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(idc, sizeof(*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(oti->oti_ins_cache,
219 oti->oti_ins_cache_used * sizeof(*idc));
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 set_file_inode(filp, 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 (fid_is_otable_it(&l->lo_header->loh_fid)) {
1404 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1405 l->lo_header->loh_attr |= LOHA_EXISTS;
1409 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1410 obj->oo_dt.do_body_ops = &osd_body_ops_new;
1411 if (result == 0 && obj->oo_inode != NULL) {
1412 struct osd_thread_info *oti = osd_oti_get(env);
1413 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1415 osd_object_init0(obj);
1416 if (unlikely(obj->oo_header))
1419 result = osd_get_lma(oti, obj->oo_inode,
1420 &oti->oti_obj_dentry, loa);
1423 * Convert LMAI flags to lustre LMA flags
1424 * and cache it to oo_lma_flags
1427 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1428 } else if (result == -ENODATA) {
1433 LINVRNT(osd_invariant(obj));
1438 * The first part of oxe_buf is xattr name, and is '\0' terminated.
1439 * The left part is for value, binary mode.
1441 struct osd_xattr_entry {
1442 struct list_head oxe_list;
1446 struct rcu_head oxe_rcu;
1450 static int osd_oxc_get(struct osd_object *obj, const char *name,
1453 struct osd_xattr_entry *tmp;
1454 struct osd_xattr_entry *oxe = NULL;
1455 size_t namelen = strlen(name);
1461 list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1462 if (namelen == tmp->oxe_namelen &&
1463 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1470 GOTO(out, rc = -ENOENT);
1472 if (!oxe->oxe_exist)
1473 GOTO(out, rc = -ENODATA);
1476 rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1479 if (buf->lb_buf == NULL)
1482 if (buf->lb_len < rc)
1483 GOTO(out, rc = -ERANGE);
1485 memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1493 static void osd_oxc_free(struct rcu_head *head)
1495 struct osd_xattr_entry *oxe;
1497 oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1498 OBD_FREE(oxe, oxe->oxe_len);
1501 static void osd_oxc_add(struct osd_object *obj, const char *name,
1502 const char *buf, int buflen)
1504 struct osd_xattr_entry *oxe;
1505 struct osd_xattr_entry *old = NULL;
1506 struct osd_xattr_entry *tmp;
1507 size_t namelen = strlen(name);
1508 size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1510 OBD_ALLOC(oxe, len);
1514 INIT_LIST_HEAD(&oxe->oxe_list);
1516 oxe->oxe_namelen = namelen;
1517 memcpy(oxe->oxe_buf, name, namelen);
1519 LASSERT(buf != NULL);
1520 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1521 oxe->oxe_exist = true;
1523 oxe->oxe_exist = false;
1526 /* this should be rarely called, just remove old and add new */
1527 spin_lock(&obj->oo_guard);
1528 list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1529 if (namelen == tmp->oxe_namelen &&
1530 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1536 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1537 call_rcu(&old->oxe_rcu, osd_oxc_free);
1539 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1541 spin_unlock(&obj->oo_guard);
1544 static void osd_oxc_del(struct osd_object *obj, const char *name)
1546 struct osd_xattr_entry *oxe;
1547 size_t namelen = strlen(name);
1549 spin_lock(&obj->oo_guard);
1550 list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1551 if (namelen == oxe->oxe_namelen &&
1552 strncmp(name, oxe->oxe_buf, namelen) == 0) {
1553 list_del_rcu(&oxe->oxe_list);
1554 call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1558 spin_unlock(&obj->oo_guard);
1561 static void osd_oxc_fini(struct osd_object *obj)
1563 struct osd_xattr_entry *oxe, *next;
1565 list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1566 list_del(&oxe->oxe_list);
1567 OBD_FREE(oxe, oxe->oxe_len);
1572 * Concurrency: no concurrent access is possible that late in object
1575 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1577 struct osd_object *obj = osd_obj(l);
1578 struct lu_object_header *h = obj->oo_header;
1580 LINVRNT(osd_invariant(obj));
1583 dt_object_fini(&obj->oo_dt);
1584 if (obj->oo_hl_head != NULL)
1585 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1588 lu_object_header_fini(h);
1594 * Concurrency: no concurrent access is possible that late in object
1597 static void osd_index_fini(struct osd_object *o)
1599 struct iam_container *bag;
1601 if (o->oo_dir != NULL) {
1602 bag = &o->oo_dir->od_container;
1603 if (o->oo_inode != NULL) {
1604 if (bag->ic_object == o->oo_inode)
1605 iam_container_fini(bag);
1607 OBD_FREE_PTR(o->oo_dir);
1613 * Concurrency: no concurrent access is possible that late in object
1614 * life-cycle (for all existing callers, that is. New callers have to provide
1615 * their own locking.)
1617 static int osd_inode_unlinked(const struct inode *inode)
1619 return inode->i_nlink == 0;
1623 OSD_TXN_OI_DELETE_CREDITS = 20,
1624 OSD_TXN_INODE_DELETE_CREDITS = 20
1631 #if OSD_THANDLE_STATS
1633 * Set time when the handle is allocated
1635 static void osd_th_alloced(struct osd_thandle *oth)
1637 oth->oth_alloced = ktime_get();
1641 * Set time when the handle started
1643 static void osd_th_started(struct osd_thandle *oth)
1645 oth->oth_started = ktime_get();
1649 * Check whether the we deal with this handle for too long.
1651 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1652 ktime_t alloced, ktime_t started,
1655 ktime_t now = ktime_get();
1657 LASSERT(dev != NULL);
1659 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1660 ktime_us_delta(started, alloced));
1661 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1662 ktime_us_delta(closed, started));
1663 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1664 ktime_us_delta(now, closed));
1666 if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1667 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1668 oth, now, alloced, started, closed);
1669 libcfs_debug_dumpstack(NULL);
1673 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
1675 ktime_t __closed = ktime_get(); \
1676 ktime_t __alloced = oth->oth_alloced; \
1677 ktime_t __started = oth->oth_started; \
1680 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
1683 #else /* OSD_THANDLE_STATS */
1685 #define osd_th_alloced(h) do {} while(0)
1686 #define osd_th_started(h) do {} while(0)
1687 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
1689 #endif /* OSD_THANDLE_STATS */
1692 * Concurrency: doesn't access mutable data.
1694 static int osd_param_is_not_sane(const struct osd_device *dev,
1695 const struct thandle *th)
1697 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1699 return oh->ot_credits > osd_transaction_size(dev);
1703 * Concurrency: shouldn't matter.
1705 static void osd_trans_commit_cb(struct super_block *sb,
1706 struct ldiskfs_journal_cb_entry *jcb, int error)
1708 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1709 struct thandle *th = &oh->ot_super;
1710 struct lu_device *lud = &th->th_dev->dd_lu_dev;
1711 struct dt_txn_commit_cb *dcb, *tmp;
1713 LASSERT(oh->ot_handle == NULL);
1716 CERROR("transaction @0x%p commit error: %d\n", th, error);
1718 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DELAY_TRANS, 40);
1719 /* call per-transaction callbacks if any */
1720 list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1722 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1723 "commit callback entry: magic=%x name='%s'\n",
1724 dcb->dcb_magic, dcb->dcb_name);
1725 list_del_init(&dcb->dcb_linkage);
1726 dcb->dcb_func(NULL, th, dcb, error);
1729 lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1736 #ifndef HAVE_SB_START_WRITE
1737 # define sb_start_write(sb) do {} while (0)
1738 # define sb_end_write(sb) do {} while (0)
1741 static struct thandle *osd_trans_create(const struct lu_env *env,
1742 struct dt_device *d)
1744 struct osd_thread_info *oti = osd_oti_get(env);
1745 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1746 struct osd_thandle *oh;
1752 CERROR("%s: someone try to start transaction under "
1753 "readonly mode, should be disabled.\n",
1754 osd_name(osd_dt_dev(d)));
1756 RETURN(ERR_PTR(-EROFS));
1759 /* on pending IO in this thread should left from prev. request */
1760 LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1762 sb_start_write(osd_sb(osd_dt_dev(d)));
1764 OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1766 sb_end_write(osd_sb(osd_dt_dev(d)));
1767 RETURN(ERR_PTR(-ENOMEM));
1770 oh->ot_quota_trans = &oti->oti_quota_trans;
1771 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1776 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1777 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1778 INIT_LIST_HEAD(&oh->ot_trunc_locks);
1781 memset(oti->oti_declare_ops, 0,
1782 sizeof(oti->oti_declare_ops));
1783 memset(oti->oti_declare_ops_cred, 0,
1784 sizeof(oti->oti_declare_ops_cred));
1785 memset(oti->oti_declare_ops_used, 0,
1786 sizeof(oti->oti_declare_ops_used));
1788 oti->oti_ins_cache_depth++;
1793 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1795 struct osd_thread_info *oti = osd_oti_get(env);
1796 struct osd_thandle *oh;
1798 oh = container_of0(th, struct osd_thandle, ot_super);
1799 LASSERT(oh != NULL);
1801 CWARN(" create: %u/%u/%u, destroy: %u/%u/%u\n",
1802 oti->oti_declare_ops[OSD_OT_CREATE],
1803 oti->oti_declare_ops_cred[OSD_OT_CREATE],
1804 oti->oti_declare_ops_used[OSD_OT_CREATE],
1805 oti->oti_declare_ops[OSD_OT_DESTROY],
1806 oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1807 oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1808 CWARN(" attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1809 oti->oti_declare_ops[OSD_OT_ATTR_SET],
1810 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1811 oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1812 oti->oti_declare_ops[OSD_OT_XATTR_SET],
1813 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1814 oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1815 CWARN(" write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1816 oti->oti_declare_ops[OSD_OT_WRITE],
1817 oti->oti_declare_ops_cred[OSD_OT_WRITE],
1818 oti->oti_declare_ops_used[OSD_OT_WRITE],
1819 oti->oti_declare_ops[OSD_OT_PUNCH],
1820 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1821 oti->oti_declare_ops_used[OSD_OT_PUNCH],
1822 oti->oti_declare_ops[OSD_OT_QUOTA],
1823 oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1824 oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1825 CWARN(" insert: %u/%u/%u, delete: %u/%u/%u\n",
1826 oti->oti_declare_ops[OSD_OT_INSERT],
1827 oti->oti_declare_ops_cred[OSD_OT_INSERT],
1828 oti->oti_declare_ops_used[OSD_OT_INSERT],
1829 oti->oti_declare_ops[OSD_OT_DELETE],
1830 oti->oti_declare_ops_cred[OSD_OT_DELETE],
1831 oti->oti_declare_ops_used[OSD_OT_DELETE]);
1832 CWARN(" ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1833 oti->oti_declare_ops[OSD_OT_REF_ADD],
1834 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1835 oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1836 oti->oti_declare_ops[OSD_OT_REF_DEL],
1837 oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1838 oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1842 * Concurrency: shouldn't matter.
1844 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1847 struct osd_thread_info *oti = osd_oti_get(env);
1848 struct osd_device *dev = osd_dt_dev(d);
1850 struct osd_thandle *oh;
1855 LASSERT(current->journal_info == NULL);
1857 oh = container_of0(th, struct osd_thandle, ot_super);
1858 LASSERT(oh != NULL);
1859 LASSERT(oh->ot_handle == NULL);
1861 rc = dt_txn_hook_start(env, d, th);
1865 if (unlikely(osd_param_is_not_sane(dev, th))) {
1866 static unsigned long last_printed;
1867 static int last_credits;
1870 * don't make noise on a tiny testing systems
1871 * actual credits misuse will be caught anyway
1873 if (last_credits != oh->ot_credits &&
1874 time_after(jiffies, last_printed +
1875 cfs_time_seconds(60)) &&
1876 osd_transaction_size(dev) > 512) {
1877 CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1878 oh->ot_credits, osd_transaction_size(dev));
1879 osd_trans_dump_creds(env, th);
1880 libcfs_debug_dumpstack(NULL);
1881 last_credits = oh->ot_credits;
1882 last_printed = jiffies;
1885 * XXX Limit the credits to 'max_transaction_buffers', and
1886 * let the underlying filesystem to catch the error if
1887 * we really need so many credits.
1889 * This should be removed when we can calculate the
1890 * credits precisely.
1892 oh->ot_credits = osd_transaction_size(dev);
1893 } else if (ldiskfs_track_declares_assert != 0) {
1895 * reserve few credits to prevent an assertion in JBD
1896 * our debugging mechanism will be able to detected
1897 * overuse. this can help to debug single-update
1900 oh->ot_credits += 10;
1901 if (unlikely(osd_param_is_not_sane(dev, th)))
1902 oh->ot_credits = osd_transaction_size(dev);
1905 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1906 GOTO(out, rc = -EIO);
1909 * XXX temporary stuff. Some abstraction layer should
1912 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1916 LASSERT(oti->oti_txns == 0);
1918 lu_device_get(&d->dd_lu_dev);
1919 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1930 static int osd_seq_exists(const struct lu_env *env,
1931 struct osd_device *osd, u64 seq)
1933 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1934 struct seq_server_site *ss = osd_seq_site(osd);
1939 LASSERT(ss != NULL);
1940 LASSERT(ss->ss_server_fld != NULL);
1942 rc = osd_fld_lookup(env, osd, seq, range);
1945 CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1946 osd_name(osd), seq, rc);
1950 RETURN(ss->ss_node_id == range->lsr_index);
1953 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1955 struct dt_txn_commit_cb *dcb;
1956 struct dt_txn_commit_cb *tmp;
1958 /* call per-transaction stop callbacks if any */
1959 list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1961 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1962 "commit callback entry: magic=%x name='%s'\n",
1963 dcb->dcb_magic, dcb->dcb_name);
1964 list_del_init(&dcb->dcb_linkage);
1965 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1970 * Concurrency: shouldn't matter.
1972 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1975 struct osd_thread_info *oti = osd_oti_get(env);
1976 struct osd_thandle *oh;
1977 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1978 struct osd_device *osd = osd_dt_dev(th->th_dev);
1979 struct qsd_instance *qsd = osd_def_qsd(osd);
1980 struct lquota_trans *qtrans;
1981 LIST_HEAD(truncates);
1982 int rc = 0, remove_agents = 0;
1986 oh = container_of0(th, struct osd_thandle, ot_super);
1988 remove_agents = oh->ot_remove_agents;
1990 qtrans = oh->ot_quota_trans;
1991 oh->ot_quota_trans = NULL;
1993 /* move locks to local list, stop tx, execute truncates */
1994 list_splice(&oh->ot_trunc_locks, &truncates);
1996 if (oh->ot_handle != NULL) {
1999 handle_t *hdl = oh->ot_handle;
2002 * add commit callback
2003 * notice we don't do this in osd_trans_start()
2004 * as underlying transaction can change during truncate
2006 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
2009 LASSERT(oti->oti_txns == 1);
2012 rc = dt_txn_hook_stop(env, th);
2014 CERROR("%s: failed in transaction hook: rc = %d\n",
2017 osd_trans_stop_cb(oh, rc);
2018 /* hook functions might modify th_sync */
2019 hdl->h_sync = th->th_sync;
2021 oh->ot_handle = NULL;
2022 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2024 CERROR("%s: failed to stop transaction: rc = %d\n",
2025 osd_name(osd), rc2);
2029 osd_process_truncates(&truncates);
2031 osd_trans_stop_cb(oh, th->th_result);
2035 osd_trunc_unlock_all(&truncates);
2037 /* inform the quota slave device that the transaction is stopping */
2038 qsd_op_end(env, qsd, qtrans);
2041 * as we want IO to journal and data IO be concurrent, we don't block
2042 * awaiting data IO completion in osd_do_bio(), instead we wait here
2043 * once transaction is submitted to the journal. all reqular requests
2044 * don't do direct IO (except read/write), thus this wait_event becomes
2047 * IMPORTANT: we have to wait till any IO submited by the thread is
2048 * completed otherwise iobuf may be corrupted by different request
2050 wait_event(iobuf->dr_wait,
2051 atomic_read(&iobuf->dr_numreqs) == 0);
2052 osd_fini_iobuf(osd, iobuf);
2054 rc = iobuf->dr_error;
2056 if (unlikely(remove_agents != 0))
2057 osd_process_scheduled_agent_removals(env, osd);
2059 oti->oti_ins_cache_depth--;
2060 /* reset OI cache for safety */
2061 if (oti->oti_ins_cache_depth == 0)
2062 oti->oti_ins_cache_used = 0;
2064 sb_end_write(osd_sb(osd));
2069 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2071 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
2074 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2075 LASSERT(&dcb->dcb_func != NULL);
2076 if (dcb->dcb_flags & DCB_TRANS_STOP)
2077 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2079 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2085 * Called just before object is freed. Releases all resources except for
2086 * object itself (that is released by osd_object_free()).
2088 * Concurrency: no concurrent access is possible that late in object
2091 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2093 struct osd_object *obj = osd_obj(l);
2094 struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2095 struct inode *inode = obj->oo_inode;
2100 LINVRNT(osd_invariant(obj));
2103 * If object is unlinked remove fid->ino mapping from object index.
2106 osd_index_fini(obj);
2111 uid = i_uid_read(inode);
2112 gid = i_gid_read(inode);
2113 projid = i_projid_read(inode);
2115 obj->oo_inode = NULL;
2118 /* do not rebalance quota if the caller needs to release memory
2119 * otherwise qsd_refresh_usage() may went into a new ldiskfs
2120 * transaction and risk to deadlock - LU-12178 */
2121 if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2124 if (!obj->oo_header && qsd) {
2125 struct osd_thread_info *info = osd_oti_get(env);
2126 struct lquota_id_info *qi = &info->oti_qi;
2128 /* Release granted quota to master if necessary */
2129 qi->lqi_id.qid_uid = uid;
2130 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2132 qi->lqi_id.qid_uid = gid;
2133 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2135 qi->lqi_id.qid_uid = projid;
2136 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2141 * Concurrency: ->loo_object_release() is called under site spin-lock.
2143 static void osd_object_release(const struct lu_env *env,
2144 struct lu_object *l)
2146 struct osd_object *o = osd_obj(l);
2149 * nobody should be releasing a non-destroyed object with nlink=0
2150 * the API allows this, but ldiskfs doesn't like and then report
2151 * this inode as deleted
2153 LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2154 o->oo_inode->i_nlink == 0));
2158 * Concurrency: shouldn't matter.
2160 static int osd_object_print(const struct lu_env *env, void *cookie,
2161 lu_printer_t p, const struct lu_object *l)
2163 struct osd_object *o = osd_obj(l);
2164 struct iam_descr *d;
2166 if (o->oo_dir != NULL)
2167 d = o->oo_dir->od_container.ic_descr;
2170 return (*p)(env, cookie,
2171 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2173 o->oo_inode ? o->oo_inode->i_ino : 0UL,
2174 o->oo_inode ? o->oo_inode->i_generation : 0,
2175 d ? d->id_ops->id_name : "plain");
2179 * Concurrency: shouldn't matter.
2181 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2182 struct obd_statfs *sfs, struct obd_statfs_info *info)
2184 struct osd_device *osd = osd_dt_dev(d);
2185 struct super_block *sb = osd_sb(osd);
2186 struct kstatfs *ksfs;
2190 if (unlikely(osd->od_mnt == NULL))
2191 return -EINPROGRESS;
2193 /* osd_lproc.c call this without env, allocate ksfs for that case */
2194 if (unlikely(env == NULL)) {
2195 OBD_ALLOC_PTR(ksfs);
2199 ksfs = &osd_oti_get(env)->oti_ksfs;
2202 result = sb->s_op->statfs(sb->s_root, ksfs);
2206 statfs_pack(sfs, ksfs);
2207 if (unlikely(sb->s_flags & SB_RDONLY))
2208 sfs->os_state |= OS_STATE_READONLY;
2210 sfs->os_state |= osd->od_nonrotational ? OS_STATE_NONROT : 0;
2212 if (ldiskfs_has_feature_extents(sb))
2213 sfs->os_maxbytes = sb->s_maxbytes;
2215 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2218 * Reserve some space so to avoid fragmenting the filesystem too much.
2219 * Fragmentation not only impacts performance, but can also increase
2220 * metadata overhead significantly, causing grant calculation to be
2223 * Reserve 0.78% of total space, at least 8MB for small filesystems.
2225 BUILD_BUG_ON(OSD_STATFS_RESERVED <= LDISKFS_MAX_BLOCK_SIZE);
2226 reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2227 if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2228 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2230 sfs->os_blocks -= reserved;
2231 sfs->os_bfree -= min(reserved, sfs->os_bfree);
2232 sfs->os_bavail -= min(reserved, sfs->os_bavail);
2235 if (unlikely(env == NULL))
2241 * Estimate space needed for file creations. We assume the largest filename
2242 * which is 2^64 - 1, hence a filename of 20 chars.
2243 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2245 #ifdef __LDISKFS_DIR_REC_LEN
2246 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2248 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2252 * Concurrency: doesn't access mutable data.
2254 static void osd_conf_get(const struct lu_env *env,
2255 const struct dt_device *dev,
2256 struct dt_device_param *param)
2258 struct osd_device *d = osd_dt_dev(dev);
2259 struct super_block *sb = osd_sb(d);
2260 struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2265 * XXX should be taken from not-yet-existing fs abstraction layer.
2267 param->ddp_max_name_len = LDISKFS_NAME_LEN;
2268 param->ddp_max_nlink = LDISKFS_LINK_MAX;
2269 param->ddp_symlink_max = sb->s_blocksize;
2270 param->ddp_mount_type = LDD_MT_LDISKFS;
2271 if (ldiskfs_has_feature_extents(sb))
2272 param->ddp_maxbytes = sb->s_maxbytes;
2274 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2276 * inode are statically allocated, so per-inode space consumption
2277 * is the space consumed by the directory entry
2279 param->ddp_inodespace = PER_OBJ_USAGE;
2281 * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2282 * is 128MB) which is unlikely to be hit in real life. Report a smaller
2283 * maximum length to not under-count the actual number of extents
2284 * needed for writing a file if there are sub-optimal block allocations.
2286 param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2287 /* worst-case extent insertion metadata overhead */
2288 param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2289 param->ddp_mntopts = 0;
2290 if (test_opt(sb, XATTR_USER))
2291 param->ddp_mntopts |= MNTOPT_USERXATTR;
2292 if (test_opt(sb, POSIX_ACL))
2293 param->ddp_mntopts |= MNTOPT_ACL;
2296 * LOD might calculate the max stripe count based on max_ea_size,
2297 * so we need take account in the overhead as well,
2298 * xattr_header + magic + xattr_entry_head
2300 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2301 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2303 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2304 if (ldiskfs_has_feature_ea_inode(sb))
2305 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2309 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2311 if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE)
2312 param->ddp_max_ea_size = OBD_MAX_EA_SIZE;
2315 * Preferred RPC size for efficient disk IO. 4MB shows good
2316 * all-around performance for ldiskfs, but use bigalloc chunk size
2317 * by default if larger.
2319 #if defined(LDISKFS_CLUSTER_SIZE)
2320 if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2321 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2324 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2326 param->ddp_t10_cksum_type = 0;
2328 unsigned short interval = blk_integrity_interval(bi);
2329 name = blk_integrity_name(bi);
2337 if (strncmp(name, "T10-DIF-TYPE",
2338 sizeof("T10-DIF-TYPE") - 1) == 0) {
2339 /* also skip "1/3-" at end */
2340 const int type_off = sizeof("T10-DIF-TYPE.");
2341 char type_number = name[type_off - 2];
2343 if (interval != 512 && interval != 4096) {
2344 CERROR("%s: unsupported T10PI sector size %u\n",
2345 d->od_svname, interval);
2346 } else if (type_number != '1' && type_number != '3') {
2347 CERROR("%s: unsupported T10PI type %s\n",
2348 d->od_svname, name);
2349 } else if (strcmp(name + type_off, "CRC") == 0) {
2350 d->od_t10_type = type_number == '1' ?
2351 OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2352 param->ddp_t10_cksum_type = interval == 512 ?
2353 OBD_CKSUM_T10CRC512 :
2355 } else if (strcmp(name + type_off, "IP") == 0) {
2356 d->od_t10_type = type_number == '1' ?
2357 OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2358 param->ddp_t10_cksum_type = interval == 512 ?
2359 OBD_CKSUM_T10IP512 :
2362 CERROR("%s: unsupported checksum type of "
2364 d->od_svname, name);
2368 CERROR("%s: unsupported T10PI type '%s'",
2369 d->od_svname, name);
2374 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2376 return osd_sb(osd_dt_dev(d));
2380 * Concurrency: shouldn't matter.
2382 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2385 struct super_block *s = osd_sb(osd_dt_dev(d));
2388 down_read(&s->s_umount);
2389 rc = s->s_op->sync_fs(s, 1);
2390 up_read(&s->s_umount);
2392 CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2399 * Start commit for OSD device.
2401 * An implementation of dt_commit_async method for OSD device.
2402 * Asychronously starts underlayng fs sync and thereby a transaction
2405 * \param env environment
2406 * \param d dt device
2408 * \see dt_device_operations
2410 static int osd_commit_async(const struct lu_env *env,
2411 struct dt_device *d)
2413 struct super_block *s = osd_sb(osd_dt_dev(d));
2418 CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2419 down_read(&s->s_umount);
2420 rc = s->s_op->sync_fs(s, 0);
2421 up_read(&s->s_umount);
2426 /* Our own copy of the set readonly functions if present, or NU if not. */
2427 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2428 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2429 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2430 static int (*priv_security_file_alloc)(struct file *file);
2432 int osd_security_file_alloc(struct file *file)
2434 if (priv_security_file_alloc)
2435 return priv_security_file_alloc(file);
2440 * Concurrency: shouldn't matter.
2442 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2444 struct super_block *sb = osd_sb(osd_dt_dev(d));
2445 struct block_device *dev = sb->s_bdev;
2446 int rc = -EOPNOTSUPP;
2450 if (priv_dev_set_rdonly) {
2451 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2454 CERROR("*** setting %s read-only ***\n",
2455 osd_dt_dev(d)->od_svname);
2457 if (sb->s_op->freeze_fs) {
2458 rc = sb->s_op->freeze_fs(sb);
2463 if (jdev && (jdev != dev)) {
2464 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2466 priv_dev_set_rdonly(jdev);
2468 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2469 priv_dev_set_rdonly(dev);
2471 if (sb->s_op->unfreeze_fs)
2472 sb->s_op->unfreeze_fs(sb);
2477 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2478 osd_dt_dev(d)->od_svname, (long)dev, rc);
2484 * Note: we do not count into QUOTA here.
2485 * If we mount with --data_journal we may need more.
2487 const int osd_dto_credits_noquota[DTO_NR] = {
2490 * INDEX_EXTRA_TRANS_BLOCKS(8) +
2491 * SINGLEDATA_TRANS_BLOCKS(8)
2492 * XXX Note: maybe iam need more, since iam have more level than
2495 [DTO_INDEX_INSERT] = 16,
2498 * just modify a single entry, probably merge few within a block
2500 [DTO_INDEX_DELETE] = 1,
2504 [DTO_INDEX_UPDATE] = 16,
2506 * 4(inode, inode bits, groups, GDT)
2507 * notice: OI updates are counted separately with DTO_INDEX_INSERT
2509 [DTO_OBJECT_CREATE] = 4,
2511 * 4(inode, inode bits, groups, GDT)
2512 * notice: OI updates are counted separately with DTO_INDEX_DELETE
2514 [DTO_OBJECT_DELETE] = 4,
2516 * Attr set credits (inode)
2518 [DTO_ATTR_SET_BASE] = 1,
2520 * Xattr set. The same as xattr of EXT3.
2521 * DATA_TRANS_BLOCKS(14)
2522 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2523 * are also counted in. Do not know why?
2525 [DTO_XATTR_SET] = 14,
2527 * credits for inode change during write.
2529 [DTO_WRITE_BASE] = 3,
2531 * credits for single block write.
2533 [DTO_WRITE_BLOCK] = 14,
2535 * Attr set credits for chown.
2536 * This is extra credits for setattr, and it is null without quota
2538 [DTO_ATTR_SET_CHOWN] = 0
2541 static const struct dt_device_operations osd_dt_ops = {
2542 .dt_root_get = osd_root_get,
2543 .dt_statfs = osd_statfs,
2544 .dt_trans_create = osd_trans_create,
2545 .dt_trans_start = osd_trans_start,
2546 .dt_trans_stop = osd_trans_stop,
2547 .dt_trans_cb_add = osd_trans_cb_add,
2548 .dt_conf_get = osd_conf_get,
2549 .dt_mnt_sb_get = osd_mnt_sb_get,
2550 .dt_sync = osd_sync,
2552 .dt_commit_async = osd_commit_async,
2555 static void osd_read_lock(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(obj->oo_owner != env);
2564 down_read_nested(&obj->oo_sem, role);
2566 LASSERT(obj->oo_owner == NULL);
2570 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2573 struct osd_object *obj = osd_dt_obj(dt);
2574 struct osd_thread_info *oti = osd_oti_get(env);
2576 LINVRNT(osd_invariant(obj));
2578 LASSERT(obj->oo_owner != env);
2579 down_write_nested(&obj->oo_sem, role);
2581 LASSERT(obj->oo_owner == NULL);
2582 obj->oo_owner = env;
2586 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2588 struct osd_object *obj = osd_dt_obj(dt);
2589 struct osd_thread_info *oti = osd_oti_get(env);
2591 LINVRNT(osd_invariant(obj));
2593 LASSERT(oti->oti_r_locks > 0);
2595 up_read(&obj->oo_sem);
2598 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2600 struct osd_object *obj = osd_dt_obj(dt);
2601 struct osd_thread_info *oti = osd_oti_get(env);
2603 LINVRNT(osd_invariant(obj));
2605 LASSERT(obj->oo_owner == env);
2606 LASSERT(oti->oti_w_locks > 0);
2608 obj->oo_owner = NULL;
2609 up_write(&obj->oo_sem);
2612 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2614 struct osd_object *obj = osd_dt_obj(dt);
2616 LINVRNT(osd_invariant(obj));
2618 return obj->oo_owner == env;
2621 static void osd_inode_getattr(const struct lu_env *env,
2622 struct inode *inode, struct lu_attr *attr)
2624 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2625 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2626 LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2627 LA_BLKSIZE | LA_TYPE;
2629 attr->la_atime = inode->i_atime.tv_sec;
2630 attr->la_mtime = inode->i_mtime.tv_sec;
2631 attr->la_ctime = inode->i_ctime.tv_sec;
2632 attr->la_mode = inode->i_mode;
2633 attr->la_size = i_size_read(inode);
2634 attr->la_blocks = inode->i_blocks;
2635 attr->la_uid = i_uid_read(inode);
2636 attr->la_gid = i_gid_read(inode);
2637 attr->la_projid = i_projid_read(inode);
2638 attr->la_flags = ll_inode_to_ext_flags(inode->i_flags);
2639 attr->la_nlink = inode->i_nlink;
2640 attr->la_rdev = inode->i_rdev;
2641 attr->la_blksize = 1 << inode->i_blkbits;
2642 attr->la_blkbits = inode->i_blkbits;
2644 * Ext4 did not transfer inherit flags from raw inode
2645 * to inode flags, and ext4 internally test raw inode
2646 * @i_flags directly. Instead of patching ext4, we do it here.
2648 if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2649 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2652 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2653 struct lu_attr *attr)
2655 struct osd_object *obj = osd_dt_obj(dt);
2657 if (unlikely(!dt_object_exists(dt)))
2659 if (unlikely(obj->oo_destroyed))
2662 LASSERT(!dt_object_remote(dt));
2663 LINVRNT(osd_invariant(obj));
2665 spin_lock(&obj->oo_guard);
2666 osd_inode_getattr(env, obj->oo_inode, attr);
2667 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2668 attr->la_valid |= LA_FLAGS;
2669 attr->la_flags |= LUSTRE_ORPHAN_FL;
2671 spin_unlock(&obj->oo_guard);
2676 static int osd_declare_attr_qid(const struct lu_env *env,
2677 struct osd_object *obj,
2678 struct osd_thandle *oh, long long bspace,
2679 qid_t old_id, qid_t new_id, bool enforce,
2680 unsigned int type, bool ignore_edquot)
2683 struct osd_thread_info *info = osd_oti_get(env);
2684 struct lquota_id_info *qi = &info->oti_qi;
2686 qi->lqi_type = type;
2687 /* inode accounting */
2688 qi->lqi_is_blk = false;
2690 /* one more inode for the new id ... */
2691 qi->lqi_id.qid_uid = new_id;
2693 /* Reserve credits for the new id */
2694 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2695 if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2700 /* and one less inode for the current id */
2701 qi->lqi_id.qid_uid = old_id;
2703 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2704 if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2709 /* block accounting */
2710 qi->lqi_is_blk = true;
2712 /* more blocks for the new id ... */
2713 qi->lqi_id.qid_uid = new_id;
2714 qi->lqi_space = bspace;
2716 * Credits for the new uid has been reserved, re-use "obj"
2717 * to save credit reservation.
2719 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2720 if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2725 /* and finally less blocks for the current uid */
2726 qi->lqi_id.qid_uid = old_id;
2727 qi->lqi_space = -bspace;
2728 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2729 if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2735 static int osd_declare_attr_set(const struct lu_env *env,
2736 struct dt_object *dt,
2737 const struct lu_attr *attr,
2738 struct thandle *handle)
2740 struct osd_thandle *oh;
2741 struct osd_object *obj;
2750 LASSERT(dt != NULL);
2751 LASSERT(handle != NULL);
2753 obj = osd_dt_obj(dt);
2754 LASSERT(osd_invariant(obj));
2756 oh = container_of0(handle, struct osd_thandle, ot_super);
2757 LASSERT(oh->ot_handle == NULL);
2759 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2760 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2762 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2763 osd_dto_credits_noquota[DTO_XATTR_SET]);
2765 if (attr == NULL || obj->oo_inode == NULL)
2768 bspace = obj->oo_inode->i_blocks << 9;
2769 bspace = toqb(bspace);
2772 * Changing ownership is always preformed by super user, it should not
2773 * fail with EDQUOT unless required explicitly.
2775 * We still need to call the osd_declare_qid() to calculate the journal
2776 * credits for updating quota accounting files and to trigger quota
2777 * space adjustment once the operation is completed.
2779 if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2780 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2783 CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2784 "(the quota space is %lld)\n",
2785 obj->oo_inode->i_sb->s_id, attr->la_uid,
2786 attr->la_gid, bspace);
2789 uid = i_uid_read(obj->oo_inode);
2790 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2791 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2792 attr->la_uid, enforce, USRQUOTA,
2797 gid = i_gid_read(obj->oo_inode);
2798 CDEBUG(D_QUOTA, "declare uid %d -> %d gid %d -> %d\n", uid,
2799 attr->la_uid, gid, attr->la_gid);
2800 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2801 rc = osd_declare_attr_qid(env, obj, oh, bspace, gid,
2802 attr->la_gid, enforce, GRPQUOTA,
2808 #ifdef HAVE_PROJECT_QUOTA
2809 if (attr->la_valid & LA_PROJID) {
2810 __u32 projid = i_projid_read(obj->oo_inode);
2812 enforce = (attr->la_valid & LA_PROJID) &&
2813 (attr->la_projid != projid);
2814 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2815 (qid_t)projid, (qid_t)attr->la_projid,
2816 enforce, PRJQUOTA, true);
2824 static int osd_inode_setattr(const struct lu_env *env,
2825 struct inode *inode, const struct lu_attr *attr)
2827 __u64 bits = attr->la_valid;
2829 /* Only allow set size for regular file */
2830 if (!S_ISREG(inode->i_mode))
2831 bits &= ~(LA_SIZE | LA_BLOCKS);
2836 if (bits & LA_ATIME)
2837 inode->i_atime = osd_inode_time(inode, attr->la_atime);
2838 if (bits & LA_CTIME)
2839 inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
2840 if (bits & LA_MTIME)
2841 inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
2842 if (bits & LA_SIZE) {
2843 spin_lock(&inode->i_lock);
2844 LDISKFS_I(inode)->i_disksize = attr->la_size;
2845 i_size_write(inode, attr->la_size);
2846 spin_unlock(&inode->i_lock);
2850 * OSD should not change "i_blocks" which is used by quota.
2851 * "i_blocks" should be changed by ldiskfs only.
2854 inode->i_mode = (inode->i_mode & S_IFMT) |
2855 (attr->la_mode & ~S_IFMT);
2857 i_uid_write(inode, attr->la_uid);
2859 i_gid_write(inode, attr->la_gid);
2860 if (bits & LA_PROJID)
2861 i_projid_write(inode, attr->la_projid);
2862 if (bits & LA_NLINK)
2863 set_nlink(inode, attr->la_nlink);
2865 inode->i_rdev = attr->la_rdev;
2867 if (bits & LA_FLAGS) {
2868 /* always keep S_NOCMTIME */
2869 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2872 * Ext4 did not transfer inherit flags from
2873 * @inode->i_flags to raw inode i_flags when writing
2874 * flags, we do it explictly here.
2876 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2877 LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2879 LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2884 #ifdef HAVE_PROJECT_QUOTA
2885 static int osd_transfer_project(struct inode *inode, __u32 projid)
2887 struct super_block *sb = inode->i_sb;
2888 struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2891 struct ldiskfs_iloc iloc;
2892 struct ldiskfs_inode *raw_inode;
2893 struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2895 if (!ldiskfs_has_feature_project(sb)) {
2896 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2897 == LDISKFS_DEF_PROJID);
2898 if (projid != LDISKFS_DEF_PROJID)
2904 if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2907 kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2908 if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2911 err = ldiskfs_get_inode_loc(inode, &iloc);
2915 raw_inode = ldiskfs_raw_inode(&iloc);
2916 if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2923 dquot_initialize(inode);
2924 transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2925 if (transfer_to[PRJQUOTA]) {
2926 err = __dquot_transfer(inode, transfer_to);
2927 dqput(transfer_to[PRJQUOTA]);
2936 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2940 if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2941 (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2945 "executing dquot_transfer inode %ld uid %d -> %d gid %d -> %d\n",
2946 inode->i_ino, i_uid_read(inode), attr->la_uid,
2947 i_gid_read(inode), attr->la_gid);
2949 dquot_initialize(inode);
2951 if (attr->la_valid & LA_UID)
2952 iattr.ia_valid |= ATTR_UID;
2953 if (attr->la_valid & LA_GID)
2954 iattr.ia_valid |= ATTR_GID;
2955 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2956 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2958 rc = dquot_transfer(inode, &iattr);
2960 CERROR("%s: quota transfer failed: rc = %d. Is quota "
2961 "enforcement enabled on the ldiskfs "
2962 "filesystem?\n", inode->i_sb->s_id, rc);
2967 /* Handle project id transfer here properly */
2968 if (attr->la_valid & LA_PROJID &&
2969 attr->la_projid != i_projid_read(inode)) {
2970 #ifdef HAVE_PROJECT_QUOTA
2971 rc = osd_transfer_project(inode, attr->la_projid);
2976 CERROR("%s: quota transfer failed: rc = %d. Is project "
2977 "enforcement enabled on the ldiskfs "
2978 "filesystem?\n", inode->i_sb->s_id, rc);
2985 static int osd_attr_set(const struct lu_env *env,
2986 struct dt_object *dt,
2987 const struct lu_attr *attr,
2988 struct thandle *handle)
2990 struct osd_object *obj = osd_dt_obj(dt);
2991 struct inode *inode;
2994 if (!dt_object_exists(dt))
2997 LASSERT(handle != NULL);
2998 LASSERT(!dt_object_remote(dt));
2999 LASSERT(osd_invariant(obj));
3001 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
3003 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
3004 !osd_obj2dev(obj)->od_is_ost) {
3005 struct osd_thread_info *oti = osd_oti_get(env);
3006 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
3007 struct lu_fid *fid1 = &oti->oti_fid;
3008 struct osd_inode_id *id = &oti->oti_id;
3009 struct iam_path_descr *ipd;
3010 struct iam_container *bag;
3011 struct osd_thandle *oh;
3014 fid_cpu_to_be(fid1, fid0);
3015 memset(id, 1, sizeof(*id));
3016 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
3017 fid0)->oi_dir.od_container;
3018 ipd = osd_idx_ipd_get(env, bag);
3019 if (unlikely(ipd == NULL))
3022 oh = container_of0(handle, struct osd_thandle, ot_super);
3023 rc = iam_update(oh->ot_handle, bag,
3024 (const struct iam_key *)fid1,
3025 (const struct iam_rec *)id, ipd);
3026 osd_ipd_put(env, bag, ipd);
3027 return(rc > 0 ? 0 : rc);
3030 inode = obj->oo_inode;
3032 rc = osd_quota_transfer(inode, attr);
3036 spin_lock(&obj->oo_guard);
3037 rc = osd_inode_setattr(env, inode, attr);
3038 spin_unlock(&obj->oo_guard);
3042 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3044 osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3046 if (!(attr->la_valid & LA_FLAGS))
3049 /* Let's check if there are extra flags need to be set into LMA */
3050 if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3051 struct osd_thread_info *info = osd_oti_get(env);
3052 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3054 LASSERT(!obj->oo_pfid_in_lma);
3056 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3057 &info->oti_ost_attrs);
3061 lma->lma_incompat |=
3062 lustre_to_lma_flags(attr->la_flags);
3063 lustre_lma_swab(lma);
3065 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3067 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3068 lma, sizeof(*lma), XATTR_REPLACE);
3070 struct osd_device *osd = osd_obj2dev(obj);
3072 CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3073 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3074 lma->lma_incompat, rc);
3077 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3079 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3086 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3087 struct osd_object *obj,
3088 const char *name, const int namelen)
3090 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3093 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3094 umode_t mode, struct dt_allocation_hint *hint,
3095 struct thandle *th, struct lu_attr *attr)
3098 struct osd_device *osd = osd_obj2dev(obj);
3099 struct osd_thandle *oth;
3100 struct dt_object *parent = NULL;
3101 struct inode *inode;
3102 uid_t owner[2] = {0, 0};
3104 if (attr->la_valid & LA_UID)
3105 owner[0] = attr->la_uid;
3106 if (attr->la_valid & LA_GID)
3107 owner[1] = attr->la_gid;
3109 LINVRNT(osd_invariant(obj));
3110 LASSERT(obj->oo_inode == NULL);
3111 LASSERT(obj->oo_hl_head == NULL);
3113 if (S_ISDIR(mode) && ldiskfs_pdo) {
3115 ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3116 if (obj->oo_hl_head == NULL)
3120 oth = container_of(th, struct osd_thandle, ot_super);
3121 LASSERT(oth->ot_handle->h_transaction != NULL);
3123 if (hint != NULL && hint->dah_parent != NULL &&
3124 !dt_object_remote(hint->dah_parent))
3125 parent = hint->dah_parent;
3127 inode = ldiskfs_create_inode(oth->ot_handle,
3128 parent ? osd_dt_obj(parent)->oo_inode :
3129 osd_sb(osd)->s_root->d_inode,
3131 if (!IS_ERR(inode)) {
3132 /* Do not update file c/mtime in ldiskfs. */
3133 inode->i_flags |= S_NOCMTIME;
3136 * For new created object, it must be consistent,
3137 * and it is unnecessary to scrub against it.
3139 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3141 obj->oo_inode = inode;
3144 if (obj->oo_hl_head != NULL) {
3145 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3146 obj->oo_hl_head = NULL;
3148 result = PTR_ERR(inode);
3150 LINVRNT(osd_invariant(obj));
3158 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3159 struct lu_attr *attr,
3160 struct dt_allocation_hint *hint,
3161 struct dt_object_format *dof,
3165 struct osd_thandle *oth;
3166 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3168 LASSERT(S_ISDIR(attr->la_mode));
3170 oth = container_of(th, struct osd_thandle, ot_super);
3171 LASSERT(oth->ot_handle->h_transaction != NULL);
3172 result = osd_mkfile(info, obj, mode, hint, th, attr);
3177 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3178 struct lu_attr *attr,
3179 struct dt_allocation_hint *hint,
3180 struct dt_object_format *dof,
3184 struct osd_thandle *oth;
3185 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3187 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3189 LASSERT(S_ISREG(attr->la_mode));
3191 oth = container_of(th, struct osd_thandle, ot_super);
3192 LASSERT(oth->ot_handle->h_transaction != NULL);
3194 result = osd_mkfile(info, obj, mode, hint, th, attr);
3196 LASSERT(obj->oo_inode != NULL);
3197 if (feat->dif_flags & DT_IND_VARKEY)
3198 result = iam_lvar_create(obj->oo_inode,
3199 feat->dif_keysize_max,
3201 feat->dif_recsize_max,
3204 result = iam_lfix_create(obj->oo_inode,
3205 feat->dif_keysize_max,
3207 feat->dif_recsize_max,
3213 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3214 struct lu_attr *attr,
3215 struct dt_allocation_hint *hint,
3216 struct dt_object_format *dof,
3219 LASSERT(S_ISREG(attr->la_mode));
3220 return osd_mkfile(info, obj, (attr->la_mode &
3221 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3225 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3226 struct lu_attr *attr,
3227 struct dt_allocation_hint *hint,
3228 struct dt_object_format *dof,
3231 LASSERT(S_ISLNK(attr->la_mode));
3232 return osd_mkfile(info, obj, (attr->la_mode &
3233 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3237 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3238 struct lu_attr *attr,
3239 struct dt_allocation_hint *hint,
3240 struct dt_object_format *dof,
3243 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3246 LINVRNT(osd_invariant(obj));
3247 LASSERT(obj->oo_inode == NULL);
3248 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3249 S_ISFIFO(mode) || S_ISSOCK(mode));
3251 result = osd_mkfile(info, obj, mode, hint, th, attr);
3253 LASSERT(obj->oo_inode != NULL);
3255 * This inode should be marked dirty for i_rdev. Currently
3256 * that is done in the osd_attr_init().
3258 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3261 LINVRNT(osd_invariant(obj));
3265 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3267 struct dt_allocation_hint *hint,
3268 struct dt_object_format *dof,
3271 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3273 osd_obj_type_f result;
3289 result = osd_mk_index;
3300 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3301 struct dt_object *parent, struct dt_object *child,
3306 ah->dah_parent = parent;
3307 ah->dah_mode = child_mode;
3309 if (parent != NULL && !dt_object_remote(parent)) {
3310 /* will help to find FID->ino at dt_insert("..") */
3311 struct osd_object *pobj = osd_dt_obj(parent);
3313 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3317 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3318 struct lu_attr *attr, struct dt_object_format *dof)
3320 struct inode *inode = obj->oo_inode;
3321 __u64 valid = attr->la_valid;
3324 attr->la_valid &= ~(LA_TYPE | LA_MODE);
3326 if (dof->dof_type != DFT_NODE)
3327 attr->la_valid &= ~LA_RDEV;
3328 if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3329 attr->la_valid &= ~LA_ATIME;
3330 if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3331 attr->la_valid &= ~LA_CTIME;
3332 if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3333 attr->la_valid &= ~LA_MTIME;
3335 result = osd_quota_transfer(inode, attr);
3339 if (attr->la_valid != 0) {
3340 result = osd_inode_setattr(info->oti_env, inode, attr);
3342 * The osd_inode_setattr() should always succeed here. The
3343 * only error that could be returned is EDQUOT when we are
3344 * trying to change the UID or GID of the inode. However, this
3345 * should not happen since quota enforcement is no longer
3346 * enabled on ldiskfs (lquota takes care of it).
3348 LASSERTF(result == 0, "%d\n", result);
3349 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3352 attr->la_valid = valid;
3356 * Helper function for osd_create()
3358 * \retval 0, on success
3360 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3361 struct lu_attr *attr, struct dt_allocation_hint *hint,
3362 struct dt_object_format *dof, struct thandle *th)
3367 osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3369 /* we drop umask so that permissions we pass are not affected */
3370 umask = current->fs->umask;
3371 current->fs->umask = 0;
3373 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3375 if (likely(obj->oo_inode != NULL)) {
3376 LASSERT(obj->oo_inode->i_state & I_NEW);
3379 * Unlock the inode before attr initialization to avoid
3380 * unnecessary dqget operations. LU-6378
3382 unlock_new_inode(obj->oo_inode);
3385 if (likely(result == 0)) {
3386 osd_attr_init(info, obj, attr, dof);
3387 osd_object_init0(obj);
3390 /* restore previous umask value */
3391 current->fs->umask = umask;
3393 osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3399 * Helper function for osd_create()
3401 * \retval 0, on success
3403 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3404 const struct lu_fid *fid, struct thandle *th)
3406 struct osd_thread_info *info = osd_oti_get(env);
3407 struct osd_inode_id *id = &info->oti_id;
3408 struct osd_device *osd = osd_obj2dev(obj);
3409 struct osd_thandle *oh;
3412 LASSERT(obj->oo_inode != NULL);
3414 oh = container_of0(th, struct osd_thandle, ot_super);
3415 LASSERT(oh->ot_handle);
3416 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3418 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3419 rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3420 OI_CHECK_FLD, NULL);
3421 if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP) && osd->od_is_ost) {
3422 struct lu_fid next_fid = *fid;
3424 /* insert next object in advance, and map to the same inode */
3426 if (next_fid.f_oid != 0) {
3427 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3428 osd_oi_insert(info, osd, &next_fid, id, oh->ot_handle,
3429 OI_CHECK_FLD, NULL);
3430 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3434 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3439 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3440 u64 seq, struct lu_seq_range *range)
3442 struct seq_server_site *ss = osd_seq_site(osd);
3444 if (fid_seq_is_idif(seq)) {
3445 fld_range_set_ost(range);
3446 range->lsr_index = idif_ost_idx(seq);
3450 if (!fid_seq_in_fldb(seq)) {
3451 fld_range_set_mdt(range);
3454 * FIXME: If ss is NULL, it suppose not get lsr_index
3457 range->lsr_index = ss->ss_node_id;
3461 LASSERT(ss != NULL);
3462 fld_range_set_any(range);
3463 /* OSD will only do local fld lookup */
3464 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3467 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3468 struct lu_attr *attr,
3469 struct dt_allocation_hint *hint,
3470 struct dt_object_format *dof,
3471 struct thandle *handle)
3473 struct osd_thandle *oh;
3478 LASSERT(handle != NULL);
3480 oh = container_of0(handle, struct osd_thandle, ot_super);
3481 LASSERT(oh->ot_handle == NULL);
3484 * EA object consumes more credits than regular object: osd_mk_index
3485 * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3486 * leaf_node, could involves the block, block bitmap, groups, GDT
3487 * change for each block, so add 4 * 2 credits in that case.
3489 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3490 osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3491 (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3493 * Reuse idle OI block may cause additional one OI block
3496 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3497 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3498 if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP))
3499 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3500 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3502 /* will help to find FID->ino mapping at dt_insert() */
3503 rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3511 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3512 attr->la_projid, 1, oh, osd_dt_obj(dt),
3513 NULL, OSD_QID_INODE);
3521 * Called to destroy on-disk representation of the object
3523 * Concurrency: must be locked
3525 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3528 struct osd_object *obj = osd_dt_obj(dt);
3529 struct inode *inode = obj->oo_inode;
3530 struct osd_thandle *oh;
3538 oh = container_of0(th, struct osd_thandle, ot_super);
3539 LASSERT(oh->ot_handle == NULL);
3541 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3542 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3544 /* For removing agent entry */
3545 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3546 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3549 * Recycle idle OI leaf may cause additional three OI blocks
3552 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3553 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3554 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3555 /* one less inode */
3556 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3557 i_projid_read(inode), -1, oh, obj, NULL,
3561 /* data to be truncated */
3562 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3563 i_projid_read(inode), 0, oh, obj, NULL,
3569 * will help to find FID->ino when this object is being
3572 rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3577 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3580 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3581 struct osd_object *obj = osd_dt_obj(dt);
3582 struct inode *inode = obj->oo_inode;
3583 struct osd_device *osd = osd_obj2dev(obj);
3584 struct osd_thandle *oh;
3589 oh = container_of0(th, struct osd_thandle, ot_super);
3590 LASSERT(oh->ot_handle);
3592 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3594 if (unlikely(fid_is_acct(fid)))
3597 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3598 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3600 CERROR("%s: remove agent entry "DFID": rc = %d\n",
3601 osd_name(osd), PFID(fid), result);
3604 if (S_ISDIR(inode->i_mode)) {
3605 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3606 inode->i_nlink == 2);
3608 spin_lock(&obj->oo_guard);
3610 spin_unlock(&obj->oo_guard);
3611 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3614 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3616 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3618 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3619 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3620 oh->ot_handle, OI_CHECK_FLD);
3622 osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3623 /* XXX: add to ext3 orphan list */
3624 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3626 /* not needed in the cache anymore */
3627 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3628 obj->oo_destroyed = 1;
3634 * Put the fid into lustre_mdt_attrs, and then place the structure
3635 * inode's ea. This fid should not be altered during the life time
3638 * \retval +ve, on success
3639 * \retval -ve, on error
3641 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3643 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3644 const struct lu_fid *fid, __u32 compat, __u32 incompat)
3646 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3647 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3652 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3655 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3658 lustre_loa_init(loa, fid, compat, incompat);
3659 lustre_loa_swab(loa, false);
3662 * For the OST device with 256 bytes inode size by default,
3663 * the PFID EA will be stored together with LMA EA to avoid
3664 * performance trouble. Otherwise the PFID EA can be stored
3665 * independently. LU-8998
3667 if ((compat & LMAC_FID_ON_OST) &&
3668 LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3669 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3670 sizeof(*loa), XATTR_CREATE);
3672 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3673 sizeof(*lma), XATTR_CREATE);
3675 * LMA may already exist, but we need to check that all the
3676 * desired compat/incompat flags have been added.
3678 if (unlikely(rc == -EEXIST)) {
3679 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3680 XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3684 if (rc < sizeof(*lma))
3687 lustre_loa_swab(loa, true);
3688 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3689 ((compat == 0 && incompat == 0) ||
3690 (!(~lma->lma_compat & compat) &&
3691 !(~lma->lma_incompat & incompat))))
3694 lma->lma_self_fid = *fid;
3695 lma->lma_compat |= compat;
3696 lma->lma_incompat |= incompat;
3697 if (rc == sizeof(*lma)) {
3698 lustre_lma_swab(lma);
3699 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3700 sizeof(*lma), XATTR_REPLACE);
3702 lustre_loa_swab(loa, false);
3703 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3704 sizeof(*loa), XATTR_REPLACE);
3712 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3713 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3714 * To have compatilibility with 1.8 ldiskfs driver we need to have
3715 * magic number at start of fid data.
3716 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3719 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3720 const struct lu_fid *fid)
3722 if (!fid_is_namespace_visible(fid) ||
3723 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3724 param->edp_magic = 0;
3728 param->edp_magic = LDISKFS_LUFID_MAGIC;
3729 param->edp_len = sizeof(struct lu_fid) + 1;
3730 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3734 * Try to read the fid from inode ea into dt_rec.
3736 * \param fid object fid.
3738 * \retval 0 on success
3740 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3741 __u32 ino, struct lu_fid *fid,
3742 struct osd_inode_id *id)
3744 struct osd_thread_info *info = osd_oti_get(env);
3745 struct inode *inode;
3749 osd_id_gen(id, ino, OSD_OII_NOGEN);
3750 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3752 RETURN(PTR_ERR(inode));
3758 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3760 struct inode *parent_dir,
3761 const struct lu_fid *dot_fid,
3762 const struct lu_fid *dot_dot_fid,
3763 struct osd_thandle *oth)
3765 struct ldiskfs_dentry_param *dot_ldp;
3766 struct ldiskfs_dentry_param *dot_dot_ldp;
3767 __u32 saved_nlink = dir->i_nlink;
3770 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3771 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3773 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3774 dot_ldp->edp_magic = 0;
3776 rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3777 dir, dot_ldp, dot_dot_ldp);
3779 * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3780 * the subseqent ref_add() will increase the dir->i_nlink
3781 * as 3. That is incorrect for new created directory.
3783 * It looks like hack, because we want to make the OSD API
3784 * to be order-independent for new created directory object
3785 * between dt_insert(..) and ref_add() operations.
3787 * Here, we only restore the in-RAM dir-inode's nlink attr,
3788 * becuase if the nlink attr is not 2, then there will be
3789 * ref_add() called following the dt_insert(..), such call
3790 * will make both the in-RAM and on-disk dir-inode's nlink
3791 * attr to be set as 2. LU-7447
3793 set_nlink(dir, saved_nlink);
3798 * Create an local agent inode for remote entry
3800 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3801 struct osd_device *osd,
3802 struct osd_object *pobj,
3803 const struct lu_fid *fid,
3807 struct osd_thread_info *info = osd_oti_get(env);
3808 struct inode *local;
3809 struct osd_thandle *oh;
3815 oh = container_of(th, struct osd_thandle, ot_super);
3816 LASSERT(oh->ot_handle->h_transaction != NULL);
3818 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type,
3820 if (IS_ERR(local)) {
3821 CERROR("%s: create local error %d\n", osd_name(osd),
3822 (int)PTR_ERR(local));
3827 * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3828 * correct gid on remote file, not agent here
3830 local->i_gid = current_fsgid();
3831 ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3833 /* e2fsck doesn't like empty symlinks. Store remote FID as symlink.
3834 * That gives e2fsck something to look at and be happy, and allows
3835 * debugging if we need to determine where this symlink came from.
3837 if (S_ISLNK(type)) {
3838 BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1);
3839 rc = snprintf((char *)LDISKFS_I(local)->i_data,
3840 LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
3842 i_size_write(local, rc);
3843 LDISKFS_I(local)->i_disksize = rc;
3845 unlock_new_inode(local);
3847 /* Agent inode should not have project ID */
3848 #ifdef HAVE_PROJECT_QUOTA
3849 if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3850 i_projid_read(pobj->oo_inode) != 0) {
3851 rc = osd_transfer_project(local, 0);
3853 CERROR("%s: quota transfer failed: rc = %d. Is project "
3854 "quota enforcement enabled on the ldiskfs "
3855 "filesystem?\n", local->i_sb->s_id, rc);
3856 RETURN(ERR_PTR(rc));
3860 /* Set special LMA flag for local agent inode */
3861 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3863 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3864 osd_name(osd), PFID(fid), rc);
3865 RETURN(ERR_PTR(rc));
3871 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3872 lu_object_fid(&pobj->oo_dt.do_lu),
3875 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3876 osd_name(osd), PFID(fid), rc);
3877 RETURN(ERR_PTR(rc));
3884 * when direntry is deleted, we have to take care of possible agent inode
3885 * referenced by that. unfortunately we can't do this at that point:
3886 * iget() within a running transaction leads to deadlock and we better do
3887 * not call that every delete declaration to save performance. so we put
3888 * a potention agent inode on a list and process that once the transaction
3889 * is over. Notice it's not any worse in terms of real orphans as regular
3890 * object destroy doesn't put inodes on the on-disk orphan list. this should
3891 * be addressed separately
3893 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3894 struct osd_thandle *oh,
3897 struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3898 struct osd_obj_orphan *oor;
3905 oor->oor_env = (struct lu_env *)env;
3906 spin_lock(&osd->od_osfs_lock);
3907 list_add(&oor->oor_list, &osd->od_orphan_list);
3908 spin_unlock(&osd->od_osfs_lock);
3910 oh->ot_remove_agents = 1;
3916 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3917 struct osd_device *osd)
3919 struct osd_thread_info *info = osd_oti_get(env);
3920 struct osd_obj_orphan *oor, *tmp;
3921 struct osd_inode_id id;
3922 struct list_head list;
3923 struct inode *inode;
3928 INIT_LIST_HEAD(&list);
3930 spin_lock(&osd->od_osfs_lock);
3931 list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3932 if (oor->oor_env == env)
3933 list_move(&oor->oor_list, &list);
3935 spin_unlock(&osd->od_osfs_lock);
3937 list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3941 list_del(&oor->oor_list);
3944 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3945 inode = osd_iget_fid(info, osd, &id, &fid);
3949 if (!osd_remote_fid(env, osd, &fid)) {
3954 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3956 mark_inode_dirty(inode);
3957 ldiskfs_journal_stop(jh);
3965 * OSD layer object create function for OST objects (b=11826).
3967 * The FID is inserted into inode xattr here.
3969 * \retval 0, on success
3970 * \retval -ve, on error
3972 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3973 struct lu_attr *attr, struct dt_allocation_hint *hint,
3974 struct dt_object_format *dof, struct thandle *th)
3976 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3977 struct osd_object *obj = osd_dt_obj(dt);
3978 struct osd_thread_info *info = osd_oti_get(env);
3979 int result, on_ost = 0;
3983 if (dt_object_exists(dt))
3986 LINVRNT(osd_invariant(obj));
3987 LASSERT(!dt_object_remote(dt));
3988 LASSERT(osd_is_write_locked(env, obj));
3989 LASSERT(th != NULL);
3991 if (unlikely(fid_is_acct(fid)))
3993 * Quota files can't be created from the kernel any more,
3994 * 'tune2fs -O quota' will take care of creating them
3998 result = __osd_create(info, obj, attr, hint, dof, th);
4000 if (fid_is_idif(fid) &&
4001 !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
4002 struct lu_fid *tfid = &info->oti_fid;
4003 struct ost_id *oi = &info->oti_ostid;
4005 fid_to_ostid(fid, oi);
4006 ostid_to_fid(tfid, oi, 0);
4008 result = osd_ea_fid_set(info, obj->oo_inode, tfid,
4009 LMAC_FID_ON_OST, 0);
4011 on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
4013 result = osd_ea_fid_set(info, obj->oo_inode, fid,
4014 on_ost ? LMAC_FID_ON_OST : 0,
4017 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
4018 obj->oo_dt.do_body_ops = &osd_body_ops;
4021 if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
4022 result = __osd_oi_insert(env, obj, fid, th);
4025 * a small optimization - dt_insert() isn't usually applied
4026 * to OST objects, so we don't need to cache OI mapping for
4029 if (result == 0 && on_ost == 0) {
4030 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4032 result = osd_idc_find_and_init(env, osd, obj);
4033 LASSERT(result == 0);
4036 LASSERT(ergo(result == 0,
4037 dt_object_exists(dt) && !dt_object_remote(dt)));
4038 LINVRNT(osd_invariant(obj));
4042 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4043 struct thandle *handle)
4045 struct osd_thandle *oh;
4047 /* it's possible that object doesn't exist yet */
4048 LASSERT(handle != NULL);
4050 oh = container_of0(handle, struct osd_thandle, ot_super);
4051 LASSERT(oh->ot_handle == NULL);
4053 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4054 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4056 osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
4062 * Concurrency: @dt is write locked.
4064 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4067 struct osd_object *obj = osd_dt_obj(dt);
4068 struct inode *inode = obj->oo_inode;
4069 struct osd_thandle *oh;
4072 if (!dt_object_exists(dt) || obj->oo_destroyed)
4075 LINVRNT(osd_invariant(obj));
4076 LASSERT(!dt_object_remote(dt));
4077 LASSERT(osd_is_write_locked(env, obj));
4078 LASSERT(th != NULL);
4080 oh = container_of0(th, struct osd_thandle, ot_super);
4081 LASSERT(oh->ot_handle != NULL);
4083 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4085 CDEBUG(D_INODE, DFID" increase nlink %d\n",
4086 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4088 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4089 * (65000) subdirectories by storing "1" in i_nlink if the link count
4090 * would otherwise overflow. Directory tranversal tools understand
4091 * that (st_nlink == 1) indicates that the filesystem dose not track
4092 * hard links count on the directory, and will not abort subdirectory
4093 * scanning early once (st_nlink - 2) subdirs have been found.
4095 * This also has to properly handle the case of inodes with nlink == 0
4096 * in case they are being linked into the PENDING directory
4098 spin_lock(&obj->oo_guard);
4099 if (unlikely(inode->i_nlink == 0))
4100 /* inc_nlink from 0 may cause WARN_ON */
4101 set_nlink(inode, 1);
4103 ldiskfs_inc_count(oh->ot_handle, inode);
4104 if (!S_ISDIR(inode->i_mode))
4105 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4107 spin_unlock(&obj->oo_guard);
4109 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4110 LINVRNT(osd_invariant(obj));
4112 osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4117 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4118 struct thandle *handle)
4120 struct osd_thandle *oh;
4122 if (!dt_object_exists(dt))
4125 LASSERT(!dt_object_remote(dt));
4126 LASSERT(handle != NULL);
4128 oh = container_of0(handle, struct osd_thandle, ot_super);
4129 LASSERT(oh->ot_handle == NULL);
4131 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4132 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4138 * Concurrency: @dt is write locked.
4140 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4143 struct osd_object *obj = osd_dt_obj(dt);
4144 struct inode *inode = obj->oo_inode;
4145 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4146 struct osd_thandle *oh;
4148 if (!dt_object_exists(dt))
4151 LINVRNT(osd_invariant(obj));
4152 LASSERT(!dt_object_remote(dt));
4153 LASSERT(osd_is_write_locked(env, obj));
4154 LASSERT(th != NULL);
4156 oh = container_of0(th, struct osd_thandle, ot_super);
4157 LASSERT(oh->ot_handle != NULL);
4159 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4161 spin_lock(&obj->oo_guard);
4163 * That can be result of upgrade from old Lustre version and
4164 * applied only to local files. Just skip this ref_del call.
4165 * ext4_unlink() only treats this as a warning, don't LASSERT here.
4167 if (inode->i_nlink == 0) {
4168 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4169 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4170 ", maybe an upgraded file? (LU-3915)\n",
4171 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4172 spin_unlock(&obj->oo_guard);
4176 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4177 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4179 ldiskfs_dec_count(oh->ot_handle, inode);
4180 spin_unlock(&obj->oo_guard);
4182 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4183 LINVRNT(osd_invariant(obj));
4185 osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4191 * Concurrency: @dt is read locked.
4193 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4194 struct lu_buf *buf, const char *name)
4196 struct osd_object *obj = osd_dt_obj(dt);
4197 struct inode *inode = obj->oo_inode;
4198 struct osd_thread_info *info = osd_oti_get(env);
4199 struct dentry *dentry = &info->oti_obj_dentry;
4200 bool cache_xattr = false;
4205 /* version get is not real XATTR but uses xattr API */
4206 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4207 dt_obj_version_t *ver = buf->lb_buf;
4210 * for version we are just using xattr API but change inode
4213 if (buf->lb_len == 0)
4214 return sizeof(dt_obj_version_t);
4216 if (buf->lb_len < sizeof(dt_obj_version_t))
4219 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4220 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4222 *ver = LDISKFS_I(inode)->i_fs_version;
4224 return sizeof(dt_obj_version_t);
4227 if (!dt_object_exists(dt))
4230 LASSERT(!dt_object_remote(dt));
4231 LASSERT(inode->i_op != NULL);
4232 #ifdef HAVE_IOP_XATTR
4233 LASSERT(inode->i_op->getxattr != NULL);
4236 if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4237 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4241 rc = osd_oxc_get(obj, name, buf);
4246 if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4247 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4248 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4249 struct filter_fid *ff;
4250 struct ost_layout *ol;
4252 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4256 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4259 if (buf->lb_len == 0 || !buf->lb_buf)
4262 if (buf->lb_len < rc)
4266 ol = &ff->ff_layout;
4267 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4268 PFID_STRIPE_IDX_BITS);
4269 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4270 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4271 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4272 if (lma->lma_compat & LMAC_COMP_INFO) {
4273 ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4274 ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4275 ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4277 ol->ol_comp_start = 0;
4278 ol->ol_comp_end = 0;
4282 rc = __osd_xattr_get(inode, dentry, name,
4283 buf->lb_buf, buf->lb_len);
4287 if (rc == -ENOENT || rc == -ENODATA)
4288 osd_oxc_add(obj, name, NULL, 0);
4289 else if (rc > 0 && buf->lb_buf != NULL)
4290 osd_oxc_add(obj, name, buf->lb_buf, rc);
4296 static int osd_declare_xattr_set(const struct lu_env *env,
4297 struct dt_object *dt,
4298 const struct lu_buf *buf, const char *name,
4299 int fl, struct thandle *handle)
4301 struct osd_thandle *oh;
4303 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4305 LASSERT(handle != NULL);
4307 oh = container_of0(handle, struct osd_thandle, ot_super);
4308 LASSERT(oh->ot_handle == NULL);
4310 if (strcmp(name, XATTR_NAME_LMA) == 0) {
4312 * For non-upgrading case, the LMA is set first and
4313 * usually fit inode. But for upgrade case, the LMA
4314 * may be in another separated EA block.
4316 if (dt_object_exists(dt)) {
4317 if (fl == LU_XATTR_REPLACE)
4322 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4324 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4325 /* We may need to delete the old PFID EA. */
4326 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4327 if (fl == LU_XATTR_REPLACE)
4333 * If some name entry resides on remote MDT, then will create
4334 * agent entry under remote parent. On the other hand, if the
4335 * remote entry will be removed, then related agent entry may
4336 * need to be removed from the remote parent. So there may be
4337 * kinds of cases, let's declare enough credits. The credits
4338 * for create agent entry is enough for remove case.
4340 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4341 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4342 if (dt_object_exists(dt))
4343 credits += 1; /* For updating LMA */
4347 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4352 if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4354 * learn xattr size from osd_xattr_get if
4355 * attribute has not been read yet
4357 buflen = __osd_xattr_get(
4358 osd_dt_obj(dt)->oo_inode,
4359 &osd_oti_get(env)->oti_obj_dentry,
4364 buflen = buf->lb_len;
4367 if (buflen > sb->s_blocksize) {
4368 credits += osd_calc_bkmap_credits(
4370 (buflen + sb->s_blocksize - 1) >>
4371 sb->s_blocksize_bits);
4375 * xattr set may involve inode quota change, reserve credits for
4376 * dquot_initialize()
4378 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4381 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4386 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4387 const struct lu_buf *buf, int fl,
4388 struct thandle *handle)
4390 struct osd_thread_info *info = osd_oti_get(env);
4391 struct dentry *dentry = &info->oti_obj_dentry;
4392 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4393 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4394 struct inode *inode = obj->oo_inode;
4395 struct filter_fid *ff = buf->lb_buf;
4396 struct ost_layout *ol = &ff->ff_layout;
4397 int flags = XATTR_REPLACE;
4402 if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4405 rc = osd_get_lma(info, inode, dentry, loa);
4406 if (rc == -ENODATA) {
4407 /* Usually for upgarding from old device */
4408 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4409 LMAC_FID_ON_OST, 0);
4410 flags = XATTR_CREATE;
4415 if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4416 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4419 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4420 /* Separate PFID EA from LMA */
4421 lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4422 lustre_lma_swab(lma);
4423 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4424 sizeof(*lma), XATTR_REPLACE);
4426 obj->oo_pfid_in_lma = 0;
4427 rc = LU_XATTR_CREATE;
4433 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4437 * Old client does not send stripe information,
4438 * then store the PFID EA on disk separatedly.
4440 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4441 ol->ol_stripe_size == 0))
4444 /* Remove old PFID EA entry firstly. */
4445 dquot_initialize(inode);
4446 rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
4447 if (rc == -ENODATA) {
4448 if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4455 fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4456 if (likely(ol->ol_stripe_size != 0)) {
4457 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4458 PFID_STRIPE_IDX_BITS;
4459 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4460 lma->lma_compat |= LMAC_STRIPE_INFO;
4461 if (ol->ol_comp_id != 0) {
4462 loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4463 loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4464 loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4465 lma->lma_compat |= LMAC_COMP_INFO;
4469 lustre_loa_swab(loa, false);
4471 /* Store the PFID EA inside LMA. */
4472 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4475 obj->oo_pfid_in_lma = 1;
4481 * In DNE environment, the object (in spite of regular file or directory)
4482 * and its name entry may reside on different MDTs. Under such case, we will
4483 * create an agent entry on the MDT where the object resides. The agent entry
4484 * references the object locally, that makes the object to be visible to the
4485 * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4486 * such as 'tar' can handle the object properly.
4488 * We handle the agent entry during set linkEA that is the common interface
4489 * for both regular file and directroy, can handle kinds of cases, such as
4490 * create/link/unlink/rename, and so on.
4492 * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4494 * XXX: There are two known issues:
4495 * 1. For one object, we will create at most one agent entry even if there
4496 * may be more than one cross-MDTs hard links on the object. So the local
4497 * e2fsck may claim that the object's nlink is larger than the name entries
4498 * that reference such inode. And in further, the e2fsck will fix the nlink
4499 * attribute to match the local references. Then it will cause the object's
4500 * nlink attribute to be inconsistent with the global references. it is bad
4501 * but not fatal. The ref_del() can handle the zero-referenced case. On the
4502 * other hand, the global namespace LFSCK can repair the object's attribute
4503 * according to the linkEA.
4504 * 2. There may be too many hard links on the object as to its linkEA overflow,
4505 * then the linkEA entry for cross-MDTs reference may be discarded. If such
4506 * case happened, then at this point, we do not know whether there are some
4507 * cross-MDTs reference. But there are local references, it guarantees that
4508 * object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4510 static int osd_xattr_handle_linkea(const struct lu_env *env,
4511 struct osd_device *osd,
4512 struct osd_object *obj,
4513 const struct lu_buf *buf,
4514 struct thandle *handle)
4516 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4517 struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4518 struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4519 struct lu_name tmpname;
4520 struct osd_thandle *oh;
4522 bool remote = false;
4526 oh = container_of0(handle, struct osd_thandle, ot_super);
4527 LASSERT(oh->ot_handle != NULL);
4529 rc = linkea_init_with_rec(&ldata);
4531 linkea_first_entry(&ldata);
4532 while (ldata.ld_lee != NULL && !remote) {
4533 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4535 if (osd_remote_fid(env, osd, tfid) > 0)
4538 linkea_next_entry(&ldata);
4540 } else if (rc == -ENODATA) {
4546 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4547 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4549 CERROR("%s: failed to remove agent entry for "DFID
4550 ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4551 } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4552 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4554 CERROR("%s: failed to create agent entry for "DFID
4555 ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4562 * Concurrency: @dt is write locked.
4564 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4565 const struct lu_buf *buf, const char *name, int fl,
4566 struct thandle *handle)
4568 struct osd_object *obj = osd_dt_obj(dt);
4569 struct osd_device *osd = osd_obj2dev(obj);
4570 struct inode *inode = obj->oo_inode;
4571 struct osd_thread_info *info = osd_oti_get(env);
4581 /* version set is not real XATTR */
4582 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4583 dt_obj_version_t *version = buf->lb_buf;
4586 * for version we are just using xattr API but change inode
4589 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4591 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4592 *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4594 LDISKFS_I(inode)->i_fs_version = *version;
4596 * Version is set after all inode operations are finished,
4597 * so we should mark it dirty here
4599 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4604 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4605 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4608 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4611 * For the OST device with 256 bytes inode size by default,
4612 * the PFID EA will be stored together with LMA EA to avoid
4613 * performance trouble. Otherwise the PFID EA can be stored
4614 * independently. LU-8998
4616 if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4617 (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4618 LASSERT(buf->lb_buf);
4620 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4623 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4624 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4625 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4627 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4631 lma->lma_incompat |= LMAI_STRIPED;
4632 lustre_lma_swab(lma);
4633 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4634 sizeof(*lma), XATTR_REPLACE);
4637 } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4638 LASSERT(!osd->od_is_ost);
4640 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4645 if (fl & LU_XATTR_REPLACE)
4646 fs_flags |= XATTR_REPLACE;
4648 if (fl & LU_XATTR_CREATE)
4649 fs_flags |= XATTR_CREATE;
4651 rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4652 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4655 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4656 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4657 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4663 * Concurrency: @dt is read locked.
4665 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4666 const struct lu_buf *buf)
4668 struct osd_object *obj = osd_dt_obj(dt);
4669 struct inode *inode = obj->oo_inode;
4670 struct osd_thread_info *info = osd_oti_get(env);
4671 struct dentry *dentry = &info->oti_obj_dentry;
4673 if (!dt_object_exists(dt))
4676 LASSERT(!dt_object_remote(dt));
4677 LASSERT(inode->i_op != NULL);
4678 LASSERT(inode->i_op->listxattr != NULL);
4680 dentry->d_inode = inode;
4681 dentry->d_sb = inode->i_sb;
4682 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4685 static int osd_declare_xattr_del(const struct lu_env *env,
4686 struct dt_object *dt, const char *name,
4687 struct thandle *handle)
4689 struct osd_thandle *oh;
4690 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4692 LASSERT(!dt_object_remote(dt));
4693 LASSERT(handle != NULL);
4695 oh = container_of0(handle, struct osd_thandle, ot_super);
4696 LASSERT(oh->ot_handle == NULL);
4698 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4699 osd_dto_credits_noquota[DTO_XATTR_SET]);
4701 * xattr del may involve inode quota change, reserve credits for
4702 * dquot_initialize()
4704 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4710 * Concurrency: @dt is write locked.
4712 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4713 const char *name, struct thandle *handle)
4715 struct osd_object *obj = osd_dt_obj(dt);
4716 struct inode *inode = obj->oo_inode;
4717 struct osd_thread_info *info = osd_oti_get(env);
4718 struct dentry *dentry = &info->oti_obj_dentry;
4721 if (!dt_object_exists(dt))
4724 LASSERT(!dt_object_remote(dt));
4725 LASSERT(inode->i_op != NULL);
4726 LASSERT(handle != NULL);
4727 #ifdef HAVE_IOP_XATTR
4728 LASSERT(inode->i_op->removexattr != NULL);
4731 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4733 if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4734 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4736 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4737 &info->oti_ost_attrs);
4739 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4741 lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4742 lustre_lma_swab(lma);
4743 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4744 sizeof(*lma), XATTR_REPLACE);
4746 obj->oo_pfid_in_lma = 0;
4749 dquot_initialize(inode);
4750 dentry->d_inode = inode;
4751 dentry->d_sb = inode->i_sb;
4752 rc = osd_removexattr(dentry, inode, name);
4755 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4758 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4759 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4760 osd_oxc_del(obj, name);
4765 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4766 __u64 start, __u64 end)
4768 struct osd_object *obj = osd_dt_obj(dt);
4769 struct inode *inode = obj->oo_inode;
4770 struct osd_thread_info *info = osd_oti_get(env);
4771 struct dentry *dentry = &info->oti_obj_dentry;
4772 struct file *file = &info->oti_file;
4777 dentry->d_inode = inode;
4778 dentry->d_sb = inode->i_sb;
4779 file->f_path.dentry = dentry;
4780 file->f_mapping = inode->i_mapping;
4781 file->f_op = inode->i_fop;
4782 set_file_inode(file, inode);
4784 rc = vfs_fsync_range(file, start, end, 0);
4789 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4798 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4799 const struct dt_index_features *feat)
4801 struct iam_descr *descr;
4803 if (osd_object_is_root(o))
4804 return feat == &dt_directory_features;
4806 LASSERT(o->oo_dir != NULL);
4808 descr = o->oo_dir->od_container.ic_descr;
4809 if (feat == &dt_directory_features) {
4810 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4815 return feat->dif_keysize_min <= descr->id_key_size &&
4816 descr->id_key_size <= feat->dif_keysize_max &&
4817 feat->dif_recsize_min <= descr->id_rec_size &&
4818 descr->id_rec_size <= feat->dif_recsize_max &&
4819 !(feat->dif_flags & (DT_IND_VARKEY |
4820 DT_IND_VARREC | DT_IND_NONUNQ)) &&
4821 ergo(feat->dif_flags & DT_IND_UPDATE,
4822 1 /* XXX check that object (and fs) is writable */);
4826 static int osd_iam_container_init(const struct lu_env *env,
4827 struct osd_object *obj,
4828 struct osd_directory *dir)
4830 struct iam_container *bag = &dir->od_container;
4833 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4837 result = iam_container_setup(bag);
4839 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4841 iam_container_fini(bag);
4848 * Concurrency: no external locking is necessary.
4850 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4851 const struct dt_index_features *feat)
4855 struct osd_object *obj = osd_dt_obj(dt);
4857 LINVRNT(osd_invariant(obj));
4859 if (osd_object_is_root(obj)) {
4860 dt->do_index_ops = &osd_index_ea_ops;
4862 } else if (feat == &dt_directory_features) {
4863 dt->do_index_ops = &osd_index_ea_ops;
4864 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4869 } else if (unlikely(feat == &dt_otable_features)) {
4870 dt->do_index_ops = &osd_otable_ops;
4872 } else if (unlikely(feat == &dt_acct_features)) {
4873 dt->do_index_ops = &osd_acct_index_ops;
4876 } else if (!osd_has_index(obj)) {
4877 struct osd_directory *dir;
4878 struct osd_device *osd = osd_obj2dev(obj);
4879 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4884 spin_lock(&obj->oo_guard);
4885 if (obj->oo_dir == NULL)
4889 * Concurrent thread allocated container data.
4892 spin_unlock(&obj->oo_guard);
4894 * Now, that we have container data, serialize its
4897 down_write(&obj->oo_ext_idx_sem);
4899 * recheck under lock.
4902 if (osd_has_index(obj)) {
4907 result = osd_iam_container_init(env, obj, obj->oo_dir);
4908 if (result || feat == &dt_lfsck_namespace_features ||
4909 feat == &dt_lfsck_layout_orphan_features ||
4910 feat == &dt_lfsck_layout_dangling_features)
4913 result = osd_index_register(osd, fid,
4914 feat->dif_keysize_max,
4915 feat->dif_recsize_max);
4917 CWARN("%s: failed to register index "
4919 osd_name(osd), PFID(fid), result);
4920 else if (result > 0)
4923 CDEBUG(D_LFSCK, "%s: index object "DFID
4924 " (%d/%d) registered\n",
4925 osd_name(osd), PFID(fid),
4926 (int)feat->dif_keysize_max,
4927 (int)feat->dif_recsize_max);
4930 up_write(&obj->oo_ext_idx_sem);
4938 if (result == 0 && skip_iam == 0) {
4939 if (!osd_iam_index_probe(env, obj, feat))
4942 LINVRNT(osd_invariant(obj));
4947 static int osd_otable_it_attr_get(const struct lu_env *env,
4948 struct dt_object *dt,
4949 struct lu_attr *attr)
4955 static const struct dt_object_operations osd_obj_ops = {
4956 .do_read_lock = osd_read_lock,
4957 .do_write_lock = osd_write_lock,
4958 .do_read_unlock = osd_read_unlock,
4959 .do_write_unlock = osd_write_unlock,
4960 .do_write_locked = osd_write_locked,
4961 .do_attr_get = osd_attr_get,
4962 .do_declare_attr_set = osd_declare_attr_set,
4963 .do_attr_set = osd_attr_set,
4964 .do_ah_init = osd_ah_init,
4965 .do_declare_create = osd_declare_create,
4966 .do_create = osd_create,
4967 .do_declare_destroy = osd_declare_destroy,
4968 .do_destroy = osd_destroy,
4969 .do_index_try = osd_index_try,
4970 .do_declare_ref_add = osd_declare_ref_add,
4971 .do_ref_add = osd_ref_add,
4972 .do_declare_ref_del = osd_declare_ref_del,
4973 .do_ref_del = osd_ref_del,
4974 .do_xattr_get = osd_xattr_get,
4975 .do_declare_xattr_set = osd_declare_xattr_set,
4976 .do_xattr_set = osd_xattr_set,
4977 .do_declare_xattr_del = osd_declare_xattr_del,
4978 .do_xattr_del = osd_xattr_del,
4979 .do_xattr_list = osd_xattr_list,
4980 .do_object_sync = osd_object_sync,
4981 .do_invalidate = osd_invalidate,
4984 static const struct dt_object_operations osd_obj_otable_it_ops = {
4985 .do_attr_get = osd_otable_it_attr_get,
4986 .do_index_try = osd_index_try,
4989 static int osd_index_declare_iam_delete(const struct lu_env *env,
4990 struct dt_object *dt,
4991 const struct dt_key *key,
4992 struct thandle *handle)
4994 struct osd_thandle *oh;
4996 oh = container_of0(handle, struct osd_thandle, ot_super);
4997 LASSERT(oh->ot_handle == NULL);
4999 /* Recycle may cause additional three blocks to be changed. */
5000 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
5001 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
5007 * delete a (key, value) pair from index \a dt specified by \a key
5009 * \param dt osd index object
5010 * \param key key for index
5011 * \param rec record reference
5012 * \param handle transaction handler
5015 * \retval -ve failure
5017 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
5018 const struct dt_key *key,
5019 struct thandle *handle)
5021 struct osd_thread_info *oti = osd_oti_get(env);
5022 struct osd_object *obj = osd_dt_obj(dt);
5023 struct osd_thandle *oh;
5024 struct iam_path_descr *ipd;
5025 struct iam_container *bag = &obj->oo_dir->od_container;
5030 if (!dt_object_exists(dt))
5033 LINVRNT(osd_invariant(obj));
5034 LASSERT(!dt_object_remote(dt));
5035 LASSERT(bag->ic_object == obj->oo_inode);
5036 LASSERT(handle != NULL);
5038 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5040 ipd = osd_idx_ipd_get(env, bag);
5041 if (unlikely(ipd == NULL))
5044 oh = container_of0(handle, struct osd_thandle, ot_super);
5045 LASSERT(oh->ot_handle != NULL);
5046 LASSERT(oh->ot_handle->h_transaction != NULL);
5048 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5049 /* swab quota uid/gid provided by caller */
5050 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5051 key = (const struct dt_key *)&oti->oti_quota_id;
5054 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5055 osd_ipd_put(env, bag, ipd);
5056 LINVRNT(osd_invariant(obj));
5057 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5061 static int osd_index_declare_ea_delete(const struct lu_env *env,
5062 struct dt_object *dt,
5063 const struct dt_key *key,
5064 struct thandle *handle)
5066 struct osd_thandle *oh;
5067 struct inode *inode;
5072 LASSERT(!dt_object_remote(dt));
5073 LASSERT(handle != NULL);
5075 oh = container_of0(handle, struct osd_thandle, ot_super);
5076 LASSERT(oh->ot_handle == NULL);
5078 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5079 osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5081 inode = osd_dt_obj(dt)->oo_inode;
5085 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5086 i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5091 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5094 struct osd_fid_pack *rec;
5097 if (de->file_type & LDISKFS_DIRENT_LUFID) {
5098 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5099 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5100 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5106 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5107 const struct lu_fid *fid)
5109 struct seq_server_site *ss = osd_seq_site(osd);
5113 /* FID seqs not in FLDB, must be local seq */
5114 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5118 * If FLD is not being initialized yet, it only happens during the
5119 * initialization, likely during mgs initialization, and we assume
5120 * this is local FID.
5122 if (ss == NULL || ss->ss_server_fld == NULL)
5125 /* Only check the local FLDB here */
5126 if (osd_seq_exists(env, osd, fid_seq(fid)))
5132 static void osd_take_care_of_agent(const struct lu_env *env,
5133 struct osd_device *osd,
5134 struct osd_thandle *oh,
5135 struct ldiskfs_dir_entry_2 *de)
5137 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5138 struct osd_idmap_cache *idc;
5139 int rc, schedule = 0;
5141 LASSERT(de != NULL);
5143 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5144 if (likely(rc == 0)) {
5145 idc = osd_idc_find_or_init(env, osd, fid);
5146 if (IS_ERR(idc) || idc->oic_remote)
5148 } else if (rc == -ENODATA) {
5150 * can't get FID, postpone to the end of the
5151 * transaction when iget() is safe
5155 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5158 osd_schedule_agent_inode_removal(env, oh,
5159 le32_to_cpu(de->inode));
5163 * Index delete function for interoperability mode (b11826).
5164 * It will remove the directory entry added by osd_index_ea_insert().
5165 * This entry is needed to maintain name->fid mapping.
5167 * \param key, key i.e. file entry to be deleted
5169 * \retval 0, on success
5170 * \retval -ve, on error
5172 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5173 const struct dt_key *key, struct thandle *handle)
5175 struct osd_object *obj = osd_dt_obj(dt);
5176 struct inode *dir = obj->oo_inode;
5177 struct dentry *dentry;
5178 struct osd_thandle *oh;
5179 struct ldiskfs_dir_entry_2 *de = NULL;
5180 struct buffer_head *bh;
5181 struct htree_lock *hlock = NULL;
5182 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5187 if (!dt_object_exists(dt))
5190 LINVRNT(osd_invariant(obj));
5191 LASSERT(!dt_object_remote(dt));
5192 LASSERT(handle != NULL);
5194 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5196 oh = container_of(handle, struct osd_thandle, ot_super);
5197 LASSERT(oh->ot_handle != NULL);
5198 LASSERT(oh->ot_handle->h_transaction != NULL);
5200 dquot_initialize(dir);
5201 dentry = osd_child_dentry_get(env, obj,
5202 (char *)key, strlen((char *)key));
5204 if (obj->oo_hl_head != NULL) {
5205 hlock = osd_oti_get(env)->oti_hlock;
5206 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5207 dir, LDISKFS_HLOCK_DEL);
5209 down_write(&obj->oo_ext_idx_sem);
5212 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5215 * If this is not the ".." entry, it might be a remote DNE
5216 * entry and we need to check if the FID is for a remote
5217 * MDT. If the FID is not in the directory entry (e.g.
5218 * upgraded 1.8 filesystem without dirdata enabled) then
5219 * we need to get the FID from the LMA. For a remote directory
5220 * there HAS to be an LMA, it cannot be an IGIF inode in this
5223 * Delete the entry before the agent inode in order to
5224 * simplify error handling. At worst an error after deleting
5225 * the entry first might leak the agent inode afterward. The
5226 * reverse would need filesystem abort in case of error deleting
5227 * the entry after the agent had been removed, or leave a
5228 * dangling entry pointing at a random inode.
5230 if (strcmp((char *)key, dotdot) != 0)
5231 osd_take_care_of_agent(env, osd, oh, de);
5232 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5238 ldiskfs_htree_unlock(hlock);
5240 up_write(&obj->oo_ext_idx_sem);
5244 LASSERT(osd_invariant(obj));
5245 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5250 * Lookup index for \a key and copy record to \a rec.
5252 * \param dt osd index object
5253 * \param key key for index
5254 * \param rec record reference
5256 * \retval +ve success : exact mach
5257 * \retval 0 return record with key not greater than \a key
5258 * \retval -ve failure
5260 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5261 struct dt_rec *rec, const struct dt_key *key)
5263 struct osd_object *obj = osd_dt_obj(dt);
5264 struct iam_path_descr *ipd;
5265 struct iam_container *bag = &obj->oo_dir->od_container;
5266 struct osd_thread_info *oti = osd_oti_get(env);
5267 struct iam_iterator *it = &oti->oti_idx_it;
5268 struct iam_rec *iam_rec;
5273 if (!dt_object_exists(dt))
5276 LASSERT(osd_invariant(obj));
5277 LASSERT(!dt_object_remote(dt));
5278 LASSERT(bag->ic_object == obj->oo_inode);
5280 ipd = osd_idx_ipd_get(env, bag);
5284 /* got ipd now we can start iterator. */
5285 iam_it_init(it, bag, 0, ipd);
5287 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5288 /* swab quota uid/gid provided by caller */
5289 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5290 key = (const struct dt_key *)&oti->oti_quota_id;
5293 rc = iam_it_get(it, (struct iam_key *)key);
5295 if (S_ISDIR(obj->oo_inode->i_mode))
5296 iam_rec = (struct iam_rec *)oti->oti_ldp;
5298 iam_rec = (struct iam_rec *)rec;
5300 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5302 if (S_ISDIR(obj->oo_inode->i_mode))
5303 osd_fid_unpack((struct lu_fid *)rec,
5304 (struct osd_fid_pack *)iam_rec);
5305 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5306 osd_quota_unpack(obj, rec);
5311 osd_ipd_put(env, bag, ipd);
5313 LINVRNT(osd_invariant(obj));
5318 static int osd_index_declare_iam_insert(const struct lu_env *env,
5319 struct dt_object *dt,
5320 const struct dt_rec *rec,
5321 const struct dt_key *key,
5322 struct thandle *handle)
5324 struct osd_thandle *oh;
5326 LASSERT(handle != NULL);
5328 oh = container_of0(handle, struct osd_thandle, ot_super);
5329 LASSERT(oh->ot_handle == NULL);
5331 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5332 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5338 * Inserts (key, value) pair in \a dt index object.
5340 * \param dt osd index object
5341 * \param key key for index
5342 * \param rec record reference
5343 * \param th transaction handler
5346 * \retval -ve failure
5348 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5349 const struct dt_rec *rec,
5350 const struct dt_key *key, struct thandle *th)
5352 struct osd_object *obj = osd_dt_obj(dt);
5353 struct iam_path_descr *ipd;
5354 struct osd_thandle *oh;
5355 struct iam_container *bag;
5356 struct osd_thread_info *oti = osd_oti_get(env);
5357 struct iam_rec *iam_rec;
5362 if (!dt_object_exists(dt))
5365 LINVRNT(osd_invariant(obj));
5366 LASSERT(!dt_object_remote(dt));
5368 bag = &obj->oo_dir->od_container;
5369 LASSERT(bag->ic_object == obj->oo_inode);
5370 LASSERT(th != NULL);
5372 osd_trans_exec_op(env, th, OSD_OT_INSERT);
5374 ipd = osd_idx_ipd_get(env, bag);
5375 if (unlikely(ipd == NULL))
5378 oh = container_of0(th, struct osd_thandle, ot_super);
5379 LASSERT(oh->ot_handle != NULL);
5380 LASSERT(oh->ot_handle->h_transaction != NULL);
5381 if (S_ISDIR(obj->oo_inode->i_mode)) {
5382 iam_rec = (struct iam_rec *)oti->oti_ldp;
5383 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5385 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5386 /* pack quota uid/gid */
5387 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5388 key = (const struct dt_key *)&oti->oti_quota_id;
5389 /* pack quota record */
5390 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5391 iam_rec = (struct iam_rec *)rec;
5393 iam_rec = (struct iam_rec *)rec;
5396 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5398 osd_ipd_put(env, bag, ipd);
5399 LINVRNT(osd_invariant(obj));
5400 osd_trans_exec_check(env, th, OSD_OT_INSERT);
5405 * Calls ldiskfs_add_entry() to add directory entry
5406 * into the directory. This is required for
5407 * interoperability mode (b11826)
5409 * \retval 0, on success
5410 * \retval -ve, on error
5412 static int __osd_ea_add_rec(struct osd_thread_info *info,
5413 struct osd_object *pobj, struct inode *cinode,
5414 const char *name, const struct lu_fid *fid,
5415 struct htree_lock *hlock, struct thandle *th)
5417 struct ldiskfs_dentry_param *ldp;
5418 struct dentry *child;
5419 struct osd_thandle *oth;
5422 oth = container_of(th, struct osd_thandle, ot_super);
5423 LASSERT(oth->ot_handle != NULL);
5424 LASSERT(oth->ot_handle->h_transaction != NULL);
5425 LASSERT(pobj->oo_inode);
5427 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5428 if (unlikely(pobj->oo_inode ==
5429 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5432 osd_get_ldiskfs_dirent_param(ldp, fid);
5433 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5434 child->d_fsdata = (void *)ldp;
5435 dquot_initialize(pobj->oo_inode);
5436 rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5437 child, cinode, hlock);
5438 if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5439 struct ldiskfs_dir_entry_2 *de;
5440 struct buffer_head *bh;
5443 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5446 rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5449 if (S_ISDIR(cinode->i_mode))
5450 de->file_type = LDISKFS_DIRENT_LUFID |
5451 LDISKFS_FT_REG_FILE;
5453 de->file_type = LDISKFS_DIRENT_LUFID |
5455 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5466 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5467 * into the directory.Also sets flags into osd object to
5468 * indicate dot and dotdot are created. This is required for
5469 * interoperability mode (b11826)
5471 * \param dir directory for dot and dotdot fixup.
5472 * \param obj child object for linking
5474 * \retval 0, on success
5475 * \retval -ve, on error
5477 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5478 struct osd_object *dir,
5479 struct inode *parent_dir, const char *name,
5480 const struct lu_fid *dot_fid,
5481 const struct lu_fid *dot_dot_fid,
5484 struct inode *inode = dir->oo_inode;
5485 struct osd_thandle *oth;
5488 oth = container_of(th, struct osd_thandle, ot_super);
5489 LASSERT(oth->ot_handle->h_transaction != NULL);
5490 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5492 if (strcmp(name, dot) == 0) {
5493 if (dir->oo_compat_dot_created) {
5496 LASSERT(inode->i_ino == parent_dir->i_ino);
5497 dir->oo_compat_dot_created = 1;
5500 } else if (strcmp(name, dotdot) == 0) {
5501 if (!dir->oo_compat_dot_created)
5503 /* in case of rename, dotdot is already created */
5504 if (dir->oo_compat_dotdot_created) {
5505 return __osd_ea_add_rec(info, dir, parent_dir, name,
5506 dot_dot_fid, NULL, th);
5509 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5510 struct lu_fid tfid = *dot_dot_fid;
5513 result = osd_add_dot_dotdot_internal(info,
5514 dir->oo_inode, parent_dir, dot_fid,
5517 result = osd_add_dot_dotdot_internal(info,
5518 dir->oo_inode, parent_dir, dot_fid,
5523 dir->oo_compat_dotdot_created = 1;
5531 * It will call the appropriate osd_add* function and return the
5532 * value, return by respective functions.
5534 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5535 struct inode *cinode, const char *name,
5536 const struct lu_fid *fid, struct thandle *th)
5538 struct osd_thread_info *info = osd_oti_get(env);
5539 struct htree_lock *hlock;
5542 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5544 if (name[0] == '.' && (name[1] == '\0' ||
5545 (name[1] == '.' && name[2] == '\0'))) {
5546 if (hlock != NULL) {
5547 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5550 down_write(&pobj->oo_ext_idx_sem);
5553 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5554 lu_object_fid(&pobj->oo_dt.do_lu),
5557 if (hlock != NULL) {
5558 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5559 pobj->oo_inode, LDISKFS_HLOCK_ADD);
5561 down_write(&pobj->oo_ext_idx_sem);
5564 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5565 struct lu_fid *tfid = &info->oti_fid;
5569 rc = __osd_ea_add_rec(info, pobj, cinode, name,
5572 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5577 ldiskfs_htree_unlock(hlock);
5579 up_write(&pobj->oo_ext_idx_sem);
5585 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5586 struct osd_idmap_cache *oic)
5588 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5589 struct lu_fid *fid = &oic->oic_fid;
5590 struct osd_inode_id *id = &oic->oic_lid;
5591 struct inode *inode = NULL;
5598 if (!fid_is_norm(fid) && !fid_is_igif(fid))
5601 if (thread_is_running(&scrub->os_thread) &&
5602 scrub->os_pos_current > id->oii_ino)
5605 if (dev->od_auto_scrub_interval == AS_NEVER ||
5606 ktime_get_real_seconds() <
5607 scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5611 rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5612 if (rc == -ENOENT) {
5613 __u32 gen = id->oii_gen;
5619 inode = osd_iget(oti, dev, id);
5620 /* The inode has been removed (by race maybe). */
5621 if (IS_ERR(inode)) {
5622 rc = PTR_ERR(inode);
5624 RETURN(rc == -ESTALE ? -ENOENT : rc);
5627 /* The OI mapping is lost. */
5628 if (gen != OSD_OII_NOGEN)
5632 * The inode may has been reused by others, we do not know,
5633 * leave it to be handled by subsequent osd_fid_lookup().
5636 } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5643 if (thread_is_running(&scrub->os_thread)) {
5644 if (inode == NULL) {
5645 inode = osd_iget(oti, dev, id);
5646 /* The inode has been removed (by race maybe). */
5647 if (IS_ERR(inode)) {
5648 rc = PTR_ERR(inode);
5650 RETURN(rc == -ESTALE ? -ENOENT : rc);
5654 rc = osd_oii_insert(dev, oic, insert);
5656 * There is race condition between osd_oi_lookup and OI scrub.
5657 * The OI scrub finished just after osd_oi_lookup() failure.
5658 * Under such case, it is unnecessary to trigger OI scrub again,
5659 * but try to call osd_oi_lookup() again.
5661 if (unlikely(rc == -EAGAIN))
5664 if (!S_ISDIR(inode->i_mode))
5667 rc = osd_check_lmv(oti, dev, inode, oic);
5672 if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5673 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5674 SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5675 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5676 "%s: trigger partial OI scrub for RPC inconsistency "
5677 "checking FID "DFID": rc = %d\n",
5678 osd_dev2name(dev), PFID(fid), rc);
5679 if (rc == 0 || rc == -EALREADY)
5692 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5693 struct osd_device *dev,
5694 struct osd_idmap_cache *oic,
5695 struct lu_fid *fid, __u32 ino)
5697 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5698 struct inode *inode;
5701 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5702 inode = osd_iget(oti, dev, &oic->oic_lid);
5703 if (IS_ERR(inode)) {
5704 fid_zero(&oic->oic_fid);
5705 return PTR_ERR(inode);
5708 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5711 fid_zero(&oic->oic_fid);
5713 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5717 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5718 struct osd_inode_id *id, const struct lu_fid *fid)
5720 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5721 id->oii_ino, id->oii_gen, info);
5722 info->oti_cache.oic_lid = *id;
5723 info->oti_cache.oic_fid = *fid;
5724 info->oti_cache.oic_dev = osd;
5728 * Get parent FID from the linkEA.
5730 * For a directory which parent resides on remote MDT, to satisfy the
5731 * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5732 * the other hand, to make the lookup(..) on the directory can return
5733 * the real parent FID, we append the real parent FID after its ".."
5734 * name entry in the /REMOTE_PARENT_DIR.
5736 * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5737 * backup. So after the restore, we cannot get the right parent FID from
5738 * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5739 * we have stored the real parent FID in the directory object's linkEA,
5740 * we can parse the linkEA for the real parent FID.
5742 * \param[in] env pointer to the thread context
5743 * \param[in] obj pointer to the object to be handled
5744 * \param[out]fid pointer to the buffer to hold the parent FID
5746 * \retval 0 for getting the real parent FID successfully
5747 * \retval negative error number on failure
5749 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5750 struct osd_object *obj,
5753 struct osd_thread_info *oti = osd_oti_get(env);
5754 struct lu_buf *buf = &oti->oti_big_buf;
5755 struct dentry *dentry = &oti->oti_obj_dentry;
5756 struct inode *inode = obj->oo_inode;
5757 struct linkea_data ldata = { NULL };
5763 if (!S_ISDIR(inode->i_mode))
5767 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5768 buf->lb_buf, buf->lb_len);
5769 if (rc == -ERANGE) {
5770 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5773 lu_buf_realloc(buf, rc);
5774 if (buf->lb_buf == NULL)
5781 if (unlikely(rc == 0))
5787 if (unlikely(buf->lb_buf == NULL)) {
5788 lu_buf_realloc(buf, rc);
5789 if (buf->lb_buf == NULL)
5796 rc = linkea_init_with_rec(&ldata);
5798 linkea_first_entry(&ldata);
5799 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5805 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5806 struct inode *inode,
5807 const struct lu_fid *pfid,
5808 const char *name, const int namelen)
5810 struct osd_thread_info *oti = osd_oti_get(env);
5811 struct lu_buf *buf = &oti->oti_big_buf;
5812 struct dentry *dentry = &oti->oti_obj_dentry;
5813 struct linkea_data ldata = { NULL };
5814 struct lu_name cname = { .ln_name = name,
5815 .ln_namelen = namelen };
5821 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5822 buf->lb_buf, buf->lb_len);
5824 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5829 if (unlikely(rc == 0))
5832 if (buf->lb_len < rc) {
5833 lu_buf_realloc(buf, rc);
5834 if (buf->lb_buf == NULL)
5841 rc = linkea_init_with_rec(&ldata);
5843 rc = linkea_links_find(&ldata, &cname, pfid);
5849 * Calls ->lookup() to find dentry. From dentry get inode and
5850 * read inode's ea to get fid. This is required for interoperability
5853 * \retval 0, on success
5854 * \retval -ve, on error
5856 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5857 struct dt_rec *rec, const struct dt_key *key)
5859 struct inode *dir = obj->oo_inode;
5860 struct dentry *dentry;
5861 struct ldiskfs_dir_entry_2 *de;
5862 struct buffer_head *bh;
5863 struct lu_fid *fid = (struct lu_fid *)rec;
5864 struct htree_lock *hlock = NULL;
5870 LASSERT(dir->i_op != NULL);
5871 LASSERT(dir->i_op->lookup != NULL);
5873 dentry = osd_child_dentry_get(env, obj,
5874 (char *)key, strlen((char *)key));
5876 if (obj->oo_hl_head != NULL) {
5877 hlock = osd_oti_get(env)->oti_hlock;
5878 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5879 dir, LDISKFS_HLOCK_LOOKUP);
5881 down_read(&obj->oo_ext_idx_sem);
5884 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5886 struct osd_thread_info *oti = osd_oti_get(env);
5887 struct osd_inode_id *id = &oti->oti_id;
5888 struct osd_idmap_cache *oic = &oti->oti_cache;
5889 struct osd_device *dev = osd_obj2dev(obj);
5891 ino = le32_to_cpu(de->inode);
5892 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5894 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5898 rc = osd_get_fid_from_dentry(de, rec);
5900 /* done with de, release bh */
5903 if (unlikely(is_remote_parent_ino(dev, ino))) {
5904 const char *name = (const char *)key;
5907 * If the parent is on remote MDT, and there
5908 * is no FID-in-dirent, then we have to get
5909 * the parent FID from the linkEA.
5911 if (likely(strlen(name) == 2 &&
5912 name[0] == '.' && name[1] == '.'))
5913 rc = osd_get_pfid_from_linkea(env, obj,
5916 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5919 osd_id_gen(id, ino, OSD_OII_NOGEN);
5922 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5923 fid_zero(&oic->oic_fid);
5928 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5929 rc = osd_consistency_check(oti, dev, oic);
5931 fid_zero(&oic->oic_fid);
5933 /* Other error should not affect lookup result. */
5943 ldiskfs_htree_unlock(hlock);
5945 up_read(&obj->oo_ext_idx_sem);
5949 static int osd_index_declare_ea_insert(const struct lu_env *env,
5950 struct dt_object *dt,
5951 const struct dt_rec *rec,
5952 const struct dt_key *key,
5953 struct thandle *handle)
5955 struct osd_thandle *oh;
5956 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5957 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5958 const struct lu_fid *fid = rec1->rec_fid;
5959 int credits, rc = 0;
5960 struct osd_idmap_cache *idc;
5964 LASSERT(!dt_object_remote(dt));
5965 LASSERT(handle != NULL);
5966 LASSERT(fid != NULL);
5967 LASSERT(rec1->rec_type != 0);
5969 oh = container_of0(handle, struct osd_thandle, ot_super);
5970 LASSERT(oh->ot_handle == NULL);
5972 credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5975 * we can't call iget() while a transactions is running
5976 * (this can lead to a deadlock), but we need to know
5977 * inum and object type. so we find this information at
5978 * declaration and cache in per-thread info
5980 idc = osd_idc_find_or_init(env, osd, fid);
5982 RETURN(PTR_ERR(idc));
5983 if (idc->oic_remote) {
5985 * a reference to remote inode is represented by an
5986 * agent inode which we have to create
5988 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5989 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5992 osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5994 if (osd_dt_obj(dt)->oo_inode != NULL) {
5995 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5998 * We ignore block quota on meta pool (MDTs), so needn't
5999 * calculate how many blocks will be consumed by this index
6002 rc = osd_declare_inode_qid(env, i_uid_read(inode),
6004 i_projid_read(inode), 0,
6005 oh, osd_dt_obj(dt), NULL,
6010 #ifdef HAVE_PROJECT_QUOTA
6012 * Reserve credits for local agent inode to transfer
6013 * to 0, quota enforcement is ignored in this case.
6015 if (idc->oic_remote &&
6016 LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
6017 i_projid_read(inode) != 0)
6018 rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
6019 0, i_projid_read(inode),
6020 0, false, PRJQUOTA, true);
6028 * Index add function for interoperability mode (b11826).
6029 * It will add the directory entry.This entry is needed to
6030 * maintain name->fid mapping.
6032 * \param key it is key i.e. file entry to be inserted
6033 * \param rec it is value of given key i.e. fid
6035 * \retval 0, on success
6036 * \retval -ve, on error
6038 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6039 const struct dt_rec *rec,
6040 const struct dt_key *key, struct thandle *th)
6042 struct osd_object *obj = osd_dt_obj(dt);
6043 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6044 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6045 const struct lu_fid *fid = rec1->rec_fid;
6046 const char *name = (const char *)key;
6047 struct osd_thread_info *oti = osd_oti_get(env);
6048 struct inode *child_inode = NULL;
6049 struct osd_idmap_cache *idc;
6054 if (!dt_object_exists(dt))
6057 LASSERT(osd_invariant(obj));
6058 LASSERT(!dt_object_remote(dt));
6059 LASSERT(th != NULL);
6061 osd_trans_exec_op(env, th, OSD_OT_INSERT);
6063 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6065 idc = osd_idc_find(env, osd, fid);
6066 if (unlikely(idc == NULL)) {
6067 idc = osd_idc_find_or_init(env, osd, fid);
6070 * this dt_insert() wasn't declared properly, so
6071 * FID is missing in OI cache. we better do not
6072 * lookup FID in FLDB/OI and don't risk to deadlock,
6073 * but in some special cases (lfsck testing, etc)
6074 * it's much simpler than fixing a caller.
6076 * normally this error should be placed after the first
6077 * find, but migrate may attach source stripes to
6078 * target, which doesn't create stripes.
6080 CERROR("%s: "DFID" wasn't declared for insert\n",
6081 osd_name(osd), PFID(fid));
6083 RETURN(PTR_ERR(idc));
6087 if (idc->oic_remote) {
6088 /* Insert remote entry */
6089 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6091 igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6093 child_inode = osd_create_local_agent_inode(env, osd,
6094 obj, fid, rec1->rec_type & S_IFMT, th);
6095 if (IS_ERR(child_inode))
6096 RETURN(PTR_ERR(child_inode));
6099 /* Insert local entry */
6100 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6101 /* for a reason OI cache wasn't filled properly */
6102 CERROR("%s: OIC for "DFID" isn't filled\n",
6103 osd_name(osd), PFID(fid));
6106 child_inode = oti->oti_inode;
6107 if (unlikely(child_inode == NULL)) {
6108 struct ldiskfs_inode_info *lii;
6113 child_inode = oti->oti_inode = &lii->vfs_inode;
6115 child_inode->i_sb = osd_sb(osd);
6116 child_inode->i_ino = idc->oic_lid.oii_ino;
6117 child_inode->i_mode = rec1->rec_type & S_IFMT;
6120 rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6122 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6123 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6125 if (child_inode && child_inode != oti->oti_inode)
6127 LASSERT(osd_invariant(obj));
6128 osd_trans_exec_check(env, th, OSD_OT_INSERT);
6133 * Initialize osd Iterator for given osd index object.
6135 * \param dt osd index object
6138 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6139 struct dt_object *dt,
6142 struct osd_it_iam *it;
6143 struct osd_object *obj = osd_dt_obj(dt);
6144 struct lu_object *lo = &dt->do_lu;
6145 struct iam_path_descr *ipd;
6146 struct iam_container *bag = &obj->oo_dir->od_container;
6148 if (!dt_object_exists(dt))
6149 return ERR_PTR(-ENOENT);
6153 return ERR_PTR(-ENOMEM);
6155 ipd = osd_it_ipd_get(env, bag);
6156 if (likely(ipd != NULL)) {
6160 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6161 return (struct dt_it *)it;
6164 return ERR_PTR(-ENOMEM);
6169 * free given Iterator.
6171 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6173 struct osd_it_iam *it = (struct osd_it_iam *)di;
6174 struct osd_object *obj = it->oi_obj;
6176 iam_it_fini(&it->oi_it);
6177 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6178 osd_object_put(env, obj);
6183 * Move Iterator to record specified by \a key
6185 * \param di osd iterator
6186 * \param key key for index
6188 * \retval +ve di points to record with least key not larger than key
6189 * \retval 0 di points to exact matched key
6190 * \retval -ve failure
6193 static int osd_it_iam_get(const struct lu_env *env,
6194 struct dt_it *di, const struct dt_key *key)
6196 struct osd_thread_info *oti = osd_oti_get(env);
6197 struct osd_it_iam *it = (struct osd_it_iam *)di;
6199 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6200 /* swab quota uid/gid */
6201 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6202 key = (struct dt_key *)&oti->oti_quota_id;
6205 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6211 * \param di osd iterator
6213 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6215 struct osd_it_iam *it = (struct osd_it_iam *)di;
6217 iam_it_put(&it->oi_it);
6221 * Move iterator by one record
6223 * \param di osd iterator
6225 * \retval +1 end of container reached
6227 * \retval -ve failure
6230 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6232 struct osd_it_iam *it = (struct osd_it_iam *)di;
6234 return iam_it_next(&it->oi_it);
6238 * Return pointer to the key under iterator.
6241 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6242 const struct dt_it *di)
6244 struct osd_thread_info *oti = osd_oti_get(env);
6245 struct osd_it_iam *it = (struct osd_it_iam *)di;
6246 struct osd_object *obj = it->oi_obj;
6249 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6251 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6252 /* swab quota uid/gid */
6253 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6254 key = (struct dt_key *)&oti->oti_quota_id;
6261 * Return size of key under iterator (in bytes)
6264 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6266 struct osd_it_iam *it = (struct osd_it_iam *)di;
6268 return iam_it_key_size(&it->oi_it);
6272 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6274 /* check if file type is required */
6275 if (ent->lde_attrs & LUDA_TYPE) {
6276 struct luda_type *lt;
6277 int align = sizeof(*lt) - 1;
6279 len = (len + align) & ~align;
6280 lt = (struct luda_type *)(ent->lde_name + len);
6281 lt->lt_type = cpu_to_le16(DTTOIF(type));
6284 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6288 * build lu direct from backend fs dirent.
6292 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6293 char *name, __u16 namelen, __u16 type, __u32 attr)
6295 ent->lde_attrs = attr | LUDA_FID;
6296 fid_cpu_to_le(&ent->lde_fid, fid);
6298 ent->lde_hash = cpu_to_le64(offset);
6299 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6301 strncpy(ent->lde_name, name, namelen);
6302 ent->lde_name[namelen] = '\0';
6303 ent->lde_namelen = cpu_to_le16(namelen);
6305 /* append lustre attributes */
6306 osd_it_append_attrs(ent, namelen, type);
6310 * Return pointer to the record under iterator.
6312 static int osd_it_iam_rec(const struct lu_env *env,
6313 const struct dt_it *di,
6314 struct dt_rec *dtrec, __u32 attr)
6316 struct osd_it_iam *it = (struct osd_it_iam *)di;
6317 struct osd_thread_info *info = osd_oti_get(env);
6321 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6322 const struct osd_fid_pack *rec;
6323 struct lu_fid *fid = &info->oti_fid;
6324 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6330 name = (char *)iam_it_key_get(&it->oi_it);
6332 RETURN(PTR_ERR(name));
6334 namelen = iam_it_key_size(&it->oi_it);
6336 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6338 RETURN(PTR_ERR(rec));
6340 rc = osd_fid_unpack(fid, rec);
6344 hash = iam_it_store(&it->oi_it);
6346 /* IAM does not store object type in IAM index (dir) */
6347 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6349 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6350 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6351 (struct iam_rec *)dtrec);
6352 osd_quota_unpack(it->oi_obj, dtrec);
6354 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6355 (struct iam_rec *)dtrec);
6362 * Returns cookie for current Iterator position.
6364 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6366 struct osd_it_iam *it = (struct osd_it_iam *)di;
6368 return iam_it_store(&it->oi_it);
6372 * Restore iterator from cookie.
6374 * \param di osd iterator
6375 * \param hash Iterator location cookie
6377 * \retval +ve di points to record with least key not larger than key.
6378 * \retval 0 di points to exact matched key
6379 * \retval -ve failure
6382 static int osd_it_iam_load(const struct lu_env *env,
6383 const struct dt_it *di, __u64 hash)
6385 struct osd_it_iam *it = (struct osd_it_iam *)di;
6387 return iam_it_load(&it->oi_it, hash);
6390 static const struct dt_index_operations osd_index_iam_ops = {
6391 .dio_lookup = osd_index_iam_lookup,
6392 .dio_declare_insert = osd_index_declare_iam_insert,
6393 .dio_insert = osd_index_iam_insert,
6394 .dio_declare_delete = osd_index_declare_iam_delete,
6395 .dio_delete = osd_index_iam_delete,
6397 .init = osd_it_iam_init,
6398 .fini = osd_it_iam_fini,
6399 .get = osd_it_iam_get,
6400 .put = osd_it_iam_put,
6401 .next = osd_it_iam_next,
6402 .key = osd_it_iam_key,
6403 .key_size = osd_it_iam_key_size,
6404 .rec = osd_it_iam_rec,
6405 .store = osd_it_iam_store,
6406 .load = osd_it_iam_load
6412 * Creates or initializes iterator context.
6414 * \retval struct osd_it_ea, iterator structure on success
6417 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6418 struct dt_object *dt,
6421 struct osd_object *obj = osd_dt_obj(dt);
6422 struct osd_thread_info *info = osd_oti_get(env);
6423 struct osd_it_ea *oie;
6425 struct lu_object *lo = &dt->do_lu;
6426 struct dentry *obj_dentry;
6430 if (!dt_object_exists(dt) || obj->oo_destroyed)
6431 RETURN(ERR_PTR(-ENOENT));
6433 OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6435 RETURN(ERR_PTR(-ENOMEM));
6436 obj_dentry = &oie->oie_dentry;
6438 obj_dentry->d_inode = obj->oo_inode;
6439 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6440 obj_dentry->d_name.hash = 0;
6442 oie->oie_rd_dirent = 0;
6443 oie->oie_it_dirent = 0;
6444 oie->oie_dirent = NULL;
6445 if (unlikely(!info->oti_it_ea_buf_used)) {
6446 oie->oie_buf = info->oti_it_ea_buf;
6447 info->oti_it_ea_buf_used = 1;
6449 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6450 if (oie->oie_buf == NULL)
6451 RETURN(ERR_PTR(-ENOMEM));
6455 file = &oie->oie_file;
6457 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6458 if (attr & LUDA_64BITHASH)
6459 file->f_mode = FMODE_64BITHASH;
6461 file->f_mode = FMODE_32BITHASH;
6462 file->f_path.dentry = obj_dentry;
6463 file->f_mapping = obj->oo_inode->i_mapping;
6464 file->f_op = obj->oo_inode->i_fop;
6465 set_file_inode(file, obj->oo_inode);
6468 RETURN((struct dt_it *)oie);
6472 * Destroy or finishes iterator context.
6474 * \param di iterator structure to be destroyed
6476 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6478 struct osd_thread_info *info = osd_oti_get(env);
6479 struct osd_it_ea *oie = (struct osd_it_ea *)di;
6480 struct osd_object *obj = oie->oie_obj;
6481 struct inode *inode = obj->oo_inode;
6484 oie->oie_file.f_op->release(inode, &oie->oie_file);
6485 osd_object_put(env, obj);
6486 if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6487 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6489 info->oti_it_ea_buf_used = 0;
6490 OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6495 * It position the iterator at given key, so that next lookup continues from
6496 * that key Or it is similar to dio_it->load() but based on a key,
6497 * rather than file position.
6499 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6502 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6504 static int osd_it_ea_get(const struct lu_env *env,
6505 struct dt_it *di, const struct dt_key *key)
6507 struct osd_it_ea *it = (struct osd_it_ea *)di;
6510 LASSERT(((const char *)key)[0] == '\0');
6511 it->oie_file.f_pos = 0;
6512 it->oie_rd_dirent = 0;
6513 it->oie_it_dirent = 0;
6514 it->oie_dirent = NULL;
6522 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6526 struct osd_filldir_cbs {
6527 struct dir_context ctx;
6528 struct osd_it_ea *it;
6531 * It is called internally by ->iterate*(). It fills the
6532 * iterator's in-memory data structure with required
6533 * information i.e. name, namelen, rec_size etc.
6535 * \param buf in which information to be filled in.
6536 * \param name name of the file in given dir
6538 * \retval 0 on success
6539 * \retval 1 on buffer full
6541 #ifdef HAVE_FILLDIR_USE_CTX
6542 static int osd_ldiskfs_filldir(struct dir_context *buf,
6544 static int osd_ldiskfs_filldir(void *buf,
6546 const char *name, int namelen,
6547 loff_t offset, __u64 ino, unsigned int d_type)
6549 struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6550 struct osd_object *obj = it->oie_obj;
6551 struct osd_it_ea_dirent *ent = it->oie_dirent;
6552 struct lu_fid *fid = &ent->oied_fid;
6553 struct osd_fid_pack *rec;
6557 /* this should never happen */
6558 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6559 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6563 if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6567 /* "." is just the object itself. */
6568 if (namelen == 1 && name[0] == '.') {
6569 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6570 } else if (d_type & LDISKFS_DIRENT_LUFID) {
6571 rec = (struct osd_fid_pack *)(name + namelen + 1);
6572 if (osd_fid_unpack(fid, rec) != 0)
6577 d_type &= ~LDISKFS_DIRENT_LUFID;
6579 /* NOT export local root. */
6580 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6581 ino = obj->oo_inode->i_ino;
6582 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6585 ent->oied_ino = ino;
6586 ent->oied_off = offset;
6587 ent->oied_namelen = namelen;
6588 ent->oied_type = d_type;
6590 memcpy(ent->oied_name, name, namelen);
6592 it->oie_rd_dirent++;
6593 it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6598 * Calls ->iterate*() to load a directory entry at a time
6599 * and stored it in iterator's in-memory data structure.
6601 * \param di iterator's in memory structure
6603 * \retval 0 on success
6604 * \retval -ve on error
6605 * \retval +1 reach the end of entry
6607 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6608 const struct dt_it *di)
6610 struct osd_it_ea *it = (struct osd_it_ea *)di;
6611 struct osd_object *obj = it->oie_obj;
6612 struct inode *inode = obj->oo_inode;
6613 struct htree_lock *hlock = NULL;
6614 struct file *filp = &it->oie_file;
6616 struct osd_filldir_cbs buf = {
6617 .ctx.actor = osd_ldiskfs_filldir,
6622 it->oie_dirent = it->oie_buf;
6623 it->oie_rd_dirent = 0;
6625 if (obj->oo_hl_head != NULL) {
6626 hlock = osd_oti_get(env)->oti_hlock;
6627 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6628 inode, LDISKFS_HLOCK_READDIR);
6630 down_read(&obj->oo_ext_idx_sem);
6633 filp->f_cred = current_cred();
6634 rc = osd_security_file_alloc(filp);
6638 filp->f_flags |= O_NOATIME;
6639 filp->f_mode |= FMODE_NONOTIFY;
6640 rc = iterate_dir(filp, &buf.ctx);
6645 ldiskfs_htree_unlock(hlock);
6647 up_read(&obj->oo_ext_idx_sem);
6649 if (it->oie_rd_dirent == 0) {
6651 * If it does not get any dirent, it means it has been reached
6652 * to the end of the dir
6654 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6658 it->oie_dirent = it->oie_buf;
6659 it->oie_it_dirent = 1;
6666 * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
6667 * to load a directory entry at a time and stored it in
6668 * iterator's in-memory data structure.
6670 * \param di iterator's in memory structure
6672 * \retval +ve iterator reached to end
6673 * \retval 0 iterator not reached to end
6674 * \retval -ve on error
6676 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6678 struct osd_it_ea *it = (struct osd_it_ea *)di;
6683 if (it->oie_it_dirent < it->oie_rd_dirent) {
6685 (void *)it->oie_dirent +
6686 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6687 it->oie_dirent->oied_namelen);
6688 it->oie_it_dirent++;
6691 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6694 rc = osd_ldiskfs_it_fill(env, di);
6701 * Returns the key at current position from iterator's in memory structure.
6703 * \param di iterator's in memory structure
6705 * \retval key i.e. struct dt_key on success
6707 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6708 const struct dt_it *di)
6710 struct osd_it_ea *it = (struct osd_it_ea *)di;
6712 return (struct dt_key *)it->oie_dirent->oied_name;
6716 * Returns key's size at current position from iterator's in memory structure.
6718 * \param di iterator's in memory structure
6720 * \retval key_size i.e. struct dt_key on success
6722 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6724 struct osd_it_ea *it = (struct osd_it_ea *)di;
6726 return it->oie_dirent->oied_namelen;
6729 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6731 if (LDISKFS_DIR_REC_LEN(de) >=
6732 __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6739 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6740 unsigned int blocksize, bool dotdot)
6743 return osd_dotdot_has_space(de);
6745 if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6746 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6753 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6754 handle_t *jh, struct dentry *dentry,
6755 const struct lu_fid *fid, struct buffer_head *bh,
6756 struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6759 struct inode *dir = dentry->d_parent->d_inode;
6760 struct inode *inode = dentry->d_inode;
6761 struct osd_fid_pack *rec;
6762 struct ldiskfs_dentry_param *ldp;
6763 int namelen = dentry->d_name.len;
6765 struct osd_thread_info *info = osd_oti_get(env);
6769 if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6772 /* There is enough space to hold the FID-in-dirent. */
6773 if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6774 rc = ldiskfs_journal_get_write_access(jh, bh);
6778 de->name[namelen] = 0;
6779 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6780 rec->fp_len = sizeof(struct lu_fid) + 1;
6781 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6782 de->file_type |= LDISKFS_DIRENT_LUFID;
6783 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6790 rc = ldiskfs_delete_entry(jh, dir, de, bh);
6794 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6795 osd_get_ldiskfs_dirent_param(ldp, fid);
6796 dentry->d_fsdata = (void *)ldp;
6797 dquot_initialize(dir);
6798 rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6800 * It is too bad, we cannot reinsert the name entry back.
6801 * That means we lose it!
6804 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6805 "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6806 osd_ino2name(inode),
6807 dir->i_ino, dir->i_generation, namelen,
6808 dentry->d_name.name, PFID(fid), rc);
6814 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6815 struct osd_it_ea *it, struct lu_fid *fid,
6816 struct osd_inode_id *id, __u32 *attr)
6818 struct osd_thread_info *info = osd_oti_get(env);
6819 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6820 struct osd_device *dev = osd_obj2dev(obj);
6821 struct super_block *sb = osd_sb(dev);
6822 const char *devname = osd_name(dev);
6823 struct osd_it_ea_dirent *ent = it->oie_dirent;
6824 struct inode *dir = obj->oo_inode;
6825 struct htree_lock *hlock = NULL;
6826 struct buffer_head *bh = NULL;
6827 handle_t *jh = NULL;
6828 struct ldiskfs_dir_entry_2 *de;
6829 struct dentry *dentry;
6830 struct inode *inode;
6831 const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6834 bool dotdot = false;
6839 if (ent->oied_name[0] == '.') {
6840 if (ent->oied_namelen == 1)
6843 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6847 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6848 inode = osd_iget(info, dev, id);
6849 if (IS_ERR(inode)) {
6850 rc = PTR_ERR(inode);
6851 if (rc == -ENOENT || rc == -ESTALE) {
6853 * Maybe dangling name entry, or
6854 * corrupted directory entry.
6856 *attr |= LUDA_UNKNOWN;
6859 CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6860 "check_repair, dir = %lu/%u, name = %.*s, "
6861 "ino = %llu, rc = %d\n",
6862 devname, dir->i_ino, dir->i_generation,
6863 ent->oied_namelen, ent->oied_name,
6870 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6872 rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6873 if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6879 * We need to ensure that the name entry is still valid.
6880 * Because it may be removed or renamed by other already.
6882 * The unlink or rename operation will start journal before PDO lock,
6883 * so to avoid deadlock, here we need to start journal handle before
6884 * related PDO lock also. But because we do not know whether there
6885 * will be something to be repaired before PDO lock, we just start
6886 * journal without conditions.
6888 * We may need to remove the name entry firstly, then insert back.
6889 * One credit is for user quota file update.
6890 * One credit is for group quota file update.
6891 * Two credits are for dirty inode.
6893 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6894 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6896 if (dev->od_dirent_journal != 0) {
6899 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6902 CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6903 "check_repair, dir = %lu/%u, credits = %d, "
6904 "name = %.*s, ino = %llu: rc = %d\n",
6905 devname, dir->i_ino, dir->i_generation, credits,
6906 ent->oied_namelen, ent->oied_name,
6909 GOTO(out_inode, rc);
6912 if (obj->oo_hl_head != NULL) {
6913 hlock = osd_oti_get(env)->oti_hlock;
6915 * "0" means exclusive lock for the whole directory.
6916 * We need to prevent others access such name entry
6917 * during the delete + insert. Neither HLOCK_ADD nor
6918 * HLOCK_DEL cannot guarantee the atomicity.
6920 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6922 down_write(&obj->oo_ext_idx_sem);
6925 if (obj->oo_hl_head != NULL) {
6926 hlock = osd_oti_get(env)->oti_hlock;
6927 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6928 LDISKFS_HLOCK_LOOKUP);
6930 down_read(&obj->oo_ext_idx_sem);
6934 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6935 if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6936 *attr |= LUDA_IGNORE;
6942 * For dotdot entry, if there is not enough space to hold the
6943 * FID-in-dirent, just keep them there. It only happens when the
6944 * device upgraded from 1.8 or restored from MDT file-level backup.
6945 * For the whole directory, only dotdot entry have no FID-in-dirent
6946 * and needs to get FID from LMA when readdir, it will not affect the
6949 if (dotdot && !osd_dotdot_has_space(de)) {
6950 *attr |= LUDA_UNKNOWN;
6956 if (lu_fid_eq(fid, &lma->lma_self_fid))
6959 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6960 struct lu_fid *tfid = &lma->lma_self_fid;
6962 if (likely(dotdot &&
6963 fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6964 fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6966 * It must be REMOTE_PARENT_DIR and as the
6967 * 'dotdot' entry of remote directory
6969 *attr |= LUDA_IGNORE;
6971 CDEBUG(D_LFSCK, "%s: expect remote agent "
6972 "parent directory, but got %.*s under "
6973 "dir = %lu/%u with the FID "DFID"\n",
6974 devname, ent->oied_namelen,
6975 ent->oied_name, dir->i_ino,
6976 dir->i_generation, PFID(tfid));
6978 *attr |= LUDA_UNKNOWN;
6985 if (!fid_is_zero(fid)) {
6986 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6988 if (rc == -ENOENT ||
6990 !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
6992 * linkEA does not recognize the dirent entry,
6993 * it may because the dirent entry corruption
6994 * and points to other's inode.
6996 CDEBUG(D_LFSCK, "%s: the target inode does not "
6997 "recognize the dirent, dir = %lu/%u, "
6998 " name = %.*s, ino = %llu, "
6999 DFID": rc = %d\n", devname, dir->i_ino,
7000 dir->i_generation, ent->oied_namelen,
7001 ent->oied_name, ent->oied_ino, PFID(fid), rc);
7002 *attr |= LUDA_UNKNOWN;
7007 if (rc && rc != -ENODATA) {
7008 CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
7009 "dir = %lu/%u, name = %.*s, ino = %llu, "
7010 DFID": rc = %d\n", devname, dir->i_ino,
7011 dir->i_generation, ent->oied_namelen,
7012 ent->oied_name, ent->oied_ino, PFID(fid), rc);
7013 *attr |= LUDA_UNKNOWN;
7021 * linkEA recognizes the dirent entry, the FID-in-LMA is
7022 * valid, trusted, in spite of fid_is_sane(fid) or not.
7024 if (*attr & LUDA_VERIFY_DRYRUN) {
7025 *fid = lma->lma_self_fid;
7026 *attr |= LUDA_REPAIR;
7033 dev->od_dirent_journal = 1;
7034 if (hlock != NULL) {
7035 ldiskfs_htree_unlock(hlock);
7038 up_read(&obj->oo_ext_idx_sem);
7044 *fid = lma->lma_self_fid;
7046 /* Update or append the FID-in-dirent. */
7047 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7048 bh, de, hlock, dotdot);
7050 *attr |= LUDA_REPAIR;
7052 CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7053 "the dirent, dir = %lu/%u, name = %.*s, "
7054 "ino = %llu, "DFID": rc = %d\n",
7055 devname, dir->i_ino, dir->i_generation,
7056 ent->oied_namelen, ent->oied_name,
7057 ent->oied_ino, PFID(fid), rc);
7059 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7060 if (*attr & LUDA_VERIFY_DRYRUN) {
7061 if (fid_is_sane(fid)) {
7062 *attr |= LUDA_REPAIR;
7063 } else if (dev->od_index == 0) {
7064 lu_igif_build(fid, inode->i_ino,
7065 inode->i_generation);
7066 *attr |= LUDA_UPGRADE;
7074 dev->od_dirent_journal = 1;
7075 if (hlock != NULL) {
7076 ldiskfs_htree_unlock(hlock);
7079 up_read(&obj->oo_ext_idx_sem);
7086 if (unlikely(fid_is_sane(fid))) {
7088 * FID-in-dirent exists, but FID-in-LMA is lost.
7089 * Trust the FID-in-dirent, and add FID-in-LMA.
7091 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7093 *attr |= LUDA_REPAIR;
7095 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7096 "update dirent, dir = %lu/%u, "
7097 "name = %.*s, ino = %llu, "
7099 devname, dir->i_ino, dir->i_generation,
7100 ent->oied_namelen, ent->oied_name,
7101 ent->oied_ino, PFID(fid), rc);
7102 } else if (dev->od_index == 0) {
7103 lu_igif_build(fid, inode->i_ino, inode->i_generation);
7105 * It is probably IGIF object. Only aappend the
7106 * FID-in-dirent. OI scrub will process FID-in-LMA.
7108 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7109 bh, de, hlock, dotdot);
7111 *attr |= LUDA_UPGRADE;
7113 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7114 "after the dirent, dir = %lu/%u, "
7115 "name = %.*s, ino = %llu, "
7117 devname, dir->i_ino, dir->i_generation,
7118 ent->oied_namelen, ent->oied_name,
7119 ent->oied_ino, PFID(fid), rc);
7128 if (hlock != NULL) {
7129 ldiskfs_htree_unlock(hlock);
7131 if (dev->od_dirent_journal != 0)
7132 up_write(&obj->oo_ext_idx_sem);
7134 up_read(&obj->oo_ext_idx_sem);
7138 ldiskfs_journal_stop(jh);
7142 if (rc >= 0 && !dirty)
7143 dev->od_dirent_journal = 0;
7149 * Returns the value at current position from iterator's in memory structure.
7151 * \param di struct osd_it_ea, iterator's in memory structure
7152 * \param attr attr requested for dirent.
7153 * \param lde lustre dirent
7155 * \retval 0 no error and \param lde has correct lustre dirent.
7156 * \retval -ve on error
7158 static inline int osd_it_ea_rec(const struct lu_env *env,
7159 const struct dt_it *di,
7160 struct dt_rec *dtrec, __u32 attr)
7162 struct osd_it_ea *it = (struct osd_it_ea *)di;
7163 struct osd_object *obj = it->oie_obj;
7164 struct osd_device *dev = osd_obj2dev(obj);
7165 struct osd_thread_info *oti = osd_oti_get(env);
7166 struct osd_inode_id *id = &oti->oti_id;
7167 struct lu_fid *fid = &it->oie_dirent->oied_fid;
7168 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
7169 __u32 ino = it->oie_dirent->oied_ino;
7174 LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7176 if (attr & LUDA_VERIFY) {
7177 if (unlikely(is_remote_parent_ino(dev, ino))) {
7178 attr |= LUDA_IGNORE;
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 (!fid_is_sane(fid) &&
7185 it->oie_dirent->oied_namelen == 2 &&
7186 it->oie_dirent->oied_name[0] == '.' &&
7187 it->oie_dirent->oied_name[1] == '.')
7188 osd_get_pfid_from_linkea(env, obj, fid);
7190 rc = osd_dirent_check_repair(env, obj, it, fid, id,
7194 if (!fid_is_sane(fid))
7195 attr |= LUDA_UNKNOWN;
7197 attr &= ~LU_DIRENT_ATTRS_MASK;
7198 if (!fid_is_sane(fid)) {
7199 bool is_dotdot = false;
7201 if (it->oie_dirent->oied_namelen == 2 &&
7202 it->oie_dirent->oied_name[0] == '.' &&
7203 it->oie_dirent->oied_name[1] == '.')
7206 * If the parent is on remote MDT, and there
7207 * is no FID-in-dirent, then we have to get
7208 * the parent FID from the linkEA.
7210 if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7211 rc = osd_get_pfid_from_linkea(env, obj, fid);
7213 if (is_dotdot == false &&
7214 OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7217 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7220 osd_id_gen(id, ino, OSD_OII_NOGEN);
7224 /* Pack the entry anyway, at least the offset is right. */
7225 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7226 it->oie_dirent->oied_name,
7227 it->oie_dirent->oied_namelen,
7228 it->oie_dirent->oied_type, attr);
7233 if (osd_remote_fid(env, dev, fid))
7236 if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7237 osd_add_oi_cache(oti, dev, id, fid);
7239 RETURN(rc > 0 ? 0 : rc);
7243 * Returns the record size size at current position.
7245 * This function will return record(lu_dirent) size in bytes.
7247 * \param[in] env execution environment
7248 * \param[in] di iterator's in memory structure
7249 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
7250 * calculate the lu_dirent size.
7252 * \retval record size(in bytes & in memory) of the current lu_dirent
7255 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7258 struct osd_it_ea *it = (struct osd_it_ea *)di;
7260 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7264 * Returns a cookie for current position of the iterator head, so that
7265 * user can use this cookie to load/start the iterator next time.
7267 * \param di iterator's in memory structure
7269 * \retval cookie for current position, on success
7271 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7273 struct osd_it_ea *it = (struct osd_it_ea *)di;
7275 return it->oie_dirent->oied_off;
7279 * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
7280 * to load a directory entry at a time and stored it i inn,
7281 * in iterator's in-memory data structure.
7283 * \param di struct osd_it_ea, iterator's in memory structure
7285 * \retval +ve on success
7286 * \retval -ve on error
7288 static int osd_it_ea_load(const struct lu_env *env,
7289 const struct dt_it *di, __u64 hash)
7291 struct osd_it_ea *it = (struct osd_it_ea *)di;
7295 it->oie_file.f_pos = hash;
7297 rc = osd_ldiskfs_it_fill(env, di);
7308 * Index lookup function for interoperability mode (b11826).
7310 * \param key, key i.e. file name to be searched
7312 * \retval +ve, on success
7313 * \retval -ve, on error
7315 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7316 struct dt_rec *rec, const struct dt_key *key)
7318 struct osd_object *obj = osd_dt_obj(dt);
7323 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7324 LINVRNT(osd_invariant(obj));
7326 rc = osd_ea_lookup_rec(env, obj, rec, key);
7333 * Index and Iterator operations for interoperability
7334 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7336 static const struct dt_index_operations osd_index_ea_ops = {
7337 .dio_lookup = osd_index_ea_lookup,
7338 .dio_declare_insert = osd_index_declare_ea_insert,
7339 .dio_insert = osd_index_ea_insert,
7340 .dio_declare_delete = osd_index_declare_ea_delete,
7341 .dio_delete = osd_index_ea_delete,
7343 .init = osd_it_ea_init,
7344 .fini = osd_it_ea_fini,
7345 .get = osd_it_ea_get,
7346 .put = osd_it_ea_put,
7347 .next = osd_it_ea_next,
7348 .key = osd_it_ea_key,
7349 .key_size = osd_it_ea_key_size,
7350 .rec = osd_it_ea_rec,
7351 .rec_size = osd_it_ea_rec_size,
7352 .store = osd_it_ea_store,
7353 .load = osd_it_ea_load
7357 static void *osd_key_init(const struct lu_context *ctx,
7358 struct lu_context_key *key)
7360 struct osd_thread_info *info;
7362 OBD_ALLOC_PTR(info);
7364 return ERR_PTR(-ENOMEM);
7366 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7367 if (info->oti_it_ea_buf == NULL)
7370 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7372 info->oti_hlock = ldiskfs_htree_lock_alloc();
7373 if (info->oti_hlock == NULL)
7379 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7382 return ERR_PTR(-ENOMEM);
7385 static void osd_key_fini(const struct lu_context *ctx,
7386 struct lu_context_key *key, void *data)
7388 struct osd_thread_info *info = data;
7389 struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7390 struct osd_idmap_cache *idc = info->oti_ins_cache;
7392 if (info->oti_dio_pages) {
7394 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7395 if (info->oti_dio_pages[i])
7396 __free_page(info->oti_dio_pages[i]);
7398 OBD_FREE(info->oti_dio_pages,
7399 sizeof(struct page *) * PTLRPC_MAX_BRW_PAGES);
7402 if (info->oti_inode != NULL)
7404 if (info->oti_hlock != NULL)
7405 ldiskfs_htree_lock_free(info->oti_hlock);
7406 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7407 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7408 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7409 lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7410 lu_buf_free(&info->oti_big_buf);
7412 LASSERT(info->oti_ins_cache_size > 0);
7413 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
7414 info->oti_ins_cache = NULL;
7415 info->oti_ins_cache_size = 0;
7420 static void osd_key_exit(const struct lu_context *ctx,
7421 struct lu_context_key *key, void *data)
7423 struct osd_thread_info *info = data;
7425 LASSERT(info->oti_r_locks == 0);
7426 LASSERT(info->oti_w_locks == 0);
7427 LASSERT(info->oti_txns == 0);
7430 /* type constructor/destructor: osd_type_init, osd_type_fini */
7431 LU_TYPE_INIT_FINI(osd, &osd_key);
7433 struct lu_context_key osd_key = {
7434 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7435 .lct_init = osd_key_init,
7436 .lct_fini = osd_key_fini,
7437 .lct_exit = osd_key_exit
7441 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7442 const char *name, struct lu_device *next)
7444 struct osd_device *osd = osd_dev(d);
7446 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7447 sizeof(osd->od_svname))
7449 return osd_procfs_init(osd, name);
7452 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7454 struct seq_server_site *ss = osd_seq_site(osd);
7459 if (osd->od_is_ost || osd->od_cl_seq != NULL)
7462 if (unlikely(ss == NULL))
7465 OBD_ALLOC_PTR(osd->od_cl_seq);
7466 if (osd->od_cl_seq == NULL)
7469 rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7470 osd->od_svname, ss->ss_server_seq);
7472 OBD_FREE_PTR(osd->od_cl_seq);
7473 osd->od_cl_seq = NULL;
7477 if (ss->ss_node_id == 0) {
7479 * If the OSD on the sequence controller(MDT0), then allocate
7480 * sequence here, otherwise allocate sequence after connected
7481 * to MDT0 (see mdt_register_lwp_callback()).
7483 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7484 &osd->od_cl_seq->lcs_space, env);
7490 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7492 if (osd->od_cl_seq == NULL)
7495 seq_client_fini(osd->od_cl_seq);
7496 OBD_FREE_PTR(osd->od_cl_seq);
7497 osd->od_cl_seq = NULL;
7500 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7504 /* shutdown quota slave instance associated with the device */
7505 if (o->od_quota_slave_md != NULL) {
7506 struct qsd_instance *qsd = o->od_quota_slave_md;
7508 o->od_quota_slave_md = NULL;
7512 if (o->od_quota_slave_dt != NULL) {
7513 struct qsd_instance *qsd = o->od_quota_slave_dt;
7515 o->od_quota_slave_dt = NULL;
7519 osd_fid_fini(env, o);
7520 osd_scrub_cleanup(env, o);
7525 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7529 if (o->od_mnt != NULL) {
7530 shrink_dcache_sb(osd_sb(o));
7531 osd_sync(env, &o->od_dt_dev);
7540 static int osd_mount(const struct lu_env *env,
7541 struct osd_device *o, struct lustre_cfg *cfg)
7543 const char *name = lustre_cfg_string(cfg, 0);
7544 const char *dev = lustre_cfg_string(cfg, 1);
7546 unsigned long page, s_flags, lmd_flags = 0;
7547 struct page *__page;
7548 struct file_system_type *type;
7549 char *options = NULL;
7551 struct osd_thread_info *info = osd_oti_get(env);
7552 struct lu_fid *fid = &info->oti_fid;
7553 struct inode *inode;
7554 int rc = 0, force_over_512tb = 0;
7558 if (o->od_mnt != NULL)
7561 if (strlen(dev) >= sizeof(o->od_mntdev))
7563 strcpy(o->od_mntdev, dev);
7565 str = lustre_cfg_string(cfg, 2);
7566 s_flags = simple_strtoul(str, NULL, 0);
7567 str = strstr(str, ":");
7569 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7570 opts = lustre_cfg_string(cfg, 3);
7572 if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7573 CERROR("%s: device %s extents feature is not guaranteed to "
7574 "work on big-endian systems. Use \"bigendian_extents\" "
7575 "mount option to override.\n", name, dev);
7579 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7580 if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7581 CWARN("force_over_128tb option is deprecated. "
7582 "Filesystems less than 512TB can be created without any "
7583 "force options. Use force_over_512tb option for "
7584 "filesystems greater than 512TB.\n");
7587 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7588 if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7589 CWARN("force_over_256tb option is deprecated. "
7590 "Filesystems less than 512TB can be created without any "
7591 "force options. Use force_over_512tb option for "
7592 "filesystems greater than 512TB.\n");
7596 if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
7597 force_over_512tb = 1;
7599 __page = alloc_page(GFP_KERNEL);
7601 GOTO(out, rc = -ENOMEM);
7602 page = (unsigned long)page_address(__page);
7603 options = (char *)page;
7606 /* strip out the options for back compatiblity */
7607 static char *sout[] = {
7614 /* strip out option we processed in osd */
7615 "bigendian_extents",
7621 strncat(options, opts, PAGE_SIZE);
7622 for (rc = 0, str = options; sout[rc]; ) {
7623 char *op = strstr(str, sout[rc]);
7630 if (op == options || *(op - 1) == ',') {
7631 str = op + strlen(sout[rc]);
7632 if (*str == ',' || *str == '\0') {
7633 *str == ',' ? str++ : str;
7634 memmove(op, str, strlen(str) + 1);
7637 for (str = op; *str != ',' && *str != '\0'; str++)
7641 strncat(options, "user_xattr,acl", PAGE_SIZE);
7644 /* Glom up mount options */
7645 if (*options != '\0')
7646 strncat(options, ",", PAGE_SIZE);
7647 strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7649 type = get_fs_type("ldiskfs");
7651 CERROR("%s: cannot find ldiskfs module\n", name);
7652 GOTO(out, rc = -ENODEV);
7655 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7656 module_put(type->owner);
7658 if (IS_ERR(o->od_mnt)) {
7659 rc = PTR_ERR(o->od_mnt);
7661 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7665 if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7666 osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7667 force_over_512tb == 0) {
7668 CERROR("%s: device %s LDISKFS does not support filesystems "
7669 "greater than 512TB and can cause data corruption. "
7670 "Use \"force_over_512tb\" mount option to override.\n",
7672 GOTO(out_mnt, rc = -EINVAL);
7675 if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7676 if (priv_dev_set_rdonly) {
7677 priv_dev_set_rdonly(osd_sb(o)->s_bdev);
7678 o->od_dt_dev.dd_rdonly = 1;
7679 LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
7682 LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7685 GOTO(out_mnt, rc = -EOPNOTSUPP);
7687 } else if (priv_dev_check_rdonly &&
7688 priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7689 CERROR("%s: underlying device %s is marked as "
7690 "read-only. Setup failed\n", name, dev);
7692 GOTO(out_mnt, rc = -EROFS);
7695 if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7696 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7697 GOTO(out_mnt, rc = -EINVAL);
7700 #ifdef LDISKFS_MOUNT_DIRDATA
7701 if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7702 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7703 else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7704 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7705 "enabling the dirdata feature. If you do not want to "
7706 "downgrade to Lustre-1.x again, you can enable it via "
7707 "'tune2fs -O dirdata device'\n", name, dev);
7709 inode = osd_sb(o)->s_root->d_inode;
7710 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7711 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7713 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7717 if (lmd_flags & LMD_FLG_NOSCRUB)
7718 o->od_auto_scrub_interval = AS_NEVER;
7720 if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7721 /* do not use pagecache with flash-backed storage */
7722 o->od_writethrough_cache = 0;
7723 o->od_read_cache = 0;
7734 __free_page(__page);
7739 static struct lu_device *osd_device_fini(const struct lu_env *env,
7740 struct lu_device *d)
7742 struct osd_device *o = osd_dev(d);
7746 osd_index_backup(env, o, false);
7747 osd_shutdown(env, o);
7749 osd_obj_map_fini(o);
7755 static int osd_device_init0(const struct lu_env *env,
7756 struct osd_device *o,
7757 struct lustre_cfg *cfg)
7759 struct lu_device *l = osd2lu_dev(o);
7760 struct osd_thread_info *info;
7764 /* if the module was re-loaded, env can loose its keys */
7765 rc = lu_env_refill((struct lu_env *)env);
7768 info = osd_oti_get(env);
7771 l->ld_ops = &osd_lu_ops;
7772 o->od_dt_dev.dd_ops = &osd_dt_ops;
7774 spin_lock_init(&o->od_osfs_lock);
7775 mutex_init(&o->od_otable_mutex);
7776 INIT_LIST_HEAD(&o->od_orphan_list);
7777 INIT_LIST_HEAD(&o->od_index_backup_list);
7778 INIT_LIST_HEAD(&o->od_index_restore_list);
7779 spin_lock_init(&o->od_lock);
7780 o->od_index_backup_policy = LIBP_NONE;
7783 o->od_read_cache = 1;
7784 o->od_writethrough_cache = 1;
7785 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7786 o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20;
7787 o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20;
7788 o->od_auto_scrub_interval = AS_DEFAULT;
7790 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7791 sizeof(o->od_svname));
7792 if (cplen >= sizeof(o->od_svname)) {
7797 o->od_index_backup_stop = 0;
7798 o->od_index = -1; /* -1 means index is invalid */
7799 rc = server_name2index(o->od_svname, &o->od_index, NULL);
7800 if (rc == LDD_F_SV_TYPE_OST)
7803 o->od_full_scrub_ratio = OFSR_DEFAULT;
7804 o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7805 rc = osd_mount(env, o, cfg);
7809 /* Can only check block device after mount */
7810 o->od_nonrotational =
7811 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
7813 rc = osd_obj_map_init(env, o);
7817 rc = lu_site_init(&o->od_site, l);
7819 GOTO(out_compat, rc);
7820 o->od_site.ls_bottom_dev = l;
7822 rc = lu_site_init_finish(&o->od_site);
7826 INIT_LIST_HEAD(&o->od_ios_list);
7827 /* setup scrub, including OI files initialization */
7829 rc = osd_scrub_setup(env, o);
7834 rc = osd_procfs_init(o, o->od_svname);
7836 CERROR("%s: can't initialize procfs: rc = %d\n",
7838 GOTO(out_scrub, rc);
7841 LASSERT(l->ld_site->ls_linkage.next != NULL);
7842 LASSERT(l->ld_site->ls_linkage.prev != NULL);
7844 /* initialize quota slave instance */
7845 /* currently it's no need to prepare qsd_instance_md for OST */
7846 if (!o->od_is_ost) {
7847 o->od_quota_slave_md = qsd_init(env, o->od_svname,
7849 o->od_proc_entry, true);
7850 if (IS_ERR(o->od_quota_slave_md)) {
7851 rc = PTR_ERR(o->od_quota_slave_md);
7852 o->od_quota_slave_md = NULL;
7853 GOTO(out_procfs, rc);
7857 o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
7858 o->od_proc_entry, false);
7860 if (IS_ERR(o->od_quota_slave_dt)) {
7861 if (o->od_quota_slave_md != NULL) {
7862 qsd_fini(env, o->od_quota_slave_md);
7863 o->od_quota_slave_md = NULL;
7866 rc = PTR_ERR(o->od_quota_slave_dt);
7867 o->od_quota_slave_dt = NULL;
7868 GOTO(out_procfs, rc);
7876 osd_scrub_cleanup(env, o);
7878 lu_site_fini(&o->od_site);
7880 osd_obj_map_fini(o);
7887 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7888 struct lu_device_type *t,
7889 struct lustre_cfg *cfg)
7891 struct osd_device *o;
7896 return ERR_PTR(-ENOMEM);
7898 rc = dt_device_init(&o->od_dt_dev, t);
7901 * Because the ctx might be revived in dt_device_init,
7902 * refill the env here
7904 lu_env_refill((struct lu_env *)env);
7905 rc = osd_device_init0(env, o, cfg);
7907 dt_device_fini(&o->od_dt_dev);
7910 if (unlikely(rc != 0))
7913 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7916 static struct lu_device *osd_device_free(const struct lu_env *env,
7917 struct lu_device *d)
7919 struct osd_device *o = osd_dev(d);
7923 /* XXX: make osd top device in order to release reference */
7924 d->ld_site->ls_top_dev = d;
7925 lu_site_purge(env, d->ld_site, -1);
7926 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7927 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7928 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7930 lu_site_fini(&o->od_site);
7931 dt_device_fini(&o->od_dt_dev);
7936 static int osd_process_config(const struct lu_env *env,
7937 struct lu_device *d, struct lustre_cfg *cfg)
7939 struct osd_device *o = osd_dev(d);
7945 switch (cfg->lcfg_command) {
7947 rc = osd_mount(env, o, cfg);
7951 * For the case LCFG_PRE_CLEANUP is not called in advance,
7952 * that may happend if hit failure during mount process.
7954 osd_index_backup(env, o, false);
7955 lu_dev_del_linkage(d->ld_site, d);
7956 rc = osd_shutdown(env, o);
7959 LASSERT(&o->od_dt_dev);
7960 count = class_modify_config(cfg, PARAM_OSD,
7961 &o->od_dt_dev.dd_kobj);
7963 count = class_modify_config(cfg, PARAM_OST,
7964 &o->od_dt_dev.dd_kobj);
7965 rc = count > 0 ? 0 : count;
7967 case LCFG_PRE_CLEANUP:
7969 osd_index_backup(env, o,
7970 o->od_index_backup_policy != LIBP_NONE);
7980 static int osd_recovery_complete(const struct lu_env *env,
7981 struct lu_device *d)
7983 struct osd_device *osd = osd_dev(d);
7988 if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
7992 * start qsd instance on recovery completion, this notifies the quota
7993 * slave code that we are about to process new requests now
7995 rc = qsd_start(env, osd->od_quota_slave_dt);
7996 if (rc == 0 && osd->od_quota_slave_md != NULL)
7997 rc = qsd_start(env, osd->od_quota_slave_md);
8003 * we use exports to track all osd users
8005 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
8006 struct obd_device *obd, struct obd_uuid *cluuid,
8007 struct obd_connect_data *data, void *localdata)
8009 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8010 struct lustre_handle conn;
8015 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
8017 rc = class_connect(&conn, obd, cluuid);
8021 *exp = class_conn2export(&conn);
8023 spin_lock(&osd->od_osfs_lock);
8025 spin_unlock(&osd->od_osfs_lock);
8031 * once last export (we don't count self-export) disappeared
8032 * osd can be released
8034 static int osd_obd_disconnect(struct obd_export *exp)
8036 struct obd_device *obd = exp->exp_obd;
8037 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8038 int rc, release = 0;
8042 /* Only disconnect the underlying layers on the final disconnect. */
8043 spin_lock(&osd->od_osfs_lock);
8045 if (osd->od_connects == 0)
8047 spin_unlock(&osd->od_osfs_lock);
8049 rc = class_disconnect(exp); /* bz 9811 */
8051 if (rc == 0 && release)
8052 class_manual_cleanup(obd);
8056 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8057 struct lu_device *dev)
8059 struct osd_device *osd = osd_dev(dev);
8060 struct lr_server_data *lsd =
8061 &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8066 if (osd->od_quota_slave_md != NULL) {
8067 /* set up quota slave objects for inode */
8068 result = qsd_prepare(env, osd->od_quota_slave_md);
8073 if (osd->od_quota_slave_dt != NULL) {
8074 /* set up quota slave objects for block */
8075 result = qsd_prepare(env, osd->od_quota_slave_dt);
8081 if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8082 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8083 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8084 osd->od_index_in_idif = 1;
8086 osd->od_index_in_idif = 0;
8087 result = osd_register_proc_index_in_idif(osd);
8092 osd->od_index_in_idif = 1;
8096 result = osd_fid_init(env, osd);
8101 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
8102 struct lu_fid *fid, struct md_op_data *op_data)
8104 struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
8106 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8109 static const struct lu_object_operations osd_lu_obj_ops = {
8110 .loo_object_init = osd_object_init,
8111 .loo_object_delete = osd_object_delete,
8112 .loo_object_release = osd_object_release,
8113 .loo_object_free = osd_object_free,
8114 .loo_object_print = osd_object_print,
8115 .loo_object_invariant = osd_object_invariant
8118 const struct lu_device_operations osd_lu_ops = {
8119 .ldo_object_alloc = osd_object_alloc,
8120 .ldo_process_config = osd_process_config,
8121 .ldo_recovery_complete = osd_recovery_complete,
8122 .ldo_prepare = osd_prepare,
8125 static const struct lu_device_type_operations osd_device_type_ops = {
8126 .ldto_init = osd_type_init,
8127 .ldto_fini = osd_type_fini,
8129 .ldto_start = osd_type_start,
8130 .ldto_stop = osd_type_stop,
8132 .ldto_device_alloc = osd_device_alloc,
8133 .ldto_device_free = osd_device_free,
8135 .ldto_device_init = osd_device_init,
8136 .ldto_device_fini = osd_device_fini
8139 static struct lu_device_type osd_device_type = {
8140 .ldt_tags = LU_DEVICE_DT,
8141 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
8142 .ldt_ops = &osd_device_type_ops,
8143 .ldt_ctx_tags = LCT_LOCAL,
8146 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8148 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8149 struct super_block *sb = osd_sb(osd);
8151 return (osd->od_mnt == NULL || sb->s_flags & SB_RDONLY);
8155 * lprocfs legacy support.
8157 static const struct obd_ops osd_obd_device_ops = {
8158 .o_owner = THIS_MODULE,
8159 .o_connect = osd_obd_connect,
8160 .o_disconnect = osd_obd_disconnect,
8161 .o_fid_alloc = osd_fid_alloc,
8162 .o_health_check = osd_health_check,
8165 static ssize_t track_declares_assert_show(struct kobject *kobj,
8166 struct attribute *attr,
8169 return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8172 static ssize_t track_declares_assert_store(struct kobject *kobj,
8173 struct attribute *attr,
8174 const char *buffer, size_t count)
8176 bool track_declares_assert;
8179 rc = kstrtobool(buffer, &track_declares_assert);
8183 ldiskfs_track_declares_assert = track_declares_assert;
8187 LUSTRE_RW_ATTR(track_declares_assert);
8189 static int __init osd_init(void)
8191 struct kobject *kobj;
8194 BUILD_BUG_ON(BH_DXLock >=
8195 sizeof(((struct buffer_head *)0)->b_state) * 8);
8196 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8197 /* please, try to keep osd_thread_info smaller than a page */
8198 BUILD_BUG_ON(sizeof(struct osd_thread_info) > PAGE_SIZE);
8203 rc = lu_kmem_init(ldiskfs_caches);
8207 #ifdef CONFIG_KALLSYMS
8208 priv_security_file_alloc =
8209 (void *)kallsyms_lookup_name("security_file_alloc");
8210 priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
8211 priv_dev_check_rdonly =
8212 (void *)kallsyms_lookup_name("dev_check_rdonly");
8215 rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL,
8216 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8218 lu_kmem_fini(ldiskfs_caches);
8222 kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8224 rc = sysfs_create_file(kobj,
8225 &lustre_attr_track_declares_assert.attr);
8228 CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8235 static void __exit osd_exit(void)
8237 struct kobject *kobj;
8239 kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8241 sysfs_remove_file(kobj,
8242 &lustre_attr_track_declares_assert.attr);
8245 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8246 lu_kmem_fini(ldiskfs_caches);
8249 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8250 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8251 MODULE_VERSION(LUSTRE_VERSION_STRING);
8252 MODULE_LICENSE("GPL");
8254 module_init(osd_init);
8255 module_exit(osd_exit);