*
*/
+#define DEBUG_SUBSYSTEM S_OSD
+
/* prerequisite for linux/xattr.h */
#include <linux/types.h>
/* prerequisite for linux/xattr.h */
#ifdef HAVE_BIO_ENDIO_USES_ONE_ARG
static void dio_complete_routine(struct bio *bio)
{
- int error = bio->bi_status;
+ int error = blk_status_to_errno(bio->bi_status);
#else
static void dio_complete_routine(struct bio *bio, int error)
{
}
static int osd_do_bio(struct osd_device *osd, struct inode *inode,
- struct osd_iobuf *iobuf)
+ struct osd_iobuf *iobuf, sector_t start_blocks,
+ sector_t count)
{
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;
struct super_block *sb = inode->i_sb;
int sector_bits = sb->s_blocksize_bits - 9;
unsigned int blocksize = sb->s_blocksize;
unsigned int page_offset;
sector_t sector;
int nblocks;
- int block_idx;
- int page_idx;
+ int block_idx, block_idx_end;
+ int page_idx, page_idx_start;
int i;
int rc = 0;
bool fault_inject;
bool integrity_enabled;
struct blk_plug plug;
+ int blocks_left_page;
+
ENTRY;
fault_inject = OBD_FAIL_CHECK(OBD_FAIL_OST_INTEGRITY_FAULT);
osd_brw_stats_update(osd, iobuf);
iobuf->dr_start_time = ktime_get();
+ if (!count)
+ count = npages * blocks_per_page;
+ block_idx_end = start_blocks + count;
+
blk_start_plug(&plug);
- for (page_idx = 0, block_idx = 0;
- page_idx < npages;
- page_idx++, block_idx += blocks_per_page) {
+
+ page_idx_start = start_blocks / blocks_per_page;
+ for (page_idx = page_idx_start, block_idx = start_blocks;
+ block_idx < block_idx_end; page_idx++,
+ block_idx += blocks_left_page) {
page = pages[page_idx];
- LASSERT(block_idx + blocks_per_page <= total_blocks);
+ LASSERT(page_idx < iobuf->dr_npages);
- for (i = 0, page_offset = 0;
- i < blocks_per_page;
+ i = block_idx % blocks_per_page;
+ blocks_left_page = blocks_per_page - i;
+ for (page_offset = i * blocksize; i < blocks_left_page;
i += nblocks, page_offset += blocksize * nblocks) {
nblocks = 1;
if (blocks[block_idx + i] == 0) { /* hole */
LASSERTF(iobuf->dr_rw == 0,
- "page_idx %u, block_idx %u, i %u\n",
- page_idx, block_idx, i);
+ "page_idx %u, block_idx %u, i %u,"
+ "start_blocks: %llu, count: %llu, npages: %d\n",
+ page_idx, block_idx, i,
+ (unsigned long long)start_blocks,
+ (unsigned long long)count, npages);
memset(kmap(page) + page_offset, 0, blocksize);
kunmap(page);
continue;
sector = (sector_t)blocks[block_idx + i] << sector_bits;
/* Additional contiguous file blocks? */
- while (i + nblocks < blocks_per_page &&
+ while (i + nblocks < blocks_left_page &&
(sector + (nblocks << sector_bits)) ==
((sector_t)blocks[block_idx + i + nblocks] <<
sector_bits))
bio_start_page_idx = page_idx;
/* allocate new bio */
bio = bio_alloc(GFP_NOIO, min(BIO_MAX_PAGES,
- (npages - page_idx) *
- blocks_per_page));
+ (block_idx_end - block_idx +
+ blocks_left_page - 1)));
if (bio == NULL) {
- CERROR("Can't allocate bio %u*%u = %u pages\n",
- (npages - page_idx), blocks_per_page,
- (npages - page_idx) * blocks_per_page);
+ CERROR("Can't allocate bio %u pages\n",
+ block_idx_end - block_idx +
+ blocks_left_page - 1);
rc = -ENOMEM;
goto out;
}
osd_bufs_put(env, dt, lnb - i, i);
return rc;
}
+/* Borrow @ext4_chunk_trans_blocks */
+static int osd_chunk_trans_blocks(struct inode *inode, int nrblocks)
+{
+ ldiskfs_group_t groups;
+ int gdpblocks;
+ int idxblocks;
+ int depth;
+ int ret;
+
+ depth = ext_depth(inode);
+ idxblocks = depth * 2;
+
+ /*
+ * Now let's see how many group bitmaps and group descriptors need
+ * to account.
+ */
+ groups = idxblocks + 1;
+ gdpblocks = groups;
+ if (groups > LDISKFS_SB(inode->i_sb)->s_groups_count)
+ groups = LDISKFS_SB(inode->i_sb)->s_groups_count;
+ if (gdpblocks > LDISKFS_SB(inode->i_sb)->s_gdb_count)
+ gdpblocks = LDISKFS_SB(inode->i_sb)->s_gdb_count;
+
+ /* bitmaps and block group descriptor blocks */
+ ret = idxblocks + groups + gdpblocks;
+
+ /* Blocks for super block, inode, quota and xattr blocks */
+ ret += LDISKFS_META_TRANS_BLOCKS(inode->i_sb);
+
+ return ret;
+}
+
+#ifdef HAVE_LDISKFS_JOURNAL_ENSURE_CREDITS
+static int osd_extend_trans(handle_t *handle, int needed,
+ struct inode *inode)
+{
+ return __ldiskfs_journal_ensure_credits(handle, needed, needed,
+ ldiskfs_trans_default_revoke_credits(inode->i_sb));
+}
+
+static int osd_extend_restart_trans(handle_t *handle, int needed,
+ struct inode *inode)
+{
+ int rc;
+
+ rc = ldiskfs_journal_ensure_credits(handle, needed,
+ ldiskfs_trans_default_revoke_credits(inode->i_sb));
+ /* this means journal has been restarted */
+ if (rc > 0)
+ rc = 0;
+
+ return rc;
+}
+#else
+static int osd_extend_trans(handle_t *handle, int needed,
+ struct inode *inode)
+{
+ if (ldiskfs_handle_has_enough_credits(handle, needed))
+ return 0;
+
+ return ldiskfs_journal_extend(handle,
+ needed - handle->h_buffer_credits);
+}
+
+static int osd_extend_restart_trans(handle_t *handle, int needed,
+ struct inode *inode)
+{
+
+ int rc = osd_extend_trans(handle, needed, inode);
+
+ if (rc <= 0)
+ return rc;
+
+ return ldiskfs_journal_restart(handle, needed);
+}
+#endif /* HAVE_LDISKFS_JOURNAL_ENSURE_CREDITS */
+
+static int osd_ldiskfs_map_write(struct inode *inode, struct osd_iobuf *iobuf,
+ struct osd_device *osd, sector_t start_blocks,
+ sector_t count, loff_t *disk_size,
+ __u64 user_size)
+{
+ /* if file has grown, take user_size into account */
+ if (user_size && *disk_size > user_size)
+ *disk_size = user_size;
+
+ spin_lock(&inode->i_lock);
+ if (*disk_size > i_size_read(inode)) {
+ i_size_write(inode, *disk_size);
+ LDISKFS_I(inode)->i_disksize = *disk_size;
+ spin_unlock(&inode->i_lock);
+ osd_dirty_inode(inode, I_DIRTY_DATASYNC);
+ } else {
+ spin_unlock(&inode->i_lock);
+ }
+
+ /*
+ * We don't do stats here as in read path because
+ * write is async: we'll do this in osd_put_bufs()
+ */
+ return osd_do_bio(osd, inode, iobuf, start_blocks, count);
+}
+
-static int osd_ldiskfs_map_inode_pages(struct inode *inode, struct page **page,
- int pages, sector_t *blocks,
- int create)
+static int osd_ldiskfs_map_inode_pages(struct inode *inode,
+ struct osd_iobuf *iobuf,
+ struct osd_device *osd,
+ int create, __u64 user_size,
+ int check_credits)
{
int blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
- int rc = 0, i = 0;
+ int rc = 0, i = 0, mapped_index = 0;
struct page *fp = NULL;
int clen = 0;
pgoff_t max_page_index;
handle_t *handle = NULL;
+ int credits;
+ sector_t start_blocks = 0, count = 0;
+ loff_t disk_size = 0;
+ struct page **page = iobuf->dr_pages;
+ int pages = iobuf->dr_npages;
+ sector_t *blocks = iobuf->dr_blocks;
+ struct niobuf_local *lnb1, *lnb2;
+ loff_t size1, size2;
max_page_index = inode->i_sb->s_maxbytes >> PAGE_SHIFT;
rc = osd_attach_jinode(inode);
if (rc)
return rc;
+ disk_size = i_size_read(inode);
+ /* if disk_size is already bigger than specified user_size,
+ * ignore user_size
+ */
+ if (disk_size > user_size)
+ user_size = 0;
}
/* pages are sorted already. so, we just have to find
* contig. space and process them properly
*/
while (i < pages) {
- long blen, total = 0;
+ long blen, total = 0, previous_total = 0;
struct ldiskfs_map_blocks map = { 0 };
if (fp == NULL) { /* start new extent */
map.m_lblk = fp->index * blocks_per_page;
map.m_len = blen = clen * blocks_per_page;
cont_map:
+ /**
+ * We might restart transaction for block allocations,
+ * in order to make sure data ordered mode, issue IO, disk
+ * size update and block allocations need be within same
+ * transaction to make sure consistency.
+ */
+ if (handle && check_credits) {
+ /*
+ * credits to insert 1 extent into extent tree.
+ */
+ credits = osd_chunk_trans_blocks(inode, blen);
+ rc = osd_extend_trans(handle, credits, inode);
+ if (rc < 0)
+ GOTO(cleanup, rc);
+ /*
+ * only issue IO if restart transaction needed,
+ * as update disk size need hold inode lock, we
+ * want to avoid that as much as possible.
+ */
+ if (rc > 0) {
+ WARN_ON_ONCE(start_blocks == 0);
+ rc = osd_ldiskfs_map_write(inode,
+ iobuf, osd, start_blocks,
+ count, &disk_size, user_size);
+ if (rc)
+ GOTO(cleanup, rc);
+#ifdef HAVE_LDISKFS_JOURNAL_ENSURE_CREDITS
+ rc = ldiskfs_journal_restart(handle, credits,
+ ldiskfs_trans_default_revoke_credits(inode->i_sb));
+#else
+ rc = ldiskfs_journal_restart(handle, credits);
+#endif
+ if (rc)
+ GOTO(cleanup, rc);
+ start_blocks += count;
+ /* reset IO block count */
+ count = 0;
+ }
+ }
rc = ldiskfs_map_blocks(handle, inode, &map, create);
if (rc >= 0) {
int c = 0;
}
rc = 0;
}
+
+ if (rc == 0 && create) {
+ count += (total - previous_total);
+ mapped_index = (count + blocks_per_page -
+ 1) / blocks_per_page - 1;
+ lnb1 = iobuf->dr_lnbs[i - clen];
+ lnb2 = iobuf->dr_lnbs[mapped_index];
+ size1 = lnb1->lnb_file_offset -
+ (lnb1->lnb_file_offset % PAGE_SIZE) +
+ (total << inode->i_blkbits);
+ size2 = lnb2->lnb_file_offset + lnb2->lnb_len;
+
+ if (size1 > size2)
+ size1 = size2;
+ if (size1 > disk_size)
+ disk_size = size1;
+ }
+
if (rc == 0 && total < blen) {
map.m_lblk = fp->index * blocks_per_page + total;
map.m_len = blen - total;
+ previous_total = total;
goto cont_map;
}
if (rc != 0)
blocks += blocks_per_page * clen;
}
cleanup:
+ if (rc == 0 && create &&
+ start_blocks < pages * blocks_per_page) {
+ rc = osd_ldiskfs_map_write(inode, iobuf, osd, start_blocks,
+ count, &disk_size, user_size);
+ LASSERT(start_blocks + count == pages * blocks_per_page);
+ }
return rc;
}
lprocfs_counter_add(osd->od_stats, LPROC_OSD_GET_PAGE, timediff);
if (iobuf->dr_npages) {
- rc = osd_ldiskfs_map_inode_pages(inode, iobuf->dr_pages,
- iobuf->dr_npages,
- iobuf->dr_blocks, 0);
+ rc = osd_ldiskfs_map_inode_pages(inode, iobuf, osd, 0,
+ 0, 0);
if (likely(rc == 0)) {
- rc = osd_do_bio(osd, inode, iobuf);
+ rc = osd_do_bio(osd, inode, iobuf, 0, 0);
/* do IO stats for preparation reads */
osd_fini_iobuf(osd, iobuf);
}
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 extents = 0;
int depth;
int i;
- int newblocks;
+ int newblocks = 0;
int rc = 0;
int credits = 0;
long long quota_space = 0;
- struct osd_fextent extent = { 0 };
+ struct osd_fextent mapped = { 0 }, extent = { 0 };
enum osd_quota_local_flags local_flags = 0;
enum osd_qid_declare_flags declare_flags = OSD_QID_BLK;
ENTRY;
oh = container_of(handle, struct osd_thandle, ot_super);
LASSERT(oh->ot_handle == NULL);
- newblocks = npages;
-
/* calculate number of extents (probably better to pass nb) */
for (i = 0; i < npages; i++) {
- if (i && lnb[i].lnb_file_offset !=
- lnb[i - 1].lnb_file_offset + lnb[i - 1].lnb_len)
- extents++;
-
- if (osd_is_mapped(dt, lnb[i].lnb_file_offset, &extent))
- lnb[i].lnb_flags |= OBD_BRW_MAPPED;
- else
- quota_space += PAGE_SIZE;
-
/* ignore quota for the whole request if any page is from
* client cache or written by root.
*
(lnb[i].lnb_flags & (OBD_BRW_FROM_GRANT | OBD_BRW_SYNC)) ==
OBD_BRW_FROM_GRANT)
declare_flags |= OSD_QID_FORCE;
+
+ if (osd_is_mapped(dt, lnb[i].lnb_file_offset, &mapped)) {
+ lnb[i].lnb_flags |= OBD_BRW_MAPPED;
+ continue;
+ }
+
+ /* count only unmapped changes */
+ newblocks++;
+ if (lnb[i].lnb_file_offset != extent.end || extent.end == 0) {
+ extents++;
+ extent.end = lnb[i].lnb_file_offset + lnb[i].lnb_len;
+ } else {
+ extent.end += lnb[i].lnb_len;
+ }
+
+ quota_space += PAGE_SIZE;
}
+ credits++; /* inode */
+ /*
+ * overwrite case, no need to modify tree and
+ * allocate blocks.
+ */
+ if (!newblocks)
+ goto out_declare;
/*
* each extent can go into new leaf causing a split
* 5 is max tree depth: inode + 4 index blocks
depth = ext_depth(inode);
depth = max(depth, 1) + 1;
newblocks += depth;
- credits++; /* inode */
credits += depth * 2 * extents;
} else {
depth = 3;
newblocks += depth;
- credits++; /* inode */
credits += depth * extents;
}
+ /*
+ * try a bit more extents to avoid restart
+ * as much as possible in normal case.
+ */
+ if (npages > 1 && extents)
+ extents <<= 1;
+
/* quota space for metadata blocks */
quota_space += depth * extents * LDISKFS_BLOCK_SIZE(osd_sb(osd));
/* each new block can go in different group (bitmap + gd) */
/* we can't dirty more bitmap blocks than exist */
- if (newblocks > LDISKFS_SB(osd_sb(osd))->s_groups_count)
+ if (extents > LDISKFS_SB(osd_sb(osd))->s_groups_count)
credits += LDISKFS_SB(osd_sb(osd))->s_groups_count;
else
- credits += newblocks;
+ credits += extents;
/* we can't dirty more gd blocks than exist */
- if (newblocks > LDISKFS_SB(osd_sb(osd))->s_gdb_count)
+ if (extents > LDISKFS_SB(osd_sb(osd))->s_gdb_count)
credits += LDISKFS_SB(osd_sb(osd))->s_gdb_count;
else
- credits += newblocks;
+ credits += extents;
+out_declare:
osd_trans_declare_op(env, oh, OSD_OT_WRITE, credits);
/* make sure the over quota flags were not set */
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));
- loff_t disk_size;
- int rc = 0, i;
+ int rc = 0, i, check_credits = 0;
+ struct osd_thandle *oh = container_of(thandle,
+ struct osd_thandle, ot_super);
+ unsigned int save_credits = oh->ot_credits;
LASSERT(inode);
if (unlikely(rc != 0))
RETURN(rc);
- disk_size = i_size_read(inode);
- /* if disk_size is already bigger than specified user_size,
- * ignore user_size
- */
- if (disk_size > user_size)
- user_size = 0;
dquot_initialize(inode);
for (i = 0; i < npages; i++) {
continue;
}
+ if (!(lnb[i].lnb_flags & OBD_BRW_MAPPED))
+ check_credits = 1;
+
LASSERT(PageLocked(lnb[i].lnb_page));
LASSERT(!PageWriteback(lnb[i].lnb_page));
- if (lnb[i].lnb_file_offset + lnb[i].lnb_len > disk_size)
- disk_size = lnb[i].lnb_file_offset + lnb[i].lnb_len;
-
/*
* Since write and truncate are serialized by oo_sem, even
* partial-page truncate should not leave dirty pages in the
osd_iobuf_add_page(iobuf, &lnb[i]);
}
- /* if file has grown, take user_size into account */
- if (user_size && disk_size > user_size)
- disk_size = user_size;
-
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) {
- rc = osd_ldiskfs_map_inode_pages(inode, iobuf->dr_pages,
- iobuf->dr_npages,
- iobuf->dr_blocks, 1);
+ rc = osd_ldiskfs_map_inode_pages(inode, iobuf, osd,
+ 1, user_size,
+ check_credits);
+ /*
+ * Write might restart transaction, extend credits
+ * if needed for operations such as attribute set.
+ */
+ if (rc == 0) {
+ handle_t *handle = ldiskfs_journal_current_handle();
+
+ LASSERT(handle != NULL);
+ rc = osd_extend_restart_trans(handle, save_credits,
+ inode);
+ }
} else {
/* no pages to write, no transno is needed */
thandle->th_local = 1;
}
- if (likely(rc == 0)) {
- spin_lock(&inode->i_lock);
- if (disk_size > i_size_read(inode)) {
- i_size_write(inode, disk_size);
- LDISKFS_I(inode)->i_disksize = disk_size;
- spin_unlock(&inode->i_lock);
- osd_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()
- */
- } else {
+ if (rc != 0)
osd_fini_iobuf(osd, iobuf);
- }
osd_trans_exec_check(env, thandle, OSD_OT_WRITE);
cache_hits + cache_misses);
if (iobuf->dr_npages) {
- rc = osd_ldiskfs_map_inode_pages(inode, iobuf->dr_pages,
- iobuf->dr_npages,
- iobuf->dr_blocks, 0);
- rc = osd_do_bio(osd, inode, iobuf);
+ rc = osd_ldiskfs_map_inode_pages(inode, iobuf, osd, 0,
+ 0, 0);
+ if (!rc)
+ rc = osd_do_bio(osd, inode, iobuf, 0, 0);
/* IO stats will be done in osd_bufs_put() */
if (unlikely(IS_ERR_OR_NULL(bh) && !sync))
CWARN(
"%s: adding bh without locking off %llu (block %lu, size %d, offs %llu)\n",
- inode->i_sb->s_id,
+ osd_ino2name(inode),
offset, block, bufsize, *offs);
if (IS_ERR_OR_NULL(bh)) {
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);
+ osd_ino2name(inode), offset, block, bufsize,
+ *offs, credits, handle->h_buffer_credits, err);
break;
}
}
static int osd_declare_fallocate(const struct lu_env *env,
- struct dt_object *dt, struct thandle *th)
+ struct dt_object *dt, __u64 start, __u64 end,
+ int mode, struct thandle *th)
{
- struct osd_thandle *oh;
- struct inode *inode;
+ struct osd_thandle *oh = container_of(th, struct osd_thandle, ot_super);
+ struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
+ struct inode *inode = osd_dt_obj(dt)->oo_inode;
+ long long quota_space = 0;
+ /* 5 is max tree depth. (inode + 4 index blocks) */
+ int depth = 5;
int rc;
+
ENTRY;
- LASSERT(th);
- oh = container_of(th, struct osd_thandle, ot_super);
+ /*
+ * Only mode == 0 (which is standard prealloc) is supported now.
+ * Rest of mode options is not supported yet.
+ */
+ if (mode & ~FALLOC_FL_KEEP_SIZE)
+ RETURN(-EOPNOTSUPP);
- osd_trans_declare_op(env, oh, OSD_OT_PREALLOC,
- osd_dto_credits_noquota[DTO_WRITE_BLOCK]);
- inode = osd_dt_obj(dt)->oo_inode;
+ LASSERT(th);
LASSERT(inode);
+ /* quota space for metadata blocks
+ * approximate metadata estimate should be good enough.
+ */
+ quota_space += PAGE_SIZE;
+ quota_space += depth * LDISKFS_BLOCK_SIZE(osd_sb(osd));
+
+ /* quota space should be reported in 1K blocks */
+ quota_space = toqb(quota_space) + toqb(end - start) +
+ LDISKFS_META_TRANS_BLOCKS(inode->i_sb);
+
+ /* We don't need to reserve credits for whole fallocate here.
+ * We reserve space only for metadata. Fallocate credits are
+ * extended as required
+ */
rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
- i_projid_read(inode), 0, oh, osd_dt_obj(dt),
- NULL, OSD_QID_BLK);
+ i_projid_read(inode), quota_space, oh,
+ osd_dt_obj(dt), NULL, OSD_QID_BLK);
RETURN(rc);
}
static int osd_fallocate(const struct lu_env *env, struct dt_object *dt,
__u64 start, __u64 end, int mode, struct thandle *th)
{
+ struct osd_thandle *oh = container_of(th, struct osd_thandle, ot_super);
+ handle_t *handle = ldiskfs_journal_current_handle();
+ unsigned int save_credits = oh->ot_credits;
struct osd_object *obj = osd_dt_obj(dt);
struct inode *inode = obj->oo_inode;
+ struct ldiskfs_map_blocks map;
+ unsigned int credits;
+ ldiskfs_lblk_t blen;
+ ldiskfs_lblk_t boff;
+ loff_t new_size = 0;
+ int depth = 0;
+ int flags;
int rc = 0;
- struct osd_thread_info *info = osd_oti_get(env);
- struct dentry *dentry = &info->oti_obj_dentry;
- struct file *file = &info->oti_file;
ENTRY;
- /*
- * Only mode == 0 (which is standard prealloc) is supported now.
- * Rest of mode options is not supported yet.
- */
- if (mode & ~FALLOC_FL_KEEP_SIZE)
- RETURN(-EOPNOTSUPP);
LASSERT(dt_object_exists(dt));
LASSERT(osd_invariant(obj));
LASSERT(inode != NULL);
+
+ CDEBUG(D_INODE, "fallocate: inode #%lu: start %llu end %llu mode %d\n",
+ inode->i_ino, start, end, mode);
+
dquot_initialize(inode);
LASSERT(th);
- osd_trans_exec_op(env, th, OSD_OT_PREALLOC);
+ boff = start >> inode->i_blkbits;
+ blen = (ALIGN(end, 1 << inode->i_blkbits) >> inode->i_blkbits) - boff;
+
+ flags = LDISKFS_GET_BLOCKS_CREATE;
+ if (mode & FALLOC_FL_KEEP_SIZE)
+ flags |= LDISKFS_GET_BLOCKS_KEEP_SIZE;
+
+ inode_lock(inode);
+
+ /*
+ * We only support preallocation for extent-based file only.
+ */
+ if (!(ldiskfs_test_inode_flag(inode, LDISKFS_INODE_EXTENTS)))
+ GOTO(out, rc = -EOPNOTSUPP);
+
+ if (!(mode & FALLOC_FL_KEEP_SIZE) && (end > i_size_read(inode) ||
+ end > LDISKFS_I(inode)->i_disksize)) {
+ new_size = end;
+ rc = inode_newsize_ok(inode, new_size);
+ if (rc)
+ GOTO(out, rc);
+ }
+
+ inode_dio_wait(inode);
+
+ map.m_lblk = boff;
+ map.m_len = blen;
+
+ /* Don't normalize the request if it can fit in one extent so
+ * that it doesn't get unnecessarily split into multiple extents.
+ */
+ if (blen <= EXT_UNWRITTEN_MAX_LEN)
+ flags |= LDISKFS_GET_BLOCKS_NO_NORMALIZE;
/*
- * Because f_op->fallocate() does not have an inode arg
+ * credits to insert 1 extent into extent tree.
*/
- dentry->d_inode = inode;
- dentry->d_sb = inode->i_sb;
- file->f_path.dentry = dentry;
- file->f_mapping = inode->i_mapping;
- file->f_op = inode->i_fop;
- file->f_inode = inode;
- rc = file->f_op->fallocate(file, mode, start, end - start);
+ credits = osd_chunk_trans_blocks(inode, blen);
+ depth = ext_depth(inode);
+
+ while (rc >= 0 && blen) {
+ loff_t epos;
+
+ /*
+ * Recalculate credits when extent tree depth changes.
+ */
+ if (depth != ext_depth(inode)) {
+ credits = osd_chunk_trans_blocks(inode, blen);
+ depth = ext_depth(inode);
+ }
+
+ /* TODO: quota check */
+ rc = osd_extend_restart_trans(handle, credits, inode);
+ if (rc)
+ break;
+
+ rc = ldiskfs_map_blocks(handle, inode, &map, flags);
+ if (rc <= 0) {
+ CDEBUG(D_INODE,
+ "inode #%lu: block %u: len %u: ldiskfs_map_blocks returned %d\n",
+ inode->i_ino, map.m_lblk, map.m_len, rc);
+ ldiskfs_mark_inode_dirty(handle, inode);
+ break;
+ }
+
+ map.m_lblk += rc;
+ map.m_len = blen = blen - rc;
+ epos = (loff_t)map.m_lblk << inode->i_blkbits;
+ inode->i_ctime = current_time(inode);
+ if (new_size) {
+ if (epos > end)
+ epos = end;
+ if (ldiskfs_update_inode_size(inode, epos) & 0x1)
+ inode->i_mtime = inode->i_ctime;
+ } else {
+ if (epos > inode->i_size)
+ ldiskfs_set_inode_flag(inode,
+ LDISKFS_INODE_EOFBLOCKS);
+ }
+
+ ldiskfs_mark_inode_dirty(handle, inode);
+ }
+
+out:
+ /* extand credits if needed for operations such as attribute set */
+ if (rc >= 0)
+ rc = osd_extend_restart_trans(handle, save_credits, inode);
+
+ inode_unlock(inode);
RETURN(rc);
}
RETURN(rc);
}
+static loff_t osd_lseek(const struct lu_env *env, struct dt_object *dt,
+ loff_t offset, int whence)
+{
+ struct osd_object *obj = osd_dt_obj(dt);
+ struct inode *inode = obj->oo_inode;
+ struct file *file;
+ loff_t result;
+
+ ENTRY;
+
+ LASSERT(dt_object_exists(dt));
+ LASSERT(osd_invariant(obj));
+ LASSERT(inode);
+ LASSERT(offset >= 0);
+
+ file = osd_quasi_file(env, inode);
+ result = file->f_op->llseek(file, offset, whence);
+
+ /*
+ * If 'offset' is beyond end of object file then treat it as not error
+ * but valid case for SEEK_HOLE and return 'offset' as result.
+ * LOV will decide if it is beyond real end of file or not.
+ */
+ if (whence == SEEK_HOLE && result == -ENXIO)
+ result = offset;
+
+ CDEBUG(D_INFO, "seek %s from %lld: %lld\n", whence == SEEK_HOLE ?
+ "hole" : "data", offset, result);
+ RETURN(result);
+}
+
/*
* in some cases we may need declare methods for objects being created
* e.g., when we create symlink
.dbo_ladvise = osd_ladvise,
.dbo_declare_fallocate = osd_declare_fallocate,
.dbo_fallocate = osd_fallocate,
+ .dbo_lseek = osd_lseek,
};
/**
git://git.whamcloud.com / fs / lustre-release.git / blobdiff