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 = cfs_time_current();
1473 * Set time when the handle started
1475 static void osd_th_started(struct osd_thandle *oth)
1477 oth->oth_started = cfs_time_current();
1481 * Helper function to convert time interval to microseconds packed in
1484 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
1488 cfs_duration_usec(cfs_time_sub(end, start), &val);
1489 return val.tv_sec * 1000000 + val.tv_usec;
1493 * Check whether the we deal with this handle for too long.
1495 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1496 cfs_time_t alloced, cfs_time_t started,
1499 cfs_time_t now = cfs_time_current();
1501 LASSERT(dev != NULL);
1503 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1504 interval_to_usec(alloced, started));
1505 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1506 interval_to_usec(started, closed));
1507 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1508 interval_to_usec(closed, now));
1510 if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
1511 CWARN("transaction handle %p was open for too long: "
1512 "now "CFS_TIME_T" ,"
1513 "alloced "CFS_TIME_T" ,"
1514 "started "CFS_TIME_T" ,"
1515 "closed "CFS_TIME_T"\n",
1516 oth, now, alloced, started, closed);
1517 libcfs_debug_dumpstack(NULL);
1521 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
1523 cfs_time_t __closed = cfs_time_current(); \
1524 cfs_time_t __alloced = oth->oth_alloced; \
1525 cfs_time_t __started = oth->oth_started; \
1528 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
1531 #else /* OSD_THANDLE_STATS */
1533 #define osd_th_alloced(h) do {} while(0)
1534 #define osd_th_started(h) do {} while(0)
1535 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
1537 #endif /* OSD_THANDLE_STATS */
1540 * Concurrency: doesn't access mutable data.
1542 static int osd_param_is_not_sane(const struct osd_device *dev,
1543 const struct thandle *th)
1545 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1547 return oh->ot_credits > osd_transaction_size(dev);
1551 * Concurrency: shouldn't matter.
1553 static void osd_trans_commit_cb(struct super_block *sb,
1554 struct ldiskfs_journal_cb_entry *jcb, int error)
1556 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1557 struct thandle *th = &oh->ot_super;
1558 struct lu_device *lud = &th->th_dev->dd_lu_dev;
1559 struct dt_txn_commit_cb *dcb, *tmp;
1561 LASSERT(oh->ot_handle == NULL);
1564 CERROR("transaction @0x%p commit error: %d\n", th, error);
1566 dt_txn_hook_commit(th);
1568 /* call per-transaction callbacks if any */
1569 list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1571 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1572 "commit callback entry: magic=%x name='%s'\n",
1573 dcb->dcb_magic, dcb->dcb_name);
1574 list_del_init(&dcb->dcb_linkage);
1575 dcb->dcb_func(NULL, th, dcb, error);
1578 lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1582 lu_context_exit(&th->th_ctx);
1583 lu_context_fini(&th->th_ctx);
1587 #ifndef HAVE_SB_START_WRITE
1588 # define sb_start_write(sb) do {} while (0)
1589 # define sb_end_write(sb) do {} while (0)
1592 static struct thandle *osd_trans_create(const struct lu_env *env,
1593 struct dt_device *d)
1595 struct osd_thread_info *oti = osd_oti_get(env);
1596 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1597 struct osd_thandle *oh;
1601 /* on pending IO in this thread should left from prev. request */
1602 LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1604 sb_start_write(osd_sb(osd_dt_dev(d)));
1606 OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
1608 oh->ot_quota_trans = &oti->oti_quota_trans;
1609 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1613 th->th_tags = LCT_TX_HANDLE;
1615 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1616 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1619 memset(oti->oti_declare_ops, 0,
1620 sizeof(oti->oti_declare_ops));
1621 memset(oti->oti_declare_ops_cred, 0,
1622 sizeof(oti->oti_declare_ops_cred));
1623 memset(oti->oti_declare_ops_used, 0,
1624 sizeof(oti->oti_declare_ops_used));
1626 sb_end_write(osd_sb(osd_dt_dev(d)));
1627 th = ERR_PTR(-ENOMEM);
1632 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1634 struct osd_thread_info *oti = osd_oti_get(env);
1635 struct osd_thandle *oh;
1637 oh = container_of0(th, struct osd_thandle, ot_super);
1638 LASSERT(oh != NULL);
1640 CWARN(" create: %u/%u/%u, destroy: %u/%u/%u\n",
1641 oti->oti_declare_ops[OSD_OT_CREATE],
1642 oti->oti_declare_ops_cred[OSD_OT_CREATE],
1643 oti->oti_declare_ops_used[OSD_OT_CREATE],
1644 oti->oti_declare_ops[OSD_OT_DESTROY],
1645 oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1646 oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1647 CWARN(" attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1648 oti->oti_declare_ops[OSD_OT_ATTR_SET],
1649 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1650 oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1651 oti->oti_declare_ops[OSD_OT_XATTR_SET],
1652 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1653 oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1654 CWARN(" write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1655 oti->oti_declare_ops[OSD_OT_WRITE],
1656 oti->oti_declare_ops_cred[OSD_OT_WRITE],
1657 oti->oti_declare_ops_used[OSD_OT_WRITE],
1658 oti->oti_declare_ops[OSD_OT_PUNCH],
1659 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1660 oti->oti_declare_ops_used[OSD_OT_PUNCH],
1661 oti->oti_declare_ops[OSD_OT_QUOTA],
1662 oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1663 oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1664 CWARN(" insert: %u/%u/%u, delete: %u/%u/%u\n",
1665 oti->oti_declare_ops[OSD_OT_INSERT],
1666 oti->oti_declare_ops_cred[OSD_OT_INSERT],
1667 oti->oti_declare_ops_used[OSD_OT_INSERT],
1668 oti->oti_declare_ops[OSD_OT_DELETE],
1669 oti->oti_declare_ops_cred[OSD_OT_DELETE],
1670 oti->oti_declare_ops_used[OSD_OT_DELETE]);
1671 CWARN(" ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1672 oti->oti_declare_ops[OSD_OT_REF_ADD],
1673 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1674 oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1675 oti->oti_declare_ops[OSD_OT_REF_DEL],
1676 oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1677 oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1681 * Concurrency: shouldn't matter.
1683 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1686 struct osd_thread_info *oti = osd_oti_get(env);
1687 struct osd_device *dev = osd_dt_dev(d);
1689 struct osd_thandle *oh;
1694 LASSERT(current->journal_info == NULL);
1696 oh = container_of0(th, struct osd_thandle, ot_super);
1697 LASSERT(oh != NULL);
1698 LASSERT(oh->ot_handle == NULL);
1700 rc = dt_txn_hook_start(env, d, th);
1704 if (unlikely(osd_param_is_not_sane(dev, th))) {
1705 static unsigned long last_printed;
1706 static int last_credits;
1708 /* don't make noise on a tiny testing systems
1709 * actual credits misuse will be caught anyway */
1710 if (last_credits != oh->ot_credits &&
1711 time_after(jiffies, last_printed +
1712 msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1713 osd_transaction_size(dev) > 512) {
1714 CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1715 oh->ot_credits, osd_transaction_size(dev));
1716 osd_trans_dump_creds(env, th);
1717 libcfs_debug_dumpstack(NULL);
1718 last_credits = oh->ot_credits;
1719 last_printed = jiffies;
1721 /* XXX Limit the credits to 'max_transaction_buffers', and
1722 * let the underlying filesystem to catch the error if
1723 * we really need so many credits.
1725 * This should be removed when we can calculate the
1726 * credits precisely. */
1727 oh->ot_credits = osd_transaction_size(dev);
1731 * XXX temporary stuff. Some abstraction layer should
1734 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1738 LASSERT(oti->oti_txns == 0);
1739 lu_context_init(&th->th_ctx, th->th_tags);
1740 lu_context_enter(&th->th_ctx);
1742 lu_device_get(&d->dd_lu_dev);
1743 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1754 static int osd_seq_exists(const struct lu_env *env,
1755 struct osd_device *osd, u64 seq)
1757 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1758 struct seq_server_site *ss = osd_seq_site(osd);
1762 LASSERT(ss != NULL);
1763 LASSERT(ss->ss_server_fld != NULL);
1765 rc = osd_fld_lookup(env, osd, seq, range);
1768 CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1769 osd_name(osd), seq, rc);
1773 RETURN(ss->ss_node_id == range->lsr_index);
1776 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1778 struct dt_txn_commit_cb *dcb;
1779 struct dt_txn_commit_cb *tmp;
1781 /* call per-transaction stop callbacks if any */
1782 list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1784 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1785 "commit callback entry: magic=%x name='%s'\n",
1786 dcb->dcb_magic, dcb->dcb_name);
1787 list_del_init(&dcb->dcb_linkage);
1788 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1793 * Concurrency: shouldn't matter.
1795 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1798 int rc = 0, remove_agents = 0;
1799 struct osd_thandle *oh;
1800 struct osd_thread_info *oti = osd_oti_get(env);
1801 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1802 struct osd_device *osd = osd_dt_dev(th->th_dev);
1803 struct qsd_instance *qsd = osd->od_quota_slave;
1804 struct lquota_trans *qtrans;
1807 oh = container_of0(th, struct osd_thandle, ot_super);
1809 /* reset OI cache for safety */
1810 oti->oti_ins_cache_used = 0;
1812 remove_agents = oh->ot_remove_agents;
1814 qtrans = oh->ot_quota_trans;
1815 oh->ot_quota_trans = NULL;
1817 if (oh->ot_handle != NULL) {
1819 handle_t *hdl = oh->ot_handle;
1822 * add commit callback
1823 * notice we don't do this in osd_trans_start()
1824 * as underlying transaction can change during truncate
1826 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1829 LASSERT(oti->oti_txns == 1);
1832 rc = dt_txn_hook_stop(env, th);
1834 CERROR("%s: failed in transaction hook: rc = %d\n",
1837 osd_trans_stop_cb(oh, rc);
1838 /* hook functions might modify th_sync */
1839 hdl->h_sync = th->th_sync;
1841 oh->ot_handle = NULL;
1842 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
1844 CERROR("%s: failed to stop transaction: rc = %d\n",
1845 osd_name(osd), rc2);
1849 osd_trans_stop_cb(oh, th->th_result);
1853 /* inform the quota slave device that the transaction is stopping */
1854 qsd_op_end(env, qsd, qtrans);
1856 /* as we want IO to journal and data IO be concurrent, we don't block
1857 * awaiting data IO completion in osd_do_bio(), instead we wait here
1858 * once transaction is submitted to the journal. all reqular requests
1859 * don't do direct IO (except read/write), thus this wait_event becomes
1862 * IMPORTANT: we have to wait till any IO submited by the thread is
1863 * completed otherwise iobuf may be corrupted by different request
1865 wait_event(iobuf->dr_wait,
1866 atomic_read(&iobuf->dr_numreqs) == 0);
1867 osd_fini_iobuf(osd, iobuf);
1869 rc = iobuf->dr_error;
1871 if (unlikely(remove_agents != 0))
1872 osd_process_scheduled_agent_removals(env, osd);
1874 sb_end_write(osd_sb(osd));
1879 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1881 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1884 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1885 LASSERT(&dcb->dcb_func != NULL);
1886 if (dcb->dcb_flags & DCB_TRANS_STOP)
1887 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1889 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1895 * Called just before object is freed. Releases all resources except for
1896 * object itself (that is released by osd_object_free()).
1898 * Concurrency: no concurrent access is possible that late in object
1901 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1903 struct osd_object *obj = osd_obj(l);
1904 struct inode *inode = obj->oo_inode;
1906 LINVRNT(osd_invariant(obj));
1909 * If object is unlinked remove fid->ino mapping from object index.
1912 osd_index_fini(obj);
1913 if (inode != NULL) {
1914 struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave;
1915 qid_t uid = i_uid_read(inode);
1916 qid_t gid = i_gid_read(inode);
1919 obj->oo_inode = NULL;
1922 struct osd_thread_info *info = osd_oti_get(env);
1923 struct lquota_id_info *qi = &info->oti_qi;
1925 /* Release granted quota to master if necessary */
1926 qi->lqi_id.qid_uid = uid;
1927 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1929 qi->lqi_id.qid_uid = gid;
1930 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1936 * Concurrency: ->loo_object_release() is called under site spin-lock.
1938 static void osd_object_release(const struct lu_env *env,
1939 struct lu_object *l)
1941 struct osd_object *o = osd_obj(l);
1942 /* nobody should be releasing a non-destroyed object with nlink=0
1943 * the API allows this, but ldiskfs doesn't like and then report
1944 * this inode as deleted */
1945 if (unlikely(!o->oo_destroyed && o->oo_inode && o->oo_inode->i_nlink == 0))
1950 * Concurrency: shouldn't matter.
1952 static int osd_object_print(const struct lu_env *env, void *cookie,
1953 lu_printer_t p, const struct lu_object *l)
1955 struct osd_object *o = osd_obj(l);
1956 struct iam_descr *d;
1958 if (o->oo_dir != NULL)
1959 d = o->oo_dir->od_container.ic_descr;
1962 return (*p)(env, cookie,
1963 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1965 o->oo_inode ? o->oo_inode->i_ino : 0UL,
1966 o->oo_inode ? o->oo_inode->i_generation : 0,
1967 d ? d->id_ops->id_name : "plain");
1971 * Concurrency: shouldn't matter.
1973 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1974 struct obd_statfs *sfs)
1976 struct osd_device *osd = osd_dt_dev(d);
1977 struct super_block *sb = osd_sb(osd);
1978 struct kstatfs *ksfs;
1982 if (unlikely(osd->od_mnt == NULL))
1983 return -EINPROGRESS;
1985 /* osd_lproc.c call this without env, allocate ksfs for that case */
1986 if (unlikely(env == NULL)) {
1987 OBD_ALLOC_PTR(ksfs);
1991 ksfs = &osd_oti_get(env)->oti_ksfs;
1994 result = sb->s_op->statfs(sb->s_root, ksfs);
1998 statfs_pack(sfs, ksfs);
1999 if (unlikely(sb->s_flags & MS_RDONLY))
2000 sfs->os_state |= OS_STATE_READONLY;
2001 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb,
2002 LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2003 sfs->os_maxbytes = sb->s_maxbytes;
2005 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2008 * Reserve some space so to avoid fragmenting the filesystem too much.
2009 * Fragmentation not only impacts performance, but can also increase
2010 * metadata overhead significantly, causing grant calculation to be
2013 * Reserve 0.78% of total space, at least 8MB for small filesystems.
2015 CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2016 reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2017 if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2018 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2020 sfs->os_blocks -= reserved;
2021 sfs->os_bfree -= min(reserved, sfs->os_bfree);
2022 sfs->os_bavail -= min(reserved, sfs->os_bavail);
2025 if (unlikely(env == NULL))
2031 * Estimate space needed for file creations. We assume the largest filename
2032 * which is 2^64 - 1, hence a filename of 20 chars.
2033 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2035 #ifdef __LDISKFS_DIR_REC_LEN
2036 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2038 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2042 * Concurrency: doesn't access mutable data.
2044 static void osd_conf_get(const struct lu_env *env,
2045 const struct dt_device *dev,
2046 struct dt_device_param *param)
2048 struct super_block *sb = osd_sb(osd_dt_dev(dev));
2052 * XXX should be taken from not-yet-existing fs abstraction layer.
2054 param->ddp_max_name_len = LDISKFS_NAME_LEN;
2055 param->ddp_max_nlink = LDISKFS_LINK_MAX;
2056 param->ddp_symlink_max = sb->s_blocksize;
2057 param->ddp_mount_type = LDD_MT_LDISKFS;
2058 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2059 param->ddp_maxbytes = sb->s_maxbytes;
2061 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2062 /* inode are statically allocated, so per-inode space consumption
2063 * is the space consumed by the directory entry */
2064 param->ddp_inodespace = PER_OBJ_USAGE;
2065 /* EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2066 * = 128MB) which is unlikely to be hit in real life. Report a smaller
2067 * maximum length to not under count the actual number of extents
2068 * needed for writing a file. */
2069 param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2;
2070 /* worst-case extent insertion metadata overhead */
2071 param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2072 param->ddp_mntopts = 0;
2073 if (test_opt(sb, XATTR_USER))
2074 param->ddp_mntopts |= MNTOPT_USERXATTR;
2075 if (test_opt(sb, POSIX_ACL))
2076 param->ddp_mntopts |= MNTOPT_ACL;
2078 /* LOD might calculate the max stripe count based on max_ea_size,
2079 * so we need take account in the overhead as well,
2080 * xattr_header + magic + xattr_entry_head */
2081 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2082 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2084 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2085 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
2086 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2090 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2094 * Concurrency: shouldn't matter.
2096 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2100 CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
2102 rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
2104 CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
2105 LUSTRE_OSD_LDISKFS_NAME, rc);
2111 * Start commit for OSD device.
2113 * An implementation of dt_commit_async method for OSD device.
2114 * Asychronously starts underlayng fs sync and thereby a transaction
2117 * \param env environment
2118 * \param d dt device
2120 * \see dt_device_operations
2122 static int osd_commit_async(const struct lu_env *env,
2123 struct dt_device *d)
2125 struct super_block *s = osd_sb(osd_dt_dev(d));
2128 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
2129 RETURN(s->s_op->sync_fs(s, 0));
2133 * Concurrency: shouldn't matter.
2136 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2138 struct super_block *sb = osd_sb(osd_dt_dev(d));
2139 struct block_device *dev = sb->s_bdev;
2140 #ifdef HAVE_DEV_SET_RDONLY
2141 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2144 int rc = -EOPNOTSUPP;
2148 #ifdef HAVE_DEV_SET_RDONLY
2149 CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
2151 if (sb->s_op->freeze_fs) {
2152 rc = sb->s_op->freeze_fs(sb);
2157 if (jdev && (jdev != dev)) {
2158 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2160 dev_set_rdonly(jdev);
2162 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2163 dev_set_rdonly(dev);
2165 if (sb->s_op->unfreeze_fs)
2166 sb->s_op->unfreeze_fs(sb);
2171 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2172 osd_dt_dev(d)->od_svname, (long)dev, rc);
2178 * Note: we do not count into QUOTA here.
2179 * If we mount with --data_journal we may need more.
2181 const int osd_dto_credits_noquota[DTO_NR] = {
2184 * INDEX_EXTRA_TRANS_BLOCKS(8) +
2185 * SINGLEDATA_TRANS_BLOCKS(8)
2186 * XXX Note: maybe iam need more, since iam have more level than
2189 [DTO_INDEX_INSERT] = 16,
2192 * just modify a single entry, probably merge few within a block
2194 [DTO_INDEX_DELETE] = 1,
2198 [DTO_INDEX_UPDATE] = 16,
2200 * 4(inode, inode bits, groups, GDT)
2201 * notice: OI updates are counted separately with DTO_INDEX_INSERT
2203 [DTO_OBJECT_CREATE] = 4,
2205 * 4(inode, inode bits, groups, GDT)
2206 * notice: OI updates are counted separately with DTO_INDEX_DELETE
2208 [DTO_OBJECT_DELETE] = 4,
2210 * Attr set credits (inode)
2212 [DTO_ATTR_SET_BASE] = 1,
2214 * Xattr set. The same as xattr of EXT3.
2215 * DATA_TRANS_BLOCKS(14)
2216 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2217 * are also counted in. Do not know why?
2219 [DTO_XATTR_SET] = 14,
2221 * credits for inode change during write.
2223 [DTO_WRITE_BASE] = 3,
2225 * credits for single block write.
2227 [DTO_WRITE_BLOCK] = 14,
2229 * Attr set credits for chown.
2230 * This is extra credits for setattr, and it is null without quota
2232 [DTO_ATTR_SET_CHOWN] = 0
2235 static const struct dt_device_operations osd_dt_ops = {
2236 .dt_root_get = osd_root_get,
2237 .dt_statfs = osd_statfs,
2238 .dt_trans_create = osd_trans_create,
2239 .dt_trans_start = osd_trans_start,
2240 .dt_trans_stop = osd_trans_stop,
2241 .dt_trans_cb_add = osd_trans_cb_add,
2242 .dt_conf_get = osd_conf_get,
2243 .dt_sync = osd_sync,
2245 .dt_commit_async = osd_commit_async,
2248 static void osd_object_read_lock(const struct lu_env *env,
2249 struct dt_object *dt, unsigned role)
2251 struct osd_object *obj = osd_dt_obj(dt);
2252 struct osd_thread_info *oti = osd_oti_get(env);
2254 LINVRNT(osd_invariant(obj));
2256 LASSERT(obj->oo_owner != env);
2257 down_read_nested(&obj->oo_sem, role);
2259 LASSERT(obj->oo_owner == NULL);
2263 static void osd_object_write_lock(const struct lu_env *env,
2264 struct dt_object *dt, unsigned role)
2266 struct osd_object *obj = osd_dt_obj(dt);
2267 struct osd_thread_info *oti = osd_oti_get(env);
2269 LINVRNT(osd_invariant(obj));
2271 LASSERT(obj->oo_owner != env);
2272 down_write_nested(&obj->oo_sem, role);
2274 LASSERT(obj->oo_owner == NULL);
2275 obj->oo_owner = env;
2279 static void osd_object_read_unlock(const struct lu_env *env,
2280 struct dt_object *dt)
2282 struct osd_object *obj = osd_dt_obj(dt);
2283 struct osd_thread_info *oti = osd_oti_get(env);
2285 LINVRNT(osd_invariant(obj));
2287 LASSERT(oti->oti_r_locks > 0);
2289 up_read(&obj->oo_sem);
2292 static void osd_object_write_unlock(const struct lu_env *env,
2293 struct dt_object *dt)
2295 struct osd_object *obj = osd_dt_obj(dt);
2296 struct osd_thread_info *oti = osd_oti_get(env);
2298 LINVRNT(osd_invariant(obj));
2300 LASSERT(obj->oo_owner == env);
2301 LASSERT(oti->oti_w_locks > 0);
2303 obj->oo_owner = NULL;
2304 up_write(&obj->oo_sem);
2307 static int osd_object_write_locked(const struct lu_env *env,
2308 struct dt_object *dt)
2310 struct osd_object *obj = osd_dt_obj(dt);
2312 LINVRNT(osd_invariant(obj));
2314 return obj->oo_owner == env;
2317 static struct timespec *osd_inode_time(const struct lu_env *env,
2318 struct inode *inode, __u64 seconds)
2320 struct osd_thread_info *oti = osd_oti_get(env);
2321 struct timespec *t = &oti->oti_time;
2323 t->tv_sec = seconds;
2325 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2329 static void osd_inode_getattr(const struct lu_env *env,
2330 struct inode *inode, struct lu_attr *attr)
2332 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2333 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2334 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
2337 attr->la_atime = LTIME_S(inode->i_atime);
2338 attr->la_mtime = LTIME_S(inode->i_mtime);
2339 attr->la_ctime = LTIME_S(inode->i_ctime);
2340 attr->la_mode = inode->i_mode;
2341 attr->la_size = i_size_read(inode);
2342 attr->la_blocks = inode->i_blocks;
2343 attr->la_uid = i_uid_read(inode);
2344 attr->la_gid = i_gid_read(inode);
2345 attr->la_flags = ll_inode_to_ext_flags(inode->i_flags);
2346 attr->la_nlink = inode->i_nlink;
2347 attr->la_rdev = inode->i_rdev;
2348 attr->la_blksize = 1 << inode->i_blkbits;
2349 attr->la_blkbits = inode->i_blkbits;
2352 static int osd_attr_get(const struct lu_env *env,
2353 struct dt_object *dt,
2354 struct lu_attr *attr)
2356 struct osd_object *obj = osd_dt_obj(dt);
2358 if (unlikely(!dt_object_exists(dt)))
2360 if (unlikely(obj->oo_destroyed))
2363 LASSERT(!dt_object_remote(dt));
2364 LINVRNT(osd_invariant(obj));
2366 spin_lock(&obj->oo_guard);
2367 osd_inode_getattr(env, obj->oo_inode, attr);
2368 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2369 attr->la_flags |= LUSTRE_ORPHAN_FL;
2370 spin_unlock(&obj->oo_guard);
2375 static int osd_declare_attr_set(const struct lu_env *env,
2376 struct dt_object *dt,
2377 const struct lu_attr *attr,
2378 struct thandle *handle)
2380 struct osd_thandle *oh;
2381 struct osd_object *obj;
2382 struct osd_thread_info *info = osd_oti_get(env);
2383 struct lquota_id_info *qi = &info->oti_qi;
2391 LASSERT(dt != NULL);
2392 LASSERT(handle != NULL);
2394 obj = osd_dt_obj(dt);
2395 LASSERT(osd_invariant(obj));
2397 oh = container_of0(handle, struct osd_thandle, ot_super);
2398 LASSERT(oh->ot_handle == NULL);
2400 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2401 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2403 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2404 osd_dto_credits_noquota[DTO_XATTR_SET]);
2406 if (attr == NULL || obj->oo_inode == NULL)
2409 bspace = obj->oo_inode->i_blocks;
2410 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
2411 bspace = toqb(bspace);
2413 /* Changing ownership is always preformed by super user, it should not
2416 * We still need to call the osd_declare_qid() to calculate the journal
2417 * credits for updating quota accounting files and to trigger quota
2418 * space adjustment once the operation is completed.*/
2419 if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2421 uid = i_uid_read(obj->oo_inode);
2422 qi->lqi_type = USRQUOTA;
2423 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2424 /* inode accounting */
2425 qi->lqi_is_blk = false;
2427 /* one more inode for the new uid ... */
2428 qi->lqi_id.qid_uid = attr->la_uid;
2430 /* Reserve credits for the new uid */
2431 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2432 if (rc == -EDQUOT || rc == -EINPROGRESS)
2437 /* and one less inode for the current uid */
2438 qi->lqi_id.qid_uid = uid;
2440 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2441 if (rc == -EDQUOT || rc == -EINPROGRESS)
2446 /* block accounting */
2447 qi->lqi_is_blk = true;
2449 /* more blocks for the new uid ... */
2450 qi->lqi_id.qid_uid = attr->la_uid;
2451 qi->lqi_space = bspace;
2453 * Credits for the new uid has been reserved, re-use "obj"
2454 * to save credit reservation.
2456 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2457 if (rc == -EDQUOT || rc == -EINPROGRESS)
2462 /* and finally less blocks for the current uid */
2463 qi->lqi_id.qid_uid = uid;
2464 qi->lqi_space = -bspace;
2465 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2466 if (rc == -EDQUOT || rc == -EINPROGRESS)
2472 gid = i_gid_read(obj->oo_inode);
2473 qi->lqi_type = GRPQUOTA;
2474 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2476 /* inode accounting */
2477 qi->lqi_is_blk = false;
2479 /* one more inode for the new gid ... */
2480 qi->lqi_id.qid_gid = attr->la_gid;
2482 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2483 if (rc == -EDQUOT || rc == -EINPROGRESS)
2488 /* and one less inode for the current gid */
2489 qi->lqi_id.qid_gid = gid;
2491 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2492 if (rc == -EDQUOT || rc == -EINPROGRESS)
2497 /* block accounting */
2498 qi->lqi_is_blk = true;
2500 /* more blocks for the new gid ... */
2501 qi->lqi_id.qid_gid = attr->la_gid;
2502 qi->lqi_space = bspace;
2503 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2504 if (rc == -EDQUOT || rc == -EINPROGRESS)
2509 /* and finally less blocks for the current gid */
2510 qi->lqi_id.qid_gid = gid;
2511 qi->lqi_space = -bspace;
2512 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2513 if (rc == -EDQUOT || rc == -EINPROGRESS)
2522 static int osd_inode_setattr(const struct lu_env *env,
2523 struct inode *inode, const struct lu_attr *attr)
2525 __u64 bits = attr->la_valid;
2527 /* Only allow set size for regular file */
2528 if (!S_ISREG(inode->i_mode))
2529 bits &= ~(LA_SIZE | LA_BLOCKS);
2534 if (bits & LA_ATIME)
2535 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
2536 if (bits & LA_CTIME)
2537 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
2538 if (bits & LA_MTIME)
2539 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
2540 if (bits & LA_SIZE) {
2541 spin_lock(&inode->i_lock);
2542 LDISKFS_I(inode)->i_disksize = attr->la_size;
2543 i_size_write(inode, attr->la_size);
2544 spin_unlock(&inode->i_lock);
2547 /* OSD should not change "i_blocks" which is used by quota.
2548 * "i_blocks" should be changed by ldiskfs only. */
2550 inode->i_mode = (inode->i_mode & S_IFMT) |
2551 (attr->la_mode & ~S_IFMT);
2553 i_uid_write(inode, attr->la_uid);
2555 i_gid_write(inode, attr->la_gid);
2556 if (bits & LA_NLINK)
2557 set_nlink(inode, attr->la_nlink);
2559 inode->i_rdev = attr->la_rdev;
2561 if (bits & LA_FLAGS) {
2562 /* always keep S_NOCMTIME */
2563 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2569 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2571 if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2572 (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2576 ll_vfs_dq_init(inode);
2578 if (attr->la_valid & LA_UID)
2579 iattr.ia_valid |= ATTR_UID;
2580 if (attr->la_valid & LA_GID)
2581 iattr.ia_valid |= ATTR_GID;
2582 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2583 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2585 rc = ll_vfs_dq_transfer(inode, &iattr);
2587 CERROR("%s: quota transfer failed: rc = %d. Is quota "
2588 "enforcement enabled on the ldiskfs "
2589 "filesystem?\n", inode->i_sb->s_id, rc);
2596 static int osd_attr_set(const struct lu_env *env,
2597 struct dt_object *dt,
2598 const struct lu_attr *attr,
2599 struct thandle *handle)
2601 struct osd_object *obj = osd_dt_obj(dt);
2602 struct inode *inode;
2605 if (!dt_object_exists(dt))
2608 LASSERT(handle != NULL);
2609 LASSERT(!dt_object_remote(dt));
2610 LASSERT(osd_invariant(obj));
2612 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2614 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
2615 struct osd_thread_info *oti = osd_oti_get(env);
2616 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
2617 struct lu_fid *fid1 = &oti->oti_fid;
2618 struct osd_inode_id *id = &oti->oti_id;
2619 struct iam_path_descr *ipd;
2620 struct iam_container *bag;
2621 struct osd_thandle *oh;
2624 fid_cpu_to_be(fid1, fid0);
2625 memset(id, 1, sizeof(*id));
2626 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2627 fid0)->oi_dir.od_container;
2628 ipd = osd_idx_ipd_get(env, bag);
2629 if (unlikely(ipd == NULL))
2632 oh = container_of0(handle, struct osd_thandle, ot_super);
2633 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2634 (const struct iam_rec *)id, ipd);
2635 osd_ipd_put(env, bag, ipd);
2636 return(rc > 0 ? 0 : rc);
2639 inode = obj->oo_inode;
2641 rc = osd_quota_transfer(inode, attr);
2645 spin_lock(&obj->oo_guard);
2646 rc = osd_inode_setattr(env, inode, attr);
2647 spin_unlock(&obj->oo_guard);
2651 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2653 if (!(attr->la_valid & LA_FLAGS))
2656 /* Let's check if there are extra flags need to be set into LMA */
2657 if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
2658 struct osd_thread_info *info = osd_oti_get(env);
2659 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2661 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
2665 lma->lma_incompat |=
2666 lustre_to_lma_flags(attr->la_flags);
2667 lustre_lma_swab(lma);
2668 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
2669 lma, sizeof(*lma), XATTR_REPLACE);
2671 struct osd_device *osd = osd_obj2dev(obj);
2673 CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
2674 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
2675 lma->lma_incompat, rc);
2678 attr->la_flags & LUSTRE_LMA_FL_MASKS;
2680 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2683 osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2688 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2689 struct osd_object *obj,
2690 const char *name, const int namelen)
2692 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2695 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2696 umode_t mode, struct dt_allocation_hint *hint,
2700 struct osd_device *osd = osd_obj2dev(obj);
2701 struct osd_thandle *oth;
2702 struct dt_object *parent = NULL;
2703 struct inode *inode;
2705 LINVRNT(osd_invariant(obj));
2706 LASSERT(obj->oo_inode == NULL);
2707 LASSERT(obj->oo_hl_head == NULL);
2709 if (S_ISDIR(mode) && ldiskfs_pdo) {
2710 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2711 if (obj->oo_hl_head == NULL)
2715 oth = container_of(th, struct osd_thandle, ot_super);
2716 LASSERT(oth->ot_handle->h_transaction != NULL);
2718 if (hint != NULL && hint->dah_parent != NULL &&
2719 !dt_object_remote(hint->dah_parent))
2720 parent = hint->dah_parent;
2722 inode = ldiskfs_create_inode(oth->ot_handle,
2723 parent ? osd_dt_obj(parent)->oo_inode :
2724 osd_sb(osd)->s_root->d_inode,
2726 if (!IS_ERR(inode)) {
2727 /* Do not update file c/mtime in ldiskfs. */
2728 inode->i_flags |= S_NOCMTIME;
2730 /* For new created object, it must be consistent,
2731 * and it is unnecessary to scrub against it. */
2732 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2734 obj->oo_inode = inode;
2737 if (obj->oo_hl_head != NULL) {
2738 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2739 obj->oo_hl_head = NULL;
2741 result = PTR_ERR(inode);
2743 LINVRNT(osd_invariant(obj));
2751 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2752 struct lu_attr *attr,
2753 struct dt_allocation_hint *hint,
2754 struct dt_object_format *dof,
2758 struct osd_thandle *oth;
2759 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
2761 LASSERT(S_ISDIR(attr->la_mode));
2763 oth = container_of(th, struct osd_thandle, ot_super);
2764 LASSERT(oth->ot_handle->h_transaction != NULL);
2765 result = osd_mkfile(info, obj, mode, hint, th);
2770 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2771 struct lu_attr *attr,
2772 struct dt_allocation_hint *hint,
2773 struct dt_object_format *dof,
2777 struct osd_thandle *oth;
2778 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2780 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2782 LASSERT(S_ISREG(attr->la_mode));
2784 oth = container_of(th, struct osd_thandle, ot_super);
2785 LASSERT(oth->ot_handle->h_transaction != NULL);
2787 result = osd_mkfile(info, obj, mode, hint, th);
2789 LASSERT(obj->oo_inode != NULL);
2790 if (feat->dif_flags & DT_IND_VARKEY)
2791 result = iam_lvar_create(obj->oo_inode,
2792 feat->dif_keysize_max,
2794 feat->dif_recsize_max,
2797 result = iam_lfix_create(obj->oo_inode,
2798 feat->dif_keysize_max,
2800 feat->dif_recsize_max,
2807 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2808 struct lu_attr *attr,
2809 struct dt_allocation_hint *hint,
2810 struct dt_object_format *dof,
2813 LASSERT(S_ISREG(attr->la_mode));
2814 return osd_mkfile(info, obj, (attr->la_mode &
2815 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2818 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2819 struct lu_attr *attr,
2820 struct dt_allocation_hint *hint,
2821 struct dt_object_format *dof,
2824 LASSERT(S_ISLNK(attr->la_mode));
2825 return osd_mkfile(info, obj, (attr->la_mode &
2826 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2829 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2830 struct lu_attr *attr,
2831 struct dt_allocation_hint *hint,
2832 struct dt_object_format *dof,
2835 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2838 LINVRNT(osd_invariant(obj));
2839 LASSERT(obj->oo_inode == NULL);
2840 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2841 S_ISFIFO(mode) || S_ISSOCK(mode));
2843 result = osd_mkfile(info, obj, mode, hint, th);
2845 LASSERT(obj->oo_inode != NULL);
2847 * This inode should be marked dirty for i_rdev. Currently
2848 * that is done in the osd_attr_init().
2850 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2853 LINVRNT(osd_invariant(obj));
2857 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2859 struct dt_allocation_hint *hint,
2860 struct dt_object_format *dof,
2863 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2865 osd_obj_type_f result;
2881 result = osd_mk_index;
2892 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2893 struct dt_object *parent, struct dt_object *child,
2898 ah->dah_parent = parent;
2899 ah->dah_mode = child_mode;
2901 if (parent != NULL && !dt_object_remote(parent)) {
2902 /* will help to find FID->ino at dt_insert("..") */
2903 struct osd_object *pobj = osd_dt_obj(parent);
2904 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
2908 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2909 struct lu_attr *attr, struct dt_object_format *dof)
2911 struct inode *inode = obj->oo_inode;
2912 __u64 valid = attr->la_valid;
2915 attr->la_valid &= ~(LA_TYPE | LA_MODE);
2917 if (dof->dof_type != DFT_NODE)
2918 attr->la_valid &= ~LA_RDEV;
2919 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2920 attr->la_valid &= ~LA_ATIME;
2921 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2922 attr->la_valid &= ~LA_CTIME;
2923 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2924 attr->la_valid &= ~LA_MTIME;
2926 result = osd_quota_transfer(inode, attr);
2930 if (attr->la_valid != 0) {
2931 result = osd_inode_setattr(info->oti_env, inode, attr);
2933 * The osd_inode_setattr() should always succeed here. The
2934 * only error that could be returned is EDQUOT when we are
2935 * trying to change the UID or GID of the inode. However, this
2936 * should not happen since quota enforcement is no longer
2937 * enabled on ldiskfs (lquota takes care of it).
2939 LASSERTF(result == 0, "%d\n", result);
2940 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2943 attr->la_valid = valid;
2947 * Helper function for osd_object_create()
2949 * \retval 0, on success
2951 static int __osd_object_create(struct osd_thread_info *info,
2952 struct osd_object *obj, struct lu_attr *attr,
2953 struct dt_allocation_hint *hint,
2954 struct dt_object_format *dof,
2960 osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
2962 /* we drop umask so that permissions we pass are not affected */
2963 umask = current->fs->umask;
2964 current->fs->umask = 0;
2966 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2968 if (likely(obj->oo_inode != NULL)) {
2969 LASSERT(obj->oo_inode->i_state & I_NEW);
2971 /* Unlock the inode before attr initialization to avoid
2972 * unnecessary dqget operations. LU-6378 */
2973 unlock_new_inode(obj->oo_inode);
2976 if (likely(result == 0)) {
2977 osd_attr_init(info, obj, attr, dof);
2978 osd_object_init0(obj);
2981 /* restore previous umask value */
2982 current->fs->umask = umask;
2984 osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
2990 * Helper function for osd_object_create()
2992 * \retval 0, on success
2994 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2995 const struct lu_fid *fid, struct thandle *th)
2997 struct osd_thread_info *info = osd_oti_get(env);
2998 struct osd_inode_id *id = &info->oti_id;
2999 struct osd_device *osd = osd_obj2dev(obj);
3000 struct osd_thandle *oh;
3003 LASSERT(obj->oo_inode != NULL);
3005 oh = container_of0(th, struct osd_thandle, ot_super);
3006 LASSERT(oh->ot_handle);
3007 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3009 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3010 rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3011 OI_CHECK_FLD, NULL);
3012 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3017 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3018 u64 seq, struct lu_seq_range *range)
3020 struct seq_server_site *ss = osd_seq_site(osd);
3022 if (fid_seq_is_idif(seq)) {
3023 fld_range_set_ost(range);
3024 range->lsr_index = idif_ost_idx(seq);
3028 if (!fid_seq_in_fldb(seq)) {
3029 fld_range_set_mdt(range);
3031 /* FIXME: If ss is NULL, it suppose not get lsr_index
3033 range->lsr_index = ss->ss_node_id;
3037 LASSERT(ss != NULL);
3038 fld_range_set_any(range);
3039 /* OSD will only do local fld lookup */
3040 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3044 * Concurrency: no external locking is necessary.
3046 static int osd_declare_object_create(const struct lu_env *env,
3047 struct dt_object *dt,
3048 struct lu_attr *attr,
3049 struct dt_allocation_hint *hint,
3050 struct dt_object_format *dof,
3051 struct thandle *handle)
3053 struct osd_thandle *oh;
3057 LASSERT(handle != NULL);
3059 oh = container_of0(handle, struct osd_thandle, ot_super);
3060 LASSERT(oh->ot_handle == NULL);
3062 /* EA object consumes more credits than regular object: osd_mk_index
3063 * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3064 * leaf_node, could involves the block, block bitmap, groups, GDT
3065 * change for each block, so add 4 * 2 credits in that case. */
3066 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3067 osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3068 (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3069 /* Reuse idle OI block may cause additional one OI block
3071 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3072 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3077 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
3078 osd_dt_obj(dt), false, NULL, false);
3082 /* will help to find FID->ino mapping at dt_insert() */
3083 rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3089 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
3090 struct lu_attr *attr,
3091 struct dt_allocation_hint *hint,
3092 struct dt_object_format *dof, struct thandle *th)
3094 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3095 struct osd_object *obj = osd_dt_obj(dt);
3096 struct osd_thread_info *info = osd_oti_get(env);
3100 if (dt_object_exists(dt))
3103 LINVRNT(osd_invariant(obj));
3104 LASSERT(!dt_object_remote(dt));
3105 LASSERT(osd_write_locked(env, obj));
3106 LASSERT(th != NULL);
3108 if (unlikely(fid_is_acct(fid)))
3109 /* Quota files can't be created from the kernel any more,
3110 * 'tune2fs -O quota' will take care of creating them */
3113 result = __osd_object_create(info, obj, attr, hint, dof, th);
3115 result = __osd_oi_insert(env, obj, fid, th);
3116 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3117 obj->oo_dt.do_body_ops = &osd_body_ops;
3119 LASSERT(ergo(result == 0,
3120 dt_object_exists(dt) && !dt_object_remote(dt)));
3122 LASSERT(osd_invariant(obj));
3127 * Called to destroy on-disk representation of the object
3129 * Concurrency: must be locked
3131 static int osd_declare_object_destroy(const struct lu_env *env,
3132 struct dt_object *dt,
3135 struct osd_object *obj = osd_dt_obj(dt);
3136 struct inode *inode = obj->oo_inode;
3137 struct osd_thandle *oh;
3144 oh = container_of0(th, struct osd_thandle, ot_super);
3145 LASSERT(oh->ot_handle == NULL);
3147 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3148 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3149 /* Recycle idle OI leaf may cause additional three OI blocks
3151 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3152 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3153 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3154 /* one less inode */
3155 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3156 -1, oh, obj, false, NULL, false);
3159 /* data to be truncated */
3160 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3161 0, oh, obj, true, NULL, false);
3165 /* will help to find FID->ino when this object is being
3166 * added to PENDING/ */
3167 rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3172 static int osd_object_destroy(const struct lu_env *env,
3173 struct dt_object *dt,
3176 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3177 struct osd_object *obj = osd_dt_obj(dt);
3178 struct inode *inode = obj->oo_inode;
3179 struct osd_device *osd = osd_obj2dev(obj);
3180 struct osd_thandle *oh;
3184 oh = container_of0(th, struct osd_thandle, ot_super);
3185 LASSERT(oh->ot_handle);
3187 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3189 if (unlikely(fid_is_acct(fid)))
3192 if (S_ISDIR(inode->i_mode)) {
3193 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3194 inode->i_nlink == 2);
3195 /* it will check/delete the inode from remote parent,
3196 * how to optimize it? unlink performance impaction XXX */
3197 result = osd_delete_from_remote_parent(env, osd, obj, oh);
3199 CERROR("%s: delete inode "DFID": rc = %d\n",
3200 osd_name(osd), PFID(fid), result);
3202 spin_lock(&obj->oo_guard);
3204 spin_unlock(&obj->oo_guard);
3205 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3208 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3210 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3212 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3213 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3214 oh->ot_handle, OI_CHECK_FLD);
3216 osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3217 /* XXX: add to ext3 orphan list */
3218 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3220 /* not needed in the cache anymore */
3221 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3222 obj->oo_destroyed = 1;
3228 * Put the fid into lustre_mdt_attrs, and then place the structure
3229 * inode's ea. This fid should not be altered during the life time
3232 * \retval +ve, on success
3233 * \retval -ve, on error
3235 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3237 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3238 const struct lu_fid *fid, __u32 compat, __u32 incompat)
3240 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3244 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3247 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3250 lustre_lma_init(lma, fid, compat, incompat);
3251 lustre_lma_swab(lma);
3253 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
3255 /* LMA may already exist, but we need to check that all the
3256 * desired compat/incompat flags have been added. */
3257 if (unlikely(rc == -EEXIST)) {
3258 if (compat == 0 && incompat == 0)
3261 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3262 XATTR_NAME_LMA, info->oti_mdt_attrs_old,
3267 lustre_lma_swab(lma);
3268 if (!(~lma->lma_compat & compat) &&
3269 !(~lma->lma_incompat & incompat))
3272 lma->lma_compat |= compat;
3273 lma->lma_incompat |= incompat;
3274 lustre_lma_swab(lma);
3275 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3276 sizeof(*lma), XATTR_REPLACE);
3283 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3284 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3285 * To have compatilibility with 1.8 ldiskfs driver we need to have
3286 * magic number at start of fid data.
3287 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3290 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3291 const struct lu_fid *fid)
3293 if (!fid_is_namespace_visible(fid) ||
3294 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3295 param->edp_magic = 0;
3299 param->edp_magic = LDISKFS_LUFID_MAGIC;
3300 param->edp_len = sizeof(struct lu_fid) + 1;
3301 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3305 * Try to read the fid from inode ea into dt_rec.
3307 * \param fid object fid.
3309 * \retval 0 on success
3311 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3312 __u32 ino, struct lu_fid *fid,
3313 struct osd_inode_id *id)
3315 struct osd_thread_info *info = osd_oti_get(env);
3316 struct inode *inode;
3319 osd_id_gen(id, ino, OSD_OII_NOGEN);
3320 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3322 RETURN(PTR_ERR(inode));
3328 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3330 struct inode *parent_dir,
3331 const struct lu_fid *dot_fid,
3332 const struct lu_fid *dot_dot_fid,
3333 struct osd_thandle *oth)
3335 struct ldiskfs_dentry_param *dot_ldp;
3336 struct ldiskfs_dentry_param *dot_dot_ldp;
3337 __u32 saved_nlink = dir->i_nlink;
3340 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3341 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3343 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3344 dot_ldp->edp_magic = 0;
3346 rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3347 dir, dot_ldp, dot_dot_ldp);
3348 /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3349 * the subseqent ref_add() will increase the dir->i_nlink
3350 * as 3. That is incorrect for new created directory.
3352 * It looks like hack, because we want to make the OSD API
3353 * to be order-independent for new created directory object
3354 * between dt_insert(..) and ref_add() operations.
3356 * Here, we only restore the in-RAM dir-inode's nlink attr,
3357 * becuase if the nlink attr is not 2, then there will be
3358 * ref_add() called following the dt_insert(..), such call
3359 * will make both the in-RAM and on-disk dir-inode's nlink
3360 * attr to be set as 2. LU-7447 */
3361 set_nlink(dir, saved_nlink);
3366 * Create an local agent inode for remote entry
3368 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3369 struct osd_device *osd,
3370 struct osd_object *pobj,
3371 const struct lu_fid *fid,
3375 struct osd_thread_info *info = osd_oti_get(env);
3376 struct inode *local;
3377 struct osd_thandle *oh;
3382 oh = container_of(th, struct osd_thandle, ot_super);
3383 LASSERT(oh->ot_handle->h_transaction != NULL);
3385 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3386 if (IS_ERR(local)) {
3387 CERROR("%s: create local error %d\n", osd_name(osd),
3388 (int)PTR_ERR(local));
3392 /* restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3393 * correct gid on remote file, not agent here */
3394 local->i_gid = current_fsgid();
3395 ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3396 unlock_new_inode(local);
3398 /* Set special LMA flag for local agent inode */
3399 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3401 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3402 osd_name(osd), PFID(fid), rc);
3403 RETURN(ERR_PTR(rc));
3409 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3410 lu_object_fid(&pobj->oo_dt.do_lu),
3413 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3414 osd_name(osd), PFID(fid), rc);
3415 RETURN(ERR_PTR(rc));
3422 * when direntry is deleted, we have to take care of possible agent inode
3423 * referenced by that. unfortunately we can't do this at that point:
3424 * iget() within a running transaction leads to deadlock and we better do
3425 * not call that every delete declaration to save performance. so we put
3426 * a potention agent inode on a list and process that once the transaction
3427 * is over. Notice it's not any worse in terms of real orphans as regular
3428 * object destroy doesn't put inodes on the on-disk orphan list. this should
3429 * be addressed separately
3431 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3432 struct osd_thandle *oh,
3435 struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3436 struct osd_obj_orphan *oor;
3443 oor->oor_env = (struct lu_env *)env;
3444 spin_lock(&osd->od_osfs_lock);
3445 list_add(&oor->oor_list, &osd->od_orphan_list);
3446 spin_unlock(&osd->od_osfs_lock);
3448 oh->ot_remove_agents = 1;
3454 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3455 struct osd_device *osd)
3457 struct osd_thread_info *info = osd_oti_get(env);
3458 struct osd_obj_orphan *oor, *tmp;
3459 struct osd_inode_id id;
3460 struct list_head list;
3461 struct inode *inode;
3466 INIT_LIST_HEAD(&list);
3468 spin_lock(&osd->od_osfs_lock);
3469 list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3470 if (oor->oor_env == env) {
3471 list_del(&oor->oor_list);
3472 list_add(&oor->oor_list, &list);
3475 spin_unlock(&osd->od_osfs_lock);
3477 list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3481 list_del(&oor->oor_list);
3484 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3485 inode = osd_iget_fid(info, osd, &id, &fid);
3489 if (!osd_remote_fid(env, osd, &fid)) {
3494 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3496 mark_inode_dirty(inode);
3497 ldiskfs_journal_stop(jh);
3505 * OSD layer object create function for interoperability mode (b11826).
3506 * This is mostly similar to osd_object_create(). Only difference being, fid is
3507 * inserted into inode ea here.
3509 * \retval 0, on success
3510 * \retval -ve, on error
3512 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
3513 struct lu_attr *attr,
3514 struct dt_allocation_hint *hint,
3515 struct dt_object_format *dof,
3518 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3519 struct osd_object *obj = osd_dt_obj(dt);
3520 struct osd_thread_info *info = osd_oti_get(env);
3521 int result, on_ost = 0;
3525 if (dt_object_exists(dt))
3528 LASSERT(osd_invariant(obj));
3529 LASSERT(!dt_object_remote(dt));
3530 LASSERT(osd_write_locked(env, obj));
3531 LASSERT(th != NULL);
3533 if (unlikely(fid_is_acct(fid)))
3534 /* Quota files can't be created from the kernel any more,
3535 * 'tune2fs -O quota' will take care of creating them */
3538 result = __osd_object_create(info, obj, attr, hint, dof, th);
3540 if (fid_is_idif(fid) &&
3541 !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3542 struct lu_fid *tfid = &info->oti_fid;
3543 struct ost_id *oi = &info->oti_ostid;
3545 fid_to_ostid(fid, oi);
3546 ostid_to_fid(tfid, oi, 0);
3548 result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3549 LMAC_FID_ON_OST, 0);
3551 on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3553 result = osd_ea_fid_set(info, obj->oo_inode, fid,
3554 on_ost ? LMAC_FID_ON_OST : 0,
3557 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3558 obj->oo_dt.do_body_ops = &osd_body_ops;
3562 result = __osd_oi_insert(env, obj, fid, th);
3564 /* a small optimization - dt_insert() isn't usually applied
3565 * to OST objects, so we don't need to cache OI mapping for
3567 if (result == 0 && on_ost == 0) {
3568 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3569 result = osd_idc_find_and_init(env, osd, obj);
3570 LASSERT(result == 0);
3573 LASSERT(ergo(result == 0,
3574 dt_object_exists(dt) && !dt_object_remote(dt)));
3575 LINVRNT(osd_invariant(obj));
3579 static int osd_declare_object_ref_add(const struct lu_env *env,
3580 struct dt_object *dt,
3581 struct thandle *handle)
3583 struct osd_thandle *oh;
3585 /* it's possible that object doesn't exist yet */
3586 LASSERT(handle != NULL);
3588 oh = container_of0(handle, struct osd_thandle, ot_super);
3589 LASSERT(oh->ot_handle == NULL);
3591 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3592 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3598 * Concurrency: @dt is write locked.
3600 static int osd_object_ref_add(const struct lu_env *env,
3601 struct dt_object *dt, struct thandle *th)
3603 struct osd_object *obj = osd_dt_obj(dt);
3604 struct inode *inode = obj->oo_inode;
3605 struct osd_thandle *oh;
3608 if (!dt_object_exists(dt) || obj->oo_destroyed)
3611 LINVRNT(osd_invariant(obj));
3612 LASSERT(!dt_object_remote(dt));
3613 LASSERT(osd_write_locked(env, obj));
3614 LASSERT(th != NULL);
3616 oh = container_of0(th, struct osd_thandle, ot_super);
3617 LASSERT(oh->ot_handle != NULL);
3619 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
3621 CDEBUG(D_INODE, DFID" increase nlink %d\n",
3622 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3624 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
3625 * (65000) subdirectories by storing "1" in i_nlink if the link count
3626 * would otherwise overflow. Directory tranversal tools understand
3627 * that (st_nlink == 1) indicates that the filesystem dose not track
3628 * hard links count on the directory, and will not abort subdirectory
3629 * scanning early once (st_nlink - 2) subdirs have been found.
3631 * This also has to properly handle the case of inodes with nlink == 0
3632 * in case they are being linked into the PENDING directory
3634 spin_lock(&obj->oo_guard);
3635 if (unlikely(inode->i_nlink == 0))
3636 /* inc_nlink from 0 may cause WARN_ON */
3637 set_nlink(inode, 1);
3639 ldiskfs_inc_count(oh->ot_handle, inode);
3640 if (!S_ISDIR(inode->i_mode))
3641 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
3643 spin_unlock(&obj->oo_guard);
3645 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3646 LINVRNT(osd_invariant(obj));
3648 osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
3653 static int osd_declare_object_ref_del(const struct lu_env *env,
3654 struct dt_object *dt,
3655 struct thandle *handle)
3657 struct osd_thandle *oh;
3659 if (!dt_object_exists(dt))
3662 LASSERT(!dt_object_remote(dt));
3663 LASSERT(handle != NULL);
3665 oh = container_of0(handle, struct osd_thandle, ot_super);
3666 LASSERT(oh->ot_handle == NULL);
3668 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
3669 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3675 * Concurrency: @dt is write locked.
3677 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
3680 struct osd_object *obj = osd_dt_obj(dt);
3681 struct inode *inode = obj->oo_inode;
3682 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3683 struct osd_thandle *oh;
3685 if (!dt_object_exists(dt))
3688 LINVRNT(osd_invariant(obj));
3689 LASSERT(!dt_object_remote(dt));
3690 LASSERT(osd_write_locked(env, obj));
3691 LASSERT(th != NULL);
3693 oh = container_of0(th, struct osd_thandle, ot_super);
3694 LASSERT(oh->ot_handle != NULL);
3696 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
3698 spin_lock(&obj->oo_guard);
3699 /* That can be result of upgrade from old Lustre version and
3700 * applied only to local files. Just skip this ref_del call.
3701 * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
3702 if (inode->i_nlink == 0) {
3703 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
3704 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
3705 ", maybe an upgraded file? (LU-3915)\n",
3706 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
3707 spin_unlock(&obj->oo_guard);
3711 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
3712 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3714 ldiskfs_dec_count(oh->ot_handle, inode);
3715 spin_unlock(&obj->oo_guard);
3717 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3718 LINVRNT(osd_invariant(obj));
3720 osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
3726 * Get the 64-bit version for an inode.
3728 static int osd_object_version_get(const struct lu_env *env,
3729 struct dt_object *dt, dt_obj_version_t *ver)
3731 struct inode *inode = osd_dt_obj(dt)->oo_inode;
3733 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
3734 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3735 *ver = LDISKFS_I(inode)->i_fs_version;
3740 * Concurrency: @dt is read locked.
3742 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
3743 struct lu_buf *buf, const char *name)
3745 struct osd_object *obj = osd_dt_obj(dt);
3746 struct inode *inode = obj->oo_inode;
3747 struct osd_thread_info *info = osd_oti_get(env);
3748 struct dentry *dentry = &info->oti_obj_dentry;
3749 bool cache_xattr = false;
3752 /* version get is not real XATTR but uses xattr API */
3753 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3754 /* for version we are just using xattr API but change inode
3756 if (buf->lb_len == 0)
3757 return sizeof(dt_obj_version_t);
3759 if (buf->lb_len < sizeof(dt_obj_version_t))
3762 osd_object_version_get(env, dt, buf->lb_buf);
3764 return sizeof(dt_obj_version_t);
3767 if (!dt_object_exists(dt))
3770 LASSERT(!dt_object_remote(dt));
3771 LASSERT(inode->i_op != NULL);
3772 LASSERT(inode->i_op->getxattr != NULL);
3774 if (strcmp(name, XATTR_NAME_LOV) == 0 ||
3775 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
3779 rc = osd_oxc_get(obj, name, buf);
3784 rc = __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
3786 if (rc == -ENOENT || rc == -ENODATA)
3787 osd_oxc_add(obj, name, NULL, 0);
3788 else if (rc > 0 && buf->lb_buf != NULL)
3789 osd_oxc_add(obj, name, buf->lb_buf, rc);
3795 static int osd_declare_xattr_set(const struct lu_env *env,
3796 struct dt_object *dt,
3797 const struct lu_buf *buf, const char *name,
3798 int fl, struct thandle *handle)
3800 struct osd_thandle *oh;
3802 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3804 LASSERT(handle != NULL);
3806 oh = container_of0(handle, struct osd_thandle, ot_super);
3807 LASSERT(oh->ot_handle == NULL);
3809 if (strcmp(name, XATTR_NAME_LMA) == 0) {
3810 /* For non-upgrading case, the LMA is set first and
3811 * usually fit inode. But for upgrade case, the LMA
3812 * may be in another separated EA block. */
3813 if (!dt_object_exists(dt))
3815 else if (fl == LU_XATTR_REPLACE)
3819 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3823 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
3828 if (buf->lb_buf == NULL && dt_object_exists(dt)) {
3829 /* learn xattr size from osd_xattr_get if
3830 attribute has not been read yet */
3831 buflen = __osd_xattr_get(
3832 osd_dt_obj(dt)->oo_inode,
3833 &osd_oti_get(env)->oti_obj_dentry,
3838 buflen = buf->lb_len;
3841 if (buflen > sb->s_blocksize) {
3842 credits += osd_calc_bkmap_credits(
3844 (buflen + sb->s_blocksize - 1) >>
3845 sb->s_blocksize_bits);
3849 * xattr set may involve inode quota change, reserve credits for
3850 * dquot_initialize()
3852 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3855 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3861 * Set the 64-bit version for object
3863 static void osd_object_version_set(const struct lu_env *env,
3864 struct dt_object *dt,
3865 dt_obj_version_t *new_version)
3867 struct inode *inode = osd_dt_obj(dt)->oo_inode;
3869 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
3870 *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3872 LDISKFS_I(inode)->i_fs_version = *new_version;
3873 /** Version is set after all inode operations are finished,
3874 * so we should mark it dirty here */
3875 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3879 * Concurrency: @dt is write locked.
3881 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3882 const struct lu_buf *buf, const char *name, int fl,
3883 struct thandle *handle)
3885 struct osd_object *obj = osd_dt_obj(dt);
3886 struct inode *inode = obj->oo_inode;
3887 struct osd_thread_info *info = osd_oti_get(env);
3892 LASSERT(handle != NULL);
3894 /* version set is not real XATTR */
3895 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3896 /* for version we are just using xattr API but change inode
3898 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3899 osd_object_version_set(env, dt, buf->lb_buf);
3900 return sizeof(dt_obj_version_t);
3903 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
3904 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3906 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3907 if (fl & LU_XATTR_REPLACE)
3908 fs_flags |= XATTR_REPLACE;
3910 if (fl & LU_XATTR_CREATE)
3911 fs_flags |= XATTR_CREATE;
3913 if (strcmp(name, XATTR_NAME_LMV) == 0) {
3914 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3916 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3920 lma->lma_incompat |= LMAI_STRIPED;
3921 lustre_lma_swab(lma);
3922 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3923 sizeof(*lma), XATTR_REPLACE);
3928 rc = __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3930 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3933 (strcmp(name, XATTR_NAME_LOV) == 0 ||
3934 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
3935 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
3941 * Concurrency: @dt is read locked.
3943 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3944 const struct lu_buf *buf)
3946 struct osd_object *obj = osd_dt_obj(dt);
3947 struct inode *inode = obj->oo_inode;
3948 struct osd_thread_info *info = osd_oti_get(env);
3949 struct dentry *dentry = &info->oti_obj_dentry;
3951 if (!dt_object_exists(dt))
3954 LASSERT(!dt_object_remote(dt));
3955 LASSERT(inode->i_op != NULL);
3956 LASSERT(inode->i_op->listxattr != NULL);
3958 dentry->d_inode = inode;
3959 dentry->d_sb = inode->i_sb;
3960 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3963 static int osd_declare_xattr_del(const struct lu_env *env,
3964 struct dt_object *dt, const char *name,
3965 struct thandle *handle)
3967 struct osd_thandle *oh;
3968 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3970 LASSERT(!dt_object_remote(dt));
3971 LASSERT(handle != NULL);
3973 oh = container_of0(handle, struct osd_thandle, ot_super);
3974 LASSERT(oh->ot_handle == NULL);
3976 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3977 osd_dto_credits_noquota[DTO_XATTR_SET]);
3979 * xattr del may involve inode quota change, reserve credits for
3980 * dquot_initialize()
3982 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3988 * Concurrency: @dt is write locked.
3990 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3991 const char *name, struct thandle *handle)
3993 struct osd_object *obj = osd_dt_obj(dt);
3994 struct inode *inode = obj->oo_inode;
3995 struct osd_thread_info *info = osd_oti_get(env);
3996 struct dentry *dentry = &info->oti_obj_dentry;
3999 if (!dt_object_exists(dt))
4002 LASSERT(!dt_object_remote(dt));
4003 LASSERT(inode->i_op != NULL);
4004 LASSERT(inode->i_op->removexattr != NULL);
4005 LASSERT(handle != NULL);
4007 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4009 ll_vfs_dq_init(inode);
4010 dentry->d_inode = inode;
4011 dentry->d_sb = inode->i_sb;
4012 rc = inode->i_op->removexattr(dentry, name);
4013 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4016 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4017 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4018 osd_oxc_del(obj, name);
4023 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4024 __u64 start, __u64 end)
4026 struct osd_object *obj = osd_dt_obj(dt);
4027 struct inode *inode = obj->oo_inode;
4028 struct osd_thread_info *info = osd_oti_get(env);
4029 struct dentry *dentry = &info->oti_obj_dentry;
4030 struct file *file = &info->oti_file;
4035 dentry->d_inode = inode;
4036 dentry->d_sb = inode->i_sb;
4037 file->f_path.dentry = dentry;
4038 file->f_mapping = inode->i_mapping;
4039 file->f_op = inode->i_fop;
4040 set_file_inode(file, inode);
4042 rc = ll_vfs_fsync_range(file, start, end, 0);
4047 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4056 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4057 const struct dt_index_features *feat)
4059 struct iam_descr *descr;
4061 if (osd_object_is_root(o))
4062 return feat == &dt_directory_features;
4064 LASSERT(o->oo_dir != NULL);
4066 descr = o->oo_dir->od_container.ic_descr;
4067 if (feat == &dt_directory_features) {
4068 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4074 feat->dif_keysize_min <= descr->id_key_size &&
4075 descr->id_key_size <= feat->dif_keysize_max &&
4076 feat->dif_recsize_min <= descr->id_rec_size &&
4077 descr->id_rec_size <= feat->dif_recsize_max &&
4078 !(feat->dif_flags & (DT_IND_VARKEY |
4079 DT_IND_VARREC | DT_IND_NONUNQ)) &&
4080 ergo(feat->dif_flags & DT_IND_UPDATE,
4081 1 /* XXX check that object (and file system) is
4086 static int osd_iam_container_init(const struct lu_env *env,
4087 struct osd_object *obj,
4088 struct osd_directory *dir)
4090 struct iam_container *bag = &dir->od_container;
4093 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4097 result = iam_container_setup(bag);
4099 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4101 iam_container_fini(bag);
4108 * Concurrency: no external locking is necessary.
4110 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4111 const struct dt_index_features *feat)
4115 struct osd_object *obj = osd_dt_obj(dt);
4117 LINVRNT(osd_invariant(obj));
4119 if (osd_object_is_root(obj)) {
4120 dt->do_index_ops = &osd_index_ea_ops;
4122 } else if (feat == &dt_directory_features) {
4123 dt->do_index_ops = &osd_index_ea_ops;
4124 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4129 } else if (unlikely(feat == &dt_otable_features)) {
4130 dt->do_index_ops = &osd_otable_ops;
4132 } else if (unlikely(feat == &dt_acct_features)) {
4133 dt->do_index_ops = &osd_acct_index_ops;
4136 } else if (!osd_has_index(obj)) {
4137 struct osd_directory *dir;
4142 spin_lock(&obj->oo_guard);
4143 if (obj->oo_dir == NULL)
4147 * Concurrent thread allocated container data.
4150 spin_unlock(&obj->oo_guard);
4152 * Now, that we have container data, serialize its
4155 down_write(&obj->oo_ext_idx_sem);
4157 * recheck under lock.
4159 if (!osd_has_index(obj))
4160 result = osd_iam_container_init(env, obj,
4164 up_write(&obj->oo_ext_idx_sem);
4172 if (result == 0 && skip_iam == 0) {
4173 if (!osd_iam_index_probe(env, obj, feat))
4176 LINVRNT(osd_invariant(obj));
4178 if (result == 0 && feat == &dt_quota_glb_features &&
4179 fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
4180 result = osd_quota_migration(env, dt);
4185 static int osd_otable_it_attr_get(const struct lu_env *env,
4186 struct dt_object *dt,
4187 struct lu_attr *attr)
4193 static const struct dt_object_operations osd_obj_ops = {
4194 .do_read_lock = osd_object_read_lock,
4195 .do_write_lock = osd_object_write_lock,
4196 .do_read_unlock = osd_object_read_unlock,
4197 .do_write_unlock = osd_object_write_unlock,
4198 .do_write_locked = osd_object_write_locked,
4199 .do_attr_get = osd_attr_get,
4200 .do_declare_attr_set = osd_declare_attr_set,
4201 .do_attr_set = osd_attr_set,
4202 .do_ah_init = osd_ah_init,
4203 .do_declare_create = osd_declare_object_create,
4204 .do_create = osd_object_create,
4205 .do_declare_destroy = osd_declare_object_destroy,
4206 .do_destroy = osd_object_destroy,
4207 .do_index_try = osd_index_try,
4208 .do_declare_ref_add = osd_declare_object_ref_add,
4209 .do_ref_add = osd_object_ref_add,
4210 .do_declare_ref_del = osd_declare_object_ref_del,
4211 .do_ref_del = osd_object_ref_del,
4212 .do_xattr_get = osd_xattr_get,
4213 .do_declare_xattr_set = osd_declare_xattr_set,
4214 .do_xattr_set = osd_xattr_set,
4215 .do_declare_xattr_del = osd_declare_xattr_del,
4216 .do_xattr_del = osd_xattr_del,
4217 .do_xattr_list = osd_xattr_list,
4218 .do_object_sync = osd_object_sync,
4219 .do_invalidate = osd_invalidate,
4223 * dt_object_operations for interoperability mode
4224 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
4226 static const struct dt_object_operations osd_obj_ea_ops = {
4227 .do_read_lock = osd_object_read_lock,
4228 .do_write_lock = osd_object_write_lock,
4229 .do_read_unlock = osd_object_read_unlock,
4230 .do_write_unlock = osd_object_write_unlock,
4231 .do_write_locked = osd_object_write_locked,
4232 .do_attr_get = osd_attr_get,
4233 .do_declare_attr_set = osd_declare_attr_set,
4234 .do_attr_set = osd_attr_set,
4235 .do_ah_init = osd_ah_init,
4236 .do_declare_create = osd_declare_object_create,
4237 .do_create = osd_object_ea_create,
4238 .do_declare_destroy = osd_declare_object_destroy,
4239 .do_destroy = osd_object_destroy,
4240 .do_index_try = osd_index_try,
4241 .do_declare_ref_add = osd_declare_object_ref_add,
4242 .do_ref_add = osd_object_ref_add,
4243 .do_declare_ref_del = osd_declare_object_ref_del,
4244 .do_ref_del = osd_object_ref_del,
4245 .do_xattr_get = osd_xattr_get,
4246 .do_declare_xattr_set = osd_declare_xattr_set,
4247 .do_xattr_set = osd_xattr_set,
4248 .do_declare_xattr_del = osd_declare_xattr_del,
4249 .do_xattr_del = osd_xattr_del,
4250 .do_xattr_list = osd_xattr_list,
4251 .do_object_sync = osd_object_sync,
4252 .do_invalidate = osd_invalidate,
4255 static const struct dt_object_operations osd_obj_otable_it_ops = {
4256 .do_attr_get = osd_otable_it_attr_get,
4257 .do_index_try = osd_index_try,
4260 static int osd_index_declare_iam_delete(const struct lu_env *env,
4261 struct dt_object *dt,
4262 const struct dt_key *key,
4263 struct thandle *handle)
4265 struct osd_thandle *oh;
4267 oh = container_of0(handle, struct osd_thandle, ot_super);
4268 LASSERT(oh->ot_handle == NULL);
4270 /* Recycle may cause additional three blocks to be changed. */
4271 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4272 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4278 * delete a (key, value) pair from index \a dt specified by \a key
4280 * \param dt osd index object
4281 * \param key key for index
4282 * \param rec record reference
4283 * \param handle transaction handler
4286 * \retval -ve failure
4288 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4289 const struct dt_key *key,
4290 struct thandle *handle)
4292 struct osd_thread_info *oti = osd_oti_get(env);
4293 struct osd_object *obj = osd_dt_obj(dt);
4294 struct osd_thandle *oh;
4295 struct iam_path_descr *ipd;
4296 struct iam_container *bag = &obj->oo_dir->od_container;
4300 if (!dt_object_exists(dt))
4303 LINVRNT(osd_invariant(obj));
4304 LASSERT(!dt_object_remote(dt));
4305 LASSERT(bag->ic_object == obj->oo_inode);
4306 LASSERT(handle != NULL);
4308 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4310 ipd = osd_idx_ipd_get(env, bag);
4311 if (unlikely(ipd == NULL))
4314 oh = container_of0(handle, struct osd_thandle, ot_super);
4315 LASSERT(oh->ot_handle != NULL);
4316 LASSERT(oh->ot_handle->h_transaction != NULL);
4318 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4319 /* swab quota uid/gid provided by caller */
4320 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4321 key = (const struct dt_key *)&oti->oti_quota_id;
4324 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4325 osd_ipd_put(env, bag, ipd);
4326 LINVRNT(osd_invariant(obj));
4327 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4331 static int osd_index_declare_ea_delete(const struct lu_env *env,
4332 struct dt_object *dt,
4333 const struct dt_key *key,
4334 struct thandle *handle)
4336 struct osd_thandle *oh;
4337 struct inode *inode;
4341 LASSERT(!dt_object_remote(dt));
4342 LASSERT(handle != NULL);
4344 oh = container_of0(handle, struct osd_thandle, ot_super);
4345 LASSERT(oh->ot_handle == NULL);
4347 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
4348 if (key != NULL && unlikely(strcmp((char *)key, dotdot) == 0)) {
4349 /* '..' to a remote object has a local representative */
4350 credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
4351 /* to reset LMAI_REMOTE_PARENT */
4354 osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
4356 inode = osd_dt_obj(dt)->oo_inode;
4360 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
4361 0, oh, osd_dt_obj(dt), true, NULL, false);
4365 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
4368 struct osd_fid_pack *rec;
4371 if (de->file_type & LDISKFS_DIRENT_LUFID) {
4372 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
4373 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
4374 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
4380 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
4381 const struct lu_fid *fid)
4383 struct seq_server_site *ss = osd_seq_site(osd);
4386 /* FID seqs not in FLDB, must be local seq */
4387 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
4390 /* If FLD is not being initialized yet, it only happens during the
4391 * initialization, likely during mgs initialization, and we assume
4392 * this is local FID. */
4393 if (ss == NULL || ss->ss_server_fld == NULL)
4396 /* Only check the local FLDB here */
4397 if (osd_seq_exists(env, osd, fid_seq(fid)))
4404 * Index delete function for interoperability mode (b11826).
4405 * It will remove the directory entry added by osd_index_ea_insert().
4406 * This entry is needed to maintain name->fid mapping.
4408 * \param key, key i.e. file entry to be deleted
4410 * \retval 0, on success
4411 * \retval -ve, on error
4413 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
4414 const struct dt_key *key, struct thandle *handle)
4416 struct osd_object *obj = osd_dt_obj(dt);
4417 struct inode *dir = obj->oo_inode;
4418 struct dentry *dentry;
4419 struct osd_thandle *oh;
4420 struct ldiskfs_dir_entry_2 *de = NULL;
4421 struct buffer_head *bh;
4422 struct htree_lock *hlock = NULL;
4423 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
4424 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4428 if (!dt_object_exists(dt))
4431 LINVRNT(osd_invariant(obj));
4432 LASSERT(!dt_object_remote(dt));
4433 LASSERT(handle != NULL);
4435 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4437 oh = container_of(handle, struct osd_thandle, ot_super);
4438 LASSERT(oh->ot_handle != NULL);
4439 LASSERT(oh->ot_handle->h_transaction != NULL);
4441 ll_vfs_dq_init(dir);
4442 dentry = osd_child_dentry_get(env, obj,
4443 (char *)key, strlen((char *)key));
4445 if (obj->oo_hl_head != NULL) {
4446 hlock = osd_oti_get(env)->oti_hlock;
4447 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4448 dir, LDISKFS_HLOCK_DEL);
4450 down_write(&obj->oo_ext_idx_sem);
4453 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4455 /* If this is not the ".." entry, it might be a remote DNE
4456 * entry and we need to check if the FID is for a remote
4457 * MDT. If the FID is not in the directory entry (e.g.
4458 * upgraded 1.8 filesystem without dirdata enabled) then
4459 * we need to get the FID from the LMA. For a remote directory
4460 * there HAS to be an LMA, it cannot be an IGIF inode in this
4463 * Delete the entry before the agent inode in order to
4464 * simplify error handling. At worst an error after deleting
4465 * the entry first might leak the agent inode afterward. The
4466 * reverse would need filesystem abort in case of error deleting
4467 * the entry after the agent had been removed, or leave a
4468 * dangling entry pointing at a random inode. */
4469 if (strcmp((char *)key, dotdot) != 0) {
4470 LASSERT(de != NULL);
4471 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
4472 if (rc == -ENODATA) {
4473 /* can't get FID, postpone to the end of the
4474 * transaction when iget() is safe */
4475 osd_schedule_agent_inode_removal(env, oh,
4476 le32_to_cpu(de->inode));
4477 } else if (rc == 0 &&
4478 unlikely(osd_remote_fid(env, osd, fid))) {
4479 osd_schedule_agent_inode_removal(env, oh,
4480 le32_to_cpu(de->inode));
4483 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
4489 ldiskfs_htree_unlock(hlock);
4491 up_write(&obj->oo_ext_idx_sem);
4496 /* For inode on the remote MDT, .. will point to
4497 * /Agent directory, Check whether it needs to delete
4498 * from agent directory */
4499 if (unlikely(strcmp((char *)key, dotdot) == 0)) {
4502 ret = osd_delete_from_remote_parent(env, osd_obj2dev(obj),
4505 /* Sigh, the entry has been deleted, and
4506 * it is not easy to revert it back, so
4507 * let's keep this error private, and let
4508 * LFSCK fix it. XXX */
4509 CERROR("%s: delete remote parent "DFID": rc = %d\n",
4510 osd_name(osd), PFID(fid), ret);
4513 LASSERT(osd_invariant(obj));
4514 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4519 * Lookup index for \a key and copy record to \a rec.
4521 * \param dt osd index object
4522 * \param key key for index
4523 * \param rec record reference
4525 * \retval +ve success : exact mach
4526 * \retval 0 return record with key not greater than \a key
4527 * \retval -ve failure
4529 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
4530 struct dt_rec *rec, const struct dt_key *key)
4532 struct osd_object *obj = osd_dt_obj(dt);
4533 struct iam_path_descr *ipd;
4534 struct iam_container *bag = &obj->oo_dir->od_container;
4535 struct osd_thread_info *oti = osd_oti_get(env);
4536 struct iam_iterator *it = &oti->oti_idx_it;
4537 struct iam_rec *iam_rec;
4541 if (!dt_object_exists(dt))
4544 LASSERT(osd_invariant(obj));
4545 LASSERT(!dt_object_remote(dt));
4546 LASSERT(bag->ic_object == obj->oo_inode);
4548 ipd = osd_idx_ipd_get(env, bag);
4552 /* got ipd now we can start iterator. */
4553 iam_it_init(it, bag, 0, ipd);
4555 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4556 /* swab quota uid/gid provided by caller */
4557 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4558 key = (const struct dt_key *)&oti->oti_quota_id;
4561 rc = iam_it_get(it, (struct iam_key *)key);
4563 if (S_ISDIR(obj->oo_inode->i_mode))
4564 iam_rec = (struct iam_rec *)oti->oti_ldp;
4566 iam_rec = (struct iam_rec *) rec;
4568 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
4570 if (S_ISDIR(obj->oo_inode->i_mode))
4571 osd_fid_unpack((struct lu_fid *) rec,
4572 (struct osd_fid_pack *)iam_rec);
4573 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
4574 osd_quota_unpack(obj, rec);
4579 osd_ipd_put(env, bag, ipd);
4581 LINVRNT(osd_invariant(obj));
4586 static int osd_index_declare_iam_insert(const struct lu_env *env,
4587 struct dt_object *dt,
4588 const struct dt_rec *rec,
4589 const struct dt_key *key,
4590 struct thandle *handle)
4592 struct osd_thandle *oh;
4594 LASSERT(handle != NULL);
4596 oh = container_of0(handle, struct osd_thandle, ot_super);
4597 LASSERT(oh->ot_handle == NULL);
4599 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4600 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4606 * Inserts (key, value) pair in \a dt index object.
4608 * \param dt osd index object
4609 * \param key key for index
4610 * \param rec record reference
4611 * \param th transaction handler
4614 * \retval -ve failure
4616 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
4617 const struct dt_rec *rec,
4618 const struct dt_key *key, struct thandle *th,
4621 struct osd_object *obj = osd_dt_obj(dt);
4622 struct iam_path_descr *ipd;
4623 struct osd_thandle *oh;
4624 struct iam_container *bag;
4625 struct osd_thread_info *oti = osd_oti_get(env);
4626 struct iam_rec *iam_rec;
4630 if (!dt_object_exists(dt))
4633 LINVRNT(osd_invariant(obj));
4634 LASSERT(!dt_object_remote(dt));
4636 bag = &obj->oo_dir->od_container;
4637 LASSERT(bag->ic_object == obj->oo_inode);
4638 LASSERT(th != NULL);
4640 osd_trans_exec_op(env, th, OSD_OT_INSERT);
4642 ipd = osd_idx_ipd_get(env, bag);
4643 if (unlikely(ipd == NULL))
4646 oh = container_of0(th, struct osd_thandle, ot_super);
4647 LASSERT(oh->ot_handle != NULL);
4648 LASSERT(oh->ot_handle->h_transaction != NULL);
4649 if (S_ISDIR(obj->oo_inode->i_mode)) {
4650 iam_rec = (struct iam_rec *)oti->oti_ldp;
4651 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
4652 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4653 /* pack quota uid/gid */
4654 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4655 key = (const struct dt_key *)&oti->oti_quota_id;
4656 /* pack quota record */
4657 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
4658 iam_rec = (struct iam_rec *)rec;
4660 iam_rec = (struct iam_rec *)rec;
4663 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
4665 osd_ipd_put(env, bag, ipd);
4666 LINVRNT(osd_invariant(obj));
4667 osd_trans_exec_check(env, th, OSD_OT_INSERT);
4672 * Calls ldiskfs_add_entry() to add directory entry
4673 * into the directory. This is required for
4674 * interoperability mode (b11826)
4676 * \retval 0, on success
4677 * \retval -ve, on error
4679 static int __osd_ea_add_rec(struct osd_thread_info *info,
4680 struct osd_object *pobj, struct inode *cinode,
4681 const char *name, const struct lu_fid *fid,
4682 struct htree_lock *hlock, struct thandle *th)
4684 struct ldiskfs_dentry_param *ldp;
4685 struct dentry *child;
4686 struct osd_thandle *oth;
4689 oth = container_of(th, struct osd_thandle, ot_super);
4690 LASSERT(oth->ot_handle != NULL);
4691 LASSERT(oth->ot_handle->h_transaction != NULL);
4692 LASSERT(pobj->oo_inode);
4694 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
4695 if (unlikely(pobj->oo_inode ==
4696 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
4699 osd_get_ldiskfs_dirent_param(ldp, fid);
4700 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
4701 child->d_fsdata = (void *)ldp;
4702 ll_vfs_dq_init(pobj->oo_inode);
4703 rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
4704 child, cinode, hlock);
4705 if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
4706 struct ldiskfs_dir_entry_2 *de;
4707 struct buffer_head *bh;
4710 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
4713 rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
4716 if (S_ISDIR(cinode->i_mode))
4717 de->file_type = LDISKFS_DIRENT_LUFID |
4718 LDISKFS_FT_REG_FILE;
4720 de->file_type = LDISKFS_DIRENT_LUFID |
4722 ldiskfs_handle_dirty_metadata(oth->ot_handle,
4733 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
4734 * into the directory.Also sets flags into osd object to
4735 * indicate dot and dotdot are created. This is required for
4736 * interoperability mode (b11826)
4738 * \param dir directory for dot and dotdot fixup.
4739 * \param obj child object for linking
4741 * \retval 0, on success
4742 * \retval -ve, on error
4744 static int osd_add_dot_dotdot(struct osd_thread_info *info,
4745 struct osd_object *dir,
4746 struct inode *parent_dir, const char *name,
4747 const struct lu_fid *dot_fid,
4748 const struct lu_fid *dot_dot_fid,
4751 struct inode *inode = dir->oo_inode;
4752 struct osd_thandle *oth;
4755 oth = container_of(th, struct osd_thandle, ot_super);
4756 LASSERT(oth->ot_handle->h_transaction != NULL);
4757 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4759 if (strcmp(name, dot) == 0) {
4760 if (dir->oo_compat_dot_created) {
4763 LASSERT(inode->i_ino == parent_dir->i_ino);
4764 dir->oo_compat_dot_created = 1;
4767 } else if (strcmp(name, dotdot) == 0) {
4768 if (!dir->oo_compat_dot_created)
4770 /* in case of rename, dotdot is already created */
4771 if (dir->oo_compat_dotdot_created) {
4772 return __osd_ea_add_rec(info, dir, parent_dir, name,
4773 dot_dot_fid, NULL, th);
4776 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
4777 struct lu_fid tfid = *dot_dot_fid;
4780 result = osd_add_dot_dotdot_internal(info,
4781 dir->oo_inode, parent_dir, dot_fid,
4784 result = osd_add_dot_dotdot_internal(info,
4785 dir->oo_inode, parent_dir, dot_fid,
4790 dir->oo_compat_dotdot_created = 1;
4798 * It will call the appropriate osd_add* function and return the
4799 * value, return by respective functions.
4801 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4802 struct inode *cinode, const char *name,
4803 const struct lu_fid *fid, struct thandle *th)
4805 struct osd_thread_info *info = osd_oti_get(env);
4806 struct htree_lock *hlock;
4809 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4811 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4813 if (hlock != NULL) {
4814 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4817 down_write(&pobj->oo_ext_idx_sem);
4820 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4821 lu_object_fid(&pobj->oo_dt.do_lu),
4824 if (hlock != NULL) {
4825 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4826 pobj->oo_inode, LDISKFS_HLOCK_ADD);
4828 down_write(&pobj->oo_ext_idx_sem);
4831 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4832 struct lu_fid *tfid = &info->oti_fid;
4836 rc = __osd_ea_add_rec(info, pobj, cinode, name,
4839 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4844 ldiskfs_htree_unlock(hlock);
4846 up_write(&pobj->oo_ext_idx_sem);
4852 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4853 struct osd_idmap_cache *oic)
4855 struct osd_scrub *scrub = &dev->od_scrub;
4856 struct lu_fid *fid = &oic->oic_fid;
4857 struct osd_inode_id *id = &oic->oic_lid;
4858 struct inode *inode = NULL;
4864 if (!fid_is_norm(fid) && !fid_is_igif(fid))
4867 if (scrub->os_pos_current > id->oii_ino)
4871 rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
4872 if (rc == -ENOENT) {
4873 __u32 gen = id->oii_gen;
4879 inode = osd_iget(oti, dev, id);
4880 /* The inode has been removed (by race maybe). */
4881 if (IS_ERR(inode)) {
4882 rc = PTR_ERR(inode);
4884 RETURN(rc == -ESTALE ? -ENOENT : rc);
4887 /* The OI mapping is lost. */
4888 if (gen != OSD_OII_NOGEN)
4892 /* The inode may has been reused by others, we do not know,
4893 * leave it to be handled by subsequent osd_fid_lookup(). */
4895 } else if (rc != 0 || osd_id_eq(id, &oti->oti_id)) {
4902 if (thread_is_running(&scrub->os_thread)) {
4903 if (inode == NULL) {
4904 inode = osd_iget(oti, dev, id);
4905 /* The inode has been removed (by race maybe). */
4906 if (IS_ERR(inode)) {
4907 rc = PTR_ERR(inode);
4909 RETURN(rc == -ESTALE ? -ENOENT : rc);
4913 rc = osd_oii_insert(dev, oic, insert);
4914 /* There is race condition between osd_oi_lookup and OI scrub.
4915 * The OI scrub finished just after osd_oi_lookup() failure.
4916 * Under such case, it is unnecessary to trigger OI scrub again,
4917 * but try to call osd_oi_lookup() again. */
4918 if (unlikely(rc == -EAGAIN))
4921 if (!S_ISDIR(inode->i_mode))
4924 rc = osd_check_lmv(oti, dev, inode, oic);
4930 if (!dev->od_noscrub && ++once == 1) {
4931 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
4933 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
4934 "%.16s: trigger partial OI scrub for RPC inconsistency "
4935 "checking FID "DFID": rc = %d\n",
4936 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4938 if (rc == 0 || rc == -EALREADY)
4948 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4949 struct osd_device *dev,
4950 struct osd_idmap_cache *oic,
4951 struct lu_fid *fid, __u32 ino)
4953 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
4954 struct inode *inode;
4957 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4958 inode = osd_iget(oti, dev, &oic->oic_lid);
4959 if (IS_ERR(inode)) {
4960 fid_zero(&oic->oic_fid);
4961 return PTR_ERR(inode);
4964 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4967 fid_zero(&oic->oic_fid);
4969 *fid = oic->oic_fid = lma->lma_self_fid;
4973 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4974 struct osd_inode_id *id, const struct lu_fid *fid)
4976 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4977 id->oii_ino, id->oii_gen, info);
4978 info->oti_cache.oic_lid = *id;
4979 info->oti_cache.oic_fid = *fid;
4980 info->oti_cache.oic_dev = osd;
4984 * Get parent FID from the linkEA.
4986 * For a directory which parent resides on remote MDT, to satisfy the
4987 * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
4988 * the other hand, to make the lookup(..) on the directory can return
4989 * the real parent FID, we append the real parent FID after its ".."
4990 * name entry in the /REMOTE_PARENT_DIR.
4992 * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
4993 * backup. So after the restore, we cannot get the right parent FID from
4994 * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
4995 * we have stored the real parent FID in the directory object's linkEA,
4996 * we can parse the linkEA for the real parent FID.
4998 * \param[in] env pointer to the thread context
4999 * \param[in] obj pointer to the object to be handled
5000 * \param[out]fid pointer to the buffer to hold the parent FID
5002 * \retval 0 for getting the real parent FID successfully
5003 * \retval negative error number on failure
5005 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5006 struct osd_object *obj,
5009 struct osd_thread_info *oti = osd_oti_get(env);
5010 struct lu_buf *buf = &oti->oti_big_buf;
5011 struct dentry *dentry = &oti->oti_obj_dentry;
5012 struct inode *inode = obj->oo_inode;
5013 struct linkea_data ldata = { NULL };
5018 if (!S_ISDIR(inode->i_mode))
5022 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5023 buf->lb_buf, buf->lb_len);
5024 if (rc == -ERANGE) {
5025 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5028 lu_buf_realloc(buf, rc);
5029 if (buf->lb_buf == NULL)
5036 if (unlikely(rc == 0))
5042 if (unlikely(buf->lb_buf == NULL)) {
5043 lu_buf_realloc(buf, rc);
5044 if (buf->lb_buf == NULL)
5051 rc = linkea_init_with_rec(&ldata);
5053 linkea_first_entry(&ldata);
5054 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5060 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5061 struct inode *inode,
5062 const struct lu_fid *pfid,
5063 const char *name, const int namelen)
5065 struct osd_thread_info *oti = osd_oti_get(env);
5066 struct lu_buf *buf = &oti->oti_big_buf;
5067 struct dentry *dentry = &oti->oti_obj_dentry;
5068 struct linkea_data ldata = { NULL };
5069 struct lu_name cname = { .ln_name = name,
5070 .ln_namelen = namelen };
5075 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5076 buf->lb_buf, buf->lb_len);
5078 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5083 if (unlikely(rc == 0))
5086 if (buf->lb_len < rc) {
5087 lu_buf_realloc(buf, rc);
5088 if (buf->lb_buf == NULL)
5095 rc = linkea_init_with_rec(&ldata);
5097 rc = linkea_links_find(&ldata, &cname, pfid);
5103 * Calls ->lookup() to find dentry. From dentry get inode and
5104 * read inode's ea to get fid. This is required for interoperability
5107 * \retval 0, on success
5108 * \retval -ve, on error
5110 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5111 struct dt_rec *rec, const struct dt_key *key)
5113 struct inode *dir = obj->oo_inode;
5114 struct dentry *dentry;
5115 struct ldiskfs_dir_entry_2 *de;
5116 struct buffer_head *bh;
5117 struct lu_fid *fid = (struct lu_fid *) rec;
5118 struct htree_lock *hlock = NULL;
5123 LASSERT(dir->i_op != NULL);
5124 LASSERT(dir->i_op->lookup != NULL);
5126 dentry = osd_child_dentry_get(env, obj,
5127 (char *)key, strlen((char *)key));
5129 if (obj->oo_hl_head != NULL) {
5130 hlock = osd_oti_get(env)->oti_hlock;
5131 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5132 dir, LDISKFS_HLOCK_LOOKUP);
5134 down_read(&obj->oo_ext_idx_sem);
5137 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5139 struct osd_thread_info *oti = osd_oti_get(env);
5140 struct osd_inode_id *id = &oti->oti_id;
5141 struct osd_idmap_cache *oic = &oti->oti_cache;
5142 struct osd_device *dev = osd_obj2dev(obj);
5144 ino = le32_to_cpu(de->inode);
5145 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5147 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5151 rc = osd_get_fid_from_dentry(de, rec);
5153 /* done with de, release bh */
5156 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5157 const char *name = (const char *)key;
5159 /* If the parent is on remote MDT, and there
5160 * is no FID-in-dirent, then we have to get
5161 * the parent FID from the linkEA. */
5162 if (likely(strlen(name) == 2 &&
5163 name[0] == '.' && name[1] == '.'))
5164 rc = osd_get_pfid_from_linkea(env, obj,
5167 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5170 osd_id_gen(id, ino, OSD_OII_NOGEN);
5173 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5174 fid_zero(&oic->oic_fid);
5179 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5180 rc = osd_consistency_check(oti, dev, oic);
5182 fid_zero(&oic->oic_fid);
5191 ldiskfs_htree_unlock(hlock);
5193 up_read(&obj->oo_ext_idx_sem);
5198 * Put the osd object once done with it.
5200 * \param obj osd object that needs to be put
5202 static inline void osd_object_put(const struct lu_env *env,
5203 struct osd_object *obj)
5205 lu_object_put(env, &obj->oo_dt.do_lu);
5208 static int osd_index_declare_ea_insert(const struct lu_env *env,
5209 struct dt_object *dt,
5210 const struct dt_rec *rec,
5211 const struct dt_key *key,
5212 struct thandle *handle)
5214 struct osd_thandle *oh;
5215 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5216 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5217 const struct lu_fid *fid = rec1->rec_fid;
5218 int credits, rc = 0;
5219 struct osd_idmap_cache *idc;
5222 LASSERT(!dt_object_remote(dt));
5223 LASSERT(handle != NULL);
5224 LASSERT(fid != NULL);
5225 LASSERT(rec1->rec_type != 0);
5227 oh = container_of0(handle, struct osd_thandle, ot_super);
5228 LASSERT(oh->ot_handle == NULL);
5230 credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5232 /* we can't call iget() while a transactions is running
5233 * (this can lead to a deadlock), but we need to know
5234 * inum and object type. so we find this information at
5235 * declaration and cache in per-thread info */
5236 idc = osd_idc_find_or_init(env, osd, fid);
5238 RETURN(PTR_ERR(idc));
5239 if (idc->oic_remote) {
5240 /* a reference to remote inode is represented by an
5241 * agent inode which we have to create */
5242 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5243 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5246 osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5248 if (osd_dt_obj(dt)->oo_inode != NULL) {
5249 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5251 /* We ignore block quota on meta pool (MDTs), so needn't
5252 * calculate how many blocks will be consumed by this index
5254 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5255 i_gid_read(inode), 0, oh,
5256 osd_dt_obj(dt), true, NULL, false);
5263 * Index add function for interoperability mode (b11826).
5264 * It will add the directory entry.This entry is needed to
5265 * maintain name->fid mapping.
5267 * \param key it is key i.e. file entry to be inserted
5268 * \param rec it is value of given key i.e. fid
5270 * \retval 0, on success
5271 * \retval -ve, on error
5273 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5274 const struct dt_rec *rec,
5275 const struct dt_key *key, struct thandle *th,
5278 struct osd_object *obj = osd_dt_obj(dt);
5279 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5280 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5281 const struct lu_fid *fid = rec1->rec_fid;
5282 const char *name = (const char *)key;
5283 struct osd_thread_info *oti = osd_oti_get(env);
5284 struct inode *child_inode = NULL;
5285 struct osd_idmap_cache *idc;
5289 if (!dt_object_exists(dt))
5292 LASSERT(osd_invariant(obj));
5293 LASSERT(!dt_object_remote(dt));
5294 LASSERT(th != NULL);
5296 osd_trans_exec_op(env, th, OSD_OT_INSERT);
5298 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
5300 idc = osd_idc_find(env, osd, fid);
5301 if (unlikely(idc == NULL)) {
5302 /* this dt_insert() wasn't declared properly, so
5303 * FID is missing in OI cache. we better do not
5304 * lookup FID in FLDB/OI and don't risk to deadlock,
5305 * but in some special cases (lfsck testing, etc)
5306 * it's much simpler than fixing a caller */
5307 CERROR("%s: "DFID" wasn't declared for insert\n",
5308 osd_name(osd), PFID(fid));
5310 idc = osd_idc_find_or_init(env, osd, fid);
5312 RETURN(PTR_ERR(idc));
5315 if (idc->oic_remote) {
5316 /* Insert remote entry */
5317 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
5318 struct osd_mdobj_map *omm = osd->od_mdt_map;
5319 struct osd_thandle *oh;
5321 /* If parent on remote MDT, we need put this object
5323 oh = container_of(th, typeof(*oh), ot_super);
5324 rc = osd_add_to_remote_parent(env, osd, obj, oh);
5326 CERROR("%s: add "DFID" error: rc = %d\n",
5328 PFID(lu_object_fid(&dt->do_lu)), rc);
5332 child_inode = igrab(omm->omm_remote_parent->d_inode);
5334 child_inode = osd_create_local_agent_inode(env, osd,
5335 obj, fid, rec1->rec_type & S_IFMT, th);
5336 if (IS_ERR(child_inode))
5337 RETURN(PTR_ERR(child_inode));
5340 /* Insert local entry */
5341 if (unlikely(idc->oic_lid.oii_ino == 0)) {
5342 /* for a reason OI cache wasn't filled properly */
5343 CERROR("%s: OIC for "DFID" isn't filled\n",
5344 osd_name(osd), PFID(fid));
5347 child_inode = oti->oti_inode;
5348 if (unlikely(child_inode == NULL)) {
5349 struct ldiskfs_inode_info *lii;
5353 child_inode = oti->oti_inode = &lii->vfs_inode;
5355 child_inode->i_sb = osd_sb(osd);
5356 child_inode->i_ino = idc->oic_lid.oii_ino;
5357 child_inode->i_mode = rec1->rec_type & S_IFMT;
5360 rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
5362 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
5363 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
5365 if (child_inode && child_inode != oti->oti_inode)
5367 LASSERT(osd_invariant(obj));
5368 osd_trans_exec_check(env, th, OSD_OT_INSERT);
5373 * Initialize osd Iterator for given osd index object.
5375 * \param dt osd index object
5378 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
5379 struct dt_object *dt,
5382 struct osd_it_iam *it;
5383 struct osd_object *obj = osd_dt_obj(dt);
5384 struct lu_object *lo = &dt->do_lu;
5385 struct iam_path_descr *ipd;
5386 struct iam_container *bag = &obj->oo_dir->od_container;
5388 if (!dt_object_exists(dt))
5389 return ERR_PTR(-ENOENT);
5393 return ERR_PTR(-ENOMEM);
5395 ipd = osd_it_ipd_get(env, bag);
5396 if (likely(ipd != NULL)) {
5400 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
5401 return (struct dt_it *)it;
5404 return ERR_PTR(-ENOMEM);
5409 * free given Iterator.
5412 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
5414 struct osd_it_iam *it = (struct osd_it_iam *)di;
5415 struct osd_object *obj = it->oi_obj;
5417 iam_it_fini(&it->oi_it);
5418 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
5419 lu_object_put(env, &obj->oo_dt.do_lu);
5424 * Move Iterator to record specified by \a key
5426 * \param di osd iterator
5427 * \param key key for index
5429 * \retval +ve di points to record with least key not larger than key
5430 * \retval 0 di points to exact matched key
5431 * \retval -ve failure
5434 static int osd_it_iam_get(const struct lu_env *env,
5435 struct dt_it *di, const struct dt_key *key)
5437 struct osd_thread_info *oti = osd_oti_get(env);
5438 struct osd_it_iam *it = (struct osd_it_iam *)di;
5440 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5441 /* swab quota uid/gid */
5442 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5443 key = (struct dt_key *)&oti->oti_quota_id;
5446 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
5452 * \param di osd iterator
5454 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
5456 struct osd_it_iam *it = (struct osd_it_iam *)di;
5458 iam_it_put(&it->oi_it);
5462 * Move iterator by one record
5464 * \param di osd iterator
5466 * \retval +1 end of container reached
5468 * \retval -ve failure
5471 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
5473 struct osd_it_iam *it = (struct osd_it_iam *)di;
5475 return iam_it_next(&it->oi_it);
5479 * Return pointer to the key under iterator.
5482 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
5483 const struct dt_it *di)
5485 struct osd_thread_info *oti = osd_oti_get(env);
5486 struct osd_it_iam *it = (struct osd_it_iam *)di;
5487 struct osd_object *obj = it->oi_obj;
5490 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
5492 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
5493 /* swab quota uid/gid */
5494 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
5495 key = (struct dt_key *)&oti->oti_quota_id;
5502 * Return size of key under iterator (in bytes)
5505 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
5507 struct osd_it_iam *it = (struct osd_it_iam *)di;
5509 return iam_it_key_size(&it->oi_it);
5513 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
5515 /* check if file type is required */
5516 if (ent->lde_attrs & LUDA_TYPE) {
5517 struct luda_type *lt;
5518 int align = sizeof(*lt) - 1;
5520 len = (len + align) & ~align;
5521 lt = (struct luda_type *)(ent->lde_name + len);
5522 lt->lt_type = cpu_to_le16(DTTOIF(type));
5525 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
5529 * build lu direct from backend fs dirent.
5533 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
5534 char *name, __u16 namelen, __u16 type, __u32 attr)
5536 ent->lde_attrs = attr | LUDA_FID;
5537 fid_cpu_to_le(&ent->lde_fid, fid);
5539 ent->lde_hash = cpu_to_le64(offset);
5540 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
5542 strncpy(ent->lde_name, name, namelen);
5543 ent->lde_name[namelen] = '\0';
5544 ent->lde_namelen = cpu_to_le16(namelen);
5546 /* append lustre attributes */
5547 osd_it_append_attrs(ent, namelen, type);
5551 * Return pointer to the record under iterator.
5553 static int osd_it_iam_rec(const struct lu_env *env,
5554 const struct dt_it *di,
5555 struct dt_rec *dtrec, __u32 attr)
5557 struct osd_it_iam *it = (struct osd_it_iam *)di;
5558 struct osd_thread_info *info = osd_oti_get(env);
5561 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
5562 const struct osd_fid_pack *rec;
5563 struct lu_fid *fid = &info->oti_fid;
5564 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
5570 name = (char *)iam_it_key_get(&it->oi_it);
5572 RETURN(PTR_ERR(name));
5574 namelen = iam_it_key_size(&it->oi_it);
5576 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
5578 RETURN(PTR_ERR(rec));
5580 rc = osd_fid_unpack(fid, rec);
5584 hash = iam_it_store(&it->oi_it);
5586 /* IAM does not store object type in IAM index (dir) */
5587 osd_it_pack_dirent(lde, fid, hash, name, namelen,
5589 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5590 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5591 (struct iam_rec *)dtrec);
5592 osd_quota_unpack(it->oi_obj, dtrec);
5594 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5595 (struct iam_rec *)dtrec);
5602 * Returns cookie for current Iterator position.
5604 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
5606 struct osd_it_iam *it = (struct osd_it_iam *)di;
5608 return iam_it_store(&it->oi_it);
5612 * Restore iterator from cookie.
5614 * \param di osd iterator
5615 * \param hash Iterator location cookie
5617 * \retval +ve di points to record with least key not larger than key.
5618 * \retval 0 di points to exact matched key
5619 * \retval -ve failure
5622 static int osd_it_iam_load(const struct lu_env *env,
5623 const struct dt_it *di, __u64 hash)
5625 struct osd_it_iam *it = (struct osd_it_iam *)di;
5627 return iam_it_load(&it->oi_it, hash);
5630 static const struct dt_index_operations osd_index_iam_ops = {
5631 .dio_lookup = osd_index_iam_lookup,
5632 .dio_declare_insert = osd_index_declare_iam_insert,
5633 .dio_insert = osd_index_iam_insert,
5634 .dio_declare_delete = osd_index_declare_iam_delete,
5635 .dio_delete = osd_index_iam_delete,
5637 .init = osd_it_iam_init,
5638 .fini = osd_it_iam_fini,
5639 .get = osd_it_iam_get,
5640 .put = osd_it_iam_put,
5641 .next = osd_it_iam_next,
5642 .key = osd_it_iam_key,
5643 .key_size = osd_it_iam_key_size,
5644 .rec = osd_it_iam_rec,
5645 .store = osd_it_iam_store,
5646 .load = osd_it_iam_load
5652 * Creates or initializes iterator context.
5654 * \retval struct osd_it_ea, iterator structure on success
5657 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
5658 struct dt_object *dt,
5661 struct osd_object *obj = osd_dt_obj(dt);
5662 struct osd_thread_info *info = osd_oti_get(env);
5663 struct osd_it_ea *oie;
5665 struct lu_object *lo = &dt->do_lu;
5666 struct dentry *obj_dentry;
5669 if (!dt_object_exists(dt) || obj->oo_destroyed)
5670 RETURN(ERR_PTR(-ENOENT));
5672 OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
5674 RETURN(ERR_PTR(-ENOMEM));
5675 obj_dentry = &oie->oie_dentry;
5677 obj_dentry->d_inode = obj->oo_inode;
5678 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
5679 obj_dentry->d_name.hash = 0;
5681 oie->oie_rd_dirent = 0;
5682 oie->oie_it_dirent = 0;
5683 oie->oie_dirent = NULL;
5684 if (unlikely(!info->oti_it_ea_buf_used)) {
5685 oie->oie_buf = info->oti_it_ea_buf;
5686 info->oti_it_ea_buf_used = 1;
5688 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5689 if (oie->oie_buf == NULL)
5690 RETURN(ERR_PTR(-ENOMEM));
5694 file = &oie->oie_file;
5696 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
5697 if (attr & LUDA_64BITHASH)
5698 file->f_mode = FMODE_64BITHASH;
5700 file->f_mode = FMODE_32BITHASH;
5701 file->f_path.dentry = obj_dentry;
5702 file->f_mapping = obj->oo_inode->i_mapping;
5703 file->f_op = obj->oo_inode->i_fop;
5704 set_file_inode(file, obj->oo_inode);
5707 RETURN((struct dt_it *) oie);
5711 * Destroy or finishes iterator context.
5713 * \param di iterator structure to be destroyed
5715 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
5717 struct osd_thread_info *info = osd_oti_get(env);
5718 struct osd_it_ea *oie = (struct osd_it_ea *)di;
5719 struct osd_object *obj = oie->oie_obj;
5720 struct inode *inode = obj->oo_inode;
5723 oie->oie_file.f_op->release(inode, &oie->oie_file);
5724 lu_object_put(env, &obj->oo_dt.do_lu);
5725 if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
5726 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5728 info->oti_it_ea_buf_used = 0;
5729 OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
5734 * It position the iterator at given key, so that next lookup continues from
5735 * that key Or it is similar to dio_it->load() but based on a key,
5736 * rather than file position.
5738 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
5741 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
5743 static int osd_it_ea_get(const struct lu_env *env,
5744 struct dt_it *di, const struct dt_key *key)
5746 struct osd_it_ea *it = (struct osd_it_ea *)di;
5749 LASSERT(((const char *)key)[0] == '\0');
5750 it->oie_file.f_pos = 0;
5751 it->oie_rd_dirent = 0;
5752 it->oie_it_dirent = 0;
5753 it->oie_dirent = NULL;
5761 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
5765 struct osd_filldir_cbs {
5766 #ifdef HAVE_DIR_CONTEXT
5767 struct dir_context ctx;
5769 struct osd_it_ea *it;
5772 * It is called internally by ->readdir(). It fills the
5773 * iterator's in-memory data structure with required
5774 * information i.e. name, namelen, rec_size etc.
5776 * \param buf in which information to be filled in.
5777 * \param name name of the file in given dir
5779 * \retval 0 on success
5780 * \retval 1 on buffer full
5782 #ifdef HAVE_FILLDIR_USE_CTX
5783 static int osd_ldiskfs_filldir(struct dir_context *buf,
5785 static int osd_ldiskfs_filldir(void *buf,
5787 const char *name, int namelen,
5788 loff_t offset, __u64 ino, unsigned d_type)
5790 struct osd_it_ea *it =
5791 ((struct osd_filldir_cbs *)buf)->it;
5792 struct osd_object *obj = it->oie_obj;
5793 struct osd_it_ea_dirent *ent = it->oie_dirent;
5794 struct lu_fid *fid = &ent->oied_fid;
5795 struct osd_fid_pack *rec;
5798 /* this should never happen */
5799 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5800 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5804 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
5808 /* "." is just the object itself. */
5809 if (namelen == 1 && name[0] == '.') {
5810 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5811 } else if (d_type & LDISKFS_DIRENT_LUFID) {
5812 rec = (struct osd_fid_pack*) (name + namelen + 1);
5813 if (osd_fid_unpack(fid, rec) != 0)
5818 d_type &= ~LDISKFS_DIRENT_LUFID;
5820 /* NOT export local root. */
5821 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5822 ino = obj->oo_inode->i_ino;
5823 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5826 ent->oied_ino = ino;
5827 ent->oied_off = offset;
5828 ent->oied_namelen = namelen;
5829 ent->oied_type = d_type;
5831 memcpy(ent->oied_name, name, namelen);
5833 it->oie_rd_dirent++;
5834 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
5839 * Calls ->readdir() to load a directory entry at a time
5840 * and stored it in iterator's in-memory data structure.
5842 * \param di iterator's in memory structure
5844 * \retval 0 on success
5845 * \retval -ve on error
5846 * \retval +1 reach the end of entry
5848 static int osd_ldiskfs_it_fill(const struct lu_env *env,
5849 const struct dt_it *di)
5851 struct osd_it_ea *it = (struct osd_it_ea *)di;
5852 struct osd_object *obj = it->oie_obj;
5853 struct inode *inode = obj->oo_inode;
5854 struct htree_lock *hlock = NULL;
5855 struct file *filp = &it->oie_file;
5857 struct osd_filldir_cbs buf = {
5858 #ifdef HAVE_DIR_CONTEXT
5859 .ctx.actor = osd_ldiskfs_filldir,
5865 it->oie_dirent = it->oie_buf;
5866 it->oie_rd_dirent = 0;
5868 if (obj->oo_hl_head != NULL) {
5869 hlock = osd_oti_get(env)->oti_hlock;
5870 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5871 inode, LDISKFS_HLOCK_READDIR);
5873 down_read(&obj->oo_ext_idx_sem);
5876 #ifdef HAVE_DIR_CONTEXT
5877 buf.ctx.pos = filp->f_pos;
5878 rc = inode->i_fop->iterate(filp, &buf.ctx);
5879 filp->f_pos = buf.ctx.pos;
5881 rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
5885 ldiskfs_htree_unlock(hlock);
5887 up_read(&obj->oo_ext_idx_sem);
5889 if (it->oie_rd_dirent == 0) {
5890 /*If it does not get any dirent, it means it has been reached
5891 *to the end of the dir */
5892 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
5896 it->oie_dirent = it->oie_buf;
5897 it->oie_it_dirent = 1;
5904 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5905 * to load a directory entry at a time and stored it in
5906 * iterator's in-memory data structure.
5908 * \param di iterator's in memory structure
5910 * \retval +ve iterator reached to end
5911 * \retval 0 iterator not reached to end
5912 * \retval -ve on error
5914 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
5916 struct osd_it_ea *it = (struct osd_it_ea *)di;
5921 if (it->oie_it_dirent < it->oie_rd_dirent) {
5923 (void *) it->oie_dirent +
5924 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
5925 it->oie_dirent->oied_namelen);
5926 it->oie_it_dirent++;
5929 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
5932 rc = osd_ldiskfs_it_fill(env, di);
5939 * Returns the key at current position from iterator's in memory structure.
5941 * \param di iterator's in memory structure
5943 * \retval key i.e. struct dt_key on success
5945 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
5946 const struct dt_it *di)
5948 struct osd_it_ea *it = (struct osd_it_ea *)di;
5950 return (struct dt_key *)it->oie_dirent->oied_name;
5954 * Returns the key's size at current position from iterator's in memory structure.
5956 * \param di iterator's in memory structure
5958 * \retval key_size i.e. struct dt_key on success
5960 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
5962 struct osd_it_ea *it = (struct osd_it_ea *)di;
5964 return it->oie_dirent->oied_namelen;
5967 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
5969 if (LDISKFS_DIR_REC_LEN(de) >=
5970 __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
5977 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
5978 unsigned blocksize, bool dotdot)
5981 return osd_dotdot_has_space(de);
5983 if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
5984 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5991 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
5992 handle_t *jh, struct dentry *dentry,
5993 const struct lu_fid *fid, struct buffer_head *bh,
5994 struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
5997 struct inode *dir = dentry->d_parent->d_inode;
5998 struct inode *inode = dentry->d_inode;
5999 struct osd_fid_pack *rec;
6000 struct ldiskfs_dentry_param *ldp;
6001 int namelen = dentry->d_name.len;
6003 struct osd_thread_info *info = osd_oti_get(env);
6006 if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
6007 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6010 /* There is enough space to hold the FID-in-dirent. */
6011 if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6012 rc = ldiskfs_journal_get_write_access(jh, bh);
6016 de->name[namelen] = 0;
6017 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6018 rec->fp_len = sizeof(struct lu_fid) + 1;
6019 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6020 de->file_type |= LDISKFS_DIRENT_LUFID;
6021 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6028 rc = ldiskfs_delete_entry(jh, dir, de, bh);
6032 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6033 osd_get_ldiskfs_dirent_param(ldp, fid);
6034 dentry->d_fsdata = (void *)ldp;
6035 ll_vfs_dq_init(dir);
6036 rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6037 /* It is too bad, we cannot reinsert the name entry back.
6038 * That means we lose it! */
6040 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
6041 "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6042 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
6043 dir->i_ino, dir->i_generation, namelen,
6044 dentry->d_name.name, PFID(fid), rc);
6050 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6051 struct osd_it_ea *it, struct lu_fid *fid,
6052 struct osd_inode_id *id, __u32 *attr)
6054 struct osd_thread_info *info = osd_oti_get(env);
6055 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
6056 struct osd_device *dev = osd_obj2dev(obj);
6057 struct super_block *sb = osd_sb(dev);
6058 const char *devname = osd_name(dev);
6059 struct osd_it_ea_dirent *ent = it->oie_dirent;
6060 struct inode *dir = obj->oo_inode;
6061 struct htree_lock *hlock = NULL;
6062 struct buffer_head *bh = NULL;
6063 handle_t *jh = NULL;
6064 struct ldiskfs_dir_entry_2 *de;
6065 struct dentry *dentry;
6066 struct inode *inode;
6067 const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6070 bool dotdot = false;
6074 if (ent->oied_name[0] == '.') {
6075 if (ent->oied_namelen == 1)
6078 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6082 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6083 inode = osd_iget(info, dev, id);
6084 if (IS_ERR(inode)) {
6085 rc = PTR_ERR(inode);
6086 if (rc == -ENOENT || rc == -ESTALE) {
6087 /* Maybe dangling name entry, or
6088 * corrupted directory entry. */
6089 *attr |= LUDA_UNKNOWN;
6092 CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6093 "check_repair, dir = %lu/%u, name = %.*s, "
6094 "ino = %llu, rc = %d\n",
6095 devname, dir->i_ino, dir->i_generation,
6096 ent->oied_namelen, ent->oied_name,
6103 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6105 rc = osd_get_lma(info, inode, dentry, lma);
6106 if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6111 /* We need to ensure that the name entry is still valid.
6112 * Because it may be removed or renamed by other already.
6114 * The unlink or rename operation will start journal before PDO lock,
6115 * so to avoid deadlock, here we need to start journal handle before
6116 * related PDO lock also. But because we do not know whether there
6117 * will be something to be repaired before PDO lock, we just start
6118 * journal without conditions.
6120 * We may need to remove the name entry firstly, then insert back.
6121 * One credit is for user quota file update.
6122 * One credit is for group quota file update.
6123 * Two credits are for dirty inode. */
6124 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6125 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6127 if (dev->od_dirent_journal != 0) {
6130 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6133 CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6134 "check_repair, dir = %lu/%u, credits = %d, "
6135 "name = %.*s, ino = %llu: rc = %d\n",
6136 devname, dir->i_ino, dir->i_generation, credits,
6137 ent->oied_namelen, ent->oied_name,
6140 GOTO(out_inode, rc);
6143 if (obj->oo_hl_head != NULL) {
6144 hlock = osd_oti_get(env)->oti_hlock;
6145 /* "0" means exclusive lock for the whole directory.
6146 * We need to prevent others access such name entry
6147 * during the delete + insert. Neither HLOCK_ADD nor
6148 * HLOCK_DEL cannot guarantee the atomicity. */
6149 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6151 down_write(&obj->oo_ext_idx_sem);
6154 if (obj->oo_hl_head != NULL) {
6155 hlock = osd_oti_get(env)->oti_hlock;
6156 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6157 LDISKFS_HLOCK_LOOKUP);
6159 down_read(&obj->oo_ext_idx_sem);
6163 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6164 if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6165 *attr |= LUDA_IGNORE;
6170 /* For dotdot entry, if there is not enough space to hold the
6171 * FID-in-dirent, just keep them there. It only happens when the
6172 * device upgraded from 1.8 or restored from MDT file-level backup.
6173 * For the whole directory, only dotdot entry have no FID-in-dirent
6174 * and needs to get FID from LMA when readdir, it will not affect the
6175 * performance much. */
6176 if (dotdot && !osd_dotdot_has_space(de)) {
6177 *attr |= LUDA_UNKNOWN;
6183 if (lu_fid_eq(fid, &lma->lma_self_fid))
6186 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6187 struct lu_fid *tfid = &lma->lma_self_fid;
6189 if (likely(dotdot &&
6190 fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6191 fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6192 /* It must be REMOTE_PARENT_DIR and as the
6193 * 'dotdot' entry of remote directory */
6194 *attr |= LUDA_IGNORE;
6196 CDEBUG(D_LFSCK, "%s: expect remote agent "
6197 "parent directory, but got %.*s under "
6198 "dir = %lu/%u with the FID "DFID"\n",
6199 devname, ent->oied_namelen,
6200 ent->oied_name, dir->i_ino,
6201 dir->i_generation, PFID(tfid));
6203 *attr |= LUDA_UNKNOWN;
6210 if (!fid_is_zero(fid)) {
6211 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6213 if (rc == -ENOENT ||
6215 !(dev->od_scrub.os_file.sf_flags & SF_UPGRADE))) {
6216 /* linkEA does not recognize the dirent entry,
6217 * it may because the dirent entry corruption
6218 * and points to other's inode. */
6219 CDEBUG(D_LFSCK, "%s: the target inode does not "
6220 "recognize the dirent, dir = %lu/%u, "
6221 " name = %.*s, ino = %llu, "
6222 DFID": rc = %d\n", devname, dir->i_ino,
6223 dir->i_generation, ent->oied_namelen,
6224 ent->oied_name, ent->oied_ino, PFID(fid), rc);
6225 *attr |= LUDA_UNKNOWN;
6230 if (rc && rc != -ENODATA) {
6231 CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6232 "dir = %lu/%u, name = %.*s, ino = %llu, "
6233 DFID": rc = %d\n", devname, dir->i_ino,
6234 dir->i_generation, ent->oied_namelen,
6235 ent->oied_name, ent->oied_ino, PFID(fid), rc);
6236 *attr |= LUDA_UNKNOWN;
6243 /* linkEA recognizes the dirent entry, the FID-in-LMA is
6244 * valid, trusted, in spite of fid_is_sane(fid) or not. */
6245 if (*attr & LUDA_VERIFY_DRYRUN) {
6246 *fid = lma->lma_self_fid;
6247 *attr |= LUDA_REPAIR;
6254 dev->od_dirent_journal = 1;
6255 if (hlock != NULL) {
6256 ldiskfs_htree_unlock(hlock);
6259 up_read(&obj->oo_ext_idx_sem);
6265 *fid = lma->lma_self_fid;
6267 /* Update or append the FID-in-dirent. */
6268 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6269 bh, de, hlock, dotdot);
6271 *attr |= LUDA_REPAIR;
6273 CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
6274 "the dirent, dir = %lu/%u, name = %.*s, "
6275 "ino = %llu, "DFID": rc = %d\n",
6276 devname, dir->i_ino, dir->i_generation,
6277 ent->oied_namelen, ent->oied_name,
6278 ent->oied_ino, PFID(fid), rc);
6280 /* lma is NULL, trust the FID-in-dirent if it is valid. */
6281 if (*attr & LUDA_VERIFY_DRYRUN) {
6282 if (fid_is_sane(fid)) {
6283 *attr |= LUDA_REPAIR;
6284 } else if (dev->od_index == 0) {
6285 lu_igif_build(fid, inode->i_ino,
6286 inode->i_generation);
6287 *attr |= LUDA_UPGRADE;
6295 dev->od_dirent_journal = 1;
6296 if (hlock != NULL) {
6297 ldiskfs_htree_unlock(hlock);
6300 up_read(&obj->oo_ext_idx_sem);
6307 if (unlikely(fid_is_sane(fid))) {
6308 /* FID-in-dirent exists, but FID-in-LMA is lost.
6309 * Trust the FID-in-dirent, and add FID-in-LMA. */
6310 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
6312 *attr |= LUDA_REPAIR;
6314 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
6315 "update dirent, dir = %lu/%u, "
6316 "name = %.*s, ino = %llu, "
6318 devname, dir->i_ino, dir->i_generation,
6319 ent->oied_namelen, ent->oied_name,
6320 ent->oied_ino, PFID(fid), rc);
6321 } else if (dev->od_index == 0) {
6322 lu_igif_build(fid, inode->i_ino, inode->i_generation);
6323 /* It is probably IGIF object. Only aappend the
6324 * FID-in-dirent. OI scrub will process FID-in-LMA. */
6325 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6326 bh, de, hlock, dotdot);
6328 *attr |= LUDA_UPGRADE;
6330 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
6331 "after the dirent, dir = %lu/%u, "
6332 "name = %.*s, ino = %llu, "
6334 devname, dir->i_ino, dir->i_generation,
6335 ent->oied_namelen, ent->oied_name,
6336 ent->oied_ino, PFID(fid), rc);
6345 if (hlock != NULL) {
6346 ldiskfs_htree_unlock(hlock);
6348 if (dev->od_dirent_journal != 0)
6349 up_write(&obj->oo_ext_idx_sem);
6351 up_read(&obj->oo_ext_idx_sem);
6355 ldiskfs_journal_stop(jh);
6359 if (rc >= 0 && !dirty)
6360 dev->od_dirent_journal = 0;
6366 * Returns the value at current position from iterator's in memory structure.
6368 * \param di struct osd_it_ea, iterator's in memory structure
6369 * \param attr attr requested for dirent.
6370 * \param lde lustre dirent
6372 * \retval 0 no error and \param lde has correct lustre dirent.
6373 * \retval -ve on error
6375 static inline int osd_it_ea_rec(const struct lu_env *env,
6376 const struct dt_it *di,
6377 struct dt_rec *dtrec, __u32 attr)
6379 struct osd_it_ea *it = (struct osd_it_ea *)di;
6380 struct osd_object *obj = it->oie_obj;
6381 struct osd_device *dev = osd_obj2dev(obj);
6382 struct osd_thread_info *oti = osd_oti_get(env);
6383 struct osd_inode_id *id = &oti->oti_id;
6384 struct lu_fid *fid = &it->oie_dirent->oied_fid;
6385 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6386 __u32 ino = it->oie_dirent->oied_ino;
6390 LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
6392 if (attr & LUDA_VERIFY) {
6393 if (unlikely(ino == osd_remote_parent_ino(dev))) {
6394 attr |= LUDA_IGNORE;
6395 /* If the parent is on remote MDT, and there
6396 * is no FID-in-dirent, then we have to get
6397 * the parent FID from the linkEA. */
6398 if (!fid_is_sane(fid) &&
6399 it->oie_dirent->oied_namelen == 2 &&
6400 it->oie_dirent->oied_name[0] == '.' &&
6401 it->oie_dirent->oied_name[1] == '.')
6402 osd_get_pfid_from_linkea(env, obj, fid);
6404 rc = osd_dirent_check_repair(env, obj, it, fid, id,
6408 if (!fid_is_sane(fid))
6409 attr |= LUDA_UNKNOWN;
6411 attr &= ~LU_DIRENT_ATTRS_MASK;
6412 if (!fid_is_sane(fid)) {
6413 bool is_dotdot = false;
6414 if (it->oie_dirent->oied_namelen == 2 &&
6415 it->oie_dirent->oied_name[0] == '.' &&
6416 it->oie_dirent->oied_name[1] == '.')
6418 /* If the parent is on remote MDT, and there
6419 * is no FID-in-dirent, then we have to get
6420 * the parent FID from the linkEA. */
6421 if (ino == osd_remote_parent_ino(dev) && is_dotdot) {
6422 rc = osd_get_pfid_from_linkea(env, obj, fid);
6424 if (is_dotdot == false &&
6425 OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
6428 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6431 osd_id_gen(id, ino, OSD_OII_NOGEN);
6435 /* Pack the entry anyway, at least the offset is right. */
6436 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
6437 it->oie_dirent->oied_name,
6438 it->oie_dirent->oied_namelen,
6439 it->oie_dirent->oied_type, attr);
6444 if (osd_remote_fid(env, dev, fid))
6447 if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
6448 osd_add_oi_cache(oti, dev, id, fid);
6450 RETURN(rc > 0 ? 0 : rc);
6454 * Returns the record size size at current position.
6456 * This function will return record(lu_dirent) size in bytes.
6458 * \param[in] env execution environment
6459 * \param[in] di iterator's in memory structure
6460 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
6461 * calculate the lu_dirent size.
6463 * \retval record size(in bytes & in memory) of the current lu_dirent
6466 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
6469 struct osd_it_ea *it = (struct osd_it_ea *)di;
6471 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
6475 * Returns a cookie for current position of the iterator head, so that
6476 * user can use this cookie to load/start the iterator next time.
6478 * \param di iterator's in memory structure
6480 * \retval cookie for current position, on success
6482 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
6484 struct osd_it_ea *it = (struct osd_it_ea *)di;
6486 return it->oie_dirent->oied_off;
6490 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6491 * to load a directory entry at a time and stored it i inn,
6492 * in iterator's in-memory data structure.
6494 * \param di struct osd_it_ea, iterator's in memory structure
6496 * \retval +ve on success
6497 * \retval -ve on error
6499 static int osd_it_ea_load(const struct lu_env *env,
6500 const struct dt_it *di, __u64 hash)
6502 struct osd_it_ea *it = (struct osd_it_ea *)di;
6506 it->oie_file.f_pos = hash;
6508 rc = osd_ldiskfs_it_fill(env, di);
6519 * Index lookup function for interoperability mode (b11826).
6521 * \param key, key i.e. file name to be searched
6523 * \retval +ve, on success
6524 * \retval -ve, on error
6526 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
6527 struct dt_rec *rec, const struct dt_key *key)
6529 struct osd_object *obj = osd_dt_obj(dt);
6534 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
6535 LINVRNT(osd_invariant(obj));
6537 rc = osd_ea_lookup_rec(env, obj, rec, key);
6544 * Index and Iterator operations for interoperability
6545 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
6547 static const struct dt_index_operations osd_index_ea_ops = {
6548 .dio_lookup = osd_index_ea_lookup,
6549 .dio_declare_insert = osd_index_declare_ea_insert,
6550 .dio_insert = osd_index_ea_insert,
6551 .dio_declare_delete = osd_index_declare_ea_delete,
6552 .dio_delete = osd_index_ea_delete,
6554 .init = osd_it_ea_init,
6555 .fini = osd_it_ea_fini,
6556 .get = osd_it_ea_get,
6557 .put = osd_it_ea_put,
6558 .next = osd_it_ea_next,
6559 .key = osd_it_ea_key,
6560 .key_size = osd_it_ea_key_size,
6561 .rec = osd_it_ea_rec,
6562 .rec_size = osd_it_ea_rec_size,
6563 .store = osd_it_ea_store,
6564 .load = osd_it_ea_load
6568 static void *osd_key_init(const struct lu_context *ctx,
6569 struct lu_context_key *key)
6571 struct osd_thread_info *info;
6573 OBD_ALLOC_PTR(info);
6575 return ERR_PTR(-ENOMEM);
6577 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6578 if (info->oti_it_ea_buf == NULL)
6581 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
6583 info->oti_hlock = ldiskfs_htree_lock_alloc();
6584 if (info->oti_hlock == NULL)
6590 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6593 return ERR_PTR(-ENOMEM);
6596 static void osd_key_fini(const struct lu_context *ctx,
6597 struct lu_context_key *key, void* data)
6599 struct osd_thread_info *info = data;
6600 struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
6601 struct osd_idmap_cache *idc = info->oti_ins_cache;
6603 if (info->oti_inode != NULL)
6605 if (info->oti_hlock != NULL)
6606 ldiskfs_htree_lock_free(info->oti_hlock);
6607 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6608 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
6609 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
6610 lu_buf_free(&info->oti_big_buf);
6612 LASSERT(info->oti_ins_cache_size > 0);
6613 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
6614 info->oti_ins_cache = NULL;
6615 info->oti_ins_cache_size = 0;
6620 static void osd_key_exit(const struct lu_context *ctx,
6621 struct lu_context_key *key, void *data)
6623 struct osd_thread_info *info = data;
6625 LASSERT(info->oti_r_locks == 0);
6626 LASSERT(info->oti_w_locks == 0);
6627 LASSERT(info->oti_txns == 0);
6630 /* type constructor/destructor: osd_type_init, osd_type_fini */
6631 LU_TYPE_INIT_FINI(osd, &osd_key);
6633 struct lu_context_key osd_key = {
6634 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
6635 .lct_init = osd_key_init,
6636 .lct_fini = osd_key_fini,
6637 .lct_exit = osd_key_exit
6641 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
6642 const char *name, struct lu_device *next)
6644 struct osd_device *osd = osd_dev(d);
6646 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
6647 >= sizeof(osd->od_svname))
6649 return osd_procfs_init(osd, name);
6652 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
6654 struct seq_server_site *ss = osd_seq_site(osd);
6658 if (osd->od_is_ost || osd->od_cl_seq != NULL)
6661 if (unlikely(ss == NULL))
6664 OBD_ALLOC_PTR(osd->od_cl_seq);
6665 if (osd->od_cl_seq == NULL)
6668 rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
6669 osd->od_svname, ss->ss_server_seq);
6671 OBD_FREE_PTR(osd->od_cl_seq);
6672 osd->od_cl_seq = NULL;
6676 if (ss->ss_node_id == 0) {
6677 /* If the OSD on the sequence controller(MDT0), then allocate
6678 * sequence here, otherwise allocate sequence after connected
6679 * to MDT0 (see mdt_register_lwp_callback()). */
6680 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
6681 &osd->od_cl_seq->lcs_space, env);
6687 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
6689 if (osd->od_cl_seq == NULL)
6692 seq_client_fini(osd->od_cl_seq);
6693 OBD_FREE_PTR(osd->od_cl_seq);
6694 osd->od_cl_seq = NULL;
6697 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
6701 /* shutdown quota slave instance associated with the device */
6702 if (o->od_quota_slave != NULL) {
6703 qsd_fini(env, o->od_quota_slave);
6704 o->od_quota_slave = NULL;
6707 osd_fid_fini(env, o);
6712 static void osd_umount(const struct lu_env *env, struct osd_device *o)
6716 if (o->od_mnt != NULL) {
6717 shrink_dcache_sb(osd_sb(o));
6718 osd_sync(env, &o->od_dt_dev);
6727 static int osd_mount(const struct lu_env *env,
6728 struct osd_device *o, struct lustre_cfg *cfg)
6730 const char *name = lustre_cfg_string(cfg, 0);
6731 const char *dev = lustre_cfg_string(cfg, 1);
6733 unsigned long page, s_flags, lmd_flags = 0;
6734 struct page *__page;
6735 struct file_system_type *type;
6736 char *options = NULL;
6738 struct osd_thread_info *info = osd_oti_get(env);
6739 struct lu_fid *fid = &info->oti_fid;
6740 struct inode *inode;
6741 int rc = 0, force_over_256tb = 0;
6744 if (o->od_mnt != NULL)
6747 if (strlen(dev) >= sizeof(o->od_mntdev))
6749 strcpy(o->od_mntdev, dev);
6751 str = lustre_cfg_string(cfg, 2);
6752 s_flags = simple_strtoul(str, NULL, 0);
6753 str = strstr(str, ":");
6755 lmd_flags = simple_strtoul(str + 1, NULL, 0);
6756 opts = lustre_cfg_string(cfg, 3);
6758 if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
6759 CERROR("%s: device %s extents feature is not guaranteed to "
6760 "work on big-endian systems. Use \"bigendian_extents\" "
6761 "mount option to override.\n", name, dev);
6765 if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
6766 CWARN("force_over_128tb option is depricated."
6767 "Filesystems less then 256TB can be created without any"
6768 "force options. Use force_over_256tb option for"
6769 "filesystems greather then 256TB.\n");
6772 if (opts != NULL && strstr(opts, "force_over_256tb") != NULL)
6773 force_over_256tb = 1;
6775 __page = alloc_page(GFP_KERNEL);
6777 GOTO(out, rc = -ENOMEM);
6778 page = (unsigned long)page_address(__page);
6779 options = (char *)page;
6782 /* strip out the options for back compatiblity */
6783 static char *sout[] = {
6790 /* strip out option we processed in osd */
6791 "bigendian_extents",
6792 #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(3,0,53,0)
6793 #warning "remove force_over_128 option"
6795 "force_over_128tb (deprecated)",
6800 strcat(options, opts);
6801 for (rc = 0, str = options; sout[rc]; ) {
6802 char *op = strstr(str, sout[rc]);
6808 if (op == options || *(op - 1) == ',') {
6809 str = op + strlen(sout[rc]);
6810 if (*str == ',' || *str == '\0') {
6811 *str == ',' ? str++ : str;
6812 memmove(op, str, strlen(str) + 1);
6815 for (str = op; *str != ',' && *str != '\0'; str++)
6819 strncat(options, "user_xattr,acl", 14);
6822 /* Glom up mount options */
6823 if (*options != '\0')
6824 strcat(options, ",");
6825 strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
6827 type = get_fs_type("ldiskfs");
6829 CERROR("%s: cannot find ldiskfs module\n", name);
6830 GOTO(out, rc = -ENODEV);
6833 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
6834 module_put(type->owner);
6836 if (IS_ERR(o->od_mnt)) {
6837 rc = PTR_ERR(o->od_mnt);
6839 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
6843 if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) > (64ULL << 30) &&
6844 force_over_256tb == 0) {
6845 CERROR("%s: device %s LDISKFS does not support filesystems "
6846 "greater than 256TB and can cause data corruption. "
6847 "Use \"force_over_256tb\" 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);