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 <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>
82 module_param(ldiskfs_pdo, int, 0644);
83 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
85 int ldiskfs_track_declares_assert;
86 module_param(ldiskfs_track_declares_assert, int, 0644);
87 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
89 /* Slab to allocate dynlocks */
90 struct kmem_cache *dynlock_cachep;
92 /* Slab to allocate osd_it_ea */
93 struct kmem_cache *osd_itea_cachep;
95 static struct lu_kmem_descr ldiskfs_caches[] = {
97 .ckd_cache = &dynlock_cachep,
98 .ckd_name = "dynlock_cache",
99 .ckd_size = sizeof(struct dynlock_handle)
102 .ckd_cache = &osd_itea_cachep,
103 .ckd_name = "osd_itea_cache",
104 .ckd_size = sizeof(struct osd_it_ea)
111 static const char dot[] = ".";
112 static const char dotdot[] = "..";
113 static const char remote_obj_dir[] = "REM_OBJ_DIR";
115 static const struct lu_object_operations osd_lu_obj_ops;
116 static const struct dt_object_operations osd_obj_ops;
117 static const struct dt_object_operations osd_obj_ea_ops;
118 static const struct dt_object_operations osd_obj_otable_it_ops;
119 static const struct dt_index_operations osd_index_iam_ops;
120 static const struct dt_index_operations osd_index_ea_ops;
122 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
123 const struct lu_fid *fid);
124 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
125 struct osd_device *osd);
127 int osd_trans_declare_op2rb[] = {
128 [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
129 [OSD_OT_PUNCH] = OSD_OT_MAX,
130 [OSD_OT_XATTR_SET] = OSD_OT_XATTR_SET,
131 [OSD_OT_CREATE] = OSD_OT_DESTROY,
132 [OSD_OT_DESTROY] = OSD_OT_CREATE,
133 [OSD_OT_REF_ADD] = OSD_OT_REF_DEL,
134 [OSD_OT_REF_DEL] = OSD_OT_REF_ADD,
135 [OSD_OT_WRITE] = OSD_OT_WRITE,
136 [OSD_OT_INSERT] = OSD_OT_DELETE,
137 [OSD_OT_DELETE] = OSD_OT_INSERT,
138 [OSD_OT_QUOTA] = OSD_OT_MAX,
141 static int osd_has_index(const struct osd_object *obj)
143 return obj->oo_dt.do_index_ops != NULL;
146 static int osd_object_invariant(const struct lu_object *l)
148 return osd_invariant(osd_obj(l));
152 * Concurrency: doesn't matter
156 * Concurrency: doesn't matter
158 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
160 struct osd_thread_info *oti = osd_oti_get(env);
161 return oti->oti_w_locks > 0 && o->oo_owner == env;
165 * Concurrency: doesn't access mutable data
167 static int osd_root_get(const struct lu_env *env,
168 struct dt_device *dev, struct lu_fid *f)
170 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
175 * the following set of functions are used to maintain per-thread
176 * cache of FID->ino mapping. this mechanism is needed to resolve
177 * FID to inode at dt_insert() which in turn stores ino in the
178 * directory entries to keep ldiskfs compatible with ext[34].
179 * due to locking-originated restrictions we can't lookup ino
180 * using LU cache (deadlock is possible). lookup using OI is quite
181 * expensive. so instead we maintain this cache and methods like
182 * dt_create() fill it. so in the majority of cases dt_insert() is
183 * able to find needed mapping in lockless manner.
185 static struct osd_idmap_cache *
186 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
187 const struct lu_fid *fid)
189 struct osd_thread_info *oti = osd_oti_get(env);
190 struct osd_idmap_cache *idc = oti->oti_ins_cache;
192 for (i = 0; i < oti->oti_ins_cache_used; i++) {
193 if (!lu_fid_eq(&idc[i].oic_fid, fid))
195 if (idc[i].oic_dev != osd)
204 static struct osd_idmap_cache *
205 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
206 const struct lu_fid *fid)
208 struct osd_thread_info *oti = osd_oti_get(env);
209 struct osd_idmap_cache *idc;
212 if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
213 i = oti->oti_ins_cache_size * 2;
215 i = OSD_INS_CACHE_SIZE;
216 OBD_ALLOC(idc, sizeof(*idc) * i);
218 return ERR_PTR(-ENOMEM);
219 if (oti->oti_ins_cache != NULL) {
220 memcpy(idc, oti->oti_ins_cache,
221 oti->oti_ins_cache_used * sizeof(*idc));
222 OBD_FREE(oti->oti_ins_cache,
223 oti->oti_ins_cache_used * sizeof(*idc));
225 oti->oti_ins_cache = idc;
226 oti->oti_ins_cache_size = i;
229 idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
232 idc->oic_lid.oii_ino = 0;
233 idc->oic_lid.oii_gen = 0;
240 * lookup mapping for the given fid in the cache, initialize a
241 * new one if not found. the initialization checks whether the
242 * object is local or remote. for local objects, OI is used to
243 * learn ino/generation. the function is used when the caller
244 * has no information about the object, e.g. at dt_insert().
246 static struct osd_idmap_cache *
247 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
248 const struct lu_fid *fid)
250 struct osd_idmap_cache *idc;
253 idc = osd_idc_find(env, osd, fid);
254 LASSERT(!IS_ERR(idc));
258 /* new mapping is needed */
259 idc = osd_idc_add(env, osd, fid);
264 rc = osd_remote_fid(env, osd, fid);
265 if (unlikely(rc < 0))
269 /* the object is local, lookup in OI */
270 /* XXX: probably cheaper to lookup in LU first? */
271 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
273 if (unlikely(rc < 0)) {
274 CERROR("can't lookup: rc = %d\n", rc);
278 /* the object is remote */
286 * lookup mapping for given FID and fill it from the given object.
287 * the object is lolcal by definition.
289 static int osd_idc_find_and_init(const struct lu_env *env,
290 struct osd_device *osd,
291 struct osd_object *obj)
293 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
294 struct osd_idmap_cache *idc;
296 idc = osd_idc_find(env, osd, fid);
297 LASSERT(!IS_ERR(idc));
299 if (obj->oo_inode == NULL)
301 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
302 LASSERT(idc->oic_lid.oii_ino == 0);
303 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
304 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
309 /* new mapping is needed */
310 idc = osd_idc_add(env, osd, fid);
314 if (obj->oo_inode != NULL) {
315 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
316 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
322 * OSD object methods.
326 * Concurrency: no concurrent access is possible that early in object
329 static struct lu_object *osd_object_alloc(const struct lu_env *env,
330 const struct lu_object_header *hdr,
333 struct osd_object *mo;
339 l = &mo->oo_dt.do_lu;
340 dt_object_init(&mo->oo_dt, NULL, d);
341 mo->oo_dt.do_ops = &osd_obj_ea_ops;
342 l->lo_ops = &osd_lu_obj_ops;
343 init_rwsem(&mo->oo_sem);
344 init_rwsem(&mo->oo_ext_idx_sem);
345 spin_lock_init(&mo->oo_guard);
346 INIT_LIST_HEAD(&mo->oo_xattr_list);
353 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
354 struct dentry *dentry, struct lustre_mdt_attrs *lma)
358 CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
359 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
360 info->oti_mdt_attrs_old, LMA_OLD_SIZE);
362 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
363 memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
365 lustre_lma_swab(lma);
366 /* Check LMA compatibility */
367 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
368 CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
369 "for fid = "DFID", ino = %lu\n",
370 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
371 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
372 PFID(&lma->lma_self_fid), inode->i_ino);
375 } else if (rc == 0) {
383 * retrieve object from backend ext fs.
385 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
386 struct osd_inode_id *id)
389 struct inode *inode = NULL;
391 /* if we look for an inode withing a running
392 * transaction, then we risk to deadlock */
393 /* osd_dirent_check_repair() breaks this */
394 /*LASSERT(current->journal_info == NULL);*/
396 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
398 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
399 id->oii_ino, PTR_ERR(inode));
400 } else if (id->oii_gen != OSD_OII_NOGEN &&
401 inode->i_generation != id->oii_gen) {
402 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
403 "i_generation = %u\n",
404 id->oii_ino, id->oii_gen, inode->i_generation);
406 inode = ERR_PTR(-ESTALE);
407 } else if (inode->i_nlink == 0) {
408 /* due to parallel readdir and unlink,
409 * we can have dead inode here. */
410 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
412 inode = ERR_PTR(-ESTALE);
413 } else if (is_bad_inode(inode)) {
414 CWARN("%.16s: bad inode: ino = %u\n",
415 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
417 inode = ERR_PTR(-ENOENT);
418 } else if ((rc = osd_attach_jinode(inode))) {
422 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
423 if (id->oii_gen == OSD_OII_NOGEN)
424 osd_id_gen(id, inode->i_ino, inode->i_generation);
426 /* Do not update file c/mtime in ldiskfs.
427 * NB: we don't have any lock to protect this because we don't
428 * have reference on osd_object now, but contention with
429 * another lookup + attr_set can't happen in the tiny window
430 * between if (...) and set S_NOCMTIME. */
431 if (!(inode->i_flags & S_NOCMTIME))
432 inode->i_flags |= S_NOCMTIME;
437 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
438 handle_t *handle, struct dentry *child,
439 struct inode *inode, struct htree_lock *hlock)
443 rc = __ldiskfs_add_entry(handle, child, inode, hlock);
444 if (rc == -ENOBUFS || rc == -ENOSPC) {
445 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
446 struct inode *parent = child->d_parent->d_inode;
447 struct lu_fid *fid = NULL;
449 rc2 = osd_get_lma(info, parent, child->d_parent, lma);
451 fid = &lma->lma_self_fid;
452 } else if (rc2 == -ENODATA) {
453 if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
454 fid = &info->oti_fid3;
455 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
456 } else if (!osd->od_is_ost && osd->od_index == 0) {
457 fid = &info->oti_fid3;
458 lu_igif_build(fid, parent->i_ino,
459 parent->i_generation);
464 CWARN("%s: directory (inode: %lu, FID: "DFID") %s "
465 "maximum entry limit\n",
466 osd_name(osd), parent->i_ino, PFID(fid),
467 rc == -ENOSPC ? "has reached" : "is approaching");
469 CWARN("%s: directory (inode: %lu, FID: unknown) %s "
470 "maximum entry limit\n",
471 osd_name(osd), parent->i_ino,
472 rc == -ENOSPC ? "has reached" : "is approaching");
474 /* ignore such error now */
483 static struct inode *
484 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
485 struct osd_inode_id *id, struct lu_fid *fid)
487 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
491 inode = osd_iget(info, dev, id);
495 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
497 *fid = lma->lma_self_fid;
498 } else if (rc == -ENODATA) {
499 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
500 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
502 lu_igif_build(fid, inode->i_ino, inode->i_generation);
510 static struct inode *osd_iget_check(struct osd_thread_info *info,
511 struct osd_device *dev,
512 const struct lu_fid *fid,
513 struct osd_inode_id *id,
520 /* The cached OI mapping is trustable. If we cannot locate the inode
521 * via the cached OI mapping, then return the failure to the caller
522 * directly without further OI checking. */
524 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
527 if (cached || (rc != -ENOENT && rc != -ESTALE)) {
528 CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
537 if (is_bad_inode(inode)) {
540 CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
548 if (id->oii_gen != OSD_OII_NOGEN &&
549 inode->i_generation != id->oii_gen) {
552 CDEBUG(D_INODE, "unmatched inode: ino = %u, "
553 "oii_gen = %u, i_generation = %u\n",
554 id->oii_ino, id->oii_gen, inode->i_generation);
562 if (inode->i_nlink == 0) {
565 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
573 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
577 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
579 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
580 /* XXX: There are four possible cases:
582 * Backup/restore caused the OI invalid.
584 * Someone unlinked the object but NOT removed
585 * the OI mapping, such as mount target device
586 * as ldiskfs, and modify something directly.
588 * Someone just removed the object between the
589 * former oi_lookup and the iget. It is normal.
590 * 4. Other failure cases.
592 * Generally, when the device is mounted, it will
593 * auto check whether the system is restored from
594 * file-level backup or not. We trust such detect
595 * to distinguish the 1st case from the 2nd case:
596 * if the OI files are consistent but may contain
597 * stale OI mappings because of case 2, if iget()
598 * returns -ENOENT or -ESTALE, then it should be
601 /* If the OI mapping was in OI file before the
602 * osd_iget_check(), but now, it is disappear,
603 * then it must be removed by race. That is a
604 * normal race case. */
607 if ((!IS_ERR(inode) && inode->i_generation != 0 &&
608 inode->i_generation == id->oii_gen) ||
609 (IS_ERR(inode) && !(dev->od_scrub.os_file.sf_flags &
615 if (id->oii_gen == OSD_OII_NOGEN)
616 osd_id_gen(id, inode->i_ino, inode->i_generation);
618 /* Do not update file c/mtime in ldiskfs.
619 * NB: we don't have any lock to protect this because we don't
620 * have reference on osd_object now, but contention with
621 * another lookup + attr_set can't happen in the tiny window
622 * between if (...) and set S_NOCMTIME. */
623 if (!(inode->i_flags & S_NOCMTIME))
624 inode->i_flags |= S_NOCMTIME;
641 * \retval +v: new filter_fid, does not contain self-fid
642 * \retval 0: filter_fid_old, contains self-fid
643 * \retval -v: other failure cases
645 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
646 struct dentry *dentry, struct lu_fid *fid)
648 struct filter_fid_old *ff = &info->oti_ff;
649 struct ost_id *ostid = &info->oti_ostid;
652 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
653 if (rc == sizeof(*ff)) {
655 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
656 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
657 /* XXX: use 0 as the index for compatibility, the caller will
658 * handle index related issues when necessarry. */
659 ostid_to_fid(fid, ostid, 0);
660 } else if (rc == sizeof(struct filter_fid)) {
662 } else if (rc >= 0) {
669 static int osd_lma_self_repair(struct osd_thread_info *info,
670 struct osd_device *osd, struct inode *inode,
671 const struct lu_fid *fid, __u32 compat)
676 LASSERT(current->journal_info == NULL);
678 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
679 osd_dto_credits_noquota[DTO_XATTR_SET]);
682 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
687 rc = osd_ea_fid_set(info, inode, fid, compat, 0);
689 CWARN("%s: cannot self repair the LMA: rc = %d\n",
691 ldiskfs_journal_stop(jh);
695 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
697 struct osd_thread_info *info = osd_oti_get(env);
698 struct osd_device *osd = osd_obj2dev(obj);
699 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
700 struct inode *inode = obj->oo_inode;
701 struct dentry *dentry = &info->oti_obj_dentry;
702 struct lu_fid *fid = NULL;
703 const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
707 CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
708 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
709 info->oti_mdt_attrs_old, LMA_OLD_SIZE);
710 if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
711 fid = &lma->lma_self_fid;
712 rc = osd_get_idif(info, inode, dentry, fid);
713 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
714 /* For the given OST-object, if it has neither LMA nor
715 * FID in XATTR_NAME_FID, then the given FID (which is
716 * contained in the @obj, from client RPC for locating
717 * the OST-object) is trusted. We use it to generate
719 osd_lma_self_repair(info, osd, inode, rfid,
730 lustre_lma_swab(lma);
731 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
732 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
733 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
734 "fid = "DFID", ino = %lu\n", osd_name(osd),
735 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
736 PFID(rfid), inode->i_ino);
739 fid = &lma->lma_self_fid;
743 if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
744 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
745 struct ost_id *oi = &info->oti_ostid;
746 struct lu_fid *fid1 = &info->oti_fid3;
747 __u32 idx = fid_idif_ost_idx(rfid);
749 /* For old IDIF, the OST index is not part of the IDIF,
750 * Means that different OSTs may have the same IDIFs.
751 * Under such case, we need to make some compatible
752 * check to make sure to trigger OI scrub properly. */
753 if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
754 /* Given @rfid is new, LMA is old. */
755 fid_to_ostid(fid, oi);
756 ostid_to_fid(fid1, oi, idx);
757 if (lu_fid_eq(fid1, rfid)) {
758 if (osd->od_index_in_idif)
759 osd_lma_self_repair(info, osd,
773 struct osd_check_lmv_buf {
774 #ifdef HAVE_DIR_CONTEXT
775 /* please keep it as first member */
776 struct dir_context ctx;
778 struct osd_thread_info *oclb_info;
779 struct osd_device *oclb_dev;
780 struct osd_idmap_cache *oclb_oic;
784 * It is called internally by ->readdir() to filter out the
785 * local slave object's FID of the striped directory.
787 * \retval 1 found the local slave's FID
788 * \retval 0 continue to check next item
789 * \retval -ve for failure
791 #ifdef HAVE_FILLDIR_USE_CTX
792 static int osd_stripe_dir_filldir(struct dir_context *buf,
794 static int osd_stripe_dir_filldir(void *buf,
796 const char *name, int namelen,
797 loff_t offset, __u64 ino, unsigned d_type)
799 struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
800 struct osd_thread_info *oti = oclb->oclb_info;
801 struct lu_fid *fid = &oti->oti_fid3;
802 struct osd_inode_id *id = &oti->oti_id3;
803 struct osd_device *dev = oclb->oclb_dev;
804 struct osd_idmap_cache *oic = oclb->oclb_oic;
812 sscanf(name + 1, SFID, RFID(fid));
813 if (!fid_is_sane(fid))
816 if (osd_remote_fid(oti->oti_env, dev, fid))
819 osd_id_gen(id, ino, OSD_OII_NOGEN);
820 inode = osd_iget(oti, dev, id);
822 return PTR_ERR(inode);
825 osd_add_oi_cache(oti, dev, id, fid);
829 rc = osd_oii_insert(dev, oic, true);
831 return rc == 0 ? 1 : rc;
834 /* When lookup item under striped directory, we need to locate the master
835 * MDT-object of the striped directory firstly, then the client will send
836 * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
837 * and the item's name. If the system is restored from MDT file level backup,
838 * then before the OI scrub completely built the OI files, the OI mappings of
839 * the master MDT-object and slave MDT-object may be invalid. Usually, it is
840 * not a problem for the master MDT-object. Because when locate the master
841 * MDT-object, we will do name based lookup (for the striped directory itself)
842 * firstly, during such process we can setup the correct OI mapping for the
843 * master MDT-object. But it will be trouble for the slave MDT-object. Because
844 * the client will not trigger name based lookup on the MDT to locate the slave
845 * MDT-object before locating item under the striped directory, then when
846 * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
847 * is invalid and does not know what the right OI mapping is, then the MDT has
848 * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
849 * the OI file, related OI mapping is unknown yet, please try again later. And
850 * then client will re-try the RPC again and again until related OI mapping has
851 * been updated. That is quite inefficient.
853 * To resolve above trouble, we will handle it as the following two cases:
855 * 1) The slave MDT-object and the master MDT-object are on different MDTs.
856 * It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
857 * is linked under /REMOTE_PARENT_DIR with the name of its FID string.
858 * We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
859 * directly. Please check osd_fid_lookup().
861 * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
862 * Under such case, during lookup the master MDT-object, we will lookup the
863 * slave MDT-object via readdir against the master MDT-object, because the
864 * slave MDT-objects information are stored as sub-directories with the name
865 * "${FID}:${index}". Then when find the local slave MDT-object, its OI
866 * mapping will be recorded. Then subsequent osd_fid_lookup() will know
867 * the correct OI mapping for the slave MDT-object. */
868 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
869 struct inode *inode, struct osd_idmap_cache *oic)
871 struct lu_buf *buf = &oti->oti_big_buf;
872 struct dentry *dentry = &oti->oti_obj_dentry;
873 struct file *filp = &oti->oti_file;
874 const struct file_operations *fops;
875 struct lmv_mds_md_v1 *lmv1;
876 struct osd_check_lmv_buf oclb = {
877 #ifdef HAVE_DIR_CONTEXT
878 .ctx.actor = osd_stripe_dir_filldir,
888 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
891 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
893 lu_buf_realloc(buf, rc);
894 if (buf->lb_buf == NULL)
895 GOTO(out, rc = -ENOMEM);
901 if (unlikely(rc == 0 || rc == -ENODATA))
907 if (unlikely(buf->lb_buf == NULL)) {
908 lu_buf_realloc(buf, rc);
909 if (buf->lb_buf == NULL)
910 GOTO(out, rc = -ENOMEM);
916 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
920 dentry->d_inode = inode;
921 dentry->d_sb = inode->i_sb;
923 filp->f_path.dentry = dentry;
924 filp->f_mode = FMODE_64BITHASH;
925 filp->f_mapping = inode->i_mapping;
927 filp->private_data = NULL;
928 set_file_inode(filp, inode);
930 #ifdef HAVE_DIR_CONTEXT
931 oclb.ctx.pos = filp->f_pos;
932 rc = fops->iterate(filp, &oclb.ctx);
933 filp->f_pos = oclb.ctx.pos;
935 rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir);
937 fops->release(inode, filp);
941 CDEBUG(D_LFSCK, "%.16s: fail to check LMV EA, inode = %lu/%u,"
943 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
944 inode->i_ino, inode->i_generation,
945 PFID(&oic->oic_fid), rc);
952 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
953 const struct lu_fid *fid,
954 const struct lu_object_conf *conf)
956 struct osd_thread_info *info;
957 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
958 struct osd_device *dev;
959 struct osd_idmap_cache *oic;
960 struct osd_inode_id *id;
961 struct osd_inode_id *tid;
962 struct inode *inode = NULL;
963 struct osd_scrub *scrub;
964 struct scrub_file *sf;
965 __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT |
975 LINVRNT(osd_invariant(obj));
976 LASSERT(obj->oo_inode == NULL);
977 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
980 scrub = &dev->od_scrub;
981 sf = &scrub->os_file;
982 info = osd_oti_get(env);
984 oic = &info->oti_cache;
986 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
989 /* For the object is created as locking anchor, or for the object to
990 * be created on disk. No need to osd_oi_lookup() at here because FID
991 * shouldn't never be re-used, if it's really a duplicate FID from
992 * unexpected reason, we should be able to detect it later by calling
993 * do_create->osd_oi_insert(). */
994 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
995 GOTO(out, result = 0);
997 /* Search order: 1. per-thread cache. */
998 if (lu_fid_eq(fid, &oic->oic_fid) &&
999 likely(oic->oic_dev == dev)) {
1005 if (!list_empty(&scrub->os_inconsistent_items)) {
1006 /* Search order: 2. OI scrub pending list. */
1007 result = osd_oii_lookup(dev, fid, id);
1013 /* Search order: 3. OI files. */
1014 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1015 if (result == -ENOENT) {
1016 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1017 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1018 !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
1020 GOTO(out, result = 0);
1029 inode = osd_iget_check(info, dev, fid, id, cached);
1030 if (IS_ERR(inode)) {
1031 result = PTR_ERR(inode);
1032 if (result == -ENOENT || result == -ESTALE)
1033 GOTO(out, result = 0);
1035 if (result == -EREMCHG) {
1038 /* We still have chance to get the valid inode: for the
1039 * object which is referenced by remote name entry, the
1040 * object on the local MDT will be linked under the dir
1041 * of "/REMOTE_PARENT_DIR" with its FID string as name.
1043 * We do not know whether the object for the given FID
1044 * is referenced by some remote name entry or not, and
1045 * especially for DNE II, a multiple-linked object may
1046 * have many name entries reside on many MDTs.
1048 * To simplify the operation, OSD will not distinguish
1049 * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1050 * only happened for the RPC from other MDT during the
1051 * OI scrub, or for the client side RPC with FID only,
1052 * such as FID to path, or from old connected client. */
1054 !fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1055 rc1 = osd_lookup_in_remote_parent(info, dev,
1060 flags |= SS_AUTO_PARTIAL;
1061 flags &= ~SS_AUTO_FULL;
1066 if (thread_is_running(&scrub->os_thread)) {
1067 if (scrub->os_partial_scan &&
1068 !scrub->os_in_join) {
1071 if (inode != NULL && !IS_ERR(inode)) {
1074 osd_add_oi_cache(info, dev, id,
1076 osd_oii_insert(dev, oic, true);
1078 result = -EINPROGRESS;
1081 } else if (!dev->od_noscrub) {
1084 rc1 = osd_scrub_start(dev, flags);
1085 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
1086 "for the "DFID" with flags 0x%x,"
1087 " rc = %d\n", osd_name(dev),
1088 PFID(fid), flags, rc1);
1089 if (rc1 == 0 || rc1 == -EALREADY) {
1090 if (inode != NULL && !IS_ERR(inode)) {
1093 osd_add_oi_cache(info, dev, id,
1095 osd_oii_insert(dev, oic, true);
1097 result = -EINPROGRESS;
1107 if (inode == NULL || IS_ERR(inode))
1109 } else if (remote) {
1113 obj->oo_inode = inode;
1114 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
1116 result = osd_check_lma(env, obj);
1120 tid = &info->oti_id3;
1123 if (result == -ENODATA) {
1125 /* The current OI mapping is from the OI file,
1126 * since the inode has been found via
1127 * osd_iget_check(), no need recheck OI. */
1130 result = osd_oi_lookup(info, dev, fid, tid, OI_CHECK_FLD);
1132 LASSERTF(tid->oii_ino == id->oii_ino &&
1133 tid->oii_gen == id->oii_gen,
1134 "OI mapping changed(1): %u/%u => %u/%u",
1135 tid->oii_ino, tid->oii_gen,
1136 id->oii_ino, id->oii_gen);
1138 LASSERTF(tid->oii_ino == inode->i_ino &&
1139 tid->oii_gen == inode->i_generation,
1140 "locate wrong inode(1): %u/%u => %ld/%u",
1141 tid->oii_ino, tid->oii_gen,
1142 inode->i_ino, inode->i_generation);
1144 /* "result == 0" means the cached OI mapping is still in
1145 * the OI file, so the target the inode is valid. */
1149 /* "result == -ENOENT" means that the OI mappinghas been removed
1150 * by race, the target inode belongs to other object.
1152 * Others error can be returned directly. */
1153 if (result == -ENOENT)
1157 saved_ino = inode->i_ino;
1158 saved_gen = inode->i_generation;
1161 obj->oo_inode = NULL;
1163 if (result != -EREMCHG)
1167 /* The current OI mapping is from the OI file,
1168 * since the inode has been found via
1169 * osd_iget_check(), no need recheck OI. */
1172 result = osd_oi_lookup(info, dev, fid, tid, OI_CHECK_FLD);
1173 /* "result == -ENOENT" means the cached OI mapping has been removed from
1174 * the OI file by race, above target inode belongs to other object.
1176 * Others error can be returned directly. */
1178 GOTO(out, result = (result == -ENOENT ? 0 : result));
1180 LASSERTF(tid->oii_ino == id->oii_ino && tid->oii_gen == id->oii_gen,
1181 "OI mapping changed(2): %u/%u => %u/%u",
1182 tid->oii_ino, tid->oii_gen, id->oii_ino, id->oii_gen);
1184 LASSERTF(tid->oii_ino == saved_ino && tid->oii_gen == saved_gen,
1185 "locate wrong inode(2): %u/%u => %u/%u",
1186 tid->oii_ino, tid->oii_gen, saved_ino, saved_gen);
1191 obj->oo_compat_dot_created = 1;
1192 obj->oo_compat_dotdot_created = 1;
1194 if (S_ISDIR(inode->i_mode) &&
1195 (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1196 osd_check_lmv(info, dev, inode, oic);
1198 result = osd_attach_jinode(inode);
1200 obj->oo_inode = NULL;
1206 GOTO(out, result = 0);
1208 LASSERT(obj->oo_hl_head == NULL);
1209 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1210 if (obj->oo_hl_head == NULL) {
1211 obj->oo_inode = NULL;
1213 GOTO(out, result = -ENOMEM);
1215 GOTO(out, result = 0);
1218 if (result != 0 && cached)
1219 fid_zero(&oic->oic_fid);
1221 LINVRNT(osd_invariant(obj));
1226 * Concurrency: shouldn't matter.
1228 static void osd_object_init0(struct osd_object *obj)
1230 LASSERT(obj->oo_inode != NULL);
1231 obj->oo_dt.do_body_ops = &osd_body_ops;
1232 obj->oo_dt.do_lu.lo_header->loh_attr |=
1233 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1237 * Concurrency: no concurrent access is possible that early in object
1240 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1241 const struct lu_object_conf *conf)
1243 struct osd_object *obj = osd_obj(l);
1246 LINVRNT(osd_invariant(obj));
1248 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1249 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1250 l->lo_header->loh_attr |= LOHA_EXISTS;
1254 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1255 obj->oo_dt.do_body_ops = &osd_body_ops_new;
1256 if (result == 0 && obj->oo_inode != NULL) {
1257 struct osd_thread_info *oti = osd_oti_get(env);
1258 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
1260 osd_object_init0(obj);
1261 result = osd_get_lma(oti, obj->oo_inode,
1262 &oti->oti_obj_dentry, lma);
1264 /* Convert LMAI flags to lustre LMA flags
1265 * and cache it to oo_lma_flags */
1267 lma_to_lustre_flags(lma->lma_incompat);
1268 } else if (result == -ENODATA) {
1273 LINVRNT(osd_invariant(obj));
1277 /* The first part of oxe_buf is xattr name, and is '\0' terminated.
1278 * The left part is for value, binary mode. */
1279 struct osd_xattr_entry {
1280 struct list_head oxe_list;
1284 struct rcu_head oxe_rcu;
1288 static int osd_oxc_get(struct osd_object *obj, const char *name,
1291 struct osd_xattr_entry *tmp;
1292 struct osd_xattr_entry *oxe = NULL;
1293 size_t namelen = strlen(name);
1298 list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1299 if (namelen == tmp->oxe_namelen &&
1300 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1307 GOTO(out, rc = -ENOENT);
1309 if (!oxe->oxe_exist)
1310 GOTO(out, rc = -ENODATA);
1313 rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1316 if (buf->lb_buf == NULL)
1319 if (buf->lb_len < rc)
1320 GOTO(out, rc = -ERANGE);
1322 memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1330 static void osd_oxc_free(struct rcu_head *head)
1332 struct osd_xattr_entry *oxe;
1334 oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1335 OBD_FREE(oxe, oxe->oxe_len);
1338 static void osd_oxc_add(struct osd_object *obj, const char *name,
1339 const char *buf, int buflen)
1341 struct osd_xattr_entry *oxe;
1342 struct osd_xattr_entry *old = NULL;
1343 struct osd_xattr_entry *tmp;
1344 size_t namelen = strlen(name);
1345 size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1347 OBD_ALLOC(oxe, len);
1351 INIT_LIST_HEAD(&oxe->oxe_list);
1353 oxe->oxe_namelen = namelen;
1354 memcpy(oxe->oxe_buf, name, namelen);
1356 LASSERT(buf != NULL);
1357 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1358 oxe->oxe_exist = true;
1360 oxe->oxe_exist = false;
1363 /* this should be rarely called, just remove old and add new */
1364 spin_lock(&obj->oo_guard);
1365 list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1366 if (namelen == tmp->oxe_namelen &&
1367 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1373 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1374 call_rcu(&old->oxe_rcu, osd_oxc_free);
1376 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1378 spin_unlock(&obj->oo_guard);
1381 static void osd_oxc_del(struct osd_object *obj, const char *name)
1383 struct osd_xattr_entry *oxe;
1384 size_t namelen = strlen(name);
1386 spin_lock(&obj->oo_guard);
1387 list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1388 if (namelen == oxe->oxe_namelen &&
1389 strncmp(name, oxe->oxe_buf, namelen) == 0) {
1390 list_del_rcu(&oxe->oxe_list);
1391 call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1395 spin_unlock(&obj->oo_guard);
1398 static void osd_oxc_fini(struct osd_object *obj)
1400 struct osd_xattr_entry *oxe, *next;
1402 list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1403 list_del(&oxe->oxe_list);
1404 OBD_FREE(oxe, oxe->oxe_len);
1409 * Concurrency: no concurrent access is possible that late in object
1412 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1414 struct osd_object *obj = osd_obj(l);
1416 LINVRNT(osd_invariant(obj));
1419 dt_object_fini(&obj->oo_dt);
1420 if (obj->oo_hl_head != NULL)
1421 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1426 * Concurrency: no concurrent access is possible that late in object
1429 static void osd_index_fini(struct osd_object *o)
1431 struct iam_container *bag;
1433 if (o->oo_dir != NULL) {
1434 bag = &o->oo_dir->od_container;
1435 if (o->oo_inode != NULL) {
1436 if (bag->ic_object == o->oo_inode)
1437 iam_container_fini(bag);
1439 OBD_FREE_PTR(o->oo_dir);
1445 * Concurrency: no concurrent access is possible that late in object
1446 * life-cycle (for all existing callers, that is. New callers have to provide
1447 * their own locking.)
1449 static int osd_inode_unlinked(const struct inode *inode)
1451 return inode->i_nlink == 0;
1455 OSD_TXN_OI_DELETE_CREDITS = 20,
1456 OSD_TXN_INODE_DELETE_CREDITS = 20
1463 #if OSD_THANDLE_STATS
1465 * Set time when the handle is allocated
1467 static void osd_th_alloced(struct osd_thandle *oth)
1469 oth->oth_alloced = ktime_get();
1473 * Set time when the handle started
1475 static void osd_th_started(struct osd_thandle *oth)
1477 oth->oth_started = ktime_get();
1481 * Check whether the we deal with this handle for too long.
1483 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1484 ktime_t alloced, ktime_t started,
1487 ktime_t now = ktime_get();
1489 LASSERT(dev != NULL);
1491 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1492 ktime_us_delta(started, alloced));
1493 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1494 ktime_us_delta(closed, started));
1495 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1496 ktime_us_delta(now, closed));
1498 if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1499 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1500 oth, now, alloced, started, closed);
1501 libcfs_debug_dumpstack(NULL);
1505 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
1507 ktime_t __closed = ktime_get(); \
1508 ktime_t __alloced = oth->oth_alloced; \
1509 ktime_t __started = oth->oth_started; \
1512 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
1515 #else /* OSD_THANDLE_STATS */
1517 #define osd_th_alloced(h) do {} while(0)
1518 #define osd_th_started(h) do {} while(0)
1519 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
1521 #endif /* OSD_THANDLE_STATS */
1524 * Concurrency: doesn't access mutable data.
1526 static int osd_param_is_not_sane(const struct osd_device *dev,
1527 const struct thandle *th)
1529 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1531 return oh->ot_credits > osd_transaction_size(dev);
1535 * Concurrency: shouldn't matter.
1537 static void osd_trans_commit_cb(struct super_block *sb,
1538 struct ldiskfs_journal_cb_entry *jcb, int error)
1540 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1541 struct thandle *th = &oh->ot_super;
1542 struct lu_device *lud = &th->th_dev->dd_lu_dev;
1543 struct dt_txn_commit_cb *dcb, *tmp;
1545 LASSERT(oh->ot_handle == NULL);
1548 CERROR("transaction @0x%p commit error: %d\n", th, error);
1550 dt_txn_hook_commit(th);
1552 /* call per-transaction callbacks if any */
1553 list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1555 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1556 "commit callback entry: magic=%x name='%s'\n",
1557 dcb->dcb_magic, dcb->dcb_name);
1558 list_del_init(&dcb->dcb_linkage);
1559 dcb->dcb_func(NULL, th, dcb, error);
1562 lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1566 lu_context_exit(&th->th_ctx);
1567 lu_context_fini(&th->th_ctx);
1571 #ifndef HAVE_SB_START_WRITE
1572 # define sb_start_write(sb) do {} while (0)
1573 # define sb_end_write(sb) do {} while (0)
1576 static struct thandle *osd_trans_create(const struct lu_env *env,
1577 struct dt_device *d)
1579 struct osd_thread_info *oti = osd_oti_get(env);
1580 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1581 struct osd_thandle *oh;
1585 /* on pending IO in this thread should left from prev. request */
1586 LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1588 sb_start_write(osd_sb(osd_dt_dev(d)));
1590 OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
1592 oh->ot_quota_trans = &oti->oti_quota_trans;
1593 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1597 th->th_tags = LCT_TX_HANDLE;
1599 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1600 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1603 memset(oti->oti_declare_ops, 0,
1604 sizeof(oti->oti_declare_ops));
1605 memset(oti->oti_declare_ops_cred, 0,
1606 sizeof(oti->oti_declare_ops_cred));
1607 memset(oti->oti_declare_ops_used, 0,
1608 sizeof(oti->oti_declare_ops_used));
1610 sb_end_write(osd_sb(osd_dt_dev(d)));
1611 th = ERR_PTR(-ENOMEM);
1616 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1618 struct osd_thread_info *oti = osd_oti_get(env);
1619 struct osd_thandle *oh;
1621 oh = container_of0(th, struct osd_thandle, ot_super);
1622 LASSERT(oh != NULL);
1624 CWARN(" create: %u/%u/%u, destroy: %u/%u/%u\n",
1625 oti->oti_declare_ops[OSD_OT_CREATE],
1626 oti->oti_declare_ops_cred[OSD_OT_CREATE],
1627 oti->oti_declare_ops_used[OSD_OT_CREATE],
1628 oti->oti_declare_ops[OSD_OT_DESTROY],
1629 oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1630 oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1631 CWARN(" attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1632 oti->oti_declare_ops[OSD_OT_ATTR_SET],
1633 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1634 oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1635 oti->oti_declare_ops[OSD_OT_XATTR_SET],
1636 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1637 oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1638 CWARN(" write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1639 oti->oti_declare_ops[OSD_OT_WRITE],
1640 oti->oti_declare_ops_cred[OSD_OT_WRITE],
1641 oti->oti_declare_ops_used[OSD_OT_WRITE],
1642 oti->oti_declare_ops[OSD_OT_PUNCH],
1643 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1644 oti->oti_declare_ops_used[OSD_OT_PUNCH],
1645 oti->oti_declare_ops[OSD_OT_QUOTA],
1646 oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1647 oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1648 CWARN(" insert: %u/%u/%u, delete: %u/%u/%u\n",
1649 oti->oti_declare_ops[OSD_OT_INSERT],
1650 oti->oti_declare_ops_cred[OSD_OT_INSERT],
1651 oti->oti_declare_ops_used[OSD_OT_INSERT],
1652 oti->oti_declare_ops[OSD_OT_DELETE],
1653 oti->oti_declare_ops_cred[OSD_OT_DELETE],
1654 oti->oti_declare_ops_used[OSD_OT_DELETE]);
1655 CWARN(" ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1656 oti->oti_declare_ops[OSD_OT_REF_ADD],
1657 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1658 oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1659 oti->oti_declare_ops[OSD_OT_REF_DEL],
1660 oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1661 oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1665 * Concurrency: shouldn't matter.
1667 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1670 struct osd_thread_info *oti = osd_oti_get(env);
1671 struct osd_device *dev = osd_dt_dev(d);
1673 struct osd_thandle *oh;
1678 LASSERT(current->journal_info == NULL);
1680 oh = container_of0(th, struct osd_thandle, ot_super);
1681 LASSERT(oh != NULL);
1682 LASSERT(oh->ot_handle == NULL);
1684 rc = dt_txn_hook_start(env, d, th);
1688 if (unlikely(osd_param_is_not_sane(dev, th))) {
1689 static unsigned long last_printed;
1690 static int last_credits;
1692 /* don't make noise on a tiny testing systems
1693 * actual credits misuse will be caught anyway */
1694 if (last_credits != oh->ot_credits &&
1695 time_after(jiffies, last_printed +
1696 msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1697 osd_transaction_size(dev) > 512) {
1698 CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1699 oh->ot_credits, osd_transaction_size(dev));
1700 osd_trans_dump_creds(env, th);
1701 libcfs_debug_dumpstack(NULL);
1702 last_credits = oh->ot_credits;
1703 last_printed = jiffies;
1705 /* XXX Limit the credits to 'max_transaction_buffers', and
1706 * let the underlying filesystem to catch the error if
1707 * we really need so many credits.
1709 * This should be removed when we can calculate the
1710 * credits precisely. */
1711 oh->ot_credits = osd_transaction_size(dev);
1712 } else if (ldiskfs_track_declares_assert != 0) {
1713 /* reserve few credits to prevent an assertion in JBD
1714 * our debugging mechanism will be able to detected
1715 * overuse. this can help to debug single-update
1717 oh->ot_credits += 10;
1718 if (unlikely(osd_param_is_not_sane(dev, th)))
1719 oh->ot_credits = osd_transaction_size(dev);
1723 * XXX temporary stuff. Some abstraction layer should
1726 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1730 LASSERT(oti->oti_txns == 0);
1731 lu_context_init(&th->th_ctx, th->th_tags);
1732 lu_context_enter(&th->th_ctx);
1734 lu_device_get(&d->dd_lu_dev);
1735 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1746 static int osd_seq_exists(const struct lu_env *env,
1747 struct osd_device *osd, u64 seq)
1749 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1750 struct seq_server_site *ss = osd_seq_site(osd);
1754 LASSERT(ss != NULL);
1755 LASSERT(ss->ss_server_fld != NULL);
1757 rc = osd_fld_lookup(env, osd, seq, range);
1760 CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1761 osd_name(osd), seq, rc);
1765 RETURN(ss->ss_node_id == range->lsr_index);
1768 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1770 struct dt_txn_commit_cb *dcb;
1771 struct dt_txn_commit_cb *tmp;
1773 /* call per-transaction stop callbacks if any */
1774 list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1776 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1777 "commit callback entry: magic=%x name='%s'\n",
1778 dcb->dcb_magic, dcb->dcb_name);
1779 list_del_init(&dcb->dcb_linkage);
1780 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1785 * Concurrency: shouldn't matter.
1787 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1790 int rc = 0, remove_agents = 0;
1791 struct osd_thandle *oh;
1792 struct osd_thread_info *oti = osd_oti_get(env);
1793 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1794 struct osd_device *osd = osd_dt_dev(th->th_dev);
1795 struct qsd_instance *qsd = osd->od_quota_slave;
1796 struct lquota_trans *qtrans;
1799 oh = container_of0(th, struct osd_thandle, ot_super);
1801 /* reset OI cache for safety */
1802 oti->oti_ins_cache_used = 0;
1804 remove_agents = oh->ot_remove_agents;
1806 qtrans = oh->ot_quota_trans;
1807 oh->ot_quota_trans = NULL;
1809 if (oh->ot_handle != NULL) {
1811 handle_t *hdl = oh->ot_handle;
1814 * add commit callback
1815 * notice we don't do this in osd_trans_start()
1816 * as underlying transaction can change during truncate
1818 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1821 LASSERT(oti->oti_txns == 1);
1824 rc = dt_txn_hook_stop(env, th);
1826 CERROR("%s: failed in transaction hook: rc = %d\n",
1829 osd_trans_stop_cb(oh, rc);
1830 /* hook functions might modify th_sync */
1831 hdl->h_sync = th->th_sync;
1833 oh->ot_handle = NULL;
1834 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
1836 CERROR("%s: failed to stop transaction: rc = %d\n",
1837 osd_name(osd), rc2);
1841 osd_trans_stop_cb(oh, th->th_result);
1845 /* inform the quota slave device that the transaction is stopping */
1846 qsd_op_end(env, qsd, qtrans);
1848 /* as we want IO to journal and data IO be concurrent, we don't block
1849 * awaiting data IO completion in osd_do_bio(), instead we wait here
1850 * once transaction is submitted to the journal. all reqular requests
1851 * don't do direct IO (except read/write), thus this wait_event becomes
1854 * IMPORTANT: we have to wait till any IO submited by the thread is
1855 * completed otherwise iobuf may be corrupted by different request
1857 wait_event(iobuf->dr_wait,
1858 atomic_read(&iobuf->dr_numreqs) == 0);
1859 osd_fini_iobuf(osd, iobuf);
1861 rc = iobuf->dr_error;
1863 if (unlikely(remove_agents != 0))
1864 osd_process_scheduled_agent_removals(env, osd);
1866 sb_end_write(osd_sb(osd));
1871 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1873 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1876 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1877 LASSERT(&dcb->dcb_func != NULL);
1878 if (dcb->dcb_flags & DCB_TRANS_STOP)
1879 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1881 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1887 * Called just before object is freed. Releases all resources except for
1888 * object itself (that is released by osd_object_free()).
1890 * Concurrency: no concurrent access is possible that late in object
1893 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1895 struct osd_object *obj = osd_obj(l);
1896 struct inode *inode = obj->oo_inode;
1898 LINVRNT(osd_invariant(obj));
1901 * If object is unlinked remove fid->ino mapping from object index.
1904 osd_index_fini(obj);
1905 if (inode != NULL) {
1906 struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave;
1907 qid_t uid = i_uid_read(inode);
1908 qid_t gid = i_gid_read(inode);
1911 obj->oo_inode = NULL;
1914 struct osd_thread_info *info = osd_oti_get(env);
1915 struct lquota_id_info *qi = &info->oti_qi;
1917 /* Release granted quota to master if necessary */
1918 qi->lqi_id.qid_uid = uid;
1919 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1921 qi->lqi_id.qid_uid = gid;
1922 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1928 * Concurrency: ->loo_object_release() is called under site spin-lock.
1930 static void osd_object_release(const struct lu_env *env,
1931 struct lu_object *l)
1933 struct osd_object *o = osd_obj(l);
1934 /* nobody should be releasing a non-destroyed object with nlink=0
1935 * the API allows this, but ldiskfs doesn't like and then report
1936 * this inode as deleted */
1937 if (unlikely(!o->oo_destroyed && o->oo_inode && o->oo_inode->i_nlink == 0))
1942 * Concurrency: shouldn't matter.
1944 static int osd_object_print(const struct lu_env *env, void *cookie,
1945 lu_printer_t p, const struct lu_object *l)
1947 struct osd_object *o = osd_obj(l);
1948 struct iam_descr *d;
1950 if (o->oo_dir != NULL)
1951 d = o->oo_dir->od_container.ic_descr;
1954 return (*p)(env, cookie,
1955 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1957 o->oo_inode ? o->oo_inode->i_ino : 0UL,
1958 o->oo_inode ? o->oo_inode->i_generation : 0,
1959 d ? d->id_ops->id_name : "plain");
1963 * Concurrency: shouldn't matter.
1965 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1966 struct obd_statfs *sfs)
1968 struct osd_device *osd = osd_dt_dev(d);
1969 struct super_block *sb = osd_sb(osd);
1970 struct kstatfs *ksfs;
1974 if (unlikely(osd->od_mnt == NULL))
1975 return -EINPROGRESS;
1977 /* osd_lproc.c call this without env, allocate ksfs for that case */
1978 if (unlikely(env == NULL)) {
1979 OBD_ALLOC_PTR(ksfs);
1983 ksfs = &osd_oti_get(env)->oti_ksfs;
1986 result = sb->s_op->statfs(sb->s_root, ksfs);
1990 statfs_pack(sfs, ksfs);
1991 if (unlikely(sb->s_flags & MS_RDONLY))
1992 sfs->os_state |= OS_STATE_READONLY;
1993 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb,
1994 LDISKFS_FEATURE_INCOMPAT_EXTENTS))
1995 sfs->os_maxbytes = sb->s_maxbytes;
1997 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2000 * Reserve some space so to avoid fragmenting the filesystem too much.
2001 * Fragmentation not only impacts performance, but can also increase
2002 * metadata overhead significantly, causing grant calculation to be
2005 * Reserve 0.78% of total space, at least 8MB for small filesystems.
2007 CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2008 reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2009 if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2010 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2012 sfs->os_blocks -= reserved;
2013 sfs->os_bfree -= min(reserved, sfs->os_bfree);
2014 sfs->os_bavail -= min(reserved, sfs->os_bavail);
2017 if (unlikely(env == NULL))
2023 * Estimate space needed for file creations. We assume the largest filename
2024 * which is 2^64 - 1, hence a filename of 20 chars.
2025 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2027 #ifdef __LDISKFS_DIR_REC_LEN
2028 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2030 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2034 * Concurrency: doesn't access mutable data.
2036 static void osd_conf_get(const struct lu_env *env,
2037 const struct dt_device *dev,
2038 struct dt_device_param *param)
2040 struct super_block *sb = osd_sb(osd_dt_dev(dev));
2044 * XXX should be taken from not-yet-existing fs abstraction layer.
2046 param->ddp_max_name_len = LDISKFS_NAME_LEN;
2047 param->ddp_max_nlink = LDISKFS_LINK_MAX;
2048 param->ddp_symlink_max = sb->s_blocksize;
2049 param->ddp_mount_type = LDD_MT_LDISKFS;
2050 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2051 param->ddp_maxbytes = sb->s_maxbytes;
2053 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2054 /* inode are statically allocated, so per-inode space consumption
2055 * is the space consumed by the directory entry */
2056 param->ddp_inodespace = PER_OBJ_USAGE;
2057 /* EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2058 * = 128MB) which is unlikely to be hit in real life. Report a smaller
2059 * maximum length to not under count the actual number of extents
2060 * needed for writing a file. */
2061 param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2;
2062 /* worst-case extent insertion metadata overhead */
2063 param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2064 param->ddp_mntopts = 0;
2065 if (test_opt(sb, XATTR_USER))
2066 param->ddp_mntopts |= MNTOPT_USERXATTR;
2067 if (test_opt(sb, POSIX_ACL))
2068 param->ddp_mntopts |= MNTOPT_ACL;
2070 /* LOD might calculate the max stripe count based on max_ea_size,
2071 * so we need take account in the overhead as well,
2072 * xattr_header + magic + xattr_entry_head */
2073 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2074 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2076 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2077 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
2078 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2082 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2086 * Concurrency: shouldn't matter.
2088 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2092 CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
2094 rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
2096 CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
2097 LUSTRE_OSD_LDISKFS_NAME, rc);
2103 * Start commit for OSD device.
2105 * An implementation of dt_commit_async method for OSD device.
2106 * Asychronously starts underlayng fs sync and thereby a transaction
2109 * \param env environment
2110 * \param d dt device
2112 * \see dt_device_operations
2114 static int osd_commit_async(const struct lu_env *env,
2115 struct dt_device *d)
2117 struct super_block *s = osd_sb(osd_dt_dev(d));
2120 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
2121 RETURN(s->s_op->sync_fs(s, 0));
2125 * Concurrency: shouldn't matter.
2128 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2130 struct super_block *sb = osd_sb(osd_dt_dev(d));
2131 struct block_device *dev = sb->s_bdev;
2132 #ifdef HAVE_DEV_SET_RDONLY
2133 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2136 int rc = -EOPNOTSUPP;
2140 #ifdef HAVE_DEV_SET_RDONLY
2141 CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
2143 if (sb->s_op->freeze_fs) {
2144 rc = sb->s_op->freeze_fs(sb);
2149 if (jdev && (jdev != dev)) {
2150 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2152 dev_set_rdonly(jdev);
2154 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2155 dev_set_rdonly(dev);
2157 if (sb->s_op->unfreeze_fs)
2158 sb->s_op->unfreeze_fs(sb);
2163 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2164 osd_dt_dev(d)->od_svname, (long)dev, rc);
2170 * Note: we do not count into QUOTA here.
2171 * If we mount with --data_journal we may need more.
2173 const int osd_dto_credits_noquota[DTO_NR] = {
2176 * INDEX_EXTRA_TRANS_BLOCKS(8) +
2177 * SINGLEDATA_TRANS_BLOCKS(8)
2178 * XXX Note: maybe iam need more, since iam have more level than
2181 [DTO_INDEX_INSERT] = 16,
2184 * just modify a single entry, probably merge few within a block
2186 [DTO_INDEX_DELETE] = 1,
2190 [DTO_INDEX_UPDATE] = 16,
2192 * 4(inode, inode bits, groups, GDT)
2193 * notice: OI updates are counted separately with DTO_INDEX_INSERT
2195 [DTO_OBJECT_CREATE] = 4,
2197 * 4(inode, inode bits, groups, GDT)
2198 * notice: OI updates are counted separately with DTO_INDEX_DELETE
2200 [DTO_OBJECT_DELETE] = 4,
2202 * Attr set credits (inode)
2204 [DTO_ATTR_SET_BASE] = 1,
2206 * Xattr set. The same as xattr of EXT3.
2207 * DATA_TRANS_BLOCKS(14)
2208 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2209 * are also counted in. Do not know why?
2211 [DTO_XATTR_SET] = 14,
2213 * credits for inode change during write.
2215 [DTO_WRITE_BASE] = 3,
2217 * credits for single block write.
2219 [DTO_WRITE_BLOCK] = 14,
2221 * Attr set credits for chown.
2222 * This is extra credits for setattr, and it is null without quota
2224 [DTO_ATTR_SET_CHOWN] = 0
2227 static const struct dt_device_operations osd_dt_ops = {
2228 .dt_root_get = osd_root_get,
2229 .dt_statfs = osd_statfs,
2230 .dt_trans_create = osd_trans_create,
2231 .dt_trans_start = osd_trans_start,
2232 .dt_trans_stop = osd_trans_stop,
2233 .dt_trans_cb_add = osd_trans_cb_add,
2234 .dt_conf_get = osd_conf_get,
2235 .dt_sync = osd_sync,
2237 .dt_commit_async = osd_commit_async,
2240 static void osd_object_read_lock(const struct lu_env *env,
2241 struct dt_object *dt, unsigned role)
2243 struct osd_object *obj = osd_dt_obj(dt);
2244 struct osd_thread_info *oti = osd_oti_get(env);
2246 LINVRNT(osd_invariant(obj));
2248 LASSERT(obj->oo_owner != env);
2249 down_read_nested(&obj->oo_sem, role);
2251 LASSERT(obj->oo_owner == NULL);
2255 static void osd_object_write_lock(const struct lu_env *env,
2256 struct dt_object *dt, unsigned role)
2258 struct osd_object *obj = osd_dt_obj(dt);
2259 struct osd_thread_info *oti = osd_oti_get(env);
2261 LINVRNT(osd_invariant(obj));
2263 LASSERT(obj->oo_owner != env);
2264 down_write_nested(&obj->oo_sem, role);
2266 LASSERT(obj->oo_owner == NULL);
2267 obj->oo_owner = env;
2271 static void osd_object_read_unlock(const struct lu_env *env,
2272 struct dt_object *dt)
2274 struct osd_object *obj = osd_dt_obj(dt);
2275 struct osd_thread_info *oti = osd_oti_get(env);
2277 LINVRNT(osd_invariant(obj));
2279 LASSERT(oti->oti_r_locks > 0);
2281 up_read(&obj->oo_sem);
2284 static void osd_object_write_unlock(const struct lu_env *env,
2285 struct dt_object *dt)
2287 struct osd_object *obj = osd_dt_obj(dt);
2288 struct osd_thread_info *oti = osd_oti_get(env);
2290 LINVRNT(osd_invariant(obj));
2292 LASSERT(obj->oo_owner == env);
2293 LASSERT(oti->oti_w_locks > 0);
2295 obj->oo_owner = NULL;
2296 up_write(&obj->oo_sem);
2299 static int osd_object_write_locked(const struct lu_env *env,
2300 struct dt_object *dt)
2302 struct osd_object *obj = osd_dt_obj(dt);
2304 LINVRNT(osd_invariant(obj));
2306 return obj->oo_owner == env;
2309 static struct timespec *osd_inode_time(const struct lu_env *env,
2310 struct inode *inode, __u64 seconds)
2312 struct osd_thread_info *oti = osd_oti_get(env);
2313 struct timespec *t = &oti->oti_time;
2315 t->tv_sec = seconds;
2317 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2321 static void osd_inode_getattr(const struct lu_env *env,
2322 struct inode *inode, struct lu_attr *attr)
2324 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2325 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2326 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
2329 attr->la_atime = LTIME_S(inode->i_atime);
2330 attr->la_mtime = LTIME_S(inode->i_mtime);
2331 attr->la_ctime = LTIME_S(inode->i_ctime);
2332 attr->la_mode = inode->i_mode;
2333 attr->la_size = i_size_read(inode);
2334 attr->la_blocks = inode->i_blocks;
2335 attr->la_uid = i_uid_read(inode);
2336 attr->la_gid = i_gid_read(inode);
2337 attr->la_flags = ll_inode_to_ext_flags(inode->i_flags);
2338 attr->la_nlink = inode->i_nlink;
2339 attr->la_rdev = inode->i_rdev;
2340 attr->la_blksize = 1 << inode->i_blkbits;
2341 attr->la_blkbits = inode->i_blkbits;
2344 static int osd_attr_get(const struct lu_env *env,
2345 struct dt_object *dt,
2346 struct lu_attr *attr)
2348 struct osd_object *obj = osd_dt_obj(dt);
2350 if (unlikely(!dt_object_exists(dt)))
2352 if (unlikely(obj->oo_destroyed))
2355 LASSERT(!dt_object_remote(dt));
2356 LINVRNT(osd_invariant(obj));
2358 spin_lock(&obj->oo_guard);
2359 osd_inode_getattr(env, obj->oo_inode, attr);
2360 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2361 attr->la_flags |= LUSTRE_ORPHAN_FL;
2362 spin_unlock(&obj->oo_guard);
2367 static int osd_declare_attr_set(const struct lu_env *env,
2368 struct dt_object *dt,
2369 const struct lu_attr *attr,
2370 struct thandle *handle)
2372 struct osd_thandle *oh;
2373 struct osd_object *obj;
2374 struct osd_thread_info *info = osd_oti_get(env);
2375 struct lquota_id_info *qi = &info->oti_qi;
2383 LASSERT(dt != NULL);
2384 LASSERT(handle != NULL);
2386 obj = osd_dt_obj(dt);
2387 LASSERT(osd_invariant(obj));
2389 oh = container_of0(handle, struct osd_thandle, ot_super);
2390 LASSERT(oh->ot_handle == NULL);
2392 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2393 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2395 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2396 osd_dto_credits_noquota[DTO_XATTR_SET]);
2398 if (attr == NULL || obj->oo_inode == NULL)
2401 bspace = obj->oo_inode->i_blocks;
2402 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
2403 bspace = toqb(bspace);
2405 /* Changing ownership is always preformed by super user, it should not
2408 * We still need to call the osd_declare_qid() to calculate the journal
2409 * credits for updating quota accounting files and to trigger quota
2410 * space adjustment once the operation is completed.*/
2411 if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2413 uid = i_uid_read(obj->oo_inode);
2414 qi->lqi_type = USRQUOTA;
2415 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2416 /* inode accounting */
2417 qi->lqi_is_blk = false;
2419 /* one more inode for the new uid ... */
2420 qi->lqi_id.qid_uid = attr->la_uid;
2422 /* Reserve credits for the new uid */
2423 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2424 if (rc == -EDQUOT || rc == -EINPROGRESS)
2429 /* and one less inode for the current uid */
2430 qi->lqi_id.qid_uid = uid;
2432 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2433 if (rc == -EDQUOT || rc == -EINPROGRESS)
2438 /* block accounting */
2439 qi->lqi_is_blk = true;
2441 /* more blocks for the new uid ... */
2442 qi->lqi_id.qid_uid = attr->la_uid;
2443 qi->lqi_space = bspace;
2445 * Credits for the new uid has been reserved, re-use "obj"
2446 * to save credit reservation.
2448 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2449 if (rc == -EDQUOT || rc == -EINPROGRESS)
2454 /* and finally less blocks for the current uid */
2455 qi->lqi_id.qid_uid = uid;
2456 qi->lqi_space = -bspace;
2457 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2458 if (rc == -EDQUOT || rc == -EINPROGRESS)
2464 gid = i_gid_read(obj->oo_inode);
2465 qi->lqi_type = GRPQUOTA;
2466 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2468 /* inode accounting */
2469 qi->lqi_is_blk = false;
2471 /* one more inode for the new gid ... */
2472 qi->lqi_id.qid_gid = attr->la_gid;
2474 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2475 if (rc == -EDQUOT || rc == -EINPROGRESS)
2480 /* and one less inode for the current gid */
2481 qi->lqi_id.qid_gid = gid;
2483 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2484 if (rc == -EDQUOT || rc == -EINPROGRESS)
2489 /* block accounting */
2490 qi->lqi_is_blk = true;
2492 /* more blocks for the new gid ... */
2493 qi->lqi_id.qid_gid = attr->la_gid;
2494 qi->lqi_space = bspace;
2495 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2496 if (rc == -EDQUOT || rc == -EINPROGRESS)
2501 /* and finally less blocks for the current gid */
2502 qi->lqi_id.qid_gid = gid;
2503 qi->lqi_space = -bspace;
2504 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2505 if (rc == -EDQUOT || rc == -EINPROGRESS)
2514 static int osd_inode_setattr(const struct lu_env *env,
2515 struct inode *inode, const struct lu_attr *attr)
2517 __u64 bits = attr->la_valid;
2519 /* Only allow set size for regular file */
2520 if (!S_ISREG(inode->i_mode))
2521 bits &= ~(LA_SIZE | LA_BLOCKS);
2526 if (bits & LA_ATIME)
2527 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
2528 if (bits & LA_CTIME)
2529 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
2530 if (bits & LA_MTIME)
2531 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
2532 if (bits & LA_SIZE) {
2533 spin_lock(&inode->i_lock);
2534 LDISKFS_I(inode)->i_disksize = attr->la_size;
2535 i_size_write(inode, attr->la_size);
2536 spin_unlock(&inode->i_lock);
2539 /* OSD should not change "i_blocks" which is used by quota.
2540 * "i_blocks" should be changed by ldiskfs only. */
2542 inode->i_mode = (inode->i_mode & S_IFMT) |
2543 (attr->la_mode & ~S_IFMT);
2545 i_uid_write(inode, attr->la_uid);
2547 i_gid_write(inode, attr->la_gid);
2548 if (bits & LA_NLINK)
2549 set_nlink(inode, attr->la_nlink);
2551 inode->i_rdev = attr->la_rdev;
2553 if (bits & LA_FLAGS) {
2554 /* always keep S_NOCMTIME */
2555 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2561 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2563 if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2564 (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2568 ll_vfs_dq_init(inode);
2570 if (attr->la_valid & LA_UID)
2571 iattr.ia_valid |= ATTR_UID;
2572 if (attr->la_valid & LA_GID)
2573 iattr.ia_valid |= ATTR_GID;
2574 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2575 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2577 rc = ll_vfs_dq_transfer(inode, &iattr);
2579 CERROR("%s: quota transfer failed: rc = %d. Is quota "
2580 "enforcement enabled on the ldiskfs "
2581 "filesystem?\n", inode->i_sb->s_id, rc);
2588 static int osd_attr_set(const struct lu_env *env,
2589 struct dt_object *dt,
2590 const struct lu_attr *attr,
2591 struct thandle *handle)
2593 struct osd_object *obj = osd_dt_obj(dt);
2594 struct inode *inode;
2597 if (!dt_object_exists(dt))
2600 LASSERT(handle != NULL);
2601 LASSERT(!dt_object_remote(dt));
2602 LASSERT(osd_invariant(obj));
2604 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2606 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
2607 struct osd_thread_info *oti = osd_oti_get(env);
2608 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
2609 struct lu_fid *fid1 = &oti->oti_fid;
2610 struct osd_inode_id *id = &oti->oti_id;
2611 struct iam_path_descr *ipd;
2612 struct iam_container *bag;
2613 struct osd_thandle *oh;
2616 fid_cpu_to_be(fid1, fid0);
2617 memset(id, 1, sizeof(*id));
2618 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2619 fid0)->oi_dir.od_container;
2620 ipd = osd_idx_ipd_get(env, bag);
2621 if (unlikely(ipd == NULL))
2624 oh = container_of0(handle, struct osd_thandle, ot_super);
2625 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2626 (const struct iam_rec *)id, ipd);
2627 osd_ipd_put(env, bag, ipd);
2628 return(rc > 0 ? 0 : rc);
2631 inode = obj->oo_inode;
2633 rc = osd_quota_transfer(inode, attr);
2637 spin_lock(&obj->oo_guard);
2638 rc = osd_inode_setattr(env, inode, attr);
2639 spin_unlock(&obj->oo_guard);
2643 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2645 if (!(attr->la_valid & LA_FLAGS))
2648 /* Let's check if there are extra flags need to be set into LMA */
2649 if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
2650 struct osd_thread_info *info = osd_oti_get(env);
2651 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2653 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
2657 lma->lma_incompat |=
2658 lustre_to_lma_flags(attr->la_flags);
2659 lustre_lma_swab(lma);
2660 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
2661 lma, sizeof(*lma), XATTR_REPLACE);
2663 struct osd_device *osd = osd_obj2dev(obj);
2665 CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
2666 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
2667 lma->lma_incompat, rc);
2670 attr->la_flags & LUSTRE_LMA_FL_MASKS;
2672 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2675 osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2680 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2681 struct osd_object *obj,
2682 const char *name, const int namelen)
2684 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2687 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2688 umode_t mode, struct dt_allocation_hint *hint,
2692 struct osd_device *osd = osd_obj2dev(obj);
2693 struct osd_thandle *oth;
2694 struct dt_object *parent = NULL;
2695 struct inode *inode;
2697 LINVRNT(osd_invariant(obj));
2698 LASSERT(obj->oo_inode == NULL);
2699 LASSERT(obj->oo_hl_head == NULL);
2701 if (S_ISDIR(mode) && ldiskfs_pdo) {
2702 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2703 if (obj->oo_hl_head == NULL)
2707 oth = container_of(th, struct osd_thandle, ot_super);
2708 LASSERT(oth->ot_handle->h_transaction != NULL);
2710 if (hint != NULL && hint->dah_parent != NULL &&
2711 !dt_object_remote(hint->dah_parent))
2712 parent = hint->dah_parent;
2714 inode = ldiskfs_create_inode(oth->ot_handle,
2715 parent ? osd_dt_obj(parent)->oo_inode :
2716 osd_sb(osd)->s_root->d_inode,
2718 if (!IS_ERR(inode)) {
2719 /* Do not update file c/mtime in ldiskfs. */
2720 inode->i_flags |= S_NOCMTIME;
2722 /* For new created object, it must be consistent,
2723 * and it is unnecessary to scrub against it. */
2724 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2726 obj->oo_inode = inode;
2729 if (obj->oo_hl_head != NULL) {
2730 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2731 obj->oo_hl_head = NULL;
2733 result = PTR_ERR(inode);
2735 LINVRNT(osd_invariant(obj));
2743 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2744 struct lu_attr *attr,
2745 struct dt_allocation_hint *hint,
2746 struct dt_object_format *dof,
2750 struct osd_thandle *oth;
2751 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
2753 LASSERT(S_ISDIR(attr->la_mode));
2755 oth = container_of(th, struct osd_thandle, ot_super);
2756 LASSERT(oth->ot_handle->h_transaction != NULL);
2757 result = osd_mkfile(info, obj, mode, hint, th);
2762 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2763 struct lu_attr *attr,
2764 struct dt_allocation_hint *hint,
2765 struct dt_object_format *dof,
2769 struct osd_thandle *oth;
2770 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2772 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2774 LASSERT(S_ISREG(attr->la_mode));
2776 oth = container_of(th, struct osd_thandle, ot_super);
2777 LASSERT(oth->ot_handle->h_transaction != NULL);
2779 result = osd_mkfile(info, obj, mode, hint, th);
2781 LASSERT(obj->oo_inode != NULL);
2782 if (feat->dif_flags & DT_IND_VARKEY)
2783 result = iam_lvar_create(obj->oo_inode,
2784 feat->dif_keysize_max,
2786 feat->dif_recsize_max,
2789 result = iam_lfix_create(obj->oo_inode,
2790 feat->dif_keysize_max,
2792 feat->dif_recsize_max,
2799 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2800 struct lu_attr *attr,
2801 struct dt_allocation_hint *hint,
2802 struct dt_object_format *dof,
2805 LASSERT(S_ISREG(attr->la_mode));
2806 return osd_mkfile(info, obj, (attr->la_mode &
2807 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2810 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2811 struct lu_attr *attr,
2812 struct dt_allocation_hint *hint,
2813 struct dt_object_format *dof,
2816 LASSERT(S_ISLNK(attr->la_mode));
2817 return osd_mkfile(info, obj, (attr->la_mode &
2818 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2821 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2822 struct lu_attr *attr,
2823 struct dt_allocation_hint *hint,
2824 struct dt_object_format *dof,
2827 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2830 LINVRNT(osd_invariant(obj));
2831 LASSERT(obj->oo_inode == NULL);
2832 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2833 S_ISFIFO(mode) || S_ISSOCK(mode));
2835 result = osd_mkfile(info, obj, mode, hint, th);
2837 LASSERT(obj->oo_inode != NULL);
2839 * This inode should be marked dirty for i_rdev. Currently
2840 * that is done in the osd_attr_init().
2842 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2845 LINVRNT(osd_invariant(obj));
2849 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2851 struct dt_allocation_hint *hint,
2852 struct dt_object_format *dof,
2855 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2857 osd_obj_type_f result;
2873 result = osd_mk_index;
2884 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2885 struct dt_object *parent, struct dt_object *child,
2890 ah->dah_parent = parent;
2891 ah->dah_mode = child_mode;
2893 if (parent != NULL && !dt_object_remote(parent)) {
2894 /* will help to find FID->ino at dt_insert("..") */
2895 struct osd_object *pobj = osd_dt_obj(parent);
2896 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
2900 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2901 struct lu_attr *attr, struct dt_object_format *dof)
2903 struct inode *inode = obj->oo_inode;
2904 __u64 valid = attr->la_valid;
2907 attr->la_valid &= ~(LA_TYPE | LA_MODE);
2909 if (dof->dof_type != DFT_NODE)
2910 attr->la_valid &= ~LA_RDEV;
2911 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2912 attr->la_valid &= ~LA_ATIME;
2913 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2914 attr->la_valid &= ~LA_CTIME;
2915 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2916 attr->la_valid &= ~LA_MTIME;
2918 result = osd_quota_transfer(inode, attr);
2922 if (attr->la_valid != 0) {
2923 result = osd_inode_setattr(info->oti_env, inode, attr);
2925 * The osd_inode_setattr() should always succeed here. The
2926 * only error that could be returned is EDQUOT when we are
2927 * trying to change the UID or GID of the inode. However, this
2928 * should not happen since quota enforcement is no longer
2929 * enabled on ldiskfs (lquota takes care of it).
2931 LASSERTF(result == 0, "%d\n", result);
2932 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2935 attr->la_valid = valid;
2939 * Helper function for osd_object_create()
2941 * \retval 0, on success
2943 static int __osd_object_create(struct osd_thread_info *info,
2944 struct osd_object *obj, struct lu_attr *attr,
2945 struct dt_allocation_hint *hint,
2946 struct dt_object_format *dof,
2952 osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
2954 /* we drop umask so that permissions we pass are not affected */
2955 umask = current->fs->umask;
2956 current->fs->umask = 0;
2958 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2960 if (likely(obj->oo_inode != NULL)) {
2961 LASSERT(obj->oo_inode->i_state & I_NEW);
2963 /* Unlock the inode before attr initialization to avoid
2964 * unnecessary dqget operations. LU-6378 */
2965 unlock_new_inode(obj->oo_inode);
2968 if (likely(result == 0)) {
2969 osd_attr_init(info, obj, attr, dof);
2970 osd_object_init0(obj);
2973 /* restore previous umask value */
2974 current->fs->umask = umask;
2976 osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
2982 * Helper function for osd_object_create()
2984 * \retval 0, on success
2986 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2987 const struct lu_fid *fid, struct thandle *th)
2989 struct osd_thread_info *info = osd_oti_get(env);
2990 struct osd_inode_id *id = &info->oti_id;
2991 struct osd_device *osd = osd_obj2dev(obj);
2992 struct osd_thandle *oh;
2995 LASSERT(obj->oo_inode != NULL);
2997 oh = container_of0(th, struct osd_thandle, ot_super);
2998 LASSERT(oh->ot_handle);
2999 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3001 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3002 rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3003 OI_CHECK_FLD, NULL);
3004 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3009 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3010 u64 seq, struct lu_seq_range *range)
3012 struct seq_server_site *ss = osd_seq_site(osd);
3014 if (fid_seq_is_idif(seq)) {
3015 fld_range_set_ost(range);
3016 range->lsr_index = idif_ost_idx(seq);
3020 if (!fid_seq_in_fldb(seq)) {
3021 fld_range_set_mdt(range);
3023 /* FIXME: If ss is NULL, it suppose not get lsr_index
3025 range->lsr_index = ss->ss_node_id;
3029 LASSERT(ss != NULL);
3030 fld_range_set_any(range);
3031 /* OSD will only do local fld lookup */
3032 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3036 * Concurrency: no external locking is necessary.
3038 static int osd_declare_object_create(const struct lu_env *env,
3039 struct dt_object *dt,
3040 struct lu_attr *attr,
3041 struct dt_allocation_hint *hint,
3042 struct dt_object_format *dof,
3043 struct thandle *handle)
3045 struct osd_thandle *oh;
3049 LASSERT(handle != NULL);
3051 oh = container_of0(handle, struct osd_thandle, ot_super);
3052 LASSERT(oh->ot_handle == NULL);
3054 /* EA object consumes more credits than regular object: osd_mk_index
3055 * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3056 * leaf_node, could involves the block, block bitmap, groups, GDT
3057 * change for each block, so add 4 * 2 credits in that case. */
3058 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3059 osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3060 (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3061 /* Reuse idle OI block may cause additional one OI block
3063 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3064 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3069 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
3070 osd_dt_obj(dt), false, NULL, false);
3074 /* will help to find FID->ino mapping at dt_insert() */
3075 rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3081 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
3082 struct lu_attr *attr,
3083 struct dt_allocation_hint *hint,
3084 struct dt_object_format *dof, struct thandle *th)
3086 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3087 struct osd_object *obj = osd_dt_obj(dt);
3088 struct osd_thread_info *info = osd_oti_get(env);
3092 if (dt_object_exists(dt))
3095 LINVRNT(osd_invariant(obj));
3096 LASSERT(!dt_object_remote(dt));
3097 LASSERT(osd_write_locked(env, obj));
3098 LASSERT(th != NULL);
3100 if (unlikely(fid_is_acct(fid)))
3101 /* Quota files can't be created from the kernel any more,
3102 * 'tune2fs -O quota' will take care of creating them */
3105 result = __osd_object_create(info, obj, attr, hint, dof, th);
3107 result = __osd_oi_insert(env, obj, fid, th);
3108 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3109 obj->oo_dt.do_body_ops = &osd_body_ops;
3111 LASSERT(ergo(result == 0,
3112 dt_object_exists(dt) && !dt_object_remote(dt)));
3114 LASSERT(osd_invariant(obj));
3119 * Called to destroy on-disk representation of the object
3121 * Concurrency: must be locked
3123 static int osd_declare_object_destroy(const struct lu_env *env,
3124 struct dt_object *dt,
3127 struct osd_object *obj = osd_dt_obj(dt);
3128 struct inode *inode = obj->oo_inode;
3129 struct osd_thandle *oh;
3136 oh = container_of0(th, struct osd_thandle, ot_super);
3137 LASSERT(oh->ot_handle == NULL);
3139 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3140 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3141 /* Recycle idle OI leaf may cause additional three OI blocks
3143 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3144 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3145 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3146 /* one less inode */
3147 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3148 -1, oh, obj, false, NULL, false);
3151 /* data to be truncated */
3152 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3153 0, oh, obj, true, NULL, false);
3157 /* will help to find FID->ino when this object is being
3158 * added to PENDING/ */
3159 rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3164 static int osd_object_destroy(const struct lu_env *env,
3165 struct dt_object *dt,
3168 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3169 struct osd_object *obj = osd_dt_obj(dt);
3170 struct inode *inode = obj->oo_inode;
3171 struct osd_device *osd = osd_obj2dev(obj);
3172 struct osd_thandle *oh;
3176 oh = container_of0(th, struct osd_thandle, ot_super);
3177 LASSERT(oh->ot_handle);
3179 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3181 if (unlikely(fid_is_acct(fid)))
3184 if (S_ISDIR(inode->i_mode)) {
3185 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3186 inode->i_nlink == 2);
3187 /* it will check/delete the inode from remote parent,
3188 * how to optimize it? unlink performance impaction XXX */
3189 result = osd_delete_from_remote_parent(env, osd, obj, oh);
3191 CERROR("%s: delete inode "DFID": rc = %d\n",
3192 osd_name(osd), PFID(fid), result);
3194 spin_lock(&obj->oo_guard);
3196 spin_unlock(&obj->oo_guard);
3197 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3200 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3202 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3204 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3205 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3206 oh->ot_handle, OI_CHECK_FLD);
3208 osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3209 /* XXX: add to ext3 orphan list */
3210 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3212 /* not needed in the cache anymore */
3213 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3214 obj->oo_destroyed = 1;
3220 * Put the fid into lustre_mdt_attrs, and then place the structure
3221 * inode's ea. This fid should not be altered during the life time
3224 * \retval +ve, on success
3225 * \retval -ve, on error
3227 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3229 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3230 const struct lu_fid *fid, __u32 compat, __u32 incompat)
3232 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3236 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3239 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3242 lustre_lma_init(lma, fid, compat, incompat);
3243 lustre_lma_swab(lma);
3245 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
3247 /* LMA may already exist, but we need to check that all the
3248 * desired compat/incompat flags have been added. */
3249 if (unlikely(rc == -EEXIST)) {
3250 if (compat == 0 && incompat == 0)
3253 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3254 XATTR_NAME_LMA, info->oti_mdt_attrs_old,
3259 lustre_lma_swab(lma);
3260 if (!(~lma->lma_compat & compat) &&
3261 !(~lma->lma_incompat & incompat))
3264 lma->lma_compat |= compat;
3265 lma->lma_incompat |= incompat;
3266 lustre_lma_swab(lma);
3267 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3268 sizeof(*lma), XATTR_REPLACE);
3275 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3276 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3277 * To have compatilibility with 1.8 ldiskfs driver we need to have
3278 * magic number at start of fid data.
3279 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3282 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3283 const struct lu_fid *fid)
3285 if (!fid_is_namespace_visible(fid) ||
3286 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3287 param->edp_magic = 0;
3291 param->edp_magic = LDISKFS_LUFID_MAGIC;
3292 param->edp_len = sizeof(struct lu_fid) + 1;
3293 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3297 * Try to read the fid from inode ea into dt_rec.
3299 * \param fid object fid.
3301 * \retval 0 on success
3303 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3304 __u32 ino, struct lu_fid *fid,
3305 struct osd_inode_id *id)
3307 struct osd_thread_info *info = osd_oti_get(env);
3308 struct inode *inode;
3311 osd_id_gen(id, ino, OSD_OII_NOGEN);
3312 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3314 RETURN(PTR_ERR(inode));
3320 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3322 struct inode *parent_dir,
3323 const struct lu_fid *dot_fid,
3324 const struct lu_fid *dot_dot_fid,
3325 struct osd_thandle *oth)
3327 struct ldiskfs_dentry_param *dot_ldp;
3328 struct ldiskfs_dentry_param *dot_dot_ldp;
3329 __u32 saved_nlink = dir->i_nlink;
3332 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3333 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3335 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3336 dot_ldp->edp_magic = 0;
3338 rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3339 dir, dot_ldp, dot_dot_ldp);
3340 /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3341 * the subseqent ref_add() will increase the dir->i_nlink
3342 * as 3. That is incorrect for new created directory.
3344 * It looks like hack, because we want to make the OSD API
3345 * to be order-independent for new created directory object
3346 * between dt_insert(..) and ref_add() operations.
3348 * Here, we only restore the in-RAM dir-inode's nlink attr,
3349 * becuase if the nlink attr is not 2, then there will be
3350 * ref_add() called following the dt_insert(..), such call
3351 * will make both the in-RAM and on-disk dir-inode's nlink
3352 * attr to be set as 2. LU-7447 */
3353 set_nlink(dir, saved_nlink);
3358 * Create an local agent inode for remote entry
3360 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3361 struct osd_device *osd,
3362 struct osd_object *pobj,
3363 const struct lu_fid *fid,
3367 struct osd_thread_info *info = osd_oti_get(env);
3368 struct inode *local;
3369 struct osd_thandle *oh;
3374 oh = container_of(th, struct osd_thandle, ot_super);
3375 LASSERT(oh->ot_handle->h_transaction != NULL);
3377 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3378 if (IS_ERR(local)) {
3379 CERROR("%s: create local error %d\n", osd_name(osd),
3380 (int)PTR_ERR(local));
3384 /* restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3385 * correct gid on remote file, not agent here */
3386 local->i_gid = current_fsgid();
3387 ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3388 unlock_new_inode(local);
3390 /* Set special LMA flag for local agent inode */
3391 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3393 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3394 osd_name(osd), PFID(fid), rc);
3395 RETURN(ERR_PTR(rc));
3401 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3402 lu_object_fid(&pobj->oo_dt.do_lu),
3405 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3406 osd_name(osd), PFID(fid), rc);
3407 RETURN(ERR_PTR(rc));
3414 * when direntry is deleted, we have to take care of possible agent inode
3415 * referenced by that. unfortunately we can't do this at that point:
3416 * iget() within a running transaction leads to deadlock and we better do
3417 * not call that every delete declaration to save performance. so we put
3418 * a potention agent inode on a list and process that once the transaction
3419 * is over. Notice it's not any worse in terms of real orphans as regular
3420 * object destroy doesn't put inodes on the on-disk orphan list. this should
3421 * be addressed separately
3423 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3424 struct osd_thandle *oh,
3427 struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3428 struct osd_obj_orphan *oor;
3435 oor->oor_env = (struct lu_env *)env;
3436 spin_lock(&osd->od_osfs_lock);
3437 list_add(&oor->oor_list, &osd->od_orphan_list);
3438 spin_unlock(&osd->od_osfs_lock);
3440 oh->ot_remove_agents = 1;
3446 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3447 struct osd_device *osd)
3449 struct osd_thread_info *info = osd_oti_get(env);
3450 struct osd_obj_orphan *oor, *tmp;
3451 struct osd_inode_id id;
3452 struct list_head list;
3453 struct inode *inode;
3458 INIT_LIST_HEAD(&list);
3460 spin_lock(&osd->od_osfs_lock);
3461 list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3462 if (oor->oor_env == env) {
3463 list_del(&oor->oor_list);
3464 list_add(&oor->oor_list, &list);
3467 spin_unlock(&osd->od_osfs_lock);
3469 list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3473 list_del(&oor->oor_list);
3476 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3477 inode = osd_iget_fid(info, osd, &id, &fid);
3481 if (!osd_remote_fid(env, osd, &fid)) {
3486 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3488 mark_inode_dirty(inode);
3489 ldiskfs_journal_stop(jh);
3497 * OSD layer object create function for interoperability mode (b11826).
3498 * This is mostly similar to osd_object_create(). Only difference being, fid is
3499 * inserted into inode ea here.
3501 * \retval 0, on success
3502 * \retval -ve, on error
3504 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
3505 struct lu_attr *attr,
3506 struct dt_allocation_hint *hint,
3507 struct dt_object_format *dof,
3510 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3511 struct osd_object *obj = osd_dt_obj(dt);
3512 struct osd_thread_info *info = osd_oti_get(env);
3513 int result, on_ost = 0;
3517 if (dt_object_exists(dt))
3520 LASSERT(osd_invariant(obj));
3521 LASSERT(!dt_object_remote(dt));
3522 LASSERT(osd_write_locked(env, obj));
3523 LASSERT(th != NULL);
3525 if (unlikely(fid_is_acct(fid)))
3526 /* Quota files can't be created from the kernel any more,
3527 * 'tune2fs -O quota' will take care of creating them */
3530 result = __osd_object_create(info, obj, attr, hint, dof, th);
3532 if (fid_is_idif(fid) &&
3533 !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3534 struct lu_fid *tfid = &info->oti_fid;
3535 struct ost_id *oi = &info->oti_ostid;
3537 fid_to_ostid(fid, oi);
3538 ostid_to_fid(tfid, oi, 0);
3540 result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3541 LMAC_FID_ON_OST, 0);
3543 on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3545 result = osd_ea_fid_set(info, obj->oo_inode, fid,
3546 on_ost ? LMAC_FID_ON_OST : 0,
3549 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3550 obj->oo_dt.do_body_ops = &osd_body_ops;
3554 result = __osd_oi_insert(env, obj, fid, th);
3556 /* a small optimization - dt_insert() isn't usually applied
3557 * to OST objects, so we don't need to cache OI mapping for
3559 if (result == 0 && on_ost == 0) {
3560 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3561 result = osd_idc_find_and_init(env, osd, obj);
3562 LASSERT(result == 0);
3565 LASSERT(ergo(result == 0,
3566 dt_object_exists(dt) && !dt_object_remote(dt)));
3567 LINVRNT(osd_invariant(obj));
3571 static int osd_declare_object_ref_add(const struct lu_env *env,
3572 struct dt_object *dt,
3573 struct thandle *handle)
3575 struct osd_thandle *oh;
3577 /* it's possible that object doesn't exist yet */
3578 LASSERT(handle != NULL);
3580 oh = container_of0(handle, struct osd_thandle, ot_super);
3581 LASSERT(oh->ot_handle == NULL);
3583 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3584 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3590 * Concurrency: @dt is write locked.
3592 static int osd_object_ref_add(const struct lu_env *env,
3593 struct dt_object *dt, struct thandle *th)
3595 struct osd_object *obj = osd_dt_obj(dt);
3596 struct inode *inode = obj->oo_inode;
3597 struct osd_thandle *oh;
3600 if (!dt_object_exists(dt) || obj->oo_destroyed)
3603 LINVRNT(osd_invariant(obj));
3604 LASSERT(!dt_object_remote(dt));
3605 LASSERT(osd_write_locked(env, obj));
3606 LASSERT(th != NULL);
3608 oh = container_of0(th, struct osd_thandle, ot_super);
3609 LASSERT(oh->ot_handle != NULL);
3611 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
3613 CDEBUG(D_INODE, DFID" increase nlink %d\n",
3614 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3616 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
3617 * (65000) subdirectories by storing "1" in i_nlink if the link count
3618 * would otherwise overflow. Directory tranversal tools understand
3619 * that (st_nlink == 1) indicates that the filesystem dose not track
3620 * hard links count on the directory, and will not abort subdirectory
3621 * scanning early once (st_nlink - 2) subdirs have been found.
3623 * This also has to properly handle the case of inodes with nlink == 0
3624 * in case they are being linked into the PENDING directory
3626 spin_lock(&obj->oo_guard);
3627 if (unlikely(inode->i_nlink == 0))
3628 /* inc_nlink from 0 may cause WARN_ON */
3629 set_nlink(inode, 1);
3631 ldiskfs_inc_count(oh->ot_handle, inode);
3632 if (!S_ISDIR(inode->i_mode))
3633 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
3635 spin_unlock(&obj->oo_guard);
3637 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3638 LINVRNT(osd_invariant(obj));
3640 osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
3645 static int osd_declare_object_ref_del(const struct lu_env *env,
3646 struct dt_object *dt,
3647 struct thandle *handle)
3649 struct osd_thandle *oh;
3651 if (!dt_object_exists(dt))
3654 LASSERT(!dt_object_remote(dt));
3655 LASSERT(handle != NULL);
3657 oh = container_of0(handle, struct osd_thandle, ot_super);
3658 LASSERT(oh->ot_handle == NULL);
3660 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
3661 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3667 * Concurrency: @dt is write locked.
3669 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
3672 struct osd_object *obj = osd_dt_obj(dt);
3673 struct inode *inode = obj->oo_inode;
3674 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3675 struct osd_thandle *oh;
3677 if (!dt_object_exists(dt))
3680 LINVRNT(osd_invariant(obj));
3681 LASSERT(!dt_object_remote(dt));
3682 LASSERT(osd_write_locked(env, obj));
3683 LASSERT(th != NULL);
3685 oh = container_of0(th, struct osd_thandle, ot_super);
3686 LASSERT(oh->ot_handle != NULL);
3688 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
3690 spin_lock(&obj->oo_guard);
3691 /* That can be result of upgrade from old Lustre version and
3692 * applied only to local files. Just skip this ref_del call.
3693 * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
3694 if (inode->i_nlink == 0) {
3695 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
3696 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
3697 ", maybe an upgraded file? (LU-3915)\n",
3698 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
3699 spin_unlock(&obj->oo_guard);
3703 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
3704 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3706 ldiskfs_dec_count(oh->ot_handle, inode);
3707 spin_unlock(&obj->oo_guard);
3709 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3710 LINVRNT(osd_invariant(obj));
3712 osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
3718 * Get the 64-bit version for an inode.
3720 static int osd_object_version_get(const struct lu_env *env,
3721 struct dt_object *dt, dt_obj_version_t *ver)
3723 struct inode *inode = osd_dt_obj(dt)->oo_inode;
3725 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
3726 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3727 *ver = LDISKFS_I(inode)->i_fs_version;
3732 * Concurrency: @dt is read locked.
3734 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
3735 struct lu_buf *buf, const char *name)
3737 struct osd_object *obj = osd_dt_obj(dt);
3738 struct inode *inode = obj->oo_inode;
3739 struct osd_thread_info *info = osd_oti_get(env);
3740 struct dentry *dentry = &info->oti_obj_dentry;
3741 bool cache_xattr = false;
3744 /* version get is not real XATTR but uses xattr API */
3745 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3746 /* for version we are just using xattr API but change inode
3748 if (buf->lb_len == 0)
3749 return sizeof(dt_obj_version_t);
3751 if (buf->lb_len < sizeof(dt_obj_version_t))
3754 osd_object_version_get(env, dt, buf->lb_buf);
3756 return sizeof(dt_obj_version_t);
3759 if (!dt_object_exists(dt))
3762 LASSERT(!dt_object_remote(dt));
3763 LASSERT(inode->i_op != NULL);
3764 LASSERT(inode->i_op->getxattr != NULL);
3766 if (strcmp(name, XATTR_NAME_LOV) == 0 ||
3767 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
3771 rc = osd_oxc_get(obj, name, buf);
3776 rc = __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
3778 if (rc == -ENOENT || rc == -ENODATA)
3779 osd_oxc_add(obj, name, NULL, 0);
3780 else if (rc > 0 && buf->lb_buf != NULL)
3781 osd_oxc_add(obj, name, buf->lb_buf, rc);
3787 static int osd_declare_xattr_set(const struct lu_env *env,
3788 struct dt_object *dt,
3789 const struct lu_buf *buf, const char *name,
3790 int fl, struct thandle *handle)
3792 struct osd_thandle *oh;
3794 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3796 LASSERT(handle != NULL);
3798 oh = container_of0(handle, struct osd_thandle, ot_super);
3799 LASSERT(oh->ot_handle == NULL);
3801 if (strcmp(name, XATTR_NAME_LMA) == 0) {
3802 /* For non-upgrading case, the LMA is set first and
3803 * usually fit inode. But for upgrade case, the LMA
3804 * may be in another separated EA block. */
3805 if (!dt_object_exists(dt))
3807 else if (fl == LU_XATTR_REPLACE)
3811 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3815 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
3820 if (buf->lb_buf == NULL && dt_object_exists(dt)) {
3821 /* learn xattr size from osd_xattr_get if
3822 attribute has not been read yet */
3823 buflen = __osd_xattr_get(
3824 osd_dt_obj(dt)->oo_inode,
3825 &osd_oti_get(env)->oti_obj_dentry,
3830 buflen = buf->lb_len;
3833 if (buflen > sb->s_blocksize) {
3834 credits += osd_calc_bkmap_credits(
3836 (buflen + sb->s_blocksize - 1) >>
3837 sb->s_blocksize_bits);
3841 * xattr set may involve inode quota change, reserve credits for
3842 * dquot_initialize()
3844 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3847 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3853 * Set the 64-bit version for object
3855 static void osd_object_version_set(const struct lu_env *env,
3856 struct dt_object *dt,
3857 dt_obj_version_t *new_version)
3859 struct inode *inode = osd_dt_obj(dt)->oo_inode;
3861 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
3862 *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3864 LDISKFS_I(inode)->i_fs_version = *new_version;
3865 /** Version is set after all inode operations are finished,
3866 * so we should mark it dirty here */
3867 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3871 * Concurrency: @dt is write locked.
3873 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3874 const struct lu_buf *buf, const char *name, int fl,
3875 struct thandle *handle)
3877 struct osd_object *obj = osd_dt_obj(dt);
3878 struct inode *inode = obj->oo_inode;
3879 struct osd_thread_info *info = osd_oti_get(env);
3884 LASSERT(handle != NULL);
3886 /* version set is not real XATTR */
3887 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3888 /* for version we are just using xattr API but change inode
3890 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3891 osd_object_version_set(env, dt, buf->lb_buf);
3892 return sizeof(dt_obj_version_t);
3895 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
3896 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3898 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3899 if (fl & LU_XATTR_REPLACE)
3900 fs_flags |= XATTR_REPLACE;
3902 if (fl & LU_XATTR_CREATE)
3903 fs_flags |= XATTR_CREATE;
3905 if (strcmp(name, XATTR_NAME_LMV) == 0) {
3906 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3908 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3912 lma->lma_incompat |= LMAI_STRIPED;
3913 lustre_lma_swab(lma);
3914 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3915 sizeof(*lma), XATTR_REPLACE);
3920 rc = __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3922 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3925 (strcmp(name, XATTR_NAME_LOV) == 0 ||
3926 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
3927 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
3933 * Concurrency: @dt is read locked.
3935 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3936 const struct lu_buf *buf)
3938 struct osd_object *obj = osd_dt_obj(dt);
3939 struct inode *inode = obj->oo_inode;
3940 struct osd_thread_info *info = osd_oti_get(env);
3941 struct dentry *dentry = &info->oti_obj_dentry;
3943 if (!dt_object_exists(dt))
3946 LASSERT(!dt_object_remote(dt));
3947 LASSERT(inode->i_op != NULL);
3948 LASSERT(inode->i_op->listxattr != NULL);
3950 dentry->d_inode = inode;
3951 dentry->d_sb = inode->i_sb;
3952 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3955 static int osd_declare_xattr_del(const struct lu_env *env,
3956 struct dt_object *dt, const char *name,
3957 struct thandle *handle)
3959 struct osd_thandle *oh;
3960 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3962 LASSERT(!dt_object_remote(dt));
3963 LASSERT(handle != NULL);
3965 oh = container_of0(handle, struct osd_thandle, ot_super);
3966 LASSERT(oh->ot_handle == NULL);
3968 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3969 osd_dto_credits_noquota[DTO_XATTR_SET]);
3971 * xattr del may involve inode quota change, reserve credits for
3972 * dquot_initialize()
3974 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3980 * Concurrency: @dt is write locked.
3982 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3983 const char *name, struct thandle *handle)
3985 struct osd_object *obj = osd_dt_obj(dt);
3986 struct inode *inode = obj->oo_inode;
3987 struct osd_thread_info *info = osd_oti_get(env);
3988 struct dentry *dentry = &info->oti_obj_dentry;
3991 if (!dt_object_exists(dt))
3994 LASSERT(!dt_object_remote(dt));
3995 LASSERT(inode->i_op != NULL);
3996 LASSERT(inode->i_op->removexattr != NULL);
3997 LASSERT(handle != NULL);
3999 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4001 ll_vfs_dq_init(inode);
4002 dentry->d_inode = inode;
4003 dentry->d_sb = inode->i_sb;
4004 rc = inode->i_op->removexattr(dentry, name);
4005 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4008 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4009 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4010 osd_oxc_del(obj, name);
4015 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4016 __u64 start, __u64 end)
4018 struct osd_object *obj = osd_dt_obj(dt);
4019 struct inode *inode = obj->oo_inode;
4020 struct osd_thread_info *info = osd_oti_get(env);
4021 struct dentry *dentry = &info->oti_obj_dentry;
4022 struct file *file = &info->oti_file;
4027 dentry->d_inode = inode;
4028 dentry->d_sb = inode->i_sb;
4029 file->f_path.dentry = dentry;
4030 file->f_mapping = inode->i_mapping;
4031 file->f_op = inode->i_fop;
4032 set_file_inode(file, inode);
4034 rc = ll_vfs_fsync_range(file, start, end, 0);
4039 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4048 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4049 const struct dt_index_features *feat)
4051 struct iam_descr *descr;
4053 if (osd_object_is_root(o))
4054 return feat == &dt_directory_features;
4056 LASSERT(o->oo_dir != NULL);
4058 descr = o->oo_dir->od_container.ic_descr;
4059 if (feat == &dt_directory_features) {
4060 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4066 feat->dif_keysize_min <= descr->id_key_size &&
4067 descr->id_key_size <= feat->dif_keysize_max &&
4068 feat->dif_recsize_min <= descr->id_rec_size &&
4069 descr->id_rec_size <= feat->dif_recsize_max &&
4070 !(feat->dif_flags & (DT_IND_VARKEY |
4071 DT_IND_VARREC | DT_IND_NONUNQ)) &&
4072 ergo(feat->dif_flags & DT_IND_UPDATE,
4073 1 /* XXX check that object (and file system) is
4078 static int osd_iam_container_init(const struct lu_env *env,
4079 struct osd_object *obj,
4080 struct osd_directory *dir)
4082 struct iam_container *bag = &dir->od_container;
4085 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4089 result = iam_container_setup(bag);
4091 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4093 iam_container_fini(bag);
4100 * Concurrency: no external locking is necessary.
4102 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4103 const struct dt_index_features *feat)
4107 struct osd_object *obj = osd_dt_obj(dt);
4109 LINVRNT(osd_invariant(obj));
4111 if (osd_object_is_root(obj)) {
4112 dt->do_index_ops = &osd_index_ea_ops;
4114 } else if (feat == &dt_directory_features) {
4115 dt->do_index_ops = &osd_index_ea_ops;
4116 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4121 } else if (unlikely(feat == &dt_otable_features)) {
4122 dt->do_index_ops = &osd_otable_ops;
4124 } else if (unlikely(feat == &dt_acct_features)) {
4125 dt->do_index_ops = &osd_acct_index_ops;
4128 } else if (!osd_has_index(obj)) {
4129 struct osd_directory *dir;
4134 spin_lock(&obj->oo_guard);
4135 if (obj->oo_dir == NULL)
4139 * Concurrent thread allocated container data.
4142 spin_unlock(&obj->oo_guard);
4144 * Now, that we have container data, serialize its
4147 down_write(&obj->oo_ext_idx_sem);
4149 * recheck under lock.
4151 if (!osd_has_index(obj))
4152 result = osd_iam_container_init(env, obj,
4156 up_write(&obj->oo_ext_idx_sem);
4164 if (result == 0 && skip_iam == 0) {
4165 if (!osd_iam_index_probe(env, obj, feat))
4168 LINVRNT(osd_invariant(obj));
4170 if (result == 0 && feat == &dt_quota_glb_features &&
4171 fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
4172 result = osd_quota_migration(env, dt);
4177 static int osd_otable_it_attr_get(const struct lu_env *env,
4178 struct dt_object *dt,
4179 struct lu_attr *attr)
4185 static const struct dt_object_operations osd_obj_ops = {
4186 .do_read_lock = osd_object_read_lock,
4187 .do_write_lock = osd_object_write_lock,
4188 .do_read_unlock = osd_object_read_unlock,
4189 .do_write_unlock = osd_object_write_unlock,
4190 .do_write_locked = osd_object_write_locked,
4191 .do_attr_get = osd_attr_get,
4192 .do_declare_attr_set = osd_declare_attr_set,
4193 .do_attr_set = osd_attr_set,
4194 .do_ah_init = osd_ah_init,
4195 .do_declare_create = osd_declare_object_create,
4196 .do_create = osd_object_create,
4197 .do_declare_destroy = osd_declare_object_destroy,
4198 .do_destroy = osd_object_destroy,
4199 .do_index_try = osd_index_try,
4200 .do_declare_ref_add = osd_declare_object_ref_add,
4201 .do_ref_add = osd_object_ref_add,
4202 .do_declare_ref_del = osd_declare_object_ref_del,
4203 .do_ref_del = osd_object_ref_del,
4204 .do_xattr_get = osd_xattr_get,
4205 .do_declare_xattr_set = osd_declare_xattr_set,
4206 .do_xattr_set = osd_xattr_set,
4207 .do_declare_xattr_del = osd_declare_xattr_del,
4208 .do_xattr_del = osd_xattr_del,
4209 .do_xattr_list = osd_xattr_list,
4210 .do_object_sync = osd_object_sync,
4211 .do_invalidate = osd_invalidate,
4215 * dt_object_operations for interoperability mode
4216 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
4218 static const struct dt_object_operations osd_obj_ea_ops = {
4219 .do_read_lock = osd_object_read_lock,
4220 .do_write_lock = osd_object_write_lock,
4221 .do_read_unlock = osd_object_read_unlock,
4222 .do_write_unlock = osd_object_write_unlock,
4223 .do_write_locked = osd_object_write_locked,
4224 .do_attr_get = osd_attr_get,
4225 .do_declare_attr_set = osd_declare_attr_set,
4226 .do_attr_set = osd_attr_set,
4227 .do_ah_init = osd_ah_init,
4228 .do_declare_create = osd_declare_object_create,
4229 .do_create = osd_object_ea_create,
4230 .do_declare_destroy = osd_declare_object_destroy,
4231 .do_destroy = osd_object_destroy,
4232 .do_index_try = osd_index_try,
4233 .do_declare_ref_add = osd_declare_object_ref_add,
4234 .do_ref_add = osd_object_ref_add,
4235 .do_declare_ref_del = osd_declare_object_ref_del,
4236 .do_ref_del = osd_object_ref_del,
4237 .do_xattr_get = osd_xattr_get,
4238 .do_declare_xattr_set = osd_declare_xattr_set,
4239 .do_xattr_set = osd_xattr_set,
4240 .do_declare_xattr_del = osd_declare_xattr_del,
4241 .do_xattr_del = osd_xattr_del,
4242 .do_xattr_list = osd_xattr_list,
4243 .do_object_sync = osd_object_sync,
4244 .do_invalidate = osd_invalidate,
4247 static const struct dt_object_operations osd_obj_otable_it_ops = {
4248 .do_attr_get = osd_otable_it_attr_get,
4249 .do_index_try = osd_index_try,
4252 static int osd_index_declare_iam_delete(const struct lu_env *env,
4253 struct dt_object *dt,
4254 const struct dt_key *key,
4255 struct thandle *handle)
4257 struct osd_thandle *oh;
4259 oh = container_of0(handle, struct osd_thandle, ot_super);
4260 LASSERT(oh->ot_handle == NULL);
4262 /* Recycle may cause additional three blocks to be changed. */
4263 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4264 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4270 * delete a (key, value) pair from index \a dt specified by \a key
4272 * \param dt osd index object
4273 * \param key key for index
4274 * \param rec record reference
4275 * \param handle transaction handler
4278 * \retval -ve failure
4280 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4281 const struct dt_key *key,
4282 struct thandle *handle)
4284 struct osd_thread_info *oti = osd_oti_get(env);
4285 struct osd_object *obj = osd_dt_obj(dt);
4286 struct osd_thandle *oh;
4287 struct iam_path_descr *ipd;
4288 struct iam_container *bag = &obj->oo_dir->od_container;
4292 if (!dt_object_exists(dt))
4295 LINVRNT(osd_invariant(obj));
4296 LASSERT(!dt_object_remote(dt));
4297 LASSERT(bag->ic_object == obj->oo_inode);
4298 LASSERT(handle != NULL);
4300 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4302 ipd = osd_idx_ipd_get(env, bag);
4303 if (unlikely(ipd == NULL))
4306 oh = container_of0(handle, struct osd_thandle, ot_super);
4307 LASSERT(oh->ot_handle != NULL);
4308 LASSERT(oh->ot_handle->h_transaction != NULL);
4310 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4311 /* swab quota uid/gid provided by caller */
4312 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4313 key = (const struct dt_key *)&oti->oti_quota_id;
4316 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4317 osd_ipd_put(env, bag, ipd);
4318 LINVRNT(osd_invariant(obj));
4319 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4323 static int osd_index_declare_ea_delete(const struct lu_env *env,
4324 struct dt_object *dt,
4325 const struct dt_key *key,
4326 struct thandle *handle)
4328 struct osd_thandle *oh;
4329 struct inode *inode;
4333 LASSERT(!dt_object_remote(dt));
4334 LASSERT(handle != NULL);
4336 oh = container_of0(handle, struct osd_thandle, ot_super);
4337 LASSERT(oh->ot_handle == NULL);
4339 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
4340 if (key != NULL && unlikely(strcmp((char *)key, dotdot) == 0)) {
4341 /* '..' to a remote object has a local representative */
4342 credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
4343 /* to reset LMAI_REMOTE_PARENT */
4346 osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
4348 inode = osd_dt_obj(dt)->oo_inode;
4352 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
4353 0, oh, osd_dt_obj(dt), true, NULL, false);
4357 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
4360 struct osd_fid_pack *rec;
4363 if (de->file_type & LDISKFS_DIRENT_LUFID) {
4364 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
4365 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
4366 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
4372 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
4373 const struct lu_fid *fid)
4375 struct seq_server_site *ss = osd_seq_site(osd);
4378 /* FID seqs not in FLDB, must be local seq */
4379 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
4382 /* If FLD is not being initialized yet, it only happens during the
4383 * initialization, likely during mgs initialization, and we assume
4384 * this is local FID. */
4385 if (ss == NULL || ss->ss_server_fld == NULL)
4388 /* Only check the local FLDB here */
4389 if (osd_seq_exists(env, osd, fid_seq(fid)))
4396 * Index delete function for interoperability mode (b11826).
4397 * It will remove the directory entry added by osd_index_ea_insert().
4398 * This entry is needed to maintain name->fid mapping.
4400 * \param key, key i.e. file entry to be deleted
4402 * \retval 0, on success
4403 * \retval -ve, on error
4405 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
4406 const struct dt_key *key, struct thandle *handle)
4408 struct osd_object *obj = osd_dt_obj(dt);
4409 struct inode *dir = obj->oo_inode;
4410 struct dentry *dentry;
4411 struct osd_thandle *oh;
4412 struct ldiskfs_dir_entry_2 *de = NULL;
4413 struct buffer_head *bh;
4414 struct htree_lock *hlock = NULL;
4415 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
4416 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4420 if (!dt_object_exists(dt))
4423 LINVRNT(osd_invariant(obj));
4424 LASSERT(!dt_object_remote(dt));
4425 LASSERT(handle != NULL);
4427 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4429 oh = container_of(handle, struct osd_thandle, ot_super);
4430 LASSERT(oh->ot_handle != NULL);
4431 LASSERT(oh->ot_handle->h_transaction != NULL);
4433 ll_vfs_dq_init(dir);
4434 dentry = osd_child_dentry_get(env, obj,
4435 (char *)key, strlen((char *)key));
4437 if (obj->oo_hl_head != NULL) {
4438 hlock = osd_oti_get(env)->oti_hlock;
4439 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4440 dir, LDISKFS_HLOCK_DEL);
4442 down_write(&obj->oo_ext_idx_sem);
4445 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4447 /* If this is not the ".." entry, it might be a remote DNE
4448 * entry and we need to check if the FID is for a remote
4449 * MDT. If the FID is not in the directory entry (e.g.
4450 * upgraded 1.8 filesystem without dirdata enabled) then
4451 * we need to get the FID from the LMA. For a remote directory
4452 * there HAS to be an LMA, it cannot be an IGIF inode in this
4455 * Delete the entry before the agent inode in order to
4456 * simplify error handling. At worst an error after deleting
4457 * the entry first might leak the agent inode afterward. The
4458 * reverse would need filesystem abort in case of error deleting
4459 * the entry after the agent had been removed, or leave a
4460 * dangling entry pointing at a random inode. */
4461 if (strcmp((char *)key, dotdot) != 0) {
4462 LASSERT(de != NULL);
4463 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
4464 if (rc == -ENODATA) {
4465 /* can't get FID, postpone to the end of the
4466 * transaction when iget() is safe */
4467 osd_schedule_agent_inode_removal(env, oh,
4468 le32_to_cpu(de->inode));
4469 } else if (rc == 0 &&
4470 unlikely(osd_remote_fid(env, osd, fid))) {
4471 osd_schedule_agent_inode_removal(env, oh,
4472 le32_to_cpu(de->inode));
4475 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
4481 ldiskfs_htree_unlock(hlock);
4483 up_write(&obj->oo_ext_idx_sem);
4488 /* For inode on the remote MDT, .. will point to
4489 * /Agent directory, Check whether it needs to delete
4490 * from agent directory */
4491 if (unlikely(strcmp((char *)key, dotdot) == 0)) {
4494 ret = osd_delete_from_remote_parent(env, osd_obj2dev(obj),
4497 /* Sigh, the entry has been deleted, and
4498 * it is not easy to revert it back, so
4499 * let's keep this error private, and let
4500 * LFSCK fix it. XXX */
4501 CERROR("%s: delete remote parent "DFID": rc = %d\n",
4502 osd_name(osd), PFID(fid), ret);
4505 LASSERT(osd_invariant(obj));
4506 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4511 * Lookup index for \a key and copy record to \a rec.
4513 * \param dt osd index object
4514 * \param key key for index
4515 * \param rec record reference
4517 * \retval +ve success : exact mach
4518 * \retval 0 return record with key not greater than \a key
4519 * \retval -ve failure
4521 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
4522 struct dt_rec *rec, const struct dt_key *key)
4524 struct osd_object *obj = osd_dt_obj(dt);
4525 struct iam_path_descr *ipd;
4526 struct iam_container *bag = &obj->oo_dir->od_container;
4527 struct osd_thread_info *oti = osd_oti_get(env);
4528 struct iam_iterator *it = &oti->oti_idx_it;
4529 struct iam_rec *iam_rec;
4533 if (!dt_object_exists(dt))
4536 LASSERT(osd_invariant(obj));
4537 LASSERT(!dt_object_remote(dt));
4538 LASSERT(bag->ic_object == obj->oo_inode);
4540 ipd = osd_idx_ipd_get(env, bag);
4544 /* got ipd now we can start iterator. */
4545 iam_it_init(it, bag, 0, ipd);
4547 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4548 /* swab quota uid/gid provided by caller */
4549 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4550 key = (const struct dt_key *)&oti->oti_quota_id;
4553 rc = iam_it_get(it, (struct iam_key *)key);
4555 if (S_ISDIR(obj->oo_inode->i_mode))
4556 iam_rec = (struct iam_rec *)oti->oti_ldp;
4558 iam_rec = (struct iam_rec *) rec;
4560 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
4562 if (S_ISDIR(obj->oo_inode->i_mode))
4563 osd_fid_unpack((struct lu_fid *) rec,
4564 (struct osd_fid_pack *)iam_rec);
4565 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
4566 osd_quota_unpack(obj, rec);
4571 osd_ipd_put(env, bag, ipd);
4573 LINVRNT(osd_invariant(obj));
4578 static int osd_index_declare_iam_insert(const struct lu_env *env,
4579 struct dt_object *dt,
4580 const struct dt_rec *rec,
4581 const struct dt_key *key,
4582 struct thandle *handle)
4584 struct osd_thandle *oh;
4586 LASSERT(handle != NULL);
4588 oh = container_of0(handle, struct osd_thandle, ot_super);
4589 LASSERT(oh->ot_handle == NULL);
4591 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4592 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4598 * Inserts (key, value) pair in \a dt index object.
4600 * \param dt osd index object
4601 * \param key key for index
4602 * \param rec record reference
4603 * \param th transaction handler
4606 * \retval -ve failure
4608 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
4609 const struct dt_rec *rec,
4610 const struct dt_key *key, struct thandle *th,
4613 struct osd_object *obj = osd_dt_obj(dt);
4614 struct iam_path_descr *ipd;
4615 struct osd_thandle *oh;
4616 struct iam_container *bag;
4617 struct osd_thread_info *oti = osd_oti_get(env);
4618 struct iam_rec *iam_rec;
4622 if (!dt_object_exists(dt))
4625 LINVRNT(osd_invariant(obj));
4626 LASSERT(!dt_object_remote(dt));
4628 bag = &obj->oo_dir->od_container;
4629 LASSERT(bag->ic_object == obj->oo_inode);
4630 LASSERT(th != NULL);
4632 osd_trans_exec_op(env, th, OSD_OT_INSERT);
4634 ipd = osd_idx_ipd_get(env, bag);
4635 if (unlikely(ipd == NULL))
4638 oh = container_of0(th, struct osd_thandle, ot_super);
4639 LASSERT(oh->ot_handle != NULL);
4640 LASSERT(oh->ot_handle->h_transaction != NULL);
4641 if (S_ISDIR(obj->oo_inode->i_mode)) {
4642 iam_rec = (struct iam_rec *)oti->oti_ldp;
4643 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
4644 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4645 /* pack quota uid/gid */
4646 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4647 key = (const struct dt_key *)&oti->oti_quota_id;
4648 /* pack quota record */
4649 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
4650 iam_rec = (struct iam_rec *)rec;
4652 iam_rec = (struct iam_rec *)rec;
4655 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
4657 osd_ipd_put(env, bag, ipd);
4658 LINVRNT(osd_invariant(obj));
4659 osd_trans_exec_check(env, th, OSD_OT_INSERT);
4664 * Calls ldiskfs_add_entry() to add directory entry
4665 * into the directory. This is required for
4666 * interoperability mode (b11826)
4668 * \retval 0, on success
4669 * \retval -ve, on error
4671 static int __osd_ea_add_rec(struct osd_thread_info *info,
4672 struct osd_object *pobj, struct inode *cinode,
4673 const char *name, const struct lu_fid *fid,
4674 struct htree_lock *hlock, struct thandle *th)
4676 struct ldiskfs_dentry_param *ldp;
4677 struct dentry *child;
4678 struct osd_thandle *oth;
4681 oth = container_of(th, struct osd_thandle, ot_super);
4682 LASSERT(oth->ot_handle != NULL);
4683 LASSERT(oth->ot_handle->h_transaction != NULL);
4684 LASSERT(pobj->oo_inode);
4686 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
4687 if (unlikely(pobj->oo_inode ==
4688 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
4691 osd_get_ldiskfs_dirent_param(ldp, fid);
4692 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
4693 child->d_fsdata = (void *)ldp;
4694 ll_vfs_dq_init(pobj->oo_inode);
4695 rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
4696 child, cinode, hlock);
4697 if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
4698 struct ldiskfs_dir_entry_2 *de;
4699 struct buffer_head *bh;
4702 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
4705 rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
4708 if (S_ISDIR(cinode->i_mode))
4709 de->file_type = LDISKFS_DIRENT_LUFID |
4710 LDISKFS_FT_REG_FILE;
4712 de->file_type = LDISKFS_DIRENT_LUFID |
4714 ldiskfs_handle_dirty_metadata(oth->ot_handle,
4725 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
4726 * into the directory.Also sets flags into osd object to
4727 * indicate dot and dotdot are created. This is required for
4728 * interoperability mode (b11826)
4730 * \param dir directory for dot and dotdot fixup.
4731 * \param obj child object for linking
4733 * \retval 0, on success
4734 * \retval -ve, on error
4736 static int osd_add_dot_dotdot(struct osd_thread_info *info,
4737 struct osd_object *dir,
4738 struct inode *parent_dir, const char *name,
4739 const struct lu_fid *dot_fid,
4740 const struct lu_fid *dot_dot_fid,
4743 struct inode *inode = dir->oo_inode;
4744 struct osd_thandle *oth;
4747 oth = container_of(th, struct osd_thandle, ot_super);
4748 LASSERT(oth->ot_handle->h_transaction != NULL);
4749 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4751 if (strcmp(name, dot) == 0) {
4752 if (dir->oo_compat_dot_created) {
4755 LASSERT(inode->i_ino == parent_dir->i_ino);
4756 dir->oo_compat_dot_created = 1;
4759 } else if (strcmp(name, dotdot) == 0) {
4760 if (!dir->oo_compat_dot_created)
4762 /* in case of rename, dotdot is already created */
4763 if (dir->oo_compat_dotdot_created) {
4764 return __osd_ea_add_rec(info, dir, parent_dir, name,
4765 dot_dot_fid, NULL, th);
4768 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
4769 struct lu_fid tfid = *dot_dot_fid;
4772 result = osd_add_dot_dotdot_internal(info,
4773 dir->oo_inode, parent_dir, dot_fid,
4776 result = osd_add_dot_dotdot_internal(info,
4777 dir->oo_inode, parent_dir, dot_fid,
4782 dir->oo_compat_dotdot_created = 1;
4790 * It will call the appropriate osd_add* function and return the
4791 * value, return by respective functions.
4793 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4794 struct inode *cinode, const char *name,
4795 const struct lu_fid *fid, struct thandle *th)
4797 struct osd_thread_info *info = osd_oti_get(env);
4798 struct htree_lock *hlock;
4801 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4803 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4805 if (hlock != NULL) {
4806 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4809 down_write(&pobj->oo_ext_idx_sem);
4812 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4813 lu_object_fid(&pobj->oo_dt.do_lu),
4816 if (hlock != NULL) {
4817 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4818 pobj->oo_inode, LDISKFS_HLOCK_ADD);
4820 down_write(&pobj->oo_ext_idx_sem);
4823 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4824 struct lu_fid *tfid = &info->oti_fid;
4828 rc = __osd_ea_add_rec(info, pobj, cinode, name,
4831 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4836 ldiskfs_htree_unlock(hlock);
4838 up_write(&pobj->oo_ext_idx_sem);
4844 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4845 struct osd_idmap_cache *oic)
4847 struct osd_scrub *scrub = &dev->od_scrub;
4848 struct lu_fid *fid = &oic->oic_fid;
4849 struct osd_inode_id *id = &oic->oic_lid;
4850 struct inode *inode = NULL;
4856 if (!fid_is_norm(fid) && !fid_is_igif(fid))
4859 if (scrub->os_pos_current > id->oii_ino)
4863 rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
4864 if (rc == -ENOENT) {
4865 __u32 gen = id->oii_gen;
4871 inode = osd_iget(oti, dev, id);
4872 /* The inode has been removed (by race maybe). */
4873 if (IS_ERR(inode)) {
4874 rc = PTR_ERR(inode);
4876 RETURN(rc == -ESTALE ? -ENOENT : rc);
4879 /* The OI mapping is lost. */
4880 if (gen != OSD_OII_NOGEN)
4884 /* The inode may has been reused by others, we do not know,
4885 * leave it to be handled by subsequent osd_fid_lookup(). */
4887 } else if (rc != 0 || osd_id_eq(id, &oti->oti_id)) {
4894 if (thread_is_running(&scrub->os_thread)) {
4895 if (inode == NULL) {
4896 inode = osd_iget(oti, dev, id);
4897 /* The inode has been removed (by race maybe). */
4898 if (IS_ERR(inode)) {
4899 rc = PTR_ERR(inode);
4901 RETURN(rc == -ESTALE ? -ENOENT : rc);
4905 rc = osd_oii_insert(dev, oic, insert);
4906 /* There is race condition between osd_oi_lookup and OI scrub.
4907 * The OI scrub finished just after osd_oi_lookup() failure.
4908 * Under such case, it is unnecessary to trigger OI scrub again,
4909 * but try to call osd_oi_lookup() again. */
4910 if (unlikely(rc == -EAGAIN))
4913 if (!S_ISDIR(inode->i_mode))
4916 rc = osd_check_lmv(oti, dev, inode, oic);
4922 if (!dev->od_noscrub && ++once == 1) {
4923 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
4925 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
4926 "%.16s: trigger partial OI scrub for RPC inconsistency "
4927 "checking FID "DFID": rc = %d\n",
4928 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4930 if (rc == 0 || rc == -EALREADY)
4940 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4941 struct osd_device *dev,
4942 struct osd_idmap_cache *oic,
4943 struct lu_fid *fid, __u32 ino)
4945 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
4946 struct inode *inode;
4949 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4950 inode = osd_iget(oti, dev, &oic->oic_lid);
4951 if (IS_ERR(inode)) {
4952 fid_zero(&oic->oic_fid);
4953 return PTR_ERR(inode);
4956 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4959 fid_zero(&oic->oic_fid);
4961 *fid = oic->oic_fid = lma->lma_self_fid;
4965 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4966 struct osd_inode_id *id, const struct lu_fid *fid)
4968 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4969 id->oii_ino, id->oii_gen, info);
4970 info->oti_cache.oic_lid = *id;
4971 info->oti_cache.oic_fid = *fid;
4972 info->oti_cache.oic_dev = osd;
4976 * Get parent FID from the linkEA.
4978 * For a directory which parent resides on remote MDT, to satisfy the
4979 * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
4980 * the other hand, to make the lookup(..) on the directory can return
4981 * the real parent FID, we append the real parent FID after its ".."
4982 * name entry in the /REMOTE_PARENT_DIR.
4984 * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
4985 * backup. So after the restore, we cannot get the right parent FID from
4986 * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
4987 * we have stored the real parent FID in the directory object's linkEA,
4988 * we can parse the linkEA for the real parent FID.
4990 * \param[in] env pointer to the thread context
4991 * \param[in] obj pointer to the object to be handled
4992 * \param[out]fid pointer to the buffer to hold the parent FID
4994 * \retval 0 for getting the real parent FID successfully
4995 * \retval negative error number on failure
4997 static int osd_get_pfid_from_linkea(const struct lu_env *env,
4998 struct osd_object *obj,
5001 struct osd_thread_info *oti = osd_oti_get(env);
5002 struct lu_buf *buf = &oti->oti_big_buf;
5003 struct dentry *dentry = &oti->oti_obj_dentry;
5004 struct inode *inode = obj->oo_inode;
5005 struct linkea_data ldata = { NULL };
5010 if (!S_ISDIR(inode->i_mode))
5014 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5015 buf->lb_buf, buf->lb_len);
5016 if (rc == -ERANGE) {
5017 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5020 lu_buf_realloc(buf, rc);
5021 if (buf->lb_buf == NULL)
5028 if (unlikely(rc == 0))
5034 if (unlikely(buf->lb_buf == NULL)) {
5035 lu_buf_realloc(buf, rc);
5036 if (buf->lb_buf == NULL)
5043 rc = linkea_init_with_rec(&ldata);
5045 linkea_first_entry(&ldata);
5046 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5052 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5053 struct inode *inode,
5054 const struct lu_fid *pfid,
5055 const char *name, const int namelen)
5057 struct osd_thread_info *oti = osd_oti_get(env);
5058 struct lu_buf *buf = &oti->oti_big_buf;
5059 struct dentry *dentry = &oti->oti_obj_dentry;
5060 struct linkea_data ldata = { NULL };
5061 struct lu_name cname = { .ln_name = name,
5062 .ln_namelen = namelen };
5067 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5068 buf->lb_buf, buf->lb_len);
5070 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5075 if (unlikely(rc == 0))
5078 if (buf->lb_len < rc) {
5079 lu_buf_realloc(buf, rc);
5080 if (buf->lb_buf == NULL)
5087 rc = linkea_init_with_rec(&ldata);
5089 rc = linkea_links_find(&ldata, &cname, pfid);
5095 * Calls ->lookup() to find dentry. From dentry get inode and
5096 * read inode's ea to get fid. This is required for interoperability
5099 * \retval 0, on success
5100 * \retval -ve, on error
5102 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5103 struct dt_rec *rec, const struct dt_key *key)
5105 struct inode *dir = obj->oo_inode;
5106 struct dentry *dentry;
5107 struct ldiskfs_dir_entry_2 *de;
5108 struct buffer_head *bh;
5109 struct lu_fid *fid = (struct lu_fid *) rec;
5110 struct htree_lock *hlock = NULL;
5115 LASSERT(dir->i_op != NULL);
5116 LASSERT(dir->i_op->lookup != NULL);
5118 dentry = osd_child_dentry_get(env, obj,
5119 (char *)key, strlen((char *)key));
5121 if (obj->oo_hl_head != NULL) {
5122 hlock = osd_oti_get(env)->oti_hlock;
5123 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5124 dir, LDISKFS_HLOCK_LOOKUP);
5126 down_read(&obj->oo_ext_idx_sem);
5129 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5131 struct osd_thread_info *oti = osd_oti_get(env);
5132 struct osd_inode_id *id = &oti->oti_id;
5133 struct osd_idmap_cache *oic = &oti->oti_cache;
5134 struct osd_device *dev = osd_obj2dev(obj);
5136 ino = le32_to_cpu(de->inode);
5137 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5139 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5143 rc = osd_get_fid_from_dentry(de, rec);
5145 /* done with de, release bh */
5148 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5149 const char *name = (const char *)key;
5151 /* If the parent is on remote MDT, and there
5152 * is no FID-in-dirent, then we have to get
5153 * the parent FID from the linkEA. */
5154 if (likely(strlen(name) == 2 &&
5155 name[0] == '.' && name[1] == '.'))
5156 rc = osd_get_pfid_from_linkea(env, obj,
5159 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5162 osd_id_gen(id, ino, OSD_OII_NOGEN);
5165 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5166 fid_zero(&oic->oic_fid);
5171 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5172 rc = osd_consistency_check(oti, dev, oic);
5174 fid_zero(&oic->oic_fid);
5183 ldiskfs_htree_unlock(hlock);
5185 up_read(&obj->oo_ext_idx_sem);
5190 * Put the osd object once done with it.
5192 * \param obj osd object that needs to be put
5194 static inline void osd_object_put(const struct lu_env *env,
5195 struct osd_object *obj)
5197 lu_object_put(env, &obj->oo_dt.do_lu);
5200 static int osd_index_declare_ea_insert(const struct lu_env *env,
5201 struct dt_object *dt,
5202 const struct dt_rec *rec,
5203 const struct dt_key *key,
5204 struct thandle *handle)
5206 struct osd_thandle *oh;
5207 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5208 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5209 const struct lu_fid *fid = rec1->rec_fid;
5210 int credits, rc = 0;
5211 struct osd_idmap_cache *idc;
5214 LASSERT(!dt_object_remote(dt));
5215 LASSERT(handle != NULL);
5216 LASSERT(fid != NULL);
5217 LASSERT(rec1->rec_type != 0);
5219 oh = container_of0(handle, struct osd_thandle, ot_super);
5220 LASSERT(oh->ot_handle == NULL);
5222 credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5224 /* we can't call iget() while a transactions is running
5225 * (this can lead to a deadlock), but we need to know
5226 * inum and object type. so we find this information at
5227 * declaration and cache in per-thread info */
5228 idc = osd_idc_find_or_init(env, osd, fid);
5230 RETURN(PTR_ERR(idc));
5231 if (idc->oic_remote) {
5232 /* a reference to remote inode is represented by an
5233 * agent inode which we have to create */
5234 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5235 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5238 osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5240 if (osd_dt_obj(dt)->oo_inode != NULL) {
5241 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5243 /* We ignore block quota on meta pool (MDTs), so needn't
5244 * calculate how many blocks will be consumed by this index
5246 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5247 i_gid_read(inode), 0, oh,
5248 osd_dt_obj(dt), true, NULL, false);
5255 * Index add function for interoperability mode (b11826).
5256 * It will add the directory entry.This entry is needed to
5257 * maintain name->fid mapping.
5259 * \param key it is key i.e. file entry to be inserted
5260 * \param rec it is value of given key i.e. fid
5262 * \retval 0, on success
5263 * \retval -ve, on error
5265 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5266 const struct dt_rec *rec,
5267 const struct dt_key *key, struct thandle *th,
5270 struct osd_object *obj = osd_dt_obj(dt);
5271 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5272 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5273 const struct lu_fid *fid = rec1->rec_fid;
5274 const char *name = (const char *)key;
5275 struct osd_thread_info *oti = osd_oti_get(env);
5276 struct inode *child_inode = NULL;
5277 struct osd_idmap_cache *idc;
5281 if (!dt_object_exists(dt))
5284 LASSERT(osd_invariant(obj));
5285 LASSERT(!dt_object_remote(dt));
5286 LASSERT(th != NULL);
5288 osd_trans_exec_op(env, th, OSD_OT_INSERT);
5290 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
5292 idc = osd_idc_find(env, osd, fid);
5293 if (unlikely(idc == NULL)) {
5294 /* this dt_insert() wasn't declared properly, so
5295 * FID is missing in OI cache. we better do not
5296 * lookup FID in FLDB/OI and don't risk to deadlock,
5297 * but in some special cases (lfsck testing, etc)
5298 * it's much simpler than fixing a caller */
5299 CERROR("%s: "DFID" wasn't declared for insert\n",
5300 osd_name(osd), PFID(fid));
5302 idc = osd_idc_find_or_init(env, osd, fid);
5304 RETURN(PTR_ERR(idc));
5307 if (idc->oic_remote) {
5308 /* Insert remote entry */
5309 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
5310 struct osd_mdobj_map *omm = osd->od_mdt_map;
5311 struct osd_thandle *oh;
5313 /* If parent on remote MDT, we need put this object
5315 oh = container_of(th, typeof(*oh), ot_super);
5316 rc = osd_add_to_remote_parent(env, osd, obj, oh);
5318 CERROR("%s: add "DFID" error: rc = %d\n",
5320 PFID(lu_object_fid(&dt->do_lu)), rc);
5324 child_inode = igrab(omm->omm_remote_parent->d_inode);
5326 child_inode = osd_create_local_agent_inode(env, osd,
5327 obj, fid, rec1->rec_type & S_IFMT, th);
5328 if (IS_ERR(child_inode))
5329 RETURN(PTR_ERR(child_inode));
5332 /* Insert local entry */
5333 if (unlikely(idc->oic_lid.oii_ino == 0)) {
5334 /* for a reason OI cache wasn't filled properly */
5335 CERROR("%s: OIC for "DFID" isn't filled\n",
5336 osd_name(osd), PFID(fid));
5339 child_inode = oti->oti_inode;
5340 if (unlikely(child_inode == NULL)) {
5341 struct ldiskfs_inode_info *lii;
5345 child_inode = oti->oti_inode = &lii->vfs_inode;
5347 child_inode->i_sb = osd_sb(osd);
5348 child_inode->i_ino = idc->oic_lid.oii_ino;
5349 child_inode->i_mode = rec1->rec_type & S_IFMT;
5352 rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
5354 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
5355 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
5357 if (child_inode && child_inode != oti->oti_inode)
5359 LASSERT(osd_invariant(obj));
5360 osd_trans_exec_check(env, th, OSD_OT_INSERT);
5365 * Initialize osd Iterator for given osd index object.
5367 * \param dt osd index object
5370 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
5371 struct dt_object *dt,
5374 struct osd_it_iam *it;
5375 struct osd_object *obj = osd_dt_obj(dt);
5376 struct lu_object *lo = &dt->do_lu;
5377 struct iam_path_descr *ipd;
5378 struct iam_container *bag = &obj->oo_dir->od_container;
5380 if (!dt_object_exists(dt))
5381 return ERR_PTR(-ENOENT);
5385 return ERR_PTR(-ENOMEM);
5387 ipd = osd_it_ipd_get(env, bag);
5388 if (likely(ipd != NULL)) {
5392 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
5393 return (struct dt_it *)it;
5396 return ERR_PTR(-ENOMEM);
5401 * free given Iterator.
5404 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
5406 struct osd_it_iam *it = (struct osd_it_iam *)di;
5407 struct osd_object *obj = it->oi_obj;
5409 iam_it_fini(&it->oi_it);
5410 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
5411 lu_object_put(env, &obj->oo_dt.do_lu);
5416 * Move Iterator to record specified by \a key
5418 * \param di osd iterator
5419 * \param key key for index
5421 * \retval +ve di points to record with least key not larger than key
5422 * \retval 0 di points to exact matched key
5423 * \retval -ve failure
5426 static int osd_it_iam_get(const struct lu_env *env,
5427 struct dt_it *di, const struct dt_key *key)
5429 struct osd_thread_info *oti = osd_oti_get(env);
5430 struct osd_it_iam *it = (struct osd_it_iam *)di;
5432 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5433 /* swab quota uid/gid */
5434 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5435 key = (struct dt_key *)&oti->oti_quota_id;
5438 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
5444 * \param di osd iterator
5446 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
5448 struct osd_it_iam *it = (struct osd_it_iam *)di;
5450 iam_it_put(&it->oi_it);
5454 * Move iterator by one record
5456 * \param di osd iterator
5458 * \retval +1 end of container reached
5460 * \retval -ve failure
5463 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
5465 struct osd_it_iam *it = (struct osd_it_iam *)di;
5467 return iam_it_next(&it->oi_it);
5471 * Return pointer to the key under iterator.
5474 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
5475 const struct dt_it *di)
5477 struct osd_thread_info *oti = osd_oti_get(env);
5478 struct osd_it_iam *it = (struct osd_it_iam *)di;
5479 struct osd_object *obj = it->oi_obj;
5482 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
5484 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
5485 /* swab quota uid/gid */
5486 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
5487 key = (struct dt_key *)&oti->oti_quota_id;
5494 * Return size of key under iterator (in bytes)
5497 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
5499 struct osd_it_iam *it = (struct osd_it_iam *)di;
5501 return iam_it_key_size(&it->oi_it);
5505 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
5507 /* check if file type is required */
5508 if (ent->lde_attrs & LUDA_TYPE) {
5509 struct luda_type *lt;
5510 int align = sizeof(*lt) - 1;
5512 len = (len + align) & ~align;
5513 lt = (struct luda_type *)(ent->lde_name + len);
5514 lt->lt_type = cpu_to_le16(DTTOIF(type));
5517 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
5521 * build lu direct from backend fs dirent.
5525 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
5526 char *name, __u16 namelen, __u16 type, __u32 attr)
5528 ent->lde_attrs = attr | LUDA_FID;
5529 fid_cpu_to_le(&ent->lde_fid, fid);
5531 ent->lde_hash = cpu_to_le64(offset);
5532 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
5534 strncpy(ent->lde_name, name, namelen);
5535 ent->lde_name[namelen] = '\0';
5536 ent->lde_namelen = cpu_to_le16(namelen);
5538 /* append lustre attributes */
5539 osd_it_append_attrs(ent, namelen, type);
5543 * Return pointer to the record under iterator.
5545 static int osd_it_iam_rec(const struct lu_env *env,
5546 const struct dt_it *di,
5547 struct dt_rec *dtrec, __u32 attr)
5549 struct osd_it_iam *it = (struct osd_it_iam *)di;
5550 struct osd_thread_info *info = osd_oti_get(env);
5553 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
5554 const struct osd_fid_pack *rec;
5555 struct lu_fid *fid = &info->oti_fid;
5556 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
5562 name = (char *)iam_it_key_get(&it->oi_it);
5564 RETURN(PTR_ERR(name));
5566 namelen = iam_it_key_size(&it->oi_it);
5568 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
5570 RETURN(PTR_ERR(rec));
5572 rc = osd_fid_unpack(fid, rec);
5576 hash = iam_it_store(&it->oi_it);
5578 /* IAM does not store object type in IAM index (dir) */
5579 osd_it_pack_dirent(lde, fid, hash, name, namelen,
5581 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5582 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5583 (struct iam_rec *)dtrec);
5584 osd_quota_unpack(it->oi_obj, dtrec);
5586 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5587 (struct iam_rec *)dtrec);
5594 * Returns cookie for current Iterator position.
5596 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
5598 struct osd_it_iam *it = (struct osd_it_iam *)di;
5600 return iam_it_store(&it->oi_it);
5604 * Restore iterator from cookie.
5606 * \param di osd iterator
5607 * \param hash Iterator location cookie
5609 * \retval +ve di points to record with least key not larger than key.
5610 * \retval 0 di points to exact matched key
5611 * \retval -ve failure
5614 static int osd_it_iam_load(const struct lu_env *env,
5615 const struct dt_it *di, __u64 hash)
5617 struct osd_it_iam *it = (struct osd_it_iam *)di;
5619 return iam_it_load(&it->oi_it, hash);
5622 static const struct dt_index_operations osd_index_iam_ops = {
5623 .dio_lookup = osd_index_iam_lookup,
5624 .dio_declare_insert = osd_index_declare_iam_insert,
5625 .dio_insert = osd_index_iam_insert,
5626 .dio_declare_delete = osd_index_declare_iam_delete,
5627 .dio_delete = osd_index_iam_delete,
5629 .init = osd_it_iam_init,
5630 .fini = osd_it_iam_fini,
5631 .get = osd_it_iam_get,
5632 .put = osd_it_iam_put,
5633 .next = osd_it_iam_next,
5634 .key = osd_it_iam_key,
5635 .key_size = osd_it_iam_key_size,
5636 .rec = osd_it_iam_rec,
5637 .store = osd_it_iam_store,
5638 .load = osd_it_iam_load
5644 * Creates or initializes iterator context.
5646 * \retval struct osd_it_ea, iterator structure on success
5649 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
5650 struct dt_object *dt,
5653 struct osd_object *obj = osd_dt_obj(dt);
5654 struct osd_thread_info *info = osd_oti_get(env);
5655 struct osd_it_ea *oie;
5657 struct lu_object *lo = &dt->do_lu;
5658 struct dentry *obj_dentry;
5661 if (!dt_object_exists(dt) || obj->oo_destroyed)
5662 RETURN(ERR_PTR(-ENOENT));
5664 OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
5666 RETURN(ERR_PTR(-ENOMEM));
5667 obj_dentry = &oie->oie_dentry;
5669 obj_dentry->d_inode = obj->oo_inode;
5670 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
5671 obj_dentry->d_name.hash = 0;
5673 oie->oie_rd_dirent = 0;
5674 oie->oie_it_dirent = 0;
5675 oie->oie_dirent = NULL;
5676 if (unlikely(!info->oti_it_ea_buf_used)) {
5677 oie->oie_buf = info->oti_it_ea_buf;
5678 info->oti_it_ea_buf_used = 1;
5680 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5681 if (oie->oie_buf == NULL)
5682 RETURN(ERR_PTR(-ENOMEM));
5686 file = &oie->oie_file;
5688 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
5689 if (attr & LUDA_64BITHASH)
5690 file->f_mode = FMODE_64BITHASH;
5692 file->f_mode = FMODE_32BITHASH;
5693 file->f_path.dentry = obj_dentry;
5694 file->f_mapping = obj->oo_inode->i_mapping;
5695 file->f_op = obj->oo_inode->i_fop;
5696 set_file_inode(file, obj->oo_inode);
5699 RETURN((struct dt_it *) oie);
5703 * Destroy or finishes iterator context.
5705 * \param di iterator structure to be destroyed
5707 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
5709 struct osd_thread_info *info = osd_oti_get(env);
5710 struct osd_it_ea *oie = (struct osd_it_ea *)di;
5711 struct osd_object *obj = oie->oie_obj;
5712 struct inode *inode = obj->oo_inode;
5715 oie->oie_file.f_op->release(inode, &oie->oie_file);
5716 lu_object_put(env, &obj->oo_dt.do_lu);
5717 if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
5718 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5720 info->oti_it_ea_buf_used = 0;
5721 OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
5726 * It position the iterator at given key, so that next lookup continues from
5727 * that key Or it is similar to dio_it->load() but based on a key,
5728 * rather than file position.
5730 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
5733 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
5735 static int osd_it_ea_get(const struct lu_env *env,
5736 struct dt_it *di, const struct dt_key *key)
5738 struct osd_it_ea *it = (struct osd_it_ea *)di;
5741 LASSERT(((const char *)key)[0] == '\0');
5742 it->oie_file.f_pos = 0;
5743 it->oie_rd_dirent = 0;
5744 it->oie_it_dirent = 0;
5745 it->oie_dirent = NULL;
5753 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
5757 struct osd_filldir_cbs {
5758 #ifdef HAVE_DIR_CONTEXT
5759 struct dir_context ctx;
5761 struct osd_it_ea *it;
5764 * It is called internally by ->readdir(). It fills the
5765 * iterator's in-memory data structure with required
5766 * information i.e. name, namelen, rec_size etc.
5768 * \param buf in which information to be filled in.
5769 * \param name name of the file in given dir
5771 * \retval 0 on success
5772 * \retval 1 on buffer full
5774 #ifdef HAVE_FILLDIR_USE_CTX
5775 static int osd_ldiskfs_filldir(struct dir_context *buf,
5777 static int osd_ldiskfs_filldir(void *buf,
5779 const char *name, int namelen,
5780 loff_t offset, __u64 ino, unsigned d_type)
5782 struct osd_it_ea *it =
5783 ((struct osd_filldir_cbs *)buf)->it;
5784 struct osd_object *obj = it->oie_obj;
5785 struct osd_it_ea_dirent *ent = it->oie_dirent;
5786 struct lu_fid *fid = &ent->oied_fid;
5787 struct osd_fid_pack *rec;
5790 /* this should never happen */
5791 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5792 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5796 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
5800 /* "." is just the object itself. */
5801 if (namelen == 1 && name[0] == '.') {
5802 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5803 } else if (d_type & LDISKFS_DIRENT_LUFID) {
5804 rec = (struct osd_fid_pack*) (name + namelen + 1);
5805 if (osd_fid_unpack(fid, rec) != 0)
5810 d_type &= ~LDISKFS_DIRENT_LUFID;
5812 /* NOT export local root. */
5813 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5814 ino = obj->oo_inode->i_ino;
5815 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5818 ent->oied_ino = ino;
5819 ent->oied_off = offset;
5820 ent->oied_namelen = namelen;
5821 ent->oied_type = d_type;
5823 memcpy(ent->oied_name, name, namelen);
5825 it->oie_rd_dirent++;
5826 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
5831 * Calls ->readdir() to load a directory entry at a time
5832 * and stored it in iterator's in-memory data structure.
5834 * \param di iterator's in memory structure
5836 * \retval 0 on success
5837 * \retval -ve on error
5838 * \retval +1 reach the end of entry
5840 static int osd_ldiskfs_it_fill(const struct lu_env *env,
5841 const struct dt_it *di)
5843 struct osd_it_ea *it = (struct osd_it_ea *)di;
5844 struct osd_object *obj = it->oie_obj;
5845 struct inode *inode = obj->oo_inode;
5846 struct htree_lock *hlock = NULL;
5847 struct file *filp = &it->oie_file;
5849 struct osd_filldir_cbs buf = {
5850 #ifdef HAVE_DIR_CONTEXT
5851 .ctx.actor = osd_ldiskfs_filldir,
5857 it->oie_dirent = it->oie_buf;
5858 it->oie_rd_dirent = 0;
5860 if (obj->oo_hl_head != NULL) {
5861 hlock = osd_oti_get(env)->oti_hlock;
5862 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5863 inode, LDISKFS_HLOCK_READDIR);
5865 down_read(&obj->oo_ext_idx_sem);
5868 #ifdef HAVE_DIR_CONTEXT
5869 buf.ctx.pos = filp->f_pos;
5870 rc = inode->i_fop->iterate(filp, &buf.ctx);
5871 filp->f_pos = buf.ctx.pos;
5873 rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
5877 ldiskfs_htree_unlock(hlock);
5879 up_read(&obj->oo_ext_idx_sem);
5881 if (it->oie_rd_dirent == 0) {
5882 /*If it does not get any dirent, it means it has been reached
5883 *to the end of the dir */
5884 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
5888 it->oie_dirent = it->oie_buf;
5889 it->oie_it_dirent = 1;
5896 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5897 * to load a directory entry at a time and stored it in
5898 * iterator's in-memory data structure.
5900 * \param di iterator's in memory structure
5902 * \retval +ve iterator reached to end
5903 * \retval 0 iterator not reached to end
5904 * \retval -ve on error
5906 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
5908 struct osd_it_ea *it = (struct osd_it_ea *)di;
5913 if (it->oie_it_dirent < it->oie_rd_dirent) {
5915 (void *) it->oie_dirent +
5916 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
5917 it->oie_dirent->oied_namelen);
5918 it->oie_it_dirent++;
5921 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
5924 rc = osd_ldiskfs_it_fill(env, di);
5931 * Returns the key at current position from iterator's in memory structure.
5933 * \param di iterator's in memory structure
5935 * \retval key i.e. struct dt_key on success
5937 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
5938 const struct dt_it *di)
5940 struct osd_it_ea *it = (struct osd_it_ea *)di;
5942 return (struct dt_key *)it->oie_dirent->oied_name;
5946 * Returns the key's size at current position from iterator's in memory structure.
5948 * \param di iterator's in memory structure
5950 * \retval key_size i.e. struct dt_key on success
5952 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
5954 struct osd_it_ea *it = (struct osd_it_ea *)di;
5956 return it->oie_dirent->oied_namelen;
5959 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
5961 if (LDISKFS_DIR_REC_LEN(de) >=
5962 __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
5969 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
5970 unsigned blocksize, bool dotdot)
5973 return osd_dotdot_has_space(de);
5975 if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
5976 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5983 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
5984 handle_t *jh, struct dentry *dentry,
5985 const struct lu_fid *fid, struct buffer_head *bh,
5986 struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
5989 struct inode *dir = dentry->d_parent->d_inode;
5990 struct inode *inode = dentry->d_inode;
5991 struct osd_fid_pack *rec;
5992 struct ldiskfs_dentry_param *ldp;
5993 int namelen = dentry->d_name.len;
5995 struct osd_thread_info *info = osd_oti_get(env);
5998 if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5999 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6002 /* There is enough space to hold the FID-in-dirent. */
6003 if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6004 rc = ldiskfs_journal_get_write_access(jh, bh);
6008 de->name[namelen] = 0;
6009 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6010 rec->fp_len = sizeof(struct lu_fid) + 1;
6011 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6012 de->file_type |= LDISKFS_DIRENT_LUFID;
6013 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6020 rc = ldiskfs_delete_entry(jh, dir, de, bh);
6024 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6025 osd_get_ldiskfs_dirent_param(ldp, fid);
6026 dentry->d_fsdata = (void *)ldp;
6027 ll_vfs_dq_init(dir);
6028 rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6029 /* It is too bad, we cannot reinsert the name entry back.
6030 * That means we lose it! */
6032 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
6033 "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6034 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
6035 dir->i_ino, dir->i_generation, namelen,
6036 dentry->d_name.name, PFID(fid), rc);
6042 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6043 struct osd_it_ea *it, struct lu_fid *fid,
6044 struct osd_inode_id *id, __u32 *attr)
6046 struct osd_thread_info *info = osd_oti_get(env);
6047 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
6048 struct osd_device *dev = osd_obj2dev(obj);
6049 struct super_block *sb = osd_sb(dev);
6050 const char *devname = osd_name(dev);
6051 struct osd_it_ea_dirent *ent = it->oie_dirent;
6052 struct inode *dir = obj->oo_inode;
6053 struct htree_lock *hlock = NULL;
6054 struct buffer_head *bh = NULL;
6055 handle_t *jh = NULL;
6056 struct ldiskfs_dir_entry_2 *de;
6057 struct dentry *dentry;
6058 struct inode *inode;
6059 const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6062 bool dotdot = false;
6066 if (ent->oied_name[0] == '.') {
6067 if (ent->oied_namelen == 1)
6070 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6074 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6075 inode = osd_iget(info, dev, id);
6076 if (IS_ERR(inode)) {
6077 rc = PTR_ERR(inode);
6078 if (rc == -ENOENT || rc == -ESTALE) {
6079 /* Maybe dangling name entry, or
6080 * corrupted directory entry. */
6081 *attr |= LUDA_UNKNOWN;
6084 CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6085 "check_repair, dir = %lu/%u, name = %.*s, "
6086 "ino = %llu, rc = %d\n",
6087 devname, dir->i_ino, dir->i_generation,
6088 ent->oied_namelen, ent->oied_name,
6095 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6097 rc = osd_get_lma(info, inode, dentry, lma);
6098 if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6103 /* We need to ensure that the name entry is still valid.
6104 * Because it may be removed or renamed by other already.
6106 * The unlink or rename operation will start journal before PDO lock,
6107 * so to avoid deadlock, here we need to start journal handle before
6108 * related PDO lock also. But because we do not know whether there
6109 * will be something to be repaired before PDO lock, we just start
6110 * journal without conditions.
6112 * We may need to remove the name entry firstly, then insert back.
6113 * One credit is for user quota file update.
6114 * One credit is for group quota file update.
6115 * Two credits are for dirty inode. */
6116 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6117 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6119 if (dev->od_dirent_journal != 0) {
6122 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6125 CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6126 "check_repair, dir = %lu/%u, credits = %d, "
6127 "name = %.*s, ino = %llu: rc = %d\n",
6128 devname, dir->i_ino, dir->i_generation, credits,
6129 ent->oied_namelen, ent->oied_name,
6132 GOTO(out_inode, rc);
6135 if (obj->oo_hl_head != NULL) {
6136 hlock = osd_oti_get(env)->oti_hlock;
6137 /* "0" means exclusive lock for the whole directory.
6138 * We need to prevent others access such name entry
6139 * during the delete + insert. Neither HLOCK_ADD nor
6140 * HLOCK_DEL cannot guarantee the atomicity. */
6141 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6143 down_write(&obj->oo_ext_idx_sem);
6146 if (obj->oo_hl_head != NULL) {
6147 hlock = osd_oti_get(env)->oti_hlock;
6148 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6149 LDISKFS_HLOCK_LOOKUP);
6151 down_read(&obj->oo_ext_idx_sem);
6155 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6156 if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6157 *attr |= LUDA_IGNORE;
6162 /* For dotdot entry, if there is not enough space to hold the
6163 * FID-in-dirent, just keep them there. It only happens when the
6164 * device upgraded from 1.8 or restored from MDT file-level backup.
6165 * For the whole directory, only dotdot entry have no FID-in-dirent
6166 * and needs to get FID from LMA when readdir, it will not affect the
6167 * performance much. */
6168 if (dotdot && !osd_dotdot_has_space(de)) {
6169 *attr |= LUDA_UNKNOWN;
6175 if (lu_fid_eq(fid, &lma->lma_self_fid))
6178 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6179 struct lu_fid *tfid = &lma->lma_self_fid;
6181 if (likely(dotdot &&
6182 fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6183 fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6184 /* It must be REMOTE_PARENT_DIR and as the
6185 * 'dotdot' entry of remote directory */
6186 *attr |= LUDA_IGNORE;
6188 CDEBUG(D_LFSCK, "%s: expect remote agent "
6189 "parent directory, but got %.*s under "
6190 "dir = %lu/%u with the FID "DFID"\n",
6191 devname, ent->oied_namelen,
6192 ent->oied_name, dir->i_ino,
6193 dir->i_generation, PFID(tfid));
6195 *attr |= LUDA_UNKNOWN;
6202 if (!fid_is_zero(fid)) {
6203 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6205 if (rc == -ENOENT ||
6207 !(dev->od_scrub.os_file.sf_flags & SF_UPGRADE))) {
6208 /* linkEA does not recognize the dirent entry,
6209 * it may because the dirent entry corruption
6210 * and points to other's inode. */
6211 CDEBUG(D_LFSCK, "%s: the target inode does not "
6212 "recognize the dirent, dir = %lu/%u, "
6213 " name = %.*s, ino = %llu, "
6214 DFID": rc = %d\n", devname, dir->i_ino,
6215 dir->i_generation, ent->oied_namelen,
6216 ent->oied_name, ent->oied_ino, PFID(fid), rc);
6217 *attr |= LUDA_UNKNOWN;
6222 if (rc && rc != -ENODATA) {
6223 CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6224 "dir = %lu/%u, name = %.*s, ino = %llu, "
6225 DFID": rc = %d\n", devname, dir->i_ino,
6226 dir->i_generation, ent->oied_namelen,
6227 ent->oied_name, ent->oied_ino, PFID(fid), rc);
6228 *attr |= LUDA_UNKNOWN;
6235 /* linkEA recognizes the dirent entry, the FID-in-LMA is
6236 * valid, trusted, in spite of fid_is_sane(fid) or not. */
6237 if (*attr & LUDA_VERIFY_DRYRUN) {
6238 *fid = lma->lma_self_fid;
6239 *attr |= LUDA_REPAIR;
6246 dev->od_dirent_journal = 1;
6247 if (hlock != NULL) {
6248 ldiskfs_htree_unlock(hlock);
6251 up_read(&obj->oo_ext_idx_sem);
6257 *fid = lma->lma_self_fid;
6259 /* Update or append the FID-in-dirent. */
6260 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6261 bh, de, hlock, dotdot);
6263 *attr |= LUDA_REPAIR;
6265 CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
6266 "the dirent, dir = %lu/%u, name = %.*s, "
6267 "ino = %llu, "DFID": rc = %d\n",
6268 devname, dir->i_ino, dir->i_generation,
6269 ent->oied_namelen, ent->oied_name,
6270 ent->oied_ino, PFID(fid), rc);
6272 /* lma is NULL, trust the FID-in-dirent if it is valid. */
6273 if (*attr & LUDA_VERIFY_DRYRUN) {
6274 if (fid_is_sane(fid)) {
6275 *attr |= LUDA_REPAIR;
6276 } else if (dev->od_index == 0) {
6277 lu_igif_build(fid, inode->i_ino,
6278 inode->i_generation);
6279 *attr |= LUDA_UPGRADE;
6287 dev->od_dirent_journal = 1;
6288 if (hlock != NULL) {
6289 ldiskfs_htree_unlock(hlock);
6292 up_read(&obj->oo_ext_idx_sem);
6299 if (unlikely(fid_is_sane(fid))) {
6300 /* FID-in-dirent exists, but FID-in-LMA is lost.
6301 * Trust the FID-in-dirent, and add FID-in-LMA. */
6302 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
6304 *attr |= LUDA_REPAIR;
6306 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
6307 "update dirent, dir = %lu/%u, "
6308 "name = %.*s, ino = %llu, "
6310 devname, dir->i_ino, dir->i_generation,
6311 ent->oied_namelen, ent->oied_name,
6312 ent->oied_ino, PFID(fid), rc);
6313 } else if (dev->od_index == 0) {
6314 lu_igif_build(fid, inode->i_ino, inode->i_generation);
6315 /* It is probably IGIF object. Only aappend the
6316 * FID-in-dirent. OI scrub will process FID-in-LMA. */
6317 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6318 bh, de, hlock, dotdot);
6320 *attr |= LUDA_UPGRADE;
6322 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
6323 "after the dirent, dir = %lu/%u, "
6324 "name = %.*s, ino = %llu, "
6326 devname, dir->i_ino, dir->i_generation,
6327 ent->oied_namelen, ent->oied_name,
6328 ent->oied_ino, PFID(fid), rc);
6337 if (hlock != NULL) {
6338 ldiskfs_htree_unlock(hlock);
6340 if (dev->od_dirent_journal != 0)
6341 up_write(&obj->oo_ext_idx_sem);
6343 up_read(&obj->oo_ext_idx_sem);
6347 ldiskfs_journal_stop(jh);
6351 if (rc >= 0 && !dirty)
6352 dev->od_dirent_journal = 0;
6358 * Returns the value at current position from iterator's in memory structure.
6360 * \param di struct osd_it_ea, iterator's in memory structure
6361 * \param attr attr requested for dirent.
6362 * \param lde lustre dirent
6364 * \retval 0 no error and \param lde has correct lustre dirent.
6365 * \retval -ve on error
6367 static inline int osd_it_ea_rec(const struct lu_env *env,
6368 const struct dt_it *di,
6369 struct dt_rec *dtrec, __u32 attr)
6371 struct osd_it_ea *it = (struct osd_it_ea *)di;
6372 struct osd_object *obj = it->oie_obj;
6373 struct osd_device *dev = osd_obj2dev(obj);
6374 struct osd_thread_info *oti = osd_oti_get(env);
6375 struct osd_inode_id *id = &oti->oti_id;
6376 struct lu_fid *fid = &it->oie_dirent->oied_fid;
6377 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6378 __u32 ino = it->oie_dirent->oied_ino;
6382 LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
6384 if (attr & LUDA_VERIFY) {
6385 if (unlikely(ino == osd_remote_parent_ino(dev))) {
6386 attr |= LUDA_IGNORE;
6387 /* If the parent is on remote MDT, and there
6388 * is no FID-in-dirent, then we have to get
6389 * the parent FID from the linkEA. */
6390 if (!fid_is_sane(fid) &&
6391 it->oie_dirent->oied_namelen == 2 &&
6392 it->oie_dirent->oied_name[0] == '.' &&
6393 it->oie_dirent->oied_name[1] == '.')
6394 osd_get_pfid_from_linkea(env, obj, fid);
6396 rc = osd_dirent_check_repair(env, obj, it, fid, id,
6400 if (!fid_is_sane(fid))
6401 attr |= LUDA_UNKNOWN;
6403 attr &= ~LU_DIRENT_ATTRS_MASK;
6404 if (!fid_is_sane(fid)) {
6405 bool is_dotdot = false;
6406 if (it->oie_dirent->oied_namelen == 2 &&
6407 it->oie_dirent->oied_name[0] == '.' &&
6408 it->oie_dirent->oied_name[1] == '.')
6410 /* If the parent is on remote MDT, and there
6411 * is no FID-in-dirent, then we have to get
6412 * the parent FID from the linkEA. */
6413 if (ino == osd_remote_parent_ino(dev) && is_dotdot) {
6414 rc = osd_get_pfid_from_linkea(env, obj, fid);
6416 if (is_dotdot == false &&
6417 OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
6420 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6423 osd_id_gen(id, ino, OSD_OII_NOGEN);
6427 /* Pack the entry anyway, at least the offset is right. */
6428 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
6429 it->oie_dirent->oied_name,
6430 it->oie_dirent->oied_namelen,
6431 it->oie_dirent->oied_type, attr);
6436 if (osd_remote_fid(env, dev, fid))
6439 if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
6440 osd_add_oi_cache(oti, dev, id, fid);
6442 RETURN(rc > 0 ? 0 : rc);
6446 * Returns the record size size at current position.
6448 * This function will return record(lu_dirent) size in bytes.
6450 * \param[in] env execution environment
6451 * \param[in] di iterator's in memory structure
6452 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
6453 * calculate the lu_dirent size.
6455 * \retval record size(in bytes & in memory) of the current lu_dirent
6458 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
6461 struct osd_it_ea *it = (struct osd_it_ea *)di;
6463 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
6467 * Returns a cookie for current position of the iterator head, so that
6468 * user can use this cookie to load/start the iterator next time.
6470 * \param di iterator's in memory structure
6472 * \retval cookie for current position, on success
6474 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
6476 struct osd_it_ea *it = (struct osd_it_ea *)di;
6478 return it->oie_dirent->oied_off;
6482 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6483 * to load a directory entry at a time and stored it i inn,
6484 * in iterator's in-memory data structure.
6486 * \param di struct osd_it_ea, iterator's in memory structure
6488 * \retval +ve on success
6489 * \retval -ve on error
6491 static int osd_it_ea_load(const struct lu_env *env,
6492 const struct dt_it *di, __u64 hash)
6494 struct osd_it_ea *it = (struct osd_it_ea *)di;
6498 it->oie_file.f_pos = hash;
6500 rc = osd_ldiskfs_it_fill(env, di);
6511 * Index lookup function for interoperability mode (b11826).
6513 * \param key, key i.e. file name to be searched
6515 * \retval +ve, on success
6516 * \retval -ve, on error
6518 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
6519 struct dt_rec *rec, const struct dt_key *key)
6521 struct osd_object *obj = osd_dt_obj(dt);
6526 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
6527 LINVRNT(osd_invariant(obj));
6529 rc = osd_ea_lookup_rec(env, obj, rec, key);
6536 * Index and Iterator operations for interoperability
6537 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
6539 static const struct dt_index_operations osd_index_ea_ops = {
6540 .dio_lookup = osd_index_ea_lookup,
6541 .dio_declare_insert = osd_index_declare_ea_insert,
6542 .dio_insert = osd_index_ea_insert,
6543 .dio_declare_delete = osd_index_declare_ea_delete,
6544 .dio_delete = osd_index_ea_delete,
6546 .init = osd_it_ea_init,
6547 .fini = osd_it_ea_fini,
6548 .get = osd_it_ea_get,
6549 .put = osd_it_ea_put,
6550 .next = osd_it_ea_next,
6551 .key = osd_it_ea_key,
6552 .key_size = osd_it_ea_key_size,
6553 .rec = osd_it_ea_rec,
6554 .rec_size = osd_it_ea_rec_size,
6555 .store = osd_it_ea_store,
6556 .load = osd_it_ea_load
6560 static void *osd_key_init(const struct lu_context *ctx,
6561 struct lu_context_key *key)
6563 struct osd_thread_info *info;
6565 OBD_ALLOC_PTR(info);
6567 return ERR_PTR(-ENOMEM);
6569 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6570 if (info->oti_it_ea_buf == NULL)
6573 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
6575 info->oti_hlock = ldiskfs_htree_lock_alloc();
6576 if (info->oti_hlock == NULL)
6582 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6585 return ERR_PTR(-ENOMEM);
6588 static void osd_key_fini(const struct lu_context *ctx,
6589 struct lu_context_key *key, void* data)
6591 struct osd_thread_info *info = data;
6592 struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
6593 struct osd_idmap_cache *idc = info->oti_ins_cache;
6595 if (info->oti_inode != NULL)
6597 if (info->oti_hlock != NULL)
6598 ldiskfs_htree_lock_free(info->oti_hlock);
6599 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6600 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
6601 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
6602 lu_buf_free(&info->oti_big_buf);
6604 LASSERT(info->oti_ins_cache_size > 0);
6605 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
6606 info->oti_ins_cache = NULL;
6607 info->oti_ins_cache_size = 0;
6612 static void osd_key_exit(const struct lu_context *ctx,
6613 struct lu_context_key *key, void *data)
6615 struct osd_thread_info *info = data;
6617 LASSERT(info->oti_r_locks == 0);
6618 LASSERT(info->oti_w_locks == 0);
6619 LASSERT(info->oti_txns == 0);
6622 /* type constructor/destructor: osd_type_init, osd_type_fini */
6623 LU_TYPE_INIT_FINI(osd, &osd_key);
6625 struct lu_context_key osd_key = {
6626 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
6627 .lct_init = osd_key_init,
6628 .lct_fini = osd_key_fini,
6629 .lct_exit = osd_key_exit
6633 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
6634 const char *name, struct lu_device *next)
6636 struct osd_device *osd = osd_dev(d);
6638 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
6639 >= sizeof(osd->od_svname))
6641 return osd_procfs_init(osd, name);
6644 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
6646 struct seq_server_site *ss = osd_seq_site(osd);
6650 if (osd->od_is_ost || osd->od_cl_seq != NULL)
6653 if (unlikely(ss == NULL))
6656 OBD_ALLOC_PTR(osd->od_cl_seq);
6657 if (osd->od_cl_seq == NULL)
6660 rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
6661 osd->od_svname, ss->ss_server_seq);
6663 OBD_FREE_PTR(osd->od_cl_seq);
6664 osd->od_cl_seq = NULL;
6668 if (ss->ss_node_id == 0) {
6669 /* If the OSD on the sequence controller(MDT0), then allocate
6670 * sequence here, otherwise allocate sequence after connected
6671 * to MDT0 (see mdt_register_lwp_callback()). */
6672 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
6673 &osd->od_cl_seq->lcs_space, env);
6679 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
6681 if (osd->od_cl_seq == NULL)
6684 seq_client_fini(osd->od_cl_seq);
6685 OBD_FREE_PTR(osd->od_cl_seq);
6686 osd->od_cl_seq = NULL;
6689 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
6693 /* shutdown quota slave instance associated with the device */
6694 if (o->od_quota_slave != NULL) {
6695 qsd_fini(env, o->od_quota_slave);
6696 o->od_quota_slave = NULL;
6699 osd_fid_fini(env, o);
6704 static void osd_umount(const struct lu_env *env, struct osd_device *o)
6708 if (o->od_mnt != NULL) {
6709 shrink_dcache_sb(osd_sb(o));
6710 osd_sync(env, &o->od_dt_dev);
6719 static int osd_mount(const struct lu_env *env,
6720 struct osd_device *o, struct lustre_cfg *cfg)
6722 const char *name = lustre_cfg_string(cfg, 0);
6723 const char *dev = lustre_cfg_string(cfg, 1);
6725 unsigned long page, s_flags, lmd_flags = 0;
6726 struct page *__page;
6727 struct file_system_type *type;
6728 char *options = NULL;
6730 struct osd_thread_info *info = osd_oti_get(env);
6731 struct lu_fid *fid = &info->oti_fid;
6732 struct inode *inode;
6733 int rc = 0, force_over_512tb = 0;
6736 if (o->od_mnt != NULL)
6739 if (strlen(dev) >= sizeof(o->od_mntdev))
6741 strcpy(o->od_mntdev, dev);
6743 str = lustre_cfg_string(cfg, 2);
6744 s_flags = simple_strtoul(str, NULL, 0);
6745 str = strstr(str, ":");
6747 lmd_flags = simple_strtoul(str + 1, NULL, 0);
6748 opts = lustre_cfg_string(cfg, 3);
6750 if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
6751 CERROR("%s: device %s extents feature is not guaranteed to "
6752 "work on big-endian systems. Use \"bigendian_extents\" "
6753 "mount option to override.\n", name, dev);
6757 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
6758 if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
6759 CWARN("force_over_128tb option is deprecated. "
6760 "Filesystems less than 512TB can be created without any "
6761 "force options. Use force_over_512tb option for "
6762 "filesystems greater than 512TB.\n");
6765 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
6766 if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
6767 CWARN("force_over_256tb option is deprecated. "
6768 "Filesystems less than 512TB can be created without any "
6769 "force options. Use force_over_512tb option for "
6770 "filesystems greater than 512TB.\n");
6774 if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
6775 force_over_512tb = 1;
6777 __page = alloc_page(GFP_KERNEL);
6779 GOTO(out, rc = -ENOMEM);
6780 page = (unsigned long)page_address(__page);
6781 options = (char *)page;
6784 /* strip out the options for back compatiblity */
6785 static char *sout[] = {
6792 /* strip out option we processed in osd */
6793 "bigendian_extents",
6799 strcat(options, opts);
6800 for (rc = 0, str = options; sout[rc]; ) {
6801 char *op = strstr(str, sout[rc]);
6807 if (op == options || *(op - 1) == ',') {
6808 str = op + strlen(sout[rc]);
6809 if (*str == ',' || *str == '\0') {
6810 *str == ',' ? str++ : str;
6811 memmove(op, str, strlen(str) + 1);
6814 for (str = op; *str != ',' && *str != '\0'; str++)
6818 strncat(options, "user_xattr,acl", 14);
6821 /* Glom up mount options */
6822 if (*options != '\0')
6823 strcat(options, ",");
6824 strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
6826 type = get_fs_type("ldiskfs");
6828 CERROR("%s: cannot find ldiskfs module\n", name);
6829 GOTO(out, rc = -ENODEV);
6832 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
6833 module_put(type->owner);
6835 if (IS_ERR(o->od_mnt)) {
6836 rc = PTR_ERR(o->od_mnt);
6838 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
6842 if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
6843 osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
6844 force_over_512tb == 0) {
6845 CERROR("%s: device %s LDISKFS does not support filesystems "
6846 "greater than 512TB and can cause data corruption. "
6847 "Use \"force_over_512tb\" mount option to override.\n",
6849 GOTO(out, rc = -EINVAL);
6852 #ifdef HAVE_DEV_SET_RDONLY
6853 if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
6854 CERROR("%s: underlying device %s is marked as read-only. "
6855 "Setup failed\n", name, dev);
6856 GOTO(out_mnt, rc = -EROFS);
6860 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
6861 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
6862 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
6863 GOTO(out_mnt, rc = -EINVAL);
6866 #ifdef LDISKFS_MOUNT_DIRDATA
6867 if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
6868 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6869 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
6870 else if (!o->od_is_ost)
6871 CWARN("%s: device %s was upgraded from Lustre-1.x without "
6872 "enabling the dirdata feature. If you do not want to "
6873 "downgrade to Lustre-1.x again, you can enable it via "
6874 "'tune2fs -O dirdata device'\n", name, dev);
6876 inode = osd_sb(o)->s_root->d_inode;
6877 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
6878 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
6880 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
6884 if (lmd_flags & LMD_FLG_NOSCRUB)
6895 __free_page(__page);
6900 static struct lu_device *osd_device_fini(const struct lu_env *env,
6901 struct lu_device *d)
6903 struct osd_device *o = osd_dev(d);
6906 osd_shutdown(env, o);
6908 osd_scrub_cleanup(env, o);
6909 osd_obj_map_fini(o);
6915 static int osd_device_init0(const struct lu_env *env,
6916 struct osd_device *o,
6917 struct lustre_cfg *cfg)
6919 struct lu_device *l = osd2lu_dev(o);
6920 struct osd_thread_info *info;
6924 /* if the module was re-loaded, env can loose its keys */
6925 rc = lu_env_refill((struct lu_env *) env);
6928 info = osd_oti_get(env);
6931 l->ld_ops = &osd_lu_ops;
6932 o->od_dt_dev.dd_ops = &osd_dt_ops;
6934 spin_lock_init(&o->od_osfs_lock);
6935 mutex_init(&o->od_otable_mutex);
6936 INIT_LIST_HEAD(&o->od_orphan_list);
6938 o->od_read_cache = 1;
6939 o->od_writethrough_cache = 1;
6940 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
6942 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
6943 sizeof(o->od_svname));
6944 if (cplen >= sizeof(o->od_svname)) {
6949 o->od_index = -1; /* -1 means index is invalid */
6950 rc = server_name2index(o->od_svname, &o->od_index, NULL);
6951 if (rc == LDD_F_SV_TYPE_OST)
6954 o->od_full_scrub_ratio = OFSR_DEFAULT;
6955 o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
6956 rc = osd_mount(env, o, cfg);
6960 rc = osd_obj_map_init(env, o);
6964 rc = lu_site_init(&o->od_site, l);
6966 GOTO(out_compat, rc);
6967 o->od_site.ls_bottom_dev = l;
6969 rc = lu_site_init_finish(&o->od_site);
6973 INIT_LIST_HEAD(&o->od_ios_list);
6974 /* setup scrub, including OI files initialization */
6975 rc = osd_scrub_setup(env, o);
6979 rc = osd_procfs_init(o, o->od_svname);
6981 CERROR("%s: can't initialize procfs: rc = %d\n",
6983 GOTO(out_scrub, rc);
6986 LASSERT(l->ld_site->ls_linkage.next != NULL);
6987 LASSERT(l->ld_site->ls_linkage.prev != NULL);
6989 /* initialize quota slave instance */
6990 o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
6992 if (IS_ERR(o->od_quota_slave)) {
6993 rc = PTR_ERR(o->od_quota_slave);
6994 o->od_quota_slave = NULL;
6995 GOTO(out_procfs, rc);
7003 osd_scrub_cleanup(env, o);
7005 lu_site_fini(&o->od_site);
7007 osd_obj_map_fini(o);
7014 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7015 struct lu_device_type *t,
7016 struct lustre_cfg *cfg)
7018 struct osd_device *o;
7023 return ERR_PTR(-ENOMEM);
7025 rc = dt_device_init(&o->od_dt_dev, t);
7027 /* Because the ctx might be revived in dt_device_init,
7028 * refill the env here */
7029 lu_env_refill((struct lu_env *)env);
7030 rc = osd_device_init0(env, o, cfg);
7032 dt_device_fini(&o->od_dt_dev);
7035 if (unlikely(rc != 0))
7038 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7041 static struct lu_device *osd_device_free(const struct lu_env *env,
7042 struct lu_device *d)
7044 struct osd_device *o = osd_dev(d);
7047 /* XXX: make osd top device in order to release reference */
7048 d->ld_site->ls_top_dev = d;
7049 lu_site_purge(env, d->ld_site, -1);
7050 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7051 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7052 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7054 lu_site_fini(&o->od_site);
7055 dt_device_fini(&o->od_dt_dev);
7060 static int osd_process_config(const struct lu_env *env,
7061 struct lu_device *d, struct lustre_cfg *cfg)
7063 struct osd_device *o = osd_dev(d);
7067 switch (cfg->lcfg_command) {
7069 rc = osd_mount(env, o, cfg);
7072 lu_dev_del_linkage(d->ld_site, d);
7073 rc = osd_shutdown(env, o);
7076 LASSERT(&o->od_dt_dev);
7077 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
7078 cfg, &o->od_dt_dev);
7079 if (rc > 0 || rc == -ENOSYS)
7080 rc = class_process_proc_param(PARAM_OST,
7081 lprocfs_osd_obd_vars,
7082 cfg, &o->od_dt_dev);
7091 static int osd_recovery_complete(const struct lu_env *env,
7092 struct lu_device *d)
7094 struct osd_device *osd = osd_dev(d);
7098 if (osd->od_quota_slave == NULL)
7101 /* start qsd instance on recovery completion, this notifies the quota
7102 * slave code that we are about to process new requests now */
7103 rc = qsd_start(env, osd->od_quota_slave);
7108 * we use exports to track all osd users
7110 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7111 struct obd_device *obd, struct obd_uuid *cluuid,
7112 struct obd_connect_data *data, void *localdata)
7114 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7115 struct lustre_handle conn;
7119 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7121 rc = class_connect(&conn, obd, cluuid);
7125 *exp = class_conn2export(&conn);
7127 spin_lock(&osd->od_osfs_lock);
7129 spin_unlock(&osd->od_osfs_lock);
7135 * once last export (we don't count self-export) disappeared
7136 * osd can be released
7138 static int osd_obd_disconnect(struct obd_export *exp)
7140 struct obd_device *obd = exp->exp_obd;
7141 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7142 int rc, release = 0;
7145 /* Only disconnect the underlying layers on the final disconnect. */
7146 spin_lock(&osd->od_osfs_lock);
7148 if (osd->od_connects == 0)
7150 spin_unlock(&osd->od_osfs_lock);
7152 rc = class_disconnect(exp); /* bz 9811 */
7154 if (rc == 0 && release)
7155 class_manual_cleanup(obd);
7159 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
7160 struct lu_device *dev)
7162 struct osd_device *osd = osd_dev(dev);
7163 struct lr_server_data *lsd =
7164 &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
7168 if (osd->od_quota_slave != NULL) {
7169 /* set up quota slave objects */
7170 result = qsd_prepare(env, osd->od_quota_slave);
7175 if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
7176 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
7177 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
7178 osd->od_index_in_idif = 1;
7180 osd->od_index_in_idif = 0;
7181 result = osd_register_proc_index_in_idif(osd);
7186 osd->od_index_in_idif = 1;
7190 result = osd_fid_init(env, osd);
7195 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
7196 struct lu_fid *fid, struct md_op_data *op_data)
7198 struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
7200 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
7203 static const struct lu_object_operations osd_lu_obj_ops = {
7204 .loo_object_init = osd_object_init,
7205 .loo_object_delete = osd_object_delete,
7206 .loo_object_release = osd_object_release,
7207 .loo_object_free = osd_object_free,
7208 .loo_object_print = osd_object_print,
7209 .loo_object_invariant = osd_object_invariant
7212 const struct lu_device_operations osd_lu_ops = {
7213 .ldo_object_alloc = osd_object_alloc,
7214 .ldo_process_config = osd_process_config,
7215 .ldo_recovery_complete = osd_recovery_complete,
7216 .ldo_prepare = osd_prepare,
7219 static const struct lu_device_type_operations osd_device_type_ops = {
7220 .ldto_init = osd_type_init,
7221 .ldto_fini = osd_type_fini,
7223 .ldto_start = osd_type_start,
7224 .ldto_stop = osd_type_stop,
7226 .ldto_device_alloc = osd_device_alloc,
7227 .ldto_device_free = osd_device_free,
7229 .ldto_device_init = osd_device_init,
7230 .ldto_device_fini = osd_device_fini
7233 static struct lu_device_type osd_device_type = {
7234 .ldt_tags = LU_DEVICE_DT,
7235 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
7236 .ldt_ops = &osd_device_type_ops,
7237 .ldt_ctx_tags = LCT_LOCAL,
7240 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
7242 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7243 struct super_block *sb = osd_sb(osd);
7245 return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
7249 * lprocfs legacy support.
7251 static struct obd_ops osd_obd_device_ops = {
7252 .o_owner = THIS_MODULE,
7253 .o_connect = osd_obd_connect,
7254 .o_disconnect = osd_obd_disconnect,
7255 .o_fid_alloc = osd_fid_alloc,
7256 .o_health_check = osd_health_check,
7259 static int __init osd_init(void)
7263 LASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
7264 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
7265 /* please, try to keep osd_thread_info smaller than a page */
7266 CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
7271 rc = lu_kmem_init(ldiskfs_caches);
7275 rc = class_register_type(&osd_obd_device_ops, NULL, true,
7276 lprocfs_osd_module_vars,
7277 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
7279 lu_kmem_fini(ldiskfs_caches);
7283 static void __exit osd_exit(void)
7285 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
7286 lu_kmem_fini(ldiskfs_caches);
7289 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7290 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
7291 MODULE_VERSION(LUSTRE_VERSION_STRING);
7292 MODULE_LICENSE("GPL");
7294 module_init(osd_init);
7295 module_exit(osd_exit);