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);
1154 osd_fini_iobuf(oti->oti_dev, iobuf);
1156 rc = iobuf->dr_error;
1161 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1163 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1166 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1167 LASSERT(&dcb->dcb_func != NULL);
1168 cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1174 * Called just before object is freed. Releases all resources except for
1175 * object itself (that is released by osd_object_free()).
1177 * Concurrency: no concurrent access is possible that late in object
1180 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1182 struct osd_object *obj = osd_obj(l);
1183 struct inode *inode = obj->oo_inode;
1185 LINVRNT(osd_invariant(obj));
1188 * If object is unlinked remove fid->ino mapping from object index.
1191 osd_index_fini(obj);
1192 if (inode != NULL) {
1193 struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave;
1194 qid_t uid = inode->i_uid;
1195 qid_t gid = inode->i_gid;
1198 obj->oo_inode = NULL;
1201 struct osd_thread_info *info = osd_oti_get(env);
1202 struct lquota_id_info *qi = &info->oti_qi;
1204 /* Release granted quota to master if necessary */
1205 qi->lqi_id.qid_uid = uid;
1206 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1208 qi->lqi_id.qid_uid = gid;
1209 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1215 * Concurrency: ->loo_object_release() is called under site spin-lock.
1217 static void osd_object_release(const struct lu_env *env,
1218 struct lu_object *l)
1223 * Concurrency: shouldn't matter.
1225 static int osd_object_print(const struct lu_env *env, void *cookie,
1226 lu_printer_t p, const struct lu_object *l)
1228 struct osd_object *o = osd_obj(l);
1229 struct iam_descr *d;
1231 if (o->oo_dir != NULL)
1232 d = o->oo_dir->od_container.ic_descr;
1235 return (*p)(env, cookie,
1236 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1238 o->oo_inode ? o->oo_inode->i_ino : 0UL,
1239 o->oo_inode ? o->oo_inode->i_generation : 0,
1240 d ? d->id_ops->id_name : "plain");
1243 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1246 * Concurrency: shouldn't matter.
1248 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1249 struct obd_statfs *sfs)
1251 struct osd_device *osd = osd_dt_dev(d);
1252 struct super_block *sb = osd_sb(osd);
1253 struct kstatfs *ksfs;
1256 if (unlikely(osd->od_mnt == NULL))
1257 return -EINPROGRESS;
1259 /* osd_lproc.c call this without env, allocate ksfs for that case */
1260 if (unlikely(env == NULL)) {
1261 OBD_ALLOC_PTR(ksfs);
1265 ksfs = &osd_oti_get(env)->oti_ksfs;
1268 spin_lock(&osd->od_osfs_lock);
1269 result = sb->s_op->statfs(sb->s_root, ksfs);
1270 if (likely(result == 0)) { /* N.B. statfs can't really fail */
1271 statfs_pack(sfs, ksfs);
1272 if (sb->s_flags & MS_RDONLY)
1273 sfs->os_state = OS_STATE_READONLY;
1276 spin_unlock(&osd->od_osfs_lock);
1278 if (unlikely(env == NULL))
1281 /* Reserve a small amount of space for local objects like last_rcvd,
1282 * llog, quota files, ... */
1283 if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1286 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1287 /** Take out metadata overhead for indirect blocks */
1288 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1295 * Estimate space needed for file creations. We assume the largest filename
1296 * which is 2^64 - 1, hence a filename of 20 chars.
1297 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1299 #ifdef __LDISKFS_DIR_REC_LEN
1300 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1302 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1306 * Concurrency: doesn't access mutable data.
1308 static void osd_conf_get(const struct lu_env *env,
1309 const struct dt_device *dev,
1310 struct dt_device_param *param)
1312 struct super_block *sb = osd_sb(osd_dt_dev(dev));
1316 * XXX should be taken from not-yet-existing fs abstraction layer.
1318 param->ddp_max_name_len = LDISKFS_NAME_LEN;
1319 param->ddp_max_nlink = LDISKFS_LINK_MAX;
1320 param->ddp_block_shift = sb->s_blocksize_bits;
1321 param->ddp_mount_type = LDD_MT_LDISKFS;
1322 param->ddp_maxbytes = sb->s_maxbytes;
1323 /* Overhead estimate should be fairly accurate, so we really take a tiny
1324 * error margin which also avoids fragmenting the filesystem too much */
1325 param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1326 /* inode are statically allocated, so per-inode space consumption
1327 * is the space consumed by the directory entry */
1328 param->ddp_inodespace = PER_OBJ_USAGE;
1329 /* per-fragment overhead to be used by the client code */
1330 param->ddp_grant_frag = 6 * LDISKFS_BLOCK_SIZE(sb);
1331 param->ddp_mntopts = 0;
1332 if (test_opt(sb, XATTR_USER))
1333 param->ddp_mntopts |= MNTOPT_USERXATTR;
1334 if (test_opt(sb, POSIX_ACL))
1335 param->ddp_mntopts |= MNTOPT_ACL;
1337 /* LOD might calculate the max stripe count based on max_ea_size,
1338 * so we need take account in the overhead as well,
1339 * xattr_header + magic + xattr_entry_head */
1340 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1341 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1343 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1344 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1345 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1349 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1353 * Concurrency: shouldn't matter.
1355 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1357 CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1358 return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1362 * Start commit for OSD device.
1364 * An implementation of dt_commit_async method for OSD device.
1365 * Asychronously starts underlayng fs sync and thereby a transaction
1368 * \param env environment
1369 * \param d dt device
1371 * \see dt_device_operations
1373 static int osd_commit_async(const struct lu_env *env,
1374 struct dt_device *d)
1376 struct super_block *s = osd_sb(osd_dt_dev(d));
1379 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1380 RETURN(s->s_op->sync_fs(s, 0));
1384 * Concurrency: shouldn't matter.
1387 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1389 struct super_block *sb = osd_sb(osd_dt_dev(d));
1390 struct block_device *dev = sb->s_bdev;
1391 #ifdef HAVE_DEV_SET_RDONLY
1392 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1395 int rc = -EOPNOTSUPP;
1399 #ifdef HAVE_DEV_SET_RDONLY
1400 CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1402 if (jdev && (jdev != dev)) {
1403 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1405 dev_set_rdonly(jdev);
1407 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1408 dev_set_rdonly(dev);
1410 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1411 osd_dt_dev(d)->od_svname, (long)dev, rc);
1417 * Concurrency: serialization provided by callers.
1419 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1420 int mode, unsigned long timeout, __u32 alg,
1421 struct lustre_capa_key *keys)
1423 struct osd_device *dev = osd_dt_dev(d);
1426 dev->od_fl_capa = mode;
1427 dev->od_capa_timeout = timeout;
1428 dev->od_capa_alg = alg;
1429 dev->od_capa_keys = keys;
1434 * Note: we do not count into QUOTA here.
1435 * If we mount with --data_journal we may need more.
1437 const int osd_dto_credits_noquota[DTO_NR] = {
1440 * INDEX_EXTRA_TRANS_BLOCKS(8) +
1441 * SINGLEDATA_TRANS_BLOCKS(8)
1442 * XXX Note: maybe iam need more, since iam have more level than
1445 [DTO_INDEX_INSERT] = 16,
1448 * just modify a single entry, probably merge few within a block
1450 [DTO_INDEX_DELETE] = 1,
1454 [DTO_INDEX_UPDATE] = 16,
1456 * 4(inode, inode bits, groups, GDT)
1457 * notice: OI updates are counted separately with DTO_INDEX_INSERT
1459 [DTO_OBJECT_CREATE] = 4,
1461 * 4(inode, inode bits, groups, GDT)
1462 * notice: OI updates are counted separately with DTO_INDEX_DELETE
1464 [DTO_OBJECT_DELETE] = 4,
1466 * Attr set credits (inode)
1468 [DTO_ATTR_SET_BASE] = 1,
1470 * Xattr set. The same as xattr of EXT3.
1471 * DATA_TRANS_BLOCKS(14)
1472 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1473 * are also counted in. Do not know why?
1475 [DTO_XATTR_SET] = 14,
1477 * credits for inode change during write.
1479 [DTO_WRITE_BASE] = 3,
1481 * credits for single block write.
1483 [DTO_WRITE_BLOCK] = 14,
1485 * Attr set credits for chown.
1486 * This is extra credits for setattr, and it is null without quota
1488 [DTO_ATTR_SET_CHOWN] = 0
1491 static const struct dt_device_operations osd_dt_ops = {
1492 .dt_root_get = osd_root_get,
1493 .dt_statfs = osd_statfs,
1494 .dt_trans_create = osd_trans_create,
1495 .dt_trans_start = osd_trans_start,
1496 .dt_trans_stop = osd_trans_stop,
1497 .dt_trans_cb_add = osd_trans_cb_add,
1498 .dt_conf_get = osd_conf_get,
1499 .dt_sync = osd_sync,
1501 .dt_commit_async = osd_commit_async,
1502 .dt_init_capa_ctxt = osd_init_capa_ctxt,
1505 static void osd_object_read_lock(const struct lu_env *env,
1506 struct dt_object *dt, unsigned role)
1508 struct osd_object *obj = osd_dt_obj(dt);
1509 struct osd_thread_info *oti = osd_oti_get(env);
1511 LINVRNT(osd_invariant(obj));
1513 LASSERT(obj->oo_owner != env);
1514 down_read_nested(&obj->oo_sem, role);
1516 LASSERT(obj->oo_owner == NULL);
1520 static void osd_object_write_lock(const struct lu_env *env,
1521 struct dt_object *dt, unsigned role)
1523 struct osd_object *obj = osd_dt_obj(dt);
1524 struct osd_thread_info *oti = osd_oti_get(env);
1526 LINVRNT(osd_invariant(obj));
1528 LASSERT(obj->oo_owner != env);
1529 down_write_nested(&obj->oo_sem, role);
1531 LASSERT(obj->oo_owner == NULL);
1532 obj->oo_owner = env;
1536 static void osd_object_read_unlock(const struct lu_env *env,
1537 struct dt_object *dt)
1539 struct osd_object *obj = osd_dt_obj(dt);
1540 struct osd_thread_info *oti = osd_oti_get(env);
1542 LINVRNT(osd_invariant(obj));
1544 LASSERT(oti->oti_r_locks > 0);
1546 up_read(&obj->oo_sem);
1549 static void osd_object_write_unlock(const struct lu_env *env,
1550 struct dt_object *dt)
1552 struct osd_object *obj = osd_dt_obj(dt);
1553 struct osd_thread_info *oti = osd_oti_get(env);
1555 LINVRNT(osd_invariant(obj));
1557 LASSERT(obj->oo_owner == env);
1558 LASSERT(oti->oti_w_locks > 0);
1560 obj->oo_owner = NULL;
1561 up_write(&obj->oo_sem);
1564 static int osd_object_write_locked(const struct lu_env *env,
1565 struct dt_object *dt)
1567 struct osd_object *obj = osd_dt_obj(dt);
1569 LINVRNT(osd_invariant(obj));
1571 return obj->oo_owner == env;
1574 static int capa_is_sane(const struct lu_env *env,
1575 struct osd_device *dev,
1576 struct lustre_capa *capa,
1577 struct lustre_capa_key *keys)
1579 struct osd_thread_info *oti = osd_oti_get(env);
1580 struct lustre_capa *tcapa = &oti->oti_capa;
1581 struct obd_capa *oc;
1585 oc = capa_lookup(dev->od_capa_hash, capa, 0);
1587 if (capa_is_expired(oc)) {
1588 DEBUG_CAPA(D_ERROR, capa, "expired");
1595 if (capa_is_expired_sec(capa)) {
1596 DEBUG_CAPA(D_ERROR, capa, "expired");
1600 spin_lock(&capa_lock);
1601 for (i = 0; i < 2; i++) {
1602 if (keys[i].lk_keyid == capa->lc_keyid) {
1603 oti->oti_capa_key = keys[i];
1607 spin_unlock(&capa_lock);
1610 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1614 rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1618 if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1619 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1623 oc = capa_add(dev->od_capa_hash, capa);
1629 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1630 struct lustre_capa *capa, __u64 opc)
1632 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1633 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
1634 struct lu_capainfo *lci;
1637 if (!osd->od_fl_capa)
1640 if (capa == BYPASS_CAPA)
1643 lci = lu_capainfo_get(env);
1644 if (unlikely(lci == NULL))
1647 if (lci->lci_auth == LC_ID_NONE)
1651 CERROR("%s: no capability provided for FID "DFID": rc = %d\n",
1652 osd_name(osd), PFID(fid), -EACCES);
1656 if (!lu_fid_eq(fid, &capa->lc_fid)) {
1657 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1662 if (!capa_opc_supported(capa, opc)) {
1663 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1667 rc = capa_is_sane(env, osd, capa, osd->od_capa_keys);
1669 DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc);
1676 static struct timespec *osd_inode_time(const struct lu_env *env,
1677 struct inode *inode, __u64 seconds)
1679 struct osd_thread_info *oti = osd_oti_get(env);
1680 struct timespec *t = &oti->oti_time;
1682 t->tv_sec = seconds;
1684 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1689 static void osd_inode_getattr(const struct lu_env *env,
1690 struct inode *inode, struct lu_attr *attr)
1692 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1693 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1694 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1697 attr->la_atime = LTIME_S(inode->i_atime);
1698 attr->la_mtime = LTIME_S(inode->i_mtime);
1699 attr->la_ctime = LTIME_S(inode->i_ctime);
1700 attr->la_mode = inode->i_mode;
1701 attr->la_size = i_size_read(inode);
1702 attr->la_blocks = inode->i_blocks;
1703 attr->la_uid = inode->i_uid;
1704 attr->la_gid = inode->i_gid;
1705 attr->la_flags = LDISKFS_I(inode)->i_flags;
1706 attr->la_nlink = inode->i_nlink;
1707 attr->la_rdev = inode->i_rdev;
1708 attr->la_blksize = 1 << inode->i_blkbits;
1709 attr->la_blkbits = inode->i_blkbits;
1712 static int osd_attr_get(const struct lu_env *env,
1713 struct dt_object *dt,
1714 struct lu_attr *attr,
1715 struct lustre_capa *capa)
1717 struct osd_object *obj = osd_dt_obj(dt);
1719 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1720 LINVRNT(osd_invariant(obj));
1722 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1725 spin_lock(&obj->oo_guard);
1726 osd_inode_getattr(env, obj->oo_inode, attr);
1727 spin_unlock(&obj->oo_guard);
1731 static int osd_declare_attr_set(const struct lu_env *env,
1732 struct dt_object *dt,
1733 const struct lu_attr *attr,
1734 struct thandle *handle)
1736 struct osd_thandle *oh;
1737 struct osd_object *obj;
1738 struct osd_thread_info *info = osd_oti_get(env);
1739 struct lquota_id_info *qi = &info->oti_qi;
1745 LASSERT(dt != NULL);
1746 LASSERT(handle != NULL);
1748 obj = osd_dt_obj(dt);
1749 LASSERT(osd_invariant(obj));
1751 oh = container_of0(handle, struct osd_thandle, ot_super);
1752 LASSERT(oh->ot_handle == NULL);
1754 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1755 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1757 if (attr == NULL || obj->oo_inode == NULL)
1760 bspace = obj->oo_inode->i_blocks;
1761 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1762 bspace = toqb(bspace);
1764 /* Changing ownership is always preformed by super user, it should not
1767 * We still need to call the osd_declare_qid() to calculate the journal
1768 * credits for updating quota accounting files and to trigger quota
1769 * space adjustment once the operation is completed.*/
1770 if ((attr->la_valid & LA_UID) != 0 &&
1771 attr->la_uid != obj->oo_inode->i_uid) {
1772 qi->lqi_type = USRQUOTA;
1774 /* inode accounting */
1775 qi->lqi_is_blk = false;
1777 /* one more inode for the new owner ... */
1778 qi->lqi_id.qid_uid = attr->la_uid;
1780 allocated = (attr->la_uid == 0) ? true : false;
1781 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1782 if (rc == -EDQUOT || rc == -EINPROGRESS)
1787 /* and one less inode for the current uid */
1788 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1790 rc = osd_declare_qid(env, oh, qi, true, NULL);
1791 if (rc == -EDQUOT || rc == -EINPROGRESS)
1796 /* block accounting */
1797 qi->lqi_is_blk = true;
1799 /* more blocks for the new owner ... */
1800 qi->lqi_id.qid_uid = attr->la_uid;
1801 qi->lqi_space = bspace;
1802 allocated = (attr->la_uid == 0) ? true : false;
1803 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1804 if (rc == -EDQUOT || rc == -EINPROGRESS)
1809 /* and finally less blocks for the current owner */
1810 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1811 qi->lqi_space = -bspace;
1812 rc = osd_declare_qid(env, oh, qi, true, NULL);
1813 if (rc == -EDQUOT || rc == -EINPROGRESS)
1819 if (attr->la_valid & LA_GID &&
1820 attr->la_gid != obj->oo_inode->i_gid) {
1821 qi->lqi_type = GRPQUOTA;
1823 /* inode accounting */
1824 qi->lqi_is_blk = false;
1826 /* one more inode for the new group owner ... */
1827 qi->lqi_id.qid_gid = attr->la_gid;
1829 allocated = (attr->la_gid == 0) ? true : false;
1830 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1831 if (rc == -EDQUOT || rc == -EINPROGRESS)
1836 /* and one less inode for the current gid */
1837 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1839 rc = osd_declare_qid(env, oh, qi, true, NULL);
1840 if (rc == -EDQUOT || rc == -EINPROGRESS)
1845 /* block accounting */
1846 qi->lqi_is_blk = true;
1848 /* more blocks for the new owner ... */
1849 qi->lqi_id.qid_gid = attr->la_gid;
1850 qi->lqi_space = bspace;
1851 allocated = (attr->la_gid == 0) ? true : false;
1852 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1853 if (rc == -EDQUOT || rc == -EINPROGRESS)
1858 /* and finally less blocks for the current owner */
1859 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1860 qi->lqi_space = -bspace;
1861 rc = osd_declare_qid(env, oh, qi, true, NULL);
1862 if (rc == -EDQUOT || rc == -EINPROGRESS)
1871 static int osd_inode_setattr(const struct lu_env *env,
1872 struct inode *inode, const struct lu_attr *attr)
1876 bits = attr->la_valid;
1878 if (bits & LA_ATIME)
1879 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
1880 if (bits & LA_CTIME)
1881 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
1882 if (bits & LA_MTIME)
1883 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
1884 if (bits & LA_SIZE) {
1885 LDISKFS_I(inode)->i_disksize = attr->la_size;
1886 i_size_write(inode, attr->la_size);
1890 /* OSD should not change "i_blocks" which is used by quota.
1891 * "i_blocks" should be changed by ldiskfs only. */
1892 if (bits & LA_BLOCKS)
1893 inode->i_blocks = attr->la_blocks;
1896 inode->i_mode = (inode->i_mode & S_IFMT) |
1897 (attr->la_mode & ~S_IFMT);
1899 inode->i_uid = attr->la_uid;
1901 inode->i_gid = attr->la_gid;
1902 if (bits & LA_NLINK)
1903 set_nlink(inode, attr->la_nlink);
1905 inode->i_rdev = attr->la_rdev;
1907 if (bits & LA_FLAGS) {
1908 /* always keep S_NOCMTIME */
1909 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1915 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1917 if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1918 (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1923 if (attr->la_valid & LA_UID)
1924 iattr.ia_valid |= ATTR_UID;
1925 if (attr->la_valid & LA_GID)
1926 iattr.ia_valid |= ATTR_GID;
1927 iattr.ia_uid = attr->la_uid;
1928 iattr.ia_gid = attr->la_gid;
1930 rc = ll_vfs_dq_transfer(inode, &iattr);
1932 CERROR("%s: quota transfer failed: rc = %d. Is quota "
1933 "enforcement enabled on the ldiskfs filesystem?",
1934 inode->i_sb->s_id, rc);
1941 static int osd_attr_set(const struct lu_env *env,
1942 struct dt_object *dt,
1943 const struct lu_attr *attr,
1944 struct thandle *handle,
1945 struct lustre_capa *capa)
1947 struct osd_object *obj = osd_dt_obj(dt);
1948 struct inode *inode;
1951 LASSERT(handle != NULL);
1952 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1953 LASSERT(osd_invariant(obj));
1955 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1958 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1960 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1961 struct osd_thread_info *oti = osd_oti_get(env);
1962 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
1963 struct lu_fid *fid1 = &oti->oti_fid;
1964 struct osd_inode_id *id = &oti->oti_id;
1965 struct iam_path_descr *ipd;
1966 struct iam_container *bag;
1967 struct osd_thandle *oh;
1970 fid_cpu_to_be(fid1, fid0);
1971 memset(id, 1, sizeof(*id));
1972 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1973 fid0)->oi_dir.od_container;
1974 ipd = osd_idx_ipd_get(env, bag);
1975 if (unlikely(ipd == NULL))
1978 oh = container_of0(handle, struct osd_thandle, ot_super);
1979 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1980 (const struct iam_rec *)id, ipd);
1981 osd_ipd_put(env, bag, ipd);
1982 return(rc > 0 ? 0 : rc);
1985 inode = obj->oo_inode;
1986 ll_vfs_dq_init(inode);
1988 rc = osd_quota_transfer(inode, attr);
1992 spin_lock(&obj->oo_guard);
1993 rc = osd_inode_setattr(env, inode, attr);
1994 spin_unlock(&obj->oo_guard);
1997 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2001 struct dentry *osd_child_dentry_get(const struct lu_env *env,
2002 struct osd_object *obj,
2003 const char *name, const int namelen)
2005 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2008 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2009 umode_t mode, struct dt_allocation_hint *hint,
2013 struct osd_device *osd = osd_obj2dev(obj);
2014 struct osd_thandle *oth;
2015 struct dt_object *parent = NULL;
2016 struct inode *inode;
2018 LINVRNT(osd_invariant(obj));
2019 LASSERT(obj->oo_inode == NULL);
2020 LASSERT(obj->oo_hl_head == NULL);
2022 if (S_ISDIR(mode) && ldiskfs_pdo) {
2023 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2024 if (obj->oo_hl_head == NULL)
2028 oth = container_of(th, struct osd_thandle, ot_super);
2029 LASSERT(oth->ot_handle->h_transaction != NULL);
2031 if (hint && hint->dah_parent)
2032 parent = hint->dah_parent;
2034 inode = ldiskfs_create_inode(oth->ot_handle,
2035 parent ? osd_dt_obj(parent)->oo_inode :
2036 osd_sb(osd)->s_root->d_inode,
2038 if (!IS_ERR(inode)) {
2039 /* Do not update file c/mtime in ldiskfs.
2040 * NB: don't need any lock because no contention at this
2042 inode->i_flags |= S_NOCMTIME;
2044 /* For new created object, it must be consistent,
2045 * and it is unnecessary to scrub against it. */
2046 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2047 obj->oo_inode = inode;
2050 if (obj->oo_hl_head != NULL) {
2051 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2052 obj->oo_hl_head = NULL;
2054 result = PTR_ERR(inode);
2056 LINVRNT(osd_invariant(obj));
2064 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2065 struct lu_attr *attr,
2066 struct dt_allocation_hint *hint,
2067 struct dt_object_format *dof,
2071 struct osd_thandle *oth;
2072 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2074 LASSERT(S_ISDIR(attr->la_mode));
2076 oth = container_of(th, struct osd_thandle, ot_super);
2077 LASSERT(oth->ot_handle->h_transaction != NULL);
2078 result = osd_mkfile(info, obj, mode, hint, th);
2083 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2084 struct lu_attr *attr,
2085 struct dt_allocation_hint *hint,
2086 struct dt_object_format *dof,
2090 struct osd_thandle *oth;
2091 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2093 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2095 LASSERT(S_ISREG(attr->la_mode));
2097 oth = container_of(th, struct osd_thandle, ot_super);
2098 LASSERT(oth->ot_handle->h_transaction != NULL);
2100 result = osd_mkfile(info, obj, mode, hint, th);
2102 LASSERT(obj->oo_inode != NULL);
2103 if (feat->dif_flags & DT_IND_VARKEY)
2104 result = iam_lvar_create(obj->oo_inode,
2105 feat->dif_keysize_max,
2107 feat->dif_recsize_max,
2110 result = iam_lfix_create(obj->oo_inode,
2111 feat->dif_keysize_max,
2113 feat->dif_recsize_max,
2120 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2121 struct lu_attr *attr,
2122 struct dt_allocation_hint *hint,
2123 struct dt_object_format *dof,
2126 LASSERT(S_ISREG(attr->la_mode));
2127 return osd_mkfile(info, obj, (attr->la_mode &
2128 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2131 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2132 struct lu_attr *attr,
2133 struct dt_allocation_hint *hint,
2134 struct dt_object_format *dof,
2137 LASSERT(S_ISLNK(attr->la_mode));
2138 return osd_mkfile(info, obj, (attr->la_mode &
2139 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2142 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2143 struct lu_attr *attr,
2144 struct dt_allocation_hint *hint,
2145 struct dt_object_format *dof,
2148 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2151 LINVRNT(osd_invariant(obj));
2152 LASSERT(obj->oo_inode == NULL);
2153 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2154 S_ISFIFO(mode) || S_ISSOCK(mode));
2156 result = osd_mkfile(info, obj, mode, hint, th);
2158 LASSERT(obj->oo_inode != NULL);
2160 * This inode should be marked dirty for i_rdev. Currently
2161 * that is done in the osd_attr_init().
2163 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2166 LINVRNT(osd_invariant(obj));
2170 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2172 struct dt_allocation_hint *hint,
2173 struct dt_object_format *dof,
2176 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2178 osd_obj_type_f result;
2194 result = osd_mk_index;
2205 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2206 struct dt_object *parent, struct dt_object *child,
2211 ah->dah_parent = parent;
2212 ah->dah_mode = child_mode;
2215 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2216 struct lu_attr *attr, struct dt_object_format *dof)
2218 struct inode *inode = obj->oo_inode;
2219 __u64 valid = attr->la_valid;
2222 attr->la_valid &= ~(LA_TYPE | LA_MODE);
2224 if (dof->dof_type != DFT_NODE)
2225 attr->la_valid &= ~LA_RDEV;
2226 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2227 attr->la_valid &= ~LA_ATIME;
2228 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2229 attr->la_valid &= ~LA_CTIME;
2230 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2231 attr->la_valid &= ~LA_MTIME;
2233 result = osd_quota_transfer(inode, attr);
2237 if (attr->la_valid != 0) {
2238 result = osd_inode_setattr(info->oti_env, inode, attr);
2240 * The osd_inode_setattr() should always succeed here. The
2241 * only error that could be returned is EDQUOT when we are
2242 * trying to change the UID or GID of the inode. However, this
2243 * should not happen since quota enforcement is no longer
2244 * enabled on ldiskfs (lquota takes care of it).
2246 LASSERTF(result == 0, "%d", result);
2247 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2250 attr->la_valid = valid;
2254 * Helper function for osd_object_create()
2256 * \retval 0, on success
2258 static int __osd_object_create(struct osd_thread_info *info,
2259 struct osd_object *obj, struct lu_attr *attr,
2260 struct dt_allocation_hint *hint,
2261 struct dt_object_format *dof,
2267 /* we drop umask so that permissions we pass are not affected */
2268 umask = current->fs->umask;
2269 current->fs->umask = 0;
2271 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2274 osd_attr_init(info, obj, attr, dof);
2275 osd_object_init0(obj);
2277 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2278 unlock_new_inode(obj->oo_inode);
2281 /* restore previous umask value */
2282 current->fs->umask = umask;
2288 * Helper function for osd_object_create()
2290 * \retval 0, on success
2292 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2293 const struct lu_fid *fid, struct thandle *th)
2295 struct osd_thread_info *info = osd_oti_get(env);
2296 struct osd_inode_id *id = &info->oti_id;
2297 struct osd_device *osd = osd_obj2dev(obj);
2298 struct osd_thandle *oh;
2300 LASSERT(obj->oo_inode != NULL);
2302 oh = container_of0(th, struct osd_thandle, ot_super);
2303 LASSERT(oh->ot_handle);
2305 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2306 return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2309 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2310 obd_seq seq, struct lu_seq_range *range)
2312 struct seq_server_site *ss = osd_seq_site(osd);
2314 if (fid_seq_is_idif(seq)) {
2315 fld_range_set_ost(range);
2316 range->lsr_index = idif_ost_idx(seq);
2320 if (!fid_seq_in_fldb(seq)) {
2321 fld_range_set_mdt(range);
2323 /* FIXME: If ss is NULL, it suppose not get lsr_index
2325 range->lsr_index = ss->ss_node_id;
2329 LASSERT(ss != NULL);
2330 fld_range_set_any(range);
2331 /* OSD will only do local fld lookup */
2332 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2336 * Concurrency: no external locking is necessary.
2338 static int osd_declare_object_create(const struct lu_env *env,
2339 struct dt_object *dt,
2340 struct lu_attr *attr,
2341 struct dt_allocation_hint *hint,
2342 struct dt_object_format *dof,
2343 struct thandle *handle)
2345 struct osd_thandle *oh;
2349 LASSERT(handle != NULL);
2351 oh = container_of0(handle, struct osd_thandle, ot_super);
2352 LASSERT(oh->ot_handle == NULL);
2354 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2355 osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2356 /* Reuse idle OI block may cause additional one OI block
2358 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2359 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2361 /* If this is directory, then we expect . and .. to be inserted as
2362 * well. The one directory block always needs to be created for the
2363 * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2364 * block), there is no danger of needing a tree for the first block.
2366 if (attr && S_ISDIR(attr->la_mode)) {
2367 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2368 osd_dto_credits_noquota[DTO_WRITE_BASE]);
2369 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2375 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2376 false, false, NULL, false);
2383 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2384 struct lu_attr *attr,
2385 struct dt_allocation_hint *hint,
2386 struct dt_object_format *dof,
2389 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2390 struct osd_object *obj = osd_dt_obj(dt);
2391 struct osd_thread_info *info = osd_oti_get(env);
2396 LINVRNT(osd_invariant(obj));
2397 LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2398 LASSERT(osd_write_locked(env, obj));
2399 LASSERT(th != NULL);
2401 if (unlikely(fid_is_acct(fid)))
2402 /* Quota files can't be created from the kernel any more,
2403 * 'tune2fs -O quota' will take care of creating them */
2406 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2407 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2409 result = __osd_object_create(info, obj, attr, hint, dof, th);
2411 result = __osd_oi_insert(env, obj, fid, th);
2413 LASSERT(ergo(result == 0,
2414 dt_object_exists(dt) && !dt_object_remote(dt)));
2416 LASSERT(osd_invariant(obj));
2421 * Called to destroy on-disk representation of the object
2423 * Concurrency: must be locked
2425 static int osd_declare_object_destroy(const struct lu_env *env,
2426 struct dt_object *dt,
2429 struct osd_object *obj = osd_dt_obj(dt);
2430 struct inode *inode = obj->oo_inode;
2431 struct osd_thandle *oh;
2435 oh = container_of0(th, struct osd_thandle, ot_super);
2436 LASSERT(oh->ot_handle == NULL);
2439 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2440 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2441 /* Recycle idle OI leaf may cause additional three OI blocks
2443 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2444 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2445 /* one less inode */
2446 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
2447 false, true, NULL, false);
2450 /* data to be truncated */
2451 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
2452 true, true, NULL, false);
2456 static int osd_object_destroy(const struct lu_env *env,
2457 struct dt_object *dt,
2460 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2461 struct osd_object *obj = osd_dt_obj(dt);
2462 struct inode *inode = obj->oo_inode;
2463 struct osd_device *osd = osd_obj2dev(obj);
2464 struct osd_thandle *oh;
2468 oh = container_of0(th, struct osd_thandle, ot_super);
2469 LASSERT(oh->ot_handle);
2471 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2473 if (unlikely(fid_is_acct(fid)))
2476 if (S_ISDIR(inode->i_mode)) {
2477 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2478 inode->i_nlink == 2);
2479 /* it will check/delete the inode from remote parent,
2480 * how to optimize it? unlink performance impaction XXX */
2481 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2482 if (result != 0 && result != -ENOENT) {
2483 CERROR("%s: delete inode "DFID": rc = %d\n",
2484 osd_name(osd), PFID(fid), result);
2486 spin_lock(&obj->oo_guard);
2488 spin_unlock(&obj->oo_guard);
2489 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2492 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2494 result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2497 /* XXX: add to ext3 orphan list */
2498 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2500 /* not needed in the cache anymore */
2501 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2507 * Put the fid into lustre_mdt_attrs, and then place the structure
2508 * inode's ea. This fid should not be altered during the life time
2511 * \retval +ve, on success
2512 * \retval -ve, on error
2514 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2516 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2517 const struct lu_fid *fid, __u32 compat, __u32 incompat)
2519 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2523 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2526 lustre_lma_init(lma, fid, compat, incompat);
2527 lustre_lma_swab(lma);
2529 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2531 /* LMA may already exist, but we need to check that all the
2532 * desired compat/incompat flags have been added. */
2533 if (unlikely(rc == -EEXIST)) {
2534 if (compat == 0 && incompat == 0)
2537 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2538 XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2543 lustre_lma_swab(lma);
2544 if (!(~lma->lma_compat & compat) &&
2545 !(~lma->lma_incompat & incompat))
2548 lma->lma_compat |= compat;
2549 lma->lma_incompat |= incompat;
2550 lustre_lma_swab(lma);
2551 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2552 sizeof(*lma), XATTR_REPLACE);
2559 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2560 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2561 * To have compatilibility with 1.8 ldiskfs driver we need to have
2562 * magic number at start of fid data.
2563 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2566 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2567 const struct dt_rec *fid)
2569 if (!fid_is_namespace_visible((const struct lu_fid *)fid) ||
2570 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2571 param->edp_magic = 0;
2575 param->edp_magic = LDISKFS_LUFID_MAGIC;
2576 param->edp_len = sizeof(struct lu_fid) + 1;
2577 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2581 * Try to read the fid from inode ea into dt_rec.
2583 * \param fid object fid.
2585 * \retval 0 on success
2587 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2588 __u32 ino, struct lu_fid *fid,
2589 struct osd_inode_id *id)
2591 struct osd_thread_info *info = osd_oti_get(env);
2592 struct inode *inode;
2595 osd_id_gen(id, ino, OSD_OII_NOGEN);
2596 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2598 RETURN(PTR_ERR(inode));
2604 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2606 struct inode *parent_dir,
2607 const struct dt_rec *dot_fid,
2608 const struct dt_rec *dot_dot_fid,
2609 struct osd_thandle *oth)
2611 struct ldiskfs_dentry_param *dot_ldp;
2612 struct ldiskfs_dentry_param *dot_dot_ldp;
2614 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2615 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2617 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2618 dot_ldp->edp_magic = 0;
2619 return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2620 dir, dot_ldp, dot_dot_ldp);
2624 * Create an local agent inode for remote entry
2626 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2627 struct osd_device *osd,
2628 struct osd_object *pobj,
2629 const struct lu_fid *fid,
2632 struct osd_thread_info *info = osd_oti_get(env);
2633 struct inode *local;
2634 struct osd_thandle *oh;
2639 oh = container_of(th, struct osd_thandle, ot_super);
2640 LASSERT(oh->ot_handle->h_transaction != NULL);
2642 /* FIXME: Insert index api needs to know the mode of
2643 * the remote object. Just use S_IFDIR for now */
2644 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR);
2645 if (IS_ERR(local)) {
2646 CERROR("%s: create local error %d\n", osd_name(osd),
2647 (int)PTR_ERR(local));
2651 /* Set special LMA flag for local agent inode */
2652 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2654 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2655 osd_name(osd), PFID(fid), rc);
2656 RETURN(ERR_PTR(rc));
2659 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2660 (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
2661 (const struct dt_rec *)fid, oh);
2663 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2664 osd_name(osd), PFID(fid), rc);
2665 RETURN(ERR_PTR(rc));
2672 * Delete local agent inode for remote entry
2674 static int osd_delete_local_agent_inode(const struct lu_env *env,
2675 struct osd_device *osd,
2676 const struct lu_fid *fid,
2677 __u32 ino, struct osd_thandle *oh)
2679 struct osd_thread_info *oti = osd_oti_get(env);
2680 struct osd_inode_id *id = &oti->oti_id;
2681 struct inode *inode;
2684 id->oii_ino = le32_to_cpu(ino);
2685 id->oii_gen = OSD_OII_NOGEN;
2686 inode = osd_iget(oti, osd, id);
2687 if (IS_ERR(inode)) {
2688 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2689 PFID(fid), id->oii_ino, id->oii_gen);
2690 RETURN(PTR_ERR(inode));
2694 mark_inode_dirty(inode);
2695 CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2696 osd_name(osd), PFID(fid), inode->i_ino);
2702 * OSD layer object create function for interoperability mode (b11826).
2703 * This is mostly similar to osd_object_create(). Only difference being, fid is
2704 * inserted into inode ea here.
2706 * \retval 0, on success
2707 * \retval -ve, on error
2709 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2710 struct lu_attr *attr,
2711 struct dt_allocation_hint *hint,
2712 struct dt_object_format *dof,
2715 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2716 struct osd_object *obj = osd_dt_obj(dt);
2717 struct osd_thread_info *info = osd_oti_get(env);
2722 LASSERT(osd_invariant(obj));
2723 LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2724 LASSERT(osd_write_locked(env, obj));
2725 LASSERT(th != NULL);
2727 if (unlikely(fid_is_acct(fid)))
2728 /* Quota files can't be created from the kernel any more,
2729 * 'tune2fs -O quota' will take care of creating them */
2732 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2733 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2735 result = __osd_object_create(info, obj, attr, hint, dof, th);
2737 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2738 fid_is_on_ost(info, osd_obj2dev(obj),
2739 fid, OI_CHECK_FLD) ?
2740 LMAC_FID_ON_OST : 0, 0);
2743 result = __osd_oi_insert(env, obj, fid, th);
2745 LASSERT(ergo(result == 0,
2746 dt_object_exists(dt) && !dt_object_remote(dt)));
2747 LINVRNT(osd_invariant(obj));
2751 static int osd_declare_object_ref_add(const struct lu_env *env,
2752 struct dt_object *dt,
2753 struct thandle *handle)
2755 struct osd_thandle *oh;
2757 /* it's possible that object doesn't exist yet */
2758 LASSERT(handle != NULL);
2760 oh = container_of0(handle, struct osd_thandle, ot_super);
2761 LASSERT(oh->ot_handle == NULL);
2763 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2764 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2770 * Concurrency: @dt is write locked.
2772 static int osd_object_ref_add(const struct lu_env *env,
2773 struct dt_object *dt, struct thandle *th)
2775 struct osd_object *obj = osd_dt_obj(dt);
2776 struct inode *inode = obj->oo_inode;
2777 struct osd_thandle *oh;
2780 LINVRNT(osd_invariant(obj));
2781 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2782 LASSERT(osd_write_locked(env, obj));
2783 LASSERT(th != NULL);
2785 oh = container_of0(th, struct osd_thandle, ot_super);
2786 LASSERT(oh->ot_handle != NULL);
2788 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2790 CDEBUG(D_INODE, DFID" increase nlink %d\n",
2791 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2793 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2794 * (65000) subdirectories by storing "1" in i_nlink if the link count
2795 * would otherwise overflow. Directory tranversal tools understand
2796 * that (st_nlink == 1) indicates that the filesystem dose not track
2797 * hard links count on the directory, and will not abort subdirectory
2798 * scanning early once (st_nlink - 2) subdirs have been found.
2800 * This also has to properly handle the case of inodes with nlink == 0
2801 * in case they are being linked into the PENDING directory
2803 spin_lock(&obj->oo_guard);
2804 ldiskfs_inc_count(oh->ot_handle, inode);
2805 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2806 spin_unlock(&obj->oo_guard);
2808 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2809 LINVRNT(osd_invariant(obj));
2814 static int osd_declare_object_ref_del(const struct lu_env *env,
2815 struct dt_object *dt,
2816 struct thandle *handle)
2818 struct osd_thandle *oh;
2820 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2821 LASSERT(handle != NULL);
2823 oh = container_of0(handle, struct osd_thandle, ot_super);
2824 LASSERT(oh->ot_handle == NULL);
2826 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2827 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2833 * Concurrency: @dt is write locked.
2835 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2838 struct osd_object *obj = osd_dt_obj(dt);
2839 struct inode *inode = obj->oo_inode;
2840 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
2841 struct osd_thandle *oh;
2843 LINVRNT(osd_invariant(obj));
2844 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2845 LASSERT(osd_write_locked(env, obj));
2846 LASSERT(th != NULL);
2848 oh = container_of0(th, struct osd_thandle, ot_super);
2849 LASSERT(oh->ot_handle != NULL);
2851 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2853 spin_lock(&obj->oo_guard);
2854 /* That can be result of upgrade from old Lustre version and
2855 * applied only to local files. Just skip this ref_del call.
2856 * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2857 if (inode->i_nlink == 0) {
2858 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2859 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2860 ", maybe an upgraded file? (LU-3915)\n",
2861 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2862 spin_unlock(&obj->oo_guard);
2866 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2867 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2869 ldiskfs_dec_count(oh->ot_handle, inode);
2870 spin_unlock(&obj->oo_guard);
2872 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2873 LINVRNT(osd_invariant(obj));
2879 * Get the 64-bit version for an inode.
2881 static int osd_object_version_get(const struct lu_env *env,
2882 struct dt_object *dt, dt_obj_version_t *ver)
2884 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2886 CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2887 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2888 *ver = LDISKFS_I(inode)->i_fs_version;
2893 * Concurrency: @dt is read locked.
2895 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2896 struct lu_buf *buf, const char *name,
2897 struct lustre_capa *capa)
2899 struct osd_object *obj = osd_dt_obj(dt);
2900 struct inode *inode = obj->oo_inode;
2901 struct osd_thread_info *info = osd_oti_get(env);
2902 struct dentry *dentry = &info->oti_obj_dentry;
2904 /* version get is not real XATTR but uses xattr API */
2905 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2906 /* for version we are just using xattr API but change inode
2908 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2909 osd_object_version_get(env, dt, buf->lb_buf);
2910 return sizeof(dt_obj_version_t);
2913 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2914 LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2916 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2919 return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2923 static int osd_declare_xattr_set(const struct lu_env *env,
2924 struct dt_object *dt,
2925 const struct lu_buf *buf, const char *name,
2926 int fl, struct thandle *handle)
2928 struct osd_thandle *oh;
2931 LASSERT(handle != NULL);
2933 oh = container_of0(handle, struct osd_thandle, ot_super);
2934 LASSERT(oh->ot_handle == NULL);
2936 /* optimistic optimization: LMA is set first and usually fit inode */
2937 if (strcmp(name, XATTR_NAME_LMA) == 0) {
2938 if (dt_object_exists(dt))
2942 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2945 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
2946 struct super_block *sb = osd_sb(osd);
2947 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2948 if (buf && buf->lb_len > sb->s_blocksize) {
2949 credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2950 sb->s_blocksize_bits;
2954 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2960 * Set the 64-bit version for object
2962 static void osd_object_version_set(const struct lu_env *env,
2963 struct dt_object *dt,
2964 dt_obj_version_t *new_version)
2966 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2968 CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2969 *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2971 LDISKFS_I(inode)->i_fs_version = *new_version;
2972 /** Version is set after all inode operations are finished,
2973 * so we should mark it dirty here */
2974 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2978 * Concurrency: @dt is write locked.
2980 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2981 const struct lu_buf *buf, const char *name, int fl,
2982 struct thandle *handle, struct lustre_capa *capa)
2984 struct osd_object *obj = osd_dt_obj(dt);
2985 struct inode *inode = obj->oo_inode;
2986 struct osd_thread_info *info = osd_oti_get(env);
2990 LASSERT(handle != NULL);
2992 /* version set is not real XATTR */
2993 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2994 /* for version we are just using xattr API but change inode
2996 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2997 osd_object_version_set(env, dt, buf->lb_buf);
2998 return sizeof(dt_obj_version_t);
3001 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3004 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n",
3005 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3007 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3008 if (fl & LU_XATTR_REPLACE)
3009 fs_flags |= XATTR_REPLACE;
3011 if (fl & LU_XATTR_CREATE)
3012 fs_flags |= XATTR_CREATE;
3014 return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3019 * Concurrency: @dt is read locked.
3021 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3022 struct lu_buf *buf, struct lustre_capa *capa)
3024 struct osd_object *obj = osd_dt_obj(dt);
3025 struct inode *inode = obj->oo_inode;
3026 struct osd_thread_info *info = osd_oti_get(env);
3027 struct dentry *dentry = &info->oti_obj_dentry;
3029 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3030 LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
3032 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3035 dentry->d_inode = inode;
3036 dentry->d_sb = inode->i_sb;
3037 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3040 static int osd_declare_xattr_del(const struct lu_env *env,
3041 struct dt_object *dt, const char *name,
3042 struct thandle *handle)
3044 struct osd_thandle *oh;
3046 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3047 LASSERT(handle != NULL);
3049 oh = container_of0(handle, struct osd_thandle, ot_super);
3050 LASSERT(oh->ot_handle == NULL);
3052 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3053 osd_dto_credits_noquota[DTO_XATTR_SET]);
3059 * Concurrency: @dt is write locked.
3061 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3062 const char *name, struct thandle *handle,
3063 struct lustre_capa *capa)
3065 struct osd_object *obj = osd_dt_obj(dt);
3066 struct inode *inode = obj->oo_inode;
3067 struct osd_thread_info *info = osd_oti_get(env);
3068 struct dentry *dentry = &info->oti_obj_dentry;
3071 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3072 LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
3073 LASSERT(handle != NULL);
3075 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3078 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3080 ll_vfs_dq_init(inode);
3081 dentry->d_inode = inode;
3082 dentry->d_sb = inode->i_sb;
3083 rc = inode->i_op->removexattr(dentry, name);
3087 static struct obd_capa *osd_capa_get(const struct lu_env *env,
3088 struct dt_object *dt,
3089 struct lustre_capa *old, __u64 opc)
3091 struct osd_thread_info *info = osd_oti_get(env);
3092 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3093 struct osd_object *obj = osd_dt_obj(dt);
3094 struct osd_device *osd = osd_obj2dev(obj);
3095 struct lustre_capa_key *key = &info->oti_capa_key;
3096 struct lustre_capa *capa = &info->oti_capa;
3097 struct obd_capa *oc;
3098 struct lu_capainfo *lci;
3102 if (!osd->od_fl_capa)
3103 RETURN(ERR_PTR(-ENOENT));
3105 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3106 LINVRNT(osd_invariant(obj));
3108 /* renewal sanity check */
3109 if (old && osd_object_auth(env, dt, old, opc))
3110 RETURN(ERR_PTR(-EACCES));
3112 lci = lu_capainfo_get(env);
3113 if (unlikely(lci == NULL))
3114 RETURN(ERR_PTR(-ENOENT));
3116 switch (lci->lci_auth) {
3120 capa->lc_uid = obj->oo_inode->i_uid;
3121 capa->lc_gid = obj->oo_inode->i_gid;
3122 capa->lc_flags = LC_ID_PLAIN;
3124 case LC_ID_CONVERT: {
3127 s[0] = obj->oo_inode->i_uid;
3128 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
3129 s[2] = obj->oo_inode->i_gid;
3130 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
3131 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
3133 RETURN(ERR_PTR(rc));
3135 capa->lc_uid = ((__u64)d[1] << 32) | d[0];
3136 capa->lc_gid = ((__u64)d[3] << 32) | d[2];
3137 capa->lc_flags = LC_ID_CONVERT;
3141 RETURN(ERR_PTR(-EINVAL));
3144 capa->lc_fid = *fid;
3146 capa->lc_flags |= osd->od_capa_alg << 24;
3147 capa->lc_timeout = osd->od_capa_timeout;
3148 capa->lc_expiry = 0;
3150 oc = capa_lookup(osd->od_capa_hash, capa, 1);
3152 LASSERT(!capa_is_expired(oc));
3156 spin_lock(&capa_lock);
3157 *key = osd->od_capa_keys[1];
3158 spin_unlock(&capa_lock);
3160 capa->lc_keyid = key->lk_keyid;
3161 capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout;
3163 rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
3165 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
3166 RETURN(ERR_PTR(rc));
3169 oc = capa_add(osd->od_capa_hash, capa);
3173 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3174 __u64 start, __u64 end)
3176 struct osd_object *obj = osd_dt_obj(dt);
3177 struct inode *inode = obj->oo_inode;
3178 struct osd_thread_info *info = osd_oti_get(env);
3179 struct dentry *dentry = &info->oti_obj_dentry;
3180 struct file *file = &info->oti_file;
3185 dentry->d_inode = inode;
3186 dentry->d_sb = inode->i_sb;
3187 file->f_dentry = dentry;
3188 file->f_mapping = inode->i_mapping;
3189 file->f_op = inode->i_fop;
3190 set_file_inode(file, inode);
3192 #ifdef HAVE_FILE_FSYNC_4ARGS
3193 rc = file->f_op->fsync(file, start, end, 0);
3194 #elif defined(HAVE_FILE_FSYNC_2ARGS)
3195 mutex_lock(&inode->i_mutex);
3196 rc = file->f_op->fsync(file, 0);
3197 mutex_unlock(&inode->i_mutex);
3199 mutex_lock(&inode->i_mutex);
3200 rc = file->f_op->fsync(file, dentry, 0);
3201 mutex_unlock(&inode->i_mutex);
3207 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
3210 struct osd_object *obj = osd_dt_obj(dt);
3213 *data = (void *)obj->oo_inode;
3221 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3222 const struct dt_index_features *feat)
3224 struct iam_descr *descr;
3226 if (osd_object_is_root(o))
3227 return feat == &dt_directory_features;
3229 LASSERT(o->oo_dir != NULL);
3231 descr = o->oo_dir->od_container.ic_descr;
3232 if (feat == &dt_directory_features) {
3233 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3239 feat->dif_keysize_min <= descr->id_key_size &&
3240 descr->id_key_size <= feat->dif_keysize_max &&
3241 feat->dif_recsize_min <= descr->id_rec_size &&
3242 descr->id_rec_size <= feat->dif_recsize_max &&
3243 !(feat->dif_flags & (DT_IND_VARKEY |
3244 DT_IND_VARREC | DT_IND_NONUNQ)) &&
3245 ergo(feat->dif_flags & DT_IND_UPDATE,
3246 1 /* XXX check that object (and file system) is
3251 static int osd_iam_container_init(const struct lu_env *env,
3252 struct osd_object *obj,
3253 struct osd_directory *dir)
3255 struct iam_container *bag = &dir->od_container;
3258 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3262 result = iam_container_setup(bag);
3264 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3266 iam_container_fini(bag);
3273 * Concurrency: no external locking is necessary.
3275 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3276 const struct dt_index_features *feat)
3280 struct osd_object *obj = osd_dt_obj(dt);
3282 LINVRNT(osd_invariant(obj));
3284 if (osd_object_is_root(obj)) {
3285 dt->do_index_ops = &osd_index_ea_ops;
3287 } else if (feat == &dt_directory_features) {
3288 dt->do_index_ops = &osd_index_ea_ops;
3289 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3294 } else if (unlikely(feat == &dt_otable_features)) {
3295 dt->do_index_ops = &osd_otable_ops;
3297 } else if (unlikely(feat == &dt_acct_features)) {
3298 dt->do_index_ops = &osd_acct_index_ops;
3301 } else if (!osd_has_index(obj)) {
3302 struct osd_directory *dir;
3307 spin_lock(&obj->oo_guard);
3308 if (obj->oo_dir == NULL)
3312 * Concurrent thread allocated container data.
3315 spin_unlock(&obj->oo_guard);
3317 * Now, that we have container data, serialize its
3320 down_write(&obj->oo_ext_idx_sem);
3322 * recheck under lock.
3324 if (!osd_has_index(obj))
3325 result = osd_iam_container_init(env, obj, dir);
3328 up_write(&obj->oo_ext_idx_sem);
3336 if (result == 0 && skip_iam == 0) {
3337 if (!osd_iam_index_probe(env, obj, feat))
3340 LINVRNT(osd_invariant(obj));
3342 if (result == 0 && is_quota_glb_feat(feat) &&
3343 fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3344 result = osd_quota_migration(env, dt, feat);
3349 static int osd_otable_it_attr_get(const struct lu_env *env,
3350 struct dt_object *dt,
3351 struct lu_attr *attr,
3352 struct lustre_capa *capa)
3358 static const struct dt_object_operations osd_obj_ops = {
3359 .do_read_lock = osd_object_read_lock,
3360 .do_write_lock = osd_object_write_lock,
3361 .do_read_unlock = osd_object_read_unlock,
3362 .do_write_unlock = osd_object_write_unlock,
3363 .do_write_locked = osd_object_write_locked,
3364 .do_attr_get = osd_attr_get,
3365 .do_declare_attr_set = osd_declare_attr_set,
3366 .do_attr_set = osd_attr_set,
3367 .do_ah_init = osd_ah_init,
3368 .do_declare_create = osd_declare_object_create,
3369 .do_create = osd_object_create,
3370 .do_declare_destroy = osd_declare_object_destroy,
3371 .do_destroy = osd_object_destroy,
3372 .do_index_try = osd_index_try,
3373 .do_declare_ref_add = osd_declare_object_ref_add,
3374 .do_ref_add = osd_object_ref_add,
3375 .do_declare_ref_del = osd_declare_object_ref_del,
3376 .do_ref_del = osd_object_ref_del,
3377 .do_xattr_get = osd_xattr_get,
3378 .do_declare_xattr_set = osd_declare_xattr_set,
3379 .do_xattr_set = osd_xattr_set,
3380 .do_declare_xattr_del = osd_declare_xattr_del,
3381 .do_xattr_del = osd_xattr_del,
3382 .do_xattr_list = osd_xattr_list,
3383 .do_capa_get = osd_capa_get,
3384 .do_object_sync = osd_object_sync,
3385 .do_data_get = osd_data_get,
3389 * dt_object_operations for interoperability mode
3390 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3392 static const struct dt_object_operations osd_obj_ea_ops = {
3393 .do_read_lock = osd_object_read_lock,
3394 .do_write_lock = osd_object_write_lock,
3395 .do_read_unlock = osd_object_read_unlock,
3396 .do_write_unlock = osd_object_write_unlock,
3397 .do_write_locked = osd_object_write_locked,
3398 .do_attr_get = osd_attr_get,
3399 .do_declare_attr_set = osd_declare_attr_set,
3400 .do_attr_set = osd_attr_set,
3401 .do_ah_init = osd_ah_init,
3402 .do_declare_create = osd_declare_object_create,
3403 .do_create = osd_object_ea_create,
3404 .do_declare_destroy = osd_declare_object_destroy,
3405 .do_destroy = osd_object_destroy,
3406 .do_index_try = osd_index_try,
3407 .do_declare_ref_add = osd_declare_object_ref_add,
3408 .do_ref_add = osd_object_ref_add,
3409 .do_declare_ref_del = osd_declare_object_ref_del,
3410 .do_ref_del = osd_object_ref_del,
3411 .do_xattr_get = osd_xattr_get,
3412 .do_declare_xattr_set = osd_declare_xattr_set,
3413 .do_xattr_set = osd_xattr_set,
3414 .do_declare_xattr_del = osd_declare_xattr_del,
3415 .do_xattr_del = osd_xattr_del,
3416 .do_xattr_list = osd_xattr_list,
3417 .do_capa_get = osd_capa_get,
3418 .do_object_sync = osd_object_sync,
3419 .do_data_get = osd_data_get,
3422 static const struct dt_object_operations osd_obj_otable_it_ops = {
3423 .do_attr_get = osd_otable_it_attr_get,
3424 .do_index_try = osd_index_try,
3427 static int osd_index_declare_iam_delete(const struct lu_env *env,
3428 struct dt_object *dt,
3429 const struct dt_key *key,
3430 struct thandle *handle)
3432 struct osd_thandle *oh;
3434 oh = container_of0(handle, struct osd_thandle, ot_super);
3435 LASSERT(oh->ot_handle == NULL);
3437 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3438 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3444 * delete a (key, value) pair from index \a dt specified by \a key
3446 * \param dt osd index object
3447 * \param key key for index
3448 * \param rec record reference
3449 * \param handle transaction handler
3452 * \retval -ve failure
3455 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3456 const struct dt_key *key,
3457 struct thandle *handle,
3458 struct lustre_capa *capa)
3460 struct osd_thread_info *oti = osd_oti_get(env);
3461 struct osd_object *obj = osd_dt_obj(dt);
3462 struct osd_thandle *oh;
3463 struct iam_path_descr *ipd;
3464 struct iam_container *bag = &obj->oo_dir->od_container;
3469 LINVRNT(osd_invariant(obj));
3470 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3471 LASSERT(bag->ic_object == obj->oo_inode);
3472 LASSERT(handle != NULL);
3474 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3477 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3479 ipd = osd_idx_ipd_get(env, bag);
3480 if (unlikely(ipd == NULL))
3483 oh = container_of0(handle, struct osd_thandle, ot_super);
3484 LASSERT(oh->ot_handle != NULL);
3485 LASSERT(oh->ot_handle->h_transaction != NULL);
3487 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3488 /* swab quota uid/gid provided by caller */
3489 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3490 key = (const struct dt_key *)&oti->oti_quota_id;
3493 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3494 osd_ipd_put(env, bag, ipd);
3495 LINVRNT(osd_invariant(obj));
3499 static int osd_index_declare_ea_delete(const struct lu_env *env,
3500 struct dt_object *dt,
3501 const struct dt_key *key,
3502 struct thandle *handle)
3504 struct osd_thandle *oh;
3505 struct inode *inode;
3509 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3510 LASSERT(handle != NULL);
3512 oh = container_of0(handle, struct osd_thandle, ot_super);
3513 LASSERT(oh->ot_handle == NULL);
3515 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3516 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3518 inode = osd_dt_obj(dt)->oo_inode;
3521 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3522 true, true, NULL, false);
3526 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3529 struct osd_fid_pack *rec;
3532 if (de->file_type & LDISKFS_DIRENT_LUFID) {
3533 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3534 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3539 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3544 /* FID seqs not in FLDB, must be local seq */
3545 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3548 if (osd_seq_exists(env, osd, fid_seq(fid)))
3555 * Index delete function for interoperability mode (b11826).
3556 * It will remove the directory entry added by osd_index_ea_insert().
3557 * This entry is needed to maintain name->fid mapping.
3559 * \param key, key i.e. file entry to be deleted
3561 * \retval 0, on success
3562 * \retval -ve, on error
3564 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3565 const struct dt_key *key,
3566 struct thandle *handle,
3567 struct lustre_capa *capa)
3569 struct osd_object *obj = osd_dt_obj(dt);
3570 struct inode *dir = obj->oo_inode;
3571 struct dentry *dentry;
3572 struct osd_thandle *oh;
3573 struct ldiskfs_dir_entry_2 *de = NULL;
3574 struct buffer_head *bh;
3575 struct htree_lock *hlock = NULL;
3576 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
3577 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3581 LINVRNT(osd_invariant(obj));
3582 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3583 LASSERT(handle != NULL);
3585 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3587 oh = container_of(handle, struct osd_thandle, ot_super);
3588 LASSERT(oh->ot_handle != NULL);
3589 LASSERT(oh->ot_handle->h_transaction != NULL);
3591 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3594 ll_vfs_dq_init(dir);
3595 dentry = osd_child_dentry_get(env, obj,
3596 (char *)key, strlen((char *)key));
3598 if (obj->oo_hl_head != NULL) {
3599 hlock = osd_oti_get(env)->oti_hlock;
3600 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3601 dir, LDISKFS_HLOCK_DEL);
3603 down_write(&obj->oo_ext_idx_sem);
3606 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3610 /* If this is not the ".." entry, it might be a remote DNE
3611 * entry and we need to check if the FID is for a remote
3612 * MDT. If the FID is not in the directory entry (e.g.
3613 * upgraded 1.8 filesystem without dirdata enabled) then
3614 * we need to get the FID from the LMA. For a remote directory
3615 * there HAS to be an LMA, it cannot be an IGIF inode in this
3618 * Delete the entry before the agent inode in order to
3619 * simplify error handling. At worst an error after deleting
3620 * the entry first might leak the agent inode afterward. The
3621 * reverse would need filesystem abort in case of error deleting
3622 * the entry after the agent had been removed, or leave a
3623 * dangling entry pointing at a random inode. */
3624 if (strcmp((char *)key, dotdot) != 0) {
3625 LASSERT(de != NULL);
3626 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3627 /* If Fid is not in dentry, try to get it from LMA */
3628 if (rc == -ENODATA) {
3629 struct osd_inode_id *id;
3630 struct inode *inode;
3632 /* Before trying to get fid from the inode,
3633 * check whether the inode is valid.
3635 * If the inode has been deleted, do not go
3636 * ahead to do osd_ea_fid_get, which will set
3637 * the inode to bad inode, which might cause
3638 * the inode to be deleted uncorrectly */
3639 inode = ldiskfs_iget(osd_sb(osd),
3640 le32_to_cpu(de->inode));
3641 if (IS_ERR(inode)) {
3642 CDEBUG(D_INODE, "%s: "DFID"get inode"
3643 "error.\n", osd_name(osd),
3645 rc = PTR_ERR(inode);
3647 if (likely(inode->i_nlink != 0)) {
3648 id = &osd_oti_get(env)->oti_id;
3649 rc = osd_ea_fid_get(env, obj,
3650 le32_to_cpu(de->inode),
3653 CDEBUG(D_INFO, "%s: %u "DFID
3656 le32_to_cpu(de->inode),
3664 unlikely(osd_remote_fid(env, osd, fid)))
3665 /* Need to delete agent inode */
3666 ino = le32_to_cpu(de->inode);
3668 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3670 if (rc == 0 && unlikely(ino != 0)) {
3671 rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3674 CERROR("%s: del local inode "DFID": rc = %d\n",
3675 osd_name(osd), PFID(fid), rc);
3681 ldiskfs_htree_unlock(hlock);
3683 up_write(&obj->oo_ext_idx_sem);
3688 /* For inode on the remote MDT, .. will point to
3689 * /Agent directory, Check whether it needs to delete
3690 * from agent directory */
3691 if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3692 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3694 if (rc != 0 && rc != -ENOENT) {
3695 CERROR("%s: delete agent inode "DFID": rc = %d\n",
3696 osd_name(osd), PFID(fid), rc);
3706 LASSERT(osd_invariant(obj));
3711 * Lookup index for \a key and copy record to \a rec.
3713 * \param dt osd index object
3714 * \param key key for index
3715 * \param rec record reference
3717 * \retval +ve success : exact mach
3718 * \retval 0 return record with key not greater than \a key
3719 * \retval -ve failure
3721 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3722 struct dt_rec *rec, const struct dt_key *key,
3723 struct lustre_capa *capa)
3725 struct osd_object *obj = osd_dt_obj(dt);
3726 struct iam_path_descr *ipd;
3727 struct iam_container *bag = &obj->oo_dir->od_container;
3728 struct osd_thread_info *oti = osd_oti_get(env);
3729 struct iam_iterator *it = &oti->oti_idx_it;
3730 struct iam_rec *iam_rec;
3735 LASSERT(osd_invariant(obj));
3736 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3737 LASSERT(bag->ic_object == obj->oo_inode);
3739 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3742 ipd = osd_idx_ipd_get(env, bag);
3746 /* got ipd now we can start iterator. */
3747 iam_it_init(it, bag, 0, ipd);
3749 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3750 /* swab quota uid/gid provided by caller */
3751 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3752 key = (const struct dt_key *)&oti->oti_quota_id;
3755 rc = iam_it_get(it, (struct iam_key *)key);
3757 if (S_ISDIR(obj->oo_inode->i_mode))
3758 iam_rec = (struct iam_rec *)oti->oti_ldp;
3760 iam_rec = (struct iam_rec *) rec;
3762 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3764 if (S_ISDIR(obj->oo_inode->i_mode))
3765 osd_fid_unpack((struct lu_fid *) rec,
3766 (struct osd_fid_pack *)iam_rec);
3767 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3768 osd_quota_unpack(obj, rec);
3773 osd_ipd_put(env, bag, ipd);
3775 LINVRNT(osd_invariant(obj));
3780 static int osd_index_declare_iam_insert(const struct lu_env *env,
3781 struct dt_object *dt,
3782 const struct dt_rec *rec,
3783 const struct dt_key *key,
3784 struct thandle *handle)
3786 struct osd_thandle *oh;
3788 LASSERT(handle != NULL);
3790 oh = container_of0(handle, struct osd_thandle, ot_super);
3791 LASSERT(oh->ot_handle == NULL);
3793 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3794 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3800 * Inserts (key, value) pair in \a dt index object.
3802 * \param dt osd index object
3803 * \param key key for index
3804 * \param rec record reference
3805 * \param th transaction handler
3808 * \retval -ve failure
3810 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3811 const struct dt_rec *rec,
3812 const struct dt_key *key, struct thandle *th,
3813 struct lustre_capa *capa, int ignore_quota)
3815 struct osd_object *obj = osd_dt_obj(dt);
3816 struct iam_path_descr *ipd;
3817 struct osd_thandle *oh;
3818 struct iam_container *bag = &obj->oo_dir->od_container;
3819 struct osd_thread_info *oti = osd_oti_get(env);
3820 struct iam_rec *iam_rec;
3825 LINVRNT(osd_invariant(obj));
3826 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3827 LASSERT(bag->ic_object == obj->oo_inode);
3828 LASSERT(th != NULL);
3830 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3833 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3835 ipd = osd_idx_ipd_get(env, bag);
3836 if (unlikely(ipd == NULL))
3839 oh = container_of0(th, struct osd_thandle, ot_super);
3840 LASSERT(oh->ot_handle != NULL);
3841 LASSERT(oh->ot_handle->h_transaction != NULL);
3842 if (S_ISDIR(obj->oo_inode->i_mode)) {
3843 iam_rec = (struct iam_rec *)oti->oti_ldp;
3844 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3845 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3846 /* pack quota uid/gid */
3847 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3848 key = (const struct dt_key *)&oti->oti_quota_id;
3849 /* pack quota record */
3850 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3851 iam_rec = (struct iam_rec *)rec;
3853 iam_rec = (struct iam_rec *)rec;
3856 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3858 osd_ipd_put(env, bag, ipd);
3859 LINVRNT(osd_invariant(obj));
3864 * Calls ldiskfs_add_entry() to add directory entry
3865 * into the directory. This is required for
3866 * interoperability mode (b11826)
3868 * \retval 0, on success
3869 * \retval -ve, on error
3871 static int __osd_ea_add_rec(struct osd_thread_info *info,
3872 struct osd_object *pobj, struct inode *cinode,
3873 const char *name, const struct dt_rec *fid,
3874 struct htree_lock *hlock, struct thandle *th)
3876 struct ldiskfs_dentry_param *ldp;
3877 struct dentry *child;
3878 struct osd_thandle *oth;
3881 oth = container_of(th, struct osd_thandle, ot_super);
3882 LASSERT(oth->ot_handle != NULL);
3883 LASSERT(oth->ot_handle->h_transaction != NULL);
3884 LASSERT(pobj->oo_inode);
3886 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3887 if (unlikely(pobj->oo_inode ==
3888 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3891 osd_get_ldiskfs_dirent_param(ldp, fid);
3892 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3893 child->d_fsdata = (void *)ldp;
3894 ll_vfs_dq_init(pobj->oo_inode);
3895 rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3901 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3902 * into the directory.Also sets flags into osd object to
3903 * indicate dot and dotdot are created. This is required for
3904 * interoperability mode (b11826)
3906 * \param dir directory for dot and dotdot fixup.
3907 * \param obj child object for linking
3909 * \retval 0, on success
3910 * \retval -ve, on error
3912 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3913 struct osd_object *dir,
3914 struct inode *parent_dir, const char *name,
3915 const struct dt_rec *dot_fid,
3916 const struct dt_rec *dot_dot_fid,
3919 struct inode *inode = dir->oo_inode;
3920 struct osd_thandle *oth;
3923 oth = container_of(th, struct osd_thandle, ot_super);
3924 LASSERT(oth->ot_handle->h_transaction != NULL);
3925 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3927 if (strcmp(name, dot) == 0) {
3928 if (dir->oo_compat_dot_created) {
3931 LASSERT(inode == parent_dir);
3932 dir->oo_compat_dot_created = 1;
3935 } else if (strcmp(name, dotdot) == 0) {
3936 if (!dir->oo_compat_dot_created)
3938 /* in case of rename, dotdot is already created */
3939 if (dir->oo_compat_dotdot_created) {
3940 return __osd_ea_add_rec(info, dir, parent_dir, name,
3941 dot_dot_fid, NULL, th);
3944 result = osd_add_dot_dotdot_internal(info, dir->oo_inode,
3945 parent_dir, dot_fid,
3948 dir->oo_compat_dotdot_created = 1;
3956 * It will call the appropriate osd_add* function and return the
3957 * value, return by respective functions.
3959 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3960 struct inode *cinode, const char *name,
3961 const struct dt_rec *fid, struct thandle *th)
3963 struct osd_thread_info *info = osd_oti_get(env);
3964 struct htree_lock *hlock;
3967 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3969 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3971 if (hlock != NULL) {
3972 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3975 down_write(&pobj->oo_ext_idx_sem);
3977 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3978 (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3981 if (hlock != NULL) {
3982 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3983 pobj->oo_inode, LDISKFS_HLOCK_ADD);
3985 down_write(&pobj->oo_ext_idx_sem);
3988 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
3989 struct lu_fid *tfid = &info->oti_fid;
3991 *tfid = *(const struct lu_fid *)fid;
3993 rc = __osd_ea_add_rec(info, pobj, cinode, name,
3994 (const struct dt_rec *)tfid,
3997 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4002 ldiskfs_htree_unlock(hlock);
4004 up_write(&pobj->oo_ext_idx_sem);
4010 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4011 struct osd_idmap_cache *oic)
4013 struct osd_scrub *scrub = &dev->od_scrub;
4014 struct lu_fid *fid = &oic->oic_fid;
4015 struct osd_inode_id *id = &oti->oti_id;
4020 if (!fid_is_norm(fid) && !fid_is_igif(fid))
4024 rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
4025 if (rc != 0 && rc != -ENOENT)
4028 if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
4031 if (thread_is_running(&scrub->os_thread)) {
4032 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
4033 /* There is race condition between osd_oi_lookup and OI scrub.
4034 * The OI scrub finished just after osd_oi_lookup() failure.
4035 * Under such case, it is unnecessary to trigger OI scrub again,
4036 * but try to call osd_oi_lookup() again. */
4037 if (unlikely(rc == -EAGAIN))
4043 if (!dev->od_noscrub && ++once == 1) {
4044 rc = osd_scrub_start(dev);
4045 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC for "DFID
4047 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4056 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4057 struct osd_device *dev,
4058 struct osd_idmap_cache *oic,
4059 struct lu_fid *fid, __u32 ino)
4061 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
4062 struct inode *inode;
4065 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4066 inode = osd_iget(oti, dev, &oic->oic_lid);
4067 if (IS_ERR(inode)) {
4068 fid_zero(&oic->oic_fid);
4069 return PTR_ERR(inode);
4072 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4075 fid_zero(&oic->oic_fid);
4077 *fid = oic->oic_fid = lma->lma_self_fid;
4081 int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4082 struct osd_inode_id *id, const struct lu_fid *fid)
4084 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4085 id->oii_ino, id->oii_gen, info);
4086 info->oti_cache.oic_lid = *id;
4087 info->oti_cache.oic_fid = *fid;
4088 info->oti_cache.oic_dev = osd;
4094 * Calls ->lookup() to find dentry. From dentry get inode and
4095 * read inode's ea to get fid. This is required for interoperability
4098 * \retval 0, on success
4099 * \retval -ve, on error
4101 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4102 struct dt_rec *rec, const struct dt_key *key)
4104 struct inode *dir = obj->oo_inode;
4105 struct dentry *dentry;
4106 struct ldiskfs_dir_entry_2 *de;
4107 struct buffer_head *bh;
4108 struct lu_fid *fid = (struct lu_fid *) rec;
4109 struct htree_lock *hlock = NULL;
4114 LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
4116 dentry = osd_child_dentry_get(env, obj,
4117 (char *)key, strlen((char *)key));
4119 if (obj->oo_hl_head != NULL) {
4120 hlock = osd_oti_get(env)->oti_hlock;
4121 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4122 dir, LDISKFS_HLOCK_LOOKUP);
4124 down_read(&obj->oo_ext_idx_sem);
4127 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4129 struct osd_thread_info *oti = osd_oti_get(env);
4130 struct osd_inode_id *id = &oti->oti_id;
4131 struct osd_idmap_cache *oic = &oti->oti_cache;
4132 struct osd_device *dev = osd_obj2dev(obj);
4133 struct osd_scrub *scrub = &dev->od_scrub;
4134 struct scrub_file *sf = &scrub->os_file;
4136 ino = le32_to_cpu(de->inode);
4137 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4139 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4143 rc = osd_get_fid_from_dentry(de, rec);
4145 /* done with de, release bh */
4148 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4150 osd_id_gen(id, ino, OSD_OII_NOGEN);
4152 fid_zero(&oic->oic_fid);
4156 if (osd_remote_fid(env, dev, fid))
4159 rc = osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id,
4163 if ((scrub->os_pos_current <= ino) &&
4164 ((sf->sf_flags & SF_INCONSISTENT) ||
4165 (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
4166 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
4168 osd_consistency_check(oti, dev, oic);
4177 ldiskfs_htree_unlock(hlock);
4179 up_read(&obj->oo_ext_idx_sem);
4184 * Find the osd object for given fid.
4186 * \param fid need to find the osd object having this fid
4188 * \retval osd_object on success
4189 * \retval -ve on error
4191 struct osd_object *osd_object_find(const struct lu_env *env,
4192 struct dt_object *dt,
4193 const struct lu_fid *fid)
4195 struct lu_device *ludev = dt->do_lu.lo_dev;
4196 struct osd_object *child = NULL;
4197 struct lu_object *luch;
4198 struct lu_object *lo;
4201 * at this point topdev might not exist yet
4202 * (i.e. MGS is preparing profiles). so we can
4203 * not rely on topdev and instead lookup with
4204 * our device passed as topdev. this can't work
4205 * if the object isn't cached yet (as osd doesn't
4206 * allocate lu_header). IOW, the object must be
4207 * in the cache, otherwise lu_object_alloc() crashes
4210 luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4211 ludev : ludev->ld_site->ls_top_dev,
4213 if (!IS_ERR(luch)) {
4214 if (lu_object_exists(luch)) {
4215 lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4217 child = osd_obj(lo);
4219 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4220 "lu_object can't be located"
4221 DFID"\n", PFID(fid));
4223 if (child == NULL) {
4224 lu_object_put(env, luch);
4225 CERROR("Unable to get osd_object\n");
4226 child = ERR_PTR(-ENOENT);
4229 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4230 "lu_object does not exists "DFID"\n",
4232 lu_object_put(env, luch);
4233 child = ERR_PTR(-ENOENT);
4236 child = ERR_CAST(luch);
4243 * Put the osd object once done with it.
4245 * \param obj osd object that needs to be put
4247 static inline void osd_object_put(const struct lu_env *env,
4248 struct osd_object *obj)
4250 lu_object_put(env, &obj->oo_dt.do_lu);
4253 static int osd_index_declare_ea_insert(const struct lu_env *env,
4254 struct dt_object *dt,
4255 const struct dt_rec *rec,
4256 const struct dt_key *key,
4257 struct thandle *handle)
4259 struct osd_thandle *oh;
4260 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4261 struct lu_fid *fid = (struct lu_fid *)rec;
4265 LASSERT(!dt_object_remote(dt));
4266 LASSERT(handle != NULL);
4268 oh = container_of0(handle, struct osd_thandle, ot_super);
4269 LASSERT(oh->ot_handle == NULL);
4271 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4272 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4274 if (osd_dt_obj(dt)->oo_inode != NULL) {
4275 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4277 /* We ignore block quota on meta pool (MDTs), so needn't
4278 * calculate how many blocks will be consumed by this index
4280 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0,
4281 oh, true, true, NULL, false);
4287 rc = osd_remote_fid(env, osd, fid);
4293 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
4294 osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
4295 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4296 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4297 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4298 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4304 * Index add function for interoperability mode (b11826).
4305 * It will add the directory entry.This entry is needed to
4306 * maintain name->fid mapping.
4308 * \param key it is key i.e. file entry to be inserted
4309 * \param rec it is value of given key i.e. fid
4311 * \retval 0, on success
4312 * \retval -ve, on error
4314 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4315 const struct dt_rec *rec,
4316 const struct dt_key *key, struct thandle *th,
4317 struct lustre_capa *capa, int ignore_quota)
4319 struct osd_object *obj = osd_dt_obj(dt);
4320 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4321 struct lu_fid *fid = (struct lu_fid *) rec;
4322 const char *name = (const char *)key;
4323 struct osd_thread_info *oti = osd_oti_get(env);
4324 struct osd_inode_id *id = &oti->oti_id;
4325 struct inode *child_inode = NULL;
4326 struct osd_object *child = NULL;
4330 LASSERT(osd_invariant(obj));
4331 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
4332 LASSERT(th != NULL);
4334 osd_trans_exec_op(env, th, OSD_OT_INSERT);
4336 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
4339 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!", PFID(fid));
4341 rc = osd_remote_fid(env, osd, fid);
4343 CERROR("%s: Can not find object "DFID" rc %d\n",
4344 osd_name(osd), PFID(fid), rc);
4349 /* Insert remote entry */
4350 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4351 struct osd_mdobj_map *omm = osd->od_mdt_map;
4352 struct osd_thandle *oh;
4354 /* If parent on remote MDT, we need put this object
4356 oh = container_of(th, typeof(*oh), ot_super);
4357 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4359 CERROR("%s: add "DFID" error: rc = %d\n",
4361 PFID(lu_object_fid(&dt->do_lu)), rc);
4365 child_inode = igrab(omm->omm_remote_parent->d_inode);
4367 child_inode = osd_create_local_agent_inode(env, osd,
4370 if (IS_ERR(child_inode))
4371 RETURN(PTR_ERR(child_inode));
4374 /* Insert local entry */
4375 child = osd_object_find(env, dt, fid);
4376 if (IS_ERR(child)) {
4377 CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4378 osd_name(osd), PFID(fid),
4379 id->oii_ino, id->oii_gen,
4380 (int)PTR_ERR(child));
4381 RETURN(PTR_ERR(child));
4383 child_inode = igrab(child->oo_inode);
4386 rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th);
4388 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4389 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4393 osd_object_put(env, child);
4394 LASSERT(osd_invariant(obj));
4399 * Initialize osd Iterator for given osd index object.
4401 * \param dt osd index object
4404 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4405 struct dt_object *dt,
4407 struct lustre_capa *capa)
4409 struct osd_it_iam *it;
4410 struct osd_thread_info *oti = osd_oti_get(env);
4411 struct osd_object *obj = osd_dt_obj(dt);
4412 struct lu_object *lo = &dt->do_lu;
4413 struct iam_path_descr *ipd;
4414 struct iam_container *bag = &obj->oo_dir->od_container;
4416 LASSERT(lu_object_exists(lo));
4418 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
4419 return ERR_PTR(-EACCES);
4422 ipd = osd_it_ipd_get(env, bag);
4423 if (likely(ipd != NULL)) {
4427 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4428 return (struct dt_it *)it;
4430 return ERR_PTR(-ENOMEM);
4434 * free given Iterator.
4437 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4439 struct osd_it_iam *it = (struct osd_it_iam *)di;
4440 struct osd_object *obj = it->oi_obj;
4442 iam_it_fini(&it->oi_it);
4443 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4444 lu_object_put(env, &obj->oo_dt.do_lu);
4448 * Move Iterator to record specified by \a key
4450 * \param di osd iterator
4451 * \param key key for index
4453 * \retval +ve di points to record with least key not larger than key
4454 * \retval 0 di points to exact matched key
4455 * \retval -ve failure
4458 static int osd_it_iam_get(const struct lu_env *env,
4459 struct dt_it *di, const struct dt_key *key)
4461 struct osd_thread_info *oti = osd_oti_get(env);
4462 struct osd_it_iam *it = (struct osd_it_iam *)di;
4464 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4465 /* swab quota uid/gid */
4466 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4467 key = (struct dt_key *)&oti->oti_quota_id;
4470 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4476 * \param di osd iterator
4478 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4480 struct osd_it_iam *it = (struct osd_it_iam *)di;
4482 iam_it_put(&it->oi_it);
4486 * Move iterator by one record
4488 * \param di osd iterator
4490 * \retval +1 end of container reached
4492 * \retval -ve failure
4495 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4497 struct osd_it_iam *it = (struct osd_it_iam *)di;
4499 return iam_it_next(&it->oi_it);
4503 * Return pointer to the key under iterator.
4506 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4507 const struct dt_it *di)
4509 struct osd_thread_info *oti = osd_oti_get(env);
4510 struct osd_it_iam *it = (struct osd_it_iam *)di;
4511 struct osd_object *obj = it->oi_obj;
4514 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4516 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4517 /* swab quota uid/gid */
4518 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4519 key = (struct dt_key *)&oti->oti_quota_id;
4526 * Return size of key under iterator (in bytes)
4529 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4531 struct osd_it_iam *it = (struct osd_it_iam *)di;
4533 return iam_it_key_size(&it->oi_it);
4537 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4539 /* check if file type is required */
4540 if (ent->lde_attrs & LUDA_TYPE) {
4541 struct luda_type *lt;
4542 int align = sizeof(*lt) - 1;
4544 len = (len + align) & ~align;
4545 lt = (struct luda_type *)(ent->lde_name + len);
4546 lt->lt_type = cpu_to_le16(DTTOIF(type));
4549 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4553 * build lu direct from backend fs dirent.
4557 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4558 char *name, __u16 namelen, __u16 type, __u32 attr)
4560 ent->lde_attrs = attr | LUDA_FID;
4561 fid_cpu_to_le(&ent->lde_fid, fid);
4563 ent->lde_hash = cpu_to_le64(offset);
4564 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4566 strncpy(ent->lde_name, name, namelen);
4567 ent->lde_name[namelen] = '\0';
4568 ent->lde_namelen = cpu_to_le16(namelen);
4570 /* append lustre attributes */
4571 osd_it_append_attrs(ent, namelen, type);
4575 * Return pointer to the record under iterator.
4577 static int osd_it_iam_rec(const struct lu_env *env,
4578 const struct dt_it *di,
4579 struct dt_rec *dtrec, __u32 attr)
4581 struct osd_it_iam *it = (struct osd_it_iam *)di;
4582 struct osd_thread_info *info = osd_oti_get(env);
4585 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4586 const struct osd_fid_pack *rec;
4587 struct lu_fid *fid = &info->oti_fid;
4588 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
4594 name = (char *)iam_it_key_get(&it->oi_it);
4596 RETURN(PTR_ERR(name));
4598 namelen = iam_it_key_size(&it->oi_it);
4600 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4602 RETURN(PTR_ERR(rec));
4604 rc = osd_fid_unpack(fid, rec);
4608 hash = iam_it_store(&it->oi_it);
4610 /* IAM does not store object type in IAM index (dir) */
4611 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4613 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4614 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4615 (struct iam_rec *)dtrec);
4616 osd_quota_unpack(it->oi_obj, dtrec);
4618 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4619 (struct iam_rec *)dtrec);
4626 * Returns cookie for current Iterator position.
4628 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4630 struct osd_it_iam *it = (struct osd_it_iam *)di;
4632 return iam_it_store(&it->oi_it);
4636 * Restore iterator from cookie.
4638 * \param di osd iterator
4639 * \param hash Iterator location cookie
4641 * \retval +ve di points to record with least key not larger than key.
4642 * \retval 0 di points to exact matched key
4643 * \retval -ve failure
4646 static int osd_it_iam_load(const struct lu_env *env,
4647 const struct dt_it *di, __u64 hash)
4649 struct osd_it_iam *it = (struct osd_it_iam *)di;
4651 return iam_it_load(&it->oi_it, hash);
4654 static const struct dt_index_operations osd_index_iam_ops = {
4655 .dio_lookup = osd_index_iam_lookup,
4656 .dio_declare_insert = osd_index_declare_iam_insert,
4657 .dio_insert = osd_index_iam_insert,
4658 .dio_declare_delete = osd_index_declare_iam_delete,
4659 .dio_delete = osd_index_iam_delete,
4661 .init = osd_it_iam_init,
4662 .fini = osd_it_iam_fini,
4663 .get = osd_it_iam_get,
4664 .put = osd_it_iam_put,
4665 .next = osd_it_iam_next,
4666 .key = osd_it_iam_key,
4667 .key_size = osd_it_iam_key_size,
4668 .rec = osd_it_iam_rec,
4669 .store = osd_it_iam_store,
4670 .load = osd_it_iam_load
4676 * Creates or initializes iterator context.
4678 * \retval struct osd_it_ea, iterator structure on success
4681 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4682 struct dt_object *dt,
4684 struct lustre_capa *capa)
4686 struct osd_object *obj = osd_dt_obj(dt);
4687 struct osd_thread_info *info = osd_oti_get(env);
4688 struct osd_it_ea *it = &info->oti_it_ea;
4689 struct file *file = &it->oie_file;
4690 struct lu_object *lo = &dt->do_lu;
4691 struct dentry *obj_dentry = &info->oti_it_dentry;
4693 LASSERT(lu_object_exists(lo));
4695 obj_dentry->d_inode = obj->oo_inode;
4696 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4697 obj_dentry->d_name.hash = 0;
4699 it->oie_rd_dirent = 0;
4700 it->oie_it_dirent = 0;
4701 it->oie_dirent = NULL;
4702 it->oie_buf = info->oti_it_ea_buf;
4705 /* Reset the "file" totally to avoid to reuse any old value from
4706 * former readdir handling, the "file->f_pos" should be zero. */
4707 memset(file, 0, sizeof(*file));
4708 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4709 if (attr & LUDA_64BITHASH)
4710 file->f_mode = FMODE_64BITHASH;
4712 file->f_mode = FMODE_32BITHASH;
4713 file->f_dentry = obj_dentry;
4714 file->f_mapping = obj->oo_inode->i_mapping;
4715 file->f_op = obj->oo_inode->i_fop;
4716 set_file_inode(file, obj->oo_inode);
4719 RETURN((struct dt_it *) it);
4723 * Destroy or finishes iterator context.
4725 * \param di iterator structure to be destroyed
4727 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4729 struct osd_it_ea *it = (struct osd_it_ea *)di;
4730 struct osd_object *obj = it->oie_obj;
4731 struct inode *inode = obj->oo_inode;
4734 it->oie_file.f_op->release(inode, &it->oie_file);
4735 lu_object_put(env, &obj->oo_dt.do_lu);
4740 * It position the iterator at given key, so that next lookup continues from
4741 * that key Or it is similar to dio_it->load() but based on a key,
4742 * rather than file position.
4744 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4747 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4749 static int osd_it_ea_get(const struct lu_env *env,
4750 struct dt_it *di, const struct dt_key *key)
4752 struct osd_it_ea *it = (struct osd_it_ea *)di;
4755 LASSERT(((const char *)key)[0] == '\0');
4756 it->oie_file.f_pos = 0;
4757 it->oie_rd_dirent = 0;
4758 it->oie_it_dirent = 0;
4759 it->oie_dirent = NULL;
4767 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4771 struct osd_filldir_cbs {
4772 #ifdef HAVE_DIR_CONTEXT
4773 struct dir_context ctx;
4775 struct osd_it_ea *it;
4778 * It is called internally by ->readdir(). It fills the
4779 * iterator's in-memory data structure with required
4780 * information i.e. name, namelen, rec_size etc.
4782 * \param buf in which information to be filled in.
4783 * \param name name of the file in given dir
4785 * \retval 0 on success
4786 * \retval 1 on buffer full
4788 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
4789 loff_t offset, __u64 ino,
4792 struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
4793 struct osd_object *obj = it->oie_obj;
4794 struct osd_it_ea_dirent *ent = it->oie_dirent;
4795 struct lu_fid *fid = &ent->oied_fid;
4796 struct osd_fid_pack *rec;
4799 /* this should never happen */
4800 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4801 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4805 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4809 /* "." is just the object itself. */
4810 if (namelen == 1 && name[0] == '.') {
4811 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4812 } else if (d_type & LDISKFS_DIRENT_LUFID) {
4813 rec = (struct osd_fid_pack*) (name + namelen + 1);
4814 if (osd_fid_unpack(fid, rec) != 0)
4819 d_type &= ~LDISKFS_DIRENT_LUFID;
4821 /* NOT export local root. */
4822 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
4823 ino = obj->oo_inode->i_ino;
4824 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4827 ent->oied_ino = ino;
4828 ent->oied_off = offset;
4829 ent->oied_namelen = namelen;
4830 ent->oied_type = d_type;
4832 memcpy(ent->oied_name, name, namelen);
4834 it->oie_rd_dirent++;
4835 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4840 * Calls ->readdir() to load a directory entry at a time
4841 * and stored it in iterator's in-memory data structure.
4843 * \param di iterator's in memory structure
4845 * \retval 0 on success
4846 * \retval -ve on error
4847 * \retval +1 reach the end of entry
4849 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4850 const struct dt_it *di)
4852 struct osd_it_ea *it = (struct osd_it_ea *)di;
4853 struct osd_object *obj = it->oie_obj;
4854 struct inode *inode = obj->oo_inode;
4855 struct htree_lock *hlock = NULL;
4856 struct file *filp = &it->oie_file;
4858 struct osd_filldir_cbs buf = {
4859 #ifdef HAVE_DIR_CONTEXT
4860 .ctx.actor = osd_ldiskfs_filldir,
4866 it->oie_dirent = it->oie_buf;
4867 it->oie_rd_dirent = 0;
4869 if (obj->oo_hl_head != NULL) {
4870 hlock = osd_oti_get(env)->oti_hlock;
4871 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4872 inode, LDISKFS_HLOCK_READDIR);
4874 down_read(&obj->oo_ext_idx_sem);
4877 #ifdef HAVE_DIR_CONTEXT
4878 buf.ctx.pos = filp->f_pos;
4879 rc = inode->i_fop->iterate(filp, &buf.ctx);
4880 filp->f_pos = buf.ctx.pos;
4882 rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
4886 ldiskfs_htree_unlock(hlock);
4888 up_read(&obj->oo_ext_idx_sem);
4890 if (it->oie_rd_dirent == 0) {
4891 /*If it does not get any dirent, it means it has been reached
4892 *to the end of the dir */
4893 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
4897 it->oie_dirent = it->oie_buf;
4898 it->oie_it_dirent = 1;
4905 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4906 * to load a directory entry at a time and stored it in
4907 * iterator's in-memory data structure.
4909 * \param di iterator's in memory structure
4911 * \retval +ve iterator reached to end
4912 * \retval 0 iterator not reached to end
4913 * \retval -ve on error
4915 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4917 struct osd_it_ea *it = (struct osd_it_ea *)di;
4922 if (it->oie_it_dirent < it->oie_rd_dirent) {
4924 (void *) it->oie_dirent +
4925 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4926 it->oie_dirent->oied_namelen);
4927 it->oie_it_dirent++;
4930 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4933 rc = osd_ldiskfs_it_fill(env, di);
4940 * Returns the key at current position from iterator's in memory structure.
4942 * \param di iterator's in memory structure
4944 * \retval key i.e. struct dt_key on success
4946 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4947 const struct dt_it *di)
4949 struct osd_it_ea *it = (struct osd_it_ea *)di;
4951 return (struct dt_key *)it->oie_dirent->oied_name;
4955 * Returns the key's size at current position from iterator's in memory structure.
4957 * \param di iterator's in memory structure
4959 * \retval key_size i.e. struct dt_key on success
4961 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4963 struct osd_it_ea *it = (struct osd_it_ea *)di;
4965 return it->oie_dirent->oied_namelen;
4969 osd_dirent_update(handle_t *jh, struct super_block *sb,
4970 struct osd_it_ea_dirent *ent, struct lu_fid *fid,
4971 struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
4973 struct osd_fid_pack *rec;
4977 LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
4978 LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
4980 rc = ldiskfs_journal_get_write_access(jh, bh);
4982 CERROR("%.16s: fail to write access for update dirent: "
4983 "name = %.*s, rc = %d\n",
4984 LDISKFS_SB(sb)->s_es->s_volume_name,
4985 ent->oied_namelen, ent->oied_name, rc);
4989 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
4990 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
4991 rc = ldiskfs_journal_dirty_metadata(jh, bh);
4993 CERROR("%.16s: fail to dirty metadata for update dirent: "
4994 "name = %.*s, rc = %d\n",
4995 LDISKFS_SB(sb)->s_es->s_volume_name,
4996 ent->oied_namelen, ent->oied_name, rc);
5002 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
5004 if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
5005 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5012 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5014 LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5015 "dot_dotdot = %d\n", dot_dotdot);
5017 if (LDISKFS_DIR_REC_LEN(de) >=
5018 __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5025 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5026 struct inode *dir, struct inode *inode,
5027 struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5028 struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5029 struct htree_lock *hlock)
5031 struct dentry *dentry;
5032 struct osd_fid_pack *rec;
5033 struct ldiskfs_dentry_param *ldp;
5037 if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5038 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5041 /* There is enough space to hold the FID-in-dirent. */
5042 if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
5043 dir->i_sb->s_blocksize)) {
5044 rc = ldiskfs_journal_get_write_access(jh, bh);
5046 CERROR("%.16s: fail to write access for reinsert "
5047 "dirent: name = %.*s, rc = %d\n",
5048 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5049 ent->oied_namelen, ent->oied_name, rc);
5053 de->name[de->name_len] = 0;
5054 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5055 rec->fp_len = sizeof(struct lu_fid) + 1;
5056 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5057 de->file_type |= LDISKFS_DIRENT_LUFID;
5059 rc = ldiskfs_journal_dirty_metadata(jh, bh);
5061 CERROR("%.16s: fail to dirty metadata for reinsert "
5062 "dirent: name = %.*s, rc = %d\n",
5063 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5064 ent->oied_namelen, ent->oied_name, rc);
5069 rc = ldiskfs_delete_entry(jh, dir, de, bh);
5071 CERROR("%.16s: fail to delete entry for reinsert dirent: "
5072 "name = %.*s, rc = %d\n",
5073 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5074 ent->oied_namelen, ent->oied_name, rc);
5078 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5080 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5081 osd_get_ldiskfs_dirent_param(ldp, (const struct dt_rec *)fid);
5082 dentry->d_fsdata = (void *)ldp;
5083 ll_vfs_dq_init(dir);
5084 rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5085 /* It is too bad, we cannot reinsert the name entry back.
5086 * That means we lose it! */
5088 CERROR("%.16s: fail to insert entry for reinsert dirent: "
5089 "name = %.*s, rc = %d\n",
5090 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5091 ent->oied_namelen, ent->oied_name, rc);
5097 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5098 struct osd_it_ea *it, struct lu_fid *fid,
5099 struct osd_inode_id *id, __u32 *attr)
5101 struct osd_thread_info *info = osd_oti_get(env);
5102 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
5103 struct osd_device *dev = osd_obj2dev(obj);
5104 struct super_block *sb = osd_sb(dev);
5105 const char *devname =
5106 LDISKFS_SB(sb)->s_es->s_volume_name;
5107 struct osd_it_ea_dirent *ent = it->oie_dirent;
5108 struct inode *dir = obj->oo_inode;
5109 struct htree_lock *hlock = NULL;
5110 struct buffer_head *bh = NULL;
5111 handle_t *jh = NULL;
5112 struct ldiskfs_dir_entry_2 *de;
5113 struct dentry *dentry;
5114 struct inode *inode;
5121 if (ent->oied_name[0] == '.') {
5122 if (ent->oied_namelen == 1)
5124 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5128 dentry = osd_child_dentry_get(env, obj, ent->oied_name,
5131 /* We need to ensure that the name entry is still valid.
5132 * Because it may be removed or renamed by other already.
5134 * The unlink or rename operation will start journal before PDO lock,
5135 * so to avoid deadlock, here we need to start journal handle before
5136 * related PDO lock also. But because we do not know whether there
5137 * will be something to be repaired before PDO lock, we just start
5138 * journal without conditions.
5140 * We may need to remove the name entry firstly, then insert back.
5141 * One credit is for user quota file update.
5142 * One credit is for group quota file update.
5143 * Two credits are for dirty inode. */
5144 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5145 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5148 if (dev->od_dirent_journal) {
5149 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5152 CERROR("%.16s: fail to start trans for dirent "
5153 "check_repair: credits %d, name %.*s, rc %d\n",
5154 devname, credits, ent->oied_namelen,
5155 ent->oied_name, rc);
5159 if (obj->oo_hl_head != NULL) {
5160 hlock = osd_oti_get(env)->oti_hlock;
5161 /* "0" means exclusive lock for the whole directory.
5162 * We need to prevent others access such name entry
5163 * during the delete + insert. Neither HLOCK_ADD nor
5164 * HLOCK_DEL cannot guarantee the atomicity. */
5165 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5167 down_write(&obj->oo_ext_idx_sem);
5170 if (obj->oo_hl_head != NULL) {
5171 hlock = osd_oti_get(env)->oti_hlock;
5172 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5173 LDISKFS_HLOCK_LOOKUP);
5175 down_read(&obj->oo_ext_idx_sem);
5179 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5180 /* For dot/dotdot entry, if there is not enough space to hold the
5181 * FID-in-dirent, just keep them there. It only happens when the
5182 * device upgraded from 1.8 or restored from MDT file-level backup.
5183 * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5184 * and needs to get FID from LMA when readdir, it will not affect the
5185 * performance much. */
5186 if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
5187 (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5188 *attr |= LUDA_IGNORE;
5189 GOTO(out_journal, rc = 0);
5192 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5193 inode = osd_iget(info, dev, id);
5194 if (IS_ERR(inode)) {
5195 rc = PTR_ERR(inode);
5196 if (rc == -ENOENT || rc == -ESTALE) {
5197 *attr |= LUDA_IGNORE;
5201 GOTO(out_journal, rc);
5204 /* skip the REMOTE_PARENT_DIR. */
5205 if (inode == dev->od_mdt_map->omm_remote_parent->d_inode)
5206 GOTO(out_inode, rc = 0);
5208 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
5210 LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
5212 if (fid_is_sane(fid)) {
5213 /* FID-in-dirent is valid. */
5214 if (lu_fid_eq(fid, &lma->lma_self_fid))
5215 GOTO(out_inode, rc = 0);
5217 /* Do not repair under dryrun mode. */
5218 if (*attr & LUDA_VERIFY_DRYRUN) {
5219 *attr |= LUDA_REPAIR;
5220 GOTO(out_inode, rc = 0);
5223 if (!dev->od_dirent_journal) {
5227 ldiskfs_htree_unlock(hlock);
5229 up_read(&obj->oo_ext_idx_sem);
5230 dev->od_dirent_journal = 1;
5234 *fid = lma->lma_self_fid;
5236 /* Update the FID-in-dirent. */
5237 rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
5239 *attr |= LUDA_REPAIR;
5241 /* Do not repair under dryrun mode. */
5242 if (*attr & LUDA_VERIFY_DRYRUN) {
5243 *fid = lma->lma_self_fid;
5244 *attr |= LUDA_REPAIR;
5245 GOTO(out_inode, rc = 0);
5248 if (!dev->od_dirent_journal) {
5252 ldiskfs_htree_unlock(hlock);
5254 up_read(&obj->oo_ext_idx_sem);
5255 dev->od_dirent_journal = 1;
5259 *fid = lma->lma_self_fid;
5261 /* Append the FID-in-dirent. */
5262 rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5263 fid, bh, de, hlock);
5265 *attr |= LUDA_REPAIR;
5267 } else if (rc == -ENODATA) {
5268 /* Do not repair under dryrun mode. */
5269 if (*attr & LUDA_VERIFY_DRYRUN) {
5270 if (fid_is_sane(fid)) {
5271 *attr |= LUDA_REPAIR;
5273 lu_igif_build(fid, inode->i_ino,
5274 inode->i_generation);
5275 *attr |= LUDA_UPGRADE;
5277 GOTO(out_inode, rc = 0);
5280 if (!dev->od_dirent_journal) {
5284 ldiskfs_htree_unlock(hlock);
5286 up_read(&obj->oo_ext_idx_sem);
5287 dev->od_dirent_journal = 1;
5292 if (unlikely(fid_is_sane(fid))) {
5293 /* FID-in-dirent exists, but FID-in-LMA is lost.
5294 * Trust the FID-in-dirent, and add FID-in-LMA. */
5295 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5297 *attr |= LUDA_REPAIR;
5299 lu_igif_build(fid, inode->i_ino, inode->i_generation);
5300 /* It is probably IGIF object. Only aappend the
5301 * FID-in-dirent. OI scrub will process FID-in-LMA. */
5302 rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5303 fid, bh, de, hlock);
5305 *attr |= LUDA_UPGRADE;
5309 GOTO(out_inode, rc);
5316 if (hlock != NULL) {
5317 ldiskfs_htree_unlock(hlock);
5319 if (dev->od_dirent_journal)
5320 up_write(&obj->oo_ext_idx_sem);
5322 up_read(&obj->oo_ext_idx_sem);
5325 ldiskfs_journal_stop(jh);
5326 if (rc >= 0 && !dirty)
5327 dev->od_dirent_journal = 0;
5332 * Returns the value at current position from iterator's in memory structure.
5334 * \param di struct osd_it_ea, iterator's in memory structure
5335 * \param attr attr requested for dirent.
5336 * \param lde lustre dirent
5338 * \retval 0 no error and \param lde has correct lustre dirent.
5339 * \retval -ve on error
5341 static inline int osd_it_ea_rec(const struct lu_env *env,
5342 const struct dt_it *di,
5343 struct dt_rec *dtrec, __u32 attr)
5345 struct osd_it_ea *it = (struct osd_it_ea *)di;
5346 struct osd_object *obj = it->oie_obj;
5347 struct osd_device *dev = osd_obj2dev(obj);
5348 struct osd_scrub *scrub = &dev->od_scrub;
5349 struct scrub_file *sf = &scrub->os_file;
5350 struct osd_thread_info *oti = osd_oti_get(env);
5351 struct osd_inode_id *id = &oti->oti_id;
5352 struct osd_idmap_cache *oic = &oti->oti_cache;
5353 struct lu_fid *fid = &it->oie_dirent->oied_fid;
5354 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
5355 __u32 ino = it->oie_dirent->oied_ino;
5359 if (attr & LUDA_VERIFY) {
5361 if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) {
5362 attr |= LUDA_IGNORE;
5365 rc = osd_dirent_check_repair(env, obj, it, fid, id,
5369 attr &= ~LU_DIRENT_ATTRS_MASK;
5370 if (!fid_is_sane(fid)) {
5371 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP) &&
5372 likely(it->oie_dirent->oied_namelen != 2 ||
5373 it->oie_dirent->oied_name[0] != '.' ||
5374 it->oie_dirent->oied_name[1] != '.'))
5377 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5379 osd_id_gen(id, ino, OSD_OII_NOGEN);
5383 /* Pack the entry anyway, at least the offset is right. */
5384 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5385 it->oie_dirent->oied_name,
5386 it->oie_dirent->oied_namelen,
5387 it->oie_dirent->oied_type, attr);
5392 if (osd_remote_fid(env, dev, fid))
5395 if (likely(!(attr & LUDA_IGNORE)))
5396 rc = osd_add_oi_cache(oti, dev, id, fid);
5398 if (!(attr & LUDA_VERIFY) &&
5399 (scrub->os_pos_current <= ino) &&
5400 ((sf->sf_flags & SF_INCONSISTENT) ||
5401 (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
5402 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
5403 osd_consistency_check(oti, dev, oic);
5409 * Returns the record size size at current position.
5411 * This function will return record(lu_dirent) size in bytes.
5413 * \param[in] env execution environment
5414 * \param[in] di iterator's in memory structure
5415 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
5416 * calculate the lu_dirent size.
5418 * \retval record size(in bytes & in memory) of the current lu_dirent
5421 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5424 struct osd_it_ea *it = (struct osd_it_ea *)di;
5426 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5430 * Returns a cookie for current position of the iterator head, so that
5431 * user can use this cookie to load/start the iterator next time.
5433 * \param di iterator's in memory structure
5435 * \retval cookie for current position, on success
5437 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5439 struct osd_it_ea *it = (struct osd_it_ea *)di;
5441 return it->oie_dirent->oied_off;
5445 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5446 * to load a directory entry at a time and stored it i inn,
5447 * in iterator's in-memory data structure.
5449 * \param di struct osd_it_ea, iterator's in memory structure
5451 * \retval +ve on success
5452 * \retval -ve on error
5454 static int osd_it_ea_load(const struct lu_env *env,
5455 const struct dt_it *di, __u64 hash)
5457 struct osd_it_ea *it = (struct osd_it_ea *)di;
5461 it->oie_file.f_pos = hash;
5463 rc = osd_ldiskfs_it_fill(env, di);
5474 * Index lookup function for interoperability mode (b11826).
5476 * \param key, key i.e. file name to be searched
5478 * \retval +ve, on success
5479 * \retval -ve, on error
5481 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5482 struct dt_rec *rec, const struct dt_key *key,
5483 struct lustre_capa *capa)
5485 struct osd_object *obj = osd_dt_obj(dt);
5490 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5491 LINVRNT(osd_invariant(obj));
5493 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
5496 rc = osd_ea_lookup_rec(env, obj, rec, key);
5503 * Index and Iterator operations for interoperability
5504 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5506 static const struct dt_index_operations osd_index_ea_ops = {
5507 .dio_lookup = osd_index_ea_lookup,
5508 .dio_declare_insert = osd_index_declare_ea_insert,
5509 .dio_insert = osd_index_ea_insert,
5510 .dio_declare_delete = osd_index_declare_ea_delete,
5511 .dio_delete = osd_index_ea_delete,
5513 .init = osd_it_ea_init,
5514 .fini = osd_it_ea_fini,
5515 .get = osd_it_ea_get,
5516 .put = osd_it_ea_put,
5517 .next = osd_it_ea_next,
5518 .key = osd_it_ea_key,
5519 .key_size = osd_it_ea_key_size,
5520 .rec = osd_it_ea_rec,
5521 .rec_size = osd_it_ea_rec_size,
5522 .store = osd_it_ea_store,
5523 .load = osd_it_ea_load
5527 static void *osd_key_init(const struct lu_context *ctx,
5528 struct lu_context_key *key)
5530 struct osd_thread_info *info;
5532 OBD_ALLOC_PTR(info);
5534 return ERR_PTR(-ENOMEM);
5536 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5537 if (info->oti_it_ea_buf == NULL)
5540 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5542 info->oti_hlock = ldiskfs_htree_lock_alloc();
5543 if (info->oti_hlock == NULL)
5549 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5552 return ERR_PTR(-ENOMEM);
5555 static void osd_key_fini(const struct lu_context *ctx,
5556 struct lu_context_key *key, void* data)
5558 struct osd_thread_info *info = data;
5560 if (info->oti_hlock != NULL)
5561 ldiskfs_htree_lock_free(info->oti_hlock);
5562 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5563 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5564 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5568 static void osd_key_exit(const struct lu_context *ctx,
5569 struct lu_context_key *key, void *data)
5571 struct osd_thread_info *info = data;
5573 LASSERT(info->oti_r_locks == 0);
5574 LASSERT(info->oti_w_locks == 0);
5575 LASSERT(info->oti_txns == 0);
5578 /* type constructor/destructor: osd_type_init, osd_type_fini */
5579 LU_TYPE_INIT_FINI(osd, &osd_key);
5581 struct lu_context_key osd_key = {
5582 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5583 .lct_init = osd_key_init,
5584 .lct_fini = osd_key_fini,
5585 .lct_exit = osd_key_exit
5589 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5590 const char *name, struct lu_device *next)
5592 struct osd_device *osd = osd_dev(d);
5594 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5595 >= sizeof(osd->od_svname))
5597 return osd_procfs_init(osd, name);
5600 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5602 struct seq_server_site *ss = osd_seq_site(osd);
5606 if (osd->od_is_ost || osd->od_cl_seq != NULL)
5609 if (unlikely(ss == NULL))
5612 OBD_ALLOC_PTR(osd->od_cl_seq);
5613 if (osd->od_cl_seq == NULL)
5616 rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5617 osd->od_svname, ss->ss_server_seq);
5620 OBD_FREE_PTR(osd->od_cl_seq);
5621 osd->od_cl_seq = NULL;
5627 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5629 if (osd->od_cl_seq == NULL)
5632 seq_client_fini(osd->od_cl_seq);
5633 OBD_FREE_PTR(osd->od_cl_seq);
5634 osd->od_cl_seq = NULL;
5637 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5641 /* shutdown quota slave instance associated with the device */
5642 if (o->od_quota_slave != NULL) {
5643 qsd_fini(env, o->od_quota_slave);
5644 o->od_quota_slave = NULL;
5647 osd_fid_fini(env, o);
5652 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5656 if (o->od_mnt != NULL) {
5657 shrink_dcache_sb(osd_sb(o));
5658 osd_sync(env, &o->od_dt_dev);
5667 static int osd_mount(const struct lu_env *env,
5668 struct osd_device *o, struct lustre_cfg *cfg)
5670 const char *name = lustre_cfg_string(cfg, 0);
5671 const char *dev = lustre_cfg_string(cfg, 1);
5673 unsigned long page, s_flags, lmd_flags = 0;
5674 struct page *__page;
5675 struct file_system_type *type;
5676 char *options = NULL;
5678 struct osd_thread_info *info = osd_oti_get(env);
5679 struct lu_fid *fid = &info->oti_fid;
5680 struct inode *inode;
5684 if (o->od_mnt != NULL)
5687 if (strlen(dev) >= sizeof(o->od_mntdev))
5689 strcpy(o->od_mntdev, dev);
5691 OBD_PAGE_ALLOC(__page, GFP_IOFS);
5693 GOTO(out, rc = -ENOMEM);
5695 str = lustre_cfg_string(cfg, 2);
5696 s_flags = simple_strtoul(str, NULL, 0);
5697 str = strstr(str, ":");
5699 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5700 opts = lustre_cfg_string(cfg, 3);
5701 page = (unsigned long)page_address(__page);
5702 options = (char *)page;
5705 strcat(options, "user_xattr,acl");
5707 strcat(options, opts);
5709 /* Glom up mount options */
5710 if (*options != '\0')
5711 strcat(options, ",");
5712 strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
5714 type = get_fs_type("ldiskfs");
5716 CERROR("%s: cannot find ldiskfs module\n", name);
5717 GOTO(out, rc = -ENODEV);
5720 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5721 module_put(type->owner);
5723 if (IS_ERR(o->od_mnt)) {
5724 rc = PTR_ERR(o->od_mnt);
5726 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5730 #ifdef HAVE_DEV_SET_RDONLY
5731 if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5732 CERROR("%s: underlying device %s is marked as read-only. "
5733 "Setup failed\n", name, dev);
5734 GOTO(out_mnt, rc = -EROFS);
5738 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5739 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5740 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5741 GOTO(out_mnt, rc = -EINVAL);
5744 #ifdef LDISKFS_MOUNT_DIRDATA
5745 if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
5746 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5747 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
5748 else if (!o->od_is_ost)
5749 CWARN("%s: device %s was upgraded from Lustre-1.x without "
5750 "enabling the dirdata feature. If you do not want to "
5751 "downgrade to Lustre-1.x again, you can enable it via "
5752 "'tune2fs -O dirdata device'\n", name, dev);
5754 inode = osd_sb(o)->s_root->d_inode;
5755 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
5756 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
5758 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
5762 if (lmd_flags & LMD_FLG_NOSCRUB)
5773 OBD_PAGE_FREE(__page);
5778 static struct lu_device *osd_device_fini(const struct lu_env *env,
5779 struct lu_device *d)
5781 struct osd_device *o = osd_dev(d);
5784 osd_shutdown(env, o);
5786 osd_scrub_cleanup(env, o);
5787 osd_obj_map_fini(o);
5793 static int osd_device_init0(const struct lu_env *env,
5794 struct osd_device *o,
5795 struct lustre_cfg *cfg)
5797 struct lu_device *l = osd2lu_dev(o);
5798 struct osd_thread_info *info;
5802 /* if the module was re-loaded, env can loose its keys */
5803 rc = lu_env_refill((struct lu_env *) env);
5806 info = osd_oti_get(env);
5809 l->ld_ops = &osd_lu_ops;
5810 o->od_dt_dev.dd_ops = &osd_dt_ops;
5812 spin_lock_init(&o->od_osfs_lock);
5813 mutex_init(&o->od_otable_mutex);
5815 o->od_capa_hash = init_capa_hash();
5816 if (o->od_capa_hash == NULL)
5817 GOTO(out, rc = -ENOMEM);
5819 o->od_read_cache = 1;
5820 o->od_writethrough_cache = 1;
5821 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
5823 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
5824 sizeof(o->od_svname));
5825 if (cplen >= sizeof(o->od_svname)) {
5830 if (server_name_is_ost(o->od_svname))
5833 rc = osd_mount(env, o, cfg);
5837 rc = osd_obj_map_init(env, o);
5841 rc = lu_site_init(&o->od_site, l);
5843 GOTO(out_compat, rc);
5844 o->od_site.ls_bottom_dev = l;
5846 rc = lu_site_init_finish(&o->od_site);
5850 /* self-repair LMA by default */
5851 o->od_lma_self_repair = 1;
5853 CFS_INIT_LIST_HEAD(&o->od_ios_list);
5854 /* setup scrub, including OI files initialization */
5855 rc = osd_scrub_setup(env, o);
5859 rc = osd_procfs_init(o, o->od_svname);
5861 CERROR("%s: can't initialize procfs: rc = %d\n",
5863 GOTO(out_scrub, rc);
5866 LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
5868 /* initialize quota slave instance */
5869 o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
5871 if (IS_ERR(o->od_quota_slave)) {
5872 rc = PTR_ERR(o->od_quota_slave);
5873 o->od_quota_slave = NULL;
5874 GOTO(out_procfs, rc);
5882 osd_scrub_cleanup(env, o);
5884 lu_site_fini(&o->od_site);
5886 osd_obj_map_fini(o);
5890 cleanup_capa_hash(o->od_capa_hash);
5895 static struct lu_device *osd_device_alloc(const struct lu_env *env,
5896 struct lu_device_type *t,
5897 struct lustre_cfg *cfg)
5899 struct osd_device *o;
5904 return ERR_PTR(-ENOMEM);
5906 rc = dt_device_init(&o->od_dt_dev, t);
5908 /* Because the ctx might be revived in dt_device_init,
5909 * refill the env here */
5910 lu_env_refill((struct lu_env *)env);
5911 rc = osd_device_init0(env, o, cfg);
5913 dt_device_fini(&o->od_dt_dev);
5916 if (unlikely(rc != 0))
5919 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
5922 static struct lu_device *osd_device_free(const struct lu_env *env,
5923 struct lu_device *d)
5925 struct osd_device *o = osd_dev(d);
5928 cleanup_capa_hash(o->od_capa_hash);
5929 /* XXX: make osd top device in order to release reference */
5930 d->ld_site->ls_top_dev = d;
5931 lu_site_purge(env, d->ld_site, -1);
5932 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
5933 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
5934 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
5936 lu_site_fini(&o->od_site);
5937 dt_device_fini(&o->od_dt_dev);
5942 static int osd_process_config(const struct lu_env *env,
5943 struct lu_device *d, struct lustre_cfg *cfg)
5945 struct osd_device *o = osd_dev(d);
5949 switch (cfg->lcfg_command) {
5951 rc = osd_mount(env, o, cfg);
5954 lu_dev_del_linkage(d->ld_site, d);
5955 rc = osd_shutdown(env, o);
5958 LASSERT(&o->od_dt_dev);
5959 rc = class_process_proc_seq_param(PARAM_OSD,
5960 lprocfs_osd_obd_vars,
5961 cfg, &o->od_dt_dev);
5962 if (rc > 0 || rc == -ENOSYS)
5963 rc = class_process_proc_seq_param(PARAM_OST,
5964 lprocfs_osd_obd_vars,
5965 cfg, &o->od_dt_dev);
5974 static int osd_recovery_complete(const struct lu_env *env,
5975 struct lu_device *d)
5977 struct osd_device *osd = osd_dev(d);
5981 if (osd->od_quota_slave == NULL)
5984 /* start qsd instance on recovery completion, this notifies the quota
5985 * slave code that we are about to process new requests now */
5986 rc = qsd_start(env, osd->od_quota_slave);
5991 * we use exports to track all osd users
5993 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
5994 struct obd_device *obd, struct obd_uuid *cluuid,
5995 struct obd_connect_data *data, void *localdata)
5997 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
5998 struct lustre_handle conn;
6002 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6004 rc = class_connect(&conn, obd, cluuid);
6008 *exp = class_conn2export(&conn);
6010 spin_lock(&osd->od_osfs_lock);
6012 spin_unlock(&osd->od_osfs_lock);
6018 * once last export (we don't count self-export) disappeared
6019 * osd can be released
6021 static int osd_obd_disconnect(struct obd_export *exp)
6023 struct obd_device *obd = exp->exp_obd;
6024 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6025 int rc, release = 0;
6028 /* Only disconnect the underlying layers on the final disconnect. */
6029 spin_lock(&osd->od_osfs_lock);
6031 if (osd->od_connects == 0)
6033 spin_unlock(&osd->od_osfs_lock);
6035 rc = class_disconnect(exp); /* bz 9811 */
6037 if (rc == 0 && release)
6038 class_manual_cleanup(obd);
6042 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6043 struct lu_device *dev)
6045 struct osd_device *osd = osd_dev(dev);
6049 if (osd->od_quota_slave != NULL) {
6050 /* set up quota slave objects */
6051 result = qsd_prepare(env, osd->od_quota_slave);
6056 result = osd_fid_init(env, osd);
6061 int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6062 struct lu_fid *fid, struct md_op_data *op_data)
6064 struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6066 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6069 static const struct lu_object_operations osd_lu_obj_ops = {
6070 .loo_object_init = osd_object_init,
6071 .loo_object_delete = osd_object_delete,
6072 .loo_object_release = osd_object_release,
6073 .loo_object_free = osd_object_free,
6074 .loo_object_print = osd_object_print,
6075 .loo_object_invariant = osd_object_invariant
6078 const struct lu_device_operations osd_lu_ops = {
6079 .ldo_object_alloc = osd_object_alloc,
6080 .ldo_process_config = osd_process_config,
6081 .ldo_recovery_complete = osd_recovery_complete,
6082 .ldo_prepare = osd_prepare,
6085 static const struct lu_device_type_operations osd_device_type_ops = {
6086 .ldto_init = osd_type_init,
6087 .ldto_fini = osd_type_fini,
6089 .ldto_start = osd_type_start,
6090 .ldto_stop = osd_type_stop,
6092 .ldto_device_alloc = osd_device_alloc,
6093 .ldto_device_free = osd_device_free,
6095 .ldto_device_init = osd_device_init,
6096 .ldto_device_fini = osd_device_fini
6099 struct lu_device_type osd_device_type = {
6100 .ldt_tags = LU_DEVICE_DT,
6101 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
6102 .ldt_ops = &osd_device_type_ops,
6103 .ldt_ctx_tags = LCT_LOCAL,
6107 * lprocfs legacy support.
6109 static struct obd_ops osd_obd_device_ops = {
6110 .o_owner = THIS_MODULE,
6111 .o_connect = osd_obd_connect,
6112 .o_disconnect = osd_obd_disconnect,
6113 .o_fid_alloc = osd_fid_alloc,
6116 static int __init osd_mod_init(void)
6122 rc = lu_kmem_init(ldiskfs_caches);
6126 rc = class_register_type(&osd_obd_device_ops, NULL, true,
6127 lprocfs_osd_module_vars,
6128 #ifndef HAVE_ONLY_PROCFS_SEQ
6131 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6133 lu_kmem_fini(ldiskfs_caches);
6137 static void __exit osd_mod_exit(void)
6139 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6140 lu_kmem_fini(ldiskfs_caches);
6143 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6144 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6145 MODULE_LICENSE("GPL");
6147 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);