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                 if (!cmpxchg(&sdp->sd_log_error, 0, (int)bio->bi_status))
 207                         fs_err(sdp, "Error %d writing to journal, jid=%u\n",
 208                                bio->bi_status, sdp->sd_jdesc->jd_jid);
 209                 gfs2_withdraw_delayed(sdp);
 210                 /* prevent more writes to the journal */
 211                 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 212                 wake_up(&sdp->sd_logd_waitq);
 213         }
 214
 215         bio_for_each_segment_all(bvec, bio, iter_all) {
 216                 page = bvec->bv_page;
 217                 if (page_has_buffers(page))
 218                         gfs2_end_log_write_bh(sdp, bvec, bio->bi_status);
 219                 else
 220                         mempool_free(page, gfs2_page_pool);
 221         }
 222
 223         bio_put(bio);
 224         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 225                 wake_up(&sdp->sd_log_flush_wait);
 226 }
 227
 228 /**
 229  * gfs2_log_submit_bio - Submit any pending log bio
 230  * @biop: Address of the bio pointer
 231  * @opf: REQ_OP | op_flags
 232  *
 233  * Submit any pending part-built or full bio to the block device. If
 234  * there is no pending bio, then this is a no-op.
 235  */
 236
 237 void gfs2_log_submit_bio(struct bio **biop, int opf)
 238 {
 239         struct bio *bio = *biop;
 240         if (bio) {
 241                 struct gfs2_sbd *sdp = bio->bi_private;
 242                 atomic_inc(&sdp->sd_log_in_flight);
 243                 bio->bi_opf = opf;
 244                 submit_bio(bio);
 245                 *biop = NULL;
 246         }
 247 }
 248
 249 /**
 250  * gfs2_log_alloc_bio - Allocate a bio
 251  * @sdp: The super block
 252  * @blkno: The device block number we want to write to
 253  * @end_io: The bi_end_io callback
 254  *
 255  * Allocate a new bio, initialize it with the given parameters and return it.
 256  *
 257  * Returns: The newly allocated bio
 258  */
 259
 260 static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno,
 261                                       bio_end_io_t *end_io)
 262 {
 263         struct super_block *sb = sdp->sd_vfs;
 264         struct bio *bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
 265
 266         bio->bi_iter.bi_sector = blkno << sdp->sd_fsb2bb_shift;
 267         bio_set_dev(bio, sb->s_bdev);
 268         bio->bi_end_io = end_io;
 269         bio->bi_private = sdp;
 270
 271         return bio;
 272 }
 273
 274 /**
 275  * gfs2_log_get_bio - Get cached log bio, or allocate a new one
 276  * @sdp: The super block
 277  * @blkno: The device block number we want to write to
 278  * @bio: The bio to get or allocate
 279  * @op: REQ_OP
 280  * @end_io: The bi_end_io callback
 281  * @flush: Always flush the current bio and allocate a new one?
 282  *
 283  * If there is a cached bio, then if the next block number is sequential
 284  * with the previous one, return it, otherwise flush the bio to the
 285  * device. If there is no cached bio, or we just flushed it, then
 286  * allocate a new one.
 287  *
 288  * Returns: The bio to use for log writes
 289  */
 290
 291 static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno,
 292                                     struct bio **biop, int op,
 293                                     bio_end_io_t *end_io, bool flush)
 294 {
 295         struct bio *bio = *biop;
 296
 297         if (bio) {
 298                 u64 nblk;
 299
 300                 nblk = bio_end_sector(bio);
 301                 nblk >>= sdp->sd_fsb2bb_shift;
 302                 if (blkno == nblk && !flush)
 303                         return bio;
 304                 gfs2_log_submit_bio(biop, op);
 305         }
 306
 307         *biop = gfs2_log_alloc_bio(sdp, blkno, end_io);
 308         return *biop;
 309 }
 310
 311 /**
 312  * gfs2_log_write - write to log
 313  * @sdp: the filesystem
 314  * @page: the page to write
 315  * @size: the size of the data to write
 316  * @offset: the offset within the page
 317  * @blkno: block number of the log entry
 318  *
 319  * Try and add the page segment to the current bio. If that fails,
 320  * submit the current bio to the device and create a new one, and
 321  * then add the page segment to that.
 322  */
 323
 324 void gfs2_log_write(struct gfs2_sbd *sdp, struct page *page,
 325                     unsigned size, unsigned offset, u64 blkno)
 326 {
 327         struct bio *bio;
 328         int ret;
 329
 330         bio = gfs2_log_get_bio(sdp, blkno, &sdp->sd_log_bio, REQ_OP_WRITE,
 331                                gfs2_end_log_write, false);
 332         ret = bio_add_page(bio, page, size, offset);
 333         if (ret == 0) {
 334                 bio = gfs2_log_get_bio(sdp, blkno, &sdp->sd_log_bio,
 335                                        REQ_OP_WRITE, gfs2_end_log_write, true);
 336                 ret = bio_add_page(bio, page, size, offset);
 337                 WARN_ON(ret == 0);
 338         }
 339 }
 340
 341 /**
 342  * gfs2_log_write_bh - write a buffer's content to the log
 343  * @sdp: The super block
 344  * @bh: The buffer pointing to the in-place location
 345  *
 346  * This writes the content of the buffer to the next available location
 347  * in the log. The buffer will be unlocked once the i/o to the log has
 348  * completed.
 349  */
 350
 351 static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh)
 352 {
 353         u64 dblock;
 354
 355         dblock = gfs2_log_bmap(sdp->sd_jdesc, sdp->sd_log_flush_head);
 356         gfs2_log_incr_head(sdp);
 357         gfs2_log_write(sdp, bh->b_page, bh->b_size, bh_offset(bh), dblock);
 358 }
 359
 360 /**
 361  * gfs2_log_write_page - write one block stored in a page, into the log
 362  * @sdp: The superblock
 363  * @page: The struct page
 364  *
 365  * This writes the first block-sized part of the page into the log. Note
 366  * that the page must have been allocated from the gfs2_page_pool mempool
 367  * and that after this has been called, ownership has been transferred and
 368  * the page may be freed at any time.
 369  */
 370
 371 void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
 372 {
 373         struct super_block *sb = sdp->sd_vfs;
 374         u64 dblock;
 375
 376         dblock = gfs2_log_bmap(sdp->sd_jdesc, sdp->sd_log_flush_head);
 377         gfs2_log_incr_head(sdp);
 378         gfs2_log_write(sdp, page, sb->s_blocksize, 0, dblock);
 379 }
 380
 381 /**
 382  * gfs2_end_log_read - end I/O callback for reads from the log
 383  * @bio: The bio
 384  *
 385  * Simply unlock the pages in the bio. The main thread will wait on them and
 386  * process them in order as necessary.
 387  */
 388
 389 static void gfs2_end_log_read(struct bio *bio)
 390 {
 391         struct page *page;
 392         struct bio_vec *bvec;
 393         struct bvec_iter_all iter_all;
 394
 395         bio_for_each_segment_all(bvec, bio, iter_all) {
 396                 page = bvec->bv_page;
 397                 if (bio->bi_status) {
 398                         int err = blk_status_to_errno(bio->bi_status);
 399
 400                         SetPageError(page);
 401                         mapping_set_error(page->mapping, err);
 402                 }
 403                 unlock_page(page);
 404         }
 405
 406         bio_put(bio);
 407 }
 408
 409 /**
 410  * gfs2_jhead_pg_srch - Look for the journal head in a given page.
 411  * @jd: The journal descriptor
 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 uninitialized_var(lh);
 423         void *kaddr = kmap_atomic(page);
 424         unsigned int offset;
 425         bool ret = false;
 426
 427         for (offset = 0; offset < PAGE_SIZE; offset += sdp->sd_sb.sb_bsize) {
 428                 if (!__get_log_header(sdp, kaddr + offset, 0, &lh)) {
 429                         if (lh.lh_sequence >= head->lh_sequence)
 430                                 *head = lh;
 431                         else {
 432                                 ret = true;
 433                                 break;
 434                         }
 435                 }
 436         }
 437         kunmap_atomic(kaddr);
 438         return ret;
 439 }
 440
 441 /**
 442  * gfs2_jhead_process_page - Search/cleanup a page
 443  * @jd: The journal descriptor
 444  * @index: Index of the page to look into
 445  * @done: If set, perform only cleanup, else search and set if found.
 446  *
 447  * Find the page with 'index' in the journal's mapping. Search the page for
 448  * the journal head if requested (cleanup == false). Release refs on the
 449  * page so the page cache can reclaim it (put_page() twice). We grabbed a
 450  * reference on this page two times, first when we did a find_or_create_page()
 451  * to obtain the page to add it to the bio and second when we do a
 452  * find_get_page() here to get the page to wait on while I/O on it is being
 453  * completed.
 454  * This function is also used to free up a page we might've grabbed but not
 455  * used. Maybe we added it to a bio, but not submitted it for I/O. Or we
 456  * submitted the I/O, but we already found the jhead so we only need to drop
 457  * our references to the page.
 458  */
 459
 460 static void gfs2_jhead_process_page(struct gfs2_jdesc *jd, unsigned long index,
 461                                     struct gfs2_log_header_host *head,
 462                                     bool *done)
 463 {
 464         struct page *page;
 465
 466         page = find_get_page(jd->jd_inode->i_mapping, index);
 467         wait_on_page_locked(page);
 468
 469         if (PageError(page))
 470                 *done = true;
 471
 472         if (!*done)
 473                 *done = gfs2_jhead_pg_srch(jd, head, page);
 474
 475         put_page(page); /* Once for find_get_page */
 476         put_page(page); /* Once more for find_or_create_page */
 477 }
 478
 479 static struct bio *gfs2_chain_bio(struct bio *prev, unsigned int nr_iovecs)
 480 {
 481         struct bio *new;
 482
 483         new = bio_alloc(GFP_NOIO, nr_iovecs);
 484         bio_copy_dev(new, prev);
 485         new->bi_iter.bi_sector = bio_end_sector(prev);
 486         new->bi_opf = prev->bi_opf;
 487         new->bi_write_hint = prev->bi_write_hint;
 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  *
 498  * Do a search of a journal by reading it in large chunks using bios and find
 499  * the valid log entry with the highest sequence number.  (i.e. the log head)
 500  *
 501  * Returns: 0 on success, errno otherwise
 502  */
 503 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head,
 504                     bool keep_cache)
 505 {
 506         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 507         struct address_space *mapping = jd->jd_inode->i_mapping;
 508         unsigned int block = 0, blocks_submitted = 0, blocks_read = 0;
 509         unsigned int bsize = sdp->sd_sb.sb_bsize, off;
 510         unsigned int bsize_shift = sdp->sd_sb.sb_bsize_shift;
 511         unsigned int shift = PAGE_SHIFT - bsize_shift;
 512         unsigned int max_bio_size = 2 * 1024 * 1024;
 513         struct gfs2_journal_extent *je;
 514         int sz, ret = 0;
 515         struct bio *bio = NULL;
 516         struct page *page = NULL;
 517         bool bio_chained = false, done = false;
 518         errseq_t since;
 519
 520         memset(head, 0, sizeof(*head));
 521         if (list_empty(&jd->extent_list))
 522                 gfs2_map_journal_extents(sdp, jd);
 523
 524         since = filemap_sample_wb_err(mapping);
 525         list_for_each_entry(je, &jd->extent_list, list) {
 526                 u64 dblock = je->dblock;
 527
 528                 for (; block < je->lblock + je->blocks; block++, dblock++) {
 529                         if (!page) {
 530                                 page = find_or_create_page(mapping,
 531                                                 block >> shift, GFP_NOFS);
 532                                 if (!page) {
 533                                         ret = -ENOMEM;
 534                                         done = true;
 535                                         goto out;
 536                                 }
 537                                 off = 0;
 538                         }
 539
 540                         if (!bio || (bio_chained && !off) ||
 541                             bio->bi_iter.bi_size >= max_bio_size) {
 542                                 /* start new bio */
 543                         } else {
 544                                 sector_t sector = dblock << sdp->sd_fsb2bb_shift;
 545
 546                                 if (bio_end_sector(bio) == sector) {
 547                                         sz = bio_add_page(bio, page, bsize, off);
 548                                         if (sz == bsize)
 549                                                 goto block_added;
 550                                 }
 551                                 if (off) {
 552                                         unsigned int blocks =
 553                                                 (PAGE_SIZE - off) >> bsize_shift;
 554
 555                                         bio = gfs2_chain_bio(bio, blocks);
 556                                         bio_chained = true;
 557                                         goto add_block_to_new_bio;
 558                                 }
 559                         }
 560
 561                         if (bio) {
 562                                 blocks_submitted = block + 1;
 563                                 submit_bio(bio);
 564                         }
 565
 566                         bio = gfs2_log_alloc_bio(sdp, dblock, gfs2_end_log_read);
 567                         bio->bi_opf = REQ_OP_READ;
 568                         bio_chained = false;
 569 add_block_to_new_bio:
 570                         sz = bio_add_page(bio, page, bsize, off);
 571                         BUG_ON(sz != bsize);
 572 block_added:
 573                         off += bsize;
 574                         if (off == PAGE_SIZE)
 575                                 page = NULL;
 576                         if (blocks_submitted < 2 * max_bio_size >> bsize_shift) {
 577                                 /* Keep at least one bio in flight */
 578                                 continue;
 579                         }
 580
 581                         gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done);
 582                         blocks_read += PAGE_SIZE >> bsize_shift;
 583                         if (done)
 584                                 goto out;  /* found */
 585                 }
 586         }
 587
 588 out:
 589         if (bio)
 590                 submit_bio(bio);
 591         while (blocks_read < block) {
 592                 gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done);
 593                 blocks_read += PAGE_SIZE >> bsize_shift;
 594         }
 595
 596         if (!ret)
 597                 ret = filemap_check_wb_err(mapping, since);
 598
 599         if (!keep_cache)
 600                 truncate_inode_pages(mapping, 0);
 601
 602         return ret;
 603 }
 604
 605 static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
 606                                       u32 ld_length, u32 ld_data1)
 607 {
 608         struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 609         struct gfs2_log_descriptor *ld = page_address(page);
 610         clear_page(ld);
 611         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 612         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 613         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 614         ld->ld_type = cpu_to_be32(ld_type);
 615         ld->ld_length = cpu_to_be32(ld_length);
 616         ld->ld_data1 = cpu_to_be32(ld_data1);
 617         ld->ld_data2 = 0;
 618         return page;
 619 }
 620
 621 static void gfs2_check_magic(struct buffer_head *bh)
 622 {
 623         void *kaddr;
 624         __be32 *ptr;
 625
 626         clear_buffer_escaped(bh);
 627         kaddr = kmap_atomic(bh->b_page);
 628         ptr = kaddr + bh_offset(bh);
 629         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 630                 set_buffer_escaped(bh);
 631         kunmap_atomic(kaddr);
 632 }
 633
 634 static int blocknr_cmp(void *priv, struct list_head *a, struct list_head *b)
 635 {
 636         struct gfs2_bufdata *bda, *bdb;
 637
 638         bda = list_entry(a, struct gfs2_bufdata, bd_list);
 639         bdb = list_entry(b, struct gfs2_bufdata, bd_list);
 640
 641         if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
 642                 return -1;
 643         if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
 644                 return 1;
 645         return 0;
 646 }
 647
 648 static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
 649                                 unsigned int total, struct list_head *blist,
 650                                 bool is_databuf)
 651 {
 652         struct gfs2_log_descriptor *ld;
 653         struct gfs2_bufdata *bd1 = NULL, *bd2;
 654         struct page *page;
 655         unsigned int num;
 656         unsigned n;
 657         __be64 *ptr;
 658
 659         gfs2_log_lock(sdp);
 660         list_sort(NULL, blist, blocknr_cmp);
 661         bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
 662         while(total) {
 663                 num = total;
 664                 if (total > limit)
 665                         num = limit;
 666                 gfs2_log_unlock(sdp);
 667                 page = gfs2_get_log_desc(sdp,
 668                                          is_databuf ? GFS2_LOG_DESC_JDATA :
 669                                          GFS2_LOG_DESC_METADATA, num + 1, num);
 670                 ld = page_address(page);
 671                 gfs2_log_lock(sdp);
 672                 ptr = (__be64 *)(ld + 1);
 673
 674                 n = 0;
 675                 list_for_each_entry_continue(bd1, blist, bd_list) {
 676                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 677                         if (is_databuf) {
 678                                 gfs2_check_magic(bd1->bd_bh);
 679                                 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
 680                         }
 681                         if (++n >= num)
 682                                 break;
 683                 }
 684
 685                 gfs2_log_unlock(sdp);
 686                 gfs2_log_write_page(sdp, page);
 687                 gfs2_log_lock(sdp);
 688
 689                 n = 0;
 690                 list_for_each_entry_continue(bd2, blist, bd_list) {
 691                         get_bh(bd2->bd_bh);
 692                         gfs2_log_unlock(sdp);
 693                         lock_buffer(bd2->bd_bh);
 694
 695                         if (buffer_escaped(bd2->bd_bh)) {
 696                                 void *kaddr;
 697                                 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 698                                 ptr = page_address(page);
 699                                 kaddr = kmap_atomic(bd2->bd_bh->b_page);
 700                                 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh),
 701                                        bd2->bd_bh->b_size);
 702                                 kunmap_atomic(kaddr);
 703                                 *(__be32 *)ptr = 0;
 704                                 clear_buffer_escaped(bd2->bd_bh);
 705                                 unlock_buffer(bd2->bd_bh);
 706                                 brelse(bd2->bd_bh);
 707                                 gfs2_log_write_page(sdp, page);
 708                         } else {
 709                                 gfs2_log_write_bh(sdp, bd2->bd_bh);
 710                         }
 711                         gfs2_log_lock(sdp);
 712                         if (++n >= num)
 713                                 break;
 714                 }
 715
 716                 BUG_ON(total < num);
 717                 total -= num;
 718         }
 719         gfs2_log_unlock(sdp);
 720 }
 721
 722 static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 723 {
 724         unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
 725         unsigned int nbuf;
 726         if (tr == NULL)
 727                 return;
 728         nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
 729         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0);
 730 }
 731
 732 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 733 {
 734         struct list_head *head;
 735         struct gfs2_bufdata *bd;
 736
 737         if (tr == NULL)
 738                 return;
 739
 740         head = &tr->tr_buf;
 741         while (!list_empty(head)) {
 742                 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
 743                 list_del_init(&bd->bd_list);
 744                 gfs2_unpin(sdp, bd->bd_bh, tr);
 745         }
 746 }
 747
 748 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 749                                struct gfs2_log_header_host *head, int pass)
 750 {
 751         if (pass != 0)
 752                 return;
 753
 754         jd->jd_found_blocks = 0;
 755         jd->jd_replayed_blocks = 0;
 756 }
 757
 758 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
 759                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 760                                 int pass)
 761 {
 762         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 763         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 764         struct gfs2_glock *gl = ip->i_gl;
 765         unsigned int blks = be32_to_cpu(ld->ld_data1);
 766         struct buffer_head *bh_log, *bh_ip;
 767         u64 blkno;
 768         int error = 0;
 769
 770         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 771                 return 0;
 772
 773         gfs2_replay_incr_blk(jd, &start);
 774
 775         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 776                 blkno = be64_to_cpu(*ptr++);
 777
 778                 jd->jd_found_blocks++;
 779
 780                 if (gfs2_revoke_check(jd, blkno, start))
 781                         continue;
 782
 783                 error = gfs2_replay_read_block(jd, start, &bh_log);
 784                 if (error)
 785                         return error;
 786
 787                 bh_ip = gfs2_meta_new(gl, blkno);
 788                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 789
 790                 if (gfs2_meta_check(sdp, bh_ip))
 791                         error = -EIO;
 792                 else {
 793                         struct gfs2_meta_header *mh =
 794                                 (struct gfs2_meta_header *)bh_ip->b_data;
 795
 796                         if (mh->mh_type == cpu_to_be32(GFS2_METATYPE_RG)) {
 797                                 struct gfs2_rgrpd *rgd;
 798
 799                                 rgd = gfs2_blk2rgrpd(sdp, blkno, false);
 800                                 if (rgd && rgd->rd_addr == blkno &&
 801                                     rgd->rd_bits && rgd->rd_bits->bi_bh) {
 802                                         fs_info(sdp, "Replaying 0x%llx but we "
 803                                                 "already have a bh!\n",
 804                                                 (unsigned long long)blkno);
 805                                         fs_info(sdp, "busy:%d, pinned:%d\n",
 806                                                 buffer_busy(rgd->rd_bits->bi_bh) ? 1 : 0,
 807                                                 buffer_pinned(rgd->rd_bits->bi_bh));
 808                                         gfs2_dump_glock(NULL, rgd->rd_gl, true);
 809                                 }
 810                         }
 811                         mark_buffer_dirty(bh_ip);
 812                 }
 813                 brelse(bh_log);
 814                 brelse(bh_ip);
 815
 816                 if (error)
 817                         break;
 818
 819                 jd->jd_replayed_blocks++;
 820         }
 821
 822         return error;
 823 }
 824
 825 /**
 826  * gfs2_meta_sync - Sync all buffers associated with a glock
 827  * @gl: The glock
 828  *
 829  */
 830
 831 static void gfs2_meta_sync(struct gfs2_glock *gl)
 832 {
 833         struct address_space *mapping = gfs2_glock2aspace(gl);
 834         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 835         int error;
 836
 837         if (mapping == NULL)
 838                 mapping = &sdp->sd_aspace;
 839
 840         filemap_fdatawrite(mapping);
 841         error = filemap_fdatawait(mapping);
 842
 843         if (error)
 844                 gfs2_io_error(gl->gl_name.ln_sbd);
 845 }
 846
 847 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 848 {
 849         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 850         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 851
 852         if (error) {
 853                 gfs2_meta_sync(ip->i_gl);
 854                 return;
 855         }
 856         if (pass != 1)
 857                 return;
 858
 859         gfs2_meta_sync(ip->i_gl);
 860
 861         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 862                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
 863 }
 864
 865 static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 866 {
 867         struct gfs2_meta_header *mh;
 868         unsigned int offset;
 869         struct list_head *head = &sdp->sd_log_revokes;
 870         struct gfs2_bufdata *bd;
 871         struct page *page;
 872         unsigned int length;
 873
 874         gfs2_write_revokes(sdp);
 875         if (!sdp->sd_log_num_revoke)
 876                 return;
 877
 878         length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke);
 879         page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
 880         offset = sizeof(struct gfs2_log_descriptor);
 881
 882         list_for_each_entry(bd, head, bd_list) {
 883                 sdp->sd_log_num_revoke--;
 884
 885                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 886
 887                         gfs2_log_write_page(sdp, page);
 888                         page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 889                         mh = page_address(page);
 890                         clear_page(mh);
 891                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 892                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 893                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 894                         offset = sizeof(struct gfs2_meta_header);
 895                 }
 896
 897                 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
 898                 offset += sizeof(u64);
 899         }
 900         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 901
 902         gfs2_log_write_page(sdp, page);
 903 }
 904
 905 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 906 {
 907         struct list_head *head = &sdp->sd_log_revokes;
 908         struct gfs2_bufdata *bd;
 909         struct gfs2_glock *gl;
 910
 911         while (!list_empty(head)) {
 912                 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
 913                 list_del_init(&bd->bd_list);
 914                 gl = bd->bd_gl;
 915                 gfs2_glock_remove_revoke(gl);
 916                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 917         }
 918 }
 919
 920 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 921                                   struct gfs2_log_header_host *head, int pass)
 922 {
 923         if (pass != 0)
 924                 return;
 925
 926         jd->jd_found_revokes = 0;
 927         jd->jd_replay_tail = head->lh_tail;
 928 }
 929
 930 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
 931                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 932                                    int pass)
 933 {
 934         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 935         unsigned int blks = be32_to_cpu(ld->ld_length);
 936         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 937         struct buffer_head *bh;
 938         unsigned int offset;
 939         u64 blkno;
 940         int first = 1;
 941         int error;
 942
 943         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 944                 return 0;
 945
 946         offset = sizeof(struct gfs2_log_descriptor);
 947
 948         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 949                 error = gfs2_replay_read_block(jd, start, &bh);
 950                 if (error)
 951                         return error;
 952
 953                 if (!first)
 954                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 955
 956                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 957                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 958
 959                         error = gfs2_revoke_add(jd, blkno, start);
 960                         if (error < 0) {
 961                                 brelse(bh);
 962                                 return error;
 963                         }
 964                         else if (error)
 965                                 jd->jd_found_revokes++;
 966
 967                         if (!--revokes)
 968                                 break;
 969                         offset += sizeof(u64);
 970                 }
 971
 972                 brelse(bh);
 973                 offset = sizeof(struct gfs2_meta_header);
 974                 first = 0;
 975         }
 976
 977         return 0;
 978 }
 979
 980 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 981 {
 982         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 983
 984         if (error) {
 985                 gfs2_revoke_clean(jd);
 986                 return;
 987         }
 988         if (pass != 1)
 989                 return;
 990
 991         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 992                 jd->jd_jid, jd->jd_found_revokes);
 993
 994         gfs2_revoke_clean(jd);
 995 }
 996
 997 /**
 998  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 999  *
1000  */
1001
1002 static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1003 {
1004         unsigned int limit = databuf_limit(sdp);
1005         unsigned int nbuf;
1006         if (tr == NULL)
1007                 return;
1008         nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
1009         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1);
1010 }
1011
1012 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
1013                                     struct gfs2_log_descriptor *ld,
1014                                     __be64 *ptr, int pass)
1015 {
1016         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1017         struct gfs2_glock *gl = ip->i_gl;
1018         unsigned int blks = be32_to_cpu(ld->ld_data1);
1019         struct buffer_head *bh_log, *bh_ip;
1020         u64 blkno;
1021         u64 esc;
1022         int error = 0;
1023
1024         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
1025                 return 0;
1026
1027         gfs2_replay_incr_blk(jd, &start);
1028         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
1029                 blkno = be64_to_cpu(*ptr++);
1030                 esc = be64_to_cpu(*ptr++);
1031
1032                 jd->jd_found_blocks++;
1033
1034                 if (gfs2_revoke_check(jd, blkno, start))
1035                         continue;
1036
1037                 error = gfs2_replay_read_block(jd, start, &bh_log);
1038                 if (error)
1039                         return error;
1040
1041                 bh_ip = gfs2_meta_new(gl, blkno);
1042                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
1043
1044                 /* Unescape */
1045                 if (esc) {
1046                         __be32 *eptr = (__be32 *)bh_ip->b_data;
1047                         *eptr = cpu_to_be32(GFS2_MAGIC);
1048                 }
1049                 mark_buffer_dirty(bh_ip);
1050
1051                 brelse(bh_log);
1052                 brelse(bh_ip);
1053
1054                 jd->jd_replayed_blocks++;
1055         }
1056
1057         return error;
1058 }
1059
1060 /* FIXME: sort out accounting for log blocks etc. */
1061
1062 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
1063 {
1064         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1065         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
1066
1067         if (error) {
1068                 gfs2_meta_sync(ip->i_gl);
1069                 return;
1070         }
1071         if (pass != 1)
1072                 return;
1073
1074         /* data sync? */
1075         gfs2_meta_sync(ip->i_gl);
1076
1077         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
1078                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
1079 }
1080
1081 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1082 {
1083         struct list_head *head;
1084         struct gfs2_bufdata *bd;
1085
1086         if (tr == NULL)
1087                 return;
1088
1089         head = &tr->tr_databuf;
1090         while (!list_empty(head)) {
1091                 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
1092                 list_del_init(&bd->bd_list);
1093                 gfs2_unpin(sdp, bd->bd_bh, tr);
1094         }
1095 }
1096
1097
1098 static const struct gfs2_log_operations gfs2_buf_lops = {
1099         .lo_before_commit = buf_lo_before_commit,
1100         .lo_after_commit = buf_lo_after_commit,
1101         .lo_before_scan = buf_lo_before_scan,
1102         .lo_scan_elements = buf_lo_scan_elements,
1103         .lo_after_scan = buf_lo_after_scan,
1104         .lo_name = "buf",
1105 };
1106
1107 static const struct gfs2_log_operations gfs2_revoke_lops = {
1108         .lo_before_commit = revoke_lo_before_commit,
1109         .lo_after_commit = revoke_lo_after_commit,
1110         .lo_before_scan = revoke_lo_before_scan,
1111         .lo_scan_elements = revoke_lo_scan_elements,
1112         .lo_after_scan = revoke_lo_after_scan,
1113         .lo_name = "revoke",
1114 };
1115
1116 static const struct gfs2_log_operations gfs2_databuf_lops = {
1117         .lo_before_commit = databuf_lo_before_commit,
1118         .lo_after_commit = databuf_lo_after_commit,
1119         .lo_scan_elements = databuf_lo_scan_elements,
1120         .lo_after_scan = databuf_lo_after_scan,
1121         .lo_name = "databuf",
1122 };
1123
1124 const struct gfs2_log_operations *gfs2_log_ops[] = {
1125         &gfs2_databuf_lops,
1126         &gfs2_buf_lops,
1127         &gfs2_revoke_lops,
1128         NULL,
1129 };
1130