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, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * lustre/osd/osd_handler.c
33 * Top-level entry points into osd module
35 * Author: Nikita Danilov <nikita@clusterfs.com>
36 * Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
39 #define DEBUG_SUBSYSTEM S_OSD
41 #include <linux/fs_struct.h>
42 #include <linux/kallsyms.h>
43 #include <linux/module.h>
44 #include <linux/user_namespace.h>
45 #include <linux/uidgid.h>
47 /* prerequisite for linux/xattr.h */
48 #include <linux/types.h>
49 /* prerequisite for linux/xattr.h */
51 /* XATTR_{REPLACE,CREATE} */
52 #include <linux/xattr.h>
54 #include <ldiskfs/ldiskfs.h>
55 #include <ldiskfs/xattr.h>
56 #include <ldiskfs/ldiskfs_extents.h>
59 * struct OBD_{ALLOC,FREE}*()
62 #include <obd_support.h>
63 /* struct ptlrpc_thread */
64 #include <lustre_net.h>
65 #include <lustre_fid.h>
67 #include <uapi/linux/lustre/lustre_param.h>
69 #include "osd_internal.h"
70 #include "osd_dynlocks.h"
72 /* llo_* api support */
73 #include <md_object.h>
74 #include <lustre_quota.h>
76 #include <lustre_linkea.h>
78 /* encoding routines */
79 #include <lustre_crypto.h>
81 /* Maximum EA size is limited by LNET_MTU for remote objects */
82 #define OSD_MAX_EA_SIZE 1048364
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[] = "..";
117 static const struct lu_object_operations osd_lu_obj_ops;
118 static const struct dt_object_operations osd_obj_ops;
119 static const struct dt_object_operations osd_obj_otable_it_ops;
120 static const struct dt_index_operations osd_index_iam_ops;
121 static const struct dt_index_operations osd_index_ea_ops;
123 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
124 const struct lu_fid *fid);
125 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
126 struct osd_device *osd);
128 int osd_trans_declare_op2rb[] = {
129 [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
130 [OSD_OT_PUNCH] = OSD_OT_MAX,
131 [OSD_OT_XATTR_SET] = OSD_OT_XATTR_SET,
132 [OSD_OT_CREATE] = OSD_OT_DESTROY,
133 [OSD_OT_DESTROY] = OSD_OT_CREATE,
134 [OSD_OT_REF_ADD] = OSD_OT_REF_DEL,
135 [OSD_OT_REF_DEL] = OSD_OT_REF_ADD,
136 [OSD_OT_WRITE] = OSD_OT_WRITE,
137 [OSD_OT_INSERT] = OSD_OT_DELETE,
138 [OSD_OT_DELETE] = OSD_OT_INSERT,
139 [OSD_OT_QUOTA] = OSD_OT_MAX,
142 static int osd_has_index(const struct osd_object *obj)
144 return obj->oo_dt.do_index_ops != NULL;
147 static int osd_object_invariant(const struct lu_object *l)
149 return osd_invariant(osd_obj(l));
153 * Concurrency: doesn't matter
155 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
157 struct osd_thread_info *oti = osd_oti_get(env);
159 return oti->oti_w_locks > 0 && o->oo_owner == env;
163 * Concurrency: doesn't access mutable data
165 static int osd_root_get(const struct lu_env *env,
166 struct dt_device *dev, struct lu_fid *f)
168 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
173 * the following set of functions are used to maintain per-thread
174 * cache of FID->ino mapping. this mechanism is needed to resolve
175 * FID to inode at dt_insert() which in turn stores ino in the
176 * directory entries to keep ldiskfs compatible with ext[34].
177 * due to locking-originated restrictions we can't lookup ino
178 * using LU cache (deadlock is possible). lookup using OI is quite
179 * expensive. so instead we maintain this cache and methods like
180 * dt_create() fill it. so in the majority of cases dt_insert() is
181 * able to find needed mapping in lockless manner.
183 static struct osd_idmap_cache *
184 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
185 const struct lu_fid *fid)
187 struct osd_thread_info *oti = osd_oti_get(env);
188 struct osd_idmap_cache *idc = oti->oti_ins_cache;
191 for (i = 0; i < oti->oti_ins_cache_used; i++) {
192 if (!lu_fid_eq(&idc[i].oic_fid, fid))
194 if (idc[i].oic_dev != osd)
203 static struct osd_idmap_cache *
204 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
205 const struct lu_fid *fid)
207 struct osd_thread_info *oti = osd_oti_get(env);
208 struct osd_idmap_cache *idc;
211 if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
212 i = oti->oti_ins_cache_size * 2;
214 i = OSD_INS_CACHE_SIZE;
215 OBD_ALLOC_PTR_ARRAY(idc, i);
217 return ERR_PTR(-ENOMEM);
218 if (oti->oti_ins_cache != NULL) {
219 memcpy(idc, oti->oti_ins_cache,
220 oti->oti_ins_cache_used * sizeof(*idc));
221 OBD_FREE_PTR_ARRAY(oti->oti_ins_cache,
222 oti->oti_ins_cache_used);
224 oti->oti_ins_cache = idc;
225 oti->oti_ins_cache_size = i;
228 idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
231 idc->oic_lid.oii_ino = 0;
232 idc->oic_lid.oii_gen = 0;
239 * lookup mapping for the given fid in the cache, initialize a
240 * new one if not found. the initialization checks whether the
241 * object is local or remote. for local objects, OI is used to
242 * learn ino/generation. the function is used when the caller
243 * has no information about the object, e.g. at dt_insert().
245 static struct osd_idmap_cache *
246 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
247 const struct lu_fid *fid)
249 struct osd_idmap_cache *idc;
252 idc = osd_idc_find(env, osd, fid);
253 LASSERT(!IS_ERR(idc));
257 CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
258 osd->od_svname, PFID(fid));
260 /* new mapping is needed */
261 idc = osd_idc_add(env, osd, fid);
263 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
264 osd->od_svname, PFID(fid), PTR_ERR(idc));
269 rc = osd_remote_fid(env, osd, fid);
270 if (unlikely(rc < 0))
274 /* the object is local, lookup in OI */
275 /* XXX: probably cheaper to lookup in LU first? */
276 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
278 if (unlikely(rc < 0)) {
279 CERROR("can't lookup: rc = %d\n", rc);
283 /* the object is remote */
290 static void osd_idc_dump_lma(const struct lu_env *env,
291 struct osd_device *osd,
295 struct osd_thread_info *info = osd_oti_get(env);
296 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
297 const struct lu_fid *fid;
298 struct osd_inode_id lid;
302 inode = osd_ldiskfs_iget(osd_sb(osd), ino);
304 CERROR("%s: can't get inode %lu: rc = %d\n",
305 osd->od_svname, ino, (int)PTR_ERR(inode));
308 if (is_bad_inode(inode)) {
309 CERROR("%s: bad inode %lu\n", osd->od_svname, ino);
312 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
314 CERROR("%s: can't get LMA for %lu: rc = %d\n",
315 osd->od_svname, ino, rc);
318 fid = &loa->loa_lma.lma_self_fid;
319 LCONSOLE(D_INFO, "%s: "DFID" in inode %lu/%u\n", osd->od_svname,
320 PFID(fid), ino, (unsigned)inode->i_generation);
323 rc = osd_oi_lookup(osd_oti_get(env), osd, fid, &lid, 0);
325 CERROR("%s: can't lookup "DFID": rc = %d\n",
326 osd->od_svname, PFID(fid), rc);
329 LCONSOLE(D_INFO, "%s: "DFID" maps to %u/%u\n", osd->od_svname,
330 PFID(fid), lid.oii_ino, lid.oii_gen);
335 static void osd_idc_dump_debug(const struct lu_env *env,
336 struct osd_device *osd,
337 const struct lu_fid *fid,
341 struct osd_inode_id lid;
345 rc = osd_oi_lookup(osd_oti_get(env), osd, fid, &lid, 0);
347 LCONSOLE(D_INFO, "%s: "DFID" maps to %u/%u\n",
348 osd->od_svname, PFID(fid), lid.oii_ino, lid.oii_gen);
349 osd_idc_dump_lma(env, osd, lid.oii_ino, false);
351 CERROR("%s: can't lookup "DFID": rc = %d\n",
352 osd->od_svname, PFID(fid), rc);
355 osd_idc_dump_lma(env, osd, ino1, true);
357 osd_idc_dump_lma(env, osd, ino2, true);
361 * lookup mapping for given FID and fill it from the given object.
362 * the object is lolcal by definition.
364 static int osd_idc_find_and_init(const struct lu_env *env,
365 struct osd_device *osd,
366 struct osd_object *obj)
368 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
369 struct osd_idmap_cache *idc;
371 idc = osd_idc_find(env, osd, fid);
372 LASSERT(!IS_ERR(idc));
374 if (obj->oo_inode == NULL)
376 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
377 if (idc->oic_lid.oii_ino) {
378 osd_idc_dump_debug(env, osd, fid,
379 idc->oic_lid.oii_ino,
380 obj->oo_inode->i_ino);
383 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
384 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
389 CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
390 osd->od_svname, PFID(fid));
392 /* new mapping is needed */
393 idc = osd_idc_add(env, osd, fid);
395 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
396 osd->od_svname, PFID(fid), PTR_ERR(idc));
400 if (obj->oo_inode != NULL) {
401 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
402 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
408 * OSD object methods.
412 * Concurrency: no concurrent access is possible that early in object
415 static struct lu_object *osd_object_alloc(const struct lu_env *env,
416 const struct lu_object_header *hdr,
419 struct osd_object *mo;
424 struct lu_object_header *h;
425 struct osd_device *o = osd_dev(d);
427 l = &mo->oo_dt.do_lu;
428 if (unlikely(o->od_in_init)) {
435 lu_object_header_init(h);
436 lu_object_init(l, h, d);
437 lu_object_add_top(h, l);
440 dt_object_init(&mo->oo_dt, NULL, d);
441 mo->oo_header = NULL;
444 mo->oo_dt.do_ops = &osd_obj_ops;
445 l->lo_ops = &osd_lu_obj_ops;
446 init_rwsem(&mo->oo_sem);
447 init_rwsem(&mo->oo_ext_idx_sem);
448 spin_lock_init(&mo->oo_guard);
449 INIT_LIST_HEAD(&mo->oo_xattr_list);
455 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
456 struct dentry *dentry, struct lustre_ost_attrs *loa)
460 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
461 (void *)loa, sizeof(*loa));
463 struct lustre_mdt_attrs *lma = &loa->loa_lma;
465 if (rc < sizeof(*lma))
469 lustre_loa_swab(loa, true);
470 /* Check LMA compatibility */
471 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
473 CWARN("%s: unsupported incompat LMA feature(s) %#x for fid = "DFID", ino = %lu: rc = %d\n",
475 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
476 PFID(&lma->lma_self_fid), inode->i_ino, rc);
478 } else if (rc == 0) {
486 * retrieve object from backend ext fs.
488 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
489 struct osd_inode_id *id)
492 struct inode *inode = NULL;
495 * if we look for an inode withing a running
496 * transaction, then we risk to deadlock
497 * osd_dirent_check_repair() breaks this
499 /* LASSERT(current->journal_info == NULL); */
501 inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
503 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
504 id->oii_ino, PTR_ERR(inode));
505 } else if (id->oii_gen != OSD_OII_NOGEN &&
506 inode->i_generation != id->oii_gen) {
507 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
508 "i_generation = %u\n",
509 id->oii_ino, id->oii_gen, inode->i_generation);
511 inode = ERR_PTR(-ESTALE);
512 } else if (inode->i_nlink == 0) {
514 * due to parallel readdir and unlink,
515 * we can have dead inode here.
517 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
519 inode = ERR_PTR(-ESTALE);
520 } else if (is_bad_inode(inode)) {
522 CWARN("%s: bad inode: ino = %u: rc = %d\n",
523 osd_dev2name(dev), id->oii_ino, rc);
526 } else if ((rc = osd_attach_jinode(inode))) {
530 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
531 if (id->oii_gen == OSD_OII_NOGEN)
532 osd_id_gen(id, inode->i_ino, inode->i_generation);
535 * Do not update file c/mtime in ldiskfs.
536 * NB: we don't have any lock to protect this because we don't
537 * have reference on osd_object now, but contention with
538 * another lookup + attr_set can't happen in the tiny window
539 * between if (...) and set S_NOCMTIME.
541 if (!(inode->i_flags & S_NOCMTIME))
542 inode->i_flags |= S_NOCMTIME;
547 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
548 handle_t *handle, struct dentry *child,
549 struct inode *inode, struct htree_lock *hlock)
553 rc = __ldiskfs_add_entry(handle, child, inode, hlock);
554 if (rc == -ENOBUFS || rc == -ENOSPC) {
555 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
556 struct inode *parent = child->d_parent->d_inode;
557 struct lu_fid *fid = NULL;
558 char fidstr[FID_LEN + 1] = "unknown";
560 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
562 fid = &loa->loa_lma.lma_self_fid;
563 } else if (rc2 == -ENODATA) {
564 if (unlikely(is_root_inode(parent))) {
565 fid = &info->oti_fid3;
566 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
567 } else if (!osd->od_is_ost && osd->od_index == 0) {
568 fid = &info->oti_fid3;
569 lu_igif_build(fid, parent->i_ino,
570 parent->i_generation);
575 snprintf(fidstr, sizeof(fidstr), DFID, PFID(fid));
577 /* below message is checked in sanity.sh test_129 */
579 CWARN("%s: directory (inode: %lu, FID: %s) has reached max size limit\n",
580 osd_name(osd), parent->i_ino, fidstr);
582 rc = 0; /* ignore such error now */
583 CWARN("%s: directory (inode: %lu, FID: %s) is approaching max size limit\n",
584 osd_name(osd), parent->i_ino, fidstr);
594 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
595 struct osd_inode_id *id, struct lu_fid *fid)
597 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
601 inode = osd_iget(info, dev, id);
605 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
607 *fid = loa->loa_lma.lma_self_fid;
608 } else if (rc == -ENODATA) {
609 if (unlikely(is_root_inode(inode)))
610 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
612 lu_igif_build(fid, inode->i_ino, inode->i_generation);
620 static struct inode *osd_iget_check(struct osd_thread_info *info,
621 struct osd_device *dev,
622 const struct lu_fid *fid,
623 struct osd_inode_id *id,
632 * The cached OI mapping is trustable. If we cannot locate the inode
633 * via the cached OI mapping, then return the failure to the caller
634 * directly without further OI checking.
638 inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
641 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
644 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
645 PFID(fid), id->oii_ino, rc);
649 if (is_bad_inode(inode)) {
654 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
655 PFID(fid), id->oii_ino);
659 if (id->oii_gen != OSD_OII_NOGEN &&
660 inode->i_generation != id->oii_gen) {
665 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
666 "oii_gen = %u, i_generation = %u\n", PFID(fid),
667 id->oii_ino, id->oii_gen, inode->i_generation);
671 if (inode->i_nlink == 0) {
676 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
677 PFID(fid), id->oii_ino);
681 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
685 __u32 saved_ino = id->oii_ino;
686 __u32 saved_gen = id->oii_gen;
689 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
691 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
693 * XXX: There are four possible cases:
695 * Backup/restore caused the OI invalid.
697 * Someone unlinked the object but NOT removed
698 * the OI mapping, such as mount target device
699 * as ldiskfs, and modify something directly.
701 * Someone just removed the object between the
702 * former oi_lookup and the iget. It is normal.
703 * 4. Other failure cases.
705 * Generally, when the device is mounted, it will
706 * auto check whether the system is restored from
707 * file-level backup or not. We trust such detect
708 * to distinguish the 1st case from the 2nd case:
709 * if the OI files are consistent but may contain
710 * stale OI mappings because of case 2, if iget()
711 * returns -ENOENT or -ESTALE, then it should be
716 * If the OI mapping was in OI file before the
717 * osd_iget_check(), but now, it is disappear,
718 * then it must be removed by race. That is a
724 * It is the OI scrub updated the OI mapping by race.
725 * The new OI mapping must be valid.
727 if (saved_ino != id->oii_ino ||
728 (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
737 if (dev->od_scrub.os_scrub.os_file.sf_flags &
740 * It still can be the case 2, but we cannot
741 * distinguish it from the case 1. So return
742 * -EREMCHG to block current operation until
743 * OI scrub rebuilt the OI mappings.
752 if (inode->i_generation == id->oii_gen)
757 if (id->oii_gen == OSD_OII_NOGEN)
758 osd_id_gen(id, inode->i_ino, inode->i_generation);
761 * Do not update file c/mtime in ldiskfs.
762 * NB: we don't have any lock to protect this because we don't
763 * have reference on osd_object now, but contention with
764 * another lookup + attr_set can't happen in the tiny window
765 * between if (...) and set S_NOCMTIME.
767 if (!(inode->i_flags & S_NOCMTIME))
768 inode->i_flags |= S_NOCMTIME;
785 * \retval +v: new filter_fid does not contain self-fid
786 * \retval 0: filter_fid_18_23, contains self-fid
787 * \retval -v: other failure cases
789 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
790 struct dentry *dentry, struct lu_fid *fid)
792 struct filter_fid *ff = &info->oti_ff;
793 struct ost_id *ostid = &info->oti_ostid;
796 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
797 if (rc == sizeof(struct filter_fid_18_23)) {
798 struct filter_fid_18_23 *ff_old = (void *)ff;
800 ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq));
801 rc = ostid_set_id(ostid, le64_to_cpu(ff_old->ff_objid));
803 * XXX: use 0 as the index for compatibility, the caller will
804 * handle index related issues when necessary.
807 ostid_to_fid(fid, ostid, 0);
808 } else if (rc >= (int)sizeof(struct filter_fid_24_29)) {
810 } else if (rc >= 0) {
817 static int osd_lma_self_repair(struct osd_thread_info *info,
818 struct osd_device *osd, struct inode *inode,
819 const struct lu_fid *fid, __u32 compat)
824 LASSERT(current->journal_info == NULL);
826 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
827 osd_dto_credits_noquota[DTO_XATTR_SET]);
830 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
835 rc = osd_ea_fid_set(info, inode, fid, compat, 0);
837 CWARN("%s: cannot self repair the LMA: rc = %d\n",
839 ldiskfs_journal_stop(jh);
843 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
845 struct osd_thread_info *info = osd_oti_get(env);
846 struct osd_device *osd = osd_obj2dev(obj);
847 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
848 struct lustre_mdt_attrs *lma = &loa->loa_lma;
849 struct inode *inode = obj->oo_inode;
850 struct dentry *dentry = &info->oti_obj_dentry;
851 struct lu_fid *fid = NULL;
852 const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
857 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
858 (void *)loa, sizeof(*loa));
859 if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
860 fid = &lma->lma_self_fid;
861 rc = osd_get_idif(info, inode, dentry, fid);
862 if (rc > 0 || (rc == -ENODATA && osd->od_index_in_idif)) {
864 * For the given OST-object, if it has neither LMA nor
865 * FID in XATTR_NAME_FID, then the given FID (which is
866 * contained in the @obj, from client RPC for locating
867 * the OST-object) is trusted. We use it to generate
870 osd_lma_self_repair(info, osd, inode, rfid,
881 lustre_lma_swab(lma);
882 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
883 (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
884 S_ISREG(inode->i_mode)))) {
885 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
886 "fid = "DFID", ino = %lu\n", osd_name(osd),
887 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
888 PFID(rfid), inode->i_ino);
891 fid = &lma->lma_self_fid;
892 if (lma->lma_compat & LMAC_STRIPE_INFO &&
894 obj->oo_pfid_in_lma = 1;
895 if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
897 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
901 if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
902 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
903 struct ost_id *oi = &info->oti_ostid;
904 struct lu_fid *fid1 = &info->oti_fid3;
905 __u32 idx = fid_idif_ost_idx(rfid);
908 * For old IDIF, the OST index is not part of the IDIF,
909 * Means that different OSTs may have the same IDIFs.
910 * Under such case, we need to make some compatible
911 * check to make sure to trigger OI scrub properly.
913 if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
914 /* Given @rfid is new, LMA is old. */
915 fid_to_ostid(fid, oi);
916 ostid_to_fid(fid1, oi, idx);
917 if (lu_fid_eq(fid1, rfid)) {
918 if (osd->od_index_in_idif)
919 osd_lma_self_repair(info, osd,
933 struct osd_check_lmv_buf {
934 /* please keep it as first member */
935 struct dir_context ctx;
936 struct osd_thread_info *oclb_info;
937 struct osd_device *oclb_dev;
943 * It is called internally by ->iterate*() to filter out the
944 * local slave object's FID of the striped directory.
946 * \retval 1 found the local slave's FID
947 * \retval 0 continue to check next item
948 * \retval -ve for failure
950 #ifdef HAVE_FILLDIR_USE_CTX
951 static int osd_stripe_dir_filldir(struct dir_context *buf,
953 static int osd_stripe_dir_filldir(void *buf,
955 const char *name, int namelen,
956 loff_t offset, __u64 ino, unsigned int d_type)
958 struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
959 struct osd_thread_info *oti = oclb->oclb_info;
960 struct lu_fid *fid = &oti->oti_fid3;
961 struct osd_inode_id *id = &oti->oti_id3;
962 struct osd_device *dev = oclb->oclb_dev;
971 sscanf(name + 1, SFID, RFID(fid));
972 if (!fid_is_sane(fid))
975 if (osd_remote_fid(oti->oti_env, dev, fid))
978 osd_id_gen(id, ino, OSD_OII_NOGEN);
979 inode = osd_iget(oti, dev, id);
981 return PTR_ERR(inode);
984 osd_add_oi_cache(oti, dev, id, fid);
985 osd_oii_insert(dev, fid, id, true);
986 oclb->oclb_found = true;
992 * When lookup item under striped directory, we need to locate the master
993 * MDT-object of the striped directory firstly, then the client will send
994 * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
995 * and the item's name. If the system is restored from MDT file level backup,
996 * then before the OI scrub completely built the OI files, the OI mappings of
997 * the master MDT-object and slave MDT-object may be invalid. Usually, it is
998 * not a problem for the master MDT-object. Because when locate the master
999 * MDT-object, we will do name based lookup (for the striped directory itself)
1000 * firstly, during such process we can setup the correct OI mapping for the
1001 * master MDT-object. But it will be trouble for the slave MDT-object. Because
1002 * the client will not trigger name based lookup on the MDT to locate the slave
1003 * MDT-object before locating item under the striped directory, then when
1004 * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
1005 * is invalid and does not know what the right OI mapping is, then the MDT has
1006 * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
1007 * the OI file, related OI mapping is unknown yet, please try again later. And
1008 * then client will re-try the RPC again and again until related OI mapping has
1009 * been updated. That is quite inefficient.
1011 * To resolve above trouble, we will handle it as the following two cases:
1013 * 1) The slave MDT-object and the master MDT-object are on different MDTs.
1014 * It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
1015 * is linked under /REMOTE_PARENT_DIR with the name of its FID string.
1016 * We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
1017 * directly. Please check osd_fid_lookup().
1019 * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
1020 * Under such case, during lookup the master MDT-object, we will lookup the
1021 * slave MDT-object via readdir against the master MDT-object, because the
1022 * slave MDT-objects information are stored as sub-directories with the name
1023 * "${FID}:${index}". Then when find the local slave MDT-object, its OI
1024 * mapping will be recorded. Then subsequent osd_fid_lookup() will know
1025 * the correct OI mapping for the slave MDT-object.
1027 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
1028 struct inode *inode)
1030 struct lu_buf *buf = &oti->oti_big_buf;
1031 struct dentry *dentry = &oti->oti_obj_dentry;
1033 struct lmv_mds_md_v1 *lmv1;
1034 struct osd_check_lmv_buf oclb = {
1035 .ctx.actor = osd_stripe_dir_filldir,
1038 .oclb_found = false,
1045 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
1047 if (rc == -ERANGE) {
1048 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
1050 lu_buf_realloc(buf, rc);
1051 if (buf->lb_buf == NULL)
1052 GOTO(out, rc = -ENOMEM);
1058 if (unlikely(rc == 0 || rc == -ENODATA))
1064 if (unlikely(buf->lb_buf == NULL)) {
1065 lu_buf_realloc(buf, rc);
1066 if (buf->lb_buf == NULL)
1067 GOTO(out, rc = -ENOMEM);
1073 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1076 filp = osd_quasi_file(oti->oti_env, inode);
1077 rc = osd_security_file_alloc(filp);
1082 oclb.oclb_items = 0;
1083 rc = iterate_dir(filp, &oclb.ctx);
1084 } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1085 filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1086 inode->i_fop->release(inode, filp);
1091 "%s: cannot check LMV, ino = %lu/%u: rc = %d\n",
1092 osd_ino2name(inode), inode->i_ino, inode->i_generation,
1100 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1101 const struct lu_fid *fid,
1102 const struct lu_object_conf *conf)
1104 struct osd_thread_info *info;
1105 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1106 struct osd_device *dev;
1107 struct osd_idmap_cache *oic;
1108 struct osd_inode_id *id;
1109 struct inode *inode = NULL;
1110 struct lustre_scrub *scrub;
1111 struct scrub_file *sf;
1112 __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1117 bool remote = false;
1118 bool trusted = true;
1119 bool updated = false;
1120 bool checked = false;
1124 LINVRNT(osd_invariant(obj));
1125 LASSERT(obj->oo_inode == NULL);
1127 if (fid_is_sane(fid) == 0) {
1128 CERROR("%s: invalid FID "DFID"\n", ldev->ld_obd->obd_name,
1134 dev = osd_dev(ldev);
1135 scrub = &dev->od_scrub.os_scrub;
1136 sf = &scrub->os_file;
1137 info = osd_oti_get(env);
1139 oic = &info->oti_cache;
1141 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1145 * For the object is created as locking anchor, or for the object to
1146 * be created on disk. No need to osd_oi_lookup() at here because FID
1147 * shouldn't never be re-used, if it's really a duplicate FID from
1148 * unexpected reason, we should be able to detect it later by calling
1149 * do_create->osd_oi_insert().
1151 if (conf && conf->loc_flags & LOC_F_NEW)
1152 GOTO(out, result = 0);
1154 /* Search order: 1. per-thread cache. */
1155 if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1161 memset(id, 0, sizeof(struct osd_inode_id));
1162 if (!list_empty(&scrub->os_inconsistent_items)) {
1163 /* Search order: 2. OI scrub pending list. */
1164 result = osd_oii_lookup(dev, fid, id);
1170 * The OI mapping in the OI file can be updated by the OI scrub
1171 * when we locate the inode via FID. So it may be not trustable.
1175 /* Search order: 3. OI files. */
1176 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1177 if (result == -ENOENT) {
1178 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1179 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1180 !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1182 GOTO(out, result = 0);
1187 /* -ESTALE is returned if inode of OST object doesn't exist */
1188 if (result == -ESTALE &&
1189 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1190 GOTO(out, result = 0);
1197 obj->oo_inode = NULL;
1198 /* for later passes through checks, not true on first pass */
1199 if (!IS_ERR_OR_NULL(inode))
1202 inode = osd_iget_check(info, dev, fid, id, trusted);
1203 if (!IS_ERR(inode)) {
1204 obj->oo_inode = inode;
1212 result = PTR_ERR(inode);
1213 if (result == -ENOENT || result == -ESTALE)
1214 GOTO(out, result = 0);
1216 if (result != -EREMCHG)
1221 * We still have chance to get the valid inode: for the
1222 * object which is referenced by remote name entry, the
1223 * object on the local MDT will be linked under the dir
1224 * of "/REMOTE_PARENT_DIR" with its FID string as name.
1226 * We do not know whether the object for the given FID
1227 * is referenced by some remote name entry or not, and
1228 * especially for DNE II, a multiple-linked object may
1229 * have many name entries reside on many MDTs.
1231 * To simplify the operation, OSD will not distinguish
1232 * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1233 * only happened for the RPC from other MDT during the
1234 * OI scrub, or for the client side RPC with FID only,
1235 * such as FID to path, or from old connected client.
1238 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1242 flags |= SS_AUTO_PARTIAL;
1243 flags &= ~SS_AUTO_FULL;
1248 if (scrub->os_running) {
1249 if (scrub->os_partial_scan && !scrub->os_in_join)
1252 if (IS_ERR_OR_NULL(inode) || result) {
1253 osd_oii_insert(dev, fid, id, result == -ENOENT);
1254 GOTO(out, result = -EINPROGRESS);
1258 LASSERT(obj->oo_inode == inode);
1260 osd_oii_insert(dev, fid, id, true);
1264 if (dev->od_scrub.os_scrub.os_auto_scrub_interval == AS_NEVER) {
1266 GOTO(out, result = -EREMCHG);
1269 LASSERT(obj->oo_inode == inode);
1271 osd_add_oi_cache(info, dev, id, fid);
1276 rc1 = osd_scrub_start(env, dev, flags);
1277 CDEBUG_LIMIT(D_LFSCK | D_CONSOLE | D_WARNING,
1278 "%s: trigger OI scrub by RPC for "DFID"/%u with flags %#x: rc = %d\n",
1279 osd_name(dev), PFID(fid), id->oii_ino, flags, rc1);
1280 if (rc1 && rc1 != -EALREADY)
1281 GOTO(out, result = -EREMCHG);
1283 if (IS_ERR_OR_NULL(inode) || result) {
1284 osd_oii_insert(dev, fid, id, result == -ENOENT);
1285 GOTO(out, result = -EINPROGRESS);
1289 LASSERT(obj->oo_inode == inode);
1291 osd_oii_insert(dev, fid, id, true);
1296 if (unlikely(obj->oo_header))
1299 result = osd_check_lma(env, obj);
1303 LASSERTF(id->oii_ino == inode->i_ino &&
1304 id->oii_gen == inode->i_generation,
1305 "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1306 PFID(fid), id->oii_ino, id->oii_gen,
1307 inode->i_ino, inode->i_generation);
1309 saved_ino = inode->i_ino;
1310 saved_gen = inode->i_generation;
1312 if (unlikely(result == -ENODATA)) {
1314 * If the OI scrub updated the OI mapping by race, it
1315 * must be valid. Trust the inode that has no LMA EA.
1320 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1323 * The OI mapping is still there, the inode is still
1324 * valid. It is just becaues the inode has no LMA EA.
1326 if (saved_ino == id->oii_ino &&
1327 saved_gen == id->oii_gen)
1331 * It is the OI scrub updated the OI mapping by race.
1332 * The new OI mapping must be valid.
1340 * "result == -ENOENT" means that the OI mappinghas been
1341 * removed by race, so the inode belongs to other object.
1343 * Others error can be returned directly.
1345 if (result == -ENOENT) {
1348 obj->oo_inode = NULL;
1353 if (result != -EREMCHG)
1359 * if two OST objects map to the same inode, and inode mode is
1360 * (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666), which means it's
1361 * reserved by precreate, and not written yet, in this case, don't
1362 * set inode for the object whose FID mismatch, so that it can create
1363 * inode and not block precreate.
1365 if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) &&
1366 inode->i_mode == (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666)) {
1367 obj->oo_inode = NULL;
1368 GOTO(out, result = 0);
1371 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1373 * "result == -ENOENT" means the cached OI mapping has been removed
1374 * from the OI file by race, above inode belongs to other object.
1376 if (result == -ENOENT) {
1379 obj->oo_inode = NULL;
1380 GOTO(out, result = 0);
1386 if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1388 osd_scrub_refresh_mapping(info, dev, fid, id, DTO_INDEX_DELETE,
1394 * It is the OI scrub updated the OI mapping by race.
1395 * The new OI mapping must be valid.
1403 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1404 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1406 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1408 if (lma->lma_compat & LMAC_STRIPE_INFO &&
1410 obj->oo_pfid_in_lma = 1;
1411 if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1413 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1414 } else if (result != -ENODATA) {
1419 obj->oo_compat_dot_created = 1;
1420 obj->oo_compat_dotdot_created = 1;
1422 if (S_ISDIR(inode->i_mode) &&
1423 (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1424 osd_check_lmv(info, dev, inode);
1426 result = osd_attach_jinode(inode);
1431 GOTO(out, result = 0);
1433 LASSERT(!obj->oo_hl_head);
1434 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1436 GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1439 if (result || !obj->oo_inode) {
1440 if (!IS_ERR_OR_NULL(inode))
1443 obj->oo_inode = NULL;
1445 fid_zero(&oic->oic_fid);
1448 LINVRNT(osd_invariant(obj));
1453 * Concurrency: shouldn't matter.
1455 static void osd_object_init0(struct osd_object *obj)
1457 LASSERT(obj->oo_inode != NULL);
1458 obj->oo_dt.do_body_ops = &osd_body_ops;
1459 obj->oo_dt.do_lu.lo_header->loh_attr |=
1460 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1464 * Concurrency: no concurrent access is possible that early in object
1467 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1468 const struct lu_object_conf *conf)
1470 struct osd_object *obj = osd_obj(l);
1473 LINVRNT(osd_invariant(obj));
1475 if (OBD_FAIL_PRECHECK(OBD_FAIL_MDS_LLOG_UMOUNT_RACE) &&
1476 cfs_fail_val == 2) {
1477 struct osd_thread_info *info = osd_oti_get(env);
1478 struct osd_idmap_cache *oic = &info->oti_cache;
1479 /* invalidate thread cache */
1480 memset(&oic->oic_fid, 0, sizeof(oic->oic_fid));
1482 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1483 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1484 l->lo_header->loh_attr |= LOHA_EXISTS;
1488 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1489 obj->oo_dt.do_body_ops = &osd_body_ops_new;
1490 if (result == 0 && obj->oo_inode != NULL) {
1491 struct osd_thread_info *oti = osd_oti_get(env);
1492 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1494 osd_object_init0(obj);
1495 if (unlikely(obj->oo_header))
1498 result = osd_get_lma(oti, obj->oo_inode,
1499 &oti->oti_obj_dentry, loa);
1502 * Convert LMAI flags to lustre LMA flags
1503 * and cache it to oo_lma_flags
1506 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1507 } else if (result == -ENODATA) {
1511 obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
1513 LINVRNT(osd_invariant(obj));
1518 * The first part of oxe_buf is xattr name, and is '\0' terminated.
1519 * The left part is for value, binary mode.
1521 struct osd_xattr_entry {
1522 struct list_head oxe_list;
1526 struct rcu_head oxe_rcu;
1530 static int osd_oxc_get(struct osd_object *obj, const char *name,
1533 struct osd_xattr_entry *tmp;
1534 struct osd_xattr_entry *oxe = NULL;
1535 size_t namelen = strlen(name);
1539 list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1540 if (namelen == tmp->oxe_namelen &&
1541 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1548 GOTO(out, rc = -ENOENT);
1550 if (!oxe->oxe_exist)
1551 GOTO(out, rc = -ENODATA);
1554 rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1557 if (buf->lb_buf == NULL)
1560 if (buf->lb_len < rc)
1561 GOTO(out, rc = -ERANGE);
1563 memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1570 static void osd_oxc_free(struct rcu_head *head)
1572 struct osd_xattr_entry *oxe;
1574 oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1575 OBD_FREE(oxe, oxe->oxe_len);
1578 static void osd_oxc_add(struct osd_object *obj, const char *name,
1579 const char *buf, int buflen)
1581 struct osd_xattr_entry *oxe;
1582 struct osd_xattr_entry *old = NULL;
1583 struct osd_xattr_entry *tmp;
1584 size_t namelen = strlen(name);
1585 size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1587 OBD_ALLOC(oxe, len);
1591 INIT_LIST_HEAD(&oxe->oxe_list);
1593 oxe->oxe_namelen = namelen;
1594 memcpy(oxe->oxe_buf, name, namelen);
1596 LASSERT(buf != NULL);
1597 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1598 oxe->oxe_exist = true;
1600 oxe->oxe_exist = false;
1603 /* this should be rarely called, just remove old and add new */
1604 spin_lock(&obj->oo_guard);
1605 list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1606 if (namelen == tmp->oxe_namelen &&
1607 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1613 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1614 call_rcu(&old->oxe_rcu, osd_oxc_free);
1616 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1618 spin_unlock(&obj->oo_guard);
1621 static void osd_oxc_del(struct osd_object *obj, const char *name)
1623 struct osd_xattr_entry *oxe;
1624 size_t namelen = strlen(name);
1626 spin_lock(&obj->oo_guard);
1627 list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1628 if (namelen == oxe->oxe_namelen &&
1629 strncmp(name, oxe->oxe_buf, namelen) == 0) {
1630 list_del_rcu(&oxe->oxe_list);
1631 call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1635 spin_unlock(&obj->oo_guard);
1638 static void osd_oxc_fini(struct osd_object *obj)
1640 struct osd_xattr_entry *oxe, *next;
1642 list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1643 list_del(&oxe->oxe_list);
1644 OBD_FREE(oxe, oxe->oxe_len);
1649 * Concurrency: no concurrent access is possible that late in object
1652 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1654 struct osd_object *obj = osd_obj(l);
1655 struct lu_object_header *h = obj->oo_header;
1657 LINVRNT(osd_invariant(obj));
1660 dt_object_fini(&obj->oo_dt);
1661 if (obj->oo_hl_head != NULL)
1662 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1663 /* obj doesn't contain an lu_object_header, so we don't need call_rcu */
1666 lu_object_header_free(h);
1670 * Concurrency: no concurrent access is possible that late in object
1673 static void osd_index_fini(struct osd_object *o)
1675 struct iam_container *bag;
1677 if (o->oo_dir != NULL) {
1678 bag = &o->oo_dir->od_container;
1679 if (o->oo_inode != NULL) {
1680 if (bag->ic_object == o->oo_inode)
1681 iam_container_fini(bag);
1683 OBD_FREE_PTR(o->oo_dir);
1689 OSD_TXN_OI_DELETE_CREDITS = 20,
1690 OSD_TXN_INODE_DELETE_CREDITS = 20
1697 #if OSD_THANDLE_STATS
1699 * Set time when the handle is allocated
1701 static void osd_th_alloced(struct osd_thandle *oth)
1703 oth->oth_alloced = ktime_get();
1707 * Set time when the handle started
1709 static void osd_th_started(struct osd_thandle *oth)
1711 oth->oth_started = ktime_get();
1715 * Check whether the we deal with this handle for too long.
1717 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1718 ktime_t alloced, ktime_t started,
1721 ktime_t now = ktime_get();
1723 LASSERT(dev != NULL);
1725 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1726 ktime_us_delta(started, alloced));
1727 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1728 ktime_us_delta(closed, started));
1729 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1730 ktime_us_delta(now, closed));
1732 if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1733 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1734 oth, now, alloced, started, closed);
1735 libcfs_debug_dumpstack(NULL);
1739 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
1741 ktime_t __closed = ktime_get(); \
1742 ktime_t __alloced = oth->oth_alloced; \
1743 ktime_t __started = oth->oth_started; \
1746 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
1749 #else /* OSD_THANDLE_STATS */
1751 #define osd_th_alloced(h) do {} while(0)
1752 #define osd_th_started(h) do {} while(0)
1753 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
1755 #endif /* OSD_THANDLE_STATS */
1758 * Concurrency: doesn't access mutable data.
1760 static int osd_param_is_not_sane(const struct osd_device *dev,
1761 const struct thandle *th)
1763 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1765 return oh->ot_credits > osd_transaction_size(dev);
1769 * Concurrency: shouldn't matter.
1771 static void osd_trans_commit_cb(struct super_block *sb,
1772 struct ldiskfs_journal_cb_entry *jcb, int error)
1774 struct osd_thandle *oh = container_of(jcb, struct osd_thandle, ot_jcb);
1775 struct thandle *th = &oh->ot_super;
1776 struct lu_device *lud = &th->th_dev->dd_lu_dev;
1777 struct osd_device *osd = osd_dev(lud);
1778 struct dt_txn_commit_cb *dcb, *tmp;
1780 LASSERT(oh->ot_handle == NULL);
1783 CERROR("transaction @0x%p commit error: %d\n", th, error);
1785 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DELAY_TRANS, 40);
1786 /* call per-transaction callbacks if any */
1787 list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1789 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1790 "commit callback entry: magic=%x name='%s'\n",
1791 dcb->dcb_magic, dcb->dcb_name);
1792 list_del_init(&dcb->dcb_linkage);
1793 dcb->dcb_func(NULL, th, dcb, error);
1796 lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1797 if (atomic_dec_and_test(&osd->od_commit_cb_in_flight))
1798 wake_up(&osd->od_commit_cb_done);
1804 static struct thandle *osd_trans_create(const struct lu_env *env,
1805 struct dt_device *d)
1807 struct osd_thread_info *oti = osd_oti_get(env);
1808 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1809 struct osd_thandle *oh;
1815 CERROR("%s: someone try to start transaction under "
1816 "readonly mode, should be disabled.\n",
1817 osd_name(osd_dt_dev(d)));
1819 RETURN(ERR_PTR(-EROFS));
1822 /* on pending IO in this thread should left from prev. request */
1823 LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1825 sb_start_write(osd_sb(osd_dt_dev(d)));
1827 OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1829 sb_end_write(osd_sb(osd_dt_dev(d)));
1830 RETURN(ERR_PTR(-ENOMEM));
1833 oh->ot_quota_trans = &oti->oti_quota_trans;
1834 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1839 oh->oh_declared_ext = 0;
1840 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1841 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1842 INIT_LIST_HEAD(&oh->ot_trunc_locks);
1845 memset(oti->oti_declare_ops, 0,
1846 sizeof(oti->oti_declare_ops));
1847 memset(oti->oti_declare_ops_cred, 0,
1848 sizeof(oti->oti_declare_ops_cred));
1849 memset(oti->oti_declare_ops_used, 0,
1850 sizeof(oti->oti_declare_ops_used));
1852 oti->oti_ins_cache_depth++;
1857 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1859 struct osd_thread_info *oti = osd_oti_get(env);
1860 struct osd_thandle *oh;
1862 oh = container_of(th, struct osd_thandle, ot_super);
1863 LASSERT(oh != NULL);
1865 CWARN(" create: %u/%u/%u, destroy: %u/%u/%u\n",
1866 oti->oti_declare_ops[OSD_OT_CREATE],
1867 oti->oti_declare_ops_cred[OSD_OT_CREATE],
1868 oti->oti_declare_ops_used[OSD_OT_CREATE],
1869 oti->oti_declare_ops[OSD_OT_DESTROY],
1870 oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1871 oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1872 CWARN(" attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1873 oti->oti_declare_ops[OSD_OT_ATTR_SET],
1874 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1875 oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1876 oti->oti_declare_ops[OSD_OT_XATTR_SET],
1877 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1878 oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1879 CWARN(" write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1880 oti->oti_declare_ops[OSD_OT_WRITE],
1881 oti->oti_declare_ops_cred[OSD_OT_WRITE],
1882 oti->oti_declare_ops_used[OSD_OT_WRITE],
1883 oti->oti_declare_ops[OSD_OT_PUNCH],
1884 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1885 oti->oti_declare_ops_used[OSD_OT_PUNCH],
1886 oti->oti_declare_ops[OSD_OT_QUOTA],
1887 oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1888 oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1889 CWARN(" insert: %u/%u/%u, delete: %u/%u/%u\n",
1890 oti->oti_declare_ops[OSD_OT_INSERT],
1891 oti->oti_declare_ops_cred[OSD_OT_INSERT],
1892 oti->oti_declare_ops_used[OSD_OT_INSERT],
1893 oti->oti_declare_ops[OSD_OT_DELETE],
1894 oti->oti_declare_ops_cred[OSD_OT_DELETE],
1895 oti->oti_declare_ops_used[OSD_OT_DELETE]);
1896 CWARN(" ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1897 oti->oti_declare_ops[OSD_OT_REF_ADD],
1898 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1899 oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1900 oti->oti_declare_ops[OSD_OT_REF_DEL],
1901 oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1902 oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1906 * Concurrency: shouldn't matter.
1908 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1911 struct osd_thread_info *oti = osd_oti_get(env);
1912 struct osd_device *dev = osd_dt_dev(d);
1914 struct osd_thandle *oh;
1919 LASSERT(current->journal_info == NULL);
1921 oh = container_of(th, struct osd_thandle, ot_super);
1922 LASSERT(oh != NULL);
1923 LASSERT(oh->ot_handle == NULL);
1925 rc = dt_txn_hook_start(env, d, th);
1929 if (unlikely(osd_param_is_not_sane(dev, th))) {
1930 static unsigned long last_printed;
1931 static int last_credits;
1934 * don't make noise on a tiny testing systems
1935 * actual credits misuse will be caught anyway
1937 if (last_credits != oh->ot_credits &&
1938 time_after(jiffies, last_printed +
1939 cfs_time_seconds(60)) &&
1940 osd_transaction_size(dev) > 512) {
1941 CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1942 oh->ot_credits, osd_transaction_size(dev));
1943 osd_trans_dump_creds(env, th);
1944 libcfs_debug_dumpstack(NULL);
1945 last_credits = oh->ot_credits;
1946 last_printed = jiffies;
1949 * XXX Limit the credits to 'max_transaction_buffers', and
1950 * let the underlying filesystem to catch the error if
1951 * we really need so many credits.
1953 * This should be removed when we can calculate the
1954 * credits precisely.
1956 oh->ot_credits = osd_transaction_size(dev);
1957 } else if (ldiskfs_track_declares_assert != 0) {
1959 * reserve few credits to prevent an assertion in JBD
1960 * our debugging mechanism will be able to detected
1961 * overuse. this can help to debug single-update
1964 oh->ot_credits += 10;
1965 if (unlikely(osd_param_is_not_sane(dev, th)))
1966 oh->ot_credits = osd_transaction_size(dev);
1969 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1970 GOTO(out, rc = -EIO);
1973 * XXX temporary stuff. Some abstraction layer should
1976 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1980 LASSERT(oti->oti_txns == 0);
1982 atomic_inc(&dev->od_commit_cb_in_flight);
1983 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1994 static int osd_seq_exists(const struct lu_env *env,
1995 struct osd_device *osd, u64 seq)
1997 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1998 struct seq_server_site *ss = osd_seq_site(osd);
2003 LASSERT(ss != NULL);
2004 LASSERT(ss->ss_server_fld != NULL);
2006 rc = osd_fld_lookup(env, osd, seq, range);
2009 CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
2010 osd_name(osd), seq, rc);
2014 RETURN(ss->ss_node_id == range->lsr_index);
2017 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
2019 struct dt_txn_commit_cb *dcb;
2020 struct dt_txn_commit_cb *tmp;
2022 /* call per-transaction stop callbacks if any */
2023 list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
2025 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
2026 "commit callback entry: magic=%x name='%s'\n",
2027 dcb->dcb_magic, dcb->dcb_name);
2028 list_del_init(&dcb->dcb_linkage);
2029 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
2034 * Concurrency: shouldn't matter.
2036 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
2039 struct osd_thread_info *oti = osd_oti_get(env);
2040 struct osd_thandle *oh;
2041 struct osd_iobuf *iobuf = &oti->oti_iobuf;
2042 struct osd_device *osd = osd_dt_dev(th->th_dev);
2043 struct qsd_instance *qsd = osd_def_qsd(osd);
2044 struct lquota_trans *qtrans;
2045 LIST_HEAD(truncates);
2046 int rc = 0, remove_agents = 0;
2050 oh = container_of(th, struct osd_thandle, ot_super);
2052 remove_agents = oh->ot_remove_agents;
2054 qtrans = oh->ot_quota_trans;
2055 oh->ot_quota_trans = NULL;
2057 /* move locks to local list, stop tx, execute truncates */
2058 list_splice(&oh->ot_trunc_locks, &truncates);
2060 if (oh->ot_handle != NULL) {
2063 handle_t *hdl = oh->ot_handle;
2066 * add commit callback
2067 * notice we don't do this in osd_trans_start()
2068 * as underlying transaction can change during truncate
2070 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
2073 LASSERT(oti->oti_txns == 1);
2076 rc = dt_txn_hook_stop(env, th);
2078 CERROR("%s: failed in transaction hook: rc = %d\n",
2081 osd_trans_stop_cb(oh, rc);
2082 /* hook functions might modify th_sync */
2083 hdl->h_sync = th->th_sync;
2085 oh->ot_handle = NULL;
2086 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2088 CERROR("%s: failed to stop transaction: rc = %d\n",
2089 osd_name(osd), rc2);
2093 osd_process_truncates(env, &truncates);
2095 osd_trans_stop_cb(oh, th->th_result);
2099 osd_trunc_unlock_all(env, &truncates);
2101 /* inform the quota slave device that the transaction is stopping */
2102 qsd_op_end(env, qsd, qtrans);
2105 * as we want IO to journal and data IO be concurrent, we don't block
2106 * awaiting data IO completion in osd_do_bio(), instead we wait here
2107 * once transaction is submitted to the journal. all reqular requests
2108 * don't do direct IO (except read/write), thus this wait_event becomes
2111 * IMPORTANT: we have to wait till any IO submited by the thread is
2112 * completed otherwise iobuf may be corrupted by different request
2114 wait_event(iobuf->dr_wait,
2115 atomic_read(&iobuf->dr_numreqs) == 0);
2116 osd_fini_iobuf(osd, iobuf);
2118 rc = iobuf->dr_error;
2120 if (unlikely(remove_agents != 0))
2121 osd_process_scheduled_agent_removals(env, osd);
2123 LASSERT(oti->oti_ins_cache_depth > 0);
2124 oti->oti_ins_cache_depth--;
2125 /* reset OI cache for safety */
2126 if (oti->oti_ins_cache_depth == 0)
2127 oti->oti_ins_cache_used = 0;
2129 sb_end_write(osd_sb(osd));
2134 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2136 struct osd_thandle *oh = container_of(th, struct osd_thandle,
2139 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2140 LASSERT(&dcb->dcb_func != NULL);
2141 if (dcb->dcb_flags & DCB_TRANS_STOP)
2142 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2144 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2150 * Called just before object is freed. Releases all resources except for
2151 * object itself (that is released by osd_object_free()).
2153 * Concurrency: no concurrent access is possible that late in object
2156 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2158 struct osd_object *obj = osd_obj(l);
2159 struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2160 struct inode *inode = obj->oo_inode;
2165 LINVRNT(osd_invariant(obj));
2168 * If object is unlinked remove fid->ino mapping from object index.
2171 osd_index_fini(obj);
2176 if (osd_has_index(obj) && obj->oo_dt.do_index_ops == &osd_index_iam_ops)
2177 ldiskfs_set_inode_flag(inode, LDISKFS_INODE_JOURNAL_DATA);
2179 uid = i_uid_read(inode);
2180 gid = i_gid_read(inode);
2181 projid = i_projid_read(inode);
2183 obj->oo_inode = NULL;
2186 /* do not rebalance quota if the caller needs to release memory
2187 * otherwise qsd_refresh_usage() may went into a new ldiskfs
2188 * transaction and risk to deadlock - LU-12178 */
2189 if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2192 if (!obj->oo_header && qsd) {
2193 struct osd_thread_info *info = osd_oti_get(env);
2194 struct lquota_id_info *qi = &info->oti_qi;
2196 /* Release granted quota to master if necessary */
2197 qi->lqi_id.qid_uid = uid;
2198 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2200 qi->lqi_id.qid_uid = gid;
2201 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2203 qi->lqi_id.qid_uid = projid;
2204 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2209 * Concurrency: ->loo_object_release() is called under site spin-lock.
2211 static void osd_object_release(const struct lu_env *env,
2212 struct lu_object *l)
2214 struct osd_object *o = osd_obj(l);
2217 * nobody should be releasing a non-destroyed object with nlink=0
2218 * the API allows this, but ldiskfs doesn't like and then report
2219 * this inode as deleted
2221 LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2222 o->oo_inode->i_nlink == 0));
2226 * Concurrency: shouldn't matter.
2228 static int osd_object_print(const struct lu_env *env, void *cookie,
2229 lu_printer_t p, const struct lu_object *l)
2231 struct osd_object *o = osd_obj(l);
2232 struct iam_descr *d;
2234 if (o->oo_dir != NULL)
2235 d = o->oo_dir->od_container.ic_descr;
2238 return (*p)(env, cookie,
2239 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2241 o->oo_inode ? o->oo_inode->i_ino : 0UL,
2242 o->oo_inode ? o->oo_inode->i_generation : 0,
2243 d ? d->id_ops->id_name : "plain");
2247 * Concurrency: shouldn't matter.
2249 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2250 struct obd_statfs *sfs, struct obd_statfs_info *info)
2252 struct osd_device *osd = osd_dt_dev(d);
2253 struct super_block *sb = osd_sb(osd);
2254 struct kstatfs *ksfs;
2258 if (unlikely(osd->od_mnt == NULL))
2259 return -EINPROGRESS;
2261 /* osd_lproc.c call this without env, allocate ksfs for that case */
2262 if (unlikely(env == NULL)) {
2263 OBD_ALLOC_PTR(ksfs);
2267 ksfs = &osd_oti_get(env)->oti_ksfs;
2270 result = sb->s_op->statfs(sb->s_root, ksfs);
2274 statfs_pack(sfs, ksfs);
2275 if (unlikely(sb->s_flags & SB_RDONLY))
2276 sfs->os_state |= OS_STATFS_READONLY;
2278 sfs->os_state |= osd->od_nonrotational ? OS_STATFS_NONROT : 0;
2280 if (ldiskfs_has_feature_extents(sb))
2281 sfs->os_maxbytes = sb->s_maxbytes;
2283 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2286 * Reserve some space so to avoid fragmenting the filesystem too much.
2287 * Fragmentation not only impacts performance, but can also increase
2288 * metadata overhead significantly, causing grant calculation to be
2291 * Reserve 0.78% of total space, at least 8MB for small filesystems.
2293 BUILD_BUG_ON(OSD_STATFS_RESERVED <= LDISKFS_MAX_BLOCK_SIZE);
2294 reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2295 if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2296 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2298 sfs->os_blocks -= reserved;
2299 sfs->os_bfree -= min(reserved, sfs->os_bfree);
2300 sfs->os_bavail -= min(reserved, sfs->os_bavail);
2303 if (unlikely(env == NULL))
2309 * Estimate space needed for file creations. We assume the largest filename
2310 * which is 2^64 - 1, hence a filename of 20 chars.
2311 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2313 #ifdef __LDISKFS_DIR_REC_LEN
2314 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2316 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2320 * Concurrency: doesn't access mutable data.
2322 static void osd_conf_get(const struct lu_env *env,
2323 const struct dt_device *dev,
2324 struct dt_device_param *param)
2326 struct osd_device *d = osd_dt_dev(dev);
2327 struct super_block *sb = osd_sb(d);
2328 struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2333 * XXX should be taken from not-yet-existing fs abstraction layer.
2335 param->ddp_max_name_len = LDISKFS_NAME_LEN;
2336 param->ddp_max_nlink = LDISKFS_LINK_MAX;
2337 param->ddp_symlink_max = sb->s_blocksize;
2338 param->ddp_mount_type = LDD_MT_LDISKFS;
2339 if (ldiskfs_has_feature_extents(sb))
2340 param->ddp_maxbytes = sb->s_maxbytes;
2342 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2344 * inode are statically allocated, so per-inode space consumption
2345 * is the space consumed by the directory entry
2347 param->ddp_inodespace = PER_OBJ_USAGE;
2349 * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2350 * is 128MB) which is unlikely to be hit in real life. Report a smaller
2351 * maximum length to not under-count the actual number of extents
2352 * needed for writing a file if there are sub-optimal block allocations.
2354 param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2355 /* worst-case extent insertion metadata overhead */
2356 param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2357 param->ddp_mntopts = 0;
2358 if (test_opt(sb, XATTR_USER))
2359 param->ddp_mntopts |= MNTOPT_USERXATTR;
2360 if (test_opt(sb, POSIX_ACL))
2361 param->ddp_mntopts |= MNTOPT_ACL;
2364 * LOD might calculate the max stripe count based on max_ea_size,
2365 * so we need take account in the overhead as well,
2366 * xattr_header + magic + xattr_entry_head
2368 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2369 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2371 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2372 if (ldiskfs_has_feature_ea_inode(sb))
2373 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2377 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2379 if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE)
2380 param->ddp_max_ea_size = OBD_MAX_EA_SIZE;
2383 * Preferred RPC size for efficient disk IO. 4MB shows good
2384 * all-around performance for ldiskfs, but use bigalloc chunk size
2385 * by default if larger.
2387 #if defined(LDISKFS_CLUSTER_SIZE)
2388 if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2389 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2392 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2394 param->ddp_t10_cksum_type = 0;
2396 unsigned short interval = blk_integrity_interval(bi);
2397 name = blk_integrity_name(bi);
2405 if (strncmp(name, "T10-DIF-TYPE",
2406 sizeof("T10-DIF-TYPE") - 1) == 0) {
2407 /* also skip "1/3-" at end */
2408 const int type_off = sizeof("T10-DIF-TYPE.");
2409 char type_number = name[type_off - 2];
2411 if (interval != 512 && interval != 4096) {
2412 CERROR("%s: unsupported T10PI sector size %u\n",
2413 d->od_svname, interval);
2414 } else if (type_number != '1' && type_number != '3') {
2415 CERROR("%s: unsupported T10PI type %s\n",
2416 d->od_svname, name);
2417 } else if (strcmp(name + type_off, "CRC") == 0) {
2418 d->od_t10_type = type_number == '1' ?
2419 OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2420 param->ddp_t10_cksum_type = interval == 512 ?
2421 OBD_CKSUM_T10CRC512 :
2423 } else if (strcmp(name + type_off, "IP") == 0) {
2424 d->od_t10_type = type_number == '1' ?
2425 OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2426 param->ddp_t10_cksum_type = interval == 512 ?
2427 OBD_CKSUM_T10IP512 :
2430 CERROR("%s: unsupported checksum type of T10PI type '%s'\n",
2431 d->od_svname, name);
2435 CERROR("%s: unsupported T10PI type '%s'\n",
2436 d->od_svname, name);
2440 param->ddp_has_lseek_data_hole = true;
2443 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2445 return osd_sb(osd_dt_dev(d));
2449 * Concurrency: shouldn't matter.
2451 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2454 struct super_block *s = osd_sb(osd_dt_dev(d));
2457 down_read(&s->s_umount);
2458 rc = s->s_op->sync_fs(s, 1);
2459 up_read(&s->s_umount);
2461 CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2468 * Start commit for OSD device.
2470 * An implementation of dt_commit_async method for OSD device.
2471 * Asychronously starts underlayng fs sync and thereby a transaction
2474 * \param env environment
2475 * \param d dt device
2477 * \see dt_device_operations
2479 static int osd_commit_async(const struct lu_env *env,
2480 struct dt_device *d)
2482 struct super_block *s = osd_sb(osd_dt_dev(d));
2487 CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2488 down_read(&s->s_umount);
2489 rc = s->s_op->sync_fs(s, 0);
2490 up_read(&s->s_umount);
2495 static int (*priv_security_file_alloc)(struct file *file);
2497 int osd_security_file_alloc(struct file *file)
2499 if (priv_security_file_alloc)
2500 return priv_security_file_alloc(file);
2505 * Concurrency: shouldn't matter.
2507 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2509 struct super_block *sb = osd_sb(osd_dt_dev(d));
2510 struct block_device *dev = sb->s_bdev;
2511 int rc = -EOPNOTSUPP;
2515 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2516 osd_dt_dev(d)->od_svname, (long)dev, rc);
2522 * Note: we do not count into QUOTA here.
2523 * If we mount with --data_journal we may need more.
2525 const int osd_dto_credits_noquota[DTO_NR] = {
2528 * INDEX_EXTRA_TRANS_BLOCKS(8) +
2529 * SINGLEDATA_TRANS_BLOCKS(8)
2530 * XXX Note: maybe iam need more, since iam have more level than
2533 [DTO_INDEX_INSERT] = 16,
2536 * just modify a single entry, probably merge few within a block
2538 [DTO_INDEX_DELETE] = 1,
2542 [DTO_INDEX_UPDATE] = 16,
2544 * 4(inode, inode bits, groups, GDT)
2545 * notice: OI updates are counted separately with DTO_INDEX_INSERT
2547 [DTO_OBJECT_CREATE] = 4,
2549 * 4(inode, inode bits, groups, GDT)
2550 * notice: OI updates are counted separately with DTO_INDEX_DELETE
2552 [DTO_OBJECT_DELETE] = 4,
2554 * Attr set credits (inode)
2556 [DTO_ATTR_SET_BASE] = 1,
2558 * Xattr set. The same as xattr of EXT3.
2559 * DATA_TRANS_BLOCKS(14)
2560 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2561 * are also counted in. Do not know why?
2563 [DTO_XATTR_SET] = 14,
2565 * credits for inode change during write.
2567 [DTO_WRITE_BASE] = 3,
2569 * credits for single block write.
2571 [DTO_WRITE_BLOCK] = 14,
2573 * Attr set credits for chown.
2574 * This is extra credits for setattr, and it is null without quota
2576 [DTO_ATTR_SET_CHOWN] = 0
2579 /* reserve or free quota for some operation */
2580 static int osd_reserve_or_free_quota(const struct lu_env *env,
2581 struct dt_device *dev,
2582 enum quota_type type, __u64 uid,
2583 __u64 gid, __s64 count, bool is_md)
2586 struct osd_device *osd = osd_dt_dev(dev);
2587 struct osd_thread_info *info = osd_oti_get(env);
2588 struct lquota_id_info *qi = &info->oti_qi;
2589 struct qsd_instance *qsd = NULL;
2594 qsd = osd->od_quota_slave_md;
2596 qsd = osd->od_quota_slave_dt;
2598 rc = quota_reserve_or_free(env, qsd, qi, type, uid, gid, count, is_md);
2602 static const struct dt_device_operations osd_dt_ops = {
2603 .dt_root_get = osd_root_get,
2604 .dt_statfs = osd_statfs,
2605 .dt_trans_create = osd_trans_create,
2606 .dt_trans_start = osd_trans_start,
2607 .dt_trans_stop = osd_trans_stop,
2608 .dt_trans_cb_add = osd_trans_cb_add,
2609 .dt_conf_get = osd_conf_get,
2610 .dt_mnt_sb_get = osd_mnt_sb_get,
2611 .dt_sync = osd_sync,
2613 .dt_commit_async = osd_commit_async,
2614 .dt_reserve_or_free_quota = osd_reserve_or_free_quota,
2617 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2620 struct osd_object *obj = osd_dt_obj(dt);
2621 struct osd_thread_info *oti = osd_oti_get(env);
2623 LINVRNT(osd_invariant(obj));
2625 LASSERT(obj->oo_owner != env);
2626 down_read_nested(&obj->oo_sem, role);
2628 LASSERT(obj->oo_owner == NULL);
2632 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2635 struct osd_object *obj = osd_dt_obj(dt);
2636 struct osd_thread_info *oti = osd_oti_get(env);
2638 LINVRNT(osd_invariant(obj));
2640 LASSERT(obj->oo_owner != env);
2641 down_write_nested(&obj->oo_sem, role);
2643 LASSERT(obj->oo_owner == NULL);
2644 obj->oo_owner = env;
2648 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2650 struct osd_object *obj = osd_dt_obj(dt);
2651 struct osd_thread_info *oti = osd_oti_get(env);
2653 LINVRNT(osd_invariant(obj));
2655 LASSERT(oti->oti_r_locks > 0);
2657 up_read(&obj->oo_sem);
2660 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2662 struct osd_object *obj = osd_dt_obj(dt);
2663 struct osd_thread_info *oti = osd_oti_get(env);
2665 LINVRNT(osd_invariant(obj));
2667 LASSERT(obj->oo_owner == env);
2668 LASSERT(oti->oti_w_locks > 0);
2670 obj->oo_owner = NULL;
2671 up_write(&obj->oo_sem);
2674 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2676 struct osd_object *obj = osd_dt_obj(dt);
2678 LINVRNT(osd_invariant(obj));
2680 return obj->oo_owner == env;
2683 static void osd_inode_getattr(const struct lu_env *env,
2684 struct inode *inode, struct lu_attr *attr)
2686 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2687 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2688 LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2689 LA_BLKSIZE | LA_TYPE | LA_BTIME;
2691 attr->la_atime = inode->i_atime.tv_sec;
2692 attr->la_mtime = inode->i_mtime.tv_sec;
2693 attr->la_ctime = inode->i_ctime.tv_sec;
2694 attr->la_btime = LDISKFS_I(inode)->i_crtime.tv_sec;
2695 attr->la_mode = inode->i_mode;
2696 attr->la_size = i_size_read(inode);
2697 attr->la_blocks = inode->i_blocks;
2698 attr->la_uid = i_uid_read(inode);
2699 attr->la_gid = i_gid_read(inode);
2700 attr->la_projid = i_projid_read(inode);
2701 attr->la_flags = ll_inode_to_ext_flags(inode->i_flags);
2702 attr->la_nlink = inode->i_nlink;
2703 attr->la_rdev = inode->i_rdev;
2704 attr->la_blksize = 1 << inode->i_blkbits;
2705 attr->la_blkbits = inode->i_blkbits;
2707 * Ext4 did not transfer inherit flags from raw inode
2708 * to inode flags, and ext4 internally test raw inode
2709 * @i_flags directly. Instead of patching ext4, we do it here.
2711 if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2712 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2715 static int osd_dirent_count(const struct lu_env *env, struct dt_object *dt,
2718 struct osd_object *obj = osd_dt_obj(dt);
2719 const struct dt_it_ops *iops;
2725 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2726 LASSERT(fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)));
2728 if (obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
2729 *count = obj->oo_dirent_count;
2733 /* directory not initialized yet */
2734 if (!dt->do_index_ops) {
2739 iops = &dt->do_index_ops->dio_it;
2740 it = iops->init(env, dt, LUDA_64BITHASH);
2742 RETURN(PTR_ERR(it));
2744 rc = iops->load(env, it, 0);
2746 if (rc == -ENODATA) {
2753 rc = iops->next(env, it);
2755 for (*count = 0; rc == 0 || rc == -ESTALE; rc = iops->next(env, it)) {
2759 if (iops->key_size(env, it) == 0)
2765 obj->oo_dirent_count = *count;
2770 iops->fini(env, it);
2775 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2776 struct lu_attr *attr)
2778 struct osd_object *obj = osd_dt_obj(dt);
2781 if (unlikely(!dt_object_exists(dt)))
2783 if (unlikely(obj->oo_destroyed))
2786 LASSERT(!dt_object_remote(dt));
2787 LINVRNT(osd_invariant(obj));
2789 spin_lock(&obj->oo_guard);
2790 osd_inode_getattr(env, obj->oo_inode, attr);
2791 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2792 attr->la_valid |= LA_FLAGS;
2793 attr->la_flags |= LUSTRE_ORPHAN_FL;
2795 if (obj->oo_lma_flags & LUSTRE_ENCRYPT_FL) {
2796 attr->la_valid |= LA_FLAGS;
2797 attr->la_flags |= LUSTRE_ENCRYPT_FL;
2799 spin_unlock(&obj->oo_guard);
2801 if (S_ISDIR(obj->oo_inode->i_mode) &&
2802 fid_is_namespace_visible(lu_object_fid(&dt->do_lu)))
2803 rc = osd_dirent_count(env, dt, &attr->la_dirent_count);
2808 static int osd_declare_attr_qid(const struct lu_env *env,
2809 struct osd_object *obj,
2810 struct osd_thandle *oh, long long bspace,
2811 qid_t old_id, qid_t new_id, bool enforce,
2815 struct osd_thread_info *info = osd_oti_get(env);
2816 struct lquota_id_info *qi = &info->oti_qi;
2818 qi->lqi_type = type;
2819 /* inode accounting */
2820 qi->lqi_is_blk = false;
2822 /* one more inode for the new id ... */
2823 qi->lqi_id.qid_uid = new_id;
2825 /* Reserve credits for the new id */
2826 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2827 if (rc == -EDQUOT || rc == -EINPROGRESS)
2832 /* and one less inode for the current id */
2833 qi->lqi_id.qid_uid = old_id;
2835 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2836 if (rc == -EDQUOT || rc == -EINPROGRESS)
2841 /* block accounting */
2842 qi->lqi_is_blk = true;
2844 /* more blocks for the new id ... */
2845 qi->lqi_id.qid_uid = new_id;
2846 qi->lqi_space = bspace;
2848 * Credits for the new uid has been reserved, re-use "obj"
2849 * to save credit reservation.
2851 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2852 if (rc == -EDQUOT || rc == -EINPROGRESS)
2857 /* and finally less blocks for the current uid */
2858 qi->lqi_id.qid_uid = old_id;
2859 qi->lqi_space = -bspace;
2860 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2861 if (rc == -EDQUOT || rc == -EINPROGRESS)
2867 static int osd_declare_attr_set(const struct lu_env *env,
2868 struct dt_object *dt,
2869 const struct lu_attr *attr,
2870 struct thandle *handle)
2872 struct osd_thandle *oh;
2873 struct osd_object *obj;
2882 LASSERT(dt != NULL);
2883 LASSERT(handle != NULL);
2885 obj = osd_dt_obj(dt);
2886 LASSERT(osd_invariant(obj));
2888 oh = container_of(handle, struct osd_thandle, ot_super);
2889 LASSERT(oh->ot_handle == NULL);
2891 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2892 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2894 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2895 osd_dto_credits_noquota[DTO_XATTR_SET]);
2897 if (attr == NULL || obj->oo_inode == NULL)
2900 bspace = obj->oo_inode->i_blocks << 9;
2901 bspace = toqb(bspace);
2904 * Changing ownership is always preformed by super user, it should not
2905 * fail with EDQUOT unless required explicitly.
2907 * We still need to call the osd_declare_qid() to calculate the journal
2908 * credits for updating quota accounting files and to trigger quota
2909 * space adjustment once the operation is completed.
2911 if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2913 uid = i_uid_read(obj->oo_inode);
2914 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2915 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2916 attr->la_uid, enforce, USRQUOTA);
2920 gid = i_gid_read(obj->oo_inode);
2921 CDEBUG(D_QUOTA, "declare uid %d -> %d gid %d -> %d\n", uid,
2922 attr->la_uid, gid, attr->la_gid);
2923 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2924 rc = osd_declare_attr_qid(env, obj, oh, bspace, gid,
2925 attr->la_gid, enforce, GRPQUOTA);
2930 #ifdef HAVE_PROJECT_QUOTA
2931 if (attr->la_valid & LA_PROJID) {
2932 __u32 projid = i_projid_read(obj->oo_inode);
2934 enforce = (attr->la_valid & LA_PROJID) &&
2935 (attr->la_projid != projid);
2936 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2937 (qid_t)projid, (qid_t)attr->la_projid,
2946 static int osd_inode_setattr(const struct lu_env *env,
2947 struct inode *inode, const struct lu_attr *attr)
2949 __u64 bits = attr->la_valid;
2951 /* Only allow set size for regular file */
2952 if (!S_ISREG(inode->i_mode))
2953 bits &= ~(LA_SIZE | LA_BLOCKS);
2958 if (bits & LA_ATIME)
2959 inode->i_atime = osd_inode_time(inode, attr->la_atime);
2960 if (bits & LA_CTIME)
2961 inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
2962 if (bits & LA_MTIME)
2963 inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
2964 if (bits & LA_SIZE) {
2965 spin_lock(&inode->i_lock);
2966 LDISKFS_I(inode)->i_disksize = attr->la_size;
2967 i_size_write(inode, attr->la_size);
2968 spin_unlock(&inode->i_lock);
2972 * OSD should not change "i_blocks" which is used by quota.
2973 * "i_blocks" should be changed by ldiskfs only.
2976 inode->i_mode = (inode->i_mode & S_IFMT) |
2977 (attr->la_mode & ~S_IFMT);
2979 i_uid_write(inode, attr->la_uid);
2981 i_gid_write(inode, attr->la_gid);
2982 if (bits & LA_PROJID)
2983 i_projid_write(inode, attr->la_projid);
2984 if (bits & LA_NLINK)
2985 set_nlink(inode, attr->la_nlink);
2987 inode->i_rdev = attr->la_rdev;
2989 if (bits & LA_FLAGS) {
2990 /* always keep S_NOCMTIME */
2991 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2993 #if defined(S_ENCRYPTED)
2994 /* Always remove S_ENCRYPTED, because ldiskfs must not be
2995 * aware of encryption status. It is just stored into LMA
2996 * so that it can be forwared to client side.
2998 inode->i_flags &= ~S_ENCRYPTED;
3001 * Ext4 did not transfer inherit flags from
3002 * @inode->i_flags to raw inode i_flags when writing
3003 * flags, we do it explictly here.
3005 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
3006 LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
3008 LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
3013 #ifdef HAVE_PROJECT_QUOTA
3014 static int osd_transfer_project(struct inode *inode, __u32 projid,
3015 struct thandle *handle)
3017 struct super_block *sb = inode->i_sb;
3018 struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
3021 struct ldiskfs_iloc iloc;
3022 struct ldiskfs_inode *raw_inode;
3023 struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
3025 if (!ldiskfs_has_feature_project(sb)) {
3026 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
3027 == LDISKFS_DEF_PROJID);
3028 if (projid != LDISKFS_DEF_PROJID)
3034 if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
3037 kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
3038 if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
3041 err = ldiskfs_get_inode_loc(inode, &iloc);
3045 raw_inode = ldiskfs_raw_inode(&iloc);
3046 if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
3047 struct osd_thandle *oh = container_of(handle,
3051 * try to expand inode size automatically.
3053 ldiskfs_mark_inode_dirty(oh->ot_handle, inode);
3054 if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
3062 dquot_initialize(inode);
3063 transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
3064 if (transfer_to[PRJQUOTA]) {
3065 lock_dquot_transfer(inode);
3066 err = __dquot_transfer(inode, transfer_to);
3067 unlock_dquot_transfer(inode);
3068 dqput(transfer_to[PRJQUOTA]);
3077 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr,
3078 struct thandle *handle)
3082 if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
3083 (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
3087 "executing dquot_transfer inode %ld uid %d -> %d gid %d -> %d\n",
3088 inode->i_ino, i_uid_read(inode), attr->la_uid,
3089 i_gid_read(inode), attr->la_gid);
3091 dquot_initialize(inode);
3093 if (attr->la_valid & LA_UID)
3094 iattr.ia_valid |= ATTR_UID;
3095 if (attr->la_valid & LA_GID)
3096 iattr.ia_valid |= ATTR_GID;
3097 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
3098 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
3100 lock_dquot_transfer(inode);
3101 rc = dquot_transfer(inode, &iattr);
3102 unlock_dquot_transfer(inode);
3104 CERROR("%s: quota transfer failed. Is quota enforcement enabled on the ldiskfs filesystem? rc = %d\n",
3105 osd_ino2name(inode), rc);
3110 /* Handle project id transfer here properly */
3111 if (attr->la_valid & LA_PROJID &&
3112 attr->la_projid != i_projid_read(inode)) {
3113 if (!projid_valid(make_kprojid(&init_user_ns, attr->la_projid)))
3115 #ifdef HAVE_PROJECT_QUOTA
3116 rc = osd_transfer_project(inode, attr->la_projid, handle);
3121 CERROR("%s: quota transfer failed. Is project enforcement enabled on the ldiskfs filesystem? rc = %d\n",
3122 osd_ino2name(inode), rc);
3129 static int osd_attr_set(const struct lu_env *env,
3130 struct dt_object *dt,
3131 const struct lu_attr *attr,
3132 struct thandle *handle)
3134 struct osd_object *obj = osd_dt_obj(dt);
3135 struct inode *inode;
3138 if (!dt_object_exists(dt))
3141 LASSERT(handle != NULL);
3142 LASSERT(!dt_object_remote(dt));
3143 LASSERT(osd_invariant(obj));
3145 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
3147 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
3148 !osd_obj2dev(obj)->od_is_ost) {
3149 struct osd_thread_info *oti = osd_oti_get(env);
3150 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
3151 struct lu_fid *fid1 = &oti->oti_fid;
3152 struct osd_inode_id *id = &oti->oti_id;
3153 struct iam_path_descr *ipd;
3154 struct iam_container *bag;
3155 struct osd_thandle *oh;
3158 fid_cpu_to_be(fid1, fid0);
3159 memset(id, 1, sizeof(*id));
3160 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
3161 fid0)->oi_dir.od_container;
3162 ipd = osd_idx_ipd_get(env, bag);
3163 if (unlikely(ipd == NULL))
3166 oh = container_of(handle, struct osd_thandle, ot_super);
3167 rc = iam_update(oh->ot_handle, bag,
3168 (const struct iam_key *)fid1,
3169 (const struct iam_rec *)id, ipd);
3170 osd_ipd_put(env, bag, ipd);
3171 return(rc > 0 ? 0 : rc);
3174 inode = obj->oo_inode;
3176 rc = osd_quota_transfer(inode, attr, handle);
3180 spin_lock(&obj->oo_guard);
3181 rc = osd_inode_setattr(env, inode, attr);
3182 spin_unlock(&obj->oo_guard);
3186 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3188 osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3190 if (!(attr->la_valid & LA_FLAGS))
3193 /* Let's check if there are extra flags need to be set into LMA */
3194 if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3195 struct osd_thread_info *info = osd_oti_get(env);
3196 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3198 LASSERT(!obj->oo_pfid_in_lma);
3200 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3201 &info->oti_ost_attrs);
3205 lma->lma_incompat |=
3206 lustre_to_lma_flags(attr->la_flags);
3207 lustre_lma_swab(lma);
3209 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3211 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3212 lma, sizeof(*lma), XATTR_REPLACE);
3214 struct osd_device *osd = osd_obj2dev(obj);
3216 CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3217 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3218 lma->lma_incompat, rc);
3221 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3223 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3230 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3231 struct osd_object *obj,
3232 const char *name, const int namelen)
3234 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3237 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3238 umode_t mode, struct dt_allocation_hint *hint,
3239 struct thandle *th, struct lu_attr *attr)
3242 struct osd_device *osd = osd_obj2dev(obj);
3243 struct osd_thandle *oth;
3244 struct dt_object *parent = NULL;
3245 struct inode *inode;
3246 struct iattr iattr = {
3247 .ia_valid = ATTR_UID | ATTR_GID |
3248 ATTR_CTIME | ATTR_MTIME | ATTR_ATIME,
3249 .ia_ctime.tv_sec = attr->la_ctime,
3250 .ia_mtime.tv_sec = attr->la_mtime,
3251 .ia_atime.tv_sec = attr->la_atime,
3252 .ia_uid = GLOBAL_ROOT_UID,
3253 .ia_gid = GLOBAL_ROOT_GID,
3255 const struct osd_timespec omit = { .tv_nsec = UTIME_OMIT };
3257 if (attr->la_valid & LA_UID)
3258 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
3259 if (attr->la_valid & LA_GID)
3260 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
3262 LINVRNT(osd_invariant(obj));
3263 LASSERT(obj->oo_inode == NULL);
3264 LASSERT(obj->oo_hl_head == NULL);
3266 if (S_ISDIR(mode) && ldiskfs_pdo) {
3268 ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3269 if (obj->oo_hl_head == NULL)
3273 oth = container_of(th, struct osd_thandle, ot_super);
3274 LASSERT(oth->ot_handle->h_transaction != NULL);
3276 if (hint != NULL && hint->dah_parent != NULL &&
3277 !dt_object_remote(hint->dah_parent))
3278 parent = hint->dah_parent;
3280 /* if a time component is not valid set it to UTIME_OMIT */
3281 if (!(attr->la_valid & LA_CTIME))
3282 iattr.ia_ctime = omit;
3283 if (!(attr->la_valid & LA_MTIME))
3284 iattr.ia_mtime = omit;
3285 if (!(attr->la_valid & LA_ATIME))
3286 iattr.ia_atime = omit;
3288 inode = ldiskfs_create_inode(oth->ot_handle,
3289 parent ? osd_dt_obj(parent)->oo_inode :
3290 osd_sb(osd)->s_root->d_inode,
3292 if (!IS_ERR(inode)) {
3293 /* Do not update file c/mtime in ldiskfs. */
3294 inode->i_flags |= S_NOCMTIME;
3297 * For new created object, it must be consistent,
3298 * and it is unnecessary to scrub against it.
3300 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3302 obj->oo_inode = inode;
3305 if (obj->oo_hl_head != NULL) {
3306 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3307 obj->oo_hl_head = NULL;
3309 result = PTR_ERR(inode);
3311 LINVRNT(osd_invariant(obj));
3319 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3320 struct lu_attr *attr,
3321 struct dt_allocation_hint *hint,
3322 struct dt_object_format *dof,
3326 struct osd_thandle *oth;
3327 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3329 LASSERT(S_ISDIR(attr->la_mode));
3331 oth = container_of(th, struct osd_thandle, ot_super);
3332 LASSERT(oth->ot_handle->h_transaction != NULL);
3333 if (fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)))
3334 obj->oo_dirent_count = 0;
3335 result = osd_mkfile(info, obj, mode, hint, th, attr);
3340 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3341 struct lu_attr *attr,
3342 struct dt_allocation_hint *hint,
3343 struct dt_object_format *dof,
3347 struct osd_thandle *oth;
3348 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3350 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3352 LASSERT(S_ISREG(attr->la_mode));
3354 oth = container_of(th, struct osd_thandle, ot_super);
3355 LASSERT(oth->ot_handle->h_transaction != NULL);
3357 result = osd_mkfile(info, obj, mode, hint, th, attr);
3359 LASSERT(obj->oo_inode != NULL);
3360 if (feat->dif_flags & DT_IND_VARKEY)
3361 result = iam_lvar_create(obj->oo_inode,
3362 feat->dif_keysize_max,
3364 feat->dif_recsize_max,
3367 result = iam_lfix_create(obj->oo_inode,
3368 feat->dif_keysize_max,
3370 feat->dif_recsize_max,
3376 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3377 struct lu_attr *attr,
3378 struct dt_allocation_hint *hint,
3379 struct dt_object_format *dof,
3382 LASSERT(S_ISREG(attr->la_mode));
3383 return osd_mkfile(info, obj, (attr->la_mode &
3384 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3388 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3389 struct lu_attr *attr,
3390 struct dt_allocation_hint *hint,
3391 struct dt_object_format *dof,
3394 LASSERT(S_ISLNK(attr->la_mode));
3395 return osd_mkfile(info, obj, (attr->la_mode &
3396 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3400 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3401 struct lu_attr *attr,
3402 struct dt_allocation_hint *hint,
3403 struct dt_object_format *dof,
3406 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3409 LINVRNT(osd_invariant(obj));
3410 LASSERT(obj->oo_inode == NULL);
3411 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3412 S_ISFIFO(mode) || S_ISSOCK(mode));
3414 result = osd_mkfile(info, obj, mode, hint, th, attr);
3416 LASSERT(obj->oo_inode != NULL);
3418 * This inode should be marked dirty for i_rdev. Currently
3419 * that is done in the osd_attr_init().
3421 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3424 LINVRNT(osd_invariant(obj));
3428 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3430 struct dt_allocation_hint *hint,
3431 struct dt_object_format *dof,
3434 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3436 osd_obj_type_f result;
3452 result = osd_mk_index;
3462 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3463 struct dt_object *parent, struct dt_object *child,
3468 ah->dah_parent = parent;
3469 ah->dah_mode = child_mode;
3471 if (parent != NULL && !dt_object_remote(parent)) {
3472 /* will help to find FID->ino at dt_insert("..") */
3473 struct osd_object *pobj = osd_dt_obj(parent);
3475 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3479 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3480 struct lu_attr *attr, struct dt_object_format *dof,
3481 struct thandle *handle)
3483 struct inode *inode = obj->oo_inode;
3484 __u64 valid = attr->la_valid;
3487 attr->la_valid &= ~(LA_TYPE | LA_MODE);
3489 if (dof->dof_type != DFT_NODE)
3490 attr->la_valid &= ~LA_RDEV;
3491 if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3492 attr->la_valid &= ~LA_ATIME;
3493 if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3494 attr->la_valid &= ~LA_CTIME;
3495 if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3496 attr->la_valid &= ~LA_MTIME;
3498 result = osd_quota_transfer(inode, attr, handle);
3502 if (attr->la_valid != 0) {
3503 result = osd_inode_setattr(info->oti_env, inode, attr);
3505 * The osd_inode_setattr() should always succeed here. The
3506 * only error that could be returned is EDQUOT when we are
3507 * trying to change the UID or GID of the inode. However, this
3508 * should not happen since quota enforcement is no longer
3509 * enabled on ldiskfs (lquota takes care of it).
3511 LASSERTF(result == 0, "%d\n", result);
3512 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3515 attr->la_valid = valid;
3519 * Helper function for osd_create()
3521 * \retval 0, on success
3523 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3524 struct lu_attr *attr, struct dt_allocation_hint *hint,
3525 struct dt_object_format *dof, struct thandle *th)
3530 osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3532 /* we drop umask so that permissions we pass are not affected */
3533 umask = current->fs->umask;
3534 current->fs->umask = 0;
3536 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3538 if (likely(obj->oo_inode != NULL)) {
3539 LASSERT(obj->oo_inode->i_state & I_NEW);
3542 * Unlock the inode before attr initialization to avoid
3543 * unnecessary dqget operations. LU-6378
3545 unlock_new_inode(obj->oo_inode);
3548 if (likely(result == 0)) {
3549 osd_attr_init(info, obj, attr, dof, th);
3550 osd_object_init0(obj);
3553 /* restore previous umask value */
3554 current->fs->umask = umask;
3556 osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3562 * Helper function for osd_create()
3564 * \retval 0, on success
3566 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3567 const struct lu_fid *fid, struct thandle *th)
3569 struct osd_thread_info *info = osd_oti_get(env);
3570 struct osd_inode_id *id = &info->oti_id;
3571 struct osd_device *osd = osd_obj2dev(obj);
3572 struct osd_thandle *oh;
3575 LASSERT(obj->oo_inode != NULL);
3577 if (CFS_FAIL_CHECK(OBD_FAIL_OSD_OI_ENOSPC))
3580 oh = container_of(th, struct osd_thandle, ot_super);
3581 LASSERT(oh->ot_handle);
3582 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3584 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3585 rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3586 OI_CHECK_FLD, NULL);
3587 if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP) && osd->od_is_ost) {
3588 struct lu_fid next_fid = *fid;
3590 /* insert next object in advance, and map to the same inode */
3592 if (next_fid.f_oid != 0) {
3593 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3594 osd_oi_insert(info, osd, &next_fid, id, oh->ot_handle,
3595 OI_CHECK_FLD, NULL);
3596 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3600 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3605 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3606 u64 seq, struct lu_seq_range *range)
3608 struct seq_server_site *ss = osd_seq_site(osd);
3610 if (fid_seq_is_idif(seq)) {
3611 fld_range_set_ost(range);
3612 range->lsr_index = idif_ost_idx(seq);
3616 if (!fid_seq_in_fldb(seq)) {
3617 fld_range_set_mdt(range);
3620 * FIXME: If ss is NULL, it suppose not get lsr_index
3623 range->lsr_index = ss->ss_node_id;
3627 LASSERT(ss != NULL);
3628 fld_range_set_any(range);
3629 /* OSD will only do local fld lookup */
3630 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3633 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3634 struct lu_attr *attr,
3635 struct dt_allocation_hint *hint,
3636 struct dt_object_format *dof,
3637 struct thandle *handle)
3639 struct osd_thandle *oh;
3640 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3646 LASSERT(handle != NULL);
3648 oh = container_of(handle, struct osd_thandle, ot_super);
3649 LASSERT(oh->ot_handle == NULL);
3652 * EA object consumes more credits than regular object: osd_mk_index
3653 * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3654 * leaf_node, could involves the block, block bitmap, groups, GDT
3655 * change for each block, so add 4 * 2 credits in that case.
3657 * The default ACL initialization may consume an additional 16 blocks
3659 credits = osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3660 ((dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3663 * While ldiskfs_new_inode() calls ldiskfs_init_acl() we have to add
3664 * credits for possible default ACL creation in new inode
3666 if (hint && hint->dah_acl_len)
3667 credits += osd_calc_bkmap_credits(sb, NULL, 0, -1,
3668 (hint->dah_acl_len + sb->s_blocksize - 1) >>
3669 sb->s_blocksize_bits);
3671 osd_trans_declare_op(env, oh, OSD_OT_CREATE, credits);
3674 * Reuse idle OI block may cause additional one OI block
3677 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3678 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3679 if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP))
3680 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3681 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3683 /* will help to find FID->ino mapping at dt_insert() */
3684 rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3692 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3693 attr->la_projid, 1, oh, osd_dt_obj(dt),
3694 NULL, OSD_QID_INODE);
3702 * Called to destroy on-disk representation of the object
3704 * Concurrency: must be locked
3706 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3709 struct osd_object *obj = osd_dt_obj(dt);
3710 struct inode *inode = obj->oo_inode;
3711 struct osd_thandle *oh;
3719 oh = container_of(th, struct osd_thandle, ot_super);
3720 LASSERT(oh->ot_handle == NULL);
3722 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3723 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3725 /* For removing agent entry */
3726 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3727 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3730 * Recycle idle OI leaf may cause additional three OI blocks
3733 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3734 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3735 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3736 /* one less inode */
3737 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3738 i_projid_read(inode), -1, oh, obj, NULL,
3742 /* data to be truncated */
3743 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3744 i_projid_read(inode), 0, oh, obj, NULL,
3750 * will help to find FID->ino when this object is being
3753 rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3758 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3761 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3762 struct osd_object *obj = osd_dt_obj(dt);
3763 struct inode *inode = obj->oo_inode;
3764 struct osd_device *osd = osd_obj2dev(obj);
3765 struct osd_thandle *oh;
3770 oh = container_of(th, struct osd_thandle, ot_super);
3771 LASSERT(oh->ot_handle);
3773 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3775 if (unlikely(fid_is_acct(fid)))
3778 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3779 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3781 CERROR("%s: remove agent entry "DFID": rc = %d\n",
3782 osd_name(osd), PFID(fid), result);
3785 if (S_ISDIR(inode->i_mode)) {
3786 if (inode->i_nlink > 2)
3787 CERROR("%s: directory "DFID" ino %lu link count is %u at unlink. run e2fsck to repair\n",
3788 osd_name(osd), PFID(fid), inode->i_ino,
3791 spin_lock(&obj->oo_guard);
3793 spin_unlock(&obj->oo_guard);
3794 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3797 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3799 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3801 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3802 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3803 oh->ot_handle, OI_CHECK_FLD);
3805 osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3806 /* XXX: add to ext3 orphan list */
3807 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3809 /* not needed in the cache anymore */
3810 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3811 obj->oo_destroyed = 1;
3817 * Put the fid into lustre_mdt_attrs, and then place the structure
3818 * inode's ea. This fid should not be altered during the life time
3821 * \retval +ve, on success
3822 * \retval -ve, on error
3824 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3826 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3827 const struct lu_fid *fid, __u32 compat, __u32 incompat)
3829 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3830 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3835 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3838 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3841 lustre_loa_init(loa, fid, compat, incompat);
3842 lustre_loa_swab(loa, false);
3845 * For the OST device with 256 bytes inode size by default,
3846 * the PFID EA will be stored together with LMA EA to avoid
3847 * performance trouble. Otherwise the PFID EA can be stored
3848 * independently. LU-8998
3850 if ((compat & LMAC_FID_ON_OST) &&
3851 LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3852 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3853 sizeof(*loa), XATTR_CREATE);
3855 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3856 sizeof(*lma), XATTR_CREATE);
3858 * LMA may already exist, but we need to check that all the
3859 * desired compat/incompat flags have been added.
3861 if (unlikely(rc == -EEXIST)) {
3862 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3863 XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3867 if (rc < sizeof(*lma))
3870 lustre_loa_swab(loa, true);
3871 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3872 ((compat == 0 && incompat == 0) ||
3873 (!(~lma->lma_compat & compat) &&
3874 !(~lma->lma_incompat & incompat))))
3877 lma->lma_self_fid = *fid;
3878 lma->lma_compat |= compat;
3879 lma->lma_incompat |= incompat;
3880 if (rc == sizeof(*lma)) {
3881 lustre_lma_swab(lma);
3882 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3883 sizeof(*lma), XATTR_REPLACE);
3885 lustre_loa_swab(loa, false);
3886 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3887 sizeof(*loa), XATTR_REPLACE);
3895 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3896 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3897 * To have compatilibility with 1.8 ldiskfs driver we need to have
3898 * magic number at start of fid data.
3899 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3902 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3903 const struct lu_fid *fid)
3905 if (!fid_is_namespace_visible(fid) ||
3906 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3907 param->edp_magic = 0;
3911 param->edp_magic = LDISKFS_LUFID_MAGIC;
3912 param->edp_len = sizeof(struct lu_fid) + 1;
3913 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3917 * Try to read the fid from inode ea into dt_rec.
3919 * \param fid object fid.
3921 * \retval 0 on success
3923 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3924 __u32 ino, struct lu_fid *fid,
3925 struct osd_inode_id *id)
3927 struct osd_thread_info *info = osd_oti_get(env);
3928 struct inode *inode;
3932 osd_id_gen(id, ino, OSD_OII_NOGEN);
3933 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3935 RETURN(PTR_ERR(inode));
3941 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3943 struct inode *parent_dir,
3944 const struct lu_fid *dot_fid,
3945 const struct lu_fid *dot_dot_fid,
3946 struct osd_thandle *oth)
3948 struct ldiskfs_dentry_param *dot_ldp;
3949 struct ldiskfs_dentry_param *dot_dot_ldp;
3950 __u32 saved_nlink = dir->i_nlink;
3953 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_DOTDOT_ENOSPC))
3956 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3957 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3959 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3960 dot_ldp->edp_magic = 0;
3962 rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3963 dir, dot_ldp, dot_dot_ldp);
3965 * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3966 * the subseqent ref_add() will increase the dir->i_nlink
3967 * as 3. That is incorrect for new created directory.
3969 * It looks like hack, because we want to make the OSD API
3970 * to be order-independent for new created directory object
3971 * between dt_insert(..) and ref_add() operations.
3973 * Here, we only restore the in-RAM dir-inode's nlink attr,
3974 * becuase if the nlink attr is not 2, then there will be
3975 * ref_add() called following the dt_insert(..), such call
3976 * will make both the in-RAM and on-disk dir-inode's nlink
3977 * attr to be set as 2. LU-7447
3979 set_nlink(dir, saved_nlink);
3984 * Create an local agent inode for remote entry
3986 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3987 struct osd_device *osd,
3988 struct osd_object *pobj,
3989 const struct lu_fid *fid,
3993 struct osd_thread_info *info = osd_oti_get(env);
3994 struct inode *local;
3995 struct osd_thandle *oh;
3996 struct iattr iattr = {
3997 .ia_valid = ATTR_UID | ATTR_GID |
3998 ATTR_CTIME | ATTR_MTIME | ATTR_ATIME,
3999 .ia_ctime.tv_nsec = UTIME_OMIT,
4000 .ia_mtime.tv_nsec = UTIME_OMIT,
4001 .ia_atime.tv_nsec = UTIME_OMIT,
4002 .ia_uid = GLOBAL_ROOT_UID,
4003 .ia_gid = GLOBAL_ROOT_GID,
4010 oh = container_of(th, struct osd_thandle, ot_super);
4011 LASSERT(oh->ot_handle->h_transaction != NULL);
4013 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode,
4015 if (IS_ERR(local)) {
4016 CERROR("%s: create local error %d\n", osd_name(osd),
4017 (int)PTR_ERR(local));
4022 * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
4023 * correct gid on remote file, not agent here
4025 local->i_gid = current_fsgid();
4026 ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
4028 /* e2fsck doesn't like empty symlinks. Store remote FID as symlink.
4029 * That gives e2fsck something to look at and be happy, and allows
4030 * debugging if we need to determine where this symlink came from.
4032 if (S_ISLNK(type)) {
4033 BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1);
4034 rc = scnprintf((char *)LDISKFS_I(local)->i_data,
4035 LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
4037 i_size_write(local, rc);
4038 LDISKFS_I(local)->i_disksize = rc;
4040 unlock_new_inode(local);
4042 /* Agent inode should not have project ID */
4043 #ifdef HAVE_PROJECT_QUOTA
4044 if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
4045 i_projid_read(pobj->oo_inode) != 0) {
4046 rc = osd_transfer_project(local, 0, th);
4048 CERROR("%s: quota transfer failed:. Is project quota enforcement enabled on the ldiskfs filesystem? rc = %d\n",
4049 osd_ino2name(local), rc);
4050 RETURN(ERR_PTR(rc));
4054 /* Set special LMA flag for local agent inode */
4055 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
4057 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
4058 osd_name(osd), PFID(fid), rc);
4059 RETURN(ERR_PTR(rc));
4065 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
4066 lu_object_fid(&pobj->oo_dt.do_lu),
4069 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
4070 osd_name(osd), PFID(fid), rc);
4071 RETURN(ERR_PTR(rc));
4078 * when direntry is deleted, we have to take care of possible agent inode
4079 * referenced by that. unfortunately we can't do this at that point:
4080 * iget() within a running transaction leads to deadlock and we better do
4081 * not call that every delete declaration to save performance. so we put
4082 * a potention agent inode on a list and process that once the transaction
4083 * is over. Notice it's not any worse in terms of real orphans as regular
4084 * object destroy doesn't put inodes on the on-disk orphan list. this should
4085 * be addressed separately
4087 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
4088 struct osd_thandle *oh,
4091 struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
4092 struct osd_obj_orphan *oor;
4099 oor->oor_env = (struct lu_env *)env;
4100 spin_lock(&osd->od_osfs_lock);
4101 list_add(&oor->oor_list, &osd->od_orphan_list);
4102 spin_unlock(&osd->od_osfs_lock);
4104 oh->ot_remove_agents = 1;
4110 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
4111 struct osd_device *osd)
4113 struct osd_thread_info *info = osd_oti_get(env);
4114 struct osd_obj_orphan *oor, *tmp;
4115 struct osd_inode_id id;
4117 struct inode *inode;
4122 spin_lock(&osd->od_osfs_lock);
4123 list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
4124 if (oor->oor_env == env)
4125 list_move(&oor->oor_list, &list);
4127 spin_unlock(&osd->od_osfs_lock);
4129 list_for_each_entry_safe(oor, tmp, &list, oor_list) {
4133 list_del(&oor->oor_list);
4136 osd_id_gen(&id, ino, OSD_OII_NOGEN);
4137 inode = osd_iget_fid(info, osd, &id, &fid);
4141 if (!osd_remote_fid(env, osd, &fid)) {
4146 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
4148 mark_inode_dirty(inode);
4149 ldiskfs_journal_stop(jh);
4157 * OSD layer object create function for OST objects (b=11826).
4159 * The FID is inserted into inode xattr here.
4161 * \retval 0, on success
4162 * \retval -ve, on error
4164 static int osd_create(const struct lu_env *env, struct dt_object *dt,
4165 struct lu_attr *attr, struct dt_allocation_hint *hint,
4166 struct dt_object_format *dof, struct thandle *th)
4168 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4169 struct osd_object *obj = osd_dt_obj(dt);
4170 struct osd_thread_info *info = osd_oti_get(env);
4171 int result, on_ost = 0;
4175 if (dt_object_exists(dt))
4178 LINVRNT(osd_invariant(obj));
4179 LASSERT(!dt_object_remote(dt));
4180 LASSERT(osd_is_write_locked(env, obj));
4181 LASSERT(th != NULL);
4183 if (unlikely(fid_is_acct(fid)))
4185 * Quota files can't be created from the kernel any more,
4186 * 'tune2fs -O quota' will take care of creating them
4190 result = __osd_create(info, obj, attr, hint, dof, th);
4192 if (fid_is_idif(fid) &&
4193 !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
4194 struct lu_fid *tfid = &info->oti_fid;
4195 struct ost_id *oi = &info->oti_ostid;
4197 fid_to_ostid(fid, oi);
4198 ostid_to_fid(tfid, oi, 0);
4200 result = osd_ea_fid_set(info, obj->oo_inode, tfid,
4201 LMAC_FID_ON_OST, 0);
4203 on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
4205 result = osd_ea_fid_set(info, obj->oo_inode, fid,
4206 on_ost ? LMAC_FID_ON_OST : 0,
4209 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
4210 obj->oo_dt.do_body_ops = &osd_body_ops;
4213 if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY)) {
4214 struct inode *inode = obj->oo_inode;
4216 result = __osd_oi_insert(env, obj, fid, th);
4217 if (result && inode) {
4218 spin_lock(&obj->oo_guard);
4220 spin_unlock(&obj->oo_guard);
4221 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4222 ldiskfs_set_inode_state(inode,
4223 LDISKFS_STATE_LUSTRE_DESTROY);
4225 obj->oo_inode = NULL;
4230 * a small optimization - dt_insert() isn't usually applied
4231 * to OST objects, so we don't need to cache OI mapping for
4234 if (result == 0 && on_ost == 0) {
4235 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4237 result = osd_idc_find_and_init(env, osd, obj);
4238 LASSERT(result == 0);
4241 LASSERT(ergo(result == 0,
4242 dt_object_exists(dt) && !dt_object_remote(dt)));
4243 LINVRNT(osd_invariant(obj));
4247 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4248 struct thandle *handle)
4250 struct osd_thandle *oh;
4253 /* it's possible that object doesn't exist yet */
4254 LASSERT(handle != NULL);
4256 oh = container_of(handle, struct osd_thandle, ot_super);
4257 LASSERT(oh->ot_handle == NULL);
4259 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4260 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4262 rc = osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev),
4269 * Concurrency: @dt is write locked.
4271 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4274 struct osd_object *obj = osd_dt_obj(dt);
4275 struct inode *inode = obj->oo_inode;
4276 struct osd_thandle *oh;
4279 if (!dt_object_exists(dt) || obj->oo_destroyed)
4282 LINVRNT(osd_invariant(obj));
4283 LASSERT(!dt_object_remote(dt));
4284 LASSERT(osd_is_write_locked(env, obj));
4285 LASSERT(th != NULL);
4287 oh = container_of(th, struct osd_thandle, ot_super);
4288 LASSERT(oh->ot_handle != NULL);
4290 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4292 CDEBUG(D_INODE, DFID" increase nlink %d\n",
4293 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4295 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4296 * (65000) subdirectories by storing "1" in i_nlink if the link count
4297 * would otherwise overflow. Directory tranversal tools understand
4298 * that (st_nlink == 1) indicates that the filesystem dose not track
4299 * hard links count on the directory, and will not abort subdirectory
4300 * scanning early once (st_nlink - 2) subdirs have been found.
4302 * This also has to properly handle the case of inodes with nlink == 0
4303 * in case they are being linked into the PENDING directory
4305 spin_lock(&obj->oo_guard);
4306 if (unlikely(inode->i_nlink == 0))
4307 /* inc_nlink from 0 may cause WARN_ON */
4308 set_nlink(inode, 1);
4310 ldiskfs_inc_count(oh->ot_handle, inode);
4311 if (!S_ISDIR(inode->i_mode))
4312 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4314 spin_unlock(&obj->oo_guard);
4316 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4317 LINVRNT(osd_invariant(obj));
4319 osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4324 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4325 struct thandle *handle)
4327 struct osd_thandle *oh;
4329 if (!dt_object_exists(dt))
4332 LASSERT(!dt_object_remote(dt));
4333 LASSERT(handle != NULL);
4335 oh = container_of(handle, struct osd_thandle, ot_super);
4336 LASSERT(oh->ot_handle == NULL);
4338 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4339 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4345 * Concurrency: @dt is write locked.
4347 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4350 struct osd_object *obj = osd_dt_obj(dt);
4351 struct inode *inode = obj->oo_inode;
4352 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4353 struct osd_thandle *oh;
4355 if (!dt_object_exists(dt))
4358 LINVRNT(osd_invariant(obj));
4359 LASSERT(!dt_object_remote(dt));
4360 LASSERT(osd_is_write_locked(env, obj));
4361 LASSERT(th != NULL);
4363 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_REF_DEL))
4366 oh = container_of(th, struct osd_thandle, ot_super);
4367 LASSERT(oh->ot_handle != NULL);
4369 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4371 spin_lock(&obj->oo_guard);
4373 * That can be result of upgrade from old Lustre version and
4374 * applied only to local files. Just skip this ref_del call.
4375 * ext4_unlink() only treats this as a warning, don't LASSERT here.
4377 if (inode->i_nlink == 0) {
4378 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4379 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4380 ", maybe an upgraded file? (LU-3915)\n",
4381 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4382 spin_unlock(&obj->oo_guard);
4386 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4387 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4389 ldiskfs_dec_count(oh->ot_handle, inode);
4390 spin_unlock(&obj->oo_guard);
4392 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4393 LINVRNT(osd_invariant(obj));
4395 osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4401 * Concurrency: @dt is read locked.
4403 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4404 struct lu_buf *buf, const char *name)
4406 struct osd_object *obj = osd_dt_obj(dt);
4407 struct inode *inode = obj->oo_inode;
4408 struct osd_thread_info *info = osd_oti_get(env);
4409 struct dentry *dentry = &info->oti_obj_dentry;
4410 bool cache_xattr = false;
4415 /* version get is not real XATTR but uses xattr API */
4416 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4417 dt_obj_version_t *ver = buf->lb_buf;
4420 * for version we are just using xattr API but change inode
4423 if (buf->lb_len == 0)
4424 return sizeof(dt_obj_version_t);
4426 if (buf->lb_len < sizeof(dt_obj_version_t))
4429 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4430 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4432 *ver = LDISKFS_I(inode)->i_fs_version;
4434 return sizeof(dt_obj_version_t);
4437 if (!dt_object_exists(dt))
4440 LASSERT(!dt_object_remote(dt));
4441 LASSERT(inode->i_op != NULL);
4442 #ifdef HAVE_IOP_XATTR
4443 LASSERT(inode->i_op->getxattr != NULL);
4446 if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4447 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4451 rc = osd_oxc_get(obj, name, buf);
4456 if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4457 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4458 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4459 struct filter_fid *ff;
4460 struct ost_layout *ol;
4462 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4466 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4469 if (buf->lb_len == 0 || !buf->lb_buf)
4472 if (buf->lb_len < rc)
4476 ol = &ff->ff_layout;
4477 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4478 PFID_STRIPE_IDX_BITS);
4479 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4480 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4481 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4482 if (lma->lma_compat & LMAC_COMP_INFO) {
4483 ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4484 ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4485 ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4487 ol->ol_comp_start = 0;
4488 ol->ol_comp_end = 0;
4492 rc = __osd_xattr_get(inode, dentry, name,
4493 buf->lb_buf, buf->lb_len);
4497 if (rc == -ENOENT || rc == -ENODATA)
4498 osd_oxc_add(obj, name, NULL, 0);
4499 else if (rc > 0 && buf->lb_buf != NULL)
4500 osd_oxc_add(obj, name, buf->lb_buf, rc);
4506 static int osd_declare_xattr_set(const struct lu_env *env,
4507 struct dt_object *dt,
4508 const struct lu_buf *buf, const char *name,
4509 int fl, struct thandle *handle)
4511 struct osd_thandle *oh;
4513 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4515 LASSERT(handle != NULL);
4517 oh = container_of(handle, struct osd_thandle, ot_super);
4518 LASSERT(oh->ot_handle == NULL);
4520 if (strcmp(name, XATTR_NAME_LMA) == 0) {
4522 * For non-upgrading case, the LMA is set first and
4523 * usually fit inode. But for upgrade case, the LMA
4524 * may be in another separated EA block.
4526 if (dt_object_exists(dt)) {
4527 if (fl == LU_XATTR_REPLACE)
4532 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4534 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4535 /* We may need to delete the old PFID EA. */
4536 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4537 if (fl == LU_XATTR_REPLACE)
4543 * If some name entry resides on remote MDT, then will create
4544 * agent entry under remote parent. On the other hand, if the
4545 * remote entry will be removed, then related agent entry may
4546 * need to be removed from the remote parent. So there may be
4547 * kinds of cases, let's declare enough credits. The credits
4548 * for create agent entry is enough for remove case.
4550 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4551 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4552 if (dt_object_exists(dt))
4553 credits += 1; /* For updating LMA */
4557 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4562 if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4564 * learn xattr size from osd_xattr_get if
4565 * attribute has not been read yet
4567 buflen = __osd_xattr_get(
4568 osd_dt_obj(dt)->oo_inode,
4569 &osd_oti_get(env)->oti_obj_dentry,
4574 buflen = buf->lb_len;
4577 if (buflen > sb->s_blocksize) {
4578 credits += osd_calc_bkmap_credits(
4580 (buflen + sb->s_blocksize - 1) >>
4581 sb->s_blocksize_bits);
4585 * xattr set may involve inode quota change, reserve credits for
4586 * dquot_initialize()
4588 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4591 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4596 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4597 const struct lu_buf *buf, int fl,
4598 struct thandle *handle)
4600 struct osd_thread_info *info = osd_oti_get(env);
4601 struct dentry *dentry = &info->oti_obj_dentry;
4602 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4603 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4604 struct inode *inode = obj->oo_inode;
4605 struct filter_fid *ff = buf->lb_buf;
4606 struct ost_layout *ol = &ff->ff_layout;
4607 int flags = XATTR_REPLACE;
4612 if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4615 rc = osd_get_lma(info, inode, dentry, loa);
4616 if (rc == -ENODATA) {
4617 /* Usually for upgarding from old device */
4618 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4619 LMAC_FID_ON_OST, 0);
4620 flags = XATTR_CREATE;
4625 if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4626 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4629 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4630 /* Separate PFID EA from LMA */
4631 lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4632 lustre_lma_swab(lma);
4633 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4634 sizeof(*lma), XATTR_REPLACE);
4636 obj->oo_pfid_in_lma = 0;
4637 rc = LU_XATTR_CREATE;
4643 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4647 * Old client does not send stripe information,
4648 * then store the PFID EA on disk separatedly.
4650 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4651 ol->ol_stripe_size == 0))
4654 /* Remove old PFID EA entry firstly. */
4655 dquot_initialize(inode);
4656 rc = ll_vfs_removexattr(dentry, inode, XATTR_NAME_FID);
4657 if (rc == -ENODATA) {
4658 /* XATTR_NAME_FID is already absent */
4665 fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4666 if (likely(ol->ol_stripe_size != 0)) {
4667 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4668 PFID_STRIPE_IDX_BITS;
4669 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4670 lma->lma_compat |= LMAC_STRIPE_INFO;
4671 if (ol->ol_comp_id != 0) {
4672 loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4673 loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4674 loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4675 lma->lma_compat |= LMAC_COMP_INFO;
4679 lustre_loa_swab(loa, false);
4681 /* Store the PFID EA inside LMA. */
4682 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4685 obj->oo_pfid_in_lma = 1;
4691 * In DNE environment, the object (in spite of regular file or directory)
4692 * and its name entry may reside on different MDTs. Under such case, we will
4693 * create an agent entry on the MDT where the object resides. The agent entry
4694 * references the object locally, that makes the object to be visible to the
4695 * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4696 * such as 'tar' can handle the object properly.
4698 * We handle the agent entry during set linkEA that is the common interface
4699 * for both regular file and directroy, can handle kinds of cases, such as
4700 * create/link/unlink/rename, and so on.
4702 * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4704 * XXX: There are two known issues:
4705 * 1. For one object, we will create at most one agent entry even if there
4706 * may be more than one cross-MDTs hard links on the object. So the local
4707 * e2fsck may claim that the object's nlink is larger than the name entries
4708 * that reference such inode. And in further, the e2fsck will fix the nlink
4709 * attribute to match the local references. Then it will cause the object's
4710 * nlink attribute to be inconsistent with the global references. it is bad
4711 * but not fatal. The ref_del() can handle the zero-referenced case. On the
4712 * other hand, the global namespace LFSCK can repair the object's attribute
4713 * according to the linkEA.
4714 * 2. There may be too many hard links on the object as to its linkEA overflow,
4715 * then the linkEA entry for cross-MDTs reference may be discarded. If such
4716 * case happened, then at this point, we do not know whether there are some
4717 * cross-MDTs reference. But there are local references, it guarantees that
4718 * object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4720 static int osd_xattr_handle_linkea(const struct lu_env *env,
4721 struct osd_device *osd,
4722 struct osd_object *obj,
4723 const struct lu_buf *buf,
4724 struct thandle *handle)
4726 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4727 struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4728 struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4729 struct lu_name tmpname;
4730 struct osd_thandle *oh;
4732 bool remote = false;
4736 oh = container_of(handle, struct osd_thandle, ot_super);
4737 LASSERT(oh->ot_handle != NULL);
4739 rc = linkea_init_with_rec(&ldata);
4741 linkea_first_entry(&ldata);
4742 while (ldata.ld_lee != NULL && !remote) {
4743 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4745 if (osd_remote_fid(env, osd, tfid) > 0)
4748 linkea_next_entry(&ldata);
4750 } else if (rc == -ENODATA) {
4756 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4757 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4759 CERROR("%s: failed to remove agent entry for "DFID
4760 ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4761 } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4762 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4764 CERROR("%s: failed to create agent entry for "DFID
4765 ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4772 * Concurrency: @dt is write locked.
4774 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4775 const struct lu_buf *buf, const char *name, int fl,
4776 struct thandle *handle)
4778 struct osd_object *obj = osd_dt_obj(dt);
4779 struct osd_device *osd = osd_obj2dev(obj);
4780 struct inode *inode = obj->oo_inode;
4781 struct osd_thread_info *info = osd_oti_get(env);
4791 /* version set is not real XATTR */
4792 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4793 dt_obj_version_t *version = buf->lb_buf;
4796 * for version we are just using xattr API but change inode
4799 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4801 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4802 *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4804 LDISKFS_I(inode)->i_fs_version = *version;
4806 * Version is set after all inode operations are finished,
4807 * so we should mark it dirty here
4809 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4814 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4815 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4818 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4821 * For the OST device with 256 bytes inode size by default,
4822 * the PFID EA will be stored together with LMA EA to avoid
4823 * performance trouble. Otherwise the PFID EA can be stored
4824 * independently. LU-8998
4826 if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4827 (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4828 LASSERT(buf->lb_buf);
4830 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4833 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4834 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4835 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4837 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4841 lma->lma_incompat |= LMAI_STRIPED;
4842 lustre_lma_swab(lma);
4843 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4844 sizeof(*lma), XATTR_REPLACE);
4847 } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4848 LASSERT(!osd->od_is_ost);
4850 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4855 if (fl & LU_XATTR_REPLACE)
4856 fs_flags |= XATTR_REPLACE;
4858 if (fl & LU_XATTR_CREATE)
4859 fs_flags |= XATTR_CREATE;
4861 rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4862 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4865 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4866 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4867 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4873 * Concurrency: @dt is read locked.
4875 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4876 const struct lu_buf *buf)
4878 struct osd_object *obj = osd_dt_obj(dt);
4879 struct inode *inode = obj->oo_inode;
4880 struct osd_thread_info *info = osd_oti_get(env);
4881 struct dentry *dentry = &info->oti_obj_dentry;
4883 if (!dt_object_exists(dt))
4886 LASSERT(!dt_object_remote(dt));
4887 LASSERT(inode->i_op != NULL);
4888 LASSERT(inode->i_op->listxattr != NULL);
4890 dentry->d_inode = inode;
4891 dentry->d_sb = inode->i_sb;
4892 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4895 static int osd_declare_xattr_del(const struct lu_env *env,
4896 struct dt_object *dt, const char *name,
4897 struct thandle *handle)
4899 struct osd_thandle *oh;
4900 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4902 LASSERT(!dt_object_remote(dt));
4903 LASSERT(handle != NULL);
4905 oh = container_of(handle, struct osd_thandle, ot_super);
4906 LASSERT(oh->ot_handle == NULL);
4908 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4909 osd_dto_credits_noquota[DTO_XATTR_SET]);
4911 * xattr del may involve inode quota change, reserve credits for
4912 * dquot_initialize()
4914 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4920 * Concurrency: @dt is write locked.
4922 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4923 const char *name, struct thandle *handle)
4925 struct osd_object *obj = osd_dt_obj(dt);
4926 struct inode *inode = obj->oo_inode;
4927 struct osd_thread_info *info = osd_oti_get(env);
4928 struct dentry *dentry = &info->oti_obj_dentry;
4931 if (!dt_object_exists(dt))
4934 LASSERT(!dt_object_remote(dt));
4935 LASSERT(inode->i_op != NULL);
4936 LASSERT(handle != NULL);
4937 #ifdef HAVE_IOP_XATTR
4938 LASSERT(inode->i_op->removexattr != NULL);
4941 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4943 if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4944 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4946 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4947 &info->oti_ost_attrs);
4949 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4951 lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4952 lustre_lma_swab(lma);
4953 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4954 sizeof(*lma), XATTR_REPLACE);
4956 obj->oo_pfid_in_lma = 0;
4959 dquot_initialize(inode);
4960 dentry->d_inode = inode;
4961 dentry->d_sb = inode->i_sb;
4962 rc = ll_vfs_removexattr(dentry, inode, name);
4965 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4968 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4969 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4970 osd_oxc_del(obj, name);
4975 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4976 __u64 start, __u64 end)
4978 struct osd_object *obj = osd_dt_obj(dt);
4979 struct inode *inode = obj->oo_inode;
4980 struct file *file = osd_quasi_file(env, inode);
4985 rc = vfs_fsync_range(file, start, end, 0);
4990 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4995 static bool osd_check_stale(struct dt_object *dt)
5003 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
5004 const struct dt_index_features *feat)
5006 struct iam_descr *descr;
5008 if (osd_object_is_root(o))
5009 return feat == &dt_directory_features;
5011 LASSERT(o->oo_dir != NULL);
5013 descr = o->oo_dir->od_container.ic_descr;
5014 if (feat == &dt_directory_features) {
5015 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
5020 return feat->dif_keysize_min <= descr->id_key_size &&
5021 descr->id_key_size <= feat->dif_keysize_max &&
5022 feat->dif_recsize_min <= descr->id_rec_size &&
5023 descr->id_rec_size <= feat->dif_recsize_max &&
5024 !(feat->dif_flags & (DT_IND_VARKEY |
5025 DT_IND_VARREC | DT_IND_NONUNQ)) &&
5026 ergo(feat->dif_flags & DT_IND_UPDATE,
5027 1 /* XXX check that object (and fs) is writable */);
5031 static int osd_iam_container_init(const struct lu_env *env,
5032 struct osd_object *obj,
5033 struct osd_directory *dir)
5035 struct iam_container *bag = &dir->od_container;
5038 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
5042 result = iam_container_setup(bag);
5044 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
5046 iam_container_fini(bag);
5053 * Concurrency: no external locking is necessary.
5055 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
5056 const struct dt_index_features *feat)
5060 struct osd_object *obj = osd_dt_obj(dt);
5062 LINVRNT(osd_invariant(obj));
5064 if (osd_object_is_root(obj)) {
5065 dt->do_index_ops = &osd_index_ea_ops;
5067 } else if (feat == &dt_directory_features) {
5068 dt->do_index_ops = &osd_index_ea_ops;
5069 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
5074 } else if (unlikely(feat == &dt_otable_features)) {
5075 dt->do_index_ops = &osd_otable_ops;
5077 } else if (unlikely(feat == &dt_acct_features)) {
5078 dt->do_index_ops = &osd_acct_index_ops;
5081 } else if (!osd_has_index(obj)) {
5082 struct osd_directory *dir;
5083 struct osd_device *osd = osd_obj2dev(obj);
5084 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
5089 spin_lock(&obj->oo_guard);
5090 if (obj->oo_dir == NULL)
5094 * Concurrent thread allocated container data.
5097 spin_unlock(&obj->oo_guard);
5099 * Now, that we have container data, serialize its
5102 down_write(&obj->oo_ext_idx_sem);
5104 * recheck under lock.
5107 if (osd_has_index(obj)) {
5112 result = osd_iam_container_init(env, obj, obj->oo_dir);
5113 if (result || feat == &dt_lfsck_namespace_features ||
5114 feat == &dt_lfsck_layout_orphan_features ||
5115 feat == &dt_lfsck_layout_dangling_features)
5118 result = osd_index_register(osd, fid,
5119 feat->dif_keysize_max,
5120 feat->dif_recsize_max);
5122 CWARN("%s: failed to register index "
5124 osd_name(osd), PFID(fid), result);
5125 else if (result > 0)
5128 CDEBUG(D_LFSCK, "%s: index object "DFID
5129 " (%d/%d) registered\n",
5130 osd_name(osd), PFID(fid),
5131 (int)feat->dif_keysize_max,
5132 (int)feat->dif_recsize_max);
5135 up_write(&obj->oo_ext_idx_sem);
5143 if (result == 0 && skip_iam == 0) {
5144 if (!osd_iam_index_probe(env, obj, feat))
5147 LINVRNT(osd_invariant(obj));
5152 static int osd_otable_it_attr_get(const struct lu_env *env,
5153 struct dt_object *dt,
5154 struct lu_attr *attr)
5160 static const struct dt_object_operations osd_obj_ops = {
5161 .do_read_lock = osd_read_lock,
5162 .do_write_lock = osd_write_lock,
5163 .do_read_unlock = osd_read_unlock,
5164 .do_write_unlock = osd_write_unlock,
5165 .do_write_locked = osd_write_locked,
5166 .do_attr_get = osd_attr_get,
5167 .do_declare_attr_set = osd_declare_attr_set,
5168 .do_attr_set = osd_attr_set,
5169 .do_ah_init = osd_ah_init,
5170 .do_declare_create = osd_declare_create,
5171 .do_create = osd_create,
5172 .do_declare_destroy = osd_declare_destroy,
5173 .do_destroy = osd_destroy,
5174 .do_index_try = osd_index_try,
5175 .do_declare_ref_add = osd_declare_ref_add,
5176 .do_ref_add = osd_ref_add,
5177 .do_declare_ref_del = osd_declare_ref_del,
5178 .do_ref_del = osd_ref_del,
5179 .do_xattr_get = osd_xattr_get,
5180 .do_declare_xattr_set = osd_declare_xattr_set,
5181 .do_xattr_set = osd_xattr_set,
5182 .do_declare_xattr_del = osd_declare_xattr_del,
5183 .do_xattr_del = osd_xattr_del,
5184 .do_xattr_list = osd_xattr_list,
5185 .do_object_sync = osd_object_sync,
5186 .do_invalidate = osd_invalidate,
5187 .do_check_stale = osd_check_stale,
5190 static const struct dt_object_operations osd_obj_otable_it_ops = {
5191 .do_attr_get = osd_otable_it_attr_get,
5192 .do_index_try = osd_index_try,
5195 static int osd_index_declare_iam_delete(const struct lu_env *env,
5196 struct dt_object *dt,
5197 const struct dt_key *key,
5198 struct thandle *handle)
5200 struct osd_thandle *oh;
5202 oh = container_of(handle, struct osd_thandle, ot_super);
5203 LASSERT(oh->ot_handle == NULL);
5205 /* Recycle may cause additional three blocks to be changed. */
5206 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
5207 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
5213 * delete a (key, value) pair from index \a dt specified by \a key
5215 * \param dt osd index object
5216 * \param key key for index
5217 * \param rec record reference
5218 * \param handle transaction handler
5221 * \retval -ve failure
5223 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
5224 const struct dt_key *key,
5225 struct thandle *handle)
5227 struct osd_thread_info *oti = osd_oti_get(env);
5228 struct osd_object *obj = osd_dt_obj(dt);
5229 struct osd_thandle *oh;
5230 struct iam_path_descr *ipd;
5231 struct iam_container *bag = &obj->oo_dir->od_container;
5236 if (!dt_object_exists(dt))
5239 LINVRNT(osd_invariant(obj));
5240 LASSERT(!dt_object_remote(dt));
5241 LASSERT(bag->ic_object == obj->oo_inode);
5242 LASSERT(handle != NULL);
5244 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5246 ipd = osd_idx_ipd_get(env, bag);
5247 if (unlikely(ipd == NULL))
5250 oh = container_of(handle, struct osd_thandle, ot_super);
5251 LASSERT(oh->ot_handle != NULL);
5252 LASSERT(oh->ot_handle->h_transaction != NULL);
5254 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5255 /* swab quota uid/gid provided by caller */
5256 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5257 key = (const struct dt_key *)&oti->oti_quota_id;
5260 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5261 osd_ipd_put(env, bag, ipd);
5262 LINVRNT(osd_invariant(obj));
5263 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5267 static int osd_index_declare_ea_delete(const struct lu_env *env,
5268 struct dt_object *dt,
5269 const struct dt_key *key,
5270 struct thandle *handle)
5272 struct osd_thandle *oh;
5273 struct inode *inode;
5278 LASSERT(!dt_object_remote(dt));
5279 LASSERT(handle != NULL);
5281 oh = container_of(handle, struct osd_thandle, ot_super);
5282 LASSERT(oh->ot_handle == NULL);
5284 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5285 osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5287 inode = osd_dt_obj(dt)->oo_inode;
5291 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5292 i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5297 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5300 struct osd_fid_pack *rec;
5303 if (de->file_type & LDISKFS_DIRENT_LUFID) {
5304 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5305 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5306 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5312 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5313 const struct lu_fid *fid)
5315 struct seq_server_site *ss = osd_seq_site(osd);
5319 /* FID seqs not in FLDB, must be local seq */
5320 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5324 * If FLD is not being initialized yet, it only happens during the
5325 * initialization, likely during mgs initialization, and we assume
5326 * this is local FID.
5328 if (ss == NULL || ss->ss_server_fld == NULL)
5331 /* Only check the local FLDB here */
5332 if (osd_seq_exists(env, osd, fid_seq(fid)))
5338 static void osd_take_care_of_agent(const struct lu_env *env,
5339 struct osd_device *osd,
5340 struct osd_thandle *oh,
5341 struct ldiskfs_dir_entry_2 *de)
5343 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5344 struct osd_idmap_cache *idc;
5345 int rc, schedule = 0;
5347 LASSERT(de != NULL);
5349 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5350 if (likely(rc == 0)) {
5351 idc = osd_idc_find_or_init(env, osd, fid);
5352 if (IS_ERR(idc) || idc->oic_remote)
5354 } else if (rc == -ENODATA) {
5356 * can't get FID, postpone to the end of the
5357 * transaction when iget() is safe
5361 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5364 osd_schedule_agent_inode_removal(env, oh,
5365 le32_to_cpu(de->inode));
5369 * Utility function to get real name from object name
5371 * \param[in] obj pointer to the object to be handled
5372 * \param[in] name object name
5373 * \param[in] len object name len
5374 * \param[out]ln pointer to the struct lu_name to hold the real name
5376 * If file is not encrypted, real name is just the object name.
5377 * If file is encrypted, object name needs to be decoded. In
5378 * this case a new buffer is allocated, and ln->ln_name needs to be freed by
5381 * \retval 0, on success
5382 * \retval -ve, on error
5384 static int obj_name2lu_name(struct osd_object *obj, const char *name,
5385 int len, struct lu_name *ln)
5387 struct lu_fid namefid;
5392 sscanf(name + 1, SFID, RFID(&namefid));
5394 if (fid_is_sane(&namefid) || name_is_dot_or_dotdot(name, len) ||
5395 !(obj->oo_lma_flags & LUSTRE_ENCRYPT_FL)) {
5397 ln->ln_namelen = len;
5399 char *buf = kmalloc(len, GFP_NOFS);
5404 len = critical_decode(name, len, buf);
5406 ln->ln_namelen = len;
5413 * Index delete function for interoperability mode (b11826).
5414 * It will remove the directory entry added by osd_index_ea_insert().
5415 * This entry is needed to maintain name->fid mapping.
5417 * \param key, key i.e. file entry to be deleted
5419 * \retval 0, on success
5420 * \retval -ve, on error
5422 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5423 const struct dt_key *key, struct thandle *handle)
5425 struct osd_object *obj = osd_dt_obj(dt);
5426 struct inode *dir = obj->oo_inode;
5427 struct dentry *dentry;
5428 struct osd_thandle *oh;
5429 struct ldiskfs_dir_entry_2 *de = NULL;
5430 struct buffer_head *bh;
5431 struct htree_lock *hlock = NULL;
5432 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5438 if (!dt_object_exists(dt))
5441 LINVRNT(osd_invariant(obj));
5442 LASSERT(!dt_object_remote(dt));
5443 LASSERT(handle != NULL);
5445 rc = obj_name2lu_name(obj, (char *)key, strlen((char *)key), &ln);
5449 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5451 oh = container_of(handle, struct osd_thandle, ot_super);
5452 LASSERT(oh->ot_handle != NULL);
5453 LASSERT(oh->ot_handle->h_transaction != NULL);
5455 dquot_initialize(dir);
5456 dentry = osd_child_dentry_get(env, obj, ln.ln_name, ln.ln_namelen);
5458 if (obj->oo_hl_head != NULL) {
5459 hlock = osd_oti_get(env)->oti_hlock;
5460 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5461 dir, LDISKFS_HLOCK_DEL);
5463 down_write(&obj->oo_ext_idx_sem);
5466 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5469 * If this is not the ".." entry, it might be a remote DNE
5470 * entry and we need to check if the FID is for a remote
5471 * MDT. If the FID is not in the directory entry (e.g.
5472 * upgraded 1.8 filesystem without dirdata enabled) then
5473 * we need to get the FID from the LMA. For a remote directory
5474 * there HAS to be an LMA, it cannot be an IGIF inode in this
5477 * Delete the entry before the agent inode in order to
5478 * simplify error handling. At worst an error after deleting
5479 * the entry first might leak the agent inode afterward. The
5480 * reverse would need filesystem abort in case of error deleting
5481 * the entry after the agent had been removed, or leave a
5482 * dangling entry pointing at a random inode.
5484 if (strcmp((char *)key, dotdot) != 0)
5485 osd_take_care_of_agent(env, osd, oh, de);
5486 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5492 if (!rc && fid_is_namespace_visible(lu_object_fid(&dt->do_lu)) &&
5493 obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
5494 /* NB, dirent count may not be accurate, because it's counted
5497 if (obj->oo_dirent_count)
5498 obj->oo_dirent_count--;
5500 obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
5503 ldiskfs_htree_unlock(hlock);
5505 up_write(&obj->oo_ext_idx_sem);
5508 LASSERT(osd_invariant(obj));
5509 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5510 if (ln.ln_name != (char *)key)
5516 * Lookup index for \a key and copy record to \a rec.
5518 * \param dt osd index object
5519 * \param key key for index
5520 * \param rec record reference
5522 * \retval +ve success : exact mach
5523 * \retval 0 return record with key not greater than \a key
5524 * \retval -ve failure
5526 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5527 struct dt_rec *rec, const struct dt_key *key)
5529 struct osd_object *obj = osd_dt_obj(dt);
5530 struct iam_path_descr *ipd;
5531 struct iam_container *bag = &obj->oo_dir->od_container;
5532 struct osd_thread_info *oti = osd_oti_get(env);
5533 struct iam_iterator *it = &oti->oti_idx_it;
5534 struct iam_rec *iam_rec;
5539 if (!dt_object_exists(dt))
5542 LASSERT(osd_invariant(obj));
5543 LASSERT(!dt_object_remote(dt));
5544 LASSERT(bag->ic_object == obj->oo_inode);
5546 ipd = osd_idx_ipd_get(env, bag);
5550 /* got ipd now we can start iterator. */
5551 iam_it_init(it, bag, 0, ipd);
5553 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5554 /* swab quota uid/gid provided by caller */
5555 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5556 key = (const struct dt_key *)&oti->oti_quota_id;
5559 rc = iam_it_get(it, (struct iam_key *)key);
5561 if (S_ISDIR(obj->oo_inode->i_mode))
5562 iam_rec = (struct iam_rec *)oti->oti_ldp;
5564 iam_rec = (struct iam_rec *)rec;
5566 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5568 if (S_ISDIR(obj->oo_inode->i_mode))
5569 osd_fid_unpack((struct lu_fid *)rec,
5570 (struct osd_fid_pack *)iam_rec);
5571 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5572 osd_quota_unpack(obj, rec);
5577 osd_ipd_put(env, bag, ipd);
5579 LINVRNT(osd_invariant(obj));
5584 static int osd_index_declare_iam_insert(const struct lu_env *env,
5585 struct dt_object *dt,
5586 const struct dt_rec *rec,
5587 const struct dt_key *key,
5588 struct thandle *handle)
5590 struct osd_thandle *oh;
5592 LASSERT(handle != NULL);
5594 oh = container_of(handle, struct osd_thandle, ot_super);
5595 LASSERT(oh->ot_handle == NULL);
5597 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5598 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5604 * Inserts (key, value) pair in \a dt index object.
5606 * \param dt osd index object
5607 * \param key key for index
5608 * \param rec record reference
5609 * \param th transaction handler
5612 * \retval -ve failure
5614 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5615 const struct dt_rec *rec,
5616 const struct dt_key *key, struct thandle *th)
5618 struct osd_object *obj = osd_dt_obj(dt);
5619 struct iam_path_descr *ipd;
5620 struct osd_thandle *oh;
5621 struct iam_container *bag;
5622 struct osd_thread_info *oti = osd_oti_get(env);
5623 struct iam_rec *iam_rec;
5628 if (!dt_object_exists(dt))
5631 LINVRNT(osd_invariant(obj));
5632 LASSERT(!dt_object_remote(dt));
5634 bag = &obj->oo_dir->od_container;
5635 LASSERT(bag->ic_object == obj->oo_inode);
5636 LASSERT(th != NULL);
5638 osd_trans_exec_op(env, th, OSD_OT_INSERT);
5640 ipd = osd_idx_ipd_get(env, bag);
5641 if (unlikely(ipd == NULL))
5644 oh = container_of(th, struct osd_thandle, ot_super);
5645 LASSERT(oh->ot_handle != NULL);
5646 LASSERT(oh->ot_handle->h_transaction != NULL);
5647 if (S_ISDIR(obj->oo_inode->i_mode)) {
5648 iam_rec = (struct iam_rec *)oti->oti_ldp;
5649 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5651 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5652 /* pack quota uid/gid */
5653 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5654 key = (const struct dt_key *)&oti->oti_quota_id;
5655 /* pack quota record */
5656 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5657 iam_rec = (struct iam_rec *)rec;
5659 iam_rec = (struct iam_rec *)rec;
5662 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5664 osd_ipd_put(env, bag, ipd);
5665 LINVRNT(osd_invariant(obj));
5666 osd_trans_exec_check(env, th, OSD_OT_INSERT);
5671 * Calls ldiskfs_add_entry() to add directory entry
5672 * into the directory. This is required for
5673 * interoperability mode (b11826)
5675 * \retval 0, on success
5676 * \retval -ve, on error
5678 static int __osd_ea_add_rec(struct osd_thread_info *info,
5679 struct osd_object *pobj, struct inode *cinode,
5680 const char *name, const struct lu_fid *fid,
5681 struct htree_lock *hlock, struct thandle *th)
5683 struct ldiskfs_dentry_param *ldp;
5684 struct dentry *child;
5685 struct osd_thandle *oth;
5689 oth = container_of(th, struct osd_thandle, ot_super);
5690 LASSERT(oth->ot_handle != NULL);
5691 LASSERT(oth->ot_handle->h_transaction != NULL);
5692 LASSERT(pobj->oo_inode);
5694 rc = obj_name2lu_name(pobj, name, strlen(name), &ln);
5698 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5699 if (unlikely(osd_object_is_root(pobj)))
5702 osd_get_ldiskfs_dirent_param(ldp, fid);
5703 child = osd_child_dentry_get(info->oti_env, pobj,
5704 ln.ln_name, ln.ln_namelen);
5705 child->d_fsdata = (void *)ldp;
5706 dquot_initialize(pobj->oo_inode);
5707 rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5708 child, cinode, hlock);
5709 if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5710 struct ldiskfs_dir_entry_2 *de;
5711 struct buffer_head *bh;
5714 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5717 rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5720 if (S_ISDIR(cinode->i_mode))
5721 de->file_type = LDISKFS_DIRENT_LUFID |
5722 LDISKFS_FT_REG_FILE;
5724 de->file_type = LDISKFS_DIRENT_LUFID |
5726 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5733 if (ln.ln_name != name)
5739 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5740 * into the directory.Also sets flags into osd object to
5741 * indicate dot and dotdot are created. This is required for
5742 * interoperability mode (b11826)
5744 * \param dir directory for dot and dotdot fixup.
5745 * \param obj child object for linking
5747 * \retval 0, on success
5748 * \retval -ve, on error
5750 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5751 struct osd_object *dir,
5752 struct inode *parent_dir, const char *name,
5753 const struct lu_fid *dot_fid,
5754 const struct lu_fid *dot_dot_fid,
5757 struct inode *inode = dir->oo_inode;
5758 struct osd_thandle *oth;
5761 oth = container_of(th, struct osd_thandle, ot_super);
5762 LASSERT(oth->ot_handle->h_transaction != NULL);
5763 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5765 if (strcmp(name, dot) == 0) {
5766 if (dir->oo_compat_dot_created) {
5769 LASSERT(inode->i_ino == parent_dir->i_ino);
5770 dir->oo_compat_dot_created = 1;
5773 } else if (strcmp(name, dotdot) == 0) {
5774 if (!dir->oo_compat_dot_created)
5776 /* in case of rename, dotdot is already created */
5777 if (dir->oo_compat_dotdot_created) {
5778 return __osd_ea_add_rec(info, dir, parent_dir, name,
5779 dot_dot_fid, NULL, th);
5782 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5783 struct lu_fid tfid = *dot_dot_fid;
5786 result = osd_add_dot_dotdot_internal(info,
5787 dir->oo_inode, parent_dir, dot_fid,
5790 result = osd_add_dot_dotdot_internal(info,
5791 dir->oo_inode, parent_dir, dot_fid,
5796 dir->oo_compat_dotdot_created = 1;
5804 * It will call the appropriate osd_add* function and return the
5805 * value, return by respective functions.
5807 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5808 struct inode *cinode, const char *name,
5809 const struct lu_fid *fid, struct thandle *th)
5811 struct osd_thread_info *info = osd_oti_get(env);
5812 struct htree_lock *hlock;
5815 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5817 if (name[0] == '.' && (name[1] == '\0' ||
5818 (name[1] == '.' && name[2] == '\0'))) {
5819 if (hlock != NULL) {
5820 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5823 down_write(&pobj->oo_ext_idx_sem);
5826 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5827 lu_object_fid(&pobj->oo_dt.do_lu),
5830 if (hlock != NULL) {
5831 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5832 pobj->oo_inode, LDISKFS_HLOCK_ADD);
5834 down_write(&pobj->oo_ext_idx_sem);
5837 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5838 struct lu_fid *tfid = &info->oti_fid;
5842 rc = __osd_ea_add_rec(info, pobj, cinode, name,
5845 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5849 if (!rc && fid_is_namespace_visible(lu_object_fid(&pobj->oo_dt.do_lu))
5850 && pobj->oo_dirent_count != LU_DIRENT_COUNT_UNSET)
5851 pobj->oo_dirent_count++;
5854 ldiskfs_htree_unlock(hlock);
5856 up_write(&pobj->oo_ext_idx_sem);
5862 osd_ldiskfs_consistency_check(struct osd_thread_info *oti,
5863 struct osd_device *dev,
5864 const struct lu_fid *fid,
5865 struct osd_inode_id *id)
5867 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5868 struct inode *inode = NULL;
5874 if (!scrub_needs_check(scrub, fid, id->oii_ino))
5877 rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5878 if (rc == -ENOENT) {
5879 __u32 gen = id->oii_gen;
5885 inode = osd_iget(oti, dev, id);
5886 /* The inode has been removed (by race maybe). */
5887 if (IS_ERR(inode)) {
5888 rc = PTR_ERR(inode);
5890 RETURN(rc == -ESTALE ? -ENOENT : rc);
5893 /* The OI mapping is lost. */
5894 if (gen != OSD_OII_NOGEN)
5898 * The inode may has been reused by others, we do not know,
5899 * leave it to be handled by subsequent osd_fid_lookup().
5902 } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5909 if (scrub->os_running) {
5910 if (inode == NULL) {
5911 inode = osd_iget(oti, dev, id);
5912 /* The inode has been removed (by race maybe). */
5913 if (IS_ERR(inode)) {
5914 rc = PTR_ERR(inode);
5916 RETURN(rc == -ESTALE ? -ENOENT : rc);
5920 rc = osd_oii_insert(dev, fid, id, insert);
5922 * There is race condition between osd_oi_lookup and OI scrub.
5923 * The OI scrub finished just after osd_oi_lookup() failure.
5924 * Under such case, it is unnecessary to trigger OI scrub again,
5925 * but try to call osd_oi_lookup() again.
5927 if (unlikely(rc == -EAGAIN))
5930 if (!S_ISDIR(inode->i_mode))
5933 rc = osd_check_lmv(oti, dev, inode);
5938 if (dev->od_scrub.os_scrub.os_auto_scrub_interval != AS_NEVER &&
5940 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5941 SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5942 CDEBUG_LIMIT(D_LFSCK | D_CONSOLE | D_WARNING,
5943 "%s: trigger partial OI scrub for RPC inconsistency, checking FID "DFID"/%u: rc = %d\n",
5944 osd_dev2name(dev), PFID(fid), id->oii_ino, rc);
5945 if (rc == 0 || rc == -EALREADY)
5957 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5958 struct osd_device *dev,
5959 struct lu_fid *fid, __u32 ino)
5961 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5962 struct osd_idmap_cache *oic = &oti->oti_cache;
5963 struct inode *inode;
5966 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5967 inode = osd_iget(oti, dev, &oic->oic_lid);
5968 if (IS_ERR(inode)) {
5969 fid_zero(&oic->oic_fid);
5970 return PTR_ERR(inode);
5973 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5976 fid_zero(&oic->oic_fid);
5978 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5982 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5983 struct osd_inode_id *id, const struct lu_fid *fid)
5985 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5986 id->oii_ino, id->oii_gen, info);
5987 info->oti_cache.oic_lid = *id;
5988 info->oti_cache.oic_fid = *fid;
5989 info->oti_cache.oic_dev = osd;
5993 * Get parent FID from the linkEA.
5995 * For a directory which parent resides on remote MDT, to satisfy the
5996 * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5997 * the other hand, to make the lookup(..) on the directory can return
5998 * the real parent FID, we append the real parent FID after its ".."
5999 * name entry in the /REMOTE_PARENT_DIR.
6001 * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
6002 * backup. So after the restore, we cannot get the right parent FID from
6003 * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
6004 * we have stored the real parent FID in the directory object's linkEA,
6005 * we can parse the linkEA for the real parent FID.
6007 * \param[in] env pointer to the thread context
6008 * \param[in] obj pointer to the object to be handled
6009 * \param[out]fid pointer to the buffer to hold the parent FID
6011 * \retval 0 for getting the real parent FID successfully
6012 * \retval negative error number on failure
6014 static int osd_get_pfid_from_linkea(const struct lu_env *env,
6015 struct osd_object *obj,
6018 struct osd_thread_info *oti = osd_oti_get(env);
6019 struct lu_buf *buf = &oti->oti_big_buf;
6020 struct dentry *dentry = &oti->oti_obj_dentry;
6021 struct inode *inode = obj->oo_inode;
6022 struct linkea_data ldata = { NULL };
6028 if (!S_ISDIR(inode->i_mode))
6032 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
6033 buf->lb_buf, buf->lb_len);
6034 if (rc == -ERANGE) {
6035 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
6038 lu_buf_realloc(buf, rc);
6039 if (buf->lb_buf == NULL)
6046 if (unlikely(rc == 0))
6052 if (unlikely(buf->lb_buf == NULL)) {
6053 lu_buf_realloc(buf, rc);
6054 if (buf->lb_buf == NULL)
6061 rc = linkea_init_with_rec(&ldata);
6063 linkea_first_entry(&ldata);
6064 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
6070 static int osd_verify_ent_by_linkea(const struct lu_env *env,
6071 struct inode *inode,
6072 const struct lu_fid *pfid,
6073 const char *name, const int namelen)
6075 struct osd_thread_info *oti = osd_oti_get(env);
6076 struct lu_buf *buf = &oti->oti_big_buf;
6077 struct dentry *dentry = &oti->oti_obj_dentry;
6078 struct linkea_data ldata = { NULL };
6079 struct lu_name cname = { .ln_name = name,
6080 .ln_namelen = namelen };
6086 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
6087 buf->lb_buf, buf->lb_len);
6089 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
6094 if (unlikely(rc == 0))
6097 if (buf->lb_len < rc) {
6098 lu_buf_realloc(buf, rc);
6099 if (buf->lb_buf == NULL)
6106 rc = linkea_init_with_rec(&ldata);
6108 rc = linkea_links_find(&ldata, &cname, pfid);
6114 * Calls ->lookup() to find dentry. From dentry get inode and
6115 * read inode's ea to get fid. This is required for interoperability
6118 * \retval 0, on success
6119 * \retval -ve, on error
6121 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
6122 struct dt_rec *rec, const struct dt_key *key)
6124 struct inode *dir = obj->oo_inode;
6125 struct dentry *dentry;
6126 struct ldiskfs_dir_entry_2 *de;
6127 struct buffer_head *bh;
6128 struct lu_fid *fid = (struct lu_fid *)rec;
6129 struct htree_lock *hlock = NULL;
6136 LASSERT(dir->i_op != NULL);
6137 LASSERT(dir->i_op->lookup != NULL);
6139 rc = obj_name2lu_name(obj, (char *)key, strlen((char *)key), &ln);
6143 dentry = osd_child_dentry_get(env, obj, ln.ln_name, ln.ln_namelen);
6145 if (obj->oo_hl_head != NULL) {
6146 hlock = osd_oti_get(env)->oti_hlock;
6147 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6148 dir, LDISKFS_HLOCK_LOOKUP);
6150 down_read(&obj->oo_ext_idx_sem);
6153 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6155 struct osd_thread_info *oti = osd_oti_get(env);
6156 struct osd_inode_id *id = &oti->oti_id;
6157 struct osd_device *dev = osd_obj2dev(obj);
6159 ino = le32_to_cpu(de->inode);
6160 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
6162 rc = osd_fail_fid_lookup(oti, dev, fid, ino);
6166 rc = osd_get_fid_from_dentry(de, rec);
6168 /* done with de, release bh */
6171 if (unlikely(is_remote_parent_ino(dev, ino))) {
6172 const char *name = (const char *)key;
6175 * If the parent is on remote MDT, and there
6176 * is no FID-in-dirent, then we have to get
6177 * the parent FID from the linkEA.
6179 if (likely(strlen(name) == 2 &&
6180 name[0] == '.' && name[1] == '.'))
6181 rc = osd_get_pfid_from_linkea(env, obj,
6184 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6187 osd_id_gen(id, ino, OSD_OII_NOGEN);
6190 if (rc != 0 || osd_remote_fid(env, dev, fid))
6193 rc = osd_ldiskfs_consistency_check(oti, dev, fid, id);
6194 if (rc != -ENOENT) {
6195 /* Other error should not affect lookup result. */
6198 /* Normal file mapping should be added into OI cache
6199 * after FID in LMA check, but for local files like
6200 * hsm_actions, their FIDs are not stored in OI files,
6201 * see osd_initial_OI_scrub(), and here is the only
6202 * place to load mapping into OI cache.
6204 if (!fid_is_namespace_visible(fid))
6205 osd_add_oi_cache(osd_oti_get(env),
6206 osd_obj2dev(obj), id, fid);
6216 ldiskfs_htree_unlock(hlock);
6218 up_read(&obj->oo_ext_idx_sem);
6219 if (ln.ln_name != (char *)key)
6224 static int osd_index_declare_ea_insert(const struct lu_env *env,
6225 struct dt_object *dt,
6226 const struct dt_rec *rec,
6227 const struct dt_key *key,
6228 struct thandle *handle)
6230 struct osd_thandle *oh;
6231 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6232 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6233 const struct lu_fid *fid = rec1->rec_fid;
6234 int credits, rc = 0;
6235 struct osd_idmap_cache *idc;
6239 LASSERT(!dt_object_remote(dt));
6240 LASSERT(handle != NULL);
6241 LASSERT(fid != NULL);
6242 LASSERT(rec1->rec_type != 0);
6244 oh = container_of(handle, struct osd_thandle, ot_super);
6245 LASSERT(oh->ot_handle == NULL);
6247 credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
6250 * we can't call iget() while a transactions is running
6251 * (this can lead to a deadlock), but we need to know
6252 * inum and object type. so we find this information at
6253 * declaration and cache in per-thread info
6255 idc = osd_idc_find_or_init(env, osd, fid);
6257 RETURN(PTR_ERR(idc));
6258 if (idc->oic_remote) {
6260 * a reference to remote inode is represented by an
6261 * agent inode which we have to create
6263 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
6264 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
6267 osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
6269 if (osd_dt_obj(dt)->oo_inode != NULL) {
6270 struct inode *inode = osd_dt_obj(dt)->oo_inode;
6273 * We ignore block quota on meta pool (MDTs), so needn't
6274 * calculate how many blocks will be consumed by this index
6277 rc = osd_declare_inode_qid(env, i_uid_read(inode),
6279 i_projid_read(inode), 0,
6280 oh, osd_dt_obj(dt), NULL,
6285 #ifdef HAVE_PROJECT_QUOTA
6287 * Reserve credits for local agent inode to transfer
6288 * to 0, quota enforcement is ignored in this case.
6290 if (idc->oic_remote &&
6291 LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
6292 i_projid_read(inode) != 0)
6293 rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
6294 0, i_projid_read(inode),
6295 0, false, PRJQUOTA);
6303 * Index add function for interoperability mode (b11826).
6304 * It will add the directory entry.This entry is needed to
6305 * maintain name->fid mapping.
6307 * \param key it is key i.e. file entry to be inserted
6308 * \param rec it is value of given key i.e. fid
6310 * \retval 0, on success
6311 * \retval -ve, on error
6313 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6314 const struct dt_rec *rec,
6315 const struct dt_key *key, struct thandle *th)
6317 struct osd_object *obj = osd_dt_obj(dt);
6318 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6319 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6320 const struct lu_fid *fid = rec1->rec_fid;
6321 const char *name = (const char *)key;
6322 struct osd_thread_info *oti = osd_oti_get(env);
6323 struct inode *child_inode = NULL;
6324 struct osd_idmap_cache *idc;
6329 if (!dt_object_exists(dt))
6332 LASSERT(osd_invariant(obj));
6333 LASSERT(!dt_object_remote(dt));
6334 LASSERT(th != NULL);
6336 osd_trans_exec_op(env, th, OSD_OT_INSERT);
6338 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6340 idc = osd_idc_find(env, osd, fid);
6341 if (unlikely(idc == NULL)) {
6342 idc = osd_idc_find_or_init(env, osd, fid);
6345 * this dt_insert() wasn't declared properly, so
6346 * FID is missing in OI cache. we better do not
6347 * lookup FID in FLDB/OI and don't risk to deadlock,
6348 * but in some special cases (lfsck testing, etc)
6349 * it's much simpler than fixing a caller.
6351 * normally this error should be placed after the first
6352 * find, but migrate may attach source stripes to
6353 * target, which doesn't create stripes.
6355 CERROR("%s: "DFID" wasn't declared for insert\n",
6356 osd_name(osd), PFID(fid));
6358 RETURN(PTR_ERR(idc));
6362 if (idc->oic_remote) {
6363 /* Insert remote entry */
6364 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6366 igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6368 child_inode = osd_create_local_agent_inode(env, osd,
6369 obj, fid, rec1->rec_type & S_IFMT, th);
6370 if (IS_ERR(child_inode))
6371 RETURN(PTR_ERR(child_inode));
6374 /* Insert local entry */
6375 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6376 /* for a reason OI cache wasn't filled properly */
6377 CERROR("%s: OIC for "DFID" isn't filled\n",
6378 osd_name(osd), PFID(fid));
6381 child_inode = oti->oti_inode;
6382 if (unlikely(child_inode == NULL)) {
6383 struct ldiskfs_inode_info *lii;
6388 child_inode = oti->oti_inode = &lii->vfs_inode;
6390 child_inode->i_sb = osd_sb(osd);
6391 child_inode->i_ino = idc->oic_lid.oii_ino;
6392 child_inode->i_mode = rec1->rec_type & S_IFMT;
6395 rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6397 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6398 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6400 if (child_inode && child_inode != oti->oti_inode)
6402 LASSERT(osd_invariant(obj));
6403 osd_trans_exec_check(env, th, OSD_OT_INSERT);
6409 * Initialize osd Iterator for given osd index object.
6411 * \param dt osd index object
6414 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6415 struct dt_object *dt,
6418 struct osd_it_iam *it;
6419 struct osd_object *obj = osd_dt_obj(dt);
6420 struct lu_object *lo = &dt->do_lu;
6421 struct iam_path_descr *ipd;
6422 struct iam_container *bag = &obj->oo_dir->od_container;
6424 if (!dt_object_exists(dt))
6425 return ERR_PTR(-ENOENT);
6429 return ERR_PTR(-ENOMEM);
6431 ipd = osd_it_ipd_get(env, bag);
6432 if (likely(ipd != NULL)) {
6436 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6437 return (struct dt_it *)it;
6440 return ERR_PTR(-ENOMEM);
6445 * free given Iterator.
6447 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6449 struct osd_it_iam *it = (struct osd_it_iam *)di;
6450 struct osd_object *obj = it->oi_obj;
6452 iam_it_fini(&it->oi_it);
6453 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6454 osd_object_put(env, obj);
6459 * Move Iterator to record specified by \a key
6461 * \param di osd iterator
6462 * \param key key for index
6464 * \retval +ve di points to record with least key not larger than key
6465 * \retval 0 di points to exact matched key
6466 * \retval -ve failure
6469 static int osd_it_iam_get(const struct lu_env *env,
6470 struct dt_it *di, const struct dt_key *key)
6472 struct osd_thread_info *oti = osd_oti_get(env);
6473 struct osd_it_iam *it = (struct osd_it_iam *)di;
6475 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6476 /* swab quota uid/gid */
6477 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6478 key = (struct dt_key *)&oti->oti_quota_id;
6481 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6487 * \param di osd iterator
6489 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6491 struct osd_it_iam *it = (struct osd_it_iam *)di;
6493 iam_it_put(&it->oi_it);
6497 * Move iterator by one record
6499 * \param di osd iterator
6501 * \retval +1 end of container reached
6503 * \retval -ve failure
6506 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6508 struct osd_it_iam *it = (struct osd_it_iam *)di;
6510 return iam_it_next(&it->oi_it);
6514 * Return pointer to the key under iterator.
6517 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6518 const struct dt_it *di)
6520 struct osd_thread_info *oti = osd_oti_get(env);
6521 struct osd_it_iam *it = (struct osd_it_iam *)di;
6522 struct osd_object *obj = it->oi_obj;
6525 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6527 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6528 /* swab quota uid/gid */
6529 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6530 key = (struct dt_key *)&oti->oti_quota_id;
6537 * Return size of key under iterator (in bytes)
6540 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6542 struct osd_it_iam *it = (struct osd_it_iam *)di;
6544 return iam_it_key_size(&it->oi_it);
6548 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6550 /* check if file type is required */
6551 if (ent->lde_attrs & LUDA_TYPE) {
6552 struct luda_type *lt;
6553 int align = sizeof(*lt) - 1;
6555 len = (len + align) & ~align;
6556 lt = (struct luda_type *)(ent->lde_name + len);
6557 lt->lt_type = cpu_to_le16(DTTOIF(type));
6560 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6564 * build lu direct from backend fs dirent.
6568 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6569 char *name, __u16 namelen, __u16 type, __u32 attr)
6571 ent->lde_attrs = attr | LUDA_FID;
6572 fid_cpu_to_le(&ent->lde_fid, fid);
6574 ent->lde_hash = cpu_to_le64(offset);
6575 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6577 strncpy(ent->lde_name, name, namelen);
6578 ent->lde_name[namelen] = '\0';
6579 ent->lde_namelen = cpu_to_le16(namelen);
6581 /* append lustre attributes */
6582 osd_it_append_attrs(ent, namelen, type);
6586 * Return pointer to the record under iterator.
6588 static int osd_it_iam_rec(const struct lu_env *env,
6589 const struct dt_it *di,
6590 struct dt_rec *dtrec, __u32 attr)
6592 struct osd_it_iam *it = (struct osd_it_iam *)di;
6593 struct osd_thread_info *info = osd_oti_get(env);
6597 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6598 const struct osd_fid_pack *rec;
6599 struct lu_fid *fid = &info->oti_fid;
6600 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6606 name = (char *)iam_it_key_get(&it->oi_it);
6608 RETURN(PTR_ERR(name));
6610 namelen = iam_it_key_size(&it->oi_it);
6612 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6614 RETURN(PTR_ERR(rec));
6616 rc = osd_fid_unpack(fid, rec);
6620 hash = iam_it_store(&it->oi_it);
6622 /* IAM does not store object type in IAM index (dir) */
6623 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6625 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6626 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6627 (struct iam_rec *)dtrec);
6628 osd_quota_unpack(it->oi_obj, dtrec);
6630 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6631 (struct iam_rec *)dtrec);
6638 * Returns cookie for current Iterator position.
6640 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6642 struct osd_it_iam *it = (struct osd_it_iam *)di;
6644 return iam_it_store(&it->oi_it);
6648 * Restore iterator from cookie.
6650 * \param di osd iterator
6651 * \param hash Iterator location cookie
6653 * \retval +ve di points to record with least key not larger than key.
6654 * \retval 0 di points to exact matched key
6655 * \retval -ve failure
6658 static int osd_it_iam_load(const struct lu_env *env,
6659 const struct dt_it *di, __u64 hash)
6661 struct osd_it_iam *it = (struct osd_it_iam *)di;
6663 return iam_it_load(&it->oi_it, hash);
6666 static const struct dt_index_operations osd_index_iam_ops = {
6667 .dio_lookup = osd_index_iam_lookup,
6668 .dio_declare_insert = osd_index_declare_iam_insert,
6669 .dio_insert = osd_index_iam_insert,
6670 .dio_declare_delete = osd_index_declare_iam_delete,
6671 .dio_delete = osd_index_iam_delete,
6673 .init = osd_it_iam_init,
6674 .fini = osd_it_iam_fini,
6675 .get = osd_it_iam_get,
6676 .put = osd_it_iam_put,
6677 .next = osd_it_iam_next,
6678 .key = osd_it_iam_key,
6679 .key_size = osd_it_iam_key_size,
6680 .rec = osd_it_iam_rec,
6681 .store = osd_it_iam_store,
6682 .load = osd_it_iam_load
6686 struct osd_it_ea *osd_it_dir_init(const struct lu_env *env,
6687 struct inode *inode, __u32 attr)
6689 struct osd_thread_info *info = osd_oti_get(env);
6690 struct osd_it_ea *oie;
6692 struct dentry *obj_dentry;
6696 OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6698 RETURN(ERR_PTR(-ENOMEM));
6699 obj_dentry = &oie->oie_dentry;
6701 obj_dentry->d_inode = inode;
6702 obj_dentry->d_sb = inode->i_sb;
6703 obj_dentry->d_name.hash = 0;
6705 oie->oie_rd_dirent = 0;
6706 oie->oie_it_dirent = 0;
6707 oie->oie_dirent = NULL;
6708 if (unlikely(!info->oti_it_ea_buf_used)) {
6709 oie->oie_buf = info->oti_it_ea_buf;
6710 info->oti_it_ea_buf_used = 1;
6712 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6713 if (oie->oie_buf == NULL)
6714 RETURN(ERR_PTR(-ENOMEM));
6716 oie->oie_obj = NULL;
6718 file = &oie->oie_file;
6720 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6721 if (attr & LUDA_64BITHASH)
6722 file->f_mode = FMODE_64BITHASH;
6724 file->f_mode = FMODE_32BITHASH;
6725 file->f_path.dentry = obj_dentry;
6726 file->f_mapping = inode->i_mapping;
6727 file->f_op = inode->i_fop;
6728 file->f_inode = inode;
6734 * Creates or initializes iterator context.
6736 * \retval struct osd_it_ea, iterator structure on success
6739 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6740 struct dt_object *dt,
6743 struct osd_object *obj = osd_dt_obj(dt);
6744 struct lu_object *lo = &dt->do_lu;
6745 struct osd_it_ea *oie;
6749 if (!dt_object_exists(dt) || obj->oo_destroyed)
6750 RETURN(ERR_PTR(-ENOENT));
6752 oie = osd_it_dir_init(env, obj->oo_inode, attr);
6754 RETURN((struct dt_it *)oie);
6758 RETURN((struct dt_it *)oie);
6761 void osd_it_dir_fini(const struct lu_env *env, struct osd_it_ea *oie,
6762 struct inode *inode)
6764 struct osd_thread_info *info = osd_oti_get(env);
6767 oie->oie_file.f_op->release(inode, &oie->oie_file);
6768 if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6769 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6771 info->oti_it_ea_buf_used = 0;
6772 OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6777 * Destroy or finishes iterator context.
6779 * \param di iterator structure to be destroyed
6781 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6783 struct osd_it_ea *oie = (struct osd_it_ea *)di;
6784 struct osd_object *obj = oie->oie_obj;
6785 struct inode *inode = obj->oo_inode;
6788 osd_it_dir_fini(env, (struct osd_it_ea *)di, inode);
6789 osd_object_put(env, obj);
6794 * It position the iterator at given key, so that next lookup continues from
6795 * that key Or it is similar to dio_it->load() but based on a key,
6796 * rather than file position.
6798 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6801 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6803 static int osd_it_ea_get(const struct lu_env *env,
6804 struct dt_it *di, const struct dt_key *key)
6806 struct osd_it_ea *it = (struct osd_it_ea *)di;
6809 LASSERT(((const char *)key)[0] == '\0');
6810 it->oie_file.f_pos = 0;
6811 it->oie_rd_dirent = 0;
6812 it->oie_it_dirent = 0;
6813 it->oie_dirent = NULL;
6821 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6825 struct osd_filldir_cbs {
6826 struct dir_context ctx;
6827 struct osd_it_ea *it;
6830 * It is called internally by ->iterate*(). It fills the
6831 * iterator's in-memory data structure with required
6832 * information i.e. name, namelen, rec_size etc.
6834 * \param buf in which information to be filled in.
6835 * \param name name of the file in given dir
6837 * \retval 0 on success
6838 * \retval 1 on buffer full
6840 #ifdef HAVE_FILLDIR_USE_CTX
6841 static int osd_ldiskfs_filldir(struct dir_context *buf,
6843 static int osd_ldiskfs_filldir(void *buf,
6845 const char *name, int namelen,
6846 loff_t offset, __u64 ino, unsigned int d_type)
6848 struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6849 struct osd_object *obj = it->oie_obj;
6850 struct osd_it_ea_dirent *ent = it->oie_dirent;
6851 struct lu_fid *fid = &ent->oied_fid;
6852 struct osd_fid_pack *rec;
6853 struct lu_fid namefid;
6857 /* this should never happen */
6858 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6859 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6863 if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6869 /* "." is just the object itself. */
6870 if (namelen == 1 && name[0] == '.') {
6872 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6873 } else if (d_type & LDISKFS_DIRENT_LUFID) {
6874 rec = (struct osd_fid_pack *)(name + namelen + 1);
6875 if (osd_fid_unpack(fid, rec) != 0)
6880 d_type &= ~LDISKFS_DIRENT_LUFID;
6882 /* NOT export local root. */
6884 unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6885 ino = obj->oo_inode->i_ino;
6886 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6890 sscanf(name + 1, SFID, RFID(&namefid));
6891 if (fid_is_sane(&namefid) || name_is_dot_or_dotdot(name, namelen) ||
6892 !obj || !(obj->oo_lma_flags & LUSTRE_ENCRYPT_FL)) {
6893 memcpy(ent->oied_name, name, namelen);
6895 int presented_len = critical_chars(name, namelen);
6897 if (presented_len == namelen)
6898 memcpy(ent->oied_name, name, namelen);
6900 namelen = critical_encode(name, namelen,
6903 ent->oied_name[namelen] = '\0';
6906 ent->oied_ino = ino;
6907 ent->oied_off = offset;
6908 ent->oied_namelen = namelen;
6909 ent->oied_type = d_type;
6911 it->oie_rd_dirent++;
6912 it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6917 * Calls ->iterate*() to load a directory entry at a time
6918 * and stored it in iterator's in-memory data structure.
6920 * \param di iterator's in memory structure
6922 * \retval 0 on success
6923 * \retval -ve on error
6924 * \retval +1 reach the end of entry
6926 int osd_ldiskfs_it_fill(const struct lu_env *env, const struct dt_it *di)
6928 struct osd_it_ea *it = (struct osd_it_ea *)di;
6929 struct osd_object *obj = it->oie_obj;
6930 struct htree_lock *hlock = NULL;
6931 struct file *filp = &it->oie_file;
6933 struct osd_filldir_cbs buf = {
6934 .ctx.actor = osd_ldiskfs_filldir,
6939 it->oie_dirent = it->oie_buf;
6940 it->oie_rd_dirent = 0;
6943 if (obj->oo_hl_head != NULL) {
6944 hlock = osd_oti_get(env)->oti_hlock;
6945 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6947 LDISKFS_HLOCK_READDIR);
6949 down_read(&obj->oo_ext_idx_sem);
6953 filp->f_cred = current_cred();
6954 rc = osd_security_file_alloc(filp);
6958 filp->f_flags |= O_NOATIME;
6959 filp->f_mode |= FMODE_NONOTIFY;
6960 rc = iterate_dir(filp, &buf.ctx);
6964 if (it->oie_rd_dirent == 0) {
6966 * If it does not get any dirent, it means it has been reached
6967 * to the end of the dir
6969 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6973 it->oie_dirent = it->oie_buf;
6974 it->oie_it_dirent = 1;
6979 ldiskfs_htree_unlock(hlock);
6981 up_read(&obj->oo_ext_idx_sem);
6988 * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
6989 * to load a directory entry at a time and stored it in
6990 * iterator's in-memory data structure.
6992 * \param di iterator's in memory structure
6994 * \retval +ve iterator reached to end
6995 * \retval 0 iterator not reached to end
6996 * \retval -ve on error
6998 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
7000 struct osd_it_ea *it = (struct osd_it_ea *)di;
7005 if (it->oie_it_dirent < it->oie_rd_dirent) {
7007 (void *)it->oie_dirent +
7008 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
7009 it->oie_dirent->oied_namelen);
7010 it->oie_it_dirent++;
7013 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
7016 rc = osd_ldiskfs_it_fill(env, di);
7023 * Returns the key at current position from iterator's in memory structure.
7025 * \param di iterator's in memory structure
7027 * \retval key i.e. struct dt_key on success
7029 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
7030 const struct dt_it *di)
7032 struct osd_it_ea *it = (struct osd_it_ea *)di;
7034 return (struct dt_key *)it->oie_dirent->oied_name;
7038 * Returns key's size at current position from iterator's in memory structure.
7040 * \param di iterator's in memory structure
7042 * \retval key_size i.e. struct dt_key on success
7044 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
7046 struct osd_it_ea *it = (struct osd_it_ea *)di;
7048 return it->oie_dirent->oied_namelen;
7051 #if defined LDISKFS_DIR_ENTRY_LEN && defined LDISKFS_DIR_ENTRY_LEN_
7052 #undef LDISKFS_DIR_REC_LEN
7053 #define LDISKFS_DIR_REC_LEN(de) LDISKFS_DIR_ENTRY_LEN_((de))
7056 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
7058 if (LDISKFS_DIR_REC_LEN(de) >=
7059 __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
7066 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
7067 unsigned int blocksize, bool dotdot)
7070 return osd_dotdot_has_space(de);
7072 if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
7073 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
7080 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
7081 handle_t *jh, struct dentry *dentry,
7082 const struct lu_fid *fid, struct buffer_head *bh,
7083 struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
7086 struct inode *dir = dentry->d_parent->d_inode;
7087 struct inode *inode = dentry->d_inode;
7088 struct osd_fid_pack *rec;
7089 struct ldiskfs_dentry_param *ldp;
7090 int namelen = dentry->d_name.len;
7092 struct osd_thread_info *info = osd_oti_get(env);
7096 if (!ldiskfs_has_feature_dirdata(inode->i_sb))
7099 /* There is enough space to hold the FID-in-dirent. */
7100 if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
7101 rc = ldiskfs_journal_get_write_access(jh, bh);
7105 de->name[namelen] = 0;
7106 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
7107 rec->fp_len = sizeof(struct lu_fid) + 1;
7108 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
7109 de->file_type |= LDISKFS_DIRENT_LUFID;
7110 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
7117 rc = ldiskfs_delete_entry(jh, dir, de, bh);
7121 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
7122 osd_get_ldiskfs_dirent_param(ldp, fid);
7123 dentry->d_fsdata = (void *)ldp;
7124 dquot_initialize(dir);
7125 rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
7127 * It is too bad, we cannot reinsert the name entry back.
7128 * That means we lose it!
7132 "%s: fail to reinsert the dirent, dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
7133 osd_ino2name(inode), dir->i_ino, dir->i_generation,
7134 namelen, dentry->d_name.name, PFID(fid), rc);
7140 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
7141 struct osd_it_ea *it, struct lu_fid *fid,
7142 struct osd_inode_id *id, __u32 *attr)
7144 struct osd_thread_info *info = osd_oti_get(env);
7145 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
7146 struct osd_device *dev = osd_obj2dev(obj);
7147 struct super_block *sb = osd_sb(dev);
7148 const char *devname = osd_name(dev);
7149 struct osd_it_ea_dirent *ent = it->oie_dirent;
7150 struct inode *dir = obj->oo_inode;
7151 struct htree_lock *hlock = NULL;
7152 struct buffer_head *bh = NULL;
7153 handle_t *jh = NULL;
7154 struct ldiskfs_dir_entry_2 *de;
7155 struct dentry *dentry;
7156 struct inode *inode;
7157 const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
7160 bool dotdot = false;
7166 if (ent->oied_name[0] == '.') {
7167 if (ent->oied_namelen == 1)
7170 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
7174 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
7175 inode = osd_iget(info, dev, id);
7176 if (IS_ERR(inode)) {
7177 rc = PTR_ERR(inode);
7178 if (rc == -ENOENT || rc == -ESTALE) {
7180 * Maybe dangling name entry, or
7181 * corrupted directory entry.
7183 *attr |= LUDA_UNKNOWN;
7186 CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
7187 "check_repair, dir = %lu/%u, name = %.*s, "
7188 "ino = %llu, rc = %d\n",
7189 devname, dir->i_ino, dir->i_generation,
7190 ent->oied_namelen, ent->oied_name,
7197 rc = obj_name2lu_name(obj, ent->oied_name, ent->oied_namelen, &ln);
7201 dentry = osd_child_dentry_by_inode(env, dir, ln.ln_name, ln.ln_namelen);
7202 rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
7203 if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
7209 * We need to ensure that the name entry is still valid.
7210 * Because it may be removed or renamed by other already.
7212 * The unlink or rename operation will start journal before PDO lock,
7213 * so to avoid deadlock, here we need to start journal handle before
7214 * related PDO lock also. But because we do not know whether there
7215 * will be something to be repaired before PDO lock, we just start
7216 * journal without conditions.
7218 * We may need to remove the name entry firstly, then insert back.
7219 * One credit is for user quota file update.
7220 * One credit is for group quota file update.
7221 * Two credits are for dirty inode.
7223 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
7224 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
7226 if (dev->od_dirent_journal != 0) {
7229 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
7232 CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
7233 "check_repair, dir = %lu/%u, credits = %d, "
7234 "name = %.*s, ino = %llu: rc = %d\n",
7235 devname, dir->i_ino, dir->i_generation, credits,
7236 ent->oied_namelen, ent->oied_name,
7239 GOTO(out_inode, rc);
7242 if (obj->oo_hl_head != NULL) {
7243 hlock = osd_oti_get(env)->oti_hlock;
7245 * "0" means exclusive lock for the whole directory.
7246 * We need to prevent others access such name entry
7247 * during the delete + insert. Neither HLOCK_ADD nor
7248 * HLOCK_DEL cannot guarantee the atomicity.
7250 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
7252 down_write(&obj->oo_ext_idx_sem);
7255 if (obj->oo_hl_head != NULL) {
7256 hlock = osd_oti_get(env)->oti_hlock;
7257 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
7258 LDISKFS_HLOCK_LOOKUP);
7260 down_read(&obj->oo_ext_idx_sem);
7264 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
7265 if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
7266 *attr |= LUDA_IGNORE;
7272 * For dotdot entry, if there is not enough space to hold the
7273 * FID-in-dirent, just keep them there. It only happens when the
7274 * device upgraded from 1.8 or restored from MDT file-level backup.
7275 * For the whole directory, only dotdot entry have no FID-in-dirent
7276 * and needs to get FID from LMA when readdir, it will not affect the
7279 if (dotdot && !osd_dotdot_has_space(de)) {
7280 *attr |= LUDA_UNKNOWN;
7286 if (lu_fid_eq(fid, &lma->lma_self_fid))
7289 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
7290 struct lu_fid *tfid = &lma->lma_self_fid;
7292 if (likely(dotdot &&
7293 fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
7294 fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
7296 * It must be REMOTE_PARENT_DIR and as the
7297 * 'dotdot' entry of remote directory
7299 *attr |= LUDA_IGNORE;
7301 CDEBUG(D_LFSCK, "%s: expect remote agent "
7302 "parent directory, but got %.*s under "
7303 "dir = %lu/%u with the FID "DFID"\n",
7304 devname, ent->oied_namelen,
7305 ent->oied_name, dir->i_ino,
7306 dir->i_generation, PFID(tfid));
7308 *attr |= LUDA_UNKNOWN;
7315 if (!fid_is_zero(fid)) {
7316 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
7318 if (rc == -ENOENT ||
7320 !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
7322 * linkEA does not recognize the dirent entry,
7323 * it may because the dirent entry corruption
7324 * and points to other's inode.
7326 CDEBUG(D_LFSCK, "%s: the target inode does not "
7327 "recognize the dirent, dir = %lu/%u, "
7328 " name = %.*s, ino = %llu, "
7329 DFID": rc = %d\n", devname, dir->i_ino,
7330 dir->i_generation, ent->oied_namelen,
7331 ent->oied_name, ent->oied_ino, PFID(fid), rc);
7332 *attr |= LUDA_UNKNOWN;
7337 if (rc && rc != -ENODATA) {
7338 CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
7339 "dir = %lu/%u, name = %.*s, ino = %llu, "
7340 DFID": rc = %d\n", devname, dir->i_ino,
7341 dir->i_generation, ent->oied_namelen,
7342 ent->oied_name, ent->oied_ino, PFID(fid), rc);
7343 *attr |= LUDA_UNKNOWN;
7351 * linkEA recognizes the dirent entry, the FID-in-LMA is
7352 * valid, trusted, in spite of fid_is_sane(fid) or not.
7354 if (*attr & LUDA_VERIFY_DRYRUN) {
7355 *fid = lma->lma_self_fid;
7356 *attr |= LUDA_REPAIR;
7363 dev->od_dirent_journal = 1;
7364 if (hlock != NULL) {
7365 ldiskfs_htree_unlock(hlock);
7368 up_read(&obj->oo_ext_idx_sem);
7374 *fid = lma->lma_self_fid;
7376 /* Update or append the FID-in-dirent. */
7377 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7378 bh, de, hlock, dotdot);
7380 *attr |= LUDA_REPAIR;
7382 CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7383 "the dirent, dir = %lu/%u, name = %.*s, "
7384 "ino = %llu, "DFID": rc = %d\n",
7385 devname, dir->i_ino, dir->i_generation,
7386 ent->oied_namelen, ent->oied_name,
7387 ent->oied_ino, PFID(fid), rc);
7389 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7390 if (*attr & LUDA_VERIFY_DRYRUN) {
7391 if (fid_is_sane(fid)) {
7392 *attr |= LUDA_REPAIR;
7393 } else if (dev->od_index == 0) {
7394 lu_igif_build(fid, inode->i_ino,
7395 inode->i_generation);
7396 *attr |= LUDA_UPGRADE;
7404 dev->od_dirent_journal = 1;
7405 if (hlock != NULL) {
7406 ldiskfs_htree_unlock(hlock);
7409 up_read(&obj->oo_ext_idx_sem);
7416 if (unlikely(fid_is_sane(fid))) {
7418 * FID-in-dirent exists, but FID-in-LMA is lost.
7419 * Trust the FID-in-dirent, and add FID-in-LMA.
7421 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7423 *attr |= LUDA_REPAIR;
7425 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7426 "update dirent, dir = %lu/%u, "
7427 "name = %.*s, ino = %llu, "
7429 devname, dir->i_ino, dir->i_generation,
7430 ent->oied_namelen, ent->oied_name,
7431 ent->oied_ino, PFID(fid), rc);
7432 } else if (dev->od_index == 0) {
7433 lu_igif_build(fid, inode->i_ino, inode->i_generation);
7435 * It is probably IGIF object. Only aappend the
7436 * FID-in-dirent. OI scrub will process FID-in-LMA.
7438 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7439 bh, de, hlock, dotdot);
7441 *attr |= LUDA_UPGRADE;
7443 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7444 "after the dirent, dir = %lu/%u, "
7445 "name = %.*s, ino = %llu, "
7447 devname, dir->i_ino, dir->i_generation,
7448 ent->oied_namelen, ent->oied_name,
7449 ent->oied_ino, PFID(fid), rc);
7458 if (hlock != NULL) {
7459 ldiskfs_htree_unlock(hlock);
7461 if (dev->od_dirent_journal != 0)
7462 up_write(&obj->oo_ext_idx_sem);
7464 up_read(&obj->oo_ext_idx_sem);
7468 ldiskfs_journal_stop(jh);
7472 if (rc >= 0 && !dirty)
7473 dev->od_dirent_journal = 0;
7474 if (ln.ln_name != ent->oied_name)
7481 * Returns the value at current position from iterator's in memory structure.
7483 * \param di struct osd_it_ea, iterator's in memory structure
7484 * \param attr attr requested for dirent.
7485 * \param lde lustre dirent
7487 * \retval 0 no error and \param lde has correct lustre dirent.
7488 * \retval -ve on error
7490 static inline int osd_it_ea_rec(const struct lu_env *env,
7491 const struct dt_it *di,
7492 struct dt_rec *dtrec, __u32 attr)
7494 struct osd_it_ea *it = (struct osd_it_ea *)di;
7495 struct osd_object *obj = it->oie_obj;
7496 struct osd_device *dev = osd_obj2dev(obj);
7497 struct osd_thread_info *oti = osd_oti_get(env);
7498 struct osd_inode_id *id = &oti->oti_id;
7499 struct lu_fid *fid = &it->oie_dirent->oied_fid;
7500 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
7501 __u32 ino = it->oie_dirent->oied_ino;
7506 LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7508 if (attr & LUDA_VERIFY) {
7509 if (unlikely(is_remote_parent_ino(dev, ino))) {
7510 attr |= LUDA_IGNORE;
7512 * If the parent is on remote MDT, and there
7513 * is no FID-in-dirent, then we have to get
7514 * the parent FID from the linkEA.
7516 if (!fid_is_sane(fid) &&
7517 it->oie_dirent->oied_namelen == 2 &&
7518 it->oie_dirent->oied_name[0] == '.' &&
7519 it->oie_dirent->oied_name[1] == '.')
7520 osd_get_pfid_from_linkea(env, obj, fid);
7522 rc = osd_dirent_check_repair(env, obj, it, fid, id,
7526 if (!fid_is_sane(fid))
7527 attr |= LUDA_UNKNOWN;
7529 attr &= ~LU_DIRENT_ATTRS_MASK;
7530 if (!fid_is_sane(fid)) {
7531 bool is_dotdot = false;
7533 if (it->oie_dirent->oied_namelen == 2 &&
7534 it->oie_dirent->oied_name[0] == '.' &&
7535 it->oie_dirent->oied_name[1] == '.')
7538 * If the parent is on remote MDT, and there
7539 * is no FID-in-dirent, then we have to get
7540 * the parent FID from the linkEA.
7542 if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7543 rc = osd_get_pfid_from_linkea(env, obj, fid);
7545 if (is_dotdot == false &&
7546 OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7549 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7554 /* Pack the entry anyway, at least the offset is right. */
7555 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7556 it->oie_dirent->oied_name,
7557 it->oie_dirent->oied_namelen,
7558 it->oie_dirent->oied_type, attr);
7560 RETURN(rc > 0 ? 0 : rc);
7564 * Returns the record size size at current position.
7566 * This function will return record(lu_dirent) size in bytes.
7568 * \param[in] env execution environment
7569 * \param[in] di iterator's in memory structure
7570 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
7571 * calculate the lu_dirent size.
7573 * \retval record size(in bytes & in memory) of the current lu_dirent
7576 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7579 struct osd_it_ea *it = (struct osd_it_ea *)di;
7581 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7585 * Returns a cookie for current position of the iterator head, so that
7586 * user can use this cookie to load/start the iterator next time.
7588 * \param di iterator's in memory structure
7590 * \retval cookie for current position, on success
7592 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7594 struct osd_it_ea *it = (struct osd_it_ea *)di;
7596 return it->oie_dirent->oied_off;
7600 * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
7601 * to load a directory entry at a time and stored it i inn,
7602 * in iterator's in-memory data structure.
7604 * \param di struct osd_it_ea, iterator's in memory structure
7606 * \retval +ve on success
7607 * \retval -ve on error
7609 static int osd_it_ea_load(const struct lu_env *env,
7610 const struct dt_it *di, __u64 hash)
7612 struct osd_it_ea *it = (struct osd_it_ea *)di;
7616 it->oie_file.f_pos = hash;
7618 rc = osd_ldiskfs_it_fill(env, di);
7629 * Index lookup function for interoperability mode (b11826).
7631 * \param key, key i.e. file name to be searched
7633 * \retval +ve, on success
7634 * \retval -ve, on error
7636 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7637 struct dt_rec *rec, const struct dt_key *key)
7639 struct osd_object *obj = osd_dt_obj(dt);
7644 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7645 LINVRNT(osd_invariant(obj));
7647 rc = osd_ea_lookup_rec(env, obj, rec, key);
7654 * Index and Iterator operations for interoperability
7655 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7657 static const struct dt_index_operations osd_index_ea_ops = {
7658 .dio_lookup = osd_index_ea_lookup,
7659 .dio_declare_insert = osd_index_declare_ea_insert,
7660 .dio_insert = osd_index_ea_insert,
7661 .dio_declare_delete = osd_index_declare_ea_delete,
7662 .dio_delete = osd_index_ea_delete,
7664 .init = osd_it_ea_init,
7665 .fini = osd_it_ea_fini,
7666 .get = osd_it_ea_get,
7667 .put = osd_it_ea_put,
7668 .next = osd_it_ea_next,
7669 .key = osd_it_ea_key,
7670 .key_size = osd_it_ea_key_size,
7671 .rec = osd_it_ea_rec,
7672 .rec_size = osd_it_ea_rec_size,
7673 .store = osd_it_ea_store,
7674 .load = osd_it_ea_load
7678 static void *osd_key_init(const struct lu_context *ctx,
7679 struct lu_context_key *key)
7681 struct osd_thread_info *info;
7683 OBD_ALLOC_PTR(info);
7685 return ERR_PTR(-ENOMEM);
7687 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7688 if (info->oti_it_ea_buf == NULL)
7691 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7693 info->oti_hlock = ldiskfs_htree_lock_alloc();
7694 if (info->oti_hlock == NULL)
7700 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7703 return ERR_PTR(-ENOMEM);
7706 static void osd_key_fini(const struct lu_context *ctx,
7707 struct lu_context_key *key, void *data)
7709 struct osd_thread_info *info = data;
7710 struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7711 struct osd_idmap_cache *idc = info->oti_ins_cache;
7713 if (info->oti_dio_pages) {
7715 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7716 struct page *page = info->oti_dio_pages[i];
7718 LASSERT(PagePrivate2(page));
7719 LASSERT(PageLocked(page));
7720 ClearPagePrivate2(page);
7725 OBD_FREE_PTR_ARRAY_LARGE(info->oti_dio_pages,
7726 PTLRPC_MAX_BRW_PAGES);
7729 if (info->oti_inode != NULL)
7731 if (info->oti_hlock != NULL)
7732 ldiskfs_htree_lock_free(info->oti_hlock);
7733 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7734 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7735 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7736 lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7737 lu_buf_free(&info->oti_big_buf);
7739 LASSERT(info->oti_ins_cache_size > 0);
7740 OBD_FREE_PTR_ARRAY(idc, info->oti_ins_cache_size);
7741 info->oti_ins_cache = NULL;
7742 info->oti_ins_cache_size = 0;
7747 static void osd_key_exit(const struct lu_context *ctx,
7748 struct lu_context_key *key, void *data)
7750 struct osd_thread_info *info = data;
7752 LASSERT(info->oti_r_locks == 0);
7753 LASSERT(info->oti_w_locks == 0);
7754 LASSERT(info->oti_txns == 0);
7757 /* type constructor/destructor: osd_type_init, osd_type_fini */
7758 LU_TYPE_INIT_FINI(osd, &osd_key);
7760 struct lu_context_key osd_key = {
7761 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7762 .lct_init = osd_key_init,
7763 .lct_fini = osd_key_fini,
7764 .lct_exit = osd_key_exit
7768 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7769 const char *name, struct lu_device *next)
7771 struct osd_device *osd = osd_dev(d);
7773 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7774 sizeof(osd->od_svname))
7776 return osd_procfs_init(osd, name);
7779 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7781 struct seq_server_site *ss = osd_seq_site(osd);
7786 if (osd->od_is_ost || osd->od_cl_seq != NULL)
7789 if (unlikely(ss == NULL))
7792 OBD_ALLOC_PTR(osd->od_cl_seq);
7793 if (osd->od_cl_seq == NULL)
7796 seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7797 osd->od_svname, ss->ss_server_seq);
7799 if (ss->ss_node_id == 0) {
7801 * If the OSD on the sequence controller(MDT0), then allocate
7802 * sequence here, otherwise allocate sequence after connected
7803 * to MDT0 (see mdt_register_lwp_callback()).
7805 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7806 &osd->od_cl_seq->lcs_space, env);
7812 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7814 if (osd->od_cl_seq == NULL)
7817 seq_client_fini(osd->od_cl_seq);
7818 OBD_FREE_PTR(osd->od_cl_seq);
7819 osd->od_cl_seq = NULL;
7822 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7826 /* shutdown quota slave instance associated with the device */
7827 if (o->od_quota_slave_md != NULL) {
7828 struct qsd_instance *qsd = o->od_quota_slave_md;
7830 o->od_quota_slave_md = NULL;
7834 if (o->od_quota_slave_dt != NULL) {
7835 struct qsd_instance *qsd = o->od_quota_slave_dt;
7837 o->od_quota_slave_dt = NULL;
7841 osd_fid_fini(env, o);
7842 osd_scrub_cleanup(env, o);
7847 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7851 if (o->od_mnt != NULL) {
7852 shrink_dcache_sb(osd_sb(o));
7853 osd_sync(env, &o->od_dt_dev);
7854 wait_event(o->od_commit_cb_done,
7855 !atomic_read(&o->od_commit_cb_in_flight));
7864 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
7865 # ifndef LDISKFS_HAS_INCOMPAT_FEATURE
7866 /* Newer kernels provide the ldiskfs_set_feature_largedir() wrapper already,
7867 * which calls ldiskfs_update_dynamic_rev() to update ancient filesystems.
7868 * All ldiskfs filesystems are already v2, so it is a no-op and unnecessary.
7869 * This avoids maintaining patches to export this otherwise-useless function.
7871 void ldiskfs_update_dynamic_rev(struct super_block *sb)
7878 static int osd_mount(const struct lu_env *env,
7879 struct osd_device *o, struct lustre_cfg *cfg)
7881 const char *name = lustre_cfg_string(cfg, 0);
7882 const char *dev = lustre_cfg_string(cfg, 1);
7884 unsigned long page, s_flags = 0, lmd_flags = 0;
7885 struct page *__page;
7886 struct file_system_type *type;
7887 char *options = NULL;
7889 struct osd_thread_info *info = osd_oti_get(env);
7890 struct lu_fid *fid = &info->oti_fid;
7891 struct inode *inode;
7892 int rc = 0, force_over_1024tb = 0;
7896 if (o->od_mnt != NULL)
7899 if (strlen(dev) >= sizeof(o->od_mntdev))
7901 strcpy(o->od_mntdev, dev);
7903 str = lustre_cfg_buf(cfg, 2);
7904 sscanf(str, "%lu:%lu", &s_flags, &lmd_flags);
7906 opts = lustre_cfg_string(cfg, 3);
7908 if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7909 CERROR("%s: device %s extents feature is not guaranteed to "
7910 "work on big-endian systems. Use \"bigendian_extents\" "
7911 "mount option to override.\n", name, dev);
7915 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7916 if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7917 CWARN("force_over_128tb option is deprecated. Filesystems smaller than 1024TB can be created without any force option. Use force_over_1024tb option for filesystems larger than 1024TB.\n");
7920 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7921 if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7922 CWARN("force_over_256tb option is deprecated. Filesystems smaller than 1024TB can be created without any force options. Use force_over_1024tb option for filesystems larger than 1024TB.\n");
7925 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
7926 if (opts != NULL && strstr(opts, "force_over_512tb") != NULL) {
7927 CWARN("force_over_512tb option is deprecated. Filesystems smaller than 1024TB can be created without any force options. Use force_over_1024tb option for filesystems larger than 1024TB.\n");
7931 if (opts != NULL && strstr(opts, "force_over_1024tb") != NULL)
7932 force_over_1024tb = 1;
7934 __page = alloc_page(GFP_KERNEL);
7936 GOTO(out, rc = -ENOMEM);
7937 page = (unsigned long)page_address(__page);
7938 options = (char *)page;
7941 /* strip out the options for back compatiblity */
7942 static const char * const sout[] = {
7949 /* strip out option we processed in osd */
7950 "bigendian_extents",
7954 "force_over_1024tb",
7958 strncat(options, opts, PAGE_SIZE);
7959 for (rc = 0, str = options; sout[rc]; ) {
7960 char *op = strstr(str, sout[rc]);
7967 if (op == options || *(op - 1) == ',') {
7968 str = op + strlen(sout[rc]);
7969 if (*str == ',' || *str == '\0') {
7970 *str == ',' ? str++ : str;
7971 memmove(op, str, strlen(str) + 1);
7974 for (str = op; *str != ',' && *str != '\0'; str++)
7978 strncat(options, "user_xattr,acl", PAGE_SIZE);
7981 /* Glom up mount options */
7982 if (*options != '\0')
7983 strncat(options, ",", PAGE_SIZE);
7984 strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7986 type = get_fs_type("ldiskfs");
7988 CERROR("%s: cannot find ldiskfs module\n", name);
7989 GOTO(out, rc = -ENODEV);
7992 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7993 module_put(type->owner);
7995 if (IS_ERR(o->od_mnt)) {
7996 rc = PTR_ERR(o->od_mnt);
7998 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
8002 if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
8003 osd_sb(o)->s_blocksize_bits > 1024ULL << 40 &&
8004 force_over_1024tb == 0) {
8005 CERROR("%s: device %s LDISKFS has not been tested on filesystems larger than 1024TB and may cause data corruption. Use 'force_over_1024tb' mount option to override.\n",
8007 GOTO(out_mnt, rc = -EINVAL);
8010 if (lmd_flags & LMD_FLG_DEV_RDONLY) {
8011 LCONSOLE_WARN("%s: not support dev_rdonly on this device\n",
8014 GOTO(out_mnt, rc = -EOPNOTSUPP);
8017 if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
8018 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
8019 GOTO(out_mnt, rc = -EINVAL);
8022 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
8023 #ifdef LDISKFS_MOUNT_DIRDATA
8024 if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
8025 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
8026 else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
8027 CWARN("%s: device %s was upgraded from Lustre-1.x without "
8028 "enabling the dirdata feature. If you do not want to "
8029 "downgrade to Lustre-1.x again, you can enable it via "
8030 "'tune2fs -O dirdata device'\n", name, dev);
8032 /* enable large_dir on MDTs to avoid REMOTE_PARENT_DIR overflow,
8033 * and on very large OSTs to avoid object directory overflow */
8034 if (unlikely(!ldiskfs_has_feature_largedir(o->od_mnt->mnt_sb) &&
8035 !strstr(name, "MGS"))) {
8036 ldiskfs_set_feature_largedir(o->od_mnt->mnt_sb);
8037 LCONSOLE_INFO("%s: enabled 'large_dir' feature on device %s\n",
8041 inode = osd_sb(o)->s_root->d_inode;
8042 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
8043 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
8045 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
8049 if (lmd_flags & LMD_FLG_NOSCRUB)
8050 o->od_scrub.os_scrub.os_auto_scrub_interval = AS_NEVER;
8052 if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
8053 /* do not use pagecache with flash-backed storage */
8054 o->od_writethrough_cache = 0;
8055 o->od_read_cache = 0;
8066 __free_page(__page);
8071 static struct lu_device *osd_device_fini(const struct lu_env *env,
8072 struct lu_device *d)
8074 struct osd_device *o = osd_dev(d);
8078 osd_index_backup(env, o, false);
8079 osd_shutdown(env, o);
8081 if (o->od_oi_table != NULL)
8082 osd_oi_fini(osd_oti_get(env), o);
8083 if (o->od_extent_bytes_percpu)
8084 free_percpu(o->od_extent_bytes_percpu);
8085 osd_obj_map_fini(o);
8091 static int osd_device_init0(const struct lu_env *env,
8092 struct osd_device *o,
8093 struct lustre_cfg *cfg)
8095 struct lu_device *l = osd2lu_dev(o);
8096 struct osd_thread_info *info;
8099 bool restored = false;
8102 /* if the module was re-loaded, env can loose its keys */
8103 rc = lu_env_refill((struct lu_env *)env);
8106 info = osd_oti_get(env);
8109 l->ld_ops = &osd_lu_ops;
8110 o->od_dt_dev.dd_ops = &osd_dt_ops;
8112 spin_lock_init(&o->od_osfs_lock);
8113 mutex_init(&o->od_otable_mutex);
8114 INIT_LIST_HEAD(&o->od_orphan_list);
8115 INIT_LIST_HEAD(&o->od_index_backup_list);
8116 INIT_LIST_HEAD(&o->od_index_restore_list);
8117 spin_lock_init(&o->od_lock);
8118 o->od_index_backup_policy = LIBP_NONE;
8120 init_waitqueue_head(&o->od_commit_cb_done);
8122 o->od_read_cache = 1;
8123 o->od_writethrough_cache = 1;
8124 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
8125 o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20;
8126 o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20;
8127 o->od_scrub.os_scrub.os_auto_scrub_interval = AS_DEFAULT;
8128 /* default fallocate to unwritten extents: LU-14326/LU-14333 */
8129 o->od_fallocate_zero_blocks = 0;
8131 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
8132 sizeof(o->od_svname));
8133 if (cplen >= sizeof(o->od_svname)) {
8138 o->od_index_backup_stop = 0;
8139 o->od_index = -1; /* -1 means index is invalid */
8140 rc = server_name2index(o->od_svname, &o->od_index, NULL);
8141 if (rc == LDD_F_SV_TYPE_OST)
8144 o->od_full_scrub_ratio = OFSR_DEFAULT;
8145 o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
8146 rc = osd_mount(env, o, cfg);
8150 /* Can only check block device after mount */
8151 o->od_nonrotational =
8152 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
8154 rc = osd_obj_map_init(env, o);
8158 rc = lu_site_init(&o->od_site, l);
8160 GOTO(out_compat, rc);
8161 o->od_site.ls_bottom_dev = l;
8163 rc = lu_site_init_finish(&o->od_site);
8167 opts = lustre_cfg_string(cfg, 3);
8168 if (opts && strstr(opts, "resetoi"))
8171 INIT_LIST_HEAD(&o->od_ios_list);
8172 /* setup scrub, including OI files initialization */
8174 rc = osd_scrub_setup(env, o, restored);
8179 rc = osd_procfs_init(o, o->od_svname);
8181 CERROR("%s: can't initialize procfs: rc = %d\n",
8183 GOTO(out_scrub, rc);
8186 LASSERT(l->ld_site->ls_linkage.next != NULL);
8187 LASSERT(l->ld_site->ls_linkage.prev != NULL);
8189 /* initialize quota slave instance */
8190 /* currently it's no need to prepare qsd_instance_md for OST */
8191 if (!o->od_is_ost) {
8192 o->od_quota_slave_md = qsd_init(env, o->od_svname,
8194 o->od_proc_entry, true);
8195 if (IS_ERR(o->od_quota_slave_md)) {
8196 rc = PTR_ERR(o->od_quota_slave_md);
8197 o->od_quota_slave_md = NULL;
8198 GOTO(out_procfs, rc);
8202 o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
8203 o->od_proc_entry, false);
8205 if (IS_ERR(o->od_quota_slave_dt)) {
8206 if (o->od_quota_slave_md != NULL) {
8207 qsd_fini(env, o->od_quota_slave_md);
8208 o->od_quota_slave_md = NULL;
8211 rc = PTR_ERR(o->od_quota_slave_dt);
8212 o->od_quota_slave_dt = NULL;
8213 GOTO(out_procfs, rc);
8216 o->od_extent_bytes_percpu = alloc_percpu(unsigned int);
8217 if (!o->od_extent_bytes_percpu) {
8219 GOTO(out_procfs, rc);
8227 osd_scrub_cleanup(env, o);
8229 lu_site_fini(&o->od_site);
8231 osd_obj_map_fini(o);
8238 static struct lu_device *osd_device_alloc(const struct lu_env *env,
8239 struct lu_device_type *t,
8240 struct lustre_cfg *cfg)
8242 struct osd_device *o;
8247 return ERR_PTR(-ENOMEM);
8249 rc = dt_device_init(&o->od_dt_dev, t);
8252 * Because the ctx might be revived in dt_device_init,
8253 * refill the env here
8255 lu_env_refill((struct lu_env *)env);
8256 rc = osd_device_init0(env, o, cfg);
8258 dt_device_fini(&o->od_dt_dev);
8261 if (unlikely(rc != 0))
8264 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
8267 static struct lu_device *osd_device_free(const struct lu_env *env,
8268 struct lu_device *d)
8270 struct osd_device *o = osd_dev(d);
8274 /* XXX: make osd top device in order to release reference */
8275 d->ld_site->ls_top_dev = d;
8276 lu_site_purge(env, d->ld_site, -1);
8277 lu_site_print(env, d->ld_site, &d->ld_site->ls_obj_hash.nelems,
8278 D_ERROR, lu_cdebug_printer);
8279 lu_site_fini(&o->od_site);
8280 dt_device_fini(&o->od_dt_dev);
8285 static int osd_process_config(const struct lu_env *env,
8286 struct lu_device *d, struct lustre_cfg *cfg)
8288 struct osd_device *o = osd_dev(d);
8294 switch (cfg->lcfg_command) {
8296 rc = osd_mount(env, o, cfg);
8300 * For the case LCFG_PRE_CLEANUP is not called in advance,
8301 * that may happend if hit failure during mount process.
8303 osd_index_backup(env, o, false);
8304 lu_dev_del_linkage(d->ld_site, d);
8305 rc = osd_shutdown(env, o);
8308 LASSERT(&o->od_dt_dev);
8309 count = class_modify_config(cfg, PARAM_OSD,
8310 &o->od_dt_dev.dd_kobj);
8312 count = class_modify_config(cfg, PARAM_OST,
8313 &o->od_dt_dev.dd_kobj);
8314 rc = count > 0 ? 0 : count;
8316 case LCFG_PRE_CLEANUP:
8318 osd_index_backup(env, o,
8319 o->od_index_backup_policy != LIBP_NONE);
8329 static int osd_recovery_complete(const struct lu_env *env,
8330 struct lu_device *d)
8332 struct osd_device *osd = osd_dev(d);
8337 if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
8341 * start qsd instance on recovery completion, this notifies the quota
8342 * slave code that we are about to process new requests now
8344 rc = qsd_start(env, osd->od_quota_slave_dt);
8345 if (rc == 0 && osd->od_quota_slave_md != NULL)
8346 rc = qsd_start(env, osd->od_quota_slave_md);
8352 * we use exports to track all osd users
8354 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
8355 struct obd_device *obd, struct obd_uuid *cluuid,
8356 struct obd_connect_data *data, void *localdata)
8358 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8359 struct lustre_handle conn;
8364 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
8366 rc = class_connect(&conn, obd, cluuid);
8370 *exp = class_conn2export(&conn);
8372 spin_lock(&osd->od_osfs_lock);
8374 spin_unlock(&osd->od_osfs_lock);
8380 * once last export (we don't count self-export) disappeared
8381 * osd can be released
8383 static int osd_obd_disconnect(struct obd_export *exp)
8385 struct obd_device *obd = exp->exp_obd;
8386 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8387 int rc, release = 0;
8391 /* Only disconnect the underlying layers on the final disconnect. */
8392 spin_lock(&osd->od_osfs_lock);
8394 if (osd->od_connects == 0)
8396 spin_unlock(&osd->od_osfs_lock);
8398 rc = class_disconnect(exp); /* bz 9811 */
8400 if (rc == 0 && release)
8401 class_manual_cleanup(obd);
8405 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8406 struct lu_device *dev)
8408 struct osd_device *osd = osd_dev(dev);
8409 struct lr_server_data *lsd =
8410 &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8415 if (osd->od_quota_slave_md != NULL) {
8416 /* set up quota slave objects for inode */
8417 result = qsd_prepare(env, osd->od_quota_slave_md);
8422 if (osd->od_quota_slave_dt != NULL) {
8423 /* set up quota slave objects for block */
8424 result = qsd_prepare(env, osd->od_quota_slave_dt);
8430 if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8431 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8432 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8433 osd->od_index_in_idif = 1;
8435 osd->od_index_in_idif = 0;
8436 result = osd_register_proc_index_in_idif(osd);
8441 osd->od_index_in_idif = 1;
8445 result = osd_fid_init(env, osd);
8451 * Implementation of lu_device_operations::ldo_fid_alloc() for OSD
8455 * see include/lu_object.h for the details.
8457 static int osd_fid_alloc(const struct lu_env *env, struct lu_device *d,
8458 struct lu_fid *fid, struct lu_object *parent,
8459 const struct lu_name *name)
8461 struct osd_device *osd = osd_dev(d);
8463 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8466 static const struct lu_object_operations osd_lu_obj_ops = {
8467 .loo_object_init = osd_object_init,
8468 .loo_object_delete = osd_object_delete,
8469 .loo_object_release = osd_object_release,
8470 .loo_object_free = osd_object_free,
8471 .loo_object_print = osd_object_print,
8472 .loo_object_invariant = osd_object_invariant
8475 const struct lu_device_operations osd_lu_ops = {
8476 .ldo_object_alloc = osd_object_alloc,
8477 .ldo_process_config = osd_process_config,
8478 .ldo_recovery_complete = osd_recovery_complete,
8479 .ldo_prepare = osd_prepare,
8480 .ldo_fid_alloc = osd_fid_alloc,
8483 static const struct lu_device_type_operations osd_device_type_ops = {
8484 .ldto_init = osd_type_init,
8485 .ldto_fini = osd_type_fini,
8487 .ldto_start = osd_type_start,
8488 .ldto_stop = osd_type_stop,
8490 .ldto_device_alloc = osd_device_alloc,
8491 .ldto_device_free = osd_device_free,
8493 .ldto_device_init = osd_device_init,
8494 .ldto_device_fini = osd_device_fini
8497 static struct lu_device_type osd_device_type = {
8498 .ldt_tags = LU_DEVICE_DT,
8499 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
8500 .ldt_ops = &osd_device_type_ops,
8501 .ldt_ctx_tags = LCT_LOCAL,
8504 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8506 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8507 struct super_block *sb = osd_sb(osd);
8509 return (osd->od_mnt == NULL || sb->s_flags & SB_RDONLY);
8513 * lprocfs legacy support.
8515 static const struct obd_ops osd_obd_device_ops = {
8516 .o_owner = THIS_MODULE,
8517 .o_connect = osd_obd_connect,
8518 .o_disconnect = osd_obd_disconnect,
8519 .o_health_check = osd_health_check,
8522 static ssize_t track_declares_assert_show(struct kobject *kobj,
8523 struct attribute *attr,
8526 return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8529 static ssize_t track_declares_assert_store(struct kobject *kobj,
8530 struct attribute *attr,
8531 const char *buffer, size_t count)
8533 bool track_declares_assert;
8536 rc = kstrtobool(buffer, &track_declares_assert);
8540 ldiskfs_track_declares_assert = track_declares_assert;
8544 LUSTRE_RW_ATTR(track_declares_assert);
8546 static int __init osd_init(void)
8548 struct kobject *kobj;
8551 BUILD_BUG_ON(BH_DXLock >=
8552 sizeof(((struct buffer_head *)0)->b_state) * 8);
8553 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8554 /* please, try to keep osd_thread_info smaller than a page */
8555 BUILD_BUG_ON(sizeof(struct osd_thread_info) > PAGE_SIZE);
8560 rc = lu_kmem_init(ldiskfs_caches);
8564 #ifdef CONFIG_KALLSYMS
8565 priv_security_file_alloc =
8566 (void *)cfs_kallsyms_lookup_name("security_file_alloc");
8569 rc = class_register_type(&osd_obd_device_ops, NULL, true,
8570 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8572 lu_kmem_fini(ldiskfs_caches);
8576 kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8578 rc = sysfs_create_file(kobj,
8579 &lustre_attr_track_declares_assert.attr);
8582 CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8589 static void __exit osd_exit(void)
8591 struct kobject *kobj;
8593 kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8595 sysfs_remove_file(kobj,
8596 &lustre_attr_track_declares_assert.attr);
8599 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8600 lu_kmem_fini(ldiskfs_caches);
8603 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8604 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8605 MODULE_VERSION(LUSTRE_VERSION_STRING);
8606 MODULE_LICENSE("GPL");
8608 module_init(osd_init);
8609 module_exit(osd_exit);