4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/osd/osd_compat.c
34 * on-disk structure for managing /O
36 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
39 /* prerequisite for linux/xattr.h */
40 #include <linux/types.h>
41 /* prerequisite for linux/xattr.h */
43 /* XATTR_{REPLACE,CREATE} */
44 #include <linux/xattr.h>
47 * struct OBD_{ALLOC,FREE}*()
50 #include <obd_support.h>
52 #include "osd_internal.h"
55 static void osd_push_ctxt(const struct osd_device *dev,
56 struct lvfs_run_ctxt *newctxt,
57 struct lvfs_run_ctxt *save)
59 OBD_SET_CTXT_MAGIC(newctxt);
60 newctxt->pwdmnt = dev->od_mnt;
61 newctxt->pwd = dev->od_mnt->mnt_root;
62 newctxt->fs = get_ds();
63 newctxt->umask = current_umask();
66 push_ctxt(save, newctxt);
69 /* utility to make a directory */
70 static struct dentry *simple_mkdir(struct dentry *dir, struct vfsmount *mnt,
71 const char *name, int mode, int fix)
73 struct dentry *dchild;
77 // ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
78 CDEBUG(D_INODE, "creating directory %.*s\n", (int)strlen(name), name);
79 dchild = ll_lookup_one_len(name, dir, strlen(name));
83 if (dchild->d_inode) {
84 int old_mode = dchild->d_inode->i_mode;
85 if (!S_ISDIR(old_mode)) {
86 CERROR("found %s (%lu/%u) is mode %o\n", name,
87 dchild->d_inode->i_ino,
88 dchild->d_inode->i_generation, old_mode);
89 GOTO(out_err, err = -ENOTDIR);
92 /* Fixup directory permissions if necessary */
93 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
95 "fixing permissions on %s from %o to %o\n",
96 name, old_mode, mode);
97 dchild->d_inode->i_mode = (mode & S_IALLUGO) |
98 (old_mode & ~S_IALLUGO);
99 mark_inode_dirty(dchild->d_inode);
101 GOTO(out_up, dchild);
104 err = vfs_mkdir(dir->d_inode, dchild, mode);
112 dchild = ERR_PTR(err);
117 static int osd_last_rcvd_subdir_count(struct osd_device *osd)
119 struct lr_server_data lsd;
120 struct dentry *dlast;
123 int count = OBJ_SUBDIR_COUNT;
127 dlast = ll_lookup_one_len(LAST_RCVD, osd_sb(osd)->s_root,
130 return PTR_ERR(dlast);
131 else if (dlast->d_inode == NULL)
135 rc = osd_ldiskfs_read(dlast->d_inode, &lsd, sizeof(lsd), &off);
136 if (rc == sizeof(lsd)) {
137 CDEBUG(D_INFO, "read last_rcvd header, uuid = %s, "
138 "subdir count = %d\n", lsd.lsd_uuid,
139 lsd.lsd_subdir_count);
140 if (le16_to_cpu(lsd.lsd_subdir_count) > 0)
141 count = le16_to_cpu(lsd.lsd_subdir_count);
142 } else if (rc != 0) {
143 CERROR("Can't read last_rcvd file, rc = %d\n", rc);
155 static const char remote_parent_dir[] = "REMOTE_PARENT_DIR";
156 static int osd_mdt_init(const struct lu_env *env, struct osd_device *dev)
158 struct lvfs_run_ctxt new;
159 struct lvfs_run_ctxt save;
160 struct dentry *parent;
161 struct osd_mdobj_map *omm;
163 struct osd_thread_info *info = osd_oti_get(env);
164 struct lu_fid *fid = &info->oti_fid3;
168 OBD_ALLOC_PTR(dev->od_mdt_map);
169 if (dev->od_mdt_map == NULL)
172 omm = dev->od_mdt_map;
174 parent = osd_sb(dev)->s_root;
175 osd_push_ctxt(dev, &new, &save);
177 d = simple_mkdir(parent, dev->od_mnt, remote_parent_dir,
180 GOTO(cleanup, rc = PTR_ERR(d));
182 omm->omm_remote_parent = d;
184 /* Set LMA for remote parent inode */
185 lu_local_obj_fid(fid, REMOTE_PARENT_DIR_OID);
186 rc = osd_ea_fid_set(info, d->d_inode, fid, LMAC_NOT_IN_OI, 0);
191 pop_ctxt(&save, &new);
193 if (omm->omm_remote_parent != NULL)
194 dput(omm->omm_remote_parent);
196 dev->od_mdt_map = NULL;
201 static void osd_mdt_fini(struct osd_device *osd)
203 struct osd_mdobj_map *omm = osd->od_mdt_map;
208 if (omm->omm_remote_parent)
209 dput(omm->omm_remote_parent);
212 osd->od_ost_map = NULL;
215 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
216 struct osd_object *obj, struct osd_thandle *oh)
218 struct osd_mdobj_map *omm = osd->od_mdt_map;
219 struct osd_thread_info *oti = osd_oti_get(env);
220 struct lustre_mdt_attrs *lma = &oti->oti_ost_attrs.loa_lma;
221 char *name = oti->oti_name;
222 struct osd_thread_info *info = osd_oti_get(env);
223 struct dentry *dentry;
224 struct dentry *parent;
227 /* Set REMOTE_PARENT in lma, so other process like unlink or lfsck
228 * can identify this object quickly */
229 rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry,
230 &oti->oti_ost_attrs);
234 lma->lma_incompat |= LMAI_REMOTE_PARENT;
235 lustre_lma_swab(lma);
236 rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
237 sizeof(*lma), XATTR_REPLACE);
241 parent = omm->omm_remote_parent;
242 sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
243 dentry = osd_child_dentry_by_inode(env, parent->d_inode,
245 mutex_lock(&parent->d_inode->i_mutex);
246 rc = osd_ldiskfs_add_entry(info, osd, oh->ot_handle, dentry,
247 obj->oo_inode, NULL);
248 CDEBUG(D_INODE, "%s: add %s:%lu to remote parent %lu.\n", osd_name(osd),
249 name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
250 ldiskfs_inc_count(oh->ot_handle, parent->d_inode);
251 mark_inode_dirty(parent->d_inode);
252 mutex_unlock(&parent->d_inode->i_mutex);
256 int osd_delete_from_remote_parent(const struct lu_env *env,
257 struct osd_device *osd,
258 struct osd_object *obj,
259 struct osd_thandle *oh)
261 struct osd_mdobj_map *omm = osd->od_mdt_map;
262 struct osd_thread_info *oti = osd_oti_get(env);
263 struct lustre_mdt_attrs *lma = &oti->oti_ost_attrs.loa_lma;
264 char *name = oti->oti_name;
265 struct dentry *dentry;
266 struct dentry *parent;
267 struct ldiskfs_dir_entry_2 *de;
268 struct buffer_head *bh;
271 /* Check lma to see whether it is remote object */
272 rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry,
273 &oti->oti_ost_attrs);
275 /* No LMA if the directory is created before 2.0 */
281 if (likely(!(lma->lma_incompat & LMAI_REMOTE_PARENT)))
284 parent = omm->omm_remote_parent;
285 sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
286 dentry = osd_child_dentry_by_inode(env, parent->d_inode,
288 mutex_lock(&parent->d_inode->i_mutex);
289 bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
292 mutex_unlock(&parent->d_inode->i_mutex);
295 CDEBUG(D_INODE, "%s: el %s:%lu to remote parent %lu.\n", osd_name(osd),
296 name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
297 rc = ldiskfs_delete_entry(oh->ot_handle, parent->d_inode, de, bh);
298 ldiskfs_dec_count(oh->ot_handle, parent->d_inode);
299 mark_inode_dirty(parent->d_inode);
300 mutex_unlock(&parent->d_inode->i_mutex);
303 /* Get rid of REMOTE_PARENT flag from incompat */
304 lma->lma_incompat &= ~LMAI_REMOTE_PARENT;
305 lustre_lma_swab(lma);
306 rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
307 sizeof(*lma), XATTR_REPLACE);
311 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
312 struct osd_device *osd,
313 const struct lu_fid *fid,
314 struct osd_inode_id *id)
316 struct osd_mdobj_map *omm = osd->od_mdt_map;
317 char *name = oti->oti_name;
318 struct dentry *parent;
319 struct dentry *dentry;
320 struct ldiskfs_dir_entry_2 *de;
321 struct buffer_head *bh;
325 parent = omm->omm_remote_parent;
326 sprintf(name, DFID_NOBRACE, PFID(fid));
327 dentry = osd_child_dentry_by_inode(oti->oti_env, parent->d_inode,
329 mutex_lock(&parent->d_inode->i_mutex);
330 bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
337 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
339 inode = osd_iget(oti, osd, id);
349 mutex_unlock(&parent->d_inode->i_mutex);
351 osd_add_oi_cache(oti, osd, id, fid);
356 * directory structure on legacy OST:
358 * O/<seq>/d0-31/<objid>
365 static int osd_ost_init(const struct lu_env *env, struct osd_device *dev)
367 struct lvfs_run_ctxt new;
368 struct lvfs_run_ctxt save;
369 struct dentry *rootd = osd_sb(dev)->s_root;
371 struct osd_thread_info *info = osd_oti_get(env);
373 struct lu_fid *fid = &info->oti_fid3;
377 OBD_ALLOC_PTR(dev->od_ost_map);
378 if (dev->od_ost_map == NULL)
381 /* to get subdir count from last_rcvd */
382 rc = osd_last_rcvd_subdir_count(dev);
384 GOTO(cleanup_alloc, rc);
386 dev->od_ost_map->om_subdir_count = rc;
389 INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list);
390 rwlock_init(&dev->od_ost_map->om_seq_list_lock);
391 mutex_init(&dev->od_ost_map->om_dir_init_mutex);
393 osd_push_ctxt(dev, &new, &save);
395 d = ll_lookup_one_len("O", rootd, strlen("O"));
397 GOTO(cleanup_ctxt, rc = PTR_ERR(d));
398 if (d->d_inode == NULL) {
400 /* The lookup() may be called again inside simple_mkdir().
401 * Since the repeated lookup() only be called for "/O" at
402 * mount time, it will not affect the whole performance. */
403 d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1);
405 GOTO(cleanup_ctxt, rc = PTR_ERR(d));
407 /* It is quite probably that the device is new formatted. */
408 dev->od_maybe_new = 1;
412 dev->od_ost_map->om_root = d;
414 /* 'What the @fid is' is not imporatant, because the object
415 * has no OI mapping, and only is visible inside the OSD.*/
416 lu_igif_build(fid, inode->i_ino, inode->i_generation);
417 rc = osd_ea_fid_set(info, inode, fid,
418 LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
420 GOTO(cleanup_dentry, rc);
422 pop_ctxt(&save, &new);
428 pop_ctxt(&save, &new);
430 OBD_FREE_PTR(dev->od_ost_map);
434 static void osd_seq_free(struct osd_obj_seq *osd_seq)
438 if (osd_seq->oos_dirs) {
439 for (j = 0; j < osd_seq->oos_subdir_count; j++) {
440 if (osd_seq->oos_dirs[j])
441 dput(osd_seq->oos_dirs[j]);
443 OBD_FREE(osd_seq->oos_dirs,
444 sizeof(struct dentry *) * osd_seq->oos_subdir_count);
447 if (osd_seq->oos_root)
448 dput(osd_seq->oos_root);
450 OBD_FREE_PTR(osd_seq);
453 static void osd_ost_fini(struct osd_device *osd)
455 struct osd_obj_seq *osd_seq;
456 struct osd_obj_seq *tmp;
457 struct osd_obj_map *map = osd->od_ost_map;
463 write_lock(&map->om_seq_list_lock);
464 list_for_each_entry_safe(osd_seq, tmp, &map->om_seq_list,
466 list_del_init(&osd_seq->oos_seq_list);
467 write_unlock(&map->om_seq_list_lock);
468 osd_seq_free(osd_seq);
469 write_lock(&map->om_seq_list_lock);
471 write_unlock(&map->om_seq_list_lock);
475 osd->od_ost_map = NULL;
479 int osd_obj_map_init(const struct lu_env *env, struct osd_device *dev)
484 /* prepare structures for OST */
485 rc = osd_ost_init(env, dev);
489 /* prepare structures for MDS */
490 rc = osd_mdt_init(env, dev);
497 static struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, u64 seq)
499 struct osd_obj_seq *osd_seq;
501 list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) {
502 if (osd_seq->oos_seq == seq)
508 static struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, u64 seq)
510 struct osd_obj_seq *osd_seq;
512 read_lock(&map->om_seq_list_lock);
513 osd_seq = osd_seq_find_locked(map, seq);
514 read_unlock(&map->om_seq_list_lock);
518 void osd_obj_map_fini(struct osd_device *dev)
525 * Update the specified OI mapping.
527 * \retval 1, changed nothing
528 * \retval 0, changed successfully
529 * \retval -ve, on error
531 static int osd_obj_update_entry(struct osd_thread_info *info,
532 struct osd_device *osd,
533 struct dentry *dir, const char *name,
534 const struct lu_fid *fid,
535 const struct osd_inode_id *id,
538 struct inode *parent = dir->d_inode;
539 struct dentry *child;
540 struct ldiskfs_dir_entry_2 *de;
541 struct buffer_head *bh;
543 struct dentry *dentry = &info->oti_obj_dentry;
544 struct osd_inode_id *oi_id = &info->oti_id3;
545 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
546 struct lu_fid *oi_fid = &lma->lma_self_fid;
551 LASSERT(th->h_transaction != NULL);
553 child = &info->oti_child_dentry;
554 child->d_parent = dir;
555 child->d_name.hash = 0;
556 child->d_name.name = name;
557 child->d_name.len = strlen(name);
559 ll_vfs_dq_init(parent);
560 mutex_lock(&parent->i_mutex);
561 bh = osd_ldiskfs_find_entry(parent, &child->d_name, &de, NULL, NULL);
563 GOTO(out, rc = PTR_ERR(bh));
565 if (le32_to_cpu(de->inode) == id->oii_ino)
568 osd_id_gen(oi_id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
569 inode = osd_iget(info, osd, oi_id);
572 if (rc == -ENOENT || rc == -ESTALE)
577 /* The EA inode should NOT be in OI, old OI scrub may added
578 * such OI mapping by wrong, replace it. */
579 if (unlikely(osd_is_ea_inode(inode))) {
584 rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
585 if (rc == -ENODATA) {
586 rc = osd_get_idif(info, inode, dentry, oi_fid);
587 if (rc > 0 || rc == -ENODATA) {
597 /* If the OST-object has neither FID-in-LMA nor FID-in-ff, it is
598 * either a crashed object or a uninitialized one. Replace it. */
599 if (oi_fid != NULL && lu_fid_eq(fid, oi_fid)) {
600 CERROR("%s: the FID "DFID" is used by two objects: "
601 "%u/%u %u/%u\n", osd_name(osd), PFID(fid),
602 oi_id->oii_ino, oi_id->oii_gen,
603 id->oii_ino, id->oii_gen);
604 GOTO(out, rc = -EEXIST);
607 if (fid_is_idif(fid) && oi_fid != NULL && fid_is_idif(oi_fid)) {
608 __u32 idx1 = fid_idif_ost_idx(fid);
609 __u32 idx2 = fid_idif_ost_idx(oi_fid);
610 struct ost_id *ostid = &info->oti_ostid;
611 struct lu_fid *tfid = &info->oti_fid3;
613 LASSERTF(idx1 == 0 || idx1 == osd->od_index,
614 "invalid given FID "DFID", not match the "
615 "device index %u\n", PFID(fid), osd->od_index);
618 if (idx1 == 0 && idx2 == osd->od_index) {
619 fid_to_ostid(fid, ostid);
620 ostid_to_fid(tfid, ostid, idx2);
621 if (lu_fid_eq(tfid, oi_fid)) {
622 CERROR("%s: the FID "DFID" is used by "
623 "two objects(2): %u/%u %u/%u\n",
624 osd_name(osd), PFID(fid),
625 oi_id->oii_ino, oi_id->oii_gen,
626 id->oii_ino, id->oii_gen);
628 GOTO(out, rc = -EEXIST);
630 } else if (idx2 == 0 && idx1 == osd->od_index) {
631 fid_to_ostid(oi_fid, ostid);
632 ostid_to_fid(tfid, ostid, idx1);
633 if (lu_fid_eq(tfid, fid)) {
634 CERROR("%s: the FID "DFID" is used by "
635 "two objects(2): %u/%u %u/%u\n",
636 osd_name(osd), PFID(fid),
637 oi_id->oii_ino, oi_id->oii_gen,
638 id->oii_ino, id->oii_gen);
640 GOTO(out, rc = -EEXIST);
647 /* There may be temporary inconsistency: On one hand, the new
648 * object may be referenced by multiple entries, which is out
649 * of our control unless we traverse the whole /O completely,
650 * which is non-flat order and inefficient, should be avoided;
651 * On the other hand, the old object may become orphan if it
652 * is still valid. Since it was referenced by an invalid entry,
653 * making it as invisible temporary may be not worse. OI scrub
654 * will process it later. */
655 rc = ldiskfs_journal_get_write_access(th, bh);
659 de->inode = cpu_to_le32(id->oii_ino);
660 rc = ldiskfs_handle_dirty_metadata(th, NULL, bh);
667 mutex_unlock(&parent->i_mutex);
671 static int osd_obj_del_entry(struct osd_thread_info *info,
672 struct osd_device *osd,
673 struct dentry *dird, char *name,
676 struct ldiskfs_dir_entry_2 *de;
677 struct buffer_head *bh;
678 struct dentry *child;
679 struct inode *dir = dird->d_inode;
684 LASSERT(th->h_transaction != NULL);
687 child = &info->oti_child_dentry;
688 child->d_name.hash = 0;
689 child->d_name.name = name;
690 child->d_name.len = strlen(name);
691 child->d_parent = dird;
692 child->d_inode = NULL;
695 mutex_lock(&dir->i_mutex);
696 bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
700 rc = ldiskfs_delete_entry(th, dir, de, bh);
703 mutex_unlock(&dir->i_mutex);
708 static int osd_obj_add_entry(struct osd_thread_info *info,
709 struct osd_device *osd,
710 struct dentry *dir, char *name,
711 const struct osd_inode_id *id,
714 struct dentry *child;
720 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY))
724 LASSERT(th->h_transaction != NULL);
726 inode = info->oti_inode;
727 if (unlikely(inode == NULL)) {
728 struct ldiskfs_inode_info *lii;
732 inode = info->oti_inode = &lii->vfs_inode;
735 inode->i_sb = osd_sb(osd);
736 osd_id_to_inode(inode, id);
737 inode->i_mode = S_IFREG; /* for type in ldiskfs dir entry */
739 child = &info->oti_child_dentry;
740 child->d_name.hash = 0;
741 child->d_name.name = name;
742 child->d_name.len = strlen(name);
743 child->d_parent = dir;
744 child->d_inode = inode;
746 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
749 ll_vfs_dq_init(dir->d_inode);
750 mutex_lock(&dir->d_inode->i_mutex);
751 rc = osd_ldiskfs_add_entry(info, osd, th, child, inode, NULL);
752 mutex_unlock(&dir->d_inode->i_mutex);
758 * Use %llu for legacy OST sequences, but use %llx for new
759 * sequences names, so that the O/{seq}/dN/{oid} more closely
760 * follows the DFID/PFID format. This makes it easier to map from
761 * debug messages to objects in the future, and the legacy space
762 * of FID_SEQ_OST_MDT0 will be unused in the future.
764 static inline void osd_seq_name(char *seq_name, size_t name_size, u64 seq)
766 snprintf(seq_name, name_size,
767 (fid_seq_is_rsvd(seq) ||
768 fid_seq_is_mdt0(seq)) ? "%llu" : "%llx",
769 fid_seq_is_idif(seq) ? 0 : seq);
772 static inline void osd_oid_name(char *name, size_t name_size,
773 const struct lu_fid *fid, u64 id)
775 snprintf(name, name_size,
776 (fid_seq_is_rsvd(fid_seq(fid)) ||
777 fid_seq_is_mdt0(fid_seq(fid)) ||
778 fid_seq_is_idif(fid_seq(fid))) ? "%llu" : "%llx", id);
781 /* external locking is required */
782 static int osd_seq_load_locked(struct osd_thread_info *info,
783 struct osd_device *osd,
784 struct osd_obj_seq *osd_seq)
786 struct osd_obj_map *map = osd->od_ost_map;
787 struct dentry *seq_dir;
789 struct lu_fid *fid = &info->oti_fid3;
795 if (osd_seq->oos_root != NULL)
799 LASSERT(map->om_root);
801 osd_seq_name(dir_name, sizeof(dir_name), osd_seq->oos_seq);
803 seq_dir = simple_mkdir(map->om_root, osd->od_mnt, dir_name, 0755, 1);
805 GOTO(out_err, rc = PTR_ERR(seq_dir));
806 else if (seq_dir->d_inode == NULL)
807 GOTO(out_put, rc = -EFAULT);
809 inode = seq_dir->d_inode;
810 osd_seq->oos_root = seq_dir;
812 /* 'What the @fid is' is not imporatant, because the object
813 * has no OI mapping, and only is visible inside the OSD.*/
814 lu_igif_build(fid, inode->i_ino, inode->i_generation);
815 rc = osd_ea_fid_set(info, inode, fid,
816 LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
820 LASSERT(osd_seq->oos_dirs == NULL);
821 OBD_ALLOC(osd_seq->oos_dirs,
822 sizeof(seq_dir) * osd_seq->oos_subdir_count);
823 if (osd_seq->oos_dirs == NULL)
824 GOTO(out_put, rc = -ENOMEM);
826 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
829 snprintf(dir_name, sizeof(dir_name), "d%u", i);
830 dir = simple_mkdir(osd_seq->oos_root, osd->od_mnt, dir_name,
833 GOTO(out_free, rc = PTR_ERR(dir));
834 } else if (dir->d_inode == NULL) {
836 GOTO(out_free, rc = -EFAULT);
839 inode = dir->d_inode;
840 osd_seq->oos_dirs[i] = dir;
842 /* 'What the @fid is' is not imporatant, because the object
843 * has no OI mapping, and only is visible inside the OSD.*/
844 lu_igif_build(fid, inode->i_ino, inode->i_generation);
845 rc = osd_ea_fid_set(info, inode, fid,
846 LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
853 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
854 if (osd_seq->oos_dirs[i] != NULL)
855 dput(osd_seq->oos_dirs[i]);
857 OBD_FREE(osd_seq->oos_dirs,
858 sizeof(seq_dir) * osd_seq->oos_subdir_count);
861 osd_seq->oos_root = NULL;
867 static struct osd_obj_seq *osd_seq_load(struct osd_thread_info *info,
868 struct osd_device *osd, u64 seq)
870 struct osd_obj_map *map;
871 struct osd_obj_seq *osd_seq;
875 map = osd->od_ost_map;
877 LASSERT(map->om_root);
879 osd_seq = osd_seq_find(map, seq);
880 if (likely(osd_seq != NULL))
883 /* Serializing init process */
884 mutex_lock(&map->om_dir_init_mutex);
886 /* Check whether the seq has been added */
887 read_lock(&map->om_seq_list_lock);
888 osd_seq = osd_seq_find_locked(map, seq);
889 if (osd_seq != NULL) {
890 read_unlock(&map->om_seq_list_lock);
891 GOTO(cleanup, rc = 0);
893 read_unlock(&map->om_seq_list_lock);
895 OBD_ALLOC_PTR(osd_seq);
897 GOTO(cleanup, rc = -ENOMEM);
899 INIT_LIST_HEAD(&osd_seq->oos_seq_list);
900 osd_seq->oos_seq = seq;
901 /* Init subdir count to be 32, but each seq can have
902 * different subdir count */
903 osd_seq->oos_subdir_count = map->om_subdir_count;
904 rc = osd_seq_load_locked(info, osd, osd_seq);
908 write_lock(&map->om_seq_list_lock);
909 list_add(&osd_seq->oos_seq_list, &map->om_seq_list);
910 write_unlock(&map->om_seq_list_lock);
913 mutex_unlock(&map->om_dir_init_mutex);
916 OBD_FREE_PTR(osd_seq);
923 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev,
924 const struct lu_fid *fid, struct osd_inode_id *id)
926 struct osd_obj_map *map;
927 struct osd_obj_seq *osd_seq;
928 struct dentry *d_seq;
929 struct dentry *child;
930 struct ost_id *ostid = &info->oti_ostid;
933 struct ldiskfs_dir_entry_2 *de;
934 struct buffer_head *bh;
939 /* on the very first lookup we find and open directories */
941 map = dev->od_ost_map;
943 LASSERT(map->om_root);
945 fid_to_ostid(fid, ostid);
946 osd_seq = osd_seq_load(info, dev, ostid_seq(ostid));
948 RETURN(PTR_ERR(osd_seq));
950 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
951 d_seq = osd_seq->oos_dirs[dirn];
954 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
956 child = &info->oti_child_dentry;
957 child->d_parent = d_seq;
958 child->d_name.hash = 0;
959 child->d_name.name = name;
960 /* XXX: we can use rc from sprintf() instead of strlen() */
961 child->d_name.len = strlen(name);
963 dir = d_seq->d_inode;
964 mutex_lock(&dir->i_mutex);
965 bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
966 mutex_unlock(&dir->i_mutex);
971 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
974 inode = osd_iget(info, dev, id);
976 int rc = PTR_ERR(inode);
978 RETURN(rc == -ENOENT ? -ESTALE : rc);
985 int osd_obj_map_insert(struct osd_thread_info *info,
986 struct osd_device *osd,
987 const struct lu_fid *fid,
988 const struct osd_inode_id *id,
991 struct osd_obj_map *map;
992 struct osd_obj_seq *osd_seq;
994 struct ost_id *ostid = &info->oti_ostid;
1000 map = osd->od_ost_map;
1003 /* map fid to seq:objid */
1004 fid_to_ostid(fid, ostid);
1006 oid = ostid_id(ostid);
1007 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1008 if (IS_ERR(osd_seq))
1009 RETURN(PTR_ERR(osd_seq));
1011 dirn = oid & (osd_seq->oos_subdir_count - 1);
1012 d = osd_seq->oos_dirs[dirn];
1015 osd_oid_name(name, sizeof(name), fid, oid);
1018 rc = osd_obj_add_entry(info, osd, d, name, id, th);
1019 if (rc == -EEXIST) {
1020 rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1021 if (unlikely(rc == -ENOENT))
1024 if (unlikely(rc == 1))
1031 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
1032 const struct lu_fid *fid, handle_t *th)
1034 struct osd_obj_map *map;
1035 struct osd_obj_seq *osd_seq;
1037 struct ost_id *ostid = &info->oti_ostid;
1042 map = osd->od_ost_map;
1045 /* map fid to seq:objid */
1046 fid_to_ostid(fid, ostid);
1048 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1049 if (IS_ERR(osd_seq))
1050 GOTO(cleanup, rc = PTR_ERR(osd_seq));
1052 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1053 d = osd_seq->oos_dirs[dirn];
1056 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1057 rc = osd_obj_del_entry(info, osd, d, name, th);
1062 int osd_obj_map_update(struct osd_thread_info *info,
1063 struct osd_device *osd,
1064 const struct lu_fid *fid,
1065 const struct osd_inode_id *id,
1068 struct osd_obj_seq *osd_seq;
1070 struct ost_id *ostid = &info->oti_ostid;
1075 fid_to_ostid(fid, ostid);
1076 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1077 if (IS_ERR(osd_seq))
1078 RETURN(PTR_ERR(osd_seq));
1080 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1081 d = osd_seq->oos_dirs[dirn];
1084 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1085 rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1090 int osd_obj_map_recover(struct osd_thread_info *info,
1091 struct osd_device *osd,
1092 struct inode *src_parent,
1093 struct dentry *src_child,
1094 const struct lu_fid *fid)
1096 struct osd_obj_seq *osd_seq;
1097 struct dentry *tgt_parent;
1098 struct dentry *tgt_child = &info->oti_child_dentry;
1100 struct inode *inode = src_child->d_inode;
1101 struct ost_id *ostid = &info->oti_ostid;
1103 struct ldiskfs_dir_entry_2 *de;
1104 struct buffer_head *bh;
1110 if (fid_is_last_id(fid)) {
1111 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1112 if (IS_ERR(osd_seq))
1113 RETURN(PTR_ERR(osd_seq));
1115 tgt_parent = osd_seq->oos_root;
1116 tgt_child->d_name.name = "LAST_ID";
1117 tgt_child->d_name.len = strlen("LAST_ID");
1119 fid_to_ostid(fid, ostid);
1120 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1121 if (IS_ERR(osd_seq))
1122 RETURN(PTR_ERR(osd_seq));
1124 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1125 tgt_parent = osd_seq->oos_dirs[dirn];
1126 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1127 tgt_child->d_name.name = name;
1128 tgt_child->d_name.len = strlen(name);
1130 LASSERT(tgt_parent != NULL);
1132 dir = tgt_parent->d_inode;
1133 tgt_child->d_name.hash = 0;
1134 tgt_child->d_parent = tgt_parent;
1135 tgt_child->d_inode = inode;
1137 /* The non-initialized src_child may be destroyed. */
1138 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
1139 osd_dto_credits_noquota[DTO_INDEX_DELETE] +
1140 osd_dto_credits_noquota[DTO_INDEX_INSERT] +
1141 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
1143 RETURN(PTR_ERR(jh));
1145 ll_vfs_dq_init(src_parent);
1146 ll_vfs_dq_init(dir);
1148 mutex_lock(&src_parent->i_mutex);
1149 mutex_lock(&dir->i_mutex);
1150 bh = osd_ldiskfs_find_entry(dir, &tgt_child->d_name, &de, NULL, NULL);
1152 /* XXX: If some other object occupied the same slot. And If such
1153 * inode is zero-sized and with SUID+SGID, then means it is
1154 * a new created one. Maybe we can remove it and insert the
1155 * original one back to the /O/<seq>/d<x>. But there are
1156 * something to be considered:
1158 * 1) The OST-object under /lost+found has crashed LMA.
1159 * So it should not conflict with the current one.
1161 * 2) There are race conditions that: someone may just want
1162 * to modify the current one. Even if the OI scrub takes
1163 * the object lock when remove the current one, it still
1164 * cause the modification to be lost becasue the target
1165 * has been removed when the RPC service thread waiting
1168 * So keep it there before we have suitable solution. */
1170 mutex_unlock(&dir->i_mutex);
1171 mutex_unlock(&src_parent->i_mutex);
1172 ldiskfs_journal_stop(jh);
1175 /* If the src object has never been modified, then remove it. */
1176 if (inode->i_size == 0 && inode->i_mode & S_ISUID &&
1177 inode->i_mode & S_ISGID) {
1178 rc = ll_vfs_unlink(src_parent, src_child);
1179 if (unlikely(rc == -ENOENT))
1185 bh = osd_ldiskfs_find_entry(src_parent, &src_child->d_name, &de,
1187 if (unlikely(IS_ERR(bh)))
1188 GOTO(unlock, rc = PTR_ERR(bh));
1190 rc = ldiskfs_delete_entry(jh, src_parent, de, bh);
1195 rc = osd_ldiskfs_add_entry(info, osd, jh, tgt_child, inode, NULL);
1200 mutex_unlock(&dir->i_mutex);
1201 mutex_unlock(&src_parent->i_mutex);
1202 ldiskfs_journal_stop(jh);
1206 static struct dentry *
1207 osd_object_spec_find(struct osd_thread_info *info, struct osd_device *osd,
1208 const struct lu_fid *fid, char **name)
1210 struct dentry *root = ERR_PTR(-ENOENT);
1212 if (fid_is_last_id(fid)) {
1213 struct osd_obj_seq *osd_seq;
1215 /* on creation of LAST_ID we create O/<seq> hierarchy */
1216 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1217 if (IS_ERR(osd_seq))
1218 RETURN((struct dentry *)osd_seq);
1221 root = osd_seq->oos_root;
1223 *name = osd_lf_fid2name(fid);
1225 CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid));
1226 else if ((*name)[0])
1227 root = osd_sb(osd)->s_root;
1233 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
1234 const struct lu_fid *fid, const struct osd_inode_id *id,
1237 struct dentry *root;
1242 root = osd_object_spec_find(info, osd, fid, &name);
1243 if (!IS_ERR(root)) {
1244 rc = osd_obj_update_entry(info, osd, root, name, fid, id, th);
1254 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
1255 const struct lu_fid *fid, const struct osd_inode_id *id,
1258 struct dentry *root;
1263 root = osd_object_spec_find(info, osd, fid, &name);
1264 if (!IS_ERR(root)) {
1265 rc = osd_obj_add_entry(info, osd, root, name, id, th);
1275 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
1276 const struct lu_fid *fid, struct osd_inode_id *id)
1278 struct dentry *root;
1279 struct dentry *dentry;
1280 struct inode *inode;
1285 if (fid_is_last_id(fid)) {
1286 struct osd_obj_seq *osd_seq;
1288 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1289 if (IS_ERR(osd_seq))
1290 RETURN(PTR_ERR(osd_seq));
1291 root = osd_seq->oos_root;
1294 root = osd_sb(osd)->s_root;
1295 name = osd_lf_fid2name(fid);
1296 if (name == NULL || strlen(name) == 0)
1300 dentry = ll_lookup_one_len(name, root, strlen(name));
1301 if (!IS_ERR(dentry)) {
1302 inode = dentry->d_inode;
1304 if (is_bad_inode(inode)) {
1307 osd_id_gen(id, inode->i_ino,
1308 inode->i_generation);
1312 /* if dentry is accessible after osd_compat_spec_insert it
1313 * will still contain NULL inode, so don't keep it in cache */
1314 d_invalidate(dentry);