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         if (bi->bi_clone == NULL)
  80                 return;
  81         if (sdp->sd_args.ar_discard)
  82                 gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi, 1, NULL);
  83         memcpy(bi->bi_clone + bi->bi_offset,
  84                bd->bd_bh->b_data + bi->bi_offset, bi->bi_bytes);
  85         clear_bit(GBF_FULL, &bi->bi_flags);
  86         rgd->rd_free_clone = rgd->rd_free;
  87         rgd->rd_extfail_pt = rgd->rd_free;
  88 }
  89
  90 /**
  91  * gfs2_unpin - Unpin a buffer
  92  * @sdp: the filesystem the buffer belongs to
  93  * @bh: The buffer to unpin
  94  * @ai:
  95  * @flags: The inode dirty flags
  96  *
  97  */
  98
  99 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
 100                        struct gfs2_trans *tr)
 101 {
 102         struct gfs2_bufdata *bd = bh->b_private;
 103
 104         BUG_ON(!buffer_uptodate(bh));
 105         BUG_ON(!buffer_pinned(bh));
 106
 107         lock_buffer(bh);
 108         mark_buffer_dirty(bh);
 109         clear_buffer_pinned(bh);
 110
 111         if (buffer_is_rgrp(bd))
 112                 maybe_release_space(bd);
 113
 114         spin_lock(&sdp->sd_ail_lock);
 115         if (bd->bd_tr) {
 116                 list_del(&bd->bd_ail_st_list);
 117                 brelse(bh);
 118         } else {
 119                 struct gfs2_glock *gl = bd->bd_gl;
 120                 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
 121                 atomic_inc(&gl->gl_ail_count);
 122         }
 123         bd->bd_tr = tr;
 124         list_add(&bd->bd_ail_st_list, &tr->tr_ail1_list);
 125         spin_unlock(&sdp->sd_ail_lock);
 126
 127         clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 128         trace_gfs2_pin(bd, 0);
 129         unlock_buffer(bh);
 130         atomic_dec(&sdp->sd_log_pinned);
 131 }
 132
 133 void gfs2_log_incr_head(struct gfs2_sbd *sdp)
 134 {
 135         BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) &&
 136                (sdp->sd_log_flush_head != sdp->sd_log_head));
 137
 138         if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks)
 139                 sdp->sd_log_flush_head = 0;
 140 }
 141
 142 u64 gfs2_log_bmap(struct gfs2_jdesc *jd, unsigned int lblock)
 143 {
 144         struct gfs2_journal_extent *je;
 145
 146         list_for_each_entry(je, &jd->extent_list, list) {
 147                 if (lblock >= je->lblock && lblock < je->lblock + je->blocks)
 148                         return je->dblock + lblock - je->lblock;
 149         }
 150
 151         return -1;
 152 }
 153
 154 /**
 155  * gfs2_end_log_write_bh - end log write of pagecache data with buffers
 156  * @sdp: The superblock
 157  * @bvec: The bio_vec
 158  * @error: The i/o status
 159  *
 160  * This finds the relevant buffers and unlocks them and sets the
 161  * error flag according to the status of the i/o request. This is
 162  * used when the log is writing data which has an in-place version
 163  * that is pinned in the pagecache.
 164  */
 165
 166 static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp,
 167                                   struct bio_vec *bvec,
 168                                   blk_status_t error)
 169 {
 170         struct buffer_head *bh, *next;
 171         struct page *page = bvec->bv_page;
 172         unsigned size;
 173
 174         bh = page_buffers(page);
 175         size = bvec->bv_len;
 176         while (bh_offset(bh) < bvec->bv_offset)
 177                 bh = bh->b_this_page;
 178         do {
 179                 if (error)
 180                         mark_buffer_write_io_error(bh);
 181                 unlock_buffer(bh);
 182                 next = bh->b_this_page;
 183                 size -= bh->b_size;
 184                 brelse(bh);
 185                 bh = next;
 186         } while(bh && size);
 187 }
 188
 189 /**
 190  * gfs2_end_log_write - end of i/o to the log
 191  * @bio: The bio
 192  *
 193  * Each bio_vec contains either data from the pagecache or data
 194  * relating to the log itself. Here we iterate over the bio_vec
 195  * array, processing both kinds of data.
 196  *
 197  */
 198
 199 static void gfs2_end_log_write(struct bio *bio)
 200 {
 201         struct gfs2_sbd *sdp = bio->bi_private;
 202         struct bio_vec *bvec;
 203         struct page *page;
 204         struct bvec_iter_all iter_all;
 205
 206         if (bio->bi_status) {
 207                 if (!cmpxchg(&sdp->sd_log_error, 0, (int)bio->bi_status))
 208                         fs_err(sdp, "Error %d writing to journal, jid=%u\n",
 209                                bio->bi_status, sdp->sd_jdesc->jd_jid);
 210                 gfs2_withdraw_delayed(sdp);
 211                 /* prevent more writes to the journal */
 212                 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 213                 wake_up(&sdp->sd_logd_waitq);
 214         }
 215
 216         bio_for_each_segment_all(bvec, bio, iter_all) {
 217                 page = bvec->bv_page;
 218                 if (page_has_buffers(page))
 219                         gfs2_end_log_write_bh(sdp, bvec, bio->bi_status);
 220                 else
 221                         mempool_free(page, gfs2_page_pool);
 222         }
 223
 224         bio_put(bio);
 225         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 226                 wake_up(&sdp->sd_log_flush_wait);
 227 }
 228
 229 /**
 230  * gfs2_log_submit_bio - Submit any pending log bio
 231  * @biop: Address of the bio pointer
 232  * @opf: REQ_OP | op_flags
 233  *
 234  * Submit any pending part-built or full bio to the block device. If
 235  * there is no pending bio, then this is a no-op.
 236  */
 237
 238 void gfs2_log_submit_bio(struct bio **biop, int opf)
 239 {
 240         struct bio *bio = *biop;
 241         if (bio) {
 242                 struct gfs2_sbd *sdp = bio->bi_private;
 243                 atomic_inc(&sdp->sd_log_in_flight);
 244                 bio->bi_opf = opf;
 245                 submit_bio(bio);
 246                 *biop = NULL;
 247         }
 248 }
 249
 250 /**
 251  * gfs2_log_alloc_bio - Allocate a bio
 252  * @sdp: The super block
 253  * @blkno: The device block number we want to write to
 254  * @end_io: The bi_end_io callback
 255  *
 256  * Allocate a new bio, initialize it with the given parameters and return it.
 257  *
 258  * Returns: The newly allocated bio
 259  */
 260
 261 static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno,
 262                                       bio_end_io_t *end_io)
 263 {
 264         struct super_block *sb = sdp->sd_vfs;
 265         struct bio *bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
 266
 267         bio->bi_iter.bi_sector = blkno << sdp->sd_fsb2bb_shift;
 268         bio_set_dev(bio, sb->s_bdev);
 269         bio->bi_end_io = end_io;
 270         bio->bi_private = sdp;
 271
 272         return bio;
 273 }
 274
 275 /**
 276  * gfs2_log_get_bio - Get cached log bio, or allocate a new one
 277  * @sdp: The super block
 278  * @blkno: The device block number we want to write to
 279  * @bio: The bio to get or allocate
 280  * @op: REQ_OP
 281  * @end_io: The bi_end_io callback
 282  * @flush: Always flush the current bio and allocate a new one?
 283  *
 284  * If there is a cached bio, then if the next block number is sequential
 285  * with the previous one, return it, otherwise flush the bio to the
 286  * device. If there is no cached bio, or we just flushed it, then
 287  * allocate a new one.
 288  *
 289  * Returns: The bio to use for log writes
 290  */
 291
 292 static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno,
 293                                     struct bio **biop, int op,
 294                                     bio_end_io_t *end_io, bool flush)
 295 {
 296         struct bio *bio = *biop;
 297
 298         if (bio) {
 299                 u64 nblk;
 300
 301                 nblk = bio_end_sector(bio);
 302                 nblk >>= sdp->sd_fsb2bb_shift;
 303                 if (blkno == nblk && !flush)
 304                         return bio;
 305                 gfs2_log_submit_bio(biop, op);
 306         }
 307
 308         *biop = gfs2_log_alloc_bio(sdp, blkno, end_io);
 309         return *biop;
 310 }
 311
 312 /**
 313  * gfs2_log_write - write to log
 314  * @sdp: the filesystem
 315  * @page: the page to write
 316  * @size: the size of the data to write
 317  * @offset: the offset within the page
 318  * @blkno: block number of the log entry
 319  *
 320  * Try and add the page segment to the current bio. If that fails,
 321  * submit the current bio to the device and create a new one, and
 322  * then add the page segment to that.
 323  */
 324
 325 void gfs2_log_write(struct gfs2_sbd *sdp, struct page *page,
 326                     unsigned size, unsigned offset, u64 blkno)
 327 {
 328         struct bio *bio;
 329         int ret;
 330
 331         bio = gfs2_log_get_bio(sdp, blkno, &sdp->sd_log_bio, REQ_OP_WRITE,
 332                                gfs2_end_log_write, false);
 333         ret = bio_add_page(bio, page, size, offset);
 334         if (ret == 0) {
 335                 bio = gfs2_log_get_bio(sdp, blkno, &sdp->sd_log_bio,
 336                                        REQ_OP_WRITE, gfs2_end_log_write, true);
 337                 ret = bio_add_page(bio, page, size, offset);
 338                 WARN_ON(ret == 0);
 339         }
 340 }
 341
 342 /**
 343  * gfs2_log_write_bh - write a buffer's content to the log
 344  * @sdp: The super block
 345  * @bh: The buffer pointing to the in-place location
 346  *
 347  * This writes the content of the buffer to the next available location
 348  * in the log. The buffer will be unlocked once the i/o to the log has
 349  * completed.
 350  */
 351
 352 static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh)
 353 {
 354         u64 dblock;
 355
 356         dblock = gfs2_log_bmap(sdp->sd_jdesc, sdp->sd_log_flush_head);
 357         gfs2_log_incr_head(sdp);
 358         gfs2_log_write(sdp, bh->b_page, bh->b_size, bh_offset(bh), dblock);
 359 }
 360
 361 /**
 362  * gfs2_log_write_page - write one block stored in a page, into the log
 363  * @sdp: The superblock
 364  * @page: The struct page
 365  *
 366  * This writes the first block-sized part of the page into the log. Note
 367  * that the page must have been allocated from the gfs2_page_pool mempool
 368  * and that after this has been called, ownership has been transferred and
 369  * the page may be freed at any time.
 370  */
 371
 372 void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
 373 {
 374         struct super_block *sb = sdp->sd_vfs;
 375         u64 dblock;
 376
 377         dblock = gfs2_log_bmap(sdp->sd_jdesc, sdp->sd_log_flush_head);
 378         gfs2_log_incr_head(sdp);
 379         gfs2_log_write(sdp, page, sb->s_blocksize, 0, dblock);
 380 }
 381
 382 /**
 383  * gfs2_end_log_read - end I/O callback for reads from the log
 384  * @bio: The bio
 385  *
 386  * Simply unlock the pages in the bio. The main thread will wait on them and
 387  * process them in order as necessary.
 388  */
 389
 390 static void gfs2_end_log_read(struct bio *bio)
 391 {
 392         struct page *page;
 393         struct bio_vec *bvec;
 394         struct bvec_iter_all iter_all;
 395
 396         bio_for_each_segment_all(bvec, bio, iter_all) {
 397                 page = bvec->bv_page;
 398                 if (bio->bi_status) {
 399                         int err = blk_status_to_errno(bio->bi_status);
 400
 401                         SetPageError(page);
 402                         mapping_set_error(page->mapping, err);
 403                 }
 404                 unlock_page(page);
 405         }
 406
 407         bio_put(bio);
 408 }
 409
 410 /**
 411  * gfs2_jhead_pg_srch - Look for the journal head in a given page.
 412  * @jd: The journal descriptor
 413  * @page: The page to look in
 414  *
 415  * Returns: 1 if found, 0 otherwise.
 416  */
 417
 418 static bool gfs2_jhead_pg_srch(struct gfs2_jdesc *jd,
 419                               struct gfs2_log_header_host *head,
 420                               struct page *page)
 421 {
 422         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 423         struct gfs2_log_header_host lh;
 424         void *kaddr = kmap_atomic(page);
 425         unsigned int offset;
 426         bool ret = false;
 427
 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_atomic(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  * @done: If set, perform only cleanup, else search and set if found.
 447  *
 448  * Find the page with 'index' in the journal's mapping. Search the page for
 449  * the journal head if requested (cleanup == false). Release refs on the
 450  * page so the page cache can reclaim it (put_page() twice). We grabbed a
 451  * reference on this page two times, first when we did a find_or_create_page()
 452  * to obtain the page to add it to the bio and second when we do a
 453  * find_get_page() here to get the page to wait on while I/O on it is being
 454  * completed.
 455  * This function is also used to free up a page we might've grabbed but not
 456  * used. Maybe we added it to a bio, but not submitted it for I/O. Or we
 457  * submitted the I/O, but we already found the jhead so we only need to drop
 458  * our references to the page.
 459  */
 460
 461 static void gfs2_jhead_process_page(struct gfs2_jdesc *jd, unsigned long index,
 462                                     struct gfs2_log_header_host *head,
 463                                     bool *done)
 464 {
 465         struct page *page;
 466
 467         page = find_get_page(jd->jd_inode->i_mapping, index);
 468         wait_on_page_locked(page);
 469
 470         if (PageError(page))
 471                 *done = true;
 472
 473         if (!*done)
 474                 *done = gfs2_jhead_pg_srch(jd, head, page);
 475
 476         put_page(page); /* Once for find_get_page */
 477         put_page(page); /* Once more for find_or_create_page */
 478 }
 479
 480 static struct bio *gfs2_chain_bio(struct bio *prev, unsigned int nr_iovecs)
 481 {
 482         struct bio *new;
 483
 484         new = bio_alloc(GFP_NOIO, nr_iovecs);
 485         bio_copy_dev(new, prev);
 486         new->bi_iter.bi_sector = bio_end_sector(prev);
 487         new->bi_opf = prev->bi_opf;
 488         new->bi_write_hint = prev->bi_write_hint;
 489         bio_chain(new, prev);
 490         submit_bio(prev);
 491         return new;
 492 }
 493
 494 /**
 495  * gfs2_find_jhead - find the head of a log
 496  * @jd: The journal descriptor
 497  * @head: The log descriptor for the head of the log is returned here
 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 = find_or_create_page(mapping,
 532                                                 block >> shift, GFP_NOFS);
 533                                 if (!page) {
 534                                         ret = -ENOMEM;
 535                                         done = true;
 536                                         goto out;
 537                                 }
 538                                 off = 0;
 539                         }
 540
 541                         if (bio && (off || block < blocks_submitted + max_blocks)) {
 542                                 sector_t sector = dblock << sdp->sd_fsb2bb_shift;
 543
 544                                 if (bio_end_sector(bio) == sector) {
 545                                         sz = bio_add_page(bio, page, bsize, off);
 546                                         if (sz == bsize)
 547                                                 goto block_added;
 548                                 }
 549                                 if (off) {
 550                                         unsigned int blocks =
 551                                                 (PAGE_SIZE - off) >> bsize_shift;
 552
 553                                         bio = gfs2_chain_bio(bio, blocks);
 554                                         goto add_block_to_new_bio;
 555                                 }
 556                         }
 557
 558                         if (bio) {
 559                                 blocks_submitted = block;
 560                                 submit_bio(bio);
 561                         }
 562
 563                         bio = gfs2_log_alloc_bio(sdp, dblock, gfs2_end_log_read);
 564                         bio->bi_opf = REQ_OP_READ;
 565 add_block_to_new_bio:
 566                         sz = bio_add_page(bio, page, bsize, off);
 567                         BUG_ON(sz != bsize);
 568 block_added:
 569                         off += bsize;
 570                         if (off == PAGE_SIZE)
 571                                 page = NULL;
 572                         if (blocks_submitted <= blocks_read + max_blocks) {
 573                                 /* Keep at least one bio in flight */
 574                                 continue;
 575                         }
 576
 577                         gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done);
 578                         blocks_read += PAGE_SIZE >> bsize_shift;
 579                         if (done)
 580                                 goto out;  /* found */
 581                 }
 582         }
 583
 584 out:
 585         if (bio)
 586                 submit_bio(bio);
 587         while (blocks_read < block) {
 588                 gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done);
 589                 blocks_read += PAGE_SIZE >> bsize_shift;
 590         }
 591
 592         if (!ret)
 593                 ret = filemap_check_wb_err(mapping, since);
 594
 595         if (!keep_cache)
 596                 truncate_inode_pages(mapping, 0);
 597
 598         return ret;
 599 }
 600
 601 static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
 602                                       u32 ld_length, u32 ld_data1)
 603 {
 604         struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 605         struct gfs2_log_descriptor *ld = page_address(page);
 606         clear_page(ld);
 607         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 608         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 609         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 610         ld->ld_type = cpu_to_be32(ld_type);
 611         ld->ld_length = cpu_to_be32(ld_length);
 612         ld->ld_data1 = cpu_to_be32(ld_data1);
 613         ld->ld_data2 = 0;
 614         return page;
 615 }
 616
 617 static void gfs2_check_magic(struct buffer_head *bh)
 618 {
 619         void *kaddr;
 620         __be32 *ptr;
 621
 622         clear_buffer_escaped(bh);
 623         kaddr = kmap_atomic(bh->b_page);
 624         ptr = kaddr + bh_offset(bh);
 625         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 626                 set_buffer_escaped(bh);
 627         kunmap_atomic(kaddr);
 628 }
 629
 630 static int blocknr_cmp(void *priv, const struct list_head *a,
 631                        const struct list_head *b)
 632 {
 633         struct gfs2_bufdata *bda, *bdb;
 634
 635         bda = list_entry(a, struct gfs2_bufdata, bd_list);
 636         bdb = list_entry(b, struct gfs2_bufdata, bd_list);
 637
 638         if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
 639                 return -1;
 640         if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
 641                 return 1;
 642         return 0;
 643 }
 644
 645 static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
 646                                 unsigned int total, struct list_head *blist,
 647                                 bool is_databuf)
 648 {
 649         struct gfs2_log_descriptor *ld;
 650         struct gfs2_bufdata *bd1 = NULL, *bd2;
 651         struct page *page;
 652         unsigned int num;
 653         unsigned n;
 654         __be64 *ptr;
 655
 656         gfs2_log_lock(sdp);
 657         list_sort(NULL, blist, blocknr_cmp);
 658         bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
 659         while(total) {
 660                 num = total;
 661                 if (total > limit)
 662                         num = limit;
 663                 gfs2_log_unlock(sdp);
 664                 page = gfs2_get_log_desc(sdp,
 665                                          is_databuf ? GFS2_LOG_DESC_JDATA :
 666                                          GFS2_LOG_DESC_METADATA, num + 1, num);
 667                 ld = page_address(page);
 668                 gfs2_log_lock(sdp);
 669                 ptr = (__be64 *)(ld + 1);
 670
 671                 n = 0;
 672                 list_for_each_entry_continue(bd1, blist, bd_list) {
 673                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 674                         if (is_databuf) {
 675                                 gfs2_check_magic(bd1->bd_bh);
 676                                 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
 677                         }
 678                         if (++n >= num)
 679                                 break;
 680                 }
 681
 682                 gfs2_log_unlock(sdp);
 683                 gfs2_log_write_page(sdp, page);
 684                 gfs2_log_lock(sdp);
 685
 686                 n = 0;
 687                 list_for_each_entry_continue(bd2, blist, bd_list) {
 688                         get_bh(bd2->bd_bh);
 689                         gfs2_log_unlock(sdp);
 690                         lock_buffer(bd2->bd_bh);
 691
 692                         if (buffer_escaped(bd2->bd_bh)) {
 693                                 void *kaddr;
 694                                 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 695                                 ptr = page_address(page);
 696                                 kaddr = kmap_atomic(bd2->bd_bh->b_page);
 697                                 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh),
 698                                        bd2->bd_bh->b_size);
 699                                 kunmap_atomic(kaddr);
 700                                 *(__be32 *)ptr = 0;
 701                                 clear_buffer_escaped(bd2->bd_bh);
 702                                 unlock_buffer(bd2->bd_bh);
 703                                 brelse(bd2->bd_bh);
 704                                 gfs2_log_write_page(sdp, page);
 705                         } else {
 706                                 gfs2_log_write_bh(sdp, bd2->bd_bh);
 707                         }
 708                         gfs2_log_lock(sdp);
 709                         if (++n >= num)
 710                                 break;
 711                 }
 712
 713                 BUG_ON(total < num);
 714                 total -= num;
 715         }
 716         gfs2_log_unlock(sdp);
 717 }
 718
 719 static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 720 {
 721         unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
 722         unsigned int nbuf;
 723         if (tr == NULL)
 724                 return;
 725         nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
 726         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0);
 727 }
 728
 729 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 730 {
 731         struct list_head *head;
 732         struct gfs2_bufdata *bd;
 733
 734         if (tr == NULL)
 735                 return;
 736
 737         head = &tr->tr_buf;
 738         while (!list_empty(head)) {
 739                 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
 740                 list_del_init(&bd->bd_list);
 741                 gfs2_unpin(sdp, bd->bd_bh, tr);
 742         }
 743 }
 744
 745 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 746                                struct gfs2_log_header_host *head, int pass)
 747 {
 748         if (pass != 0)
 749                 return;
 750
 751         jd->jd_found_blocks = 0;
 752         jd->jd_replayed_blocks = 0;
 753 }
 754
 755 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
 756                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 757                                 int pass)
 758 {
 759         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 760         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 761         struct gfs2_glock *gl = ip->i_gl;
 762         unsigned int blks = be32_to_cpu(ld->ld_data1);
 763         struct buffer_head *bh_log, *bh_ip;
 764         u64 blkno;
 765         int error = 0;
 766
 767         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 768                 return 0;
 769
 770         gfs2_replay_incr_blk(jd, &start);
 771
 772         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 773                 blkno = be64_to_cpu(*ptr++);
 774
 775                 jd->jd_found_blocks++;
 776
 777                 if (gfs2_revoke_check(jd, blkno, start))
 778                         continue;
 779
 780                 error = gfs2_replay_read_block(jd, start, &bh_log);
 781                 if (error)
 782                         return error;
 783
 784                 bh_ip = gfs2_meta_new(gl, blkno);
 785                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 786
 787                 if (gfs2_meta_check(sdp, bh_ip))
 788                         error = -EIO;
 789                 else {
 790                         struct gfs2_meta_header *mh =
 791                                 (struct gfs2_meta_header *)bh_ip->b_data;
 792
 793                         if (mh->mh_type == cpu_to_be32(GFS2_METATYPE_RG)) {
 794                                 struct gfs2_rgrpd *rgd;
 795
 796                                 rgd = gfs2_blk2rgrpd(sdp, blkno, false);
 797                                 if (rgd && rgd->rd_addr == blkno &&
 798                                     rgd->rd_bits && rgd->rd_bits->bi_bh) {
 799                                         fs_info(sdp, "Replaying 0x%llx but we "
 800                                                 "already have a bh!\n",
 801                                                 (unsigned long long)blkno);
 802                                         fs_info(sdp, "busy:%d, pinned:%d\n",
 803                                                 buffer_busy(rgd->rd_bits->bi_bh) ? 1 : 0,
 804                                                 buffer_pinned(rgd->rd_bits->bi_bh));
 805                                         gfs2_dump_glock(NULL, rgd->rd_gl, true);
 806                                 }
 807                         }
 808                         mark_buffer_dirty(bh_ip);
 809                 }
 810                 brelse(bh_log);
 811                 brelse(bh_ip);
 812
 813                 if (error)
 814                         break;
 815
 816                 jd->jd_replayed_blocks++;
 817         }
 818
 819         return error;
 820 }
 821
 822 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 823 {
 824         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 825         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 826
 827         if (error) {
 828                 gfs2_inode_metasync(ip->i_gl);
 829                 return;
 830         }
 831         if (pass != 1)
 832                 return;
 833
 834         gfs2_inode_metasync(ip->i_gl);
 835
 836         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 837                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
 838 }
 839
 840 static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 841 {
 842         struct gfs2_meta_header *mh;
 843         unsigned int offset;
 844         struct list_head *head = &sdp->sd_log_revokes;
 845         struct gfs2_bufdata *bd;
 846         struct page *page;
 847         unsigned int length;
 848
 849         gfs2_write_revokes(sdp);
 850         if (!sdp->sd_log_num_revoke)
 851                 return;
 852
 853         length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke);
 854         page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
 855         offset = sizeof(struct gfs2_log_descriptor);
 856
 857         list_for_each_entry(bd, head, bd_list) {
 858                 sdp->sd_log_num_revoke--;
 859
 860                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 861
 862                         gfs2_log_write_page(sdp, page);
 863                         page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 864                         mh = page_address(page);
 865                         clear_page(mh);
 866                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 867                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 868                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 869                         offset = sizeof(struct gfs2_meta_header);
 870                 }
 871
 872                 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
 873                 offset += sizeof(u64);
 874         }
 875         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 876
 877         gfs2_log_write_page(sdp, page);
 878 }
 879
 880 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 881 {
 882         struct list_head *head = &sdp->sd_log_revokes;
 883         struct gfs2_bufdata *bd;
 884         struct gfs2_glock *gl;
 885
 886         while (!list_empty(head)) {
 887                 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
 888                 list_del_init(&bd->bd_list);
 889                 gl = bd->bd_gl;
 890                 gfs2_glock_remove_revoke(gl);
 891                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 892         }
 893 }
 894
 895 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 896                                   struct gfs2_log_header_host *head, int pass)
 897 {
 898         if (pass != 0)
 899                 return;
 900
 901         jd->jd_found_revokes = 0;
 902         jd->jd_replay_tail = head->lh_tail;
 903 }
 904
 905 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
 906                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 907                                    int pass)
 908 {
 909         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 910         unsigned int blks = be32_to_cpu(ld->ld_length);
 911         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 912         struct buffer_head *bh;
 913         unsigned int offset;
 914         u64 blkno;
 915         int first = 1;
 916         int error;
 917
 918         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 919                 return 0;
 920
 921         offset = sizeof(struct gfs2_log_descriptor);
 922
 923         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 924                 error = gfs2_replay_read_block(jd, start, &bh);
 925                 if (error)
 926                         return error;
 927
 928                 if (!first)
 929                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 930
 931                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 932                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 933
 934                         error = gfs2_revoke_add(jd, blkno, start);
 935                         if (error < 0) {
 936                                 brelse(bh);
 937                                 return error;
 938                         }
 939                         else if (error)
 940                                 jd->jd_found_revokes++;
 941
 942                         if (!--revokes)
 943                                 break;
 944                         offset += sizeof(u64);
 945                 }
 946
 947                 brelse(bh);
 948                 offset = sizeof(struct gfs2_meta_header);
 949                 first = 0;
 950         }
 951
 952         return 0;
 953 }
 954
 955 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 956 {
 957         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 958
 959         if (error) {
 960                 gfs2_revoke_clean(jd);
 961                 return;
 962         }
 963         if (pass != 1)
 964                 return;
 965
 966         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 967                 jd->jd_jid, jd->jd_found_revokes);
 968
 969         gfs2_revoke_clean(jd);
 970 }
 971
 972 /**
 973  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 974  *
 975  */
 976
 977 static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 978 {
 979         unsigned int limit = databuf_limit(sdp);
 980         unsigned int nbuf;
 981         if (tr == NULL)
 982                 return;
 983         nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
 984         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1);
 985 }
 986
 987 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
 988                                     struct gfs2_log_descriptor *ld,
 989                                     __be64 *ptr, int pass)
 990 {
 991         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 992         struct gfs2_glock *gl = ip->i_gl;
 993         unsigned int blks = be32_to_cpu(ld->ld_data1);
 994         struct buffer_head *bh_log, *bh_ip;
 995         u64 blkno;
 996         u64 esc;
 997         int error = 0;
 998
 999         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
1000                 return 0;
1001
1002         gfs2_replay_incr_blk(jd, &start);
1003         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
1004                 blkno = be64_to_cpu(*ptr++);
1005                 esc = be64_to_cpu(*ptr++);
1006
1007                 jd->jd_found_blocks++;
1008
1009                 if (gfs2_revoke_check(jd, blkno, start))
1010                         continue;
1011
1012                 error = gfs2_replay_read_block(jd, start, &bh_log);
1013                 if (error)
1014                         return error;
1015
1016                 bh_ip = gfs2_meta_new(gl, blkno);
1017                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
1018
1019                 /* Unescape */
1020                 if (esc) {
1021                         __be32 *eptr = (__be32 *)bh_ip->b_data;
1022                         *eptr = cpu_to_be32(GFS2_MAGIC);
1023                 }
1024                 mark_buffer_dirty(bh_ip);
1025
1026                 brelse(bh_log);
1027                 brelse(bh_ip);
1028
1029                 jd->jd_replayed_blocks++;
1030         }
1031
1032         return error;
1033 }
1034
1035 /* FIXME: sort out accounting for log blocks etc. */
1036
1037 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
1038 {
1039         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1040         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
1041
1042         if (error) {
1043                 gfs2_inode_metasync(ip->i_gl);
1044                 return;
1045         }
1046         if (pass != 1)
1047                 return;
1048
1049         /* data sync? */
1050         gfs2_inode_metasync(ip->i_gl);
1051
1052         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
1053                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
1054 }
1055
1056 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1057 {
1058         struct list_head *head;
1059         struct gfs2_bufdata *bd;
1060
1061         if (tr == NULL)
1062                 return;
1063
1064         head = &tr->tr_databuf;
1065         while (!list_empty(head)) {
1066                 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
1067                 list_del_init(&bd->bd_list);
1068                 gfs2_unpin(sdp, bd->bd_bh, tr);
1069         }
1070 }
1071
1072
1073 static const struct gfs2_log_operations gfs2_buf_lops = {
1074         .lo_before_commit = buf_lo_before_commit,
1075         .lo_after_commit = buf_lo_after_commit,
1076         .lo_before_scan = buf_lo_before_scan,
1077         .lo_scan_elements = buf_lo_scan_elements,
1078         .lo_after_scan = buf_lo_after_scan,
1079         .lo_name = "buf",
1080 };
1081
1082 static const struct gfs2_log_operations gfs2_revoke_lops = {
1083         .lo_before_commit = revoke_lo_before_commit,
1084         .lo_after_commit = revoke_lo_after_commit,
1085         .lo_before_scan = revoke_lo_before_scan,
1086         .lo_scan_elements = revoke_lo_scan_elements,
1087         .lo_after_scan = revoke_lo_after_scan,
1088         .lo_name = "revoke",
1089 };
1090
1091 static const struct gfs2_log_operations gfs2_databuf_lops = {
1092         .lo_before_commit = databuf_lo_before_commit,
1093         .lo_after_commit = databuf_lo_after_commit,
1094         .lo_scan_elements = databuf_lo_scan_elements,
1095         .lo_after_scan = databuf_lo_after_scan,
1096         .lo_name = "databuf",
1097 };
1098
1099 const struct gfs2_log_operations *gfs2_log_ops[] = {
1100         &gfs2_databuf_lops,
1101         &gfs2_buf_lops,
1102         &gfs2_revoke_lops,
1103         NULL,
1104 };
1105