On filesystems with a block size smaller that the page size, a page can be split across bios. When that happens, the first bio that completes will unlock the page even though it hasn't been read completely yet. Fix that case by using a small chained bio.
While at it, clean up the sector calculation logic in gfs2_log_alloc_bio. In gfs2_find_jhead, simplify the disk block and offset calculation logic and fix a variable name.
Fixes: f4686c26ecc3 ("gfs2: read journal in large chunks") Cc: stable@vger.kernel.org # v5.2+ Signed-off-by: Andreas Gruenbacher agruenba@redhat.com --- fs/gfs2/lops.c | 54 ++++++++++++++++++++++++++++++-------------------- 1 file changed, 33 insertions(+), 21 deletions(-)
diff --git a/fs/gfs2/lops.c b/fs/gfs2/lops.c index 55fed7daf2b1..c817dee337c9 100644 --- a/fs/gfs2/lops.c +++ b/fs/gfs2/lops.c @@ -259,7 +259,7 @@ static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno, struct super_block *sb = sdp->sd_vfs; struct bio *bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
- bio->bi_iter.bi_sector = blkno * (sb->s_blocksize >> 9); + bio->bi_iter.bi_sector = blkno << (sb->s_blocksize_bits - 9); bio_set_dev(bio, sb->s_bdev); bio->bi_end_io = end_io; bio->bi_private = sdp; @@ -472,6 +472,21 @@ static void gfs2_jhead_process_page(struct gfs2_jdesc *jd, unsigned long index, put_page(page); /* Once more for find_or_create_page */ }
+static struct bio *gfs2_chain_bio(struct bio *prev) +{ + struct bio *new; + + /* one iovec only to avoid chaining the chained bio */ + new = bio_alloc(GFP_NOIO, 1); + bio_copy_dev(new, prev); + new->bi_iter.bi_sector = bio_end_sector(prev); + new->bi_opf = prev->bi_opf; + new->bi_write_hint = prev->bi_write_hint; + bio_chain(new, prev); + submit_bio(prev); + return new; +} + /** * gfs2_find_jhead - find the head of a log * @jd: The journal descriptor @@ -488,10 +503,10 @@ int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head, struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); struct address_space *mapping = jd->jd_inode->i_mapping; unsigned int block = 0, blocks_submitted = 0, blocks_read = 0; - unsigned int bsize = sdp->sd_sb.sb_bsize; + unsigned int bsize = sdp->sd_sb.sb_bsize, off; unsigned int bsize_shift = sdp->sd_sb.sb_bsize_shift; unsigned int shift = PAGE_SHIFT - bsize_shift; - unsigned int readhead_blocks = BIO_MAX_PAGES << shift; + unsigned int readahead_blocks = BIO_MAX_PAGES << shift; struct gfs2_journal_extent *je; int sz, ret = 0; struct bio *bio = NULL; @@ -505,9 +520,9 @@ int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head,
since = filemap_sample_wb_err(mapping); list_for_each_entry(je, &jd->extent_list, list) { - for (; block < je->lblock + je->blocks; block++) { - u64 dblock; + u64 dblock = je->dblock;
+ for (; block < je->lblock + je->blocks; block++, dblock++) { if (!page) { page = find_or_create_page(mapping, block >> shift, GFP_NOFS); @@ -516,35 +531,32 @@ int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head, done = true; goto out; } + off = 0; }
if (bio) { - unsigned int off; - - off = (block << bsize_shift) & ~PAGE_MASK; sz = bio_add_page(bio, page, bsize, off); - if (sz == bsize) { /* block added */ - if (off + bsize == PAGE_SIZE) { - page = NULL; - goto page_added; - } - continue; + if (sz == bsize) + goto block_added; + if (off) { + bio = gfs2_chain_bio(bio); + goto add_block_to_new_bio; } blocks_submitted = block + 1; submit_bio(bio); bio = NULL; }
- dblock = je->dblock + (block - je->lblock); bio = gfs2_log_alloc_bio(sdp, dblock, gfs2_end_log_read); bio->bi_opf = REQ_OP_READ; - sz = bio_add_page(bio, page, bsize, 0); - gfs2_assert_warn(sdp, sz == bsize); - if (bsize == PAGE_SIZE) +add_block_to_new_bio: + sz = bio_add_page(bio, page, bsize, off); + BUG_ON(sz != bsize); +block_added: + off += bsize; + if (off == PAGE_SIZE) page = NULL; - -page_added: - if (blocks_submitted < blocks_read + readhead_blocks) { + if (blocks_submitted < blocks_read + readahead_blocks) { /* Keep at least one bio in flight */ continue; }