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, 2017, 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 = KERNEL_DS;
63 newctxt->umask = current_umask();
66 push_ctxt(save, newctxt);
70 * osd_lookup_one_len_unlocked
72 * @dev: obd device we are searching
73 * @name: pathname component to lookup
74 * @base: base directory to lookup from
75 * @len: maximum length @len should be interpreted to
77 * This should be called without the parent
78 * i_mutex held, and will take the i_mutex itself.
80 struct dentry *osd_lookup_one_len_unlocked(struct osd_device *dev,
82 struct dentry *base, int len)
84 struct dentry *dchild;
86 inode_lock(base->d_inode);
87 dchild = lookup_one_len(name, base, len);
88 inode_unlock(base->d_inode);
93 if (dchild->d_inode && unlikely(is_bad_inode(dchild->d_inode))) {
94 CERROR("%s: bad inode returned %lu/%u: rc = -ENOENT\n",
95 osd_name(dev), dchild->d_inode->i_ino,
96 dchild->d_inode->i_generation);
98 dchild = ERR_PTR(-ENOENT);
105 * osd_ios_lookup_one_len - lookup single pathname component
107 * @dev: obd device we are searching
108 * @name: pathname component to lookup
109 * @base: base directory to lookup from
110 * @len: maximum length @len should be interpreted to
112 struct dentry *osd_ios_lookup_one_len(struct osd_device *dev, const char *name,
113 struct dentry *base, int len)
115 return osd_lookup_one_len_unlocked(dev, name, base, len);
118 /* utility to make a directory */
119 static struct dentry *
120 simple_mkdir(const struct lu_env *env, struct osd_device *osd,
121 struct dentry *dir, const struct lu_fid *fid,
122 const char *name, __u32 compat, int mode, bool *created)
124 struct osd_thread_info *info = osd_oti_get(env);
125 struct lu_fid *tfid = &info->oti_fid3;
127 struct dentry *dchild;
132 // ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
133 CDEBUG(D_INODE, "creating directory %.*s\n", (int)strlen(name), name);
134 dchild = osd_lookup_one_len_unlocked(osd, name, dir, strlen(name));
138 inode = dchild->d_inode;
140 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
141 int old_mode = inode->i_mode;
146 if (!S_ISDIR(old_mode)) {
147 CERROR("found %s (%lu/%u) is mode %o\n", name,
148 inode->i_ino, inode->i_generation, old_mode);
149 GOTO(out_err, err = -ENOTDIR);
152 if (unlikely(osd->od_dt_dev.dd_rdonly))
155 /* Fixup directory permissions if necessary */
156 if ((old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
158 "fixing permissions on %s from %o to %o\n",
159 name, old_mode, mode);
160 inode->i_mode = (mode & S_IALLUGO) |
161 (old_mode & ~S_IALLUGO);
162 mark_inode_dirty(inode);
165 err = osd_get_lma(info, inode, &info->oti_obj_dentry,
166 &info->oti_ost_attrs);
173 if ((fid && !lu_fid_eq(fid, &lma->lma_self_fid)) ||
174 lma->lma_compat != compat)
180 err = vfs_mkdir(dir->d_inode, dchild, mode);
184 inode = dchild->d_inode;
192 lu_igif_build(tfid, inode->i_ino, inode->i_generation);
193 err = osd_ea_fid_set(info, inode, tfid, compat, 0);
204 static int osd_last_rcvd_subdir_count(struct osd_device *osd)
206 struct lr_server_data lsd;
207 struct dentry *dlast;
210 int count = OBJ_SUBDIR_COUNT;
214 dlast = osd_lookup_one_len_unlocked(osd, LAST_RCVD, osd_sb(osd)->s_root,
217 return PTR_ERR(dlast);
218 else if (dlast->d_inode == NULL)
222 rc = osd_ldiskfs_read(dlast->d_inode, &lsd, sizeof(lsd), &off);
223 if (rc == sizeof(lsd)) {
225 "read last_rcvd header, uuid = %s, subdir count = %d\n",
226 lsd.lsd_uuid, lsd.lsd_subdir_count);
227 if (le16_to_cpu(lsd.lsd_subdir_count) > 0)
228 count = le16_to_cpu(lsd.lsd_subdir_count);
229 } else if (rc != 0) {
230 CERROR("Can't read last_rcvd file, rc = %d\n", rc);
242 static int osd_mdt_init(const struct lu_env *env, struct osd_device *dev)
244 struct lvfs_run_ctxt new;
245 struct lvfs_run_ctxt save;
246 struct dentry *parent;
247 struct osd_mdobj_map *omm;
249 struct osd_thread_info *info = osd_oti_get(env);
250 struct lu_fid *fid = &info->oti_fid3;
255 OBD_ALLOC_PTR(dev->od_mdt_map);
256 if (dev->od_mdt_map == NULL)
259 omm = dev->od_mdt_map;
261 parent = osd_sb(dev)->s_root;
262 osd_push_ctxt(dev, &new, &save);
264 lu_local_obj_fid(fid, REMOTE_PARENT_DIR_OID);
265 d = simple_mkdir(env, dev, parent, fid, REMOTE_PARENT_DIR,
266 LMAC_NOT_IN_OI, 0755, NULL);
268 GOTO(cleanup, rc = PTR_ERR(d));
270 omm->omm_remote_parent = d;
272 GOTO(cleanup, rc = 0);
275 pop_ctxt(&save, &new);
277 if (omm->omm_remote_parent != NULL)
278 dput(omm->omm_remote_parent);
280 dev->od_mdt_map = NULL;
285 static void osd_mdt_fini(struct osd_device *osd)
287 struct osd_mdobj_map *omm = osd->od_mdt_map;
292 if (omm->omm_remote_parent)
293 dput(omm->omm_remote_parent);
296 osd->od_ost_map = NULL;
299 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
300 struct osd_object *obj, struct osd_thandle *oh)
302 struct osd_mdobj_map *omm = osd->od_mdt_map;
303 struct osd_thread_info *oti = osd_oti_get(env);
304 struct lustre_mdt_attrs *lma = &oti->oti_ost_attrs.loa_lma;
305 char *name = oti->oti_name;
306 struct osd_thread_info *info = osd_oti_get(env);
307 struct dentry *dentry;
308 struct dentry *parent;
311 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_AGENTENT))
315 * Set REMOTE_PARENT in lma, so other process like unlink or lfsck
316 * can identify this object quickly
318 rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry,
319 &oti->oti_ost_attrs);
323 lma->lma_incompat |= LMAI_REMOTE_PARENT;
324 lustre_lma_swab(lma);
325 rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
326 sizeof(*lma), XATTR_REPLACE);
330 parent = omm->omm_remote_parent;
331 sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
332 dentry = osd_child_dentry_by_inode(env, parent->d_inode,
334 inode_lock(parent->d_inode);
335 rc = osd_ldiskfs_add_entry(info, osd, oh->ot_handle, dentry,
336 obj->oo_inode, NULL);
337 if (!rc && S_ISDIR(obj->oo_inode->i_mode))
338 ldiskfs_inc_count(oh->ot_handle, parent->d_inode);
339 else if (unlikely(rc == -EEXIST))
342 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
343 CDEBUG(D_INODE, "%s: create agent entry for %s: rc = %d\n",
344 osd_name(osd), name, rc);
345 mark_inode_dirty(parent->d_inode);
346 inode_unlock(parent->d_inode);
350 int osd_delete_from_remote_parent(const struct lu_env *env,
351 struct osd_device *osd,
352 struct osd_object *obj,
353 struct osd_thandle *oh, bool destroy)
355 struct osd_mdobj_map *omm = osd->od_mdt_map;
356 struct osd_thread_info *oti = osd_oti_get(env);
357 struct lustre_mdt_attrs *lma = &oti->oti_ost_attrs.loa_lma;
358 char *name = oti->oti_name;
359 struct dentry *dentry;
360 struct dentry *parent;
361 struct ldiskfs_dir_entry_2 *de;
362 struct buffer_head *bh;
365 parent = omm->omm_remote_parent;
366 sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
367 dentry = osd_child_dentry_by_inode(env, parent->d_inode,
369 inode_lock(parent->d_inode);
370 bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
373 inode_unlock(parent->d_inode);
375 if (unlikely(rc == -ENOENT))
378 rc = ldiskfs_delete_entry(oh->ot_handle, parent->d_inode,
380 if (!rc && S_ISDIR(obj->oo_inode->i_mode))
381 ldiskfs_dec_count(oh->ot_handle, parent->d_inode);
382 mark_inode_dirty(parent->d_inode);
383 inode_unlock(parent->d_inode);
385 CDEBUG(D_INODE, "%s: remove agent entry for %s: rc = %d\n",
386 osd_name(osd), name, rc);
391 lu_object_clear_agent_entry(&obj->oo_dt.do_lu);
396 rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry,
397 &oti->oti_ost_attrs);
401 /* Get rid of REMOTE_PARENT flag from incompat */
402 lma->lma_incompat &= ~LMAI_REMOTE_PARENT;
403 lustre_lma_swab(lma);
404 rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
405 sizeof(*lma), XATTR_REPLACE);
407 lu_object_clear_agent_entry(&obj->oo_dt.do_lu);
411 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
412 struct osd_device *osd,
413 const struct lu_fid *fid,
414 struct osd_inode_id *id)
416 struct osd_mdobj_map *omm = osd->od_mdt_map;
417 char *name = oti->oti_name;
418 struct dentry *parent;
419 struct dentry *dentry;
420 struct ldiskfs_dir_entry_2 *de;
421 struct buffer_head *bh;
426 if (unlikely(osd->od_is_ost))
429 parent = omm->omm_remote_parent;
430 sprintf(name, DFID_NOBRACE, PFID(fid));
431 dentry = osd_child_dentry_by_inode(oti->oti_env, parent->d_inode,
433 inode_lock(parent->d_inode);
434 bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
441 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
443 inode = osd_iget(oti, osd, id);
453 inode_unlock(parent->d_inode);
455 osd_add_oi_cache(oti, osd, id, fid);
460 * directory structure on legacy OST:
462 * O/<seq>/d0-31/<objid>
469 static int osd_ost_init(const struct lu_env *env, struct osd_device *dev)
471 struct lvfs_run_ctxt new;
472 struct lvfs_run_ctxt save;
475 bool created = false;
479 OBD_ALLOC_PTR(dev->od_ost_map);
480 if (dev->od_ost_map == NULL)
483 /* to get subdir count from last_rcvd */
484 rc = osd_last_rcvd_subdir_count(dev);
486 GOTO(cleanup_alloc, rc);
488 dev->od_ost_map->om_subdir_count = rc;
489 INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list);
490 rwlock_init(&dev->od_ost_map->om_seq_list_lock);
491 mutex_init(&dev->od_ost_map->om_dir_init_mutex);
493 osd_push_ctxt(dev, &new, &save);
494 d = simple_mkdir(env, dev, osd_sb(dev)->s_root, NULL, "O",
495 LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0755, &created);
497 GOTO(cleanup_ctxt, rc = PTR_ERR(d));
500 /* It is quite probably that the device is new formatted. */
501 dev->od_maybe_new = 1;
503 dev->od_ost_map->om_root = d;
505 pop_ctxt(&save, &new);
509 pop_ctxt(&save, &new);
511 OBD_FREE_PTR(dev->od_ost_map);
515 static void osd_seq_free(struct osd_obj_seq *osd_seq)
519 if (osd_seq->oos_dirs) {
520 for (j = 0; j < osd_seq->oos_subdir_count; j++) {
521 if (osd_seq->oos_dirs[j])
522 dput(osd_seq->oos_dirs[j]);
524 OBD_FREE(osd_seq->oos_dirs,
525 sizeof(struct dentry *) * osd_seq->oos_subdir_count);
528 if (osd_seq->oos_root)
529 dput(osd_seq->oos_root);
531 OBD_FREE_PTR(osd_seq);
534 static void osd_ost_fini(struct osd_device *osd)
536 struct osd_obj_seq *osd_seq;
537 struct osd_obj_seq *tmp;
538 struct osd_obj_map *map = osd->od_ost_map;
545 write_lock(&map->om_seq_list_lock);
546 list_for_each_entry_safe(osd_seq, tmp, &map->om_seq_list,
548 list_del_init(&osd_seq->oos_seq_list);
549 write_unlock(&map->om_seq_list_lock);
550 osd_seq_free(osd_seq);
551 write_lock(&map->om_seq_list_lock);
553 write_unlock(&map->om_seq_list_lock);
557 osd->od_ost_map = NULL;
561 static int osd_index_backup_dir_init(const struct lu_env *env,
562 struct osd_device *dev)
564 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
565 struct lvfs_run_ctxt new;
566 struct lvfs_run_ctxt save;
567 struct dentry *dentry;
572 lu_local_obj_fid(fid, INDEX_BACKUP_OID);
573 osd_push_ctxt(dev, &new, &save);
574 dentry = simple_mkdir(env, dev, osd_sb(dev)->s_root, fid,
575 INDEX_BACKUP_DIR, LMAC_NOT_IN_OI, 0755, NULL);
576 if (IS_ERR(dentry)) {
577 rc = PTR_ERR(dentry);
579 dev->od_index_backup_inode = igrab(dentry->d_inode);
582 pop_ctxt(&save, &new);
587 static void osd_index_backup_dir_fini(struct osd_device *dev)
589 if (dev->od_index_backup_inode) {
590 iput(dev->od_index_backup_inode);
591 dev->od_index_backup_inode = NULL;
595 int osd_obj_map_init(const struct lu_env *env, struct osd_device *dev)
598 bool mdt_init = false;
602 rc = osd_ost_init(env, dev);
606 if (!dev->od_is_ost) {
607 rc = osd_mdt_init(env, dev);
616 rc = osd_index_backup_dir_init(env, dev);
626 static struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, u64 seq)
628 struct osd_obj_seq *osd_seq;
630 list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) {
631 if (osd_seq->oos_seq == seq)
637 static struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, u64 seq)
639 struct osd_obj_seq *osd_seq;
641 read_lock(&map->om_seq_list_lock);
642 osd_seq = osd_seq_find_locked(map, seq);
643 read_unlock(&map->om_seq_list_lock);
647 void osd_obj_map_fini(struct osd_device *dev)
649 osd_index_backup_dir_fini(dev);
655 * Update the specified OI mapping.
657 * \retval 1, changed nothing
658 * \retval 0, changed successfully
659 * \retval -ve, on error
661 static int osd_obj_update_entry(struct osd_thread_info *info,
662 struct osd_device *osd,
663 struct dentry *dir, const char *name,
664 const struct lu_fid *fid,
665 const struct osd_inode_id *id,
668 struct inode *parent = dir->d_inode;
669 struct dentry *child;
670 struct ldiskfs_dir_entry_2 *de;
671 struct buffer_head *bh;
673 struct dentry *dentry = &info->oti_obj_dentry;
674 struct osd_inode_id *oi_id = &info->oti_id3;
675 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
676 struct lu_fid *oi_fid = &lma->lma_self_fid;
682 LASSERT(th->h_transaction != NULL);
684 child = &info->oti_child_dentry;
685 child->d_parent = dir;
686 child->d_name.hash = 0;
687 child->d_name.name = name;
688 child->d_name.len = strlen(name);
690 dquot_initialize(parent);
692 bh = osd_ldiskfs_find_entry(parent, &child->d_name, &de, NULL, NULL);
694 GOTO(out, rc = PTR_ERR(bh));
696 if (le32_to_cpu(de->inode) == id->oii_ino)
699 osd_id_gen(oi_id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
700 inode = osd_iget(info, osd, oi_id);
703 if (rc == -ENOENT || rc == -ESTALE)
709 * The EA inode should NOT be in OI, old OI scrub may added
710 * such OI mapping by wrong, replace it.
712 if (unlikely(osd_is_ea_inode(inode))) {
717 rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
718 if (rc == -ENODATA) {
719 rc = osd_get_idif(info, inode, dentry, oi_fid);
720 if (rc > 0 || rc == -ENODATA) {
731 * If the OST-object has neither FID-in-LMA nor FID-in-ff, it is
732 * either a crashed object or a uninitialized one. Replace it.
734 if (oi_fid != NULL && lu_fid_eq(fid, oi_fid)) {
735 CERROR("%s: the FID "DFID" is used by two objects: "
736 "%u/%u %u/%u\n", osd_name(osd), PFID(fid),
737 oi_id->oii_ino, oi_id->oii_gen,
738 id->oii_ino, id->oii_gen);
739 GOTO(out, rc = -EEXIST);
742 if (fid_is_idif(fid) && oi_fid != NULL && fid_is_idif(oi_fid)) {
743 __u32 idx1 = fid_idif_ost_idx(fid);
744 __u32 idx2 = fid_idif_ost_idx(oi_fid);
745 struct ost_id *ostid = &info->oti_ostid;
746 struct lu_fid *tfid = &info->oti_fid3;
748 LASSERTF(idx1 == 0 || idx1 == osd->od_index,
749 "invalid given FID "DFID", not match the "
750 "device index %u\n", PFID(fid), osd->od_index);
753 if (idx1 == 0 && idx2 == osd->od_index) {
754 fid_to_ostid(fid, ostid);
755 ostid_to_fid(tfid, ostid, idx2);
756 if (lu_fid_eq(tfid, oi_fid)) {
757 CERROR("%s: the FID "DFID" is used by "
758 "two objects(2): %u/%u %u/%u\n",
759 osd_name(osd), PFID(fid),
760 oi_id->oii_ino, oi_id->oii_gen,
761 id->oii_ino, id->oii_gen);
763 GOTO(out, rc = -EEXIST);
765 } else if (idx2 == 0 && idx1 == osd->od_index) {
766 fid_to_ostid(oi_fid, ostid);
767 ostid_to_fid(tfid, ostid, idx1);
768 if (lu_fid_eq(tfid, fid)) {
769 CERROR("%s: the FID "DFID" is used by "
770 "two objects(2): %u/%u %u/%u\n",
771 osd_name(osd), PFID(fid),
772 oi_id->oii_ino, oi_id->oii_gen,
773 id->oii_ino, id->oii_gen);
775 GOTO(out, rc = -EEXIST);
783 * There may be temporary inconsistency: On one hand, the new
784 * object may be referenced by multiple entries, which is out
785 * of our control unless we traverse the whole /O completely,
786 * which is non-flat order and inefficient, should be avoided;
787 * On the other hand, the old object may become orphan if it
788 * is still valid. Since it was referenced by an invalid entry,
789 * making it as invisible temporary may be not worse. OI scrub
790 * will process it later.
792 rc = ldiskfs_journal_get_write_access(th, bh);
796 de->inode = cpu_to_le32(id->oii_ino);
797 rc = ldiskfs_handle_dirty_metadata(th, NULL, bh);
804 inode_unlock(parent);
808 static int osd_obj_del_entry(struct osd_thread_info *info,
809 struct osd_device *osd,
810 struct dentry *dird, char *name,
813 struct ldiskfs_dir_entry_2 *de;
814 struct buffer_head *bh;
815 struct dentry *child;
816 struct inode *dir = dird->d_inode;
822 LASSERT(th->h_transaction != NULL);
824 child = &info->oti_child_dentry;
825 child->d_name.hash = 0;
826 child->d_name.name = name;
827 child->d_name.len = strlen(name);
828 child->d_parent = dird;
829 child->d_inode = NULL;
831 dquot_initialize(dir);
833 bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
837 rc = ldiskfs_delete_entry(th, dir, de, bh);
845 static int osd_obj_add_entry(struct osd_thread_info *info,
846 struct osd_device *osd,
847 struct dentry *dir, char *name,
848 const struct osd_inode_id *id,
851 struct dentry *child;
857 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY))
861 LASSERT(th->h_transaction != NULL);
863 inode = info->oti_inode;
864 if (unlikely(inode == NULL)) {
865 struct ldiskfs_inode_info *lii;
870 inode = info->oti_inode = &lii->vfs_inode;
873 inode->i_sb = osd_sb(osd);
874 osd_id_to_inode(inode, id);
875 inode->i_mode = S_IFREG; /* for type in ldiskfs dir entry */
877 child = &info->oti_child_dentry;
878 child->d_name.hash = 0;
879 child->d_name.name = name;
880 child->d_name.len = strlen(name);
881 child->d_parent = dir;
882 child->d_inode = inode;
884 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
887 dquot_initialize(dir->d_inode);
888 inode_lock(dir->d_inode);
889 rc = osd_ldiskfs_add_entry(info, osd, th, child, inode, NULL);
890 inode_unlock(dir->d_inode);
896 * Use %llu for legacy OST sequences, but use %llx for new
897 * sequences names, so that the O/{seq}/dN/{oid} more closely
898 * follows the DFID/PFID format. This makes it easier to map from
899 * debug messages to objects in the future, and the legacy space
900 * of FID_SEQ_OST_MDT0 will be unused in the future.
902 static inline void osd_seq_name(char *seq_name, size_t name_size, u64 seq)
904 snprintf(seq_name, name_size,
905 (fid_seq_is_rsvd(seq) ||
906 fid_seq_is_mdt0(seq)) ? "%llu" : "%llx",
907 fid_seq_is_idif(seq) ? 0 : seq);
910 static inline void osd_oid_name(char *name, size_t name_size,
911 const struct lu_fid *fid, u64 id)
913 snprintf(name, name_size,
914 (fid_seq_is_rsvd(fid_seq(fid)) ||
915 fid_seq_is_mdt0(fid_seq(fid)) ||
916 fid_seq_is_idif(fid_seq(fid))) ? "%llu" : "%llx", id);
919 /* external locking is required */
920 static int osd_seq_load_locked(struct osd_thread_info *info,
921 struct osd_device *osd,
922 struct osd_obj_seq *osd_seq)
924 struct osd_obj_map *map = osd->od_ost_map;
925 struct dentry *seq_dir;
932 if (osd_seq->oos_root != NULL)
936 LASSERT(map->om_root);
938 osd_seq_name(dir_name, sizeof(dir_name), osd_seq->oos_seq);
940 seq_dir = simple_mkdir(info->oti_env, osd, map->om_root, NULL, dir_name,
941 LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0755, NULL);
943 GOTO(out_err, rc = PTR_ERR(seq_dir));
944 else if (seq_dir->d_inode == NULL)
945 GOTO(out_put, rc = -EFAULT);
947 osd_seq->oos_root = seq_dir;
949 LASSERT(osd_seq->oos_dirs == NULL);
950 OBD_ALLOC(osd_seq->oos_dirs,
951 sizeof(seq_dir) * osd_seq->oos_subdir_count);
952 if (osd_seq->oos_dirs == NULL)
953 GOTO(out_put, rc = -ENOMEM);
955 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
958 snprintf(dir_name, sizeof(dir_name), "d%u", i);
959 dir = simple_mkdir(info->oti_env, osd, osd_seq->oos_root, NULL,
960 dir_name, LMAC_NOT_IN_OI | LMAC_FID_ON_OST,
963 GOTO(out_free, rc = PTR_ERR(dir));
964 } else if (dir->d_inode == NULL) {
966 GOTO(out_free, rc = -EFAULT);
969 osd_seq->oos_dirs[i] = dir;
974 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
975 if (osd_seq->oos_dirs[i] != NULL)
976 dput(osd_seq->oos_dirs[i]);
978 OBD_FREE(osd_seq->oos_dirs,
979 sizeof(seq_dir) * osd_seq->oos_subdir_count);
982 osd_seq->oos_root = NULL;
988 static struct osd_obj_seq *osd_seq_load(struct osd_thread_info *info,
989 struct osd_device *osd, u64 seq)
991 struct osd_obj_map *map;
992 struct osd_obj_seq *osd_seq;
997 map = osd->od_ost_map;
999 LASSERT(map->om_root);
1001 osd_seq = osd_seq_find(map, seq);
1002 if (likely(osd_seq != NULL))
1005 /* Serializing init process */
1006 mutex_lock(&map->om_dir_init_mutex);
1008 /* Check whether the seq has been added */
1009 read_lock(&map->om_seq_list_lock);
1010 osd_seq = osd_seq_find_locked(map, seq);
1011 if (osd_seq != NULL) {
1012 read_unlock(&map->om_seq_list_lock);
1013 GOTO(cleanup, rc = 0);
1015 read_unlock(&map->om_seq_list_lock);
1017 OBD_ALLOC_PTR(osd_seq);
1018 if (osd_seq == NULL)
1019 GOTO(cleanup, rc = -ENOMEM);
1021 INIT_LIST_HEAD(&osd_seq->oos_seq_list);
1022 osd_seq->oos_seq = seq;
1024 * Init subdir count to be 32, but each seq can have
1025 * different subdir count
1027 osd_seq->oos_subdir_count = map->om_subdir_count;
1028 rc = osd_seq_load_locked(info, osd, osd_seq);
1032 write_lock(&map->om_seq_list_lock);
1033 list_add(&osd_seq->oos_seq_list, &map->om_seq_list);
1034 write_unlock(&map->om_seq_list_lock);
1037 mutex_unlock(&map->om_dir_init_mutex);
1039 if (osd_seq != NULL)
1040 OBD_FREE_PTR(osd_seq);
1041 RETURN(ERR_PTR(rc));
1047 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev,
1048 const struct lu_fid *fid, struct osd_inode_id *id)
1050 struct osd_obj_map *map;
1051 struct osd_obj_seq *osd_seq;
1052 struct dentry *d_seq;
1053 struct dentry *child;
1054 struct ost_id *ostid = &info->oti_ostid;
1057 struct ldiskfs_dir_entry_2 *de;
1058 struct buffer_head *bh;
1060 struct inode *inode;
1064 /* on the very first lookup we find and open directories */
1065 map = dev->od_ost_map;
1067 LASSERT(map->om_root);
1069 fid_to_ostid(fid, ostid);
1070 osd_seq = osd_seq_load(info, dev, ostid_seq(ostid));
1071 if (IS_ERR(osd_seq))
1072 RETURN(PTR_ERR(osd_seq));
1074 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1075 d_seq = osd_seq->oos_dirs[dirn];
1078 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1080 child = &info->oti_child_dentry;
1081 child->d_parent = d_seq;
1082 child->d_name.hash = 0;
1083 child->d_name.name = name;
1084 /* XXX: we can use rc from sprintf() instead of strlen() */
1085 child->d_name.len = strlen(name);
1087 dir = d_seq->d_inode;
1089 bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
1093 RETURN(PTR_ERR(bh));
1095 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
1098 inode = osd_iget(info, dev, id);
1099 if (IS_ERR(inode)) {
1100 int rc = PTR_ERR(inode);
1102 RETURN(rc == -ENOENT ? -ESTALE : rc);
1109 int osd_obj_map_insert(struct osd_thread_info *info,
1110 struct osd_device *osd,
1111 const struct lu_fid *fid,
1112 const struct osd_inode_id *id,
1115 struct osd_obj_map *map;
1116 struct osd_obj_seq *osd_seq;
1118 struct ost_id *ostid = &info->oti_ostid;
1125 map = osd->od_ost_map;
1128 /* map fid to seq:objid */
1129 fid_to_ostid(fid, ostid);
1131 oid = ostid_id(ostid);
1132 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1133 if (IS_ERR(osd_seq))
1134 RETURN(PTR_ERR(osd_seq));
1136 dirn = oid & (osd_seq->oos_subdir_count - 1);
1137 d = osd_seq->oos_dirs[dirn];
1140 osd_oid_name(name, sizeof(name), fid, oid);
1143 rc = osd_obj_add_entry(info, osd, d, name, id, th);
1144 if (rc == -EEXIST) {
1145 rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1146 if (unlikely(rc == -ENOENT))
1149 if (unlikely(rc == 1))
1156 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
1157 const struct lu_fid *fid, handle_t *th)
1159 struct osd_obj_map *map;
1160 struct osd_obj_seq *osd_seq;
1162 struct ost_id *ostid = &info->oti_ostid;
1168 map = osd->od_ost_map;
1171 /* map fid to seq:objid */
1172 fid_to_ostid(fid, ostid);
1174 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1175 if (IS_ERR(osd_seq))
1176 GOTO(cleanup, rc = PTR_ERR(osd_seq));
1178 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1179 d = osd_seq->oos_dirs[dirn];
1182 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1183 rc = osd_obj_del_entry(info, osd, d, name, th);
1188 int osd_obj_map_update(struct osd_thread_info *info,
1189 struct osd_device *osd,
1190 const struct lu_fid *fid,
1191 const struct osd_inode_id *id,
1194 struct osd_obj_seq *osd_seq;
1196 struct ost_id *ostid = &info->oti_ostid;
1202 fid_to_ostid(fid, ostid);
1203 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1204 if (IS_ERR(osd_seq))
1205 RETURN(PTR_ERR(osd_seq));
1207 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1208 d = osd_seq->oos_dirs[dirn];
1211 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1212 rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1217 int osd_obj_map_recover(struct osd_thread_info *info,
1218 struct osd_device *osd,
1219 struct inode *src_parent,
1220 struct dentry *src_child,
1221 const struct lu_fid *fid)
1223 struct osd_obj_seq *osd_seq;
1224 struct dentry *tgt_parent;
1225 struct dentry *tgt_child = &info->oti_child_dentry;
1227 struct inode *inode = src_child->d_inode;
1228 struct ost_id *ostid = &info->oti_ostid;
1230 struct ldiskfs_dir_entry_2 *de;
1231 struct buffer_head *bh;
1238 if (fid_is_last_id(fid)) {
1239 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1240 if (IS_ERR(osd_seq))
1241 RETURN(PTR_ERR(osd_seq));
1243 tgt_parent = osd_seq->oos_root;
1244 tgt_child->d_name.name = "LAST_ID";
1245 tgt_child->d_name.len = strlen("LAST_ID");
1247 fid_to_ostid(fid, ostid);
1248 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1249 if (IS_ERR(osd_seq))
1250 RETURN(PTR_ERR(osd_seq));
1252 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1253 tgt_parent = osd_seq->oos_dirs[dirn];
1254 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1255 tgt_child->d_name.name = name;
1256 tgt_child->d_name.len = strlen(name);
1258 LASSERT(tgt_parent != NULL);
1260 dir = tgt_parent->d_inode;
1261 tgt_child->d_name.hash = 0;
1262 tgt_child->d_parent = tgt_parent;
1263 tgt_child->d_inode = inode;
1265 /* The non-initialized src_child may be destroyed. */
1266 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
1267 osd_dto_credits_noquota[DTO_INDEX_DELETE] +
1268 osd_dto_credits_noquota[DTO_INDEX_INSERT] +
1269 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
1271 RETURN(PTR_ERR(jh));
1273 dquot_initialize(src_parent);
1274 dquot_initialize(dir);
1276 inode_lock(src_parent);
1278 bh = osd_ldiskfs_find_entry(dir, &tgt_child->d_name, &de, NULL, NULL);
1281 * XXX: If some other object occupied the same slot. And If such
1282 * inode is zero-sized and with SUID+SGID, then means it is
1283 * a new created one. Maybe we can remove it and insert the
1284 * original one back to the /O/<seq>/d<x>. But there are
1285 * something to be considered:
1287 * 1) The OST-object under /lost+found has crashed LMA.
1288 * So it should not conflict with the current one.
1290 * 2) There are race conditions that: someone may just want
1291 * to modify the current one. Even if the OI scrub takes
1292 * the object lock when remove the current one, it still
1293 * cause the modification to be lost becasue the target
1294 * has been removed when the RPC service thread waiting
1297 * So keep it there before we have suitable solution.
1301 inode_unlock(src_parent);
1302 ldiskfs_journal_stop(jh);
1305 /* If the src object has never been modified, then remove it. */
1306 if (inode->i_size == 0 && inode->i_mode & S_ISUID &&
1307 inode->i_mode & S_ISGID) {
1308 rc = ll_vfs_unlink(src_parent, src_child);
1309 if (unlikely(rc == -ENOENT))
1316 bh = osd_ldiskfs_find_entry(src_parent, &src_child->d_name, &de,
1318 if (unlikely(IS_ERR(bh)))
1319 GOTO(unlock, rc = PTR_ERR(bh));
1321 rc = ldiskfs_delete_entry(jh, src_parent, de, bh);
1326 rc = osd_ldiskfs_add_entry(info, osd, jh, tgt_child, inode, NULL);
1332 inode_unlock(src_parent);
1333 ldiskfs_journal_stop(jh);
1337 static struct dentry *
1338 osd_object_spec_find(struct osd_thread_info *info, struct osd_device *osd,
1339 const struct lu_fid *fid, char **name)
1341 struct dentry *root = ERR_PTR(-ENOENT);
1343 if (fid_is_last_id(fid)) {
1344 struct osd_obj_seq *osd_seq;
1346 /* on creation of LAST_ID we create O/<seq> hierarchy */
1347 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1348 if (IS_ERR(osd_seq))
1349 RETURN((struct dentry *)osd_seq);
1352 root = osd_seq->oos_root;
1354 *name = osd_lf_fid2name(fid);
1356 CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid));
1357 else if ((*name)[0])
1358 root = osd_sb(osd)->s_root;
1364 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
1365 const struct lu_fid *fid, const struct osd_inode_id *id,
1368 struct dentry *root;
1374 root = osd_object_spec_find(info, osd, fid, &name);
1375 if (!IS_ERR(root)) {
1376 rc = osd_obj_update_entry(info, osd, root, name, fid, id, th);
1386 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
1387 const struct lu_fid *fid, const struct osd_inode_id *id,
1390 struct dentry *root;
1396 root = osd_object_spec_find(info, osd, fid, &name);
1397 if (!IS_ERR(root)) {
1398 rc = osd_obj_add_entry(info, osd, root, name, id, th);
1408 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
1409 const struct lu_fid *fid, struct osd_inode_id *id)
1411 struct dentry *root;
1412 struct dentry *dentry;
1413 struct inode *inode;
1419 if (fid_is_last_id(fid)) {
1420 struct osd_obj_seq *osd_seq;
1422 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1423 if (IS_ERR(osd_seq))
1424 RETURN(PTR_ERR(osd_seq));
1425 root = osd_seq->oos_root;
1428 root = osd_sb(osd)->s_root;
1429 name = osd_lf_fid2name(fid);
1430 if (name == NULL || strlen(name) == 0)
1434 dentry = osd_lookup_one_len_unlocked(osd, name, root, strlen(name));
1435 if (!IS_ERR(dentry)) {
1436 inode = dentry->d_inode;
1438 if (is_bad_inode(inode)) {
1441 osd_id_gen(id, inode->i_ino,
1442 inode->i_generation);
1447 * if dentry is accessible after osd_compat_spec_insert it
1448 * will still contain NULL inode, so don't keep it in cache
1450 d_invalidate(dentry);
1457 #ifndef HAVE_BIO_INTEGRITY_ENABLED
1458 bool bio_integrity_enabled(struct bio *bio)
1460 struct blk_integrity *bi = blk_get_integrity(bio_get_disk(bio));
1462 if (bio_op(bio) != REQ_OP_READ && bio_op(bio) != REQ_OP_WRITE)
1465 if (!bio_sectors(bio))
1468 /* Already protected? */
1469 if (bio_integrity(bio))
1475 if (bio_data_dir(bio) == READ && bi->profile->verify_fn != NULL &&
1476 (bi->flags & BLK_INTEGRITY_VERIFY))
1479 if (bio_data_dir(bio) == WRITE && bi->profile->generate_fn != NULL &&
1480 (bi->flags & BLK_INTEGRITY_GENERATE))