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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/osd/osd_handler.c
34 * Top-level entry points into osd module
36 * Author: Nikita Danilov <nikita@clusterfs.com>
37 * Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
40 #define DEBUG_SUBSYSTEM S_OSD
42 #include <linux/module.h>
43 #include <linux/user_namespace.h>
44 #ifdef HAVE_UIDGID_HEADER
45 # include <linux/uidgid.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>
57 #include <ldiskfs/ldiskfs.h>
58 #include <ldiskfs/xattr.h>
59 #include <ldiskfs/ldiskfs_extents.h>
62 * struct OBD_{ALLOC,FREE}*()
65 #include <obd_support.h>
66 /* struct ptlrpc_thread */
67 #include <lustre_net.h>
68 #include <lustre_fid.h>
70 #include <lustre_param.h>
72 #include "osd_internal.h"
73 #include "osd_dynlocks.h"
75 /* llo_* api support */
76 #include <md_object.h>
77 #include <lustre_quota.h>
79 #include <lustre_linkea.h>
81 #define PFID_STRIPE_IDX_BITS 16
82 #define PFID_STRIPE_COUNT_MASK ((1 << PFID_STRIPE_IDX_BITS) - 1)
85 module_param(ldiskfs_pdo, int, 0644);
86 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
88 int ldiskfs_track_declares_assert;
89 module_param(ldiskfs_track_declares_assert, int, 0644);
90 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
92 /* Slab to allocate dynlocks */
93 struct kmem_cache *dynlock_cachep;
95 /* Slab to allocate osd_it_ea */
96 struct kmem_cache *osd_itea_cachep;
98 static struct lu_kmem_descr ldiskfs_caches[] = {
100 .ckd_cache = &dynlock_cachep,
101 .ckd_name = "dynlock_cache",
102 .ckd_size = sizeof(struct dynlock_handle)
105 .ckd_cache = &osd_itea_cachep,
106 .ckd_name = "osd_itea_cache",
107 .ckd_size = sizeof(struct osd_it_ea)
114 static const char dot[] = ".";
115 static const char dotdot[] = "..";
116 static const char remote_obj_dir[] = "REM_OBJ_DIR";
118 static const struct lu_object_operations osd_lu_obj_ops;
119 static const struct dt_object_operations osd_obj_ops;
120 static const struct dt_object_operations osd_obj_ea_ops;
121 static const struct dt_object_operations osd_obj_otable_it_ops;
122 static const struct dt_index_operations osd_index_iam_ops;
123 static const struct dt_index_operations osd_index_ea_ops;
125 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
126 const struct lu_fid *fid);
127 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
128 struct osd_device *osd);
130 int osd_trans_declare_op2rb[] = {
131 [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
132 [OSD_OT_PUNCH] = OSD_OT_MAX,
133 [OSD_OT_XATTR_SET] = OSD_OT_XATTR_SET,
134 [OSD_OT_CREATE] = OSD_OT_DESTROY,
135 [OSD_OT_DESTROY] = OSD_OT_CREATE,
136 [OSD_OT_REF_ADD] = OSD_OT_REF_DEL,
137 [OSD_OT_REF_DEL] = OSD_OT_REF_ADD,
138 [OSD_OT_WRITE] = OSD_OT_WRITE,
139 [OSD_OT_INSERT] = OSD_OT_DELETE,
140 [OSD_OT_DELETE] = OSD_OT_INSERT,
141 [OSD_OT_QUOTA] = OSD_OT_MAX,
144 static int osd_has_index(const struct osd_object *obj)
146 return obj->oo_dt.do_index_ops != NULL;
149 static int osd_object_invariant(const struct lu_object *l)
151 return osd_invariant(osd_obj(l));
155 * Concurrency: doesn't matter
159 * Concurrency: doesn't matter
161 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
163 struct osd_thread_info *oti = osd_oti_get(env);
164 return oti->oti_w_locks > 0 && o->oo_owner == env;
168 * Concurrency: doesn't access mutable data
170 static int osd_root_get(const struct lu_env *env,
171 struct dt_device *dev, struct lu_fid *f)
173 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
178 * the following set of functions are used to maintain per-thread
179 * cache of FID->ino mapping. this mechanism is needed to resolve
180 * FID to inode at dt_insert() which in turn stores ino in the
181 * directory entries to keep ldiskfs compatible with ext[34].
182 * due to locking-originated restrictions we can't lookup ino
183 * using LU cache (deadlock is possible). lookup using OI is quite
184 * expensive. so instead we maintain this cache and methods like
185 * dt_create() fill it. so in the majority of cases dt_insert() is
186 * able to find needed mapping in lockless manner.
188 static struct osd_idmap_cache *
189 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
190 const struct lu_fid *fid)
192 struct osd_thread_info *oti = osd_oti_get(env);
193 struct osd_idmap_cache *idc = oti->oti_ins_cache;
195 for (i = 0; i < oti->oti_ins_cache_used; i++) {
196 if (!lu_fid_eq(&idc[i].oic_fid, fid))
198 if (idc[i].oic_dev != osd)
207 static struct osd_idmap_cache *
208 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
209 const struct lu_fid *fid)
211 struct osd_thread_info *oti = osd_oti_get(env);
212 struct osd_idmap_cache *idc;
215 if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
216 i = oti->oti_ins_cache_size * 2;
218 i = OSD_INS_CACHE_SIZE;
219 OBD_ALLOC(idc, sizeof(*idc) * i);
221 return ERR_PTR(-ENOMEM);
222 if (oti->oti_ins_cache != NULL) {
223 memcpy(idc, oti->oti_ins_cache,
224 oti->oti_ins_cache_used * sizeof(*idc));
225 OBD_FREE(oti->oti_ins_cache,
226 oti->oti_ins_cache_used * sizeof(*idc));
228 oti->oti_ins_cache = idc;
229 oti->oti_ins_cache_size = i;
232 idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
235 idc->oic_lid.oii_ino = 0;
236 idc->oic_lid.oii_gen = 0;
243 * lookup mapping for the given fid in the cache, initialize a
244 * new one if not found. the initialization checks whether the
245 * object is local or remote. for local objects, OI is used to
246 * learn ino/generation. the function is used when the caller
247 * has no information about the object, e.g. at dt_insert().
249 static struct osd_idmap_cache *
250 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
251 const struct lu_fid *fid)
253 struct osd_idmap_cache *idc;
256 idc = osd_idc_find(env, osd, fid);
257 LASSERT(!IS_ERR(idc));
261 /* new mapping is needed */
262 idc = osd_idc_add(env, osd, fid);
267 rc = osd_remote_fid(env, osd, fid);
268 if (unlikely(rc < 0))
272 /* the object is local, lookup in OI */
273 /* XXX: probably cheaper to lookup in LU first? */
274 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
276 if (unlikely(rc < 0)) {
277 CERROR("can't lookup: rc = %d\n", rc);
281 /* the object is remote */
289 * lookup mapping for given FID and fill it from the given object.
290 * the object is lolcal by definition.
292 static int osd_idc_find_and_init(const struct lu_env *env,
293 struct osd_device *osd,
294 struct osd_object *obj)
296 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
297 struct osd_idmap_cache *idc;
299 idc = osd_idc_find(env, osd, fid);
300 LASSERT(!IS_ERR(idc));
302 if (obj->oo_inode == NULL)
304 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
305 LASSERT(idc->oic_lid.oii_ino == 0);
306 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
307 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
312 /* new mapping is needed */
313 idc = osd_idc_add(env, osd, fid);
317 if (obj->oo_inode != NULL) {
318 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
319 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
325 * OSD object methods.
329 * Concurrency: no concurrent access is possible that early in object
332 static struct lu_object *osd_object_alloc(const struct lu_env *env,
333 const struct lu_object_header *hdr,
336 struct osd_object *mo;
342 l = &mo->oo_dt.do_lu;
343 dt_object_init(&mo->oo_dt, NULL, d);
344 mo->oo_dt.do_ops = &osd_obj_ea_ops;
345 l->lo_ops = &osd_lu_obj_ops;
346 init_rwsem(&mo->oo_sem);
347 init_rwsem(&mo->oo_ext_idx_sem);
348 spin_lock_init(&mo->oo_guard);
349 INIT_LIST_HEAD(&mo->oo_xattr_list);
356 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
357 struct dentry *dentry, struct lustre_ost_attrs *loa)
361 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
362 (void *)loa, sizeof(*loa));
364 struct lustre_mdt_attrs *lma = &loa->loa_lma;
366 if (rc < sizeof(*lma))
370 lustre_loa_swab(loa, true);
371 /* Check LMA compatibility */
372 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
373 CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
374 "for fid = "DFID", ino = %lu\n",
375 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
376 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
377 PFID(&lma->lma_self_fid), inode->i_ino);
380 } else if (rc == 0) {
388 * retrieve object from backend ext fs.
390 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
391 struct osd_inode_id *id)
394 struct inode *inode = NULL;
396 /* if we look for an inode withing a running
397 * transaction, then we risk to deadlock */
398 /* osd_dirent_check_repair() breaks this */
399 /*LASSERT(current->journal_info == NULL);*/
401 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
403 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
404 id->oii_ino, PTR_ERR(inode));
405 } else if (id->oii_gen != OSD_OII_NOGEN &&
406 inode->i_generation != id->oii_gen) {
407 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
408 "i_generation = %u\n",
409 id->oii_ino, id->oii_gen, inode->i_generation);
411 inode = ERR_PTR(-ESTALE);
412 } else if (inode->i_nlink == 0) {
413 /* due to parallel readdir and unlink,
414 * we can have dead inode here. */
415 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
417 inode = ERR_PTR(-ESTALE);
418 } else if (is_bad_inode(inode)) {
419 CWARN("%.16s: bad inode: ino = %u\n",
420 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
422 inode = ERR_PTR(-ENOENT);
423 } else if ((rc = osd_attach_jinode(inode))) {
427 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
428 if (id->oii_gen == OSD_OII_NOGEN)
429 osd_id_gen(id, inode->i_ino, inode->i_generation);
431 /* Do not update file c/mtime in ldiskfs.
432 * NB: we don't have any lock to protect this because we don't
433 * have reference on osd_object now, but contention with
434 * another lookup + attr_set can't happen in the tiny window
435 * between if (...) and set S_NOCMTIME. */
436 if (!(inode->i_flags & S_NOCMTIME))
437 inode->i_flags |= S_NOCMTIME;
442 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
443 handle_t *handle, struct dentry *child,
444 struct inode *inode, struct htree_lock *hlock)
448 rc = __ldiskfs_add_entry(handle, child, inode, hlock);
449 if (rc == -ENOBUFS || rc == -ENOSPC) {
450 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
451 struct inode *parent = child->d_parent->d_inode;
452 struct lu_fid *fid = NULL;
454 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
456 fid = &loa->loa_lma.lma_self_fid;
457 } else if (rc2 == -ENODATA) {
458 if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
459 fid = &info->oti_fid3;
460 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
461 } else if (!osd->od_is_ost && osd->od_index == 0) {
462 fid = &info->oti_fid3;
463 lu_igif_build(fid, parent->i_ino,
464 parent->i_generation);
469 CWARN("%s: directory (inode: %lu, FID: "DFID") %s "
470 "maximum entry limit\n",
471 osd_name(osd), parent->i_ino, PFID(fid),
472 rc == -ENOSPC ? "has reached" : "is approaching");
474 CWARN("%s: directory (inode: %lu, FID: unknown) %s "
475 "maximum entry limit\n",
476 osd_name(osd), parent->i_ino,
477 rc == -ENOSPC ? "has reached" : "is approaching");
479 /* ignore such error now */
488 static struct inode *
489 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
490 struct osd_inode_id *id, struct lu_fid *fid)
492 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
496 inode = osd_iget(info, dev, id);
500 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
502 *fid = loa->loa_lma.lma_self_fid;
503 } else if (rc == -ENODATA) {
504 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
505 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
507 lu_igif_build(fid, inode->i_ino, inode->i_generation);
515 static struct inode *osd_iget_check(struct osd_thread_info *info,
516 struct osd_device *dev,
517 const struct lu_fid *fid,
518 struct osd_inode_id *id,
525 /* The cached OI mapping is trustable. If we cannot locate the inode
526 * via the cached OI mapping, then return the failure to the caller
527 * directly without further OI checking. */
530 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
533 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
536 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
537 PFID(fid), id->oii_ino, rc);
541 if (is_bad_inode(inode)) {
546 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
547 PFID(fid), id->oii_ino);
551 if (id->oii_gen != OSD_OII_NOGEN &&
552 inode->i_generation != id->oii_gen) {
557 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
558 "oii_gen = %u, i_generation = %u\n", PFID(fid),
559 id->oii_ino, id->oii_gen, inode->i_generation);
563 if (inode->i_nlink == 0) {
568 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
569 PFID(fid), id->oii_ino);
573 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
577 __u32 saved_ino = id->oii_ino;
578 __u32 saved_gen = id->oii_gen;
581 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
583 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
584 /* XXX: There are four possible cases:
586 * Backup/restore caused the OI invalid.
588 * Someone unlinked the object but NOT removed
589 * the OI mapping, such as mount target device
590 * as ldiskfs, and modify something directly.
592 * Someone just removed the object between the
593 * former oi_lookup and the iget. It is normal.
594 * 4. Other failure cases.
596 * Generally, when the device is mounted, it will
597 * auto check whether the system is restored from
598 * file-level backup or not. We trust such detect
599 * to distinguish the 1st case from the 2nd case:
600 * if the OI files are consistent but may contain
601 * stale OI mappings because of case 2, if iget()
602 * returns -ENOENT or -ESTALE, then it should be
605 /* If the OI mapping was in OI file before the
606 * osd_iget_check(), but now, it is disappear,
607 * then it must be removed by race. That is a
608 * normal race case. */
611 /* It is the OI scrub updated the OI mapping by race.
612 * The new OI mapping must be valid. */
613 if (saved_ino != id->oii_ino || saved_gen != id->oii_gen) {
619 if (dev->od_scrub.os_file.sf_flags & SF_INCONSISTENT)
620 /* It still can be the case 2, but we cannot
621 * distinguish it from the case 1. So return
622 * -EREMCHG to block current operation until
623 * OI scrub rebuilt the OI mappings. */
631 if (inode->i_generation == id->oii_gen)
636 if (id->oii_gen == OSD_OII_NOGEN)
637 osd_id_gen(id, inode->i_ino, inode->i_generation);
639 /* Do not update file c/mtime in ldiskfs.
640 * NB: we don't have any lock to protect this because we don't
641 * have reference on osd_object now, but contention with
642 * another lookup + attr_set can't happen in the tiny window
643 * between if (...) and set S_NOCMTIME. */
644 if (!(inode->i_flags & S_NOCMTIME))
645 inode->i_flags |= S_NOCMTIME;
662 * \retval +v: new filter_fid, does not contain self-fid
663 * \retval 0: filter_fid_old, contains self-fid
664 * \retval -v: other failure cases
666 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
667 struct dentry *dentry, struct lu_fid *fid)
669 struct filter_fid_old *ff = &info->oti_ff;
670 struct ost_id *ostid = &info->oti_ostid;
673 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
674 if (rc == sizeof(*ff)) {
676 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
677 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
678 /* XXX: use 0 as the index for compatibility, the caller will
679 * handle index related issues when necessarry. */
680 ostid_to_fid(fid, ostid, 0);
681 } else if (rc == sizeof(struct filter_fid)) {
683 } else if (rc >= 0) {
690 static int osd_lma_self_repair(struct osd_thread_info *info,
691 struct osd_device *osd, struct inode *inode,
692 const struct lu_fid *fid, __u32 compat)
697 LASSERT(current->journal_info == NULL);
699 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
700 osd_dto_credits_noquota[DTO_XATTR_SET]);
703 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
708 rc = osd_ea_fid_set(info, inode, fid, compat, 0);
710 CWARN("%s: cannot self repair the LMA: rc = %d\n",
712 ldiskfs_journal_stop(jh);
716 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
718 struct osd_thread_info *info = osd_oti_get(env);
719 struct osd_device *osd = osd_obj2dev(obj);
720 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
721 struct lustre_mdt_attrs *lma = &loa->loa_lma;
722 struct inode *inode = obj->oo_inode;
723 struct dentry *dentry = &info->oti_obj_dentry;
724 struct lu_fid *fid = NULL;
725 const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
729 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
730 (void *)loa, sizeof(*loa));
731 if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
732 fid = &lma->lma_self_fid;
733 rc = osd_get_idif(info, inode, dentry, fid);
734 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
735 /* For the given OST-object, if it has neither LMA nor
736 * FID in XATTR_NAME_FID, then the given FID (which is
737 * contained in the @obj, from client RPC for locating
738 * the OST-object) is trusted. We use it to generate
740 osd_lma_self_repair(info, osd, inode, rfid,
751 lustre_lma_swab(lma);
752 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
753 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
754 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
755 "fid = "DFID", ino = %lu\n", osd_name(osd),
756 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
757 PFID(rfid), inode->i_ino);
760 fid = &lma->lma_self_fid;
764 if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
765 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
766 struct ost_id *oi = &info->oti_ostid;
767 struct lu_fid *fid1 = &info->oti_fid3;
768 __u32 idx = fid_idif_ost_idx(rfid);
770 /* For old IDIF, the OST index is not part of the IDIF,
771 * Means that different OSTs may have the same IDIFs.
772 * Under such case, we need to make some compatible
773 * check to make sure to trigger OI scrub properly. */
774 if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
775 /* Given @rfid is new, LMA is old. */
776 fid_to_ostid(fid, oi);
777 ostid_to_fid(fid1, oi, idx);
778 if (lu_fid_eq(fid1, rfid)) {
779 if (osd->od_index_in_idif)
780 osd_lma_self_repair(info, osd,
794 struct osd_check_lmv_buf {
795 #ifdef HAVE_DIR_CONTEXT
796 /* please keep it as first member */
797 struct dir_context ctx;
799 struct osd_thread_info *oclb_info;
800 struct osd_device *oclb_dev;
801 struct osd_idmap_cache *oclb_oic;
805 * It is called internally by ->readdir() to filter out the
806 * local slave object's FID of the striped directory.
808 * \retval 1 found the local slave's FID
809 * \retval 0 continue to check next item
810 * \retval -ve for failure
812 #ifdef HAVE_FILLDIR_USE_CTX
813 static int osd_stripe_dir_filldir(struct dir_context *buf,
815 static int osd_stripe_dir_filldir(void *buf,
817 const char *name, int namelen,
818 loff_t offset, __u64 ino, unsigned d_type)
820 struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
821 struct osd_thread_info *oti = oclb->oclb_info;
822 struct lu_fid *fid = &oti->oti_fid3;
823 struct osd_inode_id *id = &oti->oti_id3;
824 struct osd_device *dev = oclb->oclb_dev;
825 struct osd_idmap_cache *oic = oclb->oclb_oic;
833 sscanf(name + 1, SFID, RFID(fid));
834 if (!fid_is_sane(fid))
837 if (osd_remote_fid(oti->oti_env, dev, fid))
840 osd_id_gen(id, ino, OSD_OII_NOGEN);
841 inode = osd_iget(oti, dev, id);
843 return PTR_ERR(inode);
846 osd_add_oi_cache(oti, dev, id, fid);
850 rc = osd_oii_insert(dev, oic, true);
852 return rc == 0 ? 1 : rc;
855 /* When lookup item under striped directory, we need to locate the master
856 * MDT-object of the striped directory firstly, then the client will send
857 * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
858 * and the item's name. If the system is restored from MDT file level backup,
859 * then before the OI scrub completely built the OI files, the OI mappings of
860 * the master MDT-object and slave MDT-object may be invalid. Usually, it is
861 * not a problem for the master MDT-object. Because when locate the master
862 * MDT-object, we will do name based lookup (for the striped directory itself)
863 * firstly, during such process we can setup the correct OI mapping for the
864 * master MDT-object. But it will be trouble for the slave MDT-object. Because
865 * the client will not trigger name based lookup on the MDT to locate the slave
866 * MDT-object before locating item under the striped directory, then when
867 * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
868 * is invalid and does not know what the right OI mapping is, then the MDT has
869 * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
870 * the OI file, related OI mapping is unknown yet, please try again later. And
871 * then client will re-try the RPC again and again until related OI mapping has
872 * been updated. That is quite inefficient.
874 * To resolve above trouble, we will handle it as the following two cases:
876 * 1) The slave MDT-object and the master MDT-object are on different MDTs.
877 * It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
878 * is linked under /REMOTE_PARENT_DIR with the name of its FID string.
879 * We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
880 * directly. Please check osd_fid_lookup().
882 * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
883 * Under such case, during lookup the master MDT-object, we will lookup the
884 * slave MDT-object via readdir against the master MDT-object, because the
885 * slave MDT-objects information are stored as sub-directories with the name
886 * "${FID}:${index}". Then when find the local slave MDT-object, its OI
887 * mapping will be recorded. Then subsequent osd_fid_lookup() will know
888 * the correct OI mapping for the slave MDT-object. */
889 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
890 struct inode *inode, struct osd_idmap_cache *oic)
892 struct lu_buf *buf = &oti->oti_big_buf;
893 struct dentry *dentry = &oti->oti_obj_dentry;
894 struct file *filp = &oti->oti_file;
895 const struct file_operations *fops;
896 struct lmv_mds_md_v1 *lmv1;
897 struct osd_check_lmv_buf oclb = {
898 #ifdef HAVE_DIR_CONTEXT
899 .ctx.actor = osd_stripe_dir_filldir,
909 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
912 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
914 lu_buf_realloc(buf, rc);
915 if (buf->lb_buf == NULL)
916 GOTO(out, rc = -ENOMEM);
922 if (unlikely(rc == 0 || rc == -ENODATA))
928 if (unlikely(buf->lb_buf == NULL)) {
929 lu_buf_realloc(buf, rc);
930 if (buf->lb_buf == NULL)
931 GOTO(out, rc = -ENOMEM);
937 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
941 dentry->d_inode = inode;
942 dentry->d_sb = inode->i_sb;
944 filp->f_path.dentry = dentry;
945 filp->f_mode = FMODE_64BITHASH;
946 filp->f_mapping = inode->i_mapping;
948 filp->private_data = NULL;
949 set_file_inode(filp, inode);
951 #ifdef HAVE_DIR_CONTEXT
952 oclb.ctx.pos = filp->f_pos;
953 rc = fops->iterate(filp, &oclb.ctx);
954 filp->f_pos = oclb.ctx.pos;
956 rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir);
958 fops->release(inode, filp);
962 CDEBUG(D_LFSCK, "%.16s: fail to check LMV EA, inode = %lu/%u,"
964 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
965 inode->i_ino, inode->i_generation,
966 PFID(&oic->oic_fid), rc);
973 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
974 const struct lu_fid *fid,
975 const struct lu_object_conf *conf)
977 struct osd_thread_info *info;
978 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
979 struct osd_device *dev;
980 struct osd_idmap_cache *oic;
981 struct osd_inode_id *id;
982 struct inode *inode = NULL;
983 struct osd_scrub *scrub;
984 struct scrub_file *sf;
985 __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
992 bool updated = false;
995 LINVRNT(osd_invariant(obj));
996 LASSERT(obj->oo_inode == NULL);
997 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
1000 scrub = &dev->od_scrub;
1001 sf = &scrub->os_file;
1002 info = osd_oti_get(env);
1004 oic = &info->oti_cache;
1006 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1009 /* For the object is created as locking anchor, or for the object to
1010 * be created on disk. No need to osd_oi_lookup() at here because FID
1011 * shouldn't never be re-used, if it's really a duplicate FID from
1012 * unexpected reason, we should be able to detect it later by calling
1013 * do_create->osd_oi_insert(). */
1014 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
1015 GOTO(out, result = 0);
1017 /* Search order: 1. per-thread cache. */
1018 if (lu_fid_eq(fid, &oic->oic_fid) &&
1019 likely(oic->oic_dev == dev)) {
1025 if (!list_empty(&scrub->os_inconsistent_items)) {
1026 /* Search order: 2. OI scrub pending list. */
1027 result = osd_oii_lookup(dev, fid, id);
1032 /* The OI mapping in the OI file can be updated by the OI scrub
1033 * when we locate the inode via FID. So it may be not trustable. */
1036 /* Search order: 3. OI files. */
1037 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1038 if (result == -ENOENT) {
1039 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1040 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1041 !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
1043 GOTO(out, result = 0);
1052 inode = osd_iget_check(info, dev, fid, id, trusted);
1053 if (IS_ERR(inode)) {
1054 result = PTR_ERR(inode);
1055 if (result == -ENOENT || result == -ESTALE)
1056 GOTO(out, result = 0);
1058 if (result == -EREMCHG) {
1061 /* We still have chance to get the valid inode: for the
1062 * object which is referenced by remote name entry, the
1063 * object on the local MDT will be linked under the dir
1064 * of "/REMOTE_PARENT_DIR" with its FID string as name.
1066 * We do not know whether the object for the given FID
1067 * is referenced by some remote name entry or not, and
1068 * especially for DNE II, a multiple-linked object may
1069 * have many name entries reside on many MDTs.
1071 * To simplify the operation, OSD will not distinguish
1072 * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1073 * only happened for the RPC from other MDT during the
1074 * OI scrub, or for the client side RPC with FID only,
1075 * such as FID to path, or from old connected client. */
1077 !fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1078 rc1 = osd_lookup_in_remote_parent(info, dev,
1083 flags |= SS_AUTO_PARTIAL;
1084 flags &= ~SS_AUTO_FULL;
1089 if (thread_is_running(&scrub->os_thread)) {
1090 if (scrub->os_partial_scan &&
1091 !scrub->os_in_join) {
1094 if (inode != NULL && !IS_ERR(inode)) {
1097 osd_add_oi_cache(info, dev, id,
1099 osd_oii_insert(dev, oic, true);
1101 result = -EINPROGRESS;
1104 } else if (!dev->od_noscrub) {
1107 rc1 = osd_scrub_start(dev, flags);
1108 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
1109 "for the "DFID" with flags 0x%x,"
1110 " rc = %d\n", osd_name(dev),
1111 PFID(fid), flags, rc1);
1112 if (rc1 == 0 || rc1 == -EALREADY) {
1113 if (inode != NULL && !IS_ERR(inode)) {
1116 osd_add_oi_cache(info, dev, id,
1118 osd_oii_insert(dev, oic, true);
1120 result = -EINPROGRESS;
1130 if (inode == NULL || IS_ERR(inode))
1132 } else if (remote) {
1136 obj->oo_inode = inode;
1137 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
1139 result = osd_check_lma(env, obj);
1143 LASSERTF(id->oii_ino == inode->i_ino &&
1144 id->oii_gen == inode->i_generation,
1145 "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1146 PFID(fid), id->oii_ino, id->oii_gen,
1147 inode->i_ino, inode->i_generation);
1149 saved_ino = inode->i_ino;
1150 saved_gen = inode->i_generation;
1152 if (unlikely(result == -ENODATA)) {
1153 /* If the OI scrub updated the OI mapping by race, it
1154 * must be valid. Trust the inode that has no LMA EA. */
1158 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1160 /* The OI mapping is still there, the inode is still
1161 * valid. It is just becaues the inode has no LMA EA. */
1162 if (saved_ino == id->oii_ino &&
1163 saved_gen == id->oii_gen)
1166 /* It is the OI scrub updated the OI mapping by race.
1167 * The new OI mapping must be valid. */
1170 obj->oo_inode = NULL;
1176 /* "result == -ENOENT" means that the OI mappinghas been
1177 * removed by race, so the inode belongs to other object.
1179 * Others error can be returned directly. */
1180 if (result == -ENOENT) {
1189 obj->oo_inode = NULL;
1191 if (result != -EREMCHG)
1196 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1197 /* "result == -ENOENT" means the cached OI mapping has been removed
1198 * from the OI file by race, above inode belongs to other object. */
1199 if (result == -ENOENT) {
1202 GOTO(out, result = 0);
1208 if (saved_ino == id->oii_ino && saved_gen == id->oii_gen)
1211 /* It is the OI scrub updated the OI mapping by race.
1212 * The new OI mapping must be valid. */
1218 obj->oo_compat_dot_created = 1;
1219 obj->oo_compat_dotdot_created = 1;
1221 if (S_ISDIR(inode->i_mode) &&
1222 (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1223 osd_check_lmv(info, dev, inode, oic);
1225 result = osd_attach_jinode(inode);
1227 obj->oo_inode = NULL;
1233 GOTO(out, result = 0);
1235 LASSERT(obj->oo_hl_head == NULL);
1236 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1237 if (obj->oo_hl_head == NULL) {
1238 obj->oo_inode = NULL;
1240 GOTO(out, result = -ENOMEM);
1242 GOTO(out, result = 0);
1245 if (result != 0 && trusted)
1246 fid_zero(&oic->oic_fid);
1248 LINVRNT(osd_invariant(obj));
1253 * Concurrency: shouldn't matter.
1255 static void osd_object_init0(struct osd_object *obj)
1257 LASSERT(obj->oo_inode != NULL);
1258 obj->oo_dt.do_body_ops = &osd_body_ops;
1259 obj->oo_dt.do_lu.lo_header->loh_attr |=
1260 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1264 * Concurrency: no concurrent access is possible that early in object
1267 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1268 const struct lu_object_conf *conf)
1270 struct osd_object *obj = osd_obj(l);
1273 LINVRNT(osd_invariant(obj));
1275 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1276 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1277 l->lo_header->loh_attr |= LOHA_EXISTS;
1281 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1282 obj->oo_dt.do_body_ops = &osd_body_ops_new;
1283 if (result == 0 && obj->oo_inode != NULL) {
1284 struct osd_thread_info *oti = osd_oti_get(env);
1285 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1287 osd_object_init0(obj);
1288 result = osd_get_lma(oti, obj->oo_inode,
1289 &oti->oti_obj_dentry, loa);
1291 /* Convert LMAI flags to lustre LMA flags
1292 * and cache it to oo_lma_flags */
1294 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1295 } else if (result == -ENODATA) {
1300 LINVRNT(osd_invariant(obj));
1304 /* The first part of oxe_buf is xattr name, and is '\0' terminated.
1305 * The left part is for value, binary mode. */
1306 struct osd_xattr_entry {
1307 struct list_head oxe_list;
1311 struct rcu_head oxe_rcu;
1315 static int osd_oxc_get(struct osd_object *obj, const char *name,
1318 struct osd_xattr_entry *tmp;
1319 struct osd_xattr_entry *oxe = NULL;
1320 size_t namelen = strlen(name);
1325 list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1326 if (namelen == tmp->oxe_namelen &&
1327 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1334 GOTO(out, rc = -ENOENT);
1336 if (!oxe->oxe_exist)
1337 GOTO(out, rc = -ENODATA);
1340 rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1343 if (buf->lb_buf == NULL)
1346 if (buf->lb_len < rc)
1347 GOTO(out, rc = -ERANGE);
1349 memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1357 static void osd_oxc_free(struct rcu_head *head)
1359 struct osd_xattr_entry *oxe;
1361 oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1362 OBD_FREE(oxe, oxe->oxe_len);
1365 static void osd_oxc_add(struct osd_object *obj, const char *name,
1366 const char *buf, int buflen)
1368 struct osd_xattr_entry *oxe;
1369 struct osd_xattr_entry *old = NULL;
1370 struct osd_xattr_entry *tmp;
1371 size_t namelen = strlen(name);
1372 size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1374 OBD_ALLOC(oxe, len);
1378 INIT_LIST_HEAD(&oxe->oxe_list);
1380 oxe->oxe_namelen = namelen;
1381 memcpy(oxe->oxe_buf, name, namelen);
1383 LASSERT(buf != NULL);
1384 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1385 oxe->oxe_exist = true;
1387 oxe->oxe_exist = false;
1390 /* this should be rarely called, just remove old and add new */
1391 spin_lock(&obj->oo_guard);
1392 list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1393 if (namelen == tmp->oxe_namelen &&
1394 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1400 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1401 call_rcu(&old->oxe_rcu, osd_oxc_free);
1403 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1405 spin_unlock(&obj->oo_guard);
1408 static void osd_oxc_del(struct osd_object *obj, const char *name)
1410 struct osd_xattr_entry *oxe;
1411 size_t namelen = strlen(name);
1413 spin_lock(&obj->oo_guard);
1414 list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1415 if (namelen == oxe->oxe_namelen &&
1416 strncmp(name, oxe->oxe_buf, namelen) == 0) {
1417 list_del_rcu(&oxe->oxe_list);
1418 call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1422 spin_unlock(&obj->oo_guard);
1425 static void osd_oxc_fini(struct osd_object *obj)
1427 struct osd_xattr_entry *oxe, *next;
1429 list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1430 list_del(&oxe->oxe_list);
1431 OBD_FREE(oxe, oxe->oxe_len);
1436 * Concurrency: no concurrent access is possible that late in object
1439 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1441 struct osd_object *obj = osd_obj(l);
1443 LINVRNT(osd_invariant(obj));
1446 dt_object_fini(&obj->oo_dt);
1447 if (obj->oo_hl_head != NULL)
1448 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1453 * Concurrency: no concurrent access is possible that late in object
1456 static void osd_index_fini(struct osd_object *o)
1458 struct iam_container *bag;
1460 if (o->oo_dir != NULL) {
1461 bag = &o->oo_dir->od_container;
1462 if (o->oo_inode != NULL) {
1463 if (bag->ic_object == o->oo_inode)
1464 iam_container_fini(bag);
1466 OBD_FREE_PTR(o->oo_dir);
1472 * Concurrency: no concurrent access is possible that late in object
1473 * life-cycle (for all existing callers, that is. New callers have to provide
1474 * their own locking.)
1476 static int osd_inode_unlinked(const struct inode *inode)
1478 return inode->i_nlink == 0;
1482 OSD_TXN_OI_DELETE_CREDITS = 20,
1483 OSD_TXN_INODE_DELETE_CREDITS = 20
1490 #if OSD_THANDLE_STATS
1492 * Set time when the handle is allocated
1494 static void osd_th_alloced(struct osd_thandle *oth)
1496 oth->oth_alloced = ktime_get();
1500 * Set time when the handle started
1502 static void osd_th_started(struct osd_thandle *oth)
1504 oth->oth_started = ktime_get();
1508 * Check whether the we deal with this handle for too long.
1510 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1511 ktime_t alloced, ktime_t started,
1514 ktime_t now = ktime_get();
1516 LASSERT(dev != NULL);
1518 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1519 ktime_us_delta(started, alloced));
1520 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1521 ktime_us_delta(closed, started));
1522 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1523 ktime_us_delta(now, closed));
1525 if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1526 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1527 oth, now, alloced, started, closed);
1528 libcfs_debug_dumpstack(NULL);
1532 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
1534 ktime_t __closed = ktime_get(); \
1535 ktime_t __alloced = oth->oth_alloced; \
1536 ktime_t __started = oth->oth_started; \
1539 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
1542 #else /* OSD_THANDLE_STATS */
1544 #define osd_th_alloced(h) do {} while(0)
1545 #define osd_th_started(h) do {} while(0)
1546 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
1548 #endif /* OSD_THANDLE_STATS */
1551 * Concurrency: doesn't access mutable data.
1553 static int osd_param_is_not_sane(const struct osd_device *dev,
1554 const struct thandle *th)
1556 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1558 return oh->ot_credits > osd_transaction_size(dev);
1562 * Concurrency: shouldn't matter.
1564 static void osd_trans_commit_cb(struct super_block *sb,
1565 struct ldiskfs_journal_cb_entry *jcb, int error)
1567 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1568 struct thandle *th = &oh->ot_super;
1569 struct lu_device *lud = &th->th_dev->dd_lu_dev;
1570 struct dt_txn_commit_cb *dcb, *tmp;
1572 LASSERT(oh->ot_handle == NULL);
1575 CERROR("transaction @0x%p commit error: %d\n", th, error);
1577 dt_txn_hook_commit(th);
1579 /* call per-transaction callbacks if any */
1580 list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1582 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1583 "commit callback entry: magic=%x name='%s'\n",
1584 dcb->dcb_magic, dcb->dcb_name);
1585 list_del_init(&dcb->dcb_linkage);
1586 dcb->dcb_func(NULL, th, dcb, error);
1589 lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1593 lu_context_exit(&th->th_ctx);
1594 lu_context_fini(&th->th_ctx);
1598 #ifndef HAVE_SB_START_WRITE
1599 # define sb_start_write(sb) do {} while (0)
1600 # define sb_end_write(sb) do {} while (0)
1603 static struct thandle *osd_trans_create(const struct lu_env *env,
1604 struct dt_device *d)
1606 struct osd_thread_info *oti = osd_oti_get(env);
1607 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1608 struct osd_thandle *oh;
1613 CERROR("%s: someone try to start transaction under "
1614 "readonly mode, should be disabled.\n",
1615 osd_name(osd_dt_dev(d)));
1617 RETURN(ERR_PTR(-EROFS));
1620 /* on pending IO in this thread should left from prev. request */
1621 LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1623 sb_start_write(osd_sb(osd_dt_dev(d)));
1625 OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
1627 oh->ot_quota_trans = &oti->oti_quota_trans;
1628 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1632 th->th_tags = LCT_TX_HANDLE;
1634 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1635 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1638 memset(oti->oti_declare_ops, 0,
1639 sizeof(oti->oti_declare_ops));
1640 memset(oti->oti_declare_ops_cred, 0,
1641 sizeof(oti->oti_declare_ops_cred));
1642 memset(oti->oti_declare_ops_used, 0,
1643 sizeof(oti->oti_declare_ops_used));
1645 sb_end_write(osd_sb(osd_dt_dev(d)));
1646 th = ERR_PTR(-ENOMEM);
1651 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1653 struct osd_thread_info *oti = osd_oti_get(env);
1654 struct osd_thandle *oh;
1656 oh = container_of0(th, struct osd_thandle, ot_super);
1657 LASSERT(oh != NULL);
1659 CWARN(" create: %u/%u/%u, destroy: %u/%u/%u\n",
1660 oti->oti_declare_ops[OSD_OT_CREATE],
1661 oti->oti_declare_ops_cred[OSD_OT_CREATE],
1662 oti->oti_declare_ops_used[OSD_OT_CREATE],
1663 oti->oti_declare_ops[OSD_OT_DESTROY],
1664 oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1665 oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1666 CWARN(" attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1667 oti->oti_declare_ops[OSD_OT_ATTR_SET],
1668 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1669 oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1670 oti->oti_declare_ops[OSD_OT_XATTR_SET],
1671 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1672 oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1673 CWARN(" write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1674 oti->oti_declare_ops[OSD_OT_WRITE],
1675 oti->oti_declare_ops_cred[OSD_OT_WRITE],
1676 oti->oti_declare_ops_used[OSD_OT_WRITE],
1677 oti->oti_declare_ops[OSD_OT_PUNCH],
1678 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1679 oti->oti_declare_ops_used[OSD_OT_PUNCH],
1680 oti->oti_declare_ops[OSD_OT_QUOTA],
1681 oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1682 oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1683 CWARN(" insert: %u/%u/%u, delete: %u/%u/%u\n",
1684 oti->oti_declare_ops[OSD_OT_INSERT],
1685 oti->oti_declare_ops_cred[OSD_OT_INSERT],
1686 oti->oti_declare_ops_used[OSD_OT_INSERT],
1687 oti->oti_declare_ops[OSD_OT_DELETE],
1688 oti->oti_declare_ops_cred[OSD_OT_DELETE],
1689 oti->oti_declare_ops_used[OSD_OT_DELETE]);
1690 CWARN(" ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1691 oti->oti_declare_ops[OSD_OT_REF_ADD],
1692 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1693 oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1694 oti->oti_declare_ops[OSD_OT_REF_DEL],
1695 oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1696 oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1700 * Concurrency: shouldn't matter.
1702 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1705 struct osd_thread_info *oti = osd_oti_get(env);
1706 struct osd_device *dev = osd_dt_dev(d);
1708 struct osd_thandle *oh;
1713 LASSERT(current->journal_info == NULL);
1715 oh = container_of0(th, struct osd_thandle, ot_super);
1716 LASSERT(oh != NULL);
1717 LASSERT(oh->ot_handle == NULL);
1719 rc = dt_txn_hook_start(env, d, th);
1723 if (unlikely(osd_param_is_not_sane(dev, th))) {
1724 static unsigned long last_printed;
1725 static int last_credits;
1727 /* don't make noise on a tiny testing systems
1728 * actual credits misuse will be caught anyway */
1729 if (last_credits != oh->ot_credits &&
1730 time_after(jiffies, last_printed +
1731 msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1732 osd_transaction_size(dev) > 512) {
1733 CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1734 oh->ot_credits, osd_transaction_size(dev));
1735 osd_trans_dump_creds(env, th);
1736 libcfs_debug_dumpstack(NULL);
1737 last_credits = oh->ot_credits;
1738 last_printed = jiffies;
1740 /* XXX Limit the credits to 'max_transaction_buffers', and
1741 * let the underlying filesystem to catch the error if
1742 * we really need so many credits.
1744 * This should be removed when we can calculate the
1745 * credits precisely. */
1746 oh->ot_credits = osd_transaction_size(dev);
1747 } else if (ldiskfs_track_declares_assert != 0) {
1748 /* reserve few credits to prevent an assertion in JBD
1749 * our debugging mechanism will be able to detected
1750 * overuse. this can help to debug single-update
1752 oh->ot_credits += 10;
1753 if (unlikely(osd_param_is_not_sane(dev, th)))
1754 oh->ot_credits = osd_transaction_size(dev);
1758 * XXX temporary stuff. Some abstraction layer should
1761 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1765 LASSERT(oti->oti_txns == 0);
1766 lu_context_init(&th->th_ctx, th->th_tags);
1767 lu_context_enter(&th->th_ctx);
1769 lu_device_get(&d->dd_lu_dev);
1770 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1781 static int osd_seq_exists(const struct lu_env *env,
1782 struct osd_device *osd, u64 seq)
1784 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1785 struct seq_server_site *ss = osd_seq_site(osd);
1789 LASSERT(ss != NULL);
1790 LASSERT(ss->ss_server_fld != NULL);
1792 rc = osd_fld_lookup(env, osd, seq, range);
1795 CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1796 osd_name(osd), seq, rc);
1800 RETURN(ss->ss_node_id == range->lsr_index);
1803 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1805 struct dt_txn_commit_cb *dcb;
1806 struct dt_txn_commit_cb *tmp;
1808 /* call per-transaction stop callbacks if any */
1809 list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1811 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1812 "commit callback entry: magic=%x name='%s'\n",
1813 dcb->dcb_magic, dcb->dcb_name);
1814 list_del_init(&dcb->dcb_linkage);
1815 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1820 * Concurrency: shouldn't matter.
1822 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1825 int rc = 0, remove_agents = 0;
1826 struct osd_thandle *oh;
1827 struct osd_thread_info *oti = osd_oti_get(env);
1828 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1829 struct osd_device *osd = osd_dt_dev(th->th_dev);
1830 struct qsd_instance *qsd = osd->od_quota_slave;
1831 struct lquota_trans *qtrans;
1834 oh = container_of0(th, struct osd_thandle, ot_super);
1836 /* reset OI cache for safety */
1837 oti->oti_ins_cache_used = 0;
1839 remove_agents = oh->ot_remove_agents;
1841 qtrans = oh->ot_quota_trans;
1842 oh->ot_quota_trans = NULL;
1844 if (oh->ot_handle != NULL) {
1846 handle_t *hdl = oh->ot_handle;
1849 * add commit callback
1850 * notice we don't do this in osd_trans_start()
1851 * as underlying transaction can change during truncate
1853 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1856 LASSERT(oti->oti_txns == 1);
1859 rc = dt_txn_hook_stop(env, th);
1861 CERROR("%s: failed in transaction hook: rc = %d\n",
1864 osd_trans_stop_cb(oh, rc);
1865 /* hook functions might modify th_sync */
1866 hdl->h_sync = th->th_sync;
1868 oh->ot_handle = NULL;
1869 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
1871 CERROR("%s: failed to stop transaction: rc = %d\n",
1872 osd_name(osd), rc2);
1876 osd_trans_stop_cb(oh, th->th_result);
1880 /* inform the quota slave device that the transaction is stopping */
1881 qsd_op_end(env, qsd, qtrans);
1883 /* as we want IO to journal and data IO be concurrent, we don't block
1884 * awaiting data IO completion in osd_do_bio(), instead we wait here
1885 * once transaction is submitted to the journal. all reqular requests
1886 * don't do direct IO (except read/write), thus this wait_event becomes
1889 * IMPORTANT: we have to wait till any IO submited by the thread is
1890 * completed otherwise iobuf may be corrupted by different request
1892 wait_event(iobuf->dr_wait,
1893 atomic_read(&iobuf->dr_numreqs) == 0);
1894 osd_fini_iobuf(osd, iobuf);
1896 rc = iobuf->dr_error;
1898 if (unlikely(remove_agents != 0))
1899 osd_process_scheduled_agent_removals(env, osd);
1901 sb_end_write(osd_sb(osd));
1906 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1908 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1911 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1912 LASSERT(&dcb->dcb_func != NULL);
1913 if (dcb->dcb_flags & DCB_TRANS_STOP)
1914 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1916 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1922 * Called just before object is freed. Releases all resources except for
1923 * object itself (that is released by osd_object_free()).
1925 * Concurrency: no concurrent access is possible that late in object
1928 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1930 struct osd_object *obj = osd_obj(l);
1931 struct inode *inode = obj->oo_inode;
1933 LINVRNT(osd_invariant(obj));
1936 * If object is unlinked remove fid->ino mapping from object index.
1939 osd_index_fini(obj);
1940 if (inode != NULL) {
1941 struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave;
1942 qid_t uid = i_uid_read(inode);
1943 qid_t gid = i_gid_read(inode);
1946 obj->oo_inode = NULL;
1949 struct osd_thread_info *info = osd_oti_get(env);
1950 struct lquota_id_info *qi = &info->oti_qi;
1952 /* Release granted quota to master if necessary */
1953 qi->lqi_id.qid_uid = uid;
1954 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1956 qi->lqi_id.qid_uid = gid;
1957 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1963 * Concurrency: ->loo_object_release() is called under site spin-lock.
1965 static void osd_object_release(const struct lu_env *env,
1966 struct lu_object *l)
1968 struct osd_object *o = osd_obj(l);
1969 /* nobody should be releasing a non-destroyed object with nlink=0
1970 * the API allows this, but ldiskfs doesn't like and then report
1971 * this inode as deleted */
1972 if (unlikely(!o->oo_destroyed && o->oo_inode && o->oo_inode->i_nlink == 0))
1977 * Concurrency: shouldn't matter.
1979 static int osd_object_print(const struct lu_env *env, void *cookie,
1980 lu_printer_t p, const struct lu_object *l)
1982 struct osd_object *o = osd_obj(l);
1983 struct iam_descr *d;
1985 if (o->oo_dir != NULL)
1986 d = o->oo_dir->od_container.ic_descr;
1989 return (*p)(env, cookie,
1990 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1992 o->oo_inode ? o->oo_inode->i_ino : 0UL,
1993 o->oo_inode ? o->oo_inode->i_generation : 0,
1994 d ? d->id_ops->id_name : "plain");
1998 * Concurrency: shouldn't matter.
2000 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2001 struct obd_statfs *sfs)
2003 struct osd_device *osd = osd_dt_dev(d);
2004 struct super_block *sb = osd_sb(osd);
2005 struct kstatfs *ksfs;
2009 if (unlikely(osd->od_mnt == NULL))
2010 return -EINPROGRESS;
2012 /* osd_lproc.c call this without env, allocate ksfs for that case */
2013 if (unlikely(env == NULL)) {
2014 OBD_ALLOC_PTR(ksfs);
2018 ksfs = &osd_oti_get(env)->oti_ksfs;
2021 result = sb->s_op->statfs(sb->s_root, ksfs);
2025 statfs_pack(sfs, ksfs);
2026 if (unlikely(sb->s_flags & MS_RDONLY))
2027 sfs->os_state |= OS_STATE_READONLY;
2028 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb,
2029 LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2030 sfs->os_maxbytes = sb->s_maxbytes;
2032 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2035 * Reserve some space so to avoid fragmenting the filesystem too much.
2036 * Fragmentation not only impacts performance, but can also increase
2037 * metadata overhead significantly, causing grant calculation to be
2040 * Reserve 0.78% of total space, at least 8MB for small filesystems.
2042 CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2043 reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2044 if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2045 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2047 sfs->os_blocks -= reserved;
2048 sfs->os_bfree -= min(reserved, sfs->os_bfree);
2049 sfs->os_bavail -= min(reserved, sfs->os_bavail);
2052 if (unlikely(env == NULL))
2058 * Estimate space needed for file creations. We assume the largest filename
2059 * which is 2^64 - 1, hence a filename of 20 chars.
2060 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2062 #ifdef __LDISKFS_DIR_REC_LEN
2063 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2065 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2069 * Concurrency: doesn't access mutable data.
2071 static void osd_conf_get(const struct lu_env *env,
2072 const struct dt_device *dev,
2073 struct dt_device_param *param)
2075 struct super_block *sb = osd_sb(osd_dt_dev(dev));
2079 * XXX should be taken from not-yet-existing fs abstraction layer.
2081 param->ddp_max_name_len = LDISKFS_NAME_LEN;
2082 param->ddp_max_nlink = LDISKFS_LINK_MAX;
2083 param->ddp_symlink_max = sb->s_blocksize;
2084 param->ddp_mount_type = LDD_MT_LDISKFS;
2085 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2086 param->ddp_maxbytes = sb->s_maxbytes;
2088 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2089 /* inode are statically allocated, so per-inode space consumption
2090 * is the space consumed by the directory entry */
2091 param->ddp_inodespace = PER_OBJ_USAGE;
2092 /* EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2093 * = 128MB) which is unlikely to be hit in real life. Report a smaller
2094 * maximum length to not under count the actual number of extents
2095 * needed for writing a file. */
2096 param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2;
2097 /* worst-case extent insertion metadata overhead */
2098 param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2099 param->ddp_mntopts = 0;
2100 if (test_opt(sb, XATTR_USER))
2101 param->ddp_mntopts |= MNTOPT_USERXATTR;
2102 if (test_opt(sb, POSIX_ACL))
2103 param->ddp_mntopts |= MNTOPT_ACL;
2105 /* LOD might calculate the max stripe count based on max_ea_size,
2106 * so we need take account in the overhead as well,
2107 * xattr_header + magic + xattr_entry_head */
2108 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2109 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2111 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2112 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
2113 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2117 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2121 * Concurrency: shouldn't matter.
2123 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2127 CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
2129 rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
2131 CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
2132 LUSTRE_OSD_LDISKFS_NAME, rc);
2138 * Start commit for OSD device.
2140 * An implementation of dt_commit_async method for OSD device.
2141 * Asychronously starts underlayng fs sync and thereby a transaction
2144 * \param env environment
2145 * \param d dt device
2147 * \see dt_device_operations
2149 static int osd_commit_async(const struct lu_env *env,
2150 struct dt_device *d)
2152 struct super_block *s = osd_sb(osd_dt_dev(d));
2155 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
2156 RETURN(s->s_op->sync_fs(s, 0));
2160 * Concurrency: shouldn't matter.
2163 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2165 struct super_block *sb = osd_sb(osd_dt_dev(d));
2166 struct block_device *dev = sb->s_bdev;
2167 #ifdef HAVE_DEV_SET_RDONLY
2168 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2171 int rc = -EOPNOTSUPP;
2175 #ifdef HAVE_DEV_SET_RDONLY
2176 CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
2178 if (sb->s_op->freeze_fs) {
2179 rc = sb->s_op->freeze_fs(sb);
2184 if (jdev && (jdev != dev)) {
2185 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2187 dev_set_rdonly(jdev);
2189 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2190 dev_set_rdonly(dev);
2192 if (sb->s_op->unfreeze_fs)
2193 sb->s_op->unfreeze_fs(sb);
2198 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2199 osd_dt_dev(d)->od_svname, (long)dev, rc);
2205 * Note: we do not count into QUOTA here.
2206 * If we mount with --data_journal we may need more.
2208 const int osd_dto_credits_noquota[DTO_NR] = {
2211 * INDEX_EXTRA_TRANS_BLOCKS(8) +
2212 * SINGLEDATA_TRANS_BLOCKS(8)
2213 * XXX Note: maybe iam need more, since iam have more level than
2216 [DTO_INDEX_INSERT] = 16,
2219 * just modify a single entry, probably merge few within a block
2221 [DTO_INDEX_DELETE] = 1,
2225 [DTO_INDEX_UPDATE] = 16,
2227 * 4(inode, inode bits, groups, GDT)
2228 * notice: OI updates are counted separately with DTO_INDEX_INSERT
2230 [DTO_OBJECT_CREATE] = 4,
2232 * 4(inode, inode bits, groups, GDT)
2233 * notice: OI updates are counted separately with DTO_INDEX_DELETE
2235 [DTO_OBJECT_DELETE] = 4,
2237 * Attr set credits (inode)
2239 [DTO_ATTR_SET_BASE] = 1,
2241 * Xattr set. The same as xattr of EXT3.
2242 * DATA_TRANS_BLOCKS(14)
2243 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2244 * are also counted in. Do not know why?
2246 [DTO_XATTR_SET] = 14,
2248 * credits for inode change during write.
2250 [DTO_WRITE_BASE] = 3,
2252 * credits for single block write.
2254 [DTO_WRITE_BLOCK] = 14,
2256 * Attr set credits for chown.
2257 * This is extra credits for setattr, and it is null without quota
2259 [DTO_ATTR_SET_CHOWN] = 0
2262 static const struct dt_device_operations osd_dt_ops = {
2263 .dt_root_get = osd_root_get,
2264 .dt_statfs = osd_statfs,
2265 .dt_trans_create = osd_trans_create,
2266 .dt_trans_start = osd_trans_start,
2267 .dt_trans_stop = osd_trans_stop,
2268 .dt_trans_cb_add = osd_trans_cb_add,
2269 .dt_conf_get = osd_conf_get,
2270 .dt_sync = osd_sync,
2272 .dt_commit_async = osd_commit_async,
2275 static void osd_object_read_lock(const struct lu_env *env,
2276 struct dt_object *dt, unsigned role)
2278 struct osd_object *obj = osd_dt_obj(dt);
2279 struct osd_thread_info *oti = osd_oti_get(env);
2281 LINVRNT(osd_invariant(obj));
2283 LASSERT(obj->oo_owner != env);
2284 down_read_nested(&obj->oo_sem, role);
2286 LASSERT(obj->oo_owner == NULL);
2290 static void osd_object_write_lock(const struct lu_env *env,
2291 struct dt_object *dt, unsigned role)
2293 struct osd_object *obj = osd_dt_obj(dt);
2294 struct osd_thread_info *oti = osd_oti_get(env);
2296 LINVRNT(osd_invariant(obj));
2298 LASSERT(obj->oo_owner != env);
2299 down_write_nested(&obj->oo_sem, role);
2301 LASSERT(obj->oo_owner == NULL);
2302 obj->oo_owner = env;
2306 static void osd_object_read_unlock(const struct lu_env *env,
2307 struct dt_object *dt)
2309 struct osd_object *obj = osd_dt_obj(dt);
2310 struct osd_thread_info *oti = osd_oti_get(env);
2312 LINVRNT(osd_invariant(obj));
2314 LASSERT(oti->oti_r_locks > 0);
2316 up_read(&obj->oo_sem);
2319 static void osd_object_write_unlock(const struct lu_env *env,
2320 struct dt_object *dt)
2322 struct osd_object *obj = osd_dt_obj(dt);
2323 struct osd_thread_info *oti = osd_oti_get(env);
2325 LINVRNT(osd_invariant(obj));
2327 LASSERT(obj->oo_owner == env);
2328 LASSERT(oti->oti_w_locks > 0);
2330 obj->oo_owner = NULL;
2331 up_write(&obj->oo_sem);
2334 static int osd_object_write_locked(const struct lu_env *env,
2335 struct dt_object *dt)
2337 struct osd_object *obj = osd_dt_obj(dt);
2339 LINVRNT(osd_invariant(obj));
2341 return obj->oo_owner == env;
2344 static struct timespec *osd_inode_time(const struct lu_env *env,
2345 struct inode *inode, __u64 seconds)
2347 struct osd_thread_info *oti = osd_oti_get(env);
2348 struct timespec *t = &oti->oti_time;
2350 t->tv_sec = seconds;
2352 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2356 static void osd_inode_getattr(const struct lu_env *env,
2357 struct inode *inode, struct lu_attr *attr)
2359 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2360 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2361 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
2364 attr->la_atime = LTIME_S(inode->i_atime);
2365 attr->la_mtime = LTIME_S(inode->i_mtime);
2366 attr->la_ctime = LTIME_S(inode->i_ctime);
2367 attr->la_mode = inode->i_mode;
2368 attr->la_size = i_size_read(inode);
2369 attr->la_blocks = inode->i_blocks;
2370 attr->la_uid = i_uid_read(inode);
2371 attr->la_gid = i_gid_read(inode);
2372 attr->la_flags = ll_inode_to_ext_flags(inode->i_flags);
2373 attr->la_nlink = inode->i_nlink;
2374 attr->la_rdev = inode->i_rdev;
2375 attr->la_blksize = 1 << inode->i_blkbits;
2376 attr->la_blkbits = inode->i_blkbits;
2379 static int osd_attr_get(const struct lu_env *env,
2380 struct dt_object *dt,
2381 struct lu_attr *attr)
2383 struct osd_object *obj = osd_dt_obj(dt);
2385 if (unlikely(!dt_object_exists(dt)))
2387 if (unlikely(obj->oo_destroyed))
2390 LASSERT(!dt_object_remote(dt));
2391 LINVRNT(osd_invariant(obj));
2393 spin_lock(&obj->oo_guard);
2394 osd_inode_getattr(env, obj->oo_inode, attr);
2395 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2396 attr->la_flags |= LUSTRE_ORPHAN_FL;
2397 spin_unlock(&obj->oo_guard);
2402 static int osd_declare_attr_set(const struct lu_env *env,
2403 struct dt_object *dt,
2404 const struct lu_attr *attr,
2405 struct thandle *handle)
2407 struct osd_thandle *oh;
2408 struct osd_object *obj;
2409 struct osd_thread_info *info = osd_oti_get(env);
2410 struct lquota_id_info *qi = &info->oti_qi;
2418 LASSERT(dt != NULL);
2419 LASSERT(handle != NULL);
2421 obj = osd_dt_obj(dt);
2422 LASSERT(osd_invariant(obj));
2424 oh = container_of0(handle, struct osd_thandle, ot_super);
2425 LASSERT(oh->ot_handle == NULL);
2427 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2428 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2430 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2431 osd_dto_credits_noquota[DTO_XATTR_SET]);
2433 if (attr == NULL || obj->oo_inode == NULL)
2436 bspace = obj->oo_inode->i_blocks;
2437 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
2438 bspace = toqb(bspace);
2440 /* Changing ownership is always preformed by super user, it should not
2443 * We still need to call the osd_declare_qid() to calculate the journal
2444 * credits for updating quota accounting files and to trigger quota
2445 * space adjustment once the operation is completed.*/
2446 if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2448 uid = i_uid_read(obj->oo_inode);
2449 qi->lqi_type = USRQUOTA;
2450 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2451 /* inode accounting */
2452 qi->lqi_is_blk = false;
2454 /* one more inode for the new uid ... */
2455 qi->lqi_id.qid_uid = attr->la_uid;
2457 /* Reserve credits for the new uid */
2458 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2459 if (rc == -EDQUOT || rc == -EINPROGRESS)
2464 /* and one less inode for the current uid */
2465 qi->lqi_id.qid_uid = uid;
2467 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2468 if (rc == -EDQUOT || rc == -EINPROGRESS)
2473 /* block accounting */
2474 qi->lqi_is_blk = true;
2476 /* more blocks for the new uid ... */
2477 qi->lqi_id.qid_uid = attr->la_uid;
2478 qi->lqi_space = bspace;
2480 * Credits for the new uid has been reserved, re-use "obj"
2481 * to save credit reservation.
2483 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2484 if (rc == -EDQUOT || rc == -EINPROGRESS)
2489 /* and finally less blocks for the current uid */
2490 qi->lqi_id.qid_uid = uid;
2491 qi->lqi_space = -bspace;
2492 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2493 if (rc == -EDQUOT || rc == -EINPROGRESS)
2499 gid = i_gid_read(obj->oo_inode);
2500 qi->lqi_type = GRPQUOTA;
2501 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2503 /* inode accounting */
2504 qi->lqi_is_blk = false;
2506 /* one more inode for the new gid ... */
2507 qi->lqi_id.qid_gid = attr->la_gid;
2509 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2510 if (rc == -EDQUOT || rc == -EINPROGRESS)
2515 /* and one less inode for the current gid */
2516 qi->lqi_id.qid_gid = gid;
2518 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2519 if (rc == -EDQUOT || rc == -EINPROGRESS)
2524 /* block accounting */
2525 qi->lqi_is_blk = true;
2527 /* more blocks for the new gid ... */
2528 qi->lqi_id.qid_gid = attr->la_gid;
2529 qi->lqi_space = bspace;
2530 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2531 if (rc == -EDQUOT || rc == -EINPROGRESS)
2536 /* and finally less blocks for the current gid */
2537 qi->lqi_id.qid_gid = gid;
2538 qi->lqi_space = -bspace;
2539 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2540 if (rc == -EDQUOT || rc == -EINPROGRESS)
2549 static int osd_inode_setattr(const struct lu_env *env,
2550 struct inode *inode, const struct lu_attr *attr)
2552 __u64 bits = attr->la_valid;
2554 /* Only allow set size for regular file */
2555 if (!S_ISREG(inode->i_mode))
2556 bits &= ~(LA_SIZE | LA_BLOCKS);
2561 if (bits & LA_ATIME)
2562 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
2563 if (bits & LA_CTIME)
2564 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
2565 if (bits & LA_MTIME)
2566 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
2567 if (bits & LA_SIZE) {
2568 spin_lock(&inode->i_lock);
2569 LDISKFS_I(inode)->i_disksize = attr->la_size;
2570 i_size_write(inode, attr->la_size);
2571 spin_unlock(&inode->i_lock);
2574 /* OSD should not change "i_blocks" which is used by quota.
2575 * "i_blocks" should be changed by ldiskfs only. */
2577 inode->i_mode = (inode->i_mode & S_IFMT) |
2578 (attr->la_mode & ~S_IFMT);
2580 i_uid_write(inode, attr->la_uid);
2582 i_gid_write(inode, attr->la_gid);
2583 if (bits & LA_NLINK)
2584 set_nlink(inode, attr->la_nlink);
2586 inode->i_rdev = attr->la_rdev;
2588 if (bits & LA_FLAGS) {
2589 /* always keep S_NOCMTIME */
2590 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2596 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2598 if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2599 (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2603 ll_vfs_dq_init(inode);
2605 if (attr->la_valid & LA_UID)
2606 iattr.ia_valid |= ATTR_UID;
2607 if (attr->la_valid & LA_GID)
2608 iattr.ia_valid |= ATTR_GID;
2609 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2610 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2612 rc = ll_vfs_dq_transfer(inode, &iattr);
2614 CERROR("%s: quota transfer failed: rc = %d. Is quota "
2615 "enforcement enabled on the ldiskfs "
2616 "filesystem?\n", inode->i_sb->s_id, rc);
2623 static int osd_attr_set(const struct lu_env *env,
2624 struct dt_object *dt,
2625 const struct lu_attr *attr,
2626 struct thandle *handle)
2628 struct osd_object *obj = osd_dt_obj(dt);
2629 struct inode *inode;
2632 if (!dt_object_exists(dt))
2635 LASSERT(handle != NULL);
2636 LASSERT(!dt_object_remote(dt));
2637 LASSERT(osd_invariant(obj));
2639 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2641 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
2642 struct osd_thread_info *oti = osd_oti_get(env);
2643 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
2644 struct lu_fid *fid1 = &oti->oti_fid;
2645 struct osd_inode_id *id = &oti->oti_id;
2646 struct iam_path_descr *ipd;
2647 struct iam_container *bag;
2648 struct osd_thandle *oh;
2651 fid_cpu_to_be(fid1, fid0);
2652 memset(id, 1, sizeof(*id));
2653 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2654 fid0)->oi_dir.od_container;
2655 ipd = osd_idx_ipd_get(env, bag);
2656 if (unlikely(ipd == NULL))
2659 oh = container_of0(handle, struct osd_thandle, ot_super);
2660 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2661 (const struct iam_rec *)id, ipd);
2662 osd_ipd_put(env, bag, ipd);
2663 return(rc > 0 ? 0 : rc);
2666 inode = obj->oo_inode;
2668 rc = osd_quota_transfer(inode, attr);
2672 spin_lock(&obj->oo_guard);
2673 rc = osd_inode_setattr(env, inode, attr);
2674 spin_unlock(&obj->oo_guard);
2678 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2680 if (!(attr->la_valid & LA_FLAGS))
2683 /* Let's check if there are extra flags need to be set into LMA */
2684 if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
2685 struct osd_thread_info *info = osd_oti_get(env);
2686 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
2688 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
2689 &info->oti_ost_attrs);
2693 lma->lma_incompat |=
2694 lustre_to_lma_flags(attr->la_flags);
2695 lustre_lma_swab(lma);
2696 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
2697 lma, sizeof(*lma), XATTR_REPLACE);
2699 struct osd_device *osd = osd_obj2dev(obj);
2701 CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
2702 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
2703 lma->lma_incompat, rc);
2706 attr->la_flags & LUSTRE_LMA_FL_MASKS;
2708 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2711 osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2716 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2717 struct osd_object *obj,
2718 const char *name, const int namelen)
2720 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2723 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2724 umode_t mode, struct dt_allocation_hint *hint,
2728 struct osd_device *osd = osd_obj2dev(obj);
2729 struct osd_thandle *oth;
2730 struct dt_object *parent = NULL;
2731 struct inode *inode;
2733 LINVRNT(osd_invariant(obj));
2734 LASSERT(obj->oo_inode == NULL);
2735 LASSERT(obj->oo_hl_head == NULL);
2737 if (S_ISDIR(mode) && ldiskfs_pdo) {
2738 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2739 if (obj->oo_hl_head == NULL)
2743 oth = container_of(th, struct osd_thandle, ot_super);
2744 LASSERT(oth->ot_handle->h_transaction != NULL);
2746 if (hint != NULL && hint->dah_parent != NULL &&
2747 !dt_object_remote(hint->dah_parent))
2748 parent = hint->dah_parent;
2750 inode = ldiskfs_create_inode(oth->ot_handle,
2751 parent ? osd_dt_obj(parent)->oo_inode :
2752 osd_sb(osd)->s_root->d_inode,
2754 if (!IS_ERR(inode)) {
2755 /* Do not update file c/mtime in ldiskfs. */
2756 inode->i_flags |= S_NOCMTIME;
2758 /* For new created object, it must be consistent,
2759 * and it is unnecessary to scrub against it. */
2760 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2762 obj->oo_inode = inode;
2765 if (obj->oo_hl_head != NULL) {
2766 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2767 obj->oo_hl_head = NULL;
2769 result = PTR_ERR(inode);
2771 LINVRNT(osd_invariant(obj));
2779 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2780 struct lu_attr *attr,
2781 struct dt_allocation_hint *hint,
2782 struct dt_object_format *dof,
2786 struct osd_thandle *oth;
2787 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
2789 LASSERT(S_ISDIR(attr->la_mode));
2791 oth = container_of(th, struct osd_thandle, ot_super);
2792 LASSERT(oth->ot_handle->h_transaction != NULL);
2793 result = osd_mkfile(info, obj, mode, hint, th);
2798 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2799 struct lu_attr *attr,
2800 struct dt_allocation_hint *hint,
2801 struct dt_object_format *dof,
2805 struct osd_thandle *oth;
2806 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2808 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2810 LASSERT(S_ISREG(attr->la_mode));
2812 oth = container_of(th, struct osd_thandle, ot_super);
2813 LASSERT(oth->ot_handle->h_transaction != NULL);
2815 result = osd_mkfile(info, obj, mode, hint, th);
2817 LASSERT(obj->oo_inode != NULL);
2818 if (feat->dif_flags & DT_IND_VARKEY)
2819 result = iam_lvar_create(obj->oo_inode,
2820 feat->dif_keysize_max,
2822 feat->dif_recsize_max,
2825 result = iam_lfix_create(obj->oo_inode,
2826 feat->dif_keysize_max,
2828 feat->dif_recsize_max,
2835 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2836 struct lu_attr *attr,
2837 struct dt_allocation_hint *hint,
2838 struct dt_object_format *dof,
2841 LASSERT(S_ISREG(attr->la_mode));
2842 return osd_mkfile(info, obj, (attr->la_mode &
2843 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2846 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2847 struct lu_attr *attr,
2848 struct dt_allocation_hint *hint,
2849 struct dt_object_format *dof,
2852 LASSERT(S_ISLNK(attr->la_mode));
2853 return osd_mkfile(info, obj, (attr->la_mode &
2854 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2857 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2858 struct lu_attr *attr,
2859 struct dt_allocation_hint *hint,
2860 struct dt_object_format *dof,
2863 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2866 LINVRNT(osd_invariant(obj));
2867 LASSERT(obj->oo_inode == NULL);
2868 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2869 S_ISFIFO(mode) || S_ISSOCK(mode));
2871 result = osd_mkfile(info, obj, mode, hint, th);
2873 LASSERT(obj->oo_inode != NULL);
2875 * This inode should be marked dirty for i_rdev. Currently
2876 * that is done in the osd_attr_init().
2878 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2881 LINVRNT(osd_invariant(obj));
2885 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2887 struct dt_allocation_hint *hint,
2888 struct dt_object_format *dof,
2891 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2893 osd_obj_type_f result;
2909 result = osd_mk_index;
2920 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2921 struct dt_object *parent, struct dt_object *child,
2926 ah->dah_parent = parent;
2927 ah->dah_mode = child_mode;
2929 if (parent != NULL && !dt_object_remote(parent)) {
2930 /* will help to find FID->ino at dt_insert("..") */
2931 struct osd_object *pobj = osd_dt_obj(parent);
2932 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
2936 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2937 struct lu_attr *attr, struct dt_object_format *dof)
2939 struct inode *inode = obj->oo_inode;
2940 __u64 valid = attr->la_valid;
2943 attr->la_valid &= ~(LA_TYPE | LA_MODE);
2945 if (dof->dof_type != DFT_NODE)
2946 attr->la_valid &= ~LA_RDEV;
2947 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2948 attr->la_valid &= ~LA_ATIME;
2949 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2950 attr->la_valid &= ~LA_CTIME;
2951 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2952 attr->la_valid &= ~LA_MTIME;
2954 result = osd_quota_transfer(inode, attr);
2958 if (attr->la_valid != 0) {
2959 result = osd_inode_setattr(info->oti_env, inode, attr);
2961 * The osd_inode_setattr() should always succeed here. The
2962 * only error that could be returned is EDQUOT when we are
2963 * trying to change the UID or GID of the inode. However, this
2964 * should not happen since quota enforcement is no longer
2965 * enabled on ldiskfs (lquota takes care of it).
2967 LASSERTF(result == 0, "%d\n", result);
2968 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2971 attr->la_valid = valid;
2975 * Helper function for osd_object_create()
2977 * \retval 0, on success
2979 static int __osd_object_create(struct osd_thread_info *info,
2980 struct osd_object *obj, struct lu_attr *attr,
2981 struct dt_allocation_hint *hint,
2982 struct dt_object_format *dof,
2988 osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
2990 /* we drop umask so that permissions we pass are not affected */
2991 umask = current->fs->umask;
2992 current->fs->umask = 0;
2994 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2996 if (likely(obj->oo_inode != NULL)) {
2997 LASSERT(obj->oo_inode->i_state & I_NEW);
2999 /* Unlock the inode before attr initialization to avoid
3000 * unnecessary dqget operations. LU-6378 */
3001 unlock_new_inode(obj->oo_inode);
3004 if (likely(result == 0)) {
3005 osd_attr_init(info, obj, attr, dof);
3006 osd_object_init0(obj);
3009 /* restore previous umask value */
3010 current->fs->umask = umask;
3012 osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3018 * Helper function for osd_object_create()
3020 * \retval 0, on success
3022 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3023 const struct lu_fid *fid, struct thandle *th)
3025 struct osd_thread_info *info = osd_oti_get(env);
3026 struct osd_inode_id *id = &info->oti_id;
3027 struct osd_device *osd = osd_obj2dev(obj);
3028 struct osd_thandle *oh;
3031 LASSERT(obj->oo_inode != NULL);
3033 oh = container_of0(th, struct osd_thandle, ot_super);
3034 LASSERT(oh->ot_handle);
3035 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3037 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3038 rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3039 OI_CHECK_FLD, NULL);
3040 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3045 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3046 u64 seq, struct lu_seq_range *range)
3048 struct seq_server_site *ss = osd_seq_site(osd);
3050 if (fid_seq_is_idif(seq)) {
3051 fld_range_set_ost(range);
3052 range->lsr_index = idif_ost_idx(seq);
3056 if (!fid_seq_in_fldb(seq)) {
3057 fld_range_set_mdt(range);
3059 /* FIXME: If ss is NULL, it suppose not get lsr_index
3061 range->lsr_index = ss->ss_node_id;
3065 LASSERT(ss != NULL);
3066 fld_range_set_any(range);
3067 /* OSD will only do local fld lookup */
3068 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3072 * Concurrency: no external locking is necessary.
3074 static int osd_declare_object_create(const struct lu_env *env,
3075 struct dt_object *dt,
3076 struct lu_attr *attr,
3077 struct dt_allocation_hint *hint,
3078 struct dt_object_format *dof,
3079 struct thandle *handle)
3081 struct osd_thandle *oh;
3085 LASSERT(handle != NULL);
3087 oh = container_of0(handle, struct osd_thandle, ot_super);
3088 LASSERT(oh->ot_handle == NULL);
3090 /* EA object consumes more credits than regular object: osd_mk_index
3091 * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3092 * leaf_node, could involves the block, block bitmap, groups, GDT
3093 * change for each block, so add 4 * 2 credits in that case. */
3094 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3095 osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3096 (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3097 /* Reuse idle OI block may cause additional one OI block
3099 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3100 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3105 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
3106 osd_dt_obj(dt), false, NULL, false);
3110 /* will help to find FID->ino mapping at dt_insert() */
3111 rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3117 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
3118 struct lu_attr *attr,
3119 struct dt_allocation_hint *hint,
3120 struct dt_object_format *dof, struct thandle *th)
3122 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3123 struct osd_object *obj = osd_dt_obj(dt);
3124 struct osd_thread_info *info = osd_oti_get(env);
3128 if (dt_object_exists(dt))
3131 LINVRNT(osd_invariant(obj));
3132 LASSERT(!dt_object_remote(dt));
3133 LASSERT(osd_write_locked(env, obj));
3134 LASSERT(th != NULL);
3136 if (unlikely(fid_is_acct(fid)))
3137 /* Quota files can't be created from the kernel any more,
3138 * 'tune2fs -O quota' will take care of creating them */
3141 result = __osd_object_create(info, obj, attr, hint, dof, th);
3143 result = __osd_oi_insert(env, obj, fid, th);
3144 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3145 obj->oo_dt.do_body_ops = &osd_body_ops;
3147 LASSERT(ergo(result == 0,
3148 dt_object_exists(dt) && !dt_object_remote(dt)));
3150 LASSERT(osd_invariant(obj));
3155 * Called to destroy on-disk representation of the object
3157 * Concurrency: must be locked
3159 static int osd_declare_object_destroy(const struct lu_env *env,
3160 struct dt_object *dt,
3163 struct osd_object *obj = osd_dt_obj(dt);
3164 struct inode *inode = obj->oo_inode;
3165 struct osd_thandle *oh;
3172 oh = container_of0(th, struct osd_thandle, ot_super);
3173 LASSERT(oh->ot_handle == NULL);
3175 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3176 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3177 /* Recycle idle OI leaf may cause additional three OI blocks
3179 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3180 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3181 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3182 /* one less inode */
3183 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3184 -1, oh, obj, false, NULL, false);
3187 /* data to be truncated */
3188 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3189 0, oh, obj, true, NULL, false);
3193 /* will help to find FID->ino when this object is being
3194 * added to PENDING/ */
3195 rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3200 static int osd_object_destroy(const struct lu_env *env,
3201 struct dt_object *dt,
3204 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3205 struct osd_object *obj = osd_dt_obj(dt);
3206 struct inode *inode = obj->oo_inode;
3207 struct osd_device *osd = osd_obj2dev(obj);
3208 struct osd_thandle *oh;
3212 oh = container_of0(th, struct osd_thandle, ot_super);
3213 LASSERT(oh->ot_handle);
3215 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3217 if (unlikely(fid_is_acct(fid)))
3220 if (S_ISDIR(inode->i_mode)) {
3221 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3222 inode->i_nlink == 2);
3223 /* it will check/delete the inode from remote parent,
3224 * how to optimize it? unlink performance impaction XXX */
3225 result = osd_delete_from_remote_parent(env, osd, obj, oh);
3227 CERROR("%s: delete inode "DFID": rc = %d\n",
3228 osd_name(osd), PFID(fid), result);
3230 spin_lock(&obj->oo_guard);
3232 spin_unlock(&obj->oo_guard);
3233 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3236 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3238 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3240 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3241 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3242 oh->ot_handle, OI_CHECK_FLD);
3244 osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3245 /* XXX: add to ext3 orphan list */
3246 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3248 /* not needed in the cache anymore */
3249 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3250 obj->oo_destroyed = 1;
3256 * Put the fid into lustre_mdt_attrs, and then place the structure
3257 * inode's ea. This fid should not be altered during the life time
3260 * \retval +ve, on success
3261 * \retval -ve, on error
3263 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3265 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3266 const struct lu_fid *fid, __u32 compat, __u32 incompat)
3268 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3269 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3273 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3276 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3279 lustre_loa_init(loa, fid, compat, incompat);
3280 lustre_loa_swab(loa, false);
3282 /* For the OST device with 256 bytes inode size by default,
3283 * the PFID EA will be stored together with LMA EA to avoid
3284 * performance trouble. Otherwise the PFID EA can be stored
3285 * independently. LU-8998 */
3286 if ((compat & LMAC_FID_ON_OST) &&
3287 LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3288 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3289 sizeof(*loa), XATTR_CREATE);
3291 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3292 sizeof(*lma), XATTR_CREATE);
3293 /* LMA may already exist, but we need to check that all the
3294 * desired compat/incompat flags have been added. */
3295 if (unlikely(rc == -EEXIST)) {
3296 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3297 XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3301 if (rc < sizeof(*lma))
3304 lustre_loa_swab(loa, true);
3305 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3306 ((compat == 0 && incompat == 0) ||
3307 (!(~lma->lma_compat & compat) &&
3308 !(~lma->lma_incompat & incompat))))
3311 lma->lma_self_fid = *fid;
3312 lma->lma_compat |= compat;
3313 lma->lma_incompat |= incompat;
3314 if (rc == sizeof(*lma)) {
3315 lustre_lma_swab(lma);
3316 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3317 sizeof(*lma), XATTR_REPLACE);
3319 lustre_loa_swab(loa, false);
3320 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3321 sizeof(*loa), XATTR_REPLACE);
3329 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3330 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3331 * To have compatilibility with 1.8 ldiskfs driver we need to have
3332 * magic number at start of fid data.
3333 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3336 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3337 const struct lu_fid *fid)
3339 if (!fid_is_namespace_visible(fid) ||
3340 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3341 param->edp_magic = 0;
3345 param->edp_magic = LDISKFS_LUFID_MAGIC;
3346 param->edp_len = sizeof(struct lu_fid) + 1;
3347 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3351 * Try to read the fid from inode ea into dt_rec.
3353 * \param fid object fid.
3355 * \retval 0 on success
3357 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3358 __u32 ino, struct lu_fid *fid,
3359 struct osd_inode_id *id)
3361 struct osd_thread_info *info = osd_oti_get(env);
3362 struct inode *inode;
3365 osd_id_gen(id, ino, OSD_OII_NOGEN);
3366 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3368 RETURN(PTR_ERR(inode));
3374 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3376 struct inode *parent_dir,
3377 const struct lu_fid *dot_fid,
3378 const struct lu_fid *dot_dot_fid,
3379 struct osd_thandle *oth)
3381 struct ldiskfs_dentry_param *dot_ldp;
3382 struct ldiskfs_dentry_param *dot_dot_ldp;
3383 __u32 saved_nlink = dir->i_nlink;
3386 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3387 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3389 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3390 dot_ldp->edp_magic = 0;
3392 rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3393 dir, dot_ldp, dot_dot_ldp);
3394 /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3395 * the subseqent ref_add() will increase the dir->i_nlink
3396 * as 3. That is incorrect for new created directory.
3398 * It looks like hack, because we want to make the OSD API
3399 * to be order-independent for new created directory object
3400 * between dt_insert(..) and ref_add() operations.
3402 * Here, we only restore the in-RAM dir-inode's nlink attr,
3403 * becuase if the nlink attr is not 2, then there will be
3404 * ref_add() called following the dt_insert(..), such call
3405 * will make both the in-RAM and on-disk dir-inode's nlink
3406 * attr to be set as 2. LU-7447 */
3407 set_nlink(dir, saved_nlink);
3412 * Create an local agent inode for remote entry
3414 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3415 struct osd_device *osd,
3416 struct osd_object *pobj,
3417 const struct lu_fid *fid,
3421 struct osd_thread_info *info = osd_oti_get(env);
3422 struct inode *local;
3423 struct osd_thandle *oh;
3428 oh = container_of(th, struct osd_thandle, ot_super);
3429 LASSERT(oh->ot_handle->h_transaction != NULL);
3431 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3432 if (IS_ERR(local)) {
3433 CERROR("%s: create local error %d\n", osd_name(osd),
3434 (int)PTR_ERR(local));
3438 /* restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3439 * correct gid on remote file, not agent here */
3440 local->i_gid = current_fsgid();
3441 ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3442 unlock_new_inode(local);
3444 /* Set special LMA flag for local agent inode */
3445 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3447 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3448 osd_name(osd), PFID(fid), rc);
3449 RETURN(ERR_PTR(rc));
3455 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3456 lu_object_fid(&pobj->oo_dt.do_lu),
3459 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3460 osd_name(osd), PFID(fid), rc);
3461 RETURN(ERR_PTR(rc));
3468 * when direntry is deleted, we have to take care of possible agent inode
3469 * referenced by that. unfortunately we can't do this at that point:
3470 * iget() within a running transaction leads to deadlock and we better do
3471 * not call that every delete declaration to save performance. so we put
3472 * a potention agent inode on a list and process that once the transaction
3473 * is over. Notice it's not any worse in terms of real orphans as regular
3474 * object destroy doesn't put inodes on the on-disk orphan list. this should
3475 * be addressed separately
3477 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3478 struct osd_thandle *oh,
3481 struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3482 struct osd_obj_orphan *oor;
3489 oor->oor_env = (struct lu_env *)env;
3490 spin_lock(&osd->od_osfs_lock);
3491 list_add(&oor->oor_list, &osd->od_orphan_list);
3492 spin_unlock(&osd->od_osfs_lock);
3494 oh->ot_remove_agents = 1;
3500 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3501 struct osd_device *osd)
3503 struct osd_thread_info *info = osd_oti_get(env);
3504 struct osd_obj_orphan *oor, *tmp;
3505 struct osd_inode_id id;
3506 struct list_head list;
3507 struct inode *inode;
3512 INIT_LIST_HEAD(&list);
3514 spin_lock(&osd->od_osfs_lock);
3515 list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3516 if (oor->oor_env == env) {
3517 list_del(&oor->oor_list);
3518 list_add(&oor->oor_list, &list);
3521 spin_unlock(&osd->od_osfs_lock);
3523 list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3527 list_del(&oor->oor_list);
3530 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3531 inode = osd_iget_fid(info, osd, &id, &fid);
3535 if (!osd_remote_fid(env, osd, &fid)) {
3540 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3542 mark_inode_dirty(inode);
3543 ldiskfs_journal_stop(jh);
3551 * OSD layer object create function for interoperability mode (b11826).
3552 * This is mostly similar to osd_object_create(). Only difference being, fid is
3553 * inserted into inode ea here.
3555 * \retval 0, on success
3556 * \retval -ve, on error
3558 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
3559 struct lu_attr *attr,
3560 struct dt_allocation_hint *hint,
3561 struct dt_object_format *dof,
3564 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3565 struct osd_object *obj = osd_dt_obj(dt);
3566 struct osd_thread_info *info = osd_oti_get(env);
3567 int result, on_ost = 0;
3571 if (dt_object_exists(dt))
3574 LASSERT(osd_invariant(obj));
3575 LASSERT(!dt_object_remote(dt));
3576 LASSERT(osd_write_locked(env, obj));
3577 LASSERT(th != NULL);
3579 if (unlikely(fid_is_acct(fid)))
3580 /* Quota files can't be created from the kernel any more,
3581 * 'tune2fs -O quota' will take care of creating them */
3584 result = __osd_object_create(info, obj, attr, hint, dof, th);
3586 if (fid_is_idif(fid) &&
3587 !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3588 struct lu_fid *tfid = &info->oti_fid;
3589 struct ost_id *oi = &info->oti_ostid;
3591 fid_to_ostid(fid, oi);
3592 ostid_to_fid(tfid, oi, 0);
3594 result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3595 LMAC_FID_ON_OST, 0);
3597 on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3599 result = osd_ea_fid_set(info, obj->oo_inode, fid,
3600 on_ost ? LMAC_FID_ON_OST : 0,
3603 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3604 obj->oo_dt.do_body_ops = &osd_body_ops;
3608 result = __osd_oi_insert(env, obj, fid, th);
3610 /* a small optimization - dt_insert() isn't usually applied
3611 * to OST objects, so we don't need to cache OI mapping for
3613 if (result == 0 && on_ost == 0) {
3614 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3615 result = osd_idc_find_and_init(env, osd, obj);
3616 LASSERT(result == 0);
3619 LASSERT(ergo(result == 0,
3620 dt_object_exists(dt) && !dt_object_remote(dt)));
3621 LINVRNT(osd_invariant(obj));
3625 static int osd_declare_object_ref_add(const struct lu_env *env,
3626 struct dt_object *dt,
3627 struct thandle *handle)
3629 struct osd_thandle *oh;
3631 /* it's possible that object doesn't exist yet */
3632 LASSERT(handle != NULL);
3634 oh = container_of0(handle, struct osd_thandle, ot_super);
3635 LASSERT(oh->ot_handle == NULL);
3637 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3638 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3644 * Concurrency: @dt is write locked.
3646 static int osd_object_ref_add(const struct lu_env *env,
3647 struct dt_object *dt, struct thandle *th)
3649 struct osd_object *obj = osd_dt_obj(dt);
3650 struct inode *inode = obj->oo_inode;
3651 struct osd_thandle *oh;
3654 if (!dt_object_exists(dt) || obj->oo_destroyed)
3657 LINVRNT(osd_invariant(obj));
3658 LASSERT(!dt_object_remote(dt));
3659 LASSERT(osd_write_locked(env, obj));
3660 LASSERT(th != NULL);
3662 oh = container_of0(th, struct osd_thandle, ot_super);
3663 LASSERT(oh->ot_handle != NULL);
3665 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
3667 CDEBUG(D_INODE, DFID" increase nlink %d\n",
3668 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3670 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
3671 * (65000) subdirectories by storing "1" in i_nlink if the link count
3672 * would otherwise overflow. Directory tranversal tools understand
3673 * that (st_nlink == 1) indicates that the filesystem dose not track
3674 * hard links count on the directory, and will not abort subdirectory
3675 * scanning early once (st_nlink - 2) subdirs have been found.
3677 * This also has to properly handle the case of inodes with nlink == 0
3678 * in case they are being linked into the PENDING directory
3680 spin_lock(&obj->oo_guard);
3681 if (unlikely(inode->i_nlink == 0))
3682 /* inc_nlink from 0 may cause WARN_ON */
3683 set_nlink(inode, 1);
3685 ldiskfs_inc_count(oh->ot_handle, inode);
3686 if (!S_ISDIR(inode->i_mode))
3687 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
3689 spin_unlock(&obj->oo_guard);
3691 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3692 LINVRNT(osd_invariant(obj));
3694 osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
3699 static int osd_declare_object_ref_del(const struct lu_env *env,
3700 struct dt_object *dt,
3701 struct thandle *handle)
3703 struct osd_thandle *oh;
3705 if (!dt_object_exists(dt))
3708 LASSERT(!dt_object_remote(dt));
3709 LASSERT(handle != NULL);
3711 oh = container_of0(handle, struct osd_thandle, ot_super);
3712 LASSERT(oh->ot_handle == NULL);
3714 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
3715 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3721 * Concurrency: @dt is write locked.
3723 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
3726 struct osd_object *obj = osd_dt_obj(dt);
3727 struct inode *inode = obj->oo_inode;
3728 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3729 struct osd_thandle *oh;
3731 if (!dt_object_exists(dt))
3734 LINVRNT(osd_invariant(obj));
3735 LASSERT(!dt_object_remote(dt));
3736 LASSERT(osd_write_locked(env, obj));
3737 LASSERT(th != NULL);
3739 oh = container_of0(th, struct osd_thandle, ot_super);
3740 LASSERT(oh->ot_handle != NULL);
3742 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
3744 spin_lock(&obj->oo_guard);
3745 /* That can be result of upgrade from old Lustre version and
3746 * applied only to local files. Just skip this ref_del call.
3747 * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
3748 if (inode->i_nlink == 0) {
3749 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
3750 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
3751 ", maybe an upgraded file? (LU-3915)\n",
3752 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
3753 spin_unlock(&obj->oo_guard);
3757 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
3758 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3760 ldiskfs_dec_count(oh->ot_handle, inode);
3761 spin_unlock(&obj->oo_guard);
3763 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3764 LINVRNT(osd_invariant(obj));
3766 osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
3772 * Concurrency: @dt is read locked.
3774 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
3775 struct lu_buf *buf, const char *name)
3777 struct osd_object *obj = osd_dt_obj(dt);
3778 struct inode *inode = obj->oo_inode;
3779 struct osd_thread_info *info = osd_oti_get(env);
3780 struct dentry *dentry = &info->oti_obj_dentry;
3781 bool cache_xattr = false;
3786 /* version get is not real XATTR but uses xattr API */
3787 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3788 dt_obj_version_t *ver = buf->lb_buf;
3790 /* for version we are just using xattr API but change inode
3792 if (buf->lb_len == 0)
3793 return sizeof(dt_obj_version_t);
3795 if (buf->lb_len < sizeof(dt_obj_version_t))
3798 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
3799 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3801 *ver = LDISKFS_I(inode)->i_fs_version;
3803 return sizeof(dt_obj_version_t);
3806 if (!dt_object_exists(dt))
3809 LASSERT(!dt_object_remote(dt));
3810 LASSERT(inode->i_op != NULL);
3811 LASSERT(inode->i_op->getxattr != NULL);
3813 if (strcmp(name, XATTR_NAME_LOV) == 0 ||
3814 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
3818 rc = osd_oxc_get(obj, name, buf);
3823 rc = __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
3824 if (rc == -ENODATA && strcmp(name, XATTR_NAME_FID) == 0) {
3825 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3826 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3827 struct filter_fid *ff;
3828 struct ost_layout *ol;
3830 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
3832 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
3836 if (!(lma->lma_compat & LMAC_STRIPE_INFO)) {
3842 if (buf->lb_len == 0 || !buf->lb_buf)
3845 if (buf->lb_len < rc)
3849 ol = &ff->ff_layout;
3850 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
3851 PFID_STRIPE_IDX_BITS);
3852 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
3853 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
3854 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
3855 if (lma->lma_compat & LMAC_COMP_INFO) {
3856 ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
3857 ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
3858 ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
3860 ol->ol_comp_start = 0;
3861 ol->ol_comp_end = 0;
3868 if (rc == -ENOENT || rc == -ENODATA)
3869 osd_oxc_add(obj, name, NULL, 0);
3870 else if (rc > 0 && buf->lb_buf != NULL)
3871 osd_oxc_add(obj, name, buf->lb_buf, rc);
3877 static int osd_declare_xattr_set(const struct lu_env *env,
3878 struct dt_object *dt,
3879 const struct lu_buf *buf, const char *name,
3880 int fl, struct thandle *handle)
3882 struct osd_thandle *oh;
3884 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3886 LASSERT(handle != NULL);
3888 oh = container_of0(handle, struct osd_thandle, ot_super);
3889 LASSERT(oh->ot_handle == NULL);
3891 if (strcmp(name, XATTR_NAME_LMA) == 0) {
3892 /* For non-upgrading case, the LMA is set first and
3893 * usually fit inode. But for upgrade case, the LMA
3894 * may be in another separated EA block. */
3895 if (dt_object_exists(dt)) {
3896 if (fl == LU_XATTR_REPLACE)
3901 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3903 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3904 /* We may need to delete the old PFID EA. */
3905 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
3906 if (fl == LU_XATTR_REPLACE)
3913 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
3918 if (buf->lb_buf == NULL && dt_object_exists(dt)) {
3919 /* learn xattr size from osd_xattr_get if
3920 attribute has not been read yet */
3921 buflen = __osd_xattr_get(
3922 osd_dt_obj(dt)->oo_inode,
3923 &osd_oti_get(env)->oti_obj_dentry,
3928 buflen = buf->lb_len;
3931 if (buflen > sb->s_blocksize) {
3932 credits += osd_calc_bkmap_credits(
3934 (buflen + sb->s_blocksize - 1) >>
3935 sb->s_blocksize_bits);
3939 * xattr set may involve inode quota change, reserve credits for
3940 * dquot_initialize()
3942 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3945 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3951 * Concurrency: @dt is write locked.
3953 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3954 const struct lu_buf *buf, const char *name, int fl,
3955 struct thandle *handle)
3957 struct osd_object *obj = osd_dt_obj(dt);
3958 struct inode *inode = obj->oo_inode;
3959 struct osd_thread_info *info = osd_oti_get(env);
3960 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3961 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3970 /* version set is not real XATTR */
3971 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3972 dt_obj_version_t *version = buf->lb_buf;
3974 /* for version we are just using xattr API but change inode
3976 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3978 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
3979 *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3981 LDISKFS_I(inode)->i_fs_version = *version;
3982 /* Version is set after all inode operations are finished,
3983 * so we should mark it dirty here */
3984 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3989 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
3990 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3993 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3994 if (fl & LU_XATTR_REPLACE)
3995 fs_flags |= XATTR_REPLACE;
3997 if (fl & LU_XATTR_CREATE)
3998 fs_flags |= XATTR_CREATE;
4000 /* For the OST device with 256 bytes inode size by default,
4001 * the PFID EA will be stored together with LMA EA to avoid
4002 * performance trouble. Otherwise the PFID EA can be stored
4003 * independently. LU-8998 */
4004 if (strcmp(name, XATTR_NAME_FID) == 0 &&
4005 LDISKFS_INODE_SIZE(inode->i_sb) <= 256) {
4006 struct dentry *dentry = &info->oti_obj_dentry;
4007 struct filter_fid *ff;
4008 struct ost_layout *ol;
4011 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
4014 ol = &ff->ff_layout;
4015 /* Old client does not send stripe information, store
4016 * the PFID EA on disk directly. */
4017 if (buf->lb_len == sizeof(struct lu_fid) ||
4018 ol->ol_stripe_size == 0) {
4019 len = sizeof(struct lu_fid);
4023 if (buf->lb_len != sizeof(*ff))
4026 rc = osd_get_lma(info, inode, dentry, loa);
4027 if (unlikely(rc == -ENODATA)) {
4028 /* Usually for upgarding from old device */
4029 lustre_loa_init(loa, lu_object_fid(&dt->do_lu),
4030 LMAC_FID_ON_OST, 0);
4038 fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4039 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4040 PFID_STRIPE_IDX_BITS;
4041 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4042 lma->lma_compat |= LMAC_STRIPE_INFO;
4043 if (ol->ol_comp_id != 0) {
4044 loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4045 loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4046 loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4047 lma->lma_compat |= LMAC_COMP_INFO;
4050 lustre_loa_swab(loa, false);
4052 /* Remove old PFID EA entry firstly. */
4053 ll_vfs_dq_init(inode);
4054 rc = inode->i_op->removexattr(dentry, name);
4055 if (rc && rc != -ENODATA)
4058 /* Store the PFID EA inside the LMA EA. */
4059 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
4063 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4064 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4068 lma->lma_incompat |= LMAI_STRIPED;
4069 lustre_lma_swab(lma);
4070 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4071 sizeof(*lma), XATTR_REPLACE);
4077 rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4078 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4081 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4082 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4083 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4089 * Concurrency: @dt is read locked.
4091 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4092 const struct lu_buf *buf)
4094 struct osd_object *obj = osd_dt_obj(dt);
4095 struct inode *inode = obj->oo_inode;
4096 struct osd_thread_info *info = osd_oti_get(env);
4097 struct dentry *dentry = &info->oti_obj_dentry;
4099 if (!dt_object_exists(dt))
4102 LASSERT(!dt_object_remote(dt));
4103 LASSERT(inode->i_op != NULL);
4104 LASSERT(inode->i_op->listxattr != NULL);
4106 dentry->d_inode = inode;
4107 dentry->d_sb = inode->i_sb;
4108 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4111 static int osd_declare_xattr_del(const struct lu_env *env,
4112 struct dt_object *dt, const char *name,
4113 struct thandle *handle)
4115 struct osd_thandle *oh;
4116 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4118 LASSERT(!dt_object_remote(dt));
4119 LASSERT(handle != NULL);
4121 oh = container_of0(handle, struct osd_thandle, ot_super);
4122 LASSERT(oh->ot_handle == NULL);
4124 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4125 osd_dto_credits_noquota[DTO_XATTR_SET]);
4127 * xattr del may involve inode quota change, reserve credits for
4128 * dquot_initialize()
4130 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4136 * Concurrency: @dt is write locked.
4138 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4139 const char *name, struct thandle *handle)
4141 struct osd_object *obj = osd_dt_obj(dt);
4142 struct inode *inode = obj->oo_inode;
4143 struct osd_thread_info *info = osd_oti_get(env);
4144 struct dentry *dentry = &info->oti_obj_dentry;
4147 if (!dt_object_exists(dt))
4150 LASSERT(!dt_object_remote(dt));
4151 LASSERT(inode->i_op != NULL);
4152 LASSERT(inode->i_op->removexattr != NULL);
4153 LASSERT(handle != NULL);
4155 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4157 ll_vfs_dq_init(inode);
4158 dentry->d_inode = inode;
4159 dentry->d_sb = inode->i_sb;
4160 rc = inode->i_op->removexattr(dentry, name);
4161 if (rc == -ENODATA && strcmp(name, XATTR_NAME_FID) == 0) {
4162 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4164 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
4166 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4167 &info->oti_ost_attrs);
4169 if (!(lma->lma_compat & LMAC_STRIPE_INFO)) {
4174 lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4175 lustre_lma_swab(lma);
4176 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4177 sizeof(*lma), XATTR_REPLACE);
4182 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4185 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4186 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4187 osd_oxc_del(obj, name);
4192 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4193 __u64 start, __u64 end)
4195 struct osd_object *obj = osd_dt_obj(dt);
4196 struct inode *inode = obj->oo_inode;
4197 struct osd_thread_info *info = osd_oti_get(env);
4198 struct dentry *dentry = &info->oti_obj_dentry;
4199 struct file *file = &info->oti_file;
4204 dentry->d_inode = inode;
4205 dentry->d_sb = inode->i_sb;
4206 file->f_path.dentry = dentry;
4207 file->f_mapping = inode->i_mapping;
4208 file->f_op = inode->i_fop;
4209 set_file_inode(file, inode);
4211 rc = ll_vfs_fsync_range(file, start, end, 0);
4216 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4225 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4226 const struct dt_index_features *feat)
4228 struct iam_descr *descr;
4230 if (osd_object_is_root(o))
4231 return feat == &dt_directory_features;
4233 LASSERT(o->oo_dir != NULL);
4235 descr = o->oo_dir->od_container.ic_descr;
4236 if (feat == &dt_directory_features) {
4237 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4243 feat->dif_keysize_min <= descr->id_key_size &&
4244 descr->id_key_size <= feat->dif_keysize_max &&
4245 feat->dif_recsize_min <= descr->id_rec_size &&
4246 descr->id_rec_size <= feat->dif_recsize_max &&
4247 !(feat->dif_flags & (DT_IND_VARKEY |
4248 DT_IND_VARREC | DT_IND_NONUNQ)) &&
4249 ergo(feat->dif_flags & DT_IND_UPDATE,
4250 1 /* XXX check that object (and file system) is
4255 static int osd_iam_container_init(const struct lu_env *env,
4256 struct osd_object *obj,
4257 struct osd_directory *dir)
4259 struct iam_container *bag = &dir->od_container;
4262 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4266 result = iam_container_setup(bag);
4268 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4270 iam_container_fini(bag);
4277 * Concurrency: no external locking is necessary.
4279 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4280 const struct dt_index_features *feat)
4284 struct osd_object *obj = osd_dt_obj(dt);
4286 LINVRNT(osd_invariant(obj));
4288 if (osd_object_is_root(obj)) {
4289 dt->do_index_ops = &osd_index_ea_ops;
4291 } else if (feat == &dt_directory_features) {
4292 dt->do_index_ops = &osd_index_ea_ops;
4293 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4298 } else if (unlikely(feat == &dt_otable_features)) {
4299 dt->do_index_ops = &osd_otable_ops;
4301 } else if (unlikely(feat == &dt_acct_features)) {
4302 dt->do_index_ops = &osd_acct_index_ops;
4305 } else if (!osd_has_index(obj)) {
4306 struct osd_directory *dir;
4311 spin_lock(&obj->oo_guard);
4312 if (obj->oo_dir == NULL)
4316 * Concurrent thread allocated container data.
4319 spin_unlock(&obj->oo_guard);
4321 * Now, that we have container data, serialize its
4324 down_write(&obj->oo_ext_idx_sem);
4326 * recheck under lock.
4328 if (!osd_has_index(obj))
4329 result = osd_iam_container_init(env, obj,
4333 up_write(&obj->oo_ext_idx_sem);
4341 if (result == 0 && skip_iam == 0) {
4342 if (!osd_iam_index_probe(env, obj, feat))
4345 LINVRNT(osd_invariant(obj));
4347 if (result == 0 && feat == &dt_quota_glb_features &&
4348 fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
4349 result = osd_quota_migration(env, dt);
4354 static int osd_otable_it_attr_get(const struct lu_env *env,
4355 struct dt_object *dt,
4356 struct lu_attr *attr)
4362 static const struct dt_object_operations osd_obj_ops = {
4363 .do_read_lock = osd_object_read_lock,
4364 .do_write_lock = osd_object_write_lock,
4365 .do_read_unlock = osd_object_read_unlock,
4366 .do_write_unlock = osd_object_write_unlock,
4367 .do_write_locked = osd_object_write_locked,
4368 .do_attr_get = osd_attr_get,
4369 .do_declare_attr_set = osd_declare_attr_set,
4370 .do_attr_set = osd_attr_set,
4371 .do_ah_init = osd_ah_init,
4372 .do_declare_create = osd_declare_object_create,
4373 .do_create = osd_object_create,
4374 .do_declare_destroy = osd_declare_object_destroy,
4375 .do_destroy = osd_object_destroy,
4376 .do_index_try = osd_index_try,
4377 .do_declare_ref_add = osd_declare_object_ref_add,
4378 .do_ref_add = osd_object_ref_add,
4379 .do_declare_ref_del = osd_declare_object_ref_del,
4380 .do_ref_del = osd_object_ref_del,
4381 .do_xattr_get = osd_xattr_get,
4382 .do_declare_xattr_set = osd_declare_xattr_set,
4383 .do_xattr_set = osd_xattr_set,
4384 .do_declare_xattr_del = osd_declare_xattr_del,
4385 .do_xattr_del = osd_xattr_del,
4386 .do_xattr_list = osd_xattr_list,
4387 .do_object_sync = osd_object_sync,
4388 .do_invalidate = osd_invalidate,
4392 * dt_object_operations for interoperability mode
4393 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
4395 static const struct dt_object_operations osd_obj_ea_ops = {
4396 .do_read_lock = osd_object_read_lock,
4397 .do_write_lock = osd_object_write_lock,
4398 .do_read_unlock = osd_object_read_unlock,
4399 .do_write_unlock = osd_object_write_unlock,
4400 .do_write_locked = osd_object_write_locked,
4401 .do_attr_get = osd_attr_get,
4402 .do_declare_attr_set = osd_declare_attr_set,
4403 .do_attr_set = osd_attr_set,
4404 .do_ah_init = osd_ah_init,
4405 .do_declare_create = osd_declare_object_create,
4406 .do_create = osd_object_ea_create,
4407 .do_declare_destroy = osd_declare_object_destroy,
4408 .do_destroy = osd_object_destroy,
4409 .do_index_try = osd_index_try,
4410 .do_declare_ref_add = osd_declare_object_ref_add,
4411 .do_ref_add = osd_object_ref_add,
4412 .do_declare_ref_del = osd_declare_object_ref_del,
4413 .do_ref_del = osd_object_ref_del,
4414 .do_xattr_get = osd_xattr_get,
4415 .do_declare_xattr_set = osd_declare_xattr_set,
4416 .do_xattr_set = osd_xattr_set,
4417 .do_declare_xattr_del = osd_declare_xattr_del,
4418 .do_xattr_del = osd_xattr_del,
4419 .do_xattr_list = osd_xattr_list,
4420 .do_object_sync = osd_object_sync,
4421 .do_invalidate = osd_invalidate,
4424 static const struct dt_object_operations osd_obj_otable_it_ops = {
4425 .do_attr_get = osd_otable_it_attr_get,
4426 .do_index_try = osd_index_try,
4429 static int osd_index_declare_iam_delete(const struct lu_env *env,
4430 struct dt_object *dt,
4431 const struct dt_key *key,
4432 struct thandle *handle)
4434 struct osd_thandle *oh;
4436 oh = container_of0(handle, struct osd_thandle, ot_super);
4437 LASSERT(oh->ot_handle == NULL);
4439 /* Recycle may cause additional three blocks to be changed. */
4440 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4441 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4447 * delete a (key, value) pair from index \a dt specified by \a key
4449 * \param dt osd index object
4450 * \param key key for index
4451 * \param rec record reference
4452 * \param handle transaction handler
4455 * \retval -ve failure
4457 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4458 const struct dt_key *key,
4459 struct thandle *handle)
4461 struct osd_thread_info *oti = osd_oti_get(env);
4462 struct osd_object *obj = osd_dt_obj(dt);
4463 struct osd_thandle *oh;
4464 struct iam_path_descr *ipd;
4465 struct iam_container *bag = &obj->oo_dir->od_container;
4469 if (!dt_object_exists(dt))
4472 LINVRNT(osd_invariant(obj));
4473 LASSERT(!dt_object_remote(dt));
4474 LASSERT(bag->ic_object == obj->oo_inode);
4475 LASSERT(handle != NULL);
4477 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4479 ipd = osd_idx_ipd_get(env, bag);
4480 if (unlikely(ipd == NULL))
4483 oh = container_of0(handle, struct osd_thandle, ot_super);
4484 LASSERT(oh->ot_handle != NULL);
4485 LASSERT(oh->ot_handle->h_transaction != NULL);
4487 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4488 /* swab quota uid/gid provided by caller */
4489 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4490 key = (const struct dt_key *)&oti->oti_quota_id;
4493 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4494 osd_ipd_put(env, bag, ipd);
4495 LINVRNT(osd_invariant(obj));
4496 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4500 static int osd_index_declare_ea_delete(const struct lu_env *env,
4501 struct dt_object *dt,
4502 const struct dt_key *key,
4503 struct thandle *handle)
4505 struct osd_thandle *oh;
4506 struct inode *inode;
4510 LASSERT(!dt_object_remote(dt));
4511 LASSERT(handle != NULL);
4513 oh = container_of0(handle, struct osd_thandle, ot_super);
4514 LASSERT(oh->ot_handle == NULL);
4516 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
4517 if (key != NULL && unlikely(strcmp((char *)key, dotdot) == 0)) {
4518 /* '..' to a remote object has a local representative */
4519 credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
4520 /* to reset LMAI_REMOTE_PARENT */
4523 osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
4525 inode = osd_dt_obj(dt)->oo_inode;
4529 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
4530 0, oh, osd_dt_obj(dt), true, NULL, false);
4534 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
4537 struct osd_fid_pack *rec;
4540 if (de->file_type & LDISKFS_DIRENT_LUFID) {
4541 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
4542 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
4543 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
4549 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
4550 const struct lu_fid *fid)
4552 struct seq_server_site *ss = osd_seq_site(osd);
4555 /* FID seqs not in FLDB, must be local seq */
4556 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
4559 /* If FLD is not being initialized yet, it only happens during the
4560 * initialization, likely during mgs initialization, and we assume
4561 * this is local FID. */
4562 if (ss == NULL || ss->ss_server_fld == NULL)
4565 /* Only check the local FLDB here */
4566 if (osd_seq_exists(env, osd, fid_seq(fid)))
4573 * Index delete function for interoperability mode (b11826).
4574 * It will remove the directory entry added by osd_index_ea_insert().
4575 * This entry is needed to maintain name->fid mapping.
4577 * \param key, key i.e. file entry to be deleted
4579 * \retval 0, on success
4580 * \retval -ve, on error
4582 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
4583 const struct dt_key *key, struct thandle *handle)
4585 struct osd_object *obj = osd_dt_obj(dt);
4586 struct inode *dir = obj->oo_inode;
4587 struct dentry *dentry;
4588 struct osd_thandle *oh;
4589 struct ldiskfs_dir_entry_2 *de = NULL;
4590 struct buffer_head *bh;
4591 struct htree_lock *hlock = NULL;
4592 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
4593 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4597 if (!dt_object_exists(dt))
4600 LINVRNT(osd_invariant(obj));
4601 LASSERT(!dt_object_remote(dt));
4602 LASSERT(handle != NULL);
4604 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4606 oh = container_of(handle, struct osd_thandle, ot_super);
4607 LASSERT(oh->ot_handle != NULL);
4608 LASSERT(oh->ot_handle->h_transaction != NULL);
4610 ll_vfs_dq_init(dir);
4611 dentry = osd_child_dentry_get(env, obj,
4612 (char *)key, strlen((char *)key));
4614 if (obj->oo_hl_head != NULL) {
4615 hlock = osd_oti_get(env)->oti_hlock;
4616 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4617 dir, LDISKFS_HLOCK_DEL);
4619 down_write(&obj->oo_ext_idx_sem);
4622 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4624 /* If this is not the ".." entry, it might be a remote DNE
4625 * entry and we need to check if the FID is for a remote
4626 * MDT. If the FID is not in the directory entry (e.g.
4627 * upgraded 1.8 filesystem without dirdata enabled) then
4628 * we need to get the FID from the LMA. For a remote directory
4629 * there HAS to be an LMA, it cannot be an IGIF inode in this
4632 * Delete the entry before the agent inode in order to
4633 * simplify error handling. At worst an error after deleting
4634 * the entry first might leak the agent inode afterward. The
4635 * reverse would need filesystem abort in case of error deleting
4636 * the entry after the agent had been removed, or leave a
4637 * dangling entry pointing at a random inode. */
4638 if (strcmp((char *)key, dotdot) != 0) {
4639 LASSERT(de != NULL);
4640 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
4641 if (rc == -ENODATA) {
4642 /* can't get FID, postpone to the end of the
4643 * transaction when iget() is safe */
4644 osd_schedule_agent_inode_removal(env, oh,
4645 le32_to_cpu(de->inode));
4646 } else if (rc == 0 &&
4647 unlikely(osd_remote_fid(env, osd, fid))) {
4648 osd_schedule_agent_inode_removal(env, oh,
4649 le32_to_cpu(de->inode));
4652 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
4658 ldiskfs_htree_unlock(hlock);
4660 up_write(&obj->oo_ext_idx_sem);
4665 /* For inode on the remote MDT, .. will point to
4666 * /Agent directory, Check whether it needs to delete
4667 * from agent directory */
4668 if (unlikely(strcmp((char *)key, dotdot) == 0)) {
4671 ret = osd_delete_from_remote_parent(env, osd_obj2dev(obj),
4674 /* Sigh, the entry has been deleted, and
4675 * it is not easy to revert it back, so
4676 * let's keep this error private, and let
4677 * LFSCK fix it. XXX */
4678 CERROR("%s: delete remote parent "DFID": rc = %d\n",
4679 osd_name(osd), PFID(fid), ret);
4682 LASSERT(osd_invariant(obj));
4683 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4688 * Lookup index for \a key and copy record to \a rec.
4690 * \param dt osd index object
4691 * \param key key for index
4692 * \param rec record reference
4694 * \retval +ve success : exact mach
4695 * \retval 0 return record with key not greater than \a key
4696 * \retval -ve failure
4698 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
4699 struct dt_rec *rec, const struct dt_key *key)
4701 struct osd_object *obj = osd_dt_obj(dt);
4702 struct iam_path_descr *ipd;
4703 struct iam_container *bag = &obj->oo_dir->od_container;
4704 struct osd_thread_info *oti = osd_oti_get(env);
4705 struct iam_iterator *it = &oti->oti_idx_it;
4706 struct iam_rec *iam_rec;
4710 if (!dt_object_exists(dt))
4713 LASSERT(osd_invariant(obj));
4714 LASSERT(!dt_object_remote(dt));
4715 LASSERT(bag->ic_object == obj->oo_inode);
4717 ipd = osd_idx_ipd_get(env, bag);
4721 /* got ipd now we can start iterator. */
4722 iam_it_init(it, bag, 0, ipd);
4724 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4725 /* swab quota uid/gid provided by caller */
4726 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4727 key = (const struct dt_key *)&oti->oti_quota_id;
4730 rc = iam_it_get(it, (struct iam_key *)key);
4732 if (S_ISDIR(obj->oo_inode->i_mode))
4733 iam_rec = (struct iam_rec *)oti->oti_ldp;
4735 iam_rec = (struct iam_rec *) rec;
4737 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
4739 if (S_ISDIR(obj->oo_inode->i_mode))
4740 osd_fid_unpack((struct lu_fid *) rec,
4741 (struct osd_fid_pack *)iam_rec);
4742 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
4743 osd_quota_unpack(obj, rec);
4748 osd_ipd_put(env, bag, ipd);
4750 LINVRNT(osd_invariant(obj));
4755 static int osd_index_declare_iam_insert(const struct lu_env *env,
4756 struct dt_object *dt,
4757 const struct dt_rec *rec,
4758 const struct dt_key *key,
4759 struct thandle *handle)
4761 struct osd_thandle *oh;
4763 LASSERT(handle != NULL);
4765 oh = container_of0(handle, struct osd_thandle, ot_super);
4766 LASSERT(oh->ot_handle == NULL);
4768 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4769 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4775 * Inserts (key, value) pair in \a dt index object.
4777 * \param dt osd index object
4778 * \param key key for index
4779 * \param rec record reference
4780 * \param th transaction handler
4783 * \retval -ve failure
4785 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
4786 const struct dt_rec *rec,
4787 const struct dt_key *key, struct thandle *th,
4790 struct osd_object *obj = osd_dt_obj(dt);
4791 struct iam_path_descr *ipd;
4792 struct osd_thandle *oh;
4793 struct iam_container *bag;
4794 struct osd_thread_info *oti = osd_oti_get(env);
4795 struct iam_rec *iam_rec;
4799 if (!dt_object_exists(dt))
4802 LINVRNT(osd_invariant(obj));
4803 LASSERT(!dt_object_remote(dt));
4805 bag = &obj->oo_dir->od_container;
4806 LASSERT(bag->ic_object == obj->oo_inode);
4807 LASSERT(th != NULL);
4809 osd_trans_exec_op(env, th, OSD_OT_INSERT);
4811 ipd = osd_idx_ipd_get(env, bag);
4812 if (unlikely(ipd == NULL))
4815 oh = container_of0(th, struct osd_thandle, ot_super);
4816 LASSERT(oh->ot_handle != NULL);
4817 LASSERT(oh->ot_handle->h_transaction != NULL);
4818 if (S_ISDIR(obj->oo_inode->i_mode)) {
4819 iam_rec = (struct iam_rec *)oti->oti_ldp;
4820 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
4821 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4822 /* pack quota uid/gid */
4823 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4824 key = (const struct dt_key *)&oti->oti_quota_id;
4825 /* pack quota record */
4826 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
4827 iam_rec = (struct iam_rec *)rec;
4829 iam_rec = (struct iam_rec *)rec;
4832 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
4834 osd_ipd_put(env, bag, ipd);
4835 LINVRNT(osd_invariant(obj));
4836 osd_trans_exec_check(env, th, OSD_OT_INSERT);
4841 * Calls ldiskfs_add_entry() to add directory entry
4842 * into the directory. This is required for
4843 * interoperability mode (b11826)
4845 * \retval 0, on success
4846 * \retval -ve, on error
4848 static int __osd_ea_add_rec(struct osd_thread_info *info,
4849 struct osd_object *pobj, struct inode *cinode,
4850 const char *name, const struct lu_fid *fid,
4851 struct htree_lock *hlock, struct thandle *th)
4853 struct ldiskfs_dentry_param *ldp;
4854 struct dentry *child;
4855 struct osd_thandle *oth;
4858 oth = container_of(th, struct osd_thandle, ot_super);
4859 LASSERT(oth->ot_handle != NULL);
4860 LASSERT(oth->ot_handle->h_transaction != NULL);
4861 LASSERT(pobj->oo_inode);
4863 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
4864 if (unlikely(pobj->oo_inode ==
4865 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
4868 osd_get_ldiskfs_dirent_param(ldp, fid);
4869 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
4870 child->d_fsdata = (void *)ldp;
4871 ll_vfs_dq_init(pobj->oo_inode);
4872 rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
4873 child, cinode, hlock);
4874 if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
4875 struct ldiskfs_dir_entry_2 *de;
4876 struct buffer_head *bh;
4879 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
4882 rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
4885 if (S_ISDIR(cinode->i_mode))
4886 de->file_type = LDISKFS_DIRENT_LUFID |
4887 LDISKFS_FT_REG_FILE;
4889 de->file_type = LDISKFS_DIRENT_LUFID |
4891 ldiskfs_handle_dirty_metadata(oth->ot_handle,
4902 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
4903 * into the directory.Also sets flags into osd object to
4904 * indicate dot and dotdot are created. This is required for
4905 * interoperability mode (b11826)
4907 * \param dir directory for dot and dotdot fixup.
4908 * \param obj child object for linking
4910 * \retval 0, on success
4911 * \retval -ve, on error
4913 static int osd_add_dot_dotdot(struct osd_thread_info *info,
4914 struct osd_object *dir,
4915 struct inode *parent_dir, const char *name,
4916 const struct lu_fid *dot_fid,
4917 const struct lu_fid *dot_dot_fid,
4920 struct inode *inode = dir->oo_inode;
4921 struct osd_thandle *oth;
4924 oth = container_of(th, struct osd_thandle, ot_super);
4925 LASSERT(oth->ot_handle->h_transaction != NULL);
4926 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4928 if (strcmp(name, dot) == 0) {
4929 if (dir->oo_compat_dot_created) {
4932 LASSERT(inode->i_ino == parent_dir->i_ino);
4933 dir->oo_compat_dot_created = 1;
4936 } else if (strcmp(name, dotdot) == 0) {
4937 if (!dir->oo_compat_dot_created)
4939 /* in case of rename, dotdot is already created */
4940 if (dir->oo_compat_dotdot_created) {
4941 return __osd_ea_add_rec(info, dir, parent_dir, name,
4942 dot_dot_fid, NULL, th);
4945 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
4946 struct lu_fid tfid = *dot_dot_fid;
4949 result = osd_add_dot_dotdot_internal(info,
4950 dir->oo_inode, parent_dir, dot_fid,
4953 result = osd_add_dot_dotdot_internal(info,
4954 dir->oo_inode, parent_dir, dot_fid,
4959 dir->oo_compat_dotdot_created = 1;
4967 * It will call the appropriate osd_add* function and return the
4968 * value, return by respective functions.
4970 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4971 struct inode *cinode, const char *name,
4972 const struct lu_fid *fid, struct thandle *th)
4974 struct osd_thread_info *info = osd_oti_get(env);
4975 struct htree_lock *hlock;
4978 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4980 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4982 if (hlock != NULL) {
4983 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4986 down_write(&pobj->oo_ext_idx_sem);
4989 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4990 lu_object_fid(&pobj->oo_dt.do_lu),
4993 if (hlock != NULL) {
4994 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4995 pobj->oo_inode, LDISKFS_HLOCK_ADD);
4997 down_write(&pobj->oo_ext_idx_sem);
5000 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5001 struct lu_fid *tfid = &info->oti_fid;
5005 rc = __osd_ea_add_rec(info, pobj, cinode, name,
5008 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5013 ldiskfs_htree_unlock(hlock);
5015 up_write(&pobj->oo_ext_idx_sem);
5021 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5022 struct osd_idmap_cache *oic)
5024 struct osd_scrub *scrub = &dev->od_scrub;
5025 struct lu_fid *fid = &oic->oic_fid;
5026 struct osd_inode_id *id = &oic->oic_lid;
5027 struct inode *inode = NULL;
5033 if (!fid_is_norm(fid) && !fid_is_igif(fid))
5036 if (scrub->os_pos_current > id->oii_ino)
5040 rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5041 if (rc == -ENOENT) {
5042 __u32 gen = id->oii_gen;
5048 inode = osd_iget(oti, dev, id);
5049 /* The inode has been removed (by race maybe). */
5050 if (IS_ERR(inode)) {
5051 rc = PTR_ERR(inode);
5053 RETURN(rc == -ESTALE ? -ENOENT : rc);
5056 /* The OI mapping is lost. */
5057 if (gen != OSD_OII_NOGEN)
5061 /* The inode may has been reused by others, we do not know,
5062 * leave it to be handled by subsequent osd_fid_lookup(). */
5064 } else if (rc != 0 || osd_id_eq(id, &oti->oti_id)) {
5071 if (thread_is_running(&scrub->os_thread)) {
5072 if (inode == NULL) {
5073 inode = osd_iget(oti, dev, id);
5074 /* The inode has been removed (by race maybe). */
5075 if (IS_ERR(inode)) {
5076 rc = PTR_ERR(inode);
5078 RETURN(rc == -ESTALE ? -ENOENT : rc);
5082 rc = osd_oii_insert(dev, oic, insert);
5083 /* There is race condition between osd_oi_lookup and OI scrub.
5084 * The OI scrub finished just after osd_oi_lookup() failure.
5085 * Under such case, it is unnecessary to trigger OI scrub again,
5086 * but try to call osd_oi_lookup() again. */
5087 if (unlikely(rc == -EAGAIN))
5090 if (!S_ISDIR(inode->i_mode))
5093 rc = osd_check_lmv(oti, dev, inode, oic);
5099 if (!dev->od_noscrub && ++once == 1) {
5100 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
5102 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5103 "%.16s: trigger partial OI scrub for RPC inconsistency "
5104 "checking FID "DFID": rc = %d\n",
5105 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
5107 if (rc == 0 || rc == -EALREADY)
5117 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5118 struct osd_device *dev,
5119 struct osd_idmap_cache *oic,
5120 struct lu_fid *fid, __u32 ino)
5122 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5123 struct inode *inode;
5126 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5127 inode = osd_iget(oti, dev, &oic->oic_lid);
5128 if (IS_ERR(inode)) {
5129 fid_zero(&oic->oic_fid);
5130 return PTR_ERR(inode);
5133 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5136 fid_zero(&oic->oic_fid);
5138 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5142 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5143 struct osd_inode_id *id, const struct lu_fid *fid)
5145 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5146 id->oii_ino, id->oii_gen, info);
5147 info->oti_cache.oic_lid = *id;
5148 info->oti_cache.oic_fid = *fid;
5149 info->oti_cache.oic_dev = osd;
5153 * Get parent FID from the linkEA.
5155 * For a directory which parent resides on remote MDT, to satisfy the
5156 * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5157 * the other hand, to make the lookup(..) on the directory can return
5158 * the real parent FID, we append the real parent FID after its ".."
5159 * name entry in the /REMOTE_PARENT_DIR.
5161 * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5162 * backup. So after the restore, we cannot get the right parent FID from
5163 * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5164 * we have stored the real parent FID in the directory object's linkEA,
5165 * we can parse the linkEA for the real parent FID.
5167 * \param[in] env pointer to the thread context
5168 * \param[in] obj pointer to the object to be handled
5169 * \param[out]fid pointer to the buffer to hold the parent FID
5171 * \retval 0 for getting the real parent FID successfully
5172 * \retval negative error number on failure
5174 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5175 struct osd_object *obj,
5178 struct osd_thread_info *oti = osd_oti_get(env);
5179 struct lu_buf *buf = &oti->oti_big_buf;
5180 struct dentry *dentry = &oti->oti_obj_dentry;
5181 struct inode *inode = obj->oo_inode;
5182 struct linkea_data ldata = { NULL };
5187 if (!S_ISDIR(inode->i_mode))
5191 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5192 buf->lb_buf, buf->lb_len);
5193 if (rc == -ERANGE) {
5194 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5197 lu_buf_realloc(buf, rc);
5198 if (buf->lb_buf == NULL)
5205 if (unlikely(rc == 0))
5211 if (unlikely(buf->lb_buf == NULL)) {
5212 lu_buf_realloc(buf, rc);
5213 if (buf->lb_buf == NULL)
5220 rc = linkea_init_with_rec(&ldata);
5222 linkea_first_entry(&ldata);
5223 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5229 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5230 struct inode *inode,
5231 const struct lu_fid *pfid,
5232 const char *name, const int namelen)
5234 struct osd_thread_info *oti = osd_oti_get(env);
5235 struct lu_buf *buf = &oti->oti_big_buf;
5236 struct dentry *dentry = &oti->oti_obj_dentry;
5237 struct linkea_data ldata = { NULL };
5238 struct lu_name cname = { .ln_name = name,
5239 .ln_namelen = namelen };
5244 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5245 buf->lb_buf, buf->lb_len);
5247 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5252 if (unlikely(rc == 0))
5255 if (buf->lb_len < rc) {
5256 lu_buf_realloc(buf, rc);
5257 if (buf->lb_buf == NULL)
5264 rc = linkea_init_with_rec(&ldata);
5266 rc = linkea_links_find(&ldata, &cname, pfid);
5272 * Calls ->lookup() to find dentry. From dentry get inode and
5273 * read inode's ea to get fid. This is required for interoperability
5276 * \retval 0, on success
5277 * \retval -ve, on error
5279 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5280 struct dt_rec *rec, const struct dt_key *key)
5282 struct inode *dir = obj->oo_inode;
5283 struct dentry *dentry;
5284 struct ldiskfs_dir_entry_2 *de;
5285 struct buffer_head *bh;
5286 struct lu_fid *fid = (struct lu_fid *) rec;
5287 struct htree_lock *hlock = NULL;
5292 LASSERT(dir->i_op != NULL);
5293 LASSERT(dir->i_op->lookup != NULL);
5295 dentry = osd_child_dentry_get(env, obj,
5296 (char *)key, strlen((char *)key));
5298 if (obj->oo_hl_head != NULL) {
5299 hlock = osd_oti_get(env)->oti_hlock;
5300 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5301 dir, LDISKFS_HLOCK_LOOKUP);
5303 down_read(&obj->oo_ext_idx_sem);
5306 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5308 struct osd_thread_info *oti = osd_oti_get(env);
5309 struct osd_inode_id *id = &oti->oti_id;
5310 struct osd_idmap_cache *oic = &oti->oti_cache;
5311 struct osd_device *dev = osd_obj2dev(obj);
5313 ino = le32_to_cpu(de->inode);
5314 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5316 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5320 rc = osd_get_fid_from_dentry(de, rec);
5322 /* done with de, release bh */
5325 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5326 const char *name = (const char *)key;
5328 /* If the parent is on remote MDT, and there
5329 * is no FID-in-dirent, then we have to get
5330 * the parent FID from the linkEA. */
5331 if (likely(strlen(name) == 2 &&
5332 name[0] == '.' && name[1] == '.'))
5333 rc = osd_get_pfid_from_linkea(env, obj,
5336 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5339 osd_id_gen(id, ino, OSD_OII_NOGEN);
5342 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5343 fid_zero(&oic->oic_fid);
5348 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5349 rc = osd_consistency_check(oti, dev, oic);
5351 fid_zero(&oic->oic_fid);
5360 ldiskfs_htree_unlock(hlock);
5362 up_read(&obj->oo_ext_idx_sem);
5366 static int osd_index_declare_ea_insert(const struct lu_env *env,
5367 struct dt_object *dt,
5368 const struct dt_rec *rec,
5369 const struct dt_key *key,
5370 struct thandle *handle)
5372 struct osd_thandle *oh;
5373 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5374 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5375 const struct lu_fid *fid = rec1->rec_fid;
5376 int credits, rc = 0;
5377 struct osd_idmap_cache *idc;
5380 LASSERT(!dt_object_remote(dt));
5381 LASSERT(handle != NULL);
5382 LASSERT(fid != NULL);
5383 LASSERT(rec1->rec_type != 0);
5385 oh = container_of0(handle, struct osd_thandle, ot_super);
5386 LASSERT(oh->ot_handle == NULL);
5388 credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5390 /* we can't call iget() while a transactions is running
5391 * (this can lead to a deadlock), but we need to know
5392 * inum and object type. so we find this information at
5393 * declaration and cache in per-thread info */
5394 idc = osd_idc_find_or_init(env, osd, fid);
5396 RETURN(PTR_ERR(idc));
5397 if (idc->oic_remote) {
5398 /* a reference to remote inode is represented by an
5399 * agent inode which we have to create */
5400 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5401 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5404 osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5406 if (osd_dt_obj(dt)->oo_inode != NULL) {
5407 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5409 /* We ignore block quota on meta pool (MDTs), so needn't
5410 * calculate how many blocks will be consumed by this index
5412 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5413 i_gid_read(inode), 0, oh,
5414 osd_dt_obj(dt), true, NULL, false);
5421 * Index add function for interoperability mode (b11826).
5422 * It will add the directory entry.This entry is needed to
5423 * maintain name->fid mapping.
5425 * \param key it is key i.e. file entry to be inserted
5426 * \param rec it is value of given key i.e. fid
5428 * \retval 0, on success
5429 * \retval -ve, on error
5431 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5432 const struct dt_rec *rec,
5433 const struct dt_key *key, struct thandle *th,
5436 struct osd_object *obj = osd_dt_obj(dt);
5437 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5438 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5439 const struct lu_fid *fid = rec1->rec_fid;
5440 const char *name = (const char *)key;
5441 struct osd_thread_info *oti = osd_oti_get(env);
5442 struct inode *child_inode = NULL;
5443 struct osd_idmap_cache *idc;
5447 if (!dt_object_exists(dt))
5450 LASSERT(osd_invariant(obj));
5451 LASSERT(!dt_object_remote(dt));
5452 LASSERT(th != NULL);
5454 osd_trans_exec_op(env, th, OSD_OT_INSERT);
5456 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
5458 idc = osd_idc_find(env, osd, fid);
5459 if (unlikely(idc == NULL)) {
5460 /* this dt_insert() wasn't declared properly, so
5461 * FID is missing in OI cache. we better do not
5462 * lookup FID in FLDB/OI and don't risk to deadlock,
5463 * but in some special cases (lfsck testing, etc)
5464 * it's much simpler than fixing a caller */
5465 CERROR("%s: "DFID" wasn't declared for insert\n",
5466 osd_name(osd), PFID(fid));
5468 idc = osd_idc_find_or_init(env, osd, fid);
5470 RETURN(PTR_ERR(idc));
5473 if (idc->oic_remote) {
5474 /* Insert remote entry */
5475 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
5476 struct osd_mdobj_map *omm = osd->od_mdt_map;
5477 struct osd_thandle *oh;
5479 /* If parent on remote MDT, we need put this object
5481 oh = container_of(th, typeof(*oh), ot_super);
5482 rc = osd_add_to_remote_parent(env, osd, obj, oh);
5484 CERROR("%s: add "DFID" error: rc = %d\n",
5486 PFID(lu_object_fid(&dt->do_lu)), rc);
5490 child_inode = igrab(omm->omm_remote_parent->d_inode);
5492 child_inode = osd_create_local_agent_inode(env, osd,
5493 obj, fid, rec1->rec_type & S_IFMT, th);
5494 if (IS_ERR(child_inode))
5495 RETURN(PTR_ERR(child_inode));
5498 /* Insert local entry */
5499 if (unlikely(idc->oic_lid.oii_ino == 0)) {
5500 /* for a reason OI cache wasn't filled properly */
5501 CERROR("%s: OIC for "DFID" isn't filled\n",
5502 osd_name(osd), PFID(fid));
5505 child_inode = oti->oti_inode;
5506 if (unlikely(child_inode == NULL)) {
5507 struct ldiskfs_inode_info *lii;
5511 child_inode = oti->oti_inode = &lii->vfs_inode;
5513 child_inode->i_sb = osd_sb(osd);
5514 child_inode->i_ino = idc->oic_lid.oii_ino;
5515 child_inode->i_mode = rec1->rec_type & S_IFMT;
5518 rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
5520 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
5521 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
5523 if (child_inode && child_inode != oti->oti_inode)
5525 LASSERT(osd_invariant(obj));
5526 osd_trans_exec_check(env, th, OSD_OT_INSERT);
5531 * Initialize osd Iterator for given osd index object.
5533 * \param dt osd index object
5536 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
5537 struct dt_object *dt,
5540 struct osd_it_iam *it;
5541 struct osd_object *obj = osd_dt_obj(dt);
5542 struct lu_object *lo = &dt->do_lu;
5543 struct iam_path_descr *ipd;
5544 struct iam_container *bag = &obj->oo_dir->od_container;
5546 if (!dt_object_exists(dt))
5547 return ERR_PTR(-ENOENT);
5551 return ERR_PTR(-ENOMEM);
5553 ipd = osd_it_ipd_get(env, bag);
5554 if (likely(ipd != NULL)) {
5558 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
5559 return (struct dt_it *)it;
5562 return ERR_PTR(-ENOMEM);
5567 * free given Iterator.
5569 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
5571 struct osd_it_iam *it = (struct osd_it_iam *)di;
5572 struct osd_object *obj = it->oi_obj;
5574 iam_it_fini(&it->oi_it);
5575 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
5576 osd_object_put(env, obj);
5581 * Move Iterator to record specified by \a key
5583 * \param di osd iterator
5584 * \param key key for index
5586 * \retval +ve di points to record with least key not larger than key
5587 * \retval 0 di points to exact matched key
5588 * \retval -ve failure
5591 static int osd_it_iam_get(const struct lu_env *env,
5592 struct dt_it *di, const struct dt_key *key)
5594 struct osd_thread_info *oti = osd_oti_get(env);
5595 struct osd_it_iam *it = (struct osd_it_iam *)di;
5597 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5598 /* swab quota uid/gid */
5599 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5600 key = (struct dt_key *)&oti->oti_quota_id;
5603 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
5609 * \param di osd iterator
5611 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
5613 struct osd_it_iam *it = (struct osd_it_iam *)di;
5615 iam_it_put(&it->oi_it);
5619 * Move iterator by one record
5621 * \param di osd iterator
5623 * \retval +1 end of container reached
5625 * \retval -ve failure
5628 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
5630 struct osd_it_iam *it = (struct osd_it_iam *)di;
5632 return iam_it_next(&it->oi_it);
5636 * Return pointer to the key under iterator.
5639 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
5640 const struct dt_it *di)
5642 struct osd_thread_info *oti = osd_oti_get(env);
5643 struct osd_it_iam *it = (struct osd_it_iam *)di;
5644 struct osd_object *obj = it->oi_obj;
5647 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
5649 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
5650 /* swab quota uid/gid */
5651 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
5652 key = (struct dt_key *)&oti->oti_quota_id;
5659 * Return size of key under iterator (in bytes)
5662 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
5664 struct osd_it_iam *it = (struct osd_it_iam *)di;
5666 return iam_it_key_size(&it->oi_it);
5670 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
5672 /* check if file type is required */
5673 if (ent->lde_attrs & LUDA_TYPE) {
5674 struct luda_type *lt;
5675 int align = sizeof(*lt) - 1;
5677 len = (len + align) & ~align;
5678 lt = (struct luda_type *)(ent->lde_name + len);
5679 lt->lt_type = cpu_to_le16(DTTOIF(type));
5682 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
5686 * build lu direct from backend fs dirent.
5690 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
5691 char *name, __u16 namelen, __u16 type, __u32 attr)
5693 ent->lde_attrs = attr | LUDA_FID;
5694 fid_cpu_to_le(&ent->lde_fid, fid);
5696 ent->lde_hash = cpu_to_le64(offset);
5697 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
5699 strncpy(ent->lde_name, name, namelen);
5700 ent->lde_name[namelen] = '\0';
5701 ent->lde_namelen = cpu_to_le16(namelen);
5703 /* append lustre attributes */
5704 osd_it_append_attrs(ent, namelen, type);
5708 * Return pointer to the record under iterator.
5710 static int osd_it_iam_rec(const struct lu_env *env,
5711 const struct dt_it *di,
5712 struct dt_rec *dtrec, __u32 attr)
5714 struct osd_it_iam *it = (struct osd_it_iam *)di;
5715 struct osd_thread_info *info = osd_oti_get(env);
5718 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
5719 const struct osd_fid_pack *rec;
5720 struct lu_fid *fid = &info->oti_fid;
5721 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
5727 name = (char *)iam_it_key_get(&it->oi_it);
5729 RETURN(PTR_ERR(name));
5731 namelen = iam_it_key_size(&it->oi_it);
5733 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
5735 RETURN(PTR_ERR(rec));
5737 rc = osd_fid_unpack(fid, rec);
5741 hash = iam_it_store(&it->oi_it);
5743 /* IAM does not store object type in IAM index (dir) */
5744 osd_it_pack_dirent(lde, fid, hash, name, namelen,
5746 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5747 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5748 (struct iam_rec *)dtrec);
5749 osd_quota_unpack(it->oi_obj, dtrec);
5751 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5752 (struct iam_rec *)dtrec);
5759 * Returns cookie for current Iterator position.
5761 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
5763 struct osd_it_iam *it = (struct osd_it_iam *)di;
5765 return iam_it_store(&it->oi_it);
5769 * Restore iterator from cookie.
5771 * \param di osd iterator
5772 * \param hash Iterator location cookie
5774 * \retval +ve di points to record with least key not larger than key.
5775 * \retval 0 di points to exact matched key
5776 * \retval -ve failure
5779 static int osd_it_iam_load(const struct lu_env *env,
5780 const struct dt_it *di, __u64 hash)
5782 struct osd_it_iam *it = (struct osd_it_iam *)di;
5784 return iam_it_load(&it->oi_it, hash);
5787 static const struct dt_index_operations osd_index_iam_ops = {
5788 .dio_lookup = osd_index_iam_lookup,
5789 .dio_declare_insert = osd_index_declare_iam_insert,
5790 .dio_insert = osd_index_iam_insert,
5791 .dio_declare_delete = osd_index_declare_iam_delete,
5792 .dio_delete = osd_index_iam_delete,
5794 .init = osd_it_iam_init,
5795 .fini = osd_it_iam_fini,
5796 .get = osd_it_iam_get,
5797 .put = osd_it_iam_put,
5798 .next = osd_it_iam_next,
5799 .key = osd_it_iam_key,
5800 .key_size = osd_it_iam_key_size,
5801 .rec = osd_it_iam_rec,
5802 .store = osd_it_iam_store,
5803 .load = osd_it_iam_load
5809 * Creates or initializes iterator context.
5811 * \retval struct osd_it_ea, iterator structure on success
5814 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
5815 struct dt_object *dt,
5818 struct osd_object *obj = osd_dt_obj(dt);
5819 struct osd_thread_info *info = osd_oti_get(env);
5820 struct osd_it_ea *oie;
5822 struct lu_object *lo = &dt->do_lu;
5823 struct dentry *obj_dentry;
5826 if (!dt_object_exists(dt) || obj->oo_destroyed)
5827 RETURN(ERR_PTR(-ENOENT));
5829 OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
5831 RETURN(ERR_PTR(-ENOMEM));
5832 obj_dentry = &oie->oie_dentry;
5834 obj_dentry->d_inode = obj->oo_inode;
5835 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
5836 obj_dentry->d_name.hash = 0;
5838 oie->oie_rd_dirent = 0;
5839 oie->oie_it_dirent = 0;
5840 oie->oie_dirent = NULL;
5841 if (unlikely(!info->oti_it_ea_buf_used)) {
5842 oie->oie_buf = info->oti_it_ea_buf;
5843 info->oti_it_ea_buf_used = 1;
5845 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5846 if (oie->oie_buf == NULL)
5847 RETURN(ERR_PTR(-ENOMEM));
5851 file = &oie->oie_file;
5853 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
5854 if (attr & LUDA_64BITHASH)
5855 file->f_mode = FMODE_64BITHASH;
5857 file->f_mode = FMODE_32BITHASH;
5858 file->f_path.dentry = obj_dentry;
5859 file->f_mapping = obj->oo_inode->i_mapping;
5860 file->f_op = obj->oo_inode->i_fop;
5861 set_file_inode(file, obj->oo_inode);
5864 RETURN((struct dt_it *) oie);
5868 * Destroy or finishes iterator context.
5870 * \param di iterator structure to be destroyed
5872 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
5874 struct osd_thread_info *info = osd_oti_get(env);
5875 struct osd_it_ea *oie = (struct osd_it_ea *)di;
5876 struct osd_object *obj = oie->oie_obj;
5877 struct inode *inode = obj->oo_inode;
5880 oie->oie_file.f_op->release(inode, &oie->oie_file);
5881 osd_object_put(env, obj);
5882 if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
5883 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5885 info->oti_it_ea_buf_used = 0;
5886 OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
5891 * It position the iterator at given key, so that next lookup continues from
5892 * that key Or it is similar to dio_it->load() but based on a key,
5893 * rather than file position.
5895 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
5898 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
5900 static int osd_it_ea_get(const struct lu_env *env,
5901 struct dt_it *di, const struct dt_key *key)
5903 struct osd_it_ea *it = (struct osd_it_ea *)di;
5906 LASSERT(((const char *)key)[0] == '\0');
5907 it->oie_file.f_pos = 0;
5908 it->oie_rd_dirent = 0;
5909 it->oie_it_dirent = 0;
5910 it->oie_dirent = NULL;
5918 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
5922 struct osd_filldir_cbs {
5923 #ifdef HAVE_DIR_CONTEXT
5924 struct dir_context ctx;
5926 struct osd_it_ea *it;
5929 * It is called internally by ->readdir(). It fills the
5930 * iterator's in-memory data structure with required
5931 * information i.e. name, namelen, rec_size etc.
5933 * \param buf in which information to be filled in.
5934 * \param name name of the file in given dir
5936 * \retval 0 on success
5937 * \retval 1 on buffer full
5939 #ifdef HAVE_FILLDIR_USE_CTX
5940 static int osd_ldiskfs_filldir(struct dir_context *buf,
5942 static int osd_ldiskfs_filldir(void *buf,
5944 const char *name, int namelen,
5945 loff_t offset, __u64 ino, unsigned d_type)
5947 struct osd_it_ea *it =
5948 ((struct osd_filldir_cbs *)buf)->it;
5949 struct osd_object *obj = it->oie_obj;
5950 struct osd_it_ea_dirent *ent = it->oie_dirent;
5951 struct lu_fid *fid = &ent->oied_fid;
5952 struct osd_fid_pack *rec;
5955 /* this should never happen */
5956 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5957 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5961 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
5965 /* "." is just the object itself. */
5966 if (namelen == 1 && name[0] == '.') {
5967 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5968 } else if (d_type & LDISKFS_DIRENT_LUFID) {
5969 rec = (struct osd_fid_pack*) (name + namelen + 1);
5970 if (osd_fid_unpack(fid, rec) != 0)
5975 d_type &= ~LDISKFS_DIRENT_LUFID;
5977 /* NOT export local root. */
5978 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5979 ino = obj->oo_inode->i_ino;
5980 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5983 ent->oied_ino = ino;
5984 ent->oied_off = offset;
5985 ent->oied_namelen = namelen;
5986 ent->oied_type = d_type;
5988 memcpy(ent->oied_name, name, namelen);
5990 it->oie_rd_dirent++;
5991 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
5996 * Calls ->readdir() to load a directory entry at a time
5997 * and stored it in iterator's in-memory data structure.
5999 * \param di iterator's in memory structure
6001 * \retval 0 on success
6002 * \retval -ve on error
6003 * \retval +1 reach the end of entry
6005 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6006 const struct dt_it *di)
6008 struct osd_it_ea *it = (struct osd_it_ea *)di;
6009 struct osd_object *obj = it->oie_obj;
6010 struct inode *inode = obj->oo_inode;
6011 struct htree_lock *hlock = NULL;
6012 struct file *filp = &it->oie_file;
6014 struct osd_filldir_cbs buf = {
6015 #ifdef HAVE_DIR_CONTEXT
6016 .ctx.actor = osd_ldiskfs_filldir,
6022 it->oie_dirent = it->oie_buf;
6023 it->oie_rd_dirent = 0;
6025 if (obj->oo_hl_head != NULL) {
6026 hlock = osd_oti_get(env)->oti_hlock;
6027 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6028 inode, LDISKFS_HLOCK_READDIR);
6030 down_read(&obj->oo_ext_idx_sem);
6033 #ifdef HAVE_DIR_CONTEXT
6034 buf.ctx.pos = filp->f_pos;
6035 rc = inode->i_fop->iterate(filp, &buf.ctx);
6036 filp->f_pos = buf.ctx.pos;
6038 rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
6042 ldiskfs_htree_unlock(hlock);
6044 up_read(&obj->oo_ext_idx_sem);
6046 if (it->oie_rd_dirent == 0) {
6047 /*If it does not get any dirent, it means it has been reached
6048 *to the end of the dir */
6049 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6053 it->oie_dirent = it->oie_buf;
6054 it->oie_it_dirent = 1;
6061 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6062 * to load a directory entry at a time and stored it in
6063 * iterator's in-memory data structure.
6065 * \param di iterator's in memory structure
6067 * \retval +ve iterator reached to end
6068 * \retval 0 iterator not reached to end
6069 * \retval -ve on error
6071 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6073 struct osd_it_ea *it = (struct osd_it_ea *)di;
6078 if (it->oie_it_dirent < it->oie_rd_dirent) {
6080 (void *) it->oie_dirent +
6081 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6082 it->oie_dirent->oied_namelen);
6083 it->oie_it_dirent++;
6086 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6089 rc = osd_ldiskfs_it_fill(env, di);
6096 * Returns the key at current position from iterator's in memory structure.
6098 * \param di iterator's in memory structure
6100 * \retval key i.e. struct dt_key on success
6102 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6103 const struct dt_it *di)
6105 struct osd_it_ea *it = (struct osd_it_ea *)di;
6107 return (struct dt_key *)it->oie_dirent->oied_name;
6111 * Returns the key's size at current position from iterator's in memory structure.
6113 * \param di iterator's in memory structure
6115 * \retval key_size i.e. struct dt_key on success
6117 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6119 struct osd_it_ea *it = (struct osd_it_ea *)di;
6121 return it->oie_dirent->oied_namelen;
6124 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6126 if (LDISKFS_DIR_REC_LEN(de) >=
6127 __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6134 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6135 unsigned blocksize, bool dotdot)
6138 return osd_dotdot_has_space(de);
6140 if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6141 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6148 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6149 handle_t *jh, struct dentry *dentry,
6150 const struct lu_fid *fid, struct buffer_head *bh,
6151 struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6154 struct inode *dir = dentry->d_parent->d_inode;
6155 struct inode *inode = dentry->d_inode;
6156 struct osd_fid_pack *rec;
6157 struct ldiskfs_dentry_param *ldp;
6158 int namelen = dentry->d_name.len;
6160 struct osd_thread_info *info = osd_oti_get(env);
6163 if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
6164 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6167 /* There is enough space to hold the FID-in-dirent. */
6168 if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6169 rc = ldiskfs_journal_get_write_access(jh, bh);
6173 de->name[namelen] = 0;
6174 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6175 rec->fp_len = sizeof(struct lu_fid) + 1;
6176 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6177 de->file_type |= LDISKFS_DIRENT_LUFID;
6178 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6185 rc = ldiskfs_delete_entry(jh, dir, de, bh);
6189 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6190 osd_get_ldiskfs_dirent_param(ldp, fid);
6191 dentry->d_fsdata = (void *)ldp;
6192 ll_vfs_dq_init(dir);
6193 rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6194 /* It is too bad, we cannot reinsert the name entry back.
6195 * That means we lose it! */
6197 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
6198 "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6199 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
6200 dir->i_ino, dir->i_generation, namelen,
6201 dentry->d_name.name, PFID(fid), rc);
6207 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6208 struct osd_it_ea *it, struct lu_fid *fid,
6209 struct osd_inode_id *id, __u32 *attr)
6211 struct osd_thread_info *info = osd_oti_get(env);
6212 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6213 struct osd_device *dev = osd_obj2dev(obj);
6214 struct super_block *sb = osd_sb(dev);
6215 const char *devname = osd_name(dev);
6216 struct osd_it_ea_dirent *ent = it->oie_dirent;
6217 struct inode *dir = obj->oo_inode;
6218 struct htree_lock *hlock = NULL;
6219 struct buffer_head *bh = NULL;
6220 handle_t *jh = NULL;
6221 struct ldiskfs_dir_entry_2 *de;
6222 struct dentry *dentry;
6223 struct inode *inode;
6224 const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6227 bool dotdot = false;
6231 if (ent->oied_name[0] == '.') {
6232 if (ent->oied_namelen == 1)
6235 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6239 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6240 inode = osd_iget(info, dev, id);
6241 if (IS_ERR(inode)) {
6242 rc = PTR_ERR(inode);
6243 if (rc == -ENOENT || rc == -ESTALE) {
6244 /* Maybe dangling name entry, or
6245 * corrupted directory entry. */
6246 *attr |= LUDA_UNKNOWN;
6249 CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6250 "check_repair, dir = %lu/%u, name = %.*s, "
6251 "ino = %llu, rc = %d\n",
6252 devname, dir->i_ino, dir->i_generation,
6253 ent->oied_namelen, ent->oied_name,
6260 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6262 rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6263 if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6268 /* We need to ensure that the name entry is still valid.
6269 * Because it may be removed or renamed by other already.
6271 * The unlink or rename operation will start journal before PDO lock,
6272 * so to avoid deadlock, here we need to start journal handle before
6273 * related PDO lock also. But because we do not know whether there
6274 * will be something to be repaired before PDO lock, we just start
6275 * journal without conditions.
6277 * We may need to remove the name entry firstly, then insert back.
6278 * One credit is for user quota file update.
6279 * One credit is for group quota file update.
6280 * Two credits are for dirty inode. */
6281 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6282 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6284 if (dev->od_dirent_journal != 0) {
6287 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6290 CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6291 "check_repair, dir = %lu/%u, credits = %d, "
6292 "name = %.*s, ino = %llu: rc = %d\n",
6293 devname, dir->i_ino, dir->i_generation, credits,
6294 ent->oied_namelen, ent->oied_name,
6297 GOTO(out_inode, rc);
6300 if (obj->oo_hl_head != NULL) {
6301 hlock = osd_oti_get(env)->oti_hlock;
6302 /* "0" means exclusive lock for the whole directory.
6303 * We need to prevent others access such name entry
6304 * during the delete + insert. Neither HLOCK_ADD nor
6305 * HLOCK_DEL cannot guarantee the atomicity. */
6306 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6308 down_write(&obj->oo_ext_idx_sem);
6311 if (obj->oo_hl_head != NULL) {
6312 hlock = osd_oti_get(env)->oti_hlock;
6313 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6314 LDISKFS_HLOCK_LOOKUP);
6316 down_read(&obj->oo_ext_idx_sem);
6320 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6321 if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6322 *attr |= LUDA_IGNORE;
6327 /* For dotdot entry, if there is not enough space to hold the
6328 * FID-in-dirent, just keep them there. It only happens when the
6329 * device upgraded from 1.8 or restored from MDT file-level backup.
6330 * For the whole directory, only dotdot entry have no FID-in-dirent
6331 * and needs to get FID from LMA when readdir, it will not affect the
6332 * performance much. */
6333 if (dotdot && !osd_dotdot_has_space(de)) {
6334 *attr |= LUDA_UNKNOWN;
6340 if (lu_fid_eq(fid, &lma->lma_self_fid))
6343 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6344 struct lu_fid *tfid = &lma->lma_self_fid;
6346 if (likely(dotdot &&
6347 fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6348 fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6349 /* It must be REMOTE_PARENT_DIR and as the
6350 * 'dotdot' entry of remote directory */
6351 *attr |= LUDA_IGNORE;
6353 CDEBUG(D_LFSCK, "%s: expect remote agent "
6354 "parent directory, but got %.*s under "
6355 "dir = %lu/%u with the FID "DFID"\n",
6356 devname, ent->oied_namelen,
6357 ent->oied_name, dir->i_ino,
6358 dir->i_generation, PFID(tfid));
6360 *attr |= LUDA_UNKNOWN;
6367 if (!fid_is_zero(fid)) {
6368 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6370 if (rc == -ENOENT ||
6372 !(dev->od_scrub.os_file.sf_flags & SF_UPGRADE))) {
6373 /* linkEA does not recognize the dirent entry,
6374 * it may because the dirent entry corruption
6375 * and points to other's inode. */
6376 CDEBUG(D_LFSCK, "%s: the target inode does not "
6377 "recognize the dirent, dir = %lu/%u, "
6378 " name = %.*s, ino = %llu, "
6379 DFID": rc = %d\n", devname, dir->i_ino,
6380 dir->i_generation, ent->oied_namelen,
6381 ent->oied_name, ent->oied_ino, PFID(fid), rc);
6382 *attr |= LUDA_UNKNOWN;
6387 if (rc && rc != -ENODATA) {
6388 CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6389 "dir = %lu/%u, name = %.*s, ino = %llu, "
6390 DFID": rc = %d\n", devname, dir->i_ino,
6391 dir->i_generation, ent->oied_namelen,
6392 ent->oied_name, ent->oied_ino, PFID(fid), rc);
6393 *attr |= LUDA_UNKNOWN;
6400 /* linkEA recognizes the dirent entry, the FID-in-LMA is
6401 * valid, trusted, in spite of fid_is_sane(fid) or not. */
6402 if (*attr & LUDA_VERIFY_DRYRUN) {
6403 *fid = lma->lma_self_fid;
6404 *attr |= LUDA_REPAIR;
6411 dev->od_dirent_journal = 1;
6412 if (hlock != NULL) {
6413 ldiskfs_htree_unlock(hlock);
6416 up_read(&obj->oo_ext_idx_sem);
6422 *fid = lma->lma_self_fid;
6424 /* Update or append the FID-in-dirent. */
6425 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6426 bh, de, hlock, dotdot);
6428 *attr |= LUDA_REPAIR;
6430 CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
6431 "the dirent, dir = %lu/%u, name = %.*s, "
6432 "ino = %llu, "DFID": rc = %d\n",
6433 devname, dir->i_ino, dir->i_generation,
6434 ent->oied_namelen, ent->oied_name,
6435 ent->oied_ino, PFID(fid), rc);
6437 /* lma is NULL, trust the FID-in-dirent if it is valid. */
6438 if (*attr & LUDA_VERIFY_DRYRUN) {
6439 if (fid_is_sane(fid)) {
6440 *attr |= LUDA_REPAIR;
6441 } else if (dev->od_index == 0) {
6442 lu_igif_build(fid, inode->i_ino,
6443 inode->i_generation);
6444 *attr |= LUDA_UPGRADE;
6452 dev->od_dirent_journal = 1;
6453 if (hlock != NULL) {
6454 ldiskfs_htree_unlock(hlock);
6457 up_read(&obj->oo_ext_idx_sem);
6464 if (unlikely(fid_is_sane(fid))) {
6465 /* FID-in-dirent exists, but FID-in-LMA is lost.
6466 * Trust the FID-in-dirent, and add FID-in-LMA. */
6467 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
6469 *attr |= LUDA_REPAIR;
6471 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
6472 "update dirent, dir = %lu/%u, "
6473 "name = %.*s, ino = %llu, "
6475 devname, dir->i_ino, dir->i_generation,
6476 ent->oied_namelen, ent->oied_name,
6477 ent->oied_ino, PFID(fid), rc);
6478 } else if (dev->od_index == 0) {
6479 lu_igif_build(fid, inode->i_ino, inode->i_generation);
6480 /* It is probably IGIF object. Only aappend the
6481 * FID-in-dirent. OI scrub will process FID-in-LMA. */
6482 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6483 bh, de, hlock, dotdot);
6485 *attr |= LUDA_UPGRADE;
6487 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
6488 "after the dirent, dir = %lu/%u, "
6489 "name = %.*s, ino = %llu, "
6491 devname, dir->i_ino, dir->i_generation,
6492 ent->oied_namelen, ent->oied_name,
6493 ent->oied_ino, PFID(fid), rc);
6502 if (hlock != NULL) {
6503 ldiskfs_htree_unlock(hlock);
6505 if (dev->od_dirent_journal != 0)
6506 up_write(&obj->oo_ext_idx_sem);
6508 up_read(&obj->oo_ext_idx_sem);
6512 ldiskfs_journal_stop(jh);
6516 if (rc >= 0 && !dirty)
6517 dev->od_dirent_journal = 0;
6523 * Returns the value at current position from iterator's in memory structure.
6525 * \param di struct osd_it_ea, iterator's in memory structure
6526 * \param attr attr requested for dirent.
6527 * \param lde lustre dirent
6529 * \retval 0 no error and \param lde has correct lustre dirent.
6530 * \retval -ve on error
6532 static inline int osd_it_ea_rec(const struct lu_env *env,
6533 const struct dt_it *di,
6534 struct dt_rec *dtrec, __u32 attr)
6536 struct osd_it_ea *it = (struct osd_it_ea *)di;
6537 struct osd_object *obj = it->oie_obj;
6538 struct osd_device *dev = osd_obj2dev(obj);
6539 struct osd_thread_info *oti = osd_oti_get(env);
6540 struct osd_inode_id *id = &oti->oti_id;
6541 struct lu_fid *fid = &it->oie_dirent->oied_fid;
6542 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6543 __u32 ino = it->oie_dirent->oied_ino;
6547 LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
6549 if (attr & LUDA_VERIFY) {
6550 if (unlikely(ino == osd_remote_parent_ino(dev))) {
6551 attr |= LUDA_IGNORE;
6552 /* If the parent is on remote MDT, and there
6553 * is no FID-in-dirent, then we have to get
6554 * the parent FID from the linkEA. */
6555 if (!fid_is_sane(fid) &&
6556 it->oie_dirent->oied_namelen == 2 &&
6557 it->oie_dirent->oied_name[0] == '.' &&
6558 it->oie_dirent->oied_name[1] == '.')
6559 osd_get_pfid_from_linkea(env, obj, fid);
6561 rc = osd_dirent_check_repair(env, obj, it, fid, id,
6565 if (!fid_is_sane(fid))
6566 attr |= LUDA_UNKNOWN;
6568 attr &= ~LU_DIRENT_ATTRS_MASK;
6569 if (!fid_is_sane(fid)) {
6570 bool is_dotdot = false;
6571 if (it->oie_dirent->oied_namelen == 2 &&
6572 it->oie_dirent->oied_name[0] == '.' &&
6573 it->oie_dirent->oied_name[1] == '.')
6575 /* If the parent is on remote MDT, and there
6576 * is no FID-in-dirent, then we have to get
6577 * the parent FID from the linkEA. */
6578 if (ino == osd_remote_parent_ino(dev) && is_dotdot) {
6579 rc = osd_get_pfid_from_linkea(env, obj, fid);
6581 if (is_dotdot == false &&
6582 OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
6585 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6588 osd_id_gen(id, ino, OSD_OII_NOGEN);
6592 /* Pack the entry anyway, at least the offset is right. */
6593 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
6594 it->oie_dirent->oied_name,
6595 it->oie_dirent->oied_namelen,
6596 it->oie_dirent->oied_type, attr);
6601 if (osd_remote_fid(env, dev, fid))
6604 if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
6605 osd_add_oi_cache(oti, dev, id, fid);
6607 RETURN(rc > 0 ? 0 : rc);
6611 * Returns the record size size at current position.
6613 * This function will return record(lu_dirent) size in bytes.
6615 * \param[in] env execution environment
6616 * \param[in] di iterator's in memory structure
6617 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
6618 * calculate the lu_dirent size.
6620 * \retval record size(in bytes & in memory) of the current lu_dirent
6623 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
6626 struct osd_it_ea *it = (struct osd_it_ea *)di;
6628 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
6632 * Returns a cookie for current position of the iterator head, so that
6633 * user can use this cookie to load/start the iterator next time.
6635 * \param di iterator's in memory structure
6637 * \retval cookie for current position, on success
6639 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
6641 struct osd_it_ea *it = (struct osd_it_ea *)di;
6643 return it->oie_dirent->oied_off;
6647 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6648 * to load a directory entry at a time and stored it i inn,
6649 * in iterator's in-memory data structure.
6651 * \param di struct osd_it_ea, iterator's in memory structure
6653 * \retval +ve on success
6654 * \retval -ve on error
6656 static int osd_it_ea_load(const struct lu_env *env,
6657 const struct dt_it *di, __u64 hash)
6659 struct osd_it_ea *it = (struct osd_it_ea *)di;
6663 it->oie_file.f_pos = hash;
6665 rc = osd_ldiskfs_it_fill(env, di);
6676 * Index lookup function for interoperability mode (b11826).
6678 * \param key, key i.e. file name to be searched
6680 * \retval +ve, on success
6681 * \retval -ve, on error
6683 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
6684 struct dt_rec *rec, const struct dt_key *key)
6686 struct osd_object *obj = osd_dt_obj(dt);
6691 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
6692 LINVRNT(osd_invariant(obj));
6694 rc = osd_ea_lookup_rec(env, obj, rec, key);
6701 * Index and Iterator operations for interoperability
6702 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
6704 static const struct dt_index_operations osd_index_ea_ops = {
6705 .dio_lookup = osd_index_ea_lookup,
6706 .dio_declare_insert = osd_index_declare_ea_insert,
6707 .dio_insert = osd_index_ea_insert,
6708 .dio_declare_delete = osd_index_declare_ea_delete,
6709 .dio_delete = osd_index_ea_delete,
6711 .init = osd_it_ea_init,
6712 .fini = osd_it_ea_fini,
6713 .get = osd_it_ea_get,
6714 .put = osd_it_ea_put,
6715 .next = osd_it_ea_next,
6716 .key = osd_it_ea_key,
6717 .key_size = osd_it_ea_key_size,
6718 .rec = osd_it_ea_rec,
6719 .rec_size = osd_it_ea_rec_size,
6720 .store = osd_it_ea_store,
6721 .load = osd_it_ea_load
6725 static void *osd_key_init(const struct lu_context *ctx,
6726 struct lu_context_key *key)
6728 struct osd_thread_info *info;
6730 OBD_ALLOC_PTR(info);
6732 return ERR_PTR(-ENOMEM);
6734 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6735 if (info->oti_it_ea_buf == NULL)
6738 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
6740 info->oti_hlock = ldiskfs_htree_lock_alloc();
6741 if (info->oti_hlock == NULL)
6747 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6750 return ERR_PTR(-ENOMEM);
6753 static void osd_key_fini(const struct lu_context *ctx,
6754 struct lu_context_key *key, void* data)
6756 struct osd_thread_info *info = data;
6757 struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
6758 struct osd_idmap_cache *idc = info->oti_ins_cache;
6760 if (info->oti_inode != NULL)
6762 if (info->oti_hlock != NULL)
6763 ldiskfs_htree_lock_free(info->oti_hlock);
6764 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6765 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
6766 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
6767 lu_buf_free(&info->oti_big_buf);
6769 LASSERT(info->oti_ins_cache_size > 0);
6770 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
6771 info->oti_ins_cache = NULL;
6772 info->oti_ins_cache_size = 0;
6777 static void osd_key_exit(const struct lu_context *ctx,
6778 struct lu_context_key *key, void *data)
6780 struct osd_thread_info *info = data;
6782 LASSERT(info->oti_r_locks == 0);
6783 LASSERT(info->oti_w_locks == 0);
6784 LASSERT(info->oti_txns == 0);
6787 /* type constructor/destructor: osd_type_init, osd_type_fini */
6788 LU_TYPE_INIT_FINI(osd, &osd_key);
6790 struct lu_context_key osd_key = {
6791 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
6792 .lct_init = osd_key_init,
6793 .lct_fini = osd_key_fini,
6794 .lct_exit = osd_key_exit
6798 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
6799 const char *name, struct lu_device *next)
6801 struct osd_device *osd = osd_dev(d);
6803 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
6804 >= sizeof(osd->od_svname))
6806 return osd_procfs_init(osd, name);
6809 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
6811 struct seq_server_site *ss = osd_seq_site(osd);
6815 if (osd->od_is_ost || osd->od_cl_seq != NULL)
6818 if (unlikely(ss == NULL))
6821 OBD_ALLOC_PTR(osd->od_cl_seq);
6822 if (osd->od_cl_seq == NULL)
6825 rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
6826 osd->od_svname, ss->ss_server_seq);
6828 OBD_FREE_PTR(osd->od_cl_seq);
6829 osd->od_cl_seq = NULL;
6833 if (ss->ss_node_id == 0) {
6834 /* If the OSD on the sequence controller(MDT0), then allocate
6835 * sequence here, otherwise allocate sequence after connected
6836 * to MDT0 (see mdt_register_lwp_callback()). */
6837 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
6838 &osd->od_cl_seq->lcs_space, env);
6844 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
6846 if (osd->od_cl_seq == NULL)
6849 seq_client_fini(osd->od_cl_seq);
6850 OBD_FREE_PTR(osd->od_cl_seq);
6851 osd->od_cl_seq = NULL;
6854 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
6858 /* shutdown quota slave instance associated with the device */
6859 if (o->od_quota_slave != NULL) {
6860 qsd_fini(env, o->od_quota_slave);
6861 o->od_quota_slave = NULL;
6864 osd_fid_fini(env, o);
6869 static void osd_umount(const struct lu_env *env, struct osd_device *o)
6873 if (o->od_mnt != NULL) {
6874 shrink_dcache_sb(osd_sb(o));
6875 osd_sync(env, &o->od_dt_dev);
6884 static int osd_mount(const struct lu_env *env,
6885 struct osd_device *o, struct lustre_cfg *cfg)
6887 const char *name = lustre_cfg_string(cfg, 0);
6888 const char *dev = lustre_cfg_string(cfg, 1);
6890 unsigned long page, s_flags, lmd_flags = 0;
6891 struct page *__page;
6892 struct file_system_type *type;
6893 char *options = NULL;
6895 struct osd_thread_info *info = osd_oti_get(env);
6896 struct lu_fid *fid = &info->oti_fid;
6897 struct inode *inode;
6898 int rc = 0, force_over_512tb = 0;
6901 if (o->od_mnt != NULL)
6904 if (strlen(dev) >= sizeof(o->od_mntdev))
6906 strcpy(o->od_mntdev, dev);
6908 str = lustre_cfg_string(cfg, 2);
6909 s_flags = simple_strtoul(str, NULL, 0);
6910 str = strstr(str, ":");
6912 lmd_flags = simple_strtoul(str + 1, NULL, 0);
6913 opts = lustre_cfg_string(cfg, 3);
6915 if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
6916 CERROR("%s: device %s extents feature is not guaranteed to "
6917 "work on big-endian systems. Use \"bigendian_extents\" "
6918 "mount option to override.\n", name, dev);
6922 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
6923 if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
6924 CWARN("force_over_128tb option is deprecated. "
6925 "Filesystems less than 512TB can be created without any "
6926 "force options. Use force_over_512tb option for "
6927 "filesystems greater than 512TB.\n");
6930 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
6931 if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
6932 CWARN("force_over_256tb option is deprecated. "
6933 "Filesystems less than 512TB can be created without any "
6934 "force options. Use force_over_512tb option for "
6935 "filesystems greater than 512TB.\n");
6939 if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
6940 force_over_512tb = 1;
6942 __page = alloc_page(GFP_KERNEL);
6944 GOTO(out, rc = -ENOMEM);
6945 page = (unsigned long)page_address(__page);
6946 options = (char *)page;
6949 /* strip out the options for back compatiblity */
6950 static char *sout[] = {
6957 /* strip out option we processed in osd */
6958 "bigendian_extents",
6964 strcat(options, opts);
6965 for (rc = 0, str = options; sout[rc]; ) {
6966 char *op = strstr(str, sout[rc]);
6972 if (op == options || *(op - 1) == ',') {
6973 str = op + strlen(sout[rc]);
6974 if (*str == ',' || *str == '\0') {
6975 *str == ',' ? str++ : str;
6976 memmove(op, str, strlen(str) + 1);
6979 for (str = op; *str != ',' && *str != '\0'; str++)
6983 strncat(options, "user_xattr,acl", 14);
6986 /* Glom up mount options */
6987 if (*options != '\0')
6988 strcat(options, ",");
6989 strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
6991 type = get_fs_type("ldiskfs");
6993 CERROR("%s: cannot find ldiskfs module\n", name);
6994 GOTO(out, rc = -ENODEV);
6997 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
6998 module_put(type->owner);
7000 if (IS_ERR(o->od_mnt)) {
7001 rc = PTR_ERR(o->od_mnt);
7003 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7007 if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7008 osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7009 force_over_512tb == 0) {
7010 CERROR("%s: device %s LDISKFS does not support filesystems "
7011 "greater than 512TB and can cause data corruption. "
7012 "Use \"force_over_512tb\" mount option to override.\n",
7014 GOTO(out_mnt, rc = -EINVAL);
7017 if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7018 #ifdef HAVE_DEV_SET_RDONLY
7019 dev_set_rdonly(osd_sb(o)->s_bdev);
7020 o->od_dt_dev.dd_rdonly = 1;
7021 LCONSOLE_WARN("%s: set dev_rdonly on this device\n", name);
7023 LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7026 GOTO(out_mnt, rc = -EOPNOTSUPP);
7029 #ifdef HAVE_DEV_SET_RDONLY
7030 if (dev_check_rdonly(osd_sb(o)->s_bdev)) {
7031 CERROR("%s: underlying device %s is marked as "
7032 "read-only. Setup failed\n", name, dev);
7034 GOTO(out_mnt, rc = -EROFS);
7039 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
7040 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
7041 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7042 GOTO(out_mnt, rc = -EINVAL);
7045 #ifdef LDISKFS_MOUNT_DIRDATA
7046 if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
7047 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
7048 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7049 else if (!o->od_is_ost)
7050 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7051 "enabling the dirdata feature. If you do not want to "
7052 "downgrade to Lustre-1.x again, you can enable it via "
7053 "'tune2fs -O dirdata device'\n", name, dev);
7055 inode = osd_sb(o)->s_root->d_inode;
7056 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7057 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7059 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7063 if (lmd_flags & LMD_FLG_NOSCRUB)
7074 __free_page(__page);
7079 static struct lu_device *osd_device_fini(const struct lu_env *env,
7080 struct lu_device *d)
7082 struct osd_device *o = osd_dev(d);
7085 osd_shutdown(env, o);
7087 osd_scrub_cleanup(env, o);
7088 osd_obj_map_fini(o);
7094 static int osd_device_init0(const struct lu_env *env,
7095 struct osd_device *o,
7096 struct lustre_cfg *cfg)
7098 struct lu_device *l = osd2lu_dev(o);
7099 struct osd_thread_info *info;
7103 /* if the module was re-loaded, env can loose its keys */
7104 rc = lu_env_refill((struct lu_env *) env);
7107 info = osd_oti_get(env);
7110 l->ld_ops = &osd_lu_ops;
7111 o->od_dt_dev.dd_ops = &osd_dt_ops;
7113 spin_lock_init(&o->od_osfs_lock);
7114 mutex_init(&o->od_otable_mutex);
7115 INIT_LIST_HEAD(&o->od_orphan_list);
7117 o->od_read_cache = 1;
7118 o->od_writethrough_cache = 1;
7119 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7121 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7122 sizeof(o->od_svname));
7123 if (cplen >= sizeof(o->od_svname)) {
7128 o->od_index = -1; /* -1 means index is invalid */
7129 rc = server_name2index(o->od_svname, &o->od_index, NULL);
7130 if (rc == LDD_F_SV_TYPE_OST)
7133 o->od_full_scrub_ratio = OFSR_DEFAULT;
7134 o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7135 rc = osd_mount(env, o, cfg);
7139 rc = osd_obj_map_init(env, o);
7143 rc = lu_site_init(&o->od_site, l);
7145 GOTO(out_compat, rc);
7146 o->od_site.ls_bottom_dev = l;
7148 rc = lu_site_init_finish(&o->od_site);
7152 INIT_LIST_HEAD(&o->od_ios_list);
7153 /* setup scrub, including OI files initialization */
7154 rc = osd_scrub_setup(env, o);
7158 rc = osd_procfs_init(o, o->od_svname);
7160 CERROR("%s: can't initialize procfs: rc = %d\n",
7162 GOTO(out_scrub, rc);
7165 LASSERT(l->ld_site->ls_linkage.next != NULL);
7166 LASSERT(l->ld_site->ls_linkage.prev != NULL);
7168 /* initialize quota slave instance */
7169 o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
7171 if (IS_ERR(o->od_quota_slave)) {
7172 rc = PTR_ERR(o->od_quota_slave);
7173 o->od_quota_slave = NULL;
7174 GOTO(out_procfs, rc);
7182 osd_scrub_cleanup(env, o);
7184 lu_site_fini(&o->od_site);
7186 osd_obj_map_fini(o);
7193 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7194 struct lu_device_type *t,
7195 struct lustre_cfg *cfg)
7197 struct osd_device *o;
7202 return ERR_PTR(-ENOMEM);
7204 rc = dt_device_init(&o->od_dt_dev, t);
7206 /* Because the ctx might be revived in dt_device_init,
7207 * refill the env here */
7208 lu_env_refill((struct lu_env *)env);
7209 rc = osd_device_init0(env, o, cfg);
7211 dt_device_fini(&o->od_dt_dev);
7214 if (unlikely(rc != 0))
7217 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7220 static struct lu_device *osd_device_free(const struct lu_env *env,
7221 struct lu_device *d)
7223 struct osd_device *o = osd_dev(d);
7226 /* XXX: make osd top device in order to release reference */
7227 d->ld_site->ls_top_dev = d;
7228 lu_site_purge(env, d->ld_site, -1);
7229 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7230 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7231 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7233 lu_site_fini(&o->od_site);
7234 dt_device_fini(&o->od_dt_dev);
7239 static int osd_process_config(const struct lu_env *env,
7240 struct lu_device *d, struct lustre_cfg *cfg)
7242 struct osd_device *o = osd_dev(d);
7246 switch (cfg->lcfg_command) {
7248 rc = osd_mount(env, o, cfg);
7251 lu_dev_del_linkage(d->ld_site, d);
7252 rc = osd_shutdown(env, o);
7255 LASSERT(&o->od_dt_dev);
7256 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
7257 cfg, &o->od_dt_dev);
7258 if (rc > 0 || rc == -ENOSYS)
7259 rc = class_process_proc_param(PARAM_OST,
7260 lprocfs_osd_obd_vars,
7261 cfg, &o->od_dt_dev);
7270 static int osd_recovery_complete(const struct lu_env *env,
7271 struct lu_device *d)
7273 struct osd_device *osd = osd_dev(d);
7277 if (osd->od_quota_slave == NULL)
7280 /* start qsd instance on recovery completion, this notifies the quota
7281 * slave code that we are about to process new requests now */
7282 rc = qsd_start(env, osd->od_quota_slave);
7287 * we use exports to track all osd users
7289 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7290 struct obd_device *obd, struct obd_uuid *cluuid,
7291 struct obd_connect_data *data, void *localdata)
7293 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7294 struct lustre_handle conn;
7298 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7300 rc = class_connect(&conn, obd, cluuid);
7304 *exp = class_conn2export(&conn);
7306 spin_lock(&osd->od_osfs_lock);
7308 spin_unlock(&osd->od_osfs_lock);
7314 * once last export (we don't count self-export) disappeared
7315 * osd can be released
7317 static int osd_obd_disconnect(struct obd_export *exp)
7319 struct obd_device *obd = exp->exp_obd;
7320 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7321 int rc, release = 0;
7324 /* Only disconnect the underlying layers on the final disconnect. */
7325 spin_lock(&osd->od_osfs_lock);
7327 if (osd->od_connects == 0)
7329 spin_unlock(&osd->od_osfs_lock);
7331 rc = class_disconnect(exp); /* bz 9811 */
7333 if (rc == 0 && release)
7334 class_manual_cleanup(obd);
7338 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
7339 struct lu_device *dev)
7341 struct osd_device *osd = osd_dev(dev);
7342 struct lr_server_data *lsd =
7343 &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
7347 if (osd->od_quota_slave != NULL) {
7348 /* set up quota slave objects */
7349 result = qsd_prepare(env, osd->od_quota_slave);
7354 if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
7355 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
7356 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
7357 osd->od_index_in_idif = 1;
7359 osd->od_index_in_idif = 0;
7360 result = osd_register_proc_index_in_idif(osd);
7365 osd->od_index_in_idif = 1;
7369 result = osd_fid_init(env, osd);
7374 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
7375 struct lu_fid *fid, struct md_op_data *op_data)
7377 struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
7379 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
7382 static const struct lu_object_operations osd_lu_obj_ops = {
7383 .loo_object_init = osd_object_init,
7384 .loo_object_delete = osd_object_delete,
7385 .loo_object_release = osd_object_release,
7386 .loo_object_free = osd_object_free,
7387 .loo_object_print = osd_object_print,
7388 .loo_object_invariant = osd_object_invariant
7391 const struct lu_device_operations osd_lu_ops = {
7392 .ldo_object_alloc = osd_object_alloc,
7393 .ldo_process_config = osd_process_config,
7394 .ldo_recovery_complete = osd_recovery_complete,
7395 .ldo_prepare = osd_prepare,
7398 static const struct lu_device_type_operations osd_device_type_ops = {
7399 .ldto_init = osd_type_init,
7400 .ldto_fini = osd_type_fini,
7402 .ldto_start = osd_type_start,
7403 .ldto_stop = osd_type_stop,
7405 .ldto_device_alloc = osd_device_alloc,
7406 .ldto_device_free = osd_device_free,
7408 .ldto_device_init = osd_device_init,
7409 .ldto_device_fini = osd_device_fini
7412 static struct lu_device_type osd_device_type = {
7413 .ldt_tags = LU_DEVICE_DT,
7414 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
7415 .ldt_ops = &osd_device_type_ops,
7416 .ldt_ctx_tags = LCT_LOCAL,
7419 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
7421 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7422 struct super_block *sb = osd_sb(osd);
7424 return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
7428 * lprocfs legacy support.
7430 static struct obd_ops osd_obd_device_ops = {
7431 .o_owner = THIS_MODULE,
7432 .o_connect = osd_obd_connect,
7433 .o_disconnect = osd_obd_disconnect,
7434 .o_fid_alloc = osd_fid_alloc,
7435 .o_health_check = osd_health_check,
7438 static int __init osd_init(void)
7442 LASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
7443 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
7444 /* please, try to keep osd_thread_info smaller than a page */
7445 CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
7450 rc = lu_kmem_init(ldiskfs_caches);
7454 rc = class_register_type(&osd_obd_device_ops, NULL, true,
7455 lprocfs_osd_module_vars,
7456 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
7458 lu_kmem_fini(ldiskfs_caches);
7462 static void __exit osd_exit(void)
7464 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
7465 lu_kmem_fini(ldiskfs_caches);
7468 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7469 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
7470 MODULE_VERSION(LUSTRE_VERSION_STRING);
7471 MODULE_LICENSE("GPL");
7473 module_init(osd_init);
7474 module_exit(osd_exit);