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>
75 #include <ldiskfs/xattr.h>
78 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
79 "ldiskfs with parallel directory operations");
81 int ldiskfs_track_declares_assert;
82 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
83 "LBUG during tracking of declares");
85 /* Slab to allocate dynlocks */
86 struct kmem_cache *dynlock_cachep;
88 static struct lu_kmem_descr ldiskfs_caches[] = {
90 .ckd_cache = &dynlock_cachep,
91 .ckd_name = "dynlock_cache",
92 .ckd_size = sizeof(struct dynlock_handle)
99 static const char dot[] = ".";
100 static const char dotdot[] = "..";
101 static const char remote_obj_dir[] = "REM_OBJ_DIR";
103 static const struct lu_object_operations osd_lu_obj_ops;
104 static const struct dt_object_operations osd_obj_ops;
105 static const struct dt_object_operations osd_obj_ea_ops;
106 static const struct dt_object_operations osd_obj_otable_it_ops;
107 static const struct dt_index_operations osd_index_iam_ops;
108 static const struct dt_index_operations osd_index_ea_ops;
110 int osd_trans_declare_op2rb[] = {
111 [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
112 [OSD_OT_PUNCH] = OSD_OT_MAX,
113 [OSD_OT_XATTR_SET] = OSD_OT_XATTR_SET,
114 [OSD_OT_CREATE] = OSD_OT_DESTROY,
115 [OSD_OT_DESTROY] = OSD_OT_CREATE,
116 [OSD_OT_REF_ADD] = OSD_OT_REF_DEL,
117 [OSD_OT_REF_DEL] = OSD_OT_REF_ADD,
118 [OSD_OT_WRITE] = OSD_OT_WRITE,
119 [OSD_OT_INSERT] = OSD_OT_DELETE,
120 [OSD_OT_DELETE] = OSD_OT_INSERT,
121 [OSD_OT_UPDATE] = OSD_OT_MAX,
122 [OSD_OT_QUOTA] = OSD_OT_MAX,
125 static int osd_has_index(const struct osd_object *obj)
127 return obj->oo_dt.do_index_ops != NULL;
130 static int osd_object_invariant(const struct lu_object *l)
132 return osd_invariant(osd_obj(l));
136 * Concurrency: doesn't matter
140 * Concurrency: doesn't matter
142 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
144 struct osd_thread_info *oti = osd_oti_get(env);
145 return oti->oti_w_locks > 0 && o->oo_owner == env;
149 * Concurrency: doesn't access mutable data
151 static int osd_root_get(const struct lu_env *env,
152 struct dt_device *dev, struct lu_fid *f)
154 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
159 * OSD object methods.
163 * Concurrency: no concurrent access is possible that early in object
166 static struct lu_object *osd_object_alloc(const struct lu_env *env,
167 const struct lu_object_header *hdr,
170 struct osd_object *mo;
176 l = &mo->oo_dt.do_lu;
177 dt_object_init(&mo->oo_dt, NULL, d);
178 mo->oo_dt.do_ops = &osd_obj_ea_ops;
179 l->lo_ops = &osd_lu_obj_ops;
180 init_rwsem(&mo->oo_sem);
181 init_rwsem(&mo->oo_ext_idx_sem);
182 spin_lock_init(&mo->oo_guard);
189 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
190 struct dentry *dentry, struct lustre_mdt_attrs *lma)
194 CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
195 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
196 info->oti_mdt_attrs_old, LMA_OLD_SIZE);
198 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
199 memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
201 lustre_lma_swab(lma);
202 /* Check LMA compatibility */
203 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
204 CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
205 "for fid = "DFID", ino = %lu\n",
206 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
207 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
208 PFID(&lma->lma_self_fid), inode->i_ino);
211 } else if (rc == 0) {
219 * retrieve object from backend ext fs.
221 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
222 struct osd_inode_id *id)
224 struct inode *inode = NULL;
226 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
228 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
229 id->oii_ino, PTR_ERR(inode));
230 } else if (id->oii_gen != OSD_OII_NOGEN &&
231 inode->i_generation != id->oii_gen) {
232 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
233 "i_generation = %u\n",
234 id->oii_ino, id->oii_gen, inode->i_generation);
236 inode = ERR_PTR(-ESTALE);
237 } else if (inode->i_nlink == 0) {
238 /* due to parallel readdir and unlink,
239 * we can have dead inode here. */
240 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
242 inode = ERR_PTR(-ESTALE);
243 } else if (is_bad_inode(inode)) {
244 CWARN("%.16s: bad inode: ino = %u\n",
245 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
247 inode = ERR_PTR(-ENOENT);
249 if (id->oii_gen == OSD_OII_NOGEN)
250 osd_id_gen(id, inode->i_ino, inode->i_generation);
252 /* Do not update file c/mtime in ldiskfs.
253 * NB: we don't have any lock to protect this because we don't
254 * have reference on osd_object now, but contention with
255 * another lookup + attr_set can't happen in the tiny window
256 * between if (...) and set S_NOCMTIME. */
257 if (!(inode->i_flags & S_NOCMTIME))
258 inode->i_flags |= S_NOCMTIME;
263 static struct inode *
264 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
265 struct osd_inode_id *id, struct lu_fid *fid)
267 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
271 inode = osd_iget(info, dev, id);
275 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
277 *fid = lma->lma_self_fid;
278 } else if (rc == -ENODATA) {
279 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
280 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
282 lu_igif_build(fid, inode->i_ino, inode->i_generation);
290 static struct inode *osd_iget_check(struct osd_thread_info *info,
291 struct osd_device *dev,
292 const struct lu_fid *fid,
293 struct osd_inode_id *id,
300 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
303 if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) {
304 CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
313 if (is_bad_inode(inode)) {
316 CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
324 if (id->oii_gen != OSD_OII_NOGEN &&
325 inode->i_generation != id->oii_gen) {
328 CDEBUG(D_INODE, "unmatched inode: ino = %u, "
329 "oii_gen = %u, i_generation = %u\n",
330 id->oii_ino, id->oii_gen, inode->i_generation);
338 if (inode->i_nlink == 0) {
341 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
351 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
353 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
354 /* XXX: There are some possible cases:
356 * Backup/restore caused the OI invalid.
358 * Someone unlinked the object but NOT removed
359 * the OI mapping, such as mount target device
360 * as ldiskfs, and modify something directly.
362 * Someone just removed the object between the
363 * former oi_lookup and the iget. It is normal.
364 * 4. Other failure cases.
366 * Generally, when the device is mounted, it will
367 * auto check whether the system is restored from
368 * file-level backup or not. We trust such detect
369 * to distinguish the 1st case from the 2nd case. */
371 if (!IS_ERR(inode) && inode->i_generation != 0 &&
372 inode->i_generation == id->oii_gen)
378 if (id->oii_gen == OSD_OII_NOGEN)
379 osd_id_gen(id, inode->i_ino, inode->i_generation);
381 /* Do not update file c/mtime in ldiskfs.
382 * NB: we don't have any lock to protect this because we don't
383 * have reference on osd_object now, but contention with
384 * another lookup + attr_set can't happen in the tiny window
385 * between if (...) and set S_NOCMTIME. */
386 if (!(inode->i_flags & S_NOCMTIME))
387 inode->i_flags |= S_NOCMTIME;
404 * \retval +v: new filter_fid, does not contain self-fid
405 * \retval 0: filter_fid_old, contains self-fid
406 * \retval -v: other failure cases
408 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
409 struct dentry *dentry, struct lu_fid *fid)
411 struct filter_fid_old *ff = &info->oti_ff;
412 struct ost_id *ostid = &info->oti_ostid;
415 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
416 if (rc == sizeof(*ff)) {
418 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
419 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
420 /* XXX: should use real OST index in the future. LU-3569 */
421 ostid_to_fid(fid, ostid, 0);
422 } else if (rc == sizeof(struct filter_fid)) {
424 } else if (rc >= 0) {
431 static int osd_lma_self_repair(struct osd_thread_info *info,
432 struct osd_device *osd, struct inode *inode,
433 const struct lu_fid *fid, __u32 compat)
438 LASSERT(current->journal_info == NULL);
440 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
441 osd_dto_credits_noquota[DTO_XATTR_SET]);
444 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
449 rc = osd_ea_fid_set(info, inode, fid, compat, 0);
451 CWARN("%s: cannot self repair the LMA: rc = %d\n",
453 ldiskfs_journal_stop(jh);
457 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
459 struct osd_thread_info *info = osd_oti_get(env);
460 struct osd_device *osd = osd_obj2dev(obj);
461 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
462 struct inode *inode = obj->oo_inode;
463 struct dentry *dentry = &info->oti_obj_dentry;
464 struct lu_fid *fid = NULL;
465 const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
469 CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
470 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
471 info->oti_mdt_attrs_old, LMA_OLD_SIZE);
472 if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
473 fid = &lma->lma_self_fid;
474 rc = osd_get_idif(info, inode, dentry, fid);
475 if ((rc > 0) || (rc == -ENODATA && osd->od_lma_self_repair)) {
476 /* For the given OST-object, if it has neither LMA nor
477 * FID in XATTR_NAME_FID, then the given FID (which is
478 * contained in the @obj, from client RPC for locating
479 * the OST-object) is trusted. We use it to generate
481 osd_lma_self_repair(info, osd, inode, rfid,
482 fid_is_on_ost(info, osd, fid, OI_CHECK_FLD) ?
483 LMAC_FID_ON_OST : 0);
488 if (unlikely(rc == -ENODATA))
496 lustre_lma_swab(lma);
497 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
498 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
499 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
500 "fid = "DFID", ino = %lu\n", osd_name(osd),
501 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
502 PFID(rfid), inode->i_ino);
504 } else if (!(lma->lma_compat & LMAC_NOT_IN_OI)) {
505 fid = &lma->lma_self_fid;
509 if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
510 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
511 struct ost_id *oi = &info->oti_ostid;
512 struct lu_fid *fid1 = &info->oti_fid3;
513 __u32 idx = fid_idif_ost_idx(rfid);
515 /* For old IDIF, the OST index is not part of the IDIF,
516 * Means that different OSTs may have the same IDIFs.
517 * Under such case, we need to make some compatible
518 * check to make sure to trigger OI scrub properly. */
519 if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
520 /* Given @rfid is new, LMA is old. */
521 fid_to_ostid(fid, oi);
522 ostid_to_fid(fid1, oi, idx);
523 if (lu_fid_eq(fid1, rfid)) {
524 if (osd->od_lma_self_repair)
525 osd_lma_self_repair(info, osd,
533 CDEBUG(D_INODE, "%s: FID "DFID" != self_fid "DFID"\n",
534 osd_name(osd), PFID(rfid), PFID(fid));
541 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
542 const struct lu_fid *fid,
543 const struct lu_object_conf *conf)
545 struct osd_thread_info *info;
546 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
547 struct osd_device *dev;
548 struct osd_idmap_cache *oic;
549 struct osd_inode_id *id;
551 struct osd_scrub *scrub;
552 struct scrub_file *sf;
556 bool triggered = false;
559 LINVRNT(osd_invariant(obj));
560 LASSERT(obj->oo_inode == NULL);
561 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
564 scrub = &dev->od_scrub;
565 sf = &scrub->os_file;
566 info = osd_oti_get(env);
568 oic = &info->oti_cache;
570 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
573 /* For the object is created as locking anchor, or for the object to
574 * be created on disk. No need to osd_oi_lookup() at here because FID
575 * shouldn't never be re-used, if it's really a duplicate FID from
576 * unexpected reason, we should be able to detect it later by calling
577 * do_create->osd_oi_insert(). */
578 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
579 GOTO(out, result = 0);
581 /* Search order: 1. per-thread cache. */
582 if (lu_fid_eq(fid, &oic->oic_fid) &&
583 likely(oic->oic_dev == dev)) {
589 if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
590 /* Search order: 2. OI scrub pending list. */
591 result = osd_oii_lookup(dev, fid, id);
596 /* Search order: 3. OI files. */
597 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
598 if (result == -ENOENT) {
599 if (!fid_is_norm(fid) ||
600 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
601 !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
603 GOTO(out, result = 0);
614 inode = osd_iget_check(info, dev, fid, id, in_oi);
616 result = PTR_ERR(inode);
617 if (result == -ENOENT || result == -ESTALE) {
619 fid_zero(&oic->oic_fid);
621 GOTO(out, result = -ENOENT);
622 } else if (result == -EREMCHG) {
626 fid_zero(&oic->oic_fid);
628 if (unlikely(triggered))
629 GOTO(out, result = saved);
632 if (thread_is_running(&scrub->os_thread)) {
633 result = -EINPROGRESS;
634 } else if (!dev->od_noscrub) {
635 result = osd_scrub_start(dev);
636 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
637 "for "DFID", rc = %d [1]\n",
638 osd_name(dev), PFID(fid),result);
639 if (result == 0 || result == -EALREADY)
640 result = -EINPROGRESS;
645 /* We still have chance to get the valid inode: for the
646 * object which is referenced by remote name entry, the
647 * object on the local MDT will be linked under the dir
648 * of "/REMOTE_PARENT_DIR" with its FID string as name.
650 * We do not know whether the object for the given FID
651 * is referenced by some remote name entry or not, and
652 * especially for DNE II, a multiple-linked object may
653 * have many name entries reside on many MDTs.
655 * To simplify the operation, OSD will not distinguish
656 * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
657 * only happened for the RPC from other MDT during the
658 * OI scrub, or for the client side RPC with FID only,
659 * such as FID to path, or from old connected client. */
661 result = osd_lookup_in_remote_parent(info, dev,
674 obj->oo_inode = inode;
675 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
677 result = osd_check_lma(env, obj);
680 obj->oo_inode = NULL;
681 if (result == -EREMCHG) {
683 result = osd_oi_lookup(info, dev, fid, id,
686 fid_zero(&oic->oic_fid);
697 obj->oo_compat_dot_created = 1;
698 obj->oo_compat_dotdot_created = 1;
700 if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
701 GOTO(out, result = 0);
703 LASSERT(obj->oo_hl_head == NULL);
704 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
705 if (obj->oo_hl_head == NULL) {
706 obj->oo_inode = NULL;
708 GOTO(out, result = -ENOMEM);
710 GOTO(out, result = 0);
713 LINVRNT(osd_invariant(obj));
718 * Concurrency: shouldn't matter.
720 static void osd_object_init0(struct osd_object *obj)
722 LASSERT(obj->oo_inode != NULL);
723 obj->oo_dt.do_body_ops = &osd_body_ops;
724 obj->oo_dt.do_lu.lo_header->loh_attr |=
725 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
729 * Concurrency: no concurrent access is possible that early in object
732 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
733 const struct lu_object_conf *conf)
735 struct osd_object *obj = osd_obj(l);
738 LINVRNT(osd_invariant(obj));
740 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
741 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
742 l->lo_header->loh_attr |= LOHA_EXISTS;
746 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
747 obj->oo_dt.do_body_ops = &osd_body_ops_new;
748 if (result == 0 && obj->oo_inode != NULL)
749 osd_object_init0(obj);
751 LINVRNT(osd_invariant(obj));
756 * Concurrency: no concurrent access is possible that late in object
759 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
761 struct osd_object *obj = osd_obj(l);
763 LINVRNT(osd_invariant(obj));
765 dt_object_fini(&obj->oo_dt);
766 if (obj->oo_hl_head != NULL)
767 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
772 * Concurrency: no concurrent access is possible that late in object
775 static void osd_index_fini(struct osd_object *o)
777 struct iam_container *bag;
779 if (o->oo_dir != NULL) {
780 bag = &o->oo_dir->od_container;
781 if (o->oo_inode != NULL) {
782 if (bag->ic_object == o->oo_inode)
783 iam_container_fini(bag);
785 OBD_FREE_PTR(o->oo_dir);
791 * Concurrency: no concurrent access is possible that late in object
792 * life-cycle (for all existing callers, that is. New callers have to provide
793 * their own locking.)
795 static int osd_inode_unlinked(const struct inode *inode)
797 return inode->i_nlink == 0;
801 OSD_TXN_OI_DELETE_CREDITS = 20,
802 OSD_TXN_INODE_DELETE_CREDITS = 20
809 #if OSD_THANDLE_STATS
811 * Set time when the handle is allocated
813 static void osd_th_alloced(struct osd_thandle *oth)
815 oth->oth_alloced = cfs_time_current();
819 * Set time when the handle started
821 static void osd_th_started(struct osd_thandle *oth)
823 oth->oth_started = cfs_time_current();
827 * Helper function to convert time interval to microseconds packed in
830 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
834 cfs_duration_usec(cfs_time_sub(end, start), &val);
835 return val.tv_sec * 1000000 + val.tv_usec;
839 * Check whether the we deal with this handle for too long.
841 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
842 cfs_time_t alloced, cfs_time_t started,
845 cfs_time_t now = cfs_time_current();
847 LASSERT(dev != NULL);
849 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
850 interval_to_usec(alloced, started));
851 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
852 interval_to_usec(started, closed));
853 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
854 interval_to_usec(closed, now));
856 if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
857 CWARN("transaction handle %p was open for too long: "
859 "alloced "CFS_TIME_T" ,"
860 "started "CFS_TIME_T" ,"
861 "closed "CFS_TIME_T"\n",
862 oth, now, alloced, started, closed);
863 libcfs_debug_dumpstack(NULL);
867 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
869 cfs_time_t __closed = cfs_time_current(); \
870 cfs_time_t __alloced = oth->oth_alloced; \
871 cfs_time_t __started = oth->oth_started; \
874 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
877 #else /* OSD_THANDLE_STATS */
879 #define osd_th_alloced(h) do {} while(0)
880 #define osd_th_started(h) do {} while(0)
881 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
883 #endif /* OSD_THANDLE_STATS */
886 * Concurrency: doesn't access mutable data.
888 static int osd_param_is_not_sane(const struct osd_device *dev,
889 const struct thandle *th)
891 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
893 return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers;
897 * Concurrency: shouldn't matter.
899 static void osd_trans_commit_cb(struct super_block *sb,
900 struct ldiskfs_journal_cb_entry *jcb, int error)
902 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
903 struct thandle *th = &oh->ot_super;
904 struct lu_device *lud = &th->th_dev->dd_lu_dev;
905 struct dt_txn_commit_cb *dcb, *tmp;
907 LASSERT(oh->ot_handle == NULL);
910 CERROR("transaction @0x%p commit error: %d\n", th, error);
912 dt_txn_hook_commit(th);
914 /* call per-transaction callbacks if any */
915 cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
916 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
917 "commit callback entry: magic=%x name='%s'\n",
918 dcb->dcb_magic, dcb->dcb_name);
919 cfs_list_del_init(&dcb->dcb_linkage);
920 dcb->dcb_func(NULL, th, dcb, error);
923 lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
927 lu_context_exit(&th->th_ctx);
928 lu_context_fini(&th->th_ctx);
932 static struct thandle *osd_trans_create(const struct lu_env *env,
935 struct osd_thread_info *oti = osd_oti_get(env);
936 struct osd_iobuf *iobuf = &oti->oti_iobuf;
937 struct osd_thandle *oh;
941 /* on pending IO in this thread should left from prev. request */
942 LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
944 th = ERR_PTR(-ENOMEM);
945 OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
947 oh->ot_quota_trans = &oti->oti_quota_trans;
948 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
952 th->th_tags = LCT_TX_HANDLE;
954 atomic_set(&th->th_refc, 1);
955 th->th_alloc_size = sizeof(*oh);
956 oti->oti_dev = osd_dt_dev(d);
957 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
960 memset(oti->oti_declare_ops, 0,
961 sizeof(oti->oti_declare_ops));
962 memset(oti->oti_declare_ops_rb, 0,
963 sizeof(oti->oti_declare_ops_rb));
964 memset(oti->oti_declare_ops_cred, 0,
965 sizeof(oti->oti_declare_ops_cred));
966 oti->oti_rollback = false;
972 * Concurrency: shouldn't matter.
974 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
977 struct osd_thread_info *oti = osd_oti_get(env);
978 struct osd_device *dev = osd_dt_dev(d);
980 struct osd_thandle *oh;
985 LASSERT(current->journal_info == NULL);
987 oh = container_of0(th, struct osd_thandle, ot_super);
989 LASSERT(oh->ot_handle == NULL);
991 rc = dt_txn_hook_start(env, d, th);
995 if (unlikely(osd_param_is_not_sane(dev, th))) {
996 static unsigned long last_printed;
997 static int last_credits;
999 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1000 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1002 osd_journal(dev)->j_max_transaction_buffers);
1003 CWARN(" create: %u/%u, destroy: %u/%u\n",
1004 oti->oti_declare_ops[OSD_OT_CREATE],
1005 oti->oti_declare_ops_cred[OSD_OT_CREATE],
1006 oti->oti_declare_ops[OSD_OT_DESTROY],
1007 oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
1008 CWARN(" attr_set: %u/%u, xattr_set: %u/%u\n",
1009 oti->oti_declare_ops[OSD_OT_ATTR_SET],
1010 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1011 oti->oti_declare_ops[OSD_OT_XATTR_SET],
1012 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
1013 CWARN(" write: %u/%u, punch: %u/%u, quota %u/%u\n",
1014 oti->oti_declare_ops[OSD_OT_WRITE],
1015 oti->oti_declare_ops_cred[OSD_OT_WRITE],
1016 oti->oti_declare_ops[OSD_OT_PUNCH],
1017 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1018 oti->oti_declare_ops[OSD_OT_QUOTA],
1019 oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
1020 CWARN(" insert: %u/%u, delete: %u/%u\n",
1021 oti->oti_declare_ops[OSD_OT_INSERT],
1022 oti->oti_declare_ops_cred[OSD_OT_INSERT],
1023 oti->oti_declare_ops[OSD_OT_DELETE],
1024 oti->oti_declare_ops_cred[OSD_OT_DELETE]);
1025 CWARN(" ref_add: %u/%u, ref_del: %u/%u\n",
1026 oti->oti_declare_ops[OSD_OT_REF_ADD],
1027 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1028 oti->oti_declare_ops[OSD_OT_REF_DEL],
1029 oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
1031 if (last_credits != oh->ot_credits &&
1032 time_after(jiffies, last_printed + 60 * HZ)) {
1033 libcfs_debug_dumpstack(NULL);
1034 last_credits = oh->ot_credits;
1035 last_printed = jiffies;
1037 /* XXX Limit the credits to 'max_transaction_buffers', and
1038 * let the underlying filesystem to catch the error if
1039 * we really need so many credits.
1041 * This should be removed when we can calculate the
1042 * credits precisely. */
1043 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
1047 * XXX temporary stuff. Some abstraction layer should
1050 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1054 LASSERT(oti->oti_txns == 0);
1055 lu_context_init(&th->th_ctx, th->th_tags);
1056 lu_context_enter(&th->th_ctx);
1058 lu_device_get(&d->dd_lu_dev);
1059 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1070 static int osd_seq_exists(const struct lu_env *env,
1071 struct osd_device *osd, obd_seq seq)
1073 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1074 struct seq_server_site *ss = osd_seq_site(osd);
1081 rc = osd_fld_lookup(env, osd, seq, range);
1084 CERROR("%s: can't lookup FLD sequence "LPX64
1085 ": rc = %d\n", osd_name(osd), seq, rc);
1089 RETURN(ss->ss_node_id == range->lsr_index);
1093 * Concurrency: shouldn't matter.
1095 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1099 struct osd_thandle *oh;
1100 struct osd_thread_info *oti = osd_oti_get(env);
1101 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1102 struct qsd_instance *qsd = oti->oti_dev->od_quota_slave;
1103 struct lquota_trans *qtrans;
1106 oh = container_of0(th, struct osd_thandle, ot_super);
1108 qtrans = oh->ot_quota_trans;
1109 oh->ot_quota_trans = NULL;
1111 if (oh->ot_handle != NULL) {
1112 handle_t *hdl = oh->ot_handle;
1115 * add commit callback
1116 * notice we don't do this in osd_trans_start()
1117 * as underlying transaction can change during truncate
1119 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1122 LASSERT(oti->oti_txns == 1);
1124 rc = dt_txn_hook_stop(env, th);
1126 CERROR("Failure in transaction hook: %d\n", rc);
1128 /* hook functions might modify th_sync */
1129 hdl->h_sync = th->th_sync;
1131 oh->ot_handle = NULL;
1132 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
1133 rc = ldiskfs_journal_stop(hdl));
1135 CERROR("Failure to stop transaction: %d\n", rc);
1137 thandle_put(&oh->ot_super);
1140 /* inform the quota slave device that the transaction is stopping */
1141 qsd_op_end(env, qsd, qtrans);
1143 /* as we want IO to journal and data IO be concurrent, we don't block
1144 * awaiting data IO completion in osd_do_bio(), instead we wait here
1145 * once transaction is submitted to the journal. all reqular requests
1146 * don't do direct IO (except read/write), thus this wait_event becomes
1149 * IMPORTANT: we have to wait till any IO submited by the thread is
1150 * completed otherwise iobuf may be corrupted by different request
1152 wait_event(iobuf->dr_wait,
1153 atomic_read(&iobuf->dr_numreqs) == 0);
1155 rc = iobuf->dr_error;
1160 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1162 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1165 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1166 LASSERT(&dcb->dcb_func != NULL);
1167 cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1173 * Called just before object is freed. Releases all resources except for
1174 * object itself (that is released by osd_object_free()).
1176 * Concurrency: no concurrent access is possible that late in object
1179 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1181 struct osd_object *obj = osd_obj(l);
1182 struct inode *inode = obj->oo_inode;
1184 LINVRNT(osd_invariant(obj));
1187 * If object is unlinked remove fid->ino mapping from object index.
1190 osd_index_fini(obj);
1191 if (inode != NULL) {
1192 struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave;
1193 qid_t uid = inode->i_uid;
1194 qid_t gid = inode->i_gid;
1197 obj->oo_inode = NULL;
1200 struct osd_thread_info *info = osd_oti_get(env);
1201 struct lquota_id_info *qi = &info->oti_qi;
1203 /* Release granted quota to master if necessary */
1204 qi->lqi_id.qid_uid = uid;
1205 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1207 qi->lqi_id.qid_uid = gid;
1208 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1214 * Concurrency: ->loo_object_release() is called under site spin-lock.
1216 static void osd_object_release(const struct lu_env *env,
1217 struct lu_object *l)
1222 * Concurrency: shouldn't matter.
1224 static int osd_object_print(const struct lu_env *env, void *cookie,
1225 lu_printer_t p, const struct lu_object *l)
1227 struct osd_object *o = osd_obj(l);
1228 struct iam_descr *d;
1230 if (o->oo_dir != NULL)
1231 d = o->oo_dir->od_container.ic_descr;
1234 return (*p)(env, cookie,
1235 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1237 o->oo_inode ? o->oo_inode->i_ino : 0UL,
1238 o->oo_inode ? o->oo_inode->i_generation : 0,
1239 d ? d->id_ops->id_name : "plain");
1242 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1245 * Concurrency: shouldn't matter.
1247 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1248 struct obd_statfs *sfs)
1250 struct osd_device *osd = osd_dt_dev(d);
1251 struct super_block *sb = osd_sb(osd);
1252 struct kstatfs *ksfs;
1255 if (unlikely(osd->od_mnt == NULL))
1256 return -EINPROGRESS;
1258 /* osd_lproc.c call this without env, allocate ksfs for that case */
1259 if (unlikely(env == NULL)) {
1260 OBD_ALLOC_PTR(ksfs);
1264 ksfs = &osd_oti_get(env)->oti_ksfs;
1267 spin_lock(&osd->od_osfs_lock);
1268 result = sb->s_op->statfs(sb->s_root, ksfs);
1269 if (likely(result == 0)) { /* N.B. statfs can't really fail */
1270 statfs_pack(sfs, ksfs);
1271 if (sb->s_flags & MS_RDONLY)
1272 sfs->os_state = OS_STATE_READONLY;
1275 spin_unlock(&osd->od_osfs_lock);
1277 if (unlikely(env == NULL))
1280 /* Reserve a small amount of space for local objects like last_rcvd,
1281 * llog, quota files, ... */
1282 if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1285 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1286 /** Take out metadata overhead for indirect blocks */
1287 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1294 * Estimate space needed for file creations. We assume the largest filename
1295 * which is 2^64 - 1, hence a filename of 20 chars.
1296 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1298 #ifdef __LDISKFS_DIR_REC_LEN
1299 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1301 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1305 * Concurrency: doesn't access mutable data.
1307 static void osd_conf_get(const struct lu_env *env,
1308 const struct dt_device *dev,
1309 struct dt_device_param *param)
1311 struct super_block *sb = osd_sb(osd_dt_dev(dev));
1315 * XXX should be taken from not-yet-existing fs abstraction layer.
1317 param->ddp_max_name_len = LDISKFS_NAME_LEN;
1318 param->ddp_max_nlink = LDISKFS_LINK_MAX;
1319 param->ddp_block_shift = sb->s_blocksize_bits;
1320 param->ddp_mount_type = LDD_MT_LDISKFS;
1321 param->ddp_maxbytes = sb->s_maxbytes;
1322 /* Overhead estimate should be fairly accurate, so we really take a tiny
1323 * error margin which also avoids fragmenting the filesystem too much */
1324 param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1325 /* inode are statically allocated, so per-inode space consumption
1326 * is the space consumed by the directory entry */
1327 param->ddp_inodespace = PER_OBJ_USAGE;
1328 /* per-fragment overhead to be used by the client code */
1329 param->ddp_grant_frag = 6 * LDISKFS_BLOCK_SIZE(sb);
1330 param->ddp_mntopts = 0;
1331 if (test_opt(sb, XATTR_USER))
1332 param->ddp_mntopts |= MNTOPT_USERXATTR;
1333 if (test_opt(sb, POSIX_ACL))
1334 param->ddp_mntopts |= MNTOPT_ACL;
1336 /* LOD might calculate the max stripe count based on max_ea_size,
1337 * so we need take account in the overhead as well,
1338 * xattr_header + magic + xattr_entry_head */
1339 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1340 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1342 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1343 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1344 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1348 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1352 * Concurrency: shouldn't matter.
1354 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1356 CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1357 return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1361 * Start commit for OSD device.
1363 * An implementation of dt_commit_async method for OSD device.
1364 * Asychronously starts underlayng fs sync and thereby a transaction
1367 * \param env environment
1368 * \param d dt device
1370 * \see dt_device_operations
1372 static int osd_commit_async(const struct lu_env *env,
1373 struct dt_device *d)
1375 struct super_block *s = osd_sb(osd_dt_dev(d));
1378 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1379 RETURN(s->s_op->sync_fs(s, 0));
1383 * Concurrency: shouldn't matter.
1386 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1388 struct super_block *sb = osd_sb(osd_dt_dev(d));
1389 struct block_device *dev = sb->s_bdev;
1390 #ifdef HAVE_DEV_SET_RDONLY
1391 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1394 int rc = -EOPNOTSUPP;
1398 #ifdef HAVE_DEV_SET_RDONLY
1399 CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1401 if (jdev && (jdev != dev)) {
1402 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1404 dev_set_rdonly(jdev);
1406 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1407 dev_set_rdonly(dev);
1409 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1410 osd_dt_dev(d)->od_svname, (long)dev, rc);
1416 * Concurrency: serialization provided by callers.
1418 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1419 int mode, unsigned long timeout, __u32 alg,
1420 struct lustre_capa_key *keys)
1422 struct osd_device *dev = osd_dt_dev(d);
1425 dev->od_fl_capa = mode;
1426 dev->od_capa_timeout = timeout;
1427 dev->od_capa_alg = alg;
1428 dev->od_capa_keys = keys;
1433 * Note: we do not count into QUOTA here.
1434 * If we mount with --data_journal we may need more.
1436 const int osd_dto_credits_noquota[DTO_NR] = {
1439 * INDEX_EXTRA_TRANS_BLOCKS(8) +
1440 * SINGLEDATA_TRANS_BLOCKS(8)
1441 * XXX Note: maybe iam need more, since iam have more level than
1444 [DTO_INDEX_INSERT] = 16,
1447 * just modify a single entry, probably merge few within a block
1449 [DTO_INDEX_DELETE] = 1,
1453 [DTO_INDEX_UPDATE] = 16,
1455 * 4(inode, inode bits, groups, GDT)
1456 * notice: OI updates are counted separately with DTO_INDEX_INSERT
1458 [DTO_OBJECT_CREATE] = 4,
1460 * 4(inode, inode bits, groups, GDT)
1461 * notice: OI updates are counted separately with DTO_INDEX_DELETE
1463 [DTO_OBJECT_DELETE] = 4,
1465 * Attr set credits (inode)
1467 [DTO_ATTR_SET_BASE] = 1,
1469 * Xattr set. The same as xattr of EXT3.
1470 * DATA_TRANS_BLOCKS(14)
1471 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1472 * are also counted in. Do not know why?
1474 [DTO_XATTR_SET] = 14,
1476 * credits for inode change during write.
1478 [DTO_WRITE_BASE] = 3,
1480 * credits for single block write.
1482 [DTO_WRITE_BLOCK] = 14,
1484 * Attr set credits for chown.
1485 * This is extra credits for setattr, and it is null without quota
1487 [DTO_ATTR_SET_CHOWN] = 0
1490 static const struct dt_device_operations osd_dt_ops = {
1491 .dt_root_get = osd_root_get,
1492 .dt_statfs = osd_statfs,
1493 .dt_trans_create = osd_trans_create,
1494 .dt_trans_start = osd_trans_start,
1495 .dt_trans_stop = osd_trans_stop,
1496 .dt_trans_cb_add = osd_trans_cb_add,
1497 .dt_conf_get = osd_conf_get,
1498 .dt_sync = osd_sync,
1500 .dt_commit_async = osd_commit_async,
1501 .dt_init_capa_ctxt = osd_init_capa_ctxt,
1504 static void osd_object_read_lock(const struct lu_env *env,
1505 struct dt_object *dt, unsigned role)
1507 struct osd_object *obj = osd_dt_obj(dt);
1508 struct osd_thread_info *oti = osd_oti_get(env);
1510 LINVRNT(osd_invariant(obj));
1512 LASSERT(obj->oo_owner != env);
1513 down_read_nested(&obj->oo_sem, role);
1515 LASSERT(obj->oo_owner == NULL);
1519 static void osd_object_write_lock(const struct lu_env *env,
1520 struct dt_object *dt, unsigned role)
1522 struct osd_object *obj = osd_dt_obj(dt);
1523 struct osd_thread_info *oti = osd_oti_get(env);
1525 LINVRNT(osd_invariant(obj));
1527 LASSERT(obj->oo_owner != env);
1528 down_write_nested(&obj->oo_sem, role);
1530 LASSERT(obj->oo_owner == NULL);
1531 obj->oo_owner = env;
1535 static void osd_object_read_unlock(const struct lu_env *env,
1536 struct dt_object *dt)
1538 struct osd_object *obj = osd_dt_obj(dt);
1539 struct osd_thread_info *oti = osd_oti_get(env);
1541 LINVRNT(osd_invariant(obj));
1543 LASSERT(oti->oti_r_locks > 0);
1545 up_read(&obj->oo_sem);
1548 static void osd_object_write_unlock(const struct lu_env *env,
1549 struct dt_object *dt)
1551 struct osd_object *obj = osd_dt_obj(dt);
1552 struct osd_thread_info *oti = osd_oti_get(env);
1554 LINVRNT(osd_invariant(obj));
1556 LASSERT(obj->oo_owner == env);
1557 LASSERT(oti->oti_w_locks > 0);
1559 obj->oo_owner = NULL;
1560 up_write(&obj->oo_sem);
1563 static int osd_object_write_locked(const struct lu_env *env,
1564 struct dt_object *dt)
1566 struct osd_object *obj = osd_dt_obj(dt);
1568 LINVRNT(osd_invariant(obj));
1570 return obj->oo_owner == env;
1573 static int capa_is_sane(const struct lu_env *env,
1574 struct osd_device *dev,
1575 struct lustre_capa *capa,
1576 struct lustre_capa_key *keys)
1578 struct osd_thread_info *oti = osd_oti_get(env);
1579 struct lustre_capa *tcapa = &oti->oti_capa;
1580 struct obd_capa *oc;
1584 oc = capa_lookup(dev->od_capa_hash, capa, 0);
1586 if (capa_is_expired(oc)) {
1587 DEBUG_CAPA(D_ERROR, capa, "expired");
1594 if (capa_is_expired_sec(capa)) {
1595 DEBUG_CAPA(D_ERROR, capa, "expired");
1599 spin_lock(&capa_lock);
1600 for (i = 0; i < 2; i++) {
1601 if (keys[i].lk_keyid == capa->lc_keyid) {
1602 oti->oti_capa_key = keys[i];
1606 spin_unlock(&capa_lock);
1609 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1613 rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1617 if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1618 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1622 oc = capa_add(dev->od_capa_hash, capa);
1628 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1629 struct lustre_capa *capa, __u64 opc)
1631 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1632 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
1633 struct lu_capainfo *lci;
1636 if (!osd->od_fl_capa)
1639 if (capa == BYPASS_CAPA)
1642 lci = lu_capainfo_get(env);
1643 if (unlikely(lci == NULL))
1646 if (lci->lci_auth == LC_ID_NONE)
1650 CERROR("%s: no capability provided for FID "DFID": rc = %d\n",
1651 osd_name(osd), PFID(fid), -EACCES);
1655 if (!lu_fid_eq(fid, &capa->lc_fid)) {
1656 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1661 if (!capa_opc_supported(capa, opc)) {
1662 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1666 rc = capa_is_sane(env, osd, capa, osd->od_capa_keys);
1668 DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc);
1675 static struct timespec *osd_inode_time(const struct lu_env *env,
1676 struct inode *inode, __u64 seconds)
1678 struct osd_thread_info *oti = osd_oti_get(env);
1679 struct timespec *t = &oti->oti_time;
1681 t->tv_sec = seconds;
1683 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1688 static void osd_inode_getattr(const struct lu_env *env,
1689 struct inode *inode, struct lu_attr *attr)
1691 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1692 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1693 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1696 attr->la_atime = LTIME_S(inode->i_atime);
1697 attr->la_mtime = LTIME_S(inode->i_mtime);
1698 attr->la_ctime = LTIME_S(inode->i_ctime);
1699 attr->la_mode = inode->i_mode;
1700 attr->la_size = i_size_read(inode);
1701 attr->la_blocks = inode->i_blocks;
1702 attr->la_uid = inode->i_uid;
1703 attr->la_gid = inode->i_gid;
1704 attr->la_flags = LDISKFS_I(inode)->i_flags;
1705 attr->la_nlink = inode->i_nlink;
1706 attr->la_rdev = inode->i_rdev;
1707 attr->la_blksize = 1 << inode->i_blkbits;
1708 attr->la_blkbits = inode->i_blkbits;
1711 static int osd_attr_get(const struct lu_env *env,
1712 struct dt_object *dt,
1713 struct lu_attr *attr,
1714 struct lustre_capa *capa)
1716 struct osd_object *obj = osd_dt_obj(dt);
1718 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1719 LINVRNT(osd_invariant(obj));
1721 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1724 spin_lock(&obj->oo_guard);
1725 osd_inode_getattr(env, obj->oo_inode, attr);
1726 spin_unlock(&obj->oo_guard);
1730 static int osd_declare_attr_set(const struct lu_env *env,
1731 struct dt_object *dt,
1732 const struct lu_attr *attr,
1733 struct thandle *handle)
1735 struct osd_thandle *oh;
1736 struct osd_object *obj;
1737 struct osd_thread_info *info = osd_oti_get(env);
1738 struct lquota_id_info *qi = &info->oti_qi;
1744 LASSERT(dt != NULL);
1745 LASSERT(handle != NULL);
1747 obj = osd_dt_obj(dt);
1748 LASSERT(osd_invariant(obj));
1750 oh = container_of0(handle, struct osd_thandle, ot_super);
1751 LASSERT(oh->ot_handle == NULL);
1753 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1754 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1756 if (attr == NULL || obj->oo_inode == NULL)
1759 bspace = obj->oo_inode->i_blocks;
1760 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1761 bspace = toqb(bspace);
1763 /* Changing ownership is always preformed by super user, it should not
1766 * We still need to call the osd_declare_qid() to calculate the journal
1767 * credits for updating quota accounting files and to trigger quota
1768 * space adjustment once the operation is completed.*/
1769 if ((attr->la_valid & LA_UID) != 0 &&
1770 attr->la_uid != obj->oo_inode->i_uid) {
1771 qi->lqi_type = USRQUOTA;
1773 /* inode accounting */
1774 qi->lqi_is_blk = false;
1776 /* one more inode for the new owner ... */
1777 qi->lqi_id.qid_uid = attr->la_uid;
1779 allocated = (attr->la_uid == 0) ? true : false;
1780 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1781 if (rc == -EDQUOT || rc == -EINPROGRESS)
1786 /* and one less inode for the current uid */
1787 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1789 rc = osd_declare_qid(env, oh, qi, true, NULL);
1790 if (rc == -EDQUOT || rc == -EINPROGRESS)
1795 /* block accounting */
1796 qi->lqi_is_blk = true;
1798 /* more blocks for the new owner ... */
1799 qi->lqi_id.qid_uid = attr->la_uid;
1800 qi->lqi_space = bspace;
1801 allocated = (attr->la_uid == 0) ? true : false;
1802 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1803 if (rc == -EDQUOT || rc == -EINPROGRESS)
1808 /* and finally less blocks for the current owner */
1809 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1810 qi->lqi_space = -bspace;
1811 rc = osd_declare_qid(env, oh, qi, true, NULL);
1812 if (rc == -EDQUOT || rc == -EINPROGRESS)
1818 if (attr->la_valid & LA_GID &&
1819 attr->la_gid != obj->oo_inode->i_gid) {
1820 qi->lqi_type = GRPQUOTA;
1822 /* inode accounting */
1823 qi->lqi_is_blk = false;
1825 /* one more inode for the new group owner ... */
1826 qi->lqi_id.qid_gid = attr->la_gid;
1828 allocated = (attr->la_gid == 0) ? true : false;
1829 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1830 if (rc == -EDQUOT || rc == -EINPROGRESS)
1835 /* and one less inode for the current gid */
1836 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1838 rc = osd_declare_qid(env, oh, qi, true, NULL);
1839 if (rc == -EDQUOT || rc == -EINPROGRESS)
1844 /* block accounting */
1845 qi->lqi_is_blk = true;
1847 /* more blocks for the new owner ... */
1848 qi->lqi_id.qid_gid = attr->la_gid;
1849 qi->lqi_space = bspace;
1850 allocated = (attr->la_gid == 0) ? true : false;
1851 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1852 if (rc == -EDQUOT || rc == -EINPROGRESS)
1857 /* and finally less blocks for the current owner */
1858 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1859 qi->lqi_space = -bspace;
1860 rc = osd_declare_qid(env, oh, qi, true, NULL);
1861 if (rc == -EDQUOT || rc == -EINPROGRESS)
1870 static int osd_inode_setattr(const struct lu_env *env,
1871 struct inode *inode, const struct lu_attr *attr)
1875 bits = attr->la_valid;
1877 if (bits & LA_ATIME)
1878 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
1879 if (bits & LA_CTIME)
1880 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
1881 if (bits & LA_MTIME)
1882 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
1883 if (bits & LA_SIZE) {
1884 LDISKFS_I(inode)->i_disksize = attr->la_size;
1885 i_size_write(inode, attr->la_size);
1889 /* OSD should not change "i_blocks" which is used by quota.
1890 * "i_blocks" should be changed by ldiskfs only. */
1891 if (bits & LA_BLOCKS)
1892 inode->i_blocks = attr->la_blocks;
1895 inode->i_mode = (inode->i_mode & S_IFMT) |
1896 (attr->la_mode & ~S_IFMT);
1898 inode->i_uid = attr->la_uid;
1900 inode->i_gid = attr->la_gid;
1901 if (bits & LA_NLINK)
1902 set_nlink(inode, attr->la_nlink);
1904 inode->i_rdev = attr->la_rdev;
1906 if (bits & LA_FLAGS) {
1907 /* always keep S_NOCMTIME */
1908 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1914 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1916 if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1917 (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1922 if (attr->la_valid & LA_UID)
1923 iattr.ia_valid |= ATTR_UID;
1924 if (attr->la_valid & LA_GID)
1925 iattr.ia_valid |= ATTR_GID;
1926 iattr.ia_uid = attr->la_uid;
1927 iattr.ia_gid = attr->la_gid;
1929 rc = ll_vfs_dq_transfer(inode, &iattr);
1931 CERROR("%s: quota transfer failed: rc = %d. Is quota "
1932 "enforcement enabled on the ldiskfs filesystem?",
1933 inode->i_sb->s_id, rc);
1940 static int osd_attr_set(const struct lu_env *env,
1941 struct dt_object *dt,
1942 const struct lu_attr *attr,
1943 struct thandle *handle,
1944 struct lustre_capa *capa)
1946 struct osd_object *obj = osd_dt_obj(dt);
1947 struct inode *inode;
1950 LASSERT(handle != NULL);
1951 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1952 LASSERT(osd_invariant(obj));
1954 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1957 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1959 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1960 struct osd_thread_info *oti = osd_oti_get(env);
1961 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
1962 struct lu_fid *fid1 = &oti->oti_fid;
1963 struct osd_inode_id *id = &oti->oti_id;
1964 struct iam_path_descr *ipd;
1965 struct iam_container *bag;
1966 struct osd_thandle *oh;
1969 fid_cpu_to_be(fid1, fid0);
1970 memset(id, 1, sizeof(*id));
1971 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1972 fid0)->oi_dir.od_container;
1973 ipd = osd_idx_ipd_get(env, bag);
1974 if (unlikely(ipd == NULL))
1977 oh = container_of0(handle, struct osd_thandle, ot_super);
1978 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1979 (const struct iam_rec *)id, ipd);
1980 osd_ipd_put(env, bag, ipd);
1981 return(rc > 0 ? 0 : rc);
1984 inode = obj->oo_inode;
1985 ll_vfs_dq_init(inode);
1987 rc = osd_quota_transfer(inode, attr);
1991 spin_lock(&obj->oo_guard);
1992 rc = osd_inode_setattr(env, inode, attr);
1993 spin_unlock(&obj->oo_guard);
1996 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2000 struct dentry *osd_child_dentry_get(const struct lu_env *env,
2001 struct osd_object *obj,
2002 const char *name, const int namelen)
2004 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2007 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2008 umode_t mode, struct dt_allocation_hint *hint,
2012 struct osd_device *osd = osd_obj2dev(obj);
2013 struct osd_thandle *oth;
2014 struct dt_object *parent = NULL;
2015 struct inode *inode;
2017 LINVRNT(osd_invariant(obj));
2018 LASSERT(obj->oo_inode == NULL);
2019 LASSERT(obj->oo_hl_head == NULL);
2021 if (S_ISDIR(mode) && ldiskfs_pdo) {
2022 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2023 if (obj->oo_hl_head == NULL)
2027 oth = container_of(th, struct osd_thandle, ot_super);
2028 LASSERT(oth->ot_handle->h_transaction != NULL);
2030 if (hint && hint->dah_parent)
2031 parent = hint->dah_parent;
2033 inode = ldiskfs_create_inode(oth->ot_handle,
2034 parent ? osd_dt_obj(parent)->oo_inode :
2035 osd_sb(osd)->s_root->d_inode,
2037 if (!IS_ERR(inode)) {
2038 /* Do not update file c/mtime in ldiskfs.
2039 * NB: don't need any lock because no contention at this
2041 inode->i_flags |= S_NOCMTIME;
2043 /* For new created object, it must be consistent,
2044 * and it is unnecessary to scrub against it. */
2045 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2046 obj->oo_inode = inode;
2049 if (obj->oo_hl_head != NULL) {
2050 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2051 obj->oo_hl_head = NULL;
2053 result = PTR_ERR(inode);
2055 LINVRNT(osd_invariant(obj));
2063 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2064 struct lu_attr *attr,
2065 struct dt_allocation_hint *hint,
2066 struct dt_object_format *dof,
2070 struct osd_thandle *oth;
2071 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2073 LASSERT(S_ISDIR(attr->la_mode));
2075 oth = container_of(th, struct osd_thandle, ot_super);
2076 LASSERT(oth->ot_handle->h_transaction != NULL);
2077 result = osd_mkfile(info, obj, mode, hint, th);
2082 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2083 struct lu_attr *attr,
2084 struct dt_allocation_hint *hint,
2085 struct dt_object_format *dof,
2089 struct osd_thandle *oth;
2090 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2092 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2094 LASSERT(S_ISREG(attr->la_mode));
2096 oth = container_of(th, struct osd_thandle, ot_super);
2097 LASSERT(oth->ot_handle->h_transaction != NULL);
2099 result = osd_mkfile(info, obj, mode, hint, th);
2101 LASSERT(obj->oo_inode != NULL);
2102 if (feat->dif_flags & DT_IND_VARKEY)
2103 result = iam_lvar_create(obj->oo_inode,
2104 feat->dif_keysize_max,
2106 feat->dif_recsize_max,
2109 result = iam_lfix_create(obj->oo_inode,
2110 feat->dif_keysize_max,
2112 feat->dif_recsize_max,
2119 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2120 struct lu_attr *attr,
2121 struct dt_allocation_hint *hint,
2122 struct dt_object_format *dof,
2125 LASSERT(S_ISREG(attr->la_mode));
2126 return osd_mkfile(info, obj, (attr->la_mode &
2127 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2130 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2131 struct lu_attr *attr,
2132 struct dt_allocation_hint *hint,
2133 struct dt_object_format *dof,
2136 LASSERT(S_ISLNK(attr->la_mode));
2137 return osd_mkfile(info, obj, (attr->la_mode &
2138 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2141 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2142 struct lu_attr *attr,
2143 struct dt_allocation_hint *hint,
2144 struct dt_object_format *dof,
2147 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2150 LINVRNT(osd_invariant(obj));
2151 LASSERT(obj->oo_inode == NULL);
2152 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2153 S_ISFIFO(mode) || S_ISSOCK(mode));
2155 result = osd_mkfile(info, obj, mode, hint, th);
2157 LASSERT(obj->oo_inode != NULL);
2159 * This inode should be marked dirty for i_rdev. Currently
2160 * that is done in the osd_attr_init().
2162 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2165 LINVRNT(osd_invariant(obj));
2169 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2171 struct dt_allocation_hint *hint,
2172 struct dt_object_format *dof,
2175 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2177 osd_obj_type_f result;
2193 result = osd_mk_index;
2204 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2205 struct dt_object *parent, struct dt_object *child,
2210 ah->dah_parent = parent;
2211 ah->dah_mode = child_mode;
2214 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2215 struct lu_attr *attr, struct dt_object_format *dof)
2217 struct inode *inode = obj->oo_inode;
2218 __u64 valid = attr->la_valid;
2221 attr->la_valid &= ~(LA_TYPE | LA_MODE);
2223 if (dof->dof_type != DFT_NODE)
2224 attr->la_valid &= ~LA_RDEV;
2225 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2226 attr->la_valid &= ~LA_ATIME;
2227 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2228 attr->la_valid &= ~LA_CTIME;
2229 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2230 attr->la_valid &= ~LA_MTIME;
2232 result = osd_quota_transfer(inode, attr);
2236 if (attr->la_valid != 0) {
2237 result = osd_inode_setattr(info->oti_env, inode, attr);
2239 * The osd_inode_setattr() should always succeed here. The
2240 * only error that could be returned is EDQUOT when we are
2241 * trying to change the UID or GID of the inode. However, this
2242 * should not happen since quota enforcement is no longer
2243 * enabled on ldiskfs (lquota takes care of it).
2245 LASSERTF(result == 0, "%d", result);
2246 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2249 attr->la_valid = valid;
2253 * Helper function for osd_object_create()
2255 * \retval 0, on success
2257 static int __osd_object_create(struct osd_thread_info *info,
2258 struct osd_object *obj, struct lu_attr *attr,
2259 struct dt_allocation_hint *hint,
2260 struct dt_object_format *dof,
2266 /* we drop umask so that permissions we pass are not affected */
2267 umask = current->fs->umask;
2268 current->fs->umask = 0;
2270 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2273 osd_attr_init(info, obj, attr, dof);
2274 osd_object_init0(obj);
2276 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2277 unlock_new_inode(obj->oo_inode);
2280 /* restore previous umask value */
2281 current->fs->umask = umask;
2287 * Helper function for osd_object_create()
2289 * \retval 0, on success
2291 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2292 const struct lu_fid *fid, struct thandle *th)
2294 struct osd_thread_info *info = osd_oti_get(env);
2295 struct osd_inode_id *id = &info->oti_id;
2296 struct osd_device *osd = osd_obj2dev(obj);
2297 struct osd_thandle *oh;
2299 LASSERT(obj->oo_inode != NULL);
2301 oh = container_of0(th, struct osd_thandle, ot_super);
2302 LASSERT(oh->ot_handle);
2304 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2305 return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2308 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2309 obd_seq seq, struct lu_seq_range *range)
2311 struct seq_server_site *ss = osd_seq_site(osd);
2313 if (fid_seq_is_idif(seq)) {
2314 fld_range_set_ost(range);
2315 range->lsr_index = idif_ost_idx(seq);
2319 if (!fid_seq_in_fldb(seq)) {
2320 fld_range_set_mdt(range);
2322 /* FIXME: If ss is NULL, it suppose not get lsr_index
2324 range->lsr_index = ss->ss_node_id;
2328 LASSERT(ss != NULL);
2329 fld_range_set_any(range);
2330 /* OSD will only do local fld lookup */
2331 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2335 * Concurrency: no external locking is necessary.
2337 static int osd_declare_object_create(const struct lu_env *env,
2338 struct dt_object *dt,
2339 struct lu_attr *attr,
2340 struct dt_allocation_hint *hint,
2341 struct dt_object_format *dof,
2342 struct thandle *handle)
2344 struct osd_thandle *oh;
2348 LASSERT(handle != NULL);
2350 oh = container_of0(handle, struct osd_thandle, ot_super);
2351 LASSERT(oh->ot_handle == NULL);
2353 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2354 osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2355 /* Reuse idle OI block may cause additional one OI block
2357 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2358 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2360 /* If this is directory, then we expect . and .. to be inserted as
2361 * well. The one directory block always needs to be created for the
2362 * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2363 * block), there is no danger of needing a tree for the first block.
2365 if (attr && S_ISDIR(attr->la_mode)) {
2366 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2367 osd_dto_credits_noquota[DTO_WRITE_BASE]);
2368 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2374 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2375 false, false, NULL, false);
2382 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2383 struct lu_attr *attr,
2384 struct dt_allocation_hint *hint,
2385 struct dt_object_format *dof,
2388 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2389 struct osd_object *obj = osd_dt_obj(dt);
2390 struct osd_thread_info *info = osd_oti_get(env);
2395 LINVRNT(osd_invariant(obj));
2396 LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2397 LASSERT(osd_write_locked(env, obj));
2398 LASSERT(th != NULL);
2400 if (unlikely(fid_is_acct(fid)))
2401 /* Quota files can't be created from the kernel any more,
2402 * 'tune2fs -O quota' will take care of creating them */
2405 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2406 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2408 result = __osd_object_create(info, obj, attr, hint, dof, th);
2410 result = __osd_oi_insert(env, obj, fid, th);
2412 LASSERT(ergo(result == 0,
2413 dt_object_exists(dt) && !dt_object_remote(dt)));
2415 LASSERT(osd_invariant(obj));
2420 * Called to destroy on-disk representation of the object
2422 * Concurrency: must be locked
2424 static int osd_declare_object_destroy(const struct lu_env *env,
2425 struct dt_object *dt,
2428 struct osd_object *obj = osd_dt_obj(dt);
2429 struct inode *inode = obj->oo_inode;
2430 struct osd_thandle *oh;
2434 oh = container_of0(th, struct osd_thandle, ot_super);
2435 LASSERT(oh->ot_handle == NULL);
2438 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2439 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2440 /* Recycle idle OI leaf may cause additional three OI blocks
2442 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2443 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2444 /* one less inode */
2445 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
2446 false, true, NULL, false);
2449 /* data to be truncated */
2450 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
2451 true, true, NULL, false);
2455 static int osd_object_destroy(const struct lu_env *env,
2456 struct dt_object *dt,
2459 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2460 struct osd_object *obj = osd_dt_obj(dt);
2461 struct inode *inode = obj->oo_inode;
2462 struct osd_device *osd = osd_obj2dev(obj);
2463 struct osd_thandle *oh;
2467 oh = container_of0(th, struct osd_thandle, ot_super);
2468 LASSERT(oh->ot_handle);
2470 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2472 if (unlikely(fid_is_acct(fid)))
2475 if (S_ISDIR(inode->i_mode)) {
2476 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2477 inode->i_nlink == 2);
2478 /* it will check/delete the inode from remote parent,
2479 * how to optimize it? unlink performance impaction XXX */
2480 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2481 if (result != 0 && result != -ENOENT) {
2482 CERROR("%s: delete inode "DFID": rc = %d\n",
2483 osd_name(osd), PFID(fid), result);
2485 spin_lock(&obj->oo_guard);
2487 spin_unlock(&obj->oo_guard);
2488 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2491 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2493 result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2496 /* XXX: add to ext3 orphan list */
2497 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2499 /* not needed in the cache anymore */
2500 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2506 * Put the fid into lustre_mdt_attrs, and then place the structure
2507 * inode's ea. This fid should not be altered during the life time
2510 * \retval +ve, on success
2511 * \retval -ve, on error
2513 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2515 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2516 const struct lu_fid *fid, __u32 compat, __u32 incompat)
2518 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2522 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2525 lustre_lma_init(lma, fid, compat, incompat);
2526 lustre_lma_swab(lma);
2528 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2530 /* LMA may already exist, but we need to check that all the
2531 * desired compat/incompat flags have been added. */
2532 if (unlikely(rc == -EEXIST)) {
2533 if (compat == 0 && incompat == 0)
2536 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2537 XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2542 lustre_lma_swab(lma);
2543 if (!(~lma->lma_compat & compat) &&
2544 !(~lma->lma_incompat & incompat))
2547 lma->lma_compat |= compat;
2548 lma->lma_incompat |= incompat;
2549 lustre_lma_swab(lma);
2550 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2551 sizeof(*lma), XATTR_REPLACE);
2558 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2559 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2560 * To have compatilibility with 1.8 ldiskfs driver we need to have
2561 * magic number at start of fid data.
2562 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2565 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2566 const struct dt_rec *fid)
2568 if (!fid_is_namespace_visible((const struct lu_fid *)fid) ||
2569 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2570 param->edp_magic = 0;
2574 param->edp_magic = LDISKFS_LUFID_MAGIC;
2575 param->edp_len = sizeof(struct lu_fid) + 1;
2576 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2580 * Try to read the fid from inode ea into dt_rec.
2582 * \param fid object fid.
2584 * \retval 0 on success
2586 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2587 __u32 ino, struct lu_fid *fid,
2588 struct osd_inode_id *id)
2590 struct osd_thread_info *info = osd_oti_get(env);
2591 struct inode *inode;
2594 osd_id_gen(id, ino, OSD_OII_NOGEN);
2595 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2597 RETURN(PTR_ERR(inode));
2603 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2605 struct inode *parent_dir,
2606 const struct dt_rec *dot_fid,
2607 const struct dt_rec *dot_dot_fid,
2608 struct osd_thandle *oth)
2610 struct ldiskfs_dentry_param *dot_ldp;
2611 struct ldiskfs_dentry_param *dot_dot_ldp;
2613 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2614 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2616 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2617 dot_ldp->edp_magic = 0;
2618 return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2619 dir, dot_ldp, dot_dot_ldp);
2623 * Create an local agent inode for remote entry
2625 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2626 struct osd_device *osd,
2627 struct osd_object *pobj,
2628 const struct lu_fid *fid,
2631 struct osd_thread_info *info = osd_oti_get(env);
2632 struct inode *local;
2633 struct osd_thandle *oh;
2638 oh = container_of(th, struct osd_thandle, ot_super);
2639 LASSERT(oh->ot_handle->h_transaction != NULL);
2641 /* FIXME: Insert index api needs to know the mode of
2642 * the remote object. Just use S_IFDIR for now */
2643 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR);
2644 if (IS_ERR(local)) {
2645 CERROR("%s: create local error %d\n", osd_name(osd),
2646 (int)PTR_ERR(local));
2650 /* Set special LMA flag for local agent inode */
2651 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2653 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2654 osd_name(osd), PFID(fid), rc);
2655 RETURN(ERR_PTR(rc));
2658 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2659 (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
2660 (const struct dt_rec *)fid, oh);
2662 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2663 osd_name(osd), PFID(fid), rc);
2664 RETURN(ERR_PTR(rc));
2671 * Delete local agent inode for remote entry
2673 static int osd_delete_local_agent_inode(const struct lu_env *env,
2674 struct osd_device *osd,
2675 const struct lu_fid *fid,
2676 __u32 ino, struct osd_thandle *oh)
2678 struct osd_thread_info *oti = osd_oti_get(env);
2679 struct osd_inode_id *id = &oti->oti_id;
2680 struct inode *inode;
2683 id->oii_ino = le32_to_cpu(ino);
2684 id->oii_gen = OSD_OII_NOGEN;
2685 inode = osd_iget(oti, osd, id);
2686 if (IS_ERR(inode)) {
2687 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2688 PFID(fid), id->oii_ino, id->oii_gen);
2689 RETURN(PTR_ERR(inode));
2693 mark_inode_dirty(inode);
2694 CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2695 osd_name(osd), PFID(fid), inode->i_ino);
2701 * OSD layer object create function for interoperability mode (b11826).
2702 * This is mostly similar to osd_object_create(). Only difference being, fid is
2703 * inserted into inode ea here.
2705 * \retval 0, on success
2706 * \retval -ve, on error
2708 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2709 struct lu_attr *attr,
2710 struct dt_allocation_hint *hint,
2711 struct dt_object_format *dof,
2714 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2715 struct osd_object *obj = osd_dt_obj(dt);
2716 struct osd_thread_info *info = osd_oti_get(env);
2721 LASSERT(osd_invariant(obj));
2722 LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2723 LASSERT(osd_write_locked(env, obj));
2724 LASSERT(th != NULL);
2726 if (unlikely(fid_is_acct(fid)))
2727 /* Quota files can't be created from the kernel any more,
2728 * 'tune2fs -O quota' will take care of creating them */
2731 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2732 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2734 result = __osd_object_create(info, obj, attr, hint, dof, th);
2736 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2737 fid_is_on_ost(info, osd_obj2dev(obj),
2738 fid, OI_CHECK_FLD) ?
2739 LMAC_FID_ON_OST : 0, 0);
2742 result = __osd_oi_insert(env, obj, fid, th);
2744 LASSERT(ergo(result == 0,
2745 dt_object_exists(dt) && !dt_object_remote(dt)));
2746 LINVRNT(osd_invariant(obj));
2750 static int osd_declare_object_ref_add(const struct lu_env *env,
2751 struct dt_object *dt,
2752 struct thandle *handle)
2754 struct osd_thandle *oh;
2756 /* it's possible that object doesn't exist yet */
2757 LASSERT(handle != NULL);
2759 oh = container_of0(handle, struct osd_thandle, ot_super);
2760 LASSERT(oh->ot_handle == NULL);
2762 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2763 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2769 * Concurrency: @dt is write locked.
2771 static int osd_object_ref_add(const struct lu_env *env,
2772 struct dt_object *dt, struct thandle *th)
2774 struct osd_object *obj = osd_dt_obj(dt);
2775 struct inode *inode = obj->oo_inode;
2776 struct osd_thandle *oh;
2779 LINVRNT(osd_invariant(obj));
2780 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2781 LASSERT(osd_write_locked(env, obj));
2782 LASSERT(th != NULL);
2784 oh = container_of0(th, struct osd_thandle, ot_super);
2785 LASSERT(oh->ot_handle != NULL);
2787 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2789 CDEBUG(D_INODE, DFID" increase nlink %d\n",
2790 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2792 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2793 * (65000) subdirectories by storing "1" in i_nlink if the link count
2794 * would otherwise overflow. Directory tranversal tools understand
2795 * that (st_nlink == 1) indicates that the filesystem dose not track
2796 * hard links count on the directory, and will not abort subdirectory
2797 * scanning early once (st_nlink - 2) subdirs have been found.
2799 * This also has to properly handle the case of inodes with nlink == 0
2800 * in case they are being linked into the PENDING directory
2802 spin_lock(&obj->oo_guard);
2803 ldiskfs_inc_count(oh->ot_handle, inode);
2804 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2805 spin_unlock(&obj->oo_guard);
2807 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2808 LINVRNT(osd_invariant(obj));
2813 static int osd_declare_object_ref_del(const struct lu_env *env,
2814 struct dt_object *dt,
2815 struct thandle *handle)
2817 struct osd_thandle *oh;
2819 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2820 LASSERT(handle != NULL);
2822 oh = container_of0(handle, struct osd_thandle, ot_super);
2823 LASSERT(oh->ot_handle == NULL);
2825 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2826 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2832 * Concurrency: @dt is write locked.
2834 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2837 struct osd_object *obj = osd_dt_obj(dt);
2838 struct inode *inode = obj->oo_inode;
2839 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
2840 struct osd_thandle *oh;
2842 LINVRNT(osd_invariant(obj));
2843 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2844 LASSERT(osd_write_locked(env, obj));
2845 LASSERT(th != NULL);
2847 oh = container_of0(th, struct osd_thandle, ot_super);
2848 LASSERT(oh->ot_handle != NULL);
2850 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2852 spin_lock(&obj->oo_guard);
2853 /* That can be result of upgrade from old Lustre version and
2854 * applied only to local files. Just skip this ref_del call.
2855 * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2856 if (inode->i_nlink == 0) {
2857 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2858 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2859 ", maybe an upgraded file? (LU-3915)\n",
2860 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2861 spin_unlock(&obj->oo_guard);
2865 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2866 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2868 ldiskfs_dec_count(oh->ot_handle, inode);
2869 spin_unlock(&obj->oo_guard);
2871 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2872 LINVRNT(osd_invariant(obj));
2878 * Get the 64-bit version for an inode.
2880 static int osd_object_version_get(const struct lu_env *env,
2881 struct dt_object *dt, dt_obj_version_t *ver)
2883 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2885 CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2886 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2887 *ver = LDISKFS_I(inode)->i_fs_version;
2892 * Concurrency: @dt is read locked.
2894 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2895 struct lu_buf *buf, const char *name,
2896 struct lustre_capa *capa)
2898 struct osd_object *obj = osd_dt_obj(dt);
2899 struct inode *inode = obj->oo_inode;
2900 struct osd_thread_info *info = osd_oti_get(env);
2901 struct dentry *dentry = &info->oti_obj_dentry;
2903 /* version get is not real XATTR but uses xattr API */
2904 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2905 /* for version we are just using xattr API but change inode
2907 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2908 osd_object_version_get(env, dt, buf->lb_buf);
2909 return sizeof(dt_obj_version_t);
2912 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2913 LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2915 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2918 return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2922 static int osd_declare_xattr_set(const struct lu_env *env,
2923 struct dt_object *dt,
2924 const struct lu_buf *buf, const char *name,
2925 int fl, struct thandle *handle)
2927 struct osd_thandle *oh;
2930 LASSERT(handle != NULL);
2932 oh = container_of0(handle, struct osd_thandle, ot_super);
2933 LASSERT(oh->ot_handle == NULL);
2935 /* optimistic optimization: LMA is set first and usually fit inode */
2936 if (strcmp(name, XATTR_NAME_LMA) == 0) {
2937 if (dt_object_exists(dt))
2941 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2944 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
2945 struct super_block *sb = osd_sb(osd);
2946 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2947 if (buf && buf->lb_len > sb->s_blocksize) {
2948 credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2949 sb->s_blocksize_bits;
2953 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2959 * Set the 64-bit version for object
2961 static void osd_object_version_set(const struct lu_env *env,
2962 struct dt_object *dt,
2963 dt_obj_version_t *new_version)
2965 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2967 CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2968 *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2970 LDISKFS_I(inode)->i_fs_version = *new_version;
2971 /** Version is set after all inode operations are finished,
2972 * so we should mark it dirty here */
2973 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2977 * Concurrency: @dt is write locked.
2979 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2980 const struct lu_buf *buf, const char *name, int fl,
2981 struct thandle *handle, struct lustre_capa *capa)
2983 struct osd_object *obj = osd_dt_obj(dt);
2984 struct inode *inode = obj->oo_inode;
2985 struct osd_thread_info *info = osd_oti_get(env);
2989 LASSERT(handle != NULL);
2991 /* version set is not real XATTR */
2992 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2993 /* for version we are just using xattr API but change inode
2995 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2996 osd_object_version_set(env, dt, buf->lb_buf);
2997 return sizeof(dt_obj_version_t);
3000 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3003 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n",
3004 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3006 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3007 if (fl & LU_XATTR_REPLACE)
3008 fs_flags |= XATTR_REPLACE;
3010 if (fl & LU_XATTR_CREATE)
3011 fs_flags |= XATTR_CREATE;
3013 return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3018 * Concurrency: @dt is read locked.
3020 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3021 struct lu_buf *buf, struct lustre_capa *capa)
3023 struct osd_object *obj = osd_dt_obj(dt);
3024 struct inode *inode = obj->oo_inode;
3025 struct osd_thread_info *info = osd_oti_get(env);
3026 struct dentry *dentry = &info->oti_obj_dentry;
3028 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3029 LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
3031 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3034 dentry->d_inode = inode;
3035 dentry->d_sb = inode->i_sb;
3036 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3039 static int osd_declare_xattr_del(const struct lu_env *env,
3040 struct dt_object *dt, const char *name,
3041 struct thandle *handle)
3043 struct osd_thandle *oh;
3045 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3046 LASSERT(handle != NULL);
3048 oh = container_of0(handle, struct osd_thandle, ot_super);
3049 LASSERT(oh->ot_handle == NULL);
3051 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3052 osd_dto_credits_noquota[DTO_XATTR_SET]);
3058 * Concurrency: @dt is write locked.
3060 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3061 const char *name, struct thandle *handle,
3062 struct lustre_capa *capa)
3064 struct osd_object *obj = osd_dt_obj(dt);
3065 struct inode *inode = obj->oo_inode;
3066 struct osd_thread_info *info = osd_oti_get(env);
3067 struct dentry *dentry = &info->oti_obj_dentry;
3070 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3071 LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
3072 LASSERT(handle != NULL);
3074 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3077 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3079 ll_vfs_dq_init(inode);
3080 dentry->d_inode = inode;
3081 dentry->d_sb = inode->i_sb;
3082 rc = inode->i_op->removexattr(dentry, name);
3086 static struct obd_capa *osd_capa_get(const struct lu_env *env,
3087 struct dt_object *dt,
3088 struct lustre_capa *old, __u64 opc)
3090 struct osd_thread_info *info = osd_oti_get(env);
3091 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3092 struct osd_object *obj = osd_dt_obj(dt);
3093 struct osd_device *osd = osd_obj2dev(obj);
3094 struct lustre_capa_key *key = &info->oti_capa_key;
3095 struct lustre_capa *capa = &info->oti_capa;
3096 struct obd_capa *oc;
3097 struct lu_capainfo *lci;
3101 if (!osd->od_fl_capa)
3102 RETURN(ERR_PTR(-ENOENT));
3104 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3105 LINVRNT(osd_invariant(obj));
3107 /* renewal sanity check */
3108 if (old && osd_object_auth(env, dt, old, opc))
3109 RETURN(ERR_PTR(-EACCES));
3111 lci = lu_capainfo_get(env);
3112 if (unlikely(lci == NULL))
3113 RETURN(ERR_PTR(-ENOENT));
3115 switch (lci->lci_auth) {
3119 capa->lc_uid = obj->oo_inode->i_uid;
3120 capa->lc_gid = obj->oo_inode->i_gid;
3121 capa->lc_flags = LC_ID_PLAIN;
3123 case LC_ID_CONVERT: {
3126 s[0] = obj->oo_inode->i_uid;
3127 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
3128 s[2] = obj->oo_inode->i_gid;
3129 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
3130 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
3132 RETURN(ERR_PTR(rc));
3134 capa->lc_uid = ((__u64)d[1] << 32) | d[0];
3135 capa->lc_gid = ((__u64)d[3] << 32) | d[2];
3136 capa->lc_flags = LC_ID_CONVERT;
3140 RETURN(ERR_PTR(-EINVAL));
3143 capa->lc_fid = *fid;
3145 capa->lc_flags |= osd->od_capa_alg << 24;
3146 capa->lc_timeout = osd->od_capa_timeout;
3147 capa->lc_expiry = 0;
3149 oc = capa_lookup(osd->od_capa_hash, capa, 1);
3151 LASSERT(!capa_is_expired(oc));
3155 spin_lock(&capa_lock);
3156 *key = osd->od_capa_keys[1];
3157 spin_unlock(&capa_lock);
3159 capa->lc_keyid = key->lk_keyid;
3160 capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout;
3162 rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
3164 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
3165 RETURN(ERR_PTR(rc));
3168 oc = capa_add(osd->od_capa_hash, capa);
3172 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3173 __u64 start, __u64 end)
3175 struct osd_object *obj = osd_dt_obj(dt);
3176 struct inode *inode = obj->oo_inode;
3177 struct osd_thread_info *info = osd_oti_get(env);
3178 struct dentry *dentry = &info->oti_obj_dentry;
3179 struct file *file = &info->oti_file;
3184 dentry->d_inode = inode;
3185 dentry->d_sb = inode->i_sb;
3186 file->f_dentry = dentry;
3187 file->f_mapping = inode->i_mapping;
3188 file->f_op = inode->i_fop;
3189 set_file_inode(file, inode);
3191 #ifdef HAVE_FILE_FSYNC_4ARGS
3192 rc = file->f_op->fsync(file, start, end, 0);
3193 #elif defined(HAVE_FILE_FSYNC_2ARGS)
3194 mutex_lock(&inode->i_mutex);
3195 rc = file->f_op->fsync(file, 0);
3196 mutex_unlock(&inode->i_mutex);
3198 mutex_lock(&inode->i_mutex);
3199 rc = file->f_op->fsync(file, dentry, 0);
3200 mutex_unlock(&inode->i_mutex);
3206 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
3209 struct osd_object *obj = osd_dt_obj(dt);
3212 *data = (void *)obj->oo_inode;
3220 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3221 const struct dt_index_features *feat)
3223 struct iam_descr *descr;
3225 if (osd_object_is_root(o))
3226 return feat == &dt_directory_features;
3228 LASSERT(o->oo_dir != NULL);
3230 descr = o->oo_dir->od_container.ic_descr;
3231 if (feat == &dt_directory_features) {
3232 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3238 feat->dif_keysize_min <= descr->id_key_size &&
3239 descr->id_key_size <= feat->dif_keysize_max &&
3240 feat->dif_recsize_min <= descr->id_rec_size &&
3241 descr->id_rec_size <= feat->dif_recsize_max &&
3242 !(feat->dif_flags & (DT_IND_VARKEY |
3243 DT_IND_VARREC | DT_IND_NONUNQ)) &&
3244 ergo(feat->dif_flags & DT_IND_UPDATE,
3245 1 /* XXX check that object (and file system) is
3250 static int osd_iam_container_init(const struct lu_env *env,
3251 struct osd_object *obj,
3252 struct osd_directory *dir)
3254 struct iam_container *bag = &dir->od_container;
3257 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3261 result = iam_container_setup(bag);
3263 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3265 iam_container_fini(bag);
3272 * Concurrency: no external locking is necessary.
3274 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3275 const struct dt_index_features *feat)
3279 struct osd_object *obj = osd_dt_obj(dt);
3281 LINVRNT(osd_invariant(obj));
3283 if (osd_object_is_root(obj)) {
3284 dt->do_index_ops = &osd_index_ea_ops;
3286 } else if (feat == &dt_directory_features) {
3287 dt->do_index_ops = &osd_index_ea_ops;
3288 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3293 } else if (unlikely(feat == &dt_otable_features)) {
3294 dt->do_index_ops = &osd_otable_ops;
3296 } else if (unlikely(feat == &dt_acct_features)) {
3297 dt->do_index_ops = &osd_acct_index_ops;
3300 } else if (!osd_has_index(obj)) {
3301 struct osd_directory *dir;
3306 spin_lock(&obj->oo_guard);
3307 if (obj->oo_dir == NULL)
3311 * Concurrent thread allocated container data.
3314 spin_unlock(&obj->oo_guard);
3316 * Now, that we have container data, serialize its
3319 down_write(&obj->oo_ext_idx_sem);
3321 * recheck under lock.
3323 if (!osd_has_index(obj))
3324 result = osd_iam_container_init(env, obj, dir);
3327 up_write(&obj->oo_ext_idx_sem);
3335 if (result == 0 && skip_iam == 0) {
3336 if (!osd_iam_index_probe(env, obj, feat))
3339 LINVRNT(osd_invariant(obj));
3341 if (result == 0 && is_quota_glb_feat(feat) &&
3342 fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3343 result = osd_quota_migration(env, dt, feat);
3348 static int osd_otable_it_attr_get(const struct lu_env *env,
3349 struct dt_object *dt,
3350 struct lu_attr *attr,
3351 struct lustre_capa *capa)
3357 static const struct dt_object_operations osd_obj_ops = {
3358 .do_read_lock = osd_object_read_lock,
3359 .do_write_lock = osd_object_write_lock,
3360 .do_read_unlock = osd_object_read_unlock,
3361 .do_write_unlock = osd_object_write_unlock,
3362 .do_write_locked = osd_object_write_locked,
3363 .do_attr_get = osd_attr_get,
3364 .do_declare_attr_set = osd_declare_attr_set,
3365 .do_attr_set = osd_attr_set,
3366 .do_ah_init = osd_ah_init,
3367 .do_declare_create = osd_declare_object_create,
3368 .do_create = osd_object_create,
3369 .do_declare_destroy = osd_declare_object_destroy,
3370 .do_destroy = osd_object_destroy,
3371 .do_index_try = osd_index_try,
3372 .do_declare_ref_add = osd_declare_object_ref_add,
3373 .do_ref_add = osd_object_ref_add,
3374 .do_declare_ref_del = osd_declare_object_ref_del,
3375 .do_ref_del = osd_object_ref_del,
3376 .do_xattr_get = osd_xattr_get,
3377 .do_declare_xattr_set = osd_declare_xattr_set,
3378 .do_xattr_set = osd_xattr_set,
3379 .do_declare_xattr_del = osd_declare_xattr_del,
3380 .do_xattr_del = osd_xattr_del,
3381 .do_xattr_list = osd_xattr_list,
3382 .do_capa_get = osd_capa_get,
3383 .do_object_sync = osd_object_sync,
3384 .do_data_get = osd_data_get,
3388 * dt_object_operations for interoperability mode
3389 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3391 static const struct dt_object_operations osd_obj_ea_ops = {
3392 .do_read_lock = osd_object_read_lock,
3393 .do_write_lock = osd_object_write_lock,
3394 .do_read_unlock = osd_object_read_unlock,
3395 .do_write_unlock = osd_object_write_unlock,
3396 .do_write_locked = osd_object_write_locked,
3397 .do_attr_get = osd_attr_get,
3398 .do_declare_attr_set = osd_declare_attr_set,
3399 .do_attr_set = osd_attr_set,
3400 .do_ah_init = osd_ah_init,
3401 .do_declare_create = osd_declare_object_create,
3402 .do_create = osd_object_ea_create,
3403 .do_declare_destroy = osd_declare_object_destroy,
3404 .do_destroy = osd_object_destroy,
3405 .do_index_try = osd_index_try,
3406 .do_declare_ref_add = osd_declare_object_ref_add,
3407 .do_ref_add = osd_object_ref_add,
3408 .do_declare_ref_del = osd_declare_object_ref_del,
3409 .do_ref_del = osd_object_ref_del,
3410 .do_xattr_get = osd_xattr_get,
3411 .do_declare_xattr_set = osd_declare_xattr_set,
3412 .do_xattr_set = osd_xattr_set,
3413 .do_declare_xattr_del = osd_declare_xattr_del,
3414 .do_xattr_del = osd_xattr_del,
3415 .do_xattr_list = osd_xattr_list,
3416 .do_capa_get = osd_capa_get,
3417 .do_object_sync = osd_object_sync,
3418 .do_data_get = osd_data_get,
3421 static const struct dt_object_operations osd_obj_otable_it_ops = {
3422 .do_attr_get = osd_otable_it_attr_get,
3423 .do_index_try = osd_index_try,
3426 static int osd_index_declare_iam_delete(const struct lu_env *env,
3427 struct dt_object *dt,
3428 const struct dt_key *key,
3429 struct thandle *handle)
3431 struct osd_thandle *oh;
3433 oh = container_of0(handle, struct osd_thandle, ot_super);
3434 LASSERT(oh->ot_handle == NULL);
3436 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3437 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3443 * delete a (key, value) pair from index \a dt specified by \a key
3445 * \param dt osd index object
3446 * \param key key for index
3447 * \param rec record reference
3448 * \param handle transaction handler
3451 * \retval -ve failure
3454 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3455 const struct dt_key *key,
3456 struct thandle *handle,
3457 struct lustre_capa *capa)
3459 struct osd_thread_info *oti = osd_oti_get(env);
3460 struct osd_object *obj = osd_dt_obj(dt);
3461 struct osd_thandle *oh;
3462 struct iam_path_descr *ipd;
3463 struct iam_container *bag = &obj->oo_dir->od_container;
3468 LINVRNT(osd_invariant(obj));
3469 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3470 LASSERT(bag->ic_object == obj->oo_inode);
3471 LASSERT(handle != NULL);
3473 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3476 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3478 ipd = osd_idx_ipd_get(env, bag);
3479 if (unlikely(ipd == NULL))
3482 oh = container_of0(handle, struct osd_thandle, ot_super);
3483 LASSERT(oh->ot_handle != NULL);
3484 LASSERT(oh->ot_handle->h_transaction != NULL);
3486 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3487 /* swab quota uid/gid provided by caller */
3488 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3489 key = (const struct dt_key *)&oti->oti_quota_id;
3492 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3493 osd_ipd_put(env, bag, ipd);
3494 LINVRNT(osd_invariant(obj));
3498 static int osd_index_declare_ea_delete(const struct lu_env *env,
3499 struct dt_object *dt,
3500 const struct dt_key *key,
3501 struct thandle *handle)
3503 struct osd_thandle *oh;
3504 struct inode *inode;
3508 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3509 LASSERT(handle != NULL);
3511 oh = container_of0(handle, struct osd_thandle, ot_super);
3512 LASSERT(oh->ot_handle == NULL);
3514 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3515 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3517 inode = osd_dt_obj(dt)->oo_inode;
3520 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3521 true, true, NULL, false);
3525 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3528 struct osd_fid_pack *rec;
3531 if (de->file_type & LDISKFS_DIRENT_LUFID) {
3532 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3533 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3538 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3543 /* FID seqs not in FLDB, must be local seq */
3544 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3547 if (osd_seq_exists(env, osd, fid_seq(fid)))
3554 * Index delete function for interoperability mode (b11826).
3555 * It will remove the directory entry added by osd_index_ea_insert().
3556 * This entry is needed to maintain name->fid mapping.
3558 * \param key, key i.e. file entry to be deleted
3560 * \retval 0, on success
3561 * \retval -ve, on error
3563 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3564 const struct dt_key *key,
3565 struct thandle *handle,
3566 struct lustre_capa *capa)
3568 struct osd_object *obj = osd_dt_obj(dt);
3569 struct inode *dir = obj->oo_inode;
3570 struct dentry *dentry;
3571 struct osd_thandle *oh;
3572 struct ldiskfs_dir_entry_2 *de = NULL;
3573 struct buffer_head *bh;
3574 struct htree_lock *hlock = NULL;
3575 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
3576 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3580 LINVRNT(osd_invariant(obj));
3581 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3582 LASSERT(handle != NULL);
3584 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3586 oh = container_of(handle, struct osd_thandle, ot_super);
3587 LASSERT(oh->ot_handle != NULL);
3588 LASSERT(oh->ot_handle->h_transaction != NULL);
3590 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3593 ll_vfs_dq_init(dir);
3594 dentry = osd_child_dentry_get(env, obj,
3595 (char *)key, strlen((char *)key));
3597 if (obj->oo_hl_head != NULL) {
3598 hlock = osd_oti_get(env)->oti_hlock;
3599 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3600 dir, LDISKFS_HLOCK_DEL);
3602 down_write(&obj->oo_ext_idx_sem);
3605 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3609 /* If this is not the ".." entry, it might be a remote DNE
3610 * entry and we need to check if the FID is for a remote
3611 * MDT. If the FID is not in the directory entry (e.g.
3612 * upgraded 1.8 filesystem without dirdata enabled) then
3613 * we need to get the FID from the LMA. For a remote directory
3614 * there HAS to be an LMA, it cannot be an IGIF inode in this
3617 * Delete the entry before the agent inode in order to
3618 * simplify error handling. At worst an error after deleting
3619 * the entry first might leak the agent inode afterward. The
3620 * reverse would need filesystem abort in case of error deleting
3621 * the entry after the agent had been removed, or leave a
3622 * dangling entry pointing at a random inode. */
3623 if (strcmp((char *)key, dotdot) != 0) {
3624 LASSERT(de != NULL);
3625 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3626 /* If Fid is not in dentry, try to get it from LMA */
3627 if (rc == -ENODATA) {
3628 struct osd_inode_id *id;
3629 struct inode *inode;
3631 /* Before trying to get fid from the inode,
3632 * check whether the inode is valid.
3634 * If the inode has been deleted, do not go
3635 * ahead to do osd_ea_fid_get, which will set
3636 * the inode to bad inode, which might cause
3637 * the inode to be deleted uncorrectly */
3638 inode = ldiskfs_iget(osd_sb(osd),
3639 le32_to_cpu(de->inode));
3640 if (IS_ERR(inode)) {
3641 CDEBUG(D_INODE, "%s: "DFID"get inode"
3642 "error.\n", osd_name(osd),
3644 rc = PTR_ERR(inode);
3646 if (likely(inode->i_nlink != 0)) {
3647 id = &osd_oti_get(env)->oti_id;
3648 rc = osd_ea_fid_get(env, obj,
3649 le32_to_cpu(de->inode),
3652 CDEBUG(D_INFO, "%s: %u "DFID
3655 le32_to_cpu(de->inode),
3663 unlikely(osd_remote_fid(env, osd, fid)))
3664 /* Need to delete agent inode */
3665 ino = le32_to_cpu(de->inode);
3667 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3669 if (rc == 0 && unlikely(ino != 0)) {
3670 rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3673 CERROR("%s: del local inode "DFID": rc = %d\n",
3674 osd_name(osd), PFID(fid), rc);
3680 ldiskfs_htree_unlock(hlock);
3682 up_write(&obj->oo_ext_idx_sem);
3687 /* For inode on the remote MDT, .. will point to
3688 * /Agent directory, Check whether it needs to delete
3689 * from agent directory */
3690 if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3691 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3693 if (rc != 0 && rc != -ENOENT) {
3694 CERROR("%s: delete agent inode "DFID": rc = %d\n",
3695 osd_name(osd), PFID(fid), rc);
3705 LASSERT(osd_invariant(obj));
3710 * Lookup index for \a key and copy record to \a rec.
3712 * \param dt osd index object
3713 * \param key key for index
3714 * \param rec record reference
3716 * \retval +ve success : exact mach
3717 * \retval 0 return record with key not greater than \a key
3718 * \retval -ve failure
3720 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3721 struct dt_rec *rec, const struct dt_key *key,
3722 struct lustre_capa *capa)
3724 struct osd_object *obj = osd_dt_obj(dt);
3725 struct iam_path_descr *ipd;
3726 struct iam_container *bag = &obj->oo_dir->od_container;
3727 struct osd_thread_info *oti = osd_oti_get(env);
3728 struct iam_iterator *it = &oti->oti_idx_it;
3729 struct iam_rec *iam_rec;
3734 LASSERT(osd_invariant(obj));
3735 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3736 LASSERT(bag->ic_object == obj->oo_inode);
3738 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3741 ipd = osd_idx_ipd_get(env, bag);
3745 /* got ipd now we can start iterator. */
3746 iam_it_init(it, bag, 0, ipd);
3748 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3749 /* swab quota uid/gid provided by caller */
3750 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3751 key = (const struct dt_key *)&oti->oti_quota_id;
3754 rc = iam_it_get(it, (struct iam_key *)key);
3756 if (S_ISDIR(obj->oo_inode->i_mode))
3757 iam_rec = (struct iam_rec *)oti->oti_ldp;
3759 iam_rec = (struct iam_rec *) rec;
3761 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3763 if (S_ISDIR(obj->oo_inode->i_mode))
3764 osd_fid_unpack((struct lu_fid *) rec,
3765 (struct osd_fid_pack *)iam_rec);
3766 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3767 osd_quota_unpack(obj, rec);
3772 osd_ipd_put(env, bag, ipd);
3774 LINVRNT(osd_invariant(obj));
3779 static int osd_index_declare_iam_insert(const struct lu_env *env,
3780 struct dt_object *dt,
3781 const struct dt_rec *rec,
3782 const struct dt_key *key,
3783 struct thandle *handle)
3785 struct osd_thandle *oh;
3787 LASSERT(handle != NULL);
3789 oh = container_of0(handle, struct osd_thandle, ot_super);
3790 LASSERT(oh->ot_handle == NULL);
3792 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3793 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3799 * Inserts (key, value) pair in \a dt index object.
3801 * \param dt osd index object
3802 * \param key key for index
3803 * \param rec record reference
3804 * \param th transaction handler
3807 * \retval -ve failure
3809 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3810 const struct dt_rec *rec,
3811 const struct dt_key *key, struct thandle *th,
3812 struct lustre_capa *capa, int ignore_quota)
3814 struct osd_object *obj = osd_dt_obj(dt);
3815 struct iam_path_descr *ipd;
3816 struct osd_thandle *oh;
3817 struct iam_container *bag = &obj->oo_dir->od_container;
3818 struct osd_thread_info *oti = osd_oti_get(env);
3819 struct iam_rec *iam_rec;
3824 LINVRNT(osd_invariant(obj));
3825 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3826 LASSERT(bag->ic_object == obj->oo_inode);
3827 LASSERT(th != NULL);
3829 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3832 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3834 ipd = osd_idx_ipd_get(env, bag);
3835 if (unlikely(ipd == NULL))
3838 oh = container_of0(th, struct osd_thandle, ot_super);
3839 LASSERT(oh->ot_handle != NULL);
3840 LASSERT(oh->ot_handle->h_transaction != NULL);
3841 if (S_ISDIR(obj->oo_inode->i_mode)) {
3842 iam_rec = (struct iam_rec *)oti->oti_ldp;
3843 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3844 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3845 /* pack quota uid/gid */
3846 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3847 key = (const struct dt_key *)&oti->oti_quota_id;
3848 /* pack quota record */
3849 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3850 iam_rec = (struct iam_rec *)rec;
3852 iam_rec = (struct iam_rec *)rec;
3855 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3857 osd_ipd_put(env, bag, ipd);
3858 LINVRNT(osd_invariant(obj));
3863 * Calls ldiskfs_add_entry() to add directory entry
3864 * into the directory. This is required for
3865 * interoperability mode (b11826)
3867 * \retval 0, on success
3868 * \retval -ve, on error
3870 static int __osd_ea_add_rec(struct osd_thread_info *info,
3871 struct osd_object *pobj, struct inode *cinode,
3872 const char *name, const struct dt_rec *fid,
3873 struct htree_lock *hlock, struct thandle *th)
3875 struct ldiskfs_dentry_param *ldp;
3876 struct dentry *child;
3877 struct osd_thandle *oth;
3880 oth = container_of(th, struct osd_thandle, ot_super);
3881 LASSERT(oth->ot_handle != NULL);
3882 LASSERT(oth->ot_handle->h_transaction != NULL);
3883 LASSERT(pobj->oo_inode);
3885 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3886 if (unlikely(pobj->oo_inode ==
3887 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3890 osd_get_ldiskfs_dirent_param(ldp, fid);
3891 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3892 child->d_fsdata = (void *)ldp;
3893 ll_vfs_dq_init(pobj->oo_inode);
3894 rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3900 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3901 * into the directory.Also sets flags into osd object to
3902 * indicate dot and dotdot are created. This is required for
3903 * interoperability mode (b11826)
3905 * \param dir directory for dot and dotdot fixup.
3906 * \param obj child object for linking
3908 * \retval 0, on success
3909 * \retval -ve, on error
3911 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3912 struct osd_object *dir,
3913 struct inode *parent_dir, const char *name,
3914 const struct dt_rec *dot_fid,
3915 const struct dt_rec *dot_dot_fid,
3918 struct inode *inode = dir->oo_inode;
3919 struct osd_thandle *oth;
3922 oth = container_of(th, struct osd_thandle, ot_super);
3923 LASSERT(oth->ot_handle->h_transaction != NULL);
3924 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3926 if (strcmp(name, dot) == 0) {
3927 if (dir->oo_compat_dot_created) {
3930 LASSERT(inode == parent_dir);
3931 dir->oo_compat_dot_created = 1;
3934 } else if (strcmp(name, dotdot) == 0) {
3935 if (!dir->oo_compat_dot_created)
3937 /* in case of rename, dotdot is already created */
3938 if (dir->oo_compat_dotdot_created) {
3939 return __osd_ea_add_rec(info, dir, parent_dir, name,
3940 dot_dot_fid, NULL, th);
3943 result = osd_add_dot_dotdot_internal(info, dir->oo_inode,
3944 parent_dir, dot_fid,
3947 dir->oo_compat_dotdot_created = 1;
3955 * It will call the appropriate osd_add* function and return the
3956 * value, return by respective functions.
3958 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3959 struct inode *cinode, const char *name,
3960 const struct dt_rec *fid, struct thandle *th)
3962 struct osd_thread_info *info = osd_oti_get(env);
3963 struct htree_lock *hlock;
3966 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3968 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3970 if (hlock != NULL) {
3971 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3974 down_write(&pobj->oo_ext_idx_sem);
3976 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3977 (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3980 if (hlock != NULL) {
3981 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3982 pobj->oo_inode, LDISKFS_HLOCK_ADD);
3984 down_write(&pobj->oo_ext_idx_sem);
3987 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
3988 struct lu_fid *tfid = &info->oti_fid;
3990 *tfid = *(const struct lu_fid *)fid;
3992 rc = __osd_ea_add_rec(info, pobj, cinode, name,
3993 (const struct dt_rec *)tfid,
3996 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4001 ldiskfs_htree_unlock(hlock);
4003 up_write(&pobj->oo_ext_idx_sem);
4009 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4010 struct osd_idmap_cache *oic)
4012 struct osd_scrub *scrub = &dev->od_scrub;
4013 struct lu_fid *fid = &oic->oic_fid;
4014 struct osd_inode_id *id = &oti->oti_id;
4019 if (!fid_is_norm(fid) && !fid_is_igif(fid))
4023 rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
4024 if (rc != 0 && rc != -ENOENT)
4027 if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
4030 if (thread_is_running(&scrub->os_thread)) {
4031 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
4032 /* There is race condition between osd_oi_lookup and OI scrub.
4033 * The OI scrub finished just after osd_oi_lookup() failure.
4034 * Under such case, it is unnecessary to trigger OI scrub again,
4035 * but try to call osd_oi_lookup() again. */
4036 if (unlikely(rc == -EAGAIN))
4042 if (!dev->od_noscrub && ++once == 1) {
4043 rc = osd_scrub_start(dev);
4044 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC for "DFID
4046 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4055 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4056 struct osd_device *dev,
4057 struct osd_idmap_cache *oic,
4058 struct lu_fid *fid, __u32 ino)
4060 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
4061 struct inode *inode;
4064 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4065 inode = osd_iget(oti, dev, &oic->oic_lid);
4066 if (IS_ERR(inode)) {
4067 fid_zero(&oic->oic_fid);
4068 return PTR_ERR(inode);
4071 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4074 fid_zero(&oic->oic_fid);
4076 *fid = oic->oic_fid = lma->lma_self_fid;
4080 int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4081 struct osd_inode_id *id, const struct lu_fid *fid)
4083 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4084 id->oii_ino, id->oii_gen, info);
4085 info->oti_cache.oic_lid = *id;
4086 info->oti_cache.oic_fid = *fid;
4087 info->oti_cache.oic_dev = osd;
4093 * Calls ->lookup() to find dentry. From dentry get inode and
4094 * read inode's ea to get fid. This is required for interoperability
4097 * \retval 0, on success
4098 * \retval -ve, on error
4100 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4101 struct dt_rec *rec, const struct dt_key *key)
4103 struct inode *dir = obj->oo_inode;
4104 struct dentry *dentry;
4105 struct ldiskfs_dir_entry_2 *de;
4106 struct buffer_head *bh;
4107 struct lu_fid *fid = (struct lu_fid *) rec;
4108 struct htree_lock *hlock = NULL;
4113 LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
4115 dentry = osd_child_dentry_get(env, obj,
4116 (char *)key, strlen((char *)key));
4118 if (obj->oo_hl_head != NULL) {
4119 hlock = osd_oti_get(env)->oti_hlock;
4120 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4121 dir, LDISKFS_HLOCK_LOOKUP);
4123 down_read(&obj->oo_ext_idx_sem);
4126 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4128 struct osd_thread_info *oti = osd_oti_get(env);
4129 struct osd_inode_id *id = &oti->oti_id;
4130 struct osd_idmap_cache *oic = &oti->oti_cache;
4131 struct osd_device *dev = osd_obj2dev(obj);
4132 struct osd_scrub *scrub = &dev->od_scrub;
4133 struct scrub_file *sf = &scrub->os_file;
4135 ino = le32_to_cpu(de->inode);
4136 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4138 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4142 rc = osd_get_fid_from_dentry(de, rec);
4144 /* done with de, release bh */
4147 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4149 osd_id_gen(id, ino, OSD_OII_NOGEN);
4151 fid_zero(&oic->oic_fid);
4155 if (osd_remote_fid(env, dev, fid))
4158 rc = osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id,
4162 if ((scrub->os_pos_current <= ino) &&
4163 ((sf->sf_flags & SF_INCONSISTENT) ||
4164 (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
4165 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
4167 osd_consistency_check(oti, dev, oic);
4176 ldiskfs_htree_unlock(hlock);
4178 up_read(&obj->oo_ext_idx_sem);
4183 * Find the osd object for given fid.
4185 * \param fid need to find the osd object having this fid
4187 * \retval osd_object on success
4188 * \retval -ve on error
4190 struct osd_object *osd_object_find(const struct lu_env *env,
4191 struct dt_object *dt,
4192 const struct lu_fid *fid)
4194 struct lu_device *ludev = dt->do_lu.lo_dev;
4195 struct osd_object *child = NULL;
4196 struct lu_object *luch;
4197 struct lu_object *lo;
4200 * at this point topdev might not exist yet
4201 * (i.e. MGS is preparing profiles). so we can
4202 * not rely on topdev and instead lookup with
4203 * our device passed as topdev. this can't work
4204 * if the object isn't cached yet (as osd doesn't
4205 * allocate lu_header). IOW, the object must be
4206 * in the cache, otherwise lu_object_alloc() crashes
4209 luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4210 ludev : ludev->ld_site->ls_top_dev,
4212 if (!IS_ERR(luch)) {
4213 if (lu_object_exists(luch)) {
4214 lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4216 child = osd_obj(lo);
4218 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4219 "lu_object can't be located"
4220 DFID"\n", PFID(fid));
4222 if (child == NULL) {
4223 lu_object_put(env, luch);
4224 CERROR("Unable to get osd_object\n");
4225 child = ERR_PTR(-ENOENT);
4228 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4229 "lu_object does not exists "DFID"\n",
4231 lu_object_put(env, luch);
4232 child = ERR_PTR(-ENOENT);
4235 child = ERR_CAST(luch);
4242 * Put the osd object once done with it.
4244 * \param obj osd object that needs to be put
4246 static inline void osd_object_put(const struct lu_env *env,
4247 struct osd_object *obj)
4249 lu_object_put(env, &obj->oo_dt.do_lu);
4252 static int osd_index_declare_ea_insert(const struct lu_env *env,
4253 struct dt_object *dt,
4254 const struct dt_rec *rec,
4255 const struct dt_key *key,
4256 struct thandle *handle)
4258 struct osd_thandle *oh;
4259 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4260 struct lu_fid *fid = (struct lu_fid *)rec;
4264 LASSERT(!dt_object_remote(dt));
4265 LASSERT(handle != NULL);
4267 oh = container_of0(handle, struct osd_thandle, ot_super);
4268 LASSERT(oh->ot_handle == NULL);
4270 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4271 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4273 if (osd_dt_obj(dt)->oo_inode != NULL) {
4274 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4276 /* We ignore block quota on meta pool (MDTs), so needn't
4277 * calculate how many blocks will be consumed by this index
4279 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0,
4280 oh, true, true, NULL, false);
4286 rc = osd_remote_fid(env, osd, fid);
4292 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
4293 osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
4294 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4295 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4296 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4297 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4303 * Index add function for interoperability mode (b11826).
4304 * It will add the directory entry.This entry is needed to
4305 * maintain name->fid mapping.
4307 * \param key it is key i.e. file entry to be inserted
4308 * \param rec it is value of given key i.e. fid
4310 * \retval 0, on success
4311 * \retval -ve, on error
4313 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4314 const struct dt_rec *rec,
4315 const struct dt_key *key, struct thandle *th,
4316 struct lustre_capa *capa, int ignore_quota)
4318 struct osd_object *obj = osd_dt_obj(dt);
4319 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4320 struct lu_fid *fid = (struct lu_fid *) rec;
4321 const char *name = (const char *)key;
4322 struct osd_thread_info *oti = osd_oti_get(env);
4323 struct osd_inode_id *id = &oti->oti_id;
4324 struct inode *child_inode = NULL;
4325 struct osd_object *child = NULL;
4329 LASSERT(osd_invariant(obj));
4330 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
4331 LASSERT(th != NULL);
4333 osd_trans_exec_op(env, th, OSD_OT_INSERT);
4335 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
4338 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!", PFID(fid));
4340 rc = osd_remote_fid(env, osd, fid);
4342 CERROR("%s: Can not find object "DFID" rc %d\n",
4343 osd_name(osd), PFID(fid), rc);
4348 /* Insert remote entry */
4349 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4350 struct osd_mdobj_map *omm = osd->od_mdt_map;
4351 struct osd_thandle *oh;
4353 /* If parent on remote MDT, we need put this object
4355 oh = container_of(th, typeof(*oh), ot_super);
4356 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4358 CERROR("%s: add "DFID" error: rc = %d\n",
4360 PFID(lu_object_fid(&dt->do_lu)), rc);
4364 child_inode = igrab(omm->omm_remote_parent->d_inode);
4366 child_inode = osd_create_local_agent_inode(env, osd,
4369 if (IS_ERR(child_inode))
4370 RETURN(PTR_ERR(child_inode));
4373 /* Insert local entry */
4374 child = osd_object_find(env, dt, fid);
4375 if (IS_ERR(child)) {
4376 CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4377 osd_name(osd), PFID(fid),
4378 id->oii_ino, id->oii_gen,
4379 (int)PTR_ERR(child));
4380 RETURN(PTR_ERR(child));
4382 child_inode = igrab(child->oo_inode);
4385 rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th);
4387 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4388 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4392 osd_object_put(env, child);
4393 LASSERT(osd_invariant(obj));
4398 * Initialize osd Iterator for given osd index object.
4400 * \param dt osd index object
4403 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4404 struct dt_object *dt,
4406 struct lustre_capa *capa)
4408 struct osd_it_iam *it;
4409 struct osd_thread_info *oti = osd_oti_get(env);
4410 struct osd_object *obj = osd_dt_obj(dt);
4411 struct lu_object *lo = &dt->do_lu;
4412 struct iam_path_descr *ipd;
4413 struct iam_container *bag = &obj->oo_dir->od_container;
4415 LASSERT(lu_object_exists(lo));
4417 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
4418 return ERR_PTR(-EACCES);
4421 ipd = osd_it_ipd_get(env, bag);
4422 if (likely(ipd != NULL)) {
4426 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4427 return (struct dt_it *)it;
4429 return ERR_PTR(-ENOMEM);
4433 * free given Iterator.
4436 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4438 struct osd_it_iam *it = (struct osd_it_iam *)di;
4439 struct osd_object *obj = it->oi_obj;
4441 iam_it_fini(&it->oi_it);
4442 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4443 lu_object_put(env, &obj->oo_dt.do_lu);
4447 * Move Iterator to record specified by \a key
4449 * \param di osd iterator
4450 * \param key key for index
4452 * \retval +ve di points to record with least key not larger than key
4453 * \retval 0 di points to exact matched key
4454 * \retval -ve failure
4457 static int osd_it_iam_get(const struct lu_env *env,
4458 struct dt_it *di, const struct dt_key *key)
4460 struct osd_thread_info *oti = osd_oti_get(env);
4461 struct osd_it_iam *it = (struct osd_it_iam *)di;
4463 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4464 /* swab quota uid/gid */
4465 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4466 key = (struct dt_key *)&oti->oti_quota_id;
4469 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4475 * \param di osd iterator
4477 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4479 struct osd_it_iam *it = (struct osd_it_iam *)di;
4481 iam_it_put(&it->oi_it);
4485 * Move iterator by one record
4487 * \param di osd iterator
4489 * \retval +1 end of container reached
4491 * \retval -ve failure
4494 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4496 struct osd_it_iam *it = (struct osd_it_iam *)di;
4498 return iam_it_next(&it->oi_it);
4502 * Return pointer to the key under iterator.
4505 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4506 const struct dt_it *di)
4508 struct osd_thread_info *oti = osd_oti_get(env);
4509 struct osd_it_iam *it = (struct osd_it_iam *)di;
4510 struct osd_object *obj = it->oi_obj;
4513 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4515 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4516 /* swab quota uid/gid */
4517 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4518 key = (struct dt_key *)&oti->oti_quota_id;
4525 * Return size of key under iterator (in bytes)
4528 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4530 struct osd_it_iam *it = (struct osd_it_iam *)di;
4532 return iam_it_key_size(&it->oi_it);
4536 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4538 /* check if file type is required */
4539 if (ent->lde_attrs & LUDA_TYPE) {
4540 struct luda_type *lt;
4541 int align = sizeof(*lt) - 1;
4543 len = (len + align) & ~align;
4544 lt = (struct luda_type *)(ent->lde_name + len);
4545 lt->lt_type = cpu_to_le16(DTTOIF(type));
4548 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4552 * build lu direct from backend fs dirent.
4556 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4557 char *name, __u16 namelen, __u16 type, __u32 attr)
4559 ent->lde_attrs = attr | LUDA_FID;
4560 fid_cpu_to_le(&ent->lde_fid, fid);
4562 ent->lde_hash = cpu_to_le64(offset);
4563 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4565 strncpy(ent->lde_name, name, namelen);
4566 ent->lde_name[namelen] = '\0';
4567 ent->lde_namelen = cpu_to_le16(namelen);
4569 /* append lustre attributes */
4570 osd_it_append_attrs(ent, namelen, type);
4574 * Return pointer to the record under iterator.
4576 static int osd_it_iam_rec(const struct lu_env *env,
4577 const struct dt_it *di,
4578 struct dt_rec *dtrec, __u32 attr)
4580 struct osd_it_iam *it = (struct osd_it_iam *)di;
4581 struct osd_thread_info *info = osd_oti_get(env);
4584 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4585 const struct osd_fid_pack *rec;
4586 struct lu_fid *fid = &info->oti_fid;
4587 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
4593 name = (char *)iam_it_key_get(&it->oi_it);
4595 RETURN(PTR_ERR(name));
4597 namelen = iam_it_key_size(&it->oi_it);
4599 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4601 RETURN(PTR_ERR(rec));
4603 rc = osd_fid_unpack(fid, rec);
4607 hash = iam_it_store(&it->oi_it);
4609 /* IAM does not store object type in IAM index (dir) */
4610 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4612 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4613 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4614 (struct iam_rec *)dtrec);
4615 osd_quota_unpack(it->oi_obj, dtrec);
4617 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4618 (struct iam_rec *)dtrec);
4625 * Returns cookie for current Iterator position.
4627 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4629 struct osd_it_iam *it = (struct osd_it_iam *)di;
4631 return iam_it_store(&it->oi_it);
4635 * Restore iterator from cookie.
4637 * \param di osd iterator
4638 * \param hash Iterator location cookie
4640 * \retval +ve di points to record with least key not larger than key.
4641 * \retval 0 di points to exact matched key
4642 * \retval -ve failure
4645 static int osd_it_iam_load(const struct lu_env *env,
4646 const struct dt_it *di, __u64 hash)
4648 struct osd_it_iam *it = (struct osd_it_iam *)di;
4650 return iam_it_load(&it->oi_it, hash);
4653 static const struct dt_index_operations osd_index_iam_ops = {
4654 .dio_lookup = osd_index_iam_lookup,
4655 .dio_declare_insert = osd_index_declare_iam_insert,
4656 .dio_insert = osd_index_iam_insert,
4657 .dio_declare_delete = osd_index_declare_iam_delete,
4658 .dio_delete = osd_index_iam_delete,
4660 .init = osd_it_iam_init,
4661 .fini = osd_it_iam_fini,
4662 .get = osd_it_iam_get,
4663 .put = osd_it_iam_put,
4664 .next = osd_it_iam_next,
4665 .key = osd_it_iam_key,
4666 .key_size = osd_it_iam_key_size,
4667 .rec = osd_it_iam_rec,
4668 .store = osd_it_iam_store,
4669 .load = osd_it_iam_load
4675 * Creates or initializes iterator context.
4677 * \retval struct osd_it_ea, iterator structure on success
4680 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4681 struct dt_object *dt,
4683 struct lustre_capa *capa)
4685 struct osd_object *obj = osd_dt_obj(dt);
4686 struct osd_thread_info *info = osd_oti_get(env);
4687 struct osd_it_ea *it = &info->oti_it_ea;
4688 struct file *file = &it->oie_file;
4689 struct lu_object *lo = &dt->do_lu;
4690 struct dentry *obj_dentry = &info->oti_it_dentry;
4692 LASSERT(lu_object_exists(lo));
4694 obj_dentry->d_inode = obj->oo_inode;
4695 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4696 obj_dentry->d_name.hash = 0;
4698 it->oie_rd_dirent = 0;
4699 it->oie_it_dirent = 0;
4700 it->oie_dirent = NULL;
4701 it->oie_buf = info->oti_it_ea_buf;
4704 /* Reset the "file" totally to avoid to reuse any old value from
4705 * former readdir handling, the "file->f_pos" should be zero. */
4706 memset(file, 0, sizeof(*file));
4707 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4708 if (attr & LUDA_64BITHASH)
4709 file->f_mode = FMODE_64BITHASH;
4711 file->f_mode = FMODE_32BITHASH;
4712 file->f_dentry = obj_dentry;
4713 file->f_mapping = obj->oo_inode->i_mapping;
4714 file->f_op = obj->oo_inode->i_fop;
4715 set_file_inode(file, obj->oo_inode);
4718 RETURN((struct dt_it *) it);
4722 * Destroy or finishes iterator context.
4724 * \param di iterator structure to be destroyed
4726 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4728 struct osd_it_ea *it = (struct osd_it_ea *)di;
4729 struct osd_object *obj = it->oie_obj;
4730 struct inode *inode = obj->oo_inode;
4733 it->oie_file.f_op->release(inode, &it->oie_file);
4734 lu_object_put(env, &obj->oo_dt.do_lu);
4739 * It position the iterator at given key, so that next lookup continues from
4740 * that key Or it is similar to dio_it->load() but based on a key,
4741 * rather than file position.
4743 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4746 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4748 static int osd_it_ea_get(const struct lu_env *env,
4749 struct dt_it *di, const struct dt_key *key)
4751 struct osd_it_ea *it = (struct osd_it_ea *)di;
4754 LASSERT(((const char *)key)[0] == '\0');
4755 it->oie_file.f_pos = 0;
4756 it->oie_rd_dirent = 0;
4757 it->oie_it_dirent = 0;
4758 it->oie_dirent = NULL;
4766 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4770 struct osd_filldir_cbs {
4771 #ifdef HAVE_DIR_CONTEXT
4772 struct dir_context ctx;
4774 struct osd_it_ea *it;
4777 * It is called internally by ->readdir(). It fills the
4778 * iterator's in-memory data structure with required
4779 * information i.e. name, namelen, rec_size etc.
4781 * \param buf in which information to be filled in.
4782 * \param name name of the file in given dir
4784 * \retval 0 on success
4785 * \retval 1 on buffer full
4787 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
4788 loff_t offset, __u64 ino,
4791 struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
4792 struct osd_object *obj = it->oie_obj;
4793 struct osd_it_ea_dirent *ent = it->oie_dirent;
4794 struct lu_fid *fid = &ent->oied_fid;
4795 struct osd_fid_pack *rec;
4798 /* this should never happen */
4799 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4800 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4804 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4808 /* "." is just the object itself. */
4809 if (namelen == 1 && name[0] == '.') {
4810 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4811 } else if (d_type & LDISKFS_DIRENT_LUFID) {
4812 rec = (struct osd_fid_pack*) (name + namelen + 1);
4813 if (osd_fid_unpack(fid, rec) != 0)
4818 d_type &= ~LDISKFS_DIRENT_LUFID;
4820 /* NOT export local root. */
4821 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
4822 ino = obj->oo_inode->i_ino;
4823 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4826 ent->oied_ino = ino;
4827 ent->oied_off = offset;
4828 ent->oied_namelen = namelen;
4829 ent->oied_type = d_type;
4831 memcpy(ent->oied_name, name, namelen);
4833 it->oie_rd_dirent++;
4834 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4839 * Calls ->readdir() to load a directory entry at a time
4840 * and stored it in iterator's in-memory data structure.
4842 * \param di iterator's in memory structure
4844 * \retval 0 on success
4845 * \retval -ve on error
4846 * \retval +1 reach the end of entry
4848 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4849 const struct dt_it *di)
4851 struct osd_it_ea *it = (struct osd_it_ea *)di;
4852 struct osd_object *obj = it->oie_obj;
4853 struct inode *inode = obj->oo_inode;
4854 struct htree_lock *hlock = NULL;
4855 struct file *filp = &it->oie_file;
4857 struct osd_filldir_cbs buf = {
4858 #ifdef HAVE_DIR_CONTEXT
4859 .ctx.actor = osd_ldiskfs_filldir,
4865 it->oie_dirent = it->oie_buf;
4866 it->oie_rd_dirent = 0;
4868 if (obj->oo_hl_head != NULL) {
4869 hlock = osd_oti_get(env)->oti_hlock;
4870 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4871 inode, LDISKFS_HLOCK_READDIR);
4873 down_read(&obj->oo_ext_idx_sem);
4876 #ifdef HAVE_DIR_CONTEXT
4877 buf.ctx.pos = filp->f_pos;
4878 rc = inode->i_fop->iterate(filp, &buf.ctx);
4879 filp->f_pos = buf.ctx.pos;
4881 rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
4885 ldiskfs_htree_unlock(hlock);
4887 up_read(&obj->oo_ext_idx_sem);
4889 if (it->oie_rd_dirent == 0) {
4890 /*If it does not get any dirent, it means it has been reached
4891 *to the end of the dir */
4892 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
4896 it->oie_dirent = it->oie_buf;
4897 it->oie_it_dirent = 1;
4904 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4905 * to load a directory entry at a time and stored it in
4906 * iterator's in-memory data structure.
4908 * \param di iterator's in memory structure
4910 * \retval +ve iterator reached to end
4911 * \retval 0 iterator not reached to end
4912 * \retval -ve on error
4914 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4916 struct osd_it_ea *it = (struct osd_it_ea *)di;
4921 if (it->oie_it_dirent < it->oie_rd_dirent) {
4923 (void *) it->oie_dirent +
4924 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4925 it->oie_dirent->oied_namelen);
4926 it->oie_it_dirent++;
4929 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4932 rc = osd_ldiskfs_it_fill(env, di);
4939 * Returns the key at current position from iterator's in memory structure.
4941 * \param di iterator's in memory structure
4943 * \retval key i.e. struct dt_key on success
4945 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4946 const struct dt_it *di)
4948 struct osd_it_ea *it = (struct osd_it_ea *)di;
4950 return (struct dt_key *)it->oie_dirent->oied_name;
4954 * Returns the key's size at current position from iterator's in memory structure.
4956 * \param di iterator's in memory structure
4958 * \retval key_size i.e. struct dt_key on success
4960 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4962 struct osd_it_ea *it = (struct osd_it_ea *)di;
4964 return it->oie_dirent->oied_namelen;
4968 osd_dirent_update(handle_t *jh, struct super_block *sb,
4969 struct osd_it_ea_dirent *ent, struct lu_fid *fid,
4970 struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
4972 struct osd_fid_pack *rec;
4976 LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
4977 LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
4979 rc = ldiskfs_journal_get_write_access(jh, bh);
4981 CERROR("%.16s: fail to write access for update dirent: "
4982 "name = %.*s, rc = %d\n",
4983 LDISKFS_SB(sb)->s_es->s_volume_name,
4984 ent->oied_namelen, ent->oied_name, rc);
4988 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
4989 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
4990 rc = ldiskfs_journal_dirty_metadata(jh, bh);
4992 CERROR("%.16s: fail to dirty metadata for update dirent: "
4993 "name = %.*s, rc = %d\n",
4994 LDISKFS_SB(sb)->s_es->s_volume_name,
4995 ent->oied_namelen, ent->oied_name, rc);
5001 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
5003 if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
5004 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5011 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5013 LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5014 "dot_dotdot = %d\n", dot_dotdot);
5016 if (LDISKFS_DIR_REC_LEN(de) >=
5017 __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5024 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5025 struct inode *dir, struct inode *inode,
5026 struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5027 struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5028 struct htree_lock *hlock)
5030 struct dentry *dentry;
5031 struct osd_fid_pack *rec;
5032 struct ldiskfs_dentry_param *ldp;
5036 if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5037 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5040 /* There is enough space to hold the FID-in-dirent. */
5041 if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
5042 dir->i_sb->s_blocksize)) {
5043 rc = ldiskfs_journal_get_write_access(jh, bh);
5045 CERROR("%.16s: fail to write access for reinsert "
5046 "dirent: name = %.*s, rc = %d\n",
5047 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5048 ent->oied_namelen, ent->oied_name, rc);
5052 de->name[de->name_len] = 0;
5053 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5054 rec->fp_len = sizeof(struct lu_fid) + 1;
5055 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5056 de->file_type |= LDISKFS_DIRENT_LUFID;
5058 rc = ldiskfs_journal_dirty_metadata(jh, bh);
5060 CERROR("%.16s: fail to dirty metadata for reinsert "
5061 "dirent: name = %.*s, rc = %d\n",
5062 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5063 ent->oied_namelen, ent->oied_name, rc);
5068 rc = ldiskfs_delete_entry(jh, dir, de, bh);
5070 CERROR("%.16s: fail to delete entry for reinsert dirent: "
5071 "name = %.*s, rc = %d\n",
5072 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5073 ent->oied_namelen, ent->oied_name, rc);
5077 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5079 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5080 osd_get_ldiskfs_dirent_param(ldp, (const struct dt_rec *)fid);
5081 dentry->d_fsdata = (void *)ldp;
5082 ll_vfs_dq_init(dir);
5083 rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5084 /* It is too bad, we cannot reinsert the name entry back.
5085 * That means we lose it! */
5087 CERROR("%.16s: fail to insert entry for reinsert dirent: "
5088 "name = %.*s, rc = %d\n",
5089 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5090 ent->oied_namelen, ent->oied_name, rc);
5096 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5097 struct osd_it_ea *it, struct lu_fid *fid,
5098 struct osd_inode_id *id, __u32 *attr)
5100 struct osd_thread_info *info = osd_oti_get(env);
5101 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
5102 struct osd_device *dev = osd_obj2dev(obj);
5103 struct super_block *sb = osd_sb(dev);
5104 const char *devname =
5105 LDISKFS_SB(sb)->s_es->s_volume_name;
5106 struct osd_it_ea_dirent *ent = it->oie_dirent;
5107 struct inode *dir = obj->oo_inode;
5108 struct htree_lock *hlock = NULL;
5109 struct buffer_head *bh = NULL;
5110 handle_t *jh = NULL;
5111 struct ldiskfs_dir_entry_2 *de;
5112 struct dentry *dentry;
5113 struct inode *inode;
5120 if (ent->oied_name[0] == '.') {
5121 if (ent->oied_namelen == 1)
5123 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5127 dentry = osd_child_dentry_get(env, obj, ent->oied_name,
5130 /* We need to ensure that the name entry is still valid.
5131 * Because it may be removed or renamed by other already.
5133 * The unlink or rename operation will start journal before PDO lock,
5134 * so to avoid deadlock, here we need to start journal handle before
5135 * related PDO lock also. But because we do not know whether there
5136 * will be something to be repaired before PDO lock, we just start
5137 * journal without conditions.
5139 * We may need to remove the name entry firstly, then insert back.
5140 * One credit is for user quota file update.
5141 * One credit is for group quota file update.
5142 * Two credits are for dirty inode. */
5143 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5144 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5147 if (dev->od_dirent_journal) {
5148 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5151 CERROR("%.16s: fail to start trans for dirent "
5152 "check_repair: credits %d, name %.*s, rc %d\n",
5153 devname, credits, ent->oied_namelen,
5154 ent->oied_name, rc);
5158 if (obj->oo_hl_head != NULL) {
5159 hlock = osd_oti_get(env)->oti_hlock;
5160 /* "0" means exclusive lock for the whole directory.
5161 * We need to prevent others access such name entry
5162 * during the delete + insert. Neither HLOCK_ADD nor
5163 * HLOCK_DEL cannot guarantee the atomicity. */
5164 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5166 down_write(&obj->oo_ext_idx_sem);
5169 if (obj->oo_hl_head != NULL) {
5170 hlock = osd_oti_get(env)->oti_hlock;
5171 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5172 LDISKFS_HLOCK_LOOKUP);
5174 down_read(&obj->oo_ext_idx_sem);
5178 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5179 /* For dot/dotdot entry, if there is not enough space to hold the
5180 * FID-in-dirent, just keep them there. It only happens when the
5181 * device upgraded from 1.8 or restored from MDT file-level backup.
5182 * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5183 * and needs to get FID from LMA when readdir, it will not affect the
5184 * performance much. */
5185 if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
5186 (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5187 *attr |= LUDA_IGNORE;
5188 GOTO(out_journal, rc = 0);
5191 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5192 inode = osd_iget(info, dev, id);
5193 if (IS_ERR(inode)) {
5194 rc = PTR_ERR(inode);
5195 if (rc == -ENOENT || rc == -ESTALE) {
5196 *attr |= LUDA_IGNORE;
5200 GOTO(out_journal, rc);
5203 /* skip the REMOTE_PARENT_DIR. */
5204 if (inode == dev->od_mdt_map->omm_remote_parent->d_inode)
5205 GOTO(out_inode, rc = 0);
5207 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
5209 LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
5211 if (fid_is_sane(fid)) {
5212 /* FID-in-dirent is valid. */
5213 if (lu_fid_eq(fid, &lma->lma_self_fid))
5214 GOTO(out_inode, rc = 0);
5216 /* Do not repair under dryrun mode. */
5217 if (*attr & LUDA_VERIFY_DRYRUN) {
5218 *attr |= LUDA_REPAIR;
5219 GOTO(out_inode, rc = 0);
5222 if (!dev->od_dirent_journal) {
5226 ldiskfs_htree_unlock(hlock);
5228 up_read(&obj->oo_ext_idx_sem);
5229 dev->od_dirent_journal = 1;
5233 *fid = lma->lma_self_fid;
5235 /* Update the FID-in-dirent. */
5236 rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
5238 *attr |= LUDA_REPAIR;
5240 /* Do not repair under dryrun mode. */
5241 if (*attr & LUDA_VERIFY_DRYRUN) {
5242 *fid = lma->lma_self_fid;
5243 *attr |= LUDA_REPAIR;
5244 GOTO(out_inode, rc = 0);
5247 if (!dev->od_dirent_journal) {
5251 ldiskfs_htree_unlock(hlock);
5253 up_read(&obj->oo_ext_idx_sem);
5254 dev->od_dirent_journal = 1;
5258 *fid = lma->lma_self_fid;
5260 /* Append the FID-in-dirent. */
5261 rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5262 fid, bh, de, hlock);
5264 *attr |= LUDA_REPAIR;
5266 } else if (rc == -ENODATA) {
5267 /* Do not repair under dryrun mode. */
5268 if (*attr & LUDA_VERIFY_DRYRUN) {
5269 if (fid_is_sane(fid)) {
5270 *attr |= LUDA_REPAIR;
5272 lu_igif_build(fid, inode->i_ino,
5273 inode->i_generation);
5274 *attr |= LUDA_UPGRADE;
5276 GOTO(out_inode, rc = 0);
5279 if (!dev->od_dirent_journal) {
5283 ldiskfs_htree_unlock(hlock);
5285 up_read(&obj->oo_ext_idx_sem);
5286 dev->od_dirent_journal = 1;
5291 if (unlikely(fid_is_sane(fid))) {
5292 /* FID-in-dirent exists, but FID-in-LMA is lost.
5293 * Trust the FID-in-dirent, and add FID-in-LMA. */
5294 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5296 *attr |= LUDA_REPAIR;
5298 lu_igif_build(fid, inode->i_ino, inode->i_generation);
5299 /* It is probably IGIF object. Only aappend the
5300 * FID-in-dirent. OI scrub will process FID-in-LMA. */
5301 rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5302 fid, bh, de, hlock);
5304 *attr |= LUDA_UPGRADE;
5308 GOTO(out_inode, rc);
5315 if (hlock != NULL) {
5316 ldiskfs_htree_unlock(hlock);
5318 if (dev->od_dirent_journal)
5319 up_write(&obj->oo_ext_idx_sem);
5321 up_read(&obj->oo_ext_idx_sem);
5324 ldiskfs_journal_stop(jh);
5325 if (rc >= 0 && !dirty)
5326 dev->od_dirent_journal = 0;
5331 * Returns the value at current position from iterator's in memory structure.
5333 * \param di struct osd_it_ea, iterator's in memory structure
5334 * \param attr attr requested for dirent.
5335 * \param lde lustre dirent
5337 * \retval 0 no error and \param lde has correct lustre dirent.
5338 * \retval -ve on error
5340 static inline int osd_it_ea_rec(const struct lu_env *env,
5341 const struct dt_it *di,
5342 struct dt_rec *dtrec, __u32 attr)
5344 struct osd_it_ea *it = (struct osd_it_ea *)di;
5345 struct osd_object *obj = it->oie_obj;
5346 struct osd_device *dev = osd_obj2dev(obj);
5347 struct osd_scrub *scrub = &dev->od_scrub;
5348 struct scrub_file *sf = &scrub->os_file;
5349 struct osd_thread_info *oti = osd_oti_get(env);
5350 struct osd_inode_id *id = &oti->oti_id;
5351 struct osd_idmap_cache *oic = &oti->oti_cache;
5352 struct lu_fid *fid = &it->oie_dirent->oied_fid;
5353 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
5354 __u32 ino = it->oie_dirent->oied_ino;
5358 if (attr & LUDA_VERIFY) {
5360 if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) {
5361 attr |= LUDA_IGNORE;
5364 rc = osd_dirent_check_repair(env, obj, it, fid, id,
5368 attr &= ~LU_DIRENT_ATTRS_MASK;
5369 if (!fid_is_sane(fid)) {
5370 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP) &&
5371 likely(it->oie_dirent->oied_namelen != 2 ||
5372 it->oie_dirent->oied_name[0] != '.' ||
5373 it->oie_dirent->oied_name[1] != '.'))
5376 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5378 osd_id_gen(id, ino, OSD_OII_NOGEN);
5382 /* Pack the entry anyway, at least the offset is right. */
5383 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5384 it->oie_dirent->oied_name,
5385 it->oie_dirent->oied_namelen,
5386 it->oie_dirent->oied_type, attr);
5391 if (osd_remote_fid(env, dev, fid))
5394 if (likely(!(attr & LUDA_IGNORE)))
5395 rc = osd_add_oi_cache(oti, dev, id, fid);
5397 if (!(attr & LUDA_VERIFY) &&
5398 (scrub->os_pos_current <= ino) &&
5399 ((sf->sf_flags & SF_INCONSISTENT) ||
5400 (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
5401 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
5402 osd_consistency_check(oti, dev, oic);
5408 * Returns the record size size at current position.
5410 * This function will return record(lu_dirent) size in bytes.
5412 * \param[in] env execution environment
5413 * \param[in] di iterator's in memory structure
5414 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
5415 * calculate the lu_dirent size.
5417 * \retval record size(in bytes & in memory) of the current lu_dirent
5420 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5423 struct osd_it_ea *it = (struct osd_it_ea *)di;
5425 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5429 * Returns a cookie for current position of the iterator head, so that
5430 * user can use this cookie to load/start the iterator next time.
5432 * \param di iterator's in memory structure
5434 * \retval cookie for current position, on success
5436 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5438 struct osd_it_ea *it = (struct osd_it_ea *)di;
5440 return it->oie_dirent->oied_off;
5444 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5445 * to load a directory entry at a time and stored it i inn,
5446 * in iterator's in-memory data structure.
5448 * \param di struct osd_it_ea, iterator's in memory structure
5450 * \retval +ve on success
5451 * \retval -ve on error
5453 static int osd_it_ea_load(const struct lu_env *env,
5454 const struct dt_it *di, __u64 hash)
5456 struct osd_it_ea *it = (struct osd_it_ea *)di;
5460 it->oie_file.f_pos = hash;
5462 rc = osd_ldiskfs_it_fill(env, di);
5473 * Index lookup function for interoperability mode (b11826).
5475 * \param key, key i.e. file name to be searched
5477 * \retval +ve, on success
5478 * \retval -ve, on error
5480 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5481 struct dt_rec *rec, const struct dt_key *key,
5482 struct lustre_capa *capa)
5484 struct osd_object *obj = osd_dt_obj(dt);
5489 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5490 LINVRNT(osd_invariant(obj));
5492 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
5495 rc = osd_ea_lookup_rec(env, obj, rec, key);
5502 * Index and Iterator operations for interoperability
5503 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5505 static const struct dt_index_operations osd_index_ea_ops = {
5506 .dio_lookup = osd_index_ea_lookup,
5507 .dio_declare_insert = osd_index_declare_ea_insert,
5508 .dio_insert = osd_index_ea_insert,
5509 .dio_declare_delete = osd_index_declare_ea_delete,
5510 .dio_delete = osd_index_ea_delete,
5512 .init = osd_it_ea_init,
5513 .fini = osd_it_ea_fini,
5514 .get = osd_it_ea_get,
5515 .put = osd_it_ea_put,
5516 .next = osd_it_ea_next,
5517 .key = osd_it_ea_key,
5518 .key_size = osd_it_ea_key_size,
5519 .rec = osd_it_ea_rec,
5520 .rec_size = osd_it_ea_rec_size,
5521 .store = osd_it_ea_store,
5522 .load = osd_it_ea_load
5526 static void *osd_key_init(const struct lu_context *ctx,
5527 struct lu_context_key *key)
5529 struct osd_thread_info *info;
5531 OBD_ALLOC_PTR(info);
5533 return ERR_PTR(-ENOMEM);
5535 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5536 if (info->oti_it_ea_buf == NULL)
5539 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5541 info->oti_hlock = ldiskfs_htree_lock_alloc();
5542 if (info->oti_hlock == NULL)
5548 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5551 return ERR_PTR(-ENOMEM);
5554 static void osd_key_fini(const struct lu_context *ctx,
5555 struct lu_context_key *key, void* data)
5557 struct osd_thread_info *info = data;
5559 if (info->oti_hlock != NULL)
5560 ldiskfs_htree_lock_free(info->oti_hlock);
5561 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5562 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5563 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5567 static void osd_key_exit(const struct lu_context *ctx,
5568 struct lu_context_key *key, void *data)
5570 struct osd_thread_info *info = data;
5572 LASSERT(info->oti_r_locks == 0);
5573 LASSERT(info->oti_w_locks == 0);
5574 LASSERT(info->oti_txns == 0);
5577 /* type constructor/destructor: osd_type_init, osd_type_fini */
5578 LU_TYPE_INIT_FINI(osd, &osd_key);
5580 struct lu_context_key osd_key = {
5581 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5582 .lct_init = osd_key_init,
5583 .lct_fini = osd_key_fini,
5584 .lct_exit = osd_key_exit
5588 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5589 const char *name, struct lu_device *next)
5591 struct osd_device *osd = osd_dev(d);
5593 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5594 >= sizeof(osd->od_svname))
5596 return osd_procfs_init(osd, name);
5599 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5601 struct seq_server_site *ss = osd_seq_site(osd);
5605 if (osd->od_is_ost || osd->od_cl_seq != NULL)
5608 if (unlikely(ss == NULL))
5611 OBD_ALLOC_PTR(osd->od_cl_seq);
5612 if (osd->od_cl_seq == NULL)
5615 rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5616 osd->od_svname, ss->ss_server_seq);
5619 OBD_FREE_PTR(osd->od_cl_seq);
5620 osd->od_cl_seq = NULL;
5626 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5628 if (osd->od_cl_seq == NULL)
5631 seq_client_fini(osd->od_cl_seq);
5632 OBD_FREE_PTR(osd->od_cl_seq);
5633 osd->od_cl_seq = NULL;
5636 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5640 /* shutdown quota slave instance associated with the device */
5641 if (o->od_quota_slave != NULL) {
5642 qsd_fini(env, o->od_quota_slave);
5643 o->od_quota_slave = NULL;
5646 osd_fid_fini(env, o);
5651 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5655 if (o->od_mnt != NULL) {
5656 shrink_dcache_sb(osd_sb(o));
5657 osd_sync(env, &o->od_dt_dev);
5666 static int osd_mount(const struct lu_env *env,
5667 struct osd_device *o, struct lustre_cfg *cfg)
5669 const char *name = lustre_cfg_string(cfg, 0);
5670 const char *dev = lustre_cfg_string(cfg, 1);
5672 unsigned long page, s_flags, lmd_flags = 0;
5673 struct page *__page;
5674 struct file_system_type *type;
5675 char *options = NULL;
5677 struct osd_thread_info *info = osd_oti_get(env);
5678 struct lu_fid *fid = &info->oti_fid;
5679 struct inode *inode;
5683 if (o->od_mnt != NULL)
5686 if (strlen(dev) >= sizeof(o->od_mntdev))
5688 strcpy(o->od_mntdev, dev);
5690 OBD_PAGE_ALLOC(__page, GFP_IOFS);
5692 GOTO(out, rc = -ENOMEM);
5694 str = lustre_cfg_string(cfg, 2);
5695 s_flags = simple_strtoul(str, NULL, 0);
5696 str = strstr(str, ":");
5698 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5699 opts = lustre_cfg_string(cfg, 3);
5700 page = (unsigned long)page_address(__page);
5701 options = (char *)page;
5704 strcat(options, "user_xattr,acl");
5706 strcat(options, opts);
5708 /* Glom up mount options */
5709 if (*options != '\0')
5710 strcat(options, ",");
5711 strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
5713 type = get_fs_type("ldiskfs");
5715 CERROR("%s: cannot find ldiskfs module\n", name);
5716 GOTO(out, rc = -ENODEV);
5719 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5720 module_put(type->owner);
5722 if (IS_ERR(o->od_mnt)) {
5723 rc = PTR_ERR(o->od_mnt);
5725 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5729 #ifdef HAVE_DEV_SET_RDONLY
5730 if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5731 CERROR("%s: underlying device %s is marked as read-only. "
5732 "Setup failed\n", name, dev);
5733 GOTO(out_mnt, rc = -EROFS);
5737 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5738 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5739 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5740 GOTO(out_mnt, rc = -EINVAL);
5743 #ifdef LDISKFS_MOUNT_DIRDATA
5744 if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
5745 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5746 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
5747 else if (!o->od_is_ost)
5748 CWARN("%s: device %s was upgraded from Lustre-1.x without "
5749 "enabling the dirdata feature. If you do not want to "
5750 "downgrade to Lustre-1.x again, you can enable it via "
5751 "'tune2fs -O dirdata device'\n", name, dev);
5753 inode = osd_sb(o)->s_root->d_inode;
5754 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
5755 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
5757 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
5761 if (lmd_flags & LMD_FLG_NOSCRUB)
5772 OBD_PAGE_FREE(__page);
5777 static struct lu_device *osd_device_fini(const struct lu_env *env,
5778 struct lu_device *d)
5780 struct osd_device *o = osd_dev(d);
5783 osd_shutdown(env, o);
5785 osd_scrub_cleanup(env, o);
5786 osd_obj_map_fini(o);
5792 static int osd_device_init0(const struct lu_env *env,
5793 struct osd_device *o,
5794 struct lustre_cfg *cfg)
5796 struct lu_device *l = osd2lu_dev(o);
5797 struct osd_thread_info *info;
5801 /* if the module was re-loaded, env can loose its keys */
5802 rc = lu_env_refill((struct lu_env *) env);
5805 info = osd_oti_get(env);
5808 l->ld_ops = &osd_lu_ops;
5809 o->od_dt_dev.dd_ops = &osd_dt_ops;
5811 spin_lock_init(&o->od_osfs_lock);
5812 mutex_init(&o->od_otable_mutex);
5814 o->od_capa_hash = init_capa_hash();
5815 if (o->od_capa_hash == NULL)
5816 GOTO(out, rc = -ENOMEM);
5818 o->od_read_cache = 1;
5819 o->od_writethrough_cache = 1;
5820 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
5822 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
5823 sizeof(o->od_svname));
5824 if (cplen >= sizeof(o->od_svname)) {
5829 if (server_name_is_ost(o->od_svname))
5832 rc = osd_mount(env, o, cfg);
5836 rc = osd_obj_map_init(env, o);
5840 rc = lu_site_init(&o->od_site, l);
5842 GOTO(out_compat, rc);
5843 o->od_site.ls_bottom_dev = l;
5845 rc = lu_site_init_finish(&o->od_site);
5849 /* self-repair LMA by default */
5850 o->od_lma_self_repair = 1;
5852 CFS_INIT_LIST_HEAD(&o->od_ios_list);
5853 /* setup scrub, including OI files initialization */
5854 rc = osd_scrub_setup(env, o);
5858 rc = osd_procfs_init(o, o->od_svname);
5860 CERROR("%s: can't initialize procfs: rc = %d\n",
5862 GOTO(out_scrub, rc);
5865 LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
5867 /* initialize quota slave instance */
5868 o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
5870 if (IS_ERR(o->od_quota_slave)) {
5871 rc = PTR_ERR(o->od_quota_slave);
5872 o->od_quota_slave = NULL;
5873 GOTO(out_procfs, rc);
5881 osd_scrub_cleanup(env, o);
5883 lu_site_fini(&o->od_site);
5885 osd_obj_map_fini(o);
5889 cleanup_capa_hash(o->od_capa_hash);
5894 static struct lu_device *osd_device_alloc(const struct lu_env *env,
5895 struct lu_device_type *t,
5896 struct lustre_cfg *cfg)
5898 struct osd_device *o;
5903 return ERR_PTR(-ENOMEM);
5905 rc = dt_device_init(&o->od_dt_dev, t);
5907 /* Because the ctx might be revived in dt_device_init,
5908 * refill the env here */
5909 lu_env_refill((struct lu_env *)env);
5910 rc = osd_device_init0(env, o, cfg);
5912 dt_device_fini(&o->od_dt_dev);
5915 if (unlikely(rc != 0))
5918 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
5921 static struct lu_device *osd_device_free(const struct lu_env *env,
5922 struct lu_device *d)
5924 struct osd_device *o = osd_dev(d);
5927 cleanup_capa_hash(o->od_capa_hash);
5928 /* XXX: make osd top device in order to release reference */
5929 d->ld_site->ls_top_dev = d;
5930 lu_site_purge(env, d->ld_site, -1);
5931 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
5932 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
5933 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
5935 lu_site_fini(&o->od_site);
5936 dt_device_fini(&o->od_dt_dev);
5941 static int osd_process_config(const struct lu_env *env,
5942 struct lu_device *d, struct lustre_cfg *cfg)
5944 struct osd_device *o = osd_dev(d);
5948 switch (cfg->lcfg_command) {
5950 rc = osd_mount(env, o, cfg);
5953 lu_dev_del_linkage(d->ld_site, d);
5954 rc = osd_shutdown(env, o);
5957 LASSERT(&o->od_dt_dev);
5958 rc = class_process_proc_seq_param(PARAM_OSD,
5959 lprocfs_osd_obd_vars,
5960 cfg, &o->od_dt_dev);
5961 if (rc > 0 || rc == -ENOSYS)
5962 rc = class_process_proc_seq_param(PARAM_OST,
5963 lprocfs_osd_obd_vars,
5964 cfg, &o->od_dt_dev);
5973 static int osd_recovery_complete(const struct lu_env *env,
5974 struct lu_device *d)
5976 struct osd_device *osd = osd_dev(d);
5980 if (osd->od_quota_slave == NULL)
5983 /* start qsd instance on recovery completion, this notifies the quota
5984 * slave code that we are about to process new requests now */
5985 rc = qsd_start(env, osd->od_quota_slave);
5990 * we use exports to track all osd users
5992 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
5993 struct obd_device *obd, struct obd_uuid *cluuid,
5994 struct obd_connect_data *data, void *localdata)
5996 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
5997 struct lustre_handle conn;
6001 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6003 rc = class_connect(&conn, obd, cluuid);
6007 *exp = class_conn2export(&conn);
6009 spin_lock(&osd->od_osfs_lock);
6011 spin_unlock(&osd->od_osfs_lock);
6017 * once last export (we don't count self-export) disappeared
6018 * osd can be released
6020 static int osd_obd_disconnect(struct obd_export *exp)
6022 struct obd_device *obd = exp->exp_obd;
6023 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6024 int rc, release = 0;
6027 /* Only disconnect the underlying layers on the final disconnect. */
6028 spin_lock(&osd->od_osfs_lock);
6030 if (osd->od_connects == 0)
6032 spin_unlock(&osd->od_osfs_lock);
6034 rc = class_disconnect(exp); /* bz 9811 */
6036 if (rc == 0 && release)
6037 class_manual_cleanup(obd);
6041 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6042 struct lu_device *dev)
6044 struct osd_device *osd = osd_dev(dev);
6048 if (osd->od_quota_slave != NULL) {
6049 /* set up quota slave objects */
6050 result = qsd_prepare(env, osd->od_quota_slave);
6055 result = osd_fid_init(env, osd);
6060 int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6061 struct lu_fid *fid, struct md_op_data *op_data)
6063 struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6065 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6068 static const struct lu_object_operations osd_lu_obj_ops = {
6069 .loo_object_init = osd_object_init,
6070 .loo_object_delete = osd_object_delete,
6071 .loo_object_release = osd_object_release,
6072 .loo_object_free = osd_object_free,
6073 .loo_object_print = osd_object_print,
6074 .loo_object_invariant = osd_object_invariant
6077 const struct lu_device_operations osd_lu_ops = {
6078 .ldo_object_alloc = osd_object_alloc,
6079 .ldo_process_config = osd_process_config,
6080 .ldo_recovery_complete = osd_recovery_complete,
6081 .ldo_prepare = osd_prepare,
6084 static const struct lu_device_type_operations osd_device_type_ops = {
6085 .ldto_init = osd_type_init,
6086 .ldto_fini = osd_type_fini,
6088 .ldto_start = osd_type_start,
6089 .ldto_stop = osd_type_stop,
6091 .ldto_device_alloc = osd_device_alloc,
6092 .ldto_device_free = osd_device_free,
6094 .ldto_device_init = osd_device_init,
6095 .ldto_device_fini = osd_device_fini
6098 struct lu_device_type osd_device_type = {
6099 .ldt_tags = LU_DEVICE_DT,
6100 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
6101 .ldt_ops = &osd_device_type_ops,
6102 .ldt_ctx_tags = LCT_LOCAL,
6106 * lprocfs legacy support.
6108 static struct obd_ops osd_obd_device_ops = {
6109 .o_owner = THIS_MODULE,
6110 .o_connect = osd_obd_connect,
6111 .o_disconnect = osd_obd_disconnect,
6112 .o_fid_alloc = osd_fid_alloc,
6115 static int __init osd_mod_init(void)
6121 rc = lu_kmem_init(ldiskfs_caches);
6125 rc = class_register_type(&osd_obd_device_ops, NULL, true,
6126 lprocfs_osd_module_vars,
6127 #ifndef HAVE_ONLY_PROCFS_SEQ
6130 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6132 lu_kmem_fini(ldiskfs_caches);
6136 static void __exit osd_mod_exit(void)
6138 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6139 lu_kmem_fini(ldiskfs_caches);
6142 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6143 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6144 MODULE_LICENSE("GPL");
6146 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);