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, 2015, 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>
47 #include <linux/user_namespace.h>
48 #ifdef HAVE_UIDGID_HEADER
49 # include <linux/uidgid.h>
52 /* LUSTRE_VERSION_CODE */
53 #include <lustre_ver.h>
54 /* prerequisite for linux/xattr.h */
55 #include <linux/types.h>
56 /* prerequisite for linux/xattr.h */
58 /* XATTR_{REPLACE,CREATE} */
59 #include <linux/xattr.h>
61 #include <ldiskfs/ldiskfs.h>
62 #include <ldiskfs/xattr.h>
65 * struct OBD_{ALLOC,FREE}*()
68 #include <obd_support.h>
69 /* struct ptlrpc_thread */
70 #include <lustre_net.h>
71 #include <lustre_fid.h>
73 #include <lustre_param.h>
75 #include "osd_internal.h"
76 #include "osd_dynlocks.h"
78 /* llo_* api support */
79 #include <md_object.h>
80 #include <lustre_quota.h>
82 #include <lustre_linkea.h>
85 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
86 "ldiskfs with parallel directory operations");
88 int ldiskfs_track_declares_assert;
89 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
90 "LBUG during tracking of declares");
92 /* Slab to allocate dynlocks */
93 struct kmem_cache *dynlock_cachep;
95 /* Slab to allocate osd_it_ea */
96 struct kmem_cache *osd_itea_cachep;
98 static struct lu_kmem_descr ldiskfs_caches[] = {
100 .ckd_cache = &dynlock_cachep,
101 .ckd_name = "dynlock_cache",
102 .ckd_size = sizeof(struct dynlock_handle)
105 .ckd_cache = &osd_itea_cachep,
106 .ckd_name = "osd_itea_cache",
107 .ckd_size = sizeof(struct osd_it_ea)
114 static const char dot[] = ".";
115 static const char dotdot[] = "..";
116 static const char remote_obj_dir[] = "REM_OBJ_DIR";
118 static const struct lu_object_operations osd_lu_obj_ops;
119 static const struct dt_object_operations osd_obj_ops;
120 static const struct dt_object_operations osd_obj_ea_ops;
121 static const struct dt_object_operations osd_obj_otable_it_ops;
122 static const struct dt_index_operations osd_index_iam_ops;
123 static const struct dt_index_operations osd_index_ea_ops;
125 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
126 const struct lu_fid *fid);
127 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
128 struct osd_device *osd);
130 int osd_trans_declare_op2rb[] = {
131 [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
132 [OSD_OT_PUNCH] = OSD_OT_MAX,
133 [OSD_OT_XATTR_SET] = OSD_OT_XATTR_SET,
134 [OSD_OT_CREATE] = OSD_OT_DESTROY,
135 [OSD_OT_DESTROY] = OSD_OT_CREATE,
136 [OSD_OT_REF_ADD] = OSD_OT_REF_DEL,
137 [OSD_OT_REF_DEL] = OSD_OT_REF_ADD,
138 [OSD_OT_WRITE] = OSD_OT_WRITE,
139 [OSD_OT_INSERT] = OSD_OT_DELETE,
140 [OSD_OT_DELETE] = OSD_OT_INSERT,
141 [OSD_OT_QUOTA] = OSD_OT_MAX,
144 static int osd_has_index(const struct osd_object *obj)
146 return obj->oo_dt.do_index_ops != NULL;
149 static int osd_object_invariant(const struct lu_object *l)
151 return osd_invariant(osd_obj(l));
155 * Concurrency: doesn't matter
159 * Concurrency: doesn't matter
161 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
163 struct osd_thread_info *oti = osd_oti_get(env);
164 return oti->oti_w_locks > 0 && o->oo_owner == env;
168 * Concurrency: doesn't access mutable data
170 static int osd_root_get(const struct lu_env *env,
171 struct dt_device *dev, struct lu_fid *f)
173 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
178 * OSD object methods.
182 * Concurrency: no concurrent access is possible that early in object
185 static struct lu_object *osd_object_alloc(const struct lu_env *env,
186 const struct lu_object_header *hdr,
189 struct osd_object *mo;
195 l = &mo->oo_dt.do_lu;
196 dt_object_init(&mo->oo_dt, NULL, d);
197 mo->oo_dt.do_ops = &osd_obj_ea_ops;
198 l->lo_ops = &osd_lu_obj_ops;
199 init_rwsem(&mo->oo_sem);
200 init_rwsem(&mo->oo_ext_idx_sem);
201 spin_lock_init(&mo->oo_guard);
208 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
209 struct dentry *dentry, struct lustre_mdt_attrs *lma)
213 CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
214 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
215 info->oti_mdt_attrs_old, LMA_OLD_SIZE);
217 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
218 memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
220 lustre_lma_swab(lma);
221 /* Check LMA compatibility */
222 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
223 CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
224 "for fid = "DFID", ino = %lu\n",
225 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
226 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
227 PFID(&lma->lma_self_fid), inode->i_ino);
230 } else if (rc == 0) {
238 * retrieve object from backend ext fs.
240 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
241 struct osd_inode_id *id)
243 struct inode *inode = NULL;
245 /* if we look for an inode withing a running
246 * transaction, then we risk to deadlock */
247 /* osd_dirent_check_repair() breaks this */
248 /*LASSERT(current->journal_info == NULL);*/
250 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
252 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
253 id->oii_ino, PTR_ERR(inode));
254 } else if (id->oii_gen != OSD_OII_NOGEN &&
255 inode->i_generation != id->oii_gen) {
256 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
257 "i_generation = %u\n",
258 id->oii_ino, id->oii_gen, inode->i_generation);
260 inode = ERR_PTR(-ESTALE);
261 } else if (inode->i_nlink == 0) {
262 /* due to parallel readdir and unlink,
263 * we can have dead inode here. */
264 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
266 inode = ERR_PTR(-ESTALE);
267 } else if (is_bad_inode(inode)) {
268 CWARN("%.16s: bad inode: ino = %u\n",
269 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
271 inode = ERR_PTR(-ENOENT);
273 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
274 if (id->oii_gen == OSD_OII_NOGEN)
275 osd_id_gen(id, inode->i_ino, inode->i_generation);
277 /* Do not update file c/mtime in ldiskfs.
278 * NB: we don't have any lock to protect this because we don't
279 * have reference on osd_object now, but contention with
280 * another lookup + attr_set can't happen in the tiny window
281 * between if (...) and set S_NOCMTIME. */
282 if (!(inode->i_flags & S_NOCMTIME))
283 inode->i_flags |= S_NOCMTIME;
288 static struct inode *
289 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
290 struct osd_inode_id *id, struct lu_fid *fid)
292 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
296 inode = osd_iget(info, dev, id);
300 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
302 *fid = lma->lma_self_fid;
303 } else if (rc == -ENODATA) {
304 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
305 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
307 lu_igif_build(fid, inode->i_ino, inode->i_generation);
315 static struct inode *osd_iget_check(struct osd_thread_info *info,
316 struct osd_device *dev,
317 const struct lu_fid *fid,
318 struct osd_inode_id *id,
325 /* The cached OI mapping is trustable. If we cannot locate the inode
326 * via the cached OI mapping, then return the failure to the caller
327 * directly without further OI checking. */
329 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
332 if (cached || (rc != -ENOENT && rc != -ESTALE)) {
333 CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
342 if (is_bad_inode(inode)) {
345 CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
353 if (id->oii_gen != OSD_OII_NOGEN &&
354 inode->i_generation != id->oii_gen) {
357 CDEBUG(D_INODE, "unmatched inode: ino = %u, "
358 "oii_gen = %u, i_generation = %u\n",
359 id->oii_ino, id->oii_gen, inode->i_generation);
367 if (inode->i_nlink == 0) {
370 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
378 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
382 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
384 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
385 /* XXX: There are four possible cases:
387 * Backup/restore caused the OI invalid.
389 * Someone unlinked the object but NOT removed
390 * the OI mapping, such as mount target device
391 * as ldiskfs, and modify something directly.
393 * Someone just removed the object between the
394 * former oi_lookup and the iget. It is normal.
395 * 4. Other failure cases.
397 * Generally, when the device is mounted, it will
398 * auto check whether the system is restored from
399 * file-level backup or not. We trust such detect
400 * to distinguish the 1st case from the 2nd case. */
402 if (!IS_ERR(inode) && inode->i_generation != 0 &&
403 inode->i_generation == id->oii_gen)
404 /* "id->oii_gen != OSD_OII_NOGEN" is for
405 * "@cached == false" case. */
410 /* If the OI mapping was in OI file before the
411 * osd_iget_check(), but now, it is disappear,
412 * then it must be removed by race. That is a
413 * normal race case. */
416 if (id->oii_gen == OSD_OII_NOGEN)
417 osd_id_gen(id, inode->i_ino, inode->i_generation);
419 /* Do not update file c/mtime in ldiskfs.
420 * NB: we don't have any lock to protect this because we don't
421 * have reference on osd_object now, but contention with
422 * another lookup + attr_set can't happen in the tiny window
423 * between if (...) and set S_NOCMTIME. */
424 if (!(inode->i_flags & S_NOCMTIME))
425 inode->i_flags |= S_NOCMTIME;
442 * \retval +v: new filter_fid, does not contain self-fid
443 * \retval 0: filter_fid_old, contains self-fid
444 * \retval -v: other failure cases
446 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
447 struct dentry *dentry, struct lu_fid *fid)
449 struct filter_fid_old *ff = &info->oti_ff;
450 struct ost_id *ostid = &info->oti_ostid;
453 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
454 if (rc == sizeof(*ff)) {
456 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
457 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
458 /* XXX: use 0 as the index for compatibility, the caller will
459 * handle index related issues when necessarry. */
460 ostid_to_fid(fid, ostid, 0);
461 } else if (rc == sizeof(struct filter_fid)) {
463 } else if (rc >= 0) {
470 static int osd_lma_self_repair(struct osd_thread_info *info,
471 struct osd_device *osd, struct inode *inode,
472 const struct lu_fid *fid, __u32 compat)
477 LASSERT(current->journal_info == NULL);
479 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
480 osd_dto_credits_noquota[DTO_XATTR_SET]);
483 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
488 rc = osd_ea_fid_set(info, inode, fid, compat, 0);
490 CWARN("%s: cannot self repair the LMA: rc = %d\n",
492 ldiskfs_journal_stop(jh);
496 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
498 struct osd_thread_info *info = osd_oti_get(env);
499 struct osd_device *osd = osd_obj2dev(obj);
500 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
501 struct inode *inode = obj->oo_inode;
502 struct dentry *dentry = &info->oti_obj_dentry;
503 struct lu_fid *fid = NULL;
504 const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
508 CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
509 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
510 info->oti_mdt_attrs_old, LMA_OLD_SIZE);
511 if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
512 fid = &lma->lma_self_fid;
513 rc = osd_get_idif(info, inode, dentry, fid);
514 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
515 /* For the given OST-object, if it has neither LMA nor
516 * FID in XATTR_NAME_FID, then the given FID (which is
517 * contained in the @obj, from client RPC for locating
518 * the OST-object) is trusted. We use it to generate
520 osd_lma_self_repair(info, osd, inode, rfid,
531 lustre_lma_swab(lma);
532 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
533 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
534 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
535 "fid = "DFID", ino = %lu\n", osd_name(osd),
536 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
537 PFID(rfid), inode->i_ino);
540 fid = &lma->lma_self_fid;
544 if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
545 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
546 struct ost_id *oi = &info->oti_ostid;
547 struct lu_fid *fid1 = &info->oti_fid3;
548 __u32 idx = fid_idif_ost_idx(rfid);
550 /* For old IDIF, the OST index is not part of the IDIF,
551 * Means that different OSTs may have the same IDIFs.
552 * Under such case, we need to make some compatible
553 * check to make sure to trigger OI scrub properly. */
554 if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
555 /* Given @rfid is new, LMA is old. */
556 fid_to_ostid(fid, oi);
557 ostid_to_fid(fid1, oi, idx);
558 if (lu_fid_eq(fid1, rfid)) {
559 if (osd->od_index_in_idif)
560 osd_lma_self_repair(info, osd,
574 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
575 const struct lu_fid *fid,
576 const struct lu_object_conf *conf)
578 struct osd_thread_info *info;
579 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
580 struct osd_device *dev;
581 struct osd_idmap_cache *oic;
582 struct osd_inode_id *id;
584 struct osd_scrub *scrub;
585 struct scrub_file *sf;
589 bool triggered = false;
592 LINVRNT(osd_invariant(obj));
593 LASSERT(obj->oo_inode == NULL);
594 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
597 scrub = &dev->od_scrub;
598 sf = &scrub->os_file;
599 info = osd_oti_get(env);
601 oic = &info->oti_cache;
603 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
606 /* For the object is created as locking anchor, or for the object to
607 * be created on disk. No need to osd_oi_lookup() at here because FID
608 * shouldn't never be re-used, if it's really a duplicate FID from
609 * unexpected reason, we should be able to detect it later by calling
610 * do_create->osd_oi_insert(). */
611 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
612 GOTO(out, result = 0);
614 /* Search order: 1. per-thread cache. */
615 if (lu_fid_eq(fid, &oic->oic_fid) &&
616 likely(oic->oic_dev == dev)) {
622 if (!list_empty(&scrub->os_inconsistent_items)) {
623 /* Search order: 2. OI scrub pending list. */
624 result = osd_oii_lookup(dev, fid, id);
630 /* Search order: 3. OI files. */
631 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
632 if (result == -ENOENT) {
633 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
634 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
635 !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
637 GOTO(out, result = 0);
646 inode = osd_iget_check(info, dev, fid, id, cached);
648 result = PTR_ERR(inode);
649 if (result == -ENOENT || result == -ESTALE)
650 GOTO(out, result = -ENOENT);
652 if (result == -EREMCHG) {
655 if (unlikely(triggered))
656 GOTO(out, result = saved);
659 if (thread_is_running(&scrub->os_thread)) {
660 result = -EINPROGRESS;
661 } else if (!dev->od_noscrub) {
662 result = osd_scrub_start(dev, SS_AUTO_FULL |
663 SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
664 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
665 "for "DFID", rc = %d [1]\n",
666 osd_name(dev), PFID(fid), result);
667 if (result == 0 || result == -EALREADY)
668 result = -EINPROGRESS;
673 if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD))
676 /* We still have chance to get the valid inode: for the
677 * object which is referenced by remote name entry, the
678 * object on the local MDT will be linked under the dir
679 * of "/REMOTE_PARENT_DIR" with its FID string as name.
681 * We do not know whether the object for the given FID
682 * is referenced by some remote name entry or not, and
683 * especially for DNE II, a multiple-linked object may
684 * have many name entries reside on many MDTs.
686 * To simplify the operation, OSD will not distinguish
687 * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
688 * only happened for the RPC from other MDT during the
689 * OI scrub, or for the client side RPC with FID only,
690 * such as FID to path, or from old connected client. */
692 result = osd_lookup_in_remote_parent(info, dev,
705 obj->oo_inode = inode;
706 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
708 result = osd_check_lma(env, obj);
710 if (result == -ENODATA) {
712 result = osd_oi_lookup(info, dev, fid, id,
715 /* result == -ENOENT means that the OI
716 * mapping has been removed by race,
717 * the target inode belongs to other
720 * Others error also can be returned
723 obj->oo_inode = NULL;
726 /* result == 0 means the cached OI
727 * mapping is still in the OI file,
728 * the target the inode is valid. */
731 /* The current OI mapping is from the OI file,
732 * since the inode has been found via
733 * osd_iget_check(), no need recheck OI. */
740 obj->oo_inode = NULL;
741 if (result != -EREMCHG)
745 result = osd_oi_lookup(info, dev, fid, id,
747 /* result == -ENOENT means the cached OI mapping
748 * has been removed from the OI file by race,
749 * above target inode belongs to other object.
751 * Others error also can be returned directly. */
755 /* result == 0, goto trigger */
757 /* The current OI mapping is from the OI file,
758 * since the inode has been found via
759 * osd_iget_check(), no need recheck OI. */
766 obj->oo_compat_dot_created = 1;
767 obj->oo_compat_dotdot_created = 1;
769 if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
770 GOTO(out, result = 0);
772 LASSERT(obj->oo_hl_head == NULL);
773 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
774 if (obj->oo_hl_head == NULL) {
775 obj->oo_inode = NULL;
777 GOTO(out, result = -ENOMEM);
779 GOTO(out, result = 0);
782 if (result != 0 && cached)
783 fid_zero(&oic->oic_fid);
785 LINVRNT(osd_invariant(obj));
790 * Concurrency: shouldn't matter.
792 static void osd_object_init0(struct osd_object *obj)
794 LASSERT(obj->oo_inode != NULL);
795 obj->oo_dt.do_body_ops = &osd_body_ops;
796 obj->oo_dt.do_lu.lo_header->loh_attr |=
797 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
801 * Concurrency: no concurrent access is possible that early in object
804 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
805 const struct lu_object_conf *conf)
807 struct osd_object *obj = osd_obj(l);
810 LINVRNT(osd_invariant(obj));
812 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
813 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
814 l->lo_header->loh_attr |= LOHA_EXISTS;
818 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
819 obj->oo_dt.do_body_ops = &osd_body_ops_new;
820 if (result == 0 && obj->oo_inode != NULL)
821 osd_object_init0(obj);
823 LINVRNT(osd_invariant(obj));
828 * Concurrency: no concurrent access is possible that late in object
831 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
833 struct osd_object *obj = osd_obj(l);
835 LINVRNT(osd_invariant(obj));
837 dt_object_fini(&obj->oo_dt);
838 if (obj->oo_hl_head != NULL)
839 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
844 * Concurrency: no concurrent access is possible that late in object
847 static void osd_index_fini(struct osd_object *o)
849 struct iam_container *bag;
851 if (o->oo_dir != NULL) {
852 bag = &o->oo_dir->od_container;
853 if (o->oo_inode != NULL) {
854 if (bag->ic_object == o->oo_inode)
855 iam_container_fini(bag);
857 OBD_FREE_PTR(o->oo_dir);
863 * Concurrency: no concurrent access is possible that late in object
864 * life-cycle (for all existing callers, that is. New callers have to provide
865 * their own locking.)
867 static int osd_inode_unlinked(const struct inode *inode)
869 return inode->i_nlink == 0;
873 OSD_TXN_OI_DELETE_CREDITS = 20,
874 OSD_TXN_INODE_DELETE_CREDITS = 20
881 #if OSD_THANDLE_STATS
883 * Set time when the handle is allocated
885 static void osd_th_alloced(struct osd_thandle *oth)
887 oth->oth_alloced = cfs_time_current();
891 * Set time when the handle started
893 static void osd_th_started(struct osd_thandle *oth)
895 oth->oth_started = cfs_time_current();
899 * Helper function to convert time interval to microseconds packed in
902 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
906 cfs_duration_usec(cfs_time_sub(end, start), &val);
907 return val.tv_sec * 1000000 + val.tv_usec;
911 * Check whether the we deal with this handle for too long.
913 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
914 cfs_time_t alloced, cfs_time_t started,
917 cfs_time_t now = cfs_time_current();
919 LASSERT(dev != NULL);
921 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
922 interval_to_usec(alloced, started));
923 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
924 interval_to_usec(started, closed));
925 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
926 interval_to_usec(closed, now));
928 if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
929 CWARN("transaction handle %p was open for too long: "
931 "alloced "CFS_TIME_T" ,"
932 "started "CFS_TIME_T" ,"
933 "closed "CFS_TIME_T"\n",
934 oth, now, alloced, started, closed);
935 libcfs_debug_dumpstack(NULL);
939 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
941 cfs_time_t __closed = cfs_time_current(); \
942 cfs_time_t __alloced = oth->oth_alloced; \
943 cfs_time_t __started = oth->oth_started; \
946 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
949 #else /* OSD_THANDLE_STATS */
951 #define osd_th_alloced(h) do {} while(0)
952 #define osd_th_started(h) do {} while(0)
953 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
955 #endif /* OSD_THANDLE_STATS */
958 * Concurrency: doesn't access mutable data.
960 static int osd_param_is_not_sane(const struct osd_device *dev,
961 const struct thandle *th)
963 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
965 return oh->ot_credits > osd_transaction_size(dev);
969 * Concurrency: shouldn't matter.
971 static void osd_trans_commit_cb(struct super_block *sb,
972 struct ldiskfs_journal_cb_entry *jcb, int error)
974 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
975 struct thandle *th = &oh->ot_super;
976 struct lu_device *lud = &th->th_dev->dd_lu_dev;
977 struct dt_txn_commit_cb *dcb, *tmp;
979 LASSERT(oh->ot_handle == NULL);
982 CERROR("transaction @0x%p commit error: %d\n", th, error);
984 dt_txn_hook_commit(th);
986 /* call per-transaction callbacks if any */
987 list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
989 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
990 "commit callback entry: magic=%x name='%s'\n",
991 dcb->dcb_magic, dcb->dcb_name);
992 list_del_init(&dcb->dcb_linkage);
993 dcb->dcb_func(NULL, th, dcb, error);
996 lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1000 lu_context_exit(&th->th_ctx);
1001 lu_context_fini(&th->th_ctx);
1005 static struct thandle *osd_trans_create(const struct lu_env *env,
1006 struct dt_device *d)
1008 struct osd_thread_info *oti = osd_oti_get(env);
1009 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1010 struct osd_thandle *oh;
1014 /* on pending IO in this thread should left from prev. request */
1015 LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1017 th = ERR_PTR(-ENOMEM);
1018 OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
1020 oh->ot_quota_trans = &oti->oti_quota_trans;
1021 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1025 th->th_tags = LCT_TX_HANDLE;
1027 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1028 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1031 memset(oti->oti_declare_ops, 0,
1032 sizeof(oti->oti_declare_ops));
1033 memset(oti->oti_declare_ops_cred, 0,
1034 sizeof(oti->oti_declare_ops_cred));
1035 memset(oti->oti_declare_ops_used, 0,
1036 sizeof(oti->oti_declare_ops_used));
1041 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1043 struct osd_thread_info *oti = osd_oti_get(env);
1044 struct osd_thandle *oh;
1046 oh = container_of0(th, struct osd_thandle, ot_super);
1047 LASSERT(oh != NULL);
1049 CWARN(" create: %u/%u/%u, destroy: %u/%u/%u\n",
1050 oti->oti_declare_ops[OSD_OT_CREATE],
1051 oti->oti_declare_ops_cred[OSD_OT_CREATE],
1052 oti->oti_declare_ops_used[OSD_OT_CREATE],
1053 oti->oti_declare_ops[OSD_OT_DESTROY],
1054 oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1055 oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1056 CWARN(" attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1057 oti->oti_declare_ops[OSD_OT_ATTR_SET],
1058 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1059 oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1060 oti->oti_declare_ops[OSD_OT_XATTR_SET],
1061 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1062 oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1063 CWARN(" write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1064 oti->oti_declare_ops[OSD_OT_WRITE],
1065 oti->oti_declare_ops_cred[OSD_OT_WRITE],
1066 oti->oti_declare_ops_used[OSD_OT_WRITE],
1067 oti->oti_declare_ops[OSD_OT_PUNCH],
1068 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1069 oti->oti_declare_ops_used[OSD_OT_PUNCH],
1070 oti->oti_declare_ops[OSD_OT_QUOTA],
1071 oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1072 oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1073 CWARN(" insert: %u/%u/%u, delete: %u/%u/%u\n",
1074 oti->oti_declare_ops[OSD_OT_INSERT],
1075 oti->oti_declare_ops_cred[OSD_OT_INSERT],
1076 oti->oti_declare_ops_used[OSD_OT_INSERT],
1077 oti->oti_declare_ops[OSD_OT_DELETE],
1078 oti->oti_declare_ops_cred[OSD_OT_DELETE],
1079 oti->oti_declare_ops_used[OSD_OT_DELETE]);
1080 CWARN(" ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1081 oti->oti_declare_ops[OSD_OT_REF_ADD],
1082 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1083 oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1084 oti->oti_declare_ops[OSD_OT_REF_DEL],
1085 oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1086 oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1090 * Concurrency: shouldn't matter.
1092 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1095 struct osd_thread_info *oti = osd_oti_get(env);
1096 struct osd_device *dev = osd_dt_dev(d);
1098 struct osd_thandle *oh;
1103 LASSERT(current->journal_info == NULL);
1105 oh = container_of0(th, struct osd_thandle, ot_super);
1106 LASSERT(oh != NULL);
1107 LASSERT(oh->ot_handle == NULL);
1109 rc = dt_txn_hook_start(env, d, th);
1113 if (unlikely(osd_param_is_not_sane(dev, th))) {
1114 static unsigned long last_printed;
1115 static int last_credits;
1117 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1118 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1120 osd_journal(dev)->j_max_transaction_buffers);
1122 osd_trans_dump_creds(env, th);
1124 if (last_credits != oh->ot_credits &&
1125 time_after(jiffies, last_printed +
1126 msecs_to_jiffies(60 * MSEC_PER_SEC))) {
1127 libcfs_debug_dumpstack(NULL);
1128 last_credits = oh->ot_credits;
1129 last_printed = jiffies;
1131 /* XXX Limit the credits to 'max_transaction_buffers', and
1132 * let the underlying filesystem to catch the error if
1133 * we really need so many credits.
1135 * This should be removed when we can calculate the
1136 * credits precisely. */
1137 oh->ot_credits = osd_transaction_size(dev);
1141 * XXX temporary stuff. Some abstraction layer should
1144 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1148 LASSERT(oti->oti_txns == 0);
1149 lu_context_init(&th->th_ctx, th->th_tags);
1150 lu_context_enter(&th->th_ctx);
1152 lu_device_get(&d->dd_lu_dev);
1153 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1164 static int osd_seq_exists(const struct lu_env *env,
1165 struct osd_device *osd, u64 seq)
1167 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1168 struct seq_server_site *ss = osd_seq_site(osd);
1172 LASSERT(ss != NULL);
1173 LASSERT(ss->ss_server_fld != NULL);
1175 rc = osd_fld_lookup(env, osd, seq, range);
1178 CERROR("%s: can't lookup FLD sequence "LPX64
1179 ": rc = %d\n", osd_name(osd), seq, rc);
1183 RETURN(ss->ss_node_id == range->lsr_index);
1186 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1188 struct dt_txn_commit_cb *dcb;
1189 struct dt_txn_commit_cb *tmp;
1191 /* call per-transaction stop callbacks if any */
1192 list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1194 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1195 "commit callback entry: magic=%x name='%s'\n",
1196 dcb->dcb_magic, dcb->dcb_name);
1197 list_del_init(&dcb->dcb_linkage);
1198 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1203 * Concurrency: shouldn't matter.
1205 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1208 int rc = 0, remove_agents = 0;
1209 struct osd_thandle *oh;
1210 struct osd_thread_info *oti = osd_oti_get(env);
1211 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1212 struct osd_device *osd = osd_dt_dev(th->th_dev);
1213 struct qsd_instance *qsd = osd->od_quota_slave;
1214 struct lquota_trans *qtrans;
1217 oh = container_of0(th, struct osd_thandle, ot_super);
1219 remove_agents = oh->ot_remove_agents;
1221 qtrans = oh->ot_quota_trans;
1222 oh->ot_quota_trans = NULL;
1224 if (oh->ot_handle != NULL) {
1225 handle_t *hdl = oh->ot_handle;
1228 * add commit callback
1229 * notice we don't do this in osd_trans_start()
1230 * as underlying transaction can change during truncate
1232 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1235 LASSERT(oti->oti_txns == 1);
1238 rc = dt_txn_hook_stop(env, th);
1240 CERROR("%s: failed in transaction hook: rc = %d\n",
1243 osd_trans_stop_cb(oh, rc);
1244 /* hook functions might modify th_sync */
1245 hdl->h_sync = th->th_sync;
1247 oh->ot_handle = NULL;
1248 OSD_CHECK_SLOW_TH(oh, osd, rc = ldiskfs_journal_stop(hdl));
1250 CERROR("%s: failed to stop transaction: rc = %d\n",
1253 osd_trans_stop_cb(oh, th->th_result);
1257 /* inform the quota slave device that the transaction is stopping */
1258 qsd_op_end(env, qsd, qtrans);
1260 /* as we want IO to journal and data IO be concurrent, we don't block
1261 * awaiting data IO completion in osd_do_bio(), instead we wait here
1262 * once transaction is submitted to the journal. all reqular requests
1263 * don't do direct IO (except read/write), thus this wait_event becomes
1266 * IMPORTANT: we have to wait till any IO submited by the thread is
1267 * completed otherwise iobuf may be corrupted by different request
1269 wait_event(iobuf->dr_wait,
1270 atomic_read(&iobuf->dr_numreqs) == 0);
1271 osd_fini_iobuf(osd, iobuf);
1273 rc = iobuf->dr_error;
1275 if (unlikely(remove_agents != 0))
1276 osd_process_scheduled_agent_removals(env, osd);
1281 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1283 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1286 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1287 LASSERT(&dcb->dcb_func != NULL);
1288 if (dcb->dcb_flags & DCB_TRANS_STOP)
1289 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1291 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1297 * Called just before object is freed. Releases all resources except for
1298 * object itself (that is released by osd_object_free()).
1300 * Concurrency: no concurrent access is possible that late in object
1303 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1305 struct osd_object *obj = osd_obj(l);
1306 struct inode *inode = obj->oo_inode;
1308 LINVRNT(osd_invariant(obj));
1311 * If object is unlinked remove fid->ino mapping from object index.
1314 osd_index_fini(obj);
1315 if (inode != NULL) {
1316 struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave;
1317 qid_t uid = i_uid_read(inode);
1318 qid_t gid = i_gid_read(inode);
1321 obj->oo_inode = NULL;
1324 struct osd_thread_info *info = osd_oti_get(env);
1325 struct lquota_id_info *qi = &info->oti_qi;
1327 /* Release granted quota to master if necessary */
1328 qi->lqi_id.qid_uid = uid;
1329 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1331 qi->lqi_id.qid_uid = gid;
1332 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1338 * Concurrency: ->loo_object_release() is called under site spin-lock.
1340 static void osd_object_release(const struct lu_env *env,
1341 struct lu_object *l)
1346 * Concurrency: shouldn't matter.
1348 static int osd_object_print(const struct lu_env *env, void *cookie,
1349 lu_printer_t p, const struct lu_object *l)
1351 struct osd_object *o = osd_obj(l);
1352 struct iam_descr *d;
1354 if (o->oo_dir != NULL)
1355 d = o->oo_dir->od_container.ic_descr;
1358 return (*p)(env, cookie,
1359 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1361 o->oo_inode ? o->oo_inode->i_ino : 0UL,
1362 o->oo_inode ? o->oo_inode->i_generation : 0,
1363 d ? d->id_ops->id_name : "plain");
1366 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1369 * Concurrency: shouldn't matter.
1371 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1372 struct obd_statfs *sfs)
1374 struct osd_device *osd = osd_dt_dev(d);
1375 struct super_block *sb = osd_sb(osd);
1376 struct kstatfs *ksfs;
1379 if (unlikely(osd->od_mnt == NULL))
1380 return -EINPROGRESS;
1382 /* osd_lproc.c call this without env, allocate ksfs for that case */
1383 if (unlikely(env == NULL)) {
1384 OBD_ALLOC_PTR(ksfs);
1388 ksfs = &osd_oti_get(env)->oti_ksfs;
1391 spin_lock(&osd->od_osfs_lock);
1392 result = sb->s_op->statfs(sb->s_root, ksfs);
1393 if (likely(result == 0)) { /* N.B. statfs can't really fail */
1394 statfs_pack(sfs, ksfs);
1395 if (unlikely(sb->s_flags & MS_RDONLY))
1396 sfs->os_state = OS_STATE_READONLY;
1397 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb,
1398 LDISKFS_FEATURE_INCOMPAT_EXTENTS))
1399 sfs->os_maxbytes = sb->s_maxbytes;
1401 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
1403 spin_unlock(&osd->od_osfs_lock);
1405 if (unlikely(env == NULL))
1408 /* Reserve a small amount of space for local objects like last_rcvd,
1409 * llog, quota files, ... */
1410 if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1413 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1414 /** Take out metadata overhead for indirect blocks */
1415 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1422 * Estimate space needed for file creations. We assume the largest filename
1423 * which is 2^64 - 1, hence a filename of 20 chars.
1424 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1426 #ifdef __LDISKFS_DIR_REC_LEN
1427 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1429 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1433 * Concurrency: doesn't access mutable data.
1435 static void osd_conf_get(const struct lu_env *env,
1436 const struct dt_device *dev,
1437 struct dt_device_param *param)
1439 struct super_block *sb = osd_sb(osd_dt_dev(dev));
1443 * XXX should be taken from not-yet-existing fs abstraction layer.
1445 param->ddp_max_name_len = LDISKFS_NAME_LEN;
1446 param->ddp_max_nlink = LDISKFS_LINK_MAX;
1447 param->ddp_block_shift = sb->s_blocksize_bits;
1448 param->ddp_mount_type = LDD_MT_LDISKFS;
1449 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS))
1450 param->ddp_maxbytes = sb->s_maxbytes;
1452 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
1453 /* Overhead estimate should be fairly accurate, so we really take a tiny
1454 * error margin which also avoids fragmenting the filesystem too much */
1455 param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1456 /* inode are statically allocated, so per-inode space consumption
1457 * is the space consumed by the directory entry */
1458 param->ddp_inodespace = PER_OBJ_USAGE;
1459 /* per-fragment overhead to be used by the client code */
1460 param->ddp_grant_frag = 6 * LDISKFS_BLOCK_SIZE(sb);
1461 param->ddp_mntopts = 0;
1462 if (test_opt(sb, XATTR_USER))
1463 param->ddp_mntopts |= MNTOPT_USERXATTR;
1464 if (test_opt(sb, POSIX_ACL))
1465 param->ddp_mntopts |= MNTOPT_ACL;
1467 /* LOD might calculate the max stripe count based on max_ea_size,
1468 * so we need take account in the overhead as well,
1469 * xattr_header + magic + xattr_entry_head */
1470 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1471 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1473 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1474 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1475 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1479 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1483 * Concurrency: shouldn't matter.
1485 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1489 CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1491 rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1493 CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
1494 LUSTRE_OSD_LDISKFS_NAME, rc);
1500 * Start commit for OSD device.
1502 * An implementation of dt_commit_async method for OSD device.
1503 * Asychronously starts underlayng fs sync and thereby a transaction
1506 * \param env environment
1507 * \param d dt device
1509 * \see dt_device_operations
1511 static int osd_commit_async(const struct lu_env *env,
1512 struct dt_device *d)
1514 struct super_block *s = osd_sb(osd_dt_dev(d));
1517 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1518 RETURN(s->s_op->sync_fs(s, 0));
1522 * Concurrency: shouldn't matter.
1525 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1527 struct super_block *sb = osd_sb(osd_dt_dev(d));
1528 struct block_device *dev = sb->s_bdev;
1529 #ifdef HAVE_DEV_SET_RDONLY
1530 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1533 int rc = -EOPNOTSUPP;
1537 #ifdef HAVE_DEV_SET_RDONLY
1538 CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1540 if (jdev && (jdev != dev)) {
1541 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1543 dev_set_rdonly(jdev);
1545 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1546 dev_set_rdonly(dev);
1548 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1549 osd_dt_dev(d)->od_svname, (long)dev, rc);
1555 * Note: we do not count into QUOTA here.
1556 * If we mount with --data_journal we may need more.
1558 const int osd_dto_credits_noquota[DTO_NR] = {
1561 * INDEX_EXTRA_TRANS_BLOCKS(8) +
1562 * SINGLEDATA_TRANS_BLOCKS(8)
1563 * XXX Note: maybe iam need more, since iam have more level than
1566 [DTO_INDEX_INSERT] = 16,
1569 * just modify a single entry, probably merge few within a block
1571 [DTO_INDEX_DELETE] = 1,
1575 [DTO_INDEX_UPDATE] = 16,
1577 * 4(inode, inode bits, groups, GDT)
1578 * notice: OI updates are counted separately with DTO_INDEX_INSERT
1580 [DTO_OBJECT_CREATE] = 4,
1582 * 4(inode, inode bits, groups, GDT)
1583 * notice: OI updates are counted separately with DTO_INDEX_DELETE
1585 [DTO_OBJECT_DELETE] = 4,
1587 * Attr set credits (inode)
1589 [DTO_ATTR_SET_BASE] = 1,
1591 * Xattr set. The same as xattr of EXT3.
1592 * DATA_TRANS_BLOCKS(14)
1593 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1594 * are also counted in. Do not know why?
1596 [DTO_XATTR_SET] = 14,
1598 * credits for inode change during write.
1600 [DTO_WRITE_BASE] = 3,
1602 * credits for single block write.
1604 [DTO_WRITE_BLOCK] = 14,
1606 * Attr set credits for chown.
1607 * This is extra credits for setattr, and it is null without quota
1609 [DTO_ATTR_SET_CHOWN] = 0
1612 static const struct dt_device_operations osd_dt_ops = {
1613 .dt_root_get = osd_root_get,
1614 .dt_statfs = osd_statfs,
1615 .dt_trans_create = osd_trans_create,
1616 .dt_trans_start = osd_trans_start,
1617 .dt_trans_stop = osd_trans_stop,
1618 .dt_trans_cb_add = osd_trans_cb_add,
1619 .dt_conf_get = osd_conf_get,
1620 .dt_sync = osd_sync,
1622 .dt_commit_async = osd_commit_async,
1625 static void osd_object_read_lock(const struct lu_env *env,
1626 struct dt_object *dt, unsigned role)
1628 struct osd_object *obj = osd_dt_obj(dt);
1629 struct osd_thread_info *oti = osd_oti_get(env);
1631 LINVRNT(osd_invariant(obj));
1633 LASSERT(obj->oo_owner != env);
1634 down_read_nested(&obj->oo_sem, role);
1636 LASSERT(obj->oo_owner == NULL);
1640 static void osd_object_write_lock(const struct lu_env *env,
1641 struct dt_object *dt, unsigned role)
1643 struct osd_object *obj = osd_dt_obj(dt);
1644 struct osd_thread_info *oti = osd_oti_get(env);
1646 LINVRNT(osd_invariant(obj));
1648 LASSERT(obj->oo_owner != env);
1649 down_write_nested(&obj->oo_sem, role);
1651 LASSERT(obj->oo_owner == NULL);
1652 obj->oo_owner = env;
1656 static void osd_object_read_unlock(const struct lu_env *env,
1657 struct dt_object *dt)
1659 struct osd_object *obj = osd_dt_obj(dt);
1660 struct osd_thread_info *oti = osd_oti_get(env);
1662 LINVRNT(osd_invariant(obj));
1664 LASSERT(oti->oti_r_locks > 0);
1666 up_read(&obj->oo_sem);
1669 static void osd_object_write_unlock(const struct lu_env *env,
1670 struct dt_object *dt)
1672 struct osd_object *obj = osd_dt_obj(dt);
1673 struct osd_thread_info *oti = osd_oti_get(env);
1675 LINVRNT(osd_invariant(obj));
1677 LASSERT(obj->oo_owner == env);
1678 LASSERT(oti->oti_w_locks > 0);
1680 obj->oo_owner = NULL;
1681 up_write(&obj->oo_sem);
1684 static int osd_object_write_locked(const struct lu_env *env,
1685 struct dt_object *dt)
1687 struct osd_object *obj = osd_dt_obj(dt);
1689 LINVRNT(osd_invariant(obj));
1691 return obj->oo_owner == env;
1694 static struct timespec *osd_inode_time(const struct lu_env *env,
1695 struct inode *inode, __u64 seconds)
1697 struct osd_thread_info *oti = osd_oti_get(env);
1698 struct timespec *t = &oti->oti_time;
1700 t->tv_sec = seconds;
1702 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1707 static void osd_inode_getattr(const struct lu_env *env,
1708 struct inode *inode, struct lu_attr *attr)
1710 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1711 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1712 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1715 attr->la_atime = LTIME_S(inode->i_atime);
1716 attr->la_mtime = LTIME_S(inode->i_mtime);
1717 attr->la_ctime = LTIME_S(inode->i_ctime);
1718 attr->la_mode = inode->i_mode;
1719 attr->la_size = i_size_read(inode);
1720 attr->la_blocks = inode->i_blocks;
1721 attr->la_uid = i_uid_read(inode);
1722 attr->la_gid = i_gid_read(inode);
1723 attr->la_flags = LDISKFS_I(inode)->i_flags;
1724 attr->la_nlink = inode->i_nlink;
1725 attr->la_rdev = inode->i_rdev;
1726 attr->la_blksize = 1 << inode->i_blkbits;
1727 attr->la_blkbits = inode->i_blkbits;
1730 static int osd_attr_get(const struct lu_env *env,
1731 struct dt_object *dt,
1732 struct lu_attr *attr)
1734 struct osd_object *obj = osd_dt_obj(dt);
1736 if (!dt_object_exists(dt))
1739 LASSERT(!dt_object_remote(dt));
1740 LINVRNT(osd_invariant(obj));
1742 spin_lock(&obj->oo_guard);
1743 osd_inode_getattr(env, obj->oo_inode, attr);
1744 spin_unlock(&obj->oo_guard);
1748 static int osd_declare_attr_set(const struct lu_env *env,
1749 struct dt_object *dt,
1750 const struct lu_attr *attr,
1751 struct thandle *handle)
1753 struct osd_thandle *oh;
1754 struct osd_object *obj;
1755 struct osd_thread_info *info = osd_oti_get(env);
1756 struct lquota_id_info *qi = &info->oti_qi;
1764 LASSERT(dt != NULL);
1765 LASSERT(handle != NULL);
1767 obj = osd_dt_obj(dt);
1768 LASSERT(osd_invariant(obj));
1770 oh = container_of0(handle, struct osd_thandle, ot_super);
1771 LASSERT(oh->ot_handle == NULL);
1773 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1774 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1776 if (attr == NULL || obj->oo_inode == NULL)
1779 bspace = obj->oo_inode->i_blocks;
1780 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1781 bspace = toqb(bspace);
1783 /* Changing ownership is always preformed by super user, it should not
1786 * We still need to call the osd_declare_qid() to calculate the journal
1787 * credits for updating quota accounting files and to trigger quota
1788 * space adjustment once the operation is completed.*/
1789 if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
1791 uid = i_uid_read(obj->oo_inode);
1792 qi->lqi_type = USRQUOTA;
1793 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
1794 /* inode accounting */
1795 qi->lqi_is_blk = false;
1797 /* one more inode for the new uid ... */
1798 qi->lqi_id.qid_uid = attr->la_uid;
1800 /* Reserve credits for the new uid */
1801 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1802 if (rc == -EDQUOT || rc == -EINPROGRESS)
1807 /* and one less inode for the current uid */
1808 qi->lqi_id.qid_uid = uid;
1810 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1811 if (rc == -EDQUOT || rc == -EINPROGRESS)
1816 /* block accounting */
1817 qi->lqi_is_blk = true;
1819 /* more blocks for the new uid ... */
1820 qi->lqi_id.qid_uid = attr->la_uid;
1821 qi->lqi_space = bspace;
1823 * Credits for the new uid has been reserved, re-use "obj"
1824 * to save credit reservation.
1826 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1827 if (rc == -EDQUOT || rc == -EINPROGRESS)
1832 /* and finally less blocks for the current uid */
1833 qi->lqi_id.qid_uid = uid;
1834 qi->lqi_space = -bspace;
1835 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1836 if (rc == -EDQUOT || rc == -EINPROGRESS)
1842 gid = i_gid_read(obj->oo_inode);
1843 qi->lqi_type = GRPQUOTA;
1844 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
1846 /* inode accounting */
1847 qi->lqi_is_blk = false;
1849 /* one more inode for the new gid ... */
1850 qi->lqi_id.qid_gid = attr->la_gid;
1852 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1853 if (rc == -EDQUOT || rc == -EINPROGRESS)
1858 /* and one less inode for the current gid */
1859 qi->lqi_id.qid_gid = gid;
1861 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1862 if (rc == -EDQUOT || rc == -EINPROGRESS)
1867 /* block accounting */
1868 qi->lqi_is_blk = true;
1870 /* more blocks for the new gid ... */
1871 qi->lqi_id.qid_gid = attr->la_gid;
1872 qi->lqi_space = bspace;
1873 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1874 if (rc == -EDQUOT || rc == -EINPROGRESS)
1879 /* and finally less blocks for the current gid */
1880 qi->lqi_id.qid_gid = gid;
1881 qi->lqi_space = -bspace;
1882 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1883 if (rc == -EDQUOT || rc == -EINPROGRESS)
1892 static int osd_inode_setattr(const struct lu_env *env,
1893 struct inode *inode, const struct lu_attr *attr)
1895 __u64 bits = attr->la_valid;
1897 /* Only allow set size for regular file */
1898 if (!S_ISREG(inode->i_mode))
1899 bits &= ~(LA_SIZE | LA_BLOCKS);
1904 if (bits & LA_ATIME)
1905 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
1906 if (bits & LA_CTIME)
1907 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
1908 if (bits & LA_MTIME)
1909 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
1910 if (bits & LA_SIZE) {
1911 LDISKFS_I(inode)->i_disksize = attr->la_size;
1912 i_size_write(inode, attr->la_size);
1916 /* OSD should not change "i_blocks" which is used by quota.
1917 * "i_blocks" should be changed by ldiskfs only. */
1918 if (bits & LA_BLOCKS)
1919 inode->i_blocks = attr->la_blocks;
1922 inode->i_mode = (inode->i_mode & S_IFMT) |
1923 (attr->la_mode & ~S_IFMT);
1925 i_uid_write(inode, attr->la_uid);
1927 i_gid_write(inode, attr->la_gid);
1928 if (bits & LA_NLINK)
1929 set_nlink(inode, attr->la_nlink);
1931 inode->i_rdev = attr->la_rdev;
1933 if (bits & LA_FLAGS) {
1934 /* always keep S_NOCMTIME */
1935 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1941 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1943 if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1944 (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1948 ll_vfs_dq_init(inode);
1950 if (attr->la_valid & LA_UID)
1951 iattr.ia_valid |= ATTR_UID;
1952 if (attr->la_valid & LA_GID)
1953 iattr.ia_valid |= ATTR_GID;
1954 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1955 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1957 rc = ll_vfs_dq_transfer(inode, &iattr);
1959 CERROR("%s: quota transfer failed: rc = %d. Is quota "
1960 "enforcement enabled on the ldiskfs "
1961 "filesystem?\n", inode->i_sb->s_id, rc);
1968 static int osd_attr_set(const struct lu_env *env,
1969 struct dt_object *dt,
1970 const struct lu_attr *attr,
1971 struct thandle *handle)
1973 struct osd_object *obj = osd_dt_obj(dt);
1974 struct inode *inode;
1977 if (!dt_object_exists(dt))
1980 LASSERT(handle != NULL);
1981 LASSERT(!dt_object_remote(dt));
1982 LASSERT(osd_invariant(obj));
1984 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1986 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1987 struct osd_thread_info *oti = osd_oti_get(env);
1988 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
1989 struct lu_fid *fid1 = &oti->oti_fid;
1990 struct osd_inode_id *id = &oti->oti_id;
1991 struct iam_path_descr *ipd;
1992 struct iam_container *bag;
1993 struct osd_thandle *oh;
1996 fid_cpu_to_be(fid1, fid0);
1997 memset(id, 1, sizeof(*id));
1998 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1999 fid0)->oi_dir.od_container;
2000 ipd = osd_idx_ipd_get(env, bag);
2001 if (unlikely(ipd == NULL))
2004 oh = container_of0(handle, struct osd_thandle, ot_super);
2005 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2006 (const struct iam_rec *)id, ipd);
2007 osd_ipd_put(env, bag, ipd);
2008 return(rc > 0 ? 0 : rc);
2011 inode = obj->oo_inode;
2013 rc = osd_quota_transfer(inode, attr);
2017 spin_lock(&obj->oo_guard);
2018 rc = osd_inode_setattr(env, inode, attr);
2019 spin_unlock(&obj->oo_guard);
2022 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2024 osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2029 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2030 struct osd_object *obj,
2031 const char *name, const int namelen)
2033 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2036 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2037 umode_t mode, struct dt_allocation_hint *hint,
2041 struct osd_device *osd = osd_obj2dev(obj);
2042 struct osd_thandle *oth;
2043 struct dt_object *parent = NULL;
2044 struct inode *inode;
2046 LINVRNT(osd_invariant(obj));
2047 LASSERT(obj->oo_inode == NULL);
2048 LASSERT(obj->oo_hl_head == NULL);
2050 if (S_ISDIR(mode) && ldiskfs_pdo) {
2051 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2052 if (obj->oo_hl_head == NULL)
2056 oth = container_of(th, struct osd_thandle, ot_super);
2057 LASSERT(oth->ot_handle->h_transaction != NULL);
2059 if (hint != NULL && hint->dah_parent != NULL &&
2060 !dt_object_remote(hint->dah_parent))
2061 parent = hint->dah_parent;
2063 inode = ldiskfs_create_inode(oth->ot_handle,
2064 parent ? osd_dt_obj(parent)->oo_inode :
2065 osd_sb(osd)->s_root->d_inode,
2067 if (!IS_ERR(inode)) {
2068 /* Do not update file c/mtime in ldiskfs. */
2069 inode->i_flags |= S_NOCMTIME;
2071 /* For new created object, it must be consistent,
2072 * and it is unnecessary to scrub against it. */
2073 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2075 obj->oo_inode = inode;
2078 if (obj->oo_hl_head != NULL) {
2079 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2080 obj->oo_hl_head = NULL;
2082 result = PTR_ERR(inode);
2084 LINVRNT(osd_invariant(obj));
2092 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2093 struct lu_attr *attr,
2094 struct dt_allocation_hint *hint,
2095 struct dt_object_format *dof,
2099 struct osd_thandle *oth;
2100 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2102 LASSERT(S_ISDIR(attr->la_mode));
2104 oth = container_of(th, struct osd_thandle, ot_super);
2105 LASSERT(oth->ot_handle->h_transaction != NULL);
2106 result = osd_mkfile(info, obj, mode, hint, th);
2111 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2112 struct lu_attr *attr,
2113 struct dt_allocation_hint *hint,
2114 struct dt_object_format *dof,
2118 struct osd_thandle *oth;
2119 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2121 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2123 LASSERT(S_ISREG(attr->la_mode));
2125 oth = container_of(th, struct osd_thandle, ot_super);
2126 LASSERT(oth->ot_handle->h_transaction != NULL);
2128 result = osd_mkfile(info, obj, mode, hint, th);
2130 LASSERT(obj->oo_inode != NULL);
2131 if (feat->dif_flags & DT_IND_VARKEY)
2132 result = iam_lvar_create(obj->oo_inode,
2133 feat->dif_keysize_max,
2135 feat->dif_recsize_max,
2138 result = iam_lfix_create(obj->oo_inode,
2139 feat->dif_keysize_max,
2141 feat->dif_recsize_max,
2148 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2149 struct lu_attr *attr,
2150 struct dt_allocation_hint *hint,
2151 struct dt_object_format *dof,
2154 LASSERT(S_ISREG(attr->la_mode));
2155 return osd_mkfile(info, obj, (attr->la_mode &
2156 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2159 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2160 struct lu_attr *attr,
2161 struct dt_allocation_hint *hint,
2162 struct dt_object_format *dof,
2165 LASSERT(S_ISLNK(attr->la_mode));
2166 return osd_mkfile(info, obj, (attr->la_mode &
2167 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2170 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2171 struct lu_attr *attr,
2172 struct dt_allocation_hint *hint,
2173 struct dt_object_format *dof,
2176 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2179 LINVRNT(osd_invariant(obj));
2180 LASSERT(obj->oo_inode == NULL);
2181 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2182 S_ISFIFO(mode) || S_ISSOCK(mode));
2184 result = osd_mkfile(info, obj, mode, hint, th);
2186 LASSERT(obj->oo_inode != NULL);
2188 * This inode should be marked dirty for i_rdev. Currently
2189 * that is done in the osd_attr_init().
2191 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2194 LINVRNT(osd_invariant(obj));
2198 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2200 struct dt_allocation_hint *hint,
2201 struct dt_object_format *dof,
2204 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2206 osd_obj_type_f result;
2222 result = osd_mk_index;
2233 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2234 struct dt_object *parent, struct dt_object *child,
2239 ah->dah_parent = parent;
2240 ah->dah_mode = child_mode;
2243 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2244 struct lu_attr *attr, struct dt_object_format *dof)
2246 struct inode *inode = obj->oo_inode;
2247 __u64 valid = attr->la_valid;
2250 attr->la_valid &= ~(LA_TYPE | LA_MODE);
2252 if (dof->dof_type != DFT_NODE)
2253 attr->la_valid &= ~LA_RDEV;
2254 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2255 attr->la_valid &= ~LA_ATIME;
2256 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2257 attr->la_valid &= ~LA_CTIME;
2258 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2259 attr->la_valid &= ~LA_MTIME;
2261 result = osd_quota_transfer(inode, attr);
2265 if (attr->la_valid != 0) {
2266 result = osd_inode_setattr(info->oti_env, inode, attr);
2268 * The osd_inode_setattr() should always succeed here. The
2269 * only error that could be returned is EDQUOT when we are
2270 * trying to change the UID or GID of the inode. However, this
2271 * should not happen since quota enforcement is no longer
2272 * enabled on ldiskfs (lquota takes care of it).
2274 LASSERTF(result == 0, "%d\n", result);
2275 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2278 attr->la_valid = valid;
2282 * Helper function for osd_object_create()
2284 * \retval 0, on success
2286 static int __osd_object_create(struct osd_thread_info *info,
2287 struct osd_object *obj, struct lu_attr *attr,
2288 struct dt_allocation_hint *hint,
2289 struct dt_object_format *dof,
2295 osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
2297 /* we drop umask so that permissions we pass are not affected */
2298 umask = current->fs->umask;
2299 current->fs->umask = 0;
2301 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2303 if (likely(obj->oo_inode != NULL)) {
2304 LASSERT(obj->oo_inode->i_state & I_NEW);
2306 /* Unlock the inode before attr initialization to avoid
2307 * unnecessary dqget operations. LU-6378 */
2308 unlock_new_inode(obj->oo_inode);
2311 if (likely(result == 0)) {
2312 osd_attr_init(info, obj, attr, dof);
2313 osd_object_init0(obj);
2316 /* restore previous umask value */
2317 current->fs->umask = umask;
2319 osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
2325 * Helper function for osd_object_create()
2327 * \retval 0, on success
2329 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2330 const struct lu_fid *fid, struct thandle *th)
2332 struct osd_thread_info *info = osd_oti_get(env);
2333 struct osd_inode_id *id = &info->oti_id;
2334 struct osd_device *osd = osd_obj2dev(obj);
2335 struct osd_thandle *oh;
2338 LASSERT(obj->oo_inode != NULL);
2340 oh = container_of0(th, struct osd_thandle, ot_super);
2341 LASSERT(oh->ot_handle);
2342 osd_trans_exec_op(env, th, OSD_OT_INSERT);
2344 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2345 rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2346 osd_trans_exec_check(env, th, OSD_OT_INSERT);
2351 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2352 u64 seq, struct lu_seq_range *range)
2354 struct seq_server_site *ss = osd_seq_site(osd);
2356 if (fid_seq_is_idif(seq)) {
2357 fld_range_set_ost(range);
2358 range->lsr_index = idif_ost_idx(seq);
2362 if (!fid_seq_in_fldb(seq)) {
2363 fld_range_set_mdt(range);
2365 /* FIXME: If ss is NULL, it suppose not get lsr_index
2367 range->lsr_index = ss->ss_node_id;
2371 LASSERT(ss != NULL);
2372 fld_range_set_any(range);
2373 /* OSD will only do local fld lookup */
2374 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2378 * Concurrency: no external locking is necessary.
2380 static int osd_declare_object_create(const struct lu_env *env,
2381 struct dt_object *dt,
2382 struct lu_attr *attr,
2383 struct dt_allocation_hint *hint,
2384 struct dt_object_format *dof,
2385 struct thandle *handle)
2387 struct osd_thandle *oh;
2391 LASSERT(handle != NULL);
2393 oh = container_of0(handle, struct osd_thandle, ot_super);
2394 LASSERT(oh->ot_handle == NULL);
2396 /* EA object consumes more credits than regular object: osd_mk_index
2397 * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
2398 * leaf_node, could involves the block, block bitmap, groups, GDT
2399 * change for each block, so add 4 * 2 credits in that case. */
2400 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2401 osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
2402 (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
2403 /* Reuse idle OI block may cause additional one OI block
2405 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2406 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2411 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2412 osd_dt_obj(dt), false, NULL, false);
2419 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2420 struct lu_attr *attr,
2421 struct dt_allocation_hint *hint,
2422 struct dt_object_format *dof, struct thandle *th)
2424 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2425 struct osd_object *obj = osd_dt_obj(dt);
2426 struct osd_thread_info *info = osd_oti_get(env);
2430 if (dt_object_exists(dt))
2433 LINVRNT(osd_invariant(obj));
2434 LASSERT(!dt_object_remote(dt));
2435 LASSERT(osd_write_locked(env, obj));
2436 LASSERT(th != NULL);
2438 if (unlikely(fid_is_acct(fid)))
2439 /* Quota files can't be created from the kernel any more,
2440 * 'tune2fs -O quota' will take care of creating them */
2443 result = __osd_object_create(info, obj, attr, hint, dof, th);
2445 result = __osd_oi_insert(env, obj, fid, th);
2446 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
2447 obj->oo_dt.do_body_ops = &osd_body_ops;
2449 LASSERT(ergo(result == 0,
2450 dt_object_exists(dt) && !dt_object_remote(dt)));
2452 LASSERT(osd_invariant(obj));
2457 * Called to destroy on-disk representation of the object
2459 * Concurrency: must be locked
2461 static int osd_declare_object_destroy(const struct lu_env *env,
2462 struct dt_object *dt,
2465 struct osd_object *obj = osd_dt_obj(dt);
2466 struct inode *inode = obj->oo_inode;
2467 struct osd_thandle *oh;
2474 oh = container_of0(th, struct osd_thandle, ot_super);
2475 LASSERT(oh->ot_handle == NULL);
2477 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2478 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2479 /* Recycle idle OI leaf may cause additional three OI blocks
2481 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2482 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2483 /* one less inode */
2484 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2485 -1, oh, obj, false, NULL, false);
2488 /* data to be truncated */
2489 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2490 0, oh, obj, true, NULL, false);
2494 static int osd_object_destroy(const struct lu_env *env,
2495 struct dt_object *dt,
2498 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2499 struct osd_object *obj = osd_dt_obj(dt);
2500 struct inode *inode = obj->oo_inode;
2501 struct osd_device *osd = osd_obj2dev(obj);
2502 struct osd_thandle *oh;
2506 oh = container_of0(th, struct osd_thandle, ot_super);
2507 LASSERT(oh->ot_handle);
2509 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2511 if (unlikely(fid_is_acct(fid)))
2514 if (S_ISDIR(inode->i_mode)) {
2515 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2516 inode->i_nlink == 2);
2517 /* it will check/delete the inode from remote parent,
2518 * how to optimize it? unlink performance impaction XXX */
2519 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2520 if (result != 0 && result != -ENOENT) {
2521 CERROR("%s: delete inode "DFID": rc = %d\n",
2522 osd_name(osd), PFID(fid), result);
2524 spin_lock(&obj->oo_guard);
2526 spin_unlock(&obj->oo_guard);
2527 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2530 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2532 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
2533 result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2536 osd_trans_exec_check(env, th, OSD_OT_DESTROY);
2537 /* XXX: add to ext3 orphan list */
2538 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2540 /* not needed in the cache anymore */
2541 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2542 obj->oo_destroyed = 1;
2548 * Put the fid into lustre_mdt_attrs, and then place the structure
2549 * inode's ea. This fid should not be altered during the life time
2552 * \retval +ve, on success
2553 * \retval -ve, on error
2555 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2557 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2558 const struct lu_fid *fid, __u32 compat, __u32 incompat)
2560 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2564 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2567 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
2570 lustre_lma_init(lma, fid, compat, incompat);
2571 lustre_lma_swab(lma);
2573 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2575 /* LMA may already exist, but we need to check that all the
2576 * desired compat/incompat flags have been added. */
2577 if (unlikely(rc == -EEXIST)) {
2578 if (compat == 0 && incompat == 0)
2581 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2582 XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2587 lustre_lma_swab(lma);
2588 if (!(~lma->lma_compat & compat) &&
2589 !(~lma->lma_incompat & incompat))
2592 lma->lma_compat |= compat;
2593 lma->lma_incompat |= incompat;
2594 lustre_lma_swab(lma);
2595 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2596 sizeof(*lma), XATTR_REPLACE);
2603 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2604 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2605 * To have compatilibility with 1.8 ldiskfs driver we need to have
2606 * magic number at start of fid data.
2607 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2610 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2611 const struct lu_fid *fid)
2613 if (!fid_is_namespace_visible(fid) ||
2614 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2615 param->edp_magic = 0;
2619 param->edp_magic = LDISKFS_LUFID_MAGIC;
2620 param->edp_len = sizeof(struct lu_fid) + 1;
2621 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2625 * Try to read the fid from inode ea into dt_rec.
2627 * \param fid object fid.
2629 * \retval 0 on success
2631 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2632 __u32 ino, struct lu_fid *fid,
2633 struct osd_inode_id *id)
2635 struct osd_thread_info *info = osd_oti_get(env);
2636 struct inode *inode;
2639 osd_id_gen(id, ino, OSD_OII_NOGEN);
2640 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2642 RETURN(PTR_ERR(inode));
2648 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2650 struct inode *parent_dir,
2651 const struct lu_fid *dot_fid,
2652 const struct lu_fid *dot_dot_fid,
2653 struct osd_thandle *oth)
2655 struct ldiskfs_dentry_param *dot_ldp;
2656 struct ldiskfs_dentry_param *dot_dot_ldp;
2658 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2659 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2661 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2662 dot_ldp->edp_magic = 0;
2664 return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2665 dir, dot_ldp, dot_dot_ldp);
2669 * Create an local agent inode for remote entry
2671 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2672 struct osd_device *osd,
2673 struct osd_object *pobj,
2674 const struct lu_fid *fid,
2678 struct osd_thread_info *info = osd_oti_get(env);
2679 struct inode *local;
2680 struct osd_thandle *oh;
2685 oh = container_of(th, struct osd_thandle, ot_super);
2686 LASSERT(oh->ot_handle->h_transaction != NULL);
2688 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2689 if (IS_ERR(local)) {
2690 CERROR("%s: create local error %d\n", osd_name(osd),
2691 (int)PTR_ERR(local));
2695 ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
2696 unlock_new_inode(local);
2698 /* Set special LMA flag for local agent inode */
2699 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2701 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2702 osd_name(osd), PFID(fid), rc);
2703 RETURN(ERR_PTR(rc));
2709 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2710 lu_object_fid(&pobj->oo_dt.do_lu),
2713 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2714 osd_name(osd), PFID(fid), rc);
2715 RETURN(ERR_PTR(rc));
2722 * when direntry is deleted, we have to take care of possible agent inode
2723 * referenced by that. unfortunately we can't do this at that point:
2724 * iget() within a running transaction leads to deadlock and we better do
2725 * not call that every delete declaration to save performance. so we put
2726 * a potention agent inode on a list and process that once the transaction
2727 * is over. Notice it's not any worse in terms of real orphans as regular
2728 * object destroy doesn't put inodes on the on-disk orphan list. this should
2729 * be addressed separately
2731 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
2732 struct osd_thandle *oh,
2735 struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
2736 struct osd_obj_orphan *oor;
2743 oor->oor_env = (struct lu_env *)env;
2744 spin_lock(&osd->od_osfs_lock);
2745 list_add(&oor->oor_list, &osd->od_orphan_list);
2746 spin_unlock(&osd->od_osfs_lock);
2748 oh->ot_remove_agents = 1;
2754 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
2755 struct osd_device *osd)
2757 struct osd_thread_info *info = osd_oti_get(env);
2758 struct osd_obj_orphan *oor, *tmp;
2759 struct osd_inode_id id;
2760 struct list_head list;
2761 struct inode *inode;
2766 INIT_LIST_HEAD(&list);
2768 spin_lock(&osd->od_osfs_lock);
2769 list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
2770 if (oor->oor_env == env) {
2771 list_del(&oor->oor_list);
2772 list_add(&oor->oor_list, &list);
2775 spin_unlock(&osd->od_osfs_lock);
2777 list_for_each_entry_safe(oor, tmp, &list, oor_list) {
2781 list_del(&oor->oor_list);
2784 osd_id_gen(&id, ino, OSD_OII_NOGEN);
2785 inode = osd_iget_fid(info, osd, &id, &fid);
2789 if (!osd_remote_fid(env, osd, &fid)) {
2794 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
2796 mark_inode_dirty(inode);
2797 ldiskfs_journal_stop(jh);
2805 * OSD layer object create function for interoperability mode (b11826).
2806 * This is mostly similar to osd_object_create(). Only difference being, fid is
2807 * inserted into inode ea here.
2809 * \retval 0, on success
2810 * \retval -ve, on error
2812 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2813 struct lu_attr *attr,
2814 struct dt_allocation_hint *hint,
2815 struct dt_object_format *dof,
2818 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2819 struct osd_object *obj = osd_dt_obj(dt);
2820 struct osd_thread_info *info = osd_oti_get(env);
2825 if (dt_object_exists(dt))
2828 LASSERT(osd_invariant(obj));
2829 LASSERT(!dt_object_remote(dt));
2830 LASSERT(osd_write_locked(env, obj));
2831 LASSERT(th != NULL);
2833 if (unlikely(fid_is_acct(fid)))
2834 /* Quota files can't be created from the kernel any more,
2835 * 'tune2fs -O quota' will take care of creating them */
2838 result = __osd_object_create(info, obj, attr, hint, dof, th);
2840 if (fid_is_idif(fid) &&
2841 !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
2842 struct lu_fid *tfid = &info->oti_fid;
2843 struct ost_id *oi = &info->oti_ostid;
2845 fid_to_ostid(fid, oi);
2846 ostid_to_fid(tfid, oi, 0);
2847 result = osd_ea_fid_set(info, obj->oo_inode, tfid,
2848 LMAC_FID_ON_OST, 0);
2850 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2851 fid_is_on_ost(info, osd_obj2dev(obj),
2852 fid, OI_CHECK_FLD) ?
2853 LMAC_FID_ON_OST : 0, 0);
2855 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
2856 obj->oo_dt.do_body_ops = &osd_body_ops;
2860 result = __osd_oi_insert(env, obj, fid, th);
2862 LASSERT(ergo(result == 0,
2863 dt_object_exists(dt) && !dt_object_remote(dt)));
2864 LINVRNT(osd_invariant(obj));
2868 static int osd_declare_object_ref_add(const struct lu_env *env,
2869 struct dt_object *dt,
2870 struct thandle *handle)
2872 struct osd_thandle *oh;
2874 /* it's possible that object doesn't exist yet */
2875 LASSERT(handle != NULL);
2877 oh = container_of0(handle, struct osd_thandle, ot_super);
2878 LASSERT(oh->ot_handle == NULL);
2880 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2881 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2887 * Concurrency: @dt is write locked.
2889 static int osd_object_ref_add(const struct lu_env *env,
2890 struct dt_object *dt, struct thandle *th)
2892 struct osd_object *obj = osd_dt_obj(dt);
2893 struct inode *inode = obj->oo_inode;
2894 struct osd_thandle *oh;
2897 if (!dt_object_exists(dt))
2900 LINVRNT(osd_invariant(obj));
2901 LASSERT(!dt_object_remote(dt));
2902 LASSERT(osd_write_locked(env, obj));
2903 LASSERT(th != NULL);
2905 oh = container_of0(th, struct osd_thandle, ot_super);
2906 LASSERT(oh->ot_handle != NULL);
2908 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2910 CDEBUG(D_INODE, DFID" increase nlink %d\n",
2911 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2913 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2914 * (65000) subdirectories by storing "1" in i_nlink if the link count
2915 * would otherwise overflow. Directory tranversal tools understand
2916 * that (st_nlink == 1) indicates that the filesystem dose not track
2917 * hard links count on the directory, and will not abort subdirectory
2918 * scanning early once (st_nlink - 2) subdirs have been found.
2920 * This also has to properly handle the case of inodes with nlink == 0
2921 * in case they are being linked into the PENDING directory
2923 spin_lock(&obj->oo_guard);
2924 if (unlikely(inode->i_nlink == 0))
2925 /* inc_nlink from 0 may cause WARN_ON */
2926 set_nlink(inode, 1);
2928 ldiskfs_inc_count(oh->ot_handle, inode);
2929 if (!S_ISDIR(inode->i_mode))
2930 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2932 spin_unlock(&obj->oo_guard);
2934 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2935 LINVRNT(osd_invariant(obj));
2937 osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
2942 static int osd_declare_object_ref_del(const struct lu_env *env,
2943 struct dt_object *dt,
2944 struct thandle *handle)
2946 struct osd_thandle *oh;
2948 if (!dt_object_exists(dt))
2951 LASSERT(!dt_object_remote(dt));
2952 LASSERT(handle != NULL);
2954 oh = container_of0(handle, struct osd_thandle, ot_super);
2955 LASSERT(oh->ot_handle == NULL);
2957 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2958 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2964 * Concurrency: @dt is write locked.
2966 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2969 struct osd_object *obj = osd_dt_obj(dt);
2970 struct inode *inode = obj->oo_inode;
2971 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
2972 struct osd_thandle *oh;
2974 if (!dt_object_exists(dt))
2977 LINVRNT(osd_invariant(obj));
2978 LASSERT(!dt_object_remote(dt));
2979 LASSERT(osd_write_locked(env, obj));
2980 LASSERT(th != NULL);
2982 oh = container_of0(th, struct osd_thandle, ot_super);
2983 LASSERT(oh->ot_handle != NULL);
2985 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2987 spin_lock(&obj->oo_guard);
2988 /* That can be result of upgrade from old Lustre version and
2989 * applied only to local files. Just skip this ref_del call.
2990 * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2991 if (inode->i_nlink == 0) {
2992 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2993 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2994 ", maybe an upgraded file? (LU-3915)\n",
2995 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2996 spin_unlock(&obj->oo_guard);
3000 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
3001 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3003 ldiskfs_dec_count(oh->ot_handle, inode);
3004 spin_unlock(&obj->oo_guard);
3006 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3007 LINVRNT(osd_invariant(obj));
3009 osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
3015 * Get the 64-bit version for an inode.
3017 static int osd_object_version_get(const struct lu_env *env,
3018 struct dt_object *dt, dt_obj_version_t *ver)
3020 struct inode *inode = osd_dt_obj(dt)->oo_inode;
3022 CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
3023 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3024 *ver = LDISKFS_I(inode)->i_fs_version;
3029 * Concurrency: @dt is read locked.
3031 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
3032 struct lu_buf *buf, const char *name)
3034 struct osd_object *obj = osd_dt_obj(dt);
3035 struct inode *inode = obj->oo_inode;
3036 struct osd_thread_info *info = osd_oti_get(env);
3037 struct dentry *dentry = &info->oti_obj_dentry;
3039 /* version get is not real XATTR but uses xattr API */
3040 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3041 /* for version we are just using xattr API but change inode
3043 if (buf->lb_len == 0)
3044 return sizeof(dt_obj_version_t);
3046 if (buf->lb_len < sizeof(dt_obj_version_t))
3049 osd_object_version_get(env, dt, buf->lb_buf);
3051 return sizeof(dt_obj_version_t);
3054 if (!dt_object_exists(dt))
3057 LASSERT(!dt_object_remote(dt));
3058 LASSERT(inode->i_op != NULL);
3059 LASSERT(inode->i_op->getxattr != NULL);
3061 return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
3065 static int osd_declare_xattr_set(const struct lu_env *env,
3066 struct dt_object *dt,
3067 const struct lu_buf *buf, const char *name,
3068 int fl, struct thandle *handle)
3070 struct osd_thandle *oh;
3072 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3074 LASSERT(handle != NULL);
3076 oh = container_of0(handle, struct osd_thandle, ot_super);
3077 LASSERT(oh->ot_handle == NULL);
3079 if (strcmp(name, XATTR_NAME_LMA) == 0) {
3080 /* For non-upgrading case, the LMA is set first and
3081 * usually fit inode. But for upgrade case, the LMA
3082 * may be in another separated EA block. */
3083 if (!dt_object_exists(dt))
3085 else if (fl == LU_XATTR_REPLACE)
3089 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3093 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
3098 if (buf->lb_buf == NULL && dt_object_exists(dt)) {
3099 /* learn xattr size from osd_xattr_get if
3100 attribute has not been read yet */
3101 buflen = __osd_xattr_get(
3102 osd_dt_obj(dt)->oo_inode,
3103 &osd_oti_get(env)->oti_obj_dentry,
3108 buflen = buf->lb_len;
3111 if (buflen > sb->s_blocksize) {
3112 credits += osd_calc_bkmap_credits(
3114 (buflen + sb->s_blocksize - 1) >>
3115 sb->s_blocksize_bits);
3119 * xattr set may involve inode quota change, reserve credits for
3120 * dquot_initialize()
3122 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3125 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3131 * Set the 64-bit version for object
3133 static void osd_object_version_set(const struct lu_env *env,
3134 struct dt_object *dt,
3135 dt_obj_version_t *new_version)
3137 struct inode *inode = osd_dt_obj(dt)->oo_inode;
3139 CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
3140 *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3142 LDISKFS_I(inode)->i_fs_version = *new_version;
3143 /** Version is set after all inode operations are finished,
3144 * so we should mark it dirty here */
3145 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3149 * Concurrency: @dt is write locked.
3151 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3152 const struct lu_buf *buf, const char *name, int fl,
3153 struct thandle *handle)
3155 struct osd_object *obj = osd_dt_obj(dt);
3156 struct inode *inode = obj->oo_inode;
3157 struct osd_thread_info *info = osd_oti_get(env);
3162 LASSERT(handle != NULL);
3164 /* version set is not real XATTR */
3165 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3166 /* for version we are just using xattr API but change inode
3168 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3169 osd_object_version_set(env, dt, buf->lb_buf);
3170 return sizeof(dt_obj_version_t);
3173 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
3174 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3176 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3177 if (fl & LU_XATTR_REPLACE)
3178 fs_flags |= XATTR_REPLACE;
3180 if (fl & LU_XATTR_CREATE)
3181 fs_flags |= XATTR_CREATE;
3183 if (strcmp(name, XATTR_NAME_LMV) == 0) {
3184 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3187 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3191 lma->lma_incompat |= LMAI_STRIPED;
3192 lustre_lma_swab(lma);
3193 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3194 sizeof(*lma), XATTR_REPLACE);
3199 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
3200 strcmp(name, XATTR_NAME_LINK) == 0)
3203 rc = __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3205 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3211 * Concurrency: @dt is read locked.
3213 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3214 const struct lu_buf *buf)
3216 struct osd_object *obj = osd_dt_obj(dt);
3217 struct inode *inode = obj->oo_inode;
3218 struct osd_thread_info *info = osd_oti_get(env);
3219 struct dentry *dentry = &info->oti_obj_dentry;
3221 if (!dt_object_exists(dt))
3224 LASSERT(!dt_object_remote(dt));
3225 LASSERT(inode->i_op != NULL);
3226 LASSERT(inode->i_op->listxattr != NULL);
3228 dentry->d_inode = inode;
3229 dentry->d_sb = inode->i_sb;
3230 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3233 static int osd_declare_xattr_del(const struct lu_env *env,
3234 struct dt_object *dt, const char *name,
3235 struct thandle *handle)
3237 struct osd_thandle *oh;
3238 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3240 LASSERT(!dt_object_remote(dt));
3241 LASSERT(handle != NULL);
3243 oh = container_of0(handle, struct osd_thandle, ot_super);
3244 LASSERT(oh->ot_handle == NULL);
3246 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3247 osd_dto_credits_noquota[DTO_XATTR_SET]);
3249 * xattr del may involve inode quota change, reserve credits for
3250 * dquot_initialize()
3252 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3258 * Concurrency: @dt is write locked.
3260 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3261 const char *name, struct thandle *handle)
3263 struct osd_object *obj = osd_dt_obj(dt);
3264 struct inode *inode = obj->oo_inode;
3265 struct osd_thread_info *info = osd_oti_get(env);
3266 struct dentry *dentry = &info->oti_obj_dentry;
3269 if (!dt_object_exists(dt))
3272 LASSERT(!dt_object_remote(dt));
3273 LASSERT(inode->i_op != NULL);
3274 LASSERT(inode->i_op->removexattr != NULL);
3275 LASSERT(handle != NULL);
3277 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3279 ll_vfs_dq_init(inode);
3280 dentry->d_inode = inode;
3281 dentry->d_sb = inode->i_sb;
3282 rc = inode->i_op->removexattr(dentry, name);
3283 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3287 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3288 __u64 start, __u64 end)
3290 struct osd_object *obj = osd_dt_obj(dt);
3291 struct inode *inode = obj->oo_inode;
3292 struct osd_thread_info *info = osd_oti_get(env);
3293 struct dentry *dentry = &info->oti_obj_dentry;
3294 struct file *file = &info->oti_file;
3299 dentry->d_inode = inode;
3300 dentry->d_sb = inode->i_sb;
3301 file->f_path.dentry = dentry;
3302 file->f_mapping = inode->i_mapping;
3303 file->f_op = inode->i_fop;
3304 set_file_inode(file, inode);
3306 rc = ll_vfs_fsync_range(file, start, end, 0);
3315 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3316 const struct dt_index_features *feat)
3318 struct iam_descr *descr;
3320 if (osd_object_is_root(o))
3321 return feat == &dt_directory_features;
3323 LASSERT(o->oo_dir != NULL);
3325 descr = o->oo_dir->od_container.ic_descr;
3326 if (feat == &dt_directory_features) {
3327 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3333 feat->dif_keysize_min <= descr->id_key_size &&
3334 descr->id_key_size <= feat->dif_keysize_max &&
3335 feat->dif_recsize_min <= descr->id_rec_size &&
3336 descr->id_rec_size <= feat->dif_recsize_max &&
3337 !(feat->dif_flags & (DT_IND_VARKEY |
3338 DT_IND_VARREC | DT_IND_NONUNQ)) &&
3339 ergo(feat->dif_flags & DT_IND_UPDATE,
3340 1 /* XXX check that object (and file system) is
3345 static int osd_iam_container_init(const struct lu_env *env,
3346 struct osd_object *obj,
3347 struct osd_directory *dir)
3349 struct iam_container *bag = &dir->od_container;
3352 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3356 result = iam_container_setup(bag);
3358 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3360 iam_container_fini(bag);
3367 * Concurrency: no external locking is necessary.
3369 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3370 const struct dt_index_features *feat)
3374 struct osd_object *obj = osd_dt_obj(dt);
3376 LINVRNT(osd_invariant(obj));
3378 if (osd_object_is_root(obj)) {
3379 dt->do_index_ops = &osd_index_ea_ops;
3381 } else if (feat == &dt_directory_features) {
3382 dt->do_index_ops = &osd_index_ea_ops;
3383 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3388 } else if (unlikely(feat == &dt_otable_features)) {
3389 dt->do_index_ops = &osd_otable_ops;
3391 } else if (unlikely(feat == &dt_acct_features)) {
3392 dt->do_index_ops = &osd_acct_index_ops;
3395 } else if (!osd_has_index(obj)) {
3396 struct osd_directory *dir;
3401 spin_lock(&obj->oo_guard);
3402 if (obj->oo_dir == NULL)
3406 * Concurrent thread allocated container data.
3409 spin_unlock(&obj->oo_guard);
3411 * Now, that we have container data, serialize its
3414 down_write(&obj->oo_ext_idx_sem);
3416 * recheck under lock.
3418 if (!osd_has_index(obj))
3419 result = osd_iam_container_init(env, obj,
3423 up_write(&obj->oo_ext_idx_sem);
3431 if (result == 0 && skip_iam == 0) {
3432 if (!osd_iam_index_probe(env, obj, feat))
3435 LINVRNT(osd_invariant(obj));
3437 if (result == 0 && feat == &dt_quota_glb_features &&
3438 fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3439 result = osd_quota_migration(env, dt);
3444 static int osd_otable_it_attr_get(const struct lu_env *env,
3445 struct dt_object *dt,
3446 struct lu_attr *attr)
3452 static const struct dt_object_operations osd_obj_ops = {
3453 .do_read_lock = osd_object_read_lock,
3454 .do_write_lock = osd_object_write_lock,
3455 .do_read_unlock = osd_object_read_unlock,
3456 .do_write_unlock = osd_object_write_unlock,
3457 .do_write_locked = osd_object_write_locked,
3458 .do_attr_get = osd_attr_get,
3459 .do_declare_attr_set = osd_declare_attr_set,
3460 .do_attr_set = osd_attr_set,
3461 .do_ah_init = osd_ah_init,
3462 .do_declare_create = osd_declare_object_create,
3463 .do_create = osd_object_create,
3464 .do_declare_destroy = osd_declare_object_destroy,
3465 .do_destroy = osd_object_destroy,
3466 .do_index_try = osd_index_try,
3467 .do_declare_ref_add = osd_declare_object_ref_add,
3468 .do_ref_add = osd_object_ref_add,
3469 .do_declare_ref_del = osd_declare_object_ref_del,
3470 .do_ref_del = osd_object_ref_del,
3471 .do_xattr_get = osd_xattr_get,
3472 .do_declare_xattr_set = osd_declare_xattr_set,
3473 .do_xattr_set = osd_xattr_set,
3474 .do_declare_xattr_del = osd_declare_xattr_del,
3475 .do_xattr_del = osd_xattr_del,
3476 .do_xattr_list = osd_xattr_list,
3477 .do_object_sync = osd_object_sync,
3481 * dt_object_operations for interoperability mode
3482 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3484 static const struct dt_object_operations osd_obj_ea_ops = {
3485 .do_read_lock = osd_object_read_lock,
3486 .do_write_lock = osd_object_write_lock,
3487 .do_read_unlock = osd_object_read_unlock,
3488 .do_write_unlock = osd_object_write_unlock,
3489 .do_write_locked = osd_object_write_locked,
3490 .do_attr_get = osd_attr_get,
3491 .do_declare_attr_set = osd_declare_attr_set,
3492 .do_attr_set = osd_attr_set,
3493 .do_ah_init = osd_ah_init,
3494 .do_declare_create = osd_declare_object_create,
3495 .do_create = osd_object_ea_create,
3496 .do_declare_destroy = osd_declare_object_destroy,
3497 .do_destroy = osd_object_destroy,
3498 .do_index_try = osd_index_try,
3499 .do_declare_ref_add = osd_declare_object_ref_add,
3500 .do_ref_add = osd_object_ref_add,
3501 .do_declare_ref_del = osd_declare_object_ref_del,
3502 .do_ref_del = osd_object_ref_del,
3503 .do_xattr_get = osd_xattr_get,
3504 .do_declare_xattr_set = osd_declare_xattr_set,
3505 .do_xattr_set = osd_xattr_set,
3506 .do_declare_xattr_del = osd_declare_xattr_del,
3507 .do_xattr_del = osd_xattr_del,
3508 .do_xattr_list = osd_xattr_list,
3509 .do_object_sync = osd_object_sync,
3512 static const struct dt_object_operations osd_obj_otable_it_ops = {
3513 .do_attr_get = osd_otable_it_attr_get,
3514 .do_index_try = osd_index_try,
3517 static int osd_index_declare_iam_delete(const struct lu_env *env,
3518 struct dt_object *dt,
3519 const struct dt_key *key,
3520 struct thandle *handle)
3522 struct osd_thandle *oh;
3524 oh = container_of0(handle, struct osd_thandle, ot_super);
3525 LASSERT(oh->ot_handle == NULL);
3527 /* Recycle may cause additional three blocks to be changed. */
3528 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3529 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3535 * delete a (key, value) pair from index \a dt specified by \a key
3537 * \param dt osd index object
3538 * \param key key for index
3539 * \param rec record reference
3540 * \param handle transaction handler
3543 * \retval -ve failure
3545 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3546 const struct dt_key *key,
3547 struct thandle *handle)
3549 struct osd_thread_info *oti = osd_oti_get(env);
3550 struct osd_object *obj = osd_dt_obj(dt);
3551 struct osd_thandle *oh;
3552 struct iam_path_descr *ipd;
3553 struct iam_container *bag = &obj->oo_dir->od_container;
3557 if (!dt_object_exists(dt))
3560 LINVRNT(osd_invariant(obj));
3561 LASSERT(!dt_object_remote(dt));
3562 LASSERT(bag->ic_object == obj->oo_inode);
3563 LASSERT(handle != NULL);
3565 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3567 ipd = osd_idx_ipd_get(env, bag);
3568 if (unlikely(ipd == NULL))
3571 oh = container_of0(handle, struct osd_thandle, ot_super);
3572 LASSERT(oh->ot_handle != NULL);
3573 LASSERT(oh->ot_handle->h_transaction != NULL);
3575 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3576 /* swab quota uid/gid provided by caller */
3577 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3578 key = (const struct dt_key *)&oti->oti_quota_id;
3581 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3582 osd_ipd_put(env, bag, ipd);
3583 LINVRNT(osd_invariant(obj));
3584 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
3588 static int osd_index_declare_ea_delete(const struct lu_env *env,
3589 struct dt_object *dt,
3590 const struct dt_key *key,
3591 struct thandle *handle)
3593 struct osd_thandle *oh;
3594 struct inode *inode;
3598 LASSERT(!dt_object_remote(dt));
3599 LASSERT(handle != NULL);
3601 oh = container_of0(handle, struct osd_thandle, ot_super);
3602 LASSERT(oh->ot_handle == NULL);
3604 /* due to DNE we may need to remove an agent inode */
3605 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3606 osd_dto_credits_noquota[DTO_INDEX_DELETE] +
3607 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3609 inode = osd_dt_obj(dt)->oo_inode;
3613 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3614 0, oh, osd_dt_obj(dt), true, NULL, false);
3618 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3621 struct osd_fid_pack *rec;
3624 if (de->file_type & LDISKFS_DIRENT_LUFID) {
3625 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3626 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3627 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
3633 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3634 const struct lu_fid *fid)
3636 struct seq_server_site *ss = osd_seq_site(osd);
3639 /* FID seqs not in FLDB, must be local seq */
3640 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3643 /* If FLD is not being initialized yet, it only happens during the
3644 * initialization, likely during mgs initialization, and we assume
3645 * this is local FID. */
3646 if (ss == NULL || ss->ss_server_fld == NULL)
3649 /* Only check the local FLDB here */
3650 if (osd_seq_exists(env, osd, fid_seq(fid)))
3657 * Index delete function for interoperability mode (b11826).
3658 * It will remove the directory entry added by osd_index_ea_insert().
3659 * This entry is needed to maintain name->fid mapping.
3661 * \param key, key i.e. file entry to be deleted
3663 * \retval 0, on success
3664 * \retval -ve, on error
3666 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3667 const struct dt_key *key, struct thandle *handle)
3669 struct osd_object *obj = osd_dt_obj(dt);
3670 struct inode *dir = obj->oo_inode;
3671 struct dentry *dentry;
3672 struct osd_thandle *oh;
3673 struct ldiskfs_dir_entry_2 *de = NULL;
3674 struct buffer_head *bh;
3675 struct htree_lock *hlock = NULL;
3676 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
3677 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3681 if (!dt_object_exists(dt))
3684 LINVRNT(osd_invariant(obj));
3685 LASSERT(!dt_object_remote(dt));
3686 LASSERT(handle != NULL);
3688 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3690 oh = container_of(handle, struct osd_thandle, ot_super);
3691 LASSERT(oh->ot_handle != NULL);
3692 LASSERT(oh->ot_handle->h_transaction != NULL);
3694 ll_vfs_dq_init(dir);
3695 dentry = osd_child_dentry_get(env, obj,
3696 (char *)key, strlen((char *)key));
3698 if (obj->oo_hl_head != NULL) {
3699 hlock = osd_oti_get(env)->oti_hlock;
3700 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3701 dir, LDISKFS_HLOCK_DEL);
3703 down_write(&obj->oo_ext_idx_sem);
3706 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3708 /* If this is not the ".." entry, it might be a remote DNE
3709 * entry and we need to check if the FID is for a remote
3710 * MDT. If the FID is not in the directory entry (e.g.
3711 * upgraded 1.8 filesystem without dirdata enabled) then
3712 * we need to get the FID from the LMA. For a remote directory
3713 * there HAS to be an LMA, it cannot be an IGIF inode in this
3716 * Delete the entry before the agent inode in order to
3717 * simplify error handling. At worst an error after deleting
3718 * the entry first might leak the agent inode afterward. The
3719 * reverse would need filesystem abort in case of error deleting
3720 * the entry after the agent had been removed, or leave a
3721 * dangling entry pointing at a random inode. */
3722 if (strcmp((char *)key, dotdot) != 0) {
3723 LASSERT(de != NULL);
3724 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3725 if (rc == -ENODATA) {
3726 /* can't get FID, postpone to the end of the
3727 * transaction when iget() is safe */
3728 osd_schedule_agent_inode_removal(env, oh,
3729 le32_to_cpu(de->inode));
3730 } else if (rc == 0 &&
3731 unlikely(osd_remote_fid(env, osd, fid))) {
3732 osd_schedule_agent_inode_removal(env, oh,
3733 le32_to_cpu(de->inode));
3736 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3742 ldiskfs_htree_unlock(hlock);
3744 up_write(&obj->oo_ext_idx_sem);
3749 /* For inode on the remote MDT, .. will point to
3750 * /Agent directory, Check whether it needs to delete
3751 * from agent directory */
3752 if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3753 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3755 if (rc != 0 && rc != -ENOENT) {
3756 CERROR("%s: delete agent inode "DFID": rc = %d\n",
3757 osd_name(osd), PFID(fid), rc);
3767 LASSERT(osd_invariant(obj));
3768 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
3773 * Lookup index for \a key and copy record to \a rec.
3775 * \param dt osd index object
3776 * \param key key for index
3777 * \param rec record reference
3779 * \retval +ve success : exact mach
3780 * \retval 0 return record with key not greater than \a key
3781 * \retval -ve failure
3783 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3784 struct dt_rec *rec, const struct dt_key *key)
3786 struct osd_object *obj = osd_dt_obj(dt);
3787 struct iam_path_descr *ipd;
3788 struct iam_container *bag = &obj->oo_dir->od_container;
3789 struct osd_thread_info *oti = osd_oti_get(env);
3790 struct iam_iterator *it = &oti->oti_idx_it;
3791 struct iam_rec *iam_rec;
3795 if (!dt_object_exists(dt))
3798 LASSERT(osd_invariant(obj));
3799 LASSERT(!dt_object_remote(dt));
3800 LASSERT(bag->ic_object == obj->oo_inode);
3802 ipd = osd_idx_ipd_get(env, bag);
3806 /* got ipd now we can start iterator. */
3807 iam_it_init(it, bag, 0, ipd);
3809 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3810 /* swab quota uid/gid provided by caller */
3811 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3812 key = (const struct dt_key *)&oti->oti_quota_id;
3815 rc = iam_it_get(it, (struct iam_key *)key);
3817 if (S_ISDIR(obj->oo_inode->i_mode))
3818 iam_rec = (struct iam_rec *)oti->oti_ldp;
3820 iam_rec = (struct iam_rec *) rec;
3822 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3824 if (S_ISDIR(obj->oo_inode->i_mode))
3825 osd_fid_unpack((struct lu_fid *) rec,
3826 (struct osd_fid_pack *)iam_rec);
3827 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3828 osd_quota_unpack(obj, rec);
3833 osd_ipd_put(env, bag, ipd);
3835 LINVRNT(osd_invariant(obj));
3840 static int osd_index_declare_iam_insert(const struct lu_env *env,
3841 struct dt_object *dt,
3842 const struct dt_rec *rec,
3843 const struct dt_key *key,
3844 struct thandle *handle)
3846 struct osd_thandle *oh;
3848 LASSERT(handle != NULL);
3850 oh = container_of0(handle, struct osd_thandle, ot_super);
3851 LASSERT(oh->ot_handle == NULL);
3853 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3854 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3860 * Inserts (key, value) pair in \a dt index object.
3862 * \param dt osd index object
3863 * \param key key for index
3864 * \param rec record reference
3865 * \param th transaction handler
3868 * \retval -ve failure
3870 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3871 const struct dt_rec *rec,
3872 const struct dt_key *key, struct thandle *th,
3875 struct osd_object *obj = osd_dt_obj(dt);
3876 struct iam_path_descr *ipd;
3877 struct osd_thandle *oh;
3878 struct iam_container *bag;
3879 struct osd_thread_info *oti = osd_oti_get(env);
3880 struct iam_rec *iam_rec;
3884 if (!dt_object_exists(dt))
3887 LINVRNT(osd_invariant(obj));
3888 LASSERT(!dt_object_remote(dt));
3890 bag = &obj->oo_dir->od_container;
3891 LASSERT(bag->ic_object == obj->oo_inode);
3892 LASSERT(th != NULL);
3894 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3896 ipd = osd_idx_ipd_get(env, bag);
3897 if (unlikely(ipd == NULL))
3900 oh = container_of0(th, struct osd_thandle, ot_super);
3901 LASSERT(oh->ot_handle != NULL);
3902 LASSERT(oh->ot_handle->h_transaction != NULL);
3903 if (S_ISDIR(obj->oo_inode->i_mode)) {
3904 iam_rec = (struct iam_rec *)oti->oti_ldp;
3905 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3906 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3907 /* pack quota uid/gid */
3908 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3909 key = (const struct dt_key *)&oti->oti_quota_id;
3910 /* pack quota record */
3911 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3912 iam_rec = (struct iam_rec *)rec;
3914 iam_rec = (struct iam_rec *)rec;
3917 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3919 osd_ipd_put(env, bag, ipd);
3920 LINVRNT(osd_invariant(obj));
3921 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3926 * Calls ldiskfs_add_entry() to add directory entry
3927 * into the directory. This is required for
3928 * interoperability mode (b11826)
3930 * \retval 0, on success
3931 * \retval -ve, on error
3933 static int __osd_ea_add_rec(struct osd_thread_info *info,
3934 struct osd_object *pobj, struct inode *cinode,
3935 const char *name, const struct lu_fid *fid,
3936 struct htree_lock *hlock, struct thandle *th)
3938 struct ldiskfs_dentry_param *ldp;
3939 struct dentry *child;
3940 struct osd_thandle *oth;
3943 oth = container_of(th, struct osd_thandle, ot_super);
3944 LASSERT(oth->ot_handle != NULL);
3945 LASSERT(oth->ot_handle->h_transaction != NULL);
3946 LASSERT(pobj->oo_inode);
3948 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3949 if (unlikely(pobj->oo_inode ==
3950 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3953 osd_get_ldiskfs_dirent_param(ldp, fid);
3954 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3955 child->d_fsdata = (void *)ldp;
3956 ll_vfs_dq_init(pobj->oo_inode);
3957 rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3958 if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
3959 struct ldiskfs_dir_entry_2 *de;
3960 struct buffer_head *bh;
3963 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
3966 rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
3969 if (S_ISDIR(cinode->i_mode))
3970 de->file_type = LDISKFS_DIRENT_LUFID |
3971 LDISKFS_FT_REG_FILE;
3973 de->file_type = LDISKFS_DIRENT_LUFID |
3975 ldiskfs_handle_dirty_metadata(oth->ot_handle,
3986 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3987 * into the directory.Also sets flags into osd object to
3988 * indicate dot and dotdot are created. This is required for
3989 * interoperability mode (b11826)
3991 * \param dir directory for dot and dotdot fixup.
3992 * \param obj child object for linking
3994 * \retval 0, on success
3995 * \retval -ve, on error
3997 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3998 struct osd_object *dir,
3999 struct inode *parent_dir, const char *name,
4000 const struct lu_fid *dot_fid,
4001 const struct lu_fid *dot_dot_fid,
4004 struct inode *inode = dir->oo_inode;
4005 struct osd_thandle *oth;
4008 oth = container_of(th, struct osd_thandle, ot_super);
4009 LASSERT(oth->ot_handle->h_transaction != NULL);
4010 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4012 if (strcmp(name, dot) == 0) {
4013 if (dir->oo_compat_dot_created) {
4016 LASSERT(inode == parent_dir);
4017 dir->oo_compat_dot_created = 1;
4020 } else if (strcmp(name, dotdot) == 0) {
4021 if (!dir->oo_compat_dot_created)
4023 /* in case of rename, dotdot is already created */
4024 if (dir->oo_compat_dotdot_created) {
4025 return __osd_ea_add_rec(info, dir, parent_dir, name,
4026 dot_dot_fid, NULL, th);
4029 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
4030 struct lu_fid tfid = *dot_dot_fid;
4033 result = osd_add_dot_dotdot_internal(info,
4034 dir->oo_inode, parent_dir, dot_fid,
4037 result = osd_add_dot_dotdot_internal(info,
4038 dir->oo_inode, parent_dir, dot_fid,
4043 dir->oo_compat_dotdot_created = 1;
4051 * It will call the appropriate osd_add* function and return the
4052 * value, return by respective functions.
4054 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4055 struct inode *cinode, const char *name,
4056 const struct lu_fid *fid, struct thandle *th)
4058 struct osd_thread_info *info = osd_oti_get(env);
4059 struct htree_lock *hlock;
4062 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4064 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4066 if (hlock != NULL) {
4067 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4070 down_write(&pobj->oo_ext_idx_sem);
4073 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4074 lu_object_fid(&pobj->oo_dt.do_lu),
4077 if (hlock != NULL) {
4078 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4079 pobj->oo_inode, LDISKFS_HLOCK_ADD);
4081 down_write(&pobj->oo_ext_idx_sem);
4084 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4085 struct lu_fid *tfid = &info->oti_fid;
4089 rc = __osd_ea_add_rec(info, pobj, cinode, name,
4092 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4097 ldiskfs_htree_unlock(hlock);
4099 up_write(&pobj->oo_ext_idx_sem);
4105 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4106 struct osd_idmap_cache *oic)
4108 struct osd_scrub *scrub = &dev->od_scrub;
4109 struct lu_fid *fid = &oic->oic_fid;
4110 struct osd_inode_id *id = &oti->oti_id;
4115 if (!fid_is_norm(fid) && !fid_is_igif(fid))
4118 if (scrub->os_pos_current > id->oii_ino)
4122 rc = osd_oi_lookup(oti, dev, fid, id, 0);
4123 if (rc == -ENOENT) {
4124 struct inode *inode;
4127 inode = osd_iget(oti, dev, &oic->oic_lid);
4129 /* The inode has been removed (by race maybe). */
4130 if (IS_ERR(inode)) {
4131 rc = PTR_ERR(inode);
4133 RETURN(rc == -ESTALE ? -ENOENT : rc);
4137 /* The OI mapping is lost. */
4138 if (id->oii_gen != OSD_OII_NOGEN)
4141 /* The inode may has been reused by others, we do not know,
4142 * leave it to be handled by subsequent osd_fid_lookup(). */
4144 } else if (rc != 0 || osd_id_eq(id, &oic->oic_lid)) {
4149 if (thread_is_running(&scrub->os_thread)) {
4150 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
4151 /* There is race condition between osd_oi_lookup and OI scrub.
4152 * The OI scrub finished just after osd_oi_lookup() failure.
4153 * Under such case, it is unnecessary to trigger OI scrub again,
4154 * but try to call osd_oi_lookup() again. */
4155 if (unlikely(rc == -EAGAIN))
4161 if (!dev->od_noscrub && ++once == 1) {
4162 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
4164 CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI scrub by RPC "
4165 "for "DFID", rc = %d [2]\n",
4166 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4168 if (rc == 0 || rc == -EALREADY)
4175 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4176 struct osd_device *dev,
4177 struct osd_idmap_cache *oic,
4178 struct lu_fid *fid, __u32 ino)
4180 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
4181 struct inode *inode;
4184 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4185 inode = osd_iget(oti, dev, &oic->oic_lid);
4186 if (IS_ERR(inode)) {
4187 fid_zero(&oic->oic_fid);
4188 return PTR_ERR(inode);
4191 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4194 fid_zero(&oic->oic_fid);
4196 *fid = oic->oic_fid = lma->lma_self_fid;
4200 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4201 struct osd_inode_id *id, const struct lu_fid *fid)
4203 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4204 id->oii_ino, id->oii_gen, info);
4205 info->oti_cache.oic_lid = *id;
4206 info->oti_cache.oic_fid = *fid;
4207 info->oti_cache.oic_dev = osd;
4211 * Get parent FID from the linkEA.
4213 * For a directory which parent resides on remote MDT, to satisfy the
4214 * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
4215 * the other hand, to make the lookup(..) on the directory can return
4216 * the real parent FID, we append the real parent FID after its ".."
4217 * name entry in the /REMOTE_PARENT_DIR.
4219 * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
4220 * backup. So after the restore, we cannot get the right parent FID from
4221 * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
4222 * we have stored the real parent FID in the directory object's linkEA,
4223 * we can parse the linkEA for the real parent FID.
4225 * \param[in] env pointer to the thread context
4226 * \param[in] obj pointer to the object to be handled
4227 * \param[out]fid pointer to the buffer to hold the parent FID
4229 * \retval 0 for getting the real parent FID successfully
4230 * \retval negative error number on failure
4232 static int osd_get_pfid_from_linkea(const struct lu_env *env,
4233 struct osd_object *obj,
4236 struct osd_thread_info *oti = osd_oti_get(env);
4237 struct lu_buf *buf = &oti->oti_big_buf;
4238 struct dentry *dentry = &oti->oti_obj_dentry;
4239 struct inode *inode = obj->oo_inode;
4240 struct linkea_data ldata = { NULL };
4245 if (!S_ISDIR(inode->i_mode))
4249 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4250 buf->lb_buf, buf->lb_len);
4251 if (rc == -ERANGE) {
4252 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4255 lu_buf_realloc(buf, rc);
4256 if (buf->lb_buf == NULL)
4263 if (unlikely(rc == 0))
4269 if (unlikely(buf->lb_buf == NULL)) {
4270 lu_buf_realloc(buf, rc);
4271 if (buf->lb_buf == NULL)
4278 rc = linkea_init(&ldata);
4280 linkea_first_entry(&ldata);
4281 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
4288 * Calls ->lookup() to find dentry. From dentry get inode and
4289 * read inode's ea to get fid. This is required for interoperability
4292 * \retval 0, on success
4293 * \retval -ve, on error
4295 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4296 struct dt_rec *rec, const struct dt_key *key)
4298 struct inode *dir = obj->oo_inode;
4299 struct dentry *dentry;
4300 struct ldiskfs_dir_entry_2 *de;
4301 struct buffer_head *bh;
4302 struct lu_fid *fid = (struct lu_fid *) rec;
4303 struct htree_lock *hlock = NULL;
4308 LASSERT(dir->i_op != NULL);
4309 LASSERT(dir->i_op->lookup != NULL);
4311 dentry = osd_child_dentry_get(env, obj,
4312 (char *)key, strlen((char *)key));
4314 if (obj->oo_hl_head != NULL) {
4315 hlock = osd_oti_get(env)->oti_hlock;
4316 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4317 dir, LDISKFS_HLOCK_LOOKUP);
4319 down_read(&obj->oo_ext_idx_sem);
4322 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4324 struct osd_thread_info *oti = osd_oti_get(env);
4325 struct osd_inode_id *id = &oti->oti_id;
4326 struct osd_idmap_cache *oic = &oti->oti_cache;
4327 struct osd_device *dev = osd_obj2dev(obj);
4329 ino = le32_to_cpu(de->inode);
4330 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4332 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4336 rc = osd_get_fid_from_dentry(de, rec);
4338 /* done with de, release bh */
4341 if (unlikely(ino == osd_remote_parent_ino(dev))) {
4342 const char *name = (const char *)key;
4344 /* If the parent is on remote MDT, and there
4345 * is no FID-in-dirent, then we have to get
4346 * the parent FID from the linkEA. */
4347 if (likely(strlen(name) == 2 &&
4348 name[0] == '.' && name[1] == '.'))
4349 rc = osd_get_pfid_from_linkea(env, obj,
4352 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4355 osd_id_gen(id, ino, OSD_OII_NOGEN);
4358 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
4359 fid_zero(&oic->oic_fid);
4364 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
4365 rc = osd_consistency_check(oti, dev, oic);
4367 fid_zero(&oic->oic_fid);
4376 ldiskfs_htree_unlock(hlock);
4378 up_read(&obj->oo_ext_idx_sem);
4383 * Find the osd object for given fid.
4385 * \param fid need to find the osd object having this fid
4387 * \retval osd_object on success
4388 * \retval -ve on error
4390 static struct osd_object *osd_object_find(const struct lu_env *env,
4391 struct dt_object *dt,
4392 const struct lu_fid *fid)
4394 struct lu_device *ludev = dt->do_lu.lo_dev;
4395 struct osd_object *child = NULL;
4396 struct lu_object *luch;
4397 struct lu_object *lo;
4400 * at this point topdev might not exist yet
4401 * (i.e. MGS is preparing profiles). so we can
4402 * not rely on topdev and instead lookup with
4403 * our device passed as topdev. this can't work
4404 * if the object isn't cached yet (as osd doesn't
4405 * allocate lu_header). IOW, the object must be
4406 * in the cache, otherwise lu_object_alloc() crashes
4409 luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4410 ludev : ludev->ld_site->ls_top_dev,
4412 if (!IS_ERR(luch)) {
4413 if (lu_object_exists(luch)) {
4414 lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4416 child = osd_obj(lo);
4418 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4419 "lu_object can't be located"
4420 DFID"\n", PFID(fid));
4422 if (child == NULL) {
4423 lu_object_put(env, luch);
4424 CERROR("Unable to get osd_object\n");
4425 child = ERR_PTR(-ENOENT);
4428 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4429 "lu_object does not exists "DFID"\n",
4431 lu_object_put(env, luch);
4432 child = ERR_PTR(-ENOENT);
4435 child = ERR_CAST(luch);
4442 * Put the osd object once done with it.
4444 * \param obj osd object that needs to be put
4446 static inline void osd_object_put(const struct lu_env *env,
4447 struct osd_object *obj)
4449 lu_object_put(env, &obj->oo_dt.do_lu);
4452 static int osd_index_declare_ea_insert(const struct lu_env *env,
4453 struct dt_object *dt,
4454 const struct dt_rec *rec,
4455 const struct dt_key *key,
4456 struct thandle *handle)
4458 struct osd_thandle *oh;
4459 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4460 struct lu_fid *fid = (struct lu_fid *)rec;
4461 int credits, rc = 0;
4464 LASSERT(!dt_object_remote(dt));
4465 LASSERT(handle != NULL);
4467 oh = container_of0(handle, struct osd_thandle, ot_super);
4468 LASSERT(oh->ot_handle == NULL);
4470 credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
4472 rc = osd_remote_fid(env, osd, fid);
4473 if (unlikely(rc < 0))
4476 /* a reference to remote inode is represented by an
4477 * agent inode which we have to create */
4478 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
4479 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4484 osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
4486 if (osd_dt_obj(dt)->oo_inode != NULL) {
4487 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4489 /* We ignore block quota on meta pool (MDTs), so needn't
4490 * calculate how many blocks will be consumed by this index
4492 rc = osd_declare_inode_qid(env, i_uid_read(inode),
4493 i_gid_read(inode), 0, oh,
4494 osd_dt_obj(dt), true, NULL, false);
4501 * Index add function for interoperability mode (b11826).
4502 * It will add the directory entry.This entry is needed to
4503 * maintain name->fid mapping.
4505 * \param key it is key i.e. file entry to be inserted
4506 * \param rec it is value of given key i.e. fid
4508 * \retval 0, on success
4509 * \retval -ve, on error
4511 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4512 const struct dt_rec *rec,
4513 const struct dt_key *key, struct thandle *th,
4516 struct osd_object *obj = osd_dt_obj(dt);
4517 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4518 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
4519 const struct lu_fid *fid = rec1->rec_fid;
4520 const char *name = (const char *)key;
4521 struct osd_thread_info *oti = osd_oti_get(env);
4522 struct osd_inode_id *id = &oti->oti_id;
4523 struct inode *child_inode = NULL;
4524 struct osd_object *child = NULL;
4528 if (!dt_object_exists(dt))
4531 LASSERT(osd_invariant(obj));
4532 LASSERT(!dt_object_remote(dt));
4533 LASSERT(th != NULL);
4535 osd_trans_exec_op(env, th, OSD_OT_INSERT);
4537 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
4539 rc = osd_remote_fid(env, osd, fid);
4541 CERROR("%s: Can not find object "DFID" rc %d\n",
4542 osd_name(osd), PFID(fid), rc);
4547 /* Insert remote entry */
4548 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4549 struct osd_mdobj_map *omm = osd->od_mdt_map;
4550 struct osd_thandle *oh;
4552 /* If parent on remote MDT, we need put this object
4554 oh = container_of(th, typeof(*oh), ot_super);
4555 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4557 CERROR("%s: add "DFID" error: rc = %d\n",
4559 PFID(lu_object_fid(&dt->do_lu)), rc);
4563 child_inode = igrab(omm->omm_remote_parent->d_inode);
4565 child_inode = osd_create_local_agent_inode(env, osd,
4566 obj, fid, rec1->rec_type & S_IFMT, th);
4567 if (IS_ERR(child_inode))
4568 RETURN(PTR_ERR(child_inode));
4571 /* Insert local entry */
4572 child = osd_object_find(env, dt, fid);
4573 if (IS_ERR(child)) {
4574 CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4575 osd_name(osd), PFID(fid),
4576 id->oii_ino, id->oii_gen,
4577 (int)PTR_ERR(child));
4578 RETURN(PTR_ERR(child));
4580 child_inode = igrab(child->oo_inode);
4583 rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
4585 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4586 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4590 osd_object_put(env, child);
4591 LASSERT(osd_invariant(obj));
4592 osd_trans_exec_check(env, th, OSD_OT_INSERT);
4597 * Initialize osd Iterator for given osd index object.
4599 * \param dt osd index object
4602 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4603 struct dt_object *dt,
4606 struct osd_it_iam *it;
4607 struct osd_object *obj = osd_dt_obj(dt);
4608 struct lu_object *lo = &dt->do_lu;
4609 struct iam_path_descr *ipd;
4610 struct iam_container *bag = &obj->oo_dir->od_container;
4612 if (!dt_object_exists(dt))
4613 return ERR_PTR(-ENOENT);
4617 return ERR_PTR(-ENOMEM);
4619 ipd = osd_it_ipd_get(env, bag);
4620 if (likely(ipd != NULL)) {
4624 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4625 return (struct dt_it *)it;
4628 return ERR_PTR(-ENOMEM);
4633 * free given Iterator.
4636 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4638 struct osd_it_iam *it = (struct osd_it_iam *)di;
4639 struct osd_object *obj = it->oi_obj;
4641 iam_it_fini(&it->oi_it);
4642 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4643 lu_object_put(env, &obj->oo_dt.do_lu);
4648 * Move Iterator to record specified by \a key
4650 * \param di osd iterator
4651 * \param key key for index
4653 * \retval +ve di points to record with least key not larger than key
4654 * \retval 0 di points to exact matched key
4655 * \retval -ve failure
4658 static int osd_it_iam_get(const struct lu_env *env,
4659 struct dt_it *di, const struct dt_key *key)
4661 struct osd_thread_info *oti = osd_oti_get(env);
4662 struct osd_it_iam *it = (struct osd_it_iam *)di;
4664 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4665 /* swab quota uid/gid */
4666 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4667 key = (struct dt_key *)&oti->oti_quota_id;
4670 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4676 * \param di osd iterator
4678 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4680 struct osd_it_iam *it = (struct osd_it_iam *)di;
4682 iam_it_put(&it->oi_it);
4686 * Move iterator by one record
4688 * \param di osd iterator
4690 * \retval +1 end of container reached
4692 * \retval -ve failure
4695 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4697 struct osd_it_iam *it = (struct osd_it_iam *)di;
4699 return iam_it_next(&it->oi_it);
4703 * Return pointer to the key under iterator.
4706 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4707 const struct dt_it *di)
4709 struct osd_thread_info *oti = osd_oti_get(env);
4710 struct osd_it_iam *it = (struct osd_it_iam *)di;
4711 struct osd_object *obj = it->oi_obj;
4714 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4716 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4717 /* swab quota uid/gid */
4718 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4719 key = (struct dt_key *)&oti->oti_quota_id;
4726 * Return size of key under iterator (in bytes)
4729 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4731 struct osd_it_iam *it = (struct osd_it_iam *)di;
4733 return iam_it_key_size(&it->oi_it);
4737 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4739 /* check if file type is required */
4740 if (ent->lde_attrs & LUDA_TYPE) {
4741 struct luda_type *lt;
4742 int align = sizeof(*lt) - 1;
4744 len = (len + align) & ~align;
4745 lt = (struct luda_type *)(ent->lde_name + len);
4746 lt->lt_type = cpu_to_le16(DTTOIF(type));
4749 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4753 * build lu direct from backend fs dirent.
4757 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4758 char *name, __u16 namelen, __u16 type, __u32 attr)
4760 ent->lde_attrs = attr | LUDA_FID;
4761 fid_cpu_to_le(&ent->lde_fid, fid);
4763 ent->lde_hash = cpu_to_le64(offset);
4764 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4766 strncpy(ent->lde_name, name, namelen);
4767 ent->lde_name[namelen] = '\0';
4768 ent->lde_namelen = cpu_to_le16(namelen);
4770 /* append lustre attributes */
4771 osd_it_append_attrs(ent, namelen, type);
4775 * Return pointer to the record under iterator.
4777 static int osd_it_iam_rec(const struct lu_env *env,
4778 const struct dt_it *di,
4779 struct dt_rec *dtrec, __u32 attr)
4781 struct osd_it_iam *it = (struct osd_it_iam *)di;
4782 struct osd_thread_info *info = osd_oti_get(env);
4785 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4786 const struct osd_fid_pack *rec;
4787 struct lu_fid *fid = &info->oti_fid;
4788 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
4794 name = (char *)iam_it_key_get(&it->oi_it);
4796 RETURN(PTR_ERR(name));
4798 namelen = iam_it_key_size(&it->oi_it);
4800 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4802 RETURN(PTR_ERR(rec));
4804 rc = osd_fid_unpack(fid, rec);
4808 hash = iam_it_store(&it->oi_it);
4810 /* IAM does not store object type in IAM index (dir) */
4811 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4813 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4814 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4815 (struct iam_rec *)dtrec);
4816 osd_quota_unpack(it->oi_obj, dtrec);
4818 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4819 (struct iam_rec *)dtrec);
4826 * Returns cookie for current Iterator position.
4828 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4830 struct osd_it_iam *it = (struct osd_it_iam *)di;
4832 return iam_it_store(&it->oi_it);
4836 * Restore iterator from cookie.
4838 * \param di osd iterator
4839 * \param hash Iterator location cookie
4841 * \retval +ve di points to record with least key not larger than key.
4842 * \retval 0 di points to exact matched key
4843 * \retval -ve failure
4846 static int osd_it_iam_load(const struct lu_env *env,
4847 const struct dt_it *di, __u64 hash)
4849 struct osd_it_iam *it = (struct osd_it_iam *)di;
4851 return iam_it_load(&it->oi_it, hash);
4854 static const struct dt_index_operations osd_index_iam_ops = {
4855 .dio_lookup = osd_index_iam_lookup,
4856 .dio_declare_insert = osd_index_declare_iam_insert,
4857 .dio_insert = osd_index_iam_insert,
4858 .dio_declare_delete = osd_index_declare_iam_delete,
4859 .dio_delete = osd_index_iam_delete,
4861 .init = osd_it_iam_init,
4862 .fini = osd_it_iam_fini,
4863 .get = osd_it_iam_get,
4864 .put = osd_it_iam_put,
4865 .next = osd_it_iam_next,
4866 .key = osd_it_iam_key,
4867 .key_size = osd_it_iam_key_size,
4868 .rec = osd_it_iam_rec,
4869 .store = osd_it_iam_store,
4870 .load = osd_it_iam_load
4876 * Creates or initializes iterator context.
4878 * \retval struct osd_it_ea, iterator structure on success
4881 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4882 struct dt_object *dt,
4885 struct osd_object *obj = osd_dt_obj(dt);
4886 struct osd_thread_info *info = osd_oti_get(env);
4887 struct osd_it_ea *oie;
4889 struct lu_object *lo = &dt->do_lu;
4890 struct dentry *obj_dentry;
4893 if (!dt_object_exists(dt) || obj->oo_destroyed)
4894 RETURN(ERR_PTR(-ENOENT));
4896 OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
4898 RETURN(ERR_PTR(-ENOMEM));
4899 obj_dentry = &oie->oie_dentry;
4901 obj_dentry->d_inode = obj->oo_inode;
4902 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4903 obj_dentry->d_name.hash = 0;
4905 oie->oie_rd_dirent = 0;
4906 oie->oie_it_dirent = 0;
4907 oie->oie_dirent = NULL;
4908 if (unlikely(!info->oti_it_ea_buf_used)) {
4909 oie->oie_buf = info->oti_it_ea_buf;
4910 info->oti_it_ea_buf_used = 1;
4912 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
4913 if (oie->oie_buf == NULL)
4914 RETURN(ERR_PTR(-ENOMEM));
4918 file = &oie->oie_file;
4920 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4921 if (attr & LUDA_64BITHASH)
4922 file->f_mode = FMODE_64BITHASH;
4924 file->f_mode = FMODE_32BITHASH;
4925 file->f_path.dentry = obj_dentry;
4926 file->f_mapping = obj->oo_inode->i_mapping;
4927 file->f_op = obj->oo_inode->i_fop;
4928 set_file_inode(file, obj->oo_inode);
4931 RETURN((struct dt_it *) oie);
4935 * Destroy or finishes iterator context.
4937 * \param di iterator structure to be destroyed
4939 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4941 struct osd_thread_info *info = osd_oti_get(env);
4942 struct osd_it_ea *oie = (struct osd_it_ea *)di;
4943 struct osd_object *obj = oie->oie_obj;
4944 struct inode *inode = obj->oo_inode;
4947 oie->oie_file.f_op->release(inode, &oie->oie_file);
4948 lu_object_put(env, &obj->oo_dt.do_lu);
4949 if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
4950 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
4952 info->oti_it_ea_buf_used = 0;
4953 OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
4958 * It position the iterator at given key, so that next lookup continues from
4959 * that key Or it is similar to dio_it->load() but based on a key,
4960 * rather than file position.
4962 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4965 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4967 static int osd_it_ea_get(const struct lu_env *env,
4968 struct dt_it *di, const struct dt_key *key)
4970 struct osd_it_ea *it = (struct osd_it_ea *)di;
4973 LASSERT(((const char *)key)[0] == '\0');
4974 it->oie_file.f_pos = 0;
4975 it->oie_rd_dirent = 0;
4976 it->oie_it_dirent = 0;
4977 it->oie_dirent = NULL;
4985 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4989 struct osd_filldir_cbs {
4990 #ifdef HAVE_DIR_CONTEXT
4991 struct dir_context ctx;
4993 struct osd_it_ea *it;
4996 * It is called internally by ->readdir(). It fills the
4997 * iterator's in-memory data structure with required
4998 * information i.e. name, namelen, rec_size etc.
5000 * \param buf in which information to be filled in.
5001 * \param name name of the file in given dir
5003 * \retval 0 on success
5004 * \retval 1 on buffer full
5006 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
5007 loff_t offset, __u64 ino,
5010 struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
5011 struct osd_object *obj = it->oie_obj;
5012 struct osd_it_ea_dirent *ent = it->oie_dirent;
5013 struct lu_fid *fid = &ent->oied_fid;
5014 struct osd_fid_pack *rec;
5017 /* this should never happen */
5018 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5019 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5023 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
5027 /* "." is just the object itself. */
5028 if (namelen == 1 && name[0] == '.') {
5029 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5030 } else if (d_type & LDISKFS_DIRENT_LUFID) {
5031 rec = (struct osd_fid_pack*) (name + namelen + 1);
5032 if (osd_fid_unpack(fid, rec) != 0)
5037 d_type &= ~LDISKFS_DIRENT_LUFID;
5039 /* NOT export local root. */
5040 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5041 ino = obj->oo_inode->i_ino;
5042 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5045 ent->oied_ino = ino;
5046 ent->oied_off = offset;
5047 ent->oied_namelen = namelen;
5048 ent->oied_type = d_type;
5050 memcpy(ent->oied_name, name, namelen);
5052 it->oie_rd_dirent++;
5053 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
5058 * Calls ->readdir() to load a directory entry at a time
5059 * and stored it in iterator's in-memory data structure.
5061 * \param di iterator's in memory structure
5063 * \retval 0 on success
5064 * \retval -ve on error
5065 * \retval +1 reach the end of entry
5067 static int osd_ldiskfs_it_fill(const struct lu_env *env,
5068 const struct dt_it *di)
5070 struct osd_it_ea *it = (struct osd_it_ea *)di;
5071 struct osd_object *obj = it->oie_obj;
5072 struct inode *inode = obj->oo_inode;
5073 struct htree_lock *hlock = NULL;
5074 struct file *filp = &it->oie_file;
5076 struct osd_filldir_cbs buf = {
5077 #ifdef HAVE_DIR_CONTEXT
5078 .ctx.actor = osd_ldiskfs_filldir,
5084 it->oie_dirent = it->oie_buf;
5085 it->oie_rd_dirent = 0;
5087 if (obj->oo_hl_head != NULL) {
5088 hlock = osd_oti_get(env)->oti_hlock;
5089 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5090 inode, LDISKFS_HLOCK_READDIR);
5092 down_read(&obj->oo_ext_idx_sem);
5095 #ifdef HAVE_DIR_CONTEXT
5096 buf.ctx.pos = filp->f_pos;
5097 rc = inode->i_fop->iterate(filp, &buf.ctx);
5098 filp->f_pos = buf.ctx.pos;
5100 rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
5104 ldiskfs_htree_unlock(hlock);
5106 up_read(&obj->oo_ext_idx_sem);
5108 if (it->oie_rd_dirent == 0) {
5109 /*If it does not get any dirent, it means it has been reached
5110 *to the end of the dir */
5111 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
5115 it->oie_dirent = it->oie_buf;
5116 it->oie_it_dirent = 1;
5123 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5124 * to load a directory entry at a time and stored it in
5125 * iterator's in-memory data structure.
5127 * \param di iterator's in memory structure
5129 * \retval +ve iterator reached to end
5130 * \retval 0 iterator not reached to end
5131 * \retval -ve on error
5133 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
5135 struct osd_it_ea *it = (struct osd_it_ea *)di;
5140 if (it->oie_it_dirent < it->oie_rd_dirent) {
5142 (void *) it->oie_dirent +
5143 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
5144 it->oie_dirent->oied_namelen);
5145 it->oie_it_dirent++;
5148 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
5151 rc = osd_ldiskfs_it_fill(env, di);
5158 * Returns the key at current position from iterator's in memory structure.
5160 * \param di iterator's in memory structure
5162 * \retval key i.e. struct dt_key on success
5164 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
5165 const struct dt_it *di)
5167 struct osd_it_ea *it = (struct osd_it_ea *)di;
5169 return (struct dt_key *)it->oie_dirent->oied_name;
5173 * Returns the key's size at current position from iterator's in memory structure.
5175 * \param di iterator's in memory structure
5177 * \retval key_size i.e. struct dt_key on success
5179 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
5181 struct osd_it_ea *it = (struct osd_it_ea *)di;
5183 return it->oie_dirent->oied_namelen;
5187 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5189 LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5190 "dot_dotdot = %d\n", dot_dotdot);
5192 if (LDISKFS_DIR_REC_LEN(de) >=
5193 __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5200 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
5201 unsigned blocksize, int dot_dotdot)
5204 return osd_dot_dotdot_has_space(de, dot_dotdot);
5206 if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
5207 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5214 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5215 struct dentry *dentry, const struct lu_fid *fid,
5216 struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5217 struct htree_lock *hlock, int dot_dotdot)
5219 struct inode *dir = dentry->d_parent->d_inode;
5220 struct inode *inode = dentry->d_inode;
5221 struct osd_fid_pack *rec;
5222 struct ldiskfs_dentry_param *ldp;
5223 int namelen = dentry->d_name.len;
5227 if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5228 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5231 /* There is enough space to hold the FID-in-dirent. */
5232 if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize,
5234 rc = ldiskfs_journal_get_write_access(jh, bh);
5238 de->name[namelen] = 0;
5239 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
5240 rec->fp_len = sizeof(struct lu_fid) + 1;
5241 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5242 de->file_type |= LDISKFS_DIRENT_LUFID;
5243 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
5248 LASSERTF(dot_dotdot == 0, "dot_dotdot = %d\n", dot_dotdot);
5250 rc = ldiskfs_delete_entry(jh, dir, de, bh);
5254 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5255 osd_get_ldiskfs_dirent_param(ldp, fid);
5256 dentry->d_fsdata = (void *)ldp;
5257 ll_vfs_dq_init(dir);
5258 rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5259 /* It is too bad, we cannot reinsert the name entry back.
5260 * That means we lose it! */
5262 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
5263 "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
5264 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5265 dir->i_ino, dir->i_generation, namelen,
5266 dentry->d_name.name, PFID(fid), rc);
5272 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5273 struct osd_it_ea *it, struct lu_fid *fid,
5274 struct osd_inode_id *id, __u32 *attr)
5276 struct osd_thread_info *info = osd_oti_get(env);
5277 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
5278 struct osd_device *dev = osd_obj2dev(obj);
5279 struct super_block *sb = osd_sb(dev);
5280 const char *devname =
5281 LDISKFS_SB(sb)->s_es->s_volume_name;
5282 struct osd_it_ea_dirent *ent = it->oie_dirent;
5283 struct inode *dir = obj->oo_inode;
5284 struct htree_lock *hlock = NULL;
5285 struct buffer_head *bh = NULL;
5286 handle_t *jh = NULL;
5287 struct ldiskfs_dir_entry_2 *de;
5288 struct dentry *dentry;
5289 struct inode *inode;
5296 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5297 inode = osd_iget(info, dev, id);
5298 if (IS_ERR(inode)) {
5299 rc = PTR_ERR(inode);
5300 if (rc == -ENOENT || rc == -ESTALE) {
5301 *attr |= LUDA_UNKNOWN;
5304 CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
5305 "check_repair, dir = %lu/%u, name = %.*s: "
5307 devname, dir->i_ino, dir->i_generation,
5308 ent->oied_namelen, ent->oied_name, rc);
5314 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5316 rc = osd_get_lma(info, inode, dentry, lma);
5322 if (ent->oied_name[0] == '.') {
5323 if (ent->oied_namelen == 1)
5325 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5329 /* We need to ensure that the name entry is still valid.
5330 * Because it may be removed or renamed by other already.
5332 * The unlink or rename operation will start journal before PDO lock,
5333 * so to avoid deadlock, here we need to start journal handle before
5334 * related PDO lock also. But because we do not know whether there
5335 * will be something to be repaired before PDO lock, we just start
5336 * journal without conditions.
5338 * We may need to remove the name entry firstly, then insert back.
5339 * One credit is for user quota file update.
5340 * One credit is for group quota file update.
5341 * Two credits are for dirty inode. */
5342 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5343 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5345 if (dev->od_dirent_journal != 0) {
5348 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5351 CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent "
5352 "check_repair, dir = %lu/%u, credits = %d, "
5353 "name = %.*s: rc = %d\n",
5354 devname, dir->i_ino, dir->i_generation, credits,
5355 ent->oied_namelen, ent->oied_name, rc);
5357 GOTO(out_inode, rc);
5360 if (obj->oo_hl_head != NULL) {
5361 hlock = osd_oti_get(env)->oti_hlock;
5362 /* "0" means exclusive lock for the whole directory.
5363 * We need to prevent others access such name entry
5364 * during the delete + insert. Neither HLOCK_ADD nor
5365 * HLOCK_DEL cannot guarantee the atomicity. */
5366 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5368 down_write(&obj->oo_ext_idx_sem);
5371 if (obj->oo_hl_head != NULL) {
5372 hlock = osd_oti_get(env)->oti_hlock;
5373 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5374 LDISKFS_HLOCK_LOOKUP);
5376 down_read(&obj->oo_ext_idx_sem);
5380 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5381 /* For dot/dotdot entry, if there is not enough space to hold the
5382 * FID-in-dirent, just keep them there. It only happens when the
5383 * device upgraded from 1.8 or restored from MDT file-level backup.
5384 * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5385 * and needs to get FID from LMA when readdir, it will not affect the
5386 * performance much. */
5387 if ((bh == NULL) || (le32_to_cpu(de->inode) != inode->i_ino) ||
5388 (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5389 *attr |= LUDA_IGNORE;
5395 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
5396 struct lu_fid *tfid = &lma->lma_self_fid;
5398 *attr |= LUDA_IGNORE;
5399 /* It must be REMOTE_PARENT_DIR and as the
5400 * dotdot entry of remote directory */
5401 if (unlikely(dot_dotdot != 2 ||
5402 fid_seq(tfid) != FID_SEQ_LOCAL_FILE ||
5403 fid_oid(tfid) != REMOTE_PARENT_DIR_OID)) {
5404 CDEBUG(D_LFSCK, "%.16s: expect remote agent "
5405 "parent directory, but got %.*s under "
5406 "dir = %lu/%u with the FID "DFID"\n",
5407 devname, ent->oied_namelen,
5408 ent->oied_name, dir->i_ino,
5409 dir->i_generation, PFID(tfid));
5411 GOTO(out, rc = -EIO);
5417 if (fid_is_sane(fid)) {
5418 /* FID-in-dirent is valid. */
5419 if (lu_fid_eq(fid, &lma->lma_self_fid))
5422 /* Do not repair under dryrun mode. */
5423 if (*attr & LUDA_VERIFY_DRYRUN) {
5424 *attr |= LUDA_REPAIR;
5432 ldiskfs_htree_unlock(hlock);
5434 up_read(&obj->oo_ext_idx_sem);
5435 dev->od_dirent_journal = 1;
5440 *fid = lma->lma_self_fid;
5442 /* Update the FID-in-dirent. */
5443 rc = osd_dirent_reinsert(env, jh, dentry, fid, bh, de,
5446 *attr |= LUDA_REPAIR;
5448 CDEBUG(D_LFSCK, "%.16s: fail to update FID "
5449 "in the dirent, dir = %lu/%u, "
5450 "name = %.*s, "DFID": rc = %d\n",
5451 devname, dir->i_ino, dir->i_generation,
5452 ent->oied_namelen, ent->oied_name,
5455 /* Do not repair under dryrun mode. */
5456 if (*attr & LUDA_VERIFY_DRYRUN) {
5457 *fid = lma->lma_self_fid;
5458 *attr |= LUDA_REPAIR;
5466 ldiskfs_htree_unlock(hlock);
5468 up_read(&obj->oo_ext_idx_sem);
5469 dev->od_dirent_journal = 1;
5474 *fid = lma->lma_self_fid;
5476 /* Append the FID-in-dirent. */
5477 rc = osd_dirent_reinsert(env, jh, dentry, fid, bh, de,
5480 *attr |= LUDA_REPAIR;
5482 CDEBUG(D_LFSCK, "%.16s: fail to append FID "
5483 "after the dirent, dir = %lu/%u, "
5484 "name = %.*s, "DFID": rc = %d\n",
5485 devname, dir->i_ino, dir->i_generation,
5486 ent->oied_namelen, ent->oied_name,
5490 /* Do not repair under dryrun mode. */
5491 if (*attr & LUDA_VERIFY_DRYRUN) {
5492 if (fid_is_sane(fid)) {
5493 *attr |= LUDA_REPAIR;
5495 lu_igif_build(fid, inode->i_ino,
5496 inode->i_generation);
5497 *attr |= LUDA_UPGRADE;
5506 ldiskfs_htree_unlock(hlock);
5508 up_read(&obj->oo_ext_idx_sem);
5509 dev->od_dirent_journal = 1;
5515 if (unlikely(fid_is_sane(fid))) {
5516 /* FID-in-dirent exists, but FID-in-LMA is lost.
5517 * Trust the FID-in-dirent, and add FID-in-LMA. */
5518 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5520 *attr |= LUDA_REPAIR;
5522 CDEBUG(D_LFSCK, "%.16s: fail to set LMA for "
5523 "update dirent, dir = %lu/%u, "
5524 "name = %.*s, "DFID": rc = %d\n",
5525 devname, dir->i_ino, dir->i_generation,
5526 ent->oied_namelen, ent->oied_name,
5529 lu_igif_build(fid, inode->i_ino, inode->i_generation);
5530 /* It is probably IGIF object. Only aappend the
5531 * FID-in-dirent. OI scrub will process FID-in-LMA. */
5532 rc = osd_dirent_reinsert(env, jh, dentry, fid, bh, de,
5535 *attr |= LUDA_UPGRADE;
5537 CDEBUG(D_LFSCK, "%.16s: fail to append IGIF "
5538 "after the dirent, dir = %lu/%u, "
5539 "name = %.*s, "DFID": rc = %d\n",
5540 devname, dir->i_ino, dir->i_generation,
5541 ent->oied_namelen, ent->oied_name,
5550 if (hlock != NULL) {
5551 ldiskfs_htree_unlock(hlock);
5553 if (dev->od_dirent_journal != 0)
5554 up_write(&obj->oo_ext_idx_sem);
5556 up_read(&obj->oo_ext_idx_sem);
5560 ldiskfs_journal_stop(jh);
5564 if (rc >= 0 && !dirty)
5565 dev->od_dirent_journal = 0;
5571 * Returns the value at current position from iterator's in memory structure.
5573 * \param di struct osd_it_ea, iterator's in memory structure
5574 * \param attr attr requested for dirent.
5575 * \param lde lustre dirent
5577 * \retval 0 no error and \param lde has correct lustre dirent.
5578 * \retval -ve on error
5580 static inline int osd_it_ea_rec(const struct lu_env *env,
5581 const struct dt_it *di,
5582 struct dt_rec *dtrec, __u32 attr)
5584 struct osd_it_ea *it = (struct osd_it_ea *)di;
5585 struct osd_object *obj = it->oie_obj;
5586 struct osd_device *dev = osd_obj2dev(obj);
5587 struct osd_thread_info *oti = osd_oti_get(env);
5588 struct osd_inode_id *id = &oti->oti_id;
5589 struct lu_fid *fid = &it->oie_dirent->oied_fid;
5590 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
5591 __u32 ino = it->oie_dirent->oied_ino;
5595 LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
5597 if (attr & LUDA_VERIFY) {
5598 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5599 attr |= LUDA_IGNORE;
5600 /* If the parent is on remote MDT, and there
5601 * is no FID-in-dirent, then we have to get
5602 * the parent FID from the linkEA. */
5603 if (!fid_is_sane(fid) &&
5604 it->oie_dirent->oied_namelen == 2 &&
5605 it->oie_dirent->oied_name[0] == '.' &&
5606 it->oie_dirent->oied_name[1] == '.')
5607 osd_get_pfid_from_linkea(env, obj, fid);
5609 rc = osd_dirent_check_repair(env, obj, it, fid, id,
5613 if (!fid_is_sane(fid)) {
5614 attr &= ~LUDA_IGNORE;
5615 attr |= LUDA_UNKNOWN;
5618 attr &= ~LU_DIRENT_ATTRS_MASK;
5619 if (!fid_is_sane(fid)) {
5620 bool is_dotdot = false;
5621 if (it->oie_dirent->oied_namelen == 2 &&
5622 it->oie_dirent->oied_name[0] == '.' &&
5623 it->oie_dirent->oied_name[1] == '.')
5625 /* If the parent is on remote MDT, and there
5626 * is no FID-in-dirent, then we have to get
5627 * the parent FID from the linkEA. */
5628 if (ino == osd_remote_parent_ino(dev) && is_dotdot) {
5629 rc = osd_get_pfid_from_linkea(env, obj, fid);
5631 if (is_dotdot == false &&
5632 OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
5635 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5638 osd_id_gen(id, ino, OSD_OII_NOGEN);
5642 /* Pack the entry anyway, at least the offset is right. */
5643 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5644 it->oie_dirent->oied_name,
5645 it->oie_dirent->oied_namelen,
5646 it->oie_dirent->oied_type, attr);
5651 if (osd_remote_fid(env, dev, fid))
5654 if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
5655 osd_add_oi_cache(oti, dev, id, fid);
5657 RETURN(rc > 0 ? 0 : rc);
5661 * Returns the record size size at current position.
5663 * This function will return record(lu_dirent) size in bytes.
5665 * \param[in] env execution environment
5666 * \param[in] di iterator's in memory structure
5667 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
5668 * calculate the lu_dirent size.
5670 * \retval record size(in bytes & in memory) of the current lu_dirent
5673 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5676 struct osd_it_ea *it = (struct osd_it_ea *)di;
5678 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5682 * Returns a cookie for current position of the iterator head, so that
5683 * user can use this cookie to load/start the iterator next time.
5685 * \param di iterator's in memory structure
5687 * \retval cookie for current position, on success
5689 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5691 struct osd_it_ea *it = (struct osd_it_ea *)di;
5693 return it->oie_dirent->oied_off;
5697 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5698 * to load a directory entry at a time and stored it i inn,
5699 * in iterator's in-memory data structure.
5701 * \param di struct osd_it_ea, iterator's in memory structure
5703 * \retval +ve on success
5704 * \retval -ve on error
5706 static int osd_it_ea_load(const struct lu_env *env,
5707 const struct dt_it *di, __u64 hash)
5709 struct osd_it_ea *it = (struct osd_it_ea *)di;
5713 it->oie_file.f_pos = hash;
5715 rc = osd_ldiskfs_it_fill(env, di);
5726 * Index lookup function for interoperability mode (b11826).
5728 * \param key, key i.e. file name to be searched
5730 * \retval +ve, on success
5731 * \retval -ve, on error
5733 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5734 struct dt_rec *rec, const struct dt_key *key)
5736 struct osd_object *obj = osd_dt_obj(dt);
5741 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5742 LINVRNT(osd_invariant(obj));
5744 rc = osd_ea_lookup_rec(env, obj, rec, key);
5751 * Index and Iterator operations for interoperability
5752 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5754 static const struct dt_index_operations osd_index_ea_ops = {
5755 .dio_lookup = osd_index_ea_lookup,
5756 .dio_declare_insert = osd_index_declare_ea_insert,
5757 .dio_insert = osd_index_ea_insert,
5758 .dio_declare_delete = osd_index_declare_ea_delete,
5759 .dio_delete = osd_index_ea_delete,
5761 .init = osd_it_ea_init,
5762 .fini = osd_it_ea_fini,
5763 .get = osd_it_ea_get,
5764 .put = osd_it_ea_put,
5765 .next = osd_it_ea_next,
5766 .key = osd_it_ea_key,
5767 .key_size = osd_it_ea_key_size,
5768 .rec = osd_it_ea_rec,
5769 .rec_size = osd_it_ea_rec_size,
5770 .store = osd_it_ea_store,
5771 .load = osd_it_ea_load
5775 static void *osd_key_init(const struct lu_context *ctx,
5776 struct lu_context_key *key)
5778 struct osd_thread_info *info;
5780 OBD_ALLOC_PTR(info);
5782 return ERR_PTR(-ENOMEM);
5784 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5785 if (info->oti_it_ea_buf == NULL)
5788 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5790 info->oti_hlock = ldiskfs_htree_lock_alloc();
5791 if (info->oti_hlock == NULL)
5797 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5800 return ERR_PTR(-ENOMEM);
5803 static void osd_key_fini(const struct lu_context *ctx,
5804 struct lu_context_key *key, void* data)
5806 struct osd_thread_info *info = data;
5808 if (info->oti_inode != NULL)
5809 OBD_FREE_PTR(info->oti_inode);
5810 if (info->oti_hlock != NULL)
5811 ldiskfs_htree_lock_free(info->oti_hlock);
5812 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5813 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5814 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5815 lu_buf_free(&info->oti_big_buf);
5819 static void osd_key_exit(const struct lu_context *ctx,
5820 struct lu_context_key *key, void *data)
5822 struct osd_thread_info *info = data;
5824 LASSERT(info->oti_r_locks == 0);
5825 LASSERT(info->oti_w_locks == 0);
5826 LASSERT(info->oti_txns == 0);
5829 /* type constructor/destructor: osd_type_init, osd_type_fini */
5830 LU_TYPE_INIT_FINI(osd, &osd_key);
5832 struct lu_context_key osd_key = {
5833 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5834 .lct_init = osd_key_init,
5835 .lct_fini = osd_key_fini,
5836 .lct_exit = osd_key_exit
5840 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5841 const char *name, struct lu_device *next)
5843 struct osd_device *osd = osd_dev(d);
5845 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5846 >= sizeof(osd->od_svname))
5848 return osd_procfs_init(osd, name);
5851 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5853 struct seq_server_site *ss = osd_seq_site(osd);
5857 if (osd->od_is_ost || osd->od_cl_seq != NULL)
5860 if (unlikely(ss == NULL))
5863 OBD_ALLOC_PTR(osd->od_cl_seq);
5864 if (osd->od_cl_seq == NULL)
5867 rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5868 osd->od_svname, ss->ss_server_seq);
5870 OBD_FREE_PTR(osd->od_cl_seq);
5871 osd->od_cl_seq = NULL;
5875 if (ss->ss_node_id == 0) {
5876 /* If the OSD on the sequence controller(MDT0), then allocate
5877 * sequence here, otherwise allocate sequence after connected
5878 * to MDT0 (see mdt_register_lwp_callback()). */
5879 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
5880 &osd->od_cl_seq->lcs_space, env);
5886 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5888 if (osd->od_cl_seq == NULL)
5891 seq_client_fini(osd->od_cl_seq);
5892 OBD_FREE_PTR(osd->od_cl_seq);
5893 osd->od_cl_seq = NULL;
5896 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5900 /* shutdown quota slave instance associated with the device */
5901 if (o->od_quota_slave != NULL) {
5902 qsd_fini(env, o->od_quota_slave);
5903 o->od_quota_slave = NULL;
5906 osd_fid_fini(env, o);
5911 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5915 if (o->od_mnt != NULL) {
5916 shrink_dcache_sb(osd_sb(o));
5917 osd_sync(env, &o->od_dt_dev);
5926 static int osd_mount(const struct lu_env *env,
5927 struct osd_device *o, struct lustre_cfg *cfg)
5929 const char *name = lustre_cfg_string(cfg, 0);
5930 const char *dev = lustre_cfg_string(cfg, 1);
5932 unsigned long page, s_flags, lmd_flags = 0;
5933 struct page *__page;
5934 struct file_system_type *type;
5935 char *options = NULL;
5937 struct osd_thread_info *info = osd_oti_get(env);
5938 struct lu_fid *fid = &info->oti_fid;
5939 struct inode *inode;
5940 int rc = 0, force_over_128tb = 0;
5943 if (o->od_mnt != NULL)
5946 if (strlen(dev) >= sizeof(o->od_mntdev))
5948 strcpy(o->od_mntdev, dev);
5950 str = lustre_cfg_string(cfg, 2);
5951 s_flags = simple_strtoul(str, NULL, 0);
5952 str = strstr(str, ":");
5954 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5955 opts = lustre_cfg_string(cfg, 3);
5957 if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
5958 CERROR("%s: device %s extents feature is not guaranteed to "
5959 "work on big-endian systems. Use \"bigendian_extents\" "
5960 "mount option to override.\n", name, dev);
5964 if (opts != NULL && strstr(opts, "force_over_128tb") != NULL)
5965 force_over_128tb = 1;
5967 __page = alloc_page(GFP_IOFS);
5969 GOTO(out, rc = -ENOMEM);
5970 page = (unsigned long)page_address(__page);
5971 options = (char *)page;
5974 /* strip out the options for back compatiblity */
5975 static char *sout[] = {
5982 /* strip out option we processed in osd */
5983 "bigendian_extents",
5987 strcat(options, opts);
5988 for (rc = 0, str = options; sout[rc]; ) {
5989 char *op = strstr(str, sout[rc]);
5995 if (op == options || *(op - 1) == ',') {
5996 str = op + strlen(sout[rc]);
5997 if (*str == ',' || *str == '\0') {
5998 *str == ',' ? str++ : str;
5999 memmove(op, str, strlen(str) + 1);
6002 for (str = op; *str != ',' && *str != '\0'; str++)
6006 strncat(options, "user_xattr,acl", 14);
6009 /* Glom up mount options */
6010 if (*options != '\0')
6011 strcat(options, ",");
6012 strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
6014 type = get_fs_type("ldiskfs");
6016 CERROR("%s: cannot find ldiskfs module\n", name);
6017 GOTO(out, rc = -ENODEV);
6020 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
6021 module_put(type->owner);
6023 if (IS_ERR(o->od_mnt)) {
6024 rc = PTR_ERR(o->od_mnt);
6026 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
6030 if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) > (8ULL << 32) &&
6031 force_over_128tb == 0) {
6032 CERROR("%s: device %s LDISKFS does not support filesystems "
6033 "greater than 128TB and can cause data corruption. "
6034 "Use \"force_over_128tb\" mount option to override.\n",
6036 GOTO(out, rc = -EINVAL);
6039 #ifdef HAVE_DEV_SET_RDONLY
6040 if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
6041 CERROR("%s: underlying device %s is marked as read-only. "
6042 "Setup failed\n", name, dev);
6043 GOTO(out_mnt, rc = -EROFS);
6047 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
6048 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
6049 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
6050 GOTO(out_mnt, rc = -EINVAL);
6053 #ifdef LDISKFS_MOUNT_DIRDATA
6054 if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
6055 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6056 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
6057 else if (!o->od_is_ost)
6058 CWARN("%s: device %s was upgraded from Lustre-1.x without "
6059 "enabling the dirdata feature. If you do not want to "
6060 "downgrade to Lustre-1.x again, you can enable it via "
6061 "'tune2fs -O dirdata device'\n", name, dev);
6063 inode = osd_sb(o)->s_root->d_inode;
6064 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
6065 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
6067 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
6071 if (lmd_flags & LMD_FLG_NOSCRUB)
6082 __free_page(__page);
6087 static struct lu_device *osd_device_fini(const struct lu_env *env,
6088 struct lu_device *d)
6090 struct osd_device *o = osd_dev(d);
6093 osd_shutdown(env, o);
6095 osd_scrub_cleanup(env, o);
6096 osd_obj_map_fini(o);
6102 static int osd_device_init0(const struct lu_env *env,
6103 struct osd_device *o,
6104 struct lustre_cfg *cfg)
6106 struct lu_device *l = osd2lu_dev(o);
6107 struct osd_thread_info *info;
6111 /* if the module was re-loaded, env can loose its keys */
6112 rc = lu_env_refill((struct lu_env *) env);
6115 info = osd_oti_get(env);
6118 l->ld_ops = &osd_lu_ops;
6119 o->od_dt_dev.dd_ops = &osd_dt_ops;
6121 spin_lock_init(&o->od_osfs_lock);
6122 mutex_init(&o->od_otable_mutex);
6123 INIT_LIST_HEAD(&o->od_orphan_list);
6125 o->od_read_cache = 1;
6126 o->od_writethrough_cache = 1;
6127 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
6129 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
6130 sizeof(o->od_svname));
6131 if (cplen >= sizeof(o->od_svname)) {
6136 o->od_index = -1; /* -1 means index is invalid */
6137 rc = server_name2index(o->od_svname, &o->od_index, NULL);
6138 if (rc == LDD_F_SV_TYPE_OST)
6141 o->od_full_scrub_ratio = OFSR_DEFAULT;
6142 o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
6143 rc = osd_mount(env, o, cfg);
6147 rc = osd_obj_map_init(env, o);
6151 rc = lu_site_init(&o->od_site, l);
6153 GOTO(out_compat, rc);
6154 o->od_site.ls_bottom_dev = l;
6156 rc = lu_site_init_finish(&o->od_site);
6160 INIT_LIST_HEAD(&o->od_ios_list);
6161 /* setup scrub, including OI files initialization */
6162 rc = osd_scrub_setup(env, o);
6166 rc = osd_procfs_init(o, o->od_svname);
6168 CERROR("%s: can't initialize procfs: rc = %d\n",
6170 GOTO(out_scrub, rc);
6173 LASSERT(l->ld_site->ls_linkage.next != NULL);
6174 LASSERT(l->ld_site->ls_linkage.prev != NULL);
6176 /* initialize quota slave instance */
6177 o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
6179 if (IS_ERR(o->od_quota_slave)) {
6180 rc = PTR_ERR(o->od_quota_slave);
6181 o->od_quota_slave = NULL;
6182 GOTO(out_procfs, rc);
6190 osd_scrub_cleanup(env, o);
6192 lu_site_fini(&o->od_site);
6194 osd_obj_map_fini(o);
6201 static struct lu_device *osd_device_alloc(const struct lu_env *env,
6202 struct lu_device_type *t,
6203 struct lustre_cfg *cfg)
6205 struct osd_device *o;
6210 return ERR_PTR(-ENOMEM);
6212 rc = dt_device_init(&o->od_dt_dev, t);
6214 /* Because the ctx might be revived in dt_device_init,
6215 * refill the env here */
6216 lu_env_refill((struct lu_env *)env);
6217 rc = osd_device_init0(env, o, cfg);
6219 dt_device_fini(&o->od_dt_dev);
6222 if (unlikely(rc != 0))
6225 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
6228 static struct lu_device *osd_device_free(const struct lu_env *env,
6229 struct lu_device *d)
6231 struct osd_device *o = osd_dev(d);
6234 /* XXX: make osd top device in order to release reference */
6235 d->ld_site->ls_top_dev = d;
6236 lu_site_purge(env, d->ld_site, -1);
6237 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
6238 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
6239 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
6241 lu_site_fini(&o->od_site);
6242 dt_device_fini(&o->od_dt_dev);
6247 static int osd_process_config(const struct lu_env *env,
6248 struct lu_device *d, struct lustre_cfg *cfg)
6250 struct osd_device *o = osd_dev(d);
6254 switch (cfg->lcfg_command) {
6256 rc = osd_mount(env, o, cfg);
6259 lu_dev_del_linkage(d->ld_site, d);
6260 rc = osd_shutdown(env, o);
6263 LASSERT(&o->od_dt_dev);
6264 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
6265 cfg, &o->od_dt_dev);
6266 if (rc > 0 || rc == -ENOSYS)
6267 rc = class_process_proc_param(PARAM_OST,
6268 lprocfs_osd_obd_vars,
6269 cfg, &o->od_dt_dev);
6278 static int osd_recovery_complete(const struct lu_env *env,
6279 struct lu_device *d)
6281 struct osd_device *osd = osd_dev(d);
6285 if (osd->od_quota_slave == NULL)
6288 /* start qsd instance on recovery completion, this notifies the quota
6289 * slave code that we are about to process new requests now */
6290 rc = qsd_start(env, osd->od_quota_slave);
6295 * we use exports to track all osd users
6297 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
6298 struct obd_device *obd, struct obd_uuid *cluuid,
6299 struct obd_connect_data *data, void *localdata)
6301 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6302 struct lustre_handle conn;
6306 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6308 rc = class_connect(&conn, obd, cluuid);
6312 *exp = class_conn2export(&conn);
6314 spin_lock(&osd->od_osfs_lock);
6316 spin_unlock(&osd->od_osfs_lock);
6322 * once last export (we don't count self-export) disappeared
6323 * osd can be released
6325 static int osd_obd_disconnect(struct obd_export *exp)
6327 struct obd_device *obd = exp->exp_obd;
6328 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6329 int rc, release = 0;
6332 /* Only disconnect the underlying layers on the final disconnect. */
6333 spin_lock(&osd->od_osfs_lock);
6335 if (osd->od_connects == 0)
6337 spin_unlock(&osd->od_osfs_lock);
6339 rc = class_disconnect(exp); /* bz 9811 */
6341 if (rc == 0 && release)
6342 class_manual_cleanup(obd);
6346 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6347 struct lu_device *dev)
6349 struct osd_device *osd = osd_dev(dev);
6350 struct lr_server_data *lsd =
6351 &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
6355 if (osd->od_quota_slave != NULL) {
6356 /* set up quota slave objects */
6357 result = qsd_prepare(env, osd->od_quota_slave);
6362 if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
6363 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
6364 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
6365 osd->od_index_in_idif = 1;
6367 osd->od_index_in_idif = 0;
6368 result = osd_register_proc_index_in_idif(osd);
6373 osd->od_index_in_idif = 1;
6377 result = osd_fid_init(env, osd);
6382 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6383 struct lu_fid *fid, struct md_op_data *op_data)
6385 struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6387 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6390 static const struct lu_object_operations osd_lu_obj_ops = {
6391 .loo_object_init = osd_object_init,
6392 .loo_object_delete = osd_object_delete,
6393 .loo_object_release = osd_object_release,
6394 .loo_object_free = osd_object_free,
6395 .loo_object_print = osd_object_print,
6396 .loo_object_invariant = osd_object_invariant
6399 const struct lu_device_operations osd_lu_ops = {
6400 .ldo_object_alloc = osd_object_alloc,
6401 .ldo_process_config = osd_process_config,
6402 .ldo_recovery_complete = osd_recovery_complete,
6403 .ldo_prepare = osd_prepare,
6406 static const struct lu_device_type_operations osd_device_type_ops = {
6407 .ldto_init = osd_type_init,
6408 .ldto_fini = osd_type_fini,
6410 .ldto_start = osd_type_start,
6411 .ldto_stop = osd_type_stop,
6413 .ldto_device_alloc = osd_device_alloc,
6414 .ldto_device_free = osd_device_free,
6416 .ldto_device_init = osd_device_init,
6417 .ldto_device_fini = osd_device_fini
6420 static struct lu_device_type osd_device_type = {
6421 .ldt_tags = LU_DEVICE_DT,
6422 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
6423 .ldt_ops = &osd_device_type_ops,
6424 .ldt_ctx_tags = LCT_LOCAL,
6427 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
6429 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6430 struct super_block *sb = osd_sb(osd);
6432 return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
6436 * lprocfs legacy support.
6438 static struct obd_ops osd_obd_device_ops = {
6439 .o_owner = THIS_MODULE,
6440 .o_connect = osd_obd_connect,
6441 .o_disconnect = osd_obd_disconnect,
6442 .o_fid_alloc = osd_fid_alloc,
6443 .o_health_check = osd_health_check,
6446 static int __init osd_init(void)
6450 LASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
6451 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
6452 /* please, try to keep osd_thread_info smaller than a page */
6453 CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
6458 rc = lu_kmem_init(ldiskfs_caches);
6462 rc = class_register_type(&osd_obd_device_ops, NULL, true,
6463 lprocfs_osd_module_vars,
6464 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6466 lu_kmem_fini(ldiskfs_caches);
6470 static void __exit osd_exit(void)
6472 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6473 lu_kmem_fini(ldiskfs_caches);
6476 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6477 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6478 MODULE_VERSION(LUSTRE_VERSION_STRING);
6479 MODULE_LICENSE("GPL");
6481 module_init(osd_init);
6482 module_exit(osd_exit);