X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_io.c;h=961e10dc42f47b24ba9a49df4bae3c7406fa7006;hp=26d9c3b499d350530056498987c9ab33af2f6bf1;hb=83d660436a164758fd4a29c1433d11c0f4591196;hpb=8e88bbfef5795b3bf9541c8cb5b2dd8a3bb11d9f diff --git a/lustre/osd-ldiskfs/osd_io.c b/lustre/osd-ldiskfs/osd_io.c index 26d9c3b..961e10d 100644 --- a/lustre/osd-ldiskfs/osd_io.c +++ b/lustre/osd-ldiskfs/osd_io.c @@ -666,10 +666,12 @@ static struct page *osd_get_page(const struct lu_env *env, struct dt_object *dt, page = find_or_create_page(inode->i_mapping, offset >> PAGE_SHIFT, gfp_mask); - if (likely(page)) - LASSERT(!test_bit(PG_private_2, &page->flags)); - else + if (likely(page)) { + LASSERT(!PagePrivate2(page)); + wait_on_page_writeback(page); + } else { lprocfs_counter_add(d->od_stats, LPROC_OSD_NO_PAGE, 1); + } return page; } @@ -678,34 +680,29 @@ static struct page *osd_get_page(const struct lu_env *env, struct dt_object *dt, /* consult with pagecache, but do not create new pages */ /* this is normally used once */ page = find_lock_page(inode->i_mapping, offset >> PAGE_SHIFT); - if (page) + if (page) { + wait_on_page_writeback(page); return page; + } } LASSERT(oti->oti_dio_pages); cur = oti->oti_dio_pages_used; + page = oti->oti_dio_pages[cur]; - if (unlikely(!oti->oti_dio_pages[cur])) { + if (unlikely(!page)) { LASSERT(cur < PTLRPC_MAX_BRW_PAGES); page = alloc_page(gfp_mask); if (!page) return NULL; oti->oti_dio_pages[cur] = page; + SetPagePrivate2(page); + lock_page(page); } - page = oti->oti_dio_pages[cur]; - LASSERT(!test_bit(PG_private_2, &page->flags)); - set_bit(PG_private_2, &page->flags); - oti->oti_dio_pages_used++; - - LASSERT(!PageLocked(page)); - lock_page(page); - - LASSERT(!page->mapping); - LASSERT(!PageWriteback(page)); ClearPageUptodate(page); - page->index = offset >> PAGE_SHIFT; + oti->oti_dio_pages_used++; return page; } @@ -758,11 +755,7 @@ static int osd_bufs_put(const struct lu_env *env, struct dt_object *dt, /* if the page isn't cached, then reset uptodate * to prevent reuse */ - if (test_bit(PG_private_2, &page->flags)) { - clear_bit(PG_private_2, &page->flags); - ClearPageUptodate(page); - if (lnb[i].lnb_locked) - unlock_page(page); + if (PagePrivate2(page)) { oti->oti_dio_pages_used--; } else { if (lnb[i].lnb_locked) @@ -770,7 +763,6 @@ static int osd_bufs_put(const struct lu_env *env, struct dt_object *dt, if (pagevec_add(&pvec, page) == 0) pagevec_release(&pvec); } - dt_object_put(env, dt); lnb[i].lnb_page = NULL; } @@ -865,8 +857,7 @@ static int osd_bufs_get(const struct lu_env *env, struct dt_object *dt, bypass_checks: if (!cache && unlikely(!oti->oti_dio_pages)) { - OBD_ALLOC(oti->oti_dio_pages, - sizeof(struct page *) * PTLRPC_MAX_BRW_PAGES); + OBD_ALLOC_PTR_ARRAY(oti->oti_dio_pages, PTLRPC_MAX_BRW_PAGES); if (!oti->oti_dio_pages) return -ENOMEM; } @@ -881,10 +872,6 @@ bypass_checks: GOTO(cleanup, rc = -ENOMEM); lnb->lnb_locked = 1; - wait_on_page_writeback(lnb->lnb_page); - BUG_ON(PageWriteback(lnb->lnb_page)); - - lu_object_get(&dt->do_lu); } #if 0 @@ -1133,13 +1120,13 @@ static int osd_declare_write_commit(const struct lu_env *env, enum osd_qid_declare_flags declare_flags = OSD_QID_BLK; ENTRY; - LASSERT(handle != NULL); - oh = container_of0(handle, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle == NULL); + LASSERT(handle != NULL); + oh = container_of(handle, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle == NULL); - newblocks = npages; + newblocks = npages; - /* calculate number of extents (probably better to pass nb) */ + /* 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) @@ -1234,26 +1221,31 @@ static int osd_declare_write_commit(const struct lu_env *env, /* Check if a block is allocated or not */ static int osd_write_commit(const struct lu_env *env, struct dt_object *dt, - struct niobuf_local *lnb, int npages, - struct thandle *thandle) + struct niobuf_local *lnb, int npages, + struct thandle *thandle, __u64 user_size) { - struct osd_thread_info *oti = osd_oti_get(env); - 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 isize; - int rc = 0, i; + struct osd_thread_info *oti = osd_oti_get(env); + 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; - LASSERT(inode); + LASSERT(inode); rc = osd_init_iobuf(osd, iobuf, 1, npages); if (unlikely(rc != 0)) RETURN(rc); - isize = i_size_read(inode); + 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++) { + for (i = 0; i < npages; i++) { if (lnb[i].lnb_rc == -ENOSPC && (lnb[i].lnb_flags & OBD_BRW_MAPPED)) { /* Allow the write to proceed if overwriting an @@ -1273,8 +1265,8 @@ static int osd_write_commit(const struct lu_env *env, struct dt_object *dt, LASSERT(PageLocked(lnb[i].lnb_page)); LASSERT(!PageWriteback(lnb[i].lnb_page)); - if (lnb[i].lnb_file_offset + lnb[i].lnb_len > isize) - isize = lnb[i].lnb_file_offset + lnb[i].lnb_len; + 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 @@ -1286,26 +1278,29 @@ static int osd_write_commit(const struct lu_env *env, struct dt_object *dt, SetPageUptodate(lnb[i].lnb_page); 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) { + 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); - } else { - /* no pages to write, no transno is needed */ - thandle->th_local = 1; - } + } else { + /* no pages to write, no transno is needed */ + thandle->th_local = 1; + } 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; + 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 { @@ -1326,9 +1321,11 @@ static int osd_write_commit(const struct lu_env *env, struct dt_object *dt, for (i = 0; i < npages; i++) { if (lnb[i].lnb_page == NULL) continue; - LASSERT(PageLocked(lnb[i].lnb_page)); - generic_error_remove_page(inode->i_mapping, - lnb[i].lnb_page); + if (!PagePrivate2(lnb[i].lnb_page)) { + LASSERT(PageLocked(lnb[i].lnb_page)); + generic_error_remove_page(inode->i_mapping, + lnb[i].lnb_page); + } } } @@ -1411,7 +1408,8 @@ static int osd_read_prep(const struct lu_env *env, struct dt_object *dt, /* early release to let others read data during the bulk */ for (i = 0; i < iobuf->dr_npages; i++) { LASSERT(PageLocked(iobuf->dr_pages[i])); - unlock_page(iobuf->dr_pages[i]); + if (!PagePrivate2(iobuf->dr_pages[i])) + unlock_page(iobuf->dr_pages[i]); } } @@ -1498,19 +1496,27 @@ int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs) static ssize_t osd_read(const struct lu_env *env, struct dt_object *dt, struct lu_buf *buf, loff_t *pos) { - struct inode *inode = osd_dt_obj(dt)->oo_inode; - int rc; + struct inode *inode = osd_dt_obj(dt)->oo_inode; + int rc; - /* Read small symlink from inode body as we need to maintain correct - * on-disk symlinks for ldiskfs. - */ - if (S_ISLNK(dt->do_lu.lo_header->loh_attr) && - (buf->lb_len < sizeof(LDISKFS_I(inode)->i_data))) - rc = osd_ldiskfs_readlink(inode, buf->lb_buf, buf->lb_len); - else - rc = osd_ldiskfs_read(inode, buf->lb_buf, buf->lb_len, pos); + /* Read small symlink from inode body as we need to maintain correct + * on-disk symlinks for ldiskfs. + */ + if (S_ISLNK(dt->do_lu.lo_header->loh_attr)) { + loff_t size = i_size_read(inode); + + if (buf->lb_len < size) + return -EOVERFLOW; - return rc; + if (size < sizeof(LDISKFS_I(inode)->i_data)) + rc = osd_ldiskfs_readlink(inode, buf->lb_buf, size); + else + rc = osd_ldiskfs_read(inode, buf->lb_buf, size, pos); + } else { + rc = osd_ldiskfs_read(inode, buf->lb_buf, buf->lb_len, pos); + } + + return rc; } static inline int osd_extents_enabled(struct super_block *sb, @@ -1588,10 +1594,10 @@ static ssize_t osd_declare_write(const struct lu_env *env, struct dt_object *dt, ENTRY; LASSERT(buf != NULL); - LASSERT(handle != NULL); + LASSERT(handle != NULL); - oh = container_of0(handle, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle == NULL); + oh = container_of(handle, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle == NULL); size = buf->lb_len; bits = sb->s_blocksize_bits; @@ -1691,11 +1697,11 @@ static int osd_ldiskfs_write_record(struct dt_object *dt, void *buf, loff_t new_size = i_size_read(inode); unsigned long block; int blocksize = 1 << inode->i_blkbits; + struct ldiskfs_inode_info *ei = LDISKFS_I(inode); int err = 0; int size; int boffs; int dirty_inode = 0; - struct ldiskfs_inode_info *ei = LDISKFS_I(inode); bool create, sparse, sync = false; if (write_NUL) { @@ -1708,6 +1714,8 @@ static int osd_ldiskfs_write_record(struct dt_object *dt, void *buf, ++bufsize; } + dirty_inode = test_and_set_bit(LDISKFS_INODE_JOURNAL_DATA, &ei->i_flags); + /* sparse checking is racy, but sparse is very rare case, leave as is */ sparse = (new_size > 0 && (inode->i_blocks >> (inode->i_blkbits - 9)) < ((new_size - 1) >> inode->i_blkbits) + 1); @@ -1811,14 +1819,14 @@ static int osd_ldiskfs_write_record(struct dt_object *dt, void *buf, spin_lock(&inode->i_lock); if (new_size > i_size_read(inode)) i_size_write(inode, new_size); - if (i_size_read(inode) > LDISKFS_I(inode)->i_disksize) { - LDISKFS_I(inode)->i_disksize = i_size_read(inode); + if (i_size_read(inode) > ei->i_disksize) { + ei->i_disksize = i_size_read(inode); dirty_inode = 1; } spin_unlock(&inode->i_lock); - if (dirty_inode) - osd_dirty_inode(inode, I_DIRTY_DATASYNC); } + if (dirty_inode) + osd_dirty_inode(inode, I_DIRTY_DATASYNC); if (err == 0) *offs = offset; @@ -1866,6 +1874,69 @@ static ssize_t osd_write(const struct lu_env *env, struct dt_object *dt, return result; } +static int osd_declare_fallocate(const struct lu_env *env, + struct dt_object *dt, struct thandle *th) +{ + struct osd_thandle *oh; + struct inode *inode; + int rc; + ENTRY; + + LASSERT(th); + oh = container_of(th, struct osd_thandle, ot_super); + + osd_trans_declare_op(env, oh, OSD_OT_PREALLOC, + osd_dto_credits_noquota[DTO_WRITE_BLOCK]); + inode = osd_dt_obj(dt)->oo_inode; + LASSERT(inode); + + 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); + 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_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + 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); + dquot_initialize(inode); + + LASSERT(th); + + osd_trans_exec_op(env, th, OSD_OT_PREALLOC); + + /* + * Because f_op->fallocate() does not have an inode arg + */ + 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); + + RETURN(rc); +} + static int osd_declare_punch(const struct lu_env *env, struct dt_object *dt, __u64 start, __u64 end, struct thandle *th) { @@ -1913,7 +1984,6 @@ static int osd_punch(const struct lu_env *env, struct dt_object *dt, bool grow = false; ENTRY; - LASSERT(end == OBD_OBJECT_EOF); LASSERT(dt_object_exists(dt)); LASSERT(osd_invariant(obj)); LASSERT(inode != NULL); @@ -1948,10 +2018,12 @@ static int osd_punch(const struct lu_env *env, struct dt_object *dt, GOTO(out, rc); } + inode_lock(inode); /* add to orphan list to ensure truncate completion * if this transaction succeed. ldiskfs_truncate() * will take the inode out of the list */ rc = ldiskfs_orphan_add(oh->ot_handle, inode); + inode_unlock(inode); if (rc != 0) GOTO(out, rc); @@ -2086,6 +2158,8 @@ const struct dt_body_operations osd_body_ops = { .dbo_punch = osd_punch, .dbo_fiemap_get = osd_fiemap_get, .dbo_ladvise = osd_ladvise, + .dbo_declare_fallocate = osd_declare_fallocate, + .dbo_fallocate = osd_fallocate, }; /** @@ -2130,13 +2204,14 @@ int osd_trunc_lock(struct osd_object *obj, struct osd_thandle *oh, bool shared) else down_write(&obj->oo_ext_idx_sem); al->tl_shared = shared; + lu_object_get(&obj->oo_dt.do_lu); list_add(&al->tl_list, &oh->ot_trunc_locks); return 0; } -void osd_trunc_unlock_all(struct list_head *list) +void osd_trunc_unlock_all(const struct lu_env *env, struct list_head *list) { struct osd_access_lock *al, *tmp; list_for_each_entry_safe(al, tmp, list, tl_list) { @@ -2144,6 +2219,7 @@ void osd_trunc_unlock_all(struct list_head *list) up_read(&al->tl_obj->oo_ext_idx_sem); else up_write(&al->tl_obj->oo_ext_idx_sem); + osd_object_put(env, al->tl_obj); list_del(&al->tl_list); OBD_FREE_PTR(al); }