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->umask = current_umask();
65 push_ctxt(save, newctxt);
68 struct dentry *osd_lookup_one_len_common(struct osd_device *dev,
70 struct dentry *base, int len,
71 enum oi_check_flags flags)
73 struct dentry *dchild;
76 * We can't use inode_is_locked() directly since we can't know
77 * if the current thread context took the lock earlier or if
78 * another thread context took the lock. OI_LOCKED tells us
79 * if the current thread context has already taken the lock.
81 if (!(flags & OI_LOCKED)) {
82 /* If another thread took this lock already we will
83 * just have to wait until the other thread is done.
85 inode_lock(base->d_inode);
86 dchild = lookup_one_len(name, base, len);
87 inode_unlock(base->d_inode);
89 /* This thread context already has taken the lock.
90 * Other threads will have to wait until we are done.
92 dchild = lookup_one_len(name, base, len);
97 if (dchild->d_inode && unlikely(is_bad_inode(dchild->d_inode))) {
98 CERROR("%s: bad inode returned %lu/%u: rc = -ENOENT\n",
99 osd_name(dev), dchild->d_inode->i_ino,
100 dchild->d_inode->i_generation);
102 dchild = ERR_PTR(-ENOENT);
109 * osd_lookup_one_len_unlocked
111 * @dev: obd device we are searching
112 * @name: pathname component to lookup
113 * @base: base directory to lookup from
114 * @len: maximum length @len should be interpreted to
116 * Unlike osd_lookup_one_len, this should be called without the parent
117 * i_mutex held, and will take the i_mutex itself.
119 struct dentry *osd_lookup_one_len_unlocked(struct osd_device *dev,
121 struct dentry *base, int len)
123 return osd_lookup_one_len_common(dev, name, base, len, ~OI_LOCKED);
127 * osd_lookup_one_len - lookup single pathname component
129 * @dev: obd device we are searching
130 * @name: pathname component to lookup
131 * @base: base directory to lookup from
132 * @len: maximum length @len should be interpreted to
134 * The caller must hold inode lock
136 struct dentry *osd_lookup_one_len(struct osd_device *dev, const char *name,
137 struct dentry *base, int len)
139 return osd_lookup_one_len_common(dev, name, base, len, OI_LOCKED);
142 /* utility to make a directory */
143 static struct dentry *
144 simple_mkdir(const struct lu_env *env, struct osd_device *osd,
145 struct dentry *dir, const struct lu_fid *fid,
146 const char *name, __u32 compat, int mode, bool *created)
148 struct osd_thread_info *info = osd_oti_get(env);
149 struct lu_fid *tfid = &info->oti_fid3;
151 struct dentry *dchild;
156 // ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
157 CDEBUG(D_INODE, "creating directory %.*s\n", (int)strlen(name), name);
158 dchild = osd_lookup_one_len_unlocked(osd, name, dir, strlen(name));
162 inode = dchild->d_inode;
164 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
165 int old_mode = inode->i_mode;
170 if (!S_ISDIR(old_mode)) {
171 CERROR("found %s (%lu/%u) is mode %o\n", name,
172 inode->i_ino, inode->i_generation, old_mode);
173 GOTO(out_err, err = -ENOTDIR);
176 if (unlikely(osd->od_dt_dev.dd_rdonly))
179 /* Fixup directory permissions if necessary */
180 if ((old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
182 "fixing permissions on %s from %o to %o\n",
183 name, old_mode, mode);
184 inode->i_mode = (mode & S_IALLUGO) |
185 (old_mode & ~S_IALLUGO);
186 mark_inode_dirty(inode);
189 err = osd_get_lma(info, inode, &info->oti_obj_dentry,
190 &info->oti_ost_attrs);
197 if ((fid && !lu_fid_eq(fid, &lma->lma_self_fid)) ||
198 lma->lma_compat != compat)
204 err = vfs_mkdir(dir->d_inode, dchild, mode);
208 inode = dchild->d_inode;
216 lu_igif_build(tfid, inode->i_ino, inode->i_generation);
217 err = osd_ea_fid_set(info, inode, tfid, compat, 0);
228 static int osd_last_rcvd_subdir_count(struct osd_device *osd)
230 struct lr_server_data lsd;
231 struct dentry *dlast;
234 int count = OBJ_SUBDIR_COUNT;
238 dlast = osd_lookup_one_len_unlocked(osd, LAST_RCVD, osd_sb(osd)->s_root,
241 return PTR_ERR(dlast);
242 else if (dlast->d_inode == NULL)
246 rc = osd_ldiskfs_read(dlast->d_inode, &lsd, sizeof(lsd), &off);
247 if (rc == sizeof(lsd)) {
249 "read last_rcvd header, uuid = %s, subdir count = %d\n",
250 lsd.lsd_uuid, lsd.lsd_subdir_count);
251 if (le16_to_cpu(lsd.lsd_subdir_count) > 0)
252 count = le16_to_cpu(lsd.lsd_subdir_count);
253 } else if (rc != 0) {
254 CERROR("Can't read last_rcvd file, rc = %d\n", rc);
266 static int osd_mdt_init(const struct lu_env *env, struct osd_device *dev)
268 struct lvfs_run_ctxt new;
269 struct lvfs_run_ctxt save;
270 struct dentry *parent;
271 struct osd_mdobj_map *omm;
273 struct osd_thread_info *info = osd_oti_get(env);
274 struct lu_fid *fid = &info->oti_fid3;
279 OBD_ALLOC_PTR(dev->od_mdt_map);
280 if (dev->od_mdt_map == NULL)
283 omm = dev->od_mdt_map;
285 parent = osd_sb(dev)->s_root;
286 osd_push_ctxt(dev, &new, &save);
288 lu_local_obj_fid(fid, REMOTE_PARENT_DIR_OID);
289 d = simple_mkdir(env, dev, parent, fid, REMOTE_PARENT_DIR,
290 LMAC_NOT_IN_OI, 0755, NULL);
292 GOTO(cleanup, rc = PTR_ERR(d));
294 omm->omm_remote_parent = d;
296 GOTO(cleanup, rc = 0);
299 pop_ctxt(&save, &new);
301 if (omm->omm_remote_parent != NULL)
302 dput(omm->omm_remote_parent);
304 dev->od_mdt_map = NULL;
309 static void osd_mdt_fini(struct osd_device *osd)
311 struct osd_mdobj_map *omm = osd->od_mdt_map;
316 if (omm->omm_remote_parent)
317 dput(omm->omm_remote_parent);
320 osd->od_ost_map = NULL;
323 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
324 struct osd_object *obj, struct osd_thandle *oh)
326 struct osd_mdobj_map *omm = osd->od_mdt_map;
327 struct osd_thread_info *oti = osd_oti_get(env);
328 struct lustre_mdt_attrs *lma = &oti->oti_ost_attrs.loa_lma;
329 char *name = oti->oti_name;
330 struct osd_thread_info *info = osd_oti_get(env);
331 struct dentry *dentry;
332 struct dentry *parent;
335 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_AGENTENT))
339 * Set REMOTE_PARENT in lma, so other process like unlink or lfsck
340 * can identify this object quickly
342 rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry,
343 &oti->oti_ost_attrs);
347 lma->lma_incompat |= LMAI_REMOTE_PARENT;
348 lustre_lma_swab(lma);
349 rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
350 sizeof(*lma), XATTR_REPLACE);
354 parent = omm->omm_remote_parent;
355 sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
356 dentry = osd_child_dentry_by_inode(env, parent->d_inode,
358 inode_lock(parent->d_inode);
359 rc = osd_ldiskfs_add_entry(info, osd, oh->ot_handle, dentry,
360 obj->oo_inode, NULL);
361 if (!rc && S_ISDIR(obj->oo_inode->i_mode))
362 ldiskfs_inc_count(oh->ot_handle, parent->d_inode);
363 else if (unlikely(rc == -EEXIST))
366 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
367 CDEBUG(D_INODE, "%s: create agent entry for %s: rc = %d\n",
368 osd_name(osd), name, rc);
369 mark_inode_dirty(parent->d_inode);
370 inode_unlock(parent->d_inode);
374 int osd_delete_from_remote_parent(const struct lu_env *env,
375 struct osd_device *osd,
376 struct osd_object *obj,
377 struct osd_thandle *oh, bool destroy)
379 struct osd_mdobj_map *omm = osd->od_mdt_map;
380 struct osd_thread_info *oti = osd_oti_get(env);
381 struct lustre_mdt_attrs *lma = &oti->oti_ost_attrs.loa_lma;
382 char *name = oti->oti_name;
383 struct dentry *dentry;
384 struct dentry *parent;
385 struct ldiskfs_dir_entry_2 *de;
386 struct buffer_head *bh;
389 parent = omm->omm_remote_parent;
390 sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
391 dentry = osd_child_dentry_by_inode(env, parent->d_inode,
393 inode_lock(parent->d_inode);
394 bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
397 inode_unlock(parent->d_inode);
399 if (unlikely(rc == -ENOENT))
402 rc = ldiskfs_delete_entry(oh->ot_handle, parent->d_inode,
404 if (!rc && S_ISDIR(obj->oo_inode->i_mode))
405 ldiskfs_dec_count(oh->ot_handle, parent->d_inode);
406 mark_inode_dirty(parent->d_inode);
407 inode_unlock(parent->d_inode);
409 CDEBUG(D_INODE, "%s: remove agent entry for %s: rc = %d\n",
410 osd_name(osd), name, rc);
415 lu_object_clear_agent_entry(&obj->oo_dt.do_lu);
420 rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry,
421 &oti->oti_ost_attrs);
425 /* Get rid of REMOTE_PARENT flag from incompat */
426 lma->lma_incompat &= ~LMAI_REMOTE_PARENT;
427 lustre_lma_swab(lma);
428 rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
429 sizeof(*lma), XATTR_REPLACE);
431 lu_object_clear_agent_entry(&obj->oo_dt.do_lu);
435 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
436 struct osd_device *osd,
437 const struct lu_fid *fid,
438 struct osd_inode_id *id)
440 struct osd_mdobj_map *omm = osd->od_mdt_map;
441 char *name = oti->oti_name;
442 struct dentry *parent;
443 struct dentry *dentry;
444 struct ldiskfs_dir_entry_2 *de;
445 struct buffer_head *bh;
450 if (unlikely(osd->od_is_ost))
453 parent = omm->omm_remote_parent;
454 sprintf(name, DFID_NOBRACE, PFID(fid));
455 dentry = osd_child_dentry_by_inode(oti->oti_env, parent->d_inode,
457 inode_lock(parent->d_inode);
458 bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
465 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
467 inode = osd_iget(oti, osd, id);
477 inode_unlock(parent->d_inode);
479 osd_add_oi_cache(oti, osd, id, fid);
484 * directory structure on legacy OST:
486 * O/<seq>/d0-31/<objid>
493 static int osd_ost_init(const struct lu_env *env, struct osd_device *dev)
495 struct lvfs_run_ctxt new;
496 struct lvfs_run_ctxt save;
499 bool created = false;
503 OBD_ALLOC_PTR(dev->od_ost_map);
504 if (dev->od_ost_map == NULL)
507 /* to get subdir count from last_rcvd */
508 rc = osd_last_rcvd_subdir_count(dev);
510 GOTO(cleanup_alloc, rc);
512 dev->od_ost_map->om_subdir_count = rc;
513 INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list);
514 rwlock_init(&dev->od_ost_map->om_seq_list_lock);
515 mutex_init(&dev->od_ost_map->om_dir_init_mutex);
517 osd_push_ctxt(dev, &new, &save);
518 d = simple_mkdir(env, dev, osd_sb(dev)->s_root, NULL, "O",
519 LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0755, &created);
521 GOTO(cleanup_ctxt, rc = PTR_ERR(d));
524 /* It is quite probably that the device is new formatted. */
525 dev->od_maybe_new = 1;
527 dev->od_ost_map->om_root = d;
529 pop_ctxt(&save, &new);
533 pop_ctxt(&save, &new);
535 OBD_FREE_PTR(dev->od_ost_map);
539 static void osd_seq_free(struct osd_obj_seq *osd_seq)
543 if (osd_seq->oos_dirs) {
544 for (j = 0; j < osd_seq->oos_subdir_count; j++) {
545 if (osd_seq->oos_dirs[j])
546 dput(osd_seq->oos_dirs[j]);
548 OBD_FREE_PTR_ARRAY(osd_seq->oos_dirs,
549 osd_seq->oos_subdir_count);
552 if (osd_seq->oos_root)
553 dput(osd_seq->oos_root);
555 OBD_FREE_PTR(osd_seq);
558 static void osd_ost_fini(struct osd_device *osd)
560 struct osd_obj_seq *osd_seq;
561 struct osd_obj_seq *tmp;
562 struct osd_obj_map *map = osd->od_ost_map;
569 write_lock(&map->om_seq_list_lock);
570 list_for_each_entry_safe(osd_seq, tmp, &map->om_seq_list,
572 list_del_init(&osd_seq->oos_seq_list);
573 write_unlock(&map->om_seq_list_lock);
574 osd_seq_free(osd_seq);
575 write_lock(&map->om_seq_list_lock);
577 write_unlock(&map->om_seq_list_lock);
581 osd->od_ost_map = NULL;
585 static int osd_index_backup_dir_init(const struct lu_env *env,
586 struct osd_device *dev)
588 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
589 struct lvfs_run_ctxt new;
590 struct lvfs_run_ctxt save;
591 struct dentry *dentry;
596 lu_local_obj_fid(fid, INDEX_BACKUP_OID);
597 osd_push_ctxt(dev, &new, &save);
598 dentry = simple_mkdir(env, dev, osd_sb(dev)->s_root, fid,
599 INDEX_BACKUP_DIR, LMAC_NOT_IN_OI, 0755, NULL);
600 if (IS_ERR(dentry)) {
601 rc = PTR_ERR(dentry);
603 dev->od_index_backup_inode = igrab(dentry->d_inode);
606 pop_ctxt(&save, &new);
611 static void osd_index_backup_dir_fini(struct osd_device *dev)
613 if (dev->od_index_backup_inode) {
614 iput(dev->od_index_backup_inode);
615 dev->od_index_backup_inode = NULL;
619 int osd_obj_map_init(const struct lu_env *env, struct osd_device *dev)
622 bool mdt_init = false;
626 rc = osd_ost_init(env, dev);
630 if (!dev->od_is_ost) {
631 rc = osd_mdt_init(env, dev);
640 rc = osd_index_backup_dir_init(env, dev);
650 static struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, u64 seq)
652 struct osd_obj_seq *osd_seq;
654 list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) {
655 if (osd_seq->oos_seq == seq)
661 static struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, u64 seq)
663 struct osd_obj_seq *osd_seq;
665 read_lock(&map->om_seq_list_lock);
666 osd_seq = osd_seq_find_locked(map, seq);
667 read_unlock(&map->om_seq_list_lock);
671 void osd_obj_map_fini(struct osd_device *dev)
673 osd_index_backup_dir_fini(dev);
679 * Update the specified OI mapping.
681 * \retval 1, changed nothing
682 * \retval 0, changed successfully
683 * \retval -ve, on error
685 static int osd_obj_update_entry(struct osd_thread_info *info,
686 struct osd_device *osd,
687 struct dentry *dir, const char *name,
688 const struct lu_fid *fid,
689 const struct osd_inode_id *id,
692 struct inode *parent = dir->d_inode;
693 struct dentry *child;
694 struct ldiskfs_dir_entry_2 *de;
695 struct buffer_head *bh;
697 struct dentry *dentry = &info->oti_obj_dentry;
698 struct osd_inode_id *oi_id = &info->oti_id3;
699 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
700 struct lu_fid *oi_fid = &lma->lma_self_fid;
706 LASSERT(th->h_transaction != NULL);
708 child = &info->oti_child_dentry;
709 child->d_parent = dir;
710 child->d_name.hash = 0;
711 child->d_name.name = name;
712 child->d_name.len = strlen(name);
714 dquot_initialize(parent);
716 bh = osd_ldiskfs_find_entry(parent, &child->d_name, &de, NULL, NULL);
718 GOTO(out, rc = PTR_ERR(bh));
720 if (le32_to_cpu(de->inode) == id->oii_ino)
723 osd_id_gen(oi_id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
724 inode = osd_iget(info, osd, oi_id);
727 if (rc == -ENOENT || rc == -ESTALE)
733 * The EA inode should NOT be in OI, old OI scrub may added
734 * such OI mapping by wrong, replace it.
736 if (unlikely(osd_is_ea_inode(inode))) {
741 rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
742 if (rc == -ENODATA) {
743 rc = osd_get_idif(info, inode, dentry, oi_fid);
744 if (rc > 0 || rc == -ENODATA) {
755 * If the OST-object has neither FID-in-LMA nor FID-in-ff, it is
756 * either a crashed object or a uninitialized one. Replace it.
758 if (oi_fid != NULL && lu_fid_eq(fid, oi_fid)) {
759 CERROR("%s: the FID "DFID" is used by two objects: "
760 "%u/%u %u/%u\n", osd_name(osd), PFID(fid),
761 oi_id->oii_ino, oi_id->oii_gen,
762 id->oii_ino, id->oii_gen);
763 GOTO(out, rc = -EEXIST);
766 if (fid_is_idif(fid) && oi_fid != NULL && fid_is_idif(oi_fid)) {
767 __u32 idx1 = fid_idif_ost_idx(fid);
768 __u32 idx2 = fid_idif_ost_idx(oi_fid);
769 struct ost_id *ostid = &info->oti_ostid;
770 struct lu_fid *tfid = &info->oti_fid3;
772 LASSERTF(idx1 == 0 || idx1 == osd->od_index,
773 "invalid given FID "DFID", not match the "
774 "device index %u\n", PFID(fid), osd->od_index);
777 if (idx1 == 0 && idx2 == osd->od_index) {
778 fid_to_ostid(fid, ostid);
779 ostid_to_fid(tfid, ostid, idx2);
780 if (lu_fid_eq(tfid, oi_fid)) {
781 CERROR("%s: the FID "DFID" is used by "
782 "two objects(2): %u/%u %u/%u\n",
783 osd_name(osd), PFID(fid),
784 oi_id->oii_ino, oi_id->oii_gen,
785 id->oii_ino, id->oii_gen);
787 GOTO(out, rc = -EEXIST);
789 } else if (idx2 == 0 && idx1 == osd->od_index) {
790 fid_to_ostid(oi_fid, ostid);
791 ostid_to_fid(tfid, ostid, idx1);
792 if (lu_fid_eq(tfid, fid)) {
793 CERROR("%s: the FID "DFID" is used by "
794 "two objects(2): %u/%u %u/%u\n",
795 osd_name(osd), PFID(fid),
796 oi_id->oii_ino, oi_id->oii_gen,
797 id->oii_ino, id->oii_gen);
799 GOTO(out, rc = -EEXIST);
807 * There may be temporary inconsistency: On one hand, the new
808 * object may be referenced by multiple entries, which is out
809 * of our control unless we traverse the whole /O completely,
810 * which is non-flat order and inefficient, should be avoided;
811 * On the other hand, the old object may become orphan if it
812 * is still valid. Since it was referenced by an invalid entry,
813 * making it as invisible temporary may be not worse. OI scrub
814 * will process it later.
816 rc = ldiskfs_journal_get_write_access(th, bh);
820 de->inode = cpu_to_le32(id->oii_ino);
821 rc = ldiskfs_handle_dirty_metadata(th, NULL, bh);
828 inode_unlock(parent);
832 static int osd_obj_del_entry(struct osd_thread_info *info,
833 struct osd_device *osd,
834 struct dentry *dird, char *name,
837 struct ldiskfs_dir_entry_2 *de;
838 struct buffer_head *bh;
839 struct dentry *child;
840 struct inode *dir = dird->d_inode;
846 LASSERT(th->h_transaction != NULL);
848 child = &info->oti_child_dentry;
849 child->d_name.hash = 0;
850 child->d_name.name = name;
851 child->d_name.len = strlen(name);
852 child->d_parent = dird;
853 child->d_inode = NULL;
855 dquot_initialize(dir);
857 bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
861 rc = ldiskfs_delete_entry(th, dir, de, bh);
869 static int osd_obj_add_entry(struct osd_thread_info *info,
870 struct osd_device *osd,
871 struct dentry *dir, char *name,
872 const struct osd_inode_id *id,
875 struct dentry *child;
881 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY))
885 LASSERT(th->h_transaction != NULL);
887 inode = info->oti_inode;
888 if (unlikely(inode == NULL)) {
889 struct ldiskfs_inode_info *lii;
894 inode = info->oti_inode = &lii->vfs_inode;
897 inode->i_sb = osd_sb(osd);
898 osd_id_to_inode(inode, id);
899 inode->i_mode = S_IFREG; /* for type in ldiskfs dir entry */
901 child = &info->oti_child_dentry;
902 child->d_name.hash = 0;
903 child->d_name.name = name;
904 child->d_name.len = strlen(name);
905 child->d_parent = dir;
906 child->d_inode = inode;
908 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
911 dquot_initialize(dir->d_inode);
912 inode_lock(dir->d_inode);
913 rc = osd_ldiskfs_add_entry(info, osd, th, child, inode, NULL);
914 inode_unlock(dir->d_inode);
920 * Use %llu for legacy OST sequences, but use %llx for new
921 * sequences names, so that the O/{seq}/dN/{oid} more closely
922 * follows the DFID/PFID format. This makes it easier to map from
923 * debug messages to objects in the future, and the legacy space
924 * of FID_SEQ_OST_MDT0 will be unused in the future.
926 static inline void osd_seq_name(char *seq_name, size_t name_size, u64 seq)
928 snprintf(seq_name, name_size,
929 (fid_seq_is_rsvd(seq) ||
930 fid_seq_is_mdt0(seq)) ? "%llu" : "%llx",
931 fid_seq_is_idif(seq) ? 0 : seq);
934 static inline void osd_oid_name(char *name, size_t name_size,
935 const struct lu_fid *fid, u64 id)
937 snprintf(name, name_size,
938 (fid_seq_is_rsvd(fid_seq(fid)) ||
939 fid_seq_is_mdt0(fid_seq(fid)) ||
940 fid_seq_is_idif(fid_seq(fid))) ? "%llu" : "%llx", id);
943 /* external locking is required */
944 static int osd_seq_load_locked(struct osd_thread_info *info,
945 struct osd_device *osd,
946 struct osd_obj_seq *osd_seq)
948 struct osd_obj_map *map = osd->od_ost_map;
949 struct dentry *seq_dir;
956 if (osd_seq->oos_root != NULL)
960 LASSERT(map->om_root);
962 osd_seq_name(dir_name, sizeof(dir_name), osd_seq->oos_seq);
964 seq_dir = simple_mkdir(info->oti_env, osd, map->om_root, NULL, dir_name,
965 LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0755, NULL);
967 GOTO(out_err, rc = PTR_ERR(seq_dir));
968 else if (seq_dir->d_inode == NULL)
969 GOTO(out_put, rc = -EFAULT);
971 osd_seq->oos_root = seq_dir;
973 LASSERT(osd_seq->oos_dirs == NULL);
974 OBD_ALLOC_PTR_ARRAY(osd_seq->oos_dirs, osd_seq->oos_subdir_count);
975 if (osd_seq->oos_dirs == NULL)
976 GOTO(out_put, rc = -ENOMEM);
978 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
981 snprintf(dir_name, sizeof(dir_name), "d%u", i);
982 dir = simple_mkdir(info->oti_env, osd, osd_seq->oos_root, NULL,
983 dir_name, LMAC_NOT_IN_OI | LMAC_FID_ON_OST,
986 GOTO(out_free, rc = PTR_ERR(dir));
987 } else if (dir->d_inode == NULL) {
989 GOTO(out_free, rc = -EFAULT);
992 osd_seq->oos_dirs[i] = dir;
997 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
998 if (osd_seq->oos_dirs[i] != NULL)
999 dput(osd_seq->oos_dirs[i]);
1001 OBD_FREE_PTR_ARRAY(osd_seq->oos_dirs,
1002 osd_seq->oos_subdir_count);
1005 osd_seq->oos_root = NULL;
1011 static struct osd_obj_seq *osd_seq_load(struct osd_thread_info *info,
1012 struct osd_device *osd, u64 seq)
1014 struct osd_obj_map *map;
1015 struct osd_obj_seq *osd_seq;
1020 map = osd->od_ost_map;
1022 LASSERT(map->om_root);
1024 osd_seq = osd_seq_find(map, seq);
1025 if (likely(osd_seq != NULL))
1028 /* Serializing init process */
1029 mutex_lock(&map->om_dir_init_mutex);
1031 /* Check whether the seq has been added */
1032 read_lock(&map->om_seq_list_lock);
1033 osd_seq = osd_seq_find_locked(map, seq);
1034 if (osd_seq != NULL) {
1035 read_unlock(&map->om_seq_list_lock);
1036 GOTO(cleanup, rc = 0);
1038 read_unlock(&map->om_seq_list_lock);
1040 OBD_ALLOC_PTR(osd_seq);
1041 if (osd_seq == NULL)
1042 GOTO(cleanup, rc = -ENOMEM);
1044 INIT_LIST_HEAD(&osd_seq->oos_seq_list);
1045 osd_seq->oos_seq = seq;
1047 * Init subdir count to be 32, but each seq can have
1048 * different subdir count
1050 osd_seq->oos_subdir_count = map->om_subdir_count;
1051 rc = osd_seq_load_locked(info, osd, osd_seq);
1055 write_lock(&map->om_seq_list_lock);
1056 list_add(&osd_seq->oos_seq_list, &map->om_seq_list);
1057 write_unlock(&map->om_seq_list_lock);
1060 mutex_unlock(&map->om_dir_init_mutex);
1062 if (osd_seq != NULL)
1063 OBD_FREE_PTR(osd_seq);
1064 RETURN(ERR_PTR(rc));
1070 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev,
1071 const struct lu_fid *fid, struct osd_inode_id *id)
1073 struct osd_obj_map *map;
1074 struct osd_obj_seq *osd_seq;
1075 struct dentry *d_seq;
1076 struct dentry *child;
1077 struct ost_id *ostid = &info->oti_ostid;
1080 struct ldiskfs_dir_entry_2 *de;
1081 struct buffer_head *bh;
1083 struct inode *inode;
1087 /* on the very first lookup we find and open directories */
1088 map = dev->od_ost_map;
1090 LASSERT(map->om_root);
1092 fid_to_ostid(fid, ostid);
1093 osd_seq = osd_seq_load(info, dev, ostid_seq(ostid));
1094 if (IS_ERR(osd_seq))
1095 RETURN(PTR_ERR(osd_seq));
1097 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1098 d_seq = osd_seq->oos_dirs[dirn];
1101 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1103 child = &info->oti_child_dentry;
1104 child->d_parent = d_seq;
1105 child->d_name.hash = 0;
1106 child->d_name.name = name;
1107 /* XXX: we can use rc from sprintf() instead of strlen() */
1108 child->d_name.len = strlen(name);
1110 dir = d_seq->d_inode;
1112 bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
1116 RETURN(PTR_ERR(bh));
1118 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
1121 inode = osd_iget(info, dev, id);
1122 if (IS_ERR(inode)) {
1123 int rc = PTR_ERR(inode);
1125 RETURN(rc == -ENOENT ? -ESTALE : rc);
1132 int osd_obj_map_insert(struct osd_thread_info *info,
1133 struct osd_device *osd,
1134 const struct lu_fid *fid,
1135 const struct osd_inode_id *id,
1138 struct osd_obj_map *map;
1139 struct osd_obj_seq *osd_seq;
1141 struct ost_id *ostid = &info->oti_ostid;
1148 map = osd->od_ost_map;
1151 /* map fid to seq:objid */
1152 fid_to_ostid(fid, ostid);
1154 oid = ostid_id(ostid);
1155 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1156 if (IS_ERR(osd_seq))
1157 RETURN(PTR_ERR(osd_seq));
1159 dirn = oid & (osd_seq->oos_subdir_count - 1);
1160 d = osd_seq->oos_dirs[dirn];
1163 osd_oid_name(name, sizeof(name), fid, oid);
1166 rc = osd_obj_add_entry(info, osd, d, name, id, th);
1167 if (rc == -EEXIST) {
1168 rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1169 if (unlikely(rc == -ENOENT))
1172 if (unlikely(rc == 1))
1179 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
1180 const struct lu_fid *fid, handle_t *th)
1182 struct osd_obj_map *map;
1183 struct osd_obj_seq *osd_seq;
1185 struct ost_id *ostid = &info->oti_ostid;
1191 map = osd->od_ost_map;
1194 /* map fid to seq:objid */
1195 fid_to_ostid(fid, ostid);
1197 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1198 if (IS_ERR(osd_seq))
1199 GOTO(cleanup, rc = PTR_ERR(osd_seq));
1201 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1202 d = osd_seq->oos_dirs[dirn];
1205 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1206 rc = osd_obj_del_entry(info, osd, d, name, th);
1211 int osd_obj_map_update(struct osd_thread_info *info,
1212 struct osd_device *osd,
1213 const struct lu_fid *fid,
1214 const struct osd_inode_id *id,
1217 struct osd_obj_seq *osd_seq;
1219 struct ost_id *ostid = &info->oti_ostid;
1225 fid_to_ostid(fid, ostid);
1226 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1227 if (IS_ERR(osd_seq))
1228 RETURN(PTR_ERR(osd_seq));
1230 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1231 d = osd_seq->oos_dirs[dirn];
1234 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1235 rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1240 int osd_obj_map_recover(struct osd_thread_info *info,
1241 struct osd_device *osd,
1242 struct inode *src_parent,
1243 struct dentry *src_child,
1244 const struct lu_fid *fid)
1246 struct osd_obj_seq *osd_seq;
1247 struct dentry *tgt_parent;
1248 struct dentry *tgt_child = &info->oti_child_dentry;
1250 struct inode *inode = src_child->d_inode;
1251 struct ost_id *ostid = &info->oti_ostid;
1253 struct ldiskfs_dir_entry_2 *de;
1254 struct buffer_head *bh;
1261 if (fid_is_last_id(fid)) {
1262 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1263 if (IS_ERR(osd_seq))
1264 RETURN(PTR_ERR(osd_seq));
1266 tgt_parent = osd_seq->oos_root;
1267 tgt_child->d_name.name = "LAST_ID";
1268 tgt_child->d_name.len = strlen("LAST_ID");
1270 fid_to_ostid(fid, ostid);
1271 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1272 if (IS_ERR(osd_seq))
1273 RETURN(PTR_ERR(osd_seq));
1275 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1276 tgt_parent = osd_seq->oos_dirs[dirn];
1277 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1278 tgt_child->d_name.name = name;
1279 tgt_child->d_name.len = strlen(name);
1281 LASSERT(tgt_parent != NULL);
1283 dir = tgt_parent->d_inode;
1284 tgt_child->d_name.hash = 0;
1285 tgt_child->d_parent = tgt_parent;
1286 tgt_child->d_inode = inode;
1288 /* The non-initialized src_child may be destroyed. */
1289 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
1290 osd_dto_credits_noquota[DTO_INDEX_DELETE] +
1291 osd_dto_credits_noquota[DTO_INDEX_INSERT] +
1292 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
1294 RETURN(PTR_ERR(jh));
1296 dquot_initialize(src_parent);
1297 dquot_initialize(dir);
1300 bh = osd_ldiskfs_find_entry(dir, &tgt_child->d_name, &de, NULL, NULL);
1303 * XXX: If some other object occupied the same slot. And If such
1304 * inode is zero-sized and with SUID+SGID, then means it is
1305 * a new created one. Maybe we can remove it and insert the
1306 * original one back to the /O/<seq>/d<x>. But there are
1307 * something to be considered:
1309 * 1) The OST-object under /lost+found has crashed LMA.
1310 * So it should not conflict with the current one.
1312 * 2) There are race conditions that: someone may just want
1313 * to modify the current one. Even if the OI scrub takes
1314 * the object lock when remove the current one, it still
1315 * cause the modification to be lost becasue the target
1316 * has been removed when the RPC service thread waiting
1319 * So keep it there before we have suitable solution.
1323 ldiskfs_journal_stop(jh);
1326 /* If the src object has never been modified, then remove it. */
1327 if (inode->i_size == 0 && inode->i_mode & S_ISUID &&
1328 inode->i_mode & S_ISGID) {
1329 rc = ll_vfs_unlink(src_parent, src_child);
1330 if (unlikely(rc == -ENOENT))
1337 bh = osd_ldiskfs_find_entry(src_parent, &src_child->d_name, &de,
1339 if (unlikely(IS_ERR(bh)))
1340 GOTO(unlock, rc = PTR_ERR(bh));
1342 rc = ldiskfs_delete_entry(jh, src_parent, de, bh);
1347 rc = osd_ldiskfs_add_entry(info, osd, jh, tgt_child, inode, NULL);
1353 ldiskfs_journal_stop(jh);
1357 static struct dentry *
1358 osd_object_spec_find(struct osd_thread_info *info, struct osd_device *osd,
1359 const struct lu_fid *fid, char **name)
1361 struct dentry *root = ERR_PTR(-ENOENT);
1363 if (fid_is_last_id(fid)) {
1364 struct osd_obj_seq *osd_seq;
1366 /* on creation of LAST_ID we create O/<seq> hierarchy */
1367 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1368 if (IS_ERR(osd_seq))
1369 RETURN((struct dentry *)osd_seq);
1372 root = osd_seq->oos_root;
1374 *name = osd_lf_fid2name(fid);
1376 CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid));
1377 else if ((*name)[0])
1378 root = osd_sb(osd)->s_root;
1384 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
1385 const struct lu_fid *fid, const struct osd_inode_id *id,
1388 struct dentry *root;
1394 root = osd_object_spec_find(info, osd, fid, &name);
1395 if (!IS_ERR(root)) {
1396 rc = osd_obj_update_entry(info, osd, root, name, fid, id, th);
1406 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
1407 const struct lu_fid *fid, const struct osd_inode_id *id,
1410 struct dentry *root;
1416 root = osd_object_spec_find(info, osd, fid, &name);
1417 if (!IS_ERR(root)) {
1418 rc = osd_obj_add_entry(info, osd, root, name, id, th);
1428 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
1429 const struct lu_fid *fid, struct osd_inode_id *id,
1430 enum oi_check_flags flags)
1432 struct dentry *root;
1433 struct dentry *dentry;
1434 struct inode *inode;
1440 if (fid_is_last_id(fid)) {
1441 struct osd_obj_seq *osd_seq;
1443 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1444 if (IS_ERR(osd_seq))
1445 RETURN(PTR_ERR(osd_seq));
1446 root = osd_seq->oos_root;
1449 root = osd_sb(osd)->s_root;
1450 name = osd_lf_fid2name(fid);
1451 if (name == NULL || strlen(name) == 0)
1455 dentry = osd_lookup_one_len_common(osd, name, root, strlen(name),
1457 if (!IS_ERR(dentry)) {
1458 inode = dentry->d_inode;
1460 if (is_bad_inode(inode)) {
1463 osd_id_gen(id, inode->i_ino,
1464 inode->i_generation);
1469 * if dentry is accessible after osd_compat_spec_insert it
1470 * will still contain NULL inode, so don't keep it in cache
1472 d_invalidate(dentry);
1479 #ifndef HAVE_BIO_INTEGRITY_ENABLED
1480 bool bio_integrity_enabled(struct bio *bio)
1482 struct blk_integrity *bi = blk_get_integrity(bio_get_disk(bio));
1484 if (bio_op(bio) != REQ_OP_READ && bio_op(bio) != REQ_OP_WRITE)
1487 if (!bio_sectors(bio))
1490 /* Already protected? */
1491 if (bio_integrity(bio))
1497 if (bio_data_dir(bio) == READ && bi->profile->verify_fn != NULL &&
1498 (bi->flags & BLK_INTEGRITY_VERIFY))
1501 if (bio_data_dir(bio) == WRITE && bi->profile->generate_fn != NULL &&
1502 (bi->flags & BLK_INTEGRITY_GENERATE))