4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/osd/osd_handler.c
34 * Top-level entry points into osd module
36 * Author: Nikita Danilov <nikita@clusterfs.com>
37 * Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
40 #define DEBUG_SUBSYSTEM S_OSD
42 #include <linux/kallsyms.h>
43 #include <linux/module.h>
44 #include <linux/user_namespace.h>
45 #include <linux/uidgid.h>
47 /* prerequisite for linux/xattr.h */
48 #include <linux/types.h>
49 /* prerequisite for linux/xattr.h */
51 /* XATTR_{REPLACE,CREATE} */
52 #include <linux/xattr.h>
54 #include <ldiskfs/ldiskfs.h>
55 #include <ldiskfs/xattr.h>
56 #include <ldiskfs/ldiskfs_extents.h>
59 * struct OBD_{ALLOC,FREE}*()
62 #include <obd_support.h>
63 /* struct ptlrpc_thread */
64 #include <lustre_net.h>
65 #include <lustre_fid.h>
67 #include <uapi/linux/lustre/lustre_param.h>
69 #include "osd_internal.h"
70 #include "osd_dynlocks.h"
72 /* llo_* api support */
73 #include <md_object.h>
74 #include <lustre_quota.h>
76 #include <lustre_linkea.h>
78 /* Maximum EA size is limited by LNET_MTU for remote objects */
79 #define OSD_MAX_EA_SIZE 1048364
82 module_param(ldiskfs_pdo, int, 0644);
83 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
85 int ldiskfs_track_declares_assert;
86 module_param(ldiskfs_track_declares_assert, int, 0644);
87 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
89 /* Slab to allocate dynlocks */
90 struct kmem_cache *dynlock_cachep;
92 /* Slab to allocate osd_it_ea */
93 struct kmem_cache *osd_itea_cachep;
95 static struct lu_kmem_descr ldiskfs_caches[] = {
97 .ckd_cache = &dynlock_cachep,
98 .ckd_name = "dynlock_cache",
99 .ckd_size = sizeof(struct dynlock_handle)
102 .ckd_cache = &osd_itea_cachep,
103 .ckd_name = "osd_itea_cache",
104 .ckd_size = sizeof(struct osd_it_ea)
111 static const char dot[] = ".";
112 static const char dotdot[] = "..";
114 static const struct lu_object_operations osd_lu_obj_ops;
115 static const struct dt_object_operations osd_obj_ops;
116 static const struct dt_object_operations osd_obj_otable_it_ops;
117 static const struct dt_index_operations osd_index_iam_ops;
118 static const struct dt_index_operations osd_index_ea_ops;
120 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
121 const struct lu_fid *fid);
122 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
123 struct osd_device *osd);
125 int osd_trans_declare_op2rb[] = {
126 [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
127 [OSD_OT_PUNCH] = OSD_OT_MAX,
128 [OSD_OT_XATTR_SET] = OSD_OT_XATTR_SET,
129 [OSD_OT_CREATE] = OSD_OT_DESTROY,
130 [OSD_OT_DESTROY] = OSD_OT_CREATE,
131 [OSD_OT_REF_ADD] = OSD_OT_REF_DEL,
132 [OSD_OT_REF_DEL] = OSD_OT_REF_ADD,
133 [OSD_OT_WRITE] = OSD_OT_WRITE,
134 [OSD_OT_INSERT] = OSD_OT_DELETE,
135 [OSD_OT_DELETE] = OSD_OT_INSERT,
136 [OSD_OT_QUOTA] = OSD_OT_MAX,
139 static int osd_has_index(const struct osd_object *obj)
141 return obj->oo_dt.do_index_ops != NULL;
144 static int osd_object_invariant(const struct lu_object *l)
146 return osd_invariant(osd_obj(l));
150 * Concurrency: doesn't matter
152 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
154 struct osd_thread_info *oti = osd_oti_get(env);
156 return oti->oti_w_locks > 0 && o->oo_owner == env;
160 * Concurrency: doesn't access mutable data
162 static int osd_root_get(const struct lu_env *env,
163 struct dt_device *dev, struct lu_fid *f)
165 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
170 * the following set of functions are used to maintain per-thread
171 * cache of FID->ino mapping. this mechanism is needed to resolve
172 * FID to inode at dt_insert() which in turn stores ino in the
173 * directory entries to keep ldiskfs compatible with ext[34].
174 * due to locking-originated restrictions we can't lookup ino
175 * using LU cache (deadlock is possible). lookup using OI is quite
176 * expensive. so instead we maintain this cache and methods like
177 * dt_create() fill it. so in the majority of cases dt_insert() is
178 * able to find needed mapping in lockless manner.
180 static struct osd_idmap_cache *
181 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
182 const struct lu_fid *fid)
184 struct osd_thread_info *oti = osd_oti_get(env);
185 struct osd_idmap_cache *idc = oti->oti_ins_cache;
188 for (i = 0; i < oti->oti_ins_cache_used; i++) {
189 if (!lu_fid_eq(&idc[i].oic_fid, fid))
191 if (idc[i].oic_dev != osd)
200 static struct osd_idmap_cache *
201 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
202 const struct lu_fid *fid)
204 struct osd_thread_info *oti = osd_oti_get(env);
205 struct osd_idmap_cache *idc;
208 if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
209 i = oti->oti_ins_cache_size * 2;
211 i = OSD_INS_CACHE_SIZE;
212 OBD_ALLOC(idc, sizeof(*idc) * i);
214 return ERR_PTR(-ENOMEM);
215 if (oti->oti_ins_cache != NULL) {
216 memcpy(idc, oti->oti_ins_cache,
217 oti->oti_ins_cache_used * sizeof(*idc));
218 OBD_FREE(oti->oti_ins_cache,
219 oti->oti_ins_cache_used * sizeof(*idc));
221 oti->oti_ins_cache = idc;
222 oti->oti_ins_cache_size = i;
225 idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
228 idc->oic_lid.oii_ino = 0;
229 idc->oic_lid.oii_gen = 0;
236 * lookup mapping for the given fid in the cache, initialize a
237 * new one if not found. the initialization checks whether the
238 * object is local or remote. for local objects, OI is used to
239 * learn ino/generation. the function is used when the caller
240 * has no information about the object, e.g. at dt_insert().
242 static struct osd_idmap_cache *
243 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
244 const struct lu_fid *fid)
246 struct osd_idmap_cache *idc;
249 idc = osd_idc_find(env, osd, fid);
250 LASSERT(!IS_ERR(idc));
254 CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
255 osd->od_svname, PFID(fid));
257 /* new mapping is needed */
258 idc = osd_idc_add(env, osd, fid);
260 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
261 osd->od_svname, PFID(fid), PTR_ERR(idc));
266 rc = osd_remote_fid(env, osd, fid);
267 if (unlikely(rc < 0))
271 /* the object is local, lookup in OI */
272 /* XXX: probably cheaper to lookup in LU first? */
273 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
275 if (unlikely(rc < 0)) {
276 CERROR("can't lookup: rc = %d\n", rc);
280 /* the object is remote */
288 * lookup mapping for given FID and fill it from the given object.
289 * the object is lolcal by definition.
291 static int osd_idc_find_and_init(const struct lu_env *env,
292 struct osd_device *osd,
293 struct osd_object *obj)
295 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
296 struct osd_idmap_cache *idc;
298 idc = osd_idc_find(env, osd, fid);
299 LASSERT(!IS_ERR(idc));
301 if (obj->oo_inode == NULL)
303 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
304 LASSERT(idc->oic_lid.oii_ino == 0);
305 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
306 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
311 CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
312 osd->od_svname, PFID(fid));
314 /* new mapping is needed */
315 idc = osd_idc_add(env, osd, fid);
317 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
318 osd->od_svname, PFID(fid), PTR_ERR(idc));
322 if (obj->oo_inode != NULL) {
323 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
324 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
330 * OSD object methods.
334 * Concurrency: no concurrent access is possible that early in object
337 static struct lu_object *osd_object_alloc(const struct lu_env *env,
338 const struct lu_object_header *hdr,
341 struct osd_object *mo;
346 struct lu_object_header *h;
347 struct osd_device *o = osd_dev(d);
349 l = &mo->oo_dt.do_lu;
350 if (unlikely(o->od_in_init)) {
357 lu_object_header_init(h);
358 lu_object_init(l, h, d);
359 lu_object_add_top(h, l);
362 dt_object_init(&mo->oo_dt, NULL, d);
363 mo->oo_header = NULL;
366 mo->oo_dt.do_ops = &osd_obj_ops;
367 l->lo_ops = &osd_lu_obj_ops;
368 init_rwsem(&mo->oo_sem);
369 init_rwsem(&mo->oo_ext_idx_sem);
370 spin_lock_init(&mo->oo_guard);
371 INIT_LIST_HEAD(&mo->oo_xattr_list);
377 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
378 struct dentry *dentry, struct lustre_ost_attrs *loa)
382 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
383 (void *)loa, sizeof(*loa));
385 struct lustre_mdt_attrs *lma = &loa->loa_lma;
387 if (rc < sizeof(*lma))
391 lustre_loa_swab(loa, true);
392 /* Check LMA compatibility */
393 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
394 CWARN("%s: unsupported incompat LMA feature(s) %#x "
395 "for fid = "DFID", ino = %lu\n",
397 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
398 PFID(&lma->lma_self_fid), inode->i_ino);
401 } else if (rc == 0) {
409 * retrieve object from backend ext fs.
411 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
412 struct osd_inode_id *id)
415 struct inode *inode = NULL;
418 * if we look for an inode withing a running
419 * transaction, then we risk to deadlock
420 * osd_dirent_check_repair() breaks this
422 /* LASSERT(current->journal_info == NULL); */
424 inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
426 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
427 id->oii_ino, PTR_ERR(inode));
428 } else if (id->oii_gen != OSD_OII_NOGEN &&
429 inode->i_generation != id->oii_gen) {
430 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
431 "i_generation = %u\n",
432 id->oii_ino, id->oii_gen, inode->i_generation);
434 inode = ERR_PTR(-ESTALE);
435 } else if (inode->i_nlink == 0) {
437 * due to parallel readdir and unlink,
438 * we can have dead inode here.
440 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
442 inode = ERR_PTR(-ESTALE);
443 } else if (is_bad_inode(inode)) {
444 CWARN("%s: bad inode: ino = %u\n",
445 osd_dev2name(dev), id->oii_ino);
447 inode = ERR_PTR(-ENOENT);
448 } else if ((rc = osd_attach_jinode(inode))) {
452 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
453 if (id->oii_gen == OSD_OII_NOGEN)
454 osd_id_gen(id, inode->i_ino, inode->i_generation);
457 * Do not update file c/mtime in ldiskfs.
458 * NB: we don't have any lock to protect this because we don't
459 * have reference on osd_object now, but contention with
460 * another lookup + attr_set can't happen in the tiny window
461 * between if (...) and set S_NOCMTIME.
463 if (!(inode->i_flags & S_NOCMTIME))
464 inode->i_flags |= S_NOCMTIME;
469 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
470 handle_t *handle, struct dentry *child,
471 struct inode *inode, struct htree_lock *hlock)
475 rc = __ldiskfs_add_entry(handle, child, inode, hlock);
476 if (rc == -ENOBUFS || rc == -ENOSPC) {
477 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
478 struct inode *parent = child->d_parent->d_inode;
479 struct lu_fid *fid = NULL;
481 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
483 fid = &loa->loa_lma.lma_self_fid;
484 } else if (rc2 == -ENODATA) {
485 if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
486 fid = &info->oti_fid3;
487 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
488 } else if (!osd->od_is_ost && osd->od_index == 0) {
489 fid = &info->oti_fid3;
490 lu_igif_build(fid, parent->i_ino,
491 parent->i_generation);
496 /* below message is checked in sanity.sh test_129 */
497 CWARN("%s: directory (inode: %lu, FID: "DFID") %s maximum entry limit\n",
498 osd_name(osd), parent->i_ino, PFID(fid),
499 rc == -ENOSPC ? "has reached" : "is approaching");
501 /* below message is checked in sanity.sh test_129 */
502 CWARN("%s: directory (inode: %lu, FID: unknown) %s maximum entry limit\n",
503 osd_name(osd), parent->i_ino,
504 rc == -ENOSPC ? "has reached" : "is approaching");
506 /* ignore such error now */
516 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
517 struct osd_inode_id *id, struct lu_fid *fid)
519 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
523 inode = osd_iget(info, dev, id);
527 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
529 *fid = loa->loa_lma.lma_self_fid;
530 } else if (rc == -ENODATA) {
531 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
532 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
534 lu_igif_build(fid, inode->i_ino, inode->i_generation);
542 static struct inode *osd_iget_check(struct osd_thread_info *info,
543 struct osd_device *dev,
544 const struct lu_fid *fid,
545 struct osd_inode_id *id,
554 * The cached OI mapping is trustable. If we cannot locate the inode
555 * via the cached OI mapping, then return the failure to the caller
556 * directly without further OI checking.
560 inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
563 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
566 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
567 PFID(fid), id->oii_ino, rc);
571 if (is_bad_inode(inode)) {
576 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
577 PFID(fid), id->oii_ino);
581 if (id->oii_gen != OSD_OII_NOGEN &&
582 inode->i_generation != id->oii_gen) {
587 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
588 "oii_gen = %u, i_generation = %u\n", PFID(fid),
589 id->oii_ino, id->oii_gen, inode->i_generation);
593 if (inode->i_nlink == 0) {
598 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
599 PFID(fid), id->oii_ino);
603 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
607 __u32 saved_ino = id->oii_ino;
608 __u32 saved_gen = id->oii_gen;
611 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
613 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
615 * XXX: There are four possible cases:
617 * Backup/restore caused the OI invalid.
619 * Someone unlinked the object but NOT removed
620 * the OI mapping, such as mount target device
621 * as ldiskfs, and modify something directly.
623 * Someone just removed the object between the
624 * former oi_lookup and the iget. It is normal.
625 * 4. Other failure cases.
627 * Generally, when the device is mounted, it will
628 * auto check whether the system is restored from
629 * file-level backup or not. We trust such detect
630 * to distinguish the 1st case from the 2nd case:
631 * if the OI files are consistent but may contain
632 * stale OI mappings because of case 2, if iget()
633 * returns -ENOENT or -ESTALE, then it should be
638 * If the OI mapping was in OI file before the
639 * osd_iget_check(), but now, it is disappear,
640 * then it must be removed by race. That is a
646 * It is the OI scrub updated the OI mapping by race.
647 * The new OI mapping must be valid.
649 if (saved_ino != id->oii_ino ||
650 (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
659 if (dev->od_scrub.os_scrub.os_file.sf_flags &
662 * It still can be the case 2, but we cannot
663 * distinguish it from the case 1. So return
664 * -EREMCHG to block current operation until
665 * OI scrub rebuilt the OI mappings.
674 if (inode->i_generation == id->oii_gen)
679 if (id->oii_gen == OSD_OII_NOGEN)
680 osd_id_gen(id, inode->i_ino, inode->i_generation);
683 * Do not update file c/mtime in ldiskfs.
684 * NB: we don't have any lock to protect this because we don't
685 * have reference on osd_object now, but contention with
686 * another lookup + attr_set can't happen in the tiny window
687 * between if (...) and set S_NOCMTIME.
689 if (!(inode->i_flags & S_NOCMTIME))
690 inode->i_flags |= S_NOCMTIME;
707 * \retval +v: new filter_fid does not contain self-fid
708 * \retval 0: filter_fid_18_23, contains self-fid
709 * \retval -v: other failure cases
711 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
712 struct dentry *dentry, struct lu_fid *fid)
714 struct filter_fid *ff = &info->oti_ff;
715 struct ost_id *ostid = &info->oti_ostid;
718 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
719 if (rc == sizeof(struct filter_fid_18_23)) {
720 struct filter_fid_18_23 *ff_old = (void *)ff;
722 ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq));
723 rc = ostid_set_id(ostid, le64_to_cpu(ff_old->ff_objid));
725 * XXX: use 0 as the index for compatibility, the caller will
726 * handle index related issues when necessary.
729 ostid_to_fid(fid, ostid, 0);
730 } else if (rc >= (int)sizeof(struct filter_fid_24_29)) {
732 } else if (rc >= 0) {
739 static int osd_lma_self_repair(struct osd_thread_info *info,
740 struct osd_device *osd, struct inode *inode,
741 const struct lu_fid *fid, __u32 compat)
746 LASSERT(current->journal_info == NULL);
748 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
749 osd_dto_credits_noquota[DTO_XATTR_SET]);
752 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
757 rc = osd_ea_fid_set(info, inode, fid, compat, 0);
759 CWARN("%s: cannot self repair the LMA: rc = %d\n",
761 ldiskfs_journal_stop(jh);
765 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
767 struct osd_thread_info *info = osd_oti_get(env);
768 struct osd_device *osd = osd_obj2dev(obj);
769 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
770 struct lustre_mdt_attrs *lma = &loa->loa_lma;
771 struct inode *inode = obj->oo_inode;
772 struct dentry *dentry = &info->oti_obj_dentry;
773 struct lu_fid *fid = NULL;
774 const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
779 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
780 (void *)loa, sizeof(*loa));
781 if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
782 fid = &lma->lma_self_fid;
783 rc = osd_get_idif(info, inode, dentry, fid);
784 if (rc > 0 || (rc == -ENODATA && osd->od_index_in_idif)) {
786 * For the given OST-object, if it has neither LMA nor
787 * FID in XATTR_NAME_FID, then the given FID (which is
788 * contained in the @obj, from client RPC for locating
789 * the OST-object) is trusted. We use it to generate
792 osd_lma_self_repair(info, osd, inode, rfid,
803 lustre_lma_swab(lma);
804 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
805 (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
806 S_ISREG(inode->i_mode)))) {
807 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
808 "fid = "DFID", ino = %lu\n", osd_name(osd),
809 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
810 PFID(rfid), inode->i_ino);
813 fid = &lma->lma_self_fid;
814 if (lma->lma_compat & LMAC_STRIPE_INFO &&
816 obj->oo_pfid_in_lma = 1;
817 if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
819 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
823 if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
824 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
825 struct ost_id *oi = &info->oti_ostid;
826 struct lu_fid *fid1 = &info->oti_fid3;
827 __u32 idx = fid_idif_ost_idx(rfid);
830 * For old IDIF, the OST index is not part of the IDIF,
831 * Means that different OSTs may have the same IDIFs.
832 * Under such case, we need to make some compatible
833 * check to make sure to trigger OI scrub properly.
835 if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
836 /* Given @rfid is new, LMA is old. */
837 fid_to_ostid(fid, oi);
838 ostid_to_fid(fid1, oi, idx);
839 if (lu_fid_eq(fid1, rfid)) {
840 if (osd->od_index_in_idif)
841 osd_lma_self_repair(info, osd,
855 struct osd_check_lmv_buf {
856 /* please keep it as first member */
857 struct dir_context ctx;
858 struct osd_thread_info *oclb_info;
859 struct osd_device *oclb_dev;
860 struct osd_idmap_cache *oclb_oic;
866 * It is called internally by ->iterate*() to filter out the
867 * local slave object's FID of the striped directory.
869 * \retval 1 found the local slave's FID
870 * \retval 0 continue to check next item
871 * \retval -ve for failure
873 #ifdef HAVE_FILLDIR_USE_CTX
874 static int osd_stripe_dir_filldir(struct dir_context *buf,
876 static int osd_stripe_dir_filldir(void *buf,
878 const char *name, int namelen,
879 loff_t offset, __u64 ino, unsigned int d_type)
881 struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
882 struct osd_thread_info *oti = oclb->oclb_info;
883 struct lu_fid *fid = &oti->oti_fid3;
884 struct osd_inode_id *id = &oti->oti_id3;
885 struct osd_device *dev = oclb->oclb_dev;
886 struct osd_idmap_cache *oic = oclb->oclb_oic;
895 sscanf(name + 1, SFID, RFID(fid));
896 if (!fid_is_sane(fid))
899 if (osd_remote_fid(oti->oti_env, dev, fid))
902 osd_id_gen(id, ino, OSD_OII_NOGEN);
903 inode = osd_iget(oti, dev, id);
905 return PTR_ERR(inode);
908 osd_add_oi_cache(oti, dev, id, fid);
912 osd_oii_insert(dev, oic, true);
913 oclb->oclb_found = true;
919 * When lookup item under striped directory, we need to locate the master
920 * MDT-object of the striped directory firstly, then the client will send
921 * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
922 * and the item's name. If the system is restored from MDT file level backup,
923 * then before the OI scrub completely built the OI files, the OI mappings of
924 * the master MDT-object and slave MDT-object may be invalid. Usually, it is
925 * not a problem for the master MDT-object. Because when locate the master
926 * MDT-object, we will do name based lookup (for the striped directory itself)
927 * firstly, during such process we can setup the correct OI mapping for the
928 * master MDT-object. But it will be trouble for the slave MDT-object. Because
929 * the client will not trigger name based lookup on the MDT to locate the slave
930 * MDT-object before locating item under the striped directory, then when
931 * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
932 * is invalid and does not know what the right OI mapping is, then the MDT has
933 * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
934 * the OI file, related OI mapping is unknown yet, please try again later. And
935 * then client will re-try the RPC again and again until related OI mapping has
936 * been updated. That is quite inefficient.
938 * To resolve above trouble, we will handle it as the following two cases:
940 * 1) The slave MDT-object and the master MDT-object are on different MDTs.
941 * It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
942 * is linked under /REMOTE_PARENT_DIR with the name of its FID string.
943 * We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
944 * directly. Please check osd_fid_lookup().
946 * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
947 * Under such case, during lookup the master MDT-object, we will lookup the
948 * slave MDT-object via readdir against the master MDT-object, because the
949 * slave MDT-objects information are stored as sub-directories with the name
950 * "${FID}:${index}". Then when find the local slave MDT-object, its OI
951 * mapping will be recorded. Then subsequent osd_fid_lookup() will know
952 * the correct OI mapping for the slave MDT-object.
954 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
955 struct inode *inode, struct osd_idmap_cache *oic)
957 struct lu_buf *buf = &oti->oti_big_buf;
958 struct dentry *dentry = &oti->oti_obj_dentry;
959 struct file *filp = &oti->oti_file;
960 const struct file_operations *fops;
961 struct lmv_mds_md_v1 *lmv1;
962 struct osd_check_lmv_buf oclb = {
963 .ctx.actor = osd_stripe_dir_filldir,
974 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
977 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
979 lu_buf_realloc(buf, rc);
980 if (buf->lb_buf == NULL)
981 GOTO(out, rc = -ENOMEM);
987 if (unlikely(rc == 0 || rc == -ENODATA))
993 if (unlikely(buf->lb_buf == NULL)) {
994 lu_buf_realloc(buf, rc);
995 if (buf->lb_buf == NULL)
996 GOTO(out, rc = -ENOMEM);
1002 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1005 fops = inode->i_fop;
1006 dentry->d_inode = inode;
1007 dentry->d_sb = inode->i_sb;
1009 filp->f_path.dentry = dentry;
1010 filp->f_flags |= O_NOATIME;
1011 filp->f_mode = FMODE_64BITHASH | FMODE_NONOTIFY;
1012 filp->f_mapping = inode->i_mapping;
1014 filp->private_data = NULL;
1015 filp->f_cred = current_cred();
1016 set_file_inode(filp, inode);
1017 rc = osd_security_file_alloc(filp);
1022 oclb.oclb_items = 0;
1023 rc = iterate_dir(filp, &oclb.ctx);
1024 } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1025 filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1026 fops->release(inode, filp);
1030 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
1031 DFID": rc = %d\n", osd_ino2name(inode),
1032 inode->i_ino, inode->i_generation,
1033 PFID(&oic->oic_fid), rc);
1040 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1041 const struct lu_fid *fid,
1042 const struct lu_object_conf *conf)
1044 struct osd_thread_info *info;
1045 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1046 struct osd_device *dev;
1047 struct osd_idmap_cache *oic;
1048 struct osd_inode_id *id;
1049 struct inode *inode = NULL;
1050 struct lustre_scrub *scrub;
1051 struct scrub_file *sf;
1052 __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1057 bool remote = false;
1058 bool trusted = true;
1059 bool updated = false;
1060 bool checked = false;
1064 LINVRNT(osd_invariant(obj));
1065 LASSERT(obj->oo_inode == NULL);
1066 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
1068 dev = osd_dev(ldev);
1069 scrub = &dev->od_scrub.os_scrub;
1070 sf = &scrub->os_file;
1071 info = osd_oti_get(env);
1073 oic = &info->oti_cache;
1075 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1079 * For the object is created as locking anchor, or for the object to
1080 * be created on disk. No need to osd_oi_lookup() at here because FID
1081 * shouldn't never be re-used, if it's really a duplicate FID from
1082 * unexpected reason, we should be able to detect it later by calling
1083 * do_create->osd_oi_insert().
1085 if (conf && conf->loc_flags & LOC_F_NEW)
1086 GOTO(out, result = 0);
1088 /* Search order: 1. per-thread cache. */
1089 if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1095 if (!list_empty(&scrub->os_inconsistent_items)) {
1096 /* Search order: 2. OI scrub pending list. */
1097 result = osd_oii_lookup(dev, fid, id);
1103 * The OI mapping in the OI file can be updated by the OI scrub
1104 * when we locate the inode via FID. So it may be not trustable.
1108 /* Search order: 3. OI files. */
1109 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1110 if (result == -ENOENT) {
1111 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1112 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1113 !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1115 GOTO(out, result = 0);
1120 /* -ESTALE is returned if inode of OST object doesn't exist */
1121 if (result == -ESTALE &&
1122 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1123 GOTO(out, result = 0);
1130 obj->oo_inode = NULL;
1131 /* for later passes through checks, not true on first pass */
1132 if (!IS_ERR_OR_NULL(inode))
1135 inode = osd_iget_check(info, dev, fid, id, trusted);
1136 if (!IS_ERR(inode)) {
1137 obj->oo_inode = inode;
1145 result = PTR_ERR(inode);
1146 if (result == -ENOENT || result == -ESTALE)
1147 GOTO(out, result = 0);
1149 if (result != -EREMCHG)
1154 * We still have chance to get the valid inode: for the
1155 * object which is referenced by remote name entry, the
1156 * object on the local MDT will be linked under the dir
1157 * of "/REMOTE_PARENT_DIR" with its FID string as name.
1159 * We do not know whether the object for the given FID
1160 * is referenced by some remote name entry or not, and
1161 * especially for DNE II, a multiple-linked object may
1162 * have many name entries reside on many MDTs.
1164 * To simplify the operation, OSD will not distinguish
1165 * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1166 * only happened for the RPC from other MDT during the
1167 * OI scrub, or for the client side RPC with FID only,
1168 * such as FID to path, or from old connected client.
1171 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1175 flags |= SS_AUTO_PARTIAL;
1176 flags &= ~SS_AUTO_FULL;
1181 if (thread_is_running(&scrub->os_thread)) {
1182 if (scrub->os_partial_scan && !scrub->os_in_join)
1185 if (IS_ERR_OR_NULL(inode) || result)
1186 GOTO(out, result = -EINPROGRESS);
1189 LASSERT(obj->oo_inode == inode);
1191 osd_add_oi_cache(info, dev, id, fid);
1192 osd_oii_insert(dev, oic, true);
1196 if (dev->od_auto_scrub_interval == AS_NEVER) {
1198 GOTO(out, result = -EREMCHG);
1201 LASSERT(obj->oo_inode == inode);
1203 osd_add_oi_cache(info, dev, id, fid);
1208 rc1 = osd_scrub_start(env, dev, flags);
1209 LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1210 "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1211 if (rc1 && rc1 != -EALREADY)
1212 GOTO(out, result = -EREMCHG);
1214 if (IS_ERR_OR_NULL(inode) || result)
1215 GOTO(out, result = -EINPROGRESS);
1218 LASSERT(obj->oo_inode == inode);
1220 osd_add_oi_cache(info, dev, id, fid);
1221 osd_oii_insert(dev, oic, true);
1226 if (unlikely(obj->oo_header))
1229 result = osd_check_lma(env, obj);
1233 LASSERTF(id->oii_ino == inode->i_ino &&
1234 id->oii_gen == inode->i_generation,
1235 "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1236 PFID(fid), id->oii_ino, id->oii_gen,
1237 inode->i_ino, inode->i_generation);
1239 saved_ino = inode->i_ino;
1240 saved_gen = inode->i_generation;
1242 if (unlikely(result == -ENODATA)) {
1244 * If the OI scrub updated the OI mapping by race, it
1245 * must be valid. Trust the inode that has no LMA EA.
1250 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1253 * The OI mapping is still there, the inode is still
1254 * valid. It is just becaues the inode has no LMA EA.
1256 if (saved_ino == id->oii_ino &&
1257 saved_gen == id->oii_gen)
1261 * It is the OI scrub updated the OI mapping by race.
1262 * The new OI mapping must be valid.
1270 * "result == -ENOENT" means that the OI mappinghas been
1271 * removed by race, so the inode belongs to other object.
1273 * Others error can be returned directly.
1275 if (result == -ENOENT) {
1278 obj->oo_inode = NULL;
1283 if (result != -EREMCHG)
1289 * if two OST objects map to the same inode, and inode mode is
1290 * (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666), which means it's
1291 * reserved by precreate, and not written yet, in this case, don't
1292 * set inode for the object whose FID mismatch, so that it can create
1293 * inode and not block precreate.
1295 if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) &&
1296 inode->i_mode == (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666)) {
1297 obj->oo_inode = NULL;
1298 GOTO(out, result = 0);
1301 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1303 * "result == -ENOENT" means the cached OI mapping has been removed
1304 * from the OI file by race, above inode belongs to other object.
1306 if (result == -ENOENT) {
1309 obj->oo_inode = NULL;
1310 GOTO(out, result = 0);
1316 if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1322 * It is the OI scrub updated the OI mapping by race.
1323 * The new OI mapping must be valid.
1331 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1332 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1334 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1336 if (lma->lma_compat & LMAC_STRIPE_INFO &&
1338 obj->oo_pfid_in_lma = 1;
1339 if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1341 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1342 } else if (result != -ENODATA) {
1347 obj->oo_compat_dot_created = 1;
1348 obj->oo_compat_dotdot_created = 1;
1350 if (S_ISDIR(inode->i_mode) &&
1351 (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1352 osd_check_lmv(info, dev, inode, oic);
1354 result = osd_attach_jinode(inode);
1359 GOTO(out, result = 0);
1361 LASSERT(!obj->oo_hl_head);
1362 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1364 GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1367 if (result || !obj->oo_inode) {
1368 if (!IS_ERR_OR_NULL(inode))
1371 obj->oo_inode = NULL;
1373 fid_zero(&oic->oic_fid);
1376 LINVRNT(osd_invariant(obj));
1381 * Concurrency: shouldn't matter.
1383 static void osd_object_init0(struct osd_object *obj)
1385 LASSERT(obj->oo_inode != NULL);
1386 obj->oo_dt.do_body_ops = &osd_body_ops;
1387 obj->oo_dt.do_lu.lo_header->loh_attr |=
1388 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1392 * Concurrency: no concurrent access is possible that early in object
1395 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1396 const struct lu_object_conf *conf)
1398 struct osd_object *obj = osd_obj(l);
1401 LINVRNT(osd_invariant(obj));
1403 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1404 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1405 l->lo_header->loh_attr |= LOHA_EXISTS;
1409 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1410 obj->oo_dt.do_body_ops = &osd_body_ops_new;
1411 if (result == 0 && obj->oo_inode != NULL) {
1412 struct osd_thread_info *oti = osd_oti_get(env);
1413 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1415 osd_object_init0(obj);
1416 if (unlikely(obj->oo_header))
1419 result = osd_get_lma(oti, obj->oo_inode,
1420 &oti->oti_obj_dentry, loa);
1423 * Convert LMAI flags to lustre LMA flags
1424 * and cache it to oo_lma_flags
1427 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1428 } else if (result == -ENODATA) {
1433 LINVRNT(osd_invariant(obj));
1438 * The first part of oxe_buf is xattr name, and is '\0' terminated.
1439 * The left part is for value, binary mode.
1441 struct osd_xattr_entry {
1442 struct list_head oxe_list;
1446 struct rcu_head oxe_rcu;
1450 static int osd_oxc_get(struct osd_object *obj, const char *name,
1453 struct osd_xattr_entry *tmp;
1454 struct osd_xattr_entry *oxe = NULL;
1455 size_t namelen = strlen(name);
1461 list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1462 if (namelen == tmp->oxe_namelen &&
1463 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1470 GOTO(out, rc = -ENOENT);
1472 if (!oxe->oxe_exist)
1473 GOTO(out, rc = -ENODATA);
1476 rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1479 if (buf->lb_buf == NULL)
1482 if (buf->lb_len < rc)
1483 GOTO(out, rc = -ERANGE);
1485 memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1493 static void osd_oxc_free(struct rcu_head *head)
1495 struct osd_xattr_entry *oxe;
1497 oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1498 OBD_FREE(oxe, oxe->oxe_len);
1501 static void osd_oxc_add(struct osd_object *obj, const char *name,
1502 const char *buf, int buflen)
1504 struct osd_xattr_entry *oxe;
1505 struct osd_xattr_entry *old = NULL;
1506 struct osd_xattr_entry *tmp;
1507 size_t namelen = strlen(name);
1508 size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1510 OBD_ALLOC(oxe, len);
1514 INIT_LIST_HEAD(&oxe->oxe_list);
1516 oxe->oxe_namelen = namelen;
1517 memcpy(oxe->oxe_buf, name, namelen);
1519 LASSERT(buf != NULL);
1520 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1521 oxe->oxe_exist = true;
1523 oxe->oxe_exist = false;
1526 /* this should be rarely called, just remove old and add new */
1527 spin_lock(&obj->oo_guard);
1528 list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1529 if (namelen == tmp->oxe_namelen &&
1530 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1536 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1537 call_rcu(&old->oxe_rcu, osd_oxc_free);
1539 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1541 spin_unlock(&obj->oo_guard);
1544 static void osd_oxc_del(struct osd_object *obj, const char *name)
1546 struct osd_xattr_entry *oxe;
1547 size_t namelen = strlen(name);
1549 spin_lock(&obj->oo_guard);
1550 list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1551 if (namelen == oxe->oxe_namelen &&
1552 strncmp(name, oxe->oxe_buf, namelen) == 0) {
1553 list_del_rcu(&oxe->oxe_list);
1554 call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1558 spin_unlock(&obj->oo_guard);
1561 static void osd_oxc_fini(struct osd_object *obj)
1563 struct osd_xattr_entry *oxe, *next;
1565 list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1566 list_del(&oxe->oxe_list);
1567 OBD_FREE(oxe, oxe->oxe_len);
1572 * Concurrency: no concurrent access is possible that late in object
1575 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1577 struct osd_object *obj = osd_obj(l);
1578 struct lu_object_header *h = obj->oo_header;
1580 LINVRNT(osd_invariant(obj));
1583 dt_object_fini(&obj->oo_dt);
1584 if (obj->oo_hl_head != NULL)
1585 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1588 lu_object_header_fini(h);
1594 * Concurrency: no concurrent access is possible that late in object
1597 static void osd_index_fini(struct osd_object *o)
1599 struct iam_container *bag;
1601 if (o->oo_dir != NULL) {
1602 bag = &o->oo_dir->od_container;
1603 if (o->oo_inode != NULL) {
1604 if (bag->ic_object == o->oo_inode)
1605 iam_container_fini(bag);
1607 OBD_FREE_PTR(o->oo_dir);
1613 * Concurrency: no concurrent access is possible that late in object
1614 * life-cycle (for all existing callers, that is. New callers have to provide
1615 * their own locking.)
1617 static int osd_inode_unlinked(const struct inode *inode)
1619 return inode->i_nlink == 0;
1623 OSD_TXN_OI_DELETE_CREDITS = 20,
1624 OSD_TXN_INODE_DELETE_CREDITS = 20
1631 #if OSD_THANDLE_STATS
1633 * Set time when the handle is allocated
1635 static void osd_th_alloced(struct osd_thandle *oth)
1637 oth->oth_alloced = ktime_get();
1641 * Set time when the handle started
1643 static void osd_th_started(struct osd_thandle *oth)
1645 oth->oth_started = ktime_get();
1649 * Check whether the we deal with this handle for too long.
1651 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1652 ktime_t alloced, ktime_t started,
1655 ktime_t now = ktime_get();
1657 LASSERT(dev != NULL);
1659 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1660 ktime_us_delta(started, alloced));
1661 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1662 ktime_us_delta(closed, started));
1663 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1664 ktime_us_delta(now, closed));
1666 if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1667 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1668 oth, now, alloced, started, closed);
1669 libcfs_debug_dumpstack(NULL);
1673 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
1675 ktime_t __closed = ktime_get(); \
1676 ktime_t __alloced = oth->oth_alloced; \
1677 ktime_t __started = oth->oth_started; \
1680 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
1683 #else /* OSD_THANDLE_STATS */
1685 #define osd_th_alloced(h) do {} while(0)
1686 #define osd_th_started(h) do {} while(0)
1687 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
1689 #endif /* OSD_THANDLE_STATS */
1692 * Concurrency: doesn't access mutable data.
1694 static int osd_param_is_not_sane(const struct osd_device *dev,
1695 const struct thandle *th)
1697 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1699 return oh->ot_credits > osd_transaction_size(dev);
1703 * Concurrency: shouldn't matter.
1705 static void osd_trans_commit_cb(struct super_block *sb,
1706 struct ldiskfs_journal_cb_entry *jcb, int error)
1708 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1709 struct thandle *th = &oh->ot_super;
1710 struct lu_device *lud = &th->th_dev->dd_lu_dev;
1711 struct dt_txn_commit_cb *dcb, *tmp;
1713 LASSERT(oh->ot_handle == NULL);
1716 CERROR("transaction @0x%p commit error: %d\n", th, error);
1718 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DELAY_TRANS, 40);
1719 /* call per-transaction callbacks if any */
1720 list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1722 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1723 "commit callback entry: magic=%x name='%s'\n",
1724 dcb->dcb_magic, dcb->dcb_name);
1725 list_del_init(&dcb->dcb_linkage);
1726 dcb->dcb_func(NULL, th, dcb, error);
1729 lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1736 #ifndef HAVE_SB_START_WRITE
1737 # define sb_start_write(sb) do {} while (0)
1738 # define sb_end_write(sb) do {} while (0)
1741 static struct thandle *osd_trans_create(const struct lu_env *env,
1742 struct dt_device *d)
1744 struct osd_thread_info *oti = osd_oti_get(env);
1745 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1746 struct osd_thandle *oh;
1752 CERROR("%s: someone try to start transaction under "
1753 "readonly mode, should be disabled.\n",
1754 osd_name(osd_dt_dev(d)));
1756 RETURN(ERR_PTR(-EROFS));
1759 /* on pending IO in this thread should left from prev. request */
1760 LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1762 sb_start_write(osd_sb(osd_dt_dev(d)));
1764 OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1766 sb_end_write(osd_sb(osd_dt_dev(d)));
1767 RETURN(ERR_PTR(-ENOMEM));
1770 oh->ot_quota_trans = &oti->oti_quota_trans;
1771 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1776 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1777 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1778 INIT_LIST_HEAD(&oh->ot_trunc_locks);
1781 memset(oti->oti_declare_ops, 0,
1782 sizeof(oti->oti_declare_ops));
1783 memset(oti->oti_declare_ops_cred, 0,
1784 sizeof(oti->oti_declare_ops_cred));
1785 memset(oti->oti_declare_ops_used, 0,
1786 sizeof(oti->oti_declare_ops_used));
1788 oti->oti_ins_cache_depth++;
1793 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1795 struct osd_thread_info *oti = osd_oti_get(env);
1796 struct osd_thandle *oh;
1798 oh = container_of0(th, struct osd_thandle, ot_super);
1799 LASSERT(oh != NULL);
1801 CWARN(" create: %u/%u/%u, destroy: %u/%u/%u\n",
1802 oti->oti_declare_ops[OSD_OT_CREATE],
1803 oti->oti_declare_ops_cred[OSD_OT_CREATE],
1804 oti->oti_declare_ops_used[OSD_OT_CREATE],
1805 oti->oti_declare_ops[OSD_OT_DESTROY],
1806 oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1807 oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1808 CWARN(" attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1809 oti->oti_declare_ops[OSD_OT_ATTR_SET],
1810 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1811 oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1812 oti->oti_declare_ops[OSD_OT_XATTR_SET],
1813 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1814 oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1815 CWARN(" write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1816 oti->oti_declare_ops[OSD_OT_WRITE],
1817 oti->oti_declare_ops_cred[OSD_OT_WRITE],
1818 oti->oti_declare_ops_used[OSD_OT_WRITE],
1819 oti->oti_declare_ops[OSD_OT_PUNCH],
1820 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1821 oti->oti_declare_ops_used[OSD_OT_PUNCH],
1822 oti->oti_declare_ops[OSD_OT_QUOTA],
1823 oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1824 oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1825 CWARN(" insert: %u/%u/%u, delete: %u/%u/%u\n",
1826 oti->oti_declare_ops[OSD_OT_INSERT],
1827 oti->oti_declare_ops_cred[OSD_OT_INSERT],
1828 oti->oti_declare_ops_used[OSD_OT_INSERT],
1829 oti->oti_declare_ops[OSD_OT_DELETE],
1830 oti->oti_declare_ops_cred[OSD_OT_DELETE],
1831 oti->oti_declare_ops_used[OSD_OT_DELETE]);
1832 CWARN(" ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1833 oti->oti_declare_ops[OSD_OT_REF_ADD],
1834 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1835 oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1836 oti->oti_declare_ops[OSD_OT_REF_DEL],
1837 oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1838 oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1842 * Concurrency: shouldn't matter.
1844 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1847 struct osd_thread_info *oti = osd_oti_get(env);
1848 struct osd_device *dev = osd_dt_dev(d);
1850 struct osd_thandle *oh;
1855 LASSERT(current->journal_info == NULL);
1857 oh = container_of0(th, struct osd_thandle, ot_super);
1858 LASSERT(oh != NULL);
1859 LASSERT(oh->ot_handle == NULL);
1861 rc = dt_txn_hook_start(env, d, th);
1865 if (unlikely(osd_param_is_not_sane(dev, th))) {
1866 static unsigned long last_printed;
1867 static int last_credits;
1870 * don't make noise on a tiny testing systems
1871 * actual credits misuse will be caught anyway
1873 if (last_credits != oh->ot_credits &&
1874 time_after(jiffies, last_printed +
1875 cfs_time_seconds(60)) &&
1876 osd_transaction_size(dev) > 512) {
1877 CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1878 oh->ot_credits, osd_transaction_size(dev));
1879 osd_trans_dump_creds(env, th);
1880 libcfs_debug_dumpstack(NULL);
1881 last_credits = oh->ot_credits;
1882 last_printed = jiffies;
1885 * XXX Limit the credits to 'max_transaction_buffers', and
1886 * let the underlying filesystem to catch the error if
1887 * we really need so many credits.
1889 * This should be removed when we can calculate the
1890 * credits precisely.
1892 oh->ot_credits = osd_transaction_size(dev);
1893 } else if (ldiskfs_track_declares_assert != 0) {
1895 * reserve few credits to prevent an assertion in JBD
1896 * our debugging mechanism will be able to detected
1897 * overuse. this can help to debug single-update
1900 oh->ot_credits += 10;
1901 if (unlikely(osd_param_is_not_sane(dev, th)))
1902 oh->ot_credits = osd_transaction_size(dev);
1905 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1906 GOTO(out, rc = -EIO);
1909 * XXX temporary stuff. Some abstraction layer should
1912 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1916 LASSERT(oti->oti_txns == 0);
1918 lu_device_get(&d->dd_lu_dev);
1919 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1930 static int osd_seq_exists(const struct lu_env *env,
1931 struct osd_device *osd, u64 seq)
1933 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1934 struct seq_server_site *ss = osd_seq_site(osd);
1939 LASSERT(ss != NULL);
1940 LASSERT(ss->ss_server_fld != NULL);
1942 rc = osd_fld_lookup(env, osd, seq, range);
1945 CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1946 osd_name(osd), seq, rc);
1950 RETURN(ss->ss_node_id == range->lsr_index);
1953 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1955 struct dt_txn_commit_cb *dcb;
1956 struct dt_txn_commit_cb *tmp;
1958 /* call per-transaction stop callbacks if any */
1959 list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1961 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1962 "commit callback entry: magic=%x name='%s'\n",
1963 dcb->dcb_magic, dcb->dcb_name);
1964 list_del_init(&dcb->dcb_linkage);
1965 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1970 * Concurrency: shouldn't matter.
1972 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1975 struct osd_thread_info *oti = osd_oti_get(env);
1976 struct osd_thandle *oh;
1977 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1978 struct osd_device *osd = osd_dt_dev(th->th_dev);
1979 struct qsd_instance *qsd = osd_def_qsd(osd);
1980 struct lquota_trans *qtrans;
1981 LIST_HEAD(truncates);
1982 int rc = 0, remove_agents = 0;
1986 oh = container_of0(th, struct osd_thandle, ot_super);
1988 remove_agents = oh->ot_remove_agents;
1990 qtrans = oh->ot_quota_trans;
1991 oh->ot_quota_trans = NULL;
1993 /* move locks to local list, stop tx, execute truncates */
1994 list_splice(&oh->ot_trunc_locks, &truncates);
1996 if (oh->ot_handle != NULL) {
1999 handle_t *hdl = oh->ot_handle;
2002 * add commit callback
2003 * notice we don't do this in osd_trans_start()
2004 * as underlying transaction can change during truncate
2006 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
2009 LASSERT(oti->oti_txns == 1);
2012 rc = dt_txn_hook_stop(env, th);
2014 CERROR("%s: failed in transaction hook: rc = %d\n",
2017 osd_trans_stop_cb(oh, rc);
2018 /* hook functions might modify th_sync */
2019 hdl->h_sync = th->th_sync;
2021 oh->ot_handle = NULL;
2022 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2024 CERROR("%s: failed to stop transaction: rc = %d\n",
2025 osd_name(osd), rc2);
2029 osd_process_truncates(&truncates);
2031 osd_trans_stop_cb(oh, th->th_result);
2035 osd_trunc_unlock_all(&truncates);
2037 /* inform the quota slave device that the transaction is stopping */
2038 qsd_op_end(env, qsd, qtrans);
2041 * as we want IO to journal and data IO be concurrent, we don't block
2042 * awaiting data IO completion in osd_do_bio(), instead we wait here
2043 * once transaction is submitted to the journal. all reqular requests
2044 * don't do direct IO (except read/write), thus this wait_event becomes
2047 * IMPORTANT: we have to wait till any IO submited by the thread is
2048 * completed otherwise iobuf may be corrupted by different request
2050 wait_event(iobuf->dr_wait,
2051 atomic_read(&iobuf->dr_numreqs) == 0);
2052 osd_fini_iobuf(osd, iobuf);
2054 rc = iobuf->dr_error;
2056 if (unlikely(remove_agents != 0))
2057 osd_process_scheduled_agent_removals(env, osd);
2059 oti->oti_ins_cache_depth--;
2060 /* reset OI cache for safety */
2061 if (oti->oti_ins_cache_depth == 0)
2062 oti->oti_ins_cache_used = 0;
2064 sb_end_write(osd_sb(osd));
2069 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2071 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
2074 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2075 LASSERT(&dcb->dcb_func != NULL);
2076 if (dcb->dcb_flags & DCB_TRANS_STOP)
2077 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2079 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2085 * Called just before object is freed. Releases all resources except for
2086 * object itself (that is released by osd_object_free()).
2088 * Concurrency: no concurrent access is possible that late in object
2091 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2093 struct osd_object *obj = osd_obj(l);
2094 struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2095 struct inode *inode = obj->oo_inode;
2100 LINVRNT(osd_invariant(obj));
2103 * If object is unlinked remove fid->ino mapping from object index.
2106 osd_index_fini(obj);
2111 uid = i_uid_read(inode);
2112 gid = i_gid_read(inode);
2113 projid = i_projid_read(inode);
2115 obj->oo_inode = NULL;
2118 /* do not rebalance quota if the caller needs to release memory
2119 * otherwise qsd_refresh_usage() may went into a new ldiskfs
2120 * transaction and risk to deadlock - LU-12178 */
2121 if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2124 if (!obj->oo_header && qsd) {
2125 struct osd_thread_info *info = osd_oti_get(env);
2126 struct lquota_id_info *qi = &info->oti_qi;
2128 /* Release granted quota to master if necessary */
2129 qi->lqi_id.qid_uid = uid;
2130 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2132 qi->lqi_id.qid_uid = gid;
2133 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2135 qi->lqi_id.qid_uid = projid;
2136 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2141 * Concurrency: ->loo_object_release() is called under site spin-lock.
2143 static void osd_object_release(const struct lu_env *env,
2144 struct lu_object *l)
2146 struct osd_object *o = osd_obj(l);
2149 * nobody should be releasing a non-destroyed object with nlink=0
2150 * the API allows this, but ldiskfs doesn't like and then report
2151 * this inode as deleted
2153 LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2154 o->oo_inode->i_nlink == 0));
2158 * Concurrency: shouldn't matter.
2160 static int osd_object_print(const struct lu_env *env, void *cookie,
2161 lu_printer_t p, const struct lu_object *l)
2163 struct osd_object *o = osd_obj(l);
2164 struct iam_descr *d;
2166 if (o->oo_dir != NULL)
2167 d = o->oo_dir->od_container.ic_descr;
2170 return (*p)(env, cookie,
2171 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2173 o->oo_inode ? o->oo_inode->i_ino : 0UL,
2174 o->oo_inode ? o->oo_inode->i_generation : 0,
2175 d ? d->id_ops->id_name : "plain");
2179 * Concurrency: shouldn't matter.
2181 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2182 struct obd_statfs *sfs, struct obd_statfs_info *info)
2184 struct osd_device *osd = osd_dt_dev(d);
2185 struct super_block *sb = osd_sb(osd);
2186 struct kstatfs *ksfs;
2190 if (unlikely(osd->od_mnt == NULL))
2191 return -EINPROGRESS;
2193 /* osd_lproc.c call this without env, allocate ksfs for that case */
2194 if (unlikely(env == NULL)) {
2195 OBD_ALLOC_PTR(ksfs);
2199 ksfs = &osd_oti_get(env)->oti_ksfs;
2202 result = sb->s_op->statfs(sb->s_root, ksfs);
2206 statfs_pack(sfs, ksfs);
2207 if (unlikely(sb->s_flags & SB_RDONLY))
2208 sfs->os_state |= OS_STATE_READONLY;
2210 sfs->os_state |= osd->od_nonrotational ? OS_STATE_NONROT : 0;
2212 if (ldiskfs_has_feature_extents(sb))
2213 sfs->os_maxbytes = sb->s_maxbytes;
2215 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2218 * Reserve some space so to avoid fragmenting the filesystem too much.
2219 * Fragmentation not only impacts performance, but can also increase
2220 * metadata overhead significantly, causing grant calculation to be
2223 * Reserve 0.78% of total space, at least 8MB for small filesystems.
2225 BUILD_BUG_ON(OSD_STATFS_RESERVED <= LDISKFS_MAX_BLOCK_SIZE);
2226 reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2227 if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2228 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2230 sfs->os_blocks -= reserved;
2231 sfs->os_bfree -= min(reserved, sfs->os_bfree);
2232 sfs->os_bavail -= min(reserved, sfs->os_bavail);
2235 if (unlikely(env == NULL))
2241 * Estimate space needed for file creations. We assume the largest filename
2242 * which is 2^64 - 1, hence a filename of 20 chars.
2243 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2245 #ifdef __LDISKFS_DIR_REC_LEN
2246 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2248 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2252 * Concurrency: doesn't access mutable data.
2254 static void osd_conf_get(const struct lu_env *env,
2255 const struct dt_device *dev,
2256 struct dt_device_param *param)
2258 struct osd_device *d = osd_dt_dev(dev);
2259 struct super_block *sb = osd_sb(d);
2260 struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2265 * XXX should be taken from not-yet-existing fs abstraction layer.
2267 param->ddp_max_name_len = LDISKFS_NAME_LEN;
2268 param->ddp_max_nlink = LDISKFS_LINK_MAX;
2269 param->ddp_symlink_max = sb->s_blocksize;
2270 param->ddp_mount_type = LDD_MT_LDISKFS;
2271 if (ldiskfs_has_feature_extents(sb))
2272 param->ddp_maxbytes = sb->s_maxbytes;
2274 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2276 * inode are statically allocated, so per-inode space consumption
2277 * is the space consumed by the directory entry
2279 param->ddp_inodespace = PER_OBJ_USAGE;
2281 * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2282 * is 128MB) which is unlikely to be hit in real life. Report a smaller
2283 * maximum length to not under-count the actual number of extents
2284 * needed for writing a file if there are sub-optimal block allocations.
2286 param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2287 /* worst-case extent insertion metadata overhead */
2288 param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2289 param->ddp_mntopts = 0;
2290 if (test_opt(sb, XATTR_USER))
2291 param->ddp_mntopts |= MNTOPT_USERXATTR;
2292 if (test_opt(sb, POSIX_ACL))
2293 param->ddp_mntopts |= MNTOPT_ACL;
2296 * LOD might calculate the max stripe count based on max_ea_size,
2297 * so we need take account in the overhead as well,
2298 * xattr_header + magic + xattr_entry_head
2300 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2301 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2303 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2304 if (ldiskfs_has_feature_ea_inode(sb))
2305 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2309 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2311 if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE)
2312 param->ddp_max_ea_size = OBD_MAX_EA_SIZE;
2315 * Preferred RPC size for efficient disk IO. 4MB shows good
2316 * all-around performance for ldiskfs, but use bigalloc chunk size
2317 * by default if larger.
2319 #if defined(LDISKFS_CLUSTER_SIZE)
2320 if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2321 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2324 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2326 param->ddp_t10_cksum_type = 0;
2328 unsigned short interval = blk_integrity_interval(bi);
2329 name = blk_integrity_name(bi);
2337 if (strncmp(name, "T10-DIF-TYPE",
2338 sizeof("T10-DIF-TYPE") - 1) == 0) {
2339 /* also skip "1/3-" at end */
2340 const int type_off = sizeof("T10-DIF-TYPE.");
2341 char type_number = name[type_off - 2];
2343 if (interval != 512 && interval != 4096) {
2344 CERROR("%s: unsupported T10PI sector size %u\n",
2345 d->od_svname, interval);
2346 } else if (type_number != '1' && type_number != '3') {
2347 CERROR("%s: unsupported T10PI type %s\n",
2348 d->od_svname, name);
2349 } else if (strcmp(name + type_off, "CRC") == 0) {
2350 d->od_t10_type = type_number == '1' ?
2351 OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2352 param->ddp_t10_cksum_type = interval == 512 ?
2353 OBD_CKSUM_T10CRC512 :
2355 } else if (strcmp(name + type_off, "IP") == 0) {
2356 d->od_t10_type = type_number == '1' ?
2357 OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2358 param->ddp_t10_cksum_type = interval == 512 ?
2359 OBD_CKSUM_T10IP512 :
2362 CERROR("%s: unsupported checksum type of "
2364 d->od_svname, name);
2368 CERROR("%s: unsupported T10PI type '%s'",
2369 d->od_svname, name);
2374 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2376 return osd_sb(osd_dt_dev(d));
2380 * Concurrency: shouldn't matter.
2382 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2385 struct super_block *s = osd_sb(osd_dt_dev(d));
2388 down_read(&s->s_umount);
2389 rc = s->s_op->sync_fs(s, 1);
2390 up_read(&s->s_umount);
2392 CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2399 * Start commit for OSD device.
2401 * An implementation of dt_commit_async method for OSD device.
2402 * Asychronously starts underlayng fs sync and thereby a transaction
2405 * \param env environment
2406 * \param d dt device
2408 * \see dt_device_operations
2410 static int osd_commit_async(const struct lu_env *env,
2411 struct dt_device *d)
2413 struct super_block *s = osd_sb(osd_dt_dev(d));
2418 CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2419 down_read(&s->s_umount);
2420 rc = s->s_op->sync_fs(s, 0);
2421 up_read(&s->s_umount);
2426 /* Our own copy of the set readonly functions if present, or NU if not. */
2427 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2428 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2429 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2430 static int (*priv_security_file_alloc)(struct file *file);
2432 int osd_security_file_alloc(struct file *file)
2434 if (priv_security_file_alloc)
2435 return priv_security_file_alloc(file);
2440 * Concurrency: shouldn't matter.
2442 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2444 struct super_block *sb = osd_sb(osd_dt_dev(d));
2445 struct block_device *dev = sb->s_bdev;
2446 int rc = -EOPNOTSUPP;
2450 if (priv_dev_set_rdonly) {
2451 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2454 CERROR("*** setting %s read-only ***\n",
2455 osd_dt_dev(d)->od_svname);
2457 if (sb->s_op->freeze_fs) {
2458 rc = sb->s_op->freeze_fs(sb);
2463 if (jdev && (jdev != dev)) {
2464 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2466 priv_dev_set_rdonly(jdev);
2468 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2469 priv_dev_set_rdonly(dev);
2471 if (sb->s_op->unfreeze_fs)
2472 sb->s_op->unfreeze_fs(sb);
2477 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2478 osd_dt_dev(d)->od_svname, (long)dev, rc);
2484 * Note: we do not count into QUOTA here.
2485 * If we mount with --data_journal we may need more.
2487 const int osd_dto_credits_noquota[DTO_NR] = {
2490 * INDEX_EXTRA_TRANS_BLOCKS(8) +
2491 * SINGLEDATA_TRANS_BLOCKS(8)
2492 * XXX Note: maybe iam need more, since iam have more level than
2495 [DTO_INDEX_INSERT] = 16,
2498 * just modify a single entry, probably merge few within a block
2500 [DTO_INDEX_DELETE] = 1,
2504 [DTO_INDEX_UPDATE] = 16,
2506 * 4(inode, inode bits, groups, GDT)
2507 * notice: OI updates are counted separately with DTO_INDEX_INSERT
2509 [DTO_OBJECT_CREATE] = 4,
2511 * 4(inode, inode bits, groups, GDT)
2512 * notice: OI updates are counted separately with DTO_INDEX_DELETE
2514 [DTO_OBJECT_DELETE] = 4,
2516 * Attr set credits (inode)
2518 [DTO_ATTR_SET_BASE] = 1,
2520 * Xattr set. The same as xattr of EXT3.
2521 * DATA_TRANS_BLOCKS(14)
2522 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2523 * are also counted in. Do not know why?
2525 [DTO_XATTR_SET] = 14,
2527 * credits for inode change during write.
2529 [DTO_WRITE_BASE] = 3,
2531 * credits for single block write.
2533 [DTO_WRITE_BLOCK] = 14,
2535 * Attr set credits for chown.
2536 * This is extra credits for setattr, and it is null without quota
2538 [DTO_ATTR_SET_CHOWN] = 0
2541 static const struct dt_device_operations osd_dt_ops = {
2542 .dt_root_get = osd_root_get,
2543 .dt_statfs = osd_statfs,
2544 .dt_trans_create = osd_trans_create,
2545 .dt_trans_start = osd_trans_start,
2546 .dt_trans_stop = osd_trans_stop,
2547 .dt_trans_cb_add = osd_trans_cb_add,
2548 .dt_conf_get = osd_conf_get,
2549 .dt_mnt_sb_get = osd_mnt_sb_get,
2550 .dt_sync = osd_sync,
2552 .dt_commit_async = osd_commit_async,
2555 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2558 struct osd_object *obj = osd_dt_obj(dt);
2559 struct osd_thread_info *oti = osd_oti_get(env);
2561 LINVRNT(osd_invariant(obj));
2563 LASSERT(obj->oo_owner != env);
2564 down_read_nested(&obj->oo_sem, role);
2566 LASSERT(obj->oo_owner == NULL);
2570 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2573 struct osd_object *obj = osd_dt_obj(dt);
2574 struct osd_thread_info *oti = osd_oti_get(env);
2576 LINVRNT(osd_invariant(obj));
2578 LASSERT(obj->oo_owner != env);
2579 down_write_nested(&obj->oo_sem, role);
2581 LASSERT(obj->oo_owner == NULL);
2582 obj->oo_owner = env;
2586 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2588 struct osd_object *obj = osd_dt_obj(dt);
2589 struct osd_thread_info *oti = osd_oti_get(env);
2591 LINVRNT(osd_invariant(obj));
2593 LASSERT(oti->oti_r_locks > 0);
2595 up_read(&obj->oo_sem);
2598 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2600 struct osd_object *obj = osd_dt_obj(dt);
2601 struct osd_thread_info *oti = osd_oti_get(env);
2603 LINVRNT(osd_invariant(obj));
2605 LASSERT(obj->oo_owner == env);
2606 LASSERT(oti->oti_w_locks > 0);
2608 obj->oo_owner = NULL;
2609 up_write(&obj->oo_sem);
2612 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2614 struct osd_object *obj = osd_dt_obj(dt);
2616 LINVRNT(osd_invariant(obj));
2618 return obj->oo_owner == env;
2621 static void osd_inode_getattr(const struct lu_env *env,
2622 struct inode *inode, struct lu_attr *attr)
2624 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2625 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2626 LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2627 LA_BLKSIZE | LA_TYPE;
2629 attr->la_atime = inode->i_atime.tv_sec;
2630 attr->la_mtime = inode->i_mtime.tv_sec;
2631 attr->la_ctime = inode->i_ctime.tv_sec;
2632 attr->la_mode = inode->i_mode;
2633 attr->la_size = i_size_read(inode);
2634 attr->la_blocks = inode->i_blocks;
2635 attr->la_uid = i_uid_read(inode);
2636 attr->la_gid = i_gid_read(inode);
2637 attr->la_projid = i_projid_read(inode);
2638 attr->la_flags = ll_inode_to_ext_flags(inode->i_flags);
2639 attr->la_nlink = inode->i_nlink;
2640 attr->la_rdev = inode->i_rdev;
2641 attr->la_blksize = 1 << inode->i_blkbits;
2642 attr->la_blkbits = inode->i_blkbits;
2644 * Ext4 did not transfer inherit flags from raw inode
2645 * to inode flags, and ext4 internally test raw inode
2646 * @i_flags directly. Instead of patching ext4, we do it here.
2648 if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2649 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2652 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2653 struct lu_attr *attr)
2655 struct osd_object *obj = osd_dt_obj(dt);
2657 if (unlikely(!dt_object_exists(dt)))
2659 if (unlikely(obj->oo_destroyed))
2662 LASSERT(!dt_object_remote(dt));
2663 LINVRNT(osd_invariant(obj));
2665 spin_lock(&obj->oo_guard);
2666 osd_inode_getattr(env, obj->oo_inode, attr);
2667 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2668 attr->la_valid |= LA_FLAGS;
2669 attr->la_flags |= LUSTRE_ORPHAN_FL;
2671 spin_unlock(&obj->oo_guard);
2676 static int osd_declare_attr_qid(const struct lu_env *env,
2677 struct osd_object *obj,
2678 struct osd_thandle *oh, long long bspace,
2679 qid_t old_id, qid_t new_id, bool enforce,
2680 unsigned int type, bool ignore_edquot)
2683 struct osd_thread_info *info = osd_oti_get(env);
2684 struct lquota_id_info *qi = &info->oti_qi;
2686 qi->lqi_type = type;
2687 /* inode accounting */
2688 qi->lqi_is_blk = false;
2690 /* one more inode for the new id ... */
2691 qi->lqi_id.qid_uid = new_id;
2693 /* Reserve credits for the new id */
2694 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2695 if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2700 /* and one less inode for the current id */
2701 qi->lqi_id.qid_uid = old_id;
2703 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2704 if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2709 /* block accounting */
2710 qi->lqi_is_blk = true;
2712 /* more blocks for the new id ... */
2713 qi->lqi_id.qid_uid = new_id;
2714 qi->lqi_space = bspace;
2716 * Credits for the new uid has been reserved, re-use "obj"
2717 * to save credit reservation.
2719 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2720 if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2725 /* and finally less blocks for the current uid */
2726 qi->lqi_id.qid_uid = old_id;
2727 qi->lqi_space = -bspace;
2728 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2729 if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2735 static int osd_declare_attr_set(const struct lu_env *env,
2736 struct dt_object *dt,
2737 const struct lu_attr *attr,
2738 struct thandle *handle)
2740 struct osd_thandle *oh;
2741 struct osd_object *obj;
2750 LASSERT(dt != NULL);
2751 LASSERT(handle != NULL);
2753 obj = osd_dt_obj(dt);
2754 LASSERT(osd_invariant(obj));
2756 oh = container_of0(handle, struct osd_thandle, ot_super);
2757 LASSERT(oh->ot_handle == NULL);
2759 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2760 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2762 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2763 osd_dto_credits_noquota[DTO_XATTR_SET]);
2765 if (attr == NULL || obj->oo_inode == NULL)
2768 bspace = obj->oo_inode->i_blocks << 9;
2769 bspace = toqb(bspace);
2772 * Changing ownership is always preformed by super user, it should not
2773 * fail with EDQUOT unless required explicitly.
2775 * We still need to call the osd_declare_qid() to calculate the journal
2776 * credits for updating quota accounting files and to trigger quota
2777 * space adjustment once the operation is completed.
2779 if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2780 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2783 CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2784 "(the quota space is %lld)\n",
2785 obj->oo_inode->i_sb->s_id, attr->la_uid,
2786 attr->la_gid, bspace);
2789 uid = i_uid_read(obj->oo_inode);
2790 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2791 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2792 attr->la_uid, enforce, USRQUOTA,
2797 gid = i_gid_read(obj->oo_inode);
2798 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2799 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2800 i_gid_read(obj->oo_inode),
2801 attr->la_gid, enforce, GRPQUOTA,
2807 #ifdef HAVE_PROJECT_QUOTA
2808 if (attr->la_valid & LA_PROJID) {
2809 __u32 projid = i_projid_read(obj->oo_inode);
2811 enforce = (attr->la_valid & LA_PROJID) &&
2812 (attr->la_projid != projid);
2813 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2814 (qid_t)projid, (qid_t)attr->la_projid,
2815 enforce, PRJQUOTA, true);
2823 static int osd_inode_setattr(const struct lu_env *env,
2824 struct inode *inode, const struct lu_attr *attr)
2826 __u64 bits = attr->la_valid;
2828 /* Only allow set size for regular file */
2829 if (!S_ISREG(inode->i_mode))
2830 bits &= ~(LA_SIZE | LA_BLOCKS);
2835 if (bits & LA_ATIME)
2836 inode->i_atime = osd_inode_time(inode, attr->la_atime);
2837 if (bits & LA_CTIME)
2838 inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
2839 if (bits & LA_MTIME)
2840 inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
2841 if (bits & LA_SIZE) {
2842 spin_lock(&inode->i_lock);
2843 LDISKFS_I(inode)->i_disksize = attr->la_size;
2844 i_size_write(inode, attr->la_size);
2845 spin_unlock(&inode->i_lock);
2849 * OSD should not change "i_blocks" which is used by quota.
2850 * "i_blocks" should be changed by ldiskfs only.
2853 inode->i_mode = (inode->i_mode & S_IFMT) |
2854 (attr->la_mode & ~S_IFMT);
2856 i_uid_write(inode, attr->la_uid);
2858 i_gid_write(inode, attr->la_gid);
2859 if (bits & LA_PROJID)
2860 i_projid_write(inode, attr->la_projid);
2861 if (bits & LA_NLINK)
2862 set_nlink(inode, attr->la_nlink);
2864 inode->i_rdev = attr->la_rdev;
2866 if (bits & LA_FLAGS) {
2867 /* always keep S_NOCMTIME */
2868 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2871 * Ext4 did not transfer inherit flags from
2872 * @inode->i_flags to raw inode i_flags when writing
2873 * flags, we do it explictly here.
2875 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2876 LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2878 LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2883 #ifdef HAVE_PROJECT_QUOTA
2884 static int osd_transfer_project(struct inode *inode, __u32 projid)
2886 struct super_block *sb = inode->i_sb;
2887 struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2890 struct ldiskfs_iloc iloc;
2891 struct ldiskfs_inode *raw_inode;
2892 struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2894 if (!ldiskfs_has_feature_project(sb)) {
2895 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2896 == LDISKFS_DEF_PROJID);
2897 if (projid != LDISKFS_DEF_PROJID)
2903 if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2906 kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2907 if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2910 err = ldiskfs_get_inode_loc(inode, &iloc);
2914 raw_inode = ldiskfs_raw_inode(&iloc);
2915 if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2922 dquot_initialize(inode);
2923 transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2924 if (transfer_to[PRJQUOTA]) {
2925 err = __dquot_transfer(inode, transfer_to);
2926 dqput(transfer_to[PRJQUOTA]);
2935 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2939 if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2940 (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2943 dquot_initialize(inode);
2945 if (attr->la_valid & LA_UID)
2946 iattr.ia_valid |= ATTR_UID;
2947 if (attr->la_valid & LA_GID)
2948 iattr.ia_valid |= ATTR_GID;
2949 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2950 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2952 rc = dquot_transfer(inode, &iattr);
2954 CERROR("%s: quota transfer failed: rc = %d. Is quota "
2955 "enforcement enabled on the ldiskfs "
2956 "filesystem?\n", inode->i_sb->s_id, rc);
2961 /* Handle project id transfer here properly */
2962 if (attr->la_valid & LA_PROJID &&
2963 attr->la_projid != i_projid_read(inode)) {
2964 #ifdef HAVE_PROJECT_QUOTA
2965 rc = osd_transfer_project(inode, attr->la_projid);
2970 CERROR("%s: quota transfer failed: rc = %d. Is project "
2971 "enforcement enabled on the ldiskfs "
2972 "filesystem?\n", inode->i_sb->s_id, rc);
2979 static int osd_attr_set(const struct lu_env *env,
2980 struct dt_object *dt,
2981 const struct lu_attr *attr,
2982 struct thandle *handle)
2984 struct osd_object *obj = osd_dt_obj(dt);
2985 struct inode *inode;
2988 if (!dt_object_exists(dt))
2991 LASSERT(handle != NULL);
2992 LASSERT(!dt_object_remote(dt));
2993 LASSERT(osd_invariant(obj));
2995 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2997 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
2998 !osd_obj2dev(obj)->od_is_ost) {
2999 struct osd_thread_info *oti = osd_oti_get(env);
3000 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
3001 struct lu_fid *fid1 = &oti->oti_fid;
3002 struct osd_inode_id *id = &oti->oti_id;
3003 struct iam_path_descr *ipd;
3004 struct iam_container *bag;
3005 struct osd_thandle *oh;
3008 fid_cpu_to_be(fid1, fid0);
3009 memset(id, 1, sizeof(*id));
3010 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
3011 fid0)->oi_dir.od_container;
3012 ipd = osd_idx_ipd_get(env, bag);
3013 if (unlikely(ipd == NULL))
3016 oh = container_of0(handle, struct osd_thandle, ot_super);
3017 rc = iam_update(oh->ot_handle, bag,
3018 (const struct iam_key *)fid1,
3019 (const struct iam_rec *)id, ipd);
3020 osd_ipd_put(env, bag, ipd);
3021 return(rc > 0 ? 0 : rc);
3024 inode = obj->oo_inode;
3026 rc = osd_quota_transfer(inode, attr);
3030 spin_lock(&obj->oo_guard);
3031 rc = osd_inode_setattr(env, inode, attr);
3032 spin_unlock(&obj->oo_guard);
3036 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3038 if (!(attr->la_valid & LA_FLAGS))
3041 /* Let's check if there are extra flags need to be set into LMA */
3042 if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3043 struct osd_thread_info *info = osd_oti_get(env);
3044 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3046 LASSERT(!obj->oo_pfid_in_lma);
3048 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3049 &info->oti_ost_attrs);
3053 lma->lma_incompat |=
3054 lustre_to_lma_flags(attr->la_flags);
3055 lustre_lma_swab(lma);
3056 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3057 lma, sizeof(*lma), XATTR_REPLACE);
3059 struct osd_device *osd = osd_obj2dev(obj);
3061 CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3062 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3063 lma->lma_incompat, rc);
3066 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3068 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3071 osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3076 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3077 struct osd_object *obj,
3078 const char *name, const int namelen)
3080 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3083 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3084 umode_t mode, struct dt_allocation_hint *hint,
3085 struct thandle *th, struct lu_attr *attr)
3088 struct osd_device *osd = osd_obj2dev(obj);
3089 struct osd_thandle *oth;
3090 struct dt_object *parent = NULL;
3091 struct inode *inode;
3092 uid_t owner[2] = {0, 0};
3094 if (attr->la_valid & LA_UID)
3095 owner[0] = attr->la_uid;
3096 if (attr->la_valid & LA_GID)
3097 owner[1] = attr->la_gid;
3099 LINVRNT(osd_invariant(obj));
3100 LASSERT(obj->oo_inode == NULL);
3101 LASSERT(obj->oo_hl_head == NULL);
3103 if (S_ISDIR(mode) && ldiskfs_pdo) {
3105 ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3106 if (obj->oo_hl_head == NULL)
3110 oth = container_of(th, struct osd_thandle, ot_super);
3111 LASSERT(oth->ot_handle->h_transaction != NULL);
3113 if (hint != NULL && hint->dah_parent != NULL &&
3114 !dt_object_remote(hint->dah_parent))
3115 parent = hint->dah_parent;
3117 inode = ldiskfs_create_inode(oth->ot_handle,
3118 parent ? osd_dt_obj(parent)->oo_inode :
3119 osd_sb(osd)->s_root->d_inode,
3121 if (!IS_ERR(inode)) {
3122 /* Do not update file c/mtime in ldiskfs. */
3123 inode->i_flags |= S_NOCMTIME;
3126 * For new created object, it must be consistent,
3127 * and it is unnecessary to scrub against it.
3129 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3131 obj->oo_inode = inode;
3134 if (obj->oo_hl_head != NULL) {
3135 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3136 obj->oo_hl_head = NULL;
3138 result = PTR_ERR(inode);
3140 LINVRNT(osd_invariant(obj));
3148 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3149 struct lu_attr *attr,
3150 struct dt_allocation_hint *hint,
3151 struct dt_object_format *dof,
3155 struct osd_thandle *oth;
3156 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3158 LASSERT(S_ISDIR(attr->la_mode));
3160 oth = container_of(th, struct osd_thandle, ot_super);
3161 LASSERT(oth->ot_handle->h_transaction != NULL);
3162 result = osd_mkfile(info, obj, mode, hint, th, attr);
3167 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3168 struct lu_attr *attr,
3169 struct dt_allocation_hint *hint,
3170 struct dt_object_format *dof,
3174 struct osd_thandle *oth;
3175 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3177 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3179 LASSERT(S_ISREG(attr->la_mode));
3181 oth = container_of(th, struct osd_thandle, ot_super);
3182 LASSERT(oth->ot_handle->h_transaction != NULL);
3184 result = osd_mkfile(info, obj, mode, hint, th, attr);
3186 LASSERT(obj->oo_inode != NULL);
3187 if (feat->dif_flags & DT_IND_VARKEY)
3188 result = iam_lvar_create(obj->oo_inode,
3189 feat->dif_keysize_max,
3191 feat->dif_recsize_max,
3194 result = iam_lfix_create(obj->oo_inode,
3195 feat->dif_keysize_max,
3197 feat->dif_recsize_max,
3203 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3204 struct lu_attr *attr,
3205 struct dt_allocation_hint *hint,
3206 struct dt_object_format *dof,
3209 LASSERT(S_ISREG(attr->la_mode));
3210 return osd_mkfile(info, obj, (attr->la_mode &
3211 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3215 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3216 struct lu_attr *attr,
3217 struct dt_allocation_hint *hint,
3218 struct dt_object_format *dof,
3221 LASSERT(S_ISLNK(attr->la_mode));
3222 return osd_mkfile(info, obj, (attr->la_mode &
3223 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3227 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3228 struct lu_attr *attr,
3229 struct dt_allocation_hint *hint,
3230 struct dt_object_format *dof,
3233 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3236 LINVRNT(osd_invariant(obj));
3237 LASSERT(obj->oo_inode == NULL);
3238 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3239 S_ISFIFO(mode) || S_ISSOCK(mode));
3241 result = osd_mkfile(info, obj, mode, hint, th, attr);
3243 LASSERT(obj->oo_inode != NULL);
3245 * This inode should be marked dirty for i_rdev. Currently
3246 * that is done in the osd_attr_init().
3248 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3251 LINVRNT(osd_invariant(obj));
3255 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3257 struct dt_allocation_hint *hint,
3258 struct dt_object_format *dof,
3261 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3263 osd_obj_type_f result;
3279 result = osd_mk_index;
3290 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3291 struct dt_object *parent, struct dt_object *child,
3296 ah->dah_parent = parent;
3297 ah->dah_mode = child_mode;
3299 if (parent != NULL && !dt_object_remote(parent)) {
3300 /* will help to find FID->ino at dt_insert("..") */
3301 struct osd_object *pobj = osd_dt_obj(parent);
3303 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3307 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3308 struct lu_attr *attr, struct dt_object_format *dof)
3310 struct inode *inode = obj->oo_inode;
3311 __u64 valid = attr->la_valid;
3314 attr->la_valid &= ~(LA_TYPE | LA_MODE);
3316 if (dof->dof_type != DFT_NODE)
3317 attr->la_valid &= ~LA_RDEV;
3318 if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3319 attr->la_valid &= ~LA_ATIME;
3320 if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3321 attr->la_valid &= ~LA_CTIME;
3322 if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3323 attr->la_valid &= ~LA_MTIME;
3325 result = osd_quota_transfer(inode, attr);
3329 if (attr->la_valid != 0) {
3330 result = osd_inode_setattr(info->oti_env, inode, attr);
3332 * The osd_inode_setattr() should always succeed here. The
3333 * only error that could be returned is EDQUOT when we are
3334 * trying to change the UID or GID of the inode. However, this
3335 * should not happen since quota enforcement is no longer
3336 * enabled on ldiskfs (lquota takes care of it).
3338 LASSERTF(result == 0, "%d\n", result);
3339 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3342 attr->la_valid = valid;
3346 * Helper function for osd_create()
3348 * \retval 0, on success
3350 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3351 struct lu_attr *attr, struct dt_allocation_hint *hint,
3352 struct dt_object_format *dof, struct thandle *th)
3357 osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3359 /* we drop umask so that permissions we pass are not affected */
3360 umask = current->fs->umask;
3361 current->fs->umask = 0;
3363 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3365 if (likely(obj->oo_inode != NULL)) {
3366 LASSERT(obj->oo_inode->i_state & I_NEW);
3369 * Unlock the inode before attr initialization to avoid
3370 * unnecessary dqget operations. LU-6378
3372 unlock_new_inode(obj->oo_inode);
3375 if (likely(result == 0)) {
3376 osd_attr_init(info, obj, attr, dof);
3377 osd_object_init0(obj);
3380 /* restore previous umask value */
3381 current->fs->umask = umask;
3383 osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3389 * Helper function for osd_create()
3391 * \retval 0, on success
3393 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3394 const struct lu_fid *fid, struct thandle *th)
3396 struct osd_thread_info *info = osd_oti_get(env);
3397 struct osd_inode_id *id = &info->oti_id;
3398 struct osd_device *osd = osd_obj2dev(obj);
3399 struct osd_thandle *oh;
3402 LASSERT(obj->oo_inode != NULL);
3404 oh = container_of0(th, struct osd_thandle, ot_super);
3405 LASSERT(oh->ot_handle);
3406 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3408 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3409 rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3410 OI_CHECK_FLD, NULL);
3411 if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP) && osd->od_is_ost) {
3412 struct lu_fid next_fid = *fid;
3414 /* insert next object in advance, and map to the same inode */
3416 if (next_fid.f_oid != 0) {
3417 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3418 osd_oi_insert(info, osd, &next_fid, id, oh->ot_handle,
3419 OI_CHECK_FLD, NULL);
3420 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3424 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3429 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3430 u64 seq, struct lu_seq_range *range)
3432 struct seq_server_site *ss = osd_seq_site(osd);
3434 if (fid_seq_is_idif(seq)) {
3435 fld_range_set_ost(range);
3436 range->lsr_index = idif_ost_idx(seq);
3440 if (!fid_seq_in_fldb(seq)) {
3441 fld_range_set_mdt(range);
3444 * FIXME: If ss is NULL, it suppose not get lsr_index
3447 range->lsr_index = ss->ss_node_id;
3451 LASSERT(ss != NULL);
3452 fld_range_set_any(range);
3453 /* OSD will only do local fld lookup */
3454 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3457 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3458 struct lu_attr *attr,
3459 struct dt_allocation_hint *hint,
3460 struct dt_object_format *dof,
3461 struct thandle *handle)
3463 struct osd_thandle *oh;
3468 LASSERT(handle != NULL);
3470 oh = container_of0(handle, struct osd_thandle, ot_super);
3471 LASSERT(oh->ot_handle == NULL);
3474 * EA object consumes more credits than regular object: osd_mk_index
3475 * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3476 * leaf_node, could involves the block, block bitmap, groups, GDT
3477 * change for each block, so add 4 * 2 credits in that case.
3479 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3480 osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3481 (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3483 * Reuse idle OI block may cause additional one OI block
3486 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3487 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3488 if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP))
3489 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3490 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3492 /* will help to find FID->ino mapping at dt_insert() */
3493 rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3501 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3502 attr->la_projid, 1, oh, osd_dt_obj(dt),
3503 NULL, OSD_QID_INODE);
3511 * Called to destroy on-disk representation of the object
3513 * Concurrency: must be locked
3515 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3518 struct osd_object *obj = osd_dt_obj(dt);
3519 struct inode *inode = obj->oo_inode;
3520 struct osd_thandle *oh;
3528 oh = container_of0(th, struct osd_thandle, ot_super);
3529 LASSERT(oh->ot_handle == NULL);
3531 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3532 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3534 /* For removing agent entry */
3535 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3536 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3539 * Recycle idle OI leaf may cause additional three OI blocks
3542 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3543 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3544 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3545 /* one less inode */
3546 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3547 i_projid_read(inode), -1, oh, obj, NULL,
3551 /* data to be truncated */
3552 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3553 i_projid_read(inode), 0, oh, obj, NULL,
3559 * will help to find FID->ino when this object is being
3562 rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3567 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3570 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3571 struct osd_object *obj = osd_dt_obj(dt);
3572 struct inode *inode = obj->oo_inode;
3573 struct osd_device *osd = osd_obj2dev(obj);
3574 struct osd_thandle *oh;
3579 oh = container_of0(th, struct osd_thandle, ot_super);
3580 LASSERT(oh->ot_handle);
3582 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3584 if (unlikely(fid_is_acct(fid)))
3587 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3588 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3590 CERROR("%s: remove agent entry "DFID": rc = %d\n",
3591 osd_name(osd), PFID(fid), result);
3594 if (S_ISDIR(inode->i_mode)) {
3595 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3596 inode->i_nlink == 2);
3598 spin_lock(&obj->oo_guard);
3600 spin_unlock(&obj->oo_guard);
3601 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3604 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3606 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3608 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3609 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3610 oh->ot_handle, OI_CHECK_FLD);
3612 osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3613 /* XXX: add to ext3 orphan list */
3614 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3616 /* not needed in the cache anymore */
3617 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3618 obj->oo_destroyed = 1;
3624 * Put the fid into lustre_mdt_attrs, and then place the structure
3625 * inode's ea. This fid should not be altered during the life time
3628 * \retval +ve, on success
3629 * \retval -ve, on error
3631 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3633 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3634 const struct lu_fid *fid, __u32 compat, __u32 incompat)
3636 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3637 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3642 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3645 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3648 lustre_loa_init(loa, fid, compat, incompat);
3649 lustre_loa_swab(loa, false);
3652 * For the OST device with 256 bytes inode size by default,
3653 * the PFID EA will be stored together with LMA EA to avoid
3654 * performance trouble. Otherwise the PFID EA can be stored
3655 * independently. LU-8998
3657 if ((compat & LMAC_FID_ON_OST) &&
3658 LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3659 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3660 sizeof(*loa), XATTR_CREATE);
3662 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3663 sizeof(*lma), XATTR_CREATE);
3665 * LMA may already exist, but we need to check that all the
3666 * desired compat/incompat flags have been added.
3668 if (unlikely(rc == -EEXIST)) {
3669 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3670 XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3674 if (rc < sizeof(*lma))
3677 lustre_loa_swab(loa, true);
3678 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3679 ((compat == 0 && incompat == 0) ||
3680 (!(~lma->lma_compat & compat) &&
3681 !(~lma->lma_incompat & incompat))))
3684 lma->lma_self_fid = *fid;
3685 lma->lma_compat |= compat;
3686 lma->lma_incompat |= incompat;
3687 if (rc == sizeof(*lma)) {
3688 lustre_lma_swab(lma);
3689 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3690 sizeof(*lma), XATTR_REPLACE);
3692 lustre_loa_swab(loa, false);
3693 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3694 sizeof(*loa), XATTR_REPLACE);
3702 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3703 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3704 * To have compatilibility with 1.8 ldiskfs driver we need to have
3705 * magic number at start of fid data.
3706 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3709 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3710 const struct lu_fid *fid)
3712 if (!fid_is_namespace_visible(fid) ||
3713 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3714 param->edp_magic = 0;
3718 param->edp_magic = LDISKFS_LUFID_MAGIC;
3719 param->edp_len = sizeof(struct lu_fid) + 1;
3720 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3724 * Try to read the fid from inode ea into dt_rec.
3726 * \param fid object fid.
3728 * \retval 0 on success
3730 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3731 __u32 ino, struct lu_fid *fid,
3732 struct osd_inode_id *id)
3734 struct osd_thread_info *info = osd_oti_get(env);
3735 struct inode *inode;
3739 osd_id_gen(id, ino, OSD_OII_NOGEN);
3740 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3742 RETURN(PTR_ERR(inode));
3748 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3750 struct inode *parent_dir,
3751 const struct lu_fid *dot_fid,
3752 const struct lu_fid *dot_dot_fid,
3753 struct osd_thandle *oth)
3755 struct ldiskfs_dentry_param *dot_ldp;
3756 struct ldiskfs_dentry_param *dot_dot_ldp;
3757 __u32 saved_nlink = dir->i_nlink;
3760 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3761 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3763 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3764 dot_ldp->edp_magic = 0;
3766 rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3767 dir, dot_ldp, dot_dot_ldp);
3769 * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3770 * the subseqent ref_add() will increase the dir->i_nlink
3771 * as 3. That is incorrect for new created directory.
3773 * It looks like hack, because we want to make the OSD API
3774 * to be order-independent for new created directory object
3775 * between dt_insert(..) and ref_add() operations.
3777 * Here, we only restore the in-RAM dir-inode's nlink attr,
3778 * becuase if the nlink attr is not 2, then there will be
3779 * ref_add() called following the dt_insert(..), such call
3780 * will make both the in-RAM and on-disk dir-inode's nlink
3781 * attr to be set as 2. LU-7447
3783 set_nlink(dir, saved_nlink);
3788 * Create an local agent inode for remote entry
3790 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3791 struct osd_device *osd,
3792 struct osd_object *pobj,
3793 const struct lu_fid *fid,
3797 struct osd_thread_info *info = osd_oti_get(env);
3798 struct inode *local;
3799 struct osd_thandle *oh;
3805 oh = container_of(th, struct osd_thandle, ot_super);
3806 LASSERT(oh->ot_handle->h_transaction != NULL);
3808 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type,
3810 if (IS_ERR(local)) {
3811 CERROR("%s: create local error %d\n", osd_name(osd),
3812 (int)PTR_ERR(local));
3817 * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3818 * correct gid on remote file, not agent here
3820 local->i_gid = current_fsgid();
3821 ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3823 /* e2fsck doesn't like empty symlinks. Store remote FID as symlink.
3824 * That gives e2fsck something to look at and be happy, and allows
3825 * debugging if we need to determine where this symlink came from.
3827 if (S_ISLNK(type)) {
3828 BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1);
3829 rc = snprintf((char *)LDISKFS_I(local)->i_data,
3830 LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
3832 i_size_write(local, rc);
3833 LDISKFS_I(local)->i_disksize = rc;
3835 unlock_new_inode(local);
3837 /* Agent inode should not have project ID */
3838 #ifdef HAVE_PROJECT_QUOTA
3839 if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3840 i_projid_read(pobj->oo_inode) != 0) {
3841 rc = osd_transfer_project(local, 0);
3843 CERROR("%s: quota transfer failed: rc = %d. Is project "
3844 "quota enforcement enabled on the ldiskfs "
3845 "filesystem?\n", local->i_sb->s_id, rc);
3846 RETURN(ERR_PTR(rc));
3850 /* Set special LMA flag for local agent inode */
3851 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3853 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3854 osd_name(osd), PFID(fid), rc);
3855 RETURN(ERR_PTR(rc));
3861 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3862 lu_object_fid(&pobj->oo_dt.do_lu),
3865 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3866 osd_name(osd), PFID(fid), rc);
3867 RETURN(ERR_PTR(rc));
3874 * when direntry is deleted, we have to take care of possible agent inode
3875 * referenced by that. unfortunately we can't do this at that point:
3876 * iget() within a running transaction leads to deadlock and we better do
3877 * not call that every delete declaration to save performance. so we put
3878 * a potention agent inode on a list and process that once the transaction
3879 * is over. Notice it's not any worse in terms of real orphans as regular
3880 * object destroy doesn't put inodes on the on-disk orphan list. this should
3881 * be addressed separately
3883 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3884 struct osd_thandle *oh,
3887 struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3888 struct osd_obj_orphan *oor;
3895 oor->oor_env = (struct lu_env *)env;
3896 spin_lock(&osd->od_osfs_lock);
3897 list_add(&oor->oor_list, &osd->od_orphan_list);
3898 spin_unlock(&osd->od_osfs_lock);
3900 oh->ot_remove_agents = 1;
3906 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3907 struct osd_device *osd)
3909 struct osd_thread_info *info = osd_oti_get(env);
3910 struct osd_obj_orphan *oor, *tmp;
3911 struct osd_inode_id id;
3912 struct list_head list;
3913 struct inode *inode;
3918 INIT_LIST_HEAD(&list);
3920 spin_lock(&osd->od_osfs_lock);
3921 list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3922 if (oor->oor_env == env) {
3923 list_del(&oor->oor_list);
3924 list_add(&oor->oor_list, &list);
3927 spin_unlock(&osd->od_osfs_lock);
3929 list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3933 list_del(&oor->oor_list);
3936 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3937 inode = osd_iget_fid(info, osd, &id, &fid);
3941 if (!osd_remote_fid(env, osd, &fid)) {
3946 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3948 mark_inode_dirty(inode);
3949 ldiskfs_journal_stop(jh);
3957 * OSD layer object create function for OST objects (b=11826).
3959 * The FID is inserted into inode xattr here.
3961 * \retval 0, on success
3962 * \retval -ve, on error
3964 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3965 struct lu_attr *attr, struct dt_allocation_hint *hint,
3966 struct dt_object_format *dof, struct thandle *th)
3968 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3969 struct osd_object *obj = osd_dt_obj(dt);
3970 struct osd_thread_info *info = osd_oti_get(env);
3971 int result, on_ost = 0;
3975 if (dt_object_exists(dt))
3978 LINVRNT(osd_invariant(obj));
3979 LASSERT(!dt_object_remote(dt));
3980 LASSERT(osd_is_write_locked(env, obj));
3981 LASSERT(th != NULL);
3983 if (unlikely(fid_is_acct(fid)))
3985 * Quota files can't be created from the kernel any more,
3986 * 'tune2fs -O quota' will take care of creating them
3990 result = __osd_create(info, obj, attr, hint, dof, th);
3992 if (fid_is_idif(fid) &&
3993 !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3994 struct lu_fid *tfid = &info->oti_fid;
3995 struct ost_id *oi = &info->oti_ostid;
3997 fid_to_ostid(fid, oi);
3998 ostid_to_fid(tfid, oi, 0);
4000 result = osd_ea_fid_set(info, obj->oo_inode, tfid,
4001 LMAC_FID_ON_OST, 0);
4003 on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
4005 result = osd_ea_fid_set(info, obj->oo_inode, fid,
4006 on_ost ? LMAC_FID_ON_OST : 0,
4009 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
4010 obj->oo_dt.do_body_ops = &osd_body_ops;
4013 if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
4014 result = __osd_oi_insert(env, obj, fid, th);
4017 * a small optimization - dt_insert() isn't usually applied
4018 * to OST objects, so we don't need to cache OI mapping for
4021 if (result == 0 && on_ost == 0) {
4022 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4024 result = osd_idc_find_and_init(env, osd, obj);
4025 LASSERT(result == 0);
4028 LASSERT(ergo(result == 0,
4029 dt_object_exists(dt) && !dt_object_remote(dt)));
4030 LINVRNT(osd_invariant(obj));
4034 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4035 struct thandle *handle)
4037 struct osd_thandle *oh;
4039 /* it's possible that object doesn't exist yet */
4040 LASSERT(handle != NULL);
4042 oh = container_of0(handle, struct osd_thandle, ot_super);
4043 LASSERT(oh->ot_handle == NULL);
4045 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4046 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4048 osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
4054 * Concurrency: @dt is write locked.
4056 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4059 struct osd_object *obj = osd_dt_obj(dt);
4060 struct inode *inode = obj->oo_inode;
4061 struct osd_thandle *oh;
4064 if (!dt_object_exists(dt) || obj->oo_destroyed)
4067 LINVRNT(osd_invariant(obj));
4068 LASSERT(!dt_object_remote(dt));
4069 LASSERT(osd_is_write_locked(env, obj));
4070 LASSERT(th != NULL);
4072 oh = container_of0(th, struct osd_thandle, ot_super);
4073 LASSERT(oh->ot_handle != NULL);
4075 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4077 CDEBUG(D_INODE, DFID" increase nlink %d\n",
4078 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4080 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4081 * (65000) subdirectories by storing "1" in i_nlink if the link count
4082 * would otherwise overflow. Directory tranversal tools understand
4083 * that (st_nlink == 1) indicates that the filesystem dose not track
4084 * hard links count on the directory, and will not abort subdirectory
4085 * scanning early once (st_nlink - 2) subdirs have been found.
4087 * This also has to properly handle the case of inodes with nlink == 0
4088 * in case they are being linked into the PENDING directory
4090 spin_lock(&obj->oo_guard);
4091 if (unlikely(inode->i_nlink == 0))
4092 /* inc_nlink from 0 may cause WARN_ON */
4093 set_nlink(inode, 1);
4095 ldiskfs_inc_count(oh->ot_handle, inode);
4096 if (!S_ISDIR(inode->i_mode))
4097 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4099 spin_unlock(&obj->oo_guard);
4101 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4102 LINVRNT(osd_invariant(obj));
4104 osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4109 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4110 struct thandle *handle)
4112 struct osd_thandle *oh;
4114 if (!dt_object_exists(dt))
4117 LASSERT(!dt_object_remote(dt));
4118 LASSERT(handle != NULL);
4120 oh = container_of0(handle, struct osd_thandle, ot_super);
4121 LASSERT(oh->ot_handle == NULL);
4123 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4124 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4130 * Concurrency: @dt is write locked.
4132 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4135 struct osd_object *obj = osd_dt_obj(dt);
4136 struct inode *inode = obj->oo_inode;
4137 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4138 struct osd_thandle *oh;
4140 if (!dt_object_exists(dt))
4143 LINVRNT(osd_invariant(obj));
4144 LASSERT(!dt_object_remote(dt));
4145 LASSERT(osd_is_write_locked(env, obj));
4146 LASSERT(th != NULL);
4148 oh = container_of0(th, struct osd_thandle, ot_super);
4149 LASSERT(oh->ot_handle != NULL);
4151 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4153 spin_lock(&obj->oo_guard);
4155 * That can be result of upgrade from old Lustre version and
4156 * applied only to local files. Just skip this ref_del call.
4157 * ext4_unlink() only treats this as a warning, don't LASSERT here.
4159 if (inode->i_nlink == 0) {
4160 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4161 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4162 ", maybe an upgraded file? (LU-3915)\n",
4163 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4164 spin_unlock(&obj->oo_guard);
4168 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4169 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4171 ldiskfs_dec_count(oh->ot_handle, inode);
4172 spin_unlock(&obj->oo_guard);
4174 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4175 LINVRNT(osd_invariant(obj));
4177 osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4183 * Concurrency: @dt is read locked.
4185 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4186 struct lu_buf *buf, const char *name)
4188 struct osd_object *obj = osd_dt_obj(dt);
4189 struct inode *inode = obj->oo_inode;
4190 struct osd_thread_info *info = osd_oti_get(env);
4191 struct dentry *dentry = &info->oti_obj_dentry;
4192 bool cache_xattr = false;
4197 /* version get is not real XATTR but uses xattr API */
4198 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4199 dt_obj_version_t *ver = buf->lb_buf;
4202 * for version we are just using xattr API but change inode
4205 if (buf->lb_len == 0)
4206 return sizeof(dt_obj_version_t);
4208 if (buf->lb_len < sizeof(dt_obj_version_t))
4211 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4212 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4214 *ver = LDISKFS_I(inode)->i_fs_version;
4216 return sizeof(dt_obj_version_t);
4219 if (!dt_object_exists(dt))
4222 LASSERT(!dt_object_remote(dt));
4223 LASSERT(inode->i_op != NULL);
4224 #ifdef HAVE_IOP_XATTR
4225 LASSERT(inode->i_op->getxattr != NULL);
4228 if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4229 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4233 rc = osd_oxc_get(obj, name, buf);
4238 if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4239 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4240 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4241 struct filter_fid *ff;
4242 struct ost_layout *ol;
4244 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4248 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4251 if (buf->lb_len == 0 || !buf->lb_buf)
4254 if (buf->lb_len < rc)
4258 ol = &ff->ff_layout;
4259 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4260 PFID_STRIPE_IDX_BITS);
4261 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4262 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4263 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4264 if (lma->lma_compat & LMAC_COMP_INFO) {
4265 ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4266 ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4267 ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4269 ol->ol_comp_start = 0;
4270 ol->ol_comp_end = 0;
4274 rc = __osd_xattr_get(inode, dentry, name,
4275 buf->lb_buf, buf->lb_len);
4279 if (rc == -ENOENT || rc == -ENODATA)
4280 osd_oxc_add(obj, name, NULL, 0);
4281 else if (rc > 0 && buf->lb_buf != NULL)
4282 osd_oxc_add(obj, name, buf->lb_buf, rc);
4288 static int osd_declare_xattr_set(const struct lu_env *env,
4289 struct dt_object *dt,
4290 const struct lu_buf *buf, const char *name,
4291 int fl, struct thandle *handle)
4293 struct osd_thandle *oh;
4295 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4297 LASSERT(handle != NULL);
4299 oh = container_of0(handle, struct osd_thandle, ot_super);
4300 LASSERT(oh->ot_handle == NULL);
4302 if (strcmp(name, XATTR_NAME_LMA) == 0) {
4304 * For non-upgrading case, the LMA is set first and
4305 * usually fit inode. But for upgrade case, the LMA
4306 * may be in another separated EA block.
4308 if (dt_object_exists(dt)) {
4309 if (fl == LU_XATTR_REPLACE)
4314 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4316 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4317 /* We may need to delete the old PFID EA. */
4318 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4319 if (fl == LU_XATTR_REPLACE)
4325 * If some name entry resides on remote MDT, then will create
4326 * agent entry under remote parent. On the other hand, if the
4327 * remote entry will be removed, then related agent entry may
4328 * need to be removed from the remote parent. So there may be
4329 * kinds of cases, let's declare enough credits. The credits
4330 * for create agent entry is enough for remove case.
4332 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4333 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4334 if (dt_object_exists(dt))
4335 credits += 1; /* For updating LMA */
4339 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4344 if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4346 * learn xattr size from osd_xattr_get if
4347 * attribute has not been read yet
4349 buflen = __osd_xattr_get(
4350 osd_dt_obj(dt)->oo_inode,
4351 &osd_oti_get(env)->oti_obj_dentry,
4356 buflen = buf->lb_len;
4359 if (buflen > sb->s_blocksize) {
4360 credits += osd_calc_bkmap_credits(
4362 (buflen + sb->s_blocksize - 1) >>
4363 sb->s_blocksize_bits);
4367 * xattr set may involve inode quota change, reserve credits for
4368 * dquot_initialize()
4370 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4373 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4378 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4379 const struct lu_buf *buf, int fl,
4380 struct thandle *handle)
4382 struct osd_thread_info *info = osd_oti_get(env);
4383 struct dentry *dentry = &info->oti_obj_dentry;
4384 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4385 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4386 struct inode *inode = obj->oo_inode;
4387 struct filter_fid *ff = buf->lb_buf;
4388 struct ost_layout *ol = &ff->ff_layout;
4389 int flags = XATTR_REPLACE;
4394 if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4397 rc = osd_get_lma(info, inode, dentry, loa);
4398 if (rc == -ENODATA) {
4399 /* Usually for upgarding from old device */
4400 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4401 LMAC_FID_ON_OST, 0);
4402 flags = XATTR_CREATE;
4407 if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4408 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4411 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4412 /* Separate PFID EA from LMA */
4413 lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4414 lustre_lma_swab(lma);
4415 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4416 sizeof(*lma), XATTR_REPLACE);
4418 obj->oo_pfid_in_lma = 0;
4419 rc = LU_XATTR_CREATE;
4425 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4429 * Old client does not send stripe information,
4430 * then store the PFID EA on disk separatedly.
4432 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4433 ol->ol_stripe_size == 0))
4436 /* Remove old PFID EA entry firstly. */
4437 dquot_initialize(inode);
4438 rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
4439 if (rc == -ENODATA) {
4440 if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4447 fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4448 if (likely(ol->ol_stripe_size != 0)) {
4449 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4450 PFID_STRIPE_IDX_BITS;
4451 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4452 lma->lma_compat |= LMAC_STRIPE_INFO;
4453 if (ol->ol_comp_id != 0) {
4454 loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4455 loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4456 loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4457 lma->lma_compat |= LMAC_COMP_INFO;
4461 lustre_loa_swab(loa, false);
4463 /* Store the PFID EA inside LMA. */
4464 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4467 obj->oo_pfid_in_lma = 1;
4473 * In DNE environment, the object (in spite of regular file or directory)
4474 * and its name entry may reside on different MDTs. Under such case, we will
4475 * create an agent entry on the MDT where the object resides. The agent entry
4476 * references the object locally, that makes the object to be visible to the
4477 * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4478 * such as 'tar' can handle the object properly.
4480 * We handle the agent entry during set linkEA that is the common interface
4481 * for both regular file and directroy, can handle kinds of cases, such as
4482 * create/link/unlink/rename, and so on.
4484 * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4486 * XXX: There are two known issues:
4487 * 1. For one object, we will create at most one agent entry even if there
4488 * may be more than one cross-MDTs hard links on the object. So the local
4489 * e2fsck may claim that the object's nlink is larger than the name entries
4490 * that reference such inode. And in further, the e2fsck will fix the nlink
4491 * attribute to match the local references. Then it will cause the object's
4492 * nlink attribute to be inconsistent with the global references. it is bad
4493 * but not fatal. The ref_del() can handle the zero-referenced case. On the
4494 * other hand, the global namespace LFSCK can repair the object's attribute
4495 * according to the linkEA.
4496 * 2. There may be too many hard links on the object as to its linkEA overflow,
4497 * then the linkEA entry for cross-MDTs reference may be discarded. If such
4498 * case happened, then at this point, we do not know whether there are some
4499 * cross-MDTs reference. But there are local references, it guarantees that
4500 * object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4502 static int osd_xattr_handle_linkea(const struct lu_env *env,
4503 struct osd_device *osd,
4504 struct osd_object *obj,
4505 const struct lu_buf *buf,
4506 struct thandle *handle)
4508 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4509 struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4510 struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4511 struct lu_name tmpname;
4512 struct osd_thandle *oh;
4514 bool remote = false;
4518 oh = container_of0(handle, struct osd_thandle, ot_super);
4519 LASSERT(oh->ot_handle != NULL);
4521 rc = linkea_init_with_rec(&ldata);
4523 linkea_first_entry(&ldata);
4524 while (ldata.ld_lee != NULL && !remote) {
4525 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4527 if (osd_remote_fid(env, osd, tfid) > 0)
4530 linkea_next_entry(&ldata);
4532 } else if (rc == -ENODATA) {
4538 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4539 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4541 CERROR("%s: failed to remove agent entry for "DFID
4542 ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4543 } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4544 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4546 CERROR("%s: failed to create agent entry for "DFID
4547 ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4554 * Concurrency: @dt is write locked.
4556 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4557 const struct lu_buf *buf, const char *name, int fl,
4558 struct thandle *handle)
4560 struct osd_object *obj = osd_dt_obj(dt);
4561 struct osd_device *osd = osd_obj2dev(obj);
4562 struct inode *inode = obj->oo_inode;
4563 struct osd_thread_info *info = osd_oti_get(env);
4573 /* version set is not real XATTR */
4574 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4575 dt_obj_version_t *version = buf->lb_buf;
4578 * for version we are just using xattr API but change inode
4581 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4583 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4584 *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4586 LDISKFS_I(inode)->i_fs_version = *version;
4588 * Version is set after all inode operations are finished,
4589 * so we should mark it dirty here
4591 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4596 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4597 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4600 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4603 * For the OST device with 256 bytes inode size by default,
4604 * the PFID EA will be stored together with LMA EA to avoid
4605 * performance trouble. Otherwise the PFID EA can be stored
4606 * independently. LU-8998
4608 if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4609 (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4610 LASSERT(buf->lb_buf);
4612 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4615 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4616 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4617 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4619 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4623 lma->lma_incompat |= LMAI_STRIPED;
4624 lustre_lma_swab(lma);
4625 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4626 sizeof(*lma), XATTR_REPLACE);
4629 } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4630 LASSERT(!osd->od_is_ost);
4632 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4637 if (fl & LU_XATTR_REPLACE)
4638 fs_flags |= XATTR_REPLACE;
4640 if (fl & LU_XATTR_CREATE)
4641 fs_flags |= XATTR_CREATE;
4643 rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4644 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4647 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4648 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4649 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4655 * Concurrency: @dt is read locked.
4657 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4658 const struct lu_buf *buf)
4660 struct osd_object *obj = osd_dt_obj(dt);
4661 struct inode *inode = obj->oo_inode;
4662 struct osd_thread_info *info = osd_oti_get(env);
4663 struct dentry *dentry = &info->oti_obj_dentry;
4665 if (!dt_object_exists(dt))
4668 LASSERT(!dt_object_remote(dt));
4669 LASSERT(inode->i_op != NULL);
4670 LASSERT(inode->i_op->listxattr != NULL);
4672 dentry->d_inode = inode;
4673 dentry->d_sb = inode->i_sb;
4674 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4677 static int osd_declare_xattr_del(const struct lu_env *env,
4678 struct dt_object *dt, const char *name,
4679 struct thandle *handle)
4681 struct osd_thandle *oh;
4682 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4684 LASSERT(!dt_object_remote(dt));
4685 LASSERT(handle != NULL);
4687 oh = container_of0(handle, struct osd_thandle, ot_super);
4688 LASSERT(oh->ot_handle == NULL);
4690 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4691 osd_dto_credits_noquota[DTO_XATTR_SET]);
4693 * xattr del may involve inode quota change, reserve credits for
4694 * dquot_initialize()
4696 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4702 * Concurrency: @dt is write locked.
4704 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4705 const char *name, struct thandle *handle)
4707 struct osd_object *obj = osd_dt_obj(dt);
4708 struct inode *inode = obj->oo_inode;
4709 struct osd_thread_info *info = osd_oti_get(env);
4710 struct dentry *dentry = &info->oti_obj_dentry;
4713 if (!dt_object_exists(dt))
4716 LASSERT(!dt_object_remote(dt));
4717 LASSERT(inode->i_op != NULL);
4718 LASSERT(handle != NULL);
4719 #ifdef HAVE_IOP_XATTR
4720 LASSERT(inode->i_op->removexattr != NULL);
4723 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4725 if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4726 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4728 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4729 &info->oti_ost_attrs);
4731 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4733 lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4734 lustre_lma_swab(lma);
4735 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4736 sizeof(*lma), XATTR_REPLACE);
4738 obj->oo_pfid_in_lma = 0;
4741 dquot_initialize(inode);
4742 dentry->d_inode = inode;
4743 dentry->d_sb = inode->i_sb;
4744 rc = osd_removexattr(dentry, inode, name);
4747 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4750 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4751 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4752 osd_oxc_del(obj, name);
4757 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4758 __u64 start, __u64 end)
4760 struct osd_object *obj = osd_dt_obj(dt);
4761 struct inode *inode = obj->oo_inode;
4762 struct osd_thread_info *info = osd_oti_get(env);
4763 struct dentry *dentry = &info->oti_obj_dentry;
4764 struct file *file = &info->oti_file;
4769 dentry->d_inode = inode;
4770 dentry->d_sb = inode->i_sb;
4771 file->f_path.dentry = dentry;
4772 file->f_mapping = inode->i_mapping;
4773 file->f_op = inode->i_fop;
4774 set_file_inode(file, inode);
4776 rc = vfs_fsync_range(file, start, end, 0);
4781 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4790 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4791 const struct dt_index_features *feat)
4793 struct iam_descr *descr;
4795 if (osd_object_is_root(o))
4796 return feat == &dt_directory_features;
4798 LASSERT(o->oo_dir != NULL);
4800 descr = o->oo_dir->od_container.ic_descr;
4801 if (feat == &dt_directory_features) {
4802 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4807 return feat->dif_keysize_min <= descr->id_key_size &&
4808 descr->id_key_size <= feat->dif_keysize_max &&
4809 feat->dif_recsize_min <= descr->id_rec_size &&
4810 descr->id_rec_size <= feat->dif_recsize_max &&
4811 !(feat->dif_flags & (DT_IND_VARKEY |
4812 DT_IND_VARREC | DT_IND_NONUNQ)) &&
4813 ergo(feat->dif_flags & DT_IND_UPDATE,
4814 1 /* XXX check that object (and fs) is writable */);
4818 static int osd_iam_container_init(const struct lu_env *env,
4819 struct osd_object *obj,
4820 struct osd_directory *dir)
4822 struct iam_container *bag = &dir->od_container;
4825 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4829 result = iam_container_setup(bag);
4831 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4833 iam_container_fini(bag);
4840 * Concurrency: no external locking is necessary.
4842 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4843 const struct dt_index_features *feat)
4847 struct osd_object *obj = osd_dt_obj(dt);
4849 LINVRNT(osd_invariant(obj));
4851 if (osd_object_is_root(obj)) {
4852 dt->do_index_ops = &osd_index_ea_ops;
4854 } else if (feat == &dt_directory_features) {
4855 dt->do_index_ops = &osd_index_ea_ops;
4856 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4861 } else if (unlikely(feat == &dt_otable_features)) {
4862 dt->do_index_ops = &osd_otable_ops;
4864 } else if (unlikely(feat == &dt_acct_features)) {
4865 dt->do_index_ops = &osd_acct_index_ops;
4868 } else if (!osd_has_index(obj)) {
4869 struct osd_directory *dir;
4870 struct osd_device *osd = osd_obj2dev(obj);
4871 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4876 spin_lock(&obj->oo_guard);
4877 if (obj->oo_dir == NULL)
4881 * Concurrent thread allocated container data.
4884 spin_unlock(&obj->oo_guard);
4886 * Now, that we have container data, serialize its
4889 down_write(&obj->oo_ext_idx_sem);
4891 * recheck under lock.
4894 if (osd_has_index(obj)) {
4899 result = osd_iam_container_init(env, obj, obj->oo_dir);
4900 if (result || feat == &dt_lfsck_namespace_features ||
4901 feat == &dt_lfsck_layout_orphan_features ||
4902 feat == &dt_lfsck_layout_dangling_features)
4905 result = osd_index_register(osd, fid,
4906 feat->dif_keysize_max,
4907 feat->dif_recsize_max);
4909 CWARN("%s: failed to register index "
4911 osd_name(osd), PFID(fid), result);
4912 else if (result > 0)
4915 CDEBUG(D_LFSCK, "%s: index object "DFID
4916 " (%d/%d) registered\n",
4917 osd_name(osd), PFID(fid),
4918 (int)feat->dif_keysize_max,
4919 (int)feat->dif_recsize_max);
4922 up_write(&obj->oo_ext_idx_sem);
4930 if (result == 0 && skip_iam == 0) {
4931 if (!osd_iam_index_probe(env, obj, feat))
4934 LINVRNT(osd_invariant(obj));
4939 static int osd_otable_it_attr_get(const struct lu_env *env,
4940 struct dt_object *dt,
4941 struct lu_attr *attr)
4947 static const struct dt_object_operations osd_obj_ops = {
4948 .do_read_lock = osd_read_lock,
4949 .do_write_lock = osd_write_lock,
4950 .do_read_unlock = osd_read_unlock,
4951 .do_write_unlock = osd_write_unlock,
4952 .do_write_locked = osd_write_locked,
4953 .do_attr_get = osd_attr_get,
4954 .do_declare_attr_set = osd_declare_attr_set,
4955 .do_attr_set = osd_attr_set,
4956 .do_ah_init = osd_ah_init,
4957 .do_declare_create = osd_declare_create,
4958 .do_create = osd_create,
4959 .do_declare_destroy = osd_declare_destroy,
4960 .do_destroy = osd_destroy,
4961 .do_index_try = osd_index_try,
4962 .do_declare_ref_add = osd_declare_ref_add,
4963 .do_ref_add = osd_ref_add,
4964 .do_declare_ref_del = osd_declare_ref_del,
4965 .do_ref_del = osd_ref_del,
4966 .do_xattr_get = osd_xattr_get,
4967 .do_declare_xattr_set = osd_declare_xattr_set,
4968 .do_xattr_set = osd_xattr_set,
4969 .do_declare_xattr_del = osd_declare_xattr_del,
4970 .do_xattr_del = osd_xattr_del,
4971 .do_xattr_list = osd_xattr_list,
4972 .do_object_sync = osd_object_sync,
4973 .do_invalidate = osd_invalidate,
4976 static const struct dt_object_operations osd_obj_otable_it_ops = {
4977 .do_attr_get = osd_otable_it_attr_get,
4978 .do_index_try = osd_index_try,
4981 static int osd_index_declare_iam_delete(const struct lu_env *env,
4982 struct dt_object *dt,
4983 const struct dt_key *key,
4984 struct thandle *handle)
4986 struct osd_thandle *oh;
4988 oh = container_of0(handle, struct osd_thandle, ot_super);
4989 LASSERT(oh->ot_handle == NULL);
4991 /* Recycle may cause additional three blocks to be changed. */
4992 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4993 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4999 * delete a (key, value) pair from index \a dt specified by \a key
5001 * \param dt osd index object
5002 * \param key key for index
5003 * \param rec record reference
5004 * \param handle transaction handler
5007 * \retval -ve failure
5009 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
5010 const struct dt_key *key,
5011 struct thandle *handle)
5013 struct osd_thread_info *oti = osd_oti_get(env);
5014 struct osd_object *obj = osd_dt_obj(dt);
5015 struct osd_thandle *oh;
5016 struct iam_path_descr *ipd;
5017 struct iam_container *bag = &obj->oo_dir->od_container;
5022 if (!dt_object_exists(dt))
5025 LINVRNT(osd_invariant(obj));
5026 LASSERT(!dt_object_remote(dt));
5027 LASSERT(bag->ic_object == obj->oo_inode);
5028 LASSERT(handle != NULL);
5030 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5032 ipd = osd_idx_ipd_get(env, bag);
5033 if (unlikely(ipd == NULL))
5036 oh = container_of0(handle, struct osd_thandle, ot_super);
5037 LASSERT(oh->ot_handle != NULL);
5038 LASSERT(oh->ot_handle->h_transaction != NULL);
5040 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5041 /* swab quota uid/gid provided by caller */
5042 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5043 key = (const struct dt_key *)&oti->oti_quota_id;
5046 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5047 osd_ipd_put(env, bag, ipd);
5048 LINVRNT(osd_invariant(obj));
5049 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5053 static int osd_index_declare_ea_delete(const struct lu_env *env,
5054 struct dt_object *dt,
5055 const struct dt_key *key,
5056 struct thandle *handle)
5058 struct osd_thandle *oh;
5059 struct inode *inode;
5064 LASSERT(!dt_object_remote(dt));
5065 LASSERT(handle != NULL);
5067 oh = container_of0(handle, struct osd_thandle, ot_super);
5068 LASSERT(oh->ot_handle == NULL);
5070 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5071 osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5073 inode = osd_dt_obj(dt)->oo_inode;
5077 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5078 i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5083 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5086 struct osd_fid_pack *rec;
5089 if (de->file_type & LDISKFS_DIRENT_LUFID) {
5090 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5091 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5092 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5098 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5099 const struct lu_fid *fid)
5101 struct seq_server_site *ss = osd_seq_site(osd);
5105 /* FID seqs not in FLDB, must be local seq */
5106 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5110 * If FLD is not being initialized yet, it only happens during the
5111 * initialization, likely during mgs initialization, and we assume
5112 * this is local FID.
5114 if (ss == NULL || ss->ss_server_fld == NULL)
5117 /* Only check the local FLDB here */
5118 if (osd_seq_exists(env, osd, fid_seq(fid)))
5124 static void osd_take_care_of_agent(const struct lu_env *env,
5125 struct osd_device *osd,
5126 struct osd_thandle *oh,
5127 struct ldiskfs_dir_entry_2 *de)
5129 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5130 struct osd_idmap_cache *idc;
5131 int rc, schedule = 0;
5133 LASSERT(de != NULL);
5135 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5136 if (likely(rc == 0)) {
5137 idc = osd_idc_find_or_init(env, osd, fid);
5138 if (IS_ERR(idc) || idc->oic_remote)
5140 } else if (rc == -ENODATA) {
5142 * can't get FID, postpone to the end of the
5143 * transaction when iget() is safe
5147 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5150 osd_schedule_agent_inode_removal(env, oh,
5151 le32_to_cpu(de->inode));
5155 * Index delete function for interoperability mode (b11826).
5156 * It will remove the directory entry added by osd_index_ea_insert().
5157 * This entry is needed to maintain name->fid mapping.
5159 * \param key, key i.e. file entry to be deleted
5161 * \retval 0, on success
5162 * \retval -ve, on error
5164 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5165 const struct dt_key *key, struct thandle *handle)
5167 struct osd_object *obj = osd_dt_obj(dt);
5168 struct inode *dir = obj->oo_inode;
5169 struct dentry *dentry;
5170 struct osd_thandle *oh;
5171 struct ldiskfs_dir_entry_2 *de = NULL;
5172 struct buffer_head *bh;
5173 struct htree_lock *hlock = NULL;
5174 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5179 if (!dt_object_exists(dt))
5182 LINVRNT(osd_invariant(obj));
5183 LASSERT(!dt_object_remote(dt));
5184 LASSERT(handle != NULL);
5186 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5188 oh = container_of(handle, struct osd_thandle, ot_super);
5189 LASSERT(oh->ot_handle != NULL);
5190 LASSERT(oh->ot_handle->h_transaction != NULL);
5192 dquot_initialize(dir);
5193 dentry = osd_child_dentry_get(env, obj,
5194 (char *)key, strlen((char *)key));
5196 if (obj->oo_hl_head != NULL) {
5197 hlock = osd_oti_get(env)->oti_hlock;
5198 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5199 dir, LDISKFS_HLOCK_DEL);
5201 down_write(&obj->oo_ext_idx_sem);
5204 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5207 * If this is not the ".." entry, it might be a remote DNE
5208 * entry and we need to check if the FID is for a remote
5209 * MDT. If the FID is not in the directory entry (e.g.
5210 * upgraded 1.8 filesystem without dirdata enabled) then
5211 * we need to get the FID from the LMA. For a remote directory
5212 * there HAS to be an LMA, it cannot be an IGIF inode in this
5215 * Delete the entry before the agent inode in order to
5216 * simplify error handling. At worst an error after deleting
5217 * the entry first might leak the agent inode afterward. The
5218 * reverse would need filesystem abort in case of error deleting
5219 * the entry after the agent had been removed, or leave a
5220 * dangling entry pointing at a random inode.
5222 if (strcmp((char *)key, dotdot) != 0)
5223 osd_take_care_of_agent(env, osd, oh, de);
5224 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5230 ldiskfs_htree_unlock(hlock);
5232 up_write(&obj->oo_ext_idx_sem);
5236 LASSERT(osd_invariant(obj));
5237 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5242 * Lookup index for \a key and copy record to \a rec.
5244 * \param dt osd index object
5245 * \param key key for index
5246 * \param rec record reference
5248 * \retval +ve success : exact mach
5249 * \retval 0 return record with key not greater than \a key
5250 * \retval -ve failure
5252 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5253 struct dt_rec *rec, const struct dt_key *key)
5255 struct osd_object *obj = osd_dt_obj(dt);
5256 struct iam_path_descr *ipd;
5257 struct iam_container *bag = &obj->oo_dir->od_container;
5258 struct osd_thread_info *oti = osd_oti_get(env);
5259 struct iam_iterator *it = &oti->oti_idx_it;
5260 struct iam_rec *iam_rec;
5265 if (!dt_object_exists(dt))
5268 LASSERT(osd_invariant(obj));
5269 LASSERT(!dt_object_remote(dt));
5270 LASSERT(bag->ic_object == obj->oo_inode);
5272 ipd = osd_idx_ipd_get(env, bag);
5276 /* got ipd now we can start iterator. */
5277 iam_it_init(it, bag, 0, ipd);
5279 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5280 /* swab quota uid/gid provided by caller */
5281 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5282 key = (const struct dt_key *)&oti->oti_quota_id;
5285 rc = iam_it_get(it, (struct iam_key *)key);
5287 if (S_ISDIR(obj->oo_inode->i_mode))
5288 iam_rec = (struct iam_rec *)oti->oti_ldp;
5290 iam_rec = (struct iam_rec *)rec;
5292 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5294 if (S_ISDIR(obj->oo_inode->i_mode))
5295 osd_fid_unpack((struct lu_fid *)rec,
5296 (struct osd_fid_pack *)iam_rec);
5297 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5298 osd_quota_unpack(obj, rec);
5303 osd_ipd_put(env, bag, ipd);
5305 LINVRNT(osd_invariant(obj));
5310 static int osd_index_declare_iam_insert(const struct lu_env *env,
5311 struct dt_object *dt,
5312 const struct dt_rec *rec,
5313 const struct dt_key *key,
5314 struct thandle *handle)
5316 struct osd_thandle *oh;
5318 LASSERT(handle != NULL);
5320 oh = container_of0(handle, struct osd_thandle, ot_super);
5321 LASSERT(oh->ot_handle == NULL);
5323 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5324 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5330 * Inserts (key, value) pair in \a dt index object.
5332 * \param dt osd index object
5333 * \param key key for index
5334 * \param rec record reference
5335 * \param th transaction handler
5338 * \retval -ve failure
5340 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5341 const struct dt_rec *rec,
5342 const struct dt_key *key, struct thandle *th)
5344 struct osd_object *obj = osd_dt_obj(dt);
5345 struct iam_path_descr *ipd;
5346 struct osd_thandle *oh;
5347 struct iam_container *bag;
5348 struct osd_thread_info *oti = osd_oti_get(env);
5349 struct iam_rec *iam_rec;
5354 if (!dt_object_exists(dt))
5357 LINVRNT(osd_invariant(obj));
5358 LASSERT(!dt_object_remote(dt));
5360 bag = &obj->oo_dir->od_container;
5361 LASSERT(bag->ic_object == obj->oo_inode);
5362 LASSERT(th != NULL);
5364 osd_trans_exec_op(env, th, OSD_OT_INSERT);
5366 ipd = osd_idx_ipd_get(env, bag);
5367 if (unlikely(ipd == NULL))
5370 oh = container_of0(th, struct osd_thandle, ot_super);
5371 LASSERT(oh->ot_handle != NULL);
5372 LASSERT(oh->ot_handle->h_transaction != NULL);
5373 if (S_ISDIR(obj->oo_inode->i_mode)) {
5374 iam_rec = (struct iam_rec *)oti->oti_ldp;
5375 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5377 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5378 /* pack quota uid/gid */
5379 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5380 key = (const struct dt_key *)&oti->oti_quota_id;
5381 /* pack quota record */
5382 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5383 iam_rec = (struct iam_rec *)rec;
5385 iam_rec = (struct iam_rec *)rec;
5388 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5390 osd_ipd_put(env, bag, ipd);
5391 LINVRNT(osd_invariant(obj));
5392 osd_trans_exec_check(env, th, OSD_OT_INSERT);
5397 * Calls ldiskfs_add_entry() to add directory entry
5398 * into the directory. This is required for
5399 * interoperability mode (b11826)
5401 * \retval 0, on success
5402 * \retval -ve, on error
5404 static int __osd_ea_add_rec(struct osd_thread_info *info,
5405 struct osd_object *pobj, struct inode *cinode,
5406 const char *name, const struct lu_fid *fid,
5407 struct htree_lock *hlock, struct thandle *th)
5409 struct ldiskfs_dentry_param *ldp;
5410 struct dentry *child;
5411 struct osd_thandle *oth;
5414 oth = container_of(th, struct osd_thandle, ot_super);
5415 LASSERT(oth->ot_handle != NULL);
5416 LASSERT(oth->ot_handle->h_transaction != NULL);
5417 LASSERT(pobj->oo_inode);
5419 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5420 if (unlikely(pobj->oo_inode ==
5421 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5424 osd_get_ldiskfs_dirent_param(ldp, fid);
5425 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5426 child->d_fsdata = (void *)ldp;
5427 dquot_initialize(pobj->oo_inode);
5428 rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5429 child, cinode, hlock);
5430 if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5431 struct ldiskfs_dir_entry_2 *de;
5432 struct buffer_head *bh;
5435 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5438 rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5441 if (S_ISDIR(cinode->i_mode))
5442 de->file_type = LDISKFS_DIRENT_LUFID |
5443 LDISKFS_FT_REG_FILE;
5445 de->file_type = LDISKFS_DIRENT_LUFID |
5447 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5458 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5459 * into the directory.Also sets flags into osd object to
5460 * indicate dot and dotdot are created. This is required for
5461 * interoperability mode (b11826)
5463 * \param dir directory for dot and dotdot fixup.
5464 * \param obj child object for linking
5466 * \retval 0, on success
5467 * \retval -ve, on error
5469 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5470 struct osd_object *dir,
5471 struct inode *parent_dir, const char *name,
5472 const struct lu_fid *dot_fid,
5473 const struct lu_fid *dot_dot_fid,
5476 struct inode *inode = dir->oo_inode;
5477 struct osd_thandle *oth;
5480 oth = container_of(th, struct osd_thandle, ot_super);
5481 LASSERT(oth->ot_handle->h_transaction != NULL);
5482 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5484 if (strcmp(name, dot) == 0) {
5485 if (dir->oo_compat_dot_created) {
5488 LASSERT(inode->i_ino == parent_dir->i_ino);
5489 dir->oo_compat_dot_created = 1;
5492 } else if (strcmp(name, dotdot) == 0) {
5493 if (!dir->oo_compat_dot_created)
5495 /* in case of rename, dotdot is already created */
5496 if (dir->oo_compat_dotdot_created) {
5497 return __osd_ea_add_rec(info, dir, parent_dir, name,
5498 dot_dot_fid, NULL, th);
5501 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5502 struct lu_fid tfid = *dot_dot_fid;
5505 result = osd_add_dot_dotdot_internal(info,
5506 dir->oo_inode, parent_dir, dot_fid,
5509 result = osd_add_dot_dotdot_internal(info,
5510 dir->oo_inode, parent_dir, dot_fid,
5515 dir->oo_compat_dotdot_created = 1;
5523 * It will call the appropriate osd_add* function and return the
5524 * value, return by respective functions.
5526 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5527 struct inode *cinode, const char *name,
5528 const struct lu_fid *fid, struct thandle *th)
5530 struct osd_thread_info *info = osd_oti_get(env);
5531 struct htree_lock *hlock;
5534 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5536 if (name[0] == '.' && (name[1] == '\0' ||
5537 (name[1] == '.' && name[2] == '\0'))) {
5538 if (hlock != NULL) {
5539 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5542 down_write(&pobj->oo_ext_idx_sem);
5545 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5546 lu_object_fid(&pobj->oo_dt.do_lu),
5549 if (hlock != NULL) {
5550 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5551 pobj->oo_inode, LDISKFS_HLOCK_ADD);
5553 down_write(&pobj->oo_ext_idx_sem);
5556 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5557 struct lu_fid *tfid = &info->oti_fid;
5561 rc = __osd_ea_add_rec(info, pobj, cinode, name,
5564 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5569 ldiskfs_htree_unlock(hlock);
5571 up_write(&pobj->oo_ext_idx_sem);
5577 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5578 struct osd_idmap_cache *oic)
5580 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5581 struct lu_fid *fid = &oic->oic_fid;
5582 struct osd_inode_id *id = &oic->oic_lid;
5583 struct inode *inode = NULL;
5590 if (!fid_is_norm(fid) && !fid_is_igif(fid))
5593 if (thread_is_running(&scrub->os_thread) &&
5594 scrub->os_pos_current > id->oii_ino)
5597 if (dev->od_auto_scrub_interval == AS_NEVER ||
5598 ktime_get_real_seconds() <
5599 scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5603 rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5604 if (rc == -ENOENT) {
5605 __u32 gen = id->oii_gen;
5611 inode = osd_iget(oti, dev, id);
5612 /* The inode has been removed (by race maybe). */
5613 if (IS_ERR(inode)) {
5614 rc = PTR_ERR(inode);
5616 RETURN(rc == -ESTALE ? -ENOENT : rc);
5619 /* The OI mapping is lost. */
5620 if (gen != OSD_OII_NOGEN)
5624 * The inode may has been reused by others, we do not know,
5625 * leave it to be handled by subsequent osd_fid_lookup().
5628 } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5635 if (thread_is_running(&scrub->os_thread)) {
5636 if (inode == NULL) {
5637 inode = osd_iget(oti, dev, id);
5638 /* The inode has been removed (by race maybe). */
5639 if (IS_ERR(inode)) {
5640 rc = PTR_ERR(inode);
5642 RETURN(rc == -ESTALE ? -ENOENT : rc);
5646 rc = osd_oii_insert(dev, oic, insert);
5648 * There is race condition between osd_oi_lookup and OI scrub.
5649 * The OI scrub finished just after osd_oi_lookup() failure.
5650 * Under such case, it is unnecessary to trigger OI scrub again,
5651 * but try to call osd_oi_lookup() again.
5653 if (unlikely(rc == -EAGAIN))
5656 if (!S_ISDIR(inode->i_mode))
5659 rc = osd_check_lmv(oti, dev, inode, oic);
5664 if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5665 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5666 SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5667 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5668 "%s: trigger partial OI scrub for RPC inconsistency "
5669 "checking FID "DFID": rc = %d\n",
5670 osd_dev2name(dev), PFID(fid), rc);
5671 if (rc == 0 || rc == -EALREADY)
5684 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5685 struct osd_device *dev,
5686 struct osd_idmap_cache *oic,
5687 struct lu_fid *fid, __u32 ino)
5689 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5690 struct inode *inode;
5693 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5694 inode = osd_iget(oti, dev, &oic->oic_lid);
5695 if (IS_ERR(inode)) {
5696 fid_zero(&oic->oic_fid);
5697 return PTR_ERR(inode);
5700 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5703 fid_zero(&oic->oic_fid);
5705 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5709 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5710 struct osd_inode_id *id, const struct lu_fid *fid)
5712 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5713 id->oii_ino, id->oii_gen, info);
5714 info->oti_cache.oic_lid = *id;
5715 info->oti_cache.oic_fid = *fid;
5716 info->oti_cache.oic_dev = osd;
5720 * Get parent FID from the linkEA.
5722 * For a directory which parent resides on remote MDT, to satisfy the
5723 * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5724 * the other hand, to make the lookup(..) on the directory can return
5725 * the real parent FID, we append the real parent FID after its ".."
5726 * name entry in the /REMOTE_PARENT_DIR.
5728 * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5729 * backup. So after the restore, we cannot get the right parent FID from
5730 * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5731 * we have stored the real parent FID in the directory object's linkEA,
5732 * we can parse the linkEA for the real parent FID.
5734 * \param[in] env pointer to the thread context
5735 * \param[in] obj pointer to the object to be handled
5736 * \param[out]fid pointer to the buffer to hold the parent FID
5738 * \retval 0 for getting the real parent FID successfully
5739 * \retval negative error number on failure
5741 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5742 struct osd_object *obj,
5745 struct osd_thread_info *oti = osd_oti_get(env);
5746 struct lu_buf *buf = &oti->oti_big_buf;
5747 struct dentry *dentry = &oti->oti_obj_dentry;
5748 struct inode *inode = obj->oo_inode;
5749 struct linkea_data ldata = { NULL };
5755 if (!S_ISDIR(inode->i_mode))
5759 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5760 buf->lb_buf, buf->lb_len);
5761 if (rc == -ERANGE) {
5762 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5765 lu_buf_realloc(buf, rc);
5766 if (buf->lb_buf == NULL)
5773 if (unlikely(rc == 0))
5779 if (unlikely(buf->lb_buf == NULL)) {
5780 lu_buf_realloc(buf, rc);
5781 if (buf->lb_buf == NULL)
5788 rc = linkea_init_with_rec(&ldata);
5790 linkea_first_entry(&ldata);
5791 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5797 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5798 struct inode *inode,
5799 const struct lu_fid *pfid,
5800 const char *name, const int namelen)
5802 struct osd_thread_info *oti = osd_oti_get(env);
5803 struct lu_buf *buf = &oti->oti_big_buf;
5804 struct dentry *dentry = &oti->oti_obj_dentry;
5805 struct linkea_data ldata = { NULL };
5806 struct lu_name cname = { .ln_name = name,
5807 .ln_namelen = namelen };
5813 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5814 buf->lb_buf, buf->lb_len);
5816 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5821 if (unlikely(rc == 0))
5824 if (buf->lb_len < rc) {
5825 lu_buf_realloc(buf, rc);
5826 if (buf->lb_buf == NULL)
5833 rc = linkea_init_with_rec(&ldata);
5835 rc = linkea_links_find(&ldata, &cname, pfid);
5841 * Calls ->lookup() to find dentry. From dentry get inode and
5842 * read inode's ea to get fid. This is required for interoperability
5845 * \retval 0, on success
5846 * \retval -ve, on error
5848 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5849 struct dt_rec *rec, const struct dt_key *key)
5851 struct inode *dir = obj->oo_inode;
5852 struct dentry *dentry;
5853 struct ldiskfs_dir_entry_2 *de;
5854 struct buffer_head *bh;
5855 struct lu_fid *fid = (struct lu_fid *)rec;
5856 struct htree_lock *hlock = NULL;
5862 LASSERT(dir->i_op != NULL);
5863 LASSERT(dir->i_op->lookup != NULL);
5865 dentry = osd_child_dentry_get(env, obj,
5866 (char *)key, strlen((char *)key));
5868 if (obj->oo_hl_head != NULL) {
5869 hlock = osd_oti_get(env)->oti_hlock;
5870 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5871 dir, LDISKFS_HLOCK_LOOKUP);
5873 down_read(&obj->oo_ext_idx_sem);
5876 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5878 struct osd_thread_info *oti = osd_oti_get(env);
5879 struct osd_inode_id *id = &oti->oti_id;
5880 struct osd_idmap_cache *oic = &oti->oti_cache;
5881 struct osd_device *dev = osd_obj2dev(obj);
5883 ino = le32_to_cpu(de->inode);
5884 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5886 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5890 rc = osd_get_fid_from_dentry(de, rec);
5892 /* done with de, release bh */
5895 if (unlikely(is_remote_parent_ino(dev, ino))) {
5896 const char *name = (const char *)key;
5899 * If the parent is on remote MDT, and there
5900 * is no FID-in-dirent, then we have to get
5901 * the parent FID from the linkEA.
5903 if (likely(strlen(name) == 2 &&
5904 name[0] == '.' && name[1] == '.'))
5905 rc = osd_get_pfid_from_linkea(env, obj,
5908 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5911 osd_id_gen(id, ino, OSD_OII_NOGEN);
5914 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5915 fid_zero(&oic->oic_fid);
5920 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5921 rc = osd_consistency_check(oti, dev, oic);
5923 fid_zero(&oic->oic_fid);
5925 /* Other error should not affect lookup result. */
5935 ldiskfs_htree_unlock(hlock);
5937 up_read(&obj->oo_ext_idx_sem);
5941 static int osd_index_declare_ea_insert(const struct lu_env *env,
5942 struct dt_object *dt,
5943 const struct dt_rec *rec,
5944 const struct dt_key *key,
5945 struct thandle *handle)
5947 struct osd_thandle *oh;
5948 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5949 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5950 const struct lu_fid *fid = rec1->rec_fid;
5951 int credits, rc = 0;
5952 struct osd_idmap_cache *idc;
5956 LASSERT(!dt_object_remote(dt));
5957 LASSERT(handle != NULL);
5958 LASSERT(fid != NULL);
5959 LASSERT(rec1->rec_type != 0);
5961 oh = container_of0(handle, struct osd_thandle, ot_super);
5962 LASSERT(oh->ot_handle == NULL);
5964 credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5967 * we can't call iget() while a transactions is running
5968 * (this can lead to a deadlock), but we need to know
5969 * inum and object type. so we find this information at
5970 * declaration and cache in per-thread info
5972 idc = osd_idc_find_or_init(env, osd, fid);
5974 RETURN(PTR_ERR(idc));
5975 if (idc->oic_remote) {
5977 * a reference to remote inode is represented by an
5978 * agent inode which we have to create
5980 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5981 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5984 osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5986 if (osd_dt_obj(dt)->oo_inode != NULL) {
5987 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5990 * We ignore block quota on meta pool (MDTs), so needn't
5991 * calculate how many blocks will be consumed by this index
5994 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5996 i_projid_read(inode), 0,
5997 oh, osd_dt_obj(dt), NULL,
6002 #ifdef HAVE_PROJECT_QUOTA
6004 * Reserve credits for local agent inode to transfer
6005 * to 0, quota enforcement is ignored in this case.
6007 if (idc->oic_remote &&
6008 LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
6009 i_projid_read(inode) != 0)
6010 rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
6011 0, i_projid_read(inode),
6012 0, false, PRJQUOTA, true);
6020 * Index add function for interoperability mode (b11826).
6021 * It will add the directory entry.This entry is needed to
6022 * maintain name->fid mapping.
6024 * \param key it is key i.e. file entry to be inserted
6025 * \param rec it is value of given key i.e. fid
6027 * \retval 0, on success
6028 * \retval -ve, on error
6030 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6031 const struct dt_rec *rec,
6032 const struct dt_key *key, struct thandle *th)
6034 struct osd_object *obj = osd_dt_obj(dt);
6035 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6036 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6037 const struct lu_fid *fid = rec1->rec_fid;
6038 const char *name = (const char *)key;
6039 struct osd_thread_info *oti = osd_oti_get(env);
6040 struct inode *child_inode = NULL;
6041 struct osd_idmap_cache *idc;
6046 if (!dt_object_exists(dt))
6049 LASSERT(osd_invariant(obj));
6050 LASSERT(!dt_object_remote(dt));
6051 LASSERT(th != NULL);
6053 osd_trans_exec_op(env, th, OSD_OT_INSERT);
6055 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6057 idc = osd_idc_find(env, osd, fid);
6058 if (unlikely(idc == NULL)) {
6059 idc = osd_idc_find_or_init(env, osd, fid);
6062 * this dt_insert() wasn't declared properly, so
6063 * FID is missing in OI cache. we better do not
6064 * lookup FID in FLDB/OI and don't risk to deadlock,
6065 * but in some special cases (lfsck testing, etc)
6066 * it's much simpler than fixing a caller.
6068 * normally this error should be placed after the first
6069 * find, but migrate may attach source stripes to
6070 * target, which doesn't create stripes.
6072 CERROR("%s: "DFID" wasn't declared for insert\n",
6073 osd_name(osd), PFID(fid));
6075 RETURN(PTR_ERR(idc));
6079 if (idc->oic_remote) {
6080 /* Insert remote entry */
6081 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6083 igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6085 child_inode = osd_create_local_agent_inode(env, osd,
6086 obj, fid, rec1->rec_type & S_IFMT, th);
6087 if (IS_ERR(child_inode))
6088 RETURN(PTR_ERR(child_inode));
6091 /* Insert local entry */
6092 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6093 /* for a reason OI cache wasn't filled properly */
6094 CERROR("%s: OIC for "DFID" isn't filled\n",
6095 osd_name(osd), PFID(fid));
6098 child_inode = oti->oti_inode;
6099 if (unlikely(child_inode == NULL)) {
6100 struct ldiskfs_inode_info *lii;
6105 child_inode = oti->oti_inode = &lii->vfs_inode;
6107 child_inode->i_sb = osd_sb(osd);
6108 child_inode->i_ino = idc->oic_lid.oii_ino;
6109 child_inode->i_mode = rec1->rec_type & S_IFMT;
6112 rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6114 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6115 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6117 if (child_inode && child_inode != oti->oti_inode)
6119 LASSERT(osd_invariant(obj));
6120 osd_trans_exec_check(env, th, OSD_OT_INSERT);
6125 * Initialize osd Iterator for given osd index object.
6127 * \param dt osd index object
6130 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6131 struct dt_object *dt,
6134 struct osd_it_iam *it;
6135 struct osd_object *obj = osd_dt_obj(dt);
6136 struct lu_object *lo = &dt->do_lu;
6137 struct iam_path_descr *ipd;
6138 struct iam_container *bag = &obj->oo_dir->od_container;
6140 if (!dt_object_exists(dt))
6141 return ERR_PTR(-ENOENT);
6145 return ERR_PTR(-ENOMEM);
6147 ipd = osd_it_ipd_get(env, bag);
6148 if (likely(ipd != NULL)) {
6152 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6153 return (struct dt_it *)it;
6156 return ERR_PTR(-ENOMEM);
6161 * free given Iterator.
6163 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6165 struct osd_it_iam *it = (struct osd_it_iam *)di;
6166 struct osd_object *obj = it->oi_obj;
6168 iam_it_fini(&it->oi_it);
6169 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6170 osd_object_put(env, obj);
6175 * Move Iterator to record specified by \a key
6177 * \param di osd iterator
6178 * \param key key for index
6180 * \retval +ve di points to record with least key not larger than key
6181 * \retval 0 di points to exact matched key
6182 * \retval -ve failure
6185 static int osd_it_iam_get(const struct lu_env *env,
6186 struct dt_it *di, const struct dt_key *key)
6188 struct osd_thread_info *oti = osd_oti_get(env);
6189 struct osd_it_iam *it = (struct osd_it_iam *)di;
6191 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6192 /* swab quota uid/gid */
6193 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6194 key = (struct dt_key *)&oti->oti_quota_id;
6197 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6203 * \param di osd iterator
6205 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6207 struct osd_it_iam *it = (struct osd_it_iam *)di;
6209 iam_it_put(&it->oi_it);
6213 * Move iterator by one record
6215 * \param di osd iterator
6217 * \retval +1 end of container reached
6219 * \retval -ve failure
6222 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6224 struct osd_it_iam *it = (struct osd_it_iam *)di;
6226 return iam_it_next(&it->oi_it);
6230 * Return pointer to the key under iterator.
6233 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6234 const struct dt_it *di)
6236 struct osd_thread_info *oti = osd_oti_get(env);
6237 struct osd_it_iam *it = (struct osd_it_iam *)di;
6238 struct osd_object *obj = it->oi_obj;
6241 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6243 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6244 /* swab quota uid/gid */
6245 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6246 key = (struct dt_key *)&oti->oti_quota_id;
6253 * Return size of key under iterator (in bytes)
6256 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6258 struct osd_it_iam *it = (struct osd_it_iam *)di;
6260 return iam_it_key_size(&it->oi_it);
6264 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6266 /* check if file type is required */
6267 if (ent->lde_attrs & LUDA_TYPE) {
6268 struct luda_type *lt;
6269 int align = sizeof(*lt) - 1;
6271 len = (len + align) & ~align;
6272 lt = (struct luda_type *)(ent->lde_name + len);
6273 lt->lt_type = cpu_to_le16(DTTOIF(type));
6276 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6280 * build lu direct from backend fs dirent.
6284 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6285 char *name, __u16 namelen, __u16 type, __u32 attr)
6287 ent->lde_attrs = attr | LUDA_FID;
6288 fid_cpu_to_le(&ent->lde_fid, fid);
6290 ent->lde_hash = cpu_to_le64(offset);
6291 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6293 strncpy(ent->lde_name, name, namelen);
6294 ent->lde_name[namelen] = '\0';
6295 ent->lde_namelen = cpu_to_le16(namelen);
6297 /* append lustre attributes */
6298 osd_it_append_attrs(ent, namelen, type);
6302 * Return pointer to the record under iterator.
6304 static int osd_it_iam_rec(const struct lu_env *env,
6305 const struct dt_it *di,
6306 struct dt_rec *dtrec, __u32 attr)
6308 struct osd_it_iam *it = (struct osd_it_iam *)di;
6309 struct osd_thread_info *info = osd_oti_get(env);
6313 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6314 const struct osd_fid_pack *rec;
6315 struct lu_fid *fid = &info->oti_fid;
6316 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6322 name = (char *)iam_it_key_get(&it->oi_it);
6324 RETURN(PTR_ERR(name));
6326 namelen = iam_it_key_size(&it->oi_it);
6328 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6330 RETURN(PTR_ERR(rec));
6332 rc = osd_fid_unpack(fid, rec);
6336 hash = iam_it_store(&it->oi_it);
6338 /* IAM does not store object type in IAM index (dir) */
6339 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6341 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6342 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6343 (struct iam_rec *)dtrec);
6344 osd_quota_unpack(it->oi_obj, dtrec);
6346 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6347 (struct iam_rec *)dtrec);
6354 * Returns cookie for current Iterator position.
6356 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6358 struct osd_it_iam *it = (struct osd_it_iam *)di;
6360 return iam_it_store(&it->oi_it);
6364 * Restore iterator from cookie.
6366 * \param di osd iterator
6367 * \param hash Iterator location cookie
6369 * \retval +ve di points to record with least key not larger than key.
6370 * \retval 0 di points to exact matched key
6371 * \retval -ve failure
6374 static int osd_it_iam_load(const struct lu_env *env,
6375 const struct dt_it *di, __u64 hash)
6377 struct osd_it_iam *it = (struct osd_it_iam *)di;
6379 return iam_it_load(&it->oi_it, hash);
6382 static const struct dt_index_operations osd_index_iam_ops = {
6383 .dio_lookup = osd_index_iam_lookup,
6384 .dio_declare_insert = osd_index_declare_iam_insert,
6385 .dio_insert = osd_index_iam_insert,
6386 .dio_declare_delete = osd_index_declare_iam_delete,
6387 .dio_delete = osd_index_iam_delete,
6389 .init = osd_it_iam_init,
6390 .fini = osd_it_iam_fini,
6391 .get = osd_it_iam_get,
6392 .put = osd_it_iam_put,
6393 .next = osd_it_iam_next,
6394 .key = osd_it_iam_key,
6395 .key_size = osd_it_iam_key_size,
6396 .rec = osd_it_iam_rec,
6397 .store = osd_it_iam_store,
6398 .load = osd_it_iam_load
6404 * Creates or initializes iterator context.
6406 * \retval struct osd_it_ea, iterator structure on success
6409 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6410 struct dt_object *dt,
6413 struct osd_object *obj = osd_dt_obj(dt);
6414 struct osd_thread_info *info = osd_oti_get(env);
6415 struct osd_it_ea *oie;
6417 struct lu_object *lo = &dt->do_lu;
6418 struct dentry *obj_dentry;
6422 if (!dt_object_exists(dt) || obj->oo_destroyed)
6423 RETURN(ERR_PTR(-ENOENT));
6425 OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6427 RETURN(ERR_PTR(-ENOMEM));
6428 obj_dentry = &oie->oie_dentry;
6430 obj_dentry->d_inode = obj->oo_inode;
6431 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6432 obj_dentry->d_name.hash = 0;
6434 oie->oie_rd_dirent = 0;
6435 oie->oie_it_dirent = 0;
6436 oie->oie_dirent = NULL;
6437 if (unlikely(!info->oti_it_ea_buf_used)) {
6438 oie->oie_buf = info->oti_it_ea_buf;
6439 info->oti_it_ea_buf_used = 1;
6441 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6442 if (oie->oie_buf == NULL)
6443 RETURN(ERR_PTR(-ENOMEM));
6447 file = &oie->oie_file;
6449 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6450 if (attr & LUDA_64BITHASH)
6451 file->f_mode = FMODE_64BITHASH;
6453 file->f_mode = FMODE_32BITHASH;
6454 file->f_path.dentry = obj_dentry;
6455 file->f_mapping = obj->oo_inode->i_mapping;
6456 file->f_op = obj->oo_inode->i_fop;
6457 set_file_inode(file, obj->oo_inode);
6460 RETURN((struct dt_it *)oie);
6464 * Destroy or finishes iterator context.
6466 * \param di iterator structure to be destroyed
6468 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6470 struct osd_thread_info *info = osd_oti_get(env);
6471 struct osd_it_ea *oie = (struct osd_it_ea *)di;
6472 struct osd_object *obj = oie->oie_obj;
6473 struct inode *inode = obj->oo_inode;
6476 oie->oie_file.f_op->release(inode, &oie->oie_file);
6477 osd_object_put(env, obj);
6478 if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6479 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6481 info->oti_it_ea_buf_used = 0;
6482 OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6487 * It position the iterator at given key, so that next lookup continues from
6488 * that key Or it is similar to dio_it->load() but based on a key,
6489 * rather than file position.
6491 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6494 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6496 static int osd_it_ea_get(const struct lu_env *env,
6497 struct dt_it *di, const struct dt_key *key)
6499 struct osd_it_ea *it = (struct osd_it_ea *)di;
6502 LASSERT(((const char *)key)[0] == '\0');
6503 it->oie_file.f_pos = 0;
6504 it->oie_rd_dirent = 0;
6505 it->oie_it_dirent = 0;
6506 it->oie_dirent = NULL;
6514 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6518 struct osd_filldir_cbs {
6519 struct dir_context ctx;
6520 struct osd_it_ea *it;
6523 * It is called internally by ->iterate*(). It fills the
6524 * iterator's in-memory data structure with required
6525 * information i.e. name, namelen, rec_size etc.
6527 * \param buf in which information to be filled in.
6528 * \param name name of the file in given dir
6530 * \retval 0 on success
6531 * \retval 1 on buffer full
6533 #ifdef HAVE_FILLDIR_USE_CTX
6534 static int osd_ldiskfs_filldir(struct dir_context *buf,
6536 static int osd_ldiskfs_filldir(void *buf,
6538 const char *name, int namelen,
6539 loff_t offset, __u64 ino, unsigned int d_type)
6541 struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6542 struct osd_object *obj = it->oie_obj;
6543 struct osd_it_ea_dirent *ent = it->oie_dirent;
6544 struct lu_fid *fid = &ent->oied_fid;
6545 struct osd_fid_pack *rec;
6549 /* this should never happen */
6550 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6551 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6555 if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6559 /* "." is just the object itself. */
6560 if (namelen == 1 && name[0] == '.') {
6561 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6562 } else if (d_type & LDISKFS_DIRENT_LUFID) {
6563 rec = (struct osd_fid_pack *)(name + namelen + 1);
6564 if (osd_fid_unpack(fid, rec) != 0)
6569 d_type &= ~LDISKFS_DIRENT_LUFID;
6571 /* NOT export local root. */
6572 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6573 ino = obj->oo_inode->i_ino;
6574 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6577 ent->oied_ino = ino;
6578 ent->oied_off = offset;
6579 ent->oied_namelen = namelen;
6580 ent->oied_type = d_type;
6582 memcpy(ent->oied_name, name, namelen);
6584 it->oie_rd_dirent++;
6585 it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6590 * Calls ->iterate*() to load a directory entry at a time
6591 * and stored it in iterator's in-memory data structure.
6593 * \param di iterator's in memory structure
6595 * \retval 0 on success
6596 * \retval -ve on error
6597 * \retval +1 reach the end of entry
6599 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6600 const struct dt_it *di)
6602 struct osd_it_ea *it = (struct osd_it_ea *)di;
6603 struct osd_object *obj = it->oie_obj;
6604 struct inode *inode = obj->oo_inode;
6605 struct htree_lock *hlock = NULL;
6606 struct file *filp = &it->oie_file;
6608 struct osd_filldir_cbs buf = {
6609 .ctx.actor = osd_ldiskfs_filldir,
6614 it->oie_dirent = it->oie_buf;
6615 it->oie_rd_dirent = 0;
6617 if (obj->oo_hl_head != NULL) {
6618 hlock = osd_oti_get(env)->oti_hlock;
6619 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6620 inode, LDISKFS_HLOCK_READDIR);
6622 down_read(&obj->oo_ext_idx_sem);
6625 filp->f_cred = current_cred();
6626 rc = osd_security_file_alloc(filp);
6630 filp->f_flags |= O_NOATIME;
6631 filp->f_mode |= FMODE_NONOTIFY;
6632 rc = iterate_dir(filp, &buf.ctx);
6637 ldiskfs_htree_unlock(hlock);
6639 up_read(&obj->oo_ext_idx_sem);
6641 if (it->oie_rd_dirent == 0) {
6643 * If it does not get any dirent, it means it has been reached
6644 * to the end of the dir
6646 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6650 it->oie_dirent = it->oie_buf;
6651 it->oie_it_dirent = 1;
6658 * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
6659 * to load a directory entry at a time and stored it in
6660 * iterator's in-memory data structure.
6662 * \param di iterator's in memory structure
6664 * \retval +ve iterator reached to end
6665 * \retval 0 iterator not reached to end
6666 * \retval -ve on error
6668 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6670 struct osd_it_ea *it = (struct osd_it_ea *)di;
6675 if (it->oie_it_dirent < it->oie_rd_dirent) {
6677 (void *)it->oie_dirent +
6678 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6679 it->oie_dirent->oied_namelen);
6680 it->oie_it_dirent++;
6683 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6686 rc = osd_ldiskfs_it_fill(env, di);
6693 * Returns the key at current position from iterator's in memory structure.
6695 * \param di iterator's in memory structure
6697 * \retval key i.e. struct dt_key on success
6699 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6700 const struct dt_it *di)
6702 struct osd_it_ea *it = (struct osd_it_ea *)di;
6704 return (struct dt_key *)it->oie_dirent->oied_name;
6708 * Returns key's size at current position from iterator's in memory structure.
6710 * \param di iterator's in memory structure
6712 * \retval key_size i.e. struct dt_key on success
6714 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6716 struct osd_it_ea *it = (struct osd_it_ea *)di;
6718 return it->oie_dirent->oied_namelen;
6721 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6723 if (LDISKFS_DIR_REC_LEN(de) >=
6724 __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6731 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6732 unsigned int blocksize, bool dotdot)
6735 return osd_dotdot_has_space(de);
6737 if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6738 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6745 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6746 handle_t *jh, struct dentry *dentry,
6747 const struct lu_fid *fid, struct buffer_head *bh,
6748 struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6751 struct inode *dir = dentry->d_parent->d_inode;
6752 struct inode *inode = dentry->d_inode;
6753 struct osd_fid_pack *rec;
6754 struct ldiskfs_dentry_param *ldp;
6755 int namelen = dentry->d_name.len;
6757 struct osd_thread_info *info = osd_oti_get(env);
6761 if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6764 /* There is enough space to hold the FID-in-dirent. */
6765 if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6766 rc = ldiskfs_journal_get_write_access(jh, bh);
6770 de->name[namelen] = 0;
6771 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6772 rec->fp_len = sizeof(struct lu_fid) + 1;
6773 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6774 de->file_type |= LDISKFS_DIRENT_LUFID;
6775 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6782 rc = ldiskfs_delete_entry(jh, dir, de, bh);
6786 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6787 osd_get_ldiskfs_dirent_param(ldp, fid);
6788 dentry->d_fsdata = (void *)ldp;
6789 dquot_initialize(dir);
6790 rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6792 * It is too bad, we cannot reinsert the name entry back.
6793 * That means we lose it!
6796 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6797 "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6798 osd_ino2name(inode),
6799 dir->i_ino, dir->i_generation, namelen,
6800 dentry->d_name.name, PFID(fid), rc);
6806 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6807 struct osd_it_ea *it, struct lu_fid *fid,
6808 struct osd_inode_id *id, __u32 *attr)
6810 struct osd_thread_info *info = osd_oti_get(env);
6811 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6812 struct osd_device *dev = osd_obj2dev(obj);
6813 struct super_block *sb = osd_sb(dev);
6814 const char *devname = osd_name(dev);
6815 struct osd_it_ea_dirent *ent = it->oie_dirent;
6816 struct inode *dir = obj->oo_inode;
6817 struct htree_lock *hlock = NULL;
6818 struct buffer_head *bh = NULL;
6819 handle_t *jh = NULL;
6820 struct ldiskfs_dir_entry_2 *de;
6821 struct dentry *dentry;
6822 struct inode *inode;
6823 const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6826 bool dotdot = false;
6831 if (ent->oied_name[0] == '.') {
6832 if (ent->oied_namelen == 1)
6835 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6839 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6840 inode = osd_iget(info, dev, id);
6841 if (IS_ERR(inode)) {
6842 rc = PTR_ERR(inode);
6843 if (rc == -ENOENT || rc == -ESTALE) {
6845 * Maybe dangling name entry, or
6846 * corrupted directory entry.
6848 *attr |= LUDA_UNKNOWN;
6851 CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6852 "check_repair, dir = %lu/%u, name = %.*s, "
6853 "ino = %llu, rc = %d\n",
6854 devname, dir->i_ino, dir->i_generation,
6855 ent->oied_namelen, ent->oied_name,
6862 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6864 rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6865 if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6871 * We need to ensure that the name entry is still valid.
6872 * Because it may be removed or renamed by other already.
6874 * The unlink or rename operation will start journal before PDO lock,
6875 * so to avoid deadlock, here we need to start journal handle before
6876 * related PDO lock also. But because we do not know whether there
6877 * will be something to be repaired before PDO lock, we just start
6878 * journal without conditions.
6880 * We may need to remove the name entry firstly, then insert back.
6881 * One credit is for user quota file update.
6882 * One credit is for group quota file update.
6883 * Two credits are for dirty inode.
6885 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6886 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6888 if (dev->od_dirent_journal != 0) {
6891 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6894 CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6895 "check_repair, dir = %lu/%u, credits = %d, "
6896 "name = %.*s, ino = %llu: rc = %d\n",
6897 devname, dir->i_ino, dir->i_generation, credits,
6898 ent->oied_namelen, ent->oied_name,
6901 GOTO(out_inode, rc);
6904 if (obj->oo_hl_head != NULL) {
6905 hlock = osd_oti_get(env)->oti_hlock;
6907 * "0" means exclusive lock for the whole directory.
6908 * We need to prevent others access such name entry
6909 * during the delete + insert. Neither HLOCK_ADD nor
6910 * HLOCK_DEL cannot guarantee the atomicity.
6912 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6914 down_write(&obj->oo_ext_idx_sem);
6917 if (obj->oo_hl_head != NULL) {
6918 hlock = osd_oti_get(env)->oti_hlock;
6919 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6920 LDISKFS_HLOCK_LOOKUP);
6922 down_read(&obj->oo_ext_idx_sem);
6926 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6927 if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6928 *attr |= LUDA_IGNORE;
6934 * For dotdot entry, if there is not enough space to hold the
6935 * FID-in-dirent, just keep them there. It only happens when the
6936 * device upgraded from 1.8 or restored from MDT file-level backup.
6937 * For the whole directory, only dotdot entry have no FID-in-dirent
6938 * and needs to get FID from LMA when readdir, it will not affect the
6941 if (dotdot && !osd_dotdot_has_space(de)) {
6942 *attr |= LUDA_UNKNOWN;
6948 if (lu_fid_eq(fid, &lma->lma_self_fid))
6951 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6952 struct lu_fid *tfid = &lma->lma_self_fid;
6954 if (likely(dotdot &&
6955 fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6956 fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6958 * It must be REMOTE_PARENT_DIR and as the
6959 * 'dotdot' entry of remote directory
6961 *attr |= LUDA_IGNORE;
6963 CDEBUG(D_LFSCK, "%s: expect remote agent "
6964 "parent directory, but got %.*s under "
6965 "dir = %lu/%u with the FID "DFID"\n",
6966 devname, ent->oied_namelen,
6967 ent->oied_name, dir->i_ino,
6968 dir->i_generation, PFID(tfid));
6970 *attr |= LUDA_UNKNOWN;
6977 if (!fid_is_zero(fid)) {
6978 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6980 if (rc == -ENOENT ||
6982 !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
6984 * linkEA does not recognize the dirent entry,
6985 * it may because the dirent entry corruption
6986 * and points to other's inode.
6988 CDEBUG(D_LFSCK, "%s: the target inode does not "
6989 "recognize the dirent, dir = %lu/%u, "
6990 " name = %.*s, ino = %llu, "
6991 DFID": rc = %d\n", devname, dir->i_ino,
6992 dir->i_generation, ent->oied_namelen,
6993 ent->oied_name, ent->oied_ino, PFID(fid), rc);
6994 *attr |= LUDA_UNKNOWN;
6999 if (rc && rc != -ENODATA) {
7000 CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
7001 "dir = %lu/%u, name = %.*s, ino = %llu, "
7002 DFID": rc = %d\n", devname, dir->i_ino,
7003 dir->i_generation, ent->oied_namelen,
7004 ent->oied_name, ent->oied_ino, PFID(fid), rc);
7005 *attr |= LUDA_UNKNOWN;
7013 * linkEA recognizes the dirent entry, the FID-in-LMA is
7014 * valid, trusted, in spite of fid_is_sane(fid) or not.
7016 if (*attr & LUDA_VERIFY_DRYRUN) {
7017 *fid = lma->lma_self_fid;
7018 *attr |= LUDA_REPAIR;
7025 dev->od_dirent_journal = 1;
7026 if (hlock != NULL) {
7027 ldiskfs_htree_unlock(hlock);
7030 up_read(&obj->oo_ext_idx_sem);
7036 *fid = lma->lma_self_fid;
7038 /* Update or append the FID-in-dirent. */
7039 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7040 bh, de, hlock, dotdot);
7042 *attr |= LUDA_REPAIR;
7044 CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7045 "the dirent, dir = %lu/%u, name = %.*s, "
7046 "ino = %llu, "DFID": rc = %d\n",
7047 devname, dir->i_ino, dir->i_generation,
7048 ent->oied_namelen, ent->oied_name,
7049 ent->oied_ino, PFID(fid), rc);
7051 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7052 if (*attr & LUDA_VERIFY_DRYRUN) {
7053 if (fid_is_sane(fid)) {
7054 *attr |= LUDA_REPAIR;
7055 } else if (dev->od_index == 0) {
7056 lu_igif_build(fid, inode->i_ino,
7057 inode->i_generation);
7058 *attr |= LUDA_UPGRADE;
7066 dev->od_dirent_journal = 1;
7067 if (hlock != NULL) {
7068 ldiskfs_htree_unlock(hlock);
7071 up_read(&obj->oo_ext_idx_sem);
7078 if (unlikely(fid_is_sane(fid))) {
7080 * FID-in-dirent exists, but FID-in-LMA is lost.
7081 * Trust the FID-in-dirent, and add FID-in-LMA.
7083 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7085 *attr |= LUDA_REPAIR;
7087 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7088 "update dirent, dir = %lu/%u, "
7089 "name = %.*s, ino = %llu, "
7091 devname, dir->i_ino, dir->i_generation,
7092 ent->oied_namelen, ent->oied_name,
7093 ent->oied_ino, PFID(fid), rc);
7094 } else if (dev->od_index == 0) {
7095 lu_igif_build(fid, inode->i_ino, inode->i_generation);
7097 * It is probably IGIF object. Only aappend the
7098 * FID-in-dirent. OI scrub will process FID-in-LMA.
7100 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7101 bh, de, hlock, dotdot);
7103 *attr |= LUDA_UPGRADE;
7105 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7106 "after the dirent, dir = %lu/%u, "
7107 "name = %.*s, ino = %llu, "
7109 devname, dir->i_ino, dir->i_generation,
7110 ent->oied_namelen, ent->oied_name,
7111 ent->oied_ino, PFID(fid), rc);
7120 if (hlock != NULL) {
7121 ldiskfs_htree_unlock(hlock);
7123 if (dev->od_dirent_journal != 0)
7124 up_write(&obj->oo_ext_idx_sem);
7126 up_read(&obj->oo_ext_idx_sem);
7130 ldiskfs_journal_stop(jh);
7134 if (rc >= 0 && !dirty)
7135 dev->od_dirent_journal = 0;
7141 * Returns the value at current position from iterator's in memory structure.
7143 * \param di struct osd_it_ea, iterator's in memory structure
7144 * \param attr attr requested for dirent.
7145 * \param lde lustre dirent
7147 * \retval 0 no error and \param lde has correct lustre dirent.
7148 * \retval -ve on error
7150 static inline int osd_it_ea_rec(const struct lu_env *env,
7151 const struct dt_it *di,
7152 struct dt_rec *dtrec, __u32 attr)
7154 struct osd_it_ea *it = (struct osd_it_ea *)di;
7155 struct osd_object *obj = it->oie_obj;
7156 struct osd_device *dev = osd_obj2dev(obj);
7157 struct osd_thread_info *oti = osd_oti_get(env);
7158 struct osd_inode_id *id = &oti->oti_id;
7159 struct lu_fid *fid = &it->oie_dirent->oied_fid;
7160 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
7161 __u32 ino = it->oie_dirent->oied_ino;
7166 LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7168 if (attr & LUDA_VERIFY) {
7169 if (unlikely(is_remote_parent_ino(dev, ino))) {
7170 attr |= LUDA_IGNORE;
7172 * If the parent is on remote MDT, and there
7173 * is no FID-in-dirent, then we have to get
7174 * the parent FID from the linkEA.
7176 if (!fid_is_sane(fid) &&
7177 it->oie_dirent->oied_namelen == 2 &&
7178 it->oie_dirent->oied_name[0] == '.' &&
7179 it->oie_dirent->oied_name[1] == '.')
7180 osd_get_pfid_from_linkea(env, obj, fid);
7182 rc = osd_dirent_check_repair(env, obj, it, fid, id,
7186 if (!fid_is_sane(fid))
7187 attr |= LUDA_UNKNOWN;
7189 attr &= ~LU_DIRENT_ATTRS_MASK;
7190 if (!fid_is_sane(fid)) {
7191 bool is_dotdot = false;
7193 if (it->oie_dirent->oied_namelen == 2 &&
7194 it->oie_dirent->oied_name[0] == '.' &&
7195 it->oie_dirent->oied_name[1] == '.')
7198 * If the parent is on remote MDT, and there
7199 * is no FID-in-dirent, then we have to get
7200 * the parent FID from the linkEA.
7202 if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7203 rc = osd_get_pfid_from_linkea(env, obj, fid);
7205 if (is_dotdot == false &&
7206 OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7209 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7212 osd_id_gen(id, ino, OSD_OII_NOGEN);
7216 /* Pack the entry anyway, at least the offset is right. */
7217 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7218 it->oie_dirent->oied_name,
7219 it->oie_dirent->oied_namelen,
7220 it->oie_dirent->oied_type, attr);
7225 if (osd_remote_fid(env, dev, fid))
7228 if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7229 osd_add_oi_cache(oti, dev, id, fid);
7231 RETURN(rc > 0 ? 0 : rc);
7235 * Returns the record size size at current position.
7237 * This function will return record(lu_dirent) size in bytes.
7239 * \param[in] env execution environment
7240 * \param[in] di iterator's in memory structure
7241 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
7242 * calculate the lu_dirent size.
7244 * \retval record size(in bytes & in memory) of the current lu_dirent
7247 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7250 struct osd_it_ea *it = (struct osd_it_ea *)di;
7252 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7256 * Returns a cookie for current position of the iterator head, so that
7257 * user can use this cookie to load/start the iterator next time.
7259 * \param di iterator's in memory structure
7261 * \retval cookie for current position, on success
7263 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7265 struct osd_it_ea *it = (struct osd_it_ea *)di;
7267 return it->oie_dirent->oied_off;
7271 * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
7272 * to load a directory entry at a time and stored it i inn,
7273 * in iterator's in-memory data structure.
7275 * \param di struct osd_it_ea, iterator's in memory structure
7277 * \retval +ve on success
7278 * \retval -ve on error
7280 static int osd_it_ea_load(const struct lu_env *env,
7281 const struct dt_it *di, __u64 hash)
7283 struct osd_it_ea *it = (struct osd_it_ea *)di;
7287 it->oie_file.f_pos = hash;
7289 rc = osd_ldiskfs_it_fill(env, di);
7300 * Index lookup function for interoperability mode (b11826).
7302 * \param key, key i.e. file name to be searched
7304 * \retval +ve, on success
7305 * \retval -ve, on error
7307 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7308 struct dt_rec *rec, const struct dt_key *key)
7310 struct osd_object *obj = osd_dt_obj(dt);
7315 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7316 LINVRNT(osd_invariant(obj));
7318 rc = osd_ea_lookup_rec(env, obj, rec, key);
7325 * Index and Iterator operations for interoperability
7326 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7328 static const struct dt_index_operations osd_index_ea_ops = {
7329 .dio_lookup = osd_index_ea_lookup,
7330 .dio_declare_insert = osd_index_declare_ea_insert,
7331 .dio_insert = osd_index_ea_insert,
7332 .dio_declare_delete = osd_index_declare_ea_delete,
7333 .dio_delete = osd_index_ea_delete,
7335 .init = osd_it_ea_init,
7336 .fini = osd_it_ea_fini,
7337 .get = osd_it_ea_get,
7338 .put = osd_it_ea_put,
7339 .next = osd_it_ea_next,
7340 .key = osd_it_ea_key,
7341 .key_size = osd_it_ea_key_size,
7342 .rec = osd_it_ea_rec,
7343 .rec_size = osd_it_ea_rec_size,
7344 .store = osd_it_ea_store,
7345 .load = osd_it_ea_load
7349 static void *osd_key_init(const struct lu_context *ctx,
7350 struct lu_context_key *key)
7352 struct osd_thread_info *info;
7354 OBD_ALLOC_PTR(info);
7356 return ERR_PTR(-ENOMEM);
7358 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7359 if (info->oti_it_ea_buf == NULL)
7362 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7364 info->oti_hlock = ldiskfs_htree_lock_alloc();
7365 if (info->oti_hlock == NULL)
7371 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7374 return ERR_PTR(-ENOMEM);
7377 static void osd_key_fini(const struct lu_context *ctx,
7378 struct lu_context_key *key, void *data)
7380 struct osd_thread_info *info = data;
7381 struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7382 struct osd_idmap_cache *idc = info->oti_ins_cache;
7384 if (info->oti_dio_pages) {
7386 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7387 if (info->oti_dio_pages[i])
7388 __free_page(info->oti_dio_pages[i]);
7390 OBD_FREE(info->oti_dio_pages,
7391 sizeof(struct page *) * PTLRPC_MAX_BRW_PAGES);
7394 if (info->oti_inode != NULL)
7396 if (info->oti_hlock != NULL)
7397 ldiskfs_htree_lock_free(info->oti_hlock);
7398 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7399 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7400 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7401 lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7402 lu_buf_free(&info->oti_big_buf);
7404 LASSERT(info->oti_ins_cache_size > 0);
7405 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
7406 info->oti_ins_cache = NULL;
7407 info->oti_ins_cache_size = 0;
7412 static void osd_key_exit(const struct lu_context *ctx,
7413 struct lu_context_key *key, void *data)
7415 struct osd_thread_info *info = data;
7417 LASSERT(info->oti_r_locks == 0);
7418 LASSERT(info->oti_w_locks == 0);
7419 LASSERT(info->oti_txns == 0);
7422 /* type constructor/destructor: osd_type_init, osd_type_fini */
7423 LU_TYPE_INIT_FINI(osd, &osd_key);
7425 struct lu_context_key osd_key = {
7426 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7427 .lct_init = osd_key_init,
7428 .lct_fini = osd_key_fini,
7429 .lct_exit = osd_key_exit
7433 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7434 const char *name, struct lu_device *next)
7436 struct osd_device *osd = osd_dev(d);
7438 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7439 sizeof(osd->od_svname))
7441 return osd_procfs_init(osd, name);
7444 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7446 struct seq_server_site *ss = osd_seq_site(osd);
7451 if (osd->od_is_ost || osd->od_cl_seq != NULL)
7454 if (unlikely(ss == NULL))
7457 OBD_ALLOC_PTR(osd->od_cl_seq);
7458 if (osd->od_cl_seq == NULL)
7461 rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7462 osd->od_svname, ss->ss_server_seq);
7464 OBD_FREE_PTR(osd->od_cl_seq);
7465 osd->od_cl_seq = NULL;
7469 if (ss->ss_node_id == 0) {
7471 * If the OSD on the sequence controller(MDT0), then allocate
7472 * sequence here, otherwise allocate sequence after connected
7473 * to MDT0 (see mdt_register_lwp_callback()).
7475 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7476 &osd->od_cl_seq->lcs_space, env);
7482 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7484 if (osd->od_cl_seq == NULL)
7487 seq_client_fini(osd->od_cl_seq);
7488 OBD_FREE_PTR(osd->od_cl_seq);
7489 osd->od_cl_seq = NULL;
7492 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7496 /* shutdown quota slave instance associated with the device */
7497 if (o->od_quota_slave_md != NULL) {
7498 struct qsd_instance *qsd = o->od_quota_slave_md;
7500 o->od_quota_slave_md = NULL;
7504 if (o->od_quota_slave_dt != NULL) {
7505 struct qsd_instance *qsd = o->od_quota_slave_dt;
7507 o->od_quota_slave_dt = NULL;
7511 osd_fid_fini(env, o);
7512 osd_scrub_cleanup(env, o);
7517 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7521 if (o->od_mnt != NULL) {
7522 shrink_dcache_sb(osd_sb(o));
7523 osd_sync(env, &o->od_dt_dev);
7532 static int osd_mount(const struct lu_env *env,
7533 struct osd_device *o, struct lustre_cfg *cfg)
7535 const char *name = lustre_cfg_string(cfg, 0);
7536 const char *dev = lustre_cfg_string(cfg, 1);
7538 unsigned long page, s_flags, lmd_flags = 0;
7539 struct page *__page;
7540 struct file_system_type *type;
7541 char *options = NULL;
7543 struct osd_thread_info *info = osd_oti_get(env);
7544 struct lu_fid *fid = &info->oti_fid;
7545 struct inode *inode;
7546 int rc = 0, force_over_512tb = 0;
7550 if (o->od_mnt != NULL)
7553 if (strlen(dev) >= sizeof(o->od_mntdev))
7555 strcpy(o->od_mntdev, dev);
7557 str = lustre_cfg_string(cfg, 2);
7558 s_flags = simple_strtoul(str, NULL, 0);
7559 str = strstr(str, ":");
7561 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7562 opts = lustre_cfg_string(cfg, 3);
7564 if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7565 CERROR("%s: device %s extents feature is not guaranteed to "
7566 "work on big-endian systems. Use \"bigendian_extents\" "
7567 "mount option to override.\n", name, dev);
7571 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7572 if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7573 CWARN("force_over_128tb option is deprecated. "
7574 "Filesystems less than 512TB can be created without any "
7575 "force options. Use force_over_512tb option for "
7576 "filesystems greater than 512TB.\n");
7579 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7580 if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7581 CWARN("force_over_256tb option is deprecated. "
7582 "Filesystems less than 512TB can be created without any "
7583 "force options. Use force_over_512tb option for "
7584 "filesystems greater than 512TB.\n");
7588 if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
7589 force_over_512tb = 1;
7591 __page = alloc_page(GFP_KERNEL);
7593 GOTO(out, rc = -ENOMEM);
7594 page = (unsigned long)page_address(__page);
7595 options = (char *)page;
7598 /* strip out the options for back compatiblity */
7599 static char *sout[] = {
7606 /* strip out option we processed in osd */
7607 "bigendian_extents",
7613 strncat(options, opts, PAGE_SIZE);
7614 for (rc = 0, str = options; sout[rc]; ) {
7615 char *op = strstr(str, sout[rc]);
7622 if (op == options || *(op - 1) == ',') {
7623 str = op + strlen(sout[rc]);
7624 if (*str == ',' || *str == '\0') {
7625 *str == ',' ? str++ : str;
7626 memmove(op, str, strlen(str) + 1);
7629 for (str = op; *str != ',' && *str != '\0'; str++)
7633 strncat(options, "user_xattr,acl", PAGE_SIZE);
7636 /* Glom up mount options */
7637 if (*options != '\0')
7638 strncat(options, ",", PAGE_SIZE);
7639 strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7641 type = get_fs_type("ldiskfs");
7643 CERROR("%s: cannot find ldiskfs module\n", name);
7644 GOTO(out, rc = -ENODEV);
7647 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7648 module_put(type->owner);
7650 if (IS_ERR(o->od_mnt)) {
7651 rc = PTR_ERR(o->od_mnt);
7653 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7657 if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7658 osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7659 force_over_512tb == 0) {
7660 CERROR("%s: device %s LDISKFS does not support filesystems "
7661 "greater than 512TB and can cause data corruption. "
7662 "Use \"force_over_512tb\" mount option to override.\n",
7664 GOTO(out_mnt, rc = -EINVAL);
7667 if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7668 if (priv_dev_set_rdonly) {
7669 priv_dev_set_rdonly(osd_sb(o)->s_bdev);
7670 o->od_dt_dev.dd_rdonly = 1;
7671 LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
7674 LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7677 GOTO(out_mnt, rc = -EOPNOTSUPP);
7679 } else if (priv_dev_check_rdonly &&
7680 priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7681 CERROR("%s: underlying device %s is marked as "
7682 "read-only. Setup failed\n", name, dev);
7684 GOTO(out_mnt, rc = -EROFS);
7687 if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7688 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7689 GOTO(out_mnt, rc = -EINVAL);
7692 #ifdef LDISKFS_MOUNT_DIRDATA
7693 if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7694 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7695 else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7696 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7697 "enabling the dirdata feature. If you do not want to "
7698 "downgrade to Lustre-1.x again, you can enable it via "
7699 "'tune2fs -O dirdata device'\n", name, dev);
7701 inode = osd_sb(o)->s_root->d_inode;
7702 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7703 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7705 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7709 if (lmd_flags & LMD_FLG_NOSCRUB)
7710 o->od_auto_scrub_interval = AS_NEVER;
7712 if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7713 /* do not use pagecache with flash-backed storage */
7714 o->od_writethrough_cache = 0;
7715 o->od_read_cache = 0;
7726 __free_page(__page);
7731 static struct lu_device *osd_device_fini(const struct lu_env *env,
7732 struct lu_device *d)
7734 struct osd_device *o = osd_dev(d);
7738 osd_index_backup(env, o, false);
7739 osd_shutdown(env, o);
7741 osd_obj_map_fini(o);
7747 static int osd_device_init0(const struct lu_env *env,
7748 struct osd_device *o,
7749 struct lustre_cfg *cfg)
7751 struct lu_device *l = osd2lu_dev(o);
7752 struct osd_thread_info *info;
7756 /* if the module was re-loaded, env can loose its keys */
7757 rc = lu_env_refill((struct lu_env *)env);
7760 info = osd_oti_get(env);
7763 l->ld_ops = &osd_lu_ops;
7764 o->od_dt_dev.dd_ops = &osd_dt_ops;
7766 spin_lock_init(&o->od_osfs_lock);
7767 mutex_init(&o->od_otable_mutex);
7768 INIT_LIST_HEAD(&o->od_orphan_list);
7769 INIT_LIST_HEAD(&o->od_index_backup_list);
7770 INIT_LIST_HEAD(&o->od_index_restore_list);
7771 spin_lock_init(&o->od_lock);
7772 o->od_index_backup_policy = LIBP_NONE;
7775 o->od_read_cache = 1;
7776 o->od_writethrough_cache = 1;
7777 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7778 o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20;
7779 o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20;
7780 o->od_auto_scrub_interval = AS_DEFAULT;
7782 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7783 sizeof(o->od_svname));
7784 if (cplen >= sizeof(o->od_svname)) {
7789 o->od_index_backup_stop = 0;
7790 o->od_index = -1; /* -1 means index is invalid */
7791 rc = server_name2index(o->od_svname, &o->od_index, NULL);
7792 if (rc == LDD_F_SV_TYPE_OST)
7795 o->od_full_scrub_ratio = OFSR_DEFAULT;
7796 o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7797 rc = osd_mount(env, o, cfg);
7801 /* Can only check block device after mount */
7802 o->od_nonrotational =
7803 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
7805 rc = osd_obj_map_init(env, o);
7809 rc = lu_site_init(&o->od_site, l);
7811 GOTO(out_compat, rc);
7812 o->od_site.ls_bottom_dev = l;
7814 rc = lu_site_init_finish(&o->od_site);
7818 INIT_LIST_HEAD(&o->od_ios_list);
7819 /* setup scrub, including OI files initialization */
7821 rc = osd_scrub_setup(env, o);
7826 rc = osd_procfs_init(o, o->od_svname);
7828 CERROR("%s: can't initialize procfs: rc = %d\n",
7830 GOTO(out_scrub, rc);
7833 LASSERT(l->ld_site->ls_linkage.next != NULL);
7834 LASSERT(l->ld_site->ls_linkage.prev != NULL);
7836 /* initialize quota slave instance */
7837 /* currently it's no need to prepare qsd_instance_md for OST */
7838 if (!o->od_is_ost) {
7839 o->od_quota_slave_md = qsd_init(env, o->od_svname,
7841 o->od_proc_entry, true);
7842 if (IS_ERR(o->od_quota_slave_md)) {
7843 rc = PTR_ERR(o->od_quota_slave_md);
7844 o->od_quota_slave_md = NULL;
7845 GOTO(out_procfs, rc);
7849 o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
7850 o->od_proc_entry, false);
7852 if (IS_ERR(o->od_quota_slave_dt)) {
7853 if (o->od_quota_slave_md != NULL) {
7854 qsd_fini(env, o->od_quota_slave_md);
7855 o->od_quota_slave_md = NULL;
7858 rc = PTR_ERR(o->od_quota_slave_dt);
7859 o->od_quota_slave_dt = NULL;
7860 GOTO(out_procfs, rc);
7868 osd_scrub_cleanup(env, o);
7870 lu_site_fini(&o->od_site);
7872 osd_obj_map_fini(o);
7879 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7880 struct lu_device_type *t,
7881 struct lustre_cfg *cfg)
7883 struct osd_device *o;
7888 return ERR_PTR(-ENOMEM);
7890 rc = dt_device_init(&o->od_dt_dev, t);
7893 * Because the ctx might be revived in dt_device_init,
7894 * refill the env here
7896 lu_env_refill((struct lu_env *)env);
7897 rc = osd_device_init0(env, o, cfg);
7899 dt_device_fini(&o->od_dt_dev);
7902 if (unlikely(rc != 0))
7905 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7908 static struct lu_device *osd_device_free(const struct lu_env *env,
7909 struct lu_device *d)
7911 struct osd_device *o = osd_dev(d);
7915 /* XXX: make osd top device in order to release reference */
7916 d->ld_site->ls_top_dev = d;
7917 lu_site_purge(env, d->ld_site, -1);
7918 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7919 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7920 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7922 lu_site_fini(&o->od_site);
7923 dt_device_fini(&o->od_dt_dev);
7928 static int osd_process_config(const struct lu_env *env,
7929 struct lu_device *d, struct lustre_cfg *cfg)
7931 struct osd_device *o = osd_dev(d);
7937 switch (cfg->lcfg_command) {
7939 rc = osd_mount(env, o, cfg);
7943 * For the case LCFG_PRE_CLEANUP is not called in advance,
7944 * that may happend if hit failure during mount process.
7946 osd_index_backup(env, o, false);
7947 lu_dev_del_linkage(d->ld_site, d);
7948 rc = osd_shutdown(env, o);
7951 LASSERT(&o->od_dt_dev);
7952 count = class_modify_config(cfg, PARAM_OSD,
7953 &o->od_dt_dev.dd_kobj);
7955 count = class_modify_config(cfg, PARAM_OST,
7956 &o->od_dt_dev.dd_kobj);
7957 rc = count > 0 ? 0 : count;
7959 case LCFG_PRE_CLEANUP:
7961 osd_index_backup(env, o,
7962 o->od_index_backup_policy != LIBP_NONE);
7972 static int osd_recovery_complete(const struct lu_env *env,
7973 struct lu_device *d)
7975 struct osd_device *osd = osd_dev(d);
7980 if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
7984 * start qsd instance on recovery completion, this notifies the quota
7985 * slave code that we are about to process new requests now
7987 rc = qsd_start(env, osd->od_quota_slave_dt);
7988 if (rc == 0 && osd->od_quota_slave_md != NULL)
7989 rc = qsd_start(env, osd->od_quota_slave_md);
7995 * we use exports to track all osd users
7997 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7998 struct obd_device *obd, struct obd_uuid *cluuid,
7999 struct obd_connect_data *data, void *localdata)
8001 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8002 struct lustre_handle conn;
8007 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
8009 rc = class_connect(&conn, obd, cluuid);
8013 *exp = class_conn2export(&conn);
8015 spin_lock(&osd->od_osfs_lock);
8017 spin_unlock(&osd->od_osfs_lock);
8023 * once last export (we don't count self-export) disappeared
8024 * osd can be released
8026 static int osd_obd_disconnect(struct obd_export *exp)
8028 struct obd_device *obd = exp->exp_obd;
8029 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8030 int rc, release = 0;
8034 /* Only disconnect the underlying layers on the final disconnect. */
8035 spin_lock(&osd->od_osfs_lock);
8037 if (osd->od_connects == 0)
8039 spin_unlock(&osd->od_osfs_lock);
8041 rc = class_disconnect(exp); /* bz 9811 */
8043 if (rc == 0 && release)
8044 class_manual_cleanup(obd);
8048 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8049 struct lu_device *dev)
8051 struct osd_device *osd = osd_dev(dev);
8052 struct lr_server_data *lsd =
8053 &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8058 if (osd->od_quota_slave_md != NULL) {
8059 /* set up quota slave objects for inode */
8060 result = qsd_prepare(env, osd->od_quota_slave_md);
8065 if (osd->od_quota_slave_dt != NULL) {
8066 /* set up quota slave objects for block */
8067 result = qsd_prepare(env, osd->od_quota_slave_dt);
8073 if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8074 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8075 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8076 osd->od_index_in_idif = 1;
8078 osd->od_index_in_idif = 0;
8079 result = osd_register_proc_index_in_idif(osd);
8084 osd->od_index_in_idif = 1;
8088 result = osd_fid_init(env, osd);
8093 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
8094 struct lu_fid *fid, struct md_op_data *op_data)
8096 struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
8098 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8101 static const struct lu_object_operations osd_lu_obj_ops = {
8102 .loo_object_init = osd_object_init,
8103 .loo_object_delete = osd_object_delete,
8104 .loo_object_release = osd_object_release,
8105 .loo_object_free = osd_object_free,
8106 .loo_object_print = osd_object_print,
8107 .loo_object_invariant = osd_object_invariant
8110 const struct lu_device_operations osd_lu_ops = {
8111 .ldo_object_alloc = osd_object_alloc,
8112 .ldo_process_config = osd_process_config,
8113 .ldo_recovery_complete = osd_recovery_complete,
8114 .ldo_prepare = osd_prepare,
8117 static const struct lu_device_type_operations osd_device_type_ops = {
8118 .ldto_init = osd_type_init,
8119 .ldto_fini = osd_type_fini,
8121 .ldto_start = osd_type_start,
8122 .ldto_stop = osd_type_stop,
8124 .ldto_device_alloc = osd_device_alloc,
8125 .ldto_device_free = osd_device_free,
8127 .ldto_device_init = osd_device_init,
8128 .ldto_device_fini = osd_device_fini
8131 static struct lu_device_type osd_device_type = {
8132 .ldt_tags = LU_DEVICE_DT,
8133 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
8134 .ldt_ops = &osd_device_type_ops,
8135 .ldt_ctx_tags = LCT_LOCAL,
8138 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8140 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8141 struct super_block *sb = osd_sb(osd);
8143 return (osd->od_mnt == NULL || sb->s_flags & SB_RDONLY);
8147 * lprocfs legacy support.
8149 static const struct obd_ops osd_obd_device_ops = {
8150 .o_owner = THIS_MODULE,
8151 .o_connect = osd_obd_connect,
8152 .o_disconnect = osd_obd_disconnect,
8153 .o_fid_alloc = osd_fid_alloc,
8154 .o_health_check = osd_health_check,
8157 static ssize_t track_declares_assert_show(struct kobject *kobj,
8158 struct attribute *attr,
8161 return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8164 static ssize_t track_declares_assert_store(struct kobject *kobj,
8165 struct attribute *attr,
8166 const char *buffer, size_t count)
8168 bool track_declares_assert;
8171 rc = kstrtobool(buffer, &track_declares_assert);
8175 ldiskfs_track_declares_assert = track_declares_assert;
8179 LUSTRE_RW_ATTR(track_declares_assert);
8181 static int __init osd_init(void)
8183 struct kobject *kobj;
8186 BUILD_BUG_ON(BH_DXLock >=
8187 sizeof(((struct buffer_head *)0)->b_state) * 8);
8188 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8189 /* please, try to keep osd_thread_info smaller than a page */
8190 BUILD_BUG_ON(sizeof(struct osd_thread_info) > PAGE_SIZE);
8195 rc = lu_kmem_init(ldiskfs_caches);
8199 #ifdef CONFIG_KALLSYMS
8200 priv_security_file_alloc =
8201 (void *)kallsyms_lookup_name("security_file_alloc");
8202 priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
8203 priv_dev_check_rdonly =
8204 (void *)kallsyms_lookup_name("dev_check_rdonly");
8207 rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL,
8208 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8210 lu_kmem_fini(ldiskfs_caches);
8214 kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8216 rc = sysfs_create_file(kobj,
8217 &lustre_attr_track_declares_assert.attr);
8220 CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8227 static void __exit osd_exit(void)
8229 struct kobject *kobj;
8231 kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8233 sysfs_remove_file(kobj,
8234 &lustre_attr_track_declares_assert.attr);
8237 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8238 lu_kmem_fini(ldiskfs_caches);
8241 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8242 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8243 MODULE_VERSION(LUSTRE_VERSION_STRING);
8244 MODULE_LICENSE("GPL");
8246 module_init(osd_init);
8247 module_exit(osd_exit);