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, int 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(GFP_NOIO, BIO_MAX_VECS);
 269
 270         bio->bi_iter.bi_sector = blkno << sdp->sd_fsb2bb_shift;
 271         bio_set_dev(bio, sb->s_bdev);
 272         bio->bi_end_io = end_io;
 273         bio->bi_private = sdp;
 274
 275         return bio;
 276 }
 277
 278 /**
 279  * gfs2_log_get_bio - Get cached log bio, or allocate a new one
 280  * @sdp: The super block
 281  * @blkno: The device block number we want to write to
 282  * @biop: The bio to get or allocate
 283  * @op: REQ_OP
 284  * @end_io: The bi_end_io callback
 285  * @flush: Always flush the current bio and allocate a new one?
 286  *
 287  * If there is a cached bio, then if the next block number is sequential
 288  * with the previous one, return it, otherwise flush the bio to the
 289  * device. If there is no cached bio, or we just flushed it, then
 290  * allocate a new one.
 291  *
 292  * Returns: The bio to use for log writes
 293  */
 294
 295 static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno,
 296                                     struct bio **biop, int op,
 297                                     bio_end_io_t *end_io, bool flush)
 298 {
 299         struct bio *bio = *biop;
 300
 301         if (bio) {
 302                 u64 nblk;
 303
 304                 nblk = bio_end_sector(bio);
 305                 nblk >>= sdp->sd_fsb2bb_shift;
 306                 if (blkno == nblk && !flush)
 307                         return bio;
 308                 gfs2_log_submit_bio(biop, op);
 309         }
 310
 311         *biop = gfs2_log_alloc_bio(sdp, blkno, end_io);
 312         return *biop;
 313 }
 314
 315 /**
 316  * gfs2_log_write - write to log
 317  * @sdp: the filesystem
 318  * @jd: The journal descriptor
 319  * @page: the page to write
 320  * @size: the size of the data to write
 321  * @offset: the offset within the page
 322  * @blkno: block number of the log entry
 323  *
 324  * Try and add the page segment to the current bio. If that fails,
 325  * submit the current bio to the device and create a new one, and
 326  * then add the page segment to that.
 327  */
 328
 329 void gfs2_log_write(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
 330                     struct page *page, unsigned size, unsigned offset,
 331                     u64 blkno)
 332 {
 333         struct bio *bio;
 334         int ret;
 335
 336         bio = gfs2_log_get_bio(sdp, blkno, &jd->jd_log_bio, REQ_OP_WRITE,
 337                                gfs2_end_log_write, false);
 338         ret = bio_add_page(bio, page, size, offset);
 339         if (ret == 0) {
 340                 bio = gfs2_log_get_bio(sdp, blkno, &jd->jd_log_bio,
 341                                        REQ_OP_WRITE, gfs2_end_log_write, true);
 342                 ret = bio_add_page(bio, page, size, offset);
 343                 WARN_ON(ret == 0);
 344         }
 345 }
 346
 347 /**
 348  * gfs2_log_write_bh - write a buffer's content to the log
 349  * @sdp: The super block
 350  * @bh: The buffer pointing to the in-place location
 351  *
 352  * This writes the content of the buffer to the next available location
 353  * in the log. The buffer will be unlocked once the i/o to the log has
 354  * completed.
 355  */
 356
 357 static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh)
 358 {
 359         u64 dblock;
 360
 361         dblock = gfs2_log_bmap(sdp->sd_jdesc, sdp->sd_log_flush_head);
 362         gfs2_log_incr_head(sdp);
 363         gfs2_log_write(sdp, sdp->sd_jdesc, bh->b_page, bh->b_size,
 364                        bh_offset(bh), dblock);
 365 }
 366
 367 /**
 368  * gfs2_log_write_page - write one block stored in a page, into the log
 369  * @sdp: The superblock
 370  * @page: The struct page
 371  *
 372  * This writes the first block-sized part of the page into the log. Note
 373  * that the page must have been allocated from the gfs2_page_pool mempool
 374  * and that after this has been called, ownership has been transferred and
 375  * the page may be freed at any time.
 376  */
 377
 378 static void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
 379 {
 380         struct super_block *sb = sdp->sd_vfs;
 381         u64 dblock;
 382
 383         dblock = gfs2_log_bmap(sdp->sd_jdesc, sdp->sd_log_flush_head);
 384         gfs2_log_incr_head(sdp);
 385         gfs2_log_write(sdp, sdp->sd_jdesc, page, sb->s_blocksize, 0, dblock);
 386 }
 387
 388 /**
 389  * gfs2_end_log_read - end I/O callback for reads from the log
 390  * @bio: The bio
 391  *
 392  * Simply unlock the pages in the bio. The main thread will wait on them and
 393  * process them in order as necessary.
 394  */
 395
 396 static void gfs2_end_log_read(struct bio *bio)
 397 {
 398         struct page *page;
 399         struct bio_vec *bvec;
 400         struct bvec_iter_all iter_all;
 401
 402         bio_for_each_segment_all(bvec, bio, iter_all) {
 403                 page = bvec->bv_page;
 404                 if (bio->bi_status) {
 405                         int err = blk_status_to_errno(bio->bi_status);
 406
 407                         SetPageError(page);
 408                         mapping_set_error(page->mapping, err);
 409                 }
 410                 unlock_page(page);
 411         }
 412
 413         bio_put(bio);
 414 }
 415
 416 /**
 417  * gfs2_jhead_pg_srch - Look for the journal head in a given page.
 418  * @jd: The journal descriptor
 419  * @head: The journal head to start from
 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  * @head: The journal head to start from
 454  * @done: If set, perform only cleanup, else search and set if found.
 455  *
 456  * Find the page with 'index' in the journal's mapping. Search the page for
 457  * the journal head if requested (cleanup == false). Release refs on the
 458  * page so the page cache can reclaim it (put_page() twice). We grabbed a
 459  * reference on this page two times, first when we did a find_or_create_page()
 460  * to obtain the page to add it to the bio and second when we do a
 461  * find_get_page() here to get the page to wait on while I/O on it is being
 462  * completed.
 463  * This function is also used to free up a page 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 page.
 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 page *page;
 474
 475         page = find_get_page(jd->jd_inode->i_mapping, index);
 476         wait_on_page_locked(page);
 477
 478         if (PageError(page))
 479                 *done = true;
 480
 481         if (!*done)
 482                 *done = gfs2_jhead_pg_srch(jd, head, page);
 483
 484         put_page(page); /* Once for find_get_page */
 485         put_page(page); /* Once more for find_or_create_page */
 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(GFP_NOIO, nr_iovecs);
 493         bio_copy_dev(new, prev);
 494         new->bi_iter.bi_sector = bio_end_sector(prev);
 495         new->bi_opf = prev->bi_opf;
 496         new->bi_write_hint = prev->bi_write_hint;
 497         bio_chain(new, prev);
 498         submit_bio(prev);
 499         return new;
 500 }
 501
 502 /**
 503  * gfs2_find_jhead - find the head of a log
 504  * @jd: The journal descriptor
 505  * @head: The log descriptor for the head of the log is returned here
 506  * @keep_cache: If set inode pages will not be truncated
 507  *
 508  * Do a search of a journal by reading it in large chunks using bios and find
 509  * the valid log entry with the highest sequence number.  (i.e. the log head)
 510  *
 511  * Returns: 0 on success, errno otherwise
 512  */
 513 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head,
 514                     bool keep_cache)
 515 {
 516         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 517         struct address_space *mapping = jd->jd_inode->i_mapping;
 518         unsigned int block = 0, blocks_submitted = 0, blocks_read = 0;
 519         unsigned int bsize = sdp->sd_sb.sb_bsize, off;
 520         unsigned int bsize_shift = sdp->sd_sb.sb_bsize_shift;
 521         unsigned int shift = PAGE_SHIFT - bsize_shift;
 522         unsigned int max_blocks = 2 * 1024 * 1024 >> bsize_shift;
 523         struct gfs2_journal_extent *je;
 524         int sz, ret = 0;
 525         struct bio *bio = NULL;
 526         struct page *page = NULL;
 527         bool done = false;
 528         errseq_t since;
 529
 530         memset(head, 0, sizeof(*head));
 531         if (list_empty(&jd->extent_list))
 532                 gfs2_map_journal_extents(sdp, jd);
 533
 534         since = filemap_sample_wb_err(mapping);
 535         list_for_each_entry(je, &jd->extent_list, list) {
 536                 u64 dblock = je->dblock;
 537
 538                 for (; block < je->lblock + je->blocks; block++, dblock++) {
 539                         if (!page) {
 540                                 page = find_or_create_page(mapping,
 541                                                 block >> shift, GFP_NOFS);
 542                                 if (!page) {
 543                                         ret = -ENOMEM;
 544                                         done = true;
 545                                         goto out;
 546                                 }
 547                                 off = 0;
 548                         }
 549
 550                         if (bio && (off || block < blocks_submitted + max_blocks)) {
 551                                 sector_t sector = dblock << sdp->sd_fsb2bb_shift;
 552
 553                                 if (bio_end_sector(bio) == sector) {
 554                                         sz = bio_add_page(bio, page, bsize, off);
 555                                         if (sz == bsize)
 556                                                 goto block_added;
 557                                 }
 558                                 if (off) {
 559                                         unsigned int blocks =
 560                                                 (PAGE_SIZE - off) >> bsize_shift;
 561
 562                                         bio = gfs2_chain_bio(bio, blocks);
 563                                         goto add_block_to_new_bio;
 564                                 }
 565                         }
 566
 567                         if (bio) {
 568                                 blocks_submitted = block;
 569                                 submit_bio(bio);
 570                         }
 571
 572                         bio = gfs2_log_alloc_bio(sdp, dblock, gfs2_end_log_read);
 573                         bio->bi_opf = REQ_OP_READ;
 574 add_block_to_new_bio:
 575                         sz = bio_add_page(bio, page, bsize, off);
 576                         BUG_ON(sz != bsize);
 577 block_added:
 578                         off += bsize;
 579                         if (off == PAGE_SIZE)
 580                                 page = NULL;
 581                         if (blocks_submitted <= blocks_read + max_blocks) {
 582                                 /* Keep at least one bio in flight */
 583                                 continue;
 584                         }
 585
 586                         gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done);
 587                         blocks_read += PAGE_SIZE >> bsize_shift;
 588                         if (done)
 589                                 goto out;  /* found */
 590                 }
 591         }
 592
 593 out:
 594         if (bio)
 595                 submit_bio(bio);
 596         while (blocks_read < block) {
 597                 gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done);
 598                 blocks_read += PAGE_SIZE >> bsize_shift;
 599         }
 600
 601         if (!ret)
 602                 ret = filemap_check_wb_err(mapping, since);
 603
 604         if (!keep_cache)
 605                 truncate_inode_pages(mapping, 0);
 606
 607         return ret;
 608 }
 609
 610 static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
 611                                       u32 ld_length, u32 ld_data1)
 612 {
 613         struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 614         struct gfs2_log_descriptor *ld = page_address(page);
 615         clear_page(ld);
 616         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 617         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 618         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 619         ld->ld_type = cpu_to_be32(ld_type);
 620         ld->ld_length = cpu_to_be32(ld_length);
 621         ld->ld_data1 = cpu_to_be32(ld_data1);
 622         ld->ld_data2 = 0;
 623         return page;
 624 }
 625
 626 static void gfs2_check_magic(struct buffer_head *bh)
 627 {
 628         void *kaddr;
 629         __be32 *ptr;
 630
 631         clear_buffer_escaped(bh);
 632         kaddr = kmap_atomic(bh->b_page);
 633         ptr = kaddr + bh_offset(bh);
 634         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 635                 set_buffer_escaped(bh);
 636         kunmap_atomic(kaddr);
 637 }
 638
 639 static int blocknr_cmp(void *priv, const struct list_head *a,
 640                        const struct list_head *b)
 641 {
 642         struct gfs2_bufdata *bda, *bdb;
 643
 644         bda = list_entry(a, struct gfs2_bufdata, bd_list);
 645         bdb = list_entry(b, struct gfs2_bufdata, bd_list);
 646
 647         if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
 648                 return -1;
 649         if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
 650                 return 1;
 651         return 0;
 652 }
 653
 654 static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
 655                                 unsigned int total, struct list_head *blist,
 656                                 bool is_databuf)
 657 {
 658         struct gfs2_log_descriptor *ld;
 659         struct gfs2_bufdata *bd1 = NULL, *bd2;
 660         struct page *page;
 661         unsigned int num;
 662         unsigned n;
 663         __be64 *ptr;
 664
 665         gfs2_log_lock(sdp);
 666         list_sort(NULL, blist, blocknr_cmp);
 667         bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
 668         while(total) {
 669                 num = total;
 670                 if (total > limit)
 671                         num = limit;
 672                 gfs2_log_unlock(sdp);
 673                 page = gfs2_get_log_desc(sdp,
 674                                          is_databuf ? GFS2_LOG_DESC_JDATA :
 675                                          GFS2_LOG_DESC_METADATA, num + 1, num);
 676                 ld = page_address(page);
 677                 gfs2_log_lock(sdp);
 678                 ptr = (__be64 *)(ld + 1);
 679
 680                 n = 0;
 681                 list_for_each_entry_continue(bd1, blist, bd_list) {
 682                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 683                         if (is_databuf) {
 684                                 gfs2_check_magic(bd1->bd_bh);
 685                                 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
 686                         }
 687                         if (++n >= num)
 688                                 break;
 689                 }
 690
 691                 gfs2_log_unlock(sdp);
 692                 gfs2_log_write_page(sdp, page);
 693                 gfs2_log_lock(sdp);
 694
 695                 n = 0;
 696                 list_for_each_entry_continue(bd2, blist, bd_list) {
 697                         get_bh(bd2->bd_bh);
 698                         gfs2_log_unlock(sdp);
 699                         lock_buffer(bd2->bd_bh);
 700
 701                         if (buffer_escaped(bd2->bd_bh)) {
 702                                 void *kaddr;
 703                                 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 704                                 ptr = page_address(page);
 705                                 kaddr = kmap_atomic(bd2->bd_bh->b_page);
 706                                 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh),
 707                                        bd2->bd_bh->b_size);
 708                                 kunmap_atomic(kaddr);
 709                                 *(__be32 *)ptr = 0;
 710                                 clear_buffer_escaped(bd2->bd_bh);
 711                                 unlock_buffer(bd2->bd_bh);
 712                                 brelse(bd2->bd_bh);
 713                                 gfs2_log_write_page(sdp, page);
 714                         } else {
 715                                 gfs2_log_write_bh(sdp, bd2->bd_bh);
 716                         }
 717                         gfs2_log_lock(sdp);
 718                         if (++n >= num)
 719                                 break;
 720                 }
 721
 722                 BUG_ON(total < num);
 723                 total -= num;
 724         }
 725         gfs2_log_unlock(sdp);
 726 }
 727
 728 static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 729 {
 730         unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
 731         unsigned int nbuf;
 732         if (tr == NULL)
 733                 return;
 734         nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
 735         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0);
 736 }
 737
 738 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 739 {
 740         struct list_head *head;
 741         struct gfs2_bufdata *bd;
 742
 743         if (tr == NULL)
 744                 return;
 745
 746         head = &tr->tr_buf;
 747         while (!list_empty(head)) {
 748                 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
 749                 list_del_init(&bd->bd_list);
 750                 gfs2_unpin(sdp, bd->bd_bh, tr);
 751         }
 752 }
 753
 754 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 755                                struct gfs2_log_header_host *head, int pass)
 756 {
 757         if (pass != 0)
 758                 return;
 759
 760         jd->jd_found_blocks = 0;
 761         jd->jd_replayed_blocks = 0;
 762 }
 763
 764 #define obsolete_rgrp_replay \
 765 "Replaying 0x%llx from jid=%d/0x%llx but we already have a bh!\n"
 766 #define obsolete_rgrp_replay2 \
 767 "busy:%d, pinned:%d rg_gen:0x%llx, j_gen:0x%llx\n"
 768
 769 static void obsolete_rgrp(struct gfs2_jdesc *jd, struct buffer_head *bh_log,
 770                           u64 blkno)
 771 {
 772         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 773         struct gfs2_rgrpd *rgd;
 774         struct gfs2_rgrp *jrgd = (struct gfs2_rgrp *)bh_log->b_data;
 775
 776         rgd = gfs2_blk2rgrpd(sdp, blkno, false);
 777         if (rgd && rgd->rd_addr == blkno &&
 778             rgd->rd_bits && rgd->rd_bits->bi_bh) {
 779                 fs_info(sdp, obsolete_rgrp_replay, (unsigned long long)blkno,
 780                         jd->jd_jid, bh_log->b_blocknr);
 781                 fs_info(sdp, obsolete_rgrp_replay2,
 782                         buffer_busy(rgd->rd_bits->bi_bh) ? 1 : 0,
 783                         buffer_pinned(rgd->rd_bits->bi_bh),
 784                         rgd->rd_igeneration,
 785                         be64_to_cpu(jrgd->rg_igeneration));
 786                 gfs2_dump_glock(NULL, rgd->rd_gl, true);
 787         }
 788 }
 789
 790 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
 791                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 792                                 int pass)
 793 {
 794         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 795         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 796         struct gfs2_glock *gl = ip->i_gl;
 797         unsigned int blks = be32_to_cpu(ld->ld_data1);
 798         struct buffer_head *bh_log, *bh_ip;
 799         u64 blkno;
 800         int error = 0;
 801
 802         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 803                 return 0;
 804
 805         gfs2_replay_incr_blk(jd, &start);
 806
 807         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 808                 blkno = be64_to_cpu(*ptr++);
 809
 810                 jd->jd_found_blocks++;
 811
 812                 if (gfs2_revoke_check(jd, blkno, start))
 813                         continue;
 814
 815                 error = gfs2_replay_read_block(jd, start, &bh_log);
 816                 if (error)
 817                         return error;
 818
 819                 bh_ip = gfs2_meta_new(gl, blkno);
 820                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 821
 822                 if (gfs2_meta_check(sdp, bh_ip))
 823                         error = -EIO;
 824                 else {
 825                         struct gfs2_meta_header *mh =
 826                                 (struct gfs2_meta_header *)bh_ip->b_data;
 827
 828                         if (mh->mh_type == cpu_to_be32(GFS2_METATYPE_RG))
 829                                 obsolete_rgrp(jd, bh_log, blkno);
 830
 831                         mark_buffer_dirty(bh_ip);
 832                 }
 833                 brelse(bh_log);
 834                 brelse(bh_ip);
 835
 836                 if (error)
 837                         break;
 838
 839                 jd->jd_replayed_blocks++;
 840         }
 841
 842         return error;
 843 }
 844
 845 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 846 {
 847         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 848         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 849
 850         if (error) {
 851                 gfs2_inode_metasync(ip->i_gl);
 852                 return;
 853         }
 854         if (pass != 1)
 855                 return;
 856
 857         gfs2_inode_metasync(ip->i_gl);
 858
 859         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 860                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
 861 }
 862
 863 static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 864 {
 865         struct gfs2_meta_header *mh;
 866         unsigned int offset;
 867         struct list_head *head = &sdp->sd_log_revokes;
 868         struct gfs2_bufdata *bd;
 869         struct page *page;
 870         unsigned int length;
 871
 872         gfs2_flush_revokes(sdp);
 873         if (!sdp->sd_log_num_revoke)
 874                 return;
 875
 876         length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke);
 877         page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
 878         offset = sizeof(struct gfs2_log_descriptor);
 879
 880         list_for_each_entry(bd, head, bd_list) {
 881                 sdp->sd_log_num_revoke--;
 882
 883                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 884                         gfs2_log_write_page(sdp, page);
 885                         page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 886                         mh = page_address(page);
 887                         clear_page(mh);
 888                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 889                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 890                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 891                         offset = sizeof(struct gfs2_meta_header);
 892                 }
 893
 894                 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
 895                 offset += sizeof(u64);
 896         }
 897         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 898
 899         gfs2_log_write_page(sdp, page);
 900 }
 901
 902 void gfs2_drain_revokes(struct gfs2_sbd *sdp)
 903 {
 904         struct list_head *head = &sdp->sd_log_revokes;
 905         struct gfs2_bufdata *bd;
 906         struct gfs2_glock *gl;
 907
 908         while (!list_empty(head)) {
 909                 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
 910                 list_del_init(&bd->bd_list);
 911                 gl = bd->bd_gl;
 912                 gfs2_glock_remove_revoke(gl);
 913                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 914         }
 915 }
 916
 917 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 918 {
 919         gfs2_drain_revokes(sdp);
 920 }
 921
 922 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 923                                   struct gfs2_log_header_host *head, int pass)
 924 {
 925         if (pass != 0)
 926                 return;
 927
 928         jd->jd_found_revokes = 0;
 929         jd->jd_replay_tail = head->lh_tail;
 930 }
 931
 932 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
 933                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 934                                    int pass)
 935 {
 936         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 937         unsigned int blks = be32_to_cpu(ld->ld_length);
 938         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 939         struct buffer_head *bh;
 940         unsigned int offset;
 941         u64 blkno;
 942         int first = 1;
 943         int error;
 944
 945         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 946                 return 0;
 947
 948         offset = sizeof(struct gfs2_log_descriptor);
 949
 950         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 951                 error = gfs2_replay_read_block(jd, start, &bh);
 952                 if (error)
 953                         return error;
 954
 955                 if (!first)
 956                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 957
 958                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 959                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 960
 961                         error = gfs2_revoke_add(jd, blkno, start);
 962                         if (error < 0) {
 963                                 brelse(bh);
 964                                 return error;
 965                         }
 966                         else if (error)
 967                                 jd->jd_found_revokes++;
 968
 969                         if (!--revokes)
 970                                 break;
 971                         offset += sizeof(u64);
 972                 }
 973
 974                 brelse(bh);
 975                 offset = sizeof(struct gfs2_meta_header);
 976                 first = 0;
 977         }
 978
 979         return 0;
 980 }
 981
 982 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 983 {
 984         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 985
 986         if (error) {
 987                 gfs2_revoke_clean(jd);
 988                 return;
 989         }
 990         if (pass != 1)
 991                 return;
 992
 993         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 994                 jd->jd_jid, jd->jd_found_revokes);
 995
 996         gfs2_revoke_clean(jd);
 997 }
 998
 999 /**
1000  * databuf_lo_before_commit - Scan the data buffers, writing as we go
1001  * @sdp: The filesystem
1002  * @tr: The system transaction being flushed
1003  */
1004
1005 static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1006 {
1007         unsigned int limit = databuf_limit(sdp);
1008         unsigned int nbuf;
1009         if (tr == NULL)
1010                 return;
1011         nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
1012         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1);
1013 }
1014
1015 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
1016                                     struct gfs2_log_descriptor *ld,
1017                                     __be64 *ptr, int pass)
1018 {
1019         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1020         struct gfs2_glock *gl = ip->i_gl;
1021         unsigned int blks = be32_to_cpu(ld->ld_data1);
1022         struct buffer_head *bh_log, *bh_ip;
1023         u64 blkno;
1024         u64 esc;
1025         int error = 0;
1026
1027         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
1028                 return 0;
1029
1030         gfs2_replay_incr_blk(jd, &start);
1031         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
1032                 blkno = be64_to_cpu(*ptr++);
1033                 esc = be64_to_cpu(*ptr++);
1034
1035                 jd->jd_found_blocks++;
1036
1037                 if (gfs2_revoke_check(jd, blkno, start))
1038                         continue;
1039
1040                 error = gfs2_replay_read_block(jd, start, &bh_log);
1041                 if (error)
1042                         return error;
1043
1044                 bh_ip = gfs2_meta_new(gl, blkno);
1045                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
1046
1047                 /* Unescape */
1048                 if (esc) {
1049                         __be32 *eptr = (__be32 *)bh_ip->b_data;
1050                         *eptr = cpu_to_be32(GFS2_MAGIC);
1051                 }
1052                 mark_buffer_dirty(bh_ip);
1053
1054                 brelse(bh_log);
1055                 brelse(bh_ip);
1056
1057                 jd->jd_replayed_blocks++;
1058         }
1059
1060         return error;
1061 }
1062
1063 /* FIXME: sort out accounting for log blocks etc. */
1064
1065 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
1066 {
1067         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1068         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
1069
1070         if (error) {
1071                 gfs2_inode_metasync(ip->i_gl);
1072                 return;
1073         }
1074         if (pass != 1)
1075                 return;
1076
1077         /* data sync? */
1078         gfs2_inode_metasync(ip->i_gl);
1079
1080         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
1081                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
1082 }
1083
1084 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1085 {
1086         struct list_head *head;
1087         struct gfs2_bufdata *bd;
1088
1089         if (tr == NULL)
1090                 return;
1091
1092         head = &tr->tr_databuf;
1093         while (!list_empty(head)) {
1094                 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
1095                 list_del_init(&bd->bd_list);
1096                 gfs2_unpin(sdp, bd->bd_bh, tr);
1097         }
1098 }
1099
1100
1101 static const struct gfs2_log_operations gfs2_buf_lops = {
1102         .lo_before_commit = buf_lo_before_commit,
1103         .lo_after_commit = buf_lo_after_commit,
1104         .lo_before_scan = buf_lo_before_scan,
1105         .lo_scan_elements = buf_lo_scan_elements,
1106         .lo_after_scan = buf_lo_after_scan,
1107         .lo_name = "buf",
1108 };
1109
1110 static const struct gfs2_log_operations gfs2_revoke_lops = {
1111         .lo_before_commit = revoke_lo_before_commit,
1112         .lo_after_commit = revoke_lo_after_commit,
1113         .lo_before_scan = revoke_lo_before_scan,
1114         .lo_scan_elements = revoke_lo_scan_elements,
1115         .lo_after_scan = revoke_lo_after_scan,
1116         .lo_name = "revoke",
1117 };
1118
1119 static const struct gfs2_log_operations gfs2_databuf_lops = {
1120         .lo_before_commit = databuf_lo_before_commit,
1121         .lo_after_commit = databuf_lo_after_commit,
1122         .lo_scan_elements = databuf_lo_scan_elements,
1123         .lo_after_scan = databuf_lo_after_scan,
1124         .lo_name = "databuf",
1125 };
1126
1127 const struct gfs2_log_operations *gfs2_log_ops[] = {
1128         &gfs2_databuf_lops,
1129         &gfs2_buf_lops,
1130         &gfs2_revoke_lops,
1131         NULL,
1132 };
1133