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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osd/osd_handler.c
38 * Top-level entry points into osd module
40 * Author: Nikita Danilov <nikita@clusterfs.com>
41 * Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
44 #define DEBUG_SUBSYSTEM S_OSD
46 #include <linux/module.h>
47 #include <linux/user_namespace.h>
48 #ifdef HAVE_UIDGID_HEADER
49 # include <linux/uidgid.h>
52 /* LUSTRE_VERSION_CODE */
53 #include <lustre_ver.h>
54 /* prerequisite for linux/xattr.h */
55 #include <linux/types.h>
56 /* prerequisite for linux/xattr.h */
58 /* XATTR_{REPLACE,CREATE} */
59 #include <linux/xattr.h>
61 #include <ldiskfs/ldiskfs.h>
62 #include <ldiskfs/xattr.h>
63 #include <ldiskfs/ldiskfs_extents.h>
66 * struct OBD_{ALLOC,FREE}*()
69 #include <obd_support.h>
70 /* struct ptlrpc_thread */
71 #include <lustre_net.h>
72 #include <lustre_fid.h>
74 #include <lustre_param.h>
76 #include "osd_internal.h"
77 #include "osd_dynlocks.h"
79 /* llo_* api support */
80 #include <md_object.h>
81 #include <lustre_quota.h>
83 #include <lustre_linkea.h>
86 module_param(ldiskfs_pdo, int, 0644);
87 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
89 int ldiskfs_track_declares_assert;
90 module_param(ldiskfs_track_declares_assert, int, 0644);
91 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
93 /* Slab to allocate dynlocks */
94 struct kmem_cache *dynlock_cachep;
96 /* Slab to allocate osd_it_ea */
97 struct kmem_cache *osd_itea_cachep;
99 static struct lu_kmem_descr ldiskfs_caches[] = {
101 .ckd_cache = &dynlock_cachep,
102 .ckd_name = "dynlock_cache",
103 .ckd_size = sizeof(struct dynlock_handle)
106 .ckd_cache = &osd_itea_cachep,
107 .ckd_name = "osd_itea_cache",
108 .ckd_size = sizeof(struct osd_it_ea)
115 static const char dot[] = ".";
116 static const char dotdot[] = "..";
117 static const char remote_obj_dir[] = "REM_OBJ_DIR";
119 static const struct lu_object_operations osd_lu_obj_ops;
120 static const struct dt_object_operations osd_obj_ops;
121 static const struct dt_object_operations osd_obj_ea_ops;
122 static const struct dt_object_operations osd_obj_otable_it_ops;
123 static const struct dt_index_operations osd_index_iam_ops;
124 static const struct dt_index_operations osd_index_ea_ops;
126 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
127 const struct lu_fid *fid);
128 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
129 struct osd_device *osd);
131 int osd_trans_declare_op2rb[] = {
132 [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
133 [OSD_OT_PUNCH] = OSD_OT_MAX,
134 [OSD_OT_XATTR_SET] = OSD_OT_XATTR_SET,
135 [OSD_OT_CREATE] = OSD_OT_DESTROY,
136 [OSD_OT_DESTROY] = OSD_OT_CREATE,
137 [OSD_OT_REF_ADD] = OSD_OT_REF_DEL,
138 [OSD_OT_REF_DEL] = OSD_OT_REF_ADD,
139 [OSD_OT_WRITE] = OSD_OT_WRITE,
140 [OSD_OT_INSERT] = OSD_OT_DELETE,
141 [OSD_OT_DELETE] = OSD_OT_INSERT,
142 [OSD_OT_QUOTA] = OSD_OT_MAX,
145 static int osd_has_index(const struct osd_object *obj)
147 return obj->oo_dt.do_index_ops != NULL;
150 static int osd_object_invariant(const struct lu_object *l)
152 return osd_invariant(osd_obj(l));
156 * Concurrency: doesn't matter
160 * Concurrency: doesn't matter
162 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
164 struct osd_thread_info *oti = osd_oti_get(env);
165 return oti->oti_w_locks > 0 && o->oo_owner == env;
169 * Concurrency: doesn't access mutable data
171 static int osd_root_get(const struct lu_env *env,
172 struct dt_device *dev, struct lu_fid *f)
174 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
179 * the following set of functions are used to maintain per-thread
180 * cache of FID->ino mapping. this mechanism is needed to resolve
181 * FID to inode at dt_insert() which in turn stores ino in the
182 * directory entries to keep ldiskfs compatible with ext[34].
183 * due to locking-originated restrictions we can't lookup ino
184 * using LU cache (deadlock is possible). lookup using OI is quite
185 * expensive. so instead we maintain this cache and methods like
186 * dt_create() fill it. so in the majority of cases dt_insert() is
187 * able to find needed mapping in lockless manner.
189 static struct osd_idmap_cache *
190 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
191 const struct lu_fid *fid)
193 struct osd_thread_info *oti = osd_oti_get(env);
194 struct osd_idmap_cache *idc = oti->oti_ins_cache;
196 for (i = 0; i < oti->oti_ins_cache_used; i++) {
197 if (!lu_fid_eq(&idc[i].oic_fid, fid))
199 if (idc[i].oic_dev != osd)
208 static struct osd_idmap_cache *
209 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
210 const struct lu_fid *fid)
212 struct osd_thread_info *oti = osd_oti_get(env);
213 struct osd_idmap_cache *idc;
216 if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
217 i = oti->oti_ins_cache_size * 2;
219 i = OSD_INS_CACHE_SIZE;
220 OBD_ALLOC(idc, sizeof(*idc) * i);
222 return ERR_PTR(-ENOMEM);
223 if (oti->oti_ins_cache != NULL) {
224 memcpy(idc, oti->oti_ins_cache,
225 oti->oti_ins_cache_used * sizeof(*idc));
226 OBD_FREE(oti->oti_ins_cache,
227 oti->oti_ins_cache_used * sizeof(*idc));
229 oti->oti_ins_cache = idc;
230 oti->oti_ins_cache_size = i;
233 idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
236 idc->oic_lid.oii_ino = 0;
237 idc->oic_lid.oii_gen = 0;
244 * lookup mapping for the given fid in the cache, initialize a
245 * new one if not found. the initialization checks whether the
246 * object is local or remote. for local objects, OI is used to
247 * learn ino/generation. the function is used when the caller
248 * has no information about the object, e.g. at dt_insert().
250 static struct osd_idmap_cache *
251 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
252 const struct lu_fid *fid)
254 struct osd_idmap_cache *idc;
257 idc = osd_idc_find(env, osd, fid);
258 LASSERT(!IS_ERR(idc));
262 /* new mapping is needed */
263 idc = osd_idc_add(env, osd, fid);
268 rc = osd_remote_fid(env, osd, fid);
269 if (unlikely(rc < 0))
273 /* the object is local, lookup in OI */
274 /* XXX: probably cheaper to lookup in LU first? */
275 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
277 if (unlikely(rc < 0)) {
278 CERROR("can't lookup: rc = %d\n", rc);
282 /* the object is remote */
290 * lookup mapping for given FID and fill it from the given object.
291 * the object is lolcal by definition.
293 static int osd_idc_find_and_init(const struct lu_env *env,
294 struct osd_device *osd,
295 struct osd_object *obj)
297 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
298 struct osd_idmap_cache *idc;
300 idc = osd_idc_find(env, osd, fid);
301 LASSERT(!IS_ERR(idc));
303 if (obj->oo_inode == NULL)
305 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
306 LASSERT(idc->oic_lid.oii_ino == 0);
307 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
308 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
313 /* new mapping is needed */
314 idc = osd_idc_add(env, osd, fid);
318 if (obj->oo_inode != NULL) {
319 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
320 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
326 * OSD object methods.
330 * Concurrency: no concurrent access is possible that early in object
333 static struct lu_object *osd_object_alloc(const struct lu_env *env,
334 const struct lu_object_header *hdr,
337 struct osd_object *mo;
343 l = &mo->oo_dt.do_lu;
344 dt_object_init(&mo->oo_dt, NULL, d);
345 mo->oo_dt.do_ops = &osd_obj_ea_ops;
346 l->lo_ops = &osd_lu_obj_ops;
347 init_rwsem(&mo->oo_sem);
348 init_rwsem(&mo->oo_ext_idx_sem);
349 spin_lock_init(&mo->oo_guard);
356 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
357 struct dentry *dentry, struct lustre_mdt_attrs *lma)
361 CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
362 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
363 info->oti_mdt_attrs_old, LMA_OLD_SIZE);
365 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
366 memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
368 lustre_lma_swab(lma);
369 /* Check LMA compatibility */
370 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
371 CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
372 "for fid = "DFID", ino = %lu\n",
373 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
374 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
375 PFID(&lma->lma_self_fid), inode->i_ino);
378 } else if (rc == 0) {
386 * retrieve object from backend ext fs.
388 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
389 struct osd_inode_id *id)
391 struct inode *inode = NULL;
393 /* if we look for an inode withing a running
394 * transaction, then we risk to deadlock */
395 /* osd_dirent_check_repair() breaks this */
396 /*LASSERT(current->journal_info == NULL);*/
398 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
400 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
401 id->oii_ino, PTR_ERR(inode));
402 } else if (id->oii_gen != OSD_OII_NOGEN &&
403 inode->i_generation != id->oii_gen) {
404 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
405 "i_generation = %u\n",
406 id->oii_ino, id->oii_gen, inode->i_generation);
408 inode = ERR_PTR(-ESTALE);
409 } else if (inode->i_nlink == 0) {
410 /* due to parallel readdir and unlink,
411 * we can have dead inode here. */
412 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
414 inode = ERR_PTR(-ESTALE);
415 } else if (is_bad_inode(inode)) {
416 CWARN("%.16s: bad inode: ino = %u\n",
417 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
419 inode = ERR_PTR(-ENOENT);
421 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
422 if (id->oii_gen == OSD_OII_NOGEN)
423 osd_id_gen(id, inode->i_ino, inode->i_generation);
425 /* Do not update file c/mtime in ldiskfs.
426 * NB: we don't have any lock to protect this because we don't
427 * have reference on osd_object now, but contention with
428 * another lookup + attr_set can't happen in the tiny window
429 * between if (...) and set S_NOCMTIME. */
430 if (!(inode->i_flags & S_NOCMTIME))
431 inode->i_flags |= S_NOCMTIME;
436 int osd_ldiskfs_add_entry(struct osd_thread_info *info,
437 handle_t *handle, struct dentry *child,
438 struct inode *inode, struct htree_lock *hlock)
442 rc = __ldiskfs_add_entry(handle, child, inode, hlock);
443 if (rc == -ENOBUFS || rc == -ENOSPC) {
444 char fidbuf[FID_LEN + 1];
445 struct lustre_mdt_attrs lma;
446 struct lu_fid fid = { };
448 struct dentry *p_dentry = child->d_parent;
450 rc2 = osd_get_lma(info, p_dentry->d_inode, p_dentry,
453 fid = lma.lma_self_fid;
454 snprintf(fidbuf, sizeof(fidbuf), DFID, PFID(&fid));
455 } else if (rc2 == -ENODATA) {
456 if (unlikely(p_dentry->d_inode ==
457 inode->i_sb->s_root->d_inode))
458 lu_local_obj_fid(&fid, OSD_FS_ROOT_OID);
459 else if (info->oti_dev && !info->oti_dev->od_is_ost &&
460 fid_seq_is_mdt0(fid_seq(&fid)))
461 lu_igif_build(&fid, p_dentry->d_inode->i_ino,
462 p_dentry->d_inode->i_generation);
463 snprintf(fidbuf, sizeof(fidbuf), DFID, PFID(&fid));
465 snprintf(fidbuf, FID_LEN, "%s", "unknown");
469 errstr = "has reached";
471 errstr = "is approaching";
472 CWARN("%.16s: directory (inode: %lu FID: %s) %s maximum entry limit\n",
473 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
474 p_dentry->d_inode->i_ino, fidbuf, errstr);
475 /* ignore such error now */
483 static struct inode *
484 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
485 struct osd_inode_id *id, struct lu_fid *fid)
487 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
491 inode = osd_iget(info, dev, id);
495 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
497 *fid = lma->lma_self_fid;
498 } else if (rc == -ENODATA) {
499 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
500 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
502 lu_igif_build(fid, inode->i_ino, inode->i_generation);
510 static struct inode *osd_iget_check(struct osd_thread_info *info,
511 struct osd_device *dev,
512 const struct lu_fid *fid,
513 struct osd_inode_id *id,
520 /* The cached OI mapping is trustable. If we cannot locate the inode
521 * via the cached OI mapping, then return the failure to the caller
522 * directly without further OI checking. */
524 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
527 if (cached || (rc != -ENOENT && rc != -ESTALE)) {
528 CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
537 if (is_bad_inode(inode)) {
540 CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
548 if (id->oii_gen != OSD_OII_NOGEN &&
549 inode->i_generation != id->oii_gen) {
552 CDEBUG(D_INODE, "unmatched inode: ino = %u, "
553 "oii_gen = %u, i_generation = %u\n",
554 id->oii_ino, id->oii_gen, inode->i_generation);
562 if (inode->i_nlink == 0) {
565 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
573 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
577 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
579 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
580 /* XXX: There are four possible cases:
582 * Backup/restore caused the OI invalid.
584 * Someone unlinked the object but NOT removed
585 * the OI mapping, such as mount target device
586 * as ldiskfs, and modify something directly.
588 * Someone just removed the object between the
589 * former oi_lookup and the iget. It is normal.
590 * 4. Other failure cases.
592 * Generally, when the device is mounted, it will
593 * auto check whether the system is restored from
594 * file-level backup or not. We trust such detect
595 * to distinguish the 1st case from the 2nd case. */
597 if (!IS_ERR(inode) && inode->i_generation != 0 &&
598 inode->i_generation == id->oii_gen)
599 /* "id->oii_gen != OSD_OII_NOGEN" is for
600 * "@cached == false" case. */
605 /* If the OI mapping was in OI file before the
606 * osd_iget_check(), but now, it is disappear,
607 * then it must be removed by race. That is a
608 * normal race case. */
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 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
770 const struct lu_fid *fid,
771 const struct lu_object_conf *conf)
773 struct osd_thread_info *info;
774 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
775 struct osd_device *dev;
776 struct osd_idmap_cache *oic;
777 struct osd_inode_id *id;
778 struct inode *inode = NULL;
779 struct osd_scrub *scrub;
780 struct scrub_file *sf;
781 __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT |
789 LINVRNT(osd_invariant(obj));
790 LASSERT(obj->oo_inode == NULL);
791 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
794 scrub = &dev->od_scrub;
795 sf = &scrub->os_file;
796 info = osd_oti_get(env);
798 oic = &info->oti_cache;
800 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
803 /* For the object is created as locking anchor, or for the object to
804 * be created on disk. No need to osd_oi_lookup() at here because FID
805 * shouldn't never be re-used, if it's really a duplicate FID from
806 * unexpected reason, we should be able to detect it later by calling
807 * do_create->osd_oi_insert(). */
808 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
809 GOTO(out, result = 0);
811 /* Search order: 1. per-thread cache. */
812 if (lu_fid_eq(fid, &oic->oic_fid) &&
813 likely(oic->oic_dev == dev)) {
819 if (!list_empty(&scrub->os_inconsistent_items)) {
820 /* Search order: 2. OI scrub pending list. */
821 result = osd_oii_lookup(dev, fid, id);
827 /* Search order: 3. OI files. */
828 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
829 if (result == -ENOENT) {
830 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
831 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
832 !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
834 GOTO(out, result = 0);
843 inode = osd_iget_check(info, dev, fid, id, cached);
845 result = PTR_ERR(inode);
846 if (result == -ENOENT || result == -ESTALE)
847 GOTO(out, result = -ENOENT);
849 if (result == -EREMCHG) {
852 /* We still have chance to get the valid inode: for the
853 * object which is referenced by remote name entry, the
854 * object on the local MDT will be linked under the dir
855 * of "/REMOTE_PARENT_DIR" with its FID string as name.
857 * We do not know whether the object for the given FID
858 * is referenced by some remote name entry or not, and
859 * especially for DNE II, a multiple-linked object may
860 * have many name entries reside on many MDTs.
862 * To simplify the operation, OSD will not distinguish
863 * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
864 * only happened for the RPC from other MDT during the
865 * OI scrub, or for the client side RPC with FID only,
866 * such as FID to path, or from old connected client. */
868 !fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
869 rc1 = osd_lookup_in_remote_parent(info, dev,
874 flags |= SS_AUTO_PARTIAL;
875 flags &= ~SS_AUTO_FULL;
880 if (thread_is_running(&scrub->os_thread)) {
881 if (scrub->os_partial_scan &&
882 !scrub->os_in_join) {
885 if (inode != NULL && !IS_ERR(inode)) {
888 osd_add_oi_cache(info, dev, id,
890 osd_oii_insert(dev, oic, true);
892 result = -EINPROGRESS;
895 } else if (!dev->od_noscrub) {
898 rc1 = osd_scrub_start(dev, flags);
899 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
900 "for the "DFID" with flags 0x%x,"
901 " rc = %d\n", osd_name(dev),
902 PFID(fid), flags, rc1);
903 if (rc1 == 0 || rc1 == -EALREADY) {
904 if (inode != NULL && !IS_ERR(inode)) {
907 osd_add_oi_cache(info, dev, id,
909 osd_oii_insert(dev, oic, true);
911 result = -EINPROGRESS;
921 if (inode == NULL || IS_ERR(inode))
927 obj->oo_inode = inode;
928 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
930 result = osd_check_lma(env, obj);
932 if (result == -ENODATA) {
934 result = osd_oi_lookup(info, dev, fid, id,
937 /* result == -ENOENT means that the OI
938 * mapping has been removed by race,
939 * the target inode belongs to other
942 * Others error also can be returned
945 obj->oo_inode = NULL;
948 /* result == 0 means the cached OI
949 * mapping is still in the OI file,
950 * the target the inode is valid. */
953 /* The current OI mapping is from the OI file,
954 * since the inode has been found via
955 * osd_iget_check(), no need recheck OI. */
963 obj->oo_inode = NULL;
964 if (result != -EREMCHG)
968 result = osd_oi_lookup(info, dev, fid, id,
970 /* result == -ENOENT means the cached OI mapping
971 * has been removed from the OI file by race,
972 * above target inode belongs to other object.
974 * Others error also can be returned directly. */
978 /* result == 0, goto trigger */
980 /* The current OI mapping is from the OI file,
981 * since the inode has been found via
982 * osd_iget_check(), no need recheck OI. */
989 obj->oo_compat_dot_created = 1;
990 obj->oo_compat_dotdot_created = 1;
992 if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
993 GOTO(out, result = 0);
995 LASSERT(obj->oo_hl_head == NULL);
996 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
997 if (obj->oo_hl_head == NULL) {
998 obj->oo_inode = NULL;
1000 GOTO(out, result = -ENOMEM);
1002 GOTO(out, result = 0);
1005 if (result != 0 && cached)
1006 fid_zero(&oic->oic_fid);
1008 LINVRNT(osd_invariant(obj));
1013 * Concurrency: shouldn't matter.
1015 static void osd_object_init0(struct osd_object *obj)
1017 LASSERT(obj->oo_inode != NULL);
1018 obj->oo_dt.do_body_ops = &osd_body_ops;
1019 obj->oo_dt.do_lu.lo_header->loh_attr |=
1020 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1024 * Concurrency: no concurrent access is possible that early in object
1027 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1028 const struct lu_object_conf *conf)
1030 struct osd_object *obj = osd_obj(l);
1033 LINVRNT(osd_invariant(obj));
1035 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1036 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1037 l->lo_header->loh_attr |= LOHA_EXISTS;
1041 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1042 obj->oo_dt.do_body_ops = &osd_body_ops_new;
1043 if (result == 0 && obj->oo_inode != NULL) {
1044 struct osd_thread_info *oti = osd_oti_get(env);
1045 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
1047 osd_object_init0(obj);
1048 result = osd_get_lma(oti, obj->oo_inode,
1049 &oti->oti_obj_dentry, lma);
1051 /* Convert LMAI flags to lustre LMA flags
1052 * and cache it to oo_lma_flags */
1054 lma_to_lustre_flags(lma->lma_incompat);
1055 } else if (result == -ENODATA) {
1060 LINVRNT(osd_invariant(obj));
1065 * Concurrency: no concurrent access is possible that late in object
1068 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1070 struct osd_object *obj = osd_obj(l);
1072 LINVRNT(osd_invariant(obj));
1074 dt_object_fini(&obj->oo_dt);
1075 if (obj->oo_hl_head != NULL)
1076 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1081 * Concurrency: no concurrent access is possible that late in object
1084 static void osd_index_fini(struct osd_object *o)
1086 struct iam_container *bag;
1088 if (o->oo_dir != NULL) {
1089 bag = &o->oo_dir->od_container;
1090 if (o->oo_inode != NULL) {
1091 if (bag->ic_object == o->oo_inode)
1092 iam_container_fini(bag);
1094 OBD_FREE_PTR(o->oo_dir);
1100 * Concurrency: no concurrent access is possible that late in object
1101 * life-cycle (for all existing callers, that is. New callers have to provide
1102 * their own locking.)
1104 static int osd_inode_unlinked(const struct inode *inode)
1106 return inode->i_nlink == 0;
1110 OSD_TXN_OI_DELETE_CREDITS = 20,
1111 OSD_TXN_INODE_DELETE_CREDITS = 20
1118 #if OSD_THANDLE_STATS
1120 * Set time when the handle is allocated
1122 static void osd_th_alloced(struct osd_thandle *oth)
1124 oth->oth_alloced = cfs_time_current();
1128 * Set time when the handle started
1130 static void osd_th_started(struct osd_thandle *oth)
1132 oth->oth_started = cfs_time_current();
1136 * Helper function to convert time interval to microseconds packed in
1139 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
1143 cfs_duration_usec(cfs_time_sub(end, start), &val);
1144 return val.tv_sec * 1000000 + val.tv_usec;
1148 * Check whether the we deal with this handle for too long.
1150 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1151 cfs_time_t alloced, cfs_time_t started,
1154 cfs_time_t now = cfs_time_current();
1156 LASSERT(dev != NULL);
1158 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1159 interval_to_usec(alloced, started));
1160 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1161 interval_to_usec(started, closed));
1162 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1163 interval_to_usec(closed, now));
1165 if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
1166 CWARN("transaction handle %p was open for too long: "
1167 "now "CFS_TIME_T" ,"
1168 "alloced "CFS_TIME_T" ,"
1169 "started "CFS_TIME_T" ,"
1170 "closed "CFS_TIME_T"\n",
1171 oth, now, alloced, started, closed);
1172 libcfs_debug_dumpstack(NULL);
1176 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
1178 cfs_time_t __closed = cfs_time_current(); \
1179 cfs_time_t __alloced = oth->oth_alloced; \
1180 cfs_time_t __started = oth->oth_started; \
1183 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
1186 #else /* OSD_THANDLE_STATS */
1188 #define osd_th_alloced(h) do {} while(0)
1189 #define osd_th_started(h) do {} while(0)
1190 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
1192 #endif /* OSD_THANDLE_STATS */
1195 * Concurrency: doesn't access mutable data.
1197 static int osd_param_is_not_sane(const struct osd_device *dev,
1198 const struct thandle *th)
1200 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1202 return oh->ot_credits > osd_transaction_size(dev);
1206 * Concurrency: shouldn't matter.
1208 static void osd_trans_commit_cb(struct super_block *sb,
1209 struct ldiskfs_journal_cb_entry *jcb, int error)
1211 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1212 struct thandle *th = &oh->ot_super;
1213 struct lu_device *lud = &th->th_dev->dd_lu_dev;
1214 struct dt_txn_commit_cb *dcb, *tmp;
1216 LASSERT(oh->ot_handle == NULL);
1219 CERROR("transaction @0x%p commit error: %d\n", th, error);
1221 dt_txn_hook_commit(th);
1223 /* call per-transaction callbacks if any */
1224 list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1226 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1227 "commit callback entry: magic=%x name='%s'\n",
1228 dcb->dcb_magic, dcb->dcb_name);
1229 list_del_init(&dcb->dcb_linkage);
1230 dcb->dcb_func(NULL, th, dcb, error);
1233 lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1237 lu_context_exit(&th->th_ctx);
1238 lu_context_fini(&th->th_ctx);
1242 #ifndef HAVE_SB_START_WRITE
1243 # define sb_start_write(sb) do {} while (0)
1244 # define sb_end_write(sb) do {} while (0)
1247 static struct thandle *osd_trans_create(const struct lu_env *env,
1248 struct dt_device *d)
1250 struct osd_thread_info *oti = osd_oti_get(env);
1251 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1252 struct osd_thandle *oh;
1256 /* on pending IO in this thread should left from prev. request */
1257 LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1259 sb_start_write(osd_sb(osd_dt_dev(d)));
1261 OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
1263 oh->ot_quota_trans = &oti->oti_quota_trans;
1264 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1268 th->th_tags = LCT_TX_HANDLE;
1270 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1271 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1274 memset(oti->oti_declare_ops, 0,
1275 sizeof(oti->oti_declare_ops));
1276 memset(oti->oti_declare_ops_cred, 0,
1277 sizeof(oti->oti_declare_ops_cred));
1278 memset(oti->oti_declare_ops_used, 0,
1279 sizeof(oti->oti_declare_ops_used));
1281 sb_end_write(osd_sb(osd_dt_dev(d)));
1282 th = ERR_PTR(-ENOMEM);
1287 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1289 struct osd_thread_info *oti = osd_oti_get(env);
1290 struct osd_thandle *oh;
1292 oh = container_of0(th, struct osd_thandle, ot_super);
1293 LASSERT(oh != NULL);
1295 CWARN(" create: %u/%u/%u, destroy: %u/%u/%u\n",
1296 oti->oti_declare_ops[OSD_OT_CREATE],
1297 oti->oti_declare_ops_cred[OSD_OT_CREATE],
1298 oti->oti_declare_ops_used[OSD_OT_CREATE],
1299 oti->oti_declare_ops[OSD_OT_DESTROY],
1300 oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1301 oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1302 CWARN(" attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1303 oti->oti_declare_ops[OSD_OT_ATTR_SET],
1304 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1305 oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1306 oti->oti_declare_ops[OSD_OT_XATTR_SET],
1307 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1308 oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1309 CWARN(" write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1310 oti->oti_declare_ops[OSD_OT_WRITE],
1311 oti->oti_declare_ops_cred[OSD_OT_WRITE],
1312 oti->oti_declare_ops_used[OSD_OT_WRITE],
1313 oti->oti_declare_ops[OSD_OT_PUNCH],
1314 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1315 oti->oti_declare_ops_used[OSD_OT_PUNCH],
1316 oti->oti_declare_ops[OSD_OT_QUOTA],
1317 oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1318 oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1319 CWARN(" insert: %u/%u/%u, delete: %u/%u/%u\n",
1320 oti->oti_declare_ops[OSD_OT_INSERT],
1321 oti->oti_declare_ops_cred[OSD_OT_INSERT],
1322 oti->oti_declare_ops_used[OSD_OT_INSERT],
1323 oti->oti_declare_ops[OSD_OT_DELETE],
1324 oti->oti_declare_ops_cred[OSD_OT_DELETE],
1325 oti->oti_declare_ops_used[OSD_OT_DELETE]);
1326 CWARN(" ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1327 oti->oti_declare_ops[OSD_OT_REF_ADD],
1328 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1329 oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1330 oti->oti_declare_ops[OSD_OT_REF_DEL],
1331 oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1332 oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1336 * Concurrency: shouldn't matter.
1338 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1341 struct osd_thread_info *oti = osd_oti_get(env);
1342 struct osd_device *dev = osd_dt_dev(d);
1344 struct osd_thandle *oh;
1349 LASSERT(current->journal_info == NULL);
1351 oh = container_of0(th, struct osd_thandle, ot_super);
1352 LASSERT(oh != NULL);
1353 LASSERT(oh->ot_handle == NULL);
1355 rc = dt_txn_hook_start(env, d, th);
1359 if (unlikely(osd_param_is_not_sane(dev, th))) {
1360 static unsigned long last_printed;
1361 static int last_credits;
1363 /* don't make noise on a tiny testing systems
1364 * actual credits misuse will be caught anyway */
1365 if (last_credits != oh->ot_credits &&
1366 time_after(jiffies, last_printed +
1367 msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1368 osd_transaction_size(dev) > 512) {
1369 osd_trans_dump_creds(env, th);
1370 libcfs_debug_dumpstack(NULL);
1371 last_credits = oh->ot_credits;
1372 last_printed = jiffies;
1374 /* XXX Limit the credits to 'max_transaction_buffers', and
1375 * let the underlying filesystem to catch the error if
1376 * we really need so many credits.
1378 * This should be removed when we can calculate the
1379 * credits precisely. */
1380 oh->ot_credits = osd_transaction_size(dev);
1384 * XXX temporary stuff. Some abstraction layer should
1387 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1391 LASSERT(oti->oti_txns == 0);
1392 lu_context_init(&th->th_ctx, th->th_tags);
1393 lu_context_enter(&th->th_ctx);
1395 lu_device_get(&d->dd_lu_dev);
1396 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1407 static int osd_seq_exists(const struct lu_env *env,
1408 struct osd_device *osd, u64 seq)
1410 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1411 struct seq_server_site *ss = osd_seq_site(osd);
1415 LASSERT(ss != NULL);
1416 LASSERT(ss->ss_server_fld != NULL);
1418 rc = osd_fld_lookup(env, osd, seq, range);
1421 CERROR("%s: can't lookup FLD sequence "LPX64
1422 ": rc = %d\n", osd_name(osd), seq, rc);
1426 RETURN(ss->ss_node_id == range->lsr_index);
1429 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1431 struct dt_txn_commit_cb *dcb;
1432 struct dt_txn_commit_cb *tmp;
1434 /* call per-transaction stop callbacks if any */
1435 list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1437 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1438 "commit callback entry: magic=%x name='%s'\n",
1439 dcb->dcb_magic, dcb->dcb_name);
1440 list_del_init(&dcb->dcb_linkage);
1441 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1446 * Concurrency: shouldn't matter.
1448 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1451 int rc = 0, remove_agents = 0;
1452 struct osd_thandle *oh;
1453 struct osd_thread_info *oti = osd_oti_get(env);
1454 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1455 struct osd_device *osd = osd_dt_dev(th->th_dev);
1456 struct qsd_instance *qsd = osd->od_quota_slave;
1457 struct lquota_trans *qtrans;
1460 oh = container_of0(th, struct osd_thandle, ot_super);
1462 /* reset OI cache for safety */
1463 oti->oti_ins_cache_used = 0;
1465 remove_agents = oh->ot_remove_agents;
1467 qtrans = oh->ot_quota_trans;
1468 oh->ot_quota_trans = NULL;
1470 if (oh->ot_handle != NULL) {
1471 handle_t *hdl = oh->ot_handle;
1474 * add commit callback
1475 * notice we don't do this in osd_trans_start()
1476 * as underlying transaction can change during truncate
1478 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1481 LASSERT(oti->oti_txns == 1);
1484 rc = dt_txn_hook_stop(env, th);
1486 CERROR("%s: failed in transaction hook: rc = %d\n",
1489 osd_trans_stop_cb(oh, rc);
1490 /* hook functions might modify th_sync */
1491 hdl->h_sync = th->th_sync;
1493 oh->ot_handle = NULL;
1494 OSD_CHECK_SLOW_TH(oh, osd, rc = ldiskfs_journal_stop(hdl));
1496 CERROR("%s: failed to stop transaction: rc = %d\n",
1499 osd_trans_stop_cb(oh, th->th_result);
1503 /* inform the quota slave device that the transaction is stopping */
1504 qsd_op_end(env, qsd, qtrans);
1506 /* as we want IO to journal and data IO be concurrent, we don't block
1507 * awaiting data IO completion in osd_do_bio(), instead we wait here
1508 * once transaction is submitted to the journal. all reqular requests
1509 * don't do direct IO (except read/write), thus this wait_event becomes
1512 * IMPORTANT: we have to wait till any IO submited by the thread is
1513 * completed otherwise iobuf may be corrupted by different request
1515 wait_event(iobuf->dr_wait,
1516 atomic_read(&iobuf->dr_numreqs) == 0);
1517 osd_fini_iobuf(osd, iobuf);
1519 rc = iobuf->dr_error;
1521 if (unlikely(remove_agents != 0))
1522 osd_process_scheduled_agent_removals(env, osd);
1524 sb_end_write(osd_sb(osd));
1529 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1531 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1534 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1535 LASSERT(&dcb->dcb_func != NULL);
1536 if (dcb->dcb_flags & DCB_TRANS_STOP)
1537 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1539 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1545 * Called just before object is freed. Releases all resources except for
1546 * object itself (that is released by osd_object_free()).
1548 * Concurrency: no concurrent access is possible that late in object
1551 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1553 struct osd_object *obj = osd_obj(l);
1554 struct inode *inode = obj->oo_inode;
1556 LINVRNT(osd_invariant(obj));
1559 * If object is unlinked remove fid->ino mapping from object index.
1562 osd_index_fini(obj);
1563 if (inode != NULL) {
1564 struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave;
1565 qid_t uid = i_uid_read(inode);
1566 qid_t gid = i_gid_read(inode);
1569 obj->oo_inode = NULL;
1572 struct osd_thread_info *info = osd_oti_get(env);
1573 struct lquota_id_info *qi = &info->oti_qi;
1575 /* Release granted quota to master if necessary */
1576 qi->lqi_id.qid_uid = uid;
1577 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1579 qi->lqi_id.qid_uid = gid;
1580 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1586 * Concurrency: ->loo_object_release() is called under site spin-lock.
1588 static void osd_object_release(const struct lu_env *env,
1589 struct lu_object *l)
1591 struct osd_object *o = osd_obj(l);
1592 /* nobody should be releasing a non-destroyed object with nlink=0
1593 * the API allows this, but ldiskfs doesn't like and then report
1594 * this inode as deleted */
1595 if (unlikely(!o->oo_destroyed && o->oo_inode && o->oo_inode->i_nlink == 0))
1600 * Concurrency: shouldn't matter.
1602 static int osd_object_print(const struct lu_env *env, void *cookie,
1603 lu_printer_t p, const struct lu_object *l)
1605 struct osd_object *o = osd_obj(l);
1606 struct iam_descr *d;
1608 if (o->oo_dir != NULL)
1609 d = o->oo_dir->od_container.ic_descr;
1612 return (*p)(env, cookie,
1613 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1615 o->oo_inode ? o->oo_inode->i_ino : 0UL,
1616 o->oo_inode ? o->oo_inode->i_generation : 0,
1617 d ? d->id_ops->id_name : "plain");
1621 * Concurrency: shouldn't matter.
1623 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1624 struct obd_statfs *sfs)
1626 struct osd_device *osd = osd_dt_dev(d);
1627 struct super_block *sb = osd_sb(osd);
1628 struct kstatfs *ksfs;
1632 if (unlikely(osd->od_mnt == NULL))
1633 return -EINPROGRESS;
1635 /* osd_lproc.c call this without env, allocate ksfs for that case */
1636 if (unlikely(env == NULL)) {
1637 OBD_ALLOC_PTR(ksfs);
1641 ksfs = &osd_oti_get(env)->oti_ksfs;
1644 result = sb->s_op->statfs(sb->s_root, ksfs);
1648 statfs_pack(sfs, ksfs);
1649 if (unlikely(sb->s_flags & MS_RDONLY))
1650 sfs->os_state = OS_STATE_READONLY;
1651 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb,
1652 LDISKFS_FEATURE_INCOMPAT_EXTENTS))
1653 sfs->os_maxbytes = sb->s_maxbytes;
1655 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
1658 * Reserve some space so to avoid fragmenting the filesystem too much.
1659 * Fragmentation not only impacts performance, but can also increase
1660 * metadata overhead significantly, causing grant calculation to be
1663 * Reserve 0.78% of total space, at least 8MB for small filesystems.
1665 CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
1666 reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
1667 if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
1668 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
1670 sfs->os_blocks -= reserved;
1671 sfs->os_bfree -= min(reserved, sfs->os_bfree);
1672 sfs->os_bavail -= min(reserved, sfs->os_bavail);
1675 if (unlikely(env == NULL))
1681 * Estimate space needed for file creations. We assume the largest filename
1682 * which is 2^64 - 1, hence a filename of 20 chars.
1683 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1685 #ifdef __LDISKFS_DIR_REC_LEN
1686 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1688 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1692 * Concurrency: doesn't access mutable data.
1694 static void osd_conf_get(const struct lu_env *env,
1695 const struct dt_device *dev,
1696 struct dt_device_param *param)
1698 struct super_block *sb = osd_sb(osd_dt_dev(dev));
1702 * XXX should be taken from not-yet-existing fs abstraction layer.
1704 param->ddp_max_name_len = LDISKFS_NAME_LEN;
1705 param->ddp_max_nlink = LDISKFS_LINK_MAX;
1706 param->ddp_symlink_max = sb->s_blocksize;
1707 param->ddp_mount_type = LDD_MT_LDISKFS;
1708 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS))
1709 param->ddp_maxbytes = sb->s_maxbytes;
1711 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
1712 /* inode are statically allocated, so per-inode space consumption
1713 * is the space consumed by the directory entry */
1714 param->ddp_inodespace = PER_OBJ_USAGE;
1715 /* EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
1716 * = 128MB) which is unlikely to be hit in real life. Report a smaller
1717 * maximum length to not under count the actual number of extents
1718 * needed for writing a file. */
1719 param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2;
1720 /* worst-case extent insertion metadata overhead */
1721 param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
1722 param->ddp_mntopts = 0;
1723 if (test_opt(sb, XATTR_USER))
1724 param->ddp_mntopts |= MNTOPT_USERXATTR;
1725 if (test_opt(sb, POSIX_ACL))
1726 param->ddp_mntopts |= MNTOPT_ACL;
1728 /* LOD might calculate the max stripe count based on max_ea_size,
1729 * so we need take account in the overhead as well,
1730 * xattr_header + magic + xattr_entry_head */
1731 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1732 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1734 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1735 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1736 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1740 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1744 * Concurrency: shouldn't matter.
1746 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1750 CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1752 rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1754 CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
1755 LUSTRE_OSD_LDISKFS_NAME, rc);
1761 * Start commit for OSD device.
1763 * An implementation of dt_commit_async method for OSD device.
1764 * Asychronously starts underlayng fs sync and thereby a transaction
1767 * \param env environment
1768 * \param d dt device
1770 * \see dt_device_operations
1772 static int osd_commit_async(const struct lu_env *env,
1773 struct dt_device *d)
1775 struct super_block *s = osd_sb(osd_dt_dev(d));
1778 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1779 RETURN(s->s_op->sync_fs(s, 0));
1783 * Concurrency: shouldn't matter.
1786 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1788 struct super_block *sb = osd_sb(osd_dt_dev(d));
1789 struct block_device *dev = sb->s_bdev;
1790 #ifdef HAVE_DEV_SET_RDONLY
1791 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1794 int rc = -EOPNOTSUPP;
1798 #ifdef HAVE_DEV_SET_RDONLY
1799 CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1801 if (sb->s_op->freeze_fs) {
1802 rc = sb->s_op->freeze_fs(sb);
1807 if (jdev && (jdev != dev)) {
1808 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1810 dev_set_rdonly(jdev);
1812 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1813 dev_set_rdonly(dev);
1815 if (sb->s_op->unfreeze_fs)
1816 sb->s_op->unfreeze_fs(sb);
1821 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1822 osd_dt_dev(d)->od_svname, (long)dev, rc);
1828 * Note: we do not count into QUOTA here.
1829 * If we mount with --data_journal we may need more.
1831 const int osd_dto_credits_noquota[DTO_NR] = {
1834 * INDEX_EXTRA_TRANS_BLOCKS(8) +
1835 * SINGLEDATA_TRANS_BLOCKS(8)
1836 * XXX Note: maybe iam need more, since iam have more level than
1839 [DTO_INDEX_INSERT] = 16,
1842 * just modify a single entry, probably merge few within a block
1844 [DTO_INDEX_DELETE] = 1,
1848 [DTO_INDEX_UPDATE] = 16,
1850 * 4(inode, inode bits, groups, GDT)
1851 * notice: OI updates are counted separately with DTO_INDEX_INSERT
1853 [DTO_OBJECT_CREATE] = 4,
1855 * 4(inode, inode bits, groups, GDT)
1856 * notice: OI updates are counted separately with DTO_INDEX_DELETE
1858 [DTO_OBJECT_DELETE] = 4,
1860 * Attr set credits (inode)
1862 [DTO_ATTR_SET_BASE] = 1,
1864 * Xattr set. The same as xattr of EXT3.
1865 * DATA_TRANS_BLOCKS(14)
1866 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1867 * are also counted in. Do not know why?
1869 [DTO_XATTR_SET] = 14,
1871 * credits for inode change during write.
1873 [DTO_WRITE_BASE] = 3,
1875 * credits for single block write.
1877 [DTO_WRITE_BLOCK] = 14,
1879 * Attr set credits for chown.
1880 * This is extra credits for setattr, and it is null without quota
1882 [DTO_ATTR_SET_CHOWN] = 0
1885 static const struct dt_device_operations osd_dt_ops = {
1886 .dt_root_get = osd_root_get,
1887 .dt_statfs = osd_statfs,
1888 .dt_trans_create = osd_trans_create,
1889 .dt_trans_start = osd_trans_start,
1890 .dt_trans_stop = osd_trans_stop,
1891 .dt_trans_cb_add = osd_trans_cb_add,
1892 .dt_conf_get = osd_conf_get,
1893 .dt_sync = osd_sync,
1895 .dt_commit_async = osd_commit_async,
1898 static void osd_object_read_lock(const struct lu_env *env,
1899 struct dt_object *dt, unsigned role)
1901 struct osd_object *obj = osd_dt_obj(dt);
1902 struct osd_thread_info *oti = osd_oti_get(env);
1904 LINVRNT(osd_invariant(obj));
1906 LASSERT(obj->oo_owner != env);
1907 down_read_nested(&obj->oo_sem, role);
1909 LASSERT(obj->oo_owner == NULL);
1913 static void osd_object_write_lock(const struct lu_env *env,
1914 struct dt_object *dt, unsigned role)
1916 struct osd_object *obj = osd_dt_obj(dt);
1917 struct osd_thread_info *oti = osd_oti_get(env);
1919 LINVRNT(osd_invariant(obj));
1921 LASSERT(obj->oo_owner != env);
1922 down_write_nested(&obj->oo_sem, role);
1924 LASSERT(obj->oo_owner == NULL);
1925 obj->oo_owner = env;
1929 static void osd_object_read_unlock(const struct lu_env *env,
1930 struct dt_object *dt)
1932 struct osd_object *obj = osd_dt_obj(dt);
1933 struct osd_thread_info *oti = osd_oti_get(env);
1935 LINVRNT(osd_invariant(obj));
1937 LASSERT(oti->oti_r_locks > 0);
1939 up_read(&obj->oo_sem);
1942 static void osd_object_write_unlock(const struct lu_env *env,
1943 struct dt_object *dt)
1945 struct osd_object *obj = osd_dt_obj(dt);
1946 struct osd_thread_info *oti = osd_oti_get(env);
1948 LINVRNT(osd_invariant(obj));
1950 LASSERT(obj->oo_owner == env);
1951 LASSERT(oti->oti_w_locks > 0);
1953 obj->oo_owner = NULL;
1954 up_write(&obj->oo_sem);
1957 static int osd_object_write_locked(const struct lu_env *env,
1958 struct dt_object *dt)
1960 struct osd_object *obj = osd_dt_obj(dt);
1962 LINVRNT(osd_invariant(obj));
1964 return obj->oo_owner == env;
1967 static struct timespec *osd_inode_time(const struct lu_env *env,
1968 struct inode *inode, __u64 seconds)
1970 struct osd_thread_info *oti = osd_oti_get(env);
1971 struct timespec *t = &oti->oti_time;
1973 t->tv_sec = seconds;
1975 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1979 static void osd_inode_getattr(const struct lu_env *env,
1980 struct inode *inode, struct lu_attr *attr)
1982 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1983 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1984 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1987 attr->la_atime = LTIME_S(inode->i_atime);
1988 attr->la_mtime = LTIME_S(inode->i_mtime);
1989 attr->la_ctime = LTIME_S(inode->i_ctime);
1990 attr->la_mode = inode->i_mode;
1991 attr->la_size = i_size_read(inode);
1992 attr->la_blocks = inode->i_blocks;
1993 attr->la_uid = i_uid_read(inode);
1994 attr->la_gid = i_gid_read(inode);
1995 attr->la_flags = ll_inode_to_ext_flags(inode->i_flags);
1996 attr->la_nlink = inode->i_nlink;
1997 attr->la_rdev = inode->i_rdev;
1998 attr->la_blksize = 1 << inode->i_blkbits;
1999 attr->la_blkbits = inode->i_blkbits;
2002 static int osd_attr_get(const struct lu_env *env,
2003 struct dt_object *dt,
2004 struct lu_attr *attr)
2006 struct osd_object *obj = osd_dt_obj(dt);
2008 if (unlikely(!dt_object_exists(dt)))
2010 if (unlikely(obj->oo_destroyed))
2013 LASSERT(!dt_object_remote(dt));
2014 LINVRNT(osd_invariant(obj));
2016 spin_lock(&obj->oo_guard);
2017 osd_inode_getattr(env, obj->oo_inode, attr);
2018 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2019 attr->la_flags |= LUSTRE_ORPHAN_FL;
2020 spin_unlock(&obj->oo_guard);
2025 static int osd_declare_attr_set(const struct lu_env *env,
2026 struct dt_object *dt,
2027 const struct lu_attr *attr,
2028 struct thandle *handle)
2030 struct osd_thandle *oh;
2031 struct osd_object *obj;
2032 struct osd_thread_info *info = osd_oti_get(env);
2033 struct lquota_id_info *qi = &info->oti_qi;
2041 LASSERT(dt != NULL);
2042 LASSERT(handle != NULL);
2044 obj = osd_dt_obj(dt);
2045 LASSERT(osd_invariant(obj));
2047 oh = container_of0(handle, struct osd_thandle, ot_super);
2048 LASSERT(oh->ot_handle == NULL);
2050 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2051 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2053 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2054 osd_dto_credits_noquota[DTO_XATTR_SET]);
2056 if (attr == NULL || obj->oo_inode == NULL)
2059 bspace = obj->oo_inode->i_blocks;
2060 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
2061 bspace = toqb(bspace);
2063 /* Changing ownership is always preformed by super user, it should not
2066 * We still need to call the osd_declare_qid() to calculate the journal
2067 * credits for updating quota accounting files and to trigger quota
2068 * space adjustment once the operation is completed.*/
2069 if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2071 uid = i_uid_read(obj->oo_inode);
2072 qi->lqi_type = USRQUOTA;
2073 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2074 /* inode accounting */
2075 qi->lqi_is_blk = false;
2077 /* one more inode for the new uid ... */
2078 qi->lqi_id.qid_uid = attr->la_uid;
2080 /* Reserve credits for the new uid */
2081 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2082 if (rc == -EDQUOT || rc == -EINPROGRESS)
2087 /* and one less inode for the current uid */
2088 qi->lqi_id.qid_uid = uid;
2090 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2091 if (rc == -EDQUOT || rc == -EINPROGRESS)
2096 /* block accounting */
2097 qi->lqi_is_blk = true;
2099 /* more blocks for the new uid ... */
2100 qi->lqi_id.qid_uid = attr->la_uid;
2101 qi->lqi_space = bspace;
2103 * Credits for the new uid has been reserved, re-use "obj"
2104 * to save credit reservation.
2106 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2107 if (rc == -EDQUOT || rc == -EINPROGRESS)
2112 /* and finally less blocks for the current uid */
2113 qi->lqi_id.qid_uid = uid;
2114 qi->lqi_space = -bspace;
2115 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2116 if (rc == -EDQUOT || rc == -EINPROGRESS)
2122 gid = i_gid_read(obj->oo_inode);
2123 qi->lqi_type = GRPQUOTA;
2124 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2126 /* inode accounting */
2127 qi->lqi_is_blk = false;
2129 /* one more inode for the new gid ... */
2130 qi->lqi_id.qid_gid = attr->la_gid;
2132 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2133 if (rc == -EDQUOT || rc == -EINPROGRESS)
2138 /* and one less inode for the current gid */
2139 qi->lqi_id.qid_gid = gid;
2141 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2142 if (rc == -EDQUOT || rc == -EINPROGRESS)
2147 /* block accounting */
2148 qi->lqi_is_blk = true;
2150 /* more blocks for the new gid ... */
2151 qi->lqi_id.qid_gid = attr->la_gid;
2152 qi->lqi_space = bspace;
2153 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2154 if (rc == -EDQUOT || rc == -EINPROGRESS)
2159 /* and finally less blocks for the current gid */
2160 qi->lqi_id.qid_gid = gid;
2161 qi->lqi_space = -bspace;
2162 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2163 if (rc == -EDQUOT || rc == -EINPROGRESS)
2172 static int osd_inode_setattr(const struct lu_env *env,
2173 struct inode *inode, const struct lu_attr *attr)
2175 __u64 bits = attr->la_valid;
2177 /* Only allow set size for regular file */
2178 if (!S_ISREG(inode->i_mode))
2179 bits &= ~(LA_SIZE | LA_BLOCKS);
2184 if (bits & LA_ATIME)
2185 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
2186 if (bits & LA_CTIME)
2187 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
2188 if (bits & LA_MTIME)
2189 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
2190 if (bits & LA_SIZE) {
2191 LDISKFS_I(inode)->i_disksize = attr->la_size;
2192 i_size_write(inode, attr->la_size);
2196 /* OSD should not change "i_blocks" which is used by quota.
2197 * "i_blocks" should be changed by ldiskfs only. */
2198 if (bits & LA_BLOCKS)
2199 inode->i_blocks = attr->la_blocks;
2202 inode->i_mode = (inode->i_mode & S_IFMT) |
2203 (attr->la_mode & ~S_IFMT);
2205 i_uid_write(inode, attr->la_uid);
2207 i_gid_write(inode, attr->la_gid);
2208 if (bits & LA_NLINK)
2209 set_nlink(inode, attr->la_nlink);
2211 inode->i_rdev = attr->la_rdev;
2213 if (bits & LA_FLAGS) {
2214 /* always keep S_NOCMTIME */
2215 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2221 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2223 if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2224 (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2228 ll_vfs_dq_init(inode);
2230 if (attr->la_valid & LA_UID)
2231 iattr.ia_valid |= ATTR_UID;
2232 if (attr->la_valid & LA_GID)
2233 iattr.ia_valid |= ATTR_GID;
2234 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2235 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2237 rc = ll_vfs_dq_transfer(inode, &iattr);
2239 CERROR("%s: quota transfer failed: rc = %d. Is quota "
2240 "enforcement enabled on the ldiskfs "
2241 "filesystem?\n", inode->i_sb->s_id, rc);
2248 static int osd_attr_set(const struct lu_env *env,
2249 struct dt_object *dt,
2250 const struct lu_attr *attr,
2251 struct thandle *handle)
2253 struct osd_object *obj = osd_dt_obj(dt);
2254 struct inode *inode;
2257 if (!dt_object_exists(dt))
2260 LASSERT(handle != NULL);
2261 LASSERT(!dt_object_remote(dt));
2262 LASSERT(osd_invariant(obj));
2264 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2266 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
2267 struct osd_thread_info *oti = osd_oti_get(env);
2268 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
2269 struct lu_fid *fid1 = &oti->oti_fid;
2270 struct osd_inode_id *id = &oti->oti_id;
2271 struct iam_path_descr *ipd;
2272 struct iam_container *bag;
2273 struct osd_thandle *oh;
2276 fid_cpu_to_be(fid1, fid0);
2277 memset(id, 1, sizeof(*id));
2278 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2279 fid0)->oi_dir.od_container;
2280 ipd = osd_idx_ipd_get(env, bag);
2281 if (unlikely(ipd == NULL))
2284 oh = container_of0(handle, struct osd_thandle, ot_super);
2285 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2286 (const struct iam_rec *)id, ipd);
2287 osd_ipd_put(env, bag, ipd);
2288 return(rc > 0 ? 0 : rc);
2291 inode = obj->oo_inode;
2293 rc = osd_quota_transfer(inode, attr);
2297 spin_lock(&obj->oo_guard);
2298 rc = osd_inode_setattr(env, inode, attr);
2299 spin_unlock(&obj->oo_guard);
2303 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2305 if (!(attr->la_valid & LA_FLAGS))
2308 /* Let's check if there are extra flags need to be set into LMA */
2309 if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
2310 struct osd_thread_info *info = osd_oti_get(env);
2311 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2313 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
2317 lma->lma_incompat |=
2318 lustre_to_lma_flags(attr->la_flags);
2319 lustre_lma_swab(lma);
2320 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
2321 lma, sizeof(*lma), XATTR_REPLACE);
2323 struct osd_device *osd = osd_obj2dev(obj);
2325 CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
2326 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
2327 lma->lma_incompat, rc);
2330 attr->la_flags & LUSTRE_LMA_FL_MASKS;
2332 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2335 osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2340 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2341 struct osd_object *obj,
2342 const char *name, const int namelen)
2344 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2347 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2348 umode_t mode, struct dt_allocation_hint *hint,
2352 struct osd_device *osd = osd_obj2dev(obj);
2353 struct osd_thandle *oth;
2354 struct dt_object *parent = NULL;
2355 struct inode *inode;
2357 LINVRNT(osd_invariant(obj));
2358 LASSERT(obj->oo_inode == NULL);
2359 LASSERT(obj->oo_hl_head == NULL);
2361 if (S_ISDIR(mode) && ldiskfs_pdo) {
2362 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2363 if (obj->oo_hl_head == NULL)
2367 oth = container_of(th, struct osd_thandle, ot_super);
2368 LASSERT(oth->ot_handle->h_transaction != NULL);
2370 if (hint != NULL && hint->dah_parent != NULL &&
2371 !dt_object_remote(hint->dah_parent))
2372 parent = hint->dah_parent;
2374 inode = ldiskfs_create_inode(oth->ot_handle,
2375 parent ? osd_dt_obj(parent)->oo_inode :
2376 osd_sb(osd)->s_root->d_inode,
2378 if (!IS_ERR(inode)) {
2379 /* Do not update file c/mtime in ldiskfs. */
2380 inode->i_flags |= S_NOCMTIME;
2382 /* For new created object, it must be consistent,
2383 * and it is unnecessary to scrub against it. */
2384 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2386 obj->oo_inode = inode;
2389 if (obj->oo_hl_head != NULL) {
2390 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2391 obj->oo_hl_head = NULL;
2393 result = PTR_ERR(inode);
2395 LINVRNT(osd_invariant(obj));
2403 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2404 struct lu_attr *attr,
2405 struct dt_allocation_hint *hint,
2406 struct dt_object_format *dof,
2410 struct osd_thandle *oth;
2411 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2413 LASSERT(S_ISDIR(attr->la_mode));
2415 oth = container_of(th, struct osd_thandle, ot_super);
2416 LASSERT(oth->ot_handle->h_transaction != NULL);
2417 result = osd_mkfile(info, obj, mode, hint, th);
2422 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2423 struct lu_attr *attr,
2424 struct dt_allocation_hint *hint,
2425 struct dt_object_format *dof,
2429 struct osd_thandle *oth;
2430 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2432 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2434 LASSERT(S_ISREG(attr->la_mode));
2436 oth = container_of(th, struct osd_thandle, ot_super);
2437 LASSERT(oth->ot_handle->h_transaction != NULL);
2439 result = osd_mkfile(info, obj, mode, hint, th);
2441 LASSERT(obj->oo_inode != NULL);
2442 if (feat->dif_flags & DT_IND_VARKEY)
2443 result = iam_lvar_create(obj->oo_inode,
2444 feat->dif_keysize_max,
2446 feat->dif_recsize_max,
2449 result = iam_lfix_create(obj->oo_inode,
2450 feat->dif_keysize_max,
2452 feat->dif_recsize_max,
2459 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2460 struct lu_attr *attr,
2461 struct dt_allocation_hint *hint,
2462 struct dt_object_format *dof,
2465 LASSERT(S_ISREG(attr->la_mode));
2466 return osd_mkfile(info, obj, (attr->la_mode &
2467 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2470 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2471 struct lu_attr *attr,
2472 struct dt_allocation_hint *hint,
2473 struct dt_object_format *dof,
2476 LASSERT(S_ISLNK(attr->la_mode));
2477 return osd_mkfile(info, obj, (attr->la_mode &
2478 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2481 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2482 struct lu_attr *attr,
2483 struct dt_allocation_hint *hint,
2484 struct dt_object_format *dof,
2487 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2490 LINVRNT(osd_invariant(obj));
2491 LASSERT(obj->oo_inode == NULL);
2492 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2493 S_ISFIFO(mode) || S_ISSOCK(mode));
2495 result = osd_mkfile(info, obj, mode, hint, th);
2497 LASSERT(obj->oo_inode != NULL);
2499 * This inode should be marked dirty for i_rdev. Currently
2500 * that is done in the osd_attr_init().
2502 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2505 LINVRNT(osd_invariant(obj));
2509 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2511 struct dt_allocation_hint *hint,
2512 struct dt_object_format *dof,
2515 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2517 osd_obj_type_f result;
2533 result = osd_mk_index;
2544 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2545 struct dt_object *parent, struct dt_object *child,
2550 ah->dah_parent = parent;
2551 ah->dah_mode = child_mode;
2553 if (parent != NULL && !dt_object_remote(parent)) {
2554 /* will help to find FID->ino at dt_insert("..") */
2555 struct osd_object *pobj = osd_dt_obj(parent);
2556 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
2560 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2561 struct lu_attr *attr, struct dt_object_format *dof)
2563 struct inode *inode = obj->oo_inode;
2564 __u64 valid = attr->la_valid;
2567 attr->la_valid &= ~(LA_TYPE | LA_MODE);
2569 if (dof->dof_type != DFT_NODE)
2570 attr->la_valid &= ~LA_RDEV;
2571 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2572 attr->la_valid &= ~LA_ATIME;
2573 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2574 attr->la_valid &= ~LA_CTIME;
2575 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2576 attr->la_valid &= ~LA_MTIME;
2578 result = osd_quota_transfer(inode, attr);
2582 if (attr->la_valid != 0) {
2583 result = osd_inode_setattr(info->oti_env, inode, attr);
2585 * The osd_inode_setattr() should always succeed here. The
2586 * only error that could be returned is EDQUOT when we are
2587 * trying to change the UID or GID of the inode. However, this
2588 * should not happen since quota enforcement is no longer
2589 * enabled on ldiskfs (lquota takes care of it).
2591 LASSERTF(result == 0, "%d\n", result);
2592 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2595 attr->la_valid = valid;
2599 * Helper function for osd_object_create()
2601 * \retval 0, on success
2603 static int __osd_object_create(struct osd_thread_info *info,
2604 struct osd_object *obj, struct lu_attr *attr,
2605 struct dt_allocation_hint *hint,
2606 struct dt_object_format *dof,
2612 osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
2614 /* we drop umask so that permissions we pass are not affected */
2615 umask = current->fs->umask;
2616 current->fs->umask = 0;
2618 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2620 if (likely(obj->oo_inode != NULL)) {
2621 LASSERT(obj->oo_inode->i_state & I_NEW);
2623 /* Unlock the inode before attr initialization to avoid
2624 * unnecessary dqget operations. LU-6378 */
2625 unlock_new_inode(obj->oo_inode);
2628 if (likely(result == 0)) {
2629 osd_attr_init(info, obj, attr, dof);
2630 osd_object_init0(obj);
2633 /* restore previous umask value */
2634 current->fs->umask = umask;
2636 osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
2642 * Helper function for osd_object_create()
2644 * \retval 0, on success
2646 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2647 const struct lu_fid *fid, struct thandle *th)
2649 struct osd_thread_info *info = osd_oti_get(env);
2650 struct osd_inode_id *id = &info->oti_id;
2651 struct osd_device *osd = osd_obj2dev(obj);
2652 struct osd_thandle *oh;
2655 LASSERT(obj->oo_inode != NULL);
2657 oh = container_of0(th, struct osd_thandle, ot_super);
2658 LASSERT(oh->ot_handle);
2659 osd_trans_exec_op(env, th, OSD_OT_INSERT);
2661 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2662 rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
2663 OI_CHECK_FLD, NULL);
2664 osd_trans_exec_check(env, th, OSD_OT_INSERT);
2669 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2670 u64 seq, struct lu_seq_range *range)
2672 struct seq_server_site *ss = osd_seq_site(osd);
2674 if (fid_seq_is_idif(seq)) {
2675 fld_range_set_ost(range);
2676 range->lsr_index = idif_ost_idx(seq);
2680 if (!fid_seq_in_fldb(seq)) {
2681 fld_range_set_mdt(range);
2683 /* FIXME: If ss is NULL, it suppose not get lsr_index
2685 range->lsr_index = ss->ss_node_id;
2689 LASSERT(ss != NULL);
2690 fld_range_set_any(range);
2691 /* OSD will only do local fld lookup */
2692 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2696 * Concurrency: no external locking is necessary.
2698 static int osd_declare_object_create(const struct lu_env *env,
2699 struct dt_object *dt,
2700 struct lu_attr *attr,
2701 struct dt_allocation_hint *hint,
2702 struct dt_object_format *dof,
2703 struct thandle *handle)
2705 struct osd_thandle *oh;
2709 LASSERT(handle != NULL);
2711 oh = container_of0(handle, struct osd_thandle, ot_super);
2712 LASSERT(oh->ot_handle == NULL);
2714 /* EA object consumes more credits than regular object: osd_mk_index
2715 * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
2716 * leaf_node, could involves the block, block bitmap, groups, GDT
2717 * change for each block, so add 4 * 2 credits in that case. */
2718 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2719 osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
2720 (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
2721 /* Reuse idle OI block may cause additional one OI block
2723 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2724 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2729 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2730 osd_dt_obj(dt), false, NULL, false);
2734 /* will help to find FID->ino mapping at dt_insert() */
2735 rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
2741 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2742 struct lu_attr *attr,
2743 struct dt_allocation_hint *hint,
2744 struct dt_object_format *dof, struct thandle *th)
2746 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2747 struct osd_object *obj = osd_dt_obj(dt);
2748 struct osd_thread_info *info = osd_oti_get(env);
2752 if (dt_object_exists(dt))
2755 LINVRNT(osd_invariant(obj));
2756 LASSERT(!dt_object_remote(dt));
2757 LASSERT(osd_write_locked(env, obj));
2758 LASSERT(th != NULL);
2760 if (unlikely(fid_is_acct(fid)))
2761 /* Quota files can't be created from the kernel any more,
2762 * 'tune2fs -O quota' will take care of creating them */
2765 result = __osd_object_create(info, obj, attr, hint, dof, th);
2767 result = __osd_oi_insert(env, obj, fid, th);
2768 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
2769 obj->oo_dt.do_body_ops = &osd_body_ops;
2771 LASSERT(ergo(result == 0,
2772 dt_object_exists(dt) && !dt_object_remote(dt)));
2774 LASSERT(osd_invariant(obj));
2779 * Called to destroy on-disk representation of the object
2781 * Concurrency: must be locked
2783 static int osd_declare_object_destroy(const struct lu_env *env,
2784 struct dt_object *dt,
2787 struct osd_object *obj = osd_dt_obj(dt);
2788 struct inode *inode = obj->oo_inode;
2789 struct osd_thandle *oh;
2796 oh = container_of0(th, struct osd_thandle, ot_super);
2797 LASSERT(oh->ot_handle == NULL);
2799 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2800 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2801 /* Recycle idle OI leaf may cause additional three OI blocks
2803 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2804 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2805 /* one less inode */
2806 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2807 -1, oh, obj, false, NULL, false);
2810 /* data to be truncated */
2811 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2812 0, oh, obj, true, NULL, false);
2816 /* will help to find FID->ino when this object is being
2817 * added to PENDING/ */
2818 rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
2823 static int osd_object_destroy(const struct lu_env *env,
2824 struct dt_object *dt,
2827 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2828 struct osd_object *obj = osd_dt_obj(dt);
2829 struct inode *inode = obj->oo_inode;
2830 struct osd_device *osd = osd_obj2dev(obj);
2831 struct osd_thandle *oh;
2835 oh = container_of0(th, struct osd_thandle, ot_super);
2836 LASSERT(oh->ot_handle);
2838 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2840 if (unlikely(fid_is_acct(fid)))
2843 if (S_ISDIR(inode->i_mode)) {
2844 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2845 inode->i_nlink == 2);
2846 /* it will check/delete the inode from remote parent,
2847 * how to optimize it? unlink performance impaction XXX */
2848 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2849 if (result != 0 && result != -ENOENT) {
2850 CERROR("%s: delete inode "DFID": rc = %d\n",
2851 osd_name(osd), PFID(fid), result);
2853 spin_lock(&obj->oo_guard);
2855 spin_unlock(&obj->oo_guard);
2856 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2859 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2861 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
2862 result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2865 osd_trans_exec_check(env, th, OSD_OT_DESTROY);
2866 /* XXX: add to ext3 orphan list */
2867 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2869 /* not needed in the cache anymore */
2870 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2871 obj->oo_destroyed = 1;
2877 * Put the fid into lustre_mdt_attrs, and then place the structure
2878 * inode's ea. This fid should not be altered during the life time
2881 * \retval +ve, on success
2882 * \retval -ve, on error
2884 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2886 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2887 const struct lu_fid *fid, __u32 compat, __u32 incompat)
2889 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2893 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2896 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
2899 lustre_lma_init(lma, fid, compat, incompat);
2900 lustre_lma_swab(lma);
2902 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2904 /* LMA may already exist, but we need to check that all the
2905 * desired compat/incompat flags have been added. */
2906 if (unlikely(rc == -EEXIST)) {
2907 if (compat == 0 && incompat == 0)
2910 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2911 XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2916 lustre_lma_swab(lma);
2917 if (!(~lma->lma_compat & compat) &&
2918 !(~lma->lma_incompat & incompat))
2921 lma->lma_compat |= compat;
2922 lma->lma_incompat |= incompat;
2923 lustre_lma_swab(lma);
2924 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2925 sizeof(*lma), XATTR_REPLACE);
2932 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2933 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2934 * To have compatilibility with 1.8 ldiskfs driver we need to have
2935 * magic number at start of fid data.
2936 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2939 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2940 const struct lu_fid *fid)
2942 if (!fid_is_namespace_visible(fid) ||
2943 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2944 param->edp_magic = 0;
2948 param->edp_magic = LDISKFS_LUFID_MAGIC;
2949 param->edp_len = sizeof(struct lu_fid) + 1;
2950 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2954 * Try to read the fid from inode ea into dt_rec.
2956 * \param fid object fid.
2958 * \retval 0 on success
2960 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2961 __u32 ino, struct lu_fid *fid,
2962 struct osd_inode_id *id)
2964 struct osd_thread_info *info = osd_oti_get(env);
2965 struct inode *inode;
2968 osd_id_gen(id, ino, OSD_OII_NOGEN);
2969 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2971 RETURN(PTR_ERR(inode));
2977 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2979 struct inode *parent_dir,
2980 const struct lu_fid *dot_fid,
2981 const struct lu_fid *dot_dot_fid,
2982 struct osd_thandle *oth)
2984 struct ldiskfs_dentry_param *dot_ldp;
2985 struct ldiskfs_dentry_param *dot_dot_ldp;
2986 __u32 saved_nlink = dir->i_nlink;
2989 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2990 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2992 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2993 dot_ldp->edp_magic = 0;
2995 rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2996 dir, dot_ldp, dot_dot_ldp);
2997 /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
2998 * the subseqent ref_add() will increase the dir->i_nlink
2999 * as 3. That is incorrect for new created directory.
3001 * It looks like hack, because we want to make the OSD API
3002 * to be order-independent for new created directory object
3003 * between dt_insert(..) and ref_add() operations.
3005 * Here, we only restore the in-RAM dir-inode's nlink attr,
3006 * becuase if the nlink attr is not 2, then there will be
3007 * ref_add() called following the dt_insert(..), such call
3008 * will make both the in-RAM and on-disk dir-inode's nlink
3009 * attr to be set as 2. LU-7447 */
3010 set_nlink(dir, saved_nlink);
3015 * Create an local agent inode for remote entry
3017 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3018 struct osd_device *osd,
3019 struct osd_object *pobj,
3020 const struct lu_fid *fid,
3024 struct osd_thread_info *info = osd_oti_get(env);
3025 struct inode *local;
3026 struct osd_thandle *oh;
3031 oh = container_of(th, struct osd_thandle, ot_super);
3032 LASSERT(oh->ot_handle->h_transaction != NULL);
3034 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3035 if (IS_ERR(local)) {
3036 CERROR("%s: create local error %d\n", osd_name(osd),
3037 (int)PTR_ERR(local));
3041 ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3042 unlock_new_inode(local);
3044 /* Set special LMA flag for local agent inode */
3045 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3047 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3048 osd_name(osd), PFID(fid), rc);
3049 RETURN(ERR_PTR(rc));
3055 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3056 lu_object_fid(&pobj->oo_dt.do_lu),
3059 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3060 osd_name(osd), PFID(fid), rc);
3061 RETURN(ERR_PTR(rc));
3068 * when direntry is deleted, we have to take care of possible agent inode
3069 * referenced by that. unfortunately we can't do this at that point:
3070 * iget() within a running transaction leads to deadlock and we better do
3071 * not call that every delete declaration to save performance. so we put
3072 * a potention agent inode on a list and process that once the transaction
3073 * is over. Notice it's not any worse in terms of real orphans as regular
3074 * object destroy doesn't put inodes on the on-disk orphan list. this should
3075 * be addressed separately
3077 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3078 struct osd_thandle *oh,
3081 struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3082 struct osd_obj_orphan *oor;
3089 oor->oor_env = (struct lu_env *)env;
3090 spin_lock(&osd->od_osfs_lock);
3091 list_add(&oor->oor_list, &osd->od_orphan_list);
3092 spin_unlock(&osd->od_osfs_lock);
3094 oh->ot_remove_agents = 1;
3100 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3101 struct osd_device *osd)
3103 struct osd_thread_info *info = osd_oti_get(env);
3104 struct osd_obj_orphan *oor, *tmp;
3105 struct osd_inode_id id;
3106 struct list_head list;
3107 struct inode *inode;
3112 INIT_LIST_HEAD(&list);
3114 spin_lock(&osd->od_osfs_lock);
3115 list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3116 if (oor->oor_env == env) {
3117 list_del(&oor->oor_list);
3118 list_add(&oor->oor_list, &list);
3121 spin_unlock(&osd->od_osfs_lock);
3123 list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3127 list_del(&oor->oor_list);
3130 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3131 inode = osd_iget_fid(info, osd, &id, &fid);
3135 if (!osd_remote_fid(env, osd, &fid)) {
3140 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3142 mark_inode_dirty(inode);
3143 ldiskfs_journal_stop(jh);
3151 * OSD layer object create function for interoperability mode (b11826).
3152 * This is mostly similar to osd_object_create(). Only difference being, fid is
3153 * inserted into inode ea here.
3155 * \retval 0, on success
3156 * \retval -ve, on error
3158 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
3159 struct lu_attr *attr,
3160 struct dt_allocation_hint *hint,
3161 struct dt_object_format *dof,
3164 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3165 struct osd_object *obj = osd_dt_obj(dt);
3166 struct osd_thread_info *info = osd_oti_get(env);
3167 int result, on_ost = 0;
3171 if (dt_object_exists(dt))
3174 LASSERT(osd_invariant(obj));
3175 LASSERT(!dt_object_remote(dt));
3176 LASSERT(osd_write_locked(env, obj));
3177 LASSERT(th != NULL);
3179 if (unlikely(fid_is_acct(fid)))
3180 /* Quota files can't be created from the kernel any more,
3181 * 'tune2fs -O quota' will take care of creating them */
3184 result = __osd_object_create(info, obj, attr, hint, dof, th);
3186 if (fid_is_idif(fid) &&
3187 !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3188 struct lu_fid *tfid = &info->oti_fid;
3189 struct ost_id *oi = &info->oti_ostid;
3191 fid_to_ostid(fid, oi);
3192 ostid_to_fid(tfid, oi, 0);
3194 result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3195 LMAC_FID_ON_OST, 0);
3197 on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3199 result = osd_ea_fid_set(info, obj->oo_inode, fid,
3200 on_ost ? LMAC_FID_ON_OST : 0,
3203 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3204 obj->oo_dt.do_body_ops = &osd_body_ops;
3208 result = __osd_oi_insert(env, obj, fid, th);
3210 /* a small optimization - dt_insert() isn't usually applied
3211 * to OST objects, so we don't need to cache OI mapping for
3213 if (result == 0 && on_ost == 0) {
3214 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3215 result = osd_idc_find_and_init(env, osd, obj);
3216 LASSERT(result == 0);
3219 LASSERT(ergo(result == 0,
3220 dt_object_exists(dt) && !dt_object_remote(dt)));
3221 LINVRNT(osd_invariant(obj));
3225 static int osd_declare_object_ref_add(const struct lu_env *env,
3226 struct dt_object *dt,
3227 struct thandle *handle)
3229 struct osd_thandle *oh;
3231 /* it's possible that object doesn't exist yet */
3232 LASSERT(handle != NULL);
3234 oh = container_of0(handle, struct osd_thandle, ot_super);
3235 LASSERT(oh->ot_handle == NULL);
3237 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3238 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3244 * Concurrency: @dt is write locked.
3246 static int osd_object_ref_add(const struct lu_env *env,
3247 struct dt_object *dt, struct thandle *th)
3249 struct osd_object *obj = osd_dt_obj(dt);
3250 struct inode *inode = obj->oo_inode;
3251 struct osd_thandle *oh;
3254 if (!dt_object_exists(dt) || obj->oo_destroyed)
3257 LINVRNT(osd_invariant(obj));
3258 LASSERT(!dt_object_remote(dt));
3259 LASSERT(osd_write_locked(env, obj));
3260 LASSERT(th != NULL);
3262 oh = container_of0(th, struct osd_thandle, ot_super);
3263 LASSERT(oh->ot_handle != NULL);
3265 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
3267 CDEBUG(D_INODE, DFID" increase nlink %d\n",
3268 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3270 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
3271 * (65000) subdirectories by storing "1" in i_nlink if the link count
3272 * would otherwise overflow. Directory tranversal tools understand
3273 * that (st_nlink == 1) indicates that the filesystem dose not track
3274 * hard links count on the directory, and will not abort subdirectory
3275 * scanning early once (st_nlink - 2) subdirs have been found.
3277 * This also has to properly handle the case of inodes with nlink == 0
3278 * in case they are being linked into the PENDING directory
3280 spin_lock(&obj->oo_guard);
3281 if (unlikely(inode->i_nlink == 0))
3282 /* inc_nlink from 0 may cause WARN_ON */
3283 set_nlink(inode, 1);
3285 ldiskfs_inc_count(oh->ot_handle, inode);
3286 if (!S_ISDIR(inode->i_mode))
3287 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
3289 spin_unlock(&obj->oo_guard);
3291 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3292 LINVRNT(osd_invariant(obj));
3294 osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
3299 static int osd_declare_object_ref_del(const struct lu_env *env,
3300 struct dt_object *dt,
3301 struct thandle *handle)
3303 struct osd_thandle *oh;
3305 if (!dt_object_exists(dt))
3308 LASSERT(!dt_object_remote(dt));
3309 LASSERT(handle != NULL);
3311 oh = container_of0(handle, struct osd_thandle, ot_super);
3312 LASSERT(oh->ot_handle == NULL);
3314 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
3315 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3321 * Concurrency: @dt is write locked.
3323 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
3326 struct osd_object *obj = osd_dt_obj(dt);
3327 struct inode *inode = obj->oo_inode;
3328 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3329 struct osd_thandle *oh;
3331 if (!dt_object_exists(dt))
3334 LINVRNT(osd_invariant(obj));
3335 LASSERT(!dt_object_remote(dt));
3336 LASSERT(osd_write_locked(env, obj));
3337 LASSERT(th != NULL);
3339 oh = container_of0(th, struct osd_thandle, ot_super);
3340 LASSERT(oh->ot_handle != NULL);
3342 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
3344 spin_lock(&obj->oo_guard);
3345 /* That can be result of upgrade from old Lustre version and
3346 * applied only to local files. Just skip this ref_del call.
3347 * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
3348 if (inode->i_nlink == 0) {
3349 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
3350 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
3351 ", maybe an upgraded file? (LU-3915)\n",
3352 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
3353 spin_unlock(&obj->oo_guard);
3357 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
3358 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3360 ldiskfs_dec_count(oh->ot_handle, inode);
3361 spin_unlock(&obj->oo_guard);
3363 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3364 LINVRNT(osd_invariant(obj));
3366 osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
3372 * Get the 64-bit version for an inode.
3374 static int osd_object_version_get(const struct lu_env *env,
3375 struct dt_object *dt, dt_obj_version_t *ver)
3377 struct inode *inode = osd_dt_obj(dt)->oo_inode;
3379 CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
3380 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3381 *ver = LDISKFS_I(inode)->i_fs_version;
3386 * Concurrency: @dt is read locked.
3388 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
3389 struct lu_buf *buf, const char *name)
3391 struct osd_object *obj = osd_dt_obj(dt);
3392 struct inode *inode = obj->oo_inode;
3393 struct osd_thread_info *info = osd_oti_get(env);
3394 struct dentry *dentry = &info->oti_obj_dentry;
3396 /* version get is not real XATTR but uses xattr API */
3397 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3398 /* for version we are just using xattr API but change inode
3400 if (buf->lb_len == 0)
3401 return sizeof(dt_obj_version_t);
3403 if (buf->lb_len < sizeof(dt_obj_version_t))
3406 osd_object_version_get(env, dt, buf->lb_buf);
3408 return sizeof(dt_obj_version_t);
3411 if (!dt_object_exists(dt))
3414 LASSERT(!dt_object_remote(dt));
3415 LASSERT(inode->i_op != NULL);
3416 LASSERT(inode->i_op->getxattr != NULL);
3418 return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
3422 static int osd_declare_xattr_set(const struct lu_env *env,
3423 struct dt_object *dt,
3424 const struct lu_buf *buf, const char *name,
3425 int fl, struct thandle *handle)
3427 struct osd_thandle *oh;
3429 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3431 LASSERT(handle != NULL);
3433 oh = container_of0(handle, struct osd_thandle, ot_super);
3434 LASSERT(oh->ot_handle == NULL);
3436 if (strcmp(name, XATTR_NAME_LMA) == 0) {
3437 /* For non-upgrading case, the LMA is set first and
3438 * usually fit inode. But for upgrade case, the LMA
3439 * may be in another separated EA block. */
3440 if (!dt_object_exists(dt))
3442 else if (fl == LU_XATTR_REPLACE)
3446 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3450 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
3455 if (buf->lb_buf == NULL && dt_object_exists(dt)) {
3456 /* learn xattr size from osd_xattr_get if
3457 attribute has not been read yet */
3458 buflen = __osd_xattr_get(
3459 osd_dt_obj(dt)->oo_inode,
3460 &osd_oti_get(env)->oti_obj_dentry,
3465 buflen = buf->lb_len;
3468 if (buflen > sb->s_blocksize) {
3469 credits += osd_calc_bkmap_credits(
3471 (buflen + sb->s_blocksize - 1) >>
3472 sb->s_blocksize_bits);
3476 * xattr set may involve inode quota change, reserve credits for
3477 * dquot_initialize()
3479 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3482 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3488 * Set the 64-bit version for object
3490 static void osd_object_version_set(const struct lu_env *env,
3491 struct dt_object *dt,
3492 dt_obj_version_t *new_version)
3494 struct inode *inode = osd_dt_obj(dt)->oo_inode;
3496 CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
3497 *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3499 LDISKFS_I(inode)->i_fs_version = *new_version;
3500 /** Version is set after all inode operations are finished,
3501 * so we should mark it dirty here */
3502 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3506 * Concurrency: @dt is write locked.
3508 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3509 const struct lu_buf *buf, const char *name, int fl,
3510 struct thandle *handle)
3512 struct osd_object *obj = osd_dt_obj(dt);
3513 struct inode *inode = obj->oo_inode;
3514 struct osd_thread_info *info = osd_oti_get(env);
3519 LASSERT(handle != NULL);
3521 /* version set is not real XATTR */
3522 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3523 /* for version we are just using xattr API but change inode
3525 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3526 osd_object_version_set(env, dt, buf->lb_buf);
3527 return sizeof(dt_obj_version_t);
3530 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
3531 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3533 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3534 if (fl & LU_XATTR_REPLACE)
3535 fs_flags |= XATTR_REPLACE;
3537 if (fl & LU_XATTR_CREATE)
3538 fs_flags |= XATTR_CREATE;
3540 if (strcmp(name, XATTR_NAME_LMV) == 0) {
3541 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3544 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3548 lma->lma_incompat |= LMAI_STRIPED;
3549 lustre_lma_swab(lma);
3550 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3551 sizeof(*lma), XATTR_REPLACE);
3556 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
3557 strcmp(name, XATTR_NAME_LINK) == 0)
3560 rc = __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3562 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3568 * Concurrency: @dt is read locked.
3570 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3571 const struct lu_buf *buf)
3573 struct osd_object *obj = osd_dt_obj(dt);
3574 struct inode *inode = obj->oo_inode;
3575 struct osd_thread_info *info = osd_oti_get(env);
3576 struct dentry *dentry = &info->oti_obj_dentry;
3578 if (!dt_object_exists(dt))
3581 LASSERT(!dt_object_remote(dt));
3582 LASSERT(inode->i_op != NULL);
3583 LASSERT(inode->i_op->listxattr != NULL);
3585 dentry->d_inode = inode;
3586 dentry->d_sb = inode->i_sb;
3587 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3590 static int osd_declare_xattr_del(const struct lu_env *env,
3591 struct dt_object *dt, const char *name,
3592 struct thandle *handle)
3594 struct osd_thandle *oh;
3595 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3597 LASSERT(!dt_object_remote(dt));
3598 LASSERT(handle != NULL);
3600 oh = container_of0(handle, struct osd_thandle, ot_super);
3601 LASSERT(oh->ot_handle == NULL);
3603 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3604 osd_dto_credits_noquota[DTO_XATTR_SET]);
3606 * xattr del may involve inode quota change, reserve credits for
3607 * dquot_initialize()
3609 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3615 * Concurrency: @dt is write locked.
3617 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3618 const char *name, struct thandle *handle)
3620 struct osd_object *obj = osd_dt_obj(dt);
3621 struct inode *inode = obj->oo_inode;
3622 struct osd_thread_info *info = osd_oti_get(env);
3623 struct dentry *dentry = &info->oti_obj_dentry;
3626 if (!dt_object_exists(dt))
3629 LASSERT(!dt_object_remote(dt));
3630 LASSERT(inode->i_op != NULL);
3631 LASSERT(inode->i_op->removexattr != NULL);
3632 LASSERT(handle != NULL);
3634 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3636 ll_vfs_dq_init(inode);
3637 dentry->d_inode = inode;
3638 dentry->d_sb = inode->i_sb;
3639 rc = inode->i_op->removexattr(dentry, name);
3640 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3644 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3645 __u64 start, __u64 end)
3647 struct osd_object *obj = osd_dt_obj(dt);
3648 struct inode *inode = obj->oo_inode;
3649 struct osd_thread_info *info = osd_oti_get(env);
3650 struct dentry *dentry = &info->oti_obj_dentry;
3651 struct file *file = &info->oti_file;
3656 dentry->d_inode = inode;
3657 dentry->d_sb = inode->i_sb;
3658 file->f_path.dentry = dentry;
3659 file->f_mapping = inode->i_mapping;
3660 file->f_op = inode->i_fop;
3661 set_file_inode(file, inode);
3663 rc = ll_vfs_fsync_range(file, start, end, 0);
3672 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3673 const struct dt_index_features *feat)
3675 struct iam_descr *descr;
3677 if (osd_object_is_root(o))
3678 return feat == &dt_directory_features;
3680 LASSERT(o->oo_dir != NULL);
3682 descr = o->oo_dir->od_container.ic_descr;
3683 if (feat == &dt_directory_features) {
3684 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3690 feat->dif_keysize_min <= descr->id_key_size &&
3691 descr->id_key_size <= feat->dif_keysize_max &&
3692 feat->dif_recsize_min <= descr->id_rec_size &&
3693 descr->id_rec_size <= feat->dif_recsize_max &&
3694 !(feat->dif_flags & (DT_IND_VARKEY |
3695 DT_IND_VARREC | DT_IND_NONUNQ)) &&
3696 ergo(feat->dif_flags & DT_IND_UPDATE,
3697 1 /* XXX check that object (and file system) is
3702 static int osd_iam_container_init(const struct lu_env *env,
3703 struct osd_object *obj,
3704 struct osd_directory *dir)
3706 struct iam_container *bag = &dir->od_container;
3709 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3713 result = iam_container_setup(bag);
3715 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3717 iam_container_fini(bag);
3724 * Concurrency: no external locking is necessary.
3726 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3727 const struct dt_index_features *feat)
3731 struct osd_object *obj = osd_dt_obj(dt);
3733 LINVRNT(osd_invariant(obj));
3735 if (osd_object_is_root(obj)) {
3736 dt->do_index_ops = &osd_index_ea_ops;
3738 } else if (feat == &dt_directory_features) {
3739 dt->do_index_ops = &osd_index_ea_ops;
3740 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
3745 } else if (unlikely(feat == &dt_otable_features)) {
3746 dt->do_index_ops = &osd_otable_ops;
3748 } else if (unlikely(feat == &dt_acct_features)) {
3749 dt->do_index_ops = &osd_acct_index_ops;
3752 } else if (!osd_has_index(obj)) {
3753 struct osd_directory *dir;
3758 spin_lock(&obj->oo_guard);
3759 if (obj->oo_dir == NULL)
3763 * Concurrent thread allocated container data.
3766 spin_unlock(&obj->oo_guard);
3768 * Now, that we have container data, serialize its
3771 down_write(&obj->oo_ext_idx_sem);
3773 * recheck under lock.
3775 if (!osd_has_index(obj))
3776 result = osd_iam_container_init(env, obj,
3780 up_write(&obj->oo_ext_idx_sem);
3788 if (result == 0 && skip_iam == 0) {
3789 if (!osd_iam_index_probe(env, obj, feat))
3792 LINVRNT(osd_invariant(obj));
3794 if (result == 0 && feat == &dt_quota_glb_features &&
3795 fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3796 result = osd_quota_migration(env, dt);
3801 static int osd_otable_it_attr_get(const struct lu_env *env,
3802 struct dt_object *dt,
3803 struct lu_attr *attr)
3809 static const struct dt_object_operations osd_obj_ops = {
3810 .do_read_lock = osd_object_read_lock,
3811 .do_write_lock = osd_object_write_lock,
3812 .do_read_unlock = osd_object_read_unlock,
3813 .do_write_unlock = osd_object_write_unlock,
3814 .do_write_locked = osd_object_write_locked,
3815 .do_attr_get = osd_attr_get,
3816 .do_declare_attr_set = osd_declare_attr_set,
3817 .do_attr_set = osd_attr_set,
3818 .do_ah_init = osd_ah_init,
3819 .do_declare_create = osd_declare_object_create,
3820 .do_create = osd_object_create,
3821 .do_declare_destroy = osd_declare_object_destroy,
3822 .do_destroy = osd_object_destroy,
3823 .do_index_try = osd_index_try,
3824 .do_declare_ref_add = osd_declare_object_ref_add,
3825 .do_ref_add = osd_object_ref_add,
3826 .do_declare_ref_del = osd_declare_object_ref_del,
3827 .do_ref_del = osd_object_ref_del,
3828 .do_xattr_get = osd_xattr_get,
3829 .do_declare_xattr_set = osd_declare_xattr_set,
3830 .do_xattr_set = osd_xattr_set,
3831 .do_declare_xattr_del = osd_declare_xattr_del,
3832 .do_xattr_del = osd_xattr_del,
3833 .do_xattr_list = osd_xattr_list,
3834 .do_object_sync = osd_object_sync,
3838 * dt_object_operations for interoperability mode
3839 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3841 static const struct dt_object_operations osd_obj_ea_ops = {
3842 .do_read_lock = osd_object_read_lock,
3843 .do_write_lock = osd_object_write_lock,
3844 .do_read_unlock = osd_object_read_unlock,
3845 .do_write_unlock = osd_object_write_unlock,
3846 .do_write_locked = osd_object_write_locked,
3847 .do_attr_get = osd_attr_get,
3848 .do_declare_attr_set = osd_declare_attr_set,
3849 .do_attr_set = osd_attr_set,
3850 .do_ah_init = osd_ah_init,
3851 .do_declare_create = osd_declare_object_create,
3852 .do_create = osd_object_ea_create,
3853 .do_declare_destroy = osd_declare_object_destroy,
3854 .do_destroy = osd_object_destroy,
3855 .do_index_try = osd_index_try,
3856 .do_declare_ref_add = osd_declare_object_ref_add,
3857 .do_ref_add = osd_object_ref_add,
3858 .do_declare_ref_del = osd_declare_object_ref_del,
3859 .do_ref_del = osd_object_ref_del,
3860 .do_xattr_get = osd_xattr_get,
3861 .do_declare_xattr_set = osd_declare_xattr_set,
3862 .do_xattr_set = osd_xattr_set,
3863 .do_declare_xattr_del = osd_declare_xattr_del,
3864 .do_xattr_del = osd_xattr_del,
3865 .do_xattr_list = osd_xattr_list,
3866 .do_object_sync = osd_object_sync,
3869 static const struct dt_object_operations osd_obj_otable_it_ops = {
3870 .do_attr_get = osd_otable_it_attr_get,
3871 .do_index_try = osd_index_try,
3874 static int osd_index_declare_iam_delete(const struct lu_env *env,
3875 struct dt_object *dt,
3876 const struct dt_key *key,
3877 struct thandle *handle)
3879 struct osd_thandle *oh;
3881 oh = container_of0(handle, struct osd_thandle, ot_super);
3882 LASSERT(oh->ot_handle == NULL);
3884 /* Recycle may cause additional three blocks to be changed. */
3885 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3886 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3892 * delete a (key, value) pair from index \a dt specified by \a key
3894 * \param dt osd index object
3895 * \param key key for index
3896 * \param rec record reference
3897 * \param handle transaction handler
3900 * \retval -ve failure
3902 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3903 const struct dt_key *key,
3904 struct thandle *handle)
3906 struct osd_thread_info *oti = osd_oti_get(env);
3907 struct osd_object *obj = osd_dt_obj(dt);
3908 struct osd_thandle *oh;
3909 struct iam_path_descr *ipd;
3910 struct iam_container *bag = &obj->oo_dir->od_container;
3914 if (!dt_object_exists(dt))
3917 LINVRNT(osd_invariant(obj));
3918 LASSERT(!dt_object_remote(dt));
3919 LASSERT(bag->ic_object == obj->oo_inode);
3920 LASSERT(handle != NULL);
3922 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3924 ipd = osd_idx_ipd_get(env, bag);
3925 if (unlikely(ipd == NULL))
3928 oh = container_of0(handle, struct osd_thandle, ot_super);
3929 LASSERT(oh->ot_handle != NULL);
3930 LASSERT(oh->ot_handle->h_transaction != NULL);
3932 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3933 /* swab quota uid/gid provided by caller */
3934 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3935 key = (const struct dt_key *)&oti->oti_quota_id;
3938 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3939 osd_ipd_put(env, bag, ipd);
3940 LINVRNT(osd_invariant(obj));
3941 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
3945 static int osd_index_declare_ea_delete(const struct lu_env *env,
3946 struct dt_object *dt,
3947 const struct dt_key *key,
3948 struct thandle *handle)
3950 struct osd_thandle *oh;
3951 struct inode *inode;
3955 LASSERT(!dt_object_remote(dt));
3956 LASSERT(handle != NULL);
3958 oh = container_of0(handle, struct osd_thandle, ot_super);
3959 LASSERT(oh->ot_handle == NULL);
3961 /* due to DNE we may need to remove an agent inode */
3962 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3963 osd_dto_credits_noquota[DTO_INDEX_DELETE] +
3964 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3966 inode = osd_dt_obj(dt)->oo_inode;
3970 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3971 0, oh, osd_dt_obj(dt), true, NULL, false);
3975 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3978 struct osd_fid_pack *rec;
3981 if (de->file_type & LDISKFS_DIRENT_LUFID) {
3982 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3983 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3984 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
3990 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3991 const struct lu_fid *fid)
3993 struct seq_server_site *ss = osd_seq_site(osd);
3996 /* FID seqs not in FLDB, must be local seq */
3997 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
4000 /* If FLD is not being initialized yet, it only happens during the
4001 * initialization, likely during mgs initialization, and we assume
4002 * this is local FID. */
4003 if (ss == NULL || ss->ss_server_fld == NULL)
4006 /* Only check the local FLDB here */
4007 if (osd_seq_exists(env, osd, fid_seq(fid)))
4014 * Index delete function for interoperability mode (b11826).
4015 * It will remove the directory entry added by osd_index_ea_insert().
4016 * This entry is needed to maintain name->fid mapping.
4018 * \param key, key i.e. file entry to be deleted
4020 * \retval 0, on success
4021 * \retval -ve, on error
4023 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
4024 const struct dt_key *key, struct thandle *handle)
4026 struct osd_object *obj = osd_dt_obj(dt);
4027 struct inode *dir = obj->oo_inode;
4028 struct dentry *dentry;
4029 struct osd_thandle *oh;
4030 struct ldiskfs_dir_entry_2 *de = NULL;
4031 struct buffer_head *bh;
4032 struct htree_lock *hlock = NULL;
4033 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
4034 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4038 if (!dt_object_exists(dt))
4041 LINVRNT(osd_invariant(obj));
4042 LASSERT(!dt_object_remote(dt));
4043 LASSERT(handle != NULL);
4045 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4047 oh = container_of(handle, struct osd_thandle, ot_super);
4048 LASSERT(oh->ot_handle != NULL);
4049 LASSERT(oh->ot_handle->h_transaction != NULL);
4051 ll_vfs_dq_init(dir);
4052 dentry = osd_child_dentry_get(env, obj,
4053 (char *)key, strlen((char *)key));
4055 if (obj->oo_hl_head != NULL) {
4056 hlock = osd_oti_get(env)->oti_hlock;
4057 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4058 dir, LDISKFS_HLOCK_DEL);
4060 down_write(&obj->oo_ext_idx_sem);
4063 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4065 /* If this is not the ".." entry, it might be a remote DNE
4066 * entry and we need to check if the FID is for a remote
4067 * MDT. If the FID is not in the directory entry (e.g.
4068 * upgraded 1.8 filesystem without dirdata enabled) then
4069 * we need to get the FID from the LMA. For a remote directory
4070 * there HAS to be an LMA, it cannot be an IGIF inode in this
4073 * Delete the entry before the agent inode in order to
4074 * simplify error handling. At worst an error after deleting
4075 * the entry first might leak the agent inode afterward. The
4076 * reverse would need filesystem abort in case of error deleting
4077 * the entry after the agent had been removed, or leave a
4078 * dangling entry pointing at a random inode. */
4079 if (strcmp((char *)key, dotdot) != 0) {
4080 LASSERT(de != NULL);
4081 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
4082 if (rc == -ENODATA) {
4083 /* can't get FID, postpone to the end of the
4084 * transaction when iget() is safe */
4085 osd_schedule_agent_inode_removal(env, oh,
4086 le32_to_cpu(de->inode));
4087 } else if (rc == 0 &&
4088 unlikely(osd_remote_fid(env, osd, fid))) {
4089 osd_schedule_agent_inode_removal(env, oh,
4090 le32_to_cpu(de->inode));
4093 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
4099 ldiskfs_htree_unlock(hlock);
4101 up_write(&obj->oo_ext_idx_sem);
4106 /* For inode on the remote MDT, .. will point to
4107 * /Agent directory, Check whether it needs to delete
4108 * from agent directory */
4109 if (unlikely(strcmp((char *)key, dotdot) == 0)) {
4110 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
4112 if (rc != 0 && rc != -ENOENT) {
4113 CERROR("%s: delete agent inode "DFID": rc = %d\n",
4114 osd_name(osd), PFID(fid), rc);
4124 LASSERT(osd_invariant(obj));
4125 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4130 * Lookup index for \a key and copy record to \a rec.
4132 * \param dt osd index object
4133 * \param key key for index
4134 * \param rec record reference
4136 * \retval +ve success : exact mach
4137 * \retval 0 return record with key not greater than \a key
4138 * \retval -ve failure
4140 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
4141 struct dt_rec *rec, const struct dt_key *key)
4143 struct osd_object *obj = osd_dt_obj(dt);
4144 struct iam_path_descr *ipd;
4145 struct iam_container *bag = &obj->oo_dir->od_container;
4146 struct osd_thread_info *oti = osd_oti_get(env);
4147 struct iam_iterator *it = &oti->oti_idx_it;
4148 struct iam_rec *iam_rec;
4152 if (!dt_object_exists(dt))
4155 LASSERT(osd_invariant(obj));
4156 LASSERT(!dt_object_remote(dt));
4157 LASSERT(bag->ic_object == obj->oo_inode);
4159 ipd = osd_idx_ipd_get(env, bag);
4163 /* got ipd now we can start iterator. */
4164 iam_it_init(it, bag, 0, ipd);
4166 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4167 /* swab quota uid/gid provided by caller */
4168 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4169 key = (const struct dt_key *)&oti->oti_quota_id;
4172 rc = iam_it_get(it, (struct iam_key *)key);
4174 if (S_ISDIR(obj->oo_inode->i_mode))
4175 iam_rec = (struct iam_rec *)oti->oti_ldp;
4177 iam_rec = (struct iam_rec *) rec;
4179 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
4181 if (S_ISDIR(obj->oo_inode->i_mode))
4182 osd_fid_unpack((struct lu_fid *) rec,
4183 (struct osd_fid_pack *)iam_rec);
4184 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
4185 osd_quota_unpack(obj, rec);
4190 osd_ipd_put(env, bag, ipd);
4192 LINVRNT(osd_invariant(obj));
4197 static int osd_index_declare_iam_insert(const struct lu_env *env,
4198 struct dt_object *dt,
4199 const struct dt_rec *rec,
4200 const struct dt_key *key,
4201 struct thandle *handle)
4203 struct osd_thandle *oh;
4205 LASSERT(handle != NULL);
4207 oh = container_of0(handle, struct osd_thandle, ot_super);
4208 LASSERT(oh->ot_handle == NULL);
4210 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4211 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4217 * Inserts (key, value) pair in \a dt index object.
4219 * \param dt osd index object
4220 * \param key key for index
4221 * \param rec record reference
4222 * \param th transaction handler
4225 * \retval -ve failure
4227 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
4228 const struct dt_rec *rec,
4229 const struct dt_key *key, struct thandle *th,
4232 struct osd_object *obj = osd_dt_obj(dt);
4233 struct iam_path_descr *ipd;
4234 struct osd_thandle *oh;
4235 struct iam_container *bag;
4236 struct osd_thread_info *oti = osd_oti_get(env);
4237 struct iam_rec *iam_rec;
4241 if (!dt_object_exists(dt))
4244 LINVRNT(osd_invariant(obj));
4245 LASSERT(!dt_object_remote(dt));
4247 bag = &obj->oo_dir->od_container;
4248 LASSERT(bag->ic_object == obj->oo_inode);
4249 LASSERT(th != NULL);
4251 osd_trans_exec_op(env, th, OSD_OT_INSERT);
4253 ipd = osd_idx_ipd_get(env, bag);
4254 if (unlikely(ipd == NULL))
4257 oh = container_of0(th, struct osd_thandle, ot_super);
4258 LASSERT(oh->ot_handle != NULL);
4259 LASSERT(oh->ot_handle->h_transaction != NULL);
4260 if (S_ISDIR(obj->oo_inode->i_mode)) {
4261 iam_rec = (struct iam_rec *)oti->oti_ldp;
4262 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
4263 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4264 /* pack quota uid/gid */
4265 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4266 key = (const struct dt_key *)&oti->oti_quota_id;
4267 /* pack quota record */
4268 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
4269 iam_rec = (struct iam_rec *)rec;
4271 iam_rec = (struct iam_rec *)rec;
4274 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
4276 osd_ipd_put(env, bag, ipd);
4277 LINVRNT(osd_invariant(obj));
4278 osd_trans_exec_check(env, th, OSD_OT_INSERT);
4283 * Calls ldiskfs_add_entry() to add directory entry
4284 * into the directory. This is required for
4285 * interoperability mode (b11826)
4287 * \retval 0, on success
4288 * \retval -ve, on error
4290 static int __osd_ea_add_rec(struct osd_thread_info *info,
4291 struct osd_object *pobj, struct inode *cinode,
4292 const char *name, const struct lu_fid *fid,
4293 struct htree_lock *hlock, struct thandle *th)
4295 struct ldiskfs_dentry_param *ldp;
4296 struct dentry *child;
4297 struct osd_thandle *oth;
4300 oth = container_of(th, struct osd_thandle, ot_super);
4301 LASSERT(oth->ot_handle != NULL);
4302 LASSERT(oth->ot_handle->h_transaction != NULL);
4303 LASSERT(pobj->oo_inode);
4305 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
4306 if (unlikely(pobj->oo_inode ==
4307 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
4310 osd_get_ldiskfs_dirent_param(ldp, fid);
4311 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
4312 child->d_fsdata = (void *)ldp;
4313 ll_vfs_dq_init(pobj->oo_inode);
4314 rc = osd_ldiskfs_add_entry(info, oth->ot_handle, child,
4316 if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
4317 struct ldiskfs_dir_entry_2 *de;
4318 struct buffer_head *bh;
4321 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
4324 rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
4327 if (S_ISDIR(cinode->i_mode))
4328 de->file_type = LDISKFS_DIRENT_LUFID |
4329 LDISKFS_FT_REG_FILE;
4331 de->file_type = LDISKFS_DIRENT_LUFID |
4333 ldiskfs_handle_dirty_metadata(oth->ot_handle,
4344 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
4345 * into the directory.Also sets flags into osd object to
4346 * indicate dot and dotdot are created. This is required for
4347 * interoperability mode (b11826)
4349 * \param dir directory for dot and dotdot fixup.
4350 * \param obj child object for linking
4352 * \retval 0, on success
4353 * \retval -ve, on error
4355 static int osd_add_dot_dotdot(struct osd_thread_info *info,
4356 struct osd_object *dir,
4357 struct inode *parent_dir, const char *name,
4358 const struct lu_fid *dot_fid,
4359 const struct lu_fid *dot_dot_fid,
4362 struct inode *inode = dir->oo_inode;
4363 struct osd_thandle *oth;
4366 oth = container_of(th, struct osd_thandle, ot_super);
4367 LASSERT(oth->ot_handle->h_transaction != NULL);
4368 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4370 if (strcmp(name, dot) == 0) {
4371 if (dir->oo_compat_dot_created) {
4374 LASSERT(inode->i_ino == parent_dir->i_ino);
4375 dir->oo_compat_dot_created = 1;
4378 } else if (strcmp(name, dotdot) == 0) {
4379 if (!dir->oo_compat_dot_created)
4381 /* in case of rename, dotdot is already created */
4382 if (dir->oo_compat_dotdot_created) {
4383 return __osd_ea_add_rec(info, dir, parent_dir, name,
4384 dot_dot_fid, NULL, th);
4387 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
4388 struct lu_fid tfid = *dot_dot_fid;
4391 result = osd_add_dot_dotdot_internal(info,
4392 dir->oo_inode, parent_dir, dot_fid,
4395 result = osd_add_dot_dotdot_internal(info,
4396 dir->oo_inode, parent_dir, dot_fid,
4401 dir->oo_compat_dotdot_created = 1;
4409 * It will call the appropriate osd_add* function and return the
4410 * value, return by respective functions.
4412 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4413 struct inode *cinode, const char *name,
4414 const struct lu_fid *fid, struct thandle *th)
4416 struct osd_thread_info *info = osd_oti_get(env);
4417 struct htree_lock *hlock;
4420 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4422 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4424 if (hlock != NULL) {
4425 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4428 down_write(&pobj->oo_ext_idx_sem);
4431 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4432 lu_object_fid(&pobj->oo_dt.do_lu),
4435 if (hlock != NULL) {
4436 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4437 pobj->oo_inode, LDISKFS_HLOCK_ADD);
4439 down_write(&pobj->oo_ext_idx_sem);
4442 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4443 struct lu_fid *tfid = &info->oti_fid;
4447 rc = __osd_ea_add_rec(info, pobj, cinode, name,
4450 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4455 ldiskfs_htree_unlock(hlock);
4457 up_write(&pobj->oo_ext_idx_sem);
4463 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4464 struct osd_idmap_cache *oic)
4466 struct osd_scrub *scrub = &dev->od_scrub;
4467 struct lu_fid *fid = &oic->oic_fid;
4468 struct osd_inode_id *id = &oti->oti_id;
4473 if (!fid_is_norm(fid) && !fid_is_igif(fid))
4476 if (scrub->os_pos_current > id->oii_ino)
4480 rc = osd_oi_lookup(oti, dev, fid, id, 0);
4481 if (rc == -ENOENT) {
4482 struct inode *inode;
4485 inode = osd_iget(oti, dev, &oic->oic_lid);
4487 /* The inode has been removed (by race maybe). */
4488 if (IS_ERR(inode)) {
4489 rc = PTR_ERR(inode);
4491 RETURN(rc == -ESTALE ? -ENOENT : rc);
4495 /* The OI mapping is lost. */
4496 if (id->oii_gen != OSD_OII_NOGEN)
4499 /* The inode may has been reused by others, we do not know,
4500 * leave it to be handled by subsequent osd_fid_lookup(). */
4502 } else if (rc != 0 || osd_id_eq(id, &oic->oic_lid)) {
4507 if (thread_is_running(&scrub->os_thread)) {
4508 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
4509 /* There is race condition between osd_oi_lookup and OI scrub.
4510 * The OI scrub finished just after osd_oi_lookup() failure.
4511 * Under such case, it is unnecessary to trigger OI scrub again,
4512 * but try to call osd_oi_lookup() again. */
4513 if (unlikely(rc == -EAGAIN))
4519 if (!dev->od_noscrub && ++once == 1) {
4520 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
4522 CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI scrub by RPC "
4523 "for "DFID", rc = %d [2]\n",
4524 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4526 if (rc == 0 || rc == -EALREADY)
4533 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4534 struct osd_device *dev,
4535 struct osd_idmap_cache *oic,
4536 struct lu_fid *fid, __u32 ino)
4538 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
4539 struct inode *inode;
4542 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4543 inode = osd_iget(oti, dev, &oic->oic_lid);
4544 if (IS_ERR(inode)) {
4545 fid_zero(&oic->oic_fid);
4546 return PTR_ERR(inode);
4549 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4552 fid_zero(&oic->oic_fid);
4554 *fid = oic->oic_fid = lma->lma_self_fid;
4558 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4559 struct osd_inode_id *id, const struct lu_fid *fid)
4561 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4562 id->oii_ino, id->oii_gen, info);
4563 info->oti_cache.oic_lid = *id;
4564 info->oti_cache.oic_fid = *fid;
4565 info->oti_cache.oic_dev = osd;
4569 * Get parent FID from the linkEA.
4571 * For a directory which parent resides on remote MDT, to satisfy the
4572 * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
4573 * the other hand, to make the lookup(..) on the directory can return
4574 * the real parent FID, we append the real parent FID after its ".."
4575 * name entry in the /REMOTE_PARENT_DIR.
4577 * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
4578 * backup. So after the restore, we cannot get the right parent FID from
4579 * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
4580 * we have stored the real parent FID in the directory object's linkEA,
4581 * we can parse the linkEA for the real parent FID.
4583 * \param[in] env pointer to the thread context
4584 * \param[in] obj pointer to the object to be handled
4585 * \param[out]fid pointer to the buffer to hold the parent FID
4587 * \retval 0 for getting the real parent FID successfully
4588 * \retval negative error number on failure
4590 static int osd_get_pfid_from_linkea(const struct lu_env *env,
4591 struct osd_object *obj,
4594 struct osd_thread_info *oti = osd_oti_get(env);
4595 struct lu_buf *buf = &oti->oti_big_buf;
4596 struct dentry *dentry = &oti->oti_obj_dentry;
4597 struct inode *inode = obj->oo_inode;
4598 struct linkea_data ldata = { NULL };
4603 if (!S_ISDIR(inode->i_mode))
4607 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4608 buf->lb_buf, buf->lb_len);
4609 if (rc == -ERANGE) {
4610 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4613 lu_buf_realloc(buf, rc);
4614 if (buf->lb_buf == NULL)
4621 if (unlikely(rc == 0))
4627 if (unlikely(buf->lb_buf == NULL)) {
4628 lu_buf_realloc(buf, rc);
4629 if (buf->lb_buf == NULL)
4636 rc = linkea_init(&ldata);
4638 linkea_first_entry(&ldata);
4639 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
4646 * Calls ->lookup() to find dentry. From dentry get inode and
4647 * read inode's ea to get fid. This is required for interoperability
4650 * \retval 0, on success
4651 * \retval -ve, on error
4653 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4654 struct dt_rec *rec, const struct dt_key *key)
4656 struct inode *dir = obj->oo_inode;
4657 struct dentry *dentry;
4658 struct ldiskfs_dir_entry_2 *de;
4659 struct buffer_head *bh;
4660 struct lu_fid *fid = (struct lu_fid *) rec;
4661 struct htree_lock *hlock = NULL;
4666 LASSERT(dir->i_op != NULL);
4667 LASSERT(dir->i_op->lookup != NULL);
4669 dentry = osd_child_dentry_get(env, obj,
4670 (char *)key, strlen((char *)key));
4672 if (obj->oo_hl_head != NULL) {
4673 hlock = osd_oti_get(env)->oti_hlock;
4674 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4675 dir, LDISKFS_HLOCK_LOOKUP);
4677 down_read(&obj->oo_ext_idx_sem);
4680 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4682 struct osd_thread_info *oti = osd_oti_get(env);
4683 struct osd_inode_id *id = &oti->oti_id;
4684 struct osd_idmap_cache *oic = &oti->oti_cache;
4685 struct osd_device *dev = osd_obj2dev(obj);
4687 ino = le32_to_cpu(de->inode);
4688 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4690 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4694 rc = osd_get_fid_from_dentry(de, rec);
4696 /* done with de, release bh */
4699 if (unlikely(ino == osd_remote_parent_ino(dev))) {
4700 const char *name = (const char *)key;
4702 /* If the parent is on remote MDT, and there
4703 * is no FID-in-dirent, then we have to get
4704 * the parent FID from the linkEA. */
4705 if (likely(strlen(name) == 2 &&
4706 name[0] == '.' && name[1] == '.'))
4707 rc = osd_get_pfid_from_linkea(env, obj,
4710 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4713 osd_id_gen(id, ino, OSD_OII_NOGEN);
4716 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
4717 fid_zero(&oic->oic_fid);
4722 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
4723 rc = osd_consistency_check(oti, dev, oic);
4725 fid_zero(&oic->oic_fid);
4734 ldiskfs_htree_unlock(hlock);
4736 up_read(&obj->oo_ext_idx_sem);
4741 * Put the osd object once done with it.
4743 * \param obj osd object that needs to be put
4745 static inline void osd_object_put(const struct lu_env *env,
4746 struct osd_object *obj)
4748 lu_object_put(env, &obj->oo_dt.do_lu);
4751 static int osd_index_declare_ea_insert(const struct lu_env *env,
4752 struct dt_object *dt,
4753 const struct dt_rec *rec,
4754 const struct dt_key *key,
4755 struct thandle *handle)
4757 struct osd_thandle *oh;
4758 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4759 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
4760 const struct lu_fid *fid = rec1->rec_fid;
4761 int credits, rc = 0;
4762 struct osd_idmap_cache *idc;
4765 LASSERT(!dt_object_remote(dt));
4766 LASSERT(handle != NULL);
4767 LASSERT(fid != NULL);
4768 LASSERT(rec1->rec_type != 0);
4770 oh = container_of0(handle, struct osd_thandle, ot_super);
4771 LASSERT(oh->ot_handle == NULL);
4773 credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
4775 /* we can't call iget() while a transactions is running
4776 * (this can lead to a deadlock), but we need to know
4777 * inum and object type. so we find this information at
4778 * declaration and cache in per-thread info */
4779 idc = osd_idc_find_or_init(env, osd, fid);
4781 RETURN(PTR_ERR(idc));
4782 if (idc->oic_remote) {
4783 /* a reference to remote inode is represented by an
4784 * agent inode which we have to create */
4785 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
4786 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4789 osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
4791 if (osd_dt_obj(dt)->oo_inode != NULL) {
4792 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4794 /* We ignore block quota on meta pool (MDTs), so needn't
4795 * calculate how many blocks will be consumed by this index
4797 rc = osd_declare_inode_qid(env, i_uid_read(inode),
4798 i_gid_read(inode), 0, oh,
4799 osd_dt_obj(dt), true, NULL, false);
4806 * Index add function for interoperability mode (b11826).
4807 * It will add the directory entry.This entry is needed to
4808 * maintain name->fid mapping.
4810 * \param key it is key i.e. file entry to be inserted
4811 * \param rec it is value of given key i.e. fid
4813 * \retval 0, on success
4814 * \retval -ve, on error
4816 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4817 const struct dt_rec *rec,
4818 const struct dt_key *key, struct thandle *th,
4821 struct osd_object *obj = osd_dt_obj(dt);
4822 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4823 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
4824 const struct lu_fid *fid = rec1->rec_fid;
4825 const char *name = (const char *)key;
4826 struct osd_thread_info *oti = osd_oti_get(env);
4827 struct inode *child_inode = NULL;
4828 struct osd_idmap_cache *idc;
4832 if (!dt_object_exists(dt))
4835 LASSERT(osd_invariant(obj));
4836 LASSERT(!dt_object_remote(dt));
4837 LASSERT(th != NULL);
4839 osd_trans_exec_op(env, th, OSD_OT_INSERT);
4841 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
4843 idc = osd_idc_find(env, osd, fid);
4844 if (unlikely(idc == NULL)) {
4845 /* this dt_insert() wasn't declared properly, so
4846 * FID is missing in OI cache. we better do not
4847 * lookup FID in FLDB/OI and don't risk to deadlock,
4848 * but in some special cases (lfsck testing, etc)
4849 * it's much simpler than fixing a caller */
4850 CERROR("%s: "DFID" wasn't declared for insert\n",
4851 osd_name(osd), PFID(fid));
4853 idc = osd_idc_find_or_init(env, osd, fid);
4855 RETURN(PTR_ERR(idc));
4858 if (idc->oic_remote) {
4859 /* Insert remote entry */
4860 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4861 struct osd_mdobj_map *omm = osd->od_mdt_map;
4862 struct osd_thandle *oh;
4864 /* If parent on remote MDT, we need put this object
4866 oh = container_of(th, typeof(*oh), ot_super);
4867 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4869 CERROR("%s: add "DFID" error: rc = %d\n",
4871 PFID(lu_object_fid(&dt->do_lu)), rc);
4875 child_inode = igrab(omm->omm_remote_parent->d_inode);
4877 child_inode = osd_create_local_agent_inode(env, osd,
4878 obj, fid, rec1->rec_type & S_IFMT, th);
4879 if (IS_ERR(child_inode))
4880 RETURN(PTR_ERR(child_inode));
4883 /* Insert local entry */
4884 if (unlikely(idc->oic_lid.oii_ino == 0)) {
4885 /* for a reason OI cache wasn't filled properly */
4886 CERROR("%s: OIC for "DFID" isn't filled\n",
4887 osd_name(osd), PFID(fid));
4890 child_inode = oti->oti_inode;
4891 if (unlikely(child_inode == NULL)) {
4892 struct ldiskfs_inode_info *lii;
4896 child_inode = oti->oti_inode = &lii->vfs_inode;
4898 child_inode->i_sb = osd_sb(osd);
4899 child_inode->i_ino = idc->oic_lid.oii_ino;
4900 child_inode->i_mode = rec1->rec_type & S_IFMT;
4903 rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
4905 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4906 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4908 if (child_inode && child_inode != oti->oti_inode)
4910 LASSERT(osd_invariant(obj));
4911 osd_trans_exec_check(env, th, OSD_OT_INSERT);
4916 * Initialize osd Iterator for given osd index object.
4918 * \param dt osd index object
4921 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4922 struct dt_object *dt,
4925 struct osd_it_iam *it;
4926 struct osd_object *obj = osd_dt_obj(dt);
4927 struct lu_object *lo = &dt->do_lu;
4928 struct iam_path_descr *ipd;
4929 struct iam_container *bag = &obj->oo_dir->od_container;
4931 if (!dt_object_exists(dt))
4932 return ERR_PTR(-ENOENT);
4936 return ERR_PTR(-ENOMEM);
4938 ipd = osd_it_ipd_get(env, bag);
4939 if (likely(ipd != NULL)) {
4943 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4944 return (struct dt_it *)it;
4947 return ERR_PTR(-ENOMEM);
4952 * free given Iterator.
4955 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4957 struct osd_it_iam *it = (struct osd_it_iam *)di;
4958 struct osd_object *obj = it->oi_obj;
4960 iam_it_fini(&it->oi_it);
4961 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4962 lu_object_put(env, &obj->oo_dt.do_lu);
4967 * Move Iterator to record specified by \a key
4969 * \param di osd iterator
4970 * \param key key for index
4972 * \retval +ve di points to record with least key not larger than key
4973 * \retval 0 di points to exact matched key
4974 * \retval -ve failure
4977 static int osd_it_iam_get(const struct lu_env *env,
4978 struct dt_it *di, const struct dt_key *key)
4980 struct osd_thread_info *oti = osd_oti_get(env);
4981 struct osd_it_iam *it = (struct osd_it_iam *)di;
4983 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4984 /* swab quota uid/gid */
4985 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4986 key = (struct dt_key *)&oti->oti_quota_id;
4989 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4995 * \param di osd iterator
4997 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4999 struct osd_it_iam *it = (struct osd_it_iam *)di;
5001 iam_it_put(&it->oi_it);
5005 * Move iterator by one record
5007 * \param di osd iterator
5009 * \retval +1 end of container reached
5011 * \retval -ve failure
5014 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
5016 struct osd_it_iam *it = (struct osd_it_iam *)di;
5018 return iam_it_next(&it->oi_it);
5022 * Return pointer to the key under iterator.
5025 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
5026 const struct dt_it *di)
5028 struct osd_thread_info *oti = osd_oti_get(env);
5029 struct osd_it_iam *it = (struct osd_it_iam *)di;
5030 struct osd_object *obj = it->oi_obj;
5033 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
5035 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
5036 /* swab quota uid/gid */
5037 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
5038 key = (struct dt_key *)&oti->oti_quota_id;
5045 * Return size of key under iterator (in bytes)
5048 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
5050 struct osd_it_iam *it = (struct osd_it_iam *)di;
5052 return iam_it_key_size(&it->oi_it);
5056 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
5058 /* check if file type is required */
5059 if (ent->lde_attrs & LUDA_TYPE) {
5060 struct luda_type *lt;
5061 int align = sizeof(*lt) - 1;
5063 len = (len + align) & ~align;
5064 lt = (struct luda_type *)(ent->lde_name + len);
5065 lt->lt_type = cpu_to_le16(DTTOIF(type));
5068 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
5072 * build lu direct from backend fs dirent.
5076 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
5077 char *name, __u16 namelen, __u16 type, __u32 attr)
5079 ent->lde_attrs = attr | LUDA_FID;
5080 fid_cpu_to_le(&ent->lde_fid, fid);
5082 ent->lde_hash = cpu_to_le64(offset);
5083 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
5085 strncpy(ent->lde_name, name, namelen);
5086 ent->lde_name[namelen] = '\0';
5087 ent->lde_namelen = cpu_to_le16(namelen);
5089 /* append lustre attributes */
5090 osd_it_append_attrs(ent, namelen, type);
5094 * Return pointer to the record under iterator.
5096 static int osd_it_iam_rec(const struct lu_env *env,
5097 const struct dt_it *di,
5098 struct dt_rec *dtrec, __u32 attr)
5100 struct osd_it_iam *it = (struct osd_it_iam *)di;
5101 struct osd_thread_info *info = osd_oti_get(env);
5104 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
5105 const struct osd_fid_pack *rec;
5106 struct lu_fid *fid = &info->oti_fid;
5107 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
5113 name = (char *)iam_it_key_get(&it->oi_it);
5115 RETURN(PTR_ERR(name));
5117 namelen = iam_it_key_size(&it->oi_it);
5119 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
5121 RETURN(PTR_ERR(rec));
5123 rc = osd_fid_unpack(fid, rec);
5127 hash = iam_it_store(&it->oi_it);
5129 /* IAM does not store object type in IAM index (dir) */
5130 osd_it_pack_dirent(lde, fid, hash, name, namelen,
5132 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5133 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5134 (struct iam_rec *)dtrec);
5135 osd_quota_unpack(it->oi_obj, dtrec);
5137 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5138 (struct iam_rec *)dtrec);
5145 * Returns cookie for current Iterator position.
5147 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
5149 struct osd_it_iam *it = (struct osd_it_iam *)di;
5151 return iam_it_store(&it->oi_it);
5155 * Restore iterator from cookie.
5157 * \param di osd iterator
5158 * \param hash Iterator location cookie
5160 * \retval +ve di points to record with least key not larger than key.
5161 * \retval 0 di points to exact matched key
5162 * \retval -ve failure
5165 static int osd_it_iam_load(const struct lu_env *env,
5166 const struct dt_it *di, __u64 hash)
5168 struct osd_it_iam *it = (struct osd_it_iam *)di;
5170 return iam_it_load(&it->oi_it, hash);
5173 static const struct dt_index_operations osd_index_iam_ops = {
5174 .dio_lookup = osd_index_iam_lookup,
5175 .dio_declare_insert = osd_index_declare_iam_insert,
5176 .dio_insert = osd_index_iam_insert,
5177 .dio_declare_delete = osd_index_declare_iam_delete,
5178 .dio_delete = osd_index_iam_delete,
5180 .init = osd_it_iam_init,
5181 .fini = osd_it_iam_fini,
5182 .get = osd_it_iam_get,
5183 .put = osd_it_iam_put,
5184 .next = osd_it_iam_next,
5185 .key = osd_it_iam_key,
5186 .key_size = osd_it_iam_key_size,
5187 .rec = osd_it_iam_rec,
5188 .store = osd_it_iam_store,
5189 .load = osd_it_iam_load
5195 * Creates or initializes iterator context.
5197 * \retval struct osd_it_ea, iterator structure on success
5200 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
5201 struct dt_object *dt,
5204 struct osd_object *obj = osd_dt_obj(dt);
5205 struct osd_thread_info *info = osd_oti_get(env);
5206 struct osd_it_ea *oie;
5208 struct lu_object *lo = &dt->do_lu;
5209 struct dentry *obj_dentry;
5212 if (!dt_object_exists(dt) || obj->oo_destroyed)
5213 RETURN(ERR_PTR(-ENOENT));
5215 OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
5217 RETURN(ERR_PTR(-ENOMEM));
5218 obj_dentry = &oie->oie_dentry;
5220 obj_dentry->d_inode = obj->oo_inode;
5221 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
5222 obj_dentry->d_name.hash = 0;
5224 oie->oie_rd_dirent = 0;
5225 oie->oie_it_dirent = 0;
5226 oie->oie_dirent = NULL;
5227 if (unlikely(!info->oti_it_ea_buf_used)) {
5228 oie->oie_buf = info->oti_it_ea_buf;
5229 info->oti_it_ea_buf_used = 1;
5231 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5232 if (oie->oie_buf == NULL)
5233 RETURN(ERR_PTR(-ENOMEM));
5237 file = &oie->oie_file;
5239 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
5240 if (attr & LUDA_64BITHASH)
5241 file->f_mode = FMODE_64BITHASH;
5243 file->f_mode = FMODE_32BITHASH;
5244 file->f_path.dentry = obj_dentry;
5245 file->f_mapping = obj->oo_inode->i_mapping;
5246 file->f_op = obj->oo_inode->i_fop;
5247 set_file_inode(file, obj->oo_inode);
5250 RETURN((struct dt_it *) oie);
5254 * Destroy or finishes iterator context.
5256 * \param di iterator structure to be destroyed
5258 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
5260 struct osd_thread_info *info = osd_oti_get(env);
5261 struct osd_it_ea *oie = (struct osd_it_ea *)di;
5262 struct osd_object *obj = oie->oie_obj;
5263 struct inode *inode = obj->oo_inode;
5266 oie->oie_file.f_op->release(inode, &oie->oie_file);
5267 lu_object_put(env, &obj->oo_dt.do_lu);
5268 if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
5269 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5271 info->oti_it_ea_buf_used = 0;
5272 OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
5277 * It position the iterator at given key, so that next lookup continues from
5278 * that key Or it is similar to dio_it->load() but based on a key,
5279 * rather than file position.
5281 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
5284 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
5286 static int osd_it_ea_get(const struct lu_env *env,
5287 struct dt_it *di, const struct dt_key *key)
5289 struct osd_it_ea *it = (struct osd_it_ea *)di;
5292 LASSERT(((const char *)key)[0] == '\0');
5293 it->oie_file.f_pos = 0;
5294 it->oie_rd_dirent = 0;
5295 it->oie_it_dirent = 0;
5296 it->oie_dirent = NULL;
5304 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
5308 struct osd_filldir_cbs {
5309 #ifdef HAVE_DIR_CONTEXT
5310 struct dir_context ctx;
5312 struct osd_it_ea *it;
5315 * It is called internally by ->readdir(). It fills the
5316 * iterator's in-memory data structure with required
5317 * information i.e. name, namelen, rec_size etc.
5319 * \param buf in which information to be filled in.
5320 * \param name name of the file in given dir
5322 * \retval 0 on success
5323 * \retval 1 on buffer full
5325 #ifdef HAVE_FILLDIR_USE_CTX
5326 static int osd_ldiskfs_filldir(struct dir_context *buf,
5327 const char *name, int namelen,
5329 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
5331 loff_t offset, __u64 ino,
5334 struct osd_it_ea *it =
5335 ((struct osd_filldir_cbs *)buf)->it;
5336 struct osd_object *obj = it->oie_obj;
5337 struct osd_it_ea_dirent *ent = it->oie_dirent;
5338 struct lu_fid *fid = &ent->oied_fid;
5339 struct osd_fid_pack *rec;
5342 /* this should never happen */
5343 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5344 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5348 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
5352 /* "." is just the object itself. */
5353 if (namelen == 1 && name[0] == '.') {
5354 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5355 } else if (d_type & LDISKFS_DIRENT_LUFID) {
5356 rec = (struct osd_fid_pack*) (name + namelen + 1);
5357 if (osd_fid_unpack(fid, rec) != 0)
5362 d_type &= ~LDISKFS_DIRENT_LUFID;
5364 /* NOT export local root. */
5365 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5366 ino = obj->oo_inode->i_ino;
5367 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5370 ent->oied_ino = ino;
5371 ent->oied_off = offset;
5372 ent->oied_namelen = namelen;
5373 ent->oied_type = d_type;
5375 memcpy(ent->oied_name, name, namelen);
5377 it->oie_rd_dirent++;
5378 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
5383 * Calls ->readdir() to load a directory entry at a time
5384 * and stored it in iterator's in-memory data structure.
5386 * \param di iterator's in memory structure
5388 * \retval 0 on success
5389 * \retval -ve on error
5390 * \retval +1 reach the end of entry
5392 static int osd_ldiskfs_it_fill(const struct lu_env *env,
5393 const struct dt_it *di)
5395 struct osd_it_ea *it = (struct osd_it_ea *)di;
5396 struct osd_object *obj = it->oie_obj;
5397 struct inode *inode = obj->oo_inode;
5398 struct htree_lock *hlock = NULL;
5399 struct file *filp = &it->oie_file;
5401 struct osd_filldir_cbs buf = {
5402 #ifdef HAVE_DIR_CONTEXT
5403 .ctx.actor = osd_ldiskfs_filldir,
5409 it->oie_dirent = it->oie_buf;
5410 it->oie_rd_dirent = 0;
5412 if (obj->oo_hl_head != NULL) {
5413 hlock = osd_oti_get(env)->oti_hlock;
5414 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5415 inode, LDISKFS_HLOCK_READDIR);
5417 down_read(&obj->oo_ext_idx_sem);
5420 #ifdef HAVE_DIR_CONTEXT
5421 buf.ctx.pos = filp->f_pos;
5422 rc = inode->i_fop->iterate(filp, &buf.ctx);
5423 filp->f_pos = buf.ctx.pos;
5425 rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
5429 ldiskfs_htree_unlock(hlock);
5431 up_read(&obj->oo_ext_idx_sem);
5433 if (it->oie_rd_dirent == 0) {
5434 /*If it does not get any dirent, it means it has been reached
5435 *to the end of the dir */
5436 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
5440 it->oie_dirent = it->oie_buf;
5441 it->oie_it_dirent = 1;
5448 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5449 * to load a directory entry at a time and stored it in
5450 * iterator's in-memory data structure.
5452 * \param di iterator's in memory structure
5454 * \retval +ve iterator reached to end
5455 * \retval 0 iterator not reached to end
5456 * \retval -ve on error
5458 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
5460 struct osd_it_ea *it = (struct osd_it_ea *)di;
5465 if (it->oie_it_dirent < it->oie_rd_dirent) {
5467 (void *) it->oie_dirent +
5468 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
5469 it->oie_dirent->oied_namelen);
5470 it->oie_it_dirent++;
5473 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
5476 rc = osd_ldiskfs_it_fill(env, di);
5483 * Returns the key at current position from iterator's in memory structure.
5485 * \param di iterator's in memory structure
5487 * \retval key i.e. struct dt_key on success
5489 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
5490 const struct dt_it *di)
5492 struct osd_it_ea *it = (struct osd_it_ea *)di;
5494 return (struct dt_key *)it->oie_dirent->oied_name;
5498 * Returns the key's size at current position from iterator's in memory structure.
5500 * \param di iterator's in memory structure
5502 * \retval key_size i.e. struct dt_key on success
5504 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
5506 struct osd_it_ea *it = (struct osd_it_ea *)di;
5508 return it->oie_dirent->oied_namelen;
5512 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5514 LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5515 "dot_dotdot = %d\n", dot_dotdot);
5517 if (LDISKFS_DIR_REC_LEN(de) >=
5518 __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5525 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
5526 unsigned blocksize, int dot_dotdot)
5529 return osd_dot_dotdot_has_space(de, dot_dotdot);
5531 if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
5532 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5539 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5540 struct dentry *dentry, const struct lu_fid *fid,
5541 struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5542 struct htree_lock *hlock, int dot_dotdot)
5544 struct inode *dir = dentry->d_parent->d_inode;
5545 struct inode *inode = dentry->d_inode;
5546 struct osd_fid_pack *rec;
5547 struct ldiskfs_dentry_param *ldp;
5548 int namelen = dentry->d_name.len;
5550 struct osd_thread_info *info = osd_oti_get(env);
5553 if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5554 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5557 /* There is enough space to hold the FID-in-dirent. */
5558 if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize,
5560 rc = ldiskfs_journal_get_write_access(jh, bh);
5564 de->name[namelen] = 0;
5565 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
5566 rec->fp_len = sizeof(struct lu_fid) + 1;
5567 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5568 de->file_type |= LDISKFS_DIRENT_LUFID;
5569 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
5574 LASSERTF(dot_dotdot == 0, "dot_dotdot = %d\n", dot_dotdot);
5576 rc = ldiskfs_delete_entry(jh, dir, de, bh);
5580 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5581 osd_get_ldiskfs_dirent_param(ldp, fid);
5582 dentry->d_fsdata = (void *)ldp;
5583 ll_vfs_dq_init(dir);
5584 rc = osd_ldiskfs_add_entry(info, jh, dentry, inode, hlock);
5585 /* It is too bad, we cannot reinsert the name entry back.
5586 * That means we lose it! */
5588 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
5589 "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
5590 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5591 dir->i_ino, dir->i_generation, namelen,
5592 dentry->d_name.name, PFID(fid), rc);
5598 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5599 struct osd_it_ea *it, struct lu_fid *fid,
5600 struct osd_inode_id *id, __u32 *attr)
5602 struct osd_thread_info *info = osd_oti_get(env);
5603 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
5604 struct osd_device *dev = osd_obj2dev(obj);
5605 struct super_block *sb = osd_sb(dev);
5606 const char *devname =
5607 LDISKFS_SB(sb)->s_es->s_volume_name;
5608 struct osd_it_ea_dirent *ent = it->oie_dirent;
5609 struct inode *dir = obj->oo_inode;
5610 struct htree_lock *hlock = NULL;
5611 struct buffer_head *bh = NULL;
5612 handle_t *jh = NULL;
5613 struct ldiskfs_dir_entry_2 *de;
5614 struct dentry *dentry;
5615 struct inode *inode;
5622 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5623 inode = osd_iget(info, dev, id);
5624 if (IS_ERR(inode)) {
5625 rc = PTR_ERR(inode);
5626 if (rc == -ENOENT || rc == -ESTALE) {
5627 *attr |= LUDA_UNKNOWN;
5630 CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
5631 "check_repair, dir = %lu/%u, name = %.*s: "
5633 devname, dir->i_ino, dir->i_generation,
5634 ent->oied_namelen, ent->oied_name, rc);
5640 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5642 rc = osd_get_lma(info, inode, dentry, lma);
5648 if (ent->oied_name[0] == '.') {
5649 if (ent->oied_namelen == 1)
5651 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5655 /* We need to ensure that the name entry is still valid.
5656 * Because it may be removed or renamed by other already.
5658 * The unlink or rename operation will start journal before PDO lock,
5659 * so to avoid deadlock, here we need to start journal handle before
5660 * related PDO lock also. But because we do not know whether there
5661 * will be something to be repaired before PDO lock, we just start
5662 * journal without conditions.
5664 * We may need to remove the name entry firstly, then insert back.
5665 * One credit is for user quota file update.
5666 * One credit is for group quota file update.
5667 * Two credits are for dirty inode. */
5668 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5669 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5671 if (dev->od_dirent_journal != 0) {
5674 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5677 CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent "
5678 "check_repair, dir = %lu/%u, credits = %d, "
5679 "name = %.*s: rc = %d\n",
5680 devname, dir->i_ino, dir->i_generation, credits,
5681 ent->oied_namelen, ent->oied_name, rc);
5683 GOTO(out_inode, rc);
5686 if (obj->oo_hl_head != NULL) {
5687 hlock = osd_oti_get(env)->oti_hlock;
5688 /* "0" means exclusive lock for the whole directory.
5689 * We need to prevent others access such name entry
5690 * during the delete + insert. Neither HLOCK_ADD nor
5691 * HLOCK_DEL cannot guarantee the atomicity. */
5692 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5694 down_write(&obj->oo_ext_idx_sem);
5697 if (obj->oo_hl_head != NULL) {
5698 hlock = osd_oti_get(env)->oti_hlock;
5699 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5700 LDISKFS_HLOCK_LOOKUP);
5702 down_read(&obj->oo_ext_idx_sem);
5706 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5707 /* For dot/dotdot entry, if there is not enough space to hold the
5708 * FID-in-dirent, just keep them there. It only happens when the
5709 * device upgraded from 1.8 or restored from MDT file-level backup.
5710 * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5711 * and needs to get FID from LMA when readdir, it will not affect the
5712 * performance much. */
5713 if ((bh == NULL) || (le32_to_cpu(de->inode) != inode->i_ino) ||
5714 (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5715 *attr |= LUDA_IGNORE;
5721 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
5722 struct lu_fid *tfid = &lma->lma_self_fid;
5724 *attr |= LUDA_IGNORE;
5725 /* It must be REMOTE_PARENT_DIR and as the
5726 * dotdot entry of remote directory */
5727 if (unlikely(dot_dotdot != 2 ||
5728 fid_seq(tfid) != FID_SEQ_LOCAL_FILE ||
5729 fid_oid(tfid) != REMOTE_PARENT_DIR_OID)) {
5730 CDEBUG(D_LFSCK, "%.16s: expect remote agent "
5731 "parent directory, but got %.*s under "
5732 "dir = %lu/%u with the FID "DFID"\n",
5733 devname, ent->oied_namelen,
5734 ent->oied_name, dir->i_ino,
5735 dir->i_generation, PFID(tfid));
5737 GOTO(out, rc = -EIO);
5743 if (fid_is_sane(fid)) {
5744 /* FID-in-dirent is valid. */
5745 if (lu_fid_eq(fid, &lma->lma_self_fid))
5748 /* Do not repair under dryrun mode. */
5749 if (*attr & LUDA_VERIFY_DRYRUN) {
5750 *attr |= LUDA_REPAIR;
5757 dev->od_dirent_journal = 1;
5758 if (hlock != NULL) {
5759 ldiskfs_htree_unlock(hlock);
5762 up_read(&obj->oo_ext_idx_sem);
5768 *fid = lma->lma_self_fid;
5770 /* Update the FID-in-dirent. */
5771 rc = osd_dirent_reinsert(env, jh, dentry, fid, bh, de,
5774 *attr |= LUDA_REPAIR;
5776 CDEBUG(D_LFSCK, "%.16s: fail to update FID "
5777 "in the dirent, dir = %lu/%u, "
5778 "name = %.*s, "DFID": rc = %d\n",
5779 devname, dir->i_ino, dir->i_generation,
5780 ent->oied_namelen, ent->oied_name,
5783 /* Do not repair under dryrun mode. */
5784 if (*attr & LUDA_VERIFY_DRYRUN) {
5785 *fid = lma->lma_self_fid;
5786 *attr |= LUDA_REPAIR;
5793 dev->od_dirent_journal = 1;
5794 if (hlock != NULL) {
5795 ldiskfs_htree_unlock(hlock);
5798 up_read(&obj->oo_ext_idx_sem);
5804 *fid = lma->lma_self_fid;
5806 /* Append the FID-in-dirent. */
5807 rc = osd_dirent_reinsert(env, jh, dentry, fid, bh, de,
5810 *attr |= LUDA_REPAIR;
5812 CDEBUG(D_LFSCK, "%.16s: fail to append FID "
5813 "after the dirent, dir = %lu/%u, "
5814 "name = %.*s, "DFID": rc = %d\n",
5815 devname, dir->i_ino, dir->i_generation,
5816 ent->oied_namelen, ent->oied_name,
5820 /* Do not repair under dryrun mode. */
5821 if (*attr & LUDA_VERIFY_DRYRUN) {
5822 if (fid_is_sane(fid)) {
5823 *attr |= LUDA_REPAIR;
5825 lu_igif_build(fid, inode->i_ino,
5826 inode->i_generation);
5827 *attr |= LUDA_UPGRADE;
5835 dev->od_dirent_journal = 1;
5836 if (hlock != NULL) {
5837 ldiskfs_htree_unlock(hlock);
5840 up_read(&obj->oo_ext_idx_sem);
5847 if (unlikely(fid_is_sane(fid))) {
5848 /* FID-in-dirent exists, but FID-in-LMA is lost.
5849 * Trust the FID-in-dirent, and add FID-in-LMA. */
5850 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5852 *attr |= LUDA_REPAIR;
5854 CDEBUG(D_LFSCK, "%.16s: fail to set LMA for "
5855 "update dirent, dir = %lu/%u, "
5856 "name = %.*s, "DFID": rc = %d\n",
5857 devname, dir->i_ino, dir->i_generation,
5858 ent->oied_namelen, ent->oied_name,
5861 lu_igif_build(fid, inode->i_ino, inode->i_generation);
5862 /* It is probably IGIF object. Only aappend the
5863 * FID-in-dirent. OI scrub will process FID-in-LMA. */
5864 rc = osd_dirent_reinsert(env, jh, dentry, fid, bh, de,
5867 *attr |= LUDA_UPGRADE;
5869 CDEBUG(D_LFSCK, "%.16s: fail to append IGIF "
5870 "after the dirent, dir = %lu/%u, "
5871 "name = %.*s, "DFID": rc = %d\n",
5872 devname, dir->i_ino, dir->i_generation,
5873 ent->oied_namelen, ent->oied_name,
5882 if (hlock != NULL) {
5883 ldiskfs_htree_unlock(hlock);
5885 if (dev->od_dirent_journal != 0)
5886 up_write(&obj->oo_ext_idx_sem);
5888 up_read(&obj->oo_ext_idx_sem);
5892 ldiskfs_journal_stop(jh);
5896 if (rc >= 0 && !dirty)
5897 dev->od_dirent_journal = 0;
5903 * Returns the value at current position from iterator's in memory structure.
5905 * \param di struct osd_it_ea, iterator's in memory structure
5906 * \param attr attr requested for dirent.
5907 * \param lde lustre dirent
5909 * \retval 0 no error and \param lde has correct lustre dirent.
5910 * \retval -ve on error
5912 static inline int osd_it_ea_rec(const struct lu_env *env,
5913 const struct dt_it *di,
5914 struct dt_rec *dtrec, __u32 attr)
5916 struct osd_it_ea *it = (struct osd_it_ea *)di;
5917 struct osd_object *obj = it->oie_obj;
5918 struct osd_device *dev = osd_obj2dev(obj);
5919 struct osd_thread_info *oti = osd_oti_get(env);
5920 struct osd_inode_id *id = &oti->oti_id;
5921 struct lu_fid *fid = &it->oie_dirent->oied_fid;
5922 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
5923 __u32 ino = it->oie_dirent->oied_ino;
5927 LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
5929 if (attr & LUDA_VERIFY) {
5930 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5931 attr |= LUDA_IGNORE;
5932 /* If the parent is on remote MDT, and there
5933 * is no FID-in-dirent, then we have to get
5934 * the parent FID from the linkEA. */
5935 if (!fid_is_sane(fid) &&
5936 it->oie_dirent->oied_namelen == 2 &&
5937 it->oie_dirent->oied_name[0] == '.' &&
5938 it->oie_dirent->oied_name[1] == '.')
5939 osd_get_pfid_from_linkea(env, obj, fid);
5941 rc = osd_dirent_check_repair(env, obj, it, fid, id,
5945 if (!fid_is_sane(fid)) {
5946 attr &= ~LUDA_IGNORE;
5947 attr |= LUDA_UNKNOWN;
5950 attr &= ~LU_DIRENT_ATTRS_MASK;
5951 if (!fid_is_sane(fid)) {
5952 bool is_dotdot = false;
5953 if (it->oie_dirent->oied_namelen == 2 &&
5954 it->oie_dirent->oied_name[0] == '.' &&
5955 it->oie_dirent->oied_name[1] == '.')
5957 /* If the parent is on remote MDT, and there
5958 * is no FID-in-dirent, then we have to get
5959 * the parent FID from the linkEA. */
5960 if (ino == osd_remote_parent_ino(dev) && is_dotdot) {
5961 rc = osd_get_pfid_from_linkea(env, obj, fid);
5963 if (is_dotdot == false &&
5964 OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
5967 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5970 osd_id_gen(id, ino, OSD_OII_NOGEN);
5974 /* Pack the entry anyway, at least the offset is right. */
5975 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5976 it->oie_dirent->oied_name,
5977 it->oie_dirent->oied_namelen,
5978 it->oie_dirent->oied_type, attr);
5983 if (osd_remote_fid(env, dev, fid))
5986 if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
5987 osd_add_oi_cache(oti, dev, id, fid);
5989 RETURN(rc > 0 ? 0 : rc);
5993 * Returns the record size size at current position.
5995 * This function will return record(lu_dirent) size in bytes.
5997 * \param[in] env execution environment
5998 * \param[in] di iterator's in memory structure
5999 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
6000 * calculate the lu_dirent size.
6002 * \retval record size(in bytes & in memory) of the current lu_dirent
6005 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
6008 struct osd_it_ea *it = (struct osd_it_ea *)di;
6010 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
6014 * Returns a cookie for current position of the iterator head, so that
6015 * user can use this cookie to load/start the iterator next time.
6017 * \param di iterator's in memory structure
6019 * \retval cookie for current position, on success
6021 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
6023 struct osd_it_ea *it = (struct osd_it_ea *)di;
6025 return it->oie_dirent->oied_off;
6029 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6030 * to load a directory entry at a time and stored it i inn,
6031 * in iterator's in-memory data structure.
6033 * \param di struct osd_it_ea, iterator's in memory structure
6035 * \retval +ve on success
6036 * \retval -ve on error
6038 static int osd_it_ea_load(const struct lu_env *env,
6039 const struct dt_it *di, __u64 hash)
6041 struct osd_it_ea *it = (struct osd_it_ea *)di;
6045 it->oie_file.f_pos = hash;
6047 rc = osd_ldiskfs_it_fill(env, di);
6058 * Index lookup function for interoperability mode (b11826).
6060 * \param key, key i.e. file name to be searched
6062 * \retval +ve, on success
6063 * \retval -ve, on error
6065 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
6066 struct dt_rec *rec, const struct dt_key *key)
6068 struct osd_object *obj = osd_dt_obj(dt);
6073 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
6074 LINVRNT(osd_invariant(obj));
6076 rc = osd_ea_lookup_rec(env, obj, rec, key);
6083 * Index and Iterator operations for interoperability
6084 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
6086 static const struct dt_index_operations osd_index_ea_ops = {
6087 .dio_lookup = osd_index_ea_lookup,
6088 .dio_declare_insert = osd_index_declare_ea_insert,
6089 .dio_insert = osd_index_ea_insert,
6090 .dio_declare_delete = osd_index_declare_ea_delete,
6091 .dio_delete = osd_index_ea_delete,
6093 .init = osd_it_ea_init,
6094 .fini = osd_it_ea_fini,
6095 .get = osd_it_ea_get,
6096 .put = osd_it_ea_put,
6097 .next = osd_it_ea_next,
6098 .key = osd_it_ea_key,
6099 .key_size = osd_it_ea_key_size,
6100 .rec = osd_it_ea_rec,
6101 .rec_size = osd_it_ea_rec_size,
6102 .store = osd_it_ea_store,
6103 .load = osd_it_ea_load
6107 static void *osd_key_init(const struct lu_context *ctx,
6108 struct lu_context_key *key)
6110 struct osd_thread_info *info;
6112 OBD_ALLOC_PTR(info);
6114 return ERR_PTR(-ENOMEM);
6116 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6117 if (info->oti_it_ea_buf == NULL)
6120 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
6122 info->oti_hlock = ldiskfs_htree_lock_alloc();
6123 if (info->oti_hlock == NULL)
6129 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6132 return ERR_PTR(-ENOMEM);
6135 static void osd_key_fini(const struct lu_context *ctx,
6136 struct lu_context_key *key, void* data)
6138 struct osd_thread_info *info = data;
6139 struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
6140 struct osd_idmap_cache *idc = info->oti_ins_cache;
6142 if (info->oti_inode != NULL)
6144 if (info->oti_hlock != NULL)
6145 ldiskfs_htree_lock_free(info->oti_hlock);
6146 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6147 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
6148 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
6149 lu_buf_free(&info->oti_big_buf);
6151 LASSERT(info->oti_ins_cache_size > 0);
6152 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
6153 info->oti_ins_cache = NULL;
6154 info->oti_ins_cache_size = 0;
6159 static void osd_key_exit(const struct lu_context *ctx,
6160 struct lu_context_key *key, void *data)
6162 struct osd_thread_info *info = data;
6164 LASSERT(info->oti_r_locks == 0);
6165 LASSERT(info->oti_w_locks == 0);
6166 LASSERT(info->oti_txns == 0);
6169 /* type constructor/destructor: osd_type_init, osd_type_fini */
6170 LU_TYPE_INIT_FINI(osd, &osd_key);
6172 struct lu_context_key osd_key = {
6173 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
6174 .lct_init = osd_key_init,
6175 .lct_fini = osd_key_fini,
6176 .lct_exit = osd_key_exit
6180 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
6181 const char *name, struct lu_device *next)
6183 struct osd_device *osd = osd_dev(d);
6185 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
6186 >= sizeof(osd->od_svname))
6188 return osd_procfs_init(osd, name);
6191 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
6193 struct seq_server_site *ss = osd_seq_site(osd);
6197 if (osd->od_is_ost || osd->od_cl_seq != NULL)
6200 if (unlikely(ss == NULL))
6203 OBD_ALLOC_PTR(osd->od_cl_seq);
6204 if (osd->od_cl_seq == NULL)
6207 rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
6208 osd->od_svname, ss->ss_server_seq);
6210 OBD_FREE_PTR(osd->od_cl_seq);
6211 osd->od_cl_seq = NULL;
6215 if (ss->ss_node_id == 0) {
6216 /* If the OSD on the sequence controller(MDT0), then allocate
6217 * sequence here, otherwise allocate sequence after connected
6218 * to MDT0 (see mdt_register_lwp_callback()). */
6219 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
6220 &osd->od_cl_seq->lcs_space, env);
6226 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
6228 if (osd->od_cl_seq == NULL)
6231 seq_client_fini(osd->od_cl_seq);
6232 OBD_FREE_PTR(osd->od_cl_seq);
6233 osd->od_cl_seq = NULL;
6236 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
6240 /* shutdown quota slave instance associated with the device */
6241 if (o->od_quota_slave != NULL) {
6242 qsd_fini(env, o->od_quota_slave);
6243 o->od_quota_slave = NULL;
6246 osd_fid_fini(env, o);
6251 static void osd_umount(const struct lu_env *env, struct osd_device *o)
6255 if (o->od_mnt != NULL) {
6256 shrink_dcache_sb(osd_sb(o));
6257 osd_sync(env, &o->od_dt_dev);
6266 static int osd_mount(const struct lu_env *env,
6267 struct osd_device *o, struct lustre_cfg *cfg)
6269 const char *name = lustre_cfg_string(cfg, 0);
6270 const char *dev = lustre_cfg_string(cfg, 1);
6272 unsigned long page, s_flags, lmd_flags = 0;
6273 struct page *__page;
6274 struct file_system_type *type;
6275 char *options = NULL;
6277 struct osd_thread_info *info = osd_oti_get(env);
6278 struct lu_fid *fid = &info->oti_fid;
6279 struct inode *inode;
6280 int rc = 0, force_over_256tb = 0;
6283 if (o->od_mnt != NULL)
6286 if (strlen(dev) >= sizeof(o->od_mntdev))
6288 strcpy(o->od_mntdev, dev);
6290 str = lustre_cfg_string(cfg, 2);
6291 s_flags = simple_strtoul(str, NULL, 0);
6292 str = strstr(str, ":");
6294 lmd_flags = simple_strtoul(str + 1, NULL, 0);
6295 opts = lustre_cfg_string(cfg, 3);
6297 if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
6298 CERROR("%s: device %s extents feature is not guaranteed to "
6299 "work on big-endian systems. Use \"bigendian_extents\" "
6300 "mount option to override.\n", name, dev);
6304 if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
6305 CWARN("force_over_128tb option is depricated."
6306 "Filesystems less then 256TB can be created without any"
6307 "force options. Use force_over_256tb option for"
6308 "filesystems greather then 256TB.\n");
6311 if (opts != NULL && strstr(opts, "force_over_256tb") != NULL)
6312 force_over_256tb = 1;
6314 __page = alloc_page(GFP_KERNEL);
6316 GOTO(out, rc = -ENOMEM);
6317 page = (unsigned long)page_address(__page);
6318 options = (char *)page;
6321 /* strip out the options for back compatiblity */
6322 static char *sout[] = {
6329 /* strip out option we processed in osd */
6330 "bigendian_extents",
6331 #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(3,0,53,0)
6332 #warning "remove force_over_128 option"
6334 "force_over_128tb (deprecated)",
6339 strcat(options, opts);
6340 for (rc = 0, str = options; sout[rc]; ) {
6341 char *op = strstr(str, sout[rc]);
6347 if (op == options || *(op - 1) == ',') {
6348 str = op + strlen(sout[rc]);
6349 if (*str == ',' || *str == '\0') {
6350 *str == ',' ? str++ : str;
6351 memmove(op, str, strlen(str) + 1);
6354 for (str = op; *str != ',' && *str != '\0'; str++)
6358 strncat(options, "user_xattr,acl", 14);
6361 /* Glom up mount options */
6362 if (*options != '\0')
6363 strcat(options, ",");
6364 strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
6366 type = get_fs_type("ldiskfs");
6368 CERROR("%s: cannot find ldiskfs module\n", name);
6369 GOTO(out, rc = -ENODEV);
6372 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
6373 module_put(type->owner);
6375 if (IS_ERR(o->od_mnt)) {
6376 rc = PTR_ERR(o->od_mnt);
6378 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
6382 if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) > (64ULL << 30) &&
6383 force_over_256tb == 0) {
6384 CERROR("%s: device %s LDISKFS does not support filesystems "
6385 "greater than 256TB and can cause data corruption. "
6386 "Use \"force_over_256tb\" mount option to override.\n",
6388 GOTO(out, rc = -EINVAL);
6391 #ifdef HAVE_DEV_SET_RDONLY
6392 if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
6393 CERROR("%s: underlying device %s is marked as read-only. "
6394 "Setup failed\n", name, dev);
6395 GOTO(out_mnt, rc = -EROFS);
6399 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
6400 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
6401 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
6402 GOTO(out_mnt, rc = -EINVAL);
6405 #ifdef LDISKFS_MOUNT_DIRDATA
6406 if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
6407 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6408 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
6409 else if (!o->od_is_ost)
6410 CWARN("%s: device %s was upgraded from Lustre-1.x without "
6411 "enabling the dirdata feature. If you do not want to "
6412 "downgrade to Lustre-1.x again, you can enable it via "
6413 "'tune2fs -O dirdata device'\n", name, dev);
6415 inode = osd_sb(o)->s_root->d_inode;
6416 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
6417 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
6419 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
6423 if (lmd_flags & LMD_FLG_NOSCRUB)
6434 __free_page(__page);
6439 static struct lu_device *osd_device_fini(const struct lu_env *env,
6440 struct lu_device *d)
6442 struct osd_device *o = osd_dev(d);
6445 osd_shutdown(env, o);
6447 osd_scrub_cleanup(env, o);
6448 osd_obj_map_fini(o);
6454 static int osd_device_init0(const struct lu_env *env,
6455 struct osd_device *o,
6456 struct lustre_cfg *cfg)
6458 struct lu_device *l = osd2lu_dev(o);
6459 struct osd_thread_info *info;
6463 /* if the module was re-loaded, env can loose its keys */
6464 rc = lu_env_refill((struct lu_env *) env);
6467 info = osd_oti_get(env);
6470 l->ld_ops = &osd_lu_ops;
6471 o->od_dt_dev.dd_ops = &osd_dt_ops;
6473 spin_lock_init(&o->od_osfs_lock);
6474 mutex_init(&o->od_otable_mutex);
6475 INIT_LIST_HEAD(&o->od_orphan_list);
6477 o->od_read_cache = 1;
6478 o->od_writethrough_cache = 1;
6479 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
6481 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
6482 sizeof(o->od_svname));
6483 if (cplen >= sizeof(o->od_svname)) {
6488 o->od_index = -1; /* -1 means index is invalid */
6489 rc = server_name2index(o->od_svname, &o->od_index, NULL);
6490 if (rc == LDD_F_SV_TYPE_OST)
6493 o->od_full_scrub_ratio = OFSR_DEFAULT;
6494 o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
6495 rc = osd_mount(env, o, cfg);
6499 rc = osd_obj_map_init(env, o);
6503 rc = lu_site_init(&o->od_site, l);
6505 GOTO(out_compat, rc);
6506 o->od_site.ls_bottom_dev = l;
6508 rc = lu_site_init_finish(&o->od_site);
6512 INIT_LIST_HEAD(&o->od_ios_list);
6513 /* setup scrub, including OI files initialization */
6514 rc = osd_scrub_setup(env, o);
6518 rc = osd_procfs_init(o, o->od_svname);
6520 CERROR("%s: can't initialize procfs: rc = %d\n",
6522 GOTO(out_scrub, rc);
6525 LASSERT(l->ld_site->ls_linkage.next != NULL);
6526 LASSERT(l->ld_site->ls_linkage.prev != NULL);
6528 /* initialize quota slave instance */
6529 o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
6531 if (IS_ERR(o->od_quota_slave)) {
6532 rc = PTR_ERR(o->od_quota_slave);
6533 o->od_quota_slave = NULL;
6534 GOTO(out_procfs, rc);
6542 osd_scrub_cleanup(env, o);
6544 lu_site_fini(&o->od_site);
6546 osd_obj_map_fini(o);
6553 static struct lu_device *osd_device_alloc(const struct lu_env *env,
6554 struct lu_device_type *t,
6555 struct lustre_cfg *cfg)
6557 struct osd_device *o;
6562 return ERR_PTR(-ENOMEM);
6564 rc = dt_device_init(&o->od_dt_dev, t);
6566 /* Because the ctx might be revived in dt_device_init,
6567 * refill the env here */
6568 lu_env_refill((struct lu_env *)env);
6569 rc = osd_device_init0(env, o, cfg);
6571 dt_device_fini(&o->od_dt_dev);
6574 if (unlikely(rc != 0))
6577 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
6580 static struct lu_device *osd_device_free(const struct lu_env *env,
6581 struct lu_device *d)
6583 struct osd_device *o = osd_dev(d);
6586 /* XXX: make osd top device in order to release reference */
6587 d->ld_site->ls_top_dev = d;
6588 lu_site_purge(env, d->ld_site, -1);
6589 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
6590 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
6591 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
6593 lu_site_fini(&o->od_site);
6594 dt_device_fini(&o->od_dt_dev);
6599 static int osd_process_config(const struct lu_env *env,
6600 struct lu_device *d, struct lustre_cfg *cfg)
6602 struct osd_device *o = osd_dev(d);
6606 switch (cfg->lcfg_command) {
6608 rc = osd_mount(env, o, cfg);
6611 lu_dev_del_linkage(d->ld_site, d);
6612 rc = osd_shutdown(env, o);
6615 LASSERT(&o->od_dt_dev);
6616 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
6617 cfg, &o->od_dt_dev);
6618 if (rc > 0 || rc == -ENOSYS)
6619 rc = class_process_proc_param(PARAM_OST,
6620 lprocfs_osd_obd_vars,
6621 cfg, &o->od_dt_dev);
6630 static int osd_recovery_complete(const struct lu_env *env,
6631 struct lu_device *d)
6633 struct osd_device *osd = osd_dev(d);
6637 if (osd->od_quota_slave == NULL)
6640 /* start qsd instance on recovery completion, this notifies the quota
6641 * slave code that we are about to process new requests now */
6642 rc = qsd_start(env, osd->od_quota_slave);
6647 * we use exports to track all osd users
6649 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
6650 struct obd_device *obd, struct obd_uuid *cluuid,
6651 struct obd_connect_data *data, void *localdata)
6653 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6654 struct lustre_handle conn;
6658 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6660 rc = class_connect(&conn, obd, cluuid);
6664 *exp = class_conn2export(&conn);
6666 spin_lock(&osd->od_osfs_lock);
6668 spin_unlock(&osd->od_osfs_lock);
6674 * once last export (we don't count self-export) disappeared
6675 * osd can be released
6677 static int osd_obd_disconnect(struct obd_export *exp)
6679 struct obd_device *obd = exp->exp_obd;
6680 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6681 int rc, release = 0;
6684 /* Only disconnect the underlying layers on the final disconnect. */
6685 spin_lock(&osd->od_osfs_lock);
6687 if (osd->od_connects == 0)
6689 spin_unlock(&osd->od_osfs_lock);
6691 rc = class_disconnect(exp); /* bz 9811 */
6693 if (rc == 0 && release)
6694 class_manual_cleanup(obd);
6698 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6699 struct lu_device *dev)
6701 struct osd_device *osd = osd_dev(dev);
6702 struct lr_server_data *lsd =
6703 &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
6707 if (osd->od_quota_slave != NULL) {
6708 /* set up quota slave objects */
6709 result = qsd_prepare(env, osd->od_quota_slave);
6714 if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
6715 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
6716 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
6717 osd->od_index_in_idif = 1;
6719 osd->od_index_in_idif = 0;
6720 result = osd_register_proc_index_in_idif(osd);
6725 osd->od_index_in_idif = 1;
6729 result = osd_fid_init(env, osd);
6734 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6735 struct lu_fid *fid, struct md_op_data *op_data)
6737 struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6739 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6742 static const struct lu_object_operations osd_lu_obj_ops = {
6743 .loo_object_init = osd_object_init,
6744 .loo_object_delete = osd_object_delete,
6745 .loo_object_release = osd_object_release,
6746 .loo_object_free = osd_object_free,
6747 .loo_object_print = osd_object_print,
6748 .loo_object_invariant = osd_object_invariant
6751 const struct lu_device_operations osd_lu_ops = {
6752 .ldo_object_alloc = osd_object_alloc,
6753 .ldo_process_config = osd_process_config,
6754 .ldo_recovery_complete = osd_recovery_complete,
6755 .ldo_prepare = osd_prepare,
6758 static const struct lu_device_type_operations osd_device_type_ops = {
6759 .ldto_init = osd_type_init,
6760 .ldto_fini = osd_type_fini,
6762 .ldto_start = osd_type_start,
6763 .ldto_stop = osd_type_stop,
6765 .ldto_device_alloc = osd_device_alloc,
6766 .ldto_device_free = osd_device_free,
6768 .ldto_device_init = osd_device_init,
6769 .ldto_device_fini = osd_device_fini
6772 static struct lu_device_type osd_device_type = {
6773 .ldt_tags = LU_DEVICE_DT,
6774 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
6775 .ldt_ops = &osd_device_type_ops,
6776 .ldt_ctx_tags = LCT_LOCAL,
6779 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
6781 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6782 struct super_block *sb = osd_sb(osd);
6784 return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
6788 * lprocfs legacy support.
6790 static struct obd_ops osd_obd_device_ops = {
6791 .o_owner = THIS_MODULE,
6792 .o_connect = osd_obd_connect,
6793 .o_disconnect = osd_obd_disconnect,
6794 .o_fid_alloc = osd_fid_alloc,
6795 .o_health_check = osd_health_check,
6798 static int __init osd_init(void)
6802 LASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
6803 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
6804 /* please, try to keep osd_thread_info smaller than a page */
6805 CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
6810 rc = lu_kmem_init(ldiskfs_caches);
6814 rc = class_register_type(&osd_obd_device_ops, NULL, true,
6815 lprocfs_osd_module_vars,
6816 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6818 lu_kmem_fini(ldiskfs_caches);
6822 static void __exit osd_exit(void)
6824 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6825 lu_kmem_fini(ldiskfs_caches);
6828 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6829 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6830 MODULE_VERSION(LUSTRE_VERSION_STRING);
6831 MODULE_LICENSE("GPL");
6833 module_init(osd_init);
6834 module_exit(osd_exit);