fs/gfs2/lops.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   4  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
   5  */
   6
   7 #include <linux/sched.h>
   8 #include <linux/slab.h>
   9 #include <linux/spinlock.h>
  10 #include <linux/completion.h>
  11 #include <linux/buffer_head.h>
  12 #include <linux/mempool.h>
  13 #include <linux/gfs2_ondisk.h>
  14 #include <linux/bio.h>
  15 #include <linux/fs.h>
  16 #include <linux/list_sort.h>
  17 #include <linux/blkdev.h>
  18
  19 #include "bmap.h"
  20 #include "dir.h"
  21 #include "gfs2.h"
  22 #include "incore.h"
  23 #include "inode.h"
  24 #include "glock.h"
  25 #include "glops.h"
  26 #include "log.h"
  27 #include "lops.h"
  28 #include "meta_io.h"
  29 #include "recovery.h"
  30 #include "rgrp.h"
  31 #include "trans.h"
  32 #include "util.h"
  33 #include "trace_gfs2.h"
  34
  35 /**
  36  * gfs2_pin - Pin a buffer in memory
  37  * @sdp: The superblock
  38  * @bh: The buffer to be pinned
  39  *
  40  * The log lock must be held when calling this function
  41  */
  42 void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
  43 {
  44         struct gfs2_bufdata *bd;
  45
  46         BUG_ON(!current->journal_info);
  47
  48         clear_buffer_dirty(bh);
  49         if (test_set_buffer_pinned(bh))
  50                 gfs2_assert_withdraw(sdp, 0);
  51         if (!buffer_uptodate(bh))
  52                 gfs2_io_error_bh_wd(sdp, bh);
  53         bd = bh->b_private;
  54         /* If this buffer is in the AIL and it has already been written
  55          * to in-place disk block, remove it from the AIL.
  56          */
  57         spin_lock(&sdp->sd_ail_lock);
  58         if (bd->bd_tr)
  59                 list_move(&bd->bd_ail_st_list, &bd->bd_tr->tr_ail2_list);
  60         spin_unlock(&sdp->sd_ail_lock);
  61         get_bh(bh);
  62         atomic_inc(&sdp->sd_log_pinned);
  63         trace_gfs2_pin(bd, 1);
  64 }
  65
  66 static bool buffer_is_rgrp(const struct gfs2_bufdata *bd)
  67 {
  68         return bd->bd_gl->gl_name.ln_type == LM_TYPE_RGRP;
  69 }
  70
  71 static void maybe_release_space(struct gfs2_bufdata *bd)
  72 {
  73         struct gfs2_glock *gl = bd->bd_gl;
  74         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
  75         struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
  76         unsigned int index = bd->bd_bh->b_blocknr - gl->gl_name.ln_number;
  77         struct gfs2_bitmap *bi = rgd->rd_bits + index;
  78
  79         rgrp_lock_local(rgd);
  80         if (bi->bi_clone == NULL)
  81                 goto out;
  82         if (sdp->sd_args.ar_discard)
  83                 gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi, 1, NULL);
  84         memcpy(bi->bi_clone + bi->bi_offset,
  85                bd->bd_bh->b_data + bi->bi_offset, bi->bi_bytes);
  86         clear_bit(GBF_FULL, &bi->bi_flags);
  87         rgd->rd_free_clone = rgd->rd_free;
  88         BUG_ON(rgd->rd_free_clone < rgd->rd_reserved);
  89         rgd->rd_extfail_pt = rgd->rd_free;
  90
  91 out:
  92         rgrp_unlock_local(rgd);
  93 }
  94
  95 /**
  96  * gfs2_unpin - Unpin a buffer
  97  * @sdp: the filesystem the buffer belongs to
  98  * @bh: The buffer to unpin
  99  * @ai:
 100  * @flags: The inode dirty flags
 101  *
 102  */
 103
 104 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
 105                        struct gfs2_trans *tr)
 106 {
 107         struct gfs2_bufdata *bd = bh->b_private;
 108
 109         BUG_ON(!buffer_uptodate(bh));
 110         BUG_ON(!buffer_pinned(bh));
 111
 112         lock_buffer(bh);
 113         mark_buffer_dirty(bh);
 114         clear_buffer_pinned(bh);
 115
 116         if (buffer_is_rgrp(bd))
 117                 maybe_release_space(bd);
 118
 119         spin_lock(&sdp->sd_ail_lock);
 120         if (bd->bd_tr) {
 121                 list_del(&bd->bd_ail_st_list);
 122                 brelse(bh);
 123         } else {
 124                 struct gfs2_glock *gl = bd->bd_gl;
 125                 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
 126                 atomic_inc(&gl->gl_ail_count);
 127         }
 128         bd->bd_tr = tr;
 129         list_add(&bd->bd_ail_st_list, &tr->tr_ail1_list);
 130         spin_unlock(&sdp->sd_ail_lock);
 131
 132         clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 133         trace_gfs2_pin(bd, 0);
 134         unlock_buffer(bh);
 135         atomic_dec(&sdp->sd_log_pinned);
 136 }
 137
 138 void gfs2_log_incr_head(struct gfs2_sbd *sdp)
 139 {
 140         BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) &&
 141                (sdp->sd_log_flush_head != sdp->sd_log_head));
 142
 143         if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks)
 144                 sdp->sd_log_flush_head = 0;
 145 }
 146
 147 u64 gfs2_log_bmap(struct gfs2_jdesc *jd, unsigned int lblock)
 148 {
 149         struct gfs2_journal_extent *je;
 150
 151         list_for_each_entry(je, &jd->extent_list, list) {
 152                 if (lblock >= je->lblock && lblock < je->lblock + je->blocks)
 153                         return je->dblock + lblock - je->lblock;
 154         }
 155
 156         return -1;
 157 }
 158
 159 /**
 160  * gfs2_end_log_write_bh - end log write of pagecache data with buffers
 161  * @sdp: The superblock
 162  * @bvec: The bio_vec
 163  * @error: The i/o status
 164  *
 165  * This finds the relevant buffers and unlocks them and sets the
 166  * error flag according to the status of the i/o request. This is
 167  * used when the log is writing data which has an in-place version
 168  * that is pinned in the pagecache.
 169  */
 170
 171 static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp,
 172                                   struct bio_vec *bvec,
 173                                   blk_status_t error)
 174 {
 175         struct buffer_head *bh, *next;
 176         struct page *page = bvec->bv_page;
 177         unsigned size;
 178
 179         bh = page_buffers(page);
 180         size = bvec->bv_len;
 181         while (bh_offset(bh) < bvec->bv_offset)
 182                 bh = bh->b_this_page;
 183         do {
 184                 if (error)
 185                         mark_buffer_write_io_error(bh);
 186                 unlock_buffer(bh);
 187                 next = bh->b_this_page;
 188                 size -= bh->b_size;
 189                 brelse(bh);
 190                 bh = next;
 191         } while(bh && size);
 192 }
 193
 194 /**
 195  * gfs2_end_log_write - end of i/o to the log
 196  * @bio: The bio
 197  *
 198  * Each bio_vec contains either data from the pagecache or data
 199  * relating to the log itself. Here we iterate over the bio_vec
 200  * array, processing both kinds of data.
 201  *
 202  */
 203
 204 static void gfs2_end_log_write(struct bio *bio)
 205 {
 206         struct gfs2_sbd *sdp = bio->bi_private;
 207         struct bio_vec *bvec;
 208         struct page *page;
 209         struct bvec_iter_all iter_all;
 210
 211         if (bio->bi_status) {
 212                 if (!cmpxchg(&sdp->sd_log_error, 0, (int)bio->bi_status))
 213                         fs_err(sdp, "Error %d writing to journal, jid=%u\n",
 214                                bio->bi_status, sdp->sd_jdesc->jd_jid);
 215                 gfs2_withdraw_delayed(sdp);
 216                 /* prevent more writes to the journal */
 217                 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 218                 wake_up(&sdp->sd_logd_waitq);
 219         }
 220
 221         bio_for_each_segment_all(bvec, bio, iter_all) {
 222                 page = bvec->bv_page;
 223                 if (page_has_buffers(page))
 224                         gfs2_end_log_write_bh(sdp, bvec, bio->bi_status);
 225                 else
 226                         mempool_free(page, gfs2_page_pool);
 227         }
 228
 229         bio_put(bio);
 230         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 231                 wake_up(&sdp->sd_log_flush_wait);
 232 }
 233
 234 /**
 235  * gfs2_log_submit_bio - Submit any pending log bio
 236  * @biop: Address of the bio pointer
 237  * @opf: REQ_OP | op_flags
 238  *
 239  * Submit any pending part-built or full bio to the block device. If
 240  * there is no pending bio, then this is a no-op.
 241  */
 242
 243 void gfs2_log_submit_bio(struct bio **biop, int opf)
 244 {
 245         struct bio *bio = *biop;
 246         if (bio) {
 247                 struct gfs2_sbd *sdp = bio->bi_private;
 248                 atomic_inc(&sdp->sd_log_in_flight);
 249                 bio->bi_opf = opf;
 250                 submit_bio(bio);
 251                 *biop = NULL;
 252         }
 253 }
 254
 255 /**
 256  * gfs2_log_alloc_bio - Allocate a bio
 257  * @sdp: The super block
 258  * @blkno: The device block number we want to write to
 259  * @end_io: The bi_end_io callback
 260  *
 261  * Allocate a new bio, initialize it with the given parameters and return it.
 262  *
 263  * Returns: The newly allocated bio
 264  */
 265
 266 static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno,
 267                                       bio_end_io_t *end_io)
 268 {
 269         struct super_block *sb = sdp->sd_vfs;
 270         struct bio *bio = bio_alloc(GFP_NOIO, BIO_MAX_VECS);
 271
 272         bio->bi_iter.bi_sector = blkno << sdp->sd_fsb2bb_shift;
 273         bio_set_dev(bio, sb->s_bdev);
 274         bio->bi_end_io = end_io;
 275         bio->bi_private = sdp;
 276
 277         return bio;
 278 }
 279
 280 /**
 281  * gfs2_log_get_bio - Get cached log bio, or allocate a new one
 282  * @sdp: The super block
 283  * @blkno: The device block number we want to write to
 284  * @bio: The bio to get or allocate
 285  * @op: REQ_OP
 286  * @end_io: The bi_end_io callback
 287  * @flush: Always flush the current bio and allocate a new one?
 288  *
 289  * If there is a cached bio, then if the next block number is sequential
 290  * with the previous one, return it, otherwise flush the bio to the
 291  * device. If there is no cached bio, or we just flushed it, then
 292  * allocate a new one.
 293  *
 294  * Returns: The bio to use for log writes
 295  */
 296
 297 static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno,
 298                                     struct bio **biop, int op,
 299                                     bio_end_io_t *end_io, bool flush)
 300 {
 301         struct bio *bio = *biop;
 302
 303         if (bio) {
 304                 u64 nblk;
 305
 306                 nblk = bio_end_sector(bio);
 307                 nblk >>= sdp->sd_fsb2bb_shift;
 308                 if (blkno == nblk && !flush)
 309                         return bio;
 310                 gfs2_log_submit_bio(biop, op);
 311         }
 312
 313         *biop = gfs2_log_alloc_bio(sdp, blkno, end_io);
 314         return *biop;
 315 }
 316
 317 /**
 318  * gfs2_log_write - write to log
 319  * @sdp: the filesystem
 320  * @page: the page to write
 321  * @size: the size of the data to write
 322  * @offset: the offset within the page
 323  * @blkno: block number of the log entry
 324  *
 325  * Try and add the page segment to the current bio. If that fails,
 326  * submit the current bio to the device and create a new one, and
 327  * then add the page segment to that.
 328  */
 329
 330 void gfs2_log_write(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
 331                     struct page *page, unsigned size, unsigned offset,
 332                     u64 blkno)
 333 {
 334         struct bio *bio;
 335         int ret;
 336
 337         bio = gfs2_log_get_bio(sdp, blkno, &jd->jd_log_bio, REQ_OP_WRITE,
 338                                gfs2_end_log_write, false);
 339         ret = bio_add_page(bio, page, size, offset);
 340         if (ret == 0) {
 341                 bio = gfs2_log_get_bio(sdp, blkno, &jd->jd_log_bio,
 342                                        REQ_OP_WRITE, gfs2_end_log_write, true);
 343                 ret = bio_add_page(bio, page, size, offset);
 344                 WARN_ON(ret == 0);
 345         }
 346 }
 347
 348 /**
 349  * gfs2_log_write_bh - write a buffer's content to the log
 350  * @sdp: The super block
 351  * @bh: The buffer pointing to the in-place location
 352  *
 353  * This writes the content of the buffer to the next available location
 354  * in the log. The buffer will be unlocked once the i/o to the log has
 355  * completed.
 356  */
 357
 358 static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh)
 359 {
 360         u64 dblock;
 361
 362         dblock = gfs2_log_bmap(sdp->sd_jdesc, sdp->sd_log_flush_head);
 363         gfs2_log_incr_head(sdp);
 364         gfs2_log_write(sdp, sdp->sd_jdesc, bh->b_page, bh->b_size,
 365                        bh_offset(bh), dblock);
 366 }
 367
 368 /**
 369  * gfs2_log_write_page - write one block stored in a page, into the log
 370  * @sdp: The superblock
 371  * @page: The struct page
 372  *
 373  * This writes the first block-sized part of the page into the log. Note
 374  * that the page must have been allocated from the gfs2_page_pool mempool
 375  * and that after this has been called, ownership has been transferred and
 376  * the page may be freed at any time.
 377  */
 378
 379 static void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
 380 {
 381         struct super_block *sb = sdp->sd_vfs;
 382         u64 dblock;
 383
 384         dblock = gfs2_log_bmap(sdp->sd_jdesc, sdp->sd_log_flush_head);
 385         gfs2_log_incr_head(sdp);
 386         gfs2_log_write(sdp, sdp->sd_jdesc, page, sb->s_blocksize, 0, dblock);
 387 }
 388
 389 /**
 390  * gfs2_end_log_read - end I/O callback for reads from the log
 391  * @bio: The bio
 392  *
 393  * Simply unlock the pages in the bio. The main thread will wait on them and
 394  * process them in order as necessary.
 395  */
 396
 397 static void gfs2_end_log_read(struct bio *bio)
 398 {
 399         struct page *page;
 400         struct bio_vec *bvec;
 401         struct bvec_iter_all iter_all;
 402
 403         bio_for_each_segment_all(bvec, bio, iter_all) {
 404                 page = bvec->bv_page;
 405                 if (bio->bi_status) {
 406                         int err = blk_status_to_errno(bio->bi_status);
 407
 408                         SetPageError(page);
 409                         mapping_set_error(page->mapping, err);
 410                 }
 411                 unlock_page(page);
 412         }
 413
 414         bio_put(bio);
 415 }
 416
 417 /**
 418  * gfs2_jhead_pg_srch - Look for the journal head in a given page.
 419  * @jd: The journal descriptor
 420  * @page: The page to look in
 421  *
 422  * Returns: 1 if found, 0 otherwise.
 423  */
 424
 425 static bool gfs2_jhead_pg_srch(struct gfs2_jdesc *jd,
 426                               struct gfs2_log_header_host *head,
 427                               struct page *page)
 428 {
 429         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 430         struct gfs2_log_header_host lh;
 431         void *kaddr = kmap_atomic(page);
 432         unsigned int offset;
 433         bool ret = false;
 434
 435         for (offset = 0; offset < PAGE_SIZE; offset += sdp->sd_sb.sb_bsize) {
 436                 if (!__get_log_header(sdp, kaddr + offset, 0, &lh)) {
 437                         if (lh.lh_sequence >= head->lh_sequence)
 438                                 *head = lh;
 439                         else {
 440                                 ret = true;
 441                                 break;
 442                         }
 443                 }
 444         }
 445         kunmap_atomic(kaddr);
 446         return ret;
 447 }
 448
 449 /**
 450  * gfs2_jhead_process_page - Search/cleanup a page
 451  * @jd: The journal descriptor
 452  * @index: Index of the page to look into
 453  * @done: If set, perform only cleanup, else search and set if found.
 454  *
 455  * Find the page with 'index' in the journal's mapping. Search the page for
 456  * the journal head if requested (cleanup == false). Release refs on the
 457  * page so the page cache can reclaim it (put_page() twice). We grabbed a
 458  * reference on this page two times, first when we did a find_or_create_page()
 459  * to obtain the page to add it to the bio and second when we do a
 460  * find_get_page() here to get the page to wait on while I/O on it is being
 461  * completed.
 462  * This function is also used to free up a page we might've grabbed but not
 463  * used. Maybe we added it to a bio, but not submitted it for I/O. Or we
 464  * submitted the I/O, but we already found the jhead so we only need to drop
 465  * our references to the page.
 466  */
 467
 468 static void gfs2_jhead_process_page(struct gfs2_jdesc *jd, unsigned long index,
 469                                     struct gfs2_log_header_host *head,
 470                                     bool *done)
 471 {
 472         struct page *page;
 473
 474         page = find_get_page(jd->jd_inode->i_mapping, index);
 475         wait_on_page_locked(page);
 476
 477         if (PageError(page))
 478                 *done = true;
 479
 480         if (!*done)
 481                 *done = gfs2_jhead_pg_srch(jd, head, page);
 482
 483         put_page(page); /* Once for find_get_page */
 484         put_page(page); /* Once more for find_or_create_page */
 485 }
 486
 487 static struct bio *gfs2_chain_bio(struct bio *prev, unsigned int nr_iovecs)
 488 {
 489         struct bio *new;
 490
 491         new = bio_alloc(GFP_NOIO, nr_iovecs);
 492         bio_copy_dev(new, prev);
 493         new->bi_iter.bi_sector = bio_end_sector(prev);
 494         new->bi_opf = prev->bi_opf;
 495         new->bi_write_hint = prev->bi_write_hint;
 496         bio_chain(new, prev);
 497         submit_bio(prev);
 498         return new;
 499 }
 500
 501 /**
 502  * gfs2_find_jhead - find the head of a log
 503  * @jd: The journal descriptor
 504  * @head: The log descriptor for the head of the log is returned here
 505  *
 506  * Do a search of a journal by reading it in large chunks using bios and find
 507  * the valid log entry with the highest sequence number.  (i.e. the log head)
 508  *
 509  * Returns: 0 on success, errno otherwise
 510  */
 511 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head,
 512                     bool keep_cache)
 513 {
 514         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 515         struct address_space *mapping = jd->jd_inode->i_mapping;
 516         unsigned int block = 0, blocks_submitted = 0, blocks_read = 0;
 517         unsigned int bsize = sdp->sd_sb.sb_bsize, off;
 518         unsigned int bsize_shift = sdp->sd_sb.sb_bsize_shift;
 519         unsigned int shift = PAGE_SHIFT - bsize_shift;
 520         unsigned int max_blocks = 2 * 1024 * 1024 >> bsize_shift;
 521         struct gfs2_journal_extent *je;
 522         int sz, ret = 0;
 523         struct bio *bio = NULL;
 524         struct page *page = NULL;
 525         bool done = false;
 526         errseq_t since;
 527
 528         memset(head, 0, sizeof(*head));
 529         if (list_empty(&jd->extent_list))
 530                 gfs2_map_journal_extents(sdp, jd);
 531
 532         since = filemap_sample_wb_err(mapping);
 533         list_for_each_entry(je, &jd->extent_list, list) {
 534                 u64 dblock = je->dblock;
 535
 536                 for (; block < je->lblock + je->blocks; block++, dblock++) {
 537                         if (!page) {
 538                                 page = find_or_create_page(mapping,
 539                                                 block >> shift, GFP_NOFS);
 540                                 if (!page) {
 541                                         ret = -ENOMEM;
 542                                         done = true;
 543                                         goto out;
 544                                 }
 545                                 off = 0;
 546                         }
 547
 548                         if (bio && (off || block < blocks_submitted + max_blocks)) {
 549                                 sector_t sector = dblock << sdp->sd_fsb2bb_shift;
 550
 551                                 if (bio_end_sector(bio) == sector) {
 552                                         sz = bio_add_page(bio, page, bsize, off);
 553                                         if (sz == bsize)
 554                                                 goto block_added;
 555                                 }
 556                                 if (off) {
 557                                         unsigned int blocks =
 558                                                 (PAGE_SIZE - off) >> bsize_shift;
 559
 560                                         bio = gfs2_chain_bio(bio, blocks);
 561                                         goto add_block_to_new_bio;
 562                                 }
 563                         }
 564
 565                         if (bio) {
 566                                 blocks_submitted = block;
 567                                 submit_bio(bio);
 568                         }
 569
 570                         bio = gfs2_log_alloc_bio(sdp, dblock, gfs2_end_log_read);
 571                         bio->bi_opf = REQ_OP_READ;
 572 add_block_to_new_bio:
 573                         sz = bio_add_page(bio, page, bsize, off);
 574                         BUG_ON(sz != bsize);
 575 block_added:
 576                         off += bsize;
 577                         if (off == PAGE_SIZE)
 578                                 page = NULL;
 579                         if (blocks_submitted <= blocks_read + max_blocks) {
 580                                 /* Keep at least one bio in flight */
 581                                 continue;
 582                         }
 583
 584                         gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done);
 585                         blocks_read += PAGE_SIZE >> bsize_shift;
 586                         if (done)
 587                                 goto out;  /* found */
 588                 }
 589         }
 590
 591 out:
 592         if (bio)
 593                 submit_bio(bio);
 594         while (blocks_read < block) {
 595                 gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done);
 596                 blocks_read += PAGE_SIZE >> bsize_shift;
 597         }
 598
 599         if (!ret)
 600                 ret = filemap_check_wb_err(mapping, since);
 601
 602         if (!keep_cache)
 603                 truncate_inode_pages(mapping, 0);
 604
 605         return ret;
 606 }
 607
 608 static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
 609                                       u32 ld_length, u32 ld_data1)
 610 {
 611         struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 612         struct gfs2_log_descriptor *ld = page_address(page);
 613         clear_page(ld);
 614         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 615         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 616         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 617         ld->ld_type = cpu_to_be32(ld_type);
 618         ld->ld_length = cpu_to_be32(ld_length);
 619         ld->ld_data1 = cpu_to_be32(ld_data1);
 620         ld->ld_data2 = 0;
 621         return page;
 622 }
 623
 624 static void gfs2_check_magic(struct buffer_head *bh)
 625 {
 626         void *kaddr;
 627         __be32 *ptr;
 628
 629         clear_buffer_escaped(bh);
 630         kaddr = kmap_atomic(bh->b_page);
 631         ptr = kaddr + bh_offset(bh);
 632         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 633                 set_buffer_escaped(bh);
 634         kunmap_atomic(kaddr);
 635 }
 636
 637 static int blocknr_cmp(void *priv, struct list_head *a, struct list_head *b)
 638 {
 639         struct gfs2_bufdata *bda, *bdb;
 640
 641         bda = list_entry(a, struct gfs2_bufdata, bd_list);
 642         bdb = list_entry(b, struct gfs2_bufdata, bd_list);
 643
 644         if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
 645                 return -1;
 646         if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
 647                 return 1;
 648         return 0;
 649 }
 650
 651 static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
 652                                 unsigned int total, struct list_head *blist,
 653                                 bool is_databuf)
 654 {
 655         struct gfs2_log_descriptor *ld;
 656         struct gfs2_bufdata *bd1 = NULL, *bd2;
 657         struct page *page;
 658         unsigned int num;
 659         unsigned n;
 660         __be64 *ptr;
 661
 662         gfs2_log_lock(sdp);
 663         list_sort(NULL, blist, blocknr_cmp);
 664         bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
 665         while(total) {
 666                 num = total;
 667                 if (total > limit)
 668                         num = limit;
 669                 gfs2_log_unlock(sdp);
 670                 page = gfs2_get_log_desc(sdp,
 671                                          is_databuf ? GFS2_LOG_DESC_JDATA :
 672                                          GFS2_LOG_DESC_METADATA, num + 1, num);
 673                 ld = page_address(page);
 674                 gfs2_log_lock(sdp);
 675                 ptr = (__be64 *)(ld + 1);
 676
 677                 n = 0;
 678                 list_for_each_entry_continue(bd1, blist, bd_list) {
 679                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 680                         if (is_databuf) {
 681                                 gfs2_check_magic(bd1->bd_bh);
 682                                 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
 683                         }
 684                         if (++n >= num)
 685                                 break;
 686                 }
 687
 688                 gfs2_log_unlock(sdp);
 689                 gfs2_log_write_page(sdp, page);
 690                 gfs2_log_lock(sdp);
 691
 692                 n = 0;
 693                 list_for_each_entry_continue(bd2, blist, bd_list) {
 694                         get_bh(bd2->bd_bh);
 695                         gfs2_log_unlock(sdp);
 696                         lock_buffer(bd2->bd_bh);
 697
 698                         if (buffer_escaped(bd2->bd_bh)) {
 699                                 void *kaddr;
 700                                 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 701                                 ptr = page_address(page);
 702                                 kaddr = kmap_atomic(bd2->bd_bh->b_page);
 703                                 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh),
 704                                        bd2->bd_bh->b_size);
 705                                 kunmap_atomic(kaddr);
 706                                 *(__be32 *)ptr = 0;
 707                                 clear_buffer_escaped(bd2->bd_bh);
 708                                 unlock_buffer(bd2->bd_bh);
 709                                 brelse(bd2->bd_bh);
 710                                 gfs2_log_write_page(sdp, page);
 711                         } else {
 712                                 gfs2_log_write_bh(sdp, bd2->bd_bh);
 713                         }
 714                         gfs2_log_lock(sdp);
 715                         if (++n >= num)
 716                                 break;
 717                 }
 718
 719                 BUG_ON(total < num);
 720                 total -= num;
 721         }
 722         gfs2_log_unlock(sdp);
 723 }
 724
 725 static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 726 {
 727         unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
 728         unsigned int nbuf;
 729         if (tr == NULL)
 730                 return;
 731         nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
 732         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0);
 733 }
 734
 735 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 736 {
 737         struct list_head *head;
 738         struct gfs2_bufdata *bd;
 739
 740         if (tr == NULL)
 741                 return;
 742
 743         head = &tr->tr_buf;
 744         while (!list_empty(head)) {
 745                 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
 746                 list_del_init(&bd->bd_list);
 747                 gfs2_unpin(sdp, bd->bd_bh, tr);
 748         }
 749 }
 750
 751 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 752                                struct gfs2_log_header_host *head, int pass)
 753 {
 754         if (pass != 0)
 755                 return;
 756
 757         jd->jd_found_blocks = 0;
 758         jd->jd_replayed_blocks = 0;
 759 }
 760
 761 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
 762                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 763                                 int pass)
 764 {
 765         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 766         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 767         struct gfs2_glock *gl = ip->i_gl;
 768         unsigned int blks = be32_to_cpu(ld->ld_data1);
 769         struct buffer_head *bh_log, *bh_ip;
 770         u64 blkno;
 771         int error = 0;
 772
 773         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 774                 return 0;
 775
 776         gfs2_replay_incr_blk(jd, &start);
 777
 778         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 779                 blkno = be64_to_cpu(*ptr++);
 780
 781                 jd->jd_found_blocks++;
 782
 783                 if (gfs2_revoke_check(jd, blkno, start))
 784                         continue;
 785
 786                 error = gfs2_replay_read_block(jd, start, &bh_log);
 787                 if (error)
 788                         return error;
 789
 790                 bh_ip = gfs2_meta_new(gl, blkno);
 791                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 792
 793                 if (gfs2_meta_check(sdp, bh_ip))
 794                         error = -EIO;
 795                 else {
 796                         struct gfs2_meta_header *mh =
 797                                 (struct gfs2_meta_header *)bh_ip->b_data;
 798
 799                         if (mh->mh_type == cpu_to_be32(GFS2_METATYPE_RG)) {
 800                                 struct gfs2_rgrpd *rgd;
 801
 802                                 rgd = gfs2_blk2rgrpd(sdp, blkno, false);
 803                                 if (rgd && rgd->rd_addr == blkno &&
 804                                     rgd->rd_bits && rgd->rd_bits->bi_bh) {
 805                                         fs_info(sdp, "Replaying 0x%llx but we "
 806                                                 "already have a bh!\n",
 807                                                 (unsigned long long)blkno);
 808                                         fs_info(sdp, "busy:%d, pinned:%d\n",
 809                                                 buffer_busy(rgd->rd_bits->bi_bh) ? 1 : 0,
 810                                                 buffer_pinned(rgd->rd_bits->bi_bh));
 811                                         gfs2_dump_glock(NULL, rgd->rd_gl, true);
 812                                 }
 813                         }
 814                         mark_buffer_dirty(bh_ip);
 815                 }
 816                 brelse(bh_log);
 817                 brelse(bh_ip);
 818
 819                 if (error)
 820                         break;
 821
 822                 jd->jd_replayed_blocks++;
 823         }
 824
 825         return error;
 826 }
 827
 828 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 829 {
 830         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 831         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 832
 833         if (error) {
 834                 gfs2_inode_metasync(ip->i_gl);
 835                 return;
 836         }
 837         if (pass != 1)
 838                 return;
 839
 840         gfs2_inode_metasync(ip->i_gl);
 841
 842         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 843                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
 844 }
 845
 846 static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 847 {
 848         struct gfs2_meta_header *mh;
 849         unsigned int offset;
 850         struct list_head *head = &sdp->sd_log_revokes;
 851         struct gfs2_bufdata *bd;
 852         struct page *page;
 853         unsigned int length;
 854
 855         gfs2_flush_revokes(sdp);
 856         if (!sdp->sd_log_num_revoke)
 857                 return;
 858
 859         length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke);
 860         page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
 861         offset = sizeof(struct gfs2_log_descriptor);
 862
 863         list_for_each_entry(bd, head, bd_list) {
 864                 sdp->sd_log_num_revoke--;
 865
 866                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 867                         gfs2_log_write_page(sdp, page);
 868                         page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 869                         mh = page_address(page);
 870                         clear_page(mh);
 871                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 872                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 873                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 874                         offset = sizeof(struct gfs2_meta_header);
 875                 }
 876
 877                 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
 878                 offset += sizeof(u64);
 879         }
 880         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 881
 882         gfs2_log_write_page(sdp, page);
 883 }
 884
 885 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 886 {
 887         struct list_head *head = &sdp->sd_log_revokes;
 888         struct gfs2_bufdata *bd;
 889         struct gfs2_glock *gl;
 890
 891         while (!list_empty(head)) {
 892                 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
 893                 list_del_init(&bd->bd_list);
 894                 gl = bd->bd_gl;
 895                 gfs2_glock_remove_revoke(gl);
 896                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 897         }
 898 }
 899
 900 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 901                                   struct gfs2_log_header_host *head, int pass)
 902 {
 903         if (pass != 0)
 904                 return;
 905
 906         jd->jd_found_revokes = 0;
 907         jd->jd_replay_tail = head->lh_tail;
 908 }
 909
 910 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
 911                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 912                                    int pass)
 913 {
 914         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 915         unsigned int blks = be32_to_cpu(ld->ld_length);
 916         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 917         struct buffer_head *bh;
 918         unsigned int offset;
 919         u64 blkno;
 920         int first = 1;
 921         int error;
 922
 923         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 924                 return 0;
 925
 926         offset = sizeof(struct gfs2_log_descriptor);
 927
 928         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 929                 error = gfs2_replay_read_block(jd, start, &bh);
 930                 if (error)
 931                         return error;
 932
 933                 if (!first)
 934                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 935
 936                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 937                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 938
 939                         error = gfs2_revoke_add(jd, blkno, start);
 940                         if (error < 0) {
 941                                 brelse(bh);
 942                                 return error;
 943                         }
 944                         else if (error)
 945                                 jd->jd_found_revokes++;
 946
 947                         if (!--revokes)
 948                                 break;
 949                         offset += sizeof(u64);
 950                 }
 951
 952                 brelse(bh);
 953                 offset = sizeof(struct gfs2_meta_header);
 954                 first = 0;
 955         }
 956
 957         return 0;
 958 }
 959
 960 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 961 {
 962         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 963
 964         if (error) {
 965                 gfs2_revoke_clean(jd);
 966                 return;
 967         }
 968         if (pass != 1)
 969                 return;
 970
 971         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 972                 jd->jd_jid, jd->jd_found_revokes);
 973
 974         gfs2_revoke_clean(jd);
 975 }
 976
 977 /**
 978  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 979  *
 980  */
 981
 982 static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 983 {
 984         unsigned int limit = databuf_limit(sdp);
 985         unsigned int nbuf;
 986         if (tr == NULL)
 987                 return;
 988         nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
 989         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1);
 990 }
 991
 992 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
 993                                     struct gfs2_log_descriptor *ld,
 994                                     __be64 *ptr, int pass)
 995 {
 996         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 997         struct gfs2_glock *gl = ip->i_gl;
 998         unsigned int blks = be32_to_cpu(ld->ld_data1);
 999         struct buffer_head *bh_log, *bh_ip;
1000         u64 blkno;
1001         u64 esc;
1002         int error = 0;
1003
1004         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
1005                 return 0;
1006
1007         gfs2_replay_incr_blk(jd, &start);
1008         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
1009                 blkno = be64_to_cpu(*ptr++);
1010                 esc = be64_to_cpu(*ptr++);
1011
1012                 jd->jd_found_blocks++;
1013
1014                 if (gfs2_revoke_check(jd, blkno, start))
1015                         continue;
1016
1017                 error = gfs2_replay_read_block(jd, start, &bh_log);
1018                 if (error)
1019                         return error;
1020
1021                 bh_ip = gfs2_meta_new(gl, blkno);
1022                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
1023
1024                 /* Unescape */
1025                 if (esc) {
1026                         __be32 *eptr = (__be32 *)bh_ip->b_data;
1027                         *eptr = cpu_to_be32(GFS2_MAGIC);
1028                 }
1029                 mark_buffer_dirty(bh_ip);
1030
1031                 brelse(bh_log);
1032                 brelse(bh_ip);
1033
1034                 jd->jd_replayed_blocks++;
1035         }
1036
1037         return error;
1038 }
1039
1040 /* FIXME: sort out accounting for log blocks etc. */
1041
1042 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
1043 {
1044         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1045         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
1046
1047         if (error) {
1048                 gfs2_inode_metasync(ip->i_gl);
1049                 return;
1050         }
1051         if (pass != 1)
1052                 return;
1053
1054         /* data sync? */
1055         gfs2_inode_metasync(ip->i_gl);
1056
1057         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
1058                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
1059 }
1060
1061 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1062 {
1063         struct list_head *head;
1064         struct gfs2_bufdata *bd;
1065
1066         if (tr == NULL)
1067                 return;
1068
1069         head = &tr->tr_databuf;
1070         while (!list_empty(head)) {
1071                 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
1072                 list_del_init(&bd->bd_list);
1073                 gfs2_unpin(sdp, bd->bd_bh, tr);
1074         }
1075 }
1076
1077
1078 static const struct gfs2_log_operations gfs2_buf_lops = {
1079         .lo_before_commit = buf_lo_before_commit,
1080         .lo_after_commit = buf_lo_after_commit,
1081         .lo_before_scan = buf_lo_before_scan,
1082         .lo_scan_elements = buf_lo_scan_elements,
1083         .lo_after_scan = buf_lo_after_scan,
1084         .lo_name = "buf",
1085 };
1086
1087 static const struct gfs2_log_operations gfs2_revoke_lops = {
1088         .lo_before_commit = revoke_lo_before_commit,
1089         .lo_after_commit = revoke_lo_after_commit,
1090         .lo_before_scan = revoke_lo_before_scan,
1091         .lo_scan_elements = revoke_lo_scan_elements,
1092         .lo_after_scan = revoke_lo_after_scan,
1093         .lo_name = "revoke",
1094 };
1095
1096 static const struct gfs2_log_operations gfs2_databuf_lops = {
1097         .lo_before_commit = databuf_lo_before_commit,
1098         .lo_after_commit = databuf_lo_after_commit,
1099         .lo_scan_elements = databuf_lo_scan_elements,
1100         .lo_after_scan = databuf_lo_after_scan,
1101         .lo_name = "databuf",
1102 };
1103
1104 const struct gfs2_log_operations *gfs2_log_ops[] = {
1105         &gfs2_databuf_lops,
1106         &gfs2_buf_lops,
1107         &gfs2_revoke_lops,
1108         NULL,
1109 };
1110