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, 2013, 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_MDS
46 #include <linux/module.h>
48 /* LUSTRE_VERSION_CODE */
49 #include <lustre_ver.h>
50 /* prerequisite for linux/xattr.h */
51 #include <linux/types.h>
52 /* prerequisite for linux/xattr.h */
54 /* XATTR_{REPLACE,CREATE} */
55 #include <linux/xattr.h>
58 * struct OBD_{ALLOC,FREE}*()
61 #include <obd_support.h>
62 /* struct ptlrpc_thread */
63 #include <lustre_net.h>
64 #include <lustre_fid.h>
66 #include <lustre_param.h>
68 #include "osd_internal.h"
69 #include "osd_dynlocks.h"
71 /* llo_* api support */
72 #include <md_object.h>
73 #include <lustre_quota.h>
76 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
77 "ldiskfs with parallel directory operations");
79 int ldiskfs_track_declares_assert;
80 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
81 "LBUG during tracking of declares");
83 /* Slab to allocate dynlocks */
84 struct kmem_cache *dynlock_cachep;
86 static struct lu_kmem_descr ldiskfs_caches[] = {
88 .ckd_cache = &dynlock_cachep,
89 .ckd_name = "dynlock_cache",
90 .ckd_size = sizeof(struct dynlock_handle)
97 static const char dot[] = ".";
98 static const char dotdot[] = "..";
99 static const char remote_obj_dir[] = "REM_OBJ_DIR";
101 static const struct lu_object_operations osd_lu_obj_ops;
102 static const struct dt_object_operations osd_obj_ops;
103 static const struct dt_object_operations osd_obj_ea_ops;
104 static const struct dt_object_operations osd_obj_otable_it_ops;
105 static const struct dt_index_operations osd_index_iam_ops;
106 static const struct dt_index_operations osd_index_ea_ops;
108 int osd_trans_declare_op2rb[] = {
109 [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
110 [OSD_OT_PUNCH] = OSD_OT_MAX,
111 [OSD_OT_XATTR_SET] = OSD_OT_XATTR_SET,
112 [OSD_OT_CREATE] = OSD_OT_DESTROY,
113 [OSD_OT_DESTROY] = OSD_OT_CREATE,
114 [OSD_OT_REF_ADD] = OSD_OT_REF_DEL,
115 [OSD_OT_REF_DEL] = OSD_OT_REF_ADD,
116 [OSD_OT_WRITE] = OSD_OT_WRITE,
117 [OSD_OT_INSERT] = OSD_OT_DELETE,
118 [OSD_OT_DELETE] = OSD_OT_INSERT,
119 [OSD_OT_UPDATE] = OSD_OT_MAX,
120 [OSD_OT_QUOTA] = OSD_OT_MAX,
123 static int osd_has_index(const struct osd_object *obj)
125 return obj->oo_dt.do_index_ops != NULL;
128 static int osd_object_invariant(const struct lu_object *l)
130 return osd_invariant(osd_obj(l));
134 * Concurrency: doesn't matter
136 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
138 return osd_oti_get(env)->oti_r_locks > 0;
142 * Concurrency: doesn't matter
144 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
146 struct osd_thread_info *oti = osd_oti_get(env);
147 return oti->oti_w_locks > 0 && o->oo_owner == env;
151 * Concurrency: doesn't access mutable data
153 static int osd_root_get(const struct lu_env *env,
154 struct dt_device *dev, struct lu_fid *f)
156 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
161 * OSD object methods.
165 * Concurrency: no concurrent access is possible that early in object
168 static struct lu_object *osd_object_alloc(const struct lu_env *env,
169 const struct lu_object_header *hdr,
172 struct osd_object *mo;
178 l = &mo->oo_dt.do_lu;
179 dt_object_init(&mo->oo_dt, NULL, d);
180 mo->oo_dt.do_ops = &osd_obj_ea_ops;
181 l->lo_ops = &osd_lu_obj_ops;
182 init_rwsem(&mo->oo_sem);
183 init_rwsem(&mo->oo_ext_idx_sem);
184 spin_lock_init(&mo->oo_guard);
191 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
192 struct dentry *dentry, struct lustre_mdt_attrs *lma)
196 CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
197 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
198 info->oti_mdt_attrs_old, LMA_OLD_SIZE);
200 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
201 memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
203 lustre_lma_swab(lma);
204 /* Check LMA compatibility */
205 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
206 CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
207 "for fid = "DFID", ino = %lu\n",
208 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
209 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
210 PFID(&lma->lma_self_fid), inode->i_ino);
213 } else if (rc == 0) {
221 * retrieve object from backend ext fs.
223 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
224 struct osd_inode_id *id)
226 struct inode *inode = NULL;
228 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
230 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
231 id->oii_ino, PTR_ERR(inode));
232 } else if (id->oii_gen != OSD_OII_NOGEN &&
233 inode->i_generation != id->oii_gen) {
234 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
235 "i_generation = %u\n",
236 id->oii_ino, id->oii_gen, inode->i_generation);
238 inode = ERR_PTR(-ESTALE);
239 } else if (inode->i_nlink == 0) {
240 /* due to parallel readdir and unlink,
241 * we can have dead inode here. */
242 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
244 inode = ERR_PTR(-ESTALE);
245 } else if (is_bad_inode(inode)) {
246 CWARN("%.16s: bad inode: ino = %u\n",
247 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
249 inode = ERR_PTR(-ENOENT);
251 if (id->oii_gen == OSD_OII_NOGEN)
252 osd_id_gen(id, inode->i_ino, inode->i_generation);
254 /* Do not update file c/mtime in ldiskfs.
255 * NB: we don't have any lock to protect this because we don't
256 * have reference on osd_object now, but contention with
257 * another lookup + attr_set can't happen in the tiny window
258 * between if (...) and set S_NOCMTIME. */
259 if (!(inode->i_flags & S_NOCMTIME))
260 inode->i_flags |= S_NOCMTIME;
265 static struct inode *
266 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
267 struct osd_inode_id *id, struct lu_fid *fid)
269 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
273 inode = osd_iget(info, dev, id);
277 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
279 *fid = lma->lma_self_fid;
280 } else if (rc == -ENODATA) {
281 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
282 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
284 lu_igif_build(fid, inode->i_ino, inode->i_generation);
292 static struct inode *osd_iget_check(struct osd_thread_info *info,
293 struct osd_device *dev,
294 const struct lu_fid *fid,
295 struct osd_inode_id *id,
302 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
305 if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) {
306 CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
315 if (is_bad_inode(inode)) {
318 CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
326 if (id->oii_gen != OSD_OII_NOGEN &&
327 inode->i_generation != id->oii_gen) {
330 CDEBUG(D_INODE, "unmatched inode: ino = %u, "
331 "oii_gen = %u, i_generation = %u\n",
332 id->oii_ino, id->oii_gen, inode->i_generation);
340 if (inode->i_nlink == 0) {
343 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
353 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
355 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
356 /* XXX: There are some possible cases:
358 * Backup/restore caused the OI invalid.
360 * Someone unlinked the object but NOT removed
361 * the OI mapping, such as mount target device
362 * as ldiskfs, and modify something directly.
364 * Someone just removed the object between the
365 * former oi_lookup and the iget. It is normal.
366 * 4. Other failure cases.
368 * Generally, when the device is mounted, it will
369 * auto check whether the system is restored from
370 * file-level backup or not. We trust such detect
371 * to distinguish the 1st case from the 2nd case. */
373 if (!IS_ERR(inode) && inode->i_generation != 0 &&
374 inode->i_generation == id->oii_gen)
380 if (id->oii_gen == OSD_OII_NOGEN)
381 osd_id_gen(id, inode->i_ino, inode->i_generation);
383 /* Do not update file c/mtime in ldiskfs.
384 * NB: we don't have any lock to protect this because we don't
385 * have reference on osd_object now, but contention with
386 * another lookup + attr_set can't happen in the tiny window
387 * between if (...) and set S_NOCMTIME. */
388 if (!(inode->i_flags & S_NOCMTIME))
389 inode->i_flags |= S_NOCMTIME;
406 * \retval +v: new filter_fid, does not contain self-fid
407 * \retval 0: filter_fid_old, contains self-fid
408 * \retval -v: other failure cases
410 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
411 struct dentry *dentry, struct lu_fid *fid)
413 struct filter_fid_old *ff = &info->oti_ff;
414 struct ost_id *ostid = &info->oti_ostid;
417 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
418 if (rc == sizeof(*ff)) {
420 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
421 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
422 /* XXX: should use real OST index in the future. LU-3569 */
423 ostid_to_fid(fid, ostid, 0);
424 } else if (rc == sizeof(struct filter_fid)) {
426 } else if (rc >= 0) {
433 static int osd_lma_self_repair(struct osd_thread_info *info,
434 struct osd_device *osd, struct inode *inode,
435 const struct lu_fid *fid, __u32 compat)
440 LASSERT(current->journal_info == NULL);
442 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
443 osd_dto_credits_noquota[DTO_XATTR_SET]);
446 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
451 rc = osd_ea_fid_set(info, inode, fid, compat, 0);
453 CWARN("%s: cannot self repair the LMA: rc = %d\n",
455 ldiskfs_journal_stop(jh);
459 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
461 struct osd_thread_info *info = osd_oti_get(env);
462 struct osd_device *osd = osd_obj2dev(obj);
463 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
464 struct inode *inode = obj->oo_inode;
465 struct dentry *dentry = &info->oti_obj_dentry;
466 struct lu_fid *fid = NULL;
467 const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
471 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
474 CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
475 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
476 info->oti_mdt_attrs_old, LMA_OLD_SIZE);
477 if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
478 fid = &lma->lma_self_fid;
479 rc = osd_get_idif(info, inode, dentry, fid);
480 if ((rc > 0) || (rc == -ENODATA && osd->od_lma_self_repair)) {
481 /* For the given OST-object, if it has neither LMA nor
482 * FID in XATTR_NAME_FID, then the given FID (which is
483 * contained in the @obj, from client RPC for locating
484 * the OST-object) is trusted. We use it to generate
486 osd_lma_self_repair(info, osd, inode, rfid,
487 fid_is_on_ost(info, osd, fid, OI_CHECK_FLD) ?
488 LMAC_FID_ON_OST : 0);
493 if (unlikely(rc == -ENODATA))
501 lustre_lma_swab(lma);
502 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
503 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
504 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
505 "fid = "DFID", ino = %lu\n", osd_name(osd),
506 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
507 PFID(rfid), inode->i_ino);
509 } else if (!(lma->lma_compat & LMAC_NOT_IN_OI)) {
510 fid = &lma->lma_self_fid;
514 if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
515 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
516 struct ost_id *oi = &info->oti_ostid;
517 struct lu_fid *fid1 = &info->oti_fid3;
518 __u32 idx = fid_idif_ost_idx(rfid);
520 /* For old IDIF, the OST index is not part of the IDIF,
521 * Means that different OSTs may have the same IDIFs.
522 * Under such case, we need to make some compatible
523 * check to make sure to trigger OI scrub properly. */
524 if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
525 /* Given @rfid is new, LMA is old. */
526 fid_to_ostid(fid, oi);
527 ostid_to_fid(fid1, oi, idx);
528 if (lu_fid_eq(fid1, rfid)) {
529 if (osd->od_lma_self_repair)
530 osd_lma_self_repair(info, osd,
538 CDEBUG(D_INODE, "%s: FID "DFID" != self_fid "DFID"\n",
539 osd_name(osd), PFID(rfid), PFID(fid));
546 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
547 const struct lu_fid *fid,
548 const struct lu_object_conf *conf)
550 struct osd_thread_info *info;
551 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
552 struct osd_device *dev;
553 struct osd_idmap_cache *oic;
554 struct osd_inode_id *id;
556 struct osd_scrub *scrub;
557 struct scrub_file *sf;
561 bool triggered = false;
564 LINVRNT(osd_invariant(obj));
565 LASSERT(obj->oo_inode == NULL);
566 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
569 scrub = &dev->od_scrub;
570 sf = &scrub->os_file;
571 info = osd_oti_get(env);
573 oic = &info->oti_cache;
575 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
578 /* For the object is created as locking anchor, or for the object to
579 * be created on disk. No need to osd_oi_lookup() at here because FID
580 * shouldn't never be re-used, if it's really a duplicate FID from
581 * unexpected reason, we should be able to detect it later by calling
582 * do_create->osd_oi_insert(). */
583 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
584 GOTO(out, result = 0);
586 /* Search order: 1. per-thread cache. */
587 if (lu_fid_eq(fid, &oic->oic_fid) &&
588 likely(oic->oic_dev == dev)) {
594 if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
595 /* Search order: 2. OI scrub pending list. */
596 result = osd_oii_lookup(dev, fid, id);
601 /* Search order: 3. OI files. */
602 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
603 if (result == -ENOENT) {
604 if (!fid_is_norm(fid) ||
605 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
606 !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
608 GOTO(out, result = 0);
619 inode = osd_iget_check(info, dev, fid, id, in_oi);
621 result = PTR_ERR(inode);
622 if (result == -ENOENT || result == -ESTALE) {
624 fid_zero(&oic->oic_fid);
626 GOTO(out, result = -ENOENT);
627 } else if (result == -EREMCHG) {
631 fid_zero(&oic->oic_fid);
633 if (unlikely(triggered))
634 GOTO(out, result = saved);
637 if (thread_is_running(&scrub->os_thread)) {
638 result = -EINPROGRESS;
639 } else if (!dev->od_noscrub) {
640 result = osd_scrub_start(dev);
641 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
642 "for "DFID", rc = %d [1]\n",
643 osd_name(dev), PFID(fid),result);
644 if (result == 0 || result == -EALREADY)
645 result = -EINPROGRESS;
650 /* We still have chance to get the valid inode: for the
651 * object which is referenced by remote name entry, the
652 * object on the local MDT will be linked under the dir
653 * of "/REMOTE_PARENT_DIR" with its FID string as name.
655 * We do not know whether the object for the given FID
656 * is referenced by some remote name entry or not, and
657 * especially for DNE II, a multiple-linked object may
658 * have many name entries reside on many MDTs.
660 * To simplify the operation, OSD will not distinguish
661 * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
662 * only happened for the RPC from other MDT during the
663 * OI scrub, or for the client side RPC with FID only,
664 * such as FID to path, or from old connected client. */
666 result = osd_lookup_in_remote_parent(info, dev,
679 obj->oo_inode = inode;
680 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
682 result = osd_check_lma(env, obj);
685 obj->oo_inode = NULL;
686 if (result == -EREMCHG) {
688 result = osd_oi_lookup(info, dev, fid, id,
691 fid_zero(&oic->oic_fid);
702 obj->oo_compat_dot_created = 1;
703 obj->oo_compat_dotdot_created = 1;
705 if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
706 GOTO(out, result = 0);
708 LASSERT(obj->oo_hl_head == NULL);
709 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
710 if (obj->oo_hl_head == NULL) {
711 obj->oo_inode = NULL;
713 GOTO(out, result = -ENOMEM);
715 GOTO(out, result = 0);
718 LINVRNT(osd_invariant(obj));
723 * Concurrency: shouldn't matter.
725 static void osd_object_init0(struct osd_object *obj)
727 LASSERT(obj->oo_inode != NULL);
728 obj->oo_dt.do_body_ops = &osd_body_ops;
729 obj->oo_dt.do_lu.lo_header->loh_attr |=
730 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
734 * Concurrency: no concurrent access is possible that early in object
737 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
738 const struct lu_object_conf *conf)
740 struct osd_object *obj = osd_obj(l);
743 LINVRNT(osd_invariant(obj));
745 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
746 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
747 l->lo_header->loh_attr |= LOHA_EXISTS;
751 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
752 obj->oo_dt.do_body_ops = &osd_body_ops_new;
753 if (result == 0 && obj->oo_inode != NULL)
754 osd_object_init0(obj);
756 LINVRNT(osd_invariant(obj));
761 * Concurrency: no concurrent access is possible that late in object
764 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
766 struct osd_object *obj = osd_obj(l);
768 LINVRNT(osd_invariant(obj));
770 dt_object_fini(&obj->oo_dt);
771 if (obj->oo_hl_head != NULL)
772 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
777 * Concurrency: no concurrent access is possible that late in object
780 static void osd_index_fini(struct osd_object *o)
782 struct iam_container *bag;
784 if (o->oo_dir != NULL) {
785 bag = &o->oo_dir->od_container;
786 if (o->oo_inode != NULL) {
787 if (bag->ic_object == o->oo_inode)
788 iam_container_fini(bag);
790 OBD_FREE_PTR(o->oo_dir);
796 * Concurrency: no concurrent access is possible that late in object
797 * life-cycle (for all existing callers, that is. New callers have to provide
798 * their own locking.)
800 static int osd_inode_unlinked(const struct inode *inode)
802 return inode->i_nlink == 0;
806 OSD_TXN_OI_DELETE_CREDITS = 20,
807 OSD_TXN_INODE_DELETE_CREDITS = 20
814 #if OSD_THANDLE_STATS
816 * Set time when the handle is allocated
818 static void osd_th_alloced(struct osd_thandle *oth)
820 oth->oth_alloced = cfs_time_current();
824 * Set time when the handle started
826 static void osd_th_started(struct osd_thandle *oth)
828 oth->oth_started = cfs_time_current();
832 * Helper function to convert time interval to microseconds packed in
835 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
839 cfs_duration_usec(cfs_time_sub(end, start), &val);
840 return val.tv_sec * 1000000 + val.tv_usec;
844 * Check whether the we deal with this handle for too long.
846 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
847 cfs_time_t alloced, cfs_time_t started,
850 cfs_time_t now = cfs_time_current();
852 LASSERT(dev != NULL);
854 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
855 interval_to_usec(alloced, started));
856 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
857 interval_to_usec(started, closed));
858 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
859 interval_to_usec(closed, now));
861 if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
862 CWARN("transaction handle %p was open for too long: "
864 "alloced "CFS_TIME_T" ,"
865 "started "CFS_TIME_T" ,"
866 "closed "CFS_TIME_T"\n",
867 oth, now, alloced, started, closed);
868 libcfs_debug_dumpstack(NULL);
872 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
874 cfs_time_t __closed = cfs_time_current(); \
875 cfs_time_t __alloced = oth->oth_alloced; \
876 cfs_time_t __started = oth->oth_started; \
879 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
882 #else /* OSD_THANDLE_STATS */
884 #define osd_th_alloced(h) do {} while(0)
885 #define osd_th_started(h) do {} while(0)
886 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
888 #endif /* OSD_THANDLE_STATS */
891 * Concurrency: doesn't access mutable data.
893 static int osd_param_is_not_sane(const struct osd_device *dev,
894 const struct thandle *th)
896 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
898 return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers;
902 * Concurrency: shouldn't matter.
904 static void osd_trans_commit_cb(struct super_block *sb,
905 struct ldiskfs_journal_cb_entry *jcb, int error)
907 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
908 struct thandle *th = &oh->ot_super;
909 struct lu_device *lud = &th->th_dev->dd_lu_dev;
910 struct dt_txn_commit_cb *dcb, *tmp;
912 LASSERT(oh->ot_handle == NULL);
915 CERROR("transaction @0x%p commit error: %d\n", th, error);
917 dt_txn_hook_commit(th);
919 /* call per-transaction callbacks if any */
920 cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
921 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
922 "commit callback entry: magic=%x name='%s'\n",
923 dcb->dcb_magic, dcb->dcb_name);
924 cfs_list_del_init(&dcb->dcb_linkage);
925 dcb->dcb_func(NULL, th, dcb, error);
928 lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
932 lu_context_exit(&th->th_ctx);
933 lu_context_fini(&th->th_ctx);
937 static struct thandle *osd_trans_create(const struct lu_env *env,
940 struct osd_thread_info *oti = osd_oti_get(env);
941 struct osd_iobuf *iobuf = &oti->oti_iobuf;
942 struct osd_thandle *oh;
946 /* on pending IO in this thread should left from prev. request */
947 LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
949 th = ERR_PTR(-ENOMEM);
950 OBD_ALLOC_GFP(oh, sizeof *oh, __GFP_IO);
952 oh->ot_quota_trans = &oti->oti_quota_trans;
953 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
957 th->th_tags = LCT_TX_HANDLE;
959 atomic_set(&th->th_refc, 1);
960 th->th_alloc_size = sizeof(*oh);
961 oti->oti_dev = osd_dt_dev(d);
962 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
965 memset(oti->oti_declare_ops, 0,
966 sizeof(oti->oti_declare_ops));
967 memset(oti->oti_declare_ops_rb, 0,
968 sizeof(oti->oti_declare_ops_rb));
969 memset(oti->oti_declare_ops_cred, 0,
970 sizeof(oti->oti_declare_ops_cred));
971 oti->oti_rollback = false;
977 * Concurrency: shouldn't matter.
979 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
982 struct osd_thread_info *oti = osd_oti_get(env);
983 struct osd_device *dev = osd_dt_dev(d);
985 struct osd_thandle *oh;
990 LASSERT(current->journal_info == NULL);
992 oh = container_of0(th, struct osd_thandle, ot_super);
994 LASSERT(oh->ot_handle == NULL);
996 rc = dt_txn_hook_start(env, d, th);
1000 if (unlikely(osd_param_is_not_sane(dev, th))) {
1001 static unsigned long last_printed;
1002 static int last_credits;
1004 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1005 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1007 osd_journal(dev)->j_max_transaction_buffers);
1008 CWARN(" create: %u/%u, delete: %u/%u, destroy: %u/%u\n",
1009 oti->oti_declare_ops[OSD_OT_CREATE],
1010 oti->oti_declare_ops_cred[OSD_OT_CREATE],
1011 oti->oti_declare_ops[OSD_OT_DELETE],
1012 oti->oti_declare_ops_cred[OSD_OT_DELETE],
1013 oti->oti_declare_ops[OSD_OT_DESTROY],
1014 oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
1015 CWARN(" attr_set: %u/%u, xattr_set: %u/%u\n",
1016 oti->oti_declare_ops[OSD_OT_ATTR_SET],
1017 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1018 oti->oti_declare_ops[OSD_OT_XATTR_SET],
1019 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
1020 CWARN(" write: %u/%u, punch: %u/%u, quota %u/%u\n",
1021 oti->oti_declare_ops[OSD_OT_WRITE],
1022 oti->oti_declare_ops_cred[OSD_OT_WRITE],
1023 oti->oti_declare_ops[OSD_OT_PUNCH],
1024 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1025 oti->oti_declare_ops[OSD_OT_QUOTA],
1026 oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
1027 CWARN(" insert: %u/%u, delete: %u/%u\n",
1028 oti->oti_declare_ops[OSD_OT_INSERT],
1029 oti->oti_declare_ops_cred[OSD_OT_INSERT],
1030 oti->oti_declare_ops[OSD_OT_DESTROY],
1031 oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
1032 CWARN(" ref_add: %u/%u, ref_del: %u/%u\n",
1033 oti->oti_declare_ops[OSD_OT_REF_ADD],
1034 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1035 oti->oti_declare_ops[OSD_OT_REF_DEL],
1036 oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
1038 if (last_credits != oh->ot_credits &&
1039 time_after(jiffies, last_printed + 60 * HZ)) {
1040 libcfs_debug_dumpstack(NULL);
1041 last_credits = oh->ot_credits;
1042 last_printed = jiffies;
1044 /* XXX Limit the credits to 'max_transaction_buffers', and
1045 * let the underlying filesystem to catch the error if
1046 * we really need so many credits.
1048 * This should be removed when we can calculate the
1049 * credits precisely. */
1050 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
1054 * XXX temporary stuff. Some abstraction layer should
1057 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1061 LASSERT(oti->oti_txns == 0);
1062 lu_context_init(&th->th_ctx, th->th_tags);
1063 lu_context_enter(&th->th_ctx);
1065 lu_device_get(&d->dd_lu_dev);
1066 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1077 static int osd_seq_exists(const struct lu_env *env,
1078 struct osd_device *osd, obd_seq seq)
1080 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1081 struct seq_server_site *ss = osd_seq_site(osd);
1088 rc = osd_fld_lookup(env, osd, seq, range);
1091 CERROR("%s: can't lookup FLD sequence "LPX64
1092 ": rc = %d\n", osd_name(osd), seq, rc);
1096 RETURN(ss->ss_node_id == range->lsr_index);
1100 * Concurrency: shouldn't matter.
1102 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1106 struct osd_thandle *oh;
1107 struct osd_thread_info *oti = osd_oti_get(env);
1108 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1109 struct qsd_instance *qsd = oti->oti_dev->od_quota_slave;
1112 oh = container_of0(th, struct osd_thandle, ot_super);
1115 /* inform the quota slave device that the transaction is
1117 qsd_op_end(env, qsd, oh->ot_quota_trans);
1118 oh->ot_quota_trans = NULL;
1120 if (oh->ot_handle != NULL) {
1121 handle_t *hdl = oh->ot_handle;
1124 * add commit callback
1125 * notice we don't do this in osd_trans_start()
1126 * as underlying transaction can change during truncate
1128 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1131 LASSERT(oti->oti_txns == 1);
1133 rc = dt_txn_hook_stop(env, th);
1135 CERROR("Failure in transaction hook: %d\n", rc);
1137 /* hook functions might modify th_sync */
1138 hdl->h_sync = th->th_sync;
1140 oh->ot_handle = NULL;
1141 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
1142 rc = ldiskfs_journal_stop(hdl));
1144 CERROR("Failure to stop transaction: %d\n", rc);
1146 thandle_put(&oh->ot_super);
1149 /* as we want IO to journal and data IO be concurrent, we don't block
1150 * awaiting data IO completion in osd_do_bio(), instead we wait here
1151 * once transaction is submitted to the journal. all reqular requests
1152 * don't do direct IO (except read/write), thus this wait_event becomes
1155 * IMPORTANT: we have to wait till any IO submited by the thread is
1156 * completed otherwise iobuf may be corrupted by different request
1158 wait_event(iobuf->dr_wait,
1159 atomic_read(&iobuf->dr_numreqs) == 0);
1161 rc = iobuf->dr_error;
1166 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1168 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1171 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1172 LASSERT(&dcb->dcb_func != NULL);
1173 cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1179 * Called just before object is freed. Releases all resources except for
1180 * object itself (that is released by osd_object_free()).
1182 * Concurrency: no concurrent access is possible that late in object
1185 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1187 struct osd_object *obj = osd_obj(l);
1188 struct inode *inode = obj->oo_inode;
1190 LINVRNT(osd_invariant(obj));
1193 * If object is unlinked remove fid->ino mapping from object index.
1196 osd_index_fini(obj);
1197 if (inode != NULL) {
1198 struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave;
1199 qid_t uid = inode->i_uid;
1200 qid_t gid = inode->i_gid;
1203 obj->oo_inode = NULL;
1206 struct osd_thread_info *info = osd_oti_get(env);
1207 struct lquota_id_info *qi = &info->oti_qi;
1209 /* Release granted quota to master if necessary */
1210 qi->lqi_id.qid_uid = uid;
1211 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1213 qi->lqi_id.qid_uid = gid;
1214 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1220 * Concurrency: ->loo_object_release() is called under site spin-lock.
1222 static void osd_object_release(const struct lu_env *env,
1223 struct lu_object *l)
1228 * Concurrency: shouldn't matter.
1230 static int osd_object_print(const struct lu_env *env, void *cookie,
1231 lu_printer_t p, const struct lu_object *l)
1233 struct osd_object *o = osd_obj(l);
1234 struct iam_descr *d;
1236 if (o->oo_dir != NULL)
1237 d = o->oo_dir->od_container.ic_descr;
1240 return (*p)(env, cookie,
1241 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1243 o->oo_inode ? o->oo_inode->i_ino : 0UL,
1244 o->oo_inode ? o->oo_inode->i_generation : 0,
1245 d ? d->id_ops->id_name : "plain");
1248 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1251 * Concurrency: shouldn't matter.
1253 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1254 struct obd_statfs *sfs)
1256 struct osd_device *osd = osd_dt_dev(d);
1257 struct super_block *sb = osd_sb(osd);
1258 struct kstatfs *ksfs;
1261 if (unlikely(osd->od_mnt == NULL))
1262 return -EINPROGRESS;
1264 /* osd_lproc.c call this without env, allocate ksfs for that case */
1265 if (unlikely(env == NULL)) {
1266 OBD_ALLOC_PTR(ksfs);
1270 ksfs = &osd_oti_get(env)->oti_ksfs;
1273 spin_lock(&osd->od_osfs_lock);
1274 result = sb->s_op->statfs(sb->s_root, ksfs);
1275 if (likely(result == 0)) { /* N.B. statfs can't really fail */
1276 statfs_pack(sfs, ksfs);
1277 if (sb->s_flags & MS_RDONLY)
1278 sfs->os_state = OS_STATE_READONLY;
1281 spin_unlock(&osd->od_osfs_lock);
1283 if (unlikely(env == NULL))
1286 /* Reserve a small amount of space for local objects like last_rcvd,
1287 * llog, quota files, ... */
1288 if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1291 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1292 /** Take out metadata overhead for indirect blocks */
1293 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1300 * Estimate space needed for file creations. We assume the largest filename
1301 * which is 2^64 - 1, hence a filename of 20 chars.
1302 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1304 #ifdef __LDISKFS_DIR_REC_LEN
1305 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1307 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1311 * Concurrency: doesn't access mutable data.
1313 static void osd_conf_get(const struct lu_env *env,
1314 const struct dt_device *dev,
1315 struct dt_device_param *param)
1317 struct super_block *sb = osd_sb(osd_dt_dev(dev));
1320 * XXX should be taken from not-yet-existing fs abstraction layer.
1322 param->ddp_max_name_len = LDISKFS_NAME_LEN;
1323 param->ddp_max_nlink = LDISKFS_LINK_MAX;
1324 param->ddp_block_shift = sb->s_blocksize_bits;
1325 param->ddp_mount_type = LDD_MT_LDISKFS;
1326 param->ddp_maxbytes = sb->s_maxbytes;
1327 /* Overhead estimate should be fairly accurate, so we really take a tiny
1328 * error margin which also avoids fragmenting the filesystem too much */
1329 param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1330 /* inode are statically allocated, so per-inode space consumption
1331 * is the space consumed by the directory entry */
1332 param->ddp_inodespace = PER_OBJ_USAGE;
1333 /* per-fragment overhead to be used by the client code */
1334 param->ddp_grant_frag = 6 * LDISKFS_BLOCK_SIZE(sb);
1335 param->ddp_mntopts = 0;
1336 if (test_opt(sb, XATTR_USER))
1337 param->ddp_mntopts |= MNTOPT_USERXATTR;
1338 if (test_opt(sb, POSIX_ACL))
1339 param->ddp_mntopts |= MNTOPT_ACL;
1341 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1342 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1343 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE;
1346 param->ddp_max_ea_size = sb->s_blocksize;
1351 * Concurrency: shouldn't matter.
1353 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1355 CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1356 return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1360 * Start commit for OSD device.
1362 * An implementation of dt_commit_async method for OSD device.
1363 * Asychronously starts underlayng fs sync and thereby a transaction
1366 * \param env environment
1367 * \param d dt device
1369 * \see dt_device_operations
1371 static int osd_commit_async(const struct lu_env *env,
1372 struct dt_device *d)
1374 struct super_block *s = osd_sb(osd_dt_dev(d));
1377 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1378 RETURN(s->s_op->sync_fs(s, 0));
1382 * Concurrency: shouldn't matter.
1385 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1387 struct super_block *sb = osd_sb(osd_dt_dev(d));
1388 struct block_device *dev = sb->s_bdev;
1389 #ifdef HAVE_DEV_SET_RDONLY
1390 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1393 int rc = -EOPNOTSUPP;
1397 #ifdef HAVE_DEV_SET_RDONLY
1398 CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1400 if (jdev && (jdev != dev)) {
1401 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1403 dev_set_rdonly(jdev);
1405 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1406 dev_set_rdonly(dev);
1408 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1409 osd_dt_dev(d)->od_svname, (long)dev, rc);
1415 * Concurrency: serialization provided by callers.
1417 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1418 int mode, unsigned long timeout, __u32 alg,
1419 struct lustre_capa_key *keys)
1421 struct osd_device *dev = osd_dt_dev(d);
1424 dev->od_fl_capa = mode;
1425 dev->od_capa_timeout = timeout;
1426 dev->od_capa_alg = alg;
1427 dev->od_capa_keys = keys;
1432 * Note: we do not count into QUOTA here.
1433 * If we mount with --data_journal we may need more.
1435 const int osd_dto_credits_noquota[DTO_NR] = {
1438 * INDEX_EXTRA_TRANS_BLOCKS(8) +
1439 * SINGLEDATA_TRANS_BLOCKS(8)
1440 * XXX Note: maybe iam need more, since iam have more level than
1443 [DTO_INDEX_INSERT] = 16,
1444 [DTO_INDEX_DELETE] = 16,
1448 [DTO_INDEX_UPDATE] = 16,
1450 * Create a object. The same as create object in EXT3.
1451 * DATA_TRANS_BLOCKS(14) +
1452 * INDEX_EXTRA_BLOCKS(8) +
1453 * 3(inode bits, groups, GDT)
1455 [DTO_OBJECT_CREATE] = 25,
1457 * XXX: real credits to be fixed
1459 [DTO_OBJECT_DELETE] = 25,
1461 * Attr set credits (inode)
1463 [DTO_ATTR_SET_BASE] = 1,
1465 * Xattr set. The same as xattr of EXT3.
1466 * DATA_TRANS_BLOCKS(14)
1467 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1468 * are also counted in. Do not know why?
1470 [DTO_XATTR_SET] = 14,
1473 * credits for inode change during write.
1475 [DTO_WRITE_BASE] = 3,
1477 * credits for single block write.
1479 [DTO_WRITE_BLOCK] = 14,
1481 * Attr set credits for chown.
1482 * This is extra credits for setattr, and it is null without quota
1484 [DTO_ATTR_SET_CHOWN]= 0
1487 static const struct dt_device_operations osd_dt_ops = {
1488 .dt_root_get = osd_root_get,
1489 .dt_statfs = osd_statfs,
1490 .dt_trans_create = osd_trans_create,
1491 .dt_trans_start = osd_trans_start,
1492 .dt_trans_stop = osd_trans_stop,
1493 .dt_trans_cb_add = osd_trans_cb_add,
1494 .dt_conf_get = osd_conf_get,
1495 .dt_sync = osd_sync,
1497 .dt_commit_async = osd_commit_async,
1498 .dt_init_capa_ctxt = osd_init_capa_ctxt,
1501 static void osd_object_read_lock(const struct lu_env *env,
1502 struct dt_object *dt, unsigned role)
1504 struct osd_object *obj = osd_dt_obj(dt);
1505 struct osd_thread_info *oti = osd_oti_get(env);
1507 LINVRNT(osd_invariant(obj));
1509 LASSERT(obj->oo_owner != env);
1510 down_read_nested(&obj->oo_sem, role);
1512 LASSERT(obj->oo_owner == NULL);
1516 static void osd_object_write_lock(const struct lu_env *env,
1517 struct dt_object *dt, unsigned role)
1519 struct osd_object *obj = osd_dt_obj(dt);
1520 struct osd_thread_info *oti = osd_oti_get(env);
1522 LINVRNT(osd_invariant(obj));
1524 LASSERT(obj->oo_owner != env);
1525 down_write_nested(&obj->oo_sem, role);
1527 LASSERT(obj->oo_owner == NULL);
1528 obj->oo_owner = env;
1532 static void osd_object_read_unlock(const struct lu_env *env,
1533 struct dt_object *dt)
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(oti->oti_r_locks > 0);
1542 up_read(&obj->oo_sem);
1545 static void osd_object_write_unlock(const struct lu_env *env,
1546 struct dt_object *dt)
1548 struct osd_object *obj = osd_dt_obj(dt);
1549 struct osd_thread_info *oti = osd_oti_get(env);
1551 LINVRNT(osd_invariant(obj));
1553 LASSERT(obj->oo_owner == env);
1554 LASSERT(oti->oti_w_locks > 0);
1556 obj->oo_owner = NULL;
1557 up_write(&obj->oo_sem);
1560 static int osd_object_write_locked(const struct lu_env *env,
1561 struct dt_object *dt)
1563 struct osd_object *obj = osd_dt_obj(dt);
1565 LINVRNT(osd_invariant(obj));
1567 return obj->oo_owner == env;
1570 static int capa_is_sane(const struct lu_env *env,
1571 struct osd_device *dev,
1572 struct lustre_capa *capa,
1573 struct lustre_capa_key *keys)
1575 struct osd_thread_info *oti = osd_oti_get(env);
1576 struct lustre_capa *tcapa = &oti->oti_capa;
1577 struct obd_capa *oc;
1581 oc = capa_lookup(dev->od_capa_hash, capa, 0);
1583 if (capa_is_expired(oc)) {
1584 DEBUG_CAPA(D_ERROR, capa, "expired");
1591 if (capa_is_expired_sec(capa)) {
1592 DEBUG_CAPA(D_ERROR, capa, "expired");
1596 spin_lock(&capa_lock);
1597 for (i = 0; i < 2; i++) {
1598 if (keys[i].lk_keyid == capa->lc_keyid) {
1599 oti->oti_capa_key = keys[i];
1603 spin_unlock(&capa_lock);
1606 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1610 rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1614 if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1615 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1619 oc = capa_add(dev->od_capa_hash, capa);
1625 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1626 struct lustre_capa *capa, __u64 opc)
1628 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1629 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
1630 struct lu_capainfo *lci;
1633 if (!osd->od_fl_capa)
1636 if (capa == BYPASS_CAPA)
1639 lci = lu_capainfo_get(env);
1640 if (unlikely(lci == NULL))
1643 if (lci->lci_auth == LC_ID_NONE)
1647 CERROR("%s: no capability provided for FID "DFID": rc = %d\n",
1648 osd_name(osd), PFID(fid), -EACCES);
1652 if (!lu_fid_eq(fid, &capa->lc_fid)) {
1653 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1658 if (!capa_opc_supported(capa, opc)) {
1659 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1663 rc = capa_is_sane(env, osd, capa, osd->od_capa_keys);
1665 DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc);
1672 static struct timespec *osd_inode_time(const struct lu_env *env,
1673 struct inode *inode, __u64 seconds)
1675 struct osd_thread_info *oti = osd_oti_get(env);
1676 struct timespec *t = &oti->oti_time;
1678 t->tv_sec = seconds;
1680 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1685 static void osd_inode_getattr(const struct lu_env *env,
1686 struct inode *inode, struct lu_attr *attr)
1688 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1689 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1690 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1693 attr->la_atime = LTIME_S(inode->i_atime);
1694 attr->la_mtime = LTIME_S(inode->i_mtime);
1695 attr->la_ctime = LTIME_S(inode->i_ctime);
1696 attr->la_mode = inode->i_mode;
1697 attr->la_size = i_size_read(inode);
1698 attr->la_blocks = inode->i_blocks;
1699 attr->la_uid = inode->i_uid;
1700 attr->la_gid = inode->i_gid;
1701 attr->la_flags = LDISKFS_I(inode)->i_flags;
1702 attr->la_nlink = inode->i_nlink;
1703 attr->la_rdev = inode->i_rdev;
1704 attr->la_blksize = 1 << inode->i_blkbits;
1705 attr->la_blkbits = inode->i_blkbits;
1708 static int osd_attr_get(const struct lu_env *env,
1709 struct dt_object *dt,
1710 struct lu_attr *attr,
1711 struct lustre_capa *capa)
1713 struct osd_object *obj = osd_dt_obj(dt);
1715 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1716 LINVRNT(osd_invariant(obj));
1718 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1721 spin_lock(&obj->oo_guard);
1722 osd_inode_getattr(env, obj->oo_inode, attr);
1723 spin_unlock(&obj->oo_guard);
1727 static int osd_declare_attr_set(const struct lu_env *env,
1728 struct dt_object *dt,
1729 const struct lu_attr *attr,
1730 struct thandle *handle)
1732 struct osd_thandle *oh;
1733 struct osd_object *obj;
1734 struct osd_thread_info *info = osd_oti_get(env);
1735 struct lquota_id_info *qi = &info->oti_qi;
1741 LASSERT(dt != NULL);
1742 LASSERT(handle != NULL);
1744 obj = osd_dt_obj(dt);
1745 LASSERT(osd_invariant(obj));
1747 oh = container_of0(handle, struct osd_thandle, ot_super);
1748 LASSERT(oh->ot_handle == NULL);
1750 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1751 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1753 if (attr == NULL || obj->oo_inode == NULL)
1756 bspace = obj->oo_inode->i_blocks;
1757 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1758 bspace = toqb(bspace);
1760 /* Changing ownership is always preformed by super user, it should not
1763 * We still need to call the osd_declare_qid() to calculate the journal
1764 * credits for updating quota accounting files and to trigger quota
1765 * space adjustment once the operation is completed.*/
1766 if ((attr->la_valid & LA_UID) != 0 &&
1767 attr->la_uid != obj->oo_inode->i_uid) {
1768 qi->lqi_type = USRQUOTA;
1770 /* inode accounting */
1771 qi->lqi_is_blk = false;
1773 /* one more inode for the new owner ... */
1774 qi->lqi_id.qid_uid = attr->la_uid;
1776 allocated = (attr->la_uid == 0) ? true : false;
1777 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1778 if (rc == -EDQUOT || rc == -EINPROGRESS)
1783 /* and one less inode for the current uid */
1784 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1786 rc = osd_declare_qid(env, oh, qi, true, NULL);
1787 if (rc == -EDQUOT || rc == -EINPROGRESS)
1792 /* block accounting */
1793 qi->lqi_is_blk = true;
1795 /* more blocks for the new owner ... */
1796 qi->lqi_id.qid_uid = attr->la_uid;
1797 qi->lqi_space = bspace;
1798 allocated = (attr->la_uid == 0) ? true : false;
1799 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1800 if (rc == -EDQUOT || rc == -EINPROGRESS)
1805 /* and finally less blocks for the current owner */
1806 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1807 qi->lqi_space = -bspace;
1808 rc = osd_declare_qid(env, oh, qi, true, NULL);
1809 if (rc == -EDQUOT || rc == -EINPROGRESS)
1815 if (attr->la_valid & LA_GID &&
1816 attr->la_gid != obj->oo_inode->i_gid) {
1817 qi->lqi_type = GRPQUOTA;
1819 /* inode accounting */
1820 qi->lqi_is_blk = false;
1822 /* one more inode for the new group owner ... */
1823 qi->lqi_id.qid_gid = attr->la_gid;
1825 allocated = (attr->la_gid == 0) ? true : false;
1826 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1827 if (rc == -EDQUOT || rc == -EINPROGRESS)
1832 /* and one less inode for the current gid */
1833 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1835 rc = osd_declare_qid(env, oh, qi, true, NULL);
1836 if (rc == -EDQUOT || rc == -EINPROGRESS)
1841 /* block accounting */
1842 qi->lqi_is_blk = true;
1844 /* more blocks for the new owner ... */
1845 qi->lqi_id.qid_gid = attr->la_gid;
1846 qi->lqi_space = bspace;
1847 allocated = (attr->la_gid == 0) ? true : false;
1848 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1849 if (rc == -EDQUOT || rc == -EINPROGRESS)
1854 /* and finally less blocks for the current owner */
1855 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1856 qi->lqi_space = -bspace;
1857 rc = osd_declare_qid(env, oh, qi, true, NULL);
1858 if (rc == -EDQUOT || rc == -EINPROGRESS)
1867 static int osd_inode_setattr(const struct lu_env *env,
1868 struct inode *inode, const struct lu_attr *attr)
1872 bits = attr->la_valid;
1874 if (bits & LA_ATIME)
1875 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
1876 if (bits & LA_CTIME)
1877 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
1878 if (bits & LA_MTIME)
1879 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
1880 if (bits & LA_SIZE) {
1881 LDISKFS_I(inode)->i_disksize = attr->la_size;
1882 i_size_write(inode, attr->la_size);
1886 /* OSD should not change "i_blocks" which is used by quota.
1887 * "i_blocks" should be changed by ldiskfs only. */
1888 if (bits & LA_BLOCKS)
1889 inode->i_blocks = attr->la_blocks;
1892 inode->i_mode = (inode->i_mode & S_IFMT) |
1893 (attr->la_mode & ~S_IFMT);
1895 inode->i_uid = attr->la_uid;
1897 inode->i_gid = attr->la_gid;
1898 if (bits & LA_NLINK)
1899 set_nlink(inode, attr->la_nlink);
1901 inode->i_rdev = attr->la_rdev;
1903 if (bits & LA_FLAGS) {
1904 /* always keep S_NOCMTIME */
1905 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1911 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1913 if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1914 (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1919 if (attr->la_valid & LA_UID)
1920 iattr.ia_valid |= ATTR_UID;
1921 if (attr->la_valid & LA_GID)
1922 iattr.ia_valid |= ATTR_GID;
1923 iattr.ia_uid = attr->la_uid;
1924 iattr.ia_gid = attr->la_gid;
1926 rc = ll_vfs_dq_transfer(inode, &iattr);
1928 CERROR("%s: quota transfer failed: rc = %d. Is quota "
1929 "enforcement enabled on the ldiskfs filesystem?",
1930 inode->i_sb->s_id, rc);
1937 static int osd_attr_set(const struct lu_env *env,
1938 struct dt_object *dt,
1939 const struct lu_attr *attr,
1940 struct thandle *handle,
1941 struct lustre_capa *capa)
1943 struct osd_object *obj = osd_dt_obj(dt);
1944 struct inode *inode;
1947 LASSERT(handle != NULL);
1948 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1949 LASSERT(osd_invariant(obj));
1951 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1954 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1956 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1957 struct osd_thread_info *oti = osd_oti_get(env);
1958 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
1959 struct lu_fid *fid1 = &oti->oti_fid;
1960 struct osd_inode_id *id = &oti->oti_id;
1961 struct iam_path_descr *ipd;
1962 struct iam_container *bag;
1963 struct osd_thandle *oh;
1966 fid_cpu_to_be(fid1, fid0);
1967 memset(id, 1, sizeof(*id));
1968 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1969 fid0)->oi_dir.od_container;
1970 ipd = osd_idx_ipd_get(env, bag);
1971 if (unlikely(ipd == NULL))
1974 oh = container_of0(handle, struct osd_thandle, ot_super);
1975 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1976 (const struct iam_rec *)id, ipd);
1977 osd_ipd_put(env, bag, ipd);
1978 return(rc > 0 ? 0 : rc);
1981 inode = obj->oo_inode;
1982 ll_vfs_dq_init(inode);
1984 rc = osd_quota_transfer(inode, attr);
1988 spin_lock(&obj->oo_guard);
1989 rc = osd_inode_setattr(env, inode, attr);
1990 spin_unlock(&obj->oo_guard);
1993 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
1997 struct dentry *osd_child_dentry_get(const struct lu_env *env,
1998 struct osd_object *obj,
1999 const char *name, const int namelen)
2001 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2004 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2005 umode_t mode, struct dt_allocation_hint *hint,
2009 struct osd_device *osd = osd_obj2dev(obj);
2010 struct osd_thandle *oth;
2011 struct dt_object *parent = NULL;
2012 struct inode *inode;
2014 LINVRNT(osd_invariant(obj));
2015 LASSERT(obj->oo_inode == NULL);
2016 LASSERT(obj->oo_hl_head == NULL);
2018 if (S_ISDIR(mode) && ldiskfs_pdo) {
2019 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2020 if (obj->oo_hl_head == NULL)
2024 oth = container_of(th, struct osd_thandle, ot_super);
2025 LASSERT(oth->ot_handle->h_transaction != NULL);
2027 if (hint && hint->dah_parent)
2028 parent = hint->dah_parent;
2030 inode = ldiskfs_create_inode(oth->ot_handle,
2031 parent ? osd_dt_obj(parent)->oo_inode :
2032 osd_sb(osd)->s_root->d_inode,
2034 if (!IS_ERR(inode)) {
2035 /* Do not update file c/mtime in ldiskfs.
2036 * NB: don't need any lock because no contention at this
2038 inode->i_flags |= S_NOCMTIME;
2040 /* For new created object, it must be consistent,
2041 * and it is unnecessary to scrub against it. */
2042 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2043 obj->oo_inode = inode;
2046 if (obj->oo_hl_head != NULL) {
2047 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2048 obj->oo_hl_head = NULL;
2050 result = PTR_ERR(inode);
2052 LINVRNT(osd_invariant(obj));
2060 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2061 struct lu_attr *attr,
2062 struct dt_allocation_hint *hint,
2063 struct dt_object_format *dof,
2067 struct osd_thandle *oth;
2068 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2070 LASSERT(S_ISDIR(attr->la_mode));
2072 oth = container_of(th, struct osd_thandle, ot_super);
2073 LASSERT(oth->ot_handle->h_transaction != NULL);
2074 result = osd_mkfile(info, obj, mode, hint, th);
2079 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2080 struct lu_attr *attr,
2081 struct dt_allocation_hint *hint,
2082 struct dt_object_format *dof,
2086 struct osd_thandle *oth;
2087 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2089 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2091 LASSERT(S_ISREG(attr->la_mode));
2093 oth = container_of(th, struct osd_thandle, ot_super);
2094 LASSERT(oth->ot_handle->h_transaction != NULL);
2096 result = osd_mkfile(info, obj, mode, hint, th);
2098 LASSERT(obj->oo_inode != NULL);
2099 if (feat->dif_flags & DT_IND_VARKEY)
2100 result = iam_lvar_create(obj->oo_inode,
2101 feat->dif_keysize_max,
2103 feat->dif_recsize_max,
2106 result = iam_lfix_create(obj->oo_inode,
2107 feat->dif_keysize_max,
2109 feat->dif_recsize_max,
2116 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2117 struct lu_attr *attr,
2118 struct dt_allocation_hint *hint,
2119 struct dt_object_format *dof,
2122 LASSERT(S_ISREG(attr->la_mode));
2123 return osd_mkfile(info, obj, (attr->la_mode &
2124 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2127 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2128 struct lu_attr *attr,
2129 struct dt_allocation_hint *hint,
2130 struct dt_object_format *dof,
2133 LASSERT(S_ISLNK(attr->la_mode));
2134 return osd_mkfile(info, obj, (attr->la_mode &
2135 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2138 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2139 struct lu_attr *attr,
2140 struct dt_allocation_hint *hint,
2141 struct dt_object_format *dof,
2144 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2147 LINVRNT(osd_invariant(obj));
2148 LASSERT(obj->oo_inode == NULL);
2149 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2150 S_ISFIFO(mode) || S_ISSOCK(mode));
2152 result = osd_mkfile(info, obj, mode, hint, th);
2154 LASSERT(obj->oo_inode != NULL);
2156 * This inode should be marked dirty for i_rdev. Currently
2157 * that is done in the osd_attr_init().
2159 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2162 LINVRNT(osd_invariant(obj));
2166 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2168 struct dt_allocation_hint *hint,
2169 struct dt_object_format *dof,
2172 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2174 osd_obj_type_f result;
2190 result = osd_mk_index;
2201 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2202 struct dt_object *parent, struct dt_object *child,
2207 ah->dah_parent = parent;
2208 ah->dah_mode = child_mode;
2211 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2212 struct lu_attr *attr, struct dt_object_format *dof)
2214 struct inode *inode = obj->oo_inode;
2215 __u64 valid = attr->la_valid;
2218 attr->la_valid &= ~(LA_TYPE | LA_MODE);
2220 if (dof->dof_type != DFT_NODE)
2221 attr->la_valid &= ~LA_RDEV;
2222 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2223 attr->la_valid &= ~LA_ATIME;
2224 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2225 attr->la_valid &= ~LA_CTIME;
2226 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2227 attr->la_valid &= ~LA_MTIME;
2229 result = osd_quota_transfer(inode, attr);
2233 if (attr->la_valid != 0) {
2234 result = osd_inode_setattr(info->oti_env, inode, attr);
2236 * The osd_inode_setattr() should always succeed here. The
2237 * only error that could be returned is EDQUOT when we are
2238 * trying to change the UID or GID of the inode. However, this
2239 * should not happen since quota enforcement is no longer
2240 * enabled on ldiskfs (lquota takes care of it).
2242 LASSERTF(result == 0, "%d", result);
2243 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2246 attr->la_valid = valid;
2250 * Helper function for osd_object_create()
2252 * \retval 0, on success
2254 static int __osd_object_create(struct osd_thread_info *info,
2255 struct osd_object *obj, struct lu_attr *attr,
2256 struct dt_allocation_hint *hint,
2257 struct dt_object_format *dof,
2263 /* we drop umask so that permissions we pass are not affected */
2264 umask = current->fs->umask;
2265 current->fs->umask = 0;
2267 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2270 osd_attr_init(info, obj, attr, dof);
2271 osd_object_init0(obj);
2273 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2274 unlock_new_inode(obj->oo_inode);
2277 /* restore previous umask value */
2278 current->fs->umask = umask;
2284 * Helper function for osd_object_create()
2286 * \retval 0, on success
2288 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2289 const struct lu_fid *fid, struct thandle *th)
2291 struct osd_thread_info *info = osd_oti_get(env);
2292 struct osd_inode_id *id = &info->oti_id;
2293 struct osd_device *osd = osd_obj2dev(obj);
2294 struct osd_thandle *oh;
2296 LASSERT(obj->oo_inode != NULL);
2298 oh = container_of0(th, struct osd_thandle, ot_super);
2299 LASSERT(oh->ot_handle);
2301 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2302 return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2305 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2306 obd_seq seq, struct lu_seq_range *range)
2308 struct seq_server_site *ss = osd_seq_site(osd);
2310 if (fid_seq_is_idif(seq)) {
2311 fld_range_set_ost(range);
2312 range->lsr_index = idif_ost_idx(seq);
2316 if (!fid_seq_in_fldb(seq)) {
2317 fld_range_set_mdt(range);
2319 /* FIXME: If ss is NULL, it suppose not get lsr_index
2321 range->lsr_index = ss->ss_node_id;
2325 LASSERT(ss != NULL);
2326 fld_range_set_any(range);
2327 /* OSD will only do local fld lookup */
2328 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2332 * Concurrency: no external locking is necessary.
2334 static int osd_declare_object_create(const struct lu_env *env,
2335 struct dt_object *dt,
2336 struct lu_attr *attr,
2337 struct dt_allocation_hint *hint,
2338 struct dt_object_format *dof,
2339 struct thandle *handle)
2341 struct osd_thandle *oh;
2345 LASSERT(handle != NULL);
2347 oh = container_of0(handle, struct osd_thandle, ot_super);
2348 LASSERT(oh->ot_handle == NULL);
2350 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2351 osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2352 if (!fid_is_on_ost(osd_oti_get(env), osd_dt_dev(handle->th_dev),
2353 lu_object_fid(&dt->do_lu), OI_CHECK_FLD))
2354 /* Reuse idle OI block may cause additional one OI block
2356 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2357 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2359 /* If this is directory, then we expect . and .. to be inserted as
2360 * well. The one directory block always needs to be created for the
2361 * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2362 * block), there is no danger of needing a tree for the first block.
2364 if (attr && S_ISDIR(attr->la_mode)) {
2365 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2366 osd_dto_credits_noquota[DTO_WRITE_BASE]);
2367 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2373 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2374 false, false, NULL, false);
2381 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2382 struct lu_attr *attr,
2383 struct dt_allocation_hint *hint,
2384 struct dt_object_format *dof,
2387 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2388 struct osd_object *obj = osd_dt_obj(dt);
2389 struct osd_thread_info *info = osd_oti_get(env);
2394 LINVRNT(osd_invariant(obj));
2395 LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2396 LASSERT(osd_write_locked(env, obj));
2397 LASSERT(th != NULL);
2399 if (unlikely(fid_is_acct(fid)))
2400 /* Quota files can't be created from the kernel any more,
2401 * 'tune2fs -O quota' will take care of creating them */
2404 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2405 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2407 result = __osd_object_create(info, obj, attr, hint, dof, th);
2409 result = __osd_oi_insert(env, obj, fid, th);
2411 LASSERT(ergo(result == 0,
2412 dt_object_exists(dt) && !dt_object_remote(dt)));
2414 LASSERT(osd_invariant(obj));
2419 * Called to destroy on-disk representation of the object
2421 * Concurrency: must be locked
2423 static int osd_declare_object_destroy(const struct lu_env *env,
2424 struct dt_object *dt,
2427 struct osd_object *obj = osd_dt_obj(dt);
2428 struct inode *inode = obj->oo_inode;
2429 struct osd_thandle *oh;
2433 oh = container_of0(th, struct osd_thandle, ot_super);
2434 LASSERT(oh->ot_handle == NULL);
2437 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2438 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2439 /* Recycle idle OI leaf may cause additional three OI blocks
2441 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2442 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2443 /* one less inode */
2444 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
2445 false, true, NULL, false);
2448 /* data to be truncated */
2449 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
2450 true, true, NULL, false);
2454 static int osd_object_destroy(const struct lu_env *env,
2455 struct dt_object *dt,
2458 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2459 struct osd_object *obj = osd_dt_obj(dt);
2460 struct inode *inode = obj->oo_inode;
2461 struct osd_device *osd = osd_obj2dev(obj);
2462 struct osd_thandle *oh;
2466 oh = container_of0(th, struct osd_thandle, ot_super);
2467 LASSERT(oh->ot_handle);
2469 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2471 if (unlikely(fid_is_acct(fid)))
2474 if (S_ISDIR(inode->i_mode)) {
2475 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2476 inode->i_nlink == 2);
2477 /* it will check/delete the inode from remote parent,
2478 * how to optimize it? unlink performance impaction XXX */
2479 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2480 if (result != 0 && result != -ENOENT) {
2481 CERROR("%s: delete inode "DFID": rc = %d\n",
2482 osd_name(osd), PFID(fid), result);
2484 spin_lock(&obj->oo_guard);
2486 spin_unlock(&obj->oo_guard);
2487 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2490 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2492 result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2495 /* XXX: add to ext3 orphan list */
2496 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2498 /* not needed in the cache anymore */
2499 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2505 * Put the fid into lustre_mdt_attrs, and then place the structure
2506 * inode's ea. This fid should not be altered during the life time
2509 * \retval +ve, on success
2510 * \retval -ve, on error
2512 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2514 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2515 const struct lu_fid *fid, __u32 compat, __u32 incompat)
2517 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2521 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2524 lustre_lma_init(lma, fid, compat, incompat);
2525 lustre_lma_swab(lma);
2527 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2529 /* LMA may already exist, but we need to check that all the
2530 * desired compat/incompat flags have been added. */
2531 if (unlikely(rc == -EEXIST)) {
2532 if (compat == 0 && incompat == 0)
2535 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2536 XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2541 lustre_lma_swab(lma);
2542 if (!(~lma->lma_compat & compat) &&
2543 !(~lma->lma_incompat & incompat))
2546 lma->lma_compat |= compat;
2547 lma->lma_incompat |= incompat;
2548 lustre_lma_swab(lma);
2549 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2550 sizeof(*lma), XATTR_REPLACE);
2557 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2558 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2559 * To have compatilibility with 1.8 ldiskfs driver we need to have
2560 * magic number at start of fid data.
2561 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2564 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2565 const struct dt_rec *fid)
2567 if (!fid_is_namespace_visible((const struct lu_fid *)fid) ||
2568 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2569 param->edp_magic = 0;
2573 param->edp_magic = LDISKFS_LUFID_MAGIC;
2574 param->edp_len = sizeof(struct lu_fid) + 1;
2575 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2579 * Try to read the fid from inode ea into dt_rec.
2581 * \param fid object fid.
2583 * \retval 0 on success
2585 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2586 __u32 ino, struct lu_fid *fid,
2587 struct osd_inode_id *id)
2589 struct osd_thread_info *info = osd_oti_get(env);
2590 struct inode *inode;
2593 osd_id_gen(id, ino, OSD_OII_NOGEN);
2594 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2596 RETURN(PTR_ERR(inode));
2602 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2604 struct inode *parent_dir,
2605 const struct dt_rec *dot_fid,
2606 const struct dt_rec *dot_dot_fid,
2607 struct osd_thandle *oth)
2609 struct ldiskfs_dentry_param *dot_ldp;
2610 struct ldiskfs_dentry_param *dot_dot_ldp;
2612 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2613 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2615 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2616 dot_ldp->edp_magic = 0;
2617 return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2618 dir, dot_ldp, dot_dot_ldp);
2622 * Create an local agent inode for remote entry
2624 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2625 struct osd_device *osd,
2626 struct osd_object *pobj,
2627 const struct lu_fid *fid,
2630 struct osd_thread_info *info = osd_oti_get(env);
2631 struct inode *local;
2632 struct osd_thandle *oh;
2637 oh = container_of(th, struct osd_thandle, ot_super);
2638 LASSERT(oh->ot_handle->h_transaction != NULL);
2640 /* FIXME: Insert index api needs to know the mode of
2641 * the remote object. Just use S_IFDIR for now */
2642 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR);
2643 if (IS_ERR(local)) {
2644 CERROR("%s: create local error %d\n", osd_name(osd),
2645 (int)PTR_ERR(local));
2649 /* Set special LMA flag for local agent inode */
2650 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2652 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2653 osd_name(osd), PFID(fid), rc);
2654 RETURN(ERR_PTR(rc));
2657 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2658 (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
2659 (const struct dt_rec *)fid, oh);
2661 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2662 osd_name(osd), PFID(fid), rc);
2663 RETURN(ERR_PTR(rc));
2670 * Delete local agent inode for remote entry
2672 static int osd_delete_local_agent_inode(const struct lu_env *env,
2673 struct osd_device *osd,
2674 const struct lu_fid *fid,
2675 __u32 ino, struct osd_thandle *oh)
2677 struct osd_thread_info *oti = osd_oti_get(env);
2678 struct osd_inode_id *id = &oti->oti_id;
2679 struct inode *inode;
2682 id->oii_ino = le32_to_cpu(ino);
2683 id->oii_gen = OSD_OII_NOGEN;
2684 inode = osd_iget(oti, osd, id);
2685 if (IS_ERR(inode)) {
2686 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2687 PFID(fid), id->oii_ino, id->oii_gen);
2688 RETURN(PTR_ERR(inode));
2692 mark_inode_dirty(inode);
2693 CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2694 osd_name(osd), PFID(fid), inode->i_ino);
2700 * OSD layer object create function for interoperability mode (b11826).
2701 * This is mostly similar to osd_object_create(). Only difference being, fid is
2702 * inserted into inode ea here.
2704 * \retval 0, on success
2705 * \retval -ve, on error
2707 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2708 struct lu_attr *attr,
2709 struct dt_allocation_hint *hint,
2710 struct dt_object_format *dof,
2713 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2714 struct osd_object *obj = osd_dt_obj(dt);
2715 struct osd_thread_info *info = osd_oti_get(env);
2720 LASSERT(osd_invariant(obj));
2721 LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2722 LASSERT(osd_write_locked(env, obj));
2723 LASSERT(th != NULL);
2725 if (unlikely(fid_is_acct(fid)))
2726 /* Quota files can't be created from the kernel any more,
2727 * 'tune2fs -O quota' will take care of creating them */
2730 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2731 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2733 result = __osd_object_create(info, obj, attr, hint, dof, th);
2735 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2736 fid_is_on_ost(info, osd_obj2dev(obj),
2737 fid, OI_CHECK_FLD) ?
2738 LMAC_FID_ON_OST : 0, 0);
2741 result = __osd_oi_insert(env, obj, fid, th);
2743 LASSERT(ergo(result == 0,
2744 dt_object_exists(dt) && !dt_object_remote(dt)));
2745 LINVRNT(osd_invariant(obj));
2749 static int osd_declare_object_ref_add(const struct lu_env *env,
2750 struct dt_object *dt,
2751 struct thandle *handle)
2753 struct osd_thandle *oh;
2755 /* it's possible that object doesn't exist yet */
2756 LASSERT(handle != NULL);
2758 oh = container_of0(handle, struct osd_thandle, ot_super);
2759 LASSERT(oh->ot_handle == NULL);
2761 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2762 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2768 * Concurrency: @dt is write locked.
2770 static int osd_object_ref_add(const struct lu_env *env,
2771 struct dt_object *dt, struct thandle *th)
2773 struct osd_object *obj = osd_dt_obj(dt);
2774 struct inode *inode = obj->oo_inode;
2775 struct osd_thandle *oh;
2778 LINVRNT(osd_invariant(obj));
2779 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2780 LASSERT(osd_write_locked(env, obj));
2781 LASSERT(th != NULL);
2783 oh = container_of0(th, struct osd_thandle, ot_super);
2784 LASSERT(oh->ot_handle != NULL);
2786 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2789 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2790 * (65000) subdirectories by storing "1" in i_nlink if the link count
2791 * would otherwise overflow. Directory tranversal tools understand
2792 * that (st_nlink == 1) indicates that the filesystem dose not track
2793 * hard links count on the directory, and will not abort subdirectory
2794 * scanning early once (st_nlink - 2) subdirs have been found.
2796 * This also has to properly handle the case of inodes with nlink == 0
2797 * in case they are being linked into the PENDING directory
2799 spin_lock(&obj->oo_guard);
2800 ldiskfs_inc_count(oh->ot_handle, inode);
2801 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2802 spin_unlock(&obj->oo_guard);
2804 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2805 LINVRNT(osd_invariant(obj));
2810 static int osd_declare_object_ref_del(const struct lu_env *env,
2811 struct dt_object *dt,
2812 struct thandle *handle)
2814 struct osd_thandle *oh;
2816 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2817 LASSERT(handle != NULL);
2819 oh = container_of0(handle, struct osd_thandle, ot_super);
2820 LASSERT(oh->ot_handle == NULL);
2822 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2823 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2829 * Concurrency: @dt is write locked.
2831 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2834 struct osd_object *obj = osd_dt_obj(dt);
2835 struct inode *inode = obj->oo_inode;
2836 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
2837 struct osd_thandle *oh;
2839 LINVRNT(osd_invariant(obj));
2840 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2841 LASSERT(osd_write_locked(env, obj));
2842 LASSERT(th != NULL);
2844 oh = container_of0(th, struct osd_thandle, ot_super);
2845 LASSERT(oh->ot_handle != NULL);
2847 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2849 spin_lock(&obj->oo_guard);
2850 /* That can be result of upgrade from old Lustre version and
2851 * applied only to local files. Just skip this ref_del call.
2852 * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2853 if (inode->i_nlink == 0) {
2854 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2855 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2856 ", maybe an upgraded file? (LU-3915)\n",
2857 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2858 spin_unlock(&obj->oo_guard);
2862 ldiskfs_dec_count(oh->ot_handle, inode);
2863 spin_unlock(&obj->oo_guard);
2865 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2866 LINVRNT(osd_invariant(obj));
2872 * Get the 64-bit version for an inode.
2874 static int osd_object_version_get(const struct lu_env *env,
2875 struct dt_object *dt, dt_obj_version_t *ver)
2877 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2879 CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2880 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2881 *ver = LDISKFS_I(inode)->i_fs_version;
2886 * Concurrency: @dt is read locked.
2888 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2889 struct lu_buf *buf, const char *name,
2890 struct lustre_capa *capa)
2892 struct osd_object *obj = osd_dt_obj(dt);
2893 struct inode *inode = obj->oo_inode;
2894 struct osd_thread_info *info = osd_oti_get(env);
2895 struct dentry *dentry = &info->oti_obj_dentry;
2897 /* version get is not real XATTR but uses xattr API */
2898 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2899 /* for version we are just using xattr API but change inode
2901 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2902 osd_object_version_get(env, dt, buf->lb_buf);
2903 return sizeof(dt_obj_version_t);
2906 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2907 LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2909 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2912 return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2916 static int osd_declare_xattr_set(const struct lu_env *env,
2917 struct dt_object *dt,
2918 const struct lu_buf *buf, const char *name,
2919 int fl, struct thandle *handle)
2921 struct osd_thandle *oh;
2923 LASSERT(handle != NULL);
2925 oh = container_of0(handle, struct osd_thandle, ot_super);
2926 LASSERT(oh->ot_handle == NULL);
2928 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2929 strcmp(name, XATTR_NAME_VERSION) == 0 ?
2930 osd_dto_credits_noquota[DTO_ATTR_SET_BASE] :
2931 osd_dto_credits_noquota[DTO_XATTR_SET]);
2937 * Set the 64-bit version for object
2939 static void osd_object_version_set(const struct lu_env *env,
2940 struct dt_object *dt,
2941 dt_obj_version_t *new_version)
2943 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2945 CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2946 *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2948 LDISKFS_I(inode)->i_fs_version = *new_version;
2949 /** Version is set after all inode operations are finished,
2950 * so we should mark it dirty here */
2951 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2955 * Concurrency: @dt is write locked.
2957 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2958 const struct lu_buf *buf, const char *name, int fl,
2959 struct thandle *handle, struct lustre_capa *capa)
2961 struct osd_object *obj = osd_dt_obj(dt);
2962 struct inode *inode = obj->oo_inode;
2963 struct osd_thread_info *info = osd_oti_get(env);
2967 LASSERT(handle != NULL);
2969 /* version set is not real XATTR */
2970 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2971 /* for version we are just using xattr API but change inode
2973 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2974 osd_object_version_set(env, dt, buf->lb_buf);
2975 return sizeof(dt_obj_version_t);
2978 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2981 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n",
2982 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
2984 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2985 if (fl & LU_XATTR_REPLACE)
2986 fs_flags |= XATTR_REPLACE;
2988 if (fl & LU_XATTR_CREATE)
2989 fs_flags |= XATTR_CREATE;
2991 return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
2996 * Concurrency: @dt is read locked.
2998 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2999 struct lu_buf *buf, struct lustre_capa *capa)
3001 struct osd_object *obj = osd_dt_obj(dt);
3002 struct inode *inode = obj->oo_inode;
3003 struct osd_thread_info *info = osd_oti_get(env);
3004 struct dentry *dentry = &info->oti_obj_dentry;
3006 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3007 LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
3008 LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
3010 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3013 dentry->d_inode = inode;
3014 dentry->d_sb = inode->i_sb;
3015 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3018 static int osd_declare_xattr_del(const struct lu_env *env,
3019 struct dt_object *dt, const char *name,
3020 struct thandle *handle)
3022 struct osd_thandle *oh;
3024 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3025 LASSERT(handle != NULL);
3027 oh = container_of0(handle, struct osd_thandle, ot_super);
3028 LASSERT(oh->ot_handle == NULL);
3030 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3031 osd_dto_credits_noquota[DTO_XATTR_SET]);
3037 * Concurrency: @dt is write locked.
3039 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3040 const char *name, struct thandle *handle,
3041 struct lustre_capa *capa)
3043 struct osd_object *obj = osd_dt_obj(dt);
3044 struct inode *inode = obj->oo_inode;
3045 struct osd_thread_info *info = osd_oti_get(env);
3046 struct dentry *dentry = &info->oti_obj_dentry;
3049 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3050 LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
3051 LASSERT(handle != NULL);
3053 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3056 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3058 ll_vfs_dq_init(inode);
3059 dentry->d_inode = inode;
3060 dentry->d_sb = inode->i_sb;
3061 rc = inode->i_op->removexattr(dentry, name);
3065 static struct obd_capa *osd_capa_get(const struct lu_env *env,
3066 struct dt_object *dt,
3067 struct lustre_capa *old, __u64 opc)
3069 struct osd_thread_info *info = osd_oti_get(env);
3070 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3071 struct osd_object *obj = osd_dt_obj(dt);
3072 struct osd_device *osd = osd_obj2dev(obj);
3073 struct lustre_capa_key *key = &info->oti_capa_key;
3074 struct lustre_capa *capa = &info->oti_capa;
3075 struct obd_capa *oc;
3076 struct lu_capainfo *lci;
3080 if (!osd->od_fl_capa)
3081 RETURN(ERR_PTR(-ENOENT));
3083 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3084 LINVRNT(osd_invariant(obj));
3086 /* renewal sanity check */
3087 if (old && osd_object_auth(env, dt, old, opc))
3088 RETURN(ERR_PTR(-EACCES));
3090 lci = lu_capainfo_get(env);
3091 if (unlikely(lci == NULL))
3092 RETURN(ERR_PTR(-ENOENT));
3094 switch (lci->lci_auth) {
3098 capa->lc_uid = obj->oo_inode->i_uid;
3099 capa->lc_gid = obj->oo_inode->i_gid;
3100 capa->lc_flags = LC_ID_PLAIN;
3102 case LC_ID_CONVERT: {
3105 s[0] = obj->oo_inode->i_uid;
3106 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
3107 s[2] = obj->oo_inode->i_gid;
3108 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
3109 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
3111 RETURN(ERR_PTR(rc));
3113 capa->lc_uid = ((__u64)d[1] << 32) | d[0];
3114 capa->lc_gid = ((__u64)d[3] << 32) | d[2];
3115 capa->lc_flags = LC_ID_CONVERT;
3119 RETURN(ERR_PTR(-EINVAL));
3122 capa->lc_fid = *fid;
3124 capa->lc_flags |= osd->od_capa_alg << 24;
3125 capa->lc_timeout = osd->od_capa_timeout;
3126 capa->lc_expiry = 0;
3128 oc = capa_lookup(osd->od_capa_hash, capa, 1);
3130 LASSERT(!capa_is_expired(oc));
3134 spin_lock(&capa_lock);
3135 *key = osd->od_capa_keys[1];
3136 spin_unlock(&capa_lock);
3138 capa->lc_keyid = key->lk_keyid;
3139 capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout;
3141 rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
3143 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
3144 RETURN(ERR_PTR(rc));
3147 oc = capa_add(osd->od_capa_hash, capa);
3151 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
3153 struct osd_object *obj = osd_dt_obj(dt);
3154 struct inode *inode = obj->oo_inode;
3155 struct osd_thread_info *info = osd_oti_get(env);
3156 struct dentry *dentry = &info->oti_obj_dentry;
3157 struct file *file = &info->oti_file;
3162 dentry->d_inode = inode;
3163 dentry->d_sb = inode->i_sb;
3164 file->f_dentry = dentry;
3165 file->f_mapping = inode->i_mapping;
3166 file->f_op = inode->i_fop;
3167 set_file_inode(file, inode);
3168 #ifndef HAVE_FILE_FSYNC_4ARGS
3169 mutex_lock(&inode->i_mutex);
3171 rc = do_fsync(file, 0);
3172 #ifndef HAVE_FILE_FSYNC_4ARGS
3173 mutex_unlock(&inode->i_mutex);
3178 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
3181 struct osd_object *obj = osd_dt_obj(dt);
3184 *data = (void *)obj->oo_inode;
3192 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3193 const struct dt_index_features *feat)
3195 struct iam_descr *descr;
3197 if (osd_object_is_root(o))
3198 return feat == &dt_directory_features;
3200 LASSERT(o->oo_dir != NULL);
3202 descr = o->oo_dir->od_container.ic_descr;
3203 if (feat == &dt_directory_features) {
3204 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3210 feat->dif_keysize_min <= descr->id_key_size &&
3211 descr->id_key_size <= feat->dif_keysize_max &&
3212 feat->dif_recsize_min <= descr->id_rec_size &&
3213 descr->id_rec_size <= feat->dif_recsize_max &&
3214 !(feat->dif_flags & (DT_IND_VARKEY |
3215 DT_IND_VARREC | DT_IND_NONUNQ)) &&
3216 ergo(feat->dif_flags & DT_IND_UPDATE,
3217 1 /* XXX check that object (and file system) is
3222 static int osd_iam_container_init(const struct lu_env *env,
3223 struct osd_object *obj,
3224 struct osd_directory *dir)
3226 struct iam_container *bag = &dir->od_container;
3229 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3233 result = iam_container_setup(bag);
3235 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3237 iam_container_fini(bag);
3244 * Concurrency: no external locking is necessary.
3246 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3247 const struct dt_index_features *feat)
3251 struct osd_object *obj = osd_dt_obj(dt);
3253 LINVRNT(osd_invariant(obj));
3255 if (osd_object_is_root(obj)) {
3256 dt->do_index_ops = &osd_index_ea_ops;
3258 } else if (feat == &dt_directory_features) {
3259 dt->do_index_ops = &osd_index_ea_ops;
3260 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3265 } else if (unlikely(feat == &dt_otable_features)) {
3266 dt->do_index_ops = &osd_otable_ops;
3268 } else if (unlikely(feat == &dt_acct_features)) {
3269 dt->do_index_ops = &osd_acct_index_ops;
3272 } else if (!osd_has_index(obj)) {
3273 struct osd_directory *dir;
3278 spin_lock(&obj->oo_guard);
3279 if (obj->oo_dir == NULL)
3283 * Concurrent thread allocated container data.
3286 spin_unlock(&obj->oo_guard);
3288 * Now, that we have container data, serialize its
3291 down_write(&obj->oo_ext_idx_sem);
3293 * recheck under lock.
3295 if (!osd_has_index(obj))
3296 result = osd_iam_container_init(env, obj, dir);
3299 up_write(&obj->oo_ext_idx_sem);
3307 if (result == 0 && skip_iam == 0) {
3308 if (!osd_iam_index_probe(env, obj, feat))
3311 LINVRNT(osd_invariant(obj));
3313 if (result == 0 && is_quota_glb_feat(feat) &&
3314 fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3315 result = osd_quota_migration(env, dt, feat);
3320 static int osd_otable_it_attr_get(const struct lu_env *env,
3321 struct dt_object *dt,
3322 struct lu_attr *attr,
3323 struct lustre_capa *capa)
3329 static const struct dt_object_operations osd_obj_ops = {
3330 .do_read_lock = osd_object_read_lock,
3331 .do_write_lock = osd_object_write_lock,
3332 .do_read_unlock = osd_object_read_unlock,
3333 .do_write_unlock = osd_object_write_unlock,
3334 .do_write_locked = osd_object_write_locked,
3335 .do_attr_get = osd_attr_get,
3336 .do_declare_attr_set = osd_declare_attr_set,
3337 .do_attr_set = osd_attr_set,
3338 .do_ah_init = osd_ah_init,
3339 .do_declare_create = osd_declare_object_create,
3340 .do_create = osd_object_create,
3341 .do_declare_destroy = osd_declare_object_destroy,
3342 .do_destroy = osd_object_destroy,
3343 .do_index_try = osd_index_try,
3344 .do_declare_ref_add = osd_declare_object_ref_add,
3345 .do_ref_add = osd_object_ref_add,
3346 .do_declare_ref_del = osd_declare_object_ref_del,
3347 .do_ref_del = osd_object_ref_del,
3348 .do_xattr_get = osd_xattr_get,
3349 .do_declare_xattr_set = osd_declare_xattr_set,
3350 .do_xattr_set = osd_xattr_set,
3351 .do_declare_xattr_del = osd_declare_xattr_del,
3352 .do_xattr_del = osd_xattr_del,
3353 .do_xattr_list = osd_xattr_list,
3354 .do_capa_get = osd_capa_get,
3355 .do_object_sync = osd_object_sync,
3356 .do_data_get = osd_data_get,
3360 * dt_object_operations for interoperability mode
3361 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3363 static const struct dt_object_operations osd_obj_ea_ops = {
3364 .do_read_lock = osd_object_read_lock,
3365 .do_write_lock = osd_object_write_lock,
3366 .do_read_unlock = osd_object_read_unlock,
3367 .do_write_unlock = osd_object_write_unlock,
3368 .do_write_locked = osd_object_write_locked,
3369 .do_attr_get = osd_attr_get,
3370 .do_declare_attr_set = osd_declare_attr_set,
3371 .do_attr_set = osd_attr_set,
3372 .do_ah_init = osd_ah_init,
3373 .do_declare_create = osd_declare_object_create,
3374 .do_create = osd_object_ea_create,
3375 .do_declare_destroy = osd_declare_object_destroy,
3376 .do_destroy = osd_object_destroy,
3377 .do_index_try = osd_index_try,
3378 .do_declare_ref_add = osd_declare_object_ref_add,
3379 .do_ref_add = osd_object_ref_add,
3380 .do_declare_ref_del = osd_declare_object_ref_del,
3381 .do_ref_del = osd_object_ref_del,
3382 .do_xattr_get = osd_xattr_get,
3383 .do_declare_xattr_set = osd_declare_xattr_set,
3384 .do_xattr_set = osd_xattr_set,
3385 .do_declare_xattr_del = osd_declare_xattr_del,
3386 .do_xattr_del = osd_xattr_del,
3387 .do_xattr_list = osd_xattr_list,
3388 .do_capa_get = osd_capa_get,
3389 .do_object_sync = osd_object_sync,
3390 .do_data_get = osd_data_get,
3393 static const struct dt_object_operations osd_obj_otable_it_ops = {
3394 .do_attr_get = osd_otable_it_attr_get,
3395 .do_index_try = osd_index_try,
3398 static int osd_index_declare_iam_delete(const struct lu_env *env,
3399 struct dt_object *dt,
3400 const struct dt_key *key,
3401 struct thandle *handle)
3403 struct osd_thandle *oh;
3405 oh = container_of0(handle, struct osd_thandle, ot_super);
3406 LASSERT(oh->ot_handle == NULL);
3408 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3409 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3415 * delete a (key, value) pair from index \a dt specified by \a key
3417 * \param dt osd index object
3418 * \param key key for index
3419 * \param rec record reference
3420 * \param handle transaction handler
3423 * \retval -ve failure
3426 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3427 const struct dt_key *key,
3428 struct thandle *handle,
3429 struct lustre_capa *capa)
3431 struct osd_thread_info *oti = osd_oti_get(env);
3432 struct osd_object *obj = osd_dt_obj(dt);
3433 struct osd_thandle *oh;
3434 struct iam_path_descr *ipd;
3435 struct iam_container *bag = &obj->oo_dir->od_container;
3440 LINVRNT(osd_invariant(obj));
3441 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3442 LASSERT(bag->ic_object == obj->oo_inode);
3443 LASSERT(handle != NULL);
3445 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3448 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3450 ipd = osd_idx_ipd_get(env, bag);
3451 if (unlikely(ipd == NULL))
3454 oh = container_of0(handle, struct osd_thandle, ot_super);
3455 LASSERT(oh->ot_handle != NULL);
3456 LASSERT(oh->ot_handle->h_transaction != NULL);
3458 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3459 /* swab quota uid/gid provided by caller */
3460 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3461 key = (const struct dt_key *)&oti->oti_quota_id;
3464 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3465 osd_ipd_put(env, bag, ipd);
3466 LINVRNT(osd_invariant(obj));
3470 static int osd_index_declare_ea_delete(const struct lu_env *env,
3471 struct dt_object *dt,
3472 const struct dt_key *key,
3473 struct thandle *handle)
3475 struct osd_thandle *oh;
3476 struct inode *inode;
3480 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3481 LASSERT(handle != NULL);
3483 oh = container_of0(handle, struct osd_thandle, ot_super);
3484 LASSERT(oh->ot_handle == NULL);
3486 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3487 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3489 inode = osd_dt_obj(dt)->oo_inode;
3492 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3493 true, true, NULL, false);
3497 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3500 struct osd_fid_pack *rec;
3503 if (de->file_type & LDISKFS_DIRENT_LUFID) {
3504 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3505 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3510 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3515 /* FID seqs not in FLDB, must be local seq */
3516 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3519 if (osd_seq_exists(env, osd, fid_seq(fid)))
3526 * Index delete function for interoperability mode (b11826).
3527 * It will remove the directory entry added by osd_index_ea_insert().
3528 * This entry is needed to maintain name->fid mapping.
3530 * \param key, key i.e. file entry to be deleted
3532 * \retval 0, on success
3533 * \retval -ve, on error
3535 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3536 const struct dt_key *key,
3537 struct thandle *handle,
3538 struct lustre_capa *capa)
3540 struct osd_object *obj = osd_dt_obj(dt);
3541 struct inode *dir = obj->oo_inode;
3542 struct dentry *dentry;
3543 struct osd_thandle *oh;
3544 struct ldiskfs_dir_entry_2 *de = NULL;
3545 struct buffer_head *bh;
3546 struct htree_lock *hlock = NULL;
3547 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
3548 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3552 LINVRNT(osd_invariant(obj));
3553 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3554 LASSERT(handle != NULL);
3556 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3558 oh = container_of(handle, struct osd_thandle, ot_super);
3559 LASSERT(oh->ot_handle != NULL);
3560 LASSERT(oh->ot_handle->h_transaction != NULL);
3562 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3565 ll_vfs_dq_init(dir);
3566 dentry = osd_child_dentry_get(env, obj,
3567 (char *)key, strlen((char *)key));
3569 if (obj->oo_hl_head != NULL) {
3570 hlock = osd_oti_get(env)->oti_hlock;
3571 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3572 dir, LDISKFS_HLOCK_DEL);
3574 down_write(&obj->oo_ext_idx_sem);
3577 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3581 /* If this is not the ".." entry, it might be a remote DNE
3582 * entry and we need to check if the FID is for a remote
3583 * MDT. If the FID is not in the directory entry (e.g.
3584 * upgraded 1.8 filesystem without dirdata enabled) then
3585 * we need to get the FID from the LMA. For a remote directory
3586 * there HAS to be an LMA, it cannot be an IGIF inode in this
3589 * Delete the entry before the agent inode in order to
3590 * simplify error handling. At worst an error after deleting
3591 * the entry first might leak the agent inode afterward. The
3592 * reverse would need filesystem abort in case of error deleting
3593 * the entry after the agent had been removed, or leave a
3594 * dangling entry pointing at a random inode. */
3595 if (strcmp((char *)key, dotdot) != 0) {
3596 LASSERT(de != NULL);
3597 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3598 /* If Fid is not in dentry, try to get it from LMA */
3599 if (rc == -ENODATA) {
3600 struct osd_inode_id *id;
3601 struct inode *inode;
3603 /* Before trying to get fid from the inode,
3604 * check whether the inode is valid.
3606 * If the inode has been deleted, do not go
3607 * ahead to do osd_ea_fid_get, which will set
3608 * the inode to bad inode, which might cause
3609 * the inode to be deleted uncorrectly */
3610 inode = ldiskfs_iget(osd_sb(osd),
3611 le32_to_cpu(de->inode));
3612 if (IS_ERR(inode)) {
3613 CDEBUG(D_INODE, "%s: "DFID"get inode"
3614 "error.\n", osd_name(osd),
3616 rc = PTR_ERR(inode);
3618 if (likely(inode->i_nlink != 0)) {
3619 id = &osd_oti_get(env)->oti_id;
3620 rc = osd_ea_fid_get(env, obj,
3621 le32_to_cpu(de->inode),
3624 CDEBUG(D_INFO, "%s: %u "DFID
3627 le32_to_cpu(de->inode),
3635 unlikely(osd_remote_fid(env, osd, fid)))
3636 /* Need to delete agent inode */
3637 ino = le32_to_cpu(de->inode);
3639 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3641 if (rc == 0 && unlikely(ino != 0)) {
3642 rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3645 CERROR("%s: del local inode "DFID": rc = %d\n",
3646 osd_name(osd), PFID(fid), rc);
3652 ldiskfs_htree_unlock(hlock);
3654 up_write(&obj->oo_ext_idx_sem);
3659 /* For inode on the remote MDT, .. will point to
3660 * /Agent directory, Check whether it needs to delete
3661 * from agent directory */
3662 if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3663 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3665 if (rc != 0 && rc != -ENOENT) {
3666 CERROR("%s: delete agent inode "DFID": rc = %d\n",
3667 osd_name(osd), PFID(fid), rc);
3677 LASSERT(osd_invariant(obj));
3682 * Lookup index for \a key and copy record to \a rec.
3684 * \param dt osd index object
3685 * \param key key for index
3686 * \param rec record reference
3688 * \retval +ve success : exact mach
3689 * \retval 0 return record with key not greater than \a key
3690 * \retval -ve failure
3692 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3693 struct dt_rec *rec, const struct dt_key *key,
3694 struct lustre_capa *capa)
3696 struct osd_object *obj = osd_dt_obj(dt);
3697 struct iam_path_descr *ipd;
3698 struct iam_container *bag = &obj->oo_dir->od_container;
3699 struct osd_thread_info *oti = osd_oti_get(env);
3700 struct iam_iterator *it = &oti->oti_idx_it;
3701 struct iam_rec *iam_rec;
3706 LASSERT(osd_invariant(obj));
3707 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3708 LASSERT(bag->ic_object == obj->oo_inode);
3710 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3713 ipd = osd_idx_ipd_get(env, bag);
3717 /* got ipd now we can start iterator. */
3718 iam_it_init(it, bag, 0, ipd);
3720 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3721 /* swab quota uid/gid provided by caller */
3722 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3723 key = (const struct dt_key *)&oti->oti_quota_id;
3726 rc = iam_it_get(it, (struct iam_key *)key);
3728 if (S_ISDIR(obj->oo_inode->i_mode))
3729 iam_rec = (struct iam_rec *)oti->oti_ldp;
3731 iam_rec = (struct iam_rec *) rec;
3733 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3735 if (S_ISDIR(obj->oo_inode->i_mode))
3736 osd_fid_unpack((struct lu_fid *) rec,
3737 (struct osd_fid_pack *)iam_rec);
3738 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3739 osd_quota_unpack(obj, rec);
3744 osd_ipd_put(env, bag, ipd);
3746 LINVRNT(osd_invariant(obj));
3751 static int osd_index_declare_iam_insert(const struct lu_env *env,
3752 struct dt_object *dt,
3753 const struct dt_rec *rec,
3754 const struct dt_key *key,
3755 struct thandle *handle)
3757 struct osd_thandle *oh;
3759 LASSERT(handle != NULL);
3761 oh = container_of0(handle, struct osd_thandle, ot_super);
3762 LASSERT(oh->ot_handle == NULL);
3764 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3765 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3771 * Inserts (key, value) pair in \a dt index object.
3773 * \param dt osd index object
3774 * \param key key for index
3775 * \param rec record reference
3776 * \param th transaction handler
3779 * \retval -ve failure
3781 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3782 const struct dt_rec *rec,
3783 const struct dt_key *key, struct thandle *th,
3784 struct lustre_capa *capa, int ignore_quota)
3786 struct osd_object *obj = osd_dt_obj(dt);
3787 struct iam_path_descr *ipd;
3788 struct osd_thandle *oh;
3789 struct iam_container *bag = &obj->oo_dir->od_container;
3790 struct osd_thread_info *oti = osd_oti_get(env);
3791 struct iam_rec *iam_rec;
3796 LINVRNT(osd_invariant(obj));
3797 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3798 LASSERT(bag->ic_object == obj->oo_inode);
3799 LASSERT(th != NULL);
3801 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3804 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3806 ipd = osd_idx_ipd_get(env, bag);
3807 if (unlikely(ipd == NULL))
3810 oh = container_of0(th, struct osd_thandle, ot_super);
3811 LASSERT(oh->ot_handle != NULL);
3812 LASSERT(oh->ot_handle->h_transaction != NULL);
3813 if (S_ISDIR(obj->oo_inode->i_mode)) {
3814 iam_rec = (struct iam_rec *)oti->oti_ldp;
3815 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3816 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3817 /* pack quota uid/gid */
3818 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3819 key = (const struct dt_key *)&oti->oti_quota_id;
3820 /* pack quota record */
3821 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3822 iam_rec = (struct iam_rec *)rec;
3824 iam_rec = (struct iam_rec *)rec;
3827 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3829 osd_ipd_put(env, bag, ipd);
3830 LINVRNT(osd_invariant(obj));
3835 * Calls ldiskfs_add_entry() to add directory entry
3836 * into the directory. This is required for
3837 * interoperability mode (b11826)
3839 * \retval 0, on success
3840 * \retval -ve, on error
3842 static int __osd_ea_add_rec(struct osd_thread_info *info,
3843 struct osd_object *pobj, struct inode *cinode,
3844 const char *name, const struct dt_rec *fid,
3845 struct htree_lock *hlock, struct thandle *th)
3847 struct ldiskfs_dentry_param *ldp;
3848 struct dentry *child;
3849 struct osd_thandle *oth;
3852 oth = container_of(th, struct osd_thandle, ot_super);
3853 LASSERT(oth->ot_handle != NULL);
3854 LASSERT(oth->ot_handle->h_transaction != NULL);
3855 LASSERT(pobj->oo_inode);
3857 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3858 if (unlikely(pobj->oo_inode ==
3859 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3862 osd_get_ldiskfs_dirent_param(ldp, fid);
3863 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3864 child->d_fsdata = (void *)ldp;
3865 ll_vfs_dq_init(pobj->oo_inode);
3866 rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3872 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3873 * into the directory.Also sets flags into osd object to
3874 * indicate dot and dotdot are created. This is required for
3875 * interoperability mode (b11826)
3877 * \param dir directory for dot and dotdot fixup.
3878 * \param obj child object for linking
3880 * \retval 0, on success
3881 * \retval -ve, on error
3883 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3884 struct osd_object *dir,
3885 struct inode *parent_dir, const char *name,
3886 const struct dt_rec *dot_fid,
3887 const struct dt_rec *dot_dot_fid,
3890 struct inode *inode = dir->oo_inode;
3891 struct osd_thandle *oth;
3894 oth = container_of(th, struct osd_thandle, ot_super);
3895 LASSERT(oth->ot_handle->h_transaction != NULL);
3896 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3898 if (strcmp(name, dot) == 0) {
3899 if (dir->oo_compat_dot_created) {
3902 LASSERT(inode == parent_dir);
3903 dir->oo_compat_dot_created = 1;
3906 } else if (strcmp(name, dotdot) == 0) {
3907 if (!dir->oo_compat_dot_created)
3909 /* in case of rename, dotdot is already created */
3910 if (dir->oo_compat_dotdot_created) {
3911 return __osd_ea_add_rec(info, dir, parent_dir, name,
3912 dot_dot_fid, NULL, th);
3915 result = osd_add_dot_dotdot_internal(info, dir->oo_inode,
3916 parent_dir, dot_fid,
3919 dir->oo_compat_dotdot_created = 1;
3927 * It will call the appropriate osd_add* function and return the
3928 * value, return by respective functions.
3930 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3931 struct inode *cinode, const char *name,
3932 const struct dt_rec *fid, struct thandle *th)
3934 struct osd_thread_info *info = osd_oti_get(env);
3935 struct htree_lock *hlock;
3938 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3940 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3942 if (hlock != NULL) {
3943 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3946 down_write(&pobj->oo_ext_idx_sem);
3948 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3949 (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3952 if (hlock != NULL) {
3953 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3954 pobj->oo_inode, LDISKFS_HLOCK_ADD);
3956 down_write(&pobj->oo_ext_idx_sem);
3959 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
3960 struct lu_fid *tfid = &info->oti_fid;
3962 *tfid = *(const struct lu_fid *)fid;
3964 rc = __osd_ea_add_rec(info, pobj, cinode, name,
3965 (const struct dt_rec *)tfid,
3968 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3973 ldiskfs_htree_unlock(hlock);
3975 up_write(&pobj->oo_ext_idx_sem);
3981 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3982 struct osd_idmap_cache *oic)
3984 struct osd_scrub *scrub = &dev->od_scrub;
3985 struct lu_fid *fid = &oic->oic_fid;
3986 struct osd_inode_id *id = &oti->oti_id;
3991 if (!fid_is_norm(fid) && !fid_is_igif(fid))
3995 rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
3996 if (rc != 0 && rc != -ENOENT)
3999 if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
4002 if (thread_is_running(&scrub->os_thread)) {
4003 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
4004 /* There is race condition between osd_oi_lookup and OI scrub.
4005 * The OI scrub finished just after osd_oi_lookup() failure.
4006 * Under such case, it is unnecessary to trigger OI scrub again,
4007 * but try to call osd_oi_lookup() again. */
4008 if (unlikely(rc == -EAGAIN))
4014 if (!dev->od_noscrub && ++once == 1) {
4015 rc = osd_scrub_start(dev);
4016 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC for "DFID
4018 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4027 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4028 struct osd_device *dev,
4029 struct osd_idmap_cache *oic,
4030 struct lu_fid *fid, __u32 ino)
4032 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
4033 struct inode *inode;
4036 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4037 inode = osd_iget(oti, dev, &oic->oic_lid);
4038 if (IS_ERR(inode)) {
4039 fid_zero(&oic->oic_fid);
4040 return PTR_ERR(inode);
4043 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4046 fid_zero(&oic->oic_fid);
4048 *fid = oic->oic_fid = lma->lma_self_fid;
4052 int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4053 struct osd_inode_id *id, const struct lu_fid *fid)
4055 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4056 id->oii_ino, id->oii_gen, info);
4057 info->oti_cache.oic_lid = *id;
4058 info->oti_cache.oic_fid = *fid;
4059 info->oti_cache.oic_dev = osd;
4065 * Calls ->lookup() to find dentry. From dentry get inode and
4066 * read inode's ea to get fid. This is required for interoperability
4069 * \retval 0, on success
4070 * \retval -ve, on error
4072 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4073 struct dt_rec *rec, const struct dt_key *key)
4075 struct inode *dir = obj->oo_inode;
4076 struct dentry *dentry;
4077 struct ldiskfs_dir_entry_2 *de;
4078 struct buffer_head *bh;
4079 struct lu_fid *fid = (struct lu_fid *) rec;
4080 struct htree_lock *hlock = NULL;
4085 LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
4087 dentry = osd_child_dentry_get(env, obj,
4088 (char *)key, strlen((char *)key));
4090 if (obj->oo_hl_head != NULL) {
4091 hlock = osd_oti_get(env)->oti_hlock;
4092 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4093 dir, LDISKFS_HLOCK_LOOKUP);
4095 down_read(&obj->oo_ext_idx_sem);
4098 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4100 struct osd_thread_info *oti = osd_oti_get(env);
4101 struct osd_inode_id *id = &oti->oti_id;
4102 struct osd_idmap_cache *oic = &oti->oti_cache;
4103 struct osd_device *dev = osd_obj2dev(obj);
4104 struct osd_scrub *scrub = &dev->od_scrub;
4105 struct scrub_file *sf = &scrub->os_file;
4107 ino = le32_to_cpu(de->inode);
4108 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4110 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4114 rc = osd_get_fid_from_dentry(de, rec);
4116 /* done with de, release bh */
4119 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4121 osd_id_gen(id, ino, OSD_OII_NOGEN);
4123 fid_zero(&oic->oic_fid);
4127 if (osd_remote_fid(env, dev, fid))
4130 rc = osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id,
4134 if ((scrub->os_pos_current <= ino) &&
4135 ((sf->sf_flags & SF_INCONSISTENT) ||
4136 (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
4137 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
4139 osd_consistency_check(oti, dev, oic);
4148 ldiskfs_htree_unlock(hlock);
4150 up_read(&obj->oo_ext_idx_sem);
4155 * Find the osd object for given fid.
4157 * \param fid need to find the osd object having this fid
4159 * \retval osd_object on success
4160 * \retval -ve on error
4162 struct osd_object *osd_object_find(const struct lu_env *env,
4163 struct dt_object *dt,
4164 const struct lu_fid *fid)
4166 struct lu_device *ludev = dt->do_lu.lo_dev;
4167 struct osd_object *child = NULL;
4168 struct lu_object *luch;
4169 struct lu_object *lo;
4172 * at this point topdev might not exist yet
4173 * (i.e. MGS is preparing profiles). so we can
4174 * not rely on topdev and instead lookup with
4175 * our device passed as topdev. this can't work
4176 * if the object isn't cached yet (as osd doesn't
4177 * allocate lu_header). IOW, the object must be
4178 * in the cache, otherwise lu_object_alloc() crashes
4181 luch = lu_object_find_at(env, ludev, fid, NULL);
4182 if (!IS_ERR(luch)) {
4183 if (lu_object_exists(luch)) {
4184 lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4186 child = osd_obj(lo);
4188 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4189 "lu_object can't be located"
4190 DFID"\n", PFID(fid));
4192 if (child == NULL) {
4193 lu_object_put(env, luch);
4194 CERROR("Unable to get osd_object\n");
4195 child = ERR_PTR(-ENOENT);
4198 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4199 "lu_object does not exists "DFID"\n",
4201 lu_object_put(env, luch);
4202 child = ERR_PTR(-ENOENT);
4205 child = ERR_CAST(luch);
4212 * Put the osd object once done with it.
4214 * \param obj osd object that needs to be put
4216 static inline void osd_object_put(const struct lu_env *env,
4217 struct osd_object *obj)
4219 lu_object_put(env, &obj->oo_dt.do_lu);
4222 static int osd_index_declare_ea_insert(const struct lu_env *env,
4223 struct dt_object *dt,
4224 const struct dt_rec *rec,
4225 const struct dt_key *key,
4226 struct thandle *handle)
4228 struct osd_thandle *oh;
4229 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4230 struct lu_fid *fid = (struct lu_fid *)rec;
4234 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
4235 LASSERT(handle != NULL);
4237 oh = container_of0(handle, struct osd_thandle, ot_super);
4238 LASSERT(oh->ot_handle == NULL);
4240 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4241 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4243 if (osd_dt_obj(dt)->oo_inode == NULL) {
4244 const char *name = (const char *)key;
4245 /* Object is not being created yet. Only happens when
4246 * 1. declare directory create
4247 * 2. declare insert .
4248 * 3. declare insert ..
4250 LASSERT(strcmp(name, dotdot) == 0 || strcmp(name, dot) == 0);
4252 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4254 /* We ignore block quota on meta pool (MDTs), so needn't
4255 * calculate how many blocks will be consumed by this index
4257 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0,
4258 oh, true, true, NULL, false);
4264 rc = osd_remote_fid(env, osd, fid);
4270 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
4271 osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
4272 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4273 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4274 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4275 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4281 * Index add function for interoperability mode (b11826).
4282 * It will add the directory entry.This entry is needed to
4283 * maintain name->fid mapping.
4285 * \param key it is key i.e. file entry to be inserted
4286 * \param rec it is value of given key i.e. fid
4288 * \retval 0, on success
4289 * \retval -ve, on error
4291 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4292 const struct dt_rec *rec,
4293 const struct dt_key *key, struct thandle *th,
4294 struct lustre_capa *capa, int ignore_quota)
4296 struct osd_object *obj = osd_dt_obj(dt);
4297 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4298 struct lu_fid *fid = (struct lu_fid *) rec;
4299 const char *name = (const char *)key;
4300 struct osd_thread_info *oti = osd_oti_get(env);
4301 struct osd_inode_id *id = &oti->oti_id;
4302 struct inode *child_inode = NULL;
4303 struct osd_object *child = NULL;
4307 LASSERT(osd_invariant(obj));
4308 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
4309 LASSERT(th != NULL);
4311 osd_trans_exec_op(env, th, OSD_OT_INSERT);
4313 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
4316 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!", PFID(fid));
4318 rc = osd_remote_fid(env, osd, fid);
4320 CERROR("%s: Can not find object "DFID" rc %d\n",
4321 osd_name(osd), PFID(fid), rc);
4326 /* Insert remote entry */
4327 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4328 struct osd_mdobj_map *omm = osd->od_mdt_map;
4329 struct osd_thandle *oh;
4331 /* If parent on remote MDT, we need put this object
4333 oh = container_of(th, typeof(*oh), ot_super);
4334 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4336 CERROR("%s: add "DFID" error: rc = %d\n",
4338 PFID(lu_object_fid(&dt->do_lu)), rc);
4342 child_inode = igrab(omm->omm_remote_parent->d_inode);
4344 child_inode = osd_create_local_agent_inode(env, osd,
4347 if (IS_ERR(child_inode))
4348 RETURN(PTR_ERR(child_inode));
4351 /* Insert local entry */
4352 child = osd_object_find(env, dt, fid);
4353 if (IS_ERR(child)) {
4354 CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4355 osd_name(osd), PFID(fid),
4356 id->oii_ino, id->oii_gen,
4357 (int)PTR_ERR(child));
4358 RETURN(PTR_ERR(child));
4360 child_inode = igrab(child->oo_inode);
4363 rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th);
4367 osd_object_put(env, child);
4368 LASSERT(osd_invariant(obj));
4373 * Initialize osd Iterator for given osd index object.
4375 * \param dt osd index object
4378 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4379 struct dt_object *dt,
4381 struct lustre_capa *capa)
4383 struct osd_it_iam *it;
4384 struct osd_thread_info *oti = osd_oti_get(env);
4385 struct osd_object *obj = osd_dt_obj(dt);
4386 struct lu_object *lo = &dt->do_lu;
4387 struct iam_path_descr *ipd;
4388 struct iam_container *bag = &obj->oo_dir->od_container;
4390 LASSERT(lu_object_exists(lo));
4392 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
4393 return ERR_PTR(-EACCES);
4396 ipd = osd_it_ipd_get(env, bag);
4397 if (likely(ipd != NULL)) {
4401 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4402 return (struct dt_it *)it;
4404 return ERR_PTR(-ENOMEM);
4408 * free given Iterator.
4411 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4413 struct osd_it_iam *it = (struct osd_it_iam *)di;
4414 struct osd_object *obj = it->oi_obj;
4416 iam_it_fini(&it->oi_it);
4417 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4418 lu_object_put(env, &obj->oo_dt.do_lu);
4422 * Move Iterator to record specified by \a key
4424 * \param di osd iterator
4425 * \param key key for index
4427 * \retval +ve di points to record with least key not larger than key
4428 * \retval 0 di points to exact matched key
4429 * \retval -ve failure
4432 static int osd_it_iam_get(const struct lu_env *env,
4433 struct dt_it *di, const struct dt_key *key)
4435 struct osd_thread_info *oti = osd_oti_get(env);
4436 struct osd_it_iam *it = (struct osd_it_iam *)di;
4438 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4439 /* swab quota uid/gid */
4440 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4441 key = (struct dt_key *)&oti->oti_quota_id;
4444 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4450 * \param di osd iterator
4452 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4454 struct osd_it_iam *it = (struct osd_it_iam *)di;
4456 iam_it_put(&it->oi_it);
4460 * Move iterator by one record
4462 * \param di osd iterator
4464 * \retval +1 end of container reached
4466 * \retval -ve failure
4469 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4471 struct osd_it_iam *it = (struct osd_it_iam *)di;
4473 return iam_it_next(&it->oi_it);
4477 * Return pointer to the key under iterator.
4480 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4481 const struct dt_it *di)
4483 struct osd_thread_info *oti = osd_oti_get(env);
4484 struct osd_it_iam *it = (struct osd_it_iam *)di;
4485 struct osd_object *obj = it->oi_obj;
4488 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4490 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4491 /* swab quota uid/gid */
4492 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4493 key = (struct dt_key *)&oti->oti_quota_id;
4500 * Return size of key under iterator (in bytes)
4503 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4505 struct osd_it_iam *it = (struct osd_it_iam *)di;
4507 return iam_it_key_size(&it->oi_it);
4511 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4513 /* check if file type is required */
4514 if (ent->lde_attrs & LUDA_TYPE) {
4515 struct luda_type *lt;
4516 int align = sizeof(*lt) - 1;
4518 len = (len + align) & ~align;
4519 lt = (struct luda_type *)(ent->lde_name + len);
4520 lt->lt_type = cpu_to_le16(DTTOIF(type));
4523 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4527 * build lu direct from backend fs dirent.
4531 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4532 char *name, __u16 namelen, __u16 type, __u32 attr)
4534 ent->lde_attrs = attr | LUDA_FID;
4535 fid_cpu_to_le(&ent->lde_fid, fid);
4537 ent->lde_hash = cpu_to_le64(offset);
4538 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4540 strncpy(ent->lde_name, name, namelen);
4541 ent->lde_name[namelen] = '\0';
4542 ent->lde_namelen = cpu_to_le16(namelen);
4544 /* append lustre attributes */
4545 osd_it_append_attrs(ent, namelen, type);
4549 * Return pointer to the record under iterator.
4551 static int osd_it_iam_rec(const struct lu_env *env,
4552 const struct dt_it *di,
4553 struct dt_rec *dtrec, __u32 attr)
4555 struct osd_it_iam *it = (struct osd_it_iam *)di;
4556 struct osd_thread_info *info = osd_oti_get(env);
4559 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4560 const struct osd_fid_pack *rec;
4561 struct lu_fid *fid = &info->oti_fid;
4562 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
4568 name = (char *)iam_it_key_get(&it->oi_it);
4570 RETURN(PTR_ERR(name));
4572 namelen = iam_it_key_size(&it->oi_it);
4574 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4576 RETURN(PTR_ERR(rec));
4578 rc = osd_fid_unpack(fid, rec);
4582 hash = iam_it_store(&it->oi_it);
4584 /* IAM does not store object type in IAM index (dir) */
4585 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4587 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4588 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4589 (struct iam_rec *)dtrec);
4590 osd_quota_unpack(it->oi_obj, dtrec);
4592 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4593 (struct iam_rec *)dtrec);
4600 * Returns cookie for current Iterator position.
4602 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4604 struct osd_it_iam *it = (struct osd_it_iam *)di;
4606 return iam_it_store(&it->oi_it);
4610 * Restore iterator from cookie.
4612 * \param di osd iterator
4613 * \param hash Iterator location cookie
4615 * \retval +ve di points to record with least key not larger than key.
4616 * \retval 0 di points to exact matched key
4617 * \retval -ve failure
4620 static int osd_it_iam_load(const struct lu_env *env,
4621 const struct dt_it *di, __u64 hash)
4623 struct osd_it_iam *it = (struct osd_it_iam *)di;
4625 return iam_it_load(&it->oi_it, hash);
4628 static const struct dt_index_operations osd_index_iam_ops = {
4629 .dio_lookup = osd_index_iam_lookup,
4630 .dio_declare_insert = osd_index_declare_iam_insert,
4631 .dio_insert = osd_index_iam_insert,
4632 .dio_declare_delete = osd_index_declare_iam_delete,
4633 .dio_delete = osd_index_iam_delete,
4635 .init = osd_it_iam_init,
4636 .fini = osd_it_iam_fini,
4637 .get = osd_it_iam_get,
4638 .put = osd_it_iam_put,
4639 .next = osd_it_iam_next,
4640 .key = osd_it_iam_key,
4641 .key_size = osd_it_iam_key_size,
4642 .rec = osd_it_iam_rec,
4643 .store = osd_it_iam_store,
4644 .load = osd_it_iam_load
4650 * Creates or initializes iterator context.
4652 * \retval struct osd_it_ea, iterator structure on success
4655 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4656 struct dt_object *dt,
4658 struct lustre_capa *capa)
4660 struct osd_object *obj = osd_dt_obj(dt);
4661 struct osd_thread_info *info = osd_oti_get(env);
4662 struct osd_it_ea *it = &info->oti_it_ea;
4663 struct file *file = &it->oie_file;
4664 struct lu_object *lo = &dt->do_lu;
4665 struct dentry *obj_dentry = &info->oti_it_dentry;
4667 LASSERT(lu_object_exists(lo));
4669 obj_dentry->d_inode = obj->oo_inode;
4670 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4671 obj_dentry->d_name.hash = 0;
4673 it->oie_rd_dirent = 0;
4674 it->oie_it_dirent = 0;
4675 it->oie_dirent = NULL;
4676 it->oie_buf = info->oti_it_ea_buf;
4679 /* Reset the "file" totally to avoid to reuse any old value from
4680 * former readdir handling, the "file->f_pos" should be zero. */
4681 memset(file, 0, sizeof(*file));
4682 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4683 if (attr & LUDA_64BITHASH)
4684 file->f_mode = FMODE_64BITHASH;
4686 file->f_mode = FMODE_32BITHASH;
4687 file->f_dentry = obj_dentry;
4688 file->f_mapping = obj->oo_inode->i_mapping;
4689 file->f_op = obj->oo_inode->i_fop;
4690 set_file_inode(file, obj->oo_inode);
4693 RETURN((struct dt_it *) it);
4697 * Destroy or finishes iterator context.
4699 * \param di iterator structure to be destroyed
4701 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4703 struct osd_it_ea *it = (struct osd_it_ea *)di;
4704 struct osd_object *obj = it->oie_obj;
4705 struct inode *inode = obj->oo_inode;
4708 it->oie_file.f_op->release(inode, &it->oie_file);
4709 lu_object_put(env, &obj->oo_dt.do_lu);
4714 * It position the iterator at given key, so that next lookup continues from
4715 * that key Or it is similar to dio_it->load() but based on a key,
4716 * rather than file position.
4718 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4721 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4723 static int osd_it_ea_get(const struct lu_env *env,
4724 struct dt_it *di, const struct dt_key *key)
4726 struct osd_it_ea *it = (struct osd_it_ea *)di;
4729 LASSERT(((const char *)key)[0] == '\0');
4730 it->oie_file.f_pos = 0;
4731 it->oie_rd_dirent = 0;
4732 it->oie_it_dirent = 0;
4733 it->oie_dirent = NULL;
4741 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4746 * It is called internally by ->readdir(). It fills the
4747 * iterator's in-memory data structure with required
4748 * information i.e. name, namelen, rec_size etc.
4750 * \param buf in which information to be filled in.
4751 * \param name name of the file in given dir
4753 * \retval 0 on success
4754 * \retval 1 on buffer full
4756 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
4757 loff_t offset, __u64 ino,
4760 struct osd_it_ea *it = (struct osd_it_ea *)buf;
4761 struct osd_object *obj = it->oie_obj;
4762 struct osd_it_ea_dirent *ent = it->oie_dirent;
4763 struct lu_fid *fid = &ent->oied_fid;
4764 struct osd_fid_pack *rec;
4767 /* this should never happen */
4768 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4769 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4773 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4777 /* "." is just the object itself. */
4778 if (namelen == 1 && name[0] == '.') {
4779 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4780 } else if (d_type & LDISKFS_DIRENT_LUFID) {
4781 rec = (struct osd_fid_pack*) (name + namelen + 1);
4782 if (osd_fid_unpack(fid, rec) != 0)
4787 d_type &= ~LDISKFS_DIRENT_LUFID;
4789 /* NOT export local root. */
4790 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
4791 ino = obj->oo_inode->i_ino;
4792 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4795 ent->oied_ino = ino;
4796 ent->oied_off = offset;
4797 ent->oied_namelen = namelen;
4798 ent->oied_type = d_type;
4800 memcpy(ent->oied_name, name, namelen);
4802 it->oie_rd_dirent++;
4803 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4808 * Calls ->readdir() to load a directory entry at a time
4809 * and stored it in iterator's in-memory data structure.
4811 * \param di iterator's in memory structure
4813 * \retval 0 on success
4814 * \retval -ve on error
4816 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4817 const struct dt_it *di)
4819 struct osd_it_ea *it = (struct osd_it_ea *)di;
4820 struct osd_object *obj = it->oie_obj;
4821 struct inode *inode = obj->oo_inode;
4822 struct htree_lock *hlock = NULL;
4826 it->oie_dirent = it->oie_buf;
4827 it->oie_rd_dirent = 0;
4829 if (obj->oo_hl_head != NULL) {
4830 hlock = osd_oti_get(env)->oti_hlock;
4831 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4832 inode, LDISKFS_HLOCK_READDIR);
4834 down_read(&obj->oo_ext_idx_sem);
4837 result = inode->i_fop->readdir(&it->oie_file, it,
4838 (filldir_t) osd_ldiskfs_filldir);
4841 ldiskfs_htree_unlock(hlock);
4843 up_read(&obj->oo_ext_idx_sem);
4845 if (it->oie_rd_dirent == 0) {
4846 /*If it does not get any dirent, it means it has been reached
4847 *to the end of the dir */
4848 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
4850 it->oie_dirent = it->oie_buf;
4851 it->oie_it_dirent = 1;
4858 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4859 * to load a directory entry at a time and stored it in
4860 * iterator's in-memory data structure.
4862 * \param di iterator's in memory structure
4864 * \retval +ve iterator reached to end
4865 * \retval 0 iterator not reached to end
4866 * \retval -ve on error
4868 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4870 struct osd_it_ea *it = (struct osd_it_ea *)di;
4875 if (it->oie_it_dirent < it->oie_rd_dirent) {
4877 (void *) it->oie_dirent +
4878 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4879 it->oie_dirent->oied_namelen);
4880 it->oie_it_dirent++;
4883 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4886 rc = osd_ldiskfs_it_fill(env, di);
4893 * Returns the key at current position from iterator's in memory structure.
4895 * \param di iterator's in memory structure
4897 * \retval key i.e. struct dt_key on success
4899 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4900 const struct dt_it *di)
4902 struct osd_it_ea *it = (struct osd_it_ea *)di;
4904 return (struct dt_key *)it->oie_dirent->oied_name;
4908 * Returns the key's size at current position from iterator's in memory structure.
4910 * \param di iterator's in memory structure
4912 * \retval key_size i.e. struct dt_key on success
4914 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4916 struct osd_it_ea *it = (struct osd_it_ea *)di;
4918 return it->oie_dirent->oied_namelen;
4922 osd_dirent_update(handle_t *jh, struct super_block *sb,
4923 struct osd_it_ea_dirent *ent, struct lu_fid *fid,
4924 struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
4926 struct osd_fid_pack *rec;
4930 LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
4931 LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
4933 rc = ldiskfs_journal_get_write_access(jh, bh);
4935 CERROR("%.16s: fail to write access for update dirent: "
4936 "name = %.*s, rc = %d\n",
4937 LDISKFS_SB(sb)->s_es->s_volume_name,
4938 ent->oied_namelen, ent->oied_name, rc);
4942 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
4943 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
4944 rc = ldiskfs_journal_dirty_metadata(jh, bh);
4946 CERROR("%.16s: fail to dirty metadata for update dirent: "
4947 "name = %.*s, rc = %d\n",
4948 LDISKFS_SB(sb)->s_es->s_volume_name,
4949 ent->oied_namelen, ent->oied_name, rc);
4955 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
4957 if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
4958 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
4965 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
4967 LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
4968 "dot_dotdot = %d\n", dot_dotdot);
4970 if (LDISKFS_DIR_REC_LEN(de) >=
4971 __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
4978 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
4979 struct inode *dir, struct inode *inode,
4980 struct osd_it_ea_dirent *ent, struct lu_fid *fid,
4981 struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
4982 struct htree_lock *hlock)
4984 struct dentry *dentry;
4985 struct osd_fid_pack *rec;
4986 struct ldiskfs_dentry_param *ldp;
4990 if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
4991 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
4994 /* There is enough space to hold the FID-in-dirent. */
4995 if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
4996 dir->i_sb->s_blocksize)) {
4997 rc = ldiskfs_journal_get_write_access(jh, bh);
4999 CERROR("%.16s: fail to write access for reinsert "
5000 "dirent: name = %.*s, rc = %d\n",
5001 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5002 ent->oied_namelen, ent->oied_name, rc);
5006 de->name[de->name_len] = 0;
5007 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5008 rec->fp_len = sizeof(struct lu_fid) + 1;
5009 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5010 de->file_type |= LDISKFS_DIRENT_LUFID;
5012 rc = ldiskfs_journal_dirty_metadata(jh, bh);
5014 CERROR("%.16s: fail to dirty metadata for reinsert "
5015 "dirent: name = %.*s, rc = %d\n",
5016 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5017 ent->oied_namelen, ent->oied_name, rc);
5022 rc = ldiskfs_delete_entry(jh, dir, de, bh);
5024 CERROR("%.16s: fail to delete entry for reinsert dirent: "
5025 "name = %.*s, rc = %d\n",
5026 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5027 ent->oied_namelen, ent->oied_name, rc);
5031 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5033 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5034 osd_get_ldiskfs_dirent_param(ldp, (const struct dt_rec *)fid);
5035 dentry->d_fsdata = (void *)ldp;
5036 ll_vfs_dq_init(dir);
5037 rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5038 /* It is too bad, we cannot reinsert the name entry back.
5039 * That means we lose it! */
5041 CERROR("%.16s: fail to insert entry for reinsert dirent: "
5042 "name = %.*s, rc = %d\n",
5043 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5044 ent->oied_namelen, ent->oied_name, rc);
5050 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5051 struct osd_it_ea *it, struct lu_fid *fid,
5052 struct osd_inode_id *id, __u32 *attr)
5054 struct osd_thread_info *info = osd_oti_get(env);
5055 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
5056 struct osd_device *dev = osd_obj2dev(obj);
5057 struct super_block *sb = osd_sb(dev);
5058 const char *devname =
5059 LDISKFS_SB(sb)->s_es->s_volume_name;
5060 struct osd_it_ea_dirent *ent = it->oie_dirent;
5061 struct inode *dir = obj->oo_inode;
5062 struct htree_lock *hlock = NULL;
5063 struct buffer_head *bh = NULL;
5064 handle_t *jh = NULL;
5065 struct ldiskfs_dir_entry_2 *de;
5066 struct dentry *dentry;
5067 struct inode *inode;
5074 if (ent->oied_name[0] == '.') {
5075 if (ent->oied_namelen == 1)
5077 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5081 dentry = osd_child_dentry_get(env, obj, ent->oied_name,
5084 /* We need to ensure that the name entry is still valid.
5085 * Because it may be removed or renamed by other already.
5087 * The unlink or rename operation will start journal before PDO lock,
5088 * so to avoid deadlock, here we need to start journal handle before
5089 * related PDO lock also. But because we do not know whether there
5090 * will be something to be repaired before PDO lock, we just start
5091 * journal without conditions.
5093 * We may need to remove the name entry firstly, then insert back.
5094 * One credit is for user quota file update.
5095 * One credit is for group quota file update.
5096 * Two credits are for dirty inode. */
5097 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5098 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5101 if (dev->od_dirent_journal) {
5102 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5105 CERROR("%.16s: fail to start trans for dirent "
5106 "check_repair: credits %d, name %.*s, rc %d\n",
5107 devname, credits, ent->oied_namelen,
5108 ent->oied_name, rc);
5112 if (obj->oo_hl_head != NULL) {
5113 hlock = osd_oti_get(env)->oti_hlock;
5114 /* "0" means exclusive lock for the whole directory.
5115 * We need to prevent others access such name entry
5116 * during the delete + insert. Neither HLOCK_ADD nor
5117 * HLOCK_DEL cannot guarantee the atomicity. */
5118 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5120 down_write(&obj->oo_ext_idx_sem);
5123 if (obj->oo_hl_head != NULL) {
5124 hlock = osd_oti_get(env)->oti_hlock;
5125 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5126 LDISKFS_HLOCK_LOOKUP);
5128 down_read(&obj->oo_ext_idx_sem);
5132 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5133 /* For dot/dotdot entry, if there is not enough space to hold the
5134 * FID-in-dirent, just keep them there. It only happens when the
5135 * device upgraded from 1.8 or restored from MDT file-level backup.
5136 * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5137 * and needs to get FID from LMA when readdir, it will not affect the
5138 * performance much. */
5139 if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
5140 (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5141 *attr |= LUDA_IGNORE;
5142 GOTO(out_journal, rc = 0);
5145 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5146 inode = osd_iget(info, dev, id);
5147 if (IS_ERR(inode)) {
5148 rc = PTR_ERR(inode);
5149 if (rc == -ENOENT || rc == -ESTALE) {
5150 *attr |= LUDA_IGNORE;
5154 GOTO(out_journal, rc);
5157 /* skip the REMOTE_PARENT_DIR. */
5158 if (inode == dev->od_mdt_map->omm_remote_parent->d_inode)
5159 GOTO(out_inode, rc = 0);
5161 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
5163 LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
5165 if (fid_is_sane(fid)) {
5166 /* FID-in-dirent is valid. */
5167 if (lu_fid_eq(fid, &lma->lma_self_fid))
5168 GOTO(out_inode, rc = 0);
5170 /* Do not repair under dryrun mode. */
5171 if (*attr & LUDA_VERIFY_DRYRUN) {
5172 *attr |= LUDA_REPAIR;
5173 GOTO(out_inode, rc = 0);
5176 if (!dev->od_dirent_journal) {
5180 ldiskfs_htree_unlock(hlock);
5182 up_read(&obj->oo_ext_idx_sem);
5183 dev->od_dirent_journal = 1;
5187 *fid = lma->lma_self_fid;
5189 /* Update the FID-in-dirent. */
5190 rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
5192 *attr |= LUDA_REPAIR;
5194 /* Do not repair under dryrun mode. */
5195 if (*attr & LUDA_VERIFY_DRYRUN) {
5196 *fid = lma->lma_self_fid;
5197 *attr |= LUDA_REPAIR;
5198 GOTO(out_inode, rc = 0);
5201 if (!dev->od_dirent_journal) {
5205 ldiskfs_htree_unlock(hlock);
5207 up_read(&obj->oo_ext_idx_sem);
5208 dev->od_dirent_journal = 1;
5212 *fid = lma->lma_self_fid;
5214 /* Append the FID-in-dirent. */
5215 rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5216 fid, bh, de, hlock);
5218 *attr |= LUDA_REPAIR;
5220 } else if (rc == -ENODATA) {
5221 /* Do not repair under dryrun mode. */
5222 if (*attr & LUDA_VERIFY_DRYRUN) {
5223 if (fid_is_sane(fid)) {
5224 *attr |= LUDA_REPAIR;
5226 lu_igif_build(fid, inode->i_ino,
5227 inode->i_generation);
5228 *attr |= LUDA_UPGRADE;
5230 GOTO(out_inode, rc = 0);
5233 if (!dev->od_dirent_journal) {
5237 ldiskfs_htree_unlock(hlock);
5239 up_read(&obj->oo_ext_idx_sem);
5240 dev->od_dirent_journal = 1;
5245 if (unlikely(fid_is_sane(fid))) {
5246 /* FID-in-dirent exists, but FID-in-LMA is lost.
5247 * Trust the FID-in-dirent, and add FID-in-LMA. */
5248 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5250 *attr |= LUDA_REPAIR;
5252 lu_igif_build(fid, inode->i_ino, inode->i_generation);
5253 /* It is probably IGIF object. Only aappend the
5254 * FID-in-dirent. OI scrub will process FID-in-LMA. */
5255 rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5256 fid, bh, de, hlock);
5258 *attr |= LUDA_UPGRADE;
5262 GOTO(out_inode, rc);
5269 if (hlock != NULL) {
5270 ldiskfs_htree_unlock(hlock);
5272 if (dev->od_dirent_journal)
5273 up_write(&obj->oo_ext_idx_sem);
5275 up_read(&obj->oo_ext_idx_sem);
5278 ldiskfs_journal_stop(jh);
5279 if (rc >= 0 && !dirty)
5280 dev->od_dirent_journal = 0;
5285 * Returns the value at current position from iterator's in memory structure.
5287 * \param di struct osd_it_ea, iterator's in memory structure
5288 * \param attr attr requested for dirent.
5289 * \param lde lustre dirent
5291 * \retval 0 no error and \param lde has correct lustre dirent.
5292 * \retval -ve on error
5294 static inline int osd_it_ea_rec(const struct lu_env *env,
5295 const struct dt_it *di,
5296 struct dt_rec *dtrec, __u32 attr)
5298 struct osd_it_ea *it = (struct osd_it_ea *)di;
5299 struct osd_object *obj = it->oie_obj;
5300 struct osd_device *dev = osd_obj2dev(obj);
5301 struct osd_scrub *scrub = &dev->od_scrub;
5302 struct scrub_file *sf = &scrub->os_file;
5303 struct osd_thread_info *oti = osd_oti_get(env);
5304 struct osd_inode_id *id = &oti->oti_id;
5305 struct osd_idmap_cache *oic = &oti->oti_cache;
5306 struct lu_fid *fid = &it->oie_dirent->oied_fid;
5307 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
5308 __u32 ino = it->oie_dirent->oied_ino;
5312 if (attr & LUDA_VERIFY) {
5314 if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) {
5315 attr |= LUDA_IGNORE;
5318 rc = osd_dirent_check_repair(env, obj, it, fid, id,
5322 attr &= ~LU_DIRENT_ATTRS_MASK;
5323 if (!fid_is_sane(fid)) {
5324 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
5327 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5329 osd_id_gen(id, ino, OSD_OII_NOGEN);
5333 /* Pack the entry anyway, at least the offset is right. */
5334 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5335 it->oie_dirent->oied_name,
5336 it->oie_dirent->oied_namelen,
5337 it->oie_dirent->oied_type, attr);
5342 if (osd_remote_fid(env, dev, fid))
5345 if (likely(!(attr & LUDA_IGNORE)))
5346 rc = osd_add_oi_cache(oti, dev, id, fid);
5348 if (!(attr & LUDA_VERIFY) &&
5349 (scrub->os_pos_current <= ino) &&
5350 ((sf->sf_flags & SF_INCONSISTENT) ||
5351 (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
5352 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
5353 osd_consistency_check(oti, dev, oic);
5359 * Returns a cookie for current position of the iterator head, so that
5360 * user can use this cookie to load/start the iterator next time.
5362 * \param di iterator's in memory structure
5364 * \retval cookie for current position, on success
5366 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5368 struct osd_it_ea *it = (struct osd_it_ea *)di;
5370 return it->oie_dirent->oied_off;
5374 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5375 * to load a directory entry at a time and stored it i inn,
5376 * in iterator's in-memory data structure.
5378 * \param di struct osd_it_ea, iterator's in memory structure
5380 * \retval +ve on success
5381 * \retval -ve on error
5383 static int osd_it_ea_load(const struct lu_env *env,
5384 const struct dt_it *di, __u64 hash)
5386 struct osd_it_ea *it = (struct osd_it_ea *)di;
5390 it->oie_file.f_pos = hash;
5392 rc = osd_ldiskfs_it_fill(env, di);
5400 * Index lookup function for interoperability mode (b11826).
5402 * \param key, key i.e. file name to be searched
5404 * \retval +ve, on success
5405 * \retval -ve, on error
5407 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5408 struct dt_rec *rec, const struct dt_key *key,
5409 struct lustre_capa *capa)
5411 struct osd_object *obj = osd_dt_obj(dt);
5416 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5417 LINVRNT(osd_invariant(obj));
5419 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
5422 rc = osd_ea_lookup_rec(env, obj, rec, key);
5429 * Index and Iterator operations for interoperability
5430 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5432 static const struct dt_index_operations osd_index_ea_ops = {
5433 .dio_lookup = osd_index_ea_lookup,
5434 .dio_declare_insert = osd_index_declare_ea_insert,
5435 .dio_insert = osd_index_ea_insert,
5436 .dio_declare_delete = osd_index_declare_ea_delete,
5437 .dio_delete = osd_index_ea_delete,
5439 .init = osd_it_ea_init,
5440 .fini = osd_it_ea_fini,
5441 .get = osd_it_ea_get,
5442 .put = osd_it_ea_put,
5443 .next = osd_it_ea_next,
5444 .key = osd_it_ea_key,
5445 .key_size = osd_it_ea_key_size,
5446 .rec = osd_it_ea_rec,
5447 .store = osd_it_ea_store,
5448 .load = osd_it_ea_load
5452 static void *osd_key_init(const struct lu_context *ctx,
5453 struct lu_context_key *key)
5455 struct osd_thread_info *info;
5457 OBD_ALLOC_PTR(info);
5459 return ERR_PTR(-ENOMEM);
5461 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5462 if (info->oti_it_ea_buf == NULL)
5465 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5467 info->oti_hlock = ldiskfs_htree_lock_alloc();
5468 if (info->oti_hlock == NULL)
5474 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5477 return ERR_PTR(-ENOMEM);
5480 static void osd_key_fini(const struct lu_context *ctx,
5481 struct lu_context_key *key, void* data)
5483 struct osd_thread_info *info = data;
5485 if (info->oti_hlock != NULL)
5486 ldiskfs_htree_lock_free(info->oti_hlock);
5487 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5488 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5489 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5493 static void osd_key_exit(const struct lu_context *ctx,
5494 struct lu_context_key *key, void *data)
5496 struct osd_thread_info *info = data;
5498 LASSERT(info->oti_r_locks == 0);
5499 LASSERT(info->oti_w_locks == 0);
5500 LASSERT(info->oti_txns == 0);
5503 /* type constructor/destructor: osd_type_init, osd_type_fini */
5504 LU_TYPE_INIT_FINI(osd, &osd_key);
5506 struct lu_context_key osd_key = {
5507 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5508 .lct_init = osd_key_init,
5509 .lct_fini = osd_key_fini,
5510 .lct_exit = osd_key_exit
5514 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5515 const char *name, struct lu_device *next)
5517 struct osd_device *osd = osd_dev(d);
5519 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5520 >= sizeof(osd->od_svname))
5522 return osd_procfs_init(osd, name);
5525 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5529 /* shutdown quota slave instance associated with the device */
5530 if (o->od_quota_slave != NULL) {
5531 qsd_fini(env, o->od_quota_slave);
5532 o->od_quota_slave = NULL;
5538 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5542 if (o->od_mnt != NULL) {
5543 shrink_dcache_sb(osd_sb(o));
5544 osd_sync(env, &o->od_dt_dev);
5553 static int osd_mount(const struct lu_env *env,
5554 struct osd_device *o, struct lustre_cfg *cfg)
5556 const char *name = lustre_cfg_string(cfg, 0);
5557 const char *dev = lustre_cfg_string(cfg, 1);
5559 unsigned long page, s_flags, lmd_flags = 0;
5560 struct page *__page;
5561 struct file_system_type *type;
5562 char *options = NULL;
5564 struct osd_thread_info *info = osd_oti_get(env);
5565 struct lu_fid *fid = &info->oti_fid;
5566 struct inode *inode;
5570 if (o->od_mnt != NULL)
5573 if (strlen(dev) >= sizeof(o->od_mntdev))
5575 strcpy(o->od_mntdev, dev);
5577 OBD_PAGE_ALLOC(__page, GFP_IOFS);
5579 GOTO(out, rc = -ENOMEM);
5581 str = lustre_cfg_string(cfg, 2);
5582 s_flags = simple_strtoul(str, NULL, 0);
5583 str = strstr(str, ":");
5585 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5586 opts = lustre_cfg_string(cfg, 3);
5587 page = (unsigned long)page_address(__page);
5588 options = (char *)page;
5591 strcat(options, "user_xattr,acl");
5593 strcat(options, opts);
5595 /* Glom up mount options */
5596 if (*options != '\0')
5597 strcat(options, ",");
5598 strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
5600 type = get_fs_type("ldiskfs");
5602 CERROR("%s: cannot find ldiskfs module\n", name);
5603 GOTO(out, rc = -ENODEV);
5606 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5607 module_put(type->owner);
5609 if (IS_ERR(o->od_mnt)) {
5610 rc = PTR_ERR(o->od_mnt);
5612 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5616 #ifdef HAVE_DEV_SET_RDONLY
5617 if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5618 CERROR("%s: underlying device %s is marked as read-only. "
5619 "Setup failed\n", name, dev);
5620 GOTO(out_mnt, rc = -EROFS);
5624 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5625 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5626 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5627 GOTO(out_mnt, rc = -EINVAL);
5630 #ifdef LDISKFS_MOUNT_DIRDATA
5631 if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
5632 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5633 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
5635 inode = osd_sb(o)->s_root->d_inode;
5636 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
5637 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
5639 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
5643 if (lmd_flags & LMD_FLG_NOSCRUB)
5654 OBD_PAGE_FREE(__page);
5659 static struct lu_device *osd_device_fini(const struct lu_env *env,
5660 struct lu_device *d)
5662 struct osd_device *o = osd_dev(d);
5665 osd_shutdown(env, o);
5667 osd_scrub_cleanup(env, o);
5668 osd_obj_map_fini(o);
5674 static int osd_device_init0(const struct lu_env *env,
5675 struct osd_device *o,
5676 struct lustre_cfg *cfg)
5678 struct lu_device *l = osd2lu_dev(o);
5679 struct osd_thread_info *info;
5683 /* if the module was re-loaded, env can loose its keys */
5684 rc = lu_env_refill((struct lu_env *) env);
5687 info = osd_oti_get(env);
5690 l->ld_ops = &osd_lu_ops;
5691 o->od_dt_dev.dd_ops = &osd_dt_ops;
5693 spin_lock_init(&o->od_osfs_lock);
5694 mutex_init(&o->od_otable_mutex);
5696 o->od_capa_hash = init_capa_hash();
5697 if (o->od_capa_hash == NULL)
5698 GOTO(out, rc = -ENOMEM);
5700 o->od_read_cache = 1;
5701 o->od_writethrough_cache = 1;
5702 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
5704 rc = osd_mount(env, o, cfg);
5708 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
5709 sizeof(o->od_svname));
5710 if (cplen >= sizeof(o->od_svname)) {
5715 if (server_name_is_ost(o->od_svname))
5718 rc = osd_obj_map_init(env, o);
5722 rc = lu_site_init(&o->od_site, l);
5724 GOTO(out_compat, rc);
5725 o->od_site.ls_bottom_dev = l;
5727 rc = lu_site_init_finish(&o->od_site);
5731 /* self-repair LMA by default */
5732 o->od_lma_self_repair = 1;
5734 CFS_INIT_LIST_HEAD(&o->od_ios_list);
5735 /* setup scrub, including OI files initialization */
5736 rc = osd_scrub_setup(env, o);
5740 rc = osd_procfs_init(o, o->od_svname);
5742 CERROR("%s: can't initialize procfs: rc = %d\n",
5744 GOTO(out_scrub, rc);
5747 LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
5749 /* initialize quota slave instance */
5750 o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
5752 if (IS_ERR(o->od_quota_slave)) {
5753 rc = PTR_ERR(o->od_quota_slave);
5754 o->od_quota_slave = NULL;
5755 GOTO(out_procfs, rc);
5763 osd_scrub_cleanup(env, o);
5765 lu_site_fini(&o->od_site);
5767 osd_obj_map_fini(o);
5771 cleanup_capa_hash(o->od_capa_hash);
5776 static struct lu_device *osd_device_alloc(const struct lu_env *env,
5777 struct lu_device_type *t,
5778 struct lustre_cfg *cfg)
5780 struct osd_device *o;
5785 return ERR_PTR(-ENOMEM);
5787 rc = dt_device_init(&o->od_dt_dev, t);
5789 /* Because the ctx might be revived in dt_device_init,
5790 * refill the env here */
5791 lu_env_refill((struct lu_env *)env);
5792 rc = osd_device_init0(env, o, cfg);
5794 dt_device_fini(&o->od_dt_dev);
5797 if (unlikely(rc != 0))
5800 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
5803 static struct lu_device *osd_device_free(const struct lu_env *env,
5804 struct lu_device *d)
5806 struct osd_device *o = osd_dev(d);
5809 cleanup_capa_hash(o->od_capa_hash);
5810 /* XXX: make osd top device in order to release reference */
5811 d->ld_site->ls_top_dev = d;
5812 lu_site_purge(env, d->ld_site, -1);
5813 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
5814 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
5815 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
5817 lu_site_fini(&o->od_site);
5818 dt_device_fini(&o->od_dt_dev);
5823 static int osd_process_config(const struct lu_env *env,
5824 struct lu_device *d, struct lustre_cfg *cfg)
5826 struct osd_device *o = osd_dev(d);
5830 switch (cfg->lcfg_command) {
5832 rc = osd_mount(env, o, cfg);
5835 lu_dev_del_linkage(d->ld_site, d);
5836 rc = osd_shutdown(env, o);
5839 LASSERT(&o->od_dt_dev);
5840 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
5841 cfg, &o->od_dt_dev);
5842 if (rc > 0 || rc == -ENOSYS)
5843 rc = class_process_proc_param(PARAM_OST,
5844 lprocfs_osd_obd_vars,
5845 cfg, &o->od_dt_dev);
5854 static int osd_recovery_complete(const struct lu_env *env,
5855 struct lu_device *d)
5857 struct osd_device *osd = osd_dev(d);
5861 if (osd->od_quota_slave == NULL)
5864 /* start qsd instance on recovery completion, this notifies the quota
5865 * slave code that we are about to process new requests now */
5866 rc = qsd_start(env, osd->od_quota_slave);
5871 * we use exports to track all osd users
5873 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
5874 struct obd_device *obd, struct obd_uuid *cluuid,
5875 struct obd_connect_data *data, void *localdata)
5877 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
5878 struct lustre_handle conn;
5882 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
5884 rc = class_connect(&conn, obd, cluuid);
5888 *exp = class_conn2export(&conn);
5890 spin_lock(&osd->od_osfs_lock);
5892 spin_unlock(&osd->od_osfs_lock);
5898 * once last export (we don't count self-export) disappeared
5899 * osd can be released
5901 static int osd_obd_disconnect(struct obd_export *exp)
5903 struct obd_device *obd = exp->exp_obd;
5904 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
5905 int rc, release = 0;
5908 /* Only disconnect the underlying layers on the final disconnect. */
5909 spin_lock(&osd->od_osfs_lock);
5911 if (osd->od_connects == 0)
5913 spin_unlock(&osd->od_osfs_lock);
5915 rc = class_disconnect(exp); /* bz 9811 */
5917 if (rc == 0 && release)
5918 class_manual_cleanup(obd);
5922 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
5923 struct lu_device *dev)
5925 struct osd_device *osd = osd_dev(dev);
5929 if (osd->od_quota_slave != NULL)
5930 /* set up quota slave objects */
5931 result = qsd_prepare(env, osd->od_quota_slave);
5936 static const struct lu_object_operations osd_lu_obj_ops = {
5937 .loo_object_init = osd_object_init,
5938 .loo_object_delete = osd_object_delete,
5939 .loo_object_release = osd_object_release,
5940 .loo_object_free = osd_object_free,
5941 .loo_object_print = osd_object_print,
5942 .loo_object_invariant = osd_object_invariant
5945 const struct lu_device_operations osd_lu_ops = {
5946 .ldo_object_alloc = osd_object_alloc,
5947 .ldo_process_config = osd_process_config,
5948 .ldo_recovery_complete = osd_recovery_complete,
5949 .ldo_prepare = osd_prepare,
5952 static const struct lu_device_type_operations osd_device_type_ops = {
5953 .ldto_init = osd_type_init,
5954 .ldto_fini = osd_type_fini,
5956 .ldto_start = osd_type_start,
5957 .ldto_stop = osd_type_stop,
5959 .ldto_device_alloc = osd_device_alloc,
5960 .ldto_device_free = osd_device_free,
5962 .ldto_device_init = osd_device_init,
5963 .ldto_device_fini = osd_device_fini
5966 struct lu_device_type osd_device_type = {
5967 .ldt_tags = LU_DEVICE_DT,
5968 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
5969 .ldt_ops = &osd_device_type_ops,
5970 .ldt_ctx_tags = LCT_LOCAL,
5974 * lprocfs legacy support.
5976 static struct obd_ops osd_obd_device_ops = {
5977 .o_owner = THIS_MODULE,
5978 .o_connect = osd_obd_connect,
5979 .o_disconnect = osd_obd_disconnect
5982 static int __init osd_mod_init(void)
5988 rc = lu_kmem_init(ldiskfs_caches);
5992 rc = class_register_type(&osd_obd_device_ops, NULL, NULL,
5993 #ifndef HAVE_ONLY_PROCFS_SEQ
5994 lprocfs_osd_module_vars,
5996 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
5998 lu_kmem_fini(ldiskfs_caches);
6002 static void __exit osd_mod_exit(void)
6004 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6005 lu_kmem_fini(ldiskfs_caches);
6008 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6009 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6010 MODULE_LICENSE("GPL");
6012 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);