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);
1959 qi->lqi_id.qid_uid = i_projid_read(inode);
1960 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
1966 * Concurrency: ->loo_object_release() is called under site spin-lock.
1968 static void osd_object_release(const struct lu_env *env,
1969 struct lu_object *l)
1971 struct osd_object *o = osd_obj(l);
1972 /* nobody should be releasing a non-destroyed object with nlink=0
1973 * the API allows this, but ldiskfs doesn't like and then report
1974 * this inode as deleted */
1975 if (unlikely(!o->oo_destroyed && o->oo_inode && o->oo_inode->i_nlink == 0))
1980 * Concurrency: shouldn't matter.
1982 static int osd_object_print(const struct lu_env *env, void *cookie,
1983 lu_printer_t p, const struct lu_object *l)
1985 struct osd_object *o = osd_obj(l);
1986 struct iam_descr *d;
1988 if (o->oo_dir != NULL)
1989 d = o->oo_dir->od_container.ic_descr;
1992 return (*p)(env, cookie,
1993 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1995 o->oo_inode ? o->oo_inode->i_ino : 0UL,
1996 o->oo_inode ? o->oo_inode->i_generation : 0,
1997 d ? d->id_ops->id_name : "plain");
2001 * Concurrency: shouldn't matter.
2003 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2004 struct obd_statfs *sfs)
2006 struct osd_device *osd = osd_dt_dev(d);
2007 struct super_block *sb = osd_sb(osd);
2008 struct kstatfs *ksfs;
2012 if (unlikely(osd->od_mnt == NULL))
2013 return -EINPROGRESS;
2015 /* osd_lproc.c call this without env, allocate ksfs for that case */
2016 if (unlikely(env == NULL)) {
2017 OBD_ALLOC_PTR(ksfs);
2021 ksfs = &osd_oti_get(env)->oti_ksfs;
2024 result = sb->s_op->statfs(sb->s_root, ksfs);
2028 statfs_pack(sfs, ksfs);
2029 if (unlikely(sb->s_flags & MS_RDONLY))
2030 sfs->os_state |= OS_STATE_READONLY;
2031 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb,
2032 LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2033 sfs->os_maxbytes = sb->s_maxbytes;
2035 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2038 * Reserve some space so to avoid fragmenting the filesystem too much.
2039 * Fragmentation not only impacts performance, but can also increase
2040 * metadata overhead significantly, causing grant calculation to be
2043 * Reserve 0.78% of total space, at least 8MB for small filesystems.
2045 CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2046 reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2047 if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2048 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2050 sfs->os_blocks -= reserved;
2051 sfs->os_bfree -= min(reserved, sfs->os_bfree);
2052 sfs->os_bavail -= min(reserved, sfs->os_bavail);
2055 if (unlikely(env == NULL))
2061 * Estimate space needed for file creations. We assume the largest filename
2062 * which is 2^64 - 1, hence a filename of 20 chars.
2063 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2065 #ifdef __LDISKFS_DIR_REC_LEN
2066 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2068 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2072 * Concurrency: doesn't access mutable data.
2074 static void osd_conf_get(const struct lu_env *env,
2075 const struct dt_device *dev,
2076 struct dt_device_param *param)
2078 struct super_block *sb = osd_sb(osd_dt_dev(dev));
2082 * XXX should be taken from not-yet-existing fs abstraction layer.
2084 param->ddp_max_name_len = LDISKFS_NAME_LEN;
2085 param->ddp_max_nlink = LDISKFS_LINK_MAX;
2086 param->ddp_symlink_max = sb->s_blocksize;
2087 param->ddp_mount_type = LDD_MT_LDISKFS;
2088 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2089 param->ddp_maxbytes = sb->s_maxbytes;
2091 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2092 /* inode are statically allocated, so per-inode space consumption
2093 * is the space consumed by the directory entry */
2094 param->ddp_inodespace = PER_OBJ_USAGE;
2095 /* EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2096 * = 128MB) which is unlikely to be hit in real life. Report a smaller
2097 * maximum length to not under count the actual number of extents
2098 * needed for writing a file. */
2099 param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2;
2100 /* worst-case extent insertion metadata overhead */
2101 param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2102 param->ddp_mntopts = 0;
2103 if (test_opt(sb, XATTR_USER))
2104 param->ddp_mntopts |= MNTOPT_USERXATTR;
2105 if (test_opt(sb, POSIX_ACL))
2106 param->ddp_mntopts |= MNTOPT_ACL;
2108 /* LOD might calculate the max stripe count based on max_ea_size,
2109 * so we need take account in the overhead as well,
2110 * xattr_header + magic + xattr_entry_head */
2111 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2112 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2114 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2115 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
2116 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2120 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2124 * Concurrency: shouldn't matter.
2126 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2130 CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
2132 rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
2134 CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
2135 LUSTRE_OSD_LDISKFS_NAME, rc);
2141 * Start commit for OSD device.
2143 * An implementation of dt_commit_async method for OSD device.
2144 * Asychronously starts underlayng fs sync and thereby a transaction
2147 * \param env environment
2148 * \param d dt device
2150 * \see dt_device_operations
2152 static int osd_commit_async(const struct lu_env *env,
2153 struct dt_device *d)
2155 struct super_block *s = osd_sb(osd_dt_dev(d));
2158 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
2159 RETURN(s->s_op->sync_fs(s, 0));
2163 * Concurrency: shouldn't matter.
2166 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2168 struct super_block *sb = osd_sb(osd_dt_dev(d));
2169 struct block_device *dev = sb->s_bdev;
2170 #ifdef HAVE_DEV_SET_RDONLY
2171 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2174 int rc = -EOPNOTSUPP;
2178 #ifdef HAVE_DEV_SET_RDONLY
2179 CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
2181 if (sb->s_op->freeze_fs) {
2182 rc = sb->s_op->freeze_fs(sb);
2187 if (jdev && (jdev != dev)) {
2188 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2190 dev_set_rdonly(jdev);
2192 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2193 dev_set_rdonly(dev);
2195 if (sb->s_op->unfreeze_fs)
2196 sb->s_op->unfreeze_fs(sb);
2201 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2202 osd_dt_dev(d)->od_svname, (long)dev, rc);
2208 * Note: we do not count into QUOTA here.
2209 * If we mount with --data_journal we may need more.
2211 const int osd_dto_credits_noquota[DTO_NR] = {
2214 * INDEX_EXTRA_TRANS_BLOCKS(8) +
2215 * SINGLEDATA_TRANS_BLOCKS(8)
2216 * XXX Note: maybe iam need more, since iam have more level than
2219 [DTO_INDEX_INSERT] = 16,
2222 * just modify a single entry, probably merge few within a block
2224 [DTO_INDEX_DELETE] = 1,
2228 [DTO_INDEX_UPDATE] = 16,
2230 * 4(inode, inode bits, groups, GDT)
2231 * notice: OI updates are counted separately with DTO_INDEX_INSERT
2233 [DTO_OBJECT_CREATE] = 4,
2235 * 4(inode, inode bits, groups, GDT)
2236 * notice: OI updates are counted separately with DTO_INDEX_DELETE
2238 [DTO_OBJECT_DELETE] = 4,
2240 * Attr set credits (inode)
2242 [DTO_ATTR_SET_BASE] = 1,
2244 * Xattr set. The same as xattr of EXT3.
2245 * DATA_TRANS_BLOCKS(14)
2246 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2247 * are also counted in. Do not know why?
2249 [DTO_XATTR_SET] = 14,
2251 * credits for inode change during write.
2253 [DTO_WRITE_BASE] = 3,
2255 * credits for single block write.
2257 [DTO_WRITE_BLOCK] = 14,
2259 * Attr set credits for chown.
2260 * This is extra credits for setattr, and it is null without quota
2262 [DTO_ATTR_SET_CHOWN] = 0
2265 static const struct dt_device_operations osd_dt_ops = {
2266 .dt_root_get = osd_root_get,
2267 .dt_statfs = osd_statfs,
2268 .dt_trans_create = osd_trans_create,
2269 .dt_trans_start = osd_trans_start,
2270 .dt_trans_stop = osd_trans_stop,
2271 .dt_trans_cb_add = osd_trans_cb_add,
2272 .dt_conf_get = osd_conf_get,
2273 .dt_sync = osd_sync,
2275 .dt_commit_async = osd_commit_async,
2278 static void osd_object_read_lock(const struct lu_env *env,
2279 struct dt_object *dt, unsigned role)
2281 struct osd_object *obj = osd_dt_obj(dt);
2282 struct osd_thread_info *oti = osd_oti_get(env);
2284 LINVRNT(osd_invariant(obj));
2286 LASSERT(obj->oo_owner != env);
2287 down_read_nested(&obj->oo_sem, role);
2289 LASSERT(obj->oo_owner == NULL);
2293 static void osd_object_write_lock(const struct lu_env *env,
2294 struct dt_object *dt, unsigned role)
2296 struct osd_object *obj = osd_dt_obj(dt);
2297 struct osd_thread_info *oti = osd_oti_get(env);
2299 LINVRNT(osd_invariant(obj));
2301 LASSERT(obj->oo_owner != env);
2302 down_write_nested(&obj->oo_sem, role);
2304 LASSERT(obj->oo_owner == NULL);
2305 obj->oo_owner = env;
2309 static void osd_object_read_unlock(const struct lu_env *env,
2310 struct dt_object *dt)
2312 struct osd_object *obj = osd_dt_obj(dt);
2313 struct osd_thread_info *oti = osd_oti_get(env);
2315 LINVRNT(osd_invariant(obj));
2317 LASSERT(oti->oti_r_locks > 0);
2319 up_read(&obj->oo_sem);
2322 static void osd_object_write_unlock(const struct lu_env *env,
2323 struct dt_object *dt)
2325 struct osd_object *obj = osd_dt_obj(dt);
2326 struct osd_thread_info *oti = osd_oti_get(env);
2328 LINVRNT(osd_invariant(obj));
2330 LASSERT(obj->oo_owner == env);
2331 LASSERT(oti->oti_w_locks > 0);
2333 obj->oo_owner = NULL;
2334 up_write(&obj->oo_sem);
2337 static int osd_object_write_locked(const struct lu_env *env,
2338 struct dt_object *dt)
2340 struct osd_object *obj = osd_dt_obj(dt);
2342 LINVRNT(osd_invariant(obj));
2344 return obj->oo_owner == env;
2347 static struct timespec *osd_inode_time(const struct lu_env *env,
2348 struct inode *inode, __u64 seconds)
2350 struct osd_thread_info *oti = osd_oti_get(env);
2351 struct timespec *t = &oti->oti_time;
2353 t->tv_sec = seconds;
2355 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2359 static void osd_inode_getattr(const struct lu_env *env,
2360 struct inode *inode, struct lu_attr *attr)
2362 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2363 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2364 LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2365 LA_BLKSIZE | LA_TYPE;
2367 attr->la_atime = LTIME_S(inode->i_atime);
2368 attr->la_mtime = LTIME_S(inode->i_mtime);
2369 attr->la_ctime = LTIME_S(inode->i_ctime);
2370 attr->la_mode = inode->i_mode;
2371 attr->la_size = i_size_read(inode);
2372 attr->la_blocks = inode->i_blocks;
2373 attr->la_uid = i_uid_read(inode);
2374 attr->la_gid = i_gid_read(inode);
2375 attr->la_projid = i_projid_read(inode);
2376 attr->la_flags = ll_inode_to_ext_flags(inode->i_flags);
2377 attr->la_nlink = inode->i_nlink;
2378 attr->la_rdev = inode->i_rdev;
2379 attr->la_blksize = 1 << inode->i_blkbits;
2380 attr->la_blkbits = inode->i_blkbits;
2383 static int osd_attr_get(const struct lu_env *env,
2384 struct dt_object *dt,
2385 struct lu_attr *attr)
2387 struct osd_object *obj = osd_dt_obj(dt);
2389 if (unlikely(!dt_object_exists(dt)))
2391 if (unlikely(obj->oo_destroyed))
2394 LASSERT(!dt_object_remote(dt));
2395 LINVRNT(osd_invariant(obj));
2397 spin_lock(&obj->oo_guard);
2398 osd_inode_getattr(env, obj->oo_inode, attr);
2399 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2400 attr->la_flags |= LUSTRE_ORPHAN_FL;
2401 spin_unlock(&obj->oo_guard);
2406 static int osd_declare_attr_qid(const struct lu_env *env,
2407 struct osd_object *obj,
2408 struct osd_thandle *oh, long long bspace,
2409 qid_t old_id, qid_t new_id, bool enforce,
2413 struct osd_thread_info *info = osd_oti_get(env);
2414 struct lquota_id_info *qi = &info->oti_qi;
2416 qi->lqi_type = type;
2417 /* inode accounting */
2418 qi->lqi_is_blk = false;
2420 /* one more inode for the new id ... */
2421 qi->lqi_id.qid_uid = new_id;
2423 /* Reserve credits for the new id */
2424 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2425 if (rc == -EDQUOT || rc == -EINPROGRESS)
2430 /* and one less inode for the current id */
2431 qi->lqi_id.qid_uid = old_id;
2433 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2434 if (rc == -EDQUOT || rc == -EINPROGRESS)
2439 /* block accounting */
2440 qi->lqi_is_blk = true;
2442 /* more blocks for the new uid ... */
2443 qi->lqi_id.qid_uid = new_id;
2444 qi->lqi_space = bspace;
2446 * Credits for the new uid has been reserved, re-use "obj"
2447 * to save credit reservation.
2449 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2450 if (rc == -EDQUOT || rc == -EINPROGRESS)
2455 /* and finally less blocks for the current uid */
2456 qi->lqi_id.qid_uid = old_id;
2457 qi->lqi_space = -bspace;
2458 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2459 if (rc == -EDQUOT || rc == -EINPROGRESS)
2465 static int osd_declare_attr_set(const struct lu_env *env,
2466 struct dt_object *dt,
2467 const struct lu_attr *attr,
2468 struct thandle *handle)
2470 struct osd_thandle *oh;
2471 struct osd_object *obj;
2479 LASSERT(dt != NULL);
2480 LASSERT(handle != NULL);
2482 obj = osd_dt_obj(dt);
2483 LASSERT(osd_invariant(obj));
2485 oh = container_of0(handle, struct osd_thandle, ot_super);
2486 LASSERT(oh->ot_handle == NULL);
2488 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2489 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2491 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2492 osd_dto_credits_noquota[DTO_XATTR_SET]);
2494 if (attr == NULL || obj->oo_inode == NULL)
2497 bspace = obj->oo_inode->i_blocks;
2498 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
2499 bspace = toqb(bspace);
2501 /* Changing ownership is always preformed by super user, it should not
2504 * We still need to call the osd_declare_qid() to calculate the journal
2505 * credits for updating quota accounting files and to trigger quota
2506 * space adjustment once the operation is completed.*/
2507 if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2509 uid = i_uid_read(obj->oo_inode);
2510 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2511 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2512 attr->la_uid, enforce, USRQUOTA);
2516 gid = i_gid_read(obj->oo_inode);
2517 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2518 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2519 i_gid_read(obj->oo_inode), attr->la_gid,
2525 #ifdef HAVE_PROJECT_QUOTA
2526 if (attr->la_valid & LA_PROJID) {
2527 __u32 projid = i_projid_read(obj->oo_inode);
2528 enforce = (attr->la_valid & LA_PROJID) &&
2529 (attr->la_projid != projid);
2530 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2531 (qid_t)projid, (qid_t)attr->la_projid,
2540 static int osd_inode_setattr(const struct lu_env *env,
2541 struct inode *inode, const struct lu_attr *attr)
2543 __u64 bits = attr->la_valid;
2545 /* Only allow set size for regular file */
2546 if (!S_ISREG(inode->i_mode))
2547 bits &= ~(LA_SIZE | LA_BLOCKS);
2552 if (bits & LA_ATIME)
2553 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
2554 if (bits & LA_CTIME)
2555 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
2556 if (bits & LA_MTIME)
2557 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
2558 if (bits & LA_SIZE) {
2559 spin_lock(&inode->i_lock);
2560 LDISKFS_I(inode)->i_disksize = attr->la_size;
2561 i_size_write(inode, attr->la_size);
2562 spin_unlock(&inode->i_lock);
2565 /* OSD should not change "i_blocks" which is used by quota.
2566 * "i_blocks" should be changed by ldiskfs only. */
2568 inode->i_mode = (inode->i_mode & S_IFMT) |
2569 (attr->la_mode & ~S_IFMT);
2571 i_uid_write(inode, attr->la_uid);
2573 i_gid_write(inode, attr->la_gid);
2574 if (bits & LA_PROJID)
2575 i_projid_write(inode, attr->la_projid);
2576 if (bits & LA_NLINK)
2577 set_nlink(inode, attr->la_nlink);
2579 inode->i_rdev = attr->la_rdev;
2581 if (bits & LA_FLAGS) {
2582 /* always keep S_NOCMTIME */
2583 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2589 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2593 if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2594 (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2597 ll_vfs_dq_init(inode);
2599 if (attr->la_valid & LA_UID)
2600 iattr.ia_valid |= ATTR_UID;
2601 if (attr->la_valid & LA_GID)
2602 iattr.ia_valid |= ATTR_GID;
2603 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2604 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2606 rc = ll_vfs_dq_transfer(inode, &iattr);
2608 CERROR("%s: quota transfer failed: rc = %d. Is quota "
2609 "enforcement enabled on the ldiskfs "
2610 "filesystem?\n", inode->i_sb->s_id, rc);
2615 #ifdef HAVE_PROJECT_QUOTA
2616 /* Handle project id Transfer here properly */
2617 if (attr->la_valid & LA_PROJID && attr->la_projid !=
2618 i_projid_read(inode)) {
2619 rc = __ldiskfs_ioctl_setproject(inode, attr->la_projid);
2621 CERROR("%s: quota transfer failed: rc = %d. Is quota "
2622 "enforcement enabled on the ldiskfs "
2623 "filesystem?\n", inode->i_sb->s_id, rc);
2631 static int osd_attr_set(const struct lu_env *env,
2632 struct dt_object *dt,
2633 const struct lu_attr *attr,
2634 struct thandle *handle)
2636 struct osd_object *obj = osd_dt_obj(dt);
2637 struct inode *inode;
2640 if (!dt_object_exists(dt))
2643 LASSERT(handle != NULL);
2644 LASSERT(!dt_object_remote(dt));
2645 LASSERT(osd_invariant(obj));
2647 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2649 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
2650 struct osd_thread_info *oti = osd_oti_get(env);
2651 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
2652 struct lu_fid *fid1 = &oti->oti_fid;
2653 struct osd_inode_id *id = &oti->oti_id;
2654 struct iam_path_descr *ipd;
2655 struct iam_container *bag;
2656 struct osd_thandle *oh;
2659 fid_cpu_to_be(fid1, fid0);
2660 memset(id, 1, sizeof(*id));
2661 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2662 fid0)->oi_dir.od_container;
2663 ipd = osd_idx_ipd_get(env, bag);
2664 if (unlikely(ipd == NULL))
2667 oh = container_of0(handle, struct osd_thandle, ot_super);
2668 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2669 (const struct iam_rec *)id, ipd);
2670 osd_ipd_put(env, bag, ipd);
2671 return(rc > 0 ? 0 : rc);
2674 inode = obj->oo_inode;
2676 rc = osd_quota_transfer(inode, attr);
2680 spin_lock(&obj->oo_guard);
2681 rc = osd_inode_setattr(env, inode, attr);
2682 spin_unlock(&obj->oo_guard);
2686 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2688 if (!(attr->la_valid & LA_FLAGS))
2691 /* Let's check if there are extra flags need to be set into LMA */
2692 if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
2693 struct osd_thread_info *info = osd_oti_get(env);
2694 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
2696 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
2697 &info->oti_ost_attrs);
2701 lma->lma_incompat |=
2702 lustre_to_lma_flags(attr->la_flags);
2703 lustre_lma_swab(lma);
2704 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
2705 lma, sizeof(*lma), XATTR_REPLACE);
2707 struct osd_device *osd = osd_obj2dev(obj);
2709 CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
2710 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
2711 lma->lma_incompat, rc);
2714 attr->la_flags & LUSTRE_LMA_FL_MASKS;
2716 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2719 osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2724 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2725 struct osd_object *obj,
2726 const char *name, const int namelen)
2728 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2731 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2732 umode_t mode, struct dt_allocation_hint *hint,
2736 struct osd_device *osd = osd_obj2dev(obj);
2737 struct osd_thandle *oth;
2738 struct dt_object *parent = NULL;
2739 struct inode *inode;
2741 LINVRNT(osd_invariant(obj));
2742 LASSERT(obj->oo_inode == NULL);
2743 LASSERT(obj->oo_hl_head == NULL);
2745 if (S_ISDIR(mode) && ldiskfs_pdo) {
2746 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2747 if (obj->oo_hl_head == NULL)
2751 oth = container_of(th, struct osd_thandle, ot_super);
2752 LASSERT(oth->ot_handle->h_transaction != NULL);
2754 if (hint != NULL && hint->dah_parent != NULL &&
2755 !dt_object_remote(hint->dah_parent))
2756 parent = hint->dah_parent;
2758 inode = ldiskfs_create_inode(oth->ot_handle,
2759 parent ? osd_dt_obj(parent)->oo_inode :
2760 osd_sb(osd)->s_root->d_inode,
2762 if (!IS_ERR(inode)) {
2763 /* Do not update file c/mtime in ldiskfs. */
2764 inode->i_flags |= S_NOCMTIME;
2766 /* For new created object, it must be consistent,
2767 * and it is unnecessary to scrub against it. */
2768 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2770 obj->oo_inode = inode;
2773 if (obj->oo_hl_head != NULL) {
2774 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2775 obj->oo_hl_head = NULL;
2777 result = PTR_ERR(inode);
2779 LINVRNT(osd_invariant(obj));
2787 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2788 struct lu_attr *attr,
2789 struct dt_allocation_hint *hint,
2790 struct dt_object_format *dof,
2794 struct osd_thandle *oth;
2795 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
2797 LASSERT(S_ISDIR(attr->la_mode));
2799 oth = container_of(th, struct osd_thandle, ot_super);
2800 LASSERT(oth->ot_handle->h_transaction != NULL);
2801 result = osd_mkfile(info, obj, mode, hint, th);
2806 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2807 struct lu_attr *attr,
2808 struct dt_allocation_hint *hint,
2809 struct dt_object_format *dof,
2813 struct osd_thandle *oth;
2814 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2816 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2818 LASSERT(S_ISREG(attr->la_mode));
2820 oth = container_of(th, struct osd_thandle, ot_super);
2821 LASSERT(oth->ot_handle->h_transaction != NULL);
2823 result = osd_mkfile(info, obj, mode, hint, th);
2825 LASSERT(obj->oo_inode != NULL);
2826 if (feat->dif_flags & DT_IND_VARKEY)
2827 result = iam_lvar_create(obj->oo_inode,
2828 feat->dif_keysize_max,
2830 feat->dif_recsize_max,
2833 result = iam_lfix_create(obj->oo_inode,
2834 feat->dif_keysize_max,
2836 feat->dif_recsize_max,
2843 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2844 struct lu_attr *attr,
2845 struct dt_allocation_hint *hint,
2846 struct dt_object_format *dof,
2849 LASSERT(S_ISREG(attr->la_mode));
2850 return osd_mkfile(info, obj, (attr->la_mode &
2851 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2854 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2855 struct lu_attr *attr,
2856 struct dt_allocation_hint *hint,
2857 struct dt_object_format *dof,
2860 LASSERT(S_ISLNK(attr->la_mode));
2861 return osd_mkfile(info, obj, (attr->la_mode &
2862 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2865 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2866 struct lu_attr *attr,
2867 struct dt_allocation_hint *hint,
2868 struct dt_object_format *dof,
2871 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2874 LINVRNT(osd_invariant(obj));
2875 LASSERT(obj->oo_inode == NULL);
2876 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2877 S_ISFIFO(mode) || S_ISSOCK(mode));
2879 result = osd_mkfile(info, obj, mode, hint, th);
2881 LASSERT(obj->oo_inode != NULL);
2883 * This inode should be marked dirty for i_rdev. Currently
2884 * that is done in the osd_attr_init().
2886 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2889 LINVRNT(osd_invariant(obj));
2893 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2895 struct dt_allocation_hint *hint,
2896 struct dt_object_format *dof,
2899 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2901 osd_obj_type_f result;
2917 result = osd_mk_index;
2928 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2929 struct dt_object *parent, struct dt_object *child,
2934 ah->dah_parent = parent;
2935 ah->dah_mode = child_mode;
2937 if (parent != NULL && !dt_object_remote(parent)) {
2938 /* will help to find FID->ino at dt_insert("..") */
2939 struct osd_object *pobj = osd_dt_obj(parent);
2940 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
2944 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2945 struct lu_attr *attr, struct dt_object_format *dof)
2947 struct inode *inode = obj->oo_inode;
2948 __u64 valid = attr->la_valid;
2951 attr->la_valid &= ~(LA_TYPE | LA_MODE);
2953 if (dof->dof_type != DFT_NODE)
2954 attr->la_valid &= ~LA_RDEV;
2955 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2956 attr->la_valid &= ~LA_ATIME;
2957 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2958 attr->la_valid &= ~LA_CTIME;
2959 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2960 attr->la_valid &= ~LA_MTIME;
2962 result = osd_quota_transfer(inode, attr);
2966 if (attr->la_valid != 0) {
2967 result = osd_inode_setattr(info->oti_env, inode, attr);
2969 * The osd_inode_setattr() should always succeed here. The
2970 * only error that could be returned is EDQUOT when we are
2971 * trying to change the UID or GID of the inode. However, this
2972 * should not happen since quota enforcement is no longer
2973 * enabled on ldiskfs (lquota takes care of it).
2975 LASSERTF(result == 0, "%d\n", result);
2976 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2979 attr->la_valid = valid;
2983 * Helper function for osd_object_create()
2985 * \retval 0, on success
2987 static int __osd_object_create(struct osd_thread_info *info,
2988 struct osd_object *obj, struct lu_attr *attr,
2989 struct dt_allocation_hint *hint,
2990 struct dt_object_format *dof,
2996 osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
2998 /* we drop umask so that permissions we pass are not affected */
2999 umask = current->fs->umask;
3000 current->fs->umask = 0;
3002 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3004 if (likely(obj->oo_inode != NULL)) {
3005 LASSERT(obj->oo_inode->i_state & I_NEW);
3007 /* Unlock the inode before attr initialization to avoid
3008 * unnecessary dqget operations. LU-6378 */
3009 unlock_new_inode(obj->oo_inode);
3012 if (likely(result == 0)) {
3013 osd_attr_init(info, obj, attr, dof);
3014 osd_object_init0(obj);
3017 /* restore previous umask value */
3018 current->fs->umask = umask;
3020 osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3026 * Helper function for osd_object_create()
3028 * \retval 0, on success
3030 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3031 const struct lu_fid *fid, struct thandle *th)
3033 struct osd_thread_info *info = osd_oti_get(env);
3034 struct osd_inode_id *id = &info->oti_id;
3035 struct osd_device *osd = osd_obj2dev(obj);
3036 struct osd_thandle *oh;
3039 LASSERT(obj->oo_inode != NULL);
3041 oh = container_of0(th, struct osd_thandle, ot_super);
3042 LASSERT(oh->ot_handle);
3043 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3045 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3046 rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3047 OI_CHECK_FLD, NULL);
3048 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3053 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3054 u64 seq, struct lu_seq_range *range)
3056 struct seq_server_site *ss = osd_seq_site(osd);
3058 if (fid_seq_is_idif(seq)) {
3059 fld_range_set_ost(range);
3060 range->lsr_index = idif_ost_idx(seq);
3064 if (!fid_seq_in_fldb(seq)) {
3065 fld_range_set_mdt(range);
3067 /* FIXME: If ss is NULL, it suppose not get lsr_index
3069 range->lsr_index = ss->ss_node_id;
3073 LASSERT(ss != NULL);
3074 fld_range_set_any(range);
3075 /* OSD will only do local fld lookup */
3076 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3080 * Concurrency: no external locking is necessary.
3082 static int osd_declare_object_create(const struct lu_env *env,
3083 struct dt_object *dt,
3084 struct lu_attr *attr,
3085 struct dt_allocation_hint *hint,
3086 struct dt_object_format *dof,
3087 struct thandle *handle)
3089 struct osd_thandle *oh;
3093 LASSERT(handle != NULL);
3095 oh = container_of0(handle, struct osd_thandle, ot_super);
3096 LASSERT(oh->ot_handle == NULL);
3098 /* EA object consumes more credits than regular object: osd_mk_index
3099 * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3100 * leaf_node, could involves the block, block bitmap, groups, GDT
3101 * change for each block, so add 4 * 2 credits in that case. */
3102 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3103 osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3104 (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3105 /* Reuse idle OI block may cause additional one OI block
3107 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3108 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3113 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3114 attr->la_projid, 1, oh, osd_dt_obj(dt),
3115 false, NULL, false);
3119 /* will help to find FID->ino mapping at dt_insert() */
3120 rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3126 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
3127 struct lu_attr *attr,
3128 struct dt_allocation_hint *hint,
3129 struct dt_object_format *dof, struct thandle *th)
3131 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3132 struct osd_object *obj = osd_dt_obj(dt);
3133 struct osd_thread_info *info = osd_oti_get(env);
3137 if (dt_object_exists(dt))
3140 LINVRNT(osd_invariant(obj));
3141 LASSERT(!dt_object_remote(dt));
3142 LASSERT(osd_write_locked(env, obj));
3143 LASSERT(th != NULL);
3145 if (unlikely(fid_is_acct(fid)))
3146 /* Quota files can't be created from the kernel any more,
3147 * 'tune2fs -O quota' will take care of creating them */
3150 result = __osd_object_create(info, obj, attr, hint, dof, th);
3152 result = __osd_oi_insert(env, obj, fid, th);
3153 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3154 obj->oo_dt.do_body_ops = &osd_body_ops;
3156 LASSERT(ergo(result == 0,
3157 dt_object_exists(dt) && !dt_object_remote(dt)));
3159 LASSERT(osd_invariant(obj));
3164 * Called to destroy on-disk representation of the object
3166 * Concurrency: must be locked
3168 static int osd_declare_object_destroy(const struct lu_env *env,
3169 struct dt_object *dt,
3172 struct osd_object *obj = osd_dt_obj(dt);
3173 struct inode *inode = obj->oo_inode;
3174 struct osd_thandle *oh;
3181 oh = container_of0(th, struct osd_thandle, ot_super);
3182 LASSERT(oh->ot_handle == NULL);
3184 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3185 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3186 /* Recycle idle OI leaf may cause additional three OI blocks
3188 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3189 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3190 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3191 /* one less inode */
3192 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3193 i_projid_read(inode), -1, oh, obj, false,
3197 /* data to be truncated */
3198 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3199 i_projid_read(inode), 0, oh, obj, true,
3204 /* will help to find FID->ino when this object is being
3205 * added to PENDING/ */
3206 rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3211 static int osd_object_destroy(const struct lu_env *env,
3212 struct dt_object *dt,
3215 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3216 struct osd_object *obj = osd_dt_obj(dt);
3217 struct inode *inode = obj->oo_inode;
3218 struct osd_device *osd = osd_obj2dev(obj);
3219 struct osd_thandle *oh;
3223 oh = container_of0(th, struct osd_thandle, ot_super);
3224 LASSERT(oh->ot_handle);
3226 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3228 if (unlikely(fid_is_acct(fid)))
3231 if (S_ISDIR(inode->i_mode)) {
3232 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3233 inode->i_nlink == 2);
3234 /* it will check/delete the inode from remote parent,
3235 * how to optimize it? unlink performance impaction XXX */
3236 result = osd_delete_from_remote_parent(env, osd, obj, oh);
3238 CERROR("%s: delete inode "DFID": rc = %d\n",
3239 osd_name(osd), PFID(fid), result);
3241 spin_lock(&obj->oo_guard);
3243 spin_unlock(&obj->oo_guard);
3244 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3247 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3249 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3251 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3252 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3253 oh->ot_handle, OI_CHECK_FLD);
3255 osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3256 /* XXX: add to ext3 orphan list */
3257 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3259 /* not needed in the cache anymore */
3260 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3261 obj->oo_destroyed = 1;
3267 * Put the fid into lustre_mdt_attrs, and then place the structure
3268 * inode's ea. This fid should not be altered during the life time
3271 * \retval +ve, on success
3272 * \retval -ve, on error
3274 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3276 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3277 const struct lu_fid *fid, __u32 compat, __u32 incompat)
3279 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3280 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3284 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3287 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3290 lustre_loa_init(loa, fid, compat, incompat);
3291 lustre_loa_swab(loa, false);
3293 /* For the OST device with 256 bytes inode size by default,
3294 * the PFID EA will be stored together with LMA EA to avoid
3295 * performance trouble. Otherwise the PFID EA can be stored
3296 * independently. LU-8998 */
3297 if ((compat & LMAC_FID_ON_OST) &&
3298 LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3299 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3300 sizeof(*loa), XATTR_CREATE);
3302 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3303 sizeof(*lma), XATTR_CREATE);
3304 /* LMA may already exist, but we need to check that all the
3305 * desired compat/incompat flags have been added. */
3306 if (unlikely(rc == -EEXIST)) {
3307 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3308 XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3312 if (rc < sizeof(*lma))
3315 lustre_loa_swab(loa, true);
3316 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3317 ((compat == 0 && incompat == 0) ||
3318 (!(~lma->lma_compat & compat) &&
3319 !(~lma->lma_incompat & incompat))))
3322 lma->lma_self_fid = *fid;
3323 lma->lma_compat |= compat;
3324 lma->lma_incompat |= incompat;
3325 if (rc == sizeof(*lma)) {
3326 lustre_lma_swab(lma);
3327 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3328 sizeof(*lma), XATTR_REPLACE);
3330 lustre_loa_swab(loa, false);
3331 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3332 sizeof(*loa), XATTR_REPLACE);
3340 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3341 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3342 * To have compatilibility with 1.8 ldiskfs driver we need to have
3343 * magic number at start of fid data.
3344 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3347 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3348 const struct lu_fid *fid)
3350 if (!fid_is_namespace_visible(fid) ||
3351 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3352 param->edp_magic = 0;
3356 param->edp_magic = LDISKFS_LUFID_MAGIC;
3357 param->edp_len = sizeof(struct lu_fid) + 1;
3358 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3362 * Try to read the fid from inode ea into dt_rec.
3364 * \param fid object fid.
3366 * \retval 0 on success
3368 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3369 __u32 ino, struct lu_fid *fid,
3370 struct osd_inode_id *id)
3372 struct osd_thread_info *info = osd_oti_get(env);
3373 struct inode *inode;
3376 osd_id_gen(id, ino, OSD_OII_NOGEN);
3377 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3379 RETURN(PTR_ERR(inode));
3385 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3387 struct inode *parent_dir,
3388 const struct lu_fid *dot_fid,
3389 const struct lu_fid *dot_dot_fid,
3390 struct osd_thandle *oth)
3392 struct ldiskfs_dentry_param *dot_ldp;
3393 struct ldiskfs_dentry_param *dot_dot_ldp;
3394 __u32 saved_nlink = dir->i_nlink;
3397 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3398 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3400 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3401 dot_ldp->edp_magic = 0;
3403 rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3404 dir, dot_ldp, dot_dot_ldp);
3405 /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3406 * the subseqent ref_add() will increase the dir->i_nlink
3407 * as 3. That is incorrect for new created directory.
3409 * It looks like hack, because we want to make the OSD API
3410 * to be order-independent for new created directory object
3411 * between dt_insert(..) and ref_add() operations.
3413 * Here, we only restore the in-RAM dir-inode's nlink attr,
3414 * becuase if the nlink attr is not 2, then there will be
3415 * ref_add() called following the dt_insert(..), such call
3416 * will make both the in-RAM and on-disk dir-inode's nlink
3417 * attr to be set as 2. LU-7447 */
3418 set_nlink(dir, saved_nlink);
3423 * Create an local agent inode for remote entry
3425 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3426 struct osd_device *osd,
3427 struct osd_object *pobj,
3428 const struct lu_fid *fid,
3432 struct osd_thread_info *info = osd_oti_get(env);
3433 struct inode *local;
3434 struct osd_thandle *oh;
3439 oh = container_of(th, struct osd_thandle, ot_super);
3440 LASSERT(oh->ot_handle->h_transaction != NULL);
3442 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3443 if (IS_ERR(local)) {
3444 CERROR("%s: create local error %d\n", osd_name(osd),
3445 (int)PTR_ERR(local));
3449 /* restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3450 * correct gid on remote file, not agent here */
3451 local->i_gid = current_fsgid();
3452 ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3453 unlock_new_inode(local);
3455 /* Set special LMA flag for local agent inode */
3456 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3458 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3459 osd_name(osd), PFID(fid), rc);
3460 RETURN(ERR_PTR(rc));
3466 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3467 lu_object_fid(&pobj->oo_dt.do_lu),
3470 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3471 osd_name(osd), PFID(fid), rc);
3472 RETURN(ERR_PTR(rc));
3479 * when direntry is deleted, we have to take care of possible agent inode
3480 * referenced by that. unfortunately we can't do this at that point:
3481 * iget() within a running transaction leads to deadlock and we better do
3482 * not call that every delete declaration to save performance. so we put
3483 * a potention agent inode on a list and process that once the transaction
3484 * is over. Notice it's not any worse in terms of real orphans as regular
3485 * object destroy doesn't put inodes on the on-disk orphan list. this should
3486 * be addressed separately
3488 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3489 struct osd_thandle *oh,
3492 struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3493 struct osd_obj_orphan *oor;
3500 oor->oor_env = (struct lu_env *)env;
3501 spin_lock(&osd->od_osfs_lock);
3502 list_add(&oor->oor_list, &osd->od_orphan_list);
3503 spin_unlock(&osd->od_osfs_lock);
3505 oh->ot_remove_agents = 1;
3511 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3512 struct osd_device *osd)
3514 struct osd_thread_info *info = osd_oti_get(env);
3515 struct osd_obj_orphan *oor, *tmp;
3516 struct osd_inode_id id;
3517 struct list_head list;
3518 struct inode *inode;
3523 INIT_LIST_HEAD(&list);
3525 spin_lock(&osd->od_osfs_lock);
3526 list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3527 if (oor->oor_env == env) {
3528 list_del(&oor->oor_list);
3529 list_add(&oor->oor_list, &list);
3532 spin_unlock(&osd->od_osfs_lock);
3534 list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3538 list_del(&oor->oor_list);
3541 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3542 inode = osd_iget_fid(info, osd, &id, &fid);
3546 if (!osd_remote_fid(env, osd, &fid)) {
3551 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3553 mark_inode_dirty(inode);
3554 ldiskfs_journal_stop(jh);
3562 * OSD layer object create function for interoperability mode (b11826).
3563 * This is mostly similar to osd_object_create(). Only difference being, fid is
3564 * inserted into inode ea here.
3566 * \retval 0, on success
3567 * \retval -ve, on error
3569 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
3570 struct lu_attr *attr,
3571 struct dt_allocation_hint *hint,
3572 struct dt_object_format *dof,
3575 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3576 struct osd_object *obj = osd_dt_obj(dt);
3577 struct osd_thread_info *info = osd_oti_get(env);
3578 int result, on_ost = 0;
3582 if (dt_object_exists(dt))
3585 LASSERT(osd_invariant(obj));
3586 LASSERT(!dt_object_remote(dt));
3587 LASSERT(osd_write_locked(env, obj));
3588 LASSERT(th != NULL);
3590 if (unlikely(fid_is_acct(fid)))
3591 /* Quota files can't be created from the kernel any more,
3592 * 'tune2fs -O quota' will take care of creating them */
3595 result = __osd_object_create(info, obj, attr, hint, dof, th);
3597 if (fid_is_idif(fid) &&
3598 !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3599 struct lu_fid *tfid = &info->oti_fid;
3600 struct ost_id *oi = &info->oti_ostid;
3602 fid_to_ostid(fid, oi);
3603 ostid_to_fid(tfid, oi, 0);
3605 result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3606 LMAC_FID_ON_OST, 0);
3608 on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3610 result = osd_ea_fid_set(info, obj->oo_inode, fid,
3611 on_ost ? LMAC_FID_ON_OST : 0,
3614 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3615 obj->oo_dt.do_body_ops = &osd_body_ops;
3619 result = __osd_oi_insert(env, obj, fid, th);
3621 /* a small optimization - dt_insert() isn't usually applied
3622 * to OST objects, so we don't need to cache OI mapping for
3624 if (result == 0 && on_ost == 0) {
3625 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3626 result = osd_idc_find_and_init(env, osd, obj);
3627 LASSERT(result == 0);
3630 LASSERT(ergo(result == 0,
3631 dt_object_exists(dt) && !dt_object_remote(dt)));
3632 LINVRNT(osd_invariant(obj));
3636 static int osd_declare_object_ref_add(const struct lu_env *env,
3637 struct dt_object *dt,
3638 struct thandle *handle)
3640 struct osd_thandle *oh;
3642 /* it's possible that object doesn't exist yet */
3643 LASSERT(handle != NULL);
3645 oh = container_of0(handle, struct osd_thandle, ot_super);
3646 LASSERT(oh->ot_handle == NULL);
3648 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3649 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3655 * Concurrency: @dt is write locked.
3657 static int osd_object_ref_add(const struct lu_env *env,
3658 struct dt_object *dt, struct thandle *th)
3660 struct osd_object *obj = osd_dt_obj(dt);
3661 struct inode *inode = obj->oo_inode;
3662 struct osd_thandle *oh;
3665 if (!dt_object_exists(dt) || obj->oo_destroyed)
3668 LINVRNT(osd_invariant(obj));
3669 LASSERT(!dt_object_remote(dt));
3670 LASSERT(osd_write_locked(env, obj));
3671 LASSERT(th != NULL);
3673 oh = container_of0(th, struct osd_thandle, ot_super);
3674 LASSERT(oh->ot_handle != NULL);
3676 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
3678 CDEBUG(D_INODE, DFID" increase nlink %d\n",
3679 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3681 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
3682 * (65000) subdirectories by storing "1" in i_nlink if the link count
3683 * would otherwise overflow. Directory tranversal tools understand
3684 * that (st_nlink == 1) indicates that the filesystem dose not track
3685 * hard links count on the directory, and will not abort subdirectory
3686 * scanning early once (st_nlink - 2) subdirs have been found.
3688 * This also has to properly handle the case of inodes with nlink == 0
3689 * in case they are being linked into the PENDING directory
3691 spin_lock(&obj->oo_guard);
3692 if (unlikely(inode->i_nlink == 0))
3693 /* inc_nlink from 0 may cause WARN_ON */
3694 set_nlink(inode, 1);
3696 ldiskfs_inc_count(oh->ot_handle, inode);
3697 if (!S_ISDIR(inode->i_mode))
3698 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
3700 spin_unlock(&obj->oo_guard);
3702 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3703 LINVRNT(osd_invariant(obj));
3705 osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
3710 static int osd_declare_object_ref_del(const struct lu_env *env,
3711 struct dt_object *dt,
3712 struct thandle *handle)
3714 struct osd_thandle *oh;
3716 if (!dt_object_exists(dt))
3719 LASSERT(!dt_object_remote(dt));
3720 LASSERT(handle != NULL);
3722 oh = container_of0(handle, struct osd_thandle, ot_super);
3723 LASSERT(oh->ot_handle == NULL);
3725 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
3726 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3732 * Concurrency: @dt is write locked.
3734 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
3737 struct osd_object *obj = osd_dt_obj(dt);
3738 struct inode *inode = obj->oo_inode;
3739 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3740 struct osd_thandle *oh;
3742 if (!dt_object_exists(dt))
3745 LINVRNT(osd_invariant(obj));
3746 LASSERT(!dt_object_remote(dt));
3747 LASSERT(osd_write_locked(env, obj));
3748 LASSERT(th != NULL);
3750 oh = container_of0(th, struct osd_thandle, ot_super);
3751 LASSERT(oh->ot_handle != NULL);
3753 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
3755 spin_lock(&obj->oo_guard);
3756 /* That can be result of upgrade from old Lustre version and
3757 * applied only to local files. Just skip this ref_del call.
3758 * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
3759 if (inode->i_nlink == 0) {
3760 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
3761 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
3762 ", maybe an upgraded file? (LU-3915)\n",
3763 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
3764 spin_unlock(&obj->oo_guard);
3768 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
3769 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3771 ldiskfs_dec_count(oh->ot_handle, inode);
3772 spin_unlock(&obj->oo_guard);
3774 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3775 LINVRNT(osd_invariant(obj));
3777 osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
3783 * Concurrency: @dt is read locked.
3785 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
3786 struct lu_buf *buf, const char *name)
3788 struct osd_object *obj = osd_dt_obj(dt);
3789 struct inode *inode = obj->oo_inode;
3790 struct osd_thread_info *info = osd_oti_get(env);
3791 struct dentry *dentry = &info->oti_obj_dentry;
3792 bool cache_xattr = false;
3797 /* version get is not real XATTR but uses xattr API */
3798 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3799 dt_obj_version_t *ver = buf->lb_buf;
3801 /* for version we are just using xattr API but change inode
3803 if (buf->lb_len == 0)
3804 return sizeof(dt_obj_version_t);
3806 if (buf->lb_len < sizeof(dt_obj_version_t))
3809 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
3810 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3812 *ver = LDISKFS_I(inode)->i_fs_version;
3814 return sizeof(dt_obj_version_t);
3817 if (!dt_object_exists(dt))
3820 LASSERT(!dt_object_remote(dt));
3821 LASSERT(inode->i_op != NULL);
3822 LASSERT(inode->i_op->getxattr != NULL);
3824 if (strcmp(name, XATTR_NAME_LOV) == 0 ||
3825 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
3829 rc = osd_oxc_get(obj, name, buf);
3834 rc = __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
3835 if (rc == -ENODATA && strcmp(name, XATTR_NAME_FID) == 0) {
3836 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3837 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3838 struct filter_fid *ff;
3839 struct ost_layout *ol;
3841 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
3843 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
3847 if (!(lma->lma_compat & LMAC_STRIPE_INFO)) {
3853 if (buf->lb_len == 0 || !buf->lb_buf)
3856 if (buf->lb_len < rc)
3860 ol = &ff->ff_layout;
3861 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
3862 PFID_STRIPE_IDX_BITS);
3863 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
3864 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
3865 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
3866 if (lma->lma_compat & LMAC_COMP_INFO) {
3867 ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
3868 ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
3869 ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
3871 ol->ol_comp_start = 0;
3872 ol->ol_comp_end = 0;
3879 if (rc == -ENOENT || rc == -ENODATA)
3880 osd_oxc_add(obj, name, NULL, 0);
3881 else if (rc > 0 && buf->lb_buf != NULL)
3882 osd_oxc_add(obj, name, buf->lb_buf, rc);
3888 static int osd_declare_xattr_set(const struct lu_env *env,
3889 struct dt_object *dt,
3890 const struct lu_buf *buf, const char *name,
3891 int fl, struct thandle *handle)
3893 struct osd_thandle *oh;
3895 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3897 LASSERT(handle != NULL);
3899 oh = container_of0(handle, struct osd_thandle, ot_super);
3900 LASSERT(oh->ot_handle == NULL);
3902 if (strcmp(name, XATTR_NAME_LMA) == 0) {
3903 /* For non-upgrading case, the LMA is set first and
3904 * usually fit inode. But for upgrade case, the LMA
3905 * may be in another separated EA block. */
3906 if (dt_object_exists(dt)) {
3907 if (fl == LU_XATTR_REPLACE)
3912 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3914 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3915 /* We may need to delete the old PFID EA. */
3916 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
3917 if (fl == LU_XATTR_REPLACE)
3924 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
3929 if (buf->lb_buf == NULL && dt_object_exists(dt)) {
3930 /* learn xattr size from osd_xattr_get if
3931 attribute has not been read yet */
3932 buflen = __osd_xattr_get(
3933 osd_dt_obj(dt)->oo_inode,
3934 &osd_oti_get(env)->oti_obj_dentry,
3939 buflen = buf->lb_len;
3942 if (buflen > sb->s_blocksize) {
3943 credits += osd_calc_bkmap_credits(
3945 (buflen + sb->s_blocksize - 1) >>
3946 sb->s_blocksize_bits);
3950 * xattr set may involve inode quota change, reserve credits for
3951 * dquot_initialize()
3953 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3956 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3962 * Concurrency: @dt is write locked.
3964 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3965 const struct lu_buf *buf, const char *name, int fl,
3966 struct thandle *handle)
3968 struct osd_object *obj = osd_dt_obj(dt);
3969 struct inode *inode = obj->oo_inode;
3970 struct osd_thread_info *info = osd_oti_get(env);
3971 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3972 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3981 /* version set is not real XATTR */
3982 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3983 dt_obj_version_t *version = buf->lb_buf;
3985 /* for version we are just using xattr API but change inode
3987 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3989 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
3990 *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3992 LDISKFS_I(inode)->i_fs_version = *version;
3993 /* Version is set after all inode operations are finished,
3994 * so we should mark it dirty here */
3995 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4000 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4001 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4004 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4005 if (fl & LU_XATTR_REPLACE)
4006 fs_flags |= XATTR_REPLACE;
4008 if (fl & LU_XATTR_CREATE)
4009 fs_flags |= XATTR_CREATE;
4011 /* For the OST device with 256 bytes inode size by default,
4012 * the PFID EA will be stored together with LMA EA to avoid
4013 * performance trouble. Otherwise the PFID EA can be stored
4014 * independently. LU-8998 */
4015 if (strcmp(name, XATTR_NAME_FID) == 0 &&
4016 LDISKFS_INODE_SIZE(inode->i_sb) <= 256) {
4017 struct dentry *dentry = &info->oti_obj_dentry;
4018 struct filter_fid *ff;
4019 struct ost_layout *ol;
4022 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
4025 ol = &ff->ff_layout;
4026 /* Old client does not send stripe information, store
4027 * the PFID EA on disk directly. */
4028 if (buf->lb_len == sizeof(struct lu_fid) ||
4029 ol->ol_stripe_size == 0) {
4030 len = sizeof(struct lu_fid);
4034 if (buf->lb_len != sizeof(*ff))
4037 rc = osd_get_lma(info, inode, dentry, loa);
4038 if (unlikely(rc == -ENODATA)) {
4039 /* Usually for upgarding from old device */
4040 lustre_loa_init(loa, lu_object_fid(&dt->do_lu),
4041 LMAC_FID_ON_OST, 0);
4049 fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4050 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4051 PFID_STRIPE_IDX_BITS;
4052 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4053 lma->lma_compat |= LMAC_STRIPE_INFO;
4054 if (ol->ol_comp_id != 0) {
4055 loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4056 loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4057 loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4058 lma->lma_compat |= LMAC_COMP_INFO;
4061 lustre_loa_swab(loa, false);
4063 /* Remove old PFID EA entry firstly. */
4064 ll_vfs_dq_init(inode);
4065 rc = inode->i_op->removexattr(dentry, name);
4066 if (rc && rc != -ENODATA)
4069 /* Store the PFID EA inside the LMA EA. */
4070 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
4074 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4075 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4079 lma->lma_incompat |= LMAI_STRIPED;
4080 lustre_lma_swab(lma);
4081 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4082 sizeof(*lma), XATTR_REPLACE);
4088 rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4089 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4092 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4093 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4094 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4100 * Concurrency: @dt is read locked.
4102 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4103 const struct lu_buf *buf)
4105 struct osd_object *obj = osd_dt_obj(dt);
4106 struct inode *inode = obj->oo_inode;
4107 struct osd_thread_info *info = osd_oti_get(env);
4108 struct dentry *dentry = &info->oti_obj_dentry;
4110 if (!dt_object_exists(dt))
4113 LASSERT(!dt_object_remote(dt));
4114 LASSERT(inode->i_op != NULL);
4115 LASSERT(inode->i_op->listxattr != NULL);
4117 dentry->d_inode = inode;
4118 dentry->d_sb = inode->i_sb;
4119 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4122 static int osd_declare_xattr_del(const struct lu_env *env,
4123 struct dt_object *dt, const char *name,
4124 struct thandle *handle)
4126 struct osd_thandle *oh;
4127 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4129 LASSERT(!dt_object_remote(dt));
4130 LASSERT(handle != NULL);
4132 oh = container_of0(handle, struct osd_thandle, ot_super);
4133 LASSERT(oh->ot_handle == NULL);
4135 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4136 osd_dto_credits_noquota[DTO_XATTR_SET]);
4138 * xattr del may involve inode quota change, reserve credits for
4139 * dquot_initialize()
4141 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4147 * Concurrency: @dt is write locked.
4149 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4150 const char *name, struct thandle *handle)
4152 struct osd_object *obj = osd_dt_obj(dt);
4153 struct inode *inode = obj->oo_inode;
4154 struct osd_thread_info *info = osd_oti_get(env);
4155 struct dentry *dentry = &info->oti_obj_dentry;
4158 if (!dt_object_exists(dt))
4161 LASSERT(!dt_object_remote(dt));
4162 LASSERT(inode->i_op != NULL);
4163 LASSERT(inode->i_op->removexattr != NULL);
4164 LASSERT(handle != NULL);
4166 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4168 ll_vfs_dq_init(inode);
4169 dentry->d_inode = inode;
4170 dentry->d_sb = inode->i_sb;
4171 rc = inode->i_op->removexattr(dentry, name);
4172 if (rc == -ENODATA && strcmp(name, XATTR_NAME_FID) == 0) {
4173 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4175 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
4177 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4178 &info->oti_ost_attrs);
4180 if (!(lma->lma_compat & LMAC_STRIPE_INFO)) {
4185 lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4186 lustre_lma_swab(lma);
4187 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4188 sizeof(*lma), XATTR_REPLACE);
4193 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4196 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4197 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4198 osd_oxc_del(obj, name);
4203 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4204 __u64 start, __u64 end)
4206 struct osd_object *obj = osd_dt_obj(dt);
4207 struct inode *inode = obj->oo_inode;
4208 struct osd_thread_info *info = osd_oti_get(env);
4209 struct dentry *dentry = &info->oti_obj_dentry;
4210 struct file *file = &info->oti_file;
4215 dentry->d_inode = inode;
4216 dentry->d_sb = inode->i_sb;
4217 file->f_path.dentry = dentry;
4218 file->f_mapping = inode->i_mapping;
4219 file->f_op = inode->i_fop;
4220 set_file_inode(file, inode);
4222 rc = ll_vfs_fsync_range(file, start, end, 0);
4227 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4236 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4237 const struct dt_index_features *feat)
4239 struct iam_descr *descr;
4241 if (osd_object_is_root(o))
4242 return feat == &dt_directory_features;
4244 LASSERT(o->oo_dir != NULL);
4246 descr = o->oo_dir->od_container.ic_descr;
4247 if (feat == &dt_directory_features) {
4248 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4254 feat->dif_keysize_min <= descr->id_key_size &&
4255 descr->id_key_size <= feat->dif_keysize_max &&
4256 feat->dif_recsize_min <= descr->id_rec_size &&
4257 descr->id_rec_size <= feat->dif_recsize_max &&
4258 !(feat->dif_flags & (DT_IND_VARKEY |
4259 DT_IND_VARREC | DT_IND_NONUNQ)) &&
4260 ergo(feat->dif_flags & DT_IND_UPDATE,
4261 1 /* XXX check that object (and file system) is
4266 static int osd_iam_container_init(const struct lu_env *env,
4267 struct osd_object *obj,
4268 struct osd_directory *dir)
4270 struct iam_container *bag = &dir->od_container;
4273 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4277 result = iam_container_setup(bag);
4279 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4281 iam_container_fini(bag);
4288 * Concurrency: no external locking is necessary.
4290 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4291 const struct dt_index_features *feat)
4295 struct osd_object *obj = osd_dt_obj(dt);
4297 LINVRNT(osd_invariant(obj));
4299 if (osd_object_is_root(obj)) {
4300 dt->do_index_ops = &osd_index_ea_ops;
4302 } else if (feat == &dt_directory_features) {
4303 dt->do_index_ops = &osd_index_ea_ops;
4304 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4309 } else if (unlikely(feat == &dt_otable_features)) {
4310 dt->do_index_ops = &osd_otable_ops;
4312 } else if (unlikely(feat == &dt_acct_features)) {
4313 dt->do_index_ops = &osd_acct_index_ops;
4316 } else if (!osd_has_index(obj)) {
4317 struct osd_directory *dir;
4322 spin_lock(&obj->oo_guard);
4323 if (obj->oo_dir == NULL)
4327 * Concurrent thread allocated container data.
4330 spin_unlock(&obj->oo_guard);
4332 * Now, that we have container data, serialize its
4335 down_write(&obj->oo_ext_idx_sem);
4337 * recheck under lock.
4339 if (!osd_has_index(obj))
4340 result = osd_iam_container_init(env, obj,
4344 up_write(&obj->oo_ext_idx_sem);
4352 if (result == 0 && skip_iam == 0) {
4353 if (!osd_iam_index_probe(env, obj, feat))
4356 LINVRNT(osd_invariant(obj));
4361 static int osd_otable_it_attr_get(const struct lu_env *env,
4362 struct dt_object *dt,
4363 struct lu_attr *attr)
4369 static const struct dt_object_operations osd_obj_ops = {
4370 .do_read_lock = osd_object_read_lock,
4371 .do_write_lock = osd_object_write_lock,
4372 .do_read_unlock = osd_object_read_unlock,
4373 .do_write_unlock = osd_object_write_unlock,
4374 .do_write_locked = osd_object_write_locked,
4375 .do_attr_get = osd_attr_get,
4376 .do_declare_attr_set = osd_declare_attr_set,
4377 .do_attr_set = osd_attr_set,
4378 .do_ah_init = osd_ah_init,
4379 .do_declare_create = osd_declare_object_create,
4380 .do_create = osd_object_create,
4381 .do_declare_destroy = osd_declare_object_destroy,
4382 .do_destroy = osd_object_destroy,
4383 .do_index_try = osd_index_try,
4384 .do_declare_ref_add = osd_declare_object_ref_add,
4385 .do_ref_add = osd_object_ref_add,
4386 .do_declare_ref_del = osd_declare_object_ref_del,
4387 .do_ref_del = osd_object_ref_del,
4388 .do_xattr_get = osd_xattr_get,
4389 .do_declare_xattr_set = osd_declare_xattr_set,
4390 .do_xattr_set = osd_xattr_set,
4391 .do_declare_xattr_del = osd_declare_xattr_del,
4392 .do_xattr_del = osd_xattr_del,
4393 .do_xattr_list = osd_xattr_list,
4394 .do_object_sync = osd_object_sync,
4395 .do_invalidate = osd_invalidate,
4399 * dt_object_operations for interoperability mode
4400 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
4402 static const struct dt_object_operations osd_obj_ea_ops = {
4403 .do_read_lock = osd_object_read_lock,
4404 .do_write_lock = osd_object_write_lock,
4405 .do_read_unlock = osd_object_read_unlock,
4406 .do_write_unlock = osd_object_write_unlock,
4407 .do_write_locked = osd_object_write_locked,
4408 .do_attr_get = osd_attr_get,
4409 .do_declare_attr_set = osd_declare_attr_set,
4410 .do_attr_set = osd_attr_set,
4411 .do_ah_init = osd_ah_init,
4412 .do_declare_create = osd_declare_object_create,
4413 .do_create = osd_object_ea_create,
4414 .do_declare_destroy = osd_declare_object_destroy,
4415 .do_destroy = osd_object_destroy,
4416 .do_index_try = osd_index_try,
4417 .do_declare_ref_add = osd_declare_object_ref_add,
4418 .do_ref_add = osd_object_ref_add,
4419 .do_declare_ref_del = osd_declare_object_ref_del,
4420 .do_ref_del = osd_object_ref_del,
4421 .do_xattr_get = osd_xattr_get,
4422 .do_declare_xattr_set = osd_declare_xattr_set,
4423 .do_xattr_set = osd_xattr_set,
4424 .do_declare_xattr_del = osd_declare_xattr_del,
4425 .do_xattr_del = osd_xattr_del,
4426 .do_xattr_list = osd_xattr_list,
4427 .do_object_sync = osd_object_sync,
4428 .do_invalidate = osd_invalidate,
4431 static const struct dt_object_operations osd_obj_otable_it_ops = {
4432 .do_attr_get = osd_otable_it_attr_get,
4433 .do_index_try = osd_index_try,
4436 static int osd_index_declare_iam_delete(const struct lu_env *env,
4437 struct dt_object *dt,
4438 const struct dt_key *key,
4439 struct thandle *handle)
4441 struct osd_thandle *oh;
4443 oh = container_of0(handle, struct osd_thandle, ot_super);
4444 LASSERT(oh->ot_handle == NULL);
4446 /* Recycle may cause additional three blocks to be changed. */
4447 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4448 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4454 * delete a (key, value) pair from index \a dt specified by \a key
4456 * \param dt osd index object
4457 * \param key key for index
4458 * \param rec record reference
4459 * \param handle transaction handler
4462 * \retval -ve failure
4464 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4465 const struct dt_key *key,
4466 struct thandle *handle)
4468 struct osd_thread_info *oti = osd_oti_get(env);
4469 struct osd_object *obj = osd_dt_obj(dt);
4470 struct osd_thandle *oh;
4471 struct iam_path_descr *ipd;
4472 struct iam_container *bag = &obj->oo_dir->od_container;
4476 if (!dt_object_exists(dt))
4479 LINVRNT(osd_invariant(obj));
4480 LASSERT(!dt_object_remote(dt));
4481 LASSERT(bag->ic_object == obj->oo_inode);
4482 LASSERT(handle != NULL);
4484 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4486 ipd = osd_idx_ipd_get(env, bag);
4487 if (unlikely(ipd == NULL))
4490 oh = container_of0(handle, struct osd_thandle, ot_super);
4491 LASSERT(oh->ot_handle != NULL);
4492 LASSERT(oh->ot_handle->h_transaction != NULL);
4494 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4495 /* swab quota uid/gid provided by caller */
4496 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4497 key = (const struct dt_key *)&oti->oti_quota_id;
4500 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4501 osd_ipd_put(env, bag, ipd);
4502 LINVRNT(osd_invariant(obj));
4503 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4507 static int osd_index_declare_ea_delete(const struct lu_env *env,
4508 struct dt_object *dt,
4509 const struct dt_key *key,
4510 struct thandle *handle)
4512 struct osd_thandle *oh;
4513 struct inode *inode;
4517 LASSERT(!dt_object_remote(dt));
4518 LASSERT(handle != NULL);
4520 oh = container_of0(handle, struct osd_thandle, ot_super);
4521 LASSERT(oh->ot_handle == NULL);
4523 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
4524 if (key != NULL && unlikely(strcmp((char *)key, dotdot) == 0)) {
4525 /* '..' to a remote object has a local representative */
4526 credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
4527 /* to reset LMAI_REMOTE_PARENT */
4530 osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
4532 inode = osd_dt_obj(dt)->oo_inode;
4536 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
4537 i_projid_read(inode), 0, oh, osd_dt_obj(dt),
4542 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
4545 struct osd_fid_pack *rec;
4548 if (de->file_type & LDISKFS_DIRENT_LUFID) {
4549 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
4550 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
4551 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
4557 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
4558 const struct lu_fid *fid)
4560 struct seq_server_site *ss = osd_seq_site(osd);
4563 /* FID seqs not in FLDB, must be local seq */
4564 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
4567 /* If FLD is not being initialized yet, it only happens during the
4568 * initialization, likely during mgs initialization, and we assume
4569 * this is local FID. */
4570 if (ss == NULL || ss->ss_server_fld == NULL)
4573 /* Only check the local FLDB here */
4574 if (osd_seq_exists(env, osd, fid_seq(fid)))
4581 * Index delete function for interoperability mode (b11826).
4582 * It will remove the directory entry added by osd_index_ea_insert().
4583 * This entry is needed to maintain name->fid mapping.
4585 * \param key, key i.e. file entry to be deleted
4587 * \retval 0, on success
4588 * \retval -ve, on error
4590 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
4591 const struct dt_key *key, struct thandle *handle)
4593 struct osd_object *obj = osd_dt_obj(dt);
4594 struct inode *dir = obj->oo_inode;
4595 struct dentry *dentry;
4596 struct osd_thandle *oh;
4597 struct ldiskfs_dir_entry_2 *de = NULL;
4598 struct buffer_head *bh;
4599 struct htree_lock *hlock = NULL;
4600 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
4601 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4605 if (!dt_object_exists(dt))
4608 LINVRNT(osd_invariant(obj));
4609 LASSERT(!dt_object_remote(dt));
4610 LASSERT(handle != NULL);
4612 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4614 oh = container_of(handle, struct osd_thandle, ot_super);
4615 LASSERT(oh->ot_handle != NULL);
4616 LASSERT(oh->ot_handle->h_transaction != NULL);
4618 ll_vfs_dq_init(dir);
4619 dentry = osd_child_dentry_get(env, obj,
4620 (char *)key, strlen((char *)key));
4622 if (obj->oo_hl_head != NULL) {
4623 hlock = osd_oti_get(env)->oti_hlock;
4624 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4625 dir, LDISKFS_HLOCK_DEL);
4627 down_write(&obj->oo_ext_idx_sem);
4630 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4632 /* If this is not the ".." entry, it might be a remote DNE
4633 * entry and we need to check if the FID is for a remote
4634 * MDT. If the FID is not in the directory entry (e.g.
4635 * upgraded 1.8 filesystem without dirdata enabled) then
4636 * we need to get the FID from the LMA. For a remote directory
4637 * there HAS to be an LMA, it cannot be an IGIF inode in this
4640 * Delete the entry before the agent inode in order to
4641 * simplify error handling. At worst an error after deleting
4642 * the entry first might leak the agent inode afterward. The
4643 * reverse would need filesystem abort in case of error deleting
4644 * the entry after the agent had been removed, or leave a
4645 * dangling entry pointing at a random inode. */
4646 if (strcmp((char *)key, dotdot) != 0) {
4647 LASSERT(de != NULL);
4648 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
4649 if (rc == -ENODATA) {
4650 /* can't get FID, postpone to the end of the
4651 * transaction when iget() is safe */
4652 osd_schedule_agent_inode_removal(env, oh,
4653 le32_to_cpu(de->inode));
4654 } else if (rc == 0 &&
4655 unlikely(osd_remote_fid(env, osd, fid))) {
4656 osd_schedule_agent_inode_removal(env, oh,
4657 le32_to_cpu(de->inode));
4660 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
4666 ldiskfs_htree_unlock(hlock);
4668 up_write(&obj->oo_ext_idx_sem);
4673 /* For inode on the remote MDT, .. will point to
4674 * /Agent directory, Check whether it needs to delete
4675 * from agent directory */
4676 if (unlikely(strcmp((char *)key, dotdot) == 0)) {
4679 ret = osd_delete_from_remote_parent(env, osd_obj2dev(obj),
4682 /* Sigh, the entry has been deleted, and
4683 * it is not easy to revert it back, so
4684 * let's keep this error private, and let
4685 * LFSCK fix it. XXX */
4686 CERROR("%s: delete remote parent "DFID": rc = %d\n",
4687 osd_name(osd), PFID(fid), ret);
4690 LASSERT(osd_invariant(obj));
4691 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4696 * Lookup index for \a key and copy record to \a rec.
4698 * \param dt osd index object
4699 * \param key key for index
4700 * \param rec record reference
4702 * \retval +ve success : exact mach
4703 * \retval 0 return record with key not greater than \a key
4704 * \retval -ve failure
4706 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
4707 struct dt_rec *rec, const struct dt_key *key)
4709 struct osd_object *obj = osd_dt_obj(dt);
4710 struct iam_path_descr *ipd;
4711 struct iam_container *bag = &obj->oo_dir->od_container;
4712 struct osd_thread_info *oti = osd_oti_get(env);
4713 struct iam_iterator *it = &oti->oti_idx_it;
4714 struct iam_rec *iam_rec;
4718 if (!dt_object_exists(dt))
4721 LASSERT(osd_invariant(obj));
4722 LASSERT(!dt_object_remote(dt));
4723 LASSERT(bag->ic_object == obj->oo_inode);
4725 ipd = osd_idx_ipd_get(env, bag);
4729 /* got ipd now we can start iterator. */
4730 iam_it_init(it, bag, 0, ipd);
4732 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4733 /* swab quota uid/gid provided by caller */
4734 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4735 key = (const struct dt_key *)&oti->oti_quota_id;
4738 rc = iam_it_get(it, (struct iam_key *)key);
4740 if (S_ISDIR(obj->oo_inode->i_mode))
4741 iam_rec = (struct iam_rec *)oti->oti_ldp;
4743 iam_rec = (struct iam_rec *) rec;
4745 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
4747 if (S_ISDIR(obj->oo_inode->i_mode))
4748 osd_fid_unpack((struct lu_fid *) rec,
4749 (struct osd_fid_pack *)iam_rec);
4750 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
4751 osd_quota_unpack(obj, rec);
4756 osd_ipd_put(env, bag, ipd);
4758 LINVRNT(osd_invariant(obj));
4763 static int osd_index_declare_iam_insert(const struct lu_env *env,
4764 struct dt_object *dt,
4765 const struct dt_rec *rec,
4766 const struct dt_key *key,
4767 struct thandle *handle)
4769 struct osd_thandle *oh;
4771 LASSERT(handle != NULL);
4773 oh = container_of0(handle, struct osd_thandle, ot_super);
4774 LASSERT(oh->ot_handle == NULL);
4776 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4777 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4783 * Inserts (key, value) pair in \a dt index object.
4785 * \param dt osd index object
4786 * \param key key for index
4787 * \param rec record reference
4788 * \param th transaction handler
4791 * \retval -ve failure
4793 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
4794 const struct dt_rec *rec,
4795 const struct dt_key *key, struct thandle *th,
4798 struct osd_object *obj = osd_dt_obj(dt);
4799 struct iam_path_descr *ipd;
4800 struct osd_thandle *oh;
4801 struct iam_container *bag;
4802 struct osd_thread_info *oti = osd_oti_get(env);
4803 struct iam_rec *iam_rec;
4807 if (!dt_object_exists(dt))
4810 LINVRNT(osd_invariant(obj));
4811 LASSERT(!dt_object_remote(dt));
4813 bag = &obj->oo_dir->od_container;
4814 LASSERT(bag->ic_object == obj->oo_inode);
4815 LASSERT(th != NULL);
4817 osd_trans_exec_op(env, th, OSD_OT_INSERT);
4819 ipd = osd_idx_ipd_get(env, bag);
4820 if (unlikely(ipd == NULL))
4823 oh = container_of0(th, struct osd_thandle, ot_super);
4824 LASSERT(oh->ot_handle != NULL);
4825 LASSERT(oh->ot_handle->h_transaction != NULL);
4826 if (S_ISDIR(obj->oo_inode->i_mode)) {
4827 iam_rec = (struct iam_rec *)oti->oti_ldp;
4828 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
4829 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4830 /* pack quota uid/gid */
4831 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4832 key = (const struct dt_key *)&oti->oti_quota_id;
4833 /* pack quota record */
4834 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
4835 iam_rec = (struct iam_rec *)rec;
4837 iam_rec = (struct iam_rec *)rec;
4840 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
4842 osd_ipd_put(env, bag, ipd);
4843 LINVRNT(osd_invariant(obj));
4844 osd_trans_exec_check(env, th, OSD_OT_INSERT);
4849 * Calls ldiskfs_add_entry() to add directory entry
4850 * into the directory. This is required for
4851 * interoperability mode (b11826)
4853 * \retval 0, on success
4854 * \retval -ve, on error
4856 static int __osd_ea_add_rec(struct osd_thread_info *info,
4857 struct osd_object *pobj, struct inode *cinode,
4858 const char *name, const struct lu_fid *fid,
4859 struct htree_lock *hlock, struct thandle *th)
4861 struct ldiskfs_dentry_param *ldp;
4862 struct dentry *child;
4863 struct osd_thandle *oth;
4866 oth = container_of(th, struct osd_thandle, ot_super);
4867 LASSERT(oth->ot_handle != NULL);
4868 LASSERT(oth->ot_handle->h_transaction != NULL);
4869 LASSERT(pobj->oo_inode);
4871 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
4872 if (unlikely(pobj->oo_inode ==
4873 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
4876 osd_get_ldiskfs_dirent_param(ldp, fid);
4877 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
4878 child->d_fsdata = (void *)ldp;
4879 ll_vfs_dq_init(pobj->oo_inode);
4880 rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
4881 child, cinode, hlock);
4882 if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
4883 struct ldiskfs_dir_entry_2 *de;
4884 struct buffer_head *bh;
4887 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
4890 rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
4893 if (S_ISDIR(cinode->i_mode))
4894 de->file_type = LDISKFS_DIRENT_LUFID |
4895 LDISKFS_FT_REG_FILE;
4897 de->file_type = LDISKFS_DIRENT_LUFID |
4899 ldiskfs_handle_dirty_metadata(oth->ot_handle,
4910 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
4911 * into the directory.Also sets flags into osd object to
4912 * indicate dot and dotdot are created. This is required for
4913 * interoperability mode (b11826)
4915 * \param dir directory for dot and dotdot fixup.
4916 * \param obj child object for linking
4918 * \retval 0, on success
4919 * \retval -ve, on error
4921 static int osd_add_dot_dotdot(struct osd_thread_info *info,
4922 struct osd_object *dir,
4923 struct inode *parent_dir, const char *name,
4924 const struct lu_fid *dot_fid,
4925 const struct lu_fid *dot_dot_fid,
4928 struct inode *inode = dir->oo_inode;
4929 struct osd_thandle *oth;
4932 oth = container_of(th, struct osd_thandle, ot_super);
4933 LASSERT(oth->ot_handle->h_transaction != NULL);
4934 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4936 if (strcmp(name, dot) == 0) {
4937 if (dir->oo_compat_dot_created) {
4940 LASSERT(inode->i_ino == parent_dir->i_ino);
4941 dir->oo_compat_dot_created = 1;
4944 } else if (strcmp(name, dotdot) == 0) {
4945 if (!dir->oo_compat_dot_created)
4947 /* in case of rename, dotdot is already created */
4948 if (dir->oo_compat_dotdot_created) {
4949 return __osd_ea_add_rec(info, dir, parent_dir, name,
4950 dot_dot_fid, NULL, th);
4953 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
4954 struct lu_fid tfid = *dot_dot_fid;
4957 result = osd_add_dot_dotdot_internal(info,
4958 dir->oo_inode, parent_dir, dot_fid,
4961 result = osd_add_dot_dotdot_internal(info,
4962 dir->oo_inode, parent_dir, dot_fid,
4967 dir->oo_compat_dotdot_created = 1;
4975 * It will call the appropriate osd_add* function and return the
4976 * value, return by respective functions.
4978 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4979 struct inode *cinode, const char *name,
4980 const struct lu_fid *fid, struct thandle *th)
4982 struct osd_thread_info *info = osd_oti_get(env);
4983 struct htree_lock *hlock;
4986 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4988 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4990 if (hlock != NULL) {
4991 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4994 down_write(&pobj->oo_ext_idx_sem);
4997 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4998 lu_object_fid(&pobj->oo_dt.do_lu),
5001 if (hlock != NULL) {
5002 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5003 pobj->oo_inode, LDISKFS_HLOCK_ADD);
5005 down_write(&pobj->oo_ext_idx_sem);
5008 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5009 struct lu_fid *tfid = &info->oti_fid;
5013 rc = __osd_ea_add_rec(info, pobj, cinode, name,
5016 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5021 ldiskfs_htree_unlock(hlock);
5023 up_write(&pobj->oo_ext_idx_sem);
5029 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5030 struct osd_idmap_cache *oic)
5032 struct osd_scrub *scrub = &dev->od_scrub;
5033 struct lu_fid *fid = &oic->oic_fid;
5034 struct osd_inode_id *id = &oic->oic_lid;
5035 struct inode *inode = NULL;
5041 if (!fid_is_norm(fid) && !fid_is_igif(fid))
5044 if (scrub->os_pos_current > id->oii_ino)
5048 rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5049 if (rc == -ENOENT) {
5050 __u32 gen = id->oii_gen;
5056 inode = osd_iget(oti, dev, id);
5057 /* The inode has been removed (by race maybe). */
5058 if (IS_ERR(inode)) {
5059 rc = PTR_ERR(inode);
5061 RETURN(rc == -ESTALE ? -ENOENT : rc);
5064 /* The OI mapping is lost. */
5065 if (gen != OSD_OII_NOGEN)
5069 /* The inode may has been reused by others, we do not know,
5070 * leave it to be handled by subsequent osd_fid_lookup(). */
5072 } else if (rc != 0 || osd_id_eq(id, &oti->oti_id)) {
5079 if (thread_is_running(&scrub->os_thread)) {
5080 if (inode == NULL) {
5081 inode = osd_iget(oti, dev, id);
5082 /* The inode has been removed (by race maybe). */
5083 if (IS_ERR(inode)) {
5084 rc = PTR_ERR(inode);
5086 RETURN(rc == -ESTALE ? -ENOENT : rc);
5090 rc = osd_oii_insert(dev, oic, insert);
5091 /* There is race condition between osd_oi_lookup and OI scrub.
5092 * The OI scrub finished just after osd_oi_lookup() failure.
5093 * Under such case, it is unnecessary to trigger OI scrub again,
5094 * but try to call osd_oi_lookup() again. */
5095 if (unlikely(rc == -EAGAIN))
5098 if (!S_ISDIR(inode->i_mode))
5101 rc = osd_check_lmv(oti, dev, inode, oic);
5107 if (!dev->od_noscrub && ++once == 1) {
5108 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
5110 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5111 "%.16s: trigger partial OI scrub for RPC inconsistency "
5112 "checking FID "DFID": rc = %d\n",
5113 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
5115 if (rc == 0 || rc == -EALREADY)
5125 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5126 struct osd_device *dev,
5127 struct osd_idmap_cache *oic,
5128 struct lu_fid *fid, __u32 ino)
5130 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5131 struct inode *inode;
5134 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5135 inode = osd_iget(oti, dev, &oic->oic_lid);
5136 if (IS_ERR(inode)) {
5137 fid_zero(&oic->oic_fid);
5138 return PTR_ERR(inode);
5141 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5144 fid_zero(&oic->oic_fid);
5146 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5150 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5151 struct osd_inode_id *id, const struct lu_fid *fid)
5153 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5154 id->oii_ino, id->oii_gen, info);
5155 info->oti_cache.oic_lid = *id;
5156 info->oti_cache.oic_fid = *fid;
5157 info->oti_cache.oic_dev = osd;
5161 * Get parent FID from the linkEA.
5163 * For a directory which parent resides on remote MDT, to satisfy the
5164 * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5165 * the other hand, to make the lookup(..) on the directory can return
5166 * the real parent FID, we append the real parent FID after its ".."
5167 * name entry in the /REMOTE_PARENT_DIR.
5169 * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5170 * backup. So after the restore, we cannot get the right parent FID from
5171 * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5172 * we have stored the real parent FID in the directory object's linkEA,
5173 * we can parse the linkEA for the real parent FID.
5175 * \param[in] env pointer to the thread context
5176 * \param[in] obj pointer to the object to be handled
5177 * \param[out]fid pointer to the buffer to hold the parent FID
5179 * \retval 0 for getting the real parent FID successfully
5180 * \retval negative error number on failure
5182 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5183 struct osd_object *obj,
5186 struct osd_thread_info *oti = osd_oti_get(env);
5187 struct lu_buf *buf = &oti->oti_big_buf;
5188 struct dentry *dentry = &oti->oti_obj_dentry;
5189 struct inode *inode = obj->oo_inode;
5190 struct linkea_data ldata = { NULL };
5195 if (!S_ISDIR(inode->i_mode))
5199 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5200 buf->lb_buf, buf->lb_len);
5201 if (rc == -ERANGE) {
5202 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5205 lu_buf_realloc(buf, rc);
5206 if (buf->lb_buf == NULL)
5213 if (unlikely(rc == 0))
5219 if (unlikely(buf->lb_buf == NULL)) {
5220 lu_buf_realloc(buf, rc);
5221 if (buf->lb_buf == NULL)
5228 rc = linkea_init_with_rec(&ldata);
5230 linkea_first_entry(&ldata);
5231 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5237 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5238 struct inode *inode,
5239 const struct lu_fid *pfid,
5240 const char *name, const int namelen)
5242 struct osd_thread_info *oti = osd_oti_get(env);
5243 struct lu_buf *buf = &oti->oti_big_buf;
5244 struct dentry *dentry = &oti->oti_obj_dentry;
5245 struct linkea_data ldata = { NULL };
5246 struct lu_name cname = { .ln_name = name,
5247 .ln_namelen = namelen };
5252 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5253 buf->lb_buf, buf->lb_len);
5255 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5260 if (unlikely(rc == 0))
5263 if (buf->lb_len < rc) {
5264 lu_buf_realloc(buf, rc);
5265 if (buf->lb_buf == NULL)
5272 rc = linkea_init_with_rec(&ldata);
5274 rc = linkea_links_find(&ldata, &cname, pfid);
5280 * Calls ->lookup() to find dentry. From dentry get inode and
5281 * read inode's ea to get fid. This is required for interoperability
5284 * \retval 0, on success
5285 * \retval -ve, on error
5287 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5288 struct dt_rec *rec, const struct dt_key *key)
5290 struct inode *dir = obj->oo_inode;
5291 struct dentry *dentry;
5292 struct ldiskfs_dir_entry_2 *de;
5293 struct buffer_head *bh;
5294 struct lu_fid *fid = (struct lu_fid *) rec;
5295 struct htree_lock *hlock = NULL;
5300 LASSERT(dir->i_op != NULL);
5301 LASSERT(dir->i_op->lookup != NULL);
5303 dentry = osd_child_dentry_get(env, obj,
5304 (char *)key, strlen((char *)key));
5306 if (obj->oo_hl_head != NULL) {
5307 hlock = osd_oti_get(env)->oti_hlock;
5308 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5309 dir, LDISKFS_HLOCK_LOOKUP);
5311 down_read(&obj->oo_ext_idx_sem);
5314 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5316 struct osd_thread_info *oti = osd_oti_get(env);
5317 struct osd_inode_id *id = &oti->oti_id;
5318 struct osd_idmap_cache *oic = &oti->oti_cache;
5319 struct osd_device *dev = osd_obj2dev(obj);
5321 ino = le32_to_cpu(de->inode);
5322 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5324 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5328 rc = osd_get_fid_from_dentry(de, rec);
5330 /* done with de, release bh */
5333 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5334 const char *name = (const char *)key;
5336 /* If the parent is on remote MDT, and there
5337 * is no FID-in-dirent, then we have to get
5338 * the parent FID from the linkEA. */
5339 if (likely(strlen(name) == 2 &&
5340 name[0] == '.' && name[1] == '.'))
5341 rc = osd_get_pfid_from_linkea(env, obj,
5344 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5347 osd_id_gen(id, ino, OSD_OII_NOGEN);
5350 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5351 fid_zero(&oic->oic_fid);
5356 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5357 rc = osd_consistency_check(oti, dev, oic);
5359 fid_zero(&oic->oic_fid);
5368 ldiskfs_htree_unlock(hlock);
5370 up_read(&obj->oo_ext_idx_sem);
5374 static int osd_index_declare_ea_insert(const struct lu_env *env,
5375 struct dt_object *dt,
5376 const struct dt_rec *rec,
5377 const struct dt_key *key,
5378 struct thandle *handle)
5380 struct osd_thandle *oh;
5381 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5382 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5383 const struct lu_fid *fid = rec1->rec_fid;
5384 int credits, rc = 0;
5385 struct osd_idmap_cache *idc;
5388 LASSERT(!dt_object_remote(dt));
5389 LASSERT(handle != NULL);
5390 LASSERT(fid != NULL);
5391 LASSERT(rec1->rec_type != 0);
5393 oh = container_of0(handle, struct osd_thandle, ot_super);
5394 LASSERT(oh->ot_handle == NULL);
5396 credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5398 /* we can't call iget() while a transactions is running
5399 * (this can lead to a deadlock), but we need to know
5400 * inum and object type. so we find this information at
5401 * declaration and cache in per-thread info */
5402 idc = osd_idc_find_or_init(env, osd, fid);
5404 RETURN(PTR_ERR(idc));
5405 if (idc->oic_remote) {
5406 /* a reference to remote inode is represented by an
5407 * agent inode which we have to create */
5408 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5409 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5412 osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5414 if (osd_dt_obj(dt)->oo_inode != NULL) {
5415 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5417 /* We ignore block quota on meta pool (MDTs), so needn't
5418 * calculate how many blocks will be consumed by this index
5420 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5422 i_projid_read(inode), 0,
5423 oh, osd_dt_obj(dt), true,
5431 * Index add function for interoperability mode (b11826).
5432 * It will add the directory entry.This entry is needed to
5433 * maintain name->fid mapping.
5435 * \param key it is key i.e. file entry to be inserted
5436 * \param rec it is value of given key i.e. fid
5438 * \retval 0, on success
5439 * \retval -ve, on error
5441 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5442 const struct dt_rec *rec,
5443 const struct dt_key *key, struct thandle *th,
5446 struct osd_object *obj = osd_dt_obj(dt);
5447 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5448 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5449 const struct lu_fid *fid = rec1->rec_fid;
5450 const char *name = (const char *)key;
5451 struct osd_thread_info *oti = osd_oti_get(env);
5452 struct inode *child_inode = NULL;
5453 struct osd_idmap_cache *idc;
5457 if (!dt_object_exists(dt))
5460 LASSERT(osd_invariant(obj));
5461 LASSERT(!dt_object_remote(dt));
5462 LASSERT(th != NULL);
5464 osd_trans_exec_op(env, th, OSD_OT_INSERT);
5466 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
5468 idc = osd_idc_find(env, osd, fid);
5469 if (unlikely(idc == NULL)) {
5470 /* this dt_insert() wasn't declared properly, so
5471 * FID is missing in OI cache. we better do not
5472 * lookup FID in FLDB/OI and don't risk to deadlock,
5473 * but in some special cases (lfsck testing, etc)
5474 * it's much simpler than fixing a caller */
5475 CERROR("%s: "DFID" wasn't declared for insert\n",
5476 osd_name(osd), PFID(fid));
5478 idc = osd_idc_find_or_init(env, osd, fid);
5480 RETURN(PTR_ERR(idc));
5483 if (idc->oic_remote) {
5484 /* Insert remote entry */
5485 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
5486 struct osd_mdobj_map *omm = osd->od_mdt_map;
5487 struct osd_thandle *oh;
5489 /* If parent on remote MDT, we need put this object
5491 oh = container_of(th, typeof(*oh), ot_super);
5492 rc = osd_add_to_remote_parent(env, osd, obj, oh);
5494 CERROR("%s: add "DFID" error: rc = %d\n",
5496 PFID(lu_object_fid(&dt->do_lu)), rc);
5500 child_inode = igrab(omm->omm_remote_parent->d_inode);
5502 child_inode = osd_create_local_agent_inode(env, osd,
5503 obj, fid, rec1->rec_type & S_IFMT, th);
5504 if (IS_ERR(child_inode))
5505 RETURN(PTR_ERR(child_inode));
5508 /* Insert local entry */
5509 if (unlikely(idc->oic_lid.oii_ino == 0)) {
5510 /* for a reason OI cache wasn't filled properly */
5511 CERROR("%s: OIC for "DFID" isn't filled\n",
5512 osd_name(osd), PFID(fid));
5515 child_inode = oti->oti_inode;
5516 if (unlikely(child_inode == NULL)) {
5517 struct ldiskfs_inode_info *lii;
5521 child_inode = oti->oti_inode = &lii->vfs_inode;
5523 child_inode->i_sb = osd_sb(osd);
5524 child_inode->i_ino = idc->oic_lid.oii_ino;
5525 child_inode->i_mode = rec1->rec_type & S_IFMT;
5528 rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
5530 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
5531 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
5533 if (child_inode && child_inode != oti->oti_inode)
5535 LASSERT(osd_invariant(obj));
5536 osd_trans_exec_check(env, th, OSD_OT_INSERT);
5541 * Initialize osd Iterator for given osd index object.
5543 * \param dt osd index object
5546 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
5547 struct dt_object *dt,
5550 struct osd_it_iam *it;
5551 struct osd_object *obj = osd_dt_obj(dt);
5552 struct lu_object *lo = &dt->do_lu;
5553 struct iam_path_descr *ipd;
5554 struct iam_container *bag = &obj->oo_dir->od_container;
5556 if (!dt_object_exists(dt))
5557 return ERR_PTR(-ENOENT);
5561 return ERR_PTR(-ENOMEM);
5563 ipd = osd_it_ipd_get(env, bag);
5564 if (likely(ipd != NULL)) {
5568 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
5569 return (struct dt_it *)it;
5572 return ERR_PTR(-ENOMEM);
5577 * free given Iterator.
5579 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
5581 struct osd_it_iam *it = (struct osd_it_iam *)di;
5582 struct osd_object *obj = it->oi_obj;
5584 iam_it_fini(&it->oi_it);
5585 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
5586 osd_object_put(env, obj);
5591 * Move Iterator to record specified by \a key
5593 * \param di osd iterator
5594 * \param key key for index
5596 * \retval +ve di points to record with least key not larger than key
5597 * \retval 0 di points to exact matched key
5598 * \retval -ve failure
5601 static int osd_it_iam_get(const struct lu_env *env,
5602 struct dt_it *di, const struct dt_key *key)
5604 struct osd_thread_info *oti = osd_oti_get(env);
5605 struct osd_it_iam *it = (struct osd_it_iam *)di;
5607 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5608 /* swab quota uid/gid */
5609 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5610 key = (struct dt_key *)&oti->oti_quota_id;
5613 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
5619 * \param di osd iterator
5621 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
5623 struct osd_it_iam *it = (struct osd_it_iam *)di;
5625 iam_it_put(&it->oi_it);
5629 * Move iterator by one record
5631 * \param di osd iterator
5633 * \retval +1 end of container reached
5635 * \retval -ve failure
5638 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
5640 struct osd_it_iam *it = (struct osd_it_iam *)di;
5642 return iam_it_next(&it->oi_it);
5646 * Return pointer to the key under iterator.
5649 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
5650 const struct dt_it *di)
5652 struct osd_thread_info *oti = osd_oti_get(env);
5653 struct osd_it_iam *it = (struct osd_it_iam *)di;
5654 struct osd_object *obj = it->oi_obj;
5657 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
5659 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
5660 /* swab quota uid/gid */
5661 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
5662 key = (struct dt_key *)&oti->oti_quota_id;
5669 * Return size of key under iterator (in bytes)
5672 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
5674 struct osd_it_iam *it = (struct osd_it_iam *)di;
5676 return iam_it_key_size(&it->oi_it);
5680 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
5682 /* check if file type is required */
5683 if (ent->lde_attrs & LUDA_TYPE) {
5684 struct luda_type *lt;
5685 int align = sizeof(*lt) - 1;
5687 len = (len + align) & ~align;
5688 lt = (struct luda_type *)(ent->lde_name + len);
5689 lt->lt_type = cpu_to_le16(DTTOIF(type));
5692 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
5696 * build lu direct from backend fs dirent.
5700 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
5701 char *name, __u16 namelen, __u16 type, __u32 attr)
5703 ent->lde_attrs = attr | LUDA_FID;
5704 fid_cpu_to_le(&ent->lde_fid, fid);
5706 ent->lde_hash = cpu_to_le64(offset);
5707 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
5709 strncpy(ent->lde_name, name, namelen);
5710 ent->lde_name[namelen] = '\0';
5711 ent->lde_namelen = cpu_to_le16(namelen);
5713 /* append lustre attributes */
5714 osd_it_append_attrs(ent, namelen, type);
5718 * Return pointer to the record under iterator.
5720 static int osd_it_iam_rec(const struct lu_env *env,
5721 const struct dt_it *di,
5722 struct dt_rec *dtrec, __u32 attr)
5724 struct osd_it_iam *it = (struct osd_it_iam *)di;
5725 struct osd_thread_info *info = osd_oti_get(env);
5728 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
5729 const struct osd_fid_pack *rec;
5730 struct lu_fid *fid = &info->oti_fid;
5731 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
5737 name = (char *)iam_it_key_get(&it->oi_it);
5739 RETURN(PTR_ERR(name));
5741 namelen = iam_it_key_size(&it->oi_it);
5743 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
5745 RETURN(PTR_ERR(rec));
5747 rc = osd_fid_unpack(fid, rec);
5751 hash = iam_it_store(&it->oi_it);
5753 /* IAM does not store object type in IAM index (dir) */
5754 osd_it_pack_dirent(lde, fid, hash, name, namelen,
5756 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5757 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5758 (struct iam_rec *)dtrec);
5759 osd_quota_unpack(it->oi_obj, dtrec);
5761 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5762 (struct iam_rec *)dtrec);
5769 * Returns cookie for current Iterator position.
5771 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
5773 struct osd_it_iam *it = (struct osd_it_iam *)di;
5775 return iam_it_store(&it->oi_it);
5779 * Restore iterator from cookie.
5781 * \param di osd iterator
5782 * \param hash Iterator location cookie
5784 * \retval +ve di points to record with least key not larger than key.
5785 * \retval 0 di points to exact matched key
5786 * \retval -ve failure
5789 static int osd_it_iam_load(const struct lu_env *env,
5790 const struct dt_it *di, __u64 hash)
5792 struct osd_it_iam *it = (struct osd_it_iam *)di;
5794 return iam_it_load(&it->oi_it, hash);
5797 static const struct dt_index_operations osd_index_iam_ops = {
5798 .dio_lookup = osd_index_iam_lookup,
5799 .dio_declare_insert = osd_index_declare_iam_insert,
5800 .dio_insert = osd_index_iam_insert,
5801 .dio_declare_delete = osd_index_declare_iam_delete,
5802 .dio_delete = osd_index_iam_delete,
5804 .init = osd_it_iam_init,
5805 .fini = osd_it_iam_fini,
5806 .get = osd_it_iam_get,
5807 .put = osd_it_iam_put,
5808 .next = osd_it_iam_next,
5809 .key = osd_it_iam_key,
5810 .key_size = osd_it_iam_key_size,
5811 .rec = osd_it_iam_rec,
5812 .store = osd_it_iam_store,
5813 .load = osd_it_iam_load
5819 * Creates or initializes iterator context.
5821 * \retval struct osd_it_ea, iterator structure on success
5824 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
5825 struct dt_object *dt,
5828 struct osd_object *obj = osd_dt_obj(dt);
5829 struct osd_thread_info *info = osd_oti_get(env);
5830 struct osd_it_ea *oie;
5832 struct lu_object *lo = &dt->do_lu;
5833 struct dentry *obj_dentry;
5836 if (!dt_object_exists(dt) || obj->oo_destroyed)
5837 RETURN(ERR_PTR(-ENOENT));
5839 OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
5841 RETURN(ERR_PTR(-ENOMEM));
5842 obj_dentry = &oie->oie_dentry;
5844 obj_dentry->d_inode = obj->oo_inode;
5845 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
5846 obj_dentry->d_name.hash = 0;
5848 oie->oie_rd_dirent = 0;
5849 oie->oie_it_dirent = 0;
5850 oie->oie_dirent = NULL;
5851 if (unlikely(!info->oti_it_ea_buf_used)) {
5852 oie->oie_buf = info->oti_it_ea_buf;
5853 info->oti_it_ea_buf_used = 1;
5855 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5856 if (oie->oie_buf == NULL)
5857 RETURN(ERR_PTR(-ENOMEM));
5861 file = &oie->oie_file;
5863 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
5864 if (attr & LUDA_64BITHASH)
5865 file->f_mode = FMODE_64BITHASH;
5867 file->f_mode = FMODE_32BITHASH;
5868 file->f_path.dentry = obj_dentry;
5869 file->f_mapping = obj->oo_inode->i_mapping;
5870 file->f_op = obj->oo_inode->i_fop;
5871 set_file_inode(file, obj->oo_inode);
5874 RETURN((struct dt_it *) oie);
5878 * Destroy or finishes iterator context.
5880 * \param di iterator structure to be destroyed
5882 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
5884 struct osd_thread_info *info = osd_oti_get(env);
5885 struct osd_it_ea *oie = (struct osd_it_ea *)di;
5886 struct osd_object *obj = oie->oie_obj;
5887 struct inode *inode = obj->oo_inode;
5890 oie->oie_file.f_op->release(inode, &oie->oie_file);
5891 osd_object_put(env, obj);
5892 if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
5893 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5895 info->oti_it_ea_buf_used = 0;
5896 OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
5901 * It position the iterator at given key, so that next lookup continues from
5902 * that key Or it is similar to dio_it->load() but based on a key,
5903 * rather than file position.
5905 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
5908 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
5910 static int osd_it_ea_get(const struct lu_env *env,
5911 struct dt_it *di, const struct dt_key *key)
5913 struct osd_it_ea *it = (struct osd_it_ea *)di;
5916 LASSERT(((const char *)key)[0] == '\0');
5917 it->oie_file.f_pos = 0;
5918 it->oie_rd_dirent = 0;
5919 it->oie_it_dirent = 0;
5920 it->oie_dirent = NULL;
5928 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
5932 struct osd_filldir_cbs {
5933 #ifdef HAVE_DIR_CONTEXT
5934 struct dir_context ctx;
5936 struct osd_it_ea *it;
5939 * It is called internally by ->readdir(). It fills the
5940 * iterator's in-memory data structure with required
5941 * information i.e. name, namelen, rec_size etc.
5943 * \param buf in which information to be filled in.
5944 * \param name name of the file in given dir
5946 * \retval 0 on success
5947 * \retval 1 on buffer full
5949 #ifdef HAVE_FILLDIR_USE_CTX
5950 static int osd_ldiskfs_filldir(struct dir_context *buf,
5952 static int osd_ldiskfs_filldir(void *buf,
5954 const char *name, int namelen,
5955 loff_t offset, __u64 ino, unsigned d_type)
5957 struct osd_it_ea *it =
5958 ((struct osd_filldir_cbs *)buf)->it;
5959 struct osd_object *obj = it->oie_obj;
5960 struct osd_it_ea_dirent *ent = it->oie_dirent;
5961 struct lu_fid *fid = &ent->oied_fid;
5962 struct osd_fid_pack *rec;
5965 /* this should never happen */
5966 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5967 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5971 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
5975 /* "." is just the object itself. */
5976 if (namelen == 1 && name[0] == '.') {
5977 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5978 } else if (d_type & LDISKFS_DIRENT_LUFID) {
5979 rec = (struct osd_fid_pack*) (name + namelen + 1);
5980 if (osd_fid_unpack(fid, rec) != 0)
5985 d_type &= ~LDISKFS_DIRENT_LUFID;
5987 /* NOT export local root. */
5988 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5989 ino = obj->oo_inode->i_ino;
5990 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5993 ent->oied_ino = ino;
5994 ent->oied_off = offset;
5995 ent->oied_namelen = namelen;
5996 ent->oied_type = d_type;
5998 memcpy(ent->oied_name, name, namelen);
6000 it->oie_rd_dirent++;
6001 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
6006 * Calls ->readdir() to load a directory entry at a time
6007 * and stored it in iterator's in-memory data structure.
6009 * \param di iterator's in memory structure
6011 * \retval 0 on success
6012 * \retval -ve on error
6013 * \retval +1 reach the end of entry
6015 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6016 const struct dt_it *di)
6018 struct osd_it_ea *it = (struct osd_it_ea *)di;
6019 struct osd_object *obj = it->oie_obj;
6020 struct inode *inode = obj->oo_inode;
6021 struct htree_lock *hlock = NULL;
6022 struct file *filp = &it->oie_file;
6024 struct osd_filldir_cbs buf = {
6025 #ifdef HAVE_DIR_CONTEXT
6026 .ctx.actor = osd_ldiskfs_filldir,
6032 it->oie_dirent = it->oie_buf;
6033 it->oie_rd_dirent = 0;
6035 if (obj->oo_hl_head != NULL) {
6036 hlock = osd_oti_get(env)->oti_hlock;
6037 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6038 inode, LDISKFS_HLOCK_READDIR);
6040 down_read(&obj->oo_ext_idx_sem);
6043 #ifdef HAVE_DIR_CONTEXT
6044 buf.ctx.pos = filp->f_pos;
6045 rc = inode->i_fop->iterate(filp, &buf.ctx);
6046 filp->f_pos = buf.ctx.pos;
6048 rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
6052 ldiskfs_htree_unlock(hlock);
6054 up_read(&obj->oo_ext_idx_sem);
6056 if (it->oie_rd_dirent == 0) {
6057 /*If it does not get any dirent, it means it has been reached
6058 *to the end of the dir */
6059 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6063 it->oie_dirent = it->oie_buf;
6064 it->oie_it_dirent = 1;
6071 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6072 * to load a directory entry at a time and stored it in
6073 * iterator's in-memory data structure.
6075 * \param di iterator's in memory structure
6077 * \retval +ve iterator reached to end
6078 * \retval 0 iterator not reached to end
6079 * \retval -ve on error
6081 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6083 struct osd_it_ea *it = (struct osd_it_ea *)di;
6088 if (it->oie_it_dirent < it->oie_rd_dirent) {
6090 (void *) it->oie_dirent +
6091 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6092 it->oie_dirent->oied_namelen);
6093 it->oie_it_dirent++;
6096 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6099 rc = osd_ldiskfs_it_fill(env, di);
6106 * Returns the key at current position from iterator's in memory structure.
6108 * \param di iterator's in memory structure
6110 * \retval key i.e. struct dt_key on success
6112 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6113 const struct dt_it *di)
6115 struct osd_it_ea *it = (struct osd_it_ea *)di;
6117 return (struct dt_key *)it->oie_dirent->oied_name;
6121 * Returns the key's size at current position from iterator's in memory structure.
6123 * \param di iterator's in memory structure
6125 * \retval key_size i.e. struct dt_key on success
6127 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6129 struct osd_it_ea *it = (struct osd_it_ea *)di;
6131 return it->oie_dirent->oied_namelen;
6134 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6136 if (LDISKFS_DIR_REC_LEN(de) >=
6137 __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6144 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6145 unsigned blocksize, bool dotdot)
6148 return osd_dotdot_has_space(de);
6150 if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6151 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6158 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6159 handle_t *jh, struct dentry *dentry,
6160 const struct lu_fid *fid, struct buffer_head *bh,
6161 struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6164 struct inode *dir = dentry->d_parent->d_inode;
6165 struct inode *inode = dentry->d_inode;
6166 struct osd_fid_pack *rec;
6167 struct ldiskfs_dentry_param *ldp;
6168 int namelen = dentry->d_name.len;
6170 struct osd_thread_info *info = osd_oti_get(env);
6173 if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
6174 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6177 /* There is enough space to hold the FID-in-dirent. */
6178 if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6179 rc = ldiskfs_journal_get_write_access(jh, bh);
6183 de->name[namelen] = 0;
6184 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6185 rec->fp_len = sizeof(struct lu_fid) + 1;
6186 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6187 de->file_type |= LDISKFS_DIRENT_LUFID;
6188 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6195 rc = ldiskfs_delete_entry(jh, dir, de, bh);
6199 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6200 osd_get_ldiskfs_dirent_param(ldp, fid);
6201 dentry->d_fsdata = (void *)ldp;
6202 ll_vfs_dq_init(dir);
6203 rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6204 /* It is too bad, we cannot reinsert the name entry back.
6205 * That means we lose it! */
6207 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
6208 "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6209 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
6210 dir->i_ino, dir->i_generation, namelen,
6211 dentry->d_name.name, PFID(fid), rc);
6217 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6218 struct osd_it_ea *it, struct lu_fid *fid,
6219 struct osd_inode_id *id, __u32 *attr)
6221 struct osd_thread_info *info = osd_oti_get(env);
6222 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6223 struct osd_device *dev = osd_obj2dev(obj);
6224 struct super_block *sb = osd_sb(dev);
6225 const char *devname = osd_name(dev);
6226 struct osd_it_ea_dirent *ent = it->oie_dirent;
6227 struct inode *dir = obj->oo_inode;
6228 struct htree_lock *hlock = NULL;
6229 struct buffer_head *bh = NULL;
6230 handle_t *jh = NULL;
6231 struct ldiskfs_dir_entry_2 *de;
6232 struct dentry *dentry;
6233 struct inode *inode;
6234 const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6237 bool dotdot = false;
6241 if (ent->oied_name[0] == '.') {
6242 if (ent->oied_namelen == 1)
6245 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6249 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6250 inode = osd_iget(info, dev, id);
6251 if (IS_ERR(inode)) {
6252 rc = PTR_ERR(inode);
6253 if (rc == -ENOENT || rc == -ESTALE) {
6254 /* Maybe dangling name entry, or
6255 * corrupted directory entry. */
6256 *attr |= LUDA_UNKNOWN;
6259 CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6260 "check_repair, dir = %lu/%u, name = %.*s, "
6261 "ino = %llu, rc = %d\n",
6262 devname, dir->i_ino, dir->i_generation,
6263 ent->oied_namelen, ent->oied_name,
6270 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6272 rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6273 if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6278 /* We need to ensure that the name entry is still valid.
6279 * Because it may be removed or renamed by other already.
6281 * The unlink or rename operation will start journal before PDO lock,
6282 * so to avoid deadlock, here we need to start journal handle before
6283 * related PDO lock also. But because we do not know whether there
6284 * will be something to be repaired before PDO lock, we just start
6285 * journal without conditions.
6287 * We may need to remove the name entry firstly, then insert back.
6288 * One credit is for user quota file update.
6289 * One credit is for group quota file update.
6290 * Two credits are for dirty inode. */
6291 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6292 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6294 if (dev->od_dirent_journal != 0) {
6297 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6300 CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6301 "check_repair, dir = %lu/%u, credits = %d, "
6302 "name = %.*s, ino = %llu: rc = %d\n",
6303 devname, dir->i_ino, dir->i_generation, credits,
6304 ent->oied_namelen, ent->oied_name,
6307 GOTO(out_inode, rc);
6310 if (obj->oo_hl_head != NULL) {
6311 hlock = osd_oti_get(env)->oti_hlock;
6312 /* "0" means exclusive lock for the whole directory.
6313 * We need to prevent others access such name entry
6314 * during the delete + insert. Neither HLOCK_ADD nor
6315 * HLOCK_DEL cannot guarantee the atomicity. */
6316 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6318 down_write(&obj->oo_ext_idx_sem);
6321 if (obj->oo_hl_head != NULL) {
6322 hlock = osd_oti_get(env)->oti_hlock;
6323 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6324 LDISKFS_HLOCK_LOOKUP);
6326 down_read(&obj->oo_ext_idx_sem);
6330 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6331 if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6332 *attr |= LUDA_IGNORE;
6337 /* For dotdot entry, if there is not enough space to hold the
6338 * FID-in-dirent, just keep them there. It only happens when the
6339 * device upgraded from 1.8 or restored from MDT file-level backup.
6340 * For the whole directory, only dotdot entry have no FID-in-dirent
6341 * and needs to get FID from LMA when readdir, it will not affect the
6342 * performance much. */
6343 if (dotdot && !osd_dotdot_has_space(de)) {
6344 *attr |= LUDA_UNKNOWN;
6350 if (lu_fid_eq(fid, &lma->lma_self_fid))
6353 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6354 struct lu_fid *tfid = &lma->lma_self_fid;
6356 if (likely(dotdot &&
6357 fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6358 fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6359 /* It must be REMOTE_PARENT_DIR and as the
6360 * 'dotdot' entry of remote directory */
6361 *attr |= LUDA_IGNORE;
6363 CDEBUG(D_LFSCK, "%s: expect remote agent "
6364 "parent directory, but got %.*s under "
6365 "dir = %lu/%u with the FID "DFID"\n",
6366 devname, ent->oied_namelen,
6367 ent->oied_name, dir->i_ino,
6368 dir->i_generation, PFID(tfid));
6370 *attr |= LUDA_UNKNOWN;
6377 if (!fid_is_zero(fid)) {
6378 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6380 if (rc == -ENOENT ||
6382 !(dev->od_scrub.os_file.sf_flags & SF_UPGRADE))) {
6383 /* linkEA does not recognize the dirent entry,
6384 * it may because the dirent entry corruption
6385 * and points to other's inode. */
6386 CDEBUG(D_LFSCK, "%s: the target inode does not "
6387 "recognize the dirent, dir = %lu/%u, "
6388 " name = %.*s, ino = %llu, "
6389 DFID": rc = %d\n", devname, dir->i_ino,
6390 dir->i_generation, ent->oied_namelen,
6391 ent->oied_name, ent->oied_ino, PFID(fid), rc);
6392 *attr |= LUDA_UNKNOWN;
6397 if (rc && rc != -ENODATA) {
6398 CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6399 "dir = %lu/%u, name = %.*s, ino = %llu, "
6400 DFID": rc = %d\n", devname, dir->i_ino,
6401 dir->i_generation, ent->oied_namelen,
6402 ent->oied_name, ent->oied_ino, PFID(fid), rc);
6403 *attr |= LUDA_UNKNOWN;
6410 /* linkEA recognizes the dirent entry, the FID-in-LMA is
6411 * valid, trusted, in spite of fid_is_sane(fid) or not. */
6412 if (*attr & LUDA_VERIFY_DRYRUN) {
6413 *fid = lma->lma_self_fid;
6414 *attr |= LUDA_REPAIR;
6421 dev->od_dirent_journal = 1;
6422 if (hlock != NULL) {
6423 ldiskfs_htree_unlock(hlock);
6426 up_read(&obj->oo_ext_idx_sem);
6432 *fid = lma->lma_self_fid;
6434 /* Update or append the FID-in-dirent. */
6435 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6436 bh, de, hlock, dotdot);
6438 *attr |= LUDA_REPAIR;
6440 CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
6441 "the dirent, dir = %lu/%u, name = %.*s, "
6442 "ino = %llu, "DFID": rc = %d\n",
6443 devname, dir->i_ino, dir->i_generation,
6444 ent->oied_namelen, ent->oied_name,
6445 ent->oied_ino, PFID(fid), rc);
6447 /* lma is NULL, trust the FID-in-dirent if it is valid. */
6448 if (*attr & LUDA_VERIFY_DRYRUN) {
6449 if (fid_is_sane(fid)) {
6450 *attr |= LUDA_REPAIR;
6451 } else if (dev->od_index == 0) {
6452 lu_igif_build(fid, inode->i_ino,
6453 inode->i_generation);
6454 *attr |= LUDA_UPGRADE;
6462 dev->od_dirent_journal = 1;
6463 if (hlock != NULL) {
6464 ldiskfs_htree_unlock(hlock);
6467 up_read(&obj->oo_ext_idx_sem);
6474 if (unlikely(fid_is_sane(fid))) {
6475 /* FID-in-dirent exists, but FID-in-LMA is lost.
6476 * Trust the FID-in-dirent, and add FID-in-LMA. */
6477 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
6479 *attr |= LUDA_REPAIR;
6481 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
6482 "update dirent, dir = %lu/%u, "
6483 "name = %.*s, ino = %llu, "
6485 devname, dir->i_ino, dir->i_generation,
6486 ent->oied_namelen, ent->oied_name,
6487 ent->oied_ino, PFID(fid), rc);
6488 } else if (dev->od_index == 0) {
6489 lu_igif_build(fid, inode->i_ino, inode->i_generation);
6490 /* It is probably IGIF object. Only aappend the
6491 * FID-in-dirent. OI scrub will process FID-in-LMA. */
6492 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6493 bh, de, hlock, dotdot);
6495 *attr |= LUDA_UPGRADE;
6497 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
6498 "after the dirent, dir = %lu/%u, "
6499 "name = %.*s, ino = %llu, "
6501 devname, dir->i_ino, dir->i_generation,
6502 ent->oied_namelen, ent->oied_name,
6503 ent->oied_ino, PFID(fid), rc);
6512 if (hlock != NULL) {
6513 ldiskfs_htree_unlock(hlock);
6515 if (dev->od_dirent_journal != 0)
6516 up_write(&obj->oo_ext_idx_sem);
6518 up_read(&obj->oo_ext_idx_sem);
6522 ldiskfs_journal_stop(jh);
6526 if (rc >= 0 && !dirty)
6527 dev->od_dirent_journal = 0;
6533 * Returns the value at current position from iterator's in memory structure.
6535 * \param di struct osd_it_ea, iterator's in memory structure
6536 * \param attr attr requested for dirent.
6537 * \param lde lustre dirent
6539 * \retval 0 no error and \param lde has correct lustre dirent.
6540 * \retval -ve on error
6542 static inline int osd_it_ea_rec(const struct lu_env *env,
6543 const struct dt_it *di,
6544 struct dt_rec *dtrec, __u32 attr)
6546 struct osd_it_ea *it = (struct osd_it_ea *)di;
6547 struct osd_object *obj = it->oie_obj;
6548 struct osd_device *dev = osd_obj2dev(obj);
6549 struct osd_thread_info *oti = osd_oti_get(env);
6550 struct osd_inode_id *id = &oti->oti_id;
6551 struct lu_fid *fid = &it->oie_dirent->oied_fid;
6552 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6553 __u32 ino = it->oie_dirent->oied_ino;
6557 LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
6559 if (attr & LUDA_VERIFY) {
6560 if (unlikely(ino == osd_remote_parent_ino(dev))) {
6561 attr |= LUDA_IGNORE;
6562 /* If the parent is on remote MDT, and there
6563 * is no FID-in-dirent, then we have to get
6564 * the parent FID from the linkEA. */
6565 if (!fid_is_sane(fid) &&
6566 it->oie_dirent->oied_namelen == 2 &&
6567 it->oie_dirent->oied_name[0] == '.' &&
6568 it->oie_dirent->oied_name[1] == '.')
6569 osd_get_pfid_from_linkea(env, obj, fid);
6571 rc = osd_dirent_check_repair(env, obj, it, fid, id,
6575 if (!fid_is_sane(fid))
6576 attr |= LUDA_UNKNOWN;
6578 attr &= ~LU_DIRENT_ATTRS_MASK;
6579 if (!fid_is_sane(fid)) {
6580 bool is_dotdot = false;
6581 if (it->oie_dirent->oied_namelen == 2 &&
6582 it->oie_dirent->oied_name[0] == '.' &&
6583 it->oie_dirent->oied_name[1] == '.')
6585 /* If the parent is on remote MDT, and there
6586 * is no FID-in-dirent, then we have to get
6587 * the parent FID from the linkEA. */
6588 if (ino == osd_remote_parent_ino(dev) && is_dotdot) {
6589 rc = osd_get_pfid_from_linkea(env, obj, fid);
6591 if (is_dotdot == false &&
6592 OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
6595 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6598 osd_id_gen(id, ino, OSD_OII_NOGEN);
6602 /* Pack the entry anyway, at least the offset is right. */
6603 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
6604 it->oie_dirent->oied_name,
6605 it->oie_dirent->oied_namelen,
6606 it->oie_dirent->oied_type, attr);
6611 if (osd_remote_fid(env, dev, fid))
6614 if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
6615 osd_add_oi_cache(oti, dev, id, fid);
6617 RETURN(rc > 0 ? 0 : rc);
6621 * Returns the record size size at current position.
6623 * This function will return record(lu_dirent) size in bytes.
6625 * \param[in] env execution environment
6626 * \param[in] di iterator's in memory structure
6627 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
6628 * calculate the lu_dirent size.
6630 * \retval record size(in bytes & in memory) of the current lu_dirent
6633 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
6636 struct osd_it_ea *it = (struct osd_it_ea *)di;
6638 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
6642 * Returns a cookie for current position of the iterator head, so that
6643 * user can use this cookie to load/start the iterator next time.
6645 * \param di iterator's in memory structure
6647 * \retval cookie for current position, on success
6649 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
6651 struct osd_it_ea *it = (struct osd_it_ea *)di;
6653 return it->oie_dirent->oied_off;
6657 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6658 * to load a directory entry at a time and stored it i inn,
6659 * in iterator's in-memory data structure.
6661 * \param di struct osd_it_ea, iterator's in memory structure
6663 * \retval +ve on success
6664 * \retval -ve on error
6666 static int osd_it_ea_load(const struct lu_env *env,
6667 const struct dt_it *di, __u64 hash)
6669 struct osd_it_ea *it = (struct osd_it_ea *)di;
6673 it->oie_file.f_pos = hash;
6675 rc = osd_ldiskfs_it_fill(env, di);
6686 * Index lookup function for interoperability mode (b11826).
6688 * \param key, key i.e. file name to be searched
6690 * \retval +ve, on success
6691 * \retval -ve, on error
6693 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
6694 struct dt_rec *rec, const struct dt_key *key)
6696 struct osd_object *obj = osd_dt_obj(dt);
6701 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
6702 LINVRNT(osd_invariant(obj));
6704 rc = osd_ea_lookup_rec(env, obj, rec, key);
6711 * Index and Iterator operations for interoperability
6712 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
6714 static const struct dt_index_operations osd_index_ea_ops = {
6715 .dio_lookup = osd_index_ea_lookup,
6716 .dio_declare_insert = osd_index_declare_ea_insert,
6717 .dio_insert = osd_index_ea_insert,
6718 .dio_declare_delete = osd_index_declare_ea_delete,
6719 .dio_delete = osd_index_ea_delete,
6721 .init = osd_it_ea_init,
6722 .fini = osd_it_ea_fini,
6723 .get = osd_it_ea_get,
6724 .put = osd_it_ea_put,
6725 .next = osd_it_ea_next,
6726 .key = osd_it_ea_key,
6727 .key_size = osd_it_ea_key_size,
6728 .rec = osd_it_ea_rec,
6729 .rec_size = osd_it_ea_rec_size,
6730 .store = osd_it_ea_store,
6731 .load = osd_it_ea_load
6735 static void *osd_key_init(const struct lu_context *ctx,
6736 struct lu_context_key *key)
6738 struct osd_thread_info *info;
6740 OBD_ALLOC_PTR(info);
6742 return ERR_PTR(-ENOMEM);
6744 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6745 if (info->oti_it_ea_buf == NULL)
6748 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
6750 info->oti_hlock = ldiskfs_htree_lock_alloc();
6751 if (info->oti_hlock == NULL)
6757 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6760 return ERR_PTR(-ENOMEM);
6763 static void osd_key_fini(const struct lu_context *ctx,
6764 struct lu_context_key *key, void* data)
6766 struct osd_thread_info *info = data;
6767 struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
6768 struct osd_idmap_cache *idc = info->oti_ins_cache;
6770 if (info->oti_inode != NULL)
6772 if (info->oti_hlock != NULL)
6773 ldiskfs_htree_lock_free(info->oti_hlock);
6774 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6775 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
6776 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
6777 lu_buf_free(&info->oti_big_buf);
6779 LASSERT(info->oti_ins_cache_size > 0);
6780 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
6781 info->oti_ins_cache = NULL;
6782 info->oti_ins_cache_size = 0;
6787 static void osd_key_exit(const struct lu_context *ctx,
6788 struct lu_context_key *key, void *data)
6790 struct osd_thread_info *info = data;
6792 LASSERT(info->oti_r_locks == 0);
6793 LASSERT(info->oti_w_locks == 0);
6794 LASSERT(info->oti_txns == 0);
6797 /* type constructor/destructor: osd_type_init, osd_type_fini */
6798 LU_TYPE_INIT_FINI(osd, &osd_key);
6800 struct lu_context_key osd_key = {
6801 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
6802 .lct_init = osd_key_init,
6803 .lct_fini = osd_key_fini,
6804 .lct_exit = osd_key_exit
6808 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
6809 const char *name, struct lu_device *next)
6811 struct osd_device *osd = osd_dev(d);
6813 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
6814 >= sizeof(osd->od_svname))
6816 return osd_procfs_init(osd, name);
6819 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
6821 struct seq_server_site *ss = osd_seq_site(osd);
6825 if (osd->od_is_ost || osd->od_cl_seq != NULL)
6828 if (unlikely(ss == NULL))
6831 OBD_ALLOC_PTR(osd->od_cl_seq);
6832 if (osd->od_cl_seq == NULL)
6835 rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
6836 osd->od_svname, ss->ss_server_seq);
6838 OBD_FREE_PTR(osd->od_cl_seq);
6839 osd->od_cl_seq = NULL;
6843 if (ss->ss_node_id == 0) {
6844 /* If the OSD on the sequence controller(MDT0), then allocate
6845 * sequence here, otherwise allocate sequence after connected
6846 * to MDT0 (see mdt_register_lwp_callback()). */
6847 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
6848 &osd->od_cl_seq->lcs_space, env);
6854 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
6856 if (osd->od_cl_seq == NULL)
6859 seq_client_fini(osd->od_cl_seq);
6860 OBD_FREE_PTR(osd->od_cl_seq);
6861 osd->od_cl_seq = NULL;
6864 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
6868 /* shutdown quota slave instance associated with the device */
6869 if (o->od_quota_slave != NULL) {
6870 qsd_fini(env, o->od_quota_slave);
6871 o->od_quota_slave = NULL;
6874 osd_fid_fini(env, o);
6879 static void osd_umount(const struct lu_env *env, struct osd_device *o)
6883 if (o->od_mnt != NULL) {
6884 shrink_dcache_sb(osd_sb(o));
6885 osd_sync(env, &o->od_dt_dev);
6894 static int osd_mount(const struct lu_env *env,
6895 struct osd_device *o, struct lustre_cfg *cfg)
6897 const char *name = lustre_cfg_string(cfg, 0);
6898 const char *dev = lustre_cfg_string(cfg, 1);
6900 unsigned long page, s_flags, lmd_flags = 0;
6901 struct page *__page;
6902 struct file_system_type *type;
6903 char *options = NULL;
6905 struct osd_thread_info *info = osd_oti_get(env);
6906 struct lu_fid *fid = &info->oti_fid;
6907 struct inode *inode;
6908 int rc = 0, force_over_512tb = 0;
6911 if (o->od_mnt != NULL)
6914 if (strlen(dev) >= sizeof(o->od_mntdev))
6916 strcpy(o->od_mntdev, dev);
6918 str = lustre_cfg_string(cfg, 2);
6919 s_flags = simple_strtoul(str, NULL, 0);
6920 str = strstr(str, ":");
6922 lmd_flags = simple_strtoul(str + 1, NULL, 0);
6923 opts = lustre_cfg_string(cfg, 3);
6925 if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
6926 CERROR("%s: device %s extents feature is not guaranteed to "
6927 "work on big-endian systems. Use \"bigendian_extents\" "
6928 "mount option to override.\n", name, dev);
6932 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
6933 if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
6934 CWARN("force_over_128tb option is deprecated. "
6935 "Filesystems less than 512TB can be created without any "
6936 "force options. Use force_over_512tb option for "
6937 "filesystems greater than 512TB.\n");
6940 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
6941 if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
6942 CWARN("force_over_256tb option is deprecated. "
6943 "Filesystems less than 512TB can be created without any "
6944 "force options. Use force_over_512tb option for "
6945 "filesystems greater than 512TB.\n");
6949 if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
6950 force_over_512tb = 1;
6952 __page = alloc_page(GFP_KERNEL);
6954 GOTO(out, rc = -ENOMEM);
6955 page = (unsigned long)page_address(__page);
6956 options = (char *)page;
6959 /* strip out the options for back compatiblity */
6960 static char *sout[] = {
6967 /* strip out option we processed in osd */
6968 "bigendian_extents",
6974 strcat(options, opts);
6975 for (rc = 0, str = options; sout[rc]; ) {
6976 char *op = strstr(str, sout[rc]);
6982 if (op == options || *(op - 1) == ',') {
6983 str = op + strlen(sout[rc]);
6984 if (*str == ',' || *str == '\0') {
6985 *str == ',' ? str++ : str;
6986 memmove(op, str, strlen(str) + 1);
6989 for (str = op; *str != ',' && *str != '\0'; str++)
6993 strncat(options, "user_xattr,acl", 14);
6996 /* Glom up mount options */
6997 if (*options != '\0')
6998 strcat(options, ",");
6999 strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7001 type = get_fs_type("ldiskfs");
7003 CERROR("%s: cannot find ldiskfs module\n", name);
7004 GOTO(out, rc = -ENODEV);
7007 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7008 module_put(type->owner);
7010 if (IS_ERR(o->od_mnt)) {
7011 rc = PTR_ERR(o->od_mnt);
7013 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7017 if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7018 osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7019 force_over_512tb == 0) {
7020 CERROR("%s: device %s LDISKFS does not support filesystems "
7021 "greater than 512TB and can cause data corruption. "
7022 "Use \"force_over_512tb\" mount option to override.\n",
7024 GOTO(out_mnt, rc = -EINVAL);
7027 if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7028 #ifdef HAVE_DEV_SET_RDONLY
7029 dev_set_rdonly(osd_sb(o)->s_bdev);
7030 o->od_dt_dev.dd_rdonly = 1;
7031 LCONSOLE_WARN("%s: set dev_rdonly on this device\n", name);
7033 LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7036 GOTO(out_mnt, rc = -EOPNOTSUPP);
7039 #ifdef HAVE_DEV_SET_RDONLY
7040 if (dev_check_rdonly(osd_sb(o)->s_bdev)) {
7041 CERROR("%s: underlying device %s is marked as "
7042 "read-only. Setup failed\n", name, dev);
7044 GOTO(out_mnt, rc = -EROFS);
7049 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
7050 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
7051 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7052 GOTO(out_mnt, rc = -EINVAL);
7055 #ifdef LDISKFS_MOUNT_DIRDATA
7056 if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
7057 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
7058 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7059 else if (!o->od_is_ost)
7060 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7061 "enabling the dirdata feature. If you do not want to "
7062 "downgrade to Lustre-1.x again, you can enable it via "
7063 "'tune2fs -O dirdata device'\n", name, dev);
7065 inode = osd_sb(o)->s_root->d_inode;
7066 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7067 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7069 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7073 if (lmd_flags & LMD_FLG_NOSCRUB)
7084 __free_page(__page);
7089 static struct lu_device *osd_device_fini(const struct lu_env *env,
7090 struct lu_device *d)
7092 struct osd_device *o = osd_dev(d);
7095 osd_shutdown(env, o);
7097 osd_scrub_cleanup(env, o);
7098 osd_obj_map_fini(o);
7104 static int osd_device_init0(const struct lu_env *env,
7105 struct osd_device *o,
7106 struct lustre_cfg *cfg)
7108 struct lu_device *l = osd2lu_dev(o);
7109 struct osd_thread_info *info;
7113 /* if the module was re-loaded, env can loose its keys */
7114 rc = lu_env_refill((struct lu_env *) env);
7117 info = osd_oti_get(env);
7120 l->ld_ops = &osd_lu_ops;
7121 o->od_dt_dev.dd_ops = &osd_dt_ops;
7123 spin_lock_init(&o->od_osfs_lock);
7124 mutex_init(&o->od_otable_mutex);
7125 INIT_LIST_HEAD(&o->od_orphan_list);
7127 o->od_read_cache = 1;
7128 o->od_writethrough_cache = 1;
7129 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7131 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7132 sizeof(o->od_svname));
7133 if (cplen >= sizeof(o->od_svname)) {
7138 o->od_index = -1; /* -1 means index is invalid */
7139 rc = server_name2index(o->od_svname, &o->od_index, NULL);
7140 if (rc == LDD_F_SV_TYPE_OST)
7143 o->od_full_scrub_ratio = OFSR_DEFAULT;
7144 o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7145 rc = osd_mount(env, o, cfg);
7149 rc = osd_obj_map_init(env, o);
7153 rc = lu_site_init(&o->od_site, l);
7155 GOTO(out_compat, rc);
7156 o->od_site.ls_bottom_dev = l;
7158 rc = lu_site_init_finish(&o->od_site);
7162 INIT_LIST_HEAD(&o->od_ios_list);
7163 /* setup scrub, including OI files initialization */
7164 rc = osd_scrub_setup(env, o);
7168 rc = osd_procfs_init(o, o->od_svname);
7170 CERROR("%s: can't initialize procfs: rc = %d\n",
7172 GOTO(out_scrub, rc);
7175 LASSERT(l->ld_site->ls_linkage.next != NULL);
7176 LASSERT(l->ld_site->ls_linkage.prev != NULL);
7178 /* initialize quota slave instance */
7179 o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
7181 if (IS_ERR(o->od_quota_slave)) {
7182 rc = PTR_ERR(o->od_quota_slave);
7183 o->od_quota_slave = NULL;
7184 GOTO(out_procfs, rc);
7192 osd_scrub_cleanup(env, o);
7194 lu_site_fini(&o->od_site);
7196 osd_obj_map_fini(o);
7203 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7204 struct lu_device_type *t,
7205 struct lustre_cfg *cfg)
7207 struct osd_device *o;
7212 return ERR_PTR(-ENOMEM);
7214 rc = dt_device_init(&o->od_dt_dev, t);
7216 /* Because the ctx might be revived in dt_device_init,
7217 * refill the env here */
7218 lu_env_refill((struct lu_env *)env);
7219 rc = osd_device_init0(env, o, cfg);
7221 dt_device_fini(&o->od_dt_dev);
7224 if (unlikely(rc != 0))
7227 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7230 static struct lu_device *osd_device_free(const struct lu_env *env,
7231 struct lu_device *d)
7233 struct osd_device *o = osd_dev(d);
7236 /* XXX: make osd top device in order to release reference */
7237 d->ld_site->ls_top_dev = d;
7238 lu_site_purge(env, d->ld_site, -1);
7239 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7240 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7241 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7243 lu_site_fini(&o->od_site);
7244 dt_device_fini(&o->od_dt_dev);
7249 static int osd_process_config(const struct lu_env *env,
7250 struct lu_device *d, struct lustre_cfg *cfg)
7252 struct osd_device *o = osd_dev(d);
7256 switch (cfg->lcfg_command) {
7258 rc = osd_mount(env, o, cfg);
7261 lu_dev_del_linkage(d->ld_site, d);
7262 rc = osd_shutdown(env, o);
7265 LASSERT(&o->od_dt_dev);
7266 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
7267 cfg, &o->od_dt_dev);
7268 if (rc > 0 || rc == -ENOSYS)
7269 rc = class_process_proc_param(PARAM_OST,
7270 lprocfs_osd_obd_vars,
7271 cfg, &o->od_dt_dev);
7280 static int osd_recovery_complete(const struct lu_env *env,
7281 struct lu_device *d)
7283 struct osd_device *osd = osd_dev(d);
7287 if (osd->od_quota_slave == NULL)
7290 /* start qsd instance on recovery completion, this notifies the quota
7291 * slave code that we are about to process new requests now */
7292 rc = qsd_start(env, osd->od_quota_slave);
7297 * we use exports to track all osd users
7299 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7300 struct obd_device *obd, struct obd_uuid *cluuid,
7301 struct obd_connect_data *data, void *localdata)
7303 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7304 struct lustre_handle conn;
7308 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7310 rc = class_connect(&conn, obd, cluuid);
7314 *exp = class_conn2export(&conn);
7316 spin_lock(&osd->od_osfs_lock);
7318 spin_unlock(&osd->od_osfs_lock);
7324 * once last export (we don't count self-export) disappeared
7325 * osd can be released
7327 static int osd_obd_disconnect(struct obd_export *exp)
7329 struct obd_device *obd = exp->exp_obd;
7330 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7331 int rc, release = 0;
7334 /* Only disconnect the underlying layers on the final disconnect. */
7335 spin_lock(&osd->od_osfs_lock);
7337 if (osd->od_connects == 0)
7339 spin_unlock(&osd->od_osfs_lock);
7341 rc = class_disconnect(exp); /* bz 9811 */
7343 if (rc == 0 && release)
7344 class_manual_cleanup(obd);
7348 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
7349 struct lu_device *dev)
7351 struct osd_device *osd = osd_dev(dev);
7352 struct lr_server_data *lsd =
7353 &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
7357 if (osd->od_quota_slave != NULL) {
7358 /* set up quota slave objects */
7359 result = qsd_prepare(env, osd->od_quota_slave);
7364 if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
7365 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
7366 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
7367 osd->od_index_in_idif = 1;
7369 osd->od_index_in_idif = 0;
7370 result = osd_register_proc_index_in_idif(osd);
7375 osd->od_index_in_idif = 1;
7379 result = osd_fid_init(env, osd);
7384 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
7385 struct lu_fid *fid, struct md_op_data *op_data)
7387 struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
7389 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
7392 static const struct lu_object_operations osd_lu_obj_ops = {
7393 .loo_object_init = osd_object_init,
7394 .loo_object_delete = osd_object_delete,
7395 .loo_object_release = osd_object_release,
7396 .loo_object_free = osd_object_free,
7397 .loo_object_print = osd_object_print,
7398 .loo_object_invariant = osd_object_invariant
7401 const struct lu_device_operations osd_lu_ops = {
7402 .ldo_object_alloc = osd_object_alloc,
7403 .ldo_process_config = osd_process_config,
7404 .ldo_recovery_complete = osd_recovery_complete,
7405 .ldo_prepare = osd_prepare,
7408 static const struct lu_device_type_operations osd_device_type_ops = {
7409 .ldto_init = osd_type_init,
7410 .ldto_fini = osd_type_fini,
7412 .ldto_start = osd_type_start,
7413 .ldto_stop = osd_type_stop,
7415 .ldto_device_alloc = osd_device_alloc,
7416 .ldto_device_free = osd_device_free,
7418 .ldto_device_init = osd_device_init,
7419 .ldto_device_fini = osd_device_fini
7422 static struct lu_device_type osd_device_type = {
7423 .ldt_tags = LU_DEVICE_DT,
7424 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
7425 .ldt_ops = &osd_device_type_ops,
7426 .ldt_ctx_tags = LCT_LOCAL,
7429 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
7431 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7432 struct super_block *sb = osd_sb(osd);
7434 return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
7438 * lprocfs legacy support.
7440 static struct obd_ops osd_obd_device_ops = {
7441 .o_owner = THIS_MODULE,
7442 .o_connect = osd_obd_connect,
7443 .o_disconnect = osd_obd_disconnect,
7444 .o_fid_alloc = osd_fid_alloc,
7445 .o_health_check = osd_health_check,
7448 static int __init osd_init(void)
7452 LASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
7453 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
7454 /* please, try to keep osd_thread_info smaller than a page */
7455 CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
7460 rc = lu_kmem_init(ldiskfs_caches);
7464 rc = class_register_type(&osd_obd_device_ops, NULL, true,
7465 lprocfs_osd_module_vars,
7466 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
7468 lu_kmem_fini(ldiskfs_caches);
7472 static void __exit osd_exit(void)
7474 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
7475 lu_kmem_fini(ldiskfs_caches);
7478 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7479 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
7480 MODULE_VERSION(LUSTRE_VERSION_STRING);
7481 MODULE_LICENSE("GPL");
7483 module_init(osd_init);
7484 module_exit(osd_exit);