4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osd/osd_handler.c
38 * Top-level entry points into osd module
40 * Author: Nikita Danilov <nikita@clusterfs.com>
41 * Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
44 #define DEBUG_SUBSYSTEM S_MDS
46 #include <linux/module.h>
48 /* LUSTRE_VERSION_CODE */
49 #include <lustre_ver.h>
50 /* prerequisite for linux/xattr.h */
51 #include <linux/types.h>
52 /* prerequisite for linux/xattr.h */
54 /* XATTR_{REPLACE,CREATE} */
55 #include <linux/xattr.h>
58 * struct OBD_{ALLOC,FREE}*()
61 #include <obd_support.h>
62 /* struct ptlrpc_thread */
63 #include <lustre_net.h>
64 #include <lustre_fid.h>
66 #include <lustre_param.h>
68 #include "osd_internal.h"
69 #include "osd_dynlocks.h"
71 /* llo_* api support */
72 #include <md_object.h>
73 #include <lustre_quota.h>
75 #include <ldiskfs/xattr.h>
78 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
79 "ldiskfs with parallel directory operations");
81 int ldiskfs_track_declares_assert;
82 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
83 "LBUG during tracking of declares");
85 /* Slab to allocate dynlocks */
86 struct kmem_cache *dynlock_cachep;
88 static struct lu_kmem_descr ldiskfs_caches[] = {
90 .ckd_cache = &dynlock_cachep,
91 .ckd_name = "dynlock_cache",
92 .ckd_size = sizeof(struct dynlock_handle)
99 static const char dot[] = ".";
100 static const char dotdot[] = "..";
101 static const char remote_obj_dir[] = "REM_OBJ_DIR";
103 static const struct lu_object_operations osd_lu_obj_ops;
104 static const struct dt_object_operations osd_obj_ops;
105 static const struct dt_object_operations osd_obj_ea_ops;
106 static const struct dt_object_operations osd_obj_otable_it_ops;
107 static const struct dt_index_operations osd_index_iam_ops;
108 static const struct dt_index_operations osd_index_ea_ops;
110 int osd_trans_declare_op2rb[] = {
111 [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
112 [OSD_OT_PUNCH] = OSD_OT_MAX,
113 [OSD_OT_XATTR_SET] = OSD_OT_XATTR_SET,
114 [OSD_OT_CREATE] = OSD_OT_DESTROY,
115 [OSD_OT_DESTROY] = OSD_OT_CREATE,
116 [OSD_OT_REF_ADD] = OSD_OT_REF_DEL,
117 [OSD_OT_REF_DEL] = OSD_OT_REF_ADD,
118 [OSD_OT_WRITE] = OSD_OT_WRITE,
119 [OSD_OT_INSERT] = OSD_OT_DELETE,
120 [OSD_OT_DELETE] = OSD_OT_INSERT,
121 [OSD_OT_UPDATE] = OSD_OT_MAX,
122 [OSD_OT_QUOTA] = OSD_OT_MAX,
125 static int osd_has_index(const struct osd_object *obj)
127 return obj->oo_dt.do_index_ops != NULL;
130 static int osd_object_invariant(const struct lu_object *l)
132 return osd_invariant(osd_obj(l));
136 * Concurrency: doesn't matter
140 * Concurrency: doesn't matter
142 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
144 struct osd_thread_info *oti = osd_oti_get(env);
145 return oti->oti_w_locks > 0 && o->oo_owner == env;
149 * Concurrency: doesn't access mutable data
151 static int osd_root_get(const struct lu_env *env,
152 struct dt_device *dev, struct lu_fid *f)
154 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
159 * OSD object methods.
163 * Concurrency: no concurrent access is possible that early in object
166 static struct lu_object *osd_object_alloc(const struct lu_env *env,
167 const struct lu_object_header *hdr,
170 struct osd_object *mo;
176 l = &mo->oo_dt.do_lu;
177 dt_object_init(&mo->oo_dt, NULL, d);
178 mo->oo_dt.do_ops = &osd_obj_ea_ops;
179 l->lo_ops = &osd_lu_obj_ops;
180 init_rwsem(&mo->oo_sem);
181 init_rwsem(&mo->oo_ext_idx_sem);
182 spin_lock_init(&mo->oo_guard);
189 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
190 struct dentry *dentry, struct lustre_mdt_attrs *lma)
194 CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
195 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
196 info->oti_mdt_attrs_old, LMA_OLD_SIZE);
198 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
199 memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
201 lustre_lma_swab(lma);
202 /* Check LMA compatibility */
203 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
204 CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
205 "for fid = "DFID", ino = %lu\n",
206 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
207 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
208 PFID(&lma->lma_self_fid), inode->i_ino);
211 } else if (rc == 0) {
219 * retrieve object from backend ext fs.
221 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
222 struct osd_inode_id *id)
224 struct inode *inode = NULL;
226 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
228 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
229 id->oii_ino, PTR_ERR(inode));
230 } else if (id->oii_gen != OSD_OII_NOGEN &&
231 inode->i_generation != id->oii_gen) {
232 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
233 "i_generation = %u\n",
234 id->oii_ino, id->oii_gen, inode->i_generation);
236 inode = ERR_PTR(-ESTALE);
237 } else if (inode->i_nlink == 0) {
238 /* due to parallel readdir and unlink,
239 * we can have dead inode here. */
240 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
242 inode = ERR_PTR(-ESTALE);
243 } else if (is_bad_inode(inode)) {
244 CWARN("%.16s: bad inode: ino = %u\n",
245 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
247 inode = ERR_PTR(-ENOENT);
249 if (id->oii_gen == OSD_OII_NOGEN)
250 osd_id_gen(id, inode->i_ino, inode->i_generation);
252 /* Do not update file c/mtime in ldiskfs.
253 * NB: we don't have any lock to protect this because we don't
254 * have reference on osd_object now, but contention with
255 * another lookup + attr_set can't happen in the tiny window
256 * between if (...) and set S_NOCMTIME. */
257 if (!(inode->i_flags & S_NOCMTIME))
258 inode->i_flags |= S_NOCMTIME;
263 static struct inode *
264 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
265 struct osd_inode_id *id, struct lu_fid *fid)
267 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
271 inode = osd_iget(info, dev, id);
275 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
277 *fid = lma->lma_self_fid;
278 } else if (rc == -ENODATA) {
279 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
280 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
282 lu_igif_build(fid, inode->i_ino, inode->i_generation);
290 static struct inode *osd_iget_check(struct osd_thread_info *info,
291 struct osd_device *dev,
292 const struct lu_fid *fid,
293 struct osd_inode_id *id,
300 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
303 if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) {
304 CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
313 if (is_bad_inode(inode)) {
316 CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
324 if (id->oii_gen != OSD_OII_NOGEN &&
325 inode->i_generation != id->oii_gen) {
328 CDEBUG(D_INODE, "unmatched inode: ino = %u, "
329 "oii_gen = %u, i_generation = %u\n",
330 id->oii_ino, id->oii_gen, inode->i_generation);
338 if (inode->i_nlink == 0) {
341 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
351 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
353 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
354 /* XXX: There are some possible cases:
356 * Backup/restore caused the OI invalid.
358 * Someone unlinked the object but NOT removed
359 * the OI mapping, such as mount target device
360 * as ldiskfs, and modify something directly.
362 * Someone just removed the object between the
363 * former oi_lookup and the iget. It is normal.
364 * 4. Other failure cases.
366 * Generally, when the device is mounted, it will
367 * auto check whether the system is restored from
368 * file-level backup or not. We trust such detect
369 * to distinguish the 1st case from the 2nd case. */
371 if (!IS_ERR(inode) && inode->i_generation != 0 &&
372 inode->i_generation == id->oii_gen)
378 if (id->oii_gen == OSD_OII_NOGEN)
379 osd_id_gen(id, inode->i_ino, inode->i_generation);
381 /* Do not update file c/mtime in ldiskfs.
382 * NB: we don't have any lock to protect this because we don't
383 * have reference on osd_object now, but contention with
384 * another lookup + attr_set can't happen in the tiny window
385 * between if (...) and set S_NOCMTIME. */
386 if (!(inode->i_flags & S_NOCMTIME))
387 inode->i_flags |= S_NOCMTIME;
404 * \retval +v: new filter_fid, does not contain self-fid
405 * \retval 0: filter_fid_old, contains self-fid
406 * \retval -v: other failure cases
408 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
409 struct dentry *dentry, struct lu_fid *fid)
411 struct filter_fid_old *ff = &info->oti_ff;
412 struct ost_id *ostid = &info->oti_ostid;
415 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
416 if (rc == sizeof(*ff)) {
418 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
419 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
420 /* XXX: should use real OST index in the future. LU-3569 */
421 ostid_to_fid(fid, ostid, 0);
422 } else if (rc == sizeof(struct filter_fid)) {
424 } else if (rc >= 0) {
431 static int osd_lma_self_repair(struct osd_thread_info *info,
432 struct osd_device *osd, struct inode *inode,
433 const struct lu_fid *fid, __u32 compat)
438 LASSERT(current->journal_info == NULL);
440 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
441 osd_dto_credits_noquota[DTO_XATTR_SET]);
444 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
449 rc = osd_ea_fid_set(info, inode, fid, compat, 0);
451 CWARN("%s: cannot self repair the LMA: rc = %d\n",
453 ldiskfs_journal_stop(jh);
457 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
459 struct osd_thread_info *info = osd_oti_get(env);
460 struct osd_device *osd = osd_obj2dev(obj);
461 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
462 struct inode *inode = obj->oo_inode;
463 struct dentry *dentry = &info->oti_obj_dentry;
464 struct lu_fid *fid = NULL;
465 const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
469 CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
470 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
471 info->oti_mdt_attrs_old, LMA_OLD_SIZE);
472 if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
473 fid = &lma->lma_self_fid;
474 rc = osd_get_idif(info, inode, dentry, fid);
475 if ((rc > 0) || (rc == -ENODATA && osd->od_lma_self_repair)) {
476 /* For the given OST-object, if it has neither LMA nor
477 * FID in XATTR_NAME_FID, then the given FID (which is
478 * contained in the @obj, from client RPC for locating
479 * the OST-object) is trusted. We use it to generate
481 osd_lma_self_repair(info, osd, inode, rfid,
482 fid_is_on_ost(info, osd, fid, OI_CHECK_FLD) ?
483 LMAC_FID_ON_OST : 0);
488 if (unlikely(rc == -ENODATA))
496 lustre_lma_swab(lma);
497 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
498 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
499 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
500 "fid = "DFID", ino = %lu\n", osd_name(osd),
501 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
502 PFID(rfid), inode->i_ino);
504 } else if (!(lma->lma_compat & LMAC_NOT_IN_OI)) {
505 fid = &lma->lma_self_fid;
509 if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
510 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
511 struct ost_id *oi = &info->oti_ostid;
512 struct lu_fid *fid1 = &info->oti_fid3;
513 __u32 idx = fid_idif_ost_idx(rfid);
515 /* For old IDIF, the OST index is not part of the IDIF,
516 * Means that different OSTs may have the same IDIFs.
517 * Under such case, we need to make some compatible
518 * check to make sure to trigger OI scrub properly. */
519 if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
520 /* Given @rfid is new, LMA is old. */
521 fid_to_ostid(fid, oi);
522 ostid_to_fid(fid1, oi, idx);
523 if (lu_fid_eq(fid1, rfid)) {
524 if (osd->od_lma_self_repair)
525 osd_lma_self_repair(info, osd,
533 CDEBUG(D_INODE, "%s: FID "DFID" != self_fid "DFID"\n",
534 osd_name(osd), PFID(rfid), PFID(fid));
541 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
542 const struct lu_fid *fid,
543 const struct lu_object_conf *conf)
545 struct osd_thread_info *info;
546 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
547 struct osd_device *dev;
548 struct osd_idmap_cache *oic;
549 struct osd_inode_id *id;
551 struct osd_scrub *scrub;
552 struct scrub_file *sf;
556 bool triggered = false;
559 LINVRNT(osd_invariant(obj));
560 LASSERT(obj->oo_inode == NULL);
561 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
564 scrub = &dev->od_scrub;
565 sf = &scrub->os_file;
566 info = osd_oti_get(env);
568 oic = &info->oti_cache;
570 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
573 /* For the object is created as locking anchor, or for the object to
574 * be created on disk. No need to osd_oi_lookup() at here because FID
575 * shouldn't never be re-used, if it's really a duplicate FID from
576 * unexpected reason, we should be able to detect it later by calling
577 * do_create->osd_oi_insert(). */
578 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
579 GOTO(out, result = 0);
581 /* Search order: 1. per-thread cache. */
582 if (lu_fid_eq(fid, &oic->oic_fid) &&
583 likely(oic->oic_dev == dev)) {
589 if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
590 /* Search order: 2. OI scrub pending list. */
591 result = osd_oii_lookup(dev, fid, id);
596 /* Search order: 3. OI files. */
597 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
598 if (result == -ENOENT) {
599 if (!fid_is_norm(fid) ||
600 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
601 !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
603 GOTO(out, result = 0);
614 inode = osd_iget_check(info, dev, fid, id, in_oi);
616 result = PTR_ERR(inode);
617 if (result == -ENOENT || result == -ESTALE) {
619 fid_zero(&oic->oic_fid);
621 GOTO(out, result = -ENOENT);
622 } else if (result == -EREMCHG) {
626 fid_zero(&oic->oic_fid);
628 if (unlikely(triggered))
629 GOTO(out, result = saved);
632 if (thread_is_running(&scrub->os_thread)) {
633 result = -EINPROGRESS;
634 } else if (!dev->od_noscrub) {
635 result = osd_scrub_start(dev);
636 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
637 "for "DFID", rc = %d [1]\n",
638 osd_name(dev), PFID(fid),result);
639 if (result == 0 || result == -EALREADY)
640 result = -EINPROGRESS;
645 /* We still have chance to get the valid inode: for the
646 * object which is referenced by remote name entry, the
647 * object on the local MDT will be linked under the dir
648 * of "/REMOTE_PARENT_DIR" with its FID string as name.
650 * We do not know whether the object for the given FID
651 * is referenced by some remote name entry or not, and
652 * especially for DNE II, a multiple-linked object may
653 * have many name entries reside on many MDTs.
655 * To simplify the operation, OSD will not distinguish
656 * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
657 * only happened for the RPC from other MDT during the
658 * OI scrub, or for the client side RPC with FID only,
659 * such as FID to path, or from old connected client. */
661 result = osd_lookup_in_remote_parent(info, dev,
674 obj->oo_inode = inode;
675 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
677 result = osd_check_lma(env, obj);
680 obj->oo_inode = NULL;
681 if (result == -EREMCHG) {
683 result = osd_oi_lookup(info, dev, fid, id,
686 fid_zero(&oic->oic_fid);
697 obj->oo_compat_dot_created = 1;
698 obj->oo_compat_dotdot_created = 1;
700 if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
701 GOTO(out, result = 0);
703 LASSERT(obj->oo_hl_head == NULL);
704 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
705 if (obj->oo_hl_head == NULL) {
706 obj->oo_inode = NULL;
708 GOTO(out, result = -ENOMEM);
710 GOTO(out, result = 0);
713 LINVRNT(osd_invariant(obj));
718 * Concurrency: shouldn't matter.
720 static void osd_object_init0(struct osd_object *obj)
722 LASSERT(obj->oo_inode != NULL);
723 obj->oo_dt.do_body_ops = &osd_body_ops;
724 obj->oo_dt.do_lu.lo_header->loh_attr |=
725 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
729 * Concurrency: no concurrent access is possible that early in object
732 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
733 const struct lu_object_conf *conf)
735 struct osd_object *obj = osd_obj(l);
738 LINVRNT(osd_invariant(obj));
740 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
741 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
742 l->lo_header->loh_attr |= LOHA_EXISTS;
746 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
747 obj->oo_dt.do_body_ops = &osd_body_ops_new;
748 if (result == 0 && obj->oo_inode != NULL)
749 osd_object_init0(obj);
751 LINVRNT(osd_invariant(obj));
756 * Concurrency: no concurrent access is possible that late in object
759 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
761 struct osd_object *obj = osd_obj(l);
763 LINVRNT(osd_invariant(obj));
765 dt_object_fini(&obj->oo_dt);
766 if (obj->oo_hl_head != NULL)
767 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
772 * Concurrency: no concurrent access is possible that late in object
775 static void osd_index_fini(struct osd_object *o)
777 struct iam_container *bag;
779 if (o->oo_dir != NULL) {
780 bag = &o->oo_dir->od_container;
781 if (o->oo_inode != NULL) {
782 if (bag->ic_object == o->oo_inode)
783 iam_container_fini(bag);
785 OBD_FREE_PTR(o->oo_dir);
791 * Concurrency: no concurrent access is possible that late in object
792 * life-cycle (for all existing callers, that is. New callers have to provide
793 * their own locking.)
795 static int osd_inode_unlinked(const struct inode *inode)
797 return inode->i_nlink == 0;
801 OSD_TXN_OI_DELETE_CREDITS = 20,
802 OSD_TXN_INODE_DELETE_CREDITS = 20
809 #if OSD_THANDLE_STATS
811 * Set time when the handle is allocated
813 static void osd_th_alloced(struct osd_thandle *oth)
815 oth->oth_alloced = cfs_time_current();
819 * Set time when the handle started
821 static void osd_th_started(struct osd_thandle *oth)
823 oth->oth_started = cfs_time_current();
827 * Helper function to convert time interval to microseconds packed in
830 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
834 cfs_duration_usec(cfs_time_sub(end, start), &val);
835 return val.tv_sec * 1000000 + val.tv_usec;
839 * Check whether the we deal with this handle for too long.
841 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
842 cfs_time_t alloced, cfs_time_t started,
845 cfs_time_t now = cfs_time_current();
847 LASSERT(dev != NULL);
849 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
850 interval_to_usec(alloced, started));
851 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
852 interval_to_usec(started, closed));
853 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
854 interval_to_usec(closed, now));
856 if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
857 CWARN("transaction handle %p was open for too long: "
859 "alloced "CFS_TIME_T" ,"
860 "started "CFS_TIME_T" ,"
861 "closed "CFS_TIME_T"\n",
862 oth, now, alloced, started, closed);
863 libcfs_debug_dumpstack(NULL);
867 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
869 cfs_time_t __closed = cfs_time_current(); \
870 cfs_time_t __alloced = oth->oth_alloced; \
871 cfs_time_t __started = oth->oth_started; \
874 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
877 #else /* OSD_THANDLE_STATS */
879 #define osd_th_alloced(h) do {} while(0)
880 #define osd_th_started(h) do {} while(0)
881 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
883 #endif /* OSD_THANDLE_STATS */
886 * Concurrency: doesn't access mutable data.
888 static int osd_param_is_not_sane(const struct osd_device *dev,
889 const struct thandle *th)
891 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
893 return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers;
897 * Concurrency: shouldn't matter.
899 static void osd_trans_commit_cb(struct super_block *sb,
900 struct ldiskfs_journal_cb_entry *jcb, int error)
902 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
903 struct thandle *th = &oh->ot_super;
904 struct lu_device *lud = &th->th_dev->dd_lu_dev;
905 struct dt_txn_commit_cb *dcb, *tmp;
907 LASSERT(oh->ot_handle == NULL);
910 CERROR("transaction @0x%p commit error: %d\n", th, error);
912 dt_txn_hook_commit(th);
914 /* call per-transaction callbacks if any */
915 cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
916 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
917 "commit callback entry: magic=%x name='%s'\n",
918 dcb->dcb_magic, dcb->dcb_name);
919 cfs_list_del_init(&dcb->dcb_linkage);
920 dcb->dcb_func(NULL, th, dcb, error);
923 lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
927 lu_context_exit(&th->th_ctx);
928 lu_context_fini(&th->th_ctx);
932 static struct thandle *osd_trans_create(const struct lu_env *env,
935 struct osd_thread_info *oti = osd_oti_get(env);
936 struct osd_iobuf *iobuf = &oti->oti_iobuf;
937 struct osd_thandle *oh;
941 /* on pending IO in this thread should left from prev. request */
942 LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
944 th = ERR_PTR(-ENOMEM);
945 OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
947 oh->ot_quota_trans = &oti->oti_quota_trans;
948 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
952 th->th_tags = LCT_TX_HANDLE;
954 atomic_set(&th->th_refc, 1);
955 th->th_alloc_size = sizeof(*oh);
956 oti->oti_dev = osd_dt_dev(d);
957 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
960 memset(oti->oti_declare_ops, 0,
961 sizeof(oti->oti_declare_ops));
962 memset(oti->oti_declare_ops_rb, 0,
963 sizeof(oti->oti_declare_ops_rb));
964 memset(oti->oti_declare_ops_cred, 0,
965 sizeof(oti->oti_declare_ops_cred));
966 oti->oti_rollback = false;
972 * Concurrency: shouldn't matter.
974 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
977 struct osd_thread_info *oti = osd_oti_get(env);
978 struct osd_device *dev = osd_dt_dev(d);
980 struct osd_thandle *oh;
985 LASSERT(current->journal_info == NULL);
987 oh = container_of0(th, struct osd_thandle, ot_super);
989 LASSERT(oh->ot_handle == NULL);
991 rc = dt_txn_hook_start(env, d, th);
995 if (unlikely(osd_param_is_not_sane(dev, th))) {
996 static unsigned long last_printed;
997 static int last_credits;
999 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1000 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1002 osd_journal(dev)->j_max_transaction_buffers);
1003 CWARN(" create: %u/%u, destroy: %u/%u\n",
1004 oti->oti_declare_ops[OSD_OT_CREATE],
1005 oti->oti_declare_ops_cred[OSD_OT_CREATE],
1006 oti->oti_declare_ops[OSD_OT_DESTROY],
1007 oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
1008 CWARN(" attr_set: %u/%u, xattr_set: %u/%u\n",
1009 oti->oti_declare_ops[OSD_OT_ATTR_SET],
1010 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1011 oti->oti_declare_ops[OSD_OT_XATTR_SET],
1012 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
1013 CWARN(" write: %u/%u, punch: %u/%u, quota %u/%u\n",
1014 oti->oti_declare_ops[OSD_OT_WRITE],
1015 oti->oti_declare_ops_cred[OSD_OT_WRITE],
1016 oti->oti_declare_ops[OSD_OT_PUNCH],
1017 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1018 oti->oti_declare_ops[OSD_OT_QUOTA],
1019 oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
1020 CWARN(" insert: %u/%u, delete: %u/%u\n",
1021 oti->oti_declare_ops[OSD_OT_INSERT],
1022 oti->oti_declare_ops_cred[OSD_OT_INSERT],
1023 oti->oti_declare_ops[OSD_OT_DELETE],
1024 oti->oti_declare_ops_cred[OSD_OT_DELETE]);
1025 CWARN(" ref_add: %u/%u, ref_del: %u/%u\n",
1026 oti->oti_declare_ops[OSD_OT_REF_ADD],
1027 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1028 oti->oti_declare_ops[OSD_OT_REF_DEL],
1029 oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
1031 if (last_credits != oh->ot_credits &&
1032 time_after(jiffies, last_printed + 60 * HZ)) {
1033 libcfs_debug_dumpstack(NULL);
1034 last_credits = oh->ot_credits;
1035 last_printed = jiffies;
1037 /* XXX Limit the credits to 'max_transaction_buffers', and
1038 * let the underlying filesystem to catch the error if
1039 * we really need so many credits.
1041 * This should be removed when we can calculate the
1042 * credits precisely. */
1043 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
1047 * XXX temporary stuff. Some abstraction layer should
1050 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1054 LASSERT(oti->oti_txns == 0);
1055 lu_context_init(&th->th_ctx, th->th_tags);
1056 lu_context_enter(&th->th_ctx);
1058 lu_device_get(&d->dd_lu_dev);
1059 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1070 static int osd_seq_exists(const struct lu_env *env,
1071 struct osd_device *osd, obd_seq seq)
1073 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1074 struct seq_server_site *ss = osd_seq_site(osd);
1081 rc = osd_fld_lookup(env, osd, seq, range);
1084 CERROR("%s: can't lookup FLD sequence "LPX64
1085 ": rc = %d\n", osd_name(osd), seq, rc);
1089 RETURN(ss->ss_node_id == range->lsr_index);
1093 * Concurrency: shouldn't matter.
1095 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1099 struct osd_thandle *oh;
1100 struct osd_thread_info *oti = osd_oti_get(env);
1101 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1102 struct qsd_instance *qsd = oti->oti_dev->od_quota_slave;
1103 struct lquota_trans *qtrans;
1106 oh = container_of0(th, struct osd_thandle, ot_super);
1108 qtrans = oh->ot_quota_trans;
1109 oh->ot_quota_trans = NULL;
1111 if (oh->ot_handle != NULL) {
1112 handle_t *hdl = oh->ot_handle;
1115 * add commit callback
1116 * notice we don't do this in osd_trans_start()
1117 * as underlying transaction can change during truncate
1119 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1122 LASSERT(oti->oti_txns == 1);
1124 rc = dt_txn_hook_stop(env, th);
1126 CERROR("Failure in transaction hook: %d\n", rc);
1128 /* hook functions might modify th_sync */
1129 hdl->h_sync = th->th_sync;
1131 oh->ot_handle = NULL;
1132 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
1133 rc = ldiskfs_journal_stop(hdl));
1135 CERROR("Failure to stop transaction: %d\n", rc);
1137 thandle_put(&oh->ot_super);
1140 /* inform the quota slave device that the transaction is stopping */
1141 qsd_op_end(env, qsd, qtrans);
1143 /* as we want IO to journal and data IO be concurrent, we don't block
1144 * awaiting data IO completion in osd_do_bio(), instead we wait here
1145 * once transaction is submitted to the journal. all reqular requests
1146 * don't do direct IO (except read/write), thus this wait_event becomes
1149 * IMPORTANT: we have to wait till any IO submited by the thread is
1150 * completed otherwise iobuf may be corrupted by different request
1152 wait_event(iobuf->dr_wait,
1153 atomic_read(&iobuf->dr_numreqs) == 0);
1155 rc = iobuf->dr_error;
1160 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1162 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1165 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1166 LASSERT(&dcb->dcb_func != NULL);
1167 cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1173 * Called just before object is freed. Releases all resources except for
1174 * object itself (that is released by osd_object_free()).
1176 * Concurrency: no concurrent access is possible that late in object
1179 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1181 struct osd_object *obj = osd_obj(l);
1182 struct inode *inode = obj->oo_inode;
1184 LINVRNT(osd_invariant(obj));
1187 * If object is unlinked remove fid->ino mapping from object index.
1190 osd_index_fini(obj);
1191 if (inode != NULL) {
1192 struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave;
1193 qid_t uid = inode->i_uid;
1194 qid_t gid = inode->i_gid;
1197 obj->oo_inode = NULL;
1200 struct osd_thread_info *info = osd_oti_get(env);
1201 struct lquota_id_info *qi = &info->oti_qi;
1203 /* Release granted quota to master if necessary */
1204 qi->lqi_id.qid_uid = uid;
1205 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1207 qi->lqi_id.qid_uid = gid;
1208 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1214 * Concurrency: ->loo_object_release() is called under site spin-lock.
1216 static void osd_object_release(const struct lu_env *env,
1217 struct lu_object *l)
1222 * Concurrency: shouldn't matter.
1224 static int osd_object_print(const struct lu_env *env, void *cookie,
1225 lu_printer_t p, const struct lu_object *l)
1227 struct osd_object *o = osd_obj(l);
1228 struct iam_descr *d;
1230 if (o->oo_dir != NULL)
1231 d = o->oo_dir->od_container.ic_descr;
1234 return (*p)(env, cookie,
1235 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1237 o->oo_inode ? o->oo_inode->i_ino : 0UL,
1238 o->oo_inode ? o->oo_inode->i_generation : 0,
1239 d ? d->id_ops->id_name : "plain");
1242 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1245 * Concurrency: shouldn't matter.
1247 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1248 struct obd_statfs *sfs)
1250 struct osd_device *osd = osd_dt_dev(d);
1251 struct super_block *sb = osd_sb(osd);
1252 struct kstatfs *ksfs;
1255 if (unlikely(osd->od_mnt == NULL))
1256 return -EINPROGRESS;
1258 /* osd_lproc.c call this without env, allocate ksfs for that case */
1259 if (unlikely(env == NULL)) {
1260 OBD_ALLOC_PTR(ksfs);
1264 ksfs = &osd_oti_get(env)->oti_ksfs;
1267 spin_lock(&osd->od_osfs_lock);
1268 result = sb->s_op->statfs(sb->s_root, ksfs);
1269 if (likely(result == 0)) { /* N.B. statfs can't really fail */
1270 statfs_pack(sfs, ksfs);
1271 if (sb->s_flags & MS_RDONLY)
1272 sfs->os_state = OS_STATE_READONLY;
1275 spin_unlock(&osd->od_osfs_lock);
1277 if (unlikely(env == NULL))
1280 /* Reserve a small amount of space for local objects like last_rcvd,
1281 * llog, quota files, ... */
1282 if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1285 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1286 /** Take out metadata overhead for indirect blocks */
1287 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1294 * Estimate space needed for file creations. We assume the largest filename
1295 * which is 2^64 - 1, hence a filename of 20 chars.
1296 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1298 #ifdef __LDISKFS_DIR_REC_LEN
1299 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1301 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1305 * Concurrency: doesn't access mutable data.
1307 static void osd_conf_get(const struct lu_env *env,
1308 const struct dt_device *dev,
1309 struct dt_device_param *param)
1311 struct super_block *sb = osd_sb(osd_dt_dev(dev));
1315 * XXX should be taken from not-yet-existing fs abstraction layer.
1317 param->ddp_max_name_len = LDISKFS_NAME_LEN;
1318 param->ddp_max_nlink = LDISKFS_LINK_MAX;
1319 param->ddp_block_shift = sb->s_blocksize_bits;
1320 param->ddp_mount_type = LDD_MT_LDISKFS;
1321 param->ddp_maxbytes = sb->s_maxbytes;
1322 /* Overhead estimate should be fairly accurate, so we really take a tiny
1323 * error margin which also avoids fragmenting the filesystem too much */
1324 param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1325 /* inode are statically allocated, so per-inode space consumption
1326 * is the space consumed by the directory entry */
1327 param->ddp_inodespace = PER_OBJ_USAGE;
1328 /* per-fragment overhead to be used by the client code */
1329 param->ddp_grant_frag = 6 * LDISKFS_BLOCK_SIZE(sb);
1330 param->ddp_mntopts = 0;
1331 if (test_opt(sb, XATTR_USER))
1332 param->ddp_mntopts |= MNTOPT_USERXATTR;
1333 if (test_opt(sb, POSIX_ACL))
1334 param->ddp_mntopts |= MNTOPT_ACL;
1336 /* LOD might calculate the max stripe count based on max_ea_size,
1337 * so we need take account in the overhead as well,
1338 * xattr_header + magic + xattr_entry_head */
1339 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1340 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1342 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1343 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1344 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1348 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1352 * Concurrency: shouldn't matter.
1354 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1356 CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1357 return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1361 * Start commit for OSD device.
1363 * An implementation of dt_commit_async method for OSD device.
1364 * Asychronously starts underlayng fs sync and thereby a transaction
1367 * \param env environment
1368 * \param d dt device
1370 * \see dt_device_operations
1372 static int osd_commit_async(const struct lu_env *env,
1373 struct dt_device *d)
1375 struct super_block *s = osd_sb(osd_dt_dev(d));
1378 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1379 RETURN(s->s_op->sync_fs(s, 0));
1383 * Concurrency: shouldn't matter.
1386 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1388 struct super_block *sb = osd_sb(osd_dt_dev(d));
1389 struct block_device *dev = sb->s_bdev;
1390 #ifdef HAVE_DEV_SET_RDONLY
1391 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1394 int rc = -EOPNOTSUPP;
1398 #ifdef HAVE_DEV_SET_RDONLY
1399 CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1401 if (jdev && (jdev != dev)) {
1402 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1404 dev_set_rdonly(jdev);
1406 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1407 dev_set_rdonly(dev);
1409 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1410 osd_dt_dev(d)->od_svname, (long)dev, rc);
1416 * Concurrency: serialization provided by callers.
1418 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1419 int mode, unsigned long timeout, __u32 alg,
1420 struct lustre_capa_key *keys)
1422 struct osd_device *dev = osd_dt_dev(d);
1425 dev->od_fl_capa = mode;
1426 dev->od_capa_timeout = timeout;
1427 dev->od_capa_alg = alg;
1428 dev->od_capa_keys = keys;
1433 * Note: we do not count into QUOTA here.
1434 * If we mount with --data_journal we may need more.
1436 const int osd_dto_credits_noquota[DTO_NR] = {
1439 * INDEX_EXTRA_TRANS_BLOCKS(8) +
1440 * SINGLEDATA_TRANS_BLOCKS(8)
1441 * XXX Note: maybe iam need more, since iam have more level than
1444 [DTO_INDEX_INSERT] = 16,
1447 * just modify a single entry, probably merge few within a block
1449 [DTO_INDEX_DELETE] = 1,
1453 [DTO_INDEX_UPDATE] = 16,
1455 * 4(inode, inode bits, groups, GDT)
1456 * notice: OI updates are counted separately with DTO_INDEX_INSERT
1458 [DTO_OBJECT_CREATE] = 4,
1460 * 4(inode, inode bits, groups, GDT)
1461 * notice: OI updates are counted separately with DTO_INDEX_DELETE
1463 [DTO_OBJECT_DELETE] = 4,
1465 * Attr set credits (inode)
1467 [DTO_ATTR_SET_BASE] = 1,
1469 * Xattr set. The same as xattr of EXT3.
1470 * DATA_TRANS_BLOCKS(14)
1471 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1472 * are also counted in. Do not know why?
1474 [DTO_XATTR_SET] = 14,
1476 * credits for inode change during write.
1478 [DTO_WRITE_BASE] = 3,
1480 * credits for single block write.
1482 [DTO_WRITE_BLOCK] = 14,
1484 * Attr set credits for chown.
1485 * This is extra credits for setattr, and it is null without quota
1487 [DTO_ATTR_SET_CHOWN] = 0
1490 static const struct dt_device_operations osd_dt_ops = {
1491 .dt_root_get = osd_root_get,
1492 .dt_statfs = osd_statfs,
1493 .dt_trans_create = osd_trans_create,
1494 .dt_trans_start = osd_trans_start,
1495 .dt_trans_stop = osd_trans_stop,
1496 .dt_trans_cb_add = osd_trans_cb_add,
1497 .dt_conf_get = osd_conf_get,
1498 .dt_sync = osd_sync,
1500 .dt_commit_async = osd_commit_async,
1501 .dt_init_capa_ctxt = osd_init_capa_ctxt,
1504 static void osd_object_read_lock(const struct lu_env *env,
1505 struct dt_object *dt, unsigned role)
1507 struct osd_object *obj = osd_dt_obj(dt);
1508 struct osd_thread_info *oti = osd_oti_get(env);
1510 LINVRNT(osd_invariant(obj));
1512 LASSERT(obj->oo_owner != env);
1513 down_read_nested(&obj->oo_sem, role);
1515 LASSERT(obj->oo_owner == NULL);
1519 static void osd_object_write_lock(const struct lu_env *env,
1520 struct dt_object *dt, unsigned role)
1522 struct osd_object *obj = osd_dt_obj(dt);
1523 struct osd_thread_info *oti = osd_oti_get(env);
1525 LINVRNT(osd_invariant(obj));
1527 LASSERT(obj->oo_owner != env);
1528 down_write_nested(&obj->oo_sem, role);
1530 LASSERT(obj->oo_owner == NULL);
1531 obj->oo_owner = env;
1535 static void osd_object_read_unlock(const struct lu_env *env,
1536 struct dt_object *dt)
1538 struct osd_object *obj = osd_dt_obj(dt);
1539 struct osd_thread_info *oti = osd_oti_get(env);
1541 LINVRNT(osd_invariant(obj));
1543 LASSERT(oti->oti_r_locks > 0);
1545 up_read(&obj->oo_sem);
1548 static void osd_object_write_unlock(const struct lu_env *env,
1549 struct dt_object *dt)
1551 struct osd_object *obj = osd_dt_obj(dt);
1552 struct osd_thread_info *oti = osd_oti_get(env);
1554 LINVRNT(osd_invariant(obj));
1556 LASSERT(obj->oo_owner == env);
1557 LASSERT(oti->oti_w_locks > 0);
1559 obj->oo_owner = NULL;
1560 up_write(&obj->oo_sem);
1563 static int osd_object_write_locked(const struct lu_env *env,
1564 struct dt_object *dt)
1566 struct osd_object *obj = osd_dt_obj(dt);
1568 LINVRNT(osd_invariant(obj));
1570 return obj->oo_owner == env;
1573 static int capa_is_sane(const struct lu_env *env,
1574 struct osd_device *dev,
1575 struct lustre_capa *capa,
1576 struct lustre_capa_key *keys)
1578 struct osd_thread_info *oti = osd_oti_get(env);
1579 struct lustre_capa *tcapa = &oti->oti_capa;
1580 struct obd_capa *oc;
1584 oc = capa_lookup(dev->od_capa_hash, capa, 0);
1586 if (capa_is_expired(oc)) {
1587 DEBUG_CAPA(D_ERROR, capa, "expired");
1594 if (capa_is_expired_sec(capa)) {
1595 DEBUG_CAPA(D_ERROR, capa, "expired");
1599 spin_lock(&capa_lock);
1600 for (i = 0; i < 2; i++) {
1601 if (keys[i].lk_keyid == capa->lc_keyid) {
1602 oti->oti_capa_key = keys[i];
1606 spin_unlock(&capa_lock);
1609 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1613 rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1617 if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1618 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1622 oc = capa_add(dev->od_capa_hash, capa);
1628 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1629 struct lustre_capa *capa, __u64 opc)
1631 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1632 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
1633 struct lu_capainfo *lci;
1636 if (!osd->od_fl_capa)
1639 if (capa == BYPASS_CAPA)
1642 lci = lu_capainfo_get(env);
1643 if (unlikely(lci == NULL))
1646 if (lci->lci_auth == LC_ID_NONE)
1650 CERROR("%s: no capability provided for FID "DFID": rc = %d\n",
1651 osd_name(osd), PFID(fid), -EACCES);
1655 if (!lu_fid_eq(fid, &capa->lc_fid)) {
1656 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1661 if (!capa_opc_supported(capa, opc)) {
1662 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1666 rc = capa_is_sane(env, osd, capa, osd->od_capa_keys);
1668 DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc);
1675 static struct timespec *osd_inode_time(const struct lu_env *env,
1676 struct inode *inode, __u64 seconds)
1678 struct osd_thread_info *oti = osd_oti_get(env);
1679 struct timespec *t = &oti->oti_time;
1681 t->tv_sec = seconds;
1683 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1688 static void osd_inode_getattr(const struct lu_env *env,
1689 struct inode *inode, struct lu_attr *attr)
1691 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1692 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1693 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1696 attr->la_atime = LTIME_S(inode->i_atime);
1697 attr->la_mtime = LTIME_S(inode->i_mtime);
1698 attr->la_ctime = LTIME_S(inode->i_ctime);
1699 attr->la_mode = inode->i_mode;
1700 attr->la_size = i_size_read(inode);
1701 attr->la_blocks = inode->i_blocks;
1702 attr->la_uid = inode->i_uid;
1703 attr->la_gid = inode->i_gid;
1704 attr->la_flags = LDISKFS_I(inode)->i_flags;
1705 attr->la_nlink = inode->i_nlink;
1706 attr->la_rdev = inode->i_rdev;
1707 attr->la_blksize = 1 << inode->i_blkbits;
1708 attr->la_blkbits = inode->i_blkbits;
1711 static int osd_attr_get(const struct lu_env *env,
1712 struct dt_object *dt,
1713 struct lu_attr *attr,
1714 struct lustre_capa *capa)
1716 struct osd_object *obj = osd_dt_obj(dt);
1718 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1719 LINVRNT(osd_invariant(obj));
1721 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1724 spin_lock(&obj->oo_guard);
1725 osd_inode_getattr(env, obj->oo_inode, attr);
1726 spin_unlock(&obj->oo_guard);
1730 static int osd_declare_attr_set(const struct lu_env *env,
1731 struct dt_object *dt,
1732 const struct lu_attr *attr,
1733 struct thandle *handle)
1735 struct osd_thandle *oh;
1736 struct osd_object *obj;
1737 struct osd_thread_info *info = osd_oti_get(env);
1738 struct lquota_id_info *qi = &info->oti_qi;
1744 LASSERT(dt != NULL);
1745 LASSERT(handle != NULL);
1747 obj = osd_dt_obj(dt);
1748 LASSERT(osd_invariant(obj));
1750 oh = container_of0(handle, struct osd_thandle, ot_super);
1751 LASSERT(oh->ot_handle == NULL);
1753 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1754 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1756 if (attr == NULL || obj->oo_inode == NULL)
1759 bspace = obj->oo_inode->i_blocks;
1760 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1761 bspace = toqb(bspace);
1763 /* Changing ownership is always preformed by super user, it should not
1766 * We still need to call the osd_declare_qid() to calculate the journal
1767 * credits for updating quota accounting files and to trigger quota
1768 * space adjustment once the operation is completed.*/
1769 if ((attr->la_valid & LA_UID) != 0 &&
1770 attr->la_uid != obj->oo_inode->i_uid) {
1771 qi->lqi_type = USRQUOTA;
1773 /* inode accounting */
1774 qi->lqi_is_blk = false;
1776 /* one more inode for the new owner ... */
1777 qi->lqi_id.qid_uid = attr->la_uid;
1779 allocated = (attr->la_uid == 0) ? true : false;
1780 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1781 if (rc == -EDQUOT || rc == -EINPROGRESS)
1786 /* and one less inode for the current uid */
1787 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1789 rc = osd_declare_qid(env, oh, qi, true, NULL);
1790 if (rc == -EDQUOT || rc == -EINPROGRESS)
1795 /* block accounting */
1796 qi->lqi_is_blk = true;
1798 /* more blocks for the new owner ... */
1799 qi->lqi_id.qid_uid = attr->la_uid;
1800 qi->lqi_space = bspace;
1801 allocated = (attr->la_uid == 0) ? true : false;
1802 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1803 if (rc == -EDQUOT || rc == -EINPROGRESS)
1808 /* and finally less blocks for the current owner */
1809 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1810 qi->lqi_space = -bspace;
1811 rc = osd_declare_qid(env, oh, qi, true, NULL);
1812 if (rc == -EDQUOT || rc == -EINPROGRESS)
1818 if (attr->la_valid & LA_GID &&
1819 attr->la_gid != obj->oo_inode->i_gid) {
1820 qi->lqi_type = GRPQUOTA;
1822 /* inode accounting */
1823 qi->lqi_is_blk = false;
1825 /* one more inode for the new group owner ... */
1826 qi->lqi_id.qid_gid = attr->la_gid;
1828 allocated = (attr->la_gid == 0) ? true : false;
1829 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1830 if (rc == -EDQUOT || rc == -EINPROGRESS)
1835 /* and one less inode for the current gid */
1836 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1838 rc = osd_declare_qid(env, oh, qi, true, NULL);
1839 if (rc == -EDQUOT || rc == -EINPROGRESS)
1844 /* block accounting */
1845 qi->lqi_is_blk = true;
1847 /* more blocks for the new owner ... */
1848 qi->lqi_id.qid_gid = attr->la_gid;
1849 qi->lqi_space = bspace;
1850 allocated = (attr->la_gid == 0) ? true : false;
1851 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1852 if (rc == -EDQUOT || rc == -EINPROGRESS)
1857 /* and finally less blocks for the current owner */
1858 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1859 qi->lqi_space = -bspace;
1860 rc = osd_declare_qid(env, oh, qi, true, NULL);
1861 if (rc == -EDQUOT || rc == -EINPROGRESS)
1870 static int osd_inode_setattr(const struct lu_env *env,
1871 struct inode *inode, const struct lu_attr *attr)
1875 bits = attr->la_valid;
1877 if (bits & LA_ATIME)
1878 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
1879 if (bits & LA_CTIME)
1880 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
1881 if (bits & LA_MTIME)
1882 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
1883 if (bits & LA_SIZE) {
1884 LDISKFS_I(inode)->i_disksize = attr->la_size;
1885 i_size_write(inode, attr->la_size);
1889 /* OSD should not change "i_blocks" which is used by quota.
1890 * "i_blocks" should be changed by ldiskfs only. */
1891 if (bits & LA_BLOCKS)
1892 inode->i_blocks = attr->la_blocks;
1895 inode->i_mode = (inode->i_mode & S_IFMT) |
1896 (attr->la_mode & ~S_IFMT);
1898 inode->i_uid = attr->la_uid;
1900 inode->i_gid = attr->la_gid;
1901 if (bits & LA_NLINK)
1902 set_nlink(inode, attr->la_nlink);
1904 inode->i_rdev = attr->la_rdev;
1906 if (bits & LA_FLAGS) {
1907 /* always keep S_NOCMTIME */
1908 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1914 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1916 if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1917 (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1922 if (attr->la_valid & LA_UID)
1923 iattr.ia_valid |= ATTR_UID;
1924 if (attr->la_valid & LA_GID)
1925 iattr.ia_valid |= ATTR_GID;
1926 iattr.ia_uid = attr->la_uid;
1927 iattr.ia_gid = attr->la_gid;
1929 rc = ll_vfs_dq_transfer(inode, &iattr);
1931 CERROR("%s: quota transfer failed: rc = %d. Is quota "
1932 "enforcement enabled on the ldiskfs filesystem?",
1933 inode->i_sb->s_id, rc);
1940 static int osd_attr_set(const struct lu_env *env,
1941 struct dt_object *dt,
1942 const struct lu_attr *attr,
1943 struct thandle *handle,
1944 struct lustre_capa *capa)
1946 struct osd_object *obj = osd_dt_obj(dt);
1947 struct inode *inode;
1950 LASSERT(handle != NULL);
1951 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1952 LASSERT(osd_invariant(obj));
1954 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1957 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1959 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1960 struct osd_thread_info *oti = osd_oti_get(env);
1961 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
1962 struct lu_fid *fid1 = &oti->oti_fid;
1963 struct osd_inode_id *id = &oti->oti_id;
1964 struct iam_path_descr *ipd;
1965 struct iam_container *bag;
1966 struct osd_thandle *oh;
1969 fid_cpu_to_be(fid1, fid0);
1970 memset(id, 1, sizeof(*id));
1971 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1972 fid0)->oi_dir.od_container;
1973 ipd = osd_idx_ipd_get(env, bag);
1974 if (unlikely(ipd == NULL))
1977 oh = container_of0(handle, struct osd_thandle, ot_super);
1978 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1979 (const struct iam_rec *)id, ipd);
1980 osd_ipd_put(env, bag, ipd);
1981 return(rc > 0 ? 0 : rc);
1984 inode = obj->oo_inode;
1985 ll_vfs_dq_init(inode);
1987 rc = osd_quota_transfer(inode, attr);
1991 spin_lock(&obj->oo_guard);
1992 rc = osd_inode_setattr(env, inode, attr);
1993 spin_unlock(&obj->oo_guard);
1996 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2000 struct dentry *osd_child_dentry_get(const struct lu_env *env,
2001 struct osd_object *obj,
2002 const char *name, const int namelen)
2004 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2007 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2008 umode_t mode, struct dt_allocation_hint *hint,
2012 struct osd_device *osd = osd_obj2dev(obj);
2013 struct osd_thandle *oth;
2014 struct dt_object *parent = NULL;
2015 struct inode *inode;
2017 LINVRNT(osd_invariant(obj));
2018 LASSERT(obj->oo_inode == NULL);
2019 LASSERT(obj->oo_hl_head == NULL);
2021 if (S_ISDIR(mode) && ldiskfs_pdo) {
2022 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2023 if (obj->oo_hl_head == NULL)
2027 oth = container_of(th, struct osd_thandle, ot_super);
2028 LASSERT(oth->ot_handle->h_transaction != NULL);
2030 if (hint && hint->dah_parent)
2031 parent = hint->dah_parent;
2033 inode = ldiskfs_create_inode(oth->ot_handle,
2034 parent ? osd_dt_obj(parent)->oo_inode :
2035 osd_sb(osd)->s_root->d_inode,
2037 if (!IS_ERR(inode)) {
2038 /* Do not update file c/mtime in ldiskfs.
2039 * NB: don't need any lock because no contention at this
2041 inode->i_flags |= S_NOCMTIME;
2043 /* For new created object, it must be consistent,
2044 * and it is unnecessary to scrub against it. */
2045 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2046 obj->oo_inode = inode;
2049 if (obj->oo_hl_head != NULL) {
2050 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2051 obj->oo_hl_head = NULL;
2053 result = PTR_ERR(inode);
2055 LINVRNT(osd_invariant(obj));
2063 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2064 struct lu_attr *attr,
2065 struct dt_allocation_hint *hint,
2066 struct dt_object_format *dof,
2070 struct osd_thandle *oth;
2071 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2073 LASSERT(S_ISDIR(attr->la_mode));
2075 oth = container_of(th, struct osd_thandle, ot_super);
2076 LASSERT(oth->ot_handle->h_transaction != NULL);
2077 result = osd_mkfile(info, obj, mode, hint, th);
2082 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2083 struct lu_attr *attr,
2084 struct dt_allocation_hint *hint,
2085 struct dt_object_format *dof,
2089 struct osd_thandle *oth;
2090 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2092 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2094 LASSERT(S_ISREG(attr->la_mode));
2096 oth = container_of(th, struct osd_thandle, ot_super);
2097 LASSERT(oth->ot_handle->h_transaction != NULL);
2099 result = osd_mkfile(info, obj, mode, hint, th);
2101 LASSERT(obj->oo_inode != NULL);
2102 if (feat->dif_flags & DT_IND_VARKEY)
2103 result = iam_lvar_create(obj->oo_inode,
2104 feat->dif_keysize_max,
2106 feat->dif_recsize_max,
2109 result = iam_lfix_create(obj->oo_inode,
2110 feat->dif_keysize_max,
2112 feat->dif_recsize_max,
2119 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2120 struct lu_attr *attr,
2121 struct dt_allocation_hint *hint,
2122 struct dt_object_format *dof,
2125 LASSERT(S_ISREG(attr->la_mode));
2126 return osd_mkfile(info, obj, (attr->la_mode &
2127 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2130 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2131 struct lu_attr *attr,
2132 struct dt_allocation_hint *hint,
2133 struct dt_object_format *dof,
2136 LASSERT(S_ISLNK(attr->la_mode));
2137 return osd_mkfile(info, obj, (attr->la_mode &
2138 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2141 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2142 struct lu_attr *attr,
2143 struct dt_allocation_hint *hint,
2144 struct dt_object_format *dof,
2147 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2150 LINVRNT(osd_invariant(obj));
2151 LASSERT(obj->oo_inode == NULL);
2152 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2153 S_ISFIFO(mode) || S_ISSOCK(mode));
2155 result = osd_mkfile(info, obj, mode, hint, th);
2157 LASSERT(obj->oo_inode != NULL);
2159 * This inode should be marked dirty for i_rdev. Currently
2160 * that is done in the osd_attr_init().
2162 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2165 LINVRNT(osd_invariant(obj));
2169 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2171 struct dt_allocation_hint *hint,
2172 struct dt_object_format *dof,
2175 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2177 osd_obj_type_f result;
2193 result = osd_mk_index;
2204 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2205 struct dt_object *parent, struct dt_object *child,
2210 ah->dah_parent = parent;
2211 ah->dah_mode = child_mode;
2214 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2215 struct lu_attr *attr, struct dt_object_format *dof)
2217 struct inode *inode = obj->oo_inode;
2218 __u64 valid = attr->la_valid;
2221 attr->la_valid &= ~(LA_TYPE | LA_MODE);
2223 if (dof->dof_type != DFT_NODE)
2224 attr->la_valid &= ~LA_RDEV;
2225 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2226 attr->la_valid &= ~LA_ATIME;
2227 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2228 attr->la_valid &= ~LA_CTIME;
2229 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2230 attr->la_valid &= ~LA_MTIME;
2232 result = osd_quota_transfer(inode, attr);
2236 if (attr->la_valid != 0) {
2237 result = osd_inode_setattr(info->oti_env, inode, attr);
2239 * The osd_inode_setattr() should always succeed here. The
2240 * only error that could be returned is EDQUOT when we are
2241 * trying to change the UID or GID of the inode. However, this
2242 * should not happen since quota enforcement is no longer
2243 * enabled on ldiskfs (lquota takes care of it).
2245 LASSERTF(result == 0, "%d", result);
2246 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2249 attr->la_valid = valid;
2253 * Helper function for osd_object_create()
2255 * \retval 0, on success
2257 static int __osd_object_create(struct osd_thread_info *info,
2258 struct osd_object *obj, struct lu_attr *attr,
2259 struct dt_allocation_hint *hint,
2260 struct dt_object_format *dof,
2266 /* we drop umask so that permissions we pass are not affected */
2267 umask = current->fs->umask;
2268 current->fs->umask = 0;
2270 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2273 osd_attr_init(info, obj, attr, dof);
2274 osd_object_init0(obj);
2276 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2277 unlock_new_inode(obj->oo_inode);
2280 /* restore previous umask value */
2281 current->fs->umask = umask;
2287 * Helper function for osd_object_create()
2289 * \retval 0, on success
2291 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2292 const struct lu_fid *fid, struct thandle *th)
2294 struct osd_thread_info *info = osd_oti_get(env);
2295 struct osd_inode_id *id = &info->oti_id;
2296 struct osd_device *osd = osd_obj2dev(obj);
2297 struct osd_thandle *oh;
2299 LASSERT(obj->oo_inode != NULL);
2301 oh = container_of0(th, struct osd_thandle, ot_super);
2302 LASSERT(oh->ot_handle);
2304 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2305 return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2308 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2309 obd_seq seq, struct lu_seq_range *range)
2311 struct seq_server_site *ss = osd_seq_site(osd);
2313 if (fid_seq_is_idif(seq)) {
2314 fld_range_set_ost(range);
2315 range->lsr_index = idif_ost_idx(seq);
2319 if (!fid_seq_in_fldb(seq)) {
2320 fld_range_set_mdt(range);
2322 /* FIXME: If ss is NULL, it suppose not get lsr_index
2324 range->lsr_index = ss->ss_node_id;
2328 LASSERT(ss != NULL);
2329 fld_range_set_any(range);
2330 /* OSD will only do local fld lookup */
2331 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2335 * Concurrency: no external locking is necessary.
2337 static int osd_declare_object_create(const struct lu_env *env,
2338 struct dt_object *dt,
2339 struct lu_attr *attr,
2340 struct dt_allocation_hint *hint,
2341 struct dt_object_format *dof,
2342 struct thandle *handle)
2344 struct osd_thandle *oh;
2348 LASSERT(handle != NULL);
2350 oh = container_of0(handle, struct osd_thandle, ot_super);
2351 LASSERT(oh->ot_handle == NULL);
2353 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2354 osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2355 /* Reuse idle OI block may cause additional one OI block
2357 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2358 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2360 /* If this is directory, then we expect . and .. to be inserted as
2361 * well. The one directory block always needs to be created for the
2362 * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2363 * block), there is no danger of needing a tree for the first block.
2365 if (attr && S_ISDIR(attr->la_mode)) {
2366 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2367 osd_dto_credits_noquota[DTO_WRITE_BASE]);
2368 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2374 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2375 false, false, NULL, false);
2382 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2383 struct lu_attr *attr,
2384 struct dt_allocation_hint *hint,
2385 struct dt_object_format *dof,
2388 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2389 struct osd_object *obj = osd_dt_obj(dt);
2390 struct osd_thread_info *info = osd_oti_get(env);
2395 LINVRNT(osd_invariant(obj));
2396 LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2397 LASSERT(osd_write_locked(env, obj));
2398 LASSERT(th != NULL);
2400 if (unlikely(fid_is_acct(fid)))
2401 /* Quota files can't be created from the kernel any more,
2402 * 'tune2fs -O quota' will take care of creating them */
2405 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2406 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2408 result = __osd_object_create(info, obj, attr, hint, dof, th);
2410 result = __osd_oi_insert(env, obj, fid, th);
2412 LASSERT(ergo(result == 0,
2413 dt_object_exists(dt) && !dt_object_remote(dt)));
2415 LASSERT(osd_invariant(obj));
2420 * Called to destroy on-disk representation of the object
2422 * Concurrency: must be locked
2424 static int osd_declare_object_destroy(const struct lu_env *env,
2425 struct dt_object *dt,
2428 struct osd_object *obj = osd_dt_obj(dt);
2429 struct inode *inode = obj->oo_inode;
2430 struct osd_thandle *oh;
2434 oh = container_of0(th, struct osd_thandle, ot_super);
2435 LASSERT(oh->ot_handle == NULL);
2438 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2439 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2440 /* Recycle idle OI leaf may cause additional three OI blocks
2442 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2443 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2444 /* one less inode */
2445 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
2446 false, true, NULL, false);
2449 /* data to be truncated */
2450 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
2451 true, true, NULL, false);
2455 static int osd_object_destroy(const struct lu_env *env,
2456 struct dt_object *dt,
2459 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2460 struct osd_object *obj = osd_dt_obj(dt);
2461 struct inode *inode = obj->oo_inode;
2462 struct osd_device *osd = osd_obj2dev(obj);
2463 struct osd_thandle *oh;
2467 oh = container_of0(th, struct osd_thandle, ot_super);
2468 LASSERT(oh->ot_handle);
2470 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2472 if (unlikely(fid_is_acct(fid)))
2475 if (S_ISDIR(inode->i_mode)) {
2476 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2477 inode->i_nlink == 2);
2478 /* it will check/delete the inode from remote parent,
2479 * how to optimize it? unlink performance impaction XXX */
2480 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2481 if (result != 0 && result != -ENOENT) {
2482 CERROR("%s: delete inode "DFID": rc = %d\n",
2483 osd_name(osd), PFID(fid), result);
2485 spin_lock(&obj->oo_guard);
2487 spin_unlock(&obj->oo_guard);
2488 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2491 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2493 result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2496 /* XXX: add to ext3 orphan list */
2497 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2499 /* not needed in the cache anymore */
2500 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2506 * Put the fid into lustre_mdt_attrs, and then place the structure
2507 * inode's ea. This fid should not be altered during the life time
2510 * \retval +ve, on success
2511 * \retval -ve, on error
2513 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2515 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2516 const struct lu_fid *fid, __u32 compat, __u32 incompat)
2518 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2522 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2525 lustre_lma_init(lma, fid, compat, incompat);
2526 lustre_lma_swab(lma);
2528 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2530 /* LMA may already exist, but we need to check that all the
2531 * desired compat/incompat flags have been added. */
2532 if (unlikely(rc == -EEXIST)) {
2533 if (compat == 0 && incompat == 0)
2536 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2537 XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2542 lustre_lma_swab(lma);
2543 if (!(~lma->lma_compat & compat) &&
2544 !(~lma->lma_incompat & incompat))
2547 lma->lma_compat |= compat;
2548 lma->lma_incompat |= incompat;
2549 lustre_lma_swab(lma);
2550 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2551 sizeof(*lma), XATTR_REPLACE);
2558 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2559 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2560 * To have compatilibility with 1.8 ldiskfs driver we need to have
2561 * magic number at start of fid data.
2562 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2565 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2566 const struct dt_rec *fid)
2568 if (!fid_is_namespace_visible((const struct lu_fid *)fid) ||
2569 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2570 param->edp_magic = 0;
2574 param->edp_magic = LDISKFS_LUFID_MAGIC;
2575 param->edp_len = sizeof(struct lu_fid) + 1;
2576 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2580 * Try to read the fid from inode ea into dt_rec.
2582 * \param fid object fid.
2584 * \retval 0 on success
2586 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2587 __u32 ino, struct lu_fid *fid,
2588 struct osd_inode_id *id)
2590 struct osd_thread_info *info = osd_oti_get(env);
2591 struct inode *inode;
2594 osd_id_gen(id, ino, OSD_OII_NOGEN);
2595 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2597 RETURN(PTR_ERR(inode));
2603 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2605 struct inode *parent_dir,
2606 const struct dt_rec *dot_fid,
2607 const struct dt_rec *dot_dot_fid,
2608 struct osd_thandle *oth)
2610 struct ldiskfs_dentry_param *dot_ldp;
2611 struct ldiskfs_dentry_param *dot_dot_ldp;
2613 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2614 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2616 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2617 dot_ldp->edp_magic = 0;
2618 return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2619 dir, dot_ldp, dot_dot_ldp);
2623 * Create an local agent inode for remote entry
2625 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2626 struct osd_device *osd,
2627 struct osd_object *pobj,
2628 const struct lu_fid *fid,
2631 struct osd_thread_info *info = osd_oti_get(env);
2632 struct inode *local;
2633 struct osd_thandle *oh;
2638 oh = container_of(th, struct osd_thandle, ot_super);
2639 LASSERT(oh->ot_handle->h_transaction != NULL);
2641 /* FIXME: Insert index api needs to know the mode of
2642 * the remote object. Just use S_IFDIR for now */
2643 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR);
2644 if (IS_ERR(local)) {
2645 CERROR("%s: create local error %d\n", osd_name(osd),
2646 (int)PTR_ERR(local));
2650 /* Set special LMA flag for local agent inode */
2651 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2653 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2654 osd_name(osd), PFID(fid), rc);
2655 RETURN(ERR_PTR(rc));
2658 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2659 (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
2660 (const struct dt_rec *)fid, oh);
2662 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2663 osd_name(osd), PFID(fid), rc);
2664 RETURN(ERR_PTR(rc));
2671 * Delete local agent inode for remote entry
2673 static int osd_delete_local_agent_inode(const struct lu_env *env,
2674 struct osd_device *osd,
2675 const struct lu_fid *fid,
2676 __u32 ino, struct osd_thandle *oh)
2678 struct osd_thread_info *oti = osd_oti_get(env);
2679 struct osd_inode_id *id = &oti->oti_id;
2680 struct inode *inode;
2683 id->oii_ino = le32_to_cpu(ino);
2684 id->oii_gen = OSD_OII_NOGEN;
2685 inode = osd_iget(oti, osd, id);
2686 if (IS_ERR(inode)) {
2687 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2688 PFID(fid), id->oii_ino, id->oii_gen);
2689 RETURN(PTR_ERR(inode));
2693 mark_inode_dirty(inode);
2694 CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2695 osd_name(osd), PFID(fid), inode->i_ino);
2701 * OSD layer object create function for interoperability mode (b11826).
2702 * This is mostly similar to osd_object_create(). Only difference being, fid is
2703 * inserted into inode ea here.
2705 * \retval 0, on success
2706 * \retval -ve, on error
2708 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2709 struct lu_attr *attr,
2710 struct dt_allocation_hint *hint,
2711 struct dt_object_format *dof,
2714 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2715 struct osd_object *obj = osd_dt_obj(dt);
2716 struct osd_thread_info *info = osd_oti_get(env);
2721 LASSERT(osd_invariant(obj));
2722 LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2723 LASSERT(osd_write_locked(env, obj));
2724 LASSERT(th != NULL);
2726 if (unlikely(fid_is_acct(fid)))
2727 /* Quota files can't be created from the kernel any more,
2728 * 'tune2fs -O quota' will take care of creating them */
2731 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2732 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2734 result = __osd_object_create(info, obj, attr, hint, dof, th);
2736 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2737 fid_is_on_ost(info, osd_obj2dev(obj),
2738 fid, OI_CHECK_FLD) ?
2739 LMAC_FID_ON_OST : 0, 0);
2742 result = __osd_oi_insert(env, obj, fid, th);
2744 LASSERT(ergo(result == 0,
2745 dt_object_exists(dt) && !dt_object_remote(dt)));
2746 LINVRNT(osd_invariant(obj));
2750 static int osd_declare_object_ref_add(const struct lu_env *env,
2751 struct dt_object *dt,
2752 struct thandle *handle)
2754 struct osd_thandle *oh;
2756 /* it's possible that object doesn't exist yet */
2757 LASSERT(handle != NULL);
2759 oh = container_of0(handle, struct osd_thandle, ot_super);
2760 LASSERT(oh->ot_handle == NULL);
2762 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2763 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2769 * Concurrency: @dt is write locked.
2771 static int osd_object_ref_add(const struct lu_env *env,
2772 struct dt_object *dt, struct thandle *th)
2774 struct osd_object *obj = osd_dt_obj(dt);
2775 struct inode *inode = obj->oo_inode;
2776 struct osd_thandle *oh;
2779 LINVRNT(osd_invariant(obj));
2780 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2781 LASSERT(osd_write_locked(env, obj));
2782 LASSERT(th != NULL);
2784 oh = container_of0(th, struct osd_thandle, ot_super);
2785 LASSERT(oh->ot_handle != NULL);
2787 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2790 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2791 * (65000) subdirectories by storing "1" in i_nlink if the link count
2792 * would otherwise overflow. Directory tranversal tools understand
2793 * that (st_nlink == 1) indicates that the filesystem dose not track
2794 * hard links count on the directory, and will not abort subdirectory
2795 * scanning early once (st_nlink - 2) subdirs have been found.
2797 * This also has to properly handle the case of inodes with nlink == 0
2798 * in case they are being linked into the PENDING directory
2800 spin_lock(&obj->oo_guard);
2801 ldiskfs_inc_count(oh->ot_handle, inode);
2802 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2803 spin_unlock(&obj->oo_guard);
2805 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2806 LINVRNT(osd_invariant(obj));
2811 static int osd_declare_object_ref_del(const struct lu_env *env,
2812 struct dt_object *dt,
2813 struct thandle *handle)
2815 struct osd_thandle *oh;
2817 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2818 LASSERT(handle != NULL);
2820 oh = container_of0(handle, struct osd_thandle, ot_super);
2821 LASSERT(oh->ot_handle == NULL);
2823 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2824 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2830 * Concurrency: @dt is write locked.
2832 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2835 struct osd_object *obj = osd_dt_obj(dt);
2836 struct inode *inode = obj->oo_inode;
2837 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
2838 struct osd_thandle *oh;
2840 LINVRNT(osd_invariant(obj));
2841 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2842 LASSERT(osd_write_locked(env, obj));
2843 LASSERT(th != NULL);
2845 oh = container_of0(th, struct osd_thandle, ot_super);
2846 LASSERT(oh->ot_handle != NULL);
2848 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2850 spin_lock(&obj->oo_guard);
2851 /* That can be result of upgrade from old Lustre version and
2852 * applied only to local files. Just skip this ref_del call.
2853 * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2854 if (inode->i_nlink == 0) {
2855 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2856 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2857 ", maybe an upgraded file? (LU-3915)\n",
2858 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2859 spin_unlock(&obj->oo_guard);
2863 ldiskfs_dec_count(oh->ot_handle, inode);
2864 spin_unlock(&obj->oo_guard);
2866 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2867 LINVRNT(osd_invariant(obj));
2873 * Get the 64-bit version for an inode.
2875 static int osd_object_version_get(const struct lu_env *env,
2876 struct dt_object *dt, dt_obj_version_t *ver)
2878 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2880 CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2881 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2882 *ver = LDISKFS_I(inode)->i_fs_version;
2887 * Concurrency: @dt is read locked.
2889 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2890 struct lu_buf *buf, const char *name,
2891 struct lustre_capa *capa)
2893 struct osd_object *obj = osd_dt_obj(dt);
2894 struct inode *inode = obj->oo_inode;
2895 struct osd_thread_info *info = osd_oti_get(env);
2896 struct dentry *dentry = &info->oti_obj_dentry;
2898 /* version get is not real XATTR but uses xattr API */
2899 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2900 /* for version we are just using xattr API but change inode
2902 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2903 osd_object_version_get(env, dt, buf->lb_buf);
2904 return sizeof(dt_obj_version_t);
2907 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2908 LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2910 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2913 return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2917 static int osd_declare_xattr_set(const struct lu_env *env,
2918 struct dt_object *dt,
2919 const struct lu_buf *buf, const char *name,
2920 int fl, struct thandle *handle)
2922 struct osd_thandle *oh;
2925 LASSERT(handle != NULL);
2927 oh = container_of0(handle, struct osd_thandle, ot_super);
2928 LASSERT(oh->ot_handle == NULL);
2930 /* optimistic optimization: LMA is set first and usually fit inode */
2931 if (strcmp(name, XATTR_NAME_LMA) == 0) {
2932 if (dt_object_exists(dt))
2936 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2939 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
2940 struct super_block *sb = osd_sb(osd);
2941 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2942 if (buf && buf->lb_len > sb->s_blocksize) {
2943 credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2944 sb->s_blocksize_bits;
2948 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2954 * Set the 64-bit version for object
2956 static void osd_object_version_set(const struct lu_env *env,
2957 struct dt_object *dt,
2958 dt_obj_version_t *new_version)
2960 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2962 CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2963 *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2965 LDISKFS_I(inode)->i_fs_version = *new_version;
2966 /** Version is set after all inode operations are finished,
2967 * so we should mark it dirty here */
2968 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2972 * Concurrency: @dt is write locked.
2974 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2975 const struct lu_buf *buf, const char *name, int fl,
2976 struct thandle *handle, struct lustre_capa *capa)
2978 struct osd_object *obj = osd_dt_obj(dt);
2979 struct inode *inode = obj->oo_inode;
2980 struct osd_thread_info *info = osd_oti_get(env);
2984 LASSERT(handle != NULL);
2986 /* version set is not real XATTR */
2987 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2988 /* for version we are just using xattr API but change inode
2990 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2991 osd_object_version_set(env, dt, buf->lb_buf);
2992 return sizeof(dt_obj_version_t);
2995 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2998 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n",
2999 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3001 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3002 if (fl & LU_XATTR_REPLACE)
3003 fs_flags |= XATTR_REPLACE;
3005 if (fl & LU_XATTR_CREATE)
3006 fs_flags |= XATTR_CREATE;
3008 return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3013 * Concurrency: @dt is read locked.
3015 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3016 struct lu_buf *buf, struct lustre_capa *capa)
3018 struct osd_object *obj = osd_dt_obj(dt);
3019 struct inode *inode = obj->oo_inode;
3020 struct osd_thread_info *info = osd_oti_get(env);
3021 struct dentry *dentry = &info->oti_obj_dentry;
3023 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3024 LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
3026 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3029 dentry->d_inode = inode;
3030 dentry->d_sb = inode->i_sb;
3031 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3034 static int osd_declare_xattr_del(const struct lu_env *env,
3035 struct dt_object *dt, const char *name,
3036 struct thandle *handle)
3038 struct osd_thandle *oh;
3040 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3041 LASSERT(handle != NULL);
3043 oh = container_of0(handle, struct osd_thandle, ot_super);
3044 LASSERT(oh->ot_handle == NULL);
3046 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3047 osd_dto_credits_noquota[DTO_XATTR_SET]);
3053 * Concurrency: @dt is write locked.
3055 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3056 const char *name, struct thandle *handle,
3057 struct lustre_capa *capa)
3059 struct osd_object *obj = osd_dt_obj(dt);
3060 struct inode *inode = obj->oo_inode;
3061 struct osd_thread_info *info = osd_oti_get(env);
3062 struct dentry *dentry = &info->oti_obj_dentry;
3065 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3066 LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
3067 LASSERT(handle != NULL);
3069 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3072 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3074 ll_vfs_dq_init(inode);
3075 dentry->d_inode = inode;
3076 dentry->d_sb = inode->i_sb;
3077 rc = inode->i_op->removexattr(dentry, name);
3081 static struct obd_capa *osd_capa_get(const struct lu_env *env,
3082 struct dt_object *dt,
3083 struct lustre_capa *old, __u64 opc)
3085 struct osd_thread_info *info = osd_oti_get(env);
3086 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3087 struct osd_object *obj = osd_dt_obj(dt);
3088 struct osd_device *osd = osd_obj2dev(obj);
3089 struct lustre_capa_key *key = &info->oti_capa_key;
3090 struct lustre_capa *capa = &info->oti_capa;
3091 struct obd_capa *oc;
3092 struct lu_capainfo *lci;
3096 if (!osd->od_fl_capa)
3097 RETURN(ERR_PTR(-ENOENT));
3099 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3100 LINVRNT(osd_invariant(obj));
3102 /* renewal sanity check */
3103 if (old && osd_object_auth(env, dt, old, opc))
3104 RETURN(ERR_PTR(-EACCES));
3106 lci = lu_capainfo_get(env);
3107 if (unlikely(lci == NULL))
3108 RETURN(ERR_PTR(-ENOENT));
3110 switch (lci->lci_auth) {
3114 capa->lc_uid = obj->oo_inode->i_uid;
3115 capa->lc_gid = obj->oo_inode->i_gid;
3116 capa->lc_flags = LC_ID_PLAIN;
3118 case LC_ID_CONVERT: {
3121 s[0] = obj->oo_inode->i_uid;
3122 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
3123 s[2] = obj->oo_inode->i_gid;
3124 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
3125 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
3127 RETURN(ERR_PTR(rc));
3129 capa->lc_uid = ((__u64)d[1] << 32) | d[0];
3130 capa->lc_gid = ((__u64)d[3] << 32) | d[2];
3131 capa->lc_flags = LC_ID_CONVERT;
3135 RETURN(ERR_PTR(-EINVAL));
3138 capa->lc_fid = *fid;
3140 capa->lc_flags |= osd->od_capa_alg << 24;
3141 capa->lc_timeout = osd->od_capa_timeout;
3142 capa->lc_expiry = 0;
3144 oc = capa_lookup(osd->od_capa_hash, capa, 1);
3146 LASSERT(!capa_is_expired(oc));
3150 spin_lock(&capa_lock);
3151 *key = osd->od_capa_keys[1];
3152 spin_unlock(&capa_lock);
3154 capa->lc_keyid = key->lk_keyid;
3155 capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout;
3157 rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
3159 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
3160 RETURN(ERR_PTR(rc));
3163 oc = capa_add(osd->od_capa_hash, capa);
3167 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
3169 struct osd_object *obj = osd_dt_obj(dt);
3170 struct inode *inode = obj->oo_inode;
3171 struct osd_thread_info *info = osd_oti_get(env);
3172 struct dentry *dentry = &info->oti_obj_dentry;
3173 struct file *file = &info->oti_file;
3178 dentry->d_inode = inode;
3179 dentry->d_sb = inode->i_sb;
3180 file->f_dentry = dentry;
3181 file->f_mapping = inode->i_mapping;
3182 file->f_op = inode->i_fop;
3183 set_file_inode(file, inode);
3184 #ifndef HAVE_FILE_FSYNC_4ARGS
3185 mutex_lock(&inode->i_mutex);
3187 rc = do_fsync(file, 0);
3188 #ifndef HAVE_FILE_FSYNC_4ARGS
3189 mutex_unlock(&inode->i_mutex);
3194 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
3197 struct osd_object *obj = osd_dt_obj(dt);
3200 *data = (void *)obj->oo_inode;
3208 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3209 const struct dt_index_features *feat)
3211 struct iam_descr *descr;
3213 if (osd_object_is_root(o))
3214 return feat == &dt_directory_features;
3216 LASSERT(o->oo_dir != NULL);
3218 descr = o->oo_dir->od_container.ic_descr;
3219 if (feat == &dt_directory_features) {
3220 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3226 feat->dif_keysize_min <= descr->id_key_size &&
3227 descr->id_key_size <= feat->dif_keysize_max &&
3228 feat->dif_recsize_min <= descr->id_rec_size &&
3229 descr->id_rec_size <= feat->dif_recsize_max &&
3230 !(feat->dif_flags & (DT_IND_VARKEY |
3231 DT_IND_VARREC | DT_IND_NONUNQ)) &&
3232 ergo(feat->dif_flags & DT_IND_UPDATE,
3233 1 /* XXX check that object (and file system) is
3238 static int osd_iam_container_init(const struct lu_env *env,
3239 struct osd_object *obj,
3240 struct osd_directory *dir)
3242 struct iam_container *bag = &dir->od_container;
3245 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3249 result = iam_container_setup(bag);
3251 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3253 iam_container_fini(bag);
3260 * Concurrency: no external locking is necessary.
3262 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3263 const struct dt_index_features *feat)
3267 struct osd_object *obj = osd_dt_obj(dt);
3269 LINVRNT(osd_invariant(obj));
3271 if (osd_object_is_root(obj)) {
3272 dt->do_index_ops = &osd_index_ea_ops;
3274 } else if (feat == &dt_directory_features) {
3275 dt->do_index_ops = &osd_index_ea_ops;
3276 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3281 } else if (unlikely(feat == &dt_otable_features)) {
3282 dt->do_index_ops = &osd_otable_ops;
3284 } else if (unlikely(feat == &dt_acct_features)) {
3285 dt->do_index_ops = &osd_acct_index_ops;
3288 } else if (!osd_has_index(obj)) {
3289 struct osd_directory *dir;
3294 spin_lock(&obj->oo_guard);
3295 if (obj->oo_dir == NULL)
3299 * Concurrent thread allocated container data.
3302 spin_unlock(&obj->oo_guard);
3304 * Now, that we have container data, serialize its
3307 down_write(&obj->oo_ext_idx_sem);
3309 * recheck under lock.
3311 if (!osd_has_index(obj))
3312 result = osd_iam_container_init(env, obj, dir);
3315 up_write(&obj->oo_ext_idx_sem);
3323 if (result == 0 && skip_iam == 0) {
3324 if (!osd_iam_index_probe(env, obj, feat))
3327 LINVRNT(osd_invariant(obj));
3329 if (result == 0 && is_quota_glb_feat(feat) &&
3330 fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3331 result = osd_quota_migration(env, dt, feat);
3336 static int osd_otable_it_attr_get(const struct lu_env *env,
3337 struct dt_object *dt,
3338 struct lu_attr *attr,
3339 struct lustre_capa *capa)
3345 static const struct dt_object_operations osd_obj_ops = {
3346 .do_read_lock = osd_object_read_lock,
3347 .do_write_lock = osd_object_write_lock,
3348 .do_read_unlock = osd_object_read_unlock,
3349 .do_write_unlock = osd_object_write_unlock,
3350 .do_write_locked = osd_object_write_locked,
3351 .do_attr_get = osd_attr_get,
3352 .do_declare_attr_set = osd_declare_attr_set,
3353 .do_attr_set = osd_attr_set,
3354 .do_ah_init = osd_ah_init,
3355 .do_declare_create = osd_declare_object_create,
3356 .do_create = osd_object_create,
3357 .do_declare_destroy = osd_declare_object_destroy,
3358 .do_destroy = osd_object_destroy,
3359 .do_index_try = osd_index_try,
3360 .do_declare_ref_add = osd_declare_object_ref_add,
3361 .do_ref_add = osd_object_ref_add,
3362 .do_declare_ref_del = osd_declare_object_ref_del,
3363 .do_ref_del = osd_object_ref_del,
3364 .do_xattr_get = osd_xattr_get,
3365 .do_declare_xattr_set = osd_declare_xattr_set,
3366 .do_xattr_set = osd_xattr_set,
3367 .do_declare_xattr_del = osd_declare_xattr_del,
3368 .do_xattr_del = osd_xattr_del,
3369 .do_xattr_list = osd_xattr_list,
3370 .do_capa_get = osd_capa_get,
3371 .do_object_sync = osd_object_sync,
3372 .do_data_get = osd_data_get,
3376 * dt_object_operations for interoperability mode
3377 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3379 static const struct dt_object_operations osd_obj_ea_ops = {
3380 .do_read_lock = osd_object_read_lock,
3381 .do_write_lock = osd_object_write_lock,
3382 .do_read_unlock = osd_object_read_unlock,
3383 .do_write_unlock = osd_object_write_unlock,
3384 .do_write_locked = osd_object_write_locked,
3385 .do_attr_get = osd_attr_get,
3386 .do_declare_attr_set = osd_declare_attr_set,
3387 .do_attr_set = osd_attr_set,
3388 .do_ah_init = osd_ah_init,
3389 .do_declare_create = osd_declare_object_create,
3390 .do_create = osd_object_ea_create,
3391 .do_declare_destroy = osd_declare_object_destroy,
3392 .do_destroy = osd_object_destroy,
3393 .do_index_try = osd_index_try,
3394 .do_declare_ref_add = osd_declare_object_ref_add,
3395 .do_ref_add = osd_object_ref_add,
3396 .do_declare_ref_del = osd_declare_object_ref_del,
3397 .do_ref_del = osd_object_ref_del,
3398 .do_xattr_get = osd_xattr_get,
3399 .do_declare_xattr_set = osd_declare_xattr_set,
3400 .do_xattr_set = osd_xattr_set,
3401 .do_declare_xattr_del = osd_declare_xattr_del,
3402 .do_xattr_del = osd_xattr_del,
3403 .do_xattr_list = osd_xattr_list,
3404 .do_capa_get = osd_capa_get,
3405 .do_object_sync = osd_object_sync,
3406 .do_data_get = osd_data_get,
3409 static const struct dt_object_operations osd_obj_otable_it_ops = {
3410 .do_attr_get = osd_otable_it_attr_get,
3411 .do_index_try = osd_index_try,
3414 static int osd_index_declare_iam_delete(const struct lu_env *env,
3415 struct dt_object *dt,
3416 const struct dt_key *key,
3417 struct thandle *handle)
3419 struct osd_thandle *oh;
3421 oh = container_of0(handle, struct osd_thandle, ot_super);
3422 LASSERT(oh->ot_handle == NULL);
3424 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3425 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3431 * delete a (key, value) pair from index \a dt specified by \a key
3433 * \param dt osd index object
3434 * \param key key for index
3435 * \param rec record reference
3436 * \param handle transaction handler
3439 * \retval -ve failure
3442 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3443 const struct dt_key *key,
3444 struct thandle *handle,
3445 struct lustre_capa *capa)
3447 struct osd_thread_info *oti = osd_oti_get(env);
3448 struct osd_object *obj = osd_dt_obj(dt);
3449 struct osd_thandle *oh;
3450 struct iam_path_descr *ipd;
3451 struct iam_container *bag = &obj->oo_dir->od_container;
3456 LINVRNT(osd_invariant(obj));
3457 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3458 LASSERT(bag->ic_object == obj->oo_inode);
3459 LASSERT(handle != NULL);
3461 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3464 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3466 ipd = osd_idx_ipd_get(env, bag);
3467 if (unlikely(ipd == NULL))
3470 oh = container_of0(handle, struct osd_thandle, ot_super);
3471 LASSERT(oh->ot_handle != NULL);
3472 LASSERT(oh->ot_handle->h_transaction != NULL);
3474 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3475 /* swab quota uid/gid provided by caller */
3476 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3477 key = (const struct dt_key *)&oti->oti_quota_id;
3480 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3481 osd_ipd_put(env, bag, ipd);
3482 LINVRNT(osd_invariant(obj));
3486 static int osd_index_declare_ea_delete(const struct lu_env *env,
3487 struct dt_object *dt,
3488 const struct dt_key *key,
3489 struct thandle *handle)
3491 struct osd_thandle *oh;
3492 struct inode *inode;
3496 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3497 LASSERT(handle != NULL);
3499 oh = container_of0(handle, struct osd_thandle, ot_super);
3500 LASSERT(oh->ot_handle == NULL);
3502 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3503 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3505 inode = osd_dt_obj(dt)->oo_inode;
3508 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3509 true, true, NULL, false);
3513 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3516 struct osd_fid_pack *rec;
3519 if (de->file_type & LDISKFS_DIRENT_LUFID) {
3520 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3521 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3526 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3531 /* FID seqs not in FLDB, must be local seq */
3532 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3535 if (osd_seq_exists(env, osd, fid_seq(fid)))
3542 * Index delete function for interoperability mode (b11826).
3543 * It will remove the directory entry added by osd_index_ea_insert().
3544 * This entry is needed to maintain name->fid mapping.
3546 * \param key, key i.e. file entry to be deleted
3548 * \retval 0, on success
3549 * \retval -ve, on error
3551 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3552 const struct dt_key *key,
3553 struct thandle *handle,
3554 struct lustre_capa *capa)
3556 struct osd_object *obj = osd_dt_obj(dt);
3557 struct inode *dir = obj->oo_inode;
3558 struct dentry *dentry;
3559 struct osd_thandle *oh;
3560 struct ldiskfs_dir_entry_2 *de = NULL;
3561 struct buffer_head *bh;
3562 struct htree_lock *hlock = NULL;
3563 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
3564 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3568 LINVRNT(osd_invariant(obj));
3569 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3570 LASSERT(handle != NULL);
3572 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3574 oh = container_of(handle, struct osd_thandle, ot_super);
3575 LASSERT(oh->ot_handle != NULL);
3576 LASSERT(oh->ot_handle->h_transaction != NULL);
3578 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3581 ll_vfs_dq_init(dir);
3582 dentry = osd_child_dentry_get(env, obj,
3583 (char *)key, strlen((char *)key));
3585 if (obj->oo_hl_head != NULL) {
3586 hlock = osd_oti_get(env)->oti_hlock;
3587 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3588 dir, LDISKFS_HLOCK_DEL);
3590 down_write(&obj->oo_ext_idx_sem);
3593 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3597 /* If this is not the ".." entry, it might be a remote DNE
3598 * entry and we need to check if the FID is for a remote
3599 * MDT. If the FID is not in the directory entry (e.g.
3600 * upgraded 1.8 filesystem without dirdata enabled) then
3601 * we need to get the FID from the LMA. For a remote directory
3602 * there HAS to be an LMA, it cannot be an IGIF inode in this
3605 * Delete the entry before the agent inode in order to
3606 * simplify error handling. At worst an error after deleting
3607 * the entry first might leak the agent inode afterward. The
3608 * reverse would need filesystem abort in case of error deleting
3609 * the entry after the agent had been removed, or leave a
3610 * dangling entry pointing at a random inode. */
3611 if (strcmp((char *)key, dotdot) != 0) {
3612 LASSERT(de != NULL);
3613 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3614 /* If Fid is not in dentry, try to get it from LMA */
3615 if (rc == -ENODATA) {
3616 struct osd_inode_id *id;
3617 struct inode *inode;
3619 /* Before trying to get fid from the inode,
3620 * check whether the inode is valid.
3622 * If the inode has been deleted, do not go
3623 * ahead to do osd_ea_fid_get, which will set
3624 * the inode to bad inode, which might cause
3625 * the inode to be deleted uncorrectly */
3626 inode = ldiskfs_iget(osd_sb(osd),
3627 le32_to_cpu(de->inode));
3628 if (IS_ERR(inode)) {
3629 CDEBUG(D_INODE, "%s: "DFID"get inode"
3630 "error.\n", osd_name(osd),
3632 rc = PTR_ERR(inode);
3634 if (likely(inode->i_nlink != 0)) {
3635 id = &osd_oti_get(env)->oti_id;
3636 rc = osd_ea_fid_get(env, obj,
3637 le32_to_cpu(de->inode),
3640 CDEBUG(D_INFO, "%s: %u "DFID
3643 le32_to_cpu(de->inode),
3651 unlikely(osd_remote_fid(env, osd, fid)))
3652 /* Need to delete agent inode */
3653 ino = le32_to_cpu(de->inode);
3655 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3657 if (rc == 0 && unlikely(ino != 0)) {
3658 rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3661 CERROR("%s: del local inode "DFID": rc = %d\n",
3662 osd_name(osd), PFID(fid), rc);
3668 ldiskfs_htree_unlock(hlock);
3670 up_write(&obj->oo_ext_idx_sem);
3675 /* For inode on the remote MDT, .. will point to
3676 * /Agent directory, Check whether it needs to delete
3677 * from agent directory */
3678 if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3679 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3681 if (rc != 0 && rc != -ENOENT) {
3682 CERROR("%s: delete agent inode "DFID": rc = %d\n",
3683 osd_name(osd), PFID(fid), rc);
3693 LASSERT(osd_invariant(obj));
3698 * Lookup index for \a key and copy record to \a rec.
3700 * \param dt osd index object
3701 * \param key key for index
3702 * \param rec record reference
3704 * \retval +ve success : exact mach
3705 * \retval 0 return record with key not greater than \a key
3706 * \retval -ve failure
3708 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3709 struct dt_rec *rec, const struct dt_key *key,
3710 struct lustre_capa *capa)
3712 struct osd_object *obj = osd_dt_obj(dt);
3713 struct iam_path_descr *ipd;
3714 struct iam_container *bag = &obj->oo_dir->od_container;
3715 struct osd_thread_info *oti = osd_oti_get(env);
3716 struct iam_iterator *it = &oti->oti_idx_it;
3717 struct iam_rec *iam_rec;
3722 LASSERT(osd_invariant(obj));
3723 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3724 LASSERT(bag->ic_object == obj->oo_inode);
3726 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3729 ipd = osd_idx_ipd_get(env, bag);
3733 /* got ipd now we can start iterator. */
3734 iam_it_init(it, bag, 0, ipd);
3736 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3737 /* swab quota uid/gid provided by caller */
3738 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3739 key = (const struct dt_key *)&oti->oti_quota_id;
3742 rc = iam_it_get(it, (struct iam_key *)key);
3744 if (S_ISDIR(obj->oo_inode->i_mode))
3745 iam_rec = (struct iam_rec *)oti->oti_ldp;
3747 iam_rec = (struct iam_rec *) rec;
3749 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3751 if (S_ISDIR(obj->oo_inode->i_mode))
3752 osd_fid_unpack((struct lu_fid *) rec,
3753 (struct osd_fid_pack *)iam_rec);
3754 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3755 osd_quota_unpack(obj, rec);
3760 osd_ipd_put(env, bag, ipd);
3762 LINVRNT(osd_invariant(obj));
3767 static int osd_index_declare_iam_insert(const struct lu_env *env,
3768 struct dt_object *dt,
3769 const struct dt_rec *rec,
3770 const struct dt_key *key,
3771 struct thandle *handle)
3773 struct osd_thandle *oh;
3775 LASSERT(handle != NULL);
3777 oh = container_of0(handle, struct osd_thandle, ot_super);
3778 LASSERT(oh->ot_handle == NULL);
3780 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3781 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3787 * Inserts (key, value) pair in \a dt index object.
3789 * \param dt osd index object
3790 * \param key key for index
3791 * \param rec record reference
3792 * \param th transaction handler
3795 * \retval -ve failure
3797 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3798 const struct dt_rec *rec,
3799 const struct dt_key *key, struct thandle *th,
3800 struct lustre_capa *capa, int ignore_quota)
3802 struct osd_object *obj = osd_dt_obj(dt);
3803 struct iam_path_descr *ipd;
3804 struct osd_thandle *oh;
3805 struct iam_container *bag = &obj->oo_dir->od_container;
3806 struct osd_thread_info *oti = osd_oti_get(env);
3807 struct iam_rec *iam_rec;
3812 LINVRNT(osd_invariant(obj));
3813 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3814 LASSERT(bag->ic_object == obj->oo_inode);
3815 LASSERT(th != NULL);
3817 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3820 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3822 ipd = osd_idx_ipd_get(env, bag);
3823 if (unlikely(ipd == NULL))
3826 oh = container_of0(th, struct osd_thandle, ot_super);
3827 LASSERT(oh->ot_handle != NULL);
3828 LASSERT(oh->ot_handle->h_transaction != NULL);
3829 if (S_ISDIR(obj->oo_inode->i_mode)) {
3830 iam_rec = (struct iam_rec *)oti->oti_ldp;
3831 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3832 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3833 /* pack quota uid/gid */
3834 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3835 key = (const struct dt_key *)&oti->oti_quota_id;
3836 /* pack quota record */
3837 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3838 iam_rec = (struct iam_rec *)rec;
3840 iam_rec = (struct iam_rec *)rec;
3843 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3845 osd_ipd_put(env, bag, ipd);
3846 LINVRNT(osd_invariant(obj));
3851 * Calls ldiskfs_add_entry() to add directory entry
3852 * into the directory. This is required for
3853 * interoperability mode (b11826)
3855 * \retval 0, on success
3856 * \retval -ve, on error
3858 static int __osd_ea_add_rec(struct osd_thread_info *info,
3859 struct osd_object *pobj, struct inode *cinode,
3860 const char *name, const struct dt_rec *fid,
3861 struct htree_lock *hlock, struct thandle *th)
3863 struct ldiskfs_dentry_param *ldp;
3864 struct dentry *child;
3865 struct osd_thandle *oth;
3868 oth = container_of(th, struct osd_thandle, ot_super);
3869 LASSERT(oth->ot_handle != NULL);
3870 LASSERT(oth->ot_handle->h_transaction != NULL);
3871 LASSERT(pobj->oo_inode);
3873 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3874 if (unlikely(pobj->oo_inode ==
3875 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3878 osd_get_ldiskfs_dirent_param(ldp, fid);
3879 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3880 child->d_fsdata = (void *)ldp;
3881 ll_vfs_dq_init(pobj->oo_inode);
3882 rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3888 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3889 * into the directory.Also sets flags into osd object to
3890 * indicate dot and dotdot are created. This is required for
3891 * interoperability mode (b11826)
3893 * \param dir directory for dot and dotdot fixup.
3894 * \param obj child object for linking
3896 * \retval 0, on success
3897 * \retval -ve, on error
3899 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3900 struct osd_object *dir,
3901 struct inode *parent_dir, const char *name,
3902 const struct dt_rec *dot_fid,
3903 const struct dt_rec *dot_dot_fid,
3906 struct inode *inode = dir->oo_inode;
3907 struct osd_thandle *oth;
3910 oth = container_of(th, struct osd_thandle, ot_super);
3911 LASSERT(oth->ot_handle->h_transaction != NULL);
3912 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3914 if (strcmp(name, dot) == 0) {
3915 if (dir->oo_compat_dot_created) {
3918 LASSERT(inode == parent_dir);
3919 dir->oo_compat_dot_created = 1;
3922 } else if (strcmp(name, dotdot) == 0) {
3923 if (!dir->oo_compat_dot_created)
3925 /* in case of rename, dotdot is already created */
3926 if (dir->oo_compat_dotdot_created) {
3927 return __osd_ea_add_rec(info, dir, parent_dir, name,
3928 dot_dot_fid, NULL, th);
3931 result = osd_add_dot_dotdot_internal(info, dir->oo_inode,
3932 parent_dir, dot_fid,
3935 dir->oo_compat_dotdot_created = 1;
3943 * It will call the appropriate osd_add* function and return the
3944 * value, return by respective functions.
3946 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3947 struct inode *cinode, const char *name,
3948 const struct dt_rec *fid, struct thandle *th)
3950 struct osd_thread_info *info = osd_oti_get(env);
3951 struct htree_lock *hlock;
3954 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3956 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3958 if (hlock != NULL) {
3959 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3962 down_write(&pobj->oo_ext_idx_sem);
3964 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3965 (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3968 if (hlock != NULL) {
3969 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3970 pobj->oo_inode, LDISKFS_HLOCK_ADD);
3972 down_write(&pobj->oo_ext_idx_sem);
3975 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
3976 struct lu_fid *tfid = &info->oti_fid;
3978 *tfid = *(const struct lu_fid *)fid;
3980 rc = __osd_ea_add_rec(info, pobj, cinode, name,
3981 (const struct dt_rec *)tfid,
3984 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3989 ldiskfs_htree_unlock(hlock);
3991 up_write(&pobj->oo_ext_idx_sem);
3997 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3998 struct osd_idmap_cache *oic)
4000 struct osd_scrub *scrub = &dev->od_scrub;
4001 struct lu_fid *fid = &oic->oic_fid;
4002 struct osd_inode_id *id = &oti->oti_id;
4007 if (!fid_is_norm(fid) && !fid_is_igif(fid))
4011 rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
4012 if (rc != 0 && rc != -ENOENT)
4015 if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
4018 if (thread_is_running(&scrub->os_thread)) {
4019 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
4020 /* There is race condition between osd_oi_lookup and OI scrub.
4021 * The OI scrub finished just after osd_oi_lookup() failure.
4022 * Under such case, it is unnecessary to trigger OI scrub again,
4023 * but try to call osd_oi_lookup() again. */
4024 if (unlikely(rc == -EAGAIN))
4030 if (!dev->od_noscrub && ++once == 1) {
4031 rc = osd_scrub_start(dev);
4032 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC for "DFID
4034 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4043 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4044 struct osd_device *dev,
4045 struct osd_idmap_cache *oic,
4046 struct lu_fid *fid, __u32 ino)
4048 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
4049 struct inode *inode;
4052 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4053 inode = osd_iget(oti, dev, &oic->oic_lid);
4054 if (IS_ERR(inode)) {
4055 fid_zero(&oic->oic_fid);
4056 return PTR_ERR(inode);
4059 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4062 fid_zero(&oic->oic_fid);
4064 *fid = oic->oic_fid = lma->lma_self_fid;
4068 int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4069 struct osd_inode_id *id, const struct lu_fid *fid)
4071 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4072 id->oii_ino, id->oii_gen, info);
4073 info->oti_cache.oic_lid = *id;
4074 info->oti_cache.oic_fid = *fid;
4075 info->oti_cache.oic_dev = osd;
4081 * Calls ->lookup() to find dentry. From dentry get inode and
4082 * read inode's ea to get fid. This is required for interoperability
4085 * \retval 0, on success
4086 * \retval -ve, on error
4088 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4089 struct dt_rec *rec, const struct dt_key *key)
4091 struct inode *dir = obj->oo_inode;
4092 struct dentry *dentry;
4093 struct ldiskfs_dir_entry_2 *de;
4094 struct buffer_head *bh;
4095 struct lu_fid *fid = (struct lu_fid *) rec;
4096 struct htree_lock *hlock = NULL;
4101 LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
4103 dentry = osd_child_dentry_get(env, obj,
4104 (char *)key, strlen((char *)key));
4106 if (obj->oo_hl_head != NULL) {
4107 hlock = osd_oti_get(env)->oti_hlock;
4108 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4109 dir, LDISKFS_HLOCK_LOOKUP);
4111 down_read(&obj->oo_ext_idx_sem);
4114 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4116 struct osd_thread_info *oti = osd_oti_get(env);
4117 struct osd_inode_id *id = &oti->oti_id;
4118 struct osd_idmap_cache *oic = &oti->oti_cache;
4119 struct osd_device *dev = osd_obj2dev(obj);
4120 struct osd_scrub *scrub = &dev->od_scrub;
4121 struct scrub_file *sf = &scrub->os_file;
4123 ino = le32_to_cpu(de->inode);
4124 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4126 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4130 rc = osd_get_fid_from_dentry(de, rec);
4132 /* done with de, release bh */
4135 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4137 osd_id_gen(id, ino, OSD_OII_NOGEN);
4139 fid_zero(&oic->oic_fid);
4143 if (osd_remote_fid(env, dev, fid))
4146 rc = osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id,
4150 if ((scrub->os_pos_current <= ino) &&
4151 ((sf->sf_flags & SF_INCONSISTENT) ||
4152 (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
4153 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
4155 osd_consistency_check(oti, dev, oic);
4164 ldiskfs_htree_unlock(hlock);
4166 up_read(&obj->oo_ext_idx_sem);
4171 * Find the osd object for given fid.
4173 * \param fid need to find the osd object having this fid
4175 * \retval osd_object on success
4176 * \retval -ve on error
4178 struct osd_object *osd_object_find(const struct lu_env *env,
4179 struct dt_object *dt,
4180 const struct lu_fid *fid)
4182 struct lu_device *ludev = dt->do_lu.lo_dev;
4183 struct osd_object *child = NULL;
4184 struct lu_object *luch;
4185 struct lu_object *lo;
4188 * at this point topdev might not exist yet
4189 * (i.e. MGS is preparing profiles). so we can
4190 * not rely on topdev and instead lookup with
4191 * our device passed as topdev. this can't work
4192 * if the object isn't cached yet (as osd doesn't
4193 * allocate lu_header). IOW, the object must be
4194 * in the cache, otherwise lu_object_alloc() crashes
4197 luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4198 ludev : ludev->ld_site->ls_top_dev,
4200 if (!IS_ERR(luch)) {
4201 if (lu_object_exists(luch)) {
4202 lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4204 child = osd_obj(lo);
4206 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4207 "lu_object can't be located"
4208 DFID"\n", PFID(fid));
4210 if (child == NULL) {
4211 lu_object_put(env, luch);
4212 CERROR("Unable to get osd_object\n");
4213 child = ERR_PTR(-ENOENT);
4216 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4217 "lu_object does not exists "DFID"\n",
4219 lu_object_put(env, luch);
4220 child = ERR_PTR(-ENOENT);
4223 child = ERR_CAST(luch);
4230 * Put the osd object once done with it.
4232 * \param obj osd object that needs to be put
4234 static inline void osd_object_put(const struct lu_env *env,
4235 struct osd_object *obj)
4237 lu_object_put(env, &obj->oo_dt.do_lu);
4240 static int osd_index_declare_ea_insert(const struct lu_env *env,
4241 struct dt_object *dt,
4242 const struct dt_rec *rec,
4243 const struct dt_key *key,
4244 struct thandle *handle)
4246 struct osd_thandle *oh;
4247 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4248 struct lu_fid *fid = (struct lu_fid *)rec;
4252 LASSERT(!dt_object_remote(dt));
4253 LASSERT(handle != NULL);
4255 oh = container_of0(handle, struct osd_thandle, ot_super);
4256 LASSERT(oh->ot_handle == NULL);
4258 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4259 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4261 if (osd_dt_obj(dt)->oo_inode == NULL) {
4262 const char *name = (const char *)key;
4263 /* Object is not being created yet. Only happens when
4264 * 1. declare directory create
4265 * 2. declare insert .
4266 * 3. declare insert ..
4268 LASSERT(strcmp(name, dotdot) == 0 || strcmp(name, dot) == 0);
4270 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4272 /* We ignore block quota on meta pool (MDTs), so needn't
4273 * calculate how many blocks will be consumed by this index
4275 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0,
4276 oh, true, true, NULL, false);
4282 rc = osd_remote_fid(env, osd, fid);
4288 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
4289 osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
4290 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4291 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4292 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4293 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4299 * Index add function for interoperability mode (b11826).
4300 * It will add the directory entry.This entry is needed to
4301 * maintain name->fid mapping.
4303 * \param key it is key i.e. file entry to be inserted
4304 * \param rec it is value of given key i.e. fid
4306 * \retval 0, on success
4307 * \retval -ve, on error
4309 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4310 const struct dt_rec *rec,
4311 const struct dt_key *key, struct thandle *th,
4312 struct lustre_capa *capa, int ignore_quota)
4314 struct osd_object *obj = osd_dt_obj(dt);
4315 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4316 struct lu_fid *fid = (struct lu_fid *) rec;
4317 const char *name = (const char *)key;
4318 struct osd_thread_info *oti = osd_oti_get(env);
4319 struct osd_inode_id *id = &oti->oti_id;
4320 struct inode *child_inode = NULL;
4321 struct osd_object *child = NULL;
4325 LASSERT(osd_invariant(obj));
4326 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
4327 LASSERT(th != NULL);
4329 osd_trans_exec_op(env, th, OSD_OT_INSERT);
4331 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
4334 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!", PFID(fid));
4336 rc = osd_remote_fid(env, osd, fid);
4338 CERROR("%s: Can not find object "DFID" rc %d\n",
4339 osd_name(osd), PFID(fid), rc);
4344 /* Insert remote entry */
4345 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4346 struct osd_mdobj_map *omm = osd->od_mdt_map;
4347 struct osd_thandle *oh;
4349 /* If parent on remote MDT, we need put this object
4351 oh = container_of(th, typeof(*oh), ot_super);
4352 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4354 CERROR("%s: add "DFID" error: rc = %d\n",
4356 PFID(lu_object_fid(&dt->do_lu)), rc);
4360 child_inode = igrab(omm->omm_remote_parent->d_inode);
4362 child_inode = osd_create_local_agent_inode(env, osd,
4365 if (IS_ERR(child_inode))
4366 RETURN(PTR_ERR(child_inode));
4369 /* Insert local entry */
4370 child = osd_object_find(env, dt, fid);
4371 if (IS_ERR(child)) {
4372 CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4373 osd_name(osd), PFID(fid),
4374 id->oii_ino, id->oii_gen,
4375 (int)PTR_ERR(child));
4376 RETURN(PTR_ERR(child));
4378 child_inode = igrab(child->oo_inode);
4381 rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th);
4385 osd_object_put(env, child);
4386 LASSERT(osd_invariant(obj));
4391 * Initialize osd Iterator for given osd index object.
4393 * \param dt osd index object
4396 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4397 struct dt_object *dt,
4399 struct lustre_capa *capa)
4401 struct osd_it_iam *it;
4402 struct osd_thread_info *oti = osd_oti_get(env);
4403 struct osd_object *obj = osd_dt_obj(dt);
4404 struct lu_object *lo = &dt->do_lu;
4405 struct iam_path_descr *ipd;
4406 struct iam_container *bag = &obj->oo_dir->od_container;
4408 LASSERT(lu_object_exists(lo));
4410 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
4411 return ERR_PTR(-EACCES);
4414 ipd = osd_it_ipd_get(env, bag);
4415 if (likely(ipd != NULL)) {
4419 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4420 return (struct dt_it *)it;
4422 return ERR_PTR(-ENOMEM);
4426 * free given Iterator.
4429 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4431 struct osd_it_iam *it = (struct osd_it_iam *)di;
4432 struct osd_object *obj = it->oi_obj;
4434 iam_it_fini(&it->oi_it);
4435 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4436 lu_object_put(env, &obj->oo_dt.do_lu);
4440 * Move Iterator to record specified by \a key
4442 * \param di osd iterator
4443 * \param key key for index
4445 * \retval +ve di points to record with least key not larger than key
4446 * \retval 0 di points to exact matched key
4447 * \retval -ve failure
4450 static int osd_it_iam_get(const struct lu_env *env,
4451 struct dt_it *di, const struct dt_key *key)
4453 struct osd_thread_info *oti = osd_oti_get(env);
4454 struct osd_it_iam *it = (struct osd_it_iam *)di;
4456 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4457 /* swab quota uid/gid */
4458 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4459 key = (struct dt_key *)&oti->oti_quota_id;
4462 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4468 * \param di osd iterator
4470 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4472 struct osd_it_iam *it = (struct osd_it_iam *)di;
4474 iam_it_put(&it->oi_it);
4478 * Move iterator by one record
4480 * \param di osd iterator
4482 * \retval +1 end of container reached
4484 * \retval -ve failure
4487 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4489 struct osd_it_iam *it = (struct osd_it_iam *)di;
4491 return iam_it_next(&it->oi_it);
4495 * Return pointer to the key under iterator.
4498 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4499 const struct dt_it *di)
4501 struct osd_thread_info *oti = osd_oti_get(env);
4502 struct osd_it_iam *it = (struct osd_it_iam *)di;
4503 struct osd_object *obj = it->oi_obj;
4506 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4508 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4509 /* swab quota uid/gid */
4510 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4511 key = (struct dt_key *)&oti->oti_quota_id;
4518 * Return size of key under iterator (in bytes)
4521 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4523 struct osd_it_iam *it = (struct osd_it_iam *)di;
4525 return iam_it_key_size(&it->oi_it);
4529 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4531 /* check if file type is required */
4532 if (ent->lde_attrs & LUDA_TYPE) {
4533 struct luda_type *lt;
4534 int align = sizeof(*lt) - 1;
4536 len = (len + align) & ~align;
4537 lt = (struct luda_type *)(ent->lde_name + len);
4538 lt->lt_type = cpu_to_le16(DTTOIF(type));
4541 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4545 * build lu direct from backend fs dirent.
4549 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4550 char *name, __u16 namelen, __u16 type, __u32 attr)
4552 ent->lde_attrs = attr | LUDA_FID;
4553 fid_cpu_to_le(&ent->lde_fid, fid);
4555 ent->lde_hash = cpu_to_le64(offset);
4556 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4558 strncpy(ent->lde_name, name, namelen);
4559 ent->lde_name[namelen] = '\0';
4560 ent->lde_namelen = cpu_to_le16(namelen);
4562 /* append lustre attributes */
4563 osd_it_append_attrs(ent, namelen, type);
4567 * Return pointer to the record under iterator.
4569 static int osd_it_iam_rec(const struct lu_env *env,
4570 const struct dt_it *di,
4571 struct dt_rec *dtrec, __u32 attr)
4573 struct osd_it_iam *it = (struct osd_it_iam *)di;
4574 struct osd_thread_info *info = osd_oti_get(env);
4577 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4578 const struct osd_fid_pack *rec;
4579 struct lu_fid *fid = &info->oti_fid;
4580 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
4586 name = (char *)iam_it_key_get(&it->oi_it);
4588 RETURN(PTR_ERR(name));
4590 namelen = iam_it_key_size(&it->oi_it);
4592 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4594 RETURN(PTR_ERR(rec));
4596 rc = osd_fid_unpack(fid, rec);
4600 hash = iam_it_store(&it->oi_it);
4602 /* IAM does not store object type in IAM index (dir) */
4603 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4605 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4606 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4607 (struct iam_rec *)dtrec);
4608 osd_quota_unpack(it->oi_obj, dtrec);
4610 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4611 (struct iam_rec *)dtrec);
4618 * Returns cookie for current Iterator position.
4620 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4622 struct osd_it_iam *it = (struct osd_it_iam *)di;
4624 return iam_it_store(&it->oi_it);
4628 * Restore iterator from cookie.
4630 * \param di osd iterator
4631 * \param hash Iterator location cookie
4633 * \retval +ve di points to record with least key not larger than key.
4634 * \retval 0 di points to exact matched key
4635 * \retval -ve failure
4638 static int osd_it_iam_load(const struct lu_env *env,
4639 const struct dt_it *di, __u64 hash)
4641 struct osd_it_iam *it = (struct osd_it_iam *)di;
4643 return iam_it_load(&it->oi_it, hash);
4646 static const struct dt_index_operations osd_index_iam_ops = {
4647 .dio_lookup = osd_index_iam_lookup,
4648 .dio_declare_insert = osd_index_declare_iam_insert,
4649 .dio_insert = osd_index_iam_insert,
4650 .dio_declare_delete = osd_index_declare_iam_delete,
4651 .dio_delete = osd_index_iam_delete,
4653 .init = osd_it_iam_init,
4654 .fini = osd_it_iam_fini,
4655 .get = osd_it_iam_get,
4656 .put = osd_it_iam_put,
4657 .next = osd_it_iam_next,
4658 .key = osd_it_iam_key,
4659 .key_size = osd_it_iam_key_size,
4660 .rec = osd_it_iam_rec,
4661 .store = osd_it_iam_store,
4662 .load = osd_it_iam_load
4668 * Creates or initializes iterator context.
4670 * \retval struct osd_it_ea, iterator structure on success
4673 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4674 struct dt_object *dt,
4676 struct lustre_capa *capa)
4678 struct osd_object *obj = osd_dt_obj(dt);
4679 struct osd_thread_info *info = osd_oti_get(env);
4680 struct osd_it_ea *it = &info->oti_it_ea;
4681 struct file *file = &it->oie_file;
4682 struct lu_object *lo = &dt->do_lu;
4683 struct dentry *obj_dentry = &info->oti_it_dentry;
4685 LASSERT(lu_object_exists(lo));
4687 obj_dentry->d_inode = obj->oo_inode;
4688 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4689 obj_dentry->d_name.hash = 0;
4691 it->oie_rd_dirent = 0;
4692 it->oie_it_dirent = 0;
4693 it->oie_dirent = NULL;
4694 it->oie_buf = info->oti_it_ea_buf;
4697 /* Reset the "file" totally to avoid to reuse any old value from
4698 * former readdir handling, the "file->f_pos" should be zero. */
4699 memset(file, 0, sizeof(*file));
4700 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4701 if (attr & LUDA_64BITHASH)
4702 file->f_mode = FMODE_64BITHASH;
4704 file->f_mode = FMODE_32BITHASH;
4705 file->f_dentry = obj_dentry;
4706 file->f_mapping = obj->oo_inode->i_mapping;
4707 file->f_op = obj->oo_inode->i_fop;
4708 set_file_inode(file, obj->oo_inode);
4711 RETURN((struct dt_it *) it);
4715 * Destroy or finishes iterator context.
4717 * \param di iterator structure to be destroyed
4719 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4721 struct osd_it_ea *it = (struct osd_it_ea *)di;
4722 struct osd_object *obj = it->oie_obj;
4723 struct inode *inode = obj->oo_inode;
4726 it->oie_file.f_op->release(inode, &it->oie_file);
4727 lu_object_put(env, &obj->oo_dt.do_lu);
4732 * It position the iterator at given key, so that next lookup continues from
4733 * that key Or it is similar to dio_it->load() but based on a key,
4734 * rather than file position.
4736 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4739 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4741 static int osd_it_ea_get(const struct lu_env *env,
4742 struct dt_it *di, const struct dt_key *key)
4744 struct osd_it_ea *it = (struct osd_it_ea *)di;
4747 LASSERT(((const char *)key)[0] == '\0');
4748 it->oie_file.f_pos = 0;
4749 it->oie_rd_dirent = 0;
4750 it->oie_it_dirent = 0;
4751 it->oie_dirent = NULL;
4759 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4763 struct osd_filldir_cbs {
4764 #ifdef HAVE_DIR_CONTEXT
4765 struct dir_context ctx;
4767 struct osd_it_ea *it;
4770 * It is called internally by ->readdir(). It fills the
4771 * iterator's in-memory data structure with required
4772 * information i.e. name, namelen, rec_size etc.
4774 * \param buf in which information to be filled in.
4775 * \param name name of the file in given dir
4777 * \retval 0 on success
4778 * \retval 1 on buffer full
4780 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
4781 loff_t offset, __u64 ino,
4784 struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
4785 struct osd_object *obj = it->oie_obj;
4786 struct osd_it_ea_dirent *ent = it->oie_dirent;
4787 struct lu_fid *fid = &ent->oied_fid;
4788 struct osd_fid_pack *rec;
4791 /* this should never happen */
4792 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4793 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4797 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4801 /* "." is just the object itself. */
4802 if (namelen == 1 && name[0] == '.') {
4803 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4804 } else if (d_type & LDISKFS_DIRENT_LUFID) {
4805 rec = (struct osd_fid_pack*) (name + namelen + 1);
4806 if (osd_fid_unpack(fid, rec) != 0)
4811 d_type &= ~LDISKFS_DIRENT_LUFID;
4813 /* NOT export local root. */
4814 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
4815 ino = obj->oo_inode->i_ino;
4816 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4819 ent->oied_ino = ino;
4820 ent->oied_off = offset;
4821 ent->oied_namelen = namelen;
4822 ent->oied_type = d_type;
4824 memcpy(ent->oied_name, name, namelen);
4826 it->oie_rd_dirent++;
4827 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4832 * Calls ->readdir() to load a directory entry at a time
4833 * and stored it in iterator's in-memory data structure.
4835 * \param di iterator's in memory structure
4837 * \retval 0 on success
4838 * \retval -ve on error
4840 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4841 const struct dt_it *di)
4843 struct osd_it_ea *it = (struct osd_it_ea *)di;
4844 struct osd_object *obj = it->oie_obj;
4845 struct inode *inode = obj->oo_inode;
4846 struct htree_lock *hlock = NULL;
4847 struct file *filp = &it->oie_file;
4849 struct osd_filldir_cbs buf = {
4850 #ifdef HAVE_DIR_CONTEXT
4851 .ctx.actor = osd_ldiskfs_filldir,
4857 it->oie_dirent = it->oie_buf;
4858 it->oie_rd_dirent = 0;
4860 if (obj->oo_hl_head != NULL) {
4861 hlock = osd_oti_get(env)->oti_hlock;
4862 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4863 inode, LDISKFS_HLOCK_READDIR);
4865 down_read(&obj->oo_ext_idx_sem);
4868 #ifdef HAVE_DIR_CONTEXT
4869 buf.ctx.pos = filp->f_pos;
4870 rc = inode->i_fop->iterate(filp, &buf.ctx);
4871 filp->f_pos = buf.ctx.pos;
4873 rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
4877 ldiskfs_htree_unlock(hlock);
4879 up_read(&obj->oo_ext_idx_sem);
4881 if (it->oie_rd_dirent == 0) {
4882 /*If it does not get any dirent, it means it has been reached
4883 *to the end of the dir */
4884 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
4886 it->oie_dirent = it->oie_buf;
4887 it->oie_it_dirent = 1;
4894 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4895 * to load a directory entry at a time and stored it in
4896 * iterator's in-memory data structure.
4898 * \param di iterator's in memory structure
4900 * \retval +ve iterator reached to end
4901 * \retval 0 iterator not reached to end
4902 * \retval -ve on error
4904 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4906 struct osd_it_ea *it = (struct osd_it_ea *)di;
4911 if (it->oie_it_dirent < it->oie_rd_dirent) {
4913 (void *) it->oie_dirent +
4914 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4915 it->oie_dirent->oied_namelen);
4916 it->oie_it_dirent++;
4919 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4922 rc = osd_ldiskfs_it_fill(env, di);
4929 * Returns the key at current position from iterator's in memory structure.
4931 * \param di iterator's in memory structure
4933 * \retval key i.e. struct dt_key on success
4935 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4936 const struct dt_it *di)
4938 struct osd_it_ea *it = (struct osd_it_ea *)di;
4940 return (struct dt_key *)it->oie_dirent->oied_name;
4944 * Returns the key's size at current position from iterator's in memory structure.
4946 * \param di iterator's in memory structure
4948 * \retval key_size i.e. struct dt_key on success
4950 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4952 struct osd_it_ea *it = (struct osd_it_ea *)di;
4954 return it->oie_dirent->oied_namelen;
4958 osd_dirent_update(handle_t *jh, struct super_block *sb,
4959 struct osd_it_ea_dirent *ent, struct lu_fid *fid,
4960 struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
4962 struct osd_fid_pack *rec;
4966 LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
4967 LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
4969 rc = ldiskfs_journal_get_write_access(jh, bh);
4971 CERROR("%.16s: fail to write access for update dirent: "
4972 "name = %.*s, rc = %d\n",
4973 LDISKFS_SB(sb)->s_es->s_volume_name,
4974 ent->oied_namelen, ent->oied_name, rc);
4978 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
4979 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
4980 rc = ldiskfs_journal_dirty_metadata(jh, bh);
4982 CERROR("%.16s: fail to dirty metadata for update dirent: "
4983 "name = %.*s, rc = %d\n",
4984 LDISKFS_SB(sb)->s_es->s_volume_name,
4985 ent->oied_namelen, ent->oied_name, rc);
4991 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
4993 if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
4994 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5001 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5003 LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5004 "dot_dotdot = %d\n", dot_dotdot);
5006 if (LDISKFS_DIR_REC_LEN(de) >=
5007 __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5014 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5015 struct inode *dir, struct inode *inode,
5016 struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5017 struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5018 struct htree_lock *hlock)
5020 struct dentry *dentry;
5021 struct osd_fid_pack *rec;
5022 struct ldiskfs_dentry_param *ldp;
5026 if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5027 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5030 /* There is enough space to hold the FID-in-dirent. */
5031 if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
5032 dir->i_sb->s_blocksize)) {
5033 rc = ldiskfs_journal_get_write_access(jh, bh);
5035 CERROR("%.16s: fail to write access for reinsert "
5036 "dirent: name = %.*s, rc = %d\n",
5037 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5038 ent->oied_namelen, ent->oied_name, rc);
5042 de->name[de->name_len] = 0;
5043 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5044 rec->fp_len = sizeof(struct lu_fid) + 1;
5045 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5046 de->file_type |= LDISKFS_DIRENT_LUFID;
5048 rc = ldiskfs_journal_dirty_metadata(jh, bh);
5050 CERROR("%.16s: fail to dirty metadata for reinsert "
5051 "dirent: name = %.*s, rc = %d\n",
5052 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5053 ent->oied_namelen, ent->oied_name, rc);
5058 rc = ldiskfs_delete_entry(jh, dir, de, bh);
5060 CERROR("%.16s: fail to delete entry for reinsert dirent: "
5061 "name = %.*s, rc = %d\n",
5062 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5063 ent->oied_namelen, ent->oied_name, rc);
5067 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5069 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5070 osd_get_ldiskfs_dirent_param(ldp, (const struct dt_rec *)fid);
5071 dentry->d_fsdata = (void *)ldp;
5072 ll_vfs_dq_init(dir);
5073 rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5074 /* It is too bad, we cannot reinsert the name entry back.
5075 * That means we lose it! */
5077 CERROR("%.16s: fail to insert entry for reinsert dirent: "
5078 "name = %.*s, rc = %d\n",
5079 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5080 ent->oied_namelen, ent->oied_name, rc);
5086 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5087 struct osd_it_ea *it, struct lu_fid *fid,
5088 struct osd_inode_id *id, __u32 *attr)
5090 struct osd_thread_info *info = osd_oti_get(env);
5091 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
5092 struct osd_device *dev = osd_obj2dev(obj);
5093 struct super_block *sb = osd_sb(dev);
5094 const char *devname =
5095 LDISKFS_SB(sb)->s_es->s_volume_name;
5096 struct osd_it_ea_dirent *ent = it->oie_dirent;
5097 struct inode *dir = obj->oo_inode;
5098 struct htree_lock *hlock = NULL;
5099 struct buffer_head *bh = NULL;
5100 handle_t *jh = NULL;
5101 struct ldiskfs_dir_entry_2 *de;
5102 struct dentry *dentry;
5103 struct inode *inode;
5110 if (ent->oied_name[0] == '.') {
5111 if (ent->oied_namelen == 1)
5113 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5117 dentry = osd_child_dentry_get(env, obj, ent->oied_name,
5120 /* We need to ensure that the name entry is still valid.
5121 * Because it may be removed or renamed by other already.
5123 * The unlink or rename operation will start journal before PDO lock,
5124 * so to avoid deadlock, here we need to start journal handle before
5125 * related PDO lock also. But because we do not know whether there
5126 * will be something to be repaired before PDO lock, we just start
5127 * journal without conditions.
5129 * We may need to remove the name entry firstly, then insert back.
5130 * One credit is for user quota file update.
5131 * One credit is for group quota file update.
5132 * Two credits are for dirty inode. */
5133 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5134 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5137 if (dev->od_dirent_journal) {
5138 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5141 CERROR("%.16s: fail to start trans for dirent "
5142 "check_repair: credits %d, name %.*s, rc %d\n",
5143 devname, credits, ent->oied_namelen,
5144 ent->oied_name, rc);
5148 if (obj->oo_hl_head != NULL) {
5149 hlock = osd_oti_get(env)->oti_hlock;
5150 /* "0" means exclusive lock for the whole directory.
5151 * We need to prevent others access such name entry
5152 * during the delete + insert. Neither HLOCK_ADD nor
5153 * HLOCK_DEL cannot guarantee the atomicity. */
5154 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5156 down_write(&obj->oo_ext_idx_sem);
5159 if (obj->oo_hl_head != NULL) {
5160 hlock = osd_oti_get(env)->oti_hlock;
5161 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5162 LDISKFS_HLOCK_LOOKUP);
5164 down_read(&obj->oo_ext_idx_sem);
5168 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5169 /* For dot/dotdot entry, if there is not enough space to hold the
5170 * FID-in-dirent, just keep them there. It only happens when the
5171 * device upgraded from 1.8 or restored from MDT file-level backup.
5172 * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5173 * and needs to get FID from LMA when readdir, it will not affect the
5174 * performance much. */
5175 if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
5176 (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5177 *attr |= LUDA_IGNORE;
5178 GOTO(out_journal, rc = 0);
5181 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5182 inode = osd_iget(info, dev, id);
5183 if (IS_ERR(inode)) {
5184 rc = PTR_ERR(inode);
5185 if (rc == -ENOENT || rc == -ESTALE) {
5186 *attr |= LUDA_IGNORE;
5190 GOTO(out_journal, rc);
5193 /* skip the REMOTE_PARENT_DIR. */
5194 if (inode == dev->od_mdt_map->omm_remote_parent->d_inode)
5195 GOTO(out_inode, rc = 0);
5197 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
5199 LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
5201 if (fid_is_sane(fid)) {
5202 /* FID-in-dirent is valid. */
5203 if (lu_fid_eq(fid, &lma->lma_self_fid))
5204 GOTO(out_inode, rc = 0);
5206 /* Do not repair under dryrun mode. */
5207 if (*attr & LUDA_VERIFY_DRYRUN) {
5208 *attr |= LUDA_REPAIR;
5209 GOTO(out_inode, rc = 0);
5212 if (!dev->od_dirent_journal) {
5216 ldiskfs_htree_unlock(hlock);
5218 up_read(&obj->oo_ext_idx_sem);
5219 dev->od_dirent_journal = 1;
5223 *fid = lma->lma_self_fid;
5225 /* Update the FID-in-dirent. */
5226 rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
5228 *attr |= LUDA_REPAIR;
5230 /* Do not repair under dryrun mode. */
5231 if (*attr & LUDA_VERIFY_DRYRUN) {
5232 *fid = lma->lma_self_fid;
5233 *attr |= LUDA_REPAIR;
5234 GOTO(out_inode, rc = 0);
5237 if (!dev->od_dirent_journal) {
5241 ldiskfs_htree_unlock(hlock);
5243 up_read(&obj->oo_ext_idx_sem);
5244 dev->od_dirent_journal = 1;
5248 *fid = lma->lma_self_fid;
5250 /* Append the FID-in-dirent. */
5251 rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5252 fid, bh, de, hlock);
5254 *attr |= LUDA_REPAIR;
5256 } else if (rc == -ENODATA) {
5257 /* Do not repair under dryrun mode. */
5258 if (*attr & LUDA_VERIFY_DRYRUN) {
5259 if (fid_is_sane(fid)) {
5260 *attr |= LUDA_REPAIR;
5262 lu_igif_build(fid, inode->i_ino,
5263 inode->i_generation);
5264 *attr |= LUDA_UPGRADE;
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;
5281 if (unlikely(fid_is_sane(fid))) {
5282 /* FID-in-dirent exists, but FID-in-LMA is lost.
5283 * Trust the FID-in-dirent, and add FID-in-LMA. */
5284 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5286 *attr |= LUDA_REPAIR;
5288 lu_igif_build(fid, inode->i_ino, inode->i_generation);
5289 /* It is probably IGIF object. Only aappend the
5290 * FID-in-dirent. OI scrub will process FID-in-LMA. */
5291 rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5292 fid, bh, de, hlock);
5294 *attr |= LUDA_UPGRADE;
5298 GOTO(out_inode, rc);
5305 if (hlock != NULL) {
5306 ldiskfs_htree_unlock(hlock);
5308 if (dev->od_dirent_journal)
5309 up_write(&obj->oo_ext_idx_sem);
5311 up_read(&obj->oo_ext_idx_sem);
5314 ldiskfs_journal_stop(jh);
5315 if (rc >= 0 && !dirty)
5316 dev->od_dirent_journal = 0;
5321 * Returns the value at current position from iterator's in memory structure.
5323 * \param di struct osd_it_ea, iterator's in memory structure
5324 * \param attr attr requested for dirent.
5325 * \param lde lustre dirent
5327 * \retval 0 no error and \param lde has correct lustre dirent.
5328 * \retval -ve on error
5330 static inline int osd_it_ea_rec(const struct lu_env *env,
5331 const struct dt_it *di,
5332 struct dt_rec *dtrec, __u32 attr)
5334 struct osd_it_ea *it = (struct osd_it_ea *)di;
5335 struct osd_object *obj = it->oie_obj;
5336 struct osd_device *dev = osd_obj2dev(obj);
5337 struct osd_scrub *scrub = &dev->od_scrub;
5338 struct scrub_file *sf = &scrub->os_file;
5339 struct osd_thread_info *oti = osd_oti_get(env);
5340 struct osd_inode_id *id = &oti->oti_id;
5341 struct osd_idmap_cache *oic = &oti->oti_cache;
5342 struct lu_fid *fid = &it->oie_dirent->oied_fid;
5343 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
5344 __u32 ino = it->oie_dirent->oied_ino;
5348 if (attr & LUDA_VERIFY) {
5350 if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) {
5351 attr |= LUDA_IGNORE;
5354 rc = osd_dirent_check_repair(env, obj, it, fid, id,
5358 attr &= ~LU_DIRENT_ATTRS_MASK;
5359 if (!fid_is_sane(fid)) {
5360 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP) &&
5361 likely(it->oie_dirent->oied_namelen != 2 ||
5362 it->oie_dirent->oied_name[0] != '.' ||
5363 it->oie_dirent->oied_name[1] != '.'))
5366 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5368 osd_id_gen(id, ino, OSD_OII_NOGEN);
5372 /* Pack the entry anyway, at least the offset is right. */
5373 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5374 it->oie_dirent->oied_name,
5375 it->oie_dirent->oied_namelen,
5376 it->oie_dirent->oied_type, attr);
5381 if (osd_remote_fid(env, dev, fid))
5384 if (likely(!(attr & LUDA_IGNORE)))
5385 rc = osd_add_oi_cache(oti, dev, id, fid);
5387 if (!(attr & LUDA_VERIFY) &&
5388 (scrub->os_pos_current <= ino) &&
5389 ((sf->sf_flags & SF_INCONSISTENT) ||
5390 (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
5391 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
5392 osd_consistency_check(oti, dev, oic);
5398 * Returns a cookie for current position of the iterator head, so that
5399 * user can use this cookie to load/start the iterator next time.
5401 * \param di iterator's in memory structure
5403 * \retval cookie for current position, on success
5405 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5407 struct osd_it_ea *it = (struct osd_it_ea *)di;
5409 return it->oie_dirent->oied_off;
5413 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5414 * to load a directory entry at a time and stored it i inn,
5415 * in iterator's in-memory data structure.
5417 * \param di struct osd_it_ea, iterator's in memory structure
5419 * \retval +ve on success
5420 * \retval -ve on error
5422 static int osd_it_ea_load(const struct lu_env *env,
5423 const struct dt_it *di, __u64 hash)
5425 struct osd_it_ea *it = (struct osd_it_ea *)di;
5429 it->oie_file.f_pos = hash;
5431 rc = osd_ldiskfs_it_fill(env, di);
5439 * Index lookup function for interoperability mode (b11826).
5441 * \param key, key i.e. file name to be searched
5443 * \retval +ve, on success
5444 * \retval -ve, on error
5446 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5447 struct dt_rec *rec, const struct dt_key *key,
5448 struct lustre_capa *capa)
5450 struct osd_object *obj = osd_dt_obj(dt);
5455 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5456 LINVRNT(osd_invariant(obj));
5458 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
5461 rc = osd_ea_lookup_rec(env, obj, rec, key);
5468 * Index and Iterator operations for interoperability
5469 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5471 static const struct dt_index_operations osd_index_ea_ops = {
5472 .dio_lookup = osd_index_ea_lookup,
5473 .dio_declare_insert = osd_index_declare_ea_insert,
5474 .dio_insert = osd_index_ea_insert,
5475 .dio_declare_delete = osd_index_declare_ea_delete,
5476 .dio_delete = osd_index_ea_delete,
5478 .init = osd_it_ea_init,
5479 .fini = osd_it_ea_fini,
5480 .get = osd_it_ea_get,
5481 .put = osd_it_ea_put,
5482 .next = osd_it_ea_next,
5483 .key = osd_it_ea_key,
5484 .key_size = osd_it_ea_key_size,
5485 .rec = osd_it_ea_rec,
5486 .store = osd_it_ea_store,
5487 .load = osd_it_ea_load
5491 static void *osd_key_init(const struct lu_context *ctx,
5492 struct lu_context_key *key)
5494 struct osd_thread_info *info;
5496 OBD_ALLOC_PTR(info);
5498 return ERR_PTR(-ENOMEM);
5500 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5501 if (info->oti_it_ea_buf == NULL)
5504 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5506 info->oti_hlock = ldiskfs_htree_lock_alloc();
5507 if (info->oti_hlock == NULL)
5513 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5516 return ERR_PTR(-ENOMEM);
5519 static void osd_key_fini(const struct lu_context *ctx,
5520 struct lu_context_key *key, void* data)
5522 struct osd_thread_info *info = data;
5524 if (info->oti_hlock != NULL)
5525 ldiskfs_htree_lock_free(info->oti_hlock);
5526 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5527 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5528 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5532 static void osd_key_exit(const struct lu_context *ctx,
5533 struct lu_context_key *key, void *data)
5535 struct osd_thread_info *info = data;
5537 LASSERT(info->oti_r_locks == 0);
5538 LASSERT(info->oti_w_locks == 0);
5539 LASSERT(info->oti_txns == 0);
5542 /* type constructor/destructor: osd_type_init, osd_type_fini */
5543 LU_TYPE_INIT_FINI(osd, &osd_key);
5545 struct lu_context_key osd_key = {
5546 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5547 .lct_init = osd_key_init,
5548 .lct_fini = osd_key_fini,
5549 .lct_exit = osd_key_exit
5553 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5554 const char *name, struct lu_device *next)
5556 struct osd_device *osd = osd_dev(d);
5558 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5559 >= sizeof(osd->od_svname))
5561 return osd_procfs_init(osd, name);
5564 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5568 /* shutdown quota slave instance associated with the device */
5569 if (o->od_quota_slave != NULL) {
5570 qsd_fini(env, o->od_quota_slave);
5571 o->od_quota_slave = NULL;
5577 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5581 if (o->od_mnt != NULL) {
5582 shrink_dcache_sb(osd_sb(o));
5583 osd_sync(env, &o->od_dt_dev);
5592 static int osd_mount(const struct lu_env *env,
5593 struct osd_device *o, struct lustre_cfg *cfg)
5595 const char *name = lustre_cfg_string(cfg, 0);
5596 const char *dev = lustre_cfg_string(cfg, 1);
5598 unsigned long page, s_flags, lmd_flags = 0;
5599 struct page *__page;
5600 struct file_system_type *type;
5601 char *options = NULL;
5603 struct osd_thread_info *info = osd_oti_get(env);
5604 struct lu_fid *fid = &info->oti_fid;
5605 struct inode *inode;
5609 if (o->od_mnt != NULL)
5612 if (strlen(dev) >= sizeof(o->od_mntdev))
5614 strcpy(o->od_mntdev, dev);
5616 OBD_PAGE_ALLOC(__page, GFP_IOFS);
5618 GOTO(out, rc = -ENOMEM);
5620 str = lustre_cfg_string(cfg, 2);
5621 s_flags = simple_strtoul(str, NULL, 0);
5622 str = strstr(str, ":");
5624 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5625 opts = lustre_cfg_string(cfg, 3);
5626 page = (unsigned long)page_address(__page);
5627 options = (char *)page;
5630 strcat(options, "user_xattr,acl");
5632 strcat(options, opts);
5634 /* Glom up mount options */
5635 if (*options != '\0')
5636 strcat(options, ",");
5637 strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
5639 type = get_fs_type("ldiskfs");
5641 CERROR("%s: cannot find ldiskfs module\n", name);
5642 GOTO(out, rc = -ENODEV);
5645 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5646 module_put(type->owner);
5648 if (IS_ERR(o->od_mnt)) {
5649 rc = PTR_ERR(o->od_mnt);
5651 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5655 #ifdef HAVE_DEV_SET_RDONLY
5656 if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5657 CERROR("%s: underlying device %s is marked as read-only. "
5658 "Setup failed\n", name, dev);
5659 GOTO(out_mnt, rc = -EROFS);
5663 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5664 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5665 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5666 GOTO(out_mnt, rc = -EINVAL);
5669 #ifdef LDISKFS_MOUNT_DIRDATA
5670 if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
5671 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5672 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
5674 inode = osd_sb(o)->s_root->d_inode;
5675 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
5676 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
5678 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
5682 if (lmd_flags & LMD_FLG_NOSCRUB)
5693 OBD_PAGE_FREE(__page);
5698 static struct lu_device *osd_device_fini(const struct lu_env *env,
5699 struct lu_device *d)
5701 struct osd_device *o = osd_dev(d);
5704 osd_shutdown(env, o);
5706 osd_scrub_cleanup(env, o);
5707 osd_obj_map_fini(o);
5713 static int osd_device_init0(const struct lu_env *env,
5714 struct osd_device *o,
5715 struct lustre_cfg *cfg)
5717 struct lu_device *l = osd2lu_dev(o);
5718 struct osd_thread_info *info;
5722 /* if the module was re-loaded, env can loose its keys */
5723 rc = lu_env_refill((struct lu_env *) env);
5726 info = osd_oti_get(env);
5729 l->ld_ops = &osd_lu_ops;
5730 o->od_dt_dev.dd_ops = &osd_dt_ops;
5732 spin_lock_init(&o->od_osfs_lock);
5733 mutex_init(&o->od_otable_mutex);
5735 o->od_capa_hash = init_capa_hash();
5736 if (o->od_capa_hash == NULL)
5737 GOTO(out, rc = -ENOMEM);
5739 o->od_read_cache = 1;
5740 o->od_writethrough_cache = 1;
5741 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
5743 rc = osd_mount(env, o, cfg);
5747 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
5748 sizeof(o->od_svname));
5749 if (cplen >= sizeof(o->od_svname)) {
5754 if (server_name_is_ost(o->od_svname))
5757 rc = osd_obj_map_init(env, o);
5761 rc = lu_site_init(&o->od_site, l);
5763 GOTO(out_compat, rc);
5764 o->od_site.ls_bottom_dev = l;
5766 rc = lu_site_init_finish(&o->od_site);
5770 /* self-repair LMA by default */
5771 o->od_lma_self_repair = 1;
5773 CFS_INIT_LIST_HEAD(&o->od_ios_list);
5774 /* setup scrub, including OI files initialization */
5775 rc = osd_scrub_setup(env, o);
5779 rc = osd_procfs_init(o, o->od_svname);
5781 CERROR("%s: can't initialize procfs: rc = %d\n",
5783 GOTO(out_scrub, rc);
5786 LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
5788 /* initialize quota slave instance */
5789 o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
5791 if (IS_ERR(o->od_quota_slave)) {
5792 rc = PTR_ERR(o->od_quota_slave);
5793 o->od_quota_slave = NULL;
5794 GOTO(out_procfs, rc);
5802 osd_scrub_cleanup(env, o);
5804 lu_site_fini(&o->od_site);
5806 osd_obj_map_fini(o);
5810 cleanup_capa_hash(o->od_capa_hash);
5815 static struct lu_device *osd_device_alloc(const struct lu_env *env,
5816 struct lu_device_type *t,
5817 struct lustre_cfg *cfg)
5819 struct osd_device *o;
5824 return ERR_PTR(-ENOMEM);
5826 rc = dt_device_init(&o->od_dt_dev, t);
5828 /* Because the ctx might be revived in dt_device_init,
5829 * refill the env here */
5830 lu_env_refill((struct lu_env *)env);
5831 rc = osd_device_init0(env, o, cfg);
5833 dt_device_fini(&o->od_dt_dev);
5836 if (unlikely(rc != 0))
5839 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
5842 static struct lu_device *osd_device_free(const struct lu_env *env,
5843 struct lu_device *d)
5845 struct osd_device *o = osd_dev(d);
5848 cleanup_capa_hash(o->od_capa_hash);
5849 /* XXX: make osd top device in order to release reference */
5850 d->ld_site->ls_top_dev = d;
5851 lu_site_purge(env, d->ld_site, -1);
5852 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
5853 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
5854 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
5856 lu_site_fini(&o->od_site);
5857 dt_device_fini(&o->od_dt_dev);
5862 static int osd_process_config(const struct lu_env *env,
5863 struct lu_device *d, struct lustre_cfg *cfg)
5865 struct osd_device *o = osd_dev(d);
5869 switch (cfg->lcfg_command) {
5871 rc = osd_mount(env, o, cfg);
5874 lu_dev_del_linkage(d->ld_site, d);
5875 rc = osd_shutdown(env, o);
5878 LASSERT(&o->od_dt_dev);
5879 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
5880 cfg, &o->od_dt_dev);
5881 if (rc > 0 || rc == -ENOSYS)
5882 rc = class_process_proc_param(PARAM_OST,
5883 lprocfs_osd_obd_vars,
5884 cfg, &o->od_dt_dev);
5893 static int osd_recovery_complete(const struct lu_env *env,
5894 struct lu_device *d)
5896 struct osd_device *osd = osd_dev(d);
5900 if (osd->od_quota_slave == NULL)
5903 /* start qsd instance on recovery completion, this notifies the quota
5904 * slave code that we are about to process new requests now */
5905 rc = qsd_start(env, osd->od_quota_slave);
5910 * we use exports to track all osd users
5912 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
5913 struct obd_device *obd, struct obd_uuid *cluuid,
5914 struct obd_connect_data *data, void *localdata)
5916 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
5917 struct lustre_handle conn;
5921 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
5923 rc = class_connect(&conn, obd, cluuid);
5927 *exp = class_conn2export(&conn);
5929 spin_lock(&osd->od_osfs_lock);
5931 spin_unlock(&osd->od_osfs_lock);
5937 * once last export (we don't count self-export) disappeared
5938 * osd can be released
5940 static int osd_obd_disconnect(struct obd_export *exp)
5942 struct obd_device *obd = exp->exp_obd;
5943 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
5944 int rc, release = 0;
5947 /* Only disconnect the underlying layers on the final disconnect. */
5948 spin_lock(&osd->od_osfs_lock);
5950 if (osd->od_connects == 0)
5952 spin_unlock(&osd->od_osfs_lock);
5954 rc = class_disconnect(exp); /* bz 9811 */
5956 if (rc == 0 && release)
5957 class_manual_cleanup(obd);
5961 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
5962 struct lu_device *dev)
5964 struct osd_device *osd = osd_dev(dev);
5968 if (osd->od_quota_slave != NULL)
5969 /* set up quota slave objects */
5970 result = qsd_prepare(env, osd->od_quota_slave);
5975 static const struct lu_object_operations osd_lu_obj_ops = {
5976 .loo_object_init = osd_object_init,
5977 .loo_object_delete = osd_object_delete,
5978 .loo_object_release = osd_object_release,
5979 .loo_object_free = osd_object_free,
5980 .loo_object_print = osd_object_print,
5981 .loo_object_invariant = osd_object_invariant
5984 const struct lu_device_operations osd_lu_ops = {
5985 .ldo_object_alloc = osd_object_alloc,
5986 .ldo_process_config = osd_process_config,
5987 .ldo_recovery_complete = osd_recovery_complete,
5988 .ldo_prepare = osd_prepare,
5991 static const struct lu_device_type_operations osd_device_type_ops = {
5992 .ldto_init = osd_type_init,
5993 .ldto_fini = osd_type_fini,
5995 .ldto_start = osd_type_start,
5996 .ldto_stop = osd_type_stop,
5998 .ldto_device_alloc = osd_device_alloc,
5999 .ldto_device_free = osd_device_free,
6001 .ldto_device_init = osd_device_init,
6002 .ldto_device_fini = osd_device_fini
6005 struct lu_device_type osd_device_type = {
6006 .ldt_tags = LU_DEVICE_DT,
6007 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
6008 .ldt_ops = &osd_device_type_ops,
6009 .ldt_ctx_tags = LCT_LOCAL,
6013 * lprocfs legacy support.
6015 static struct obd_ops osd_obd_device_ops = {
6016 .o_owner = THIS_MODULE,
6017 .o_connect = osd_obd_connect,
6018 .o_disconnect = osd_obd_disconnect
6021 static int __init osd_mod_init(void)
6027 rc = lu_kmem_init(ldiskfs_caches);
6031 rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL,
6032 #ifndef HAVE_ONLY_PROCFS_SEQ
6033 lprocfs_osd_module_vars,
6035 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6037 lu_kmem_fini(ldiskfs_caches);
6041 static void __exit osd_mod_exit(void)
6043 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6044 lu_kmem_fini(ldiskfs_caches);
6047 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6048 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6049 MODULE_LICENSE("GPL");
6051 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);