4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/osd/osd_handler.c
34 * Top-level entry points into osd module
36 * Author: Nikita Danilov <nikita@clusterfs.com>
37 * Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
40 #define DEBUG_SUBSYSTEM S_OSD
42 #include <linux/kallsyms.h>
43 #include <linux/module.h>
44 #include <linux/user_namespace.h>
45 #include <linux/uidgid.h>
47 /* prerequisite for linux/xattr.h */
48 #include <linux/types.h>
49 /* prerequisite for linux/xattr.h */
51 /* XATTR_{REPLACE,CREATE} */
52 #include <linux/xattr.h>
54 #include <ldiskfs/ldiskfs.h>
55 #include <ldiskfs/xattr.h>
56 #include <ldiskfs/ldiskfs_extents.h>
59 * struct OBD_{ALLOC,FREE}*()
62 #include <obd_support.h>
63 /* struct ptlrpc_thread */
64 #include <lustre_net.h>
65 #include <lustre_fid.h>
67 #include <uapi/linux/lustre/lustre_param.h>
69 #include "osd_internal.h"
70 #include "osd_dynlocks.h"
72 /* llo_* api support */
73 #include <md_object.h>
74 #include <lustre_quota.h>
76 #include <lustre_linkea.h>
78 /* Maximum EA size is limited by LNET_MTU for remote objects */
79 #define OSD_MAX_EA_SIZE 1048364
82 module_param(ldiskfs_pdo, int, 0644);
83 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
85 int ldiskfs_track_declares_assert;
86 module_param(ldiskfs_track_declares_assert, int, 0644);
87 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
89 /* Slab to allocate dynlocks */
90 struct kmem_cache *dynlock_cachep;
92 /* Slab to allocate osd_it_ea */
93 struct kmem_cache *osd_itea_cachep;
95 static struct lu_kmem_descr ldiskfs_caches[] = {
97 .ckd_cache = &dynlock_cachep,
98 .ckd_name = "dynlock_cache",
99 .ckd_size = sizeof(struct dynlock_handle)
102 .ckd_cache = &osd_itea_cachep,
103 .ckd_name = "osd_itea_cache",
104 .ckd_size = sizeof(struct osd_it_ea)
111 static const char dot[] = ".";
112 static const char dotdot[] = "..";
114 static const struct lu_object_operations osd_lu_obj_ops;
115 static const struct dt_object_operations osd_obj_ops;
116 static const struct dt_object_operations osd_obj_otable_it_ops;
117 static const struct dt_index_operations osd_index_iam_ops;
118 static const struct dt_index_operations osd_index_ea_ops;
120 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
121 const struct lu_fid *fid);
122 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
123 struct osd_device *osd);
125 int osd_trans_declare_op2rb[] = {
126 [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
127 [OSD_OT_PUNCH] = OSD_OT_MAX,
128 [OSD_OT_XATTR_SET] = OSD_OT_XATTR_SET,
129 [OSD_OT_CREATE] = OSD_OT_DESTROY,
130 [OSD_OT_DESTROY] = OSD_OT_CREATE,
131 [OSD_OT_REF_ADD] = OSD_OT_REF_DEL,
132 [OSD_OT_REF_DEL] = OSD_OT_REF_ADD,
133 [OSD_OT_WRITE] = OSD_OT_WRITE,
134 [OSD_OT_INSERT] = OSD_OT_DELETE,
135 [OSD_OT_DELETE] = OSD_OT_INSERT,
136 [OSD_OT_QUOTA] = OSD_OT_MAX,
139 static int osd_has_index(const struct osd_object *obj)
141 return obj->oo_dt.do_index_ops != NULL;
144 static int osd_object_invariant(const struct lu_object *l)
146 return osd_invariant(osd_obj(l));
150 * Concurrency: doesn't matter
152 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
154 struct osd_thread_info *oti = osd_oti_get(env);
156 return oti->oti_w_locks > 0 && o->oo_owner == env;
160 * Concurrency: doesn't access mutable data
162 static int osd_root_get(const struct lu_env *env,
163 struct dt_device *dev, struct lu_fid *f)
165 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
170 * the following set of functions are used to maintain per-thread
171 * cache of FID->ino mapping. this mechanism is needed to resolve
172 * FID to inode at dt_insert() which in turn stores ino in the
173 * directory entries to keep ldiskfs compatible with ext[34].
174 * due to locking-originated restrictions we can't lookup ino
175 * using LU cache (deadlock is possible). lookup using OI is quite
176 * expensive. so instead we maintain this cache and methods like
177 * dt_create() fill it. so in the majority of cases dt_insert() is
178 * able to find needed mapping in lockless manner.
180 static struct osd_idmap_cache *
181 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
182 const struct lu_fid *fid)
184 struct osd_thread_info *oti = osd_oti_get(env);
185 struct osd_idmap_cache *idc = oti->oti_ins_cache;
188 for (i = 0; i < oti->oti_ins_cache_used; i++) {
189 if (!lu_fid_eq(&idc[i].oic_fid, fid))
191 if (idc[i].oic_dev != osd)
200 static struct osd_idmap_cache *
201 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
202 const struct lu_fid *fid)
204 struct osd_thread_info *oti = osd_oti_get(env);
205 struct osd_idmap_cache *idc;
208 if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
209 i = oti->oti_ins_cache_size * 2;
211 i = OSD_INS_CACHE_SIZE;
212 OBD_ALLOC_PTR_ARRAY(idc, i);
214 return ERR_PTR(-ENOMEM);
215 if (oti->oti_ins_cache != NULL) {
216 memcpy(idc, oti->oti_ins_cache,
217 oti->oti_ins_cache_used * sizeof(*idc));
218 OBD_FREE_PTR_ARRAY(oti->oti_ins_cache,
219 oti->oti_ins_cache_used);
221 oti->oti_ins_cache = idc;
222 oti->oti_ins_cache_size = i;
225 idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
228 idc->oic_lid.oii_ino = 0;
229 idc->oic_lid.oii_gen = 0;
236 * lookup mapping for the given fid in the cache, initialize a
237 * new one if not found. the initialization checks whether the
238 * object is local or remote. for local objects, OI is used to
239 * learn ino/generation. the function is used when the caller
240 * has no information about the object, e.g. at dt_insert().
242 static struct osd_idmap_cache *
243 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
244 const struct lu_fid *fid)
246 struct osd_idmap_cache *idc;
249 idc = osd_idc_find(env, osd, fid);
250 LASSERT(!IS_ERR(idc));
254 CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
255 osd->od_svname, PFID(fid));
257 /* new mapping is needed */
258 idc = osd_idc_add(env, osd, fid);
260 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
261 osd->od_svname, PFID(fid), PTR_ERR(idc));
266 rc = osd_remote_fid(env, osd, fid);
267 if (unlikely(rc < 0))
271 /* the object is local, lookup in OI */
272 /* XXX: probably cheaper to lookup in LU first? */
273 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
275 if (unlikely(rc < 0)) {
276 CERROR("can't lookup: rc = %d\n", rc);
280 /* the object is remote */
288 * lookup mapping for given FID and fill it from the given object.
289 * the object is lolcal by definition.
291 static int osd_idc_find_and_init(const struct lu_env *env,
292 struct osd_device *osd,
293 struct osd_object *obj)
295 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
296 struct osd_idmap_cache *idc;
298 idc = osd_idc_find(env, osd, fid);
299 LASSERT(!IS_ERR(idc));
301 if (obj->oo_inode == NULL)
303 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
304 LASSERT(idc->oic_lid.oii_ino == 0);
305 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
306 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
311 CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
312 osd->od_svname, PFID(fid));
314 /* new mapping is needed */
315 idc = osd_idc_add(env, osd, fid);
317 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
318 osd->od_svname, PFID(fid), PTR_ERR(idc));
322 if (obj->oo_inode != NULL) {
323 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
324 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
330 * OSD object methods.
334 * Concurrency: no concurrent access is possible that early in object
337 static struct lu_object *osd_object_alloc(const struct lu_env *env,
338 const struct lu_object_header *hdr,
341 struct osd_object *mo;
346 struct lu_object_header *h;
347 struct osd_device *o = osd_dev(d);
349 l = &mo->oo_dt.do_lu;
350 if (unlikely(o->od_in_init)) {
357 lu_object_header_init(h);
358 lu_object_init(l, h, d);
359 lu_object_add_top(h, l);
362 dt_object_init(&mo->oo_dt, NULL, d);
363 mo->oo_header = NULL;
366 mo->oo_dt.do_ops = &osd_obj_ops;
367 l->lo_ops = &osd_lu_obj_ops;
368 init_rwsem(&mo->oo_sem);
369 init_rwsem(&mo->oo_ext_idx_sem);
370 spin_lock_init(&mo->oo_guard);
371 INIT_LIST_HEAD(&mo->oo_xattr_list);
377 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
378 struct dentry *dentry, struct lustre_ost_attrs *loa)
382 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
383 (void *)loa, sizeof(*loa));
385 struct lustre_mdt_attrs *lma = &loa->loa_lma;
387 if (rc < sizeof(*lma))
391 lustre_loa_swab(loa, true);
392 /* Check LMA compatibility */
393 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
394 CWARN("%s: unsupported incompat LMA feature(s) %#x "
395 "for fid = "DFID", ino = %lu\n",
397 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
398 PFID(&lma->lma_self_fid), inode->i_ino);
401 } else if (rc == 0) {
409 * retrieve object from backend ext fs.
411 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
412 struct osd_inode_id *id)
415 struct inode *inode = NULL;
418 * if we look for an inode withing a running
419 * transaction, then we risk to deadlock
420 * osd_dirent_check_repair() breaks this
422 /* LASSERT(current->journal_info == NULL); */
424 inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
426 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
427 id->oii_ino, PTR_ERR(inode));
428 } else if (id->oii_gen != OSD_OII_NOGEN &&
429 inode->i_generation != id->oii_gen) {
430 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
431 "i_generation = %u\n",
432 id->oii_ino, id->oii_gen, inode->i_generation);
434 inode = ERR_PTR(-ESTALE);
435 } else if (inode->i_nlink == 0) {
437 * due to parallel readdir and unlink,
438 * we can have dead inode here.
440 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
442 inode = ERR_PTR(-ESTALE);
443 } else if (is_bad_inode(inode)) {
444 CWARN("%s: bad inode: ino = %u\n",
445 osd_dev2name(dev), id->oii_ino);
447 inode = ERR_PTR(-ENOENT);
448 } else if ((rc = osd_attach_jinode(inode))) {
452 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
453 if (id->oii_gen == OSD_OII_NOGEN)
454 osd_id_gen(id, inode->i_ino, inode->i_generation);
457 * Do not update file c/mtime in ldiskfs.
458 * NB: we don't have any lock to protect this because we don't
459 * have reference on osd_object now, but contention with
460 * another lookup + attr_set can't happen in the tiny window
461 * between if (...) and set S_NOCMTIME.
463 if (!(inode->i_flags & S_NOCMTIME))
464 inode->i_flags |= S_NOCMTIME;
469 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
470 handle_t *handle, struct dentry *child,
471 struct inode *inode, struct htree_lock *hlock)
475 rc = __ldiskfs_add_entry(handle, child, inode, hlock);
476 if (rc == -ENOBUFS || rc == -ENOSPC) {
477 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
478 struct inode *parent = child->d_parent->d_inode;
479 struct lu_fid *fid = NULL;
480 char fidstr[FID_LEN + 1] = "unknown";
482 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
484 fid = &loa->loa_lma.lma_self_fid;
485 } else if (rc2 == -ENODATA) {
486 if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
487 fid = &info->oti_fid3;
488 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
489 } else if (!osd->od_is_ost && osd->od_index == 0) {
490 fid = &info->oti_fid3;
491 lu_igif_build(fid, parent->i_ino,
492 parent->i_generation);
497 snprintf(fidstr, sizeof(fidstr), DFID, PFID(fid));
499 /* below message is checked in sanity.sh test_129 */
501 CWARN("%s: directory (inode: %lu, FID: %s) has reached max size limit\n",
502 osd_name(osd), parent->i_ino, fidstr);
504 rc = 0; /* ignore such error now */
505 CWARN("%s: directory (inode: %lu, FID: %s) is approaching max size limit\n",
506 osd_name(osd), parent->i_ino, fidstr);
516 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
517 struct osd_inode_id *id, struct lu_fid *fid)
519 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
523 inode = osd_iget(info, dev, id);
527 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
529 *fid = loa->loa_lma.lma_self_fid;
530 } else if (rc == -ENODATA) {
531 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
532 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
534 lu_igif_build(fid, inode->i_ino, inode->i_generation);
542 static struct inode *osd_iget_check(struct osd_thread_info *info,
543 struct osd_device *dev,
544 const struct lu_fid *fid,
545 struct osd_inode_id *id,
554 * The cached OI mapping is trustable. If we cannot locate the inode
555 * via the cached OI mapping, then return the failure to the caller
556 * directly without further OI checking.
560 inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
563 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
566 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
567 PFID(fid), id->oii_ino, rc);
571 if (is_bad_inode(inode)) {
576 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
577 PFID(fid), id->oii_ino);
581 if (id->oii_gen != OSD_OII_NOGEN &&
582 inode->i_generation != id->oii_gen) {
587 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
588 "oii_gen = %u, i_generation = %u\n", PFID(fid),
589 id->oii_ino, id->oii_gen, inode->i_generation);
593 if (inode->i_nlink == 0) {
598 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
599 PFID(fid), id->oii_ino);
603 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
607 __u32 saved_ino = id->oii_ino;
608 __u32 saved_gen = id->oii_gen;
611 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
613 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
615 * XXX: There are four possible cases:
617 * Backup/restore caused the OI invalid.
619 * Someone unlinked the object but NOT removed
620 * the OI mapping, such as mount target device
621 * as ldiskfs, and modify something directly.
623 * Someone just removed the object between the
624 * former oi_lookup and the iget. It is normal.
625 * 4. Other failure cases.
627 * Generally, when the device is mounted, it will
628 * auto check whether the system is restored from
629 * file-level backup or not. We trust such detect
630 * to distinguish the 1st case from the 2nd case:
631 * if the OI files are consistent but may contain
632 * stale OI mappings because of case 2, if iget()
633 * returns -ENOENT or -ESTALE, then it should be
638 * If the OI mapping was in OI file before the
639 * osd_iget_check(), but now, it is disappear,
640 * then it must be removed by race. That is a
646 * It is the OI scrub updated the OI mapping by race.
647 * The new OI mapping must be valid.
649 if (saved_ino != id->oii_ino ||
650 (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
659 if (dev->od_scrub.os_scrub.os_file.sf_flags &
662 * It still can be the case 2, but we cannot
663 * distinguish it from the case 1. So return
664 * -EREMCHG to block current operation until
665 * OI scrub rebuilt the OI mappings.
674 if (inode->i_generation == id->oii_gen)
679 if (id->oii_gen == OSD_OII_NOGEN)
680 osd_id_gen(id, inode->i_ino, inode->i_generation);
683 * Do not update file c/mtime in ldiskfs.
684 * NB: we don't have any lock to protect this because we don't
685 * have reference on osd_object now, but contention with
686 * another lookup + attr_set can't happen in the tiny window
687 * between if (...) and set S_NOCMTIME.
689 if (!(inode->i_flags & S_NOCMTIME))
690 inode->i_flags |= S_NOCMTIME;
707 * \retval +v: new filter_fid does not contain self-fid
708 * \retval 0: filter_fid_18_23, contains self-fid
709 * \retval -v: other failure cases
711 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
712 struct dentry *dentry, struct lu_fid *fid)
714 struct filter_fid *ff = &info->oti_ff;
715 struct ost_id *ostid = &info->oti_ostid;
718 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
719 if (rc == sizeof(struct filter_fid_18_23)) {
720 struct filter_fid_18_23 *ff_old = (void *)ff;
722 ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq));
723 rc = ostid_set_id(ostid, le64_to_cpu(ff_old->ff_objid));
725 * XXX: use 0 as the index for compatibility, the caller will
726 * handle index related issues when necessary.
729 ostid_to_fid(fid, ostid, 0);
730 } else if (rc >= (int)sizeof(struct filter_fid_24_29)) {
732 } else if (rc >= 0) {
739 static int osd_lma_self_repair(struct osd_thread_info *info,
740 struct osd_device *osd, struct inode *inode,
741 const struct lu_fid *fid, __u32 compat)
746 LASSERT(current->journal_info == NULL);
748 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
749 osd_dto_credits_noquota[DTO_XATTR_SET]);
752 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
757 rc = osd_ea_fid_set(info, inode, fid, compat, 0);
759 CWARN("%s: cannot self repair the LMA: rc = %d\n",
761 ldiskfs_journal_stop(jh);
765 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
767 struct osd_thread_info *info = osd_oti_get(env);
768 struct osd_device *osd = osd_obj2dev(obj);
769 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
770 struct lustre_mdt_attrs *lma = &loa->loa_lma;
771 struct inode *inode = obj->oo_inode;
772 struct dentry *dentry = &info->oti_obj_dentry;
773 struct lu_fid *fid = NULL;
774 const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
779 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
780 (void *)loa, sizeof(*loa));
781 if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
782 fid = &lma->lma_self_fid;
783 rc = osd_get_idif(info, inode, dentry, fid);
784 if (rc > 0 || (rc == -ENODATA && osd->od_index_in_idif)) {
786 * For the given OST-object, if it has neither LMA nor
787 * FID in XATTR_NAME_FID, then the given FID (which is
788 * contained in the @obj, from client RPC for locating
789 * the OST-object) is trusted. We use it to generate
792 osd_lma_self_repair(info, osd, inode, rfid,
803 lustre_lma_swab(lma);
804 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
805 (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
806 S_ISREG(inode->i_mode)))) {
807 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
808 "fid = "DFID", ino = %lu\n", osd_name(osd),
809 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
810 PFID(rfid), inode->i_ino);
813 fid = &lma->lma_self_fid;
814 if (lma->lma_compat & LMAC_STRIPE_INFO &&
816 obj->oo_pfid_in_lma = 1;
817 if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
819 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
823 if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
824 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
825 struct ost_id *oi = &info->oti_ostid;
826 struct lu_fid *fid1 = &info->oti_fid3;
827 __u32 idx = fid_idif_ost_idx(rfid);
830 * For old IDIF, the OST index is not part of the IDIF,
831 * Means that different OSTs may have the same IDIFs.
832 * Under such case, we need to make some compatible
833 * check to make sure to trigger OI scrub properly.
835 if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
836 /* Given @rfid is new, LMA is old. */
837 fid_to_ostid(fid, oi);
838 ostid_to_fid(fid1, oi, idx);
839 if (lu_fid_eq(fid1, rfid)) {
840 if (osd->od_index_in_idif)
841 osd_lma_self_repair(info, osd,
855 struct osd_check_lmv_buf {
856 /* please keep it as first member */
857 struct dir_context ctx;
858 struct osd_thread_info *oclb_info;
859 struct osd_device *oclb_dev;
860 struct osd_idmap_cache *oclb_oic;
866 * It is called internally by ->iterate*() to filter out the
867 * local slave object's FID of the striped directory.
869 * \retval 1 found the local slave's FID
870 * \retval 0 continue to check next item
871 * \retval -ve for failure
873 #ifdef HAVE_FILLDIR_USE_CTX
874 static int osd_stripe_dir_filldir(struct dir_context *buf,
876 static int osd_stripe_dir_filldir(void *buf,
878 const char *name, int namelen,
879 loff_t offset, __u64 ino, unsigned int d_type)
881 struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
882 struct osd_thread_info *oti = oclb->oclb_info;
883 struct lu_fid *fid = &oti->oti_fid3;
884 struct osd_inode_id *id = &oti->oti_id3;
885 struct osd_device *dev = oclb->oclb_dev;
886 struct osd_idmap_cache *oic = oclb->oclb_oic;
895 sscanf(name + 1, SFID, RFID(fid));
896 if (!fid_is_sane(fid))
899 if (osd_remote_fid(oti->oti_env, dev, fid))
902 osd_id_gen(id, ino, OSD_OII_NOGEN);
903 inode = osd_iget(oti, dev, id);
905 return PTR_ERR(inode);
908 osd_add_oi_cache(oti, dev, id, fid);
912 osd_oii_insert(dev, oic, true);
913 oclb->oclb_found = true;
919 * When lookup item under striped directory, we need to locate the master
920 * MDT-object of the striped directory firstly, then the client will send
921 * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
922 * and the item's name. If the system is restored from MDT file level backup,
923 * then before the OI scrub completely built the OI files, the OI mappings of
924 * the master MDT-object and slave MDT-object may be invalid. Usually, it is
925 * not a problem for the master MDT-object. Because when locate the master
926 * MDT-object, we will do name based lookup (for the striped directory itself)
927 * firstly, during such process we can setup the correct OI mapping for the
928 * master MDT-object. But it will be trouble for the slave MDT-object. Because
929 * the client will not trigger name based lookup on the MDT to locate the slave
930 * MDT-object before locating item under the striped directory, then when
931 * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
932 * is invalid and does not know what the right OI mapping is, then the MDT has
933 * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
934 * the OI file, related OI mapping is unknown yet, please try again later. And
935 * then client will re-try the RPC again and again until related OI mapping has
936 * been updated. That is quite inefficient.
938 * To resolve above trouble, we will handle it as the following two cases:
940 * 1) The slave MDT-object and the master MDT-object are on different MDTs.
941 * It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
942 * is linked under /REMOTE_PARENT_DIR with the name of its FID string.
943 * We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
944 * directly. Please check osd_fid_lookup().
946 * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
947 * Under such case, during lookup the master MDT-object, we will lookup the
948 * slave MDT-object via readdir against the master MDT-object, because the
949 * slave MDT-objects information are stored as sub-directories with the name
950 * "${FID}:${index}". Then when find the local slave MDT-object, its OI
951 * mapping will be recorded. Then subsequent osd_fid_lookup() will know
952 * the correct OI mapping for the slave MDT-object.
954 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
955 struct inode *inode, struct osd_idmap_cache *oic)
957 struct lu_buf *buf = &oti->oti_big_buf;
958 struct dentry *dentry = &oti->oti_obj_dentry;
959 struct file *filp = &oti->oti_file;
960 const struct file_operations *fops;
961 struct lmv_mds_md_v1 *lmv1;
962 struct osd_check_lmv_buf oclb = {
963 .ctx.actor = osd_stripe_dir_filldir,
974 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
977 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
979 lu_buf_realloc(buf, rc);
980 if (buf->lb_buf == NULL)
981 GOTO(out, rc = -ENOMEM);
987 if (unlikely(rc == 0 || rc == -ENODATA))
993 if (unlikely(buf->lb_buf == NULL)) {
994 lu_buf_realloc(buf, rc);
995 if (buf->lb_buf == NULL)
996 GOTO(out, rc = -ENOMEM);
1002 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1005 fops = inode->i_fop;
1006 dentry->d_inode = inode;
1007 dentry->d_sb = inode->i_sb;
1009 filp->f_path.dentry = dentry;
1010 filp->f_flags |= O_NOATIME;
1011 filp->f_mode = FMODE_64BITHASH | FMODE_NONOTIFY;
1012 filp->f_mapping = inode->i_mapping;
1014 filp->private_data = NULL;
1015 filp->f_cred = current_cred();
1016 filp->f_inode = inode;
1017 rc = osd_security_file_alloc(filp);
1022 oclb.oclb_items = 0;
1023 rc = iterate_dir(filp, &oclb.ctx);
1024 } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1025 filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1026 fops->release(inode, filp);
1030 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
1031 DFID": rc = %d\n", osd_ino2name(inode),
1032 inode->i_ino, inode->i_generation,
1033 PFID(&oic->oic_fid), rc);
1040 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1041 const struct lu_fid *fid,
1042 const struct lu_object_conf *conf)
1044 struct osd_thread_info *info;
1045 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1046 struct osd_device *dev;
1047 struct osd_idmap_cache *oic;
1048 struct osd_inode_id *id;
1049 struct inode *inode = NULL;
1050 struct lustre_scrub *scrub;
1051 struct scrub_file *sf;
1052 __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1057 bool remote = false;
1058 bool trusted = true;
1059 bool updated = false;
1060 bool checked = false;
1064 LINVRNT(osd_invariant(obj));
1065 LASSERT(obj->oo_inode == NULL);
1066 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
1068 dev = osd_dev(ldev);
1069 scrub = &dev->od_scrub.os_scrub;
1070 sf = &scrub->os_file;
1071 info = osd_oti_get(env);
1073 oic = &info->oti_cache;
1075 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1079 * For the object is created as locking anchor, or for the object to
1080 * be created on disk. No need to osd_oi_lookup() at here because FID
1081 * shouldn't never be re-used, if it's really a duplicate FID from
1082 * unexpected reason, we should be able to detect it later by calling
1083 * do_create->osd_oi_insert().
1085 if (conf && conf->loc_flags & LOC_F_NEW)
1086 GOTO(out, result = 0);
1088 /* Search order: 1. per-thread cache. */
1089 if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1095 if (!list_empty(&scrub->os_inconsistent_items)) {
1096 /* Search order: 2. OI scrub pending list. */
1097 result = osd_oii_lookup(dev, fid, id);
1103 * The OI mapping in the OI file can be updated by the OI scrub
1104 * when we locate the inode via FID. So it may be not trustable.
1108 /* Search order: 3. OI files. */
1109 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1110 if (result == -ENOENT) {
1111 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1112 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1113 !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1115 GOTO(out, result = 0);
1120 /* -ESTALE is returned if inode of OST object doesn't exist */
1121 if (result == -ESTALE &&
1122 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1123 GOTO(out, result = 0);
1130 obj->oo_inode = NULL;
1131 /* for later passes through checks, not true on first pass */
1132 if (!IS_ERR_OR_NULL(inode))
1135 inode = osd_iget_check(info, dev, fid, id, trusted);
1136 if (!IS_ERR(inode)) {
1137 obj->oo_inode = inode;
1145 result = PTR_ERR(inode);
1146 if (result == -ENOENT || result == -ESTALE)
1147 GOTO(out, result = 0);
1149 if (result != -EREMCHG)
1154 * We still have chance to get the valid inode: for the
1155 * object which is referenced by remote name entry, the
1156 * object on the local MDT will be linked under the dir
1157 * of "/REMOTE_PARENT_DIR" with its FID string as name.
1159 * We do not know whether the object for the given FID
1160 * is referenced by some remote name entry or not, and
1161 * especially for DNE II, a multiple-linked object may
1162 * have many name entries reside on many MDTs.
1164 * To simplify the operation, OSD will not distinguish
1165 * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1166 * only happened for the RPC from other MDT during the
1167 * OI scrub, or for the client side RPC with FID only,
1168 * such as FID to path, or from old connected client.
1171 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1175 flags |= SS_AUTO_PARTIAL;
1176 flags &= ~SS_AUTO_FULL;
1181 if (thread_is_running(&scrub->os_thread)) {
1182 if (scrub->os_partial_scan && !scrub->os_in_join)
1185 if (IS_ERR_OR_NULL(inode) || result)
1186 GOTO(out, result = -EINPROGRESS);
1189 LASSERT(obj->oo_inode == inode);
1191 osd_add_oi_cache(info, dev, id, fid);
1192 osd_oii_insert(dev, oic, true);
1196 if (dev->od_auto_scrub_interval == AS_NEVER) {
1198 GOTO(out, result = -EREMCHG);
1201 LASSERT(obj->oo_inode == inode);
1203 osd_add_oi_cache(info, dev, id, fid);
1208 rc1 = osd_scrub_start(env, dev, flags);
1209 LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1210 "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1211 if (rc1 && rc1 != -EALREADY)
1212 GOTO(out, result = -EREMCHG);
1214 if (IS_ERR_OR_NULL(inode) || result)
1215 GOTO(out, result = -EINPROGRESS);
1218 LASSERT(obj->oo_inode == inode);
1220 osd_add_oi_cache(info, dev, id, fid);
1221 osd_oii_insert(dev, oic, true);
1226 if (unlikely(obj->oo_header))
1229 result = osd_check_lma(env, obj);
1233 LASSERTF(id->oii_ino == inode->i_ino &&
1234 id->oii_gen == inode->i_generation,
1235 "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1236 PFID(fid), id->oii_ino, id->oii_gen,
1237 inode->i_ino, inode->i_generation);
1239 saved_ino = inode->i_ino;
1240 saved_gen = inode->i_generation;
1242 if (unlikely(result == -ENODATA)) {
1244 * If the OI scrub updated the OI mapping by race, it
1245 * must be valid. Trust the inode that has no LMA EA.
1250 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1253 * The OI mapping is still there, the inode is still
1254 * valid. It is just becaues the inode has no LMA EA.
1256 if (saved_ino == id->oii_ino &&
1257 saved_gen == id->oii_gen)
1261 * It is the OI scrub updated the OI mapping by race.
1262 * The new OI mapping must be valid.
1270 * "result == -ENOENT" means that the OI mappinghas been
1271 * removed by race, so the inode belongs to other object.
1273 * Others error can be returned directly.
1275 if (result == -ENOENT) {
1278 obj->oo_inode = NULL;
1283 if (result != -EREMCHG)
1289 * if two OST objects map to the same inode, and inode mode is
1290 * (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666), which means it's
1291 * reserved by precreate, and not written yet, in this case, don't
1292 * set inode for the object whose FID mismatch, so that it can create
1293 * inode and not block precreate.
1295 if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) &&
1296 inode->i_mode == (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666)) {
1297 obj->oo_inode = NULL;
1298 GOTO(out, result = 0);
1301 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1303 * "result == -ENOENT" means the cached OI mapping has been removed
1304 * from the OI file by race, above inode belongs to other object.
1306 if (result == -ENOENT) {
1309 obj->oo_inode = NULL;
1310 GOTO(out, result = 0);
1316 if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1322 * It is the OI scrub updated the OI mapping by race.
1323 * The new OI mapping must be valid.
1331 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1332 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1334 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1336 if (lma->lma_compat & LMAC_STRIPE_INFO &&
1338 obj->oo_pfid_in_lma = 1;
1339 if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1341 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1342 } else if (result != -ENODATA) {
1347 obj->oo_compat_dot_created = 1;
1348 obj->oo_compat_dotdot_created = 1;
1350 if (S_ISDIR(inode->i_mode) &&
1351 (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1352 osd_check_lmv(info, dev, inode, oic);
1354 result = osd_attach_jinode(inode);
1359 GOTO(out, result = 0);
1361 LASSERT(!obj->oo_hl_head);
1362 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1364 GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1367 if (result || !obj->oo_inode) {
1368 if (!IS_ERR_OR_NULL(inode))
1371 obj->oo_inode = NULL;
1373 fid_zero(&oic->oic_fid);
1376 LINVRNT(osd_invariant(obj));
1381 * Concurrency: shouldn't matter.
1383 static void osd_object_init0(struct osd_object *obj)
1385 LASSERT(obj->oo_inode != NULL);
1386 obj->oo_dt.do_body_ops = &osd_body_ops;
1387 obj->oo_dt.do_lu.lo_header->loh_attr |=
1388 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1392 * Concurrency: no concurrent access is possible that early in object
1395 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1396 const struct lu_object_conf *conf)
1398 struct osd_object *obj = osd_obj(l);
1401 LINVRNT(osd_invariant(obj));
1403 if (OBD_FAIL_PRECHECK(OBD_FAIL_MDS_LLOG_UMOUNT_RACE) &&
1404 cfs_fail_val == 2) {
1405 struct osd_thread_info *info = osd_oti_get(env);
1406 struct osd_idmap_cache *oic = &info->oti_cache;
1407 /* invalidate thread cache */
1408 memset(&oic->oic_fid, 0, sizeof(oic->oic_fid));
1410 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1411 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1412 l->lo_header->loh_attr |= LOHA_EXISTS;
1416 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1417 obj->oo_dt.do_body_ops = &osd_body_ops_new;
1418 if (result == 0 && obj->oo_inode != NULL) {
1419 struct osd_thread_info *oti = osd_oti_get(env);
1420 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1422 osd_object_init0(obj);
1423 if (unlikely(obj->oo_header))
1426 result = osd_get_lma(oti, obj->oo_inode,
1427 &oti->oti_obj_dentry, loa);
1430 * Convert LMAI flags to lustre LMA flags
1431 * and cache it to oo_lma_flags
1434 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1435 } else if (result == -ENODATA) {
1439 obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
1441 LINVRNT(osd_invariant(obj));
1446 * The first part of oxe_buf is xattr name, and is '\0' terminated.
1447 * The left part is for value, binary mode.
1449 struct osd_xattr_entry {
1450 struct list_head oxe_list;
1454 struct rcu_head oxe_rcu;
1458 static int osd_oxc_get(struct osd_object *obj, const char *name,
1461 struct osd_xattr_entry *tmp;
1462 struct osd_xattr_entry *oxe = NULL;
1463 size_t namelen = strlen(name);
1467 list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1468 if (namelen == tmp->oxe_namelen &&
1469 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1476 GOTO(out, rc = -ENOENT);
1478 if (!oxe->oxe_exist)
1479 GOTO(out, rc = -ENODATA);
1482 rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1485 if (buf->lb_buf == NULL)
1488 if (buf->lb_len < rc)
1489 GOTO(out, rc = -ERANGE);
1491 memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1498 static void osd_oxc_free(struct rcu_head *head)
1500 struct osd_xattr_entry *oxe;
1502 oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1503 OBD_FREE(oxe, oxe->oxe_len);
1506 static void osd_oxc_add(struct osd_object *obj, const char *name,
1507 const char *buf, int buflen)
1509 struct osd_xattr_entry *oxe;
1510 struct osd_xattr_entry *old = NULL;
1511 struct osd_xattr_entry *tmp;
1512 size_t namelen = strlen(name);
1513 size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1515 OBD_ALLOC(oxe, len);
1519 INIT_LIST_HEAD(&oxe->oxe_list);
1521 oxe->oxe_namelen = namelen;
1522 memcpy(oxe->oxe_buf, name, namelen);
1524 LASSERT(buf != NULL);
1525 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1526 oxe->oxe_exist = true;
1528 oxe->oxe_exist = false;
1531 /* this should be rarely called, just remove old and add new */
1532 spin_lock(&obj->oo_guard);
1533 list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1534 if (namelen == tmp->oxe_namelen &&
1535 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1541 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1542 call_rcu(&old->oxe_rcu, osd_oxc_free);
1544 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1546 spin_unlock(&obj->oo_guard);
1549 static void osd_oxc_del(struct osd_object *obj, const char *name)
1551 struct osd_xattr_entry *oxe;
1552 size_t namelen = strlen(name);
1554 spin_lock(&obj->oo_guard);
1555 list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1556 if (namelen == oxe->oxe_namelen &&
1557 strncmp(name, oxe->oxe_buf, namelen) == 0) {
1558 list_del_rcu(&oxe->oxe_list);
1559 call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1563 spin_unlock(&obj->oo_guard);
1566 static void osd_oxc_fini(struct osd_object *obj)
1568 struct osd_xattr_entry *oxe, *next;
1570 list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1571 list_del(&oxe->oxe_list);
1572 OBD_FREE(oxe, oxe->oxe_len);
1577 * Concurrency: no concurrent access is possible that late in object
1580 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1582 struct osd_object *obj = osd_obj(l);
1583 struct lu_object_header *h = obj->oo_header;
1585 LINVRNT(osd_invariant(obj));
1588 dt_object_fini(&obj->oo_dt);
1589 if (obj->oo_hl_head != NULL)
1590 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1591 /* obj doesn't contain an lu_object_header, so we don't need call_rcu */
1594 lu_object_header_fini(h);
1595 OBD_FREE_PRE(h, sizeof(*h), "kfreed");
1596 kfree_rcu(h, loh_rcu);
1601 * Concurrency: no concurrent access is possible that late in object
1604 static void osd_index_fini(struct osd_object *o)
1606 struct iam_container *bag;
1608 if (o->oo_dir != NULL) {
1609 bag = &o->oo_dir->od_container;
1610 if (o->oo_inode != NULL) {
1611 if (bag->ic_object == o->oo_inode)
1612 iam_container_fini(bag);
1614 OBD_FREE_PTR(o->oo_dir);
1620 * Concurrency: no concurrent access is possible that late in object
1621 * life-cycle (for all existing callers, that is. New callers have to provide
1622 * their own locking.)
1624 static int osd_inode_unlinked(const struct inode *inode)
1626 return inode->i_nlink == 0;
1630 OSD_TXN_OI_DELETE_CREDITS = 20,
1631 OSD_TXN_INODE_DELETE_CREDITS = 20
1638 #if OSD_THANDLE_STATS
1640 * Set time when the handle is allocated
1642 static void osd_th_alloced(struct osd_thandle *oth)
1644 oth->oth_alloced = ktime_get();
1648 * Set time when the handle started
1650 static void osd_th_started(struct osd_thandle *oth)
1652 oth->oth_started = ktime_get();
1656 * Check whether the we deal with this handle for too long.
1658 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1659 ktime_t alloced, ktime_t started,
1662 ktime_t now = ktime_get();
1664 LASSERT(dev != NULL);
1666 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1667 ktime_us_delta(started, alloced));
1668 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1669 ktime_us_delta(closed, started));
1670 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1671 ktime_us_delta(now, closed));
1673 if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1674 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1675 oth, now, alloced, started, closed);
1676 libcfs_debug_dumpstack(NULL);
1680 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
1682 ktime_t __closed = ktime_get(); \
1683 ktime_t __alloced = oth->oth_alloced; \
1684 ktime_t __started = oth->oth_started; \
1687 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
1690 #else /* OSD_THANDLE_STATS */
1692 #define osd_th_alloced(h) do {} while(0)
1693 #define osd_th_started(h) do {} while(0)
1694 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
1696 #endif /* OSD_THANDLE_STATS */
1699 * Concurrency: doesn't access mutable data.
1701 static int osd_param_is_not_sane(const struct osd_device *dev,
1702 const struct thandle *th)
1704 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1706 return oh->ot_credits > osd_transaction_size(dev);
1710 * Concurrency: shouldn't matter.
1712 static void osd_trans_commit_cb(struct super_block *sb,
1713 struct ldiskfs_journal_cb_entry *jcb, int error)
1715 struct osd_thandle *oh = container_of(jcb, struct osd_thandle, ot_jcb);
1716 struct thandle *th = &oh->ot_super;
1717 struct lu_device *lud = &th->th_dev->dd_lu_dev;
1718 struct dt_txn_commit_cb *dcb, *tmp;
1720 LASSERT(oh->ot_handle == NULL);
1723 CERROR("transaction @0x%p commit error: %d\n", th, error);
1725 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DELAY_TRANS, 40);
1726 /* call per-transaction callbacks if any */
1727 list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1729 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1730 "commit callback entry: magic=%x name='%s'\n",
1731 dcb->dcb_magic, dcb->dcb_name);
1732 list_del_init(&dcb->dcb_linkage);
1733 dcb->dcb_func(NULL, th, dcb, error);
1736 lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1743 static struct thandle *osd_trans_create(const struct lu_env *env,
1744 struct dt_device *d)
1746 struct osd_thread_info *oti = osd_oti_get(env);
1747 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1748 struct osd_thandle *oh;
1754 CERROR("%s: someone try to start transaction under "
1755 "readonly mode, should be disabled.\n",
1756 osd_name(osd_dt_dev(d)));
1758 RETURN(ERR_PTR(-EROFS));
1761 /* on pending IO in this thread should left from prev. request */
1762 LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1764 sb_start_write(osd_sb(osd_dt_dev(d)));
1766 OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1768 sb_end_write(osd_sb(osd_dt_dev(d)));
1769 RETURN(ERR_PTR(-ENOMEM));
1772 oh->ot_quota_trans = &oti->oti_quota_trans;
1773 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1778 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1779 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1780 INIT_LIST_HEAD(&oh->ot_trunc_locks);
1783 memset(oti->oti_declare_ops, 0,
1784 sizeof(oti->oti_declare_ops));
1785 memset(oti->oti_declare_ops_cred, 0,
1786 sizeof(oti->oti_declare_ops_cred));
1787 memset(oti->oti_declare_ops_used, 0,
1788 sizeof(oti->oti_declare_ops_used));
1790 oti->oti_ins_cache_depth++;
1795 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1797 struct osd_thread_info *oti = osd_oti_get(env);
1798 struct osd_thandle *oh;
1800 oh = container_of(th, struct osd_thandle, ot_super);
1801 LASSERT(oh != NULL);
1803 CWARN(" create: %u/%u/%u, destroy: %u/%u/%u\n",
1804 oti->oti_declare_ops[OSD_OT_CREATE],
1805 oti->oti_declare_ops_cred[OSD_OT_CREATE],
1806 oti->oti_declare_ops_used[OSD_OT_CREATE],
1807 oti->oti_declare_ops[OSD_OT_DESTROY],
1808 oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1809 oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1810 CWARN(" attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1811 oti->oti_declare_ops[OSD_OT_ATTR_SET],
1812 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1813 oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1814 oti->oti_declare_ops[OSD_OT_XATTR_SET],
1815 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1816 oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1817 CWARN(" write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1818 oti->oti_declare_ops[OSD_OT_WRITE],
1819 oti->oti_declare_ops_cred[OSD_OT_WRITE],
1820 oti->oti_declare_ops_used[OSD_OT_WRITE],
1821 oti->oti_declare_ops[OSD_OT_PUNCH],
1822 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1823 oti->oti_declare_ops_used[OSD_OT_PUNCH],
1824 oti->oti_declare_ops[OSD_OT_QUOTA],
1825 oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1826 oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1827 CWARN(" insert: %u/%u/%u, delete: %u/%u/%u\n",
1828 oti->oti_declare_ops[OSD_OT_INSERT],
1829 oti->oti_declare_ops_cred[OSD_OT_INSERT],
1830 oti->oti_declare_ops_used[OSD_OT_INSERT],
1831 oti->oti_declare_ops[OSD_OT_DELETE],
1832 oti->oti_declare_ops_cred[OSD_OT_DELETE],
1833 oti->oti_declare_ops_used[OSD_OT_DELETE]);
1834 CWARN(" ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1835 oti->oti_declare_ops[OSD_OT_REF_ADD],
1836 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1837 oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1838 oti->oti_declare_ops[OSD_OT_REF_DEL],
1839 oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1840 oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1844 * Concurrency: shouldn't matter.
1846 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1849 struct osd_thread_info *oti = osd_oti_get(env);
1850 struct osd_device *dev = osd_dt_dev(d);
1852 struct osd_thandle *oh;
1857 LASSERT(current->journal_info == NULL);
1859 oh = container_of(th, struct osd_thandle, ot_super);
1860 LASSERT(oh != NULL);
1861 LASSERT(oh->ot_handle == NULL);
1863 rc = dt_txn_hook_start(env, d, th);
1867 if (unlikely(osd_param_is_not_sane(dev, th))) {
1868 static unsigned long last_printed;
1869 static int last_credits;
1872 * don't make noise on a tiny testing systems
1873 * actual credits misuse will be caught anyway
1875 if (last_credits != oh->ot_credits &&
1876 time_after(jiffies, last_printed +
1877 cfs_time_seconds(60)) &&
1878 osd_transaction_size(dev) > 512) {
1879 CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1880 oh->ot_credits, osd_transaction_size(dev));
1881 osd_trans_dump_creds(env, th);
1882 libcfs_debug_dumpstack(NULL);
1883 last_credits = oh->ot_credits;
1884 last_printed = jiffies;
1887 * XXX Limit the credits to 'max_transaction_buffers', and
1888 * let the underlying filesystem to catch the error if
1889 * we really need so many credits.
1891 * This should be removed when we can calculate the
1892 * credits precisely.
1894 oh->ot_credits = osd_transaction_size(dev);
1895 } else if (ldiskfs_track_declares_assert != 0) {
1897 * reserve few credits to prevent an assertion in JBD
1898 * our debugging mechanism will be able to detected
1899 * overuse. this can help to debug single-update
1902 oh->ot_credits += 10;
1903 if (unlikely(osd_param_is_not_sane(dev, th)))
1904 oh->ot_credits = osd_transaction_size(dev);
1907 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1908 GOTO(out, rc = -EIO);
1911 * XXX temporary stuff. Some abstraction layer should
1914 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1918 LASSERT(oti->oti_txns == 0);
1920 lu_device_get(&d->dd_lu_dev);
1921 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1932 static int osd_seq_exists(const struct lu_env *env,
1933 struct osd_device *osd, u64 seq)
1935 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1936 struct seq_server_site *ss = osd_seq_site(osd);
1941 LASSERT(ss != NULL);
1942 LASSERT(ss->ss_server_fld != NULL);
1944 rc = osd_fld_lookup(env, osd, seq, range);
1947 CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1948 osd_name(osd), seq, rc);
1952 RETURN(ss->ss_node_id == range->lsr_index);
1955 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1957 struct dt_txn_commit_cb *dcb;
1958 struct dt_txn_commit_cb *tmp;
1960 /* call per-transaction stop callbacks if any */
1961 list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1963 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1964 "commit callback entry: magic=%x name='%s'\n",
1965 dcb->dcb_magic, dcb->dcb_name);
1966 list_del_init(&dcb->dcb_linkage);
1967 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1972 * Concurrency: shouldn't matter.
1974 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1977 struct osd_thread_info *oti = osd_oti_get(env);
1978 struct osd_thandle *oh;
1979 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1980 struct osd_device *osd = osd_dt_dev(th->th_dev);
1981 struct qsd_instance *qsd = osd_def_qsd(osd);
1982 struct lquota_trans *qtrans;
1983 LIST_HEAD(truncates);
1984 int rc = 0, remove_agents = 0;
1988 oh = container_of(th, struct osd_thandle, ot_super);
1990 remove_agents = oh->ot_remove_agents;
1992 qtrans = oh->ot_quota_trans;
1993 oh->ot_quota_trans = NULL;
1995 /* move locks to local list, stop tx, execute truncates */
1996 list_splice(&oh->ot_trunc_locks, &truncates);
1998 if (oh->ot_handle != NULL) {
2001 handle_t *hdl = oh->ot_handle;
2004 * add commit callback
2005 * notice we don't do this in osd_trans_start()
2006 * as underlying transaction can change during truncate
2008 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
2011 LASSERT(oti->oti_txns == 1);
2014 rc = dt_txn_hook_stop(env, th);
2016 CERROR("%s: failed in transaction hook: rc = %d\n",
2019 osd_trans_stop_cb(oh, rc);
2020 /* hook functions might modify th_sync */
2021 hdl->h_sync = th->th_sync;
2023 oh->ot_handle = NULL;
2024 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2026 CERROR("%s: failed to stop transaction: rc = %d\n",
2027 osd_name(osd), rc2);
2031 osd_process_truncates(&truncates);
2033 osd_trans_stop_cb(oh, th->th_result);
2037 osd_trunc_unlock_all(env, &truncates);
2039 /* inform the quota slave device that the transaction is stopping */
2040 qsd_op_end(env, qsd, qtrans);
2043 * as we want IO to journal and data IO be concurrent, we don't block
2044 * awaiting data IO completion in osd_do_bio(), instead we wait here
2045 * once transaction is submitted to the journal. all reqular requests
2046 * don't do direct IO (except read/write), thus this wait_event becomes
2049 * IMPORTANT: we have to wait till any IO submited by the thread is
2050 * completed otherwise iobuf may be corrupted by different request
2052 wait_event(iobuf->dr_wait,
2053 atomic_read(&iobuf->dr_numreqs) == 0);
2054 osd_fini_iobuf(osd, iobuf);
2056 rc = iobuf->dr_error;
2058 if (unlikely(remove_agents != 0))
2059 osd_process_scheduled_agent_removals(env, osd);
2061 oti->oti_ins_cache_depth--;
2062 /* reset OI cache for safety */
2063 if (oti->oti_ins_cache_depth == 0)
2064 oti->oti_ins_cache_used = 0;
2066 sb_end_write(osd_sb(osd));
2071 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2073 struct osd_thandle *oh = container_of(th, struct osd_thandle,
2076 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2077 LASSERT(&dcb->dcb_func != NULL);
2078 if (dcb->dcb_flags & DCB_TRANS_STOP)
2079 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2081 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2087 * Called just before object is freed. Releases all resources except for
2088 * object itself (that is released by osd_object_free()).
2090 * Concurrency: no concurrent access is possible that late in object
2093 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2095 struct osd_object *obj = osd_obj(l);
2096 struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2097 struct inode *inode = obj->oo_inode;
2102 LINVRNT(osd_invariant(obj));
2105 * If object is unlinked remove fid->ino mapping from object index.
2108 osd_index_fini(obj);
2113 if (osd_has_index(obj) && obj->oo_dt.do_index_ops == &osd_index_iam_ops)
2114 ldiskfs_set_inode_flag(inode, LDISKFS_INODE_JOURNAL_DATA);
2116 uid = i_uid_read(inode);
2117 gid = i_gid_read(inode);
2118 projid = i_projid_read(inode);
2120 obj->oo_inode = NULL;
2123 /* do not rebalance quota if the caller needs to release memory
2124 * otherwise qsd_refresh_usage() may went into a new ldiskfs
2125 * transaction and risk to deadlock - LU-12178 */
2126 if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2129 if (!obj->oo_header && qsd) {
2130 struct osd_thread_info *info = osd_oti_get(env);
2131 struct lquota_id_info *qi = &info->oti_qi;
2133 /* Release granted quota to master if necessary */
2134 qi->lqi_id.qid_uid = uid;
2135 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2137 qi->lqi_id.qid_uid = gid;
2138 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2140 qi->lqi_id.qid_uid = projid;
2141 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2146 * Concurrency: ->loo_object_release() is called under site spin-lock.
2148 static void osd_object_release(const struct lu_env *env,
2149 struct lu_object *l)
2151 struct osd_object *o = osd_obj(l);
2154 * nobody should be releasing a non-destroyed object with nlink=0
2155 * the API allows this, but ldiskfs doesn't like and then report
2156 * this inode as deleted
2158 LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2159 o->oo_inode->i_nlink == 0));
2163 * Concurrency: shouldn't matter.
2165 static int osd_object_print(const struct lu_env *env, void *cookie,
2166 lu_printer_t p, const struct lu_object *l)
2168 struct osd_object *o = osd_obj(l);
2169 struct iam_descr *d;
2171 if (o->oo_dir != NULL)
2172 d = o->oo_dir->od_container.ic_descr;
2175 return (*p)(env, cookie,
2176 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2178 o->oo_inode ? o->oo_inode->i_ino : 0UL,
2179 o->oo_inode ? o->oo_inode->i_generation : 0,
2180 d ? d->id_ops->id_name : "plain");
2184 * Concurrency: shouldn't matter.
2186 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2187 struct obd_statfs *sfs, struct obd_statfs_info *info)
2189 struct osd_device *osd = osd_dt_dev(d);
2190 struct super_block *sb = osd_sb(osd);
2191 struct kstatfs *ksfs;
2195 if (unlikely(osd->od_mnt == NULL))
2196 return -EINPROGRESS;
2198 /* osd_lproc.c call this without env, allocate ksfs for that case */
2199 if (unlikely(env == NULL)) {
2200 OBD_ALLOC_PTR(ksfs);
2204 ksfs = &osd_oti_get(env)->oti_ksfs;
2207 result = sb->s_op->statfs(sb->s_root, ksfs);
2211 statfs_pack(sfs, ksfs);
2212 if (unlikely(sb->s_flags & SB_RDONLY))
2213 sfs->os_state |= OS_STATFS_READONLY;
2215 sfs->os_state |= osd->od_nonrotational ? OS_STATFS_NONROT : 0;
2217 if (ldiskfs_has_feature_extents(sb))
2218 sfs->os_maxbytes = sb->s_maxbytes;
2220 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2223 * Reserve some space so to avoid fragmenting the filesystem too much.
2224 * Fragmentation not only impacts performance, but can also increase
2225 * metadata overhead significantly, causing grant calculation to be
2228 * Reserve 0.78% of total space, at least 8MB for small filesystems.
2230 BUILD_BUG_ON(OSD_STATFS_RESERVED <= LDISKFS_MAX_BLOCK_SIZE);
2231 reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2232 if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2233 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2235 sfs->os_blocks -= reserved;
2236 sfs->os_bfree -= min(reserved, sfs->os_bfree);
2237 sfs->os_bavail -= min(reserved, sfs->os_bavail);
2240 if (unlikely(env == NULL))
2246 * Estimate space needed for file creations. We assume the largest filename
2247 * which is 2^64 - 1, hence a filename of 20 chars.
2248 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2250 #ifdef __LDISKFS_DIR_REC_LEN
2251 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2253 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2257 * Concurrency: doesn't access mutable data.
2259 static void osd_conf_get(const struct lu_env *env,
2260 const struct dt_device *dev,
2261 struct dt_device_param *param)
2263 struct osd_device *d = osd_dt_dev(dev);
2264 struct super_block *sb = osd_sb(d);
2265 struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2270 * XXX should be taken from not-yet-existing fs abstraction layer.
2272 param->ddp_max_name_len = LDISKFS_NAME_LEN;
2273 param->ddp_max_nlink = LDISKFS_LINK_MAX;
2274 param->ddp_symlink_max = sb->s_blocksize;
2275 param->ddp_mount_type = LDD_MT_LDISKFS;
2276 if (ldiskfs_has_feature_extents(sb))
2277 param->ddp_maxbytes = sb->s_maxbytes;
2279 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2281 * inode are statically allocated, so per-inode space consumption
2282 * is the space consumed by the directory entry
2284 param->ddp_inodespace = PER_OBJ_USAGE;
2286 * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2287 * is 128MB) which is unlikely to be hit in real life. Report a smaller
2288 * maximum length to not under-count the actual number of extents
2289 * needed for writing a file if there are sub-optimal block allocations.
2291 param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2292 /* worst-case extent insertion metadata overhead */
2293 param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2294 param->ddp_mntopts = 0;
2295 if (test_opt(sb, XATTR_USER))
2296 param->ddp_mntopts |= MNTOPT_USERXATTR;
2297 if (test_opt(sb, POSIX_ACL))
2298 param->ddp_mntopts |= MNTOPT_ACL;
2301 * LOD might calculate the max stripe count based on max_ea_size,
2302 * so we need take account in the overhead as well,
2303 * xattr_header + magic + xattr_entry_head
2305 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2306 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2308 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2309 if (ldiskfs_has_feature_ea_inode(sb))
2310 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2314 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2316 if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE)
2317 param->ddp_max_ea_size = OBD_MAX_EA_SIZE;
2320 * Preferred RPC size for efficient disk IO. 4MB shows good
2321 * all-around performance for ldiskfs, but use bigalloc chunk size
2322 * by default if larger.
2324 #if defined(LDISKFS_CLUSTER_SIZE)
2325 if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2326 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2329 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2331 param->ddp_t10_cksum_type = 0;
2333 unsigned short interval = blk_integrity_interval(bi);
2334 name = blk_integrity_name(bi);
2342 if (strncmp(name, "T10-DIF-TYPE",
2343 sizeof("T10-DIF-TYPE") - 1) == 0) {
2344 /* also skip "1/3-" at end */
2345 const int type_off = sizeof("T10-DIF-TYPE.");
2346 char type_number = name[type_off - 2];
2348 if (interval != 512 && interval != 4096) {
2349 CERROR("%s: unsupported T10PI sector size %u\n",
2350 d->od_svname, interval);
2351 } else if (type_number != '1' && type_number != '3') {
2352 CERROR("%s: unsupported T10PI type %s\n",
2353 d->od_svname, name);
2354 } else if (strcmp(name + type_off, "CRC") == 0) {
2355 d->od_t10_type = type_number == '1' ?
2356 OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2357 param->ddp_t10_cksum_type = interval == 512 ?
2358 OBD_CKSUM_T10CRC512 :
2360 } else if (strcmp(name + type_off, "IP") == 0) {
2361 d->od_t10_type = type_number == '1' ?
2362 OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2363 param->ddp_t10_cksum_type = interval == 512 ?
2364 OBD_CKSUM_T10IP512 :
2367 CERROR("%s: unsupported checksum type of "
2369 d->od_svname, name);
2373 CERROR("%s: unsupported T10PI type '%s'",
2374 d->od_svname, name);
2379 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2381 return osd_sb(osd_dt_dev(d));
2385 * Concurrency: shouldn't matter.
2387 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2390 struct super_block *s = osd_sb(osd_dt_dev(d));
2393 down_read(&s->s_umount);
2394 rc = s->s_op->sync_fs(s, 1);
2395 up_read(&s->s_umount);
2397 CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2404 * Start commit for OSD device.
2406 * An implementation of dt_commit_async method for OSD device.
2407 * Asychronously starts underlayng fs sync and thereby a transaction
2410 * \param env environment
2411 * \param d dt device
2413 * \see dt_device_operations
2415 static int osd_commit_async(const struct lu_env *env,
2416 struct dt_device *d)
2418 struct super_block *s = osd_sb(osd_dt_dev(d));
2423 CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2424 down_read(&s->s_umount);
2425 rc = s->s_op->sync_fs(s, 0);
2426 up_read(&s->s_umount);
2431 static int (*priv_security_file_alloc)(struct file *file);
2433 int osd_security_file_alloc(struct file *file)
2435 if (priv_security_file_alloc)
2436 return priv_security_file_alloc(file);
2441 * Concurrency: shouldn't matter.
2443 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2445 struct super_block *sb = osd_sb(osd_dt_dev(d));
2446 struct block_device *dev = sb->s_bdev;
2447 int rc = -EOPNOTSUPP;
2451 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2452 osd_dt_dev(d)->od_svname, (long)dev, rc);
2458 * Note: we do not count into QUOTA here.
2459 * If we mount with --data_journal we may need more.
2461 const int osd_dto_credits_noquota[DTO_NR] = {
2464 * INDEX_EXTRA_TRANS_BLOCKS(8) +
2465 * SINGLEDATA_TRANS_BLOCKS(8)
2466 * XXX Note: maybe iam need more, since iam have more level than
2469 [DTO_INDEX_INSERT] = 16,
2472 * just modify a single entry, probably merge few within a block
2474 [DTO_INDEX_DELETE] = 1,
2478 [DTO_INDEX_UPDATE] = 16,
2480 * 4(inode, inode bits, groups, GDT)
2481 * notice: OI updates are counted separately with DTO_INDEX_INSERT
2483 [DTO_OBJECT_CREATE] = 4,
2485 * 4(inode, inode bits, groups, GDT)
2486 * notice: OI updates are counted separately with DTO_INDEX_DELETE
2488 [DTO_OBJECT_DELETE] = 4,
2490 * Attr set credits (inode)
2492 [DTO_ATTR_SET_BASE] = 1,
2494 * Xattr set. The same as xattr of EXT3.
2495 * DATA_TRANS_BLOCKS(14)
2496 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2497 * are also counted in. Do not know why?
2499 [DTO_XATTR_SET] = 14,
2501 * credits for inode change during write.
2503 [DTO_WRITE_BASE] = 3,
2505 * credits for single block write.
2507 [DTO_WRITE_BLOCK] = 14,
2509 * Attr set credits for chown.
2510 * This is extra credits for setattr, and it is null without quota
2512 [DTO_ATTR_SET_CHOWN] = 0
2515 static const struct dt_device_operations osd_dt_ops = {
2516 .dt_root_get = osd_root_get,
2517 .dt_statfs = osd_statfs,
2518 .dt_trans_create = osd_trans_create,
2519 .dt_trans_start = osd_trans_start,
2520 .dt_trans_stop = osd_trans_stop,
2521 .dt_trans_cb_add = osd_trans_cb_add,
2522 .dt_conf_get = osd_conf_get,
2523 .dt_mnt_sb_get = osd_mnt_sb_get,
2524 .dt_sync = osd_sync,
2526 .dt_commit_async = osd_commit_async,
2529 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2532 struct osd_object *obj = osd_dt_obj(dt);
2533 struct osd_thread_info *oti = osd_oti_get(env);
2535 LINVRNT(osd_invariant(obj));
2537 LASSERT(obj->oo_owner != env);
2538 down_read_nested(&obj->oo_sem, role);
2540 LASSERT(obj->oo_owner == NULL);
2544 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2547 struct osd_object *obj = osd_dt_obj(dt);
2548 struct osd_thread_info *oti = osd_oti_get(env);
2550 LINVRNT(osd_invariant(obj));
2552 LASSERT(obj->oo_owner != env);
2553 down_write_nested(&obj->oo_sem, role);
2555 LASSERT(obj->oo_owner == NULL);
2556 obj->oo_owner = env;
2560 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2562 struct osd_object *obj = osd_dt_obj(dt);
2563 struct osd_thread_info *oti = osd_oti_get(env);
2565 LINVRNT(osd_invariant(obj));
2567 LASSERT(oti->oti_r_locks > 0);
2569 up_read(&obj->oo_sem);
2572 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2574 struct osd_object *obj = osd_dt_obj(dt);
2575 struct osd_thread_info *oti = osd_oti_get(env);
2577 LINVRNT(osd_invariant(obj));
2579 LASSERT(obj->oo_owner == env);
2580 LASSERT(oti->oti_w_locks > 0);
2582 obj->oo_owner = NULL;
2583 up_write(&obj->oo_sem);
2586 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2588 struct osd_object *obj = osd_dt_obj(dt);
2590 LINVRNT(osd_invariant(obj));
2592 return obj->oo_owner == env;
2595 static void osd_inode_getattr(const struct lu_env *env,
2596 struct inode *inode, struct lu_attr *attr)
2598 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2599 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2600 LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2601 LA_BLKSIZE | LA_TYPE | LA_BTIME;
2603 attr->la_atime = inode->i_atime.tv_sec;
2604 attr->la_mtime = inode->i_mtime.tv_sec;
2605 attr->la_ctime = inode->i_ctime.tv_sec;
2606 attr->la_btime = LDISKFS_I(inode)->i_crtime.tv_sec;
2607 attr->la_mode = inode->i_mode;
2608 attr->la_size = i_size_read(inode);
2609 attr->la_blocks = inode->i_blocks;
2610 attr->la_uid = i_uid_read(inode);
2611 attr->la_gid = i_gid_read(inode);
2612 attr->la_projid = i_projid_read(inode);
2613 attr->la_flags = ll_inode_to_ext_flags(inode->i_flags);
2614 attr->la_nlink = inode->i_nlink;
2615 attr->la_rdev = inode->i_rdev;
2616 attr->la_blksize = 1 << inode->i_blkbits;
2617 attr->la_blkbits = inode->i_blkbits;
2619 * Ext4 did not transfer inherit flags from raw inode
2620 * to inode flags, and ext4 internally test raw inode
2621 * @i_flags directly. Instead of patching ext4, we do it here.
2623 if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2624 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2627 static int osd_dirent_count(const struct lu_env *env, struct dt_object *dt,
2630 struct osd_object *obj = osd_dt_obj(dt);
2631 const struct dt_it_ops *iops;
2637 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2638 LASSERT(fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)));
2640 if (obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
2641 *count = obj->oo_dirent_count;
2645 /* directory not initialized yet */
2646 if (!dt->do_index_ops) {
2651 iops = &dt->do_index_ops->dio_it;
2652 it = iops->init(env, dt, LUDA_64BITHASH);
2654 RETURN(PTR_ERR(it));
2656 rc = iops->load(env, it, 0);
2658 if (rc == -ENODATA) {
2665 rc = iops->next(env, it);
2667 for (*count = 0; rc == 0 || rc == -ESTALE; rc = iops->next(env, it)) {
2671 if (iops->key_size(env, it) == 0)
2677 obj->oo_dirent_count = *count;
2682 iops->fini(env, it);
2687 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2688 struct lu_attr *attr)
2690 struct osd_object *obj = osd_dt_obj(dt);
2693 if (unlikely(!dt_object_exists(dt)))
2695 if (unlikely(obj->oo_destroyed))
2698 LASSERT(!dt_object_remote(dt));
2699 LINVRNT(osd_invariant(obj));
2701 spin_lock(&obj->oo_guard);
2702 osd_inode_getattr(env, obj->oo_inode, attr);
2703 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2704 attr->la_valid |= LA_FLAGS;
2705 attr->la_flags |= LUSTRE_ORPHAN_FL;
2707 if (obj->oo_lma_flags & LUSTRE_ENCRYPT_FL) {
2708 attr->la_valid |= LA_FLAGS;
2709 attr->la_flags |= LUSTRE_ENCRYPT_FL;
2711 spin_unlock(&obj->oo_guard);
2713 if (S_ISDIR(obj->oo_inode->i_mode) &&
2714 fid_is_namespace_visible(lu_object_fid(&dt->do_lu)))
2715 rc = osd_dirent_count(env, dt, &attr->la_dirent_count);
2720 static int osd_declare_attr_qid(const struct lu_env *env,
2721 struct osd_object *obj,
2722 struct osd_thandle *oh, long long bspace,
2723 qid_t old_id, qid_t new_id, bool enforce,
2724 unsigned int type, bool ignore_edquot)
2727 struct osd_thread_info *info = osd_oti_get(env);
2728 struct lquota_id_info *qi = &info->oti_qi;
2730 qi->lqi_type = type;
2731 /* inode accounting */
2732 qi->lqi_is_blk = false;
2734 /* one more inode for the new id ... */
2735 qi->lqi_id.qid_uid = new_id;
2737 /* Reserve credits for the new id */
2738 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2739 if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2744 /* and one less inode for the current id */
2745 qi->lqi_id.qid_uid = old_id;
2747 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2748 if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2753 /* block accounting */
2754 qi->lqi_is_blk = true;
2756 /* more blocks for the new id ... */
2757 qi->lqi_id.qid_uid = new_id;
2758 qi->lqi_space = bspace;
2760 * Credits for the new uid has been reserved, re-use "obj"
2761 * to save credit reservation.
2763 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2764 if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2769 /* and finally less blocks for the current uid */
2770 qi->lqi_id.qid_uid = old_id;
2771 qi->lqi_space = -bspace;
2772 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2773 if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2779 static int osd_declare_attr_set(const struct lu_env *env,
2780 struct dt_object *dt,
2781 const struct lu_attr *attr,
2782 struct thandle *handle)
2784 struct osd_thandle *oh;
2785 struct osd_object *obj;
2794 LASSERT(dt != NULL);
2795 LASSERT(handle != NULL);
2797 obj = osd_dt_obj(dt);
2798 LASSERT(osd_invariant(obj));
2800 oh = container_of(handle, struct osd_thandle, ot_super);
2801 LASSERT(oh->ot_handle == NULL);
2803 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2804 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2806 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2807 osd_dto_credits_noquota[DTO_XATTR_SET]);
2809 if (attr == NULL || obj->oo_inode == NULL)
2812 bspace = obj->oo_inode->i_blocks << 9;
2813 bspace = toqb(bspace);
2816 * Changing ownership is always preformed by super user, it should not
2817 * fail with EDQUOT unless required explicitly.
2819 * We still need to call the osd_declare_qid() to calculate the journal
2820 * credits for updating quota accounting files and to trigger quota
2821 * space adjustment once the operation is completed.
2823 if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2824 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2827 CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2828 "(the quota space is %lld)\n",
2829 obj->oo_inode->i_sb->s_id, attr->la_uid,
2830 attr->la_gid, bspace);
2833 uid = i_uid_read(obj->oo_inode);
2834 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2835 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2836 attr->la_uid, enforce, USRQUOTA,
2841 gid = i_gid_read(obj->oo_inode);
2842 CDEBUG(D_QUOTA, "declare uid %d -> %d gid %d -> %d\n", uid,
2843 attr->la_uid, gid, attr->la_gid);
2844 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2845 rc = osd_declare_attr_qid(env, obj, oh, bspace, gid,
2846 attr->la_gid, enforce, GRPQUOTA,
2852 #ifdef HAVE_PROJECT_QUOTA
2853 if (attr->la_valid & LA_PROJID) {
2854 __u32 projid = i_projid_read(obj->oo_inode);
2856 enforce = (attr->la_valid & LA_PROJID) &&
2857 (attr->la_projid != projid);
2858 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2859 (qid_t)projid, (qid_t)attr->la_projid,
2860 enforce, PRJQUOTA, true);
2868 static int osd_inode_setattr(const struct lu_env *env,
2869 struct inode *inode, const struct lu_attr *attr)
2871 __u64 bits = attr->la_valid;
2873 /* Only allow set size for regular file */
2874 if (!S_ISREG(inode->i_mode))
2875 bits &= ~(LA_SIZE | LA_BLOCKS);
2880 if (bits & LA_ATIME)
2881 inode->i_atime = osd_inode_time(inode, attr->la_atime);
2882 if (bits & LA_CTIME)
2883 inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
2884 if (bits & LA_MTIME)
2885 inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
2886 if (bits & LA_SIZE) {
2887 spin_lock(&inode->i_lock);
2888 LDISKFS_I(inode)->i_disksize = attr->la_size;
2889 i_size_write(inode, attr->la_size);
2890 spin_unlock(&inode->i_lock);
2894 * OSD should not change "i_blocks" which is used by quota.
2895 * "i_blocks" should be changed by ldiskfs only.
2898 inode->i_mode = (inode->i_mode & S_IFMT) |
2899 (attr->la_mode & ~S_IFMT);
2901 i_uid_write(inode, attr->la_uid);
2903 i_gid_write(inode, attr->la_gid);
2904 if (bits & LA_PROJID)
2905 i_projid_write(inode, attr->la_projid);
2906 if (bits & LA_NLINK)
2907 set_nlink(inode, attr->la_nlink);
2909 inode->i_rdev = attr->la_rdev;
2911 if (bits & LA_FLAGS) {
2912 /* always keep S_NOCMTIME */
2913 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2916 * Ext4 did not transfer inherit flags from
2917 * @inode->i_flags to raw inode i_flags when writing
2918 * flags, we do it explictly here.
2920 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2921 LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2923 LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2928 #ifdef HAVE_PROJECT_QUOTA
2929 static int osd_transfer_project(struct inode *inode, __u32 projid,
2930 struct thandle *handle)
2932 struct super_block *sb = inode->i_sb;
2933 struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2936 struct ldiskfs_iloc iloc;
2937 struct ldiskfs_inode *raw_inode;
2938 struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2940 if (!ldiskfs_has_feature_project(sb)) {
2941 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2942 == LDISKFS_DEF_PROJID);
2943 if (projid != LDISKFS_DEF_PROJID)
2949 if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2952 kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2953 if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2956 err = ldiskfs_get_inode_loc(inode, &iloc);
2960 raw_inode = ldiskfs_raw_inode(&iloc);
2961 if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2962 struct osd_thandle *oh =
2963 container_of0(handle, struct osd_thandle,
2966 * try to expand inode size automatically.
2968 ldiskfs_mark_inode_dirty(oh->ot_handle, inode);
2969 if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2977 dquot_initialize(inode);
2978 transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2979 if (transfer_to[PRJQUOTA]) {
2980 err = __dquot_transfer(inode, transfer_to);
2981 dqput(transfer_to[PRJQUOTA]);
2990 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr,
2991 struct thandle *handle)
2995 if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2996 (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
3000 "executing dquot_transfer inode %ld uid %d -> %d gid %d -> %d\n",
3001 inode->i_ino, i_uid_read(inode), attr->la_uid,
3002 i_gid_read(inode), attr->la_gid);
3004 dquot_initialize(inode);
3006 if (attr->la_valid & LA_UID)
3007 iattr.ia_valid |= ATTR_UID;
3008 if (attr->la_valid & LA_GID)
3009 iattr.ia_valid |= ATTR_GID;
3010 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
3011 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
3013 rc = dquot_transfer(inode, &iattr);
3015 CERROR("%s: quota transfer failed: rc = %d. Is quota "
3016 "enforcement enabled on the ldiskfs "
3017 "filesystem?\n", inode->i_sb->s_id, rc);
3022 /* Handle project id transfer here properly */
3023 if (attr->la_valid & LA_PROJID &&
3024 attr->la_projid != i_projid_read(inode)) {
3025 #ifdef HAVE_PROJECT_QUOTA
3026 rc = osd_transfer_project(inode, attr->la_projid, handle);
3031 CERROR("%s: quota transfer failed: rc = %d. Is project "
3032 "enforcement enabled on the ldiskfs "
3033 "filesystem?\n", inode->i_sb->s_id, rc);
3040 static int osd_attr_set(const struct lu_env *env,
3041 struct dt_object *dt,
3042 const struct lu_attr *attr,
3043 struct thandle *handle)
3045 struct osd_object *obj = osd_dt_obj(dt);
3046 struct inode *inode;
3049 if (!dt_object_exists(dt))
3052 LASSERT(handle != NULL);
3053 LASSERT(!dt_object_remote(dt));
3054 LASSERT(osd_invariant(obj));
3056 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
3058 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
3059 !osd_obj2dev(obj)->od_is_ost) {
3060 struct osd_thread_info *oti = osd_oti_get(env);
3061 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
3062 struct lu_fid *fid1 = &oti->oti_fid;
3063 struct osd_inode_id *id = &oti->oti_id;
3064 struct iam_path_descr *ipd;
3065 struct iam_container *bag;
3066 struct osd_thandle *oh;
3069 fid_cpu_to_be(fid1, fid0);
3070 memset(id, 1, sizeof(*id));
3071 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
3072 fid0)->oi_dir.od_container;
3073 ipd = osd_idx_ipd_get(env, bag);
3074 if (unlikely(ipd == NULL))
3077 oh = container_of(handle, struct osd_thandle, ot_super);
3078 rc = iam_update(oh->ot_handle, bag,
3079 (const struct iam_key *)fid1,
3080 (const struct iam_rec *)id, ipd);
3081 osd_ipd_put(env, bag, ipd);
3082 return(rc > 0 ? 0 : rc);
3085 inode = obj->oo_inode;
3087 rc = osd_quota_transfer(inode, attr, handle);
3091 spin_lock(&obj->oo_guard);
3092 rc = osd_inode_setattr(env, inode, attr);
3093 spin_unlock(&obj->oo_guard);
3097 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3099 osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3101 if (!(attr->la_valid & LA_FLAGS))
3104 /* Let's check if there are extra flags need to be set into LMA */
3105 if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3106 struct osd_thread_info *info = osd_oti_get(env);
3107 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3109 LASSERT(!obj->oo_pfid_in_lma);
3111 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3112 &info->oti_ost_attrs);
3116 lma->lma_incompat |=
3117 lustre_to_lma_flags(attr->la_flags);
3118 lustre_lma_swab(lma);
3120 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3122 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3123 lma, sizeof(*lma), XATTR_REPLACE);
3125 struct osd_device *osd = osd_obj2dev(obj);
3127 CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3128 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3129 lma->lma_incompat, rc);
3132 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3134 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3141 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3142 struct osd_object *obj,
3143 const char *name, const int namelen)
3145 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3148 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3149 umode_t mode, struct dt_allocation_hint *hint,
3150 struct thandle *th, struct lu_attr *attr)
3153 struct osd_device *osd = osd_obj2dev(obj);
3154 struct osd_thandle *oth;
3155 struct dt_object *parent = NULL;
3156 struct inode *inode;
3157 uid_t owner[2] = {0, 0};
3159 if (attr->la_valid & LA_UID)
3160 owner[0] = attr->la_uid;
3161 if (attr->la_valid & LA_GID)
3162 owner[1] = attr->la_gid;
3164 LINVRNT(osd_invariant(obj));
3165 LASSERT(obj->oo_inode == NULL);
3166 LASSERT(obj->oo_hl_head == NULL);
3168 if (S_ISDIR(mode) && ldiskfs_pdo) {
3170 ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3171 if (obj->oo_hl_head == NULL)
3175 oth = container_of(th, struct osd_thandle, ot_super);
3176 LASSERT(oth->ot_handle->h_transaction != NULL);
3178 if (hint != NULL && hint->dah_parent != NULL &&
3179 !dt_object_remote(hint->dah_parent))
3180 parent = hint->dah_parent;
3182 inode = ldiskfs_create_inode(oth->ot_handle,
3183 parent ? osd_dt_obj(parent)->oo_inode :
3184 osd_sb(osd)->s_root->d_inode,
3186 if (!IS_ERR(inode)) {
3187 /* Do not update file c/mtime in ldiskfs. */
3188 inode->i_flags |= S_NOCMTIME;
3191 * For new created object, it must be consistent,
3192 * and it is unnecessary to scrub against it.
3194 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3196 obj->oo_inode = inode;
3199 if (obj->oo_hl_head != NULL) {
3200 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3201 obj->oo_hl_head = NULL;
3203 result = PTR_ERR(inode);
3205 LINVRNT(osd_invariant(obj));
3213 static int osd_mkdir(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,
3220 struct osd_thandle *oth;
3221 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3223 LASSERT(S_ISDIR(attr->la_mode));
3225 oth = container_of(th, struct osd_thandle, ot_super);
3226 LASSERT(oth->ot_handle->h_transaction != NULL);
3227 if (fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)))
3228 obj->oo_dirent_count = 0;
3229 result = osd_mkfile(info, obj, mode, hint, th, attr);
3234 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3235 struct lu_attr *attr,
3236 struct dt_allocation_hint *hint,
3237 struct dt_object_format *dof,
3241 struct osd_thandle *oth;
3242 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3244 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3246 LASSERT(S_ISREG(attr->la_mode));
3248 oth = container_of(th, struct osd_thandle, ot_super);
3249 LASSERT(oth->ot_handle->h_transaction != NULL);
3251 result = osd_mkfile(info, obj, mode, hint, th, attr);
3253 LASSERT(obj->oo_inode != NULL);
3254 if (feat->dif_flags & DT_IND_VARKEY)
3255 result = iam_lvar_create(obj->oo_inode,
3256 feat->dif_keysize_max,
3258 feat->dif_recsize_max,
3261 result = iam_lfix_create(obj->oo_inode,
3262 feat->dif_keysize_max,
3264 feat->dif_recsize_max,
3270 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3271 struct lu_attr *attr,
3272 struct dt_allocation_hint *hint,
3273 struct dt_object_format *dof,
3276 LASSERT(S_ISREG(attr->la_mode));
3277 return osd_mkfile(info, obj, (attr->la_mode &
3278 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3282 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3283 struct lu_attr *attr,
3284 struct dt_allocation_hint *hint,
3285 struct dt_object_format *dof,
3288 LASSERT(S_ISLNK(attr->la_mode));
3289 return osd_mkfile(info, obj, (attr->la_mode &
3290 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3294 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3295 struct lu_attr *attr,
3296 struct dt_allocation_hint *hint,
3297 struct dt_object_format *dof,
3300 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3303 LINVRNT(osd_invariant(obj));
3304 LASSERT(obj->oo_inode == NULL);
3305 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3306 S_ISFIFO(mode) || S_ISSOCK(mode));
3308 result = osd_mkfile(info, obj, mode, hint, th, attr);
3310 LASSERT(obj->oo_inode != NULL);
3312 * This inode should be marked dirty for i_rdev. Currently
3313 * that is done in the osd_attr_init().
3315 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3318 LINVRNT(osd_invariant(obj));
3322 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3324 struct dt_allocation_hint *hint,
3325 struct dt_object_format *dof,
3328 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3330 osd_obj_type_f result;
3346 result = osd_mk_index;
3357 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3358 struct dt_object *parent, struct dt_object *child,
3363 ah->dah_parent = parent;
3364 ah->dah_mode = child_mode;
3366 if (parent != NULL && !dt_object_remote(parent)) {
3367 /* will help to find FID->ino at dt_insert("..") */
3368 struct osd_object *pobj = osd_dt_obj(parent);
3370 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3374 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3375 struct lu_attr *attr, struct dt_object_format *dof,
3376 struct thandle *handle)
3378 struct inode *inode = obj->oo_inode;
3379 __u64 valid = attr->la_valid;
3382 attr->la_valid &= ~(LA_TYPE | LA_MODE);
3384 if (dof->dof_type != DFT_NODE)
3385 attr->la_valid &= ~LA_RDEV;
3386 if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3387 attr->la_valid &= ~LA_ATIME;
3388 if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3389 attr->la_valid &= ~LA_CTIME;
3390 if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3391 attr->la_valid &= ~LA_MTIME;
3393 result = osd_quota_transfer(inode, attr, handle);
3397 if (attr->la_valid != 0) {
3398 result = osd_inode_setattr(info->oti_env, inode, attr);
3400 * The osd_inode_setattr() should always succeed here. The
3401 * only error that could be returned is EDQUOT when we are
3402 * trying to change the UID or GID of the inode. However, this
3403 * should not happen since quota enforcement is no longer
3404 * enabled on ldiskfs (lquota takes care of it).
3406 LASSERTF(result == 0, "%d\n", result);
3407 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3410 attr->la_valid = valid;
3414 * Helper function for osd_create()
3416 * \retval 0, on success
3418 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3419 struct lu_attr *attr, struct dt_allocation_hint *hint,
3420 struct dt_object_format *dof, struct thandle *th)
3425 osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3427 /* we drop umask so that permissions we pass are not affected */
3428 umask = current->fs->umask;
3429 current->fs->umask = 0;
3431 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3433 if (likely(obj->oo_inode != NULL)) {
3434 LASSERT(obj->oo_inode->i_state & I_NEW);
3437 * Unlock the inode before attr initialization to avoid
3438 * unnecessary dqget operations. LU-6378
3440 unlock_new_inode(obj->oo_inode);
3443 if (likely(result == 0)) {
3444 osd_attr_init(info, obj, attr, dof, th);
3445 osd_object_init0(obj);
3448 /* restore previous umask value */
3449 current->fs->umask = umask;
3451 osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3457 * Helper function for osd_create()
3459 * \retval 0, on success
3461 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3462 const struct lu_fid *fid, struct thandle *th)
3464 struct osd_thread_info *info = osd_oti_get(env);
3465 struct osd_inode_id *id = &info->oti_id;
3466 struct osd_device *osd = osd_obj2dev(obj);
3467 struct osd_thandle *oh;
3470 LASSERT(obj->oo_inode != NULL);
3472 oh = container_of(th, struct osd_thandle, ot_super);
3473 LASSERT(oh->ot_handle);
3474 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3476 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3477 rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3478 OI_CHECK_FLD, NULL);
3479 if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP) && osd->od_is_ost) {
3480 struct lu_fid next_fid = *fid;
3482 /* insert next object in advance, and map to the same inode */
3484 if (next_fid.f_oid != 0) {
3485 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3486 osd_oi_insert(info, osd, &next_fid, id, oh->ot_handle,
3487 OI_CHECK_FLD, NULL);
3488 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3492 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3497 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3498 u64 seq, struct lu_seq_range *range)
3500 struct seq_server_site *ss = osd_seq_site(osd);
3502 if (fid_seq_is_idif(seq)) {
3503 fld_range_set_ost(range);
3504 range->lsr_index = idif_ost_idx(seq);
3508 if (!fid_seq_in_fldb(seq)) {
3509 fld_range_set_mdt(range);
3512 * FIXME: If ss is NULL, it suppose not get lsr_index
3515 range->lsr_index = ss->ss_node_id;
3519 LASSERT(ss != NULL);
3520 fld_range_set_any(range);
3521 /* OSD will only do local fld lookup */
3522 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3525 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3526 struct lu_attr *attr,
3527 struct dt_allocation_hint *hint,
3528 struct dt_object_format *dof,
3529 struct thandle *handle)
3531 struct osd_thandle *oh;
3536 LASSERT(handle != NULL);
3538 oh = container_of(handle, struct osd_thandle, ot_super);
3539 LASSERT(oh->ot_handle == NULL);
3542 * EA object consumes more credits than regular object: osd_mk_index
3543 * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3544 * leaf_node, could involves the block, block bitmap, groups, GDT
3545 * change for each block, so add 4 * 2 credits in that case.
3547 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3548 osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3549 (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3551 * Reuse idle OI block may cause additional one OI block
3554 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3555 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3556 if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP))
3557 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3558 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3560 /* will help to find FID->ino mapping at dt_insert() */
3561 rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3569 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3570 attr->la_projid, 1, oh, osd_dt_obj(dt),
3571 NULL, OSD_QID_INODE);
3579 * Called to destroy on-disk representation of the object
3581 * Concurrency: must be locked
3583 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3586 struct osd_object *obj = osd_dt_obj(dt);
3587 struct inode *inode = obj->oo_inode;
3588 struct osd_thandle *oh;
3596 oh = container_of(th, struct osd_thandle, ot_super);
3597 LASSERT(oh->ot_handle == NULL);
3599 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3600 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3602 /* For removing agent entry */
3603 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3604 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3607 * Recycle idle OI leaf may cause additional three OI blocks
3610 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3611 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3612 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3613 /* one less inode */
3614 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3615 i_projid_read(inode), -1, oh, obj, NULL,
3619 /* data to be truncated */
3620 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3621 i_projid_read(inode), 0, oh, obj, NULL,
3627 * will help to find FID->ino when this object is being
3630 rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3635 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3638 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3639 struct osd_object *obj = osd_dt_obj(dt);
3640 struct inode *inode = obj->oo_inode;
3641 struct osd_device *osd = osd_obj2dev(obj);
3642 struct osd_thandle *oh;
3647 oh = container_of(th, struct osd_thandle, ot_super);
3648 LASSERT(oh->ot_handle);
3650 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3652 if (unlikely(fid_is_acct(fid)))
3655 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3656 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3658 CERROR("%s: remove agent entry "DFID": rc = %d\n",
3659 osd_name(osd), PFID(fid), result);
3662 if (S_ISDIR(inode->i_mode)) {
3663 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3664 inode->i_nlink == 2);
3666 spin_lock(&obj->oo_guard);
3668 spin_unlock(&obj->oo_guard);
3669 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3672 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3674 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3676 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3677 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3678 oh->ot_handle, OI_CHECK_FLD);
3680 osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3681 /* XXX: add to ext3 orphan list */
3682 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3684 /* not needed in the cache anymore */
3685 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3686 obj->oo_destroyed = 1;
3692 * Put the fid into lustre_mdt_attrs, and then place the structure
3693 * inode's ea. This fid should not be altered during the life time
3696 * \retval +ve, on success
3697 * \retval -ve, on error
3699 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3701 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3702 const struct lu_fid *fid, __u32 compat, __u32 incompat)
3704 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3705 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3710 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3713 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3716 lustre_loa_init(loa, fid, compat, incompat);
3717 lustre_loa_swab(loa, false);
3720 * For the OST device with 256 bytes inode size by default,
3721 * the PFID EA will be stored together with LMA EA to avoid
3722 * performance trouble. Otherwise the PFID EA can be stored
3723 * independently. LU-8998
3725 if ((compat & LMAC_FID_ON_OST) &&
3726 LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3727 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3728 sizeof(*loa), XATTR_CREATE);
3730 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3731 sizeof(*lma), XATTR_CREATE);
3733 * LMA may already exist, but we need to check that all the
3734 * desired compat/incompat flags have been added.
3736 if (unlikely(rc == -EEXIST)) {
3737 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3738 XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3742 if (rc < sizeof(*lma))
3745 lustre_loa_swab(loa, true);
3746 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3747 ((compat == 0 && incompat == 0) ||
3748 (!(~lma->lma_compat & compat) &&
3749 !(~lma->lma_incompat & incompat))))
3752 lma->lma_self_fid = *fid;
3753 lma->lma_compat |= compat;
3754 lma->lma_incompat |= incompat;
3755 if (rc == sizeof(*lma)) {
3756 lustre_lma_swab(lma);
3757 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3758 sizeof(*lma), XATTR_REPLACE);
3760 lustre_loa_swab(loa, false);
3761 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3762 sizeof(*loa), XATTR_REPLACE);
3770 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3771 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3772 * To have compatilibility with 1.8 ldiskfs driver we need to have
3773 * magic number at start of fid data.
3774 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3777 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3778 const struct lu_fid *fid)
3780 if (!fid_is_namespace_visible(fid) ||
3781 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3782 param->edp_magic = 0;
3786 param->edp_magic = LDISKFS_LUFID_MAGIC;
3787 param->edp_len = sizeof(struct lu_fid) + 1;
3788 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3792 * Try to read the fid from inode ea into dt_rec.
3794 * \param fid object fid.
3796 * \retval 0 on success
3798 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3799 __u32 ino, struct lu_fid *fid,
3800 struct osd_inode_id *id)
3802 struct osd_thread_info *info = osd_oti_get(env);
3803 struct inode *inode;
3807 osd_id_gen(id, ino, OSD_OII_NOGEN);
3808 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3810 RETURN(PTR_ERR(inode));
3816 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3818 struct inode *parent_dir,
3819 const struct lu_fid *dot_fid,
3820 const struct lu_fid *dot_dot_fid,
3821 struct osd_thandle *oth)
3823 struct ldiskfs_dentry_param *dot_ldp;
3824 struct ldiskfs_dentry_param *dot_dot_ldp;
3825 __u32 saved_nlink = dir->i_nlink;
3828 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3829 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3831 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3832 dot_ldp->edp_magic = 0;
3834 rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3835 dir, dot_ldp, dot_dot_ldp);
3837 * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3838 * the subseqent ref_add() will increase the dir->i_nlink
3839 * as 3. That is incorrect for new created directory.
3841 * It looks like hack, because we want to make the OSD API
3842 * to be order-independent for new created directory object
3843 * between dt_insert(..) and ref_add() operations.
3845 * Here, we only restore the in-RAM dir-inode's nlink attr,
3846 * becuase if the nlink attr is not 2, then there will be
3847 * ref_add() called following the dt_insert(..), such call
3848 * will make both the in-RAM and on-disk dir-inode's nlink
3849 * attr to be set as 2. LU-7447
3851 set_nlink(dir, saved_nlink);
3856 * Create an local agent inode for remote entry
3858 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3859 struct osd_device *osd,
3860 struct osd_object *pobj,
3861 const struct lu_fid *fid,
3865 struct osd_thread_info *info = osd_oti_get(env);
3866 struct inode *local;
3867 struct osd_thandle *oh;
3873 oh = container_of(th, struct osd_thandle, ot_super);
3874 LASSERT(oh->ot_handle->h_transaction != NULL);
3876 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type,
3878 if (IS_ERR(local)) {
3879 CERROR("%s: create local error %d\n", osd_name(osd),
3880 (int)PTR_ERR(local));
3885 * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3886 * correct gid on remote file, not agent here
3888 local->i_gid = current_fsgid();
3889 ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3891 /* e2fsck doesn't like empty symlinks. Store remote FID as symlink.
3892 * That gives e2fsck something to look at and be happy, and allows
3893 * debugging if we need to determine where this symlink came from.
3895 if (S_ISLNK(type)) {
3896 BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1);
3897 rc = snprintf((char *)LDISKFS_I(local)->i_data,
3898 LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
3900 i_size_write(local, rc);
3901 LDISKFS_I(local)->i_disksize = rc;
3903 unlock_new_inode(local);
3905 /* Agent inode should not have project ID */
3906 #ifdef HAVE_PROJECT_QUOTA
3907 if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3908 i_projid_read(pobj->oo_inode) != 0) {
3909 rc = osd_transfer_project(local, 0, th);
3911 CERROR("%s: quota transfer failed: rc = %d. Is project "
3912 "quota enforcement enabled on the ldiskfs "
3913 "filesystem?\n", local->i_sb->s_id, rc);
3914 RETURN(ERR_PTR(rc));
3918 /* Set special LMA flag for local agent inode */
3919 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3921 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3922 osd_name(osd), PFID(fid), rc);
3923 RETURN(ERR_PTR(rc));
3929 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3930 lu_object_fid(&pobj->oo_dt.do_lu),
3933 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3934 osd_name(osd), PFID(fid), rc);
3935 RETURN(ERR_PTR(rc));
3942 * when direntry is deleted, we have to take care of possible agent inode
3943 * referenced by that. unfortunately we can't do this at that point:
3944 * iget() within a running transaction leads to deadlock and we better do
3945 * not call that every delete declaration to save performance. so we put
3946 * a potention agent inode on a list and process that once the transaction
3947 * is over. Notice it's not any worse in terms of real orphans as regular
3948 * object destroy doesn't put inodes on the on-disk orphan list. this should
3949 * be addressed separately
3951 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3952 struct osd_thandle *oh,
3955 struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3956 struct osd_obj_orphan *oor;
3963 oor->oor_env = (struct lu_env *)env;
3964 spin_lock(&osd->od_osfs_lock);
3965 list_add(&oor->oor_list, &osd->od_orphan_list);
3966 spin_unlock(&osd->od_osfs_lock);
3968 oh->ot_remove_agents = 1;
3974 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3975 struct osd_device *osd)
3977 struct osd_thread_info *info = osd_oti_get(env);
3978 struct osd_obj_orphan *oor, *tmp;
3979 struct osd_inode_id id;
3981 struct inode *inode;
3986 spin_lock(&osd->od_osfs_lock);
3987 list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3988 if (oor->oor_env == env)
3989 list_move(&oor->oor_list, &list);
3991 spin_unlock(&osd->od_osfs_lock);
3993 list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3997 list_del(&oor->oor_list);
4000 osd_id_gen(&id, ino, OSD_OII_NOGEN);
4001 inode = osd_iget_fid(info, osd, &id, &fid);
4005 if (!osd_remote_fid(env, osd, &fid)) {
4010 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
4012 mark_inode_dirty(inode);
4013 ldiskfs_journal_stop(jh);
4021 * OSD layer object create function for OST objects (b=11826).
4023 * The FID is inserted into inode xattr here.
4025 * \retval 0, on success
4026 * \retval -ve, on error
4028 static int osd_create(const struct lu_env *env, struct dt_object *dt,
4029 struct lu_attr *attr, struct dt_allocation_hint *hint,
4030 struct dt_object_format *dof, struct thandle *th)
4032 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4033 struct osd_object *obj = osd_dt_obj(dt);
4034 struct osd_thread_info *info = osd_oti_get(env);
4035 int result, on_ost = 0;
4039 if (dt_object_exists(dt))
4042 LINVRNT(osd_invariant(obj));
4043 LASSERT(!dt_object_remote(dt));
4044 LASSERT(osd_is_write_locked(env, obj));
4045 LASSERT(th != NULL);
4047 if (unlikely(fid_is_acct(fid)))
4049 * Quota files can't be created from the kernel any more,
4050 * 'tune2fs -O quota' will take care of creating them
4054 result = __osd_create(info, obj, attr, hint, dof, th);
4056 if (fid_is_idif(fid) &&
4057 !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
4058 struct lu_fid *tfid = &info->oti_fid;
4059 struct ost_id *oi = &info->oti_ostid;
4061 fid_to_ostid(fid, oi);
4062 ostid_to_fid(tfid, oi, 0);
4064 result = osd_ea_fid_set(info, obj->oo_inode, tfid,
4065 LMAC_FID_ON_OST, 0);
4067 on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
4069 result = osd_ea_fid_set(info, obj->oo_inode, fid,
4070 on_ost ? LMAC_FID_ON_OST : 0,
4073 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
4074 obj->oo_dt.do_body_ops = &osd_body_ops;
4077 if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
4078 result = __osd_oi_insert(env, obj, fid, th);
4081 * a small optimization - dt_insert() isn't usually applied
4082 * to OST objects, so we don't need to cache OI mapping for
4085 if (result == 0 && on_ost == 0) {
4086 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4088 result = osd_idc_find_and_init(env, osd, obj);
4089 LASSERT(result == 0);
4092 LASSERT(ergo(result == 0,
4093 dt_object_exists(dt) && !dt_object_remote(dt)));
4094 LINVRNT(osd_invariant(obj));
4098 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4099 struct thandle *handle)
4101 struct osd_thandle *oh;
4103 /* it's possible that object doesn't exist yet */
4104 LASSERT(handle != NULL);
4106 oh = container_of(handle, struct osd_thandle, ot_super);
4107 LASSERT(oh->ot_handle == NULL);
4109 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4110 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4112 osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
4118 * Concurrency: @dt is write locked.
4120 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4123 struct osd_object *obj = osd_dt_obj(dt);
4124 struct inode *inode = obj->oo_inode;
4125 struct osd_thandle *oh;
4128 if (!dt_object_exists(dt) || obj->oo_destroyed)
4131 LINVRNT(osd_invariant(obj));
4132 LASSERT(!dt_object_remote(dt));
4133 LASSERT(osd_is_write_locked(env, obj));
4134 LASSERT(th != NULL);
4136 oh = container_of(th, struct osd_thandle, ot_super);
4137 LASSERT(oh->ot_handle != NULL);
4139 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4141 CDEBUG(D_INODE, DFID" increase nlink %d\n",
4142 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4144 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4145 * (65000) subdirectories by storing "1" in i_nlink if the link count
4146 * would otherwise overflow. Directory tranversal tools understand
4147 * that (st_nlink == 1) indicates that the filesystem dose not track
4148 * hard links count on the directory, and will not abort subdirectory
4149 * scanning early once (st_nlink - 2) subdirs have been found.
4151 * This also has to properly handle the case of inodes with nlink == 0
4152 * in case they are being linked into the PENDING directory
4154 spin_lock(&obj->oo_guard);
4155 if (unlikely(inode->i_nlink == 0))
4156 /* inc_nlink from 0 may cause WARN_ON */
4157 set_nlink(inode, 1);
4159 ldiskfs_inc_count(oh->ot_handle, inode);
4160 if (!S_ISDIR(inode->i_mode))
4161 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4163 spin_unlock(&obj->oo_guard);
4165 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4166 LINVRNT(osd_invariant(obj));
4168 osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4173 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4174 struct thandle *handle)
4176 struct osd_thandle *oh;
4178 if (!dt_object_exists(dt))
4181 LASSERT(!dt_object_remote(dt));
4182 LASSERT(handle != NULL);
4184 oh = container_of(handle, struct osd_thandle, ot_super);
4185 LASSERT(oh->ot_handle == NULL);
4187 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4188 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4194 * Concurrency: @dt is write locked.
4196 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4199 struct osd_object *obj = osd_dt_obj(dt);
4200 struct inode *inode = obj->oo_inode;
4201 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4202 struct osd_thandle *oh;
4204 if (!dt_object_exists(dt))
4207 LINVRNT(osd_invariant(obj));
4208 LASSERT(!dt_object_remote(dt));
4209 LASSERT(osd_is_write_locked(env, obj));
4210 LASSERT(th != NULL);
4212 oh = container_of(th, struct osd_thandle, ot_super);
4213 LASSERT(oh->ot_handle != NULL);
4215 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4217 spin_lock(&obj->oo_guard);
4219 * That can be result of upgrade from old Lustre version and
4220 * applied only to local files. Just skip this ref_del call.
4221 * ext4_unlink() only treats this as a warning, don't LASSERT here.
4223 if (inode->i_nlink == 0) {
4224 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4225 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4226 ", maybe an upgraded file? (LU-3915)\n",
4227 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4228 spin_unlock(&obj->oo_guard);
4232 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4233 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4235 ldiskfs_dec_count(oh->ot_handle, inode);
4236 spin_unlock(&obj->oo_guard);
4238 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4239 LINVRNT(osd_invariant(obj));
4241 osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4247 * Concurrency: @dt is read locked.
4249 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4250 struct lu_buf *buf, const char *name)
4252 struct osd_object *obj = osd_dt_obj(dt);
4253 struct inode *inode = obj->oo_inode;
4254 struct osd_thread_info *info = osd_oti_get(env);
4255 struct dentry *dentry = &info->oti_obj_dentry;
4256 bool cache_xattr = false;
4261 /* version get is not real XATTR but uses xattr API */
4262 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4263 dt_obj_version_t *ver = buf->lb_buf;
4266 * for version we are just using xattr API but change inode
4269 if (buf->lb_len == 0)
4270 return sizeof(dt_obj_version_t);
4272 if (buf->lb_len < sizeof(dt_obj_version_t))
4275 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4276 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4278 *ver = LDISKFS_I(inode)->i_fs_version;
4280 return sizeof(dt_obj_version_t);
4283 if (!dt_object_exists(dt))
4286 LASSERT(!dt_object_remote(dt));
4287 LASSERT(inode->i_op != NULL);
4288 #ifdef HAVE_IOP_XATTR
4289 LASSERT(inode->i_op->getxattr != NULL);
4292 if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4293 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4297 rc = osd_oxc_get(obj, name, buf);
4302 if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4303 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4304 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4305 struct filter_fid *ff;
4306 struct ost_layout *ol;
4308 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4312 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4315 if (buf->lb_len == 0 || !buf->lb_buf)
4318 if (buf->lb_len < rc)
4322 ol = &ff->ff_layout;
4323 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4324 PFID_STRIPE_IDX_BITS);
4325 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4326 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4327 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4328 if (lma->lma_compat & LMAC_COMP_INFO) {
4329 ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4330 ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4331 ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4333 ol->ol_comp_start = 0;
4334 ol->ol_comp_end = 0;
4338 rc = __osd_xattr_get(inode, dentry, name,
4339 buf->lb_buf, buf->lb_len);
4343 if (rc == -ENOENT || rc == -ENODATA)
4344 osd_oxc_add(obj, name, NULL, 0);
4345 else if (rc > 0 && buf->lb_buf != NULL)
4346 osd_oxc_add(obj, name, buf->lb_buf, rc);
4352 static int osd_declare_xattr_set(const struct lu_env *env,
4353 struct dt_object *dt,
4354 const struct lu_buf *buf, const char *name,
4355 int fl, struct thandle *handle)
4357 struct osd_thandle *oh;
4359 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4361 LASSERT(handle != NULL);
4363 oh = container_of(handle, struct osd_thandle, ot_super);
4364 LASSERT(oh->ot_handle == NULL);
4366 if (strcmp(name, XATTR_NAME_LMA) == 0) {
4368 * For non-upgrading case, the LMA is set first and
4369 * usually fit inode. But for upgrade case, the LMA
4370 * may be in another separated EA block.
4372 if (dt_object_exists(dt)) {
4373 if (fl == LU_XATTR_REPLACE)
4378 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4380 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4381 /* We may need to delete the old PFID EA. */
4382 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4383 if (fl == LU_XATTR_REPLACE)
4389 * If some name entry resides on remote MDT, then will create
4390 * agent entry under remote parent. On the other hand, if the
4391 * remote entry will be removed, then related agent entry may
4392 * need to be removed from the remote parent. So there may be
4393 * kinds of cases, let's declare enough credits. The credits
4394 * for create agent entry is enough for remove case.
4396 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4397 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4398 if (dt_object_exists(dt))
4399 credits += 1; /* For updating LMA */
4403 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4408 if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4410 * learn xattr size from osd_xattr_get if
4411 * attribute has not been read yet
4413 buflen = __osd_xattr_get(
4414 osd_dt_obj(dt)->oo_inode,
4415 &osd_oti_get(env)->oti_obj_dentry,
4420 buflen = buf->lb_len;
4423 if (buflen > sb->s_blocksize) {
4424 credits += osd_calc_bkmap_credits(
4426 (buflen + sb->s_blocksize - 1) >>
4427 sb->s_blocksize_bits);
4431 * xattr set may involve inode quota change, reserve credits for
4432 * dquot_initialize()
4434 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4437 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4442 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4443 const struct lu_buf *buf, int fl,
4444 struct thandle *handle)
4446 struct osd_thread_info *info = osd_oti_get(env);
4447 struct dentry *dentry = &info->oti_obj_dentry;
4448 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4449 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4450 struct inode *inode = obj->oo_inode;
4451 struct filter_fid *ff = buf->lb_buf;
4452 struct ost_layout *ol = &ff->ff_layout;
4453 int flags = XATTR_REPLACE;
4458 if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4461 rc = osd_get_lma(info, inode, dentry, loa);
4462 if (rc == -ENODATA) {
4463 /* Usually for upgarding from old device */
4464 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4465 LMAC_FID_ON_OST, 0);
4466 flags = XATTR_CREATE;
4471 if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4472 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4475 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4476 /* Separate PFID EA from LMA */
4477 lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4478 lustre_lma_swab(lma);
4479 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4480 sizeof(*lma), XATTR_REPLACE);
4482 obj->oo_pfid_in_lma = 0;
4483 rc = LU_XATTR_CREATE;
4489 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4493 * Old client does not send stripe information,
4494 * then store the PFID EA on disk separatedly.
4496 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4497 ol->ol_stripe_size == 0))
4500 /* Remove old PFID EA entry firstly. */
4501 dquot_initialize(inode);
4502 rc = ll_vfs_removexattr(dentry, inode, XATTR_NAME_FID);
4503 if (rc == -ENODATA) {
4504 if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4511 fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4512 if (likely(ol->ol_stripe_size != 0)) {
4513 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4514 PFID_STRIPE_IDX_BITS;
4515 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4516 lma->lma_compat |= LMAC_STRIPE_INFO;
4517 if (ol->ol_comp_id != 0) {
4518 loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4519 loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4520 loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4521 lma->lma_compat |= LMAC_COMP_INFO;
4525 lustre_loa_swab(loa, false);
4527 /* Store the PFID EA inside LMA. */
4528 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4531 obj->oo_pfid_in_lma = 1;
4537 * In DNE environment, the object (in spite of regular file or directory)
4538 * and its name entry may reside on different MDTs. Under such case, we will
4539 * create an agent entry on the MDT where the object resides. The agent entry
4540 * references the object locally, that makes the object to be visible to the
4541 * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4542 * such as 'tar' can handle the object properly.
4544 * We handle the agent entry during set linkEA that is the common interface
4545 * for both regular file and directroy, can handle kinds of cases, such as
4546 * create/link/unlink/rename, and so on.
4548 * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4550 * XXX: There are two known issues:
4551 * 1. For one object, we will create at most one agent entry even if there
4552 * may be more than one cross-MDTs hard links on the object. So the local
4553 * e2fsck may claim that the object's nlink is larger than the name entries
4554 * that reference such inode. And in further, the e2fsck will fix the nlink
4555 * attribute to match the local references. Then it will cause the object's
4556 * nlink attribute to be inconsistent with the global references. it is bad
4557 * but not fatal. The ref_del() can handle the zero-referenced case. On the
4558 * other hand, the global namespace LFSCK can repair the object's attribute
4559 * according to the linkEA.
4560 * 2. There may be too many hard links on the object as to its linkEA overflow,
4561 * then the linkEA entry for cross-MDTs reference may be discarded. If such
4562 * case happened, then at this point, we do not know whether there are some
4563 * cross-MDTs reference. But there are local references, it guarantees that
4564 * object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4566 static int osd_xattr_handle_linkea(const struct lu_env *env,
4567 struct osd_device *osd,
4568 struct osd_object *obj,
4569 const struct lu_buf *buf,
4570 struct thandle *handle)
4572 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4573 struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4574 struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4575 struct lu_name tmpname;
4576 struct osd_thandle *oh;
4578 bool remote = false;
4582 oh = container_of(handle, struct osd_thandle, ot_super);
4583 LASSERT(oh->ot_handle != NULL);
4585 rc = linkea_init_with_rec(&ldata);
4587 linkea_first_entry(&ldata);
4588 while (ldata.ld_lee != NULL && !remote) {
4589 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4591 if (osd_remote_fid(env, osd, tfid) > 0)
4594 linkea_next_entry(&ldata);
4596 } else if (rc == -ENODATA) {
4602 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4603 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4605 CERROR("%s: failed to remove agent entry for "DFID
4606 ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4607 } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4608 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4610 CERROR("%s: failed to create agent entry for "DFID
4611 ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4618 * Concurrency: @dt is write locked.
4620 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4621 const struct lu_buf *buf, const char *name, int fl,
4622 struct thandle *handle)
4624 struct osd_object *obj = osd_dt_obj(dt);
4625 struct osd_device *osd = osd_obj2dev(obj);
4626 struct inode *inode = obj->oo_inode;
4627 struct osd_thread_info *info = osd_oti_get(env);
4637 /* version set is not real XATTR */
4638 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4639 dt_obj_version_t *version = buf->lb_buf;
4642 * for version we are just using xattr API but change inode
4645 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4647 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4648 *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4650 LDISKFS_I(inode)->i_fs_version = *version;
4652 * Version is set after all inode operations are finished,
4653 * so we should mark it dirty here
4655 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4660 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4661 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4664 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4667 * For the OST device with 256 bytes inode size by default,
4668 * the PFID EA will be stored together with LMA EA to avoid
4669 * performance trouble. Otherwise the PFID EA can be stored
4670 * independently. LU-8998
4672 if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4673 (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4674 LASSERT(buf->lb_buf);
4676 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4679 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4680 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4681 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4683 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4687 lma->lma_incompat |= LMAI_STRIPED;
4688 lustre_lma_swab(lma);
4689 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4690 sizeof(*lma), XATTR_REPLACE);
4693 } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4694 LASSERT(!osd->od_is_ost);
4696 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4701 if (fl & LU_XATTR_REPLACE)
4702 fs_flags |= XATTR_REPLACE;
4704 if (fl & LU_XATTR_CREATE)
4705 fs_flags |= XATTR_CREATE;
4707 rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4708 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4711 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4712 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4713 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4719 * Concurrency: @dt is read locked.
4721 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4722 const struct lu_buf *buf)
4724 struct osd_object *obj = osd_dt_obj(dt);
4725 struct inode *inode = obj->oo_inode;
4726 struct osd_thread_info *info = osd_oti_get(env);
4727 struct dentry *dentry = &info->oti_obj_dentry;
4729 if (!dt_object_exists(dt))
4732 LASSERT(!dt_object_remote(dt));
4733 LASSERT(inode->i_op != NULL);
4734 LASSERT(inode->i_op->listxattr != NULL);
4736 dentry->d_inode = inode;
4737 dentry->d_sb = inode->i_sb;
4738 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4741 static int osd_declare_xattr_del(const struct lu_env *env,
4742 struct dt_object *dt, const char *name,
4743 struct thandle *handle)
4745 struct osd_thandle *oh;
4746 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4748 LASSERT(!dt_object_remote(dt));
4749 LASSERT(handle != NULL);
4751 oh = container_of(handle, struct osd_thandle, ot_super);
4752 LASSERT(oh->ot_handle == NULL);
4754 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4755 osd_dto_credits_noquota[DTO_XATTR_SET]);
4757 * xattr del may involve inode quota change, reserve credits for
4758 * dquot_initialize()
4760 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4766 * Concurrency: @dt is write locked.
4768 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4769 const char *name, struct thandle *handle)
4771 struct osd_object *obj = osd_dt_obj(dt);
4772 struct inode *inode = obj->oo_inode;
4773 struct osd_thread_info *info = osd_oti_get(env);
4774 struct dentry *dentry = &info->oti_obj_dentry;
4777 if (!dt_object_exists(dt))
4780 LASSERT(!dt_object_remote(dt));
4781 LASSERT(inode->i_op != NULL);
4782 LASSERT(handle != NULL);
4783 #ifdef HAVE_IOP_XATTR
4784 LASSERT(inode->i_op->removexattr != NULL);
4787 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4789 if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4790 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4792 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4793 &info->oti_ost_attrs);
4795 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4797 lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4798 lustre_lma_swab(lma);
4799 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4800 sizeof(*lma), XATTR_REPLACE);
4802 obj->oo_pfid_in_lma = 0;
4805 dquot_initialize(inode);
4806 dentry->d_inode = inode;
4807 dentry->d_sb = inode->i_sb;
4808 rc = ll_vfs_removexattr(dentry, inode, name);
4811 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4814 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4815 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4816 osd_oxc_del(obj, name);
4821 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4822 __u64 start, __u64 end)
4824 struct osd_object *obj = osd_dt_obj(dt);
4825 struct inode *inode = obj->oo_inode;
4826 struct osd_thread_info *info = osd_oti_get(env);
4827 struct dentry *dentry = &info->oti_obj_dentry;
4828 struct file *file = &info->oti_file;
4833 dentry->d_inode = inode;
4834 dentry->d_sb = inode->i_sb;
4835 file->f_path.dentry = dentry;
4836 file->f_mapping = inode->i_mapping;
4837 file->f_op = inode->i_fop;
4838 file->f_inode = inode;
4840 rc = vfs_fsync_range(file, start, end, 0);
4845 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4854 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4855 const struct dt_index_features *feat)
4857 struct iam_descr *descr;
4859 if (osd_object_is_root(o))
4860 return feat == &dt_directory_features;
4862 LASSERT(o->oo_dir != NULL);
4864 descr = o->oo_dir->od_container.ic_descr;
4865 if (feat == &dt_directory_features) {
4866 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4871 return feat->dif_keysize_min <= descr->id_key_size &&
4872 descr->id_key_size <= feat->dif_keysize_max &&
4873 feat->dif_recsize_min <= descr->id_rec_size &&
4874 descr->id_rec_size <= feat->dif_recsize_max &&
4875 !(feat->dif_flags & (DT_IND_VARKEY |
4876 DT_IND_VARREC | DT_IND_NONUNQ)) &&
4877 ergo(feat->dif_flags & DT_IND_UPDATE,
4878 1 /* XXX check that object (and fs) is writable */);
4882 static int osd_iam_container_init(const struct lu_env *env,
4883 struct osd_object *obj,
4884 struct osd_directory *dir)
4886 struct iam_container *bag = &dir->od_container;
4889 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4893 result = iam_container_setup(bag);
4895 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4897 iam_container_fini(bag);
4904 * Concurrency: no external locking is necessary.
4906 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4907 const struct dt_index_features *feat)
4911 struct osd_object *obj = osd_dt_obj(dt);
4913 LINVRNT(osd_invariant(obj));
4915 if (osd_object_is_root(obj)) {
4916 dt->do_index_ops = &osd_index_ea_ops;
4918 } else if (feat == &dt_directory_features) {
4919 dt->do_index_ops = &osd_index_ea_ops;
4920 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4925 } else if (unlikely(feat == &dt_otable_features)) {
4926 dt->do_index_ops = &osd_otable_ops;
4928 } else if (unlikely(feat == &dt_acct_features)) {
4929 dt->do_index_ops = &osd_acct_index_ops;
4932 } else if (!osd_has_index(obj)) {
4933 struct osd_directory *dir;
4934 struct osd_device *osd = osd_obj2dev(obj);
4935 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4940 spin_lock(&obj->oo_guard);
4941 if (obj->oo_dir == NULL)
4945 * Concurrent thread allocated container data.
4948 spin_unlock(&obj->oo_guard);
4950 * Now, that we have container data, serialize its
4953 down_write(&obj->oo_ext_idx_sem);
4955 * recheck under lock.
4958 if (osd_has_index(obj)) {
4963 result = osd_iam_container_init(env, obj, obj->oo_dir);
4964 if (result || feat == &dt_lfsck_namespace_features ||
4965 feat == &dt_lfsck_layout_orphan_features ||
4966 feat == &dt_lfsck_layout_dangling_features)
4969 result = osd_index_register(osd, fid,
4970 feat->dif_keysize_max,
4971 feat->dif_recsize_max);
4973 CWARN("%s: failed to register index "
4975 osd_name(osd), PFID(fid), result);
4976 else if (result > 0)
4979 CDEBUG(D_LFSCK, "%s: index object "DFID
4980 " (%d/%d) registered\n",
4981 osd_name(osd), PFID(fid),
4982 (int)feat->dif_keysize_max,
4983 (int)feat->dif_recsize_max);
4986 up_write(&obj->oo_ext_idx_sem);
4994 if (result == 0 && skip_iam == 0) {
4995 if (!osd_iam_index_probe(env, obj, feat))
4998 LINVRNT(osd_invariant(obj));
5003 static int osd_otable_it_attr_get(const struct lu_env *env,
5004 struct dt_object *dt,
5005 struct lu_attr *attr)
5011 static const struct dt_object_operations osd_obj_ops = {
5012 .do_read_lock = osd_read_lock,
5013 .do_write_lock = osd_write_lock,
5014 .do_read_unlock = osd_read_unlock,
5015 .do_write_unlock = osd_write_unlock,
5016 .do_write_locked = osd_write_locked,
5017 .do_attr_get = osd_attr_get,
5018 .do_declare_attr_set = osd_declare_attr_set,
5019 .do_attr_set = osd_attr_set,
5020 .do_ah_init = osd_ah_init,
5021 .do_declare_create = osd_declare_create,
5022 .do_create = osd_create,
5023 .do_declare_destroy = osd_declare_destroy,
5024 .do_destroy = osd_destroy,
5025 .do_index_try = osd_index_try,
5026 .do_declare_ref_add = osd_declare_ref_add,
5027 .do_ref_add = osd_ref_add,
5028 .do_declare_ref_del = osd_declare_ref_del,
5029 .do_ref_del = osd_ref_del,
5030 .do_xattr_get = osd_xattr_get,
5031 .do_declare_xattr_set = osd_declare_xattr_set,
5032 .do_xattr_set = osd_xattr_set,
5033 .do_declare_xattr_del = osd_declare_xattr_del,
5034 .do_xattr_del = osd_xattr_del,
5035 .do_xattr_list = osd_xattr_list,
5036 .do_object_sync = osd_object_sync,
5037 .do_invalidate = osd_invalidate,
5040 static const struct dt_object_operations osd_obj_otable_it_ops = {
5041 .do_attr_get = osd_otable_it_attr_get,
5042 .do_index_try = osd_index_try,
5045 static int osd_index_declare_iam_delete(const struct lu_env *env,
5046 struct dt_object *dt,
5047 const struct dt_key *key,
5048 struct thandle *handle)
5050 struct osd_thandle *oh;
5052 oh = container_of(handle, struct osd_thandle, ot_super);
5053 LASSERT(oh->ot_handle == NULL);
5055 /* Recycle may cause additional three blocks to be changed. */
5056 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
5057 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
5063 * delete a (key, value) pair from index \a dt specified by \a key
5065 * \param dt osd index object
5066 * \param key key for index
5067 * \param rec record reference
5068 * \param handle transaction handler
5071 * \retval -ve failure
5073 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
5074 const struct dt_key *key,
5075 struct thandle *handle)
5077 struct osd_thread_info *oti = osd_oti_get(env);
5078 struct osd_object *obj = osd_dt_obj(dt);
5079 struct osd_thandle *oh;
5080 struct iam_path_descr *ipd;
5081 struct iam_container *bag = &obj->oo_dir->od_container;
5086 if (!dt_object_exists(dt))
5089 LINVRNT(osd_invariant(obj));
5090 LASSERT(!dt_object_remote(dt));
5091 LASSERT(bag->ic_object == obj->oo_inode);
5092 LASSERT(handle != NULL);
5094 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5096 ipd = osd_idx_ipd_get(env, bag);
5097 if (unlikely(ipd == NULL))
5100 oh = container_of(handle, struct osd_thandle, ot_super);
5101 LASSERT(oh->ot_handle != NULL);
5102 LASSERT(oh->ot_handle->h_transaction != NULL);
5104 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5105 /* swab quota uid/gid provided by caller */
5106 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5107 key = (const struct dt_key *)&oti->oti_quota_id;
5110 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5111 osd_ipd_put(env, bag, ipd);
5112 LINVRNT(osd_invariant(obj));
5113 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5117 static int osd_index_declare_ea_delete(const struct lu_env *env,
5118 struct dt_object *dt,
5119 const struct dt_key *key,
5120 struct thandle *handle)
5122 struct osd_thandle *oh;
5123 struct inode *inode;
5128 LASSERT(!dt_object_remote(dt));
5129 LASSERT(handle != NULL);
5131 oh = container_of(handle, struct osd_thandle, ot_super);
5132 LASSERT(oh->ot_handle == NULL);
5134 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5135 osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5137 inode = osd_dt_obj(dt)->oo_inode;
5141 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5142 i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5147 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5150 struct osd_fid_pack *rec;
5153 if (de->file_type & LDISKFS_DIRENT_LUFID) {
5154 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5155 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5156 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5162 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5163 const struct lu_fid *fid)
5165 struct seq_server_site *ss = osd_seq_site(osd);
5169 /* FID seqs not in FLDB, must be local seq */
5170 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5174 * If FLD is not being initialized yet, it only happens during the
5175 * initialization, likely during mgs initialization, and we assume
5176 * this is local FID.
5178 if (ss == NULL || ss->ss_server_fld == NULL)
5181 /* Only check the local FLDB here */
5182 if (osd_seq_exists(env, osd, fid_seq(fid)))
5188 static void osd_take_care_of_agent(const struct lu_env *env,
5189 struct osd_device *osd,
5190 struct osd_thandle *oh,
5191 struct ldiskfs_dir_entry_2 *de)
5193 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5194 struct osd_idmap_cache *idc;
5195 int rc, schedule = 0;
5197 LASSERT(de != NULL);
5199 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5200 if (likely(rc == 0)) {
5201 idc = osd_idc_find_or_init(env, osd, fid);
5202 if (IS_ERR(idc) || idc->oic_remote)
5204 } else if (rc == -ENODATA) {
5206 * can't get FID, postpone to the end of the
5207 * transaction when iget() is safe
5211 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5214 osd_schedule_agent_inode_removal(env, oh,
5215 le32_to_cpu(de->inode));
5219 * Index delete function for interoperability mode (b11826).
5220 * It will remove the directory entry added by osd_index_ea_insert().
5221 * This entry is needed to maintain name->fid mapping.
5223 * \param key, key i.e. file entry to be deleted
5225 * \retval 0, on success
5226 * \retval -ve, on error
5228 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5229 const struct dt_key *key, struct thandle *handle)
5231 struct osd_object *obj = osd_dt_obj(dt);
5232 struct inode *dir = obj->oo_inode;
5233 struct dentry *dentry;
5234 struct osd_thandle *oh;
5235 struct ldiskfs_dir_entry_2 *de = NULL;
5236 struct buffer_head *bh;
5237 struct htree_lock *hlock = NULL;
5238 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5243 if (!dt_object_exists(dt))
5246 LINVRNT(osd_invariant(obj));
5247 LASSERT(!dt_object_remote(dt));
5248 LASSERT(handle != NULL);
5250 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5252 oh = container_of(handle, struct osd_thandle, ot_super);
5253 LASSERT(oh->ot_handle != NULL);
5254 LASSERT(oh->ot_handle->h_transaction != NULL);
5256 dquot_initialize(dir);
5257 dentry = osd_child_dentry_get(env, obj,
5258 (char *)key, strlen((char *)key));
5260 if (obj->oo_hl_head != NULL) {
5261 hlock = osd_oti_get(env)->oti_hlock;
5262 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5263 dir, LDISKFS_HLOCK_DEL);
5265 down_write(&obj->oo_ext_idx_sem);
5268 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5271 * If this is not the ".." entry, it might be a remote DNE
5272 * entry and we need to check if the FID is for a remote
5273 * MDT. If the FID is not in the directory entry (e.g.
5274 * upgraded 1.8 filesystem without dirdata enabled) then
5275 * we need to get the FID from the LMA. For a remote directory
5276 * there HAS to be an LMA, it cannot be an IGIF inode in this
5279 * Delete the entry before the agent inode in order to
5280 * simplify error handling. At worst an error after deleting
5281 * the entry first might leak the agent inode afterward. The
5282 * reverse would need filesystem abort in case of error deleting
5283 * the entry after the agent had been removed, or leave a
5284 * dangling entry pointing at a random inode.
5286 if (strcmp((char *)key, dotdot) != 0)
5287 osd_take_care_of_agent(env, osd, oh, de);
5288 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5294 if (!rc && fid_is_namespace_visible(lu_object_fid(&dt->do_lu)) &&
5295 obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
5296 /* NB, dirent count may not be accurate, because it's counted
5299 if (obj->oo_dirent_count)
5300 obj->oo_dirent_count--;
5302 obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
5305 ldiskfs_htree_unlock(hlock);
5307 up_write(&obj->oo_ext_idx_sem);
5310 LASSERT(osd_invariant(obj));
5311 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5316 * Lookup index for \a key and copy record to \a rec.
5318 * \param dt osd index object
5319 * \param key key for index
5320 * \param rec record reference
5322 * \retval +ve success : exact mach
5323 * \retval 0 return record with key not greater than \a key
5324 * \retval -ve failure
5326 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5327 struct dt_rec *rec, const struct dt_key *key)
5329 struct osd_object *obj = osd_dt_obj(dt);
5330 struct iam_path_descr *ipd;
5331 struct iam_container *bag = &obj->oo_dir->od_container;
5332 struct osd_thread_info *oti = osd_oti_get(env);
5333 struct iam_iterator *it = &oti->oti_idx_it;
5334 struct iam_rec *iam_rec;
5339 if (!dt_object_exists(dt))
5342 LASSERT(osd_invariant(obj));
5343 LASSERT(!dt_object_remote(dt));
5344 LASSERT(bag->ic_object == obj->oo_inode);
5346 ipd = osd_idx_ipd_get(env, bag);
5350 /* got ipd now we can start iterator. */
5351 iam_it_init(it, bag, 0, ipd);
5353 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5354 /* swab quota uid/gid provided by caller */
5355 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5356 key = (const struct dt_key *)&oti->oti_quota_id;
5359 rc = iam_it_get(it, (struct iam_key *)key);
5361 if (S_ISDIR(obj->oo_inode->i_mode))
5362 iam_rec = (struct iam_rec *)oti->oti_ldp;
5364 iam_rec = (struct iam_rec *)rec;
5366 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5368 if (S_ISDIR(obj->oo_inode->i_mode))
5369 osd_fid_unpack((struct lu_fid *)rec,
5370 (struct osd_fid_pack *)iam_rec);
5371 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5372 osd_quota_unpack(obj, rec);
5377 osd_ipd_put(env, bag, ipd);
5379 LINVRNT(osd_invariant(obj));
5384 static int osd_index_declare_iam_insert(const struct lu_env *env,
5385 struct dt_object *dt,
5386 const struct dt_rec *rec,
5387 const struct dt_key *key,
5388 struct thandle *handle)
5390 struct osd_thandle *oh;
5392 LASSERT(handle != NULL);
5394 oh = container_of(handle, struct osd_thandle, ot_super);
5395 LASSERT(oh->ot_handle == NULL);
5397 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5398 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5404 * Inserts (key, value) pair in \a dt index object.
5406 * \param dt osd index object
5407 * \param key key for index
5408 * \param rec record reference
5409 * \param th transaction handler
5412 * \retval -ve failure
5414 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5415 const struct dt_rec *rec,
5416 const struct dt_key *key, struct thandle *th)
5418 struct osd_object *obj = osd_dt_obj(dt);
5419 struct iam_path_descr *ipd;
5420 struct osd_thandle *oh;
5421 struct iam_container *bag;
5422 struct osd_thread_info *oti = osd_oti_get(env);
5423 struct iam_rec *iam_rec;
5428 if (!dt_object_exists(dt))
5431 LINVRNT(osd_invariant(obj));
5432 LASSERT(!dt_object_remote(dt));
5434 bag = &obj->oo_dir->od_container;
5435 LASSERT(bag->ic_object == obj->oo_inode);
5436 LASSERT(th != NULL);
5438 osd_trans_exec_op(env, th, OSD_OT_INSERT);
5440 ipd = osd_idx_ipd_get(env, bag);
5441 if (unlikely(ipd == NULL))
5444 oh = container_of(th, struct osd_thandle, ot_super);
5445 LASSERT(oh->ot_handle != NULL);
5446 LASSERT(oh->ot_handle->h_transaction != NULL);
5447 if (S_ISDIR(obj->oo_inode->i_mode)) {
5448 iam_rec = (struct iam_rec *)oti->oti_ldp;
5449 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5451 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5452 /* pack quota uid/gid */
5453 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5454 key = (const struct dt_key *)&oti->oti_quota_id;
5455 /* pack quota record */
5456 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5457 iam_rec = (struct iam_rec *)rec;
5459 iam_rec = (struct iam_rec *)rec;
5462 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5464 osd_ipd_put(env, bag, ipd);
5465 LINVRNT(osd_invariant(obj));
5466 osd_trans_exec_check(env, th, OSD_OT_INSERT);
5471 * Calls ldiskfs_add_entry() to add directory entry
5472 * into the directory. This is required for
5473 * interoperability mode (b11826)
5475 * \retval 0, on success
5476 * \retval -ve, on error
5478 static int __osd_ea_add_rec(struct osd_thread_info *info,
5479 struct osd_object *pobj, struct inode *cinode,
5480 const char *name, const struct lu_fid *fid,
5481 struct htree_lock *hlock, struct thandle *th)
5483 struct ldiskfs_dentry_param *ldp;
5484 struct dentry *child;
5485 struct osd_thandle *oth;
5488 oth = container_of(th, struct osd_thandle, ot_super);
5489 LASSERT(oth->ot_handle != NULL);
5490 LASSERT(oth->ot_handle->h_transaction != NULL);
5491 LASSERT(pobj->oo_inode);
5493 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5494 if (unlikely(pobj->oo_inode ==
5495 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5498 osd_get_ldiskfs_dirent_param(ldp, fid);
5499 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5500 child->d_fsdata = (void *)ldp;
5501 dquot_initialize(pobj->oo_inode);
5502 rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5503 child, cinode, hlock);
5504 if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5505 struct ldiskfs_dir_entry_2 *de;
5506 struct buffer_head *bh;
5509 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5512 rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5515 if (S_ISDIR(cinode->i_mode))
5516 de->file_type = LDISKFS_DIRENT_LUFID |
5517 LDISKFS_FT_REG_FILE;
5519 de->file_type = LDISKFS_DIRENT_LUFID |
5521 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5532 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5533 * into the directory.Also sets flags into osd object to
5534 * indicate dot and dotdot are created. This is required for
5535 * interoperability mode (b11826)
5537 * \param dir directory for dot and dotdot fixup.
5538 * \param obj child object for linking
5540 * \retval 0, on success
5541 * \retval -ve, on error
5543 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5544 struct osd_object *dir,
5545 struct inode *parent_dir, const char *name,
5546 const struct lu_fid *dot_fid,
5547 const struct lu_fid *dot_dot_fid,
5550 struct inode *inode = dir->oo_inode;
5551 struct osd_thandle *oth;
5554 oth = container_of(th, struct osd_thandle, ot_super);
5555 LASSERT(oth->ot_handle->h_transaction != NULL);
5556 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5558 if (strcmp(name, dot) == 0) {
5559 if (dir->oo_compat_dot_created) {
5562 LASSERT(inode->i_ino == parent_dir->i_ino);
5563 dir->oo_compat_dot_created = 1;
5566 } else if (strcmp(name, dotdot) == 0) {
5567 if (!dir->oo_compat_dot_created)
5569 /* in case of rename, dotdot is already created */
5570 if (dir->oo_compat_dotdot_created) {
5571 return __osd_ea_add_rec(info, dir, parent_dir, name,
5572 dot_dot_fid, NULL, th);
5575 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5576 struct lu_fid tfid = *dot_dot_fid;
5579 result = osd_add_dot_dotdot_internal(info,
5580 dir->oo_inode, parent_dir, dot_fid,
5583 result = osd_add_dot_dotdot_internal(info,
5584 dir->oo_inode, parent_dir, dot_fid,
5589 dir->oo_compat_dotdot_created = 1;
5597 * It will call the appropriate osd_add* function and return the
5598 * value, return by respective functions.
5600 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5601 struct inode *cinode, const char *name,
5602 const struct lu_fid *fid, struct thandle *th)
5604 struct osd_thread_info *info = osd_oti_get(env);
5605 struct htree_lock *hlock;
5608 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5610 if (name[0] == '.' && (name[1] == '\0' ||
5611 (name[1] == '.' && name[2] == '\0'))) {
5612 if (hlock != NULL) {
5613 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5616 down_write(&pobj->oo_ext_idx_sem);
5619 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5620 lu_object_fid(&pobj->oo_dt.do_lu),
5623 if (hlock != NULL) {
5624 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5625 pobj->oo_inode, LDISKFS_HLOCK_ADD);
5627 down_write(&pobj->oo_ext_idx_sem);
5630 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5631 struct lu_fid *tfid = &info->oti_fid;
5635 rc = __osd_ea_add_rec(info, pobj, cinode, name,
5638 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5642 if (!rc && fid_is_namespace_visible(lu_object_fid(&pobj->oo_dt.do_lu))
5643 && pobj->oo_dirent_count != LU_DIRENT_COUNT_UNSET)
5644 pobj->oo_dirent_count++;
5647 ldiskfs_htree_unlock(hlock);
5649 up_write(&pobj->oo_ext_idx_sem);
5655 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5656 struct osd_idmap_cache *oic)
5658 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5659 struct lu_fid *fid = &oic->oic_fid;
5660 struct osd_inode_id *id = &oic->oic_lid;
5661 struct inode *inode = NULL;
5668 if (!fid_is_norm(fid) && !fid_is_igif(fid))
5671 if (thread_is_running(&scrub->os_thread) &&
5672 scrub->os_pos_current > id->oii_ino)
5675 if (dev->od_auto_scrub_interval == AS_NEVER ||
5676 ktime_get_real_seconds() <
5677 scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5681 rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5682 if (rc == -ENOENT) {
5683 __u32 gen = id->oii_gen;
5689 inode = osd_iget(oti, dev, id);
5690 /* The inode has been removed (by race maybe). */
5691 if (IS_ERR(inode)) {
5692 rc = PTR_ERR(inode);
5694 RETURN(rc == -ESTALE ? -ENOENT : rc);
5697 /* The OI mapping is lost. */
5698 if (gen != OSD_OII_NOGEN)
5702 * The inode may has been reused by others, we do not know,
5703 * leave it to be handled by subsequent osd_fid_lookup().
5706 } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5713 if (thread_is_running(&scrub->os_thread)) {
5714 if (inode == NULL) {
5715 inode = osd_iget(oti, dev, id);
5716 /* The inode has been removed (by race maybe). */
5717 if (IS_ERR(inode)) {
5718 rc = PTR_ERR(inode);
5720 RETURN(rc == -ESTALE ? -ENOENT : rc);
5724 rc = osd_oii_insert(dev, oic, insert);
5726 * There is race condition between osd_oi_lookup and OI scrub.
5727 * The OI scrub finished just after osd_oi_lookup() failure.
5728 * Under such case, it is unnecessary to trigger OI scrub again,
5729 * but try to call osd_oi_lookup() again.
5731 if (unlikely(rc == -EAGAIN))
5734 if (!S_ISDIR(inode->i_mode))
5737 rc = osd_check_lmv(oti, dev, inode, oic);
5742 if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5743 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5744 SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5745 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5746 "%s: trigger partial OI scrub for RPC inconsistency "
5747 "checking FID "DFID": rc = %d\n",
5748 osd_dev2name(dev), PFID(fid), rc);
5749 if (rc == 0 || rc == -EALREADY)
5762 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5763 struct osd_device *dev,
5764 struct osd_idmap_cache *oic,
5765 struct lu_fid *fid, __u32 ino)
5767 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5768 struct inode *inode;
5771 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5772 inode = osd_iget(oti, dev, &oic->oic_lid);
5773 if (IS_ERR(inode)) {
5774 fid_zero(&oic->oic_fid);
5775 return PTR_ERR(inode);
5778 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5781 fid_zero(&oic->oic_fid);
5783 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5787 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5788 struct osd_inode_id *id, const struct lu_fid *fid)
5790 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5791 id->oii_ino, id->oii_gen, info);
5792 info->oti_cache.oic_lid = *id;
5793 info->oti_cache.oic_fid = *fid;
5794 info->oti_cache.oic_dev = osd;
5798 * Get parent FID from the linkEA.
5800 * For a directory which parent resides on remote MDT, to satisfy the
5801 * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5802 * the other hand, to make the lookup(..) on the directory can return
5803 * the real parent FID, we append the real parent FID after its ".."
5804 * name entry in the /REMOTE_PARENT_DIR.
5806 * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5807 * backup. So after the restore, we cannot get the right parent FID from
5808 * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5809 * we have stored the real parent FID in the directory object's linkEA,
5810 * we can parse the linkEA for the real parent FID.
5812 * \param[in] env pointer to the thread context
5813 * \param[in] obj pointer to the object to be handled
5814 * \param[out]fid pointer to the buffer to hold the parent FID
5816 * \retval 0 for getting the real parent FID successfully
5817 * \retval negative error number on failure
5819 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5820 struct osd_object *obj,
5823 struct osd_thread_info *oti = osd_oti_get(env);
5824 struct lu_buf *buf = &oti->oti_big_buf;
5825 struct dentry *dentry = &oti->oti_obj_dentry;
5826 struct inode *inode = obj->oo_inode;
5827 struct linkea_data ldata = { NULL };
5833 if (!S_ISDIR(inode->i_mode))
5837 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5838 buf->lb_buf, buf->lb_len);
5839 if (rc == -ERANGE) {
5840 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5843 lu_buf_realloc(buf, rc);
5844 if (buf->lb_buf == NULL)
5851 if (unlikely(rc == 0))
5857 if (unlikely(buf->lb_buf == NULL)) {
5858 lu_buf_realloc(buf, rc);
5859 if (buf->lb_buf == NULL)
5866 rc = linkea_init_with_rec(&ldata);
5868 linkea_first_entry(&ldata);
5869 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5875 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5876 struct inode *inode,
5877 const struct lu_fid *pfid,
5878 const char *name, const int namelen)
5880 struct osd_thread_info *oti = osd_oti_get(env);
5881 struct lu_buf *buf = &oti->oti_big_buf;
5882 struct dentry *dentry = &oti->oti_obj_dentry;
5883 struct linkea_data ldata = { NULL };
5884 struct lu_name cname = { .ln_name = name,
5885 .ln_namelen = namelen };
5891 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5892 buf->lb_buf, buf->lb_len);
5894 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5899 if (unlikely(rc == 0))
5902 if (buf->lb_len < rc) {
5903 lu_buf_realloc(buf, rc);
5904 if (buf->lb_buf == NULL)
5911 rc = linkea_init_with_rec(&ldata);
5913 rc = linkea_links_find(&ldata, &cname, pfid);
5919 * Calls ->lookup() to find dentry. From dentry get inode and
5920 * read inode's ea to get fid. This is required for interoperability
5923 * \retval 0, on success
5924 * \retval -ve, on error
5926 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5927 struct dt_rec *rec, const struct dt_key *key)
5929 struct inode *dir = obj->oo_inode;
5930 struct dentry *dentry;
5931 struct ldiskfs_dir_entry_2 *de;
5932 struct buffer_head *bh;
5933 struct lu_fid *fid = (struct lu_fid *)rec;
5934 struct htree_lock *hlock = NULL;
5940 LASSERT(dir->i_op != NULL);
5941 LASSERT(dir->i_op->lookup != NULL);
5943 dentry = osd_child_dentry_get(env, obj,
5944 (char *)key, strlen((char *)key));
5946 if (obj->oo_hl_head != NULL) {
5947 hlock = osd_oti_get(env)->oti_hlock;
5948 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5949 dir, LDISKFS_HLOCK_LOOKUP);
5951 down_read(&obj->oo_ext_idx_sem);
5954 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5956 struct osd_thread_info *oti = osd_oti_get(env);
5957 struct osd_inode_id *id = &oti->oti_id;
5958 struct osd_idmap_cache *oic = &oti->oti_cache;
5959 struct osd_device *dev = osd_obj2dev(obj);
5961 ino = le32_to_cpu(de->inode);
5962 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5964 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5968 rc = osd_get_fid_from_dentry(de, rec);
5970 /* done with de, release bh */
5973 if (unlikely(is_remote_parent_ino(dev, ino))) {
5974 const char *name = (const char *)key;
5977 * If the parent is on remote MDT, and there
5978 * is no FID-in-dirent, then we have to get
5979 * the parent FID from the linkEA.
5981 if (likely(strlen(name) == 2 &&
5982 name[0] == '.' && name[1] == '.'))
5983 rc = osd_get_pfid_from_linkea(env, obj,
5986 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5989 osd_id_gen(id, ino, OSD_OII_NOGEN);
5992 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5993 fid_zero(&oic->oic_fid);
5998 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5999 rc = osd_consistency_check(oti, dev, oic);
6001 fid_zero(&oic->oic_fid);
6003 /* Other error should not affect lookup result. */
6013 ldiskfs_htree_unlock(hlock);
6015 up_read(&obj->oo_ext_idx_sem);
6019 static int osd_index_declare_ea_insert(const struct lu_env *env,
6020 struct dt_object *dt,
6021 const struct dt_rec *rec,
6022 const struct dt_key *key,
6023 struct thandle *handle)
6025 struct osd_thandle *oh;
6026 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6027 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6028 const struct lu_fid *fid = rec1->rec_fid;
6029 int credits, rc = 0;
6030 struct osd_idmap_cache *idc;
6034 LASSERT(!dt_object_remote(dt));
6035 LASSERT(handle != NULL);
6036 LASSERT(fid != NULL);
6037 LASSERT(rec1->rec_type != 0);
6039 oh = container_of(handle, struct osd_thandle, ot_super);
6040 LASSERT(oh->ot_handle == NULL);
6042 credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
6045 * we can't call iget() while a transactions is running
6046 * (this can lead to a deadlock), but we need to know
6047 * inum and object type. so we find this information at
6048 * declaration and cache in per-thread info
6050 idc = osd_idc_find_or_init(env, osd, fid);
6052 RETURN(PTR_ERR(idc));
6053 if (idc->oic_remote) {
6055 * a reference to remote inode is represented by an
6056 * agent inode which we have to create
6058 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
6059 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
6062 osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
6064 if (osd_dt_obj(dt)->oo_inode != NULL) {
6065 struct inode *inode = osd_dt_obj(dt)->oo_inode;
6068 * We ignore block quota on meta pool (MDTs), so needn't
6069 * calculate how many blocks will be consumed by this index
6072 rc = osd_declare_inode_qid(env, i_uid_read(inode),
6074 i_projid_read(inode), 0,
6075 oh, osd_dt_obj(dt), NULL,
6080 #ifdef HAVE_PROJECT_QUOTA
6082 * Reserve credits for local agent inode to transfer
6083 * to 0, quota enforcement is ignored in this case.
6085 if (idc->oic_remote &&
6086 LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
6087 i_projid_read(inode) != 0)
6088 rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
6089 0, i_projid_read(inode),
6090 0, false, PRJQUOTA, true);
6098 * Index add function for interoperability mode (b11826).
6099 * It will add the directory entry.This entry is needed to
6100 * maintain name->fid mapping.
6102 * \param key it is key i.e. file entry to be inserted
6103 * \param rec it is value of given key i.e. fid
6105 * \retval 0, on success
6106 * \retval -ve, on error
6108 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6109 const struct dt_rec *rec,
6110 const struct dt_key *key, struct thandle *th)
6112 struct osd_object *obj = osd_dt_obj(dt);
6113 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6114 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6115 const struct lu_fid *fid = rec1->rec_fid;
6116 const char *name = (const char *)key;
6117 struct osd_thread_info *oti = osd_oti_get(env);
6118 struct inode *child_inode = NULL;
6119 struct osd_idmap_cache *idc;
6124 if (!dt_object_exists(dt))
6127 LASSERT(osd_invariant(obj));
6128 LASSERT(!dt_object_remote(dt));
6129 LASSERT(th != NULL);
6131 osd_trans_exec_op(env, th, OSD_OT_INSERT);
6133 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6135 idc = osd_idc_find(env, osd, fid);
6136 if (unlikely(idc == NULL)) {
6137 idc = osd_idc_find_or_init(env, osd, fid);
6140 * this dt_insert() wasn't declared properly, so
6141 * FID is missing in OI cache. we better do not
6142 * lookup FID in FLDB/OI and don't risk to deadlock,
6143 * but in some special cases (lfsck testing, etc)
6144 * it's much simpler than fixing a caller.
6146 * normally this error should be placed after the first
6147 * find, but migrate may attach source stripes to
6148 * target, which doesn't create stripes.
6150 CERROR("%s: "DFID" wasn't declared for insert\n",
6151 osd_name(osd), PFID(fid));
6153 RETURN(PTR_ERR(idc));
6157 if (idc->oic_remote) {
6158 /* Insert remote entry */
6159 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6161 igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6163 child_inode = osd_create_local_agent_inode(env, osd,
6164 obj, fid, rec1->rec_type & S_IFMT, th);
6165 if (IS_ERR(child_inode))
6166 RETURN(PTR_ERR(child_inode));
6169 /* Insert local entry */
6170 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6171 /* for a reason OI cache wasn't filled properly */
6172 CERROR("%s: OIC for "DFID" isn't filled\n",
6173 osd_name(osd), PFID(fid));
6176 child_inode = oti->oti_inode;
6177 if (unlikely(child_inode == NULL)) {
6178 struct ldiskfs_inode_info *lii;
6183 child_inode = oti->oti_inode = &lii->vfs_inode;
6185 child_inode->i_sb = osd_sb(osd);
6186 child_inode->i_ino = idc->oic_lid.oii_ino;
6187 child_inode->i_mode = rec1->rec_type & S_IFMT;
6190 rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6192 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6193 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6195 if (child_inode && child_inode != oti->oti_inode)
6197 LASSERT(osd_invariant(obj));
6198 osd_trans_exec_check(env, th, OSD_OT_INSERT);
6204 * Initialize osd Iterator for given osd index object.
6206 * \param dt osd index object
6209 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6210 struct dt_object *dt,
6213 struct osd_it_iam *it;
6214 struct osd_object *obj = osd_dt_obj(dt);
6215 struct lu_object *lo = &dt->do_lu;
6216 struct iam_path_descr *ipd;
6217 struct iam_container *bag = &obj->oo_dir->od_container;
6219 if (!dt_object_exists(dt))
6220 return ERR_PTR(-ENOENT);
6224 return ERR_PTR(-ENOMEM);
6226 ipd = osd_it_ipd_get(env, bag);
6227 if (likely(ipd != NULL)) {
6231 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6232 return (struct dt_it *)it;
6235 return ERR_PTR(-ENOMEM);
6240 * free given Iterator.
6242 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6244 struct osd_it_iam *it = (struct osd_it_iam *)di;
6245 struct osd_object *obj = it->oi_obj;
6247 iam_it_fini(&it->oi_it);
6248 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6249 osd_object_put(env, obj);
6254 * Move Iterator to record specified by \a key
6256 * \param di osd iterator
6257 * \param key key for index
6259 * \retval +ve di points to record with least key not larger than key
6260 * \retval 0 di points to exact matched key
6261 * \retval -ve failure
6264 static int osd_it_iam_get(const struct lu_env *env,
6265 struct dt_it *di, const struct dt_key *key)
6267 struct osd_thread_info *oti = osd_oti_get(env);
6268 struct osd_it_iam *it = (struct osd_it_iam *)di;
6270 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6271 /* swab quota uid/gid */
6272 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6273 key = (struct dt_key *)&oti->oti_quota_id;
6276 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6282 * \param di osd iterator
6284 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6286 struct osd_it_iam *it = (struct osd_it_iam *)di;
6288 iam_it_put(&it->oi_it);
6292 * Move iterator by one record
6294 * \param di osd iterator
6296 * \retval +1 end of container reached
6298 * \retval -ve failure
6301 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6303 struct osd_it_iam *it = (struct osd_it_iam *)di;
6305 return iam_it_next(&it->oi_it);
6309 * Return pointer to the key under iterator.
6312 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6313 const struct dt_it *di)
6315 struct osd_thread_info *oti = osd_oti_get(env);
6316 struct osd_it_iam *it = (struct osd_it_iam *)di;
6317 struct osd_object *obj = it->oi_obj;
6320 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6322 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6323 /* swab quota uid/gid */
6324 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6325 key = (struct dt_key *)&oti->oti_quota_id;
6332 * Return size of key under iterator (in bytes)
6335 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6337 struct osd_it_iam *it = (struct osd_it_iam *)di;
6339 return iam_it_key_size(&it->oi_it);
6343 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6345 /* check if file type is required */
6346 if (ent->lde_attrs & LUDA_TYPE) {
6347 struct luda_type *lt;
6348 int align = sizeof(*lt) - 1;
6350 len = (len + align) & ~align;
6351 lt = (struct luda_type *)(ent->lde_name + len);
6352 lt->lt_type = cpu_to_le16(DTTOIF(type));
6355 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6359 * build lu direct from backend fs dirent.
6363 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6364 char *name, __u16 namelen, __u16 type, __u32 attr)
6366 ent->lde_attrs = attr | LUDA_FID;
6367 fid_cpu_to_le(&ent->lde_fid, fid);
6369 ent->lde_hash = cpu_to_le64(offset);
6370 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6372 strncpy(ent->lde_name, name, namelen);
6373 ent->lde_name[namelen] = '\0';
6374 ent->lde_namelen = cpu_to_le16(namelen);
6376 /* append lustre attributes */
6377 osd_it_append_attrs(ent, namelen, type);
6381 * Return pointer to the record under iterator.
6383 static int osd_it_iam_rec(const struct lu_env *env,
6384 const struct dt_it *di,
6385 struct dt_rec *dtrec, __u32 attr)
6387 struct osd_it_iam *it = (struct osd_it_iam *)di;
6388 struct osd_thread_info *info = osd_oti_get(env);
6392 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6393 const struct osd_fid_pack *rec;
6394 struct lu_fid *fid = &info->oti_fid;
6395 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6401 name = (char *)iam_it_key_get(&it->oi_it);
6403 RETURN(PTR_ERR(name));
6405 namelen = iam_it_key_size(&it->oi_it);
6407 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6409 RETURN(PTR_ERR(rec));
6411 rc = osd_fid_unpack(fid, rec);
6415 hash = iam_it_store(&it->oi_it);
6417 /* IAM does not store object type in IAM index (dir) */
6418 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6420 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6421 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6422 (struct iam_rec *)dtrec);
6423 osd_quota_unpack(it->oi_obj, dtrec);
6425 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6426 (struct iam_rec *)dtrec);
6433 * Returns cookie for current Iterator position.
6435 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6437 struct osd_it_iam *it = (struct osd_it_iam *)di;
6439 return iam_it_store(&it->oi_it);
6443 * Restore iterator from cookie.
6445 * \param di osd iterator
6446 * \param hash Iterator location cookie
6448 * \retval +ve di points to record with least key not larger than key.
6449 * \retval 0 di points to exact matched key
6450 * \retval -ve failure
6453 static int osd_it_iam_load(const struct lu_env *env,
6454 const struct dt_it *di, __u64 hash)
6456 struct osd_it_iam *it = (struct osd_it_iam *)di;
6458 return iam_it_load(&it->oi_it, hash);
6461 static const struct dt_index_operations osd_index_iam_ops = {
6462 .dio_lookup = osd_index_iam_lookup,
6463 .dio_declare_insert = osd_index_declare_iam_insert,
6464 .dio_insert = osd_index_iam_insert,
6465 .dio_declare_delete = osd_index_declare_iam_delete,
6466 .dio_delete = osd_index_iam_delete,
6468 .init = osd_it_iam_init,
6469 .fini = osd_it_iam_fini,
6470 .get = osd_it_iam_get,
6471 .put = osd_it_iam_put,
6472 .next = osd_it_iam_next,
6473 .key = osd_it_iam_key,
6474 .key_size = osd_it_iam_key_size,
6475 .rec = osd_it_iam_rec,
6476 .store = osd_it_iam_store,
6477 .load = osd_it_iam_load
6483 * Creates or initializes iterator context.
6485 * \retval struct osd_it_ea, iterator structure on success
6488 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6489 struct dt_object *dt,
6492 struct osd_object *obj = osd_dt_obj(dt);
6493 struct osd_thread_info *info = osd_oti_get(env);
6494 struct osd_it_ea *oie;
6496 struct lu_object *lo = &dt->do_lu;
6497 struct dentry *obj_dentry;
6501 if (!dt_object_exists(dt) || obj->oo_destroyed)
6502 RETURN(ERR_PTR(-ENOENT));
6504 OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6506 RETURN(ERR_PTR(-ENOMEM));
6507 obj_dentry = &oie->oie_dentry;
6509 obj_dentry->d_inode = obj->oo_inode;
6510 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6511 obj_dentry->d_name.hash = 0;
6513 oie->oie_rd_dirent = 0;
6514 oie->oie_it_dirent = 0;
6515 oie->oie_dirent = NULL;
6516 if (unlikely(!info->oti_it_ea_buf_used)) {
6517 oie->oie_buf = info->oti_it_ea_buf;
6518 info->oti_it_ea_buf_used = 1;
6520 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6521 if (oie->oie_buf == NULL)
6522 RETURN(ERR_PTR(-ENOMEM));
6526 file = &oie->oie_file;
6528 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6529 if (attr & LUDA_64BITHASH)
6530 file->f_mode = FMODE_64BITHASH;
6532 file->f_mode = FMODE_32BITHASH;
6533 file->f_path.dentry = obj_dentry;
6534 file->f_mapping = obj->oo_inode->i_mapping;
6535 file->f_op = obj->oo_inode->i_fop;
6536 file->f_inode = obj->oo_inode;
6539 RETURN((struct dt_it *)oie);
6543 * Destroy or finishes iterator context.
6545 * \param di iterator structure to be destroyed
6547 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6549 struct osd_thread_info *info = osd_oti_get(env);
6550 struct osd_it_ea *oie = (struct osd_it_ea *)di;
6551 struct osd_object *obj = oie->oie_obj;
6552 struct inode *inode = obj->oo_inode;
6555 oie->oie_file.f_op->release(inode, &oie->oie_file);
6556 osd_object_put(env, obj);
6557 if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6558 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6560 info->oti_it_ea_buf_used = 0;
6561 OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6566 * It position the iterator at given key, so that next lookup continues from
6567 * that key Or it is similar to dio_it->load() but based on a key,
6568 * rather than file position.
6570 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6573 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6575 static int osd_it_ea_get(const struct lu_env *env,
6576 struct dt_it *di, const struct dt_key *key)
6578 struct osd_it_ea *it = (struct osd_it_ea *)di;
6581 LASSERT(((const char *)key)[0] == '\0');
6582 it->oie_file.f_pos = 0;
6583 it->oie_rd_dirent = 0;
6584 it->oie_it_dirent = 0;
6585 it->oie_dirent = NULL;
6593 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6597 struct osd_filldir_cbs {
6598 struct dir_context ctx;
6599 struct osd_it_ea *it;
6602 * It is called internally by ->iterate*(). It fills the
6603 * iterator's in-memory data structure with required
6604 * information i.e. name, namelen, rec_size etc.
6606 * \param buf in which information to be filled in.
6607 * \param name name of the file in given dir
6609 * \retval 0 on success
6610 * \retval 1 on buffer full
6612 #ifdef HAVE_FILLDIR_USE_CTX
6613 static int osd_ldiskfs_filldir(struct dir_context *buf,
6615 static int osd_ldiskfs_filldir(void *buf,
6617 const char *name, int namelen,
6618 loff_t offset, __u64 ino, unsigned int d_type)
6620 struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6621 struct osd_object *obj = it->oie_obj;
6622 struct osd_it_ea_dirent *ent = it->oie_dirent;
6623 struct lu_fid *fid = &ent->oied_fid;
6624 struct osd_fid_pack *rec;
6628 /* this should never happen */
6629 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6630 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6634 if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6638 /* "." is just the object itself. */
6639 if (namelen == 1 && name[0] == '.') {
6640 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6641 } else if (d_type & LDISKFS_DIRENT_LUFID) {
6642 rec = (struct osd_fid_pack *)(name + namelen + 1);
6643 if (osd_fid_unpack(fid, rec) != 0)
6648 d_type &= ~LDISKFS_DIRENT_LUFID;
6650 /* NOT export local root. */
6651 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6652 ino = obj->oo_inode->i_ino;
6653 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6656 ent->oied_ino = ino;
6657 ent->oied_off = offset;
6658 ent->oied_namelen = namelen;
6659 ent->oied_type = d_type;
6661 memcpy(ent->oied_name, name, namelen);
6663 it->oie_rd_dirent++;
6664 it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6669 * Calls ->iterate*() to load a directory entry at a time
6670 * and stored it in iterator's in-memory data structure.
6672 * \param di iterator's in memory structure
6674 * \retval 0 on success
6675 * \retval -ve on error
6676 * \retval +1 reach the end of entry
6678 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6679 const struct dt_it *di)
6681 struct osd_it_ea *it = (struct osd_it_ea *)di;
6682 struct osd_object *obj = it->oie_obj;
6683 struct inode *inode = obj->oo_inode;
6684 struct htree_lock *hlock = NULL;
6685 struct file *filp = &it->oie_file;
6687 struct osd_filldir_cbs buf = {
6688 .ctx.actor = osd_ldiskfs_filldir,
6693 it->oie_dirent = it->oie_buf;
6694 it->oie_rd_dirent = 0;
6696 if (obj->oo_hl_head != NULL) {
6697 hlock = osd_oti_get(env)->oti_hlock;
6698 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6699 inode, LDISKFS_HLOCK_READDIR);
6701 down_read(&obj->oo_ext_idx_sem);
6704 filp->f_cred = current_cred();
6705 rc = osd_security_file_alloc(filp);
6709 filp->f_flags |= O_NOATIME;
6710 filp->f_mode |= FMODE_NONOTIFY;
6711 rc = iterate_dir(filp, &buf.ctx);
6716 ldiskfs_htree_unlock(hlock);
6718 up_read(&obj->oo_ext_idx_sem);
6720 if (it->oie_rd_dirent == 0) {
6722 * If it does not get any dirent, it means it has been reached
6723 * to the end of the dir
6725 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6729 it->oie_dirent = it->oie_buf;
6730 it->oie_it_dirent = 1;
6737 * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
6738 * to load a directory entry at a time and stored it in
6739 * iterator's in-memory data structure.
6741 * \param di iterator's in memory structure
6743 * \retval +ve iterator reached to end
6744 * \retval 0 iterator not reached to end
6745 * \retval -ve on error
6747 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6749 struct osd_it_ea *it = (struct osd_it_ea *)di;
6754 if (it->oie_it_dirent < it->oie_rd_dirent) {
6756 (void *)it->oie_dirent +
6757 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6758 it->oie_dirent->oied_namelen);
6759 it->oie_it_dirent++;
6762 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6765 rc = osd_ldiskfs_it_fill(env, di);
6772 * Returns the key at current position from iterator's in memory structure.
6774 * \param di iterator's in memory structure
6776 * \retval key i.e. struct dt_key on success
6778 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6779 const struct dt_it *di)
6781 struct osd_it_ea *it = (struct osd_it_ea *)di;
6783 return (struct dt_key *)it->oie_dirent->oied_name;
6787 * Returns key's size at current position from iterator's in memory structure.
6789 * \param di iterator's in memory structure
6791 * \retval key_size i.e. struct dt_key on success
6793 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6795 struct osd_it_ea *it = (struct osd_it_ea *)di;
6797 return it->oie_dirent->oied_namelen;
6800 #if defined LDISKFS_DIR_ENTRY_LEN && defined LDISKFS_DIR_ENTRY_LEN_
6801 #undef LDISKFS_DIR_REC_LEN
6802 #define LDISKFS_DIR_REC_LEN(de) LDISKFS_DIR_ENTRY_LEN_((de))
6805 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6807 if (LDISKFS_DIR_REC_LEN(de) >=
6808 __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6815 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6816 unsigned int blocksize, bool dotdot)
6819 return osd_dotdot_has_space(de);
6821 if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6822 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6829 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6830 handle_t *jh, struct dentry *dentry,
6831 const struct lu_fid *fid, struct buffer_head *bh,
6832 struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6835 struct inode *dir = dentry->d_parent->d_inode;
6836 struct inode *inode = dentry->d_inode;
6837 struct osd_fid_pack *rec;
6838 struct ldiskfs_dentry_param *ldp;
6839 int namelen = dentry->d_name.len;
6841 struct osd_thread_info *info = osd_oti_get(env);
6845 if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6848 /* There is enough space to hold the FID-in-dirent. */
6849 if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6850 rc = ldiskfs_journal_get_write_access(jh, bh);
6854 de->name[namelen] = 0;
6855 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6856 rec->fp_len = sizeof(struct lu_fid) + 1;
6857 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6858 de->file_type |= LDISKFS_DIRENT_LUFID;
6859 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6866 rc = ldiskfs_delete_entry(jh, dir, de, bh);
6870 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6871 osd_get_ldiskfs_dirent_param(ldp, fid);
6872 dentry->d_fsdata = (void *)ldp;
6873 dquot_initialize(dir);
6874 rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6876 * It is too bad, we cannot reinsert the name entry back.
6877 * That means we lose it!
6880 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6881 "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6882 osd_ino2name(inode),
6883 dir->i_ino, dir->i_generation, namelen,
6884 dentry->d_name.name, PFID(fid), rc);
6890 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6891 struct osd_it_ea *it, struct lu_fid *fid,
6892 struct osd_inode_id *id, __u32 *attr)
6894 struct osd_thread_info *info = osd_oti_get(env);
6895 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6896 struct osd_device *dev = osd_obj2dev(obj);
6897 struct super_block *sb = osd_sb(dev);
6898 const char *devname = osd_name(dev);
6899 struct osd_it_ea_dirent *ent = it->oie_dirent;
6900 struct inode *dir = obj->oo_inode;
6901 struct htree_lock *hlock = NULL;
6902 struct buffer_head *bh = NULL;
6903 handle_t *jh = NULL;
6904 struct ldiskfs_dir_entry_2 *de;
6905 struct dentry *dentry;
6906 struct inode *inode;
6907 const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6910 bool dotdot = false;
6915 if (ent->oied_name[0] == '.') {
6916 if (ent->oied_namelen == 1)
6919 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6923 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6924 inode = osd_iget(info, dev, id);
6925 if (IS_ERR(inode)) {
6926 rc = PTR_ERR(inode);
6927 if (rc == -ENOENT || rc == -ESTALE) {
6929 * Maybe dangling name entry, or
6930 * corrupted directory entry.
6932 *attr |= LUDA_UNKNOWN;
6935 CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6936 "check_repair, dir = %lu/%u, name = %.*s, "
6937 "ino = %llu, rc = %d\n",
6938 devname, dir->i_ino, dir->i_generation,
6939 ent->oied_namelen, ent->oied_name,
6946 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6948 rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6949 if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6955 * We need to ensure that the name entry is still valid.
6956 * Because it may be removed or renamed by other already.
6958 * The unlink or rename operation will start journal before PDO lock,
6959 * so to avoid deadlock, here we need to start journal handle before
6960 * related PDO lock also. But because we do not know whether there
6961 * will be something to be repaired before PDO lock, we just start
6962 * journal without conditions.
6964 * We may need to remove the name entry firstly, then insert back.
6965 * One credit is for user quota file update.
6966 * One credit is for group quota file update.
6967 * Two credits are for dirty inode.
6969 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6970 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6972 if (dev->od_dirent_journal != 0) {
6975 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6978 CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6979 "check_repair, dir = %lu/%u, credits = %d, "
6980 "name = %.*s, ino = %llu: rc = %d\n",
6981 devname, dir->i_ino, dir->i_generation, credits,
6982 ent->oied_namelen, ent->oied_name,
6985 GOTO(out_inode, rc);
6988 if (obj->oo_hl_head != NULL) {
6989 hlock = osd_oti_get(env)->oti_hlock;
6991 * "0" means exclusive lock for the whole directory.
6992 * We need to prevent others access such name entry
6993 * during the delete + insert. Neither HLOCK_ADD nor
6994 * HLOCK_DEL cannot guarantee the atomicity.
6996 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6998 down_write(&obj->oo_ext_idx_sem);
7001 if (obj->oo_hl_head != NULL) {
7002 hlock = osd_oti_get(env)->oti_hlock;
7003 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
7004 LDISKFS_HLOCK_LOOKUP);
7006 down_read(&obj->oo_ext_idx_sem);
7010 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
7011 if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
7012 *attr |= LUDA_IGNORE;
7018 * For dotdot entry, if there is not enough space to hold the
7019 * FID-in-dirent, just keep them there. It only happens when the
7020 * device upgraded from 1.8 or restored from MDT file-level backup.
7021 * For the whole directory, only dotdot entry have no FID-in-dirent
7022 * and needs to get FID from LMA when readdir, it will not affect the
7025 if (dotdot && !osd_dotdot_has_space(de)) {
7026 *attr |= LUDA_UNKNOWN;
7032 if (lu_fid_eq(fid, &lma->lma_self_fid))
7035 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
7036 struct lu_fid *tfid = &lma->lma_self_fid;
7038 if (likely(dotdot &&
7039 fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
7040 fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
7042 * It must be REMOTE_PARENT_DIR and as the
7043 * 'dotdot' entry of remote directory
7045 *attr |= LUDA_IGNORE;
7047 CDEBUG(D_LFSCK, "%s: expect remote agent "
7048 "parent directory, but got %.*s under "
7049 "dir = %lu/%u with the FID "DFID"\n",
7050 devname, ent->oied_namelen,
7051 ent->oied_name, dir->i_ino,
7052 dir->i_generation, PFID(tfid));
7054 *attr |= LUDA_UNKNOWN;
7061 if (!fid_is_zero(fid)) {
7062 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
7064 if (rc == -ENOENT ||
7066 !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
7068 * linkEA does not recognize the dirent entry,
7069 * it may because the dirent entry corruption
7070 * and points to other's inode.
7072 CDEBUG(D_LFSCK, "%s: the target inode does not "
7073 "recognize the dirent, dir = %lu/%u, "
7074 " name = %.*s, ino = %llu, "
7075 DFID": rc = %d\n", devname, dir->i_ino,
7076 dir->i_generation, ent->oied_namelen,
7077 ent->oied_name, ent->oied_ino, PFID(fid), rc);
7078 *attr |= LUDA_UNKNOWN;
7083 if (rc && rc != -ENODATA) {
7084 CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
7085 "dir = %lu/%u, name = %.*s, ino = %llu, "
7086 DFID": rc = %d\n", devname, dir->i_ino,
7087 dir->i_generation, ent->oied_namelen,
7088 ent->oied_name, ent->oied_ino, PFID(fid), rc);
7089 *attr |= LUDA_UNKNOWN;
7097 * linkEA recognizes the dirent entry, the FID-in-LMA is
7098 * valid, trusted, in spite of fid_is_sane(fid) or not.
7100 if (*attr & LUDA_VERIFY_DRYRUN) {
7101 *fid = lma->lma_self_fid;
7102 *attr |= LUDA_REPAIR;
7109 dev->od_dirent_journal = 1;
7110 if (hlock != NULL) {
7111 ldiskfs_htree_unlock(hlock);
7114 up_read(&obj->oo_ext_idx_sem);
7120 *fid = lma->lma_self_fid;
7122 /* Update or append the FID-in-dirent. */
7123 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7124 bh, de, hlock, dotdot);
7126 *attr |= LUDA_REPAIR;
7128 CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7129 "the dirent, dir = %lu/%u, name = %.*s, "
7130 "ino = %llu, "DFID": rc = %d\n",
7131 devname, dir->i_ino, dir->i_generation,
7132 ent->oied_namelen, ent->oied_name,
7133 ent->oied_ino, PFID(fid), rc);
7135 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7136 if (*attr & LUDA_VERIFY_DRYRUN) {
7137 if (fid_is_sane(fid)) {
7138 *attr |= LUDA_REPAIR;
7139 } else if (dev->od_index == 0) {
7140 lu_igif_build(fid, inode->i_ino,
7141 inode->i_generation);
7142 *attr |= LUDA_UPGRADE;
7150 dev->od_dirent_journal = 1;
7151 if (hlock != NULL) {
7152 ldiskfs_htree_unlock(hlock);
7155 up_read(&obj->oo_ext_idx_sem);
7162 if (unlikely(fid_is_sane(fid))) {
7164 * FID-in-dirent exists, but FID-in-LMA is lost.
7165 * Trust the FID-in-dirent, and add FID-in-LMA.
7167 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7169 *attr |= LUDA_REPAIR;
7171 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7172 "update dirent, dir = %lu/%u, "
7173 "name = %.*s, ino = %llu, "
7175 devname, dir->i_ino, dir->i_generation,
7176 ent->oied_namelen, ent->oied_name,
7177 ent->oied_ino, PFID(fid), rc);
7178 } else if (dev->od_index == 0) {
7179 lu_igif_build(fid, inode->i_ino, inode->i_generation);
7181 * It is probably IGIF object. Only aappend the
7182 * FID-in-dirent. OI scrub will process FID-in-LMA.
7184 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7185 bh, de, hlock, dotdot);
7187 *attr |= LUDA_UPGRADE;
7189 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7190 "after the dirent, dir = %lu/%u, "
7191 "name = %.*s, ino = %llu, "
7193 devname, dir->i_ino, dir->i_generation,
7194 ent->oied_namelen, ent->oied_name,
7195 ent->oied_ino, PFID(fid), rc);
7204 if (hlock != NULL) {
7205 ldiskfs_htree_unlock(hlock);
7207 if (dev->od_dirent_journal != 0)
7208 up_write(&obj->oo_ext_idx_sem);
7210 up_read(&obj->oo_ext_idx_sem);
7214 ldiskfs_journal_stop(jh);
7218 if (rc >= 0 && !dirty)
7219 dev->od_dirent_journal = 0;
7225 * Returns the value at current position from iterator's in memory structure.
7227 * \param di struct osd_it_ea, iterator's in memory structure
7228 * \param attr attr requested for dirent.
7229 * \param lde lustre dirent
7231 * \retval 0 no error and \param lde has correct lustre dirent.
7232 * \retval -ve on error
7234 static inline int osd_it_ea_rec(const struct lu_env *env,
7235 const struct dt_it *di,
7236 struct dt_rec *dtrec, __u32 attr)
7238 struct osd_it_ea *it = (struct osd_it_ea *)di;
7239 struct osd_object *obj = it->oie_obj;
7240 struct osd_device *dev = osd_obj2dev(obj);
7241 struct osd_thread_info *oti = osd_oti_get(env);
7242 struct osd_inode_id *id = &oti->oti_id;
7243 struct lu_fid *fid = &it->oie_dirent->oied_fid;
7244 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
7245 __u32 ino = it->oie_dirent->oied_ino;
7250 LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7252 if (attr & LUDA_VERIFY) {
7253 if (unlikely(is_remote_parent_ino(dev, ino))) {
7254 attr |= LUDA_IGNORE;
7256 * If the parent is on remote MDT, and there
7257 * is no FID-in-dirent, then we have to get
7258 * the parent FID from the linkEA.
7260 if (!fid_is_sane(fid) &&
7261 it->oie_dirent->oied_namelen == 2 &&
7262 it->oie_dirent->oied_name[0] == '.' &&
7263 it->oie_dirent->oied_name[1] == '.')
7264 osd_get_pfid_from_linkea(env, obj, fid);
7266 rc = osd_dirent_check_repair(env, obj, it, fid, id,
7270 if (!fid_is_sane(fid))
7271 attr |= LUDA_UNKNOWN;
7273 attr &= ~LU_DIRENT_ATTRS_MASK;
7274 if (!fid_is_sane(fid)) {
7275 bool is_dotdot = false;
7277 if (it->oie_dirent->oied_namelen == 2 &&
7278 it->oie_dirent->oied_name[0] == '.' &&
7279 it->oie_dirent->oied_name[1] == '.')
7282 * If the parent is on remote MDT, and there
7283 * is no FID-in-dirent, then we have to get
7284 * the parent FID from the linkEA.
7286 if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7287 rc = osd_get_pfid_from_linkea(env, obj, fid);
7289 if (is_dotdot == false &&
7290 OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7293 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7296 osd_id_gen(id, ino, OSD_OII_NOGEN);
7300 /* Pack the entry anyway, at least the offset is right. */
7301 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7302 it->oie_dirent->oied_name,
7303 it->oie_dirent->oied_namelen,
7304 it->oie_dirent->oied_type, attr);
7309 if (osd_remote_fid(env, dev, fid))
7312 if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7313 osd_add_oi_cache(oti, dev, id, fid);
7315 RETURN(rc > 0 ? 0 : rc);
7319 * Returns the record size size at current position.
7321 * This function will return record(lu_dirent) size in bytes.
7323 * \param[in] env execution environment
7324 * \param[in] di iterator's in memory structure
7325 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
7326 * calculate the lu_dirent size.
7328 * \retval record size(in bytes & in memory) of the current lu_dirent
7331 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7334 struct osd_it_ea *it = (struct osd_it_ea *)di;
7336 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7340 * Returns a cookie for current position of the iterator head, so that
7341 * user can use this cookie to load/start the iterator next time.
7343 * \param di iterator's in memory structure
7345 * \retval cookie for current position, on success
7347 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7349 struct osd_it_ea *it = (struct osd_it_ea *)di;
7351 return it->oie_dirent->oied_off;
7355 * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
7356 * to load a directory entry at a time and stored it i inn,
7357 * in iterator's in-memory data structure.
7359 * \param di struct osd_it_ea, iterator's in memory structure
7361 * \retval +ve on success
7362 * \retval -ve on error
7364 static int osd_it_ea_load(const struct lu_env *env,
7365 const struct dt_it *di, __u64 hash)
7367 struct osd_it_ea *it = (struct osd_it_ea *)di;
7371 it->oie_file.f_pos = hash;
7373 rc = osd_ldiskfs_it_fill(env, di);
7384 * Index lookup function for interoperability mode (b11826).
7386 * \param key, key i.e. file name to be searched
7388 * \retval +ve, on success
7389 * \retval -ve, on error
7391 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7392 struct dt_rec *rec, const struct dt_key *key)
7394 struct osd_object *obj = osd_dt_obj(dt);
7399 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7400 LINVRNT(osd_invariant(obj));
7402 rc = osd_ea_lookup_rec(env, obj, rec, key);
7409 * Index and Iterator operations for interoperability
7410 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7412 static const struct dt_index_operations osd_index_ea_ops = {
7413 .dio_lookup = osd_index_ea_lookup,
7414 .dio_declare_insert = osd_index_declare_ea_insert,
7415 .dio_insert = osd_index_ea_insert,
7416 .dio_declare_delete = osd_index_declare_ea_delete,
7417 .dio_delete = osd_index_ea_delete,
7419 .init = osd_it_ea_init,
7420 .fini = osd_it_ea_fini,
7421 .get = osd_it_ea_get,
7422 .put = osd_it_ea_put,
7423 .next = osd_it_ea_next,
7424 .key = osd_it_ea_key,
7425 .key_size = osd_it_ea_key_size,
7426 .rec = osd_it_ea_rec,
7427 .rec_size = osd_it_ea_rec_size,
7428 .store = osd_it_ea_store,
7429 .load = osd_it_ea_load
7433 static void *osd_key_init(const struct lu_context *ctx,
7434 struct lu_context_key *key)
7436 struct osd_thread_info *info;
7438 OBD_ALLOC_PTR(info);
7440 return ERR_PTR(-ENOMEM);
7442 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7443 if (info->oti_it_ea_buf == NULL)
7446 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7448 info->oti_hlock = ldiskfs_htree_lock_alloc();
7449 if (info->oti_hlock == NULL)
7455 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7458 return ERR_PTR(-ENOMEM);
7461 static void osd_key_fini(const struct lu_context *ctx,
7462 struct lu_context_key *key, void *data)
7464 struct osd_thread_info *info = data;
7465 struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7466 struct osd_idmap_cache *idc = info->oti_ins_cache;
7468 if (info->oti_dio_pages) {
7470 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7471 struct page *page = info->oti_dio_pages[i];
7473 LASSERT(PagePrivate2(page));
7474 LASSERT(PageLocked(page));
7475 ClearPagePrivate2(page);
7480 OBD_FREE_PTR_ARRAY(info->oti_dio_pages, PTLRPC_MAX_BRW_PAGES);
7483 if (info->oti_inode != NULL)
7485 if (info->oti_hlock != NULL)
7486 ldiskfs_htree_lock_free(info->oti_hlock);
7487 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7488 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7489 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7490 lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7491 lu_buf_free(&info->oti_big_buf);
7493 LASSERT(info->oti_ins_cache_size > 0);
7494 OBD_FREE_PTR_ARRAY(idc, info->oti_ins_cache_size);
7495 info->oti_ins_cache = NULL;
7496 info->oti_ins_cache_size = 0;
7501 static void osd_key_exit(const struct lu_context *ctx,
7502 struct lu_context_key *key, void *data)
7504 struct osd_thread_info *info = data;
7506 LASSERT(info->oti_r_locks == 0);
7507 LASSERT(info->oti_w_locks == 0);
7508 LASSERT(info->oti_txns == 0);
7511 /* type constructor/destructor: osd_type_init, osd_type_fini */
7512 LU_TYPE_INIT_FINI(osd, &osd_key);
7514 struct lu_context_key osd_key = {
7515 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7516 .lct_init = osd_key_init,
7517 .lct_fini = osd_key_fini,
7518 .lct_exit = osd_key_exit
7522 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7523 const char *name, struct lu_device *next)
7525 struct osd_device *osd = osd_dev(d);
7527 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7528 sizeof(osd->od_svname))
7530 return osd_procfs_init(osd, name);
7533 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7535 struct seq_server_site *ss = osd_seq_site(osd);
7540 if (osd->od_is_ost || osd->od_cl_seq != NULL)
7543 if (unlikely(ss == NULL))
7546 OBD_ALLOC_PTR(osd->od_cl_seq);
7547 if (osd->od_cl_seq == NULL)
7550 seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7551 osd->od_svname, ss->ss_server_seq);
7553 if (ss->ss_node_id == 0) {
7555 * If the OSD on the sequence controller(MDT0), then allocate
7556 * sequence here, otherwise allocate sequence after connected
7557 * to MDT0 (see mdt_register_lwp_callback()).
7559 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7560 &osd->od_cl_seq->lcs_space, env);
7566 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7568 if (osd->od_cl_seq == NULL)
7571 seq_client_fini(osd->od_cl_seq);
7572 OBD_FREE_PTR(osd->od_cl_seq);
7573 osd->od_cl_seq = NULL;
7576 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7580 /* shutdown quota slave instance associated with the device */
7581 if (o->od_quota_slave_md != NULL) {
7582 struct qsd_instance *qsd = o->od_quota_slave_md;
7584 o->od_quota_slave_md = NULL;
7588 if (o->od_quota_slave_dt != NULL) {
7589 struct qsd_instance *qsd = o->od_quota_slave_dt;
7591 o->od_quota_slave_dt = NULL;
7595 osd_fid_fini(env, o);
7596 osd_scrub_cleanup(env, o);
7601 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7605 if (o->od_mnt != NULL) {
7606 shrink_dcache_sb(osd_sb(o));
7607 osd_sync(env, &o->od_dt_dev);
7616 static int osd_mount(const struct lu_env *env,
7617 struct osd_device *o, struct lustre_cfg *cfg)
7619 const char *name = lustre_cfg_string(cfg, 0);
7620 const char *dev = lustre_cfg_string(cfg, 1);
7622 unsigned long page, s_flags, lmd_flags = 0;
7623 struct page *__page;
7624 struct file_system_type *type;
7625 char *options = NULL;
7627 struct osd_thread_info *info = osd_oti_get(env);
7628 struct lu_fid *fid = &info->oti_fid;
7629 struct inode *inode;
7630 int rc = 0, force_over_1024tb = 0;
7634 if (o->od_mnt != NULL)
7637 if (strlen(dev) >= sizeof(o->od_mntdev))
7639 strcpy(o->od_mntdev, dev);
7641 str = lustre_cfg_string(cfg, 2);
7642 s_flags = simple_strtoul(str, NULL, 0);
7643 str = strstr(str, ":");
7645 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7646 opts = lustre_cfg_string(cfg, 3);
7648 if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7649 CERROR("%s: device %s extents feature is not guaranteed to "
7650 "work on big-endian systems. Use \"bigendian_extents\" "
7651 "mount option to override.\n", name, dev);
7655 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7656 if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7657 CWARN("force_over_128tb option is deprecated. Filesystems smaller than 1024TB can be created without any force option. Use force_over_1024tb option for filesystems larger than 1024TB.\n");
7660 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7661 if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7662 CWARN("force_over_256tb option is deprecated. Filesystems smaller than 1024TB can be created without any force options. Use force_over_1024tb option for filesystems larger than 1024TB.\n");
7665 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
7666 if (opts != NULL && strstr(opts, "force_over_512tb") != NULL) {
7667 CWARN("force_over_512tb option is deprecated. Filesystems smaller than 1024TB can be created without any force options. Use force_over_1024tb option for filesystems larger than 1024TB.\n");
7671 if (opts != NULL && strstr(opts, "force_over_1024tb") != NULL)
7672 force_over_1024tb = 1;
7674 __page = alloc_page(GFP_KERNEL);
7676 GOTO(out, rc = -ENOMEM);
7677 page = (unsigned long)page_address(__page);
7678 options = (char *)page;
7681 /* strip out the options for back compatiblity */
7682 static const char * const sout[] = {
7689 /* strip out option we processed in osd */
7690 "bigendian_extents",
7694 "force_over_1024tb",
7697 strncat(options, opts, PAGE_SIZE);
7698 for (rc = 0, str = options; sout[rc]; ) {
7699 char *op = strstr(str, sout[rc]);
7706 if (op == options || *(op - 1) == ',') {
7707 str = op + strlen(sout[rc]);
7708 if (*str == ',' || *str == '\0') {
7709 *str == ',' ? str++ : str;
7710 memmove(op, str, strlen(str) + 1);
7713 for (str = op; *str != ',' && *str != '\0'; str++)
7717 strncat(options, "user_xattr,acl", PAGE_SIZE);
7720 /* Glom up mount options */
7721 if (*options != '\0')
7722 strncat(options, ",", PAGE_SIZE);
7723 strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7725 type = get_fs_type("ldiskfs");
7727 CERROR("%s: cannot find ldiskfs module\n", name);
7728 GOTO(out, rc = -ENODEV);
7731 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7732 module_put(type->owner);
7734 if (IS_ERR(o->od_mnt)) {
7735 rc = PTR_ERR(o->od_mnt);
7737 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7741 if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7742 osd_sb(o)->s_blocksize_bits > 1024ULL << 40 &&
7743 force_over_1024tb == 0) {
7744 CERROR("%s: device %s LDISKFS has not been tested on filesystems larger than 1024TB and may cause data corruption. Use 'force_over_1024tb' mount option to override.\n",
7746 GOTO(out_mnt, rc = -EINVAL);
7749 if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7750 LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7753 GOTO(out_mnt, rc = -EOPNOTSUPP);
7756 if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7757 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7758 GOTO(out_mnt, rc = -EINVAL);
7761 #ifdef LDISKFS_MOUNT_DIRDATA
7762 if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7763 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7764 else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7765 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7766 "enabling the dirdata feature. If you do not want to "
7767 "downgrade to Lustre-1.x again, you can enable it via "
7768 "'tune2fs -O dirdata device'\n", name, dev);
7770 inode = osd_sb(o)->s_root->d_inode;
7771 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7772 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7774 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7778 if (lmd_flags & LMD_FLG_NOSCRUB)
7779 o->od_auto_scrub_interval = AS_NEVER;
7781 if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7782 /* do not use pagecache with flash-backed storage */
7783 o->od_writethrough_cache = 0;
7784 o->od_read_cache = 0;
7795 __free_page(__page);
7800 static struct lu_device *osd_device_fini(const struct lu_env *env,
7801 struct lu_device *d)
7803 struct osd_device *o = osd_dev(d);
7807 osd_index_backup(env, o, false);
7808 osd_shutdown(env, o);
7810 if (o->od_oi_table != NULL)
7811 osd_oi_fini(osd_oti_get(env), o);
7812 osd_obj_map_fini(o);
7818 static int osd_device_init0(const struct lu_env *env,
7819 struct osd_device *o,
7820 struct lustre_cfg *cfg)
7822 struct lu_device *l = osd2lu_dev(o);
7823 struct osd_thread_info *info;
7827 /* if the module was re-loaded, env can loose its keys */
7828 rc = lu_env_refill((struct lu_env *)env);
7831 info = osd_oti_get(env);
7834 l->ld_ops = &osd_lu_ops;
7835 o->od_dt_dev.dd_ops = &osd_dt_ops;
7837 spin_lock_init(&o->od_osfs_lock);
7838 mutex_init(&o->od_otable_mutex);
7839 INIT_LIST_HEAD(&o->od_orphan_list);
7840 INIT_LIST_HEAD(&o->od_index_backup_list);
7841 INIT_LIST_HEAD(&o->od_index_restore_list);
7842 spin_lock_init(&o->od_lock);
7843 o->od_index_backup_policy = LIBP_NONE;
7846 o->od_read_cache = 1;
7847 o->od_writethrough_cache = 1;
7848 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7849 o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20;
7850 o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20;
7851 o->od_auto_scrub_interval = AS_DEFAULT;
7853 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7854 sizeof(o->od_svname));
7855 if (cplen >= sizeof(o->od_svname)) {
7860 o->od_index_backup_stop = 0;
7861 o->od_index = -1; /* -1 means index is invalid */
7862 rc = server_name2index(o->od_svname, &o->od_index, NULL);
7863 if (rc == LDD_F_SV_TYPE_OST)
7866 o->od_full_scrub_ratio = OFSR_DEFAULT;
7867 o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7868 rc = osd_mount(env, o, cfg);
7872 /* Can only check block device after mount */
7873 o->od_nonrotational =
7874 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
7876 rc = osd_obj_map_init(env, o);
7880 rc = lu_site_init(&o->od_site, l);
7882 GOTO(out_compat, rc);
7883 o->od_site.ls_bottom_dev = l;
7885 rc = lu_site_init_finish(&o->od_site);
7889 INIT_LIST_HEAD(&o->od_ios_list);
7890 /* setup scrub, including OI files initialization */
7892 rc = osd_scrub_setup(env, o);
7897 rc = osd_procfs_init(o, o->od_svname);
7899 CERROR("%s: can't initialize procfs: rc = %d\n",
7901 GOTO(out_scrub, rc);
7904 LASSERT(l->ld_site->ls_linkage.next != NULL);
7905 LASSERT(l->ld_site->ls_linkage.prev != NULL);
7907 /* initialize quota slave instance */
7908 /* currently it's no need to prepare qsd_instance_md for OST */
7909 if (!o->od_is_ost) {
7910 o->od_quota_slave_md = qsd_init(env, o->od_svname,
7912 o->od_proc_entry, true);
7913 if (IS_ERR(o->od_quota_slave_md)) {
7914 rc = PTR_ERR(o->od_quota_slave_md);
7915 o->od_quota_slave_md = NULL;
7916 GOTO(out_procfs, rc);
7920 o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
7921 o->od_proc_entry, false);
7923 if (IS_ERR(o->od_quota_slave_dt)) {
7924 if (o->od_quota_slave_md != NULL) {
7925 qsd_fini(env, o->od_quota_slave_md);
7926 o->od_quota_slave_md = NULL;
7929 rc = PTR_ERR(o->od_quota_slave_dt);
7930 o->od_quota_slave_dt = NULL;
7931 GOTO(out_procfs, rc);
7939 osd_scrub_cleanup(env, o);
7941 lu_site_fini(&o->od_site);
7943 osd_obj_map_fini(o);
7950 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7951 struct lu_device_type *t,
7952 struct lustre_cfg *cfg)
7954 struct osd_device *o;
7959 return ERR_PTR(-ENOMEM);
7961 rc = dt_device_init(&o->od_dt_dev, t);
7964 * Because the ctx might be revived in dt_device_init,
7965 * refill the env here
7967 lu_env_refill((struct lu_env *)env);
7968 rc = osd_device_init0(env, o, cfg);
7970 dt_device_fini(&o->od_dt_dev);
7973 if (unlikely(rc != 0))
7976 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7979 static struct lu_device *osd_device_free(const struct lu_env *env,
7980 struct lu_device *d)
7982 struct osd_device *o = osd_dev(d);
7986 /* XXX: make osd top device in order to release reference */
7987 d->ld_site->ls_top_dev = d;
7988 lu_site_purge(env, d->ld_site, -1);
7989 lu_site_print(env, d->ld_site, &d->ld_site->ls_obj_hash.nelems,
7990 D_ERROR, lu_cdebug_printer);
7991 lu_site_fini(&o->od_site);
7992 dt_device_fini(&o->od_dt_dev);
7997 static int osd_process_config(const struct lu_env *env,
7998 struct lu_device *d, struct lustre_cfg *cfg)
8000 struct osd_device *o = osd_dev(d);
8006 switch (cfg->lcfg_command) {
8008 rc = osd_mount(env, o, cfg);
8012 * For the case LCFG_PRE_CLEANUP is not called in advance,
8013 * that may happend if hit failure during mount process.
8015 osd_index_backup(env, o, false);
8016 lu_dev_del_linkage(d->ld_site, d);
8017 rc = osd_shutdown(env, o);
8020 LASSERT(&o->od_dt_dev);
8021 count = class_modify_config(cfg, PARAM_OSD,
8022 &o->od_dt_dev.dd_kobj);
8024 count = class_modify_config(cfg, PARAM_OST,
8025 &o->od_dt_dev.dd_kobj);
8026 rc = count > 0 ? 0 : count;
8028 case LCFG_PRE_CLEANUP:
8030 osd_index_backup(env, o,
8031 o->od_index_backup_policy != LIBP_NONE);
8041 static int osd_recovery_complete(const struct lu_env *env,
8042 struct lu_device *d)
8044 struct osd_device *osd = osd_dev(d);
8049 if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
8053 * start qsd instance on recovery completion, this notifies the quota
8054 * slave code that we are about to process new requests now
8056 rc = qsd_start(env, osd->od_quota_slave_dt);
8057 if (rc == 0 && osd->od_quota_slave_md != NULL)
8058 rc = qsd_start(env, osd->od_quota_slave_md);
8064 * we use exports to track all osd users
8066 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
8067 struct obd_device *obd, struct obd_uuid *cluuid,
8068 struct obd_connect_data *data, void *localdata)
8070 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8071 struct lustre_handle conn;
8076 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
8078 rc = class_connect(&conn, obd, cluuid);
8082 *exp = class_conn2export(&conn);
8084 spin_lock(&osd->od_osfs_lock);
8086 spin_unlock(&osd->od_osfs_lock);
8092 * once last export (we don't count self-export) disappeared
8093 * osd can be released
8095 static int osd_obd_disconnect(struct obd_export *exp)
8097 struct obd_device *obd = exp->exp_obd;
8098 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8099 int rc, release = 0;
8103 /* Only disconnect the underlying layers on the final disconnect. */
8104 spin_lock(&osd->od_osfs_lock);
8106 if (osd->od_connects == 0)
8108 spin_unlock(&osd->od_osfs_lock);
8110 rc = class_disconnect(exp); /* bz 9811 */
8112 if (rc == 0 && release)
8113 class_manual_cleanup(obd);
8117 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8118 struct lu_device *dev)
8120 struct osd_device *osd = osd_dev(dev);
8121 struct lr_server_data *lsd =
8122 &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8127 if (osd->od_quota_slave_md != NULL) {
8128 /* set up quota slave objects for inode */
8129 result = qsd_prepare(env, osd->od_quota_slave_md);
8134 if (osd->od_quota_slave_dt != NULL) {
8135 /* set up quota slave objects for block */
8136 result = qsd_prepare(env, osd->od_quota_slave_dt);
8142 if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8143 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8144 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8145 osd->od_index_in_idif = 1;
8147 osd->od_index_in_idif = 0;
8148 result = osd_register_proc_index_in_idif(osd);
8153 osd->od_index_in_idif = 1;
8157 result = osd_fid_init(env, osd);
8163 * Implementation of lu_device_operations::ldo_fid_alloc() for OSD
8167 * see include/lu_object.h for the details.
8169 static int osd_fid_alloc(const struct lu_env *env, struct lu_device *d,
8170 struct lu_fid *fid, struct lu_object *parent,
8171 const struct lu_name *name)
8173 struct osd_device *osd = osd_dev(d);
8175 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8178 static const struct lu_object_operations osd_lu_obj_ops = {
8179 .loo_object_init = osd_object_init,
8180 .loo_object_delete = osd_object_delete,
8181 .loo_object_release = osd_object_release,
8182 .loo_object_free = osd_object_free,
8183 .loo_object_print = osd_object_print,
8184 .loo_object_invariant = osd_object_invariant
8187 const struct lu_device_operations osd_lu_ops = {
8188 .ldo_object_alloc = osd_object_alloc,
8189 .ldo_process_config = osd_process_config,
8190 .ldo_recovery_complete = osd_recovery_complete,
8191 .ldo_prepare = osd_prepare,
8192 .ldo_fid_alloc = osd_fid_alloc,
8195 static const struct lu_device_type_operations osd_device_type_ops = {
8196 .ldto_init = osd_type_init,
8197 .ldto_fini = osd_type_fini,
8199 .ldto_start = osd_type_start,
8200 .ldto_stop = osd_type_stop,
8202 .ldto_device_alloc = osd_device_alloc,
8203 .ldto_device_free = osd_device_free,
8205 .ldto_device_init = osd_device_init,
8206 .ldto_device_fini = osd_device_fini
8209 static struct lu_device_type osd_device_type = {
8210 .ldt_tags = LU_DEVICE_DT,
8211 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
8212 .ldt_ops = &osd_device_type_ops,
8213 .ldt_ctx_tags = LCT_LOCAL,
8216 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8218 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8219 struct super_block *sb = osd_sb(osd);
8221 return (osd->od_mnt == NULL || sb->s_flags & SB_RDONLY);
8225 * lprocfs legacy support.
8227 static const struct obd_ops osd_obd_device_ops = {
8228 .o_owner = THIS_MODULE,
8229 .o_connect = osd_obd_connect,
8230 .o_disconnect = osd_obd_disconnect,
8231 .o_health_check = osd_health_check,
8234 static ssize_t track_declares_assert_show(struct kobject *kobj,
8235 struct attribute *attr,
8238 return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8241 static ssize_t track_declares_assert_store(struct kobject *kobj,
8242 struct attribute *attr,
8243 const char *buffer, size_t count)
8245 bool track_declares_assert;
8248 rc = kstrtobool(buffer, &track_declares_assert);
8252 ldiskfs_track_declares_assert = track_declares_assert;
8256 LUSTRE_RW_ATTR(track_declares_assert);
8258 static int __init osd_init(void)
8260 struct kobject *kobj;
8263 BUILD_BUG_ON(BH_DXLock >=
8264 sizeof(((struct buffer_head *)0)->b_state) * 8);
8265 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8266 /* please, try to keep osd_thread_info smaller than a page */
8267 BUILD_BUG_ON(sizeof(struct osd_thread_info) > PAGE_SIZE);
8272 rc = lu_kmem_init(ldiskfs_caches);
8276 #ifdef CONFIG_KALLSYMS
8277 priv_security_file_alloc =
8278 (void *)kallsyms_lookup_name("security_file_alloc");
8281 rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL,
8282 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8284 lu_kmem_fini(ldiskfs_caches);
8288 kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8290 rc = sysfs_create_file(kobj,
8291 &lustre_attr_track_declares_assert.attr);
8294 CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8301 static void __exit osd_exit(void)
8303 struct kobject *kobj;
8305 kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8307 sysfs_remove_file(kobj,
8308 &lustre_attr_track_declares_assert.attr);
8311 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8312 lu_kmem_fini(ldiskfs_caches);
8315 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8316 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8317 MODULE_VERSION(LUSTRE_VERSION_STRING);
8318 MODULE_LICENSE("GPL");
8320 module_init(osd_init);
8321 module_exit(osd_exit);