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, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osd/osd_handler.c
38 * Top-level entry points into osd module
40 * Author: Nikita Danilov <nikita@clusterfs.com>
41 * Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
44 #define DEBUG_SUBSYSTEM S_OSD
46 #include <linux/module.h>
48 /* LUSTRE_VERSION_CODE */
49 #include <lustre_ver.h>
50 /* prerequisite for linux/xattr.h */
51 #include <linux/types.h>
52 /* prerequisite for linux/xattr.h */
54 /* XATTR_{REPLACE,CREATE} */
55 #include <linux/xattr.h>
58 * struct OBD_{ALLOC,FREE}*()
61 #include <obd_support.h>
62 /* struct ptlrpc_thread */
63 #include <lustre_net.h>
64 #include <lustre_fid.h>
66 #include <lustre_param.h>
68 #include "osd_internal.h"
69 #include "osd_dynlocks.h"
71 /* llo_* api support */
72 #include <md_object.h>
73 #include <lustre_quota.h>
75 #include <ldiskfs/xattr.h>
76 #include <lustre_linkea.h>
79 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
80 "ldiskfs with parallel directory operations");
82 int ldiskfs_track_declares_assert;
83 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
84 "LBUG during tracking of declares");
86 /* Slab to allocate dynlocks */
87 struct kmem_cache *dynlock_cachep;
89 /* Slab to allocate osd_it_ea */
90 struct kmem_cache *osd_itea_cachep;
92 static struct lu_kmem_descr ldiskfs_caches[] = {
94 .ckd_cache = &dynlock_cachep,
95 .ckd_name = "dynlock_cache",
96 .ckd_size = sizeof(struct dynlock_handle)
99 .ckd_cache = &osd_itea_cachep,
100 .ckd_name = "osd_itea_cache",
101 .ckd_size = sizeof(struct osd_it_ea)
108 static const char dot[] = ".";
109 static const char dotdot[] = "..";
110 static const char remote_obj_dir[] = "REM_OBJ_DIR";
112 static const struct lu_object_operations osd_lu_obj_ops;
113 static const struct dt_object_operations osd_obj_ops;
114 static const struct dt_object_operations osd_obj_ea_ops;
115 static const struct dt_object_operations osd_obj_otable_it_ops;
116 static const struct dt_index_operations osd_index_iam_ops;
117 static const struct dt_index_operations osd_index_ea_ops;
119 int osd_trans_declare_op2rb[] = {
120 [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
121 [OSD_OT_PUNCH] = OSD_OT_MAX,
122 [OSD_OT_XATTR_SET] = OSD_OT_XATTR_SET,
123 [OSD_OT_CREATE] = OSD_OT_DESTROY,
124 [OSD_OT_DESTROY] = OSD_OT_CREATE,
125 [OSD_OT_REF_ADD] = OSD_OT_REF_DEL,
126 [OSD_OT_REF_DEL] = OSD_OT_REF_ADD,
127 [OSD_OT_WRITE] = OSD_OT_WRITE,
128 [OSD_OT_INSERT] = OSD_OT_DELETE,
129 [OSD_OT_DELETE] = OSD_OT_INSERT,
130 [OSD_OT_UPDATE] = OSD_OT_MAX,
131 [OSD_OT_QUOTA] = OSD_OT_MAX,
134 static int osd_has_index(const struct osd_object *obj)
136 return obj->oo_dt.do_index_ops != NULL;
139 static int osd_object_invariant(const struct lu_object *l)
141 return osd_invariant(osd_obj(l));
145 * Concurrency: doesn't matter
149 * Concurrency: doesn't matter
151 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
153 struct osd_thread_info *oti = osd_oti_get(env);
154 return oti->oti_w_locks > 0 && o->oo_owner == env;
158 * Concurrency: doesn't access mutable data
160 static int osd_root_get(const struct lu_env *env,
161 struct dt_device *dev, struct lu_fid *f)
163 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
168 * OSD object methods.
172 * Concurrency: no concurrent access is possible that early in object
175 static struct lu_object *osd_object_alloc(const struct lu_env *env,
176 const struct lu_object_header *hdr,
179 struct osd_object *mo;
185 l = &mo->oo_dt.do_lu;
186 dt_object_init(&mo->oo_dt, NULL, d);
187 mo->oo_dt.do_ops = &osd_obj_ea_ops;
188 l->lo_ops = &osd_lu_obj_ops;
189 init_rwsem(&mo->oo_sem);
190 init_rwsem(&mo->oo_ext_idx_sem);
191 spin_lock_init(&mo->oo_guard);
198 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
199 struct dentry *dentry, struct lustre_mdt_attrs *lma)
203 CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
204 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
205 info->oti_mdt_attrs_old, LMA_OLD_SIZE);
207 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
208 memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
210 lustre_lma_swab(lma);
211 /* Check LMA compatibility */
212 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
213 CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
214 "for fid = "DFID", ino = %lu\n",
215 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
216 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
217 PFID(&lma->lma_self_fid), inode->i_ino);
220 } else if (rc == 0) {
228 * retrieve object from backend ext fs.
230 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
231 struct osd_inode_id *id)
233 struct inode *inode = NULL;
235 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
237 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
238 id->oii_ino, PTR_ERR(inode));
239 } else if (id->oii_gen != OSD_OII_NOGEN &&
240 inode->i_generation != id->oii_gen) {
241 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
242 "i_generation = %u\n",
243 id->oii_ino, id->oii_gen, inode->i_generation);
245 inode = ERR_PTR(-ESTALE);
246 } else if (inode->i_nlink == 0) {
247 /* due to parallel readdir and unlink,
248 * we can have dead inode here. */
249 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
251 inode = ERR_PTR(-ESTALE);
252 } else if (is_bad_inode(inode)) {
253 CWARN("%.16s: bad inode: ino = %u\n",
254 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
256 inode = ERR_PTR(-ENOENT);
258 if (id->oii_gen == OSD_OII_NOGEN)
259 osd_id_gen(id, inode->i_ino, inode->i_generation);
261 /* Do not update file c/mtime in ldiskfs.
262 * NB: we don't have any lock to protect this because we don't
263 * have reference on osd_object now, but contention with
264 * another lookup + attr_set can't happen in the tiny window
265 * between if (...) and set S_NOCMTIME. */
266 if (!(inode->i_flags & S_NOCMTIME))
267 inode->i_flags |= S_NOCMTIME;
272 static struct inode *
273 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
274 struct osd_inode_id *id, struct lu_fid *fid)
276 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
280 inode = osd_iget(info, dev, id);
284 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
286 *fid = lma->lma_self_fid;
287 } else if (rc == -ENODATA) {
288 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
289 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
291 lu_igif_build(fid, inode->i_ino, inode->i_generation);
299 static struct inode *osd_iget_check(struct osd_thread_info *info,
300 struct osd_device *dev,
301 const struct lu_fid *fid,
302 struct osd_inode_id *id,
309 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
312 if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) {
313 CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
322 if (is_bad_inode(inode)) {
325 CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
333 if (id->oii_gen != OSD_OII_NOGEN &&
334 inode->i_generation != id->oii_gen) {
337 CDEBUG(D_INODE, "unmatched inode: ino = %u, "
338 "oii_gen = %u, i_generation = %u\n",
339 id->oii_ino, id->oii_gen, inode->i_generation);
347 if (inode->i_nlink == 0) {
350 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
360 struct osd_inode_id saved_id = *id;
362 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
364 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
365 /* XXX: There are some possible cases:
367 * Backup/restore caused the OI invalid.
369 * Someone unlinked the object but NOT removed
370 * the OI mapping, such as mount target device
371 * as ldiskfs, and modify something directly.
373 * Someone just removed the object between the
374 * former oi_lookup and the iget. It is normal.
375 * 4. Other failure cases.
377 * Generally, when the device is mounted, it will
378 * auto check whether the system is restored from
379 * file-level backup or not. We trust such detect
380 * to distinguish the 1st case from the 2nd case. */
382 if (!IS_ERR(inode) && inode->i_generation != 0 &&
383 inode->i_generation == id->oii_gen) {
386 __u32 level = D_LFSCK;
389 if (!thread_is_running(&dev->od_scrub.os_thread))
392 CDEBUG(level, "%s: the OI mapping for the FID "
393 DFID" become inconsistent, the given ID "
394 "%u/%u, the ID in OI mapping %u/%u\n",
395 osd_name(dev), PFID(fid),
396 saved_id.oii_ino, saved_id.oii_gen,
397 id->oii_ino, id->oii_ino);
401 if (id->oii_gen == OSD_OII_NOGEN)
402 osd_id_gen(id, inode->i_ino, inode->i_generation);
404 /* Do not update file c/mtime in ldiskfs.
405 * NB: we don't have any lock to protect this because we don't
406 * have reference on osd_object now, but contention with
407 * another lookup + attr_set can't happen in the tiny window
408 * between if (...) and set S_NOCMTIME. */
409 if (!(inode->i_flags & S_NOCMTIME))
410 inode->i_flags |= S_NOCMTIME;
427 * \retval +v: new filter_fid, does not contain self-fid
428 * \retval 0: filter_fid_old, contains self-fid
429 * \retval -v: other failure cases
431 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
432 struct dentry *dentry, struct lu_fid *fid)
434 struct filter_fid_old *ff = &info->oti_ff;
435 struct ost_id *ostid = &info->oti_ostid;
438 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
439 if (rc == sizeof(*ff)) {
441 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
442 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
443 /* XXX: should use real OST index in the future. LU-3569 */
444 ostid_to_fid(fid, ostid, 0);
445 } else if (rc == sizeof(struct filter_fid)) {
447 } else if (rc >= 0) {
454 static int osd_lma_self_repair(struct osd_thread_info *info,
455 struct osd_device *osd, struct inode *inode,
456 const struct lu_fid *fid, __u32 compat)
461 LASSERT(current->journal_info == NULL);
463 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
464 osd_dto_credits_noquota[DTO_XATTR_SET]);
467 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
472 rc = osd_ea_fid_set(info, inode, fid, compat, 0);
474 CWARN("%s: cannot self repair the LMA: rc = %d\n",
476 ldiskfs_journal_stop(jh);
480 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
482 struct osd_thread_info *info = osd_oti_get(env);
483 struct osd_device *osd = osd_obj2dev(obj);
484 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
485 struct inode *inode = obj->oo_inode;
486 struct dentry *dentry = &info->oti_obj_dentry;
487 struct lu_fid *fid = NULL;
488 const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
492 CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
493 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
494 info->oti_mdt_attrs_old, LMA_OLD_SIZE);
495 if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
496 fid = &lma->lma_self_fid;
497 rc = osd_get_idif(info, inode, dentry, fid);
498 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
499 /* For the given OST-object, if it has neither LMA nor
500 * FID in XATTR_NAME_FID, then the given FID (which is
501 * contained in the @obj, from client RPC for locating
502 * the OST-object) is trusted. We use it to generate
504 osd_lma_self_repair(info, osd, inode, rfid,
510 if (unlikely(rc == -ENODATA))
518 lustre_lma_swab(lma);
519 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
520 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
521 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
522 "fid = "DFID", ino = %lu\n", osd_name(osd),
523 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
524 PFID(rfid), inode->i_ino);
526 } else if (!(lma->lma_compat & LMAC_NOT_IN_OI)) {
527 fid = &lma->lma_self_fid;
531 if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
532 __u32 level = D_LFSCK;
534 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
535 struct ost_id *oi = &info->oti_ostid;
536 struct lu_fid *fid1 = &info->oti_fid3;
537 __u32 idx = fid_idif_ost_idx(rfid);
539 /* For old IDIF, the OST index is not part of the IDIF,
540 * Means that different OSTs may have the same IDIFs.
541 * Under such case, we need to make some compatible
542 * check to make sure to trigger OI scrub properly. */
543 if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
544 /* Given @rfid is new, LMA is old. */
545 fid_to_ostid(fid, oi);
546 ostid_to_fid(fid1, oi, idx);
547 if (lu_fid_eq(fid1, rfid)) {
548 if (osd->od_index_in_idif)
549 osd_lma_self_repair(info, osd,
559 if (!thread_is_running(&osd->od_scrub.os_thread))
562 CDEBUG(level, "%s: FID "DFID" != self_fid "DFID"\n",
563 osd_name(osd), PFID(rfid), PFID(fid));
569 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
570 const struct lu_fid *fid,
571 const struct lu_object_conf *conf)
573 struct osd_thread_info *info;
574 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
575 struct osd_device *dev;
576 struct osd_idmap_cache *oic;
577 struct osd_inode_id *id;
579 struct osd_scrub *scrub;
580 struct scrub_file *sf;
584 bool triggered = false;
587 LINVRNT(osd_invariant(obj));
588 LASSERT(obj->oo_inode == NULL);
589 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
592 scrub = &dev->od_scrub;
593 sf = &scrub->os_file;
594 info = osd_oti_get(env);
596 oic = &info->oti_cache;
598 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
601 /* For the object is created as locking anchor, or for the object to
602 * be created on disk. No need to osd_oi_lookup() at here because FID
603 * shouldn't never be re-used, if it's really a duplicate FID from
604 * unexpected reason, we should be able to detect it later by calling
605 * do_create->osd_oi_insert(). */
606 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
607 GOTO(out, result = 0);
609 /* Search order: 1. per-thread cache. */
610 if (lu_fid_eq(fid, &oic->oic_fid) &&
611 likely(oic->oic_dev == dev)) {
617 if (!list_empty(&scrub->os_inconsistent_items)) {
618 /* Search order: 2. OI scrub pending list. */
619 result = osd_oii_lookup(dev, fid, id);
624 /* Search order: 3. OI files. */
625 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
626 if (result == -ENOENT) {
627 if (!fid_is_norm(fid) ||
628 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
629 !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
631 GOTO(out, result = 0);
642 inode = osd_iget_check(info, dev, fid, id, in_oi);
644 result = PTR_ERR(inode);
645 if (result == -ENOENT || result == -ESTALE) {
647 fid_zero(&oic->oic_fid);
649 GOTO(out, result = -ENOENT);
650 } else if (result == -EREMCHG) {
654 fid_zero(&oic->oic_fid);
656 if (unlikely(triggered))
657 GOTO(out, result = saved);
660 if (thread_is_running(&scrub->os_thread)) {
661 result = -EINPROGRESS;
662 } else if (!dev->od_noscrub) {
663 /* Since we do not know the right OI mapping,
664 * we have to trigger OI scrub to scan the
666 result = osd_scrub_start(dev, SS_AUTO_FULL |
667 SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
668 CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI "
669 "scrub by RPC for "DFID", rc = %d [1]\n",
670 osd_name(dev), PFID(fid),result);
671 if (result == 0 || result == -EALREADY)
672 result = -EINPROGRESS;
677 /* We still have chance to get the valid inode: for the
678 * object which is referenced by remote name entry, the
679 * object on the local MDT will be linked under the dir
680 * of "/REMOTE_PARENT_DIR" with its FID string as name.
682 * We do not know whether the object for the given FID
683 * is referenced by some remote name entry or not, and
684 * especially for DNE II, a multiple-linked object may
685 * have many name entries reside on many MDTs.
687 * To simplify the operation, OSD will not distinguish
688 * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
689 * only happened for the RPC from other MDT during the
690 * OI scrub, or for the client side RPC with FID only,
691 * such as FID to path, or from old connected client. */
693 result = osd_lookup_in_remote_parent(info, dev,
706 obj->oo_inode = inode;
707 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
709 result = osd_check_lma(env, obj);
712 obj->oo_inode = NULL;
713 if (result == -EREMCHG) {
715 result = osd_oi_lookup(info, dev, fid, id,
718 fid_zero(&oic->oic_fid);
729 obj->oo_compat_dot_created = 1;
730 obj->oo_compat_dotdot_created = 1;
732 if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
733 GOTO(out, result = 0);
735 LASSERT(obj->oo_hl_head == NULL);
736 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
737 if (obj->oo_hl_head == NULL) {
738 obj->oo_inode = NULL;
740 GOTO(out, result = -ENOMEM);
742 GOTO(out, result = 0);
745 LINVRNT(osd_invariant(obj));
750 * Concurrency: shouldn't matter.
752 static void osd_object_init0(struct osd_object *obj)
754 LASSERT(obj->oo_inode != NULL);
755 obj->oo_dt.do_body_ops = &osd_body_ops;
756 obj->oo_dt.do_lu.lo_header->loh_attr |=
757 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
761 * Concurrency: no concurrent access is possible that early in object
764 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
765 const struct lu_object_conf *conf)
767 struct osd_object *obj = osd_obj(l);
770 LINVRNT(osd_invariant(obj));
772 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
773 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
774 l->lo_header->loh_attr |= LOHA_EXISTS;
778 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
779 obj->oo_dt.do_body_ops = &osd_body_ops_new;
780 if (result == 0 && obj->oo_inode != NULL)
781 osd_object_init0(obj);
783 LINVRNT(osd_invariant(obj));
788 * Concurrency: no concurrent access is possible that late in object
791 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
793 struct osd_object *obj = osd_obj(l);
795 LINVRNT(osd_invariant(obj));
797 dt_object_fini(&obj->oo_dt);
798 if (obj->oo_hl_head != NULL)
799 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
804 * Concurrency: no concurrent access is possible that late in object
807 static void osd_index_fini(struct osd_object *o)
809 struct iam_container *bag;
811 if (o->oo_dir != NULL) {
812 bag = &o->oo_dir->od_container;
813 if (o->oo_inode != NULL) {
814 if (bag->ic_object == o->oo_inode)
815 iam_container_fini(bag);
817 OBD_FREE_PTR(o->oo_dir);
823 * Concurrency: no concurrent access is possible that late in object
824 * life-cycle (for all existing callers, that is. New callers have to provide
825 * their own locking.)
827 static int osd_inode_unlinked(const struct inode *inode)
829 return inode->i_nlink == 0;
833 OSD_TXN_OI_DELETE_CREDITS = 20,
834 OSD_TXN_INODE_DELETE_CREDITS = 20
841 #if OSD_THANDLE_STATS
843 * Set time when the handle is allocated
845 static void osd_th_alloced(struct osd_thandle *oth)
847 oth->oth_alloced = cfs_time_current();
851 * Set time when the handle started
853 static void osd_th_started(struct osd_thandle *oth)
855 oth->oth_started = cfs_time_current();
859 * Helper function to convert time interval to microseconds packed in
862 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
866 cfs_duration_usec(cfs_time_sub(end, start), &val);
867 return val.tv_sec * 1000000 + val.tv_usec;
871 * Check whether the we deal with this handle for too long.
873 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
874 cfs_time_t alloced, cfs_time_t started,
877 cfs_time_t now = cfs_time_current();
879 LASSERT(dev != NULL);
881 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
882 interval_to_usec(alloced, started));
883 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
884 interval_to_usec(started, closed));
885 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
886 interval_to_usec(closed, now));
888 if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
889 CWARN("transaction handle %p was open for too long: "
891 "alloced "CFS_TIME_T" ,"
892 "started "CFS_TIME_T" ,"
893 "closed "CFS_TIME_T"\n",
894 oth, now, alloced, started, closed);
895 libcfs_debug_dumpstack(NULL);
899 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
901 cfs_time_t __closed = cfs_time_current(); \
902 cfs_time_t __alloced = oth->oth_alloced; \
903 cfs_time_t __started = oth->oth_started; \
906 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
909 #else /* OSD_THANDLE_STATS */
911 #define osd_th_alloced(h) do {} while(0)
912 #define osd_th_started(h) do {} while(0)
913 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
915 #endif /* OSD_THANDLE_STATS */
918 * Concurrency: doesn't access mutable data.
920 static int osd_param_is_not_sane(const struct osd_device *dev,
921 const struct thandle *th)
923 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
925 return oh->ot_credits > osd_transaction_size(dev);
929 * Concurrency: shouldn't matter.
931 static void osd_trans_commit_cb(struct super_block *sb,
932 struct ldiskfs_journal_cb_entry *jcb, int error)
934 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
935 struct thandle *th = &oh->ot_super;
936 struct lu_device *lud = &th->th_dev->dd_lu_dev;
937 struct dt_txn_commit_cb *dcb, *tmp;
939 LASSERT(oh->ot_handle == NULL);
942 CERROR("transaction @0x%p commit error: %d\n", th, error);
944 dt_txn_hook_commit(th);
946 /* call per-transaction callbacks if any */
947 list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
948 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
949 "commit callback entry: magic=%x name='%s'\n",
950 dcb->dcb_magic, dcb->dcb_name);
951 list_del_init(&dcb->dcb_linkage);
952 dcb->dcb_func(NULL, th, dcb, error);
955 lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
959 lu_context_exit(&th->th_ctx);
960 lu_context_fini(&th->th_ctx);
964 static struct thandle *osd_trans_create(const struct lu_env *env,
967 struct osd_thread_info *oti = osd_oti_get(env);
968 struct osd_iobuf *iobuf = &oti->oti_iobuf;
969 struct osd_thandle *oh;
973 /* on pending IO in this thread should left from prev. request */
974 LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
976 th = ERR_PTR(-ENOMEM);
977 OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
979 oh->ot_quota_trans = &oti->oti_quota_trans;
980 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
984 th->th_tags = LCT_TX_HANDLE;
986 atomic_set(&th->th_refc, 1);
987 th->th_alloc_size = sizeof(*oh);
988 oti->oti_dev = osd_dt_dev(d);
989 INIT_LIST_HEAD(&oh->ot_dcb_list);
992 memset(oti->oti_declare_ops, 0,
993 sizeof(oti->oti_declare_ops));
994 memset(oti->oti_declare_ops_rb, 0,
995 sizeof(oti->oti_declare_ops_rb));
996 memset(oti->oti_declare_ops_cred, 0,
997 sizeof(oti->oti_declare_ops_cred));
998 oti->oti_rollback = false;
1004 * Concurrency: shouldn't matter.
1006 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1009 struct osd_thread_info *oti = osd_oti_get(env);
1010 struct osd_device *dev = osd_dt_dev(d);
1012 struct osd_thandle *oh;
1017 LASSERT(current->journal_info == NULL);
1019 oh = container_of0(th, struct osd_thandle, ot_super);
1020 LASSERT(oh != NULL);
1021 LASSERT(oh->ot_handle == NULL);
1023 rc = dt_txn_hook_start(env, d, th);
1027 if (unlikely(osd_param_is_not_sane(dev, th))) {
1028 static unsigned long last_printed;
1029 static int last_credits;
1031 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1032 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1034 osd_journal(dev)->j_max_transaction_buffers);
1035 CWARN(" create: %u/%u, destroy: %u/%u\n",
1036 oti->oti_declare_ops[OSD_OT_CREATE],
1037 oti->oti_declare_ops_cred[OSD_OT_CREATE],
1038 oti->oti_declare_ops[OSD_OT_DESTROY],
1039 oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
1040 CWARN(" attr_set: %u/%u, xattr_set: %u/%u\n",
1041 oti->oti_declare_ops[OSD_OT_ATTR_SET],
1042 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1043 oti->oti_declare_ops[OSD_OT_XATTR_SET],
1044 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
1045 CWARN(" write: %u/%u, punch: %u/%u, quota %u/%u\n",
1046 oti->oti_declare_ops[OSD_OT_WRITE],
1047 oti->oti_declare_ops_cred[OSD_OT_WRITE],
1048 oti->oti_declare_ops[OSD_OT_PUNCH],
1049 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1050 oti->oti_declare_ops[OSD_OT_QUOTA],
1051 oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
1052 CWARN(" insert: %u/%u, delete: %u/%u\n",
1053 oti->oti_declare_ops[OSD_OT_INSERT],
1054 oti->oti_declare_ops_cred[OSD_OT_INSERT],
1055 oti->oti_declare_ops[OSD_OT_DELETE],
1056 oti->oti_declare_ops_cred[OSD_OT_DELETE]);
1057 CWARN(" ref_add: %u/%u, ref_del: %u/%u\n",
1058 oti->oti_declare_ops[OSD_OT_REF_ADD],
1059 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1060 oti->oti_declare_ops[OSD_OT_REF_DEL],
1061 oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
1063 if (last_credits != oh->ot_credits &&
1064 time_after(jiffies, last_printed +
1065 msecs_to_jiffies(60 * MSEC_PER_SEC))) {
1066 libcfs_debug_dumpstack(NULL);
1067 last_credits = oh->ot_credits;
1068 last_printed = jiffies;
1070 /* XXX Limit the credits to 'max_transaction_buffers', and
1071 * let the underlying filesystem to catch the error if
1072 * we really need so many credits.
1074 * This should be removed when we can calculate the
1075 * credits precisely. */
1076 oh->ot_credits = osd_transaction_size(dev);
1080 * XXX temporary stuff. Some abstraction layer should
1083 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1087 LASSERT(oti->oti_txns == 0);
1088 lu_context_init(&th->th_ctx, th->th_tags);
1089 lu_context_enter(&th->th_ctx);
1091 lu_device_get(&d->dd_lu_dev);
1092 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1103 static int osd_seq_exists(const struct lu_env *env,
1104 struct osd_device *osd, u64 seq)
1106 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1107 struct seq_server_site *ss = osd_seq_site(osd);
1111 LASSERT(ss != NULL);
1112 LASSERT(ss->ss_server_fld != NULL);
1114 rc = osd_fld_lookup(env, osd, seq, range);
1117 CERROR("%s: can't lookup FLD sequence "LPX64
1118 ": rc = %d\n", osd_name(osd), seq, rc);
1122 RETURN(ss->ss_node_id == range->lsr_index);
1126 * Concurrency: shouldn't matter.
1128 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1132 struct osd_thandle *oh;
1133 struct osd_thread_info *oti = osd_oti_get(env);
1134 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1135 struct qsd_instance *qsd = oti->oti_dev->od_quota_slave;
1136 struct lquota_trans *qtrans;
1139 oh = container_of0(th, struct osd_thandle, ot_super);
1141 qtrans = oh->ot_quota_trans;
1142 oh->ot_quota_trans = NULL;
1144 if (oh->ot_handle != NULL) {
1145 handle_t *hdl = oh->ot_handle;
1148 * add commit callback
1149 * notice we don't do this in osd_trans_start()
1150 * as underlying transaction can change during truncate
1152 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1155 LASSERT(oti->oti_txns == 1);
1157 rc = dt_txn_hook_stop(env, th);
1159 CERROR("Failure in transaction hook: %d\n", rc);
1161 /* hook functions might modify th_sync */
1162 hdl->h_sync = th->th_sync;
1164 oh->ot_handle = NULL;
1165 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
1166 rc = ldiskfs_journal_stop(hdl));
1168 CERROR("Failure to stop transaction: %d\n", rc);
1170 thandle_put(&oh->ot_super);
1173 /* inform the quota slave device that the transaction is stopping */
1174 qsd_op_end(env, qsd, qtrans);
1176 /* as we want IO to journal and data IO be concurrent, we don't block
1177 * awaiting data IO completion in osd_do_bio(), instead we wait here
1178 * once transaction is submitted to the journal. all reqular requests
1179 * don't do direct IO (except read/write), thus this wait_event becomes
1182 * IMPORTANT: we have to wait till any IO submited by the thread is
1183 * completed otherwise iobuf may be corrupted by different request
1185 wait_event(iobuf->dr_wait,
1186 atomic_read(&iobuf->dr_numreqs) == 0);
1187 osd_fini_iobuf(oti->oti_dev, iobuf);
1189 rc = iobuf->dr_error;
1194 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1196 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1199 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1200 LASSERT(&dcb->dcb_func != NULL);
1201 list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1207 * Called just before object is freed. Releases all resources except for
1208 * object itself (that is released by osd_object_free()).
1210 * Concurrency: no concurrent access is possible that late in object
1213 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1215 struct osd_object *obj = osd_obj(l);
1216 struct inode *inode = obj->oo_inode;
1218 LINVRNT(osd_invariant(obj));
1221 * If object is unlinked remove fid->ino mapping from object index.
1224 osd_index_fini(obj);
1225 if (inode != NULL) {
1226 struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave;
1227 qid_t uid = i_uid_read(inode);
1228 qid_t gid = i_gid_read(inode);
1231 obj->oo_inode = NULL;
1234 struct osd_thread_info *info = osd_oti_get(env);
1235 struct lquota_id_info *qi = &info->oti_qi;
1237 /* Release granted quota to master if necessary */
1238 qi->lqi_id.qid_uid = uid;
1239 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1241 qi->lqi_id.qid_uid = gid;
1242 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1248 * Concurrency: ->loo_object_release() is called under site spin-lock.
1250 static void osd_object_release(const struct lu_env *env,
1251 struct lu_object *l)
1256 * Concurrency: shouldn't matter.
1258 static int osd_object_print(const struct lu_env *env, void *cookie,
1259 lu_printer_t p, const struct lu_object *l)
1261 struct osd_object *o = osd_obj(l);
1262 struct iam_descr *d;
1264 if (o->oo_dir != NULL)
1265 d = o->oo_dir->od_container.ic_descr;
1268 return (*p)(env, cookie,
1269 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1271 o->oo_inode ? o->oo_inode->i_ino : 0UL,
1272 o->oo_inode ? o->oo_inode->i_generation : 0,
1273 d ? d->id_ops->id_name : "plain");
1276 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1279 * Concurrency: shouldn't matter.
1281 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1282 struct obd_statfs *sfs)
1284 struct osd_device *osd = osd_dt_dev(d);
1285 struct super_block *sb = osd_sb(osd);
1286 struct kstatfs *ksfs;
1289 if (unlikely(osd->od_mnt == NULL))
1290 return -EINPROGRESS;
1292 /* osd_lproc.c call this without env, allocate ksfs for that case */
1293 if (unlikely(env == NULL)) {
1294 OBD_ALLOC_PTR(ksfs);
1298 ksfs = &osd_oti_get(env)->oti_ksfs;
1301 spin_lock(&osd->od_osfs_lock);
1302 result = sb->s_op->statfs(sb->s_root, ksfs);
1303 if (likely(result == 0)) { /* N.B. statfs can't really fail */
1304 statfs_pack(sfs, ksfs);
1305 if (sb->s_flags & MS_RDONLY)
1306 sfs->os_state = OS_STATE_READONLY;
1309 spin_unlock(&osd->od_osfs_lock);
1311 if (unlikely(env == NULL))
1314 /* Reserve a small amount of space for local objects like last_rcvd,
1315 * llog, quota files, ... */
1316 if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1319 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1320 /** Take out metadata overhead for indirect blocks */
1321 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1328 * Estimate space needed for file creations. We assume the largest filename
1329 * which is 2^64 - 1, hence a filename of 20 chars.
1330 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1332 #ifdef __LDISKFS_DIR_REC_LEN
1333 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1335 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1339 * Concurrency: doesn't access mutable data.
1341 static void osd_conf_get(const struct lu_env *env,
1342 const struct dt_device *dev,
1343 struct dt_device_param *param)
1345 struct super_block *sb = osd_sb(osd_dt_dev(dev));
1349 * XXX should be taken from not-yet-existing fs abstraction layer.
1351 param->ddp_max_name_len = LDISKFS_NAME_LEN;
1352 param->ddp_max_nlink = LDISKFS_LINK_MAX;
1353 param->ddp_block_shift = sb->s_blocksize_bits;
1354 param->ddp_mount_type = LDD_MT_LDISKFS;
1355 param->ddp_maxbytes = sb->s_maxbytes;
1356 /* Overhead estimate should be fairly accurate, so we really take a tiny
1357 * error margin which also avoids fragmenting the filesystem too much */
1358 param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1359 /* inode are statically allocated, so per-inode space consumption
1360 * is the space consumed by the directory entry */
1361 param->ddp_inodespace = PER_OBJ_USAGE;
1362 /* per-fragment overhead to be used by the client code */
1363 param->ddp_grant_frag = 6 * LDISKFS_BLOCK_SIZE(sb);
1364 param->ddp_mntopts = 0;
1365 if (test_opt(sb, XATTR_USER))
1366 param->ddp_mntopts |= MNTOPT_USERXATTR;
1367 if (test_opt(sb, POSIX_ACL))
1368 param->ddp_mntopts |= MNTOPT_ACL;
1370 /* LOD might calculate the max stripe count based on max_ea_size,
1371 * so we need take account in the overhead as well,
1372 * xattr_header + magic + xattr_entry_head */
1373 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1374 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1376 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1377 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1378 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1382 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1386 * Concurrency: shouldn't matter.
1388 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1392 CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1394 rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1396 CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
1397 LUSTRE_OSD_LDISKFS_NAME, rc);
1403 * Start commit for OSD device.
1405 * An implementation of dt_commit_async method for OSD device.
1406 * Asychronously starts underlayng fs sync and thereby a transaction
1409 * \param env environment
1410 * \param d dt device
1412 * \see dt_device_operations
1414 static int osd_commit_async(const struct lu_env *env,
1415 struct dt_device *d)
1417 struct super_block *s = osd_sb(osd_dt_dev(d));
1420 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1421 RETURN(s->s_op->sync_fs(s, 0));
1425 * Concurrency: shouldn't matter.
1428 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1430 struct super_block *sb = osd_sb(osd_dt_dev(d));
1431 struct block_device *dev = sb->s_bdev;
1432 #ifdef HAVE_DEV_SET_RDONLY
1433 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1436 int rc = -EOPNOTSUPP;
1440 #ifdef HAVE_DEV_SET_RDONLY
1441 CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1443 if (jdev && (jdev != dev)) {
1444 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1446 dev_set_rdonly(jdev);
1448 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1449 dev_set_rdonly(dev);
1451 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1452 osd_dt_dev(d)->od_svname, (long)dev, rc);
1458 * Concurrency: serialization provided by callers.
1460 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1461 int mode, unsigned long timeout, __u32 alg,
1462 struct lustre_capa_key *keys)
1464 struct osd_device *dev = osd_dt_dev(d);
1467 dev->od_fl_capa = mode;
1468 dev->od_capa_timeout = timeout;
1469 dev->od_capa_alg = alg;
1470 dev->od_capa_keys = keys;
1475 * Note: we do not count into QUOTA here.
1476 * If we mount with --data_journal we may need more.
1478 const int osd_dto_credits_noquota[DTO_NR] = {
1481 * INDEX_EXTRA_TRANS_BLOCKS(8) +
1482 * SINGLEDATA_TRANS_BLOCKS(8)
1483 * XXX Note: maybe iam need more, since iam have more level than
1486 [DTO_INDEX_INSERT] = 16,
1489 * just modify a single entry, probably merge few within a block
1491 [DTO_INDEX_DELETE] = 1,
1495 [DTO_INDEX_UPDATE] = 16,
1497 * 4(inode, inode bits, groups, GDT)
1498 * notice: OI updates are counted separately with DTO_INDEX_INSERT
1500 [DTO_OBJECT_CREATE] = 4,
1502 * 4(inode, inode bits, groups, GDT)
1503 * notice: OI updates are counted separately with DTO_INDEX_DELETE
1505 [DTO_OBJECT_DELETE] = 4,
1507 * Attr set credits (inode)
1509 [DTO_ATTR_SET_BASE] = 1,
1511 * Xattr set. The same as xattr of EXT3.
1512 * DATA_TRANS_BLOCKS(14)
1513 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1514 * are also counted in. Do not know why?
1516 [DTO_XATTR_SET] = 14,
1518 * credits for inode change during write.
1520 [DTO_WRITE_BASE] = 3,
1522 * credits for single block write.
1524 [DTO_WRITE_BLOCK] = 14,
1526 * Attr set credits for chown.
1527 * This is extra credits for setattr, and it is null without quota
1529 [DTO_ATTR_SET_CHOWN] = 0
1532 static const struct dt_device_operations osd_dt_ops = {
1533 .dt_root_get = osd_root_get,
1534 .dt_statfs = osd_statfs,
1535 .dt_trans_create = osd_trans_create,
1536 .dt_trans_start = osd_trans_start,
1537 .dt_trans_stop = osd_trans_stop,
1538 .dt_trans_cb_add = osd_trans_cb_add,
1539 .dt_conf_get = osd_conf_get,
1540 .dt_sync = osd_sync,
1542 .dt_commit_async = osd_commit_async,
1543 .dt_init_capa_ctxt = osd_init_capa_ctxt,
1546 static void osd_object_read_lock(const struct lu_env *env,
1547 struct dt_object *dt, unsigned role)
1549 struct osd_object *obj = osd_dt_obj(dt);
1550 struct osd_thread_info *oti = osd_oti_get(env);
1552 LINVRNT(osd_invariant(obj));
1554 LASSERT(obj->oo_owner != env);
1555 down_read_nested(&obj->oo_sem, role);
1557 LASSERT(obj->oo_owner == NULL);
1561 static void osd_object_write_lock(const struct lu_env *env,
1562 struct dt_object *dt, unsigned role)
1564 struct osd_object *obj = osd_dt_obj(dt);
1565 struct osd_thread_info *oti = osd_oti_get(env);
1567 LINVRNT(osd_invariant(obj));
1569 LASSERT(obj->oo_owner != env);
1570 down_write_nested(&obj->oo_sem, role);
1572 LASSERT(obj->oo_owner == NULL);
1573 obj->oo_owner = env;
1577 static void osd_object_read_unlock(const struct lu_env *env,
1578 struct dt_object *dt)
1580 struct osd_object *obj = osd_dt_obj(dt);
1581 struct osd_thread_info *oti = osd_oti_get(env);
1583 LINVRNT(osd_invariant(obj));
1585 LASSERT(oti->oti_r_locks > 0);
1587 up_read(&obj->oo_sem);
1590 static void osd_object_write_unlock(const struct lu_env *env,
1591 struct dt_object *dt)
1593 struct osd_object *obj = osd_dt_obj(dt);
1594 struct osd_thread_info *oti = osd_oti_get(env);
1596 LINVRNT(osd_invariant(obj));
1598 LASSERT(obj->oo_owner == env);
1599 LASSERT(oti->oti_w_locks > 0);
1601 obj->oo_owner = NULL;
1602 up_write(&obj->oo_sem);
1605 static int osd_object_write_locked(const struct lu_env *env,
1606 struct dt_object *dt)
1608 struct osd_object *obj = osd_dt_obj(dt);
1610 LINVRNT(osd_invariant(obj));
1612 return obj->oo_owner == env;
1615 static int capa_is_sane(const struct lu_env *env,
1616 struct osd_device *dev,
1617 struct lustre_capa *capa,
1618 struct lustre_capa_key *keys)
1620 struct osd_thread_info *oti = osd_oti_get(env);
1621 struct lustre_capa *tcapa = &oti->oti_capa;
1622 struct obd_capa *oc;
1626 oc = capa_lookup(dev->od_capa_hash, capa, 0);
1628 if (capa_is_expired(oc)) {
1629 DEBUG_CAPA(D_ERROR, capa, "expired");
1636 if (capa_is_expired_sec(capa)) {
1637 DEBUG_CAPA(D_ERROR, capa, "expired");
1641 spin_lock(&capa_lock);
1642 for (i = 0; i < 2; i++) {
1643 if (keys[i].lk_keyid == capa->lc_keyid) {
1644 oti->oti_capa_key = keys[i];
1648 spin_unlock(&capa_lock);
1651 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1655 rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1659 if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1660 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1664 oc = capa_add(dev->od_capa_hash, capa);
1670 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1671 struct lustre_capa *capa, __u64 opc)
1673 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1674 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
1675 struct lu_capainfo *lci;
1678 if (!osd->od_fl_capa)
1681 if (capa == BYPASS_CAPA)
1684 lci = lu_capainfo_get(env);
1685 if (unlikely(lci == NULL))
1688 if (lci->lci_auth == LC_ID_NONE)
1692 CERROR("%s: no capability provided for FID "DFID": rc = %d\n",
1693 osd_name(osd), PFID(fid), -EACCES);
1697 if (!lu_fid_eq(fid, &capa->lc_fid)) {
1698 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1703 if (!capa_opc_supported(capa, opc)) {
1704 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1708 rc = capa_is_sane(env, osd, capa, osd->od_capa_keys);
1710 DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc);
1717 static struct timespec *osd_inode_time(const struct lu_env *env,
1718 struct inode *inode, __u64 seconds)
1720 struct osd_thread_info *oti = osd_oti_get(env);
1721 struct timespec *t = &oti->oti_time;
1723 t->tv_sec = seconds;
1725 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1730 static void osd_inode_getattr(const struct lu_env *env,
1731 struct inode *inode, struct lu_attr *attr)
1733 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1734 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1735 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1738 attr->la_atime = LTIME_S(inode->i_atime);
1739 attr->la_mtime = LTIME_S(inode->i_mtime);
1740 attr->la_ctime = LTIME_S(inode->i_ctime);
1741 attr->la_mode = inode->i_mode;
1742 attr->la_size = i_size_read(inode);
1743 attr->la_blocks = inode->i_blocks;
1744 attr->la_uid = i_uid_read(inode);
1745 attr->la_gid = i_gid_read(inode);
1746 attr->la_flags = LDISKFS_I(inode)->i_flags;
1747 attr->la_nlink = inode->i_nlink;
1748 attr->la_rdev = inode->i_rdev;
1749 attr->la_blksize = 1 << inode->i_blkbits;
1750 attr->la_blkbits = inode->i_blkbits;
1753 static int osd_attr_get(const struct lu_env *env,
1754 struct dt_object *dt,
1755 struct lu_attr *attr,
1756 struct lustre_capa *capa)
1758 struct osd_object *obj = osd_dt_obj(dt);
1760 if (!dt_object_exists(dt))
1763 LASSERT(!dt_object_remote(dt));
1764 LINVRNT(osd_invariant(obj));
1766 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1769 spin_lock(&obj->oo_guard);
1770 osd_inode_getattr(env, obj->oo_inode, attr);
1771 spin_unlock(&obj->oo_guard);
1775 static int osd_declare_attr_set(const struct lu_env *env,
1776 struct dt_object *dt,
1777 const struct lu_attr *attr,
1778 struct thandle *handle)
1780 struct osd_thandle *oh;
1781 struct osd_object *obj;
1782 struct osd_thread_info *info = osd_oti_get(env);
1783 struct lquota_id_info *qi = &info->oti_qi;
1791 LASSERT(dt != NULL);
1792 LASSERT(handle != NULL);
1794 obj = osd_dt_obj(dt);
1795 LASSERT(osd_invariant(obj));
1797 oh = container_of0(handle, struct osd_thandle, ot_super);
1798 LASSERT(oh->ot_handle == NULL);
1800 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1801 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1803 if (attr == NULL || obj->oo_inode == NULL)
1806 bspace = obj->oo_inode->i_blocks;
1807 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1808 bspace = toqb(bspace);
1810 /* Changing ownership is always preformed by super user, it should not
1813 * We still need to call the osd_declare_qid() to calculate the journal
1814 * credits for updating quota accounting files and to trigger quota
1815 * space adjustment once the operation is completed.*/
1816 if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
1818 uid = i_uid_read(obj->oo_inode);
1819 qi->lqi_type = USRQUOTA;
1820 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
1821 /* inode accounting */
1822 qi->lqi_is_blk = false;
1824 /* one more inode for the new uid ... */
1825 qi->lqi_id.qid_uid = attr->la_uid;
1827 /* Reserve credits for the new uid */
1828 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1829 if (rc == -EDQUOT || rc == -EINPROGRESS)
1834 /* and one less inode for the current uid */
1835 qi->lqi_id.qid_uid = uid;
1837 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1838 if (rc == -EDQUOT || rc == -EINPROGRESS)
1843 /* block accounting */
1844 qi->lqi_is_blk = true;
1846 /* more blocks for the new uid ... */
1847 qi->lqi_id.qid_uid = attr->la_uid;
1848 qi->lqi_space = bspace;
1850 * Credits for the new uid has been reserved, re-use "obj"
1851 * to save credit reservation.
1853 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1854 if (rc == -EDQUOT || rc == -EINPROGRESS)
1859 /* and finally less blocks for the current uid */
1860 qi->lqi_id.qid_uid = uid;
1861 qi->lqi_space = -bspace;
1862 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1863 if (rc == -EDQUOT || rc == -EINPROGRESS)
1869 gid = i_gid_read(obj->oo_inode);
1870 qi->lqi_type = GRPQUOTA;
1871 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
1873 /* inode accounting */
1874 qi->lqi_is_blk = false;
1876 /* one more inode for the new gid ... */
1877 qi->lqi_id.qid_gid = attr->la_gid;
1879 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1880 if (rc == -EDQUOT || rc == -EINPROGRESS)
1885 /* and one less inode for the current gid */
1886 qi->lqi_id.qid_gid = gid;
1888 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1889 if (rc == -EDQUOT || rc == -EINPROGRESS)
1894 /* block accounting */
1895 qi->lqi_is_blk = true;
1897 /* more blocks for the new gid ... */
1898 qi->lqi_id.qid_gid = attr->la_gid;
1899 qi->lqi_space = bspace;
1900 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1901 if (rc == -EDQUOT || rc == -EINPROGRESS)
1906 /* and finally less blocks for the current gid */
1907 qi->lqi_id.qid_gid = gid;
1908 qi->lqi_space = -bspace;
1909 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1910 if (rc == -EDQUOT || rc == -EINPROGRESS)
1919 static int osd_inode_setattr(const struct lu_env *env,
1920 struct inode *inode, const struct lu_attr *attr)
1922 __u64 bits = attr->la_valid;
1924 /* Only allow set size for regular file */
1925 if (!S_ISREG(inode->i_mode))
1926 bits &= ~(LA_SIZE | LA_BLOCKS);
1931 if (bits & LA_ATIME)
1932 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
1933 if (bits & LA_CTIME)
1934 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
1935 if (bits & LA_MTIME)
1936 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
1937 if (bits & LA_SIZE) {
1938 LDISKFS_I(inode)->i_disksize = attr->la_size;
1939 i_size_write(inode, attr->la_size);
1943 /* OSD should not change "i_blocks" which is used by quota.
1944 * "i_blocks" should be changed by ldiskfs only. */
1945 if (bits & LA_BLOCKS)
1946 inode->i_blocks = attr->la_blocks;
1949 inode->i_mode = (inode->i_mode & S_IFMT) |
1950 (attr->la_mode & ~S_IFMT);
1952 i_uid_write(inode, attr->la_uid);
1954 i_gid_write(inode, attr->la_gid);
1955 if (bits & LA_NLINK)
1956 set_nlink(inode, attr->la_nlink);
1958 inode->i_rdev = attr->la_rdev;
1960 if (bits & LA_FLAGS) {
1961 /* always keep S_NOCMTIME */
1962 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1968 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1970 if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1971 (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1975 ll_vfs_dq_init(inode);
1977 if (attr->la_valid & LA_UID)
1978 iattr.ia_valid |= ATTR_UID;
1979 if (attr->la_valid & LA_GID)
1980 iattr.ia_valid |= ATTR_GID;
1981 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1982 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1984 rc = ll_vfs_dq_transfer(inode, &iattr);
1986 CERROR("%s: quota transfer failed: rc = %d. Is quota "
1987 "enforcement enabled on the ldiskfs "
1988 "filesystem?\n", inode->i_sb->s_id, rc);
1995 static int osd_attr_set(const struct lu_env *env,
1996 struct dt_object *dt,
1997 const struct lu_attr *attr,
1998 struct thandle *handle,
1999 struct lustre_capa *capa)
2001 struct osd_object *obj = osd_dt_obj(dt);
2002 struct inode *inode;
2005 if (!dt_object_exists(dt))
2008 LASSERT(handle != NULL);
2009 LASSERT(!dt_object_remote(dt));
2010 LASSERT(osd_invariant(obj));
2012 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2015 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2017 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
2018 struct osd_thread_info *oti = osd_oti_get(env);
2019 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
2020 struct lu_fid *fid1 = &oti->oti_fid;
2021 struct osd_inode_id *id = &oti->oti_id;
2022 struct iam_path_descr *ipd;
2023 struct iam_container *bag;
2024 struct osd_thandle *oh;
2027 fid_cpu_to_be(fid1, fid0);
2028 memset(id, 1, sizeof(*id));
2029 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2030 fid0)->oi_dir.od_container;
2031 ipd = osd_idx_ipd_get(env, bag);
2032 if (unlikely(ipd == NULL))
2035 oh = container_of0(handle, struct osd_thandle, ot_super);
2036 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2037 (const struct iam_rec *)id, ipd);
2038 osd_ipd_put(env, bag, ipd);
2039 return(rc > 0 ? 0 : rc);
2042 inode = obj->oo_inode;
2044 rc = osd_quota_transfer(inode, attr);
2048 spin_lock(&obj->oo_guard);
2049 rc = osd_inode_setattr(env, inode, attr);
2050 spin_unlock(&obj->oo_guard);
2053 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2057 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2058 struct osd_object *obj,
2059 const char *name, const int namelen)
2061 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2064 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2065 umode_t mode, struct dt_allocation_hint *hint,
2069 struct osd_device *osd = osd_obj2dev(obj);
2070 struct osd_thandle *oth;
2071 struct dt_object *parent = NULL;
2072 struct inode *inode;
2074 LINVRNT(osd_invariant(obj));
2075 LASSERT(obj->oo_inode == NULL);
2076 LASSERT(obj->oo_hl_head == NULL);
2078 if (S_ISDIR(mode) && ldiskfs_pdo) {
2079 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2080 if (obj->oo_hl_head == NULL)
2084 oth = container_of(th, struct osd_thandle, ot_super);
2085 LASSERT(oth->ot_handle->h_transaction != NULL);
2087 if (hint && hint->dah_parent)
2088 parent = hint->dah_parent;
2090 inode = ldiskfs_create_inode(oth->ot_handle,
2091 parent ? osd_dt_obj(parent)->oo_inode :
2092 osd_sb(osd)->s_root->d_inode,
2094 if (!IS_ERR(inode)) {
2095 /* Do not update file c/mtime in ldiskfs.
2096 * NB: don't need any lock because no contention at this
2098 inode->i_flags |= S_NOCMTIME;
2100 /* For new created object, it must be consistent,
2101 * and it is unnecessary to scrub against it. */
2102 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2103 obj->oo_inode = inode;
2106 if (obj->oo_hl_head != NULL) {
2107 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2108 obj->oo_hl_head = NULL;
2110 result = PTR_ERR(inode);
2112 LINVRNT(osd_invariant(obj));
2120 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2121 struct lu_attr *attr,
2122 struct dt_allocation_hint *hint,
2123 struct dt_object_format *dof,
2127 struct osd_thandle *oth;
2128 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2130 LASSERT(S_ISDIR(attr->la_mode));
2132 oth = container_of(th, struct osd_thandle, ot_super);
2133 LASSERT(oth->ot_handle->h_transaction != NULL);
2134 result = osd_mkfile(info, obj, mode, hint, th);
2139 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2140 struct lu_attr *attr,
2141 struct dt_allocation_hint *hint,
2142 struct dt_object_format *dof,
2146 struct osd_thandle *oth;
2147 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2149 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2151 LASSERT(S_ISREG(attr->la_mode));
2153 oth = container_of(th, struct osd_thandle, ot_super);
2154 LASSERT(oth->ot_handle->h_transaction != NULL);
2156 result = osd_mkfile(info, obj, mode, hint, th);
2158 LASSERT(obj->oo_inode != NULL);
2159 if (feat->dif_flags & DT_IND_VARKEY)
2160 result = iam_lvar_create(obj->oo_inode,
2161 feat->dif_keysize_max,
2163 feat->dif_recsize_max,
2166 result = iam_lfix_create(obj->oo_inode,
2167 feat->dif_keysize_max,
2169 feat->dif_recsize_max,
2176 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2177 struct lu_attr *attr,
2178 struct dt_allocation_hint *hint,
2179 struct dt_object_format *dof,
2182 LASSERT(S_ISREG(attr->la_mode));
2183 return osd_mkfile(info, obj, (attr->la_mode &
2184 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2187 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2188 struct lu_attr *attr,
2189 struct dt_allocation_hint *hint,
2190 struct dt_object_format *dof,
2193 LASSERT(S_ISLNK(attr->la_mode));
2194 return osd_mkfile(info, obj, (attr->la_mode &
2195 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2198 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2199 struct lu_attr *attr,
2200 struct dt_allocation_hint *hint,
2201 struct dt_object_format *dof,
2204 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2207 LINVRNT(osd_invariant(obj));
2208 LASSERT(obj->oo_inode == NULL);
2209 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2210 S_ISFIFO(mode) || S_ISSOCK(mode));
2212 result = osd_mkfile(info, obj, mode, hint, th);
2214 LASSERT(obj->oo_inode != NULL);
2216 * This inode should be marked dirty for i_rdev. Currently
2217 * that is done in the osd_attr_init().
2219 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2222 LINVRNT(osd_invariant(obj));
2226 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2228 struct dt_allocation_hint *hint,
2229 struct dt_object_format *dof,
2232 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2234 osd_obj_type_f result;
2250 result = osd_mk_index;
2261 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2262 struct dt_object *parent, struct dt_object *child,
2267 ah->dah_parent = parent;
2268 ah->dah_mode = child_mode;
2271 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2272 struct lu_attr *attr, struct dt_object_format *dof)
2274 struct inode *inode = obj->oo_inode;
2275 __u64 valid = attr->la_valid;
2278 attr->la_valid &= ~(LA_TYPE | LA_MODE);
2280 if (dof->dof_type != DFT_NODE)
2281 attr->la_valid &= ~LA_RDEV;
2282 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2283 attr->la_valid &= ~LA_ATIME;
2284 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2285 attr->la_valid &= ~LA_CTIME;
2286 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2287 attr->la_valid &= ~LA_MTIME;
2289 result = osd_quota_transfer(inode, attr);
2293 if (attr->la_valid != 0) {
2294 result = osd_inode_setattr(info->oti_env, inode, attr);
2296 * The osd_inode_setattr() should always succeed here. The
2297 * only error that could be returned is EDQUOT when we are
2298 * trying to change the UID or GID of the inode. However, this
2299 * should not happen since quota enforcement is no longer
2300 * enabled on ldiskfs (lquota takes care of it).
2302 LASSERTF(result == 0, "%d\n", result);
2303 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2306 attr->la_valid = valid;
2310 * Helper function for osd_object_create()
2312 * \retval 0, on success
2314 static int __osd_object_create(struct osd_thread_info *info,
2315 struct osd_object *obj, struct lu_attr *attr,
2316 struct dt_allocation_hint *hint,
2317 struct dt_object_format *dof,
2323 /* we drop umask so that permissions we pass are not affected */
2324 umask = current->fs->umask;
2325 current->fs->umask = 0;
2327 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2330 osd_attr_init(info, obj, attr, dof);
2331 osd_object_init0(obj);
2333 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2334 unlock_new_inode(obj->oo_inode);
2337 /* restore previous umask value */
2338 current->fs->umask = umask;
2344 * Helper function for osd_object_create()
2346 * \retval 0, on success
2348 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2349 const struct lu_fid *fid, struct thandle *th)
2351 struct osd_thread_info *info = osd_oti_get(env);
2352 struct osd_inode_id *id = &info->oti_id;
2353 struct osd_device *osd = osd_obj2dev(obj);
2354 struct osd_thandle *oh;
2356 LASSERT(obj->oo_inode != NULL);
2358 oh = container_of0(th, struct osd_thandle, ot_super);
2359 LASSERT(oh->ot_handle);
2361 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2362 return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2365 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2366 u64 seq, struct lu_seq_range *range)
2368 struct seq_server_site *ss = osd_seq_site(osd);
2370 if (fid_seq_is_idif(seq)) {
2371 fld_range_set_ost(range);
2372 range->lsr_index = idif_ost_idx(seq);
2376 if (!fid_seq_in_fldb(seq)) {
2377 fld_range_set_mdt(range);
2379 /* FIXME: If ss is NULL, it suppose not get lsr_index
2381 range->lsr_index = ss->ss_node_id;
2385 LASSERT(ss != NULL);
2386 fld_range_set_any(range);
2387 /* OSD will only do local fld lookup */
2388 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2392 * Concurrency: no external locking is necessary.
2394 static int osd_declare_object_create(const struct lu_env *env,
2395 struct dt_object *dt,
2396 struct lu_attr *attr,
2397 struct dt_allocation_hint *hint,
2398 struct dt_object_format *dof,
2399 struct thandle *handle)
2401 struct osd_thandle *oh;
2405 LASSERT(handle != NULL);
2407 oh = container_of0(handle, struct osd_thandle, ot_super);
2408 LASSERT(oh->ot_handle == NULL);
2410 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2411 osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2412 /* Reuse idle OI block may cause additional one OI block
2414 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2415 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2417 /* If this is directory, then we expect . and .. to be inserted as
2418 * well. The one directory block always needs to be created for the
2419 * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2420 * block), there is no danger of needing a tree for the first block.
2422 if (attr && S_ISDIR(attr->la_mode)) {
2423 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2424 osd_dto_credits_noquota[DTO_WRITE_BASE]);
2425 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2431 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2432 osd_dt_obj(dt), false, NULL, false);
2439 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2440 struct lu_attr *attr,
2441 struct dt_allocation_hint *hint,
2442 struct dt_object_format *dof, struct thandle *th)
2444 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2445 struct osd_object *obj = osd_dt_obj(dt);
2446 struct osd_thread_info *info = osd_oti_get(env);
2450 if (dt_object_exists(dt))
2453 LINVRNT(osd_invariant(obj));
2454 LASSERT(!dt_object_remote(dt));
2455 LASSERT(osd_write_locked(env, obj));
2456 LASSERT(th != NULL);
2458 if (unlikely(fid_is_acct(fid)))
2459 /* Quota files can't be created from the kernel any more,
2460 * 'tune2fs -O quota' will take care of creating them */
2463 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2464 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2466 result = __osd_object_create(info, obj, attr, hint, dof, th);
2468 result = __osd_oi_insert(env, obj, fid, th);
2470 LASSERT(ergo(result == 0,
2471 dt_object_exists(dt) && !dt_object_remote(dt)));
2473 LASSERT(osd_invariant(obj));
2478 * Called to destroy on-disk representation of the object
2480 * Concurrency: must be locked
2482 static int osd_declare_object_destroy(const struct lu_env *env,
2483 struct dt_object *dt,
2486 struct osd_object *obj = osd_dt_obj(dt);
2487 struct inode *inode = obj->oo_inode;
2488 struct osd_thandle *oh;
2492 oh = container_of0(th, struct osd_thandle, ot_super);
2493 LASSERT(oh->ot_handle == NULL);
2496 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2497 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2498 /* Recycle idle OI leaf may cause additional three OI blocks
2500 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2501 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2502 /* one less inode */
2503 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2504 -1, oh, obj, false, NULL, false);
2507 /* data to be truncated */
2508 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2509 0, oh, obj, true, NULL, false);
2513 static int osd_object_destroy(const struct lu_env *env,
2514 struct dt_object *dt,
2517 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2518 struct osd_object *obj = osd_dt_obj(dt);
2519 struct inode *inode = obj->oo_inode;
2520 struct osd_device *osd = osd_obj2dev(obj);
2521 struct osd_thandle *oh;
2525 oh = container_of0(th, struct osd_thandle, ot_super);
2526 LASSERT(oh->ot_handle);
2528 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2530 if (unlikely(fid_is_acct(fid)))
2533 if (S_ISDIR(inode->i_mode)) {
2534 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2535 inode->i_nlink == 2);
2536 /* it will check/delete the inode from remote parent,
2537 * how to optimize it? unlink performance impaction XXX */
2538 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2539 if (result != 0 && result != -ENOENT) {
2540 CERROR("%s: delete inode "DFID": rc = %d\n",
2541 osd_name(osd), PFID(fid), result);
2543 spin_lock(&obj->oo_guard);
2545 spin_unlock(&obj->oo_guard);
2546 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2549 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2551 result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2554 /* XXX: add to ext3 orphan list */
2555 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2557 /* not needed in the cache anymore */
2558 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2564 * Put the fid into lustre_mdt_attrs, and then place the structure
2565 * inode's ea. This fid should not be altered during the life time
2568 * \retval +ve, on success
2569 * \retval -ve, on error
2571 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2573 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2574 const struct lu_fid *fid, __u32 compat, __u32 incompat)
2576 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2580 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2583 lustre_lma_init(lma, fid, compat, incompat);
2584 lustre_lma_swab(lma);
2586 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2588 /* LMA may already exist, but we need to check that all the
2589 * desired compat/incompat flags have been added. */
2590 if (unlikely(rc == -EEXIST)) {
2591 if (compat == 0 && incompat == 0)
2594 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2595 XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2600 lustre_lma_swab(lma);
2601 if (!(~lma->lma_compat & compat) &&
2602 !(~lma->lma_incompat & incompat))
2605 lma->lma_compat |= compat;
2606 lma->lma_incompat |= incompat;
2607 lustre_lma_swab(lma);
2608 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2609 sizeof(*lma), XATTR_REPLACE);
2616 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2617 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2618 * To have compatilibility with 1.8 ldiskfs driver we need to have
2619 * magic number at start of fid data.
2620 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2623 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2624 const struct lu_fid *fid)
2626 if (!fid_is_namespace_visible(fid) ||
2627 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2628 param->edp_magic = 0;
2632 param->edp_magic = LDISKFS_LUFID_MAGIC;
2633 param->edp_len = sizeof(struct lu_fid) + 1;
2634 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2638 * Try to read the fid from inode ea into dt_rec.
2640 * \param fid object fid.
2642 * \retval 0 on success
2644 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2645 __u32 ino, struct lu_fid *fid,
2646 struct osd_inode_id *id)
2648 struct osd_thread_info *info = osd_oti_get(env);
2649 struct inode *inode;
2652 osd_id_gen(id, ino, OSD_OII_NOGEN);
2653 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2655 RETURN(PTR_ERR(inode));
2661 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2663 struct inode *parent_dir,
2664 const struct lu_fid *dot_fid,
2665 const struct lu_fid *dot_dot_fid,
2666 struct osd_thandle *oth)
2668 struct ldiskfs_dentry_param *dot_ldp;
2669 struct ldiskfs_dentry_param *dot_dot_ldp;
2671 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2672 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2674 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2675 dot_ldp->edp_magic = 0;
2676 return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2677 dir, dot_ldp, dot_dot_ldp);
2681 * Create an local agent inode for remote entry
2683 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2684 struct osd_device *osd,
2685 struct osd_object *pobj,
2686 const struct lu_fid *fid,
2690 struct osd_thread_info *info = osd_oti_get(env);
2691 struct inode *local;
2692 struct osd_thandle *oh;
2697 oh = container_of(th, struct osd_thandle, ot_super);
2698 LASSERT(oh->ot_handle->h_transaction != NULL);
2700 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2701 if (IS_ERR(local)) {
2702 CERROR("%s: create local error %d\n", osd_name(osd),
2703 (int)PTR_ERR(local));
2707 /* Set special LMA flag for local agent inode */
2708 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2710 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2711 osd_name(osd), PFID(fid), rc);
2712 RETURN(ERR_PTR(rc));
2718 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2719 lu_object_fid(&pobj->oo_dt.do_lu),
2722 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2723 osd_name(osd), PFID(fid), rc);
2724 RETURN(ERR_PTR(rc));
2731 * Delete local agent inode for remote entry
2733 static int osd_delete_local_agent_inode(const struct lu_env *env,
2734 struct osd_device *osd,
2735 const struct lu_fid *fid,
2736 __u32 ino, struct osd_thandle *oh)
2738 struct osd_thread_info *oti = osd_oti_get(env);
2739 struct osd_inode_id *id = &oti->oti_id;
2740 struct inode *inode;
2743 id->oii_ino = le32_to_cpu(ino);
2744 id->oii_gen = OSD_OII_NOGEN;
2745 inode = osd_iget(oti, osd, id);
2746 if (IS_ERR(inode)) {
2747 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2748 PFID(fid), id->oii_ino, id->oii_gen);
2749 RETURN(PTR_ERR(inode));
2753 mark_inode_dirty(inode);
2754 CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2755 osd_name(osd), PFID(fid), inode->i_ino);
2761 * OSD layer object create function for interoperability mode (b11826).
2762 * This is mostly similar to osd_object_create(). Only difference being, fid is
2763 * inserted into inode ea here.
2765 * \retval 0, on success
2766 * \retval -ve, on error
2768 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2769 struct lu_attr *attr,
2770 struct dt_allocation_hint *hint,
2771 struct dt_object_format *dof,
2774 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2775 struct osd_object *obj = osd_dt_obj(dt);
2776 struct osd_thread_info *info = osd_oti_get(env);
2781 if (dt_object_exists(dt))
2784 LASSERT(osd_invariant(obj));
2785 LASSERT(!dt_object_remote(dt));
2786 LASSERT(osd_write_locked(env, obj));
2787 LASSERT(th != NULL);
2789 if (unlikely(fid_is_acct(fid)))
2790 /* Quota files can't be created from the kernel any more,
2791 * 'tune2fs -O quota' will take care of creating them */
2794 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2795 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2797 result = __osd_object_create(info, obj, attr, hint, dof, th);
2799 if (fid_is_idif(fid) &&
2800 !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
2801 struct lu_fid *tfid = &info->oti_fid;
2802 struct ost_id *oi = &info->oti_ostid;
2804 fid_to_ostid(fid, oi);
2805 ostid_to_fid(tfid, oi, 0);
2806 result = osd_ea_fid_set(info, obj->oo_inode, tfid,
2807 LMAC_FID_ON_OST, 0);
2809 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2810 fid_is_on_ost(info, osd_obj2dev(obj),
2811 fid, OI_CHECK_FLD) ?
2812 LMAC_FID_ON_OST : 0, 0);
2817 result = __osd_oi_insert(env, obj, fid, th);
2819 LASSERT(ergo(result == 0,
2820 dt_object_exists(dt) && !dt_object_remote(dt)));
2821 LINVRNT(osd_invariant(obj));
2825 static int osd_declare_object_ref_add(const struct lu_env *env,
2826 struct dt_object *dt,
2827 struct thandle *handle)
2829 struct osd_thandle *oh;
2831 /* it's possible that object doesn't exist yet */
2832 LASSERT(handle != NULL);
2834 oh = container_of0(handle, struct osd_thandle, ot_super);
2835 LASSERT(oh->ot_handle == NULL);
2837 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2838 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2844 * Concurrency: @dt is write locked.
2846 static int osd_object_ref_add(const struct lu_env *env,
2847 struct dt_object *dt, struct thandle *th)
2849 struct osd_object *obj = osd_dt_obj(dt);
2850 struct inode *inode = obj->oo_inode;
2851 struct osd_thandle *oh;
2854 if (!dt_object_exists(dt))
2857 LINVRNT(osd_invariant(obj));
2858 LASSERT(!dt_object_remote(dt));
2859 LASSERT(osd_write_locked(env, obj));
2860 LASSERT(th != NULL);
2862 oh = container_of0(th, struct osd_thandle, ot_super);
2863 LASSERT(oh->ot_handle != NULL);
2865 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2867 CDEBUG(D_INODE, DFID" increase nlink %d\n",
2868 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2870 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2871 * (65000) subdirectories by storing "1" in i_nlink if the link count
2872 * would otherwise overflow. Directory tranversal tools understand
2873 * that (st_nlink == 1) indicates that the filesystem dose not track
2874 * hard links count on the directory, and will not abort subdirectory
2875 * scanning early once (st_nlink - 2) subdirs have been found.
2877 * This also has to properly handle the case of inodes with nlink == 0
2878 * in case they are being linked into the PENDING directory
2880 spin_lock(&obj->oo_guard);
2881 if (unlikely(inode->i_nlink == 0))
2882 /* inc_nlink from 0 may cause WARN_ON */
2883 set_nlink(inode, 1);
2885 ldiskfs_inc_count(oh->ot_handle, inode);
2886 if (!S_ISDIR(inode->i_mode))
2887 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2889 spin_unlock(&obj->oo_guard);
2891 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2892 LINVRNT(osd_invariant(obj));
2897 static int osd_declare_object_ref_del(const struct lu_env *env,
2898 struct dt_object *dt,
2899 struct thandle *handle)
2901 struct osd_thandle *oh;
2903 LASSERT(!dt_object_remote(dt));
2904 LASSERT(handle != NULL);
2906 oh = container_of0(handle, struct osd_thandle, ot_super);
2907 LASSERT(oh->ot_handle == NULL);
2909 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2910 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2916 * Concurrency: @dt is write locked.
2918 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2921 struct osd_object *obj = osd_dt_obj(dt);
2922 struct inode *inode = obj->oo_inode;
2923 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
2924 struct osd_thandle *oh;
2926 if (!dt_object_exists(dt))
2929 LINVRNT(osd_invariant(obj));
2930 LASSERT(!dt_object_remote(dt));
2931 LASSERT(osd_write_locked(env, obj));
2932 LASSERT(th != NULL);
2934 oh = container_of0(th, struct osd_thandle, ot_super);
2935 LASSERT(oh->ot_handle != NULL);
2937 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2939 spin_lock(&obj->oo_guard);
2940 /* That can be result of upgrade from old Lustre version and
2941 * applied only to local files. Just skip this ref_del call.
2942 * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2943 if (inode->i_nlink == 0) {
2944 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2945 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2946 ", maybe an upgraded file? (LU-3915)\n",
2947 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2948 spin_unlock(&obj->oo_guard);
2952 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2953 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2955 ldiskfs_dec_count(oh->ot_handle, inode);
2956 spin_unlock(&obj->oo_guard);
2958 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2959 LINVRNT(osd_invariant(obj));
2965 * Get the 64-bit version for an inode.
2967 static int osd_object_version_get(const struct lu_env *env,
2968 struct dt_object *dt, dt_obj_version_t *ver)
2970 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2972 CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2973 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2974 *ver = LDISKFS_I(inode)->i_fs_version;
2979 * Concurrency: @dt is read locked.
2981 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2982 struct lu_buf *buf, const char *name,
2983 struct lustre_capa *capa)
2985 struct osd_object *obj = osd_dt_obj(dt);
2986 struct inode *inode = obj->oo_inode;
2987 struct osd_thread_info *info = osd_oti_get(env);
2988 struct dentry *dentry = &info->oti_obj_dentry;
2990 /* version get is not real XATTR but uses xattr API */
2991 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2992 /* for version we are just using xattr API but change inode
2994 if (buf->lb_len == 0)
2995 return sizeof(dt_obj_version_t);
2997 if (buf->lb_len < sizeof(dt_obj_version_t))
3000 osd_object_version_get(env, dt, buf->lb_buf);
3002 return sizeof(dt_obj_version_t);
3005 if (!dt_object_exists(dt))
3008 LASSERT(!dt_object_remote(dt));
3009 LASSERT(inode->i_op != NULL);
3010 LASSERT(inode->i_op->getxattr != NULL);
3012 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3015 return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
3019 static int osd_declare_xattr_set(const struct lu_env *env,
3020 struct dt_object *dt,
3021 const struct lu_buf *buf, const char *name,
3022 int fl, struct thandle *handle)
3024 struct osd_thandle *oh;
3026 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3028 LASSERT(handle != NULL);
3030 oh = container_of0(handle, struct osd_thandle, ot_super);
3031 LASSERT(oh->ot_handle == NULL);
3033 /* optimistic optimization: LMA is set first and usually fit inode */
3034 if (strcmp(name, XATTR_NAME_LMA) == 0) {
3035 if (dt_object_exists(dt))
3039 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3042 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
3043 if (buf && buf->lb_len > sb->s_blocksize) {
3044 credits *= (buf->lb_len + sb->s_blocksize - 1) >>
3045 sb->s_blocksize_bits;
3048 * xattr set may involve inode quota change, reserve credits for
3049 * dquot_initialize()
3051 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3054 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3060 * Set the 64-bit version for object
3062 static void osd_object_version_set(const struct lu_env *env,
3063 struct dt_object *dt,
3064 dt_obj_version_t *new_version)
3066 struct inode *inode = osd_dt_obj(dt)->oo_inode;
3068 CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
3069 *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3071 LDISKFS_I(inode)->i_fs_version = *new_version;
3072 /** Version is set after all inode operations are finished,
3073 * so we should mark it dirty here */
3074 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3078 * Concurrency: @dt is write locked.
3080 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3081 const struct lu_buf *buf, const char *name, int fl,
3082 struct thandle *handle, struct lustre_capa *capa)
3084 struct osd_object *obj = osd_dt_obj(dt);
3085 struct inode *inode = obj->oo_inode;
3086 struct osd_thread_info *info = osd_oti_get(env);
3090 LASSERT(handle != NULL);
3092 /* version set is not real XATTR */
3093 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3094 /* for version we are just using xattr API but change inode
3096 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3097 osd_object_version_set(env, dt, buf->lb_buf);
3098 return sizeof(dt_obj_version_t);
3101 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3104 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
3105 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3107 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3108 if (fl & LU_XATTR_REPLACE)
3109 fs_flags |= XATTR_REPLACE;
3111 if (fl & LU_XATTR_CREATE)
3112 fs_flags |= XATTR_CREATE;
3114 if (strcmp(name, XATTR_NAME_LMV) == 0) {
3115 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3118 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3122 lma->lma_incompat |= LMAI_STRIPED;
3123 lustre_lma_swab(lma);
3124 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3125 sizeof(*lma), XATTR_REPLACE);
3130 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
3131 strcmp(name, XATTR_NAME_LINK) == 0)
3134 return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3139 * Concurrency: @dt is read locked.
3141 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3142 const struct lu_buf *buf, struct lustre_capa *capa)
3144 struct osd_object *obj = osd_dt_obj(dt);
3145 struct inode *inode = obj->oo_inode;
3146 struct osd_thread_info *info = osd_oti_get(env);
3147 struct dentry *dentry = &info->oti_obj_dentry;
3149 if (!dt_object_exists(dt))
3152 LASSERT(!dt_object_remote(dt));
3153 LASSERT(inode->i_op != NULL);
3154 LASSERT(inode->i_op->listxattr != NULL);
3156 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3159 dentry->d_inode = inode;
3160 dentry->d_sb = inode->i_sb;
3161 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3164 static int osd_declare_xattr_del(const struct lu_env *env,
3165 struct dt_object *dt, const char *name,
3166 struct thandle *handle)
3168 struct osd_thandle *oh;
3169 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3171 LASSERT(!dt_object_remote(dt));
3172 LASSERT(handle != NULL);
3174 oh = container_of0(handle, struct osd_thandle, ot_super);
3175 LASSERT(oh->ot_handle == NULL);
3177 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3178 osd_dto_credits_noquota[DTO_XATTR_SET]);
3180 * xattr del may involve inode quota change, reserve credits for
3181 * dquot_initialize()
3183 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3189 * Concurrency: @dt is write locked.
3191 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3192 const char *name, struct thandle *handle,
3193 struct lustre_capa *capa)
3195 struct osd_object *obj = osd_dt_obj(dt);
3196 struct inode *inode = obj->oo_inode;
3197 struct osd_thread_info *info = osd_oti_get(env);
3198 struct dentry *dentry = &info->oti_obj_dentry;
3201 if (!dt_object_exists(dt))
3204 LASSERT(!dt_object_remote(dt));
3205 LASSERT(inode->i_op != NULL);
3206 LASSERT(inode->i_op->removexattr != NULL);
3207 LASSERT(handle != NULL);
3209 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3212 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3214 ll_vfs_dq_init(inode);
3215 dentry->d_inode = inode;
3216 dentry->d_sb = inode->i_sb;
3217 rc = inode->i_op->removexattr(dentry, name);
3221 static struct obd_capa *osd_capa_get(const struct lu_env *env,
3222 struct dt_object *dt,
3223 struct lustre_capa *old, __u64 opc)
3225 struct osd_thread_info *info = osd_oti_get(env);
3226 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3227 struct osd_object *obj = osd_dt_obj(dt);
3228 struct osd_device *osd = osd_obj2dev(obj);
3229 struct lustre_capa_key *key = &info->oti_capa_key;
3230 struct lustre_capa *capa = &info->oti_capa;
3231 struct obd_capa *oc;
3232 struct lu_capainfo *lci;
3236 if (!osd->od_fl_capa)
3237 RETURN(ERR_PTR(-ENOENT));
3239 if (!dt_object_exists(dt))
3240 RETURN(ERR_PTR(-ENOENT));
3242 LASSERT(!dt_object_remote(dt));
3243 LINVRNT(osd_invariant(obj));
3245 /* renewal sanity check */
3246 if (old && osd_object_auth(env, dt, old, opc))
3247 RETURN(ERR_PTR(-EACCES));
3249 lci = lu_capainfo_get(env);
3250 if (unlikely(lci == NULL))
3251 RETURN(ERR_PTR(-ENOENT));
3253 switch (lci->lci_auth) {
3257 capa->lc_uid = i_uid_read(obj->oo_inode);
3258 capa->lc_gid = i_gid_read(obj->oo_inode);
3259 capa->lc_flags = LC_ID_PLAIN;
3261 case LC_ID_CONVERT: {
3264 s[0] = i_uid_read(obj->oo_inode);
3265 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
3266 s[2] = i_uid_read(obj->oo_inode);
3267 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
3268 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
3270 RETURN(ERR_PTR(rc));
3272 capa->lc_uid = ((__u64)d[1] << 32) | d[0];
3273 capa->lc_gid = ((__u64)d[3] << 32) | d[2];
3274 capa->lc_flags = LC_ID_CONVERT;
3278 RETURN(ERR_PTR(-EINVAL));
3281 capa->lc_fid = *fid;
3283 capa->lc_flags |= osd->od_capa_alg << 24;
3284 capa->lc_timeout = osd->od_capa_timeout;
3285 capa->lc_expiry = 0;
3287 oc = capa_lookup(osd->od_capa_hash, capa, 1);
3289 LASSERT(!capa_is_expired(oc));
3293 spin_lock(&capa_lock);
3294 *key = osd->od_capa_keys[1];
3295 spin_unlock(&capa_lock);
3297 capa->lc_keyid = key->lk_keyid;
3298 capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout;
3300 rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
3302 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
3303 RETURN(ERR_PTR(rc));
3306 oc = capa_add(osd->od_capa_hash, capa);
3310 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3311 __u64 start, __u64 end)
3313 struct osd_object *obj = osd_dt_obj(dt);
3314 struct inode *inode = obj->oo_inode;
3315 struct osd_thread_info *info = osd_oti_get(env);
3316 struct dentry *dentry = &info->oti_obj_dentry;
3317 struct file *file = &info->oti_file;
3322 dentry->d_inode = inode;
3323 dentry->d_sb = inode->i_sb;
3324 file->f_dentry = dentry;
3325 file->f_mapping = inode->i_mapping;
3326 file->f_op = inode->i_fop;
3327 set_file_inode(file, inode);
3329 rc = ll_vfs_fsync_range(file, start, end, 0);
3334 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
3337 struct osd_object *obj = osd_dt_obj(dt);
3340 *data = (void *)obj->oo_inode;
3348 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3349 const struct dt_index_features *feat)
3351 struct iam_descr *descr;
3353 if (osd_object_is_root(o))
3354 return feat == &dt_directory_features;
3356 LASSERT(o->oo_dir != NULL);
3358 descr = o->oo_dir->od_container.ic_descr;
3359 if (feat == &dt_directory_features) {
3360 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3366 feat->dif_keysize_min <= descr->id_key_size &&
3367 descr->id_key_size <= feat->dif_keysize_max &&
3368 feat->dif_recsize_min <= descr->id_rec_size &&
3369 descr->id_rec_size <= feat->dif_recsize_max &&
3370 !(feat->dif_flags & (DT_IND_VARKEY |
3371 DT_IND_VARREC | DT_IND_NONUNQ)) &&
3372 ergo(feat->dif_flags & DT_IND_UPDATE,
3373 1 /* XXX check that object (and file system) is
3378 static int osd_iam_container_init(const struct lu_env *env,
3379 struct osd_object *obj,
3380 struct osd_directory *dir)
3382 struct iam_container *bag = &dir->od_container;
3385 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3389 result = iam_container_setup(bag);
3391 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3393 iam_container_fini(bag);
3400 * Concurrency: no external locking is necessary.
3402 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3403 const struct dt_index_features *feat)
3407 struct osd_object *obj = osd_dt_obj(dt);
3409 LINVRNT(osd_invariant(obj));
3411 if (osd_object_is_root(obj)) {
3412 dt->do_index_ops = &osd_index_ea_ops;
3414 } else if (feat == &dt_directory_features) {
3415 dt->do_index_ops = &osd_index_ea_ops;
3416 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3421 } else if (unlikely(feat == &dt_otable_features)) {
3422 dt->do_index_ops = &osd_otable_ops;
3424 } else if (unlikely(feat == &dt_acct_features)) {
3425 dt->do_index_ops = &osd_acct_index_ops;
3428 } else if (!osd_has_index(obj)) {
3429 struct osd_directory *dir;
3434 spin_lock(&obj->oo_guard);
3435 if (obj->oo_dir == NULL)
3439 * Concurrent thread allocated container data.
3442 spin_unlock(&obj->oo_guard);
3444 * Now, that we have container data, serialize its
3447 down_write(&obj->oo_ext_idx_sem);
3449 * recheck under lock.
3451 if (!osd_has_index(obj))
3452 result = osd_iam_container_init(env, obj,
3456 up_write(&obj->oo_ext_idx_sem);
3464 if (result == 0 && skip_iam == 0) {
3465 if (!osd_iam_index_probe(env, obj, feat))
3468 LINVRNT(osd_invariant(obj));
3470 if (result == 0 && is_quota_glb_feat(feat) &&
3471 fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3472 result = osd_quota_migration(env, dt, feat);
3477 static int osd_otable_it_attr_get(const struct lu_env *env,
3478 struct dt_object *dt,
3479 struct lu_attr *attr,
3480 struct lustre_capa *capa)
3486 static const struct dt_object_operations osd_obj_ops = {
3487 .do_read_lock = osd_object_read_lock,
3488 .do_write_lock = osd_object_write_lock,
3489 .do_read_unlock = osd_object_read_unlock,
3490 .do_write_unlock = osd_object_write_unlock,
3491 .do_write_locked = osd_object_write_locked,
3492 .do_attr_get = osd_attr_get,
3493 .do_declare_attr_set = osd_declare_attr_set,
3494 .do_attr_set = osd_attr_set,
3495 .do_ah_init = osd_ah_init,
3496 .do_declare_create = osd_declare_object_create,
3497 .do_create = osd_object_create,
3498 .do_declare_destroy = osd_declare_object_destroy,
3499 .do_destroy = osd_object_destroy,
3500 .do_index_try = osd_index_try,
3501 .do_declare_ref_add = osd_declare_object_ref_add,
3502 .do_ref_add = osd_object_ref_add,
3503 .do_declare_ref_del = osd_declare_object_ref_del,
3504 .do_ref_del = osd_object_ref_del,
3505 .do_xattr_get = osd_xattr_get,
3506 .do_declare_xattr_set = osd_declare_xattr_set,
3507 .do_xattr_set = osd_xattr_set,
3508 .do_declare_xattr_del = osd_declare_xattr_del,
3509 .do_xattr_del = osd_xattr_del,
3510 .do_xattr_list = osd_xattr_list,
3511 .do_capa_get = osd_capa_get,
3512 .do_object_sync = osd_object_sync,
3513 .do_data_get = osd_data_get,
3517 * dt_object_operations for interoperability mode
3518 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3520 static const struct dt_object_operations osd_obj_ea_ops = {
3521 .do_read_lock = osd_object_read_lock,
3522 .do_write_lock = osd_object_write_lock,
3523 .do_read_unlock = osd_object_read_unlock,
3524 .do_write_unlock = osd_object_write_unlock,
3525 .do_write_locked = osd_object_write_locked,
3526 .do_attr_get = osd_attr_get,
3527 .do_declare_attr_set = osd_declare_attr_set,
3528 .do_attr_set = osd_attr_set,
3529 .do_ah_init = osd_ah_init,
3530 .do_declare_create = osd_declare_object_create,
3531 .do_create = osd_object_ea_create,
3532 .do_declare_destroy = osd_declare_object_destroy,
3533 .do_destroy = osd_object_destroy,
3534 .do_index_try = osd_index_try,
3535 .do_declare_ref_add = osd_declare_object_ref_add,
3536 .do_ref_add = osd_object_ref_add,
3537 .do_declare_ref_del = osd_declare_object_ref_del,
3538 .do_ref_del = osd_object_ref_del,
3539 .do_xattr_get = osd_xattr_get,
3540 .do_declare_xattr_set = osd_declare_xattr_set,
3541 .do_xattr_set = osd_xattr_set,
3542 .do_declare_xattr_del = osd_declare_xattr_del,
3543 .do_xattr_del = osd_xattr_del,
3544 .do_xattr_list = osd_xattr_list,
3545 .do_capa_get = osd_capa_get,
3546 .do_object_sync = osd_object_sync,
3547 .do_data_get = osd_data_get,
3550 static const struct dt_object_operations osd_obj_otable_it_ops = {
3551 .do_attr_get = osd_otable_it_attr_get,
3552 .do_index_try = osd_index_try,
3555 static int osd_index_declare_iam_delete(const struct lu_env *env,
3556 struct dt_object *dt,
3557 const struct dt_key *key,
3558 struct thandle *handle)
3560 struct osd_thandle *oh;
3562 oh = container_of0(handle, struct osd_thandle, ot_super);
3563 LASSERT(oh->ot_handle == NULL);
3565 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3566 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3572 * delete a (key, value) pair from index \a dt specified by \a key
3574 * \param dt osd index object
3575 * \param key key for index
3576 * \param rec record reference
3577 * \param handle transaction handler
3580 * \retval -ve failure
3582 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3583 const struct dt_key *key,
3584 struct thandle *handle,
3585 struct lustre_capa *capa)
3587 struct osd_thread_info *oti = osd_oti_get(env);
3588 struct osd_object *obj = osd_dt_obj(dt);
3589 struct osd_thandle *oh;
3590 struct iam_path_descr *ipd;
3591 struct iam_container *bag = &obj->oo_dir->od_container;
3595 if (!dt_object_exists(dt))
3598 LINVRNT(osd_invariant(obj));
3599 LASSERT(!dt_object_remote(dt));
3600 LASSERT(bag->ic_object == obj->oo_inode);
3601 LASSERT(handle != NULL);
3603 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3606 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3608 ipd = osd_idx_ipd_get(env, bag);
3609 if (unlikely(ipd == NULL))
3612 oh = container_of0(handle, struct osd_thandle, ot_super);
3613 LASSERT(oh->ot_handle != NULL);
3614 LASSERT(oh->ot_handle->h_transaction != NULL);
3616 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3617 /* swab quota uid/gid provided by caller */
3618 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3619 key = (const struct dt_key *)&oti->oti_quota_id;
3622 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3623 osd_ipd_put(env, bag, ipd);
3624 LINVRNT(osd_invariant(obj));
3628 static int osd_index_declare_ea_delete(const struct lu_env *env,
3629 struct dt_object *dt,
3630 const struct dt_key *key,
3631 struct thandle *handle)
3633 struct osd_thandle *oh;
3634 struct inode *inode;
3638 LASSERT(!dt_object_remote(dt));
3639 LASSERT(handle != NULL);
3641 oh = container_of0(handle, struct osd_thandle, ot_super);
3642 LASSERT(oh->ot_handle == NULL);
3644 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3645 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3647 inode = osd_dt_obj(dt)->oo_inode;
3650 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3651 0, oh, osd_dt_obj(dt), true, NULL, false);
3655 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3658 struct osd_fid_pack *rec;
3661 if (de->file_type & LDISKFS_DIRENT_LUFID) {
3662 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3663 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3668 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3669 const struct lu_fid *fid)
3671 struct seq_server_site *ss = osd_seq_site(osd);
3674 /* FID seqs not in FLDB, must be local seq */
3675 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3678 /* If FLD is not being initialized yet, it only happens during the
3679 * initialization, likely during mgs initialization, and we assume
3680 * this is local FID. */
3681 if (ss == NULL || ss->ss_server_fld == NULL)
3684 /* Only check the local FLDB here */
3685 if (osd_seq_exists(env, osd, fid_seq(fid)))
3692 * Index delete function for interoperability mode (b11826).
3693 * It will remove the directory entry added by osd_index_ea_insert().
3694 * This entry is needed to maintain name->fid mapping.
3696 * \param key, key i.e. file entry to be deleted
3698 * \retval 0, on success
3699 * \retval -ve, on error
3701 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3702 const struct dt_key *key, struct thandle *handle,
3703 struct lustre_capa *capa)
3705 struct osd_object *obj = osd_dt_obj(dt);
3706 struct inode *dir = obj->oo_inode;
3707 struct dentry *dentry;
3708 struct osd_thandle *oh;
3709 struct ldiskfs_dir_entry_2 *de = NULL;
3710 struct buffer_head *bh;
3711 struct htree_lock *hlock = NULL;
3712 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
3713 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3717 if (!dt_object_exists(dt))
3720 LINVRNT(osd_invariant(obj));
3721 LASSERT(!dt_object_remote(dt));
3722 LASSERT(handle != NULL);
3724 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3726 oh = container_of(handle, struct osd_thandle, ot_super);
3727 LASSERT(oh->ot_handle != NULL);
3728 LASSERT(oh->ot_handle->h_transaction != NULL);
3730 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3733 ll_vfs_dq_init(dir);
3734 dentry = osd_child_dentry_get(env, obj,
3735 (char *)key, strlen((char *)key));
3737 if (obj->oo_hl_head != NULL) {
3738 hlock = osd_oti_get(env)->oti_hlock;
3739 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3740 dir, LDISKFS_HLOCK_DEL);
3742 down_write(&obj->oo_ext_idx_sem);
3745 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3749 /* If this is not the ".." entry, it might be a remote DNE
3750 * entry and we need to check if the FID is for a remote
3751 * MDT. If the FID is not in the directory entry (e.g.
3752 * upgraded 1.8 filesystem without dirdata enabled) then
3753 * we need to get the FID from the LMA. For a remote directory
3754 * there HAS to be an LMA, it cannot be an IGIF inode in this
3757 * Delete the entry before the agent inode in order to
3758 * simplify error handling. At worst an error after deleting
3759 * the entry first might leak the agent inode afterward. The
3760 * reverse would need filesystem abort in case of error deleting
3761 * the entry after the agent had been removed, or leave a
3762 * dangling entry pointing at a random inode. */
3763 if (strcmp((char *)key, dotdot) != 0) {
3764 LASSERT(de != NULL);
3765 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3766 /* If Fid is not in dentry, try to get it from LMA */
3767 if (rc == -ENODATA) {
3768 struct osd_inode_id *id;
3769 struct inode *inode;
3771 /* Before trying to get fid from the inode,
3772 * check whether the inode is valid.
3774 * If the inode has been deleted, do not go
3775 * ahead to do osd_ea_fid_get, which will set
3776 * the inode to bad inode, which might cause
3777 * the inode to be deleted uncorrectly */
3778 inode = ldiskfs_iget(osd_sb(osd),
3779 le32_to_cpu(de->inode));
3780 if (IS_ERR(inode)) {
3781 CDEBUG(D_INODE, "%s: "DFID"get inode"
3782 "error.\n", osd_name(osd),
3784 rc = PTR_ERR(inode);
3786 if (likely(inode->i_nlink != 0)) {
3787 id = &osd_oti_get(env)->oti_id;
3788 rc = osd_ea_fid_get(env, obj,
3789 le32_to_cpu(de->inode),
3792 CDEBUG(D_INFO, "%s: %u "DFID
3795 le32_to_cpu(de->inode),
3803 unlikely(osd_remote_fid(env, osd, fid)))
3804 /* Need to delete agent inode */
3805 ino = le32_to_cpu(de->inode);
3807 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3809 if (rc == 0 && unlikely(ino != 0)) {
3810 rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3813 CERROR("%s: del local inode "DFID": rc = %d\n",
3814 osd_name(osd), PFID(fid), rc);
3820 ldiskfs_htree_unlock(hlock);
3822 up_write(&obj->oo_ext_idx_sem);
3827 /* For inode on the remote MDT, .. will point to
3828 * /Agent directory, Check whether it needs to delete
3829 * from agent directory */
3830 if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3831 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3833 if (rc != 0 && rc != -ENOENT) {
3834 CERROR("%s: delete agent inode "DFID": rc = %d\n",
3835 osd_name(osd), PFID(fid), rc);
3845 LASSERT(osd_invariant(obj));
3850 * Lookup index for \a key and copy record to \a rec.
3852 * \param dt osd index object
3853 * \param key key for index
3854 * \param rec record reference
3856 * \retval +ve success : exact mach
3857 * \retval 0 return record with key not greater than \a key
3858 * \retval -ve failure
3860 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3861 struct dt_rec *rec, const struct dt_key *key,
3862 struct lustre_capa *capa)
3864 struct osd_object *obj = osd_dt_obj(dt);
3865 struct iam_path_descr *ipd;
3866 struct iam_container *bag = &obj->oo_dir->od_container;
3867 struct osd_thread_info *oti = osd_oti_get(env);
3868 struct iam_iterator *it = &oti->oti_idx_it;
3869 struct iam_rec *iam_rec;
3873 if (!dt_object_exists(dt))
3876 LASSERT(osd_invariant(obj));
3877 LASSERT(!dt_object_remote(dt));
3878 LASSERT(bag->ic_object == obj->oo_inode);
3880 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3883 ipd = osd_idx_ipd_get(env, bag);
3887 /* got ipd now we can start iterator. */
3888 iam_it_init(it, bag, 0, ipd);
3890 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3891 /* swab quota uid/gid provided by caller */
3892 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3893 key = (const struct dt_key *)&oti->oti_quota_id;
3896 rc = iam_it_get(it, (struct iam_key *)key);
3898 if (S_ISDIR(obj->oo_inode->i_mode))
3899 iam_rec = (struct iam_rec *)oti->oti_ldp;
3901 iam_rec = (struct iam_rec *) rec;
3903 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3905 if (S_ISDIR(obj->oo_inode->i_mode))
3906 osd_fid_unpack((struct lu_fid *) rec,
3907 (struct osd_fid_pack *)iam_rec);
3908 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3909 osd_quota_unpack(obj, rec);
3914 osd_ipd_put(env, bag, ipd);
3916 LINVRNT(osd_invariant(obj));
3921 static int osd_index_declare_iam_insert(const struct lu_env *env,
3922 struct dt_object *dt,
3923 const struct dt_rec *rec,
3924 const struct dt_key *key,
3925 struct thandle *handle)
3927 struct osd_thandle *oh;
3929 LASSERT(handle != NULL);
3931 oh = container_of0(handle, struct osd_thandle, ot_super);
3932 LASSERT(oh->ot_handle == NULL);
3934 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3935 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3941 * Inserts (key, value) pair in \a dt index object.
3943 * \param dt osd index object
3944 * \param key key for index
3945 * \param rec record reference
3946 * \param th transaction handler
3949 * \retval -ve failure
3951 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3952 const struct dt_rec *rec,
3953 const struct dt_key *key, struct thandle *th,
3954 struct lustre_capa *capa, int ignore_quota)
3956 struct osd_object *obj = osd_dt_obj(dt);
3957 struct iam_path_descr *ipd;
3958 struct osd_thandle *oh;
3959 struct iam_container *bag;
3960 struct osd_thread_info *oti = osd_oti_get(env);
3961 struct iam_rec *iam_rec;
3965 if (!dt_object_exists(dt))
3968 LINVRNT(osd_invariant(obj));
3969 LASSERT(!dt_object_remote(dt));
3971 bag = &obj->oo_dir->od_container;
3972 LASSERT(bag->ic_object == obj->oo_inode);
3973 LASSERT(th != NULL);
3975 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3978 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3980 ipd = osd_idx_ipd_get(env, bag);
3981 if (unlikely(ipd == NULL))
3984 oh = container_of0(th, struct osd_thandle, ot_super);
3985 LASSERT(oh->ot_handle != NULL);
3986 LASSERT(oh->ot_handle->h_transaction != NULL);
3987 if (S_ISDIR(obj->oo_inode->i_mode)) {
3988 iam_rec = (struct iam_rec *)oti->oti_ldp;
3989 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3990 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3991 /* pack quota uid/gid */
3992 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3993 key = (const struct dt_key *)&oti->oti_quota_id;
3994 /* pack quota record */
3995 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3996 iam_rec = (struct iam_rec *)rec;
3998 iam_rec = (struct iam_rec *)rec;
4001 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
4003 osd_ipd_put(env, bag, ipd);
4004 LINVRNT(osd_invariant(obj));
4009 * Calls ldiskfs_add_entry() to add directory entry
4010 * into the directory. This is required for
4011 * interoperability mode (b11826)
4013 * \retval 0, on success
4014 * \retval -ve, on error
4016 static int __osd_ea_add_rec(struct osd_thread_info *info,
4017 struct osd_object *pobj, struct inode *cinode,
4018 const char *name, const struct lu_fid *fid,
4019 struct htree_lock *hlock, struct thandle *th)
4021 struct ldiskfs_dentry_param *ldp;
4022 struct dentry *child;
4023 struct osd_thandle *oth;
4026 oth = container_of(th, struct osd_thandle, ot_super);
4027 LASSERT(oth->ot_handle != NULL);
4028 LASSERT(oth->ot_handle->h_transaction != NULL);
4029 LASSERT(pobj->oo_inode);
4031 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
4032 if (unlikely(pobj->oo_inode ==
4033 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
4036 osd_get_ldiskfs_dirent_param(ldp, fid);
4037 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
4038 child->d_fsdata = (void *)ldp;
4039 ll_vfs_dq_init(pobj->oo_inode);
4040 rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
4041 if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
4042 struct ldiskfs_dir_entry_2 *de;
4043 struct buffer_head *bh;
4046 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
4049 rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
4052 if (S_ISDIR(cinode->i_mode))
4053 de->file_type = LDISKFS_DIRENT_LUFID |
4054 LDISKFS_FT_REG_FILE;
4056 de->file_type = LDISKFS_DIRENT_LUFID |
4058 ldiskfs_handle_dirty_metadata(oth->ot_handle,
4069 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
4070 * into the directory.Also sets flags into osd object to
4071 * indicate dot and dotdot are created. This is required for
4072 * interoperability mode (b11826)
4074 * \param dir directory for dot and dotdot fixup.
4075 * \param obj child object for linking
4077 * \retval 0, on success
4078 * \retval -ve, on error
4080 static int osd_add_dot_dotdot(struct osd_thread_info *info,
4081 struct osd_object *dir,
4082 struct inode *parent_dir, const char *name,
4083 const struct lu_fid *dot_fid,
4084 const struct lu_fid *dot_dot_fid,
4087 struct inode *inode = dir->oo_inode;
4088 struct osd_thandle *oth;
4091 oth = container_of(th, struct osd_thandle, ot_super);
4092 LASSERT(oth->ot_handle->h_transaction != NULL);
4093 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4095 if (strcmp(name, dot) == 0) {
4096 if (dir->oo_compat_dot_created) {
4099 LASSERT(inode == parent_dir);
4100 dir->oo_compat_dot_created = 1;
4103 } else if (strcmp(name, dotdot) == 0) {
4104 if (!dir->oo_compat_dot_created)
4106 /* in case of rename, dotdot is already created */
4107 if (dir->oo_compat_dotdot_created) {
4108 return __osd_ea_add_rec(info, dir, parent_dir, name,
4109 dot_dot_fid, NULL, th);
4112 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
4113 struct lu_fid tfid = *dot_dot_fid;
4116 result = osd_add_dot_dotdot_internal(info,
4117 dir->oo_inode, parent_dir, dot_fid,
4120 result = osd_add_dot_dotdot_internal(info,
4121 dir->oo_inode, parent_dir, dot_fid,
4126 dir->oo_compat_dotdot_created = 1;
4134 * It will call the appropriate osd_add* function and return the
4135 * value, return by respective functions.
4137 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4138 struct inode *cinode, const char *name,
4139 const struct lu_fid *fid, struct thandle *th)
4141 struct osd_thread_info *info = osd_oti_get(env);
4142 struct htree_lock *hlock;
4145 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4147 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4149 if (hlock != NULL) {
4150 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4153 down_write(&pobj->oo_ext_idx_sem);
4156 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4157 lu_object_fid(&pobj->oo_dt.do_lu),
4160 if (hlock != NULL) {
4161 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4162 pobj->oo_inode, LDISKFS_HLOCK_ADD);
4164 down_write(&pobj->oo_ext_idx_sem);
4167 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4168 struct lu_fid *tfid = &info->oti_fid;
4172 rc = __osd_ea_add_rec(info, pobj, cinode, name,
4175 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4180 ldiskfs_htree_unlock(hlock);
4182 up_write(&pobj->oo_ext_idx_sem);
4188 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4189 struct osd_idmap_cache *oic)
4191 struct osd_scrub *scrub = &dev->od_scrub;
4192 struct lu_fid *fid = &oic->oic_fid;
4193 struct osd_inode_id *id = &oti->oti_id;
4198 if (!fid_is_norm(fid) && !fid_is_igif(fid))
4201 if (scrub->os_pos_current > id->oii_ino)
4205 rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
4206 if (rc != 0 && rc != -ENOENT)
4209 if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
4212 if (thread_is_running(&scrub->os_thread)) {
4213 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
4214 /* There is race condition between osd_oi_lookup and OI scrub.
4215 * The OI scrub finished just after osd_oi_lookup() failure.
4216 * Under such case, it is unnecessary to trigger OI scrub again,
4217 * but try to call osd_oi_lookup() again. */
4218 if (unlikely(rc == -EAGAIN))
4224 if (!dev->od_noscrub && ++once == 1) {
4225 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
4227 CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI scrub by RPC "
4228 "for "DFID", rc = %d [2]\n",
4229 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4231 if (rc == 0 || rc == -EALREADY)
4238 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4239 struct osd_device *dev,
4240 struct osd_idmap_cache *oic,
4241 struct lu_fid *fid, __u32 ino)
4243 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
4244 struct inode *inode;
4247 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4248 inode = osd_iget(oti, dev, &oic->oic_lid);
4249 if (IS_ERR(inode)) {
4250 fid_zero(&oic->oic_fid);
4251 return PTR_ERR(inode);
4254 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4257 fid_zero(&oic->oic_fid);
4259 *fid = oic->oic_fid = lma->lma_self_fid;
4263 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4264 struct osd_inode_id *id, const struct lu_fid *fid)
4266 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4267 id->oii_ino, id->oii_gen, info);
4268 info->oti_cache.oic_lid = *id;
4269 info->oti_cache.oic_fid = *fid;
4270 info->oti_cache.oic_dev = osd;
4274 * Get parent FID from the linkEA.
4276 * For a directory which parent resides on remote MDT, to satisfy the
4277 * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
4278 * the other hand, to make the lookup(..) on the directory can return
4279 * the real parent FID, we append the real parent FID after its ".."
4280 * name entry in the /REMOTE_PARENT_DIR.
4282 * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
4283 * backup. So after the restore, we cannot get the right parent FID from
4284 * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
4285 * we have stored the real parent FID in the directory object's linkEA,
4286 * we can parse the linkEA for the real parent FID.
4288 * \param[in] env pointer to the thread context
4289 * \param[in] obj pointer to the object to be handled
4290 * \param[out]fid pointer to the buffer to hold the parent FID
4292 * \retval 0 for getting the real parent FID successfully
4293 * \retval negative error number on failure
4295 static int osd_get_pfid_from_linkea(const struct lu_env *env,
4296 struct osd_object *obj,
4299 struct osd_thread_info *oti = osd_oti_get(env);
4300 struct lu_buf *buf = &oti->oti_big_buf;
4301 struct dentry *dentry = &oti->oti_obj_dentry;
4302 struct inode *inode = obj->oo_inode;
4303 struct linkea_data ldata = { NULL };
4308 if (!S_ISDIR(inode->i_mode))
4312 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4313 buf->lb_buf, buf->lb_len);
4314 if (rc == -ERANGE) {
4315 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4318 lu_buf_realloc(buf, rc);
4319 if (buf->lb_buf == NULL)
4326 if (unlikely(rc == 0))
4332 if (unlikely(buf->lb_buf == NULL)) {
4333 lu_buf_realloc(buf, rc);
4334 if (buf->lb_buf == NULL)
4341 rc = linkea_init(&ldata);
4343 linkea_first_entry(&ldata);
4344 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
4351 * Calls ->lookup() to find dentry. From dentry get inode and
4352 * read inode's ea to get fid. This is required for interoperability
4355 * \retval 0, on success
4356 * \retval -ve, on error
4358 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4359 struct dt_rec *rec, const struct dt_key *key)
4361 struct inode *dir = obj->oo_inode;
4362 struct dentry *dentry;
4363 struct ldiskfs_dir_entry_2 *de;
4364 struct buffer_head *bh;
4365 struct lu_fid *fid = (struct lu_fid *) rec;
4366 struct htree_lock *hlock = NULL;
4371 LASSERT(dir->i_op != NULL);
4372 LASSERT(dir->i_op->lookup != NULL);
4374 dentry = osd_child_dentry_get(env, obj,
4375 (char *)key, strlen((char *)key));
4377 if (obj->oo_hl_head != NULL) {
4378 hlock = osd_oti_get(env)->oti_hlock;
4379 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4380 dir, LDISKFS_HLOCK_LOOKUP);
4382 down_read(&obj->oo_ext_idx_sem);
4385 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4387 struct osd_thread_info *oti = osd_oti_get(env);
4388 struct osd_inode_id *id = &oti->oti_id;
4389 struct osd_idmap_cache *oic = &oti->oti_cache;
4390 struct osd_device *dev = osd_obj2dev(obj);
4392 ino = le32_to_cpu(de->inode);
4393 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4395 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4399 rc = osd_get_fid_from_dentry(de, rec);
4401 /* done with de, release bh */
4404 if (unlikely(ino == osd_remote_parent_ino(dev)))
4405 /* If the parent is on remote MDT, and there
4406 * is no FID-in-dirent, then we have to get
4407 * the parent FID from the linkEA. */
4408 rc = osd_get_pfid_from_linkea(env, obj, fid);
4410 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4412 osd_id_gen(id, ino, OSD_OII_NOGEN);
4416 fid_zero(&oic->oic_fid);
4420 if (osd_remote_fid(env, dev, fid))
4423 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
4424 osd_consistency_check(oti, dev, oic);
4433 ldiskfs_htree_unlock(hlock);
4435 up_read(&obj->oo_ext_idx_sem);
4440 * Find the osd object for given fid.
4442 * \param fid need to find the osd object having this fid
4444 * \retval osd_object on success
4445 * \retval -ve on error
4447 static struct osd_object *osd_object_find(const struct lu_env *env,
4448 struct dt_object *dt,
4449 const struct lu_fid *fid)
4451 struct lu_device *ludev = dt->do_lu.lo_dev;
4452 struct osd_object *child = NULL;
4453 struct lu_object *luch;
4454 struct lu_object *lo;
4457 * at this point topdev might not exist yet
4458 * (i.e. MGS is preparing profiles). so we can
4459 * not rely on topdev and instead lookup with
4460 * our device passed as topdev. this can't work
4461 * if the object isn't cached yet (as osd doesn't
4462 * allocate lu_header). IOW, the object must be
4463 * in the cache, otherwise lu_object_alloc() crashes
4466 luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4467 ludev : ludev->ld_site->ls_top_dev,
4469 if (!IS_ERR(luch)) {
4470 if (lu_object_exists(luch)) {
4471 lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4473 child = osd_obj(lo);
4475 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4476 "lu_object can't be located"
4477 DFID"\n", PFID(fid));
4479 if (child == NULL) {
4480 lu_object_put(env, luch);
4481 CERROR("Unable to get osd_object\n");
4482 child = ERR_PTR(-ENOENT);
4485 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4486 "lu_object does not exists "DFID"\n",
4488 lu_object_put(env, luch);
4489 child = ERR_PTR(-ENOENT);
4492 child = ERR_CAST(luch);
4499 * Put the osd object once done with it.
4501 * \param obj osd object that needs to be put
4503 static inline void osd_object_put(const struct lu_env *env,
4504 struct osd_object *obj)
4506 lu_object_put(env, &obj->oo_dt.do_lu);
4509 static int osd_index_declare_ea_insert(const struct lu_env *env,
4510 struct dt_object *dt,
4511 const struct dt_rec *rec,
4512 const struct dt_key *key,
4513 struct thandle *handle)
4515 struct osd_thandle *oh;
4516 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4517 struct lu_fid *fid = (struct lu_fid *)rec;
4521 LASSERT(!dt_object_remote(dt));
4522 LASSERT(handle != NULL);
4524 oh = container_of0(handle, struct osd_thandle, ot_super);
4525 LASSERT(oh->ot_handle == NULL);
4527 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4528 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4530 if (osd_dt_obj(dt)->oo_inode != NULL) {
4531 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4533 /* We ignore block quota on meta pool (MDTs), so needn't
4534 * calculate how many blocks will be consumed by this index
4536 rc = osd_declare_inode_qid(env, i_uid_read(inode),
4537 i_gid_read(inode), 0, oh,
4538 osd_dt_obj(dt), true, NULL, false);
4544 rc = osd_remote_fid(env, osd, fid);
4550 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
4551 osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
4552 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4553 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4554 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4555 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
4561 * Index add function for interoperability mode (b11826).
4562 * It will add the directory entry.This entry is needed to
4563 * maintain name->fid mapping.
4565 * \param key it is key i.e. file entry to be inserted
4566 * \param rec it is value of given key i.e. fid
4568 * \retval 0, on success
4569 * \retval -ve, on error
4571 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4572 const struct dt_rec *rec,
4573 const struct dt_key *key, struct thandle *th,
4574 struct lustre_capa *capa, int ignore_quota)
4576 struct osd_object *obj = osd_dt_obj(dt);
4577 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4578 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
4579 const struct lu_fid *fid = rec1->rec_fid;
4580 const char *name = (const char *)key;
4581 struct osd_thread_info *oti = osd_oti_get(env);
4582 struct osd_inode_id *id = &oti->oti_id;
4583 struct inode *child_inode = NULL;
4584 struct osd_object *child = NULL;
4588 if (!dt_object_exists(dt))
4591 LASSERT(osd_invariant(obj));
4592 LASSERT(!dt_object_remote(dt));
4593 LASSERT(th != NULL);
4595 osd_trans_exec_op(env, th, OSD_OT_INSERT);
4597 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
4600 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
4602 rc = osd_remote_fid(env, osd, fid);
4604 CERROR("%s: Can not find object "DFID" rc %d\n",
4605 osd_name(osd), PFID(fid), rc);
4610 /* Insert remote entry */
4611 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4612 struct osd_mdobj_map *omm = osd->od_mdt_map;
4613 struct osd_thandle *oh;
4615 /* If parent on remote MDT, we need put this object
4617 oh = container_of(th, typeof(*oh), ot_super);
4618 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4620 CERROR("%s: add "DFID" error: rc = %d\n",
4622 PFID(lu_object_fid(&dt->do_lu)), rc);
4626 child_inode = igrab(omm->omm_remote_parent->d_inode);
4628 child_inode = osd_create_local_agent_inode(env, osd,
4629 obj, fid, rec1->rec_type & S_IFMT, th);
4630 if (IS_ERR(child_inode))
4631 RETURN(PTR_ERR(child_inode));
4634 /* Insert local entry */
4635 child = osd_object_find(env, dt, fid);
4636 if (IS_ERR(child)) {
4637 CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4638 osd_name(osd), PFID(fid),
4639 id->oii_ino, id->oii_gen,
4640 (int)PTR_ERR(child));
4641 RETURN(PTR_ERR(child));
4643 child_inode = igrab(child->oo_inode);
4646 rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
4648 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4649 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4653 osd_object_put(env, child);
4654 LASSERT(osd_invariant(obj));
4659 * Initialize osd Iterator for given osd index object.
4661 * \param dt osd index object
4664 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4665 struct dt_object *dt,
4667 struct lustre_capa *capa)
4669 struct osd_it_iam *it;
4670 struct osd_object *obj = osd_dt_obj(dt);
4671 struct lu_object *lo = &dt->do_lu;
4672 struct iam_path_descr *ipd;
4673 struct iam_container *bag = &obj->oo_dir->od_container;
4675 if (!dt_object_exists(dt))
4676 return ERR_PTR(-ENOENT);
4678 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
4679 return ERR_PTR(-EACCES);
4683 return ERR_PTR(-ENOMEM);
4685 ipd = osd_it_ipd_get(env, bag);
4686 if (likely(ipd != NULL)) {
4690 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4691 return (struct dt_it *)it;
4694 return ERR_PTR(-ENOMEM);
4699 * free given Iterator.
4702 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4704 struct osd_it_iam *it = (struct osd_it_iam *)di;
4705 struct osd_object *obj = it->oi_obj;
4707 iam_it_fini(&it->oi_it);
4708 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4709 lu_object_put(env, &obj->oo_dt.do_lu);
4714 * Move Iterator to record specified by \a key
4716 * \param di osd iterator
4717 * \param key key for index
4719 * \retval +ve di points to record with least key not larger than key
4720 * \retval 0 di points to exact matched key
4721 * \retval -ve failure
4724 static int osd_it_iam_get(const struct lu_env *env,
4725 struct dt_it *di, const struct dt_key *key)
4727 struct osd_thread_info *oti = osd_oti_get(env);
4728 struct osd_it_iam *it = (struct osd_it_iam *)di;
4730 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4731 /* swab quota uid/gid */
4732 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4733 key = (struct dt_key *)&oti->oti_quota_id;
4736 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4742 * \param di osd iterator
4744 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4746 struct osd_it_iam *it = (struct osd_it_iam *)di;
4748 iam_it_put(&it->oi_it);
4752 * Move iterator by one record
4754 * \param di osd iterator
4756 * \retval +1 end of container reached
4758 * \retval -ve failure
4761 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4763 struct osd_it_iam *it = (struct osd_it_iam *)di;
4765 return iam_it_next(&it->oi_it);
4769 * Return pointer to the key under iterator.
4772 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4773 const struct dt_it *di)
4775 struct osd_thread_info *oti = osd_oti_get(env);
4776 struct osd_it_iam *it = (struct osd_it_iam *)di;
4777 struct osd_object *obj = it->oi_obj;
4780 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4782 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4783 /* swab quota uid/gid */
4784 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4785 key = (struct dt_key *)&oti->oti_quota_id;
4792 * Return size of key under iterator (in bytes)
4795 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4797 struct osd_it_iam *it = (struct osd_it_iam *)di;
4799 return iam_it_key_size(&it->oi_it);
4803 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4805 /* check if file type is required */
4806 if (ent->lde_attrs & LUDA_TYPE) {
4807 struct luda_type *lt;
4808 int align = sizeof(*lt) - 1;
4810 len = (len + align) & ~align;
4811 lt = (struct luda_type *)(ent->lde_name + len);
4812 lt->lt_type = cpu_to_le16(DTTOIF(type));
4815 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4819 * build lu direct from backend fs dirent.
4823 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4824 char *name, __u16 namelen, __u16 type, __u32 attr)
4826 ent->lde_attrs = attr | LUDA_FID;
4827 fid_cpu_to_le(&ent->lde_fid, fid);
4829 ent->lde_hash = cpu_to_le64(offset);
4830 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4832 strncpy(ent->lde_name, name, namelen);
4833 ent->lde_name[namelen] = '\0';
4834 ent->lde_namelen = cpu_to_le16(namelen);
4836 /* append lustre attributes */
4837 osd_it_append_attrs(ent, namelen, type);
4841 * Return pointer to the record under iterator.
4843 static int osd_it_iam_rec(const struct lu_env *env,
4844 const struct dt_it *di,
4845 struct dt_rec *dtrec, __u32 attr)
4847 struct osd_it_iam *it = (struct osd_it_iam *)di;
4848 struct osd_thread_info *info = osd_oti_get(env);
4851 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4852 const struct osd_fid_pack *rec;
4853 struct lu_fid *fid = &info->oti_fid;
4854 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
4860 name = (char *)iam_it_key_get(&it->oi_it);
4862 RETURN(PTR_ERR(name));
4864 namelen = iam_it_key_size(&it->oi_it);
4866 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4868 RETURN(PTR_ERR(rec));
4870 rc = osd_fid_unpack(fid, rec);
4874 hash = iam_it_store(&it->oi_it);
4876 /* IAM does not store object type in IAM index (dir) */
4877 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4879 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4880 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4881 (struct iam_rec *)dtrec);
4882 osd_quota_unpack(it->oi_obj, dtrec);
4884 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4885 (struct iam_rec *)dtrec);
4892 * Returns cookie for current Iterator position.
4894 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4896 struct osd_it_iam *it = (struct osd_it_iam *)di;
4898 return iam_it_store(&it->oi_it);
4902 * Restore iterator from cookie.
4904 * \param di osd iterator
4905 * \param hash Iterator location cookie
4907 * \retval +ve di points to record with least key not larger than key.
4908 * \retval 0 di points to exact matched key
4909 * \retval -ve failure
4912 static int osd_it_iam_load(const struct lu_env *env,
4913 const struct dt_it *di, __u64 hash)
4915 struct osd_it_iam *it = (struct osd_it_iam *)di;
4917 return iam_it_load(&it->oi_it, hash);
4920 static const struct dt_index_operations osd_index_iam_ops = {
4921 .dio_lookup = osd_index_iam_lookup,
4922 .dio_declare_insert = osd_index_declare_iam_insert,
4923 .dio_insert = osd_index_iam_insert,
4924 .dio_declare_delete = osd_index_declare_iam_delete,
4925 .dio_delete = osd_index_iam_delete,
4927 .init = osd_it_iam_init,
4928 .fini = osd_it_iam_fini,
4929 .get = osd_it_iam_get,
4930 .put = osd_it_iam_put,
4931 .next = osd_it_iam_next,
4932 .key = osd_it_iam_key,
4933 .key_size = osd_it_iam_key_size,
4934 .rec = osd_it_iam_rec,
4935 .store = osd_it_iam_store,
4936 .load = osd_it_iam_load
4942 * Creates or initializes iterator context.
4944 * \retval struct osd_it_ea, iterator structure on success
4947 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4948 struct dt_object *dt,
4950 struct lustre_capa *capa)
4952 struct osd_object *obj = osd_dt_obj(dt);
4953 struct osd_thread_info *info = osd_oti_get(env);
4954 struct osd_it_ea *oie;
4956 struct lu_object *lo = &dt->do_lu;
4957 struct dentry *obj_dentry;
4960 if (!dt_object_exists(dt))
4961 RETURN(ERR_PTR(-ENOENT));
4963 OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
4965 RETURN(ERR_PTR(-ENOMEM));
4966 obj_dentry = &oie->oie_dentry;
4968 obj_dentry->d_inode = obj->oo_inode;
4969 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4970 obj_dentry->d_name.hash = 0;
4972 oie->oie_rd_dirent = 0;
4973 oie->oie_it_dirent = 0;
4974 oie->oie_dirent = NULL;
4975 if (unlikely(!info->oti_it_ea_buf_used)) {
4976 oie->oie_buf = info->oti_it_ea_buf;
4977 info->oti_it_ea_buf_used = 1;
4979 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
4980 if (oie->oie_buf == NULL)
4981 RETURN(ERR_PTR(-ENOMEM));
4985 file = &oie->oie_file;
4987 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4988 if (attr & LUDA_64BITHASH)
4989 file->f_mode = FMODE_64BITHASH;
4991 file->f_mode = FMODE_32BITHASH;
4992 file->f_dentry = obj_dentry;
4993 file->f_mapping = obj->oo_inode->i_mapping;
4994 file->f_op = obj->oo_inode->i_fop;
4995 set_file_inode(file, obj->oo_inode);
4998 RETURN((struct dt_it *) oie);
5002 * Destroy or finishes iterator context.
5004 * \param di iterator structure to be destroyed
5006 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
5008 struct osd_thread_info *info = osd_oti_get(env);
5009 struct osd_it_ea *oie = (struct osd_it_ea *)di;
5010 struct osd_object *obj = oie->oie_obj;
5011 struct inode *inode = obj->oo_inode;
5014 oie->oie_file.f_op->release(inode, &oie->oie_file);
5015 lu_object_put(env, &obj->oo_dt.do_lu);
5016 if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
5017 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5019 info->oti_it_ea_buf_used = 0;
5020 OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
5025 * It position the iterator at given key, so that next lookup continues from
5026 * that key Or it is similar to dio_it->load() but based on a key,
5027 * rather than file position.
5029 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
5032 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
5034 static int osd_it_ea_get(const struct lu_env *env,
5035 struct dt_it *di, const struct dt_key *key)
5037 struct osd_it_ea *it = (struct osd_it_ea *)di;
5040 LASSERT(((const char *)key)[0] == '\0');
5041 it->oie_file.f_pos = 0;
5042 it->oie_rd_dirent = 0;
5043 it->oie_it_dirent = 0;
5044 it->oie_dirent = NULL;
5052 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
5056 struct osd_filldir_cbs {
5057 #ifdef HAVE_DIR_CONTEXT
5058 struct dir_context ctx;
5060 struct osd_it_ea *it;
5063 * It is called internally by ->readdir(). It fills the
5064 * iterator's in-memory data structure with required
5065 * information i.e. name, namelen, rec_size etc.
5067 * \param buf in which information to be filled in.
5068 * \param name name of the file in given dir
5070 * \retval 0 on success
5071 * \retval 1 on buffer full
5073 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
5074 loff_t offset, __u64 ino,
5077 struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
5078 struct osd_object *obj = it->oie_obj;
5079 struct osd_it_ea_dirent *ent = it->oie_dirent;
5080 struct lu_fid *fid = &ent->oied_fid;
5081 struct osd_fid_pack *rec;
5084 /* this should never happen */
5085 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5086 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5090 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
5094 /* "." is just the object itself. */
5095 if (namelen == 1 && name[0] == '.') {
5096 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5097 } else if (d_type & LDISKFS_DIRENT_LUFID) {
5098 rec = (struct osd_fid_pack*) (name + namelen + 1);
5099 if (osd_fid_unpack(fid, rec) != 0)
5104 d_type &= ~LDISKFS_DIRENT_LUFID;
5106 /* NOT export local root. */
5107 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5108 ino = obj->oo_inode->i_ino;
5109 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5112 ent->oied_ino = ino;
5113 ent->oied_off = offset;
5114 ent->oied_namelen = namelen;
5115 ent->oied_type = d_type;
5117 memcpy(ent->oied_name, name, namelen);
5119 it->oie_rd_dirent++;
5120 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
5125 * Calls ->readdir() to load a directory entry at a time
5126 * and stored it in iterator's in-memory data structure.
5128 * \param di iterator's in memory structure
5130 * \retval 0 on success
5131 * \retval -ve on error
5132 * \retval +1 reach the end of entry
5134 static int osd_ldiskfs_it_fill(const struct lu_env *env,
5135 const struct dt_it *di)
5137 struct osd_it_ea *it = (struct osd_it_ea *)di;
5138 struct osd_object *obj = it->oie_obj;
5139 struct inode *inode = obj->oo_inode;
5140 struct htree_lock *hlock = NULL;
5141 struct file *filp = &it->oie_file;
5143 struct osd_filldir_cbs buf = {
5144 #ifdef HAVE_DIR_CONTEXT
5145 .ctx.actor = osd_ldiskfs_filldir,
5151 it->oie_dirent = it->oie_buf;
5152 it->oie_rd_dirent = 0;
5154 if (obj->oo_hl_head != NULL) {
5155 hlock = osd_oti_get(env)->oti_hlock;
5156 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5157 inode, LDISKFS_HLOCK_READDIR);
5159 down_read(&obj->oo_ext_idx_sem);
5162 #ifdef HAVE_DIR_CONTEXT
5163 buf.ctx.pos = filp->f_pos;
5164 rc = inode->i_fop->iterate(filp, &buf.ctx);
5165 filp->f_pos = buf.ctx.pos;
5167 rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
5171 ldiskfs_htree_unlock(hlock);
5173 up_read(&obj->oo_ext_idx_sem);
5175 if (it->oie_rd_dirent == 0) {
5176 /*If it does not get any dirent, it means it has been reached
5177 *to the end of the dir */
5178 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
5182 it->oie_dirent = it->oie_buf;
5183 it->oie_it_dirent = 1;
5190 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5191 * to load a directory entry at a time and stored it in
5192 * iterator's in-memory data structure.
5194 * \param di iterator's in memory structure
5196 * \retval +ve iterator reached to end
5197 * \retval 0 iterator not reached to end
5198 * \retval -ve on error
5200 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
5202 struct osd_it_ea *it = (struct osd_it_ea *)di;
5207 if (it->oie_it_dirent < it->oie_rd_dirent) {
5209 (void *) it->oie_dirent +
5210 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
5211 it->oie_dirent->oied_namelen);
5212 it->oie_it_dirent++;
5215 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
5218 rc = osd_ldiskfs_it_fill(env, di);
5225 * Returns the key at current position from iterator's in memory structure.
5227 * \param di iterator's in memory structure
5229 * \retval key i.e. struct dt_key on success
5231 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
5232 const struct dt_it *di)
5234 struct osd_it_ea *it = (struct osd_it_ea *)di;
5236 return (struct dt_key *)it->oie_dirent->oied_name;
5240 * Returns the key's size at current position from iterator's in memory structure.
5242 * \param di iterator's in memory structure
5244 * \retval key_size i.e. struct dt_key on success
5246 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
5248 struct osd_it_ea *it = (struct osd_it_ea *)di;
5250 return it->oie_dirent->oied_namelen;
5254 osd_dirent_update(handle_t *jh, struct super_block *sb,
5255 struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5256 struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
5258 struct osd_fid_pack *rec;
5262 LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
5263 LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
5265 rc = ldiskfs_journal_get_write_access(jh, bh);
5269 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5270 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5271 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
5277 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
5279 if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
5280 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5287 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5289 LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5290 "dot_dotdot = %d\n", dot_dotdot);
5292 if (LDISKFS_DIR_REC_LEN(de) >=
5293 __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5300 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5301 struct inode *dir, struct inode *inode,
5302 struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5303 struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5304 struct htree_lock *hlock)
5306 struct dentry *dentry;
5307 struct osd_fid_pack *rec;
5308 struct ldiskfs_dentry_param *ldp;
5312 if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5313 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5316 /* There is enough space to hold the FID-in-dirent. */
5317 if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
5318 dir->i_sb->s_blocksize)) {
5319 rc = ldiskfs_journal_get_write_access(jh, bh);
5323 de->name[de->name_len] = 0;
5324 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5325 rec->fp_len = sizeof(struct lu_fid) + 1;
5326 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5327 de->file_type |= LDISKFS_DIRENT_LUFID;
5329 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
5334 rc = ldiskfs_delete_entry(jh, dir, de, bh);
5338 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5340 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5341 osd_get_ldiskfs_dirent_param(ldp, fid);
5342 dentry->d_fsdata = (void *)ldp;
5343 ll_vfs_dq_init(dir);
5344 rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5345 /* It is too bad, we cannot reinsert the name entry back.
5346 * That means we lose it! */
5348 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
5349 "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
5350 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5351 dir->i_ino, dir->i_generation,
5352 ent->oied_namelen, ent->oied_name, PFID(fid), rc);
5358 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5359 struct osd_it_ea *it, struct lu_fid *fid,
5360 struct osd_inode_id *id, __u32 *attr)
5362 struct osd_thread_info *info = osd_oti_get(env);
5363 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
5364 struct osd_device *dev = osd_obj2dev(obj);
5365 struct super_block *sb = osd_sb(dev);
5366 const char *devname =
5367 LDISKFS_SB(sb)->s_es->s_volume_name;
5368 struct osd_it_ea_dirent *ent = it->oie_dirent;
5369 struct inode *dir = obj->oo_inode;
5370 struct htree_lock *hlock = NULL;
5371 struct buffer_head *bh = NULL;
5372 handle_t *jh = NULL;
5373 struct ldiskfs_dir_entry_2 *de;
5374 struct dentry *dentry;
5375 struct inode *inode;
5382 if (ent->oied_name[0] == '.') {
5383 if (ent->oied_namelen == 1)
5385 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5389 dentry = osd_child_dentry_get(env, obj, ent->oied_name,
5392 /* We need to ensure that the name entry is still valid.
5393 * Because it may be removed or renamed by other already.
5395 * The unlink or rename operation will start journal before PDO lock,
5396 * so to avoid deadlock, here we need to start journal handle before
5397 * related PDO lock also. But because we do not know whether there
5398 * will be something to be repaired before PDO lock, we just start
5399 * journal without conditions.
5401 * We may need to remove the name entry firstly, then insert back.
5402 * One credit is for user quota file update.
5403 * One credit is for group quota file update.
5404 * Two credits are for dirty inode. */
5405 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5406 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5409 if (dev->od_dirent_journal != 0) {
5410 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5413 CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent "
5414 "check_repair, dir = %lu/%u, credits = %d, "
5415 "name = %.*s: rc = %d\n",
5416 devname, dir->i_ino, dir->i_generation, credits,
5417 ent->oied_namelen, ent->oied_name, rc);
5421 if (obj->oo_hl_head != NULL) {
5422 hlock = osd_oti_get(env)->oti_hlock;
5423 /* "0" means exclusive lock for the whole directory.
5424 * We need to prevent others access such name entry
5425 * during the delete + insert. Neither HLOCK_ADD nor
5426 * HLOCK_DEL cannot guarantee the atomicity. */
5427 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5429 down_write(&obj->oo_ext_idx_sem);
5432 if (obj->oo_hl_head != NULL) {
5433 hlock = osd_oti_get(env)->oti_hlock;
5434 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5435 LDISKFS_HLOCK_LOOKUP);
5437 down_read(&obj->oo_ext_idx_sem);
5441 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5442 /* For dot/dotdot entry, if there is not enough space to hold the
5443 * FID-in-dirent, just keep them there. It only happens when the
5444 * device upgraded from 1.8 or restored from MDT file-level backup.
5445 * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5446 * and needs to get FID from LMA when readdir, it will not affect the
5447 * performance much. */
5448 if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
5449 (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5450 *attr |= LUDA_IGNORE;
5451 GOTO(out_journal, rc = 0);
5454 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5455 inode = osd_iget(info, dev, id);
5456 if (IS_ERR(inode)) {
5457 rc = PTR_ERR(inode);
5458 if (rc == -ENOENT || rc == -ESTALE)
5461 CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
5462 "check_repair, dir = %lu/%u, name = %.*s: "
5464 devname, dir->i_ino, dir->i_generation,
5465 ent->oied_namelen, ent->oied_name, rc);
5467 GOTO(out_journal, rc);
5470 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
5472 LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
5474 if (fid_is_sane(fid)) {
5475 /* FID-in-dirent is valid. */
5476 if (lu_fid_eq(fid, &lma->lma_self_fid))
5477 GOTO(out_inode, rc = 0);
5479 /* Do not repair under dryrun mode. */
5480 if (*attr & LUDA_VERIFY_DRYRUN) {
5481 *attr |= LUDA_REPAIR;
5482 GOTO(out_inode, rc = 0);
5485 if (dev->od_dirent_journal == 0) {
5489 ldiskfs_htree_unlock(hlock);
5491 up_read(&obj->oo_ext_idx_sem);
5492 dev->od_dirent_journal = 1;
5496 *fid = lma->lma_self_fid;
5498 /* Update the FID-in-dirent. */
5499 rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
5501 *attr |= LUDA_REPAIR;
5503 CDEBUG(D_LFSCK, "%.16s: fail to update FID "
5504 "in the dirent, dir = %lu/%u, "
5505 "name = %.*s, "DFID": rc = %d\n",
5506 devname, dir->i_ino, dir->i_generation,
5507 ent->oied_namelen, ent->oied_name,
5510 /* Do not repair under dryrun mode. */
5511 if (*attr & LUDA_VERIFY_DRYRUN) {
5512 *fid = lma->lma_self_fid;
5513 *attr |= LUDA_REPAIR;
5514 GOTO(out_inode, rc = 0);
5517 if (dev->od_dirent_journal == 0) {
5521 ldiskfs_htree_unlock(hlock);
5523 up_read(&obj->oo_ext_idx_sem);
5524 dev->od_dirent_journal = 1;
5528 *fid = lma->lma_self_fid;
5530 /* Append the FID-in-dirent. */
5531 rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5532 fid, bh, de, hlock);
5534 *attr |= LUDA_REPAIR;
5536 CDEBUG(D_LFSCK, "%.16s: fail to append FID "
5537 "after the dirent, dir = %lu/%u, "
5538 "name = %.*s, "DFID": rc = %d\n",
5539 devname, dir->i_ino, dir->i_generation,
5540 ent->oied_namelen, ent->oied_name,
5543 } else if (rc == -ENODATA) {
5544 /* Do not repair under dryrun mode. */
5545 if (*attr & LUDA_VERIFY_DRYRUN) {
5546 if (fid_is_sane(fid)) {
5547 *attr |= LUDA_REPAIR;
5549 lu_igif_build(fid, inode->i_ino,
5550 inode->i_generation);
5551 *attr |= LUDA_UPGRADE;
5553 GOTO(out_inode, rc = 0);
5556 if (dev->od_dirent_journal == 0) {
5560 ldiskfs_htree_unlock(hlock);
5562 up_read(&obj->oo_ext_idx_sem);
5563 dev->od_dirent_journal = 1;
5568 if (unlikely(fid_is_sane(fid))) {
5569 /* FID-in-dirent exists, but FID-in-LMA is lost.
5570 * Trust the FID-in-dirent, and add FID-in-LMA. */
5571 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5573 *attr |= LUDA_REPAIR;
5575 CDEBUG(D_LFSCK, "%.16s: fail to set LMA for "
5576 "update dirent, dir = %lu/%u, "
5577 "name = %.*s, "DFID": rc = %d\n",
5578 devname, dir->i_ino, dir->i_generation,
5579 ent->oied_namelen, ent->oied_name,
5582 lu_igif_build(fid, inode->i_ino, inode->i_generation);
5583 /* It is probably IGIF object. Only aappend the
5584 * FID-in-dirent. OI scrub will process FID-in-LMA. */
5585 rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5586 fid, bh, de, hlock);
5588 *attr |= LUDA_UPGRADE;
5590 CDEBUG(D_LFSCK, "%.16s: fail to append IGIF "
5591 "after the dirent, dir = %lu/%u, "
5592 "name = %.*s, "DFID": rc = %d\n",
5593 devname, dir->i_ino, dir->i_generation,
5594 ent->oied_namelen, ent->oied_name,
5599 GOTO(out_inode, rc);
5606 if (hlock != NULL) {
5607 ldiskfs_htree_unlock(hlock);
5609 if (dev->od_dirent_journal != 0)
5610 up_write(&obj->oo_ext_idx_sem);
5612 up_read(&obj->oo_ext_idx_sem);
5615 ldiskfs_journal_stop(jh);
5616 if (rc >= 0 && !dirty)
5617 dev->od_dirent_journal = 0;
5622 * Returns the value at current position from iterator's in memory structure.
5624 * \param di struct osd_it_ea, iterator's in memory structure
5625 * \param attr attr requested for dirent.
5626 * \param lde lustre dirent
5628 * \retval 0 no error and \param lde has correct lustre dirent.
5629 * \retval -ve on error
5631 static inline int osd_it_ea_rec(const struct lu_env *env,
5632 const struct dt_it *di,
5633 struct dt_rec *dtrec, __u32 attr)
5635 struct osd_it_ea *it = (struct osd_it_ea *)di;
5636 struct osd_object *obj = it->oie_obj;
5637 struct osd_device *dev = osd_obj2dev(obj);
5638 struct osd_thread_info *oti = osd_oti_get(env);
5639 struct osd_inode_id *id = &oti->oti_id;
5640 struct lu_fid *fid = &it->oie_dirent->oied_fid;
5641 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
5642 __u32 ino = it->oie_dirent->oied_ino;
5646 LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
5648 if (attr & LUDA_VERIFY) {
5649 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5650 attr |= LUDA_IGNORE;
5651 /* If the parent is on remote MDT, and there
5652 * is no FID-in-dirent, then we have to get
5653 * the parent FID from the linkEA. */
5654 osd_get_pfid_from_linkea(env, obj, fid);
5656 rc = osd_dirent_check_repair(env, obj, it, fid, id,
5660 attr &= ~LU_DIRENT_ATTRS_MASK;
5661 if (!fid_is_sane(fid)) {
5662 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP) &&
5663 likely(it->oie_dirent->oied_namelen != 2 ||
5664 it->oie_dirent->oied_name[0] != '.' ||
5665 it->oie_dirent->oied_name[1] != '.'))
5668 if (unlikely(ino == osd_remote_parent_ino(dev)))
5669 /* If the parent is on remote MDT, and there
5670 * is no FID-in-dirent, then we have to get
5671 * the parent FID from the linkEA. */
5672 rc = osd_get_pfid_from_linkea(env, obj, fid);
5674 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5676 osd_id_gen(id, ino, OSD_OII_NOGEN);
5680 /* Pack the entry anyway, at least the offset is right. */
5681 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5682 it->oie_dirent->oied_name,
5683 it->oie_dirent->oied_namelen,
5684 it->oie_dirent->oied_type, attr);
5689 if (osd_remote_fid(env, dev, fid))
5692 if (likely(!(attr & LUDA_IGNORE) && rc == 0))
5693 osd_add_oi_cache(oti, dev, id, fid);
5695 RETURN(rc > 0 ? 0 : rc);
5699 * Returns the record size size at current position.
5701 * This function will return record(lu_dirent) size in bytes.
5703 * \param[in] env execution environment
5704 * \param[in] di iterator's in memory structure
5705 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
5706 * calculate the lu_dirent size.
5708 * \retval record size(in bytes & in memory) of the current lu_dirent
5711 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5714 struct osd_it_ea *it = (struct osd_it_ea *)di;
5716 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5720 * Returns a cookie for current position of the iterator head, so that
5721 * user can use this cookie to load/start the iterator next time.
5723 * \param di iterator's in memory structure
5725 * \retval cookie for current position, on success
5727 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5729 struct osd_it_ea *it = (struct osd_it_ea *)di;
5731 return it->oie_dirent->oied_off;
5735 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5736 * to load a directory entry at a time and stored it i inn,
5737 * in iterator's in-memory data structure.
5739 * \param di struct osd_it_ea, iterator's in memory structure
5741 * \retval +ve on success
5742 * \retval -ve on error
5744 static int osd_it_ea_load(const struct lu_env *env,
5745 const struct dt_it *di, __u64 hash)
5747 struct osd_it_ea *it = (struct osd_it_ea *)di;
5751 it->oie_file.f_pos = hash;
5753 rc = osd_ldiskfs_it_fill(env, di);
5764 * Index lookup function for interoperability mode (b11826).
5766 * \param key, key i.e. file name to be searched
5768 * \retval +ve, on success
5769 * \retval -ve, on error
5771 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5772 struct dt_rec *rec, const struct dt_key *key,
5773 struct lustre_capa *capa)
5775 struct osd_object *obj = osd_dt_obj(dt);
5780 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5781 LINVRNT(osd_invariant(obj));
5783 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
5786 rc = osd_ea_lookup_rec(env, obj, rec, key);
5793 * Index and Iterator operations for interoperability
5794 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5796 static const struct dt_index_operations osd_index_ea_ops = {
5797 .dio_lookup = osd_index_ea_lookup,
5798 .dio_declare_insert = osd_index_declare_ea_insert,
5799 .dio_insert = osd_index_ea_insert,
5800 .dio_declare_delete = osd_index_declare_ea_delete,
5801 .dio_delete = osd_index_ea_delete,
5803 .init = osd_it_ea_init,
5804 .fini = osd_it_ea_fini,
5805 .get = osd_it_ea_get,
5806 .put = osd_it_ea_put,
5807 .next = osd_it_ea_next,
5808 .key = osd_it_ea_key,
5809 .key_size = osd_it_ea_key_size,
5810 .rec = osd_it_ea_rec,
5811 .rec_size = osd_it_ea_rec_size,
5812 .store = osd_it_ea_store,
5813 .load = osd_it_ea_load
5817 static void *osd_key_init(const struct lu_context *ctx,
5818 struct lu_context_key *key)
5820 struct osd_thread_info *info;
5822 OBD_ALLOC_PTR(info);
5824 return ERR_PTR(-ENOMEM);
5826 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5827 if (info->oti_it_ea_buf == NULL)
5830 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5832 info->oti_hlock = ldiskfs_htree_lock_alloc();
5833 if (info->oti_hlock == NULL)
5839 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5842 return ERR_PTR(-ENOMEM);
5845 static void osd_key_fini(const struct lu_context *ctx,
5846 struct lu_context_key *key, void* data)
5848 struct osd_thread_info *info = data;
5850 if (info->oti_hlock != NULL)
5851 ldiskfs_htree_lock_free(info->oti_hlock);
5852 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5853 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5854 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5855 lu_buf_free(&info->oti_big_buf);
5859 static void osd_key_exit(const struct lu_context *ctx,
5860 struct lu_context_key *key, void *data)
5862 struct osd_thread_info *info = data;
5864 LASSERT(info->oti_r_locks == 0);
5865 LASSERT(info->oti_w_locks == 0);
5866 LASSERT(info->oti_txns == 0);
5869 /* type constructor/destructor: osd_type_init, osd_type_fini */
5870 LU_TYPE_INIT_FINI(osd, &osd_key);
5872 struct lu_context_key osd_key = {
5873 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5874 .lct_init = osd_key_init,
5875 .lct_fini = osd_key_fini,
5876 .lct_exit = osd_key_exit
5880 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5881 const char *name, struct lu_device *next)
5883 struct osd_device *osd = osd_dev(d);
5885 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5886 >= sizeof(osd->od_svname))
5888 return osd_procfs_init(osd, name);
5891 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5893 struct seq_server_site *ss = osd_seq_site(osd);
5897 if (osd->od_is_ost || osd->od_cl_seq != NULL)
5900 if (unlikely(ss == NULL))
5903 OBD_ALLOC_PTR(osd->od_cl_seq);
5904 if (osd->od_cl_seq == NULL)
5907 rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5908 osd->od_svname, ss->ss_server_seq);
5911 OBD_FREE_PTR(osd->od_cl_seq);
5912 osd->od_cl_seq = NULL;
5918 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5920 if (osd->od_cl_seq == NULL)
5923 seq_client_fini(osd->od_cl_seq);
5924 OBD_FREE_PTR(osd->od_cl_seq);
5925 osd->od_cl_seq = NULL;
5928 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5932 /* shutdown quota slave instance associated with the device */
5933 if (o->od_quota_slave != NULL) {
5934 qsd_fini(env, o->od_quota_slave);
5935 o->od_quota_slave = NULL;
5938 osd_fid_fini(env, o);
5943 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5947 if (o->od_mnt != NULL) {
5948 shrink_dcache_sb(osd_sb(o));
5949 osd_sync(env, &o->od_dt_dev);
5958 static int osd_mount(const struct lu_env *env,
5959 struct osd_device *o, struct lustre_cfg *cfg)
5961 const char *name = lustre_cfg_string(cfg, 0);
5962 const char *dev = lustre_cfg_string(cfg, 1);
5964 unsigned long page, s_flags, lmd_flags = 0;
5965 struct page *__page;
5966 struct file_system_type *type;
5967 char *options = NULL;
5969 struct osd_thread_info *info = osd_oti_get(env);
5970 struct lu_fid *fid = &info->oti_fid;
5971 struct inode *inode;
5975 if (o->od_mnt != NULL)
5978 if (strlen(dev) >= sizeof(o->od_mntdev))
5980 strcpy(o->od_mntdev, dev);
5982 OBD_PAGE_ALLOC(__page, GFP_IOFS);
5984 GOTO(out, rc = -ENOMEM);
5986 str = lustre_cfg_string(cfg, 2);
5987 s_flags = simple_strtoul(str, NULL, 0);
5988 str = strstr(str, ":");
5990 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5991 opts = lustre_cfg_string(cfg, 3);
5992 page = (unsigned long)page_address(__page);
5993 options = (char *)page;
5996 strcat(options, "user_xattr,acl");
5998 strcat(options, opts);
6000 /* Glom up mount options */
6001 if (*options != '\0')
6002 strcat(options, ",");
6003 strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
6005 type = get_fs_type("ldiskfs");
6007 CERROR("%s: cannot find ldiskfs module\n", name);
6008 GOTO(out, rc = -ENODEV);
6011 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
6012 module_put(type->owner);
6014 if (IS_ERR(o->od_mnt)) {
6015 rc = PTR_ERR(o->od_mnt);
6017 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
6021 #ifdef HAVE_DEV_SET_RDONLY
6022 if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
6023 CERROR("%s: underlying device %s is marked as read-only. "
6024 "Setup failed\n", name, dev);
6025 GOTO(out_mnt, rc = -EROFS);
6029 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
6030 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
6031 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
6032 GOTO(out_mnt, rc = -EINVAL);
6035 #ifdef LDISKFS_MOUNT_DIRDATA
6036 if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
6037 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6038 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
6039 else if (!o->od_is_ost)
6040 CWARN("%s: device %s was upgraded from Lustre-1.x without "
6041 "enabling the dirdata feature. If you do not want to "
6042 "downgrade to Lustre-1.x again, you can enable it via "
6043 "'tune2fs -O dirdata device'\n", name, dev);
6045 inode = osd_sb(o)->s_root->d_inode;
6046 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
6047 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
6049 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
6053 if (lmd_flags & LMD_FLG_NOSCRUB)
6064 OBD_PAGE_FREE(__page);
6069 static struct lu_device *osd_device_fini(const struct lu_env *env,
6070 struct lu_device *d)
6072 struct osd_device *o = osd_dev(d);
6075 osd_shutdown(env, o);
6077 osd_scrub_cleanup(env, o);
6078 osd_obj_map_fini(o);
6084 static int osd_device_init0(const struct lu_env *env,
6085 struct osd_device *o,
6086 struct lustre_cfg *cfg)
6088 struct lu_device *l = osd2lu_dev(o);
6089 struct osd_thread_info *info;
6093 /* if the module was re-loaded, env can loose its keys */
6094 rc = lu_env_refill((struct lu_env *) env);
6097 info = osd_oti_get(env);
6100 l->ld_ops = &osd_lu_ops;
6101 o->od_dt_dev.dd_ops = &osd_dt_ops;
6103 spin_lock_init(&o->od_osfs_lock);
6104 mutex_init(&o->od_otable_mutex);
6106 o->od_capa_hash = init_capa_hash();
6107 if (o->od_capa_hash == NULL)
6108 GOTO(out, rc = -ENOMEM);
6110 o->od_read_cache = 1;
6111 o->od_writethrough_cache = 1;
6112 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
6114 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
6115 sizeof(o->od_svname));
6116 if (cplen >= sizeof(o->od_svname)) {
6121 if (server_name_is_ost(o->od_svname))
6124 o->od_full_scrub_ratio = OFSR_DEFAULT;
6125 o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
6126 rc = osd_mount(env, o, cfg);
6130 rc = osd_obj_map_init(env, o);
6134 rc = lu_site_init(&o->od_site, l);
6136 GOTO(out_compat, rc);
6137 o->od_site.ls_bottom_dev = l;
6139 rc = lu_site_init_finish(&o->od_site);
6143 INIT_LIST_HEAD(&o->od_ios_list);
6144 /* setup scrub, including OI files initialization */
6145 rc = osd_scrub_setup(env, o);
6149 rc = osd_procfs_init(o, o->od_svname);
6151 CERROR("%s: can't initialize procfs: rc = %d\n",
6153 GOTO(out_scrub, rc);
6156 LASSERT(l->ld_site->ls_linkage.next != NULL);
6157 LASSERT(l->ld_site->ls_linkage.prev != NULL);
6159 /* initialize quota slave instance */
6160 o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
6162 if (IS_ERR(o->od_quota_slave)) {
6163 rc = PTR_ERR(o->od_quota_slave);
6164 o->od_quota_slave = NULL;
6165 GOTO(out_procfs, rc);
6173 osd_scrub_cleanup(env, o);
6175 lu_site_fini(&o->od_site);
6177 osd_obj_map_fini(o);
6181 cleanup_capa_hash(o->od_capa_hash);
6186 static struct lu_device *osd_device_alloc(const struct lu_env *env,
6187 struct lu_device_type *t,
6188 struct lustre_cfg *cfg)
6190 struct osd_device *o;
6195 return ERR_PTR(-ENOMEM);
6197 rc = dt_device_init(&o->od_dt_dev, t);
6199 /* Because the ctx might be revived in dt_device_init,
6200 * refill the env here */
6201 lu_env_refill((struct lu_env *)env);
6202 rc = osd_device_init0(env, o, cfg);
6204 dt_device_fini(&o->od_dt_dev);
6207 if (unlikely(rc != 0))
6210 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
6213 static struct lu_device *osd_device_free(const struct lu_env *env,
6214 struct lu_device *d)
6216 struct osd_device *o = osd_dev(d);
6219 cleanup_capa_hash(o->od_capa_hash);
6220 /* XXX: make osd top device in order to release reference */
6221 d->ld_site->ls_top_dev = d;
6222 lu_site_purge(env, d->ld_site, -1);
6223 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
6224 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
6225 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
6227 lu_site_fini(&o->od_site);
6228 dt_device_fini(&o->od_dt_dev);
6233 static int osd_process_config(const struct lu_env *env,
6234 struct lu_device *d, struct lustre_cfg *cfg)
6236 struct osd_device *o = osd_dev(d);
6240 switch (cfg->lcfg_command) {
6242 rc = osd_mount(env, o, cfg);
6245 lu_dev_del_linkage(d->ld_site, d);
6246 rc = osd_shutdown(env, o);
6249 LASSERT(&o->od_dt_dev);
6250 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
6251 cfg, &o->od_dt_dev);
6252 if (rc > 0 || rc == -ENOSYS)
6253 rc = class_process_proc_param(PARAM_OST,
6254 lprocfs_osd_obd_vars,
6255 cfg, &o->od_dt_dev);
6264 static int osd_recovery_complete(const struct lu_env *env,
6265 struct lu_device *d)
6267 struct osd_device *osd = osd_dev(d);
6271 if (osd->od_quota_slave == NULL)
6274 /* start qsd instance on recovery completion, this notifies the quota
6275 * slave code that we are about to process new requests now */
6276 rc = qsd_start(env, osd->od_quota_slave);
6281 * we use exports to track all osd users
6283 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
6284 struct obd_device *obd, struct obd_uuid *cluuid,
6285 struct obd_connect_data *data, void *localdata)
6287 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6288 struct lustre_handle conn;
6292 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6294 rc = class_connect(&conn, obd, cluuid);
6298 *exp = class_conn2export(&conn);
6300 spin_lock(&osd->od_osfs_lock);
6302 spin_unlock(&osd->od_osfs_lock);
6308 * once last export (we don't count self-export) disappeared
6309 * osd can be released
6311 static int osd_obd_disconnect(struct obd_export *exp)
6313 struct obd_device *obd = exp->exp_obd;
6314 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6315 int rc, release = 0;
6318 /* Only disconnect the underlying layers on the final disconnect. */
6319 spin_lock(&osd->od_osfs_lock);
6321 if (osd->od_connects == 0)
6323 spin_unlock(&osd->od_osfs_lock);
6325 rc = class_disconnect(exp); /* bz 9811 */
6327 if (rc == 0 && release)
6328 class_manual_cleanup(obd);
6332 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6333 struct lu_device *dev)
6335 struct osd_device *osd = osd_dev(dev);
6336 struct lr_server_data *lsd =
6337 &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
6341 if (osd->od_quota_slave != NULL) {
6342 /* set up quota slave objects */
6343 result = qsd_prepare(env, osd->od_quota_slave);
6348 if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
6349 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
6350 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
6351 osd->od_index_in_idif = 1;
6353 osd->od_index_in_idif = 0;
6354 result = osd_register_proc_index_in_idif(osd);
6359 osd->od_index_in_idif = 1;
6363 result = osd_fid_init(env, osd);
6368 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6369 struct lu_fid *fid, struct md_op_data *op_data)
6371 struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6373 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6376 static const struct lu_object_operations osd_lu_obj_ops = {
6377 .loo_object_init = osd_object_init,
6378 .loo_object_delete = osd_object_delete,
6379 .loo_object_release = osd_object_release,
6380 .loo_object_free = osd_object_free,
6381 .loo_object_print = osd_object_print,
6382 .loo_object_invariant = osd_object_invariant
6385 const struct lu_device_operations osd_lu_ops = {
6386 .ldo_object_alloc = osd_object_alloc,
6387 .ldo_process_config = osd_process_config,
6388 .ldo_recovery_complete = osd_recovery_complete,
6389 .ldo_prepare = osd_prepare,
6392 static const struct lu_device_type_operations osd_device_type_ops = {
6393 .ldto_init = osd_type_init,
6394 .ldto_fini = osd_type_fini,
6396 .ldto_start = osd_type_start,
6397 .ldto_stop = osd_type_stop,
6399 .ldto_device_alloc = osd_device_alloc,
6400 .ldto_device_free = osd_device_free,
6402 .ldto_device_init = osd_device_init,
6403 .ldto_device_fini = osd_device_fini
6406 static struct lu_device_type osd_device_type = {
6407 .ldt_tags = LU_DEVICE_DT,
6408 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
6409 .ldt_ops = &osd_device_type_ops,
6410 .ldt_ctx_tags = LCT_LOCAL,
6414 * lprocfs legacy support.
6416 static struct obd_ops osd_obd_device_ops = {
6417 .o_owner = THIS_MODULE,
6418 .o_connect = osd_obd_connect,
6419 .o_disconnect = osd_obd_disconnect,
6420 .o_fid_alloc = osd_fid_alloc,
6423 static int __init osd_mod_init(void)
6429 rc = lu_kmem_init(ldiskfs_caches);
6433 rc = class_register_type(&osd_obd_device_ops, NULL, true,
6434 lprocfs_osd_module_vars,
6435 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6437 lu_kmem_fini(ldiskfs_caches);
6441 static void __exit osd_mod_exit(void)
6443 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6444 lu_kmem_fini(ldiskfs_caches);
6447 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6448 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6449 MODULE_LICENSE("GPL");
6451 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);