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, 2016, 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/module.h>
43 #include <linux/user_namespace.h>
44 #ifdef HAVE_UIDGID_HEADER
45 # include <linux/uidgid.h>
48 /* LUSTRE_VERSION_CODE */
49 #include <lustre_ver.h>
50 /* prerequisite for linux/xattr.h */
51 #include <linux/types.h>
52 /* prerequisite for linux/xattr.h */
54 /* XATTR_{REPLACE,CREATE} */
55 #include <linux/xattr.h>
57 #include <ldiskfs/ldiskfs.h>
58 #include <ldiskfs/xattr.h>
59 #include <ldiskfs/ldiskfs_extents.h>
62 * struct OBD_{ALLOC,FREE}*()
65 #include <obd_support.h>
66 /* struct ptlrpc_thread */
67 #include <lustre_net.h>
68 #include <lustre_fid.h>
70 #include <uapi/linux/lustre_param.h>
72 #include "osd_internal.h"
73 #include "osd_dynlocks.h"
75 /* llo_* api support */
76 #include <md_object.h>
77 #include <lustre_quota.h>
79 #include <lustre_linkea.h>
81 #define PFID_STRIPE_IDX_BITS 16
82 #define PFID_STRIPE_COUNT_MASK ((1 << PFID_STRIPE_IDX_BITS) - 1)
85 module_param(ldiskfs_pdo, int, 0644);
86 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
88 int ldiskfs_track_declares_assert;
89 module_param(ldiskfs_track_declares_assert, int, 0644);
90 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
92 /* Slab to allocate dynlocks */
93 struct kmem_cache *dynlock_cachep;
95 /* Slab to allocate osd_it_ea */
96 struct kmem_cache *osd_itea_cachep;
98 static struct lu_kmem_descr ldiskfs_caches[] = {
100 .ckd_cache = &dynlock_cachep,
101 .ckd_name = "dynlock_cache",
102 .ckd_size = sizeof(struct dynlock_handle)
105 .ckd_cache = &osd_itea_cachep,
106 .ckd_name = "osd_itea_cache",
107 .ckd_size = sizeof(struct osd_it_ea)
114 static const char dot[] = ".";
115 static const char dotdot[] = "..";
117 static const struct lu_object_operations osd_lu_obj_ops;
118 static const struct dt_object_operations osd_obj_ops;
119 static const struct dt_object_operations osd_obj_otable_it_ops;
120 static const struct dt_index_operations osd_index_iam_ops;
121 static const struct dt_index_operations osd_index_ea_ops;
123 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
124 const struct lu_fid *fid);
125 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
126 struct osd_device *osd);
128 int osd_trans_declare_op2rb[] = {
129 [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
130 [OSD_OT_PUNCH] = OSD_OT_MAX,
131 [OSD_OT_XATTR_SET] = OSD_OT_XATTR_SET,
132 [OSD_OT_CREATE] = OSD_OT_DESTROY,
133 [OSD_OT_DESTROY] = OSD_OT_CREATE,
134 [OSD_OT_REF_ADD] = OSD_OT_REF_DEL,
135 [OSD_OT_REF_DEL] = OSD_OT_REF_ADD,
136 [OSD_OT_WRITE] = OSD_OT_WRITE,
137 [OSD_OT_INSERT] = OSD_OT_DELETE,
138 [OSD_OT_DELETE] = OSD_OT_INSERT,
139 [OSD_OT_QUOTA] = OSD_OT_MAX,
142 static int osd_has_index(const struct osd_object *obj)
144 return obj->oo_dt.do_index_ops != NULL;
147 static int osd_object_invariant(const struct lu_object *l)
149 return osd_invariant(osd_obj(l));
153 * Concurrency: doesn't matter
155 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
157 struct osd_thread_info *oti = osd_oti_get(env);
158 return oti->oti_w_locks > 0 && o->oo_owner == env;
162 * Concurrency: doesn't access mutable data
164 static int osd_root_get(const struct lu_env *env,
165 struct dt_device *dev, struct lu_fid *f)
167 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
172 * the following set of functions are used to maintain per-thread
173 * cache of FID->ino mapping. this mechanism is needed to resolve
174 * FID to inode at dt_insert() which in turn stores ino in the
175 * directory entries to keep ldiskfs compatible with ext[34].
176 * due to locking-originated restrictions we can't lookup ino
177 * using LU cache (deadlock is possible). lookup using OI is quite
178 * expensive. so instead we maintain this cache and methods like
179 * dt_create() fill it. so in the majority of cases dt_insert() is
180 * able to find needed mapping in lockless manner.
182 static struct osd_idmap_cache *
183 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
184 const struct lu_fid *fid)
186 struct osd_thread_info *oti = osd_oti_get(env);
187 struct osd_idmap_cache *idc = oti->oti_ins_cache;
189 for (i = 0; i < oti->oti_ins_cache_used; i++) {
190 if (!lu_fid_eq(&idc[i].oic_fid, fid))
192 if (idc[i].oic_dev != osd)
201 static struct osd_idmap_cache *
202 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
203 const struct lu_fid *fid)
205 struct osd_thread_info *oti = osd_oti_get(env);
206 struct osd_idmap_cache *idc;
209 if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
210 i = oti->oti_ins_cache_size * 2;
212 i = OSD_INS_CACHE_SIZE;
213 OBD_ALLOC(idc, sizeof(*idc) * i);
215 return ERR_PTR(-ENOMEM);
216 if (oti->oti_ins_cache != NULL) {
217 memcpy(idc, oti->oti_ins_cache,
218 oti->oti_ins_cache_used * sizeof(*idc));
219 OBD_FREE(oti->oti_ins_cache,
220 oti->oti_ins_cache_used * sizeof(*idc));
222 oti->oti_ins_cache = idc;
223 oti->oti_ins_cache_size = i;
226 idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
229 idc->oic_lid.oii_ino = 0;
230 idc->oic_lid.oii_gen = 0;
237 * lookup mapping for the given fid in the cache, initialize a
238 * new one if not found. the initialization checks whether the
239 * object is local or remote. for local objects, OI is used to
240 * learn ino/generation. the function is used when the caller
241 * has no information about the object, e.g. at dt_insert().
243 static struct osd_idmap_cache *
244 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
245 const struct lu_fid *fid)
247 struct osd_idmap_cache *idc;
250 idc = osd_idc_find(env, osd, fid);
251 LASSERT(!IS_ERR(idc));
255 /* new mapping is needed */
256 idc = osd_idc_add(env, osd, fid);
261 rc = osd_remote_fid(env, osd, fid);
262 if (unlikely(rc < 0))
266 /* the object is local, lookup in OI */
267 /* XXX: probably cheaper to lookup in LU first? */
268 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
270 if (unlikely(rc < 0)) {
271 CERROR("can't lookup: rc = %d\n", rc);
275 /* the object is remote */
283 * lookup mapping for given FID and fill it from the given object.
284 * the object is lolcal by definition.
286 static int osd_idc_find_and_init(const struct lu_env *env,
287 struct osd_device *osd,
288 struct osd_object *obj)
290 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
291 struct osd_idmap_cache *idc;
293 idc = osd_idc_find(env, osd, fid);
294 LASSERT(!IS_ERR(idc));
296 if (obj->oo_inode == NULL)
298 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
299 LASSERT(idc->oic_lid.oii_ino == 0);
300 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
301 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
306 /* new mapping is needed */
307 idc = osd_idc_add(env, osd, fid);
311 if (obj->oo_inode != NULL) {
312 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
313 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
319 * OSD object methods.
323 * Concurrency: no concurrent access is possible that early in object
326 static struct lu_object *osd_object_alloc(const struct lu_env *env,
327 const struct lu_object_header *hdr,
330 struct osd_object *mo;
336 l = &mo->oo_dt.do_lu;
337 dt_object_init(&mo->oo_dt, NULL, d);
338 mo->oo_dt.do_ops = &osd_obj_ops;
339 l->lo_ops = &osd_lu_obj_ops;
340 init_rwsem(&mo->oo_sem);
341 init_rwsem(&mo->oo_ext_idx_sem);
342 spin_lock_init(&mo->oo_guard);
343 INIT_LIST_HEAD(&mo->oo_xattr_list);
350 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
351 struct dentry *dentry, struct lustre_ost_attrs *loa)
355 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
356 (void *)loa, sizeof(*loa));
358 struct lustre_mdt_attrs *lma = &loa->loa_lma;
360 if (rc < sizeof(*lma))
364 lustre_loa_swab(loa, true);
365 /* Check LMA compatibility */
366 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
367 CWARN("%s: unsupported incompat LMA feature(s) %#x "
368 "for fid = "DFID", ino = %lu\n",
370 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
371 PFID(&lma->lma_self_fid), inode->i_ino);
374 } else if (rc == 0) {
382 * retrieve object from backend ext fs.
384 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
385 struct osd_inode_id *id)
388 struct inode *inode = NULL;
390 /* if we look for an inode withing a running
391 * transaction, then we risk to deadlock */
392 /* osd_dirent_check_repair() breaks this */
393 /*LASSERT(current->journal_info == NULL);*/
395 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
397 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
398 id->oii_ino, PTR_ERR(inode));
399 } else if (id->oii_gen != OSD_OII_NOGEN &&
400 inode->i_generation != id->oii_gen) {
401 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
402 "i_generation = %u\n",
403 id->oii_ino, id->oii_gen, inode->i_generation);
405 inode = ERR_PTR(-ESTALE);
406 } else if (inode->i_nlink == 0) {
407 /* due to parallel readdir and unlink,
408 * we can have dead inode here. */
409 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
411 inode = ERR_PTR(-ESTALE);
412 } else if (is_bad_inode(inode)) {
413 CWARN("%s: bad inode: ino = %u\n",
414 osd_dev2name(dev), id->oii_ino);
416 inode = ERR_PTR(-ENOENT);
417 } else if ((rc = osd_attach_jinode(inode))) {
421 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
422 if (id->oii_gen == OSD_OII_NOGEN)
423 osd_id_gen(id, inode->i_ino, inode->i_generation);
425 /* Do not update file c/mtime in ldiskfs.
426 * NB: we don't have any lock to protect this because we don't
427 * have reference on osd_object now, but contention with
428 * another lookup + attr_set can't happen in the tiny window
429 * between if (...) and set S_NOCMTIME. */
430 if (!(inode->i_flags & S_NOCMTIME))
431 inode->i_flags |= S_NOCMTIME;
436 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
437 handle_t *handle, struct dentry *child,
438 struct inode *inode, struct htree_lock *hlock)
442 rc = __ldiskfs_add_entry(handle, child, inode, hlock);
443 if (rc == -ENOBUFS || rc == -ENOSPC) {
444 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
445 struct inode *parent = child->d_parent->d_inode;
446 struct lu_fid *fid = NULL;
448 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
450 fid = &loa->loa_lma.lma_self_fid;
451 } else if (rc2 == -ENODATA) {
452 if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
453 fid = &info->oti_fid3;
454 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
455 } else if (!osd->od_is_ost && osd->od_index == 0) {
456 fid = &info->oti_fid3;
457 lu_igif_build(fid, parent->i_ino,
458 parent->i_generation);
463 CWARN("%s: directory (inode: %lu, FID: "DFID") %s "
464 "maximum entry limit\n",
465 osd_name(osd), parent->i_ino, PFID(fid),
466 rc == -ENOSPC ? "has reached" : "is approaching");
468 CWARN("%s: directory (inode: %lu, FID: unknown) %s "
469 "maximum entry limit\n",
470 osd_name(osd), parent->i_ino,
471 rc == -ENOSPC ? "has reached" : "is approaching");
473 /* ignore such error now */
482 static struct inode *
483 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
484 struct osd_inode_id *id, struct lu_fid *fid)
486 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
490 inode = osd_iget(info, dev, id);
494 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
496 *fid = loa->loa_lma.lma_self_fid;
497 } else if (rc == -ENODATA) {
498 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
499 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
501 lu_igif_build(fid, inode->i_ino, inode->i_generation);
509 static struct inode *osd_iget_check(struct osd_thread_info *info,
510 struct osd_device *dev,
511 const struct lu_fid *fid,
512 struct osd_inode_id *id,
519 /* The cached OI mapping is trustable. If we cannot locate the inode
520 * via the cached OI mapping, then return the failure to the caller
521 * directly without further OI checking. */
524 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
527 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
530 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
531 PFID(fid), id->oii_ino, rc);
535 if (is_bad_inode(inode)) {
540 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
541 PFID(fid), id->oii_ino);
545 if (id->oii_gen != OSD_OII_NOGEN &&
546 inode->i_generation != id->oii_gen) {
551 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
552 "oii_gen = %u, i_generation = %u\n", PFID(fid),
553 id->oii_ino, id->oii_gen, inode->i_generation);
557 if (inode->i_nlink == 0) {
562 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
563 PFID(fid), id->oii_ino);
567 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
571 __u32 saved_ino = id->oii_ino;
572 __u32 saved_gen = id->oii_gen;
575 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
577 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
578 /* XXX: There are four possible cases:
580 * Backup/restore caused the OI invalid.
582 * Someone unlinked the object but NOT removed
583 * the OI mapping, such as mount target device
584 * as ldiskfs, and modify something directly.
586 * Someone just removed the object between the
587 * former oi_lookup and the iget. It is normal.
588 * 4. Other failure cases.
590 * Generally, when the device is mounted, it will
591 * auto check whether the system is restored from
592 * file-level backup or not. We trust such detect
593 * to distinguish the 1st case from the 2nd case:
594 * if the OI files are consistent but may contain
595 * stale OI mappings because of case 2, if iget()
596 * returns -ENOENT or -ESTALE, then it should be
599 /* If the OI mapping was in OI file before the
600 * osd_iget_check(), but now, it is disappear,
601 * then it must be removed by race. That is a
602 * normal race case. */
605 /* It is the OI scrub updated the OI mapping by race.
606 * The new OI mapping must be valid. */
607 if (saved_ino != id->oii_ino || saved_gen != id->oii_gen) {
613 if (dev->od_scrub.os_file.sf_flags & SF_INCONSISTENT)
614 /* It still can be the case 2, but we cannot
615 * distinguish it from the case 1. So return
616 * -EREMCHG to block current operation until
617 * OI scrub rebuilt the OI mappings. */
625 if (inode->i_generation == id->oii_gen)
630 if (id->oii_gen == OSD_OII_NOGEN)
631 osd_id_gen(id, inode->i_ino, inode->i_generation);
633 /* Do not update file c/mtime in ldiskfs.
634 * NB: we don't have any lock to protect this because we don't
635 * have reference on osd_object now, but contention with
636 * another lookup + attr_set can't happen in the tiny window
637 * between if (...) and set S_NOCMTIME. */
638 if (!(inode->i_flags & S_NOCMTIME))
639 inode->i_flags |= S_NOCMTIME;
656 * \retval +v: new filter_fid, does not contain self-fid
657 * \retval 0: filter_fid_old, contains self-fid
658 * \retval -v: other failure cases
660 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
661 struct dentry *dentry, struct lu_fid *fid)
663 struct filter_fid_old *ff = &info->oti_ff;
664 struct ost_id *ostid = &info->oti_ostid;
667 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
668 if (rc == sizeof(*ff)) {
670 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
671 rc = ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
673 * XXX: use 0 as the index for compatibility, the caller will
674 * handle index related issues when necessary.
677 ostid_to_fid(fid, ostid, 0);
678 } else if (rc == sizeof(struct filter_fid)) {
680 } else if (rc >= 0) {
687 static int osd_lma_self_repair(struct osd_thread_info *info,
688 struct osd_device *osd, struct inode *inode,
689 const struct lu_fid *fid, __u32 compat)
694 LASSERT(current->journal_info == NULL);
696 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
697 osd_dto_credits_noquota[DTO_XATTR_SET]);
700 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
705 rc = osd_ea_fid_set(info, inode, fid, compat, 0);
707 CWARN("%s: cannot self repair the LMA: rc = %d\n",
709 ldiskfs_journal_stop(jh);
713 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
715 struct osd_thread_info *info = osd_oti_get(env);
716 struct osd_device *osd = osd_obj2dev(obj);
717 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
718 struct lustre_mdt_attrs *lma = &loa->loa_lma;
719 struct inode *inode = obj->oo_inode;
720 struct dentry *dentry = &info->oti_obj_dentry;
721 struct lu_fid *fid = NULL;
722 const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
726 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
727 (void *)loa, sizeof(*loa));
728 if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
729 fid = &lma->lma_self_fid;
730 rc = osd_get_idif(info, inode, dentry, fid);
731 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
732 /* For the given OST-object, if it has neither LMA nor
733 * FID in XATTR_NAME_FID, then the given FID (which is
734 * contained in the @obj, from client RPC for locating
735 * the OST-object) is trusted. We use it to generate
737 osd_lma_self_repair(info, osd, inode, rfid,
748 lustre_lma_swab(lma);
749 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
750 (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
751 S_ISREG(inode->i_mode)))) {
752 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
753 "fid = "DFID", ino = %lu\n", osd_name(osd),
754 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
755 PFID(rfid), inode->i_ino);
758 fid = &lma->lma_self_fid;
762 if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
763 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
764 struct ost_id *oi = &info->oti_ostid;
765 struct lu_fid *fid1 = &info->oti_fid3;
766 __u32 idx = fid_idif_ost_idx(rfid);
768 /* For old IDIF, the OST index is not part of the IDIF,
769 * Means that different OSTs may have the same IDIFs.
770 * Under such case, we need to make some compatible
771 * check to make sure to trigger OI scrub properly. */
772 if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
773 /* Given @rfid is new, LMA is old. */
774 fid_to_ostid(fid, oi);
775 ostid_to_fid(fid1, oi, idx);
776 if (lu_fid_eq(fid1, rfid)) {
777 if (osd->od_index_in_idif)
778 osd_lma_self_repair(info, osd,
792 struct osd_check_lmv_buf {
793 #ifdef HAVE_DIR_CONTEXT
794 /* please keep it as first member */
795 struct dir_context ctx;
797 struct osd_thread_info *oclb_info;
798 struct osd_device *oclb_dev;
799 struct osd_idmap_cache *oclb_oic;
803 * It is called internally by ->readdir() to filter out the
804 * local slave object's FID of the striped directory.
806 * \retval 1 found the local slave's FID
807 * \retval 0 continue to check next item
808 * \retval -ve for failure
810 #ifdef HAVE_FILLDIR_USE_CTX
811 static int osd_stripe_dir_filldir(struct dir_context *buf,
813 static int osd_stripe_dir_filldir(void *buf,
815 const char *name, int namelen,
816 loff_t offset, __u64 ino, unsigned d_type)
818 struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
819 struct osd_thread_info *oti = oclb->oclb_info;
820 struct lu_fid *fid = &oti->oti_fid3;
821 struct osd_inode_id *id = &oti->oti_id3;
822 struct osd_device *dev = oclb->oclb_dev;
823 struct osd_idmap_cache *oic = oclb->oclb_oic;
831 sscanf(name + 1, SFID, RFID(fid));
832 if (!fid_is_sane(fid))
835 if (osd_remote_fid(oti->oti_env, dev, fid))
838 osd_id_gen(id, ino, OSD_OII_NOGEN);
839 inode = osd_iget(oti, dev, id);
841 return PTR_ERR(inode);
844 osd_add_oi_cache(oti, dev, id, fid);
848 rc = osd_oii_insert(dev, oic, true);
850 return rc == 0 ? 1 : rc;
853 /* When lookup item under striped directory, we need to locate the master
854 * MDT-object of the striped directory firstly, then the client will send
855 * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
856 * and the item's name. If the system is restored from MDT file level backup,
857 * then before the OI scrub completely built the OI files, the OI mappings of
858 * the master MDT-object and slave MDT-object may be invalid. Usually, it is
859 * not a problem for the master MDT-object. Because when locate the master
860 * MDT-object, we will do name based lookup (for the striped directory itself)
861 * firstly, during such process we can setup the correct OI mapping for the
862 * master MDT-object. But it will be trouble for the slave MDT-object. Because
863 * the client will not trigger name based lookup on the MDT to locate the slave
864 * MDT-object before locating item under the striped directory, then when
865 * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
866 * is invalid and does not know what the right OI mapping is, then the MDT has
867 * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
868 * the OI file, related OI mapping is unknown yet, please try again later. And
869 * then client will re-try the RPC again and again until related OI mapping has
870 * been updated. That is quite inefficient.
872 * To resolve above trouble, we will handle it as the following two cases:
874 * 1) The slave MDT-object and the master MDT-object are on different MDTs.
875 * It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
876 * is linked under /REMOTE_PARENT_DIR with the name of its FID string.
877 * We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
878 * directly. Please check osd_fid_lookup().
880 * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
881 * Under such case, during lookup the master MDT-object, we will lookup the
882 * slave MDT-object via readdir against the master MDT-object, because the
883 * slave MDT-objects information are stored as sub-directories with the name
884 * "${FID}:${index}". Then when find the local slave MDT-object, its OI
885 * mapping will be recorded. Then subsequent osd_fid_lookup() will know
886 * the correct OI mapping for the slave MDT-object. */
887 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
888 struct inode *inode, struct osd_idmap_cache *oic)
890 struct lu_buf *buf = &oti->oti_big_buf;
891 struct dentry *dentry = &oti->oti_obj_dentry;
892 struct file *filp = &oti->oti_file;
893 const struct file_operations *fops;
894 struct lmv_mds_md_v1 *lmv1;
895 struct osd_check_lmv_buf oclb = {
896 #ifdef HAVE_DIR_CONTEXT
897 .ctx.actor = osd_stripe_dir_filldir,
907 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
910 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
912 lu_buf_realloc(buf, rc);
913 if (buf->lb_buf == NULL)
914 GOTO(out, rc = -ENOMEM);
920 if (unlikely(rc == 0 || rc == -ENODATA))
926 if (unlikely(buf->lb_buf == NULL)) {
927 lu_buf_realloc(buf, rc);
928 if (buf->lb_buf == NULL)
929 GOTO(out, rc = -ENOMEM);
935 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
939 dentry->d_inode = inode;
940 dentry->d_sb = inode->i_sb;
942 filp->f_path.dentry = dentry;
943 filp->f_mode = FMODE_64BITHASH;
944 filp->f_mapping = inode->i_mapping;
946 filp->private_data = NULL;
947 set_file_inode(filp, inode);
949 #ifdef HAVE_DIR_CONTEXT
950 oclb.ctx.pos = filp->f_pos;
951 rc = fops->iterate(filp, &oclb.ctx);
952 filp->f_pos = oclb.ctx.pos;
954 rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir);
956 fops->release(inode, filp);
960 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
961 DFID": rc = %d\n", osd_ino2name(inode),
962 inode->i_ino, inode->i_generation,
963 PFID(&oic->oic_fid), rc);
970 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
971 const struct lu_fid *fid,
972 const struct lu_object_conf *conf)
974 struct osd_thread_info *info;
975 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
976 struct osd_device *dev;
977 struct osd_idmap_cache *oic;
978 struct osd_inode_id *id;
979 struct inode *inode = NULL;
980 struct osd_scrub *scrub;
981 struct scrub_file *sf;
982 __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
989 bool updated = false;
992 LINVRNT(osd_invariant(obj));
993 LASSERT(obj->oo_inode == NULL);
994 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
997 scrub = &dev->od_scrub;
998 sf = &scrub->os_file;
999 info = osd_oti_get(env);
1001 oic = &info->oti_cache;
1003 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1006 /* For the object is created as locking anchor, or for the object to
1007 * be created on disk. No need to osd_oi_lookup() at here because FID
1008 * shouldn't never be re-used, if it's really a duplicate FID from
1009 * unexpected reason, we should be able to detect it later by calling
1010 * do_create->osd_oi_insert(). */
1011 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
1012 GOTO(out, result = 0);
1014 /* Search order: 1. per-thread cache. */
1015 if (lu_fid_eq(fid, &oic->oic_fid) &&
1016 likely(oic->oic_dev == dev)) {
1022 if (!list_empty(&scrub->os_inconsistent_items)) {
1023 /* Search order: 2. OI scrub pending list. */
1024 result = osd_oii_lookup(dev, fid, id);
1029 /* The OI mapping in the OI file can be updated by the OI scrub
1030 * when we locate the inode via FID. So it may be not trustable. */
1033 /* Search order: 3. OI files. */
1034 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1035 if (result == -ENOENT) {
1036 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1037 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1038 !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
1040 GOTO(out, result = 0);
1049 inode = osd_iget_check(info, dev, fid, id, trusted);
1050 if (IS_ERR(inode)) {
1051 result = PTR_ERR(inode);
1052 if (result == -ENOENT || result == -ESTALE)
1053 GOTO(out, result = 0);
1055 if (result == -EREMCHG) {
1058 /* We still have chance to get the valid inode: for the
1059 * object which is referenced by remote name entry, the
1060 * object on the local MDT will be linked under the dir
1061 * of "/REMOTE_PARENT_DIR" with its FID string as name.
1063 * We do not know whether the object for the given FID
1064 * is referenced by some remote name entry or not, and
1065 * especially for DNE II, a multiple-linked object may
1066 * have many name entries reside on many MDTs.
1068 * To simplify the operation, OSD will not distinguish
1069 * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1070 * only happened for the RPC from other MDT during the
1071 * OI scrub, or for the client side RPC with FID only,
1072 * such as FID to path, or from old connected client. */
1074 !fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1075 rc1 = osd_lookup_in_remote_parent(info, dev,
1080 flags |= SS_AUTO_PARTIAL;
1081 flags &= ~SS_AUTO_FULL;
1086 if (thread_is_running(&scrub->os_thread)) {
1087 if (scrub->os_partial_scan &&
1088 !scrub->os_in_join) {
1091 if (inode != NULL && !IS_ERR(inode)) {
1094 osd_add_oi_cache(info, dev, id,
1096 osd_oii_insert(dev, oic, true);
1098 result = -EINPROGRESS;
1101 } else if (!dev->od_noscrub) {
1104 rc1 = osd_scrub_start(dev, flags);
1105 LCONSOLE_WARN("%s: trigger OI scrub by RPC "
1106 "for the "DFID" with flags 0x%x,"
1107 " rc = %d\n", osd_name(dev),
1108 PFID(fid), flags, rc1);
1109 if (rc1 == 0 || rc1 == -EALREADY) {
1110 if (inode != NULL && !IS_ERR(inode)) {
1113 osd_add_oi_cache(info, dev, id,
1115 osd_oii_insert(dev, oic, true);
1117 result = -EINPROGRESS;
1127 if (inode == NULL || IS_ERR(inode))
1129 } else if (remote) {
1133 obj->oo_inode = inode;
1134 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
1136 result = osd_check_lma(env, obj);
1140 LASSERTF(id->oii_ino == inode->i_ino &&
1141 id->oii_gen == inode->i_generation,
1142 "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1143 PFID(fid), id->oii_ino, id->oii_gen,
1144 inode->i_ino, inode->i_generation);
1146 saved_ino = inode->i_ino;
1147 saved_gen = inode->i_generation;
1149 if (unlikely(result == -ENODATA)) {
1150 /* If the OI scrub updated the OI mapping by race, it
1151 * must be valid. Trust the inode that has no LMA EA. */
1155 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1157 /* The OI mapping is still there, the inode is still
1158 * valid. It is just becaues the inode has no LMA EA. */
1159 if (saved_ino == id->oii_ino &&
1160 saved_gen == id->oii_gen)
1163 /* It is the OI scrub updated the OI mapping by race.
1164 * The new OI mapping must be valid. */
1167 obj->oo_inode = NULL;
1173 /* "result == -ENOENT" means that the OI mappinghas been
1174 * removed by race, so the inode belongs to other object.
1176 * Others error can be returned directly. */
1177 if (result == -ENOENT) {
1186 obj->oo_inode = NULL;
1188 if (result != -EREMCHG)
1193 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1194 /* "result == -ENOENT" means the cached OI mapping has been removed
1195 * from the OI file by race, above inode belongs to other object. */
1196 if (result == -ENOENT) {
1199 GOTO(out, result = 0);
1205 if (saved_ino == id->oii_ino && saved_gen == id->oii_gen)
1208 /* It is the OI scrub updated the OI mapping by race.
1209 * The new OI mapping must be valid. */
1215 obj->oo_compat_dot_created = 1;
1216 obj->oo_compat_dotdot_created = 1;
1218 if (S_ISDIR(inode->i_mode) &&
1219 (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1220 osd_check_lmv(info, dev, inode, oic);
1222 result = osd_attach_jinode(inode);
1224 obj->oo_inode = NULL;
1230 GOTO(out, result = 0);
1232 LASSERT(obj->oo_hl_head == NULL);
1233 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1234 if (obj->oo_hl_head == NULL) {
1235 obj->oo_inode = NULL;
1237 GOTO(out, result = -ENOMEM);
1239 GOTO(out, result = 0);
1242 if (result != 0 && trusted)
1243 fid_zero(&oic->oic_fid);
1245 LINVRNT(osd_invariant(obj));
1250 * Concurrency: shouldn't matter.
1252 static void osd_object_init0(struct osd_object *obj)
1254 LASSERT(obj->oo_inode != NULL);
1255 obj->oo_dt.do_body_ops = &osd_body_ops;
1256 obj->oo_dt.do_lu.lo_header->loh_attr |=
1257 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1261 * Concurrency: no concurrent access is possible that early in object
1264 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1265 const struct lu_object_conf *conf)
1267 struct osd_object *obj = osd_obj(l);
1270 LINVRNT(osd_invariant(obj));
1272 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1273 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1274 l->lo_header->loh_attr |= LOHA_EXISTS;
1278 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1279 obj->oo_dt.do_body_ops = &osd_body_ops_new;
1280 if (result == 0 && obj->oo_inode != NULL) {
1281 struct osd_thread_info *oti = osd_oti_get(env);
1282 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1284 osd_object_init0(obj);
1285 result = osd_get_lma(oti, obj->oo_inode,
1286 &oti->oti_obj_dentry, loa);
1288 /* Convert LMAI flags to lustre LMA flags
1289 * and cache it to oo_lma_flags */
1291 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1292 } else if (result == -ENODATA) {
1297 LINVRNT(osd_invariant(obj));
1301 /* The first part of oxe_buf is xattr name, and is '\0' terminated.
1302 * The left part is for value, binary mode. */
1303 struct osd_xattr_entry {
1304 struct list_head oxe_list;
1308 struct rcu_head oxe_rcu;
1312 static int osd_oxc_get(struct osd_object *obj, const char *name,
1315 struct osd_xattr_entry *tmp;
1316 struct osd_xattr_entry *oxe = NULL;
1317 size_t namelen = strlen(name);
1322 list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1323 if (namelen == tmp->oxe_namelen &&
1324 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1331 GOTO(out, rc = -ENOENT);
1333 if (!oxe->oxe_exist)
1334 GOTO(out, rc = -ENODATA);
1337 rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1340 if (buf->lb_buf == NULL)
1343 if (buf->lb_len < rc)
1344 GOTO(out, rc = -ERANGE);
1346 memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1354 static void osd_oxc_free(struct rcu_head *head)
1356 struct osd_xattr_entry *oxe;
1358 oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1359 OBD_FREE(oxe, oxe->oxe_len);
1362 static void osd_oxc_add(struct osd_object *obj, const char *name,
1363 const char *buf, int buflen)
1365 struct osd_xattr_entry *oxe;
1366 struct osd_xattr_entry *old = NULL;
1367 struct osd_xattr_entry *tmp;
1368 size_t namelen = strlen(name);
1369 size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1371 OBD_ALLOC(oxe, len);
1375 INIT_LIST_HEAD(&oxe->oxe_list);
1377 oxe->oxe_namelen = namelen;
1378 memcpy(oxe->oxe_buf, name, namelen);
1380 LASSERT(buf != NULL);
1381 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1382 oxe->oxe_exist = true;
1384 oxe->oxe_exist = false;
1387 /* this should be rarely called, just remove old and add new */
1388 spin_lock(&obj->oo_guard);
1389 list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1390 if (namelen == tmp->oxe_namelen &&
1391 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1397 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1398 call_rcu(&old->oxe_rcu, osd_oxc_free);
1400 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1402 spin_unlock(&obj->oo_guard);
1405 static void osd_oxc_del(struct osd_object *obj, const char *name)
1407 struct osd_xattr_entry *oxe;
1408 size_t namelen = strlen(name);
1410 spin_lock(&obj->oo_guard);
1411 list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1412 if (namelen == oxe->oxe_namelen &&
1413 strncmp(name, oxe->oxe_buf, namelen) == 0) {
1414 list_del_rcu(&oxe->oxe_list);
1415 call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1419 spin_unlock(&obj->oo_guard);
1422 static void osd_oxc_fini(struct osd_object *obj)
1424 struct osd_xattr_entry *oxe, *next;
1426 list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1427 list_del(&oxe->oxe_list);
1428 OBD_FREE(oxe, oxe->oxe_len);
1433 * Concurrency: no concurrent access is possible that late in object
1436 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1438 struct osd_object *obj = osd_obj(l);
1440 LINVRNT(osd_invariant(obj));
1443 dt_object_fini(&obj->oo_dt);
1444 if (obj->oo_hl_head != NULL)
1445 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1450 * Concurrency: no concurrent access is possible that late in object
1453 static void osd_index_fini(struct osd_object *o)
1455 struct iam_container *bag;
1457 if (o->oo_dir != NULL) {
1458 bag = &o->oo_dir->od_container;
1459 if (o->oo_inode != NULL) {
1460 if (bag->ic_object == o->oo_inode)
1461 iam_container_fini(bag);
1463 OBD_FREE_PTR(o->oo_dir);
1469 * Concurrency: no concurrent access is possible that late in object
1470 * life-cycle (for all existing callers, that is. New callers have to provide
1471 * their own locking.)
1473 static int osd_inode_unlinked(const struct inode *inode)
1475 return inode->i_nlink == 0;
1479 OSD_TXN_OI_DELETE_CREDITS = 20,
1480 OSD_TXN_INODE_DELETE_CREDITS = 20
1487 #if OSD_THANDLE_STATS
1489 * Set time when the handle is allocated
1491 static void osd_th_alloced(struct osd_thandle *oth)
1493 oth->oth_alloced = ktime_get();
1497 * Set time when the handle started
1499 static void osd_th_started(struct osd_thandle *oth)
1501 oth->oth_started = ktime_get();
1505 * Check whether the we deal with this handle for too long.
1507 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1508 ktime_t alloced, ktime_t started,
1511 ktime_t now = ktime_get();
1513 LASSERT(dev != NULL);
1515 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1516 ktime_us_delta(started, alloced));
1517 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1518 ktime_us_delta(closed, started));
1519 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1520 ktime_us_delta(now, closed));
1522 if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1523 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1524 oth, now, alloced, started, closed);
1525 libcfs_debug_dumpstack(NULL);
1529 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
1531 ktime_t __closed = ktime_get(); \
1532 ktime_t __alloced = oth->oth_alloced; \
1533 ktime_t __started = oth->oth_started; \
1536 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
1539 #else /* OSD_THANDLE_STATS */
1541 #define osd_th_alloced(h) do {} while(0)
1542 #define osd_th_started(h) do {} while(0)
1543 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
1545 #endif /* OSD_THANDLE_STATS */
1548 * Concurrency: doesn't access mutable data.
1550 static int osd_param_is_not_sane(const struct osd_device *dev,
1551 const struct thandle *th)
1553 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1555 return oh->ot_credits > osd_transaction_size(dev);
1559 * Concurrency: shouldn't matter.
1561 static void osd_trans_commit_cb(struct super_block *sb,
1562 struct ldiskfs_journal_cb_entry *jcb, int error)
1564 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1565 struct thandle *th = &oh->ot_super;
1566 struct lu_device *lud = &th->th_dev->dd_lu_dev;
1567 struct dt_txn_commit_cb *dcb, *tmp;
1569 LASSERT(oh->ot_handle == NULL);
1572 CERROR("transaction @0x%p commit error: %d\n", th, error);
1574 dt_txn_hook_commit(th);
1576 /* call per-transaction callbacks if any */
1577 list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1579 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1580 "commit callback entry: magic=%x name='%s'\n",
1581 dcb->dcb_magic, dcb->dcb_name);
1582 list_del_init(&dcb->dcb_linkage);
1583 dcb->dcb_func(NULL, th, dcb, error);
1586 lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1590 lu_context_exit(&th->th_ctx);
1591 lu_context_fini(&th->th_ctx);
1595 #ifndef HAVE_SB_START_WRITE
1596 # define sb_start_write(sb) do {} while (0)
1597 # define sb_end_write(sb) do {} while (0)
1600 static struct thandle *osd_trans_create(const struct lu_env *env,
1601 struct dt_device *d)
1603 struct osd_thread_info *oti = osd_oti_get(env);
1604 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1605 struct osd_thandle *oh;
1610 CERROR("%s: someone try to start transaction under "
1611 "readonly mode, should be disabled.\n",
1612 osd_name(osd_dt_dev(d)));
1614 RETURN(ERR_PTR(-EROFS));
1617 /* on pending IO in this thread should left from prev. request */
1618 LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1620 sb_start_write(osd_sb(osd_dt_dev(d)));
1622 OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
1624 oh->ot_quota_trans = &oti->oti_quota_trans;
1625 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1629 th->th_tags = LCT_TX_HANDLE;
1631 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1632 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1635 memset(oti->oti_declare_ops, 0,
1636 sizeof(oti->oti_declare_ops));
1637 memset(oti->oti_declare_ops_cred, 0,
1638 sizeof(oti->oti_declare_ops_cred));
1639 memset(oti->oti_declare_ops_used, 0,
1640 sizeof(oti->oti_declare_ops_used));
1642 sb_end_write(osd_sb(osd_dt_dev(d)));
1643 th = ERR_PTR(-ENOMEM);
1648 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1650 struct osd_thread_info *oti = osd_oti_get(env);
1651 struct osd_thandle *oh;
1653 oh = container_of0(th, struct osd_thandle, ot_super);
1654 LASSERT(oh != NULL);
1656 CWARN(" create: %u/%u/%u, destroy: %u/%u/%u\n",
1657 oti->oti_declare_ops[OSD_OT_CREATE],
1658 oti->oti_declare_ops_cred[OSD_OT_CREATE],
1659 oti->oti_declare_ops_used[OSD_OT_CREATE],
1660 oti->oti_declare_ops[OSD_OT_DESTROY],
1661 oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1662 oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1663 CWARN(" attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1664 oti->oti_declare_ops[OSD_OT_ATTR_SET],
1665 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1666 oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1667 oti->oti_declare_ops[OSD_OT_XATTR_SET],
1668 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1669 oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1670 CWARN(" write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1671 oti->oti_declare_ops[OSD_OT_WRITE],
1672 oti->oti_declare_ops_cred[OSD_OT_WRITE],
1673 oti->oti_declare_ops_used[OSD_OT_WRITE],
1674 oti->oti_declare_ops[OSD_OT_PUNCH],
1675 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1676 oti->oti_declare_ops_used[OSD_OT_PUNCH],
1677 oti->oti_declare_ops[OSD_OT_QUOTA],
1678 oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1679 oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1680 CWARN(" insert: %u/%u/%u, delete: %u/%u/%u\n",
1681 oti->oti_declare_ops[OSD_OT_INSERT],
1682 oti->oti_declare_ops_cred[OSD_OT_INSERT],
1683 oti->oti_declare_ops_used[OSD_OT_INSERT],
1684 oti->oti_declare_ops[OSD_OT_DELETE],
1685 oti->oti_declare_ops_cred[OSD_OT_DELETE],
1686 oti->oti_declare_ops_used[OSD_OT_DELETE]);
1687 CWARN(" ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1688 oti->oti_declare_ops[OSD_OT_REF_ADD],
1689 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1690 oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1691 oti->oti_declare_ops[OSD_OT_REF_DEL],
1692 oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1693 oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1697 * Concurrency: shouldn't matter.
1699 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1702 struct osd_thread_info *oti = osd_oti_get(env);
1703 struct osd_device *dev = osd_dt_dev(d);
1705 struct osd_thandle *oh;
1710 LASSERT(current->journal_info == NULL);
1712 oh = container_of0(th, struct osd_thandle, ot_super);
1713 LASSERT(oh != NULL);
1714 LASSERT(oh->ot_handle == NULL);
1716 rc = dt_txn_hook_start(env, d, th);
1720 if (unlikely(osd_param_is_not_sane(dev, th))) {
1721 static unsigned long last_printed;
1722 static int last_credits;
1724 /* don't make noise on a tiny testing systems
1725 * actual credits misuse will be caught anyway */
1726 if (last_credits != oh->ot_credits &&
1727 time_after(jiffies, last_printed +
1728 msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1729 osd_transaction_size(dev) > 512) {
1730 CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1731 oh->ot_credits, osd_transaction_size(dev));
1732 osd_trans_dump_creds(env, th);
1733 libcfs_debug_dumpstack(NULL);
1734 last_credits = oh->ot_credits;
1735 last_printed = jiffies;
1737 /* XXX Limit the credits to 'max_transaction_buffers', and
1738 * let the underlying filesystem to catch the error if
1739 * we really need so many credits.
1741 * This should be removed when we can calculate the
1742 * credits precisely. */
1743 oh->ot_credits = osd_transaction_size(dev);
1744 } else if (ldiskfs_track_declares_assert != 0) {
1745 /* reserve few credits to prevent an assertion in JBD
1746 * our debugging mechanism will be able to detected
1747 * overuse. this can help to debug single-update
1749 oh->ot_credits += 10;
1750 if (unlikely(osd_param_is_not_sane(dev, th)))
1751 oh->ot_credits = osd_transaction_size(dev);
1755 * XXX temporary stuff. Some abstraction layer should
1758 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1762 LASSERT(oti->oti_txns == 0);
1763 lu_context_init(&th->th_ctx, th->th_tags);
1764 lu_context_enter(&th->th_ctx);
1766 lu_device_get(&d->dd_lu_dev);
1767 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1778 static int osd_seq_exists(const struct lu_env *env,
1779 struct osd_device *osd, u64 seq)
1781 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1782 struct seq_server_site *ss = osd_seq_site(osd);
1786 LASSERT(ss != NULL);
1787 LASSERT(ss->ss_server_fld != NULL);
1789 rc = osd_fld_lookup(env, osd, seq, range);
1792 CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1793 osd_name(osd), seq, rc);
1797 RETURN(ss->ss_node_id == range->lsr_index);
1800 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1802 struct dt_txn_commit_cb *dcb;
1803 struct dt_txn_commit_cb *tmp;
1805 /* call per-transaction stop callbacks if any */
1806 list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1808 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1809 "commit callback entry: magic=%x name='%s'\n",
1810 dcb->dcb_magic, dcb->dcb_name);
1811 list_del_init(&dcb->dcb_linkage);
1812 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1817 * Concurrency: shouldn't matter.
1819 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1822 int rc = 0, remove_agents = 0;
1823 struct osd_thandle *oh;
1824 struct osd_thread_info *oti = osd_oti_get(env);
1825 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1826 struct osd_device *osd = osd_dt_dev(th->th_dev);
1827 struct qsd_instance *qsd = osd->od_quota_slave;
1828 struct lquota_trans *qtrans;
1831 oh = container_of0(th, struct osd_thandle, ot_super);
1833 /* reset OI cache for safety */
1834 oti->oti_ins_cache_used = 0;
1836 remove_agents = oh->ot_remove_agents;
1838 qtrans = oh->ot_quota_trans;
1839 oh->ot_quota_trans = NULL;
1841 if (oh->ot_handle != NULL) {
1843 handle_t *hdl = oh->ot_handle;
1846 * add commit callback
1847 * notice we don't do this in osd_trans_start()
1848 * as underlying transaction can change during truncate
1850 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1853 LASSERT(oti->oti_txns == 1);
1856 rc = dt_txn_hook_stop(env, th);
1858 CERROR("%s: failed in transaction hook: rc = %d\n",
1861 osd_trans_stop_cb(oh, rc);
1862 /* hook functions might modify th_sync */
1863 hdl->h_sync = th->th_sync;
1865 oh->ot_handle = NULL;
1866 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
1868 CERROR("%s: failed to stop transaction: rc = %d\n",
1869 osd_name(osd), rc2);
1873 osd_trans_stop_cb(oh, th->th_result);
1877 /* inform the quota slave device that the transaction is stopping */
1878 qsd_op_end(env, qsd, qtrans);
1880 /* as we want IO to journal and data IO be concurrent, we don't block
1881 * awaiting data IO completion in osd_do_bio(), instead we wait here
1882 * once transaction is submitted to the journal. all reqular requests
1883 * don't do direct IO (except read/write), thus this wait_event becomes
1886 * IMPORTANT: we have to wait till any IO submited by the thread is
1887 * completed otherwise iobuf may be corrupted by different request
1889 wait_event(iobuf->dr_wait,
1890 atomic_read(&iobuf->dr_numreqs) == 0);
1891 osd_fini_iobuf(osd, iobuf);
1893 rc = iobuf->dr_error;
1895 if (unlikely(remove_agents != 0))
1896 osd_process_scheduled_agent_removals(env, osd);
1898 sb_end_write(osd_sb(osd));
1903 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1905 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1908 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1909 LASSERT(&dcb->dcb_func != NULL);
1910 if (dcb->dcb_flags & DCB_TRANS_STOP)
1911 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1913 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1919 * Called just before object is freed. Releases all resources except for
1920 * object itself (that is released by osd_object_free()).
1922 * Concurrency: no concurrent access is possible that late in object
1925 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1927 struct osd_object *obj = osd_obj(l);
1928 struct inode *inode = obj->oo_inode;
1930 LINVRNT(osd_invariant(obj));
1933 * If object is unlinked remove fid->ino mapping from object index.
1936 osd_index_fini(obj);
1937 if (inode != NULL) {
1938 struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave;
1939 qid_t uid = i_uid_read(inode);
1940 qid_t gid = i_gid_read(inode);
1943 obj->oo_inode = NULL;
1946 struct osd_thread_info *info = osd_oti_get(env);
1947 struct lquota_id_info *qi = &info->oti_qi;
1949 /* Release granted quota to master if necessary */
1950 qi->lqi_id.qid_uid = uid;
1951 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1953 qi->lqi_id.qid_uid = gid;
1954 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1956 qi->lqi_id.qid_uid = i_projid_read(inode);
1957 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
1963 * Concurrency: ->loo_object_release() is called under site spin-lock.
1965 static void osd_object_release(const struct lu_env *env,
1966 struct lu_object *l)
1968 struct osd_object *o = osd_obj(l);
1969 /* nobody should be releasing a non-destroyed object with nlink=0
1970 * the API allows this, but ldiskfs doesn't like and then report
1971 * this inode as deleted */
1972 if (unlikely(!o->oo_destroyed && o->oo_inode && o->oo_inode->i_nlink == 0))
1977 * Concurrency: shouldn't matter.
1979 static int osd_object_print(const struct lu_env *env, void *cookie,
1980 lu_printer_t p, const struct lu_object *l)
1982 struct osd_object *o = osd_obj(l);
1983 struct iam_descr *d;
1985 if (o->oo_dir != NULL)
1986 d = o->oo_dir->od_container.ic_descr;
1989 return (*p)(env, cookie,
1990 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1992 o->oo_inode ? o->oo_inode->i_ino : 0UL,
1993 o->oo_inode ? o->oo_inode->i_generation : 0,
1994 d ? d->id_ops->id_name : "plain");
1998 * Concurrency: shouldn't matter.
2000 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2001 struct obd_statfs *sfs)
2003 struct osd_device *osd = osd_dt_dev(d);
2004 struct super_block *sb = osd_sb(osd);
2005 struct kstatfs *ksfs;
2009 if (unlikely(osd->od_mnt == NULL))
2010 return -EINPROGRESS;
2012 /* osd_lproc.c call this without env, allocate ksfs for that case */
2013 if (unlikely(env == NULL)) {
2014 OBD_ALLOC_PTR(ksfs);
2018 ksfs = &osd_oti_get(env)->oti_ksfs;
2021 result = sb->s_op->statfs(sb->s_root, ksfs);
2025 statfs_pack(sfs, ksfs);
2026 if (unlikely(sb->s_flags & MS_RDONLY))
2027 sfs->os_state |= OS_STATE_READONLY;
2028 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb,
2029 LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2030 sfs->os_maxbytes = sb->s_maxbytes;
2032 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2035 * Reserve some space so to avoid fragmenting the filesystem too much.
2036 * Fragmentation not only impacts performance, but can also increase
2037 * metadata overhead significantly, causing grant calculation to be
2040 * Reserve 0.78% of total space, at least 8MB for small filesystems.
2042 CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2043 reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2044 if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2045 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2047 sfs->os_blocks -= reserved;
2048 sfs->os_bfree -= min(reserved, sfs->os_bfree);
2049 sfs->os_bavail -= min(reserved, sfs->os_bavail);
2052 if (unlikely(env == NULL))
2058 * Estimate space needed for file creations. We assume the largest filename
2059 * which is 2^64 - 1, hence a filename of 20 chars.
2060 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2062 #ifdef __LDISKFS_DIR_REC_LEN
2063 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2065 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2069 * Concurrency: doesn't access mutable data.
2071 static void osd_conf_get(const struct lu_env *env,
2072 const struct dt_device *dev,
2073 struct dt_device_param *param)
2075 struct super_block *sb = osd_sb(osd_dt_dev(dev));
2079 * XXX should be taken from not-yet-existing fs abstraction layer.
2081 param->ddp_max_name_len = LDISKFS_NAME_LEN;
2082 param->ddp_max_nlink = LDISKFS_LINK_MAX;
2083 param->ddp_symlink_max = sb->s_blocksize;
2084 param->ddp_mount_type = LDD_MT_LDISKFS;
2085 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2086 param->ddp_maxbytes = sb->s_maxbytes;
2088 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2089 /* inode are statically allocated, so per-inode space consumption
2090 * is the space consumed by the directory entry */
2091 param->ddp_inodespace = PER_OBJ_USAGE;
2092 /* EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2093 * = 128MB) which is unlikely to be hit in real life. Report a smaller
2094 * maximum length to not under count the actual number of extents
2095 * needed for writing a file. */
2096 param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2;
2097 /* worst-case extent insertion metadata overhead */
2098 param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2099 param->ddp_mntopts = 0;
2100 if (test_opt(sb, XATTR_USER))
2101 param->ddp_mntopts |= MNTOPT_USERXATTR;
2102 if (test_opt(sb, POSIX_ACL))
2103 param->ddp_mntopts |= MNTOPT_ACL;
2105 /* LOD might calculate the max stripe count based on max_ea_size,
2106 * so we need take account in the overhead as well,
2107 * xattr_header + magic + xattr_entry_head */
2108 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2109 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2111 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2112 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
2113 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2117 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2121 * Concurrency: shouldn't matter.
2123 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2127 CDEBUG(D_CACHE, "%s: syncing OSD\n", osd_dt_dev(d)->od_svname);
2129 rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
2131 CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2138 * Start commit for OSD device.
2140 * An implementation of dt_commit_async method for OSD device.
2141 * Asychronously starts underlayng fs sync and thereby a transaction
2144 * \param env environment
2145 * \param d dt device
2147 * \see dt_device_operations
2149 static int osd_commit_async(const struct lu_env *env,
2150 struct dt_device *d)
2152 struct super_block *s = osd_sb(osd_dt_dev(d));
2155 CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2156 RETURN(s->s_op->sync_fs(s, 0));
2160 * Concurrency: shouldn't matter.
2163 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2165 struct super_block *sb = osd_sb(osd_dt_dev(d));
2166 struct block_device *dev = sb->s_bdev;
2167 #ifdef HAVE_DEV_SET_RDONLY
2168 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2171 int rc = -EOPNOTSUPP;
2175 #ifdef HAVE_DEV_SET_RDONLY
2176 CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
2178 if (sb->s_op->freeze_fs) {
2179 rc = sb->s_op->freeze_fs(sb);
2184 if (jdev && (jdev != dev)) {
2185 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2187 dev_set_rdonly(jdev);
2189 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2190 dev_set_rdonly(dev);
2192 if (sb->s_op->unfreeze_fs)
2193 sb->s_op->unfreeze_fs(sb);
2198 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2199 osd_dt_dev(d)->od_svname, (long)dev, rc);
2205 * Note: we do not count into QUOTA here.
2206 * If we mount with --data_journal we may need more.
2208 const int osd_dto_credits_noquota[DTO_NR] = {
2211 * INDEX_EXTRA_TRANS_BLOCKS(8) +
2212 * SINGLEDATA_TRANS_BLOCKS(8)
2213 * XXX Note: maybe iam need more, since iam have more level than
2216 [DTO_INDEX_INSERT] = 16,
2219 * just modify a single entry, probably merge few within a block
2221 [DTO_INDEX_DELETE] = 1,
2225 [DTO_INDEX_UPDATE] = 16,
2227 * 4(inode, inode bits, groups, GDT)
2228 * notice: OI updates are counted separately with DTO_INDEX_INSERT
2230 [DTO_OBJECT_CREATE] = 4,
2232 * 4(inode, inode bits, groups, GDT)
2233 * notice: OI updates are counted separately with DTO_INDEX_DELETE
2235 [DTO_OBJECT_DELETE] = 4,
2237 * Attr set credits (inode)
2239 [DTO_ATTR_SET_BASE] = 1,
2241 * Xattr set. The same as xattr of EXT3.
2242 * DATA_TRANS_BLOCKS(14)
2243 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2244 * are also counted in. Do not know why?
2246 [DTO_XATTR_SET] = 14,
2248 * credits for inode change during write.
2250 [DTO_WRITE_BASE] = 3,
2252 * credits for single block write.
2254 [DTO_WRITE_BLOCK] = 14,
2256 * Attr set credits for chown.
2257 * This is extra credits for setattr, and it is null without quota
2259 [DTO_ATTR_SET_CHOWN] = 0
2262 static const struct dt_device_operations osd_dt_ops = {
2263 .dt_root_get = osd_root_get,
2264 .dt_statfs = osd_statfs,
2265 .dt_trans_create = osd_trans_create,
2266 .dt_trans_start = osd_trans_start,
2267 .dt_trans_stop = osd_trans_stop,
2268 .dt_trans_cb_add = osd_trans_cb_add,
2269 .dt_conf_get = osd_conf_get,
2270 .dt_sync = osd_sync,
2272 .dt_commit_async = osd_commit_async,
2275 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2278 struct osd_object *obj = osd_dt_obj(dt);
2279 struct osd_thread_info *oti = osd_oti_get(env);
2281 LINVRNT(osd_invariant(obj));
2283 LASSERT(obj->oo_owner != env);
2284 down_read_nested(&obj->oo_sem, role);
2286 LASSERT(obj->oo_owner == NULL);
2290 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2293 struct osd_object *obj = osd_dt_obj(dt);
2294 struct osd_thread_info *oti = osd_oti_get(env);
2296 LINVRNT(osd_invariant(obj));
2298 LASSERT(obj->oo_owner != env);
2299 down_write_nested(&obj->oo_sem, role);
2301 LASSERT(obj->oo_owner == NULL);
2302 obj->oo_owner = env;
2306 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2308 struct osd_object *obj = osd_dt_obj(dt);
2309 struct osd_thread_info *oti = osd_oti_get(env);
2311 LINVRNT(osd_invariant(obj));
2313 LASSERT(oti->oti_r_locks > 0);
2315 up_read(&obj->oo_sem);
2318 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2320 struct osd_object *obj = osd_dt_obj(dt);
2321 struct osd_thread_info *oti = osd_oti_get(env);
2323 LINVRNT(osd_invariant(obj));
2325 LASSERT(obj->oo_owner == env);
2326 LASSERT(oti->oti_w_locks > 0);
2328 obj->oo_owner = NULL;
2329 up_write(&obj->oo_sem);
2332 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2334 struct osd_object *obj = osd_dt_obj(dt);
2336 LINVRNT(osd_invariant(obj));
2338 return obj->oo_owner == env;
2341 static struct timespec *osd_inode_time(const struct lu_env *env,
2342 struct inode *inode, __u64 seconds)
2344 struct osd_thread_info *oti = osd_oti_get(env);
2345 struct timespec *t = &oti->oti_time;
2347 t->tv_sec = seconds;
2349 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2353 static void osd_inode_getattr(const struct lu_env *env,
2354 struct inode *inode, struct lu_attr *attr)
2356 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2357 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2358 LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2359 LA_BLKSIZE | LA_TYPE;
2361 attr->la_atime = LTIME_S(inode->i_atime);
2362 attr->la_mtime = LTIME_S(inode->i_mtime);
2363 attr->la_ctime = LTIME_S(inode->i_ctime);
2364 attr->la_mode = inode->i_mode;
2365 attr->la_size = i_size_read(inode);
2366 attr->la_blocks = inode->i_blocks;
2367 attr->la_uid = i_uid_read(inode);
2368 attr->la_gid = i_gid_read(inode);
2369 attr->la_projid = i_projid_read(inode);
2370 attr->la_flags = ll_inode_to_ext_flags(inode->i_flags);
2371 attr->la_nlink = inode->i_nlink;
2372 attr->la_rdev = inode->i_rdev;
2373 attr->la_blksize = 1 << inode->i_blkbits;
2374 attr->la_blkbits = inode->i_blkbits;
2376 * Ext4 did not transfer inherit flags from raw inode
2377 * to inode flags, and ext4 internally test raw inode
2378 * @i_flags directly. Instead of patching ext4, we do it here.
2380 if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2381 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2384 static int osd_attr_get(const struct lu_env *env,
2385 struct dt_object *dt,
2386 struct lu_attr *attr)
2388 struct osd_object *obj = osd_dt_obj(dt);
2390 if (unlikely(!dt_object_exists(dt)))
2392 if (unlikely(obj->oo_destroyed))
2395 LASSERT(!dt_object_remote(dt));
2396 LINVRNT(osd_invariant(obj));
2398 spin_lock(&obj->oo_guard);
2399 osd_inode_getattr(env, obj->oo_inode, attr);
2400 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2401 attr->la_flags |= LUSTRE_ORPHAN_FL;
2402 spin_unlock(&obj->oo_guard);
2407 static int osd_declare_attr_qid(const struct lu_env *env,
2408 struct osd_object *obj,
2409 struct osd_thandle *oh, long long bspace,
2410 qid_t old_id, qid_t new_id, bool enforce,
2414 struct osd_thread_info *info = osd_oti_get(env);
2415 struct lquota_id_info *qi = &info->oti_qi;
2417 qi->lqi_type = type;
2418 /* inode accounting */
2419 qi->lqi_is_blk = false;
2421 /* one more inode for the new id ... */
2422 qi->lqi_id.qid_uid = new_id;
2424 /* Reserve credits for the new id */
2425 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2426 if (rc == -EDQUOT || rc == -EINPROGRESS)
2431 /* and one less inode for the current id */
2432 qi->lqi_id.qid_uid = old_id;
2434 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2435 if (rc == -EDQUOT || rc == -EINPROGRESS)
2440 /* block accounting */
2441 qi->lqi_is_blk = true;
2443 /* more blocks for the new id ... */
2444 qi->lqi_id.qid_uid = new_id;
2445 qi->lqi_space = bspace;
2447 * Credits for the new uid has been reserved, re-use "obj"
2448 * to save credit reservation.
2450 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2451 if (rc == -EDQUOT || rc == -EINPROGRESS)
2456 /* and finally less blocks for the current uid */
2457 qi->lqi_id.qid_uid = old_id;
2458 qi->lqi_space = -bspace;
2459 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2460 if (rc == -EDQUOT || rc == -EINPROGRESS)
2466 static int osd_declare_attr_set(const struct lu_env *env,
2467 struct dt_object *dt,
2468 const struct lu_attr *attr,
2469 struct thandle *handle)
2471 struct osd_thandle *oh;
2472 struct osd_object *obj;
2480 LASSERT(dt != NULL);
2481 LASSERT(handle != NULL);
2483 obj = osd_dt_obj(dt);
2484 LASSERT(osd_invariant(obj));
2486 oh = container_of0(handle, struct osd_thandle, ot_super);
2487 LASSERT(oh->ot_handle == NULL);
2489 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2490 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2492 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2493 osd_dto_credits_noquota[DTO_XATTR_SET]);
2495 if (attr == NULL || obj->oo_inode == NULL)
2498 bspace = obj->oo_inode->i_blocks;
2499 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
2500 bspace = toqb(bspace);
2502 /* Changing ownership is always preformed by super user, it should not
2505 * We still need to call the osd_declare_qid() to calculate the journal
2506 * credits for updating quota accounting files and to trigger quota
2507 * space adjustment once the operation is completed.*/
2508 if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2510 uid = i_uid_read(obj->oo_inode);
2511 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2512 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2513 attr->la_uid, enforce, USRQUOTA);
2517 gid = i_gid_read(obj->oo_inode);
2518 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2519 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2520 i_gid_read(obj->oo_inode),
2521 attr->la_gid, enforce, GRPQUOTA);
2526 #ifdef HAVE_PROJECT_QUOTA
2527 if (attr->la_valid & LA_PROJID) {
2528 __u32 projid = i_projid_read(obj->oo_inode);
2529 enforce = (attr->la_valid & LA_PROJID) &&
2530 (attr->la_projid != projid);
2531 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2532 (qid_t)projid, (qid_t)attr->la_projid,
2541 static int osd_inode_setattr(const struct lu_env *env,
2542 struct inode *inode, const struct lu_attr *attr)
2544 __u64 bits = attr->la_valid;
2546 /* Only allow set size for regular file */
2547 if (!S_ISREG(inode->i_mode))
2548 bits &= ~(LA_SIZE | LA_BLOCKS);
2553 if (bits & LA_ATIME)
2554 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
2555 if (bits & LA_CTIME)
2556 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
2557 if (bits & LA_MTIME)
2558 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
2559 if (bits & LA_SIZE) {
2560 spin_lock(&inode->i_lock);
2561 LDISKFS_I(inode)->i_disksize = attr->la_size;
2562 i_size_write(inode, attr->la_size);
2563 spin_unlock(&inode->i_lock);
2566 /* OSD should not change "i_blocks" which is used by quota.
2567 * "i_blocks" should be changed by ldiskfs only. */
2569 inode->i_mode = (inode->i_mode & S_IFMT) |
2570 (attr->la_mode & ~S_IFMT);
2572 i_uid_write(inode, attr->la_uid);
2574 i_gid_write(inode, attr->la_gid);
2575 if (bits & LA_PROJID)
2576 i_projid_write(inode, attr->la_projid);
2577 if (bits & LA_NLINK)
2578 set_nlink(inode, attr->la_nlink);
2580 inode->i_rdev = attr->la_rdev;
2582 if (bits & LA_FLAGS) {
2583 /* always keep S_NOCMTIME */
2584 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2587 * Ext4 did not transfer inherit flags from
2588 * @inode->i_flags to raw inode i_flags when writing
2589 * flags, we do it explictly here.
2591 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2592 LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2597 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2601 if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2602 (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2605 ll_vfs_dq_init(inode);
2607 if (attr->la_valid & LA_UID)
2608 iattr.ia_valid |= ATTR_UID;
2609 if (attr->la_valid & LA_GID)
2610 iattr.ia_valid |= ATTR_GID;
2611 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2612 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2614 rc = ll_vfs_dq_transfer(inode, &iattr);
2616 CERROR("%s: quota transfer failed: rc = %d. Is quota "
2617 "enforcement enabled on the ldiskfs "
2618 "filesystem?\n", inode->i_sb->s_id, rc);
2623 #ifdef HAVE_PROJECT_QUOTA
2624 /* Handle project id transfer here properly */
2625 if (attr->la_valid & LA_PROJID &&
2626 attr->la_projid != i_projid_read(inode)) {
2627 rc = __ldiskfs_ioctl_setproject(inode, attr->la_projid);
2629 CERROR("%s: quota transfer failed: rc = %d. Is quota "
2630 "enforcement enabled on the ldiskfs "
2631 "filesystem?\n", inode->i_sb->s_id, rc);
2639 static int osd_attr_set(const struct lu_env *env,
2640 struct dt_object *dt,
2641 const struct lu_attr *attr,
2642 struct thandle *handle)
2644 struct osd_object *obj = osd_dt_obj(dt);
2645 struct inode *inode;
2648 if (!dt_object_exists(dt))
2651 LASSERT(handle != NULL);
2652 LASSERT(!dt_object_remote(dt));
2653 LASSERT(osd_invariant(obj));
2655 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2657 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
2658 struct osd_thread_info *oti = osd_oti_get(env);
2659 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
2660 struct lu_fid *fid1 = &oti->oti_fid;
2661 struct osd_inode_id *id = &oti->oti_id;
2662 struct iam_path_descr *ipd;
2663 struct iam_container *bag;
2664 struct osd_thandle *oh;
2667 fid_cpu_to_be(fid1, fid0);
2668 memset(id, 1, sizeof(*id));
2669 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2670 fid0)->oi_dir.od_container;
2671 ipd = osd_idx_ipd_get(env, bag);
2672 if (unlikely(ipd == NULL))
2675 oh = container_of0(handle, struct osd_thandle, ot_super);
2676 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2677 (const struct iam_rec *)id, ipd);
2678 osd_ipd_put(env, bag, ipd);
2679 return(rc > 0 ? 0 : rc);
2682 inode = obj->oo_inode;
2684 rc = osd_quota_transfer(inode, attr);
2688 spin_lock(&obj->oo_guard);
2689 rc = osd_inode_setattr(env, inode, attr);
2690 spin_unlock(&obj->oo_guard);
2694 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2696 if (!(attr->la_valid & LA_FLAGS))
2699 /* Let's check if there are extra flags need to be set into LMA */
2700 if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
2701 struct osd_thread_info *info = osd_oti_get(env);
2702 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
2704 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
2705 &info->oti_ost_attrs);
2709 lma->lma_incompat |=
2710 lustre_to_lma_flags(attr->la_flags);
2711 lustre_lma_swab(lma);
2712 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
2713 lma, sizeof(*lma), XATTR_REPLACE);
2715 struct osd_device *osd = osd_obj2dev(obj);
2717 CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
2718 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
2719 lma->lma_incompat, rc);
2722 attr->la_flags & LUSTRE_LMA_FL_MASKS;
2724 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2727 osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2732 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2733 struct osd_object *obj,
2734 const char *name, const int namelen)
2736 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2739 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2740 umode_t mode, struct dt_allocation_hint *hint,
2744 struct osd_device *osd = osd_obj2dev(obj);
2745 struct osd_thandle *oth;
2746 struct dt_object *parent = NULL;
2747 struct inode *inode;
2749 LINVRNT(osd_invariant(obj));
2750 LASSERT(obj->oo_inode == NULL);
2751 LASSERT(obj->oo_hl_head == NULL);
2753 if (S_ISDIR(mode) && ldiskfs_pdo) {
2754 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2755 if (obj->oo_hl_head == NULL)
2759 oth = container_of(th, struct osd_thandle, ot_super);
2760 LASSERT(oth->ot_handle->h_transaction != NULL);
2762 if (hint != NULL && hint->dah_parent != NULL &&
2763 !dt_object_remote(hint->dah_parent))
2764 parent = hint->dah_parent;
2766 inode = ldiskfs_create_inode(oth->ot_handle,
2767 parent ? osd_dt_obj(parent)->oo_inode :
2768 osd_sb(osd)->s_root->d_inode,
2770 if (!IS_ERR(inode)) {
2771 /* Do not update file c/mtime in ldiskfs. */
2772 inode->i_flags |= S_NOCMTIME;
2774 /* For new created object, it must be consistent,
2775 * and it is unnecessary to scrub against it. */
2776 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2778 obj->oo_inode = inode;
2781 if (obj->oo_hl_head != NULL) {
2782 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2783 obj->oo_hl_head = NULL;
2785 result = PTR_ERR(inode);
2787 LINVRNT(osd_invariant(obj));
2795 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2796 struct lu_attr *attr,
2797 struct dt_allocation_hint *hint,
2798 struct dt_object_format *dof,
2802 struct osd_thandle *oth;
2803 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
2805 LASSERT(S_ISDIR(attr->la_mode));
2807 oth = container_of(th, struct osd_thandle, ot_super);
2808 LASSERT(oth->ot_handle->h_transaction != NULL);
2809 result = osd_mkfile(info, obj, mode, hint, th);
2814 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2815 struct lu_attr *attr,
2816 struct dt_allocation_hint *hint,
2817 struct dt_object_format *dof,
2821 struct osd_thandle *oth;
2822 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2824 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2826 LASSERT(S_ISREG(attr->la_mode));
2828 oth = container_of(th, struct osd_thandle, ot_super);
2829 LASSERT(oth->ot_handle->h_transaction != NULL);
2831 result = osd_mkfile(info, obj, mode, hint, th);
2833 LASSERT(obj->oo_inode != NULL);
2834 if (feat->dif_flags & DT_IND_VARKEY)
2835 result = iam_lvar_create(obj->oo_inode,
2836 feat->dif_keysize_max,
2838 feat->dif_recsize_max,
2841 result = iam_lfix_create(obj->oo_inode,
2842 feat->dif_keysize_max,
2844 feat->dif_recsize_max,
2851 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2852 struct lu_attr *attr,
2853 struct dt_allocation_hint *hint,
2854 struct dt_object_format *dof,
2857 LASSERT(S_ISREG(attr->la_mode));
2858 return osd_mkfile(info, obj, (attr->la_mode &
2859 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2862 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2863 struct lu_attr *attr,
2864 struct dt_allocation_hint *hint,
2865 struct dt_object_format *dof,
2868 LASSERT(S_ISLNK(attr->la_mode));
2869 return osd_mkfile(info, obj, (attr->la_mode &
2870 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2873 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2874 struct lu_attr *attr,
2875 struct dt_allocation_hint *hint,
2876 struct dt_object_format *dof,
2879 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2882 LINVRNT(osd_invariant(obj));
2883 LASSERT(obj->oo_inode == NULL);
2884 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2885 S_ISFIFO(mode) || S_ISSOCK(mode));
2887 result = osd_mkfile(info, obj, mode, hint, th);
2889 LASSERT(obj->oo_inode != NULL);
2891 * This inode should be marked dirty for i_rdev. Currently
2892 * that is done in the osd_attr_init().
2894 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2897 LINVRNT(osd_invariant(obj));
2901 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2903 struct dt_allocation_hint *hint,
2904 struct dt_object_format *dof,
2907 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2909 osd_obj_type_f result;
2925 result = osd_mk_index;
2936 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2937 struct dt_object *parent, struct dt_object *child,
2942 ah->dah_parent = parent;
2943 ah->dah_mode = child_mode;
2945 if (parent != NULL && !dt_object_remote(parent)) {
2946 /* will help to find FID->ino at dt_insert("..") */
2947 struct osd_object *pobj = osd_dt_obj(parent);
2948 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
2952 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2953 struct lu_attr *attr, struct dt_object_format *dof)
2955 struct inode *inode = obj->oo_inode;
2956 __u64 valid = attr->la_valid;
2959 attr->la_valid &= ~(LA_TYPE | LA_MODE);
2961 if (dof->dof_type != DFT_NODE)
2962 attr->la_valid &= ~LA_RDEV;
2963 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2964 attr->la_valid &= ~LA_ATIME;
2965 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2966 attr->la_valid &= ~LA_CTIME;
2967 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2968 attr->la_valid &= ~LA_MTIME;
2970 result = osd_quota_transfer(inode, attr);
2974 if (attr->la_valid != 0) {
2975 result = osd_inode_setattr(info->oti_env, inode, attr);
2977 * The osd_inode_setattr() should always succeed here. The
2978 * only error that could be returned is EDQUOT when we are
2979 * trying to change the UID or GID of the inode. However, this
2980 * should not happen since quota enforcement is no longer
2981 * enabled on ldiskfs (lquota takes care of it).
2983 LASSERTF(result == 0, "%d\n", result);
2984 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2987 attr->la_valid = valid;
2991 * Helper function for osd_create()
2993 * \retval 0, on success
2995 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
2996 struct lu_attr *attr, struct dt_allocation_hint *hint,
2997 struct dt_object_format *dof, struct thandle *th)
3002 osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3004 /* we drop umask so that permissions we pass are not affected */
3005 umask = current->fs->umask;
3006 current->fs->umask = 0;
3008 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3010 if (likely(obj->oo_inode != NULL)) {
3011 LASSERT(obj->oo_inode->i_state & I_NEW);
3013 /* Unlock the inode before attr initialization to avoid
3014 * unnecessary dqget operations. LU-6378 */
3015 unlock_new_inode(obj->oo_inode);
3018 if (likely(result == 0)) {
3019 osd_attr_init(info, obj, attr, dof);
3020 osd_object_init0(obj);
3023 /* restore previous umask value */
3024 current->fs->umask = umask;
3026 osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3032 * Helper function for osd_create()
3034 * \retval 0, on success
3036 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3037 const struct lu_fid *fid, struct thandle *th)
3039 struct osd_thread_info *info = osd_oti_get(env);
3040 struct osd_inode_id *id = &info->oti_id;
3041 struct osd_device *osd = osd_obj2dev(obj);
3042 struct osd_thandle *oh;
3045 LASSERT(obj->oo_inode != NULL);
3047 oh = container_of0(th, struct osd_thandle, ot_super);
3048 LASSERT(oh->ot_handle);
3049 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3051 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3052 rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3053 OI_CHECK_FLD, NULL);
3054 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3059 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3060 u64 seq, struct lu_seq_range *range)
3062 struct seq_server_site *ss = osd_seq_site(osd);
3064 if (fid_seq_is_idif(seq)) {
3065 fld_range_set_ost(range);
3066 range->lsr_index = idif_ost_idx(seq);
3070 if (!fid_seq_in_fldb(seq)) {
3071 fld_range_set_mdt(range);
3073 /* FIXME: If ss is NULL, it suppose not get lsr_index
3075 range->lsr_index = ss->ss_node_id;
3079 LASSERT(ss != NULL);
3080 fld_range_set_any(range);
3081 /* OSD will only do local fld lookup */
3082 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3085 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3086 struct lu_attr *attr,
3087 struct dt_allocation_hint *hint,
3088 struct dt_object_format *dof,
3089 struct thandle *handle)
3091 struct osd_thandle *oh;
3095 LASSERT(handle != NULL);
3097 oh = container_of0(handle, struct osd_thandle, ot_super);
3098 LASSERT(oh->ot_handle == NULL);
3100 /* EA object consumes more credits than regular object: osd_mk_index
3101 * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3102 * leaf_node, could involves the block, block bitmap, groups, GDT
3103 * change for each block, so add 4 * 2 credits in that case. */
3104 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3105 osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3106 (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3107 /* Reuse idle OI block may cause additional one OI block
3109 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3110 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3115 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3116 attr->la_projid, 1, oh, osd_dt_obj(dt),
3117 NULL, OSD_QID_INODE);
3121 /* will help to find FID->ino mapping at dt_insert() */
3122 rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3129 * Called to destroy on-disk representation of the object
3131 * Concurrency: must be locked
3133 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3136 struct osd_object *obj = osd_dt_obj(dt);
3137 struct inode *inode = obj->oo_inode;
3138 struct osd_thandle *oh;
3145 oh = container_of0(th, struct osd_thandle, ot_super);
3146 LASSERT(oh->ot_handle == NULL);
3148 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3149 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3150 /* Recycle idle OI leaf may cause additional three OI blocks
3152 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3153 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3154 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3155 /* one less inode */
3156 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3157 i_projid_read(inode), -1, oh, obj, NULL,
3161 /* data to be truncated */
3162 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3163 i_projid_read(inode), 0, oh, obj, NULL,
3168 /* will help to find FID->ino when this object is being
3169 * added to PENDING/ */
3170 rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3175 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3178 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3179 struct osd_object *obj = osd_dt_obj(dt);
3180 struct inode *inode = obj->oo_inode;
3181 struct osd_device *osd = osd_obj2dev(obj);
3182 struct osd_thandle *oh;
3186 oh = container_of0(th, struct osd_thandle, ot_super);
3187 LASSERT(oh->ot_handle);
3189 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3191 if (unlikely(fid_is_acct(fid)))
3194 if (S_ISDIR(inode->i_mode)) {
3195 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3196 inode->i_nlink == 2);
3197 /* it will check/delete the inode from remote parent,
3198 * how to optimize it? unlink performance impaction XXX */
3199 result = osd_delete_from_remote_parent(env, osd, obj, oh);
3201 CERROR("%s: delete inode "DFID": rc = %d\n",
3202 osd_name(osd), PFID(fid), result);
3204 spin_lock(&obj->oo_guard);
3206 spin_unlock(&obj->oo_guard);
3207 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3210 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3212 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3214 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3215 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3216 oh->ot_handle, OI_CHECK_FLD);
3218 osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3219 /* XXX: add to ext3 orphan list */
3220 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3222 /* not needed in the cache anymore */
3223 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3224 obj->oo_destroyed = 1;
3230 * Put the fid into lustre_mdt_attrs, and then place the structure
3231 * inode's ea. This fid should not be altered during the life time
3234 * \retval +ve, on success
3235 * \retval -ve, on error
3237 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3239 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3240 const struct lu_fid *fid, __u32 compat, __u32 incompat)
3242 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3243 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3247 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3250 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3253 lustre_loa_init(loa, fid, compat, incompat);
3254 lustre_loa_swab(loa, false);
3256 /* For the OST device with 256 bytes inode size by default,
3257 * the PFID EA will be stored together with LMA EA to avoid
3258 * performance trouble. Otherwise the PFID EA can be stored
3259 * independently. LU-8998 */
3260 if ((compat & LMAC_FID_ON_OST) &&
3261 LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3262 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3263 sizeof(*loa), XATTR_CREATE);
3265 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3266 sizeof(*lma), XATTR_CREATE);
3267 /* LMA may already exist, but we need to check that all the
3268 * desired compat/incompat flags have been added. */
3269 if (unlikely(rc == -EEXIST)) {
3270 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3271 XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3275 if (rc < sizeof(*lma))
3278 lustre_loa_swab(loa, true);
3279 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3280 ((compat == 0 && incompat == 0) ||
3281 (!(~lma->lma_compat & compat) &&
3282 !(~lma->lma_incompat & incompat))))
3285 lma->lma_self_fid = *fid;
3286 lma->lma_compat |= compat;
3287 lma->lma_incompat |= incompat;
3288 if (rc == sizeof(*lma)) {
3289 lustre_lma_swab(lma);
3290 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3291 sizeof(*lma), XATTR_REPLACE);
3293 lustre_loa_swab(loa, false);
3294 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3295 sizeof(*loa), XATTR_REPLACE);
3303 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3304 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3305 * To have compatilibility with 1.8 ldiskfs driver we need to have
3306 * magic number at start of fid data.
3307 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3310 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3311 const struct lu_fid *fid)
3313 if (!fid_is_namespace_visible(fid) ||
3314 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3315 param->edp_magic = 0;
3319 param->edp_magic = LDISKFS_LUFID_MAGIC;
3320 param->edp_len = sizeof(struct lu_fid) + 1;
3321 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3325 * Try to read the fid from inode ea into dt_rec.
3327 * \param fid object fid.
3329 * \retval 0 on success
3331 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3332 __u32 ino, struct lu_fid *fid,
3333 struct osd_inode_id *id)
3335 struct osd_thread_info *info = osd_oti_get(env);
3336 struct inode *inode;
3339 osd_id_gen(id, ino, OSD_OII_NOGEN);
3340 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3342 RETURN(PTR_ERR(inode));
3348 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3350 struct inode *parent_dir,
3351 const struct lu_fid *dot_fid,
3352 const struct lu_fid *dot_dot_fid,
3353 struct osd_thandle *oth)
3355 struct ldiskfs_dentry_param *dot_ldp;
3356 struct ldiskfs_dentry_param *dot_dot_ldp;
3357 __u32 saved_nlink = dir->i_nlink;
3360 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3361 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3363 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3364 dot_ldp->edp_magic = 0;
3366 rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3367 dir, dot_ldp, dot_dot_ldp);
3368 /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3369 * the subseqent ref_add() will increase the dir->i_nlink
3370 * as 3. That is incorrect for new created directory.
3372 * It looks like hack, because we want to make the OSD API
3373 * to be order-independent for new created directory object
3374 * between dt_insert(..) and ref_add() operations.
3376 * Here, we only restore the in-RAM dir-inode's nlink attr,
3377 * becuase if the nlink attr is not 2, then there will be
3378 * ref_add() called following the dt_insert(..), such call
3379 * will make both the in-RAM and on-disk dir-inode's nlink
3380 * attr to be set as 2. LU-7447 */
3381 set_nlink(dir, saved_nlink);
3386 * Create an local agent inode for remote entry
3388 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3389 struct osd_device *osd,
3390 struct osd_object *pobj,
3391 const struct lu_fid *fid,
3395 struct osd_thread_info *info = osd_oti_get(env);
3396 struct inode *local;
3397 struct osd_thandle *oh;
3402 oh = container_of(th, struct osd_thandle, ot_super);
3403 LASSERT(oh->ot_handle->h_transaction != NULL);
3405 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3406 if (IS_ERR(local)) {
3407 CERROR("%s: create local error %d\n", osd_name(osd),
3408 (int)PTR_ERR(local));
3412 /* restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3413 * correct gid on remote file, not agent here */
3414 local->i_gid = current_fsgid();
3415 ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3416 unlock_new_inode(local);
3418 /* Agent inode should not have project ID */
3419 #ifdef HAVE_PROJECT_QUOTA
3420 if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL) {
3421 rc = __ldiskfs_ioctl_setproject(local, 0);
3423 CERROR("%s: quota transfer failed: rc = %d. Is project "
3424 "quota enforcement enabled on the ldiskfs "
3425 "filesystem?\n", local->i_sb->s_id, rc);
3426 RETURN(ERR_PTR(rc));
3430 /* Set special LMA flag for local agent inode */
3431 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3433 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3434 osd_name(osd), PFID(fid), rc);
3435 RETURN(ERR_PTR(rc));
3441 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3442 lu_object_fid(&pobj->oo_dt.do_lu),
3445 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3446 osd_name(osd), PFID(fid), rc);
3447 RETURN(ERR_PTR(rc));
3454 * when direntry is deleted, we have to take care of possible agent inode
3455 * referenced by that. unfortunately we can't do this at that point:
3456 * iget() within a running transaction leads to deadlock and we better do
3457 * not call that every delete declaration to save performance. so we put
3458 * a potention agent inode on a list and process that once the transaction
3459 * is over. Notice it's not any worse in terms of real orphans as regular
3460 * object destroy doesn't put inodes on the on-disk orphan list. this should
3461 * be addressed separately
3463 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3464 struct osd_thandle *oh,
3467 struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3468 struct osd_obj_orphan *oor;
3475 oor->oor_env = (struct lu_env *)env;
3476 spin_lock(&osd->od_osfs_lock);
3477 list_add(&oor->oor_list, &osd->od_orphan_list);
3478 spin_unlock(&osd->od_osfs_lock);
3480 oh->ot_remove_agents = 1;
3486 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3487 struct osd_device *osd)
3489 struct osd_thread_info *info = osd_oti_get(env);
3490 struct osd_obj_orphan *oor, *tmp;
3491 struct osd_inode_id id;
3492 struct list_head list;
3493 struct inode *inode;
3498 INIT_LIST_HEAD(&list);
3500 spin_lock(&osd->od_osfs_lock);
3501 list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3502 if (oor->oor_env == env) {
3503 list_del(&oor->oor_list);
3504 list_add(&oor->oor_list, &list);
3507 spin_unlock(&osd->od_osfs_lock);
3509 list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3513 list_del(&oor->oor_list);
3516 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3517 inode = osd_iget_fid(info, osd, &id, &fid);
3521 if (!osd_remote_fid(env, osd, &fid)) {
3526 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3528 mark_inode_dirty(inode);
3529 ldiskfs_journal_stop(jh);
3537 * OSD layer object create function for OST objects (b=11826).
3539 * The FID is inserted into inode xattr here.
3541 * \retval 0, on success
3542 * \retval -ve, on error
3544 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3545 struct lu_attr *attr, struct dt_allocation_hint *hint,
3546 struct dt_object_format *dof, struct thandle *th)
3548 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3549 struct osd_object *obj = osd_dt_obj(dt);
3550 struct osd_thread_info *info = osd_oti_get(env);
3551 int result, on_ost = 0;
3555 if (dt_object_exists(dt))
3558 LINVRNT(osd_invariant(obj));
3559 LASSERT(!dt_object_remote(dt));
3560 LASSERT(osd_is_write_locked(env, obj));
3561 LASSERT(th != NULL);
3563 if (unlikely(fid_is_acct(fid)))
3564 /* Quota files can't be created from the kernel any more,
3565 * 'tune2fs -O quota' will take care of creating them */
3568 result = __osd_create(info, obj, attr, hint, dof, th);
3570 if (fid_is_idif(fid) &&
3571 !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3572 struct lu_fid *tfid = &info->oti_fid;
3573 struct ost_id *oi = &info->oti_ostid;
3575 fid_to_ostid(fid, oi);
3576 ostid_to_fid(tfid, oi, 0);
3578 result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3579 LMAC_FID_ON_OST, 0);
3581 on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3583 result = osd_ea_fid_set(info, obj->oo_inode, fid,
3584 on_ost ? LMAC_FID_ON_OST : 0,
3587 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3588 obj->oo_dt.do_body_ops = &osd_body_ops;
3592 result = __osd_oi_insert(env, obj, fid, th);
3594 /* a small optimization - dt_insert() isn't usually applied
3595 * to OST objects, so we don't need to cache OI mapping for
3597 if (result == 0 && on_ost == 0) {
3598 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3599 result = osd_idc_find_and_init(env, osd, obj);
3600 LASSERT(result == 0);
3603 LASSERT(ergo(result == 0,
3604 dt_object_exists(dt) && !dt_object_remote(dt)));
3605 LINVRNT(osd_invariant(obj));
3609 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
3610 struct thandle *handle)
3612 struct osd_thandle *oh;
3614 /* it's possible that object doesn't exist yet */
3615 LASSERT(handle != NULL);
3617 oh = container_of0(handle, struct osd_thandle, ot_super);
3618 LASSERT(oh->ot_handle == NULL);
3620 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3621 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3627 * Concurrency: @dt is write locked.
3629 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
3632 struct osd_object *obj = osd_dt_obj(dt);
3633 struct inode *inode = obj->oo_inode;
3634 struct osd_thandle *oh;
3637 if (!dt_object_exists(dt) || obj->oo_destroyed)
3640 LINVRNT(osd_invariant(obj));
3641 LASSERT(!dt_object_remote(dt));
3642 LASSERT(osd_is_write_locked(env, obj));
3643 LASSERT(th != NULL);
3645 oh = container_of0(th, struct osd_thandle, ot_super);
3646 LASSERT(oh->ot_handle != NULL);
3648 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
3650 CDEBUG(D_INODE, DFID" increase nlink %d\n",
3651 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3653 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
3654 * (65000) subdirectories by storing "1" in i_nlink if the link count
3655 * would otherwise overflow. Directory tranversal tools understand
3656 * that (st_nlink == 1) indicates that the filesystem dose not track
3657 * hard links count on the directory, and will not abort subdirectory
3658 * scanning early once (st_nlink - 2) subdirs have been found.
3660 * This also has to properly handle the case of inodes with nlink == 0
3661 * in case they are being linked into the PENDING directory
3663 spin_lock(&obj->oo_guard);
3664 if (unlikely(inode->i_nlink == 0))
3665 /* inc_nlink from 0 may cause WARN_ON */
3666 set_nlink(inode, 1);
3668 ldiskfs_inc_count(oh->ot_handle, inode);
3669 if (!S_ISDIR(inode->i_mode))
3670 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
3672 spin_unlock(&obj->oo_guard);
3674 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3675 LINVRNT(osd_invariant(obj));
3677 osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
3682 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
3683 struct thandle *handle)
3685 struct osd_thandle *oh;
3687 if (!dt_object_exists(dt))
3690 LASSERT(!dt_object_remote(dt));
3691 LASSERT(handle != NULL);
3693 oh = container_of0(handle, struct osd_thandle, ot_super);
3694 LASSERT(oh->ot_handle == NULL);
3696 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
3697 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3703 * Concurrency: @dt is write locked.
3705 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
3708 struct osd_object *obj = osd_dt_obj(dt);
3709 struct inode *inode = obj->oo_inode;
3710 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3711 struct osd_thandle *oh;
3713 if (!dt_object_exists(dt))
3716 LINVRNT(osd_invariant(obj));
3717 LASSERT(!dt_object_remote(dt));
3718 LASSERT(osd_is_write_locked(env, obj));
3719 LASSERT(th != NULL);
3721 oh = container_of0(th, struct osd_thandle, ot_super);
3722 LASSERT(oh->ot_handle != NULL);
3724 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
3726 spin_lock(&obj->oo_guard);
3727 /* That can be result of upgrade from old Lustre version and
3728 * applied only to local files. Just skip this ref_del call.
3729 * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
3730 if (inode->i_nlink == 0) {
3731 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
3732 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
3733 ", maybe an upgraded file? (LU-3915)\n",
3734 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
3735 spin_unlock(&obj->oo_guard);
3739 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
3740 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3742 ldiskfs_dec_count(oh->ot_handle, inode);
3743 spin_unlock(&obj->oo_guard);
3745 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3746 LINVRNT(osd_invariant(obj));
3748 osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
3754 * Concurrency: @dt is read locked.
3756 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
3757 struct lu_buf *buf, const char *name)
3759 struct osd_object *obj = osd_dt_obj(dt);
3760 struct inode *inode = obj->oo_inode;
3761 struct osd_thread_info *info = osd_oti_get(env);
3762 struct dentry *dentry = &info->oti_obj_dentry;
3763 bool cache_xattr = false;
3768 /* version get is not real XATTR but uses xattr API */
3769 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3770 dt_obj_version_t *ver = buf->lb_buf;
3772 /* for version we are just using xattr API but change inode
3774 if (buf->lb_len == 0)
3775 return sizeof(dt_obj_version_t);
3777 if (buf->lb_len < sizeof(dt_obj_version_t))
3780 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
3781 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3783 *ver = LDISKFS_I(inode)->i_fs_version;
3785 return sizeof(dt_obj_version_t);
3788 if (!dt_object_exists(dt))
3791 LASSERT(!dt_object_remote(dt));
3792 LASSERT(inode->i_op != NULL);
3793 LASSERT(inode->i_op->getxattr != NULL);
3795 if (strcmp(name, XATTR_NAME_LOV) == 0 ||
3796 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
3800 rc = osd_oxc_get(obj, name, buf);
3805 rc = __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
3806 if (rc == -ENODATA && strcmp(name, XATTR_NAME_FID) == 0) {
3807 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3808 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3809 struct filter_fid *ff;
3810 struct ost_layout *ol;
3812 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
3814 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
3818 if (!(lma->lma_compat & LMAC_STRIPE_INFO)) {
3824 if (buf->lb_len == 0 || !buf->lb_buf)
3827 if (buf->lb_len < rc)
3831 ol = &ff->ff_layout;
3832 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
3833 PFID_STRIPE_IDX_BITS);
3834 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
3835 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
3836 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
3837 if (lma->lma_compat & LMAC_COMP_INFO) {
3838 ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
3839 ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
3840 ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
3842 ol->ol_comp_start = 0;
3843 ol->ol_comp_end = 0;
3850 if (rc == -ENOENT || rc == -ENODATA)
3851 osd_oxc_add(obj, name, NULL, 0);
3852 else if (rc > 0 && buf->lb_buf != NULL)
3853 osd_oxc_add(obj, name, buf->lb_buf, rc);
3859 static int osd_declare_xattr_set(const struct lu_env *env,
3860 struct dt_object *dt,
3861 const struct lu_buf *buf, const char *name,
3862 int fl, struct thandle *handle)
3864 struct osd_thandle *oh;
3866 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3868 LASSERT(handle != NULL);
3870 oh = container_of0(handle, struct osd_thandle, ot_super);
3871 LASSERT(oh->ot_handle == NULL);
3873 if (strcmp(name, XATTR_NAME_LMA) == 0) {
3874 /* For non-upgrading case, the LMA is set first and
3875 * usually fit inode. But for upgrade case, the LMA
3876 * may be in another separated EA block. */
3877 if (dt_object_exists(dt)) {
3878 if (fl == LU_XATTR_REPLACE)
3883 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3885 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3886 /* We may need to delete the old PFID EA. */
3887 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
3888 if (fl == LU_XATTR_REPLACE)
3895 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
3900 if (buf->lb_buf == NULL && dt_object_exists(dt)) {
3901 /* learn xattr size from osd_xattr_get if
3902 attribute has not been read yet */
3903 buflen = __osd_xattr_get(
3904 osd_dt_obj(dt)->oo_inode,
3905 &osd_oti_get(env)->oti_obj_dentry,
3910 buflen = buf->lb_len;
3913 if (buflen > sb->s_blocksize) {
3914 credits += osd_calc_bkmap_credits(
3916 (buflen + sb->s_blocksize - 1) >>
3917 sb->s_blocksize_bits);
3921 * xattr set may involve inode quota change, reserve credits for
3922 * dquot_initialize()
3924 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3927 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3933 * Concurrency: @dt is write locked.
3935 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3936 const struct lu_buf *buf, const char *name, int fl,
3937 struct thandle *handle)
3939 struct osd_object *obj = osd_dt_obj(dt);
3940 struct inode *inode = obj->oo_inode;
3941 struct osd_thread_info *info = osd_oti_get(env);
3942 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3943 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3952 /* version set is not real XATTR */
3953 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3954 dt_obj_version_t *version = buf->lb_buf;
3956 /* for version we are just using xattr API but change inode
3958 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3960 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
3961 *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3963 LDISKFS_I(inode)->i_fs_version = *version;
3964 /* Version is set after all inode operations are finished,
3965 * so we should mark it dirty here */
3966 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3971 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
3972 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3975 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3976 if (fl & LU_XATTR_REPLACE)
3977 fs_flags |= XATTR_REPLACE;
3979 if (fl & LU_XATTR_CREATE)
3980 fs_flags |= XATTR_CREATE;
3982 /* For the OST device with 256 bytes inode size by default,
3983 * the PFID EA will be stored together with LMA EA to avoid
3984 * performance trouble. Otherwise the PFID EA can be stored
3985 * independently. LU-8998 */
3986 if (strcmp(name, XATTR_NAME_FID) == 0 &&
3987 LDISKFS_INODE_SIZE(inode->i_sb) <= 256) {
3988 struct dentry *dentry = &info->oti_obj_dentry;
3989 struct filter_fid *ff;
3990 struct ost_layout *ol;
3993 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
3996 ol = &ff->ff_layout;
3997 /* Old client does not send stripe information, store
3998 * the PFID EA on disk directly. */
3999 if (buf->lb_len == sizeof(struct lu_fid) ||
4000 ol->ol_stripe_size == 0) {
4001 len = sizeof(struct lu_fid);
4005 if (buf->lb_len != sizeof(*ff))
4008 rc = osd_get_lma(info, inode, dentry, loa);
4009 if (unlikely(rc == -ENODATA)) {
4010 /* Usually for upgarding from old device */
4011 lustre_loa_init(loa, lu_object_fid(&dt->do_lu),
4012 LMAC_FID_ON_OST, 0);
4020 fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4021 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4022 PFID_STRIPE_IDX_BITS;
4023 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4024 lma->lma_compat |= LMAC_STRIPE_INFO;
4025 if (ol->ol_comp_id != 0) {
4026 loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4027 loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4028 loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4029 lma->lma_compat |= LMAC_COMP_INFO;
4032 lustre_loa_swab(loa, false);
4034 /* Remove old PFID EA entry firstly. */
4035 ll_vfs_dq_init(inode);
4036 rc = inode->i_op->removexattr(dentry, name);
4037 if (rc && rc != -ENODATA)
4040 /* Store the PFID EA inside the LMA EA. */
4041 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
4045 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4046 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4050 lma->lma_incompat |= LMAI_STRIPED;
4051 lustre_lma_swab(lma);
4052 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4053 sizeof(*lma), XATTR_REPLACE);
4059 rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4060 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4063 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4064 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4065 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4071 * Concurrency: @dt is read locked.
4073 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4074 const struct lu_buf *buf)
4076 struct osd_object *obj = osd_dt_obj(dt);
4077 struct inode *inode = obj->oo_inode;
4078 struct osd_thread_info *info = osd_oti_get(env);
4079 struct dentry *dentry = &info->oti_obj_dentry;
4081 if (!dt_object_exists(dt))
4084 LASSERT(!dt_object_remote(dt));
4085 LASSERT(inode->i_op != NULL);
4086 LASSERT(inode->i_op->listxattr != NULL);
4088 dentry->d_inode = inode;
4089 dentry->d_sb = inode->i_sb;
4090 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4093 static int osd_declare_xattr_del(const struct lu_env *env,
4094 struct dt_object *dt, const char *name,
4095 struct thandle *handle)
4097 struct osd_thandle *oh;
4098 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4100 LASSERT(!dt_object_remote(dt));
4101 LASSERT(handle != NULL);
4103 oh = container_of0(handle, struct osd_thandle, ot_super);
4104 LASSERT(oh->ot_handle == NULL);
4106 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4107 osd_dto_credits_noquota[DTO_XATTR_SET]);
4109 * xattr del may involve inode quota change, reserve credits for
4110 * dquot_initialize()
4112 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4118 * Concurrency: @dt is write locked.
4120 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4121 const char *name, struct thandle *handle)
4123 struct osd_object *obj = osd_dt_obj(dt);
4124 struct inode *inode = obj->oo_inode;
4125 struct osd_thread_info *info = osd_oti_get(env);
4126 struct dentry *dentry = &info->oti_obj_dentry;
4129 if (!dt_object_exists(dt))
4132 LASSERT(!dt_object_remote(dt));
4133 LASSERT(inode->i_op != NULL);
4134 LASSERT(inode->i_op->removexattr != NULL);
4135 LASSERT(handle != NULL);
4137 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4139 ll_vfs_dq_init(inode);
4140 dentry->d_inode = inode;
4141 dentry->d_sb = inode->i_sb;
4142 rc = inode->i_op->removexattr(dentry, name);
4143 if (rc == -ENODATA && strcmp(name, XATTR_NAME_FID) == 0) {
4144 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4146 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
4148 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4149 &info->oti_ost_attrs);
4151 if (!(lma->lma_compat & LMAC_STRIPE_INFO)) {
4156 lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4157 lustre_lma_swab(lma);
4158 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4159 sizeof(*lma), XATTR_REPLACE);
4164 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4167 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4168 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4169 osd_oxc_del(obj, name);
4174 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4175 __u64 start, __u64 end)
4177 struct osd_object *obj = osd_dt_obj(dt);
4178 struct inode *inode = obj->oo_inode;
4179 struct osd_thread_info *info = osd_oti_get(env);
4180 struct dentry *dentry = &info->oti_obj_dentry;
4181 struct file *file = &info->oti_file;
4186 dentry->d_inode = inode;
4187 dentry->d_sb = inode->i_sb;
4188 file->f_path.dentry = dentry;
4189 file->f_mapping = inode->i_mapping;
4190 file->f_op = inode->i_fop;
4191 set_file_inode(file, inode);
4193 rc = ll_vfs_fsync_range(file, start, end, 0);
4198 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4207 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4208 const struct dt_index_features *feat)
4210 struct iam_descr *descr;
4212 if (osd_object_is_root(o))
4213 return feat == &dt_directory_features;
4215 LASSERT(o->oo_dir != NULL);
4217 descr = o->oo_dir->od_container.ic_descr;
4218 if (feat == &dt_directory_features) {
4219 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4225 feat->dif_keysize_min <= descr->id_key_size &&
4226 descr->id_key_size <= feat->dif_keysize_max &&
4227 feat->dif_recsize_min <= descr->id_rec_size &&
4228 descr->id_rec_size <= feat->dif_recsize_max &&
4229 !(feat->dif_flags & (DT_IND_VARKEY |
4230 DT_IND_VARREC | DT_IND_NONUNQ)) &&
4231 ergo(feat->dif_flags & DT_IND_UPDATE,
4232 1 /* XXX check that object (and file system) is
4237 static int osd_iam_container_init(const struct lu_env *env,
4238 struct osd_object *obj,
4239 struct osd_directory *dir)
4241 struct iam_container *bag = &dir->od_container;
4244 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4248 result = iam_container_setup(bag);
4250 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4252 iam_container_fini(bag);
4259 * Concurrency: no external locking is necessary.
4261 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4262 const struct dt_index_features *feat)
4266 struct osd_object *obj = osd_dt_obj(dt);
4268 LINVRNT(osd_invariant(obj));
4270 if (osd_object_is_root(obj)) {
4271 dt->do_index_ops = &osd_index_ea_ops;
4273 } else if (feat == &dt_directory_features) {
4274 dt->do_index_ops = &osd_index_ea_ops;
4275 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4280 } else if (unlikely(feat == &dt_otable_features)) {
4281 dt->do_index_ops = &osd_otable_ops;
4283 } else if (unlikely(feat == &dt_acct_features)) {
4284 dt->do_index_ops = &osd_acct_index_ops;
4287 } else if (!osd_has_index(obj)) {
4288 struct osd_directory *dir;
4293 spin_lock(&obj->oo_guard);
4294 if (obj->oo_dir == NULL)
4298 * Concurrent thread allocated container data.
4301 spin_unlock(&obj->oo_guard);
4303 * Now, that we have container data, serialize its
4306 down_write(&obj->oo_ext_idx_sem);
4308 * recheck under lock.
4310 if (!osd_has_index(obj))
4311 result = osd_iam_container_init(env, obj,
4315 up_write(&obj->oo_ext_idx_sem);
4323 if (result == 0 && skip_iam == 0) {
4324 if (!osd_iam_index_probe(env, obj, feat))
4327 LINVRNT(osd_invariant(obj));
4332 static int osd_otable_it_attr_get(const struct lu_env *env,
4333 struct dt_object *dt,
4334 struct lu_attr *attr)
4340 static const struct dt_object_operations osd_obj_ops = {
4341 .do_read_lock = osd_read_lock,
4342 .do_write_lock = osd_write_lock,
4343 .do_read_unlock = osd_read_unlock,
4344 .do_write_unlock = osd_write_unlock,
4345 .do_write_locked = osd_write_locked,
4346 .do_attr_get = osd_attr_get,
4347 .do_declare_attr_set = osd_declare_attr_set,
4348 .do_attr_set = osd_attr_set,
4349 .do_ah_init = osd_ah_init,
4350 .do_declare_create = osd_declare_create,
4351 .do_create = osd_create,
4352 .do_declare_destroy = osd_declare_destroy,
4353 .do_destroy = osd_destroy,
4354 .do_index_try = osd_index_try,
4355 .do_declare_ref_add = osd_declare_ref_add,
4356 .do_ref_add = osd_ref_add,
4357 .do_declare_ref_del = osd_declare_ref_del,
4358 .do_ref_del = osd_ref_del,
4359 .do_xattr_get = osd_xattr_get,
4360 .do_declare_xattr_set = osd_declare_xattr_set,
4361 .do_xattr_set = osd_xattr_set,
4362 .do_declare_xattr_del = osd_declare_xattr_del,
4363 .do_xattr_del = osd_xattr_del,
4364 .do_xattr_list = osd_xattr_list,
4365 .do_object_sync = osd_object_sync,
4366 .do_invalidate = osd_invalidate,
4369 static const struct dt_object_operations osd_obj_otable_it_ops = {
4370 .do_attr_get = osd_otable_it_attr_get,
4371 .do_index_try = osd_index_try,
4374 static int osd_index_declare_iam_delete(const struct lu_env *env,
4375 struct dt_object *dt,
4376 const struct dt_key *key,
4377 struct thandle *handle)
4379 struct osd_thandle *oh;
4381 oh = container_of0(handle, struct osd_thandle, ot_super);
4382 LASSERT(oh->ot_handle == NULL);
4384 /* Recycle may cause additional three blocks to be changed. */
4385 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4386 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4392 * delete a (key, value) pair from index \a dt specified by \a key
4394 * \param dt osd index object
4395 * \param key key for index
4396 * \param rec record reference
4397 * \param handle transaction handler
4400 * \retval -ve failure
4402 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4403 const struct dt_key *key,
4404 struct thandle *handle)
4406 struct osd_thread_info *oti = osd_oti_get(env);
4407 struct osd_object *obj = osd_dt_obj(dt);
4408 struct osd_thandle *oh;
4409 struct iam_path_descr *ipd;
4410 struct iam_container *bag = &obj->oo_dir->od_container;
4414 if (!dt_object_exists(dt))
4417 LINVRNT(osd_invariant(obj));
4418 LASSERT(!dt_object_remote(dt));
4419 LASSERT(bag->ic_object == obj->oo_inode);
4420 LASSERT(handle != NULL);
4422 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4424 ipd = osd_idx_ipd_get(env, bag);
4425 if (unlikely(ipd == NULL))
4428 oh = container_of0(handle, struct osd_thandle, ot_super);
4429 LASSERT(oh->ot_handle != NULL);
4430 LASSERT(oh->ot_handle->h_transaction != NULL);
4432 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4433 /* swab quota uid/gid provided by caller */
4434 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4435 key = (const struct dt_key *)&oti->oti_quota_id;
4438 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4439 osd_ipd_put(env, bag, ipd);
4440 LINVRNT(osd_invariant(obj));
4441 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4445 static int osd_index_declare_ea_delete(const struct lu_env *env,
4446 struct dt_object *dt,
4447 const struct dt_key *key,
4448 struct thandle *handle)
4450 struct osd_thandle *oh;
4451 struct inode *inode;
4455 LASSERT(!dt_object_remote(dt));
4456 LASSERT(handle != NULL);
4458 oh = container_of0(handle, struct osd_thandle, ot_super);
4459 LASSERT(oh->ot_handle == NULL);
4461 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
4462 if (key != NULL && unlikely(strcmp((char *)key, dotdot) == 0)) {
4463 /* '..' to a remote object has a local representative */
4464 credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
4465 /* to reset LMAI_REMOTE_PARENT */
4468 osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
4470 inode = osd_dt_obj(dt)->oo_inode;
4474 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
4475 i_projid_read(inode), 0, oh, osd_dt_obj(dt),
4480 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
4483 struct osd_fid_pack *rec;
4486 if (de->file_type & LDISKFS_DIRENT_LUFID) {
4487 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
4488 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
4489 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
4495 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
4496 const struct lu_fid *fid)
4498 struct seq_server_site *ss = osd_seq_site(osd);
4501 /* FID seqs not in FLDB, must be local seq */
4502 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
4505 /* If FLD is not being initialized yet, it only happens during the
4506 * initialization, likely during mgs initialization, and we assume
4507 * this is local FID. */
4508 if (ss == NULL || ss->ss_server_fld == NULL)
4511 /* Only check the local FLDB here */
4512 if (osd_seq_exists(env, osd, fid_seq(fid)))
4519 * Index delete function for interoperability mode (b11826).
4520 * It will remove the directory entry added by osd_index_ea_insert().
4521 * This entry is needed to maintain name->fid mapping.
4523 * \param key, key i.e. file entry to be deleted
4525 * \retval 0, on success
4526 * \retval -ve, on error
4528 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
4529 const struct dt_key *key, struct thandle *handle)
4531 struct osd_object *obj = osd_dt_obj(dt);
4532 struct inode *dir = obj->oo_inode;
4533 struct dentry *dentry;
4534 struct osd_thandle *oh;
4535 struct ldiskfs_dir_entry_2 *de = NULL;
4536 struct buffer_head *bh;
4537 struct htree_lock *hlock = NULL;
4538 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
4539 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4543 if (!dt_object_exists(dt))
4546 LINVRNT(osd_invariant(obj));
4547 LASSERT(!dt_object_remote(dt));
4548 LASSERT(handle != NULL);
4550 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4552 oh = container_of(handle, struct osd_thandle, ot_super);
4553 LASSERT(oh->ot_handle != NULL);
4554 LASSERT(oh->ot_handle->h_transaction != NULL);
4556 ll_vfs_dq_init(dir);
4557 dentry = osd_child_dentry_get(env, obj,
4558 (char *)key, strlen((char *)key));
4560 if (obj->oo_hl_head != NULL) {
4561 hlock = osd_oti_get(env)->oti_hlock;
4562 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4563 dir, LDISKFS_HLOCK_DEL);
4565 down_write(&obj->oo_ext_idx_sem);
4568 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4570 /* If this is not the ".." entry, it might be a remote DNE
4571 * entry and we need to check if the FID is for a remote
4572 * MDT. If the FID is not in the directory entry (e.g.
4573 * upgraded 1.8 filesystem without dirdata enabled) then
4574 * we need to get the FID from the LMA. For a remote directory
4575 * there HAS to be an LMA, it cannot be an IGIF inode in this
4578 * Delete the entry before the agent inode in order to
4579 * simplify error handling. At worst an error after deleting
4580 * the entry first might leak the agent inode afterward. The
4581 * reverse would need filesystem abort in case of error deleting
4582 * the entry after the agent had been removed, or leave a
4583 * dangling entry pointing at a random inode. */
4584 if (strcmp((char *)key, dotdot) != 0) {
4585 LASSERT(de != NULL);
4586 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
4587 if (rc == -ENODATA) {
4588 /* can't get FID, postpone to the end of the
4589 * transaction when iget() is safe */
4590 osd_schedule_agent_inode_removal(env, oh,
4591 le32_to_cpu(de->inode));
4592 } else if (rc == 0 &&
4593 unlikely(osd_remote_fid(env, osd, fid))) {
4594 osd_schedule_agent_inode_removal(env, oh,
4595 le32_to_cpu(de->inode));
4598 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
4604 ldiskfs_htree_unlock(hlock);
4606 up_write(&obj->oo_ext_idx_sem);
4611 /* For inode on the remote MDT, .. will point to
4612 * /Agent directory, Check whether it needs to delete
4613 * from agent directory */
4614 if (unlikely(strcmp((char *)key, dotdot) == 0)) {
4617 ret = osd_delete_from_remote_parent(env, osd_obj2dev(obj),
4620 /* Sigh, the entry has been deleted, and
4621 * it is not easy to revert it back, so
4622 * let's keep this error private, and let
4623 * LFSCK fix it. XXX */
4624 CERROR("%s: delete remote parent "DFID": rc = %d\n",
4625 osd_name(osd), PFID(fid), ret);
4628 LASSERT(osd_invariant(obj));
4629 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4634 * Lookup index for \a key and copy record to \a rec.
4636 * \param dt osd index object
4637 * \param key key for index
4638 * \param rec record reference
4640 * \retval +ve success : exact mach
4641 * \retval 0 return record with key not greater than \a key
4642 * \retval -ve failure
4644 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
4645 struct dt_rec *rec, const struct dt_key *key)
4647 struct osd_object *obj = osd_dt_obj(dt);
4648 struct iam_path_descr *ipd;
4649 struct iam_container *bag = &obj->oo_dir->od_container;
4650 struct osd_thread_info *oti = osd_oti_get(env);
4651 struct iam_iterator *it = &oti->oti_idx_it;
4652 struct iam_rec *iam_rec;
4656 if (!dt_object_exists(dt))
4659 LASSERT(osd_invariant(obj));
4660 LASSERT(!dt_object_remote(dt));
4661 LASSERT(bag->ic_object == obj->oo_inode);
4663 ipd = osd_idx_ipd_get(env, bag);
4667 /* got ipd now we can start iterator. */
4668 iam_it_init(it, bag, 0, ipd);
4670 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4671 /* swab quota uid/gid provided by caller */
4672 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4673 key = (const struct dt_key *)&oti->oti_quota_id;
4676 rc = iam_it_get(it, (struct iam_key *)key);
4678 if (S_ISDIR(obj->oo_inode->i_mode))
4679 iam_rec = (struct iam_rec *)oti->oti_ldp;
4681 iam_rec = (struct iam_rec *) rec;
4683 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
4685 if (S_ISDIR(obj->oo_inode->i_mode))
4686 osd_fid_unpack((struct lu_fid *) rec,
4687 (struct osd_fid_pack *)iam_rec);
4688 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
4689 osd_quota_unpack(obj, rec);
4694 osd_ipd_put(env, bag, ipd);
4696 LINVRNT(osd_invariant(obj));
4701 static int osd_index_declare_iam_insert(const struct lu_env *env,
4702 struct dt_object *dt,
4703 const struct dt_rec *rec,
4704 const struct dt_key *key,
4705 struct thandle *handle)
4707 struct osd_thandle *oh;
4709 LASSERT(handle != NULL);
4711 oh = container_of0(handle, struct osd_thandle, ot_super);
4712 LASSERT(oh->ot_handle == NULL);
4714 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4715 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4721 * Inserts (key, value) pair in \a dt index object.
4723 * \param dt osd index object
4724 * \param key key for index
4725 * \param rec record reference
4726 * \param th transaction handler
4729 * \retval -ve failure
4731 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
4732 const struct dt_rec *rec,
4733 const struct dt_key *key, struct thandle *th,
4736 struct osd_object *obj = osd_dt_obj(dt);
4737 struct iam_path_descr *ipd;
4738 struct osd_thandle *oh;
4739 struct iam_container *bag;
4740 struct osd_thread_info *oti = osd_oti_get(env);
4741 struct iam_rec *iam_rec;
4745 if (!dt_object_exists(dt))
4748 LINVRNT(osd_invariant(obj));
4749 LASSERT(!dt_object_remote(dt));
4751 bag = &obj->oo_dir->od_container;
4752 LASSERT(bag->ic_object == obj->oo_inode);
4753 LASSERT(th != NULL);
4755 osd_trans_exec_op(env, th, OSD_OT_INSERT);
4757 ipd = osd_idx_ipd_get(env, bag);
4758 if (unlikely(ipd == NULL))
4761 oh = container_of0(th, struct osd_thandle, ot_super);
4762 LASSERT(oh->ot_handle != NULL);
4763 LASSERT(oh->ot_handle->h_transaction != NULL);
4764 if (S_ISDIR(obj->oo_inode->i_mode)) {
4765 iam_rec = (struct iam_rec *)oti->oti_ldp;
4766 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
4767 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4768 /* pack quota uid/gid */
4769 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4770 key = (const struct dt_key *)&oti->oti_quota_id;
4771 /* pack quota record */
4772 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
4773 iam_rec = (struct iam_rec *)rec;
4775 iam_rec = (struct iam_rec *)rec;
4778 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
4780 osd_ipd_put(env, bag, ipd);
4781 LINVRNT(osd_invariant(obj));
4782 osd_trans_exec_check(env, th, OSD_OT_INSERT);
4787 * Calls ldiskfs_add_entry() to add directory entry
4788 * into the directory. This is required for
4789 * interoperability mode (b11826)
4791 * \retval 0, on success
4792 * \retval -ve, on error
4794 static int __osd_ea_add_rec(struct osd_thread_info *info,
4795 struct osd_object *pobj, struct inode *cinode,
4796 const char *name, const struct lu_fid *fid,
4797 struct htree_lock *hlock, struct thandle *th)
4799 struct ldiskfs_dentry_param *ldp;
4800 struct dentry *child;
4801 struct osd_thandle *oth;
4804 oth = container_of(th, struct osd_thandle, ot_super);
4805 LASSERT(oth->ot_handle != NULL);
4806 LASSERT(oth->ot_handle->h_transaction != NULL);
4807 LASSERT(pobj->oo_inode);
4809 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
4810 if (unlikely(pobj->oo_inode ==
4811 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
4814 osd_get_ldiskfs_dirent_param(ldp, fid);
4815 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
4816 child->d_fsdata = (void *)ldp;
4817 ll_vfs_dq_init(pobj->oo_inode);
4818 rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
4819 child, cinode, hlock);
4820 if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
4821 struct ldiskfs_dir_entry_2 *de;
4822 struct buffer_head *bh;
4825 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
4828 rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
4831 if (S_ISDIR(cinode->i_mode))
4832 de->file_type = LDISKFS_DIRENT_LUFID |
4833 LDISKFS_FT_REG_FILE;
4835 de->file_type = LDISKFS_DIRENT_LUFID |
4837 ldiskfs_handle_dirty_metadata(oth->ot_handle,
4848 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
4849 * into the directory.Also sets flags into osd object to
4850 * indicate dot and dotdot are created. This is required for
4851 * interoperability mode (b11826)
4853 * \param dir directory for dot and dotdot fixup.
4854 * \param obj child object for linking
4856 * \retval 0, on success
4857 * \retval -ve, on error
4859 static int osd_add_dot_dotdot(struct osd_thread_info *info,
4860 struct osd_object *dir,
4861 struct inode *parent_dir, const char *name,
4862 const struct lu_fid *dot_fid,
4863 const struct lu_fid *dot_dot_fid,
4866 struct inode *inode = dir->oo_inode;
4867 struct osd_thandle *oth;
4870 oth = container_of(th, struct osd_thandle, ot_super);
4871 LASSERT(oth->ot_handle->h_transaction != NULL);
4872 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4874 if (strcmp(name, dot) == 0) {
4875 if (dir->oo_compat_dot_created) {
4878 LASSERT(inode->i_ino == parent_dir->i_ino);
4879 dir->oo_compat_dot_created = 1;
4882 } else if (strcmp(name, dotdot) == 0) {
4883 if (!dir->oo_compat_dot_created)
4885 /* in case of rename, dotdot is already created */
4886 if (dir->oo_compat_dotdot_created) {
4887 return __osd_ea_add_rec(info, dir, parent_dir, name,
4888 dot_dot_fid, NULL, th);
4891 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
4892 struct lu_fid tfid = *dot_dot_fid;
4895 result = osd_add_dot_dotdot_internal(info,
4896 dir->oo_inode, parent_dir, dot_fid,
4899 result = osd_add_dot_dotdot_internal(info,
4900 dir->oo_inode, parent_dir, dot_fid,
4905 dir->oo_compat_dotdot_created = 1;
4913 * It will call the appropriate osd_add* function and return the
4914 * value, return by respective functions.
4916 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4917 struct inode *cinode, const char *name,
4918 const struct lu_fid *fid, struct thandle *th)
4920 struct osd_thread_info *info = osd_oti_get(env);
4921 struct htree_lock *hlock;
4924 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4926 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4928 if (hlock != NULL) {
4929 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4932 down_write(&pobj->oo_ext_idx_sem);
4935 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4936 lu_object_fid(&pobj->oo_dt.do_lu),
4939 if (hlock != NULL) {
4940 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4941 pobj->oo_inode, LDISKFS_HLOCK_ADD);
4943 down_write(&pobj->oo_ext_idx_sem);
4946 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4947 struct lu_fid *tfid = &info->oti_fid;
4951 rc = __osd_ea_add_rec(info, pobj, cinode, name,
4954 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4959 ldiskfs_htree_unlock(hlock);
4961 up_write(&pobj->oo_ext_idx_sem);
4967 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4968 struct osd_idmap_cache *oic)
4970 struct osd_scrub *scrub = &dev->od_scrub;
4971 struct lu_fid *fid = &oic->oic_fid;
4972 struct osd_inode_id *id = &oic->oic_lid;
4973 struct inode *inode = NULL;
4979 if (!fid_is_norm(fid) && !fid_is_igif(fid))
4982 if (scrub->os_pos_current > id->oii_ino)
4986 rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
4987 if (rc == -ENOENT) {
4988 __u32 gen = id->oii_gen;
4994 inode = osd_iget(oti, dev, id);
4995 /* The inode has been removed (by race maybe). */
4996 if (IS_ERR(inode)) {
4997 rc = PTR_ERR(inode);
4999 RETURN(rc == -ESTALE ? -ENOENT : rc);
5002 /* The OI mapping is lost. */
5003 if (gen != OSD_OII_NOGEN)
5007 /* The inode may has been reused by others, we do not know,
5008 * leave it to be handled by subsequent osd_fid_lookup(). */
5010 } else if (rc != 0 || osd_id_eq(id, &oti->oti_id)) {
5017 if (thread_is_running(&scrub->os_thread)) {
5018 if (inode == NULL) {
5019 inode = osd_iget(oti, dev, id);
5020 /* The inode has been removed (by race maybe). */
5021 if (IS_ERR(inode)) {
5022 rc = PTR_ERR(inode);
5024 RETURN(rc == -ESTALE ? -ENOENT : rc);
5028 rc = osd_oii_insert(dev, oic, insert);
5029 /* There is race condition between osd_oi_lookup and OI scrub.
5030 * The OI scrub finished just after osd_oi_lookup() failure.
5031 * Under such case, it is unnecessary to trigger OI scrub again,
5032 * but try to call osd_oi_lookup() again. */
5033 if (unlikely(rc == -EAGAIN))
5036 if (!S_ISDIR(inode->i_mode))
5039 rc = osd_check_lmv(oti, dev, inode, oic);
5045 if (!dev->od_noscrub && ++once == 1) {
5046 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
5048 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5049 "%s: trigger partial OI scrub for RPC inconsistency "
5050 "checking FID "DFID": rc = %d\n",
5051 osd_dev2name(dev), PFID(fid), rc);
5052 if (rc == 0 || rc == -EALREADY)
5062 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5063 struct osd_device *dev,
5064 struct osd_idmap_cache *oic,
5065 struct lu_fid *fid, __u32 ino)
5067 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5068 struct inode *inode;
5071 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5072 inode = osd_iget(oti, dev, &oic->oic_lid);
5073 if (IS_ERR(inode)) {
5074 fid_zero(&oic->oic_fid);
5075 return PTR_ERR(inode);
5078 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5081 fid_zero(&oic->oic_fid);
5083 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5087 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5088 struct osd_inode_id *id, const struct lu_fid *fid)
5090 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5091 id->oii_ino, id->oii_gen, info);
5092 info->oti_cache.oic_lid = *id;
5093 info->oti_cache.oic_fid = *fid;
5094 info->oti_cache.oic_dev = osd;
5098 * Get parent FID from the linkEA.
5100 * For a directory which parent resides on remote MDT, to satisfy the
5101 * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5102 * the other hand, to make the lookup(..) on the directory can return
5103 * the real parent FID, we append the real parent FID after its ".."
5104 * name entry in the /REMOTE_PARENT_DIR.
5106 * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5107 * backup. So after the restore, we cannot get the right parent FID from
5108 * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5109 * we have stored the real parent FID in the directory object's linkEA,
5110 * we can parse the linkEA for the real parent FID.
5112 * \param[in] env pointer to the thread context
5113 * \param[in] obj pointer to the object to be handled
5114 * \param[out]fid pointer to the buffer to hold the parent FID
5116 * \retval 0 for getting the real parent FID successfully
5117 * \retval negative error number on failure
5119 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5120 struct osd_object *obj,
5123 struct osd_thread_info *oti = osd_oti_get(env);
5124 struct lu_buf *buf = &oti->oti_big_buf;
5125 struct dentry *dentry = &oti->oti_obj_dentry;
5126 struct inode *inode = obj->oo_inode;
5127 struct linkea_data ldata = { NULL };
5132 if (!S_ISDIR(inode->i_mode))
5136 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5137 buf->lb_buf, buf->lb_len);
5138 if (rc == -ERANGE) {
5139 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5142 lu_buf_realloc(buf, rc);
5143 if (buf->lb_buf == NULL)
5150 if (unlikely(rc == 0))
5156 if (unlikely(buf->lb_buf == NULL)) {
5157 lu_buf_realloc(buf, rc);
5158 if (buf->lb_buf == NULL)
5165 rc = linkea_init_with_rec(&ldata);
5167 linkea_first_entry(&ldata);
5168 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5174 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5175 struct inode *inode,
5176 const struct lu_fid *pfid,
5177 const char *name, const int namelen)
5179 struct osd_thread_info *oti = osd_oti_get(env);
5180 struct lu_buf *buf = &oti->oti_big_buf;
5181 struct dentry *dentry = &oti->oti_obj_dentry;
5182 struct linkea_data ldata = { NULL };
5183 struct lu_name cname = { .ln_name = name,
5184 .ln_namelen = namelen };
5189 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5190 buf->lb_buf, buf->lb_len);
5192 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5197 if (unlikely(rc == 0))
5200 if (buf->lb_len < rc) {
5201 lu_buf_realloc(buf, rc);
5202 if (buf->lb_buf == NULL)
5209 rc = linkea_init_with_rec(&ldata);
5211 rc = linkea_links_find(&ldata, &cname, pfid);
5217 * Calls ->lookup() to find dentry. From dentry get inode and
5218 * read inode's ea to get fid. This is required for interoperability
5221 * \retval 0, on success
5222 * \retval -ve, on error
5224 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5225 struct dt_rec *rec, const struct dt_key *key)
5227 struct inode *dir = obj->oo_inode;
5228 struct dentry *dentry;
5229 struct ldiskfs_dir_entry_2 *de;
5230 struct buffer_head *bh;
5231 struct lu_fid *fid = (struct lu_fid *) rec;
5232 struct htree_lock *hlock = NULL;
5237 LASSERT(dir->i_op != NULL);
5238 LASSERT(dir->i_op->lookup != NULL);
5240 dentry = osd_child_dentry_get(env, obj,
5241 (char *)key, strlen((char *)key));
5243 if (obj->oo_hl_head != NULL) {
5244 hlock = osd_oti_get(env)->oti_hlock;
5245 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5246 dir, LDISKFS_HLOCK_LOOKUP);
5248 down_read(&obj->oo_ext_idx_sem);
5251 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5253 struct osd_thread_info *oti = osd_oti_get(env);
5254 struct osd_inode_id *id = &oti->oti_id;
5255 struct osd_idmap_cache *oic = &oti->oti_cache;
5256 struct osd_device *dev = osd_obj2dev(obj);
5258 ino = le32_to_cpu(de->inode);
5259 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5261 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5265 rc = osd_get_fid_from_dentry(de, rec);
5267 /* done with de, release bh */
5270 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5271 const char *name = (const char *)key;
5273 /* If the parent is on remote MDT, and there
5274 * is no FID-in-dirent, then we have to get
5275 * the parent FID from the linkEA. */
5276 if (likely(strlen(name) == 2 &&
5277 name[0] == '.' && name[1] == '.'))
5278 rc = osd_get_pfid_from_linkea(env, obj,
5281 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5284 osd_id_gen(id, ino, OSD_OII_NOGEN);
5287 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5288 fid_zero(&oic->oic_fid);
5293 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5294 rc = osd_consistency_check(oti, dev, oic);
5296 fid_zero(&oic->oic_fid);
5305 ldiskfs_htree_unlock(hlock);
5307 up_read(&obj->oo_ext_idx_sem);
5311 static int osd_index_declare_ea_insert(const struct lu_env *env,
5312 struct dt_object *dt,
5313 const struct dt_rec *rec,
5314 const struct dt_key *key,
5315 struct thandle *handle)
5317 struct osd_thandle *oh;
5318 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5319 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5320 const struct lu_fid *fid = rec1->rec_fid;
5321 int credits, rc = 0;
5322 struct osd_idmap_cache *idc;
5325 LASSERT(!dt_object_remote(dt));
5326 LASSERT(handle != NULL);
5327 LASSERT(fid != NULL);
5328 LASSERT(rec1->rec_type != 0);
5330 oh = container_of0(handle, struct osd_thandle, ot_super);
5331 LASSERT(oh->ot_handle == NULL);
5333 credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5335 /* we can't call iget() while a transactions is running
5336 * (this can lead to a deadlock), but we need to know
5337 * inum and object type. so we find this information at
5338 * declaration and cache in per-thread info */
5339 idc = osd_idc_find_or_init(env, osd, fid);
5341 RETURN(PTR_ERR(idc));
5342 if (idc->oic_remote) {
5343 /* a reference to remote inode is represented by an
5344 * agent inode which we have to create */
5345 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5346 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5349 osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5351 if (osd_dt_obj(dt)->oo_inode != NULL) {
5352 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5354 /* We ignore block quota on meta pool (MDTs), so needn't
5355 * calculate how many blocks will be consumed by this index
5357 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5359 i_projid_read(inode), 0,
5360 oh, osd_dt_obj(dt), NULL,
5368 * Index add function for interoperability mode (b11826).
5369 * It will add the directory entry.This entry is needed to
5370 * maintain name->fid mapping.
5372 * \param key it is key i.e. file entry to be inserted
5373 * \param rec it is value of given key i.e. fid
5375 * \retval 0, on success
5376 * \retval -ve, on error
5378 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5379 const struct dt_rec *rec,
5380 const struct dt_key *key, struct thandle *th,
5383 struct osd_object *obj = osd_dt_obj(dt);
5384 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5385 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5386 const struct lu_fid *fid = rec1->rec_fid;
5387 const char *name = (const char *)key;
5388 struct osd_thread_info *oti = osd_oti_get(env);
5389 struct inode *child_inode = NULL;
5390 struct osd_idmap_cache *idc;
5394 if (!dt_object_exists(dt))
5397 LASSERT(osd_invariant(obj));
5398 LASSERT(!dt_object_remote(dt));
5399 LASSERT(th != NULL);
5401 osd_trans_exec_op(env, th, OSD_OT_INSERT);
5403 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
5405 idc = osd_idc_find(env, osd, fid);
5406 if (unlikely(idc == NULL)) {
5407 /* this dt_insert() wasn't declared properly, so
5408 * FID is missing in OI cache. we better do not
5409 * lookup FID in FLDB/OI and don't risk to deadlock,
5410 * but in some special cases (lfsck testing, etc)
5411 * it's much simpler than fixing a caller */
5412 CERROR("%s: "DFID" wasn't declared for insert\n",
5413 osd_name(osd), PFID(fid));
5415 idc = osd_idc_find_or_init(env, osd, fid);
5417 RETURN(PTR_ERR(idc));
5420 if (idc->oic_remote) {
5421 /* Insert remote entry */
5422 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
5423 struct osd_mdobj_map *omm = osd->od_mdt_map;
5424 struct osd_thandle *oh;
5426 /* If parent on remote MDT, we need put this object
5428 oh = container_of(th, typeof(*oh), ot_super);
5429 rc = osd_add_to_remote_parent(env, osd, obj, oh);
5431 CERROR("%s: add "DFID" error: rc = %d\n",
5433 PFID(lu_object_fid(&dt->do_lu)), rc);
5437 child_inode = igrab(omm->omm_remote_parent->d_inode);
5439 child_inode = osd_create_local_agent_inode(env, osd,
5440 obj, fid, rec1->rec_type & S_IFMT, th);
5441 if (IS_ERR(child_inode))
5442 RETURN(PTR_ERR(child_inode));
5445 /* Insert local entry */
5446 if (unlikely(idc->oic_lid.oii_ino == 0)) {
5447 /* for a reason OI cache wasn't filled properly */
5448 CERROR("%s: OIC for "DFID" isn't filled\n",
5449 osd_name(osd), PFID(fid));
5452 child_inode = oti->oti_inode;
5453 if (unlikely(child_inode == NULL)) {
5454 struct ldiskfs_inode_info *lii;
5458 child_inode = oti->oti_inode = &lii->vfs_inode;
5460 child_inode->i_sb = osd_sb(osd);
5461 child_inode->i_ino = idc->oic_lid.oii_ino;
5462 child_inode->i_mode = rec1->rec_type & S_IFMT;
5465 rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
5467 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
5468 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
5470 if (child_inode && child_inode != oti->oti_inode)
5472 LASSERT(osd_invariant(obj));
5473 osd_trans_exec_check(env, th, OSD_OT_INSERT);
5478 * Initialize osd Iterator for given osd index object.
5480 * \param dt osd index object
5483 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
5484 struct dt_object *dt,
5487 struct osd_it_iam *it;
5488 struct osd_object *obj = osd_dt_obj(dt);
5489 struct lu_object *lo = &dt->do_lu;
5490 struct iam_path_descr *ipd;
5491 struct iam_container *bag = &obj->oo_dir->od_container;
5493 if (!dt_object_exists(dt))
5494 return ERR_PTR(-ENOENT);
5498 return ERR_PTR(-ENOMEM);
5500 ipd = osd_it_ipd_get(env, bag);
5501 if (likely(ipd != NULL)) {
5505 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
5506 return (struct dt_it *)it;
5509 return ERR_PTR(-ENOMEM);
5514 * free given Iterator.
5516 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
5518 struct osd_it_iam *it = (struct osd_it_iam *)di;
5519 struct osd_object *obj = it->oi_obj;
5521 iam_it_fini(&it->oi_it);
5522 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
5523 osd_object_put(env, obj);
5528 * Move Iterator to record specified by \a key
5530 * \param di osd iterator
5531 * \param key key for index
5533 * \retval +ve di points to record with least key not larger than key
5534 * \retval 0 di points to exact matched key
5535 * \retval -ve failure
5538 static int osd_it_iam_get(const struct lu_env *env,
5539 struct dt_it *di, const struct dt_key *key)
5541 struct osd_thread_info *oti = osd_oti_get(env);
5542 struct osd_it_iam *it = (struct osd_it_iam *)di;
5544 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5545 /* swab quota uid/gid */
5546 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5547 key = (struct dt_key *)&oti->oti_quota_id;
5550 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
5556 * \param di osd iterator
5558 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
5560 struct osd_it_iam *it = (struct osd_it_iam *)di;
5562 iam_it_put(&it->oi_it);
5566 * Move iterator by one record
5568 * \param di osd iterator
5570 * \retval +1 end of container reached
5572 * \retval -ve failure
5575 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
5577 struct osd_it_iam *it = (struct osd_it_iam *)di;
5579 return iam_it_next(&it->oi_it);
5583 * Return pointer to the key under iterator.
5586 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
5587 const struct dt_it *di)
5589 struct osd_thread_info *oti = osd_oti_get(env);
5590 struct osd_it_iam *it = (struct osd_it_iam *)di;
5591 struct osd_object *obj = it->oi_obj;
5594 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
5596 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
5597 /* swab quota uid/gid */
5598 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
5599 key = (struct dt_key *)&oti->oti_quota_id;
5606 * Return size of key under iterator (in bytes)
5609 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
5611 struct osd_it_iam *it = (struct osd_it_iam *)di;
5613 return iam_it_key_size(&it->oi_it);
5617 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
5619 /* check if file type is required */
5620 if (ent->lde_attrs & LUDA_TYPE) {
5621 struct luda_type *lt;
5622 int align = sizeof(*lt) - 1;
5624 len = (len + align) & ~align;
5625 lt = (struct luda_type *)(ent->lde_name + len);
5626 lt->lt_type = cpu_to_le16(DTTOIF(type));
5629 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
5633 * build lu direct from backend fs dirent.
5637 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
5638 char *name, __u16 namelen, __u16 type, __u32 attr)
5640 ent->lde_attrs = attr | LUDA_FID;
5641 fid_cpu_to_le(&ent->lde_fid, fid);
5643 ent->lde_hash = cpu_to_le64(offset);
5644 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
5646 strncpy(ent->lde_name, name, namelen);
5647 ent->lde_name[namelen] = '\0';
5648 ent->lde_namelen = cpu_to_le16(namelen);
5650 /* append lustre attributes */
5651 osd_it_append_attrs(ent, namelen, type);
5655 * Return pointer to the record under iterator.
5657 static int osd_it_iam_rec(const struct lu_env *env,
5658 const struct dt_it *di,
5659 struct dt_rec *dtrec, __u32 attr)
5661 struct osd_it_iam *it = (struct osd_it_iam *)di;
5662 struct osd_thread_info *info = osd_oti_get(env);
5665 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
5666 const struct osd_fid_pack *rec;
5667 struct lu_fid *fid = &info->oti_fid;
5668 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
5674 name = (char *)iam_it_key_get(&it->oi_it);
5676 RETURN(PTR_ERR(name));
5678 namelen = iam_it_key_size(&it->oi_it);
5680 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
5682 RETURN(PTR_ERR(rec));
5684 rc = osd_fid_unpack(fid, rec);
5688 hash = iam_it_store(&it->oi_it);
5690 /* IAM does not store object type in IAM index (dir) */
5691 osd_it_pack_dirent(lde, fid, hash, name, namelen,
5693 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5694 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5695 (struct iam_rec *)dtrec);
5696 osd_quota_unpack(it->oi_obj, dtrec);
5698 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5699 (struct iam_rec *)dtrec);
5706 * Returns cookie for current Iterator position.
5708 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
5710 struct osd_it_iam *it = (struct osd_it_iam *)di;
5712 return iam_it_store(&it->oi_it);
5716 * Restore iterator from cookie.
5718 * \param di osd iterator
5719 * \param hash Iterator location cookie
5721 * \retval +ve di points to record with least key not larger than key.
5722 * \retval 0 di points to exact matched key
5723 * \retval -ve failure
5726 static int osd_it_iam_load(const struct lu_env *env,
5727 const struct dt_it *di, __u64 hash)
5729 struct osd_it_iam *it = (struct osd_it_iam *)di;
5731 return iam_it_load(&it->oi_it, hash);
5734 static const struct dt_index_operations osd_index_iam_ops = {
5735 .dio_lookup = osd_index_iam_lookup,
5736 .dio_declare_insert = osd_index_declare_iam_insert,
5737 .dio_insert = osd_index_iam_insert,
5738 .dio_declare_delete = osd_index_declare_iam_delete,
5739 .dio_delete = osd_index_iam_delete,
5741 .init = osd_it_iam_init,
5742 .fini = osd_it_iam_fini,
5743 .get = osd_it_iam_get,
5744 .put = osd_it_iam_put,
5745 .next = osd_it_iam_next,
5746 .key = osd_it_iam_key,
5747 .key_size = osd_it_iam_key_size,
5748 .rec = osd_it_iam_rec,
5749 .store = osd_it_iam_store,
5750 .load = osd_it_iam_load
5756 * Creates or initializes iterator context.
5758 * \retval struct osd_it_ea, iterator structure on success
5761 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
5762 struct dt_object *dt,
5765 struct osd_object *obj = osd_dt_obj(dt);
5766 struct osd_thread_info *info = osd_oti_get(env);
5767 struct osd_it_ea *oie;
5769 struct lu_object *lo = &dt->do_lu;
5770 struct dentry *obj_dentry;
5773 if (!dt_object_exists(dt) || obj->oo_destroyed)
5774 RETURN(ERR_PTR(-ENOENT));
5776 OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
5778 RETURN(ERR_PTR(-ENOMEM));
5779 obj_dentry = &oie->oie_dentry;
5781 obj_dentry->d_inode = obj->oo_inode;
5782 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
5783 obj_dentry->d_name.hash = 0;
5785 oie->oie_rd_dirent = 0;
5786 oie->oie_it_dirent = 0;
5787 oie->oie_dirent = NULL;
5788 if (unlikely(!info->oti_it_ea_buf_used)) {
5789 oie->oie_buf = info->oti_it_ea_buf;
5790 info->oti_it_ea_buf_used = 1;
5792 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5793 if (oie->oie_buf == NULL)
5794 RETURN(ERR_PTR(-ENOMEM));
5798 file = &oie->oie_file;
5800 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
5801 if (attr & LUDA_64BITHASH)
5802 file->f_mode = FMODE_64BITHASH;
5804 file->f_mode = FMODE_32BITHASH;
5805 file->f_path.dentry = obj_dentry;
5806 file->f_mapping = obj->oo_inode->i_mapping;
5807 file->f_op = obj->oo_inode->i_fop;
5808 set_file_inode(file, obj->oo_inode);
5811 RETURN((struct dt_it *) oie);
5815 * Destroy or finishes iterator context.
5817 * \param di iterator structure to be destroyed
5819 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
5821 struct osd_thread_info *info = osd_oti_get(env);
5822 struct osd_it_ea *oie = (struct osd_it_ea *)di;
5823 struct osd_object *obj = oie->oie_obj;
5824 struct inode *inode = obj->oo_inode;
5827 oie->oie_file.f_op->release(inode, &oie->oie_file);
5828 osd_object_put(env, obj);
5829 if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
5830 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5832 info->oti_it_ea_buf_used = 0;
5833 OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
5838 * It position the iterator at given key, so that next lookup continues from
5839 * that key Or it is similar to dio_it->load() but based on a key,
5840 * rather than file position.
5842 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
5845 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
5847 static int osd_it_ea_get(const struct lu_env *env,
5848 struct dt_it *di, const struct dt_key *key)
5850 struct osd_it_ea *it = (struct osd_it_ea *)di;
5853 LASSERT(((const char *)key)[0] == '\0');
5854 it->oie_file.f_pos = 0;
5855 it->oie_rd_dirent = 0;
5856 it->oie_it_dirent = 0;
5857 it->oie_dirent = NULL;
5865 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
5869 struct osd_filldir_cbs {
5870 #ifdef HAVE_DIR_CONTEXT
5871 struct dir_context ctx;
5873 struct osd_it_ea *it;
5876 * It is called internally by ->readdir(). It fills the
5877 * iterator's in-memory data structure with required
5878 * information i.e. name, namelen, rec_size etc.
5880 * \param buf in which information to be filled in.
5881 * \param name name of the file in given dir
5883 * \retval 0 on success
5884 * \retval 1 on buffer full
5886 #ifdef HAVE_FILLDIR_USE_CTX
5887 static int osd_ldiskfs_filldir(struct dir_context *buf,
5889 static int osd_ldiskfs_filldir(void *buf,
5891 const char *name, int namelen,
5892 loff_t offset, __u64 ino, unsigned d_type)
5894 struct osd_it_ea *it =
5895 ((struct osd_filldir_cbs *)buf)->it;
5896 struct osd_object *obj = it->oie_obj;
5897 struct osd_it_ea_dirent *ent = it->oie_dirent;
5898 struct lu_fid *fid = &ent->oied_fid;
5899 struct osd_fid_pack *rec;
5902 /* this should never happen */
5903 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5904 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5908 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
5912 /* "." is just the object itself. */
5913 if (namelen == 1 && name[0] == '.') {
5914 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5915 } else if (d_type & LDISKFS_DIRENT_LUFID) {
5916 rec = (struct osd_fid_pack*) (name + namelen + 1);
5917 if (osd_fid_unpack(fid, rec) != 0)
5922 d_type &= ~LDISKFS_DIRENT_LUFID;
5924 /* NOT export local root. */
5925 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5926 ino = obj->oo_inode->i_ino;
5927 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5930 ent->oied_ino = ino;
5931 ent->oied_off = offset;
5932 ent->oied_namelen = namelen;
5933 ent->oied_type = d_type;
5935 memcpy(ent->oied_name, name, namelen);
5937 it->oie_rd_dirent++;
5938 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
5943 * Calls ->readdir() to load a directory entry at a time
5944 * and stored it in iterator's in-memory data structure.
5946 * \param di iterator's in memory structure
5948 * \retval 0 on success
5949 * \retval -ve on error
5950 * \retval +1 reach the end of entry
5952 static int osd_ldiskfs_it_fill(const struct lu_env *env,
5953 const struct dt_it *di)
5955 struct osd_it_ea *it = (struct osd_it_ea *)di;
5956 struct osd_object *obj = it->oie_obj;
5957 struct inode *inode = obj->oo_inode;
5958 struct htree_lock *hlock = NULL;
5959 struct file *filp = &it->oie_file;
5961 struct osd_filldir_cbs buf = {
5962 #ifdef HAVE_DIR_CONTEXT
5963 .ctx.actor = osd_ldiskfs_filldir,
5969 it->oie_dirent = it->oie_buf;
5970 it->oie_rd_dirent = 0;
5972 if (obj->oo_hl_head != NULL) {
5973 hlock = osd_oti_get(env)->oti_hlock;
5974 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5975 inode, LDISKFS_HLOCK_READDIR);
5977 down_read(&obj->oo_ext_idx_sem);
5980 #ifdef HAVE_DIR_CONTEXT
5981 buf.ctx.pos = filp->f_pos;
5982 rc = inode->i_fop->iterate(filp, &buf.ctx);
5983 filp->f_pos = buf.ctx.pos;
5985 rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
5989 ldiskfs_htree_unlock(hlock);
5991 up_read(&obj->oo_ext_idx_sem);
5993 if (it->oie_rd_dirent == 0) {
5994 /*If it does not get any dirent, it means it has been reached
5995 *to the end of the dir */
5996 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6000 it->oie_dirent = it->oie_buf;
6001 it->oie_it_dirent = 1;
6008 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6009 * to load a directory entry at a time and stored it in
6010 * iterator's in-memory data structure.
6012 * \param di iterator's in memory structure
6014 * \retval +ve iterator reached to end
6015 * \retval 0 iterator not reached to end
6016 * \retval -ve on error
6018 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6020 struct osd_it_ea *it = (struct osd_it_ea *)di;
6025 if (it->oie_it_dirent < it->oie_rd_dirent) {
6027 (void *) it->oie_dirent +
6028 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6029 it->oie_dirent->oied_namelen);
6030 it->oie_it_dirent++;
6033 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6036 rc = osd_ldiskfs_it_fill(env, di);
6043 * Returns the key at current position from iterator's in memory structure.
6045 * \param di iterator's in memory structure
6047 * \retval key i.e. struct dt_key on success
6049 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6050 const struct dt_it *di)
6052 struct osd_it_ea *it = (struct osd_it_ea *)di;
6054 return (struct dt_key *)it->oie_dirent->oied_name;
6058 * Returns the key's size at current position from iterator's in memory structure.
6060 * \param di iterator's in memory structure
6062 * \retval key_size i.e. struct dt_key on success
6064 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6066 struct osd_it_ea *it = (struct osd_it_ea *)di;
6068 return it->oie_dirent->oied_namelen;
6071 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6073 if (LDISKFS_DIR_REC_LEN(de) >=
6074 __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6081 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6082 unsigned blocksize, bool dotdot)
6085 return osd_dotdot_has_space(de);
6087 if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6088 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6095 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6096 handle_t *jh, struct dentry *dentry,
6097 const struct lu_fid *fid, struct buffer_head *bh,
6098 struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6101 struct inode *dir = dentry->d_parent->d_inode;
6102 struct inode *inode = dentry->d_inode;
6103 struct osd_fid_pack *rec;
6104 struct ldiskfs_dentry_param *ldp;
6105 int namelen = dentry->d_name.len;
6107 struct osd_thread_info *info = osd_oti_get(env);
6110 if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
6111 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6114 /* There is enough space to hold the FID-in-dirent. */
6115 if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6116 rc = ldiskfs_journal_get_write_access(jh, bh);
6120 de->name[namelen] = 0;
6121 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6122 rec->fp_len = sizeof(struct lu_fid) + 1;
6123 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6124 de->file_type |= LDISKFS_DIRENT_LUFID;
6125 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6132 rc = ldiskfs_delete_entry(jh, dir, de, bh);
6136 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6137 osd_get_ldiskfs_dirent_param(ldp, fid);
6138 dentry->d_fsdata = (void *)ldp;
6139 ll_vfs_dq_init(dir);
6140 rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6141 /* It is too bad, we cannot reinsert the name entry back.
6142 * That means we lose it! */
6144 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6145 "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6146 osd_ino2name(inode),
6147 dir->i_ino, dir->i_generation, namelen,
6148 dentry->d_name.name, PFID(fid), rc);
6154 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6155 struct osd_it_ea *it, struct lu_fid *fid,
6156 struct osd_inode_id *id, __u32 *attr)
6158 struct osd_thread_info *info = osd_oti_get(env);
6159 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6160 struct osd_device *dev = osd_obj2dev(obj);
6161 struct super_block *sb = osd_sb(dev);
6162 const char *devname = osd_name(dev);
6163 struct osd_it_ea_dirent *ent = it->oie_dirent;
6164 struct inode *dir = obj->oo_inode;
6165 struct htree_lock *hlock = NULL;
6166 struct buffer_head *bh = NULL;
6167 handle_t *jh = NULL;
6168 struct ldiskfs_dir_entry_2 *de;
6169 struct dentry *dentry;
6170 struct inode *inode;
6171 const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6174 bool dotdot = false;
6178 if (ent->oied_name[0] == '.') {
6179 if (ent->oied_namelen == 1)
6182 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6186 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6187 inode = osd_iget(info, dev, id);
6188 if (IS_ERR(inode)) {
6189 rc = PTR_ERR(inode);
6190 if (rc == -ENOENT || rc == -ESTALE) {
6191 /* Maybe dangling name entry, or
6192 * corrupted directory entry. */
6193 *attr |= LUDA_UNKNOWN;
6196 CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6197 "check_repair, dir = %lu/%u, name = %.*s, "
6198 "ino = %llu, rc = %d\n",
6199 devname, dir->i_ino, dir->i_generation,
6200 ent->oied_namelen, ent->oied_name,
6207 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6209 rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6210 if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6215 /* We need to ensure that the name entry is still valid.
6216 * Because it may be removed or renamed by other already.
6218 * The unlink or rename operation will start journal before PDO lock,
6219 * so to avoid deadlock, here we need to start journal handle before
6220 * related PDO lock also. But because we do not know whether there
6221 * will be something to be repaired before PDO lock, we just start
6222 * journal without conditions.
6224 * We may need to remove the name entry firstly, then insert back.
6225 * One credit is for user quota file update.
6226 * One credit is for group quota file update.
6227 * Two credits are for dirty inode. */
6228 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6229 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6231 if (dev->od_dirent_journal != 0) {
6234 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6237 CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6238 "check_repair, dir = %lu/%u, credits = %d, "
6239 "name = %.*s, ino = %llu: rc = %d\n",
6240 devname, dir->i_ino, dir->i_generation, credits,
6241 ent->oied_namelen, ent->oied_name,
6244 GOTO(out_inode, rc);
6247 if (obj->oo_hl_head != NULL) {
6248 hlock = osd_oti_get(env)->oti_hlock;
6249 /* "0" means exclusive lock for the whole directory.
6250 * We need to prevent others access such name entry
6251 * during the delete + insert. Neither HLOCK_ADD nor
6252 * HLOCK_DEL cannot guarantee the atomicity. */
6253 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6255 down_write(&obj->oo_ext_idx_sem);
6258 if (obj->oo_hl_head != NULL) {
6259 hlock = osd_oti_get(env)->oti_hlock;
6260 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6261 LDISKFS_HLOCK_LOOKUP);
6263 down_read(&obj->oo_ext_idx_sem);
6267 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6268 if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6269 *attr |= LUDA_IGNORE;
6274 /* For dotdot entry, if there is not enough space to hold the
6275 * FID-in-dirent, just keep them there. It only happens when the
6276 * device upgraded from 1.8 or restored from MDT file-level backup.
6277 * For the whole directory, only dotdot entry have no FID-in-dirent
6278 * and needs to get FID from LMA when readdir, it will not affect the
6279 * performance much. */
6280 if (dotdot && !osd_dotdot_has_space(de)) {
6281 *attr |= LUDA_UNKNOWN;
6287 if (lu_fid_eq(fid, &lma->lma_self_fid))
6290 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6291 struct lu_fid *tfid = &lma->lma_self_fid;
6293 if (likely(dotdot &&
6294 fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6295 fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6296 /* It must be REMOTE_PARENT_DIR and as the
6297 * 'dotdot' entry of remote directory */
6298 *attr |= LUDA_IGNORE;
6300 CDEBUG(D_LFSCK, "%s: expect remote agent "
6301 "parent directory, but got %.*s under "
6302 "dir = %lu/%u with the FID "DFID"\n",
6303 devname, ent->oied_namelen,
6304 ent->oied_name, dir->i_ino,
6305 dir->i_generation, PFID(tfid));
6307 *attr |= LUDA_UNKNOWN;
6314 if (!fid_is_zero(fid)) {
6315 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6317 if (rc == -ENOENT ||
6319 !(dev->od_scrub.os_file.sf_flags & SF_UPGRADE))) {
6320 /* linkEA does not recognize the dirent entry,
6321 * it may because the dirent entry corruption
6322 * and points to other's inode. */
6323 CDEBUG(D_LFSCK, "%s: the target inode does not "
6324 "recognize the dirent, dir = %lu/%u, "
6325 " name = %.*s, ino = %llu, "
6326 DFID": rc = %d\n", devname, dir->i_ino,
6327 dir->i_generation, ent->oied_namelen,
6328 ent->oied_name, ent->oied_ino, PFID(fid), rc);
6329 *attr |= LUDA_UNKNOWN;
6334 if (rc && rc != -ENODATA) {
6335 CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6336 "dir = %lu/%u, name = %.*s, ino = %llu, "
6337 DFID": rc = %d\n", devname, dir->i_ino,
6338 dir->i_generation, ent->oied_namelen,
6339 ent->oied_name, ent->oied_ino, PFID(fid), rc);
6340 *attr |= LUDA_UNKNOWN;
6347 /* linkEA recognizes the dirent entry, the FID-in-LMA is
6348 * valid, trusted, in spite of fid_is_sane(fid) or not. */
6349 if (*attr & LUDA_VERIFY_DRYRUN) {
6350 *fid = lma->lma_self_fid;
6351 *attr |= LUDA_REPAIR;
6358 dev->od_dirent_journal = 1;
6359 if (hlock != NULL) {
6360 ldiskfs_htree_unlock(hlock);
6363 up_read(&obj->oo_ext_idx_sem);
6369 *fid = lma->lma_self_fid;
6371 /* Update or append the FID-in-dirent. */
6372 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6373 bh, de, hlock, dotdot);
6375 *attr |= LUDA_REPAIR;
6377 CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
6378 "the dirent, dir = %lu/%u, name = %.*s, "
6379 "ino = %llu, "DFID": rc = %d\n",
6380 devname, dir->i_ino, dir->i_generation,
6381 ent->oied_namelen, ent->oied_name,
6382 ent->oied_ino, PFID(fid), rc);
6384 /* lma is NULL, trust the FID-in-dirent if it is valid. */
6385 if (*attr & LUDA_VERIFY_DRYRUN) {
6386 if (fid_is_sane(fid)) {
6387 *attr |= LUDA_REPAIR;
6388 } else if (dev->od_index == 0) {
6389 lu_igif_build(fid, inode->i_ino,
6390 inode->i_generation);
6391 *attr |= LUDA_UPGRADE;
6399 dev->od_dirent_journal = 1;
6400 if (hlock != NULL) {
6401 ldiskfs_htree_unlock(hlock);
6404 up_read(&obj->oo_ext_idx_sem);
6411 if (unlikely(fid_is_sane(fid))) {
6412 /* FID-in-dirent exists, but FID-in-LMA is lost.
6413 * Trust the FID-in-dirent, and add FID-in-LMA. */
6414 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
6416 *attr |= LUDA_REPAIR;
6418 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
6419 "update dirent, dir = %lu/%u, "
6420 "name = %.*s, ino = %llu, "
6422 devname, dir->i_ino, dir->i_generation,
6423 ent->oied_namelen, ent->oied_name,
6424 ent->oied_ino, PFID(fid), rc);
6425 } else if (dev->od_index == 0) {
6426 lu_igif_build(fid, inode->i_ino, inode->i_generation);
6427 /* It is probably IGIF object. Only aappend the
6428 * FID-in-dirent. OI scrub will process FID-in-LMA. */
6429 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6430 bh, de, hlock, dotdot);
6432 *attr |= LUDA_UPGRADE;
6434 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
6435 "after the dirent, dir = %lu/%u, "
6436 "name = %.*s, ino = %llu, "
6438 devname, dir->i_ino, dir->i_generation,
6439 ent->oied_namelen, ent->oied_name,
6440 ent->oied_ino, PFID(fid), rc);
6449 if (hlock != NULL) {
6450 ldiskfs_htree_unlock(hlock);
6452 if (dev->od_dirent_journal != 0)
6453 up_write(&obj->oo_ext_idx_sem);
6455 up_read(&obj->oo_ext_idx_sem);
6459 ldiskfs_journal_stop(jh);
6463 if (rc >= 0 && !dirty)
6464 dev->od_dirent_journal = 0;
6470 * Returns the value at current position from iterator's in memory structure.
6472 * \param di struct osd_it_ea, iterator's in memory structure
6473 * \param attr attr requested for dirent.
6474 * \param lde lustre dirent
6476 * \retval 0 no error and \param lde has correct lustre dirent.
6477 * \retval -ve on error
6479 static inline int osd_it_ea_rec(const struct lu_env *env,
6480 const struct dt_it *di,
6481 struct dt_rec *dtrec, __u32 attr)
6483 struct osd_it_ea *it = (struct osd_it_ea *)di;
6484 struct osd_object *obj = it->oie_obj;
6485 struct osd_device *dev = osd_obj2dev(obj);
6486 struct osd_thread_info *oti = osd_oti_get(env);
6487 struct osd_inode_id *id = &oti->oti_id;
6488 struct lu_fid *fid = &it->oie_dirent->oied_fid;
6489 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6490 __u32 ino = it->oie_dirent->oied_ino;
6494 LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
6496 if (attr & LUDA_VERIFY) {
6497 if (unlikely(ino == osd_remote_parent_ino(dev))) {
6498 attr |= LUDA_IGNORE;
6499 /* If the parent is on remote MDT, and there
6500 * is no FID-in-dirent, then we have to get
6501 * the parent FID from the linkEA. */
6502 if (!fid_is_sane(fid) &&
6503 it->oie_dirent->oied_namelen == 2 &&
6504 it->oie_dirent->oied_name[0] == '.' &&
6505 it->oie_dirent->oied_name[1] == '.')
6506 osd_get_pfid_from_linkea(env, obj, fid);
6508 rc = osd_dirent_check_repair(env, obj, it, fid, id,
6512 if (!fid_is_sane(fid))
6513 attr |= LUDA_UNKNOWN;
6515 attr &= ~LU_DIRENT_ATTRS_MASK;
6516 if (!fid_is_sane(fid)) {
6517 bool is_dotdot = false;
6518 if (it->oie_dirent->oied_namelen == 2 &&
6519 it->oie_dirent->oied_name[0] == '.' &&
6520 it->oie_dirent->oied_name[1] == '.')
6522 /* If the parent is on remote MDT, and there
6523 * is no FID-in-dirent, then we have to get
6524 * the parent FID from the linkEA. */
6525 if (ino == osd_remote_parent_ino(dev) && is_dotdot) {
6526 rc = osd_get_pfid_from_linkea(env, obj, fid);
6528 if (is_dotdot == false &&
6529 OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
6532 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6535 osd_id_gen(id, ino, OSD_OII_NOGEN);
6539 /* Pack the entry anyway, at least the offset is right. */
6540 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
6541 it->oie_dirent->oied_name,
6542 it->oie_dirent->oied_namelen,
6543 it->oie_dirent->oied_type, attr);
6548 if (osd_remote_fid(env, dev, fid))
6551 if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
6552 osd_add_oi_cache(oti, dev, id, fid);
6554 RETURN(rc > 0 ? 0 : rc);
6558 * Returns the record size size at current position.
6560 * This function will return record(lu_dirent) size in bytes.
6562 * \param[in] env execution environment
6563 * \param[in] di iterator's in memory structure
6564 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
6565 * calculate the lu_dirent size.
6567 * \retval record size(in bytes & in memory) of the current lu_dirent
6570 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
6573 struct osd_it_ea *it = (struct osd_it_ea *)di;
6575 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
6579 * Returns a cookie for current position of the iterator head, so that
6580 * user can use this cookie to load/start the iterator next time.
6582 * \param di iterator's in memory structure
6584 * \retval cookie for current position, on success
6586 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
6588 struct osd_it_ea *it = (struct osd_it_ea *)di;
6590 return it->oie_dirent->oied_off;
6594 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6595 * to load a directory entry at a time and stored it i inn,
6596 * in iterator's in-memory data structure.
6598 * \param di struct osd_it_ea, iterator's in memory structure
6600 * \retval +ve on success
6601 * \retval -ve on error
6603 static int osd_it_ea_load(const struct lu_env *env,
6604 const struct dt_it *di, __u64 hash)
6606 struct osd_it_ea *it = (struct osd_it_ea *)di;
6610 it->oie_file.f_pos = hash;
6612 rc = osd_ldiskfs_it_fill(env, di);
6623 * Index lookup function for interoperability mode (b11826).
6625 * \param key, key i.e. file name to be searched
6627 * \retval +ve, on success
6628 * \retval -ve, on error
6630 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
6631 struct dt_rec *rec, const struct dt_key *key)
6633 struct osd_object *obj = osd_dt_obj(dt);
6638 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
6639 LINVRNT(osd_invariant(obj));
6641 rc = osd_ea_lookup_rec(env, obj, rec, key);
6648 * Index and Iterator operations for interoperability
6649 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
6651 static const struct dt_index_operations osd_index_ea_ops = {
6652 .dio_lookup = osd_index_ea_lookup,
6653 .dio_declare_insert = osd_index_declare_ea_insert,
6654 .dio_insert = osd_index_ea_insert,
6655 .dio_declare_delete = osd_index_declare_ea_delete,
6656 .dio_delete = osd_index_ea_delete,
6658 .init = osd_it_ea_init,
6659 .fini = osd_it_ea_fini,
6660 .get = osd_it_ea_get,
6661 .put = osd_it_ea_put,
6662 .next = osd_it_ea_next,
6663 .key = osd_it_ea_key,
6664 .key_size = osd_it_ea_key_size,
6665 .rec = osd_it_ea_rec,
6666 .rec_size = osd_it_ea_rec_size,
6667 .store = osd_it_ea_store,
6668 .load = osd_it_ea_load
6672 static void *osd_key_init(const struct lu_context *ctx,
6673 struct lu_context_key *key)
6675 struct osd_thread_info *info;
6677 OBD_ALLOC_PTR(info);
6679 return ERR_PTR(-ENOMEM);
6681 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6682 if (info->oti_it_ea_buf == NULL)
6685 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
6687 info->oti_hlock = ldiskfs_htree_lock_alloc();
6688 if (info->oti_hlock == NULL)
6694 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6697 return ERR_PTR(-ENOMEM);
6700 static void osd_key_fini(const struct lu_context *ctx,
6701 struct lu_context_key *key, void* data)
6703 struct osd_thread_info *info = data;
6704 struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
6705 struct osd_idmap_cache *idc = info->oti_ins_cache;
6707 if (info->oti_inode != NULL)
6709 if (info->oti_hlock != NULL)
6710 ldiskfs_htree_lock_free(info->oti_hlock);
6711 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6712 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
6713 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
6714 lu_buf_free(&info->oti_big_buf);
6716 LASSERT(info->oti_ins_cache_size > 0);
6717 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
6718 info->oti_ins_cache = NULL;
6719 info->oti_ins_cache_size = 0;
6724 static void osd_key_exit(const struct lu_context *ctx,
6725 struct lu_context_key *key, void *data)
6727 struct osd_thread_info *info = data;
6729 LASSERT(info->oti_r_locks == 0);
6730 LASSERT(info->oti_w_locks == 0);
6731 LASSERT(info->oti_txns == 0);
6734 /* type constructor/destructor: osd_type_init, osd_type_fini */
6735 LU_TYPE_INIT_FINI(osd, &osd_key);
6737 struct lu_context_key osd_key = {
6738 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
6739 .lct_init = osd_key_init,
6740 .lct_fini = osd_key_fini,
6741 .lct_exit = osd_key_exit
6745 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
6746 const char *name, struct lu_device *next)
6748 struct osd_device *osd = osd_dev(d);
6750 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
6751 >= sizeof(osd->od_svname))
6753 return osd_procfs_init(osd, name);
6756 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
6758 struct seq_server_site *ss = osd_seq_site(osd);
6762 if (osd->od_is_ost || osd->od_cl_seq != NULL)
6765 if (unlikely(ss == NULL))
6768 OBD_ALLOC_PTR(osd->od_cl_seq);
6769 if (osd->od_cl_seq == NULL)
6772 rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
6773 osd->od_svname, ss->ss_server_seq);
6775 OBD_FREE_PTR(osd->od_cl_seq);
6776 osd->od_cl_seq = NULL;
6780 if (ss->ss_node_id == 0) {
6781 /* If the OSD on the sequence controller(MDT0), then allocate
6782 * sequence here, otherwise allocate sequence after connected
6783 * to MDT0 (see mdt_register_lwp_callback()). */
6784 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
6785 &osd->od_cl_seq->lcs_space, env);
6791 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
6793 if (osd->od_cl_seq == NULL)
6796 seq_client_fini(osd->od_cl_seq);
6797 OBD_FREE_PTR(osd->od_cl_seq);
6798 osd->od_cl_seq = NULL;
6801 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
6805 /* shutdown quota slave instance associated with the device */
6806 if (o->od_quota_slave != NULL) {
6807 qsd_fini(env, o->od_quota_slave);
6808 o->od_quota_slave = NULL;
6811 osd_fid_fini(env, o);
6816 static void osd_umount(const struct lu_env *env, struct osd_device *o)
6820 if (o->od_mnt != NULL) {
6821 shrink_dcache_sb(osd_sb(o));
6822 osd_sync(env, &o->od_dt_dev);
6831 static int osd_mount(const struct lu_env *env,
6832 struct osd_device *o, struct lustre_cfg *cfg)
6834 const char *name = lustre_cfg_string(cfg, 0);
6835 const char *dev = lustre_cfg_string(cfg, 1);
6837 unsigned long page, s_flags, lmd_flags = 0;
6838 struct page *__page;
6839 struct file_system_type *type;
6840 char *options = NULL;
6842 struct osd_thread_info *info = osd_oti_get(env);
6843 struct lu_fid *fid = &info->oti_fid;
6844 struct inode *inode;
6845 int rc = 0, force_over_512tb = 0;
6848 if (o->od_mnt != NULL)
6851 if (strlen(dev) >= sizeof(o->od_mntdev))
6853 strcpy(o->od_mntdev, dev);
6855 str = lustre_cfg_string(cfg, 2);
6856 s_flags = simple_strtoul(str, NULL, 0);
6857 str = strstr(str, ":");
6859 lmd_flags = simple_strtoul(str + 1, NULL, 0);
6860 opts = lustre_cfg_string(cfg, 3);
6862 if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
6863 CERROR("%s: device %s extents feature is not guaranteed to "
6864 "work on big-endian systems. Use \"bigendian_extents\" "
6865 "mount option to override.\n", name, dev);
6869 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
6870 if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
6871 CWARN("force_over_128tb option is deprecated. "
6872 "Filesystems less than 512TB can be created without any "
6873 "force options. Use force_over_512tb option for "
6874 "filesystems greater than 512TB.\n");
6877 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
6878 if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
6879 CWARN("force_over_256tb option is deprecated. "
6880 "Filesystems less than 512TB can be created without any "
6881 "force options. Use force_over_512tb option for "
6882 "filesystems greater than 512TB.\n");
6886 if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
6887 force_over_512tb = 1;
6889 __page = alloc_page(GFP_KERNEL);
6891 GOTO(out, rc = -ENOMEM);
6892 page = (unsigned long)page_address(__page);
6893 options = (char *)page;
6896 /* strip out the options for back compatiblity */
6897 static char *sout[] = {
6904 /* strip out option we processed in osd */
6905 "bigendian_extents",
6911 strcat(options, opts);
6912 for (rc = 0, str = options; sout[rc]; ) {
6913 char *op = strstr(str, sout[rc]);
6919 if (op == options || *(op - 1) == ',') {
6920 str = op + strlen(sout[rc]);
6921 if (*str == ',' || *str == '\0') {
6922 *str == ',' ? str++ : str;
6923 memmove(op, str, strlen(str) + 1);
6926 for (str = op; *str != ',' && *str != '\0'; str++)
6930 strncat(options, "user_xattr,acl", 14);
6933 /* Glom up mount options */
6934 if (*options != '\0')
6935 strcat(options, ",");
6936 strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
6938 type = get_fs_type("ldiskfs");
6940 CERROR("%s: cannot find ldiskfs module\n", name);
6941 GOTO(out, rc = -ENODEV);
6944 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
6945 module_put(type->owner);
6947 if (IS_ERR(o->od_mnt)) {
6948 rc = PTR_ERR(o->od_mnt);
6950 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
6954 if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
6955 osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
6956 force_over_512tb == 0) {
6957 CERROR("%s: device %s LDISKFS does not support filesystems "
6958 "greater than 512TB and can cause data corruption. "
6959 "Use \"force_over_512tb\" mount option to override.\n",
6961 GOTO(out_mnt, rc = -EINVAL);
6964 if (lmd_flags & LMD_FLG_DEV_RDONLY) {
6965 #ifdef HAVE_DEV_SET_RDONLY
6966 dev_set_rdonly(osd_sb(o)->s_bdev);
6967 o->od_dt_dev.dd_rdonly = 1;
6968 LCONSOLE_WARN("%s: set dev_rdonly on this device\n", name);
6970 LCONSOLE_WARN("%s: not support dev_rdonly on this device",
6973 GOTO(out_mnt, rc = -EOPNOTSUPP);
6976 #ifdef HAVE_DEV_SET_RDONLY
6977 if (dev_check_rdonly(osd_sb(o)->s_bdev)) {
6978 CERROR("%s: underlying device %s is marked as "
6979 "read-only. Setup failed\n", name, dev);
6981 GOTO(out_mnt, rc = -EROFS);
6986 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
6987 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
6988 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
6989 GOTO(out_mnt, rc = -EINVAL);
6992 #ifdef LDISKFS_MOUNT_DIRDATA
6993 if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
6994 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6995 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
6996 else if (!o->od_is_ost)
6997 CWARN("%s: device %s was upgraded from Lustre-1.x without "
6998 "enabling the dirdata feature. If you do not want to "
6999 "downgrade to Lustre-1.x again, you can enable it via "
7000 "'tune2fs -O dirdata device'\n", name, dev);
7002 inode = osd_sb(o)->s_root->d_inode;
7003 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7004 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7006 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7010 if (lmd_flags & LMD_FLG_NOSCRUB)
7021 __free_page(__page);
7026 static struct lu_device *osd_device_fini(const struct lu_env *env,
7027 struct lu_device *d)
7029 struct osd_device *o = osd_dev(d);
7032 osd_shutdown(env, o);
7034 osd_scrub_cleanup(env, o);
7035 osd_obj_map_fini(o);
7041 static int osd_device_init0(const struct lu_env *env,
7042 struct osd_device *o,
7043 struct lustre_cfg *cfg)
7045 struct lu_device *l = osd2lu_dev(o);
7046 struct osd_thread_info *info;
7050 /* if the module was re-loaded, env can loose its keys */
7051 rc = lu_env_refill((struct lu_env *) env);
7054 info = osd_oti_get(env);
7057 l->ld_ops = &osd_lu_ops;
7058 o->od_dt_dev.dd_ops = &osd_dt_ops;
7060 spin_lock_init(&o->od_osfs_lock);
7061 mutex_init(&o->od_otable_mutex);
7062 INIT_LIST_HEAD(&o->od_orphan_list);
7064 o->od_read_cache = 1;
7065 o->od_writethrough_cache = 1;
7066 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7068 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7069 sizeof(o->od_svname));
7070 if (cplen >= sizeof(o->od_svname)) {
7075 o->od_index = -1; /* -1 means index is invalid */
7076 rc = server_name2index(o->od_svname, &o->od_index, NULL);
7077 if (rc == LDD_F_SV_TYPE_OST)
7080 o->od_full_scrub_ratio = OFSR_DEFAULT;
7081 o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7082 rc = osd_mount(env, o, cfg);
7086 rc = osd_obj_map_init(env, o);
7090 rc = lu_site_init(&o->od_site, l);
7092 GOTO(out_compat, rc);
7093 o->od_site.ls_bottom_dev = l;
7095 rc = lu_site_init_finish(&o->od_site);
7099 INIT_LIST_HEAD(&o->od_ios_list);
7100 /* setup scrub, including OI files initialization */
7101 rc = osd_scrub_setup(env, o);
7105 rc = osd_procfs_init(o, o->od_svname);
7107 CERROR("%s: can't initialize procfs: rc = %d\n",
7109 GOTO(out_scrub, rc);
7112 LASSERT(l->ld_site->ls_linkage.next != NULL);
7113 LASSERT(l->ld_site->ls_linkage.prev != NULL);
7115 /* initialize quota slave instance */
7116 o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
7118 if (IS_ERR(o->od_quota_slave)) {
7119 rc = PTR_ERR(o->od_quota_slave);
7120 o->od_quota_slave = NULL;
7121 GOTO(out_procfs, rc);
7129 osd_scrub_cleanup(env, o);
7131 lu_site_fini(&o->od_site);
7133 osd_obj_map_fini(o);
7140 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7141 struct lu_device_type *t,
7142 struct lustre_cfg *cfg)
7144 struct osd_device *o;
7149 return ERR_PTR(-ENOMEM);
7151 rc = dt_device_init(&o->od_dt_dev, t);
7153 /* Because the ctx might be revived in dt_device_init,
7154 * refill the env here */
7155 lu_env_refill((struct lu_env *)env);
7156 rc = osd_device_init0(env, o, cfg);
7158 dt_device_fini(&o->od_dt_dev);
7161 if (unlikely(rc != 0))
7164 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7167 static struct lu_device *osd_device_free(const struct lu_env *env,
7168 struct lu_device *d)
7170 struct osd_device *o = osd_dev(d);
7173 /* XXX: make osd top device in order to release reference */
7174 d->ld_site->ls_top_dev = d;
7175 lu_site_purge(env, d->ld_site, -1);
7176 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7177 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7178 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7180 lu_site_fini(&o->od_site);
7181 dt_device_fini(&o->od_dt_dev);
7186 static int osd_process_config(const struct lu_env *env,
7187 struct lu_device *d, struct lustre_cfg *cfg)
7189 struct osd_device *o = osd_dev(d);
7193 switch (cfg->lcfg_command) {
7195 rc = osd_mount(env, o, cfg);
7198 lu_dev_del_linkage(d->ld_site, d);
7199 rc = osd_shutdown(env, o);
7202 LASSERT(&o->od_dt_dev);
7203 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
7204 cfg, &o->od_dt_dev);
7205 if (rc > 0 || rc == -ENOSYS)
7206 rc = class_process_proc_param(PARAM_OST,
7207 lprocfs_osd_obd_vars,
7208 cfg, &o->od_dt_dev);
7217 static int osd_recovery_complete(const struct lu_env *env,
7218 struct lu_device *d)
7220 struct osd_device *osd = osd_dev(d);
7224 if (osd->od_quota_slave == NULL)
7227 /* start qsd instance on recovery completion, this notifies the quota
7228 * slave code that we are about to process new requests now */
7229 rc = qsd_start(env, osd->od_quota_slave);
7234 * we use exports to track all osd users
7236 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7237 struct obd_device *obd, struct obd_uuid *cluuid,
7238 struct obd_connect_data *data, void *localdata)
7240 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7241 struct lustre_handle conn;
7245 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7247 rc = class_connect(&conn, obd, cluuid);
7251 *exp = class_conn2export(&conn);
7253 spin_lock(&osd->od_osfs_lock);
7255 spin_unlock(&osd->od_osfs_lock);
7261 * once last export (we don't count self-export) disappeared
7262 * osd can be released
7264 static int osd_obd_disconnect(struct obd_export *exp)
7266 struct obd_device *obd = exp->exp_obd;
7267 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7268 int rc, release = 0;
7271 /* Only disconnect the underlying layers on the final disconnect. */
7272 spin_lock(&osd->od_osfs_lock);
7274 if (osd->od_connects == 0)
7276 spin_unlock(&osd->od_osfs_lock);
7278 rc = class_disconnect(exp); /* bz 9811 */
7280 if (rc == 0 && release)
7281 class_manual_cleanup(obd);
7285 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
7286 struct lu_device *dev)
7288 struct osd_device *osd = osd_dev(dev);
7289 struct lr_server_data *lsd =
7290 &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
7294 if (osd->od_quota_slave != NULL) {
7295 /* set up quota slave objects */
7296 result = qsd_prepare(env, osd->od_quota_slave);
7301 if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
7302 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
7303 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
7304 osd->od_index_in_idif = 1;
7306 osd->od_index_in_idif = 0;
7307 result = osd_register_proc_index_in_idif(osd);
7312 osd->od_index_in_idif = 1;
7316 result = osd_fid_init(env, osd);
7321 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
7322 struct lu_fid *fid, struct md_op_data *op_data)
7324 struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
7326 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
7329 static const struct lu_object_operations osd_lu_obj_ops = {
7330 .loo_object_init = osd_object_init,
7331 .loo_object_delete = osd_object_delete,
7332 .loo_object_release = osd_object_release,
7333 .loo_object_free = osd_object_free,
7334 .loo_object_print = osd_object_print,
7335 .loo_object_invariant = osd_object_invariant
7338 const struct lu_device_operations osd_lu_ops = {
7339 .ldo_object_alloc = osd_object_alloc,
7340 .ldo_process_config = osd_process_config,
7341 .ldo_recovery_complete = osd_recovery_complete,
7342 .ldo_prepare = osd_prepare,
7345 static const struct lu_device_type_operations osd_device_type_ops = {
7346 .ldto_init = osd_type_init,
7347 .ldto_fini = osd_type_fini,
7349 .ldto_start = osd_type_start,
7350 .ldto_stop = osd_type_stop,
7352 .ldto_device_alloc = osd_device_alloc,
7353 .ldto_device_free = osd_device_free,
7355 .ldto_device_init = osd_device_init,
7356 .ldto_device_fini = osd_device_fini
7359 static struct lu_device_type osd_device_type = {
7360 .ldt_tags = LU_DEVICE_DT,
7361 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
7362 .ldt_ops = &osd_device_type_ops,
7363 .ldt_ctx_tags = LCT_LOCAL,
7366 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
7368 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7369 struct super_block *sb = osd_sb(osd);
7371 return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
7375 * lprocfs legacy support.
7377 static struct obd_ops osd_obd_device_ops = {
7378 .o_owner = THIS_MODULE,
7379 .o_connect = osd_obd_connect,
7380 .o_disconnect = osd_obd_disconnect,
7381 .o_fid_alloc = osd_fid_alloc,
7382 .o_health_check = osd_health_check,
7385 static int __init osd_init(void)
7389 LASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
7390 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
7391 /* please, try to keep osd_thread_info smaller than a page */
7392 CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
7397 rc = lu_kmem_init(ldiskfs_caches);
7401 rc = class_register_type(&osd_obd_device_ops, NULL, true,
7402 lprocfs_osd_module_vars,
7403 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
7405 lu_kmem_fini(ldiskfs_caches);
7409 static void __exit osd_exit(void)
7411 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
7412 lu_kmem_fini(ldiskfs_caches);
7415 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7416 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
7417 MODULE_VERSION(LUSTRE_VERSION_STRING);
7418 MODULE_LICENSE("GPL");
7420 module_init(osd_init);
7421 module_exit(osd_exit);