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);
126 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
127 const struct lu_buf *buf, const char *name, int fl,
128 struct thandle *handle);
130 int osd_trans_declare_op2rb[] = {
131 [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
132 [OSD_OT_PUNCH] = OSD_OT_MAX,
133 [OSD_OT_XATTR_SET] = OSD_OT_XATTR_SET,
134 [OSD_OT_CREATE] = OSD_OT_DESTROY,
135 [OSD_OT_DESTROY] = OSD_OT_CREATE,
136 [OSD_OT_REF_ADD] = OSD_OT_REF_DEL,
137 [OSD_OT_REF_DEL] = OSD_OT_REF_ADD,
138 [OSD_OT_WRITE] = OSD_OT_WRITE,
139 [OSD_OT_INSERT] = OSD_OT_DELETE,
140 [OSD_OT_DELETE] = OSD_OT_INSERT,
141 [OSD_OT_QUOTA] = OSD_OT_MAX,
144 static int osd_has_index(const struct osd_object *obj)
146 return obj->oo_dt.do_index_ops != NULL;
149 static int osd_object_invariant(const struct lu_object *l)
151 return osd_invariant(osd_obj(l));
155 * Concurrency: doesn't matter
157 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
159 struct osd_thread_info *oti = osd_oti_get(env);
161 return oti->oti_w_locks > 0 && o->oo_owner == env;
165 * Concurrency: doesn't access mutable data
167 static int osd_root_get(const struct lu_env *env,
168 struct dt_device *dev, struct lu_fid *f)
170 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
175 * the following set of functions are used to maintain per-thread
176 * cache of FID->ino mapping. this mechanism is needed to resolve
177 * FID to inode at dt_insert() which in turn stores ino in the
178 * directory entries to keep ldiskfs compatible with ext[34].
179 * due to locking-originated restrictions we can't lookup ino
180 * using LU cache (deadlock is possible). lookup using OI is quite
181 * expensive. so instead we maintain this cache and methods like
182 * dt_create() fill it. so in the majority of cases dt_insert() is
183 * able to find needed mapping in lockless manner.
185 static struct osd_idmap_cache *
186 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
187 const struct lu_fid *fid)
189 struct osd_thread_info *oti = osd_oti_get(env);
190 struct osd_idmap_cache *idc = oti->oti_ins_cache;
193 for (i = 0; i < oti->oti_ins_cache_used; i++) {
194 if (!lu_fid_eq(&idc[i].oic_fid, fid))
196 if (idc[i].oic_dev != osd)
205 static struct osd_idmap_cache *
206 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
207 const struct lu_fid *fid)
209 struct osd_thread_info *oti = osd_oti_get(env);
210 struct osd_idmap_cache *idc;
213 if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
214 i = oti->oti_ins_cache_size * 2;
216 i = OSD_INS_CACHE_SIZE;
217 OBD_ALLOC_PTR_ARRAY_LARGE(idc, i);
219 return ERR_PTR(-ENOMEM);
220 if (oti->oti_ins_cache != NULL) {
221 memcpy(idc, oti->oti_ins_cache,
222 oti->oti_ins_cache_used * sizeof(*idc));
223 OBD_FREE_PTR_ARRAY_LARGE(oti->oti_ins_cache,
224 oti->oti_ins_cache_used);
226 oti->oti_ins_cache = idc;
227 oti->oti_ins_cache_size = i;
230 idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
233 idc->oic_lid.oii_ino = 0;
234 idc->oic_lid.oii_gen = 0;
241 * lookup mapping for the given fid in the cache, initialize a
242 * new one if not found. the initialization checks whether the
243 * object is local or remote. for local objects, OI is used to
244 * learn ino/generation. the function is used when the caller
245 * has no information about the object, e.g. at dt_insert().
247 static struct osd_idmap_cache *
248 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
249 const struct lu_fid *fid)
251 struct osd_idmap_cache *idc;
254 idc = osd_idc_find(env, osd, fid);
255 LASSERT(!IS_ERR(idc));
259 CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
260 osd->od_svname, PFID(fid));
262 /* new mapping is needed */
263 idc = osd_idc_add(env, osd, fid);
265 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
266 osd->od_svname, PFID(fid), PTR_ERR(idc));
271 rc = osd_remote_fid(env, osd, fid);
272 if (unlikely(rc < 0))
276 /* the object is local, lookup in OI */
277 /* XXX: probably cheaper to lookup in LU first? */
278 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
280 if (unlikely(rc < 0)) {
281 CERROR("can't lookup: rc = %d\n", rc);
285 /* the object is remote */
292 static void osd_idc_dump_lma(const struct lu_env *env,
293 struct osd_device *osd,
297 struct osd_thread_info *info = osd_oti_get(env);
298 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
299 const struct lu_fid *fid;
300 struct osd_inode_id lid;
304 inode = osd_ldiskfs_iget(osd_sb(osd), ino);
306 CERROR("%s: can't get inode %lu: rc = %d\n",
307 osd->od_svname, ino, (int)PTR_ERR(inode));
310 if (is_bad_inode(inode)) {
311 CERROR("%s: bad inode %lu\n", osd->od_svname, ino);
314 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
316 CERROR("%s: can't get LMA for %lu: rc = %d\n",
317 osd->od_svname, ino, rc);
320 fid = &loa->loa_lma.lma_self_fid;
321 LCONSOLE(D_INFO, "%s: "DFID" in inode %lu/%u\n", osd->od_svname,
322 PFID(fid), ino, (unsigned)inode->i_generation);
325 rc = osd_oi_lookup(osd_oti_get(env), osd, fid, &lid, 0);
327 CERROR("%s: can't lookup "DFID": rc = %d\n",
328 osd->od_svname, PFID(fid), rc);
331 LCONSOLE(D_INFO, "%s: "DFID" maps to %u/%u\n", osd->od_svname,
332 PFID(fid), lid.oii_ino, lid.oii_gen);
337 static void osd_idc_dump_debug(const struct lu_env *env,
338 struct osd_device *osd,
339 const struct lu_fid *fid,
343 struct osd_inode_id lid;
347 rc = osd_oi_lookup(osd_oti_get(env), osd, fid, &lid, 0);
349 LCONSOLE(D_INFO, "%s: "DFID" maps to %u/%u\n",
350 osd->od_svname, PFID(fid), lid.oii_ino, lid.oii_gen);
351 osd_idc_dump_lma(env, osd, lid.oii_ino, false);
353 CERROR("%s: can't lookup "DFID": rc = %d\n",
354 osd->od_svname, PFID(fid), rc);
357 osd_idc_dump_lma(env, osd, ino1, true);
359 osd_idc_dump_lma(env, osd, ino2, true);
363 * lookup mapping for given FID and fill it from the given object.
364 * the object is lolcal by definition.
366 static int osd_idc_find_and_init(const struct lu_env *env,
367 struct osd_device *osd,
368 struct osd_object *obj)
370 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
371 struct osd_idmap_cache *idc;
373 idc = osd_idc_find(env, osd, fid);
374 LASSERT(!IS_ERR(idc));
376 if (obj->oo_inode == NULL)
378 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
379 if (idc->oic_lid.oii_ino) {
380 osd_idc_dump_debug(env, osd, fid,
381 idc->oic_lid.oii_ino,
382 obj->oo_inode->i_ino);
385 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
386 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
391 CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
392 osd->od_svname, PFID(fid));
394 /* new mapping is needed */
395 idc = osd_idc_add(env, osd, fid);
397 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
398 osd->od_svname, PFID(fid), PTR_ERR(idc));
402 if (obj->oo_inode != NULL) {
403 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
404 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
410 * OSD object methods.
414 * Concurrency: no concurrent access is possible that early in object
417 static struct lu_object *osd_object_alloc(const struct lu_env *env,
418 const struct lu_object_header *hdr,
421 struct osd_object *mo;
426 struct lu_object_header *h;
427 struct osd_device *o = osd_dev(d);
429 l = &mo->oo_dt.do_lu;
430 if (unlikely(o->od_in_init)) {
437 lu_object_header_init(h);
438 lu_object_init(l, h, d);
439 lu_object_add_top(h, l);
442 dt_object_init(&mo->oo_dt, NULL, d);
443 mo->oo_header = NULL;
446 mo->oo_dt.do_ops = &osd_obj_ops;
447 l->lo_ops = &osd_lu_obj_ops;
448 init_rwsem(&mo->oo_sem);
449 init_rwsem(&mo->oo_ext_idx_sem);
450 spin_lock_init(&mo->oo_guard);
451 INIT_LIST_HEAD(&mo->oo_xattr_list);
457 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
458 struct dentry *dentry, struct lustre_ost_attrs *loa)
462 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
463 (void *)loa, sizeof(*loa));
465 struct lustre_mdt_attrs *lma = &loa->loa_lma;
467 if (rc < sizeof(*lma))
471 lustre_loa_swab(loa, true);
472 /* Check LMA compatibility */
473 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
475 CWARN("%s: unsupported incompat LMA feature(s) %#x for fid = "DFID", ino = %lu: rc = %d\n",
477 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
478 PFID(&lma->lma_self_fid), inode->i_ino, rc);
480 } else if (rc == 0) {
488 * retrieve object from backend ext fs.
490 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
491 struct osd_inode_id *id)
494 struct inode *inode = NULL;
497 * if we look for an inode withing a running
498 * transaction, then we risk to deadlock
499 * osd_dirent_check_repair() breaks this
501 /* LASSERT(current->journal_info == NULL); */
503 inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
505 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
506 id->oii_ino, PTR_ERR(inode));
507 } else if (id->oii_gen != OSD_OII_NOGEN &&
508 inode->i_generation != id->oii_gen) {
509 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
510 "i_generation = %u\n",
511 id->oii_ino, id->oii_gen, inode->i_generation);
513 inode = ERR_PTR(-ESTALE);
514 } else if (inode->i_nlink == 0) {
516 * due to parallel readdir and unlink,
517 * we can have dead inode here.
519 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
521 inode = ERR_PTR(-ESTALE);
522 } else if (is_bad_inode(inode)) {
523 CWARN("%s: bad inode: ino = %u: rc = %d\n",
524 osd_dev2name(dev), id->oii_ino, -ENOENT);
526 inode = ERR_PTR(-ENOENT);
527 } else if (osd_is_ea_inode(inode)) {
529 * EA inode is internal ldiskfs object, should don't visible
532 CDEBUG(D_INODE, "EA inode: ino = %u\n", id->oii_ino);
534 inode = ERR_PTR(-ENOENT);
535 } else if ((rc = osd_attach_jinode(inode))) {
539 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
540 if (id->oii_gen == OSD_OII_NOGEN)
541 osd_id_gen(id, inode->i_ino, inode->i_generation);
544 * Do not update file c/mtime in ldiskfs.
545 * NB: we don't have any lock to protect this because we don't
546 * have reference on osd_object now, but contention with
547 * another lookup + attr_set can't happen in the tiny window
548 * between if (...) and set S_NOCMTIME.
550 if (!(inode->i_flags & S_NOCMTIME))
551 inode->i_flags |= S_NOCMTIME;
556 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
557 handle_t *handle, struct dentry *child,
558 struct inode *inode, struct htree_lock *hlock)
562 rc = __ldiskfs_add_entry(handle, child, inode, hlock);
563 if (rc == -ENOBUFS || rc == -ENOSPC) {
564 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
565 struct inode *parent = child->d_parent->d_inode;
566 struct lu_fid *fid = NULL;
567 char fidstr[FID_LEN + 1] = "unknown";
569 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
571 fid = &loa->loa_lma.lma_self_fid;
572 } else if (rc2 == -ENODATA) {
573 if (unlikely(is_root_inode(parent))) {
574 fid = &info->oti_fid3;
575 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
576 } else if (!osd->od_is_ost && osd->od_index == 0) {
577 fid = &info->oti_fid3;
578 lu_igif_build(fid, parent->i_ino,
579 parent->i_generation);
584 snprintf(fidstr, sizeof(fidstr), DFID, PFID(fid));
586 /* below message is checked in sanity.sh test_129 */
588 CWARN("%s: directory (inode: %lu, FID: %s) has reached max size limit\n",
589 osd_name(osd), parent->i_ino, fidstr);
591 rc = 0; /* ignore such error now */
592 CWARN("%s: directory (inode: %lu, FID: %s) is approaching max size limit\n",
593 osd_name(osd), parent->i_ino, fidstr);
603 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
604 struct osd_inode_id *id, struct lu_fid *fid)
606 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
610 inode = osd_iget(info, dev, id);
614 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
616 *fid = loa->loa_lma.lma_self_fid;
617 } else if (rc == -ENODATA) {
618 if (unlikely(is_root_inode(inode)))
619 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
621 lu_igif_build(fid, inode->i_ino, inode->i_generation);
629 static struct inode *osd_iget_check(struct osd_thread_info *info,
630 struct osd_device *dev,
631 const struct lu_fid *fid,
632 struct osd_inode_id *id,
641 * The cached OI mapping is trustable. If we cannot locate the inode
642 * via the cached OI mapping, then return the failure to the caller
643 * directly without further OI checking.
647 inode = osd_iget(info, dev, id);
650 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
653 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
654 PFID(fid), id->oii_ino, rc);
660 __u32 saved_ino = id->oii_ino;
661 __u32 saved_gen = id->oii_gen;
664 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
666 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
668 * XXX: There are four possible cases:
670 * Backup/restore caused the OI invalid.
672 * Someone unlinked the object but NOT removed
673 * the OI mapping, such as mount target device
674 * as ldiskfs, and modify something directly.
676 * Someone just removed the object between the
677 * former oi_lookup and the iget. It is normal.
678 * 4. Other failure cases.
680 * Generally, when the device is mounted, it will
681 * auto check whether the system is restored from
682 * file-level backup or not. We trust such detect
683 * to distinguish the 1st case from the 2nd case:
684 * if the OI files are consistent but may contain
685 * stale OI mappings because of case 2, if iget()
686 * returns -ENOENT or -ESTALE, then it should be
691 * If the OI mapping was in OI file before the
692 * osd_iget_check(), but now, it is disappear,
693 * then it must be removed by race. That is a
699 * It is the OI scrub updated the OI mapping by race.
700 * The new OI mapping must be valid.
702 if (saved_ino != id->oii_ino ||
703 (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
712 if (dev->od_scrub.os_scrub.os_file.sf_flags &
715 * It still can be the case 2, but we cannot
716 * distinguish it from the case 1. So return
717 * -EREMCHG to block current operation until
718 * OI scrub rebuilt the OI mappings.
727 if (inode->i_generation == id->oii_gen)
747 * \retval +v: new filter_fid does not contain self-fid
748 * \retval 0: filter_fid_18_23, contains self-fid
749 * \retval -v: other failure cases
751 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
752 struct dentry *dentry, struct lu_fid *fid)
754 struct filter_fid *ff = &info->oti_ff;
755 struct ost_id *ostid = &info->oti_ostid;
758 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
759 if (rc == sizeof(struct filter_fid_18_23)) {
760 struct filter_fid_18_23 *ff_old = (void *)ff;
762 ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq));
763 rc = ostid_set_id(ostid, le64_to_cpu(ff_old->ff_objid));
765 * XXX: use 0 as the index for compatibility, the caller will
766 * handle index related issues when necessary.
769 ostid_to_fid(fid, ostid, 0);
770 } else if (rc >= (int)sizeof(struct filter_fid_24_29)) {
772 } else if (rc >= 0) {
779 static int osd_lma_self_repair(struct osd_thread_info *info,
780 struct osd_device *osd, struct inode *inode,
781 const struct lu_fid *fid, __u32 compat)
786 LASSERT(current->journal_info == NULL);
788 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
789 osd_dto_credits_noquota[DTO_XATTR_SET]);
792 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
797 rc = osd_ea_fid_set(info, inode, fid, compat, 0);
799 CWARN("%s: cannot self repair the LMA: rc = %d\n",
801 ldiskfs_journal_stop(jh);
805 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
807 struct osd_thread_info *info = osd_oti_get(env);
808 struct osd_device *osd = osd_obj2dev(obj);
809 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
810 struct lustre_mdt_attrs *lma = &loa->loa_lma;
811 struct inode *inode = obj->oo_inode;
812 struct dentry *dentry = &info->oti_obj_dentry;
813 struct lu_fid *fid = NULL;
814 const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
819 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
820 (void *)loa, sizeof(*loa));
821 if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
822 fid = &lma->lma_self_fid;
823 rc = osd_get_idif(info, inode, dentry, fid);
824 if (rc > 0 || (rc == -ENODATA && osd->od_index_in_idif)) {
826 * For the given OST-object, if it has neither LMA nor
827 * FID in XATTR_NAME_FID, then the given FID (which is
828 * contained in the @obj, from client RPC for locating
829 * the OST-object) is trusted. We use it to generate
832 osd_lma_self_repair(info, osd, inode, rfid,
843 lustre_lma_swab(lma);
844 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
845 (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
846 S_ISREG(inode->i_mode)))) {
847 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
848 "fid = "DFID", ino = %lu\n", osd_name(osd),
849 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
850 PFID(rfid), inode->i_ino);
853 fid = &lma->lma_self_fid;
854 if (lma->lma_compat & LMAC_STRIPE_INFO &&
856 obj->oo_pfid_in_lma = 1;
857 if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
859 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
863 if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
864 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
865 struct ost_id *oi = &info->oti_ostid;
866 struct lu_fid *fid1 = &info->oti_fid3;
867 __u32 idx = fid_idif_ost_idx(rfid);
870 * For old IDIF, the OST index is not part of the IDIF,
871 * Means that different OSTs may have the same IDIFs.
872 * Under such case, we need to make some compatible
873 * check to make sure to trigger OI scrub properly.
875 if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
876 /* Given @rfid is new, LMA is old. */
877 fid_to_ostid(fid, oi);
878 ostid_to_fid(fid1, oi, idx);
879 if (lu_fid_eq(fid1, rfid)) {
880 if (osd->od_index_in_idif)
881 osd_lma_self_repair(info, osd,
895 struct osd_check_lmv_buf {
896 /* please keep it as first member */
897 struct dir_context ctx;
898 struct osd_thread_info *oclb_info;
899 struct osd_device *oclb_dev;
905 * It is called internally by ->iterate*() to filter out the
906 * local slave object's FID of the striped directory.
908 * \retval 1 found the local slave's FID
909 * \retval 0 continue to check next item
910 * \retval -ve for failure
912 #ifdef HAVE_FILLDIR_USE_CTX
913 static int osd_stripe_dir_filldir(struct dir_context *buf,
915 static int osd_stripe_dir_filldir(void *buf,
917 const char *name, int namelen,
918 loff_t offset, __u64 ino, unsigned int d_type)
920 struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
921 struct osd_thread_info *oti = oclb->oclb_info;
922 struct lu_fid *fid = &oti->oti_fid3;
923 struct osd_inode_id *id = &oti->oti_id3;
924 struct osd_device *dev = oclb->oclb_dev;
933 sscanf(name + 1, SFID, RFID(fid));
934 if (!fid_is_sane(fid))
937 if (osd_remote_fid(oti->oti_env, dev, fid))
940 osd_id_gen(id, ino, OSD_OII_NOGEN);
941 inode = osd_iget(oti, dev, id);
943 return PTR_ERR(inode);
946 osd_add_oi_cache(oti, dev, id, fid);
947 osd_scrub_oi_insert(dev, fid, id, true);
948 oclb->oclb_found = true;
954 * When lookup item under striped directory, we need to locate the master
955 * MDT-object of the striped directory firstly, then the client will send
956 * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
957 * and the item's name. If the system is restored from MDT file level backup,
958 * then before the OI scrub completely built the OI files, the OI mappings of
959 * the master MDT-object and slave MDT-object may be invalid. Usually, it is
960 * not a problem for the master MDT-object. Because when locate the master
961 * MDT-object, we will do name based lookup (for the striped directory itself)
962 * firstly, during such process we can setup the correct OI mapping for the
963 * master MDT-object. But it will be trouble for the slave MDT-object. Because
964 * the client will not trigger name based lookup on the MDT to locate the slave
965 * MDT-object before locating item under the striped directory, then when
966 * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
967 * is invalid and does not know what the right OI mapping is, then the MDT has
968 * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
969 * the OI file, related OI mapping is unknown yet, please try again later. And
970 * then client will re-try the RPC again and again until related OI mapping has
971 * been updated. That is quite inefficient.
973 * To resolve above trouble, we will handle it as the following two cases:
975 * 1) The slave MDT-object and the master MDT-object are on different MDTs.
976 * It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
977 * is linked under /REMOTE_PARENT_DIR with the name of its FID string.
978 * We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
979 * directly. Please check osd_fid_lookup().
981 * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
982 * Under such case, during lookup the master MDT-object, we will lookup the
983 * slave MDT-object via readdir against the master MDT-object, because the
984 * slave MDT-objects information are stored as sub-directories with the name
985 * "${FID}:${index}". Then when find the local slave MDT-object, its OI
986 * mapping will be recorded. Then subsequent osd_fid_lookup() will know
987 * the correct OI mapping for the slave MDT-object.
989 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
992 struct lu_buf *buf = &oti->oti_big_buf;
994 struct lmv_mds_md_v1 *lmv1;
995 struct osd_check_lmv_buf oclb = {
996 .ctx.actor = osd_stripe_dir_filldir,
1004 /* We should use the VFS layer to create a real dentry. */
1005 oti->oti_obj_dentry.d_inode = inode;
1006 oti->oti_obj_dentry.d_sb = inode->i_sb;
1008 filp = alloc_file_pseudo(inode, dev->od_mnt, "/", O_NOATIME,
1013 filp->f_mode |= FMODE_64BITHASH;
1017 rc = __osd_xattr_get(inode, filp->f_path.dentry, XATTR_NAME_LMV,
1018 buf->lb_buf, buf->lb_len);
1019 if (rc == -ERANGE) {
1020 rc = __osd_xattr_get(inode, filp->f_path.dentry,
1021 XATTR_NAME_LMV, NULL, 0);
1023 lu_buf_realloc(buf, rc);
1024 if (buf->lb_buf == NULL)
1025 GOTO(out, rc = -ENOMEM);
1031 if (unlikely(rc == 0 || rc == -ENODATA))
1037 if (unlikely(buf->lb_buf == NULL)) {
1038 lu_buf_realloc(buf, rc);
1039 if (buf->lb_buf == NULL)
1040 GOTO(out, rc = -ENOMEM);
1046 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1050 oclb.oclb_items = 0;
1051 rc = iterate_dir(filp, &oclb.ctx);
1052 } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1053 filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1058 "%s: cannot check LMV, ino = %lu/%u: rc = %d\n",
1059 osd_ino2name(inode), inode->i_ino, inode->i_generation,
1068 * Is object in scrub inconsistent/stale list.
1070 * \a scrub has two lists, os_inconsistent_items contains mappings to fix, while
1071 * os_stale_items contains mappings failed to fix.
1073 static bool fid_in_scrub_list(struct lustre_scrub *scrub,
1074 const struct list_head *list,
1075 const struct lu_fid *fid)
1077 struct osd_inconsistent_item *oii;
1079 if (list_empty(list))
1082 spin_lock(&scrub->os_lock);
1083 list_for_each_entry(oii, list, oii_list) {
1084 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
1085 spin_unlock(&scrub->os_lock);
1089 spin_unlock(&scrub->os_lock);
1094 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1095 const struct lu_fid *fid,
1096 const struct lu_object_conf *conf)
1098 struct osd_thread_info *info;
1099 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1100 struct osd_device *dev;
1101 struct osd_idmap_cache *oic;
1102 struct osd_inode_id *id;
1103 struct inode *inode = NULL;
1104 struct lustre_scrub *scrub;
1105 struct scrub_file *sf;
1106 __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1111 bool remote = false;
1112 bool trusted = true;
1113 bool updated = false;
1114 bool checked = false;
1119 LINVRNT(osd_invariant(obj));
1120 LASSERT(obj->oo_inode == NULL);
1122 if (fid_is_sane(fid) == 0) {
1123 CERROR("%s: invalid FID "DFID"\n", ldev->ld_obd->obd_name,
1129 dev = osd_dev(ldev);
1130 scrub = &dev->od_scrub.os_scrub;
1131 sf = &scrub->os_file;
1132 info = osd_oti_get(env);
1134 oic = &info->oti_cache;
1136 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1140 * For the object is created as locking anchor, or for the object to
1141 * be created on disk. No need to osd_oi_lookup() at here because FID
1142 * shouldn't never be re-used, if it's really a duplicate FID from
1143 * unexpected reason, we should be able to detect it later by calling
1144 * do_create->osd_oi_insert().
1146 if (conf && conf->loc_flags & LOC_F_NEW)
1149 /* Search order: 1. per-thread cache. */
1150 if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1155 /* Search order: 2. OI scrub pending list. */
1157 memset(id, 0, sizeof(struct osd_inode_id));
1158 if (fid_in_scrub_list(scrub, &scrub->os_inconsistent_items, fid) &&
1160 RETURN(-EINPROGRESS);
1162 stale = fid_in_scrub_list(scrub, &scrub->os_stale_items, fid);
1163 if (stale && OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_STALE))
1167 * The OI mapping in the OI file can be updated by the OI scrub
1168 * when we locate the inode via FID. So it may be not trustable.
1172 /* Search order: 3. OI files. */
1173 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1174 if (result == -ENOENT) {
1175 if (!fid_is_norm(fid) ||
1176 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1177 !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1179 GOTO(out, result = 0);
1184 /* -ESTALE is returned if inode of OST object doesn't exist */
1185 if (result == -ESTALE &&
1186 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1187 GOTO(out, result = 0);
1194 obj->oo_inode = NULL;
1195 /* for later passes through checks, not true on first pass */
1196 if (!IS_ERR_OR_NULL(inode))
1199 inode = osd_iget_check(info, dev, fid, id, trusted);
1200 if (!IS_ERR(inode)) {
1201 obj->oo_inode = inode;
1209 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_STALE))
1212 result = PTR_ERR(inode);
1213 if (result == -ENOENT || result == -ESTALE)
1214 GOTO(out, result = 0);
1216 if (result != -EREMCHG)
1220 /* don't trigger repeatedly for stale mapping */
1222 GOTO(out, result = -ESTALE);
1225 * We still have chance to get the valid inode: for the
1226 * object which is referenced by remote name entry, the
1227 * object on the local MDT will be linked under the dir
1228 * of "/REMOTE_PARENT_DIR" with its FID string as name.
1230 * We do not know whether the object for the given FID
1231 * is referenced by some remote name entry or not, and
1232 * especially for DNE II, a multiple-linked object may
1233 * have many name entries reside on many MDTs.
1235 * To simplify the operation, OSD will not distinguish
1236 * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1237 * only happened for the RPC from other MDT during the
1238 * OI scrub, or for the client side RPC with FID only,
1239 * such as FID to path, or from old connected client.
1242 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1246 flags |= SS_AUTO_PARTIAL;
1247 flags &= ~SS_AUTO_FULL;
1252 if (scrub->os_running) {
1253 if (scrub->os_partial_scan && !scrub->os_in_join)
1256 if (IS_ERR_OR_NULL(inode) || result) {
1257 osd_scrub_oi_insert(dev, fid, id, result == -ENOENT);
1258 GOTO(out, result = -EINPROGRESS);
1262 LASSERT(obj->oo_inode == inode);
1264 osd_scrub_oi_insert(dev, fid, id, true);
1268 if (dev->od_scrub.os_scrub.os_auto_scrub_interval == AS_NEVER) {
1270 GOTO(out, result = -EREMCHG);
1273 LASSERT(obj->oo_inode == inode);
1275 osd_add_oi_cache(info, dev, id, fid);
1280 if (IS_ERR_OR_NULL(inode) || result)
1281 osd_scrub_oi_insert(dev, fid, id, result == -ENOENT);
1283 rc1 = osd_scrub_start(env, dev, flags);
1284 CDEBUG_LIMIT(D_LFSCK | D_CONSOLE | D_WARNING,
1285 "%s: trigger OI scrub by RPC for "DFID"/%u with flags %#x: rc = %d\n",
1286 osd_name(dev), PFID(fid), id->oii_ino, flags, rc1);
1287 if (rc1 && rc1 != -EALREADY)
1288 GOTO(out, result = -EREMCHG);
1290 if (IS_ERR_OR_NULL(inode) || result)
1291 GOTO(out, result = -EINPROGRESS);
1294 LASSERT(obj->oo_inode == inode);
1299 if (unlikely(obj->oo_header))
1302 result = osd_check_lma(env, obj);
1306 LASSERTF(id->oii_ino == inode->i_ino &&
1307 id->oii_gen == inode->i_generation,
1308 "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1309 PFID(fid), id->oii_ino, id->oii_gen,
1310 inode->i_ino, inode->i_generation);
1312 saved_ino = inode->i_ino;
1313 saved_gen = inode->i_generation;
1315 if (unlikely(result == -ENODATA)) {
1317 * If the OI scrub updated the OI mapping by race, it
1318 * must be valid. Trust the inode that has no LMA EA.
1323 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1326 * The OI mapping is still there, the inode is still
1327 * valid. It is just becaues the inode has no LMA EA.
1329 if (saved_ino == id->oii_ino &&
1330 saved_gen == id->oii_gen)
1334 * It is the OI scrub updated the OI mapping by race.
1335 * The new OI mapping must be valid.
1343 * "result == -ENOENT" means that the OI mappinghas been
1344 * removed by race, so the inode belongs to other object.
1346 * Others error can be returned directly.
1348 if (result == -ENOENT) {
1351 obj->oo_inode = NULL;
1356 if (result != -EREMCHG)
1362 * if two OST objects map to the same inode, and inode mode is
1363 * (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666), which means it's
1364 * reserved by precreate, and not written yet, in this case, don't
1365 * set inode for the object whose FID mismatch, so that it can create
1366 * inode and not block precreate.
1368 if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) &&
1369 inode->i_mode == (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666)) {
1370 obj->oo_inode = NULL;
1371 GOTO(out, result = 0);
1374 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1376 * "result == -ENOENT" means the cached OI mapping has been removed
1377 * from the OI file by race, above inode belongs to other object.
1379 if (result == -ENOENT) {
1382 obj->oo_inode = NULL;
1383 GOTO(out, result = 0);
1389 if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1391 osd_scrub_refresh_mapping(info, dev, fid, id, DTO_INDEX_DELETE,
1397 * It is the OI scrub updated the OI mapping by race.
1398 * The new OI mapping must be valid.
1406 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1407 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1409 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1411 if (lma->lma_compat & LMAC_STRIPE_INFO &&
1413 obj->oo_pfid_in_lma = 1;
1414 if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1416 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1417 } else if (result != -ENODATA) {
1422 obj->oo_compat_dot_created = 1;
1423 obj->oo_compat_dotdot_created = 1;
1425 if (S_ISDIR(inode->i_mode) &&
1426 (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1427 osd_check_lmv(info, dev, inode);
1429 result = osd_attach_jinode(inode);
1434 GOTO(out, result = 0);
1436 LASSERT(!obj->oo_hl_head);
1437 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1439 GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1442 if (!result && stale)
1443 osd_scrub_oi_resurrect(scrub, fid);
1445 if (result || !obj->oo_inode) {
1446 if (!IS_ERR_OR_NULL(inode))
1449 obj->oo_inode = NULL;
1451 fid_zero(&oic->oic_fid);
1454 LINVRNT(osd_invariant(obj));
1459 * Concurrency: shouldn't matter.
1461 static void osd_object_init0(struct osd_object *obj)
1463 LASSERT(obj->oo_inode != NULL);
1464 obj->oo_dt.do_body_ops = &osd_body_ops;
1465 obj->oo_dt.do_lu.lo_header->loh_attr |=
1466 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1470 * Concurrency: no concurrent access is possible that early in object
1473 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1474 const struct lu_object_conf *conf)
1476 struct osd_object *obj = osd_obj(l);
1479 LINVRNT(osd_invariant(obj));
1481 if (OBD_FAIL_PRECHECK(OBD_FAIL_MDS_LLOG_UMOUNT_RACE) &&
1482 cfs_fail_val == 2) {
1483 struct osd_thread_info *info = osd_oti_get(env);
1484 struct osd_idmap_cache *oic = &info->oti_cache;
1485 /* invalidate thread cache */
1486 memset(&oic->oic_fid, 0, sizeof(oic->oic_fid));
1488 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1489 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1490 l->lo_header->loh_attr |= LOHA_EXISTS;
1494 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1495 obj->oo_dt.do_body_ops = &osd_body_ops_new;
1496 if (result == 0 && obj->oo_inode != NULL) {
1497 struct osd_thread_info *oti = osd_oti_get(env);
1498 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1500 osd_object_init0(obj);
1501 if (unlikely(obj->oo_header))
1504 result = osd_get_lma(oti, obj->oo_inode,
1505 &oti->oti_obj_dentry, loa);
1508 * Convert LMAI flags to lustre LMA flags
1509 * and cache it to oo_lma_flags
1512 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1513 } else if (result == -ENODATA) {
1517 obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
1519 LINVRNT(osd_invariant(obj));
1524 * The first part of oxe_buf is xattr name, and is '\0' terminated.
1525 * The left part is for value, binary mode.
1527 struct osd_xattr_entry {
1528 struct list_head oxe_list;
1532 struct rcu_head oxe_rcu;
1536 static int osd_oxc_get(struct osd_object *obj, const char *name,
1539 struct osd_xattr_entry *tmp;
1540 struct osd_xattr_entry *oxe = NULL;
1541 size_t namelen = strlen(name);
1545 list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1546 if (namelen == tmp->oxe_namelen &&
1547 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1554 GOTO(out, rc = -ENOENT);
1556 if (!oxe->oxe_exist)
1557 GOTO(out, rc = -ENODATA);
1560 rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1563 if (buf->lb_buf == NULL)
1566 if (buf->lb_len < rc)
1567 GOTO(out, rc = -ERANGE);
1569 memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1576 static void osd_oxc_free(struct rcu_head *head)
1578 struct osd_xattr_entry *oxe;
1580 oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1581 OBD_FREE(oxe, oxe->oxe_len);
1584 static void osd_oxc_add(struct osd_object *obj, const char *name,
1585 const char *buf, int buflen)
1587 struct osd_xattr_entry *oxe;
1588 struct osd_xattr_entry *old = NULL;
1589 struct osd_xattr_entry *tmp;
1590 size_t namelen = strlen(name);
1591 size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1593 OBD_ALLOC(oxe, len);
1597 INIT_LIST_HEAD(&oxe->oxe_list);
1599 oxe->oxe_namelen = namelen;
1600 memcpy(oxe->oxe_buf, name, namelen);
1602 LASSERT(buf != NULL);
1603 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1604 oxe->oxe_exist = true;
1606 oxe->oxe_exist = false;
1609 /* this should be rarely called, just remove old and add new */
1610 spin_lock(&obj->oo_guard);
1611 list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1612 if (namelen == tmp->oxe_namelen &&
1613 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1619 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1620 call_rcu(&old->oxe_rcu, osd_oxc_free);
1622 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1624 spin_unlock(&obj->oo_guard);
1627 static void osd_oxc_del(struct osd_object *obj, const char *name)
1629 struct osd_xattr_entry *oxe;
1630 size_t namelen = strlen(name);
1632 spin_lock(&obj->oo_guard);
1633 list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1634 if (namelen == oxe->oxe_namelen &&
1635 strncmp(name, oxe->oxe_buf, namelen) == 0) {
1636 list_del_rcu(&oxe->oxe_list);
1637 call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1641 spin_unlock(&obj->oo_guard);
1644 static void osd_oxc_fini(struct osd_object *obj)
1646 struct osd_xattr_entry *oxe, *next;
1648 list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1649 list_del(&oxe->oxe_list);
1650 OBD_FREE(oxe, oxe->oxe_len);
1655 * Concurrency: no concurrent access is possible that late in object
1658 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1660 struct osd_object *obj = osd_obj(l);
1661 struct lu_object_header *h = obj->oo_header;
1663 LINVRNT(osd_invariant(obj));
1666 dt_object_fini(&obj->oo_dt);
1667 if (obj->oo_hl_head != NULL)
1668 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1669 /* obj doesn't contain an lu_object_header, so we don't need call_rcu */
1672 lu_object_header_free(h);
1676 * Concurrency: no concurrent access is possible that late in object
1679 static void osd_index_fini(struct osd_object *o)
1681 struct iam_container *bag;
1683 if (o->oo_dir != NULL) {
1684 bag = &o->oo_dir->od_container;
1685 if (o->oo_inode != NULL) {
1686 if (bag->ic_object == o->oo_inode)
1687 iam_container_fini(bag);
1689 OBD_FREE_PTR(o->oo_dir);
1695 OSD_TXN_OI_DELETE_CREDITS = 20,
1696 OSD_TXN_INODE_DELETE_CREDITS = 20
1703 #if OSD_THANDLE_STATS
1705 * Set time when the handle is allocated
1707 static void osd_th_alloced(struct osd_thandle *oth)
1709 oth->oth_alloced = ktime_get();
1713 * Set time when the handle started
1715 static void osd_th_started(struct osd_thandle *oth)
1717 oth->oth_started = ktime_get();
1721 * Check whether the we deal with this handle for too long.
1723 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1724 ktime_t alloced, ktime_t started,
1727 ktime_t now = ktime_get();
1729 LASSERT(dev != NULL);
1731 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1732 ktime_us_delta(started, alloced));
1733 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1734 ktime_us_delta(closed, started));
1735 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1736 ktime_us_delta(now, closed));
1738 if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1739 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1740 oth, now, alloced, started, closed);
1741 libcfs_debug_dumpstack(NULL);
1745 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
1747 ktime_t __closed = ktime_get(); \
1748 ktime_t __alloced = oth->oth_alloced; \
1749 ktime_t __started = oth->oth_started; \
1752 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
1755 #else /* OSD_THANDLE_STATS */
1757 #define osd_th_alloced(h) do {} while(0)
1758 #define osd_th_started(h) do {} while(0)
1759 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
1761 #endif /* OSD_THANDLE_STATS */
1764 * in some cases (like overstriped files) the same operations on the same
1765 * objects are declared many times and this may lead to huge number of
1766 * credits which can be a problem and/or cause performance degradation.
1767 * this function is to remember what declarations have been made within
1768 * a given thandle and then skip duplications.
1769 * limit it's scope so that regular small transactions don't need all
1770 * this overhead with allocations, lists.
1771 * also, limit scope to the specific objects like llogs, etc.
1773 static inline bool osd_check_special_fid(const struct lu_fid *f)
1775 if (fid_seq_is_llog(f->f_seq))
1777 if (f->f_seq == FID_SEQ_LOCAL_FILE &&
1778 f->f_oid == MDD_LOV_OBJ_OID)
1783 bool osd_tx_was_declared(const struct lu_env *env, struct osd_thandle *oth,
1784 struct dt_object *dt, enum dt_txn_op op, loff_t pos)
1786 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1787 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1788 struct osd_thread_info *oti = osd_oti_get(env);
1789 struct osd_obj_declare *old;
1794 /* small transactions don't need this overhead */
1795 if (oti->oti_declare_ops[DTO_OBJECT_CREATE] < 10 &&
1796 oti->oti_declare_ops[DTO_WRITE_BASE] < 10)
1799 if (osd_check_special_fid(fid) == 0)
1802 list_for_each_entry(old, &oth->ot_declare_list, old_list) {
1803 if (old->old_op == op && old->old_pos == pos &&
1804 lu_fid_eq(&old->old_fid, fid))
1808 if (unlikely(old == NULL))
1810 old->old_fid = *lu_object_fid(&dt->do_lu);
1813 list_add(&old->old_list, &oth->ot_declare_list);
1817 void osd_tx_declaration_free(struct osd_thandle *oth)
1819 struct osd_obj_declare *old, *tmp;
1821 list_for_each_entry_safe(old, tmp, &oth->ot_declare_list, old_list) {
1822 list_del_init(&old->old_list);
1828 * Concurrency: doesn't access mutable data.
1830 static int osd_param_is_not_sane(const struct osd_device *dev,
1831 const struct thandle *th)
1833 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1835 return oh->ot_credits > osd_transaction_size(dev);
1839 * Concurrency: shouldn't matter.
1841 static void osd_trans_commit_cb(struct super_block *sb,
1842 struct ldiskfs_journal_cb_entry *jcb, int error)
1844 struct osd_thandle *oh = container_of(jcb, struct osd_thandle, ot_jcb);
1845 struct thandle *th = &oh->ot_super;
1846 struct lu_device *lud = &th->th_dev->dd_lu_dev;
1847 struct osd_device *osd = osd_dev(lud);
1848 struct dt_txn_commit_cb *dcb, *tmp;
1850 LASSERT(oh->ot_handle == NULL);
1853 CERROR("transaction @0x%p commit error: %d\n", th, error);
1855 /* call per-transaction callbacks if any */
1856 list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1858 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1859 "commit callback entry: magic=%x name='%s'\n",
1860 dcb->dcb_magic, dcb->dcb_name);
1861 list_del_init(&dcb->dcb_linkage);
1862 dcb->dcb_func(NULL, th, dcb, error);
1865 lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1866 if (atomic_dec_and_test(&osd->od_commit_cb_in_flight))
1867 wake_up(&osd->od_commit_cb_done);
1873 static struct thandle *osd_trans_create(const struct lu_env *env,
1874 struct dt_device *d)
1876 struct osd_thread_info *oti = osd_oti_get(env);
1877 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1878 struct osd_thandle *oh;
1884 CERROR("%s: someone try to start transaction under "
1885 "readonly mode, should be disabled.\n",
1886 osd_name(osd_dt_dev(d)));
1888 RETURN(ERR_PTR(-EROFS));
1891 /* on pending IO in this thread should left from prev. request */
1892 LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1894 sb_start_write(osd_sb(osd_dt_dev(d)));
1896 OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1898 sb_end_write(osd_sb(osd_dt_dev(d)));
1899 RETURN(ERR_PTR(-ENOMEM));
1902 oh->ot_quota_trans = &oti->oti_quota_trans;
1903 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1908 oh->oh_declared_ext = 0;
1909 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1910 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1911 INIT_LIST_HEAD(&oh->ot_trunc_locks);
1912 INIT_LIST_HEAD(&oh->ot_declare_list);
1915 memset(oti->oti_declare_ops, 0,
1916 sizeof(oti->oti_declare_ops));
1917 memset(oti->oti_declare_ops_cred, 0,
1918 sizeof(oti->oti_declare_ops_cred));
1919 memset(oti->oti_declare_ops_used, 0,
1920 sizeof(oti->oti_declare_ops_used));
1922 oti->oti_ins_cache_depth++;
1927 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1929 struct osd_thread_info *oti = osd_oti_get(env);
1930 struct osd_thandle *oh;
1932 oh = container_of(th, struct osd_thandle, ot_super);
1933 LASSERT(oh != NULL);
1935 CWARN(" create: %u/%u/%u, destroy: %u/%u/%u\n",
1936 oti->oti_declare_ops[OSD_OT_CREATE],
1937 oti->oti_declare_ops_cred[OSD_OT_CREATE],
1938 oti->oti_declare_ops_used[OSD_OT_CREATE],
1939 oti->oti_declare_ops[OSD_OT_DESTROY],
1940 oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1941 oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1942 CWARN(" attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1943 oti->oti_declare_ops[OSD_OT_ATTR_SET],
1944 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1945 oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1946 oti->oti_declare_ops[OSD_OT_XATTR_SET],
1947 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1948 oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1949 CWARN(" write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1950 oti->oti_declare_ops[OSD_OT_WRITE],
1951 oti->oti_declare_ops_cred[OSD_OT_WRITE],
1952 oti->oti_declare_ops_used[OSD_OT_WRITE],
1953 oti->oti_declare_ops[OSD_OT_PUNCH],
1954 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1955 oti->oti_declare_ops_used[OSD_OT_PUNCH],
1956 oti->oti_declare_ops[OSD_OT_QUOTA],
1957 oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1958 oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1959 CWARN(" insert: %u/%u/%u, delete: %u/%u/%u\n",
1960 oti->oti_declare_ops[OSD_OT_INSERT],
1961 oti->oti_declare_ops_cred[OSD_OT_INSERT],
1962 oti->oti_declare_ops_used[OSD_OT_INSERT],
1963 oti->oti_declare_ops[OSD_OT_DELETE],
1964 oti->oti_declare_ops_cred[OSD_OT_DELETE],
1965 oti->oti_declare_ops_used[OSD_OT_DELETE]);
1966 CWARN(" ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1967 oti->oti_declare_ops[OSD_OT_REF_ADD],
1968 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1969 oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1970 oti->oti_declare_ops[OSD_OT_REF_DEL],
1971 oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1972 oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1976 * Concurrency: shouldn't matter.
1978 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1981 struct osd_thread_info *oti = osd_oti_get(env);
1982 struct osd_device *dev = osd_dt_dev(d);
1984 struct osd_thandle *oh;
1989 LASSERT(current->journal_info == NULL);
1991 oh = container_of(th, struct osd_thandle, ot_super);
1992 LASSERT(oh != NULL);
1993 LASSERT(oh->ot_handle == NULL);
1994 if (unlikely(ldiskfs_track_declares_assert != 0)) {
1995 LASSERT(oti->oti_r_locks == 0);
1996 LASSERT(oti->oti_w_locks == 0);
1999 rc = dt_txn_hook_start(env, d, th);
2003 if (unlikely(osd_param_is_not_sane(dev, th))) {
2004 static unsigned long last_printed;
2005 static int last_credits;
2007 lprocfs_counter_add(dev->od_stats,
2008 LPROC_OSD_TOO_MANY_CREDITS, 1);
2011 * don't make noise on a tiny testing systems
2012 * actual credits misuse will be caught anyway
2014 if (last_credits != oh->ot_credits &&
2015 time_after(jiffies, last_printed +
2016 cfs_time_seconds(60)) &&
2017 osd_transaction_size(dev) > 512) {
2018 CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
2019 oh->ot_credits, osd_transaction_size(dev));
2020 osd_trans_dump_creds(env, th);
2021 libcfs_debug_dumpstack(NULL);
2022 last_credits = oh->ot_credits;
2023 last_printed = jiffies;
2026 * XXX Limit the credits to 'max_transaction_buffers', and
2027 * let the underlying filesystem to catch the error if
2028 * we really need so many credits.
2030 * This should be removed when we can calculate the
2031 * credits precisely.
2033 oh->ot_credits = osd_transaction_size(dev);
2034 } else if (ldiskfs_track_declares_assert != 0) {
2036 * reserve few credits to prevent an assertion in JBD
2037 * our debugging mechanism will be able to detected
2038 * overuse. this can help to debug single-update
2041 oh->ot_credits += 10;
2042 if (unlikely(osd_param_is_not_sane(dev, th)))
2043 oh->ot_credits = osd_transaction_size(dev);
2046 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
2047 GOTO(out, rc = -EIO);
2050 * we ignore quota checks for system-owned files, but still
2051 * need to count blocks for uid/gid/projid
2053 osd_trans_declare_op(env, oh, OSD_OT_QUOTA, 3);
2056 * XXX temporary stuff. Some abstraction layer should
2059 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
2063 LASSERT(oti->oti_txns == 0);
2065 atomic_inc(&dev->od_commit_cb_in_flight);
2066 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
2077 static int osd_seq_exists(const struct lu_env *env,
2078 struct osd_device *osd, u64 seq)
2080 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
2081 struct seq_server_site *ss = osd_seq_site(osd);
2086 LASSERT(ss != NULL);
2087 LASSERT(ss->ss_server_fld != NULL);
2089 rc = osd_fld_lookup(env, osd, seq, range);
2092 CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
2093 osd_name(osd), seq, rc);
2097 RETURN(ss->ss_node_id == range->lsr_index);
2100 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
2102 struct dt_txn_commit_cb *dcb;
2103 struct dt_txn_commit_cb *tmp;
2105 /* call per-transaction stop callbacks if any */
2106 list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
2108 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
2109 "commit callback entry: magic=%x name='%s'\n",
2110 dcb->dcb_magic, dcb->dcb_name);
2111 list_del_init(&dcb->dcb_linkage);
2112 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
2117 * Concurrency: shouldn't matter.
2119 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
2122 struct osd_thread_info *oti = osd_oti_get(env);
2123 struct osd_thandle *oh;
2124 struct osd_iobuf *iobuf = &oti->oti_iobuf;
2125 struct osd_device *osd = osd_dt_dev(th->th_dev);
2126 struct qsd_instance *qsd = osd_def_qsd(osd);
2127 struct lquota_trans *qtrans;
2128 LIST_HEAD(truncates);
2129 int rc = 0, remove_agents = 0;
2133 oh = container_of(th, struct osd_thandle, ot_super);
2135 remove_agents = oh->ot_remove_agents;
2137 qtrans = oh->ot_quota_trans;
2138 oh->ot_quota_trans = NULL;
2140 osd_tx_declaration_free(oh);
2142 /* move locks to local list, stop tx, execute truncates */
2143 list_splice(&oh->ot_trunc_locks, &truncates);
2145 if (oh->ot_handle != NULL) {
2148 handle_t *hdl = oh->ot_handle;
2151 * add commit callback
2152 * notice we don't do this in osd_trans_start()
2153 * as underlying transaction can change during truncate
2155 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
2158 LASSERT(oti->oti_txns == 1);
2161 rc = dt_txn_hook_stop(env, th);
2163 CERROR("%s: failed in transaction hook: rc = %d\n",
2166 osd_trans_stop_cb(oh, rc);
2167 /* hook functions might modify th_sync */
2168 hdl->h_sync = th->th_sync;
2170 oh->ot_handle = NULL;
2171 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2173 CERROR("%s: failed to stop transaction: rc = %d\n",
2174 osd_name(osd), rc2);
2178 /* We preserve the origin behavior of ignoring any
2179 * failures with the underlying punch / truncate
2180 * operation. We do record for debugging if an error
2181 * does occur in the lctl dk logs.
2183 rc2 = osd_process_truncates(env, &truncates);
2185 CERROR("%s: failed truncate process: rc = %d\n",
2186 osd_name(osd), rc2);
2188 osd_trans_stop_cb(oh, th->th_result);
2192 osd_trunc_unlock_all(env, &truncates);
2194 /* inform the quota slave device that the transaction is stopping */
2195 qsd_op_end(env, qsd, qtrans);
2198 * as we want IO to journal and data IO be concurrent, we don't block
2199 * awaiting data IO completion in osd_do_bio(), instead we wait here
2200 * once transaction is submitted to the journal. all reqular requests
2201 * don't do direct IO (except read/write), thus this wait_event becomes
2204 * IMPORTANT: we have to wait till any IO submited by the thread is
2205 * completed otherwise iobuf may be corrupted by different request
2207 wait_event(iobuf->dr_wait,
2208 atomic_read(&iobuf->dr_numreqs) == 0);
2209 osd_fini_iobuf(osd, iobuf);
2211 rc = iobuf->dr_error;
2213 if (unlikely(remove_agents != 0))
2214 osd_process_scheduled_agent_removals(env, osd);
2216 LASSERT(oti->oti_ins_cache_depth > 0);
2217 oti->oti_ins_cache_depth--;
2218 /* reset OI cache for safety */
2219 if (oti->oti_ins_cache_depth == 0)
2220 oti->oti_ins_cache_used = 0;
2222 sb_end_write(osd_sb(osd));
2227 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2229 struct osd_thandle *oh = container_of(th, struct osd_thandle,
2232 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2233 LASSERT(&dcb->dcb_func != NULL);
2234 if (dcb->dcb_flags & DCB_TRANS_STOP)
2235 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2237 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2243 * Called just before object is freed. Releases all resources except for
2244 * object itself (that is released by osd_object_free()).
2246 * Concurrency: no concurrent access is possible that late in object
2249 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2251 struct osd_object *obj = osd_obj(l);
2252 struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2253 struct inode *inode = obj->oo_inode;
2258 LINVRNT(osd_invariant(obj));
2261 * If object is unlinked remove fid->ino mapping from object index.
2264 osd_index_fini(obj);
2269 if (osd_has_index(obj) && obj->oo_dt.do_index_ops == &osd_index_iam_ops)
2270 ldiskfs_set_inode_flag(inode, LDISKFS_INODE_JOURNAL_DATA);
2272 uid = i_uid_read(inode);
2273 gid = i_gid_read(inode);
2274 projid = i_projid_read(inode);
2276 obj->oo_inode = NULL;
2279 /* do not rebalance quota if the caller needs to release memory
2280 * otherwise qsd_refresh_usage() may went into a new ldiskfs
2281 * transaction and risk to deadlock - LU-12178 */
2282 if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2285 if (!obj->oo_header && qsd) {
2286 struct osd_thread_info *info = osd_oti_get(env);
2287 struct lquota_id_info *qi = &info->oti_qi;
2289 /* Release granted quota to master if necessary */
2290 qi->lqi_id.qid_uid = uid;
2291 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2293 qi->lqi_id.qid_uid = gid;
2294 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2296 qi->lqi_id.qid_uid = projid;
2297 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2302 * Concurrency: ->loo_object_release() is called under site spin-lock.
2304 static void osd_object_release(const struct lu_env *env,
2305 struct lu_object *l)
2307 struct osd_object *o = osd_obj(l);
2310 * nobody should be releasing a non-destroyed object with nlink=0
2311 * the API allows this, but ldiskfs doesn't like and then report
2312 * this inode as deleted
2314 LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2315 o->oo_inode->i_nlink == 0));
2319 * Concurrency: shouldn't matter.
2321 static int osd_object_print(const struct lu_env *env, void *cookie,
2322 lu_printer_t p, const struct lu_object *l)
2324 struct osd_object *o = osd_obj(l);
2325 struct iam_descr *d;
2327 if (o->oo_dir != NULL)
2328 d = o->oo_dir->od_container.ic_descr;
2331 return (*p)(env, cookie,
2332 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2334 o->oo_inode ? o->oo_inode->i_ino : 0UL,
2335 o->oo_inode ? o->oo_inode->i_generation : 0,
2336 d ? d->id_ops->id_name : "plain");
2340 * Concurrency: shouldn't matter.
2342 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2343 struct obd_statfs *sfs, struct obd_statfs_info *info)
2345 struct osd_device *osd = osd_dt_dev(d);
2346 struct super_block *sb = osd_sb(osd);
2347 struct kstatfs *ksfs;
2351 if (unlikely(osd->od_mnt == NULL))
2352 return -EINPROGRESS;
2354 /* osd_lproc.c call this without env, allocate ksfs for that case */
2355 if (unlikely(env == NULL)) {
2356 OBD_ALLOC_PTR(ksfs);
2360 ksfs = &osd_oti_get(env)->oti_ksfs;
2363 result = sb->s_op->statfs(sb->s_root, ksfs);
2367 statfs_pack(sfs, ksfs);
2368 if (unlikely(sb->s_flags & SB_RDONLY))
2369 sfs->os_state |= OS_STATFS_READONLY;
2371 sfs->os_state |= osd->od_nonrotational ? OS_STATFS_NONROT : 0;
2373 if (ldiskfs_has_feature_extents(sb))
2374 sfs->os_maxbytes = sb->s_maxbytes;
2376 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2379 * Reserve some space so to avoid fragmenting the filesystem too much.
2380 * Fragmentation not only impacts performance, but can also increase
2381 * metadata overhead significantly, causing grant calculation to be
2384 * Reserve 0.78% of total space, at least 8MB for small filesystems.
2386 BUILD_BUG_ON(OSD_STATFS_RESERVED <= LDISKFS_MAX_BLOCK_SIZE);
2387 reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2388 if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2389 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2391 sfs->os_blocks -= reserved;
2392 sfs->os_bfree -= min(reserved, sfs->os_bfree);
2393 sfs->os_bavail -= min(reserved, sfs->os_bavail);
2396 if (unlikely(env == NULL))
2402 * Estimate space needed for file creations. We assume the largest filename
2403 * which is 2^64 - 1, hence a filename of 20 chars.
2404 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2406 #ifdef __LDISKFS_DIR_REC_LEN
2407 # define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2408 #elif defined LDISKFS_DIR_REC_LEN_WITH_DIR
2409 # define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20, NULL)
2411 # define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2415 * Concurrency: doesn't access mutable data.
2417 static void osd_conf_get(const struct lu_env *env,
2418 const struct dt_device *dev,
2419 struct dt_device_param *param)
2421 struct osd_device *d = osd_dt_dev(dev);
2422 struct super_block *sb = osd_sb(d);
2423 struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2428 * XXX should be taken from not-yet-existing fs abstraction layer.
2430 param->ddp_max_name_len = LDISKFS_NAME_LEN;
2431 param->ddp_max_nlink = LDISKFS_LINK_MAX;
2432 param->ddp_symlink_max = sb->s_blocksize;
2433 param->ddp_mount_type = LDD_MT_LDISKFS;
2434 if (ldiskfs_has_feature_extents(sb))
2435 param->ddp_maxbytes = sb->s_maxbytes;
2437 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2439 * inode are statically allocated, so per-inode space consumption
2440 * is the space consumed by the directory entry
2442 param->ddp_inodespace = PER_OBJ_USAGE;
2444 * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2445 * is 128MB) which is unlikely to be hit in real life. Report a smaller
2446 * maximum length to not under-count the actual number of extents
2447 * needed for writing a file if there are sub-optimal block allocations.
2449 param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2450 /* worst-case extent insertion metadata overhead */
2451 param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2452 param->ddp_mntopts = 0;
2453 if (test_opt(sb, XATTR_USER))
2454 param->ddp_mntopts |= MNTOPT_USERXATTR;
2455 if (test_opt(sb, POSIX_ACL))
2456 param->ddp_mntopts |= MNTOPT_ACL;
2459 * LOD might calculate the max stripe count based on max_ea_size,
2460 * so we need take account in the overhead as well,
2461 * xattr_header + magic + xattr_entry_head
2463 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2464 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2466 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2467 if (ldiskfs_has_feature_ea_inode(sb))
2468 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2472 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2474 if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE)
2475 param->ddp_max_ea_size = OBD_MAX_EA_SIZE;
2478 * Preferred RPC size for efficient disk IO. 4MB shows good
2479 * all-around performance for ldiskfs, but use bigalloc chunk size
2480 * by default if larger.
2482 #if defined(LDISKFS_CLUSTER_SIZE)
2483 if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2484 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2487 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2489 param->ddp_t10_cksum_type = 0;
2491 unsigned short interval = blk_integrity_interval(bi);
2492 name = blk_integrity_name(bi);
2502 if (strncmp(name, "T10-DIF-TYPE",
2503 sizeof("T10-DIF-TYPE") - 1) == 0) {
2504 /* also skip "1/2/3-" at end */
2505 const int type_off = sizeof("T10-DIF-TYPE.");
2506 char type_number = name[type_off - 2];
2508 if (interval != 512 && interval != 4096) {
2509 CERROR("%s: unsupported T10PI sector size %u\n",
2510 d->od_svname, interval);
2513 switch (type_number) {
2515 d->od_t10_type = OSD_T10_TYPE1;
2518 d->od_t10_type = OSD_T10_TYPE2;
2521 d->od_t10_type = OSD_T10_TYPE3;
2524 CERROR("%s: unsupported T10PI type %s\n",
2525 d->od_svname, name);
2528 if (strcmp(name + type_off, "CRC") == 0) {
2529 d->od_t10_type |= OSD_T10_TYPE_CRC;
2530 param->ddp_t10_cksum_type = interval == 512 ?
2531 OBD_CKSUM_T10CRC512 :
2533 } else if (strcmp(name + type_off, "IP") == 0) {
2534 d->od_t10_type |= OSD_T10_TYPE_IP;
2535 param->ddp_t10_cksum_type = interval == 512 ?
2536 OBD_CKSUM_T10IP512 :
2539 CERROR("%s: unsupported checksum type of T10PI type '%s'\n",
2540 d->od_svname, name);
2545 CERROR("%s: unsupported T10PI type '%s'\n",
2546 d->od_svname, name);
2551 param->ddp_has_lseek_data_hole = true;
2554 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2556 return osd_sb(osd_dt_dev(d));
2560 * Concurrency: shouldn't matter.
2562 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2565 struct super_block *s = osd_sb(osd_dt_dev(d));
2568 down_read(&s->s_umount);
2569 rc = s->s_op->sync_fs(s, 1);
2570 up_read(&s->s_umount);
2572 CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2579 * Start commit for OSD device.
2581 * An implementation of dt_commit_async method for OSD device.
2582 * Asychronously starts underlayng fs sync and thereby a transaction
2585 * \param env environment
2586 * \param d dt device
2588 * \see dt_device_operations
2590 static int osd_commit_async(const struct lu_env *env,
2591 struct dt_device *d)
2593 struct super_block *s = osd_sb(osd_dt_dev(d));
2598 CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2599 down_read(&s->s_umount);
2600 rc = s->s_op->sync_fs(s, 0);
2601 up_read(&s->s_umount);
2607 * Concurrency: shouldn't matter.
2609 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2611 struct super_block *sb = osd_sb(osd_dt_dev(d));
2612 struct block_device *dev = sb->s_bdev;
2613 int rc = -EOPNOTSUPP;
2617 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2618 osd_dt_dev(d)->od_svname, (long)dev, rc);
2624 * Note: we do not count into QUOTA here.
2625 * If we mount with --data_journal we may need more.
2627 const int osd_dto_credits_noquota[DTO_NR] = {
2630 * INDEX_EXTRA_TRANS_BLOCKS(8) +
2631 * SINGLEDATA_TRANS_BLOCKS(8)
2632 * XXX Note: maybe iam need more, since iam have more level than
2635 [DTO_INDEX_INSERT] = 16,
2638 * just modify a single entry, probably merge few within a block
2640 [DTO_INDEX_DELETE] = 1,
2644 [DTO_INDEX_UPDATE] = 16,
2646 * 4(inode, inode bits, groups, GDT)
2647 * notice: OI updates are counted separately with DTO_INDEX_INSERT
2649 [DTO_OBJECT_CREATE] = 4,
2651 * 4(inode, inode bits, groups, GDT)
2652 * notice: OI updates are counted separately with DTO_INDEX_DELETE
2654 [DTO_OBJECT_DELETE] = 4,
2656 * Attr set credits (inode)
2658 [DTO_ATTR_SET_BASE] = 1,
2660 * Xattr set. The same as xattr of EXT3.
2661 * DATA_TRANS_BLOCKS(14)
2662 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2663 * are also counted in. Do not know why?
2665 [DTO_XATTR_SET] = 14,
2667 * credits for inode change during write.
2669 [DTO_WRITE_BASE] = 3,
2671 * credits for single block write.
2673 [DTO_WRITE_BLOCK] = 14,
2675 * Attr set credits for chown.
2676 * This is extra credits for setattr, and it is null without quota
2678 [DTO_ATTR_SET_CHOWN] = 0
2681 /* reserve or free quota for some operation */
2682 static int osd_reserve_or_free_quota(const struct lu_env *env,
2683 struct dt_device *dev,
2684 struct lquota_id_info *qi)
2686 struct osd_device *osd = osd_dt_dev(dev);
2687 struct qsd_instance *qsd = NULL;
2693 qsd = osd->od_quota_slave_dt;
2695 qsd = osd->od_quota_slave_md;
2697 rc = qsd_reserve_or_free_quota(env, qsd, qi);
2701 static const struct dt_device_operations osd_dt_ops = {
2702 .dt_root_get = osd_root_get,
2703 .dt_statfs = osd_statfs,
2704 .dt_trans_create = osd_trans_create,
2705 .dt_trans_start = osd_trans_start,
2706 .dt_trans_stop = osd_trans_stop,
2707 .dt_trans_cb_add = osd_trans_cb_add,
2708 .dt_conf_get = osd_conf_get,
2709 .dt_mnt_sb_get = osd_mnt_sb_get,
2710 .dt_sync = osd_sync,
2712 .dt_commit_async = osd_commit_async,
2713 .dt_reserve_or_free_quota = osd_reserve_or_free_quota,
2716 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2719 struct osd_object *obj = osd_dt_obj(dt);
2720 struct osd_thread_info *oti = osd_oti_get(env);
2722 LINVRNT(osd_invariant(obj));
2724 LASSERT(obj->oo_owner != env);
2725 down_read_nested(&obj->oo_sem, role);
2727 LASSERT(obj->oo_owner == NULL);
2731 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2734 struct osd_object *obj = osd_dt_obj(dt);
2735 struct osd_thread_info *oti = osd_oti_get(env);
2737 LINVRNT(osd_invariant(obj));
2739 LASSERT(obj->oo_owner != env);
2740 down_write_nested(&obj->oo_sem, role);
2742 LASSERT(obj->oo_owner == NULL);
2743 obj->oo_owner = env;
2747 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2749 struct osd_object *obj = osd_dt_obj(dt);
2750 struct osd_thread_info *oti = osd_oti_get(env);
2752 LINVRNT(osd_invariant(obj));
2754 LASSERT(oti->oti_r_locks > 0);
2756 up_read(&obj->oo_sem);
2759 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2761 struct osd_object *obj = osd_dt_obj(dt);
2762 struct osd_thread_info *oti = osd_oti_get(env);
2764 LINVRNT(osd_invariant(obj));
2766 LASSERT(obj->oo_owner == env);
2767 LASSERT(oti->oti_w_locks > 0);
2769 obj->oo_owner = NULL;
2770 up_write(&obj->oo_sem);
2773 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2775 struct osd_object *obj = osd_dt_obj(dt);
2777 LINVRNT(osd_invariant(obj));
2779 return obj->oo_owner == env;
2782 static void osd_inode_getattr(const struct lu_env *env,
2783 struct inode *inode, struct lu_attr *attr)
2785 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2786 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2787 LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2788 LA_BLKSIZE | LA_TYPE | LA_BTIME;
2790 attr->la_atime = inode->i_atime.tv_sec;
2791 attr->la_mtime = inode->i_mtime.tv_sec;
2792 attr->la_ctime = inode->i_ctime.tv_sec;
2793 attr->la_btime = LDISKFS_I(inode)->i_crtime.tv_sec;
2794 attr->la_mode = inode->i_mode;
2795 attr->la_size = i_size_read(inode);
2796 attr->la_blocks = inode->i_blocks;
2797 attr->la_uid = i_uid_read(inode);
2798 attr->la_gid = i_gid_read(inode);
2799 attr->la_projid = i_projid_read(inode);
2800 attr->la_flags = ll_inode_to_ext_flags(inode->i_flags);
2801 attr->la_nlink = inode->i_nlink;
2802 attr->la_rdev = inode->i_rdev;
2803 attr->la_blksize = 1 << inode->i_blkbits;
2804 attr->la_blkbits = inode->i_blkbits;
2806 * Ext4 did not transfer inherit flags from raw inode
2807 * to inode flags, and ext4 internally test raw inode
2808 * @i_flags directly. Instead of patching ext4, we do it here.
2810 if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2811 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2814 static int osd_dirent_count(const struct lu_env *env, struct dt_object *dt,
2817 struct osd_object *obj = osd_dt_obj(dt);
2818 const struct dt_it_ops *iops;
2824 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2825 LASSERT(fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)));
2827 if (obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
2828 *count = obj->oo_dirent_count;
2832 /* directory not initialized yet */
2833 if (!dt->do_index_ops) {
2838 iops = &dt->do_index_ops->dio_it;
2839 it = iops->init(env, dt, LUDA_64BITHASH);
2841 RETURN(PTR_ERR(it));
2843 rc = iops->load(env, it, 0);
2845 if (rc == -ENODATA) {
2852 rc = iops->next(env, it);
2854 for (*count = 0; rc == 0 || rc == -ESTALE; rc = iops->next(env, it)) {
2858 if (iops->key_size(env, it) == 0)
2864 obj->oo_dirent_count = *count;
2869 iops->fini(env, it);
2874 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2875 struct lu_attr *attr)
2877 struct osd_object *obj = osd_dt_obj(dt);
2880 if (unlikely(!dt_object_exists(dt)))
2882 if (unlikely(obj->oo_destroyed))
2885 LASSERT(!dt_object_remote(dt));
2886 LINVRNT(osd_invariant(obj));
2888 spin_lock(&obj->oo_guard);
2889 osd_inode_getattr(env, obj->oo_inode, attr);
2890 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2891 attr->la_valid |= LA_FLAGS;
2892 attr->la_flags |= LUSTRE_ORPHAN_FL;
2894 if (obj->oo_lma_flags & LUSTRE_ENCRYPT_FL) {
2895 attr->la_valid |= LA_FLAGS;
2896 attr->la_flags |= LUSTRE_ENCRYPT_FL;
2898 spin_unlock(&obj->oo_guard);
2900 if (S_ISDIR(obj->oo_inode->i_mode) &&
2901 fid_is_namespace_visible(lu_object_fid(&dt->do_lu)))
2902 rc = osd_dirent_count(env, dt, &attr->la_dirent_count);
2907 static int osd_declare_attr_qid(const struct lu_env *env,
2908 struct osd_object *obj,
2909 struct osd_thandle *oh, long long bspace,
2910 qid_t old_id, qid_t new_id, bool enforce,
2914 struct osd_thread_info *info = osd_oti_get(env);
2915 struct lquota_id_info *qi = &info->oti_qi;
2917 qi->lqi_type = type;
2918 /* inode accounting */
2919 qi->lqi_is_blk = false;
2921 /* one more inode for the new id ... */
2922 qi->lqi_id.qid_uid = new_id;
2924 /* Reserve credits for the new id */
2925 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2926 if (rc == -EDQUOT || rc == -EINPROGRESS)
2931 /* and one less inode for the current id */
2932 qi->lqi_id.qid_uid = old_id;
2934 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2935 if (rc == -EDQUOT || rc == -EINPROGRESS)
2940 /* block accounting */
2941 qi->lqi_is_blk = true;
2943 /* more blocks for the new id ... */
2944 qi->lqi_id.qid_uid = new_id;
2945 qi->lqi_space = bspace;
2947 * Credits for the new uid has been reserved, re-use "obj"
2948 * to save credit reservation.
2950 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2951 if (rc == -EDQUOT || rc == -EINPROGRESS)
2956 /* and finally less blocks for the current uid */
2957 qi->lqi_id.qid_uid = old_id;
2958 qi->lqi_space = -bspace;
2959 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2960 if (rc == -EDQUOT || rc == -EINPROGRESS)
2966 static int osd_declare_attr_set(const struct lu_env *env,
2967 struct dt_object *dt,
2968 const struct lu_attr *attr,
2969 struct thandle *handle)
2971 struct osd_thandle *oh;
2972 struct osd_object *obj;
2981 LASSERT(dt != NULL);
2982 LASSERT(handle != NULL);
2984 obj = osd_dt_obj(dt);
2985 LASSERT(osd_invariant(obj));
2987 oh = container_of(handle, struct osd_thandle, ot_super);
2988 LASSERT(oh->ot_handle == NULL);
2990 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2991 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2993 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2994 osd_dto_credits_noquota[DTO_XATTR_SET]);
2996 if (attr == NULL || obj->oo_inode == NULL)
2999 bspace = obj->oo_inode->i_blocks << 9;
3000 bspace = toqb(bspace);
3003 * Changing ownership is always preformed by super user, it should not
3004 * fail with EDQUOT unless required explicitly.
3006 * We still need to call the osd_declare_qid() to calculate the journal
3007 * credits for updating quota accounting files and to trigger quota
3008 * space adjustment once the operation is completed.
3010 if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
3012 uid = i_uid_read(obj->oo_inode);
3013 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
3014 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
3015 attr->la_uid, enforce, USRQUOTA);
3019 gid = i_gid_read(obj->oo_inode);
3020 CDEBUG(D_QUOTA, "declare uid %d -> %d gid %d -> %d\n", uid,
3021 attr->la_uid, gid, attr->la_gid);
3022 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
3023 rc = osd_declare_attr_qid(env, obj, oh, bspace, gid,
3024 attr->la_gid, enforce, GRPQUOTA);
3029 #ifdef HAVE_PROJECT_QUOTA
3030 if (attr->la_valid & LA_PROJID) {
3031 __u32 projid = i_projid_read(obj->oo_inode);
3033 enforce = (attr->la_valid & LA_PROJID) &&
3034 (attr->la_projid != projid);
3035 rc = osd_declare_attr_qid(env, obj, oh, bspace,
3036 (qid_t)projid, (qid_t)attr->la_projid,
3042 /* punch must be aware we are dealing with an encrypted file */
3043 if (attr->la_valid & LA_FLAGS && attr->la_flags & LUSTRE_ENCRYPT_FL)
3044 obj->oo_lma_flags |= LUSTRE_ENCRYPT_FL;
3049 static int osd_inode_setattr(const struct lu_env *env,
3050 struct inode *inode, const struct lu_attr *attr)
3052 __u64 bits = attr->la_valid;
3054 /* Only allow set size for regular file */
3055 if (!S_ISREG(inode->i_mode))
3056 bits &= ~(LA_SIZE | LA_BLOCKS);
3061 if (bits & LA_ATIME)
3062 inode->i_atime = osd_inode_time(inode, attr->la_atime);
3063 if (bits & LA_CTIME)
3064 inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
3065 if (bits & LA_MTIME)
3066 inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
3067 if (bits & LA_SIZE) {
3068 spin_lock(&inode->i_lock);
3069 LDISKFS_I(inode)->i_disksize = attr->la_size;
3070 i_size_write(inode, attr->la_size);
3071 spin_unlock(&inode->i_lock);
3075 * OSD should not change "i_blocks" which is used by quota.
3076 * "i_blocks" should be changed by ldiskfs only.
3079 inode->i_mode = (inode->i_mode & S_IFMT) |
3080 (attr->la_mode & ~S_IFMT);
3082 i_uid_write(inode, attr->la_uid);
3084 i_gid_write(inode, attr->la_gid);
3085 if (bits & LA_PROJID)
3086 i_projid_write(inode, attr->la_projid);
3087 if (bits & LA_NLINK)
3088 set_nlink(inode, attr->la_nlink);
3090 inode->i_rdev = attr->la_rdev;
3092 if (bits & LA_FLAGS) {
3093 /* always keep S_NOCMTIME */
3094 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
3096 #if defined(S_ENCRYPTED)
3097 /* Always remove S_ENCRYPTED, because ldiskfs must not be
3098 * aware of encryption status. It is just stored into LMA
3099 * so that it can be forwared to client side.
3101 inode->i_flags &= ~S_ENCRYPTED;
3104 * Ext4 did not transfer inherit flags from
3105 * @inode->i_flags to raw inode i_flags when writing
3106 * flags, we do it explictly here.
3108 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
3109 LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
3111 LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
3116 #ifdef HAVE_PROJECT_QUOTA
3117 static int osd_transfer_project(struct inode *inode, __u32 projid,
3118 struct thandle *handle)
3120 struct super_block *sb = inode->i_sb;
3121 struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
3124 struct ldiskfs_iloc iloc;
3125 struct ldiskfs_inode *raw_inode;
3126 struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
3128 if (!ldiskfs_has_feature_project(sb)) {
3129 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
3130 == LDISKFS_DEF_PROJID);
3131 if (projid != LDISKFS_DEF_PROJID)
3137 if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
3140 kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
3141 if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
3144 err = ldiskfs_get_inode_loc(inode, &iloc);
3148 raw_inode = ldiskfs_raw_inode(&iloc);
3149 if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
3150 struct osd_thandle *oh = container_of(handle,
3154 * try to expand inode size automatically.
3156 ldiskfs_mark_inode_dirty(oh->ot_handle, inode);
3157 if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
3165 dquot_initialize(inode);
3166 transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
3167 if (transfer_to[PRJQUOTA]) {
3168 lock_dquot_transfer(inode);
3169 err = __dquot_transfer(inode, transfer_to);
3170 unlock_dquot_transfer(inode);
3171 dqput(transfer_to[PRJQUOTA]);
3180 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr,
3181 struct thandle *handle)
3185 if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
3186 (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
3190 "executing dquot_transfer inode %ld uid %d -> %d gid %d -> %d\n",
3191 inode->i_ino, i_uid_read(inode), attr->la_uid,
3192 i_gid_read(inode), attr->la_gid);
3194 dquot_initialize(inode);
3196 if (attr->la_valid & LA_UID)
3197 iattr.ia_valid |= ATTR_UID;
3198 if (attr->la_valid & LA_GID)
3199 iattr.ia_valid |= ATTR_GID;
3200 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
3201 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
3203 lock_dquot_transfer(inode);
3204 rc = osd_dquot_transfer(&init_user_ns, inode, &iattr);
3205 unlock_dquot_transfer(inode);
3207 CERROR("%s: quota transfer failed. Is quota enforcement enabled on the ldiskfs filesystem? rc = %d\n",
3208 osd_ino2name(inode), rc);
3213 /* Handle project id transfer here properly */
3214 if (attr->la_valid & LA_PROJID &&
3215 attr->la_projid != i_projid_read(inode)) {
3216 if (!projid_valid(make_kprojid(&init_user_ns, attr->la_projid)))
3218 #ifdef HAVE_PROJECT_QUOTA
3219 rc = osd_transfer_project(inode, attr->la_projid, handle);
3224 CERROR("%s: quota transfer failed. Is project enforcement enabled on the ldiskfs filesystem? rc = %d\n",
3225 osd_ino2name(inode), rc);
3232 static int osd_attr_set(const struct lu_env *env,
3233 struct dt_object *dt,
3234 const struct lu_attr *attr,
3235 struct thandle *handle)
3237 struct osd_thread_info *info = osd_oti_get(env);
3238 struct osd_object *obj = osd_dt_obj(dt);
3239 struct osd_device *osd = osd_obj2dev(obj);
3240 struct inode *inode;
3243 if (!dt_object_exists(dt))
3246 LASSERT(handle != NULL);
3247 LASSERT(!dt_object_remote(dt));
3248 LASSERT(osd_invariant(obj));
3250 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
3252 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
3253 !osd_obj2dev(obj)->od_is_ost) {
3254 struct osd_thread_info *oti = osd_oti_get(env);
3255 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
3256 struct lu_fid *fid1 = &oti->oti_fid;
3257 struct osd_inode_id *id = &oti->oti_id;
3258 struct iam_path_descr *ipd;
3259 struct iam_container *bag;
3260 struct osd_thandle *oh;
3263 fid_cpu_to_be(fid1, fid0);
3264 memset(id, 1, sizeof(*id));
3265 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
3266 fid0)->oi_dir.od_container;
3267 ipd = osd_idx_ipd_get(env, bag);
3268 if (unlikely(ipd == NULL))
3271 oh = container_of(handle, struct osd_thandle, ot_super);
3272 rc = iam_update(oh->ot_handle, bag,
3273 (const struct iam_key *)fid1,
3274 (const struct iam_rec *)id, ipd);
3275 osd_ipd_put(env, bag, ipd);
3276 return(rc > 0 ? 0 : rc);
3279 inode = obj->oo_inode;
3281 rc = osd_quota_transfer(inode, attr, handle);
3285 spin_lock(&obj->oo_guard);
3286 rc = osd_inode_setattr(env, inode, attr);
3287 spin_unlock(&obj->oo_guard);
3291 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3293 osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3295 if (!(attr->la_valid & LA_FLAGS))
3298 /* If setting LUSTRE_ENCRYPT_FL on an OST object, also set a dummy
3299 * enc ctx xattr, with 2 benefits:
3300 * - setting the LL_XATTR_NAME_ENCRYPTION_CONTEXT xattr internally sets
3301 * the LDISKFS_ENCRYPT_FL flag on the on-disk inode;
3302 * - it makes e2fsprogs happy to see an enc ctx for an inode that has
3303 * the LDISKFS_ENCRYPT_FL flag
3304 * We do not need the actual encryption context on OST objects, it is
3305 * only stored on MDT inodes, at file creation time.
3307 if (!(LDISKFS_I(obj->oo_inode)->i_flags & LDISKFS_ENCRYPT_FL) &&
3308 attr->la_flags & LUSTRE_ENCRYPT_FL && osd->od_is_ost &&
3309 !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_ENCFLAG)) {
3312 /* use a dummy enc ctx, fine with e2fsprogs */
3313 buf.lb_buf = "\xFF";
3315 rc = osd_xattr_set(env, dt, &buf,
3316 LL_XATTR_NAME_ENCRYPTION_CONTEXT,
3319 CWARN("%s: set "DFID" enc ctx failed: rc = %d\n",
3320 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3324 /* Let's check if there are extra flags need to be set into LMA */
3325 if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3326 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3328 LASSERT(!obj->oo_pfid_in_lma);
3330 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3331 &info->oti_ost_attrs);
3335 if ((lma->lma_incompat & lustre_to_lma_flags(attr->la_flags)) ==
3336 lustre_to_lma_flags(attr->la_flags))
3337 /* if lma incompat already has the flags,
3338 * save a useless call to xattr_set
3342 lma->lma_incompat |=
3343 lustre_to_lma_flags(attr->la_flags);
3344 lustre_lma_swab(lma);
3346 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3348 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3349 lma, sizeof(*lma), XATTR_REPLACE);
3351 CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3352 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3353 lma->lma_incompat, rc);
3356 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3357 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3364 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3365 struct osd_object *obj,
3366 const char *name, const int namelen)
3368 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3371 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3372 umode_t mode, struct dt_allocation_hint *hint,
3373 struct thandle *th, struct lu_attr *attr)
3376 struct osd_device *osd = osd_obj2dev(obj);
3377 struct osd_thandle *oth;
3378 struct dt_object *parent = NULL;
3379 struct inode *inode;
3380 struct iattr iattr = {
3381 .ia_valid = ATTR_UID | ATTR_GID |
3382 ATTR_CTIME | ATTR_MTIME | ATTR_ATIME,
3383 .ia_ctime.tv_sec = attr->la_ctime,
3384 .ia_mtime.tv_sec = attr->la_mtime,
3385 .ia_atime.tv_sec = attr->la_atime,
3386 .ia_uid = GLOBAL_ROOT_UID,
3387 .ia_gid = GLOBAL_ROOT_GID,
3389 const struct osd_timespec omit = { .tv_nsec = UTIME_OMIT };
3391 if (attr->la_valid & LA_UID)
3392 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
3393 if (attr->la_valid & LA_GID)
3394 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
3396 LINVRNT(osd_invariant(obj));
3397 LASSERT(obj->oo_inode == NULL);
3398 LASSERT(obj->oo_hl_head == NULL);
3400 if (S_ISDIR(mode) && ldiskfs_pdo) {
3402 ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3403 if (obj->oo_hl_head == NULL)
3407 oth = container_of(th, struct osd_thandle, ot_super);
3408 LASSERT(oth->ot_handle->h_transaction != NULL);
3410 if (hint != NULL && hint->dah_parent != NULL &&
3411 !dt_object_remote(hint->dah_parent))
3412 parent = hint->dah_parent;
3414 /* if a time component is not valid set it to UTIME_OMIT */
3415 if (!(attr->la_valid & LA_CTIME))
3416 iattr.ia_ctime = omit;
3417 if (!(attr->la_valid & LA_MTIME))
3418 iattr.ia_mtime = omit;
3419 if (!(attr->la_valid & LA_ATIME))
3420 iattr.ia_atime = omit;
3422 inode = ldiskfs_create_inode(oth->ot_handle,
3423 parent ? osd_dt_obj(parent)->oo_inode :
3424 osd_sb(osd)->s_root->d_inode,
3426 if (!IS_ERR(inode)) {
3427 /* Do not update file c/mtime in ldiskfs. */
3428 inode->i_flags |= S_NOCMTIME;
3431 * For new created object, it must be consistent,
3432 * and it is unnecessary to scrub against it.
3434 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3436 obj->oo_inode = inode;
3439 if (obj->oo_hl_head != NULL) {
3440 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3441 obj->oo_hl_head = NULL;
3443 result = PTR_ERR(inode);
3445 LINVRNT(osd_invariant(obj));
3453 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3454 struct lu_attr *attr,
3455 struct dt_allocation_hint *hint,
3456 struct dt_object_format *dof,
3460 struct osd_thandle *oth;
3461 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3463 LASSERT(S_ISDIR(attr->la_mode));
3465 oth = container_of(th, struct osd_thandle, ot_super);
3466 LASSERT(oth->ot_handle->h_transaction != NULL);
3467 if (fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)))
3468 obj->oo_dirent_count = 0;
3469 result = osd_mkfile(info, obj, mode, hint, th, attr);
3474 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3475 struct lu_attr *attr,
3476 struct dt_allocation_hint *hint,
3477 struct dt_object_format *dof,
3481 struct osd_thandle *oth;
3482 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3484 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3486 LASSERT(S_ISREG(attr->la_mode));
3488 oth = container_of(th, struct osd_thandle, ot_super);
3489 LASSERT(oth->ot_handle->h_transaction != NULL);
3491 result = osd_mkfile(info, obj, mode, hint, th, attr);
3493 LASSERT(obj->oo_inode != NULL);
3494 if (feat->dif_flags & DT_IND_VARKEY)
3495 result = iam_lvar_create(obj->oo_inode,
3496 feat->dif_keysize_max,
3498 feat->dif_recsize_max,
3501 result = iam_lfix_create(obj->oo_inode,
3502 feat->dif_keysize_max,
3504 feat->dif_recsize_max,
3510 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3511 struct lu_attr *attr,
3512 struct dt_allocation_hint *hint,
3513 struct dt_object_format *dof,
3516 LASSERT(S_ISREG(attr->la_mode));
3517 return osd_mkfile(info, obj, (attr->la_mode &
3518 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3522 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3523 struct lu_attr *attr,
3524 struct dt_allocation_hint *hint,
3525 struct dt_object_format *dof,
3528 LASSERT(S_ISLNK(attr->la_mode));
3529 return osd_mkfile(info, obj, (attr->la_mode &
3530 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3534 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3535 struct lu_attr *attr,
3536 struct dt_allocation_hint *hint,
3537 struct dt_object_format *dof,
3540 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3543 LINVRNT(osd_invariant(obj));
3544 LASSERT(obj->oo_inode == NULL);
3545 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3546 S_ISFIFO(mode) || S_ISSOCK(mode));
3548 result = osd_mkfile(info, obj, mode, hint, th, attr);
3550 LASSERT(obj->oo_inode != NULL);
3552 * This inode should be marked dirty for i_rdev. Currently
3553 * that is done in the osd_attr_init().
3555 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3558 LINVRNT(osd_invariant(obj));
3562 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3564 struct dt_allocation_hint *hint,
3565 struct dt_object_format *dof,
3568 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3570 osd_obj_type_f result;
3586 result = osd_mk_index;
3596 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3597 struct dt_object *parent, struct dt_object *child,
3602 ah->dah_parent = parent;
3604 if (parent != NULL && !dt_object_remote(parent)) {
3605 /* will help to find FID->ino at dt_insert("..") */
3606 struct osd_object *pobj = osd_dt_obj(parent);
3608 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3612 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3613 struct lu_attr *attr, struct dt_object_format *dof,
3614 struct thandle *handle)
3616 struct inode *inode = obj->oo_inode;
3617 __u64 valid = attr->la_valid;
3620 attr->la_valid &= ~(LA_TYPE | LA_MODE);
3622 if (dof->dof_type != DFT_NODE)
3623 attr->la_valid &= ~LA_RDEV;
3624 if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3625 attr->la_valid &= ~LA_ATIME;
3626 if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3627 attr->la_valid &= ~LA_CTIME;
3628 if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3629 attr->la_valid &= ~LA_MTIME;
3631 result = osd_quota_transfer(inode, attr, handle);
3635 if (attr->la_valid != 0) {
3636 result = osd_inode_setattr(info->oti_env, inode, attr);
3638 * The osd_inode_setattr() should always succeed here. The
3639 * only error that could be returned is EDQUOT when we are
3640 * trying to change the UID or GID of the inode. However, this
3641 * should not happen since quota enforcement is no longer
3642 * enabled on ldiskfs (lquota takes care of it).
3644 LASSERTF(result == 0, "%d\n", result);
3645 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3648 attr->la_valid = valid;
3652 * Helper function for osd_create()
3654 * \retval 0, on success
3656 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3657 struct lu_attr *attr, struct dt_allocation_hint *hint,
3658 struct dt_object_format *dof, struct thandle *th)
3663 osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3665 /* we drop umask so that permissions we pass are not affected */
3666 umask = current->fs->umask;
3667 current->fs->umask = 0;
3669 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3671 if (likely(obj->oo_inode != NULL)) {
3672 LASSERT(obj->oo_inode->i_state & I_NEW);
3675 * Unlock the inode before attr initialization to avoid
3676 * unnecessary dqget operations. LU-6378
3678 unlock_new_inode(obj->oo_inode);
3681 if (likely(result == 0)) {
3682 osd_attr_init(info, obj, attr, dof, th);
3683 osd_object_init0(obj);
3686 /* restore previous umask value */
3687 current->fs->umask = umask;
3689 osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3695 * Helper function for osd_create()
3697 * \retval 0, on success
3699 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3700 const struct lu_fid *fid, struct thandle *th)
3702 struct osd_thread_info *info = osd_oti_get(env);
3703 struct osd_inode_id *id = &info->oti_id;
3704 struct osd_device *osd = osd_obj2dev(obj);
3705 struct osd_thandle *oh;
3708 LASSERT(obj->oo_inode != NULL);
3710 if (CFS_FAIL_CHECK(OBD_FAIL_OSD_OI_ENOSPC))
3713 oh = container_of(th, struct osd_thandle, ot_super);
3714 LASSERT(oh->ot_handle);
3715 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3717 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3718 rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3719 OI_CHECK_FLD, NULL);
3720 if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP) && osd->od_is_ost) {
3721 struct lu_fid next_fid = *fid;
3723 /* insert next object in advance, and map to the same inode */
3725 if (next_fid.f_oid != 0) {
3726 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3727 osd_oi_insert(info, osd, &next_fid, id, oh->ot_handle,
3728 OI_CHECK_FLD, NULL);
3729 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3733 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3738 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3739 u64 seq, struct lu_seq_range *range)
3741 struct seq_server_site *ss = osd_seq_site(osd);
3743 if (fid_seq_is_idif(seq)) {
3744 fld_range_set_ost(range);
3745 range->lsr_index = idif_ost_idx(seq);
3749 if (!fid_seq_in_fldb(seq)) {
3750 fld_range_set_mdt(range);
3753 * FIXME: If ss is NULL, it suppose not get lsr_index
3756 range->lsr_index = ss->ss_node_id;
3760 LASSERT(ss != NULL);
3761 fld_range_set_any(range);
3762 /* OSD will only do local fld lookup */
3763 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3766 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3767 struct lu_attr *attr,
3768 struct dt_allocation_hint *hint,
3769 struct dt_object_format *dof,
3770 struct thandle *handle)
3772 struct osd_thandle *oh;
3773 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3779 LASSERT(handle != NULL);
3781 oh = container_of(handle, struct osd_thandle, ot_super);
3782 LASSERT(oh->ot_handle == NULL);
3784 if (osd_tx_was_declared(env, oh, dt, DTO_OBJECT_CREATE, 0))
3788 * EA object consumes more credits than regular object: osd_mk_index
3789 * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3790 * leaf_node, could involves the block, block bitmap, groups, GDT
3791 * change for each block, so add 4 * 2 credits in that case.
3793 * The default ACL initialization may consume an additional 16 blocks
3795 credits = osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3796 ((dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3799 * While ldiskfs_new_inode() calls ldiskfs_init_acl() we have to add
3800 * credits for possible default ACL creation in new inode
3802 if (hint && hint->dah_acl_len)
3803 credits += osd_calc_bkmap_credits(sb, NULL, 0, -1,
3804 (hint->dah_acl_len + sb->s_blocksize - 1) >>
3805 sb->s_blocksize_bits);
3807 osd_trans_declare_op(env, oh, OSD_OT_CREATE, credits);
3810 * Reuse idle OI block may cause additional one OI block
3813 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3814 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3815 if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP))
3816 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3817 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3819 /* will help to find FID->ino mapping at dt_insert() */
3820 rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3828 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3829 attr->la_projid, 1, oh, osd_dt_obj(dt),
3830 NULL, OSD_QID_INODE);
3838 * Called to destroy on-disk representation of the object
3840 * Concurrency: must be locked
3842 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3845 struct osd_object *obj = osd_dt_obj(dt);
3846 struct inode *inode = obj->oo_inode;
3847 struct osd_thandle *oh;
3855 oh = container_of(th, struct osd_thandle, ot_super);
3856 LASSERT(oh->ot_handle == NULL);
3858 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3859 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3861 /* For removing agent entry */
3862 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3863 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3866 * Recycle idle OI leaf may cause additional three OI blocks
3869 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3870 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3871 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3872 /* one less inode */
3873 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3874 i_projid_read(inode), -1, oh, obj, NULL,
3878 /* data to be truncated */
3879 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3880 i_projid_read(inode), 0, oh, obj, NULL,
3886 * will help to find FID->ino when this object is being
3889 rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3894 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3897 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3898 struct osd_object *obj = osd_dt_obj(dt);
3899 struct inode *inode = obj->oo_inode;
3900 struct osd_device *osd = osd_obj2dev(obj);
3901 struct osd_thandle *oh;
3906 oh = container_of(th, struct osd_thandle, ot_super);
3907 LASSERT(oh->ot_handle);
3909 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3911 if (unlikely(fid_is_acct(fid)))
3914 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3915 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3917 CERROR("%s: remove agent entry "DFID": rc = %d\n",
3918 osd_name(osd), PFID(fid), result);
3921 if (S_ISDIR(inode->i_mode)) {
3922 if (inode->i_nlink > 2)
3923 CERROR("%s: directory "DFID" ino %lu link count is %u at unlink. run e2fsck to repair\n",
3924 osd_name(osd), PFID(fid), inode->i_ino,
3927 spin_lock(&obj->oo_guard);
3929 spin_unlock(&obj->oo_guard);
3930 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3933 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3935 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3937 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3938 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3939 oh->ot_handle, OI_CHECK_FLD);
3941 osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3942 /* XXX: add to ext3 orphan list */
3943 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3945 /* not needed in the cache anymore */
3946 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3947 obj->oo_destroyed = 1;
3953 * Put the fid into lustre_mdt_attrs, and then place the structure
3954 * inode's ea. This fid should not be altered during the life time
3957 * \retval +ve, on success
3958 * \retval -ve, on error
3960 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3962 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3963 const struct lu_fid *fid, __u32 compat, __u32 incompat)
3965 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3966 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3971 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3974 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3977 lustre_loa_init(loa, fid, compat, incompat);
3978 lustre_loa_swab(loa, false);
3981 * For the OST device with 256 bytes inode size by default,
3982 * the PFID EA will be stored together with LMA EA to avoid
3983 * performance trouble. Otherwise the PFID EA can be stored
3984 * independently. LU-8998
3986 if ((compat & LMAC_FID_ON_OST) &&
3987 LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3988 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3989 sizeof(*loa), XATTR_CREATE);
3991 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3992 sizeof(*lma), XATTR_CREATE);
3994 * LMA may already exist, but we need to check that all the
3995 * desired compat/incompat flags have been added.
3997 if (unlikely(rc == -EEXIST)) {
3998 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3999 XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
4003 if (rc < sizeof(*lma))
4006 lustre_loa_swab(loa, true);
4007 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
4008 ((compat == 0 && incompat == 0) ||
4009 (!(~lma->lma_compat & compat) &&
4010 !(~lma->lma_incompat & incompat))))
4013 lma->lma_self_fid = *fid;
4014 lma->lma_compat |= compat;
4015 lma->lma_incompat |= incompat;
4016 if (rc == sizeof(*lma)) {
4017 lustre_lma_swab(lma);
4018 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4019 sizeof(*lma), XATTR_REPLACE);
4021 lustre_loa_swab(loa, false);
4022 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
4023 sizeof(*loa), XATTR_REPLACE);
4031 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
4032 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
4033 * To have compatilibility with 1.8 ldiskfs driver we need to have
4034 * magic number at start of fid data.
4035 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
4038 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
4039 const struct lu_fid *fid)
4041 if (!fid_is_namespace_visible(fid) ||
4042 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
4043 param->edp_magic = 0;
4047 param->edp_magic = LDISKFS_LUFID_MAGIC;
4048 param->edp_len = sizeof(struct lu_fid) + 1;
4049 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
4053 * Try to read the fid from inode ea into dt_rec.
4055 * \param fid object fid.
4057 * \retval 0 on success
4059 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
4060 __u32 ino, struct lu_fid *fid,
4061 struct osd_inode_id *id)
4063 struct osd_thread_info *info = osd_oti_get(env);
4064 struct inode *inode;
4068 osd_id_gen(id, ino, OSD_OII_NOGEN);
4069 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
4071 RETURN(PTR_ERR(inode));
4077 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
4079 struct inode *parent_dir,
4080 const struct lu_fid *dot_fid,
4081 const struct lu_fid *dot_dot_fid,
4082 struct osd_thandle *oth)
4084 struct ldiskfs_dentry_param *dot_ldp;
4085 struct ldiskfs_dentry_param *dot_dot_ldp;
4086 __u32 saved_nlink = dir->i_nlink;
4089 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_DOTDOT_ENOSPC))
4092 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
4093 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
4095 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
4096 dot_ldp->edp_magic = 0;
4098 rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
4099 dir, dot_ldp, dot_dot_ldp);
4101 * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
4102 * the subseqent ref_add() will increase the dir->i_nlink
4103 * as 3. That is incorrect for new created directory.
4105 * It looks like hack, because we want to make the OSD API
4106 * to be order-independent for new created directory object
4107 * between dt_insert(..) and ref_add() operations.
4109 * Here, we only restore the in-RAM dir-inode's nlink attr,
4110 * becuase if the nlink attr is not 2, then there will be
4111 * ref_add() called following the dt_insert(..), such call
4112 * will make both the in-RAM and on-disk dir-inode's nlink
4113 * attr to be set as 2. LU-7447
4115 set_nlink(dir, saved_nlink);
4120 * Create an local agent inode for remote entry
4122 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
4123 struct osd_device *osd,
4124 struct osd_object *pobj,
4125 const struct lu_fid *fid,
4129 struct osd_thread_info *info = osd_oti_get(env);
4130 struct inode *local;
4131 struct osd_thandle *oh;
4132 struct iattr iattr = {
4133 .ia_valid = ATTR_UID | ATTR_GID |
4134 ATTR_CTIME | ATTR_MTIME | ATTR_ATIME,
4135 .ia_ctime.tv_nsec = UTIME_OMIT,
4136 .ia_mtime.tv_nsec = UTIME_OMIT,
4137 .ia_atime.tv_nsec = UTIME_OMIT,
4138 .ia_uid = GLOBAL_ROOT_UID,
4139 .ia_gid = GLOBAL_ROOT_GID,
4146 oh = container_of(th, struct osd_thandle, ot_super);
4147 LASSERT(oh->ot_handle->h_transaction != NULL);
4149 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode,
4151 if (IS_ERR(local)) {
4152 CERROR("%s: create local error %d\n", osd_name(osd),
4153 (int)PTR_ERR(local));
4158 * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
4159 * correct gid on remote file, not agent here
4161 local->i_gid = current_fsgid();
4162 ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
4164 /* e2fsck doesn't like empty symlinks. Store remote FID as symlink.
4165 * That gives e2fsck something to look at and be happy, and allows
4166 * debugging if we need to determine where this symlink came from.
4168 if (S_ISLNK(type)) {
4169 BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1);
4170 ldiskfs_clear_inode_flag(local, LDISKFS_INODE_EXTENTS);
4171 rc = scnprintf((char *)LDISKFS_I(local)->i_data,
4172 LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
4174 i_size_write(local, rc);
4175 LDISKFS_I(local)->i_disksize = rc;
4177 unlock_new_inode(local);
4179 /* Agent inode should not have project ID */
4180 #ifdef HAVE_PROJECT_QUOTA
4181 if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
4182 i_projid_read(pobj->oo_inode) != 0) {
4183 rc = osd_transfer_project(local, 0, th);
4185 CERROR("%s: quota transfer failed:. Is project quota enforcement enabled on the ldiskfs filesystem? rc = %d\n",
4186 osd_ino2name(local), rc);
4187 RETURN(ERR_PTR(rc));
4191 /* Set special LMA flag for local agent inode */
4192 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
4194 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
4195 osd_name(osd), PFID(fid), rc);
4196 RETURN(ERR_PTR(rc));
4202 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode, fid,
4203 lu_object_fid(&pobj->oo_dt.do_lu),
4206 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
4207 osd_name(osd), PFID(fid), rc);
4208 RETURN(ERR_PTR(rc));
4215 * when direntry is deleted, we have to take care of possible agent inode
4216 * referenced by that. unfortunately we can't do this at that point:
4217 * iget() within a running transaction leads to deadlock and we better do
4218 * not call that every delete declaration to save performance. so we put
4219 * a potention agent inode on a list and process that once the transaction
4220 * is over. Notice it's not any worse in terms of real orphans as regular
4221 * object destroy doesn't put inodes on the on-disk orphan list. this should
4222 * be addressed separately
4224 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
4225 struct osd_thandle *oh,
4228 struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
4229 struct osd_obj_orphan *oor;
4236 oor->oor_env = (struct lu_env *)env;
4237 spin_lock(&osd->od_osfs_lock);
4238 list_add(&oor->oor_list, &osd->od_orphan_list);
4239 spin_unlock(&osd->od_osfs_lock);
4241 oh->ot_remove_agents = 1;
4247 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
4248 struct osd_device *osd)
4250 struct osd_thread_info *info = osd_oti_get(env);
4251 struct osd_obj_orphan *oor, *tmp;
4252 struct osd_inode_id id;
4254 struct inode *inode;
4259 spin_lock(&osd->od_osfs_lock);
4260 list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
4261 if (oor->oor_env == env)
4262 list_move(&oor->oor_list, &list);
4264 spin_unlock(&osd->od_osfs_lock);
4266 list_for_each_entry_safe(oor, tmp, &list, oor_list) {
4270 list_del(&oor->oor_list);
4273 osd_id_gen(&id, ino, OSD_OII_NOGEN);
4274 inode = osd_iget_fid(info, osd, &id, &fid);
4278 if (!osd_remote_fid(env, osd, &fid)) {
4283 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
4285 mark_inode_dirty(inode);
4286 ldiskfs_journal_stop(jh);
4294 * OSD layer object create function for OST objects (b=11826).
4296 * The FID is inserted into inode xattr here.
4298 * \retval 0, on success
4299 * \retval -ve, on error
4301 static int osd_create(const struct lu_env *env, struct dt_object *dt,
4302 struct lu_attr *attr, struct dt_allocation_hint *hint,
4303 struct dt_object_format *dof, struct thandle *th)
4305 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4306 struct osd_object *obj = osd_dt_obj(dt);
4307 struct osd_thread_info *info = osd_oti_get(env);
4308 int result, on_ost = 0;
4312 if (dt_object_exists(dt))
4315 LINVRNT(osd_invariant(obj));
4316 LASSERT(!dt_object_remote(dt));
4317 LASSERT(osd_is_write_locked(env, obj));
4318 LASSERT(th != NULL);
4320 if (unlikely(fid_is_acct(fid)))
4322 * Quota files can't be created from the kernel any more,
4323 * 'tune2fs -O quota' will take care of creating them
4327 result = __osd_create(info, obj, attr, hint, dof, th);
4329 if (fid_is_idif(fid) &&
4330 !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
4331 struct lu_fid *tfid = &info->oti_fid;
4332 struct ost_id *oi = &info->oti_ostid;
4334 fid_to_ostid(fid, oi);
4335 ostid_to_fid(tfid, oi, 0);
4337 result = osd_ea_fid_set(info, obj->oo_inode, tfid,
4338 LMAC_FID_ON_OST, 0);
4340 on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
4342 result = osd_ea_fid_set(info, obj->oo_inode, fid,
4343 on_ost ? LMAC_FID_ON_OST : 0,
4346 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
4347 obj->oo_dt.do_body_ops = &osd_body_ops;
4350 if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY)) {
4351 struct inode *inode = obj->oo_inode;
4353 result = __osd_oi_insert(env, obj, fid, th);
4354 if (result && inode) {
4355 spin_lock(&obj->oo_guard);
4357 spin_unlock(&obj->oo_guard);
4358 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4359 ldiskfs_set_inode_state(inode,
4360 LDISKFS_STATE_LUSTRE_DESTROY);
4362 obj->oo_inode = NULL;
4367 * a small optimization - dt_insert() isn't usually applied
4368 * to OST objects, so we don't need to cache OI mapping for
4371 if (result == 0 && on_ost == 0) {
4372 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4374 result = osd_idc_find_and_init(env, osd, obj);
4375 LASSERT(result == 0);
4378 LASSERT(ergo(result == 0,
4379 dt_object_exists(dt) && !dt_object_remote(dt)));
4380 LINVRNT(osd_invariant(obj));
4384 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4385 struct thandle *handle)
4387 struct osd_thandle *oh;
4390 /* it's possible that object doesn't exist yet */
4391 LASSERT(handle != NULL);
4393 oh = container_of(handle, struct osd_thandle, ot_super);
4394 LASSERT(oh->ot_handle == NULL);
4396 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4397 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4399 rc = osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev),
4406 * Concurrency: @dt is write locked.
4408 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4411 struct osd_object *obj = osd_dt_obj(dt);
4412 struct inode *inode = obj->oo_inode;
4413 struct osd_thandle *oh;
4416 if (!dt_object_exists(dt) || obj->oo_destroyed)
4419 LINVRNT(osd_invariant(obj));
4420 LASSERT(!dt_object_remote(dt));
4421 LASSERT(osd_is_write_locked(env, obj));
4422 LASSERT(th != NULL);
4424 oh = container_of(th, struct osd_thandle, ot_super);
4425 LASSERT(oh->ot_handle != NULL);
4427 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4429 CDEBUG(D_INODE, DFID" increase nlink %d\n",
4430 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4432 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4433 * (65000) subdirectories by storing "1" in i_nlink if the link count
4434 * would otherwise overflow. Directory tranversal tools understand
4435 * that (st_nlink == 1) indicates that the filesystem dose not track
4436 * hard links count on the directory, and will not abort subdirectory
4437 * scanning early once (st_nlink - 2) subdirs have been found.
4439 * This also has to properly handle the case of inodes with nlink == 0
4440 * in case they are being linked into the PENDING directory
4442 spin_lock(&obj->oo_guard);
4443 if (unlikely(inode->i_nlink == 0))
4444 /* inc_nlink from 0 may cause WARN_ON */
4445 set_nlink(inode, 1);
4447 osd_ldiskfs_inc_count(oh->ot_handle, inode);
4448 if (!S_ISDIR(inode->i_mode))
4449 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4451 spin_unlock(&obj->oo_guard);
4453 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4454 LINVRNT(osd_invariant(obj));
4456 osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4461 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4462 struct thandle *handle)
4464 struct osd_thandle *oh;
4466 if (!dt_object_exists(dt))
4469 LASSERT(!dt_object_remote(dt));
4470 LASSERT(handle != NULL);
4472 oh = container_of(handle, struct osd_thandle, ot_super);
4473 LASSERT(oh->ot_handle == NULL);
4475 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4476 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4482 * Concurrency: @dt is write locked.
4484 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4487 struct osd_object *obj = osd_dt_obj(dt);
4488 struct inode *inode = obj->oo_inode;
4489 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4490 struct osd_thandle *oh;
4492 if (!dt_object_exists(dt))
4495 LINVRNT(osd_invariant(obj));
4496 LASSERT(!dt_object_remote(dt));
4497 LASSERT(osd_is_write_locked(env, obj));
4498 LASSERT(th != NULL);
4500 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_REF_DEL))
4503 oh = container_of(th, struct osd_thandle, ot_super);
4504 LASSERT(oh->ot_handle != NULL);
4506 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4508 spin_lock(&obj->oo_guard);
4510 * That can be result of upgrade from old Lustre version and
4511 * applied only to local files. Just skip this ref_del call.
4512 * ext4_unlink() only treats this as a warning, don't LASSERT here.
4514 if (inode->i_nlink == 0) {
4515 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4516 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4517 ", maybe an upgraded file? (LU-3915)\n",
4518 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4519 spin_unlock(&obj->oo_guard);
4523 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4524 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4526 osd_ldiskfs_dec_count(oh->ot_handle, inode);
4527 spin_unlock(&obj->oo_guard);
4529 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4530 LINVRNT(osd_invariant(obj));
4532 osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4538 * Concurrency: @dt is read locked.
4540 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4541 struct lu_buf *buf, const char *name)
4543 struct osd_object *obj = osd_dt_obj(dt);
4544 struct inode *inode = obj->oo_inode;
4545 struct osd_thread_info *info = osd_oti_get(env);
4546 struct dentry *dentry = &info->oti_obj_dentry;
4547 bool cache_xattr = false;
4552 /* version get is not real XATTR but uses xattr API */
4553 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4554 dt_obj_version_t *ver = buf->lb_buf;
4557 * for version we are just using xattr API but change inode
4560 if (buf->lb_len == 0)
4561 return sizeof(dt_obj_version_t);
4563 if (buf->lb_len < sizeof(dt_obj_version_t))
4566 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4567 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4569 *ver = LDISKFS_I(inode)->i_fs_version;
4571 return sizeof(dt_obj_version_t);
4574 if (!dt_object_exists(dt))
4577 LASSERT(!dt_object_remote(dt));
4578 LASSERT(inode->i_op != NULL);
4579 #ifdef HAVE_IOP_XATTR
4580 LASSERT(inode->i_op->getxattr != NULL);
4583 if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4584 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4588 rc = osd_oxc_get(obj, name, buf);
4593 if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4594 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4595 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4596 struct filter_fid *ff;
4597 struct ost_layout *ol;
4599 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4603 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4606 if (buf->lb_len == 0 || !buf->lb_buf)
4609 if (buf->lb_len < rc)
4613 ol = &ff->ff_layout;
4614 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4615 PFID_STRIPE_IDX_BITS);
4616 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4617 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4618 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4619 if (lma->lma_compat & LMAC_COMP_INFO) {
4620 ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4621 ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4622 ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4624 ol->ol_comp_start = 0;
4625 ol->ol_comp_end = 0;
4629 /* Get enc context xattr directly from ldiskfs instead of going
4630 * through the VFS, as there is no xattr handler for
4633 if (strcmp(name, LL_XATTR_NAME_ENCRYPTION_CONTEXT) == 0)
4634 rc = ldiskfs_xattr_get(inode,
4635 LDISKFS_XATTR_INDEX_ENCRYPTION,
4636 LDISKFS_XATTR_NAME_ENCRYPTION_CONTEXT,
4637 buf->lb_buf, buf->lb_len);
4639 rc = __osd_xattr_get(inode, dentry, name,
4640 buf->lb_buf, buf->lb_len);
4644 if (rc == -ENOENT || rc == -ENODATA)
4645 osd_oxc_add(obj, name, NULL, 0);
4646 else if (rc > 0 && buf->lb_buf != NULL)
4647 osd_oxc_add(obj, name, buf->lb_buf, rc);
4653 static int osd_declare_xattr_set(const struct lu_env *env,
4654 struct dt_object *dt,
4655 const struct lu_buf *buf, const char *name,
4656 int fl, struct thandle *handle)
4658 struct osd_thandle *oh;
4660 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4662 LASSERT(handle != NULL);
4664 oh = container_of(handle, struct osd_thandle, ot_super);
4665 LASSERT(oh->ot_handle == NULL);
4667 if (strcmp(name, XATTR_NAME_LMA) == 0) {
4669 * For non-upgrading case, the LMA is set first and
4670 * usually fit inode. But for upgrade case, the LMA
4671 * may be in another separated EA block.
4673 if (dt_object_exists(dt)) {
4674 if (fl == LU_XATTR_REPLACE)
4679 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4681 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4682 /* We may need to delete the old PFID EA. */
4683 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4684 if (fl == LU_XATTR_REPLACE)
4690 * If some name entry resides on remote MDT, then will create
4691 * agent entry under remote parent. On the other hand, if the
4692 * remote entry will be removed, then related agent entry may
4693 * need to be removed from the remote parent. So there may be
4694 * kinds of cases, let's declare enough credits. The credits
4695 * for create agent entry is enough for remove case.
4697 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4698 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4699 if (dt_object_exists(dt))
4700 credits += 1; /* For updating LMA */
4704 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4709 if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4711 * learn xattr size from osd_xattr_get if
4712 * attribute has not been read yet
4714 buflen = __osd_xattr_get(
4715 osd_dt_obj(dt)->oo_inode,
4716 &osd_oti_get(env)->oti_obj_dentry,
4721 buflen = buf->lb_len;
4724 if (buflen > sb->s_blocksize) {
4725 credits += osd_calc_bkmap_credits(
4727 (buflen + sb->s_blocksize - 1) >>
4728 sb->s_blocksize_bits);
4732 * xattr set may involve inode quota change, reserve credits for
4733 * dquot_initialize()
4735 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4738 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4743 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4744 const struct lu_buf *buf, int fl,
4745 struct thandle *handle)
4747 struct osd_thread_info *info = osd_oti_get(env);
4748 struct dentry *dentry = &info->oti_obj_dentry;
4749 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4750 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4751 struct inode *inode = obj->oo_inode;
4752 struct filter_fid *ff = buf->lb_buf;
4753 struct ost_layout *ol = &ff->ff_layout;
4754 int flags = XATTR_REPLACE;
4759 if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4762 rc = osd_get_lma(info, inode, dentry, loa);
4763 if (rc == -ENODATA) {
4764 /* Usually for upgarding from old device */
4765 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4766 LMAC_FID_ON_OST, 0);
4767 flags = XATTR_CREATE;
4772 if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4773 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4776 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4777 /* Separate PFID EA from LMA */
4778 lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4779 lustre_lma_swab(lma);
4780 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4781 sizeof(*lma), XATTR_REPLACE);
4783 obj->oo_pfid_in_lma = 0;
4784 rc = LU_XATTR_CREATE;
4790 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4794 * Old client does not send stripe information,
4795 * then store the PFID EA on disk separatedly.
4797 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4798 ol->ol_stripe_size == 0))
4801 /* Remove old PFID EA entry firstly. */
4802 dquot_initialize(inode);
4803 rc = ll_vfs_removexattr(dentry, inode, XATTR_NAME_FID);
4804 if (rc == -ENODATA) {
4805 /* XATTR_NAME_FID is already absent */
4812 fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4813 if (likely(ol->ol_stripe_size != 0)) {
4814 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4815 PFID_STRIPE_IDX_BITS;
4816 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4817 lma->lma_compat |= LMAC_STRIPE_INFO;
4818 if (ol->ol_comp_id != 0) {
4819 loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4820 loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4821 loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4822 lma->lma_compat |= LMAC_COMP_INFO;
4826 lustre_loa_swab(loa, false);
4828 /* Store the PFID EA inside LMA. */
4829 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4832 obj->oo_pfid_in_lma = 1;
4838 * In DNE environment, the object (in spite of regular file or directory)
4839 * and its name entry may reside on different MDTs. Under such case, we will
4840 * create an agent entry on the MDT where the object resides. The agent entry
4841 * references the object locally, that makes the object to be visible to the
4842 * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4843 * such as 'tar' can handle the object properly.
4845 * We handle the agent entry during set linkEA that is the common interface
4846 * for both regular file and directroy, can handle kinds of cases, such as
4847 * create/link/unlink/rename, and so on.
4849 * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4851 * XXX: There are two known issues:
4852 * 1. For one object, we will create at most one agent entry even if there
4853 * may be more than one cross-MDTs hard links on the object. So the local
4854 * e2fsck may claim that the object's nlink is larger than the name entries
4855 * that reference such inode. And in further, the e2fsck will fix the nlink
4856 * attribute to match the local references. Then it will cause the object's
4857 * nlink attribute to be inconsistent with the global references. it is bad
4858 * but not fatal. The ref_del() can handle the zero-referenced case. On the
4859 * other hand, the global namespace LFSCK can repair the object's attribute
4860 * according to the linkEA.
4861 * 2. There may be too many hard links on the object as to its linkEA overflow,
4862 * then the linkEA entry for cross-MDTs reference may be discarded. If such
4863 * case happened, then at this point, we do not know whether there are some
4864 * cross-MDTs reference. But there are local references, it guarantees that
4865 * object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4867 static int osd_xattr_handle_linkea(const struct lu_env *env,
4868 struct osd_device *osd,
4869 struct osd_object *obj,
4870 const struct lu_buf *buf,
4871 struct thandle *handle)
4873 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4874 struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4875 struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4876 struct lu_name tmpname;
4877 struct osd_thandle *oh;
4879 bool remote = false;
4883 oh = container_of(handle, struct osd_thandle, ot_super);
4884 LASSERT(oh->ot_handle != NULL);
4886 rc = linkea_init_with_rec(&ldata);
4888 linkea_first_entry(&ldata);
4889 while (ldata.ld_lee != NULL && !remote) {
4890 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4892 if (osd_remote_fid(env, osd, tfid) > 0)
4895 linkea_next_entry(&ldata);
4897 } else if (rc == -ENODATA) {
4903 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4904 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4906 CERROR("%s: failed to remove agent entry for "DFID
4907 ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4908 } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4909 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4911 CERROR("%s: failed to create agent entry for "DFID
4912 ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4919 * Concurrency: @dt is write locked.
4921 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4922 const struct lu_buf *buf, const char *name, int fl,
4923 struct thandle *handle)
4925 struct osd_object *obj = osd_dt_obj(dt);
4926 struct osd_device *osd = osd_obj2dev(obj);
4927 struct inode *inode = obj->oo_inode;
4928 struct osd_thread_info *info = osd_oti_get(env);
4938 /* version set is not real XATTR */
4939 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4940 dt_obj_version_t *version = buf->lb_buf;
4943 * for version we are just using xattr API but change inode
4946 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4948 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4949 *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4951 LDISKFS_I(inode)->i_fs_version = *version;
4953 * Version is set after all inode operations are finished,
4954 * so we should mark it dirty here
4956 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4961 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4962 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4965 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4968 * For the OST device with 256 bytes inode size by default,
4969 * the PFID EA will be stored together with LMA EA to avoid
4970 * performance trouble. Otherwise the PFID EA can be stored
4971 * independently. LU-8998
4973 if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4974 (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4975 LASSERT(buf->lb_buf);
4977 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4980 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4981 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4982 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4984 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4988 lma->lma_incompat |= LMAI_STRIPED;
4989 lustre_lma_swab(lma);
4990 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4991 sizeof(*lma), XATTR_REPLACE);
4994 } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4995 LASSERT(!osd->od_is_ost);
4997 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
5002 if (fl & LU_XATTR_REPLACE)
5003 fs_flags |= XATTR_REPLACE;
5005 if (fl & LU_XATTR_CREATE)
5006 fs_flags |= XATTR_CREATE;
5008 if (strcmp(name, LL_XATTR_NAME_ENCRYPTION_CONTEXT) == 0) {
5009 /* Set enc context xattr directly in ldiskfs instead of going
5010 * through the VFS, as there is no xattr handler for
5013 struct osd_thandle *oth = container_of(handle,
5017 if (!oth->ot_handle)
5018 /* this should be already part of a transaction */
5021 rc = ldiskfs_xattr_set_handle(oth->ot_handle, inode,
5022 LDISKFS_XATTR_INDEX_ENCRYPTION,
5023 LDISKFS_XATTR_NAME_ENCRYPTION_CONTEXT,
5024 buf->lb_buf, len, fs_flags);
5026 rc = __osd_xattr_set(info, inode, name,
5027 buf->lb_buf, len, fs_flags);
5029 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
5032 (strcmp(name, XATTR_NAME_LOV) == 0 ||
5033 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
5034 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
5040 * Concurrency: @dt is read locked.
5042 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
5043 const struct lu_buf *buf)
5045 struct osd_object *obj = osd_dt_obj(dt);
5046 struct osd_device *dev = osd_obj2dev(obj);
5047 struct inode *inode = obj->oo_inode;
5048 struct osd_thread_info *info = osd_oti_get(env);
5049 struct dentry *dentry = &info->oti_obj_dentry;
5052 if (!dt_object_exists(dt))
5055 LASSERT(!dt_object_remote(dt));
5056 LASSERT(inode->i_op != NULL);
5057 LASSERT(inode->i_op->listxattr != NULL);
5059 dentry->d_inode = inode;
5060 dentry->d_sb = inode->i_sb;
5061 rc = inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
5063 if (rc < 0 || buf->lb_buf == NULL)
5066 /* Hide virtual project ID xattr from list if disabled */
5067 if (!dev->od_enable_projid_xattr) {
5068 char *end = (char *)buf->lb_buf + rc;
5069 char *p = buf->lb_buf;
5072 char *next = p + strlen(p) + 1;
5074 if (strcmp(p, XATTR_NAME_PROJID) == 0) {
5076 memmove(p, next, end - next);
5088 static int osd_declare_xattr_del(const struct lu_env *env,
5089 struct dt_object *dt, const char *name,
5090 struct thandle *handle)
5092 struct osd_thandle *oh;
5093 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
5095 LASSERT(!dt_object_remote(dt));
5096 LASSERT(handle != NULL);
5098 oh = container_of(handle, struct osd_thandle, ot_super);
5099 LASSERT(oh->ot_handle == NULL);
5101 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
5102 osd_dto_credits_noquota[DTO_XATTR_SET]);
5104 * xattr del may involve inode quota change, reserve credits for
5105 * dquot_initialize()
5107 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
5113 * Concurrency: @dt is write locked.
5115 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
5116 const char *name, struct thandle *handle)
5118 struct osd_object *obj = osd_dt_obj(dt);
5119 struct inode *inode = obj->oo_inode;
5120 struct osd_thread_info *info = osd_oti_get(env);
5121 struct dentry *dentry = &info->oti_obj_dentry;
5124 if (!dt_object_exists(dt))
5127 LASSERT(!dt_object_remote(dt));
5128 LASSERT(inode->i_op != NULL);
5129 LASSERT(handle != NULL);
5130 #ifdef HAVE_IOP_XATTR
5131 LASSERT(inode->i_op->removexattr != NULL);
5134 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
5136 if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
5137 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
5139 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
5140 &info->oti_ost_attrs);
5142 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
5144 lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
5145 lustre_lma_swab(lma);
5146 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
5147 sizeof(*lma), XATTR_REPLACE);
5149 obj->oo_pfid_in_lma = 0;
5152 dquot_initialize(inode);
5153 dentry->d_inode = inode;
5154 dentry->d_sb = inode->i_sb;
5155 rc = ll_vfs_removexattr(dentry, inode, name);
5158 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
5161 (strcmp(name, XATTR_NAME_LOV) == 0 ||
5162 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
5163 osd_oxc_del(obj, name);
5168 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
5169 __u64 start, __u64 end)
5171 struct osd_object *obj = osd_dt_obj(dt);
5172 struct osd_device *dev = osd_obj2dev(obj);
5173 struct inode *inode = obj->oo_inode;
5178 file = alloc_file_pseudo(inode, dev->od_mnt, "/", O_NOATIME,
5181 RETURN(PTR_ERR(file));
5183 file->f_mode |= FMODE_64BITHASH;
5184 rc = vfs_fsync_range(file, start, end, 0);
5191 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
5196 static bool osd_check_stale(struct dt_object *dt)
5204 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
5205 const struct dt_index_features *feat)
5207 struct iam_descr *descr;
5209 if (osd_object_is_root(o))
5210 return feat == &dt_directory_features;
5212 LASSERT(o->oo_dir != NULL);
5214 descr = o->oo_dir->od_container.ic_descr;
5215 if (feat == &dt_directory_features) {
5216 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
5221 return feat->dif_keysize_min <= descr->id_key_size &&
5222 descr->id_key_size <= feat->dif_keysize_max &&
5223 feat->dif_recsize_min <= descr->id_rec_size &&
5224 descr->id_rec_size <= feat->dif_recsize_max &&
5225 !(feat->dif_flags & (DT_IND_VARKEY |
5226 DT_IND_VARREC | DT_IND_NONUNQ)) &&
5227 ergo(feat->dif_flags & DT_IND_UPDATE,
5228 1 /* XXX check that object (and fs) is writable */);
5232 static int osd_iam_container_init(const struct lu_env *env,
5233 struct osd_object *obj,
5234 struct osd_directory *dir)
5236 struct iam_container *bag = &dir->od_container;
5239 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
5243 result = iam_container_setup(bag);
5245 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
5247 iam_container_fini(bag);
5254 * Concurrency: no external locking is necessary.
5256 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
5257 const struct dt_index_features *feat)
5261 struct osd_object *obj = osd_dt_obj(dt);
5263 LINVRNT(osd_invariant(obj));
5265 if (osd_object_is_root(obj)) {
5266 dt->do_index_ops = &osd_index_ea_ops;
5268 } else if (feat == &dt_directory_features) {
5269 dt->do_index_ops = &osd_index_ea_ops;
5270 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
5275 } else if (unlikely(feat == &dt_otable_features)) {
5276 dt->do_index_ops = &osd_otable_ops;
5278 } else if (unlikely(feat == &dt_acct_features)) {
5279 dt->do_index_ops = &osd_acct_index_ops;
5282 } else if (!osd_has_index(obj)) {
5283 struct osd_directory *dir;
5284 struct osd_device *osd = osd_obj2dev(obj);
5285 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
5290 spin_lock(&obj->oo_guard);
5291 if (obj->oo_dir == NULL)
5295 * Concurrent thread allocated container data.
5298 spin_unlock(&obj->oo_guard);
5300 * Now, that we have container data, serialize its
5303 down_write(&obj->oo_ext_idx_sem);
5305 * recheck under lock.
5308 if (osd_has_index(obj)) {
5313 result = osd_iam_container_init(env, obj, obj->oo_dir);
5314 if (result || feat == &dt_lfsck_namespace_features ||
5315 feat == &dt_lfsck_layout_orphan_features ||
5316 feat == &dt_lfsck_layout_dangling_features)
5319 result = osd_index_register(osd, fid,
5320 feat->dif_keysize_max,
5321 feat->dif_recsize_max);
5323 CWARN("%s: failed to register index "
5325 osd_name(osd), PFID(fid), result);
5326 else if (result > 0)
5329 CDEBUG(D_LFSCK, "%s: index object "DFID
5330 " (%d/%d) registered\n",
5331 osd_name(osd), PFID(fid),
5332 (int)feat->dif_keysize_max,
5333 (int)feat->dif_recsize_max);
5336 up_write(&obj->oo_ext_idx_sem);
5344 if (result == 0 && skip_iam == 0) {
5345 if (!osd_iam_index_probe(env, obj, feat))
5348 LINVRNT(osd_invariant(obj));
5353 static int osd_otable_it_attr_get(const struct lu_env *env,
5354 struct dt_object *dt,
5355 struct lu_attr *attr)
5361 static const struct dt_object_operations osd_obj_ops = {
5362 .do_read_lock = osd_read_lock,
5363 .do_write_lock = osd_write_lock,
5364 .do_read_unlock = osd_read_unlock,
5365 .do_write_unlock = osd_write_unlock,
5366 .do_write_locked = osd_write_locked,
5367 .do_attr_get = osd_attr_get,
5368 .do_declare_attr_set = osd_declare_attr_set,
5369 .do_attr_set = osd_attr_set,
5370 .do_ah_init = osd_ah_init,
5371 .do_declare_create = osd_declare_create,
5372 .do_create = osd_create,
5373 .do_declare_destroy = osd_declare_destroy,
5374 .do_destroy = osd_destroy,
5375 .do_index_try = osd_index_try,
5376 .do_declare_ref_add = osd_declare_ref_add,
5377 .do_ref_add = osd_ref_add,
5378 .do_declare_ref_del = osd_declare_ref_del,
5379 .do_ref_del = osd_ref_del,
5380 .do_xattr_get = osd_xattr_get,
5381 .do_declare_xattr_set = osd_declare_xattr_set,
5382 .do_xattr_set = osd_xattr_set,
5383 .do_declare_xattr_del = osd_declare_xattr_del,
5384 .do_xattr_del = osd_xattr_del,
5385 .do_xattr_list = osd_xattr_list,
5386 .do_object_sync = osd_object_sync,
5387 .do_invalidate = osd_invalidate,
5388 .do_check_stale = osd_check_stale,
5391 static const struct dt_object_operations osd_obj_otable_it_ops = {
5392 .do_attr_get = osd_otable_it_attr_get,
5393 .do_index_try = osd_index_try,
5396 static int osd_index_declare_iam_delete(const struct lu_env *env,
5397 struct dt_object *dt,
5398 const struct dt_key *key,
5399 struct thandle *handle)
5401 struct osd_thandle *oh;
5403 oh = container_of(handle, struct osd_thandle, ot_super);
5404 LASSERT(oh->ot_handle == NULL);
5406 /* Recycle may cause additional three blocks to be changed. */
5407 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
5408 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
5414 * delete a (key, value) pair from index \a dt specified by \a key
5416 * \param dt osd index object
5417 * \param key key for index
5418 * \param rec record reference
5419 * \param handle transaction handler
5422 * \retval -ve failure
5424 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
5425 const struct dt_key *key,
5426 struct thandle *handle)
5428 struct osd_thread_info *oti = osd_oti_get(env);
5429 struct osd_object *obj = osd_dt_obj(dt);
5430 struct osd_thandle *oh;
5431 struct iam_path_descr *ipd;
5432 struct iam_container *bag = &obj->oo_dir->od_container;
5437 if (!dt_object_exists(dt))
5440 LINVRNT(osd_invariant(obj));
5441 LASSERT(!dt_object_remote(dt));
5442 LASSERT(bag->ic_object == obj->oo_inode);
5443 LASSERT(handle != NULL);
5445 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5447 ipd = osd_idx_ipd_get(env, bag);
5448 if (unlikely(ipd == NULL))
5451 oh = container_of(handle, struct osd_thandle, ot_super);
5452 LASSERT(oh->ot_handle != NULL);
5453 LASSERT(oh->ot_handle->h_transaction != NULL);
5455 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5456 /* swab quota uid/gid provided by caller */
5457 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5458 key = (const struct dt_key *)&oti->oti_quota_id;
5461 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5462 osd_ipd_put(env, bag, ipd);
5463 LINVRNT(osd_invariant(obj));
5464 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5468 static int osd_index_declare_ea_delete(const struct lu_env *env,
5469 struct dt_object *dt,
5470 const struct dt_key *key,
5471 struct thandle *handle)
5473 struct osd_thandle *oh;
5474 struct inode *inode;
5479 LASSERT(!dt_object_remote(dt));
5480 LASSERT(handle != NULL);
5482 oh = container_of(handle, struct osd_thandle, ot_super);
5483 LASSERT(oh->ot_handle == NULL);
5485 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5486 osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5488 inode = osd_dt_obj(dt)->oo_inode;
5492 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5493 i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5498 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5501 struct osd_fid_pack *rec;
5504 if (de->file_type & LDISKFS_DIRENT_LUFID) {
5505 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5506 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5507 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5513 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5514 const struct lu_fid *fid)
5516 struct seq_server_site *ss = osd_seq_site(osd);
5520 /* FID seqs not in FLDB, must be local seq */
5521 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5525 * If FLD is not being initialized yet, it only happens during the
5526 * initialization, likely during mgs initialization, and we assume
5527 * this is local FID.
5529 if (ss == NULL || ss->ss_server_fld == NULL)
5532 /* Only check the local FLDB here */
5533 if (osd_seq_exists(env, osd, fid_seq(fid)))
5539 static void osd_take_care_of_agent(const struct lu_env *env,
5540 struct osd_device *osd,
5541 struct osd_thandle *oh,
5542 struct ldiskfs_dir_entry_2 *de)
5544 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5545 struct osd_idmap_cache *idc;
5546 int rc, schedule = 0;
5548 LASSERT(de != NULL);
5550 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5551 if (likely(rc == 0)) {
5552 idc = osd_idc_find_or_init(env, osd, fid);
5553 if (IS_ERR(idc) || idc->oic_remote)
5555 } else if (rc == -ENODATA) {
5557 * can't get FID, postpone to the end of the
5558 * transaction when iget() is safe
5562 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5565 osd_schedule_agent_inode_removal(env, oh,
5566 le32_to_cpu(de->inode));
5570 * Utility function to get real name from object name
5572 * \param[in] obj pointer to the object to be handled
5573 * \param[in] name object name
5574 * \param[in] len object name len
5575 * \param[out]ln pointer to the struct lu_name to hold the real name
5577 * If file is not encrypted, real name is just the object name.
5578 * If file is encrypted, object name needs to be decoded. In
5579 * this case a new buffer is allocated, and ln->ln_name needs to be freed by
5582 * \retval 0, on success
5583 * \retval -ve, on error
5585 static int obj_name2lu_name(struct osd_object *obj, const char *name,
5586 int len, struct lu_name *ln)
5588 if (!(obj->oo_lma_flags & LUSTRE_ENCRYPT_FL)) {
5590 ln->ln_namelen = len;
5592 char *buf = kmalloc(len, GFP_NOFS);
5597 len = critical_decode(name, len, buf);
5599 ln->ln_namelen = len;
5606 * Index delete function for interoperability mode (b11826).
5607 * It will remove the directory entry added by osd_index_ea_insert().
5608 * This entry is needed to maintain name->fid mapping.
5610 * \param key, key i.e. file entry to be deleted
5612 * \retval 0, on success
5613 * \retval -ve, on error
5615 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5616 const struct dt_key *key, struct thandle *handle)
5618 struct osd_object *obj = osd_dt_obj(dt);
5619 struct inode *dir = obj->oo_inode;
5620 struct dentry *dentry;
5621 struct osd_thandle *oh;
5622 struct ldiskfs_dir_entry_2 *de = NULL;
5623 struct buffer_head *bh;
5624 struct htree_lock *hlock = NULL;
5625 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5631 if (!dt_object_exists(dt))
5634 LINVRNT(osd_invariant(obj));
5635 LASSERT(!dt_object_remote(dt));
5636 LASSERT(handle != NULL);
5638 rc = obj_name2lu_name(obj, (char *)key, strlen((char *)key), &ln);
5642 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5644 oh = container_of(handle, struct osd_thandle, ot_super);
5645 LASSERT(oh->ot_handle != NULL);
5646 LASSERT(oh->ot_handle->h_transaction != NULL);
5648 dquot_initialize(dir);
5649 dentry = osd_child_dentry_get(env, obj, ln.ln_name, ln.ln_namelen);
5651 if (obj->oo_hl_head != NULL) {
5652 hlock = osd_oti_get(env)->oti_hlock;
5653 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5654 dir, LDISKFS_HLOCK_DEL);
5656 down_write(&obj->oo_ext_idx_sem);
5659 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5662 * If this is not the ".." entry, it might be a remote DNE
5663 * entry and we need to check if the FID is for a remote
5664 * MDT. If the FID is not in the directory entry (e.g.
5665 * upgraded 1.8 filesystem without dirdata enabled) then
5666 * we need to get the FID from the LMA. For a remote directory
5667 * there HAS to be an LMA, it cannot be an IGIF inode in this
5670 * Delete the entry before the agent inode in order to
5671 * simplify error handling. At worst an error after deleting
5672 * the entry first might leak the agent inode afterward. The
5673 * reverse would need filesystem abort in case of error deleting
5674 * the entry after the agent had been removed, or leave a
5675 * dangling entry pointing at a random inode.
5677 if (strcmp((char *)key, dotdot) != 0)
5678 osd_take_care_of_agent(env, osd, oh, de);
5679 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5685 if (!rc && fid_is_namespace_visible(lu_object_fid(&dt->do_lu)) &&
5686 obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
5687 /* NB, dirent count may not be accurate, because it's counted
5690 if (obj->oo_dirent_count)
5691 obj->oo_dirent_count--;
5693 obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
5696 ldiskfs_htree_unlock(hlock);
5698 up_write(&obj->oo_ext_idx_sem);
5701 LASSERT(osd_invariant(obj));
5702 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5703 if (ln.ln_name != (char *)key)
5709 * Lookup index for \a key and copy record to \a rec.
5711 * \param dt osd index object
5712 * \param key key for index
5713 * \param rec record reference
5715 * \retval +ve success : exact mach
5716 * \retval 0 return record with key not greater than \a key
5717 * \retval -ve failure
5719 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5720 struct dt_rec *rec, const struct dt_key *key)
5722 struct osd_object *obj = osd_dt_obj(dt);
5723 struct iam_path_descr *ipd;
5724 struct iam_container *bag = &obj->oo_dir->od_container;
5725 struct osd_thread_info *oti = osd_oti_get(env);
5726 struct iam_iterator *it = &oti->oti_idx_it;
5727 struct iam_rec *iam_rec;
5732 if (!dt_object_exists(dt))
5735 LASSERT(osd_invariant(obj));
5736 LASSERT(!dt_object_remote(dt));
5737 LASSERT(bag->ic_object == obj->oo_inode);
5739 ipd = osd_idx_ipd_get(env, bag);
5743 /* got ipd now we can start iterator. */
5744 iam_it_init(it, bag, 0, ipd);
5746 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5747 /* swab quota uid/gid provided by caller */
5748 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5749 key = (const struct dt_key *)&oti->oti_quota_id;
5752 rc = iam_it_get(it, (struct iam_key *)key);
5754 if (S_ISDIR(obj->oo_inode->i_mode))
5755 iam_rec = (struct iam_rec *)oti->oti_ldp;
5757 iam_rec = (struct iam_rec *)rec;
5759 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5761 if (S_ISDIR(obj->oo_inode->i_mode))
5762 osd_fid_unpack((struct lu_fid *)rec,
5763 (struct osd_fid_pack *)iam_rec);
5764 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5765 osd_quota_unpack(obj, rec);
5770 osd_ipd_put(env, bag, ipd);
5772 LINVRNT(osd_invariant(obj));
5777 static int osd_index_declare_iam_insert(const struct lu_env *env,
5778 struct dt_object *dt,
5779 const struct dt_rec *rec,
5780 const struct dt_key *key,
5781 struct thandle *handle)
5783 struct osd_thandle *oh;
5785 LASSERT(handle != NULL);
5787 oh = container_of(handle, struct osd_thandle, ot_super);
5788 LASSERT(oh->ot_handle == NULL);
5790 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5791 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5797 * Inserts (key, value) pair in \a dt index object.
5799 * \param dt osd index object
5800 * \param key key for index
5801 * \param rec record reference
5802 * \param th transaction handler
5805 * \retval -ve failure
5807 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5808 const struct dt_rec *rec,
5809 const struct dt_key *key, struct thandle *th)
5811 struct osd_object *obj = osd_dt_obj(dt);
5812 struct iam_path_descr *ipd;
5813 struct osd_thandle *oh;
5814 struct iam_container *bag;
5815 struct osd_thread_info *oti = osd_oti_get(env);
5816 struct iam_rec *iam_rec;
5821 if (!dt_object_exists(dt))
5824 LINVRNT(osd_invariant(obj));
5825 LASSERT(!dt_object_remote(dt));
5827 bag = &obj->oo_dir->od_container;
5828 LASSERT(bag->ic_object == obj->oo_inode);
5829 LASSERT(th != NULL);
5831 osd_trans_exec_op(env, th, OSD_OT_INSERT);
5833 ipd = osd_idx_ipd_get(env, bag);
5834 if (unlikely(ipd == NULL))
5837 oh = container_of(th, struct osd_thandle, ot_super);
5838 LASSERT(oh->ot_handle != NULL);
5839 LASSERT(oh->ot_handle->h_transaction != NULL);
5840 if (S_ISDIR(obj->oo_inode->i_mode)) {
5841 iam_rec = (struct iam_rec *)oti->oti_ldp;
5842 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5844 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5845 /* pack quota uid/gid */
5846 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5847 key = (const struct dt_key *)&oti->oti_quota_id;
5848 /* pack quota record */
5849 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5850 iam_rec = (struct iam_rec *)rec;
5852 iam_rec = (struct iam_rec *)rec;
5855 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5857 osd_ipd_put(env, bag, ipd);
5858 LINVRNT(osd_invariant(obj));
5859 osd_trans_exec_check(env, th, OSD_OT_INSERT);
5864 * Calls ldiskfs_add_entry() to add directory entry
5865 * into the directory. This is required for
5866 * interoperability mode (b11826)
5868 * \retval 0, on success
5869 * \retval -ve, on error
5871 static int __osd_ea_add_rec(struct osd_thread_info *info,
5872 struct osd_object *pobj, struct inode *cinode,
5873 const char *name, const struct lu_fid *fid,
5874 struct htree_lock *hlock, struct thandle *th)
5876 struct ldiskfs_dentry_param *ldp;
5877 struct dentry *child;
5878 struct osd_thandle *oth;
5882 oth = container_of(th, struct osd_thandle, ot_super);
5883 LASSERT(oth->ot_handle != NULL);
5884 LASSERT(oth->ot_handle->h_transaction != NULL);
5885 LASSERT(pobj->oo_inode);
5887 rc = obj_name2lu_name(pobj, name, strlen(name), &ln);
5891 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5892 if (unlikely(osd_object_is_root(pobj)))
5895 osd_get_ldiskfs_dirent_param(ldp, fid);
5896 child = osd_child_dentry_get(info->oti_env, pobj,
5897 ln.ln_name, ln.ln_namelen);
5898 child->d_fsdata = (void *)ldp;
5899 dquot_initialize(pobj->oo_inode);
5900 rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5901 child, cinode, hlock);
5902 if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5903 struct ldiskfs_dir_entry_2 *de;
5904 struct buffer_head *bh;
5907 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5910 rc1 = osd_ldiskfs_journal_get_write_access(
5911 oth->ot_handle, pobj->oo_inode->i_sb, bh,
5914 if (S_ISDIR(cinode->i_mode))
5915 de->file_type = LDISKFS_DIRENT_LUFID |
5916 LDISKFS_FT_REG_FILE;
5918 de->file_type = LDISKFS_DIRENT_LUFID |
5920 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5927 if (ln.ln_name != name)
5933 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5934 * into the directory.Also sets flags into osd object to
5935 * indicate dot and dotdot are created. This is required for
5936 * interoperability mode (b11826)
5938 * \param dir directory for dot and dotdot fixup.
5939 * \param obj child object for linking
5941 * \retval 0, on success
5942 * \retval -ve, on error
5944 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5945 struct osd_object *dir,
5946 struct inode *parent_dir, const char *name,
5947 const struct lu_fid *dot_fid,
5948 const struct lu_fid *dot_dot_fid,
5951 struct inode *inode = dir->oo_inode;
5952 struct osd_thandle *oth;
5955 oth = container_of(th, struct osd_thandle, ot_super);
5956 LASSERT(oth->ot_handle->h_transaction != NULL);
5957 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5959 if (strcmp(name, dot) == 0) {
5960 if (dir->oo_compat_dot_created) {
5963 LASSERT(inode->i_ino == parent_dir->i_ino);
5964 dir->oo_compat_dot_created = 1;
5967 } else if (strcmp(name, dotdot) == 0) {
5968 if (!dir->oo_compat_dot_created)
5970 /* in case of rename, dotdot is already created */
5971 if (dir->oo_compat_dotdot_created) {
5972 return __osd_ea_add_rec(info, dir, parent_dir, name,
5973 dot_dot_fid, NULL, th);
5976 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5977 struct lu_fid tfid = *dot_dot_fid;
5980 result = osd_add_dot_dotdot_internal(info,
5981 dir->oo_inode, parent_dir, dot_fid,
5984 result = osd_add_dot_dotdot_internal(info,
5985 dir->oo_inode, parent_dir, dot_fid,
5990 dir->oo_compat_dotdot_created = 1;
5998 * It will call the appropriate osd_add* function and return the
5999 * value, return by respective functions.
6001 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
6002 struct inode *cinode, const char *name,
6003 const struct lu_fid *fid, struct thandle *th)
6005 struct osd_thread_info *info = osd_oti_get(env);
6006 struct htree_lock *hlock;
6009 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
6011 if (name[0] == '.' && (name[1] == '\0' ||
6012 (name[1] == '.' && name[2] == '\0'))) {
6013 if (hlock != NULL) {
6014 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
6017 down_write(&pobj->oo_ext_idx_sem);
6020 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
6021 lu_object_fid(&pobj->oo_dt.do_lu),
6024 if (hlock != NULL) {
6025 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
6026 pobj->oo_inode, LDISKFS_HLOCK_ADD);
6028 down_write(&pobj->oo_ext_idx_sem);
6031 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
6032 struct lu_fid *tfid = &info->oti_fid;
6036 rc = __osd_ea_add_rec(info, pobj, cinode, name,
6039 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
6043 if (!rc && fid_is_namespace_visible(lu_object_fid(&pobj->oo_dt.do_lu))
6044 && pobj->oo_dirent_count != LU_DIRENT_COUNT_UNSET)
6045 pobj->oo_dirent_count++;
6048 ldiskfs_htree_unlock(hlock);
6050 up_write(&pobj->oo_ext_idx_sem);
6056 osd_ldiskfs_consistency_check(struct osd_thread_info *oti,
6057 struct osd_device *dev,
6058 const struct lu_fid *fid,
6059 struct osd_inode_id *id)
6061 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
6062 struct inode *inode = NULL;
6068 if (!scrub_needs_check(scrub, fid, id->oii_ino))
6071 rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
6072 if (rc == -ENOENT) {
6073 __u32 gen = id->oii_gen;
6079 inode = osd_iget(oti, dev, id);
6080 /* The inode has been removed (by race maybe). */
6081 if (IS_ERR(inode)) {
6082 rc = PTR_ERR(inode);
6084 RETURN(rc == -ESTALE ? -ENOENT : rc);
6087 /* The OI mapping is lost. */
6088 if (gen != OSD_OII_NOGEN)
6092 * The inode may has been reused by others, we do not know,
6093 * leave it to be handled by subsequent osd_fid_lookup().
6096 } else if (rc || osd_id_eq(id, &oti->oti_id)) {
6103 if (scrub->os_running) {
6104 if (inode == NULL) {
6105 inode = osd_iget(oti, dev, id);
6106 /* The inode has been removed (by race maybe). */
6107 if (IS_ERR(inode)) {
6108 rc = PTR_ERR(inode);
6110 RETURN(rc == -ESTALE ? -ENOENT : rc);
6114 rc = osd_scrub_oi_insert(dev, fid, id, insert);
6116 * There is race condition between osd_oi_lookup and OI scrub.
6117 * The OI scrub finished just after osd_oi_lookup() failure.
6118 * Under such case, it is unnecessary to trigger OI scrub again,
6119 * but try to call osd_oi_lookup() again.
6121 if (unlikely(rc == -EAGAIN))
6124 if (!S_ISDIR(inode->i_mode))
6127 rc = osd_check_lmv(oti, dev, inode);
6132 if (dev->od_scrub.os_scrub.os_auto_scrub_interval != AS_NEVER &&
6134 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
6135 SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
6136 CDEBUG_LIMIT(D_LFSCK | D_CONSOLE | D_WARNING,
6137 "%s: trigger partial OI scrub for RPC inconsistency, checking FID "DFID"/%u: rc = %d\n",
6138 osd_dev2name(dev), PFID(fid), id->oii_ino, rc);
6139 if (rc == 0 || rc == -EALREADY)
6151 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
6152 struct osd_device *dev,
6153 struct lu_fid *fid, __u32 ino)
6155 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
6156 struct osd_idmap_cache *oic = &oti->oti_cache;
6157 struct inode *inode;
6160 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
6161 inode = osd_iget(oti, dev, &oic->oic_lid);
6162 if (IS_ERR(inode)) {
6163 fid_zero(&oic->oic_fid);
6164 return PTR_ERR(inode);
6167 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
6170 fid_zero(&oic->oic_fid);
6172 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
6176 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
6177 struct osd_inode_id *id, const struct lu_fid *fid)
6179 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
6180 id->oii_ino, id->oii_gen, info);
6181 info->oti_cache.oic_lid = *id;
6182 info->oti_cache.oic_fid = *fid;
6183 info->oti_cache.oic_dev = osd;
6187 * Get parent FID from the linkEA.
6189 * For a directory which parent resides on remote MDT, to satisfy the
6190 * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
6191 * the other hand, to make the lookup(..) on the directory can return
6192 * the real parent FID, we append the real parent FID after its ".."
6193 * name entry in the /REMOTE_PARENT_DIR.
6195 * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
6196 * backup. So after the restore, we cannot get the right parent FID from
6197 * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
6198 * we have stored the real parent FID in the directory object's linkEA,
6199 * we can parse the linkEA for the real parent FID.
6201 * \param[in] env pointer to the thread context
6202 * \param[in] obj pointer to the object to be handled
6203 * \param[out]fid pointer to the buffer to hold the parent FID
6205 * \retval 0 for getting the real parent FID successfully
6206 * \retval negative error number on failure
6208 static int osd_get_pfid_from_linkea(const struct lu_env *env,
6209 struct osd_object *obj,
6212 struct osd_thread_info *oti = osd_oti_get(env);
6213 struct lu_buf *buf = &oti->oti_big_buf;
6214 struct dentry *dentry = &oti->oti_obj_dentry;
6215 struct inode *inode = obj->oo_inode;
6216 struct linkea_data ldata = { NULL };
6222 if (!S_ISDIR(inode->i_mode))
6226 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
6227 buf->lb_buf, buf->lb_len);
6228 if (rc == -ERANGE) {
6229 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
6232 lu_buf_realloc(buf, rc);
6233 if (buf->lb_buf == NULL)
6240 if (unlikely(rc == 0))
6246 if (unlikely(buf->lb_buf == NULL)) {
6247 lu_buf_realloc(buf, rc);
6248 if (buf->lb_buf == NULL)
6255 rc = linkea_init_with_rec(&ldata);
6257 linkea_first_entry(&ldata);
6258 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
6264 static int osd_verify_ent_by_linkea(const struct lu_env *env,
6265 struct inode *inode,
6266 const struct lu_fid *pfid,
6267 const char *name, const int namelen)
6269 struct osd_thread_info *oti = osd_oti_get(env);
6270 struct lu_buf *buf = &oti->oti_big_buf;
6271 struct dentry *dentry = &oti->oti_obj_dentry;
6272 struct linkea_data ldata = { NULL };
6273 struct lu_name cname = { .ln_name = name,
6274 .ln_namelen = namelen };
6280 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
6281 buf->lb_buf, buf->lb_len);
6283 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
6288 if (unlikely(rc == 0))
6291 if (buf->lb_len < rc) {
6292 lu_buf_realloc(buf, rc);
6293 if (buf->lb_buf == NULL)
6300 rc = linkea_init_with_rec(&ldata);
6302 rc = linkea_links_find(&ldata, &cname, pfid);
6308 * Calls ->lookup() to find dentry. From dentry get inode and
6309 * read inode's ea to get fid. This is required for interoperability
6312 * \retval 0, on success
6313 * \retval -ve, on error
6315 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
6316 struct dt_rec *rec, const struct dt_key *key)
6318 struct inode *dir = obj->oo_inode;
6319 struct dentry *dentry;
6320 struct ldiskfs_dir_entry_2 *de;
6321 struct buffer_head *bh;
6322 struct lu_fid *fid = (struct lu_fid *)rec;
6323 struct htree_lock *hlock = NULL;
6330 LASSERT(dir->i_op != NULL);
6331 LASSERT(dir->i_op->lookup != NULL);
6333 rc = obj_name2lu_name(obj, (char *)key, strlen((char *)key), &ln);
6337 dentry = osd_child_dentry_get(env, obj, ln.ln_name, ln.ln_namelen);
6339 if (obj->oo_hl_head != NULL) {
6340 hlock = osd_oti_get(env)->oti_hlock;
6341 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6342 dir, LDISKFS_HLOCK_LOOKUP);
6344 down_read(&obj->oo_ext_idx_sem);
6347 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6349 struct osd_thread_info *oti = osd_oti_get(env);
6350 struct osd_inode_id *id = &oti->oti_id;
6351 struct osd_device *dev = osd_obj2dev(obj);
6353 ino = le32_to_cpu(de->inode);
6354 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
6356 rc = osd_fail_fid_lookup(oti, dev, fid, ino);
6360 rc = osd_get_fid_from_dentry(de, rec);
6362 /* done with de, release bh */
6365 if (unlikely(is_remote_parent_ino(dev, ino))) {
6366 const char *name = (const char *)key;
6369 * If the parent is on remote MDT, and there
6370 * is no FID-in-dirent, then we have to get
6371 * the parent FID from the linkEA.
6373 if (likely(strlen(name) == 2 &&
6374 name[0] == '.' && name[1] == '.'))
6375 rc = osd_get_pfid_from_linkea(env, obj,
6378 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6381 osd_id_gen(id, ino, OSD_OII_NOGEN);
6384 if (rc != 0 || osd_remote_fid(env, dev, fid))
6387 rc = osd_ldiskfs_consistency_check(oti, dev, fid, id);
6388 if (rc != -ENOENT) {
6389 /* Other error should not affect lookup result. */
6392 /* Normal file mapping should be added into OI cache
6393 * after FID in LMA check, but for local files like
6394 * hsm_actions, their FIDs are not stored in OI files,
6395 * see osd_initial_OI_scrub(), and here is the only
6396 * place to load mapping into OI cache.
6398 if (!fid_is_namespace_visible(fid))
6399 osd_add_oi_cache(osd_oti_get(env),
6400 osd_obj2dev(obj), id, fid);
6402 CDEBUG(D_INODE, DFID"/"DNAME" => "DFID"\n",
6403 PFID(lu_object_fid(&obj->oo_dt.do_lu)), PNAME(&ln),
6413 ldiskfs_htree_unlock(hlock);
6415 up_read(&obj->oo_ext_idx_sem);
6416 if (ln.ln_name != (char *)key)
6421 static int osd_index_declare_ea_insert(const struct lu_env *env,
6422 struct dt_object *dt,
6423 const struct dt_rec *rec,
6424 const struct dt_key *key,
6425 struct thandle *handle)
6427 struct osd_thandle *oh;
6428 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6429 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6430 const struct lu_fid *fid = rec1->rec_fid;
6431 int credits, rc = 0;
6432 struct osd_idmap_cache *idc;
6436 LASSERT(!dt_object_remote(dt));
6437 LASSERT(handle != NULL);
6438 LASSERT(fid != NULL);
6439 LASSERT(rec1->rec_type != 0);
6441 oh = container_of(handle, struct osd_thandle, ot_super);
6442 LASSERT(oh->ot_handle == NULL);
6444 credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
6447 * we can't call iget() while a transactions is running
6448 * (this can lead to a deadlock), but we need to know
6449 * inum and object type. so we find this information at
6450 * declaration and cache in per-thread info
6452 idc = osd_idc_find_or_init(env, osd, fid);
6454 RETURN(PTR_ERR(idc));
6455 if (idc->oic_remote) {
6457 * a reference to remote inode is represented by an
6458 * agent inode which we have to create
6460 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
6461 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
6464 osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
6466 if (osd_dt_obj(dt)->oo_inode != NULL) {
6467 struct inode *inode = osd_dt_obj(dt)->oo_inode;
6470 * We ignore block quota on meta pool (MDTs), so needn't
6471 * calculate how many blocks will be consumed by this index
6474 rc = osd_declare_inode_qid(env, i_uid_read(inode),
6476 i_projid_read(inode), 0,
6477 oh, osd_dt_obj(dt), NULL,
6482 #ifdef HAVE_PROJECT_QUOTA
6484 * Reserve credits for local agent inode to transfer
6485 * to 0, quota enforcement is ignored in this case.
6487 if (idc->oic_remote &&
6488 LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
6489 i_projid_read(inode) != 0)
6490 rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
6491 0, i_projid_read(inode),
6492 0, false, PRJQUOTA);
6500 * Index add function for interoperability mode (b11826).
6501 * It will add the directory entry.This entry is needed to
6502 * maintain name->fid mapping.
6504 * \param key it is key i.e. file entry to be inserted
6505 * \param rec it is value of given key i.e. fid
6507 * \retval 0, on success
6508 * \retval -ve, on error
6510 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6511 const struct dt_rec *rec,
6512 const struct dt_key *key, struct thandle *th)
6514 struct osd_object *obj = osd_dt_obj(dt);
6515 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6516 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6517 const struct lu_fid *fid = rec1->rec_fid;
6518 const char *name = (const char *)key;
6519 struct osd_thread_info *oti = osd_oti_get(env);
6520 struct inode *child_inode = NULL;
6521 struct osd_idmap_cache *idc;
6526 if (!dt_object_exists(dt))
6529 LASSERT(osd_invariant(obj));
6530 LASSERT(!dt_object_remote(dt));
6531 LASSERT(th != NULL);
6533 osd_trans_exec_op(env, th, OSD_OT_INSERT);
6535 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6537 idc = osd_idc_find(env, osd, fid);
6538 if (unlikely(idc == NULL)) {
6539 idc = osd_idc_find_or_init(env, osd, fid);
6542 * this dt_insert() wasn't declared properly, so
6543 * FID is missing in OI cache. we better do not
6544 * lookup FID in FLDB/OI and don't risk to deadlock,
6545 * but in some special cases (lfsck testing, etc)
6546 * it's much simpler than fixing a caller.
6548 * normally this error should be placed after the first
6549 * find, but migrate may attach source stripes to
6550 * target, which doesn't create stripes.
6552 CERROR("%s: "DFID" wasn't declared for insert\n",
6553 osd_name(osd), PFID(fid));
6555 RETURN(PTR_ERR(idc));
6559 if (idc->oic_remote) {
6560 /* Insert remote entry */
6561 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6563 igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6565 child_inode = osd_create_local_agent_inode(env, osd,
6566 obj, fid, rec1->rec_type & S_IFMT, th);
6567 if (IS_ERR(child_inode))
6568 RETURN(PTR_ERR(child_inode));
6571 /* Insert local entry */
6572 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6573 /* for a reason OI cache wasn't filled properly */
6574 CERROR("%s: OIC for "DFID" isn't filled\n",
6575 osd_name(osd), PFID(fid));
6578 child_inode = oti->oti_inode;
6579 if (unlikely(child_inode == NULL)) {
6580 struct ldiskfs_inode_info *lii;
6585 child_inode = oti->oti_inode = &lii->vfs_inode;
6587 child_inode->i_sb = osd_sb(osd);
6588 child_inode->i_ino = idc->oic_lid.oii_ino;
6589 child_inode->i_mode = rec1->rec_type & S_IFMT;
6592 rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6594 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6595 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6597 if (child_inode && child_inode != oti->oti_inode)
6599 LASSERT(osd_invariant(obj));
6600 osd_trans_exec_check(env, th, OSD_OT_INSERT);
6606 * Initialize osd Iterator for given osd index object.
6608 * \param dt osd index object
6611 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6612 struct dt_object *dt,
6615 struct osd_it_iam *it;
6616 struct osd_object *obj = osd_dt_obj(dt);
6617 struct lu_object *lo = &dt->do_lu;
6618 struct iam_path_descr *ipd;
6619 struct iam_container *bag = &obj->oo_dir->od_container;
6621 if (!dt_object_exists(dt))
6622 return ERR_PTR(-ENOENT);
6626 return ERR_PTR(-ENOMEM);
6628 ipd = osd_it_ipd_get(env, bag);
6629 if (likely(ipd != NULL)) {
6633 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6634 return (struct dt_it *)it;
6637 return ERR_PTR(-ENOMEM);
6642 * free given Iterator.
6644 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6646 struct osd_it_iam *it = (struct osd_it_iam *)di;
6647 struct osd_object *obj = it->oi_obj;
6649 iam_it_fini(&it->oi_it);
6650 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6651 osd_object_put(env, obj);
6656 * Move Iterator to record specified by \a key
6658 * \param di osd iterator
6659 * \param key key for index
6661 * \retval +ve di points to record with least key not larger than key
6662 * \retval 0 di points to exact matched key
6663 * \retval -ve failure
6666 static int osd_it_iam_get(const struct lu_env *env,
6667 struct dt_it *di, const struct dt_key *key)
6669 struct osd_thread_info *oti = osd_oti_get(env);
6670 struct osd_it_iam *it = (struct osd_it_iam *)di;
6672 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6673 /* swab quota uid/gid */
6674 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6675 key = (struct dt_key *)&oti->oti_quota_id;
6678 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6684 * \param di osd iterator
6686 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6688 struct osd_it_iam *it = (struct osd_it_iam *)di;
6690 iam_it_put(&it->oi_it);
6694 * Move iterator by one record
6696 * \param di osd iterator
6698 * \retval +1 end of container reached
6700 * \retval -ve failure
6703 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6705 struct osd_it_iam *it = (struct osd_it_iam *)di;
6707 return iam_it_next(&it->oi_it);
6711 * Return pointer to the key under iterator.
6714 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6715 const struct dt_it *di)
6717 struct osd_thread_info *oti = osd_oti_get(env);
6718 struct osd_it_iam *it = (struct osd_it_iam *)di;
6719 struct osd_object *obj = it->oi_obj;
6722 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6724 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6725 /* swab quota uid/gid */
6726 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6727 key = (struct dt_key *)&oti->oti_quota_id;
6734 * Return size of key under iterator (in bytes)
6737 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6739 struct osd_it_iam *it = (struct osd_it_iam *)di;
6741 return iam_it_key_size(&it->oi_it);
6745 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6747 /* check if file type is required */
6748 if (ent->lde_attrs & LUDA_TYPE) {
6749 struct luda_type *lt;
6750 int align = sizeof(*lt) - 1;
6752 len = (len + align) & ~align;
6753 lt = (struct luda_type *)(ent->lde_name + len);
6754 lt->lt_type = cpu_to_le16(DTTOIF(type));
6757 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6761 * build lu direct from backend fs dirent.
6765 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6766 char *name, __u16 namelen, __u16 type, __u32 attr)
6768 ent->lde_attrs = attr | LUDA_FID;
6769 fid_cpu_to_le(&ent->lde_fid, fid);
6771 ent->lde_hash = cpu_to_le64(offset);
6772 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6774 strncpy(ent->lde_name, name, namelen);
6775 ent->lde_name[namelen] = '\0';
6776 ent->lde_namelen = cpu_to_le16(namelen);
6778 /* append lustre attributes */
6779 osd_it_append_attrs(ent, namelen, type);
6783 * Return pointer to the record under iterator.
6785 static int osd_it_iam_rec(const struct lu_env *env,
6786 const struct dt_it *di,
6787 struct dt_rec *dtrec, __u32 attr)
6789 struct osd_it_iam *it = (struct osd_it_iam *)di;
6790 struct osd_thread_info *info = osd_oti_get(env);
6794 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6795 const struct osd_fid_pack *rec;
6796 struct lu_fid *fid = &info->oti_fid;
6797 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6803 name = (char *)iam_it_key_get(&it->oi_it);
6805 RETURN(PTR_ERR(name));
6807 namelen = iam_it_key_size(&it->oi_it);
6809 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6811 RETURN(PTR_ERR(rec));
6813 rc = osd_fid_unpack(fid, rec);
6817 hash = iam_it_store(&it->oi_it);
6819 /* IAM does not store object type in IAM index (dir) */
6820 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6822 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6823 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6824 (struct iam_rec *)dtrec);
6825 osd_quota_unpack(it->oi_obj, dtrec);
6827 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6828 (struct iam_rec *)dtrec);
6835 * Returns cookie for current Iterator position.
6837 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6839 struct osd_it_iam *it = (struct osd_it_iam *)di;
6841 return iam_it_store(&it->oi_it);
6845 * Restore iterator from cookie.
6847 * \param di osd iterator
6848 * \param hash Iterator location cookie
6850 * \retval +ve di points to record with least key not larger than key.
6851 * \retval 0 di points to exact matched key
6852 * \retval -ve failure
6855 static int osd_it_iam_load(const struct lu_env *env,
6856 const struct dt_it *di, __u64 hash)
6858 struct osd_it_iam *it = (struct osd_it_iam *)di;
6860 return iam_it_load(&it->oi_it, hash);
6863 static const struct dt_index_operations osd_index_iam_ops = {
6864 .dio_lookup = osd_index_iam_lookup,
6865 .dio_declare_insert = osd_index_declare_iam_insert,
6866 .dio_insert = osd_index_iam_insert,
6867 .dio_declare_delete = osd_index_declare_iam_delete,
6868 .dio_delete = osd_index_iam_delete,
6870 .init = osd_it_iam_init,
6871 .fini = osd_it_iam_fini,
6872 .get = osd_it_iam_get,
6873 .put = osd_it_iam_put,
6874 .next = osd_it_iam_next,
6875 .key = osd_it_iam_key,
6876 .key_size = osd_it_iam_key_size,
6877 .rec = osd_it_iam_rec,
6878 .store = osd_it_iam_store,
6879 .load = osd_it_iam_load
6883 struct osd_it_ea *osd_it_dir_init(const struct lu_env *env,
6884 struct osd_device *dev,
6885 struct inode *inode, u32 attr)
6887 struct osd_thread_info *info = osd_oti_get(env);
6888 struct osd_it_ea *oie;
6892 file = alloc_file_pseudo(inode, dev->od_mnt, "/", O_NOATIME,
6895 RETURN(ERR_CAST(file));
6897 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6898 if (attr & LUDA_64BITHASH)
6899 file->f_mode |= FMODE_64BITHASH;
6901 file->f_mode |= FMODE_32BITHASH;
6904 OBD_SLAB_ALLOC_PTR(oie, osd_itea_cachep);
6908 oie->oie_rd_dirent = 0;
6909 oie->oie_it_dirent = 0;
6910 oie->oie_dirent = NULL;
6911 if (unlikely(!info->oti_it_ea_buf_used)) {
6912 oie->oie_buf = info->oti_it_ea_buf;
6913 info->oti_it_ea_buf_used = 1;
6915 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6919 oie->oie_obj = NULL;
6920 oie->oie_file = file;
6925 OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6929 return ERR_PTR(-ENOMEM);
6933 * Creates or initializes iterator context.
6935 * \retval struct osd_it_ea, iterator structure on success
6938 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6939 struct dt_object *dt,
6942 struct osd_object *obj = osd_dt_obj(dt);
6943 struct osd_device *dev = osd_obj2dev(obj);
6944 struct lu_object *lo = &dt->do_lu;
6945 struct osd_it_ea *oie;
6949 if (!dt_object_exists(dt) || obj->oo_destroyed)
6950 RETURN(ERR_PTR(-ENOENT));
6952 oie = osd_it_dir_init(env, dev, obj->oo_inode, attr);
6954 RETURN(ERR_CAST(oie));
6958 RETURN((struct dt_it *)oie);
6961 void osd_it_dir_fini(const struct lu_env *env, struct osd_it_ea *oie,
6962 struct inode *inode)
6964 struct osd_thread_info *info = osd_oti_get(env);
6967 fput(oie->oie_file);
6968 if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6969 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6971 info->oti_it_ea_buf_used = 0;
6972 OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6977 * Destroy or finishes iterator context.
6979 * \param di iterator structure to be destroyed
6981 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6983 struct osd_it_ea *oie = (struct osd_it_ea *)di;
6984 struct osd_object *obj = oie->oie_obj;
6985 struct inode *inode = obj->oo_inode;
6988 osd_it_dir_fini(env, (struct osd_it_ea *)di, inode);
6989 osd_object_put(env, obj);
6994 * It position the iterator at given key, so that next lookup continues from
6995 * that key Or it is similar to dio_it->load() but based on a key,
6996 * rather than file position.
6998 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
7001 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
7003 static int osd_it_ea_get(const struct lu_env *env,
7004 struct dt_it *di, const struct dt_key *key)
7006 struct osd_it_ea *it = (struct osd_it_ea *)di;
7009 LASSERT(((const char *)key)[0] == '\0');
7010 it->oie_file->f_pos = 0;
7011 it->oie_rd_dirent = 0;
7012 it->oie_it_dirent = 0;
7013 it->oie_dirent = NULL;
7021 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
7025 struct osd_filldir_cbs {
7026 struct dir_context ctx;
7027 struct osd_it_ea *it;
7030 * It is called internally by ->iterate*(). It fills the
7031 * iterator's in-memory data structure with required
7032 * information i.e. name, namelen, rec_size etc.
7034 * \param buf in which information to be filled in.
7035 * \param name name of the file in given dir
7037 * \retval 0 on success
7038 * \retval 1 on buffer full
7040 #ifdef HAVE_FILLDIR_USE_CTX
7041 static int osd_ldiskfs_filldir(struct dir_context *ctx,
7043 static int osd_ldiskfs_filldir(void *ctx,
7045 const char *name, int namelen,
7046 loff_t offset, __u64 ino, unsigned int d_type)
7048 struct osd_it_ea *it = ((struct osd_filldir_cbs *)ctx)->it;
7049 struct osd_object *obj = it->oie_obj;
7050 struct osd_it_ea_dirent *ent = it->oie_dirent;
7051 struct lu_fid *fid = &ent->oied_fid;
7052 char *buf = it->oie_buf;
7053 struct osd_fid_pack *rec;
7056 /* this should never happen */
7057 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
7058 CERROR("ldiskfs return invalid namelen %d\n", namelen);
7062 /* Check for enough space. Note oied_name is not NUL terminated. */
7063 if (&ent->oied_name[namelen] > buf + OSD_IT_EA_BUFSIZE)
7066 /* "." is just the object itself. */
7067 if (namelen == 1 && name[0] == '.') {
7069 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
7070 } else if (d_type & LDISKFS_DIRENT_LUFID) {
7071 rec = (struct osd_fid_pack *)(name + namelen + 1);
7072 if (osd_fid_unpack(fid, rec) != 0)
7077 d_type &= ~LDISKFS_DIRENT_LUFID;
7079 /* NOT export local root. */
7081 unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
7082 ino = obj->oo_inode->i_ino;
7083 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
7086 if (obj == NULL || !(obj->oo_lma_flags & LUSTRE_ENCRYPT_FL)) {
7087 ent->oied_namelen = namelen;
7088 memcpy(ent->oied_name, name, namelen);
7090 int encoded_namelen = critical_chars(name, namelen);
7092 /* Check again for enough space. */
7093 if (&ent->oied_name[encoded_namelen] > buf + OSD_IT_EA_BUFSIZE)
7096 ent->oied_namelen = encoded_namelen;
7098 if (encoded_namelen == namelen)
7099 memcpy(ent->oied_name, name, namelen);
7101 critical_encode(name, namelen, ent->oied_name);
7104 ent->oied_ino = ino;
7105 ent->oied_off = offset;
7106 ent->oied_type = d_type;
7108 it->oie_rd_dirent++;
7109 it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + ent->oied_namelen);
7114 * Calls ->iterate*() to load a directory entry at a time
7115 * and stored it in iterator's in-memory data structure.
7117 * \param di iterator's in memory structure
7119 * \retval 0 on success
7120 * \retval -ve on error
7121 * \retval +1 reach the end of entry
7123 int osd_ldiskfs_it_fill(const struct lu_env *env, const struct dt_it *di)
7125 struct osd_it_ea *it = (struct osd_it_ea *)di;
7126 struct osd_object *obj = it->oie_obj;
7127 struct htree_lock *hlock = NULL;
7128 struct file *filp = it->oie_file;
7130 struct osd_filldir_cbs buf = {
7131 .ctx.actor = osd_ldiskfs_filldir,
7136 it->oie_dirent = it->oie_buf;
7137 it->oie_rd_dirent = 0;
7140 if (obj->oo_hl_head != NULL) {
7141 hlock = osd_oti_get(env)->oti_hlock;
7142 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
7144 LDISKFS_HLOCK_READDIR);
7146 down_read(&obj->oo_ext_idx_sem);
7150 rc = iterate_dir(filp, &buf.ctx);
7154 if (it->oie_rd_dirent == 0) {
7156 * If it does not get any dirent, it means it has been reached
7157 * to the end of the dir
7159 it->oie_file->f_pos = ldiskfs_get_htree_eof(it->oie_file);
7163 it->oie_dirent = it->oie_buf;
7164 it->oie_it_dirent = 1;
7169 ldiskfs_htree_unlock(hlock);
7171 up_read(&obj->oo_ext_idx_sem);
7178 * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
7179 * to load a directory entry at a time and stored it in
7180 * iterator's in-memory data structure.
7182 * \param di iterator's in memory structure
7184 * \retval +ve iterator reached to end
7185 * \retval 0 iterator not reached to end
7186 * \retval -ve on error
7188 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
7190 struct osd_it_ea *it = (struct osd_it_ea *)di;
7195 if (it->oie_it_dirent < it->oie_rd_dirent) {
7197 (void *)it->oie_dirent +
7198 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
7199 it->oie_dirent->oied_namelen);
7200 it->oie_it_dirent++;
7203 if (it->oie_file->f_pos == ldiskfs_get_htree_eof(it->oie_file))
7206 rc = osd_ldiskfs_it_fill(env, di);
7213 * Returns the key at current position from iterator's in memory structure.
7215 * \param di iterator's in memory structure
7217 * \retval key i.e. struct dt_key on success
7219 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
7220 const struct dt_it *di)
7222 struct osd_it_ea *it = (struct osd_it_ea *)di;
7224 return (struct dt_key *)it->oie_dirent->oied_name;
7228 * Returns key's size at current position from iterator's in memory structure.
7230 * \param di iterator's in memory structure
7232 * \retval key_size i.e. struct dt_key on success
7234 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
7236 struct osd_it_ea *it = (struct osd_it_ea *)di;
7238 return it->oie_dirent->oied_namelen;
7241 #if defined LDISKFS_DIR_ENTRY_LEN && defined LDISKFS_DIR_ENTRY_LEN_
7242 #undef LDISKFS_DIR_REC_LEN
7243 # if defined LDISKFS_DIR_REC_LEN_WITH_DIR
7244 # define LDISKFS_DIR_REC_LEN(de, dir) LDISKFS_DIR_ENTRY_LEN_((de), (dir))
7246 # define LDISKFS_DIR_REC_LEN(de) LDISKFS_DIR_ENTRY_LEN_((de))
7250 #if defined LDISKFS_DIR_REC_LEN_WITH_DIR
7251 # define LDISKFS_DIR_REC_LEN_DIR(de) LDISKFS_DIR_REC_LEN((de), NULL)
7253 # define LDISKFS_DIR_REC_LEN_DIR(de) LDISKFS_DIR_REC_LEN((de))
7256 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
7258 if (LDISKFS_DIR_REC_LEN_DIR(de) >=
7259 __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
7266 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
7267 unsigned int blocksize, bool dotdot)
7270 return osd_dotdot_has_space(de);
7272 if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
7273 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
7280 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
7281 handle_t *jh, struct dentry *dentry,
7282 const struct lu_fid *fid, struct buffer_head *bh,
7283 struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
7286 struct inode *dir = dentry->d_parent->d_inode;
7287 struct inode *inode = dentry->d_inode;
7288 struct osd_fid_pack *rec;
7289 struct ldiskfs_dentry_param *ldp;
7290 int namelen = dentry->d_name.len;
7292 struct osd_thread_info *info = osd_oti_get(env);
7296 if (!ldiskfs_has_feature_dirdata(inode->i_sb))
7299 /* There is enough space to hold the FID-in-dirent. */
7300 if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
7301 rc = osd_ldiskfs_journal_get_write_access(jh, dir->i_sb, bh,
7306 de->name[namelen] = 0;
7307 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
7308 rec->fp_len = sizeof(struct lu_fid) + 1;
7309 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
7310 de->file_type |= LDISKFS_DIRENT_LUFID;
7311 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
7318 rc = ldiskfs_delete_entry(jh, dir, de, bh);
7322 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
7323 osd_get_ldiskfs_dirent_param(ldp, fid);
7324 dentry->d_fsdata = (void *)ldp;
7325 dquot_initialize(dir);
7326 rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
7328 * It is too bad, we cannot reinsert the name entry back.
7329 * That means we lose it!
7333 "%s: fail to reinsert the dirent, dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
7334 osd_ino2name(inode), dir->i_ino, dir->i_generation,
7335 namelen, dentry->d_name.name, PFID(fid), rc);
7341 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
7342 struct osd_it_ea *it, struct lu_fid *fid,
7343 struct osd_inode_id *id, __u32 *attr)
7345 struct osd_thread_info *info = osd_oti_get(env);
7346 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
7347 struct osd_device *dev = osd_obj2dev(obj);
7348 struct super_block *sb = osd_sb(dev);
7349 const char *devname = osd_name(dev);
7350 struct osd_it_ea_dirent *ent = it->oie_dirent;
7351 struct inode *dir = obj->oo_inode;
7352 struct htree_lock *hlock = NULL;
7353 struct buffer_head *bh = NULL;
7354 handle_t *jh = NULL;
7355 struct ldiskfs_dir_entry_2 *de;
7356 struct dentry *dentry;
7357 struct inode *inode;
7358 const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
7361 bool dotdot = false;
7367 if (ent->oied_name[0] == '.') {
7368 if (ent->oied_namelen == 1)
7371 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
7375 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
7376 inode = osd_iget(info, dev, id);
7377 if (IS_ERR(inode)) {
7378 rc = PTR_ERR(inode);
7379 if (rc == -ENOENT || rc == -ESTALE) {
7381 * Maybe dangling name entry, or
7382 * corrupted directory entry.
7384 *attr |= LUDA_UNKNOWN;
7387 CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
7388 "check_repair, dir = %lu/%u, name = %.*s, "
7389 "ino = %llu, rc = %d\n",
7390 devname, dir->i_ino, dir->i_generation,
7391 ent->oied_namelen, ent->oied_name,
7398 rc = obj_name2lu_name(obj, ent->oied_name, ent->oied_namelen, &ln);
7402 dentry = osd_child_dentry_by_inode(env, dir, ln.ln_name, ln.ln_namelen);
7403 rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
7404 if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
7410 * We need to ensure that the name entry is still valid.
7411 * Because it may be removed or renamed by other already.
7413 * The unlink or rename operation will start journal before PDO lock,
7414 * so to avoid deadlock, here we need to start journal handle before
7415 * related PDO lock also. But because we do not know whether there
7416 * will be something to be repaired before PDO lock, we just start
7417 * journal without conditions.
7419 * We may need to remove the name entry firstly, then insert back.
7420 * One credit is for user quota file update.
7421 * One credit is for group quota file update.
7422 * Two credits are for dirty inode.
7424 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
7425 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
7427 if (dev->od_dirent_journal != 0) {
7430 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
7433 CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
7434 "check_repair, dir = %lu/%u, credits = %d, "
7435 "name = %.*s, ino = %llu: rc = %d\n",
7436 devname, dir->i_ino, dir->i_generation, credits,
7437 ent->oied_namelen, ent->oied_name,
7440 GOTO(out_inode, rc);
7443 if (obj->oo_hl_head != NULL) {
7444 hlock = osd_oti_get(env)->oti_hlock;
7446 * "0" means exclusive lock for the whole directory.
7447 * We need to prevent others access such name entry
7448 * during the delete + insert. Neither HLOCK_ADD nor
7449 * HLOCK_DEL cannot guarantee the atomicity.
7451 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
7453 down_write(&obj->oo_ext_idx_sem);
7456 if (obj->oo_hl_head != NULL) {
7457 hlock = osd_oti_get(env)->oti_hlock;
7458 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
7459 LDISKFS_HLOCK_LOOKUP);
7461 down_read(&obj->oo_ext_idx_sem);
7465 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
7466 if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
7467 *attr |= LUDA_IGNORE;
7473 * For dotdot entry, if there is not enough space to hold the
7474 * FID-in-dirent, just keep them there. It only happens when the
7475 * device upgraded from 1.8 or restored from MDT file-level backup.
7476 * For the whole directory, only dotdot entry have no FID-in-dirent
7477 * and needs to get FID from LMA when readdir, it will not affect the
7480 if (dotdot && !osd_dotdot_has_space(de)) {
7481 *attr |= LUDA_UNKNOWN;
7487 if (lu_fid_eq(fid, &lma->lma_self_fid))
7490 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
7491 struct lu_fid *tfid = &lma->lma_self_fid;
7493 if (likely(dotdot &&
7494 fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
7495 fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
7497 * It must be REMOTE_PARENT_DIR and as the
7498 * 'dotdot' entry of remote directory
7500 *attr |= LUDA_IGNORE;
7502 CDEBUG(D_LFSCK, "%s: expect remote agent "
7503 "parent directory, but got %.*s under "
7504 "dir = %lu/%u with the FID "DFID"\n",
7505 devname, ent->oied_namelen,
7506 ent->oied_name, dir->i_ino,
7507 dir->i_generation, PFID(tfid));
7509 *attr |= LUDA_UNKNOWN;
7516 if (!fid_is_zero(fid)) {
7517 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
7519 if (rc == -ENOENT ||
7521 !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
7523 * linkEA does not recognize the dirent entry,
7524 * it may because the dirent entry corruption
7525 * and points to other's inode.
7527 CDEBUG(D_LFSCK, "%s: the target inode does not "
7528 "recognize the dirent, dir = %lu/%u, "
7529 " name = %.*s, ino = %llu, "
7530 DFID": rc = %d\n", devname, dir->i_ino,
7531 dir->i_generation, ent->oied_namelen,
7532 ent->oied_name, ent->oied_ino, PFID(fid), rc);
7533 *attr |= LUDA_UNKNOWN;
7538 if (rc && rc != -ENODATA) {
7539 CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
7540 "dir = %lu/%u, name = %.*s, ino = %llu, "
7541 DFID": rc = %d\n", devname, dir->i_ino,
7542 dir->i_generation, ent->oied_namelen,
7543 ent->oied_name, ent->oied_ino, PFID(fid), rc);
7544 *attr |= LUDA_UNKNOWN;
7552 * linkEA recognizes the dirent entry, the FID-in-LMA is
7553 * valid, trusted, in spite of fid_is_sane(fid) or not.
7555 if (*attr & LUDA_VERIFY_DRYRUN) {
7556 *fid = lma->lma_self_fid;
7557 *attr |= LUDA_REPAIR;
7564 dev->od_dirent_journal = 1;
7565 if (hlock != NULL) {
7566 ldiskfs_htree_unlock(hlock);
7569 up_read(&obj->oo_ext_idx_sem);
7575 *fid = lma->lma_self_fid;
7577 /* Update or append the FID-in-dirent. */
7578 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7579 bh, de, hlock, dotdot);
7581 *attr |= LUDA_REPAIR;
7583 CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7584 "the dirent, dir = %lu/%u, name = %.*s, "
7585 "ino = %llu, "DFID": rc = %d\n",
7586 devname, dir->i_ino, dir->i_generation,
7587 ent->oied_namelen, ent->oied_name,
7588 ent->oied_ino, PFID(fid), rc);
7590 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7591 if (*attr & LUDA_VERIFY_DRYRUN) {
7592 if (fid_is_sane(fid)) {
7593 *attr |= LUDA_REPAIR;
7594 } else if (dev->od_index == 0) {
7595 lu_igif_build(fid, inode->i_ino,
7596 inode->i_generation);
7597 *attr |= LUDA_UPGRADE;
7605 dev->od_dirent_journal = 1;
7606 if (hlock != NULL) {
7607 ldiskfs_htree_unlock(hlock);
7610 up_read(&obj->oo_ext_idx_sem);
7617 if (unlikely(fid_is_sane(fid))) {
7619 * FID-in-dirent exists, but FID-in-LMA is lost.
7620 * Trust the FID-in-dirent, and add FID-in-LMA.
7622 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7624 *attr |= LUDA_REPAIR;
7626 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7627 "update dirent, dir = %lu/%u, "
7628 "name = %.*s, ino = %llu, "
7630 devname, dir->i_ino, dir->i_generation,
7631 ent->oied_namelen, ent->oied_name,
7632 ent->oied_ino, PFID(fid), rc);
7633 } else if (dev->od_index == 0) {
7634 lu_igif_build(fid, inode->i_ino, inode->i_generation);
7636 * It is probably IGIF object. Only aappend the
7637 * FID-in-dirent. OI scrub will process FID-in-LMA.
7639 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7640 bh, de, hlock, dotdot);
7642 *attr |= LUDA_UPGRADE;
7644 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7645 "after the dirent, dir = %lu/%u, "
7646 "name = %.*s, ino = %llu, "
7648 devname, dir->i_ino, dir->i_generation,
7649 ent->oied_namelen, ent->oied_name,
7650 ent->oied_ino, PFID(fid), rc);
7659 if (hlock != NULL) {
7660 ldiskfs_htree_unlock(hlock);
7662 if (dev->od_dirent_journal != 0)
7663 up_write(&obj->oo_ext_idx_sem);
7665 up_read(&obj->oo_ext_idx_sem);
7669 ldiskfs_journal_stop(jh);
7673 if (rc >= 0 && !dirty)
7674 dev->od_dirent_journal = 0;
7675 if (ln.ln_name != ent->oied_name)
7682 * Returns the value at current position from iterator's in memory structure.
7684 * \param di struct osd_it_ea, iterator's in memory structure
7685 * \param attr attr requested for dirent.
7686 * \param lde lustre dirent
7688 * \retval 0 no error and \param lde has correct lustre dirent.
7689 * \retval -ve on error
7691 static inline int osd_it_ea_rec(const struct lu_env *env,
7692 const struct dt_it *di,
7693 struct dt_rec *dtrec, __u32 attr)
7695 struct osd_it_ea *it = (struct osd_it_ea *)di;
7696 struct osd_object *obj = it->oie_obj;
7697 struct osd_device *dev = osd_obj2dev(obj);
7698 struct osd_thread_info *oti = osd_oti_get(env);
7699 struct osd_inode_id *id = &oti->oti_id;
7700 struct lu_fid *fid = &it->oie_dirent->oied_fid;
7701 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
7702 __u32 ino = it->oie_dirent->oied_ino;
7707 LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7709 if (attr & LUDA_VERIFY) {
7710 if (unlikely(is_remote_parent_ino(dev, ino))) {
7711 attr |= LUDA_IGNORE;
7713 * If the parent is on remote MDT, and there
7714 * is no FID-in-dirent, then we have to get
7715 * the parent FID from the linkEA.
7717 if (!fid_is_sane(fid) &&
7718 it->oie_dirent->oied_namelen == 2 &&
7719 it->oie_dirent->oied_name[0] == '.' &&
7720 it->oie_dirent->oied_name[1] == '.')
7721 osd_get_pfid_from_linkea(env, obj, fid);
7723 rc = osd_dirent_check_repair(env, obj, it, fid, id,
7727 if (!fid_is_sane(fid))
7728 attr |= LUDA_UNKNOWN;
7730 attr &= ~LU_DIRENT_ATTRS_MASK;
7731 if (!fid_is_sane(fid)) {
7732 bool is_dotdot = false;
7734 if (it->oie_dirent->oied_namelen == 2 &&
7735 it->oie_dirent->oied_name[0] == '.' &&
7736 it->oie_dirent->oied_name[1] == '.')
7739 * If the parent is on remote MDT, and there
7740 * is no FID-in-dirent, then we have to get
7741 * the parent FID from the linkEA.
7743 if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7744 rc = osd_get_pfid_from_linkea(env, obj, fid);
7746 if (is_dotdot == false &&
7747 OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7750 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7755 /* Pack the entry anyway, at least the offset is right. */
7756 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7757 it->oie_dirent->oied_name,
7758 it->oie_dirent->oied_namelen,
7759 it->oie_dirent->oied_type, attr);
7761 RETURN(rc > 0 ? 0 : rc);
7765 * Returns the record size size at current position.
7767 * This function will return record(lu_dirent) size in bytes.
7769 * \param[in] env execution environment
7770 * \param[in] di iterator's in memory structure
7771 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
7772 * calculate the lu_dirent size.
7774 * \retval record size(in bytes & in memory) of the current lu_dirent
7777 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7780 struct osd_it_ea *it = (struct osd_it_ea *)di;
7782 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7786 * Returns a cookie for current position of the iterator head, so that
7787 * user can use this cookie to load/start the iterator next time.
7789 * \param di iterator's in memory structure
7791 * \retval cookie for current position, on success
7793 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7795 struct osd_it_ea *it = (struct osd_it_ea *)di;
7797 return it->oie_dirent->oied_off;
7801 * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
7802 * to load a directory entry at a time and stored it i inn,
7803 * in iterator's in-memory data structure.
7805 * \param di struct osd_it_ea, iterator's in memory structure
7807 * \retval +ve on success
7808 * \retval -ve on error
7810 static int osd_it_ea_load(const struct lu_env *env,
7811 const struct dt_it *di, __u64 hash)
7813 struct osd_it_ea *it = (struct osd_it_ea *)di;
7817 it->oie_file->f_pos = hash;
7819 rc = osd_ldiskfs_it_fill(env, di);
7830 * Index lookup function for interoperability mode (b11826).
7832 * \param key, key i.e. file name to be searched
7834 * \retval +ve, on success
7835 * \retval -ve, on error
7837 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7838 struct dt_rec *rec, const struct dt_key *key)
7840 struct osd_object *obj = osd_dt_obj(dt);
7845 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7846 LINVRNT(osd_invariant(obj));
7848 rc = osd_ea_lookup_rec(env, obj, rec, key);
7855 * Index and Iterator operations for interoperability
7856 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7858 static const struct dt_index_operations osd_index_ea_ops = {
7859 .dio_lookup = osd_index_ea_lookup,
7860 .dio_declare_insert = osd_index_declare_ea_insert,
7861 .dio_insert = osd_index_ea_insert,
7862 .dio_declare_delete = osd_index_declare_ea_delete,
7863 .dio_delete = osd_index_ea_delete,
7865 .init = osd_it_ea_init,
7866 .fini = osd_it_ea_fini,
7867 .get = osd_it_ea_get,
7868 .put = osd_it_ea_put,
7869 .next = osd_it_ea_next,
7870 .key = osd_it_ea_key,
7871 .key_size = osd_it_ea_key_size,
7872 .rec = osd_it_ea_rec,
7873 .rec_size = osd_it_ea_rec_size,
7874 .store = osd_it_ea_store,
7875 .load = osd_it_ea_load
7879 static void *osd_key_init(const struct lu_context *ctx,
7880 struct lu_context_key *key)
7882 struct osd_thread_info *info;
7884 OBD_ALLOC_PTR(info);
7886 return ERR_PTR(-ENOMEM);
7888 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7889 if (info->oti_it_ea_buf == NULL)
7892 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7894 info->oti_hlock = ldiskfs_htree_lock_alloc();
7895 if (info->oti_hlock == NULL)
7901 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7904 return ERR_PTR(-ENOMEM);
7907 static void osd_key_fini(const struct lu_context *ctx,
7908 struct lu_context_key *key, void *data)
7910 struct osd_thread_info *info = data;
7911 struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7912 struct osd_idmap_cache *idc = info->oti_ins_cache;
7914 if (info->oti_dio_pages) {
7916 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7917 struct page *page = info->oti_dio_pages[i];
7919 LASSERT(PagePrivate2(page));
7920 LASSERT(PageLocked(page));
7921 ClearPagePrivate2(page);
7926 OBD_FREE_PTR_ARRAY_LARGE(info->oti_dio_pages,
7927 PTLRPC_MAX_BRW_PAGES);
7930 if (info->oti_inode != NULL)
7932 if (info->oti_hlock != NULL)
7933 ldiskfs_htree_lock_free(info->oti_hlock);
7934 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7935 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7936 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7937 lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7938 lu_buf_free(&info->oti_big_buf);
7940 LASSERT(info->oti_ins_cache_size > 0);
7941 OBD_FREE_PTR_ARRAY_LARGE(idc, info->oti_ins_cache_size);
7942 info->oti_ins_cache = NULL;
7943 info->oti_ins_cache_size = 0;
7948 static void osd_key_exit(const struct lu_context *ctx,
7949 struct lu_context_key *key, void *data)
7951 struct osd_thread_info *info = data;
7953 LASSERT(info->oti_r_locks == 0);
7954 LASSERT(info->oti_w_locks == 0);
7955 LASSERT(info->oti_txns == 0);
7956 LASSERTF(info->oti_dio_pages_used == 0, "%d\n",
7957 info->oti_dio_pages_used);
7960 /* type constructor/destructor: osd_type_init, osd_type_fini */
7961 LU_TYPE_INIT_FINI(osd, &osd_key);
7963 struct lu_context_key osd_key = {
7964 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7965 .lct_init = osd_key_init,
7966 .lct_fini = osd_key_fini,
7967 .lct_exit = osd_key_exit
7971 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7972 const char *name, struct lu_device *next)
7974 struct osd_device *osd = osd_dev(d);
7976 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7977 sizeof(osd->od_svname))
7979 return osd_procfs_init(osd, name);
7982 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7984 struct seq_server_site *ss = osd_seq_site(osd);
7989 if (osd->od_is_ost || osd->od_cl_seq != NULL)
7992 if (unlikely(ss == NULL))
7995 OBD_ALLOC_PTR(osd->od_cl_seq);
7996 if (osd->od_cl_seq == NULL)
7999 seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
8000 osd->od_svname, ss->ss_server_seq);
8002 if (ss->ss_node_id == 0) {
8004 * If the OSD on the sequence controller(MDT0), then allocate
8005 * sequence here, otherwise allocate sequence after connected
8006 * to MDT0 (see mdt_register_lwp_callback()).
8008 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
8009 &osd->od_cl_seq->lcs_space, env);
8015 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
8017 if (osd->od_cl_seq == NULL)
8020 seq_client_fini(osd->od_cl_seq);
8021 OBD_FREE_PTR(osd->od_cl_seq);
8022 osd->od_cl_seq = NULL;
8025 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
8029 /* shutdown quota slave instance associated with the device */
8030 if (o->od_quota_slave_md != NULL) {
8031 struct qsd_instance *qsd = o->od_quota_slave_md;
8033 o->od_quota_slave_md = NULL;
8037 if (o->od_quota_slave_dt != NULL) {
8038 struct qsd_instance *qsd = o->od_quota_slave_dt;
8040 o->od_quota_slave_dt = NULL;
8044 osd_fid_fini(env, o);
8045 osd_scrub_cleanup(env, o);
8050 static void osd_umount(const struct lu_env *env, struct osd_device *o)
8054 if (o->od_mnt != NULL) {
8055 shrink_dcache_sb(osd_sb(o));
8056 osd_sync(env, &o->od_dt_dev);
8057 wait_event(o->od_commit_cb_done,
8058 !atomic_read(&o->od_commit_cb_in_flight));
8067 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
8068 # ifndef LDISKFS_HAS_INCOMPAT_FEATURE
8069 /* Newer kernels provide the ldiskfs_set_feature_largedir() wrapper already,
8070 * which calls ldiskfs_update_dynamic_rev() to update ancient filesystems.
8071 * All ldiskfs filesystems are already v2, so it is a no-op and unnecessary.
8072 * This avoids maintaining patches to export this otherwise-useless function.
8074 void ldiskfs_update_dynamic_rev(struct super_block *sb)
8081 static int osd_mount(const struct lu_env *env,
8082 struct osd_device *o, struct lustre_cfg *cfg)
8084 const char *name = lustre_cfg_string(cfg, 0);
8085 const char *dev = lustre_cfg_string(cfg, 1);
8087 unsigned long page, s_flags = 0, lmd_flags = 0;
8088 struct page *__page;
8089 struct file_system_type *type;
8090 char *options = NULL;
8092 struct osd_thread_info *info = osd_oti_get(env);
8093 struct lu_fid *fid = &info->oti_fid;
8094 struct inode *inode;
8095 int rc = 0, force_over_1024tb = 0;
8099 if (o->od_mnt != NULL)
8102 if (strlen(dev) >= sizeof(o->od_mntdev))
8104 strcpy(o->od_mntdev, dev);
8106 str = lustre_cfg_buf(cfg, 2);
8107 sscanf(str, "%lu:%lu", &s_flags, &lmd_flags);
8109 opts = lustre_cfg_string(cfg, 3);
8111 if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
8112 CERROR("%s: device %s extents feature is not guaranteed to "
8113 "work on big-endian systems. Use \"bigendian_extents\" "
8114 "mount option to override.\n", name, dev);
8118 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
8119 if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
8120 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");
8123 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
8124 if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
8125 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");
8128 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
8129 if (opts != NULL && strstr(opts, "force_over_512tb") != NULL) {
8130 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");
8134 if (opts != NULL && strstr(opts, "force_over_1024tb") != NULL)
8135 force_over_1024tb = 1;
8137 __page = alloc_page(GFP_KERNEL);
8139 GOTO(out, rc = -ENOMEM);
8140 page = (unsigned long)page_address(__page);
8141 options = (char *)page;
8144 /* strip out the options for back compatiblity */
8145 static const char * const sout[] = {
8152 /* strip out option we processed in osd */
8153 "bigendian_extents",
8157 "force_over_1024tb",
8161 strncat(options, opts, PAGE_SIZE);
8162 for (rc = 0, str = options; sout[rc]; ) {
8163 char *op = strstr(str, sout[rc]);
8170 if (op == options || *(op - 1) == ',') {
8171 str = op + strlen(sout[rc]);
8172 if (*str == ',' || *str == '\0') {
8173 *str == ',' ? str++ : str;
8174 memmove(op, str, strlen(str) + 1);
8177 for (str = op; *str != ',' && *str != '\0'; str++)
8181 strncat(options, "user_xattr,acl", PAGE_SIZE);
8184 /* Glom up mount options */
8185 if (*options != '\0')
8186 strncat(options, ",", PAGE_SIZE);
8187 strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
8189 type = get_fs_type("ldiskfs");
8191 CERROR("%s: cannot find ldiskfs module\n", name);
8192 GOTO(out, rc = -ENODEV);
8195 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
8196 module_put(type->owner);
8198 if (IS_ERR(o->od_mnt)) {
8199 rc = PTR_ERR(o->od_mnt);
8201 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
8205 if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
8206 osd_sb(o)->s_blocksize_bits > 1024ULL << 40 &&
8207 force_over_1024tb == 0) {
8208 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",
8210 GOTO(out_mnt, rc = -EINVAL);
8213 if (lmd_flags & LMD_FLG_DEV_RDONLY) {
8214 LCONSOLE_WARN("%s: not support dev_rdonly on this device\n",
8217 GOTO(out_mnt, rc = -EOPNOTSUPP);
8220 if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
8221 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
8222 GOTO(out_mnt, rc = -EINVAL);
8225 if (ldiskfs_has_feature_fast_commit(o->od_mnt->mnt_sb)) {
8226 CERROR("%s: device %s is mounted with fast_commit that breaks recovery\n",
8228 GOTO(out_mnt, rc = -EOPNOTSUPP);
8231 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
8232 #ifdef LDISKFS_MOUNT_DIRDATA
8233 if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
8234 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
8235 else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
8236 CWARN("%s: device %s was upgraded from Lustre-1.x without "
8237 "enabling the dirdata feature. If you do not want to "
8238 "downgrade to Lustre-1.x again, you can enable it via "
8239 "'tune2fs -O dirdata device'\n", name, dev);
8241 /* enable large_dir on MDTs to avoid REMOTE_PARENT_DIR overflow,
8242 * and on very large OSTs to avoid object directory overflow */
8243 if (unlikely(!ldiskfs_has_feature_largedir(o->od_mnt->mnt_sb) &&
8244 !strstr(name, "MGS"))) {
8245 ldiskfs_set_feature_largedir(o->od_mnt->mnt_sb);
8246 LCONSOLE_INFO("%s: enabled 'large_dir' feature on device %s\n",
8250 inode = osd_sb(o)->s_root->d_inode;
8251 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
8252 if (!o->od_dt_dev.dd_rdonly) {
8253 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
8255 CERROR("%s: failed to set lma on %s root inode\n",
8261 if (lmd_flags & LMD_FLG_NOSCRUB)
8262 o->od_scrub.os_scrub.os_auto_scrub_interval = AS_NEVER;
8264 if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
8265 /* do not use pagecache with flash-backed storage */
8266 o->od_writethrough_cache = 0;
8267 o->od_read_cache = 0;
8278 __free_page(__page);
8283 static struct lu_device *osd_device_fini(const struct lu_env *env,
8284 struct lu_device *d)
8286 struct osd_device *o = osd_dev(d);
8290 osd_index_backup(env, o, false);
8291 osd_shutdown(env, o);
8293 if (o->od_oi_table != NULL)
8294 osd_oi_fini(osd_oti_get(env), o);
8295 if (o->od_extent_bytes_percpu)
8296 free_percpu(o->od_extent_bytes_percpu);
8297 osd_obj_map_fini(o);
8303 static int osd_device_init0(const struct lu_env *env,
8304 struct osd_device *o,
8305 struct lustre_cfg *cfg)
8307 struct lu_device *l = osd2lu_dev(o);
8308 struct osd_thread_info *info;
8311 bool restored = false;
8314 /* if the module was re-loaded, env can loose its keys */
8315 rc = lu_env_refill((struct lu_env *)env);
8318 info = osd_oti_get(env);
8321 l->ld_ops = &osd_lu_ops;
8322 o->od_dt_dev.dd_ops = &osd_dt_ops;
8324 spin_lock_init(&o->od_osfs_lock);
8325 mutex_init(&o->od_otable_mutex);
8326 INIT_LIST_HEAD(&o->od_orphan_list);
8327 INIT_LIST_HEAD(&o->od_index_backup_list);
8328 INIT_LIST_HEAD(&o->od_index_restore_list);
8329 spin_lock_init(&o->od_lock);
8330 o->od_index_backup_policy = LIBP_NONE;
8332 init_waitqueue_head(&o->od_commit_cb_done);
8334 o->od_read_cache = 1;
8335 o->od_writethrough_cache = 1;
8336 o->od_enable_projid_xattr = 0;
8337 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
8338 o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20;
8339 o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20;
8340 o->od_scrub.os_scrub.os_auto_scrub_interval = AS_DEFAULT;
8341 /* default fallocate to unwritten extents: LU-14326/LU-14333 */
8342 o->od_fallocate_zero_blocks = 0;
8344 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
8345 sizeof(o->od_svname));
8346 if (cplen >= sizeof(o->od_svname)) {
8351 o->od_index_backup_stop = 0;
8352 o->od_index = -1; /* -1 means index is invalid */
8353 rc = server_name2index(o->od_svname, &o->od_index, NULL);
8354 if (rc == LDD_F_SV_TYPE_OST)
8357 o->od_full_scrub_ratio = OFSR_DEFAULT;
8358 o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
8359 rc = osd_mount(env, o, cfg);
8363 /* Can only check block device after mount */
8364 o->od_nonrotational =
8365 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
8367 rc = osd_obj_map_init(env, o);
8371 rc = lu_site_init(&o->od_site, l);
8373 GOTO(out_compat, rc);
8374 o->od_site.ls_bottom_dev = l;
8376 rc = lu_site_init_finish(&o->od_site);
8380 opts = lustre_cfg_string(cfg, 3);
8381 if (opts && strstr(opts, "resetoi"))
8384 INIT_LIST_HEAD(&o->od_ios_list);
8386 rc = lprocfs_init_brw_stats(&o->od_brw_stats);
8388 GOTO(out_brw_stats, rc);
8390 /* setup scrub, including OI files initialization */
8392 rc = osd_scrub_setup(env, o, restored);
8395 GOTO(out_brw_stats, rc);
8397 rc = osd_procfs_init(o, o->od_svname);
8399 CERROR("%s: can't initialize procfs: rc = %d\n",
8401 GOTO(out_scrub, rc);
8404 LASSERT(l->ld_site->ls_linkage.next != NULL);
8405 LASSERT(l->ld_site->ls_linkage.prev != NULL);
8407 /* initialize quota slave instance */
8408 /* currently it's no need to prepare qsd_instance_md for OST */
8409 if (!o->od_is_ost) {
8410 o->od_quota_slave_md = qsd_init(env, o->od_svname,
8411 &o->od_dt_dev, o->od_proc_entry,
8413 if (IS_ERR(o->od_quota_slave_md)) {
8414 rc = PTR_ERR(o->od_quota_slave_md);
8415 o->od_quota_slave_md = NULL;
8416 GOTO(out_procfs, rc);
8420 o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
8421 o->od_proc_entry, false, true);
8423 if (IS_ERR(o->od_quota_slave_dt)) {
8424 if (o->od_quota_slave_md != NULL) {
8425 qsd_fini(env, o->od_quota_slave_md);
8426 o->od_quota_slave_md = NULL;
8429 rc = PTR_ERR(o->od_quota_slave_dt);
8430 o->od_quota_slave_dt = NULL;
8431 GOTO(out_procfs, rc);
8434 o->od_extent_bytes_percpu = alloc_percpu(unsigned int);
8435 if (!o->od_extent_bytes_percpu) {
8437 GOTO(out_procfs, rc);
8445 osd_scrub_cleanup(env, o);
8447 lprocfs_fini_brw_stats(&o->od_brw_stats);
8449 lu_site_fini(&o->od_site);
8451 osd_obj_map_fini(o);
8458 static struct lu_device *osd_device_alloc(const struct lu_env *env,
8459 struct lu_device_type *t,
8460 struct lustre_cfg *cfg)
8462 struct osd_device *o;
8467 return ERR_PTR(-ENOMEM);
8469 rc = dt_device_init(&o->od_dt_dev, t);
8472 * Because the ctx might be revived in dt_device_init,
8473 * refill the env here
8475 lu_env_refill((struct lu_env *)env);
8476 rc = osd_device_init0(env, o, cfg);
8478 dt_device_fini(&o->od_dt_dev);
8481 if (unlikely(rc != 0))
8484 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
8487 static struct lu_device *osd_device_free(const struct lu_env *env,
8488 struct lu_device *d)
8490 struct osd_device *o = osd_dev(d);
8494 /* XXX: make osd top device in order to release reference */
8495 d->ld_site->ls_top_dev = d;
8496 lu_site_purge(env, d->ld_site, -1);
8497 lu_site_print(env, d->ld_site, &d->ld_site->ls_obj_hash.nelems,
8498 D_ERROR, lu_cdebug_printer);
8499 lu_site_fini(&o->od_site);
8500 dt_device_fini(&o->od_dt_dev);
8505 static int osd_process_config(const struct lu_env *env,
8506 struct lu_device *d, struct lustre_cfg *cfg)
8508 struct osd_device *o = osd_dev(d);
8514 switch (cfg->lcfg_command) {
8516 rc = osd_mount(env, o, cfg);
8520 * For the case LCFG_PRE_CLEANUP is not called in advance,
8521 * that may happend if hit failure during mount process.
8523 osd_index_backup(env, o, false);
8524 lu_dev_del_linkage(d->ld_site, d);
8525 rc = osd_shutdown(env, o);
8528 LASSERT(&o->od_dt_dev);
8529 count = class_modify_config(cfg, PARAM_OSD,
8530 &o->od_dt_dev.dd_kobj);
8532 count = class_modify_config(cfg, PARAM_OST,
8533 &o->od_dt_dev.dd_kobj);
8534 rc = count > 0 ? 0 : count;
8536 case LCFG_PRE_CLEANUP:
8538 osd_index_backup(env, o,
8539 o->od_index_backup_policy != LIBP_NONE);
8549 static int osd_recovery_complete(const struct lu_env *env,
8550 struct lu_device *d)
8552 struct osd_device *osd = osd_dev(d);
8557 if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
8561 * start qsd instance on recovery completion, this notifies the quota
8562 * slave code that we are about to process new requests now
8564 rc = qsd_start(env, osd->od_quota_slave_dt);
8565 if (rc == 0 && osd->od_quota_slave_md != NULL)
8566 rc = qsd_start(env, osd->od_quota_slave_md);
8572 * we use exports to track all osd users
8574 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
8575 struct obd_device *obd, struct obd_uuid *cluuid,
8576 struct obd_connect_data *data, void *localdata)
8578 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8579 struct lustre_handle conn;
8584 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
8586 rc = class_connect(&conn, obd, cluuid);
8590 *exp = class_conn2export(&conn);
8592 spin_lock(&osd->od_osfs_lock);
8594 spin_unlock(&osd->od_osfs_lock);
8600 * once last export (we don't count self-export) disappeared
8601 * osd can be released
8603 static int osd_obd_disconnect(struct obd_export *exp)
8605 struct obd_device *obd = exp->exp_obd;
8606 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8607 int rc, release = 0;
8611 /* Only disconnect the underlying layers on the final disconnect. */
8612 spin_lock(&osd->od_osfs_lock);
8614 if (osd->od_connects == 0)
8616 spin_unlock(&osd->od_osfs_lock);
8618 rc = class_disconnect(exp); /* bz 9811 */
8620 if (rc == 0 && release)
8621 class_manual_cleanup(obd);
8625 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8626 struct lu_device *dev)
8628 struct osd_device *osd = osd_dev(dev);
8629 struct lr_server_data *lsd =
8630 &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8635 if (osd->od_quota_slave_md != NULL) {
8636 /* set up quota slave objects for inode */
8637 result = qsd_prepare(env, osd->od_quota_slave_md);
8642 if (osd->od_quota_slave_dt != NULL) {
8643 /* set up quota slave objects for block */
8644 result = qsd_prepare(env, osd->od_quota_slave_dt);
8650 if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8651 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8652 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8653 osd->od_index_in_idif = 1;
8655 osd->od_index_in_idif = 0;
8656 result = osd_register_proc_index_in_idif(osd);
8661 osd->od_index_in_idif = 1;
8665 result = osd_fid_init(env, osd);
8671 * Implementation of lu_device_operations::ldo_fid_alloc() for OSD
8675 * see include/lu_object.h for the details.
8677 static int osd_fid_alloc(const struct lu_env *env, struct lu_device *d,
8678 struct lu_fid *fid, struct lu_object *parent,
8679 const struct lu_name *name)
8681 struct osd_device *osd = osd_dev(d);
8683 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8686 static const struct lu_object_operations osd_lu_obj_ops = {
8687 .loo_object_init = osd_object_init,
8688 .loo_object_delete = osd_object_delete,
8689 .loo_object_release = osd_object_release,
8690 .loo_object_free = osd_object_free,
8691 .loo_object_print = osd_object_print,
8692 .loo_object_invariant = osd_object_invariant
8695 const struct lu_device_operations osd_lu_ops = {
8696 .ldo_object_alloc = osd_object_alloc,
8697 .ldo_process_config = osd_process_config,
8698 .ldo_recovery_complete = osd_recovery_complete,
8699 .ldo_prepare = osd_prepare,
8700 .ldo_fid_alloc = osd_fid_alloc,
8703 static const struct lu_device_type_operations osd_device_type_ops = {
8704 .ldto_init = osd_type_init,
8705 .ldto_fini = osd_type_fini,
8707 .ldto_start = osd_type_start,
8708 .ldto_stop = osd_type_stop,
8710 .ldto_device_alloc = osd_device_alloc,
8711 .ldto_device_free = osd_device_free,
8713 .ldto_device_init = osd_device_init,
8714 .ldto_device_fini = osd_device_fini
8717 static struct lu_device_type osd_device_type = {
8718 .ldt_tags = LU_DEVICE_DT,
8719 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
8720 .ldt_ops = &osd_device_type_ops,
8721 .ldt_ctx_tags = LCT_LOCAL,
8724 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8726 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8727 struct super_block *sb = osd_sb(osd);
8729 return (osd->od_mnt == NULL || sb->s_flags & SB_RDONLY);
8733 * lprocfs legacy support.
8735 static const struct obd_ops osd_obd_device_ops = {
8736 .o_owner = THIS_MODULE,
8737 .o_connect = osd_obd_connect,
8738 .o_disconnect = osd_obd_disconnect,
8739 .o_health_check = osd_health_check,
8742 static ssize_t track_declares_assert_show(struct kobject *kobj,
8743 struct attribute *attr,
8746 return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8749 static ssize_t track_declares_assert_store(struct kobject *kobj,
8750 struct attribute *attr,
8751 const char *buffer, size_t count)
8753 bool track_declares_assert;
8756 rc = kstrtobool(buffer, &track_declares_assert);
8760 ldiskfs_track_declares_assert = track_declares_assert;
8764 LUSTRE_RW_ATTR(track_declares_assert);
8766 static int __init osd_init(void)
8768 struct kobject *kobj;
8771 BUILD_BUG_ON(BH_DXLock >=
8772 sizeof(((struct buffer_head *)0)->b_state) * 8);
8773 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8774 /* please, try to keep osd_thread_info smaller than a page */
8775 BUILD_BUG_ON(sizeof(struct osd_thread_info) > PAGE_SIZE);
8780 rc = lu_kmem_init(ldiskfs_caches);
8784 rc = class_register_type(&osd_obd_device_ops, NULL, true,
8785 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8787 lu_kmem_fini(ldiskfs_caches);
8791 kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8793 rc = sysfs_create_file(kobj,
8794 &lustre_attr_track_declares_assert.attr);
8797 CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8805 static void __exit osd_exit(void)
8807 struct kobject *kobj;
8809 kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8811 sysfs_remove_file(kobj,
8812 &lustre_attr_track_declares_assert.attr);
8815 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8816 lu_kmem_fini(ldiskfs_caches);
8819 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8820 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8821 MODULE_VERSION(LUSTRE_VERSION_STRING);
8822 MODULE_LICENSE("GPL");
8824 module_init(osd_init);
8825 module_exit(osd_exit);