 continue;
 550                         }
 551
 552                         gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done);
 553                         blocks_read += PAGE_SIZE >> bsize_shift;
 554                         if (done)
 555                                 goto out;  /* found */
 556                 }
 557         }
 558
 559 out:
 560         if (bio)
 561                 submit_bio(bio);
 562         while (blocks_read < block) {
 563                 gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done);
 564                 blocks_read += PAGE_SIZE >> bsize_shift;
 565         }
 566
 567         if (!ret)
 568                 ret = filemap_check_wb_err(mapping, since);
 569
 570         if (!keep_cache)
 571                 truncate_inode_pages(mapping, 0);
 572
 573         return ret;
 574 }
 575
 576 static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
 577                                       u32 ld_length, u32 ld_data1)
 578 {
 579         struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 580         struct gfs2_log_descriptor *ld = page_address(page);
 581         clear_page(ld);
 582         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 583         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 584         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 585         ld->ld_type = cpu_to_be32(ld_type);
 586         ld->ld_length = cpu_to_be32(ld_length);
 587         ld->ld_data1 = cpu_to_be32(ld_data1);
 588         ld->ld_data2 = 0;
 589         return page;
 590 }
 591
 592 static void gfs2_check_magic(struct buffer_head *bh)
 593 {
 594         void *kaddr;
 595         __be32 *ptr;
 596
 597         clear_buffer_escaped(bh);
 598         kaddr = kmap_atomic(bh->b_page);
 599         ptr = kaddr + bh_offset(bh);
 600         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 601                 set_buffer_escaped(bh);
 602         kunmap_atomic(kaddr);
 603 }
 604
 605 static int blocknr_cmp(void *priv, struct list_head *a, struct list_head *b)
 606 {
 607         struct gfs2_bufdata *bda, *bdb;
 608
 609         bda = list_entry(a, struct gfs2_bufdata, bd_list);
 610         bdb = list_entry(b, struct gfs2_bufdata, bd_list);
 611
 612         if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
 613                 return -1;
 614         if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
 615                 return 1;
 616         return 0;
 617 }
 618
 619 static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
 620                                 unsigned int total, struct list_head *blist,
 621                                 bool is_databuf)
 622 {
 623         struct gfs2_log_descriptor *ld;
 624         struct gfs2_bufdata *bd1 = NULL, *bd2;
 625         struct page *page;
 626         unsigned int num;
 627         unsigned n;
 628         __be64 *ptr;
 629
 630         gfs2_log_lock(sdp);
 631         list_sort(NULL, blist, blocknr_cmp);
 632         bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
 633         while(total) {
 634                 num = total;
 635                 if (total > limit)
 636                         num = limit;
 637                 gfs2_log_unlock(sdp);
 638                 page = gfs2_get_log_desc(sdp,
 639                                          is_databuf ? GFS2_LOG_DESC_JDATA :
 640                                          GFS2_LOG_DESC_METADATA, num + 1, num);
 641                 ld = page_address(page);
 642                 gfs2_log_lock(sdp);
 643                 ptr = (__be64 *)(ld + 1);
 644
 645                 n = 0;
 646                 list_for_each_entry_continue(bd1, blist, bd_list) {
 647                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 648                         if (is_databuf) {
 649                                 gfs2_check_magic(bd1->bd_bh);
 650                                 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
 651                         }
 652                         if (++n >= num)
 653                                 break;
 654                 }
 655
 656                 gfs2_log_unlock(sdp);
 657                 gfs2_log_write_page(sdp, page);
 658                 gfs2_log_lock(sdp);
 659
 660                 n = 0;
 661                 list_for_each_entry_continue(bd2, blist, bd_list) {
 662                         get_bh(bd2->bd_bh);
 663                         gfs2_log_unlock(sdp);
 664                         lock_buffer(bd2->bd_bh);
 665
 666                         if (buffer_escaped(bd2->bd_bh)) {
 667                                 void *kaddr;
 668                                 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 669                                 ptr = page_address(page);
 670                                 kaddr = kmap_atomic(bd2->bd_bh->b_page);
 671                                 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh),
 672                                        bd2->bd_bh->b_size);
 673                                 kunmap_atomic(kaddr);
 674                                 *(__be32 *)ptr = 0;
 675                                 clear_buffer_escaped(bd2->bd_bh);
 676                                 unlock_buffer(bd2->bd_bh);
 677                                 brelse(bd2->bd_bh);
 678                                 gfs2_log_write_page(sdp, page);
 679                         } else {
 680                                 gfs2_log_write_bh(sdp, bd2->bd_bh);
 681                         }
 682                         gfs2_log_lock(sdp);
 683                         if (++n >= num)
 684                                 break;
 685                 }
 686
 687                 BUG_ON(total < num);
 688                 total -= num;
 689         }
 690         gfs2_log_unlock(sdp);
 691 }
 692
 693 static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 694 {
 695         unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
 696         unsigned int nbuf;
 697         if (tr == NULL)
 698                 return;
 699         nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
 700         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0);
 701 }
 702
 703 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 704 {
 705         struct list_head *head;
 706         struct gfs2_bufdata *bd;
 707
 708         if (tr == NULL)
 709                 return;
 710
 711         head = &tr->tr_buf;
 712         while (!list_empty(head)) {
 713                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 714                 list_del_init(&bd->bd_list);
 715                 gfs2_unpin(sdp, bd->bd_bh, tr);
 716         }
 717 }
 718
 719 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 720                                struct gfs2_log_header_host *head, int pass)
 721 {
 722         if (pass != 0)
 723                 return;
 724
 725         jd->jd_found_blocks = 0;
 726         jd->jd_replayed_blocks = 0;
 727 }
 728
 729 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
 730                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 731                                 int pass)
 732 {
 733         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 734         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 735         struct gfs2_glock *gl = ip->i_gl;
 736         unsigned int blks = be32_to_cpu(ld->ld_data1);
 737         struct buffer_head *bh_log, *bh_ip;
 738         u64 blkno;
 739         int error = 0;
 740
 741         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 742                 return 0;
 743
 744         gfs2_replay_incr_blk(jd, &start);
 745
 746         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 747                 blkno = be64_to_cpu(*ptr++);
 748
 749                 jd->jd_found_blocks++;
 750
 751                 if (gfs2_revoke_check(jd, blkno, start))
 752                         continue;
 753
 754                 error = gfs2_replay_read_block(jd, start, &bh_log);
 755                 if (error)
 756                         return error;
 757
 758                 bh_ip = gfs2_meta_new(gl, blkno);
 759                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 760
 761                 if (gfs2_meta_check(sdp, bh_ip))
 762                         error = -EIO;
 763                 else {
 764                         struct gfs2_meta_header *mh =
 765                                 (struct gfs2_meta_header *)bh_ip->b_data;
 766
 767                         if (mh->mh_type == cpu_to_be32(GFS2_METATYPE_RG)) {
 768                                 struct gfs2_rgrpd *rgd;
 769
 770                                 rgd = gfs2_blk2rgrpd(sdp, blkno, false);
 771                                 if (rgd && rgd->rd_addr == blkno &&
 772                                     rgd->rd_bits && rgd->rd_bits->bi_bh) {
 773                                         fs_info(sdp, "Replaying 0x%llx but we "
 774                                                 "already have a bh!\n",
 775                                                 (unsigned long long)blkno);
 776                                         fs_info(sdp, "busy:%d, pinned:%d\n",
 777                                                 buffer_busy(rgd->rd_bits->bi_bh) ? 1 : 0,
 778                                                 buffer_pinned(rgd->rd_bits->bi_bh));
 779                                         gfs2_dump_glock(NULL, rgd->rd_gl, true);
 780                                 }
 781                         }
 782                         mark_buffer_dirty(bh_ip);
 783                 }
 784                 brelse(bh_log);
 785                 brelse(bh_ip);
 786
 787                 if (error)
 788                         break;
 789
 790                 jd->jd_replayed_blocks++;
 791         }
 792
 793         return error;
 794 }
 795
 796 /**
 797  * gfs2_meta_sync - Sync all buffers associated with a glock
 798  * @gl: The glock
 799  *
 800  */
 801
 802 static void gfs2_meta_sync(struct gfs2_glock *gl)
 803 {
 804         struct address_space *mapping = gfs2_glock2aspace(gl);
 805         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 806         int error;
 807
 808         if (mapping == NULL)
 809                 mapping = &sdp->sd_aspace;
 810
 811         filemap_fdatawrite(mapping);
 812         error = filemap_fdatawait(mapping);
 813
 814         if (error)
 815                 gfs2_io_error(gl->gl_name.ln_sbd);
 816 }
 817
 818 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 819 {
 820         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 821         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 822
 823         if (error) {
 824                 gfs2_meta_sync(ip->i_gl);
 825                 return;
 826         }
 827         if (pass != 1)
 828                 return;
 829
 830         gfs2_meta_sync(ip->i_gl);
 831
 832         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 833                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
 834 }
 835
 836 static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 837 {
 838         struct gfs2_meta_header *mh;
 839         unsigned int offset;
 840         struct list_head *head = &sdp->sd_log_revokes;
 841         struct gfs2_bufdata *bd;
 842         struct page *page;
 843         unsigned int length;
 844
 845         gfs2_write_revokes(sdp);
 846         if (!sdp->sd_log_num_revoke)
 847                 return;
 848
 849         length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(u64));
 850         page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
 851         offset = sizeof(struct gfs2_log_descriptor);
 852
 853         list_for_each_entry(bd, head, bd_list) {
 854                 sdp->sd_log_num_revoke--;
 855
 856                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 857
 858                         gfs2_log_write_page(sdp, page);
 859                         page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 860                         mh = page_address(page);
 861                         clear_page(mh);
 862                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 863                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 864                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 865                         offset = sizeof(struct gfs2_meta_header);
 866                 }
 867
 868                 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
 869                 offset += sizeof(u64);
 870         }
 871         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 872
 873         gfs2_log_write_page(sdp, page);
 874 }
 875
 876 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 877 {
 878         struct list_head *head = &sdp->sd_log_revokes;
 879         struct gfs2_bufdata *bd;
 880         struct gfs2_glock *gl;
 881
 882         while (!list_empty(head)) {
 883                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 884                 list_del_init(&bd->bd_list);
 885                 gl = bd->bd_gl;
 886                 gfs2_glock_remove_revoke(gl);
 887                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 888         }
 889 }
 890
 891 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 892                                   struct gfs2_log_header_host *head, int pass)
 893 {
 894         if (pass != 0)
 895                 return;
 896
 897         jd->jd_found_revokes = 0;
 898         jd->jd_replay_tail = head->lh_tail;
 899 }
 900
 901 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
 902                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 903                                    int pass)
 904 {
 905         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 906         unsigned int blks = be32_to_cpu(ld->ld_length);
 907         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 908         struct buffer_head *bh;
 909         unsigned int offset;
 910         u64 blkno;
 911         int first = 1;
 912         int error;
 913
 914         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 915                 return 0;
 916
 917         offset = sizeof(struct gfs2_log_descriptor);
 918
 919         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 920                 error = gfs2_replay_read_block(jd, start, &bh);
 921                 if (error)
 922                         return error;
 923
 924                 if (!first)
 925                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 926
 927                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 928                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 929
 930                         error = gfs2_revoke_add(jd, blkno, start);
 931                         if (error < 0) {
 932                                 brelse(bh);
 933                                 return error;
 934                         }
 935                         else if (error)
 936                                 jd->jd_found_revokes++;
 937
 938                         if (!--revokes)
 939                                 break;
 940                         offset += sizeof(u64);
 941                 }
 942
 943                 brelse(bh);
 944                 offset = sizeof(struct gfs2_meta_header);
 945                 first = 0;
 946         }
 947
 948         return 0;
 949 }
 950
 951 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 952 {
 953         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 954
 955         if (error) {
 956                 gfs2_revoke_clean(jd);
 957                 return;
 958         }
 959         if (pass != 1)
 960                 return;
 961
 962         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 963                 jd->jd_jid, jd->jd_found_revokes);
 964
 965         gfs2_revoke_clean(jd);
 966 }
 967
 968 /**
 969  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 970  *
 971  */
 972
 973 static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 974 {
 975         unsigned int limit = databuf_limit(sdp);
 976         unsigned int nbuf;
 977         if (tr == NULL)
 978                 return;
 979         nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
 980         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1);
 981 }
 982
 983 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
 984                                     struct gfs2_log_descriptor *ld,
 985                                     __be64 *ptr, int pass)
 986 {
 987         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 988         struct gfs2_glock *gl = ip->i_gl;
 989         unsigned int blks = be32_to_cpu(ld->ld_data1);
 990         struct buffer_head *bh_log, *bh_ip;
 991         u64 blkno;
 992         u64 esc;
 993         int error = 0;
 994
 995         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
 996                 return 0;
 997
 998         gfs2_replay_incr_blk(jd, &start);
 999         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
1000                 blkno = be64_to_cpu(*ptr++);
1001                 esc = be64_to_cpu(*ptr++);
1002
1003                 jd->jd_found_blocks++;
1004
1005                 if (gfs2_revoke_check(jd, blkno, start))
1006                         continue;
1007
1008                 error = gfs2_replay_read_block(jd, start, &bh_log);
1009                 if (error)
1010                         return error;
1011
1012                 bh_ip = gfs2_meta_new(gl, blkno);
1013                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
1014
1015                 /* Unescape */
1016                 if (esc) {
1017                         __be32 *eptr = (__be32 *)bh_ip->b_data;
1018                         *eptr = cpu_to_be32(GFS2_MAGIC);
1019                 }
1020                 mark_buffer_dirty(bh_ip);
1021
1022                 brelse(bh_log);
1023                 brelse(bh_ip);
1024
1025                 jd->jd_replayed_blocks++;
1026         }
1027
1028         return error;
1029 }
1030
1031 /* FIXME: sort out accounting for log blocks etc. */
1032
1033 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
1034 {
1035         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1036         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
1037
1038         if (error) {
1039                 gfs2_meta_sync(ip->i_gl);
1040                 return;
1041         }
1042         if (pass != 1)
1043                 return;
1044
1045         /* data sync? */
1046         gfs2_meta_sync(ip->i_gl);
1047
1048         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
1049                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
1050 }
1051
1052 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1053 {
1054         struct list_head *head;
1055         struct gfs2_bufdata *bd;
1056
1057         if (tr == NULL)
1058                 return;
1059
1060         head = &tr->tr_databuf;
1061         while (!list_empty(head)) {
1062                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
1063                 list_del_init(&bd->bd_list);
1064                 gfs2_unpin(sdp, bd->bd_bh, tr);
1065         }
1066 }
1067
1068
1069 static const struct gfs2_log_operations gfs2_buf_lops = {
1070         .lo_before_commit = buf_lo_before_commit,
1071         .lo_after_commit = buf_lo_after_commit,
1072         .lo_before_scan = buf_lo_before_scan,
1073         .lo_scan_elements = buf_lo_scan_elements,
1074         .lo_after_scan = buf_lo_after_scan,
1075         .lo_name = "buf",
1076 };
1077
1078 static const struct gfs2_log_operations gfs2_revoke_lops = {
1079         .lo_before_commit = revoke_lo_before_commit,
1080         .lo_after_commit = revoke_lo_after_commit,
1081         .lo_before_scan = revoke_lo_before_scan,
1082         .lo_scan_elements = revoke_lo_scan_elements,
1083         .lo_after_scan = revoke_lo_after_scan,
1084         .lo_name = "revoke",
1085 };
1086
1087 static const struct gfs2_log_operations gfs2_databuf_lops = {
1088         .lo_before_commit = databuf_lo_before_commit,
1089         .lo_after_commit = databuf_lo_after_commit,
1090         .lo_scan_elements = databuf_lo_scan_elements,
1091         .lo_after_scan = databuf_lo_after_scan,
1092         .lo_name = "databuf",
1093 };
1094
1095 const struct gfs2_log_operations *gfs2_log_ops[] = {
1096         &gfs2_databuf_lops,
1097         &gfs2_buf_lops,
1098         &gfs2_revoke_lops,
1099         NULL,
1100 };
1101