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 (!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 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 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 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 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 = i_uid_read(inode);
1195 qid_t gid = i_gid_read(inode);
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 = i_uid_read(inode);
1704 attr->la_gid = i_gid_read(inode);
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;
1747 LASSERT(dt != NULL);
1748 LASSERT(handle != NULL);
1750 obj = osd_dt_obj(dt);
1751 LASSERT(osd_invariant(obj));
1753 oh = container_of0(handle, struct osd_thandle, ot_super);
1754 LASSERT(oh->ot_handle == NULL);
1756 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1757 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1759 if (attr == NULL || obj->oo_inode == NULL)
1762 bspace = obj->oo_inode->i_blocks;
1763 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1764 bspace = toqb(bspace);
1766 /* Changing ownership is always preformed by super user, it should not
1769 * We still need to call the osd_declare_qid() to calculate the journal
1770 * credits for updating quota accounting files and to trigger quota
1771 * space adjustment once the operation is completed.*/
1772 if ((attr->la_valid & LA_UID) != 0 &&
1773 attr->la_uid != (uid = i_uid_read(obj->oo_inode))) {
1774 qi->lqi_type = USRQUOTA;
1776 /* inode accounting */
1777 qi->lqi_is_blk = false;
1779 /* one more inode for the new owner ... */
1780 qi->lqi_id.qid_uid = attr->la_uid;
1782 allocated = (attr->la_uid == 0) ? true : false;
1783 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1784 if (rc == -EDQUOT || rc == -EINPROGRESS)
1789 /* and one less inode for the current uid */
1790 qi->lqi_id.qid_uid = uid;
1792 rc = osd_declare_qid(env, oh, qi, true, NULL);
1793 if (rc == -EDQUOT || rc == -EINPROGRESS)
1798 /* block accounting */
1799 qi->lqi_is_blk = true;
1801 /* more blocks for the new owner ... */
1802 qi->lqi_id.qid_uid = attr->la_uid;
1803 qi->lqi_space = bspace;
1804 allocated = (attr->la_uid == 0) ? true : false;
1805 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1806 if (rc == -EDQUOT || rc == -EINPROGRESS)
1811 /* and finally less blocks for the current owner */
1812 qi->lqi_id.qid_uid = uid;
1813 qi->lqi_space = -bspace;
1814 rc = osd_declare_qid(env, oh, qi, true, NULL);
1815 if (rc == -EDQUOT || rc == -EINPROGRESS)
1821 if (attr->la_valid & LA_GID &&
1822 attr->la_gid != (gid = i_gid_read(obj->oo_inode))) {
1823 qi->lqi_type = GRPQUOTA;
1825 /* inode accounting */
1826 qi->lqi_is_blk = false;
1828 /* one more inode for the new group owner ... */
1829 qi->lqi_id.qid_gid = attr->la_gid;
1831 allocated = (attr->la_gid == 0) ? true : false;
1832 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1833 if (rc == -EDQUOT || rc == -EINPROGRESS)
1838 /* and one less inode for the current gid */
1839 qi->lqi_id.qid_gid = gid;
1841 rc = osd_declare_qid(env, oh, qi, true, NULL);
1842 if (rc == -EDQUOT || rc == -EINPROGRESS)
1847 /* block accounting */
1848 qi->lqi_is_blk = true;
1850 /* more blocks for the new owner ... */
1851 qi->lqi_id.qid_gid = attr->la_gid;
1852 qi->lqi_space = bspace;
1853 allocated = (attr->la_gid == 0) ? true : false;
1854 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1855 if (rc == -EDQUOT || rc == -EINPROGRESS)
1860 /* and finally less blocks for the current owner */
1861 qi->lqi_id.qid_gid = gid;
1862 qi->lqi_space = -bspace;
1863 rc = osd_declare_qid(env, oh, qi, true, NULL);
1864 if (rc == -EDQUOT || rc == -EINPROGRESS)
1873 static int osd_inode_setattr(const struct lu_env *env,
1874 struct inode *inode, const struct lu_attr *attr)
1878 bits = attr->la_valid;
1880 if (bits & LA_ATIME)
1881 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
1882 if (bits & LA_CTIME)
1883 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
1884 if (bits & LA_MTIME)
1885 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
1886 if (bits & LA_SIZE) {
1887 LDISKFS_I(inode)->i_disksize = attr->la_size;
1888 i_size_write(inode, attr->la_size);
1892 /* OSD should not change "i_blocks" which is used by quota.
1893 * "i_blocks" should be changed by ldiskfs only. */
1894 if (bits & LA_BLOCKS)
1895 inode->i_blocks = attr->la_blocks;
1898 inode->i_mode = (inode->i_mode & S_IFMT) |
1899 (attr->la_mode & ~S_IFMT);
1901 i_uid_write(inode, attr->la_uid);
1903 i_gid_write(inode, attr->la_gid);
1904 if (bits & LA_NLINK)
1905 set_nlink(inode, attr->la_nlink);
1907 inode->i_rdev = attr->la_rdev;
1909 if (bits & LA_FLAGS) {
1910 /* always keep S_NOCMTIME */
1911 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1917 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1919 if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1920 (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1925 if (attr->la_valid & LA_UID)
1926 iattr.ia_valid |= ATTR_UID;
1927 if (attr->la_valid & LA_GID)
1928 iattr.ia_valid |= ATTR_GID;
1929 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1930 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1932 rc = ll_vfs_dq_transfer(inode, &iattr);
1934 CERROR("%s: quota transfer failed: rc = %d. Is quota "
1935 "enforcement enabled on the ldiskfs filesystem?",
1936 inode->i_sb->s_id, rc);
1943 static int osd_attr_set(const struct lu_env *env,
1944 struct dt_object *dt,
1945 const struct lu_attr *attr,
1946 struct thandle *handle,
1947 struct lustre_capa *capa)
1949 struct osd_object *obj = osd_dt_obj(dt);
1950 struct inode *inode;
1953 LASSERT(handle != NULL);
1954 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1955 LASSERT(osd_invariant(obj));
1957 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1960 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1962 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1963 struct osd_thread_info *oti = osd_oti_get(env);
1964 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
1965 struct lu_fid *fid1 = &oti->oti_fid;
1966 struct osd_inode_id *id = &oti->oti_id;
1967 struct iam_path_descr *ipd;
1968 struct iam_container *bag;
1969 struct osd_thandle *oh;
1972 fid_cpu_to_be(fid1, fid0);
1973 memset(id, 1, sizeof(*id));
1974 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1975 fid0)->oi_dir.od_container;
1976 ipd = osd_idx_ipd_get(env, bag);
1977 if (unlikely(ipd == NULL))
1980 oh = container_of0(handle, struct osd_thandle, ot_super);
1981 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1982 (const struct iam_rec *)id, ipd);
1983 osd_ipd_put(env, bag, ipd);
1984 return(rc > 0 ? 0 : rc);
1987 inode = obj->oo_inode;
1988 ll_vfs_dq_init(inode);
1990 rc = osd_quota_transfer(inode, attr);
1994 spin_lock(&obj->oo_guard);
1995 rc = osd_inode_setattr(env, inode, attr);
1996 spin_unlock(&obj->oo_guard);
1999 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2003 struct dentry *osd_child_dentry_get(const struct lu_env *env,
2004 struct osd_object *obj,
2005 const char *name, const int namelen)
2007 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2010 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2011 umode_t mode, struct dt_allocation_hint *hint,
2015 struct osd_device *osd = osd_obj2dev(obj);
2016 struct osd_thandle *oth;
2017 struct dt_object *parent = NULL;
2018 struct inode *inode;
2020 LINVRNT(osd_invariant(obj));
2021 LASSERT(obj->oo_inode == NULL);
2022 LASSERT(obj->oo_hl_head == NULL);
2024 if (S_ISDIR(mode) && ldiskfs_pdo) {
2025 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2026 if (obj->oo_hl_head == NULL)
2030 oth = container_of(th, struct osd_thandle, ot_super);
2031 LASSERT(oth->ot_handle->h_transaction != NULL);
2033 if (hint && hint->dah_parent)
2034 parent = hint->dah_parent;
2036 inode = ldiskfs_create_inode(oth->ot_handle,
2037 parent ? osd_dt_obj(parent)->oo_inode :
2038 osd_sb(osd)->s_root->d_inode,
2040 if (!IS_ERR(inode)) {
2041 /* Do not update file c/mtime in ldiskfs.
2042 * NB: don't need any lock because no contention at this
2044 inode->i_flags |= S_NOCMTIME;
2046 /* For new created object, it must be consistent,
2047 * and it is unnecessary to scrub against it. */
2048 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2049 obj->oo_inode = inode;
2052 if (obj->oo_hl_head != NULL) {
2053 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2054 obj->oo_hl_head = NULL;
2056 result = PTR_ERR(inode);
2058 LINVRNT(osd_invariant(obj));
2066 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2067 struct lu_attr *attr,
2068 struct dt_allocation_hint *hint,
2069 struct dt_object_format *dof,
2073 struct osd_thandle *oth;
2074 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2076 LASSERT(S_ISDIR(attr->la_mode));
2078 oth = container_of(th, struct osd_thandle, ot_super);
2079 LASSERT(oth->ot_handle->h_transaction != NULL);
2080 result = osd_mkfile(info, obj, mode, hint, th);
2085 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2086 struct lu_attr *attr,
2087 struct dt_allocation_hint *hint,
2088 struct dt_object_format *dof,
2092 struct osd_thandle *oth;
2093 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2095 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2097 LASSERT(S_ISREG(attr->la_mode));
2099 oth = container_of(th, struct osd_thandle, ot_super);
2100 LASSERT(oth->ot_handle->h_transaction != NULL);
2102 result = osd_mkfile(info, obj, mode, hint, th);
2104 LASSERT(obj->oo_inode != NULL);
2105 if (feat->dif_flags & DT_IND_VARKEY)
2106 result = iam_lvar_create(obj->oo_inode,
2107 feat->dif_keysize_max,
2109 feat->dif_recsize_max,
2112 result = iam_lfix_create(obj->oo_inode,
2113 feat->dif_keysize_max,
2115 feat->dif_recsize_max,
2122 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2123 struct lu_attr *attr,
2124 struct dt_allocation_hint *hint,
2125 struct dt_object_format *dof,
2128 LASSERT(S_ISREG(attr->la_mode));
2129 return osd_mkfile(info, obj, (attr->la_mode &
2130 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2133 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2134 struct lu_attr *attr,
2135 struct dt_allocation_hint *hint,
2136 struct dt_object_format *dof,
2139 LASSERT(S_ISLNK(attr->la_mode));
2140 return osd_mkfile(info, obj, (attr->la_mode &
2141 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2144 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2145 struct lu_attr *attr,
2146 struct dt_allocation_hint *hint,
2147 struct dt_object_format *dof,
2150 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2153 LINVRNT(osd_invariant(obj));
2154 LASSERT(obj->oo_inode == NULL);
2155 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2156 S_ISFIFO(mode) || S_ISSOCK(mode));
2158 result = osd_mkfile(info, obj, mode, hint, th);
2160 LASSERT(obj->oo_inode != NULL);
2162 * This inode should be marked dirty for i_rdev. Currently
2163 * that is done in the osd_attr_init().
2165 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2168 LINVRNT(osd_invariant(obj));
2172 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2174 struct dt_allocation_hint *hint,
2175 struct dt_object_format *dof,
2178 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2180 osd_obj_type_f result;
2196 result = osd_mk_index;
2207 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2208 struct dt_object *parent, struct dt_object *child,
2213 ah->dah_parent = parent;
2214 ah->dah_mode = child_mode;
2217 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2218 struct lu_attr *attr, struct dt_object_format *dof)
2220 struct inode *inode = obj->oo_inode;
2221 __u64 valid = attr->la_valid;
2224 attr->la_valid &= ~(LA_TYPE | LA_MODE);
2226 if (dof->dof_type != DFT_NODE)
2227 attr->la_valid &= ~LA_RDEV;
2228 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2229 attr->la_valid &= ~LA_ATIME;
2230 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2231 attr->la_valid &= ~LA_CTIME;
2232 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2233 attr->la_valid &= ~LA_MTIME;
2235 result = osd_quota_transfer(inode, attr);
2239 if (attr->la_valid != 0) {
2240 result = osd_inode_setattr(info->oti_env, inode, attr);
2242 * The osd_inode_setattr() should always succeed here. The
2243 * only error that could be returned is EDQUOT when we are
2244 * trying to change the UID or GID of the inode. However, this
2245 * should not happen since quota enforcement is no longer
2246 * enabled on ldiskfs (lquota takes care of it).
2248 LASSERTF(result == 0, "%d", result);
2249 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2252 attr->la_valid = valid;
2256 * Helper function for osd_object_create()
2258 * \retval 0, on success
2260 static int __osd_object_create(struct osd_thread_info *info,
2261 struct osd_object *obj, struct lu_attr *attr,
2262 struct dt_allocation_hint *hint,
2263 struct dt_object_format *dof,
2269 /* we drop umask so that permissions we pass are not affected */
2270 umask = current->fs->umask;
2271 current->fs->umask = 0;
2273 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2276 osd_attr_init(info, obj, attr, dof);
2277 osd_object_init0(obj);
2279 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2280 unlock_new_inode(obj->oo_inode);
2283 /* restore previous umask value */
2284 current->fs->umask = umask;
2290 * Helper function for osd_object_create()
2292 * \retval 0, on success
2294 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2295 const struct lu_fid *fid, struct thandle *th)
2297 struct osd_thread_info *info = osd_oti_get(env);
2298 struct osd_inode_id *id = &info->oti_id;
2299 struct osd_device *osd = osd_obj2dev(obj);
2300 struct osd_thandle *oh;
2302 LASSERT(obj->oo_inode != NULL);
2304 oh = container_of0(th, struct osd_thandle, ot_super);
2305 LASSERT(oh->ot_handle);
2307 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2308 return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2311 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2312 obd_seq seq, struct lu_seq_range *range)
2314 struct seq_server_site *ss = osd_seq_site(osd);
2316 if (fid_seq_is_idif(seq)) {
2317 fld_range_set_ost(range);
2318 range->lsr_index = idif_ost_idx(seq);
2322 if (!fid_seq_in_fldb(seq)) {
2323 fld_range_set_mdt(range);
2325 /* FIXME: If ss is NULL, it suppose not get lsr_index
2327 range->lsr_index = ss->ss_node_id;
2331 LASSERT(ss != NULL);
2332 fld_range_set_any(range);
2333 /* OSD will only do local fld lookup */
2334 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2338 * Concurrency: no external locking is necessary.
2340 static int osd_declare_object_create(const struct lu_env *env,
2341 struct dt_object *dt,
2342 struct lu_attr *attr,
2343 struct dt_allocation_hint *hint,
2344 struct dt_object_format *dof,
2345 struct thandle *handle)
2347 struct osd_thandle *oh;
2351 LASSERT(handle != NULL);
2353 oh = container_of0(handle, struct osd_thandle, ot_super);
2354 LASSERT(oh->ot_handle == NULL);
2356 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2357 osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2358 /* Reuse idle OI block may cause additional one OI block
2360 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2361 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2363 /* If this is directory, then we expect . and .. to be inserted as
2364 * well. The one directory block always needs to be created for the
2365 * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2366 * block), there is no danger of needing a tree for the first block.
2368 if (attr && S_ISDIR(attr->la_mode)) {
2369 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2370 osd_dto_credits_noquota[DTO_WRITE_BASE]);
2371 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2377 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2378 false, false, NULL, false);
2385 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2386 struct lu_attr *attr,
2387 struct dt_allocation_hint *hint,
2388 struct dt_object_format *dof,
2391 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2392 struct osd_object *obj = osd_dt_obj(dt);
2393 struct osd_thread_info *info = osd_oti_get(env);
2398 LINVRNT(osd_invariant(obj));
2399 LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2400 LASSERT(osd_write_locked(env, obj));
2401 LASSERT(th != NULL);
2403 if (unlikely(fid_is_acct(fid)))
2404 /* Quota files can't be created from the kernel any more,
2405 * 'tune2fs -O quota' will take care of creating them */
2408 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2409 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2411 result = __osd_object_create(info, obj, attr, hint, dof, th);
2413 result = __osd_oi_insert(env, obj, fid, th);
2415 LASSERT(ergo(result == 0,
2416 dt_object_exists(dt) && !dt_object_remote(dt)));
2418 LASSERT(osd_invariant(obj));
2423 * Called to destroy on-disk representation of the object
2425 * Concurrency: must be locked
2427 static int osd_declare_object_destroy(const struct lu_env *env,
2428 struct dt_object *dt,
2431 struct osd_object *obj = osd_dt_obj(dt);
2432 struct inode *inode = obj->oo_inode;
2433 struct osd_thandle *oh;
2437 oh = container_of0(th, struct osd_thandle, ot_super);
2438 LASSERT(oh->ot_handle == NULL);
2441 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2442 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2443 /* Recycle idle OI leaf may cause additional three OI blocks
2445 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2446 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2447 /* one less inode */
2448 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2449 -1, oh, false, true, NULL, false);
2452 /* data to be truncated */
2453 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2454 0, oh, true, true, NULL, false);
2458 static int osd_object_destroy(const struct lu_env *env,
2459 struct dt_object *dt,
2462 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2463 struct osd_object *obj = osd_dt_obj(dt);
2464 struct inode *inode = obj->oo_inode;
2465 struct osd_device *osd = osd_obj2dev(obj);
2466 struct osd_thandle *oh;
2470 oh = container_of0(th, struct osd_thandle, ot_super);
2471 LASSERT(oh->ot_handle);
2473 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2475 if (unlikely(fid_is_acct(fid)))
2478 if (S_ISDIR(inode->i_mode)) {
2479 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2480 inode->i_nlink == 2);
2481 /* it will check/delete the inode from remote parent,
2482 * how to optimize it? unlink performance impaction XXX */
2483 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2484 if (result != 0 && result != -ENOENT) {
2485 CERROR("%s: delete inode "DFID": rc = %d\n",
2486 osd_name(osd), PFID(fid), result);
2488 spin_lock(&obj->oo_guard);
2490 spin_unlock(&obj->oo_guard);
2491 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2494 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2496 result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2499 /* XXX: add to ext3 orphan list */
2500 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2502 /* not needed in the cache anymore */
2503 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2509 * Put the fid into lustre_mdt_attrs, and then place the structure
2510 * inode's ea. This fid should not be altered during the life time
2513 * \retval +ve, on success
2514 * \retval -ve, on error
2516 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2518 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2519 const struct lu_fid *fid, __u32 compat, __u32 incompat)
2521 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2525 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2528 lustre_lma_init(lma, fid, compat, incompat);
2529 lustre_lma_swab(lma);
2531 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2533 /* LMA may already exist, but we need to check that all the
2534 * desired compat/incompat flags have been added. */
2535 if (unlikely(rc == -EEXIST)) {
2536 if (compat == 0 && incompat == 0)
2539 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2540 XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2545 lustre_lma_swab(lma);
2546 if (!(~lma->lma_compat & compat) &&
2547 !(~lma->lma_incompat & incompat))
2550 lma->lma_compat |= compat;
2551 lma->lma_incompat |= incompat;
2552 lustre_lma_swab(lma);
2553 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2554 sizeof(*lma), XATTR_REPLACE);
2561 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2562 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2563 * To have compatilibility with 1.8 ldiskfs driver we need to have
2564 * magic number at start of fid data.
2565 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2568 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2569 const struct dt_rec *fid)
2571 if (!fid_is_namespace_visible((const struct lu_fid *)fid) ||
2572 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2573 param->edp_magic = 0;
2577 param->edp_magic = LDISKFS_LUFID_MAGIC;
2578 param->edp_len = sizeof(struct lu_fid) + 1;
2579 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2583 * Try to read the fid from inode ea into dt_rec.
2585 * \param fid object fid.
2587 * \retval 0 on success
2589 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2590 __u32 ino, struct lu_fid *fid,
2591 struct osd_inode_id *id)
2593 struct osd_thread_info *info = osd_oti_get(env);
2594 struct inode *inode;
2597 osd_id_gen(id, ino, OSD_OII_NOGEN);
2598 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2600 RETURN(PTR_ERR(inode));
2606 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2608 struct inode *parent_dir,
2609 const struct dt_rec *dot_fid,
2610 const struct dt_rec *dot_dot_fid,
2611 struct osd_thandle *oth)
2613 struct ldiskfs_dentry_param *dot_ldp;
2614 struct ldiskfs_dentry_param *dot_dot_ldp;
2616 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2617 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2619 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2620 dot_ldp->edp_magic = 0;
2621 return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2622 dir, dot_ldp, dot_dot_ldp);
2626 * Create an local agent inode for remote entry
2628 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2629 struct osd_device *osd,
2630 struct osd_object *pobj,
2631 const struct lu_fid *fid,
2634 struct osd_thread_info *info = osd_oti_get(env);
2635 struct inode *local;
2636 struct osd_thandle *oh;
2641 oh = container_of(th, struct osd_thandle, ot_super);
2642 LASSERT(oh->ot_handle->h_transaction != NULL);
2644 /* FIXME: Insert index api needs to know the mode of
2645 * the remote object. Just use S_IFDIR for now */
2646 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR);
2647 if (IS_ERR(local)) {
2648 CERROR("%s: create local error %d\n", osd_name(osd),
2649 (int)PTR_ERR(local));
2653 /* Set special LMA flag for local agent inode */
2654 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2656 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2657 osd_name(osd), PFID(fid), rc);
2658 RETURN(ERR_PTR(rc));
2661 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2662 (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
2663 (const struct dt_rec *)fid, oh);
2665 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2666 osd_name(osd), PFID(fid), rc);
2667 RETURN(ERR_PTR(rc));
2674 * Delete local agent inode for remote entry
2676 static int osd_delete_local_agent_inode(const struct lu_env *env,
2677 struct osd_device *osd,
2678 const struct lu_fid *fid,
2679 __u32 ino, struct osd_thandle *oh)
2681 struct osd_thread_info *oti = osd_oti_get(env);
2682 struct osd_inode_id *id = &oti->oti_id;
2683 struct inode *inode;
2686 id->oii_ino = le32_to_cpu(ino);
2687 id->oii_gen = OSD_OII_NOGEN;
2688 inode = osd_iget(oti, osd, id);
2689 if (IS_ERR(inode)) {
2690 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2691 PFID(fid), id->oii_ino, id->oii_gen);
2692 RETURN(PTR_ERR(inode));
2696 mark_inode_dirty(inode);
2697 CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2698 osd_name(osd), PFID(fid), inode->i_ino);
2704 * OSD layer object create function for interoperability mode (b11826).
2705 * This is mostly similar to osd_object_create(). Only difference being, fid is
2706 * inserted into inode ea here.
2708 * \retval 0, on success
2709 * \retval -ve, on error
2711 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2712 struct lu_attr *attr,
2713 struct dt_allocation_hint *hint,
2714 struct dt_object_format *dof,
2717 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2718 struct osd_object *obj = osd_dt_obj(dt);
2719 struct osd_thread_info *info = osd_oti_get(env);
2724 LASSERT(osd_invariant(obj));
2725 LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2726 LASSERT(osd_write_locked(env, obj));
2727 LASSERT(th != NULL);
2729 if (unlikely(fid_is_acct(fid)))
2730 /* Quota files can't be created from the kernel any more,
2731 * 'tune2fs -O quota' will take care of creating them */
2734 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2735 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2737 result = __osd_object_create(info, obj, attr, hint, dof, th);
2739 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2740 fid_is_on_ost(info, osd_obj2dev(obj),
2741 fid, OI_CHECK_FLD) ?
2742 LMAC_FID_ON_OST : 0, 0);
2745 result = __osd_oi_insert(env, obj, fid, th);
2747 LASSERT(ergo(result == 0,
2748 dt_object_exists(dt) && !dt_object_remote(dt)));
2749 LINVRNT(osd_invariant(obj));
2753 static int osd_declare_object_ref_add(const struct lu_env *env,
2754 struct dt_object *dt,
2755 struct thandle *handle)
2757 struct osd_thandle *oh;
2759 /* it's possible that object doesn't exist yet */
2760 LASSERT(handle != NULL);
2762 oh = container_of0(handle, struct osd_thandle, ot_super);
2763 LASSERT(oh->ot_handle == NULL);
2765 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2766 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2772 * Concurrency: @dt is write locked.
2774 static int osd_object_ref_add(const struct lu_env *env,
2775 struct dt_object *dt, struct thandle *th)
2777 struct osd_object *obj = osd_dt_obj(dt);
2778 struct inode *inode = obj->oo_inode;
2779 struct osd_thandle *oh;
2782 LINVRNT(osd_invariant(obj));
2783 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2784 LASSERT(osd_write_locked(env, obj));
2785 LASSERT(th != NULL);
2787 oh = container_of0(th, struct osd_thandle, ot_super);
2788 LASSERT(oh->ot_handle != NULL);
2790 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2792 CDEBUG(D_INODE, DFID" increase nlink %d\n",
2793 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2795 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2796 * (65000) subdirectories by storing "1" in i_nlink if the link count
2797 * would otherwise overflow. Directory tranversal tools understand
2798 * that (st_nlink == 1) indicates that the filesystem dose not track
2799 * hard links count on the directory, and will not abort subdirectory
2800 * scanning early once (st_nlink - 2) subdirs have been found.
2802 * This also has to properly handle the case of inodes with nlink == 0
2803 * in case they are being linked into the PENDING directory
2805 spin_lock(&obj->oo_guard);
2806 ldiskfs_inc_count(oh->ot_handle, inode);
2807 if (!S_ISDIR(inode->i_mode))
2808 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2809 spin_unlock(&obj->oo_guard);
2811 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2812 LINVRNT(osd_invariant(obj));
2817 static int osd_declare_object_ref_del(const struct lu_env *env,
2818 struct dt_object *dt,
2819 struct thandle *handle)
2821 struct osd_thandle *oh;
2823 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2824 LASSERT(handle != NULL);
2826 oh = container_of0(handle, struct osd_thandle, ot_super);
2827 LASSERT(oh->ot_handle == NULL);
2829 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2830 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2836 * Concurrency: @dt is write locked.
2838 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2841 struct osd_object *obj = osd_dt_obj(dt);
2842 struct inode *inode = obj->oo_inode;
2843 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
2844 struct osd_thandle *oh;
2846 LINVRNT(osd_invariant(obj));
2847 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2848 LASSERT(osd_write_locked(env, obj));
2849 LASSERT(th != NULL);
2851 oh = container_of0(th, struct osd_thandle, ot_super);
2852 LASSERT(oh->ot_handle != NULL);
2854 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2856 spin_lock(&obj->oo_guard);
2857 /* That can be result of upgrade from old Lustre version and
2858 * applied only to local files. Just skip this ref_del call.
2859 * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2860 if (inode->i_nlink == 0) {
2861 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2862 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2863 ", maybe an upgraded file? (LU-3915)\n",
2864 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2865 spin_unlock(&obj->oo_guard);
2869 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2870 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2872 ldiskfs_dec_count(oh->ot_handle, inode);
2873 spin_unlock(&obj->oo_guard);
2875 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2876 LINVRNT(osd_invariant(obj));
2882 * Get the 64-bit version for an inode.
2884 static int osd_object_version_get(const struct lu_env *env,
2885 struct dt_object *dt, dt_obj_version_t *ver)
2887 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2889 CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2890 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2891 *ver = LDISKFS_I(inode)->i_fs_version;
2896 * Concurrency: @dt is read locked.
2898 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2899 struct lu_buf *buf, const char *name,
2900 struct lustre_capa *capa)
2902 struct osd_object *obj = osd_dt_obj(dt);
2903 struct inode *inode = obj->oo_inode;
2904 struct osd_thread_info *info = osd_oti_get(env);
2905 struct dentry *dentry = &info->oti_obj_dentry;
2907 /* version get is not real XATTR but uses xattr API */
2908 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2909 /* for version we are just using xattr API but change inode
2911 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2912 osd_object_version_get(env, dt, buf->lb_buf);
2913 return sizeof(dt_obj_version_t);
2916 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2917 LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2919 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2922 return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2926 static int osd_declare_xattr_set(const struct lu_env *env,
2927 struct dt_object *dt,
2928 const struct lu_buf *buf, const char *name,
2929 int fl, struct thandle *handle)
2931 struct osd_thandle *oh;
2934 LASSERT(handle != NULL);
2936 oh = container_of0(handle, struct osd_thandle, ot_super);
2937 LASSERT(oh->ot_handle == NULL);
2939 /* optimistic optimization: LMA is set first and usually fit inode */
2940 if (strcmp(name, XATTR_NAME_LMA) == 0) {
2941 if (dt_object_exists(dt))
2945 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2948 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
2949 struct super_block *sb = osd_sb(osd);
2950 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2951 if (buf && buf->lb_len > sb->s_blocksize) {
2952 credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2953 sb->s_blocksize_bits;
2957 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2963 * Set the 64-bit version for object
2965 static void osd_object_version_set(const struct lu_env *env,
2966 struct dt_object *dt,
2967 dt_obj_version_t *new_version)
2969 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2971 CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2972 *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2974 LDISKFS_I(inode)->i_fs_version = *new_version;
2975 /** Version is set after all inode operations are finished,
2976 * so we should mark it dirty here */
2977 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2981 * Concurrency: @dt is write locked.
2983 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2984 const struct lu_buf *buf, const char *name, int fl,
2985 struct thandle *handle, struct lustre_capa *capa)
2987 struct osd_object *obj = osd_dt_obj(dt);
2988 struct inode *inode = obj->oo_inode;
2989 struct osd_thread_info *info = osd_oti_get(env);
2993 LASSERT(handle != NULL);
2995 /* version set is not real XATTR */
2996 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2997 /* for version we are just using xattr API but change inode
2999 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3000 osd_object_version_set(env, dt, buf->lb_buf);
3001 return sizeof(dt_obj_version_t);
3004 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3007 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n",
3008 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3010 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3011 if (fl & LU_XATTR_REPLACE)
3012 fs_flags |= XATTR_REPLACE;
3014 if (fl & LU_XATTR_CREATE)
3015 fs_flags |= XATTR_CREATE;
3017 if (strcmp(name, XATTR_NAME_LMV) == 0) {
3018 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3021 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3025 lma->lma_incompat |= LMAI_STRIPED;
3026 lustre_lma_swab(lma);
3027 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3028 sizeof(*lma), XATTR_REPLACE);
3033 return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3038 * Concurrency: @dt is read locked.
3040 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3041 struct lu_buf *buf, struct lustre_capa *capa)
3043 struct osd_object *obj = osd_dt_obj(dt);
3044 struct inode *inode = obj->oo_inode;
3045 struct osd_thread_info *info = osd_oti_get(env);
3046 struct dentry *dentry = &info->oti_obj_dentry;
3048 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3049 LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
3051 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3054 dentry->d_inode = inode;
3055 dentry->d_sb = inode->i_sb;
3056 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3059 static int osd_declare_xattr_del(const struct lu_env *env,
3060 struct dt_object *dt, const char *name,
3061 struct thandle *handle)
3063 struct osd_thandle *oh;
3065 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3066 LASSERT(handle != NULL);
3068 oh = container_of0(handle, struct osd_thandle, ot_super);
3069 LASSERT(oh->ot_handle == NULL);
3071 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3072 osd_dto_credits_noquota[DTO_XATTR_SET]);
3078 * Concurrency: @dt is write locked.
3080 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3081 const char *name, struct thandle *handle,
3082 struct lustre_capa *capa)
3084 struct osd_object *obj = osd_dt_obj(dt);
3085 struct inode *inode = obj->oo_inode;
3086 struct osd_thread_info *info = osd_oti_get(env);
3087 struct dentry *dentry = &info->oti_obj_dentry;
3090 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3091 LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
3092 LASSERT(handle != NULL);
3094 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3097 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3099 ll_vfs_dq_init(inode);
3100 dentry->d_inode = inode;
3101 dentry->d_sb = inode->i_sb;
3102 rc = inode->i_op->removexattr(dentry, name);
3106 static struct obd_capa *osd_capa_get(const struct lu_env *env,
3107 struct dt_object *dt,
3108 struct lustre_capa *old, __u64 opc)
3110 struct osd_thread_info *info = osd_oti_get(env);
3111 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3112 struct osd_object *obj = osd_dt_obj(dt);
3113 struct osd_device *osd = osd_obj2dev(obj);
3114 struct lustre_capa_key *key = &info->oti_capa_key;
3115 struct lustre_capa *capa = &info->oti_capa;
3116 struct obd_capa *oc;
3117 struct lu_capainfo *lci;
3121 if (!osd->od_fl_capa)
3122 RETURN(ERR_PTR(-ENOENT));
3124 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3125 LINVRNT(osd_invariant(obj));
3127 /* renewal sanity check */
3128 if (old && osd_object_auth(env, dt, old, opc))
3129 RETURN(ERR_PTR(-EACCES));
3131 lci = lu_capainfo_get(env);
3132 if (unlikely(lci == NULL))
3133 RETURN(ERR_PTR(-ENOENT));
3135 switch (lci->lci_auth) {
3139 capa->lc_uid = i_uid_read(obj->oo_inode);
3140 capa->lc_gid = i_gid_read(obj->oo_inode);
3141 capa->lc_flags = LC_ID_PLAIN;
3143 case LC_ID_CONVERT: {
3146 s[0] = i_uid_read(obj->oo_inode);
3147 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
3148 s[2] = i_uid_read(obj->oo_inode);
3149 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
3150 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
3152 RETURN(ERR_PTR(rc));
3154 capa->lc_uid = ((__u64)d[1] << 32) | d[0];
3155 capa->lc_gid = ((__u64)d[3] << 32) | d[2];
3156 capa->lc_flags = LC_ID_CONVERT;
3160 RETURN(ERR_PTR(-EINVAL));
3163 capa->lc_fid = *fid;
3165 capa->lc_flags |= osd->od_capa_alg << 24;
3166 capa->lc_timeout = osd->od_capa_timeout;
3167 capa->lc_expiry = 0;
3169 oc = capa_lookup(osd->od_capa_hash, capa, 1);
3171 LASSERT(!capa_is_expired(oc));
3175 spin_lock(&capa_lock);
3176 *key = osd->od_capa_keys[1];
3177 spin_unlock(&capa_lock);
3179 capa->lc_keyid = key->lk_keyid;
3180 capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout;
3182 rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
3184 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
3185 RETURN(ERR_PTR(rc));
3188 oc = capa_add(osd->od_capa_hash, capa);
3192 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3193 __u64 start, __u64 end)
3195 struct osd_object *obj = osd_dt_obj(dt);
3196 struct inode *inode = obj->oo_inode;
3197 struct osd_thread_info *info = osd_oti_get(env);
3198 struct dentry *dentry = &info->oti_obj_dentry;
3199 struct file *file = &info->oti_file;
3204 dentry->d_inode = inode;
3205 dentry->d_sb = inode->i_sb;
3206 file->f_dentry = dentry;
3207 file->f_mapping = inode->i_mapping;
3208 file->f_op = inode->i_fop;
3209 set_file_inode(file, inode);
3211 #ifdef HAVE_FILE_FSYNC_4ARGS
3212 rc = file->f_op->fsync(file, start, end, 0);
3213 #elif defined(HAVE_FILE_FSYNC_2ARGS)
3214 mutex_lock(&inode->i_mutex);
3215 rc = file->f_op->fsync(file, 0);
3216 mutex_unlock(&inode->i_mutex);
3218 mutex_lock(&inode->i_mutex);
3219 rc = file->f_op->fsync(file, dentry, 0);
3220 mutex_unlock(&inode->i_mutex);
3226 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
3229 struct osd_object *obj = osd_dt_obj(dt);
3232 *data = (void *)obj->oo_inode;
3240 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3241 const struct dt_index_features *feat)
3243 struct iam_descr *descr;
3245 if (osd_object_is_root(o))
3246 return feat == &dt_directory_features;
3248 LASSERT(o->oo_dir != NULL);
3250 descr = o->oo_dir->od_container.ic_descr;
3251 if (feat == &dt_directory_features) {
3252 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3258 feat->dif_keysize_min <= descr->id_key_size &&
3259 descr->id_key_size <= feat->dif_keysize_max &&
3260 feat->dif_recsize_min <= descr->id_rec_size &&
3261 descr->id_rec_size <= feat->dif_recsize_max &&
3262 !(feat->dif_flags & (DT_IND_VARKEY |
3263 DT_IND_VARREC | DT_IND_NONUNQ)) &&
3264 ergo(feat->dif_flags & DT_IND_UPDATE,
3265 1 /* XXX check that object (and file system) is
3270 static int osd_iam_container_init(const struct lu_env *env,
3271 struct osd_object *obj,
3272 struct osd_directory *dir)
3274 struct iam_container *bag = &dir->od_container;
3277 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3281 result = iam_container_setup(bag);
3283 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3285 iam_container_fini(bag);
3292 * Concurrency: no external locking is necessary.
3294 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3295 const struct dt_index_features *feat)
3299 struct osd_object *obj = osd_dt_obj(dt);
3301 LINVRNT(osd_invariant(obj));
3303 if (osd_object_is_root(obj)) {
3304 dt->do_index_ops = &osd_index_ea_ops;
3306 } else if (feat == &dt_directory_features) {
3307 dt->do_index_ops = &osd_index_ea_ops;
3308 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3313 } else if (unlikely(feat == &dt_otable_features)) {
3314 dt->do_index_ops = &osd_otable_ops;
3316 } else if (unlikely(feat == &dt_acct_features)) {
3317 dt->do_index_ops = &osd_acct_index_ops;
3320 } else if (!osd_has_index(obj)) {
3321 struct osd_directory *dir;
3326 spin_lock(&obj->oo_guard);
3327 if (obj->oo_dir == NULL)
3331 * Concurrent thread allocated container data.
3334 spin_unlock(&obj->oo_guard);
3336 * Now, that we have container data, serialize its
3339 down_write(&obj->oo_ext_idx_sem);
3341 * recheck under lock.
3343 if (!osd_has_index(obj))
3344 result = osd_iam_container_init(env, obj,
3348 up_write(&obj->oo_ext_idx_sem);
3356 if (result == 0 && skip_iam == 0) {
3357 if (!osd_iam_index_probe(env, obj, feat))
3360 LINVRNT(osd_invariant(obj));
3362 if (result == 0 && is_quota_glb_feat(feat) &&
3363 fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3364 result = osd_quota_migration(env, dt, feat);
3369 static int osd_otable_it_attr_get(const struct lu_env *env,
3370 struct dt_object *dt,
3371 struct lu_attr *attr,
3372 struct lustre_capa *capa)
3378 static const struct dt_object_operations osd_obj_ops = {
3379 .do_read_lock = osd_object_read_lock,
3380 .do_write_lock = osd_object_write_lock,
3381 .do_read_unlock = osd_object_read_unlock,
3382 .do_write_unlock = osd_object_write_unlock,
3383 .do_write_locked = osd_object_write_locked,
3384 .do_attr_get = osd_attr_get,
3385 .do_declare_attr_set = osd_declare_attr_set,
3386 .do_attr_set = osd_attr_set,
3387 .do_ah_init = osd_ah_init,
3388 .do_declare_create = osd_declare_object_create,
3389 .do_create = osd_object_create,
3390 .do_declare_destroy = osd_declare_object_destroy,
3391 .do_destroy = osd_object_destroy,
3392 .do_index_try = osd_index_try,
3393 .do_declare_ref_add = osd_declare_object_ref_add,
3394 .do_ref_add = osd_object_ref_add,
3395 .do_declare_ref_del = osd_declare_object_ref_del,
3396 .do_ref_del = osd_object_ref_del,
3397 .do_xattr_get = osd_xattr_get,
3398 .do_declare_xattr_set = osd_declare_xattr_set,
3399 .do_xattr_set = osd_xattr_set,
3400 .do_declare_xattr_del = osd_declare_xattr_del,
3401 .do_xattr_del = osd_xattr_del,
3402 .do_xattr_list = osd_xattr_list,
3403 .do_capa_get = osd_capa_get,
3404 .do_object_sync = osd_object_sync,
3405 .do_data_get = osd_data_get,
3409 * dt_object_operations for interoperability mode
3410 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3412 static const struct dt_object_operations osd_obj_ea_ops = {
3413 .do_read_lock = osd_object_read_lock,
3414 .do_write_lock = osd_object_write_lock,
3415 .do_read_unlock = osd_object_read_unlock,
3416 .do_write_unlock = osd_object_write_unlock,
3417 .do_write_locked = osd_object_write_locked,
3418 .do_attr_get = osd_attr_get,
3419 .do_declare_attr_set = osd_declare_attr_set,
3420 .do_attr_set = osd_attr_set,
3421 .do_ah_init = osd_ah_init,
3422 .do_declare_create = osd_declare_object_create,
3423 .do_create = osd_object_ea_create,
3424 .do_declare_destroy = osd_declare_object_destroy,
3425 .do_destroy = osd_object_destroy,
3426 .do_index_try = osd_index_try,
3427 .do_declare_ref_add = osd_declare_object_ref_add,
3428 .do_ref_add = osd_object_ref_add,
3429 .do_declare_ref_del = osd_declare_object_ref_del,
3430 .do_ref_del = osd_object_ref_del,
3431 .do_xattr_get = osd_xattr_get,
3432 .do_declare_xattr_set = osd_declare_xattr_set,
3433 .do_xattr_set = osd_xattr_set,
3434 .do_declare_xattr_del = osd_declare_xattr_del,
3435 .do_xattr_del = osd_xattr_del,
3436 .do_xattr_list = osd_xattr_list,
3437 .do_capa_get = osd_capa_get,
3438 .do_object_sync = osd_object_sync,
3439 .do_data_get = osd_data_get,
3442 static const struct dt_object_operations osd_obj_otable_it_ops = {
3443 .do_attr_get = osd_otable_it_attr_get,
3444 .do_index_try = osd_index_try,
3447 static int osd_index_declare_iam_delete(const struct lu_env *env,
3448 struct dt_object *dt,
3449 const struct dt_key *key,
3450 struct thandle *handle)
3452 struct osd_thandle *oh;
3454 oh = container_of0(handle, struct osd_thandle, ot_super);
3455 LASSERT(oh->ot_handle == NULL);
3457 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3458 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3464 * delete a (key, value) pair from index \a dt specified by \a key
3466 * \param dt osd index object
3467 * \param key key for index
3468 * \param rec record reference
3469 * \param handle transaction handler
3472 * \retval -ve failure
3475 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3476 const struct dt_key *key,
3477 struct thandle *handle,
3478 struct lustre_capa *capa)
3480 struct osd_thread_info *oti = osd_oti_get(env);
3481 struct osd_object *obj = osd_dt_obj(dt);
3482 struct osd_thandle *oh;
3483 struct iam_path_descr *ipd;
3484 struct iam_container *bag = &obj->oo_dir->od_container;
3489 LINVRNT(osd_invariant(obj));
3490 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3491 LASSERT(bag->ic_object == obj->oo_inode);
3492 LASSERT(handle != NULL);
3494 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3497 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3499 ipd = osd_idx_ipd_get(env, bag);
3500 if (unlikely(ipd == NULL))
3503 oh = container_of0(handle, struct osd_thandle, ot_super);
3504 LASSERT(oh->ot_handle != NULL);
3505 LASSERT(oh->ot_handle->h_transaction != NULL);
3507 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3508 /* swab quota uid/gid provided by caller */
3509 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3510 key = (const struct dt_key *)&oti->oti_quota_id;
3513 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3514 osd_ipd_put(env, bag, ipd);
3515 LINVRNT(osd_invariant(obj));
3519 static int osd_index_declare_ea_delete(const struct lu_env *env,
3520 struct dt_object *dt,
3521 const struct dt_key *key,
3522 struct thandle *handle)
3524 struct osd_thandle *oh;
3525 struct inode *inode;
3529 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3530 LASSERT(handle != NULL);
3532 oh = container_of0(handle, struct osd_thandle, ot_super);
3533 LASSERT(oh->ot_handle == NULL);
3535 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3536 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3538 inode = osd_dt_obj(dt)->oo_inode;
3541 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3542 0, oh, true, true, NULL, false);
3546 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3549 struct osd_fid_pack *rec;
3552 if (de->file_type & LDISKFS_DIRENT_LUFID) {
3553 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3554 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3559 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3564 /* FID seqs not in FLDB, must be local seq */
3565 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3568 if (osd_seq_exists(env, osd, fid_seq(fid)))
3575 * Index delete function for interoperability mode (b11826).
3576 * It will remove the directory entry added by osd_index_ea_insert().
3577 * This entry is needed to maintain name->fid mapping.
3579 * \param key, key i.e. file entry to be deleted
3581 * \retval 0, on success
3582 * \retval -ve, on error
3584 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3585 const struct dt_key *key,
3586 struct thandle *handle,
3587 struct lustre_capa *capa)
3589 struct osd_object *obj = osd_dt_obj(dt);
3590 struct inode *dir = obj->oo_inode;
3591 struct dentry *dentry;
3592 struct osd_thandle *oh;
3593 struct ldiskfs_dir_entry_2 *de = NULL;
3594 struct buffer_head *bh;
3595 struct htree_lock *hlock = NULL;
3596 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
3597 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3601 LINVRNT(osd_invariant(obj));
3602 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3603 LASSERT(handle != NULL);
3605 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3607 oh = container_of(handle, struct osd_thandle, ot_super);
3608 LASSERT(oh->ot_handle != NULL);
3609 LASSERT(oh->ot_handle->h_transaction != NULL);
3611 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3614 ll_vfs_dq_init(dir);
3615 dentry = osd_child_dentry_get(env, obj,
3616 (char *)key, strlen((char *)key));
3618 if (obj->oo_hl_head != NULL) {
3619 hlock = osd_oti_get(env)->oti_hlock;
3620 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3621 dir, LDISKFS_HLOCK_DEL);
3623 down_write(&obj->oo_ext_idx_sem);
3626 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3630 /* If this is not the ".." entry, it might be a remote DNE
3631 * entry and we need to check if the FID is for a remote
3632 * MDT. If the FID is not in the directory entry (e.g.
3633 * upgraded 1.8 filesystem without dirdata enabled) then
3634 * we need to get the FID from the LMA. For a remote directory
3635 * there HAS to be an LMA, it cannot be an IGIF inode in this
3638 * Delete the entry before the agent inode in order to
3639 * simplify error handling. At worst an error after deleting
3640 * the entry first might leak the agent inode afterward. The
3641 * reverse would need filesystem abort in case of error deleting
3642 * the entry after the agent had been removed, or leave a
3643 * dangling entry pointing at a random inode. */
3644 if (strcmp((char *)key, dotdot) != 0) {
3645 LASSERT(de != NULL);
3646 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3647 /* If Fid is not in dentry, try to get it from LMA */
3648 if (rc == -ENODATA) {
3649 struct osd_inode_id *id;
3650 struct inode *inode;
3652 /* Before trying to get fid from the inode,
3653 * check whether the inode is valid.
3655 * If the inode has been deleted, do not go
3656 * ahead to do osd_ea_fid_get, which will set
3657 * the inode to bad inode, which might cause
3658 * the inode to be deleted uncorrectly */
3659 inode = ldiskfs_iget(osd_sb(osd),
3660 le32_to_cpu(de->inode));
3661 if (IS_ERR(inode)) {
3662 CDEBUG(D_INODE, "%s: "DFID"get inode"
3663 "error.\n", osd_name(osd),
3665 rc = PTR_ERR(inode);
3667 if (likely(inode->i_nlink != 0)) {
3668 id = &osd_oti_get(env)->oti_id;
3669 rc = osd_ea_fid_get(env, obj,
3670 le32_to_cpu(de->inode),
3673 CDEBUG(D_INFO, "%s: %u "DFID
3676 le32_to_cpu(de->inode),
3684 unlikely(osd_remote_fid(env, osd, fid)))
3685 /* Need to delete agent inode */
3686 ino = le32_to_cpu(de->inode);
3688 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3690 if (rc == 0 && unlikely(ino != 0)) {
3691 rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3694 CERROR("%s: del local inode "DFID": rc = %d\n",
3695 osd_name(osd), PFID(fid), rc);
3701 ldiskfs_htree_unlock(hlock);
3703 up_write(&obj->oo_ext_idx_sem);
3708 /* For inode on the remote MDT, .. will point to
3709 * /Agent directory, Check whether it needs to delete
3710 * from agent directory */
3711 if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3712 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3714 if (rc != 0 && rc != -ENOENT) {
3715 CERROR("%s: delete agent inode "DFID": rc = %d\n",
3716 osd_name(osd), PFID(fid), rc);
3726 LASSERT(osd_invariant(obj));
3731 * Lookup index for \a key and copy record to \a rec.
3733 * \param dt osd index object
3734 * \param key key for index
3735 * \param rec record reference
3737 * \retval +ve success : exact mach
3738 * \retval 0 return record with key not greater than \a key
3739 * \retval -ve failure
3741 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3742 struct dt_rec *rec, const struct dt_key *key,
3743 struct lustre_capa *capa)
3745 struct osd_object *obj = osd_dt_obj(dt);
3746 struct iam_path_descr *ipd;
3747 struct iam_container *bag = &obj->oo_dir->od_container;
3748 struct osd_thread_info *oti = osd_oti_get(env);
3749 struct iam_iterator *it = &oti->oti_idx_it;
3750 struct iam_rec *iam_rec;
3755 LASSERT(osd_invariant(obj));
3756 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3757 LASSERT(bag->ic_object == obj->oo_inode);
3759 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3762 ipd = osd_idx_ipd_get(env, bag);
3766 /* got ipd now we can start iterator. */
3767 iam_it_init(it, bag, 0, ipd);
3769 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3770 /* swab quota uid/gid provided by caller */
3771 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3772 key = (const struct dt_key *)&oti->oti_quota_id;
3775 rc = iam_it_get(it, (struct iam_key *)key);
3777 if (S_ISDIR(obj->oo_inode->i_mode))
3778 iam_rec = (struct iam_rec *)oti->oti_ldp;
3780 iam_rec = (struct iam_rec *) rec;
3782 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3784 if (S_ISDIR(obj->oo_inode->i_mode))
3785 osd_fid_unpack((struct lu_fid *) rec,
3786 (struct osd_fid_pack *)iam_rec);
3787 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3788 osd_quota_unpack(obj, rec);
3793 osd_ipd_put(env, bag, ipd);
3795 LINVRNT(osd_invariant(obj));
3800 static int osd_index_declare_iam_insert(const struct lu_env *env,
3801 struct dt_object *dt,
3802 const struct dt_rec *rec,
3803 const struct dt_key *key,
3804 struct thandle *handle)
3806 struct osd_thandle *oh;
3808 LASSERT(handle != NULL);
3810 oh = container_of0(handle, struct osd_thandle, ot_super);
3811 LASSERT(oh->ot_handle == NULL);
3813 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3814 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3820 * Inserts (key, value) pair in \a dt index object.
3822 * \param dt osd index object
3823 * \param key key for index
3824 * \param rec record reference
3825 * \param th transaction handler
3828 * \retval -ve failure
3830 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3831 const struct dt_rec *rec,
3832 const struct dt_key *key, struct thandle *th,
3833 struct lustre_capa *capa, int ignore_quota)
3835 struct osd_object *obj = osd_dt_obj(dt);
3836 struct iam_path_descr *ipd;
3837 struct osd_thandle *oh;
3838 struct iam_container *bag = &obj->oo_dir->od_container;
3839 struct osd_thread_info *oti = osd_oti_get(env);
3840 struct iam_rec *iam_rec;
3845 LINVRNT(osd_invariant(obj));
3846 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3847 LASSERT(bag->ic_object == obj->oo_inode);
3848 LASSERT(th != NULL);
3850 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3853 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3855 ipd = osd_idx_ipd_get(env, bag);
3856 if (unlikely(ipd == NULL))
3859 oh = container_of0(th, struct osd_thandle, ot_super);
3860 LASSERT(oh->ot_handle != NULL);
3861 LASSERT(oh->ot_handle->h_transaction != NULL);
3862 if (S_ISDIR(obj->oo_inode->i_mode)) {
3863 iam_rec = (struct iam_rec *)oti->oti_ldp;
3864 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3865 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3866 /* pack quota uid/gid */
3867 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3868 key = (const struct dt_key *)&oti->oti_quota_id;
3869 /* pack quota record */
3870 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3871 iam_rec = (struct iam_rec *)rec;
3873 iam_rec = (struct iam_rec *)rec;
3876 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3878 osd_ipd_put(env, bag, ipd);
3879 LINVRNT(osd_invariant(obj));
3884 * Calls ldiskfs_add_entry() to add directory entry
3885 * into the directory. This is required for
3886 * interoperability mode (b11826)
3888 * \retval 0, on success
3889 * \retval -ve, on error
3891 static int __osd_ea_add_rec(struct osd_thread_info *info,
3892 struct osd_object *pobj, struct inode *cinode,
3893 const char *name, const struct dt_rec *fid,
3894 struct htree_lock *hlock, struct thandle *th)
3896 struct ldiskfs_dentry_param *ldp;
3897 struct dentry *child;
3898 struct osd_thandle *oth;
3901 oth = container_of(th, struct osd_thandle, ot_super);
3902 LASSERT(oth->ot_handle != NULL);
3903 LASSERT(oth->ot_handle->h_transaction != NULL);
3904 LASSERT(pobj->oo_inode);
3906 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3907 if (unlikely(pobj->oo_inode ==
3908 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3911 osd_get_ldiskfs_dirent_param(ldp, fid);
3912 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3913 child->d_fsdata = (void *)ldp;
3914 ll_vfs_dq_init(pobj->oo_inode);
3915 rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3921 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3922 * into the directory.Also sets flags into osd object to
3923 * indicate dot and dotdot are created. This is required for
3924 * interoperability mode (b11826)
3926 * \param dir directory for dot and dotdot fixup.
3927 * \param obj child object for linking
3929 * \retval 0, on success
3930 * \retval -ve, on error
3932 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3933 struct osd_object *dir,
3934 struct inode *parent_dir, const char *name,
3935 const struct dt_rec *dot_fid,
3936 const struct dt_rec *dot_dot_fid,
3939 struct inode *inode = dir->oo_inode;
3940 struct osd_thandle *oth;
3943 oth = container_of(th, struct osd_thandle, ot_super);
3944 LASSERT(oth->ot_handle->h_transaction != NULL);
3945 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3947 if (strcmp(name, dot) == 0) {
3948 if (dir->oo_compat_dot_created) {
3951 LASSERT(inode == parent_dir);
3952 dir->oo_compat_dot_created = 1;
3955 } else if (strcmp(name, dotdot) == 0) {
3956 if (!dir->oo_compat_dot_created)
3958 /* in case of rename, dotdot is already created */
3959 if (dir->oo_compat_dotdot_created) {
3960 return __osd_ea_add_rec(info, dir, parent_dir, name,
3961 dot_dot_fid, NULL, th);
3964 result = osd_add_dot_dotdot_internal(info, dir->oo_inode,
3965 parent_dir, dot_fid,
3968 dir->oo_compat_dotdot_created = 1;
3976 * It will call the appropriate osd_add* function and return the
3977 * value, return by respective functions.
3979 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3980 struct inode *cinode, const char *name,
3981 const struct dt_rec *fid, struct thandle *th)
3983 struct osd_thread_info *info = osd_oti_get(env);
3984 struct htree_lock *hlock;
3987 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3989 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3991 if (hlock != NULL) {
3992 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3995 down_write(&pobj->oo_ext_idx_sem);
3997 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3998 (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
4001 if (hlock != NULL) {
4002 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4003 pobj->oo_inode, LDISKFS_HLOCK_ADD);
4005 down_write(&pobj->oo_ext_idx_sem);
4008 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4009 struct lu_fid *tfid = &info->oti_fid;
4011 *tfid = *(const struct lu_fid *)fid;
4013 rc = __osd_ea_add_rec(info, pobj, cinode, name,
4014 (const struct dt_rec *)tfid,
4017 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4022 ldiskfs_htree_unlock(hlock);
4024 up_write(&pobj->oo_ext_idx_sem);
4030 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4031 struct osd_idmap_cache *oic)
4033 struct osd_scrub *scrub = &dev->od_scrub;
4034 struct lu_fid *fid = &oic->oic_fid;
4035 struct osd_inode_id *id = &oti->oti_id;
4040 if (!fid_is_norm(fid) && !fid_is_igif(fid))
4044 rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
4045 if (rc != 0 && rc != -ENOENT)
4048 if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
4051 if (thread_is_running(&scrub->os_thread)) {
4052 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
4053 /* There is race condition between osd_oi_lookup and OI scrub.
4054 * The OI scrub finished just after osd_oi_lookup() failure.
4055 * Under such case, it is unnecessary to trigger OI scrub again,
4056 * but try to call osd_oi_lookup() again. */
4057 if (unlikely(rc == -EAGAIN))
4063 if (!dev->od_noscrub && ++once == 1) {
4064 rc = osd_scrub_start(dev);
4065 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC for "DFID
4067 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4076 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4077 struct osd_device *dev,
4078 struct osd_idmap_cache *oic,
4079 struct lu_fid *fid, __u32 ino)
4081 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
4082 struct inode *inode;
4085 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4086 inode = osd_iget(oti, dev, &oic->oic_lid);
4087 if (IS_ERR(inode)) {
4088 fid_zero(&oic->oic_fid);
4089 return PTR_ERR(inode);
4092 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4095 fid_zero(&oic->oic_fid);
4097 *fid = oic->oic_fid = lma->lma_self_fid;
4101 int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4102 struct osd_inode_id *id, const struct lu_fid *fid)
4104 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4105 id->oii_ino, id->oii_gen, info);
4106 info->oti_cache.oic_lid = *id;
4107 info->oti_cache.oic_fid = *fid;
4108 info->oti_cache.oic_dev = osd;
4114 * Calls ->lookup() to find dentry. From dentry get inode and
4115 * read inode's ea to get fid. This is required for interoperability
4118 * \retval 0, on success
4119 * \retval -ve, on error
4121 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4122 struct dt_rec *rec, const struct dt_key *key)
4124 struct inode *dir = obj->oo_inode;
4125 struct dentry *dentry;
4126 struct ldiskfs_dir_entry_2 *de;
4127 struct buffer_head *bh;
4128 struct lu_fid *fid = (struct lu_fid *) rec;
4129 struct htree_lock *hlock = NULL;
4134 LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
4136 dentry = osd_child_dentry_get(env, obj,
4137 (char *)key, strlen((char *)key));
4139 if (obj->oo_hl_head != NULL) {
4140 hlock = osd_oti_get(env)->oti_hlock;
4141 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4142 dir, LDISKFS_HLOCK_LOOKUP);
4144 down_read(&obj->oo_ext_idx_sem);
4147 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4149 struct osd_thread_info *oti = osd_oti_get(env);
4150 struct osd_inode_id *id = &oti->oti_id;
4151 struct osd_idmap_cache *oic = &oti->oti_cache;
4152 struct osd_device *dev = osd_obj2dev(obj);
4153 struct osd_scrub *scrub = &dev->od_scrub;
4154 struct scrub_file *sf = &scrub->os_file;
4156 ino = le32_to_cpu(de->inode);
4157 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4159 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4163 rc = osd_get_fid_from_dentry(de, rec);
4165 /* done with de, release bh */
4168 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4170 osd_id_gen(id, ino, OSD_OII_NOGEN);
4172 fid_zero(&oic->oic_fid);
4176 if (osd_remote_fid(env, dev, fid))
4179 rc = osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id,
4183 if ((scrub->os_pos_current <= ino) &&
4184 ((sf->sf_flags & SF_INCONSISTENT) ||
4185 (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
4186 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
4188 osd_consistency_check(oti, dev, oic);
4197 ldiskfs_htree_unlock(hlock);
4199 up_read(&obj->oo_ext_idx_sem);
4204 * Find the osd object for given fid.
4206 * \param fid need to find the osd object having this fid
4208 * \retval osd_object on success
4209 * \retval -ve on error
4211 struct osd_object *osd_object_find(const struct lu_env *env,
4212 struct dt_object *dt,
4213 const struct lu_fid *fid)
4215 struct lu_device *ludev = dt->do_lu.lo_dev;
4216 struct osd_object *child = NULL;
4217 struct lu_object *luch;
4218 struct lu_object *lo;
4221 * at this point topdev might not exist yet
4222 * (i.e. MGS is preparing profiles). so we can
4223 * not rely on topdev and instead lookup with
4224 * our device passed as topdev. this can't work
4225 * if the object isn't cached yet (as osd doesn't
4226 * allocate lu_header). IOW, the object must be
4227 * in the cache, otherwise lu_object_alloc() crashes
4230 luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4231 ludev : ludev->ld_site->ls_top_dev,
4233 if (!IS_ERR(luch)) {
4234 if (lu_object_exists(luch)) {
4235 lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4237 child = osd_obj(lo);
4239 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4240 "lu_object can't be located"
4241 DFID"\n", PFID(fid));
4243 if (child == NULL) {
4244 lu_object_put(env, luch);
4245 CERROR("Unable to get osd_object\n");
4246 child = ERR_PTR(-ENOENT);
4249 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4250 "lu_object does not exists "DFID"\n",
4252 lu_object_put(env, luch);
4253 child = ERR_PTR(-ENOENT);
4256 child = ERR_CAST(luch);
4263 * Put the osd object once done with it.
4265 * \param obj osd object that needs to be put
4267 static inline void osd_object_put(const struct lu_env *env,
4268 struct osd_object *obj)
4270 lu_object_put(env, &obj->oo_dt.do_lu);
4273 static int osd_index_declare_ea_insert(const struct lu_env *env,
4274 struct dt_object *dt,
4275 const struct dt_rec *rec,
4276 const struct dt_key *key,
4277 struct thandle *handle)
4279 struct osd_thandle *oh;
4280 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4281 struct lu_fid *fid = (struct lu_fid *)rec;
4285 LASSERT(!dt_object_remote(dt));
4286 LASSERT(handle != NULL);
4288 oh = container_of0(handle, struct osd_thandle, ot_super);
4289 LASSERT(oh->ot_handle == NULL);
4291 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4292 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4294 if (osd_dt_obj(dt)->oo_inode != NULL) {
4295 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4297 /* We ignore block quota on meta pool (MDTs), so needn't
4298 * calculate how many blocks will be consumed by this index
4300 rc = osd_declare_inode_qid(env, i_uid_read(inode),
4301 i_gid_read(inode), 0,
4302 oh, true, true, NULL, false);
4308 rc = osd_remote_fid(env, osd, fid);
4314 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
4315 osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
4316 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4317 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4318 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4319 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4325 * Index add function for interoperability mode (b11826).
4326 * It will add the directory entry.This entry is needed to
4327 * maintain name->fid mapping.
4329 * \param key it is key i.e. file entry to be inserted
4330 * \param rec it is value of given key i.e. fid
4332 * \retval 0, on success
4333 * \retval -ve, on error
4335 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4336 const struct dt_rec *rec,
4337 const struct dt_key *key, struct thandle *th,
4338 struct lustre_capa *capa, int ignore_quota)
4340 struct osd_object *obj = osd_dt_obj(dt);
4341 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4342 struct lu_fid *fid = (struct lu_fid *) rec;
4343 const char *name = (const char *)key;
4344 struct osd_thread_info *oti = osd_oti_get(env);
4345 struct osd_inode_id *id = &oti->oti_id;
4346 struct inode *child_inode = NULL;
4347 struct osd_object *child = NULL;
4351 LASSERT(osd_invariant(obj));
4352 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
4353 LASSERT(th != NULL);
4355 osd_trans_exec_op(env, th, OSD_OT_INSERT);
4357 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
4360 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!", PFID(fid));
4362 rc = osd_remote_fid(env, osd, fid);
4364 CERROR("%s: Can not find object "DFID" rc %d\n",
4365 osd_name(osd), PFID(fid), rc);
4370 /* Insert remote entry */
4371 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4372 struct osd_mdobj_map *omm = osd->od_mdt_map;
4373 struct osd_thandle *oh;
4375 /* If parent on remote MDT, we need put this object
4377 oh = container_of(th, typeof(*oh), ot_super);
4378 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4380 CERROR("%s: add "DFID" error: rc = %d\n",
4382 PFID(lu_object_fid(&dt->do_lu)), rc);
4386 child_inode = igrab(omm->omm_remote_parent->d_inode);
4388 child_inode = osd_create_local_agent_inode(env, osd,
4391 if (IS_ERR(child_inode))
4392 RETURN(PTR_ERR(child_inode));
4395 /* Insert local entry */
4396 child = osd_object_find(env, dt, fid);
4397 if (IS_ERR(child)) {
4398 CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4399 osd_name(osd), PFID(fid),
4400 id->oii_ino, id->oii_gen,
4401 (int)PTR_ERR(child));
4402 RETURN(PTR_ERR(child));
4404 child_inode = igrab(child->oo_inode);
4407 rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th);
4409 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4410 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4414 osd_object_put(env, child);
4415 LASSERT(osd_invariant(obj));
4420 * Initialize osd Iterator for given osd index object.
4422 * \param dt osd index object
4425 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4426 struct dt_object *dt,
4428 struct lustre_capa *capa)
4430 struct osd_it_iam *it;
4431 struct osd_thread_info *oti = osd_oti_get(env);
4432 struct osd_object *obj = osd_dt_obj(dt);
4433 struct lu_object *lo = &dt->do_lu;
4434 struct iam_path_descr *ipd;
4435 struct iam_container *bag = &obj->oo_dir->od_container;
4437 LASSERT(lu_object_exists(lo));
4439 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
4440 return ERR_PTR(-EACCES);
4443 ipd = osd_it_ipd_get(env, bag);
4444 if (likely(ipd != NULL)) {
4448 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4449 return (struct dt_it *)it;
4451 return ERR_PTR(-ENOMEM);
4455 * free given Iterator.
4458 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4460 struct osd_it_iam *it = (struct osd_it_iam *)di;
4461 struct osd_object *obj = it->oi_obj;
4463 iam_it_fini(&it->oi_it);
4464 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4465 lu_object_put(env, &obj->oo_dt.do_lu);
4469 * Move Iterator to record specified by \a key
4471 * \param di osd iterator
4472 * \param key key for index
4474 * \retval +ve di points to record with least key not larger than key
4475 * \retval 0 di points to exact matched key
4476 * \retval -ve failure
4479 static int osd_it_iam_get(const struct lu_env *env,
4480 struct dt_it *di, const struct dt_key *key)
4482 struct osd_thread_info *oti = osd_oti_get(env);
4483 struct osd_it_iam *it = (struct osd_it_iam *)di;
4485 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4486 /* swab quota uid/gid */
4487 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4488 key = (struct dt_key *)&oti->oti_quota_id;
4491 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4497 * \param di osd iterator
4499 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4501 struct osd_it_iam *it = (struct osd_it_iam *)di;
4503 iam_it_put(&it->oi_it);
4507 * Move iterator by one record
4509 * \param di osd iterator
4511 * \retval +1 end of container reached
4513 * \retval -ve failure
4516 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4518 struct osd_it_iam *it = (struct osd_it_iam *)di;
4520 return iam_it_next(&it->oi_it);
4524 * Return pointer to the key under iterator.
4527 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4528 const struct dt_it *di)
4530 struct osd_thread_info *oti = osd_oti_get(env);
4531 struct osd_it_iam *it = (struct osd_it_iam *)di;
4532 struct osd_object *obj = it->oi_obj;
4535 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4537 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4538 /* swab quota uid/gid */
4539 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4540 key = (struct dt_key *)&oti->oti_quota_id;
4547 * Return size of key under iterator (in bytes)
4550 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4552 struct osd_it_iam *it = (struct osd_it_iam *)di;
4554 return iam_it_key_size(&it->oi_it);
4558 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4560 /* check if file type is required */
4561 if (ent->lde_attrs & LUDA_TYPE) {
4562 struct luda_type *lt;
4563 int align = sizeof(*lt) - 1;
4565 len = (len + align) & ~align;
4566 lt = (struct luda_type *)(ent->lde_name + len);
4567 lt->lt_type = cpu_to_le16(DTTOIF(type));
4570 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4574 * build lu direct from backend fs dirent.
4578 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4579 char *name, __u16 namelen, __u16 type, __u32 attr)
4581 ent->lde_attrs = attr | LUDA_FID;
4582 fid_cpu_to_le(&ent->lde_fid, fid);
4584 ent->lde_hash = cpu_to_le64(offset);
4585 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4587 strncpy(ent->lde_name, name, namelen);
4588 ent->lde_name[namelen] = '\0';
4589 ent->lde_namelen = cpu_to_le16(namelen);
4591 /* append lustre attributes */
4592 osd_it_append_attrs(ent, namelen, type);
4596 * Return pointer to the record under iterator.
4598 static int osd_it_iam_rec(const struct lu_env *env,
4599 const struct dt_it *di,
4600 struct dt_rec *dtrec, __u32 attr)
4602 struct osd_it_iam *it = (struct osd_it_iam *)di;
4603 struct osd_thread_info *info = osd_oti_get(env);
4606 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4607 const struct osd_fid_pack *rec;
4608 struct lu_fid *fid = &info->oti_fid;
4609 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
4615 name = (char *)iam_it_key_get(&it->oi_it);
4617 RETURN(PTR_ERR(name));
4619 namelen = iam_it_key_size(&it->oi_it);
4621 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4623 RETURN(PTR_ERR(rec));
4625 rc = osd_fid_unpack(fid, rec);
4629 hash = iam_it_store(&it->oi_it);
4631 /* IAM does not store object type in IAM index (dir) */
4632 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4634 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4635 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4636 (struct iam_rec *)dtrec);
4637 osd_quota_unpack(it->oi_obj, dtrec);
4639 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4640 (struct iam_rec *)dtrec);
4647 * Returns cookie for current Iterator position.
4649 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4651 struct osd_it_iam *it = (struct osd_it_iam *)di;
4653 return iam_it_store(&it->oi_it);
4657 * Restore iterator from cookie.
4659 * \param di osd iterator
4660 * \param hash Iterator location cookie
4662 * \retval +ve di points to record with least key not larger than key.
4663 * \retval 0 di points to exact matched key
4664 * \retval -ve failure
4667 static int osd_it_iam_load(const struct lu_env *env,
4668 const struct dt_it *di, __u64 hash)
4670 struct osd_it_iam *it = (struct osd_it_iam *)di;
4672 return iam_it_load(&it->oi_it, hash);
4675 static const struct dt_index_operations osd_index_iam_ops = {
4676 .dio_lookup = osd_index_iam_lookup,
4677 .dio_declare_insert = osd_index_declare_iam_insert,
4678 .dio_insert = osd_index_iam_insert,
4679 .dio_declare_delete = osd_index_declare_iam_delete,
4680 .dio_delete = osd_index_iam_delete,
4682 .init = osd_it_iam_init,
4683 .fini = osd_it_iam_fini,
4684 .get = osd_it_iam_get,
4685 .put = osd_it_iam_put,
4686 .next = osd_it_iam_next,
4687 .key = osd_it_iam_key,
4688 .key_size = osd_it_iam_key_size,
4689 .rec = osd_it_iam_rec,
4690 .store = osd_it_iam_store,
4691 .load = osd_it_iam_load
4697 * Creates or initializes iterator context.
4699 * \retval struct osd_it_ea, iterator structure on success
4702 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4703 struct dt_object *dt,
4705 struct lustre_capa *capa)
4707 struct osd_object *obj = osd_dt_obj(dt);
4708 struct osd_thread_info *info = osd_oti_get(env);
4709 struct osd_it_ea *it = &info->oti_it_ea;
4710 struct file *file = &it->oie_file;
4711 struct lu_object *lo = &dt->do_lu;
4712 struct dentry *obj_dentry = &info->oti_it_dentry;
4714 LASSERT(lu_object_exists(lo));
4716 obj_dentry->d_inode = obj->oo_inode;
4717 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4718 obj_dentry->d_name.hash = 0;
4720 it->oie_rd_dirent = 0;
4721 it->oie_it_dirent = 0;
4722 it->oie_dirent = NULL;
4723 it->oie_buf = info->oti_it_ea_buf;
4726 /* Reset the "file" totally to avoid to reuse any old value from
4727 * former readdir handling, the "file->f_pos" should be zero. */
4728 memset(file, 0, sizeof(*file));
4729 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4730 if (attr & LUDA_64BITHASH)
4731 file->f_mode = FMODE_64BITHASH;
4733 file->f_mode = FMODE_32BITHASH;
4734 file->f_dentry = obj_dentry;
4735 file->f_mapping = obj->oo_inode->i_mapping;
4736 file->f_op = obj->oo_inode->i_fop;
4737 set_file_inode(file, obj->oo_inode);
4740 RETURN((struct dt_it *) it);
4744 * Destroy or finishes iterator context.
4746 * \param di iterator structure to be destroyed
4748 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4750 struct osd_it_ea *it = (struct osd_it_ea *)di;
4751 struct osd_object *obj = it->oie_obj;
4752 struct inode *inode = obj->oo_inode;
4755 it->oie_file.f_op->release(inode, &it->oie_file);
4756 lu_object_put(env, &obj->oo_dt.do_lu);
4761 * It position the iterator at given key, so that next lookup continues from
4762 * that key Or it is similar to dio_it->load() but based on a key,
4763 * rather than file position.
4765 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4768 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4770 static int osd_it_ea_get(const struct lu_env *env,
4771 struct dt_it *di, const struct dt_key *key)
4773 struct osd_it_ea *it = (struct osd_it_ea *)di;
4776 LASSERT(((const char *)key)[0] == '\0');
4777 it->oie_file.f_pos = 0;
4778 it->oie_rd_dirent = 0;
4779 it->oie_it_dirent = 0;
4780 it->oie_dirent = NULL;
4788 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4792 struct osd_filldir_cbs {
4793 #ifdef HAVE_DIR_CONTEXT
4794 struct dir_context ctx;
4796 struct osd_it_ea *it;
4799 * It is called internally by ->readdir(). It fills the
4800 * iterator's in-memory data structure with required
4801 * information i.e. name, namelen, rec_size etc.
4803 * \param buf in which information to be filled in.
4804 * \param name name of the file in given dir
4806 * \retval 0 on success
4807 * \retval 1 on buffer full
4809 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
4810 loff_t offset, __u64 ino,
4813 struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
4814 struct osd_object *obj = it->oie_obj;
4815 struct osd_it_ea_dirent *ent = it->oie_dirent;
4816 struct lu_fid *fid = &ent->oied_fid;
4817 struct osd_fid_pack *rec;
4820 /* this should never happen */
4821 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4822 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4826 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4830 /* "." is just the object itself. */
4831 if (namelen == 1 && name[0] == '.') {
4832 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4833 } else if (d_type & LDISKFS_DIRENT_LUFID) {
4834 rec = (struct osd_fid_pack*) (name + namelen + 1);
4835 if (osd_fid_unpack(fid, rec) != 0)
4840 d_type &= ~LDISKFS_DIRENT_LUFID;
4842 /* NOT export local root. */
4843 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
4844 ino = obj->oo_inode->i_ino;
4845 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4848 ent->oied_ino = ino;
4849 ent->oied_off = offset;
4850 ent->oied_namelen = namelen;
4851 ent->oied_type = d_type;
4853 memcpy(ent->oied_name, name, namelen);
4855 it->oie_rd_dirent++;
4856 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4861 * Calls ->readdir() to load a directory entry at a time
4862 * and stored it in iterator's in-memory data structure.
4864 * \param di iterator's in memory structure
4866 * \retval 0 on success
4867 * \retval -ve on error
4868 * \retval +1 reach the end of entry
4870 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4871 const struct dt_it *di)
4873 struct osd_it_ea *it = (struct osd_it_ea *)di;
4874 struct osd_object *obj = it->oie_obj;
4875 struct inode *inode = obj->oo_inode;
4876 struct htree_lock *hlock = NULL;
4877 struct file *filp = &it->oie_file;
4879 struct osd_filldir_cbs buf = {
4880 #ifdef HAVE_DIR_CONTEXT
4881 .ctx.actor = osd_ldiskfs_filldir,
4887 it->oie_dirent = it->oie_buf;
4888 it->oie_rd_dirent = 0;
4890 if (obj->oo_hl_head != NULL) {
4891 hlock = osd_oti_get(env)->oti_hlock;
4892 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4893 inode, LDISKFS_HLOCK_READDIR);
4895 down_read(&obj->oo_ext_idx_sem);
4898 #ifdef HAVE_DIR_CONTEXT
4899 buf.ctx.pos = filp->f_pos;
4900 rc = inode->i_fop->iterate(filp, &buf.ctx);
4901 filp->f_pos = buf.ctx.pos;
4903 rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
4907 ldiskfs_htree_unlock(hlock);
4909 up_read(&obj->oo_ext_idx_sem);
4911 if (it->oie_rd_dirent == 0) {
4912 /*If it does not get any dirent, it means it has been reached
4913 *to the end of the dir */
4914 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
4918 it->oie_dirent = it->oie_buf;
4919 it->oie_it_dirent = 1;
4926 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4927 * to load a directory entry at a time and stored it in
4928 * iterator's in-memory data structure.
4930 * \param di iterator's in memory structure
4932 * \retval +ve iterator reached to end
4933 * \retval 0 iterator not reached to end
4934 * \retval -ve on error
4936 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4938 struct osd_it_ea *it = (struct osd_it_ea *)di;
4943 if (it->oie_it_dirent < it->oie_rd_dirent) {
4945 (void *) it->oie_dirent +
4946 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4947 it->oie_dirent->oied_namelen);
4948 it->oie_it_dirent++;
4951 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4954 rc = osd_ldiskfs_it_fill(env, di);
4961 * Returns the key at current position from iterator's in memory structure.
4963 * \param di iterator's in memory structure
4965 * \retval key i.e. struct dt_key on success
4967 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4968 const struct dt_it *di)
4970 struct osd_it_ea *it = (struct osd_it_ea *)di;
4972 return (struct dt_key *)it->oie_dirent->oied_name;
4976 * Returns the key's size at current position from iterator's in memory structure.
4978 * \param di iterator's in memory structure
4980 * \retval key_size i.e. struct dt_key on success
4982 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4984 struct osd_it_ea *it = (struct osd_it_ea *)di;
4986 return it->oie_dirent->oied_namelen;
4990 osd_dirent_update(handle_t *jh, struct super_block *sb,
4991 struct osd_it_ea_dirent *ent, struct lu_fid *fid,
4992 struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
4994 struct osd_fid_pack *rec;
4998 LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
4999 LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
5001 rc = ldiskfs_journal_get_write_access(jh, bh);
5003 CERROR("%.16s: fail to write access for update dirent: "
5004 "name = %.*s, rc = %d\n",
5005 LDISKFS_SB(sb)->s_es->s_volume_name,
5006 ent->oied_namelen, ent->oied_name, rc);
5010 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5011 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5012 rc = ldiskfs_journal_dirty_metadata(jh, bh);
5014 CERROR("%.16s: fail to dirty metadata for update dirent: "
5015 "name = %.*s, rc = %d\n",
5016 LDISKFS_SB(sb)->s_es->s_volume_name,
5017 ent->oied_namelen, ent->oied_name, rc);
5023 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
5025 if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
5026 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5033 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5035 LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5036 "dot_dotdot = %d\n", dot_dotdot);
5038 if (LDISKFS_DIR_REC_LEN(de) >=
5039 __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5046 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5047 struct inode *dir, struct inode *inode,
5048 struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5049 struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5050 struct htree_lock *hlock)
5052 struct dentry *dentry;
5053 struct osd_fid_pack *rec;
5054 struct ldiskfs_dentry_param *ldp;
5058 if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5059 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5062 /* There is enough space to hold the FID-in-dirent. */
5063 if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
5064 dir->i_sb->s_blocksize)) {
5065 rc = ldiskfs_journal_get_write_access(jh, bh);
5067 CERROR("%.16s: fail to write access for reinsert "
5068 "dirent: name = %.*s, rc = %d\n",
5069 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5070 ent->oied_namelen, ent->oied_name, rc);
5074 de->name[de->name_len] = 0;
5075 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5076 rec->fp_len = sizeof(struct lu_fid) + 1;
5077 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5078 de->file_type |= LDISKFS_DIRENT_LUFID;
5080 rc = ldiskfs_journal_dirty_metadata(jh, bh);
5082 CERROR("%.16s: fail to dirty metadata for reinsert "
5083 "dirent: name = %.*s, rc = %d\n",
5084 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5085 ent->oied_namelen, ent->oied_name, rc);
5090 rc = ldiskfs_delete_entry(jh, dir, de, bh);
5092 CERROR("%.16s: fail to delete entry for reinsert dirent: "
5093 "name = %.*s, rc = %d\n",
5094 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5095 ent->oied_namelen, ent->oied_name, rc);
5099 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5101 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5102 osd_get_ldiskfs_dirent_param(ldp, (const struct dt_rec *)fid);
5103 dentry->d_fsdata = (void *)ldp;
5104 ll_vfs_dq_init(dir);
5105 rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5106 /* It is too bad, we cannot reinsert the name entry back.
5107 * That means we lose it! */
5109 CERROR("%.16s: fail to insert entry for reinsert dirent: "
5110 "name = %.*s, rc = %d\n",
5111 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5112 ent->oied_namelen, ent->oied_name, rc);
5118 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5119 struct osd_it_ea *it, struct lu_fid *fid,
5120 struct osd_inode_id *id, __u32 *attr)
5122 struct osd_thread_info *info = osd_oti_get(env);
5123 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
5124 struct osd_device *dev = osd_obj2dev(obj);
5125 struct super_block *sb = osd_sb(dev);
5126 const char *devname =
5127 LDISKFS_SB(sb)->s_es->s_volume_name;
5128 struct osd_it_ea_dirent *ent = it->oie_dirent;
5129 struct inode *dir = obj->oo_inode;
5130 struct htree_lock *hlock = NULL;
5131 struct buffer_head *bh = NULL;
5132 handle_t *jh = NULL;
5133 struct ldiskfs_dir_entry_2 *de;
5134 struct dentry *dentry;
5135 struct inode *inode;
5142 if (ent->oied_name[0] == '.') {
5143 if (ent->oied_namelen == 1)
5145 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5149 dentry = osd_child_dentry_get(env, obj, ent->oied_name,
5152 /* We need to ensure that the name entry is still valid.
5153 * Because it may be removed or renamed by other already.
5155 * The unlink or rename operation will start journal before PDO lock,
5156 * so to avoid deadlock, here we need to start journal handle before
5157 * related PDO lock also. But because we do not know whether there
5158 * will be something to be repaired before PDO lock, we just start
5159 * journal without conditions.
5161 * We may need to remove the name entry firstly, then insert back.
5162 * One credit is for user quota file update.
5163 * One credit is for group quota file update.
5164 * Two credits are for dirty inode. */
5165 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5166 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5169 if (dev->od_dirent_journal) {
5170 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5173 CERROR("%.16s: fail to start trans for dirent "
5174 "check_repair: credits %d, name %.*s, rc %d\n",
5175 devname, credits, ent->oied_namelen,
5176 ent->oied_name, rc);
5180 if (obj->oo_hl_head != NULL) {
5181 hlock = osd_oti_get(env)->oti_hlock;
5182 /* "0" means exclusive lock for the whole directory.
5183 * We need to prevent others access such name entry
5184 * during the delete + insert. Neither HLOCK_ADD nor
5185 * HLOCK_DEL cannot guarantee the atomicity. */
5186 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5188 down_write(&obj->oo_ext_idx_sem);
5191 if (obj->oo_hl_head != NULL) {
5192 hlock = osd_oti_get(env)->oti_hlock;
5193 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5194 LDISKFS_HLOCK_LOOKUP);
5196 down_read(&obj->oo_ext_idx_sem);
5200 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5201 /* For dot/dotdot entry, if there is not enough space to hold the
5202 * FID-in-dirent, just keep them there. It only happens when the
5203 * device upgraded from 1.8 or restored from MDT file-level backup.
5204 * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5205 * and needs to get FID from LMA when readdir, it will not affect the
5206 * performance much. */
5207 if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
5208 (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5209 *attr |= LUDA_IGNORE;
5210 GOTO(out_journal, rc = 0);
5213 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5214 inode = osd_iget(info, dev, id);
5215 if (IS_ERR(inode)) {
5216 rc = PTR_ERR(inode);
5217 if (rc == -ENOENT || rc == -ESTALE) {
5218 *attr |= LUDA_IGNORE;
5222 GOTO(out_journal, rc);
5225 /* skip the REMOTE_PARENT_DIR. */
5226 if (inode == dev->od_mdt_map->omm_remote_parent->d_inode)
5227 GOTO(out_inode, rc = 0);
5229 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
5231 LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
5233 if (fid_is_sane(fid)) {
5234 /* FID-in-dirent is valid. */
5235 if (lu_fid_eq(fid, &lma->lma_self_fid))
5236 GOTO(out_inode, rc = 0);
5238 /* Do not repair under dryrun mode. */
5239 if (*attr & LUDA_VERIFY_DRYRUN) {
5240 *attr |= LUDA_REPAIR;
5241 GOTO(out_inode, rc = 0);
5244 if (!dev->od_dirent_journal) {
5248 ldiskfs_htree_unlock(hlock);
5250 up_read(&obj->oo_ext_idx_sem);
5251 dev->od_dirent_journal = 1;
5255 *fid = lma->lma_self_fid;
5257 /* Update the FID-in-dirent. */
5258 rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
5260 *attr |= LUDA_REPAIR;
5262 /* Do not repair under dryrun mode. */
5263 if (*attr & LUDA_VERIFY_DRYRUN) {
5264 *fid = lma->lma_self_fid;
5265 *attr |= LUDA_REPAIR;
5266 GOTO(out_inode, rc = 0);
5269 if (!dev->od_dirent_journal) {
5273 ldiskfs_htree_unlock(hlock);
5275 up_read(&obj->oo_ext_idx_sem);
5276 dev->od_dirent_journal = 1;
5280 *fid = lma->lma_self_fid;
5282 /* Append the FID-in-dirent. */
5283 rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5284 fid, bh, de, hlock);
5286 *attr |= LUDA_REPAIR;
5288 } else if (rc == -ENODATA) {
5289 /* Do not repair under dryrun mode. */
5290 if (*attr & LUDA_VERIFY_DRYRUN) {
5291 if (fid_is_sane(fid)) {
5292 *attr |= LUDA_REPAIR;
5294 lu_igif_build(fid, inode->i_ino,
5295 inode->i_generation);
5296 *attr |= LUDA_UPGRADE;
5298 GOTO(out_inode, rc = 0);
5301 if (!dev->od_dirent_journal) {
5305 ldiskfs_htree_unlock(hlock);
5307 up_read(&obj->oo_ext_idx_sem);
5308 dev->od_dirent_journal = 1;
5313 if (unlikely(fid_is_sane(fid))) {
5314 /* FID-in-dirent exists, but FID-in-LMA is lost.
5315 * Trust the FID-in-dirent, and add FID-in-LMA. */
5316 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5318 *attr |= LUDA_REPAIR;
5320 lu_igif_build(fid, inode->i_ino, inode->i_generation);
5321 /* It is probably IGIF object. Only aappend the
5322 * FID-in-dirent. OI scrub will process FID-in-LMA. */
5323 rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5324 fid, bh, de, hlock);
5326 *attr |= LUDA_UPGRADE;
5330 GOTO(out_inode, rc);
5337 if (hlock != NULL) {
5338 ldiskfs_htree_unlock(hlock);
5340 if (dev->od_dirent_journal)
5341 up_write(&obj->oo_ext_idx_sem);
5343 up_read(&obj->oo_ext_idx_sem);
5346 ldiskfs_journal_stop(jh);
5347 if (rc >= 0 && !dirty)
5348 dev->od_dirent_journal = 0;
5353 * Returns the value at current position from iterator's in memory structure.
5355 * \param di struct osd_it_ea, iterator's in memory structure
5356 * \param attr attr requested for dirent.
5357 * \param lde lustre dirent
5359 * \retval 0 no error and \param lde has correct lustre dirent.
5360 * \retval -ve on error
5362 static inline int osd_it_ea_rec(const struct lu_env *env,
5363 const struct dt_it *di,
5364 struct dt_rec *dtrec, __u32 attr)
5366 struct osd_it_ea *it = (struct osd_it_ea *)di;
5367 struct osd_object *obj = it->oie_obj;
5368 struct osd_device *dev = osd_obj2dev(obj);
5369 struct osd_scrub *scrub = &dev->od_scrub;
5370 struct scrub_file *sf = &scrub->os_file;
5371 struct osd_thread_info *oti = osd_oti_get(env);
5372 struct osd_inode_id *id = &oti->oti_id;
5373 struct osd_idmap_cache *oic = &oti->oti_cache;
5374 struct lu_fid *fid = &it->oie_dirent->oied_fid;
5375 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
5376 __u32 ino = it->oie_dirent->oied_ino;
5380 if (attr & LUDA_VERIFY) {
5382 if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) {
5383 attr |= LUDA_IGNORE;
5386 rc = osd_dirent_check_repair(env, obj, it, fid, id,
5390 attr &= ~LU_DIRENT_ATTRS_MASK;
5391 if (!fid_is_sane(fid)) {
5392 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP) &&
5393 likely(it->oie_dirent->oied_namelen != 2 ||
5394 it->oie_dirent->oied_name[0] != '.' ||
5395 it->oie_dirent->oied_name[1] != '.'))
5398 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5400 osd_id_gen(id, ino, OSD_OII_NOGEN);
5404 /* Pack the entry anyway, at least the offset is right. */
5405 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5406 it->oie_dirent->oied_name,
5407 it->oie_dirent->oied_namelen,
5408 it->oie_dirent->oied_type, attr);
5413 if (osd_remote_fid(env, dev, fid))
5416 if (likely(!(attr & LUDA_IGNORE)))
5417 rc = osd_add_oi_cache(oti, dev, id, fid);
5419 if (!(attr & LUDA_VERIFY) &&
5420 (scrub->os_pos_current <= ino) &&
5421 ((sf->sf_flags & SF_INCONSISTENT) ||
5422 (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
5423 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
5424 osd_consistency_check(oti, dev, oic);
5430 * Returns the record size size at current position.
5432 * This function will return record(lu_dirent) size in bytes.
5434 * \param[in] env execution environment
5435 * \param[in] di iterator's in memory structure
5436 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
5437 * calculate the lu_dirent size.
5439 * \retval record size(in bytes & in memory) of the current lu_dirent
5442 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5445 struct osd_it_ea *it = (struct osd_it_ea *)di;
5447 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5451 * Returns a cookie for current position of the iterator head, so that
5452 * user can use this cookie to load/start the iterator next time.
5454 * \param di iterator's in memory structure
5456 * \retval cookie for current position, on success
5458 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5460 struct osd_it_ea *it = (struct osd_it_ea *)di;
5462 return it->oie_dirent->oied_off;
5466 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5467 * to load a directory entry at a time and stored it i inn,
5468 * in iterator's in-memory data structure.
5470 * \param di struct osd_it_ea, iterator's in memory structure
5472 * \retval +ve on success
5473 * \retval -ve on error
5475 static int osd_it_ea_load(const struct lu_env *env,
5476 const struct dt_it *di, __u64 hash)
5478 struct osd_it_ea *it = (struct osd_it_ea *)di;
5482 it->oie_file.f_pos = hash;
5484 rc = osd_ldiskfs_it_fill(env, di);
5495 * Index lookup function for interoperability mode (b11826).
5497 * \param key, key i.e. file name to be searched
5499 * \retval +ve, on success
5500 * \retval -ve, on error
5502 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5503 struct dt_rec *rec, const struct dt_key *key,
5504 struct lustre_capa *capa)
5506 struct osd_object *obj = osd_dt_obj(dt);
5511 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5512 LINVRNT(osd_invariant(obj));
5514 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
5517 rc = osd_ea_lookup_rec(env, obj, rec, key);
5524 * Index and Iterator operations for interoperability
5525 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5527 static const struct dt_index_operations osd_index_ea_ops = {
5528 .dio_lookup = osd_index_ea_lookup,
5529 .dio_declare_insert = osd_index_declare_ea_insert,
5530 .dio_insert = osd_index_ea_insert,
5531 .dio_declare_delete = osd_index_declare_ea_delete,
5532 .dio_delete = osd_index_ea_delete,
5534 .init = osd_it_ea_init,
5535 .fini = osd_it_ea_fini,
5536 .get = osd_it_ea_get,
5537 .put = osd_it_ea_put,
5538 .next = osd_it_ea_next,
5539 .key = osd_it_ea_key,
5540 .key_size = osd_it_ea_key_size,
5541 .rec = osd_it_ea_rec,
5542 .rec_size = osd_it_ea_rec_size,
5543 .store = osd_it_ea_store,
5544 .load = osd_it_ea_load
5548 static void *osd_key_init(const struct lu_context *ctx,
5549 struct lu_context_key *key)
5551 struct osd_thread_info *info;
5553 OBD_ALLOC_PTR(info);
5555 return ERR_PTR(-ENOMEM);
5557 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5558 if (info->oti_it_ea_buf == NULL)
5561 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5563 info->oti_hlock = ldiskfs_htree_lock_alloc();
5564 if (info->oti_hlock == NULL)
5570 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5573 return ERR_PTR(-ENOMEM);
5576 static void osd_key_fini(const struct lu_context *ctx,
5577 struct lu_context_key *key, void* data)
5579 struct osd_thread_info *info = data;
5581 if (info->oti_hlock != NULL)
5582 ldiskfs_htree_lock_free(info->oti_hlock);
5583 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5584 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5585 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5589 static void osd_key_exit(const struct lu_context *ctx,
5590 struct lu_context_key *key, void *data)
5592 struct osd_thread_info *info = data;
5594 LASSERT(info->oti_r_locks == 0);
5595 LASSERT(info->oti_w_locks == 0);
5596 LASSERT(info->oti_txns == 0);
5599 /* type constructor/destructor: osd_type_init, osd_type_fini */
5600 LU_TYPE_INIT_FINI(osd, &osd_key);
5602 struct lu_context_key osd_key = {
5603 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5604 .lct_init = osd_key_init,
5605 .lct_fini = osd_key_fini,
5606 .lct_exit = osd_key_exit
5610 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5611 const char *name, struct lu_device *next)
5613 struct osd_device *osd = osd_dev(d);
5615 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5616 >= sizeof(osd->od_svname))
5618 return osd_procfs_init(osd, name);
5621 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5623 struct seq_server_site *ss = osd_seq_site(osd);
5627 if (osd->od_is_ost || osd->od_cl_seq != NULL)
5630 if (unlikely(ss == NULL))
5633 OBD_ALLOC_PTR(osd->od_cl_seq);
5634 if (osd->od_cl_seq == NULL)
5637 rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5638 osd->od_svname, ss->ss_server_seq);
5641 OBD_FREE_PTR(osd->od_cl_seq);
5642 osd->od_cl_seq = NULL;
5648 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5650 if (osd->od_cl_seq == NULL)
5653 seq_client_fini(osd->od_cl_seq);
5654 OBD_FREE_PTR(osd->od_cl_seq);
5655 osd->od_cl_seq = NULL;
5658 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5662 /* shutdown quota slave instance associated with the device */
5663 if (o->od_quota_slave != NULL) {
5664 qsd_fini(env, o->od_quota_slave);
5665 o->od_quota_slave = NULL;
5668 osd_fid_fini(env, o);
5673 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5677 if (o->od_mnt != NULL) {
5678 shrink_dcache_sb(osd_sb(o));
5679 osd_sync(env, &o->od_dt_dev);
5688 static int osd_mount(const struct lu_env *env,
5689 struct osd_device *o, struct lustre_cfg *cfg)
5691 const char *name = lustre_cfg_string(cfg, 0);
5692 const char *dev = lustre_cfg_string(cfg, 1);
5694 unsigned long page, s_flags, lmd_flags = 0;
5695 struct page *__page;
5696 struct file_system_type *type;
5697 char *options = NULL;
5699 struct osd_thread_info *info = osd_oti_get(env);
5700 struct lu_fid *fid = &info->oti_fid;
5701 struct inode *inode;
5705 if (o->od_mnt != NULL)
5708 if (strlen(dev) >= sizeof(o->od_mntdev))
5710 strcpy(o->od_mntdev, dev);
5712 OBD_PAGE_ALLOC(__page, GFP_IOFS);
5714 GOTO(out, rc = -ENOMEM);
5716 str = lustre_cfg_string(cfg, 2);
5717 s_flags = simple_strtoul(str, NULL, 0);
5718 str = strstr(str, ":");
5720 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5721 opts = lustre_cfg_string(cfg, 3);
5722 page = (unsigned long)page_address(__page);
5723 options = (char *)page;
5726 strcat(options, "user_xattr,acl");
5728 strcat(options, opts);
5730 /* Glom up mount options */
5731 if (*options != '\0')
5732 strcat(options, ",");
5733 strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
5735 type = get_fs_type("ldiskfs");
5737 CERROR("%s: cannot find ldiskfs module\n", name);
5738 GOTO(out, rc = -ENODEV);
5741 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5742 module_put(type->owner);
5744 if (IS_ERR(o->od_mnt)) {
5745 rc = PTR_ERR(o->od_mnt);
5747 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5751 #ifdef HAVE_DEV_SET_RDONLY
5752 if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5753 CERROR("%s: underlying device %s is marked as read-only. "
5754 "Setup failed\n", name, dev);
5755 GOTO(out_mnt, rc = -EROFS);
5759 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5760 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5761 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5762 GOTO(out_mnt, rc = -EINVAL);
5765 #ifdef LDISKFS_MOUNT_DIRDATA
5766 if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
5767 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5768 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
5769 else if (!o->od_is_ost)
5770 CWARN("%s: device %s was upgraded from Lustre-1.x without "
5771 "enabling the dirdata feature. If you do not want to "
5772 "downgrade to Lustre-1.x again, you can enable it via "
5773 "'tune2fs -O dirdata device'\n", name, dev);
5775 inode = osd_sb(o)->s_root->d_inode;
5776 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
5777 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
5779 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
5783 if (lmd_flags & LMD_FLG_NOSCRUB)
5794 OBD_PAGE_FREE(__page);
5799 static struct lu_device *osd_device_fini(const struct lu_env *env,
5800 struct lu_device *d)
5802 struct osd_device *o = osd_dev(d);
5805 osd_shutdown(env, o);
5807 osd_scrub_cleanup(env, o);
5808 osd_obj_map_fini(o);
5814 static int osd_device_init0(const struct lu_env *env,
5815 struct osd_device *o,
5816 struct lustre_cfg *cfg)
5818 struct lu_device *l = osd2lu_dev(o);
5819 struct osd_thread_info *info;
5823 /* if the module was re-loaded, env can loose its keys */
5824 rc = lu_env_refill((struct lu_env *) env);
5827 info = osd_oti_get(env);
5830 l->ld_ops = &osd_lu_ops;
5831 o->od_dt_dev.dd_ops = &osd_dt_ops;
5833 spin_lock_init(&o->od_osfs_lock);
5834 mutex_init(&o->od_otable_mutex);
5836 o->od_capa_hash = init_capa_hash();
5837 if (o->od_capa_hash == NULL)
5838 GOTO(out, rc = -ENOMEM);
5840 o->od_read_cache = 1;
5841 o->od_writethrough_cache = 1;
5842 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
5844 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
5845 sizeof(o->od_svname));
5846 if (cplen >= sizeof(o->od_svname)) {
5851 if (server_name_is_ost(o->od_svname))
5854 rc = osd_mount(env, o, cfg);
5858 rc = osd_obj_map_init(env, o);
5862 rc = lu_site_init(&o->od_site, l);
5864 GOTO(out_compat, rc);
5865 o->od_site.ls_bottom_dev = l;
5867 rc = lu_site_init_finish(&o->od_site);
5871 /* self-repair LMA by default */
5872 o->od_lma_self_repair = 1;
5874 INIT_LIST_HEAD(&o->od_ios_list);
5875 /* setup scrub, including OI files initialization */
5876 rc = osd_scrub_setup(env, o);
5880 rc = osd_procfs_init(o, o->od_svname);
5882 CERROR("%s: can't initialize procfs: rc = %d\n",
5884 GOTO(out_scrub, rc);
5887 LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
5889 /* initialize quota slave instance */
5890 o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
5892 if (IS_ERR(o->od_quota_slave)) {
5893 rc = PTR_ERR(o->od_quota_slave);
5894 o->od_quota_slave = NULL;
5895 GOTO(out_procfs, rc);
5903 osd_scrub_cleanup(env, o);
5905 lu_site_fini(&o->od_site);
5907 osd_obj_map_fini(o);
5911 cleanup_capa_hash(o->od_capa_hash);
5916 static struct lu_device *osd_device_alloc(const struct lu_env *env,
5917 struct lu_device_type *t,
5918 struct lustre_cfg *cfg)
5920 struct osd_device *o;
5925 return ERR_PTR(-ENOMEM);
5927 rc = dt_device_init(&o->od_dt_dev, t);
5929 /* Because the ctx might be revived in dt_device_init,
5930 * refill the env here */
5931 lu_env_refill((struct lu_env *)env);
5932 rc = osd_device_init0(env, o, cfg);
5934 dt_device_fini(&o->od_dt_dev);
5937 if (unlikely(rc != 0))
5940 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
5943 static struct lu_device *osd_device_free(const struct lu_env *env,
5944 struct lu_device *d)
5946 struct osd_device *o = osd_dev(d);
5949 cleanup_capa_hash(o->od_capa_hash);
5950 /* XXX: make osd top device in order to release reference */
5951 d->ld_site->ls_top_dev = d;
5952 lu_site_purge(env, d->ld_site, -1);
5953 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
5954 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
5955 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
5957 lu_site_fini(&o->od_site);
5958 dt_device_fini(&o->od_dt_dev);
5963 static int osd_process_config(const struct lu_env *env,
5964 struct lu_device *d, struct lustre_cfg *cfg)
5966 struct osd_device *o = osd_dev(d);
5970 switch (cfg->lcfg_command) {
5972 rc = osd_mount(env, o, cfg);
5975 lu_dev_del_linkage(d->ld_site, d);
5976 rc = osd_shutdown(env, o);
5979 LASSERT(&o->od_dt_dev);
5980 rc = class_process_proc_seq_param(PARAM_OSD,
5981 lprocfs_osd_obd_vars,
5982 cfg, &o->od_dt_dev);
5983 if (rc > 0 || rc == -ENOSYS)
5984 rc = class_process_proc_seq_param(PARAM_OST,
5985 lprocfs_osd_obd_vars,
5986 cfg, &o->od_dt_dev);
5995 static int osd_recovery_complete(const struct lu_env *env,
5996 struct lu_device *d)
5998 struct osd_device *osd = osd_dev(d);
6002 if (osd->od_quota_slave == NULL)
6005 /* start qsd instance on recovery completion, this notifies the quota
6006 * slave code that we are about to process new requests now */
6007 rc = qsd_start(env, osd->od_quota_slave);
6012 * we use exports to track all osd users
6014 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
6015 struct obd_device *obd, struct obd_uuid *cluuid,
6016 struct obd_connect_data *data, void *localdata)
6018 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6019 struct lustre_handle conn;
6023 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6025 rc = class_connect(&conn, obd, cluuid);
6029 *exp = class_conn2export(&conn);
6031 spin_lock(&osd->od_osfs_lock);
6033 spin_unlock(&osd->od_osfs_lock);
6039 * once last export (we don't count self-export) disappeared
6040 * osd can be released
6042 static int osd_obd_disconnect(struct obd_export *exp)
6044 struct obd_device *obd = exp->exp_obd;
6045 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6046 int rc, release = 0;
6049 /* Only disconnect the underlying layers on the final disconnect. */
6050 spin_lock(&osd->od_osfs_lock);
6052 if (osd->od_connects == 0)
6054 spin_unlock(&osd->od_osfs_lock);
6056 rc = class_disconnect(exp); /* bz 9811 */
6058 if (rc == 0 && release)
6059 class_manual_cleanup(obd);
6063 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6064 struct lu_device *dev)
6066 struct osd_device *osd = osd_dev(dev);
6070 if (osd->od_quota_slave != NULL) {
6071 /* set up quota slave objects */
6072 result = qsd_prepare(env, osd->od_quota_slave);
6077 result = osd_fid_init(env, osd);
6082 int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6083 struct lu_fid *fid, struct md_op_data *op_data)
6085 struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6087 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6090 static const struct lu_object_operations osd_lu_obj_ops = {
6091 .loo_object_init = osd_object_init,
6092 .loo_object_delete = osd_object_delete,
6093 .loo_object_release = osd_object_release,
6094 .loo_object_free = osd_object_free,
6095 .loo_object_print = osd_object_print,
6096 .loo_object_invariant = osd_object_invariant
6099 const struct lu_device_operations osd_lu_ops = {
6100 .ldo_object_alloc = osd_object_alloc,
6101 .ldo_process_config = osd_process_config,
6102 .ldo_recovery_complete = osd_recovery_complete,
6103 .ldo_prepare = osd_prepare,
6106 static const struct lu_device_type_operations osd_device_type_ops = {
6107 .ldto_init = osd_type_init,
6108 .ldto_fini = osd_type_fini,
6110 .ldto_start = osd_type_start,
6111 .ldto_stop = osd_type_stop,
6113 .ldto_device_alloc = osd_device_alloc,
6114 .ldto_device_free = osd_device_free,
6116 .ldto_device_init = osd_device_init,
6117 .ldto_device_fini = osd_device_fini
6120 struct lu_device_type osd_device_type = {
6121 .ldt_tags = LU_DEVICE_DT,
6122 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
6123 .ldt_ops = &osd_device_type_ops,
6124 .ldt_ctx_tags = LCT_LOCAL,
6128 * lprocfs legacy support.
6130 static struct obd_ops osd_obd_device_ops = {
6131 .o_owner = THIS_MODULE,
6132 .o_connect = osd_obd_connect,
6133 .o_disconnect = osd_obd_disconnect,
6134 .o_fid_alloc = osd_fid_alloc,
6137 static int __init osd_mod_init(void)
6143 rc = lu_kmem_init(ldiskfs_caches);
6147 rc = class_register_type(&osd_obd_device_ops, NULL, true,
6148 lprocfs_osd_module_vars,
6149 #ifndef HAVE_ONLY_PROCFS_SEQ
6152 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6154 lu_kmem_fini(ldiskfs_caches);
6158 static void __exit osd_mod_exit(void)
6160 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6161 lu_kmem_fini(ldiskfs_caches);
6164 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6165 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6166 MODULE_LICENSE("GPL");
6168 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);