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, 2015, 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)
388 struct inode *inode = NULL;
390 /* if we look for an inode withing a running
391 * transaction, then we risk to deadlock */
392 /* osd_dirent_check_repair() breaks this */
393 /*LASSERT(current->journal_info == NULL);*/
395 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
397 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
398 id->oii_ino, PTR_ERR(inode));
399 } else if (id->oii_gen != OSD_OII_NOGEN &&
400 inode->i_generation != id->oii_gen) {
401 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
402 "i_generation = %u\n",
403 id->oii_ino, id->oii_gen, inode->i_generation);
405 inode = ERR_PTR(-ESTALE);
406 } else if (inode->i_nlink == 0) {
407 /* due to parallel readdir and unlink,
408 * we can have dead inode here. */
409 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
411 inode = ERR_PTR(-ESTALE);
412 } else if (is_bad_inode(inode)) {
413 CWARN("%.16s: bad inode: ino = %u\n",
414 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
416 inode = ERR_PTR(-ENOENT);
418 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
419 if (id->oii_gen == OSD_OII_NOGEN)
420 osd_id_gen(id, inode->i_ino, inode->i_generation);
422 /* Do not update file c/mtime in ldiskfs.
423 * NB: we don't have any lock to protect this because we don't
424 * have reference on osd_object now, but contention with
425 * another lookup + attr_set can't happen in the tiny window
426 * between if (...) and set S_NOCMTIME. */
427 if (!(inode->i_flags & S_NOCMTIME))
428 inode->i_flags |= S_NOCMTIME;
433 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
434 handle_t *handle, struct dentry *child,
435 struct inode *inode, struct htree_lock *hlock)
439 rc = __ldiskfs_add_entry(handle, child, inode, hlock);
440 if (rc == -ENOBUFS || rc == -ENOSPC) {
441 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
442 struct inode *parent = child->d_parent->d_inode;
443 struct lu_fid *fid = NULL;
445 rc2 = osd_get_lma(info, parent, child->d_parent, lma);
447 fid = &lma->lma_self_fid;
448 } else if (rc2 == -ENODATA) {
449 if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
450 fid = &info->oti_fid3;
451 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
452 } else if (!osd->od_is_ost && osd->od_index == 0) {
453 fid = &info->oti_fid3;
454 lu_igif_build(fid, parent->i_ino,
455 parent->i_generation);
460 CWARN("%s: directory (inode: %lu, FID: "DFID") %s "
461 "maximum entry limit\n",
462 osd_name(osd), parent->i_ino, PFID(fid),
463 rc == -ENOSPC ? "has reached" : "is approaching");
465 CWARN("%s: directory (inode: %lu, FID: unknown) %s "
466 "maximum entry limit\n",
467 osd_name(osd), parent->i_ino,
468 rc == -ENOSPC ? "has reached" : "is approaching");
470 /* ignore such error now */
479 static struct inode *
480 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
481 struct osd_inode_id *id, struct lu_fid *fid)
483 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
487 inode = osd_iget(info, dev, id);
491 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
493 *fid = lma->lma_self_fid;
494 } else if (rc == -ENODATA) {
495 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
496 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
498 lu_igif_build(fid, inode->i_ino, inode->i_generation);
506 static struct inode *osd_iget_check(struct osd_thread_info *info,
507 struct osd_device *dev,
508 const struct lu_fid *fid,
509 struct osd_inode_id *id,
516 /* The cached OI mapping is trustable. If we cannot locate the inode
517 * via the cached OI mapping, then return the failure to the caller
518 * directly without further OI checking. */
520 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
523 if (cached || (rc != -ENOENT && rc != -ESTALE)) {
524 CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
533 if (is_bad_inode(inode)) {
536 CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
544 if (id->oii_gen != OSD_OII_NOGEN &&
545 inode->i_generation != id->oii_gen) {
548 CDEBUG(D_INODE, "unmatched inode: ino = %u, "
549 "oii_gen = %u, i_generation = %u\n",
550 id->oii_ino, id->oii_gen, inode->i_generation);
558 if (inode->i_nlink == 0) {
561 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
569 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
573 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
575 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
576 /* XXX: There are four possible cases:
578 * Backup/restore caused the OI invalid.
580 * Someone unlinked the object but NOT removed
581 * the OI mapping, such as mount target device
582 * as ldiskfs, and modify something directly.
584 * Someone just removed the object between the
585 * former oi_lookup and the iget. It is normal.
586 * 4. Other failure cases.
588 * Generally, when the device is mounted, it will
589 * auto check whether the system is restored from
590 * file-level backup or not. We trust such detect
591 * to distinguish the 1st case from the 2nd case:
592 * if the OI files are consistent but may contain
593 * stale OI mappings because of case 2, if iget()
594 * returns -ENOENT or -ESTALE, then it should be
597 /* If the OI mapping was in OI file before the
598 * osd_iget_check(), but now, it is disappear,
599 * then it must be removed by race. That is a
600 * normal race case. */
603 if ((!IS_ERR(inode) && inode->i_generation != 0 &&
604 inode->i_generation == id->oii_gen) ||
605 (IS_ERR(inode) && !(dev->od_scrub.os_file.sf_flags &
611 if (id->oii_gen == OSD_OII_NOGEN)
612 osd_id_gen(id, inode->i_ino, inode->i_generation);
614 /* Do not update file c/mtime in ldiskfs.
615 * NB: we don't have any lock to protect this because we don't
616 * have reference on osd_object now, but contention with
617 * another lookup + attr_set can't happen in the tiny window
618 * between if (...) and set S_NOCMTIME. */
619 if (!(inode->i_flags & S_NOCMTIME))
620 inode->i_flags |= S_NOCMTIME;
637 * \retval +v: new filter_fid, does not contain self-fid
638 * \retval 0: filter_fid_old, contains self-fid
639 * \retval -v: other failure cases
641 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
642 struct dentry *dentry, struct lu_fid *fid)
644 struct filter_fid_old *ff = &info->oti_ff;
645 struct ost_id *ostid = &info->oti_ostid;
648 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
649 if (rc == sizeof(*ff)) {
651 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
652 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
653 /* XXX: use 0 as the index for compatibility, the caller will
654 * handle index related issues when necessarry. */
655 ostid_to_fid(fid, ostid, 0);
656 } else if (rc == sizeof(struct filter_fid)) {
658 } else if (rc >= 0) {
665 static int osd_lma_self_repair(struct osd_thread_info *info,
666 struct osd_device *osd, struct inode *inode,
667 const struct lu_fid *fid, __u32 compat)
672 LASSERT(current->journal_info == NULL);
674 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
675 osd_dto_credits_noquota[DTO_XATTR_SET]);
678 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
683 rc = osd_ea_fid_set(info, inode, fid, compat, 0);
685 CWARN("%s: cannot self repair the LMA: rc = %d\n",
687 ldiskfs_journal_stop(jh);
691 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
693 struct osd_thread_info *info = osd_oti_get(env);
694 struct osd_device *osd = osd_obj2dev(obj);
695 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
696 struct inode *inode = obj->oo_inode;
697 struct dentry *dentry = &info->oti_obj_dentry;
698 struct lu_fid *fid = NULL;
699 const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
703 CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
704 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
705 info->oti_mdt_attrs_old, LMA_OLD_SIZE);
706 if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
707 fid = &lma->lma_self_fid;
708 rc = osd_get_idif(info, inode, dentry, fid);
709 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
710 /* For the given OST-object, if it has neither LMA nor
711 * FID in XATTR_NAME_FID, then the given FID (which is
712 * contained in the @obj, from client RPC for locating
713 * the OST-object) is trusted. We use it to generate
715 osd_lma_self_repair(info, osd, inode, rfid,
726 lustre_lma_swab(lma);
727 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
728 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
729 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
730 "fid = "DFID", ino = %lu\n", osd_name(osd),
731 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
732 PFID(rfid), inode->i_ino);
735 fid = &lma->lma_self_fid;
739 if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
740 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
741 struct ost_id *oi = &info->oti_ostid;
742 struct lu_fid *fid1 = &info->oti_fid3;
743 __u32 idx = fid_idif_ost_idx(rfid);
745 /* For old IDIF, the OST index is not part of the IDIF,
746 * Means that different OSTs may have the same IDIFs.
747 * Under such case, we need to make some compatible
748 * check to make sure to trigger OI scrub properly. */
749 if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
750 /* Given @rfid is new, LMA is old. */
751 fid_to_ostid(fid, oi);
752 ostid_to_fid(fid1, oi, idx);
753 if (lu_fid_eq(fid1, rfid)) {
754 if (osd->od_index_in_idif)
755 osd_lma_self_repair(info, osd,
769 struct osd_check_lmv_buf {
770 #ifdef HAVE_DIR_CONTEXT
771 /* please keep it as first member */
772 struct dir_context ctx;
774 struct osd_thread_info *oclb_info;
775 struct osd_device *oclb_dev;
776 struct osd_idmap_cache *oclb_oic;
780 * It is called internally by ->readdir() to filter out the
781 * local slave object's FID of the striped directory.
783 * \retval 1 found the local slave's FID
784 * \retval 0 continue to check next item
785 * \retval -ve for failure
787 #ifdef HAVE_FILLDIR_USE_CTX
788 static int osd_stripe_dir_filldir(struct dir_context *buf,
790 static int osd_stripe_dir_filldir(void *buf,
792 const char *name, int namelen,
793 loff_t offset, __u64 ino, unsigned d_type)
795 struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
796 struct osd_thread_info *oti = oclb->oclb_info;
797 struct lu_fid *fid = &oti->oti_fid3;
798 struct osd_inode_id *id = &oti->oti_id3;
799 struct osd_device *dev = oclb->oclb_dev;
800 struct osd_idmap_cache *oic = oclb->oclb_oic;
808 sscanf(name + 1, SFID, RFID(fid));
809 if (!fid_is_sane(fid))
812 if (osd_remote_fid(oti->oti_env, dev, fid))
815 osd_id_gen(id, ino, OSD_OII_NOGEN);
816 inode = osd_iget(oti, dev, id);
818 return PTR_ERR(inode);
821 osd_add_oi_cache(oti, dev, id, fid);
825 rc = osd_oii_insert(dev, oic, true);
827 return rc == 0 ? 1 : rc;
830 /* When lookup item under striped directory, we need to locate the master
831 * MDT-object of the striped directory firstly, then the client will send
832 * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
833 * and the item's name. If the system is restored from MDT file level backup,
834 * then before the OI scrub completely built the OI files, the OI mappings of
835 * the master MDT-object and slave MDT-object may be invalid. Usually, it is
836 * not a problem for the master MDT-object. Because when locate the master
837 * MDT-object, we will do name based lookup (for the striped directory itself)
838 * firstly, during such process we can setup the correct OI mapping for the
839 * master MDT-object. But it will be trouble for the slave MDT-object. Because
840 * the client will not trigger name based lookup on the MDT to locate the slave
841 * MDT-object before locating item under the striped directory, then when
842 * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
843 * is invalid and does not know what the right OI mapping is, then the MDT has
844 * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
845 * the OI file, related OI mapping is unknown yet, please try again later. And
846 * then client will re-try the RPC again and again until related OI mapping has
847 * been updated. That is quite inefficient.
849 * To resolve above trouble, we will handle it as the following two cases:
851 * 1) The slave MDT-object and the master MDT-object are on different MDTs.
852 * It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
853 * is linked under /REMOTE_PARENT_DIR with the name of its FID string.
854 * We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
855 * directly. Please check osd_fid_lookup().
857 * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
858 * Under such case, during lookup the master MDT-object, we will lookup the
859 * slave MDT-object via readdir against the master MDT-object, because the
860 * slave MDT-objects information are stored as sub-directories with the name
861 * "${FID}:${index}". Then when find the local slave MDT-object, its OI
862 * mapping will be recorded. Then subsequent osd_fid_lookup() will know
863 * the correct OI mapping for the slave MDT-object. */
864 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
865 struct inode *inode, struct osd_idmap_cache *oic)
867 struct lu_buf *buf = &oti->oti_big_buf;
868 struct dentry *dentry = &oti->oti_obj_dentry;
869 struct file *filp = &oti->oti_file;
870 const struct file_operations *fops;
871 struct lmv_mds_md_v1 *lmv1;
872 struct osd_check_lmv_buf oclb = {
873 #ifdef HAVE_DIR_CONTEXT
874 .ctx.actor = osd_stripe_dir_filldir,
884 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
887 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
889 lu_buf_realloc(buf, rc);
890 if (buf->lb_buf == NULL)
891 GOTO(out, rc = -ENOMEM);
897 if (unlikely(rc == 0 || rc == -ENODATA))
903 if (unlikely(buf->lb_buf == NULL)) {
904 lu_buf_realloc(buf, rc);
905 if (buf->lb_buf == NULL)
906 GOTO(out, rc = -ENOMEM);
912 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
916 dentry->d_inode = inode;
917 dentry->d_sb = inode->i_sb;
919 filp->f_path.dentry = dentry;
920 filp->f_mode = FMODE_64BITHASH;
921 filp->f_mapping = inode->i_mapping;
923 filp->private_data = NULL;
924 set_file_inode(filp, inode);
926 #ifdef HAVE_DIR_CONTEXT
927 oclb.ctx.pos = filp->f_pos;
928 rc = fops->iterate(filp, &oclb.ctx);
929 filp->f_pos = oclb.ctx.pos;
931 rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir);
933 fops->release(inode, filp);
937 CDEBUG(D_LFSCK, "%.16s: fail to check LMV EA, inode = %lu/%u,"
939 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
940 inode->i_ino, inode->i_generation,
941 PFID(&oic->oic_fid), rc);
948 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
949 const struct lu_fid *fid,
950 const struct lu_object_conf *conf)
952 struct osd_thread_info *info;
953 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
954 struct osd_device *dev;
955 struct osd_idmap_cache *oic;
956 struct osd_inode_id *id;
957 struct osd_inode_id *tid;
958 struct inode *inode = NULL;
959 struct osd_scrub *scrub;
960 struct scrub_file *sf;
961 __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT |
971 LINVRNT(osd_invariant(obj));
972 LASSERT(obj->oo_inode == NULL);
973 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
976 scrub = &dev->od_scrub;
977 sf = &scrub->os_file;
978 info = osd_oti_get(env);
980 oic = &info->oti_cache;
982 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
985 /* For the object is created as locking anchor, or for the object to
986 * be created on disk. No need to osd_oi_lookup() at here because FID
987 * shouldn't never be re-used, if it's really a duplicate FID from
988 * unexpected reason, we should be able to detect it later by calling
989 * do_create->osd_oi_insert(). */
990 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
991 GOTO(out, result = 0);
993 /* Search order: 1. per-thread cache. */
994 if (lu_fid_eq(fid, &oic->oic_fid) &&
995 likely(oic->oic_dev == dev)) {
1001 if (!list_empty(&scrub->os_inconsistent_items)) {
1002 /* Search order: 2. OI scrub pending list. */
1003 result = osd_oii_lookup(dev, fid, id);
1009 /* Search order: 3. OI files. */
1010 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1011 if (result == -ENOENT) {
1012 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1013 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1014 !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
1016 GOTO(out, result = 0);
1025 inode = osd_iget_check(info, dev, fid, id, cached);
1026 if (IS_ERR(inode)) {
1027 result = PTR_ERR(inode);
1028 if (result == -ENOENT || result == -ESTALE)
1029 GOTO(out, result = 0);
1031 if (result == -EREMCHG) {
1034 /* We still have chance to get the valid inode: for the
1035 * object which is referenced by remote name entry, the
1036 * object on the local MDT will be linked under the dir
1037 * of "/REMOTE_PARENT_DIR" with its FID string as name.
1039 * We do not know whether the object for the given FID
1040 * is referenced by some remote name entry or not, and
1041 * especially for DNE II, a multiple-linked object may
1042 * have many name entries reside on many MDTs.
1044 * To simplify the operation, OSD will not distinguish
1045 * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1046 * only happened for the RPC from other MDT during the
1047 * OI scrub, or for the client side RPC with FID only,
1048 * such as FID to path, or from old connected client. */
1050 !fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1051 rc1 = osd_lookup_in_remote_parent(info, dev,
1056 flags |= SS_AUTO_PARTIAL;
1057 flags &= ~SS_AUTO_FULL;
1062 if (thread_is_running(&scrub->os_thread)) {
1063 if (scrub->os_partial_scan &&
1064 !scrub->os_in_join) {
1067 if (inode != NULL && !IS_ERR(inode)) {
1070 osd_add_oi_cache(info, dev, id,
1072 osd_oii_insert(dev, oic, true);
1074 result = -EINPROGRESS;
1077 } else if (!dev->od_noscrub) {
1080 rc1 = osd_scrub_start(dev, flags);
1081 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
1082 "for the "DFID" with flags 0x%x,"
1083 " rc = %d\n", osd_name(dev),
1084 PFID(fid), flags, rc1);
1085 if (rc1 == 0 || rc1 == -EALREADY) {
1086 if (inode != NULL && !IS_ERR(inode)) {
1089 osd_add_oi_cache(info, dev, id,
1091 osd_oii_insert(dev, oic, true);
1093 result = -EINPROGRESS;
1103 if (inode == NULL || IS_ERR(inode))
1105 } else if (remote) {
1109 obj->oo_inode = inode;
1110 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
1112 result = osd_check_lma(env, obj);
1116 tid = &info->oti_id3;
1119 if (result == -ENODATA) {
1121 /* The current OI mapping is from the OI file,
1122 * since the inode has been found via
1123 * osd_iget_check(), no need recheck OI. */
1126 result = osd_oi_lookup(info, dev, fid, tid, OI_CHECK_FLD);
1128 LASSERTF(tid->oii_ino == id->oii_ino &&
1129 tid->oii_gen == id->oii_gen,
1130 "OI mapping changed(1): %u/%u => %u/%u",
1131 tid->oii_ino, tid->oii_gen,
1132 id->oii_ino, id->oii_gen);
1134 LASSERTF(tid->oii_ino == inode->i_ino &&
1135 tid->oii_gen == inode->i_generation,
1136 "locate wrong inode(1): %u/%u => %ld/%u",
1137 tid->oii_ino, tid->oii_gen,
1138 inode->i_ino, inode->i_generation);
1140 /* "result == 0" means the cached OI mapping is still in
1141 * the OI file, so the target the inode is valid. */
1145 /* "result == -ENOENT" means that the OI mappinghas been removed
1146 * by race, the target inode belongs to other object.
1148 * Others error can be returned directly. */
1149 if (result == -ENOENT)
1153 saved_ino = inode->i_ino;
1154 saved_gen = inode->i_generation;
1157 obj->oo_inode = NULL;
1159 if (result != -EREMCHG)
1163 /* The current OI mapping is from the OI file,
1164 * since the inode has been found via
1165 * osd_iget_check(), no need recheck OI. */
1168 result = osd_oi_lookup(info, dev, fid, tid, OI_CHECK_FLD);
1169 /* "result == -ENOENT" means the cached OI mapping has been removed from
1170 * the OI file by race, above target inode belongs to other object.
1172 * Others error can be returned directly. */
1174 GOTO(out, result = (result == -ENOENT ? 0 : result));
1176 LASSERTF(tid->oii_ino == id->oii_ino && tid->oii_gen == id->oii_gen,
1177 "OI mapping changed(2): %u/%u => %u/%u",
1178 tid->oii_ino, tid->oii_gen, id->oii_ino, id->oii_gen);
1180 LASSERTF(tid->oii_ino == saved_ino && tid->oii_gen == saved_gen,
1181 "locate wrong inode(2): %u/%u => %u/%u",
1182 tid->oii_ino, tid->oii_gen, saved_ino, saved_gen);
1187 obj->oo_compat_dot_created = 1;
1188 obj->oo_compat_dotdot_created = 1;
1190 if (S_ISDIR(inode->i_mode) &&
1191 (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1192 osd_check_lmv(info, dev, inode, oic);
1195 GOTO(out, result = 0);
1197 LASSERT(obj->oo_hl_head == NULL);
1198 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1199 if (obj->oo_hl_head == NULL) {
1200 obj->oo_inode = NULL;
1202 GOTO(out, result = -ENOMEM);
1204 GOTO(out, result = 0);
1207 if (result != 0 && cached)
1208 fid_zero(&oic->oic_fid);
1210 LINVRNT(osd_invariant(obj));
1215 * Concurrency: shouldn't matter.
1217 static void osd_object_init0(struct osd_object *obj)
1219 LASSERT(obj->oo_inode != NULL);
1220 obj->oo_dt.do_body_ops = &osd_body_ops;
1221 obj->oo_dt.do_lu.lo_header->loh_attr |=
1222 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1226 * Concurrency: no concurrent access is possible that early in object
1229 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1230 const struct lu_object_conf *conf)
1232 struct osd_object *obj = osd_obj(l);
1235 LINVRNT(osd_invariant(obj));
1237 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1238 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1239 l->lo_header->loh_attr |= LOHA_EXISTS;
1243 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1244 obj->oo_dt.do_body_ops = &osd_body_ops_new;
1245 if (result == 0 && obj->oo_inode != NULL) {
1246 struct osd_thread_info *oti = osd_oti_get(env);
1247 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
1249 osd_object_init0(obj);
1250 result = osd_get_lma(oti, obj->oo_inode,
1251 &oti->oti_obj_dentry, lma);
1253 /* Convert LMAI flags to lustre LMA flags
1254 * and cache it to oo_lma_flags */
1256 lma_to_lustre_flags(lma->lma_incompat);
1257 } else if (result == -ENODATA) {
1262 LINVRNT(osd_invariant(obj));
1266 /* The first part of oxe_buf is xattr name, and is '\0' terminated.
1267 * The left part is for value, binary mode. */
1268 struct osd_xattr_entry {
1269 struct list_head oxe_list;
1273 struct rcu_head oxe_rcu;
1277 static int osd_oxc_get(struct osd_object *obj, const char *name,
1280 struct osd_xattr_entry *tmp;
1281 struct osd_xattr_entry *oxe = NULL;
1282 size_t namelen = strlen(name);
1287 list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1288 if (namelen == tmp->oxe_namelen &&
1289 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1296 GOTO(out, rc = -ENOENT);
1298 if (!oxe->oxe_exist)
1299 GOTO(out, rc = -ENODATA);
1302 rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1305 if (buf->lb_buf == NULL)
1308 if (buf->lb_len < rc)
1309 GOTO(out, rc = -ERANGE);
1311 memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1319 static void osd_oxc_free(struct rcu_head *head)
1321 struct osd_xattr_entry *oxe;
1323 oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1324 OBD_FREE(oxe, oxe->oxe_len);
1327 static void osd_oxc_add(struct osd_object *obj, const char *name,
1328 const char *buf, int buflen)
1330 struct osd_xattr_entry *oxe;
1331 struct osd_xattr_entry *old = NULL;
1332 struct osd_xattr_entry *tmp;
1333 size_t namelen = strlen(name);
1334 size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1336 OBD_ALLOC(oxe, len);
1340 INIT_LIST_HEAD(&oxe->oxe_list);
1342 oxe->oxe_namelen = namelen;
1343 memcpy(oxe->oxe_buf, name, namelen);
1345 LASSERT(buf != NULL);
1346 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1347 oxe->oxe_exist = true;
1349 oxe->oxe_exist = false;
1352 /* this should be rarely called, just remove old and add new */
1353 spin_lock(&obj->oo_guard);
1354 list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1355 if (namelen == tmp->oxe_namelen &&
1356 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1362 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1363 call_rcu(&old->oxe_rcu, osd_oxc_free);
1365 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1367 spin_unlock(&obj->oo_guard);
1370 static void osd_oxc_del(struct osd_object *obj, const char *name)
1372 struct osd_xattr_entry *oxe;
1373 size_t namelen = strlen(name);
1375 spin_lock(&obj->oo_guard);
1376 list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1377 if (namelen == oxe->oxe_namelen &&
1378 strncmp(name, oxe->oxe_buf, namelen) == 0) {
1379 list_del_rcu(&oxe->oxe_list);
1380 call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1384 spin_unlock(&obj->oo_guard);
1387 static void osd_oxc_fini(struct osd_object *obj)
1389 struct osd_xattr_entry *oxe, *next;
1391 list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1392 list_del(&oxe->oxe_list);
1393 OBD_FREE(oxe, oxe->oxe_len);
1398 * Concurrency: no concurrent access is possible that late in object
1401 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1403 struct osd_object *obj = osd_obj(l);
1405 LINVRNT(osd_invariant(obj));
1408 dt_object_fini(&obj->oo_dt);
1409 if (obj->oo_hl_head != NULL)
1410 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1415 * Concurrency: no concurrent access is possible that late in object
1418 static void osd_index_fini(struct osd_object *o)
1420 struct iam_container *bag;
1422 if (o->oo_dir != NULL) {
1423 bag = &o->oo_dir->od_container;
1424 if (o->oo_inode != NULL) {
1425 if (bag->ic_object == o->oo_inode)
1426 iam_container_fini(bag);
1428 OBD_FREE_PTR(o->oo_dir);
1434 * Concurrency: no concurrent access is possible that late in object
1435 * life-cycle (for all existing callers, that is. New callers have to provide
1436 * their own locking.)
1438 static int osd_inode_unlinked(const struct inode *inode)
1440 return inode->i_nlink == 0;
1444 OSD_TXN_OI_DELETE_CREDITS = 20,
1445 OSD_TXN_INODE_DELETE_CREDITS = 20
1452 #if OSD_THANDLE_STATS
1454 * Set time when the handle is allocated
1456 static void osd_th_alloced(struct osd_thandle *oth)
1458 oth->oth_alloced = cfs_time_current();
1462 * Set time when the handle started
1464 static void osd_th_started(struct osd_thandle *oth)
1466 oth->oth_started = cfs_time_current();
1470 * Helper function to convert time interval to microseconds packed in
1473 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
1477 cfs_duration_usec(cfs_time_sub(end, start), &val);
1478 return val.tv_sec * 1000000 + val.tv_usec;
1482 * Check whether the we deal with this handle for too long.
1484 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1485 cfs_time_t alloced, cfs_time_t started,
1488 cfs_time_t now = cfs_time_current();
1490 LASSERT(dev != NULL);
1492 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1493 interval_to_usec(alloced, started));
1494 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1495 interval_to_usec(started, closed));
1496 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1497 interval_to_usec(closed, now));
1499 if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
1500 CWARN("transaction handle %p was open for too long: "
1501 "now "CFS_TIME_T" ,"
1502 "alloced "CFS_TIME_T" ,"
1503 "started "CFS_TIME_T" ,"
1504 "closed "CFS_TIME_T"\n",
1505 oth, now, alloced, started, closed);
1506 libcfs_debug_dumpstack(NULL);
1510 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
1512 cfs_time_t __closed = cfs_time_current(); \
1513 cfs_time_t __alloced = oth->oth_alloced; \
1514 cfs_time_t __started = oth->oth_started; \
1517 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
1520 #else /* OSD_THANDLE_STATS */
1522 #define osd_th_alloced(h) do {} while(0)
1523 #define osd_th_started(h) do {} while(0)
1524 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
1526 #endif /* OSD_THANDLE_STATS */
1529 * Concurrency: doesn't access mutable data.
1531 static int osd_param_is_not_sane(const struct osd_device *dev,
1532 const struct thandle *th)
1534 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1536 return oh->ot_credits > osd_transaction_size(dev);
1540 * Concurrency: shouldn't matter.
1542 static void osd_trans_commit_cb(struct super_block *sb,
1543 struct ldiskfs_journal_cb_entry *jcb, int error)
1545 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1546 struct thandle *th = &oh->ot_super;
1547 struct lu_device *lud = &th->th_dev->dd_lu_dev;
1548 struct dt_txn_commit_cb *dcb, *tmp;
1550 LASSERT(oh->ot_handle == NULL);
1553 CERROR("transaction @0x%p commit error: %d\n", th, error);
1555 dt_txn_hook_commit(th);
1557 /* call per-transaction callbacks if any */
1558 list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1560 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1561 "commit callback entry: magic=%x name='%s'\n",
1562 dcb->dcb_magic, dcb->dcb_name);
1563 list_del_init(&dcb->dcb_linkage);
1564 dcb->dcb_func(NULL, th, dcb, error);
1567 lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1571 lu_context_exit(&th->th_ctx);
1572 lu_context_fini(&th->th_ctx);
1576 #ifndef HAVE_SB_START_WRITE
1577 # define sb_start_write(sb) do {} while (0)
1578 # define sb_end_write(sb) do {} while (0)
1581 static struct thandle *osd_trans_create(const struct lu_env *env,
1582 struct dt_device *d)
1584 struct osd_thread_info *oti = osd_oti_get(env);
1585 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1586 struct osd_thandle *oh;
1590 /* on pending IO in this thread should left from prev. request */
1591 LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1593 sb_start_write(osd_sb(osd_dt_dev(d)));
1595 OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
1597 oh->ot_quota_trans = &oti->oti_quota_trans;
1598 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1602 th->th_tags = LCT_TX_HANDLE;
1604 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1605 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1608 memset(oti->oti_declare_ops, 0,
1609 sizeof(oti->oti_declare_ops));
1610 memset(oti->oti_declare_ops_cred, 0,
1611 sizeof(oti->oti_declare_ops_cred));
1612 memset(oti->oti_declare_ops_used, 0,
1613 sizeof(oti->oti_declare_ops_used));
1615 sb_end_write(osd_sb(osd_dt_dev(d)));
1616 th = ERR_PTR(-ENOMEM);
1621 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1623 struct osd_thread_info *oti = osd_oti_get(env);
1624 struct osd_thandle *oh;
1626 oh = container_of0(th, struct osd_thandle, ot_super);
1627 LASSERT(oh != NULL);
1629 CWARN(" create: %u/%u/%u, destroy: %u/%u/%u\n",
1630 oti->oti_declare_ops[OSD_OT_CREATE],
1631 oti->oti_declare_ops_cred[OSD_OT_CREATE],
1632 oti->oti_declare_ops_used[OSD_OT_CREATE],
1633 oti->oti_declare_ops[OSD_OT_DESTROY],
1634 oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1635 oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1636 CWARN(" attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1637 oti->oti_declare_ops[OSD_OT_ATTR_SET],
1638 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1639 oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1640 oti->oti_declare_ops[OSD_OT_XATTR_SET],
1641 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1642 oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1643 CWARN(" write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1644 oti->oti_declare_ops[OSD_OT_WRITE],
1645 oti->oti_declare_ops_cred[OSD_OT_WRITE],
1646 oti->oti_declare_ops_used[OSD_OT_WRITE],
1647 oti->oti_declare_ops[OSD_OT_PUNCH],
1648 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1649 oti->oti_declare_ops_used[OSD_OT_PUNCH],
1650 oti->oti_declare_ops[OSD_OT_QUOTA],
1651 oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1652 oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1653 CWARN(" insert: %u/%u/%u, delete: %u/%u/%u\n",
1654 oti->oti_declare_ops[OSD_OT_INSERT],
1655 oti->oti_declare_ops_cred[OSD_OT_INSERT],
1656 oti->oti_declare_ops_used[OSD_OT_INSERT],
1657 oti->oti_declare_ops[OSD_OT_DELETE],
1658 oti->oti_declare_ops_cred[OSD_OT_DELETE],
1659 oti->oti_declare_ops_used[OSD_OT_DELETE]);
1660 CWARN(" ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1661 oti->oti_declare_ops[OSD_OT_REF_ADD],
1662 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1663 oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1664 oti->oti_declare_ops[OSD_OT_REF_DEL],
1665 oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1666 oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1670 * Concurrency: shouldn't matter.
1672 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1675 struct osd_thread_info *oti = osd_oti_get(env);
1676 struct osd_device *dev = osd_dt_dev(d);
1678 struct osd_thandle *oh;
1683 LASSERT(current->journal_info == NULL);
1685 oh = container_of0(th, struct osd_thandle, ot_super);
1686 LASSERT(oh != NULL);
1687 LASSERT(oh->ot_handle == NULL);
1689 rc = dt_txn_hook_start(env, d, th);
1693 if (unlikely(osd_param_is_not_sane(dev, th))) {
1694 static unsigned long last_printed;
1695 static int last_credits;
1697 /* don't make noise on a tiny testing systems
1698 * actual credits misuse will be caught anyway */
1699 if (last_credits != oh->ot_credits &&
1700 time_after(jiffies, last_printed +
1701 msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1702 osd_transaction_size(dev) > 512) {
1703 CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1704 oh->ot_credits, osd_transaction_size(dev));
1705 osd_trans_dump_creds(env, th);
1706 libcfs_debug_dumpstack(NULL);
1707 last_credits = oh->ot_credits;
1708 last_printed = jiffies;
1710 /* XXX Limit the credits to 'max_transaction_buffers', and
1711 * let the underlying filesystem to catch the error if
1712 * we really need so many credits.
1714 * This should be removed when we can calculate the
1715 * credits precisely. */
1716 oh->ot_credits = osd_transaction_size(dev);
1720 * XXX temporary stuff. Some abstraction layer should
1723 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1727 LASSERT(oti->oti_txns == 0);
1728 lu_context_init(&th->th_ctx, th->th_tags);
1729 lu_context_enter(&th->th_ctx);
1731 lu_device_get(&d->dd_lu_dev);
1732 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1743 static int osd_seq_exists(const struct lu_env *env,
1744 struct osd_device *osd, u64 seq)
1746 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1747 struct seq_server_site *ss = osd_seq_site(osd);
1751 LASSERT(ss != NULL);
1752 LASSERT(ss->ss_server_fld != NULL);
1754 rc = osd_fld_lookup(env, osd, seq, range);
1757 CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1758 osd_name(osd), seq, rc);
1762 RETURN(ss->ss_node_id == range->lsr_index);
1765 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1767 struct dt_txn_commit_cb *dcb;
1768 struct dt_txn_commit_cb *tmp;
1770 /* call per-transaction stop callbacks if any */
1771 list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1773 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1774 "commit callback entry: magic=%x name='%s'\n",
1775 dcb->dcb_magic, dcb->dcb_name);
1776 list_del_init(&dcb->dcb_linkage);
1777 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1782 * Concurrency: shouldn't matter.
1784 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1787 int rc = 0, remove_agents = 0;
1788 struct osd_thandle *oh;
1789 struct osd_thread_info *oti = osd_oti_get(env);
1790 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1791 struct osd_device *osd = osd_dt_dev(th->th_dev);
1792 struct qsd_instance *qsd = osd->od_quota_slave;
1793 struct lquota_trans *qtrans;
1796 oh = container_of0(th, struct osd_thandle, ot_super);
1798 /* reset OI cache for safety */
1799 oti->oti_ins_cache_used = 0;
1801 remove_agents = oh->ot_remove_agents;
1803 qtrans = oh->ot_quota_trans;
1804 oh->ot_quota_trans = NULL;
1806 if (oh->ot_handle != NULL) {
1807 handle_t *hdl = oh->ot_handle;
1810 * add commit callback
1811 * notice we don't do this in osd_trans_start()
1812 * as underlying transaction can change during truncate
1814 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1817 LASSERT(oti->oti_txns == 1);
1820 rc = dt_txn_hook_stop(env, th);
1822 CERROR("%s: failed in transaction hook: rc = %d\n",
1825 osd_trans_stop_cb(oh, rc);
1826 /* hook functions might modify th_sync */
1827 hdl->h_sync = th->th_sync;
1829 oh->ot_handle = NULL;
1830 OSD_CHECK_SLOW_TH(oh, osd, rc = ldiskfs_journal_stop(hdl));
1832 CERROR("%s: failed to stop transaction: rc = %d\n",
1835 osd_trans_stop_cb(oh, th->th_result);
1839 /* inform the quota slave device that the transaction is stopping */
1840 qsd_op_end(env, qsd, qtrans);
1842 /* as we want IO to journal and data IO be concurrent, we don't block
1843 * awaiting data IO completion in osd_do_bio(), instead we wait here
1844 * once transaction is submitted to the journal. all reqular requests
1845 * don't do direct IO (except read/write), thus this wait_event becomes
1848 * IMPORTANT: we have to wait till any IO submited by the thread is
1849 * completed otherwise iobuf may be corrupted by different request
1851 wait_event(iobuf->dr_wait,
1852 atomic_read(&iobuf->dr_numreqs) == 0);
1853 osd_fini_iobuf(osd, iobuf);
1855 rc = iobuf->dr_error;
1857 if (unlikely(remove_agents != 0))
1858 osd_process_scheduled_agent_removals(env, osd);
1860 sb_end_write(osd_sb(osd));
1865 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1867 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1870 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1871 LASSERT(&dcb->dcb_func != NULL);
1872 if (dcb->dcb_flags & DCB_TRANS_STOP)
1873 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1875 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1881 * Called just before object is freed. Releases all resources except for
1882 * object itself (that is released by osd_object_free()).
1884 * Concurrency: no concurrent access is possible that late in object
1887 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1889 struct osd_object *obj = osd_obj(l);
1890 struct inode *inode = obj->oo_inode;
1892 LINVRNT(osd_invariant(obj));
1895 * If object is unlinked remove fid->ino mapping from object index.
1898 osd_index_fini(obj);
1899 if (inode != NULL) {
1900 struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave;
1901 qid_t uid = i_uid_read(inode);
1902 qid_t gid = i_gid_read(inode);
1905 obj->oo_inode = NULL;
1908 struct osd_thread_info *info = osd_oti_get(env);
1909 struct lquota_id_info *qi = &info->oti_qi;
1911 /* Release granted quota to master if necessary */
1912 qi->lqi_id.qid_uid = uid;
1913 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1915 qi->lqi_id.qid_uid = gid;
1916 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1922 * Concurrency: ->loo_object_release() is called under site spin-lock.
1924 static void osd_object_release(const struct lu_env *env,
1925 struct lu_object *l)
1927 struct osd_object *o = osd_obj(l);
1928 /* nobody should be releasing a non-destroyed object with nlink=0
1929 * the API allows this, but ldiskfs doesn't like and then report
1930 * this inode as deleted */
1931 if (unlikely(!o->oo_destroyed && o->oo_inode && o->oo_inode->i_nlink == 0))
1936 * Concurrency: shouldn't matter.
1938 static int osd_object_print(const struct lu_env *env, void *cookie,
1939 lu_printer_t p, const struct lu_object *l)
1941 struct osd_object *o = osd_obj(l);
1942 struct iam_descr *d;
1944 if (o->oo_dir != NULL)
1945 d = o->oo_dir->od_container.ic_descr;
1948 return (*p)(env, cookie,
1949 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1951 o->oo_inode ? o->oo_inode->i_ino : 0UL,
1952 o->oo_inode ? o->oo_inode->i_generation : 0,
1953 d ? d->id_ops->id_name : "plain");
1957 * Concurrency: shouldn't matter.
1959 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1960 struct obd_statfs *sfs)
1962 struct osd_device *osd = osd_dt_dev(d);
1963 struct super_block *sb = osd_sb(osd);
1964 struct kstatfs *ksfs;
1968 if (unlikely(osd->od_mnt == NULL))
1969 return -EINPROGRESS;
1971 /* osd_lproc.c call this without env, allocate ksfs for that case */
1972 if (unlikely(env == NULL)) {
1973 OBD_ALLOC_PTR(ksfs);
1977 ksfs = &osd_oti_get(env)->oti_ksfs;
1980 result = sb->s_op->statfs(sb->s_root, ksfs);
1984 statfs_pack(sfs, ksfs);
1985 if (unlikely(sb->s_flags & MS_RDONLY))
1986 sfs->os_state |= OS_STATE_READONLY;
1987 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb,
1988 LDISKFS_FEATURE_INCOMPAT_EXTENTS))
1989 sfs->os_maxbytes = sb->s_maxbytes;
1991 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
1994 * Reserve some space so to avoid fragmenting the filesystem too much.
1995 * Fragmentation not only impacts performance, but can also increase
1996 * metadata overhead significantly, causing grant calculation to be
1999 * Reserve 0.78% of total space, at least 8MB for small filesystems.
2001 CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2002 reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2003 if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2004 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2006 sfs->os_blocks -= reserved;
2007 sfs->os_bfree -= min(reserved, sfs->os_bfree);
2008 sfs->os_bavail -= min(reserved, sfs->os_bavail);
2011 if (unlikely(env == NULL))
2017 * Estimate space needed for file creations. We assume the largest filename
2018 * which is 2^64 - 1, hence a filename of 20 chars.
2019 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2021 #ifdef __LDISKFS_DIR_REC_LEN
2022 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2024 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2028 * Concurrency: doesn't access mutable data.
2030 static void osd_conf_get(const struct lu_env *env,
2031 const struct dt_device *dev,
2032 struct dt_device_param *param)
2034 struct super_block *sb = osd_sb(osd_dt_dev(dev));
2038 * XXX should be taken from not-yet-existing fs abstraction layer.
2040 param->ddp_max_name_len = LDISKFS_NAME_LEN;
2041 param->ddp_max_nlink = LDISKFS_LINK_MAX;
2042 param->ddp_symlink_max = sb->s_blocksize;
2043 param->ddp_mount_type = LDD_MT_LDISKFS;
2044 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2045 param->ddp_maxbytes = sb->s_maxbytes;
2047 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2048 /* inode are statically allocated, so per-inode space consumption
2049 * is the space consumed by the directory entry */
2050 param->ddp_inodespace = PER_OBJ_USAGE;
2051 /* EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2052 * = 128MB) which is unlikely to be hit in real life. Report a smaller
2053 * maximum length to not under count the actual number of extents
2054 * needed for writing a file. */
2055 param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2;
2056 /* worst-case extent insertion metadata overhead */
2057 param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2058 param->ddp_mntopts = 0;
2059 if (test_opt(sb, XATTR_USER))
2060 param->ddp_mntopts |= MNTOPT_USERXATTR;
2061 if (test_opt(sb, POSIX_ACL))
2062 param->ddp_mntopts |= MNTOPT_ACL;
2064 /* LOD might calculate the max stripe count based on max_ea_size,
2065 * so we need take account in the overhead as well,
2066 * xattr_header + magic + xattr_entry_head */
2067 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2068 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2070 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2071 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
2072 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2076 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2080 * Concurrency: shouldn't matter.
2082 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2086 CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
2088 rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
2090 CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
2091 LUSTRE_OSD_LDISKFS_NAME, rc);
2097 * Start commit for OSD device.
2099 * An implementation of dt_commit_async method for OSD device.
2100 * Asychronously starts underlayng fs sync and thereby a transaction
2103 * \param env environment
2104 * \param d dt device
2106 * \see dt_device_operations
2108 static int osd_commit_async(const struct lu_env *env,
2109 struct dt_device *d)
2111 struct super_block *s = osd_sb(osd_dt_dev(d));
2114 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
2115 RETURN(s->s_op->sync_fs(s, 0));
2119 * Concurrency: shouldn't matter.
2122 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2124 struct super_block *sb = osd_sb(osd_dt_dev(d));
2125 struct block_device *dev = sb->s_bdev;
2126 #ifdef HAVE_DEV_SET_RDONLY
2127 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2130 int rc = -EOPNOTSUPP;
2134 #ifdef HAVE_DEV_SET_RDONLY
2135 CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
2137 if (sb->s_op->freeze_fs) {
2138 rc = sb->s_op->freeze_fs(sb);
2143 if (jdev && (jdev != dev)) {
2144 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2146 dev_set_rdonly(jdev);
2148 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2149 dev_set_rdonly(dev);
2151 if (sb->s_op->unfreeze_fs)
2152 sb->s_op->unfreeze_fs(sb);
2157 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2158 osd_dt_dev(d)->od_svname, (long)dev, rc);
2164 * Note: we do not count into QUOTA here.
2165 * If we mount with --data_journal we may need more.
2167 const int osd_dto_credits_noquota[DTO_NR] = {
2170 * INDEX_EXTRA_TRANS_BLOCKS(8) +
2171 * SINGLEDATA_TRANS_BLOCKS(8)
2172 * XXX Note: maybe iam need more, since iam have more level than
2175 [DTO_INDEX_INSERT] = 16,
2178 * just modify a single entry, probably merge few within a block
2180 [DTO_INDEX_DELETE] = 1,
2184 [DTO_INDEX_UPDATE] = 16,
2186 * 4(inode, inode bits, groups, GDT)
2187 * notice: OI updates are counted separately with DTO_INDEX_INSERT
2189 [DTO_OBJECT_CREATE] = 4,
2191 * 4(inode, inode bits, groups, GDT)
2192 * notice: OI updates are counted separately with DTO_INDEX_DELETE
2194 [DTO_OBJECT_DELETE] = 4,
2196 * Attr set credits (inode)
2198 [DTO_ATTR_SET_BASE] = 1,
2200 * Xattr set. The same as xattr of EXT3.
2201 * DATA_TRANS_BLOCKS(14)
2202 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2203 * are also counted in. Do not know why?
2205 [DTO_XATTR_SET] = 14,
2207 * credits for inode change during write.
2209 [DTO_WRITE_BASE] = 3,
2211 * credits for single block write.
2213 [DTO_WRITE_BLOCK] = 14,
2215 * Attr set credits for chown.
2216 * This is extra credits for setattr, and it is null without quota
2218 [DTO_ATTR_SET_CHOWN] = 0
2221 static const struct dt_device_operations osd_dt_ops = {
2222 .dt_root_get = osd_root_get,
2223 .dt_statfs = osd_statfs,
2224 .dt_trans_create = osd_trans_create,
2225 .dt_trans_start = osd_trans_start,
2226 .dt_trans_stop = osd_trans_stop,
2227 .dt_trans_cb_add = osd_trans_cb_add,
2228 .dt_conf_get = osd_conf_get,
2229 .dt_sync = osd_sync,
2231 .dt_commit_async = osd_commit_async,
2234 static void osd_object_read_lock(const struct lu_env *env,
2235 struct dt_object *dt, unsigned role)
2237 struct osd_object *obj = osd_dt_obj(dt);
2238 struct osd_thread_info *oti = osd_oti_get(env);
2240 LINVRNT(osd_invariant(obj));
2242 LASSERT(obj->oo_owner != env);
2243 down_read_nested(&obj->oo_sem, role);
2245 LASSERT(obj->oo_owner == NULL);
2249 static void osd_object_write_lock(const struct lu_env *env,
2250 struct dt_object *dt, unsigned role)
2252 struct osd_object *obj = osd_dt_obj(dt);
2253 struct osd_thread_info *oti = osd_oti_get(env);
2255 LINVRNT(osd_invariant(obj));
2257 LASSERT(obj->oo_owner != env);
2258 down_write_nested(&obj->oo_sem, role);
2260 LASSERT(obj->oo_owner == NULL);
2261 obj->oo_owner = env;
2265 static void osd_object_read_unlock(const struct lu_env *env,
2266 struct dt_object *dt)
2268 struct osd_object *obj = osd_dt_obj(dt);
2269 struct osd_thread_info *oti = osd_oti_get(env);
2271 LINVRNT(osd_invariant(obj));
2273 LASSERT(oti->oti_r_locks > 0);
2275 up_read(&obj->oo_sem);
2278 static void osd_object_write_unlock(const struct lu_env *env,
2279 struct dt_object *dt)
2281 struct osd_object *obj = osd_dt_obj(dt);
2282 struct osd_thread_info *oti = osd_oti_get(env);
2284 LINVRNT(osd_invariant(obj));
2286 LASSERT(obj->oo_owner == env);
2287 LASSERT(oti->oti_w_locks > 0);
2289 obj->oo_owner = NULL;
2290 up_write(&obj->oo_sem);
2293 static int osd_object_write_locked(const struct lu_env *env,
2294 struct dt_object *dt)
2296 struct osd_object *obj = osd_dt_obj(dt);
2298 LINVRNT(osd_invariant(obj));
2300 return obj->oo_owner == env;
2303 static struct timespec *osd_inode_time(const struct lu_env *env,
2304 struct inode *inode, __u64 seconds)
2306 struct osd_thread_info *oti = osd_oti_get(env);
2307 struct timespec *t = &oti->oti_time;
2309 t->tv_sec = seconds;
2311 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2315 static void osd_inode_getattr(const struct lu_env *env,
2316 struct inode *inode, struct lu_attr *attr)
2318 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2319 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2320 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
2323 attr->la_atime = LTIME_S(inode->i_atime);
2324 attr->la_mtime = LTIME_S(inode->i_mtime);
2325 attr->la_ctime = LTIME_S(inode->i_ctime);
2326 attr->la_mode = inode->i_mode;
2327 attr->la_size = i_size_read(inode);
2328 attr->la_blocks = inode->i_blocks;
2329 attr->la_uid = i_uid_read(inode);
2330 attr->la_gid = i_gid_read(inode);
2331 attr->la_flags = ll_inode_to_ext_flags(inode->i_flags);
2332 attr->la_nlink = inode->i_nlink;
2333 attr->la_rdev = inode->i_rdev;
2334 attr->la_blksize = 1 << inode->i_blkbits;
2335 attr->la_blkbits = inode->i_blkbits;
2338 static int osd_attr_get(const struct lu_env *env,
2339 struct dt_object *dt,
2340 struct lu_attr *attr)
2342 struct osd_object *obj = osd_dt_obj(dt);
2344 if (unlikely(!dt_object_exists(dt)))
2346 if (unlikely(obj->oo_destroyed))
2349 LASSERT(!dt_object_remote(dt));
2350 LINVRNT(osd_invariant(obj));
2352 spin_lock(&obj->oo_guard);
2353 osd_inode_getattr(env, obj->oo_inode, attr);
2354 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2355 attr->la_flags |= LUSTRE_ORPHAN_FL;
2356 spin_unlock(&obj->oo_guard);
2361 static int osd_declare_attr_set(const struct lu_env *env,
2362 struct dt_object *dt,
2363 const struct lu_attr *attr,
2364 struct thandle *handle)
2366 struct osd_thandle *oh;
2367 struct osd_object *obj;
2368 struct osd_thread_info *info = osd_oti_get(env);
2369 struct lquota_id_info *qi = &info->oti_qi;
2377 LASSERT(dt != NULL);
2378 LASSERT(handle != NULL);
2380 obj = osd_dt_obj(dt);
2381 LASSERT(osd_invariant(obj));
2383 oh = container_of0(handle, struct osd_thandle, ot_super);
2384 LASSERT(oh->ot_handle == NULL);
2386 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2387 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2389 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2390 osd_dto_credits_noquota[DTO_XATTR_SET]);
2392 if (attr == NULL || obj->oo_inode == NULL)
2395 bspace = obj->oo_inode->i_blocks;
2396 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
2397 bspace = toqb(bspace);
2399 /* Changing ownership is always preformed by super user, it should not
2402 * We still need to call the osd_declare_qid() to calculate the journal
2403 * credits for updating quota accounting files and to trigger quota
2404 * space adjustment once the operation is completed.*/
2405 if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2407 uid = i_uid_read(obj->oo_inode);
2408 qi->lqi_type = USRQUOTA;
2409 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2410 /* inode accounting */
2411 qi->lqi_is_blk = false;
2413 /* one more inode for the new uid ... */
2414 qi->lqi_id.qid_uid = attr->la_uid;
2416 /* Reserve credits for the new uid */
2417 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2418 if (rc == -EDQUOT || rc == -EINPROGRESS)
2423 /* and one less inode for the current uid */
2424 qi->lqi_id.qid_uid = uid;
2426 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2427 if (rc == -EDQUOT || rc == -EINPROGRESS)
2432 /* block accounting */
2433 qi->lqi_is_blk = true;
2435 /* more blocks for the new uid ... */
2436 qi->lqi_id.qid_uid = attr->la_uid;
2437 qi->lqi_space = bspace;
2439 * Credits for the new uid has been reserved, re-use "obj"
2440 * to save credit reservation.
2442 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2443 if (rc == -EDQUOT || rc == -EINPROGRESS)
2448 /* and finally less blocks for the current uid */
2449 qi->lqi_id.qid_uid = uid;
2450 qi->lqi_space = -bspace;
2451 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2452 if (rc == -EDQUOT || rc == -EINPROGRESS)
2458 gid = i_gid_read(obj->oo_inode);
2459 qi->lqi_type = GRPQUOTA;
2460 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2462 /* inode accounting */
2463 qi->lqi_is_blk = false;
2465 /* one more inode for the new gid ... */
2466 qi->lqi_id.qid_gid = attr->la_gid;
2468 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2469 if (rc == -EDQUOT || rc == -EINPROGRESS)
2474 /* and one less inode for the current gid */
2475 qi->lqi_id.qid_gid = gid;
2477 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2478 if (rc == -EDQUOT || rc == -EINPROGRESS)
2483 /* block accounting */
2484 qi->lqi_is_blk = true;
2486 /* more blocks for the new gid ... */
2487 qi->lqi_id.qid_gid = attr->la_gid;
2488 qi->lqi_space = bspace;
2489 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2490 if (rc == -EDQUOT || rc == -EINPROGRESS)
2495 /* and finally less blocks for the current gid */
2496 qi->lqi_id.qid_gid = gid;
2497 qi->lqi_space = -bspace;
2498 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2499 if (rc == -EDQUOT || rc == -EINPROGRESS)
2508 static int osd_inode_setattr(const struct lu_env *env,
2509 struct inode *inode, const struct lu_attr *attr)
2511 __u64 bits = attr->la_valid;
2513 /* Only allow set size for regular file */
2514 if (!S_ISREG(inode->i_mode))
2515 bits &= ~(LA_SIZE | LA_BLOCKS);
2520 if (bits & LA_ATIME)
2521 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
2522 if (bits & LA_CTIME)
2523 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
2524 if (bits & LA_MTIME)
2525 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
2526 if (bits & LA_SIZE) {
2527 spin_lock(&inode->i_lock);
2528 LDISKFS_I(inode)->i_disksize = attr->la_size;
2529 i_size_write(inode, attr->la_size);
2530 spin_unlock(&inode->i_lock);
2533 /* OSD should not change "i_blocks" which is used by quota.
2534 * "i_blocks" should be changed by ldiskfs only. */
2536 inode->i_mode = (inode->i_mode & S_IFMT) |
2537 (attr->la_mode & ~S_IFMT);
2539 i_uid_write(inode, attr->la_uid);
2541 i_gid_write(inode, attr->la_gid);
2542 if (bits & LA_NLINK)
2543 set_nlink(inode, attr->la_nlink);
2545 inode->i_rdev = attr->la_rdev;
2547 if (bits & LA_FLAGS) {
2548 /* always keep S_NOCMTIME */
2549 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2555 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2557 if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2558 (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2562 ll_vfs_dq_init(inode);
2564 if (attr->la_valid & LA_UID)
2565 iattr.ia_valid |= ATTR_UID;
2566 if (attr->la_valid & LA_GID)
2567 iattr.ia_valid |= ATTR_GID;
2568 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2569 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2571 rc = ll_vfs_dq_transfer(inode, &iattr);
2573 CERROR("%s: quota transfer failed: rc = %d. Is quota "
2574 "enforcement enabled on the ldiskfs "
2575 "filesystem?\n", inode->i_sb->s_id, rc);
2582 static int osd_attr_set(const struct lu_env *env,
2583 struct dt_object *dt,
2584 const struct lu_attr *attr,
2585 struct thandle *handle)
2587 struct osd_object *obj = osd_dt_obj(dt);
2588 struct inode *inode;
2591 if (!dt_object_exists(dt))
2594 LASSERT(handle != NULL);
2595 LASSERT(!dt_object_remote(dt));
2596 LASSERT(osd_invariant(obj));
2598 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2600 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
2601 struct osd_thread_info *oti = osd_oti_get(env);
2602 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
2603 struct lu_fid *fid1 = &oti->oti_fid;
2604 struct osd_inode_id *id = &oti->oti_id;
2605 struct iam_path_descr *ipd;
2606 struct iam_container *bag;
2607 struct osd_thandle *oh;
2610 fid_cpu_to_be(fid1, fid0);
2611 memset(id, 1, sizeof(*id));
2612 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2613 fid0)->oi_dir.od_container;
2614 ipd = osd_idx_ipd_get(env, bag);
2615 if (unlikely(ipd == NULL))
2618 oh = container_of0(handle, struct osd_thandle, ot_super);
2619 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2620 (const struct iam_rec *)id, ipd);
2621 osd_ipd_put(env, bag, ipd);
2622 return(rc > 0 ? 0 : rc);
2625 inode = obj->oo_inode;
2627 rc = osd_quota_transfer(inode, attr);
2631 spin_lock(&obj->oo_guard);
2632 rc = osd_inode_setattr(env, inode, attr);
2633 spin_unlock(&obj->oo_guard);
2637 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2639 if (!(attr->la_valid & LA_FLAGS))
2642 /* Let's check if there are extra flags need to be set into LMA */
2643 if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
2644 struct osd_thread_info *info = osd_oti_get(env);
2645 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2647 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
2651 lma->lma_incompat |=
2652 lustre_to_lma_flags(attr->la_flags);
2653 lustre_lma_swab(lma);
2654 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
2655 lma, sizeof(*lma), XATTR_REPLACE);
2657 struct osd_device *osd = osd_obj2dev(obj);
2659 CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
2660 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
2661 lma->lma_incompat, rc);
2664 attr->la_flags & LUSTRE_LMA_FL_MASKS;
2666 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2669 osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2674 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2675 struct osd_object *obj,
2676 const char *name, const int namelen)
2678 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2681 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2682 umode_t mode, struct dt_allocation_hint *hint,
2686 struct osd_device *osd = osd_obj2dev(obj);
2687 struct osd_thandle *oth;
2688 struct dt_object *parent = NULL;
2689 struct inode *inode;
2691 LINVRNT(osd_invariant(obj));
2692 LASSERT(obj->oo_inode == NULL);
2693 LASSERT(obj->oo_hl_head == NULL);
2695 if (S_ISDIR(mode) && ldiskfs_pdo) {
2696 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2697 if (obj->oo_hl_head == NULL)
2701 oth = container_of(th, struct osd_thandle, ot_super);
2702 LASSERT(oth->ot_handle->h_transaction != NULL);
2704 if (hint != NULL && hint->dah_parent != NULL &&
2705 !dt_object_remote(hint->dah_parent))
2706 parent = hint->dah_parent;
2708 inode = ldiskfs_create_inode(oth->ot_handle,
2709 parent ? osd_dt_obj(parent)->oo_inode :
2710 osd_sb(osd)->s_root->d_inode,
2712 if (!IS_ERR(inode)) {
2713 /* Do not update file c/mtime in ldiskfs. */
2714 inode->i_flags |= S_NOCMTIME;
2716 /* For new created object, it must be consistent,
2717 * and it is unnecessary to scrub against it. */
2718 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2720 obj->oo_inode = inode;
2723 if (obj->oo_hl_head != NULL) {
2724 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2725 obj->oo_hl_head = NULL;
2727 result = PTR_ERR(inode);
2729 LINVRNT(osd_invariant(obj));
2737 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2738 struct lu_attr *attr,
2739 struct dt_allocation_hint *hint,
2740 struct dt_object_format *dof,
2744 struct osd_thandle *oth;
2745 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2747 LASSERT(S_ISDIR(attr->la_mode));
2749 oth = container_of(th, struct osd_thandle, ot_super);
2750 LASSERT(oth->ot_handle->h_transaction != NULL);
2751 result = osd_mkfile(info, obj, mode, hint, th);
2756 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2757 struct lu_attr *attr,
2758 struct dt_allocation_hint *hint,
2759 struct dt_object_format *dof,
2763 struct osd_thandle *oth;
2764 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2766 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2768 LASSERT(S_ISREG(attr->la_mode));
2770 oth = container_of(th, struct osd_thandle, ot_super);
2771 LASSERT(oth->ot_handle->h_transaction != NULL);
2773 result = osd_mkfile(info, obj, mode, hint, th);
2775 LASSERT(obj->oo_inode != NULL);
2776 if (feat->dif_flags & DT_IND_VARKEY)
2777 result = iam_lvar_create(obj->oo_inode,
2778 feat->dif_keysize_max,
2780 feat->dif_recsize_max,
2783 result = iam_lfix_create(obj->oo_inode,
2784 feat->dif_keysize_max,
2786 feat->dif_recsize_max,
2793 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2794 struct lu_attr *attr,
2795 struct dt_allocation_hint *hint,
2796 struct dt_object_format *dof,
2799 LASSERT(S_ISREG(attr->la_mode));
2800 return osd_mkfile(info, obj, (attr->la_mode &
2801 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2804 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2805 struct lu_attr *attr,
2806 struct dt_allocation_hint *hint,
2807 struct dt_object_format *dof,
2810 LASSERT(S_ISLNK(attr->la_mode));
2811 return osd_mkfile(info, obj, (attr->la_mode &
2812 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2815 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2816 struct lu_attr *attr,
2817 struct dt_allocation_hint *hint,
2818 struct dt_object_format *dof,
2821 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2824 LINVRNT(osd_invariant(obj));
2825 LASSERT(obj->oo_inode == NULL);
2826 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2827 S_ISFIFO(mode) || S_ISSOCK(mode));
2829 result = osd_mkfile(info, obj, mode, hint, th);
2831 LASSERT(obj->oo_inode != NULL);
2833 * This inode should be marked dirty for i_rdev. Currently
2834 * that is done in the osd_attr_init().
2836 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2839 LINVRNT(osd_invariant(obj));
2843 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2845 struct dt_allocation_hint *hint,
2846 struct dt_object_format *dof,
2849 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2851 osd_obj_type_f result;
2867 result = osd_mk_index;
2878 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2879 struct dt_object *parent, struct dt_object *child,
2884 ah->dah_parent = parent;
2885 ah->dah_mode = child_mode;
2887 if (parent != NULL && !dt_object_remote(parent)) {
2888 /* will help to find FID->ino at dt_insert("..") */
2889 struct osd_object *pobj = osd_dt_obj(parent);
2890 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
2894 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2895 struct lu_attr *attr, struct dt_object_format *dof)
2897 struct inode *inode = obj->oo_inode;
2898 __u64 valid = attr->la_valid;
2901 attr->la_valid &= ~(LA_TYPE | LA_MODE);
2903 if (dof->dof_type != DFT_NODE)
2904 attr->la_valid &= ~LA_RDEV;
2905 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2906 attr->la_valid &= ~LA_ATIME;
2907 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2908 attr->la_valid &= ~LA_CTIME;
2909 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2910 attr->la_valid &= ~LA_MTIME;
2912 result = osd_quota_transfer(inode, attr);
2916 if (attr->la_valid != 0) {
2917 result = osd_inode_setattr(info->oti_env, inode, attr);
2919 * The osd_inode_setattr() should always succeed here. The
2920 * only error that could be returned is EDQUOT when we are
2921 * trying to change the UID or GID of the inode. However, this
2922 * should not happen since quota enforcement is no longer
2923 * enabled on ldiskfs (lquota takes care of it).
2925 LASSERTF(result == 0, "%d\n", result);
2926 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2929 attr->la_valid = valid;
2933 * Helper function for osd_object_create()
2935 * \retval 0, on success
2937 static int __osd_object_create(struct osd_thread_info *info,
2938 struct osd_object *obj, struct lu_attr *attr,
2939 struct dt_allocation_hint *hint,
2940 struct dt_object_format *dof,
2946 osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
2948 /* we drop umask so that permissions we pass are not affected */
2949 umask = current->fs->umask;
2950 current->fs->umask = 0;
2952 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2954 if (likely(obj->oo_inode != NULL)) {
2955 LASSERT(obj->oo_inode->i_state & I_NEW);
2957 /* Unlock the inode before attr initialization to avoid
2958 * unnecessary dqget operations. LU-6378 */
2959 unlock_new_inode(obj->oo_inode);
2962 if (likely(result == 0)) {
2963 osd_attr_init(info, obj, attr, dof);
2964 osd_object_init0(obj);
2967 /* restore previous umask value */
2968 current->fs->umask = umask;
2970 osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
2976 * Helper function for osd_object_create()
2978 * \retval 0, on success
2980 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2981 const struct lu_fid *fid, struct thandle *th)
2983 struct osd_thread_info *info = osd_oti_get(env);
2984 struct osd_inode_id *id = &info->oti_id;
2985 struct osd_device *osd = osd_obj2dev(obj);
2986 struct osd_thandle *oh;
2989 LASSERT(obj->oo_inode != NULL);
2991 oh = container_of0(th, struct osd_thandle, ot_super);
2992 LASSERT(oh->ot_handle);
2993 osd_trans_exec_op(env, th, OSD_OT_INSERT);
2995 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2996 rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
2997 OI_CHECK_FLD, NULL);
2998 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3003 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3004 u64 seq, struct lu_seq_range *range)
3006 struct seq_server_site *ss = osd_seq_site(osd);
3008 if (fid_seq_is_idif(seq)) {
3009 fld_range_set_ost(range);
3010 range->lsr_index = idif_ost_idx(seq);
3014 if (!fid_seq_in_fldb(seq)) {
3015 fld_range_set_mdt(range);
3017 /* FIXME: If ss is NULL, it suppose not get lsr_index
3019 range->lsr_index = ss->ss_node_id;
3023 LASSERT(ss != NULL);
3024 fld_range_set_any(range);
3025 /* OSD will only do local fld lookup */
3026 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3030 * Concurrency: no external locking is necessary.
3032 static int osd_declare_object_create(const struct lu_env *env,
3033 struct dt_object *dt,
3034 struct lu_attr *attr,
3035 struct dt_allocation_hint *hint,
3036 struct dt_object_format *dof,
3037 struct thandle *handle)
3039 struct osd_thandle *oh;
3043 LASSERT(handle != NULL);
3045 oh = container_of0(handle, struct osd_thandle, ot_super);
3046 LASSERT(oh->ot_handle == NULL);
3048 /* EA object consumes more credits than regular object: osd_mk_index
3049 * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3050 * leaf_node, could involves the block, block bitmap, groups, GDT
3051 * change for each block, so add 4 * 2 credits in that case. */
3052 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3053 osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3054 (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3055 /* Reuse idle OI block may cause additional one OI block
3057 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3058 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3063 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
3064 osd_dt_obj(dt), false, NULL, false);
3068 /* will help to find FID->ino mapping at dt_insert() */
3069 rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3075 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
3076 struct lu_attr *attr,
3077 struct dt_allocation_hint *hint,
3078 struct dt_object_format *dof, struct thandle *th)
3080 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3081 struct osd_object *obj = osd_dt_obj(dt);
3082 struct osd_thread_info *info = osd_oti_get(env);
3086 if (dt_object_exists(dt))
3089 LINVRNT(osd_invariant(obj));
3090 LASSERT(!dt_object_remote(dt));
3091 LASSERT(osd_write_locked(env, obj));
3092 LASSERT(th != NULL);
3094 if (unlikely(fid_is_acct(fid)))
3095 /* Quota files can't be created from the kernel any more,
3096 * 'tune2fs -O quota' will take care of creating them */
3099 result = __osd_object_create(info, obj, attr, hint, dof, th);
3101 result = __osd_oi_insert(env, obj, fid, th);
3102 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3103 obj->oo_dt.do_body_ops = &osd_body_ops;
3105 LASSERT(ergo(result == 0,
3106 dt_object_exists(dt) && !dt_object_remote(dt)));
3108 LASSERT(osd_invariant(obj));
3113 * Called to destroy on-disk representation of the object
3115 * Concurrency: must be locked
3117 static int osd_declare_object_destroy(const struct lu_env *env,
3118 struct dt_object *dt,
3121 struct osd_object *obj = osd_dt_obj(dt);
3122 struct inode *inode = obj->oo_inode;
3123 struct osd_thandle *oh;
3130 oh = container_of0(th, struct osd_thandle, ot_super);
3131 LASSERT(oh->ot_handle == NULL);
3133 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3134 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3135 /* Recycle idle OI leaf may cause additional three OI blocks
3137 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3138 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3139 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3140 /* one less inode */
3141 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3142 -1, oh, obj, false, NULL, false);
3145 /* data to be truncated */
3146 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3147 0, oh, obj, true, NULL, false);
3151 /* will help to find FID->ino when this object is being
3152 * added to PENDING/ */
3153 rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3158 static int osd_object_destroy(const struct lu_env *env,
3159 struct dt_object *dt,
3162 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3163 struct osd_object *obj = osd_dt_obj(dt);
3164 struct inode *inode = obj->oo_inode;
3165 struct osd_device *osd = osd_obj2dev(obj);
3166 struct osd_thandle *oh;
3170 oh = container_of0(th, struct osd_thandle, ot_super);
3171 LASSERT(oh->ot_handle);
3173 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3175 if (unlikely(fid_is_acct(fid)))
3178 if (S_ISDIR(inode->i_mode)) {
3179 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3180 inode->i_nlink == 2);
3181 /* it will check/delete the inode from remote parent,
3182 * how to optimize it? unlink performance impaction XXX */
3183 result = osd_delete_from_remote_parent(env, osd, obj, oh);
3185 CERROR("%s: delete inode "DFID": rc = %d\n",
3186 osd_name(osd), PFID(fid), result);
3188 spin_lock(&obj->oo_guard);
3190 spin_unlock(&obj->oo_guard);
3191 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3194 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3196 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3198 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3199 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3200 oh->ot_handle, OI_CHECK_FLD);
3202 osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3203 /* XXX: add to ext3 orphan list */
3204 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3206 /* not needed in the cache anymore */
3207 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3208 obj->oo_destroyed = 1;
3214 * Put the fid into lustre_mdt_attrs, and then place the structure
3215 * inode's ea. This fid should not be altered during the life time
3218 * \retval +ve, on success
3219 * \retval -ve, on error
3221 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3223 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3224 const struct lu_fid *fid, __u32 compat, __u32 incompat)
3226 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3230 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3233 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3236 lustre_lma_init(lma, fid, compat, incompat);
3237 lustre_lma_swab(lma);
3239 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
3241 /* LMA may already exist, but we need to check that all the
3242 * desired compat/incompat flags have been added. */
3243 if (unlikely(rc == -EEXIST)) {
3244 if (compat == 0 && incompat == 0)
3247 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3248 XATTR_NAME_LMA, info->oti_mdt_attrs_old,
3253 lustre_lma_swab(lma);
3254 if (!(~lma->lma_compat & compat) &&
3255 !(~lma->lma_incompat & incompat))
3258 lma->lma_compat |= compat;
3259 lma->lma_incompat |= incompat;
3260 lustre_lma_swab(lma);
3261 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3262 sizeof(*lma), XATTR_REPLACE);
3269 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3270 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3271 * To have compatilibility with 1.8 ldiskfs driver we need to have
3272 * magic number at start of fid data.
3273 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3276 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3277 const struct lu_fid *fid)
3279 if (!fid_is_namespace_visible(fid) ||
3280 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3281 param->edp_magic = 0;
3285 param->edp_magic = LDISKFS_LUFID_MAGIC;
3286 param->edp_len = sizeof(struct lu_fid) + 1;
3287 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3291 * Try to read the fid from inode ea into dt_rec.
3293 * \param fid object fid.
3295 * \retval 0 on success
3297 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3298 __u32 ino, struct lu_fid *fid,
3299 struct osd_inode_id *id)
3301 struct osd_thread_info *info = osd_oti_get(env);
3302 struct inode *inode;
3305 osd_id_gen(id, ino, OSD_OII_NOGEN);
3306 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3308 RETURN(PTR_ERR(inode));
3314 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3316 struct inode *parent_dir,
3317 const struct lu_fid *dot_fid,
3318 const struct lu_fid *dot_dot_fid,
3319 struct osd_thandle *oth)
3321 struct ldiskfs_dentry_param *dot_ldp;
3322 struct ldiskfs_dentry_param *dot_dot_ldp;
3323 __u32 saved_nlink = dir->i_nlink;
3326 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3327 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3329 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3330 dot_ldp->edp_magic = 0;
3332 rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3333 dir, dot_ldp, dot_dot_ldp);
3334 /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3335 * the subseqent ref_add() will increase the dir->i_nlink
3336 * as 3. That is incorrect for new created directory.
3338 * It looks like hack, because we want to make the OSD API
3339 * to be order-independent for new created directory object
3340 * between dt_insert(..) and ref_add() operations.
3342 * Here, we only restore the in-RAM dir-inode's nlink attr,
3343 * becuase if the nlink attr is not 2, then there will be
3344 * ref_add() called following the dt_insert(..), such call
3345 * will make both the in-RAM and on-disk dir-inode's nlink
3346 * attr to be set as 2. LU-7447 */
3347 set_nlink(dir, saved_nlink);
3352 * Create an local agent inode for remote entry
3354 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3355 struct osd_device *osd,
3356 struct osd_object *pobj,
3357 const struct lu_fid *fid,
3361 struct osd_thread_info *info = osd_oti_get(env);
3362 struct inode *local;
3363 struct osd_thandle *oh;
3368 oh = container_of(th, struct osd_thandle, ot_super);
3369 LASSERT(oh->ot_handle->h_transaction != NULL);
3371 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3372 if (IS_ERR(local)) {
3373 CERROR("%s: create local error %d\n", osd_name(osd),
3374 (int)PTR_ERR(local));
3378 ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3379 unlock_new_inode(local);
3381 /* Set special LMA flag for local agent inode */
3382 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3384 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3385 osd_name(osd), PFID(fid), rc);
3386 RETURN(ERR_PTR(rc));
3392 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3393 lu_object_fid(&pobj->oo_dt.do_lu),
3396 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3397 osd_name(osd), PFID(fid), rc);
3398 RETURN(ERR_PTR(rc));
3405 * when direntry is deleted, we have to take care of possible agent inode
3406 * referenced by that. unfortunately we can't do this at that point:
3407 * iget() within a running transaction leads to deadlock and we better do
3408 * not call that every delete declaration to save performance. so we put
3409 * a potention agent inode on a list and process that once the transaction
3410 * is over. Notice it's not any worse in terms of real orphans as regular
3411 * object destroy doesn't put inodes on the on-disk orphan list. this should
3412 * be addressed separately
3414 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3415 struct osd_thandle *oh,
3418 struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3419 struct osd_obj_orphan *oor;
3426 oor->oor_env = (struct lu_env *)env;
3427 spin_lock(&osd->od_osfs_lock);
3428 list_add(&oor->oor_list, &osd->od_orphan_list);
3429 spin_unlock(&osd->od_osfs_lock);
3431 oh->ot_remove_agents = 1;
3437 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3438 struct osd_device *osd)
3440 struct osd_thread_info *info = osd_oti_get(env);
3441 struct osd_obj_orphan *oor, *tmp;
3442 struct osd_inode_id id;
3443 struct list_head list;
3444 struct inode *inode;
3449 INIT_LIST_HEAD(&list);
3451 spin_lock(&osd->od_osfs_lock);
3452 list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3453 if (oor->oor_env == env) {
3454 list_del(&oor->oor_list);
3455 list_add(&oor->oor_list, &list);
3458 spin_unlock(&osd->od_osfs_lock);
3460 list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3464 list_del(&oor->oor_list);
3467 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3468 inode = osd_iget_fid(info, osd, &id, &fid);
3472 if (!osd_remote_fid(env, osd, &fid)) {
3477 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3479 mark_inode_dirty(inode);
3480 ldiskfs_journal_stop(jh);
3488 * OSD layer object create function for interoperability mode (b11826).
3489 * This is mostly similar to osd_object_create(). Only difference being, fid is
3490 * inserted into inode ea here.
3492 * \retval 0, on success
3493 * \retval -ve, on error
3495 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
3496 struct lu_attr *attr,
3497 struct dt_allocation_hint *hint,
3498 struct dt_object_format *dof,
3501 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3502 struct osd_object *obj = osd_dt_obj(dt);
3503 struct osd_thread_info *info = osd_oti_get(env);
3504 int result, on_ost = 0;
3508 if (dt_object_exists(dt))
3511 LASSERT(osd_invariant(obj));
3512 LASSERT(!dt_object_remote(dt));
3513 LASSERT(osd_write_locked(env, obj));
3514 LASSERT(th != NULL);
3516 if (unlikely(fid_is_acct(fid)))
3517 /* Quota files can't be created from the kernel any more,
3518 * 'tune2fs -O quota' will take care of creating them */
3521 result = __osd_object_create(info, obj, attr, hint, dof, th);
3523 if (fid_is_idif(fid) &&
3524 !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3525 struct lu_fid *tfid = &info->oti_fid;
3526 struct ost_id *oi = &info->oti_ostid;
3528 fid_to_ostid(fid, oi);
3529 ostid_to_fid(tfid, oi, 0);
3531 result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3532 LMAC_FID_ON_OST, 0);
3534 on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3536 result = osd_ea_fid_set(info, obj->oo_inode, fid,
3537 on_ost ? LMAC_FID_ON_OST : 0,
3540 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3541 obj->oo_dt.do_body_ops = &osd_body_ops;
3545 result = __osd_oi_insert(env, obj, fid, th);
3547 /* a small optimization - dt_insert() isn't usually applied
3548 * to OST objects, so we don't need to cache OI mapping for
3550 if (result == 0 && on_ost == 0) {
3551 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3552 result = osd_idc_find_and_init(env, osd, obj);
3553 LASSERT(result == 0);
3556 LASSERT(ergo(result == 0,
3557 dt_object_exists(dt) && !dt_object_remote(dt)));
3558 LINVRNT(osd_invariant(obj));
3562 static int osd_declare_object_ref_add(const struct lu_env *env,
3563 struct dt_object *dt,
3564 struct thandle *handle)
3566 struct osd_thandle *oh;
3568 /* it's possible that object doesn't exist yet */
3569 LASSERT(handle != NULL);
3571 oh = container_of0(handle, struct osd_thandle, ot_super);
3572 LASSERT(oh->ot_handle == NULL);
3574 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3575 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3581 * Concurrency: @dt is write locked.
3583 static int osd_object_ref_add(const struct lu_env *env,
3584 struct dt_object *dt, struct thandle *th)
3586 struct osd_object *obj = osd_dt_obj(dt);
3587 struct inode *inode = obj->oo_inode;
3588 struct osd_thandle *oh;
3591 if (!dt_object_exists(dt) || obj->oo_destroyed)
3594 LINVRNT(osd_invariant(obj));
3595 LASSERT(!dt_object_remote(dt));
3596 LASSERT(osd_write_locked(env, obj));
3597 LASSERT(th != NULL);
3599 oh = container_of0(th, struct osd_thandle, ot_super);
3600 LASSERT(oh->ot_handle != NULL);
3602 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
3604 CDEBUG(D_INODE, DFID" increase nlink %d\n",
3605 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3607 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
3608 * (65000) subdirectories by storing "1" in i_nlink if the link count
3609 * would otherwise overflow. Directory tranversal tools understand
3610 * that (st_nlink == 1) indicates that the filesystem dose not track
3611 * hard links count on the directory, and will not abort subdirectory
3612 * scanning early once (st_nlink - 2) subdirs have been found.
3614 * This also has to properly handle the case of inodes with nlink == 0
3615 * in case they are being linked into the PENDING directory
3617 spin_lock(&obj->oo_guard);
3618 if (unlikely(inode->i_nlink == 0))
3619 /* inc_nlink from 0 may cause WARN_ON */
3620 set_nlink(inode, 1);
3622 ldiskfs_inc_count(oh->ot_handle, inode);
3623 if (!S_ISDIR(inode->i_mode))
3624 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
3626 spin_unlock(&obj->oo_guard);
3628 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3629 LINVRNT(osd_invariant(obj));
3631 osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
3636 static int osd_declare_object_ref_del(const struct lu_env *env,
3637 struct dt_object *dt,
3638 struct thandle *handle)
3640 struct osd_thandle *oh;
3642 if (!dt_object_exists(dt))
3645 LASSERT(!dt_object_remote(dt));
3646 LASSERT(handle != NULL);
3648 oh = container_of0(handle, struct osd_thandle, ot_super);
3649 LASSERT(oh->ot_handle == NULL);
3651 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
3652 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3658 * Concurrency: @dt is write locked.
3660 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
3663 struct osd_object *obj = osd_dt_obj(dt);
3664 struct inode *inode = obj->oo_inode;
3665 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3666 struct osd_thandle *oh;
3668 if (!dt_object_exists(dt))
3671 LINVRNT(osd_invariant(obj));
3672 LASSERT(!dt_object_remote(dt));
3673 LASSERT(osd_write_locked(env, obj));
3674 LASSERT(th != NULL);
3676 oh = container_of0(th, struct osd_thandle, ot_super);
3677 LASSERT(oh->ot_handle != NULL);
3679 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
3681 spin_lock(&obj->oo_guard);
3682 /* That can be result of upgrade from old Lustre version and
3683 * applied only to local files. Just skip this ref_del call.
3684 * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
3685 if (inode->i_nlink == 0) {
3686 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
3687 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
3688 ", maybe an upgraded file? (LU-3915)\n",
3689 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
3690 spin_unlock(&obj->oo_guard);
3694 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
3695 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3697 ldiskfs_dec_count(oh->ot_handle, inode);
3698 spin_unlock(&obj->oo_guard);
3700 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3701 LINVRNT(osd_invariant(obj));
3703 osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
3709 * Get the 64-bit version for an inode.
3711 static int osd_object_version_get(const struct lu_env *env,
3712 struct dt_object *dt, dt_obj_version_t *ver)
3714 struct inode *inode = osd_dt_obj(dt)->oo_inode;
3716 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
3717 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3718 *ver = LDISKFS_I(inode)->i_fs_version;
3723 * Concurrency: @dt is read locked.
3725 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
3726 struct lu_buf *buf, const char *name)
3728 struct osd_object *obj = osd_dt_obj(dt);
3729 struct inode *inode = obj->oo_inode;
3730 struct osd_thread_info *info = osd_oti_get(env);
3731 struct dentry *dentry = &info->oti_obj_dentry;
3732 bool cache_xattr = false;
3735 /* version get is not real XATTR but uses xattr API */
3736 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3737 /* for version we are just using xattr API but change inode
3739 if (buf->lb_len == 0)
3740 return sizeof(dt_obj_version_t);
3742 if (buf->lb_len < sizeof(dt_obj_version_t))
3745 osd_object_version_get(env, dt, buf->lb_buf);
3747 return sizeof(dt_obj_version_t);
3750 if (!dt_object_exists(dt))
3753 LASSERT(!dt_object_remote(dt));
3754 LASSERT(inode->i_op != NULL);
3755 LASSERT(inode->i_op->getxattr != NULL);
3757 if (strcmp(name, XATTR_NAME_LOV) == 0 ||
3758 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
3762 rc = osd_oxc_get(obj, name, buf);
3767 rc = __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
3769 if (rc == -ENOENT || rc == -ENODATA)
3770 osd_oxc_add(obj, name, NULL, 0);
3771 else if (rc > 0 && buf->lb_buf != NULL)
3772 osd_oxc_add(obj, name, buf->lb_buf, rc);
3778 static int osd_declare_xattr_set(const struct lu_env *env,
3779 struct dt_object *dt,
3780 const struct lu_buf *buf, const char *name,
3781 int fl, struct thandle *handle)
3783 struct osd_thandle *oh;
3785 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3787 LASSERT(handle != NULL);
3789 oh = container_of0(handle, struct osd_thandle, ot_super);
3790 LASSERT(oh->ot_handle == NULL);
3792 if (strcmp(name, XATTR_NAME_LMA) == 0) {
3793 /* For non-upgrading case, the LMA is set first and
3794 * usually fit inode. But for upgrade case, the LMA
3795 * may be in another separated EA block. */
3796 if (!dt_object_exists(dt))
3798 else if (fl == LU_XATTR_REPLACE)
3802 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3806 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
3811 if (buf->lb_buf == NULL && dt_object_exists(dt)) {
3812 /* learn xattr size from osd_xattr_get if
3813 attribute has not been read yet */
3814 buflen = __osd_xattr_get(
3815 osd_dt_obj(dt)->oo_inode,
3816 &osd_oti_get(env)->oti_obj_dentry,
3821 buflen = buf->lb_len;
3824 if (buflen > sb->s_blocksize) {
3825 credits += osd_calc_bkmap_credits(
3827 (buflen + sb->s_blocksize - 1) >>
3828 sb->s_blocksize_bits);
3832 * xattr set may involve inode quota change, reserve credits for
3833 * dquot_initialize()
3835 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3838 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3844 * Set the 64-bit version for object
3846 static void osd_object_version_set(const struct lu_env *env,
3847 struct dt_object *dt,
3848 dt_obj_version_t *new_version)
3850 struct inode *inode = osd_dt_obj(dt)->oo_inode;
3852 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
3853 *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3855 LDISKFS_I(inode)->i_fs_version = *new_version;
3856 /** Version is set after all inode operations are finished,
3857 * so we should mark it dirty here */
3858 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3862 * Concurrency: @dt is write locked.
3864 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3865 const struct lu_buf *buf, const char *name, int fl,
3866 struct thandle *handle)
3868 struct osd_object *obj = osd_dt_obj(dt);
3869 struct inode *inode = obj->oo_inode;
3870 struct osd_thread_info *info = osd_oti_get(env);
3875 LASSERT(handle != NULL);
3877 /* version set is not real XATTR */
3878 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3879 /* for version we are just using xattr API but change inode
3881 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3882 osd_object_version_set(env, dt, buf->lb_buf);
3883 return sizeof(dt_obj_version_t);
3886 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
3887 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3889 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3890 if (fl & LU_XATTR_REPLACE)
3891 fs_flags |= XATTR_REPLACE;
3893 if (fl & LU_XATTR_CREATE)
3894 fs_flags |= XATTR_CREATE;
3896 if (strcmp(name, XATTR_NAME_LMV) == 0) {
3897 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3900 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3904 lma->lma_incompat |= LMAI_STRIPED;
3905 lustre_lma_swab(lma);
3906 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3907 sizeof(*lma), XATTR_REPLACE);
3912 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
3913 strcmp(name, XATTR_NAME_LINK) == 0)
3916 rc = __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3918 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3921 (strcmp(name, XATTR_NAME_LOV) == 0 ||
3922 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
3923 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
3929 * Concurrency: @dt is read locked.
3931 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3932 const struct lu_buf *buf)
3934 struct osd_object *obj = osd_dt_obj(dt);
3935 struct inode *inode = obj->oo_inode;
3936 struct osd_thread_info *info = osd_oti_get(env);
3937 struct dentry *dentry = &info->oti_obj_dentry;
3939 if (!dt_object_exists(dt))
3942 LASSERT(!dt_object_remote(dt));
3943 LASSERT(inode->i_op != NULL);
3944 LASSERT(inode->i_op->listxattr != NULL);
3946 dentry->d_inode = inode;
3947 dentry->d_sb = inode->i_sb;
3948 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3951 static int osd_declare_xattr_del(const struct lu_env *env,
3952 struct dt_object *dt, const char *name,
3953 struct thandle *handle)
3955 struct osd_thandle *oh;
3956 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3958 LASSERT(!dt_object_remote(dt));
3959 LASSERT(handle != NULL);
3961 oh = container_of0(handle, struct osd_thandle, ot_super);
3962 LASSERT(oh->ot_handle == NULL);
3964 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3965 osd_dto_credits_noquota[DTO_XATTR_SET]);
3967 * xattr del may involve inode quota change, reserve credits for
3968 * dquot_initialize()
3970 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3976 * Concurrency: @dt is write locked.
3978 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3979 const char *name, struct thandle *handle)
3981 struct osd_object *obj = osd_dt_obj(dt);
3982 struct inode *inode = obj->oo_inode;
3983 struct osd_thread_info *info = osd_oti_get(env);
3984 struct dentry *dentry = &info->oti_obj_dentry;
3987 if (!dt_object_exists(dt))
3990 LASSERT(!dt_object_remote(dt));
3991 LASSERT(inode->i_op != NULL);
3992 LASSERT(inode->i_op->removexattr != NULL);
3993 LASSERT(handle != NULL);
3995 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3997 ll_vfs_dq_init(inode);
3998 dentry->d_inode = inode;
3999 dentry->d_sb = inode->i_sb;
4000 rc = inode->i_op->removexattr(dentry, name);
4001 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4004 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4005 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4006 osd_oxc_del(obj, name);
4011 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4012 __u64 start, __u64 end)
4014 struct osd_object *obj = osd_dt_obj(dt);
4015 struct inode *inode = obj->oo_inode;
4016 struct osd_thread_info *info = osd_oti_get(env);
4017 struct dentry *dentry = &info->oti_obj_dentry;
4018 struct file *file = &info->oti_file;
4023 dentry->d_inode = inode;
4024 dentry->d_sb = inode->i_sb;
4025 file->f_path.dentry = dentry;
4026 file->f_mapping = inode->i_mapping;
4027 file->f_op = inode->i_fop;
4028 set_file_inode(file, inode);
4030 rc = ll_vfs_fsync_range(file, start, end, 0);
4035 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4044 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4045 const struct dt_index_features *feat)
4047 struct iam_descr *descr;
4049 if (osd_object_is_root(o))
4050 return feat == &dt_directory_features;
4052 LASSERT(o->oo_dir != NULL);
4054 descr = o->oo_dir->od_container.ic_descr;
4055 if (feat == &dt_directory_features) {
4056 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4062 feat->dif_keysize_min <= descr->id_key_size &&
4063 descr->id_key_size <= feat->dif_keysize_max &&
4064 feat->dif_recsize_min <= descr->id_rec_size &&
4065 descr->id_rec_size <= feat->dif_recsize_max &&
4066 !(feat->dif_flags & (DT_IND_VARKEY |
4067 DT_IND_VARREC | DT_IND_NONUNQ)) &&
4068 ergo(feat->dif_flags & DT_IND_UPDATE,
4069 1 /* XXX check that object (and file system) is
4074 static int osd_iam_container_init(const struct lu_env *env,
4075 struct osd_object *obj,
4076 struct osd_directory *dir)
4078 struct iam_container *bag = &dir->od_container;
4081 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4085 result = iam_container_setup(bag);
4087 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4089 iam_container_fini(bag);
4096 * Concurrency: no external locking is necessary.
4098 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4099 const struct dt_index_features *feat)
4103 struct osd_object *obj = osd_dt_obj(dt);
4105 LINVRNT(osd_invariant(obj));
4107 if (osd_object_is_root(obj)) {
4108 dt->do_index_ops = &osd_index_ea_ops;
4110 } else if (feat == &dt_directory_features) {
4111 dt->do_index_ops = &osd_index_ea_ops;
4112 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4117 } else if (unlikely(feat == &dt_otable_features)) {
4118 dt->do_index_ops = &osd_otable_ops;
4120 } else if (unlikely(feat == &dt_acct_features)) {
4121 dt->do_index_ops = &osd_acct_index_ops;
4124 } else if (!osd_has_index(obj)) {
4125 struct osd_directory *dir;
4130 spin_lock(&obj->oo_guard);
4131 if (obj->oo_dir == NULL)
4135 * Concurrent thread allocated container data.
4138 spin_unlock(&obj->oo_guard);
4140 * Now, that we have container data, serialize its
4143 down_write(&obj->oo_ext_idx_sem);
4145 * recheck under lock.
4147 if (!osd_has_index(obj))
4148 result = osd_iam_container_init(env, obj,
4152 up_write(&obj->oo_ext_idx_sem);
4160 if (result == 0 && skip_iam == 0) {
4161 if (!osd_iam_index_probe(env, obj, feat))
4164 LINVRNT(osd_invariant(obj));
4166 if (result == 0 && feat == &dt_quota_glb_features &&
4167 fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
4168 result = osd_quota_migration(env, dt);
4173 static int osd_otable_it_attr_get(const struct lu_env *env,
4174 struct dt_object *dt,
4175 struct lu_attr *attr)
4181 static const struct dt_object_operations osd_obj_ops = {
4182 .do_read_lock = osd_object_read_lock,
4183 .do_write_lock = osd_object_write_lock,
4184 .do_read_unlock = osd_object_read_unlock,
4185 .do_write_unlock = osd_object_write_unlock,
4186 .do_write_locked = osd_object_write_locked,
4187 .do_attr_get = osd_attr_get,
4188 .do_declare_attr_set = osd_declare_attr_set,
4189 .do_attr_set = osd_attr_set,
4190 .do_ah_init = osd_ah_init,
4191 .do_declare_create = osd_declare_object_create,
4192 .do_create = osd_object_create,
4193 .do_declare_destroy = osd_declare_object_destroy,
4194 .do_destroy = osd_object_destroy,
4195 .do_index_try = osd_index_try,
4196 .do_declare_ref_add = osd_declare_object_ref_add,
4197 .do_ref_add = osd_object_ref_add,
4198 .do_declare_ref_del = osd_declare_object_ref_del,
4199 .do_ref_del = osd_object_ref_del,
4200 .do_xattr_get = osd_xattr_get,
4201 .do_declare_xattr_set = osd_declare_xattr_set,
4202 .do_xattr_set = osd_xattr_set,
4203 .do_declare_xattr_del = osd_declare_xattr_del,
4204 .do_xattr_del = osd_xattr_del,
4205 .do_xattr_list = osd_xattr_list,
4206 .do_object_sync = osd_object_sync,
4207 .do_invalidate = osd_invalidate,
4211 * dt_object_operations for interoperability mode
4212 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
4214 static const struct dt_object_operations osd_obj_ea_ops = {
4215 .do_read_lock = osd_object_read_lock,
4216 .do_write_lock = osd_object_write_lock,
4217 .do_read_unlock = osd_object_read_unlock,
4218 .do_write_unlock = osd_object_write_unlock,
4219 .do_write_locked = osd_object_write_locked,
4220 .do_attr_get = osd_attr_get,
4221 .do_declare_attr_set = osd_declare_attr_set,
4222 .do_attr_set = osd_attr_set,
4223 .do_ah_init = osd_ah_init,
4224 .do_declare_create = osd_declare_object_create,
4225 .do_create = osd_object_ea_create,
4226 .do_declare_destroy = osd_declare_object_destroy,
4227 .do_destroy = osd_object_destroy,
4228 .do_index_try = osd_index_try,
4229 .do_declare_ref_add = osd_declare_object_ref_add,
4230 .do_ref_add = osd_object_ref_add,
4231 .do_declare_ref_del = osd_declare_object_ref_del,
4232 .do_ref_del = osd_object_ref_del,
4233 .do_xattr_get = osd_xattr_get,
4234 .do_declare_xattr_set = osd_declare_xattr_set,
4235 .do_xattr_set = osd_xattr_set,
4236 .do_declare_xattr_del = osd_declare_xattr_del,
4237 .do_xattr_del = osd_xattr_del,
4238 .do_xattr_list = osd_xattr_list,
4239 .do_object_sync = osd_object_sync,
4240 .do_invalidate = osd_invalidate,
4243 static const struct dt_object_operations osd_obj_otable_it_ops = {
4244 .do_attr_get = osd_otable_it_attr_get,
4245 .do_index_try = osd_index_try,
4248 static int osd_index_declare_iam_delete(const struct lu_env *env,
4249 struct dt_object *dt,
4250 const struct dt_key *key,
4251 struct thandle *handle)
4253 struct osd_thandle *oh;
4255 oh = container_of0(handle, struct osd_thandle, ot_super);
4256 LASSERT(oh->ot_handle == NULL);
4258 /* Recycle may cause additional three blocks to be changed. */
4259 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4260 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4266 * delete a (key, value) pair from index \a dt specified by \a key
4268 * \param dt osd index object
4269 * \param key key for index
4270 * \param rec record reference
4271 * \param handle transaction handler
4274 * \retval -ve failure
4276 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4277 const struct dt_key *key,
4278 struct thandle *handle)
4280 struct osd_thread_info *oti = osd_oti_get(env);
4281 struct osd_object *obj = osd_dt_obj(dt);
4282 struct osd_thandle *oh;
4283 struct iam_path_descr *ipd;
4284 struct iam_container *bag = &obj->oo_dir->od_container;
4288 if (!dt_object_exists(dt))
4291 LINVRNT(osd_invariant(obj));
4292 LASSERT(!dt_object_remote(dt));
4293 LASSERT(bag->ic_object == obj->oo_inode);
4294 LASSERT(handle != NULL);
4296 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4298 ipd = osd_idx_ipd_get(env, bag);
4299 if (unlikely(ipd == NULL))
4302 oh = container_of0(handle, struct osd_thandle, ot_super);
4303 LASSERT(oh->ot_handle != NULL);
4304 LASSERT(oh->ot_handle->h_transaction != NULL);
4306 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4307 /* swab quota uid/gid provided by caller */
4308 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4309 key = (const struct dt_key *)&oti->oti_quota_id;
4312 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4313 osd_ipd_put(env, bag, ipd);
4314 LINVRNT(osd_invariant(obj));
4315 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4319 static int osd_index_declare_ea_delete(const struct lu_env *env,
4320 struct dt_object *dt,
4321 const struct dt_key *key,
4322 struct thandle *handle)
4324 struct osd_thandle *oh;
4325 struct inode *inode;
4329 LASSERT(!dt_object_remote(dt));
4330 LASSERT(handle != NULL);
4332 oh = container_of0(handle, struct osd_thandle, ot_super);
4333 LASSERT(oh->ot_handle == NULL);
4335 /* due to DNE we may need to remove an agent inode */
4336 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4337 osd_dto_credits_noquota[DTO_INDEX_DELETE] +
4338 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
4340 inode = osd_dt_obj(dt)->oo_inode;
4344 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
4345 0, oh, osd_dt_obj(dt), true, NULL, false);
4349 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
4352 struct osd_fid_pack *rec;
4355 if (de->file_type & LDISKFS_DIRENT_LUFID) {
4356 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
4357 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
4358 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
4364 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
4365 const struct lu_fid *fid)
4367 struct seq_server_site *ss = osd_seq_site(osd);
4370 /* FID seqs not in FLDB, must be local seq */
4371 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
4374 /* If FLD is not being initialized yet, it only happens during the
4375 * initialization, likely during mgs initialization, and we assume
4376 * this is local FID. */
4377 if (ss == NULL || ss->ss_server_fld == NULL)
4380 /* Only check the local FLDB here */
4381 if (osd_seq_exists(env, osd, fid_seq(fid)))
4388 * Index delete function for interoperability mode (b11826).
4389 * It will remove the directory entry added by osd_index_ea_insert().
4390 * This entry is needed to maintain name->fid mapping.
4392 * \param key, key i.e. file entry to be deleted
4394 * \retval 0, on success
4395 * \retval -ve, on error
4397 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
4398 const struct dt_key *key, struct thandle *handle)
4400 struct osd_object *obj = osd_dt_obj(dt);
4401 struct inode *dir = obj->oo_inode;
4402 struct dentry *dentry;
4403 struct osd_thandle *oh;
4404 struct ldiskfs_dir_entry_2 *de = NULL;
4405 struct buffer_head *bh;
4406 struct htree_lock *hlock = NULL;
4407 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
4408 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4412 if (!dt_object_exists(dt))
4415 LINVRNT(osd_invariant(obj));
4416 LASSERT(!dt_object_remote(dt));
4417 LASSERT(handle != NULL);
4419 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4421 oh = container_of(handle, struct osd_thandle, ot_super);
4422 LASSERT(oh->ot_handle != NULL);
4423 LASSERT(oh->ot_handle->h_transaction != NULL);
4425 ll_vfs_dq_init(dir);
4426 dentry = osd_child_dentry_get(env, obj,
4427 (char *)key, strlen((char *)key));
4429 if (obj->oo_hl_head != NULL) {
4430 hlock = osd_oti_get(env)->oti_hlock;
4431 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4432 dir, LDISKFS_HLOCK_DEL);
4434 down_write(&obj->oo_ext_idx_sem);
4437 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4439 /* If this is not the ".." entry, it might be a remote DNE
4440 * entry and we need to check if the FID is for a remote
4441 * MDT. If the FID is not in the directory entry (e.g.
4442 * upgraded 1.8 filesystem without dirdata enabled) then
4443 * we need to get the FID from the LMA. For a remote directory
4444 * there HAS to be an LMA, it cannot be an IGIF inode in this
4447 * Delete the entry before the agent inode in order to
4448 * simplify error handling. At worst an error after deleting
4449 * the entry first might leak the agent inode afterward. The
4450 * reverse would need filesystem abort in case of error deleting
4451 * the entry after the agent had been removed, or leave a
4452 * dangling entry pointing at a random inode. */
4453 if (strcmp((char *)key, dotdot) != 0) {
4454 LASSERT(de != NULL);
4455 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
4456 if (rc == -ENODATA) {
4457 /* can't get FID, postpone to the end of the
4458 * transaction when iget() is safe */
4459 osd_schedule_agent_inode_removal(env, oh,
4460 le32_to_cpu(de->inode));
4461 } else if (rc == 0 &&
4462 unlikely(osd_remote_fid(env, osd, fid))) {
4463 osd_schedule_agent_inode_removal(env, oh,
4464 le32_to_cpu(de->inode));
4467 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
4473 ldiskfs_htree_unlock(hlock);
4475 up_write(&obj->oo_ext_idx_sem);
4480 /* For inode on the remote MDT, .. will point to
4481 * /Agent directory, Check whether it needs to delete
4482 * from agent directory */
4483 if (unlikely(strcmp((char *)key, dotdot) == 0)) {
4486 ret = osd_delete_from_remote_parent(env, osd_obj2dev(obj),
4489 /* Sigh, the entry has been deleted, and
4490 * it is not easy to revert it back, so
4491 * let's keep this error private, and let
4492 * LFSCK fix it. XXX */
4493 CERROR("%s: delete remote parent "DFID": rc = %d\n",
4494 osd_name(osd), PFID(fid), ret);
4497 LASSERT(osd_invariant(obj));
4498 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4503 * Lookup index for \a key and copy record to \a rec.
4505 * \param dt osd index object
4506 * \param key key for index
4507 * \param rec record reference
4509 * \retval +ve success : exact mach
4510 * \retval 0 return record with key not greater than \a key
4511 * \retval -ve failure
4513 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
4514 struct dt_rec *rec, const struct dt_key *key)
4516 struct osd_object *obj = osd_dt_obj(dt);
4517 struct iam_path_descr *ipd;
4518 struct iam_container *bag = &obj->oo_dir->od_container;
4519 struct osd_thread_info *oti = osd_oti_get(env);
4520 struct iam_iterator *it = &oti->oti_idx_it;
4521 struct iam_rec *iam_rec;
4525 if (!dt_object_exists(dt))
4528 LASSERT(osd_invariant(obj));
4529 LASSERT(!dt_object_remote(dt));
4530 LASSERT(bag->ic_object == obj->oo_inode);
4532 ipd = osd_idx_ipd_get(env, bag);
4536 /* got ipd now we can start iterator. */
4537 iam_it_init(it, bag, 0, ipd);
4539 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4540 /* swab quota uid/gid provided by caller */
4541 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4542 key = (const struct dt_key *)&oti->oti_quota_id;
4545 rc = iam_it_get(it, (struct iam_key *)key);
4547 if (S_ISDIR(obj->oo_inode->i_mode))
4548 iam_rec = (struct iam_rec *)oti->oti_ldp;
4550 iam_rec = (struct iam_rec *) rec;
4552 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
4554 if (S_ISDIR(obj->oo_inode->i_mode))
4555 osd_fid_unpack((struct lu_fid *) rec,
4556 (struct osd_fid_pack *)iam_rec);
4557 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
4558 osd_quota_unpack(obj, rec);
4563 osd_ipd_put(env, bag, ipd);
4565 LINVRNT(osd_invariant(obj));
4570 static int osd_index_declare_iam_insert(const struct lu_env *env,
4571 struct dt_object *dt,
4572 const struct dt_rec *rec,
4573 const struct dt_key *key,
4574 struct thandle *handle)
4576 struct osd_thandle *oh;
4578 LASSERT(handle != NULL);
4580 oh = container_of0(handle, struct osd_thandle, ot_super);
4581 LASSERT(oh->ot_handle == NULL);
4583 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4584 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4590 * Inserts (key, value) pair in \a dt index object.
4592 * \param dt osd index object
4593 * \param key key for index
4594 * \param rec record reference
4595 * \param th transaction handler
4598 * \retval -ve failure
4600 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
4601 const struct dt_rec *rec,
4602 const struct dt_key *key, struct thandle *th,
4605 struct osd_object *obj = osd_dt_obj(dt);
4606 struct iam_path_descr *ipd;
4607 struct osd_thandle *oh;
4608 struct iam_container *bag;
4609 struct osd_thread_info *oti = osd_oti_get(env);
4610 struct iam_rec *iam_rec;
4614 if (!dt_object_exists(dt))
4617 LINVRNT(osd_invariant(obj));
4618 LASSERT(!dt_object_remote(dt));
4620 bag = &obj->oo_dir->od_container;
4621 LASSERT(bag->ic_object == obj->oo_inode);
4622 LASSERT(th != NULL);
4624 osd_trans_exec_op(env, th, OSD_OT_INSERT);
4626 ipd = osd_idx_ipd_get(env, bag);
4627 if (unlikely(ipd == NULL))
4630 oh = container_of0(th, struct osd_thandle, ot_super);
4631 LASSERT(oh->ot_handle != NULL);
4632 LASSERT(oh->ot_handle->h_transaction != NULL);
4633 if (S_ISDIR(obj->oo_inode->i_mode)) {
4634 iam_rec = (struct iam_rec *)oti->oti_ldp;
4635 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
4636 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4637 /* pack quota uid/gid */
4638 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4639 key = (const struct dt_key *)&oti->oti_quota_id;
4640 /* pack quota record */
4641 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
4642 iam_rec = (struct iam_rec *)rec;
4644 iam_rec = (struct iam_rec *)rec;
4647 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
4649 osd_ipd_put(env, bag, ipd);
4650 LINVRNT(osd_invariant(obj));
4651 osd_trans_exec_check(env, th, OSD_OT_INSERT);
4656 * Calls ldiskfs_add_entry() to add directory entry
4657 * into the directory. This is required for
4658 * interoperability mode (b11826)
4660 * \retval 0, on success
4661 * \retval -ve, on error
4663 static int __osd_ea_add_rec(struct osd_thread_info *info,
4664 struct osd_object *pobj, struct inode *cinode,
4665 const char *name, const struct lu_fid *fid,
4666 struct htree_lock *hlock, struct thandle *th)
4668 struct ldiskfs_dentry_param *ldp;
4669 struct dentry *child;
4670 struct osd_thandle *oth;
4673 oth = container_of(th, struct osd_thandle, ot_super);
4674 LASSERT(oth->ot_handle != NULL);
4675 LASSERT(oth->ot_handle->h_transaction != NULL);
4676 LASSERT(pobj->oo_inode);
4678 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
4679 if (unlikely(pobj->oo_inode ==
4680 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
4683 osd_get_ldiskfs_dirent_param(ldp, fid);
4684 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
4685 child->d_fsdata = (void *)ldp;
4686 ll_vfs_dq_init(pobj->oo_inode);
4687 rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
4688 child, cinode, hlock);
4689 if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
4690 struct ldiskfs_dir_entry_2 *de;
4691 struct buffer_head *bh;
4694 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
4697 rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
4700 if (S_ISDIR(cinode->i_mode))
4701 de->file_type = LDISKFS_DIRENT_LUFID |
4702 LDISKFS_FT_REG_FILE;
4704 de->file_type = LDISKFS_DIRENT_LUFID |
4706 ldiskfs_handle_dirty_metadata(oth->ot_handle,
4717 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
4718 * into the directory.Also sets flags into osd object to
4719 * indicate dot and dotdot are created. This is required for
4720 * interoperability mode (b11826)
4722 * \param dir directory for dot and dotdot fixup.
4723 * \param obj child object for linking
4725 * \retval 0, on success
4726 * \retval -ve, on error
4728 static int osd_add_dot_dotdot(struct osd_thread_info *info,
4729 struct osd_object *dir,
4730 struct inode *parent_dir, const char *name,
4731 const struct lu_fid *dot_fid,
4732 const struct lu_fid *dot_dot_fid,
4735 struct inode *inode = dir->oo_inode;
4736 struct osd_thandle *oth;
4739 oth = container_of(th, struct osd_thandle, ot_super);
4740 LASSERT(oth->ot_handle->h_transaction != NULL);
4741 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4743 if (strcmp(name, dot) == 0) {
4744 if (dir->oo_compat_dot_created) {
4747 LASSERT(inode->i_ino == parent_dir->i_ino);
4748 dir->oo_compat_dot_created = 1;
4751 } else if (strcmp(name, dotdot) == 0) {
4752 if (!dir->oo_compat_dot_created)
4754 /* in case of rename, dotdot is already created */
4755 if (dir->oo_compat_dotdot_created) {
4756 return __osd_ea_add_rec(info, dir, parent_dir, name,
4757 dot_dot_fid, NULL, th);
4760 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
4761 struct lu_fid tfid = *dot_dot_fid;
4764 result = osd_add_dot_dotdot_internal(info,
4765 dir->oo_inode, parent_dir, dot_fid,
4768 result = osd_add_dot_dotdot_internal(info,
4769 dir->oo_inode, parent_dir, dot_fid,
4774 dir->oo_compat_dotdot_created = 1;
4782 * It will call the appropriate osd_add* function and return the
4783 * value, return by respective functions.
4785 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4786 struct inode *cinode, const char *name,
4787 const struct lu_fid *fid, struct thandle *th)
4789 struct osd_thread_info *info = osd_oti_get(env);
4790 struct htree_lock *hlock;
4793 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4795 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4797 if (hlock != NULL) {
4798 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4801 down_write(&pobj->oo_ext_idx_sem);
4804 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4805 lu_object_fid(&pobj->oo_dt.do_lu),
4808 if (hlock != NULL) {
4809 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4810 pobj->oo_inode, LDISKFS_HLOCK_ADD);
4812 down_write(&pobj->oo_ext_idx_sem);
4815 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4816 struct lu_fid *tfid = &info->oti_fid;
4820 rc = __osd_ea_add_rec(info, pobj, cinode, name,
4823 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4828 ldiskfs_htree_unlock(hlock);
4830 up_write(&pobj->oo_ext_idx_sem);
4836 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4837 struct osd_idmap_cache *oic)
4839 struct osd_scrub *scrub = &dev->od_scrub;
4840 struct lu_fid *fid = &oic->oic_fid;
4841 struct osd_inode_id *id = &oic->oic_lid;
4842 struct inode *inode = NULL;
4848 if (!fid_is_norm(fid) && !fid_is_igif(fid))
4851 if (scrub->os_pos_current > id->oii_ino)
4855 rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
4856 if (rc == -ENOENT) {
4857 __u32 gen = id->oii_gen;
4863 inode = osd_iget(oti, dev, id);
4864 /* The inode has been removed (by race maybe). */
4865 if (IS_ERR(inode)) {
4866 rc = PTR_ERR(inode);
4868 RETURN(rc == -ESTALE ? -ENOENT : rc);
4871 /* The OI mapping is lost. */
4872 if (gen != OSD_OII_NOGEN)
4876 /* The inode may has been reused by others, we do not know,
4877 * leave it to be handled by subsequent osd_fid_lookup(). */
4879 } else if (rc != 0 || osd_id_eq(id, &oti->oti_id)) {
4886 if (thread_is_running(&scrub->os_thread)) {
4887 if (inode == NULL) {
4888 inode = osd_iget(oti, dev, id);
4889 /* The inode has been removed (by race maybe). */
4890 if (IS_ERR(inode)) {
4891 rc = PTR_ERR(inode);
4893 RETURN(rc == -ESTALE ? -ENOENT : rc);
4897 rc = osd_oii_insert(dev, oic, insert);
4898 /* There is race condition between osd_oi_lookup and OI scrub.
4899 * The OI scrub finished just after osd_oi_lookup() failure.
4900 * Under such case, it is unnecessary to trigger OI scrub again,
4901 * but try to call osd_oi_lookup() again. */
4902 if (unlikely(rc == -EAGAIN))
4905 if (!S_ISDIR(inode->i_mode))
4908 rc = osd_check_lmv(oti, dev, inode, oic);
4914 if (!dev->od_noscrub && ++once == 1) {
4915 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
4917 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
4918 "%.16s: trigger partial OI scrub for RPC inconsistency "
4919 "checking FID "DFID": rc = %d\n",
4920 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4922 if (rc == 0 || rc == -EALREADY)
4932 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4933 struct osd_device *dev,
4934 struct osd_idmap_cache *oic,
4935 struct lu_fid *fid, __u32 ino)
4937 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
4938 struct inode *inode;
4941 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4942 inode = osd_iget(oti, dev, &oic->oic_lid);
4943 if (IS_ERR(inode)) {
4944 fid_zero(&oic->oic_fid);
4945 return PTR_ERR(inode);
4948 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4951 fid_zero(&oic->oic_fid);
4953 *fid = oic->oic_fid = lma->lma_self_fid;
4957 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4958 struct osd_inode_id *id, const struct lu_fid *fid)
4960 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4961 id->oii_ino, id->oii_gen, info);
4962 info->oti_cache.oic_lid = *id;
4963 info->oti_cache.oic_fid = *fid;
4964 info->oti_cache.oic_dev = osd;
4968 * Get parent FID from the linkEA.
4970 * For a directory which parent resides on remote MDT, to satisfy the
4971 * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
4972 * the other hand, to make the lookup(..) on the directory can return
4973 * the real parent FID, we append the real parent FID after its ".."
4974 * name entry in the /REMOTE_PARENT_DIR.
4976 * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
4977 * backup. So after the restore, we cannot get the right parent FID from
4978 * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
4979 * we have stored the real parent FID in the directory object's linkEA,
4980 * we can parse the linkEA for the real parent FID.
4982 * \param[in] env pointer to the thread context
4983 * \param[in] obj pointer to the object to be handled
4984 * \param[out]fid pointer to the buffer to hold the parent FID
4986 * \retval 0 for getting the real parent FID successfully
4987 * \retval negative error number on failure
4989 static int osd_get_pfid_from_linkea(const struct lu_env *env,
4990 struct osd_object *obj,
4993 struct osd_thread_info *oti = osd_oti_get(env);
4994 struct lu_buf *buf = &oti->oti_big_buf;
4995 struct dentry *dentry = &oti->oti_obj_dentry;
4996 struct inode *inode = obj->oo_inode;
4997 struct linkea_data ldata = { NULL };
5002 if (!S_ISDIR(inode->i_mode))
5006 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5007 buf->lb_buf, buf->lb_len);
5008 if (rc == -ERANGE) {
5009 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5012 lu_buf_realloc(buf, rc);
5013 if (buf->lb_buf == NULL)
5020 if (unlikely(rc == 0))
5026 if (unlikely(buf->lb_buf == NULL)) {
5027 lu_buf_realloc(buf, rc);
5028 if (buf->lb_buf == NULL)
5035 rc = linkea_init(&ldata);
5037 linkea_first_entry(&ldata);
5038 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5044 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5045 struct inode *inode,
5046 const struct lu_fid *pfid,
5047 const char *name, const int namelen)
5049 struct osd_thread_info *oti = osd_oti_get(env);
5050 struct lu_buf *buf = &oti->oti_big_buf;
5051 struct dentry *dentry = &oti->oti_obj_dentry;
5052 struct linkea_data ldata = { NULL };
5053 struct lu_name cname = { .ln_name = name,
5054 .ln_namelen = namelen };
5059 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5060 buf->lb_buf, buf->lb_len);
5062 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5067 if (unlikely(rc == 0))
5070 if (buf->lb_len < rc) {
5071 lu_buf_realloc(buf, rc);
5072 if (buf->lb_buf == NULL)
5079 rc = linkea_init(&ldata);
5081 rc = linkea_links_find(&ldata, &cname, pfid);
5087 * Calls ->lookup() to find dentry. From dentry get inode and
5088 * read inode's ea to get fid. This is required for interoperability
5091 * \retval 0, on success
5092 * \retval -ve, on error
5094 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5095 struct dt_rec *rec, const struct dt_key *key)
5097 struct inode *dir = obj->oo_inode;
5098 struct dentry *dentry;
5099 struct ldiskfs_dir_entry_2 *de;
5100 struct buffer_head *bh;
5101 struct lu_fid *fid = (struct lu_fid *) rec;
5102 struct htree_lock *hlock = NULL;
5107 LASSERT(dir->i_op != NULL);
5108 LASSERT(dir->i_op->lookup != NULL);
5110 dentry = osd_child_dentry_get(env, obj,
5111 (char *)key, strlen((char *)key));
5113 if (obj->oo_hl_head != NULL) {
5114 hlock = osd_oti_get(env)->oti_hlock;
5115 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5116 dir, LDISKFS_HLOCK_LOOKUP);
5118 down_read(&obj->oo_ext_idx_sem);
5121 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5123 struct osd_thread_info *oti = osd_oti_get(env);
5124 struct osd_inode_id *id = &oti->oti_id;
5125 struct osd_idmap_cache *oic = &oti->oti_cache;
5126 struct osd_device *dev = osd_obj2dev(obj);
5128 ino = le32_to_cpu(de->inode);
5129 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5131 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5135 rc = osd_get_fid_from_dentry(de, rec);
5137 /* done with de, release bh */
5140 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5141 const char *name = (const char *)key;
5143 /* If the parent is on remote MDT, and there
5144 * is no FID-in-dirent, then we have to get
5145 * the parent FID from the linkEA. */
5146 if (likely(strlen(name) == 2 &&
5147 name[0] == '.' && name[1] == '.'))
5148 rc = osd_get_pfid_from_linkea(env, obj,
5151 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5154 osd_id_gen(id, ino, OSD_OII_NOGEN);
5157 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5158 fid_zero(&oic->oic_fid);
5163 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5164 rc = osd_consistency_check(oti, dev, oic);
5166 fid_zero(&oic->oic_fid);
5175 ldiskfs_htree_unlock(hlock);
5177 up_read(&obj->oo_ext_idx_sem);
5182 * Put the osd object once done with it.
5184 * \param obj osd object that needs to be put
5186 static inline void osd_object_put(const struct lu_env *env,
5187 struct osd_object *obj)
5189 lu_object_put(env, &obj->oo_dt.do_lu);
5192 static int osd_index_declare_ea_insert(const struct lu_env *env,
5193 struct dt_object *dt,
5194 const struct dt_rec *rec,
5195 const struct dt_key *key,
5196 struct thandle *handle)
5198 struct osd_thandle *oh;
5199 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5200 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5201 const struct lu_fid *fid = rec1->rec_fid;
5202 int credits, rc = 0;
5203 struct osd_idmap_cache *idc;
5206 LASSERT(!dt_object_remote(dt));
5207 LASSERT(handle != NULL);
5208 LASSERT(fid != NULL);
5209 LASSERT(rec1->rec_type != 0);
5211 oh = container_of0(handle, struct osd_thandle, ot_super);
5212 LASSERT(oh->ot_handle == NULL);
5214 credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5216 /* we can't call iget() while a transactions is running
5217 * (this can lead to a deadlock), but we need to know
5218 * inum and object type. so we find this information at
5219 * declaration and cache in per-thread info */
5220 idc = osd_idc_find_or_init(env, osd, fid);
5222 RETURN(PTR_ERR(idc));
5223 if (idc->oic_remote) {
5224 /* a reference to remote inode is represented by an
5225 * agent inode which we have to create */
5226 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5227 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5230 osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5232 if (osd_dt_obj(dt)->oo_inode != NULL) {
5233 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5235 /* We ignore block quota on meta pool (MDTs), so needn't
5236 * calculate how many blocks will be consumed by this index
5238 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5239 i_gid_read(inode), 0, oh,
5240 osd_dt_obj(dt), true, NULL, false);
5247 * Index add function for interoperability mode (b11826).
5248 * It will add the directory entry.This entry is needed to
5249 * maintain name->fid mapping.
5251 * \param key it is key i.e. file entry to be inserted
5252 * \param rec it is value of given key i.e. fid
5254 * \retval 0, on success
5255 * \retval -ve, on error
5257 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5258 const struct dt_rec *rec,
5259 const struct dt_key *key, struct thandle *th,
5262 struct osd_object *obj = osd_dt_obj(dt);
5263 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5264 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5265 const struct lu_fid *fid = rec1->rec_fid;
5266 const char *name = (const char *)key;
5267 struct osd_thread_info *oti = osd_oti_get(env);
5268 struct inode *child_inode = NULL;
5269 struct osd_idmap_cache *idc;
5273 if (!dt_object_exists(dt))
5276 LASSERT(osd_invariant(obj));
5277 LASSERT(!dt_object_remote(dt));
5278 LASSERT(th != NULL);
5280 osd_trans_exec_op(env, th, OSD_OT_INSERT);
5282 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
5284 idc = osd_idc_find(env, osd, fid);
5285 if (unlikely(idc == NULL)) {
5286 /* this dt_insert() wasn't declared properly, so
5287 * FID is missing in OI cache. we better do not
5288 * lookup FID in FLDB/OI and don't risk to deadlock,
5289 * but in some special cases (lfsck testing, etc)
5290 * it's much simpler than fixing a caller */
5291 CERROR("%s: "DFID" wasn't declared for insert\n",
5292 osd_name(osd), PFID(fid));
5294 idc = osd_idc_find_or_init(env, osd, fid);
5296 RETURN(PTR_ERR(idc));
5299 if (idc->oic_remote) {
5300 /* Insert remote entry */
5301 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
5302 struct osd_mdobj_map *omm = osd->od_mdt_map;
5303 struct osd_thandle *oh;
5305 /* If parent on remote MDT, we need put this object
5307 oh = container_of(th, typeof(*oh), ot_super);
5308 rc = osd_add_to_remote_parent(env, osd, obj, oh);
5310 CERROR("%s: add "DFID" error: rc = %d\n",
5312 PFID(lu_object_fid(&dt->do_lu)), rc);
5316 child_inode = igrab(omm->omm_remote_parent->d_inode);
5318 child_inode = osd_create_local_agent_inode(env, osd,
5319 obj, fid, rec1->rec_type & S_IFMT, th);
5320 if (IS_ERR(child_inode))
5321 RETURN(PTR_ERR(child_inode));
5324 /* Insert local entry */
5325 if (unlikely(idc->oic_lid.oii_ino == 0)) {
5326 /* for a reason OI cache wasn't filled properly */
5327 CERROR("%s: OIC for "DFID" isn't filled\n",
5328 osd_name(osd), PFID(fid));
5331 child_inode = oti->oti_inode;
5332 if (unlikely(child_inode == NULL)) {
5333 struct ldiskfs_inode_info *lii;
5337 child_inode = oti->oti_inode = &lii->vfs_inode;
5339 child_inode->i_sb = osd_sb(osd);
5340 child_inode->i_ino = idc->oic_lid.oii_ino;
5341 child_inode->i_mode = rec1->rec_type & S_IFMT;
5344 rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
5346 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
5347 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
5349 if (child_inode && child_inode != oti->oti_inode)
5351 LASSERT(osd_invariant(obj));
5352 osd_trans_exec_check(env, th, OSD_OT_INSERT);
5357 * Initialize osd Iterator for given osd index object.
5359 * \param dt osd index object
5362 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
5363 struct dt_object *dt,
5366 struct osd_it_iam *it;
5367 struct osd_object *obj = osd_dt_obj(dt);
5368 struct lu_object *lo = &dt->do_lu;
5369 struct iam_path_descr *ipd;
5370 struct iam_container *bag = &obj->oo_dir->od_container;
5372 if (!dt_object_exists(dt))
5373 return ERR_PTR(-ENOENT);
5377 return ERR_PTR(-ENOMEM);
5379 ipd = osd_it_ipd_get(env, bag);
5380 if (likely(ipd != NULL)) {
5384 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
5385 return (struct dt_it *)it;
5388 return ERR_PTR(-ENOMEM);
5393 * free given Iterator.
5396 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
5398 struct osd_it_iam *it = (struct osd_it_iam *)di;
5399 struct osd_object *obj = it->oi_obj;
5401 iam_it_fini(&it->oi_it);
5402 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
5403 lu_object_put(env, &obj->oo_dt.do_lu);
5408 * Move Iterator to record specified by \a key
5410 * \param di osd iterator
5411 * \param key key for index
5413 * \retval +ve di points to record with least key not larger than key
5414 * \retval 0 di points to exact matched key
5415 * \retval -ve failure
5418 static int osd_it_iam_get(const struct lu_env *env,
5419 struct dt_it *di, const struct dt_key *key)
5421 struct osd_thread_info *oti = osd_oti_get(env);
5422 struct osd_it_iam *it = (struct osd_it_iam *)di;
5424 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5425 /* swab quota uid/gid */
5426 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5427 key = (struct dt_key *)&oti->oti_quota_id;
5430 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
5436 * \param di osd iterator
5438 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
5440 struct osd_it_iam *it = (struct osd_it_iam *)di;
5442 iam_it_put(&it->oi_it);
5446 * Move iterator by one record
5448 * \param di osd iterator
5450 * \retval +1 end of container reached
5452 * \retval -ve failure
5455 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
5457 struct osd_it_iam *it = (struct osd_it_iam *)di;
5459 return iam_it_next(&it->oi_it);
5463 * Return pointer to the key under iterator.
5466 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
5467 const struct dt_it *di)
5469 struct osd_thread_info *oti = osd_oti_get(env);
5470 struct osd_it_iam *it = (struct osd_it_iam *)di;
5471 struct osd_object *obj = it->oi_obj;
5474 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
5476 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
5477 /* swab quota uid/gid */
5478 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
5479 key = (struct dt_key *)&oti->oti_quota_id;
5486 * Return size of key under iterator (in bytes)
5489 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
5491 struct osd_it_iam *it = (struct osd_it_iam *)di;
5493 return iam_it_key_size(&it->oi_it);
5497 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
5499 /* check if file type is required */
5500 if (ent->lde_attrs & LUDA_TYPE) {
5501 struct luda_type *lt;
5502 int align = sizeof(*lt) - 1;
5504 len = (len + align) & ~align;
5505 lt = (struct luda_type *)(ent->lde_name + len);
5506 lt->lt_type = cpu_to_le16(DTTOIF(type));
5509 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
5513 * build lu direct from backend fs dirent.
5517 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
5518 char *name, __u16 namelen, __u16 type, __u32 attr)
5520 ent->lde_attrs = attr | LUDA_FID;
5521 fid_cpu_to_le(&ent->lde_fid, fid);
5523 ent->lde_hash = cpu_to_le64(offset);
5524 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
5526 strncpy(ent->lde_name, name, namelen);
5527 ent->lde_name[namelen] = '\0';
5528 ent->lde_namelen = cpu_to_le16(namelen);
5530 /* append lustre attributes */
5531 osd_it_append_attrs(ent, namelen, type);
5535 * Return pointer to the record under iterator.
5537 static int osd_it_iam_rec(const struct lu_env *env,
5538 const struct dt_it *di,
5539 struct dt_rec *dtrec, __u32 attr)
5541 struct osd_it_iam *it = (struct osd_it_iam *)di;
5542 struct osd_thread_info *info = osd_oti_get(env);
5545 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
5546 const struct osd_fid_pack *rec;
5547 struct lu_fid *fid = &info->oti_fid;
5548 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
5554 name = (char *)iam_it_key_get(&it->oi_it);
5556 RETURN(PTR_ERR(name));
5558 namelen = iam_it_key_size(&it->oi_it);
5560 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
5562 RETURN(PTR_ERR(rec));
5564 rc = osd_fid_unpack(fid, rec);
5568 hash = iam_it_store(&it->oi_it);
5570 /* IAM does not store object type in IAM index (dir) */
5571 osd_it_pack_dirent(lde, fid, hash, name, namelen,
5573 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5574 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5575 (struct iam_rec *)dtrec);
5576 osd_quota_unpack(it->oi_obj, dtrec);
5578 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5579 (struct iam_rec *)dtrec);
5586 * Returns cookie for current Iterator position.
5588 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
5590 struct osd_it_iam *it = (struct osd_it_iam *)di;
5592 return iam_it_store(&it->oi_it);
5596 * Restore iterator from cookie.
5598 * \param di osd iterator
5599 * \param hash Iterator location cookie
5601 * \retval +ve di points to record with least key not larger than key.
5602 * \retval 0 di points to exact matched key
5603 * \retval -ve failure
5606 static int osd_it_iam_load(const struct lu_env *env,
5607 const struct dt_it *di, __u64 hash)
5609 struct osd_it_iam *it = (struct osd_it_iam *)di;
5611 return iam_it_load(&it->oi_it, hash);
5614 static const struct dt_index_operations osd_index_iam_ops = {
5615 .dio_lookup = osd_index_iam_lookup,
5616 .dio_declare_insert = osd_index_declare_iam_insert,
5617 .dio_insert = osd_index_iam_insert,
5618 .dio_declare_delete = osd_index_declare_iam_delete,
5619 .dio_delete = osd_index_iam_delete,
5621 .init = osd_it_iam_init,
5622 .fini = osd_it_iam_fini,
5623 .get = osd_it_iam_get,
5624 .put = osd_it_iam_put,
5625 .next = osd_it_iam_next,
5626 .key = osd_it_iam_key,
5627 .key_size = osd_it_iam_key_size,
5628 .rec = osd_it_iam_rec,
5629 .store = osd_it_iam_store,
5630 .load = osd_it_iam_load
5636 * Creates or initializes iterator context.
5638 * \retval struct osd_it_ea, iterator structure on success
5641 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
5642 struct dt_object *dt,
5645 struct osd_object *obj = osd_dt_obj(dt);
5646 struct osd_thread_info *info = osd_oti_get(env);
5647 struct osd_it_ea *oie;
5649 struct lu_object *lo = &dt->do_lu;
5650 struct dentry *obj_dentry;
5653 if (!dt_object_exists(dt) || obj->oo_destroyed)
5654 RETURN(ERR_PTR(-ENOENT));
5656 OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
5658 RETURN(ERR_PTR(-ENOMEM));
5659 obj_dentry = &oie->oie_dentry;
5661 obj_dentry->d_inode = obj->oo_inode;
5662 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
5663 obj_dentry->d_name.hash = 0;
5665 oie->oie_rd_dirent = 0;
5666 oie->oie_it_dirent = 0;
5667 oie->oie_dirent = NULL;
5668 if (unlikely(!info->oti_it_ea_buf_used)) {
5669 oie->oie_buf = info->oti_it_ea_buf;
5670 info->oti_it_ea_buf_used = 1;
5672 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5673 if (oie->oie_buf == NULL)
5674 RETURN(ERR_PTR(-ENOMEM));
5678 file = &oie->oie_file;
5680 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
5681 if (attr & LUDA_64BITHASH)
5682 file->f_mode = FMODE_64BITHASH;
5684 file->f_mode = FMODE_32BITHASH;
5685 file->f_path.dentry = obj_dentry;
5686 file->f_mapping = obj->oo_inode->i_mapping;
5687 file->f_op = obj->oo_inode->i_fop;
5688 set_file_inode(file, obj->oo_inode);
5691 RETURN((struct dt_it *) oie);
5695 * Destroy or finishes iterator context.
5697 * \param di iterator structure to be destroyed
5699 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
5701 struct osd_thread_info *info = osd_oti_get(env);
5702 struct osd_it_ea *oie = (struct osd_it_ea *)di;
5703 struct osd_object *obj = oie->oie_obj;
5704 struct inode *inode = obj->oo_inode;
5707 oie->oie_file.f_op->release(inode, &oie->oie_file);
5708 lu_object_put(env, &obj->oo_dt.do_lu);
5709 if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
5710 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5712 info->oti_it_ea_buf_used = 0;
5713 OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
5718 * It position the iterator at given key, so that next lookup continues from
5719 * that key Or it is similar to dio_it->load() but based on a key,
5720 * rather than file position.
5722 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
5725 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
5727 static int osd_it_ea_get(const struct lu_env *env,
5728 struct dt_it *di, const struct dt_key *key)
5730 struct osd_it_ea *it = (struct osd_it_ea *)di;
5733 LASSERT(((const char *)key)[0] == '\0');
5734 it->oie_file.f_pos = 0;
5735 it->oie_rd_dirent = 0;
5736 it->oie_it_dirent = 0;
5737 it->oie_dirent = NULL;
5745 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
5749 struct osd_filldir_cbs {
5750 #ifdef HAVE_DIR_CONTEXT
5751 struct dir_context ctx;
5753 struct osd_it_ea *it;
5756 * It is called internally by ->readdir(). It fills the
5757 * iterator's in-memory data structure with required
5758 * information i.e. name, namelen, rec_size etc.
5760 * \param buf in which information to be filled in.
5761 * \param name name of the file in given dir
5763 * \retval 0 on success
5764 * \retval 1 on buffer full
5766 #ifdef HAVE_FILLDIR_USE_CTX
5767 static int osd_ldiskfs_filldir(struct dir_context *buf,
5769 static int osd_ldiskfs_filldir(void *buf,
5771 const char *name, int namelen,
5772 loff_t offset, __u64 ino, unsigned d_type)
5774 struct osd_it_ea *it =
5775 ((struct osd_filldir_cbs *)buf)->it;
5776 struct osd_object *obj = it->oie_obj;
5777 struct osd_it_ea_dirent *ent = it->oie_dirent;
5778 struct lu_fid *fid = &ent->oied_fid;
5779 struct osd_fid_pack *rec;
5782 /* this should never happen */
5783 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5784 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5788 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
5792 /* "." is just the object itself. */
5793 if (namelen == 1 && name[0] == '.') {
5794 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5795 } else if (d_type & LDISKFS_DIRENT_LUFID) {
5796 rec = (struct osd_fid_pack*) (name + namelen + 1);
5797 if (osd_fid_unpack(fid, rec) != 0)
5802 d_type &= ~LDISKFS_DIRENT_LUFID;
5804 /* NOT export local root. */
5805 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5806 ino = obj->oo_inode->i_ino;
5807 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5810 ent->oied_ino = ino;
5811 ent->oied_off = offset;
5812 ent->oied_namelen = namelen;
5813 ent->oied_type = d_type;
5815 memcpy(ent->oied_name, name, namelen);
5817 it->oie_rd_dirent++;
5818 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
5823 * Calls ->readdir() to load a directory entry at a time
5824 * and stored it in iterator's in-memory data structure.
5826 * \param di iterator's in memory structure
5828 * \retval 0 on success
5829 * \retval -ve on error
5830 * \retval +1 reach the end of entry
5832 static int osd_ldiskfs_it_fill(const struct lu_env *env,
5833 const struct dt_it *di)
5835 struct osd_it_ea *it = (struct osd_it_ea *)di;
5836 struct osd_object *obj = it->oie_obj;
5837 struct inode *inode = obj->oo_inode;
5838 struct htree_lock *hlock = NULL;
5839 struct file *filp = &it->oie_file;
5841 struct osd_filldir_cbs buf = {
5842 #ifdef HAVE_DIR_CONTEXT
5843 .ctx.actor = osd_ldiskfs_filldir,
5849 it->oie_dirent = it->oie_buf;
5850 it->oie_rd_dirent = 0;
5852 if (obj->oo_hl_head != NULL) {
5853 hlock = osd_oti_get(env)->oti_hlock;
5854 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5855 inode, LDISKFS_HLOCK_READDIR);
5857 down_read(&obj->oo_ext_idx_sem);
5860 #ifdef HAVE_DIR_CONTEXT
5861 buf.ctx.pos = filp->f_pos;
5862 rc = inode->i_fop->iterate(filp, &buf.ctx);
5863 filp->f_pos = buf.ctx.pos;
5865 rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
5869 ldiskfs_htree_unlock(hlock);
5871 up_read(&obj->oo_ext_idx_sem);
5873 if (it->oie_rd_dirent == 0) {
5874 /*If it does not get any dirent, it means it has been reached
5875 *to the end of the dir */
5876 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
5880 it->oie_dirent = it->oie_buf;
5881 it->oie_it_dirent = 1;
5888 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5889 * to load a directory entry at a time and stored it in
5890 * iterator's in-memory data structure.
5892 * \param di iterator's in memory structure
5894 * \retval +ve iterator reached to end
5895 * \retval 0 iterator not reached to end
5896 * \retval -ve on error
5898 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
5900 struct osd_it_ea *it = (struct osd_it_ea *)di;
5905 if (it->oie_it_dirent < it->oie_rd_dirent) {
5907 (void *) it->oie_dirent +
5908 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
5909 it->oie_dirent->oied_namelen);
5910 it->oie_it_dirent++;
5913 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
5916 rc = osd_ldiskfs_it_fill(env, di);
5923 * Returns the key at current position from iterator's in memory structure.
5925 * \param di iterator's in memory structure
5927 * \retval key i.e. struct dt_key on success
5929 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
5930 const struct dt_it *di)
5932 struct osd_it_ea *it = (struct osd_it_ea *)di;
5934 return (struct dt_key *)it->oie_dirent->oied_name;
5938 * Returns the key's size at current position from iterator's in memory structure.
5940 * \param di iterator's in memory structure
5942 * \retval key_size i.e. struct dt_key on success
5944 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
5946 struct osd_it_ea *it = (struct osd_it_ea *)di;
5948 return it->oie_dirent->oied_namelen;
5951 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
5953 if (LDISKFS_DIR_REC_LEN(de) >=
5954 __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
5961 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
5962 unsigned blocksize, bool dotdot)
5965 return osd_dotdot_has_space(de);
5967 if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
5968 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5975 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
5976 handle_t *jh, struct dentry *dentry,
5977 const struct lu_fid *fid, struct buffer_head *bh,
5978 struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
5981 struct inode *dir = dentry->d_parent->d_inode;
5982 struct inode *inode = dentry->d_inode;
5983 struct osd_fid_pack *rec;
5984 struct ldiskfs_dentry_param *ldp;
5985 int namelen = dentry->d_name.len;
5987 struct osd_thread_info *info = osd_oti_get(env);
5990 if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5991 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5994 /* There is enough space to hold the FID-in-dirent. */
5995 if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
5996 rc = ldiskfs_journal_get_write_access(jh, bh);
6000 de->name[namelen] = 0;
6001 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6002 rec->fp_len = sizeof(struct lu_fid) + 1;
6003 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6004 de->file_type |= LDISKFS_DIRENT_LUFID;
6005 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6012 rc = ldiskfs_delete_entry(jh, dir, de, bh);
6016 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6017 osd_get_ldiskfs_dirent_param(ldp, fid);
6018 dentry->d_fsdata = (void *)ldp;
6019 ll_vfs_dq_init(dir);
6020 rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6021 /* It is too bad, we cannot reinsert the name entry back.
6022 * That means we lose it! */
6024 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
6025 "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6026 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
6027 dir->i_ino, dir->i_generation, namelen,
6028 dentry->d_name.name, PFID(fid), rc);
6034 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6035 struct osd_it_ea *it, struct lu_fid *fid,
6036 struct osd_inode_id *id, __u32 *attr)
6038 struct osd_thread_info *info = osd_oti_get(env);
6039 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
6040 struct osd_device *dev = osd_obj2dev(obj);
6041 struct super_block *sb = osd_sb(dev);
6042 const char *devname = osd_name(dev);
6043 struct osd_it_ea_dirent *ent = it->oie_dirent;
6044 struct inode *dir = obj->oo_inode;
6045 struct htree_lock *hlock = NULL;
6046 struct buffer_head *bh = NULL;
6047 handle_t *jh = NULL;
6048 struct ldiskfs_dir_entry_2 *de;
6049 struct dentry *dentry;
6050 struct inode *inode;
6051 const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6054 bool dotdot = false;
6058 if (ent->oied_name[0] == '.') {
6059 if (ent->oied_namelen == 1)
6062 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6066 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6067 inode = osd_iget(info, dev, id);
6068 if (IS_ERR(inode)) {
6069 rc = PTR_ERR(inode);
6070 if (rc == -ENOENT || rc == -ESTALE) {
6071 /* Maybe dangling name entry, or
6072 * corrupted directory entry. */
6073 *attr |= LUDA_UNKNOWN;
6076 CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6077 "check_repair, dir = %lu/%u, name = %.*s, "
6078 "ino = %llu, rc = %d\n",
6079 devname, dir->i_ino, dir->i_generation,
6080 ent->oied_namelen, ent->oied_name,
6087 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6089 rc = osd_get_lma(info, inode, dentry, lma);
6090 if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6095 /* We need to ensure that the name entry is still valid.
6096 * Because it may be removed or renamed by other already.
6098 * The unlink or rename operation will start journal before PDO lock,
6099 * so to avoid deadlock, here we need to start journal handle before
6100 * related PDO lock also. But because we do not know whether there
6101 * will be something to be repaired before PDO lock, we just start
6102 * journal without conditions.
6104 * We may need to remove the name entry firstly, then insert back.
6105 * One credit is for user quota file update.
6106 * One credit is for group quota file update.
6107 * Two credits are for dirty inode. */
6108 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6109 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6111 if (dev->od_dirent_journal != 0) {
6114 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6117 CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6118 "check_repair, dir = %lu/%u, credits = %d, "
6119 "name = %.*s, ino = %llu: rc = %d\n",
6120 devname, dir->i_ino, dir->i_generation, credits,
6121 ent->oied_namelen, ent->oied_name,
6124 GOTO(out_inode, rc);
6127 if (obj->oo_hl_head != NULL) {
6128 hlock = osd_oti_get(env)->oti_hlock;
6129 /* "0" means exclusive lock for the whole directory.
6130 * We need to prevent others access such name entry
6131 * during the delete + insert. Neither HLOCK_ADD nor
6132 * HLOCK_DEL cannot guarantee the atomicity. */
6133 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6135 down_write(&obj->oo_ext_idx_sem);
6138 if (obj->oo_hl_head != NULL) {
6139 hlock = osd_oti_get(env)->oti_hlock;
6140 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6141 LDISKFS_HLOCK_LOOKUP);
6143 down_read(&obj->oo_ext_idx_sem);
6147 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6148 if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6149 *attr |= LUDA_IGNORE;
6154 /* For dotdot entry, if there is not enough space to hold the
6155 * FID-in-dirent, just keep them there. It only happens when the
6156 * device upgraded from 1.8 or restored from MDT file-level backup.
6157 * For the whole directory, only dotdot entry have no FID-in-dirent
6158 * and needs to get FID from LMA when readdir, it will not affect the
6159 * performance much. */
6160 if (dotdot && !osd_dotdot_has_space(de)) {
6161 *attr |= LUDA_UNKNOWN;
6167 if (lu_fid_eq(fid, &lma->lma_self_fid))
6170 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6171 struct lu_fid *tfid = &lma->lma_self_fid;
6173 if (likely(dotdot &&
6174 fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6175 fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6176 /* It must be REMOTE_PARENT_DIR and as the
6177 * 'dotdot' entry of remote directory */
6178 *attr |= LUDA_IGNORE;
6180 CDEBUG(D_LFSCK, "%s: expect remote agent "
6181 "parent directory, but got %.*s under "
6182 "dir = %lu/%u with the FID "DFID"\n",
6183 devname, ent->oied_namelen,
6184 ent->oied_name, dir->i_ino,
6185 dir->i_generation, PFID(tfid));
6187 *attr |= LUDA_UNKNOWN;
6194 if (!fid_is_zero(fid)) {
6195 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6197 if (rc == -ENOENT ||
6199 !(dev->od_scrub.os_file.sf_flags & SF_UPGRADE))) {
6200 /* linkEA does not recognize the dirent entry,
6201 * it may because the dirent entry corruption
6202 * and points to other's inode. */
6203 CDEBUG(D_LFSCK, "%s: the target inode does not "
6204 "recognize the dirent, dir = %lu/%u, "
6205 " name = %.*s, ino = %llu, "
6206 DFID": rc = %d\n", devname, dir->i_ino,
6207 dir->i_generation, ent->oied_namelen,
6208 ent->oied_name, ent->oied_ino, PFID(fid), rc);
6209 *attr |= LUDA_UNKNOWN;
6214 if (rc && rc != -ENODATA) {
6215 CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6216 "dir = %lu/%u, name = %.*s, ino = %llu, "
6217 DFID": rc = %d\n", devname, dir->i_ino,
6218 dir->i_generation, ent->oied_namelen,
6219 ent->oied_name, ent->oied_ino, PFID(fid), rc);
6220 *attr |= LUDA_UNKNOWN;
6227 /* linkEA recognizes the dirent entry, the FID-in-LMA is
6228 * valid, trusted, in spite of fid_is_sane(fid) or not. */
6229 if (*attr & LUDA_VERIFY_DRYRUN) {
6230 *fid = lma->lma_self_fid;
6231 *attr |= LUDA_REPAIR;
6238 dev->od_dirent_journal = 1;
6239 if (hlock != NULL) {
6240 ldiskfs_htree_unlock(hlock);
6243 up_read(&obj->oo_ext_idx_sem);
6249 *fid = lma->lma_self_fid;
6251 /* Update or append the FID-in-dirent. */
6252 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6253 bh, de, hlock, dotdot);
6255 *attr |= LUDA_REPAIR;
6257 CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
6258 "the dirent, dir = %lu/%u, name = %.*s, "
6259 "ino = %llu, "DFID": rc = %d\n",
6260 devname, dir->i_ino, dir->i_generation,
6261 ent->oied_namelen, ent->oied_name,
6262 ent->oied_ino, PFID(fid), rc);
6264 /* lma is NULL, trust the FID-in-dirent if it is valid. */
6265 if (*attr & LUDA_VERIFY_DRYRUN) {
6266 if (fid_is_sane(fid)) {
6267 *attr |= LUDA_REPAIR;
6268 } else if (dev->od_index == 0) {
6269 lu_igif_build(fid, inode->i_ino,
6270 inode->i_generation);
6271 *attr |= LUDA_UPGRADE;
6279 dev->od_dirent_journal = 1;
6280 if (hlock != NULL) {
6281 ldiskfs_htree_unlock(hlock);
6284 up_read(&obj->oo_ext_idx_sem);
6291 if (unlikely(fid_is_sane(fid))) {
6292 /* FID-in-dirent exists, but FID-in-LMA is lost.
6293 * Trust the FID-in-dirent, and add FID-in-LMA. */
6294 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
6296 *attr |= LUDA_REPAIR;
6298 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
6299 "update dirent, dir = %lu/%u, "
6300 "name = %.*s, ino = %llu, "
6302 devname, dir->i_ino, dir->i_generation,
6303 ent->oied_namelen, ent->oied_name,
6304 ent->oied_ino, PFID(fid), rc);
6305 } else if (dev->od_index == 0) {
6306 lu_igif_build(fid, inode->i_ino, inode->i_generation);
6307 /* It is probably IGIF object. Only aappend the
6308 * FID-in-dirent. OI scrub will process FID-in-LMA. */
6309 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6310 bh, de, hlock, dotdot);
6312 *attr |= LUDA_UPGRADE;
6314 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
6315 "after the 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);
6329 if (hlock != NULL) {
6330 ldiskfs_htree_unlock(hlock);
6332 if (dev->od_dirent_journal != 0)
6333 up_write(&obj->oo_ext_idx_sem);
6335 up_read(&obj->oo_ext_idx_sem);
6339 ldiskfs_journal_stop(jh);
6343 if (rc >= 0 && !dirty)
6344 dev->od_dirent_journal = 0;
6350 * Returns the value at current position from iterator's in memory structure.
6352 * \param di struct osd_it_ea, iterator's in memory structure
6353 * \param attr attr requested for dirent.
6354 * \param lde lustre dirent
6356 * \retval 0 no error and \param lde has correct lustre dirent.
6357 * \retval -ve on error
6359 static inline int osd_it_ea_rec(const struct lu_env *env,
6360 const struct dt_it *di,
6361 struct dt_rec *dtrec, __u32 attr)
6363 struct osd_it_ea *it = (struct osd_it_ea *)di;
6364 struct osd_object *obj = it->oie_obj;
6365 struct osd_device *dev = osd_obj2dev(obj);
6366 struct osd_thread_info *oti = osd_oti_get(env);
6367 struct osd_inode_id *id = &oti->oti_id;
6368 struct lu_fid *fid = &it->oie_dirent->oied_fid;
6369 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6370 __u32 ino = it->oie_dirent->oied_ino;
6374 LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
6376 if (attr & LUDA_VERIFY) {
6377 if (unlikely(ino == osd_remote_parent_ino(dev))) {
6378 attr |= LUDA_IGNORE;
6379 /* If the parent is on remote MDT, and there
6380 * is no FID-in-dirent, then we have to get
6381 * the parent FID from the linkEA. */
6382 if (!fid_is_sane(fid) &&
6383 it->oie_dirent->oied_namelen == 2 &&
6384 it->oie_dirent->oied_name[0] == '.' &&
6385 it->oie_dirent->oied_name[1] == '.')
6386 osd_get_pfid_from_linkea(env, obj, fid);
6388 rc = osd_dirent_check_repair(env, obj, it, fid, id,
6392 if (!fid_is_sane(fid))
6393 attr |= LUDA_UNKNOWN;
6395 attr &= ~LU_DIRENT_ATTRS_MASK;
6396 if (!fid_is_sane(fid)) {
6397 bool is_dotdot = false;
6398 if (it->oie_dirent->oied_namelen == 2 &&
6399 it->oie_dirent->oied_name[0] == '.' &&
6400 it->oie_dirent->oied_name[1] == '.')
6402 /* If the parent is on remote MDT, and there
6403 * is no FID-in-dirent, then we have to get
6404 * the parent FID from the linkEA. */
6405 if (ino == osd_remote_parent_ino(dev) && is_dotdot) {
6406 rc = osd_get_pfid_from_linkea(env, obj, fid);
6408 if (is_dotdot == false &&
6409 OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
6412 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6415 osd_id_gen(id, ino, OSD_OII_NOGEN);
6419 /* Pack the entry anyway, at least the offset is right. */
6420 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
6421 it->oie_dirent->oied_name,
6422 it->oie_dirent->oied_namelen,
6423 it->oie_dirent->oied_type, attr);
6428 if (osd_remote_fid(env, dev, fid))
6431 if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
6432 osd_add_oi_cache(oti, dev, id, fid);
6434 RETURN(rc > 0 ? 0 : rc);
6438 * Returns the record size size at current position.
6440 * This function will return record(lu_dirent) size in bytes.
6442 * \param[in] env execution environment
6443 * \param[in] di iterator's in memory structure
6444 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
6445 * calculate the lu_dirent size.
6447 * \retval record size(in bytes & in memory) of the current lu_dirent
6450 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
6453 struct osd_it_ea *it = (struct osd_it_ea *)di;
6455 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
6459 * Returns a cookie for current position of the iterator head, so that
6460 * user can use this cookie to load/start the iterator next time.
6462 * \param di iterator's in memory structure
6464 * \retval cookie for current position, on success
6466 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
6468 struct osd_it_ea *it = (struct osd_it_ea *)di;
6470 return it->oie_dirent->oied_off;
6474 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6475 * to load a directory entry at a time and stored it i inn,
6476 * in iterator's in-memory data structure.
6478 * \param di struct osd_it_ea, iterator's in memory structure
6480 * \retval +ve on success
6481 * \retval -ve on error
6483 static int osd_it_ea_load(const struct lu_env *env,
6484 const struct dt_it *di, __u64 hash)
6486 struct osd_it_ea *it = (struct osd_it_ea *)di;
6490 it->oie_file.f_pos = hash;
6492 rc = osd_ldiskfs_it_fill(env, di);
6503 * Index lookup function for interoperability mode (b11826).
6505 * \param key, key i.e. file name to be searched
6507 * \retval +ve, on success
6508 * \retval -ve, on error
6510 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
6511 struct dt_rec *rec, const struct dt_key *key)
6513 struct osd_object *obj = osd_dt_obj(dt);
6518 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
6519 LINVRNT(osd_invariant(obj));
6521 rc = osd_ea_lookup_rec(env, obj, rec, key);
6528 * Index and Iterator operations for interoperability
6529 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
6531 static const struct dt_index_operations osd_index_ea_ops = {
6532 .dio_lookup = osd_index_ea_lookup,
6533 .dio_declare_insert = osd_index_declare_ea_insert,
6534 .dio_insert = osd_index_ea_insert,
6535 .dio_declare_delete = osd_index_declare_ea_delete,
6536 .dio_delete = osd_index_ea_delete,
6538 .init = osd_it_ea_init,
6539 .fini = osd_it_ea_fini,
6540 .get = osd_it_ea_get,
6541 .put = osd_it_ea_put,
6542 .next = osd_it_ea_next,
6543 .key = osd_it_ea_key,
6544 .key_size = osd_it_ea_key_size,
6545 .rec = osd_it_ea_rec,
6546 .rec_size = osd_it_ea_rec_size,
6547 .store = osd_it_ea_store,
6548 .load = osd_it_ea_load
6552 static void *osd_key_init(const struct lu_context *ctx,
6553 struct lu_context_key *key)
6555 struct osd_thread_info *info;
6557 OBD_ALLOC_PTR(info);
6559 return ERR_PTR(-ENOMEM);
6561 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6562 if (info->oti_it_ea_buf == NULL)
6565 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
6567 info->oti_hlock = ldiskfs_htree_lock_alloc();
6568 if (info->oti_hlock == NULL)
6574 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6577 return ERR_PTR(-ENOMEM);
6580 static void osd_key_fini(const struct lu_context *ctx,
6581 struct lu_context_key *key, void* data)
6583 struct osd_thread_info *info = data;
6584 struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
6585 struct osd_idmap_cache *idc = info->oti_ins_cache;
6587 if (info->oti_inode != NULL)
6589 if (info->oti_hlock != NULL)
6590 ldiskfs_htree_lock_free(info->oti_hlock);
6591 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6592 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
6593 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
6594 lu_buf_free(&info->oti_big_buf);
6596 LASSERT(info->oti_ins_cache_size > 0);
6597 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
6598 info->oti_ins_cache = NULL;
6599 info->oti_ins_cache_size = 0;
6604 static void osd_key_exit(const struct lu_context *ctx,
6605 struct lu_context_key *key, void *data)
6607 struct osd_thread_info *info = data;
6609 LASSERT(info->oti_r_locks == 0);
6610 LASSERT(info->oti_w_locks == 0);
6611 LASSERT(info->oti_txns == 0);
6614 /* type constructor/destructor: osd_type_init, osd_type_fini */
6615 LU_TYPE_INIT_FINI(osd, &osd_key);
6617 struct lu_context_key osd_key = {
6618 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
6619 .lct_init = osd_key_init,
6620 .lct_fini = osd_key_fini,
6621 .lct_exit = osd_key_exit
6625 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
6626 const char *name, struct lu_device *next)
6628 struct osd_device *osd = osd_dev(d);
6630 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
6631 >= sizeof(osd->od_svname))
6633 return osd_procfs_init(osd, name);
6636 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
6638 struct seq_server_site *ss = osd_seq_site(osd);
6642 if (osd->od_is_ost || osd->od_cl_seq != NULL)
6645 if (unlikely(ss == NULL))
6648 OBD_ALLOC_PTR(osd->od_cl_seq);
6649 if (osd->od_cl_seq == NULL)
6652 rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
6653 osd->od_svname, ss->ss_server_seq);
6655 OBD_FREE_PTR(osd->od_cl_seq);
6656 osd->od_cl_seq = NULL;
6660 if (ss->ss_node_id == 0) {
6661 /* If the OSD on the sequence controller(MDT0), then allocate
6662 * sequence here, otherwise allocate sequence after connected
6663 * to MDT0 (see mdt_register_lwp_callback()). */
6664 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
6665 &osd->od_cl_seq->lcs_space, env);
6671 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
6673 if (osd->od_cl_seq == NULL)
6676 seq_client_fini(osd->od_cl_seq);
6677 OBD_FREE_PTR(osd->od_cl_seq);
6678 osd->od_cl_seq = NULL;
6681 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
6685 /* shutdown quota slave instance associated with the device */
6686 if (o->od_quota_slave != NULL) {
6687 qsd_fini(env, o->od_quota_slave);
6688 o->od_quota_slave = NULL;
6691 osd_fid_fini(env, o);
6696 static void osd_umount(const struct lu_env *env, struct osd_device *o)
6700 if (o->od_mnt != NULL) {
6701 shrink_dcache_sb(osd_sb(o));
6702 osd_sync(env, &o->od_dt_dev);
6711 static int osd_mount(const struct lu_env *env,
6712 struct osd_device *o, struct lustre_cfg *cfg)
6714 const char *name = lustre_cfg_string(cfg, 0);
6715 const char *dev = lustre_cfg_string(cfg, 1);
6717 unsigned long page, s_flags, lmd_flags = 0;
6718 struct page *__page;
6719 struct file_system_type *type;
6720 char *options = NULL;
6722 struct osd_thread_info *info = osd_oti_get(env);
6723 struct lu_fid *fid = &info->oti_fid;
6724 struct inode *inode;
6725 int rc = 0, force_over_256tb = 0;
6728 if (o->od_mnt != NULL)
6731 if (strlen(dev) >= sizeof(o->od_mntdev))
6733 strcpy(o->od_mntdev, dev);
6735 str = lustre_cfg_string(cfg, 2);
6736 s_flags = simple_strtoul(str, NULL, 0);
6737 str = strstr(str, ":");
6739 lmd_flags = simple_strtoul(str + 1, NULL, 0);
6740 opts = lustre_cfg_string(cfg, 3);
6742 if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
6743 CERROR("%s: device %s extents feature is not guaranteed to "
6744 "work on big-endian systems. Use \"bigendian_extents\" "
6745 "mount option to override.\n", name, dev);
6749 if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
6750 CWARN("force_over_128tb option is depricated."
6751 "Filesystems less then 256TB can be created without any"
6752 "force options. Use force_over_256tb option for"
6753 "filesystems greather then 256TB.\n");
6756 if (opts != NULL && strstr(opts, "force_over_256tb") != NULL)
6757 force_over_256tb = 1;
6759 __page = alloc_page(GFP_KERNEL);
6761 GOTO(out, rc = -ENOMEM);
6762 page = (unsigned long)page_address(__page);
6763 options = (char *)page;
6766 /* strip out the options for back compatiblity */
6767 static char *sout[] = {
6774 /* strip out option we processed in osd */
6775 "bigendian_extents",
6776 #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(3,0,53,0)
6777 #warning "remove force_over_128 option"
6779 "force_over_128tb (deprecated)",
6784 strcat(options, opts);
6785 for (rc = 0, str = options; sout[rc]; ) {
6786 char *op = strstr(str, sout[rc]);
6792 if (op == options || *(op - 1) == ',') {
6793 str = op + strlen(sout[rc]);
6794 if (*str == ',' || *str == '\0') {
6795 *str == ',' ? str++ : str;
6796 memmove(op, str, strlen(str) + 1);
6799 for (str = op; *str != ',' && *str != '\0'; str++)
6803 strncat(options, "user_xattr,acl", 14);
6806 /* Glom up mount options */
6807 if (*options != '\0')
6808 strcat(options, ",");
6809 strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
6811 type = get_fs_type("ldiskfs");
6813 CERROR("%s: cannot find ldiskfs module\n", name);
6814 GOTO(out, rc = -ENODEV);
6817 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
6818 module_put(type->owner);
6820 if (IS_ERR(o->od_mnt)) {
6821 rc = PTR_ERR(o->od_mnt);
6823 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
6827 if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) > (64ULL << 30) &&
6828 force_over_256tb == 0) {
6829 CERROR("%s: device %s LDISKFS does not support filesystems "
6830 "greater than 256TB and can cause data corruption. "
6831 "Use \"force_over_256tb\" mount option to override.\n",
6833 GOTO(out, rc = -EINVAL);
6836 #ifdef HAVE_DEV_SET_RDONLY
6837 if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
6838 CERROR("%s: underlying device %s is marked as read-only. "
6839 "Setup failed\n", name, dev);
6840 GOTO(out_mnt, rc = -EROFS);
6844 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
6845 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
6846 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
6847 GOTO(out_mnt, rc = -EINVAL);
6850 #ifdef LDISKFS_MOUNT_DIRDATA
6851 if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
6852 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6853 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
6854 else if (!o->od_is_ost)
6855 CWARN("%s: device %s was upgraded from Lustre-1.x without "
6856 "enabling the dirdata feature. If you do not want to "
6857 "downgrade to Lustre-1.x again, you can enable it via "
6858 "'tune2fs -O dirdata device'\n", name, dev);
6860 inode = osd_sb(o)->s_root->d_inode;
6861 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
6862 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
6864 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
6868 if (lmd_flags & LMD_FLG_NOSCRUB)
6879 __free_page(__page);
6884 static struct lu_device *osd_device_fini(const struct lu_env *env,
6885 struct lu_device *d)
6887 struct osd_device *o = osd_dev(d);
6890 osd_shutdown(env, o);
6892 osd_scrub_cleanup(env, o);
6893 osd_obj_map_fini(o);
6899 static int osd_device_init0(const struct lu_env *env,
6900 struct osd_device *o,
6901 struct lustre_cfg *cfg)
6903 struct lu_device *l = osd2lu_dev(o);
6904 struct osd_thread_info *info;
6908 /* if the module was re-loaded, env can loose its keys */
6909 rc = lu_env_refill((struct lu_env *) env);
6912 info = osd_oti_get(env);
6915 l->ld_ops = &osd_lu_ops;
6916 o->od_dt_dev.dd_ops = &osd_dt_ops;
6918 spin_lock_init(&o->od_osfs_lock);
6919 mutex_init(&o->od_otable_mutex);
6920 INIT_LIST_HEAD(&o->od_orphan_list);
6922 o->od_read_cache = 1;
6923 o->od_writethrough_cache = 1;
6924 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
6926 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
6927 sizeof(o->od_svname));
6928 if (cplen >= sizeof(o->od_svname)) {
6933 o->od_index = -1; /* -1 means index is invalid */
6934 rc = server_name2index(o->od_svname, &o->od_index, NULL);
6935 if (rc == LDD_F_SV_TYPE_OST)
6938 o->od_full_scrub_ratio = OFSR_DEFAULT;
6939 o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
6940 rc = osd_mount(env, o, cfg);
6944 rc = osd_obj_map_init(env, o);
6948 rc = lu_site_init(&o->od_site, l);
6950 GOTO(out_compat, rc);
6951 o->od_site.ls_bottom_dev = l;
6953 rc = lu_site_init_finish(&o->od_site);
6957 INIT_LIST_HEAD(&o->od_ios_list);
6958 /* setup scrub, including OI files initialization */
6959 rc = osd_scrub_setup(env, o);
6963 rc = osd_procfs_init(o, o->od_svname);
6965 CERROR("%s: can't initialize procfs: rc = %d\n",
6967 GOTO(out_scrub, rc);
6970 LASSERT(l->ld_site->ls_linkage.next != NULL);
6971 LASSERT(l->ld_site->ls_linkage.prev != NULL);
6973 /* initialize quota slave instance */
6974 o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
6976 if (IS_ERR(o->od_quota_slave)) {
6977 rc = PTR_ERR(o->od_quota_slave);
6978 o->od_quota_slave = NULL;
6979 GOTO(out_procfs, rc);
6987 osd_scrub_cleanup(env, o);
6989 lu_site_fini(&o->od_site);
6991 osd_obj_map_fini(o);
6998 static struct lu_device *osd_device_alloc(const struct lu_env *env,
6999 struct lu_device_type *t,
7000 struct lustre_cfg *cfg)
7002 struct osd_device *o;
7007 return ERR_PTR(-ENOMEM);
7009 rc = dt_device_init(&o->od_dt_dev, t);
7011 /* Because the ctx might be revived in dt_device_init,
7012 * refill the env here */
7013 lu_env_refill((struct lu_env *)env);
7014 rc = osd_device_init0(env, o, cfg);
7016 dt_device_fini(&o->od_dt_dev);
7019 if (unlikely(rc != 0))
7022 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7025 static struct lu_device *osd_device_free(const struct lu_env *env,
7026 struct lu_device *d)
7028 struct osd_device *o = osd_dev(d);
7031 /* XXX: make osd top device in order to release reference */
7032 d->ld_site->ls_top_dev = d;
7033 lu_site_purge(env, d->ld_site, -1);
7034 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7035 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7036 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7038 lu_site_fini(&o->od_site);
7039 dt_device_fini(&o->od_dt_dev);
7044 static int osd_process_config(const struct lu_env *env,
7045 struct lu_device *d, struct lustre_cfg *cfg)
7047 struct osd_device *o = osd_dev(d);
7051 switch (cfg->lcfg_command) {
7053 rc = osd_mount(env, o, cfg);
7056 lu_dev_del_linkage(d->ld_site, d);
7057 rc = osd_shutdown(env, o);
7060 LASSERT(&o->od_dt_dev);
7061 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
7062 cfg, &o->od_dt_dev);
7063 if (rc > 0 || rc == -ENOSYS)
7064 rc = class_process_proc_param(PARAM_OST,
7065 lprocfs_osd_obd_vars,
7066 cfg, &o->od_dt_dev);
7075 static int osd_recovery_complete(const struct lu_env *env,
7076 struct lu_device *d)
7078 struct osd_device *osd = osd_dev(d);
7082 if (osd->od_quota_slave == NULL)
7085 /* start qsd instance on recovery completion, this notifies the quota
7086 * slave code that we are about to process new requests now */
7087 rc = qsd_start(env, osd->od_quota_slave);
7092 * we use exports to track all osd users
7094 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7095 struct obd_device *obd, struct obd_uuid *cluuid,
7096 struct obd_connect_data *data, void *localdata)
7098 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7099 struct lustre_handle conn;
7103 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7105 rc = class_connect(&conn, obd, cluuid);
7109 *exp = class_conn2export(&conn);
7111 spin_lock(&osd->od_osfs_lock);
7113 spin_unlock(&osd->od_osfs_lock);
7119 * once last export (we don't count self-export) disappeared
7120 * osd can be released
7122 static int osd_obd_disconnect(struct obd_export *exp)
7124 struct obd_device *obd = exp->exp_obd;
7125 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7126 int rc, release = 0;
7129 /* Only disconnect the underlying layers on the final disconnect. */
7130 spin_lock(&osd->od_osfs_lock);
7132 if (osd->od_connects == 0)
7134 spin_unlock(&osd->od_osfs_lock);
7136 rc = class_disconnect(exp); /* bz 9811 */
7138 if (rc == 0 && release)
7139 class_manual_cleanup(obd);
7143 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
7144 struct lu_device *dev)
7146 struct osd_device *osd = osd_dev(dev);
7147 struct lr_server_data *lsd =
7148 &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
7152 if (osd->od_quota_slave != NULL) {
7153 /* set up quota slave objects */
7154 result = qsd_prepare(env, osd->od_quota_slave);
7159 if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
7160 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
7161 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
7162 osd->od_index_in_idif = 1;
7164 osd->od_index_in_idif = 0;
7165 result = osd_register_proc_index_in_idif(osd);
7170 osd->od_index_in_idif = 1;
7174 result = osd_fid_init(env, osd);
7179 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
7180 struct lu_fid *fid, struct md_op_data *op_data)
7182 struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
7184 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
7187 static const struct lu_object_operations osd_lu_obj_ops = {
7188 .loo_object_init = osd_object_init,
7189 .loo_object_delete = osd_object_delete,
7190 .loo_object_release = osd_object_release,
7191 .loo_object_free = osd_object_free,
7192 .loo_object_print = osd_object_print,
7193 .loo_object_invariant = osd_object_invariant
7196 const struct lu_device_operations osd_lu_ops = {
7197 .ldo_object_alloc = osd_object_alloc,
7198 .ldo_process_config = osd_process_config,
7199 .ldo_recovery_complete = osd_recovery_complete,
7200 .ldo_prepare = osd_prepare,
7203 static const struct lu_device_type_operations osd_device_type_ops = {
7204 .ldto_init = osd_type_init,
7205 .ldto_fini = osd_type_fini,
7207 .ldto_start = osd_type_start,
7208 .ldto_stop = osd_type_stop,
7210 .ldto_device_alloc = osd_device_alloc,
7211 .ldto_device_free = osd_device_free,
7213 .ldto_device_init = osd_device_init,
7214 .ldto_device_fini = osd_device_fini
7217 static struct lu_device_type osd_device_type = {
7218 .ldt_tags = LU_DEVICE_DT,
7219 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
7220 .ldt_ops = &osd_device_type_ops,
7221 .ldt_ctx_tags = LCT_LOCAL,
7224 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
7226 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7227 struct super_block *sb = osd_sb(osd);
7229 return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
7233 * lprocfs legacy support.
7235 static struct obd_ops osd_obd_device_ops = {
7236 .o_owner = THIS_MODULE,
7237 .o_connect = osd_obd_connect,
7238 .o_disconnect = osd_obd_disconnect,
7239 .o_fid_alloc = osd_fid_alloc,
7240 .o_health_check = osd_health_check,
7243 static int __init osd_init(void)
7247 LASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
7248 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
7249 /* please, try to keep osd_thread_info smaller than a page */
7250 CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
7255 rc = lu_kmem_init(ldiskfs_caches);
7259 rc = class_register_type(&osd_obd_device_ops, NULL, true,
7260 lprocfs_osd_module_vars,
7261 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
7263 lu_kmem_fini(ldiskfs_caches);
7267 static void __exit osd_exit(void)
7269 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
7270 lu_kmem_fini(ldiskfs_caches);
7273 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7274 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
7275 MODULE_VERSION(LUSTRE_VERSION_STRING);
7276 MODULE_LICENSE("GPL");
7278 module_init(osd_init);
7279 module_exit(osd_exit);