* Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2014, Intel Corporation.
+ * Copyright (c) 2012, 2015, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
/* the check is outside of the cycle for performance reason -bzzz */
if (!test_bit(__REQ_WRITE, &bio->bi_rw)) {
- bio_for_each_segment(bvl, bio, iter) {
+ bio_for_each_segment_all(bvl, bio, iter) {
if (likely(error == 0))
- SetPageUptodate(bvec_iter_page(&bvl, iter));
- LASSERT(PageLocked(bvec_iter_page(&bvl, iter)));
+ SetPageUptodate(bvl_to_page(bvl));
+ LASSERT(PageLocked(bvl_to_page(bvl)));
}
atomic_dec(&iobuf->dr_dev->od_r_in_flight);
} else {
struct osd_iobuf *iobuf)
{
int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits;
- struct page **pages = iobuf->dr_pages;
- int npages = iobuf->dr_npages;
- unsigned long *blocks = iobuf->dr_blocks;
- int total_blocks = npages * blocks_per_page;
- int sector_bits = inode->i_sb->s_blocksize_bits - 9;
- unsigned int blocksize = inode->i_sb->s_blocksize;
- struct bio *bio = NULL;
- struct page *page;
- unsigned int page_offset;
- sector_t sector;
- int nblocks;
- int block_idx;
- int page_idx;
- int i;
- int rc = 0;
- ENTRY;
+ struct page **pages = iobuf->dr_pages;
+ int npages = iobuf->dr_npages;
+ sector_t *blocks = iobuf->dr_blocks;
+ int total_blocks = npages * blocks_per_page;
+ int sector_bits = inode->i_sb->s_blocksize_bits - 9;
+ unsigned int blocksize = inode->i_sb->s_blocksize;
+ struct bio *bio = NULL;
+ struct page *page;
+ unsigned int page_offset;
+ sector_t sector;
+ int nblocks;
+ int block_idx;
+ int page_idx;
+ int i;
+ int rc = 0;
+ ENTRY;
LASSERT(iobuf->dr_npages == npages);
#endif
struct bpointers {
- unsigned long *blocks;
+ sector_t *blocks;
unsigned long start;
int num;
int init_num;
{
struct bpointers *bp = cbdata;
struct ldiskfs_extent nex;
- unsigned long pblock;
+ unsigned long pblock = 0;
unsigned long tgen;
int err, i;
unsigned long count;
i, cex->ec_len);
for (; i < cex->ec_len && bp->num; i++) {
*(bp->blocks) = cex->ec_start + i;
-#ifdef LDISKFS_EXT_CACHE_EXTENT /* until kernel 2.6.37 */
- if (cex->ec_type != LDISKFS_EXT_CACHE_EXTENT) {
-#else
- if ((cex->ec_len == 0) || (cex->ec_start == 0)) {
-#endif
+ if (pblock != 0) {
/* unmap any possible underlying metadata from
* the block device mapping. bug 6998. */
unmap_underlying_metadata(inode->i_sb->s_bdev,
return err;
}
-static int osd_ldiskfs_map_nblocks(struct inode *inode, unsigned long block,
- unsigned long num, unsigned long *blocks,
- int create)
+static int osd_ldiskfs_map_nblocks(struct inode *inode, unsigned long index,
+ int clen, sector_t *blocks, int create)
{
+ int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits;
struct bpointers bp;
int err;
- CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
- block, block + num - 1, (unsigned) inode->i_ino);
+ if (index + clen >= inode->i_sb->s_maxbytes >> PAGE_SHIFT)
+ return -EFBIG;
bp.blocks = blocks;
- bp.start = block;
- bp.init_num = bp.num = num;
+ bp.start = index * blocks_per_page;
+ bp.init_num = bp.num = clen * blocks_per_page;
bp.create = create;
- err = ldiskfs_ext_walk_space(inode, block, num,
- ldiskfs_ext_new_extent_cb, &bp);
+ CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
+ bp.start, bp.start + bp.num - 1, (unsigned)inode->i_ino);
+
+ err = ldiskfs_ext_walk_space(inode, bp.start, bp.num,
+ ldiskfs_ext_new_extent_cb, &bp);
ldiskfs_ext_invalidate_cache(inode);
return err;
static int osd_ldiskfs_map_bm_inode_pages(struct inode *inode,
struct page **page, int pages,
- unsigned long *blocks, int create)
+ sector_t *blocks, int create)
{
int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits;
pgoff_t bitmap_max_page_index;
- unsigned long *b;
+ sector_t *b;
int rc = 0, i;
bitmap_max_page_index = LDISKFS_SB(inode->i_sb)->s_bitmap_maxbytes >>
}
rc = ldiskfs_map_inode_page(inode, *page, b, create);
if (rc) {
- CERROR("ino %lu, blk %lu create %d: rc %d\n",
- inode->i_ino, *b, create, rc);
+ CERROR("ino %lu, blk %llu create %d: rc %d\n",
+ inode->i_ino,
+ (unsigned long long)*b, create, rc);
break;
}
b += blocks_per_page;
static int osd_ldiskfs_map_ext_inode_pages(struct inode *inode,
struct page **page,
- int pages, unsigned long *blocks,
+ int pages, sector_t *blocks,
int create)
{
- int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits;
- int rc = 0, i = 0;
+ int rc = 0, i = 0, clen = 0;
struct page *fp = NULL;
- int clen = 0;
- pgoff_t extent_max_page_index;
-
- extent_max_page_index = inode->i_sb->s_maxbytes >> PAGE_SHIFT;
CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
inode->i_ino, pages, (*page)->index);
continue;
}
- if (fp->index + i >= extent_max_page_index)
- GOTO(cleanup, rc = -EFBIG);
-
/* process found extent */
- rc = osd_ldiskfs_map_nblocks(inode, fp->index * blocks_per_page,
- clen * blocks_per_page, blocks,
- create);
+ rc = osd_ldiskfs_map_nblocks(inode, fp->index, clen,
+ blocks, create);
if (rc)
GOTO(cleanup, rc);
/* look for next extent */
fp = NULL;
- blocks += blocks_per_page * clen;
+ blocks += clen * (PAGE_CACHE_SIZE >> inode->i_blkbits);
}
if (fp)
- rc = osd_ldiskfs_map_nblocks(inode, fp->index * blocks_per_page,
- clen * blocks_per_page, blocks,
- create);
+ rc = osd_ldiskfs_map_nblocks(inode, fp->index, clen,
+ blocks, create);
+
cleanup:
return rc;
}
static int osd_ldiskfs_map_inode_pages(struct inode *inode, struct page **page,
- int pages, unsigned long *blocks,
+ int pages, sector_t *blocks,
int create)
{
int rc;
}
#else
static int osd_ldiskfs_map_inode_pages(struct inode *inode, struct page **page,
- int pages, unsigned long *blocks,
+ int pages, sector_t *blocks,
int create)
{
int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits;
if (++i != pages)
continue;
}
- if (fp->index + i >= max_page_index)
+ if (fp->index + clen >= max_page_index)
GOTO(cleanup, rc = -EFBIG);
/* process found extent */
map.m_lblk = fp->index * blocks_per_page;
RETURN(rc);
}
-/* Check if a block is allocated or not */
-static int osd_is_mapped(struct inode *inode, u64 offset)
-{
- sector_t (*fs_bmap)(struct address_space *, sector_t);
+struct osd_fextent {
+ sector_t start;
+ sector_t end;
+ unsigned int mapped:1;
+};
- fs_bmap = inode->i_mapping->a_ops->bmap;
+static int osd_is_mapped(struct dt_object *dt, __u64 offset,
+ struct osd_fextent *cached_extent)
+{
+ struct inode *inode = osd_dt_obj(dt)->oo_inode;
+ sector_t block = offset >> inode->i_blkbits;
+ sector_t start;
+ struct fiemap_extent_info fei = { 0 };
+ struct fiemap_extent fe = { 0 };
+ mm_segment_t saved_fs;
+ int rc;
- /* We can't know if we are overwriting or not */
- if (unlikely(fs_bmap == NULL))
- return 0;
+ if (block >= cached_extent->start && block < cached_extent->end)
+ return cached_extent->mapped;
if (i_size_read(inode) == 0)
return 0;
/* Beyond EOF, must not be mapped */
- if (((i_size_read(inode) - 1) >> inode->i_blkbits) <
- (offset >> inode->i_blkbits))
+ if (((i_size_read(inode) - 1) >> inode->i_blkbits) < block)
return 0;
- if (fs_bmap(inode->i_mapping, offset >> inode->i_blkbits) == 0)
+ fei.fi_extents_max = 1;
+ fei.fi_extents_start = &fe;
+
+ saved_fs = get_fs();
+ set_fs(get_ds());
+ rc = inode->i_op->fiemap(inode, &fei, offset, FIEMAP_MAX_OFFSET-offset);
+ set_fs(saved_fs);
+ if (rc != 0)
return 0;
- return 1;
+ start = fe.fe_logical >> inode->i_blkbits;
+
+ if (start > block) {
+ cached_extent->start = block;
+ cached_extent->end = start;
+ cached_extent->mapped = 0;
+ } else {
+ cached_extent->start = start;
+ cached_extent->end = (fe.fe_logical + fe.fe_length) >>
+ inode->i_blkbits;
+ cached_extent->mapped = 1;
+ }
+
+ return cached_extent->mapped;
}
static int osd_declare_write_commit(const struct lu_env *env,
int newblocks;
int rc = 0;
int flags = 0;
+ int credits = 0;
bool ignore_quota = false;
long long quota_space = 0;
+ struct osd_fextent extent = { 0 };
ENTRY;
LASSERT(handle != NULL);
lnb[i - 1].lnb_file_offset + lnb[i - 1].lnb_len)
extents++;
- if (!osd_is_mapped(inode, lnb[i].lnb_file_offset))
+ if (!osd_is_mapped(dt, lnb[i].lnb_file_offset, &extent))
quota_space += PAGE_CACHE_SIZE;
/* ignore quota for the whole request if any page is from
depth = ext_depth(inode);
depth = max(depth, 1) + 1;
newblocks += depth;
- oh->ot_credits++; /* inode */
- oh->ot_credits += depth * 2 * extents;
- } else {
- depth = 3;
- newblocks += depth;
- oh->ot_credits++; /* inode */
- oh->ot_credits += depth * extents;
- }
+ credits++; /* inode */
+ credits += depth * 2 * extents;
+ } else {
+ depth = 3;
+ newblocks += depth;
+ credits++; /* inode */
+ credits += depth * extents;
+ }
/* quota space for metadata blocks */
quota_space += depth * extents * LDISKFS_BLOCK_SIZE(osd_sb(osd));
/* we can't dirty more bitmap blocks than exist */
if (newblocks > LDISKFS_SB(osd_sb(osd))->s_groups_count)
- oh->ot_credits += LDISKFS_SB(osd_sb(osd))->s_groups_count;
+ credits += LDISKFS_SB(osd_sb(osd))->s_groups_count;
else
- oh->ot_credits += newblocks;
+ credits += newblocks;
- /* we can't dirty more gd blocks than exist */
- if (newblocks > LDISKFS_SB(osd_sb(osd))->s_gdb_count)
- oh->ot_credits += LDISKFS_SB(osd_sb(osd))->s_gdb_count;
- else
- oh->ot_credits += newblocks;
+ /* we can't dirty more gd blocks than exist */
+ if (newblocks > LDISKFS_SB(osd_sb(osd))->s_gdb_count)
+ credits += LDISKFS_SB(osd_sb(osd))->s_gdb_count;
+ else
+ credits += newblocks;
+
+ osd_trans_declare_op(env, oh, OSD_OT_WRITE, credits);
/* make sure the over quota flags were not set */
lnb[0].lnb_flags &= ~(OBD_BRW_OVER_USRQUOTA | OBD_BRW_OVER_GRPQUOTA);
struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
loff_t isize;
int rc = 0, i;
+ struct osd_fextent extent = { 0 };
LASSERT(inode);
for (i = 0; i < npages; i++) {
if (lnb[i].lnb_rc == -ENOSPC &&
- osd_is_mapped(inode, lnb[i].lnb_file_offset)) {
+ osd_is_mapped(dt, lnb[i].lnb_file_offset, &extent)) {
/* Allow the write to proceed if overwriting an
* existing block */
lnb[i].lnb_rc = 0;
osd_iobuf_add_page(iobuf, lnb[i].lnb_page);
}
+ osd_trans_exec_op(env, thandle, OSD_OT_WRITE);
+
if (OBD_FAIL_CHECK(OBD_FAIL_OST_MAPBLK_ENOSPC)) {
rc = -ENOSPC;
} else if (iobuf->dr_npages > 0) {
osd_fini_iobuf(osd, iobuf);
}
- if (unlikely(rc != 0)) {
- /* if write fails, we should drop pages from the cache */
- for (i = 0; i < npages; i++) {
+ osd_trans_exec_check(env, thandle, OSD_OT_WRITE);
+
+ if (unlikely(rc != 0)) {
+ /* if write fails, we should drop pages from the cache */
+ for (i = 0; i < npages; i++) {
if (lnb[i].lnb_page == NULL)
continue;
LASSERT(PageLocked(lnb[i].lnb_page));
* lnb->lnb_rc == 0, so it's easy to detect later. */
break;
- if (isize < lnb[i].lnb_file_offset + lnb[i].lnb_len - 1)
+ if (isize < lnb[i].lnb_file_offset + lnb[i].lnb_len)
lnb[i].lnb_rc = isize - lnb[i].lnb_file_offset;
else
lnb[i].lnb_rc = lnb[i].lnb_len;
int blocksize;
int csize;
int boffs;
- int err;
+ int err = 0;
/* prevent reading after eof */
spin_lock(&inode->i_lock);
return 0;
}
-static inline int osd_calc_bkmap_credits(struct super_block *sb,
- struct inode *inode,
- const loff_t size,
- const loff_t pos,
- const int blocks)
+int osd_calc_bkmap_credits(struct super_block *sb, struct inode *inode,
+ const loff_t size, const loff_t pos,
+ const int blocks)
{
int credits, bits, bs, i;
} else if (pos + size <= (LDISKFS_NDIR_BLOCKS + 1024) * bs) {
/* single indirect */
credits = blocks * 3;
- /* probably indirect block has been allocated already */
- if (!inode || LDISKFS_I(inode)->i_data[LDISKFS_IND_BLOCK])
+ if (inode == NULL ||
+ LDISKFS_I(inode)->i_data[LDISKFS_IND_BLOCK] == 0)
credits += 3;
+ else
+ /* The indirect block may be modified. */
+ credits += 1;
}
return credits;
size = min(blocksize - boffs, bufsize);
bh = ldiskfs_bread(handle, inode, block, 1, &err);
if (!bh) {
+ err = err ? err : -EIO;
CERROR("%s: error reading offset %llu (block %lu): "
"rc = %d\n",
inode->i_sb->s_id, offset, block, err);
oh = container_of(handle, struct osd_thandle, ot_super);
LASSERT(oh->ot_handle->h_transaction != NULL);
+ osd_trans_exec_op(env, handle, OSD_OT_WRITE);
+
/* Write small symlink to inode body as we need to maintain correct
* on-disk symlinks for ldiskfs.
* Note: the buf->lb_buf contains a NUL terminator while buf->lb_len
result = osd_ldiskfs_write_record(inode, buf->lb_buf,
buf->lb_len, is_link, pos,
oh->ot_handle);
- if (result == 0)
- result = buf->lb_len;
- return result;
+ if (result == 0)
+ result = buf->lb_len;
+
+ osd_trans_exec_check(env, handle, OSD_OT_WRITE);
+
+ return result;
}
static int osd_declare_punch(const struct lu_env *env, struct dt_object *dt,
LASSERT(h != NULL);
LASSERT(h == oh->ot_handle);
+ /* do not check credits with osd_trans_exec_check() as the truncate
+ * can restart the transaction internally and we restart the
+ * transaction in this case */
+
if (tid != h->h_transaction->t_tid) {
int credits = oh->ot_credits;
/*
#define FIEMAP_MAX_EXTENTS (UINT_MAX / sizeof(struct fiemap_extent))
static int osd_fiemap_get(const struct lu_env *env, struct dt_object *dt,
- struct ll_user_fiemap *fm)
+ struct fiemap *fm)
{
struct fiemap_extent_info fieinfo = {0, };
struct inode *inode = osd_dt_obj(dt)->oo_inode;
return rc;
}
+static int osd_ladvise(const struct lu_env *env, struct dt_object *dt,
+ __u64 start, __u64 end, enum lu_ladvise_type advice)
+{
+ int rc;
+ ENTRY;
+
+ switch (advice) {
+ default:
+ rc = -ENOTSUPP;
+ break;
+ }
+
+ RETURN(rc);
+}
+
/*
* in some cases we may need declare methods for objects being created
* e.g., when we create symlink
};
const struct dt_body_operations osd_body_ops = {
- .dbo_read = osd_read,
- .dbo_declare_write = osd_declare_write,
- .dbo_write = osd_write,
- .dbo_bufs_get = osd_bufs_get,
- .dbo_bufs_put = osd_bufs_put,
- .dbo_write_prep = osd_write_prep,
- .dbo_declare_write_commit = osd_declare_write_commit,
- .dbo_write_commit = osd_write_commit,
- .dbo_read_prep = osd_read_prep,
- .dbo_declare_punch = osd_declare_punch,
- .dbo_punch = osd_punch,
- .dbo_fiemap_get = osd_fiemap_get,
+ .dbo_read = osd_read,
+ .dbo_declare_write = osd_declare_write,
+ .dbo_write = osd_write,
+ .dbo_bufs_get = osd_bufs_get,
+ .dbo_bufs_put = osd_bufs_put,
+ .dbo_write_prep = osd_write_prep,
+ .dbo_declare_write_commit = osd_declare_write_commit,
+ .dbo_write_commit = osd_write_commit,
+ .dbo_read_prep = osd_read_prep,
+ .dbo_declare_punch = osd_declare_punch,
+ .dbo_punch = osd_punch,
+ .dbo_fiemap_get = osd_fiemap_get,
+ .dbo_ladvise = osd_ladvise,
};
git://git.whamcloud.com / fs / lustre-release.git / blobdiff