fs/gfs2/lops.c

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
gfs2_pin
buffer_is_rgrp
maybe_release_space
gfs2_unpin
gfs2_log_incr_head
gfs2_log_bmap
gfs2_end_log_write_bh
gfs2_end_log_write
gfs2_log_submit_bio
gfs2_log_alloc_bio
gfs2_log_get_bio
gfs2_log_write
gfs2_log_write_bh
gfs2_log_write_page
gfs2_end_log_read
gfs2_jhead_pg_srch
gfs2_jhead_process_page
gfs2_chain_bio
gfs2_find_jhead
gfs2_get_log_desc
gfs2_check_magic
blocknr_cmp
gfs2_before_commit
buf_lo_before_commit
buf_lo_after_commit
buf_lo_before_scan
buf_lo_scan_elements
gfs2_meta_sync
buf_lo_after_scan
revoke_lo_before_commit
revoke_lo_after_commit
revoke_lo_before_scan
revoke_lo_scan_elements
revoke_lo_after_scan
databuf_lo_before_commit
databuf_lo_scan_elements
databuf_lo_after_scan
databuf_lo_after_commit
   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   4  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
   5  */
   6 
   7 #include <linux/sched.h>
   8 #include <linux/slab.h>
   9 #include <linux/spinlock.h>
  10 #include <linux/completion.h>
  11 #include <linux/buffer_head.h>
  12 #include <linux/mempool.h>
  13 #include <linux/gfs2_ondisk.h>
  14 #include <linux/bio.h>
  15 #include <linux/fs.h>
  16 #include <linux/list_sort.h>
  17 #include <linux/blkdev.h>
  18 
  19 #include "bmap.h"
  20 #include "dir.h"
  21 #include "gfs2.h"
  22 #include "incore.h"
  23 #include "inode.h"
  24 #include "glock.h"
  25 #include "log.h"
  26 #include "lops.h"
  27 #include "meta_io.h"
  28 #include "recovery.h"
  29 #include "rgrp.h"
  30 #include "trans.h"
  31 #include "util.h"
  32 #include "trace_gfs2.h"
  33 
  34 /**
  35  * gfs2_pin - Pin a buffer in memory
  36  * @sdp: The superblock
  37  * @bh: The buffer to be pinned
  38  *
  39  * The log lock must be held when calling this function
  40  */
  41 void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
  42 {
  43         struct gfs2_bufdata *bd;
  44 
  45         BUG_ON(!current->journal_info);
  46 
  47         clear_buffer_dirty(bh);
  48         if (test_set_buffer_pinned(bh))
  49                 gfs2_assert_withdraw(sdp, 0);
  50         if (!buffer_uptodate(bh))
  51                 gfs2_io_error_bh_wd(sdp, bh);
  52         bd = bh->b_private;
  53         /* If this buffer is in the AIL and it has already been written
  54          * to in-place disk block, remove it from the AIL.
  55          */
  56         spin_lock(&sdp->sd_ail_lock);
  57         if (bd->bd_tr)
  58                 list_move(&bd->bd_ail_st_list, &bd->bd_tr->tr_ail2_list);
  59         spin_unlock(&sdp->sd_ail_lock);
  60         get_bh(bh);
  61         atomic_inc(&sdp->sd_log_pinned);
  62         trace_gfs2_pin(bd, 1);
  63 }
  64 
  65 static bool buffer_is_rgrp(const struct gfs2_bufdata *bd)
  66 {
  67         return bd->bd_gl->gl_name.ln_type == LM_TYPE_RGRP;
  68 }
  69 
  70 static void maybe_release_space(struct gfs2_bufdata *bd)
  71 {
  72         struct gfs2_glock *gl = bd->bd_gl;
  73         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
  74         struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
  75         unsigned int index = bd->bd_bh->b_blocknr - gl->gl_name.ln_number;
  76         struct gfs2_bitmap *bi = rgd->rd_bits + index;
  77 
  78         if (bi->bi_clone == NULL)
  79                 return;
  80         if (sdp->sd_args.ar_discard)
  81                 gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi, 1, NULL);
  82         memcpy(bi->bi_clone + bi->bi_offset,
  83                bd->bd_bh->b_data + bi->bi_offset, bi->bi_bytes);
  84         clear_bit(GBF_FULL, &bi->bi_flags);
  85         rgd->rd_free_clone = rgd->rd_free;
  86         rgd->rd_extfail_pt = rgd->rd_free;
  87 }
  88 
  89 /**
  90  * gfs2_unpin - Unpin a buffer
  91  * @sdp: the filesystem the buffer belongs to
  92  * @bh: The buffer to unpin
  93  * @ai:
  94  * @flags: The inode dirty flags
  95  *
  96  */
  97 
  98 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
  99                        struct gfs2_trans *tr)
 100 {
 101         struct gfs2_bufdata *bd = bh->b_private;
 102 
 103         BUG_ON(!buffer_uptodate(bh));
 104         BUG_ON(!buffer_pinned(bh));
 105 
 106         lock_buffer(bh);
 107         mark_buffer_dirty(bh);
 108         clear_buffer_pinned(bh);
 109 
 110         if (buffer_is_rgrp(bd))
 111                 maybe_release_space(bd);
 112 
 113         spin_lock(&sdp->sd_ail_lock);
 114         if (bd->bd_tr) {
 115                 list_del(&bd->bd_ail_st_list);
 116                 brelse(bh);
 117         } else {
 118                 struct gfs2_glock *gl = bd->bd_gl;
 119                 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
 120                 atomic_inc(&gl->gl_ail_count);
 121         }
 122         bd->bd_tr = tr;
 123         list_add(&bd->bd_ail_st_list, &tr->tr_ail1_list);
 124         spin_unlock(&sdp->sd_ail_lock);
 125 
 126         clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 127         trace_gfs2_pin(bd, 0);
 128         unlock_buffer(bh);
 129         atomic_dec(&sdp->sd_log_pinned);
 130 }
 131 
 132 static void gfs2_log_incr_head(struct gfs2_sbd *sdp)
 133 {
 134         BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) &&
 135                (sdp->sd_log_flush_head != sdp->sd_log_head));
 136 
 137         if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks)
 138                 sdp->sd_log_flush_head = 0;
 139 }
 140 
 141 u64 gfs2_log_bmap(struct gfs2_sbd *sdp)
 142 {
 143         unsigned int lbn = sdp->sd_log_flush_head;
 144         struct gfs2_journal_extent *je;
 145         u64 block;
 146 
 147         list_for_each_entry(je, &sdp->sd_jdesc->extent_list, list) {
 148                 if ((lbn >= je->lblock) && (lbn < (je->lblock + je->blocks))) {
 149                         block = je->dblock + lbn - je->lblock;
 150                         gfs2_log_incr_head(sdp);
 151                         return block;
 152                 }
 153         }
 154 
 155         return -1;
 156 }
 157 
 158 /**
 159  * gfs2_end_log_write_bh - end log write of pagecache data with buffers
 160  * @sdp: The superblock
 161  * @bvec: The bio_vec
 162  * @error: The i/o status
 163  *
 164  * This finds the relevant buffers and unlocks them and sets the
 165  * error flag according to the status of the i/o request. This is
 166  * used when the log is writing data which has an in-place version
 167  * that is pinned in the pagecache.
 168  */
 169 
 170 static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp,
 171                                   struct bio_vec *bvec,
 172                                   blk_status_t error)
 173 {
 174         struct buffer_head *bh, *next;
 175         struct page *page = bvec->bv_page;
 176         unsigned size;
 177 
 178         bh = page_buffers(page);
 179         size = bvec->bv_len;
 180         while (bh_offset(bh) < bvec->bv_offset)
 181                 bh = bh->b_this_page;
 182         do {
 183                 if (error)
 184                         mark_buffer_write_io_error(bh);
 185                 unlock_buffer(bh);
 186                 next = bh->b_this_page;
 187                 size -= bh->b_size;
 188                 brelse(bh);
 189                 bh = next;
 190         } while(bh && size);
 191 }
 192 
 193 /**
 194  * gfs2_end_log_write - end of i/o to the log
 195  * @bio: The bio
 196  *
 197  * Each bio_vec contains either data from the pagecache or data
 198  * relating to the log itself. Here we iterate over the bio_vec
 199  * array, processing both kinds of data.
 200  *
 201  */
 202 
 203 static void gfs2_end_log_write(struct bio *bio)
 204 {
 205         struct gfs2_sbd *sdp = bio->bi_private;
 206         struct bio_vec *bvec;
 207         struct page *page;
 208         struct bvec_iter_all iter_all;
 209 
 210         if (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                 wake_up(&sdp->sd_logd_waitq);
 214         }
 215 
 216         bio_for_each_segment_all(bvec, bio, iter_all) {
 217                 page = bvec->bv_page;
 218                 if (page_has_buffers(page))
 219                         gfs2_end_log_write_bh(sdp, bvec, bio->bi_status);
 220                 else
 221                         mempool_free(page, gfs2_page_pool);
 222         }
 223 
 224         bio_put(bio);
 225         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 226                 wake_up(&sdp->sd_log_flush_wait);
 227 }
 228 
 229 /**
 230  * gfs2_log_submit_bio - Submit any pending log bio
 231  * @biop: Address of the bio pointer
 232  * @opf: REQ_OP | op_flags
 233  *
 234  * Submit any pending part-built or full bio to the block device. If
 235  * there is no pending bio, then this is a no-op.
 236  */
 237 
 238 void gfs2_log_submit_bio(struct bio **biop, int opf)
 239 {
 240         struct bio *bio = *biop;
 241         if (bio) {
 242                 struct gfs2_sbd *sdp = bio->bi_private;
 243                 atomic_inc(&sdp->sd_log_in_flight);
 244                 bio->bi_opf = opf;
 245                 submit_bio(bio);
 246                 *biop = NULL;
 247         }
 248 }
 249 
 250 /**
 251  * gfs2_log_alloc_bio - Allocate a bio
 252  * @sdp: The super block
 253  * @blkno: The device block number we want to write to
 254  * @end_io: The bi_end_io callback
 255  *
 256  * Allocate a new bio, initialize it with the given parameters and return it.
 257  *
 258  * Returns: The newly allocated bio
 259  */
 260 
 261 static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno,
 262                                       bio_end_io_t *end_io)
 263 {
 264         struct super_block *sb = sdp->sd_vfs;
 265         struct bio *bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
 266 
 267         bio->bi_iter.bi_sector = blkno << sdp->sd_fsb2bb_shift;
 268         bio_set_dev(bio, sb->s_bdev);
 269         bio->bi_end_io = end_io;
 270         bio->bi_private = sdp;
 271 
 272         return bio;
 273 }
 274 
 275 /**
 276  * gfs2_log_get_bio - Get cached log bio, or allocate a new one
 277  * @sdp: The super block
 278  * @blkno: The device block number we want to write to
 279  * @bio: The bio to get or allocate
 280  * @op: REQ_OP
 281  * @end_io: The bi_end_io callback
 282  * @flush: Always flush the current bio and allocate a new one?
 283  *
 284  * If there is a cached bio, then if the next block number is sequential
 285  * with the previous one, return it, otherwise flush the bio to the
 286  * device. If there is no cached bio, or we just flushed it, then
 287  * allocate a new one.
 288  *
 289  * Returns: The bio to use for log writes
 290  */
 291 
 292 static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno,
 293                                     struct bio **biop, int op,
 294                                     bio_end_io_t *end_io, bool flush)
 295 {
 296         struct bio *bio = *biop;
 297 
 298         if (bio) {
 299                 u64 nblk;
 300 
 301                 nblk = bio_end_sector(bio);
 302                 nblk >>= sdp->sd_fsb2bb_shift;
 303                 if (blkno == nblk && !flush)
 304                         return bio;
 305                 gfs2_log_submit_bio(biop, op);
 306         }
 307 
 308         *biop = gfs2_log_alloc_bio(sdp, blkno, end_io);
 309         return *biop;
 310 }
 311 
 312 /**
 313  * gfs2_log_write - write to log
 314  * @sdp: the filesystem
 315  * @page: the page to write
 316  * @size: the size of the data to write
 317  * @offset: the offset within the page 
 318  * @blkno: block number of the log entry
 319  *
 320  * Try and add the page segment to the current bio. If that fails,
 321  * submit the current bio to the device and create a new one, and
 322  * then add the page segment to that.
 323  */
 324 
 325 void gfs2_log_write(struct gfs2_sbd *sdp, struct page *page,
 326                     unsigned size, unsigned offset, u64 blkno)
 327 {
 328         struct bio *bio;
 329         int ret;
 330 
 331         bio = gfs2_log_get_bio(sdp, blkno, &sdp->sd_log_bio, REQ_OP_WRITE,
 332                                gfs2_end_log_write, false);
 333         ret = bio_add_page(bio, page, size, offset);
 334         if (ret == 0) {
 335                 bio = gfs2_log_get_bio(sdp, blkno, &sdp->sd_log_bio,
 336                                        REQ_OP_WRITE, gfs2_end_log_write, true);
 337                 ret = bio_add_page(bio, page, size, offset);
 338                 WARN_ON(ret == 0);
 339         }
 340 }
 341 
 342 /**
 343  * gfs2_log_write_bh - write a buffer's content to the log
 344  * @sdp: The super block
 345  * @bh: The buffer pointing to the in-place location
 346  * 
 347  * This writes the content of the buffer to the next available location
 348  * in the log. The buffer will be unlocked once the i/o to the log has
 349  * completed.
 350  */
 351 
 352 static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh)
 353 {
 354         gfs2_log_write(sdp, bh->b_page, bh->b_size, bh_offset(bh),
 355                        gfs2_log_bmap(sdp));
 356 }
 357 
 358 /**
 359  * gfs2_log_write_page - write one block stored in a page, into the log
 360  * @sdp: The superblock
 361  * @page: The struct page
 362  *
 363  * This writes the first block-sized part of the page into the log. Note
 364  * that the page must have been allocated from the gfs2_page_pool mempool
 365  * and that after this has been called, ownership has been transferred and
 366  * the page may be freed at any time.
 367  */
 368 
 369 void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
 370 {
 371         struct super_block *sb = sdp->sd_vfs;
 372         gfs2_log_write(sdp, page, sb->s_blocksize, 0,
 373                        gfs2_log_bmap(sdp));
 374 }
 375 
 376 /**
 377  * gfs2_end_log_read - end I/O callback for reads from the log
 378  * @bio: The bio
 379  *
 380  * Simply unlock the pages in the bio. The main thread will wait on them and
 381  * process them in order as necessary.
 382  */
 383 
 384 static void gfs2_end_log_read(struct bio *bio)
 385 {
 386         struct page *page;
 387         struct bio_vec *bvec;
 388         struct bvec_iter_all iter_all;
 389 
 390         bio_for_each_segment_all(bvec, bio, iter_all) {
 391                 page = bvec->bv_page;
 392                 if (bio->bi_status) {
 393                         int err = blk_status_to_errno(bio->bi_status);
 394 
 395                         SetPageError(page);
 396                         mapping_set_error(page->mapping, err);
 397                 }
 398                 unlock_page(page);
 399         }
 400 
 401         bio_put(bio);
 402 }
 403 
 404 /**
 405  * gfs2_jhead_pg_srch - Look for the journal head in a given page.
 406  * @jd: The journal descriptor
 407  * @page: The page to look in
 408  *
 409  * Returns: 1 if found, 0 otherwise.
 410  */
 411 
 412 static bool gfs2_jhead_pg_srch(struct gfs2_jdesc *jd,
 413                               struct gfs2_log_header_host *head,
 414                               struct page *page)
 415 {
 416         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 417         struct gfs2_log_header_host uninitialized_var(lh);
 418         void *kaddr = kmap_atomic(page);
 419         unsigned int offset;
 420         bool ret = false;
 421 
 422         for (offset = 0; offset < PAGE_SIZE; offset += sdp->sd_sb.sb_bsize) {
 423                 if (!__get_log_header(sdp, kaddr + offset, 0, &lh)) {
 424                         if (lh.lh_sequence >= head->lh_sequence)
 425                                 *head = lh;
 426                         else {
 427                                 ret = true;
 428                                 break;
 429                         }
 430                 }
 431         }
 432         kunmap_atomic(kaddr);
 433         return ret;
 434 }
 435 
 436 /**
 437  * gfs2_jhead_process_page - Search/cleanup a page
 438  * @jd: The journal descriptor
 439  * @index: Index of the page to look into
 440  * @done: If set, perform only cleanup, else search and set if found.
 441  *
 442  * Find the page with 'index' in the journal's mapping. Search the page for
 443  * the journal head if requested (cleanup == false). Release refs on the
 444  * page so the page cache can reclaim it (put_page() twice). We grabbed a
 445  * reference on this page two times, first when we did a find_or_create_page()
 446  * to obtain the page to add it to the bio and second when we do a
 447  * find_get_page() here to get the page to wait on while I/O on it is being
 448  * completed.
 449  * This function is also used to free up a page we might've grabbed but not
 450  * used. Maybe we added it to a bio, but not submitted it for I/O. Or we
 451  * submitted the I/O, but we already found the jhead so we only need to drop
 452  * our references to the page.
 453  */
 454 
 455 static void gfs2_jhead_process_page(struct gfs2_jdesc *jd, unsigned long index,
 456                                     struct gfs2_log_header_host *head,
 457                                     bool *done)
 458 {
 459         struct page *page;
 460 
 461         page = find_get_page(jd->jd_inode->i_mapping, index);
 462         wait_on_page_locked(page);
 463 
 464         if (PageError(page))
 465                 *done = true;
 466 
 467         if (!*done)
 468                 *done = gfs2_jhead_pg_srch(jd, head, page);
 469 
 470         put_page(page); /* Once for find_get_page */
 471         put_page(page); /* Once more for find_or_create_page */
 472 }
 473 
 474 static struct bio *gfs2_chain_bio(struct bio *prev, unsigned int nr_iovecs)
 475 {
 476         struct bio *new;
 477 
 478         new = bio_alloc(GFP_NOIO, nr_iovecs);
 479         bio_copy_dev(new, prev);
 480         new->bi_iter.bi_sector = bio_end_sector(prev);
 481         new->bi_opf = prev->bi_opf;
 482         new->bi_write_hint = prev->bi_write_hint;
 483         bio_chain(new, prev);
 484         submit_bio(prev);
 485         return new;
 486 }
 487 
 488 /**
 489  * gfs2_find_jhead - find the head of a log
 490  * @jd: The journal descriptor
 491  * @head: The log descriptor for the head of the log is returned here
 492  *
 493  * Do a search of a journal by reading it in large chunks using bios and find
 494  * the valid log entry with the highest sequence number.  (i.e. the log head)
 495  *
 496  * Returns: 0 on success, errno otherwise
 497  */
 498 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head,
 499                     bool keep_cache)
 500 {
 501         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 502         struct address_space *mapping = jd->jd_inode->i_mapping;
 503         unsigned int block = 0, blocks_submitted = 0, blocks_read = 0;
 504         unsigned int bsize = sdp->sd_sb.sb_bsize, off;
 505         unsigned int bsize_shift = sdp->sd_sb.sb_bsize_shift;
 506         unsigned int shift = PAGE_SHIFT - bsize_shift;
 507         unsigned int max_blocks = 2 * 1024 * 1024 >> bsize_shift;
 508         struct gfs2_journal_extent *je;
 509         int sz, ret = 0;
 510         struct bio *bio = NULL;
 511         struct page *page = NULL;
 512         bool done = false;
 513         errseq_t since;
 514 
 515         memset(head, 0, sizeof(*head));
 516         if (list_empty(&jd->extent_list))
 517                 gfs2_map_journal_extents(sdp, jd);
 518 
 519         since = filemap_sample_wb_err(mapping);
 520         list_for_each_entry(je, &jd->extent_list, list) {
 521                 u64 dblock = je->dblock;
 522 
 523                 for (; block < je->lblock + je->blocks; block++, dblock++) {
 524                         if (!page) {
 525                                 page = find_or_create_page(mapping,
 526                                                 block >> shift, GFP_NOFS);
 527                                 if (!page) {
 528                                         ret = -ENOMEM;
 529                                         done = true;
 530                                         goto out;
 531                                 }
 532                                 off = 0;
 533                         }
 534 
 535                         if (bio && (off || block < blocks_submitted + max_blocks)) {
 536                                 sector_t sector = dblock << sdp->sd_fsb2bb_shift;
 537 
 538                                 if (bio_end_sector(bio) == sector) {
 539                                         sz = bio_add_page(bio, page, bsize, off);
 540                                         if (sz == bsize)
 541                                                 goto block_added;
 542                                 }
 543                                 if (off) {
 544                                         unsigned int blocks =
 545                                                 (PAGE_SIZE - off) >> bsize_shift;
 546 
 547                                         bio = gfs2_chain_bio(bio, blocks);
 548                                         goto add_block_to_new_bio;
 549                                 }
 550                         }
 551 
 552                         if (bio) {
 553                                 blocks_submitted = block;
 554                                 submit_bio(bio);
 555                         }
 556 
 557                         bio = gfs2_log_alloc_bio(sdp, dblock, gfs2_end_log_read);
 558                         bio->bi_opf = REQ_OP_READ;
 559 add_block_to_new_bio:
 560                         sz = bio_add_page(bio, page, bsize, off);
 561                         BUG_ON(sz != bsize);
 562 block_added:
 563                         off += bsize;
 564                         if (off == PAGE_SIZE)
 565                                 page = NULL;
 566                         if (blocks_submitted <= blocks_read + max_blocks) {
 567                                 /* Keep at least one bio in flight */
 568                                 continue;
 569                         }
 570 
 571                         gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done);
 572                         blocks_read += PAGE_SIZE >> bsize_shift;
 573                         if (done)
 574                                 goto out;  /* found */
 575                 }
 576         }
 577 
 578 out:
 579         if (bio)
 580                 submit_bio(bio);
 581         while (blocks_read < block) {
 582                 gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done);
 583                 blocks_read += PAGE_SIZE >> bsize_shift;
 584         }
 585 
 586         if (!ret)
 587                 ret = filemap_check_wb_err(mapping, since);
 588 
 589         if (!keep_cache)
 590                 truncate_inode_pages(mapping, 0);
 591 
 592         return ret;
 593 }
 594 
 595 static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
 596                                       u32 ld_length, u32 ld_data1)
 597 {
 598         struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 599         struct gfs2_log_descriptor *ld = page_address(page);
 600         clear_page(ld);
 601         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 602         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 603         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 604         ld->ld_type = cpu_to_be32(ld_type);
 605         ld->ld_length = cpu_to_be32(ld_length);
 606         ld->ld_data1 = cpu_to_be32(ld_data1);
 607         ld->ld_data2 = 0;
 608         return page;
 609 }
 610 
 611 static void gfs2_check_magic(struct buffer_head *bh)
 612 {
 613         void *kaddr;
 614         __be32 *ptr;
 615 
 616         clear_buffer_escaped(bh);
 617         kaddr = kmap_atomic(bh->b_page);
 618         ptr = kaddr + bh_offset(bh);
 619         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 620                 set_buffer_escaped(bh);
 621         kunmap_atomic(kaddr);
 622 }
 623 
 624 static int blocknr_cmp(void *priv, struct list_head *a, struct list_head *b)
 625 {
 626         struct gfs2_bufdata *bda, *bdb;
 627 
 628         bda = list_entry(a, struct gfs2_bufdata, bd_list);
 629         bdb = list_entry(b, struct gfs2_bufdata, bd_list);
 630 
 631         if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
 632                 return -1;
 633         if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
 634                 return 1;
 635         return 0;
 636 }
 637 
 638 static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
 639                                 unsigned int total, struct list_head *blist,
 640                                 bool is_databuf)
 641 {
 642         struct gfs2_log_descriptor *ld;
 643         struct gfs2_bufdata *bd1 = NULL, *bd2;
 644         struct page *page;
 645         unsigned int num;
 646         unsigned n;
 647         __be64 *ptr;
 648 
 649         gfs2_log_lock(sdp);
 650         list_sort(NULL, blist, blocknr_cmp);
 651         bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
 652         while(total) {
 653                 num = total;
 654                 if (total > limit)
 655                         num = limit;
 656                 gfs2_log_unlock(sdp);
 657                 page = gfs2_get_log_desc(sdp,
 658                                          is_databuf ? GFS2_LOG_DESC_JDATA :
 659                                          GFS2_LOG_DESC_METADATA, num + 1, num);
 660                 ld = page_address(page);
 661                 gfs2_log_lock(sdp);
 662                 ptr = (__be64 *)(ld + 1);
 663 
 664                 n = 0;
 665                 list_for_each_entry_continue(bd1, blist, bd_list) {
 666                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 667                         if (is_databuf) {
 668                                 gfs2_check_magic(bd1->bd_bh);
 669                                 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
 670                         }
 671                         if (++n >= num)
 672                                 break;
 673                 }
 674 
 675                 gfs2_log_unlock(sdp);
 676                 gfs2_log_write_page(sdp, page);
 677                 gfs2_log_lock(sdp);
 678 
 679                 n = 0;
 680                 list_for_each_entry_continue(bd2, blist, bd_list) {
 681                         get_bh(bd2->bd_bh);
 682                         gfs2_log_unlock(sdp);
 683                         lock_buffer(bd2->bd_bh);
 684 
 685                         if (buffer_escaped(bd2->bd_bh)) {
 686                                 void *kaddr;
 687                                 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 688                                 ptr = page_address(page);
 689                                 kaddr = kmap_atomic(bd2->bd_bh->b_page);
 690                                 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh),
 691                                        bd2->bd_bh->b_size);
 692                                 kunmap_atomic(kaddr);
 693                                 *(__be32 *)ptr = 0;
 694                                 clear_buffer_escaped(bd2->bd_bh);
 695                                 unlock_buffer(bd2->bd_bh);
 696                                 brelse(bd2->bd_bh);
 697                                 gfs2_log_write_page(sdp, page);
 698                         } else {
 699                                 gfs2_log_write_bh(sdp, bd2->bd_bh);
 700                         }
 701                         gfs2_log_lock(sdp);
 702                         if (++n >= num)
 703                                 break;
 704                 }
 705 
 706                 BUG_ON(total < num);
 707                 total -= num;
 708         }
 709         gfs2_log_unlock(sdp);
 710 }
 711 
 712 static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 713 {
 714         unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
 715         unsigned int nbuf;
 716         if (tr == NULL)
 717                 return;
 718         nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
 719         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0);
 720 }
 721 
 722 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 723 {
 724         struct list_head *head;
 725         struct gfs2_bufdata *bd;
 726 
 727         if (tr == NULL)
 728                 return;
 729 
 730         head = &tr->tr_buf;
 731         while (!list_empty(head)) {
 732                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 733                 list_del_init(&bd->bd_list);
 734                 gfs2_unpin(sdp, bd->bd_bh, tr);
 735         }
 736 }
 737 
 738 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 739                                struct gfs2_log_header_host *head, int pass)
 740 {
 741         if (pass != 0)
 742                 return;
 743 
 744         jd->jd_found_blocks = 0;
 745         jd->jd_replayed_blocks = 0;
 746 }
 747 
 748 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
 749                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 750                                 int pass)
 751 {
 752         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 753         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 754         struct gfs2_glock *gl = ip->i_gl;
 755         unsigned int blks = be32_to_cpu(ld->ld_data1);
 756         struct buffer_head *bh_log, *bh_ip;
 757         u64 blkno;
 758         int error = 0;
 759 
 760         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 761                 return 0;
 762 
 763         gfs2_replay_incr_blk(jd, &start);
 764 
 765         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 766                 blkno = be64_to_cpu(*ptr++);
 767 
 768                 jd->jd_found_blocks++;
 769 
 770                 if (gfs2_revoke_check(jd, blkno, start))
 771                         continue;
 772 
 773                 error = gfs2_replay_read_block(jd, start, &bh_log);
 774                 if (error)
 775                         return error;
 776 
 777                 bh_ip = gfs2_meta_new(gl, blkno);
 778                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 779 
 780                 if (gfs2_meta_check(sdp, bh_ip))
 781                         error = -EIO;
 782                 else {
 783                         struct gfs2_meta_header *mh =
 784                                 (struct gfs2_meta_header *)bh_ip->b_data;
 785 
 786                         if (mh->mh_type == cpu_to_be32(GFS2_METATYPE_RG)) {
 787                                 struct gfs2_rgrpd *rgd;
 788 
 789                                 rgd = gfs2_blk2rgrpd(sdp, blkno, false);
 790                                 if (rgd && rgd->rd_addr == blkno &&
 791                                     rgd->rd_bits && rgd->rd_bits->bi_bh) {
 792                                         fs_info(sdp, "Replaying 0x%llx but we "
 793                                                 "already have a bh!\n",
 794                                                 (unsigned long long)blkno);
 795                                         fs_info(sdp, "busy:%d, pinned:%d\n",
 796                                                 buffer_busy(rgd->rd_bits->bi_bh) ? 1 : 0,
 797                                                 buffer_pinned(rgd->rd_bits->bi_bh));
 798                                         gfs2_dump_glock(NULL, rgd->rd_gl, true);
 799                                 }
 800                         }
 801                         mark_buffer_dirty(bh_ip);
 802                 }
 803                 brelse(bh_log);
 804                 brelse(bh_ip);
 805 
 806                 if (error)
 807                         break;
 808 
 809                 jd->jd_replayed_blocks++;
 810         }
 811 
 812         return error;
 813 }
 814 
 815 /**
 816  * gfs2_meta_sync - Sync all buffers associated with a glock
 817  * @gl: The glock
 818  *
 819  */
 820 
 821 static void gfs2_meta_sync(struct gfs2_glock *gl)
 822 {
 823         struct address_space *mapping = gfs2_glock2aspace(gl);
 824         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
 825         int error;
 826 
 827         if (mapping == NULL)
 828                 mapping = &sdp->sd_aspace;
 829 
 830         filemap_fdatawrite(mapping);
 831         error = filemap_fdatawait(mapping);
 832 
 833         if (error)
 834                 gfs2_io_error(gl->gl_name.ln_sbd);
 835 }
 836 
 837 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 838 {
 839         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 840         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 841 
 842         if (error) {
 843                 gfs2_meta_sync(ip->i_gl);
 844                 return;
 845         }
 846         if (pass != 1)
 847                 return;
 848 
 849         gfs2_meta_sync(ip->i_gl);
 850 
 851         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 852                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
 853 }
 854 
 855 static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 856 {
 857         struct gfs2_meta_header *mh;
 858         unsigned int offset;
 859         struct list_head *head = &sdp->sd_log_revokes;
 860         struct gfs2_bufdata *bd;
 861         struct page *page;
 862         unsigned int length;
 863 
 864         gfs2_write_revokes(sdp);
 865         if (!sdp->sd_log_num_revoke)
 866                 return;
 867 
 868         length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(u64));
 869         page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
 870         offset = sizeof(struct gfs2_log_descriptor);
 871 
 872         list_for_each_entry(bd, head, bd_list) {
 873                 sdp->sd_log_num_revoke--;
 874 
 875                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 876 
 877                         gfs2_log_write_page(sdp, page);
 878                         page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 879                         mh = page_address(page);
 880                         clear_page(mh);
 881                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 882                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 883                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 884                         offset = sizeof(struct gfs2_meta_header);
 885                 }
 886 
 887                 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
 888                 offset += sizeof(u64);
 889         }
 890         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 891 
 892         gfs2_log_write_page(sdp, page);
 893 }
 894 
 895 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 896 {
 897         struct list_head *head = &sdp->sd_log_revokes;
 898         struct gfs2_bufdata *bd;
 899         struct gfs2_glock *gl;
 900 
 901         while (!list_empty(head)) {
 902                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 903                 list_del_init(&bd->bd_list);
 904                 gl = bd->bd_gl;
 905                 gfs2_glock_remove_revoke(gl);
 906                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 907         }
 908 }
 909 
 910 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 911                                   struct gfs2_log_header_host *head, int pass)
 912 {
 913         if (pass != 0)
 914                 return;
 915 
 916         jd->jd_found_revokes = 0;
 917         jd->jd_replay_tail = head->lh_tail;
 918 }
 919 
 920 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
 921                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 922                                    int pass)
 923 {
 924         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 925         unsigned int blks = be32_to_cpu(ld->ld_length);
 926         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 927         struct buffer_head *bh;
 928         unsigned int offset;
 929         u64 blkno;
 930         int first = 1;
 931         int error;
 932 
 933         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 934                 return 0;
 935 
 936         offset = sizeof(struct gfs2_log_descriptor);
 937 
 938         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
 939                 error = gfs2_replay_read_block(jd, start, &bh);
 940                 if (error)
 941                         return error;
 942 
 943                 if (!first)
 944                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 945 
 946                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 947                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 948 
 949                         error = gfs2_revoke_add(jd, blkno, start);
 950                         if (error < 0) {
 951                                 brelse(bh);
 952                                 return error;
 953                         }
 954                         else if (error)
 955                                 jd->jd_found_revokes++;
 956 
 957                         if (!--revokes)
 958                                 break;
 959                         offset += sizeof(u64);
 960                 }
 961 
 962                 brelse(bh);
 963                 offset = sizeof(struct gfs2_meta_header);
 964                 first = 0;
 965         }
 966 
 967         return 0;
 968 }
 969 
 970 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 971 {
 972         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 973 
 974         if (error) {
 975                 gfs2_revoke_clean(jd);
 976                 return;
 977         }
 978         if (pass != 1)
 979                 return;
 980 
 981         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 982                 jd->jd_jid, jd->jd_found_revokes);
 983 
 984         gfs2_revoke_clean(jd);
 985 }
 986 
 987 /**
 988  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 989  *
 990  */
 991 
 992 static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 993 {
 994         unsigned int limit = databuf_limit(sdp);
 995         unsigned int nbuf;
 996         if (tr == NULL)
 997                 return;
 998         nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
 999         gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1);
1000 }
1001 
1002 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
1003                                     struct gfs2_log_descriptor *ld,
1004                                     __be64 *ptr, int pass)
1005 {
1006         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1007         struct gfs2_glock *gl = ip->i_gl;
1008         unsigned int blks = be32_to_cpu(ld->ld_data1);
1009         struct buffer_head *bh_log, *bh_ip;
1010         u64 blkno;
1011         u64 esc;
1012         int error = 0;
1013 
1014         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
1015                 return 0;
1016 
1017         gfs2_replay_incr_blk(jd, &start);
1018         for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
1019                 blkno = be64_to_cpu(*ptr++);
1020                 esc = be64_to_cpu(*ptr++);
1021 
1022                 jd->jd_found_blocks++;
1023 
1024                 if (gfs2_revoke_check(jd, blkno, start))
1025                         continue;
1026 
1027                 error = gfs2_replay_read_block(jd, start, &bh_log);
1028                 if (error)
1029                         return error;
1030 
1031                 bh_ip = gfs2_meta_new(gl, blkno);
1032                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
1033 
1034                 /* Unescape */
1035                 if (esc) {
1036                         __be32 *eptr = (__be32 *)bh_ip->b_data;
1037                         *eptr = cpu_to_be32(GFS2_MAGIC);
1038                 }
1039                 mark_buffer_dirty(bh_ip);
1040 
1041                 brelse(bh_log);
1042                 brelse(bh_ip);
1043 
1044                 jd->jd_replayed_blocks++;
1045         }
1046 
1047         return error;
1048 }
1049 
1050 /* FIXME: sort out accounting for log blocks etc. */
1051 
1052 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
1053 {
1054         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1055         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
1056 
1057         if (error) {
1058                 gfs2_meta_sync(ip->i_gl);
1059                 return;
1060         }
1061         if (pass != 1)
1062                 return;
1063 
1064         /* data sync? */
1065         gfs2_meta_sync(ip->i_gl);
1066 
1067         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
1068                 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
1069 }
1070 
1071 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1072 {
1073         struct list_head *head;
1074         struct gfs2_bufdata *bd;
1075 
1076         if (tr == NULL)
1077                 return;
1078 
1079         head = &tr->tr_databuf;
1080         while (!list_empty(head)) {
1081                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
1082                 list_del_init(&bd->bd_list);
1083                 gfs2_unpin(sdp, bd->bd_bh, tr);
1084         }
1085 }
1086 
1087 
1088 static const struct gfs2_log_operations gfs2_buf_lops = {
1089         .lo_before_commit = buf_lo_before_commit,
1090         .lo_after_commit = buf_lo_after_commit,
1091         .lo_before_scan = buf_lo_before_scan,
1092         .lo_scan_elements = buf_lo_scan_elements,
1093         .lo_after_scan = buf_lo_after_scan,
1094         .lo_name = "buf",
1095 };
1096 
1097 static const struct gfs2_log_operations gfs2_revoke_lops = {
1098         .lo_before_commit = revoke_lo_before_commit,
1099         .lo_after_commit = revoke_lo_after_commit,
1100         .lo_before_scan = revoke_lo_before_scan,
1101         .lo_scan_elements = revoke_lo_scan_elements,
1102         .lo_after_scan = revoke_lo_after_scan,
1103         .lo_name = "revoke",
1104 };
1105 
1106 static const struct gfs2_log_operations gfs2_databuf_lops = {
1107         .lo_before_commit = databuf_lo_before_commit,
1108         .lo_after_commit = databuf_lo_after_commit,
1109         .lo_scan_elements = databuf_lo_scan_elements,
1110         .lo_after_scan = databuf_lo_after_scan,
1111         .lo_name = "databuf",
1112 };
1113 
1114 const struct gfs2_log_operations *gfs2_log_ops[] = {
1115         &gfs2_databuf_lops,
1116         &gfs2_buf_lops,
1117         &gfs2_revoke_lops,
1118         NULL,
1119 };
1120
/* [<][>][^][v][top][bottom][index][help] */
root/fs/gfs2/lops.c

DEFINITIONS
