*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
* Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2013, Intel Corporation.
+ * Copyright (c) 2012, 2016, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
iobuf->dr_rw = rw;
iobuf->dr_init_at = line;
- blocks = pages * (PAGE_CACHE_SIZE >> osd_sb(d)->s_blocksize_bits);
+ blocks = pages * (PAGE_SIZE >> osd_sb(d)->s_blocksize_bits);
if (iobuf->dr_bl_buf.lb_len >= blocks * sizeof(iobuf->dr_blocks[0])) {
LASSERT(iobuf->dr_pg_buf.lb_len >=
pages * sizeof(iobuf->dr_pages[0]));
CDEBUG(D_OTHER, "realloc %u for %u (%u) pages\n",
(unsigned)(pages * sizeof(iobuf->dr_pages[0])), i, pages);
pages = i;
- blocks = pages * (PAGE_CACHE_SIZE >> osd_sb(d)->s_blocksize_bits);
+ blocks = pages * (PAGE_SIZE >> osd_sb(d)->s_blocksize_bits);
iobuf->dr_max_pages = 0;
CDEBUG(D_OTHER, "realloc %u for %u blocks\n",
(unsigned)(blocks * sizeof(iobuf->dr_blocks[0])), blocks);
#define __REQ_WRITE BIO_RW
#endif
+#ifdef HAVE_BIO_ENDIO_USES_ONE_ARG
+static void dio_complete_routine(struct bio *bio)
+{
+ int error = bio->bi_error;
+#else
static void dio_complete_routine(struct bio *bio, int error)
{
+#endif
struct osd_iobuf *iobuf = bio->bi_private;
-#ifdef HAVE_BVEC_ITER
- struct bvec_iter iter;
- struct bio_vec bvl;
-#else
int iter;
struct bio_vec *bvl;
-#endif
/* CAVEAT EMPTOR: possibly in IRQ context
* DO NOT record procfs stats here!!! */
"IO - you will probably have to reboot this node.\n");
CERROR("bi_next: %p, bi_flags: %lx, bi_rw: %lu, bi_vcnt: %d, "
"bi_idx: %d, bi->size: %d, bi_end_io: %p, bi_cnt: %d, "
- "bi_private: %p\n", bio->bi_next, bio->bi_flags,
+ "bi_private: %p\n", bio->bi_next,
+ (unsigned long)bio->bi_flags,
bio->bi_rw, bio->bi_vcnt, bio_idx(bio),
bio_sectors(bio) << 9, bio->bi_end_io,
- atomic_read(&bio->bi_cnt), bio->bi_private);
+#ifdef HAVE_BI_CNT
+ atomic_read(&bio->bi_cnt),
+#else
+ atomic_read(&bio->__bi_cnt),
+#endif
+ bio->bi_private);
return;
}
/* 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 {
static int osd_do_bio(struct osd_device *osd, struct inode *inode,
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;
+ int blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
+ 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;
+ DECLARE_PLUG(plug);
+ ENTRY;
LASSERT(iobuf->dr_npages == npages);
osd_brw_stats_update(osd, iobuf);
iobuf->dr_start_time = cfs_time_current();
+ blk_start_plug(&plug);
for (page_idx = 0, block_idx = 0;
page_idx < npages;
page_idx++, block_idx += blocks_per_page) {
}
out:
+ blk_finish_plug(&plug);
+
/* in order to achieve better IO throughput, we don't wait for writes
* completion here. instead we proceed with transaction commit in
* parallel and wait for IO completion once transaction is stopped
*nrpages = 0;
while (len > 0) {
- int poff = offset & (PAGE_CACHE_SIZE - 1);
- int plen = PAGE_CACHE_SIZE - poff;
+ int poff = offset & (PAGE_SIZE - 1);
+ int plen = PAGE_SIZE - poff;
if (plen > len)
plen = len;
RETURN(0);
}
-static struct page *osd_get_page(struct dt_object *dt, loff_t offset, int rw)
+static struct page *osd_get_page(struct dt_object *dt, loff_t offset,
+ gfp_t gfp_mask)
{
- struct inode *inode = osd_dt_obj(dt)->oo_inode;
- struct osd_device *d = osd_obj2dev(osd_dt_obj(dt));
- struct page *page;
+ struct inode *inode = osd_dt_obj(dt)->oo_inode;
+ struct osd_device *d = osd_obj2dev(osd_dt_obj(dt));
+ struct page *page;
LASSERT(inode);
- page = find_or_create_page(inode->i_mapping, offset >> PAGE_CACHE_SHIFT,
- GFP_NOFS | __GFP_HIGHMEM);
+ page = find_or_create_page(inode->i_mapping, offset >> PAGE_SHIFT,
+ gfp_mask);
+
if (unlikely(page == NULL))
lprocfs_counter_add(d->od_stats, LPROC_OSD_NO_PAGE, 1);
continue;
LASSERT(PageLocked(lnb[i].lnb_page));
unlock_page(lnb[i].lnb_page);
- page_cache_release(lnb[i].lnb_page);
- lu_object_put(env, &dt->do_lu);
+ put_page(lnb[i].lnb_page);
+ dt_object_put(env, dt);
lnb[i].lnb_page = NULL;
}
* \param pos byte offset of IO start
* \param len number of bytes of IO
* \param lnb array of extents undergoing IO
- * \param rw read or write operation?
+ * \param rw read or write operation, and other flags
* \param capa capabilities
*
* \retval pages (zero or more) loaded successfully
*/
static int osd_bufs_get(const struct lu_env *env, struct dt_object *dt,
loff_t pos, ssize_t len, struct niobuf_local *lnb,
- int rw, struct lustre_capa *capa)
+ enum dt_bufs_type rw)
{
- struct osd_object *obj = osd_dt_obj(dt);
+ struct osd_object *obj = osd_dt_obj(dt);
int npages, i, rc = 0;
+ gfp_t gfp_mask;
LASSERT(obj->oo_inode);
osd_map_remote_to_local(pos, len, &npages, lnb);
+ /* this could also try less hard for DT_BUFS_TYPE_READAHEAD pages */
+ gfp_mask = rw & DT_BUFS_TYPE_LOCAL ? (GFP_NOFS | __GFP_HIGHMEM) :
+ GFP_HIGHUSER;
for (i = 0; i < npages; i++, lnb++) {
- lnb->lnb_page = osd_get_page(dt, lnb->lnb_file_offset, rw);
+ lnb->lnb_page = osd_get_page(dt, lnb->lnb_file_offset,
+ gfp_mask);
if (lnb->lnb_page == NULL)
GOTO(cleanup, rc = -ENOMEM);
#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_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;
- unsigned long *b;
+ int blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
+ pgoff_t bitmap_max_page_index;
+ sector_t *b;
int rc = 0, i;
+ bitmap_max_page_index = LDISKFS_SB(inode->i_sb)->s_bitmap_maxbytes >>
+ PAGE_SHIFT;
for (i = 0, b = blocks; i < pages; i++, page++) {
+ if ((*page)->index + 1 >= bitmap_max_page_index) {
+ rc = -EFBIG;
+ break;
+ }
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;
CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
inode->i_ino, pages, (*page)->index);
}
/* 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_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;
+ int blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
int rc = 0, i = 0;
struct page *fp = NULL;
int clen = 0;
+ pgoff_t max_page_index;
+ handle_t *handle = NULL;
+
+ 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);
+ if (create) {
+ create = LDISKFS_GET_BLOCKS_CREATE;
+ handle = ldiskfs_journal_current_handle();
+ LASSERT(handle != NULL);
+ rc = osd_attach_jinode(inode);
+ if (rc)
+ return rc;
+ }
/* pages are sorted already. so, we just have to find
* contig. space and process them properly */
while (i < pages) {
long blen, total = 0;
- handle_t *handle = NULL;
struct ldiskfs_map_blocks map = { 0 };
if (fp == NULL) { /* start new extent */
if (++i != pages)
continue;
}
+ if (fp->index + clen >= max_page_index)
+ GOTO(cleanup, rc = -EFBIG);
/* process found extent */
map.m_lblk = fp->index * blocks_per_page;
map.m_len = blen = clen * blocks_per_page;
- if (create) {
- create = LDISKFS_GET_BLOCKS_CREATE;
- handle = ldiskfs_journal_current_handle();
- LASSERT(handle != NULL);
- }
cont_map:
rc = ldiskfs_map_blocks(handle, inode, &map, create);
if (rc >= 0) {
struct osd_iobuf *iobuf = &oti->oti_iobuf;
struct inode *inode = osd_dt_obj(dt)->oo_inode;
struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
- struct timeval start;
- struct timeval end;
- unsigned long timediff;
+ ktime_t start;
+ ktime_t end;
+ s64 timediff;
ssize_t isize;
__s64 maxidx;
int rc = 0;
RETURN(rc);
isize = i_size_read(inode);
- maxidx = ((isize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) - 1;
+ maxidx = ((isize + PAGE_SIZE - 1) >> PAGE_SHIFT) - 1;
if (osd->od_writethrough_cache)
cache = 1;
if (isize > osd->od_readcache_max_filesize)
cache = 0;
- do_gettimeofday(&start);
+ start = ktime_get();
for (i = 0; i < npages; i++) {
if (cache == 0)
*/
ClearPageUptodate(lnb[i].lnb_page);
- if (lnb[i].lnb_len == PAGE_CACHE_SIZE)
+ if (lnb[i].lnb_len == PAGE_SIZE)
continue;
if (maxidx >= lnb[i].lnb_page->index) {
if (off)
memset(p, 0, off);
off = (lnb[i].lnb_page_offset + lnb[i].lnb_len) &
- ~CFS_PAGE_MASK;
+ ~PAGE_MASK;
if (off)
- memset(p + off, 0, PAGE_CACHE_SIZE - off);
+ memset(p + off, 0, PAGE_SIZE - off);
kunmap(lnb[i].lnb_page);
}
}
- do_gettimeofday(&end);
- timediff = cfs_timeval_sub(&end, &start, NULL);
+ end = ktime_get();
+ timediff = ktime_us_delta(end, start);
lprocfs_counter_add(osd->od_stats, LPROC_OSD_GET_PAGE, timediff);
if (iobuf->dr_npages) {
RETURN(rc);
}
-/* Check if a block is allocated or not */
-static int osd_is_mapped(struct inode *inode, obd_size 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,
struct niobuf_local *lnb, int npages,
struct thandle *handle)
{
- const struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
- struct inode *inode = osd_dt_obj(dt)->oo_inode;
- struct osd_thandle *oh;
- int extents = 1;
- int depth;
- int i;
- int newblocks;
- int rc = 0;
- int flags = 0;
- bool ignore_quota = false;
- long long quota_space = 0;
+ const struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
+ struct inode *inode = osd_dt_obj(dt)->oo_inode;
+ struct osd_thandle *oh;
+ int extents = 1;
+ int depth;
+ int i;
+ int newblocks;
+ int rc = 0;
+ int flags = 0;
+ int credits = 0;
+ long long quota_space = 0;
+ struct osd_fextent extent = { 0 };
+ enum osd_qid_declare_flags declare_flags = OSD_QID_BLK;
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))
- quota_space += PAGE_CACHE_SIZE;
+ if (!osd_is_mapped(dt, lnb[i].lnb_file_offset, &extent))
+ quota_space += PAGE_SIZE;
/* ignore quota for the whole request if any page is from
* client cache or written by root.
if ((lnb[i].lnb_flags & OBD_BRW_NOQUOTA) ||
(lnb[i].lnb_flags & (OBD_BRW_FROM_GRANT | OBD_BRW_SYNC)) ==
OBD_BRW_FROM_GRANT)
- ignore_quota = true;
+ declare_flags |= OSD_QID_FORCE;
}
/*
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);
+ lnb[0].lnb_flags &= ~OBD_BRW_OVER_ALLQUOTA;
rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
- quota_space, oh, osd_dt_obj(dt), true,
- &flags, ignore_quota);
+ i_projid_read(inode), quota_space, oh,
+ osd_dt_obj(dt), &flags, declare_flags);
/* we need only to store the overquota flags in the first lnb for
* now, once we support multiple objects BRW, this code needs be
lnb[0].lnb_flags |= OBD_BRW_OVER_USRQUOTA;
if (flags & QUOTA_FL_OVER_GRPQUOTA)
lnb[0].lnb_flags |= OBD_BRW_OVER_GRPQUOTA;
+ if (flags & QUOTA_FL_OVER_PRJQUOTA)
+ lnb[0].lnb_flags |= OBD_BRW_OVER_PRJQUOTA;
RETURN(rc);
}
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) {
thandle->th_local = 1;
}
- if (likely(rc == 0)) {
- if (isize > i_size_read(inode)) {
- i_size_write(inode, isize);
- LDISKFS_I(inode)->i_disksize = isize;
+ if (likely(rc == 0)) {
+ spin_lock(&inode->i_lock);
+ if (isize > i_size_read(inode)) {
+ i_size_write(inode, isize);
+ LDISKFS_I(inode)->i_disksize = isize;
+ spin_unlock(&inode->i_lock);
ll_dirty_inode(inode, I_DIRTY_DATASYNC);
- }
+ } else {
+ spin_unlock(&inode->i_lock);
+ }
- rc = osd_do_bio(osd, inode, iobuf);
- /* we don't do stats here as in read path because
- * write is async: we'll do this in osd_put_bufs() */
+ rc = osd_do_bio(osd, inode, iobuf);
+ /* we don't do stats here as in read path because
+ * write is async: we'll do this in osd_put_bufs() */
} else {
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));
struct osd_iobuf *iobuf = &oti->oti_iobuf;
struct inode *inode = osd_dt_obj(dt)->oo_inode;
struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
- struct timeval start, end;
- unsigned long timediff;
- int rc = 0, i, m = 0, cache = 0, cache_hits = 0, cache_misses = 0;
+ int rc = 0, i, cache = 0, cache_hits = 0, cache_misses = 0;
+ ktime_t start, end;
+ s64 timediff;
+ loff_t isize;
LASSERT(inode);
if (unlikely(rc != 0))
RETURN(rc);
+ isize = i_size_read(inode);
+
if (osd->od_read_cache)
cache = 1;
- if (i_size_read(inode) > osd->od_readcache_max_filesize)
+ if (isize > osd->od_readcache_max_filesize)
cache = 0;
- do_gettimeofday(&start);
+ start = ktime_get();
for (i = 0; i < npages; i++) {
- if (i_size_read(inode) <= lnb[i].lnb_file_offset)
+ if (isize <= lnb[i].lnb_file_offset)
/* If there's no more data, abort early.
* lnb->lnb_rc == 0, so it's easy to detect later. */
break;
- if (i_size_read(inode) <
- lnb[i].lnb_file_offset + lnb[i].lnb_len - 1)
- lnb[i].lnb_rc = i_size_read(inode) -
- lnb[i].lnb_file_offset;
+ 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;
- m += lnb[i].lnb_len;
+
+ /* Bypass disk read if fail_loc is set properly */
+ if (OBD_FAIL_CHECK(OBD_FAIL_OST_FAKE_RW))
+ SetPageUptodate(lnb[i].lnb_page);
if (PageUptodate(lnb[i].lnb_page)) {
cache_hits++;
generic_error_remove_page(inode->i_mapping,
lnb[i].lnb_page);
}
- do_gettimeofday(&end);
- timediff = cfs_timeval_sub(&end, &start, NULL);
+ end = ktime_get();
+ timediff = ktime_us_delta(end, start);
lprocfs_counter_add(osd->od_stats, LPROC_OSD_GET_PAGE, timediff);
if (cache_hits != 0)
int blocksize;
int csize;
int boffs;
- int err;
/* prevent reading after eof */
spin_lock(&inode->i_lock);
block = *offs >> inode->i_blkbits;
boffs = *offs & (blocksize - 1);
csize = min(blocksize - boffs, size);
- bh = ldiskfs_bread(NULL, inode, block, 0, &err);
- if (err != 0) {
- CERROR("%s: can't read %u@%llu on ino %lu: rc = %d\n",
- LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
- csize, *offs, inode->i_ino, err);
- if (bh != NULL)
- brelse(bh);
- return err;
+ bh = __ldiskfs_bread(NULL, inode, block, 0);
+ if (IS_ERR(bh)) {
+ CERROR("%s: can't read %u@%llu on ino %lu: "
+ "rc = %ld\n", osd_ino2name(inode),
+ csize, *offs, inode->i_ino,
+ PTR_ERR(bh));
+ return PTR_ERR(bh);
}
if (bh != NULL) {
}
static ssize_t osd_read(const struct lu_env *env, struct dt_object *dt,
- struct lu_buf *buf, loff_t *pos,
- struct lustre_capa *capa)
+ struct lu_buf *buf, loff_t *pos)
{
struct inode *inode = osd_dt_obj(dt)->oo_inode;
int rc;
- if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
- RETURN(-EACCES);
-
/* Read small symlink from inode body as we need to maintain correct
* on-disk symlinks for ldiskfs.
*/
if (inode != NULL) {
if (LDISKFS_I(inode)->i_flags & LDISKFS_EXTENTS_FL)
return 1;
- } else if (test_opt(sb, EXTENTS)) {
+ } else if (LDISKFS_HAS_INCOMPAT_FEATURE(sb,
+ LDISKFS_FEATURE_INCOMPAT_EXTENTS)) {
return 1;
}
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;
credits = depth;
/* if not append, then split may need to modify
* existing blocks moving entries into the new ones */
- if (_pos == -1)
+ if (_pos != -1)
credits += depth;
/* blocks to store data: bitmap,gd,itself */
credits += blocks * 3;
* objects, so always set the lqi_space as 0. */
if (inode != NULL)
rc = osd_declare_inode_qid(env, i_uid_read(inode),
- i_gid_read(inode), 0, oh, obj, true,
- NULL, false);
+ i_gid_read(inode),
+ i_projid_read(inode), 0,
+ oh, obj, NULL, OSD_QID_BLK);
RETURN(rc);
}
/* LU-2634: clear the extent format for fast symlink */
ldiskfs_clear_inode_flag(inode, LDISKFS_INODE_EXTENTS);
- memcpy((char *)&LDISKFS_I(inode)->i_data, (char *)buffer, buflen);
- LDISKFS_I(inode)->i_disksize = buflen;
- i_size_write(inode, buflen);
+ memcpy((char *)&LDISKFS_I(inode)->i_data, (char *)buffer, buflen);
+ spin_lock(&inode->i_lock);
+ LDISKFS_I(inode)->i_disksize = buflen;
+ i_size_write(inode, buflen);
+ spin_unlock(&inode->i_lock);
ll_dirty_inode(inode, I_DIRTY_DATASYNC);
- return 0;
+ return 0;
}
int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
((char *)buf)[bufsize] = '\0';
++bufsize;
}
- while (bufsize > 0) {
- if (bh != NULL)
- brelse(bh);
-
- block = offset >> inode->i_blkbits;
- boffs = offset & (blocksize - 1);
- size = min(blocksize - boffs, bufsize);
- bh = ldiskfs_bread(handle, inode, block, 1, &err);
- if (!bh) {
- CERROR("%s: error reading offset %llu (block %lu): "
- "rc = %d\n",
- inode->i_sb->s_id, offset, block, err);
+
+ while (bufsize > 0) {
+ int credits = handle->h_buffer_credits;
+
+ if (bh)
+ brelse(bh);
+
+ block = offset >> inode->i_blkbits;
+ boffs = offset & (blocksize - 1);
+ size = min(blocksize - boffs, bufsize);
+ bh = __ldiskfs_bread(handle, inode, block, 1);
+ if (IS_ERR_OR_NULL(bh)) {
+ if (bh == NULL) {
+ err = -EIO;
+ } else {
+ err = PTR_ERR(bh);
+ bh = NULL;
+ }
+
+ CERROR("%s: error reading offset %llu (block %lu, "
+ "size %d, offs %llu), credits %d/%d: rc = %d\n",
+ inode->i_sb->s_id, offset, block, bufsize, *offs,
+ credits, handle->h_buffer_credits, err);
break;
}
"boffs %d size %d bh->b_size %lu\n",
boffs, size, (unsigned long)bh->b_size);
memcpy(bh->b_data + boffs, buf, size);
- err = ldiskfs_journal_dirty_metadata(handle, bh);
+ err = ldiskfs_handle_dirty_metadata(handle, NULL, bh);
if (err)
break;
if (write_NUL)
--new_size;
- /* correct in-core and on-disk sizes */
- if (new_size > i_size_read(inode)) {
+ /* correct in-core and on-disk sizes */
+ if (new_size > i_size_read(inode)) {
spin_lock(&inode->i_lock);
if (new_size > i_size_read(inode))
i_size_write(inode, new_size);
}
static ssize_t osd_write(const struct lu_env *env, struct dt_object *dt,
- const struct lu_buf *buf, loff_t *pos,
- struct thandle *handle, struct lustre_capa *capa,
- int ignore_quota)
+ const struct lu_buf *buf, loff_t *pos,
+ struct thandle *handle, int ignore_quota)
{
struct inode *inode = osd_dt_obj(dt)->oo_inode;
struct osd_thandle *oh;
LASSERT(dt_object_exists(dt));
- if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_WRITE))
- return -EACCES;
-
LASSERT(handle != NULL);
LASSERT(inode != NULL);
ll_vfs_dq_init(inode);
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(inode);
rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
- 0, oh, osd_dt_obj(dt), true, NULL, false);
+ i_projid_read(inode), 0, oh, osd_dt_obj(dt),
+ NULL, OSD_QID_BLK);
RETURN(rc);
}
static int osd_punch(const struct lu_env *env, struct dt_object *dt,
- __u64 start, __u64 end, struct thandle *th,
- struct lustre_capa *capa)
+ __u64 start, __u64 end, struct thandle *th)
{
struct osd_thandle *oh;
struct osd_object *obj = osd_dt_obj(dt);
tid = oh->ot_handle->h_transaction->t_tid;
+ spin_lock(&inode->i_lock);
i_size_write(inode, start);
+ spin_unlock(&inode->i_lock);
ll_truncate_pagecache(inode, start);
#ifdef HAVE_INODEOPS_TRUNCATE
if (inode->i_op->truncate) {
* For a partial-page truncate, flush the page to disk immediately to
* avoid data corruption during direct disk write. b=17397
*/
- if ((start & ~CFS_PAGE_MASK) != 0)
+ if ((start & ~PAGE_MASK) != 0)
rc = filemap_fdatawrite_range(inode->i_mapping, start, start+1);
h = journal_current_handle();
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;
/*
RETURN(rc == 0 ? rc2 : rc);
}
+static int fiemap_check_ranges(struct inode *inode,
+ u64 start, u64 len, u64 *new_len)
+{
+ loff_t maxbytes;
+
+ *new_len = len;
+
+ if (len == 0)
+ return -EINVAL;
+
+ if (ldiskfs_test_inode_flag(inode, LDISKFS_INODE_EXTENTS))
+ maxbytes = inode->i_sb->s_maxbytes;
+ else
+ maxbytes = LDISKFS_SB(inode->i_sb)->s_bitmap_maxbytes;
+
+ if (start > maxbytes)
+ return -EFBIG;
+
+ /*
+ * Shrink request scope to what the fs can actually handle.
+ */
+ if (len > maxbytes || (maxbytes - len) < start)
+ *new_len = maxbytes - start;
+
+ return 0;
+}
+
+/* So that the fiemap access checks can't overflow on 32 bit machines. */
+#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 inode *inode = osd_dt_obj(dt)->oo_inode;
- struct osd_thread_info *info = osd_oti_get(env);
- struct dentry *dentry = &info->oti_obj_dentry;
- struct file *file = &info->oti_file;
- mm_segment_t saved_fs;
- int rc;
+ struct fiemap_extent_info fieinfo = {0, };
+ struct inode *inode = osd_dt_obj(dt)->oo_inode;
+ u64 len;
+ int rc;
- LASSERT(inode);
- dentry->d_inode = inode;
- dentry->d_sb = inode->i_sb;
- file->f_dentry = dentry;
- file->f_mapping = inode->i_mapping;
- file->f_op = inode->i_fop;
- set_file_inode(file, inode);
-
- saved_fs = get_fs();
- set_fs(get_ds());
- /* ldiskfs_ioctl does not have a inode argument */
- if (inode->i_fop->unlocked_ioctl)
- rc = inode->i_fop->unlocked_ioctl(file, FSFILT_IOC_FIEMAP,
- (long)fm);
- else
- rc = -ENOTTY;
- set_fs(saved_fs);
- return rc;
+
+ LASSERT(inode);
+ if (inode->i_op->fiemap == NULL)
+ return -EOPNOTSUPP;
+
+ if (fm->fm_extent_count > FIEMAP_MAX_EXTENTS)
+ return -EINVAL;
+
+ rc = fiemap_check_ranges(inode, fm->fm_start, fm->fm_length, &len);
+ if (rc)
+ return rc;
+
+ fieinfo.fi_flags = fm->fm_flags;
+ fieinfo.fi_extents_max = fm->fm_extent_count;
+ fieinfo.fi_extents_start = fm->fm_extents;
+
+ if (fieinfo.fi_flags & FIEMAP_FLAG_SYNC)
+ filemap_write_and_wait(inode->i_mapping);
+
+ rc = inode->i_op->fiemap(inode, &fieinfo, fm->fm_start, len);
+ fm->fm_flags = fieinfo.fi_flags;
+ fm->fm_mapped_extents = fieinfo.fi_extents_mapped;
+
+ 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 = 0;
+ struct inode *inode = osd_dt_obj(dt)->oo_inode;
+ ENTRY;
+
+ switch (advice) {
+ case LU_LADVISE_DONTNEED:
+ if (end == 0)
+ break;
+ invalidate_mapping_pages(inode->i_mapping,
+ start >> PAGE_CACHE_SHIFT,
+ (end - 1) >> PAGE_CACHE_SHIFT);
+ break;
+ default:
+ rc = -ENOTSUPP;
+ break;
+ }
+
+ RETURN(rc);
}
/*
};
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