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/module.h>
43 #include <linux/user_namespace.h>
44 #include <linux/uidgid.h>
46 /* prerequisite for linux/xattr.h */
47 #include <linux/types.h>
48 /* prerequisite for linux/xattr.h */
50 /* XATTR_{REPLACE,CREATE} */
51 #include <linux/xattr.h>
53 #include <ldiskfs/ldiskfs.h>
54 #include <ldiskfs/xattr.h>
55 #include <ldiskfs/ldiskfs_extents.h>
58 * struct OBD_{ALLOC,FREE}*()
61 #include <obd_support.h>
62 /* struct ptlrpc_thread */
63 #include <lustre_net.h>
64 #include <lustre_fid.h>
66 #include <uapi/linux/lustre/lustre_param.h>
68 #include "osd_internal.h"
69 #include "osd_dynlocks.h"
71 /* llo_* api support */
72 #include <md_object.h>
73 #include <lustre_quota.h>
75 #include <lustre_linkea.h>
77 /* encoding routines */
78 #include <lustre_crypto.h>
80 /* Maximum EA size is limited by LNET_MTU for remote objects */
81 #define OSD_MAX_EA_SIZE 1048364
84 module_param(ldiskfs_pdo, int, 0644);
85 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
87 int ldiskfs_track_declares_assert;
88 module_param(ldiskfs_track_declares_assert, int, 0644);
89 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
91 /* Slab to allocate dynlocks */
92 struct kmem_cache *dynlock_cachep;
94 /* Slab to allocate osd_it_ea */
95 struct kmem_cache *osd_itea_cachep;
97 static struct lu_kmem_descr ldiskfs_caches[] = {
99 .ckd_cache = &dynlock_cachep,
100 .ckd_name = "dynlock_cache",
101 .ckd_size = sizeof(struct dynlock_handle)
104 .ckd_cache = &osd_itea_cachep,
105 .ckd_name = "osd_itea_cache",
106 .ckd_size = sizeof(struct osd_it_ea)
113 static const char dot[] = ".";
114 static const char dotdot[] = "..";
116 static const struct lu_object_operations osd_lu_obj_ops;
117 static const struct dt_object_operations osd_obj_ops;
118 static const struct dt_object_operations osd_obj_otable_it_ops;
119 static const struct dt_index_operations osd_index_iam_ops;
120 static const struct dt_index_operations osd_index_ea_ops;
122 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
123 const struct lu_fid *fid);
124 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
125 struct osd_device *osd);
127 int osd_trans_declare_op2rb[] = {
128 [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
129 [OSD_OT_PUNCH] = OSD_OT_MAX,
130 [OSD_OT_XATTR_SET] = OSD_OT_XATTR_SET,
131 [OSD_OT_CREATE] = OSD_OT_DESTROY,
132 [OSD_OT_DESTROY] = OSD_OT_CREATE,
133 [OSD_OT_REF_ADD] = OSD_OT_REF_DEL,
134 [OSD_OT_REF_DEL] = OSD_OT_REF_ADD,
135 [OSD_OT_WRITE] = OSD_OT_WRITE,
136 [OSD_OT_INSERT] = OSD_OT_DELETE,
137 [OSD_OT_DELETE] = OSD_OT_INSERT,
138 [OSD_OT_QUOTA] = OSD_OT_MAX,
141 static int osd_has_index(const struct osd_object *obj)
143 return obj->oo_dt.do_index_ops != NULL;
146 static int osd_object_invariant(const struct lu_object *l)
148 return osd_invariant(osd_obj(l));
152 * Concurrency: doesn't matter
154 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
156 struct osd_thread_info *oti = osd_oti_get(env);
158 return oti->oti_w_locks > 0 && o->oo_owner == env;
162 * Concurrency: doesn't access mutable data
164 static int osd_root_get(const struct lu_env *env,
165 struct dt_device *dev, struct lu_fid *f)
167 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
172 * the following set of functions are used to maintain per-thread
173 * cache of FID->ino mapping. this mechanism is needed to resolve
174 * FID to inode at dt_insert() which in turn stores ino in the
175 * directory entries to keep ldiskfs compatible with ext[34].
176 * due to locking-originated restrictions we can't lookup ino
177 * using LU cache (deadlock is possible). lookup using OI is quite
178 * expensive. so instead we maintain this cache and methods like
179 * dt_create() fill it. so in the majority of cases dt_insert() is
180 * able to find needed mapping in lockless manner.
182 static struct osd_idmap_cache *
183 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
184 const struct lu_fid *fid)
186 struct osd_thread_info *oti = osd_oti_get(env);
187 struct osd_idmap_cache *idc = oti->oti_ins_cache;
190 for (i = 0; i < oti->oti_ins_cache_used; i++) {
191 if (!lu_fid_eq(&idc[i].oic_fid, fid))
193 if (idc[i].oic_dev != osd)
202 static struct osd_idmap_cache *
203 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
204 const struct lu_fid *fid)
206 struct osd_thread_info *oti = osd_oti_get(env);
207 struct osd_idmap_cache *idc;
210 if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
211 i = oti->oti_ins_cache_size * 2;
213 i = OSD_INS_CACHE_SIZE;
214 OBD_ALLOC_PTR_ARRAY_LARGE(idc, i);
216 return ERR_PTR(-ENOMEM);
217 if (oti->oti_ins_cache != NULL) {
218 memcpy(idc, oti->oti_ins_cache,
219 oti->oti_ins_cache_used * sizeof(*idc));
220 OBD_FREE_PTR_ARRAY_LARGE(oti->oti_ins_cache,
221 oti->oti_ins_cache_used);
223 oti->oti_ins_cache = idc;
224 oti->oti_ins_cache_size = i;
227 idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
230 idc->oic_lid.oii_ino = 0;
231 idc->oic_lid.oii_gen = 0;
238 * lookup mapping for the given fid in the cache, initialize a
239 * new one if not found. the initialization checks whether the
240 * object is local or remote. for local objects, OI is used to
241 * learn ino/generation. the function is used when the caller
242 * has no information about the object, e.g. at dt_insert().
244 static struct osd_idmap_cache *
245 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
246 const struct lu_fid *fid)
248 struct osd_idmap_cache *idc;
251 idc = osd_idc_find(env, osd, fid);
252 LASSERT(!IS_ERR(idc));
256 CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
257 osd->od_svname, PFID(fid));
259 /* new mapping is needed */
260 idc = osd_idc_add(env, osd, fid);
262 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
263 osd->od_svname, PFID(fid), PTR_ERR(idc));
268 rc = osd_remote_fid(env, osd, fid);
269 if (unlikely(rc < 0))
273 /* the object is local, lookup in OI */
274 /* XXX: probably cheaper to lookup in LU first? */
275 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
277 if (unlikely(rc < 0)) {
278 CERROR("can't lookup: rc = %d\n", rc);
282 /* the object is remote */
289 static void osd_idc_dump_lma(const struct lu_env *env,
290 struct osd_device *osd,
294 struct osd_thread_info *info = osd_oti_get(env);
295 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
296 const struct lu_fid *fid;
297 struct osd_inode_id lid;
301 inode = osd_ldiskfs_iget(osd_sb(osd), ino);
303 CERROR("%s: can't get inode %lu: rc = %d\n",
304 osd->od_svname, ino, (int)PTR_ERR(inode));
307 if (is_bad_inode(inode)) {
308 CERROR("%s: bad inode %lu\n", osd->od_svname, ino);
311 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
313 CERROR("%s: can't get LMA for %lu: rc = %d\n",
314 osd->od_svname, ino, rc);
317 fid = &loa->loa_lma.lma_self_fid;
318 LCONSOLE(D_INFO, "%s: "DFID" in inode %lu/%u\n", osd->od_svname,
319 PFID(fid), ino, (unsigned)inode->i_generation);
322 rc = osd_oi_lookup(osd_oti_get(env), osd, fid, &lid, 0);
324 CERROR("%s: can't lookup "DFID": rc = %d\n",
325 osd->od_svname, PFID(fid), rc);
328 LCONSOLE(D_INFO, "%s: "DFID" maps to %u/%u\n", osd->od_svname,
329 PFID(fid), lid.oii_ino, lid.oii_gen);
334 static void osd_idc_dump_debug(const struct lu_env *env,
335 struct osd_device *osd,
336 const struct lu_fid *fid,
340 struct osd_inode_id lid;
344 rc = osd_oi_lookup(osd_oti_get(env), osd, fid, &lid, 0);
346 LCONSOLE(D_INFO, "%s: "DFID" maps to %u/%u\n",
347 osd->od_svname, PFID(fid), lid.oii_ino, lid.oii_gen);
348 osd_idc_dump_lma(env, osd, lid.oii_ino, false);
350 CERROR("%s: can't lookup "DFID": rc = %d\n",
351 osd->od_svname, PFID(fid), rc);
354 osd_idc_dump_lma(env, osd, ino1, true);
356 osd_idc_dump_lma(env, osd, ino2, true);
360 * lookup mapping for given FID and fill it from the given object.
361 * the object is lolcal by definition.
363 static int osd_idc_find_and_init(const struct lu_env *env,
364 struct osd_device *osd,
365 struct osd_object *obj)
367 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
368 struct osd_idmap_cache *idc;
370 idc = osd_idc_find(env, osd, fid);
371 LASSERT(!IS_ERR(idc));
373 if (obj->oo_inode == NULL)
375 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
376 if (idc->oic_lid.oii_ino) {
377 osd_idc_dump_debug(env, osd, fid,
378 idc->oic_lid.oii_ino,
379 obj->oo_inode->i_ino);
382 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
383 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
388 CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
389 osd->od_svname, PFID(fid));
391 /* new mapping is needed */
392 idc = osd_idc_add(env, osd, fid);
394 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
395 osd->od_svname, PFID(fid), PTR_ERR(idc));
399 if (obj->oo_inode != NULL) {
400 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
401 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
407 * OSD object methods.
411 * Concurrency: no concurrent access is possible that early in object
414 static struct lu_object *osd_object_alloc(const struct lu_env *env,
415 const struct lu_object_header *hdr,
418 struct osd_object *mo;
423 struct lu_object_header *h;
424 struct osd_device *o = osd_dev(d);
426 l = &mo->oo_dt.do_lu;
427 if (unlikely(o->od_in_init)) {
434 lu_object_header_init(h);
435 lu_object_init(l, h, d);
436 lu_object_add_top(h, l);
439 dt_object_init(&mo->oo_dt, NULL, d);
440 mo->oo_header = NULL;
443 mo->oo_dt.do_ops = &osd_obj_ops;
444 l->lo_ops = &osd_lu_obj_ops;
445 init_rwsem(&mo->oo_sem);
446 init_rwsem(&mo->oo_ext_idx_sem);
447 spin_lock_init(&mo->oo_guard);
448 INIT_LIST_HEAD(&mo->oo_xattr_list);
454 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
455 struct dentry *dentry, struct lustre_ost_attrs *loa)
459 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
460 (void *)loa, sizeof(*loa));
462 struct lustre_mdt_attrs *lma = &loa->loa_lma;
464 if (rc < sizeof(*lma))
468 lustre_loa_swab(loa, true);
469 /* Check LMA compatibility */
470 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
472 CWARN("%s: unsupported incompat LMA feature(s) %#x for fid = "DFID", ino = %lu: rc = %d\n",
474 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
475 PFID(&lma->lma_self_fid), inode->i_ino, rc);
477 } else if (rc == 0) {
485 * retrieve object from backend ext fs.
487 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
488 struct osd_inode_id *id)
491 struct inode *inode = NULL;
494 * if we look for an inode withing a running
495 * transaction, then we risk to deadlock
496 * osd_dirent_check_repair() breaks this
498 /* LASSERT(current->journal_info == NULL); */
500 inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
502 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
503 id->oii_ino, PTR_ERR(inode));
504 } else if (id->oii_gen != OSD_OII_NOGEN &&
505 inode->i_generation != id->oii_gen) {
506 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
507 "i_generation = %u\n",
508 id->oii_ino, id->oii_gen, inode->i_generation);
510 inode = ERR_PTR(-ESTALE);
511 } else if (inode->i_nlink == 0) {
513 * due to parallel readdir and unlink,
514 * we can have dead inode here.
516 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
518 inode = ERR_PTR(-ESTALE);
519 } else if (is_bad_inode(inode)) {
520 CWARN("%s: bad inode: ino = %u: rc = %d\n",
521 osd_dev2name(dev), id->oii_ino, -ENOENT);
523 inode = ERR_PTR(-ENOENT);
524 } else if (osd_is_ea_inode(inode)) {
526 * EA inode is internal ldiskfs object, should don't visible
529 CDEBUG(D_INODE, "EA inode: ino = %u\n", id->oii_ino);
531 inode = ERR_PTR(-ENOENT);
532 } else if ((rc = osd_attach_jinode(inode))) {
536 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
537 if (id->oii_gen == OSD_OII_NOGEN)
538 osd_id_gen(id, inode->i_ino, inode->i_generation);
541 * Do not update file c/mtime in ldiskfs.
542 * NB: we don't have any lock to protect this because we don't
543 * have reference on osd_object now, but contention with
544 * another lookup + attr_set can't happen in the tiny window
545 * between if (...) and set S_NOCMTIME.
547 if (!(inode->i_flags & S_NOCMTIME))
548 inode->i_flags |= S_NOCMTIME;
553 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
554 handle_t *handle, struct dentry *child,
555 struct inode *inode, struct htree_lock *hlock)
559 rc = __ldiskfs_add_entry(handle, child, inode, hlock);
560 if (rc == -ENOBUFS || rc == -ENOSPC) {
561 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
562 struct inode *parent = child->d_parent->d_inode;
563 struct lu_fid *fid = NULL;
564 char fidstr[FID_LEN + 1] = "unknown";
566 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
568 fid = &loa->loa_lma.lma_self_fid;
569 } else if (rc2 == -ENODATA) {
570 if (unlikely(is_root_inode(parent))) {
571 fid = &info->oti_fid3;
572 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
573 } else if (!osd->od_is_ost && osd->od_index == 0) {
574 fid = &info->oti_fid3;
575 lu_igif_build(fid, parent->i_ino,
576 parent->i_generation);
581 snprintf(fidstr, sizeof(fidstr), DFID, PFID(fid));
583 /* below message is checked in sanity.sh test_129 */
585 CWARN("%s: directory (inode: %lu, FID: %s) has reached max size limit\n",
586 osd_name(osd), parent->i_ino, fidstr);
588 rc = 0; /* ignore such error now */
589 CWARN("%s: directory (inode: %lu, FID: %s) is approaching max size limit\n",
590 osd_name(osd), parent->i_ino, fidstr);
600 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
601 struct osd_inode_id *id, struct lu_fid *fid)
603 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
607 inode = osd_iget(info, dev, id);
611 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
613 *fid = loa->loa_lma.lma_self_fid;
614 } else if (rc == -ENODATA) {
615 if (unlikely(is_root_inode(inode)))
616 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
618 lu_igif_build(fid, inode->i_ino, inode->i_generation);
626 static struct inode *osd_iget_check(struct osd_thread_info *info,
627 struct osd_device *dev,
628 const struct lu_fid *fid,
629 struct osd_inode_id *id,
638 * The cached OI mapping is trustable. If we cannot locate the inode
639 * via the cached OI mapping, then return the failure to the caller
640 * directly without further OI checking.
644 inode = osd_iget(info, dev, id);
647 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
650 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
651 PFID(fid), id->oii_ino, rc);
657 __u32 saved_ino = id->oii_ino;
658 __u32 saved_gen = id->oii_gen;
661 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
663 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
665 * XXX: There are four possible cases:
667 * Backup/restore caused the OI invalid.
669 * Someone unlinked the object but NOT removed
670 * the OI mapping, such as mount target device
671 * as ldiskfs, and modify something directly.
673 * Someone just removed the object between the
674 * former oi_lookup and the iget. It is normal.
675 * 4. Other failure cases.
677 * Generally, when the device is mounted, it will
678 * auto check whether the system is restored from
679 * file-level backup or not. We trust such detect
680 * to distinguish the 1st case from the 2nd case:
681 * if the OI files are consistent but may contain
682 * stale OI mappings because of case 2, if iget()
683 * returns -ENOENT or -ESTALE, then it should be
688 * If the OI mapping was in OI file before the
689 * osd_iget_check(), but now, it is disappear,
690 * then it must be removed by race. That is a
696 * It is the OI scrub updated the OI mapping by race.
697 * The new OI mapping must be valid.
699 if (saved_ino != id->oii_ino ||
700 (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
709 if (dev->od_scrub.os_scrub.os_file.sf_flags &
712 * It still can be the case 2, but we cannot
713 * distinguish it from the case 1. So return
714 * -EREMCHG to block current operation until
715 * OI scrub rebuilt the OI mappings.
724 if (inode->i_generation == id->oii_gen)
744 * \retval +v: new filter_fid does not contain self-fid
745 * \retval 0: filter_fid_18_23, contains self-fid
746 * \retval -v: other failure cases
748 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
749 struct dentry *dentry, struct lu_fid *fid)
751 struct filter_fid *ff = &info->oti_ff;
752 struct ost_id *ostid = &info->oti_ostid;
755 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
756 if (rc == sizeof(struct filter_fid_18_23)) {
757 struct filter_fid_18_23 *ff_old = (void *)ff;
759 ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq));
760 rc = ostid_set_id(ostid, le64_to_cpu(ff_old->ff_objid));
762 * XXX: use 0 as the index for compatibility, the caller will
763 * handle index related issues when necessary.
766 ostid_to_fid(fid, ostid, 0);
767 } else if (rc >= (int)sizeof(struct filter_fid_24_29)) {
769 } else if (rc >= 0) {
776 static int osd_lma_self_repair(struct osd_thread_info *info,
777 struct osd_device *osd, struct inode *inode,
778 const struct lu_fid *fid, __u32 compat)
783 LASSERT(current->journal_info == NULL);
785 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
786 osd_dto_credits_noquota[DTO_XATTR_SET]);
789 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
794 rc = osd_ea_fid_set(info, inode, fid, compat, 0);
796 CWARN("%s: cannot self repair the LMA: rc = %d\n",
798 ldiskfs_journal_stop(jh);
802 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
804 struct osd_thread_info *info = osd_oti_get(env);
805 struct osd_device *osd = osd_obj2dev(obj);
806 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
807 struct lustre_mdt_attrs *lma = &loa->loa_lma;
808 struct inode *inode = obj->oo_inode;
809 struct dentry *dentry = &info->oti_obj_dentry;
810 struct lu_fid *fid = NULL;
811 const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
816 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
817 (void *)loa, sizeof(*loa));
818 if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
819 fid = &lma->lma_self_fid;
820 rc = osd_get_idif(info, inode, dentry, fid);
821 if (rc > 0 || (rc == -ENODATA && osd->od_index_in_idif)) {
823 * For the given OST-object, if it has neither LMA nor
824 * FID in XATTR_NAME_FID, then the given FID (which is
825 * contained in the @obj, from client RPC for locating
826 * the OST-object) is trusted. We use it to generate
829 osd_lma_self_repair(info, osd, inode, rfid,
840 lustre_lma_swab(lma);
841 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
842 (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
843 S_ISREG(inode->i_mode)))) {
844 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
845 "fid = "DFID", ino = %lu\n", osd_name(osd),
846 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
847 PFID(rfid), inode->i_ino);
850 fid = &lma->lma_self_fid;
851 if (lma->lma_compat & LMAC_STRIPE_INFO &&
853 obj->oo_pfid_in_lma = 1;
854 if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
856 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
860 if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
861 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
862 struct ost_id *oi = &info->oti_ostid;
863 struct lu_fid *fid1 = &info->oti_fid3;
864 __u32 idx = fid_idif_ost_idx(rfid);
867 * For old IDIF, the OST index is not part of the IDIF,
868 * Means that different OSTs may have the same IDIFs.
869 * Under such case, we need to make some compatible
870 * check to make sure to trigger OI scrub properly.
872 if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
873 /* Given @rfid is new, LMA is old. */
874 fid_to_ostid(fid, oi);
875 ostid_to_fid(fid1, oi, idx);
876 if (lu_fid_eq(fid1, rfid)) {
877 if (osd->od_index_in_idif)
878 osd_lma_self_repair(info, osd,
892 struct osd_check_lmv_buf {
893 /* please keep it as first member */
894 struct dir_context ctx;
895 struct osd_thread_info *oclb_info;
896 struct osd_device *oclb_dev;
902 * It is called internally by ->iterate*() to filter out the
903 * local slave object's FID of the striped directory.
905 * \retval 1 found the local slave's FID
906 * \retval 0 continue to check next item
907 * \retval -ve for failure
909 #ifdef HAVE_FILLDIR_USE_CTX
910 static int osd_stripe_dir_filldir(struct dir_context *buf,
912 static int osd_stripe_dir_filldir(void *buf,
914 const char *name, int namelen,
915 loff_t offset, __u64 ino, unsigned int d_type)
917 struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
918 struct osd_thread_info *oti = oclb->oclb_info;
919 struct lu_fid *fid = &oti->oti_fid3;
920 struct osd_inode_id *id = &oti->oti_id3;
921 struct osd_device *dev = oclb->oclb_dev;
930 sscanf(name + 1, SFID, RFID(fid));
931 if (!fid_is_sane(fid))
934 if (osd_remote_fid(oti->oti_env, dev, fid))
937 osd_id_gen(id, ino, OSD_OII_NOGEN);
938 inode = osd_iget(oti, dev, id);
940 return PTR_ERR(inode);
943 osd_add_oi_cache(oti, dev, id, fid);
944 osd_oii_insert(dev, fid, id, true);
945 oclb->oclb_found = true;
951 * When lookup item under striped directory, we need to locate the master
952 * MDT-object of the striped directory firstly, then the client will send
953 * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
954 * and the item's name. If the system is restored from MDT file level backup,
955 * then before the OI scrub completely built the OI files, the OI mappings of
956 * the master MDT-object and slave MDT-object may be invalid. Usually, it is
957 * not a problem for the master MDT-object. Because when locate the master
958 * MDT-object, we will do name based lookup (for the striped directory itself)
959 * firstly, during such process we can setup the correct OI mapping for the
960 * master MDT-object. But it will be trouble for the slave MDT-object. Because
961 * the client will not trigger name based lookup on the MDT to locate the slave
962 * MDT-object before locating item under the striped directory, then when
963 * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
964 * is invalid and does not know what the right OI mapping is, then the MDT has
965 * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
966 * the OI file, related OI mapping is unknown yet, please try again later. And
967 * then client will re-try the RPC again and again until related OI mapping has
968 * been updated. That is quite inefficient.
970 * To resolve above trouble, we will handle it as the following two cases:
972 * 1) The slave MDT-object and the master MDT-object are on different MDTs.
973 * It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
974 * is linked under /REMOTE_PARENT_DIR with the name of its FID string.
975 * We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
976 * directly. Please check osd_fid_lookup().
978 * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
979 * Under such case, during lookup the master MDT-object, we will lookup the
980 * slave MDT-object via readdir against the master MDT-object, because the
981 * slave MDT-objects information are stored as sub-directories with the name
982 * "${FID}:${index}". Then when find the local slave MDT-object, its OI
983 * mapping will be recorded. Then subsequent osd_fid_lookup() will know
984 * the correct OI mapping for the slave MDT-object.
986 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
989 struct lu_buf *buf = &oti->oti_big_buf;
991 struct lmv_mds_md_v1 *lmv1;
992 struct osd_check_lmv_buf oclb = {
993 .ctx.actor = osd_stripe_dir_filldir,
1001 /* We should use the VFS layer to create a real dentry. */
1002 oti->oti_obj_dentry.d_inode = inode;
1003 oti->oti_obj_dentry.d_sb = inode->i_sb;
1005 filp = alloc_file_pseudo(inode, dev->od_mnt, "/", O_NOATIME,
1010 filp->f_mode |= FMODE_64BITHASH;
1014 rc = __osd_xattr_get(inode, filp->f_path.dentry, XATTR_NAME_LMV,
1015 buf->lb_buf, buf->lb_len);
1016 if (rc == -ERANGE) {
1017 rc = __osd_xattr_get(inode, filp->f_path.dentry,
1018 XATTR_NAME_LMV, NULL, 0);
1020 lu_buf_realloc(buf, rc);
1021 if (buf->lb_buf == NULL)
1022 GOTO(out, rc = -ENOMEM);
1028 if (unlikely(rc == 0 || rc == -ENODATA))
1034 if (unlikely(buf->lb_buf == NULL)) {
1035 lu_buf_realloc(buf, rc);
1036 if (buf->lb_buf == NULL)
1037 GOTO(out, rc = -ENOMEM);
1043 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1047 oclb.oclb_items = 0;
1048 rc = iterate_dir(filp, &oclb.ctx);
1049 } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1050 filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1055 "%s: cannot check LMV, ino = %lu/%u: rc = %d\n",
1056 osd_ino2name(inode), inode->i_ino, inode->i_generation,
1064 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1065 const struct lu_fid *fid,
1066 const struct lu_object_conf *conf)
1068 struct osd_thread_info *info;
1069 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1070 struct osd_device *dev;
1071 struct osd_idmap_cache *oic;
1072 struct osd_inode_id *id;
1073 struct inode *inode = NULL;
1074 struct lustre_scrub *scrub;
1075 struct scrub_file *sf;
1076 __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1081 bool remote = false;
1082 bool trusted = true;
1083 bool updated = false;
1084 bool checked = false;
1088 LINVRNT(osd_invariant(obj));
1089 LASSERT(obj->oo_inode == NULL);
1091 if (fid_is_sane(fid) == 0) {
1092 CERROR("%s: invalid FID "DFID"\n", ldev->ld_obd->obd_name,
1098 dev = osd_dev(ldev);
1099 scrub = &dev->od_scrub.os_scrub;
1100 sf = &scrub->os_file;
1101 info = osd_oti_get(env);
1103 oic = &info->oti_cache;
1105 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1109 * For the object is created as locking anchor, or for the object to
1110 * be created on disk. No need to osd_oi_lookup() at here because FID
1111 * shouldn't never be re-used, if it's really a duplicate FID from
1112 * unexpected reason, we should be able to detect it later by calling
1113 * do_create->osd_oi_insert().
1115 if (conf && conf->loc_flags & LOC_F_NEW)
1116 GOTO(out, result = 0);
1118 /* Search order: 1. per-thread cache. */
1119 if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1125 memset(id, 0, sizeof(struct osd_inode_id));
1126 if (!list_empty(&scrub->os_inconsistent_items)) {
1127 /* Search order: 2. OI scrub pending list. */
1128 result = osd_oii_lookup(dev, fid, id);
1134 * The OI mapping in the OI file can be updated by the OI scrub
1135 * when we locate the inode via FID. So it may be not trustable.
1139 /* Search order: 3. OI files. */
1140 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1141 if (result == -ENOENT) {
1142 if (!fid_is_norm(fid) ||
1143 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1144 !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1146 GOTO(out, result = 0);
1151 /* -ESTALE is returned if inode of OST object doesn't exist */
1152 if (result == -ESTALE &&
1153 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1154 GOTO(out, result = 0);
1161 obj->oo_inode = NULL;
1162 /* for later passes through checks, not true on first pass */
1163 if (!IS_ERR_OR_NULL(inode))
1166 inode = osd_iget_check(info, dev, fid, id, trusted);
1167 if (!IS_ERR(inode)) {
1168 obj->oo_inode = inode;
1176 result = PTR_ERR(inode);
1177 if (result == -ENOENT || result == -ESTALE)
1178 GOTO(out, result = 0);
1180 if (result != -EREMCHG)
1185 * We still have chance to get the valid inode: for the
1186 * object which is referenced by remote name entry, the
1187 * object on the local MDT will be linked under the dir
1188 * of "/REMOTE_PARENT_DIR" with its FID string as name.
1190 * We do not know whether the object for the given FID
1191 * is referenced by some remote name entry or not, and
1192 * especially for DNE II, a multiple-linked object may
1193 * have many name entries reside on many MDTs.
1195 * To simplify the operation, OSD will not distinguish
1196 * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1197 * only happened for the RPC from other MDT during the
1198 * OI scrub, or for the client side RPC with FID only,
1199 * such as FID to path, or from old connected client.
1202 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1206 flags |= SS_AUTO_PARTIAL;
1207 flags &= ~SS_AUTO_FULL;
1212 if (scrub->os_running) {
1213 if (scrub->os_partial_scan && !scrub->os_in_join)
1216 if (IS_ERR_OR_NULL(inode) || result) {
1217 osd_oii_insert(dev, fid, id, result == -ENOENT);
1218 GOTO(out, result = -EINPROGRESS);
1222 LASSERT(obj->oo_inode == inode);
1224 osd_oii_insert(dev, fid, id, true);
1228 if (dev->od_scrub.os_scrub.os_auto_scrub_interval == AS_NEVER) {
1230 GOTO(out, result = -EREMCHG);
1233 LASSERT(obj->oo_inode == inode);
1235 osd_add_oi_cache(info, dev, id, fid);
1240 rc1 = osd_scrub_start(env, dev, flags);
1241 CDEBUG_LIMIT(D_LFSCK | D_CONSOLE | D_WARNING,
1242 "%s: trigger OI scrub by RPC for "DFID"/%u with flags %#x: rc = %d\n",
1243 osd_name(dev), PFID(fid), id->oii_ino, flags, rc1);
1244 if (rc1 && rc1 != -EALREADY)
1245 GOTO(out, result = -EREMCHG);
1247 if (IS_ERR_OR_NULL(inode) || result) {
1248 osd_oii_insert(dev, fid, id, result == -ENOENT);
1249 GOTO(out, result = -EINPROGRESS);
1253 LASSERT(obj->oo_inode == inode);
1255 osd_oii_insert(dev, fid, id, true);
1260 if (unlikely(obj->oo_header))
1263 result = osd_check_lma(env, obj);
1267 LASSERTF(id->oii_ino == inode->i_ino &&
1268 id->oii_gen == inode->i_generation,
1269 "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1270 PFID(fid), id->oii_ino, id->oii_gen,
1271 inode->i_ino, inode->i_generation);
1273 saved_ino = inode->i_ino;
1274 saved_gen = inode->i_generation;
1276 if (unlikely(result == -ENODATA)) {
1278 * If the OI scrub updated the OI mapping by race, it
1279 * must be valid. Trust the inode that has no LMA EA.
1284 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1287 * The OI mapping is still there, the inode is still
1288 * valid. It is just becaues the inode has no LMA EA.
1290 if (saved_ino == id->oii_ino &&
1291 saved_gen == id->oii_gen)
1295 * It is the OI scrub updated the OI mapping by race.
1296 * The new OI mapping must be valid.
1304 * "result == -ENOENT" means that the OI mappinghas been
1305 * removed by race, so the inode belongs to other object.
1307 * Others error can be returned directly.
1309 if (result == -ENOENT) {
1312 obj->oo_inode = NULL;
1317 if (result != -EREMCHG)
1323 * if two OST objects map to the same inode, and inode mode is
1324 * (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666), which means it's
1325 * reserved by precreate, and not written yet, in this case, don't
1326 * set inode for the object whose FID mismatch, so that it can create
1327 * inode and not block precreate.
1329 if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) &&
1330 inode->i_mode == (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666)) {
1331 obj->oo_inode = NULL;
1332 GOTO(out, result = 0);
1335 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1337 * "result == -ENOENT" means the cached OI mapping has been removed
1338 * from the OI file by race, above inode belongs to other object.
1340 if (result == -ENOENT) {
1343 obj->oo_inode = NULL;
1344 GOTO(out, result = 0);
1350 if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1352 osd_scrub_refresh_mapping(info, dev, fid, id, DTO_INDEX_DELETE,
1358 * It is the OI scrub updated the OI mapping by race.
1359 * The new OI mapping must be valid.
1367 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1368 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1370 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1372 if (lma->lma_compat & LMAC_STRIPE_INFO &&
1374 obj->oo_pfid_in_lma = 1;
1375 if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1377 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1378 } else if (result != -ENODATA) {
1383 obj->oo_compat_dot_created = 1;
1384 obj->oo_compat_dotdot_created = 1;
1386 if (S_ISDIR(inode->i_mode) &&
1387 (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1388 osd_check_lmv(info, dev, inode);
1390 result = osd_attach_jinode(inode);
1395 GOTO(out, result = 0);
1397 LASSERT(!obj->oo_hl_head);
1398 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1400 GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1403 if (result || !obj->oo_inode) {
1404 if (!IS_ERR_OR_NULL(inode))
1407 obj->oo_inode = NULL;
1409 fid_zero(&oic->oic_fid);
1412 LINVRNT(osd_invariant(obj));
1417 * Concurrency: shouldn't matter.
1419 static void osd_object_init0(struct osd_object *obj)
1421 LASSERT(obj->oo_inode != NULL);
1422 obj->oo_dt.do_body_ops = &osd_body_ops;
1423 obj->oo_dt.do_lu.lo_header->loh_attr |=
1424 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1428 * Concurrency: no concurrent access is possible that early in object
1431 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1432 const struct lu_object_conf *conf)
1434 struct osd_object *obj = osd_obj(l);
1437 LINVRNT(osd_invariant(obj));
1439 if (OBD_FAIL_PRECHECK(OBD_FAIL_MDS_LLOG_UMOUNT_RACE) &&
1440 cfs_fail_val == 2) {
1441 struct osd_thread_info *info = osd_oti_get(env);
1442 struct osd_idmap_cache *oic = &info->oti_cache;
1443 /* invalidate thread cache */
1444 memset(&oic->oic_fid, 0, sizeof(oic->oic_fid));
1446 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1447 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1448 l->lo_header->loh_attr |= LOHA_EXISTS;
1452 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1453 obj->oo_dt.do_body_ops = &osd_body_ops_new;
1454 if (result == 0 && obj->oo_inode != NULL) {
1455 struct osd_thread_info *oti = osd_oti_get(env);
1456 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1458 osd_object_init0(obj);
1459 if (unlikely(obj->oo_header))
1462 result = osd_get_lma(oti, obj->oo_inode,
1463 &oti->oti_obj_dentry, loa);
1466 * Convert LMAI flags to lustre LMA flags
1467 * and cache it to oo_lma_flags
1470 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1471 } else if (result == -ENODATA) {
1475 obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
1477 LINVRNT(osd_invariant(obj));
1482 * The first part of oxe_buf is xattr name, and is '\0' terminated.
1483 * The left part is for value, binary mode.
1485 struct osd_xattr_entry {
1486 struct list_head oxe_list;
1490 struct rcu_head oxe_rcu;
1494 static int osd_oxc_get(struct osd_object *obj, const char *name,
1497 struct osd_xattr_entry *tmp;
1498 struct osd_xattr_entry *oxe = NULL;
1499 size_t namelen = strlen(name);
1503 list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1504 if (namelen == tmp->oxe_namelen &&
1505 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1512 GOTO(out, rc = -ENOENT);
1514 if (!oxe->oxe_exist)
1515 GOTO(out, rc = -ENODATA);
1518 rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1521 if (buf->lb_buf == NULL)
1524 if (buf->lb_len < rc)
1525 GOTO(out, rc = -ERANGE);
1527 memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1534 static void osd_oxc_free(struct rcu_head *head)
1536 struct osd_xattr_entry *oxe;
1538 oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1539 OBD_FREE(oxe, oxe->oxe_len);
1542 static void osd_oxc_add(struct osd_object *obj, const char *name,
1543 const char *buf, int buflen)
1545 struct osd_xattr_entry *oxe;
1546 struct osd_xattr_entry *old = NULL;
1547 struct osd_xattr_entry *tmp;
1548 size_t namelen = strlen(name);
1549 size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1551 OBD_ALLOC(oxe, len);
1555 INIT_LIST_HEAD(&oxe->oxe_list);
1557 oxe->oxe_namelen = namelen;
1558 memcpy(oxe->oxe_buf, name, namelen);
1560 LASSERT(buf != NULL);
1561 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1562 oxe->oxe_exist = true;
1564 oxe->oxe_exist = false;
1567 /* this should be rarely called, just remove old and add new */
1568 spin_lock(&obj->oo_guard);
1569 list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1570 if (namelen == tmp->oxe_namelen &&
1571 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1577 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1578 call_rcu(&old->oxe_rcu, osd_oxc_free);
1580 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1582 spin_unlock(&obj->oo_guard);
1585 static void osd_oxc_del(struct osd_object *obj, const char *name)
1587 struct osd_xattr_entry *oxe;
1588 size_t namelen = strlen(name);
1590 spin_lock(&obj->oo_guard);
1591 list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1592 if (namelen == oxe->oxe_namelen &&
1593 strncmp(name, oxe->oxe_buf, namelen) == 0) {
1594 list_del_rcu(&oxe->oxe_list);
1595 call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1599 spin_unlock(&obj->oo_guard);
1602 static void osd_oxc_fini(struct osd_object *obj)
1604 struct osd_xattr_entry *oxe, *next;
1606 list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1607 list_del(&oxe->oxe_list);
1608 OBD_FREE(oxe, oxe->oxe_len);
1613 * Concurrency: no concurrent access is possible that late in object
1616 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1618 struct osd_object *obj = osd_obj(l);
1619 struct lu_object_header *h = obj->oo_header;
1621 LINVRNT(osd_invariant(obj));
1624 dt_object_fini(&obj->oo_dt);
1625 if (obj->oo_hl_head != NULL)
1626 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1627 /* obj doesn't contain an lu_object_header, so we don't need call_rcu */
1630 lu_object_header_free(h);
1634 * Concurrency: no concurrent access is possible that late in object
1637 static void osd_index_fini(struct osd_object *o)
1639 struct iam_container *bag;
1641 if (o->oo_dir != NULL) {
1642 bag = &o->oo_dir->od_container;
1643 if (o->oo_inode != NULL) {
1644 if (bag->ic_object == o->oo_inode)
1645 iam_container_fini(bag);
1647 OBD_FREE_PTR(o->oo_dir);
1653 OSD_TXN_OI_DELETE_CREDITS = 20,
1654 OSD_TXN_INODE_DELETE_CREDITS = 20
1661 #if OSD_THANDLE_STATS
1663 * Set time when the handle is allocated
1665 static void osd_th_alloced(struct osd_thandle *oth)
1667 oth->oth_alloced = ktime_get();
1671 * Set time when the handle started
1673 static void osd_th_started(struct osd_thandle *oth)
1675 oth->oth_started = ktime_get();
1679 * Check whether the we deal with this handle for too long.
1681 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1682 ktime_t alloced, ktime_t started,
1685 ktime_t now = ktime_get();
1687 LASSERT(dev != NULL);
1689 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1690 ktime_us_delta(started, alloced));
1691 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1692 ktime_us_delta(closed, started));
1693 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1694 ktime_us_delta(now, closed));
1696 if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1697 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1698 oth, now, alloced, started, closed);
1699 libcfs_debug_dumpstack(NULL);
1703 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
1705 ktime_t __closed = ktime_get(); \
1706 ktime_t __alloced = oth->oth_alloced; \
1707 ktime_t __started = oth->oth_started; \
1710 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
1713 #else /* OSD_THANDLE_STATS */
1715 #define osd_th_alloced(h) do {} while(0)
1716 #define osd_th_started(h) do {} while(0)
1717 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
1719 #endif /* OSD_THANDLE_STATS */
1722 * Concurrency: doesn't access mutable data.
1724 static int osd_param_is_not_sane(const struct osd_device *dev,
1725 const struct thandle *th)
1727 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1729 return oh->ot_credits > osd_transaction_size(dev);
1733 * Concurrency: shouldn't matter.
1735 static void osd_trans_commit_cb(struct super_block *sb,
1736 struct ldiskfs_journal_cb_entry *jcb, int error)
1738 struct osd_thandle *oh = container_of(jcb, struct osd_thandle, ot_jcb);
1739 struct thandle *th = &oh->ot_super;
1740 struct lu_device *lud = &th->th_dev->dd_lu_dev;
1741 struct osd_device *osd = osd_dev(lud);
1742 struct dt_txn_commit_cb *dcb, *tmp;
1744 LASSERT(oh->ot_handle == NULL);
1747 CERROR("transaction @0x%p commit error: %d\n", th, error);
1749 /* call per-transaction callbacks if any */
1750 list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1752 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1753 "commit callback entry: magic=%x name='%s'\n",
1754 dcb->dcb_magic, dcb->dcb_name);
1755 list_del_init(&dcb->dcb_linkage);
1756 dcb->dcb_func(NULL, th, dcb, error);
1759 lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1760 if (atomic_dec_and_test(&osd->od_commit_cb_in_flight))
1761 wake_up(&osd->od_commit_cb_done);
1767 static struct thandle *osd_trans_create(const struct lu_env *env,
1768 struct dt_device *d)
1770 struct osd_thread_info *oti = osd_oti_get(env);
1771 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1772 struct osd_thandle *oh;
1778 CERROR("%s: someone try to start transaction under "
1779 "readonly mode, should be disabled.\n",
1780 osd_name(osd_dt_dev(d)));
1782 RETURN(ERR_PTR(-EROFS));
1785 /* on pending IO in this thread should left from prev. request */
1786 LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1788 sb_start_write(osd_sb(osd_dt_dev(d)));
1790 OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1792 sb_end_write(osd_sb(osd_dt_dev(d)));
1793 RETURN(ERR_PTR(-ENOMEM));
1796 oh->ot_quota_trans = &oti->oti_quota_trans;
1797 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1802 oh->oh_declared_ext = 0;
1803 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1804 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1805 INIT_LIST_HEAD(&oh->ot_trunc_locks);
1808 memset(oti->oti_declare_ops, 0,
1809 sizeof(oti->oti_declare_ops));
1810 memset(oti->oti_declare_ops_cred, 0,
1811 sizeof(oti->oti_declare_ops_cred));
1812 memset(oti->oti_declare_ops_used, 0,
1813 sizeof(oti->oti_declare_ops_used));
1815 oti->oti_ins_cache_depth++;
1820 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1822 struct osd_thread_info *oti = osd_oti_get(env);
1823 struct osd_thandle *oh;
1825 oh = container_of(th, struct osd_thandle, ot_super);
1826 LASSERT(oh != NULL);
1828 CWARN(" create: %u/%u/%u, destroy: %u/%u/%u\n",
1829 oti->oti_declare_ops[OSD_OT_CREATE],
1830 oti->oti_declare_ops_cred[OSD_OT_CREATE],
1831 oti->oti_declare_ops_used[OSD_OT_CREATE],
1832 oti->oti_declare_ops[OSD_OT_DESTROY],
1833 oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1834 oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1835 CWARN(" attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1836 oti->oti_declare_ops[OSD_OT_ATTR_SET],
1837 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1838 oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1839 oti->oti_declare_ops[OSD_OT_XATTR_SET],
1840 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1841 oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1842 CWARN(" write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1843 oti->oti_declare_ops[OSD_OT_WRITE],
1844 oti->oti_declare_ops_cred[OSD_OT_WRITE],
1845 oti->oti_declare_ops_used[OSD_OT_WRITE],
1846 oti->oti_declare_ops[OSD_OT_PUNCH],
1847 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1848 oti->oti_declare_ops_used[OSD_OT_PUNCH],
1849 oti->oti_declare_ops[OSD_OT_QUOTA],
1850 oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1851 oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1852 CWARN(" insert: %u/%u/%u, delete: %u/%u/%u\n",
1853 oti->oti_declare_ops[OSD_OT_INSERT],
1854 oti->oti_declare_ops_cred[OSD_OT_INSERT],
1855 oti->oti_declare_ops_used[OSD_OT_INSERT],
1856 oti->oti_declare_ops[OSD_OT_DELETE],
1857 oti->oti_declare_ops_cred[OSD_OT_DELETE],
1858 oti->oti_declare_ops_used[OSD_OT_DELETE]);
1859 CWARN(" ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1860 oti->oti_declare_ops[OSD_OT_REF_ADD],
1861 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1862 oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1863 oti->oti_declare_ops[OSD_OT_REF_DEL],
1864 oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1865 oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1869 * Concurrency: shouldn't matter.
1871 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1874 struct osd_thread_info *oti = osd_oti_get(env);
1875 struct osd_device *dev = osd_dt_dev(d);
1877 struct osd_thandle *oh;
1882 LASSERT(current->journal_info == NULL);
1884 oh = container_of(th, struct osd_thandle, ot_super);
1885 LASSERT(oh != NULL);
1886 LASSERT(oh->ot_handle == NULL);
1888 rc = dt_txn_hook_start(env, d, th);
1892 if (unlikely(osd_param_is_not_sane(dev, th))) {
1893 static unsigned long last_printed;
1894 static int last_credits;
1897 * don't make noise on a tiny testing systems
1898 * actual credits misuse will be caught anyway
1900 if (last_credits != oh->ot_credits &&
1901 time_after(jiffies, last_printed +
1902 cfs_time_seconds(60)) &&
1903 osd_transaction_size(dev) > 512) {
1904 CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1905 oh->ot_credits, osd_transaction_size(dev));
1906 osd_trans_dump_creds(env, th);
1907 libcfs_debug_dumpstack(NULL);
1908 last_credits = oh->ot_credits;
1909 last_printed = jiffies;
1912 * XXX Limit the credits to 'max_transaction_buffers', and
1913 * let the underlying filesystem to catch the error if
1914 * we really need so many credits.
1916 * This should be removed when we can calculate the
1917 * credits precisely.
1919 oh->ot_credits = osd_transaction_size(dev);
1920 } else if (ldiskfs_track_declares_assert != 0) {
1922 * reserve few credits to prevent an assertion in JBD
1923 * our debugging mechanism will be able to detected
1924 * overuse. this can help to debug single-update
1927 oh->ot_credits += 10;
1928 if (unlikely(osd_param_is_not_sane(dev, th)))
1929 oh->ot_credits = osd_transaction_size(dev);
1932 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1933 GOTO(out, rc = -EIO);
1936 * XXX temporary stuff. Some abstraction layer should
1939 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1943 LASSERT(oti->oti_txns == 0);
1945 atomic_inc(&dev->od_commit_cb_in_flight);
1946 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1957 static int osd_seq_exists(const struct lu_env *env,
1958 struct osd_device *osd, u64 seq)
1960 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1961 struct seq_server_site *ss = osd_seq_site(osd);
1966 LASSERT(ss != NULL);
1967 LASSERT(ss->ss_server_fld != NULL);
1969 rc = osd_fld_lookup(env, osd, seq, range);
1972 CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1973 osd_name(osd), seq, rc);
1977 RETURN(ss->ss_node_id == range->lsr_index);
1980 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1982 struct dt_txn_commit_cb *dcb;
1983 struct dt_txn_commit_cb *tmp;
1985 /* call per-transaction stop callbacks if any */
1986 list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1988 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1989 "commit callback entry: magic=%x name='%s'\n",
1990 dcb->dcb_magic, dcb->dcb_name);
1991 list_del_init(&dcb->dcb_linkage);
1992 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1997 * Concurrency: shouldn't matter.
1999 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
2002 struct osd_thread_info *oti = osd_oti_get(env);
2003 struct osd_thandle *oh;
2004 struct osd_iobuf *iobuf = &oti->oti_iobuf;
2005 struct osd_device *osd = osd_dt_dev(th->th_dev);
2006 struct qsd_instance *qsd = osd_def_qsd(osd);
2007 struct lquota_trans *qtrans;
2008 LIST_HEAD(truncates);
2009 int rc = 0, remove_agents = 0;
2013 oh = container_of(th, struct osd_thandle, ot_super);
2015 remove_agents = oh->ot_remove_agents;
2017 qtrans = oh->ot_quota_trans;
2018 oh->ot_quota_trans = NULL;
2020 /* move locks to local list, stop tx, execute truncates */
2021 list_splice(&oh->ot_trunc_locks, &truncates);
2023 if (oh->ot_handle != NULL) {
2026 handle_t *hdl = oh->ot_handle;
2029 * add commit callback
2030 * notice we don't do this in osd_trans_start()
2031 * as underlying transaction can change during truncate
2033 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
2036 LASSERT(oti->oti_txns == 1);
2039 rc = dt_txn_hook_stop(env, th);
2041 CERROR("%s: failed in transaction hook: rc = %d\n",
2044 osd_trans_stop_cb(oh, rc);
2045 /* hook functions might modify th_sync */
2046 hdl->h_sync = th->th_sync;
2048 oh->ot_handle = NULL;
2049 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2051 CERROR("%s: failed to stop transaction: rc = %d\n",
2052 osd_name(osd), rc2);
2056 /* We preserve the origin behavior of ignoring any
2057 * failures with the underlying punch / truncate
2058 * operation. We do record for debugging if an error
2059 * does occur in the lctl dk logs.
2061 rc2 = osd_process_truncates(env, &truncates);
2063 CERROR("%s: failed truncate process: rc = %d\n",
2064 osd_name(osd), rc2);
2066 osd_trans_stop_cb(oh, th->th_result);
2070 osd_trunc_unlock_all(env, &truncates);
2072 /* inform the quota slave device that the transaction is stopping */
2073 qsd_op_end(env, qsd, qtrans);
2076 * as we want IO to journal and data IO be concurrent, we don't block
2077 * awaiting data IO completion in osd_do_bio(), instead we wait here
2078 * once transaction is submitted to the journal. all reqular requests
2079 * don't do direct IO (except read/write), thus this wait_event becomes
2082 * IMPORTANT: we have to wait till any IO submited by the thread is
2083 * completed otherwise iobuf may be corrupted by different request
2085 wait_event(iobuf->dr_wait,
2086 atomic_read(&iobuf->dr_numreqs) == 0);
2087 osd_fini_iobuf(osd, iobuf);
2089 rc = iobuf->dr_error;
2091 if (unlikely(remove_agents != 0))
2092 osd_process_scheduled_agent_removals(env, osd);
2094 LASSERT(oti->oti_ins_cache_depth > 0);
2095 oti->oti_ins_cache_depth--;
2096 /* reset OI cache for safety */
2097 if (oti->oti_ins_cache_depth == 0)
2098 oti->oti_ins_cache_used = 0;
2100 sb_end_write(osd_sb(osd));
2105 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2107 struct osd_thandle *oh = container_of(th, struct osd_thandle,
2110 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2111 LASSERT(&dcb->dcb_func != NULL);
2112 if (dcb->dcb_flags & DCB_TRANS_STOP)
2113 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2115 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2121 * Called just before object is freed. Releases all resources except for
2122 * object itself (that is released by osd_object_free()).
2124 * Concurrency: no concurrent access is possible that late in object
2127 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2129 struct osd_object *obj = osd_obj(l);
2130 struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2131 struct inode *inode = obj->oo_inode;
2136 LINVRNT(osd_invariant(obj));
2139 * If object is unlinked remove fid->ino mapping from object index.
2142 osd_index_fini(obj);
2147 if (osd_has_index(obj) && obj->oo_dt.do_index_ops == &osd_index_iam_ops)
2148 ldiskfs_set_inode_flag(inode, LDISKFS_INODE_JOURNAL_DATA);
2150 uid = i_uid_read(inode);
2151 gid = i_gid_read(inode);
2152 projid = i_projid_read(inode);
2154 obj->oo_inode = NULL;
2157 /* do not rebalance quota if the caller needs to release memory
2158 * otherwise qsd_refresh_usage() may went into a new ldiskfs
2159 * transaction and risk to deadlock - LU-12178 */
2160 if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2163 if (!obj->oo_header && qsd) {
2164 struct osd_thread_info *info = osd_oti_get(env);
2165 struct lquota_id_info *qi = &info->oti_qi;
2167 /* Release granted quota to master if necessary */
2168 qi->lqi_id.qid_uid = uid;
2169 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2171 qi->lqi_id.qid_uid = gid;
2172 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2174 qi->lqi_id.qid_uid = projid;
2175 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2180 * Concurrency: ->loo_object_release() is called under site spin-lock.
2182 static void osd_object_release(const struct lu_env *env,
2183 struct lu_object *l)
2185 struct osd_object *o = osd_obj(l);
2188 * nobody should be releasing a non-destroyed object with nlink=0
2189 * the API allows this, but ldiskfs doesn't like and then report
2190 * this inode as deleted
2192 LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2193 o->oo_inode->i_nlink == 0));
2197 * Concurrency: shouldn't matter.
2199 static int osd_object_print(const struct lu_env *env, void *cookie,
2200 lu_printer_t p, const struct lu_object *l)
2202 struct osd_object *o = osd_obj(l);
2203 struct iam_descr *d;
2205 if (o->oo_dir != NULL)
2206 d = o->oo_dir->od_container.ic_descr;
2209 return (*p)(env, cookie,
2210 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2212 o->oo_inode ? o->oo_inode->i_ino : 0UL,
2213 o->oo_inode ? o->oo_inode->i_generation : 0,
2214 d ? d->id_ops->id_name : "plain");
2218 * Concurrency: shouldn't matter.
2220 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2221 struct obd_statfs *sfs, struct obd_statfs_info *info)
2223 struct osd_device *osd = osd_dt_dev(d);
2224 struct super_block *sb = osd_sb(osd);
2225 struct kstatfs *ksfs;
2229 if (unlikely(osd->od_mnt == NULL))
2230 return -EINPROGRESS;
2232 /* osd_lproc.c call this without env, allocate ksfs for that case */
2233 if (unlikely(env == NULL)) {
2234 OBD_ALLOC_PTR(ksfs);
2238 ksfs = &osd_oti_get(env)->oti_ksfs;
2241 result = sb->s_op->statfs(sb->s_root, ksfs);
2245 statfs_pack(sfs, ksfs);
2246 if (unlikely(sb->s_flags & SB_RDONLY))
2247 sfs->os_state |= OS_STATFS_READONLY;
2249 sfs->os_state |= osd->od_nonrotational ? OS_STATFS_NONROT : 0;
2251 if (ldiskfs_has_feature_extents(sb))
2252 sfs->os_maxbytes = sb->s_maxbytes;
2254 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2257 * Reserve some space so to avoid fragmenting the filesystem too much.
2258 * Fragmentation not only impacts performance, but can also increase
2259 * metadata overhead significantly, causing grant calculation to be
2262 * Reserve 0.78% of total space, at least 8MB for small filesystems.
2264 BUILD_BUG_ON(OSD_STATFS_RESERVED <= LDISKFS_MAX_BLOCK_SIZE);
2265 reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2266 if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2267 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2269 sfs->os_blocks -= reserved;
2270 sfs->os_bfree -= min(reserved, sfs->os_bfree);
2271 sfs->os_bavail -= min(reserved, sfs->os_bavail);
2274 if (unlikely(env == NULL))
2280 * Estimate space needed for file creations. We assume the largest filename
2281 * which is 2^64 - 1, hence a filename of 20 chars.
2282 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2284 #ifdef __LDISKFS_DIR_REC_LEN
2285 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2287 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2291 * Concurrency: doesn't access mutable data.
2293 static void osd_conf_get(const struct lu_env *env,
2294 const struct dt_device *dev,
2295 struct dt_device_param *param)
2297 struct osd_device *d = osd_dt_dev(dev);
2298 struct super_block *sb = osd_sb(d);
2299 struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2304 * XXX should be taken from not-yet-existing fs abstraction layer.
2306 param->ddp_max_name_len = LDISKFS_NAME_LEN;
2307 param->ddp_max_nlink = LDISKFS_LINK_MAX;
2308 param->ddp_symlink_max = sb->s_blocksize;
2309 param->ddp_mount_type = LDD_MT_LDISKFS;
2310 if (ldiskfs_has_feature_extents(sb))
2311 param->ddp_maxbytes = sb->s_maxbytes;
2313 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2315 * inode are statically allocated, so per-inode space consumption
2316 * is the space consumed by the directory entry
2318 param->ddp_inodespace = PER_OBJ_USAGE;
2320 * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2321 * is 128MB) which is unlikely to be hit in real life. Report a smaller
2322 * maximum length to not under-count the actual number of extents
2323 * needed for writing a file if there are sub-optimal block allocations.
2325 param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2326 /* worst-case extent insertion metadata overhead */
2327 param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2328 param->ddp_mntopts = 0;
2329 if (test_opt(sb, XATTR_USER))
2330 param->ddp_mntopts |= MNTOPT_USERXATTR;
2331 if (test_opt(sb, POSIX_ACL))
2332 param->ddp_mntopts |= MNTOPT_ACL;
2335 * LOD might calculate the max stripe count based on max_ea_size,
2336 * so we need take account in the overhead as well,
2337 * xattr_header + magic + xattr_entry_head
2339 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2340 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2342 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2343 if (ldiskfs_has_feature_ea_inode(sb))
2344 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2348 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2350 if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE)
2351 param->ddp_max_ea_size = OBD_MAX_EA_SIZE;
2354 * Preferred RPC size for efficient disk IO. 4MB shows good
2355 * all-around performance for ldiskfs, but use bigalloc chunk size
2356 * by default if larger.
2358 #if defined(LDISKFS_CLUSTER_SIZE)
2359 if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2360 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2363 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2365 param->ddp_t10_cksum_type = 0;
2367 unsigned short interval = blk_integrity_interval(bi);
2368 name = blk_integrity_name(bi);
2376 if (strncmp(name, "T10-DIF-TYPE",
2377 sizeof("T10-DIF-TYPE") - 1) == 0) {
2378 /* also skip "1/3-" at end */
2379 const int type_off = sizeof("T10-DIF-TYPE.");
2380 char type_number = name[type_off - 2];
2382 if (interval != 512 && interval != 4096) {
2383 CERROR("%s: unsupported T10PI sector size %u\n",
2384 d->od_svname, interval);
2385 } else if (type_number != '1' && type_number != '3') {
2386 CERROR("%s: unsupported T10PI type %s\n",
2387 d->od_svname, name);
2388 } else if (strcmp(name + type_off, "CRC") == 0) {
2389 d->od_t10_type = type_number == '1' ?
2390 OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2391 param->ddp_t10_cksum_type = interval == 512 ?
2392 OBD_CKSUM_T10CRC512 :
2394 } else if (strcmp(name + type_off, "IP") == 0) {
2395 d->od_t10_type = type_number == '1' ?
2396 OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2397 param->ddp_t10_cksum_type = interval == 512 ?
2398 OBD_CKSUM_T10IP512 :
2401 CERROR("%s: unsupported checksum type of T10PI type '%s'\n",
2402 d->od_svname, name);
2406 CERROR("%s: unsupported T10PI type '%s'\n",
2407 d->od_svname, name);
2411 param->ddp_has_lseek_data_hole = true;
2414 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2416 return osd_sb(osd_dt_dev(d));
2420 * Concurrency: shouldn't matter.
2422 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2425 struct super_block *s = osd_sb(osd_dt_dev(d));
2428 down_read(&s->s_umount);
2429 rc = s->s_op->sync_fs(s, 1);
2430 up_read(&s->s_umount);
2432 CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2439 * Start commit for OSD device.
2441 * An implementation of dt_commit_async method for OSD device.
2442 * Asychronously starts underlayng fs sync and thereby a transaction
2445 * \param env environment
2446 * \param d dt device
2448 * \see dt_device_operations
2450 static int osd_commit_async(const struct lu_env *env,
2451 struct dt_device *d)
2453 struct super_block *s = osd_sb(osd_dt_dev(d));
2458 CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2459 down_read(&s->s_umount);
2460 rc = s->s_op->sync_fs(s, 0);
2461 up_read(&s->s_umount);
2467 * Concurrency: shouldn't matter.
2469 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2471 struct super_block *sb = osd_sb(osd_dt_dev(d));
2472 struct block_device *dev = sb->s_bdev;
2473 int rc = -EOPNOTSUPP;
2477 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2478 osd_dt_dev(d)->od_svname, (long)dev, rc);
2484 * Note: we do not count into QUOTA here.
2485 * If we mount with --data_journal we may need more.
2487 const int osd_dto_credits_noquota[DTO_NR] = {
2490 * INDEX_EXTRA_TRANS_BLOCKS(8) +
2491 * SINGLEDATA_TRANS_BLOCKS(8)
2492 * XXX Note: maybe iam need more, since iam have more level than
2495 [DTO_INDEX_INSERT] = 16,
2498 * just modify a single entry, probably merge few within a block
2500 [DTO_INDEX_DELETE] = 1,
2504 [DTO_INDEX_UPDATE] = 16,
2506 * 4(inode, inode bits, groups, GDT)
2507 * notice: OI updates are counted separately with DTO_INDEX_INSERT
2509 [DTO_OBJECT_CREATE] = 4,
2511 * 4(inode, inode bits, groups, GDT)
2512 * notice: OI updates are counted separately with DTO_INDEX_DELETE
2514 [DTO_OBJECT_DELETE] = 4,
2516 * Attr set credits (inode)
2518 [DTO_ATTR_SET_BASE] = 1,
2520 * Xattr set. The same as xattr of EXT3.
2521 * DATA_TRANS_BLOCKS(14)
2522 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2523 * are also counted in. Do not know why?
2525 [DTO_XATTR_SET] = 14,
2527 * credits for inode change during write.
2529 [DTO_WRITE_BASE] = 3,
2531 * credits for single block write.
2533 [DTO_WRITE_BLOCK] = 14,
2535 * Attr set credits for chown.
2536 * This is extra credits for setattr, and it is null without quota
2538 [DTO_ATTR_SET_CHOWN] = 0
2541 /* reserve or free quota for some operation */
2542 static int osd_reserve_or_free_quota(const struct lu_env *env,
2543 struct dt_device *dev,
2544 enum quota_type type, __u64 uid,
2545 __u64 gid, __s64 count, bool is_md)
2548 struct osd_device *osd = osd_dt_dev(dev);
2549 struct osd_thread_info *info = osd_oti_get(env);
2550 struct lquota_id_info *qi = &info->oti_qi;
2551 struct qsd_instance *qsd = NULL;
2556 qsd = osd->od_quota_slave_md;
2558 qsd = osd->od_quota_slave_dt;
2560 rc = quota_reserve_or_free(env, qsd, qi, type, uid, gid, count, is_md);
2564 static const struct dt_device_operations osd_dt_ops = {
2565 .dt_root_get = osd_root_get,
2566 .dt_statfs = osd_statfs,
2567 .dt_trans_create = osd_trans_create,
2568 .dt_trans_start = osd_trans_start,
2569 .dt_trans_stop = osd_trans_stop,
2570 .dt_trans_cb_add = osd_trans_cb_add,
2571 .dt_conf_get = osd_conf_get,
2572 .dt_mnt_sb_get = osd_mnt_sb_get,
2573 .dt_sync = osd_sync,
2575 .dt_commit_async = osd_commit_async,
2576 .dt_reserve_or_free_quota = osd_reserve_or_free_quota,
2579 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2582 struct osd_object *obj = osd_dt_obj(dt);
2583 struct osd_thread_info *oti = osd_oti_get(env);
2585 LINVRNT(osd_invariant(obj));
2587 LASSERT(obj->oo_owner != env);
2588 down_read_nested(&obj->oo_sem, role);
2590 LASSERT(obj->oo_owner == NULL);
2594 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2597 struct osd_object *obj = osd_dt_obj(dt);
2598 struct osd_thread_info *oti = osd_oti_get(env);
2600 LINVRNT(osd_invariant(obj));
2602 LASSERT(obj->oo_owner != env);
2603 down_write_nested(&obj->oo_sem, role);
2605 LASSERT(obj->oo_owner == NULL);
2606 obj->oo_owner = env;
2610 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2612 struct osd_object *obj = osd_dt_obj(dt);
2613 struct osd_thread_info *oti = osd_oti_get(env);
2615 LINVRNT(osd_invariant(obj));
2617 LASSERT(oti->oti_r_locks > 0);
2619 up_read(&obj->oo_sem);
2622 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2624 struct osd_object *obj = osd_dt_obj(dt);
2625 struct osd_thread_info *oti = osd_oti_get(env);
2627 LINVRNT(osd_invariant(obj));
2629 LASSERT(obj->oo_owner == env);
2630 LASSERT(oti->oti_w_locks > 0);
2632 obj->oo_owner = NULL;
2633 up_write(&obj->oo_sem);
2636 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2638 struct osd_object *obj = osd_dt_obj(dt);
2640 LINVRNT(osd_invariant(obj));
2642 return obj->oo_owner == env;
2645 static void osd_inode_getattr(const struct lu_env *env,
2646 struct inode *inode, struct lu_attr *attr)
2648 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2649 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2650 LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2651 LA_BLKSIZE | LA_TYPE | LA_BTIME;
2653 attr->la_atime = inode->i_atime.tv_sec;
2654 attr->la_mtime = inode->i_mtime.tv_sec;
2655 attr->la_ctime = inode->i_ctime.tv_sec;
2656 attr->la_btime = LDISKFS_I(inode)->i_crtime.tv_sec;
2657 attr->la_mode = inode->i_mode;
2658 attr->la_size = i_size_read(inode);
2659 attr->la_blocks = inode->i_blocks;
2660 attr->la_uid = i_uid_read(inode);
2661 attr->la_gid = i_gid_read(inode);
2662 attr->la_projid = i_projid_read(inode);
2663 attr->la_flags = ll_inode_to_ext_flags(inode->i_flags);
2664 attr->la_nlink = inode->i_nlink;
2665 attr->la_rdev = inode->i_rdev;
2666 attr->la_blksize = 1 << inode->i_blkbits;
2667 attr->la_blkbits = inode->i_blkbits;
2669 * Ext4 did not transfer inherit flags from raw inode
2670 * to inode flags, and ext4 internally test raw inode
2671 * @i_flags directly. Instead of patching ext4, we do it here.
2673 if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2674 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2677 static int osd_dirent_count(const struct lu_env *env, struct dt_object *dt,
2680 struct osd_object *obj = osd_dt_obj(dt);
2681 const struct dt_it_ops *iops;
2687 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2688 LASSERT(fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)));
2690 if (obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
2691 *count = obj->oo_dirent_count;
2695 /* directory not initialized yet */
2696 if (!dt->do_index_ops) {
2701 iops = &dt->do_index_ops->dio_it;
2702 it = iops->init(env, dt, LUDA_64BITHASH);
2704 RETURN(PTR_ERR(it));
2706 rc = iops->load(env, it, 0);
2708 if (rc == -ENODATA) {
2715 rc = iops->next(env, it);
2717 for (*count = 0; rc == 0 || rc == -ESTALE; rc = iops->next(env, it)) {
2721 if (iops->key_size(env, it) == 0)
2727 obj->oo_dirent_count = *count;
2732 iops->fini(env, it);
2737 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2738 struct lu_attr *attr)
2740 struct osd_object *obj = osd_dt_obj(dt);
2743 if (unlikely(!dt_object_exists(dt)))
2745 if (unlikely(obj->oo_destroyed))
2748 LASSERT(!dt_object_remote(dt));
2749 LINVRNT(osd_invariant(obj));
2751 spin_lock(&obj->oo_guard);
2752 osd_inode_getattr(env, obj->oo_inode, attr);
2753 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2754 attr->la_valid |= LA_FLAGS;
2755 attr->la_flags |= LUSTRE_ORPHAN_FL;
2757 if (obj->oo_lma_flags & LUSTRE_ENCRYPT_FL) {
2758 attr->la_valid |= LA_FLAGS;
2759 attr->la_flags |= LUSTRE_ENCRYPT_FL;
2761 spin_unlock(&obj->oo_guard);
2763 if (S_ISDIR(obj->oo_inode->i_mode) &&
2764 fid_is_namespace_visible(lu_object_fid(&dt->do_lu)))
2765 rc = osd_dirent_count(env, dt, &attr->la_dirent_count);
2770 static int osd_declare_attr_qid(const struct lu_env *env,
2771 struct osd_object *obj,
2772 struct osd_thandle *oh, long long bspace,
2773 qid_t old_id, qid_t new_id, bool enforce,
2777 struct osd_thread_info *info = osd_oti_get(env);
2778 struct lquota_id_info *qi = &info->oti_qi;
2780 qi->lqi_type = type;
2781 /* inode accounting */
2782 qi->lqi_is_blk = false;
2784 /* one more inode for the new id ... */
2785 qi->lqi_id.qid_uid = new_id;
2787 /* Reserve credits for the new id */
2788 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2789 if (rc == -EDQUOT || rc == -EINPROGRESS)
2794 /* and one less inode for the current id */
2795 qi->lqi_id.qid_uid = old_id;
2797 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2798 if (rc == -EDQUOT || rc == -EINPROGRESS)
2803 /* block accounting */
2804 qi->lqi_is_blk = true;
2806 /* more blocks for the new id ... */
2807 qi->lqi_id.qid_uid = new_id;
2808 qi->lqi_space = bspace;
2810 * Credits for the new uid has been reserved, re-use "obj"
2811 * to save credit reservation.
2813 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2814 if (rc == -EDQUOT || rc == -EINPROGRESS)
2819 /* and finally less blocks for the current uid */
2820 qi->lqi_id.qid_uid = old_id;
2821 qi->lqi_space = -bspace;
2822 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2823 if (rc == -EDQUOT || rc == -EINPROGRESS)
2829 static int osd_declare_attr_set(const struct lu_env *env,
2830 struct dt_object *dt,
2831 const struct lu_attr *attr,
2832 struct thandle *handle)
2834 struct osd_thandle *oh;
2835 struct osd_object *obj;
2844 LASSERT(dt != NULL);
2845 LASSERT(handle != NULL);
2847 obj = osd_dt_obj(dt);
2848 LASSERT(osd_invariant(obj));
2850 oh = container_of(handle, struct osd_thandle, ot_super);
2851 LASSERT(oh->ot_handle == NULL);
2853 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2854 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2856 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2857 osd_dto_credits_noquota[DTO_XATTR_SET]);
2859 if (attr == NULL || obj->oo_inode == NULL)
2862 bspace = obj->oo_inode->i_blocks << 9;
2863 bspace = toqb(bspace);
2866 * Changing ownership is always preformed by super user, it should not
2867 * fail with EDQUOT unless required explicitly.
2869 * We still need to call the osd_declare_qid() to calculate the journal
2870 * credits for updating quota accounting files and to trigger quota
2871 * space adjustment once the operation is completed.
2873 if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2875 uid = i_uid_read(obj->oo_inode);
2876 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2877 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2878 attr->la_uid, enforce, USRQUOTA);
2882 gid = i_gid_read(obj->oo_inode);
2883 CDEBUG(D_QUOTA, "declare uid %d -> %d gid %d -> %d\n", uid,
2884 attr->la_uid, gid, attr->la_gid);
2885 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2886 rc = osd_declare_attr_qid(env, obj, oh, bspace, gid,
2887 attr->la_gid, enforce, GRPQUOTA);
2892 #ifdef HAVE_PROJECT_QUOTA
2893 if (attr->la_valid & LA_PROJID) {
2894 __u32 projid = i_projid_read(obj->oo_inode);
2896 enforce = (attr->la_valid & LA_PROJID) &&
2897 (attr->la_projid != projid);
2898 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2899 (qid_t)projid, (qid_t)attr->la_projid,
2908 static int osd_inode_setattr(const struct lu_env *env,
2909 struct inode *inode, const struct lu_attr *attr)
2911 __u64 bits = attr->la_valid;
2913 /* Only allow set size for regular file */
2914 if (!S_ISREG(inode->i_mode))
2915 bits &= ~(LA_SIZE | LA_BLOCKS);
2920 if (bits & LA_ATIME)
2921 inode->i_atime = osd_inode_time(inode, attr->la_atime);
2922 if (bits & LA_CTIME)
2923 inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
2924 if (bits & LA_MTIME)
2925 inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
2926 if (bits & LA_SIZE) {
2927 spin_lock(&inode->i_lock);
2928 LDISKFS_I(inode)->i_disksize = attr->la_size;
2929 i_size_write(inode, attr->la_size);
2930 spin_unlock(&inode->i_lock);
2934 * OSD should not change "i_blocks" which is used by quota.
2935 * "i_blocks" should be changed by ldiskfs only.
2938 inode->i_mode = (inode->i_mode & S_IFMT) |
2939 (attr->la_mode & ~S_IFMT);
2941 i_uid_write(inode, attr->la_uid);
2943 i_gid_write(inode, attr->la_gid);
2944 if (bits & LA_PROJID)
2945 i_projid_write(inode, attr->la_projid);
2946 if (bits & LA_NLINK)
2947 set_nlink(inode, attr->la_nlink);
2949 inode->i_rdev = attr->la_rdev;
2951 if (bits & LA_FLAGS) {
2952 /* always keep S_NOCMTIME */
2953 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2955 #if defined(S_ENCRYPTED)
2956 /* Always remove S_ENCRYPTED, because ldiskfs must not be
2957 * aware of encryption status. It is just stored into LMA
2958 * so that it can be forwared to client side.
2960 inode->i_flags &= ~S_ENCRYPTED;
2963 * Ext4 did not transfer inherit flags from
2964 * @inode->i_flags to raw inode i_flags when writing
2965 * flags, we do it explictly here.
2967 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2968 LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2970 LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2975 #ifdef HAVE_PROJECT_QUOTA
2976 static int osd_transfer_project(struct inode *inode, __u32 projid,
2977 struct thandle *handle)
2979 struct super_block *sb = inode->i_sb;
2980 struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2983 struct ldiskfs_iloc iloc;
2984 struct ldiskfs_inode *raw_inode;
2985 struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2987 if (!ldiskfs_has_feature_project(sb)) {
2988 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2989 == LDISKFS_DEF_PROJID);
2990 if (projid != LDISKFS_DEF_PROJID)
2996 if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2999 kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
3000 if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
3003 err = ldiskfs_get_inode_loc(inode, &iloc);
3007 raw_inode = ldiskfs_raw_inode(&iloc);
3008 if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
3009 struct osd_thandle *oh = container_of(handle,
3013 * try to expand inode size automatically.
3015 ldiskfs_mark_inode_dirty(oh->ot_handle, inode);
3016 if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
3024 dquot_initialize(inode);
3025 transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
3026 if (transfer_to[PRJQUOTA]) {
3027 lock_dquot_transfer(inode);
3028 err = __dquot_transfer(inode, transfer_to);
3029 unlock_dquot_transfer(inode);
3030 dqput(transfer_to[PRJQUOTA]);
3039 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr,
3040 struct thandle *handle)
3044 if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
3045 (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
3049 "executing dquot_transfer inode %ld uid %d -> %d gid %d -> %d\n",
3050 inode->i_ino, i_uid_read(inode), attr->la_uid,
3051 i_gid_read(inode), attr->la_gid);
3053 dquot_initialize(inode);
3055 if (attr->la_valid & LA_UID)
3056 iattr.ia_valid |= ATTR_UID;
3057 if (attr->la_valid & LA_GID)
3058 iattr.ia_valid |= ATTR_GID;
3059 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
3060 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
3062 lock_dquot_transfer(inode);
3063 rc = dquot_transfer(inode, &iattr);
3064 unlock_dquot_transfer(inode);
3066 CERROR("%s: quota transfer failed. Is quota enforcement enabled on the ldiskfs filesystem? rc = %d\n",
3067 osd_ino2name(inode), rc);
3072 /* Handle project id transfer here properly */
3073 if (attr->la_valid & LA_PROJID &&
3074 attr->la_projid != i_projid_read(inode)) {
3075 if (!projid_valid(make_kprojid(&init_user_ns, attr->la_projid)))
3077 #ifdef HAVE_PROJECT_QUOTA
3078 rc = osd_transfer_project(inode, attr->la_projid, handle);
3083 CERROR("%s: quota transfer failed. Is project enforcement enabled on the ldiskfs filesystem? rc = %d\n",
3084 osd_ino2name(inode), rc);
3091 static int osd_attr_set(const struct lu_env *env,
3092 struct dt_object *dt,
3093 const struct lu_attr *attr,
3094 struct thandle *handle)
3096 struct osd_object *obj = osd_dt_obj(dt);
3097 struct inode *inode;
3100 if (!dt_object_exists(dt))
3103 LASSERT(handle != NULL);
3104 LASSERT(!dt_object_remote(dt));
3105 LASSERT(osd_invariant(obj));
3107 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
3109 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
3110 !osd_obj2dev(obj)->od_is_ost) {
3111 struct osd_thread_info *oti = osd_oti_get(env);
3112 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
3113 struct lu_fid *fid1 = &oti->oti_fid;
3114 struct osd_inode_id *id = &oti->oti_id;
3115 struct iam_path_descr *ipd;
3116 struct iam_container *bag;
3117 struct osd_thandle *oh;
3120 fid_cpu_to_be(fid1, fid0);
3121 memset(id, 1, sizeof(*id));
3122 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
3123 fid0)->oi_dir.od_container;
3124 ipd = osd_idx_ipd_get(env, bag);
3125 if (unlikely(ipd == NULL))
3128 oh = container_of(handle, struct osd_thandle, ot_super);
3129 rc = iam_update(oh->ot_handle, bag,
3130 (const struct iam_key *)fid1,
3131 (const struct iam_rec *)id, ipd);
3132 osd_ipd_put(env, bag, ipd);
3133 return(rc > 0 ? 0 : rc);
3136 inode = obj->oo_inode;
3138 rc = osd_quota_transfer(inode, attr, handle);
3142 spin_lock(&obj->oo_guard);
3143 rc = osd_inode_setattr(env, inode, attr);
3144 spin_unlock(&obj->oo_guard);
3148 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3150 osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3152 if (!(attr->la_valid & LA_FLAGS))
3155 /* Let's check if there are extra flags need to be set into LMA */
3156 if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3157 struct osd_thread_info *info = osd_oti_get(env);
3158 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3160 LASSERT(!obj->oo_pfid_in_lma);
3162 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3163 &info->oti_ost_attrs);
3167 lma->lma_incompat |=
3168 lustre_to_lma_flags(attr->la_flags);
3169 lustre_lma_swab(lma);
3171 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3173 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3174 lma, sizeof(*lma), XATTR_REPLACE);
3176 struct osd_device *osd = osd_obj2dev(obj);
3178 CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3179 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3180 lma->lma_incompat, rc);
3183 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3185 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3192 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3193 struct osd_object *obj,
3194 const char *name, const int namelen)
3196 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3199 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3200 umode_t mode, struct dt_allocation_hint *hint,
3201 struct thandle *th, struct lu_attr *attr)
3204 struct osd_device *osd = osd_obj2dev(obj);
3205 struct osd_thandle *oth;
3206 struct dt_object *parent = NULL;
3207 struct inode *inode;
3208 struct iattr iattr = {
3209 .ia_valid = ATTR_UID | ATTR_GID |
3210 ATTR_CTIME | ATTR_MTIME | ATTR_ATIME,
3211 .ia_ctime.tv_sec = attr->la_ctime,
3212 .ia_mtime.tv_sec = attr->la_mtime,
3213 .ia_atime.tv_sec = attr->la_atime,
3214 .ia_uid = GLOBAL_ROOT_UID,
3215 .ia_gid = GLOBAL_ROOT_GID,
3217 const struct osd_timespec omit = { .tv_nsec = UTIME_OMIT };
3219 if (attr->la_valid & LA_UID)
3220 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
3221 if (attr->la_valid & LA_GID)
3222 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
3224 LINVRNT(osd_invariant(obj));
3225 LASSERT(obj->oo_inode == NULL);
3226 LASSERT(obj->oo_hl_head == NULL);
3228 if (S_ISDIR(mode) && ldiskfs_pdo) {
3230 ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3231 if (obj->oo_hl_head == NULL)
3235 oth = container_of(th, struct osd_thandle, ot_super);
3236 LASSERT(oth->ot_handle->h_transaction != NULL);
3238 if (hint != NULL && hint->dah_parent != NULL &&
3239 !dt_object_remote(hint->dah_parent))
3240 parent = hint->dah_parent;
3242 /* if a time component is not valid set it to UTIME_OMIT */
3243 if (!(attr->la_valid & LA_CTIME))
3244 iattr.ia_ctime = omit;
3245 if (!(attr->la_valid & LA_MTIME))
3246 iattr.ia_mtime = omit;
3247 if (!(attr->la_valid & LA_ATIME))
3248 iattr.ia_atime = omit;
3250 inode = ldiskfs_create_inode(oth->ot_handle,
3251 parent ? osd_dt_obj(parent)->oo_inode :
3252 osd_sb(osd)->s_root->d_inode,
3254 if (!IS_ERR(inode)) {
3255 /* Do not update file c/mtime in ldiskfs. */
3256 inode->i_flags |= S_NOCMTIME;
3259 * For new created object, it must be consistent,
3260 * and it is unnecessary to scrub against it.
3262 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3264 obj->oo_inode = inode;
3267 if (obj->oo_hl_head != NULL) {
3268 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3269 obj->oo_hl_head = NULL;
3271 result = PTR_ERR(inode);
3273 LINVRNT(osd_invariant(obj));
3281 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3282 struct lu_attr *attr,
3283 struct dt_allocation_hint *hint,
3284 struct dt_object_format *dof,
3288 struct osd_thandle *oth;
3289 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3291 LASSERT(S_ISDIR(attr->la_mode));
3293 oth = container_of(th, struct osd_thandle, ot_super);
3294 LASSERT(oth->ot_handle->h_transaction != NULL);
3295 if (fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)))
3296 obj->oo_dirent_count = 0;
3297 result = osd_mkfile(info, obj, mode, hint, th, attr);
3302 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3303 struct lu_attr *attr,
3304 struct dt_allocation_hint *hint,
3305 struct dt_object_format *dof,
3309 struct osd_thandle *oth;
3310 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3312 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3314 LASSERT(S_ISREG(attr->la_mode));
3316 oth = container_of(th, struct osd_thandle, ot_super);
3317 LASSERT(oth->ot_handle->h_transaction != NULL);
3319 result = osd_mkfile(info, obj, mode, hint, th, attr);
3321 LASSERT(obj->oo_inode != NULL);
3322 if (feat->dif_flags & DT_IND_VARKEY)
3323 result = iam_lvar_create(obj->oo_inode,
3324 feat->dif_keysize_max,
3326 feat->dif_recsize_max,
3329 result = iam_lfix_create(obj->oo_inode,
3330 feat->dif_keysize_max,
3332 feat->dif_recsize_max,
3338 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3339 struct lu_attr *attr,
3340 struct dt_allocation_hint *hint,
3341 struct dt_object_format *dof,
3344 LASSERT(S_ISREG(attr->la_mode));
3345 return osd_mkfile(info, obj, (attr->la_mode &
3346 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3350 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3351 struct lu_attr *attr,
3352 struct dt_allocation_hint *hint,
3353 struct dt_object_format *dof,
3356 LASSERT(S_ISLNK(attr->la_mode));
3357 return osd_mkfile(info, obj, (attr->la_mode &
3358 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3362 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3363 struct lu_attr *attr,
3364 struct dt_allocation_hint *hint,
3365 struct dt_object_format *dof,
3368 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3371 LINVRNT(osd_invariant(obj));
3372 LASSERT(obj->oo_inode == NULL);
3373 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3374 S_ISFIFO(mode) || S_ISSOCK(mode));
3376 result = osd_mkfile(info, obj, mode, hint, th, attr);
3378 LASSERT(obj->oo_inode != NULL);
3380 * This inode should be marked dirty for i_rdev. Currently
3381 * that is done in the osd_attr_init().
3383 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3386 LINVRNT(osd_invariant(obj));
3390 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3392 struct dt_allocation_hint *hint,
3393 struct dt_object_format *dof,
3396 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3398 osd_obj_type_f result;
3414 result = osd_mk_index;
3424 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3425 struct dt_object *parent, struct dt_object *child,
3430 ah->dah_parent = parent;
3431 ah->dah_mode = child_mode;
3433 if (parent != NULL && !dt_object_remote(parent)) {
3434 /* will help to find FID->ino at dt_insert("..") */
3435 struct osd_object *pobj = osd_dt_obj(parent);
3437 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3441 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3442 struct lu_attr *attr, struct dt_object_format *dof,
3443 struct thandle *handle)
3445 struct inode *inode = obj->oo_inode;
3446 __u64 valid = attr->la_valid;
3449 attr->la_valid &= ~(LA_TYPE | LA_MODE);
3451 if (dof->dof_type != DFT_NODE)
3452 attr->la_valid &= ~LA_RDEV;
3453 if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3454 attr->la_valid &= ~LA_ATIME;
3455 if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3456 attr->la_valid &= ~LA_CTIME;
3457 if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3458 attr->la_valid &= ~LA_MTIME;
3460 result = osd_quota_transfer(inode, attr, handle);
3464 if (attr->la_valid != 0) {
3465 result = osd_inode_setattr(info->oti_env, inode, attr);
3467 * The osd_inode_setattr() should always succeed here. The
3468 * only error that could be returned is EDQUOT when we are
3469 * trying to change the UID or GID of the inode. However, this
3470 * should not happen since quota enforcement is no longer
3471 * enabled on ldiskfs (lquota takes care of it).
3473 LASSERTF(result == 0, "%d\n", result);
3474 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3477 attr->la_valid = valid;
3481 * Helper function for osd_create()
3483 * \retval 0, on success
3485 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3486 struct lu_attr *attr, struct dt_allocation_hint *hint,
3487 struct dt_object_format *dof, struct thandle *th)
3492 osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3494 /* we drop umask so that permissions we pass are not affected */
3495 umask = current->fs->umask;
3496 current->fs->umask = 0;
3498 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3500 if (likely(obj->oo_inode != NULL)) {
3501 LASSERT(obj->oo_inode->i_state & I_NEW);
3504 * Unlock the inode before attr initialization to avoid
3505 * unnecessary dqget operations. LU-6378
3507 unlock_new_inode(obj->oo_inode);
3510 if (likely(result == 0)) {
3511 osd_attr_init(info, obj, attr, dof, th);
3512 osd_object_init0(obj);
3515 /* restore previous umask value */
3516 current->fs->umask = umask;
3518 osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3524 * Helper function for osd_create()
3526 * \retval 0, on success
3528 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3529 const struct lu_fid *fid, struct thandle *th)
3531 struct osd_thread_info *info = osd_oti_get(env);
3532 struct osd_inode_id *id = &info->oti_id;
3533 struct osd_device *osd = osd_obj2dev(obj);
3534 struct osd_thandle *oh;
3537 LASSERT(obj->oo_inode != NULL);
3539 if (CFS_FAIL_CHECK(OBD_FAIL_OSD_OI_ENOSPC))
3542 oh = container_of(th, struct osd_thandle, ot_super);
3543 LASSERT(oh->ot_handle);
3544 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3546 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3547 rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3548 OI_CHECK_FLD, NULL);
3549 if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP) && osd->od_is_ost) {
3550 struct lu_fid next_fid = *fid;
3552 /* insert next object in advance, and map to the same inode */
3554 if (next_fid.f_oid != 0) {
3555 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3556 osd_oi_insert(info, osd, &next_fid, id, oh->ot_handle,
3557 OI_CHECK_FLD, NULL);
3558 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3562 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3567 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3568 u64 seq, struct lu_seq_range *range)
3570 struct seq_server_site *ss = osd_seq_site(osd);
3572 if (fid_seq_is_idif(seq)) {
3573 fld_range_set_ost(range);
3574 range->lsr_index = idif_ost_idx(seq);
3578 if (!fid_seq_in_fldb(seq)) {
3579 fld_range_set_mdt(range);
3582 * FIXME: If ss is NULL, it suppose not get lsr_index
3585 range->lsr_index = ss->ss_node_id;
3589 LASSERT(ss != NULL);
3590 fld_range_set_any(range);
3591 /* OSD will only do local fld lookup */
3592 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3595 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3596 struct lu_attr *attr,
3597 struct dt_allocation_hint *hint,
3598 struct dt_object_format *dof,
3599 struct thandle *handle)
3601 struct osd_thandle *oh;
3602 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3608 LASSERT(handle != NULL);
3610 oh = container_of(handle, struct osd_thandle, ot_super);
3611 LASSERT(oh->ot_handle == NULL);
3614 * EA object consumes more credits than regular object: osd_mk_index
3615 * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3616 * leaf_node, could involves the block, block bitmap, groups, GDT
3617 * change for each block, so add 4 * 2 credits in that case.
3619 * The default ACL initialization may consume an additional 16 blocks
3621 credits = osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3622 ((dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3625 * While ldiskfs_new_inode() calls ldiskfs_init_acl() we have to add
3626 * credits for possible default ACL creation in new inode
3628 if (hint && hint->dah_acl_len)
3629 credits += osd_calc_bkmap_credits(sb, NULL, 0, -1,
3630 (hint->dah_acl_len + sb->s_blocksize - 1) >>
3631 sb->s_blocksize_bits);
3633 osd_trans_declare_op(env, oh, OSD_OT_CREATE, credits);
3636 * Reuse idle OI block may cause additional one OI block
3639 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3640 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3641 if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP))
3642 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3643 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3645 /* will help to find FID->ino mapping at dt_insert() */
3646 rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3654 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3655 attr->la_projid, 1, oh, osd_dt_obj(dt),
3656 NULL, OSD_QID_INODE);
3664 * Called to destroy on-disk representation of the object
3666 * Concurrency: must be locked
3668 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3671 struct osd_object *obj = osd_dt_obj(dt);
3672 struct inode *inode = obj->oo_inode;
3673 struct osd_thandle *oh;
3681 oh = container_of(th, struct osd_thandle, ot_super);
3682 LASSERT(oh->ot_handle == NULL);
3684 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3685 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3687 /* For removing agent entry */
3688 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3689 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3692 * Recycle idle OI leaf may cause additional three OI blocks
3695 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3696 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3697 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3698 /* one less inode */
3699 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3700 i_projid_read(inode), -1, oh, obj, NULL,
3704 /* data to be truncated */
3705 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3706 i_projid_read(inode), 0, oh, obj, NULL,
3712 * will help to find FID->ino when this object is being
3715 rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3720 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3723 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3724 struct osd_object *obj = osd_dt_obj(dt);
3725 struct inode *inode = obj->oo_inode;
3726 struct osd_device *osd = osd_obj2dev(obj);
3727 struct osd_thandle *oh;
3732 oh = container_of(th, struct osd_thandle, ot_super);
3733 LASSERT(oh->ot_handle);
3735 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3737 if (unlikely(fid_is_acct(fid)))
3740 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3741 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3743 CERROR("%s: remove agent entry "DFID": rc = %d\n",
3744 osd_name(osd), PFID(fid), result);
3747 if (S_ISDIR(inode->i_mode)) {
3748 if (inode->i_nlink > 2)
3749 CERROR("%s: directory "DFID" ino %lu link count is %u at unlink. run e2fsck to repair\n",
3750 osd_name(osd), PFID(fid), inode->i_ino,
3753 spin_lock(&obj->oo_guard);
3755 spin_unlock(&obj->oo_guard);
3756 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3759 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3761 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3763 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3764 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3765 oh->ot_handle, OI_CHECK_FLD);
3767 osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3768 /* XXX: add to ext3 orphan list */
3769 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3771 /* not needed in the cache anymore */
3772 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3773 obj->oo_destroyed = 1;
3779 * Put the fid into lustre_mdt_attrs, and then place the structure
3780 * inode's ea. This fid should not be altered during the life time
3783 * \retval +ve, on success
3784 * \retval -ve, on error
3786 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3788 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3789 const struct lu_fid *fid, __u32 compat, __u32 incompat)
3791 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3792 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3797 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3800 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3803 lustre_loa_init(loa, fid, compat, incompat);
3804 lustre_loa_swab(loa, false);
3807 * For the OST device with 256 bytes inode size by default,
3808 * the PFID EA will be stored together with LMA EA to avoid
3809 * performance trouble. Otherwise the PFID EA can be stored
3810 * independently. LU-8998
3812 if ((compat & LMAC_FID_ON_OST) &&
3813 LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3814 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3815 sizeof(*loa), XATTR_CREATE);
3817 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3818 sizeof(*lma), XATTR_CREATE);
3820 * LMA may already exist, but we need to check that all the
3821 * desired compat/incompat flags have been added.
3823 if (unlikely(rc == -EEXIST)) {
3824 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3825 XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3829 if (rc < sizeof(*lma))
3832 lustre_loa_swab(loa, true);
3833 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3834 ((compat == 0 && incompat == 0) ||
3835 (!(~lma->lma_compat & compat) &&
3836 !(~lma->lma_incompat & incompat))))
3839 lma->lma_self_fid = *fid;
3840 lma->lma_compat |= compat;
3841 lma->lma_incompat |= incompat;
3842 if (rc == sizeof(*lma)) {
3843 lustre_lma_swab(lma);
3844 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3845 sizeof(*lma), XATTR_REPLACE);
3847 lustre_loa_swab(loa, false);
3848 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3849 sizeof(*loa), XATTR_REPLACE);
3857 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3858 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3859 * To have compatilibility with 1.8 ldiskfs driver we need to have
3860 * magic number at start of fid data.
3861 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3864 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3865 const struct lu_fid *fid)
3867 if (!fid_is_namespace_visible(fid) ||
3868 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3869 param->edp_magic = 0;
3873 param->edp_magic = LDISKFS_LUFID_MAGIC;
3874 param->edp_len = sizeof(struct lu_fid) + 1;
3875 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3879 * Try to read the fid from inode ea into dt_rec.
3881 * \param fid object fid.
3883 * \retval 0 on success
3885 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3886 __u32 ino, struct lu_fid *fid,
3887 struct osd_inode_id *id)
3889 struct osd_thread_info *info = osd_oti_get(env);
3890 struct inode *inode;
3894 osd_id_gen(id, ino, OSD_OII_NOGEN);
3895 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3897 RETURN(PTR_ERR(inode));
3903 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3905 struct inode *parent_dir,
3906 const struct lu_fid *dot_fid,
3907 const struct lu_fid *dot_dot_fid,
3908 struct osd_thandle *oth)
3910 struct ldiskfs_dentry_param *dot_ldp;
3911 struct ldiskfs_dentry_param *dot_dot_ldp;
3912 __u32 saved_nlink = dir->i_nlink;
3915 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_DOTDOT_ENOSPC))
3918 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3919 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3921 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3922 dot_ldp->edp_magic = 0;
3924 rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3925 dir, dot_ldp, dot_dot_ldp);
3927 * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3928 * the subseqent ref_add() will increase the dir->i_nlink
3929 * as 3. That is incorrect for new created directory.
3931 * It looks like hack, because we want to make the OSD API
3932 * to be order-independent for new created directory object
3933 * between dt_insert(..) and ref_add() operations.
3935 * Here, we only restore the in-RAM dir-inode's nlink attr,
3936 * becuase if the nlink attr is not 2, then there will be
3937 * ref_add() called following the dt_insert(..), such call
3938 * will make both the in-RAM and on-disk dir-inode's nlink
3939 * attr to be set as 2. LU-7447
3941 set_nlink(dir, saved_nlink);
3946 * Create an local agent inode for remote entry
3948 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3949 struct osd_device *osd,
3950 struct osd_object *pobj,
3951 const struct lu_fid *fid,
3955 struct osd_thread_info *info = osd_oti_get(env);
3956 struct inode *local;
3957 struct osd_thandle *oh;
3958 struct iattr iattr = {
3959 .ia_valid = ATTR_UID | ATTR_GID |
3960 ATTR_CTIME | ATTR_MTIME | ATTR_ATIME,
3961 .ia_ctime.tv_nsec = UTIME_OMIT,
3962 .ia_mtime.tv_nsec = UTIME_OMIT,
3963 .ia_atime.tv_nsec = UTIME_OMIT,
3964 .ia_uid = GLOBAL_ROOT_UID,
3965 .ia_gid = GLOBAL_ROOT_GID,
3972 oh = container_of(th, struct osd_thandle, ot_super);
3973 LASSERT(oh->ot_handle->h_transaction != NULL);
3975 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode,
3977 if (IS_ERR(local)) {
3978 CERROR("%s: create local error %d\n", osd_name(osd),
3979 (int)PTR_ERR(local));
3984 * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3985 * correct gid on remote file, not agent here
3987 local->i_gid = current_fsgid();
3988 ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3990 /* e2fsck doesn't like empty symlinks. Store remote FID as symlink.
3991 * That gives e2fsck something to look at and be happy, and allows
3992 * debugging if we need to determine where this symlink came from.
3994 if (S_ISLNK(type)) {
3995 BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1);
3996 rc = scnprintf((char *)LDISKFS_I(local)->i_data,
3997 LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
3999 i_size_write(local, rc);
4000 LDISKFS_I(local)->i_disksize = rc;
4002 unlock_new_inode(local);
4004 /* Agent inode should not have project ID */
4005 #ifdef HAVE_PROJECT_QUOTA
4006 if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
4007 i_projid_read(pobj->oo_inode) != 0) {
4008 rc = osd_transfer_project(local, 0, th);
4010 CERROR("%s: quota transfer failed:. Is project quota enforcement enabled on the ldiskfs filesystem? rc = %d\n",
4011 osd_ino2name(local), rc);
4012 RETURN(ERR_PTR(rc));
4016 /* Set special LMA flag for local agent inode */
4017 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
4019 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
4020 osd_name(osd), PFID(fid), rc);
4021 RETURN(ERR_PTR(rc));
4027 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
4028 lu_object_fid(&pobj->oo_dt.do_lu),
4031 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
4032 osd_name(osd), PFID(fid), rc);
4033 RETURN(ERR_PTR(rc));
4040 * when direntry is deleted, we have to take care of possible agent inode
4041 * referenced by that. unfortunately we can't do this at that point:
4042 * iget() within a running transaction leads to deadlock and we better do
4043 * not call that every delete declaration to save performance. so we put
4044 * a potention agent inode on a list and process that once the transaction
4045 * is over. Notice it's not any worse in terms of real orphans as regular
4046 * object destroy doesn't put inodes on the on-disk orphan list. this should
4047 * be addressed separately
4049 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
4050 struct osd_thandle *oh,
4053 struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
4054 struct osd_obj_orphan *oor;
4061 oor->oor_env = (struct lu_env *)env;
4062 spin_lock(&osd->od_osfs_lock);
4063 list_add(&oor->oor_list, &osd->od_orphan_list);
4064 spin_unlock(&osd->od_osfs_lock);
4066 oh->ot_remove_agents = 1;
4072 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
4073 struct osd_device *osd)
4075 struct osd_thread_info *info = osd_oti_get(env);
4076 struct osd_obj_orphan *oor, *tmp;
4077 struct osd_inode_id id;
4079 struct inode *inode;
4084 spin_lock(&osd->od_osfs_lock);
4085 list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
4086 if (oor->oor_env == env)
4087 list_move(&oor->oor_list, &list);
4089 spin_unlock(&osd->od_osfs_lock);
4091 list_for_each_entry_safe(oor, tmp, &list, oor_list) {
4095 list_del(&oor->oor_list);
4098 osd_id_gen(&id, ino, OSD_OII_NOGEN);
4099 inode = osd_iget_fid(info, osd, &id, &fid);
4103 if (!osd_remote_fid(env, osd, &fid)) {
4108 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
4110 mark_inode_dirty(inode);
4111 ldiskfs_journal_stop(jh);
4119 * OSD layer object create function for OST objects (b=11826).
4121 * The FID is inserted into inode xattr here.
4123 * \retval 0, on success
4124 * \retval -ve, on error
4126 static int osd_create(const struct lu_env *env, struct dt_object *dt,
4127 struct lu_attr *attr, struct dt_allocation_hint *hint,
4128 struct dt_object_format *dof, struct thandle *th)
4130 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4131 struct osd_object *obj = osd_dt_obj(dt);
4132 struct osd_thread_info *info = osd_oti_get(env);
4133 int result, on_ost = 0;
4137 if (dt_object_exists(dt))
4140 LINVRNT(osd_invariant(obj));
4141 LASSERT(!dt_object_remote(dt));
4142 LASSERT(osd_is_write_locked(env, obj));
4143 LASSERT(th != NULL);
4145 if (unlikely(fid_is_acct(fid)))
4147 * Quota files can't be created from the kernel any more,
4148 * 'tune2fs -O quota' will take care of creating them
4152 result = __osd_create(info, obj, attr, hint, dof, th);
4154 if (fid_is_idif(fid) &&
4155 !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
4156 struct lu_fid *tfid = &info->oti_fid;
4157 struct ost_id *oi = &info->oti_ostid;
4159 fid_to_ostid(fid, oi);
4160 ostid_to_fid(tfid, oi, 0);
4162 result = osd_ea_fid_set(info, obj->oo_inode, tfid,
4163 LMAC_FID_ON_OST, 0);
4165 on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
4167 result = osd_ea_fid_set(info, obj->oo_inode, fid,
4168 on_ost ? LMAC_FID_ON_OST : 0,
4171 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
4172 obj->oo_dt.do_body_ops = &osd_body_ops;
4175 if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY)) {
4176 struct inode *inode = obj->oo_inode;
4178 result = __osd_oi_insert(env, obj, fid, th);
4179 if (result && inode) {
4180 spin_lock(&obj->oo_guard);
4182 spin_unlock(&obj->oo_guard);
4183 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4184 ldiskfs_set_inode_state(inode,
4185 LDISKFS_STATE_LUSTRE_DESTROY);
4187 obj->oo_inode = NULL;
4192 * a small optimization - dt_insert() isn't usually applied
4193 * to OST objects, so we don't need to cache OI mapping for
4196 if (result == 0 && on_ost == 0) {
4197 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4199 result = osd_idc_find_and_init(env, osd, obj);
4200 LASSERT(result == 0);
4203 LASSERT(ergo(result == 0,
4204 dt_object_exists(dt) && !dt_object_remote(dt)));
4205 LINVRNT(osd_invariant(obj));
4209 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4210 struct thandle *handle)
4212 struct osd_thandle *oh;
4215 /* it's possible that object doesn't exist yet */
4216 LASSERT(handle != NULL);
4218 oh = container_of(handle, struct osd_thandle, ot_super);
4219 LASSERT(oh->ot_handle == NULL);
4221 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4222 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4224 rc = osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev),
4231 * Concurrency: @dt is write locked.
4233 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4236 struct osd_object *obj = osd_dt_obj(dt);
4237 struct inode *inode = obj->oo_inode;
4238 struct osd_thandle *oh;
4241 if (!dt_object_exists(dt) || obj->oo_destroyed)
4244 LINVRNT(osd_invariant(obj));
4245 LASSERT(!dt_object_remote(dt));
4246 LASSERT(osd_is_write_locked(env, obj));
4247 LASSERT(th != NULL);
4249 oh = container_of(th, struct osd_thandle, ot_super);
4250 LASSERT(oh->ot_handle != NULL);
4252 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4254 CDEBUG(D_INODE, DFID" increase nlink %d\n",
4255 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4257 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4258 * (65000) subdirectories by storing "1" in i_nlink if the link count
4259 * would otherwise overflow. Directory tranversal tools understand
4260 * that (st_nlink == 1) indicates that the filesystem dose not track
4261 * hard links count on the directory, and will not abort subdirectory
4262 * scanning early once (st_nlink - 2) subdirs have been found.
4264 * This also has to properly handle the case of inodes with nlink == 0
4265 * in case they are being linked into the PENDING directory
4267 spin_lock(&obj->oo_guard);
4268 if (unlikely(inode->i_nlink == 0))
4269 /* inc_nlink from 0 may cause WARN_ON */
4270 set_nlink(inode, 1);
4272 osd_ldiskfs_inc_count(oh->ot_handle, inode);
4273 if (!S_ISDIR(inode->i_mode))
4274 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4276 spin_unlock(&obj->oo_guard);
4278 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4279 LINVRNT(osd_invariant(obj));
4281 osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4286 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4287 struct thandle *handle)
4289 struct osd_thandle *oh;
4291 if (!dt_object_exists(dt))
4294 LASSERT(!dt_object_remote(dt));
4295 LASSERT(handle != NULL);
4297 oh = container_of(handle, struct osd_thandle, ot_super);
4298 LASSERT(oh->ot_handle == NULL);
4300 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4301 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4307 * Concurrency: @dt is write locked.
4309 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4312 struct osd_object *obj = osd_dt_obj(dt);
4313 struct inode *inode = obj->oo_inode;
4314 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4315 struct osd_thandle *oh;
4317 if (!dt_object_exists(dt))
4320 LINVRNT(osd_invariant(obj));
4321 LASSERT(!dt_object_remote(dt));
4322 LASSERT(osd_is_write_locked(env, obj));
4323 LASSERT(th != NULL);
4325 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_REF_DEL))
4328 oh = container_of(th, struct osd_thandle, ot_super);
4329 LASSERT(oh->ot_handle != NULL);
4331 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4333 spin_lock(&obj->oo_guard);
4335 * That can be result of upgrade from old Lustre version and
4336 * applied only to local files. Just skip this ref_del call.
4337 * ext4_unlink() only treats this as a warning, don't LASSERT here.
4339 if (inode->i_nlink == 0) {
4340 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4341 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4342 ", maybe an upgraded file? (LU-3915)\n",
4343 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4344 spin_unlock(&obj->oo_guard);
4348 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4349 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4351 osd_ldiskfs_dec_count(oh->ot_handle, inode);
4352 spin_unlock(&obj->oo_guard);
4354 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4355 LINVRNT(osd_invariant(obj));
4357 osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4363 * Concurrency: @dt is read locked.
4365 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4366 struct lu_buf *buf, const char *name)
4368 struct osd_object *obj = osd_dt_obj(dt);
4369 struct inode *inode = obj->oo_inode;
4370 struct osd_thread_info *info = osd_oti_get(env);
4371 struct dentry *dentry = &info->oti_obj_dentry;
4372 bool cache_xattr = false;
4377 /* version get is not real XATTR but uses xattr API */
4378 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4379 dt_obj_version_t *ver = buf->lb_buf;
4382 * for version we are just using xattr API but change inode
4385 if (buf->lb_len == 0)
4386 return sizeof(dt_obj_version_t);
4388 if (buf->lb_len < sizeof(dt_obj_version_t))
4391 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4392 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4394 *ver = LDISKFS_I(inode)->i_fs_version;
4396 return sizeof(dt_obj_version_t);
4399 if (!dt_object_exists(dt))
4402 LASSERT(!dt_object_remote(dt));
4403 LASSERT(inode->i_op != NULL);
4404 #ifdef HAVE_IOP_XATTR
4405 LASSERT(inode->i_op->getxattr != NULL);
4408 if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4409 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4413 rc = osd_oxc_get(obj, name, buf);
4418 if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4419 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4420 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4421 struct filter_fid *ff;
4422 struct ost_layout *ol;
4424 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4428 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4431 if (buf->lb_len == 0 || !buf->lb_buf)
4434 if (buf->lb_len < rc)
4438 ol = &ff->ff_layout;
4439 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4440 PFID_STRIPE_IDX_BITS);
4441 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4442 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4443 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4444 if (lma->lma_compat & LMAC_COMP_INFO) {
4445 ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4446 ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4447 ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4449 ol->ol_comp_start = 0;
4450 ol->ol_comp_end = 0;
4454 /* Get enc context xattr directly from ldiskfs instead of going
4455 * through the VFS, as there is no xattr handler for
4458 if (strcmp(name, LL_XATTR_NAME_ENCRYPTION_CONTEXT) == 0)
4459 rc = ldiskfs_xattr_get(inode,
4460 LDISKFS_XATTR_INDEX_ENCRYPTION,
4461 LDISKFS_XATTR_NAME_ENCRYPTION_CONTEXT,
4462 buf->lb_buf, buf->lb_len);
4464 rc = __osd_xattr_get(inode, dentry, name,
4465 buf->lb_buf, buf->lb_len);
4469 if (rc == -ENOENT || rc == -ENODATA)
4470 osd_oxc_add(obj, name, NULL, 0);
4471 else if (rc > 0 && buf->lb_buf != NULL)
4472 osd_oxc_add(obj, name, buf->lb_buf, rc);
4478 static int osd_declare_xattr_set(const struct lu_env *env,
4479 struct dt_object *dt,
4480 const struct lu_buf *buf, const char *name,
4481 int fl, struct thandle *handle)
4483 struct osd_thandle *oh;
4485 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4487 LASSERT(handle != NULL);
4489 oh = container_of(handle, struct osd_thandle, ot_super);
4490 LASSERT(oh->ot_handle == NULL);
4492 if (strcmp(name, XATTR_NAME_LMA) == 0) {
4494 * For non-upgrading case, the LMA is set first and
4495 * usually fit inode. But for upgrade case, the LMA
4496 * may be in another separated EA block.
4498 if (dt_object_exists(dt)) {
4499 if (fl == LU_XATTR_REPLACE)
4504 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4506 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4507 /* We may need to delete the old PFID EA. */
4508 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4509 if (fl == LU_XATTR_REPLACE)
4515 * If some name entry resides on remote MDT, then will create
4516 * agent entry under remote parent. On the other hand, if the
4517 * remote entry will be removed, then related agent entry may
4518 * need to be removed from the remote parent. So there may be
4519 * kinds of cases, let's declare enough credits. The credits
4520 * for create agent entry is enough for remove case.
4522 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4523 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4524 if (dt_object_exists(dt))
4525 credits += 1; /* For updating LMA */
4529 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4534 if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4536 * learn xattr size from osd_xattr_get if
4537 * attribute has not been read yet
4539 buflen = __osd_xattr_get(
4540 osd_dt_obj(dt)->oo_inode,
4541 &osd_oti_get(env)->oti_obj_dentry,
4546 buflen = buf->lb_len;
4549 if (buflen > sb->s_blocksize) {
4550 credits += osd_calc_bkmap_credits(
4552 (buflen + sb->s_blocksize - 1) >>
4553 sb->s_blocksize_bits);
4557 * xattr set may involve inode quota change, reserve credits for
4558 * dquot_initialize()
4560 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4563 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4568 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4569 const struct lu_buf *buf, int fl,
4570 struct thandle *handle)
4572 struct osd_thread_info *info = osd_oti_get(env);
4573 struct dentry *dentry = &info->oti_obj_dentry;
4574 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4575 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4576 struct inode *inode = obj->oo_inode;
4577 struct filter_fid *ff = buf->lb_buf;
4578 struct ost_layout *ol = &ff->ff_layout;
4579 int flags = XATTR_REPLACE;
4584 if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4587 rc = osd_get_lma(info, inode, dentry, loa);
4588 if (rc == -ENODATA) {
4589 /* Usually for upgarding from old device */
4590 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4591 LMAC_FID_ON_OST, 0);
4592 flags = XATTR_CREATE;
4597 if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4598 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4601 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4602 /* Separate PFID EA from LMA */
4603 lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4604 lustre_lma_swab(lma);
4605 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4606 sizeof(*lma), XATTR_REPLACE);
4608 obj->oo_pfid_in_lma = 0;
4609 rc = LU_XATTR_CREATE;
4615 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4619 * Old client does not send stripe information,
4620 * then store the PFID EA on disk separatedly.
4622 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4623 ol->ol_stripe_size == 0))
4626 /* Remove old PFID EA entry firstly. */
4627 dquot_initialize(inode);
4628 rc = ll_vfs_removexattr(dentry, inode, XATTR_NAME_FID);
4629 if (rc == -ENODATA) {
4630 /* XATTR_NAME_FID is already absent */
4637 fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4638 if (likely(ol->ol_stripe_size != 0)) {
4639 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4640 PFID_STRIPE_IDX_BITS;
4641 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4642 lma->lma_compat |= LMAC_STRIPE_INFO;
4643 if (ol->ol_comp_id != 0) {
4644 loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4645 loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4646 loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4647 lma->lma_compat |= LMAC_COMP_INFO;
4651 lustre_loa_swab(loa, false);
4653 /* Store the PFID EA inside LMA. */
4654 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4657 obj->oo_pfid_in_lma = 1;
4663 * In DNE environment, the object (in spite of regular file or directory)
4664 * and its name entry may reside on different MDTs. Under such case, we will
4665 * create an agent entry on the MDT where the object resides. The agent entry
4666 * references the object locally, that makes the object to be visible to the
4667 * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4668 * such as 'tar' can handle the object properly.
4670 * We handle the agent entry during set linkEA that is the common interface
4671 * for both regular file and directroy, can handle kinds of cases, such as
4672 * create/link/unlink/rename, and so on.
4674 * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4676 * XXX: There are two known issues:
4677 * 1. For one object, we will create at most one agent entry even if there
4678 * may be more than one cross-MDTs hard links on the object. So the local
4679 * e2fsck may claim that the object's nlink is larger than the name entries
4680 * that reference such inode. And in further, the e2fsck will fix the nlink
4681 * attribute to match the local references. Then it will cause the object's
4682 * nlink attribute to be inconsistent with the global references. it is bad
4683 * but not fatal. The ref_del() can handle the zero-referenced case. On the
4684 * other hand, the global namespace LFSCK can repair the object's attribute
4685 * according to the linkEA.
4686 * 2. There may be too many hard links on the object as to its linkEA overflow,
4687 * then the linkEA entry for cross-MDTs reference may be discarded. If such
4688 * case happened, then at this point, we do not know whether there are some
4689 * cross-MDTs reference. But there are local references, it guarantees that
4690 * object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4692 static int osd_xattr_handle_linkea(const struct lu_env *env,
4693 struct osd_device *osd,
4694 struct osd_object *obj,
4695 const struct lu_buf *buf,
4696 struct thandle *handle)
4698 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4699 struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4700 struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4701 struct lu_name tmpname;
4702 struct osd_thandle *oh;
4704 bool remote = false;
4708 oh = container_of(handle, struct osd_thandle, ot_super);
4709 LASSERT(oh->ot_handle != NULL);
4711 rc = linkea_init_with_rec(&ldata);
4713 linkea_first_entry(&ldata);
4714 while (ldata.ld_lee != NULL && !remote) {
4715 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4717 if (osd_remote_fid(env, osd, tfid) > 0)
4720 linkea_next_entry(&ldata);
4722 } else if (rc == -ENODATA) {
4728 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4729 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4731 CERROR("%s: failed to remove agent entry for "DFID
4732 ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4733 } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4734 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4736 CERROR("%s: failed to create agent entry for "DFID
4737 ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4744 * Concurrency: @dt is write locked.
4746 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4747 const struct lu_buf *buf, const char *name, int fl,
4748 struct thandle *handle)
4750 struct osd_object *obj = osd_dt_obj(dt);
4751 struct osd_device *osd = osd_obj2dev(obj);
4752 struct inode *inode = obj->oo_inode;
4753 struct osd_thread_info *info = osd_oti_get(env);
4763 /* version set is not real XATTR */
4764 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4765 dt_obj_version_t *version = buf->lb_buf;
4768 * for version we are just using xattr API but change inode
4771 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4773 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4774 *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4776 LDISKFS_I(inode)->i_fs_version = *version;
4778 * Version is set after all inode operations are finished,
4779 * so we should mark it dirty here
4781 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4786 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4787 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4790 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4793 * For the OST device with 256 bytes inode size by default,
4794 * the PFID EA will be stored together with LMA EA to avoid
4795 * performance trouble. Otherwise the PFID EA can be stored
4796 * independently. LU-8998
4798 if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4799 (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4800 LASSERT(buf->lb_buf);
4802 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4805 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4806 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4807 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4809 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4813 lma->lma_incompat |= LMAI_STRIPED;
4814 lustre_lma_swab(lma);
4815 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4816 sizeof(*lma), XATTR_REPLACE);
4819 } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4820 LASSERT(!osd->od_is_ost);
4822 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4827 if (fl & LU_XATTR_REPLACE)
4828 fs_flags |= XATTR_REPLACE;
4830 if (fl & LU_XATTR_CREATE)
4831 fs_flags |= XATTR_CREATE;
4833 if (strcmp(name, LL_XATTR_NAME_ENCRYPTION_CONTEXT) == 0) {
4834 /* Set enc context xattr directly in ldiskfs instead of going
4835 * through the VFS, as there is no xattr handler for
4838 struct osd_thandle *oth = container_of(handle,
4842 if (!oth->ot_handle)
4843 /* this should be already part of a transaction */
4846 rc = ldiskfs_xattr_set_handle(oth->ot_handle, inode,
4847 LDISKFS_XATTR_INDEX_ENCRYPTION,
4848 LDISKFS_XATTR_NAME_ENCRYPTION_CONTEXT,
4849 buf->lb_buf, len, fs_flags);
4851 rc = __osd_xattr_set(info, inode, name,
4852 buf->lb_buf, len, fs_flags);
4854 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4857 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4858 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4859 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4865 * Concurrency: @dt is read locked.
4867 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4868 const struct lu_buf *buf)
4870 struct osd_object *obj = osd_dt_obj(dt);
4871 struct osd_device *dev = osd_obj2dev(obj);
4872 struct inode *inode = obj->oo_inode;
4873 struct osd_thread_info *info = osd_oti_get(env);
4874 struct dentry *dentry = &info->oti_obj_dentry;
4877 if (!dt_object_exists(dt))
4880 LASSERT(!dt_object_remote(dt));
4881 LASSERT(inode->i_op != NULL);
4882 LASSERT(inode->i_op->listxattr != NULL);
4884 dentry->d_inode = inode;
4885 dentry->d_sb = inode->i_sb;
4886 rc = inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4888 if (rc < 0 || buf->lb_buf == NULL)
4891 /* Hide virtual project ID xattr from list if disabled */
4892 if (!dev->od_enable_projid_xattr) {
4893 char *end = (char *)buf->lb_buf + rc;
4894 char *p = buf->lb_buf;
4897 char *next = p + strlen(p) + 1;
4899 if (strcmp(p, XATTR_NAME_PROJID) == 0) {
4901 memmove(p, next, end - next);
4913 static int osd_declare_xattr_del(const struct lu_env *env,
4914 struct dt_object *dt, const char *name,
4915 struct thandle *handle)
4917 struct osd_thandle *oh;
4918 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4920 LASSERT(!dt_object_remote(dt));
4921 LASSERT(handle != NULL);
4923 oh = container_of(handle, struct osd_thandle, ot_super);
4924 LASSERT(oh->ot_handle == NULL);
4926 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4927 osd_dto_credits_noquota[DTO_XATTR_SET]);
4929 * xattr del may involve inode quota change, reserve credits for
4930 * dquot_initialize()
4932 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4938 * Concurrency: @dt is write locked.
4940 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4941 const char *name, struct thandle *handle)
4943 struct osd_object *obj = osd_dt_obj(dt);
4944 struct inode *inode = obj->oo_inode;
4945 struct osd_thread_info *info = osd_oti_get(env);
4946 struct dentry *dentry = &info->oti_obj_dentry;
4949 if (!dt_object_exists(dt))
4952 LASSERT(!dt_object_remote(dt));
4953 LASSERT(inode->i_op != NULL);
4954 LASSERT(handle != NULL);
4955 #ifdef HAVE_IOP_XATTR
4956 LASSERT(inode->i_op->removexattr != NULL);
4959 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4961 if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4962 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4964 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4965 &info->oti_ost_attrs);
4967 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4969 lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4970 lustre_lma_swab(lma);
4971 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4972 sizeof(*lma), XATTR_REPLACE);
4974 obj->oo_pfid_in_lma = 0;
4977 dquot_initialize(inode);
4978 dentry->d_inode = inode;
4979 dentry->d_sb = inode->i_sb;
4980 rc = ll_vfs_removexattr(dentry, inode, name);
4983 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4986 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4987 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4988 osd_oxc_del(obj, name);
4993 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4994 __u64 start, __u64 end)
4996 struct osd_object *obj = osd_dt_obj(dt);
4997 struct osd_device *dev = osd_obj2dev(obj);
4998 struct inode *inode = obj->oo_inode;
5003 file = alloc_file_pseudo(inode, dev->od_mnt, "/", O_NOATIME,
5006 RETURN(PTR_ERR(file));
5008 file->f_mode |= FMODE_64BITHASH;
5009 rc = vfs_fsync_range(file, start, end, 0);
5016 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
5021 static bool osd_check_stale(struct dt_object *dt)
5029 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
5030 const struct dt_index_features *feat)
5032 struct iam_descr *descr;
5034 if (osd_object_is_root(o))
5035 return feat == &dt_directory_features;
5037 LASSERT(o->oo_dir != NULL);
5039 descr = o->oo_dir->od_container.ic_descr;
5040 if (feat == &dt_directory_features) {
5041 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
5046 return feat->dif_keysize_min <= descr->id_key_size &&
5047 descr->id_key_size <= feat->dif_keysize_max &&
5048 feat->dif_recsize_min <= descr->id_rec_size &&
5049 descr->id_rec_size <= feat->dif_recsize_max &&
5050 !(feat->dif_flags & (DT_IND_VARKEY |
5051 DT_IND_VARREC | DT_IND_NONUNQ)) &&
5052 ergo(feat->dif_flags & DT_IND_UPDATE,
5053 1 /* XXX check that object (and fs) is writable */);
5057 static int osd_iam_container_init(const struct lu_env *env,
5058 struct osd_object *obj,
5059 struct osd_directory *dir)
5061 struct iam_container *bag = &dir->od_container;
5064 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
5068 result = iam_container_setup(bag);
5070 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
5072 iam_container_fini(bag);
5079 * Concurrency: no external locking is necessary.
5081 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
5082 const struct dt_index_features *feat)
5086 struct osd_object *obj = osd_dt_obj(dt);
5088 LINVRNT(osd_invariant(obj));
5090 if (osd_object_is_root(obj)) {
5091 dt->do_index_ops = &osd_index_ea_ops;
5093 } else if (feat == &dt_directory_features) {
5094 dt->do_index_ops = &osd_index_ea_ops;
5095 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
5100 } else if (unlikely(feat == &dt_otable_features)) {
5101 dt->do_index_ops = &osd_otable_ops;
5103 } else if (unlikely(feat == &dt_acct_features)) {
5104 dt->do_index_ops = &osd_acct_index_ops;
5107 } else if (!osd_has_index(obj)) {
5108 struct osd_directory *dir;
5109 struct osd_device *osd = osd_obj2dev(obj);
5110 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
5115 spin_lock(&obj->oo_guard);
5116 if (obj->oo_dir == NULL)
5120 * Concurrent thread allocated container data.
5123 spin_unlock(&obj->oo_guard);
5125 * Now, that we have container data, serialize its
5128 down_write(&obj->oo_ext_idx_sem);
5130 * recheck under lock.
5133 if (osd_has_index(obj)) {
5138 result = osd_iam_container_init(env, obj, obj->oo_dir);
5139 if (result || feat == &dt_lfsck_namespace_features ||
5140 feat == &dt_lfsck_layout_orphan_features ||
5141 feat == &dt_lfsck_layout_dangling_features)
5144 result = osd_index_register(osd, fid,
5145 feat->dif_keysize_max,
5146 feat->dif_recsize_max);
5148 CWARN("%s: failed to register index "
5150 osd_name(osd), PFID(fid), result);
5151 else if (result > 0)
5154 CDEBUG(D_LFSCK, "%s: index object "DFID
5155 " (%d/%d) registered\n",
5156 osd_name(osd), PFID(fid),
5157 (int)feat->dif_keysize_max,
5158 (int)feat->dif_recsize_max);
5161 up_write(&obj->oo_ext_idx_sem);
5169 if (result == 0 && skip_iam == 0) {
5170 if (!osd_iam_index_probe(env, obj, feat))
5173 LINVRNT(osd_invariant(obj));
5178 static int osd_otable_it_attr_get(const struct lu_env *env,
5179 struct dt_object *dt,
5180 struct lu_attr *attr)
5186 static const struct dt_object_operations osd_obj_ops = {
5187 .do_read_lock = osd_read_lock,
5188 .do_write_lock = osd_write_lock,
5189 .do_read_unlock = osd_read_unlock,
5190 .do_write_unlock = osd_write_unlock,
5191 .do_write_locked = osd_write_locked,
5192 .do_attr_get = osd_attr_get,
5193 .do_declare_attr_set = osd_declare_attr_set,
5194 .do_attr_set = osd_attr_set,
5195 .do_ah_init = osd_ah_init,
5196 .do_declare_create = osd_declare_create,
5197 .do_create = osd_create,
5198 .do_declare_destroy = osd_declare_destroy,
5199 .do_destroy = osd_destroy,
5200 .do_index_try = osd_index_try,
5201 .do_declare_ref_add = osd_declare_ref_add,
5202 .do_ref_add = osd_ref_add,
5203 .do_declare_ref_del = osd_declare_ref_del,
5204 .do_ref_del = osd_ref_del,
5205 .do_xattr_get = osd_xattr_get,
5206 .do_declare_xattr_set = osd_declare_xattr_set,
5207 .do_xattr_set = osd_xattr_set,
5208 .do_declare_xattr_del = osd_declare_xattr_del,
5209 .do_xattr_del = osd_xattr_del,
5210 .do_xattr_list = osd_xattr_list,
5211 .do_object_sync = osd_object_sync,
5212 .do_invalidate = osd_invalidate,
5213 .do_check_stale = osd_check_stale,
5216 static const struct dt_object_operations osd_obj_otable_it_ops = {
5217 .do_attr_get = osd_otable_it_attr_get,
5218 .do_index_try = osd_index_try,
5221 static int osd_index_declare_iam_delete(const struct lu_env *env,
5222 struct dt_object *dt,
5223 const struct dt_key *key,
5224 struct thandle *handle)
5226 struct osd_thandle *oh;
5228 oh = container_of(handle, struct osd_thandle, ot_super);
5229 LASSERT(oh->ot_handle == NULL);
5231 /* Recycle may cause additional three blocks to be changed. */
5232 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
5233 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
5239 * delete a (key, value) pair from index \a dt specified by \a key
5241 * \param dt osd index object
5242 * \param key key for index
5243 * \param rec record reference
5244 * \param handle transaction handler
5247 * \retval -ve failure
5249 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
5250 const struct dt_key *key,
5251 struct thandle *handle)
5253 struct osd_thread_info *oti = osd_oti_get(env);
5254 struct osd_object *obj = osd_dt_obj(dt);
5255 struct osd_thandle *oh;
5256 struct iam_path_descr *ipd;
5257 struct iam_container *bag = &obj->oo_dir->od_container;
5262 if (!dt_object_exists(dt))
5265 LINVRNT(osd_invariant(obj));
5266 LASSERT(!dt_object_remote(dt));
5267 LASSERT(bag->ic_object == obj->oo_inode);
5268 LASSERT(handle != NULL);
5270 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5272 ipd = osd_idx_ipd_get(env, bag);
5273 if (unlikely(ipd == NULL))
5276 oh = container_of(handle, struct osd_thandle, ot_super);
5277 LASSERT(oh->ot_handle != NULL);
5278 LASSERT(oh->ot_handle->h_transaction != NULL);
5280 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5281 /* swab quota uid/gid provided by caller */
5282 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5283 key = (const struct dt_key *)&oti->oti_quota_id;
5286 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5287 osd_ipd_put(env, bag, ipd);
5288 LINVRNT(osd_invariant(obj));
5289 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5293 static int osd_index_declare_ea_delete(const struct lu_env *env,
5294 struct dt_object *dt,
5295 const struct dt_key *key,
5296 struct thandle *handle)
5298 struct osd_thandle *oh;
5299 struct inode *inode;
5304 LASSERT(!dt_object_remote(dt));
5305 LASSERT(handle != NULL);
5307 oh = container_of(handle, struct osd_thandle, ot_super);
5308 LASSERT(oh->ot_handle == NULL);
5310 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5311 osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5313 inode = osd_dt_obj(dt)->oo_inode;
5317 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5318 i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5323 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5326 struct osd_fid_pack *rec;
5329 if (de->file_type & LDISKFS_DIRENT_LUFID) {
5330 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5331 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5332 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5338 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5339 const struct lu_fid *fid)
5341 struct seq_server_site *ss = osd_seq_site(osd);
5345 /* FID seqs not in FLDB, must be local seq */
5346 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5350 * If FLD is not being initialized yet, it only happens during the
5351 * initialization, likely during mgs initialization, and we assume
5352 * this is local FID.
5354 if (ss == NULL || ss->ss_server_fld == NULL)
5357 /* Only check the local FLDB here */
5358 if (osd_seq_exists(env, osd, fid_seq(fid)))
5364 static void osd_take_care_of_agent(const struct lu_env *env,
5365 struct osd_device *osd,
5366 struct osd_thandle *oh,
5367 struct ldiskfs_dir_entry_2 *de)
5369 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5370 struct osd_idmap_cache *idc;
5371 int rc, schedule = 0;
5373 LASSERT(de != NULL);
5375 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5376 if (likely(rc == 0)) {
5377 idc = osd_idc_find_or_init(env, osd, fid);
5378 if (IS_ERR(idc) || idc->oic_remote)
5380 } else if (rc == -ENODATA) {
5382 * can't get FID, postpone to the end of the
5383 * transaction when iget() is safe
5387 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5390 osd_schedule_agent_inode_removal(env, oh,
5391 le32_to_cpu(de->inode));
5395 * Utility function to get real name from object name
5397 * \param[in] obj pointer to the object to be handled
5398 * \param[in] name object name
5399 * \param[in] len object name len
5400 * \param[out]ln pointer to the struct lu_name to hold the real name
5402 * If file is not encrypted, real name is just the object name.
5403 * If file is encrypted, object name needs to be decoded. In
5404 * this case a new buffer is allocated, and ln->ln_name needs to be freed by
5407 * \retval 0, on success
5408 * \retval -ve, on error
5410 static int obj_name2lu_name(struct osd_object *obj, const char *name,
5411 int len, struct lu_name *ln)
5413 if (!(obj->oo_lma_flags & LUSTRE_ENCRYPT_FL)) {
5415 ln->ln_namelen = len;
5417 char *buf = kmalloc(len, GFP_NOFS);
5422 len = critical_decode(name, len, buf);
5424 ln->ln_namelen = len;
5431 * Index delete function for interoperability mode (b11826).
5432 * It will remove the directory entry added by osd_index_ea_insert().
5433 * This entry is needed to maintain name->fid mapping.
5435 * \param key, key i.e. file entry to be deleted
5437 * \retval 0, on success
5438 * \retval -ve, on error
5440 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5441 const struct dt_key *key, struct thandle *handle)
5443 struct osd_object *obj = osd_dt_obj(dt);
5444 struct inode *dir = obj->oo_inode;
5445 struct dentry *dentry;
5446 struct osd_thandle *oh;
5447 struct ldiskfs_dir_entry_2 *de = NULL;
5448 struct buffer_head *bh;
5449 struct htree_lock *hlock = NULL;
5450 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5456 if (!dt_object_exists(dt))
5459 LINVRNT(osd_invariant(obj));
5460 LASSERT(!dt_object_remote(dt));
5461 LASSERT(handle != NULL);
5463 rc = obj_name2lu_name(obj, (char *)key, strlen((char *)key), &ln);
5467 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5469 oh = container_of(handle, struct osd_thandle, ot_super);
5470 LASSERT(oh->ot_handle != NULL);
5471 LASSERT(oh->ot_handle->h_transaction != NULL);
5473 dquot_initialize(dir);
5474 dentry = osd_child_dentry_get(env, obj, ln.ln_name, ln.ln_namelen);
5476 if (obj->oo_hl_head != NULL) {
5477 hlock = osd_oti_get(env)->oti_hlock;
5478 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5479 dir, LDISKFS_HLOCK_DEL);
5481 down_write(&obj->oo_ext_idx_sem);
5484 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5487 * If this is not the ".." entry, it might be a remote DNE
5488 * entry and we need to check if the FID is for a remote
5489 * MDT. If the FID is not in the directory entry (e.g.
5490 * upgraded 1.8 filesystem without dirdata enabled) then
5491 * we need to get the FID from the LMA. For a remote directory
5492 * there HAS to be an LMA, it cannot be an IGIF inode in this
5495 * Delete the entry before the agent inode in order to
5496 * simplify error handling. At worst an error after deleting
5497 * the entry first might leak the agent inode afterward. The
5498 * reverse would need filesystem abort in case of error deleting
5499 * the entry after the agent had been removed, or leave a
5500 * dangling entry pointing at a random inode.
5502 if (strcmp((char *)key, dotdot) != 0)
5503 osd_take_care_of_agent(env, osd, oh, de);
5504 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5510 if (!rc && fid_is_namespace_visible(lu_object_fid(&dt->do_lu)) &&
5511 obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
5512 /* NB, dirent count may not be accurate, because it's counted
5515 if (obj->oo_dirent_count)
5516 obj->oo_dirent_count--;
5518 obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
5521 ldiskfs_htree_unlock(hlock);
5523 up_write(&obj->oo_ext_idx_sem);
5526 LASSERT(osd_invariant(obj));
5527 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5528 if (ln.ln_name != (char *)key)
5534 * Lookup index for \a key and copy record to \a rec.
5536 * \param dt osd index object
5537 * \param key key for index
5538 * \param rec record reference
5540 * \retval +ve success : exact mach
5541 * \retval 0 return record with key not greater than \a key
5542 * \retval -ve failure
5544 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5545 struct dt_rec *rec, const struct dt_key *key)
5547 struct osd_object *obj = osd_dt_obj(dt);
5548 struct iam_path_descr *ipd;
5549 struct iam_container *bag = &obj->oo_dir->od_container;
5550 struct osd_thread_info *oti = osd_oti_get(env);
5551 struct iam_iterator *it = &oti->oti_idx_it;
5552 struct iam_rec *iam_rec;
5557 if (!dt_object_exists(dt))
5560 LASSERT(osd_invariant(obj));
5561 LASSERT(!dt_object_remote(dt));
5562 LASSERT(bag->ic_object == obj->oo_inode);
5564 ipd = osd_idx_ipd_get(env, bag);
5568 /* got ipd now we can start iterator. */
5569 iam_it_init(it, bag, 0, ipd);
5571 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5572 /* swab quota uid/gid provided by caller */
5573 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5574 key = (const struct dt_key *)&oti->oti_quota_id;
5577 rc = iam_it_get(it, (struct iam_key *)key);
5579 if (S_ISDIR(obj->oo_inode->i_mode))
5580 iam_rec = (struct iam_rec *)oti->oti_ldp;
5582 iam_rec = (struct iam_rec *)rec;
5584 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5586 if (S_ISDIR(obj->oo_inode->i_mode))
5587 osd_fid_unpack((struct lu_fid *)rec,
5588 (struct osd_fid_pack *)iam_rec);
5589 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5590 osd_quota_unpack(obj, rec);
5595 osd_ipd_put(env, bag, ipd);
5597 LINVRNT(osd_invariant(obj));
5602 static int osd_index_declare_iam_insert(const struct lu_env *env,
5603 struct dt_object *dt,
5604 const struct dt_rec *rec,
5605 const struct dt_key *key,
5606 struct thandle *handle)
5608 struct osd_thandle *oh;
5610 LASSERT(handle != NULL);
5612 oh = container_of(handle, struct osd_thandle, ot_super);
5613 LASSERT(oh->ot_handle == NULL);
5615 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5616 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5622 * Inserts (key, value) pair in \a dt index object.
5624 * \param dt osd index object
5625 * \param key key for index
5626 * \param rec record reference
5627 * \param th transaction handler
5630 * \retval -ve failure
5632 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5633 const struct dt_rec *rec,
5634 const struct dt_key *key, struct thandle *th)
5636 struct osd_object *obj = osd_dt_obj(dt);
5637 struct iam_path_descr *ipd;
5638 struct osd_thandle *oh;
5639 struct iam_container *bag;
5640 struct osd_thread_info *oti = osd_oti_get(env);
5641 struct iam_rec *iam_rec;
5646 if (!dt_object_exists(dt))
5649 LINVRNT(osd_invariant(obj));
5650 LASSERT(!dt_object_remote(dt));
5652 bag = &obj->oo_dir->od_container;
5653 LASSERT(bag->ic_object == obj->oo_inode);
5654 LASSERT(th != NULL);
5656 osd_trans_exec_op(env, th, OSD_OT_INSERT);
5658 ipd = osd_idx_ipd_get(env, bag);
5659 if (unlikely(ipd == NULL))
5662 oh = container_of(th, struct osd_thandle, ot_super);
5663 LASSERT(oh->ot_handle != NULL);
5664 LASSERT(oh->ot_handle->h_transaction != NULL);
5665 if (S_ISDIR(obj->oo_inode->i_mode)) {
5666 iam_rec = (struct iam_rec *)oti->oti_ldp;
5667 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5669 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5670 /* pack quota uid/gid */
5671 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5672 key = (const struct dt_key *)&oti->oti_quota_id;
5673 /* pack quota record */
5674 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5675 iam_rec = (struct iam_rec *)rec;
5677 iam_rec = (struct iam_rec *)rec;
5680 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5682 osd_ipd_put(env, bag, ipd);
5683 LINVRNT(osd_invariant(obj));
5684 osd_trans_exec_check(env, th, OSD_OT_INSERT);
5689 * Calls ldiskfs_add_entry() to add directory entry
5690 * into the directory. This is required for
5691 * interoperability mode (b11826)
5693 * \retval 0, on success
5694 * \retval -ve, on error
5696 static int __osd_ea_add_rec(struct osd_thread_info *info,
5697 struct osd_object *pobj, struct inode *cinode,
5698 const char *name, const struct lu_fid *fid,
5699 struct htree_lock *hlock, struct thandle *th)
5701 struct ldiskfs_dentry_param *ldp;
5702 struct dentry *child;
5703 struct osd_thandle *oth;
5707 oth = container_of(th, struct osd_thandle, ot_super);
5708 LASSERT(oth->ot_handle != NULL);
5709 LASSERT(oth->ot_handle->h_transaction != NULL);
5710 LASSERT(pobj->oo_inode);
5712 rc = obj_name2lu_name(pobj, name, strlen(name), &ln);
5716 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5717 if (unlikely(osd_object_is_root(pobj)))
5720 osd_get_ldiskfs_dirent_param(ldp, fid);
5721 child = osd_child_dentry_get(info->oti_env, pobj,
5722 ln.ln_name, ln.ln_namelen);
5723 child->d_fsdata = (void *)ldp;
5724 dquot_initialize(pobj->oo_inode);
5725 rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5726 child, cinode, hlock);
5727 if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5728 struct ldiskfs_dir_entry_2 *de;
5729 struct buffer_head *bh;
5732 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5735 rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5738 if (S_ISDIR(cinode->i_mode))
5739 de->file_type = LDISKFS_DIRENT_LUFID |
5740 LDISKFS_FT_REG_FILE;
5742 de->file_type = LDISKFS_DIRENT_LUFID |
5744 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5751 if (ln.ln_name != name)
5757 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5758 * into the directory.Also sets flags into osd object to
5759 * indicate dot and dotdot are created. This is required for
5760 * interoperability mode (b11826)
5762 * \param dir directory for dot and dotdot fixup.
5763 * \param obj child object for linking
5765 * \retval 0, on success
5766 * \retval -ve, on error
5768 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5769 struct osd_object *dir,
5770 struct inode *parent_dir, const char *name,
5771 const struct lu_fid *dot_fid,
5772 const struct lu_fid *dot_dot_fid,
5775 struct inode *inode = dir->oo_inode;
5776 struct osd_thandle *oth;
5779 oth = container_of(th, struct osd_thandle, ot_super);
5780 LASSERT(oth->ot_handle->h_transaction != NULL);
5781 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5783 if (strcmp(name, dot) == 0) {
5784 if (dir->oo_compat_dot_created) {
5787 LASSERT(inode->i_ino == parent_dir->i_ino);
5788 dir->oo_compat_dot_created = 1;
5791 } else if (strcmp(name, dotdot) == 0) {
5792 if (!dir->oo_compat_dot_created)
5794 /* in case of rename, dotdot is already created */
5795 if (dir->oo_compat_dotdot_created) {
5796 return __osd_ea_add_rec(info, dir, parent_dir, name,
5797 dot_dot_fid, NULL, th);
5800 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5801 struct lu_fid tfid = *dot_dot_fid;
5804 result = osd_add_dot_dotdot_internal(info,
5805 dir->oo_inode, parent_dir, dot_fid,
5808 result = osd_add_dot_dotdot_internal(info,
5809 dir->oo_inode, parent_dir, dot_fid,
5814 dir->oo_compat_dotdot_created = 1;
5822 * It will call the appropriate osd_add* function and return the
5823 * value, return by respective functions.
5825 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5826 struct inode *cinode, const char *name,
5827 const struct lu_fid *fid, struct thandle *th)
5829 struct osd_thread_info *info = osd_oti_get(env);
5830 struct htree_lock *hlock;
5833 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5835 if (name[0] == '.' && (name[1] == '\0' ||
5836 (name[1] == '.' && name[2] == '\0'))) {
5837 if (hlock != NULL) {
5838 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5841 down_write(&pobj->oo_ext_idx_sem);
5844 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5845 lu_object_fid(&pobj->oo_dt.do_lu),
5848 if (hlock != NULL) {
5849 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5850 pobj->oo_inode, LDISKFS_HLOCK_ADD);
5852 down_write(&pobj->oo_ext_idx_sem);
5855 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5856 struct lu_fid *tfid = &info->oti_fid;
5860 rc = __osd_ea_add_rec(info, pobj, cinode, name,
5863 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5867 if (!rc && fid_is_namespace_visible(lu_object_fid(&pobj->oo_dt.do_lu))
5868 && pobj->oo_dirent_count != LU_DIRENT_COUNT_UNSET)
5869 pobj->oo_dirent_count++;
5872 ldiskfs_htree_unlock(hlock);
5874 up_write(&pobj->oo_ext_idx_sem);
5880 osd_ldiskfs_consistency_check(struct osd_thread_info *oti,
5881 struct osd_device *dev,
5882 const struct lu_fid *fid,
5883 struct osd_inode_id *id)
5885 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5886 struct inode *inode = NULL;
5892 if (!scrub_needs_check(scrub, fid, id->oii_ino))
5895 rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5896 if (rc == -ENOENT) {
5897 __u32 gen = id->oii_gen;
5903 inode = osd_iget(oti, dev, id);
5904 /* The inode has been removed (by race maybe). */
5905 if (IS_ERR(inode)) {
5906 rc = PTR_ERR(inode);
5908 RETURN(rc == -ESTALE ? -ENOENT : rc);
5911 /* The OI mapping is lost. */
5912 if (gen != OSD_OII_NOGEN)
5916 * The inode may has been reused by others, we do not know,
5917 * leave it to be handled by subsequent osd_fid_lookup().
5920 } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5927 if (scrub->os_running) {
5928 if (inode == NULL) {
5929 inode = osd_iget(oti, dev, id);
5930 /* The inode has been removed (by race maybe). */
5931 if (IS_ERR(inode)) {
5932 rc = PTR_ERR(inode);
5934 RETURN(rc == -ESTALE ? -ENOENT : rc);
5938 rc = osd_oii_insert(dev, fid, id, insert);
5940 * There is race condition between osd_oi_lookup and OI scrub.
5941 * The OI scrub finished just after osd_oi_lookup() failure.
5942 * Under such case, it is unnecessary to trigger OI scrub again,
5943 * but try to call osd_oi_lookup() again.
5945 if (unlikely(rc == -EAGAIN))
5948 if (!S_ISDIR(inode->i_mode))
5951 rc = osd_check_lmv(oti, dev, inode);
5956 if (dev->od_scrub.os_scrub.os_auto_scrub_interval != AS_NEVER &&
5958 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5959 SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5960 CDEBUG_LIMIT(D_LFSCK | D_CONSOLE | D_WARNING,
5961 "%s: trigger partial OI scrub for RPC inconsistency, checking FID "DFID"/%u: rc = %d\n",
5962 osd_dev2name(dev), PFID(fid), id->oii_ino, rc);
5963 if (rc == 0 || rc == -EALREADY)
5975 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5976 struct osd_device *dev,
5977 struct lu_fid *fid, __u32 ino)
5979 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5980 struct osd_idmap_cache *oic = &oti->oti_cache;
5981 struct inode *inode;
5984 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5985 inode = osd_iget(oti, dev, &oic->oic_lid);
5986 if (IS_ERR(inode)) {
5987 fid_zero(&oic->oic_fid);
5988 return PTR_ERR(inode);
5991 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5994 fid_zero(&oic->oic_fid);
5996 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
6000 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
6001 struct osd_inode_id *id, const struct lu_fid *fid)
6003 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
6004 id->oii_ino, id->oii_gen, info);
6005 info->oti_cache.oic_lid = *id;
6006 info->oti_cache.oic_fid = *fid;
6007 info->oti_cache.oic_dev = osd;
6011 * Get parent FID from the linkEA.
6013 * For a directory which parent resides on remote MDT, to satisfy the
6014 * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
6015 * the other hand, to make the lookup(..) on the directory can return
6016 * the real parent FID, we append the real parent FID after its ".."
6017 * name entry in the /REMOTE_PARENT_DIR.
6019 * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
6020 * backup. So after the restore, we cannot get the right parent FID from
6021 * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
6022 * we have stored the real parent FID in the directory object's linkEA,
6023 * we can parse the linkEA for the real parent FID.
6025 * \param[in] env pointer to the thread context
6026 * \param[in] obj pointer to the object to be handled
6027 * \param[out]fid pointer to the buffer to hold the parent FID
6029 * \retval 0 for getting the real parent FID successfully
6030 * \retval negative error number on failure
6032 static int osd_get_pfid_from_linkea(const struct lu_env *env,
6033 struct osd_object *obj,
6036 struct osd_thread_info *oti = osd_oti_get(env);
6037 struct lu_buf *buf = &oti->oti_big_buf;
6038 struct dentry *dentry = &oti->oti_obj_dentry;
6039 struct inode *inode = obj->oo_inode;
6040 struct linkea_data ldata = { NULL };
6046 if (!S_ISDIR(inode->i_mode))
6050 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
6051 buf->lb_buf, buf->lb_len);
6052 if (rc == -ERANGE) {
6053 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
6056 lu_buf_realloc(buf, rc);
6057 if (buf->lb_buf == NULL)
6064 if (unlikely(rc == 0))
6070 if (unlikely(buf->lb_buf == NULL)) {
6071 lu_buf_realloc(buf, rc);
6072 if (buf->lb_buf == NULL)
6079 rc = linkea_init_with_rec(&ldata);
6081 linkea_first_entry(&ldata);
6082 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
6088 static int osd_verify_ent_by_linkea(const struct lu_env *env,
6089 struct inode *inode,
6090 const struct lu_fid *pfid,
6091 const char *name, const int namelen)
6093 struct osd_thread_info *oti = osd_oti_get(env);
6094 struct lu_buf *buf = &oti->oti_big_buf;
6095 struct dentry *dentry = &oti->oti_obj_dentry;
6096 struct linkea_data ldata = { NULL };
6097 struct lu_name cname = { .ln_name = name,
6098 .ln_namelen = namelen };
6104 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
6105 buf->lb_buf, buf->lb_len);
6107 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
6112 if (unlikely(rc == 0))
6115 if (buf->lb_len < rc) {
6116 lu_buf_realloc(buf, rc);
6117 if (buf->lb_buf == NULL)
6124 rc = linkea_init_with_rec(&ldata);
6126 rc = linkea_links_find(&ldata, &cname, pfid);
6132 * Calls ->lookup() to find dentry. From dentry get inode and
6133 * read inode's ea to get fid. This is required for interoperability
6136 * \retval 0, on success
6137 * \retval -ve, on error
6139 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
6140 struct dt_rec *rec, const struct dt_key *key)
6142 struct inode *dir = obj->oo_inode;
6143 struct dentry *dentry;
6144 struct ldiskfs_dir_entry_2 *de;
6145 struct buffer_head *bh;
6146 struct lu_fid *fid = (struct lu_fid *)rec;
6147 struct htree_lock *hlock = NULL;
6154 LASSERT(dir->i_op != NULL);
6155 LASSERT(dir->i_op->lookup != NULL);
6157 rc = obj_name2lu_name(obj, (char *)key, strlen((char *)key), &ln);
6161 dentry = osd_child_dentry_get(env, obj, ln.ln_name, ln.ln_namelen);
6163 if (obj->oo_hl_head != NULL) {
6164 hlock = osd_oti_get(env)->oti_hlock;
6165 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6166 dir, LDISKFS_HLOCK_LOOKUP);
6168 down_read(&obj->oo_ext_idx_sem);
6171 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6173 struct osd_thread_info *oti = osd_oti_get(env);
6174 struct osd_inode_id *id = &oti->oti_id;
6175 struct osd_device *dev = osd_obj2dev(obj);
6177 ino = le32_to_cpu(de->inode);
6178 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
6180 rc = osd_fail_fid_lookup(oti, dev, fid, ino);
6184 rc = osd_get_fid_from_dentry(de, rec);
6186 /* done with de, release bh */
6189 if (unlikely(is_remote_parent_ino(dev, ino))) {
6190 const char *name = (const char *)key;
6193 * If the parent is on remote MDT, and there
6194 * is no FID-in-dirent, then we have to get
6195 * the parent FID from the linkEA.
6197 if (likely(strlen(name) == 2 &&
6198 name[0] == '.' && name[1] == '.'))
6199 rc = osd_get_pfid_from_linkea(env, obj,
6202 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6205 osd_id_gen(id, ino, OSD_OII_NOGEN);
6208 if (rc != 0 || osd_remote_fid(env, dev, fid))
6211 rc = osd_ldiskfs_consistency_check(oti, dev, fid, id);
6212 if (rc != -ENOENT) {
6213 /* Other error should not affect lookup result. */
6216 /* Normal file mapping should be added into OI cache
6217 * after FID in LMA check, but for local files like
6218 * hsm_actions, their FIDs are not stored in OI files,
6219 * see osd_initial_OI_scrub(), and here is the only
6220 * place to load mapping into OI cache.
6222 if (!fid_is_namespace_visible(fid))
6223 osd_add_oi_cache(osd_oti_get(env),
6224 osd_obj2dev(obj), id, fid);
6234 ldiskfs_htree_unlock(hlock);
6236 up_read(&obj->oo_ext_idx_sem);
6237 if (ln.ln_name != (char *)key)
6242 static int osd_index_declare_ea_insert(const struct lu_env *env,
6243 struct dt_object *dt,
6244 const struct dt_rec *rec,
6245 const struct dt_key *key,
6246 struct thandle *handle)
6248 struct osd_thandle *oh;
6249 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6250 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6251 const struct lu_fid *fid = rec1->rec_fid;
6252 int credits, rc = 0;
6253 struct osd_idmap_cache *idc;
6257 LASSERT(!dt_object_remote(dt));
6258 LASSERT(handle != NULL);
6259 LASSERT(fid != NULL);
6260 LASSERT(rec1->rec_type != 0);
6262 oh = container_of(handle, struct osd_thandle, ot_super);
6263 LASSERT(oh->ot_handle == NULL);
6265 credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
6268 * we can't call iget() while a transactions is running
6269 * (this can lead to a deadlock), but we need to know
6270 * inum and object type. so we find this information at
6271 * declaration and cache in per-thread info
6273 idc = osd_idc_find_or_init(env, osd, fid);
6275 RETURN(PTR_ERR(idc));
6276 if (idc->oic_remote) {
6278 * a reference to remote inode is represented by an
6279 * agent inode which we have to create
6281 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
6282 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
6285 osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
6287 if (osd_dt_obj(dt)->oo_inode != NULL) {
6288 struct inode *inode = osd_dt_obj(dt)->oo_inode;
6291 * We ignore block quota on meta pool (MDTs), so needn't
6292 * calculate how many blocks will be consumed by this index
6295 rc = osd_declare_inode_qid(env, i_uid_read(inode),
6297 i_projid_read(inode), 0,
6298 oh, osd_dt_obj(dt), NULL,
6303 #ifdef HAVE_PROJECT_QUOTA
6305 * Reserve credits for local agent inode to transfer
6306 * to 0, quota enforcement is ignored in this case.
6308 if (idc->oic_remote &&
6309 LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
6310 i_projid_read(inode) != 0)
6311 rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
6312 0, i_projid_read(inode),
6313 0, false, PRJQUOTA);
6321 * Index add function for interoperability mode (b11826).
6322 * It will add the directory entry.This entry is needed to
6323 * maintain name->fid mapping.
6325 * \param key it is key i.e. file entry to be inserted
6326 * \param rec it is value of given key i.e. fid
6328 * \retval 0, on success
6329 * \retval -ve, on error
6331 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6332 const struct dt_rec *rec,
6333 const struct dt_key *key, struct thandle *th)
6335 struct osd_object *obj = osd_dt_obj(dt);
6336 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6337 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6338 const struct lu_fid *fid = rec1->rec_fid;
6339 const char *name = (const char *)key;
6340 struct osd_thread_info *oti = osd_oti_get(env);
6341 struct inode *child_inode = NULL;
6342 struct osd_idmap_cache *idc;
6347 if (!dt_object_exists(dt))
6350 LASSERT(osd_invariant(obj));
6351 LASSERT(!dt_object_remote(dt));
6352 LASSERT(th != NULL);
6354 osd_trans_exec_op(env, th, OSD_OT_INSERT);
6356 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6358 idc = osd_idc_find(env, osd, fid);
6359 if (unlikely(idc == NULL)) {
6360 idc = osd_idc_find_or_init(env, osd, fid);
6363 * this dt_insert() wasn't declared properly, so
6364 * FID is missing in OI cache. we better do not
6365 * lookup FID in FLDB/OI and don't risk to deadlock,
6366 * but in some special cases (lfsck testing, etc)
6367 * it's much simpler than fixing a caller.
6369 * normally this error should be placed after the first
6370 * find, but migrate may attach source stripes to
6371 * target, which doesn't create stripes.
6373 CERROR("%s: "DFID" wasn't declared for insert\n",
6374 osd_name(osd), PFID(fid));
6376 RETURN(PTR_ERR(idc));
6380 if (idc->oic_remote) {
6381 /* Insert remote entry */
6382 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6384 igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6386 child_inode = osd_create_local_agent_inode(env, osd,
6387 obj, fid, rec1->rec_type & S_IFMT, th);
6388 if (IS_ERR(child_inode))
6389 RETURN(PTR_ERR(child_inode));
6392 /* Insert local entry */
6393 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6394 /* for a reason OI cache wasn't filled properly */
6395 CERROR("%s: OIC for "DFID" isn't filled\n",
6396 osd_name(osd), PFID(fid));
6399 child_inode = oti->oti_inode;
6400 if (unlikely(child_inode == NULL)) {
6401 struct ldiskfs_inode_info *lii;
6406 child_inode = oti->oti_inode = &lii->vfs_inode;
6408 child_inode->i_sb = osd_sb(osd);
6409 child_inode->i_ino = idc->oic_lid.oii_ino;
6410 child_inode->i_mode = rec1->rec_type & S_IFMT;
6413 rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6415 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6416 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6418 if (child_inode && child_inode != oti->oti_inode)
6420 LASSERT(osd_invariant(obj));
6421 osd_trans_exec_check(env, th, OSD_OT_INSERT);
6427 * Initialize osd Iterator for given osd index object.
6429 * \param dt osd index object
6432 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6433 struct dt_object *dt,
6436 struct osd_it_iam *it;
6437 struct osd_object *obj = osd_dt_obj(dt);
6438 struct lu_object *lo = &dt->do_lu;
6439 struct iam_path_descr *ipd;
6440 struct iam_container *bag = &obj->oo_dir->od_container;
6442 if (!dt_object_exists(dt))
6443 return ERR_PTR(-ENOENT);
6447 return ERR_PTR(-ENOMEM);
6449 ipd = osd_it_ipd_get(env, bag);
6450 if (likely(ipd != NULL)) {
6454 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6455 return (struct dt_it *)it;
6458 return ERR_PTR(-ENOMEM);
6463 * free given Iterator.
6465 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6467 struct osd_it_iam *it = (struct osd_it_iam *)di;
6468 struct osd_object *obj = it->oi_obj;
6470 iam_it_fini(&it->oi_it);
6471 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6472 osd_object_put(env, obj);
6477 * Move Iterator to record specified by \a key
6479 * \param di osd iterator
6480 * \param key key for index
6482 * \retval +ve di points to record with least key not larger than key
6483 * \retval 0 di points to exact matched key
6484 * \retval -ve failure
6487 static int osd_it_iam_get(const struct lu_env *env,
6488 struct dt_it *di, const struct dt_key *key)
6490 struct osd_thread_info *oti = osd_oti_get(env);
6491 struct osd_it_iam *it = (struct osd_it_iam *)di;
6493 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6494 /* swab quota uid/gid */
6495 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6496 key = (struct dt_key *)&oti->oti_quota_id;
6499 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6505 * \param di osd iterator
6507 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6509 struct osd_it_iam *it = (struct osd_it_iam *)di;
6511 iam_it_put(&it->oi_it);
6515 * Move iterator by one record
6517 * \param di osd iterator
6519 * \retval +1 end of container reached
6521 * \retval -ve failure
6524 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6526 struct osd_it_iam *it = (struct osd_it_iam *)di;
6528 return iam_it_next(&it->oi_it);
6532 * Return pointer to the key under iterator.
6535 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6536 const struct dt_it *di)
6538 struct osd_thread_info *oti = osd_oti_get(env);
6539 struct osd_it_iam *it = (struct osd_it_iam *)di;
6540 struct osd_object *obj = it->oi_obj;
6543 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6545 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6546 /* swab quota uid/gid */
6547 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6548 key = (struct dt_key *)&oti->oti_quota_id;
6555 * Return size of key under iterator (in bytes)
6558 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6560 struct osd_it_iam *it = (struct osd_it_iam *)di;
6562 return iam_it_key_size(&it->oi_it);
6566 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6568 /* check if file type is required */
6569 if (ent->lde_attrs & LUDA_TYPE) {
6570 struct luda_type *lt;
6571 int align = sizeof(*lt) - 1;
6573 len = (len + align) & ~align;
6574 lt = (struct luda_type *)(ent->lde_name + len);
6575 lt->lt_type = cpu_to_le16(DTTOIF(type));
6578 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6582 * build lu direct from backend fs dirent.
6586 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6587 char *name, __u16 namelen, __u16 type, __u32 attr)
6589 ent->lde_attrs = attr | LUDA_FID;
6590 fid_cpu_to_le(&ent->lde_fid, fid);
6592 ent->lde_hash = cpu_to_le64(offset);
6593 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6595 strncpy(ent->lde_name, name, namelen);
6596 ent->lde_name[namelen] = '\0';
6597 ent->lde_namelen = cpu_to_le16(namelen);
6599 /* append lustre attributes */
6600 osd_it_append_attrs(ent, namelen, type);
6604 * Return pointer to the record under iterator.
6606 static int osd_it_iam_rec(const struct lu_env *env,
6607 const struct dt_it *di,
6608 struct dt_rec *dtrec, __u32 attr)
6610 struct osd_it_iam *it = (struct osd_it_iam *)di;
6611 struct osd_thread_info *info = osd_oti_get(env);
6615 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6616 const struct osd_fid_pack *rec;
6617 struct lu_fid *fid = &info->oti_fid;
6618 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6624 name = (char *)iam_it_key_get(&it->oi_it);
6626 RETURN(PTR_ERR(name));
6628 namelen = iam_it_key_size(&it->oi_it);
6630 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6632 RETURN(PTR_ERR(rec));
6634 rc = osd_fid_unpack(fid, rec);
6638 hash = iam_it_store(&it->oi_it);
6640 /* IAM does not store object type in IAM index (dir) */
6641 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6643 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6644 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6645 (struct iam_rec *)dtrec);
6646 osd_quota_unpack(it->oi_obj, dtrec);
6648 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6649 (struct iam_rec *)dtrec);
6656 * Returns cookie for current Iterator position.
6658 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6660 struct osd_it_iam *it = (struct osd_it_iam *)di;
6662 return iam_it_store(&it->oi_it);
6666 * Restore iterator from cookie.
6668 * \param di osd iterator
6669 * \param hash Iterator location cookie
6671 * \retval +ve di points to record with least key not larger than key.
6672 * \retval 0 di points to exact matched key
6673 * \retval -ve failure
6676 static int osd_it_iam_load(const struct lu_env *env,
6677 const struct dt_it *di, __u64 hash)
6679 struct osd_it_iam *it = (struct osd_it_iam *)di;
6681 return iam_it_load(&it->oi_it, hash);
6684 static const struct dt_index_operations osd_index_iam_ops = {
6685 .dio_lookup = osd_index_iam_lookup,
6686 .dio_declare_insert = osd_index_declare_iam_insert,
6687 .dio_insert = osd_index_iam_insert,
6688 .dio_declare_delete = osd_index_declare_iam_delete,
6689 .dio_delete = osd_index_iam_delete,
6691 .init = osd_it_iam_init,
6692 .fini = osd_it_iam_fini,
6693 .get = osd_it_iam_get,
6694 .put = osd_it_iam_put,
6695 .next = osd_it_iam_next,
6696 .key = osd_it_iam_key,
6697 .key_size = osd_it_iam_key_size,
6698 .rec = osd_it_iam_rec,
6699 .store = osd_it_iam_store,
6700 .load = osd_it_iam_load
6704 struct osd_it_ea *osd_it_dir_init(const struct lu_env *env,
6705 struct osd_device *dev,
6706 struct inode *inode, u32 attr)
6708 struct osd_thread_info *info = osd_oti_get(env);
6709 struct osd_it_ea *oie;
6713 file = alloc_file_pseudo(inode, dev->od_mnt, "/", O_NOATIME,
6716 RETURN(ERR_CAST(file));
6718 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6719 if (attr & LUDA_64BITHASH)
6720 file->f_mode |= FMODE_64BITHASH;
6722 file->f_mode |= FMODE_32BITHASH;
6725 OBD_SLAB_ALLOC_PTR(oie, osd_itea_cachep);
6729 oie->oie_rd_dirent = 0;
6730 oie->oie_it_dirent = 0;
6731 oie->oie_dirent = NULL;
6732 if (unlikely(!info->oti_it_ea_buf_used)) {
6733 oie->oie_buf = info->oti_it_ea_buf;
6734 info->oti_it_ea_buf_used = 1;
6736 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6740 oie->oie_obj = NULL;
6741 oie->oie_file = file;
6746 OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6750 return ERR_PTR(-ENOMEM);
6754 * Creates or initializes iterator context.
6756 * \retval struct osd_it_ea, iterator structure on success
6759 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6760 struct dt_object *dt,
6763 struct osd_object *obj = osd_dt_obj(dt);
6764 struct osd_device *dev = osd_obj2dev(obj);
6765 struct lu_object *lo = &dt->do_lu;
6766 struct osd_it_ea *oie;
6770 if (!dt_object_exists(dt) || obj->oo_destroyed)
6771 RETURN(ERR_PTR(-ENOENT));
6773 oie = osd_it_dir_init(env, dev, obj->oo_inode, attr);
6775 RETURN(ERR_CAST(oie));
6779 RETURN((struct dt_it *)oie);
6782 void osd_it_dir_fini(const struct lu_env *env, struct osd_it_ea *oie,
6783 struct inode *inode)
6785 struct osd_thread_info *info = osd_oti_get(env);
6788 fput(oie->oie_file);
6789 if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6790 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6792 info->oti_it_ea_buf_used = 0;
6793 OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6798 * Destroy or finishes iterator context.
6800 * \param di iterator structure to be destroyed
6802 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6804 struct osd_it_ea *oie = (struct osd_it_ea *)di;
6805 struct osd_object *obj = oie->oie_obj;
6806 struct inode *inode = obj->oo_inode;
6809 osd_it_dir_fini(env, (struct osd_it_ea *)di, inode);
6810 osd_object_put(env, obj);
6815 * It position the iterator at given key, so that next lookup continues from
6816 * that key Or it is similar to dio_it->load() but based on a key,
6817 * rather than file position.
6819 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6822 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6824 static int osd_it_ea_get(const struct lu_env *env,
6825 struct dt_it *di, const struct dt_key *key)
6827 struct osd_it_ea *it = (struct osd_it_ea *)di;
6830 LASSERT(((const char *)key)[0] == '\0');
6831 it->oie_file->f_pos = 0;
6832 it->oie_rd_dirent = 0;
6833 it->oie_it_dirent = 0;
6834 it->oie_dirent = NULL;
6842 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6846 struct osd_filldir_cbs {
6847 struct dir_context ctx;
6848 struct osd_it_ea *it;
6851 * It is called internally by ->iterate*(). It fills the
6852 * iterator's in-memory data structure with required
6853 * information i.e. name, namelen, rec_size etc.
6855 * \param buf in which information to be filled in.
6856 * \param name name of the file in given dir
6858 * \retval 0 on success
6859 * \retval 1 on buffer full
6861 #ifdef HAVE_FILLDIR_USE_CTX
6862 static int osd_ldiskfs_filldir(struct dir_context *ctx,
6864 static int osd_ldiskfs_filldir(void *ctx,
6866 const char *name, int namelen,
6867 loff_t offset, __u64 ino, unsigned int d_type)
6869 struct osd_it_ea *it = ((struct osd_filldir_cbs *)ctx)->it;
6870 struct osd_object *obj = it->oie_obj;
6871 struct osd_it_ea_dirent *ent = it->oie_dirent;
6872 struct lu_fid *fid = &ent->oied_fid;
6873 char *buf = it->oie_buf;
6874 struct osd_fid_pack *rec;
6877 /* this should never happen */
6878 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6879 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6883 /* Check for enough space. Note oied_name is not NUL terminated. */
6884 if (&ent->oied_name[namelen] > buf + OSD_IT_EA_BUFSIZE)
6887 /* "." is just the object itself. */
6888 if (namelen == 1 && name[0] == '.') {
6890 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6891 } else if (d_type & LDISKFS_DIRENT_LUFID) {
6892 rec = (struct osd_fid_pack *)(name + namelen + 1);
6893 if (osd_fid_unpack(fid, rec) != 0)
6898 d_type &= ~LDISKFS_DIRENT_LUFID;
6900 /* NOT export local root. */
6902 unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6903 ino = obj->oo_inode->i_ino;
6904 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6907 if (obj == NULL || !(obj->oo_lma_flags & LUSTRE_ENCRYPT_FL)) {
6908 ent->oied_namelen = namelen;
6909 memcpy(ent->oied_name, name, namelen);
6911 int encoded_namelen = critical_chars(name, namelen);
6913 /* Check again for enough space. */
6914 if (&ent->oied_name[encoded_namelen] > buf + OSD_IT_EA_BUFSIZE)
6917 ent->oied_namelen = encoded_namelen;
6919 if (encoded_namelen == namelen)
6920 memcpy(ent->oied_name, name, namelen);
6922 critical_encode(name, namelen, ent->oied_name);
6925 ent->oied_ino = ino;
6926 ent->oied_off = offset;
6927 ent->oied_type = d_type;
6929 it->oie_rd_dirent++;
6930 it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + ent->oied_namelen);
6935 * Calls ->iterate*() to load a directory entry at a time
6936 * and stored it in iterator's in-memory data structure.
6938 * \param di iterator's in memory structure
6940 * \retval 0 on success
6941 * \retval -ve on error
6942 * \retval +1 reach the end of entry
6944 int osd_ldiskfs_it_fill(const struct lu_env *env, const struct dt_it *di)
6946 struct osd_it_ea *it = (struct osd_it_ea *)di;
6947 struct osd_object *obj = it->oie_obj;
6948 struct htree_lock *hlock = NULL;
6949 struct file *filp = it->oie_file;
6951 struct osd_filldir_cbs buf = {
6952 .ctx.actor = osd_ldiskfs_filldir,
6957 it->oie_dirent = it->oie_buf;
6958 it->oie_rd_dirent = 0;
6961 if (obj->oo_hl_head != NULL) {
6962 hlock = osd_oti_get(env)->oti_hlock;
6963 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6965 LDISKFS_HLOCK_READDIR);
6967 down_read(&obj->oo_ext_idx_sem);
6971 rc = iterate_dir(filp, &buf.ctx);
6975 if (it->oie_rd_dirent == 0) {
6977 * If it does not get any dirent, it means it has been reached
6978 * to the end of the dir
6980 it->oie_file->f_pos = ldiskfs_get_htree_eof(it->oie_file);
6984 it->oie_dirent = it->oie_buf;
6985 it->oie_it_dirent = 1;
6990 ldiskfs_htree_unlock(hlock);
6992 up_read(&obj->oo_ext_idx_sem);
6999 * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
7000 * to load a directory entry at a time and stored it in
7001 * iterator's in-memory data structure.
7003 * \param di iterator's in memory structure
7005 * \retval +ve iterator reached to end
7006 * \retval 0 iterator not reached to end
7007 * \retval -ve on error
7009 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
7011 struct osd_it_ea *it = (struct osd_it_ea *)di;
7016 if (it->oie_it_dirent < it->oie_rd_dirent) {
7018 (void *)it->oie_dirent +
7019 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
7020 it->oie_dirent->oied_namelen);
7021 it->oie_it_dirent++;
7024 if (it->oie_file->f_pos == ldiskfs_get_htree_eof(it->oie_file))
7027 rc = osd_ldiskfs_it_fill(env, di);
7034 * Returns the key at current position from iterator's in memory structure.
7036 * \param di iterator's in memory structure
7038 * \retval key i.e. struct dt_key on success
7040 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
7041 const struct dt_it *di)
7043 struct osd_it_ea *it = (struct osd_it_ea *)di;
7045 return (struct dt_key *)it->oie_dirent->oied_name;
7049 * Returns key's size at current position from iterator's in memory structure.
7051 * \param di iterator's in memory structure
7053 * \retval key_size i.e. struct dt_key on success
7055 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
7057 struct osd_it_ea *it = (struct osd_it_ea *)di;
7059 return it->oie_dirent->oied_namelen;
7062 #if defined LDISKFS_DIR_ENTRY_LEN && defined LDISKFS_DIR_ENTRY_LEN_
7063 #undef LDISKFS_DIR_REC_LEN
7064 #define LDISKFS_DIR_REC_LEN(de) LDISKFS_DIR_ENTRY_LEN_((de))
7067 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
7069 if (LDISKFS_DIR_REC_LEN(de) >=
7070 __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
7077 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
7078 unsigned int blocksize, bool dotdot)
7081 return osd_dotdot_has_space(de);
7083 if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
7084 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
7091 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
7092 handle_t *jh, struct dentry *dentry,
7093 const struct lu_fid *fid, struct buffer_head *bh,
7094 struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
7097 struct inode *dir = dentry->d_parent->d_inode;
7098 struct inode *inode = dentry->d_inode;
7099 struct osd_fid_pack *rec;
7100 struct ldiskfs_dentry_param *ldp;
7101 int namelen = dentry->d_name.len;
7103 struct osd_thread_info *info = osd_oti_get(env);
7107 if (!ldiskfs_has_feature_dirdata(inode->i_sb))
7110 /* There is enough space to hold the FID-in-dirent. */
7111 if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
7112 rc = ldiskfs_journal_get_write_access(jh, bh);
7116 de->name[namelen] = 0;
7117 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
7118 rec->fp_len = sizeof(struct lu_fid) + 1;
7119 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
7120 de->file_type |= LDISKFS_DIRENT_LUFID;
7121 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
7128 rc = ldiskfs_delete_entry(jh, dir, de, bh);
7132 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
7133 osd_get_ldiskfs_dirent_param(ldp, fid);
7134 dentry->d_fsdata = (void *)ldp;
7135 dquot_initialize(dir);
7136 rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
7138 * It is too bad, we cannot reinsert the name entry back.
7139 * That means we lose it!
7143 "%s: fail to reinsert the dirent, dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
7144 osd_ino2name(inode), dir->i_ino, dir->i_generation,
7145 namelen, dentry->d_name.name, PFID(fid), rc);
7151 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
7152 struct osd_it_ea *it, struct lu_fid *fid,
7153 struct osd_inode_id *id, __u32 *attr)
7155 struct osd_thread_info *info = osd_oti_get(env);
7156 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
7157 struct osd_device *dev = osd_obj2dev(obj);
7158 struct super_block *sb = osd_sb(dev);
7159 const char *devname = osd_name(dev);
7160 struct osd_it_ea_dirent *ent = it->oie_dirent;
7161 struct inode *dir = obj->oo_inode;
7162 struct htree_lock *hlock = NULL;
7163 struct buffer_head *bh = NULL;
7164 handle_t *jh = NULL;
7165 struct ldiskfs_dir_entry_2 *de;
7166 struct dentry *dentry;
7167 struct inode *inode;
7168 const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
7171 bool dotdot = false;
7177 if (ent->oied_name[0] == '.') {
7178 if (ent->oied_namelen == 1)
7181 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
7185 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
7186 inode = osd_iget(info, dev, id);
7187 if (IS_ERR(inode)) {
7188 rc = PTR_ERR(inode);
7189 if (rc == -ENOENT || rc == -ESTALE) {
7191 * Maybe dangling name entry, or
7192 * corrupted directory entry.
7194 *attr |= LUDA_UNKNOWN;
7197 CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
7198 "check_repair, dir = %lu/%u, name = %.*s, "
7199 "ino = %llu, rc = %d\n",
7200 devname, dir->i_ino, dir->i_generation,
7201 ent->oied_namelen, ent->oied_name,
7208 rc = obj_name2lu_name(obj, ent->oied_name, ent->oied_namelen, &ln);
7212 dentry = osd_child_dentry_by_inode(env, dir, ln.ln_name, ln.ln_namelen);
7213 rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
7214 if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
7220 * We need to ensure that the name entry is still valid.
7221 * Because it may be removed or renamed by other already.
7223 * The unlink or rename operation will start journal before PDO lock,
7224 * so to avoid deadlock, here we need to start journal handle before
7225 * related PDO lock also. But because we do not know whether there
7226 * will be something to be repaired before PDO lock, we just start
7227 * journal without conditions.
7229 * We may need to remove the name entry firstly, then insert back.
7230 * One credit is for user quota file update.
7231 * One credit is for group quota file update.
7232 * Two credits are for dirty inode.
7234 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
7235 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
7237 if (dev->od_dirent_journal != 0) {
7240 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
7243 CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
7244 "check_repair, dir = %lu/%u, credits = %d, "
7245 "name = %.*s, ino = %llu: rc = %d\n",
7246 devname, dir->i_ino, dir->i_generation, credits,
7247 ent->oied_namelen, ent->oied_name,
7250 GOTO(out_inode, rc);
7253 if (obj->oo_hl_head != NULL) {
7254 hlock = osd_oti_get(env)->oti_hlock;
7256 * "0" means exclusive lock for the whole directory.
7257 * We need to prevent others access such name entry
7258 * during the delete + insert. Neither HLOCK_ADD nor
7259 * HLOCK_DEL cannot guarantee the atomicity.
7261 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
7263 down_write(&obj->oo_ext_idx_sem);
7266 if (obj->oo_hl_head != NULL) {
7267 hlock = osd_oti_get(env)->oti_hlock;
7268 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
7269 LDISKFS_HLOCK_LOOKUP);
7271 down_read(&obj->oo_ext_idx_sem);
7275 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
7276 if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
7277 *attr |= LUDA_IGNORE;
7283 * For dotdot entry, if there is not enough space to hold the
7284 * FID-in-dirent, just keep them there. It only happens when the
7285 * device upgraded from 1.8 or restored from MDT file-level backup.
7286 * For the whole directory, only dotdot entry have no FID-in-dirent
7287 * and needs to get FID from LMA when readdir, it will not affect the
7290 if (dotdot && !osd_dotdot_has_space(de)) {
7291 *attr |= LUDA_UNKNOWN;
7297 if (lu_fid_eq(fid, &lma->lma_self_fid))
7300 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
7301 struct lu_fid *tfid = &lma->lma_self_fid;
7303 if (likely(dotdot &&
7304 fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
7305 fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
7307 * It must be REMOTE_PARENT_DIR and as the
7308 * 'dotdot' entry of remote directory
7310 *attr |= LUDA_IGNORE;
7312 CDEBUG(D_LFSCK, "%s: expect remote agent "
7313 "parent directory, but got %.*s under "
7314 "dir = %lu/%u with the FID "DFID"\n",
7315 devname, ent->oied_namelen,
7316 ent->oied_name, dir->i_ino,
7317 dir->i_generation, PFID(tfid));
7319 *attr |= LUDA_UNKNOWN;
7326 if (!fid_is_zero(fid)) {
7327 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
7329 if (rc == -ENOENT ||
7331 !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
7333 * linkEA does not recognize the dirent entry,
7334 * it may because the dirent entry corruption
7335 * and points to other's inode.
7337 CDEBUG(D_LFSCK, "%s: the target inode does not "
7338 "recognize the dirent, dir = %lu/%u, "
7339 " 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;
7348 if (rc && rc != -ENODATA) {
7349 CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
7350 "dir = %lu/%u, name = %.*s, ino = %llu, "
7351 DFID": rc = %d\n", devname, dir->i_ino,
7352 dir->i_generation, ent->oied_namelen,
7353 ent->oied_name, ent->oied_ino, PFID(fid), rc);
7354 *attr |= LUDA_UNKNOWN;
7362 * linkEA recognizes the dirent entry, the FID-in-LMA is
7363 * valid, trusted, in spite of fid_is_sane(fid) or not.
7365 if (*attr & LUDA_VERIFY_DRYRUN) {
7366 *fid = lma->lma_self_fid;
7367 *attr |= LUDA_REPAIR;
7374 dev->od_dirent_journal = 1;
7375 if (hlock != NULL) {
7376 ldiskfs_htree_unlock(hlock);
7379 up_read(&obj->oo_ext_idx_sem);
7385 *fid = lma->lma_self_fid;
7387 /* Update or append the FID-in-dirent. */
7388 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7389 bh, de, hlock, dotdot);
7391 *attr |= LUDA_REPAIR;
7393 CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7394 "the dirent, dir = %lu/%u, name = %.*s, "
7395 "ino = %llu, "DFID": rc = %d\n",
7396 devname, dir->i_ino, dir->i_generation,
7397 ent->oied_namelen, ent->oied_name,
7398 ent->oied_ino, PFID(fid), rc);
7400 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7401 if (*attr & LUDA_VERIFY_DRYRUN) {
7402 if (fid_is_sane(fid)) {
7403 *attr |= LUDA_REPAIR;
7404 } else if (dev->od_index == 0) {
7405 lu_igif_build(fid, inode->i_ino,
7406 inode->i_generation);
7407 *attr |= LUDA_UPGRADE;
7415 dev->od_dirent_journal = 1;
7416 if (hlock != NULL) {
7417 ldiskfs_htree_unlock(hlock);
7420 up_read(&obj->oo_ext_idx_sem);
7427 if (unlikely(fid_is_sane(fid))) {
7429 * FID-in-dirent exists, but FID-in-LMA is lost.
7430 * Trust the FID-in-dirent, and add FID-in-LMA.
7432 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7434 *attr |= LUDA_REPAIR;
7436 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7437 "update dirent, dir = %lu/%u, "
7438 "name = %.*s, ino = %llu, "
7440 devname, dir->i_ino, dir->i_generation,
7441 ent->oied_namelen, ent->oied_name,
7442 ent->oied_ino, PFID(fid), rc);
7443 } else if (dev->od_index == 0) {
7444 lu_igif_build(fid, inode->i_ino, inode->i_generation);
7446 * It is probably IGIF object. Only aappend the
7447 * FID-in-dirent. OI scrub will process FID-in-LMA.
7449 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7450 bh, de, hlock, dotdot);
7452 *attr |= LUDA_UPGRADE;
7454 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7455 "after the dirent, dir = %lu/%u, "
7456 "name = %.*s, ino = %llu, "
7458 devname, dir->i_ino, dir->i_generation,
7459 ent->oied_namelen, ent->oied_name,
7460 ent->oied_ino, PFID(fid), rc);
7469 if (hlock != NULL) {
7470 ldiskfs_htree_unlock(hlock);
7472 if (dev->od_dirent_journal != 0)
7473 up_write(&obj->oo_ext_idx_sem);
7475 up_read(&obj->oo_ext_idx_sem);
7479 ldiskfs_journal_stop(jh);
7483 if (rc >= 0 && !dirty)
7484 dev->od_dirent_journal = 0;
7485 if (ln.ln_name != ent->oied_name)
7492 * Returns the value at current position from iterator's in memory structure.
7494 * \param di struct osd_it_ea, iterator's in memory structure
7495 * \param attr attr requested for dirent.
7496 * \param lde lustre dirent
7498 * \retval 0 no error and \param lde has correct lustre dirent.
7499 * \retval -ve on error
7501 static inline int osd_it_ea_rec(const struct lu_env *env,
7502 const struct dt_it *di,
7503 struct dt_rec *dtrec, __u32 attr)
7505 struct osd_it_ea *it = (struct osd_it_ea *)di;
7506 struct osd_object *obj = it->oie_obj;
7507 struct osd_device *dev = osd_obj2dev(obj);
7508 struct osd_thread_info *oti = osd_oti_get(env);
7509 struct osd_inode_id *id = &oti->oti_id;
7510 struct lu_fid *fid = &it->oie_dirent->oied_fid;
7511 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
7512 __u32 ino = it->oie_dirent->oied_ino;
7517 LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7519 if (attr & LUDA_VERIFY) {
7520 if (unlikely(is_remote_parent_ino(dev, ino))) {
7521 attr |= LUDA_IGNORE;
7523 * If the parent is on remote MDT, and there
7524 * is no FID-in-dirent, then we have to get
7525 * the parent FID from the linkEA.
7527 if (!fid_is_sane(fid) &&
7528 it->oie_dirent->oied_namelen == 2 &&
7529 it->oie_dirent->oied_name[0] == '.' &&
7530 it->oie_dirent->oied_name[1] == '.')
7531 osd_get_pfid_from_linkea(env, obj, fid);
7533 rc = osd_dirent_check_repair(env, obj, it, fid, id,
7537 if (!fid_is_sane(fid))
7538 attr |= LUDA_UNKNOWN;
7540 attr &= ~LU_DIRENT_ATTRS_MASK;
7541 if (!fid_is_sane(fid)) {
7542 bool is_dotdot = false;
7544 if (it->oie_dirent->oied_namelen == 2 &&
7545 it->oie_dirent->oied_name[0] == '.' &&
7546 it->oie_dirent->oied_name[1] == '.')
7549 * If the parent is on remote MDT, and there
7550 * is no FID-in-dirent, then we have to get
7551 * the parent FID from the linkEA.
7553 if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7554 rc = osd_get_pfid_from_linkea(env, obj, fid);
7556 if (is_dotdot == false &&
7557 OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7560 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7565 /* Pack the entry anyway, at least the offset is right. */
7566 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7567 it->oie_dirent->oied_name,
7568 it->oie_dirent->oied_namelen,
7569 it->oie_dirent->oied_type, attr);
7571 RETURN(rc > 0 ? 0 : rc);
7575 * Returns the record size size at current position.
7577 * This function will return record(lu_dirent) size in bytes.
7579 * \param[in] env execution environment
7580 * \param[in] di iterator's in memory structure
7581 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
7582 * calculate the lu_dirent size.
7584 * \retval record size(in bytes & in memory) of the current lu_dirent
7587 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7590 struct osd_it_ea *it = (struct osd_it_ea *)di;
7592 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7596 * Returns a cookie for current position of the iterator head, so that
7597 * user can use this cookie to load/start the iterator next time.
7599 * \param di iterator's in memory structure
7601 * \retval cookie for current position, on success
7603 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7605 struct osd_it_ea *it = (struct osd_it_ea *)di;
7607 return it->oie_dirent->oied_off;
7611 * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
7612 * to load a directory entry at a time and stored it i inn,
7613 * in iterator's in-memory data structure.
7615 * \param di struct osd_it_ea, iterator's in memory structure
7617 * \retval +ve on success
7618 * \retval -ve on error
7620 static int osd_it_ea_load(const struct lu_env *env,
7621 const struct dt_it *di, __u64 hash)
7623 struct osd_it_ea *it = (struct osd_it_ea *)di;
7627 it->oie_file->f_pos = hash;
7629 rc = osd_ldiskfs_it_fill(env, di);
7640 * Index lookup function for interoperability mode (b11826).
7642 * \param key, key i.e. file name to be searched
7644 * \retval +ve, on success
7645 * \retval -ve, on error
7647 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7648 struct dt_rec *rec, const struct dt_key *key)
7650 struct osd_object *obj = osd_dt_obj(dt);
7655 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7656 LINVRNT(osd_invariant(obj));
7658 rc = osd_ea_lookup_rec(env, obj, rec, key);
7665 * Index and Iterator operations for interoperability
7666 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7668 static const struct dt_index_operations osd_index_ea_ops = {
7669 .dio_lookup = osd_index_ea_lookup,
7670 .dio_declare_insert = osd_index_declare_ea_insert,
7671 .dio_insert = osd_index_ea_insert,
7672 .dio_declare_delete = osd_index_declare_ea_delete,
7673 .dio_delete = osd_index_ea_delete,
7675 .init = osd_it_ea_init,
7676 .fini = osd_it_ea_fini,
7677 .get = osd_it_ea_get,
7678 .put = osd_it_ea_put,
7679 .next = osd_it_ea_next,
7680 .key = osd_it_ea_key,
7681 .key_size = osd_it_ea_key_size,
7682 .rec = osd_it_ea_rec,
7683 .rec_size = osd_it_ea_rec_size,
7684 .store = osd_it_ea_store,
7685 .load = osd_it_ea_load
7689 static void *osd_key_init(const struct lu_context *ctx,
7690 struct lu_context_key *key)
7692 struct osd_thread_info *info;
7694 OBD_ALLOC_PTR(info);
7696 return ERR_PTR(-ENOMEM);
7698 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7699 if (info->oti_it_ea_buf == NULL)
7702 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7704 info->oti_hlock = ldiskfs_htree_lock_alloc();
7705 if (info->oti_hlock == NULL)
7711 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7714 return ERR_PTR(-ENOMEM);
7717 static void osd_key_fini(const struct lu_context *ctx,
7718 struct lu_context_key *key, void *data)
7720 struct osd_thread_info *info = data;
7721 struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7722 struct osd_idmap_cache *idc = info->oti_ins_cache;
7724 if (info->oti_dio_pages) {
7726 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7727 struct page *page = info->oti_dio_pages[i];
7729 LASSERT(PagePrivate2(page));
7730 LASSERT(PageLocked(page));
7731 ClearPagePrivate2(page);
7736 OBD_FREE_PTR_ARRAY_LARGE(info->oti_dio_pages,
7737 PTLRPC_MAX_BRW_PAGES);
7740 if (info->oti_inode != NULL)
7742 if (info->oti_hlock != NULL)
7743 ldiskfs_htree_lock_free(info->oti_hlock);
7744 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7745 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7746 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7747 lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7748 lu_buf_free(&info->oti_big_buf);
7750 LASSERT(info->oti_ins_cache_size > 0);
7751 OBD_FREE_PTR_ARRAY_LARGE(idc, info->oti_ins_cache_size);
7752 info->oti_ins_cache = NULL;
7753 info->oti_ins_cache_size = 0;
7758 static void osd_key_exit(const struct lu_context *ctx,
7759 struct lu_context_key *key, void *data)
7761 struct osd_thread_info *info = data;
7763 LASSERT(info->oti_r_locks == 0);
7764 LASSERT(info->oti_w_locks == 0);
7765 LASSERT(info->oti_txns == 0);
7768 /* type constructor/destructor: osd_type_init, osd_type_fini */
7769 LU_TYPE_INIT_FINI(osd, &osd_key);
7771 struct lu_context_key osd_key = {
7772 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7773 .lct_init = osd_key_init,
7774 .lct_fini = osd_key_fini,
7775 .lct_exit = osd_key_exit
7779 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7780 const char *name, struct lu_device *next)
7782 struct osd_device *osd = osd_dev(d);
7784 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7785 sizeof(osd->od_svname))
7787 return osd_procfs_init(osd, name);
7790 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7792 struct seq_server_site *ss = osd_seq_site(osd);
7797 if (osd->od_is_ost || osd->od_cl_seq != NULL)
7800 if (unlikely(ss == NULL))
7803 OBD_ALLOC_PTR(osd->od_cl_seq);
7804 if (osd->od_cl_seq == NULL)
7807 seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7808 osd->od_svname, ss->ss_server_seq);
7810 if (ss->ss_node_id == 0) {
7812 * If the OSD on the sequence controller(MDT0), then allocate
7813 * sequence here, otherwise allocate sequence after connected
7814 * to MDT0 (see mdt_register_lwp_callback()).
7816 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7817 &osd->od_cl_seq->lcs_space, env);
7823 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7825 if (osd->od_cl_seq == NULL)
7828 seq_client_fini(osd->od_cl_seq);
7829 OBD_FREE_PTR(osd->od_cl_seq);
7830 osd->od_cl_seq = NULL;
7833 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7837 /* shutdown quota slave instance associated with the device */
7838 if (o->od_quota_slave_md != NULL) {
7839 struct qsd_instance *qsd = o->od_quota_slave_md;
7841 o->od_quota_slave_md = NULL;
7845 if (o->od_quota_slave_dt != NULL) {
7846 struct qsd_instance *qsd = o->od_quota_slave_dt;
7848 o->od_quota_slave_dt = NULL;
7852 osd_fid_fini(env, o);
7853 osd_scrub_cleanup(env, o);
7858 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7862 if (o->od_mnt != NULL) {
7863 shrink_dcache_sb(osd_sb(o));
7864 osd_sync(env, &o->od_dt_dev);
7865 wait_event(o->od_commit_cb_done,
7866 !atomic_read(&o->od_commit_cb_in_flight));
7875 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
7876 # ifndef LDISKFS_HAS_INCOMPAT_FEATURE
7877 /* Newer kernels provide the ldiskfs_set_feature_largedir() wrapper already,
7878 * which calls ldiskfs_update_dynamic_rev() to update ancient filesystems.
7879 * All ldiskfs filesystems are already v2, so it is a no-op and unnecessary.
7880 * This avoids maintaining patches to export this otherwise-useless function.
7882 void ldiskfs_update_dynamic_rev(struct super_block *sb)
7889 static int osd_mount(const struct lu_env *env,
7890 struct osd_device *o, struct lustre_cfg *cfg)
7892 const char *name = lustre_cfg_string(cfg, 0);
7893 const char *dev = lustre_cfg_string(cfg, 1);
7895 unsigned long page, s_flags = 0, lmd_flags = 0;
7896 struct page *__page;
7897 struct file_system_type *type;
7898 char *options = NULL;
7900 struct osd_thread_info *info = osd_oti_get(env);
7901 struct lu_fid *fid = &info->oti_fid;
7902 struct inode *inode;
7903 int rc = 0, force_over_1024tb = 0;
7907 if (o->od_mnt != NULL)
7910 if (strlen(dev) >= sizeof(o->od_mntdev))
7912 strcpy(o->od_mntdev, dev);
7914 str = lustre_cfg_buf(cfg, 2);
7915 sscanf(str, "%lu:%lu", &s_flags, &lmd_flags);
7917 opts = lustre_cfg_string(cfg, 3);
7919 if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7920 CERROR("%s: device %s extents feature is not guaranteed to "
7921 "work on big-endian systems. Use \"bigendian_extents\" "
7922 "mount option to override.\n", name, dev);
7926 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7927 if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7928 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");
7931 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7932 if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7933 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");
7936 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
7937 if (opts != NULL && strstr(opts, "force_over_512tb") != NULL) {
7938 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");
7942 if (opts != NULL && strstr(opts, "force_over_1024tb") != NULL)
7943 force_over_1024tb = 1;
7945 __page = alloc_page(GFP_KERNEL);
7947 GOTO(out, rc = -ENOMEM);
7948 page = (unsigned long)page_address(__page);
7949 options = (char *)page;
7952 /* strip out the options for back compatiblity */
7953 static const char * const sout[] = {
7960 /* strip out option we processed in osd */
7961 "bigendian_extents",
7965 "force_over_1024tb",
7969 strncat(options, opts, PAGE_SIZE);
7970 for (rc = 0, str = options; sout[rc]; ) {
7971 char *op = strstr(str, sout[rc]);
7978 if (op == options || *(op - 1) == ',') {
7979 str = op + strlen(sout[rc]);
7980 if (*str == ',' || *str == '\0') {
7981 *str == ',' ? str++ : str;
7982 memmove(op, str, strlen(str) + 1);
7985 for (str = op; *str != ',' && *str != '\0'; str++)
7989 strncat(options, "user_xattr,acl", PAGE_SIZE);
7992 /* Glom up mount options */
7993 if (*options != '\0')
7994 strncat(options, ",", PAGE_SIZE);
7995 strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7997 type = get_fs_type("ldiskfs");
7999 CERROR("%s: cannot find ldiskfs module\n", name);
8000 GOTO(out, rc = -ENODEV);
8003 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
8004 module_put(type->owner);
8006 if (IS_ERR(o->od_mnt)) {
8007 rc = PTR_ERR(o->od_mnt);
8009 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
8013 if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
8014 osd_sb(o)->s_blocksize_bits > 1024ULL << 40 &&
8015 force_over_1024tb == 0) {
8016 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",
8018 GOTO(out_mnt, rc = -EINVAL);
8021 if (lmd_flags & LMD_FLG_DEV_RDONLY) {
8022 LCONSOLE_WARN("%s: not support dev_rdonly on this device\n",
8025 GOTO(out_mnt, rc = -EOPNOTSUPP);
8028 if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
8029 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
8030 GOTO(out_mnt, rc = -EINVAL);
8033 if (ldiskfs_has_feature_fast_commit(o->od_mnt->mnt_sb)) {
8034 CERROR("%s: device %s is mounted with fast_commit that breaks recovery\n",
8036 GOTO(out_mnt, rc = -EOPNOTSUPP);
8039 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
8040 #ifdef LDISKFS_MOUNT_DIRDATA
8041 if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
8042 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
8043 else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
8044 CWARN("%s: device %s was upgraded from Lustre-1.x without "
8045 "enabling the dirdata feature. If you do not want to "
8046 "downgrade to Lustre-1.x again, you can enable it via "
8047 "'tune2fs -O dirdata device'\n", name, dev);
8049 /* enable large_dir on MDTs to avoid REMOTE_PARENT_DIR overflow,
8050 * and on very large OSTs to avoid object directory overflow */
8051 if (unlikely(!ldiskfs_has_feature_largedir(o->od_mnt->mnt_sb) &&
8052 !strstr(name, "MGS"))) {
8053 ldiskfs_set_feature_largedir(o->od_mnt->mnt_sb);
8054 LCONSOLE_INFO("%s: enabled 'large_dir' feature on device %s\n",
8058 inode = osd_sb(o)->s_root->d_inode;
8059 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
8060 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
8062 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
8066 if (lmd_flags & LMD_FLG_NOSCRUB)
8067 o->od_scrub.os_scrub.os_auto_scrub_interval = AS_NEVER;
8069 if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
8070 /* do not use pagecache with flash-backed storage */
8071 o->od_writethrough_cache = 0;
8072 o->od_read_cache = 0;
8083 __free_page(__page);
8088 static struct lu_device *osd_device_fini(const struct lu_env *env,
8089 struct lu_device *d)
8091 struct osd_device *o = osd_dev(d);
8095 osd_index_backup(env, o, false);
8096 osd_shutdown(env, o);
8098 if (o->od_oi_table != NULL)
8099 osd_oi_fini(osd_oti_get(env), o);
8100 if (o->od_extent_bytes_percpu)
8101 free_percpu(o->od_extent_bytes_percpu);
8102 osd_obj_map_fini(o);
8108 static int osd_device_init0(const struct lu_env *env,
8109 struct osd_device *o,
8110 struct lustre_cfg *cfg)
8112 struct lu_device *l = osd2lu_dev(o);
8113 struct osd_thread_info *info;
8116 bool restored = false;
8119 /* if the module was re-loaded, env can loose its keys */
8120 rc = lu_env_refill((struct lu_env *)env);
8123 info = osd_oti_get(env);
8126 l->ld_ops = &osd_lu_ops;
8127 o->od_dt_dev.dd_ops = &osd_dt_ops;
8129 spin_lock_init(&o->od_osfs_lock);
8130 mutex_init(&o->od_otable_mutex);
8131 INIT_LIST_HEAD(&o->od_orphan_list);
8132 INIT_LIST_HEAD(&o->od_index_backup_list);
8133 INIT_LIST_HEAD(&o->od_index_restore_list);
8134 spin_lock_init(&o->od_lock);
8135 o->od_index_backup_policy = LIBP_NONE;
8137 init_waitqueue_head(&o->od_commit_cb_done);
8139 o->od_read_cache = 1;
8140 o->od_writethrough_cache = 1;
8141 o->od_enable_projid_xattr = 0;
8142 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
8143 o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20;
8144 o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20;
8145 o->od_scrub.os_scrub.os_auto_scrub_interval = AS_DEFAULT;
8146 /* default fallocate to unwritten extents: LU-14326/LU-14333 */
8147 o->od_fallocate_zero_blocks = 0;
8149 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
8150 sizeof(o->od_svname));
8151 if (cplen >= sizeof(o->od_svname)) {
8156 o->od_index_backup_stop = 0;
8157 o->od_index = -1; /* -1 means index is invalid */
8158 rc = server_name2index(o->od_svname, &o->od_index, NULL);
8159 if (rc == LDD_F_SV_TYPE_OST)
8162 o->od_full_scrub_ratio = OFSR_DEFAULT;
8163 o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
8164 rc = osd_mount(env, o, cfg);
8168 /* Can only check block device after mount */
8169 o->od_nonrotational =
8170 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
8172 rc = osd_obj_map_init(env, o);
8176 rc = lu_site_init(&o->od_site, l);
8178 GOTO(out_compat, rc);
8179 o->od_site.ls_bottom_dev = l;
8181 rc = lu_site_init_finish(&o->od_site);
8185 opts = lustre_cfg_string(cfg, 3);
8186 if (opts && strstr(opts, "resetoi"))
8189 INIT_LIST_HEAD(&o->od_ios_list);
8191 rc = lprocfs_init_brw_stats(&o->od_brw_stats);
8193 GOTO(out_brw_stats, rc);
8195 /* setup scrub, including OI files initialization */
8197 rc = osd_scrub_setup(env, o, restored);
8200 GOTO(out_brw_stats, rc);
8202 rc = osd_procfs_init(o, o->od_svname);
8204 CERROR("%s: can't initialize procfs: rc = %d\n",
8206 GOTO(out_scrub, rc);
8209 LASSERT(l->ld_site->ls_linkage.next != NULL);
8210 LASSERT(l->ld_site->ls_linkage.prev != NULL);
8212 /* initialize quota slave instance */
8213 /* currently it's no need to prepare qsd_instance_md for OST */
8214 if (!o->od_is_ost) {
8215 o->od_quota_slave_md = qsd_init(env, o->od_svname,
8216 &o->od_dt_dev, o->od_proc_entry,
8218 if (IS_ERR(o->od_quota_slave_md)) {
8219 rc = PTR_ERR(o->od_quota_slave_md);
8220 o->od_quota_slave_md = NULL;
8221 GOTO(out_procfs, rc);
8225 o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
8226 o->od_proc_entry, false, true);
8228 if (IS_ERR(o->od_quota_slave_dt)) {
8229 if (o->od_quota_slave_md != NULL) {
8230 qsd_fini(env, o->od_quota_slave_md);
8231 o->od_quota_slave_md = NULL;
8234 rc = PTR_ERR(o->od_quota_slave_dt);
8235 o->od_quota_slave_dt = NULL;
8236 GOTO(out_procfs, rc);
8239 o->od_extent_bytes_percpu = alloc_percpu(unsigned int);
8240 if (!o->od_extent_bytes_percpu) {
8242 GOTO(out_procfs, rc);
8250 osd_scrub_cleanup(env, o);
8252 lprocfs_fini_brw_stats(&o->od_brw_stats);
8254 lu_site_fini(&o->od_site);
8256 osd_obj_map_fini(o);
8263 static struct lu_device *osd_device_alloc(const struct lu_env *env,
8264 struct lu_device_type *t,
8265 struct lustre_cfg *cfg)
8267 struct osd_device *o;
8272 return ERR_PTR(-ENOMEM);
8274 rc = dt_device_init(&o->od_dt_dev, t);
8277 * Because the ctx might be revived in dt_device_init,
8278 * refill the env here
8280 lu_env_refill((struct lu_env *)env);
8281 rc = osd_device_init0(env, o, cfg);
8283 dt_device_fini(&o->od_dt_dev);
8286 if (unlikely(rc != 0))
8289 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
8292 static struct lu_device *osd_device_free(const struct lu_env *env,
8293 struct lu_device *d)
8295 struct osd_device *o = osd_dev(d);
8299 /* XXX: make osd top device in order to release reference */
8300 d->ld_site->ls_top_dev = d;
8301 lu_site_purge(env, d->ld_site, -1);
8302 lu_site_print(env, d->ld_site, &d->ld_site->ls_obj_hash.nelems,
8303 D_ERROR, lu_cdebug_printer);
8304 lu_site_fini(&o->od_site);
8305 dt_device_fini(&o->od_dt_dev);
8310 static int osd_process_config(const struct lu_env *env,
8311 struct lu_device *d, struct lustre_cfg *cfg)
8313 struct osd_device *o = osd_dev(d);
8319 switch (cfg->lcfg_command) {
8321 rc = osd_mount(env, o, cfg);
8325 * For the case LCFG_PRE_CLEANUP is not called in advance,
8326 * that may happend if hit failure during mount process.
8328 osd_index_backup(env, o, false);
8329 lu_dev_del_linkage(d->ld_site, d);
8330 rc = osd_shutdown(env, o);
8333 LASSERT(&o->od_dt_dev);
8334 count = class_modify_config(cfg, PARAM_OSD,
8335 &o->od_dt_dev.dd_kobj);
8337 count = class_modify_config(cfg, PARAM_OST,
8338 &o->od_dt_dev.dd_kobj);
8339 rc = count > 0 ? 0 : count;
8341 case LCFG_PRE_CLEANUP:
8343 osd_index_backup(env, o,
8344 o->od_index_backup_policy != LIBP_NONE);
8354 static int osd_recovery_complete(const struct lu_env *env,
8355 struct lu_device *d)
8357 struct osd_device *osd = osd_dev(d);
8362 if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
8366 * start qsd instance on recovery completion, this notifies the quota
8367 * slave code that we are about to process new requests now
8369 rc = qsd_start(env, osd->od_quota_slave_dt);
8370 if (rc == 0 && osd->od_quota_slave_md != NULL)
8371 rc = qsd_start(env, osd->od_quota_slave_md);
8377 * we use exports to track all osd users
8379 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
8380 struct obd_device *obd, struct obd_uuid *cluuid,
8381 struct obd_connect_data *data, void *localdata)
8383 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8384 struct lustre_handle conn;
8389 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
8391 rc = class_connect(&conn, obd, cluuid);
8395 *exp = class_conn2export(&conn);
8397 spin_lock(&osd->od_osfs_lock);
8399 spin_unlock(&osd->od_osfs_lock);
8405 * once last export (we don't count self-export) disappeared
8406 * osd can be released
8408 static int osd_obd_disconnect(struct obd_export *exp)
8410 struct obd_device *obd = exp->exp_obd;
8411 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8412 int rc, release = 0;
8416 /* Only disconnect the underlying layers on the final disconnect. */
8417 spin_lock(&osd->od_osfs_lock);
8419 if (osd->od_connects == 0)
8421 spin_unlock(&osd->od_osfs_lock);
8423 rc = class_disconnect(exp); /* bz 9811 */
8425 if (rc == 0 && release)
8426 class_manual_cleanup(obd);
8430 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8431 struct lu_device *dev)
8433 struct osd_device *osd = osd_dev(dev);
8434 struct lr_server_data *lsd =
8435 &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8440 if (osd->od_quota_slave_md != NULL) {
8441 /* set up quota slave objects for inode */
8442 result = qsd_prepare(env, osd->od_quota_slave_md);
8447 if (osd->od_quota_slave_dt != NULL) {
8448 /* set up quota slave objects for block */
8449 result = qsd_prepare(env, osd->od_quota_slave_dt);
8455 if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8456 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8457 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8458 osd->od_index_in_idif = 1;
8460 osd->od_index_in_idif = 0;
8461 result = osd_register_proc_index_in_idif(osd);
8466 osd->od_index_in_idif = 1;
8470 result = osd_fid_init(env, osd);
8476 * Implementation of lu_device_operations::ldo_fid_alloc() for OSD
8480 * see include/lu_object.h for the details.
8482 static int osd_fid_alloc(const struct lu_env *env, struct lu_device *d,
8483 struct lu_fid *fid, struct lu_object *parent,
8484 const struct lu_name *name)
8486 struct osd_device *osd = osd_dev(d);
8488 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8491 static const struct lu_object_operations osd_lu_obj_ops = {
8492 .loo_object_init = osd_object_init,
8493 .loo_object_delete = osd_object_delete,
8494 .loo_object_release = osd_object_release,
8495 .loo_object_free = osd_object_free,
8496 .loo_object_print = osd_object_print,
8497 .loo_object_invariant = osd_object_invariant
8500 const struct lu_device_operations osd_lu_ops = {
8501 .ldo_object_alloc = osd_object_alloc,
8502 .ldo_process_config = osd_process_config,
8503 .ldo_recovery_complete = osd_recovery_complete,
8504 .ldo_prepare = osd_prepare,
8505 .ldo_fid_alloc = osd_fid_alloc,
8508 static const struct lu_device_type_operations osd_device_type_ops = {
8509 .ldto_init = osd_type_init,
8510 .ldto_fini = osd_type_fini,
8512 .ldto_start = osd_type_start,
8513 .ldto_stop = osd_type_stop,
8515 .ldto_device_alloc = osd_device_alloc,
8516 .ldto_device_free = osd_device_free,
8518 .ldto_device_init = osd_device_init,
8519 .ldto_device_fini = osd_device_fini
8522 static struct lu_device_type osd_device_type = {
8523 .ldt_tags = LU_DEVICE_DT,
8524 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
8525 .ldt_ops = &osd_device_type_ops,
8526 .ldt_ctx_tags = LCT_LOCAL,
8529 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8531 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8532 struct super_block *sb = osd_sb(osd);
8534 return (osd->od_mnt == NULL || sb->s_flags & SB_RDONLY);
8538 * lprocfs legacy support.
8540 static const struct obd_ops osd_obd_device_ops = {
8541 .o_owner = THIS_MODULE,
8542 .o_connect = osd_obd_connect,
8543 .o_disconnect = osd_obd_disconnect,
8544 .o_health_check = osd_health_check,
8547 static ssize_t track_declares_assert_show(struct kobject *kobj,
8548 struct attribute *attr,
8551 return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8554 static ssize_t track_declares_assert_store(struct kobject *kobj,
8555 struct attribute *attr,
8556 const char *buffer, size_t count)
8558 bool track_declares_assert;
8561 rc = kstrtobool(buffer, &track_declares_assert);
8565 ldiskfs_track_declares_assert = track_declares_assert;
8569 LUSTRE_RW_ATTR(track_declares_assert);
8571 static int __init osd_init(void)
8573 struct kobject *kobj;
8576 BUILD_BUG_ON(BH_DXLock >=
8577 sizeof(((struct buffer_head *)0)->b_state) * 8);
8578 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8579 /* please, try to keep osd_thread_info smaller than a page */
8580 BUILD_BUG_ON(sizeof(struct osd_thread_info) > PAGE_SIZE);
8585 rc = lu_kmem_init(ldiskfs_caches);
8589 rc = class_register_type(&osd_obd_device_ops, NULL, true,
8590 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8592 lu_kmem_fini(ldiskfs_caches);
8596 kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8598 rc = sysfs_create_file(kobj,
8599 &lustre_attr_track_declares_assert.attr);
8602 CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8610 static void __exit osd_exit(void)
8612 struct kobject *kobj;
8614 kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8616 sysfs_remove_file(kobj,
8617 &lustre_attr_track_declares_assert.attr);
8620 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8621 lu_kmem_fini(ldiskfs_caches);
8624 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8625 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8626 MODULE_VERSION(LUSTRE_VERSION_STRING);
8627 MODULE_LICENSE("GPL");
8629 module_init(osd_init);
8630 module_exit(osd_exit);