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_OSD
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>
76 #include <lustre_linkea.h>
79 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
80 "ldiskfs with parallel directory operations");
82 int ldiskfs_track_declares_assert;
83 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
84 "LBUG during tracking of declares");
86 /* Slab to allocate dynlocks */
87 struct kmem_cache *dynlock_cachep;
89 static struct lu_kmem_descr ldiskfs_caches[] = {
91 .ckd_cache = &dynlock_cachep,
92 .ckd_name = "dynlock_cache",
93 .ckd_size = sizeof(struct dynlock_handle)
100 static const char dot[] = ".";
101 static const char dotdot[] = "..";
102 static const char remote_obj_dir[] = "REM_OBJ_DIR";
104 static const struct lu_object_operations osd_lu_obj_ops;
105 static const struct dt_object_operations osd_obj_ops;
106 static const struct dt_object_operations osd_obj_ea_ops;
107 static const struct dt_object_operations osd_obj_otable_it_ops;
108 static const struct dt_index_operations osd_index_iam_ops;
109 static const struct dt_index_operations osd_index_ea_ops;
111 int osd_trans_declare_op2rb[] = {
112 [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
113 [OSD_OT_PUNCH] = OSD_OT_MAX,
114 [OSD_OT_XATTR_SET] = OSD_OT_XATTR_SET,
115 [OSD_OT_CREATE] = OSD_OT_DESTROY,
116 [OSD_OT_DESTROY] = OSD_OT_CREATE,
117 [OSD_OT_REF_ADD] = OSD_OT_REF_DEL,
118 [OSD_OT_REF_DEL] = OSD_OT_REF_ADD,
119 [OSD_OT_WRITE] = OSD_OT_WRITE,
120 [OSD_OT_INSERT] = OSD_OT_DELETE,
121 [OSD_OT_DELETE] = OSD_OT_INSERT,
122 [OSD_OT_UPDATE] = OSD_OT_MAX,
123 [OSD_OT_QUOTA] = OSD_OT_MAX,
126 static int osd_has_index(const struct osd_object *obj)
128 return obj->oo_dt.do_index_ops != NULL;
131 static int osd_object_invariant(const struct lu_object *l)
133 return osd_invariant(osd_obj(l));
137 * Concurrency: doesn't matter
141 * Concurrency: doesn't matter
143 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
145 struct osd_thread_info *oti = osd_oti_get(env);
146 return oti->oti_w_locks > 0 && o->oo_owner == env;
150 * Concurrency: doesn't access mutable data
152 static int osd_root_get(const struct lu_env *env,
153 struct dt_device *dev, struct lu_fid *f)
155 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
160 * OSD object methods.
164 * Concurrency: no concurrent access is possible that early in object
167 static struct lu_object *osd_object_alloc(const struct lu_env *env,
168 const struct lu_object_header *hdr,
171 struct osd_object *mo;
177 l = &mo->oo_dt.do_lu;
178 dt_object_init(&mo->oo_dt, NULL, d);
179 mo->oo_dt.do_ops = &osd_obj_ea_ops;
180 l->lo_ops = &osd_lu_obj_ops;
181 init_rwsem(&mo->oo_sem);
182 init_rwsem(&mo->oo_ext_idx_sem);
183 spin_lock_init(&mo->oo_guard);
190 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
191 struct dentry *dentry, struct lustre_mdt_attrs *lma)
195 CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
196 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
197 info->oti_mdt_attrs_old, LMA_OLD_SIZE);
199 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
200 memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
202 lustre_lma_swab(lma);
203 /* Check LMA compatibility */
204 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
205 CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
206 "for fid = "DFID", ino = %lu\n",
207 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
208 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
209 PFID(&lma->lma_self_fid), inode->i_ino);
212 } else if (rc == 0) {
220 * retrieve object from backend ext fs.
222 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
223 struct osd_inode_id *id)
225 struct inode *inode = NULL;
227 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
229 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
230 id->oii_ino, PTR_ERR(inode));
231 } else if (id->oii_gen != OSD_OII_NOGEN &&
232 inode->i_generation != id->oii_gen) {
233 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
234 "i_generation = %u\n",
235 id->oii_ino, id->oii_gen, inode->i_generation);
237 inode = ERR_PTR(-ESTALE);
238 } else if (inode->i_nlink == 0) {
239 /* due to parallel readdir and unlink,
240 * we can have dead inode here. */
241 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
243 inode = ERR_PTR(-ESTALE);
244 } else if (is_bad_inode(inode)) {
245 CWARN("%.16s: bad inode: ino = %u\n",
246 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
248 inode = ERR_PTR(-ENOENT);
250 if (id->oii_gen == OSD_OII_NOGEN)
251 osd_id_gen(id, inode->i_ino, inode->i_generation);
253 /* Do not update file c/mtime in ldiskfs.
254 * NB: we don't have any lock to protect this because we don't
255 * have reference on osd_object now, but contention with
256 * another lookup + attr_set can't happen in the tiny window
257 * between if (...) and set S_NOCMTIME. */
258 if (!(inode->i_flags & S_NOCMTIME))
259 inode->i_flags |= S_NOCMTIME;
264 static struct inode *
265 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
266 struct osd_inode_id *id, struct lu_fid *fid)
268 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
272 inode = osd_iget(info, dev, id);
276 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
278 *fid = lma->lma_self_fid;
279 } else if (rc == -ENODATA) {
280 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
281 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
283 lu_igif_build(fid, inode->i_ino, inode->i_generation);
291 static struct inode *osd_iget_check(struct osd_thread_info *info,
292 struct osd_device *dev,
293 const struct lu_fid *fid,
294 struct osd_inode_id *id,
301 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
304 if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) {
305 CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
314 if (is_bad_inode(inode)) {
317 CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
325 if (id->oii_gen != OSD_OII_NOGEN &&
326 inode->i_generation != id->oii_gen) {
329 CDEBUG(D_INODE, "unmatched inode: ino = %u, "
330 "oii_gen = %u, i_generation = %u\n",
331 id->oii_ino, id->oii_gen, inode->i_generation);
339 if (inode->i_nlink == 0) {
342 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
352 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
354 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
355 /* XXX: There are some possible cases:
357 * Backup/restore caused the OI invalid.
359 * Someone unlinked the object but NOT removed
360 * the OI mapping, such as mount target device
361 * as ldiskfs, and modify something directly.
363 * Someone just removed the object between the
364 * former oi_lookup and the iget. It is normal.
365 * 4. Other failure cases.
367 * Generally, when the device is mounted, it will
368 * auto check whether the system is restored from
369 * file-level backup or not. We trust such detect
370 * to distinguish the 1st case from the 2nd case. */
372 if (!IS_ERR(inode) && inode->i_generation != 0 &&
373 inode->i_generation == id->oii_gen)
379 if (id->oii_gen == OSD_OII_NOGEN)
380 osd_id_gen(id, inode->i_ino, inode->i_generation);
382 /* Do not update file c/mtime in ldiskfs.
383 * NB: we don't have any lock to protect this because we don't
384 * have reference on osd_object now, but contention with
385 * another lookup + attr_set can't happen in the tiny window
386 * between if (...) and set S_NOCMTIME. */
387 if (!(inode->i_flags & S_NOCMTIME))
388 inode->i_flags |= S_NOCMTIME;
405 * \retval +v: new filter_fid, does not contain self-fid
406 * \retval 0: filter_fid_old, contains self-fid
407 * \retval -v: other failure cases
409 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
410 struct dentry *dentry, struct lu_fid *fid)
412 struct filter_fid_old *ff = &info->oti_ff;
413 struct ost_id *ostid = &info->oti_ostid;
416 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
417 if (rc == sizeof(*ff)) {
419 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
420 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
421 /* XXX: should use real OST index in the future. LU-3569 */
422 ostid_to_fid(fid, ostid, 0);
423 } else if (rc == sizeof(struct filter_fid)) {
425 } else if (rc >= 0) {
432 static int osd_lma_self_repair(struct osd_thread_info *info,
433 struct osd_device *osd, struct inode *inode,
434 const struct lu_fid *fid, __u32 compat)
439 LASSERT(current->journal_info == NULL);
441 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
442 osd_dto_credits_noquota[DTO_XATTR_SET]);
445 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
450 rc = osd_ea_fid_set(info, inode, fid, compat, 0);
452 CWARN("%s: cannot self repair the LMA: rc = %d\n",
454 ldiskfs_journal_stop(jh);
458 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
460 struct osd_thread_info *info = osd_oti_get(env);
461 struct osd_device *osd = osd_obj2dev(obj);
462 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
463 struct inode *inode = obj->oo_inode;
464 struct dentry *dentry = &info->oti_obj_dentry;
465 struct lu_fid *fid = NULL;
466 const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
470 CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
471 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
472 info->oti_mdt_attrs_old, LMA_OLD_SIZE);
473 if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
474 fid = &lma->lma_self_fid;
475 rc = osd_get_idif(info, inode, dentry, fid);
476 if ((rc > 0) || (rc == -ENODATA && osd->od_lma_self_repair)) {
477 /* For the given OST-object, if it has neither LMA nor
478 * FID in XATTR_NAME_FID, then the given FID (which is
479 * contained in the @obj, from client RPC for locating
480 * the OST-object) is trusted. We use it to generate
482 osd_lma_self_repair(info, osd, inode, rfid,
483 fid_is_on_ost(info, osd, fid, OI_CHECK_FLD) ?
484 LMAC_FID_ON_OST : 0);
489 if (unlikely(rc == -ENODATA))
497 lustre_lma_swab(lma);
498 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
499 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
500 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
501 "fid = "DFID", ino = %lu\n", osd_name(osd),
502 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
503 PFID(rfid), inode->i_ino);
505 } else if (!(lma->lma_compat & LMAC_NOT_IN_OI)) {
506 fid = &lma->lma_self_fid;
510 if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
511 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
512 struct ost_id *oi = &info->oti_ostid;
513 struct lu_fid *fid1 = &info->oti_fid3;
514 __u32 idx = fid_idif_ost_idx(rfid);
516 /* For old IDIF, the OST index is not part of the IDIF,
517 * Means that different OSTs may have the same IDIFs.
518 * Under such case, we need to make some compatible
519 * check to make sure to trigger OI scrub properly. */
520 if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
521 /* Given @rfid is new, LMA is old. */
522 fid_to_ostid(fid, oi);
523 ostid_to_fid(fid1, oi, idx);
524 if (lu_fid_eq(fid1, rfid)) {
525 if (osd->od_lma_self_repair)
526 osd_lma_self_repair(info, osd,
534 CDEBUG(D_INODE, "%s: FID "DFID" != self_fid "DFID"\n",
535 osd_name(osd), PFID(rfid), PFID(fid));
542 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
543 const struct lu_fid *fid,
544 const struct lu_object_conf *conf)
546 struct osd_thread_info *info;
547 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
548 struct osd_device *dev;
549 struct osd_idmap_cache *oic;
550 struct osd_inode_id *id;
552 struct osd_scrub *scrub;
553 struct scrub_file *sf;
557 bool triggered = false;
560 LINVRNT(osd_invariant(obj));
561 LASSERT(obj->oo_inode == NULL);
562 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
565 scrub = &dev->od_scrub;
566 sf = &scrub->os_file;
567 info = osd_oti_get(env);
569 oic = &info->oti_cache;
571 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
574 /* For the object is created as locking anchor, or for the object to
575 * be created on disk. No need to osd_oi_lookup() at here because FID
576 * shouldn't never be re-used, if it's really a duplicate FID from
577 * unexpected reason, we should be able to detect it later by calling
578 * do_create->osd_oi_insert(). */
579 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
580 GOTO(out, result = 0);
582 /* Search order: 1. per-thread cache. */
583 if (lu_fid_eq(fid, &oic->oic_fid) &&
584 likely(oic->oic_dev == dev)) {
590 if (!list_empty(&scrub->os_inconsistent_items)) {
591 /* Search order: 2. OI scrub pending list. */
592 result = osd_oii_lookup(dev, fid, id);
597 /* Search order: 3. OI files. */
598 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
599 if (result == -ENOENT) {
600 if (!fid_is_norm(fid) ||
601 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
602 !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
604 GOTO(out, result = 0);
615 inode = osd_iget_check(info, dev, fid, id, in_oi);
617 result = PTR_ERR(inode);
618 if (result == -ENOENT || result == -ESTALE) {
620 fid_zero(&oic->oic_fid);
622 GOTO(out, result = -ENOENT);
623 } else if (result == -EREMCHG) {
627 fid_zero(&oic->oic_fid);
629 if (unlikely(triggered))
630 GOTO(out, result = saved);
633 if (thread_is_running(&scrub->os_thread)) {
634 result = -EINPROGRESS;
635 } else if (!dev->od_noscrub) {
636 result = osd_scrub_start(dev);
637 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
638 "for "DFID", rc = %d [1]\n",
639 osd_name(dev), PFID(fid),result);
640 if (result == 0 || result == -EALREADY)
641 result = -EINPROGRESS;
646 /* We still have chance to get the valid inode: for the
647 * object which is referenced by remote name entry, the
648 * object on the local MDT will be linked under the dir
649 * of "/REMOTE_PARENT_DIR" with its FID string as name.
651 * We do not know whether the object for the given FID
652 * is referenced by some remote name entry or not, and
653 * especially for DNE II, a multiple-linked object may
654 * have many name entries reside on many MDTs.
656 * To simplify the operation, OSD will not distinguish
657 * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
658 * only happened for the RPC from other MDT during the
659 * OI scrub, or for the client side RPC with FID only,
660 * such as FID to path, or from old connected client. */
662 result = osd_lookup_in_remote_parent(info, dev,
675 obj->oo_inode = inode;
676 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
678 result = osd_check_lma(env, obj);
681 obj->oo_inode = NULL;
682 if (result == -EREMCHG) {
684 result = osd_oi_lookup(info, dev, fid, id,
687 fid_zero(&oic->oic_fid);
698 obj->oo_compat_dot_created = 1;
699 obj->oo_compat_dotdot_created = 1;
701 if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
702 GOTO(out, result = 0);
704 LASSERT(obj->oo_hl_head == NULL);
705 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
706 if (obj->oo_hl_head == NULL) {
707 obj->oo_inode = NULL;
709 GOTO(out, result = -ENOMEM);
711 GOTO(out, result = 0);
714 LINVRNT(osd_invariant(obj));
719 * Concurrency: shouldn't matter.
721 static void osd_object_init0(struct osd_object *obj)
723 LASSERT(obj->oo_inode != NULL);
724 obj->oo_dt.do_body_ops = &osd_body_ops;
725 obj->oo_dt.do_lu.lo_header->loh_attr |=
726 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
730 * Concurrency: no concurrent access is possible that early in object
733 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
734 const struct lu_object_conf *conf)
736 struct osd_object *obj = osd_obj(l);
739 LINVRNT(osd_invariant(obj));
741 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
742 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
743 l->lo_header->loh_attr |= LOHA_EXISTS;
747 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
748 obj->oo_dt.do_body_ops = &osd_body_ops_new;
749 if (result == 0 && obj->oo_inode != NULL)
750 osd_object_init0(obj);
752 LINVRNT(osd_invariant(obj));
757 * Concurrency: no concurrent access is possible that late in object
760 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
762 struct osd_object *obj = osd_obj(l);
764 LINVRNT(osd_invariant(obj));
766 dt_object_fini(&obj->oo_dt);
767 if (obj->oo_hl_head != NULL)
768 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
773 * Concurrency: no concurrent access is possible that late in object
776 static void osd_index_fini(struct osd_object *o)
778 struct iam_container *bag;
780 if (o->oo_dir != NULL) {
781 bag = &o->oo_dir->od_container;
782 if (o->oo_inode != NULL) {
783 if (bag->ic_object == o->oo_inode)
784 iam_container_fini(bag);
786 OBD_FREE_PTR(o->oo_dir);
792 * Concurrency: no concurrent access is possible that late in object
793 * life-cycle (for all existing callers, that is. New callers have to provide
794 * their own locking.)
796 static int osd_inode_unlinked(const struct inode *inode)
798 return inode->i_nlink == 0;
802 OSD_TXN_OI_DELETE_CREDITS = 20,
803 OSD_TXN_INODE_DELETE_CREDITS = 20
810 #if OSD_THANDLE_STATS
812 * Set time when the handle is allocated
814 static void osd_th_alloced(struct osd_thandle *oth)
816 oth->oth_alloced = cfs_time_current();
820 * Set time when the handle started
822 static void osd_th_started(struct osd_thandle *oth)
824 oth->oth_started = cfs_time_current();
828 * Helper function to convert time interval to microseconds packed in
831 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
835 cfs_duration_usec(cfs_time_sub(end, start), &val);
836 return val.tv_sec * 1000000 + val.tv_usec;
840 * Check whether the we deal with this handle for too long.
842 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
843 cfs_time_t alloced, cfs_time_t started,
846 cfs_time_t now = cfs_time_current();
848 LASSERT(dev != NULL);
850 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
851 interval_to_usec(alloced, started));
852 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
853 interval_to_usec(started, closed));
854 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
855 interval_to_usec(closed, now));
857 if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
858 CWARN("transaction handle %p was open for too long: "
860 "alloced "CFS_TIME_T" ,"
861 "started "CFS_TIME_T" ,"
862 "closed "CFS_TIME_T"\n",
863 oth, now, alloced, started, closed);
864 libcfs_debug_dumpstack(NULL);
868 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
870 cfs_time_t __closed = cfs_time_current(); \
871 cfs_time_t __alloced = oth->oth_alloced; \
872 cfs_time_t __started = oth->oth_started; \
875 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
878 #else /* OSD_THANDLE_STATS */
880 #define osd_th_alloced(h) do {} while(0)
881 #define osd_th_started(h) do {} while(0)
882 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
884 #endif /* OSD_THANDLE_STATS */
887 * Concurrency: doesn't access mutable data.
889 static int osd_param_is_not_sane(const struct osd_device *dev,
890 const struct thandle *th)
892 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
894 return oh->ot_credits > osd_transaction_size(dev);
898 * Concurrency: shouldn't matter.
900 static void osd_trans_commit_cb(struct super_block *sb,
901 struct ldiskfs_journal_cb_entry *jcb, int error)
903 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
904 struct thandle *th = &oh->ot_super;
905 struct lu_device *lud = &th->th_dev->dd_lu_dev;
906 struct dt_txn_commit_cb *dcb, *tmp;
908 LASSERT(oh->ot_handle == NULL);
911 CERROR("transaction @0x%p commit error: %d\n", th, error);
913 dt_txn_hook_commit(th);
915 /* call per-transaction callbacks if any */
916 list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
917 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
918 "commit callback entry: magic=%x name='%s'\n",
919 dcb->dcb_magic, dcb->dcb_name);
920 list_del_init(&dcb->dcb_linkage);
921 dcb->dcb_func(NULL, th, dcb, error);
924 lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
928 lu_context_exit(&th->th_ctx);
929 lu_context_fini(&th->th_ctx);
933 static struct thandle *osd_trans_create(const struct lu_env *env,
936 struct osd_thread_info *oti = osd_oti_get(env);
937 struct osd_iobuf *iobuf = &oti->oti_iobuf;
938 struct osd_thandle *oh;
942 /* on pending IO in this thread should left from prev. request */
943 LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
945 th = ERR_PTR(-ENOMEM);
946 OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
948 oh->ot_quota_trans = &oti->oti_quota_trans;
949 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
953 th->th_tags = LCT_TX_HANDLE;
955 atomic_set(&th->th_refc, 1);
956 th->th_alloc_size = sizeof(*oh);
957 oti->oti_dev = osd_dt_dev(d);
958 INIT_LIST_HEAD(&oh->ot_dcb_list);
961 memset(oti->oti_declare_ops, 0,
962 sizeof(oti->oti_declare_ops));
963 memset(oti->oti_declare_ops_rb, 0,
964 sizeof(oti->oti_declare_ops_rb));
965 memset(oti->oti_declare_ops_cred, 0,
966 sizeof(oti->oti_declare_ops_cred));
967 oti->oti_rollback = false;
973 * Concurrency: shouldn't matter.
975 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
978 struct osd_thread_info *oti = osd_oti_get(env);
979 struct osd_device *dev = osd_dt_dev(d);
981 struct osd_thandle *oh;
986 LASSERT(current->journal_info == NULL);
988 oh = container_of0(th, struct osd_thandle, ot_super);
990 LASSERT(oh->ot_handle == NULL);
992 rc = dt_txn_hook_start(env, d, th);
996 if (unlikely(osd_param_is_not_sane(dev, th))) {
997 static unsigned long last_printed;
998 static int last_credits;
1000 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1001 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1003 osd_journal(dev)->j_max_transaction_buffers);
1004 CWARN(" create: %u/%u, destroy: %u/%u\n",
1005 oti->oti_declare_ops[OSD_OT_CREATE],
1006 oti->oti_declare_ops_cred[OSD_OT_CREATE],
1007 oti->oti_declare_ops[OSD_OT_DESTROY],
1008 oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
1009 CWARN(" attr_set: %u/%u, xattr_set: %u/%u\n",
1010 oti->oti_declare_ops[OSD_OT_ATTR_SET],
1011 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1012 oti->oti_declare_ops[OSD_OT_XATTR_SET],
1013 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
1014 CWARN(" write: %u/%u, punch: %u/%u, quota %u/%u\n",
1015 oti->oti_declare_ops[OSD_OT_WRITE],
1016 oti->oti_declare_ops_cred[OSD_OT_WRITE],
1017 oti->oti_declare_ops[OSD_OT_PUNCH],
1018 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1019 oti->oti_declare_ops[OSD_OT_QUOTA],
1020 oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
1021 CWARN(" insert: %u/%u, delete: %u/%u\n",
1022 oti->oti_declare_ops[OSD_OT_INSERT],
1023 oti->oti_declare_ops_cred[OSD_OT_INSERT],
1024 oti->oti_declare_ops[OSD_OT_DELETE],
1025 oti->oti_declare_ops_cred[OSD_OT_DELETE]);
1026 CWARN(" ref_add: %u/%u, ref_del: %u/%u\n",
1027 oti->oti_declare_ops[OSD_OT_REF_ADD],
1028 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1029 oti->oti_declare_ops[OSD_OT_REF_DEL],
1030 oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
1032 if (last_credits != oh->ot_credits &&
1033 time_after(jiffies, last_printed + 60 * HZ)) {
1034 libcfs_debug_dumpstack(NULL);
1035 last_credits = oh->ot_credits;
1036 last_printed = jiffies;
1038 /* XXX Limit the credits to 'max_transaction_buffers', and
1039 * let the underlying filesystem to catch the error if
1040 * we really need so many credits.
1042 * This should be removed when we can calculate the
1043 * credits precisely. */
1044 oh->ot_credits = osd_transaction_size(dev);
1048 * XXX temporary stuff. Some abstraction layer should
1051 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1055 LASSERT(oti->oti_txns == 0);
1056 lu_context_init(&th->th_ctx, th->th_tags);
1057 lu_context_enter(&th->th_ctx);
1059 lu_device_get(&d->dd_lu_dev);
1060 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1071 static int osd_seq_exists(const struct lu_env *env,
1072 struct osd_device *osd, obd_seq seq)
1074 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1075 struct seq_server_site *ss = osd_seq_site(osd);
1079 LASSERT(ss != NULL);
1080 LASSERT(ss->ss_server_fld != NULL);
1082 rc = osd_fld_lookup(env, osd, seq, range);
1085 CERROR("%s: can't lookup FLD sequence "LPX64
1086 ": rc = %d\n", osd_name(osd), seq, rc);
1090 RETURN(ss->ss_node_id == range->lsr_index);
1094 * Concurrency: shouldn't matter.
1096 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1100 struct osd_thandle *oh;
1101 struct osd_thread_info *oti = osd_oti_get(env);
1102 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1103 struct qsd_instance *qsd = oti->oti_dev->od_quota_slave;
1104 struct lquota_trans *qtrans;
1107 oh = container_of0(th, struct osd_thandle, ot_super);
1109 qtrans = oh->ot_quota_trans;
1110 oh->ot_quota_trans = NULL;
1112 if (oh->ot_handle != NULL) {
1113 handle_t *hdl = oh->ot_handle;
1116 * add commit callback
1117 * notice we don't do this in osd_trans_start()
1118 * as underlying transaction can change during truncate
1120 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1123 LASSERT(oti->oti_txns == 1);
1125 rc = dt_txn_hook_stop(env, th);
1127 CERROR("Failure in transaction hook: %d\n", rc);
1129 /* hook functions might modify th_sync */
1130 hdl->h_sync = th->th_sync;
1132 oh->ot_handle = NULL;
1133 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
1134 rc = ldiskfs_journal_stop(hdl));
1136 CERROR("Failure to stop transaction: %d\n", rc);
1138 thandle_put(&oh->ot_super);
1141 /* inform the quota slave device that the transaction is stopping */
1142 qsd_op_end(env, qsd, qtrans);
1144 /* as we want IO to journal and data IO be concurrent, we don't block
1145 * awaiting data IO completion in osd_do_bio(), instead we wait here
1146 * once transaction is submitted to the journal. all reqular requests
1147 * don't do direct IO (except read/write), thus this wait_event becomes
1150 * IMPORTANT: we have to wait till any IO submited by the thread is
1151 * completed otherwise iobuf may be corrupted by different request
1153 wait_event(iobuf->dr_wait,
1154 atomic_read(&iobuf->dr_numreqs) == 0);
1155 osd_fini_iobuf(oti->oti_dev, iobuf);
1157 rc = iobuf->dr_error;
1162 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1164 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1167 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1168 LASSERT(&dcb->dcb_func != NULL);
1169 list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1175 * Called just before object is freed. Releases all resources except for
1176 * object itself (that is released by osd_object_free()).
1178 * Concurrency: no concurrent access is possible that late in object
1181 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1183 struct osd_object *obj = osd_obj(l);
1184 struct inode *inode = obj->oo_inode;
1186 LINVRNT(osd_invariant(obj));
1189 * If object is unlinked remove fid->ino mapping from object index.
1192 osd_index_fini(obj);
1193 if (inode != NULL) {
1194 struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave;
1195 qid_t uid = i_uid_read(inode);
1196 qid_t gid = i_gid_read(inode);
1199 obj->oo_inode = NULL;
1202 struct osd_thread_info *info = osd_oti_get(env);
1203 struct lquota_id_info *qi = &info->oti_qi;
1205 /* Release granted quota to master if necessary */
1206 qi->lqi_id.qid_uid = uid;
1207 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1209 qi->lqi_id.qid_uid = gid;
1210 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1216 * Concurrency: ->loo_object_release() is called under site spin-lock.
1218 static void osd_object_release(const struct lu_env *env,
1219 struct lu_object *l)
1224 * Concurrency: shouldn't matter.
1226 static int osd_object_print(const struct lu_env *env, void *cookie,
1227 lu_printer_t p, const struct lu_object *l)
1229 struct osd_object *o = osd_obj(l);
1230 struct iam_descr *d;
1232 if (o->oo_dir != NULL)
1233 d = o->oo_dir->od_container.ic_descr;
1236 return (*p)(env, cookie,
1237 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1239 o->oo_inode ? o->oo_inode->i_ino : 0UL,
1240 o->oo_inode ? o->oo_inode->i_generation : 0,
1241 d ? d->id_ops->id_name : "plain");
1244 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1247 * Concurrency: shouldn't matter.
1249 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1250 struct obd_statfs *sfs)
1252 struct osd_device *osd = osd_dt_dev(d);
1253 struct super_block *sb = osd_sb(osd);
1254 struct kstatfs *ksfs;
1257 if (unlikely(osd->od_mnt == NULL))
1258 return -EINPROGRESS;
1260 /* osd_lproc.c call this without env, allocate ksfs for that case */
1261 if (unlikely(env == NULL)) {
1262 OBD_ALLOC_PTR(ksfs);
1266 ksfs = &osd_oti_get(env)->oti_ksfs;
1269 spin_lock(&osd->od_osfs_lock);
1270 result = sb->s_op->statfs(sb->s_root, ksfs);
1271 if (likely(result == 0)) { /* N.B. statfs can't really fail */
1272 statfs_pack(sfs, ksfs);
1273 if (sb->s_flags & MS_RDONLY)
1274 sfs->os_state = OS_STATE_READONLY;
1277 spin_unlock(&osd->od_osfs_lock);
1279 if (unlikely(env == NULL))
1282 /* Reserve a small amount of space for local objects like last_rcvd,
1283 * llog, quota files, ... */
1284 if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1287 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1288 /** Take out metadata overhead for indirect blocks */
1289 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1296 * Estimate space needed for file creations. We assume the largest filename
1297 * which is 2^64 - 1, hence a filename of 20 chars.
1298 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1300 #ifdef __LDISKFS_DIR_REC_LEN
1301 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1303 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1307 * Concurrency: doesn't access mutable data.
1309 static void osd_conf_get(const struct lu_env *env,
1310 const struct dt_device *dev,
1311 struct dt_device_param *param)
1313 struct super_block *sb = osd_sb(osd_dt_dev(dev));
1317 * XXX should be taken from not-yet-existing fs abstraction layer.
1319 param->ddp_max_name_len = LDISKFS_NAME_LEN;
1320 param->ddp_max_nlink = LDISKFS_LINK_MAX;
1321 param->ddp_block_shift = sb->s_blocksize_bits;
1322 param->ddp_mount_type = LDD_MT_LDISKFS;
1323 param->ddp_maxbytes = sb->s_maxbytes;
1324 /* Overhead estimate should be fairly accurate, so we really take a tiny
1325 * error margin which also avoids fragmenting the filesystem too much */
1326 param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1327 /* inode are statically allocated, so per-inode space consumption
1328 * is the space consumed by the directory entry */
1329 param->ddp_inodespace = PER_OBJ_USAGE;
1330 /* per-fragment overhead to be used by the client code */
1331 param->ddp_grant_frag = 6 * LDISKFS_BLOCK_SIZE(sb);
1332 param->ddp_mntopts = 0;
1333 if (test_opt(sb, XATTR_USER))
1334 param->ddp_mntopts |= MNTOPT_USERXATTR;
1335 if (test_opt(sb, POSIX_ACL))
1336 param->ddp_mntopts |= MNTOPT_ACL;
1338 /* LOD might calculate the max stripe count based on max_ea_size,
1339 * so we need take account in the overhead as well,
1340 * xattr_header + magic + xattr_entry_head */
1341 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1342 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1344 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1345 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1346 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1350 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1354 * Concurrency: shouldn't matter.
1356 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1358 CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1359 return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1363 * Start commit for OSD device.
1365 * An implementation of dt_commit_async method for OSD device.
1366 * Asychronously starts underlayng fs sync and thereby a transaction
1369 * \param env environment
1370 * \param d dt device
1372 * \see dt_device_operations
1374 static int osd_commit_async(const struct lu_env *env,
1375 struct dt_device *d)
1377 struct super_block *s = osd_sb(osd_dt_dev(d));
1380 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1381 RETURN(s->s_op->sync_fs(s, 0));
1385 * Concurrency: shouldn't matter.
1388 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1390 struct super_block *sb = osd_sb(osd_dt_dev(d));
1391 struct block_device *dev = sb->s_bdev;
1392 #ifdef HAVE_DEV_SET_RDONLY
1393 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1396 int rc = -EOPNOTSUPP;
1400 #ifdef HAVE_DEV_SET_RDONLY
1401 CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1403 if (jdev && (jdev != dev)) {
1404 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1406 dev_set_rdonly(jdev);
1408 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1409 dev_set_rdonly(dev);
1411 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1412 osd_dt_dev(d)->od_svname, (long)dev, rc);
1418 * Concurrency: serialization provided by callers.
1420 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1421 int mode, unsigned long timeout, __u32 alg,
1422 struct lustre_capa_key *keys)
1424 struct osd_device *dev = osd_dt_dev(d);
1427 dev->od_fl_capa = mode;
1428 dev->od_capa_timeout = timeout;
1429 dev->od_capa_alg = alg;
1430 dev->od_capa_keys = keys;
1435 * Note: we do not count into QUOTA here.
1436 * If we mount with --data_journal we may need more.
1438 const int osd_dto_credits_noquota[DTO_NR] = {
1441 * INDEX_EXTRA_TRANS_BLOCKS(8) +
1442 * SINGLEDATA_TRANS_BLOCKS(8)
1443 * XXX Note: maybe iam need more, since iam have more level than
1446 [DTO_INDEX_INSERT] = 16,
1449 * just modify a single entry, probably merge few within a block
1451 [DTO_INDEX_DELETE] = 1,
1455 [DTO_INDEX_UPDATE] = 16,
1457 * 4(inode, inode bits, groups, GDT)
1458 * notice: OI updates are counted separately with DTO_INDEX_INSERT
1460 [DTO_OBJECT_CREATE] = 4,
1462 * 4(inode, inode bits, groups, GDT)
1463 * notice: OI updates are counted separately with DTO_INDEX_DELETE
1465 [DTO_OBJECT_DELETE] = 4,
1467 * Attr set credits (inode)
1469 [DTO_ATTR_SET_BASE] = 1,
1471 * Xattr set. The same as xattr of EXT3.
1472 * DATA_TRANS_BLOCKS(14)
1473 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1474 * are also counted in. Do not know why?
1476 [DTO_XATTR_SET] = 14,
1478 * credits for inode change during write.
1480 [DTO_WRITE_BASE] = 3,
1482 * credits for single block write.
1484 [DTO_WRITE_BLOCK] = 14,
1486 * Attr set credits for chown.
1487 * This is extra credits for setattr, and it is null without quota
1489 [DTO_ATTR_SET_CHOWN] = 0
1492 static const struct dt_device_operations osd_dt_ops = {
1493 .dt_root_get = osd_root_get,
1494 .dt_statfs = osd_statfs,
1495 .dt_trans_create = osd_trans_create,
1496 .dt_trans_start = osd_trans_start,
1497 .dt_trans_stop = osd_trans_stop,
1498 .dt_trans_cb_add = osd_trans_cb_add,
1499 .dt_conf_get = osd_conf_get,
1500 .dt_sync = osd_sync,
1502 .dt_commit_async = osd_commit_async,
1503 .dt_init_capa_ctxt = osd_init_capa_ctxt,
1506 static void osd_object_read_lock(const struct lu_env *env,
1507 struct dt_object *dt, unsigned role)
1509 struct osd_object *obj = osd_dt_obj(dt);
1510 struct osd_thread_info *oti = osd_oti_get(env);
1512 LINVRNT(osd_invariant(obj));
1514 LASSERT(obj->oo_owner != env);
1515 down_read_nested(&obj->oo_sem, role);
1517 LASSERT(obj->oo_owner == NULL);
1521 static void osd_object_write_lock(const struct lu_env *env,
1522 struct dt_object *dt, unsigned role)
1524 struct osd_object *obj = osd_dt_obj(dt);
1525 struct osd_thread_info *oti = osd_oti_get(env);
1527 LINVRNT(osd_invariant(obj));
1529 LASSERT(obj->oo_owner != env);
1530 down_write_nested(&obj->oo_sem, role);
1532 LASSERT(obj->oo_owner == NULL);
1533 obj->oo_owner = env;
1537 static void osd_object_read_unlock(const struct lu_env *env,
1538 struct dt_object *dt)
1540 struct osd_object *obj = osd_dt_obj(dt);
1541 struct osd_thread_info *oti = osd_oti_get(env);
1543 LINVRNT(osd_invariant(obj));
1545 LASSERT(oti->oti_r_locks > 0);
1547 up_read(&obj->oo_sem);
1550 static void osd_object_write_unlock(const struct lu_env *env,
1551 struct dt_object *dt)
1553 struct osd_object *obj = osd_dt_obj(dt);
1554 struct osd_thread_info *oti = osd_oti_get(env);
1556 LINVRNT(osd_invariant(obj));
1558 LASSERT(obj->oo_owner == env);
1559 LASSERT(oti->oti_w_locks > 0);
1561 obj->oo_owner = NULL;
1562 up_write(&obj->oo_sem);
1565 static int osd_object_write_locked(const struct lu_env *env,
1566 struct dt_object *dt)
1568 struct osd_object *obj = osd_dt_obj(dt);
1570 LINVRNT(osd_invariant(obj));
1572 return obj->oo_owner == env;
1575 static int capa_is_sane(const struct lu_env *env,
1576 struct osd_device *dev,
1577 struct lustre_capa *capa,
1578 struct lustre_capa_key *keys)
1580 struct osd_thread_info *oti = osd_oti_get(env);
1581 struct lustre_capa *tcapa = &oti->oti_capa;
1582 struct obd_capa *oc;
1586 oc = capa_lookup(dev->od_capa_hash, capa, 0);
1588 if (capa_is_expired(oc)) {
1589 DEBUG_CAPA(D_ERROR, capa, "expired");
1596 if (capa_is_expired_sec(capa)) {
1597 DEBUG_CAPA(D_ERROR, capa, "expired");
1601 spin_lock(&capa_lock);
1602 for (i = 0; i < 2; i++) {
1603 if (keys[i].lk_keyid == capa->lc_keyid) {
1604 oti->oti_capa_key = keys[i];
1608 spin_unlock(&capa_lock);
1611 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1615 rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1619 if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1620 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1624 oc = capa_add(dev->od_capa_hash, capa);
1630 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1631 struct lustre_capa *capa, __u64 opc)
1633 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1634 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
1635 struct lu_capainfo *lci;
1638 if (!osd->od_fl_capa)
1641 if (capa == BYPASS_CAPA)
1644 lci = lu_capainfo_get(env);
1645 if (unlikely(lci == NULL))
1648 if (lci->lci_auth == LC_ID_NONE)
1652 CERROR("%s: no capability provided for FID "DFID": rc = %d\n",
1653 osd_name(osd), PFID(fid), -EACCES);
1657 if (!lu_fid_eq(fid, &capa->lc_fid)) {
1658 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1663 if (!capa_opc_supported(capa, opc)) {
1664 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1668 rc = capa_is_sane(env, osd, capa, osd->od_capa_keys);
1670 DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc);
1677 static struct timespec *osd_inode_time(const struct lu_env *env,
1678 struct inode *inode, __u64 seconds)
1680 struct osd_thread_info *oti = osd_oti_get(env);
1681 struct timespec *t = &oti->oti_time;
1683 t->tv_sec = seconds;
1685 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1690 static void osd_inode_getattr(const struct lu_env *env,
1691 struct inode *inode, struct lu_attr *attr)
1693 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1694 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1695 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1698 attr->la_atime = LTIME_S(inode->i_atime);
1699 attr->la_mtime = LTIME_S(inode->i_mtime);
1700 attr->la_ctime = LTIME_S(inode->i_ctime);
1701 attr->la_mode = inode->i_mode;
1702 attr->la_size = i_size_read(inode);
1703 attr->la_blocks = inode->i_blocks;
1704 attr->la_uid = i_uid_read(inode);
1705 attr->la_gid = i_gid_read(inode);
1706 attr->la_flags = LDISKFS_I(inode)->i_flags;
1707 attr->la_nlink = inode->i_nlink;
1708 attr->la_rdev = inode->i_rdev;
1709 attr->la_blksize = 1 << inode->i_blkbits;
1710 attr->la_blkbits = inode->i_blkbits;
1713 static int osd_attr_get(const struct lu_env *env,
1714 struct dt_object *dt,
1715 struct lu_attr *attr,
1716 struct lustre_capa *capa)
1718 struct osd_object *obj = osd_dt_obj(dt);
1720 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1721 LINVRNT(osd_invariant(obj));
1723 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1726 spin_lock(&obj->oo_guard);
1727 osd_inode_getattr(env, obj->oo_inode, attr);
1728 spin_unlock(&obj->oo_guard);
1732 static int osd_declare_attr_set(const struct lu_env *env,
1733 struct dt_object *dt,
1734 const struct lu_attr *attr,
1735 struct thandle *handle)
1737 struct osd_thandle *oh;
1738 struct osd_object *obj;
1739 struct osd_thread_info *info = osd_oti_get(env);
1740 struct lquota_id_info *qi = &info->oti_qi;
1748 LASSERT(dt != NULL);
1749 LASSERT(handle != NULL);
1751 obj = osd_dt_obj(dt);
1752 LASSERT(osd_invariant(obj));
1754 oh = container_of0(handle, struct osd_thandle, ot_super);
1755 LASSERT(oh->ot_handle == NULL);
1757 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1758 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1760 if (attr == NULL || obj->oo_inode == NULL)
1763 bspace = obj->oo_inode->i_blocks;
1764 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1765 bspace = toqb(bspace);
1767 /* Changing ownership is always preformed by super user, it should not
1770 * We still need to call the osd_declare_qid() to calculate the journal
1771 * credits for updating quota accounting files and to trigger quota
1772 * space adjustment once the operation is completed.*/
1773 if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
1775 uid = i_uid_read(obj->oo_inode);
1776 qi->lqi_type = USRQUOTA;
1777 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
1778 /* inode accounting */
1779 qi->lqi_is_blk = false;
1781 /* one more inode for the new uid ... */
1782 qi->lqi_id.qid_uid = attr->la_uid;
1784 /* Reserve credits for the new uid */
1785 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1786 if (rc == -EDQUOT || rc == -EINPROGRESS)
1791 /* and one less inode for the current uid */
1792 qi->lqi_id.qid_uid = uid;
1794 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1795 if (rc == -EDQUOT || rc == -EINPROGRESS)
1800 /* block accounting */
1801 qi->lqi_is_blk = true;
1803 /* more blocks for the new uid ... */
1804 qi->lqi_id.qid_uid = attr->la_uid;
1805 qi->lqi_space = bspace;
1807 * Credits for the new uid has been reserved, re-use "obj"
1808 * to save credit reservation.
1810 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1811 if (rc == -EDQUOT || rc == -EINPROGRESS)
1816 /* and finally less blocks for the current uid */
1817 qi->lqi_id.qid_uid = uid;
1818 qi->lqi_space = -bspace;
1819 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1820 if (rc == -EDQUOT || rc == -EINPROGRESS)
1826 gid = i_gid_read(obj->oo_inode);
1827 qi->lqi_type = GRPQUOTA;
1828 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
1830 /* inode accounting */
1831 qi->lqi_is_blk = false;
1833 /* one more inode for the new gid ... */
1834 qi->lqi_id.qid_gid = attr->la_gid;
1836 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1837 if (rc == -EDQUOT || rc == -EINPROGRESS)
1842 /* and one less inode for the current gid */
1843 qi->lqi_id.qid_gid = gid;
1845 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1846 if (rc == -EDQUOT || rc == -EINPROGRESS)
1851 /* block accounting */
1852 qi->lqi_is_blk = true;
1854 /* more blocks for the new gid ... */
1855 qi->lqi_id.qid_gid = attr->la_gid;
1856 qi->lqi_space = bspace;
1857 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1858 if (rc == -EDQUOT || rc == -EINPROGRESS)
1863 /* and finally less blocks for the current gid */
1864 qi->lqi_id.qid_gid = gid;
1865 qi->lqi_space = -bspace;
1866 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1867 if (rc == -EDQUOT || rc == -EINPROGRESS)
1876 static int osd_inode_setattr(const struct lu_env *env,
1877 struct inode *inode, const struct lu_attr *attr)
1879 __u64 bits = attr->la_valid;
1881 /* Only allow set size for regular file */
1882 if (!S_ISREG(inode->i_mode))
1883 bits &= ~(LA_SIZE | LA_BLOCKS);
1888 if (bits & LA_ATIME)
1889 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
1890 if (bits & LA_CTIME)
1891 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
1892 if (bits & LA_MTIME)
1893 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
1894 if (bits & LA_SIZE) {
1895 LDISKFS_I(inode)->i_disksize = attr->la_size;
1896 i_size_write(inode, attr->la_size);
1900 /* OSD should not change "i_blocks" which is used by quota.
1901 * "i_blocks" should be changed by ldiskfs only. */
1902 if (bits & LA_BLOCKS)
1903 inode->i_blocks = attr->la_blocks;
1906 inode->i_mode = (inode->i_mode & S_IFMT) |
1907 (attr->la_mode & ~S_IFMT);
1909 i_uid_write(inode, attr->la_uid);
1911 i_gid_write(inode, attr->la_gid);
1912 if (bits & LA_NLINK)
1913 set_nlink(inode, attr->la_nlink);
1915 inode->i_rdev = attr->la_rdev;
1917 if (bits & LA_FLAGS) {
1918 /* always keep S_NOCMTIME */
1919 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1925 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1927 if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1928 (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1932 ll_vfs_dq_init(inode);
1934 if (attr->la_valid & LA_UID)
1935 iattr.ia_valid |= ATTR_UID;
1936 if (attr->la_valid & LA_GID)
1937 iattr.ia_valid |= ATTR_GID;
1938 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1939 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1941 rc = ll_vfs_dq_transfer(inode, &iattr);
1943 CERROR("%s: quota transfer failed: rc = %d. Is quota "
1944 "enforcement enabled on the ldiskfs "
1945 "filesystem?\n", inode->i_sb->s_id, rc);
1952 static int osd_attr_set(const struct lu_env *env,
1953 struct dt_object *dt,
1954 const struct lu_attr *attr,
1955 struct thandle *handle,
1956 struct lustre_capa *capa)
1958 struct osd_object *obj = osd_dt_obj(dt);
1959 struct inode *inode;
1962 LASSERT(handle != NULL);
1963 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1964 LASSERT(osd_invariant(obj));
1966 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1969 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1971 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1972 struct osd_thread_info *oti = osd_oti_get(env);
1973 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
1974 struct lu_fid *fid1 = &oti->oti_fid;
1975 struct osd_inode_id *id = &oti->oti_id;
1976 struct iam_path_descr *ipd;
1977 struct iam_container *bag;
1978 struct osd_thandle *oh;
1981 fid_cpu_to_be(fid1, fid0);
1982 memset(id, 1, sizeof(*id));
1983 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1984 fid0)->oi_dir.od_container;
1985 ipd = osd_idx_ipd_get(env, bag);
1986 if (unlikely(ipd == NULL))
1989 oh = container_of0(handle, struct osd_thandle, ot_super);
1990 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1991 (const struct iam_rec *)id, ipd);
1992 osd_ipd_put(env, bag, ipd);
1993 return(rc > 0 ? 0 : rc);
1996 inode = obj->oo_inode;
1998 rc = osd_quota_transfer(inode, attr);
2002 spin_lock(&obj->oo_guard);
2003 rc = osd_inode_setattr(env, inode, attr);
2004 spin_unlock(&obj->oo_guard);
2007 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2011 struct dentry *osd_child_dentry_get(const struct lu_env *env,
2012 struct osd_object *obj,
2013 const char *name, const int namelen)
2015 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2018 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2019 umode_t mode, struct dt_allocation_hint *hint,
2023 struct osd_device *osd = osd_obj2dev(obj);
2024 struct osd_thandle *oth;
2025 struct dt_object *parent = NULL;
2026 struct inode *inode;
2028 LINVRNT(osd_invariant(obj));
2029 LASSERT(obj->oo_inode == NULL);
2030 LASSERT(obj->oo_hl_head == NULL);
2032 if (S_ISDIR(mode) && ldiskfs_pdo) {
2033 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2034 if (obj->oo_hl_head == NULL)
2038 oth = container_of(th, struct osd_thandle, ot_super);
2039 LASSERT(oth->ot_handle->h_transaction != NULL);
2041 if (hint && hint->dah_parent)
2042 parent = hint->dah_parent;
2044 inode = ldiskfs_create_inode(oth->ot_handle,
2045 parent ? osd_dt_obj(parent)->oo_inode :
2046 osd_sb(osd)->s_root->d_inode,
2048 if (!IS_ERR(inode)) {
2049 /* Do not update file c/mtime in ldiskfs.
2050 * NB: don't need any lock because no contention at this
2052 inode->i_flags |= S_NOCMTIME;
2054 /* For new created object, it must be consistent,
2055 * and it is unnecessary to scrub against it. */
2056 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2057 obj->oo_inode = inode;
2060 if (obj->oo_hl_head != NULL) {
2061 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2062 obj->oo_hl_head = NULL;
2064 result = PTR_ERR(inode);
2066 LINVRNT(osd_invariant(obj));
2074 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2075 struct lu_attr *attr,
2076 struct dt_allocation_hint *hint,
2077 struct dt_object_format *dof,
2081 struct osd_thandle *oth;
2082 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2084 LASSERT(S_ISDIR(attr->la_mode));
2086 oth = container_of(th, struct osd_thandle, ot_super);
2087 LASSERT(oth->ot_handle->h_transaction != NULL);
2088 result = osd_mkfile(info, obj, mode, hint, th);
2093 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2094 struct lu_attr *attr,
2095 struct dt_allocation_hint *hint,
2096 struct dt_object_format *dof,
2100 struct osd_thandle *oth;
2101 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2103 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2105 LASSERT(S_ISREG(attr->la_mode));
2107 oth = container_of(th, struct osd_thandle, ot_super);
2108 LASSERT(oth->ot_handle->h_transaction != NULL);
2110 result = osd_mkfile(info, obj, mode, hint, th);
2112 LASSERT(obj->oo_inode != NULL);
2113 if (feat->dif_flags & DT_IND_VARKEY)
2114 result = iam_lvar_create(obj->oo_inode,
2115 feat->dif_keysize_max,
2117 feat->dif_recsize_max,
2120 result = iam_lfix_create(obj->oo_inode,
2121 feat->dif_keysize_max,
2123 feat->dif_recsize_max,
2130 static int osd_mkreg(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_ISREG(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_mksym(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 LASSERT(S_ISLNK(attr->la_mode));
2148 return osd_mkfile(info, obj, (attr->la_mode &
2149 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2152 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2153 struct lu_attr *attr,
2154 struct dt_allocation_hint *hint,
2155 struct dt_object_format *dof,
2158 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2161 LINVRNT(osd_invariant(obj));
2162 LASSERT(obj->oo_inode == NULL);
2163 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2164 S_ISFIFO(mode) || S_ISSOCK(mode));
2166 result = osd_mkfile(info, obj, mode, hint, th);
2168 LASSERT(obj->oo_inode != NULL);
2170 * This inode should be marked dirty for i_rdev. Currently
2171 * that is done in the osd_attr_init().
2173 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2176 LINVRNT(osd_invariant(obj));
2180 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2182 struct dt_allocation_hint *hint,
2183 struct dt_object_format *dof,
2186 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2188 osd_obj_type_f result;
2204 result = osd_mk_index;
2215 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2216 struct dt_object *parent, struct dt_object *child,
2221 ah->dah_parent = parent;
2222 ah->dah_mode = child_mode;
2225 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2226 struct lu_attr *attr, struct dt_object_format *dof)
2228 struct inode *inode = obj->oo_inode;
2229 __u64 valid = attr->la_valid;
2232 attr->la_valid &= ~(LA_TYPE | LA_MODE);
2234 if (dof->dof_type != DFT_NODE)
2235 attr->la_valid &= ~LA_RDEV;
2236 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2237 attr->la_valid &= ~LA_ATIME;
2238 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2239 attr->la_valid &= ~LA_CTIME;
2240 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2241 attr->la_valid &= ~LA_MTIME;
2243 result = osd_quota_transfer(inode, attr);
2247 if (attr->la_valid != 0) {
2248 result = osd_inode_setattr(info->oti_env, inode, attr);
2250 * The osd_inode_setattr() should always succeed here. The
2251 * only error that could be returned is EDQUOT when we are
2252 * trying to change the UID or GID of the inode. However, this
2253 * should not happen since quota enforcement is no longer
2254 * enabled on ldiskfs (lquota takes care of it).
2256 LASSERTF(result == 0, "%d\n", result);
2257 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2260 attr->la_valid = valid;
2264 * Helper function for osd_object_create()
2266 * \retval 0, on success
2268 static int __osd_object_create(struct osd_thread_info *info,
2269 struct osd_object *obj, struct lu_attr *attr,
2270 struct dt_allocation_hint *hint,
2271 struct dt_object_format *dof,
2277 /* we drop umask so that permissions we pass are not affected */
2278 umask = current->fs->umask;
2279 current->fs->umask = 0;
2281 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2284 osd_attr_init(info, obj, attr, dof);
2285 osd_object_init0(obj);
2287 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2288 unlock_new_inode(obj->oo_inode);
2291 /* restore previous umask value */
2292 current->fs->umask = umask;
2298 * Helper function for osd_object_create()
2300 * \retval 0, on success
2302 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2303 const struct lu_fid *fid, struct thandle *th)
2305 struct osd_thread_info *info = osd_oti_get(env);
2306 struct osd_inode_id *id = &info->oti_id;
2307 struct osd_device *osd = osd_obj2dev(obj);
2308 struct osd_thandle *oh;
2310 LASSERT(obj->oo_inode != NULL);
2312 oh = container_of0(th, struct osd_thandle, ot_super);
2313 LASSERT(oh->ot_handle);
2315 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2316 return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2319 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2320 obd_seq seq, struct lu_seq_range *range)
2322 struct seq_server_site *ss = osd_seq_site(osd);
2324 if (fid_seq_is_idif(seq)) {
2325 fld_range_set_ost(range);
2326 range->lsr_index = idif_ost_idx(seq);
2330 if (!fid_seq_in_fldb(seq)) {
2331 fld_range_set_mdt(range);
2333 /* FIXME: If ss is NULL, it suppose not get lsr_index
2335 range->lsr_index = ss->ss_node_id;
2339 LASSERT(ss != NULL);
2340 fld_range_set_any(range);
2341 /* OSD will only do local fld lookup */
2342 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2346 * Concurrency: no external locking is necessary.
2348 static int osd_declare_object_create(const struct lu_env *env,
2349 struct dt_object *dt,
2350 struct lu_attr *attr,
2351 struct dt_allocation_hint *hint,
2352 struct dt_object_format *dof,
2353 struct thandle *handle)
2355 struct osd_thandle *oh;
2359 LASSERT(handle != NULL);
2361 oh = container_of0(handle, struct osd_thandle, ot_super);
2362 LASSERT(oh->ot_handle == NULL);
2364 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2365 osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2366 /* Reuse idle OI block may cause additional one OI block
2368 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2369 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2371 /* If this is directory, then we expect . and .. to be inserted as
2372 * well. The one directory block always needs to be created for the
2373 * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2374 * block), there is no danger of needing a tree for the first block.
2376 if (attr && S_ISDIR(attr->la_mode)) {
2377 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2378 osd_dto_credits_noquota[DTO_WRITE_BASE]);
2379 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2385 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2386 osd_dt_obj(dt), false, NULL, false);
2393 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2394 struct lu_attr *attr,
2395 struct dt_allocation_hint *hint,
2396 struct dt_object_format *dof,
2399 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2400 struct osd_object *obj = osd_dt_obj(dt);
2401 struct osd_thread_info *info = osd_oti_get(env);
2406 LINVRNT(osd_invariant(obj));
2407 LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2408 LASSERT(osd_write_locked(env, obj));
2409 LASSERT(th != NULL);
2411 if (unlikely(fid_is_acct(fid)))
2412 /* Quota files can't be created from the kernel any more,
2413 * 'tune2fs -O quota' will take care of creating them */
2416 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2417 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2419 result = __osd_object_create(info, obj, attr, hint, dof, th);
2421 result = __osd_oi_insert(env, obj, fid, th);
2423 LASSERT(ergo(result == 0,
2424 dt_object_exists(dt) && !dt_object_remote(dt)));
2426 LASSERT(osd_invariant(obj));
2431 * Called to destroy on-disk representation of the object
2433 * Concurrency: must be locked
2435 static int osd_declare_object_destroy(const struct lu_env *env,
2436 struct dt_object *dt,
2439 struct osd_object *obj = osd_dt_obj(dt);
2440 struct inode *inode = obj->oo_inode;
2441 struct osd_thandle *oh;
2445 oh = container_of0(th, struct osd_thandle, ot_super);
2446 LASSERT(oh->ot_handle == NULL);
2449 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2450 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2451 /* Recycle idle OI leaf may cause additional three OI blocks
2453 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2454 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2455 /* one less inode */
2456 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2457 -1, oh, obj, false, NULL, false);
2460 /* data to be truncated */
2461 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2462 0, oh, obj, true, NULL, false);
2466 static int osd_object_destroy(const struct lu_env *env,
2467 struct dt_object *dt,
2470 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2471 struct osd_object *obj = osd_dt_obj(dt);
2472 struct inode *inode = obj->oo_inode;
2473 struct osd_device *osd = osd_obj2dev(obj);
2474 struct osd_thandle *oh;
2478 oh = container_of0(th, struct osd_thandle, ot_super);
2479 LASSERT(oh->ot_handle);
2481 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2483 if (unlikely(fid_is_acct(fid)))
2486 if (S_ISDIR(inode->i_mode)) {
2487 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2488 inode->i_nlink == 2);
2489 /* it will check/delete the inode from remote parent,
2490 * how to optimize it? unlink performance impaction XXX */
2491 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2492 if (result != 0 && result != -ENOENT) {
2493 CERROR("%s: delete inode "DFID": rc = %d\n",
2494 osd_name(osd), PFID(fid), result);
2496 spin_lock(&obj->oo_guard);
2498 spin_unlock(&obj->oo_guard);
2499 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2502 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2504 result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2507 /* XXX: add to ext3 orphan list */
2508 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2510 /* not needed in the cache anymore */
2511 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2517 * Put the fid into lustre_mdt_attrs, and then place the structure
2518 * inode's ea. This fid should not be altered during the life time
2521 * \retval +ve, on success
2522 * \retval -ve, on error
2524 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2526 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2527 const struct lu_fid *fid, __u32 compat, __u32 incompat)
2529 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2533 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2536 lustre_lma_init(lma, fid, compat, incompat);
2537 lustre_lma_swab(lma);
2539 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2541 /* LMA may already exist, but we need to check that all the
2542 * desired compat/incompat flags have been added. */
2543 if (unlikely(rc == -EEXIST)) {
2544 if (compat == 0 && incompat == 0)
2547 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2548 XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2553 lustre_lma_swab(lma);
2554 if (!(~lma->lma_compat & compat) &&
2555 !(~lma->lma_incompat & incompat))
2558 lma->lma_compat |= compat;
2559 lma->lma_incompat |= incompat;
2560 lustre_lma_swab(lma);
2561 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2562 sizeof(*lma), XATTR_REPLACE);
2569 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2570 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2571 * To have compatilibility with 1.8 ldiskfs driver we need to have
2572 * magic number at start of fid data.
2573 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2576 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2577 const struct lu_fid *fid)
2579 if (!fid_is_namespace_visible(fid) ||
2580 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2581 param->edp_magic = 0;
2585 param->edp_magic = LDISKFS_LUFID_MAGIC;
2586 param->edp_len = sizeof(struct lu_fid) + 1;
2587 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2591 * Try to read the fid from inode ea into dt_rec.
2593 * \param fid object fid.
2595 * \retval 0 on success
2597 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2598 __u32 ino, struct lu_fid *fid,
2599 struct osd_inode_id *id)
2601 struct osd_thread_info *info = osd_oti_get(env);
2602 struct inode *inode;
2605 osd_id_gen(id, ino, OSD_OII_NOGEN);
2606 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2608 RETURN(PTR_ERR(inode));
2614 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2616 struct inode *parent_dir,
2617 const struct lu_fid *dot_fid,
2618 const struct lu_fid *dot_dot_fid,
2619 struct osd_thandle *oth)
2621 struct ldiskfs_dentry_param *dot_ldp;
2622 struct ldiskfs_dentry_param *dot_dot_ldp;
2624 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2625 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2627 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2628 dot_ldp->edp_magic = 0;
2629 return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2630 dir, dot_ldp, dot_dot_ldp);
2634 * Create an local agent inode for remote entry
2636 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2637 struct osd_device *osd,
2638 struct osd_object *pobj,
2639 const struct lu_fid *fid,
2643 struct osd_thread_info *info = osd_oti_get(env);
2644 struct inode *local;
2645 struct osd_thandle *oh;
2650 oh = container_of(th, struct osd_thandle, ot_super);
2651 LASSERT(oh->ot_handle->h_transaction != NULL);
2653 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2654 if (IS_ERR(local)) {
2655 CERROR("%s: create local error %d\n", osd_name(osd),
2656 (int)PTR_ERR(local));
2660 /* Set special LMA flag for local agent inode */
2661 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2663 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2664 osd_name(osd), PFID(fid), rc);
2665 RETURN(ERR_PTR(rc));
2671 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2672 lu_object_fid(&pobj->oo_dt.do_lu),
2675 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2676 osd_name(osd), PFID(fid), rc);
2677 RETURN(ERR_PTR(rc));
2684 * Delete local agent inode for remote entry
2686 static int osd_delete_local_agent_inode(const struct lu_env *env,
2687 struct osd_device *osd,
2688 const struct lu_fid *fid,
2689 __u32 ino, struct osd_thandle *oh)
2691 struct osd_thread_info *oti = osd_oti_get(env);
2692 struct osd_inode_id *id = &oti->oti_id;
2693 struct inode *inode;
2696 id->oii_ino = le32_to_cpu(ino);
2697 id->oii_gen = OSD_OII_NOGEN;
2698 inode = osd_iget(oti, osd, id);
2699 if (IS_ERR(inode)) {
2700 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2701 PFID(fid), id->oii_ino, id->oii_gen);
2702 RETURN(PTR_ERR(inode));
2706 mark_inode_dirty(inode);
2707 CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2708 osd_name(osd), PFID(fid), inode->i_ino);
2714 * OSD layer object create function for interoperability mode (b11826).
2715 * This is mostly similar to osd_object_create(). Only difference being, fid is
2716 * inserted into inode ea here.
2718 * \retval 0, on success
2719 * \retval -ve, on error
2721 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2722 struct lu_attr *attr,
2723 struct dt_allocation_hint *hint,
2724 struct dt_object_format *dof,
2727 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2728 struct osd_object *obj = osd_dt_obj(dt);
2729 struct osd_thread_info *info = osd_oti_get(env);
2734 LASSERT(osd_invariant(obj));
2735 LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2736 LASSERT(osd_write_locked(env, obj));
2737 LASSERT(th != NULL);
2739 if (unlikely(fid_is_acct(fid)))
2740 /* Quota files can't be created from the kernel any more,
2741 * 'tune2fs -O quota' will take care of creating them */
2744 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2745 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2747 result = __osd_object_create(info, obj, attr, hint, dof, th);
2749 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2750 fid_is_on_ost(info, osd_obj2dev(obj),
2751 fid, OI_CHECK_FLD) ?
2752 LMAC_FID_ON_OST : 0, 0);
2755 result = __osd_oi_insert(env, obj, fid, th);
2757 LASSERT(ergo(result == 0,
2758 dt_object_exists(dt) && !dt_object_remote(dt)));
2759 LINVRNT(osd_invariant(obj));
2763 static int osd_declare_object_ref_add(const struct lu_env *env,
2764 struct dt_object *dt,
2765 struct thandle *handle)
2767 struct osd_thandle *oh;
2769 /* it's possible that object doesn't exist yet */
2770 LASSERT(handle != NULL);
2772 oh = container_of0(handle, struct osd_thandle, ot_super);
2773 LASSERT(oh->ot_handle == NULL);
2775 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2776 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2782 * Concurrency: @dt is write locked.
2784 static int osd_object_ref_add(const struct lu_env *env,
2785 struct dt_object *dt, struct thandle *th)
2787 struct osd_object *obj = osd_dt_obj(dt);
2788 struct inode *inode = obj->oo_inode;
2789 struct osd_thandle *oh;
2792 LINVRNT(osd_invariant(obj));
2793 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2794 LASSERT(osd_write_locked(env, obj));
2795 LASSERT(th != NULL);
2797 oh = container_of0(th, struct osd_thandle, ot_super);
2798 LASSERT(oh->ot_handle != NULL);
2800 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2802 CDEBUG(D_INODE, DFID" increase nlink %d\n",
2803 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2805 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2806 * (65000) subdirectories by storing "1" in i_nlink if the link count
2807 * would otherwise overflow. Directory tranversal tools understand
2808 * that (st_nlink == 1) indicates that the filesystem dose not track
2809 * hard links count on the directory, and will not abort subdirectory
2810 * scanning early once (st_nlink - 2) subdirs have been found.
2812 * This also has to properly handle the case of inodes with nlink == 0
2813 * in case they are being linked into the PENDING directory
2815 spin_lock(&obj->oo_guard);
2816 if (unlikely(inode->i_nlink == 0))
2817 /* inc_nlink from 0 may cause WARN_ON */
2818 set_nlink(inode, 1);
2820 ldiskfs_inc_count(oh->ot_handle, inode);
2821 if (!S_ISDIR(inode->i_mode))
2822 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2824 spin_unlock(&obj->oo_guard);
2826 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2827 LINVRNT(osd_invariant(obj));
2832 static int osd_declare_object_ref_del(const struct lu_env *env,
2833 struct dt_object *dt,
2834 struct thandle *handle)
2836 struct osd_thandle *oh;
2838 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2839 LASSERT(handle != NULL);
2841 oh = container_of0(handle, struct osd_thandle, ot_super);
2842 LASSERT(oh->ot_handle == NULL);
2844 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2845 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2851 * Concurrency: @dt is write locked.
2853 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2856 struct osd_object *obj = osd_dt_obj(dt);
2857 struct inode *inode = obj->oo_inode;
2858 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
2859 struct osd_thandle *oh;
2861 LINVRNT(osd_invariant(obj));
2862 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2863 LASSERT(osd_write_locked(env, obj));
2864 LASSERT(th != NULL);
2866 oh = container_of0(th, struct osd_thandle, ot_super);
2867 LASSERT(oh->ot_handle != NULL);
2869 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2871 spin_lock(&obj->oo_guard);
2872 /* That can be result of upgrade from old Lustre version and
2873 * applied only to local files. Just skip this ref_del call.
2874 * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2875 if (inode->i_nlink == 0) {
2876 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2877 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2878 ", maybe an upgraded file? (LU-3915)\n",
2879 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2880 spin_unlock(&obj->oo_guard);
2884 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2885 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2887 ldiskfs_dec_count(oh->ot_handle, inode);
2888 spin_unlock(&obj->oo_guard);
2890 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2891 LINVRNT(osd_invariant(obj));
2897 * Get the 64-bit version for an inode.
2899 static int osd_object_version_get(const struct lu_env *env,
2900 struct dt_object *dt, dt_obj_version_t *ver)
2902 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2904 CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2905 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2906 *ver = LDISKFS_I(inode)->i_fs_version;
2911 * Concurrency: @dt is read locked.
2913 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2914 struct lu_buf *buf, const char *name,
2915 struct lustre_capa *capa)
2917 struct osd_object *obj = osd_dt_obj(dt);
2918 struct inode *inode = obj->oo_inode;
2919 struct osd_thread_info *info = osd_oti_get(env);
2920 struct dentry *dentry = &info->oti_obj_dentry;
2922 /* version get is not real XATTR but uses xattr API */
2923 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2924 /* for version we are just using xattr API but change inode
2926 if (buf->lb_len == 0)
2927 return sizeof(dt_obj_version_t);
2929 if (buf->lb_len < sizeof(dt_obj_version_t))
2932 osd_object_version_get(env, dt, buf->lb_buf);
2934 return sizeof(dt_obj_version_t);
2937 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2938 LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2940 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2943 return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2947 static int osd_declare_xattr_set(const struct lu_env *env,
2948 struct dt_object *dt,
2949 const struct lu_buf *buf, const char *name,
2950 int fl, struct thandle *handle)
2952 struct osd_thandle *oh;
2954 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
2956 LASSERT(handle != NULL);
2958 oh = container_of0(handle, struct osd_thandle, ot_super);
2959 LASSERT(oh->ot_handle == NULL);
2961 /* optimistic optimization: LMA is set first and usually fit inode */
2962 if (strcmp(name, XATTR_NAME_LMA) == 0) {
2963 if (dt_object_exists(dt))
2967 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2970 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2971 if (buf && buf->lb_len > sb->s_blocksize) {
2972 credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2973 sb->s_blocksize_bits;
2976 * xattr set may involve inode quota change, reserve credits for
2977 * dquot_initialize()
2979 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
2982 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2988 * Set the 64-bit version for object
2990 static void osd_object_version_set(const struct lu_env *env,
2991 struct dt_object *dt,
2992 dt_obj_version_t *new_version)
2994 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2996 CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2997 *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2999 LDISKFS_I(inode)->i_fs_version = *new_version;
3000 /** Version is set after all inode operations are finished,
3001 * so we should mark it dirty here */
3002 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3006 * Concurrency: @dt is write locked.
3008 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3009 const struct lu_buf *buf, const char *name, int fl,
3010 struct thandle *handle, struct lustre_capa *capa)
3012 struct osd_object *obj = osd_dt_obj(dt);
3013 struct inode *inode = obj->oo_inode;
3014 struct osd_thread_info *info = osd_oti_get(env);
3018 LASSERT(handle != NULL);
3020 /* version set is not real XATTR */
3021 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3022 /* for version we are just using xattr API but change inode
3024 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3025 osd_object_version_set(env, dt, buf->lb_buf);
3026 return sizeof(dt_obj_version_t);
3029 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3032 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n",
3033 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3035 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3036 if (fl & LU_XATTR_REPLACE)
3037 fs_flags |= XATTR_REPLACE;
3039 if (fl & LU_XATTR_CREATE)
3040 fs_flags |= XATTR_CREATE;
3042 if (strcmp(name, XATTR_NAME_LMV) == 0) {
3043 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3046 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3050 lma->lma_incompat |= LMAI_STRIPED;
3051 lustre_lma_swab(lma);
3052 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3053 sizeof(*lma), XATTR_REPLACE);
3058 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
3059 strcmp(name, XATTR_NAME_LINK) == 0)
3062 return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3067 * Concurrency: @dt is read locked.
3069 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3070 struct lu_buf *buf, struct lustre_capa *capa)
3072 struct osd_object *obj = osd_dt_obj(dt);
3073 struct inode *inode = obj->oo_inode;
3074 struct osd_thread_info *info = osd_oti_get(env);
3075 struct dentry *dentry = &info->oti_obj_dentry;
3077 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3078 LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
3080 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3083 dentry->d_inode = inode;
3084 dentry->d_sb = inode->i_sb;
3085 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3088 static int osd_declare_xattr_del(const struct lu_env *env,
3089 struct dt_object *dt, const char *name,
3090 struct thandle *handle)
3092 struct osd_thandle *oh;
3093 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3095 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3096 LASSERT(handle != NULL);
3098 oh = container_of0(handle, struct osd_thandle, ot_super);
3099 LASSERT(oh->ot_handle == NULL);
3101 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3102 osd_dto_credits_noquota[DTO_XATTR_SET]);
3104 * xattr del may involve inode quota change, reserve credits for
3105 * dquot_initialize()
3107 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3113 * Concurrency: @dt is write locked.
3115 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3116 const char *name, struct thandle *handle,
3117 struct lustre_capa *capa)
3119 struct osd_object *obj = osd_dt_obj(dt);
3120 struct inode *inode = obj->oo_inode;
3121 struct osd_thread_info *info = osd_oti_get(env);
3122 struct dentry *dentry = &info->oti_obj_dentry;
3125 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3126 LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
3127 LASSERT(handle != NULL);
3129 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3132 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3134 ll_vfs_dq_init(inode);
3135 dentry->d_inode = inode;
3136 dentry->d_sb = inode->i_sb;
3137 rc = inode->i_op->removexattr(dentry, name);
3141 static struct obd_capa *osd_capa_get(const struct lu_env *env,
3142 struct dt_object *dt,
3143 struct lustre_capa *old, __u64 opc)
3145 struct osd_thread_info *info = osd_oti_get(env);
3146 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3147 struct osd_object *obj = osd_dt_obj(dt);
3148 struct osd_device *osd = osd_obj2dev(obj);
3149 struct lustre_capa_key *key = &info->oti_capa_key;
3150 struct lustre_capa *capa = &info->oti_capa;
3151 struct obd_capa *oc;
3152 struct lu_capainfo *lci;
3156 if (!osd->od_fl_capa)
3157 RETURN(ERR_PTR(-ENOENT));
3159 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3160 LINVRNT(osd_invariant(obj));
3162 /* renewal sanity check */
3163 if (old && osd_object_auth(env, dt, old, opc))
3164 RETURN(ERR_PTR(-EACCES));
3166 lci = lu_capainfo_get(env);
3167 if (unlikely(lci == NULL))
3168 RETURN(ERR_PTR(-ENOENT));
3170 switch (lci->lci_auth) {
3174 capa->lc_uid = i_uid_read(obj->oo_inode);
3175 capa->lc_gid = i_gid_read(obj->oo_inode);
3176 capa->lc_flags = LC_ID_PLAIN;
3178 case LC_ID_CONVERT: {
3181 s[0] = i_uid_read(obj->oo_inode);
3182 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
3183 s[2] = i_uid_read(obj->oo_inode);
3184 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
3185 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
3187 RETURN(ERR_PTR(rc));
3189 capa->lc_uid = ((__u64)d[1] << 32) | d[0];
3190 capa->lc_gid = ((__u64)d[3] << 32) | d[2];
3191 capa->lc_flags = LC_ID_CONVERT;
3195 RETURN(ERR_PTR(-EINVAL));
3198 capa->lc_fid = *fid;
3200 capa->lc_flags |= osd->od_capa_alg << 24;
3201 capa->lc_timeout = osd->od_capa_timeout;
3202 capa->lc_expiry = 0;
3204 oc = capa_lookup(osd->od_capa_hash, capa, 1);
3206 LASSERT(!capa_is_expired(oc));
3210 spin_lock(&capa_lock);
3211 *key = osd->od_capa_keys[1];
3212 spin_unlock(&capa_lock);
3214 capa->lc_keyid = key->lk_keyid;
3215 capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout;
3217 rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
3219 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
3220 RETURN(ERR_PTR(rc));
3223 oc = capa_add(osd->od_capa_hash, capa);
3227 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3228 __u64 start, __u64 end)
3230 struct osd_object *obj = osd_dt_obj(dt);
3231 struct inode *inode = obj->oo_inode;
3232 struct osd_thread_info *info = osd_oti_get(env);
3233 struct dentry *dentry = &info->oti_obj_dentry;
3234 struct file *file = &info->oti_file;
3239 dentry->d_inode = inode;
3240 dentry->d_sb = inode->i_sb;
3241 file->f_dentry = dentry;
3242 file->f_mapping = inode->i_mapping;
3243 file->f_op = inode->i_fop;
3244 set_file_inode(file, inode);
3246 #ifdef HAVE_FILE_FSYNC_4ARGS
3247 rc = file->f_op->fsync(file, start, end, 0);
3248 #elif defined(HAVE_FILE_FSYNC_2ARGS)
3249 mutex_lock(&inode->i_mutex);
3250 rc = file->f_op->fsync(file, 0);
3251 mutex_unlock(&inode->i_mutex);
3253 mutex_lock(&inode->i_mutex);
3254 rc = file->f_op->fsync(file, dentry, 0);
3255 mutex_unlock(&inode->i_mutex);
3261 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
3264 struct osd_object *obj = osd_dt_obj(dt);
3267 *data = (void *)obj->oo_inode;
3275 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3276 const struct dt_index_features *feat)
3278 struct iam_descr *descr;
3280 if (osd_object_is_root(o))
3281 return feat == &dt_directory_features;
3283 LASSERT(o->oo_dir != NULL);
3285 descr = o->oo_dir->od_container.ic_descr;
3286 if (feat == &dt_directory_features) {
3287 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3293 feat->dif_keysize_min <= descr->id_key_size &&
3294 descr->id_key_size <= feat->dif_keysize_max &&
3295 feat->dif_recsize_min <= descr->id_rec_size &&
3296 descr->id_rec_size <= feat->dif_recsize_max &&
3297 !(feat->dif_flags & (DT_IND_VARKEY |
3298 DT_IND_VARREC | DT_IND_NONUNQ)) &&
3299 ergo(feat->dif_flags & DT_IND_UPDATE,
3300 1 /* XXX check that object (and file system) is
3305 static int osd_iam_container_init(const struct lu_env *env,
3306 struct osd_object *obj,
3307 struct osd_directory *dir)
3309 struct iam_container *bag = &dir->od_container;
3312 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3316 result = iam_container_setup(bag);
3318 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3320 iam_container_fini(bag);
3327 * Concurrency: no external locking is necessary.
3329 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3330 const struct dt_index_features *feat)
3334 struct osd_object *obj = osd_dt_obj(dt);
3336 LINVRNT(osd_invariant(obj));
3338 if (osd_object_is_root(obj)) {
3339 dt->do_index_ops = &osd_index_ea_ops;
3341 } else if (feat == &dt_directory_features) {
3342 dt->do_index_ops = &osd_index_ea_ops;
3343 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3348 } else if (unlikely(feat == &dt_otable_features)) {
3349 dt->do_index_ops = &osd_otable_ops;
3351 } else if (unlikely(feat == &dt_acct_features)) {
3352 dt->do_index_ops = &osd_acct_index_ops;
3355 } else if (!osd_has_index(obj)) {
3356 struct osd_directory *dir;
3361 spin_lock(&obj->oo_guard);
3362 if (obj->oo_dir == NULL)
3366 * Concurrent thread allocated container data.
3369 spin_unlock(&obj->oo_guard);
3371 * Now, that we have container data, serialize its
3374 down_write(&obj->oo_ext_idx_sem);
3376 * recheck under lock.
3378 if (!osd_has_index(obj))
3379 result = osd_iam_container_init(env, obj,
3383 up_write(&obj->oo_ext_idx_sem);
3391 if (result == 0 && skip_iam == 0) {
3392 if (!osd_iam_index_probe(env, obj, feat))
3395 LINVRNT(osd_invariant(obj));
3397 if (result == 0 && is_quota_glb_feat(feat) &&
3398 fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3399 result = osd_quota_migration(env, dt, feat);
3404 static int osd_otable_it_attr_get(const struct lu_env *env,
3405 struct dt_object *dt,
3406 struct lu_attr *attr,
3407 struct lustre_capa *capa)
3413 static const struct dt_object_operations osd_obj_ops = {
3414 .do_read_lock = osd_object_read_lock,
3415 .do_write_lock = osd_object_write_lock,
3416 .do_read_unlock = osd_object_read_unlock,
3417 .do_write_unlock = osd_object_write_unlock,
3418 .do_write_locked = osd_object_write_locked,
3419 .do_attr_get = osd_attr_get,
3420 .do_declare_attr_set = osd_declare_attr_set,
3421 .do_attr_set = osd_attr_set,
3422 .do_ah_init = osd_ah_init,
3423 .do_declare_create = osd_declare_object_create,
3424 .do_create = osd_object_create,
3425 .do_declare_destroy = osd_declare_object_destroy,
3426 .do_destroy = osd_object_destroy,
3427 .do_index_try = osd_index_try,
3428 .do_declare_ref_add = osd_declare_object_ref_add,
3429 .do_ref_add = osd_object_ref_add,
3430 .do_declare_ref_del = osd_declare_object_ref_del,
3431 .do_ref_del = osd_object_ref_del,
3432 .do_xattr_get = osd_xattr_get,
3433 .do_declare_xattr_set = osd_declare_xattr_set,
3434 .do_xattr_set = osd_xattr_set,
3435 .do_declare_xattr_del = osd_declare_xattr_del,
3436 .do_xattr_del = osd_xattr_del,
3437 .do_xattr_list = osd_xattr_list,
3438 .do_capa_get = osd_capa_get,
3439 .do_object_sync = osd_object_sync,
3440 .do_data_get = osd_data_get,
3444 * dt_object_operations for interoperability mode
3445 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3447 static const struct dt_object_operations osd_obj_ea_ops = {
3448 .do_read_lock = osd_object_read_lock,
3449 .do_write_lock = osd_object_write_lock,
3450 .do_read_unlock = osd_object_read_unlock,
3451 .do_write_unlock = osd_object_write_unlock,
3452 .do_write_locked = osd_object_write_locked,
3453 .do_attr_get = osd_attr_get,
3454 .do_declare_attr_set = osd_declare_attr_set,
3455 .do_attr_set = osd_attr_set,
3456 .do_ah_init = osd_ah_init,
3457 .do_declare_create = osd_declare_object_create,
3458 .do_create = osd_object_ea_create,
3459 .do_declare_destroy = osd_declare_object_destroy,
3460 .do_destroy = osd_object_destroy,
3461 .do_index_try = osd_index_try,
3462 .do_declare_ref_add = osd_declare_object_ref_add,
3463 .do_ref_add = osd_object_ref_add,
3464 .do_declare_ref_del = osd_declare_object_ref_del,
3465 .do_ref_del = osd_object_ref_del,
3466 .do_xattr_get = osd_xattr_get,
3467 .do_declare_xattr_set = osd_declare_xattr_set,
3468 .do_xattr_set = osd_xattr_set,
3469 .do_declare_xattr_del = osd_declare_xattr_del,
3470 .do_xattr_del = osd_xattr_del,
3471 .do_xattr_list = osd_xattr_list,
3472 .do_capa_get = osd_capa_get,
3473 .do_object_sync = osd_object_sync,
3474 .do_data_get = osd_data_get,
3477 static const struct dt_object_operations osd_obj_otable_it_ops = {
3478 .do_attr_get = osd_otable_it_attr_get,
3479 .do_index_try = osd_index_try,
3482 static int osd_index_declare_iam_delete(const struct lu_env *env,
3483 struct dt_object *dt,
3484 const struct dt_key *key,
3485 struct thandle *handle)
3487 struct osd_thandle *oh;
3489 oh = container_of0(handle, struct osd_thandle, ot_super);
3490 LASSERT(oh->ot_handle == NULL);
3492 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3493 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3499 * delete a (key, value) pair from index \a dt specified by \a key
3501 * \param dt osd index object
3502 * \param key key for index
3503 * \param rec record reference
3504 * \param handle transaction handler
3507 * \retval -ve failure
3510 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3511 const struct dt_key *key,
3512 struct thandle *handle,
3513 struct lustre_capa *capa)
3515 struct osd_thread_info *oti = osd_oti_get(env);
3516 struct osd_object *obj = osd_dt_obj(dt);
3517 struct osd_thandle *oh;
3518 struct iam_path_descr *ipd;
3519 struct iam_container *bag = &obj->oo_dir->od_container;
3524 LINVRNT(osd_invariant(obj));
3525 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3526 LASSERT(bag->ic_object == obj->oo_inode);
3527 LASSERT(handle != NULL);
3529 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3532 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3534 ipd = osd_idx_ipd_get(env, bag);
3535 if (unlikely(ipd == NULL))
3538 oh = container_of0(handle, struct osd_thandle, ot_super);
3539 LASSERT(oh->ot_handle != NULL);
3540 LASSERT(oh->ot_handle->h_transaction != NULL);
3542 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3543 /* swab quota uid/gid provided by caller */
3544 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3545 key = (const struct dt_key *)&oti->oti_quota_id;
3548 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3549 osd_ipd_put(env, bag, ipd);
3550 LINVRNT(osd_invariant(obj));
3554 static int osd_index_declare_ea_delete(const struct lu_env *env,
3555 struct dt_object *dt,
3556 const struct dt_key *key,
3557 struct thandle *handle)
3559 struct osd_thandle *oh;
3560 struct inode *inode;
3564 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3565 LASSERT(handle != NULL);
3567 oh = container_of0(handle, struct osd_thandle, ot_super);
3568 LASSERT(oh->ot_handle == NULL);
3570 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3571 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3573 inode = osd_dt_obj(dt)->oo_inode;
3576 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3577 0, oh, osd_dt_obj(dt), true, NULL, false);
3581 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3584 struct osd_fid_pack *rec;
3587 if (de->file_type & LDISKFS_DIRENT_LUFID) {
3588 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3589 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3594 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3595 const struct lu_fid *fid)
3597 struct seq_server_site *ss = osd_seq_site(osd);
3600 /* FID seqs not in FLDB, must be local seq */
3601 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3604 /* If FLD is not being initialized yet, it only happens during the
3605 * initialization, likely during mgs initialization, and we assume
3606 * this is local FID. */
3607 if (ss == NULL || ss->ss_server_fld == NULL)
3610 /* Only check the local FLDB here */
3611 if (osd_seq_exists(env, osd, fid_seq(fid)))
3618 * Index delete function for interoperability mode (b11826).
3619 * It will remove the directory entry added by osd_index_ea_insert().
3620 * This entry is needed to maintain name->fid mapping.
3622 * \param key, key i.e. file entry to be deleted
3624 * \retval 0, on success
3625 * \retval -ve, on error
3627 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3628 const struct dt_key *key,
3629 struct thandle *handle,
3630 struct lustre_capa *capa)
3632 struct osd_object *obj = osd_dt_obj(dt);
3633 struct inode *dir = obj->oo_inode;
3634 struct dentry *dentry;
3635 struct osd_thandle *oh;
3636 struct ldiskfs_dir_entry_2 *de = NULL;
3637 struct buffer_head *bh;
3638 struct htree_lock *hlock = NULL;
3639 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
3640 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3644 LINVRNT(osd_invariant(obj));
3645 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3646 LASSERT(handle != NULL);
3648 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3650 oh = container_of(handle, struct osd_thandle, ot_super);
3651 LASSERT(oh->ot_handle != NULL);
3652 LASSERT(oh->ot_handle->h_transaction != NULL);
3654 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3657 ll_vfs_dq_init(dir);
3658 dentry = osd_child_dentry_get(env, obj,
3659 (char *)key, strlen((char *)key));
3661 if (obj->oo_hl_head != NULL) {
3662 hlock = osd_oti_get(env)->oti_hlock;
3663 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3664 dir, LDISKFS_HLOCK_DEL);
3666 down_write(&obj->oo_ext_idx_sem);
3669 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3673 /* If this is not the ".." entry, it might be a remote DNE
3674 * entry and we need to check if the FID is for a remote
3675 * MDT. If the FID is not in the directory entry (e.g.
3676 * upgraded 1.8 filesystem without dirdata enabled) then
3677 * we need to get the FID from the LMA. For a remote directory
3678 * there HAS to be an LMA, it cannot be an IGIF inode in this
3681 * Delete the entry before the agent inode in order to
3682 * simplify error handling. At worst an error after deleting
3683 * the entry first might leak the agent inode afterward. The
3684 * reverse would need filesystem abort in case of error deleting
3685 * the entry after the agent had been removed, or leave a
3686 * dangling entry pointing at a random inode. */
3687 if (strcmp((char *)key, dotdot) != 0) {
3688 LASSERT(de != NULL);
3689 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3690 /* If Fid is not in dentry, try to get it from LMA */
3691 if (rc == -ENODATA) {
3692 struct osd_inode_id *id;
3693 struct inode *inode;
3695 /* Before trying to get fid from the inode,
3696 * check whether the inode is valid.
3698 * If the inode has been deleted, do not go
3699 * ahead to do osd_ea_fid_get, which will set
3700 * the inode to bad inode, which might cause
3701 * the inode to be deleted uncorrectly */
3702 inode = ldiskfs_iget(osd_sb(osd),
3703 le32_to_cpu(de->inode));
3704 if (IS_ERR(inode)) {
3705 CDEBUG(D_INODE, "%s: "DFID"get inode"
3706 "error.\n", osd_name(osd),
3708 rc = PTR_ERR(inode);
3710 if (likely(inode->i_nlink != 0)) {
3711 id = &osd_oti_get(env)->oti_id;
3712 rc = osd_ea_fid_get(env, obj,
3713 le32_to_cpu(de->inode),
3716 CDEBUG(D_INFO, "%s: %u "DFID
3719 le32_to_cpu(de->inode),
3727 unlikely(osd_remote_fid(env, osd, fid)))
3728 /* Need to delete agent inode */
3729 ino = le32_to_cpu(de->inode);
3731 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3733 if (rc == 0 && unlikely(ino != 0)) {
3734 rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3737 CERROR("%s: del local inode "DFID": rc = %d\n",
3738 osd_name(osd), PFID(fid), rc);
3744 ldiskfs_htree_unlock(hlock);
3746 up_write(&obj->oo_ext_idx_sem);
3751 /* For inode on the remote MDT, .. will point to
3752 * /Agent directory, Check whether it needs to delete
3753 * from agent directory */
3754 if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3755 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3757 if (rc != 0 && rc != -ENOENT) {
3758 CERROR("%s: delete agent inode "DFID": rc = %d\n",
3759 osd_name(osd), PFID(fid), rc);
3769 LASSERT(osd_invariant(obj));
3774 * Lookup index for \a key and copy record to \a rec.
3776 * \param dt osd index object
3777 * \param key key for index
3778 * \param rec record reference
3780 * \retval +ve success : exact mach
3781 * \retval 0 return record with key not greater than \a key
3782 * \retval -ve failure
3784 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3785 struct dt_rec *rec, const struct dt_key *key,
3786 struct lustre_capa *capa)
3788 struct osd_object *obj = osd_dt_obj(dt);
3789 struct iam_path_descr *ipd;
3790 struct iam_container *bag = &obj->oo_dir->od_container;
3791 struct osd_thread_info *oti = osd_oti_get(env);
3792 struct iam_iterator *it = &oti->oti_idx_it;
3793 struct iam_rec *iam_rec;
3798 LASSERT(osd_invariant(obj));
3799 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3800 LASSERT(bag->ic_object == obj->oo_inode);
3802 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3805 ipd = osd_idx_ipd_get(env, bag);
3809 /* got ipd now we can start iterator. */
3810 iam_it_init(it, bag, 0, ipd);
3812 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3813 /* swab quota uid/gid provided by caller */
3814 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3815 key = (const struct dt_key *)&oti->oti_quota_id;
3818 rc = iam_it_get(it, (struct iam_key *)key);
3820 if (S_ISDIR(obj->oo_inode->i_mode))
3821 iam_rec = (struct iam_rec *)oti->oti_ldp;
3823 iam_rec = (struct iam_rec *) rec;
3825 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3827 if (S_ISDIR(obj->oo_inode->i_mode))
3828 osd_fid_unpack((struct lu_fid *) rec,
3829 (struct osd_fid_pack *)iam_rec);
3830 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3831 osd_quota_unpack(obj, rec);
3836 osd_ipd_put(env, bag, ipd);
3838 LINVRNT(osd_invariant(obj));
3843 static int osd_index_declare_iam_insert(const struct lu_env *env,
3844 struct dt_object *dt,
3845 const struct dt_rec *rec,
3846 const struct dt_key *key,
3847 struct thandle *handle)
3849 struct osd_thandle *oh;
3851 LASSERT(handle != NULL);
3853 oh = container_of0(handle, struct osd_thandle, ot_super);
3854 LASSERT(oh->ot_handle == NULL);
3856 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3857 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3863 * Inserts (key, value) pair in \a dt index object.
3865 * \param dt osd index object
3866 * \param key key for index
3867 * \param rec record reference
3868 * \param th transaction handler
3871 * \retval -ve failure
3873 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3874 const struct dt_rec *rec,
3875 const struct dt_key *key, struct thandle *th,
3876 struct lustre_capa *capa, int ignore_quota)
3878 struct osd_object *obj = osd_dt_obj(dt);
3879 struct iam_path_descr *ipd;
3880 struct osd_thandle *oh;
3881 struct iam_container *bag = &obj->oo_dir->od_container;
3882 struct osd_thread_info *oti = osd_oti_get(env);
3883 struct iam_rec *iam_rec;
3888 LINVRNT(osd_invariant(obj));
3889 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3890 LASSERT(bag->ic_object == obj->oo_inode);
3891 LASSERT(th != NULL);
3893 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3896 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3898 ipd = osd_idx_ipd_get(env, bag);
3899 if (unlikely(ipd == NULL))
3902 oh = container_of0(th, struct osd_thandle, ot_super);
3903 LASSERT(oh->ot_handle != NULL);
3904 LASSERT(oh->ot_handle->h_transaction != NULL);
3905 if (S_ISDIR(obj->oo_inode->i_mode)) {
3906 iam_rec = (struct iam_rec *)oti->oti_ldp;
3907 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3908 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3909 /* pack quota uid/gid */
3910 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3911 key = (const struct dt_key *)&oti->oti_quota_id;
3912 /* pack quota record */
3913 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3914 iam_rec = (struct iam_rec *)rec;
3916 iam_rec = (struct iam_rec *)rec;
3919 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3921 osd_ipd_put(env, bag, ipd);
3922 LINVRNT(osd_invariant(obj));
3927 * Calls ldiskfs_add_entry() to add directory entry
3928 * into the directory. This is required for
3929 * interoperability mode (b11826)
3931 * \retval 0, on success
3932 * \retval -ve, on error
3934 static int __osd_ea_add_rec(struct osd_thread_info *info,
3935 struct osd_object *pobj, struct inode *cinode,
3936 const char *name, const struct lu_fid *fid,
3937 struct htree_lock *hlock, struct thandle *th)
3939 struct ldiskfs_dentry_param *ldp;
3940 struct dentry *child;
3941 struct osd_thandle *oth;
3944 oth = container_of(th, struct osd_thandle, ot_super);
3945 LASSERT(oth->ot_handle != NULL);
3946 LASSERT(oth->ot_handle->h_transaction != NULL);
3947 LASSERT(pobj->oo_inode);
3949 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3950 if (unlikely(pobj->oo_inode ==
3951 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3954 osd_get_ldiskfs_dirent_param(ldp, fid);
3955 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3956 child->d_fsdata = (void *)ldp;
3957 ll_vfs_dq_init(pobj->oo_inode);
3958 rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3959 if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
3960 struct ldiskfs_dir_entry_2 *de;
3961 struct buffer_head *bh;
3964 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
3967 rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
3970 if (S_ISDIR(cinode->i_mode))
3971 de->file_type = LDISKFS_DIRENT_LUFID |
3972 LDISKFS_FT_REG_FILE;
3974 de->file_type = LDISKFS_DIRENT_LUFID |
3976 ldiskfs_journal_dirty_metadata(oth->ot_handle,
3987 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3988 * into the directory.Also sets flags into osd object to
3989 * indicate dot and dotdot are created. This is required for
3990 * interoperability mode (b11826)
3992 * \param dir directory for dot and dotdot fixup.
3993 * \param obj child object for linking
3995 * \retval 0, on success
3996 * \retval -ve, on error
3998 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3999 struct osd_object *dir,
4000 struct inode *parent_dir, const char *name,
4001 const struct lu_fid *dot_fid,
4002 const struct lu_fid *dot_dot_fid,
4005 struct inode *inode = dir->oo_inode;
4006 struct osd_thandle *oth;
4009 oth = container_of(th, struct osd_thandle, ot_super);
4010 LASSERT(oth->ot_handle->h_transaction != NULL);
4011 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4013 if (strcmp(name, dot) == 0) {
4014 if (dir->oo_compat_dot_created) {
4017 LASSERT(inode == parent_dir);
4018 dir->oo_compat_dot_created = 1;
4021 } else if (strcmp(name, dotdot) == 0) {
4022 if (!dir->oo_compat_dot_created)
4024 /* in case of rename, dotdot is already created */
4025 if (dir->oo_compat_dotdot_created) {
4026 return __osd_ea_add_rec(info, dir, parent_dir, name,
4027 dot_dot_fid, NULL, th);
4030 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT) ||
4031 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT2)) {
4032 struct lu_fid tfid = *dot_dot_fid;
4035 result = osd_add_dot_dotdot_internal(info,
4036 dir->oo_inode, parent_dir, dot_fid,
4039 result = osd_add_dot_dotdot_internal(info,
4040 dir->oo_inode, parent_dir, dot_fid,
4045 dir->oo_compat_dotdot_created = 1;
4053 * It will call the appropriate osd_add* function and return the
4054 * value, return by respective functions.
4056 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4057 struct inode *cinode, const char *name,
4058 const struct lu_fid *fid, struct thandle *th)
4060 struct osd_thread_info *info = osd_oti_get(env);
4061 struct htree_lock *hlock;
4064 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4066 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4068 if (hlock != NULL) {
4069 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4072 down_write(&pobj->oo_ext_idx_sem);
4075 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4076 lu_object_fid(&pobj->oo_dt.do_lu),
4079 if (hlock != NULL) {
4080 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4081 pobj->oo_inode, LDISKFS_HLOCK_ADD);
4083 down_write(&pobj->oo_ext_idx_sem);
4086 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4087 struct lu_fid *tfid = &info->oti_fid;
4091 rc = __osd_ea_add_rec(info, pobj, cinode, name,
4094 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4099 ldiskfs_htree_unlock(hlock);
4101 up_write(&pobj->oo_ext_idx_sem);
4107 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4108 struct osd_idmap_cache *oic)
4110 struct osd_scrub *scrub = &dev->od_scrub;
4111 struct lu_fid *fid = &oic->oic_fid;
4112 struct osd_inode_id *id = &oti->oti_id;
4117 if (!fid_is_norm(fid) && !fid_is_igif(fid))
4121 rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
4122 if (rc != 0 && rc != -ENOENT)
4125 if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
4128 if (thread_is_running(&scrub->os_thread)) {
4129 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
4130 /* There is race condition between osd_oi_lookup and OI scrub.
4131 * The OI scrub finished just after osd_oi_lookup() failure.
4132 * Under such case, it is unnecessary to trigger OI scrub again,
4133 * but try to call osd_oi_lookup() again. */
4134 if (unlikely(rc == -EAGAIN))
4140 if (!dev->od_noscrub && ++once == 1) {
4141 rc = osd_scrub_start(dev);
4142 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC for "DFID
4144 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4153 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4154 struct osd_device *dev,
4155 struct osd_idmap_cache *oic,
4156 struct lu_fid *fid, __u32 ino)
4158 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
4159 struct inode *inode;
4162 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4163 inode = osd_iget(oti, dev, &oic->oic_lid);
4164 if (IS_ERR(inode)) {
4165 fid_zero(&oic->oic_fid);
4166 return PTR_ERR(inode);
4169 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4172 fid_zero(&oic->oic_fid);
4174 *fid = oic->oic_fid = lma->lma_self_fid;
4178 int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4179 struct osd_inode_id *id, const struct lu_fid *fid)
4181 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4182 id->oii_ino, id->oii_gen, info);
4183 info->oti_cache.oic_lid = *id;
4184 info->oti_cache.oic_fid = *fid;
4185 info->oti_cache.oic_dev = osd;
4191 * Get parent FID from the linkEA.
4193 * For a directory which parent resides on remote MDT, to satisfy the
4194 * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
4195 * the other hand, to make the lookup(..) on the directory can return
4196 * the real parent FID, we append the real parent FID after its ".."
4197 * name entry in the /REMOTE_PARENT_DIR.
4199 * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
4200 * backup. So after the restore, we cannot get the right parent FID from
4201 * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
4202 * we have stored the real parent FID in the directory object's linkEA,
4203 * we can parse the linkEA for the real parent FID.
4205 * \param[in] env pointer to the thread context
4206 * \param[in] obj pointer to the object to be handled
4207 * \param[out]fid pointer to the buffer to hold the parent FID
4209 * \retval 0 for getting the real parent FID successfully
4210 * \retval negative error number on failure
4212 static int osd_get_pfid_from_linkea(const struct lu_env *env,
4213 struct osd_object *obj,
4216 struct osd_thread_info *oti = osd_oti_get(env);
4217 struct lu_buf *buf = &oti->oti_big_buf;
4218 struct dentry *dentry = &oti->oti_obj_dentry;
4219 struct inode *inode = obj->oo_inode;
4220 struct linkea_data ldata = { 0 };
4225 if (!S_ISDIR(inode->i_mode))
4229 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4230 buf->lb_buf, buf->lb_len);
4231 if (rc == -ERANGE) {
4232 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4235 lu_buf_realloc(buf, rc);
4236 if (buf->lb_buf == NULL)
4243 if (unlikely(rc == 0))
4249 if (unlikely(buf->lb_buf == NULL)) {
4250 lu_buf_realloc(buf, rc);
4251 if (buf->lb_buf == NULL)
4258 rc = linkea_init(&ldata);
4260 linkea_first_entry(&ldata);
4261 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
4268 * Calls ->lookup() to find dentry. From dentry get inode and
4269 * read inode's ea to get fid. This is required for interoperability
4272 * \retval 0, on success
4273 * \retval -ve, on error
4275 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4276 struct dt_rec *rec, const struct dt_key *key)
4278 struct inode *dir = obj->oo_inode;
4279 struct dentry *dentry;
4280 struct ldiskfs_dir_entry_2 *de;
4281 struct buffer_head *bh;
4282 struct lu_fid *fid = (struct lu_fid *) rec;
4283 struct htree_lock *hlock = NULL;
4288 LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
4290 dentry = osd_child_dentry_get(env, obj,
4291 (char *)key, strlen((char *)key));
4293 if (obj->oo_hl_head != NULL) {
4294 hlock = osd_oti_get(env)->oti_hlock;
4295 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4296 dir, LDISKFS_HLOCK_LOOKUP);
4298 down_read(&obj->oo_ext_idx_sem);
4301 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4303 struct osd_thread_info *oti = osd_oti_get(env);
4304 struct osd_inode_id *id = &oti->oti_id;
4305 struct osd_idmap_cache *oic = &oti->oti_cache;
4306 struct osd_device *dev = osd_obj2dev(obj);
4307 struct osd_scrub *scrub = &dev->od_scrub;
4308 struct scrub_file *sf = &scrub->os_file;
4310 ino = le32_to_cpu(de->inode);
4311 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4313 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4317 rc = osd_get_fid_from_dentry(de, rec);
4319 /* done with de, release bh */
4322 if (unlikely(ino == osd_remote_parent_ino(dev)))
4323 /* If the parent is on remote MDT, and there
4324 * is no FID-in-dirent, then we have to get
4325 * the parent FID from the linkEA. */
4326 rc = osd_get_pfid_from_linkea(env, obj, fid);
4328 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4330 osd_id_gen(id, ino, OSD_OII_NOGEN);
4334 fid_zero(&oic->oic_fid);
4338 if (osd_remote_fid(env, dev, fid))
4341 rc = osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id,
4345 if ((scrub->os_pos_current <= ino) &&
4346 ((sf->sf_flags & SF_INCONSISTENT) ||
4347 (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
4348 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
4350 osd_consistency_check(oti, dev, oic);
4359 ldiskfs_htree_unlock(hlock);
4361 up_read(&obj->oo_ext_idx_sem);
4366 * Find the osd object for given fid.
4368 * \param fid need to find the osd object having this fid
4370 * \retval osd_object on success
4371 * \retval -ve on error
4373 struct osd_object *osd_object_find(const struct lu_env *env,
4374 struct dt_object *dt,
4375 const struct lu_fid *fid)
4377 struct lu_device *ludev = dt->do_lu.lo_dev;
4378 struct osd_object *child = NULL;
4379 struct lu_object *luch;
4380 struct lu_object *lo;
4383 * at this point topdev might not exist yet
4384 * (i.e. MGS is preparing profiles). so we can
4385 * not rely on topdev and instead lookup with
4386 * our device passed as topdev. this can't work
4387 * if the object isn't cached yet (as osd doesn't
4388 * allocate lu_header). IOW, the object must be
4389 * in the cache, otherwise lu_object_alloc() crashes
4392 luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4393 ludev : ludev->ld_site->ls_top_dev,
4395 if (!IS_ERR(luch)) {
4396 if (lu_object_exists(luch)) {
4397 lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4399 child = osd_obj(lo);
4401 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4402 "lu_object can't be located"
4403 DFID"\n", PFID(fid));
4405 if (child == NULL) {
4406 lu_object_put(env, luch);
4407 CERROR("Unable to get osd_object\n");
4408 child = ERR_PTR(-ENOENT);
4411 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4412 "lu_object does not exists "DFID"\n",
4414 lu_object_put(env, luch);
4415 child = ERR_PTR(-ENOENT);
4418 child = ERR_CAST(luch);
4425 * Put the osd object once done with it.
4427 * \param obj osd object that needs to be put
4429 static inline void osd_object_put(const struct lu_env *env,
4430 struct osd_object *obj)
4432 lu_object_put(env, &obj->oo_dt.do_lu);
4435 static int osd_index_declare_ea_insert(const struct lu_env *env,
4436 struct dt_object *dt,
4437 const struct dt_rec *rec,
4438 const struct dt_key *key,
4439 struct thandle *handle)
4441 struct osd_thandle *oh;
4442 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4443 struct lu_fid *fid = (struct lu_fid *)rec;
4447 LASSERT(!dt_object_remote(dt));
4448 LASSERT(handle != NULL);
4450 oh = container_of0(handle, struct osd_thandle, ot_super);
4451 LASSERT(oh->ot_handle == NULL);
4453 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4454 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4456 if (osd_dt_obj(dt)->oo_inode != NULL) {
4457 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4459 /* We ignore block quota on meta pool (MDTs), so needn't
4460 * calculate how many blocks will be consumed by this index
4462 rc = osd_declare_inode_qid(env, i_uid_read(inode),
4463 i_gid_read(inode), 0, oh,
4464 osd_dt_obj(dt), true, NULL, false);
4470 rc = osd_remote_fid(env, osd, fid);
4476 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
4477 osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
4478 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4479 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4480 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4481 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4487 * Index add function for interoperability mode (b11826).
4488 * It will add the directory entry.This entry is needed to
4489 * maintain name->fid mapping.
4491 * \param key it is key i.e. file entry to be inserted
4492 * \param rec it is value of given key i.e. fid
4494 * \retval 0, on success
4495 * \retval -ve, on error
4497 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4498 const struct dt_rec *rec,
4499 const struct dt_key *key, struct thandle *th,
4500 struct lustre_capa *capa, int ignore_quota)
4502 struct osd_object *obj = osd_dt_obj(dt);
4503 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4504 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
4505 const struct lu_fid *fid = rec1->rec_fid;
4506 const char *name = (const char *)key;
4507 struct osd_thread_info *oti = osd_oti_get(env);
4508 struct osd_inode_id *id = &oti->oti_id;
4509 struct inode *child_inode = NULL;
4510 struct osd_object *child = NULL;
4514 LASSERT(osd_invariant(obj));
4515 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
4516 LASSERT(th != NULL);
4518 osd_trans_exec_op(env, th, OSD_OT_INSERT);
4520 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
4523 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
4525 rc = osd_remote_fid(env, osd, fid);
4527 CERROR("%s: Can not find object "DFID" rc %d\n",
4528 osd_name(osd), PFID(fid), rc);
4533 /* Insert remote entry */
4534 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4535 struct osd_mdobj_map *omm = osd->od_mdt_map;
4536 struct osd_thandle *oh;
4538 /* If parent on remote MDT, we need put this object
4540 oh = container_of(th, typeof(*oh), ot_super);
4541 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4543 CERROR("%s: add "DFID" error: rc = %d\n",
4545 PFID(lu_object_fid(&dt->do_lu)), rc);
4549 child_inode = igrab(omm->omm_remote_parent->d_inode);
4551 child_inode = osd_create_local_agent_inode(env, osd,
4552 obj, fid, rec1->rec_type & S_IFMT, th);
4553 if (IS_ERR(child_inode))
4554 RETURN(PTR_ERR(child_inode));
4557 /* Insert local entry */
4558 child = osd_object_find(env, dt, fid);
4559 if (IS_ERR(child)) {
4560 CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4561 osd_name(osd), PFID(fid),
4562 id->oii_ino, id->oii_gen,
4563 (int)PTR_ERR(child));
4564 RETURN(PTR_ERR(child));
4566 child_inode = igrab(child->oo_inode);
4569 rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
4571 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4572 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4576 osd_object_put(env, child);
4577 LASSERT(osd_invariant(obj));
4582 * Initialize osd Iterator for given osd index object.
4584 * \param dt osd index object
4587 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4588 struct dt_object *dt,
4590 struct lustre_capa *capa)
4592 struct osd_it_iam *it;
4593 struct osd_thread_info *oti = osd_oti_get(env);
4594 struct osd_object *obj = osd_dt_obj(dt);
4595 struct lu_object *lo = &dt->do_lu;
4596 struct iam_path_descr *ipd;
4597 struct iam_container *bag = &obj->oo_dir->od_container;
4599 LASSERT(lu_object_exists(lo));
4601 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
4602 return ERR_PTR(-EACCES);
4605 ipd = osd_it_ipd_get(env, bag);
4606 if (likely(ipd != NULL)) {
4610 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4611 return (struct dt_it *)it;
4613 return ERR_PTR(-ENOMEM);
4617 * free given Iterator.
4620 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4622 struct osd_it_iam *it = (struct osd_it_iam *)di;
4623 struct osd_object *obj = it->oi_obj;
4625 iam_it_fini(&it->oi_it);
4626 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4627 lu_object_put(env, &obj->oo_dt.do_lu);
4631 * Move Iterator to record specified by \a key
4633 * \param di osd iterator
4634 * \param key key for index
4636 * \retval +ve di points to record with least key not larger than key
4637 * \retval 0 di points to exact matched key
4638 * \retval -ve failure
4641 static int osd_it_iam_get(const struct lu_env *env,
4642 struct dt_it *di, const struct dt_key *key)
4644 struct osd_thread_info *oti = osd_oti_get(env);
4645 struct osd_it_iam *it = (struct osd_it_iam *)di;
4647 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4648 /* swab quota uid/gid */
4649 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4650 key = (struct dt_key *)&oti->oti_quota_id;
4653 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4659 * \param di osd iterator
4661 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4663 struct osd_it_iam *it = (struct osd_it_iam *)di;
4665 iam_it_put(&it->oi_it);
4669 * Move iterator by one record
4671 * \param di osd iterator
4673 * \retval +1 end of container reached
4675 * \retval -ve failure
4678 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4680 struct osd_it_iam *it = (struct osd_it_iam *)di;
4682 return iam_it_next(&it->oi_it);
4686 * Return pointer to the key under iterator.
4689 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4690 const struct dt_it *di)
4692 struct osd_thread_info *oti = osd_oti_get(env);
4693 struct osd_it_iam *it = (struct osd_it_iam *)di;
4694 struct osd_object *obj = it->oi_obj;
4697 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4699 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4700 /* swab quota uid/gid */
4701 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4702 key = (struct dt_key *)&oti->oti_quota_id;
4709 * Return size of key under iterator (in bytes)
4712 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4714 struct osd_it_iam *it = (struct osd_it_iam *)di;
4716 return iam_it_key_size(&it->oi_it);
4720 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4722 /* check if file type is required */
4723 if (ent->lde_attrs & LUDA_TYPE) {
4724 struct luda_type *lt;
4725 int align = sizeof(*lt) - 1;
4727 len = (len + align) & ~align;
4728 lt = (struct luda_type *)(ent->lde_name + len);
4729 lt->lt_type = cpu_to_le16(DTTOIF(type));
4732 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4736 * build lu direct from backend fs dirent.
4740 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4741 char *name, __u16 namelen, __u16 type, __u32 attr)
4743 ent->lde_attrs = attr | LUDA_FID;
4744 fid_cpu_to_le(&ent->lde_fid, fid);
4746 ent->lde_hash = cpu_to_le64(offset);
4747 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4749 strncpy(ent->lde_name, name, namelen);
4750 ent->lde_name[namelen] = '\0';
4751 ent->lde_namelen = cpu_to_le16(namelen);
4753 /* append lustre attributes */
4754 osd_it_append_attrs(ent, namelen, type);
4758 * Return pointer to the record under iterator.
4760 static int osd_it_iam_rec(const struct lu_env *env,
4761 const struct dt_it *di,
4762 struct dt_rec *dtrec, __u32 attr)
4764 struct osd_it_iam *it = (struct osd_it_iam *)di;
4765 struct osd_thread_info *info = osd_oti_get(env);
4768 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4769 const struct osd_fid_pack *rec;
4770 struct lu_fid *fid = &info->oti_fid;
4771 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
4777 name = (char *)iam_it_key_get(&it->oi_it);
4779 RETURN(PTR_ERR(name));
4781 namelen = iam_it_key_size(&it->oi_it);
4783 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4785 RETURN(PTR_ERR(rec));
4787 rc = osd_fid_unpack(fid, rec);
4791 hash = iam_it_store(&it->oi_it);
4793 /* IAM does not store object type in IAM index (dir) */
4794 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4796 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4797 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4798 (struct iam_rec *)dtrec);
4799 osd_quota_unpack(it->oi_obj, dtrec);
4801 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4802 (struct iam_rec *)dtrec);
4809 * Returns cookie for current Iterator position.
4811 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4813 struct osd_it_iam *it = (struct osd_it_iam *)di;
4815 return iam_it_store(&it->oi_it);
4819 * Restore iterator from cookie.
4821 * \param di osd iterator
4822 * \param hash Iterator location cookie
4824 * \retval +ve di points to record with least key not larger than key.
4825 * \retval 0 di points to exact matched key
4826 * \retval -ve failure
4829 static int osd_it_iam_load(const struct lu_env *env,
4830 const struct dt_it *di, __u64 hash)
4832 struct osd_it_iam *it = (struct osd_it_iam *)di;
4834 return iam_it_load(&it->oi_it, hash);
4837 static const struct dt_index_operations osd_index_iam_ops = {
4838 .dio_lookup = osd_index_iam_lookup,
4839 .dio_declare_insert = osd_index_declare_iam_insert,
4840 .dio_insert = osd_index_iam_insert,
4841 .dio_declare_delete = osd_index_declare_iam_delete,
4842 .dio_delete = osd_index_iam_delete,
4844 .init = osd_it_iam_init,
4845 .fini = osd_it_iam_fini,
4846 .get = osd_it_iam_get,
4847 .put = osd_it_iam_put,
4848 .next = osd_it_iam_next,
4849 .key = osd_it_iam_key,
4850 .key_size = osd_it_iam_key_size,
4851 .rec = osd_it_iam_rec,
4852 .store = osd_it_iam_store,
4853 .load = osd_it_iam_load
4859 * Creates or initializes iterator context.
4861 * \retval struct osd_it_ea, iterator structure on success
4864 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4865 struct dt_object *dt,
4867 struct lustre_capa *capa)
4869 struct osd_object *obj = osd_dt_obj(dt);
4870 struct osd_thread_info *info = osd_oti_get(env);
4871 struct osd_it_ea *it = &info->oti_it_ea;
4872 struct file *file = &it->oie_file;
4873 struct lu_object *lo = &dt->do_lu;
4874 struct dentry *obj_dentry = &info->oti_it_dentry;
4876 LASSERT(lu_object_exists(lo));
4878 obj_dentry->d_inode = obj->oo_inode;
4879 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4880 obj_dentry->d_name.hash = 0;
4882 it->oie_rd_dirent = 0;
4883 it->oie_it_dirent = 0;
4884 it->oie_dirent = NULL;
4885 it->oie_buf = info->oti_it_ea_buf;
4888 /* Reset the "file" totally to avoid to reuse any old value from
4889 * former readdir handling, the "file->f_pos" should be zero. */
4890 memset(file, 0, sizeof(*file));
4891 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4892 if (attr & LUDA_64BITHASH)
4893 file->f_mode = FMODE_64BITHASH;
4895 file->f_mode = FMODE_32BITHASH;
4896 file->f_dentry = obj_dentry;
4897 file->f_mapping = obj->oo_inode->i_mapping;
4898 file->f_op = obj->oo_inode->i_fop;
4899 set_file_inode(file, obj->oo_inode);
4902 RETURN((struct dt_it *) it);
4906 * Destroy or finishes iterator context.
4908 * \param di iterator structure to be destroyed
4910 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4912 struct osd_it_ea *it = (struct osd_it_ea *)di;
4913 struct osd_object *obj = it->oie_obj;
4914 struct inode *inode = obj->oo_inode;
4917 it->oie_file.f_op->release(inode, &it->oie_file);
4918 lu_object_put(env, &obj->oo_dt.do_lu);
4923 * It position the iterator at given key, so that next lookup continues from
4924 * that key Or it is similar to dio_it->load() but based on a key,
4925 * rather than file position.
4927 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4930 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4932 static int osd_it_ea_get(const struct lu_env *env,
4933 struct dt_it *di, const struct dt_key *key)
4935 struct osd_it_ea *it = (struct osd_it_ea *)di;
4938 LASSERT(((const char *)key)[0] == '\0');
4939 it->oie_file.f_pos = 0;
4940 it->oie_rd_dirent = 0;
4941 it->oie_it_dirent = 0;
4942 it->oie_dirent = NULL;
4950 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4954 struct osd_filldir_cbs {
4955 #ifdef HAVE_DIR_CONTEXT
4956 struct dir_context ctx;
4958 struct osd_it_ea *it;
4961 * It is called internally by ->readdir(). It fills the
4962 * iterator's in-memory data structure with required
4963 * information i.e. name, namelen, rec_size etc.
4965 * \param buf in which information to be filled in.
4966 * \param name name of the file in given dir
4968 * \retval 0 on success
4969 * \retval 1 on buffer full
4971 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
4972 loff_t offset, __u64 ino,
4975 struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
4976 struct osd_object *obj = it->oie_obj;
4977 struct osd_it_ea_dirent *ent = it->oie_dirent;
4978 struct lu_fid *fid = &ent->oied_fid;
4979 struct osd_fid_pack *rec;
4982 /* this should never happen */
4983 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4984 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4988 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4992 /* "." is just the object itself. */
4993 if (namelen == 1 && name[0] == '.') {
4994 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4995 } else if (d_type & LDISKFS_DIRENT_LUFID) {
4996 rec = (struct osd_fid_pack*) (name + namelen + 1);
4997 if (osd_fid_unpack(fid, rec) != 0)
5002 d_type &= ~LDISKFS_DIRENT_LUFID;
5004 /* NOT export local root. */
5005 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5006 ino = obj->oo_inode->i_ino;
5007 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5010 ent->oied_ino = ino;
5011 ent->oied_off = offset;
5012 ent->oied_namelen = namelen;
5013 ent->oied_type = d_type;
5015 memcpy(ent->oied_name, name, namelen);
5017 it->oie_rd_dirent++;
5018 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
5023 * Calls ->readdir() to load a directory entry at a time
5024 * and stored it in iterator's in-memory data structure.
5026 * \param di iterator's in memory structure
5028 * \retval 0 on success
5029 * \retval -ve on error
5030 * \retval +1 reach the end of entry
5032 static int osd_ldiskfs_it_fill(const struct lu_env *env,
5033 const struct dt_it *di)
5035 struct osd_it_ea *it = (struct osd_it_ea *)di;
5036 struct osd_object *obj = it->oie_obj;
5037 struct inode *inode = obj->oo_inode;
5038 struct htree_lock *hlock = NULL;
5039 struct file *filp = &it->oie_file;
5041 struct osd_filldir_cbs buf = {
5042 #ifdef HAVE_DIR_CONTEXT
5043 .ctx.actor = osd_ldiskfs_filldir,
5049 it->oie_dirent = it->oie_buf;
5050 it->oie_rd_dirent = 0;
5052 if (obj->oo_hl_head != NULL) {
5053 hlock = osd_oti_get(env)->oti_hlock;
5054 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5055 inode, LDISKFS_HLOCK_READDIR);
5057 down_read(&obj->oo_ext_idx_sem);
5060 #ifdef HAVE_DIR_CONTEXT
5061 buf.ctx.pos = filp->f_pos;
5062 rc = inode->i_fop->iterate(filp, &buf.ctx);
5063 filp->f_pos = buf.ctx.pos;
5065 rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
5069 ldiskfs_htree_unlock(hlock);
5071 up_read(&obj->oo_ext_idx_sem);
5073 if (it->oie_rd_dirent == 0) {
5074 /*If it does not get any dirent, it means it has been reached
5075 *to the end of the dir */
5076 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
5080 it->oie_dirent = it->oie_buf;
5081 it->oie_it_dirent = 1;
5088 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5089 * to load a directory entry at a time and stored it in
5090 * iterator's in-memory data structure.
5092 * \param di iterator's in memory structure
5094 * \retval +ve iterator reached to end
5095 * \retval 0 iterator not reached to end
5096 * \retval -ve on error
5098 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
5100 struct osd_it_ea *it = (struct osd_it_ea *)di;
5105 if (it->oie_it_dirent < it->oie_rd_dirent) {
5107 (void *) it->oie_dirent +
5108 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
5109 it->oie_dirent->oied_namelen);
5110 it->oie_it_dirent++;
5113 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
5116 rc = osd_ldiskfs_it_fill(env, di);
5123 * Returns the key at current position from iterator's in memory structure.
5125 * \param di iterator's in memory structure
5127 * \retval key i.e. struct dt_key on success
5129 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
5130 const struct dt_it *di)
5132 struct osd_it_ea *it = (struct osd_it_ea *)di;
5134 return (struct dt_key *)it->oie_dirent->oied_name;
5138 * Returns the key's size at current position from iterator's in memory structure.
5140 * \param di iterator's in memory structure
5142 * \retval key_size i.e. struct dt_key on success
5144 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
5146 struct osd_it_ea *it = (struct osd_it_ea *)di;
5148 return it->oie_dirent->oied_namelen;
5152 osd_dirent_update(handle_t *jh, struct super_block *sb,
5153 struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5154 struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
5156 struct osd_fid_pack *rec;
5160 LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
5161 LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
5163 rc = ldiskfs_journal_get_write_access(jh, bh);
5167 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5168 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5169 rc = ldiskfs_journal_dirty_metadata(jh, bh);
5175 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
5177 if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
5178 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5185 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5187 LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5188 "dot_dotdot = %d\n", dot_dotdot);
5190 if (LDISKFS_DIR_REC_LEN(de) >=
5191 __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5198 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5199 struct inode *dir, struct inode *inode,
5200 struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5201 struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5202 struct htree_lock *hlock)
5204 struct dentry *dentry;
5205 struct osd_fid_pack *rec;
5206 struct ldiskfs_dentry_param *ldp;
5210 if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5211 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5214 /* There is enough space to hold the FID-in-dirent. */
5215 if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
5216 dir->i_sb->s_blocksize)) {
5217 rc = ldiskfs_journal_get_write_access(jh, bh);
5221 de->name[de->name_len] = 0;
5222 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5223 rec->fp_len = sizeof(struct lu_fid) + 1;
5224 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5225 de->file_type |= LDISKFS_DIRENT_LUFID;
5227 rc = ldiskfs_journal_dirty_metadata(jh, bh);
5232 rc = ldiskfs_delete_entry(jh, dir, de, bh);
5236 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5238 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5239 osd_get_ldiskfs_dirent_param(ldp, fid);
5240 dentry->d_fsdata = (void *)ldp;
5241 ll_vfs_dq_init(dir);
5242 rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5243 /* It is too bad, we cannot reinsert the name entry back.
5244 * That means we lose it! */
5246 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
5247 "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
5248 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5249 dir->i_ino, dir->i_generation,
5250 ent->oied_namelen, ent->oied_name, PFID(fid), rc);
5256 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5257 struct osd_it_ea *it, struct lu_fid *fid,
5258 struct osd_inode_id *id, __u32 *attr)
5260 struct osd_thread_info *info = osd_oti_get(env);
5261 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
5262 struct osd_device *dev = osd_obj2dev(obj);
5263 struct super_block *sb = osd_sb(dev);
5264 const char *devname =
5265 LDISKFS_SB(sb)->s_es->s_volume_name;
5266 struct osd_it_ea_dirent *ent = it->oie_dirent;
5267 struct inode *dir = obj->oo_inode;
5268 struct htree_lock *hlock = NULL;
5269 struct buffer_head *bh = NULL;
5270 handle_t *jh = NULL;
5271 struct ldiskfs_dir_entry_2 *de;
5272 struct dentry *dentry;
5273 struct inode *inode;
5280 if (ent->oied_name[0] == '.') {
5281 if (ent->oied_namelen == 1)
5283 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5287 dentry = osd_child_dentry_get(env, obj, ent->oied_name,
5290 /* We need to ensure that the name entry is still valid.
5291 * Because it may be removed or renamed by other already.
5293 * The unlink or rename operation will start journal before PDO lock,
5294 * so to avoid deadlock, here we need to start journal handle before
5295 * related PDO lock also. But because we do not know whether there
5296 * will be something to be repaired before PDO lock, we just start
5297 * journal without conditions.
5299 * We may need to remove the name entry firstly, then insert back.
5300 * One credit is for user quota file update.
5301 * One credit is for group quota file update.
5302 * Two credits are for dirty inode. */
5303 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5304 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5307 if (dev->od_dirent_journal) {
5308 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5311 CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent "
5312 "check_repair, dir = %lu/%u, credits = %d, "
5313 "name = %.*s: rc = %d\n",
5314 devname, dir->i_ino, dir->i_generation, credits,
5315 ent->oied_namelen, ent->oied_name, rc);
5319 if (obj->oo_hl_head != NULL) {
5320 hlock = osd_oti_get(env)->oti_hlock;
5321 /* "0" means exclusive lock for the whole directory.
5322 * We need to prevent others access such name entry
5323 * during the delete + insert. Neither HLOCK_ADD nor
5324 * HLOCK_DEL cannot guarantee the atomicity. */
5325 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5327 down_write(&obj->oo_ext_idx_sem);
5330 if (obj->oo_hl_head != NULL) {
5331 hlock = osd_oti_get(env)->oti_hlock;
5332 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5333 LDISKFS_HLOCK_LOOKUP);
5335 down_read(&obj->oo_ext_idx_sem);
5339 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5340 /* For dot/dotdot entry, if there is not enough space to hold the
5341 * FID-in-dirent, just keep them there. It only happens when the
5342 * device upgraded from 1.8 or restored from MDT file-level backup.
5343 * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5344 * and needs to get FID from LMA when readdir, it will not affect the
5345 * performance much. */
5346 if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
5347 (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5348 *attr |= LUDA_IGNORE;
5349 GOTO(out_journal, rc = 0);
5352 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5353 inode = osd_iget(info, dev, id);
5354 if (IS_ERR(inode)) {
5355 rc = PTR_ERR(inode);
5356 if (rc == -ENOENT || rc == -ESTALE)
5359 CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
5360 "check_repair, dir = %lu/%u, name = %.*s: "
5362 devname, dir->i_ino, dir->i_generation,
5363 ent->oied_namelen, ent->oied_name, rc);
5365 GOTO(out_journal, rc);
5368 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
5370 LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
5372 if (fid_is_sane(fid)) {
5373 /* FID-in-dirent is valid. */
5374 if (lu_fid_eq(fid, &lma->lma_self_fid))
5375 GOTO(out_inode, rc = 0);
5377 /* Do not repair under dryrun mode. */
5378 if (*attr & LUDA_VERIFY_DRYRUN) {
5379 *attr |= LUDA_REPAIR;
5380 GOTO(out_inode, rc = 0);
5383 if (!dev->od_dirent_journal) {
5387 ldiskfs_htree_unlock(hlock);
5389 up_read(&obj->oo_ext_idx_sem);
5390 dev->od_dirent_journal = 1;
5394 *fid = lma->lma_self_fid;
5396 /* Update the FID-in-dirent. */
5397 rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
5399 *attr |= LUDA_REPAIR;
5401 CDEBUG(D_LFSCK, "%.16s: fail to update FID "
5402 "in the dirent, dir = %lu/%u, "
5403 "name = %.*s, "DFID": rc = %d\n",
5404 devname, dir->i_ino, dir->i_generation,
5405 ent->oied_namelen, ent->oied_name,
5408 /* Do not repair under dryrun mode. */
5409 if (*attr & LUDA_VERIFY_DRYRUN) {
5410 *fid = lma->lma_self_fid;
5411 *attr |= LUDA_REPAIR;
5412 GOTO(out_inode, rc = 0);
5415 if (!dev->od_dirent_journal) {
5419 ldiskfs_htree_unlock(hlock);
5421 up_read(&obj->oo_ext_idx_sem);
5422 dev->od_dirent_journal = 1;
5426 *fid = lma->lma_self_fid;
5428 /* Append the FID-in-dirent. */
5429 rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5430 fid, bh, de, hlock);
5432 *attr |= LUDA_REPAIR;
5434 CDEBUG(D_LFSCK, "%.16s: fail to append FID "
5435 "after the dirent, dir = %lu/%u, "
5436 "name = %.*s, "DFID": rc = %d\n",
5437 devname, dir->i_ino, dir->i_generation,
5438 ent->oied_namelen, ent->oied_name,
5441 } else if (rc == -ENODATA) {
5442 /* Do not repair under dryrun mode. */
5443 if (*attr & LUDA_VERIFY_DRYRUN) {
5444 if (fid_is_sane(fid)) {
5445 *attr |= LUDA_REPAIR;
5447 lu_igif_build(fid, inode->i_ino,
5448 inode->i_generation);
5449 *attr |= LUDA_UPGRADE;
5451 GOTO(out_inode, rc = 0);
5454 if (!dev->od_dirent_journal) {
5458 ldiskfs_htree_unlock(hlock);
5460 up_read(&obj->oo_ext_idx_sem);
5461 dev->od_dirent_journal = 1;
5466 if (unlikely(fid_is_sane(fid))) {
5467 /* FID-in-dirent exists, but FID-in-LMA is lost.
5468 * Trust the FID-in-dirent, and add FID-in-LMA. */
5469 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5471 *attr |= LUDA_REPAIR;
5473 CDEBUG(D_LFSCK, "%.16s: fail to set LMA for "
5474 "update dirent, dir = %lu/%u, "
5475 "name = %.*s, "DFID": rc = %d\n",
5476 devname, dir->i_ino, dir->i_generation,
5477 ent->oied_namelen, ent->oied_name,
5480 lu_igif_build(fid, inode->i_ino, inode->i_generation);
5481 /* It is probably IGIF object. Only aappend the
5482 * FID-in-dirent. OI scrub will process FID-in-LMA. */
5483 rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5484 fid, bh, de, hlock);
5486 *attr |= LUDA_UPGRADE;
5488 CDEBUG(D_LFSCK, "%.16s: fail to append IGIF "
5489 "after the dirent, dir = %lu/%u, "
5490 "name = %.*s, "DFID": rc = %d\n",
5491 devname, dir->i_ino, dir->i_generation,
5492 ent->oied_namelen, ent->oied_name,
5497 GOTO(out_inode, rc);
5504 if (hlock != NULL) {
5505 ldiskfs_htree_unlock(hlock);
5507 if (dev->od_dirent_journal)
5508 up_write(&obj->oo_ext_idx_sem);
5510 up_read(&obj->oo_ext_idx_sem);
5513 ldiskfs_journal_stop(jh);
5514 if (rc >= 0 && !dirty)
5515 dev->od_dirent_journal = 0;
5520 * Returns the value at current position from iterator's in memory structure.
5522 * \param di struct osd_it_ea, iterator's in memory structure
5523 * \param attr attr requested for dirent.
5524 * \param lde lustre dirent
5526 * \retval 0 no error and \param lde has correct lustre dirent.
5527 * \retval -ve on error
5529 static inline int osd_it_ea_rec(const struct lu_env *env,
5530 const struct dt_it *di,
5531 struct dt_rec *dtrec, __u32 attr)
5533 struct osd_it_ea *it = (struct osd_it_ea *)di;
5534 struct osd_object *obj = it->oie_obj;
5535 struct osd_device *dev = osd_obj2dev(obj);
5536 struct osd_scrub *scrub = &dev->od_scrub;
5537 struct scrub_file *sf = &scrub->os_file;
5538 struct osd_thread_info *oti = osd_oti_get(env);
5539 struct osd_inode_id *id = &oti->oti_id;
5540 struct osd_idmap_cache *oic = &oti->oti_cache;
5541 struct lu_fid *fid = &it->oie_dirent->oied_fid;
5542 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
5543 __u32 ino = it->oie_dirent->oied_ino;
5547 LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
5549 if (attr & LUDA_VERIFY) {
5550 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5551 attr |= LUDA_IGNORE;
5552 /* If the parent is on remote MDT, and there
5553 * is no FID-in-dirent, then we have to get
5554 * the parent FID from the linkEA. */
5555 osd_get_pfid_from_linkea(env, obj, fid);
5557 rc = osd_dirent_check_repair(env, obj, it, fid, id,
5561 attr &= ~LU_DIRENT_ATTRS_MASK;
5562 if (!fid_is_sane(fid)) {
5563 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP) &&
5564 likely(it->oie_dirent->oied_namelen != 2 ||
5565 it->oie_dirent->oied_name[0] != '.' ||
5566 it->oie_dirent->oied_name[1] != '.'))
5569 if (unlikely(ino == osd_remote_parent_ino(dev)))
5570 /* If the parent is on remote MDT, and there
5571 * is no FID-in-dirent, then we have to get
5572 * the parent FID from the linkEA. */
5573 rc = osd_get_pfid_from_linkea(env, obj, fid);
5575 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5577 osd_id_gen(id, ino, OSD_OII_NOGEN);
5581 /* Pack the entry anyway, at least the offset is right. */
5582 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5583 it->oie_dirent->oied_name,
5584 it->oie_dirent->oied_namelen,
5585 it->oie_dirent->oied_type, attr);
5590 if (osd_remote_fid(env, dev, fid))
5593 if (likely(!(attr & LUDA_IGNORE) && rc == 0))
5594 rc = osd_add_oi_cache(oti, dev, id, fid);
5596 if (!(attr & LUDA_VERIFY) &&
5597 (scrub->os_pos_current <= ino) &&
5598 ((sf->sf_flags & SF_INCONSISTENT) ||
5599 (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
5600 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
5601 osd_consistency_check(oti, dev, oic);
5603 RETURN(rc > 0 ? 0 : rc);
5607 * Returns the record size size at current position.
5609 * This function will return record(lu_dirent) size in bytes.
5611 * \param[in] env execution environment
5612 * \param[in] di iterator's in memory structure
5613 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
5614 * calculate the lu_dirent size.
5616 * \retval record size(in bytes & in memory) of the current lu_dirent
5619 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5622 struct osd_it_ea *it = (struct osd_it_ea *)di;
5624 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5628 * Returns a cookie for current position of the iterator head, so that
5629 * user can use this cookie to load/start the iterator next time.
5631 * \param di iterator's in memory structure
5633 * \retval cookie for current position, on success
5635 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5637 struct osd_it_ea *it = (struct osd_it_ea *)di;
5639 return it->oie_dirent->oied_off;
5643 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5644 * to load a directory entry at a time and stored it i inn,
5645 * in iterator's in-memory data structure.
5647 * \param di struct osd_it_ea, iterator's in memory structure
5649 * \retval +ve on success
5650 * \retval -ve on error
5652 static int osd_it_ea_load(const struct lu_env *env,
5653 const struct dt_it *di, __u64 hash)
5655 struct osd_it_ea *it = (struct osd_it_ea *)di;
5659 it->oie_file.f_pos = hash;
5661 rc = osd_ldiskfs_it_fill(env, di);
5672 * Index lookup function for interoperability mode (b11826).
5674 * \param key, key i.e. file name to be searched
5676 * \retval +ve, on success
5677 * \retval -ve, on error
5679 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5680 struct dt_rec *rec, const struct dt_key *key,
5681 struct lustre_capa *capa)
5683 struct osd_object *obj = osd_dt_obj(dt);
5688 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5689 LINVRNT(osd_invariant(obj));
5691 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
5694 rc = osd_ea_lookup_rec(env, obj, rec, key);
5701 * Index and Iterator operations for interoperability
5702 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5704 static const struct dt_index_operations osd_index_ea_ops = {
5705 .dio_lookup = osd_index_ea_lookup,
5706 .dio_declare_insert = osd_index_declare_ea_insert,
5707 .dio_insert = osd_index_ea_insert,
5708 .dio_declare_delete = osd_index_declare_ea_delete,
5709 .dio_delete = osd_index_ea_delete,
5711 .init = osd_it_ea_init,
5712 .fini = osd_it_ea_fini,
5713 .get = osd_it_ea_get,
5714 .put = osd_it_ea_put,
5715 .next = osd_it_ea_next,
5716 .key = osd_it_ea_key,
5717 .key_size = osd_it_ea_key_size,
5718 .rec = osd_it_ea_rec,
5719 .rec_size = osd_it_ea_rec_size,
5720 .store = osd_it_ea_store,
5721 .load = osd_it_ea_load
5725 static void *osd_key_init(const struct lu_context *ctx,
5726 struct lu_context_key *key)
5728 struct osd_thread_info *info;
5730 OBD_ALLOC_PTR(info);
5732 return ERR_PTR(-ENOMEM);
5734 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5735 if (info->oti_it_ea_buf == NULL)
5738 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5740 info->oti_hlock = ldiskfs_htree_lock_alloc();
5741 if (info->oti_hlock == NULL)
5747 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5750 return ERR_PTR(-ENOMEM);
5753 static void osd_key_fini(const struct lu_context *ctx,
5754 struct lu_context_key *key, void* data)
5756 struct osd_thread_info *info = data;
5758 if (info->oti_hlock != NULL)
5759 ldiskfs_htree_lock_free(info->oti_hlock);
5760 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5761 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5762 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5763 lu_buf_free(&info->oti_big_buf);
5767 static void osd_key_exit(const struct lu_context *ctx,
5768 struct lu_context_key *key, void *data)
5770 struct osd_thread_info *info = data;
5772 LASSERT(info->oti_r_locks == 0);
5773 LASSERT(info->oti_w_locks == 0);
5774 LASSERT(info->oti_txns == 0);
5777 /* type constructor/destructor: osd_type_init, osd_type_fini */
5778 LU_TYPE_INIT_FINI(osd, &osd_key);
5780 struct lu_context_key osd_key = {
5781 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5782 .lct_init = osd_key_init,
5783 .lct_fini = osd_key_fini,
5784 .lct_exit = osd_key_exit
5788 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5789 const char *name, struct lu_device *next)
5791 struct osd_device *osd = osd_dev(d);
5793 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5794 >= sizeof(osd->od_svname))
5796 return osd_procfs_init(osd, name);
5799 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5801 struct seq_server_site *ss = osd_seq_site(osd);
5805 if (osd->od_is_ost || osd->od_cl_seq != NULL)
5808 if (unlikely(ss == NULL))
5811 OBD_ALLOC_PTR(osd->od_cl_seq);
5812 if (osd->od_cl_seq == NULL)
5815 rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5816 osd->od_svname, ss->ss_server_seq);
5819 OBD_FREE_PTR(osd->od_cl_seq);
5820 osd->od_cl_seq = NULL;
5826 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5828 if (osd->od_cl_seq == NULL)
5831 seq_client_fini(osd->od_cl_seq);
5832 OBD_FREE_PTR(osd->od_cl_seq);
5833 osd->od_cl_seq = NULL;
5836 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5840 /* shutdown quota slave instance associated with the device */
5841 if (o->od_quota_slave != NULL) {
5842 qsd_fini(env, o->od_quota_slave);
5843 o->od_quota_slave = NULL;
5846 osd_fid_fini(env, o);
5851 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5855 if (o->od_mnt != NULL) {
5856 shrink_dcache_sb(osd_sb(o));
5857 osd_sync(env, &o->od_dt_dev);
5866 static int osd_mount(const struct lu_env *env,
5867 struct osd_device *o, struct lustre_cfg *cfg)
5869 const char *name = lustre_cfg_string(cfg, 0);
5870 const char *dev = lustre_cfg_string(cfg, 1);
5872 unsigned long page, s_flags, lmd_flags = 0;
5873 struct page *__page;
5874 struct file_system_type *type;
5875 char *options = NULL;
5877 struct osd_thread_info *info = osd_oti_get(env);
5878 struct lu_fid *fid = &info->oti_fid;
5879 struct inode *inode;
5883 if (o->od_mnt != NULL)
5886 if (strlen(dev) >= sizeof(o->od_mntdev))
5888 strcpy(o->od_mntdev, dev);
5890 OBD_PAGE_ALLOC(__page, GFP_IOFS);
5892 GOTO(out, rc = -ENOMEM);
5894 str = lustre_cfg_string(cfg, 2);
5895 s_flags = simple_strtoul(str, NULL, 0);
5896 str = strstr(str, ":");
5898 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5899 opts = lustre_cfg_string(cfg, 3);
5900 page = (unsigned long)page_address(__page);
5901 options = (char *)page;
5904 strcat(options, "user_xattr,acl");
5906 strcat(options, opts);
5908 /* Glom up mount options */
5909 if (*options != '\0')
5910 strcat(options, ",");
5911 strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
5913 type = get_fs_type("ldiskfs");
5915 CERROR("%s: cannot find ldiskfs module\n", name);
5916 GOTO(out, rc = -ENODEV);
5919 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5920 module_put(type->owner);
5922 if (IS_ERR(o->od_mnt)) {
5923 rc = PTR_ERR(o->od_mnt);
5925 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5929 #ifdef HAVE_DEV_SET_RDONLY
5930 if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5931 CERROR("%s: underlying device %s is marked as read-only. "
5932 "Setup failed\n", name, dev);
5933 GOTO(out_mnt, rc = -EROFS);
5937 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5938 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5939 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5940 GOTO(out_mnt, rc = -EINVAL);
5943 #ifdef LDISKFS_MOUNT_DIRDATA
5944 if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
5945 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5946 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
5947 else if (!o->od_is_ost)
5948 CWARN("%s: device %s was upgraded from Lustre-1.x without "
5949 "enabling the dirdata feature. If you do not want to "
5950 "downgrade to Lustre-1.x again, you can enable it via "
5951 "'tune2fs -O dirdata device'\n", name, dev);
5953 inode = osd_sb(o)->s_root->d_inode;
5954 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
5955 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
5957 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
5961 if (lmd_flags & LMD_FLG_NOSCRUB)
5972 OBD_PAGE_FREE(__page);
5977 static struct lu_device *osd_device_fini(const struct lu_env *env,
5978 struct lu_device *d)
5980 struct osd_device *o = osd_dev(d);
5983 osd_shutdown(env, o);
5985 osd_scrub_cleanup(env, o);
5986 osd_obj_map_fini(o);
5992 static int osd_device_init0(const struct lu_env *env,
5993 struct osd_device *o,
5994 struct lustre_cfg *cfg)
5996 struct lu_device *l = osd2lu_dev(o);
5997 struct osd_thread_info *info;
6001 /* if the module was re-loaded, env can loose its keys */
6002 rc = lu_env_refill((struct lu_env *) env);
6005 info = osd_oti_get(env);
6008 l->ld_ops = &osd_lu_ops;
6009 o->od_dt_dev.dd_ops = &osd_dt_ops;
6011 spin_lock_init(&o->od_osfs_lock);
6012 mutex_init(&o->od_otable_mutex);
6014 o->od_capa_hash = init_capa_hash();
6015 if (o->od_capa_hash == NULL)
6016 GOTO(out, rc = -ENOMEM);
6018 o->od_read_cache = 1;
6019 o->od_writethrough_cache = 1;
6020 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
6022 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
6023 sizeof(o->od_svname));
6024 if (cplen >= sizeof(o->od_svname)) {
6029 if (server_name_is_ost(o->od_svname))
6032 rc = osd_mount(env, o, cfg);
6036 rc = osd_obj_map_init(env, o);
6040 rc = lu_site_init(&o->od_site, l);
6042 GOTO(out_compat, rc);
6043 o->od_site.ls_bottom_dev = l;
6045 rc = lu_site_init_finish(&o->od_site);
6049 /* self-repair LMA by default */
6050 o->od_lma_self_repair = 1;
6052 INIT_LIST_HEAD(&o->od_ios_list);
6053 /* setup scrub, including OI files initialization */
6054 rc = osd_scrub_setup(env, o);
6058 rc = osd_procfs_init(o, o->od_svname);
6060 CERROR("%s: can't initialize procfs: rc = %d\n",
6062 GOTO(out_scrub, rc);
6065 LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
6067 /* initialize quota slave instance */
6068 o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
6070 if (IS_ERR(o->od_quota_slave)) {
6071 rc = PTR_ERR(o->od_quota_slave);
6072 o->od_quota_slave = NULL;
6073 GOTO(out_procfs, rc);
6081 osd_scrub_cleanup(env, o);
6083 lu_site_fini(&o->od_site);
6085 osd_obj_map_fini(o);
6089 cleanup_capa_hash(o->od_capa_hash);
6094 static struct lu_device *osd_device_alloc(const struct lu_env *env,
6095 struct lu_device_type *t,
6096 struct lustre_cfg *cfg)
6098 struct osd_device *o;
6103 return ERR_PTR(-ENOMEM);
6105 rc = dt_device_init(&o->od_dt_dev, t);
6107 /* Because the ctx might be revived in dt_device_init,
6108 * refill the env here */
6109 lu_env_refill((struct lu_env *)env);
6110 rc = osd_device_init0(env, o, cfg);
6112 dt_device_fini(&o->od_dt_dev);
6115 if (unlikely(rc != 0))
6118 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
6121 static struct lu_device *osd_device_free(const struct lu_env *env,
6122 struct lu_device *d)
6124 struct osd_device *o = osd_dev(d);
6127 cleanup_capa_hash(o->od_capa_hash);
6128 /* XXX: make osd top device in order to release reference */
6129 d->ld_site->ls_top_dev = d;
6130 lu_site_purge(env, d->ld_site, -1);
6131 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
6132 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
6133 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
6135 lu_site_fini(&o->od_site);
6136 dt_device_fini(&o->od_dt_dev);
6141 static int osd_process_config(const struct lu_env *env,
6142 struct lu_device *d, struct lustre_cfg *cfg)
6144 struct osd_device *o = osd_dev(d);
6148 switch (cfg->lcfg_command) {
6150 rc = osd_mount(env, o, cfg);
6153 lu_dev_del_linkage(d->ld_site, d);
6154 rc = osd_shutdown(env, o);
6157 LASSERT(&o->od_dt_dev);
6158 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
6159 cfg, &o->od_dt_dev);
6160 if (rc > 0 || rc == -ENOSYS)
6161 rc = class_process_proc_param(PARAM_OST,
6162 lprocfs_osd_obd_vars,
6163 cfg, &o->od_dt_dev);
6172 static int osd_recovery_complete(const struct lu_env *env,
6173 struct lu_device *d)
6175 struct osd_device *osd = osd_dev(d);
6179 if (osd->od_quota_slave == NULL)
6182 /* start qsd instance on recovery completion, this notifies the quota
6183 * slave code that we are about to process new requests now */
6184 rc = qsd_start(env, osd->od_quota_slave);
6189 * we use exports to track all osd users
6191 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
6192 struct obd_device *obd, struct obd_uuid *cluuid,
6193 struct obd_connect_data *data, void *localdata)
6195 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6196 struct lustre_handle conn;
6200 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6202 rc = class_connect(&conn, obd, cluuid);
6206 *exp = class_conn2export(&conn);
6208 spin_lock(&osd->od_osfs_lock);
6210 spin_unlock(&osd->od_osfs_lock);
6216 * once last export (we don't count self-export) disappeared
6217 * osd can be released
6219 static int osd_obd_disconnect(struct obd_export *exp)
6221 struct obd_device *obd = exp->exp_obd;
6222 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6223 int rc, release = 0;
6226 /* Only disconnect the underlying layers on the final disconnect. */
6227 spin_lock(&osd->od_osfs_lock);
6229 if (osd->od_connects == 0)
6231 spin_unlock(&osd->od_osfs_lock);
6233 rc = class_disconnect(exp); /* bz 9811 */
6235 if (rc == 0 && release)
6236 class_manual_cleanup(obd);
6240 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6241 struct lu_device *dev)
6243 struct osd_device *osd = osd_dev(dev);
6247 if (osd->od_quota_slave != NULL) {
6248 /* set up quota slave objects */
6249 result = qsd_prepare(env, osd->od_quota_slave);
6254 result = osd_fid_init(env, osd);
6259 int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6260 struct lu_fid *fid, struct md_op_data *op_data)
6262 struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6264 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6267 static const struct lu_object_operations osd_lu_obj_ops = {
6268 .loo_object_init = osd_object_init,
6269 .loo_object_delete = osd_object_delete,
6270 .loo_object_release = osd_object_release,
6271 .loo_object_free = osd_object_free,
6272 .loo_object_print = osd_object_print,
6273 .loo_object_invariant = osd_object_invariant
6276 const struct lu_device_operations osd_lu_ops = {
6277 .ldo_object_alloc = osd_object_alloc,
6278 .ldo_process_config = osd_process_config,
6279 .ldo_recovery_complete = osd_recovery_complete,
6280 .ldo_prepare = osd_prepare,
6283 static const struct lu_device_type_operations osd_device_type_ops = {
6284 .ldto_init = osd_type_init,
6285 .ldto_fini = osd_type_fini,
6287 .ldto_start = osd_type_start,
6288 .ldto_stop = osd_type_stop,
6290 .ldto_device_alloc = osd_device_alloc,
6291 .ldto_device_free = osd_device_free,
6293 .ldto_device_init = osd_device_init,
6294 .ldto_device_fini = osd_device_fini
6297 struct lu_device_type osd_device_type = {
6298 .ldt_tags = LU_DEVICE_DT,
6299 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
6300 .ldt_ops = &osd_device_type_ops,
6301 .ldt_ctx_tags = LCT_LOCAL,
6305 * lprocfs legacy support.
6307 static struct obd_ops osd_obd_device_ops = {
6308 .o_owner = THIS_MODULE,
6309 .o_connect = osd_obd_connect,
6310 .o_disconnect = osd_obd_disconnect,
6311 .o_fid_alloc = osd_fid_alloc,
6314 static int __init osd_mod_init(void)
6320 rc = lu_kmem_init(ldiskfs_caches);
6324 rc = class_register_type(&osd_obd_device_ops, NULL, true,
6325 lprocfs_osd_module_vars,
6326 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6328 lu_kmem_fini(ldiskfs_caches);
6332 static void __exit osd_mod_exit(void)
6334 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6335 lu_kmem_fini(ldiskfs_caches);
6338 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6339 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6340 MODULE_LICENSE("GPL");
6342 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);