fs/gfs2/lops.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   4  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
   5  */
   6
   7 #include <linux/sched.h>
   8 #include <linux/slab.h>
   9 #include <linux/spinlock.h>
  10 #include <linux/completion.h>
  11 #include <linux/buffer_head.h>
  12 #include <linux/mempool.h>
  13 #include <linux/gfs2_ondisk.h>
  14 #include <linux/bio.h>
  15 #include <linux/fs.h>
  16 #include <linux/list_sort.h>
  17 #include <linux/blkdev.h>
  18
  19 #include "bmap.h"
  20 #include "dir.h"
  21 #include "gfs2.h"
  22 #include "incore.h"
  23 #include "inode.h"
  24 #include "glock.h"
  25 #include "glops.h"
  26 #include "log.h"
  27 #include "lops.h"
  28 #include "meta_io.h"
  29 #include "recovery.h"
  30 #include "rgrp.h"
  31 #include "trans.h"
  32 #include "util.h"
  33 #include "trace_gfs2.h"
  34
  35 /**
  36  * gfs2_pin - Pin a buffer in memory
  37  * @sdp: The superblock
  38  * @bh: The buffer to be pinned
  39  *
  40  * The log lock must be held when calling this function
  41  */
  42 void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
  43 {
  44         struct gfs2_bufdata *bd;
  45
  46         BUG_ON(!current->journal_info);
  47
  48         clear_buffer_dirty(bh);
  49         if (test_set_buffer_pinned(bh))
  50                 gfs2_assert_withdraw(sdp, 0);
  51         if (!buffer_uptodate(bh))
  52                 gfs2_io_error_bh_wd(sdp, bh);
  53         bd = bh->b_private;
  54         /* If this buffer is in the AIL and it has already been written
  55          * to in-place disk block, remove it from the AIL.
  56          */
  57         spin_lock(&sdp->sd_ail_lock);
  58         if (bd->bd_tr)
  59                 list_move(&bd->bd_ail_st_list, &bd->bd_tr->tr_ail2_list);
  60         spin_unlock(&sdp->sd_ail_lock);
  61         get_bh(bh);
  62         atomic_inc(&sdp->sd_log_pinned);
  63         trace_gfs2_pin(bd, 1);
  64 }
  65
  66 static bool buffer_is_rgrp(const struct gfs2_bufdata *bd)
  67 {
  68         return bd->bd_gl->gl_name.ln_type == LM_TYPE_RGRP;
  69 }
  70
  71 static void maybe_release_space(struct gfs2_bufdata *bd)
  72 {
  73         struct gfs2_glock *gl = bd->bd_gl;
  74         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
  75         struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
  76         unsigned int index = bd->bd_bh->b_blocknr - gl->gl_name.ln_number;
  77         struct gfs2_bitmap *bi = rgd->rd_bits + index;
  78
  79         rgrp_lock_local(rgd);
  80         if (bi->bi_clone == NULL)
  81                 goto out;
  82         if (sdp->sd_args.ar_discard)
  83                 gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi, 1, NULL);
  84         memcpy(bi->bi_clone + bi->bi_offset,
  85                bd->bd_bh->b_data + bi->bi_offset, bi->bi_bytes);
  86         clear_bit(GBF_FULL, &bi->bi_flags);
  87         rgd->rd_free_clone = rgd->rd_free;
  88         BUG_ON(rgd->rd_free_clone < rgd->rd_reserved);
  89         rgd->rd_extfail_pt = rgd->rd_free;
  90
  91 out:
  92         rgrp_unlock_local(rgd);
  93 }
  94
  95 /**
  96  * gfs2_unpin - Unpin a buffer
  97  * @sdp: the filesystem the buffer belongs to
  98  * @bh: The buffer to unpin
  99  * @tr: The system transaction being flushed
 100  */
 101
 102 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
 103                        struct gfs2_trans *tr)
 104 {
 105         struct gfs2_bufdata *bd = bh->b_private;
 106
 107         BUG_ON(!buffer_uptodate(bh));
 108         BUG_ON(!buffer_pinned(bh));
 109
 110         lock_buffer(bh);
 111         mark_buffer_dirty(bh);
 112         clear_buffer_pinned(bh);
 113
 114         if (buffer_is_rgrp(bd))
 115                 maybe_release_space(bd);
 116
 117         spin_lock(&sdp->sd_ail_lock);
 118         if (bd->bd_tr) {
 119                 list_del(&bd->bd_ail_st_list);
 120                 brelse(bh);
 121         } else {
 122                 struct gfs2_glock *gl = bd->bd_gl;
 123                 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
 124                 atomic_inc(&gl->gl_ail_count);
 125         }
 126         bd->bd_tr = tr;
 127         list_add(&bd->bd_ail_st_list, &tr->tr_ail1_list);
 128         spin_unlock(&sdp->sd_ail_lock);
 129
 130         clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 131         trace_gfs2_pin(bd, 0);
 132         unlock_buffer(bh);
 133         atomic_dec(&sdp->sd_log_pinned);
 134 }
 135
 136 void gfs2_log_incr_head(struct gfs2_sbd *sdp)
 137 {
 138         BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) &&
 139                (sdp->sd_log_flush_head != sdp->sd_log_head));
 140
 141         if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks)
 142                 sdp->sd_log_flush_head = 0;
 143 }
 144
 145 u64 gfs2_log_bmap(struct gfs2_jdesc *jd, unsigned int lblock)
 146 {
 147         struct gfs2_journal_extent *je;
 148
 149         list_for_each_entry(je, &jd->extent_list, list) {
 150                 if (lblock >= je->lblock && lblock < je->lblock + je->blocks)
 151                         return je->dblock + lblock - je->lblock;
 152         }
 153
 154         return -1;
 155 }
 156
 157 /**
 158  * gfs2_end_log_write_bh - end log write of pagecache data with buffers
 159  * @sdp: The superblock
 160  * @bvec: The bio_vec
 161  * @error: The i/o status
 162  *
 163  * This finds the relevant buffers and unlocks them and sets the
 164  * error flag according to the status of the i/o request. This is
 165  * used when the log is writing data which has an in-place version
 166  * that is pinned in the pagecache.
 167  */
 168
 169 static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp,
 170                                   struct bio_vec *bvec,
 171                                   blk_status_t error)
 172 {
 173         struct buffer_head *bh, *next;
 174         struct page *page = bvec->bv_page;
 175         unsigned size;
 176
 177         bh = page_buffers(page);
 178         size = bvec->bv_len;
 179         while (bh_offset(bh) < bvec->bv_offset)
 180                 bh = bh->b_this_page;
 181         do {
 182                 if (error)
 183                         mark_buffer_write_io_error(bh);
 184                 unlock_buffer(bh);
 185                 next = bh->b_this_page;
 186                 size -= bh->b_size;
 187                 brelse(bh);
 188                 bh = next;
 189         } while(bh && size);
 190 }
 191
 192 /**
 193  * gfs2_end_log_write - end of i/o to the log
 194  * @bio: The bio
 195  *
 196  * Each bio_vec contains either data from the pagecache or data
 197  * relating to the log itself. Here we iterate over the bio_vec
 198  * array, processing both kinds of data.
 199  *
 200  */
 201
 202 static void gfs2_end_log_write(struct bio *bio)
 203 {
 204         struct gfs2_sbd *sdp = bio->bi_private;
 205         struct bio_vec *bvec;
 206         struct page *page;
 207         struct bvec_iter_all iter_all;
 208
 209         if (bio->bi_status) {
 210                 if (!cmpxchg(&sdp->sd_log_error, 0, (int)bio->bi_status))
 211                         fs_err(sdp, "Error %d writing to journal, jid=%u\n",
 212                                bio->bi_status, sdp->sd_jdesc->jd_jid);
 213                 gfs2_withdraw_delayed(sdp);
 214                 /* prevent more writes to the journal */
 215                 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 216                 wake_up(&sdp->sd_logd_waitq);
 217         }
 218
 219         bio_for_each_segment_all(bvec, bio, iter_all) {
 220                 page = bvec->bv_page;
 221                 if (page_has_buffers(page))
 222                         gfs2_end_log_write_bh(sdp, bvec, bio->bi_status);
 223                 else
 224                         mempool_free(page, gfs2_page_pool);
 225         }
 226
 227         bio_put(bio);
 228         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 229                 wake_up(&sdp->sd_log_flush_wait);
 230 }
 231
 232 /**
 233  * gfs2_log_submit_bio - Submit any pending log bio
 234  * @biop: Address of the bio pointer
 235  * @opf: REQ_OP | op_flags
 236  *
 237  * Submit any pending part-built or full bio to the block device. If
 238  * there is no pending bio, then this is a no-op.
 239  */
 240
 241 void gfs2_log_submit_bio(struct bio **biop, blk_opf_t opf)
 242 {
 243         struct bio *bio = *biop;
 244         if (bio) {
 245                 struct gfs2_sbd *sdp = bio->bi_private;
 246                 atomic_inc(&sdp->sd_log_in_flight);
 247                 bio->bi_opf = opf;
 248                 submit_bio(bio);
 249                 *biop = NULL;
 250         }
 251 }
 252
 253 /**
 254  * gfs2_log_alloc_bio - Allocate a bio
 255  * @sdp: The super block
 256  * @blkno: The device block number we want to write to
 257  * @end_io: The bi_end_io callback
 258  *
 259  * Allocate a new bio, initialize it with the given parameters and return it.
 260  *
 261  * Returns: The newly allocated bio
 262  */
 263
 264 static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno,
 265                                       bio_end_io_t *end_io)
 266 {
 267         struct super_block *sb = sdp->sd_vfs;
 268         struct bio *bio = bio_alloc(sb->s_bdev, BIO_MAX_VECS, 0, GFP_NOIO);
 269
 270         bio->bi_iter.bi_sector = blkno << sdp->sd_fsb2bb_shift;
 271         bio->bi_end_io = end_io;
 272         bio->bi_private = sdp;
 273
 274         return bio;
 275 }
 276
 277 /**
 278  * gfs2_log_get_bio - Get cached log bio, or allocate a new one
 279  * @sdp: The super block
 280  * @blkno: The device block number we want to write to
 281  * @biop: The bio to get or allocate
 282  * @op: REQ_OP
 283  * @end_io: The bi_end_io callback
 284  * @flush: Always flush the current bio and allocate a new one?
 285  *
 286  * If there is a cached bio, then if the next block number is sequential
 287  * with the previous one, return it, otherwise flush the bio to the
 288  * device. If there is no cached bio, or we just flushed it, then
 289  * allocate a new one.
 290  *
 291  * Returns: The bio to use for log writes
 292  */
 293
 294 static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno,
 295                                     struct bio **biop, enum req_op op,
 296                                     bio_end_io_t *end_io, bool flush)
 297 {
 298         struct bio *bio = *biop;
 299
 300         if (bio) {
 301                 u64 nblk;
 302
 303                 nblk = bio_end_sector(bio);
 304                 nblk >>= sdp->sd_fsb2bb_shift;
 305                 if (blkno == nblk && !flush)
 306                         return bio;
 307                 gfs2_log_submit_bio(biop, op);
 308         }
 309
 310         *biop = gfs2_log_alloc_bio(sdp, blkno, end_io);
 311         return *biop;
 312 }
 313
 314 /**
 315  * gfs2_log_write - write to log
 316  * @sdp: the filesystem
 317  * @jd: The journal descriptor
 318  * @page: the page to write
 319  * @size: the size of the data to write
 320  * @offset: the offset within the page
 321  * @blkno: block number of the log entry
 322  *
 323  * Try and add the page segment to the current bio. If that fails,
 324  * submit the current bio to the device and create a new one, and
 325  * then add the page segment to that.
 326  */
 327
 328 void gfs2_log_write(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
 329                     struct page *page, unsigned size, unsigned offset,
 330                     u64 blkno)
 331 {
 332         struct bio *bio;
 333         int ret;
 334
 335         bio = gfs2_log_get_bio(sdp, blkno, &jd->jd_log_bio, REQ_OP_WRITE,
 336                                gfs2_end_log_write, false);
 337         ret = bio_add_page(bio, page, size, offset);
 338         if (ret == 0) {
 339                 bio = gfs2_log_get_bio(sdp, blkno, &jd->jd_log_bio,
 340                                        REQ_OP_WRITE, gfs2_end_log_write, true);
 341                 ret = bio_add_page(bio, page, size, offset);
 342                 WARN_ON(ret == 0);
 343         }
 344 }
 345
 346 /**
 347  * gfs2_log_write_bh - write a buffer's content to the log
 348  * @sdp: The super block
 349  * @bh: The buffer pointing to the in-place location
 350  *
 351  * This writes the content of the buffer to the next available location
 352  * in the log. The buffer will be unlocked once the i/o to the log has
 353  * completed.
 354  */
 355
 356 static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh)
 357 {
 358         u64 dblock;
 359
 360         dblock = gfs2_log_bmap(sdp->sd_jdesc, sdp->sd_log_flush_head);
 361         gfs2_log_incr_head(sdp);
 362         gfs2_log_write(sdp, sdp->sd_jdesc, bh->b_page, bh->b_size,
 363                        bh_offset(bh), dblock);
 364 }
 365
 366 /**
 367  * gfs2_log_write_page - write one block stored in a page, into the log
 368  * @sdp: The superblock
 369  * @page: The struct page
 370  *
 371  * This writes the first block-sized part of the page into the log. Note
 372  * that the page must have been allocated from the gfs2_page_pool mempool
 373  * and that after this has been called, ownership has been transferred and
 374  * the page may be freed at any time.
 375  */
 376
 377 static void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
 378 {
 379         struct super_block *sb = sdp->sd_vfs;
 380         u64 dblock;
 381
 382         dblock = gfs2_log_bmap(sdp->sd_jdesc, sdp->sd_log_flush_head);
 383         gfs2_log_incr_head(sdp);
 384         gfs2_log_write(sdp, sdp->sd_jdesc, page, sb->s_blocksize, 0, dblock);
 385 }
 386
 387 /**
 388  * gfs2_end_log_read - end I/O callback for reads from the log
 389  * @bio: The bio
 390  *
 391  * Simply unlock the pages in the bio. The main thread will wait on them and
 392  * process them in order as necessary.
 393  */
 394
 395 static void gfs2_end_log_read(struct bio *bio)
 396 {
 397         struct page *page;
 398         struct bio_vec *bvec;
 399         struct bvec_iter_all iter_all;
 400
 401         bio_for_each_segment_all(bvec, bio, iter_all) {
 402                 page = bvec->bv_page;
 403                 if (bio->bi_status) {
 404                         int err = blk_status_to_errno(bio->bi_status);
 405
 406                         SetPageError(page);
 407                         mapping_set_error(page->mapping, err);
 408                 }
 409                 unlock_page(page);
 410         }
 411
 412         bio_put(bio);
 413 }
 414
 415 /**
 416  * gfs2_jhead_pg_srch - Look for the journal head in a given page.
 417  * @jd: The journal descriptor
 418  * @head: The journal head to start from
 419  * @page: The page to look in
 420  *
 421  * Returns: 1 if found, 0 otherwise.
 422  */
 423
 424 static bool gfs2_jhead_pg_srch(struct gfs2_jdesc *jd,
 425                               struct gfs2_log_header_host *head,
 426                               struct page *page)
 427 {
 428         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 429         struct gfs2_log_header_host lh;
 430         void *kaddr;
 431         unsigned int offset;
 432         bool ret = false;
 433
 434         kaddr = kmap_local_page(page);
 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_local(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  * @head: The journal head to start from
 454  * @done: If set, perform only cleanup, else search and set if found.
 455  *
 456  * Find the folio with 'index' in the journal's mapping. Search the folio for
 457  * the journal head if requested (cleanup == false). Release refs on the
 458  * folio so the page cache can reclaim it. We grabbed a
 459  * reference on this folio twice, first when we did a find_or_create_page()
 460  * to obtain the folio to add it to the bio and second when we do a
 461  * filemap_get_folio() here to get the folio to wait on while I/O on it is being
 462  * completed.
 463  * This function is also used to free up a folio we might've grabbed but not
 464  * used. Maybe we added it to a bio, but not submitted it for I/O. Or we
 465  * submitted the I/O, but we already found the jhead so we only need to drop
 466  * our references to the folio.
 467  */
 468
 469 static void gfs2_jhead_process_page(struct gfs2_jdesc *jd, unsigned long index,
 470                                     struct gfs2_log_header_host *head,
 471                                     bool *done)
 472 {
 473         struct folio *folio;
 474
 475         folio = filemap_get_folio(jd->jd_inode->i_mapping, index);
 476
 477         folio_wait_locked(folio);
 478         if (folio_test_error(folio))
 479                 *done = true;
 480
 481         if (!*done)
 482                 *done = gfs2_jhead_pg_srch(jd, head, &folio->page);
 483
 484         /* filemap_get_folio() and the earlier find_or_create_page() */
 485         folio_put_refs(folio, 2);
 486 }
 487
 488 static struct bio *gfs2_chain_bio(struct bio *prev, unsigned int nr_iovecs)
 489 {
 490         struct bio *new;
 491
 492         new = bio_alloc(prev->bi_bdev, nr_iovecs, prev->bi_opf, GFP_NOIO);
 493         bio_clone_blkg_association(new, prev);
 494         new->bi_iter.bi_sector = bio_end_sector(prev);
 495         bio_chain(new, prev);
 496         submit_bio(prev);
 497         return new;
 498 }
 499
 500 /**
 501  * gfs2_find_jhead - find the head of a log
 502  * @jd: The journal descriptor
 503  * @head: The log descriptor for the head of the log is returned here
 504  * @keep_cache: If set inode pages will not be truncated
 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_local_page(bh->b_page);
 631         ptr = kaddr + bh_offset(bh);
 632         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 633                 set_buffer_escaped(bh);
 634         kunmap_local(kaddr);
 635 }
 636
 637 static int blocknr_cmp(void *priv, const struct list_head *a,
 638                        const struct list_head *b)
 639 {
 640         struct gfs2_bufdata *bda, *bdb;
 641
 642         bda = list_entry(a, struct gfs2_bufdata, bd_list);
 643         bdb = list_entry(b, struct gfs2_bufdata, bd_list);
 644
 645         if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
 646                 return -1;
 647         if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
 648                 return 1;
 649         return 0;
 650 }
 651
 652 static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
 653                                 unsigned int total, struct list_head *blist,
 654                                 bool is_databuf)
 655 {
 656         struct gfs2_log_descriptor *ld;
 657         struct gfs2_bufdata *bd1 = NULL, *bd2;
 658         struct page *page;
 659         unsigned int num;
 660         unsigned n;
 661         __be64 *ptr;
 662
 663         gfs2_log_lock(sdp);
 664         list_sort(NULL, blist, blocknr_cmp);
 665         bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
 666         while(total) {
 667                 num = total;
 668                 if (total > limit)
 669                         num = limit;
 670                 gfs2_log_unlock(sdp);
 671                 page = gfs2_get_log_desc(sdp,
 672                                          is_databuf ? GFS2_LOG_DESC_JDATA :
 673                                          GFS2_LOG_DESC_METADATA, num + 1, num);
 674                 ld = page_address(page);
 675                 gfs2_log_lock(sdp);
 676                 ptr = (__be64 *)(ld + 1);
 677
 678                 n = 0;
 679                 list_for_each_entry_continue(bd1, blist, bd_list) {
 680                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 681                         if (is_databuf) {
 682                                 gfs2_check_magic(bd1->bd_bh);
 683                                 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
 684                         }
 685                         if (++n >= num)
 686                                 break;
 687                 }
 688
 689                 gfs2_log_unlock(sdp);
 690                 gfs2_log_write_page(sdp, page);
 691                 gfs2_log_lock(sdp);
 692
 693                 n = 0;
 694                 list_for_each_entry_continue(bd2, blist, bd_list) {
 695                         get_bh(bd2->bd_bh);
 696                         gfs2_log_unlock(sdp);
 697                         lock_buffer(bd2->bd_bh);
 698
 699                         if (buffer_escaped(bd2->bd_bh)) {
 700                                 void *p;
 701
 702                                 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 703                                 p = page_address(page);
 704                                 memcpy_from_page(p, page, bh_offset(bd2->bd_bh), bd2->bd_bh->b_size);
 705                                 *(__be32 *)p = 0;
 706                                 clear_buffer_escaped(bd2->bd_bh);
 707                                 unlock_buffer(bd2->bd_bh);
 708                                 brelse(bd2->bd_bh);
 709                                 gfs2_log_write_page(sdp, page);
 710                         } else {
 711                                 gfs2_log_write_bh(sdp, bd2->bd_bh);
 712                         }
 713                         gfs2_log_lock(sdp);
 714                         if (++n >= num)
 715                                 break;
 716                 }
 717
 718                 BUG_ON(total < num);
 719                 total -= num;
 720         }
 721         gfs2_log_unlock(sdp);
 722 }
 723
 724 static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 725 {
 726         unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
 727         unsigned int nbuf;
 728         if (tr == NULL)
 729                 return;
 730         nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
 731         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0);
 732 }
 733
 734 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 735 {
 736         struct list_head *head;
 737         struct gfs2_bufdata *bd;
 738
 739         if (tr == NULL)
 740                 return;
 741
 742         head = &tr->tr_buf;
 743         while (!list_empty(head)) {
 744                 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
 745                 list_del_init(&bd->bd_list);
 746                 gfs2_unpin(sdp, bd->bd_bh, tr);
 747         }
 748 }
 749
 750 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 751                                struct gfs2_log_header_host *head, int pass)
 752 {
 753         if (pass != 0)
 754                 return;
 755
 756         jd->jd_found_blocks = 0;
 757         jd->jd_replayed_blocks = 0;
 758 }
 759
 760 #define obsolete_rgrp_replay \
 761 "Replaying 0x%llx from jid=%d/0x%llx but we already have a bh!\n"
 762 #define obsolete_rgrp_replay2 \
 763 "busy:%d, pinned:%d rg_gen:0x%llx, j_gen:0x%llx\n"
 764
 765 static void obsolete_rgrp(struct gfs2_jdesc *jd, struct buffer_head *bh_log,
 766                           u64 blkno)
 767 {
 768         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 769         struct gfs2_rgrpd *rgd;
 770         struct gfs2_rgrp *jrgd = (struct gfs2_rgrp *)bh_log->b_data;
 771
 772         rgd = gfs2_blk2rgrpd(sdp, blkno, false);
 773         if (rgd && rgd->rd_addr == blkno &&
 774             rgd->rd_bits && rgd->rd_bits->bi_bh) {
 775                 fs_info(sdp, obsolete_rgrp_replay, (unsigned long long)blkno,
 776                         jd->jd_jid, bh_log->b_blocknr);
 777                 fs_info(sdp, obsolete_rgrp_replay2,
 778                         buffer_busy(rgd->rd_bits->bi_bh) ? 1 : 0,
 779                         buffer_pinned(rgd->rd_bits->bi_bh),
 780                         rgd->rd_igeneration,
 781                         be64_to_cpu(jrgd->rg_igeneration));
 782                 gfs2_dump_glock(NULL, rgd->rd_gl, true);
 783         }
 784 }
 785
 786 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
 787                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 788                                 int pass)
 789 {
 790         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 791         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 792         struct gfs2_glock *gl = ip->i_gl;
 793         unsigned int blks = be32_to_cpu(ld->ld_data1);
 794         struct buffer_head *bh_log, *bh_ip;
 795         u64 blkno;
 796         int error = 0;
 797
 798         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 799                 return 0;
 800
 801         gfs2_replay_incr_blk(jd, &start);
 802
 803         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 804                 blkno = be64_to_cpu(*ptr++);
 805
 806                 jd->jd_found_blocks++;
 807
 808                 if (gfs2_revoke_check(jd, blkno, start))
 809                         continue;
 810
 811                 error = gfs2_replay_read_block(jd, start, &bh_log);
 812                 if (error)
 813                         return error;
 814
 815                 bh_ip = gfs2_meta_new(gl, blkno);
 816                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 817
 818                 if (gfs2_meta_check(sdp, bh_ip))
 819                         error = -EIO;
 820                 else {
 821                         struct gfs2_meta_header *mh =
 822                                 (struct gfs2_meta_header *)bh_ip->b_data;
 823
 824                         if (mh->mh_type == cpu_to_be32(GFS2_METATYPE_RG))
 825                                 obsolete_rgrp(jd, bh_log, blkno);
 826
 827                         mark_buffer_dirty(bh_ip);
 828                 }
 829                 brelse(bh_log);
 830                 brelse(bh_ip);
 831
 832                 if (error)
 833                         break;
 834
 835                 jd->jd_replayed_blocks++;
 836         }
 837
 838         return error;
 839 }
 840
 841 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 842 {
 843         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 844         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 845
 846         if (error) {
 847                 gfs2_inode_metasync(ip->i_gl);
 848                 return;
 849         }
 850         if (pass != 1)
 851                 return;
 852
 853         gfs2_inode_metasync(ip->i_gl);
 854
 855         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 856                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
 857 }
 858
 859 static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 860 {
 861         struct gfs2_meta_header *mh;
 862         unsigned int offset;
 863         struct list_head *head = &sdp->sd_log_revokes;
 864         struct gfs2_bufdata *bd;
 865         struct page *page;
 866         unsigned int length;
 867
 868         gfs2_flush_revokes(sdp);
 869         if (!sdp->sd_log_num_revoke)
 870                 return;
 871
 872         length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke);
 873         page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
 874         offset = sizeof(struct gfs2_log_descriptor);
 875
 876         list_for_each_entry(bd, head, bd_list) {
 877                 sdp->sd_log_num_revoke--;
 878
 879                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 880                         gfs2_log_write_page(sdp, page);
 881                         page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 882                         mh = page_address(page);
 883                         clear_page(mh);
 884                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 885                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 886                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 887                         offset = sizeof(struct gfs2_meta_header);
 888                 }
 889
 890                 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
 891                 offset += sizeof(u64);
 892         }
 893         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 894
 895         gfs2_log_write_page(sdp, page);
 896 }
 897
 898 void gfs2_drain_revokes(struct gfs2_sbd *sdp)
 899 {
 900         struct list_head *head = &sdp->sd_log_revokes;
 901         struct gfs2_bufdata *bd;
 902         struct gfs2_glock *gl;
 903
 904         while (!list_empty(head)) {
 905                 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
 906                 list_del_init(&bd->bd_list);
 907                 gl = bd->bd_gl;
 908                 gfs2_glock_remove_revoke(gl);
 909                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 910         }
 911 }
 912
 913 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 914 {
 915         gfs2_drain_revokes(sdp);
 916 }
 917
 918 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 919                                   struct gfs2_log_header_host *head, int pass)
 920 {
 921         if (pass != 0)
 922                 return;
 923
 924         jd->jd_found_revokes = 0;
 925         jd->jd_replay_tail = head->lh_tail;
 926 }
 927
 928 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
 929                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 930                                    int pass)
 931 {
 932         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 933         unsigned int blks = be32_to_cpu(ld->ld_length);
 934         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 935         struct buffer_head *bh;
 936         unsigned int offset;
 937         u64 blkno;
 938         int first = 1;
 939         int error;
 940
 941         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 942                 return 0;
 943
 944         offset = sizeof(struct gfs2_log_descriptor);
 945
 946         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 947                 error = gfs2_replay_read_block(jd, start, &bh);
 948                 if (error)
 949                         return error;
 950
 951                 if (!first)
 952                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 953
 954                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 955                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 956
 957                         error = gfs2_revoke_add(jd, blkno, start);
 958                         if (error < 0) {
 959                                 brelse(bh);
 960                                 return error;
 961                         }
 962                         else if (error)
 963                                 jd->jd_found_revokes++;
 964
 965                         if (!--revokes)
 966                                 break;
 967                         offset += sizeof(u64);
 968                 }
 969
 970                 brelse(bh);
 971                 offset = sizeof(struct gfs2_meta_header);
 972                 first = 0;
 973         }
 974
 975         return 0;
 976 }
 977
 978 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 979 {
 980         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 981
 982         if (error) {
 983                 gfs2_revoke_clean(jd);
 984                 return;
 985         }
 986         if (pass != 1)
 987                 return;
 988
 989         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 990                 jd->jd_jid, jd->jd_found_revokes);
 991
 992         gfs2_revoke_clean(jd);
 993 }
 994
 995 /**
 996  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 997  * @sdp: The filesystem
 998  * @tr: The system transaction being flushed
 999  */
1000
1001 static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1002 {
1003         unsigned int limit = databuf_limit(sdp);
1004         unsigned int nbuf;
1005         if (tr == NULL)
1006                 return;
1007         nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
1008         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1);
1009 }
1010
1011 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
1012                                     struct gfs2_log_descriptor *ld,
1013                                     __be64 *ptr, int pass)
1014 {
1015         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1016         struct gfs2_glock *gl = ip->i_gl;
1017         unsigned int blks = be32_to_cpu(ld->ld_data1);
1018         struct buffer_head *bh_log, *bh_ip;
1019         u64 blkno;
1020         u64 esc;
1021         int error = 0;
1022
1023         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
1024                 return 0;
1025
1026         gfs2_replay_incr_blk(jd, &start);
1027         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
1028                 blkno = be64_to_cpu(*ptr++);
1029                 esc = be64_to_cpu(*ptr++);
1030
1031                 jd->jd_found_blocks++;
1032
1033                 if (gfs2_revoke_check(jd, blkno, start))
1034                         continue;
1035
1036                 error = gfs2_replay_read_block(jd, start, &bh_log);
1037                 if (error)
1038                         return error;
1039
1040                 bh_ip = gfs2_meta_new(gl, blkno);
1041                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
1042
1043                 /* Unescape */
1044                 if (esc) {
1045                         __be32 *eptr = (__be32 *)bh_ip->b_data;
1046                         *eptr = cpu_to_be32(GFS2_MAGIC);
1047                 }
1048                 mark_buffer_dirty(bh_ip);
1049
1050                 brelse(bh_log);
1051                 brelse(bh_ip);
1052
1053                 jd->jd_replayed_blocks++;
1054         }
1055
1056         return error;
1057 }
1058
1059 /* FIXME: sort out accounting for log blocks etc. */
1060
1061 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
1062 {
1063         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1064         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
1065
1066         if (error) {
1067                 gfs2_inode_metasync(ip->i_gl);
1068                 return;
1069         }
1070         if (pass != 1)
1071                 return;
1072
1073         /* data sync? */
1074         gfs2_inode_metasync(ip->i_gl);
1075
1076         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
1077                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
1078 }
1079
1080 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1081 {
1082         struct list_head *head;
1083         struct gfs2_bufdata *bd;
1084
1085         if (tr == NULL)
1086                 return;
1087
1088         head = &tr->tr_databuf;
1089         while (!list_empty(head)) {
1090                 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
1091                 list_del_init(&bd->bd_list);
1092                 gfs2_unpin(sdp, bd->bd_bh, tr);
1093         }
1094 }
1095
1096
1097 static const struct gfs2_log_operations gfs2_buf_lops = {
1098         .lo_before_commit = buf_lo_before_commit,
1099         .lo_after_commit = buf_lo_after_commit,
1100         .lo_before_scan = buf_lo_before_scan,
1101         .lo_scan_elements = buf_lo_scan_elements,
1102         .lo_after_scan = buf_lo_after_scan,
1103         .lo_name = "buf",
1104 };
1105
1106 static const struct gfs2_log_operations gfs2_revoke_lops = {
1107         .lo_before_commit = revoke_lo_before_commit,
1108         .lo_after_commit = revoke_lo_after_commit,
1109         .lo_before_scan = revoke_lo_before_scan,
1110         .lo_scan_elements = revoke_lo_scan_elements,
1111         .lo_after_scan = revoke_lo_after_scan,
1112         .lo_name = "revoke",
1113 };
1114
1115 static const struct gfs2_log_operations gfs2_databuf_lops = {
1116         .lo_before_commit = databuf_lo_before_commit,
1117         .lo_after_commit = databuf_lo_after_commit,
1118         .lo_scan_elements = databuf_lo_scan_elements,
1119         .lo_after_scan = databuf_lo_after_scan,
1120         .lo_name = "databuf",
1121 };
1122
1123 const struct gfs2_log_operations *gfs2_log_ops[] = {
1124         &gfs2_databuf_lops,
1125         &gfs2_buf_lops,
1126         &gfs2_revoke_lops,
1127         NULL,
1128 };
1129