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/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/osd/osd_handler.c
34 * Top-level entry points into osd module
36 * Author: Nikita Danilov <nikita@clusterfs.com>
37 * Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
40 #define DEBUG_SUBSYSTEM S_OSD
42 #include <linux/kallsyms.h>
43 #include <linux/module.h>
44 #include <linux/user_namespace.h>
45 #include <linux/uidgid.h>
47 /* prerequisite for linux/xattr.h */
48 #include <linux/types.h>
49 /* prerequisite for linux/xattr.h */
51 /* XATTR_{REPLACE,CREATE} */
52 #include <linux/xattr.h>
54 #include <ldiskfs/ldiskfs.h>
55 #include <ldiskfs/xattr.h>
56 #include <ldiskfs/ldiskfs_extents.h>
59 * struct OBD_{ALLOC,FREE}*()
62 #include <obd_support.h>
63 /* struct ptlrpc_thread */
64 #include <lustre_net.h>
65 #include <lustre_fid.h>
67 #include <uapi/linux/lustre/lustre_param.h>
69 #include "osd_internal.h"
70 #include "osd_dynlocks.h"
72 /* llo_* api support */
73 #include <md_object.h>
74 #include <lustre_quota.h>
76 #include <lustre_linkea.h>
78 /* Maximum EA size is limited by LNET_MTU for remote objects */
79 #define OSD_MAX_EA_SIZE 1048364
82 module_param(ldiskfs_pdo, int, 0644);
83 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
85 int ldiskfs_track_declares_assert;
86 module_param(ldiskfs_track_declares_assert, int, 0644);
87 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
89 /* Slab to allocate dynlocks */
90 struct kmem_cache *dynlock_cachep;
92 /* Slab to allocate osd_it_ea */
93 struct kmem_cache *osd_itea_cachep;
95 static struct lu_kmem_descr ldiskfs_caches[] = {
97 .ckd_cache = &dynlock_cachep,
98 .ckd_name = "dynlock_cache",
99 .ckd_size = sizeof(struct dynlock_handle)
102 .ckd_cache = &osd_itea_cachep,
103 .ckd_name = "osd_itea_cache",
104 .ckd_size = sizeof(struct osd_it_ea)
111 static const char dot[] = ".";
112 static const char dotdot[] = "..";
114 static const struct lu_object_operations osd_lu_obj_ops;
115 static const struct dt_object_operations osd_obj_ops;
116 static const struct dt_object_operations osd_obj_otable_it_ops;
117 static const struct dt_index_operations osd_index_iam_ops;
118 static const struct dt_index_operations osd_index_ea_ops;
120 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
121 const struct lu_fid *fid);
122 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
123 struct osd_device *osd);
125 int osd_trans_declare_op2rb[] = {
126 [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
127 [OSD_OT_PUNCH] = OSD_OT_MAX,
128 [OSD_OT_XATTR_SET] = OSD_OT_XATTR_SET,
129 [OSD_OT_CREATE] = OSD_OT_DESTROY,
130 [OSD_OT_DESTROY] = OSD_OT_CREATE,
131 [OSD_OT_REF_ADD] = OSD_OT_REF_DEL,
132 [OSD_OT_REF_DEL] = OSD_OT_REF_ADD,
133 [OSD_OT_WRITE] = OSD_OT_WRITE,
134 [OSD_OT_INSERT] = OSD_OT_DELETE,
135 [OSD_OT_DELETE] = OSD_OT_INSERT,
136 [OSD_OT_QUOTA] = OSD_OT_MAX,
139 static int osd_has_index(const struct osd_object *obj)
141 return obj->oo_dt.do_index_ops != NULL;
144 static int osd_object_invariant(const struct lu_object *l)
146 return osd_invariant(osd_obj(l));
150 * Concurrency: doesn't matter
152 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
154 struct osd_thread_info *oti = osd_oti_get(env);
156 return oti->oti_w_locks > 0 && o->oo_owner == env;
160 * Concurrency: doesn't access mutable data
162 static int osd_root_get(const struct lu_env *env,
163 struct dt_device *dev, struct lu_fid *f)
165 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
170 * the following set of functions are used to maintain per-thread
171 * cache of FID->ino mapping. this mechanism is needed to resolve
172 * FID to inode at dt_insert() which in turn stores ino in the
173 * directory entries to keep ldiskfs compatible with ext[34].
174 * due to locking-originated restrictions we can't lookup ino
175 * using LU cache (deadlock is possible). lookup using OI is quite
176 * expensive. so instead we maintain this cache and methods like
177 * dt_create() fill it. so in the majority of cases dt_insert() is
178 * able to find needed mapping in lockless manner.
180 static struct osd_idmap_cache *
181 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
182 const struct lu_fid *fid)
184 struct osd_thread_info *oti = osd_oti_get(env);
185 struct osd_idmap_cache *idc = oti->oti_ins_cache;
188 for (i = 0; i < oti->oti_ins_cache_used; i++) {
189 if (!lu_fid_eq(&idc[i].oic_fid, fid))
191 if (idc[i].oic_dev != osd)
200 static struct osd_idmap_cache *
201 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
202 const struct lu_fid *fid)
204 struct osd_thread_info *oti = osd_oti_get(env);
205 struct osd_idmap_cache *idc;
208 if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
209 i = oti->oti_ins_cache_size * 2;
211 i = OSD_INS_CACHE_SIZE;
212 OBD_ALLOC_PTR_ARRAY(idc, i);
214 return ERR_PTR(-ENOMEM);
215 if (oti->oti_ins_cache != NULL) {
216 memcpy(idc, oti->oti_ins_cache,
217 oti->oti_ins_cache_used * sizeof(*idc));
218 OBD_FREE_PTR_ARRAY(oti->oti_ins_cache,
219 oti->oti_ins_cache_used);
221 oti->oti_ins_cache = idc;
222 oti->oti_ins_cache_size = i;
225 idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
228 idc->oic_lid.oii_ino = 0;
229 idc->oic_lid.oii_gen = 0;
236 * lookup mapping for the given fid in the cache, initialize a
237 * new one if not found. the initialization checks whether the
238 * object is local or remote. for local objects, OI is used to
239 * learn ino/generation. the function is used when the caller
240 * has no information about the object, e.g. at dt_insert().
242 static struct osd_idmap_cache *
243 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
244 const struct lu_fid *fid)
246 struct osd_idmap_cache *idc;
249 idc = osd_idc_find(env, osd, fid);
250 LASSERT(!IS_ERR(idc));
254 CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
255 osd->od_svname, PFID(fid));
257 /* new mapping is needed */
258 idc = osd_idc_add(env, osd, fid);
260 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
261 osd->od_svname, PFID(fid), PTR_ERR(idc));
266 rc = osd_remote_fid(env, osd, fid);
267 if (unlikely(rc < 0))
271 /* the object is local, lookup in OI */
272 /* XXX: probably cheaper to lookup in LU first? */
273 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
275 if (unlikely(rc < 0)) {
276 CERROR("can't lookup: rc = %d\n", rc);
280 /* the object is remote */
288 * lookup mapping for given FID and fill it from the given object.
289 * the object is lolcal by definition.
291 static int osd_idc_find_and_init(const struct lu_env *env,
292 struct osd_device *osd,
293 struct osd_object *obj)
295 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
296 struct osd_idmap_cache *idc;
298 idc = osd_idc_find(env, osd, fid);
299 LASSERT(!IS_ERR(idc));
301 if (obj->oo_inode == NULL)
303 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
304 LASSERT(idc->oic_lid.oii_ino == 0);
305 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
306 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
311 CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
312 osd->od_svname, PFID(fid));
314 /* new mapping is needed */
315 idc = osd_idc_add(env, osd, fid);
317 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
318 osd->od_svname, PFID(fid), PTR_ERR(idc));
322 if (obj->oo_inode != NULL) {
323 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
324 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
330 * OSD object methods.
334 * Concurrency: no concurrent access is possible that early in object
337 static struct lu_object *osd_object_alloc(const struct lu_env *env,
338 const struct lu_object_header *hdr,
341 struct osd_object *mo;
346 struct lu_object_header *h;
347 struct osd_device *o = osd_dev(d);
349 l = &mo->oo_dt.do_lu;
350 if (unlikely(o->od_in_init)) {
357 lu_object_header_init(h);
358 lu_object_init(l, h, d);
359 lu_object_add_top(h, l);
362 dt_object_init(&mo->oo_dt, NULL, d);
363 mo->oo_header = NULL;
366 mo->oo_dt.do_ops = &osd_obj_ops;
367 l->lo_ops = &osd_lu_obj_ops;
368 init_rwsem(&mo->oo_sem);
369 init_rwsem(&mo->oo_ext_idx_sem);
370 spin_lock_init(&mo->oo_guard);
371 INIT_LIST_HEAD(&mo->oo_xattr_list);
377 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
378 struct dentry *dentry, struct lustre_ost_attrs *loa)
382 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
383 (void *)loa, sizeof(*loa));
385 struct lustre_mdt_attrs *lma = &loa->loa_lma;
387 if (rc < sizeof(*lma))
391 lustre_loa_swab(loa, true);
392 /* Check LMA compatibility */
393 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
394 CWARN("%s: unsupported incompat LMA feature(s) %#x "
395 "for fid = "DFID", ino = %lu\n",
397 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
398 PFID(&lma->lma_self_fid), inode->i_ino);
401 } else if (rc == 0) {
409 * retrieve object from backend ext fs.
411 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
412 struct osd_inode_id *id)
415 struct inode *inode = NULL;
418 * if we look for an inode withing a running
419 * transaction, then we risk to deadlock
420 * osd_dirent_check_repair() breaks this
422 /* LASSERT(current->journal_info == NULL); */
424 inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
426 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
427 id->oii_ino, PTR_ERR(inode));
428 } else if (id->oii_gen != OSD_OII_NOGEN &&
429 inode->i_generation != id->oii_gen) {
430 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
431 "i_generation = %u\n",
432 id->oii_ino, id->oii_gen, inode->i_generation);
434 inode = ERR_PTR(-ESTALE);
435 } else if (inode->i_nlink == 0) {
437 * due to parallel readdir and unlink,
438 * we can have dead inode here.
440 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
442 inode = ERR_PTR(-ESTALE);
443 } else if (is_bad_inode(inode)) {
444 CWARN("%s: bad inode: ino = %u\n",
445 osd_dev2name(dev), id->oii_ino);
447 inode = ERR_PTR(-ENOENT);
448 } else if ((rc = osd_attach_jinode(inode))) {
452 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
453 if (id->oii_gen == OSD_OII_NOGEN)
454 osd_id_gen(id, inode->i_ino, inode->i_generation);
457 * Do not update file c/mtime in ldiskfs.
458 * NB: we don't have any lock to protect this because we don't
459 * have reference on osd_object now, but contention with
460 * another lookup + attr_set can't happen in the tiny window
461 * between if (...) and set S_NOCMTIME.
463 if (!(inode->i_flags & S_NOCMTIME))
464 inode->i_flags |= S_NOCMTIME;
469 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
470 handle_t *handle, struct dentry *child,
471 struct inode *inode, struct htree_lock *hlock)
475 rc = __ldiskfs_add_entry(handle, child, inode, hlock);
476 if (rc == -ENOBUFS || rc == -ENOSPC) {
477 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
478 struct inode *parent = child->d_parent->d_inode;
479 struct lu_fid *fid = NULL;
481 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
483 fid = &loa->loa_lma.lma_self_fid;
484 } else if (rc2 == -ENODATA) {
485 if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
486 fid = &info->oti_fid3;
487 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
488 } else if (!osd->od_is_ost && osd->od_index == 0) {
489 fid = &info->oti_fid3;
490 lu_igif_build(fid, parent->i_ino,
491 parent->i_generation);
496 /* below message is checked in sanity.sh test_129 */
497 CWARN("%s: directory (inode: %lu, FID: "DFID") %s maximum entry limit\n",
498 osd_name(osd), parent->i_ino, PFID(fid),
499 rc == -ENOSPC ? "has reached" : "is approaching");
501 /* below message is checked in sanity.sh test_129 */
502 CWARN("%s: directory (inode: %lu, FID: unknown) %s maximum entry limit\n",
503 osd_name(osd), parent->i_ino,
504 rc == -ENOSPC ? "has reached" : "is approaching");
506 /* ignore such error now */
516 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
517 struct osd_inode_id *id, struct lu_fid *fid)
519 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
523 inode = osd_iget(info, dev, id);
527 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
529 *fid = loa->loa_lma.lma_self_fid;
530 } else if (rc == -ENODATA) {
531 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
532 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
534 lu_igif_build(fid, inode->i_ino, inode->i_generation);
542 static struct inode *osd_iget_check(struct osd_thread_info *info,
543 struct osd_device *dev,
544 const struct lu_fid *fid,
545 struct osd_inode_id *id,
554 * The cached OI mapping is trustable. If we cannot locate the inode
555 * via the cached OI mapping, then return the failure to the caller
556 * directly without further OI checking.
560 inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
563 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
566 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
567 PFID(fid), id->oii_ino, rc);
571 if (is_bad_inode(inode)) {
576 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
577 PFID(fid), id->oii_ino);
581 if (id->oii_gen != OSD_OII_NOGEN &&
582 inode->i_generation != id->oii_gen) {
587 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
588 "oii_gen = %u, i_generation = %u\n", PFID(fid),
589 id->oii_ino, id->oii_gen, inode->i_generation);
593 if (inode->i_nlink == 0) {
598 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
599 PFID(fid), id->oii_ino);
603 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
607 __u32 saved_ino = id->oii_ino;
608 __u32 saved_gen = id->oii_gen;
611 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
613 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
615 * XXX: There are four possible cases:
617 * Backup/restore caused the OI invalid.
619 * Someone unlinked the object but NOT removed
620 * the OI mapping, such as mount target device
621 * as ldiskfs, and modify something directly.
623 * Someone just removed the object between the
624 * former oi_lookup and the iget. It is normal.
625 * 4. Other failure cases.
627 * Generally, when the device is mounted, it will
628 * auto check whether the system is restored from
629 * file-level backup or not. We trust such detect
630 * to distinguish the 1st case from the 2nd case:
631 * if the OI files are consistent but may contain
632 * stale OI mappings because of case 2, if iget()
633 * returns -ENOENT or -ESTALE, then it should be
638 * If the OI mapping was in OI file before the
639 * osd_iget_check(), but now, it is disappear,
640 * then it must be removed by race. That is a
646 * It is the OI scrub updated the OI mapping by race.
647 * The new OI mapping must be valid.
649 if (saved_ino != id->oii_ino ||
650 (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
659 if (dev->od_scrub.os_scrub.os_file.sf_flags &
662 * It still can be the case 2, but we cannot
663 * distinguish it from the case 1. So return
664 * -EREMCHG to block current operation until
665 * OI scrub rebuilt the OI mappings.
674 if (inode->i_generation == id->oii_gen)
679 if (id->oii_gen == OSD_OII_NOGEN)
680 osd_id_gen(id, inode->i_ino, inode->i_generation);
683 * Do not update file c/mtime in ldiskfs.
684 * NB: we don't have any lock to protect this because we don't
685 * have reference on osd_object now, but contention with
686 * another lookup + attr_set can't happen in the tiny window
687 * between if (...) and set S_NOCMTIME.
689 if (!(inode->i_flags & S_NOCMTIME))
690 inode->i_flags |= S_NOCMTIME;
707 * \retval +v: new filter_fid does not contain self-fid
708 * \retval 0: filter_fid_18_23, contains self-fid
709 * \retval -v: other failure cases
711 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
712 struct dentry *dentry, struct lu_fid *fid)
714 struct filter_fid *ff = &info->oti_ff;
715 struct ost_id *ostid = &info->oti_ostid;
718 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
719 if (rc == sizeof(struct filter_fid_18_23)) {
720 struct filter_fid_18_23 *ff_old = (void *)ff;
722 ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq));
723 rc = ostid_set_id(ostid, le64_to_cpu(ff_old->ff_objid));
725 * XXX: use 0 as the index for compatibility, the caller will
726 * handle index related issues when necessary.
729 ostid_to_fid(fid, ostid, 0);
730 } else if (rc >= (int)sizeof(struct filter_fid_24_29)) {
732 } else if (rc >= 0) {
739 static int osd_lma_self_repair(struct osd_thread_info *info,
740 struct osd_device *osd, struct inode *inode,
741 const struct lu_fid *fid, __u32 compat)
746 LASSERT(current->journal_info == NULL);
748 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
749 osd_dto_credits_noquota[DTO_XATTR_SET]);
752 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
757 rc = osd_ea_fid_set(info, inode, fid, compat, 0);
759 CWARN("%s: cannot self repair the LMA: rc = %d\n",
761 ldiskfs_journal_stop(jh);
765 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
767 struct osd_thread_info *info = osd_oti_get(env);
768 struct osd_device *osd = osd_obj2dev(obj);
769 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
770 struct lustre_mdt_attrs *lma = &loa->loa_lma;
771 struct inode *inode = obj->oo_inode;
772 struct dentry *dentry = &info->oti_obj_dentry;
773 struct lu_fid *fid = NULL;
774 const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
779 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
780 (void *)loa, sizeof(*loa));
781 if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
782 fid = &lma->lma_self_fid;
783 rc = osd_get_idif(info, inode, dentry, fid);
784 if (rc > 0 || (rc == -ENODATA && osd->od_index_in_idif)) {
786 * For the given OST-object, if it has neither LMA nor
787 * FID in XATTR_NAME_FID, then the given FID (which is
788 * contained in the @obj, from client RPC for locating
789 * the OST-object) is trusted. We use it to generate
792 osd_lma_self_repair(info, osd, inode, rfid,
803 lustre_lma_swab(lma);
804 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
805 (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
806 S_ISREG(inode->i_mode)))) {
807 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
808 "fid = "DFID", ino = %lu\n", osd_name(osd),
809 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
810 PFID(rfid), inode->i_ino);
813 fid = &lma->lma_self_fid;
814 if (lma->lma_compat & LMAC_STRIPE_INFO &&
816 obj->oo_pfid_in_lma = 1;
817 if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
819 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
823 if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
824 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
825 struct ost_id *oi = &info->oti_ostid;
826 struct lu_fid *fid1 = &info->oti_fid3;
827 __u32 idx = fid_idif_ost_idx(rfid);
830 * For old IDIF, the OST index is not part of the IDIF,
831 * Means that different OSTs may have the same IDIFs.
832 * Under such case, we need to make some compatible
833 * check to make sure to trigger OI scrub properly.
835 if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
836 /* Given @rfid is new, LMA is old. */
837 fid_to_ostid(fid, oi);
838 ostid_to_fid(fid1, oi, idx);
839 if (lu_fid_eq(fid1, rfid)) {
840 if (osd->od_index_in_idif)
841 osd_lma_self_repair(info, osd,
855 struct osd_check_lmv_buf {
856 /* please keep it as first member */
857 struct dir_context ctx;
858 struct osd_thread_info *oclb_info;
859 struct osd_device *oclb_dev;
860 struct osd_idmap_cache *oclb_oic;
866 * It is called internally by ->iterate*() to filter out the
867 * local slave object's FID of the striped directory.
869 * \retval 1 found the local slave's FID
870 * \retval 0 continue to check next item
871 * \retval -ve for failure
873 #ifdef HAVE_FILLDIR_USE_CTX
874 static int osd_stripe_dir_filldir(struct dir_context *buf,
876 static int osd_stripe_dir_filldir(void *buf,
878 const char *name, int namelen,
879 loff_t offset, __u64 ino, unsigned int d_type)
881 struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
882 struct osd_thread_info *oti = oclb->oclb_info;
883 struct lu_fid *fid = &oti->oti_fid3;
884 struct osd_inode_id *id = &oti->oti_id3;
885 struct osd_device *dev = oclb->oclb_dev;
886 struct osd_idmap_cache *oic = oclb->oclb_oic;
895 sscanf(name + 1, SFID, RFID(fid));
896 if (!fid_is_sane(fid))
899 if (osd_remote_fid(oti->oti_env, dev, fid))
902 osd_id_gen(id, ino, OSD_OII_NOGEN);
903 inode = osd_iget(oti, dev, id);
905 return PTR_ERR(inode);
908 osd_add_oi_cache(oti, dev, id, fid);
912 osd_oii_insert(dev, oic, true);
913 oclb->oclb_found = true;
919 * When lookup item under striped directory, we need to locate the master
920 * MDT-object of the striped directory firstly, then the client will send
921 * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
922 * and the item's name. If the system is restored from MDT file level backup,
923 * then before the OI scrub completely built the OI files, the OI mappings of
924 * the master MDT-object and slave MDT-object may be invalid. Usually, it is
925 * not a problem for the master MDT-object. Because when locate the master
926 * MDT-object, we will do name based lookup (for the striped directory itself)
927 * firstly, during such process we can setup the correct OI mapping for the
928 * master MDT-object. But it will be trouble for the slave MDT-object. Because
929 * the client will not trigger name based lookup on the MDT to locate the slave
930 * MDT-object before locating item under the striped directory, then when
931 * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
932 * is invalid and does not know what the right OI mapping is, then the MDT has
933 * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
934 * the OI file, related OI mapping is unknown yet, please try again later. And
935 * then client will re-try the RPC again and again until related OI mapping has
936 * been updated. That is quite inefficient.
938 * To resolve above trouble, we will handle it as the following two cases:
940 * 1) The slave MDT-object and the master MDT-object are on different MDTs.
941 * It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
942 * is linked under /REMOTE_PARENT_DIR with the name of its FID string.
943 * We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
944 * directly. Please check osd_fid_lookup().
946 * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
947 * Under such case, during lookup the master MDT-object, we will lookup the
948 * slave MDT-object via readdir against the master MDT-object, because the
949 * slave MDT-objects information are stored as sub-directories with the name
950 * "${FID}:${index}". Then when find the local slave MDT-object, its OI
951 * mapping will be recorded. Then subsequent osd_fid_lookup() will know
952 * the correct OI mapping for the slave MDT-object.
954 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
955 struct inode *inode, struct osd_idmap_cache *oic)
957 struct lu_buf *buf = &oti->oti_big_buf;
958 struct dentry *dentry = &oti->oti_obj_dentry;
959 struct file *filp = &oti->oti_file;
960 const struct file_operations *fops;
961 struct lmv_mds_md_v1 *lmv1;
962 struct osd_check_lmv_buf oclb = {
963 .ctx.actor = osd_stripe_dir_filldir,
974 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
977 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
979 lu_buf_realloc(buf, rc);
980 if (buf->lb_buf == NULL)
981 GOTO(out, rc = -ENOMEM);
987 if (unlikely(rc == 0 || rc == -ENODATA))
993 if (unlikely(buf->lb_buf == NULL)) {
994 lu_buf_realloc(buf, rc);
995 if (buf->lb_buf == NULL)
996 GOTO(out, rc = -ENOMEM);
1002 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1005 fops = inode->i_fop;
1006 dentry->d_inode = inode;
1007 dentry->d_sb = inode->i_sb;
1009 filp->f_path.dentry = dentry;
1010 filp->f_flags |= O_NOATIME;
1011 filp->f_mode = FMODE_64BITHASH | FMODE_NONOTIFY;
1012 filp->f_mapping = inode->i_mapping;
1014 filp->private_data = NULL;
1015 filp->f_cred = current_cred();
1016 filp->f_inode = inode;
1017 rc = osd_security_file_alloc(filp);
1022 oclb.oclb_items = 0;
1023 rc = iterate_dir(filp, &oclb.ctx);
1024 } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1025 filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1026 fops->release(inode, filp);
1030 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
1031 DFID": rc = %d\n", osd_ino2name(inode),
1032 inode->i_ino, inode->i_generation,
1033 PFID(&oic->oic_fid), rc);
1040 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1041 const struct lu_fid *fid,
1042 const struct lu_object_conf *conf)
1044 struct osd_thread_info *info;
1045 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1046 struct osd_device *dev;
1047 struct osd_idmap_cache *oic;
1048 struct osd_inode_id *id;
1049 struct inode *inode = NULL;
1050 struct lustre_scrub *scrub;
1051 struct scrub_file *sf;
1052 __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1057 bool remote = false;
1058 bool trusted = true;
1059 bool updated = false;
1060 bool checked = false;
1064 LINVRNT(osd_invariant(obj));
1065 LASSERT(obj->oo_inode == NULL);
1066 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
1068 dev = osd_dev(ldev);
1069 scrub = &dev->od_scrub.os_scrub;
1070 sf = &scrub->os_file;
1071 info = osd_oti_get(env);
1073 oic = &info->oti_cache;
1075 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1079 * For the object is created as locking anchor, or for the object to
1080 * be created on disk. No need to osd_oi_lookup() at here because FID
1081 * shouldn't never be re-used, if it's really a duplicate FID from
1082 * unexpected reason, we should be able to detect it later by calling
1083 * do_create->osd_oi_insert().
1085 if (conf && conf->loc_flags & LOC_F_NEW)
1086 GOTO(out, result = 0);
1088 /* Search order: 1. per-thread cache. */
1089 if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1095 if (!list_empty(&scrub->os_inconsistent_items)) {
1096 /* Search order: 2. OI scrub pending list. */
1097 result = osd_oii_lookup(dev, fid, id);
1103 * The OI mapping in the OI file can be updated by the OI scrub
1104 * when we locate the inode via FID. So it may be not trustable.
1108 /* Search order: 3. OI files. */
1109 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1110 if (result == -ENOENT) {
1111 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1112 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1113 !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1115 GOTO(out, result = 0);
1120 /* -ESTALE is returned if inode of OST object doesn't exist */
1121 if (result == -ESTALE &&
1122 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1123 GOTO(out, result = 0);
1130 obj->oo_inode = NULL;
1131 /* for later passes through checks, not true on first pass */
1132 if (!IS_ERR_OR_NULL(inode))
1135 inode = osd_iget_check(info, dev, fid, id, trusted);
1136 if (!IS_ERR(inode)) {
1137 obj->oo_inode = inode;
1145 result = PTR_ERR(inode);
1146 if (result == -ENOENT || result == -ESTALE)
1147 GOTO(out, result = 0);
1149 if (result != -EREMCHG)
1154 * We still have chance to get the valid inode: for the
1155 * object which is referenced by remote name entry, the
1156 * object on the local MDT will be linked under the dir
1157 * of "/REMOTE_PARENT_DIR" with its FID string as name.
1159 * We do not know whether the object for the given FID
1160 * is referenced by some remote name entry or not, and
1161 * especially for DNE II, a multiple-linked object may
1162 * have many name entries reside on many MDTs.
1164 * To simplify the operation, OSD will not distinguish
1165 * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1166 * only happened for the RPC from other MDT during the
1167 * OI scrub, or for the client side RPC with FID only,
1168 * such as FID to path, or from old connected client.
1171 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1175 flags |= SS_AUTO_PARTIAL;
1176 flags &= ~SS_AUTO_FULL;
1181 if (thread_is_running(&scrub->os_thread)) {
1182 if (scrub->os_partial_scan && !scrub->os_in_join)
1185 if (IS_ERR_OR_NULL(inode) || result)
1186 GOTO(out, result = -EINPROGRESS);
1189 LASSERT(obj->oo_inode == inode);
1191 osd_add_oi_cache(info, dev, id, fid);
1192 osd_oii_insert(dev, oic, true);
1196 if (dev->od_auto_scrub_interval == AS_NEVER) {
1198 GOTO(out, result = -EREMCHG);
1201 LASSERT(obj->oo_inode == inode);
1203 osd_add_oi_cache(info, dev, id, fid);
1208 rc1 = osd_scrub_start(env, dev, flags);
1209 LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1210 "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1211 if (rc1 && rc1 != -EALREADY)
1212 GOTO(out, result = -EREMCHG);
1214 if (IS_ERR_OR_NULL(inode) || result)
1215 GOTO(out, result = -EINPROGRESS);
1218 LASSERT(obj->oo_inode == inode);
1220 osd_add_oi_cache(info, dev, id, fid);
1221 osd_oii_insert(dev, oic, true);
1226 if (unlikely(obj->oo_header))
1229 result = osd_check_lma(env, obj);
1233 LASSERTF(id->oii_ino == inode->i_ino &&
1234 id->oii_gen == inode->i_generation,
1235 "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1236 PFID(fid), id->oii_ino, id->oii_gen,
1237 inode->i_ino, inode->i_generation);
1239 saved_ino = inode->i_ino;
1240 saved_gen = inode->i_generation;
1242 if (unlikely(result == -ENODATA)) {
1244 * If the OI scrub updated the OI mapping by race, it
1245 * must be valid. Trust the inode that has no LMA EA.
1250 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1253 * The OI mapping is still there, the inode is still
1254 * valid. It is just becaues the inode has no LMA EA.
1256 if (saved_ino == id->oii_ino &&
1257 saved_gen == id->oii_gen)
1261 * It is the OI scrub updated the OI mapping by race.
1262 * The new OI mapping must be valid.
1270 * "result == -ENOENT" means that the OI mappinghas been
1271 * removed by race, so the inode belongs to other object.
1273 * Others error can be returned directly.
1275 if (result == -ENOENT) {
1278 obj->oo_inode = NULL;
1283 if (result != -EREMCHG)
1289 * if two OST objects map to the same inode, and inode mode is
1290 * (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666), which means it's
1291 * reserved by precreate, and not written yet, in this case, don't
1292 * set inode for the object whose FID mismatch, so that it can create
1293 * inode and not block precreate.
1295 if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) &&
1296 inode->i_mode == (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666)) {
1297 obj->oo_inode = NULL;
1298 GOTO(out, result = 0);
1301 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1303 * "result == -ENOENT" means the cached OI mapping has been removed
1304 * from the OI file by race, above inode belongs to other object.
1306 if (result == -ENOENT) {
1309 obj->oo_inode = NULL;
1310 GOTO(out, result = 0);
1316 if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1322 * It is the OI scrub updated the OI mapping by race.
1323 * The new OI mapping must be valid.
1331 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1332 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1334 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1336 if (lma->lma_compat & LMAC_STRIPE_INFO &&
1338 obj->oo_pfid_in_lma = 1;
1339 if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1341 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1342 } else if (result != -ENODATA) {
1347 obj->oo_compat_dot_created = 1;
1348 obj->oo_compat_dotdot_created = 1;
1350 if (S_ISDIR(inode->i_mode) &&
1351 (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1352 osd_check_lmv(info, dev, inode, oic);
1354 result = osd_attach_jinode(inode);
1359 GOTO(out, result = 0);
1361 LASSERT(!obj->oo_hl_head);
1362 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1364 GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1367 if (result || !obj->oo_inode) {
1368 if (!IS_ERR_OR_NULL(inode))
1371 obj->oo_inode = NULL;
1373 fid_zero(&oic->oic_fid);
1376 LINVRNT(osd_invariant(obj));
1381 * Concurrency: shouldn't matter.
1383 static void osd_object_init0(struct osd_object *obj)
1385 LASSERT(obj->oo_inode != NULL);
1386 obj->oo_dt.do_body_ops = &osd_body_ops;
1387 obj->oo_dt.do_lu.lo_header->loh_attr |=
1388 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1392 * Concurrency: no concurrent access is possible that early in object
1395 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1396 const struct lu_object_conf *conf)
1398 struct osd_object *obj = osd_obj(l);
1401 LINVRNT(osd_invariant(obj));
1403 if (OBD_FAIL_PRECHECK(OBD_FAIL_MDS_LLOG_UMOUNT_RACE) &&
1404 cfs_fail_val == 2) {
1405 struct osd_thread_info *info = osd_oti_get(env);
1406 struct osd_idmap_cache *oic = &info->oti_cache;
1407 /* invalidate thread cache */
1408 memset(&oic->oic_fid, 0, sizeof(oic->oic_fid));
1410 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1411 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1412 l->lo_header->loh_attr |= LOHA_EXISTS;
1416 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1417 obj->oo_dt.do_body_ops = &osd_body_ops_new;
1418 if (result == 0 && obj->oo_inode != NULL) {
1419 struct osd_thread_info *oti = osd_oti_get(env);
1420 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1422 osd_object_init0(obj);
1423 if (unlikely(obj->oo_header))
1426 result = osd_get_lma(oti, obj->oo_inode,
1427 &oti->oti_obj_dentry, loa);
1430 * Convert LMAI flags to lustre LMA flags
1431 * and cache it to oo_lma_flags
1434 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1435 } else if (result == -ENODATA) {
1440 LINVRNT(osd_invariant(obj));
1445 * The first part of oxe_buf is xattr name, and is '\0' terminated.
1446 * The left part is for value, binary mode.
1448 struct osd_xattr_entry {
1449 struct list_head oxe_list;
1453 struct rcu_head oxe_rcu;
1457 static int osd_oxc_get(struct osd_object *obj, const char *name,
1460 struct osd_xattr_entry *tmp;
1461 struct osd_xattr_entry *oxe = NULL;
1462 size_t namelen = strlen(name);
1466 list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1467 if (namelen == tmp->oxe_namelen &&
1468 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1475 GOTO(out, rc = -ENOENT);
1477 if (!oxe->oxe_exist)
1478 GOTO(out, rc = -ENODATA);
1481 rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1484 if (buf->lb_buf == NULL)
1487 if (buf->lb_len < rc)
1488 GOTO(out, rc = -ERANGE);
1490 memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1497 static void osd_oxc_free(struct rcu_head *head)
1499 struct osd_xattr_entry *oxe;
1501 oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1502 OBD_FREE(oxe, oxe->oxe_len);
1505 static void osd_oxc_add(struct osd_object *obj, const char *name,
1506 const char *buf, int buflen)
1508 struct osd_xattr_entry *oxe;
1509 struct osd_xattr_entry *old = NULL;
1510 struct osd_xattr_entry *tmp;
1511 size_t namelen = strlen(name);
1512 size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1514 OBD_ALLOC(oxe, len);
1518 INIT_LIST_HEAD(&oxe->oxe_list);
1520 oxe->oxe_namelen = namelen;
1521 memcpy(oxe->oxe_buf, name, namelen);
1523 LASSERT(buf != NULL);
1524 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1525 oxe->oxe_exist = true;
1527 oxe->oxe_exist = false;
1530 /* this should be rarely called, just remove old and add new */
1531 spin_lock(&obj->oo_guard);
1532 list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1533 if (namelen == tmp->oxe_namelen &&
1534 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1540 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1541 call_rcu(&old->oxe_rcu, osd_oxc_free);
1543 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1545 spin_unlock(&obj->oo_guard);
1548 static void osd_oxc_del(struct osd_object *obj, const char *name)
1550 struct osd_xattr_entry *oxe;
1551 size_t namelen = strlen(name);
1553 spin_lock(&obj->oo_guard);
1554 list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1555 if (namelen == oxe->oxe_namelen &&
1556 strncmp(name, oxe->oxe_buf, namelen) == 0) {
1557 list_del_rcu(&oxe->oxe_list);
1558 call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1562 spin_unlock(&obj->oo_guard);
1565 static void osd_oxc_fini(struct osd_object *obj)
1567 struct osd_xattr_entry *oxe, *next;
1569 list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1570 list_del(&oxe->oxe_list);
1571 OBD_FREE(oxe, oxe->oxe_len);
1576 * Concurrency: no concurrent access is possible that late in object
1579 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1581 struct osd_object *obj = osd_obj(l);
1582 struct lu_object_header *h = obj->oo_header;
1584 LINVRNT(osd_invariant(obj));
1587 dt_object_fini(&obj->oo_dt);
1588 if (obj->oo_hl_head != NULL)
1589 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1590 /* obj doesn't contain an lu_object_header, so we don't need call_rcu */
1593 lu_object_header_fini(h);
1594 OBD_FREE_PRE(h, sizeof(*h), "kfreed");
1595 kfree_rcu(h, loh_rcu);
1600 * Concurrency: no concurrent access is possible that late in object
1603 static void osd_index_fini(struct osd_object *o)
1605 struct iam_container *bag;
1607 if (o->oo_dir != NULL) {
1608 bag = &o->oo_dir->od_container;
1609 if (o->oo_inode != NULL) {
1610 if (bag->ic_object == o->oo_inode)
1611 iam_container_fini(bag);
1613 OBD_FREE_PTR(o->oo_dir);
1619 * Concurrency: no concurrent access is possible that late in object
1620 * life-cycle (for all existing callers, that is. New callers have to provide
1621 * their own locking.)
1623 static int osd_inode_unlinked(const struct inode *inode)
1625 return inode->i_nlink == 0;
1629 OSD_TXN_OI_DELETE_CREDITS = 20,
1630 OSD_TXN_INODE_DELETE_CREDITS = 20
1637 #if OSD_THANDLE_STATS
1639 * Set time when the handle is allocated
1641 static void osd_th_alloced(struct osd_thandle *oth)
1643 oth->oth_alloced = ktime_get();
1647 * Set time when the handle started
1649 static void osd_th_started(struct osd_thandle *oth)
1651 oth->oth_started = ktime_get();
1655 * Check whether the we deal with this handle for too long.
1657 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1658 ktime_t alloced, ktime_t started,
1661 ktime_t now = ktime_get();
1663 LASSERT(dev != NULL);
1665 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1666 ktime_us_delta(started, alloced));
1667 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1668 ktime_us_delta(closed, started));
1669 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1670 ktime_us_delta(now, closed));
1672 if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1673 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1674 oth, now, alloced, started, closed);
1675 libcfs_debug_dumpstack(NULL);
1679 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
1681 ktime_t __closed = ktime_get(); \
1682 ktime_t __alloced = oth->oth_alloced; \
1683 ktime_t __started = oth->oth_started; \
1686 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
1689 #else /* OSD_THANDLE_STATS */
1691 #define osd_th_alloced(h) do {} while(0)
1692 #define osd_th_started(h) do {} while(0)
1693 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
1695 #endif /* OSD_THANDLE_STATS */
1698 * Concurrency: doesn't access mutable data.
1700 static int osd_param_is_not_sane(const struct osd_device *dev,
1701 const struct thandle *th)
1703 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1705 return oh->ot_credits > osd_transaction_size(dev);
1709 * Concurrency: shouldn't matter.
1711 static void osd_trans_commit_cb(struct super_block *sb,
1712 struct ldiskfs_journal_cb_entry *jcb, int error)
1714 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1715 struct thandle *th = &oh->ot_super;
1716 struct lu_device *lud = &th->th_dev->dd_lu_dev;
1717 struct dt_txn_commit_cb *dcb, *tmp;
1719 LASSERT(oh->ot_handle == NULL);
1722 CERROR("transaction @0x%p commit error: %d\n", th, error);
1724 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DELAY_TRANS, 40);
1725 /* call per-transaction callbacks if any */
1726 list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1728 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1729 "commit callback entry: magic=%x name='%s'\n",
1730 dcb->dcb_magic, dcb->dcb_name);
1731 list_del_init(&dcb->dcb_linkage);
1732 dcb->dcb_func(NULL, th, dcb, error);
1735 lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1742 static struct thandle *osd_trans_create(const struct lu_env *env,
1743 struct dt_device *d)
1745 struct osd_thread_info *oti = osd_oti_get(env);
1746 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1747 struct osd_thandle *oh;
1753 CERROR("%s: someone try to start transaction under "
1754 "readonly mode, should be disabled.\n",
1755 osd_name(osd_dt_dev(d)));
1757 RETURN(ERR_PTR(-EROFS));
1760 /* on pending IO in this thread should left from prev. request */
1761 LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1763 sb_start_write(osd_sb(osd_dt_dev(d)));
1765 OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1767 sb_end_write(osd_sb(osd_dt_dev(d)));
1768 RETURN(ERR_PTR(-ENOMEM));
1771 oh->ot_quota_trans = &oti->oti_quota_trans;
1772 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1777 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1778 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1779 INIT_LIST_HEAD(&oh->ot_trunc_locks);
1782 memset(oti->oti_declare_ops, 0,
1783 sizeof(oti->oti_declare_ops));
1784 memset(oti->oti_declare_ops_cred, 0,
1785 sizeof(oti->oti_declare_ops_cred));
1786 memset(oti->oti_declare_ops_used, 0,
1787 sizeof(oti->oti_declare_ops_used));
1789 oti->oti_ins_cache_depth++;
1794 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1796 struct osd_thread_info *oti = osd_oti_get(env);
1797 struct osd_thandle *oh;
1799 oh = container_of0(th, struct osd_thandle, ot_super);
1800 LASSERT(oh != NULL);
1802 CWARN(" create: %u/%u/%u, destroy: %u/%u/%u\n",
1803 oti->oti_declare_ops[OSD_OT_CREATE],
1804 oti->oti_declare_ops_cred[OSD_OT_CREATE],
1805 oti->oti_declare_ops_used[OSD_OT_CREATE],
1806 oti->oti_declare_ops[OSD_OT_DESTROY],
1807 oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1808 oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1809 CWARN(" attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1810 oti->oti_declare_ops[OSD_OT_ATTR_SET],
1811 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1812 oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1813 oti->oti_declare_ops[OSD_OT_XATTR_SET],
1814 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1815 oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1816 CWARN(" write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1817 oti->oti_declare_ops[OSD_OT_WRITE],
1818 oti->oti_declare_ops_cred[OSD_OT_WRITE],
1819 oti->oti_declare_ops_used[OSD_OT_WRITE],
1820 oti->oti_declare_ops[OSD_OT_PUNCH],
1821 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1822 oti->oti_declare_ops_used[OSD_OT_PUNCH],
1823 oti->oti_declare_ops[OSD_OT_QUOTA],
1824 oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1825 oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1826 CWARN(" insert: %u/%u/%u, delete: %u/%u/%u\n",
1827 oti->oti_declare_ops[OSD_OT_INSERT],
1828 oti->oti_declare_ops_cred[OSD_OT_INSERT],
1829 oti->oti_declare_ops_used[OSD_OT_INSERT],
1830 oti->oti_declare_ops[OSD_OT_DELETE],
1831 oti->oti_declare_ops_cred[OSD_OT_DELETE],
1832 oti->oti_declare_ops_used[OSD_OT_DELETE]);
1833 CWARN(" ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1834 oti->oti_declare_ops[OSD_OT_REF_ADD],
1835 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1836 oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1837 oti->oti_declare_ops[OSD_OT_REF_DEL],
1838 oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1839 oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1843 * Concurrency: shouldn't matter.
1845 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1848 struct osd_thread_info *oti = osd_oti_get(env);
1849 struct osd_device *dev = osd_dt_dev(d);
1851 struct osd_thandle *oh;
1856 LASSERT(current->journal_info == NULL);
1858 oh = container_of0(th, struct osd_thandle, ot_super);
1859 LASSERT(oh != NULL);
1860 LASSERT(oh->ot_handle == NULL);
1862 rc = dt_txn_hook_start(env, d, th);
1866 if (unlikely(osd_param_is_not_sane(dev, th))) {
1867 static unsigned long last_printed;
1868 static int last_credits;
1871 * don't make noise on a tiny testing systems
1872 * actual credits misuse will be caught anyway
1874 if (last_credits != oh->ot_credits &&
1875 time_after(jiffies, last_printed +
1876 cfs_time_seconds(60)) &&
1877 osd_transaction_size(dev) > 512) {
1878 CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1879 oh->ot_credits, osd_transaction_size(dev));
1880 osd_trans_dump_creds(env, th);
1881 libcfs_debug_dumpstack(NULL);
1882 last_credits = oh->ot_credits;
1883 last_printed = jiffies;
1886 * XXX Limit the credits to 'max_transaction_buffers', and
1887 * let the underlying filesystem to catch the error if
1888 * we really need so many credits.
1890 * This should be removed when we can calculate the
1891 * credits precisely.
1893 oh->ot_credits = osd_transaction_size(dev);
1894 } else if (ldiskfs_track_declares_assert != 0) {
1896 * reserve few credits to prevent an assertion in JBD
1897 * our debugging mechanism will be able to detected
1898 * overuse. this can help to debug single-update
1901 oh->ot_credits += 10;
1902 if (unlikely(osd_param_is_not_sane(dev, th)))
1903 oh->ot_credits = osd_transaction_size(dev);
1906 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1907 GOTO(out, rc = -EIO);
1910 * XXX temporary stuff. Some abstraction layer should
1913 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1917 LASSERT(oti->oti_txns == 0);
1919 lu_device_get(&d->dd_lu_dev);
1920 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1931 static int osd_seq_exists(const struct lu_env *env,
1932 struct osd_device *osd, u64 seq)
1934 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1935 struct seq_server_site *ss = osd_seq_site(osd);
1940 LASSERT(ss != NULL);
1941 LASSERT(ss->ss_server_fld != NULL);
1943 rc = osd_fld_lookup(env, osd, seq, range);
1946 CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1947 osd_name(osd), seq, rc);
1951 RETURN(ss->ss_node_id == range->lsr_index);
1954 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1956 struct dt_txn_commit_cb *dcb;
1957 struct dt_txn_commit_cb *tmp;
1959 /* call per-transaction stop callbacks if any */
1960 list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1962 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1963 "commit callback entry: magic=%x name='%s'\n",
1964 dcb->dcb_magic, dcb->dcb_name);
1965 list_del_init(&dcb->dcb_linkage);
1966 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1971 * Concurrency: shouldn't matter.
1973 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1976 struct osd_thread_info *oti = osd_oti_get(env);
1977 struct osd_thandle *oh;
1978 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1979 struct osd_device *osd = osd_dt_dev(th->th_dev);
1980 struct qsd_instance *qsd = osd_def_qsd(osd);
1981 struct lquota_trans *qtrans;
1982 LIST_HEAD(truncates);
1983 int rc = 0, remove_agents = 0;
1987 oh = container_of0(th, struct osd_thandle, ot_super);
1989 remove_agents = oh->ot_remove_agents;
1991 qtrans = oh->ot_quota_trans;
1992 oh->ot_quota_trans = NULL;
1994 /* move locks to local list, stop tx, execute truncates */
1995 list_splice(&oh->ot_trunc_locks, &truncates);
1997 if (oh->ot_handle != NULL) {
2000 handle_t *hdl = oh->ot_handle;
2003 * add commit callback
2004 * notice we don't do this in osd_trans_start()
2005 * as underlying transaction can change during truncate
2007 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
2010 LASSERT(oti->oti_txns == 1);
2013 rc = dt_txn_hook_stop(env, th);
2015 CERROR("%s: failed in transaction hook: rc = %d\n",
2018 osd_trans_stop_cb(oh, rc);
2019 /* hook functions might modify th_sync */
2020 hdl->h_sync = th->th_sync;
2022 oh->ot_handle = NULL;
2023 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2025 CERROR("%s: failed to stop transaction: rc = %d\n",
2026 osd_name(osd), rc2);
2030 osd_process_truncates(&truncates);
2032 osd_trans_stop_cb(oh, th->th_result);
2036 osd_trunc_unlock_all(&truncates);
2038 /* inform the quota slave device that the transaction is stopping */
2039 qsd_op_end(env, qsd, qtrans);
2042 * as we want IO to journal and data IO be concurrent, we don't block
2043 * awaiting data IO completion in osd_do_bio(), instead we wait here
2044 * once transaction is submitted to the journal. all reqular requests
2045 * don't do direct IO (except read/write), thus this wait_event becomes
2048 * IMPORTANT: we have to wait till any IO submited by the thread is
2049 * completed otherwise iobuf may be corrupted by different request
2051 wait_event(iobuf->dr_wait,
2052 atomic_read(&iobuf->dr_numreqs) == 0);
2053 osd_fini_iobuf(osd, iobuf);
2055 rc = iobuf->dr_error;
2057 if (unlikely(remove_agents != 0))
2058 osd_process_scheduled_agent_removals(env, osd);
2060 oti->oti_ins_cache_depth--;
2061 /* reset OI cache for safety */
2062 if (oti->oti_ins_cache_depth == 0)
2063 oti->oti_ins_cache_used = 0;
2065 sb_end_write(osd_sb(osd));
2070 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2072 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
2075 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2076 LASSERT(&dcb->dcb_func != NULL);
2077 if (dcb->dcb_flags & DCB_TRANS_STOP)
2078 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2080 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2086 * Called just before object is freed. Releases all resources except for
2087 * object itself (that is released by osd_object_free()).
2089 * Concurrency: no concurrent access is possible that late in object
2092 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2094 struct osd_object *obj = osd_obj(l);
2095 struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2096 struct inode *inode = obj->oo_inode;
2101 LINVRNT(osd_invariant(obj));
2104 * If object is unlinked remove fid->ino mapping from object index.
2107 osd_index_fini(obj);
2112 uid = i_uid_read(inode);
2113 gid = i_gid_read(inode);
2114 projid = i_projid_read(inode);
2116 obj->oo_inode = NULL;
2119 /* do not rebalance quota if the caller needs to release memory
2120 * otherwise qsd_refresh_usage() may went into a new ldiskfs
2121 * transaction and risk to deadlock - LU-12178 */
2122 if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2125 if (!obj->oo_header && qsd) {
2126 struct osd_thread_info *info = osd_oti_get(env);
2127 struct lquota_id_info *qi = &info->oti_qi;
2129 /* Release granted quota to master if necessary */
2130 qi->lqi_id.qid_uid = uid;
2131 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2133 qi->lqi_id.qid_uid = gid;
2134 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2136 qi->lqi_id.qid_uid = projid;
2137 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2142 * Concurrency: ->loo_object_release() is called under site spin-lock.
2144 static void osd_object_release(const struct lu_env *env,
2145 struct lu_object *l)
2147 struct osd_object *o = osd_obj(l);
2150 * nobody should be releasing a non-destroyed object with nlink=0
2151 * the API allows this, but ldiskfs doesn't like and then report
2152 * this inode as deleted
2154 LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2155 o->oo_inode->i_nlink == 0));
2159 * Concurrency: shouldn't matter.
2161 static int osd_object_print(const struct lu_env *env, void *cookie,
2162 lu_printer_t p, const struct lu_object *l)
2164 struct osd_object *o = osd_obj(l);
2165 struct iam_descr *d;
2167 if (o->oo_dir != NULL)
2168 d = o->oo_dir->od_container.ic_descr;
2171 return (*p)(env, cookie,
2172 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2174 o->oo_inode ? o->oo_inode->i_ino : 0UL,
2175 o->oo_inode ? o->oo_inode->i_generation : 0,
2176 d ? d->id_ops->id_name : "plain");
2180 * Concurrency: shouldn't matter.
2182 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2183 struct obd_statfs *sfs, struct obd_statfs_info *info)
2185 struct osd_device *osd = osd_dt_dev(d);
2186 struct super_block *sb = osd_sb(osd);
2187 struct kstatfs *ksfs;
2191 if (unlikely(osd->od_mnt == NULL))
2192 return -EINPROGRESS;
2194 /* osd_lproc.c call this without env, allocate ksfs for that case */
2195 if (unlikely(env == NULL)) {
2196 OBD_ALLOC_PTR(ksfs);
2200 ksfs = &osd_oti_get(env)->oti_ksfs;
2203 result = sb->s_op->statfs(sb->s_root, ksfs);
2207 statfs_pack(sfs, ksfs);
2208 if (unlikely(sb->s_flags & SB_RDONLY))
2209 sfs->os_state |= OS_STATE_READONLY;
2211 sfs->os_state |= osd->od_nonrotational ? OS_STATE_NONROT : 0;
2213 if (ldiskfs_has_feature_extents(sb))
2214 sfs->os_maxbytes = sb->s_maxbytes;
2216 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2219 * Reserve some space so to avoid fragmenting the filesystem too much.
2220 * Fragmentation not only impacts performance, but can also increase
2221 * metadata overhead significantly, causing grant calculation to be
2224 * Reserve 0.78% of total space, at least 8MB for small filesystems.
2226 BUILD_BUG_ON(OSD_STATFS_RESERVED <= LDISKFS_MAX_BLOCK_SIZE);
2227 reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2228 if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2229 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2231 sfs->os_blocks -= reserved;
2232 sfs->os_bfree -= min(reserved, sfs->os_bfree);
2233 sfs->os_bavail -= min(reserved, sfs->os_bavail);
2236 if (unlikely(env == NULL))
2242 * Estimate space needed for file creations. We assume the largest filename
2243 * which is 2^64 - 1, hence a filename of 20 chars.
2244 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2246 #ifdef __LDISKFS_DIR_REC_LEN
2247 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2249 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2253 * Concurrency: doesn't access mutable data.
2255 static void osd_conf_get(const struct lu_env *env,
2256 const struct dt_device *dev,
2257 struct dt_device_param *param)
2259 struct osd_device *d = osd_dt_dev(dev);
2260 struct super_block *sb = osd_sb(d);
2261 struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2266 * XXX should be taken from not-yet-existing fs abstraction layer.
2268 param->ddp_max_name_len = LDISKFS_NAME_LEN;
2269 param->ddp_max_nlink = LDISKFS_LINK_MAX;
2270 param->ddp_symlink_max = sb->s_blocksize;
2271 param->ddp_mount_type = LDD_MT_LDISKFS;
2272 if (ldiskfs_has_feature_extents(sb))
2273 param->ddp_maxbytes = sb->s_maxbytes;
2275 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2277 * inode are statically allocated, so per-inode space consumption
2278 * is the space consumed by the directory entry
2280 param->ddp_inodespace = PER_OBJ_USAGE;
2282 * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2283 * is 128MB) which is unlikely to be hit in real life. Report a smaller
2284 * maximum length to not under-count the actual number of extents
2285 * needed for writing a file if there are sub-optimal block allocations.
2287 param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2288 /* worst-case extent insertion metadata overhead */
2289 param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2290 param->ddp_mntopts = 0;
2291 if (test_opt(sb, XATTR_USER))
2292 param->ddp_mntopts |= MNTOPT_USERXATTR;
2293 if (test_opt(sb, POSIX_ACL))
2294 param->ddp_mntopts |= MNTOPT_ACL;
2297 * LOD might calculate the max stripe count based on max_ea_size,
2298 * so we need take account in the overhead as well,
2299 * xattr_header + magic + xattr_entry_head
2301 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2302 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2304 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2305 if (ldiskfs_has_feature_ea_inode(sb))
2306 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2310 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2312 if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE)
2313 param->ddp_max_ea_size = OBD_MAX_EA_SIZE;
2316 * Preferred RPC size for efficient disk IO. 4MB shows good
2317 * all-around performance for ldiskfs, but use bigalloc chunk size
2318 * by default if larger.
2320 #if defined(LDISKFS_CLUSTER_SIZE)
2321 if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2322 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2325 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2327 param->ddp_t10_cksum_type = 0;
2329 unsigned short interval = blk_integrity_interval(bi);
2330 name = blk_integrity_name(bi);
2338 if (strncmp(name, "T10-DIF-TYPE",
2339 sizeof("T10-DIF-TYPE") - 1) == 0) {
2340 /* also skip "1/3-" at end */
2341 const int type_off = sizeof("T10-DIF-TYPE.");
2342 char type_number = name[type_off - 2];
2344 if (interval != 512 && interval != 4096) {
2345 CERROR("%s: unsupported T10PI sector size %u\n",
2346 d->od_svname, interval);
2347 } else if (type_number != '1' && type_number != '3') {
2348 CERROR("%s: unsupported T10PI type %s\n",
2349 d->od_svname, name);
2350 } else if (strcmp(name + type_off, "CRC") == 0) {
2351 d->od_t10_type = type_number == '1' ?
2352 OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2353 param->ddp_t10_cksum_type = interval == 512 ?
2354 OBD_CKSUM_T10CRC512 :
2356 } else if (strcmp(name + type_off, "IP") == 0) {
2357 d->od_t10_type = type_number == '1' ?
2358 OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2359 param->ddp_t10_cksum_type = interval == 512 ?
2360 OBD_CKSUM_T10IP512 :
2363 CERROR("%s: unsupported checksum type of "
2365 d->od_svname, name);
2369 CERROR("%s: unsupported T10PI type '%s'",
2370 d->od_svname, name);
2375 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2377 return osd_sb(osd_dt_dev(d));
2381 * Concurrency: shouldn't matter.
2383 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2386 struct super_block *s = osd_sb(osd_dt_dev(d));
2389 down_read(&s->s_umount);
2390 rc = s->s_op->sync_fs(s, 1);
2391 up_read(&s->s_umount);
2393 CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2400 * Start commit for OSD device.
2402 * An implementation of dt_commit_async method for OSD device.
2403 * Asychronously starts underlayng fs sync and thereby a transaction
2406 * \param env environment
2407 * \param d dt device
2409 * \see dt_device_operations
2411 static int osd_commit_async(const struct lu_env *env,
2412 struct dt_device *d)
2414 struct super_block *s = osd_sb(osd_dt_dev(d));
2419 CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2420 down_read(&s->s_umount);
2421 rc = s->s_op->sync_fs(s, 0);
2422 up_read(&s->s_umount);
2427 static int (*priv_security_file_alloc)(struct file *file);
2429 int osd_security_file_alloc(struct file *file)
2431 if (priv_security_file_alloc)
2432 return priv_security_file_alloc(file);
2437 * Concurrency: shouldn't matter.
2439 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2441 struct super_block *sb = osd_sb(osd_dt_dev(d));
2442 struct block_device *dev = sb->s_bdev;
2443 int rc = -EOPNOTSUPP;
2447 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2448 osd_dt_dev(d)->od_svname, (long)dev, rc);
2454 * Note: we do not count into QUOTA here.
2455 * If we mount with --data_journal we may need more.
2457 const int osd_dto_credits_noquota[DTO_NR] = {
2460 * INDEX_EXTRA_TRANS_BLOCKS(8) +
2461 * SINGLEDATA_TRANS_BLOCKS(8)
2462 * XXX Note: maybe iam need more, since iam have more level than
2465 [DTO_INDEX_INSERT] = 16,
2468 * just modify a single entry, probably merge few within a block
2470 [DTO_INDEX_DELETE] = 1,
2474 [DTO_INDEX_UPDATE] = 16,
2476 * 4(inode, inode bits, groups, GDT)
2477 * notice: OI updates are counted separately with DTO_INDEX_INSERT
2479 [DTO_OBJECT_CREATE] = 4,
2481 * 4(inode, inode bits, groups, GDT)
2482 * notice: OI updates are counted separately with DTO_INDEX_DELETE
2484 [DTO_OBJECT_DELETE] = 4,
2486 * Attr set credits (inode)
2488 [DTO_ATTR_SET_BASE] = 1,
2490 * Xattr set. The same as xattr of EXT3.
2491 * DATA_TRANS_BLOCKS(14)
2492 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2493 * are also counted in. Do not know why?
2495 [DTO_XATTR_SET] = 14,
2497 * credits for inode change during write.
2499 [DTO_WRITE_BASE] = 3,
2501 * credits for single block write.
2503 [DTO_WRITE_BLOCK] = 14,
2505 * Attr set credits for chown.
2506 * This is extra credits for setattr, and it is null without quota
2508 [DTO_ATTR_SET_CHOWN] = 0
2511 static const struct dt_device_operations osd_dt_ops = {
2512 .dt_root_get = osd_root_get,
2513 .dt_statfs = osd_statfs,
2514 .dt_trans_create = osd_trans_create,
2515 .dt_trans_start = osd_trans_start,
2516 .dt_trans_stop = osd_trans_stop,
2517 .dt_trans_cb_add = osd_trans_cb_add,
2518 .dt_conf_get = osd_conf_get,
2519 .dt_mnt_sb_get = osd_mnt_sb_get,
2520 .dt_sync = osd_sync,
2522 .dt_commit_async = osd_commit_async,
2525 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2528 struct osd_object *obj = osd_dt_obj(dt);
2529 struct osd_thread_info *oti = osd_oti_get(env);
2531 LINVRNT(osd_invariant(obj));
2533 LASSERT(obj->oo_owner != env);
2534 down_read_nested(&obj->oo_sem, role);
2536 LASSERT(obj->oo_owner == NULL);
2540 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2543 struct osd_object *obj = osd_dt_obj(dt);
2544 struct osd_thread_info *oti = osd_oti_get(env);
2546 LINVRNT(osd_invariant(obj));
2548 LASSERT(obj->oo_owner != env);
2549 down_write_nested(&obj->oo_sem, role);
2551 LASSERT(obj->oo_owner == NULL);
2552 obj->oo_owner = env;
2556 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2558 struct osd_object *obj = osd_dt_obj(dt);
2559 struct osd_thread_info *oti = osd_oti_get(env);
2561 LINVRNT(osd_invariant(obj));
2563 LASSERT(oti->oti_r_locks > 0);
2565 up_read(&obj->oo_sem);
2568 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2570 struct osd_object *obj = osd_dt_obj(dt);
2571 struct osd_thread_info *oti = osd_oti_get(env);
2573 LINVRNT(osd_invariant(obj));
2575 LASSERT(obj->oo_owner == env);
2576 LASSERT(oti->oti_w_locks > 0);
2578 obj->oo_owner = NULL;
2579 up_write(&obj->oo_sem);
2582 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2584 struct osd_object *obj = osd_dt_obj(dt);
2586 LINVRNT(osd_invariant(obj));
2588 return obj->oo_owner == env;
2591 static void osd_inode_getattr(const struct lu_env *env,
2592 struct inode *inode, struct lu_attr *attr)
2594 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2595 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2596 LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2597 LA_BLKSIZE | LA_TYPE | LA_BTIME;
2599 attr->la_atime = inode->i_atime.tv_sec;
2600 attr->la_mtime = inode->i_mtime.tv_sec;
2601 attr->la_ctime = inode->i_ctime.tv_sec;
2602 attr->la_btime = LDISKFS_I(inode)->i_crtime.tv_sec;
2603 attr->la_mode = inode->i_mode;
2604 attr->la_size = i_size_read(inode);
2605 attr->la_blocks = inode->i_blocks;
2606 attr->la_uid = i_uid_read(inode);
2607 attr->la_gid = i_gid_read(inode);
2608 attr->la_projid = i_projid_read(inode);
2609 attr->la_flags = ll_inode_to_ext_flags(inode->i_flags);
2610 attr->la_nlink = inode->i_nlink;
2611 attr->la_rdev = inode->i_rdev;
2612 attr->la_blksize = 1 << inode->i_blkbits;
2613 attr->la_blkbits = inode->i_blkbits;
2615 * Ext4 did not transfer inherit flags from raw inode
2616 * to inode flags, and ext4 internally test raw inode
2617 * @i_flags directly. Instead of patching ext4, we do it here.
2619 if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2620 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2623 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2624 struct lu_attr *attr)
2626 struct osd_object *obj = osd_dt_obj(dt);
2628 if (unlikely(!dt_object_exists(dt)))
2630 if (unlikely(obj->oo_destroyed))
2633 LASSERT(!dt_object_remote(dt));
2634 LINVRNT(osd_invariant(obj));
2636 spin_lock(&obj->oo_guard);
2637 osd_inode_getattr(env, obj->oo_inode, attr);
2638 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2639 attr->la_valid |= LA_FLAGS;
2640 attr->la_flags |= LUSTRE_ORPHAN_FL;
2642 spin_unlock(&obj->oo_guard);
2647 static int osd_declare_attr_qid(const struct lu_env *env,
2648 struct osd_object *obj,
2649 struct osd_thandle *oh, long long bspace,
2650 qid_t old_id, qid_t new_id, bool enforce,
2651 unsigned int type, bool ignore_edquot)
2654 struct osd_thread_info *info = osd_oti_get(env);
2655 struct lquota_id_info *qi = &info->oti_qi;
2657 qi->lqi_type = type;
2658 /* inode accounting */
2659 qi->lqi_is_blk = false;
2661 /* one more inode for the new id ... */
2662 qi->lqi_id.qid_uid = new_id;
2664 /* Reserve credits for the new id */
2665 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2666 if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2671 /* and one less inode for the current id */
2672 qi->lqi_id.qid_uid = old_id;
2674 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2675 if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2680 /* block accounting */
2681 qi->lqi_is_blk = true;
2683 /* more blocks for the new id ... */
2684 qi->lqi_id.qid_uid = new_id;
2685 qi->lqi_space = bspace;
2687 * Credits for the new uid has been reserved, re-use "obj"
2688 * to save credit reservation.
2690 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2691 if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2696 /* and finally less blocks for the current uid */
2697 qi->lqi_id.qid_uid = old_id;
2698 qi->lqi_space = -bspace;
2699 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2700 if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2706 static int osd_declare_attr_set(const struct lu_env *env,
2707 struct dt_object *dt,
2708 const struct lu_attr *attr,
2709 struct thandle *handle)
2711 struct osd_thandle *oh;
2712 struct osd_object *obj;
2721 LASSERT(dt != NULL);
2722 LASSERT(handle != NULL);
2724 obj = osd_dt_obj(dt);
2725 LASSERT(osd_invariant(obj));
2727 oh = container_of0(handle, struct osd_thandle, ot_super);
2728 LASSERT(oh->ot_handle == NULL);
2730 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2731 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2733 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2734 osd_dto_credits_noquota[DTO_XATTR_SET]);
2736 if (attr == NULL || obj->oo_inode == NULL)
2739 bspace = obj->oo_inode->i_blocks << 9;
2740 bspace = toqb(bspace);
2743 * Changing ownership is always preformed by super user, it should not
2744 * fail with EDQUOT unless required explicitly.
2746 * We still need to call the osd_declare_qid() to calculate the journal
2747 * credits for updating quota accounting files and to trigger quota
2748 * space adjustment once the operation is completed.
2750 if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2751 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2754 CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2755 "(the quota space is %lld)\n",
2756 obj->oo_inode->i_sb->s_id, attr->la_uid,
2757 attr->la_gid, bspace);
2760 uid = i_uid_read(obj->oo_inode);
2761 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2762 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2763 attr->la_uid, enforce, USRQUOTA,
2768 gid = i_gid_read(obj->oo_inode);
2769 CDEBUG(D_QUOTA, "declare uid %d -> %d gid %d -> %d\n", uid,
2770 attr->la_uid, gid, attr->la_gid);
2771 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2772 rc = osd_declare_attr_qid(env, obj, oh, bspace, gid,
2773 attr->la_gid, enforce, GRPQUOTA,
2779 #ifdef HAVE_PROJECT_QUOTA
2780 if (attr->la_valid & LA_PROJID) {
2781 __u32 projid = i_projid_read(obj->oo_inode);
2783 enforce = (attr->la_valid & LA_PROJID) &&
2784 (attr->la_projid != projid);
2785 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2786 (qid_t)projid, (qid_t)attr->la_projid,
2787 enforce, PRJQUOTA, true);
2795 static int osd_inode_setattr(const struct lu_env *env,
2796 struct inode *inode, const struct lu_attr *attr)
2798 __u64 bits = attr->la_valid;
2800 /* Only allow set size for regular file */
2801 if (!S_ISREG(inode->i_mode))
2802 bits &= ~(LA_SIZE | LA_BLOCKS);
2807 if (bits & LA_ATIME)
2808 inode->i_atime = osd_inode_time(inode, attr->la_atime);
2809 if (bits & LA_CTIME)
2810 inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
2811 if (bits & LA_MTIME)
2812 inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
2813 if (bits & LA_SIZE) {
2814 spin_lock(&inode->i_lock);
2815 LDISKFS_I(inode)->i_disksize = attr->la_size;
2816 i_size_write(inode, attr->la_size);
2817 spin_unlock(&inode->i_lock);
2821 * OSD should not change "i_blocks" which is used by quota.
2822 * "i_blocks" should be changed by ldiskfs only.
2825 inode->i_mode = (inode->i_mode & S_IFMT) |
2826 (attr->la_mode & ~S_IFMT);
2828 i_uid_write(inode, attr->la_uid);
2830 i_gid_write(inode, attr->la_gid);
2831 if (bits & LA_PROJID)
2832 i_projid_write(inode, attr->la_projid);
2833 if (bits & LA_NLINK)
2834 set_nlink(inode, attr->la_nlink);
2836 inode->i_rdev = attr->la_rdev;
2838 if (bits & LA_FLAGS) {
2839 /* always keep S_NOCMTIME */
2840 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2843 * Ext4 did not transfer inherit flags from
2844 * @inode->i_flags to raw inode i_flags when writing
2845 * flags, we do it explictly here.
2847 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2848 LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2850 LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2855 #ifdef HAVE_PROJECT_QUOTA
2856 static int osd_transfer_project(struct inode *inode, __u32 projid)
2858 struct super_block *sb = inode->i_sb;
2859 struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2862 struct ldiskfs_iloc iloc;
2863 struct ldiskfs_inode *raw_inode;
2864 struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2866 if (!ldiskfs_has_feature_project(sb)) {
2867 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2868 == LDISKFS_DEF_PROJID);
2869 if (projid != LDISKFS_DEF_PROJID)
2875 if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2878 kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2879 if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2882 err = ldiskfs_get_inode_loc(inode, &iloc);
2886 raw_inode = ldiskfs_raw_inode(&iloc);
2887 if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2894 dquot_initialize(inode);
2895 transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2896 if (transfer_to[PRJQUOTA]) {
2897 err = __dquot_transfer(inode, transfer_to);
2898 dqput(transfer_to[PRJQUOTA]);
2907 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2911 if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2912 (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2916 "executing dquot_transfer inode %ld uid %d -> %d gid %d -> %d\n",
2917 inode->i_ino, i_uid_read(inode), attr->la_uid,
2918 i_gid_read(inode), attr->la_gid);
2920 dquot_initialize(inode);
2922 if (attr->la_valid & LA_UID)
2923 iattr.ia_valid |= ATTR_UID;
2924 if (attr->la_valid & LA_GID)
2925 iattr.ia_valid |= ATTR_GID;
2926 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2927 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2929 rc = dquot_transfer(inode, &iattr);
2931 CERROR("%s: quota transfer failed: rc = %d. Is quota "
2932 "enforcement enabled on the ldiskfs "
2933 "filesystem?\n", inode->i_sb->s_id, rc);
2938 /* Handle project id transfer here properly */
2939 if (attr->la_valid & LA_PROJID &&
2940 attr->la_projid != i_projid_read(inode)) {
2941 #ifdef HAVE_PROJECT_QUOTA
2942 rc = osd_transfer_project(inode, attr->la_projid);
2947 CERROR("%s: quota transfer failed: rc = %d. Is project "
2948 "enforcement enabled on the ldiskfs "
2949 "filesystem?\n", inode->i_sb->s_id, rc);
2956 static int osd_attr_set(const struct lu_env *env,
2957 struct dt_object *dt,
2958 const struct lu_attr *attr,
2959 struct thandle *handle)
2961 struct osd_object *obj = osd_dt_obj(dt);
2962 struct inode *inode;
2965 if (!dt_object_exists(dt))
2968 LASSERT(handle != NULL);
2969 LASSERT(!dt_object_remote(dt));
2970 LASSERT(osd_invariant(obj));
2972 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2974 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
2975 !osd_obj2dev(obj)->od_is_ost) {
2976 struct osd_thread_info *oti = osd_oti_get(env);
2977 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
2978 struct lu_fid *fid1 = &oti->oti_fid;
2979 struct osd_inode_id *id = &oti->oti_id;
2980 struct iam_path_descr *ipd;
2981 struct iam_container *bag;
2982 struct osd_thandle *oh;
2985 fid_cpu_to_be(fid1, fid0);
2986 memset(id, 1, sizeof(*id));
2987 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2988 fid0)->oi_dir.od_container;
2989 ipd = osd_idx_ipd_get(env, bag);
2990 if (unlikely(ipd == NULL))
2993 oh = container_of0(handle, struct osd_thandle, ot_super);
2994 rc = iam_update(oh->ot_handle, bag,
2995 (const struct iam_key *)fid1,
2996 (const struct iam_rec *)id, ipd);
2997 osd_ipd_put(env, bag, ipd);
2998 return(rc > 0 ? 0 : rc);
3001 inode = obj->oo_inode;
3003 rc = osd_quota_transfer(inode, attr);
3007 spin_lock(&obj->oo_guard);
3008 rc = osd_inode_setattr(env, inode, attr);
3009 spin_unlock(&obj->oo_guard);
3013 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3015 osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3017 if (!(attr->la_valid & LA_FLAGS))
3020 /* Let's check if there are extra flags need to be set into LMA */
3021 if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3022 struct osd_thread_info *info = osd_oti_get(env);
3023 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3025 LASSERT(!obj->oo_pfid_in_lma);
3027 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3028 &info->oti_ost_attrs);
3032 lma->lma_incompat |=
3033 lustre_to_lma_flags(attr->la_flags);
3034 lustre_lma_swab(lma);
3036 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3038 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3039 lma, sizeof(*lma), XATTR_REPLACE);
3041 struct osd_device *osd = osd_obj2dev(obj);
3043 CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3044 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3045 lma->lma_incompat, rc);
3048 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3050 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3057 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3058 struct osd_object *obj,
3059 const char *name, const int namelen)
3061 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3064 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3065 umode_t mode, struct dt_allocation_hint *hint,
3066 struct thandle *th, struct lu_attr *attr)
3069 struct osd_device *osd = osd_obj2dev(obj);
3070 struct osd_thandle *oth;
3071 struct dt_object *parent = NULL;
3072 struct inode *inode;
3073 uid_t owner[2] = {0, 0};
3075 if (attr->la_valid & LA_UID)
3076 owner[0] = attr->la_uid;
3077 if (attr->la_valid & LA_GID)
3078 owner[1] = attr->la_gid;
3080 LINVRNT(osd_invariant(obj));
3081 LASSERT(obj->oo_inode == NULL);
3082 LASSERT(obj->oo_hl_head == NULL);
3084 if (S_ISDIR(mode) && ldiskfs_pdo) {
3086 ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3087 if (obj->oo_hl_head == NULL)
3091 oth = container_of(th, struct osd_thandle, ot_super);
3092 LASSERT(oth->ot_handle->h_transaction != NULL);
3094 if (hint != NULL && hint->dah_parent != NULL &&
3095 !dt_object_remote(hint->dah_parent))
3096 parent = hint->dah_parent;
3098 inode = ldiskfs_create_inode(oth->ot_handle,
3099 parent ? osd_dt_obj(parent)->oo_inode :
3100 osd_sb(osd)->s_root->d_inode,
3102 if (!IS_ERR(inode)) {
3103 /* Do not update file c/mtime in ldiskfs. */
3104 inode->i_flags |= S_NOCMTIME;
3107 * For new created object, it must be consistent,
3108 * and it is unnecessary to scrub against it.
3110 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3112 obj->oo_inode = inode;
3115 if (obj->oo_hl_head != NULL) {
3116 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3117 obj->oo_hl_head = NULL;
3119 result = PTR_ERR(inode);
3121 LINVRNT(osd_invariant(obj));
3129 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3130 struct lu_attr *attr,
3131 struct dt_allocation_hint *hint,
3132 struct dt_object_format *dof,
3136 struct osd_thandle *oth;
3137 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3139 LASSERT(S_ISDIR(attr->la_mode));
3141 oth = container_of(th, struct osd_thandle, ot_super);
3142 LASSERT(oth->ot_handle->h_transaction != NULL);
3143 result = osd_mkfile(info, obj, mode, hint, th, attr);
3148 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3149 struct lu_attr *attr,
3150 struct dt_allocation_hint *hint,
3151 struct dt_object_format *dof,
3155 struct osd_thandle *oth;
3156 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3158 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3160 LASSERT(S_ISREG(attr->la_mode));
3162 oth = container_of(th, struct osd_thandle, ot_super);
3163 LASSERT(oth->ot_handle->h_transaction != NULL);
3165 result = osd_mkfile(info, obj, mode, hint, th, attr);
3167 LASSERT(obj->oo_inode != NULL);
3168 if (feat->dif_flags & DT_IND_VARKEY)
3169 result = iam_lvar_create(obj->oo_inode,
3170 feat->dif_keysize_max,
3172 feat->dif_recsize_max,
3175 result = iam_lfix_create(obj->oo_inode,
3176 feat->dif_keysize_max,
3178 feat->dif_recsize_max,
3184 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3185 struct lu_attr *attr,
3186 struct dt_allocation_hint *hint,
3187 struct dt_object_format *dof,
3190 LASSERT(S_ISREG(attr->la_mode));
3191 return osd_mkfile(info, obj, (attr->la_mode &
3192 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3196 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3197 struct lu_attr *attr,
3198 struct dt_allocation_hint *hint,
3199 struct dt_object_format *dof,
3202 LASSERT(S_ISLNK(attr->la_mode));
3203 return osd_mkfile(info, obj, (attr->la_mode &
3204 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3208 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3209 struct lu_attr *attr,
3210 struct dt_allocation_hint *hint,
3211 struct dt_object_format *dof,
3214 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3217 LINVRNT(osd_invariant(obj));
3218 LASSERT(obj->oo_inode == NULL);
3219 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3220 S_ISFIFO(mode) || S_ISSOCK(mode));
3222 result = osd_mkfile(info, obj, mode, hint, th, attr);
3224 LASSERT(obj->oo_inode != NULL);
3226 * This inode should be marked dirty for i_rdev. Currently
3227 * that is done in the osd_attr_init().
3229 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3232 LINVRNT(osd_invariant(obj));
3236 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3238 struct dt_allocation_hint *hint,
3239 struct dt_object_format *dof,
3242 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3244 osd_obj_type_f result;
3260 result = osd_mk_index;
3271 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3272 struct dt_object *parent, struct dt_object *child,
3277 ah->dah_parent = parent;
3278 ah->dah_mode = child_mode;
3280 if (parent != NULL && !dt_object_remote(parent)) {
3281 /* will help to find FID->ino at dt_insert("..") */
3282 struct osd_object *pobj = osd_dt_obj(parent);
3284 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3288 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3289 struct lu_attr *attr, struct dt_object_format *dof)
3291 struct inode *inode = obj->oo_inode;
3292 __u64 valid = attr->la_valid;
3295 attr->la_valid &= ~(LA_TYPE | LA_MODE);
3297 if (dof->dof_type != DFT_NODE)
3298 attr->la_valid &= ~LA_RDEV;
3299 if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3300 attr->la_valid &= ~LA_ATIME;
3301 if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3302 attr->la_valid &= ~LA_CTIME;
3303 if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3304 attr->la_valid &= ~LA_MTIME;
3306 result = osd_quota_transfer(inode, attr);
3310 if (attr->la_valid != 0) {
3311 result = osd_inode_setattr(info->oti_env, inode, attr);
3313 * The osd_inode_setattr() should always succeed here. The
3314 * only error that could be returned is EDQUOT when we are
3315 * trying to change the UID or GID of the inode. However, this
3316 * should not happen since quota enforcement is no longer
3317 * enabled on ldiskfs (lquota takes care of it).
3319 LASSERTF(result == 0, "%d\n", result);
3320 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3323 attr->la_valid = valid;
3327 * Helper function for osd_create()
3329 * \retval 0, on success
3331 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3332 struct lu_attr *attr, struct dt_allocation_hint *hint,
3333 struct dt_object_format *dof, struct thandle *th)
3338 osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3340 /* we drop umask so that permissions we pass are not affected */
3341 umask = current->fs->umask;
3342 current->fs->umask = 0;
3344 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3346 if (likely(obj->oo_inode != NULL)) {
3347 LASSERT(obj->oo_inode->i_state & I_NEW);
3350 * Unlock the inode before attr initialization to avoid
3351 * unnecessary dqget operations. LU-6378
3353 unlock_new_inode(obj->oo_inode);
3356 if (likely(result == 0)) {
3357 osd_attr_init(info, obj, attr, dof);
3358 osd_object_init0(obj);
3361 /* restore previous umask value */
3362 current->fs->umask = umask;
3364 osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3370 * Helper function for osd_create()
3372 * \retval 0, on success
3374 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3375 const struct lu_fid *fid, struct thandle *th)
3377 struct osd_thread_info *info = osd_oti_get(env);
3378 struct osd_inode_id *id = &info->oti_id;
3379 struct osd_device *osd = osd_obj2dev(obj);
3380 struct osd_thandle *oh;
3383 LASSERT(obj->oo_inode != NULL);
3385 oh = container_of0(th, struct osd_thandle, ot_super);
3386 LASSERT(oh->ot_handle);
3387 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3389 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3390 rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3391 OI_CHECK_FLD, NULL);
3392 if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP) && osd->od_is_ost) {
3393 struct lu_fid next_fid = *fid;
3395 /* insert next object in advance, and map to the same inode */
3397 if (next_fid.f_oid != 0) {
3398 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3399 osd_oi_insert(info, osd, &next_fid, id, oh->ot_handle,
3400 OI_CHECK_FLD, NULL);
3401 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3405 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3410 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3411 u64 seq, struct lu_seq_range *range)
3413 struct seq_server_site *ss = osd_seq_site(osd);
3415 if (fid_seq_is_idif(seq)) {
3416 fld_range_set_ost(range);
3417 range->lsr_index = idif_ost_idx(seq);
3421 if (!fid_seq_in_fldb(seq)) {
3422 fld_range_set_mdt(range);
3425 * FIXME: If ss is NULL, it suppose not get lsr_index
3428 range->lsr_index = ss->ss_node_id;
3432 LASSERT(ss != NULL);
3433 fld_range_set_any(range);
3434 /* OSD will only do local fld lookup */
3435 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3438 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3439 struct lu_attr *attr,
3440 struct dt_allocation_hint *hint,
3441 struct dt_object_format *dof,
3442 struct thandle *handle)
3444 struct osd_thandle *oh;
3449 LASSERT(handle != NULL);
3451 oh = container_of0(handle, struct osd_thandle, ot_super);
3452 LASSERT(oh->ot_handle == NULL);
3455 * EA object consumes more credits than regular object: osd_mk_index
3456 * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3457 * leaf_node, could involves the block, block bitmap, groups, GDT
3458 * change for each block, so add 4 * 2 credits in that case.
3460 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3461 osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3462 (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3464 * Reuse idle OI block may cause additional one OI block
3467 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3468 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3469 if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP))
3470 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3471 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3473 /* will help to find FID->ino mapping at dt_insert() */
3474 rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3482 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3483 attr->la_projid, 1, oh, osd_dt_obj(dt),
3484 NULL, OSD_QID_INODE);
3492 * Called to destroy on-disk representation of the object
3494 * Concurrency: must be locked
3496 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3499 struct osd_object *obj = osd_dt_obj(dt);
3500 struct inode *inode = obj->oo_inode;
3501 struct osd_thandle *oh;
3509 oh = container_of0(th, struct osd_thandle, ot_super);
3510 LASSERT(oh->ot_handle == NULL);
3512 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3513 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3515 /* For removing agent entry */
3516 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3517 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3520 * Recycle idle OI leaf may cause additional three OI blocks
3523 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3524 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3525 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3526 /* one less inode */
3527 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3528 i_projid_read(inode), -1, oh, obj, NULL,
3532 /* data to be truncated */
3533 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3534 i_projid_read(inode), 0, oh, obj, NULL,
3540 * will help to find FID->ino when this object is being
3543 rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3548 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3551 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3552 struct osd_object *obj = osd_dt_obj(dt);
3553 struct inode *inode = obj->oo_inode;
3554 struct osd_device *osd = osd_obj2dev(obj);
3555 struct osd_thandle *oh;
3560 oh = container_of0(th, struct osd_thandle, ot_super);
3561 LASSERT(oh->ot_handle);
3563 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3565 if (unlikely(fid_is_acct(fid)))
3568 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3569 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3571 CERROR("%s: remove agent entry "DFID": rc = %d\n",
3572 osd_name(osd), PFID(fid), result);
3575 if (S_ISDIR(inode->i_mode)) {
3576 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3577 inode->i_nlink == 2);
3579 spin_lock(&obj->oo_guard);
3581 spin_unlock(&obj->oo_guard);
3582 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3585 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3587 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3589 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3590 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3591 oh->ot_handle, OI_CHECK_FLD);
3593 osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3594 /* XXX: add to ext3 orphan list */
3595 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3597 /* not needed in the cache anymore */
3598 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3599 obj->oo_destroyed = 1;
3605 * Put the fid into lustre_mdt_attrs, and then place the structure
3606 * inode's ea. This fid should not be altered during the life time
3609 * \retval +ve, on success
3610 * \retval -ve, on error
3612 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3614 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3615 const struct lu_fid *fid, __u32 compat, __u32 incompat)
3617 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3618 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3623 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3626 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3629 lustre_loa_init(loa, fid, compat, incompat);
3630 lustre_loa_swab(loa, false);
3633 * For the OST device with 256 bytes inode size by default,
3634 * the PFID EA will be stored together with LMA EA to avoid
3635 * performance trouble. Otherwise the PFID EA can be stored
3636 * independently. LU-8998
3638 if ((compat & LMAC_FID_ON_OST) &&
3639 LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3640 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3641 sizeof(*loa), XATTR_CREATE);
3643 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3644 sizeof(*lma), XATTR_CREATE);
3646 * LMA may already exist, but we need to check that all the
3647 * desired compat/incompat flags have been added.
3649 if (unlikely(rc == -EEXIST)) {
3650 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3651 XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3655 if (rc < sizeof(*lma))
3658 lustre_loa_swab(loa, true);
3659 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3660 ((compat == 0 && incompat == 0) ||
3661 (!(~lma->lma_compat & compat) &&
3662 !(~lma->lma_incompat & incompat))))
3665 lma->lma_self_fid = *fid;
3666 lma->lma_compat |= compat;
3667 lma->lma_incompat |= incompat;
3668 if (rc == sizeof(*lma)) {
3669 lustre_lma_swab(lma);
3670 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3671 sizeof(*lma), XATTR_REPLACE);
3673 lustre_loa_swab(loa, false);
3674 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3675 sizeof(*loa), XATTR_REPLACE);
3683 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3684 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3685 * To have compatilibility with 1.8 ldiskfs driver we need to have
3686 * magic number at start of fid data.
3687 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3690 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3691 const struct lu_fid *fid)
3693 if (!fid_is_namespace_visible(fid) ||
3694 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3695 param->edp_magic = 0;
3699 param->edp_magic = LDISKFS_LUFID_MAGIC;
3700 param->edp_len = sizeof(struct lu_fid) + 1;
3701 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3705 * Try to read the fid from inode ea into dt_rec.
3707 * \param fid object fid.
3709 * \retval 0 on success
3711 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3712 __u32 ino, struct lu_fid *fid,
3713 struct osd_inode_id *id)
3715 struct osd_thread_info *info = osd_oti_get(env);
3716 struct inode *inode;
3720 osd_id_gen(id, ino, OSD_OII_NOGEN);
3721 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3723 RETURN(PTR_ERR(inode));
3729 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3731 struct inode *parent_dir,
3732 const struct lu_fid *dot_fid,
3733 const struct lu_fid *dot_dot_fid,
3734 struct osd_thandle *oth)
3736 struct ldiskfs_dentry_param *dot_ldp;
3737 struct ldiskfs_dentry_param *dot_dot_ldp;
3738 __u32 saved_nlink = dir->i_nlink;
3741 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3742 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3744 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3745 dot_ldp->edp_magic = 0;
3747 rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3748 dir, dot_ldp, dot_dot_ldp);
3750 * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3751 * the subseqent ref_add() will increase the dir->i_nlink
3752 * as 3. That is incorrect for new created directory.
3754 * It looks like hack, because we want to make the OSD API
3755 * to be order-independent for new created directory object
3756 * between dt_insert(..) and ref_add() operations.
3758 * Here, we only restore the in-RAM dir-inode's nlink attr,
3759 * becuase if the nlink attr is not 2, then there will be
3760 * ref_add() called following the dt_insert(..), such call
3761 * will make both the in-RAM and on-disk dir-inode's nlink
3762 * attr to be set as 2. LU-7447
3764 set_nlink(dir, saved_nlink);
3769 * Create an local agent inode for remote entry
3771 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3772 struct osd_device *osd,
3773 struct osd_object *pobj,
3774 const struct lu_fid *fid,
3778 struct osd_thread_info *info = osd_oti_get(env);
3779 struct inode *local;
3780 struct osd_thandle *oh;
3786 oh = container_of(th, struct osd_thandle, ot_super);
3787 LASSERT(oh->ot_handle->h_transaction != NULL);
3789 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type,
3791 if (IS_ERR(local)) {
3792 CERROR("%s: create local error %d\n", osd_name(osd),
3793 (int)PTR_ERR(local));
3798 * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3799 * correct gid on remote file, not agent here
3801 local->i_gid = current_fsgid();
3802 ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3804 /* e2fsck doesn't like empty symlinks. Store remote FID as symlink.
3805 * That gives e2fsck something to look at and be happy, and allows
3806 * debugging if we need to determine where this symlink came from.
3808 if (S_ISLNK(type)) {
3809 BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1);
3810 rc = snprintf((char *)LDISKFS_I(local)->i_data,
3811 LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
3813 i_size_write(local, rc);
3814 LDISKFS_I(local)->i_disksize = rc;
3816 unlock_new_inode(local);
3818 /* Agent inode should not have project ID */
3819 #ifdef HAVE_PROJECT_QUOTA
3820 if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3821 i_projid_read(pobj->oo_inode) != 0) {
3822 rc = osd_transfer_project(local, 0);
3824 CERROR("%s: quota transfer failed: rc = %d. Is project "
3825 "quota enforcement enabled on the ldiskfs "
3826 "filesystem?\n", local->i_sb->s_id, rc);
3827 RETURN(ERR_PTR(rc));
3831 /* Set special LMA flag for local agent inode */
3832 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3834 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3835 osd_name(osd), PFID(fid), rc);
3836 RETURN(ERR_PTR(rc));
3842 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3843 lu_object_fid(&pobj->oo_dt.do_lu),
3846 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3847 osd_name(osd), PFID(fid), rc);
3848 RETURN(ERR_PTR(rc));
3855 * when direntry is deleted, we have to take care of possible agent inode
3856 * referenced by that. unfortunately we can't do this at that point:
3857 * iget() within a running transaction leads to deadlock and we better do
3858 * not call that every delete declaration to save performance. so we put
3859 * a potention agent inode on a list and process that once the transaction
3860 * is over. Notice it's not any worse in terms of real orphans as regular
3861 * object destroy doesn't put inodes on the on-disk orphan list. this should
3862 * be addressed separately
3864 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3865 struct osd_thandle *oh,
3868 struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3869 struct osd_obj_orphan *oor;
3876 oor->oor_env = (struct lu_env *)env;
3877 spin_lock(&osd->od_osfs_lock);
3878 list_add(&oor->oor_list, &osd->od_orphan_list);
3879 spin_unlock(&osd->od_osfs_lock);
3881 oh->ot_remove_agents = 1;
3887 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3888 struct osd_device *osd)
3890 struct osd_thread_info *info = osd_oti_get(env);
3891 struct osd_obj_orphan *oor, *tmp;
3892 struct osd_inode_id id;
3894 struct inode *inode;
3899 spin_lock(&osd->od_osfs_lock);
3900 list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3901 if (oor->oor_env == env)
3902 list_move(&oor->oor_list, &list);
3904 spin_unlock(&osd->od_osfs_lock);
3906 list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3910 list_del(&oor->oor_list);
3913 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3914 inode = osd_iget_fid(info, osd, &id, &fid);
3918 if (!osd_remote_fid(env, osd, &fid)) {
3923 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3925 mark_inode_dirty(inode);
3926 ldiskfs_journal_stop(jh);
3934 * OSD layer object create function for OST objects (b=11826).
3936 * The FID is inserted into inode xattr here.
3938 * \retval 0, on success
3939 * \retval -ve, on error
3941 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3942 struct lu_attr *attr, struct dt_allocation_hint *hint,
3943 struct dt_object_format *dof, struct thandle *th)
3945 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3946 struct osd_object *obj = osd_dt_obj(dt);
3947 struct osd_thread_info *info = osd_oti_get(env);
3948 int result, on_ost = 0;
3952 if (dt_object_exists(dt))
3955 LINVRNT(osd_invariant(obj));
3956 LASSERT(!dt_object_remote(dt));
3957 LASSERT(osd_is_write_locked(env, obj));
3958 LASSERT(th != NULL);
3960 if (unlikely(fid_is_acct(fid)))
3962 * Quota files can't be created from the kernel any more,
3963 * 'tune2fs -O quota' will take care of creating them
3967 result = __osd_create(info, obj, attr, hint, dof, th);
3969 if (fid_is_idif(fid) &&
3970 !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3971 struct lu_fid *tfid = &info->oti_fid;
3972 struct ost_id *oi = &info->oti_ostid;
3974 fid_to_ostid(fid, oi);
3975 ostid_to_fid(tfid, oi, 0);
3977 result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3978 LMAC_FID_ON_OST, 0);
3980 on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3982 result = osd_ea_fid_set(info, obj->oo_inode, fid,
3983 on_ost ? LMAC_FID_ON_OST : 0,
3986 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3987 obj->oo_dt.do_body_ops = &osd_body_ops;
3990 if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
3991 result = __osd_oi_insert(env, obj, fid, th);
3994 * a small optimization - dt_insert() isn't usually applied
3995 * to OST objects, so we don't need to cache OI mapping for
3998 if (result == 0 && on_ost == 0) {
3999 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4001 result = osd_idc_find_and_init(env, osd, obj);
4002 LASSERT(result == 0);
4005 LASSERT(ergo(result == 0,
4006 dt_object_exists(dt) && !dt_object_remote(dt)));
4007 LINVRNT(osd_invariant(obj));
4011 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4012 struct thandle *handle)
4014 struct osd_thandle *oh;
4016 /* it's possible that object doesn't exist yet */
4017 LASSERT(handle != NULL);
4019 oh = container_of0(handle, struct osd_thandle, ot_super);
4020 LASSERT(oh->ot_handle == NULL);
4022 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4023 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4025 osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
4031 * Concurrency: @dt is write locked.
4033 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4036 struct osd_object *obj = osd_dt_obj(dt);
4037 struct inode *inode = obj->oo_inode;
4038 struct osd_thandle *oh;
4041 if (!dt_object_exists(dt) || obj->oo_destroyed)
4044 LINVRNT(osd_invariant(obj));
4045 LASSERT(!dt_object_remote(dt));
4046 LASSERT(osd_is_write_locked(env, obj));
4047 LASSERT(th != NULL);
4049 oh = container_of0(th, struct osd_thandle, ot_super);
4050 LASSERT(oh->ot_handle != NULL);
4052 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4054 CDEBUG(D_INODE, DFID" increase nlink %d\n",
4055 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4057 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4058 * (65000) subdirectories by storing "1" in i_nlink if the link count
4059 * would otherwise overflow. Directory tranversal tools understand
4060 * that (st_nlink == 1) indicates that the filesystem dose not track
4061 * hard links count on the directory, and will not abort subdirectory
4062 * scanning early once (st_nlink - 2) subdirs have been found.
4064 * This also has to properly handle the case of inodes with nlink == 0
4065 * in case they are being linked into the PENDING directory
4067 spin_lock(&obj->oo_guard);
4068 if (unlikely(inode->i_nlink == 0))
4069 /* inc_nlink from 0 may cause WARN_ON */
4070 set_nlink(inode, 1);
4072 ldiskfs_inc_count(oh->ot_handle, inode);
4073 if (!S_ISDIR(inode->i_mode))
4074 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4076 spin_unlock(&obj->oo_guard);
4078 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4079 LINVRNT(osd_invariant(obj));
4081 osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4086 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4087 struct thandle *handle)
4089 struct osd_thandle *oh;
4091 if (!dt_object_exists(dt))
4094 LASSERT(!dt_object_remote(dt));
4095 LASSERT(handle != NULL);
4097 oh = container_of0(handle, struct osd_thandle, ot_super);
4098 LASSERT(oh->ot_handle == NULL);
4100 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4101 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4107 * Concurrency: @dt is write locked.
4109 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4112 struct osd_object *obj = osd_dt_obj(dt);
4113 struct inode *inode = obj->oo_inode;
4114 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4115 struct osd_thandle *oh;
4117 if (!dt_object_exists(dt))
4120 LINVRNT(osd_invariant(obj));
4121 LASSERT(!dt_object_remote(dt));
4122 LASSERT(osd_is_write_locked(env, obj));
4123 LASSERT(th != NULL);
4125 oh = container_of0(th, struct osd_thandle, ot_super);
4126 LASSERT(oh->ot_handle != NULL);
4128 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4130 spin_lock(&obj->oo_guard);
4132 * That can be result of upgrade from old Lustre version and
4133 * applied only to local files. Just skip this ref_del call.
4134 * ext4_unlink() only treats this as a warning, don't LASSERT here.
4136 if (inode->i_nlink == 0) {
4137 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4138 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4139 ", maybe an upgraded file? (LU-3915)\n",
4140 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4141 spin_unlock(&obj->oo_guard);
4145 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4146 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4148 ldiskfs_dec_count(oh->ot_handle, inode);
4149 spin_unlock(&obj->oo_guard);
4151 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4152 LINVRNT(osd_invariant(obj));
4154 osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4160 * Concurrency: @dt is read locked.
4162 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4163 struct lu_buf *buf, const char *name)
4165 struct osd_object *obj = osd_dt_obj(dt);
4166 struct inode *inode = obj->oo_inode;
4167 struct osd_thread_info *info = osd_oti_get(env);
4168 struct dentry *dentry = &info->oti_obj_dentry;
4169 bool cache_xattr = false;
4174 /* version get is not real XATTR but uses xattr API */
4175 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4176 dt_obj_version_t *ver = buf->lb_buf;
4179 * for version we are just using xattr API but change inode
4182 if (buf->lb_len == 0)
4183 return sizeof(dt_obj_version_t);
4185 if (buf->lb_len < sizeof(dt_obj_version_t))
4188 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4189 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4191 *ver = LDISKFS_I(inode)->i_fs_version;
4193 return sizeof(dt_obj_version_t);
4196 if (!dt_object_exists(dt))
4199 LASSERT(!dt_object_remote(dt));
4200 LASSERT(inode->i_op != NULL);
4201 #ifdef HAVE_IOP_XATTR
4202 LASSERT(inode->i_op->getxattr != NULL);
4205 if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4206 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4210 rc = osd_oxc_get(obj, name, buf);
4215 if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4216 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4217 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4218 struct filter_fid *ff;
4219 struct ost_layout *ol;
4221 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4225 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4228 if (buf->lb_len == 0 || !buf->lb_buf)
4231 if (buf->lb_len < rc)
4235 ol = &ff->ff_layout;
4236 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4237 PFID_STRIPE_IDX_BITS);
4238 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4239 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4240 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4241 if (lma->lma_compat & LMAC_COMP_INFO) {
4242 ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4243 ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4244 ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4246 ol->ol_comp_start = 0;
4247 ol->ol_comp_end = 0;
4251 rc = __osd_xattr_get(inode, dentry, name,
4252 buf->lb_buf, buf->lb_len);
4256 if (rc == -ENOENT || rc == -ENODATA)
4257 osd_oxc_add(obj, name, NULL, 0);
4258 else if (rc > 0 && buf->lb_buf != NULL)
4259 osd_oxc_add(obj, name, buf->lb_buf, rc);
4265 static int osd_declare_xattr_set(const struct lu_env *env,
4266 struct dt_object *dt,
4267 const struct lu_buf *buf, const char *name,
4268 int fl, struct thandle *handle)
4270 struct osd_thandle *oh;
4272 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4274 LASSERT(handle != NULL);
4276 oh = container_of0(handle, struct osd_thandle, ot_super);
4277 LASSERT(oh->ot_handle == NULL);
4279 if (strcmp(name, XATTR_NAME_LMA) == 0) {
4281 * For non-upgrading case, the LMA is set first and
4282 * usually fit inode. But for upgrade case, the LMA
4283 * may be in another separated EA block.
4285 if (dt_object_exists(dt)) {
4286 if (fl == LU_XATTR_REPLACE)
4291 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4293 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4294 /* We may need to delete the old PFID EA. */
4295 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4296 if (fl == LU_XATTR_REPLACE)
4302 * If some name entry resides on remote MDT, then will create
4303 * agent entry under remote parent. On the other hand, if the
4304 * remote entry will be removed, then related agent entry may
4305 * need to be removed from the remote parent. So there may be
4306 * kinds of cases, let's declare enough credits. The credits
4307 * for create agent entry is enough for remove case.
4309 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4310 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4311 if (dt_object_exists(dt))
4312 credits += 1; /* For updating LMA */
4316 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4321 if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4323 * learn xattr size from osd_xattr_get if
4324 * attribute has not been read yet
4326 buflen = __osd_xattr_get(
4327 osd_dt_obj(dt)->oo_inode,
4328 &osd_oti_get(env)->oti_obj_dentry,
4333 buflen = buf->lb_len;
4336 if (buflen > sb->s_blocksize) {
4337 credits += osd_calc_bkmap_credits(
4339 (buflen + sb->s_blocksize - 1) >>
4340 sb->s_blocksize_bits);
4344 * xattr set may involve inode quota change, reserve credits for
4345 * dquot_initialize()
4347 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4350 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4355 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4356 const struct lu_buf *buf, int fl,
4357 struct thandle *handle)
4359 struct osd_thread_info *info = osd_oti_get(env);
4360 struct dentry *dentry = &info->oti_obj_dentry;
4361 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4362 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4363 struct inode *inode = obj->oo_inode;
4364 struct filter_fid *ff = buf->lb_buf;
4365 struct ost_layout *ol = &ff->ff_layout;
4366 int flags = XATTR_REPLACE;
4371 if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4374 rc = osd_get_lma(info, inode, dentry, loa);
4375 if (rc == -ENODATA) {
4376 /* Usually for upgarding from old device */
4377 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4378 LMAC_FID_ON_OST, 0);
4379 flags = XATTR_CREATE;
4384 if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4385 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4388 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4389 /* Separate PFID EA from LMA */
4390 lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4391 lustre_lma_swab(lma);
4392 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4393 sizeof(*lma), XATTR_REPLACE);
4395 obj->oo_pfid_in_lma = 0;
4396 rc = LU_XATTR_CREATE;
4402 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4406 * Old client does not send stripe information,
4407 * then store the PFID EA on disk separatedly.
4409 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4410 ol->ol_stripe_size == 0))
4413 /* Remove old PFID EA entry firstly. */
4414 dquot_initialize(inode);
4415 rc = ll_vfs_removexattr(dentry, inode, XATTR_NAME_FID);
4416 if (rc == -ENODATA) {
4417 if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4424 fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4425 if (likely(ol->ol_stripe_size != 0)) {
4426 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4427 PFID_STRIPE_IDX_BITS;
4428 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4429 lma->lma_compat |= LMAC_STRIPE_INFO;
4430 if (ol->ol_comp_id != 0) {
4431 loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4432 loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4433 loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4434 lma->lma_compat |= LMAC_COMP_INFO;
4438 lustre_loa_swab(loa, false);
4440 /* Store the PFID EA inside LMA. */
4441 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4444 obj->oo_pfid_in_lma = 1;
4450 * In DNE environment, the object (in spite of regular file or directory)
4451 * and its name entry may reside on different MDTs. Under such case, we will
4452 * create an agent entry on the MDT where the object resides. The agent entry
4453 * references the object locally, that makes the object to be visible to the
4454 * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4455 * such as 'tar' can handle the object properly.
4457 * We handle the agent entry during set linkEA that is the common interface
4458 * for both regular file and directroy, can handle kinds of cases, such as
4459 * create/link/unlink/rename, and so on.
4461 * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4463 * XXX: There are two known issues:
4464 * 1. For one object, we will create at most one agent entry even if there
4465 * may be more than one cross-MDTs hard links on the object. So the local
4466 * e2fsck may claim that the object's nlink is larger than the name entries
4467 * that reference such inode. And in further, the e2fsck will fix the nlink
4468 * attribute to match the local references. Then it will cause the object's
4469 * nlink attribute to be inconsistent with the global references. it is bad
4470 * but not fatal. The ref_del() can handle the zero-referenced case. On the
4471 * other hand, the global namespace LFSCK can repair the object's attribute
4472 * according to the linkEA.
4473 * 2. There may be too many hard links on the object as to its linkEA overflow,
4474 * then the linkEA entry for cross-MDTs reference may be discarded. If such
4475 * case happened, then at this point, we do not know whether there are some
4476 * cross-MDTs reference. But there are local references, it guarantees that
4477 * object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4479 static int osd_xattr_handle_linkea(const struct lu_env *env,
4480 struct osd_device *osd,
4481 struct osd_object *obj,
4482 const struct lu_buf *buf,
4483 struct thandle *handle)
4485 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4486 struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4487 struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4488 struct lu_name tmpname;
4489 struct osd_thandle *oh;
4491 bool remote = false;
4495 oh = container_of0(handle, struct osd_thandle, ot_super);
4496 LASSERT(oh->ot_handle != NULL);
4498 rc = linkea_init_with_rec(&ldata);
4500 linkea_first_entry(&ldata);
4501 while (ldata.ld_lee != NULL && !remote) {
4502 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4504 if (osd_remote_fid(env, osd, tfid) > 0)
4507 linkea_next_entry(&ldata);
4509 } else if (rc == -ENODATA) {
4515 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4516 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4518 CERROR("%s: failed to remove agent entry for "DFID
4519 ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4520 } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4521 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4523 CERROR("%s: failed to create agent entry for "DFID
4524 ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4531 * Concurrency: @dt is write locked.
4533 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4534 const struct lu_buf *buf, const char *name, int fl,
4535 struct thandle *handle)
4537 struct osd_object *obj = osd_dt_obj(dt);
4538 struct osd_device *osd = osd_obj2dev(obj);
4539 struct inode *inode = obj->oo_inode;
4540 struct osd_thread_info *info = osd_oti_get(env);
4550 /* version set is not real XATTR */
4551 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4552 dt_obj_version_t *version = buf->lb_buf;
4555 * for version we are just using xattr API but change inode
4558 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4560 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4561 *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4563 LDISKFS_I(inode)->i_fs_version = *version;
4565 * Version is set after all inode operations are finished,
4566 * so we should mark it dirty here
4568 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4573 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4574 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4577 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4580 * For the OST device with 256 bytes inode size by default,
4581 * the PFID EA will be stored together with LMA EA to avoid
4582 * performance trouble. Otherwise the PFID EA can be stored
4583 * independently. LU-8998
4585 if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4586 (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4587 LASSERT(buf->lb_buf);
4589 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4592 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4593 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4594 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4596 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4600 lma->lma_incompat |= LMAI_STRIPED;
4601 lustre_lma_swab(lma);
4602 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4603 sizeof(*lma), XATTR_REPLACE);
4606 } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4607 LASSERT(!osd->od_is_ost);
4609 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4614 if (fl & LU_XATTR_REPLACE)
4615 fs_flags |= XATTR_REPLACE;
4617 if (fl & LU_XATTR_CREATE)
4618 fs_flags |= XATTR_CREATE;
4620 rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4621 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4624 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4625 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4626 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4632 * Concurrency: @dt is read locked.
4634 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4635 const struct lu_buf *buf)
4637 struct osd_object *obj = osd_dt_obj(dt);
4638 struct inode *inode = obj->oo_inode;
4639 struct osd_thread_info *info = osd_oti_get(env);
4640 struct dentry *dentry = &info->oti_obj_dentry;
4642 if (!dt_object_exists(dt))
4645 LASSERT(!dt_object_remote(dt));
4646 LASSERT(inode->i_op != NULL);
4647 LASSERT(inode->i_op->listxattr != NULL);
4649 dentry->d_inode = inode;
4650 dentry->d_sb = inode->i_sb;
4651 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4654 static int osd_declare_xattr_del(const struct lu_env *env,
4655 struct dt_object *dt, const char *name,
4656 struct thandle *handle)
4658 struct osd_thandle *oh;
4659 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4661 LASSERT(!dt_object_remote(dt));
4662 LASSERT(handle != NULL);
4664 oh = container_of0(handle, struct osd_thandle, ot_super);
4665 LASSERT(oh->ot_handle == NULL);
4667 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4668 osd_dto_credits_noquota[DTO_XATTR_SET]);
4670 * xattr del may involve inode quota change, reserve credits for
4671 * dquot_initialize()
4673 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4679 * Concurrency: @dt is write locked.
4681 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4682 const char *name, struct thandle *handle)
4684 struct osd_object *obj = osd_dt_obj(dt);
4685 struct inode *inode = obj->oo_inode;
4686 struct osd_thread_info *info = osd_oti_get(env);
4687 struct dentry *dentry = &info->oti_obj_dentry;
4690 if (!dt_object_exists(dt))
4693 LASSERT(!dt_object_remote(dt));
4694 LASSERT(inode->i_op != NULL);
4695 LASSERT(handle != NULL);
4696 #ifdef HAVE_IOP_XATTR
4697 LASSERT(inode->i_op->removexattr != NULL);
4700 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4702 if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4703 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4705 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4706 &info->oti_ost_attrs);
4708 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4710 lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4711 lustre_lma_swab(lma);
4712 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4713 sizeof(*lma), XATTR_REPLACE);
4715 obj->oo_pfid_in_lma = 0;
4718 dquot_initialize(inode);
4719 dentry->d_inode = inode;
4720 dentry->d_sb = inode->i_sb;
4721 rc = ll_vfs_removexattr(dentry, inode, name);
4724 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4727 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4728 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4729 osd_oxc_del(obj, name);
4734 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4735 __u64 start, __u64 end)
4737 struct osd_object *obj = osd_dt_obj(dt);
4738 struct inode *inode = obj->oo_inode;
4739 struct osd_thread_info *info = osd_oti_get(env);
4740 struct dentry *dentry = &info->oti_obj_dentry;
4741 struct file *file = &info->oti_file;
4746 dentry->d_inode = inode;
4747 dentry->d_sb = inode->i_sb;
4748 file->f_path.dentry = dentry;
4749 file->f_mapping = inode->i_mapping;
4750 file->f_op = inode->i_fop;
4751 file->f_inode = inode;
4753 rc = vfs_fsync_range(file, start, end, 0);
4758 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4767 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4768 const struct dt_index_features *feat)
4770 struct iam_descr *descr;
4772 if (osd_object_is_root(o))
4773 return feat == &dt_directory_features;
4775 LASSERT(o->oo_dir != NULL);
4777 descr = o->oo_dir->od_container.ic_descr;
4778 if (feat == &dt_directory_features) {
4779 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4784 return feat->dif_keysize_min <= descr->id_key_size &&
4785 descr->id_key_size <= feat->dif_keysize_max &&
4786 feat->dif_recsize_min <= descr->id_rec_size &&
4787 descr->id_rec_size <= feat->dif_recsize_max &&
4788 !(feat->dif_flags & (DT_IND_VARKEY |
4789 DT_IND_VARREC | DT_IND_NONUNQ)) &&
4790 ergo(feat->dif_flags & DT_IND_UPDATE,
4791 1 /* XXX check that object (and fs) is writable */);
4795 static int osd_iam_container_init(const struct lu_env *env,
4796 struct osd_object *obj,
4797 struct osd_directory *dir)
4799 struct iam_container *bag = &dir->od_container;
4802 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4806 result = iam_container_setup(bag);
4808 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4810 iam_container_fini(bag);
4817 * Concurrency: no external locking is necessary.
4819 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4820 const struct dt_index_features *feat)
4824 struct osd_object *obj = osd_dt_obj(dt);
4826 LINVRNT(osd_invariant(obj));
4828 if (osd_object_is_root(obj)) {
4829 dt->do_index_ops = &osd_index_ea_ops;
4831 } else if (feat == &dt_directory_features) {
4832 dt->do_index_ops = &osd_index_ea_ops;
4833 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4838 } else if (unlikely(feat == &dt_otable_features)) {
4839 dt->do_index_ops = &osd_otable_ops;
4841 } else if (unlikely(feat == &dt_acct_features)) {
4842 dt->do_index_ops = &osd_acct_index_ops;
4845 } else if (!osd_has_index(obj)) {
4846 struct osd_directory *dir;
4847 struct osd_device *osd = osd_obj2dev(obj);
4848 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4853 spin_lock(&obj->oo_guard);
4854 if (obj->oo_dir == NULL)
4858 * Concurrent thread allocated container data.
4861 spin_unlock(&obj->oo_guard);
4863 * Now, that we have container data, serialize its
4866 down_write(&obj->oo_ext_idx_sem);
4868 * recheck under lock.
4871 if (osd_has_index(obj)) {
4876 result = osd_iam_container_init(env, obj, obj->oo_dir);
4877 if (result || feat == &dt_lfsck_namespace_features ||
4878 feat == &dt_lfsck_layout_orphan_features ||
4879 feat == &dt_lfsck_layout_dangling_features)
4882 result = osd_index_register(osd, fid,
4883 feat->dif_keysize_max,
4884 feat->dif_recsize_max);
4886 CWARN("%s: failed to register index "
4888 osd_name(osd), PFID(fid), result);
4889 else if (result > 0)
4892 CDEBUG(D_LFSCK, "%s: index object "DFID
4893 " (%d/%d) registered\n",
4894 osd_name(osd), PFID(fid),
4895 (int)feat->dif_keysize_max,
4896 (int)feat->dif_recsize_max);
4899 up_write(&obj->oo_ext_idx_sem);
4907 if (result == 0 && skip_iam == 0) {
4908 if (!osd_iam_index_probe(env, obj, feat))
4911 LINVRNT(osd_invariant(obj));
4916 static int osd_otable_it_attr_get(const struct lu_env *env,
4917 struct dt_object *dt,
4918 struct lu_attr *attr)
4924 static const struct dt_object_operations osd_obj_ops = {
4925 .do_read_lock = osd_read_lock,
4926 .do_write_lock = osd_write_lock,
4927 .do_read_unlock = osd_read_unlock,
4928 .do_write_unlock = osd_write_unlock,
4929 .do_write_locked = osd_write_locked,
4930 .do_attr_get = osd_attr_get,
4931 .do_declare_attr_set = osd_declare_attr_set,
4932 .do_attr_set = osd_attr_set,
4933 .do_ah_init = osd_ah_init,
4934 .do_declare_create = osd_declare_create,
4935 .do_create = osd_create,
4936 .do_declare_destroy = osd_declare_destroy,
4937 .do_destroy = osd_destroy,
4938 .do_index_try = osd_index_try,
4939 .do_declare_ref_add = osd_declare_ref_add,
4940 .do_ref_add = osd_ref_add,
4941 .do_declare_ref_del = osd_declare_ref_del,
4942 .do_ref_del = osd_ref_del,
4943 .do_xattr_get = osd_xattr_get,
4944 .do_declare_xattr_set = osd_declare_xattr_set,
4945 .do_xattr_set = osd_xattr_set,
4946 .do_declare_xattr_del = osd_declare_xattr_del,
4947 .do_xattr_del = osd_xattr_del,
4948 .do_xattr_list = osd_xattr_list,
4949 .do_object_sync = osd_object_sync,
4950 .do_invalidate = osd_invalidate,
4953 static const struct dt_object_operations osd_obj_otable_it_ops = {
4954 .do_attr_get = osd_otable_it_attr_get,
4955 .do_index_try = osd_index_try,
4958 static int osd_index_declare_iam_delete(const struct lu_env *env,
4959 struct dt_object *dt,
4960 const struct dt_key *key,
4961 struct thandle *handle)
4963 struct osd_thandle *oh;
4965 oh = container_of0(handle, struct osd_thandle, ot_super);
4966 LASSERT(oh->ot_handle == NULL);
4968 /* Recycle may cause additional three blocks to be changed. */
4969 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4970 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4976 * delete a (key, value) pair from index \a dt specified by \a key
4978 * \param dt osd index object
4979 * \param key key for index
4980 * \param rec record reference
4981 * \param handle transaction handler
4984 * \retval -ve failure
4986 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4987 const struct dt_key *key,
4988 struct thandle *handle)
4990 struct osd_thread_info *oti = osd_oti_get(env);
4991 struct osd_object *obj = osd_dt_obj(dt);
4992 struct osd_thandle *oh;
4993 struct iam_path_descr *ipd;
4994 struct iam_container *bag = &obj->oo_dir->od_container;
4999 if (!dt_object_exists(dt))
5002 LINVRNT(osd_invariant(obj));
5003 LASSERT(!dt_object_remote(dt));
5004 LASSERT(bag->ic_object == obj->oo_inode);
5005 LASSERT(handle != NULL);
5007 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5009 ipd = osd_idx_ipd_get(env, bag);
5010 if (unlikely(ipd == NULL))
5013 oh = container_of0(handle, struct osd_thandle, ot_super);
5014 LASSERT(oh->ot_handle != NULL);
5015 LASSERT(oh->ot_handle->h_transaction != NULL);
5017 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5018 /* swab quota uid/gid provided by caller */
5019 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5020 key = (const struct dt_key *)&oti->oti_quota_id;
5023 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5024 osd_ipd_put(env, bag, ipd);
5025 LINVRNT(osd_invariant(obj));
5026 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5030 static int osd_index_declare_ea_delete(const struct lu_env *env,
5031 struct dt_object *dt,
5032 const struct dt_key *key,
5033 struct thandle *handle)
5035 struct osd_thandle *oh;
5036 struct inode *inode;
5041 LASSERT(!dt_object_remote(dt));
5042 LASSERT(handle != NULL);
5044 oh = container_of0(handle, struct osd_thandle, ot_super);
5045 LASSERT(oh->ot_handle == NULL);
5047 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5048 osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5050 inode = osd_dt_obj(dt)->oo_inode;
5054 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5055 i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5060 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5063 struct osd_fid_pack *rec;
5066 if (de->file_type & LDISKFS_DIRENT_LUFID) {
5067 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5068 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5069 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5075 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5076 const struct lu_fid *fid)
5078 struct seq_server_site *ss = osd_seq_site(osd);
5082 /* FID seqs not in FLDB, must be local seq */
5083 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5087 * If FLD is not being initialized yet, it only happens during the
5088 * initialization, likely during mgs initialization, and we assume
5089 * this is local FID.
5091 if (ss == NULL || ss->ss_server_fld == NULL)
5094 /* Only check the local FLDB here */
5095 if (osd_seq_exists(env, osd, fid_seq(fid)))
5101 static void osd_take_care_of_agent(const struct lu_env *env,
5102 struct osd_device *osd,
5103 struct osd_thandle *oh,
5104 struct ldiskfs_dir_entry_2 *de)
5106 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5107 struct osd_idmap_cache *idc;
5108 int rc, schedule = 0;
5110 LASSERT(de != NULL);
5112 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5113 if (likely(rc == 0)) {
5114 idc = osd_idc_find_or_init(env, osd, fid);
5115 if (IS_ERR(idc) || idc->oic_remote)
5117 } else if (rc == -ENODATA) {
5119 * can't get FID, postpone to the end of the
5120 * transaction when iget() is safe
5124 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5127 osd_schedule_agent_inode_removal(env, oh,
5128 le32_to_cpu(de->inode));
5132 * Index delete function for interoperability mode (b11826).
5133 * It will remove the directory entry added by osd_index_ea_insert().
5134 * This entry is needed to maintain name->fid mapping.
5136 * \param key, key i.e. file entry to be deleted
5138 * \retval 0, on success
5139 * \retval -ve, on error
5141 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5142 const struct dt_key *key, struct thandle *handle)
5144 struct osd_object *obj = osd_dt_obj(dt);
5145 struct inode *dir = obj->oo_inode;
5146 struct dentry *dentry;
5147 struct osd_thandle *oh;
5148 struct ldiskfs_dir_entry_2 *de = NULL;
5149 struct buffer_head *bh;
5150 struct htree_lock *hlock = NULL;
5151 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5156 if (!dt_object_exists(dt))
5159 LINVRNT(osd_invariant(obj));
5160 LASSERT(!dt_object_remote(dt));
5161 LASSERT(handle != NULL);
5163 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5165 oh = container_of(handle, struct osd_thandle, ot_super);
5166 LASSERT(oh->ot_handle != NULL);
5167 LASSERT(oh->ot_handle->h_transaction != NULL);
5169 dquot_initialize(dir);
5170 dentry = osd_child_dentry_get(env, obj,
5171 (char *)key, strlen((char *)key));
5173 if (obj->oo_hl_head != NULL) {
5174 hlock = osd_oti_get(env)->oti_hlock;
5175 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5176 dir, LDISKFS_HLOCK_DEL);
5178 down_write(&obj->oo_ext_idx_sem);
5181 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5184 * If this is not the ".." entry, it might be a remote DNE
5185 * entry and we need to check if the FID is for a remote
5186 * MDT. If the FID is not in the directory entry (e.g.
5187 * upgraded 1.8 filesystem without dirdata enabled) then
5188 * we need to get the FID from the LMA. For a remote directory
5189 * there HAS to be an LMA, it cannot be an IGIF inode in this
5192 * Delete the entry before the agent inode in order to
5193 * simplify error handling. At worst an error after deleting
5194 * the entry first might leak the agent inode afterward. The
5195 * reverse would need filesystem abort in case of error deleting
5196 * the entry after the agent had been removed, or leave a
5197 * dangling entry pointing at a random inode.
5199 if (strcmp((char *)key, dotdot) != 0)
5200 osd_take_care_of_agent(env, osd, oh, de);
5201 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5207 ldiskfs_htree_unlock(hlock);
5209 up_write(&obj->oo_ext_idx_sem);
5213 LASSERT(osd_invariant(obj));
5214 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5219 * Lookup index for \a key and copy record to \a rec.
5221 * \param dt osd index object
5222 * \param key key for index
5223 * \param rec record reference
5225 * \retval +ve success : exact mach
5226 * \retval 0 return record with key not greater than \a key
5227 * \retval -ve failure
5229 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5230 struct dt_rec *rec, const struct dt_key *key)
5232 struct osd_object *obj = osd_dt_obj(dt);
5233 struct iam_path_descr *ipd;
5234 struct iam_container *bag = &obj->oo_dir->od_container;
5235 struct osd_thread_info *oti = osd_oti_get(env);
5236 struct iam_iterator *it = &oti->oti_idx_it;
5237 struct iam_rec *iam_rec;
5242 if (!dt_object_exists(dt))
5245 LASSERT(osd_invariant(obj));
5246 LASSERT(!dt_object_remote(dt));
5247 LASSERT(bag->ic_object == obj->oo_inode);
5249 ipd = osd_idx_ipd_get(env, bag);
5253 /* got ipd now we can start iterator. */
5254 iam_it_init(it, bag, 0, ipd);
5256 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5257 /* swab quota uid/gid provided by caller */
5258 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5259 key = (const struct dt_key *)&oti->oti_quota_id;
5262 rc = iam_it_get(it, (struct iam_key *)key);
5264 if (S_ISDIR(obj->oo_inode->i_mode))
5265 iam_rec = (struct iam_rec *)oti->oti_ldp;
5267 iam_rec = (struct iam_rec *)rec;
5269 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5271 if (S_ISDIR(obj->oo_inode->i_mode))
5272 osd_fid_unpack((struct lu_fid *)rec,
5273 (struct osd_fid_pack *)iam_rec);
5274 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5275 osd_quota_unpack(obj, rec);
5280 osd_ipd_put(env, bag, ipd);
5282 LINVRNT(osd_invariant(obj));
5287 static int osd_index_declare_iam_insert(const struct lu_env *env,
5288 struct dt_object *dt,
5289 const struct dt_rec *rec,
5290 const struct dt_key *key,
5291 struct thandle *handle)
5293 struct osd_thandle *oh;
5295 LASSERT(handle != NULL);
5297 oh = container_of0(handle, struct osd_thandle, ot_super);
5298 LASSERT(oh->ot_handle == NULL);
5300 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5301 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5307 * Inserts (key, value) pair in \a dt index object.
5309 * \param dt osd index object
5310 * \param key key for index
5311 * \param rec record reference
5312 * \param th transaction handler
5315 * \retval -ve failure
5317 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5318 const struct dt_rec *rec,
5319 const struct dt_key *key, struct thandle *th)
5321 struct osd_object *obj = osd_dt_obj(dt);
5322 struct iam_path_descr *ipd;
5323 struct osd_thandle *oh;
5324 struct iam_container *bag;
5325 struct osd_thread_info *oti = osd_oti_get(env);
5326 struct iam_rec *iam_rec;
5331 if (!dt_object_exists(dt))
5334 LINVRNT(osd_invariant(obj));
5335 LASSERT(!dt_object_remote(dt));
5337 bag = &obj->oo_dir->od_container;
5338 LASSERT(bag->ic_object == obj->oo_inode);
5339 LASSERT(th != NULL);
5341 osd_trans_exec_op(env, th, OSD_OT_INSERT);
5343 ipd = osd_idx_ipd_get(env, bag);
5344 if (unlikely(ipd == NULL))
5347 oh = container_of0(th, struct osd_thandle, ot_super);
5348 LASSERT(oh->ot_handle != NULL);
5349 LASSERT(oh->ot_handle->h_transaction != NULL);
5350 if (S_ISDIR(obj->oo_inode->i_mode)) {
5351 iam_rec = (struct iam_rec *)oti->oti_ldp;
5352 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5354 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5355 /* pack quota uid/gid */
5356 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5357 key = (const struct dt_key *)&oti->oti_quota_id;
5358 /* pack quota record */
5359 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5360 iam_rec = (struct iam_rec *)rec;
5362 iam_rec = (struct iam_rec *)rec;
5365 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5367 osd_ipd_put(env, bag, ipd);
5368 LINVRNT(osd_invariant(obj));
5369 osd_trans_exec_check(env, th, OSD_OT_INSERT);
5374 * Calls ldiskfs_add_entry() to add directory entry
5375 * into the directory. This is required for
5376 * interoperability mode (b11826)
5378 * \retval 0, on success
5379 * \retval -ve, on error
5381 static int __osd_ea_add_rec(struct osd_thread_info *info,
5382 struct osd_object *pobj, struct inode *cinode,
5383 const char *name, const struct lu_fid *fid,
5384 struct htree_lock *hlock, struct thandle *th)
5386 struct ldiskfs_dentry_param *ldp;
5387 struct dentry *child;
5388 struct osd_thandle *oth;
5391 oth = container_of(th, struct osd_thandle, ot_super);
5392 LASSERT(oth->ot_handle != NULL);
5393 LASSERT(oth->ot_handle->h_transaction != NULL);
5394 LASSERT(pobj->oo_inode);
5396 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5397 if (unlikely(pobj->oo_inode ==
5398 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5401 osd_get_ldiskfs_dirent_param(ldp, fid);
5402 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5403 child->d_fsdata = (void *)ldp;
5404 dquot_initialize(pobj->oo_inode);
5405 rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5406 child, cinode, hlock);
5407 if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5408 struct ldiskfs_dir_entry_2 *de;
5409 struct buffer_head *bh;
5412 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5415 rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5418 if (S_ISDIR(cinode->i_mode))
5419 de->file_type = LDISKFS_DIRENT_LUFID |
5420 LDISKFS_FT_REG_FILE;
5422 de->file_type = LDISKFS_DIRENT_LUFID |
5424 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5435 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5436 * into the directory.Also sets flags into osd object to
5437 * indicate dot and dotdot are created. This is required for
5438 * interoperability mode (b11826)
5440 * \param dir directory for dot and dotdot fixup.
5441 * \param obj child object for linking
5443 * \retval 0, on success
5444 * \retval -ve, on error
5446 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5447 struct osd_object *dir,
5448 struct inode *parent_dir, const char *name,
5449 const struct lu_fid *dot_fid,
5450 const struct lu_fid *dot_dot_fid,
5453 struct inode *inode = dir->oo_inode;
5454 struct osd_thandle *oth;
5457 oth = container_of(th, struct osd_thandle, ot_super);
5458 LASSERT(oth->ot_handle->h_transaction != NULL);
5459 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5461 if (strcmp(name, dot) == 0) {
5462 if (dir->oo_compat_dot_created) {
5465 LASSERT(inode->i_ino == parent_dir->i_ino);
5466 dir->oo_compat_dot_created = 1;
5469 } else if (strcmp(name, dotdot) == 0) {
5470 if (!dir->oo_compat_dot_created)
5472 /* in case of rename, dotdot is already created */
5473 if (dir->oo_compat_dotdot_created) {
5474 return __osd_ea_add_rec(info, dir, parent_dir, name,
5475 dot_dot_fid, NULL, th);
5478 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5479 struct lu_fid tfid = *dot_dot_fid;
5482 result = osd_add_dot_dotdot_internal(info,
5483 dir->oo_inode, parent_dir, dot_fid,
5486 result = osd_add_dot_dotdot_internal(info,
5487 dir->oo_inode, parent_dir, dot_fid,
5492 dir->oo_compat_dotdot_created = 1;
5500 * It will call the appropriate osd_add* function and return the
5501 * value, return by respective functions.
5503 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5504 struct inode *cinode, const char *name,
5505 const struct lu_fid *fid, struct thandle *th)
5507 struct osd_thread_info *info = osd_oti_get(env);
5508 struct htree_lock *hlock;
5511 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5513 if (name[0] == '.' && (name[1] == '\0' ||
5514 (name[1] == '.' && name[2] == '\0'))) {
5515 if (hlock != NULL) {
5516 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5519 down_write(&pobj->oo_ext_idx_sem);
5522 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5523 lu_object_fid(&pobj->oo_dt.do_lu),
5526 if (hlock != NULL) {
5527 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5528 pobj->oo_inode, LDISKFS_HLOCK_ADD);
5530 down_write(&pobj->oo_ext_idx_sem);
5533 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5534 struct lu_fid *tfid = &info->oti_fid;
5538 rc = __osd_ea_add_rec(info, pobj, cinode, name,
5541 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5546 ldiskfs_htree_unlock(hlock);
5548 up_write(&pobj->oo_ext_idx_sem);
5554 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5555 struct osd_idmap_cache *oic)
5557 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5558 struct lu_fid *fid = &oic->oic_fid;
5559 struct osd_inode_id *id = &oic->oic_lid;
5560 struct inode *inode = NULL;
5567 if (!fid_is_norm(fid) && !fid_is_igif(fid))
5570 if (thread_is_running(&scrub->os_thread) &&
5571 scrub->os_pos_current > id->oii_ino)
5574 if (dev->od_auto_scrub_interval == AS_NEVER ||
5575 ktime_get_real_seconds() <
5576 scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5580 rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5581 if (rc == -ENOENT) {
5582 __u32 gen = id->oii_gen;
5588 inode = osd_iget(oti, dev, id);
5589 /* The inode has been removed (by race maybe). */
5590 if (IS_ERR(inode)) {
5591 rc = PTR_ERR(inode);
5593 RETURN(rc == -ESTALE ? -ENOENT : rc);
5596 /* The OI mapping is lost. */
5597 if (gen != OSD_OII_NOGEN)
5601 * The inode may has been reused by others, we do not know,
5602 * leave it to be handled by subsequent osd_fid_lookup().
5605 } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5612 if (thread_is_running(&scrub->os_thread)) {
5613 if (inode == NULL) {
5614 inode = osd_iget(oti, dev, id);
5615 /* The inode has been removed (by race maybe). */
5616 if (IS_ERR(inode)) {
5617 rc = PTR_ERR(inode);
5619 RETURN(rc == -ESTALE ? -ENOENT : rc);
5623 rc = osd_oii_insert(dev, oic, insert);
5625 * There is race condition between osd_oi_lookup and OI scrub.
5626 * The OI scrub finished just after osd_oi_lookup() failure.
5627 * Under such case, it is unnecessary to trigger OI scrub again,
5628 * but try to call osd_oi_lookup() again.
5630 if (unlikely(rc == -EAGAIN))
5633 if (!S_ISDIR(inode->i_mode))
5636 rc = osd_check_lmv(oti, dev, inode, oic);
5641 if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5642 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5643 SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5644 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5645 "%s: trigger partial OI scrub for RPC inconsistency "
5646 "checking FID "DFID": rc = %d\n",
5647 osd_dev2name(dev), PFID(fid), rc);
5648 if (rc == 0 || rc == -EALREADY)
5661 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5662 struct osd_device *dev,
5663 struct osd_idmap_cache *oic,
5664 struct lu_fid *fid, __u32 ino)
5666 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5667 struct inode *inode;
5670 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5671 inode = osd_iget(oti, dev, &oic->oic_lid);
5672 if (IS_ERR(inode)) {
5673 fid_zero(&oic->oic_fid);
5674 return PTR_ERR(inode);
5677 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5680 fid_zero(&oic->oic_fid);
5682 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5686 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5687 struct osd_inode_id *id, const struct lu_fid *fid)
5689 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5690 id->oii_ino, id->oii_gen, info);
5691 info->oti_cache.oic_lid = *id;
5692 info->oti_cache.oic_fid = *fid;
5693 info->oti_cache.oic_dev = osd;
5697 * Get parent FID from the linkEA.
5699 * For a directory which parent resides on remote MDT, to satisfy the
5700 * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5701 * the other hand, to make the lookup(..) on the directory can return
5702 * the real parent FID, we append the real parent FID after its ".."
5703 * name entry in the /REMOTE_PARENT_DIR.
5705 * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5706 * backup. So after the restore, we cannot get the right parent FID from
5707 * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5708 * we have stored the real parent FID in the directory object's linkEA,
5709 * we can parse the linkEA for the real parent FID.
5711 * \param[in] env pointer to the thread context
5712 * \param[in] obj pointer to the object to be handled
5713 * \param[out]fid pointer to the buffer to hold the parent FID
5715 * \retval 0 for getting the real parent FID successfully
5716 * \retval negative error number on failure
5718 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5719 struct osd_object *obj,
5722 struct osd_thread_info *oti = osd_oti_get(env);
5723 struct lu_buf *buf = &oti->oti_big_buf;
5724 struct dentry *dentry = &oti->oti_obj_dentry;
5725 struct inode *inode = obj->oo_inode;
5726 struct linkea_data ldata = { NULL };
5732 if (!S_ISDIR(inode->i_mode))
5736 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5737 buf->lb_buf, buf->lb_len);
5738 if (rc == -ERANGE) {
5739 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5742 lu_buf_realloc(buf, rc);
5743 if (buf->lb_buf == NULL)
5750 if (unlikely(rc == 0))
5756 if (unlikely(buf->lb_buf == NULL)) {
5757 lu_buf_realloc(buf, rc);
5758 if (buf->lb_buf == NULL)
5765 rc = linkea_init_with_rec(&ldata);
5767 linkea_first_entry(&ldata);
5768 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5774 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5775 struct inode *inode,
5776 const struct lu_fid *pfid,
5777 const char *name, const int namelen)
5779 struct osd_thread_info *oti = osd_oti_get(env);
5780 struct lu_buf *buf = &oti->oti_big_buf;
5781 struct dentry *dentry = &oti->oti_obj_dentry;
5782 struct linkea_data ldata = { NULL };
5783 struct lu_name cname = { .ln_name = name,
5784 .ln_namelen = namelen };
5790 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5791 buf->lb_buf, buf->lb_len);
5793 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5798 if (unlikely(rc == 0))
5801 if (buf->lb_len < rc) {
5802 lu_buf_realloc(buf, rc);
5803 if (buf->lb_buf == NULL)
5810 rc = linkea_init_with_rec(&ldata);
5812 rc = linkea_links_find(&ldata, &cname, pfid);
5818 * Calls ->lookup() to find dentry. From dentry get inode and
5819 * read inode's ea to get fid. This is required for interoperability
5822 * \retval 0, on success
5823 * \retval -ve, on error
5825 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5826 struct dt_rec *rec, const struct dt_key *key)
5828 struct inode *dir = obj->oo_inode;
5829 struct dentry *dentry;
5830 struct ldiskfs_dir_entry_2 *de;
5831 struct buffer_head *bh;
5832 struct lu_fid *fid = (struct lu_fid *)rec;
5833 struct htree_lock *hlock = NULL;
5839 LASSERT(dir->i_op != NULL);
5840 LASSERT(dir->i_op->lookup != NULL);
5842 dentry = osd_child_dentry_get(env, obj,
5843 (char *)key, strlen((char *)key));
5845 if (obj->oo_hl_head != NULL) {
5846 hlock = osd_oti_get(env)->oti_hlock;
5847 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5848 dir, LDISKFS_HLOCK_LOOKUP);
5850 down_read(&obj->oo_ext_idx_sem);
5853 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5855 struct osd_thread_info *oti = osd_oti_get(env);
5856 struct osd_inode_id *id = &oti->oti_id;
5857 struct osd_idmap_cache *oic = &oti->oti_cache;
5858 struct osd_device *dev = osd_obj2dev(obj);
5860 ino = le32_to_cpu(de->inode);
5861 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5863 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5867 rc = osd_get_fid_from_dentry(de, rec);
5869 /* done with de, release bh */
5872 if (unlikely(is_remote_parent_ino(dev, ino))) {
5873 const char *name = (const char *)key;
5876 * If the parent is on remote MDT, and there
5877 * is no FID-in-dirent, then we have to get
5878 * the parent FID from the linkEA.
5880 if (likely(strlen(name) == 2 &&
5881 name[0] == '.' && name[1] == '.'))
5882 rc = osd_get_pfid_from_linkea(env, obj,
5885 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5888 osd_id_gen(id, ino, OSD_OII_NOGEN);
5891 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5892 fid_zero(&oic->oic_fid);
5897 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5898 rc = osd_consistency_check(oti, dev, oic);
5900 fid_zero(&oic->oic_fid);
5902 /* Other error should not affect lookup result. */
5912 ldiskfs_htree_unlock(hlock);
5914 up_read(&obj->oo_ext_idx_sem);
5918 static int osd_index_declare_ea_insert(const struct lu_env *env,
5919 struct dt_object *dt,
5920 const struct dt_rec *rec,
5921 const struct dt_key *key,
5922 struct thandle *handle)
5924 struct osd_thandle *oh;
5925 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5926 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5927 const struct lu_fid *fid = rec1->rec_fid;
5928 int credits, rc = 0;
5929 struct osd_idmap_cache *idc;
5933 LASSERT(!dt_object_remote(dt));
5934 LASSERT(handle != NULL);
5935 LASSERT(fid != NULL);
5936 LASSERT(rec1->rec_type != 0);
5938 oh = container_of0(handle, struct osd_thandle, ot_super);
5939 LASSERT(oh->ot_handle == NULL);
5941 credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5944 * we can't call iget() while a transactions is running
5945 * (this can lead to a deadlock), but we need to know
5946 * inum and object type. so we find this information at
5947 * declaration and cache in per-thread info
5949 idc = osd_idc_find_or_init(env, osd, fid);
5951 RETURN(PTR_ERR(idc));
5952 if (idc->oic_remote) {
5954 * a reference to remote inode is represented by an
5955 * agent inode which we have to create
5957 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5958 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5961 osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5963 if (osd_dt_obj(dt)->oo_inode != NULL) {
5964 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5967 * We ignore block quota on meta pool (MDTs), so needn't
5968 * calculate how many blocks will be consumed by this index
5971 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5973 i_projid_read(inode), 0,
5974 oh, osd_dt_obj(dt), NULL,
5979 #ifdef HAVE_PROJECT_QUOTA
5981 * Reserve credits for local agent inode to transfer
5982 * to 0, quota enforcement is ignored in this case.
5984 if (idc->oic_remote &&
5985 LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
5986 i_projid_read(inode) != 0)
5987 rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
5988 0, i_projid_read(inode),
5989 0, false, PRJQUOTA, true);
5997 * Index add function for interoperability mode (b11826).
5998 * It will add the directory entry.This entry is needed to
5999 * maintain name->fid mapping.
6001 * \param key it is key i.e. file entry to be inserted
6002 * \param rec it is value of given key i.e. fid
6004 * \retval 0, on success
6005 * \retval -ve, on error
6007 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6008 const struct dt_rec *rec,
6009 const struct dt_key *key, struct thandle *th)
6011 struct osd_object *obj = osd_dt_obj(dt);
6012 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6013 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6014 const struct lu_fid *fid = rec1->rec_fid;
6015 const char *name = (const char *)key;
6016 struct osd_thread_info *oti = osd_oti_get(env);
6017 struct inode *child_inode = NULL;
6018 struct osd_idmap_cache *idc;
6023 if (!dt_object_exists(dt))
6026 LASSERT(osd_invariant(obj));
6027 LASSERT(!dt_object_remote(dt));
6028 LASSERT(th != NULL);
6030 osd_trans_exec_op(env, th, OSD_OT_INSERT);
6032 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6034 idc = osd_idc_find(env, osd, fid);
6035 if (unlikely(idc == NULL)) {
6036 idc = osd_idc_find_or_init(env, osd, fid);
6039 * this dt_insert() wasn't declared properly, so
6040 * FID is missing in OI cache. we better do not
6041 * lookup FID in FLDB/OI and don't risk to deadlock,
6042 * but in some special cases (lfsck testing, etc)
6043 * it's much simpler than fixing a caller.
6045 * normally this error should be placed after the first
6046 * find, but migrate may attach source stripes to
6047 * target, which doesn't create stripes.
6049 CERROR("%s: "DFID" wasn't declared for insert\n",
6050 osd_name(osd), PFID(fid));
6052 RETURN(PTR_ERR(idc));
6056 if (idc->oic_remote) {
6057 /* Insert remote entry */
6058 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6060 igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6062 child_inode = osd_create_local_agent_inode(env, osd,
6063 obj, fid, rec1->rec_type & S_IFMT, th);
6064 if (IS_ERR(child_inode))
6065 RETURN(PTR_ERR(child_inode));
6068 /* Insert local entry */
6069 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6070 /* for a reason OI cache wasn't filled properly */
6071 CERROR("%s: OIC for "DFID" isn't filled\n",
6072 osd_name(osd), PFID(fid));
6075 child_inode = oti->oti_inode;
6076 if (unlikely(child_inode == NULL)) {
6077 struct ldiskfs_inode_info *lii;
6082 child_inode = oti->oti_inode = &lii->vfs_inode;
6084 child_inode->i_sb = osd_sb(osd);
6085 child_inode->i_ino = idc->oic_lid.oii_ino;
6086 child_inode->i_mode = rec1->rec_type & S_IFMT;
6089 rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6091 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6092 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6094 if (child_inode && child_inode != oti->oti_inode)
6096 LASSERT(osd_invariant(obj));
6097 osd_trans_exec_check(env, th, OSD_OT_INSERT);
6102 * Initialize osd Iterator for given osd index object.
6104 * \param dt osd index object
6107 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6108 struct dt_object *dt,
6111 struct osd_it_iam *it;
6112 struct osd_object *obj = osd_dt_obj(dt);
6113 struct lu_object *lo = &dt->do_lu;
6114 struct iam_path_descr *ipd;
6115 struct iam_container *bag = &obj->oo_dir->od_container;
6117 if (!dt_object_exists(dt))
6118 return ERR_PTR(-ENOENT);
6122 return ERR_PTR(-ENOMEM);
6124 ipd = osd_it_ipd_get(env, bag);
6125 if (likely(ipd != NULL)) {
6129 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6130 return (struct dt_it *)it;
6133 return ERR_PTR(-ENOMEM);
6138 * free given Iterator.
6140 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6142 struct osd_it_iam *it = (struct osd_it_iam *)di;
6143 struct osd_object *obj = it->oi_obj;
6145 iam_it_fini(&it->oi_it);
6146 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6147 osd_object_put(env, obj);
6152 * Move Iterator to record specified by \a key
6154 * \param di osd iterator
6155 * \param key key for index
6157 * \retval +ve di points to record with least key not larger than key
6158 * \retval 0 di points to exact matched key
6159 * \retval -ve failure
6162 static int osd_it_iam_get(const struct lu_env *env,
6163 struct dt_it *di, const struct dt_key *key)
6165 struct osd_thread_info *oti = osd_oti_get(env);
6166 struct osd_it_iam *it = (struct osd_it_iam *)di;
6168 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6169 /* swab quota uid/gid */
6170 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6171 key = (struct dt_key *)&oti->oti_quota_id;
6174 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6180 * \param di osd iterator
6182 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6184 struct osd_it_iam *it = (struct osd_it_iam *)di;
6186 iam_it_put(&it->oi_it);
6190 * Move iterator by one record
6192 * \param di osd iterator
6194 * \retval +1 end of container reached
6196 * \retval -ve failure
6199 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6201 struct osd_it_iam *it = (struct osd_it_iam *)di;
6203 return iam_it_next(&it->oi_it);
6207 * Return pointer to the key under iterator.
6210 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6211 const struct dt_it *di)
6213 struct osd_thread_info *oti = osd_oti_get(env);
6214 struct osd_it_iam *it = (struct osd_it_iam *)di;
6215 struct osd_object *obj = it->oi_obj;
6218 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6220 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6221 /* swab quota uid/gid */
6222 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6223 key = (struct dt_key *)&oti->oti_quota_id;
6230 * Return size of key under iterator (in bytes)
6233 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6235 struct osd_it_iam *it = (struct osd_it_iam *)di;
6237 return iam_it_key_size(&it->oi_it);
6241 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6243 /* check if file type is required */
6244 if (ent->lde_attrs & LUDA_TYPE) {
6245 struct luda_type *lt;
6246 int align = sizeof(*lt) - 1;
6248 len = (len + align) & ~align;
6249 lt = (struct luda_type *)(ent->lde_name + len);
6250 lt->lt_type = cpu_to_le16(DTTOIF(type));
6253 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6257 * build lu direct from backend fs dirent.
6261 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6262 char *name, __u16 namelen, __u16 type, __u32 attr)
6264 ent->lde_attrs = attr | LUDA_FID;
6265 fid_cpu_to_le(&ent->lde_fid, fid);
6267 ent->lde_hash = cpu_to_le64(offset);
6268 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6270 strncpy(ent->lde_name, name, namelen);
6271 ent->lde_name[namelen] = '\0';
6272 ent->lde_namelen = cpu_to_le16(namelen);
6274 /* append lustre attributes */
6275 osd_it_append_attrs(ent, namelen, type);
6279 * Return pointer to the record under iterator.
6281 static int osd_it_iam_rec(const struct lu_env *env,
6282 const struct dt_it *di,
6283 struct dt_rec *dtrec, __u32 attr)
6285 struct osd_it_iam *it = (struct osd_it_iam *)di;
6286 struct osd_thread_info *info = osd_oti_get(env);
6290 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6291 const struct osd_fid_pack *rec;
6292 struct lu_fid *fid = &info->oti_fid;
6293 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6299 name = (char *)iam_it_key_get(&it->oi_it);
6301 RETURN(PTR_ERR(name));
6303 namelen = iam_it_key_size(&it->oi_it);
6305 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6307 RETURN(PTR_ERR(rec));
6309 rc = osd_fid_unpack(fid, rec);
6313 hash = iam_it_store(&it->oi_it);
6315 /* IAM does not store object type in IAM index (dir) */
6316 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6318 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6319 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6320 (struct iam_rec *)dtrec);
6321 osd_quota_unpack(it->oi_obj, dtrec);
6323 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6324 (struct iam_rec *)dtrec);
6331 * Returns cookie for current Iterator position.
6333 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6335 struct osd_it_iam *it = (struct osd_it_iam *)di;
6337 return iam_it_store(&it->oi_it);
6341 * Restore iterator from cookie.
6343 * \param di osd iterator
6344 * \param hash Iterator location cookie
6346 * \retval +ve di points to record with least key not larger than key.
6347 * \retval 0 di points to exact matched key
6348 * \retval -ve failure
6351 static int osd_it_iam_load(const struct lu_env *env,
6352 const struct dt_it *di, __u64 hash)
6354 struct osd_it_iam *it = (struct osd_it_iam *)di;
6356 return iam_it_load(&it->oi_it, hash);
6359 static const struct dt_index_operations osd_index_iam_ops = {
6360 .dio_lookup = osd_index_iam_lookup,
6361 .dio_declare_insert = osd_index_declare_iam_insert,
6362 .dio_insert = osd_index_iam_insert,
6363 .dio_declare_delete = osd_index_declare_iam_delete,
6364 .dio_delete = osd_index_iam_delete,
6366 .init = osd_it_iam_init,
6367 .fini = osd_it_iam_fini,
6368 .get = osd_it_iam_get,
6369 .put = osd_it_iam_put,
6370 .next = osd_it_iam_next,
6371 .key = osd_it_iam_key,
6372 .key_size = osd_it_iam_key_size,
6373 .rec = osd_it_iam_rec,
6374 .store = osd_it_iam_store,
6375 .load = osd_it_iam_load
6381 * Creates or initializes iterator context.
6383 * \retval struct osd_it_ea, iterator structure on success
6386 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6387 struct dt_object *dt,
6390 struct osd_object *obj = osd_dt_obj(dt);
6391 struct osd_thread_info *info = osd_oti_get(env);
6392 struct osd_it_ea *oie;
6394 struct lu_object *lo = &dt->do_lu;
6395 struct dentry *obj_dentry;
6399 if (!dt_object_exists(dt) || obj->oo_destroyed)
6400 RETURN(ERR_PTR(-ENOENT));
6402 OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6404 RETURN(ERR_PTR(-ENOMEM));
6405 obj_dentry = &oie->oie_dentry;
6407 obj_dentry->d_inode = obj->oo_inode;
6408 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6409 obj_dentry->d_name.hash = 0;
6411 oie->oie_rd_dirent = 0;
6412 oie->oie_it_dirent = 0;
6413 oie->oie_dirent = NULL;
6414 if (unlikely(!info->oti_it_ea_buf_used)) {
6415 oie->oie_buf = info->oti_it_ea_buf;
6416 info->oti_it_ea_buf_used = 1;
6418 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6419 if (oie->oie_buf == NULL)
6420 RETURN(ERR_PTR(-ENOMEM));
6424 file = &oie->oie_file;
6426 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6427 if (attr & LUDA_64BITHASH)
6428 file->f_mode = FMODE_64BITHASH;
6430 file->f_mode = FMODE_32BITHASH;
6431 file->f_path.dentry = obj_dentry;
6432 file->f_mapping = obj->oo_inode->i_mapping;
6433 file->f_op = obj->oo_inode->i_fop;
6434 file->f_inode = obj->oo_inode;
6437 RETURN((struct dt_it *)oie);
6441 * Destroy or finishes iterator context.
6443 * \param di iterator structure to be destroyed
6445 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6447 struct osd_thread_info *info = osd_oti_get(env);
6448 struct osd_it_ea *oie = (struct osd_it_ea *)di;
6449 struct osd_object *obj = oie->oie_obj;
6450 struct inode *inode = obj->oo_inode;
6453 oie->oie_file.f_op->release(inode, &oie->oie_file);
6454 osd_object_put(env, obj);
6455 if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6456 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6458 info->oti_it_ea_buf_used = 0;
6459 OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6464 * It position the iterator at given key, so that next lookup continues from
6465 * that key Or it is similar to dio_it->load() but based on a key,
6466 * rather than file position.
6468 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6471 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6473 static int osd_it_ea_get(const struct lu_env *env,
6474 struct dt_it *di, const struct dt_key *key)
6476 struct osd_it_ea *it = (struct osd_it_ea *)di;
6479 LASSERT(((const char *)key)[0] == '\0');
6480 it->oie_file.f_pos = 0;
6481 it->oie_rd_dirent = 0;
6482 it->oie_it_dirent = 0;
6483 it->oie_dirent = NULL;
6491 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6495 struct osd_filldir_cbs {
6496 struct dir_context ctx;
6497 struct osd_it_ea *it;
6500 * It is called internally by ->iterate*(). It fills the
6501 * iterator's in-memory data structure with required
6502 * information i.e. name, namelen, rec_size etc.
6504 * \param buf in which information to be filled in.
6505 * \param name name of the file in given dir
6507 * \retval 0 on success
6508 * \retval 1 on buffer full
6510 #ifdef HAVE_FILLDIR_USE_CTX
6511 static int osd_ldiskfs_filldir(struct dir_context *buf,
6513 static int osd_ldiskfs_filldir(void *buf,
6515 const char *name, int namelen,
6516 loff_t offset, __u64 ino, unsigned int d_type)
6518 struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6519 struct osd_object *obj = it->oie_obj;
6520 struct osd_it_ea_dirent *ent = it->oie_dirent;
6521 struct lu_fid *fid = &ent->oied_fid;
6522 struct osd_fid_pack *rec;
6526 /* this should never happen */
6527 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6528 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6532 if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6536 /* "." is just the object itself. */
6537 if (namelen == 1 && name[0] == '.') {
6538 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6539 } else if (d_type & LDISKFS_DIRENT_LUFID) {
6540 rec = (struct osd_fid_pack *)(name + namelen + 1);
6541 if (osd_fid_unpack(fid, rec) != 0)
6546 d_type &= ~LDISKFS_DIRENT_LUFID;
6548 /* NOT export local root. */
6549 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6550 ino = obj->oo_inode->i_ino;
6551 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6554 ent->oied_ino = ino;
6555 ent->oied_off = offset;
6556 ent->oied_namelen = namelen;
6557 ent->oied_type = d_type;
6559 memcpy(ent->oied_name, name, namelen);
6561 it->oie_rd_dirent++;
6562 it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6567 * Calls ->iterate*() to load a directory entry at a time
6568 * and stored it in iterator's in-memory data structure.
6570 * \param di iterator's in memory structure
6572 * \retval 0 on success
6573 * \retval -ve on error
6574 * \retval +1 reach the end of entry
6576 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6577 const struct dt_it *di)
6579 struct osd_it_ea *it = (struct osd_it_ea *)di;
6580 struct osd_object *obj = it->oie_obj;
6581 struct inode *inode = obj->oo_inode;
6582 struct htree_lock *hlock = NULL;
6583 struct file *filp = &it->oie_file;
6585 struct osd_filldir_cbs buf = {
6586 .ctx.actor = osd_ldiskfs_filldir,
6591 it->oie_dirent = it->oie_buf;
6592 it->oie_rd_dirent = 0;
6594 if (obj->oo_hl_head != NULL) {
6595 hlock = osd_oti_get(env)->oti_hlock;
6596 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6597 inode, LDISKFS_HLOCK_READDIR);
6599 down_read(&obj->oo_ext_idx_sem);
6602 filp->f_cred = current_cred();
6603 rc = osd_security_file_alloc(filp);
6607 filp->f_flags |= O_NOATIME;
6608 filp->f_mode |= FMODE_NONOTIFY;
6609 rc = iterate_dir(filp, &buf.ctx);
6614 ldiskfs_htree_unlock(hlock);
6616 up_read(&obj->oo_ext_idx_sem);
6618 if (it->oie_rd_dirent == 0) {
6620 * If it does not get any dirent, it means it has been reached
6621 * to the end of the dir
6623 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6627 it->oie_dirent = it->oie_buf;
6628 it->oie_it_dirent = 1;
6635 * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
6636 * to load a directory entry at a time and stored it in
6637 * iterator's in-memory data structure.
6639 * \param di iterator's in memory structure
6641 * \retval +ve iterator reached to end
6642 * \retval 0 iterator not reached to end
6643 * \retval -ve on error
6645 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6647 struct osd_it_ea *it = (struct osd_it_ea *)di;
6652 if (it->oie_it_dirent < it->oie_rd_dirent) {
6654 (void *)it->oie_dirent +
6655 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6656 it->oie_dirent->oied_namelen);
6657 it->oie_it_dirent++;
6660 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6663 rc = osd_ldiskfs_it_fill(env, di);
6670 * Returns the key at current position from iterator's in memory structure.
6672 * \param di iterator's in memory structure
6674 * \retval key i.e. struct dt_key on success
6676 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6677 const struct dt_it *di)
6679 struct osd_it_ea *it = (struct osd_it_ea *)di;
6681 return (struct dt_key *)it->oie_dirent->oied_name;
6685 * Returns key's size at current position from iterator's in memory structure.
6687 * \param di iterator's in memory structure
6689 * \retval key_size i.e. struct dt_key on success
6691 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6693 struct osd_it_ea *it = (struct osd_it_ea *)di;
6695 return it->oie_dirent->oied_namelen;
6698 #if defined LDISKFS_DIR_ENTRY_LEN && defined LDISKFS_DIR_ENTRY_LEN_
6699 #undef LDISKFS_DIR_REC_LEN
6700 #define LDISKFS_DIR_REC_LEN(de) LDISKFS_DIR_ENTRY_LEN_((de))
6703 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6705 if (LDISKFS_DIR_REC_LEN(de) >=
6706 __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6713 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6714 unsigned int blocksize, bool dotdot)
6717 return osd_dotdot_has_space(de);
6719 if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6720 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6727 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6728 handle_t *jh, struct dentry *dentry,
6729 const struct lu_fid *fid, struct buffer_head *bh,
6730 struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6733 struct inode *dir = dentry->d_parent->d_inode;
6734 struct inode *inode = dentry->d_inode;
6735 struct osd_fid_pack *rec;
6736 struct ldiskfs_dentry_param *ldp;
6737 int namelen = dentry->d_name.len;
6739 struct osd_thread_info *info = osd_oti_get(env);
6743 if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6746 /* There is enough space to hold the FID-in-dirent. */
6747 if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6748 rc = ldiskfs_journal_get_write_access(jh, bh);
6752 de->name[namelen] = 0;
6753 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6754 rec->fp_len = sizeof(struct lu_fid) + 1;
6755 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6756 de->file_type |= LDISKFS_DIRENT_LUFID;
6757 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6764 rc = ldiskfs_delete_entry(jh, dir, de, bh);
6768 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6769 osd_get_ldiskfs_dirent_param(ldp, fid);
6770 dentry->d_fsdata = (void *)ldp;
6771 dquot_initialize(dir);
6772 rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6774 * It is too bad, we cannot reinsert the name entry back.
6775 * That means we lose it!
6778 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6779 "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6780 osd_ino2name(inode),
6781 dir->i_ino, dir->i_generation, namelen,
6782 dentry->d_name.name, PFID(fid), rc);
6788 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6789 struct osd_it_ea *it, struct lu_fid *fid,
6790 struct osd_inode_id *id, __u32 *attr)
6792 struct osd_thread_info *info = osd_oti_get(env);
6793 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6794 struct osd_device *dev = osd_obj2dev(obj);
6795 struct super_block *sb = osd_sb(dev);
6796 const char *devname = osd_name(dev);
6797 struct osd_it_ea_dirent *ent = it->oie_dirent;
6798 struct inode *dir = obj->oo_inode;
6799 struct htree_lock *hlock = NULL;
6800 struct buffer_head *bh = NULL;
6801 handle_t *jh = NULL;
6802 struct ldiskfs_dir_entry_2 *de;
6803 struct dentry *dentry;
6804 struct inode *inode;
6805 const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6808 bool dotdot = false;
6813 if (ent->oied_name[0] == '.') {
6814 if (ent->oied_namelen == 1)
6817 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6821 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6822 inode = osd_iget(info, dev, id);
6823 if (IS_ERR(inode)) {
6824 rc = PTR_ERR(inode);
6825 if (rc == -ENOENT || rc == -ESTALE) {
6827 * Maybe dangling name entry, or
6828 * corrupted directory entry.
6830 *attr |= LUDA_UNKNOWN;
6833 CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6834 "check_repair, dir = %lu/%u, name = %.*s, "
6835 "ino = %llu, rc = %d\n",
6836 devname, dir->i_ino, dir->i_generation,
6837 ent->oied_namelen, ent->oied_name,
6844 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6846 rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6847 if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6853 * We need to ensure that the name entry is still valid.
6854 * Because it may be removed or renamed by other already.
6856 * The unlink or rename operation will start journal before PDO lock,
6857 * so to avoid deadlock, here we need to start journal handle before
6858 * related PDO lock also. But because we do not know whether there
6859 * will be something to be repaired before PDO lock, we just start
6860 * journal without conditions.
6862 * We may need to remove the name entry firstly, then insert back.
6863 * One credit is for user quota file update.
6864 * One credit is for group quota file update.
6865 * Two credits are for dirty inode.
6867 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6868 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6870 if (dev->od_dirent_journal != 0) {
6873 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6876 CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6877 "check_repair, dir = %lu/%u, credits = %d, "
6878 "name = %.*s, ino = %llu: rc = %d\n",
6879 devname, dir->i_ino, dir->i_generation, credits,
6880 ent->oied_namelen, ent->oied_name,
6883 GOTO(out_inode, rc);
6886 if (obj->oo_hl_head != NULL) {
6887 hlock = osd_oti_get(env)->oti_hlock;
6889 * "0" means exclusive lock for the whole directory.
6890 * We need to prevent others access such name entry
6891 * during the delete + insert. Neither HLOCK_ADD nor
6892 * HLOCK_DEL cannot guarantee the atomicity.
6894 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6896 down_write(&obj->oo_ext_idx_sem);
6899 if (obj->oo_hl_head != NULL) {
6900 hlock = osd_oti_get(env)->oti_hlock;
6901 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6902 LDISKFS_HLOCK_LOOKUP);
6904 down_read(&obj->oo_ext_idx_sem);
6908 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6909 if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6910 *attr |= LUDA_IGNORE;
6916 * For dotdot entry, if there is not enough space to hold the
6917 * FID-in-dirent, just keep them there. It only happens when the
6918 * device upgraded from 1.8 or restored from MDT file-level backup.
6919 * For the whole directory, only dotdot entry have no FID-in-dirent
6920 * and needs to get FID from LMA when readdir, it will not affect the
6923 if (dotdot && !osd_dotdot_has_space(de)) {
6924 *attr |= LUDA_UNKNOWN;
6930 if (lu_fid_eq(fid, &lma->lma_self_fid))
6933 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6934 struct lu_fid *tfid = &lma->lma_self_fid;
6936 if (likely(dotdot &&
6937 fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6938 fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6940 * It must be REMOTE_PARENT_DIR and as the
6941 * 'dotdot' entry of remote directory
6943 *attr |= LUDA_IGNORE;
6945 CDEBUG(D_LFSCK, "%s: expect remote agent "
6946 "parent directory, but got %.*s under "
6947 "dir = %lu/%u with the FID "DFID"\n",
6948 devname, ent->oied_namelen,
6949 ent->oied_name, dir->i_ino,
6950 dir->i_generation, PFID(tfid));
6952 *attr |= LUDA_UNKNOWN;
6959 if (!fid_is_zero(fid)) {
6960 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6962 if (rc == -ENOENT ||
6964 !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
6966 * linkEA does not recognize the dirent entry,
6967 * it may because the dirent entry corruption
6968 * and points to other's inode.
6970 CDEBUG(D_LFSCK, "%s: the target inode does not "
6971 "recognize the dirent, dir = %lu/%u, "
6972 " name = %.*s, ino = %llu, "
6973 DFID": rc = %d\n", devname, dir->i_ino,
6974 dir->i_generation, ent->oied_namelen,
6975 ent->oied_name, ent->oied_ino, PFID(fid), rc);
6976 *attr |= LUDA_UNKNOWN;
6981 if (rc && rc != -ENODATA) {
6982 CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6983 "dir = %lu/%u, name = %.*s, ino = %llu, "
6984 DFID": rc = %d\n", devname, dir->i_ino,
6985 dir->i_generation, ent->oied_namelen,
6986 ent->oied_name, ent->oied_ino, PFID(fid), rc);
6987 *attr |= LUDA_UNKNOWN;
6995 * linkEA recognizes the dirent entry, the FID-in-LMA is
6996 * valid, trusted, in spite of fid_is_sane(fid) or not.
6998 if (*attr & LUDA_VERIFY_DRYRUN) {
6999 *fid = lma->lma_self_fid;
7000 *attr |= LUDA_REPAIR;
7007 dev->od_dirent_journal = 1;
7008 if (hlock != NULL) {
7009 ldiskfs_htree_unlock(hlock);
7012 up_read(&obj->oo_ext_idx_sem);
7018 *fid = lma->lma_self_fid;
7020 /* Update or append the FID-in-dirent. */
7021 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7022 bh, de, hlock, dotdot);
7024 *attr |= LUDA_REPAIR;
7026 CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7027 "the dirent, dir = %lu/%u, name = %.*s, "
7028 "ino = %llu, "DFID": rc = %d\n",
7029 devname, dir->i_ino, dir->i_generation,
7030 ent->oied_namelen, ent->oied_name,
7031 ent->oied_ino, PFID(fid), rc);
7033 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7034 if (*attr & LUDA_VERIFY_DRYRUN) {
7035 if (fid_is_sane(fid)) {
7036 *attr |= LUDA_REPAIR;
7037 } else if (dev->od_index == 0) {
7038 lu_igif_build(fid, inode->i_ino,
7039 inode->i_generation);
7040 *attr |= LUDA_UPGRADE;
7048 dev->od_dirent_journal = 1;
7049 if (hlock != NULL) {
7050 ldiskfs_htree_unlock(hlock);
7053 up_read(&obj->oo_ext_idx_sem);
7060 if (unlikely(fid_is_sane(fid))) {
7062 * FID-in-dirent exists, but FID-in-LMA is lost.
7063 * Trust the FID-in-dirent, and add FID-in-LMA.
7065 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7067 *attr |= LUDA_REPAIR;
7069 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7070 "update dirent, dir = %lu/%u, "
7071 "name = %.*s, ino = %llu, "
7073 devname, dir->i_ino, dir->i_generation,
7074 ent->oied_namelen, ent->oied_name,
7075 ent->oied_ino, PFID(fid), rc);
7076 } else if (dev->od_index == 0) {
7077 lu_igif_build(fid, inode->i_ino, inode->i_generation);
7079 * It is probably IGIF object. Only aappend the
7080 * FID-in-dirent. OI scrub will process FID-in-LMA.
7082 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7083 bh, de, hlock, dotdot);
7085 *attr |= LUDA_UPGRADE;
7087 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7088 "after the dirent, dir = %lu/%u, "
7089 "name = %.*s, ino = %llu, "
7091 devname, dir->i_ino, dir->i_generation,
7092 ent->oied_namelen, ent->oied_name,
7093 ent->oied_ino, PFID(fid), rc);
7102 if (hlock != NULL) {
7103 ldiskfs_htree_unlock(hlock);
7105 if (dev->od_dirent_journal != 0)
7106 up_write(&obj->oo_ext_idx_sem);
7108 up_read(&obj->oo_ext_idx_sem);
7112 ldiskfs_journal_stop(jh);
7116 if (rc >= 0 && !dirty)
7117 dev->od_dirent_journal = 0;
7123 * Returns the value at current position from iterator's in memory structure.
7125 * \param di struct osd_it_ea, iterator's in memory structure
7126 * \param attr attr requested for dirent.
7127 * \param lde lustre dirent
7129 * \retval 0 no error and \param lde has correct lustre dirent.
7130 * \retval -ve on error
7132 static inline int osd_it_ea_rec(const struct lu_env *env,
7133 const struct dt_it *di,
7134 struct dt_rec *dtrec, __u32 attr)
7136 struct osd_it_ea *it = (struct osd_it_ea *)di;
7137 struct osd_object *obj = it->oie_obj;
7138 struct osd_device *dev = osd_obj2dev(obj);
7139 struct osd_thread_info *oti = osd_oti_get(env);
7140 struct osd_inode_id *id = &oti->oti_id;
7141 struct lu_fid *fid = &it->oie_dirent->oied_fid;
7142 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
7143 __u32 ino = it->oie_dirent->oied_ino;
7148 LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7150 if (attr & LUDA_VERIFY) {
7151 if (unlikely(is_remote_parent_ino(dev, ino))) {
7152 attr |= LUDA_IGNORE;
7154 * If the parent is on remote MDT, and there
7155 * is no FID-in-dirent, then we have to get
7156 * the parent FID from the linkEA.
7158 if (!fid_is_sane(fid) &&
7159 it->oie_dirent->oied_namelen == 2 &&
7160 it->oie_dirent->oied_name[0] == '.' &&
7161 it->oie_dirent->oied_name[1] == '.')
7162 osd_get_pfid_from_linkea(env, obj, fid);
7164 rc = osd_dirent_check_repair(env, obj, it, fid, id,
7168 if (!fid_is_sane(fid))
7169 attr |= LUDA_UNKNOWN;
7171 attr &= ~LU_DIRENT_ATTRS_MASK;
7172 if (!fid_is_sane(fid)) {
7173 bool is_dotdot = false;
7175 if (it->oie_dirent->oied_namelen == 2 &&
7176 it->oie_dirent->oied_name[0] == '.' &&
7177 it->oie_dirent->oied_name[1] == '.')
7180 * If the parent is on remote MDT, and there
7181 * is no FID-in-dirent, then we have to get
7182 * the parent FID from the linkEA.
7184 if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7185 rc = osd_get_pfid_from_linkea(env, obj, fid);
7187 if (is_dotdot == false &&
7188 OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7191 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7194 osd_id_gen(id, ino, OSD_OII_NOGEN);
7198 /* Pack the entry anyway, at least the offset is right. */
7199 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7200 it->oie_dirent->oied_name,
7201 it->oie_dirent->oied_namelen,
7202 it->oie_dirent->oied_type, attr);
7207 if (osd_remote_fid(env, dev, fid))
7210 if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7211 osd_add_oi_cache(oti, dev, id, fid);
7213 RETURN(rc > 0 ? 0 : rc);
7217 * Returns the record size size at current position.
7219 * This function will return record(lu_dirent) size in bytes.
7221 * \param[in] env execution environment
7222 * \param[in] di iterator's in memory structure
7223 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
7224 * calculate the lu_dirent size.
7226 * \retval record size(in bytes & in memory) of the current lu_dirent
7229 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7232 struct osd_it_ea *it = (struct osd_it_ea *)di;
7234 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7238 * Returns a cookie for current position of the iterator head, so that
7239 * user can use this cookie to load/start the iterator next time.
7241 * \param di iterator's in memory structure
7243 * \retval cookie for current position, on success
7245 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7247 struct osd_it_ea *it = (struct osd_it_ea *)di;
7249 return it->oie_dirent->oied_off;
7253 * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
7254 * to load a directory entry at a time and stored it i inn,
7255 * in iterator's in-memory data structure.
7257 * \param di struct osd_it_ea, iterator's in memory structure
7259 * \retval +ve on success
7260 * \retval -ve on error
7262 static int osd_it_ea_load(const struct lu_env *env,
7263 const struct dt_it *di, __u64 hash)
7265 struct osd_it_ea *it = (struct osd_it_ea *)di;
7269 it->oie_file.f_pos = hash;
7271 rc = osd_ldiskfs_it_fill(env, di);
7282 * Index lookup function for interoperability mode (b11826).
7284 * \param key, key i.e. file name to be searched
7286 * \retval +ve, on success
7287 * \retval -ve, on error
7289 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7290 struct dt_rec *rec, const struct dt_key *key)
7292 struct osd_object *obj = osd_dt_obj(dt);
7297 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7298 LINVRNT(osd_invariant(obj));
7300 rc = osd_ea_lookup_rec(env, obj, rec, key);
7307 * Index and Iterator operations for interoperability
7308 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7310 static const struct dt_index_operations osd_index_ea_ops = {
7311 .dio_lookup = osd_index_ea_lookup,
7312 .dio_declare_insert = osd_index_declare_ea_insert,
7313 .dio_insert = osd_index_ea_insert,
7314 .dio_declare_delete = osd_index_declare_ea_delete,
7315 .dio_delete = osd_index_ea_delete,
7317 .init = osd_it_ea_init,
7318 .fini = osd_it_ea_fini,
7319 .get = osd_it_ea_get,
7320 .put = osd_it_ea_put,
7321 .next = osd_it_ea_next,
7322 .key = osd_it_ea_key,
7323 .key_size = osd_it_ea_key_size,
7324 .rec = osd_it_ea_rec,
7325 .rec_size = osd_it_ea_rec_size,
7326 .store = osd_it_ea_store,
7327 .load = osd_it_ea_load
7331 static void *osd_key_init(const struct lu_context *ctx,
7332 struct lu_context_key *key)
7334 struct osd_thread_info *info;
7336 OBD_ALLOC_PTR(info);
7338 return ERR_PTR(-ENOMEM);
7340 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7341 if (info->oti_it_ea_buf == NULL)
7344 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7346 info->oti_hlock = ldiskfs_htree_lock_alloc();
7347 if (info->oti_hlock == NULL)
7353 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7356 return ERR_PTR(-ENOMEM);
7359 static void osd_key_fini(const struct lu_context *ctx,
7360 struct lu_context_key *key, void *data)
7362 struct osd_thread_info *info = data;
7363 struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7364 struct osd_idmap_cache *idc = info->oti_ins_cache;
7366 if (info->oti_dio_pages) {
7368 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7369 struct page *page = info->oti_dio_pages[i];
7371 LASSERT(PagePrivate2(page));
7372 LASSERT(PageLocked(page));
7373 ClearPagePrivate2(page);
7378 OBD_FREE_PTR_ARRAY(info->oti_dio_pages, PTLRPC_MAX_BRW_PAGES);
7381 if (info->oti_inode != NULL)
7383 if (info->oti_hlock != NULL)
7384 ldiskfs_htree_lock_free(info->oti_hlock);
7385 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7386 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7387 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7388 lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7389 lu_buf_free(&info->oti_big_buf);
7391 LASSERT(info->oti_ins_cache_size > 0);
7392 OBD_FREE_PTR_ARRAY(idc, info->oti_ins_cache_size);
7393 info->oti_ins_cache = NULL;
7394 info->oti_ins_cache_size = 0;
7399 static void osd_key_exit(const struct lu_context *ctx,
7400 struct lu_context_key *key, void *data)
7402 struct osd_thread_info *info = data;
7404 LASSERT(info->oti_r_locks == 0);
7405 LASSERT(info->oti_w_locks == 0);
7406 LASSERT(info->oti_txns == 0);
7409 /* type constructor/destructor: osd_type_init, osd_type_fini */
7410 LU_TYPE_INIT_FINI(osd, &osd_key);
7412 struct lu_context_key osd_key = {
7413 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7414 .lct_init = osd_key_init,
7415 .lct_fini = osd_key_fini,
7416 .lct_exit = osd_key_exit
7420 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7421 const char *name, struct lu_device *next)
7423 struct osd_device *osd = osd_dev(d);
7425 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7426 sizeof(osd->od_svname))
7428 return osd_procfs_init(osd, name);
7431 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7433 struct seq_server_site *ss = osd_seq_site(osd);
7438 if (osd->od_is_ost || osd->od_cl_seq != NULL)
7441 if (unlikely(ss == NULL))
7444 OBD_ALLOC_PTR(osd->od_cl_seq);
7445 if (osd->od_cl_seq == NULL)
7448 seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7449 osd->od_svname, ss->ss_server_seq);
7451 if (ss->ss_node_id == 0) {
7453 * If the OSD on the sequence controller(MDT0), then allocate
7454 * sequence here, otherwise allocate sequence after connected
7455 * to MDT0 (see mdt_register_lwp_callback()).
7457 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7458 &osd->od_cl_seq->lcs_space, env);
7464 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7466 if (osd->od_cl_seq == NULL)
7469 seq_client_fini(osd->od_cl_seq);
7470 OBD_FREE_PTR(osd->od_cl_seq);
7471 osd->od_cl_seq = NULL;
7474 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7478 /* shutdown quota slave instance associated with the device */
7479 if (o->od_quota_slave_md != NULL) {
7480 struct qsd_instance *qsd = o->od_quota_slave_md;
7482 o->od_quota_slave_md = NULL;
7486 if (o->od_quota_slave_dt != NULL) {
7487 struct qsd_instance *qsd = o->od_quota_slave_dt;
7489 o->od_quota_slave_dt = NULL;
7493 osd_fid_fini(env, o);
7494 osd_scrub_cleanup(env, o);
7499 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7503 if (o->od_mnt != NULL) {
7504 shrink_dcache_sb(osd_sb(o));
7505 osd_sync(env, &o->od_dt_dev);
7514 static int osd_mount(const struct lu_env *env,
7515 struct osd_device *o, struct lustre_cfg *cfg)
7517 const char *name = lustre_cfg_string(cfg, 0);
7518 const char *dev = lustre_cfg_string(cfg, 1);
7520 unsigned long page, s_flags, lmd_flags = 0;
7521 struct page *__page;
7522 struct file_system_type *type;
7523 char *options = NULL;
7525 struct osd_thread_info *info = osd_oti_get(env);
7526 struct lu_fid *fid = &info->oti_fid;
7527 struct inode *inode;
7528 int rc = 0, force_over_1024tb = 0;
7532 if (o->od_mnt != NULL)
7535 if (strlen(dev) >= sizeof(o->od_mntdev))
7537 strcpy(o->od_mntdev, dev);
7539 str = lustre_cfg_string(cfg, 2);
7540 s_flags = simple_strtoul(str, NULL, 0);
7541 str = strstr(str, ":");
7543 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7544 opts = lustre_cfg_string(cfg, 3);
7546 if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7547 CERROR("%s: device %s extents feature is not guaranteed to "
7548 "work on big-endian systems. Use \"bigendian_extents\" "
7549 "mount option to override.\n", name, dev);
7553 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7554 if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7555 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");
7558 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7559 if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7560 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");
7563 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
7564 if (opts != NULL && strstr(opts, "force_over_512tb") != NULL) {
7565 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");
7569 if (opts != NULL && strstr(opts, "force_over_1024tb") != NULL)
7570 force_over_1024tb = 1;
7572 __page = alloc_page(GFP_KERNEL);
7574 GOTO(out, rc = -ENOMEM);
7575 page = (unsigned long)page_address(__page);
7576 options = (char *)page;
7579 /* strip out the options for back compatiblity */
7580 static const char * const sout[] = {
7587 /* strip out option we processed in osd */
7588 "bigendian_extents",
7592 "force_over_1024tb",
7595 strncat(options, opts, PAGE_SIZE);
7596 for (rc = 0, str = options; sout[rc]; ) {
7597 char *op = strstr(str, sout[rc]);
7604 if (op == options || *(op - 1) == ',') {
7605 str = op + strlen(sout[rc]);
7606 if (*str == ',' || *str == '\0') {
7607 *str == ',' ? str++ : str;
7608 memmove(op, str, strlen(str) + 1);
7611 for (str = op; *str != ',' && *str != '\0'; str++)
7615 strncat(options, "user_xattr,acl", PAGE_SIZE);
7618 /* Glom up mount options */
7619 if (*options != '\0')
7620 strncat(options, ",", PAGE_SIZE);
7621 strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7623 type = get_fs_type("ldiskfs");
7625 CERROR("%s: cannot find ldiskfs module\n", name);
7626 GOTO(out, rc = -ENODEV);
7629 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7630 module_put(type->owner);
7632 if (IS_ERR(o->od_mnt)) {
7633 rc = PTR_ERR(o->od_mnt);
7635 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7639 if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7640 osd_sb(o)->s_blocksize_bits > 1024ULL << 40 &&
7641 force_over_1024tb == 0) {
7642 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",
7644 GOTO(out_mnt, rc = -EINVAL);
7647 if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7648 LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7651 GOTO(out_mnt, rc = -EOPNOTSUPP);
7654 if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7655 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7656 GOTO(out_mnt, rc = -EINVAL);
7659 #ifdef LDISKFS_MOUNT_DIRDATA
7660 if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7661 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7662 else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7663 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7664 "enabling the dirdata feature. If you do not want to "
7665 "downgrade to Lustre-1.x again, you can enable it via "
7666 "'tune2fs -O dirdata device'\n", name, dev);
7668 inode = osd_sb(o)->s_root->d_inode;
7669 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7670 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7672 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7676 if (lmd_flags & LMD_FLG_NOSCRUB)
7677 o->od_auto_scrub_interval = AS_NEVER;
7679 if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7680 /* do not use pagecache with flash-backed storage */
7681 o->od_writethrough_cache = 0;
7682 o->od_read_cache = 0;
7693 __free_page(__page);
7698 static struct lu_device *osd_device_fini(const struct lu_env *env,
7699 struct lu_device *d)
7701 struct osd_device *o = osd_dev(d);
7705 osd_index_backup(env, o, false);
7706 osd_shutdown(env, o);
7708 if (o->od_oi_table != NULL)
7709 osd_oi_fini(osd_oti_get(env), o);
7710 osd_obj_map_fini(o);
7716 static int osd_device_init0(const struct lu_env *env,
7717 struct osd_device *o,
7718 struct lustre_cfg *cfg)
7720 struct lu_device *l = osd2lu_dev(o);
7721 struct osd_thread_info *info;
7725 /* if the module was re-loaded, env can loose its keys */
7726 rc = lu_env_refill((struct lu_env *)env);
7729 info = osd_oti_get(env);
7732 l->ld_ops = &osd_lu_ops;
7733 o->od_dt_dev.dd_ops = &osd_dt_ops;
7735 spin_lock_init(&o->od_osfs_lock);
7736 mutex_init(&o->od_otable_mutex);
7737 INIT_LIST_HEAD(&o->od_orphan_list);
7738 INIT_LIST_HEAD(&o->od_index_backup_list);
7739 INIT_LIST_HEAD(&o->od_index_restore_list);
7740 spin_lock_init(&o->od_lock);
7741 o->od_index_backup_policy = LIBP_NONE;
7744 o->od_read_cache = 1;
7745 o->od_writethrough_cache = 1;
7746 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7747 o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20;
7748 o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20;
7749 o->od_auto_scrub_interval = AS_DEFAULT;
7751 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7752 sizeof(o->od_svname));
7753 if (cplen >= sizeof(o->od_svname)) {
7758 o->od_index_backup_stop = 0;
7759 o->od_index = -1; /* -1 means index is invalid */
7760 rc = server_name2index(o->od_svname, &o->od_index, NULL);
7761 if (rc == LDD_F_SV_TYPE_OST)
7764 o->od_full_scrub_ratio = OFSR_DEFAULT;
7765 o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7766 rc = osd_mount(env, o, cfg);
7770 /* Can only check block device after mount */
7771 o->od_nonrotational =
7772 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
7774 rc = osd_obj_map_init(env, o);
7778 rc = lu_site_init(&o->od_site, l);
7780 GOTO(out_compat, rc);
7781 o->od_site.ls_bottom_dev = l;
7783 rc = lu_site_init_finish(&o->od_site);
7787 INIT_LIST_HEAD(&o->od_ios_list);
7788 /* setup scrub, including OI files initialization */
7790 rc = osd_scrub_setup(env, o);
7795 rc = osd_procfs_init(o, o->od_svname);
7797 CERROR("%s: can't initialize procfs: rc = %d\n",
7799 GOTO(out_scrub, rc);
7802 LASSERT(l->ld_site->ls_linkage.next != NULL);
7803 LASSERT(l->ld_site->ls_linkage.prev != NULL);
7805 /* initialize quota slave instance */
7806 /* currently it's no need to prepare qsd_instance_md for OST */
7807 if (!o->od_is_ost) {
7808 o->od_quota_slave_md = qsd_init(env, o->od_svname,
7810 o->od_proc_entry, true);
7811 if (IS_ERR(o->od_quota_slave_md)) {
7812 rc = PTR_ERR(o->od_quota_slave_md);
7813 o->od_quota_slave_md = NULL;
7814 GOTO(out_procfs, rc);
7818 o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
7819 o->od_proc_entry, false);
7821 if (IS_ERR(o->od_quota_slave_dt)) {
7822 if (o->od_quota_slave_md != NULL) {
7823 qsd_fini(env, o->od_quota_slave_md);
7824 o->od_quota_slave_md = NULL;
7827 rc = PTR_ERR(o->od_quota_slave_dt);
7828 o->od_quota_slave_dt = NULL;
7829 GOTO(out_procfs, rc);
7837 osd_scrub_cleanup(env, o);
7839 lu_site_fini(&o->od_site);
7841 osd_obj_map_fini(o);
7848 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7849 struct lu_device_type *t,
7850 struct lustre_cfg *cfg)
7852 struct osd_device *o;
7857 return ERR_PTR(-ENOMEM);
7859 rc = dt_device_init(&o->od_dt_dev, t);
7862 * Because the ctx might be revived in dt_device_init,
7863 * refill the env here
7865 lu_env_refill((struct lu_env *)env);
7866 rc = osd_device_init0(env, o, cfg);
7868 dt_device_fini(&o->od_dt_dev);
7871 if (unlikely(rc != 0))
7874 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7877 static struct lu_device *osd_device_free(const struct lu_env *env,
7878 struct lu_device *d)
7880 struct osd_device *o = osd_dev(d);
7884 /* XXX: make osd top device in order to release reference */
7885 d->ld_site->ls_top_dev = d;
7886 lu_site_purge(env, d->ld_site, -1);
7887 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7888 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7889 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7891 lu_site_fini(&o->od_site);
7892 dt_device_fini(&o->od_dt_dev);
7897 static int osd_process_config(const struct lu_env *env,
7898 struct lu_device *d, struct lustre_cfg *cfg)
7900 struct osd_device *o = osd_dev(d);
7906 switch (cfg->lcfg_command) {
7908 rc = osd_mount(env, o, cfg);
7912 * For the case LCFG_PRE_CLEANUP is not called in advance,
7913 * that may happend if hit failure during mount process.
7915 osd_index_backup(env, o, false);
7916 lu_dev_del_linkage(d->ld_site, d);
7917 rc = osd_shutdown(env, o);
7920 LASSERT(&o->od_dt_dev);
7921 count = class_modify_config(cfg, PARAM_OSD,
7922 &o->od_dt_dev.dd_kobj);
7924 count = class_modify_config(cfg, PARAM_OST,
7925 &o->od_dt_dev.dd_kobj);
7926 rc = count > 0 ? 0 : count;
7928 case LCFG_PRE_CLEANUP:
7930 osd_index_backup(env, o,
7931 o->od_index_backup_policy != LIBP_NONE);
7941 static int osd_recovery_complete(const struct lu_env *env,
7942 struct lu_device *d)
7944 struct osd_device *osd = osd_dev(d);
7949 if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
7953 * start qsd instance on recovery completion, this notifies the quota
7954 * slave code that we are about to process new requests now
7956 rc = qsd_start(env, osd->od_quota_slave_dt);
7957 if (rc == 0 && osd->od_quota_slave_md != NULL)
7958 rc = qsd_start(env, osd->od_quota_slave_md);
7964 * we use exports to track all osd users
7966 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7967 struct obd_device *obd, struct obd_uuid *cluuid,
7968 struct obd_connect_data *data, void *localdata)
7970 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7971 struct lustre_handle conn;
7976 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7978 rc = class_connect(&conn, obd, cluuid);
7982 *exp = class_conn2export(&conn);
7984 spin_lock(&osd->od_osfs_lock);
7986 spin_unlock(&osd->od_osfs_lock);
7992 * once last export (we don't count self-export) disappeared
7993 * osd can be released
7995 static int osd_obd_disconnect(struct obd_export *exp)
7997 struct obd_device *obd = exp->exp_obd;
7998 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7999 int rc, release = 0;
8003 /* Only disconnect the underlying layers on the final disconnect. */
8004 spin_lock(&osd->od_osfs_lock);
8006 if (osd->od_connects == 0)
8008 spin_unlock(&osd->od_osfs_lock);
8010 rc = class_disconnect(exp); /* bz 9811 */
8012 if (rc == 0 && release)
8013 class_manual_cleanup(obd);
8017 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8018 struct lu_device *dev)
8020 struct osd_device *osd = osd_dev(dev);
8021 struct lr_server_data *lsd =
8022 &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8027 if (osd->od_quota_slave_md != NULL) {
8028 /* set up quota slave objects for inode */
8029 result = qsd_prepare(env, osd->od_quota_slave_md);
8034 if (osd->od_quota_slave_dt != NULL) {
8035 /* set up quota slave objects for block */
8036 result = qsd_prepare(env, osd->od_quota_slave_dt);
8042 if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8043 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8044 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8045 osd->od_index_in_idif = 1;
8047 osd->od_index_in_idif = 0;
8048 result = osd_register_proc_index_in_idif(osd);
8053 osd->od_index_in_idif = 1;
8057 result = osd_fid_init(env, osd);
8062 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
8063 struct lu_fid *fid, struct md_op_data *op_data)
8065 struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
8067 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8070 static const struct lu_object_operations osd_lu_obj_ops = {
8071 .loo_object_init = osd_object_init,
8072 .loo_object_delete = osd_object_delete,
8073 .loo_object_release = osd_object_release,
8074 .loo_object_free = osd_object_free,
8075 .loo_object_print = osd_object_print,
8076 .loo_object_invariant = osd_object_invariant
8079 const struct lu_device_operations osd_lu_ops = {
8080 .ldo_object_alloc = osd_object_alloc,
8081 .ldo_process_config = osd_process_config,
8082 .ldo_recovery_complete = osd_recovery_complete,
8083 .ldo_prepare = osd_prepare,
8086 static const struct lu_device_type_operations osd_device_type_ops = {
8087 .ldto_init = osd_type_init,
8088 .ldto_fini = osd_type_fini,
8090 .ldto_start = osd_type_start,
8091 .ldto_stop = osd_type_stop,
8093 .ldto_device_alloc = osd_device_alloc,
8094 .ldto_device_free = osd_device_free,
8096 .ldto_device_init = osd_device_init,
8097 .ldto_device_fini = osd_device_fini
8100 static struct lu_device_type osd_device_type = {
8101 .ldt_tags = LU_DEVICE_DT,
8102 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
8103 .ldt_ops = &osd_device_type_ops,
8104 .ldt_ctx_tags = LCT_LOCAL,
8107 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8109 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8110 struct super_block *sb = osd_sb(osd);
8112 return (osd->od_mnt == NULL || sb->s_flags & SB_RDONLY);
8116 * lprocfs legacy support.
8118 static const struct obd_ops osd_obd_device_ops = {
8119 .o_owner = THIS_MODULE,
8120 .o_connect = osd_obd_connect,
8121 .o_disconnect = osd_obd_disconnect,
8122 .o_fid_alloc = osd_fid_alloc,
8123 .o_health_check = osd_health_check,
8126 static ssize_t track_declares_assert_show(struct kobject *kobj,
8127 struct attribute *attr,
8130 return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8133 static ssize_t track_declares_assert_store(struct kobject *kobj,
8134 struct attribute *attr,
8135 const char *buffer, size_t count)
8137 bool track_declares_assert;
8140 rc = kstrtobool(buffer, &track_declares_assert);
8144 ldiskfs_track_declares_assert = track_declares_assert;
8148 LUSTRE_RW_ATTR(track_declares_assert);
8150 static int __init osd_init(void)
8152 struct kobject *kobj;
8155 BUILD_BUG_ON(BH_DXLock >=
8156 sizeof(((struct buffer_head *)0)->b_state) * 8);
8157 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8158 /* please, try to keep osd_thread_info smaller than a page */
8159 BUILD_BUG_ON(sizeof(struct osd_thread_info) > PAGE_SIZE);
8164 rc = lu_kmem_init(ldiskfs_caches);
8168 #ifdef CONFIG_KALLSYMS
8169 priv_security_file_alloc =
8170 (void *)kallsyms_lookup_name("security_file_alloc");
8173 rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL,
8174 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8176 lu_kmem_fini(ldiskfs_caches);
8180 kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8182 rc = sysfs_create_file(kobj,
8183 &lustre_attr_track_declares_assert.attr);
8186 CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8193 static void __exit osd_exit(void)
8195 struct kobject *kobj;
8197 kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8199 sysfs_remove_file(kobj,
8200 &lustre_attr_track_declares_assert.attr);
8203 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8204 lu_kmem_fini(ldiskfs_caches);
8207 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8208 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8209 MODULE_VERSION(LUSTRE_VERSION_STRING);
8210 MODULE_LICENSE("GPL");
8212 module_init(osd_init);
8213 module_exit(osd_exit);