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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osd/osd_handler.c
38 * Top-level entry points into osd module
40 * Author: Nikita Danilov <nikita@clusterfs.com>
41 * Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
44 #define DEBUG_SUBSYSTEM S_OSD
46 #include <linux/module.h>
47 #include <linux/user_namespace.h>
48 #ifdef HAVE_UIDGID_HEADER
49 # include <linux/uidgid.h>
52 /* LUSTRE_VERSION_CODE */
53 #include <lustre_ver.h>
54 /* prerequisite for linux/xattr.h */
55 #include <linux/types.h>
56 /* prerequisite for linux/xattr.h */
58 /* XATTR_{REPLACE,CREATE} */
59 #include <linux/xattr.h>
61 #include <ldiskfs/ldiskfs.h>
62 #include <ldiskfs/xattr.h>
65 * struct OBD_{ALLOC,FREE}*()
68 #include <obd_support.h>
69 /* struct ptlrpc_thread */
70 #include <lustre_net.h>
71 #include <lustre_fid.h>
73 #include <lustre_param.h>
75 #include "osd_internal.h"
76 #include "osd_dynlocks.h"
78 /* llo_* api support */
79 #include <md_object.h>
80 #include <lustre_quota.h>
82 #include <lustre_linkea.h>
85 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
86 "ldiskfs with parallel directory operations");
88 int ldiskfs_track_declares_assert;
89 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
90 "LBUG during tracking of declares");
92 /* Slab to allocate dynlocks */
93 struct kmem_cache *dynlock_cachep;
95 /* Slab to allocate osd_it_ea */
96 struct kmem_cache *osd_itea_cachep;
98 static struct lu_kmem_descr ldiskfs_caches[] = {
100 .ckd_cache = &dynlock_cachep,
101 .ckd_name = "dynlock_cache",
102 .ckd_size = sizeof(struct dynlock_handle)
105 .ckd_cache = &osd_itea_cachep,
106 .ckd_name = "osd_itea_cache",
107 .ckd_size = sizeof(struct osd_it_ea)
114 static const char dot[] = ".";
115 static const char dotdot[] = "..";
116 static const char remote_obj_dir[] = "REM_OBJ_DIR";
118 static const struct lu_object_operations osd_lu_obj_ops;
119 static const struct dt_object_operations osd_obj_ops;
120 static const struct dt_object_operations osd_obj_ea_ops;
121 static const struct dt_object_operations osd_obj_otable_it_ops;
122 static const struct dt_index_operations osd_index_iam_ops;
123 static const struct dt_index_operations osd_index_ea_ops;
125 int osd_trans_declare_op2rb[] = {
126 [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
127 [OSD_OT_PUNCH] = OSD_OT_MAX,
128 [OSD_OT_XATTR_SET] = OSD_OT_XATTR_SET,
129 [OSD_OT_CREATE] = OSD_OT_DESTROY,
130 [OSD_OT_DESTROY] = OSD_OT_CREATE,
131 [OSD_OT_REF_ADD] = OSD_OT_REF_DEL,
132 [OSD_OT_REF_DEL] = OSD_OT_REF_ADD,
133 [OSD_OT_WRITE] = OSD_OT_WRITE,
134 [OSD_OT_INSERT] = OSD_OT_DELETE,
135 [OSD_OT_DELETE] = OSD_OT_INSERT,
136 [OSD_OT_QUOTA] = OSD_OT_MAX,
139 static int osd_has_index(const struct osd_object *obj)
141 return obj->oo_dt.do_index_ops != NULL;
144 static int osd_object_invariant(const struct lu_object *l)
146 return osd_invariant(osd_obj(l));
150 * Concurrency: doesn't matter
154 * Concurrency: doesn't matter
156 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
158 struct osd_thread_info *oti = osd_oti_get(env);
159 return oti->oti_w_locks > 0 && o->oo_owner == env;
163 * Concurrency: doesn't access mutable data
165 static int osd_root_get(const struct lu_env *env,
166 struct dt_device *dev, struct lu_fid *f)
168 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
173 * OSD object methods.
177 * Concurrency: no concurrent access is possible that early in object
180 static struct lu_object *osd_object_alloc(const struct lu_env *env,
181 const struct lu_object_header *hdr,
184 struct osd_object *mo;
190 l = &mo->oo_dt.do_lu;
191 dt_object_init(&mo->oo_dt, NULL, d);
192 mo->oo_dt.do_ops = &osd_obj_ea_ops;
193 l->lo_ops = &osd_lu_obj_ops;
194 init_rwsem(&mo->oo_sem);
195 init_rwsem(&mo->oo_ext_idx_sem);
196 spin_lock_init(&mo->oo_guard);
203 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
204 struct dentry *dentry, struct lustre_mdt_attrs *lma)
208 CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
209 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
210 info->oti_mdt_attrs_old, LMA_OLD_SIZE);
212 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
213 memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
215 lustre_lma_swab(lma);
216 /* Check LMA compatibility */
217 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
218 CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
219 "for fid = "DFID", ino = %lu\n",
220 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
221 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
222 PFID(&lma->lma_self_fid), inode->i_ino);
225 } else if (rc == 0) {
233 * retrieve object from backend ext fs.
235 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
236 struct osd_inode_id *id)
238 struct inode *inode = NULL;
240 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
242 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
243 id->oii_ino, PTR_ERR(inode));
244 } else if (id->oii_gen != OSD_OII_NOGEN &&
245 inode->i_generation != id->oii_gen) {
246 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
247 "i_generation = %u\n",
248 id->oii_ino, id->oii_gen, inode->i_generation);
250 inode = ERR_PTR(-ESTALE);
251 } else if (inode->i_nlink == 0) {
252 /* due to parallel readdir and unlink,
253 * we can have dead inode here. */
254 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
256 inode = ERR_PTR(-ESTALE);
257 } else if (is_bad_inode(inode)) {
258 CWARN("%.16s: bad inode: ino = %u\n",
259 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
261 inode = ERR_PTR(-ENOENT);
263 if (id->oii_gen == OSD_OII_NOGEN)
264 osd_id_gen(id, inode->i_ino, inode->i_generation);
266 /* Do not update file c/mtime in ldiskfs.
267 * NB: we don't have any lock to protect this because we don't
268 * have reference on osd_object now, but contention with
269 * another lookup + attr_set can't happen in the tiny window
270 * between if (...) and set S_NOCMTIME. */
271 if (!(inode->i_flags & S_NOCMTIME))
272 inode->i_flags |= S_NOCMTIME;
277 static struct inode *
278 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
279 struct osd_inode_id *id, struct lu_fid *fid)
281 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
285 inode = osd_iget(info, dev, id);
289 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
291 *fid = lma->lma_self_fid;
292 } else if (rc == -ENODATA) {
293 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
294 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
296 lu_igif_build(fid, inode->i_ino, inode->i_generation);
304 static struct inode *osd_iget_check(struct osd_thread_info *info,
305 struct osd_device *dev,
306 const struct lu_fid *fid,
307 struct osd_inode_id *id,
314 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
317 if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) {
318 CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
327 if (is_bad_inode(inode)) {
330 CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
338 if (id->oii_gen != OSD_OII_NOGEN &&
339 inode->i_generation != id->oii_gen) {
342 CDEBUG(D_INODE, "unmatched inode: ino = %u, "
343 "oii_gen = %u, i_generation = %u\n",
344 id->oii_ino, id->oii_gen, inode->i_generation);
352 if (inode->i_nlink == 0) {
355 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
365 struct osd_inode_id saved_id = *id;
367 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
369 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
370 /* XXX: There are some possible cases:
372 * Backup/restore caused the OI invalid.
374 * Someone unlinked the object but NOT removed
375 * the OI mapping, such as mount target device
376 * as ldiskfs, and modify something directly.
378 * Someone just removed the object between the
379 * former oi_lookup and the iget. It is normal.
380 * 4. Other failure cases.
382 * Generally, when the device is mounted, it will
383 * auto check whether the system is restored from
384 * file-level backup or not. We trust such detect
385 * to distinguish the 1st case from the 2nd case. */
387 if (!IS_ERR(inode) && inode->i_generation != 0 &&
388 inode->i_generation == id->oii_gen) {
391 __u32 level = D_LFSCK;
394 if (!thread_is_running(&dev->od_scrub.os_thread))
397 CDEBUG(level, "%s: the OI mapping for the FID "
398 DFID" become inconsistent, the given ID "
399 "%u/%u, the ID in OI mapping %u/%u\n",
400 osd_name(dev), PFID(fid),
401 saved_id.oii_ino, saved_id.oii_gen,
402 id->oii_ino, id->oii_ino);
406 if (id->oii_gen == OSD_OII_NOGEN)
407 osd_id_gen(id, inode->i_ino, inode->i_generation);
409 /* Do not update file c/mtime in ldiskfs.
410 * NB: we don't have any lock to protect this because we don't
411 * have reference on osd_object now, but contention with
412 * another lookup + attr_set can't happen in the tiny window
413 * between if (...) and set S_NOCMTIME. */
414 if (!(inode->i_flags & S_NOCMTIME))
415 inode->i_flags |= S_NOCMTIME;
432 * \retval +v: new filter_fid, does not contain self-fid
433 * \retval 0: filter_fid_old, contains self-fid
434 * \retval -v: other failure cases
436 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
437 struct dentry *dentry, struct lu_fid *fid)
439 struct filter_fid_old *ff = &info->oti_ff;
440 struct ost_id *ostid = &info->oti_ostid;
443 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
444 if (rc == sizeof(*ff)) {
446 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
447 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
448 /* XXX: should use real OST index in the future. LU-3569 */
449 ostid_to_fid(fid, ostid, 0);
450 } else if (rc == sizeof(struct filter_fid)) {
452 } else if (rc >= 0) {
459 static int osd_lma_self_repair(struct osd_thread_info *info,
460 struct osd_device *osd, struct inode *inode,
461 const struct lu_fid *fid, __u32 compat)
466 LASSERT(current->journal_info == NULL);
468 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
469 osd_dto_credits_noquota[DTO_XATTR_SET]);
472 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
477 rc = osd_ea_fid_set(info, inode, fid, compat, 0);
479 CWARN("%s: cannot self repair the LMA: rc = %d\n",
481 ldiskfs_journal_stop(jh);
485 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
487 struct osd_thread_info *info = osd_oti_get(env);
488 struct osd_device *osd = osd_obj2dev(obj);
489 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
490 struct inode *inode = obj->oo_inode;
491 struct dentry *dentry = &info->oti_obj_dentry;
492 struct lu_fid *fid = NULL;
493 const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
497 CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
498 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
499 info->oti_mdt_attrs_old, LMA_OLD_SIZE);
500 if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
501 fid = &lma->lma_self_fid;
502 rc = osd_get_idif(info, inode, dentry, fid);
503 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
504 /* For the given OST-object, if it has neither LMA nor
505 * FID in XATTR_NAME_FID, then the given FID (which is
506 * contained in the @obj, from client RPC for locating
507 * the OST-object) is trusted. We use it to generate
509 osd_lma_self_repair(info, osd, inode, rfid,
515 if (unlikely(rc == -ENODATA))
523 lustre_lma_swab(lma);
524 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
525 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
526 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
527 "fid = "DFID", ino = %lu\n", osd_name(osd),
528 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
529 PFID(rfid), inode->i_ino);
532 fid = &lma->lma_self_fid;
536 if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
537 __u32 level = D_LFSCK;
539 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
540 struct ost_id *oi = &info->oti_ostid;
541 struct lu_fid *fid1 = &info->oti_fid3;
542 __u32 idx = fid_idif_ost_idx(rfid);
544 /* For old IDIF, the OST index is not part of the IDIF,
545 * Means that different OSTs may have the same IDIFs.
546 * Under such case, we need to make some compatible
547 * check to make sure to trigger OI scrub properly. */
548 if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
549 /* Given @rfid is new, LMA is old. */
550 fid_to_ostid(fid, oi);
551 ostid_to_fid(fid1, oi, idx);
552 if (lu_fid_eq(fid1, rfid)) {
553 if (osd->od_index_in_idif)
554 osd_lma_self_repair(info, osd,
564 if (!thread_is_running(&osd->od_scrub.os_thread))
567 CDEBUG(level, "%s: FID "DFID" != self_fid "DFID"\n",
568 osd_name(osd), PFID(rfid), PFID(fid));
574 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
575 const struct lu_fid *fid,
576 const struct lu_object_conf *conf)
578 struct osd_thread_info *info;
579 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
580 struct osd_device *dev;
581 struct osd_idmap_cache *oic;
582 struct osd_inode_id *id;
584 struct osd_scrub *scrub;
585 struct scrub_file *sf;
589 bool triggered = false;
592 LINVRNT(osd_invariant(obj));
593 LASSERT(obj->oo_inode == NULL);
594 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
597 scrub = &dev->od_scrub;
598 sf = &scrub->os_file;
599 info = osd_oti_get(env);
601 oic = &info->oti_cache;
603 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
606 /* For the object is created as locking anchor, or for the object to
607 * be created on disk. No need to osd_oi_lookup() at here because FID
608 * shouldn't never be re-used, if it's really a duplicate FID from
609 * unexpected reason, we should be able to detect it later by calling
610 * do_create->osd_oi_insert(). */
611 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
612 GOTO(out, result = 0);
614 /* Search order: 1. per-thread cache. */
615 if (lu_fid_eq(fid, &oic->oic_fid) &&
616 likely(oic->oic_dev == dev)) {
622 if (!list_empty(&scrub->os_inconsistent_items)) {
623 /* Search order: 2. OI scrub pending list. */
624 result = osd_oii_lookup(dev, fid, id);
629 /* Search order: 3. OI files. */
630 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
631 if (result == -ENOENT) {
632 if (!fid_is_norm(fid) ||
633 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
634 !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
636 GOTO(out, result = 0);
647 inode = osd_iget_check(info, dev, fid, id, in_oi);
649 result = PTR_ERR(inode);
650 if (result == -ENOENT || result == -ESTALE) {
652 fid_zero(&oic->oic_fid);
654 GOTO(out, result = -ENOENT);
655 } else if (result == -EREMCHG) {
659 fid_zero(&oic->oic_fid);
661 if (unlikely(triggered))
662 GOTO(out, result = saved);
665 if (thread_is_running(&scrub->os_thread)) {
666 result = -EINPROGRESS;
667 } else if (!dev->od_noscrub) {
668 /* Since we do not know the right OI mapping,
669 * we have to trigger OI scrub to scan the
671 result = osd_scrub_start(dev, SS_AUTO_FULL |
672 SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
673 CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI "
674 "scrub by RPC for "DFID", rc = %d [1]\n",
675 osd_name(dev), PFID(fid),result);
676 if (result == 0 || result == -EALREADY)
677 result = -EINPROGRESS;
682 /* We still have chance to get the valid inode: for the
683 * object which is referenced by remote name entry, the
684 * object on the local MDT will be linked under the dir
685 * of "/REMOTE_PARENT_DIR" with its FID string as name.
687 * We do not know whether the object for the given FID
688 * is referenced by some remote name entry or not, and
689 * especially for DNE II, a multiple-linked object may
690 * have many name entries reside on many MDTs.
692 * To simplify the operation, OSD will not distinguish
693 * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
694 * only happened for the RPC from other MDT during the
695 * OI scrub, or for the client side RPC with FID only,
696 * such as FID to path, or from old connected client. */
698 result = osd_lookup_in_remote_parent(info, dev,
711 obj->oo_inode = inode;
712 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
714 result = osd_check_lma(env, obj);
717 obj->oo_inode = NULL;
718 if (result == -EREMCHG) {
720 result = osd_oi_lookup(info, dev, fid, id,
723 fid_zero(&oic->oic_fid);
734 obj->oo_compat_dot_created = 1;
735 obj->oo_compat_dotdot_created = 1;
737 if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
738 GOTO(out, result = 0);
740 LASSERT(obj->oo_hl_head == NULL);
741 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
742 if (obj->oo_hl_head == NULL) {
743 obj->oo_inode = NULL;
745 GOTO(out, result = -ENOMEM);
747 GOTO(out, result = 0);
750 LINVRNT(osd_invariant(obj));
755 * Concurrency: shouldn't matter.
757 static void osd_object_init0(struct osd_object *obj)
759 LASSERT(obj->oo_inode != NULL);
760 obj->oo_dt.do_body_ops = &osd_body_ops;
761 obj->oo_dt.do_lu.lo_header->loh_attr |=
762 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
766 * Concurrency: no concurrent access is possible that early in object
769 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
770 const struct lu_object_conf *conf)
772 struct osd_object *obj = osd_obj(l);
775 LINVRNT(osd_invariant(obj));
777 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
778 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
779 l->lo_header->loh_attr |= LOHA_EXISTS;
783 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
784 obj->oo_dt.do_body_ops = &osd_body_ops_new;
785 if (result == 0 && obj->oo_inode != NULL)
786 osd_object_init0(obj);
788 LINVRNT(osd_invariant(obj));
793 * Concurrency: no concurrent access is possible that late in object
796 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
798 struct osd_object *obj = osd_obj(l);
800 LINVRNT(osd_invariant(obj));
802 dt_object_fini(&obj->oo_dt);
803 if (obj->oo_hl_head != NULL)
804 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
809 * Concurrency: no concurrent access is possible that late in object
812 static void osd_index_fini(struct osd_object *o)
814 struct iam_container *bag;
816 if (o->oo_dir != NULL) {
817 bag = &o->oo_dir->od_container;
818 if (o->oo_inode != NULL) {
819 if (bag->ic_object == o->oo_inode)
820 iam_container_fini(bag);
822 OBD_FREE_PTR(o->oo_dir);
828 * Concurrency: no concurrent access is possible that late in object
829 * life-cycle (for all existing callers, that is. New callers have to provide
830 * their own locking.)
832 static int osd_inode_unlinked(const struct inode *inode)
834 return inode->i_nlink == 0;
838 OSD_TXN_OI_DELETE_CREDITS = 20,
839 OSD_TXN_INODE_DELETE_CREDITS = 20
846 #if OSD_THANDLE_STATS
848 * Set time when the handle is allocated
850 static void osd_th_alloced(struct osd_thandle *oth)
852 oth->oth_alloced = cfs_time_current();
856 * Set time when the handle started
858 static void osd_th_started(struct osd_thandle *oth)
860 oth->oth_started = cfs_time_current();
864 * Helper function to convert time interval to microseconds packed in
867 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
871 cfs_duration_usec(cfs_time_sub(end, start), &val);
872 return val.tv_sec * 1000000 + val.tv_usec;
876 * Check whether the we deal with this handle for too long.
878 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
879 cfs_time_t alloced, cfs_time_t started,
882 cfs_time_t now = cfs_time_current();
884 LASSERT(dev != NULL);
886 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
887 interval_to_usec(alloced, started));
888 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
889 interval_to_usec(started, closed));
890 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
891 interval_to_usec(closed, now));
893 if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
894 CWARN("transaction handle %p was open for too long: "
896 "alloced "CFS_TIME_T" ,"
897 "started "CFS_TIME_T" ,"
898 "closed "CFS_TIME_T"\n",
899 oth, now, alloced, started, closed);
900 libcfs_debug_dumpstack(NULL);
904 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
906 cfs_time_t __closed = cfs_time_current(); \
907 cfs_time_t __alloced = oth->oth_alloced; \
908 cfs_time_t __started = oth->oth_started; \
911 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
914 #else /* OSD_THANDLE_STATS */
916 #define osd_th_alloced(h) do {} while(0)
917 #define osd_th_started(h) do {} while(0)
918 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
920 #endif /* OSD_THANDLE_STATS */
923 * Concurrency: doesn't access mutable data.
925 static int osd_param_is_not_sane(const struct osd_device *dev,
926 const struct thandle *th)
928 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
930 return oh->ot_credits > osd_transaction_size(dev);
934 * Concurrency: shouldn't matter.
936 static void osd_trans_commit_cb(struct super_block *sb,
937 struct ldiskfs_journal_cb_entry *jcb, int error)
939 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
940 struct thandle *th = &oh->ot_super;
941 struct lu_device *lud = &th->th_dev->dd_lu_dev;
942 struct dt_txn_commit_cb *dcb, *tmp;
944 LASSERT(oh->ot_handle == NULL);
947 CERROR("transaction @0x%p commit error: %d\n", th, error);
949 dt_txn_hook_commit(th);
951 /* call per-transaction callbacks if any */
952 list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
953 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
954 "commit callback entry: magic=%x name='%s'\n",
955 dcb->dcb_magic, dcb->dcb_name);
956 list_del_init(&dcb->dcb_linkage);
957 dcb->dcb_func(NULL, th, dcb, error);
960 lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
964 lu_context_exit(&th->th_ctx);
965 lu_context_fini(&th->th_ctx);
969 static struct thandle *osd_trans_create(const struct lu_env *env,
972 struct osd_thread_info *oti = osd_oti_get(env);
973 struct osd_iobuf *iobuf = &oti->oti_iobuf;
974 struct osd_thandle *oh;
978 /* on pending IO in this thread should left from prev. request */
979 LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
981 th = ERR_PTR(-ENOMEM);
982 OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
984 oh->ot_quota_trans = &oti->oti_quota_trans;
985 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
989 th->th_tags = LCT_TX_HANDLE;
991 INIT_LIST_HEAD(&oh->ot_dcb_list);
994 memset(oti->oti_declare_ops, 0,
995 sizeof(oti->oti_declare_ops));
996 memset(oti->oti_declare_ops_rb, 0,
997 sizeof(oti->oti_declare_ops_rb));
998 memset(oti->oti_declare_ops_cred, 0,
999 sizeof(oti->oti_declare_ops_cred));
1000 oti->oti_rollback = false;
1006 * Concurrency: shouldn't matter.
1008 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1011 struct osd_thread_info *oti = osd_oti_get(env);
1012 struct osd_device *dev = osd_dt_dev(d);
1014 struct osd_thandle *oh;
1019 LASSERT(current->journal_info == NULL);
1021 oh = container_of0(th, struct osd_thandle, ot_super);
1022 LASSERT(oh != NULL);
1023 LASSERT(oh->ot_handle == NULL);
1025 rc = dt_txn_hook_start(env, d, th);
1029 if (unlikely(osd_param_is_not_sane(dev, th))) {
1030 static unsigned long last_printed;
1031 static int last_credits;
1033 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1034 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1036 osd_journal(dev)->j_max_transaction_buffers);
1037 CWARN(" create: %u/%u, destroy: %u/%u\n",
1038 oti->oti_declare_ops[OSD_OT_CREATE],
1039 oti->oti_declare_ops_cred[OSD_OT_CREATE],
1040 oti->oti_declare_ops[OSD_OT_DESTROY],
1041 oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
1042 CWARN(" attr_set: %u/%u, xattr_set: %u/%u\n",
1043 oti->oti_declare_ops[OSD_OT_ATTR_SET],
1044 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1045 oti->oti_declare_ops[OSD_OT_XATTR_SET],
1046 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
1047 CWARN(" write: %u/%u, punch: %u/%u, quota %u/%u\n",
1048 oti->oti_declare_ops[OSD_OT_WRITE],
1049 oti->oti_declare_ops_cred[OSD_OT_WRITE],
1050 oti->oti_declare_ops[OSD_OT_PUNCH],
1051 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1052 oti->oti_declare_ops[OSD_OT_QUOTA],
1053 oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
1054 CWARN(" insert: %u/%u, delete: %u/%u\n",
1055 oti->oti_declare_ops[OSD_OT_INSERT],
1056 oti->oti_declare_ops_cred[OSD_OT_INSERT],
1057 oti->oti_declare_ops[OSD_OT_DELETE],
1058 oti->oti_declare_ops_cred[OSD_OT_DELETE]);
1059 CWARN(" ref_add: %u/%u, ref_del: %u/%u\n",
1060 oti->oti_declare_ops[OSD_OT_REF_ADD],
1061 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1062 oti->oti_declare_ops[OSD_OT_REF_DEL],
1063 oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
1065 if (last_credits != oh->ot_credits &&
1066 time_after(jiffies, last_printed +
1067 msecs_to_jiffies(60 * MSEC_PER_SEC))) {
1068 libcfs_debug_dumpstack(NULL);
1069 last_credits = oh->ot_credits;
1070 last_printed = jiffies;
1072 /* XXX Limit the credits to 'max_transaction_buffers', and
1073 * let the underlying filesystem to catch the error if
1074 * we really need so many credits.
1076 * This should be removed when we can calculate the
1077 * credits precisely. */
1078 oh->ot_credits = osd_transaction_size(dev);
1082 * XXX temporary stuff. Some abstraction layer should
1085 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1089 LASSERT(oti->oti_txns == 0);
1090 lu_context_init(&th->th_ctx, th->th_tags);
1091 lu_context_enter(&th->th_ctx);
1093 lu_device_get(&d->dd_lu_dev);
1094 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1105 static int osd_seq_exists(const struct lu_env *env,
1106 struct osd_device *osd, u64 seq)
1108 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1109 struct seq_server_site *ss = osd_seq_site(osd);
1113 LASSERT(ss != NULL);
1114 LASSERT(ss->ss_server_fld != NULL);
1116 rc = osd_fld_lookup(env, osd, seq, range);
1119 CERROR("%s: can't lookup FLD sequence "LPX64
1120 ": rc = %d\n", osd_name(osd), seq, rc);
1124 RETURN(ss->ss_node_id == range->lsr_index);
1128 * Concurrency: shouldn't matter.
1130 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1134 struct osd_thandle *oh;
1135 struct osd_thread_info *oti = osd_oti_get(env);
1136 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1137 struct osd_device *osd = osd_dt_dev(th->th_dev);
1138 struct qsd_instance *qsd = osd->od_quota_slave;
1139 struct lquota_trans *qtrans;
1142 oh = container_of0(th, struct osd_thandle, ot_super);
1144 qtrans = oh->ot_quota_trans;
1145 oh->ot_quota_trans = NULL;
1147 if (oh->ot_handle != NULL) {
1148 handle_t *hdl = oh->ot_handle;
1151 * add commit callback
1152 * notice we don't do this in osd_trans_start()
1153 * as underlying transaction can change during truncate
1155 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1158 LASSERT(oti->oti_txns == 1);
1160 rc = dt_txn_hook_stop(env, th);
1162 CERROR("%s: failed in transaction hook: rc = %d\n",
1165 /* hook functions might modify th_sync */
1166 hdl->h_sync = th->th_sync;
1168 oh->ot_handle = NULL;
1169 OSD_CHECK_SLOW_TH(oh, osd, rc = ldiskfs_journal_stop(hdl));
1171 CERROR("%s: failed to stop transaction: rc = %d\n",
1177 /* inform the quota slave device that the transaction is stopping */
1178 qsd_op_end(env, qsd, qtrans);
1180 /* as we want IO to journal and data IO be concurrent, we don't block
1181 * awaiting data IO completion in osd_do_bio(), instead we wait here
1182 * once transaction is submitted to the journal. all reqular requests
1183 * don't do direct IO (except read/write), thus this wait_event becomes
1186 * IMPORTANT: we have to wait till any IO submited by the thread is
1187 * completed otherwise iobuf may be corrupted by different request
1189 wait_event(iobuf->dr_wait,
1190 atomic_read(&iobuf->dr_numreqs) == 0);
1191 osd_fini_iobuf(osd, iobuf);
1193 rc = iobuf->dr_error;
1198 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1200 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1203 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1204 LASSERT(&dcb->dcb_func != NULL);
1205 list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1211 * Called just before object is freed. Releases all resources except for
1212 * object itself (that is released by osd_object_free()).
1214 * Concurrency: no concurrent access is possible that late in object
1217 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1219 struct osd_object *obj = osd_obj(l);
1220 struct inode *inode = obj->oo_inode;
1222 LINVRNT(osd_invariant(obj));
1225 * If object is unlinked remove fid->ino mapping from object index.
1228 osd_index_fini(obj);
1229 if (inode != NULL) {
1230 struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave;
1231 qid_t uid = i_uid_read(inode);
1232 qid_t gid = i_gid_read(inode);
1235 obj->oo_inode = NULL;
1238 struct osd_thread_info *info = osd_oti_get(env);
1239 struct lquota_id_info *qi = &info->oti_qi;
1241 /* Release granted quota to master if necessary */
1242 qi->lqi_id.qid_uid = uid;
1243 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1245 qi->lqi_id.qid_uid = gid;
1246 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1252 * Concurrency: ->loo_object_release() is called under site spin-lock.
1254 static void osd_object_release(const struct lu_env *env,
1255 struct lu_object *l)
1260 * Concurrency: shouldn't matter.
1262 static int osd_object_print(const struct lu_env *env, void *cookie,
1263 lu_printer_t p, const struct lu_object *l)
1265 struct osd_object *o = osd_obj(l);
1266 struct iam_descr *d;
1268 if (o->oo_dir != NULL)
1269 d = o->oo_dir->od_container.ic_descr;
1272 return (*p)(env, cookie,
1273 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1275 o->oo_inode ? o->oo_inode->i_ino : 0UL,
1276 o->oo_inode ? o->oo_inode->i_generation : 0,
1277 d ? d->id_ops->id_name : "plain");
1280 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1283 * Concurrency: shouldn't matter.
1285 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1286 struct obd_statfs *sfs)
1288 struct osd_device *osd = osd_dt_dev(d);
1289 struct super_block *sb = osd_sb(osd);
1290 struct kstatfs *ksfs;
1293 if (unlikely(osd->od_mnt == NULL))
1294 return -EINPROGRESS;
1296 /* osd_lproc.c call this without env, allocate ksfs for that case */
1297 if (unlikely(env == NULL)) {
1298 OBD_ALLOC_PTR(ksfs);
1302 ksfs = &osd_oti_get(env)->oti_ksfs;
1305 spin_lock(&osd->od_osfs_lock);
1306 result = sb->s_op->statfs(sb->s_root, ksfs);
1307 if (likely(result == 0)) { /* N.B. statfs can't really fail */
1308 statfs_pack(sfs, ksfs);
1309 if (sb->s_flags & MS_RDONLY)
1310 sfs->os_state = OS_STATE_READONLY;
1313 spin_unlock(&osd->od_osfs_lock);
1315 if (unlikely(env == NULL))
1318 /* Reserve a small amount of space for local objects like last_rcvd,
1319 * llog, quota files, ... */
1320 if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1323 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1324 /** Take out metadata overhead for indirect blocks */
1325 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1332 * Estimate space needed for file creations. We assume the largest filename
1333 * which is 2^64 - 1, hence a filename of 20 chars.
1334 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1336 #ifdef __LDISKFS_DIR_REC_LEN
1337 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1339 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1343 * Concurrency: doesn't access mutable data.
1345 static void osd_conf_get(const struct lu_env *env,
1346 const struct dt_device *dev,
1347 struct dt_device_param *param)
1349 struct super_block *sb = osd_sb(osd_dt_dev(dev));
1353 * XXX should be taken from not-yet-existing fs abstraction layer.
1355 param->ddp_max_name_len = LDISKFS_NAME_LEN;
1356 param->ddp_max_nlink = LDISKFS_LINK_MAX;
1357 param->ddp_block_shift = sb->s_blocksize_bits;
1358 param->ddp_mount_type = LDD_MT_LDISKFS;
1359 param->ddp_maxbytes = sb->s_maxbytes;
1360 /* Overhead estimate should be fairly accurate, so we really take a tiny
1361 * error margin which also avoids fragmenting the filesystem too much */
1362 param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1363 /* inode are statically allocated, so per-inode space consumption
1364 * is the space consumed by the directory entry */
1365 param->ddp_inodespace = PER_OBJ_USAGE;
1366 /* per-fragment overhead to be used by the client code */
1367 param->ddp_grant_frag = 6 * LDISKFS_BLOCK_SIZE(sb);
1368 param->ddp_mntopts = 0;
1369 if (test_opt(sb, XATTR_USER))
1370 param->ddp_mntopts |= MNTOPT_USERXATTR;
1371 if (test_opt(sb, POSIX_ACL))
1372 param->ddp_mntopts |= MNTOPT_ACL;
1374 /* LOD might calculate the max stripe count based on max_ea_size,
1375 * so we need take account in the overhead as well,
1376 * xattr_header + magic + xattr_entry_head */
1377 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1378 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1380 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1381 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1382 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1386 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1390 * Concurrency: shouldn't matter.
1392 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1396 CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1398 rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1400 CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
1401 LUSTRE_OSD_LDISKFS_NAME, rc);
1407 * Start commit for OSD device.
1409 * An implementation of dt_commit_async method for OSD device.
1410 * Asychronously starts underlayng fs sync and thereby a transaction
1413 * \param env environment
1414 * \param d dt device
1416 * \see dt_device_operations
1418 static int osd_commit_async(const struct lu_env *env,
1419 struct dt_device *d)
1421 struct super_block *s = osd_sb(osd_dt_dev(d));
1424 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1425 RETURN(s->s_op->sync_fs(s, 0));
1429 * Concurrency: shouldn't matter.
1432 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1434 struct super_block *sb = osd_sb(osd_dt_dev(d));
1435 struct block_device *dev = sb->s_bdev;
1436 #ifdef HAVE_DEV_SET_RDONLY
1437 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1440 int rc = -EOPNOTSUPP;
1444 #ifdef HAVE_DEV_SET_RDONLY
1445 CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1447 if (jdev && (jdev != dev)) {
1448 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1450 dev_set_rdonly(jdev);
1452 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1453 dev_set_rdonly(dev);
1455 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1456 osd_dt_dev(d)->od_svname, (long)dev, rc);
1462 * Note: we do not count into QUOTA here.
1463 * If we mount with --data_journal we may need more.
1465 const int osd_dto_credits_noquota[DTO_NR] = {
1468 * INDEX_EXTRA_TRANS_BLOCKS(8) +
1469 * SINGLEDATA_TRANS_BLOCKS(8)
1470 * XXX Note: maybe iam need more, since iam have more level than
1473 [DTO_INDEX_INSERT] = 16,
1476 * just modify a single entry, probably merge few within a block
1478 [DTO_INDEX_DELETE] = 1,
1482 [DTO_INDEX_UPDATE] = 16,
1484 * 4(inode, inode bits, groups, GDT)
1485 * notice: OI updates are counted separately with DTO_INDEX_INSERT
1487 [DTO_OBJECT_CREATE] = 4,
1489 * 4(inode, inode bits, groups, GDT)
1490 * notice: OI updates are counted separately with DTO_INDEX_DELETE
1492 [DTO_OBJECT_DELETE] = 4,
1494 * Attr set credits (inode)
1496 [DTO_ATTR_SET_BASE] = 1,
1498 * Xattr set. The same as xattr of EXT3.
1499 * DATA_TRANS_BLOCKS(14)
1500 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1501 * are also counted in. Do not know why?
1503 [DTO_XATTR_SET] = 14,
1505 * credits for inode change during write.
1507 [DTO_WRITE_BASE] = 3,
1509 * credits for single block write.
1511 [DTO_WRITE_BLOCK] = 14,
1513 * Attr set credits for chown.
1514 * This is extra credits for setattr, and it is null without quota
1516 [DTO_ATTR_SET_CHOWN] = 0
1519 static const struct dt_device_operations osd_dt_ops = {
1520 .dt_root_get = osd_root_get,
1521 .dt_statfs = osd_statfs,
1522 .dt_trans_create = osd_trans_create,
1523 .dt_trans_start = osd_trans_start,
1524 .dt_trans_stop = osd_trans_stop,
1525 .dt_trans_cb_add = osd_trans_cb_add,
1526 .dt_conf_get = osd_conf_get,
1527 .dt_sync = osd_sync,
1529 .dt_commit_async = osd_commit_async,
1532 static void osd_object_read_lock(const struct lu_env *env,
1533 struct dt_object *dt, unsigned role)
1535 struct osd_object *obj = osd_dt_obj(dt);
1536 struct osd_thread_info *oti = osd_oti_get(env);
1538 LINVRNT(osd_invariant(obj));
1540 LASSERT(obj->oo_owner != env);
1541 down_read_nested(&obj->oo_sem, role);
1543 LASSERT(obj->oo_owner == NULL);
1547 static void osd_object_write_lock(const struct lu_env *env,
1548 struct dt_object *dt, unsigned role)
1550 struct osd_object *obj = osd_dt_obj(dt);
1551 struct osd_thread_info *oti = osd_oti_get(env);
1553 LINVRNT(osd_invariant(obj));
1555 LASSERT(obj->oo_owner != env);
1556 down_write_nested(&obj->oo_sem, role);
1558 LASSERT(obj->oo_owner == NULL);
1559 obj->oo_owner = env;
1563 static void osd_object_read_unlock(const struct lu_env *env,
1564 struct dt_object *dt)
1566 struct osd_object *obj = osd_dt_obj(dt);
1567 struct osd_thread_info *oti = osd_oti_get(env);
1569 LINVRNT(osd_invariant(obj));
1571 LASSERT(oti->oti_r_locks > 0);
1573 up_read(&obj->oo_sem);
1576 static void osd_object_write_unlock(const struct lu_env *env,
1577 struct dt_object *dt)
1579 struct osd_object *obj = osd_dt_obj(dt);
1580 struct osd_thread_info *oti = osd_oti_get(env);
1582 LINVRNT(osd_invariant(obj));
1584 LASSERT(obj->oo_owner == env);
1585 LASSERT(oti->oti_w_locks > 0);
1587 obj->oo_owner = NULL;
1588 up_write(&obj->oo_sem);
1591 static int osd_object_write_locked(const struct lu_env *env,
1592 struct dt_object *dt)
1594 struct osd_object *obj = osd_dt_obj(dt);
1596 LINVRNT(osd_invariant(obj));
1598 return obj->oo_owner == env;
1601 static struct timespec *osd_inode_time(const struct lu_env *env,
1602 struct inode *inode, __u64 seconds)
1604 struct osd_thread_info *oti = osd_oti_get(env);
1605 struct timespec *t = &oti->oti_time;
1607 t->tv_sec = seconds;
1609 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1614 static void osd_inode_getattr(const struct lu_env *env,
1615 struct inode *inode, struct lu_attr *attr)
1617 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1618 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1619 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1622 attr->la_atime = LTIME_S(inode->i_atime);
1623 attr->la_mtime = LTIME_S(inode->i_mtime);
1624 attr->la_ctime = LTIME_S(inode->i_ctime);
1625 attr->la_mode = inode->i_mode;
1626 attr->la_size = i_size_read(inode);
1627 attr->la_blocks = inode->i_blocks;
1628 attr->la_uid = i_uid_read(inode);
1629 attr->la_gid = i_gid_read(inode);
1630 attr->la_flags = LDISKFS_I(inode)->i_flags;
1631 attr->la_nlink = inode->i_nlink;
1632 attr->la_rdev = inode->i_rdev;
1633 attr->la_blksize = 1 << inode->i_blkbits;
1634 attr->la_blkbits = inode->i_blkbits;
1637 static int osd_attr_get(const struct lu_env *env,
1638 struct dt_object *dt,
1639 struct lu_attr *attr)
1641 struct osd_object *obj = osd_dt_obj(dt);
1643 if (!dt_object_exists(dt))
1646 LASSERT(!dt_object_remote(dt));
1647 LINVRNT(osd_invariant(obj));
1649 spin_lock(&obj->oo_guard);
1650 osd_inode_getattr(env, obj->oo_inode, attr);
1651 spin_unlock(&obj->oo_guard);
1655 static int osd_declare_attr_set(const struct lu_env *env,
1656 struct dt_object *dt,
1657 const struct lu_attr *attr,
1658 struct thandle *handle)
1660 struct osd_thandle *oh;
1661 struct osd_object *obj;
1662 struct osd_thread_info *info = osd_oti_get(env);
1663 struct lquota_id_info *qi = &info->oti_qi;
1671 LASSERT(dt != NULL);
1672 LASSERT(handle != NULL);
1674 obj = osd_dt_obj(dt);
1675 LASSERT(osd_invariant(obj));
1677 oh = container_of0(handle, struct osd_thandle, ot_super);
1678 LASSERT(oh->ot_handle == NULL);
1680 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1681 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1683 if (attr == NULL || obj->oo_inode == NULL)
1686 bspace = obj->oo_inode->i_blocks;
1687 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1688 bspace = toqb(bspace);
1690 /* Changing ownership is always preformed by super user, it should not
1693 * We still need to call the osd_declare_qid() to calculate the journal
1694 * credits for updating quota accounting files and to trigger quota
1695 * space adjustment once the operation is completed.*/
1696 if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
1698 uid = i_uid_read(obj->oo_inode);
1699 qi->lqi_type = USRQUOTA;
1700 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
1701 /* inode accounting */
1702 qi->lqi_is_blk = false;
1704 /* one more inode for the new uid ... */
1705 qi->lqi_id.qid_uid = attr->la_uid;
1707 /* Reserve credits for the new uid */
1708 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1709 if (rc == -EDQUOT || rc == -EINPROGRESS)
1714 /* and one less inode for the current uid */
1715 qi->lqi_id.qid_uid = uid;
1717 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1718 if (rc == -EDQUOT || rc == -EINPROGRESS)
1723 /* block accounting */
1724 qi->lqi_is_blk = true;
1726 /* more blocks for the new uid ... */
1727 qi->lqi_id.qid_uid = attr->la_uid;
1728 qi->lqi_space = bspace;
1730 * Credits for the new uid has been reserved, re-use "obj"
1731 * to save credit reservation.
1733 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1734 if (rc == -EDQUOT || rc == -EINPROGRESS)
1739 /* and finally less blocks for the current uid */
1740 qi->lqi_id.qid_uid = uid;
1741 qi->lqi_space = -bspace;
1742 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1743 if (rc == -EDQUOT || rc == -EINPROGRESS)
1749 gid = i_gid_read(obj->oo_inode);
1750 qi->lqi_type = GRPQUOTA;
1751 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
1753 /* inode accounting */
1754 qi->lqi_is_blk = false;
1756 /* one more inode for the new gid ... */
1757 qi->lqi_id.qid_gid = attr->la_gid;
1759 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1760 if (rc == -EDQUOT || rc == -EINPROGRESS)
1765 /* and one less inode for the current gid */
1766 qi->lqi_id.qid_gid = gid;
1768 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1769 if (rc == -EDQUOT || rc == -EINPROGRESS)
1774 /* block accounting */
1775 qi->lqi_is_blk = true;
1777 /* more blocks for the new gid ... */
1778 qi->lqi_id.qid_gid = attr->la_gid;
1779 qi->lqi_space = bspace;
1780 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1781 if (rc == -EDQUOT || rc == -EINPROGRESS)
1786 /* and finally less blocks for the current gid */
1787 qi->lqi_id.qid_gid = gid;
1788 qi->lqi_space = -bspace;
1789 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1790 if (rc == -EDQUOT || rc == -EINPROGRESS)
1799 static int osd_inode_setattr(const struct lu_env *env,
1800 struct inode *inode, const struct lu_attr *attr)
1802 __u64 bits = attr->la_valid;
1804 /* Only allow set size for regular file */
1805 if (!S_ISREG(inode->i_mode))
1806 bits &= ~(LA_SIZE | LA_BLOCKS);
1811 if (bits & LA_ATIME)
1812 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
1813 if (bits & LA_CTIME)
1814 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
1815 if (bits & LA_MTIME)
1816 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
1817 if (bits & LA_SIZE) {
1818 LDISKFS_I(inode)->i_disksize = attr->la_size;
1819 i_size_write(inode, attr->la_size);
1823 /* OSD should not change "i_blocks" which is used by quota.
1824 * "i_blocks" should be changed by ldiskfs only. */
1825 if (bits & LA_BLOCKS)
1826 inode->i_blocks = attr->la_blocks;
1829 inode->i_mode = (inode->i_mode & S_IFMT) |
1830 (attr->la_mode & ~S_IFMT);
1832 i_uid_write(inode, attr->la_uid);
1834 i_gid_write(inode, attr->la_gid);
1835 if (bits & LA_NLINK)
1836 set_nlink(inode, attr->la_nlink);
1838 inode->i_rdev = attr->la_rdev;
1840 if (bits & LA_FLAGS) {
1841 /* always keep S_NOCMTIME */
1842 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1848 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1850 if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1851 (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1855 ll_vfs_dq_init(inode);
1857 if (attr->la_valid & LA_UID)
1858 iattr.ia_valid |= ATTR_UID;
1859 if (attr->la_valid & LA_GID)
1860 iattr.ia_valid |= ATTR_GID;
1861 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1862 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1864 rc = ll_vfs_dq_transfer(inode, &iattr);
1866 CERROR("%s: quota transfer failed: rc = %d. Is quota "
1867 "enforcement enabled on the ldiskfs "
1868 "filesystem?\n", inode->i_sb->s_id, rc);
1875 static int osd_attr_set(const struct lu_env *env,
1876 struct dt_object *dt,
1877 const struct lu_attr *attr,
1878 struct thandle *handle)
1880 struct osd_object *obj = osd_dt_obj(dt);
1881 struct inode *inode;
1884 if (!dt_object_exists(dt))
1887 LASSERT(handle != NULL);
1888 LASSERT(!dt_object_remote(dt));
1889 LASSERT(osd_invariant(obj));
1891 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1893 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1894 struct osd_thread_info *oti = osd_oti_get(env);
1895 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
1896 struct lu_fid *fid1 = &oti->oti_fid;
1897 struct osd_inode_id *id = &oti->oti_id;
1898 struct iam_path_descr *ipd;
1899 struct iam_container *bag;
1900 struct osd_thandle *oh;
1903 fid_cpu_to_be(fid1, fid0);
1904 memset(id, 1, sizeof(*id));
1905 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1906 fid0)->oi_dir.od_container;
1907 ipd = osd_idx_ipd_get(env, bag);
1908 if (unlikely(ipd == NULL))
1911 oh = container_of0(handle, struct osd_thandle, ot_super);
1912 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1913 (const struct iam_rec *)id, ipd);
1914 osd_ipd_put(env, bag, ipd);
1915 return(rc > 0 ? 0 : rc);
1918 inode = obj->oo_inode;
1920 rc = osd_quota_transfer(inode, attr);
1924 spin_lock(&obj->oo_guard);
1925 rc = osd_inode_setattr(env, inode, attr);
1926 spin_unlock(&obj->oo_guard);
1929 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
1933 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
1934 struct osd_object *obj,
1935 const char *name, const int namelen)
1937 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1940 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1941 umode_t mode, struct dt_allocation_hint *hint,
1945 struct osd_device *osd = osd_obj2dev(obj);
1946 struct osd_thandle *oth;
1947 struct dt_object *parent = NULL;
1948 struct inode *inode;
1950 LINVRNT(osd_invariant(obj));
1951 LASSERT(obj->oo_inode == NULL);
1952 LASSERT(obj->oo_hl_head == NULL);
1954 if (S_ISDIR(mode) && ldiskfs_pdo) {
1955 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1956 if (obj->oo_hl_head == NULL)
1960 oth = container_of(th, struct osd_thandle, ot_super);
1961 LASSERT(oth->ot_handle->h_transaction != NULL);
1963 if (hint && hint->dah_parent)
1964 parent = hint->dah_parent;
1966 inode = ldiskfs_create_inode(oth->ot_handle,
1967 parent ? osd_dt_obj(parent)->oo_inode :
1968 osd_sb(osd)->s_root->d_inode,
1970 if (!IS_ERR(inode)) {
1971 /* Do not update file c/mtime in ldiskfs. */
1972 inode->i_flags |= S_NOCMTIME;
1974 /* For new created object, it must be consistent,
1975 * and it is unnecessary to scrub against it. */
1976 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
1978 obj->oo_inode = inode;
1981 if (obj->oo_hl_head != NULL) {
1982 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1983 obj->oo_hl_head = NULL;
1985 result = PTR_ERR(inode);
1987 LINVRNT(osd_invariant(obj));
1995 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1996 struct lu_attr *attr,
1997 struct dt_allocation_hint *hint,
1998 struct dt_object_format *dof,
2002 struct osd_thandle *oth;
2003 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2005 LASSERT(S_ISDIR(attr->la_mode));
2007 oth = container_of(th, struct osd_thandle, ot_super);
2008 LASSERT(oth->ot_handle->h_transaction != NULL);
2009 result = osd_mkfile(info, obj, mode, hint, th);
2014 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2015 struct lu_attr *attr,
2016 struct dt_allocation_hint *hint,
2017 struct dt_object_format *dof,
2021 struct osd_thandle *oth;
2022 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2024 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2026 LASSERT(S_ISREG(attr->la_mode));
2028 oth = container_of(th, struct osd_thandle, ot_super);
2029 LASSERT(oth->ot_handle->h_transaction != NULL);
2031 result = osd_mkfile(info, obj, mode, hint, th);
2033 LASSERT(obj->oo_inode != NULL);
2034 if (feat->dif_flags & DT_IND_VARKEY)
2035 result = iam_lvar_create(obj->oo_inode,
2036 feat->dif_keysize_max,
2038 feat->dif_recsize_max,
2041 result = iam_lfix_create(obj->oo_inode,
2042 feat->dif_keysize_max,
2044 feat->dif_recsize_max,
2051 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2052 struct lu_attr *attr,
2053 struct dt_allocation_hint *hint,
2054 struct dt_object_format *dof,
2057 LASSERT(S_ISREG(attr->la_mode));
2058 return osd_mkfile(info, obj, (attr->la_mode &
2059 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2062 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2063 struct lu_attr *attr,
2064 struct dt_allocation_hint *hint,
2065 struct dt_object_format *dof,
2068 LASSERT(S_ISLNK(attr->la_mode));
2069 return osd_mkfile(info, obj, (attr->la_mode &
2070 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2073 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2074 struct lu_attr *attr,
2075 struct dt_allocation_hint *hint,
2076 struct dt_object_format *dof,
2079 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2082 LINVRNT(osd_invariant(obj));
2083 LASSERT(obj->oo_inode == NULL);
2084 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2085 S_ISFIFO(mode) || S_ISSOCK(mode));
2087 result = osd_mkfile(info, obj, mode, hint, th);
2089 LASSERT(obj->oo_inode != NULL);
2091 * This inode should be marked dirty for i_rdev. Currently
2092 * that is done in the osd_attr_init().
2094 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2097 LINVRNT(osd_invariant(obj));
2101 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2103 struct dt_allocation_hint *hint,
2104 struct dt_object_format *dof,
2107 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2109 osd_obj_type_f result;
2125 result = osd_mk_index;
2136 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2137 struct dt_object *parent, struct dt_object *child,
2142 ah->dah_parent = parent;
2143 ah->dah_mode = child_mode;
2146 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2147 struct lu_attr *attr, struct dt_object_format *dof)
2149 struct inode *inode = obj->oo_inode;
2150 __u64 valid = attr->la_valid;
2153 attr->la_valid &= ~(LA_TYPE | LA_MODE);
2155 if (dof->dof_type != DFT_NODE)
2156 attr->la_valid &= ~LA_RDEV;
2157 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2158 attr->la_valid &= ~LA_ATIME;
2159 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2160 attr->la_valid &= ~LA_CTIME;
2161 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2162 attr->la_valid &= ~LA_MTIME;
2164 result = osd_quota_transfer(inode, attr);
2168 if (attr->la_valid != 0) {
2169 result = osd_inode_setattr(info->oti_env, inode, attr);
2171 * The osd_inode_setattr() should always succeed here. The
2172 * only error that could be returned is EDQUOT when we are
2173 * trying to change the UID or GID of the inode. However, this
2174 * should not happen since quota enforcement is no longer
2175 * enabled on ldiskfs (lquota takes care of it).
2177 LASSERTF(result == 0, "%d\n", result);
2178 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2181 attr->la_valid = valid;
2185 * Helper function for osd_object_create()
2187 * \retval 0, on success
2189 static int __osd_object_create(struct osd_thread_info *info,
2190 struct osd_object *obj, struct lu_attr *attr,
2191 struct dt_allocation_hint *hint,
2192 struct dt_object_format *dof,
2198 /* we drop umask so that permissions we pass are not affected */
2199 umask = current->fs->umask;
2200 current->fs->umask = 0;
2202 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2204 if (likely(obj->oo_inode != NULL)) {
2205 LASSERT(obj->oo_inode->i_state & I_NEW);
2207 /* Unlock the inode before attr initialization to avoid
2208 * unnecessary dqget operations. LU-6378 */
2209 unlock_new_inode(obj->oo_inode);
2212 if (likely(result == 0)) {
2213 osd_attr_init(info, obj, attr, dof);
2214 osd_object_init0(obj);
2217 /* restore previous umask value */
2218 current->fs->umask = umask;
2224 * Helper function for osd_object_create()
2226 * \retval 0, on success
2228 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2229 const struct lu_fid *fid, struct thandle *th)
2231 struct osd_thread_info *info = osd_oti_get(env);
2232 struct osd_inode_id *id = &info->oti_id;
2233 struct osd_device *osd = osd_obj2dev(obj);
2234 struct osd_thandle *oh;
2236 LASSERT(obj->oo_inode != NULL);
2238 oh = container_of0(th, struct osd_thandle, ot_super);
2239 LASSERT(oh->ot_handle);
2241 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2242 return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2245 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2246 u64 seq, struct lu_seq_range *range)
2248 struct seq_server_site *ss = osd_seq_site(osd);
2250 if (fid_seq_is_idif(seq)) {
2251 fld_range_set_ost(range);
2252 range->lsr_index = idif_ost_idx(seq);
2256 if (!fid_seq_in_fldb(seq)) {
2257 fld_range_set_mdt(range);
2259 /* FIXME: If ss is NULL, it suppose not get lsr_index
2261 range->lsr_index = ss->ss_node_id;
2265 LASSERT(ss != NULL);
2266 fld_range_set_any(range);
2267 /* OSD will only do local fld lookup */
2268 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2272 * Concurrency: no external locking is necessary.
2274 static int osd_declare_object_create(const struct lu_env *env,
2275 struct dt_object *dt,
2276 struct lu_attr *attr,
2277 struct dt_allocation_hint *hint,
2278 struct dt_object_format *dof,
2279 struct thandle *handle)
2281 struct osd_thandle *oh;
2285 LASSERT(handle != NULL);
2287 oh = container_of0(handle, struct osd_thandle, ot_super);
2288 LASSERT(oh->ot_handle == NULL);
2290 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2291 osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2292 /* Reuse idle OI block may cause additional one OI block
2294 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2295 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2297 /* If this is directory, then we expect . and .. to be inserted as
2298 * well. The one directory block always needs to be created for the
2299 * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2300 * block), there is no danger of needing a tree for the first block.
2302 if (attr && S_ISDIR(attr->la_mode)) {
2303 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2304 osd_dto_credits_noquota[DTO_WRITE_BASE]);
2305 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2311 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2312 osd_dt_obj(dt), false, NULL, false);
2319 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2320 struct lu_attr *attr,
2321 struct dt_allocation_hint *hint,
2322 struct dt_object_format *dof, struct thandle *th)
2324 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2325 struct osd_object *obj = osd_dt_obj(dt);
2326 struct osd_thread_info *info = osd_oti_get(env);
2330 if (dt_object_exists(dt))
2333 LINVRNT(osd_invariant(obj));
2334 LASSERT(!dt_object_remote(dt));
2335 LASSERT(osd_write_locked(env, obj));
2336 LASSERT(th != NULL);
2338 if (unlikely(fid_is_acct(fid)))
2339 /* Quota files can't be created from the kernel any more,
2340 * 'tune2fs -O quota' will take care of creating them */
2343 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2344 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2346 result = __osd_object_create(info, obj, attr, hint, dof, th);
2348 result = __osd_oi_insert(env, obj, fid, th);
2350 LASSERT(ergo(result == 0,
2351 dt_object_exists(dt) && !dt_object_remote(dt)));
2353 LASSERT(osd_invariant(obj));
2358 * Called to destroy on-disk representation of the object
2360 * Concurrency: must be locked
2362 static int osd_declare_object_destroy(const struct lu_env *env,
2363 struct dt_object *dt,
2366 struct osd_object *obj = osd_dt_obj(dt);
2367 struct inode *inode = obj->oo_inode;
2368 struct osd_thandle *oh;
2372 oh = container_of0(th, struct osd_thandle, ot_super);
2373 LASSERT(oh->ot_handle == NULL);
2376 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2377 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2378 /* Recycle idle OI leaf may cause additional three OI blocks
2380 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2381 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2382 /* one less inode */
2383 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2384 -1, oh, obj, false, NULL, false);
2387 /* data to be truncated */
2388 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2389 0, oh, obj, true, NULL, false);
2393 static int osd_object_destroy(const struct lu_env *env,
2394 struct dt_object *dt,
2397 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2398 struct osd_object *obj = osd_dt_obj(dt);
2399 struct inode *inode = obj->oo_inode;
2400 struct osd_device *osd = osd_obj2dev(obj);
2401 struct osd_thandle *oh;
2405 oh = container_of0(th, struct osd_thandle, ot_super);
2406 LASSERT(oh->ot_handle);
2408 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2410 if (unlikely(fid_is_acct(fid)))
2413 if (S_ISDIR(inode->i_mode)) {
2414 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2415 inode->i_nlink == 2);
2416 /* it will check/delete the inode from remote parent,
2417 * how to optimize it? unlink performance impaction XXX */
2418 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2419 if (result != 0 && result != -ENOENT) {
2420 CERROR("%s: delete inode "DFID": rc = %d\n",
2421 osd_name(osd), PFID(fid), result);
2423 spin_lock(&obj->oo_guard);
2425 spin_unlock(&obj->oo_guard);
2426 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2429 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2431 result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2434 /* XXX: add to ext3 orphan list */
2435 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2437 /* not needed in the cache anymore */
2438 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2444 * Put the fid into lustre_mdt_attrs, and then place the structure
2445 * inode's ea. This fid should not be altered during the life time
2448 * \retval +ve, on success
2449 * \retval -ve, on error
2451 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2453 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2454 const struct lu_fid *fid, __u32 compat, __u32 incompat)
2456 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2460 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2463 lustre_lma_init(lma, fid, compat, incompat);
2464 lustre_lma_swab(lma);
2466 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2468 /* LMA may already exist, but we need to check that all the
2469 * desired compat/incompat flags have been added. */
2470 if (unlikely(rc == -EEXIST)) {
2471 if (compat == 0 && incompat == 0)
2474 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2475 XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2480 lustre_lma_swab(lma);
2481 if (!(~lma->lma_compat & compat) &&
2482 !(~lma->lma_incompat & incompat))
2485 lma->lma_compat |= compat;
2486 lma->lma_incompat |= incompat;
2487 lustre_lma_swab(lma);
2488 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2489 sizeof(*lma), XATTR_REPLACE);
2496 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2497 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2498 * To have compatilibility with 1.8 ldiskfs driver we need to have
2499 * magic number at start of fid data.
2500 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2503 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2504 const struct lu_fid *fid)
2506 if (!fid_is_namespace_visible(fid) ||
2507 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2508 param->edp_magic = 0;
2512 param->edp_magic = LDISKFS_LUFID_MAGIC;
2513 param->edp_len = sizeof(struct lu_fid) + 1;
2514 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2518 * Try to read the fid from inode ea into dt_rec.
2520 * \param fid object fid.
2522 * \retval 0 on success
2524 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2525 __u32 ino, struct lu_fid *fid,
2526 struct osd_inode_id *id)
2528 struct osd_thread_info *info = osd_oti_get(env);
2529 struct inode *inode;
2532 osd_id_gen(id, ino, OSD_OII_NOGEN);
2533 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2535 RETURN(PTR_ERR(inode));
2541 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2543 struct inode *parent_dir,
2544 const struct lu_fid *dot_fid,
2545 const struct lu_fid *dot_dot_fid,
2546 struct osd_thandle *oth)
2548 struct ldiskfs_dentry_param *dot_ldp;
2549 struct ldiskfs_dentry_param *dot_dot_ldp;
2551 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2552 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2554 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2555 dot_ldp->edp_magic = 0;
2556 return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2557 dir, dot_ldp, dot_dot_ldp);
2561 * Create an local agent inode for remote entry
2563 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2564 struct osd_device *osd,
2565 struct osd_object *pobj,
2566 const struct lu_fid *fid,
2570 struct osd_thread_info *info = osd_oti_get(env);
2571 struct inode *local;
2572 struct osd_thandle *oh;
2577 oh = container_of(th, struct osd_thandle, ot_super);
2578 LASSERT(oh->ot_handle->h_transaction != NULL);
2580 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2581 if (IS_ERR(local)) {
2582 CERROR("%s: create local error %d\n", osd_name(osd),
2583 (int)PTR_ERR(local));
2587 ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
2588 unlock_new_inode(local);
2590 /* Set special LMA flag for local agent inode */
2591 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2593 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2594 osd_name(osd), PFID(fid), rc);
2595 RETURN(ERR_PTR(rc));
2601 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2602 lu_object_fid(&pobj->oo_dt.do_lu),
2605 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2606 osd_name(osd), PFID(fid), rc);
2607 RETURN(ERR_PTR(rc));
2614 * Delete local agent inode for remote entry
2616 static int osd_delete_local_agent_inode(const struct lu_env *env,
2617 struct osd_device *osd,
2618 const struct lu_fid *fid,
2619 __u32 ino, struct osd_thandle *oh)
2621 struct osd_thread_info *oti = osd_oti_get(env);
2622 struct osd_inode_id *id = &oti->oti_id;
2623 struct inode *inode;
2626 id->oii_ino = le32_to_cpu(ino);
2627 id->oii_gen = OSD_OII_NOGEN;
2628 inode = osd_iget(oti, osd, id);
2629 if (IS_ERR(inode)) {
2630 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2631 PFID(fid), id->oii_ino, id->oii_gen);
2632 RETURN(PTR_ERR(inode));
2636 mark_inode_dirty(inode);
2637 CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2638 osd_name(osd), PFID(fid), inode->i_ino);
2644 * OSD layer object create function for interoperability mode (b11826).
2645 * This is mostly similar to osd_object_create(). Only difference being, fid is
2646 * inserted into inode ea here.
2648 * \retval 0, on success
2649 * \retval -ve, on error
2651 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2652 struct lu_attr *attr,
2653 struct dt_allocation_hint *hint,
2654 struct dt_object_format *dof,
2657 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2658 struct osd_object *obj = osd_dt_obj(dt);
2659 struct osd_thread_info *info = osd_oti_get(env);
2664 if (dt_object_exists(dt))
2667 LASSERT(osd_invariant(obj));
2668 LASSERT(!dt_object_remote(dt));
2669 LASSERT(osd_write_locked(env, obj));
2670 LASSERT(th != NULL);
2672 if (unlikely(fid_is_acct(fid)))
2673 /* Quota files can't be created from the kernel any more,
2674 * 'tune2fs -O quota' will take care of creating them */
2677 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2678 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2680 result = __osd_object_create(info, obj, attr, hint, dof, th);
2682 if (fid_is_idif(fid) &&
2683 !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
2684 struct lu_fid *tfid = &info->oti_fid;
2685 struct ost_id *oi = &info->oti_ostid;
2687 fid_to_ostid(fid, oi);
2688 ostid_to_fid(tfid, oi, 0);
2689 result = osd_ea_fid_set(info, obj->oo_inode, tfid,
2690 LMAC_FID_ON_OST, 0);
2692 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2693 fid_is_on_ost(info, osd_obj2dev(obj),
2694 fid, OI_CHECK_FLD) ?
2695 LMAC_FID_ON_OST : 0, 0);
2700 result = __osd_oi_insert(env, obj, fid, th);
2702 LASSERT(ergo(result == 0,
2703 dt_object_exists(dt) && !dt_object_remote(dt)));
2704 LINVRNT(osd_invariant(obj));
2708 static int osd_declare_object_ref_add(const struct lu_env *env,
2709 struct dt_object *dt,
2710 struct thandle *handle)
2712 struct osd_thandle *oh;
2714 /* it's possible that object doesn't exist yet */
2715 LASSERT(handle != NULL);
2717 oh = container_of0(handle, struct osd_thandle, ot_super);
2718 LASSERT(oh->ot_handle == NULL);
2720 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2721 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2727 * Concurrency: @dt is write locked.
2729 static int osd_object_ref_add(const struct lu_env *env,
2730 struct dt_object *dt, struct thandle *th)
2732 struct osd_object *obj = osd_dt_obj(dt);
2733 struct inode *inode = obj->oo_inode;
2734 struct osd_thandle *oh;
2737 if (!dt_object_exists(dt))
2740 LINVRNT(osd_invariant(obj));
2741 LASSERT(!dt_object_remote(dt));
2742 LASSERT(osd_write_locked(env, obj));
2743 LASSERT(th != NULL);
2745 oh = container_of0(th, struct osd_thandle, ot_super);
2746 LASSERT(oh->ot_handle != NULL);
2748 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2750 CDEBUG(D_INODE, DFID" increase nlink %d\n",
2751 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2753 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2754 * (65000) subdirectories by storing "1" in i_nlink if the link count
2755 * would otherwise overflow. Directory tranversal tools understand
2756 * that (st_nlink == 1) indicates that the filesystem dose not track
2757 * hard links count on the directory, and will not abort subdirectory
2758 * scanning early once (st_nlink - 2) subdirs have been found.
2760 * This also has to properly handle the case of inodes with nlink == 0
2761 * in case they are being linked into the PENDING directory
2763 spin_lock(&obj->oo_guard);
2764 if (unlikely(inode->i_nlink == 0))
2765 /* inc_nlink from 0 may cause WARN_ON */
2766 set_nlink(inode, 1);
2768 ldiskfs_inc_count(oh->ot_handle, inode);
2769 if (!S_ISDIR(inode->i_mode))
2770 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2772 spin_unlock(&obj->oo_guard);
2774 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2775 LINVRNT(osd_invariant(obj));
2780 static int osd_declare_object_ref_del(const struct lu_env *env,
2781 struct dt_object *dt,
2782 struct thandle *handle)
2784 struct osd_thandle *oh;
2786 LASSERT(!dt_object_remote(dt));
2787 LASSERT(handle != NULL);
2789 oh = container_of0(handle, struct osd_thandle, ot_super);
2790 LASSERT(oh->ot_handle == NULL);
2792 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2793 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2799 * Concurrency: @dt is write locked.
2801 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2804 struct osd_object *obj = osd_dt_obj(dt);
2805 struct inode *inode = obj->oo_inode;
2806 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
2807 struct osd_thandle *oh;
2809 if (!dt_object_exists(dt))
2812 LINVRNT(osd_invariant(obj));
2813 LASSERT(!dt_object_remote(dt));
2814 LASSERT(osd_write_locked(env, obj));
2815 LASSERT(th != NULL);
2817 oh = container_of0(th, struct osd_thandle, ot_super);
2818 LASSERT(oh->ot_handle != NULL);
2820 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2822 spin_lock(&obj->oo_guard);
2823 /* That can be result of upgrade from old Lustre version and
2824 * applied only to local files. Just skip this ref_del call.
2825 * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2826 if (inode->i_nlink == 0) {
2827 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2828 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2829 ", maybe an upgraded file? (LU-3915)\n",
2830 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2831 spin_unlock(&obj->oo_guard);
2835 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2836 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2838 ldiskfs_dec_count(oh->ot_handle, inode);
2839 spin_unlock(&obj->oo_guard);
2841 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2842 LINVRNT(osd_invariant(obj));
2848 * Get the 64-bit version for an inode.
2850 static int osd_object_version_get(const struct lu_env *env,
2851 struct dt_object *dt, dt_obj_version_t *ver)
2853 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2855 CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2856 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2857 *ver = LDISKFS_I(inode)->i_fs_version;
2862 * Concurrency: @dt is read locked.
2864 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2865 struct lu_buf *buf, const char *name)
2867 struct osd_object *obj = osd_dt_obj(dt);
2868 struct inode *inode = obj->oo_inode;
2869 struct osd_thread_info *info = osd_oti_get(env);
2870 struct dentry *dentry = &info->oti_obj_dentry;
2872 /* version get is not real XATTR but uses xattr API */
2873 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2874 /* for version we are just using xattr API but change inode
2876 if (buf->lb_len == 0)
2877 return sizeof(dt_obj_version_t);
2879 if (buf->lb_len < sizeof(dt_obj_version_t))
2882 osd_object_version_get(env, dt, buf->lb_buf);
2884 return sizeof(dt_obj_version_t);
2887 if (!dt_object_exists(dt))
2890 LASSERT(!dt_object_remote(dt));
2891 LASSERT(inode->i_op != NULL);
2892 LASSERT(inode->i_op->getxattr != NULL);
2894 return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2898 static int osd_declare_xattr_set(const struct lu_env *env,
2899 struct dt_object *dt,
2900 const struct lu_buf *buf, const char *name,
2901 int fl, struct thandle *handle)
2903 struct osd_thandle *oh;
2905 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
2907 LASSERT(handle != NULL);
2909 oh = container_of0(handle, struct osd_thandle, ot_super);
2910 LASSERT(oh->ot_handle == NULL);
2912 /* optimistic optimization: LMA is set first and usually fit inode */
2913 if (strcmp(name, XATTR_NAME_LMA) == 0) {
2914 if (dt_object_exists(dt))
2918 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2921 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2922 if (buf && buf->lb_len > sb->s_blocksize) {
2923 credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2924 sb->s_blocksize_bits;
2927 * xattr set may involve inode quota change, reserve credits for
2928 * dquot_initialize()
2930 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
2933 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2939 * Set the 64-bit version for object
2941 static void osd_object_version_set(const struct lu_env *env,
2942 struct dt_object *dt,
2943 dt_obj_version_t *new_version)
2945 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2947 CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2948 *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2950 LDISKFS_I(inode)->i_fs_version = *new_version;
2951 /** Version is set after all inode operations are finished,
2952 * so we should mark it dirty here */
2953 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2957 * Concurrency: @dt is write locked.
2959 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2960 const struct lu_buf *buf, const char *name, int fl,
2961 struct thandle *handle)
2963 struct osd_object *obj = osd_dt_obj(dt);
2964 struct inode *inode = obj->oo_inode;
2965 struct osd_thread_info *info = osd_oti_get(env);
2969 LASSERT(handle != NULL);
2971 /* version set is not real XATTR */
2972 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2973 /* for version we are just using xattr API but change inode
2975 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2976 osd_object_version_set(env, dt, buf->lb_buf);
2977 return sizeof(dt_obj_version_t);
2980 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
2981 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
2983 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2984 if (fl & LU_XATTR_REPLACE)
2985 fs_flags |= XATTR_REPLACE;
2987 if (fl & LU_XATTR_CREATE)
2988 fs_flags |= XATTR_CREATE;
2990 if (strcmp(name, XATTR_NAME_LMV) == 0) {
2991 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2994 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
2998 lma->lma_incompat |= LMAI_STRIPED;
2999 lustre_lma_swab(lma);
3000 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3001 sizeof(*lma), XATTR_REPLACE);
3006 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
3007 strcmp(name, XATTR_NAME_LINK) == 0)
3010 return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3015 * Concurrency: @dt is read locked.
3017 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3018 const struct lu_buf *buf)
3020 struct osd_object *obj = osd_dt_obj(dt);
3021 struct inode *inode = obj->oo_inode;
3022 struct osd_thread_info *info = osd_oti_get(env);
3023 struct dentry *dentry = &info->oti_obj_dentry;
3025 if (!dt_object_exists(dt))
3028 LASSERT(!dt_object_remote(dt));
3029 LASSERT(inode->i_op != NULL);
3030 LASSERT(inode->i_op->listxattr != NULL);
3032 dentry->d_inode = inode;
3033 dentry->d_sb = inode->i_sb;
3034 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3037 static int osd_declare_xattr_del(const struct lu_env *env,
3038 struct dt_object *dt, const char *name,
3039 struct thandle *handle)
3041 struct osd_thandle *oh;
3042 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3044 LASSERT(!dt_object_remote(dt));
3045 LASSERT(handle != NULL);
3047 oh = container_of0(handle, struct osd_thandle, ot_super);
3048 LASSERT(oh->ot_handle == NULL);
3050 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3051 osd_dto_credits_noquota[DTO_XATTR_SET]);
3053 * xattr del may involve inode quota change, reserve credits for
3054 * dquot_initialize()
3056 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3062 * Concurrency: @dt is write locked.
3064 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3065 const char *name, struct thandle *handle)
3067 struct osd_object *obj = osd_dt_obj(dt);
3068 struct inode *inode = obj->oo_inode;
3069 struct osd_thread_info *info = osd_oti_get(env);
3070 struct dentry *dentry = &info->oti_obj_dentry;
3073 if (!dt_object_exists(dt))
3076 LASSERT(!dt_object_remote(dt));
3077 LASSERT(inode->i_op != NULL);
3078 LASSERT(inode->i_op->removexattr != NULL);
3079 LASSERT(handle != NULL);
3081 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3083 ll_vfs_dq_init(inode);
3084 dentry->d_inode = inode;
3085 dentry->d_sb = inode->i_sb;
3086 rc = inode->i_op->removexattr(dentry, name);
3090 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3091 __u64 start, __u64 end)
3093 struct osd_object *obj = osd_dt_obj(dt);
3094 struct inode *inode = obj->oo_inode;
3095 struct osd_thread_info *info = osd_oti_get(env);
3096 struct dentry *dentry = &info->oti_obj_dentry;
3097 struct file *file = &info->oti_file;
3102 dentry->d_inode = inode;
3103 dentry->d_sb = inode->i_sb;
3104 file->f_dentry = dentry;
3105 file->f_mapping = inode->i_mapping;
3106 file->f_op = inode->i_fop;
3107 set_file_inode(file, inode);
3109 rc = ll_vfs_fsync_range(file, start, end, 0);
3118 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3119 const struct dt_index_features *feat)
3121 struct iam_descr *descr;
3123 if (osd_object_is_root(o))
3124 return feat == &dt_directory_features;
3126 LASSERT(o->oo_dir != NULL);
3128 descr = o->oo_dir->od_container.ic_descr;
3129 if (feat == &dt_directory_features) {
3130 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3136 feat->dif_keysize_min <= descr->id_key_size &&
3137 descr->id_key_size <= feat->dif_keysize_max &&
3138 feat->dif_recsize_min <= descr->id_rec_size &&
3139 descr->id_rec_size <= feat->dif_recsize_max &&
3140 !(feat->dif_flags & (DT_IND_VARKEY |
3141 DT_IND_VARREC | DT_IND_NONUNQ)) &&
3142 ergo(feat->dif_flags & DT_IND_UPDATE,
3143 1 /* XXX check that object (and file system) is
3148 static int osd_iam_container_init(const struct lu_env *env,
3149 struct osd_object *obj,
3150 struct osd_directory *dir)
3152 struct iam_container *bag = &dir->od_container;
3155 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3159 result = iam_container_setup(bag);
3161 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3163 iam_container_fini(bag);
3170 * Concurrency: no external locking is necessary.
3172 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3173 const struct dt_index_features *feat)
3177 struct osd_object *obj = osd_dt_obj(dt);
3179 LINVRNT(osd_invariant(obj));
3181 if (osd_object_is_root(obj)) {
3182 dt->do_index_ops = &osd_index_ea_ops;
3184 } else if (feat == &dt_directory_features) {
3185 dt->do_index_ops = &osd_index_ea_ops;
3186 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3191 } else if (unlikely(feat == &dt_otable_features)) {
3192 dt->do_index_ops = &osd_otable_ops;
3194 } else if (unlikely(feat == &dt_acct_features)) {
3195 dt->do_index_ops = &osd_acct_index_ops;
3198 } else if (!osd_has_index(obj)) {
3199 struct osd_directory *dir;
3204 spin_lock(&obj->oo_guard);
3205 if (obj->oo_dir == NULL)
3209 * Concurrent thread allocated container data.
3212 spin_unlock(&obj->oo_guard);
3214 * Now, that we have container data, serialize its
3217 down_write(&obj->oo_ext_idx_sem);
3219 * recheck under lock.
3221 if (!osd_has_index(obj))
3222 result = osd_iam_container_init(env, obj,
3226 up_write(&obj->oo_ext_idx_sem);
3234 if (result == 0 && skip_iam == 0) {
3235 if (!osd_iam_index_probe(env, obj, feat))
3238 LINVRNT(osd_invariant(obj));
3240 if (result == 0 && is_quota_glb_feat(feat) &&
3241 fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3242 result = osd_quota_migration(env, dt, feat);
3247 static int osd_otable_it_attr_get(const struct lu_env *env,
3248 struct dt_object *dt,
3249 struct lu_attr *attr)
3255 static const struct dt_object_operations osd_obj_ops = {
3256 .do_read_lock = osd_object_read_lock,
3257 .do_write_lock = osd_object_write_lock,
3258 .do_read_unlock = osd_object_read_unlock,
3259 .do_write_unlock = osd_object_write_unlock,
3260 .do_write_locked = osd_object_write_locked,
3261 .do_attr_get = osd_attr_get,
3262 .do_declare_attr_set = osd_declare_attr_set,
3263 .do_attr_set = osd_attr_set,
3264 .do_ah_init = osd_ah_init,
3265 .do_declare_create = osd_declare_object_create,
3266 .do_create = osd_object_create,
3267 .do_declare_destroy = osd_declare_object_destroy,
3268 .do_destroy = osd_object_destroy,
3269 .do_index_try = osd_index_try,
3270 .do_declare_ref_add = osd_declare_object_ref_add,
3271 .do_ref_add = osd_object_ref_add,
3272 .do_declare_ref_del = osd_declare_object_ref_del,
3273 .do_ref_del = osd_object_ref_del,
3274 .do_xattr_get = osd_xattr_get,
3275 .do_declare_xattr_set = osd_declare_xattr_set,
3276 .do_xattr_set = osd_xattr_set,
3277 .do_declare_xattr_del = osd_declare_xattr_del,
3278 .do_xattr_del = osd_xattr_del,
3279 .do_xattr_list = osd_xattr_list,
3280 .do_object_sync = osd_object_sync,
3284 * dt_object_operations for interoperability mode
3285 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3287 static const struct dt_object_operations osd_obj_ea_ops = {
3288 .do_read_lock = osd_object_read_lock,
3289 .do_write_lock = osd_object_write_lock,
3290 .do_read_unlock = osd_object_read_unlock,
3291 .do_write_unlock = osd_object_write_unlock,
3292 .do_write_locked = osd_object_write_locked,
3293 .do_attr_get = osd_attr_get,
3294 .do_declare_attr_set = osd_declare_attr_set,
3295 .do_attr_set = osd_attr_set,
3296 .do_ah_init = osd_ah_init,
3297 .do_declare_create = osd_declare_object_create,
3298 .do_create = osd_object_ea_create,
3299 .do_declare_destroy = osd_declare_object_destroy,
3300 .do_destroy = osd_object_destroy,
3301 .do_index_try = osd_index_try,
3302 .do_declare_ref_add = osd_declare_object_ref_add,
3303 .do_ref_add = osd_object_ref_add,
3304 .do_declare_ref_del = osd_declare_object_ref_del,
3305 .do_ref_del = osd_object_ref_del,
3306 .do_xattr_get = osd_xattr_get,
3307 .do_declare_xattr_set = osd_declare_xattr_set,
3308 .do_xattr_set = osd_xattr_set,
3309 .do_declare_xattr_del = osd_declare_xattr_del,
3310 .do_xattr_del = osd_xattr_del,
3311 .do_xattr_list = osd_xattr_list,
3312 .do_object_sync = osd_object_sync,
3315 static const struct dt_object_operations osd_obj_otable_it_ops = {
3316 .do_attr_get = osd_otable_it_attr_get,
3317 .do_index_try = osd_index_try,
3320 static int osd_index_declare_iam_delete(const struct lu_env *env,
3321 struct dt_object *dt,
3322 const struct dt_key *key,
3323 struct thandle *handle)
3325 struct osd_thandle *oh;
3327 oh = container_of0(handle, struct osd_thandle, ot_super);
3328 LASSERT(oh->ot_handle == NULL);
3330 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3331 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3337 * delete a (key, value) pair from index \a dt specified by \a key
3339 * \param dt osd index object
3340 * \param key key for index
3341 * \param rec record reference
3342 * \param handle transaction handler
3345 * \retval -ve failure
3347 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3348 const struct dt_key *key,
3349 struct thandle *handle)
3351 struct osd_thread_info *oti = osd_oti_get(env);
3352 struct osd_object *obj = osd_dt_obj(dt);
3353 struct osd_thandle *oh;
3354 struct iam_path_descr *ipd;
3355 struct iam_container *bag = &obj->oo_dir->od_container;
3359 if (!dt_object_exists(dt))
3362 LINVRNT(osd_invariant(obj));
3363 LASSERT(!dt_object_remote(dt));
3364 LASSERT(bag->ic_object == obj->oo_inode);
3365 LASSERT(handle != NULL);
3367 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3369 ipd = osd_idx_ipd_get(env, bag);
3370 if (unlikely(ipd == NULL))
3373 oh = container_of0(handle, struct osd_thandle, ot_super);
3374 LASSERT(oh->ot_handle != NULL);
3375 LASSERT(oh->ot_handle->h_transaction != NULL);
3377 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3378 /* swab quota uid/gid provided by caller */
3379 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3380 key = (const struct dt_key *)&oti->oti_quota_id;
3383 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3384 osd_ipd_put(env, bag, ipd);
3385 LINVRNT(osd_invariant(obj));
3389 static int osd_index_declare_ea_delete(const struct lu_env *env,
3390 struct dt_object *dt,
3391 const struct dt_key *key,
3392 struct thandle *handle)
3394 struct osd_thandle *oh;
3395 struct inode *inode;
3399 LASSERT(!dt_object_remote(dt));
3400 LASSERT(handle != NULL);
3402 oh = container_of0(handle, struct osd_thandle, ot_super);
3403 LASSERT(oh->ot_handle == NULL);
3405 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3406 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3408 inode = osd_dt_obj(dt)->oo_inode;
3411 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3412 0, oh, osd_dt_obj(dt), true, NULL, false);
3416 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3419 struct osd_fid_pack *rec;
3422 if (de->file_type & LDISKFS_DIRENT_LUFID) {
3423 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3424 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3429 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3430 const struct lu_fid *fid)
3432 struct seq_server_site *ss = osd_seq_site(osd);
3435 /* FID seqs not in FLDB, must be local seq */
3436 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3439 /* If FLD is not being initialized yet, it only happens during the
3440 * initialization, likely during mgs initialization, and we assume
3441 * this is local FID. */
3442 if (ss == NULL || ss->ss_server_fld == NULL)
3445 /* Only check the local FLDB here */
3446 if (osd_seq_exists(env, osd, fid_seq(fid)))
3453 * Index delete function for interoperability mode (b11826).
3454 * It will remove the directory entry added by osd_index_ea_insert().
3455 * This entry is needed to maintain name->fid mapping.
3457 * \param key, key i.e. file entry to be deleted
3459 * \retval 0, on success
3460 * \retval -ve, on error
3462 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3463 const struct dt_key *key, struct thandle *handle)
3465 struct osd_object *obj = osd_dt_obj(dt);
3466 struct inode *dir = obj->oo_inode;
3467 struct dentry *dentry;
3468 struct osd_thandle *oh;
3469 struct ldiskfs_dir_entry_2 *de = NULL;
3470 struct buffer_head *bh;
3471 struct htree_lock *hlock = NULL;
3472 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
3473 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3477 if (!dt_object_exists(dt))
3480 LINVRNT(osd_invariant(obj));
3481 LASSERT(!dt_object_remote(dt));
3482 LASSERT(handle != NULL);
3484 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3486 oh = container_of(handle, struct osd_thandle, ot_super);
3487 LASSERT(oh->ot_handle != NULL);
3488 LASSERT(oh->ot_handle->h_transaction != NULL);
3490 ll_vfs_dq_init(dir);
3491 dentry = osd_child_dentry_get(env, obj,
3492 (char *)key, strlen((char *)key));
3494 if (obj->oo_hl_head != NULL) {
3495 hlock = osd_oti_get(env)->oti_hlock;
3496 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3497 dir, LDISKFS_HLOCK_DEL);
3499 down_write(&obj->oo_ext_idx_sem);
3502 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3506 /* If this is not the ".." entry, it might be a remote DNE
3507 * entry and we need to check if the FID is for a remote
3508 * MDT. If the FID is not in the directory entry (e.g.
3509 * upgraded 1.8 filesystem without dirdata enabled) then
3510 * we need to get the FID from the LMA. For a remote directory
3511 * there HAS to be an LMA, it cannot be an IGIF inode in this
3514 * Delete the entry before the agent inode in order to
3515 * simplify error handling. At worst an error after deleting
3516 * the entry first might leak the agent inode afterward. The
3517 * reverse would need filesystem abort in case of error deleting
3518 * the entry after the agent had been removed, or leave a
3519 * dangling entry pointing at a random inode. */
3520 if (strcmp((char *)key, dotdot) != 0) {
3521 LASSERT(de != NULL);
3522 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3523 /* If Fid is not in dentry, try to get it from LMA */
3524 if (rc == -ENODATA) {
3525 struct osd_inode_id *id;
3526 struct inode *inode;
3528 /* Before trying to get fid from the inode,
3529 * check whether the inode is valid.
3531 * If the inode has been deleted, do not go
3532 * ahead to do osd_ea_fid_get, which will set
3533 * the inode to bad inode, which might cause
3534 * the inode to be deleted uncorrectly */
3535 inode = ldiskfs_iget(osd_sb(osd),
3536 le32_to_cpu(de->inode));
3537 if (IS_ERR(inode)) {
3538 CDEBUG(D_INODE, "%s: "DFID"get inode"
3539 "error.\n", osd_name(osd),
3541 rc = PTR_ERR(inode);
3543 if (likely(inode->i_nlink != 0)) {
3544 id = &osd_oti_get(env)->oti_id;
3545 rc = osd_ea_fid_get(env, obj,
3546 le32_to_cpu(de->inode),
3549 CDEBUG(D_INFO, "%s: %u "DFID
3552 le32_to_cpu(de->inode),
3560 unlikely(osd_remote_fid(env, osd, fid)))
3561 /* Need to delete agent inode */
3562 ino = le32_to_cpu(de->inode);
3564 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3566 if (rc == 0 && unlikely(ino != 0)) {
3567 rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3570 CERROR("%s: del local inode "DFID": rc = %d\n",
3571 osd_name(osd), PFID(fid), rc);
3577 ldiskfs_htree_unlock(hlock);
3579 up_write(&obj->oo_ext_idx_sem);
3584 /* For inode on the remote MDT, .. will point to
3585 * /Agent directory, Check whether it needs to delete
3586 * from agent directory */
3587 if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3588 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3590 if (rc != 0 && rc != -ENOENT) {
3591 CERROR("%s: delete agent inode "DFID": rc = %d\n",
3592 osd_name(osd), PFID(fid), rc);
3602 LASSERT(osd_invariant(obj));
3607 * Lookup index for \a key and copy record to \a rec.
3609 * \param dt osd index object
3610 * \param key key for index
3611 * \param rec record reference
3613 * \retval +ve success : exact mach
3614 * \retval 0 return record with key not greater than \a key
3615 * \retval -ve failure
3617 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3618 struct dt_rec *rec, const struct dt_key *key)
3620 struct osd_object *obj = osd_dt_obj(dt);
3621 struct iam_path_descr *ipd;
3622 struct iam_container *bag = &obj->oo_dir->od_container;
3623 struct osd_thread_info *oti = osd_oti_get(env);
3624 struct iam_iterator *it = &oti->oti_idx_it;
3625 struct iam_rec *iam_rec;
3629 if (!dt_object_exists(dt))
3632 LASSERT(osd_invariant(obj));
3633 LASSERT(!dt_object_remote(dt));
3634 LASSERT(bag->ic_object == obj->oo_inode);
3636 ipd = osd_idx_ipd_get(env, bag);
3640 /* got ipd now we can start iterator. */
3641 iam_it_init(it, bag, 0, ipd);
3643 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3644 /* swab quota uid/gid provided by caller */
3645 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3646 key = (const struct dt_key *)&oti->oti_quota_id;
3649 rc = iam_it_get(it, (struct iam_key *)key);
3651 if (S_ISDIR(obj->oo_inode->i_mode))
3652 iam_rec = (struct iam_rec *)oti->oti_ldp;
3654 iam_rec = (struct iam_rec *) rec;
3656 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3658 if (S_ISDIR(obj->oo_inode->i_mode))
3659 osd_fid_unpack((struct lu_fid *) rec,
3660 (struct osd_fid_pack *)iam_rec);
3661 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3662 osd_quota_unpack(obj, rec);
3667 osd_ipd_put(env, bag, ipd);
3669 LINVRNT(osd_invariant(obj));
3674 static int osd_index_declare_iam_insert(const struct lu_env *env,
3675 struct dt_object *dt,
3676 const struct dt_rec *rec,
3677 const struct dt_key *key,
3678 struct thandle *handle)
3680 struct osd_thandle *oh;
3682 LASSERT(handle != NULL);
3684 oh = container_of0(handle, struct osd_thandle, ot_super);
3685 LASSERT(oh->ot_handle == NULL);
3687 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3688 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3694 * Inserts (key, value) pair in \a dt index object.
3696 * \param dt osd index object
3697 * \param key key for index
3698 * \param rec record reference
3699 * \param th transaction handler
3702 * \retval -ve failure
3704 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3705 const struct dt_rec *rec,
3706 const struct dt_key *key, struct thandle *th,
3709 struct osd_object *obj = osd_dt_obj(dt);
3710 struct iam_path_descr *ipd;
3711 struct osd_thandle *oh;
3712 struct iam_container *bag;
3713 struct osd_thread_info *oti = osd_oti_get(env);
3714 struct iam_rec *iam_rec;
3718 if (!dt_object_exists(dt))
3721 LINVRNT(osd_invariant(obj));
3722 LASSERT(!dt_object_remote(dt));
3724 bag = &obj->oo_dir->od_container;
3725 LASSERT(bag->ic_object == obj->oo_inode);
3726 LASSERT(th != NULL);
3728 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3730 ipd = osd_idx_ipd_get(env, bag);
3731 if (unlikely(ipd == NULL))
3734 oh = container_of0(th, struct osd_thandle, ot_super);
3735 LASSERT(oh->ot_handle != NULL);
3736 LASSERT(oh->ot_handle->h_transaction != NULL);
3737 if (S_ISDIR(obj->oo_inode->i_mode)) {
3738 iam_rec = (struct iam_rec *)oti->oti_ldp;
3739 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3740 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3741 /* pack quota uid/gid */
3742 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3743 key = (const struct dt_key *)&oti->oti_quota_id;
3744 /* pack quota record */
3745 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3746 iam_rec = (struct iam_rec *)rec;
3748 iam_rec = (struct iam_rec *)rec;
3751 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3753 osd_ipd_put(env, bag, ipd);
3754 LINVRNT(osd_invariant(obj));
3759 * Calls ldiskfs_add_entry() to add directory entry
3760 * into the directory. This is required for
3761 * interoperability mode (b11826)
3763 * \retval 0, on success
3764 * \retval -ve, on error
3766 static int __osd_ea_add_rec(struct osd_thread_info *info,
3767 struct osd_object *pobj, struct inode *cinode,
3768 const char *name, const struct lu_fid *fid,
3769 struct htree_lock *hlock, struct thandle *th)
3771 struct ldiskfs_dentry_param *ldp;
3772 struct dentry *child;
3773 struct osd_thandle *oth;
3776 oth = container_of(th, struct osd_thandle, ot_super);
3777 LASSERT(oth->ot_handle != NULL);
3778 LASSERT(oth->ot_handle->h_transaction != NULL);
3779 LASSERT(pobj->oo_inode);
3781 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3782 if (unlikely(pobj->oo_inode ==
3783 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3786 osd_get_ldiskfs_dirent_param(ldp, fid);
3787 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3788 child->d_fsdata = (void *)ldp;
3789 ll_vfs_dq_init(pobj->oo_inode);
3790 rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3791 if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
3792 struct ldiskfs_dir_entry_2 *de;
3793 struct buffer_head *bh;
3796 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
3799 rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
3802 if (S_ISDIR(cinode->i_mode))
3803 de->file_type = LDISKFS_DIRENT_LUFID |
3804 LDISKFS_FT_REG_FILE;
3806 de->file_type = LDISKFS_DIRENT_LUFID |
3808 ldiskfs_handle_dirty_metadata(oth->ot_handle,
3819 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3820 * into the directory.Also sets flags into osd object to
3821 * indicate dot and dotdot are created. This is required for
3822 * interoperability mode (b11826)
3824 * \param dir directory for dot and dotdot fixup.
3825 * \param obj child object for linking
3827 * \retval 0, on success
3828 * \retval -ve, on error
3830 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3831 struct osd_object *dir,
3832 struct inode *parent_dir, const char *name,
3833 const struct lu_fid *dot_fid,
3834 const struct lu_fid *dot_dot_fid,
3837 struct inode *inode = dir->oo_inode;
3838 struct osd_thandle *oth;
3841 oth = container_of(th, struct osd_thandle, ot_super);
3842 LASSERT(oth->ot_handle->h_transaction != NULL);
3843 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3845 if (strcmp(name, dot) == 0) {
3846 if (dir->oo_compat_dot_created) {
3849 LASSERT(inode == parent_dir);
3850 dir->oo_compat_dot_created = 1;
3853 } else if (strcmp(name, dotdot) == 0) {
3854 if (!dir->oo_compat_dot_created)
3856 /* in case of rename, dotdot is already created */
3857 if (dir->oo_compat_dotdot_created) {
3858 return __osd_ea_add_rec(info, dir, parent_dir, name,
3859 dot_dot_fid, NULL, th);
3862 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
3863 struct lu_fid tfid = *dot_dot_fid;
3866 result = osd_add_dot_dotdot_internal(info,
3867 dir->oo_inode, parent_dir, dot_fid,
3870 result = osd_add_dot_dotdot_internal(info,
3871 dir->oo_inode, parent_dir, dot_fid,
3876 dir->oo_compat_dotdot_created = 1;
3884 * It will call the appropriate osd_add* function and return the
3885 * value, return by respective functions.
3887 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3888 struct inode *cinode, const char *name,
3889 const struct lu_fid *fid, struct thandle *th)
3891 struct osd_thread_info *info = osd_oti_get(env);
3892 struct htree_lock *hlock;
3895 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3897 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3899 if (hlock != NULL) {
3900 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3903 down_write(&pobj->oo_ext_idx_sem);
3906 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3907 lu_object_fid(&pobj->oo_dt.do_lu),
3910 if (hlock != NULL) {
3911 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3912 pobj->oo_inode, LDISKFS_HLOCK_ADD);
3914 down_write(&pobj->oo_ext_idx_sem);
3917 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
3918 struct lu_fid *tfid = &info->oti_fid;
3922 rc = __osd_ea_add_rec(info, pobj, cinode, name,
3925 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3930 ldiskfs_htree_unlock(hlock);
3932 up_write(&pobj->oo_ext_idx_sem);
3938 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3939 struct osd_idmap_cache *oic)
3941 struct osd_scrub *scrub = &dev->od_scrub;
3942 struct lu_fid *fid = &oic->oic_fid;
3943 struct osd_inode_id *id = &oti->oti_id;
3948 if (!fid_is_norm(fid) && !fid_is_igif(fid))
3951 if (scrub->os_pos_current > id->oii_ino)
3955 rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
3956 if (rc != 0 && rc != -ENOENT)
3959 if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
3962 if (thread_is_running(&scrub->os_thread)) {
3963 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
3964 /* There is race condition between osd_oi_lookup and OI scrub.
3965 * The OI scrub finished just after osd_oi_lookup() failure.
3966 * Under such case, it is unnecessary to trigger OI scrub again,
3967 * but try to call osd_oi_lookup() again. */
3968 if (unlikely(rc == -EAGAIN))
3974 if (!dev->od_noscrub && ++once == 1) {
3975 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
3977 CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI scrub by RPC "
3978 "for "DFID", rc = %d [2]\n",
3979 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
3981 if (rc == 0 || rc == -EALREADY)
3988 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
3989 struct osd_device *dev,
3990 struct osd_idmap_cache *oic,
3991 struct lu_fid *fid, __u32 ino)
3993 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
3994 struct inode *inode;
3997 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
3998 inode = osd_iget(oti, dev, &oic->oic_lid);
3999 if (IS_ERR(inode)) {
4000 fid_zero(&oic->oic_fid);
4001 return PTR_ERR(inode);
4004 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4007 fid_zero(&oic->oic_fid);
4009 *fid = oic->oic_fid = lma->lma_self_fid;
4013 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4014 struct osd_inode_id *id, const struct lu_fid *fid)
4016 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4017 id->oii_ino, id->oii_gen, info);
4018 info->oti_cache.oic_lid = *id;
4019 info->oti_cache.oic_fid = *fid;
4020 info->oti_cache.oic_dev = osd;
4024 * Get parent FID from the linkEA.
4026 * For a directory which parent resides on remote MDT, to satisfy the
4027 * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
4028 * the other hand, to make the lookup(..) on the directory can return
4029 * the real parent FID, we append the real parent FID after its ".."
4030 * name entry in the /REMOTE_PARENT_DIR.
4032 * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
4033 * backup. So after the restore, we cannot get the right parent FID from
4034 * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
4035 * we have stored the real parent FID in the directory object's linkEA,
4036 * we can parse the linkEA for the real parent FID.
4038 * \param[in] env pointer to the thread context
4039 * \param[in] obj pointer to the object to be handled
4040 * \param[out]fid pointer to the buffer to hold the parent FID
4042 * \retval 0 for getting the real parent FID successfully
4043 * \retval negative error number on failure
4045 static int osd_get_pfid_from_linkea(const struct lu_env *env,
4046 struct osd_object *obj,
4049 struct osd_thread_info *oti = osd_oti_get(env);
4050 struct lu_buf *buf = &oti->oti_big_buf;
4051 struct dentry *dentry = &oti->oti_obj_dentry;
4052 struct inode *inode = obj->oo_inode;
4053 struct linkea_data ldata = { NULL };
4058 if (!S_ISDIR(inode->i_mode))
4062 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4063 buf->lb_buf, buf->lb_len);
4064 if (rc == -ERANGE) {
4065 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4068 lu_buf_realloc(buf, rc);
4069 if (buf->lb_buf == NULL)
4076 if (unlikely(rc == 0))
4082 if (unlikely(buf->lb_buf == NULL)) {
4083 lu_buf_realloc(buf, rc);
4084 if (buf->lb_buf == NULL)
4091 rc = linkea_init(&ldata);
4093 linkea_first_entry(&ldata);
4094 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
4101 * Calls ->lookup() to find dentry. From dentry get inode and
4102 * read inode's ea to get fid. This is required for interoperability
4105 * \retval 0, on success
4106 * \retval -ve, on error
4108 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4109 struct dt_rec *rec, const struct dt_key *key)
4111 struct inode *dir = obj->oo_inode;
4112 struct dentry *dentry;
4113 struct ldiskfs_dir_entry_2 *de;
4114 struct buffer_head *bh;
4115 struct lu_fid *fid = (struct lu_fid *) rec;
4116 struct htree_lock *hlock = NULL;
4121 LASSERT(dir->i_op != NULL);
4122 LASSERT(dir->i_op->lookup != NULL);
4124 dentry = osd_child_dentry_get(env, obj,
4125 (char *)key, strlen((char *)key));
4127 if (obj->oo_hl_head != NULL) {
4128 hlock = osd_oti_get(env)->oti_hlock;
4129 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4130 dir, LDISKFS_HLOCK_LOOKUP);
4132 down_read(&obj->oo_ext_idx_sem);
4135 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4137 struct osd_thread_info *oti = osd_oti_get(env);
4138 struct osd_inode_id *id = &oti->oti_id;
4139 struct osd_idmap_cache *oic = &oti->oti_cache;
4140 struct osd_device *dev = osd_obj2dev(obj);
4142 ino = le32_to_cpu(de->inode);
4143 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4145 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4149 rc = osd_get_fid_from_dentry(de, rec);
4151 /* done with de, release bh */
4154 if (unlikely(ino == osd_remote_parent_ino(dev))) {
4155 const char *name = (const char *)key;
4157 /* If the parent is on remote MDT, and there
4158 * is no FID-in-dirent, then we have to get
4159 * the parent FID from the linkEA. */
4160 if (likely(strlen(name) == 2 &&
4161 name[0] == '.' && name[1] == '.'))
4162 rc = osd_get_pfid_from_linkea(env, obj,
4165 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4168 osd_id_gen(id, ino, OSD_OII_NOGEN);
4172 fid_zero(&oic->oic_fid);
4176 if (osd_remote_fid(env, dev, fid))
4179 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
4180 osd_consistency_check(oti, dev, oic);
4189 ldiskfs_htree_unlock(hlock);
4191 up_read(&obj->oo_ext_idx_sem);
4196 * Find the osd object for given fid.
4198 * \param fid need to find the osd object having this fid
4200 * \retval osd_object on success
4201 * \retval -ve on error
4203 static struct osd_object *osd_object_find(const struct lu_env *env,
4204 struct dt_object *dt,
4205 const struct lu_fid *fid)
4207 struct lu_device *ludev = dt->do_lu.lo_dev;
4208 struct osd_object *child = NULL;
4209 struct lu_object *luch;
4210 struct lu_object *lo;
4213 * at this point topdev might not exist yet
4214 * (i.e. MGS is preparing profiles). so we can
4215 * not rely on topdev and instead lookup with
4216 * our device passed as topdev. this can't work
4217 * if the object isn't cached yet (as osd doesn't
4218 * allocate lu_header). IOW, the object must be
4219 * in the cache, otherwise lu_object_alloc() crashes
4222 luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4223 ludev : ludev->ld_site->ls_top_dev,
4225 if (!IS_ERR(luch)) {
4226 if (lu_object_exists(luch)) {
4227 lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4229 child = osd_obj(lo);
4231 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4232 "lu_object can't be located"
4233 DFID"\n", PFID(fid));
4235 if (child == NULL) {
4236 lu_object_put(env, luch);
4237 CERROR("Unable to get osd_object\n");
4238 child = ERR_PTR(-ENOENT);
4241 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4242 "lu_object does not exists "DFID"\n",
4244 lu_object_put(env, luch);
4245 child = ERR_PTR(-ENOENT);
4248 child = ERR_CAST(luch);
4255 * Put the osd object once done with it.
4257 * \param obj osd object that needs to be put
4259 static inline void osd_object_put(const struct lu_env *env,
4260 struct osd_object *obj)
4262 lu_object_put(env, &obj->oo_dt.do_lu);
4265 static int osd_index_declare_ea_insert(const struct lu_env *env,
4266 struct dt_object *dt,
4267 const struct dt_rec *rec,
4268 const struct dt_key *key,
4269 struct thandle *handle)
4271 struct osd_thandle *oh;
4272 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4273 struct lu_fid *fid = (struct lu_fid *)rec;
4277 LASSERT(!dt_object_remote(dt));
4278 LASSERT(handle != NULL);
4280 oh = container_of0(handle, struct osd_thandle, ot_super);
4281 LASSERT(oh->ot_handle == NULL);
4283 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4284 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4286 if (osd_dt_obj(dt)->oo_inode != NULL) {
4287 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4289 /* We ignore block quota on meta pool (MDTs), so needn't
4290 * calculate how many blocks will be consumed by this index
4292 rc = osd_declare_inode_qid(env, i_uid_read(inode),
4293 i_gid_read(inode), 0, oh,
4294 osd_dt_obj(dt), true, NULL, false);
4300 rc = osd_remote_fid(env, osd, fid);
4306 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
4307 osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
4308 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4309 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4310 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4311 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4317 * Index add function for interoperability mode (b11826).
4318 * It will add the directory entry.This entry is needed to
4319 * maintain name->fid mapping.
4321 * \param key it is key i.e. file entry to be inserted
4322 * \param rec it is value of given key i.e. fid
4324 * \retval 0, on success
4325 * \retval -ve, on error
4327 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4328 const struct dt_rec *rec,
4329 const struct dt_key *key, struct thandle *th,
4332 struct osd_object *obj = osd_dt_obj(dt);
4333 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4334 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
4335 const struct lu_fid *fid = rec1->rec_fid;
4336 const char *name = (const char *)key;
4337 struct osd_thread_info *oti = osd_oti_get(env);
4338 struct osd_inode_id *id = &oti->oti_id;
4339 struct inode *child_inode = NULL;
4340 struct osd_object *child = NULL;
4344 if (!dt_object_exists(dt))
4347 LASSERT(osd_invariant(obj));
4348 LASSERT(!dt_object_remote(dt));
4349 LASSERT(th != NULL);
4351 osd_trans_exec_op(env, th, OSD_OT_INSERT);
4353 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
4355 rc = osd_remote_fid(env, osd, fid);
4357 CERROR("%s: Can not find object "DFID" rc %d\n",
4358 osd_name(osd), PFID(fid), rc);
4363 /* Insert remote entry */
4364 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4365 struct osd_mdobj_map *omm = osd->od_mdt_map;
4366 struct osd_thandle *oh;
4368 /* If parent on remote MDT, we need put this object
4370 oh = container_of(th, typeof(*oh), ot_super);
4371 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4373 CERROR("%s: add "DFID" error: rc = %d\n",
4375 PFID(lu_object_fid(&dt->do_lu)), rc);
4379 child_inode = igrab(omm->omm_remote_parent->d_inode);
4381 child_inode = osd_create_local_agent_inode(env, osd,
4382 obj, fid, rec1->rec_type & S_IFMT, th);
4383 if (IS_ERR(child_inode))
4384 RETURN(PTR_ERR(child_inode));
4387 /* Insert local entry */
4388 child = osd_object_find(env, dt, fid);
4389 if (IS_ERR(child)) {
4390 CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4391 osd_name(osd), PFID(fid),
4392 id->oii_ino, id->oii_gen,
4393 (int)PTR_ERR(child));
4394 RETURN(PTR_ERR(child));
4396 child_inode = igrab(child->oo_inode);
4399 rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
4401 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4402 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4406 osd_object_put(env, child);
4407 LASSERT(osd_invariant(obj));
4412 * Initialize osd Iterator for given osd index object.
4414 * \param dt osd index object
4417 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4418 struct dt_object *dt,
4421 struct osd_it_iam *it;
4422 struct osd_object *obj = osd_dt_obj(dt);
4423 struct lu_object *lo = &dt->do_lu;
4424 struct iam_path_descr *ipd;
4425 struct iam_container *bag = &obj->oo_dir->od_container;
4427 if (!dt_object_exists(dt))
4428 return ERR_PTR(-ENOENT);
4432 return ERR_PTR(-ENOMEM);
4434 ipd = osd_it_ipd_get(env, bag);
4435 if (likely(ipd != NULL)) {
4439 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4440 return (struct dt_it *)it;
4443 return ERR_PTR(-ENOMEM);
4448 * free given Iterator.
4451 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4453 struct osd_it_iam *it = (struct osd_it_iam *)di;
4454 struct osd_object *obj = it->oi_obj;
4456 iam_it_fini(&it->oi_it);
4457 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4458 lu_object_put(env, &obj->oo_dt.do_lu);
4463 * Move Iterator to record specified by \a key
4465 * \param di osd iterator
4466 * \param key key for index
4468 * \retval +ve di points to record with least key not larger than key
4469 * \retval 0 di points to exact matched key
4470 * \retval -ve failure
4473 static int osd_it_iam_get(const struct lu_env *env,
4474 struct dt_it *di, const struct dt_key *key)
4476 struct osd_thread_info *oti = osd_oti_get(env);
4477 struct osd_it_iam *it = (struct osd_it_iam *)di;
4479 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4480 /* swab quota uid/gid */
4481 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4482 key = (struct dt_key *)&oti->oti_quota_id;
4485 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4491 * \param di osd iterator
4493 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4495 struct osd_it_iam *it = (struct osd_it_iam *)di;
4497 iam_it_put(&it->oi_it);
4501 * Move iterator by one record
4503 * \param di osd iterator
4505 * \retval +1 end of container reached
4507 * \retval -ve failure
4510 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4512 struct osd_it_iam *it = (struct osd_it_iam *)di;
4514 return iam_it_next(&it->oi_it);
4518 * Return pointer to the key under iterator.
4521 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4522 const struct dt_it *di)
4524 struct osd_thread_info *oti = osd_oti_get(env);
4525 struct osd_it_iam *it = (struct osd_it_iam *)di;
4526 struct osd_object *obj = it->oi_obj;
4529 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4531 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4532 /* swab quota uid/gid */
4533 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4534 key = (struct dt_key *)&oti->oti_quota_id;
4541 * Return size of key under iterator (in bytes)
4544 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4546 struct osd_it_iam *it = (struct osd_it_iam *)di;
4548 return iam_it_key_size(&it->oi_it);
4552 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4554 /* check if file type is required */
4555 if (ent->lde_attrs & LUDA_TYPE) {
4556 struct luda_type *lt;
4557 int align = sizeof(*lt) - 1;
4559 len = (len + align) & ~align;
4560 lt = (struct luda_type *)(ent->lde_name + len);
4561 lt->lt_type = cpu_to_le16(DTTOIF(type));
4564 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4568 * build lu direct from backend fs dirent.
4572 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4573 char *name, __u16 namelen, __u16 type, __u32 attr)
4575 ent->lde_attrs = attr | LUDA_FID;
4576 fid_cpu_to_le(&ent->lde_fid, fid);
4578 ent->lde_hash = cpu_to_le64(offset);
4579 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4581 strncpy(ent->lde_name, name, namelen);
4582 ent->lde_name[namelen] = '\0';
4583 ent->lde_namelen = cpu_to_le16(namelen);
4585 /* append lustre attributes */
4586 osd_it_append_attrs(ent, namelen, type);
4590 * Return pointer to the record under iterator.
4592 static int osd_it_iam_rec(const struct lu_env *env,
4593 const struct dt_it *di,
4594 struct dt_rec *dtrec, __u32 attr)
4596 struct osd_it_iam *it = (struct osd_it_iam *)di;
4597 struct osd_thread_info *info = osd_oti_get(env);
4600 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4601 const struct osd_fid_pack *rec;
4602 struct lu_fid *fid = &info->oti_fid;
4603 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
4609 name = (char *)iam_it_key_get(&it->oi_it);
4611 RETURN(PTR_ERR(name));
4613 namelen = iam_it_key_size(&it->oi_it);
4615 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4617 RETURN(PTR_ERR(rec));
4619 rc = osd_fid_unpack(fid, rec);
4623 hash = iam_it_store(&it->oi_it);
4625 /* IAM does not store object type in IAM index (dir) */
4626 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4628 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4629 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4630 (struct iam_rec *)dtrec);
4631 osd_quota_unpack(it->oi_obj, dtrec);
4633 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4634 (struct iam_rec *)dtrec);
4641 * Returns cookie for current Iterator position.
4643 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4645 struct osd_it_iam *it = (struct osd_it_iam *)di;
4647 return iam_it_store(&it->oi_it);
4651 * Restore iterator from cookie.
4653 * \param di osd iterator
4654 * \param hash Iterator location cookie
4656 * \retval +ve di points to record with least key not larger than key.
4657 * \retval 0 di points to exact matched key
4658 * \retval -ve failure
4661 static int osd_it_iam_load(const struct lu_env *env,
4662 const struct dt_it *di, __u64 hash)
4664 struct osd_it_iam *it = (struct osd_it_iam *)di;
4666 return iam_it_load(&it->oi_it, hash);
4669 static const struct dt_index_operations osd_index_iam_ops = {
4670 .dio_lookup = osd_index_iam_lookup,
4671 .dio_declare_insert = osd_index_declare_iam_insert,
4672 .dio_insert = osd_index_iam_insert,
4673 .dio_declare_delete = osd_index_declare_iam_delete,
4674 .dio_delete = osd_index_iam_delete,
4676 .init = osd_it_iam_init,
4677 .fini = osd_it_iam_fini,
4678 .get = osd_it_iam_get,
4679 .put = osd_it_iam_put,
4680 .next = osd_it_iam_next,
4681 .key = osd_it_iam_key,
4682 .key_size = osd_it_iam_key_size,
4683 .rec = osd_it_iam_rec,
4684 .store = osd_it_iam_store,
4685 .load = osd_it_iam_load
4691 * Creates or initializes iterator context.
4693 * \retval struct osd_it_ea, iterator structure on success
4696 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4697 struct dt_object *dt,
4700 struct osd_object *obj = osd_dt_obj(dt);
4701 struct osd_thread_info *info = osd_oti_get(env);
4702 struct osd_it_ea *oie;
4704 struct lu_object *lo = &dt->do_lu;
4705 struct dentry *obj_dentry;
4708 if (!dt_object_exists(dt))
4709 RETURN(ERR_PTR(-ENOENT));
4711 OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
4713 RETURN(ERR_PTR(-ENOMEM));
4714 obj_dentry = &oie->oie_dentry;
4716 obj_dentry->d_inode = obj->oo_inode;
4717 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4718 obj_dentry->d_name.hash = 0;
4720 oie->oie_rd_dirent = 0;
4721 oie->oie_it_dirent = 0;
4722 oie->oie_dirent = NULL;
4723 if (unlikely(!info->oti_it_ea_buf_used)) {
4724 oie->oie_buf = info->oti_it_ea_buf;
4725 info->oti_it_ea_buf_used = 1;
4727 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
4728 if (oie->oie_buf == NULL)
4729 RETURN(ERR_PTR(-ENOMEM));
4733 file = &oie->oie_file;
4735 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4736 if (attr & LUDA_64BITHASH)
4737 file->f_mode = FMODE_64BITHASH;
4739 file->f_mode = FMODE_32BITHASH;
4740 file->f_dentry = obj_dentry;
4741 file->f_mapping = obj->oo_inode->i_mapping;
4742 file->f_op = obj->oo_inode->i_fop;
4743 set_file_inode(file, obj->oo_inode);
4746 RETURN((struct dt_it *) oie);
4750 * Destroy or finishes iterator context.
4752 * \param di iterator structure to be destroyed
4754 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4756 struct osd_thread_info *info = osd_oti_get(env);
4757 struct osd_it_ea *oie = (struct osd_it_ea *)di;
4758 struct osd_object *obj = oie->oie_obj;
4759 struct inode *inode = obj->oo_inode;
4762 oie->oie_file.f_op->release(inode, &oie->oie_file);
4763 lu_object_put(env, &obj->oo_dt.do_lu);
4764 if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
4765 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
4767 info->oti_it_ea_buf_used = 0;
4768 OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
4773 * It position the iterator at given key, so that next lookup continues from
4774 * that key Or it is similar to dio_it->load() but based on a key,
4775 * rather than file position.
4777 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4780 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4782 static int osd_it_ea_get(const struct lu_env *env,
4783 struct dt_it *di, const struct dt_key *key)
4785 struct osd_it_ea *it = (struct osd_it_ea *)di;
4788 LASSERT(((const char *)key)[0] == '\0');
4789 it->oie_file.f_pos = 0;
4790 it->oie_rd_dirent = 0;
4791 it->oie_it_dirent = 0;
4792 it->oie_dirent = NULL;
4800 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4804 struct osd_filldir_cbs {
4805 #ifdef HAVE_DIR_CONTEXT
4806 struct dir_context ctx;
4808 struct osd_it_ea *it;
4811 * It is called internally by ->readdir(). It fills the
4812 * iterator's in-memory data structure with required
4813 * information i.e. name, namelen, rec_size etc.
4815 * \param buf in which information to be filled in.
4816 * \param name name of the file in given dir
4818 * \retval 0 on success
4819 * \retval 1 on buffer full
4821 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
4822 loff_t offset, __u64 ino,
4825 struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
4826 struct osd_object *obj = it->oie_obj;
4827 struct osd_it_ea_dirent *ent = it->oie_dirent;
4828 struct lu_fid *fid = &ent->oied_fid;
4829 struct osd_fid_pack *rec;
4832 /* this should never happen */
4833 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4834 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4838 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4842 /* "." is just the object itself. */
4843 if (namelen == 1 && name[0] == '.') {
4844 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4845 } else if (d_type & LDISKFS_DIRENT_LUFID) {
4846 rec = (struct osd_fid_pack*) (name + namelen + 1);
4847 if (osd_fid_unpack(fid, rec) != 0)
4852 d_type &= ~LDISKFS_DIRENT_LUFID;
4854 /* NOT export local root. */
4855 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
4856 ino = obj->oo_inode->i_ino;
4857 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4860 ent->oied_ino = ino;
4861 ent->oied_off = offset;
4862 ent->oied_namelen = namelen;
4863 ent->oied_type = d_type;
4865 memcpy(ent->oied_name, name, namelen);
4867 it->oie_rd_dirent++;
4868 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4873 * Calls ->readdir() to load a directory entry at a time
4874 * and stored it in iterator's in-memory data structure.
4876 * \param di iterator's in memory structure
4878 * \retval 0 on success
4879 * \retval -ve on error
4880 * \retval +1 reach the end of entry
4882 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4883 const struct dt_it *di)
4885 struct osd_it_ea *it = (struct osd_it_ea *)di;
4886 struct osd_object *obj = it->oie_obj;
4887 struct inode *inode = obj->oo_inode;
4888 struct htree_lock *hlock = NULL;
4889 struct file *filp = &it->oie_file;
4891 struct osd_filldir_cbs buf = {
4892 #ifdef HAVE_DIR_CONTEXT
4893 .ctx.actor = osd_ldiskfs_filldir,
4899 it->oie_dirent = it->oie_buf;
4900 it->oie_rd_dirent = 0;
4902 if (obj->oo_hl_head != NULL) {
4903 hlock = osd_oti_get(env)->oti_hlock;
4904 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4905 inode, LDISKFS_HLOCK_READDIR);
4907 down_read(&obj->oo_ext_idx_sem);
4910 #ifdef HAVE_DIR_CONTEXT
4911 buf.ctx.pos = filp->f_pos;
4912 rc = inode->i_fop->iterate(filp, &buf.ctx);
4913 filp->f_pos = buf.ctx.pos;
4915 rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
4919 ldiskfs_htree_unlock(hlock);
4921 up_read(&obj->oo_ext_idx_sem);
4923 if (it->oie_rd_dirent == 0) {
4924 /*If it does not get any dirent, it means it has been reached
4925 *to the end of the dir */
4926 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
4930 it->oie_dirent = it->oie_buf;
4931 it->oie_it_dirent = 1;
4938 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4939 * to load a directory entry at a time and stored it in
4940 * iterator's in-memory data structure.
4942 * \param di iterator's in memory structure
4944 * \retval +ve iterator reached to end
4945 * \retval 0 iterator not reached to end
4946 * \retval -ve on error
4948 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4950 struct osd_it_ea *it = (struct osd_it_ea *)di;
4955 if (it->oie_it_dirent < it->oie_rd_dirent) {
4957 (void *) it->oie_dirent +
4958 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4959 it->oie_dirent->oied_namelen);
4960 it->oie_it_dirent++;
4963 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4966 rc = osd_ldiskfs_it_fill(env, di);
4973 * Returns the key at current position from iterator's in memory structure.
4975 * \param di iterator's in memory structure
4977 * \retval key i.e. struct dt_key on success
4979 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4980 const struct dt_it *di)
4982 struct osd_it_ea *it = (struct osd_it_ea *)di;
4984 return (struct dt_key *)it->oie_dirent->oied_name;
4988 * Returns the key's size at current position from iterator's in memory structure.
4990 * \param di iterator's in memory structure
4992 * \retval key_size i.e. struct dt_key on success
4994 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4996 struct osd_it_ea *it = (struct osd_it_ea *)di;
4998 return it->oie_dirent->oied_namelen;
5002 osd_dirent_update(handle_t *jh, struct super_block *sb,
5003 struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5004 struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
5006 struct osd_fid_pack *rec;
5010 LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
5011 LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
5013 rc = ldiskfs_journal_get_write_access(jh, bh);
5017 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5018 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5019 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
5025 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
5027 if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
5028 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5035 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5037 LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5038 "dot_dotdot = %d\n", dot_dotdot);
5040 if (LDISKFS_DIR_REC_LEN(de) >=
5041 __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5048 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5049 struct inode *dir, struct inode *inode,
5050 struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5051 struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5052 struct htree_lock *hlock)
5054 struct dentry *dentry;
5055 struct osd_fid_pack *rec;
5056 struct ldiskfs_dentry_param *ldp;
5060 if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5061 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5064 /* There is enough space to hold the FID-in-dirent. */
5065 if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
5066 dir->i_sb->s_blocksize)) {
5067 rc = ldiskfs_journal_get_write_access(jh, bh);
5071 de->name[de->name_len] = 0;
5072 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5073 rec->fp_len = sizeof(struct lu_fid) + 1;
5074 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5075 de->file_type |= LDISKFS_DIRENT_LUFID;
5077 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
5082 rc = ldiskfs_delete_entry(jh, dir, de, bh);
5086 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5088 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5089 osd_get_ldiskfs_dirent_param(ldp, fid);
5090 dentry->d_fsdata = (void *)ldp;
5091 ll_vfs_dq_init(dir);
5092 rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5093 /* It is too bad, we cannot reinsert the name entry back.
5094 * That means we lose it! */
5096 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
5097 "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
5098 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5099 dir->i_ino, dir->i_generation,
5100 ent->oied_namelen, ent->oied_name, PFID(fid), rc);
5106 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5107 struct osd_it_ea *it, struct lu_fid *fid,
5108 struct osd_inode_id *id, __u32 *attr)
5110 struct osd_thread_info *info = osd_oti_get(env);
5111 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
5112 struct osd_device *dev = osd_obj2dev(obj);
5113 struct super_block *sb = osd_sb(dev);
5114 const char *devname =
5115 LDISKFS_SB(sb)->s_es->s_volume_name;
5116 struct osd_it_ea_dirent *ent = it->oie_dirent;
5117 struct inode *dir = obj->oo_inode;
5118 struct htree_lock *hlock = NULL;
5119 struct buffer_head *bh = NULL;
5120 handle_t *jh = NULL;
5121 struct ldiskfs_dir_entry_2 *de;
5122 struct dentry *dentry;
5123 struct inode *inode;
5130 if (ent->oied_name[0] == '.') {
5131 if (ent->oied_namelen == 1)
5133 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5137 dentry = osd_child_dentry_get(env, obj, ent->oied_name,
5140 /* We need to ensure that the name entry is still valid.
5141 * Because it may be removed or renamed by other already.
5143 * The unlink or rename operation will start journal before PDO lock,
5144 * so to avoid deadlock, here we need to start journal handle before
5145 * related PDO lock also. But because we do not know whether there
5146 * will be something to be repaired before PDO lock, we just start
5147 * journal without conditions.
5149 * We may need to remove the name entry firstly, then insert back.
5150 * One credit is for user quota file update.
5151 * One credit is for group quota file update.
5152 * Two credits are for dirty inode. */
5153 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5154 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5157 if (dev->od_dirent_journal != 0) {
5158 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5161 CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent "
5162 "check_repair, dir = %lu/%u, credits = %d, "
5163 "name = %.*s: rc = %d\n",
5164 devname, dir->i_ino, dir->i_generation, credits,
5165 ent->oied_namelen, ent->oied_name, rc);
5169 if (obj->oo_hl_head != NULL) {
5170 hlock = osd_oti_get(env)->oti_hlock;
5171 /* "0" means exclusive lock for the whole directory.
5172 * We need to prevent others access such name entry
5173 * during the delete + insert. Neither HLOCK_ADD nor
5174 * HLOCK_DEL cannot guarantee the atomicity. */
5175 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5177 down_write(&obj->oo_ext_idx_sem);
5180 if (obj->oo_hl_head != NULL) {
5181 hlock = osd_oti_get(env)->oti_hlock;
5182 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5183 LDISKFS_HLOCK_LOOKUP);
5185 down_read(&obj->oo_ext_idx_sem);
5189 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5190 /* For dot/dotdot entry, if there is not enough space to hold the
5191 * FID-in-dirent, just keep them there. It only happens when the
5192 * device upgraded from 1.8 or restored from MDT file-level backup.
5193 * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5194 * and needs to get FID from LMA when readdir, it will not affect the
5195 * performance much. */
5196 if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
5197 (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5198 *attr |= LUDA_IGNORE;
5199 GOTO(out_journal, rc = 0);
5202 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5203 inode = osd_iget(info, dev, id);
5204 if (IS_ERR(inode)) {
5205 rc = PTR_ERR(inode);
5206 if (rc == -ENOENT || rc == -ESTALE)
5209 CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
5210 "check_repair, dir = %lu/%u, name = %.*s: "
5212 devname, dir->i_ino, dir->i_generation,
5213 ent->oied_namelen, ent->oied_name, rc);
5215 GOTO(out_journal, rc);
5218 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
5220 LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
5222 if (fid_is_sane(fid)) {
5223 /* FID-in-dirent is valid. */
5224 if (lu_fid_eq(fid, &lma->lma_self_fid))
5225 GOTO(out_inode, rc = 0);
5227 /* Do not repair under dryrun mode. */
5228 if (*attr & LUDA_VERIFY_DRYRUN) {
5229 *attr |= LUDA_REPAIR;
5230 GOTO(out_inode, rc = 0);
5233 if (dev->od_dirent_journal == 0) {
5237 ldiskfs_htree_unlock(hlock);
5239 up_read(&obj->oo_ext_idx_sem);
5240 dev->od_dirent_journal = 1;
5244 *fid = lma->lma_self_fid;
5246 /* Update the FID-in-dirent. */
5247 rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
5249 *attr |= LUDA_REPAIR;
5251 CDEBUG(D_LFSCK, "%.16s: fail to update FID "
5252 "in the dirent, dir = %lu/%u, "
5253 "name = %.*s, "DFID": rc = %d\n",
5254 devname, dir->i_ino, dir->i_generation,
5255 ent->oied_namelen, ent->oied_name,
5258 /* Do not repair under dryrun mode. */
5259 if (*attr & LUDA_VERIFY_DRYRUN) {
5260 *fid = lma->lma_self_fid;
5261 *attr |= LUDA_REPAIR;
5262 GOTO(out_inode, rc = 0);
5265 if (dev->od_dirent_journal == 0) {
5269 ldiskfs_htree_unlock(hlock);
5271 up_read(&obj->oo_ext_idx_sem);
5272 dev->od_dirent_journal = 1;
5276 *fid = lma->lma_self_fid;
5278 /* Append the FID-in-dirent. */
5279 rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5280 fid, bh, de, hlock);
5282 *attr |= LUDA_REPAIR;
5284 CDEBUG(D_LFSCK, "%.16s: fail to append FID "
5285 "after the dirent, dir = %lu/%u, "
5286 "name = %.*s, "DFID": rc = %d\n",
5287 devname, dir->i_ino, dir->i_generation,
5288 ent->oied_namelen, ent->oied_name,
5291 } else if (rc == -ENODATA) {
5292 /* Do not repair under dryrun mode. */
5293 if (*attr & LUDA_VERIFY_DRYRUN) {
5294 if (fid_is_sane(fid)) {
5295 *attr |= LUDA_REPAIR;
5297 lu_igif_build(fid, inode->i_ino,
5298 inode->i_generation);
5299 *attr |= LUDA_UPGRADE;
5301 GOTO(out_inode, rc = 0);
5304 if (dev->od_dirent_journal == 0) {
5308 ldiskfs_htree_unlock(hlock);
5310 up_read(&obj->oo_ext_idx_sem);
5311 dev->od_dirent_journal = 1;
5316 if (unlikely(fid_is_sane(fid))) {
5317 /* FID-in-dirent exists, but FID-in-LMA is lost.
5318 * Trust the FID-in-dirent, and add FID-in-LMA. */
5319 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5321 *attr |= LUDA_REPAIR;
5323 CDEBUG(D_LFSCK, "%.16s: fail to set LMA for "
5324 "update dirent, dir = %lu/%u, "
5325 "name = %.*s, "DFID": rc = %d\n",
5326 devname, dir->i_ino, dir->i_generation,
5327 ent->oied_namelen, ent->oied_name,
5330 lu_igif_build(fid, inode->i_ino, inode->i_generation);
5331 /* It is probably IGIF object. Only aappend the
5332 * FID-in-dirent. OI scrub will process FID-in-LMA. */
5333 rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5334 fid, bh, de, hlock);
5336 *attr |= LUDA_UPGRADE;
5338 CDEBUG(D_LFSCK, "%.16s: fail to append IGIF "
5339 "after the dirent, dir = %lu/%u, "
5340 "name = %.*s, "DFID": rc = %d\n",
5341 devname, dir->i_ino, dir->i_generation,
5342 ent->oied_namelen, ent->oied_name,
5347 GOTO(out_inode, rc);
5354 if (hlock != NULL) {
5355 ldiskfs_htree_unlock(hlock);
5357 if (dev->od_dirent_journal != 0)
5358 up_write(&obj->oo_ext_idx_sem);
5360 up_read(&obj->oo_ext_idx_sem);
5363 ldiskfs_journal_stop(jh);
5364 if (rc >= 0 && !dirty)
5365 dev->od_dirent_journal = 0;
5370 * Returns the value at current position from iterator's in memory structure.
5372 * \param di struct osd_it_ea, iterator's in memory structure
5373 * \param attr attr requested for dirent.
5374 * \param lde lustre dirent
5376 * \retval 0 no error and \param lde has correct lustre dirent.
5377 * \retval -ve on error
5379 static inline int osd_it_ea_rec(const struct lu_env *env,
5380 const struct dt_it *di,
5381 struct dt_rec *dtrec, __u32 attr)
5383 struct osd_it_ea *it = (struct osd_it_ea *)di;
5384 struct osd_object *obj = it->oie_obj;
5385 struct osd_device *dev = osd_obj2dev(obj);
5386 struct osd_thread_info *oti = osd_oti_get(env);
5387 struct osd_inode_id *id = &oti->oti_id;
5388 struct lu_fid *fid = &it->oie_dirent->oied_fid;
5389 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
5390 __u32 ino = it->oie_dirent->oied_ino;
5394 LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
5396 if (attr & LUDA_VERIFY) {
5397 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5398 attr |= LUDA_IGNORE;
5399 /* If the parent is on remote MDT, and there
5400 * is no FID-in-dirent, then we have to get
5401 * the parent FID from the linkEA. */
5402 if (!fid_is_sane(fid) &&
5403 it->oie_dirent->oied_namelen == 2 &&
5404 it->oie_dirent->oied_name[0] == '.' &&
5405 it->oie_dirent->oied_name[1] == '.')
5406 osd_get_pfid_from_linkea(env, obj, fid);
5408 rc = osd_dirent_check_repair(env, obj, it, fid, id,
5412 attr &= ~LU_DIRENT_ATTRS_MASK;
5413 if (!fid_is_sane(fid)) {
5414 if (it->oie_dirent->oied_namelen == 2 &&
5415 it->oie_dirent->oied_name[0] == '.' &&
5416 it->oie_dirent->oied_name[1] == '.') {
5417 /* If the parent is on remote MDT, and there
5418 * is no FID-in-dirent, then we have to get
5419 * the parent FID from the linkEA. */
5420 if (ino == osd_remote_parent_ino(dev))
5421 rc = osd_get_pfid_from_linkea(env, obj,
5424 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
5427 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5430 osd_id_gen(id, ino, OSD_OII_NOGEN);
5434 /* Pack the entry anyway, at least the offset is right. */
5435 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5436 it->oie_dirent->oied_name,
5437 it->oie_dirent->oied_namelen,
5438 it->oie_dirent->oied_type, attr);
5443 if (osd_remote_fid(env, dev, fid))
5446 if (likely(!(attr & LUDA_IGNORE) && rc == 0))
5447 osd_add_oi_cache(oti, dev, id, fid);
5449 RETURN(rc > 0 ? 0 : rc);
5453 * Returns the record size size at current position.
5455 * This function will return record(lu_dirent) size in bytes.
5457 * \param[in] env execution environment
5458 * \param[in] di iterator's in memory structure
5459 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
5460 * calculate the lu_dirent size.
5462 * \retval record size(in bytes & in memory) of the current lu_dirent
5465 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5468 struct osd_it_ea *it = (struct osd_it_ea *)di;
5470 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5474 * Returns a cookie for current position of the iterator head, so that
5475 * user can use this cookie to load/start the iterator next time.
5477 * \param di iterator's in memory structure
5479 * \retval cookie for current position, on success
5481 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5483 struct osd_it_ea *it = (struct osd_it_ea *)di;
5485 return it->oie_dirent->oied_off;
5489 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5490 * to load a directory entry at a time and stored it i inn,
5491 * in iterator's in-memory data structure.
5493 * \param di struct osd_it_ea, iterator's in memory structure
5495 * \retval +ve on success
5496 * \retval -ve on error
5498 static int osd_it_ea_load(const struct lu_env *env,
5499 const struct dt_it *di, __u64 hash)
5501 struct osd_it_ea *it = (struct osd_it_ea *)di;
5505 it->oie_file.f_pos = hash;
5507 rc = osd_ldiskfs_it_fill(env, di);
5518 * Index lookup function for interoperability mode (b11826).
5520 * \param key, key i.e. file name to be searched
5522 * \retval +ve, on success
5523 * \retval -ve, on error
5525 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5526 struct dt_rec *rec, const struct dt_key *key)
5528 struct osd_object *obj = osd_dt_obj(dt);
5533 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5534 LINVRNT(osd_invariant(obj));
5536 rc = osd_ea_lookup_rec(env, obj, rec, key);
5543 * Index and Iterator operations for interoperability
5544 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5546 static const struct dt_index_operations osd_index_ea_ops = {
5547 .dio_lookup = osd_index_ea_lookup,
5548 .dio_declare_insert = osd_index_declare_ea_insert,
5549 .dio_insert = osd_index_ea_insert,
5550 .dio_declare_delete = osd_index_declare_ea_delete,
5551 .dio_delete = osd_index_ea_delete,
5553 .init = osd_it_ea_init,
5554 .fini = osd_it_ea_fini,
5555 .get = osd_it_ea_get,
5556 .put = osd_it_ea_put,
5557 .next = osd_it_ea_next,
5558 .key = osd_it_ea_key,
5559 .key_size = osd_it_ea_key_size,
5560 .rec = osd_it_ea_rec,
5561 .rec_size = osd_it_ea_rec_size,
5562 .store = osd_it_ea_store,
5563 .load = osd_it_ea_load
5567 static void *osd_key_init(const struct lu_context *ctx,
5568 struct lu_context_key *key)
5570 struct osd_thread_info *info;
5572 OBD_ALLOC_PTR(info);
5574 return ERR_PTR(-ENOMEM);
5576 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5577 if (info->oti_it_ea_buf == NULL)
5580 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5582 info->oti_hlock = ldiskfs_htree_lock_alloc();
5583 if (info->oti_hlock == NULL)
5589 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5592 return ERR_PTR(-ENOMEM);
5595 static void osd_key_fini(const struct lu_context *ctx,
5596 struct lu_context_key *key, void* data)
5598 struct osd_thread_info *info = data;
5600 if (info->oti_inode != NULL)
5601 OBD_FREE_PTR(info->oti_inode);
5602 if (info->oti_hlock != NULL)
5603 ldiskfs_htree_lock_free(info->oti_hlock);
5604 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5605 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5606 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5607 lu_buf_free(&info->oti_big_buf);
5611 static void osd_key_exit(const struct lu_context *ctx,
5612 struct lu_context_key *key, void *data)
5614 struct osd_thread_info *info = data;
5616 LASSERT(info->oti_r_locks == 0);
5617 LASSERT(info->oti_w_locks == 0);
5618 LASSERT(info->oti_txns == 0);
5621 /* type constructor/destructor: osd_type_init, osd_type_fini */
5622 LU_TYPE_INIT_FINI(osd, &osd_key);
5624 struct lu_context_key osd_key = {
5625 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5626 .lct_init = osd_key_init,
5627 .lct_fini = osd_key_fini,
5628 .lct_exit = osd_key_exit
5632 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5633 const char *name, struct lu_device *next)
5635 struct osd_device *osd = osd_dev(d);
5637 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5638 >= sizeof(osd->od_svname))
5640 return osd_procfs_init(osd, name);
5643 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5645 struct seq_server_site *ss = osd_seq_site(osd);
5649 if (osd->od_is_ost || osd->od_cl_seq != NULL)
5652 if (unlikely(ss == NULL))
5655 OBD_ALLOC_PTR(osd->od_cl_seq);
5656 if (osd->od_cl_seq == NULL)
5659 rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5660 osd->od_svname, ss->ss_server_seq);
5663 OBD_FREE_PTR(osd->od_cl_seq);
5664 osd->od_cl_seq = NULL;
5670 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5672 if (osd->od_cl_seq == NULL)
5675 seq_client_fini(osd->od_cl_seq);
5676 OBD_FREE_PTR(osd->od_cl_seq);
5677 osd->od_cl_seq = NULL;
5680 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5684 /* shutdown quota slave instance associated with the device */
5685 if (o->od_quota_slave != NULL) {
5686 qsd_fini(env, o->od_quota_slave);
5687 o->od_quota_slave = NULL;
5690 osd_fid_fini(env, o);
5695 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5699 if (o->od_mnt != NULL) {
5700 shrink_dcache_sb(osd_sb(o));
5701 osd_sync(env, &o->od_dt_dev);
5710 static int osd_mount(const struct lu_env *env,
5711 struct osd_device *o, struct lustre_cfg *cfg)
5713 const char *name = lustre_cfg_string(cfg, 0);
5714 const char *dev = lustre_cfg_string(cfg, 1);
5716 unsigned long page, s_flags, lmd_flags = 0;
5717 struct page *__page;
5718 struct file_system_type *type;
5719 char *options = NULL;
5721 struct osd_thread_info *info = osd_oti_get(env);
5722 struct lu_fid *fid = &info->oti_fid;
5723 struct inode *inode;
5724 int rc = 0, force_over_128tb = 0;
5727 if (o->od_mnt != NULL)
5730 if (strlen(dev) >= sizeof(o->od_mntdev))
5732 strcpy(o->od_mntdev, dev);
5734 str = lustre_cfg_string(cfg, 2);
5735 s_flags = simple_strtoul(str, NULL, 0);
5736 str = strstr(str, ":");
5738 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5739 opts = lustre_cfg_string(cfg, 3);
5741 if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
5742 CERROR("%s: device %s extents feature is not guaranteed to "
5743 "work on big-endian systems. Use \"bigendian_extents\" "
5744 "mount option to override.\n", name, dev);
5748 if (opts != NULL && strstr(opts, "force_over_128tb") != NULL)
5749 force_over_128tb = 1;
5751 OBD_PAGE_ALLOC(__page, GFP_IOFS);
5753 GOTO(out, rc = -ENOMEM);
5754 page = (unsigned long)page_address(__page);
5755 options = (char *)page;
5758 /* strip out the options for back compatiblity */
5759 static char *sout[] = {
5766 /* strip out option we processed in osd */
5767 "bigendian_extents",
5771 strcat(options, opts);
5772 for (rc = 0, str = options; sout[rc]; ) {
5773 char *op = strstr(str, sout[rc]);
5779 if (op == options || *(op - 1) == ',') {
5780 str = op + strlen(sout[rc]);
5781 if (*str == ',' || *str == '\0') {
5782 *str == ',' ? str++ : str;
5783 memmove(op, str, strlen(str) + 1);
5786 for (str = op; *str != ',' && *str != '\0'; str++)
5790 strncat(options, "user_xattr,acl", 14);
5793 /* Glom up mount options */
5794 if (*options != '\0')
5795 strcat(options, ",");
5796 strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
5798 type = get_fs_type("ldiskfs");
5800 CERROR("%s: cannot find ldiskfs module\n", name);
5801 GOTO(out, rc = -ENODEV);
5804 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5805 module_put(type->owner);
5807 if (IS_ERR(o->od_mnt)) {
5808 rc = PTR_ERR(o->od_mnt);
5810 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5814 if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) > (8ULL << 32) &&
5815 force_over_128tb == 0) {
5816 CERROR("%s: device %s LDISKFS does not support filesystems "
5817 "greater than 128TB and can cause data corruption. "
5818 "Use \"force_over_128tb\" mount option to override.\n",
5820 GOTO(out, rc = -EINVAL);
5823 #ifdef HAVE_DEV_SET_RDONLY
5824 if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5825 CERROR("%s: underlying device %s is marked as read-only. "
5826 "Setup failed\n", name, dev);
5827 GOTO(out_mnt, rc = -EROFS);
5831 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5832 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5833 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5834 GOTO(out_mnt, rc = -EINVAL);
5837 #ifdef LDISKFS_MOUNT_DIRDATA
5838 if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
5839 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5840 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
5841 else if (!o->od_is_ost)
5842 CWARN("%s: device %s was upgraded from Lustre-1.x without "
5843 "enabling the dirdata feature. If you do not want to "
5844 "downgrade to Lustre-1.x again, you can enable it via "
5845 "'tune2fs -O dirdata device'\n", name, dev);
5847 inode = osd_sb(o)->s_root->d_inode;
5848 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
5849 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
5851 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
5855 if (lmd_flags & LMD_FLG_NOSCRUB)
5866 OBD_PAGE_FREE(__page);
5871 static struct lu_device *osd_device_fini(const struct lu_env *env,
5872 struct lu_device *d)
5874 struct osd_device *o = osd_dev(d);
5877 osd_shutdown(env, o);
5879 osd_scrub_cleanup(env, o);
5880 osd_obj_map_fini(o);
5886 static int osd_device_init0(const struct lu_env *env,
5887 struct osd_device *o,
5888 struct lustre_cfg *cfg)
5890 struct lu_device *l = osd2lu_dev(o);
5891 struct osd_thread_info *info;
5895 /* if the module was re-loaded, env can loose its keys */
5896 rc = lu_env_refill((struct lu_env *) env);
5899 info = osd_oti_get(env);
5902 l->ld_ops = &osd_lu_ops;
5903 o->od_dt_dev.dd_ops = &osd_dt_ops;
5905 spin_lock_init(&o->od_osfs_lock);
5906 mutex_init(&o->od_otable_mutex);
5908 o->od_read_cache = 1;
5909 o->od_writethrough_cache = 1;
5910 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
5912 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
5913 sizeof(o->od_svname));
5914 if (cplen >= sizeof(o->od_svname)) {
5919 if (server_name_is_ost(o->od_svname))
5922 o->od_full_scrub_ratio = OFSR_DEFAULT;
5923 o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
5924 rc = osd_mount(env, o, cfg);
5928 rc = osd_obj_map_init(env, o);
5932 rc = lu_site_init(&o->od_site, l);
5934 GOTO(out_compat, rc);
5935 o->od_site.ls_bottom_dev = l;
5937 rc = lu_site_init_finish(&o->od_site);
5941 INIT_LIST_HEAD(&o->od_ios_list);
5942 /* setup scrub, including OI files initialization */
5943 rc = osd_scrub_setup(env, o);
5947 rc = osd_procfs_init(o, o->od_svname);
5949 CERROR("%s: can't initialize procfs: rc = %d\n",
5951 GOTO(out_scrub, rc);
5954 LASSERT(l->ld_site->ls_linkage.next != NULL);
5955 LASSERT(l->ld_site->ls_linkage.prev != NULL);
5957 /* initialize quota slave instance */
5958 o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
5960 if (IS_ERR(o->od_quota_slave)) {
5961 rc = PTR_ERR(o->od_quota_slave);
5962 o->od_quota_slave = NULL;
5963 GOTO(out_procfs, rc);
5971 osd_scrub_cleanup(env, o);
5973 lu_site_fini(&o->od_site);
5975 osd_obj_map_fini(o);
5982 static struct lu_device *osd_device_alloc(const struct lu_env *env,
5983 struct lu_device_type *t,
5984 struct lustre_cfg *cfg)
5986 struct osd_device *o;
5991 return ERR_PTR(-ENOMEM);
5993 rc = dt_device_init(&o->od_dt_dev, t);
5995 /* Because the ctx might be revived in dt_device_init,
5996 * refill the env here */
5997 lu_env_refill((struct lu_env *)env);
5998 rc = osd_device_init0(env, o, cfg);
6000 dt_device_fini(&o->od_dt_dev);
6003 if (unlikely(rc != 0))
6006 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
6009 static struct lu_device *osd_device_free(const struct lu_env *env,
6010 struct lu_device *d)
6012 struct osd_device *o = osd_dev(d);
6015 /* XXX: make osd top device in order to release reference */
6016 d->ld_site->ls_top_dev = d;
6017 lu_site_purge(env, d->ld_site, -1);
6018 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
6019 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
6020 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
6022 lu_site_fini(&o->od_site);
6023 dt_device_fini(&o->od_dt_dev);
6028 static int osd_process_config(const struct lu_env *env,
6029 struct lu_device *d, struct lustre_cfg *cfg)
6031 struct osd_device *o = osd_dev(d);
6035 switch (cfg->lcfg_command) {
6037 rc = osd_mount(env, o, cfg);
6040 lu_dev_del_linkage(d->ld_site, d);
6041 rc = osd_shutdown(env, o);
6044 LASSERT(&o->od_dt_dev);
6045 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
6046 cfg, &o->od_dt_dev);
6047 if (rc > 0 || rc == -ENOSYS)
6048 rc = class_process_proc_param(PARAM_OST,
6049 lprocfs_osd_obd_vars,
6050 cfg, &o->od_dt_dev);
6059 static int osd_recovery_complete(const struct lu_env *env,
6060 struct lu_device *d)
6062 struct osd_device *osd = osd_dev(d);
6066 if (osd->od_quota_slave == NULL)
6069 /* start qsd instance on recovery completion, this notifies the quota
6070 * slave code that we are about to process new requests now */
6071 rc = qsd_start(env, osd->od_quota_slave);
6076 * we use exports to track all osd users
6078 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
6079 struct obd_device *obd, struct obd_uuid *cluuid,
6080 struct obd_connect_data *data, void *localdata)
6082 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6083 struct lustre_handle conn;
6087 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6089 rc = class_connect(&conn, obd, cluuid);
6093 *exp = class_conn2export(&conn);
6095 spin_lock(&osd->od_osfs_lock);
6097 spin_unlock(&osd->od_osfs_lock);
6103 * once last export (we don't count self-export) disappeared
6104 * osd can be released
6106 static int osd_obd_disconnect(struct obd_export *exp)
6108 struct obd_device *obd = exp->exp_obd;
6109 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6110 int rc, release = 0;
6113 /* Only disconnect the underlying layers on the final disconnect. */
6114 spin_lock(&osd->od_osfs_lock);
6116 if (osd->od_connects == 0)
6118 spin_unlock(&osd->od_osfs_lock);
6120 rc = class_disconnect(exp); /* bz 9811 */
6122 if (rc == 0 && release)
6123 class_manual_cleanup(obd);
6127 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6128 struct lu_device *dev)
6130 struct osd_device *osd = osd_dev(dev);
6131 struct lr_server_data *lsd =
6132 &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
6136 if (osd->od_quota_slave != NULL) {
6137 /* set up quota slave objects */
6138 result = qsd_prepare(env, osd->od_quota_slave);
6143 if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
6144 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
6145 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
6146 osd->od_index_in_idif = 1;
6148 osd->od_index_in_idif = 0;
6149 result = osd_register_proc_index_in_idif(osd);
6154 osd->od_index_in_idif = 1;
6158 result = osd_fid_init(env, osd);
6163 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6164 struct lu_fid *fid, struct md_op_data *op_data)
6166 struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6168 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6171 static const struct lu_object_operations osd_lu_obj_ops = {
6172 .loo_object_init = osd_object_init,
6173 .loo_object_delete = osd_object_delete,
6174 .loo_object_release = osd_object_release,
6175 .loo_object_free = osd_object_free,
6176 .loo_object_print = osd_object_print,
6177 .loo_object_invariant = osd_object_invariant
6180 const struct lu_device_operations osd_lu_ops = {
6181 .ldo_object_alloc = osd_object_alloc,
6182 .ldo_process_config = osd_process_config,
6183 .ldo_recovery_complete = osd_recovery_complete,
6184 .ldo_prepare = osd_prepare,
6187 static const struct lu_device_type_operations osd_device_type_ops = {
6188 .ldto_init = osd_type_init,
6189 .ldto_fini = osd_type_fini,
6191 .ldto_start = osd_type_start,
6192 .ldto_stop = osd_type_stop,
6194 .ldto_device_alloc = osd_device_alloc,
6195 .ldto_device_free = osd_device_free,
6197 .ldto_device_init = osd_device_init,
6198 .ldto_device_fini = osd_device_fini
6201 static struct lu_device_type osd_device_type = {
6202 .ldt_tags = LU_DEVICE_DT,
6203 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
6204 .ldt_ops = &osd_device_type_ops,
6205 .ldt_ctx_tags = LCT_LOCAL,
6208 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
6210 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6211 struct super_block *sb = osd_sb(osd);
6213 return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
6217 * lprocfs legacy support.
6219 static struct obd_ops osd_obd_device_ops = {
6220 .o_owner = THIS_MODULE,
6221 .o_connect = osd_obd_connect,
6222 .o_disconnect = osd_obd_disconnect,
6223 .o_fid_alloc = osd_fid_alloc,
6224 .o_health_check = osd_health_check,
6227 static int __init osd_mod_init(void)
6231 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
6232 /* please, try to keep osd_thread_info smaller than a page */
6233 CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
6238 rc = lu_kmem_init(ldiskfs_caches);
6242 rc = class_register_type(&osd_obd_device_ops, NULL, true,
6243 lprocfs_osd_module_vars,
6244 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6246 lu_kmem_fini(ldiskfs_caches);
6250 static void __exit osd_mod_exit(void)
6252 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6253 lu_kmem_fini(ldiskfs_caches);
6256 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6257 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6258 MODULE_VERSION(LUSTRE_VERSION_STRING);
6259 MODULE_LICENSE("GPL");
6261 module_init(osd_mod_init);
6262 module_exit(osd_mod_exit);